# BIOLOGY OF BRAIN DISORDERS

EDITED BY: Daniela Tropea and Andrew Harkin PUBLISHED IN: Frontiers in Cellular Neuroscience, Frontiers in Neuroscience, Frontiers in Molecular Neuroscience, Frontiers in Aging Neuroscience, Frontiers in Physiology and Frontiers in Neurology

#### *Frontiers Copyright Statement*

*© Copyright 2007-2018 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

*The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.*

*Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.*

*Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.*

*As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.*

> *All copyright, and all rights therein, are protected by national and international copyright laws.*

> *The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use.*

ISSN 1664-8714 ISBN 978-2-88945-380-1 DOI 10.3389/978-2-88945-380-1

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# **BIOLOGY OF BRAIN DISORDERS**

Topic Editors: **Daniela Tropea,** Trinity College Dublin, Ireland **Andrew Harkin,** Trinity College Dublin, Ireland

The graphics represents a brain where functional pathways are represented as the path of an underground train. External elements influence brain development and function: genes, environment, early development.

The editors would like to acknowledge Albert Sanfeliu for his assistance in the production of this illustration.

Image by Daniela Tropea and Albert Sanfeliu.

Brain disorders, including neurological and neuropsychiatric conditions, represent a challenge for public health systems and society at large. The limited knowledge of their biology hampers the development of diagnostic tools and effective therapeutics. A clear understanding of the mechanisms that underlie the onset and progression of brain disorders is required in order to identify new avenues for therapeutic intervention.

Overlapping genetic risk factors across different brain disorders suggest common linkages and pathophysiological mechanisms that underlie brain disorders. Methodological and technological advances are leading to new insights that go beyond traditional hypotheses. Taking account of underlying molecular, cellular and systems biology underlying brain function will play an important role in the classification of brain disorders in future.

In this Research Topic, the latest advances in our understanding of biological mechanisms across different brain disorders are presented. The areas covered include developments in neurogenetics, epigenetics, plasticity, glial cell biology, neuroimmune interactions and new technologies associated with the study of brain function. Examples of how understanding of biological mechanisms are translating into research strategies that aim to advance diagnoses and treatment of brain disorders are discussed.

**Citation:** Tropea, D., Harkin, A., eds. (2018). Biology of Brain Disorders. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-380-1

# Table of Contents

#### **Editorial**

*08 Editorial: Biology of Brain Disorders* Daniela Tropea and Andrew Harkin

#### **Brain Imaging**

*10 Recent Advances in Translational Magnetic Resonance Imaging in Animal Models of Stress and Depression*

Allison L. McIntosh, Shane Gormley, Leonardo Tozzi, Thomas Frodl and Andrew Harkin

*25 Richness in Functional Connectivity Depends on the Neuronal Integrity within the Posterior Cingulate Cortex*

Anton R. Lord, Meng Li, Liliana R. Demenescu, Johan van den Meer, Viola Borchardt, Anna Linda Krause, Hans-Jochen Heinze, Michael Breakspear and Martin Walter

*34 Investigating the Role of Glutamate and GABA in the Modulation of Transthalamic Activity: A Combined fMRI-fMRS Study* Nathalie Just and Sarah Sonnay

#### **Damaged Brain**


Yu Shanshan, Jiang Beibei, Tan Li, Gao Minna, Lei Shipeng, Peng Li and Zhao Yong

*97 Calpastatin Overexpression Preserves Cognitive Function Following Seizures, While Maintaining Post-Injury Neurogenesis*

Vanessa M. Machado, Ana Sofia Lourenço, Cláudia Florindo, Raquel Fernandes, Caetana M. Carvalho and Inês M. Araújo

*113 Neuroplastic Changes Following Brain Ischemia and their Contribution to Stroke Recovery: Novel Approaches in Neurorehabilitation*

Claudia Alia, Cristina Spalletti, Stefano Lai, Alessandro Panarese, Giuseppe Lamola, Federica Bertolucci, Fabio Vallone, Angelo Di Garbo, Carmelo Chisari, Silvestro Micera and Matteo Caleo

*135 Influence of Aerobic Training and Combinations of Interventions on Cognition and Neuroplasticity after Stroke*

Annabelle Constans, Caroline Pin-barre, Jean-Jacques Temprado, Patrick Decherchi and Jérôme Laurin

*152 C1q/Tumor Necrosis Factor-Related Protein-3 Attenuates Brain Injury after Intracerebral Hemorrhage via AMPK-Dependent Pathway in Rat*

Shaohua Wang, Yang Zhou, Yang Bo, Lingyu Li, Shanshan Yu, Yanlin Chen, Jin Zhu and Yong Zhao

#### **Genetic Risk Factors**

*162 Intracellular Fibroblast Growth Factor 14: Emerging Risk Factor for Brain Disorders*

Jessica Di Re, Paul A. Wadsworth and Fernanda Laezza

*169 Early Origin and Evolution of the Angelman Syndrome Ubiquitin Ligase Gene*  **Ube3a**

Masaaki Sato


Chujun Wu and Dongsheng Fan

*199 Slitrk Missense Mutations Associated with Neuropsychiatric Disorders Distinctively Impair Slitrk Trafficking and Synapse Formation* Hyeyeon Kang, Kyung Ah Han, Seoung Youn Won, Ho Min Kim, Young-Ho Lee, Jaewon Ko and Ji Won Um

#### **Bioenergetics**

*217 Bioenergetic Failure in Rat Oligodendrocyte Progenitor Cells Treated with Cerebrospinal Fluid Derived from Multiple Sclerosis Patients*

Deepali Mathur, Angela L. Riffo-Campos, Josefa Castillo, Jeffery D. Haines, Oscar G. Vidaurre, Fan Zhang, Francisco Coret-Ferrer, Patrizia Casaccia, Bonaventura Casanova and Gerardo Lopez-Rodas

*235 Lactate Shuttles in Neuroenergetics—Homeostasis, Allostasis and Beyond* Shayne Mason

#### **Growth Factors in Brain Function**


Dongxue Yang, Wenbo Zhang, Arshad Padhiar, Yao Yue, Yonghui Shi, Tiezheng Zheng, Kaspar Davis, Yu Zhang, Min Huang, Yuyuan Li and Li Sha

#### **Epigenetic Determinants**


#### **RNA Related Mechanisms**

*322 Changes in the Transcriptome of Human Astrocytes Accompanying Oxidative Stress-Induced Senescence*

Elizabeth P. Crowe, Ferit Tuzer, Brian D. Gregory, Greg Donahue, Sager J. Gosai, Justin Cohen, Yuk Y. Leung, Emre Yetkin, Raffaella Nativio, Li-San Wang, Christian Sell, Nancy M. Bonini, Shelley L. Berger, F. Brad Johnson and Claudio Torres


Chuanjun Zhuo, Weihong Hou, Lirong Hu, Chongguang Lin, Ce Chen and Xiaodong Lin

*357 Transcriptomics of Environmental Enrichment Reveals a Role for Retinoic Acid Signaling in Addiction*

Yafang Zhang, Fanping Kong, Elizabeth J. Crofton, Steven N. Dragosljvich, Mala Sinha, Dingge Li, Xiuzhen Fan, Shyny Koshy, Jonathan D. Hommel, Heidi M. Spratt, Bruce A. Luxon and Thomas A. Green

*374 The Many Faces of Elongator in Neurodevelopment and Disease* Marija Kojic and Brandon Wainwright

#### **Neurotrasmitters Systems**


#### **Ageing and Alzheimer Disease**

*418 Synaptic Dysfunction in Alzheimer's Disease and Glaucoma: From Common Degenerative Mechanisms Toward Neuroprotection*

Chiara Criscuolo, Carlotta Fabiani, Elisa Cerri and Luciano Domenici


Eva Martínez-Pinilla, Cristina Ordóñez, Eva del Valle, Ana Navarro and Jorge Tolivia

#### **Analysis of Complex Disorders**


#### **Synaptic Proteins Related Mechanisms**

*471 New Insights into the Crosstalk between NMDARs and Iron: Implications for Understanding Pathology of Neurological Diseases*

Huamin Xu, Hong Jiang and Junxia Xie

*481 Efficient Binding of the NOS1AP C-Terminus to the nNOS PDZ Pocket Requires the Concerted Action of the PDZ Ligand Motif, the Internal ExF Site and Structural Integrity of an Independent Element*

Li-Li Li, Katryna Cisek and Michael J. Courtney

## **Early Post-Natal Mechanisms**


Sean M. Riordan, Douglas C. Bittel, Jean-Baptiste Le Pichon, Silvia Gazzin, Claudio Tiribelli, Jon F. Watchko, Richard P. Wennberg and Steven M. Shapiro

#### **Neuronal Death**

*516 Association of Cell Adhesion Molecules Contactin-6 and Latrophilin-1 Regulates Neuronal Apoptosis*

Amila Zuko, Asami Oguro-Ando, Harm Post, Renske L. R. E. Taggenbrock, Roland E. van Dijk, A. F. Maarten Altelaar, Albert J. R. Heck, Alexander G. Petrenko, Bert van der Zwaag, Yasushi Shimoda, R. Jeroen Pasterkamp and J. Peter H. Burbach

*532 Immununochemical Markers of the Amyloid Cascade in the Hippocampus in Motor Neuron Diseases*

Ulises Gómez-Pinedo, Rocio N. Villar-Quiles, Lucia Galán, Jordi A. Matías-Guiu, Maria S. Benito-Martin, Antonio Guerrero-Sola, Teresa Moreno-Ramos and Jorge Matías-Guiu

# **Glial Cell Related Mechanisms**

#### *545 A Novel Approach for Amplification and Purification of Mouse Oligodendrocyte Progenitor Cells*

Junlin Yang, Xuejun Cheng, Jiaxi Shen, Binghua Xie, Xiaofeng Zhao, Zunyi Zhang, Qilin Cao, Ying Shen and Mengsheng Qiu

*555 Sustained HSP25 Expression Induces Clasmatodendrosis via ER Stress in the Rat Hippocampus*

Ji-Eun Kim, Hye-Won Hyun, Su-Ji Min and Tae-Cheon Kang

*570 Alterations in CD200-CD200R1 System during EAE Already Manifest at Presymptomatic Stages*

Tony Valente, Joan Serratosa, Unai Perpiñá, Josep Saura and Carme Solà

# Editorial: Biology of Brain Disorders

Daniela Tropea1, 2, 3 \* and Andrew Harkin3, 4 \*

*<sup>1</sup> Neuropsychiatric Genetics, Functional Genomics, Psychiatry, Trinity College, Dublin, Ireland, <sup>2</sup> Trinity Center for Health Sciences St. James Hospital, Dublin, Ireland, <sup>3</sup> Trinity College Institute of Neuroscience, Trinity College, Dublin, Ireland, <sup>4</sup> School of Pharmacy and Pharmaceutical Sciences, Trinity College, Dublin, Ireland*

Keywords: brain, disorders, neurodegenerative, neuropsychiatric, injury

**Editorial on the Research Topic**

#### **Biology of Brain Disorders**

Brain disorders are a major global healthcare problem. The scale of the problem is growing as the population ages and the numbers of people affected, directly or indirectly, is on the rise. The need for ongoing research to gain a greater understanding of the brain, structure, and function in health and disease, has never been more important. In the past, progress has been made largely by technological advances e.g., the clinico-pathological description of degenerative disorders followed by the discovery of chemical transmitters and introduction of pharmacological treatments. The introduction of analytical methods allowed for the identification of chemical transmitters affected providing insights into phenotypes of cells lost in degenerative diseases. More recent times have heralded the genetic analysis of causation leading to the recognition of disease genes for most brain disorders. The greater significance is that genetic analysis allowed scientists to access the endogenous origins of a disease. The current era of genomics in addition to analysis of the transcriptome, proteome, and metabolome allows for the study of all genes at all stages from initiation to death to provide a more complete picture. Genetic analysis has further allowed for the development of models to understand pathways and improved earlier diagnosis in presymptomatic stages. However, no treatment based on genetic analysis has yet to reach clinical utility although ongoing efforts are getting close to finding potential solutions such as amyloid therapies for Alzheimer's disease.

Edited and reviewed by: *Christian Hansel, University of Chicago, United States*

> \*Correspondence: *Daniela Tropea tropead@tcd.ie Andrew Harkin aharkin@tcd.ie*

Received: *24 October 2017* Accepted: *06 November 2017* Published: *21 November 2017*

#### Citation:

*Tropea D and Harkin A (2017) Editorial: Biology of Brain Disorders. Front. Cell. Neurosci. 11:366. doi: 10.3389/fncel.2017.00366*

Despite these advances in understanding brain disorders, significant challenges remain. The human brain is the most complicated machine we know. There are numerous cell types including neurons, astrocytes, microglia, and oligodendrocytes including further variations in the types of neurons and glia. There is a specificity for particular regions of the nervous system or cell type in different disorders. Why this is the case and how this comes about remain a mystery and is part of the challenge that scientists face. Furthermore, although there is clearly a genetic component to many brain disorders, it is equally clear that there are environmental risk or attenuating factors. This remains a critical issue in genetic forms of brain disorders when many mutations show incomplete penetrance. To address these complex challenges, there is no one best approach. Reductionism to the level of molecules and cells is inevitable but eventually scientists must return to the human brain at a systems level to determine neuronal networks, regional and whole brain structure and function. There is a reliance on cellular models, cell lines, primary cells, and stem cells and model organisms, from flies to rodents, to gain invasive and fundamental insights not possible in human experiments. There are a myriad of new technologies available heralding a next generation of progress ranging from advances in molecular biology, inducible pluripotent stem cells, genomic editing, optogenetics, and structural and functional MRI. MRI is of particular value as it one of a few tools that may be used both in the clinic and in animal experiments and can therefore provide a translational framework for the combination of model organisms with multimodal imaging and other modalities such as EEG, behavior and post-mortem cell and molecular analysis to elucidate mechanisms underlying imaging phenomena in patients with brain disorders (McIntosh et al.).

This collection of research articles represents a cross sectional sample of current approaches to study brain disorders. Submissions range from scientists working on genetic, epigenetic, and transcriptional aspects, on molecular and cellular mechanisms associated with neuronal plasticity and degeneration, on elucidating a role for brain glia, brain imaging in animals and humans in addition to translational and interventional research in clinical populations. There is a growing consensus that a multi-level approach is required to uncover mechanisms in brain function to provide pathways for the development of new treatments for brain disorders (Carcone and Ruocco).

In some cases genetic analyses across different brain disorders (schizophrenia, depression, anxiety, neurodegenerative disorders) point to common factors (Di Re et al.). In other instances, as revealed by genome wide association studies (GWAS), there are some diseases which are polygenic with multiple genes contributing to cumulative risk (Redenšek et al.). Some mechanisms and factors are common across diseases such as processes leading to synaptic dysfunction indicating that common protective strategies may be used (Criscudo et al.). Another example is given by the growth factor insulin-like growth factor 1 (IGF1) which is a common modulator of neurodevelopmental disorders and aging (Wrigley et al.), or molecular determinants of epigenetic controls: HDAC3 and HDAC4—enzymes controlling histone deacetylation—activate mechanisms for neurodegeneration, cognition, and brain injury (Wu et al.; Yang et al.).

The field of bioenergetics is emerging as common ground in the pathophysiology of several brain disorders: Mathur et al. show that gene pathways linked to carbohydrate metabolism are linked to the severity of multiple sclerosis. Lactate shuttles are viewed as operators for homeostasis in brain trauma and neurodegenerative diseases (Mason).

Brain plasticity, the brain's ability to adapt and capacity to repair itself following injury, is also a critical area of research at present. The rescue of brain tissue following ischemic damage (Alia et al.) and the improvement in function following rehabilitation after stroke (Constans et al.) are representative of this largely untapped area. There is a growing awareness of the importance of plasticity as a central mechanism underlying adaptive change associated with the pathophysiology and treatment of brain disorders.

Insight into mechanisms that underlie the onset and progression of brain disorders may help identify new avenues for therapeutic intervention. Emerging evidence that common deficits are present in several brain disorders point to mechanisms that generalize across disorders. Further progress will no doubt be related to technological advances in molecular screening and detection, the manipulation and study of neuronal circuits, tools for assessment of systems/regional connectivity, and whole brain function which may be used in both animal and human experiments and the analysis of big data drawn from multi-centered strategic partnerships in the development of projects of greater scale.

#### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

# ACKNOWLEDGMENTS

AH would like to acknowledge support from Marie Curie Initial Training Networks (ITN): FP7-PEOPLE-2012-ITN; Brain Imaging Return to Health "reBIRTH."

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Tropea and Harkin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Recent Advances in Translational Magnetic Resonance Imaging in Animal Models of Stress and Depression

Allison L. McIntosh<sup>1</sup> , Shane Gormley <sup>1</sup> , Leonardo Tozzi <sup>1</sup> , Thomas Frodl 1, 2 and Andrew Harkin1, 3 \*

1 Institute of Neuroscience, Trinity College Dublin, Dublin, Ireland, <sup>2</sup> Universitätsklinikum A.ö.R, Universitätsklinik für Psychiatrie und Psychotherapie, Medizinische Fakultät, Otto von Guericke Universität, Magdeburg, Germany, <sup>3</sup> School of Pharmacy and Pharmaceutical sciences, Trinity College Dublin, Dublin, Ireland

Magnetic resonance imaging (MRI) is a valuable translational tool that can be used to investigate alterations in brain structure and function in both patients and animal models of disease. Regional changes in brain structure, functional connectivity, and metabolite concentrations have been reported in depressed patients, giving insight into the networks and brain regions involved, however preclinical models are less well characterized. The development of more effective treatments depends upon animal models that best translate to the human condition and animal models may be exploited to assess the molecular and cellular alterations that accompany neuroimaging changes. Recent advances in preclinical imaging have facilitated significant developments within the field, particularly relating to high resolution structural imaging and resting-state functional imaging which are emerging techniques in clinical research. This review aims to bring together the current literature on preclinical neuroimaging in animal models of stress and depression, highlighting promising avenues of research toward understanding the pathological basis of this hugely prevalent disorder.

#### Edited by:

Edna Grünblatt, University of Zurich, Switzerland

#### Reviewed by:

Bechara John Saab, University of Zurich Hospital for Psychiatry, Switzerland Zoe Hughes, Pfizer Inc, United States

> \*Correspondence: Andrew Harkin aharkin@tcd.ie

Received: 13 February 2017 Accepted: 09 May 2017 Published: 24 May 2017

#### Citation:

McIntosh AL, Gormley S, Tozzi L, Frodl T and Harkin A (2017) Recent Advances in Translational Magnetic Resonance Imaging in Animal Models of Stress and Depression. Front. Cell. Neurosci. 11:150. doi: 10.3389/fncel.2017.00150 Keywords: depression, stress, animal models, MRI, neuroimaging

# INTRODUCTION

Major depressive disorder (MDD) is a chronic, heterogeneous disorder with a diverse range of risk factors including genetic predisposition, stress, early life experiences, and environmental factors. MDD often presents with co-morbid anxiety and is linked to several somatic disorders including pain, insomnia, cardiovascular disease, and gastritis. Current antidepressant treatments are inadequate with a delayed onset and a large proportion of patients being treatment resistant. The pathophysiology of MDD remains to be fully elucidated and the development of more effective therapeutics depends on a better understanding of the molecular underpinnings of the disorder. Magnetic resonance imaging (MRI) is a safe, non-invasive technique that can provide valuable information on structural and functional changes in the human brain. MRI is commonly used in patients with MDD to assess changes in brain structure and function relating to symptom severity and treatment response. Alterations in the morphology of brain structures, particularly the hippocampus, have been frequently reported, in addition to differences in microstructural integrity, network connectivity, cerebral blood flow, and metabolite concentrations (Sacher et al., 2012). At present however, the pathophysiology underlying these neuroimaging findings remain unclear which significantly limits the development of novel therapeutic strategies. Preclinical studies in animal models of disease are vital to investigate the molecular and cellular changes that may underpin the neuroimaging hallmarks in MDD. The translation of preclinical research to the clinical setting is often challenging, however the ability to investigate comparable MRI modalities in both humans and animals is particularly valuable. The characterization of animal models of depression by MRI has already provided important insights into potential environmental factors associated with, and physiological mechanisms which may underlie the observed changes in brain structure and function in depressed patients. In addition, investigating neuroimaging markers in animal models allows the examination of molecular and cellular changes post-mortem. Furthermore, MRI in animal models may also provide valuable biomarkers that can be used clinically for treatment response and patient stratification. The advantages and challenges of preclinical neuroimaging as well as methodological concepts have previously been reviewed elsewhere (Hoyer et al., 2014; Jonckers et al., 2015) however a review of the preclinical neuroimaging findings relating to stress and depression is currently lacking. This review aims to address this by discussing the current MRI data in animal models of stress and depression, covering all MRI modalities including structural data, resting-state fMRI, manganese-enhanced MRI, cerebral blood perfusion measurements, and magnetic resonance spectroscopy (MRS). A comparison of the neuroimaging findings in animal models has also been summarized in **Table 1**.

#### ANIMAL MODELS OF DEPRESSION

Animal models of depression are important to screen novel compounds for antidepressant action, and to investigate neurochemical and physiological processes that may underlie depressive behaviors. Modeling depressive disorders in rodents is challenging, particularly as many of the symptoms cannot be convincingly determined in rodents. Furthermore, as the pathophysiological processes and genetic risk factors underlying the human disorder and the mechanism of antidepressant action are currently not well understood, developing valid models is problematic. Animal models are therefore unlikely to mirror the full extent of the disorder, however they can be validated through three sets of criteria: face, construct, and predictive validity. The model should therefore reflect features of the human condition (face validity), have some relevance to the etiological basis of the disorder (construct validity) and finally respond similarly to treatments (predictive validity). Animal models can therefore be a useful tool to investigate the underlying pathophysiology or potential of novel therapeutics.

Several techniques have been used to induce depressive-like symptoms in animals including acute exposure to stressors as well as more chronic interventions that reflect aspects of depression etiology such as stress or genetic manipulation. The forced swim test (FST) is a paradigm based on the observation that animals who are faced with an inescapable stressor will desist from attempting to escape after a period of time. Once a rodent is placed in a large container of water from which they cannot escape they will initially vigorously attempt to escape, but will then cease escape orientated behaviors and the immobility seen in the FST has been reported to reflect "behavioral despair." Antidepressant treatments have been shown to reduce immobility time and therefore the FST has been used as a rapid screen for antidepressant activity in novel compounds. The FST test however has very little construct validity for depression as it does not have any etiological relevance to the disorder. Therefore, more relevant interventions have been developed to induce a depressive-like phenotype in animals and the FST is used as a behavioral measure in these. A well characterized model for depression is the olfactory bulbectomized (OB) rat which involves bilateral removal of the olfactory bulbs, resulting in retrograde degeneration of neurons that project from the olfactory bulbs and loss of efferent connections to sub-cortical brain structures. This, combined with loss of olfaction which may constitute a profound stressor, produces a depressive-like and highly agitated behavioral phenotype which is associated with alterations in endocrine and neurotransmitter levels similar to those reported in depressed human patients (for review see, Kelly et al., 1997; Harkin et al., 2003; Song and Leonard, 2005). The OB model also has good predictive validity as OB-induced hyperactivity in the open field, indicative of a highly anxiety-like phenotype, is sensitive to chronic, but not acute antidepressant treatment. Chronic mild stress (CMS) is also commonly used to induce depressive-like phenotype in animals and has better construct validity than the OB rat. Although the exact protocol can vary, in general CMS involves exposing rodents to several stressors of different intensity and duration over a period of several days and induces a robust depressive-like behavioral phenotype. Several inbred rat strains have been proposed as models of genetic vulnerability to depression, including the Wistar-Kyoto (WKY) rat and also the congenital learned helpless (cLH) rat. Originally used as a normotensive control for the spontaneously hypertensive rat (SHR), the Wistar-Kyoto rat strain endogenously expresses some of the behavioral, endocrine, and neurotransmitter changes found in depressed human patients (Redei et al., 2001). Similarly, the cLH rats demonstrate congenital learned helpless behavior and anhedonia, predictive validity with antidepressant treatments and biochemical changes including alterations in hippocampal glutamatergic systems. Genetic manipulations have also been used to investigate the alterations and/or underlying mechanisms associated with stress, anxiety, and depressive-like symptoms. Given the fact that serotoninergic dysfunction is implicated in stress and depression with many of the current antidepressants modulating serotonin activity, studies have used genetic knockouts of serotonin receptors to further elucidate the effects of serotonin dysregulation (Savitz et al., 2009). The serotonin 5-HT1A receptor knockout (5-HT1A-R KO) mouse, although not considered a model of depression, is one such manipulation used to investigate the neural circuits affected by altered serotonin transmission.


(Continued)


TABLE

1


Continued

imaging; SCx, somatosensory

 cortex; SN, substantia nigra; VCx, visual cortex; vHipp, ventral hippocampus;

 VTA, ventral tegmental area.

All of these models demonstrate some face, construct, and predictive validity for depression and using translational approaches such as MRI may serve to increase our understanding of the physiological basis of the functional and structural brain alterations observed in depressed patients by allowing examination of molecular and cellular alterations post-mortem.

#### MRI IN ANIMAL MODELS OF DEPRESSION

#### Structural Changes Volumetric MRI

Recent meta-analyses of structural imaging studies in the clinical literature have detected volumetric changes in a number of brain regions in MDD relative to controls. The most robust finding in the literature is reduced hippocampal volume in patients (Campbell et al., 2004; Videbech, 2004; Kempton et al., 2011; Arnone et al., 2012; Schmaal et al., 2016), but some reports have highlighted a decrease in the volume of prefrontal, dorsomedialprefrontal, orbitofrontal and cingulate cortices, and striatum (Arnone et al., 2012; Bora et al., 2012; Sacher et al., 2012; Lai, 2013) as well as increased volume in the lateral ventricles (Kempton et al., 2011).

Although hippocampal volume loss might be dependent on patient age and disease duration (McKinnon et al., 2009; Schmaal et al., 2016), a meta-analysis of studies in first episode patients revealed significant hippocampal volume reductions, suggesting that smaller hippocampal volume may be a possible risk factor for depression, rather than a marker of disease progression (Cole et al., 2011). This is in agreement with a 3-year longitudinal study, which reported no significant reduction in hippocampal volume of patients, but did show that smaller baseline volume was associated with poorer clinical outcome (Frodl et al., 2008a). It should be noted that, following a whole brain voxel based morphometry analysis, a decrease in gray matter density in the hippocampus in patients with ongoing depression over a 3 year period was detected, thus highlighting a potential impact of the method of analysis on the findings (Frodl et al., 2008b).

Changes in hippocampal volume also appear to be sensitive to antidepressant treatment, showing an increase in volume following traditional antidepressant (Frodl et al., 2008a) and electroconvulsive (ECT; Abbott et al., 2014) therapies. Fu et al. (2013) suggested that lower right hippocampal volume could even be a predictor of poor antidepressant response. Reduced subgenual anterior cingulate cortex volume persists despite successful treatment with antidepressant drugs (Drevets et al., 1997) but chronic lithium treatment, which exerts robust neurotrophic effects in animal models, has been associated with an increase in gray mater volume in treatment responders in this and other prefrontal areas (Drevets et al., 2008; Moore et al., 2009).

Interestingly, smaller hippocampi in depressed patients have been found to be associated with increased markers of glucocorticoid receptor activation in peripheral plasma, suggesting that gray matter loss in this structure might be related to changes in HPA axis function (Frodl et al., 2012).

Translational neuroimaging studies have reported structural alterations in the rodent brain in animal models of depression. The Wistar-kyoto (WKY) rat is an inbred rat strain with an inherent anxiety- and depressive-like behavioral phenotype. In vivo MRI has shown increased lateral ventricular volume and decreased hippocampal volume in adult WKY rats compared to the Wistar comparison strain with no change in total cortical volume (Gormley et al., 2016). Alterations in T1 and T2 relaxation times can provide information about tissue characteristics and increased T2 relaxation times were evident in the hippocampus of WKY rats compared to Wistar controls (Gormley et al., 2016), a finding that may be linked to oedema or changes in vascular permeability. Work in our group has also investigated MRI changes in the olfactory bulbectomized rat (OB) and showed opposite changes with reductions in T2 relaxation times in the hippocampus, visual cortex and left retrosplenial cortex (Gigliucci et al., 2014). This highlights the heterogeneity of animal models and may indicate alterations in relaxation times could be linked to specific phenotypes. Furthermore, this emphasizes the importance of selecting an appropriate animal model as the OB rat may be more related to anxiety rather than depressive-like behaviors. Structural analysis revealed a trend toward increased ventricular volume in OB rats however no changes in hippocampal volume were observed (Gigliucci et al., 2014). Reductions in T2 relaxation times have been linked to microglial activation and also increased cell packing density in the brain parenchyma (Ding et al., 2008; Blau et al., 2012). Given the lack of overt inflammation, it is most likely that changes in cell density or morphology may account for the decreased T2 relaxation times reported in the OB rats.

Reduced hippocampal volume has also been linked to anxietylike behavior, with mice bred for high anxiety showing reduced hippocampal volume compared to those bred for low anxiety traits (Kalisch et al., 2006). The structural changes may also be linked to the depressive-like behavioral phenotype observed in these animals, however the authors show that in control animals, hippocampal volume positively correlated with anxiety-like but not depressive-like behavior and therefore suggest the differences in hippocampal volume are primarily linked to anxiety-like traits.

Chronic stress paradigms have also been shown to produce changes in rodent brain morphology. Structural MRI of post-mortem brains of Wistar rats that were subjected to a 10 day immobilization stress, have shown an increase in lateral ventricular volume without any change in hippocampal volume or morphology when compared to non-stressed controls (Henckens et al., 2015). Delgado y Palacios et al. (2011) also report no change in hippocampal volume in rats exposed to 8 weeks of CMS, although subtle changes in hippocampal morphology were evident. In contrast, Lee et al. (2009)report that 21 days of immobilization stress in rats results in a 3% reduction in baseline hippocampal volume, assessed by longitudinal MRI, a change which was not observed in control animals. There was no change observed in anterior cingulate or retrosplenial cortical volumes reported (Lee et al., 2009). The contradictory reports relating to hippocampal volume may be related to stressor severity and/or the ex vivo acquisition of structural data. Ex vivo MRI on post-mortem brains following 10 days of social defeat in mice shows that the volume of the ventral tegmental area (VTA), bed nucleus of the stria terminalis (BNST) and the dorsal raphe nucleus (DRN) positively correlate with social avoidance scores in animals, indicating that these areas may be linked to stress vulnerability (Anacker et al., 2016). Social avoidance scores also negatively correlated with size of both the cingulate cortex and nucleus accumbens suggesting decreased volume in stresssensitive mice. Interestingly DTI metrics largely correlated in opposite directions to volumetric changes indicating increased volume is accompanied by reduced diffusion in the brain regions investigated.

Volumetric alterations have also been associated with antidepressant effects in offspring following administration of a ketogenic (high ketone) diet (KD) during pregnancy. Whilst small structural changes were observed at weaning (PND 21.5), KD offspring showed increased frontal cortical and cerebellar volume, but interestingly, decreased volume of the hippocampus, striatum, and some cortical areas in adulthood (Sussman et al., 2015). The observed decreased hippocampal volume is somewhat counterintuitive to an antidepressant effect given that reductions in hippocampal volume are associated with depressive symptoms. The authors suggest this may be due to reduced dendritic branching or abnormal neuronal differentiation resulting from low protein consumption however as this is a single study examining MRI related KD effects, further work is required to comprehensively assess the long-term effects of KD diet on behavior and brain structure.

Together these data suggest that increased lateral ventricular volume and decreased hippocampal volume are relatively consistent hallmarks in both patients with MDD and animal models, adding to their face validity. Although contrasting findings may be due to study design, analysis method or method of acquisition they could also reflect subtypes of stress/depression and the diversity of patient populations that may be modeled in animals.

#### Diffusion Imaging

Diffusion tensor imaging (DTI) allows for an assessment of in vivo white matter microstructural integrity through measurement of the restriction in diffusion of water molecules in brain tissue (Moser et al., 2009). Several DTI studies in patients with unipolar depression have reported decreased cortical fractional anisotropy (FA), a measure of white matter integrity and fiber directionality in the frontal and temporal lobes (for review see Sexton et al., 2009) as well as in the amygdala (Arnold et al., 2012). This seems to be evident in both late life depression (Bae et al., 2006; Nobuhara et al., 2006) and in young adults (Li et al., 2007). Data suggests that altered FA values may be a marker of pre-disposition to depression or may occur early in the course of disease, as patients with first onset depression have been shown to exhibit decreased FA values in frontal cortical regions such as the anterior cingulate cortex (Zhu et al., 2011) and the frontal gyrus (Ma et al., 2007).

Currently, tractography studies show that first episode depressed patients exhibit increased cortico-limbic structural connectivity (Fang et al., 2012) and that depression is associated with a disruption in the pathways linking subcortical regions, including the hippocampus, to the prefrontal cortex (Liao et al., 2013) as well as alterations in the superior longitudinal and fronto-occipital fascicule (Murphy and Frodl, 2011). Finally, preliminary results from the largest study to date have also highlighted alterations in the corpus callosum of MDD subjects (Kelly et al., 2016).

DTI, based on Gaussian distribution of water molecules, and more recently diffusion kurtosis imaging (DKI) which attempts to account for biological variation in water diffusion and provide a more accurate model of diffusion, have both identified alterations in microstructural integrity in animal models of depression. Common metrics reported in DTI and DKI studies are mean diffusion or kurtosis across all directions (MD or MK), radial diffusion or kurtosis (RD or RK; movement in the transverse direction) and axial diffusion or kurtosis (AD or AK; movement along the main axis of diffusion). Although, interpretation of diffusion imaging has not been fully explored, microstructural organization of the brain, including cellular membranes, myelinated axons, intracellular organelles, and dendritic architecture, affects the orientation of water diffusion.

DTI and DKI studies have identified alterations in the microstructural integrity of gray matter regions in rats exposed to CMS. Increased mean diffusion values in the bilateral frontal cortex, hippocampus, hypothalamus, caudate putamen and corpus callosum were reported in rats exposed to CMS compared to control animals, which may relate to reduced membrane density or tissue degeneration (Hemanth Kumar et al., 2014). This was accompanied by increased radial and axial diffusion in the frontal cortex and increased AD in the hypothalamus. It is suggested that increased RD may be due to demyelination while the AD alterations may be attributed to axonal loss. Conversely, decreased mean and radial kurtosis (magnitude of restrictions perpendicular to the principle direction of diffusion) has been reported in the hippocampus of animals exposed to CMS (Delgado y Palacios et al., 2011) irrespective of sensitivity to the stress protocol. Further work by the same group expanded the neuroimaging markers relating to stress sensitivity and resilience. Specifically, increased axial diffusion in the caudate putamen (CPu) demonstrates microstructural alterations within the CPu (Delgado y Palacios et al., 2014). Furthermore, increases in radial diffusion were observed in the amygdala (Delgado y Palacios et al., 2011) as well as frontal cortex and hypothalamus (Hemanth Kumar et al., 2014) of CMS animals. This may reflect dendritic atrophy, augmented arborization or demyelination as alterations in hippocampal kurtosis parameters were accompanied by decreased staining of microtubule associated protein 2, which is involved in stabilization of neuronal cytoarchitecture (Delgado y Palacios et al., 2011). In addition, by subdividing the CMS group into those sensitive or resilient to CMS Delgado et al. showed decreased mean kurtosis values in the caudate putamen of animals exhibiting stress-induced behavioral anhedonia compared to stress-resilient counterparts. Such alterations in diffusion metrics may be attributed to differences in dendritic arborization or potential alterations in microglial morphology due to activation states. Although, further studies are required to elucidate the molecular and cellular mechanisms underlying these microstructural changes, these data highlight the CPu as a region altered by stress and potentially valuable in identifying individuals vulnerable to stress. Furthermore, this highlights the importance of analyzing cohorts based on their behavioral responses to stressors as failure to identify possible "resilient" subgroups may result in misleading conclusions.

The effect of stress on tissue architecture in both control and stress sensitive WKY rats has also been investigated. WKY rats show decreased FA in the corpus callosum and anterior commissures and increased mean diffusion in both the fornix and corpus callosum compared to the Wistar comparator strain (Zalsman et al., 2016). Furthermore, exposing WKY rats to early-life stress produced significantly higher mean diffusion values compared to control in the amygdala, cingulate cortex and external capsule. Conversely, early stress decreased mean diffusion in the cingulate cortex, amygdala, and external capsule in Wistar rats (Zalsman et al., 2015). This suggests that stress alters the structural integrity of brain areas linked to emotional regulation differently in animals with genetic vulnerability.

DTI studies in depressed patients have shown decreased fractional anisotropy in the frontal cortex, cingulate and corpus callosum, changes which are mirrored in the preclinical research supporting the translational relevance of these measures. Interestingly, diffusion kurtosis imaging is not widely used in clinical neuroimaging, however preclinical results from CMS studies suggest kurtosis parameters, particularly in the caudate putamen, may be useful in identifying susceptible from resilient phenotypes and highlights the importance of animal models.

#### Functional MRI

#### Magnetic Resonance Spectroscopy

Throughout the years, the investigation of the biochemical alterations associated with MDD has highlighted changes in neurotransmitter systems of depressed patients (Crow et al., 1984; Cheetham et al., 1989; Yildiz-Yesiloglu and Ankerst, 2006; Rajkowska and Stockmeier, 2013). In particular, proton MRS is a non-invasive technique that has been often used to detect differences in tissue concentration of various chemical compounds in vivo (Jansen et al., 2006). Even on routinely available 1.5 T scanners, this technique can identify Nacetylaspartate (NAA), choline (Cho), and creatine (Cr). At higher field strengths (3-7 T), glutamate (Glu), glutamine (Gln) and myoinositol (Ins), can also be individually resolved. Finally, γ-Aminobutyric acid (GABA) as well as glutathione (GSH) are optimally quantified at the highest field strengths with special editing and acquisition techniques (Bustillo, 2013; Rae, 2014), although GSH has been shown to be reliably measured at lower fields using fitting routines such as LCModel after acquisition (Terpstra et al., 2003).

MRS studies have often reported a reduction of glutamate, the most common excitatory neurotransmitter, in the prefrontal cortex of MDD compared to controls (Yildiz-Yesiloglu and Ankerst, 2006; Luykx et al., 2012; Miladinovic et al., 2015). Glutamine (Gln), an amino acid synthetized from glutamate in the glia (Martinez-Hernandez et al., 1977), and GABA have also been found to be decreased in MDD in this region (Walter et al., 2009; Pehrson and Sanchez, 2015). Glutathione, which is involved in synthesis and degradation of proteins and DNA (Maher, 2005), has recently been found to be reduced in the occipital cortex of depressed patients, suggesting reduced anti-oxidative capacity in this cohort (Godlewska et al., 2015). Choline, which is related to membrane metabolism and turnover, has been found to correlate with mood in healthy subjects (Jung et al., 2002) and to be increased in MDD patients compared to healthy controls in the basal ganglia (Yildiz-Yesiloglu and Ankerst, 2006). Most studies have not shown any difference in NAA levels in depressed patients (Yildiz-Yesiloglu and Ankerst, 2006) and inconsistent results for Cr levels (Venkatraman et al., 2009). Thus, the most consistent findings in MDD are decreased excitatory and inhibitory neurotransmitters as well as glutathione, and elevated Ch levels in some regions, indicative of increased membrane phospholipid turnover.

The low signal to noise ratio and large voxel size required has hindered the progression of preclinical MRS studies, however with greater sensitivity achieved with magnetic fields as high as 11.7T currently available, several recent studies have shown alterations in a range of metabolites in animal models of depression.

Exposing the offspring of dams subjected to CMS to an acute stressor in adolescence (a single forced swim test) decreased glutamate MR signal in the right hippocampus (Huang et al., 2016). Similarly, chronic exposure to the FST has been reported to decrease the glutamate signal in both the prefrontal cortex and hippocampus in rats (Li et al., 2008). Reductions in the combined glutamate and glutamine signal have also been reported in the prefrontal cortex of mice exposed to CMS (Hemanth Kumar et al., 2012). In contrast, Delgado y Palacios et al. (2011) report increased hippocampal glutamate levels in animals susceptible to CMS compared to both resilient and control groups. These conflicting changes in hippocampal glutamate may be due to the type of stressor used (acute compared to chronic) or regional differences within the hippocampus. Delgado et al. positioned a 2 mm<sup>3</sup> voxel in the ventral part of the left hippocampus whereas Huang et al. (2016) positioned their 2.5 × 4 × 4 mm voxel in the dorsal area of the right hippocampus. Similar decreases in dorsal hippocampal glutamate have been reported in rats bred for learned helplessness (cLH) compared to Sprague-Dawley controls and also following CMS (Hemanth Kumar et al., 2012; Schulz et al., 2013).

Schulz et al. also report reductions in choline in both the hippocampus and frontal cortex (encompassing the cingulated and motor cortex) of cLH rats (Schulz et al., 2013). Conversely, a recent study reported increased choline levels in the prefrontal cortex following chronic psychosocial stress (Grandjean et al., 2014), which may reflect differences in preclinical models used (genetic vs. stress) or regional differences between the frontal and prefrontal cortex. Alterations in choline levels have also been shown by Han et al. with increased levels in both the hippocampus and amygdala following a single prolonged restraint stress (Han et al., 2015). Choline has also been shown to be sensitive to antidepressant treatments as Sartorius et al. (2003) showed electroconvulsive stimulation (ECS), the preclinical model of ECT, increased choline levels in the hippocampus of normal animals. Similarly, ECS increased the absolute concentrations of choline in the hippocampus of cLH rats but in contrast to the study by Sartorius et al., there was no effect in wild type control animals (Biedermann et al., 2012). This may be due to normalization of the data or comparison to a baseline scan in Sprague-Dawley control rats. Interestingly, choline MRS signal was also increased in the prefrontal cortex of patients treated with the SSRI paroxetine (Zhang et al., 2015).

Previous studies have examined the effect of early life adversity in the form of maternal separation (MS) on metabolite concentrations, however no difference between MS animals and controls was observed however both environmental enrichment and escitalopram treatment increased NAA levels in the (Hui et al., 2010, 2011).

Myoinositol (mI) levels can be indicative of glial cell function or proliferation and it has been considered a proxy marker of inflammation (Kousi et al., 2013). Preclinical MRS studies have shown increased hippocampal myoinositol levels following both CMS (Hemanth Kumar et al., 2012) and the acute FST (Kim et al., 2010). Similarly, increased mI levels have been observed in the amygdala following chronic psychosocial stress (Grandjean et al., 2016). Together, these data suggest a possible link between glial function and depression pathology.

Increased levels of GABA have been reported in the hippocampus of CMS-offspring exposed to an acute stressor (Huang et al., 2016). Ex vivo <sup>1</sup>H-MRS studies support elevated GABA levels following CMS and restraint stress with increases observed in the anterior cingulate and prefrontal cortex, respectively (Perrine et al., 2014; Drouet et al., 2015) suggesting that altered inhibitory neurotransmission may be involved in the pathophysiology of CMS.

Alterations in the N-acetyl aspartic acid (NAA), a marker commonly used to indicate neuronal integrity have also been reported, with decreased hippocampal NAA levels following both mild and unpredictable chronic stress paradigms (Xi et al., 2011; Hemanth Kumar et al., 2012) as well as chronic exposure to the FST (Li et al., 2008). Xi et al. (2011)report that hippocampal NAA reductions were normalized by chronic escitalopram treatment (4 weeks). Conversely, Han et al. (2015) showed increased NAA in both the amygdala and hippocampus following a single prolonged stress. This may be due to single vs. repeated stress protocols used. In addition, MRS studies in three mouse lines bred for HPA axis stress reactivity have revealed decreased Nacetylaspartate (NAA) levels in the dorsal hippocampus and prefrontal cortex which is coupled with decreased levels of performance in hippocampal dependent spatial memory tasks when compared with mice with low HPA axis stress reactivity (Knapman et al., 2012).

It is worth noting that an advantage of animal models is that metabolite concentrations in can also be assessed through invasive or post-mortem methods such as in vivo microdialysis or ex vivo assessment of tissue concentrations. These techniques could be used to compliment or validate in vivo MRS findings.

Results from both preclinical and clinical MRS studies implicate altered glutamatergic signaling in depression with studies in both depressed patients and animal models reporting decreased glutamate and/or glutamine signal, particularly in the prefrontal cortex (Yildiz-Yesiloglu and Ankerst, 2006; Li et al., 2008; Walter et al., 2009; Hemanth Kumar et al., 2012). Clinical MRS results for other metabolites are inconsistent and depend upon subregion investigated, disease duration, depression subtype, and medication (Yildiz-Yesiloglu and Ankerst, 2006) however some studies report increased choline signal in the frontal cortex (Steingard et al., 2000; Farchione et al., 2002), which is also shown in animal models (Han et al., 2015; Grandjean et al., 2016). Interestingly, reductions in the hippocampal glutamate signal differentiated susceptible from resilient animals when exposed to CMS (Delgado y Palacios et al., 2011) and similar alterations have been shown in depressed patients (Block et al., 2009). Alterations in certain metabolite concentrations in preclinical models could be validated through in vivo microdialysis or ex vivo assessment of tissue concentrations.

#### Resting-State Functional MRI (rs-fMRI)

Functional MRI (fMRI) provides a tool to map regional changes in cerebral blood flow at resting state, i.e., in the absence of task execution. Depressed patients have been reported to have decreased dorsolateral prefrontal cortex activation (Alcaro et al., 2010) as well as increased baseline activity in the orbitofrontal and subgenual anterior cingulate cortices (for review see Hasler and Northoff, 2011) relative to controls. Increased resting state perfusion as assessed via arterial spin labeling (ASL) in the subgenual anterior cingulate cortex has also been shown specifically in treatment resistant depressed patients relative to controls (Duhameau et al., 2010). These alterations are sensitive to treatment: resting state perfusion as assessed via ASL in the anterior cingulate cortex, for example, was reduced (Clark et al., 2006) and dorsomedial prefrontal cortical activity was increased following chronic venlafaxine and fluoxetine treatment in depressed patients (Savitz and Drevets, 2009). Hence, unlike some of the structural alterations discussed above, brain functional alterations appear to be state dependent markers of MDD. This is potentially unsurprising as quite modest changes in plasticity may induce marked changes in blood flow or neuronal activity however alterations in brain structure at a resolution detectable would be harder to induce.

The default mode network is a set of brain regions which are more active at rest than during task execution, which include the orbital frontal cortex, the medial prefrontal/anterior cingulate cortex, the lateral temporal cortex, the inferior parietal lobe, the posterior cingulate and retrosplenial cortex, the hippocampus and parahippocampal cortex (Lu et al., 2012). Hyperactivity in the default mode network during task performance in depressed patients has been reported and interpreted as a possible MRI marker for rumination (for review see Whitfield-Gabrieli and Ford, 2012). Hyper connectivity between the subgenual anterior cingulate and default mode network at rest has also been observed in MDD (Greicius et al., 2007; Berman et al., 2011), and increased default mode network connectivity with the dorsolateral prefrontal cortex are thought to underlie some aspects of emotional dysregulation characteristic of the disorder (Sheline et al., 2010).

Preclinical studies have begun to employ fMRI to investigate regional activity and functional connectivity in vivo and extraction of the blood oxygen level dependent (BOLD) signal is commonly used as surrogate marker of neuronal activation. Recent studies have shown altered BOLD signal intensity in several brain regions involved in mood regulation following exposure to early-life adversity or stress. Hui et al. (2010) recently showed that animals subjected to the early life maternal separation protocol exhibited increased BOLD signal in the insular lobe, hypothalamus, limbic system, hippocampus, and frontal lobe in adulthood. Similarly, increased BOLD activation was observed in the hippocampus, limbic system and temporal lobe of rats exposed to CMS.

Recent advances in preclinical neuroimaging have allowed the identification of similar resting-state networks in rodents and humans (Lu et al., 2012; Sierakowiak et al., 2015; Zerbi et al., 2015; Gozzi and Schwarz, 2016) and the analysis of these networks in disease models.

Grandjean et al. have shown that chronic psychosocial stress increased within-network functional connectivity in the cingulate cortex and sensory cortical networks, specifically the supplementary, barrel field 1 and 2, and visual cortices in mice (Grandjean et al., 2016). In addition there was also increased between-network functional connectivity between the prefrontal cortex and both the amygdala and piriform cortex; ventral hippocampus and the amygdala and between the cingulate cortex and both the piriform cortex and the amygdala following psychosocial stress. Rodents have also been shown to have a default mode network similar to that seen in humans (Lu et al., 2012) and 10 days of immobilization stress has been reported to increase resting state functional connectivity within this network in addition to the visual and somatosensory networks (Henckens et al., 2015). Altered functional connectivity has also been shown in the congenital learned helplessness rat with increased correlation between activity in the DRN and the somatosensory cortex, orbital cortex, frontal cortex, and caudate-putamen. Enhanced correlations were also observed between the ventral hippocampus and the retrosplenial cortex and caudate putamen and also between the retrosplenial cortex and area 1 of the cingulate cortex (Gass et al., 2014a). Interestingly hippocampalcingulate connectivity has been positively correlated with depression severity in patients, drawing parallels with the increased hippocampal-retrosplenial connectivity reported in cLH rats. By contrast, there are reductions in inter-hemispheric connectivity, particularly in the sensory, motor, cingulated and infralimbic cortices, the nucleus accumbens and the raphe nucleus of cLH rats (Ben-Shimol et al., 2015). Similarly, the WKY rat has shown increased functional connectivity between the hippocampus and the left frontal association cortex/dorsolateral orbital cortex in a "more immobile" cohort compared to "less immobile" counterparts (Williams et al., 2014). In contrast, hypoconnectivity was observed with the hippocampus and the left somatosensory cortex, the left ventral striatum partially inclusive of the nucleus accumbens core, the bilateral cingulate cortex, the bilateral lateral septum and the left caudate.

A study investigating resting state fMRI in serotonin receptor 1A knockout mice showed reduced functional connectivity between cortical areas (including prefrontal, retrosplenial, and entorhinal cortices) and the hippocampus (mainly CA1 and dentate gyrus, DG). Similarly, reduced functional correlations were observed between the dorsomedial thalamus and both the cortex and hippocampus (Razoux et al., 2013).

Alterations in functional connectivity reported in animal models of depression are displayed in **Figure 1**. Studies in both depressed patients and animal models of depression consistently report hyperactivity within the default mode network, comprised of the orbital, prelimbic, auditory/temporal association, parietal, and retrosplenial cortices and the dorsal hippocampus in the rat. In addition, studies in depressed patients have also shown hyperactivity between the cingulate cortex and the amygdala, parts of the affective network, which is also reported in the rodent literature. Interestingly, sub-anaesthetic doses of ketamine increased functional connectivity between frontocortical regions and also the prefrontal cortex and the hippocampus in both rats and humans (Grimm et al., 2015; Becker et al., 2016) suggesting that resting state fMRI may be a valuable translational tool for investigating the actions of novel antidepressants.

#### Manganese-Enhanced MRI

Manganese-enhanced MRI (MEMRI) is an emerging neuroimaging technique that exploits the capability of manganese (Mn) ions to enter excitable cells through voltagegated calcium channels. As Mn ions decrease water longitudinal relaxation time this uptake can be visualized by MRI and as such MEMRI is being increasingly used in preclinical studies to investigate alterations in functional networks. Daducci et al. (2014) showed that chronic administration of interferon alpha to rats, commonly used to model chemotherapy-induced depression, reduced the pituitary volume and also reduced activation in cortical areas, particularly the visual and sensory cortices. Reduced MEMRI signal following manganese injection into the raphe nucleus has been shown in rats exposed to CMS in the substantia nigra, hippocampus, entorhinal cortex and the insular cortex, while increased signal was observed in the medial septal nucleus compared to non-stressed control animals (Gordon and Goelman, 2016) suggesting impaired serotonergic connectivity between the raphe and the substantia nigra and hippocampal areas.

#### Cerebral Perfusion

High resolution MRI has revealed discrete reductions in regional cerebral blood volume (rCBV) in the left habenula, particularly the lateral area, and increased rCBV in the bed nucleus of the stria terminalis (BNST) in the congenital learned helplessness (cLH) model of depression (Gass et al., 2014a). Altered cerebral blood perfusion in preclinical models of depression has also been measured by bolus-tracking arterial spin labeling (bt-ASL). This non-invasive method directly assesses blood perfusion and provides quantitative measures relating to blood flow. Decreased cerebral perfusion was observed in the striatum and pre-limbic cortex of WKY rats compared to Wistar controls and interestingly, cerebral perfusion correlated with GFAPpositive cell number such that increased GFAP expression corresponded with increased cerebral perfusion (Gormley et al., 2016). In contrast no alterations in regional cerebral perfusion were reported in the OB rat model of depression (Gigliucci et al., 2014), again highlighting that many MR markers associated with depressed phenotypes may be "model" or patient specific.

#### PharmacofMRI

Pharmacological MRI is becoming a more widely used tool in both preclinical and clinical drug research (Jonckers et al., 2015). Insights gained from clinical MRI studies are frequently confounded by illness, chronicity and medication use when attempting to determine if antidepressant drug treatment may influence gray matter volumes or functional MR parameters. PharmacoMRI coupled with in vivo structural MRI in animal models represents an approach which allows for effects of acute and chronic drug treatment and subsequent withdrawal, with clinically relevant dosing to be determined on brain structure and function. Findings may then be further evaluated and confirmed in the post-mortem brain using stereological histochemistry. Vernon et al. (2012) reported that chronic lithium treatment in rats induced an increase in whole-brain volume and cortical gray matter without a significant effect on striatal volume. Increased total brain volumes were subsequently confirmed post-mortem. Comparisons with the antipsychotic haloperidol indicated that the distribution of changes was topographically distinct.

Reductions in brain activation are observed with phMRI following acute treatment with the serotonin (5-HT) reuptake inhibitor (SSRI) fluoxetine (10 mg/kg oral or 5 mg/kg i.v.) in rats (Bouet et al., 2012). Subsequently Klomp et al. (2012) employed pharmacoMRI to assess the effects of chronic treatment with fluoxetine (5 mg/kg orally for 3 weeks) in juvenile (post natal day 25) and adult rats (post natal day 65) following a 1 week washout, using an acute fluoxetine challenge (5 mg/kg i.v.) to trigger the serotonergic system. A significant age by treatment interaction was observed in several subcortical brain regions related to 5-HT neurotransmission, suggestive of differential neuronal effects of SSRI treatment in juveniles that may underlie emotional disturbances seen in adolescents treated with fluoxetine. Harris and Reynell (2016), using data derived from animal studies, consider the underlying mechanisms of antidepressant treatment-related changes in BOLD including alterations in neurovascular coupling and brain energetics and metabolism.

To aid interpretation of spatially distributed activation patterns in response to pharmacological stimuli, Bruns et al. (2015) have proposed a set of multivariate metrics termed domain gauges which are calibrated based on different classes of reference drugs including antidepressants. The profile provides quantitative activation patterns with high biological plausibility with the potential to be developed as a valuable analytical tool for interpretation and decision making in drug development.

A number of recent investigations have assessed the effects of sub-anaesthetic doses of ketamine on intrinsic BOLD connectivity within hippocampal-prefrontal circuits in the rat (Gass et al., 2014b; Becker et al., 2016). Hippocampal-prefrontal cortical connectivity is a key substrate hypothesized to be associated with cognitive and emotional state in central nervous system disorders. Ketamine, a NMDA receptor antagonist, is a psychomimetic agent and rapidly acting antidepressant in the clinic (Zarate et al., 2006) and thus understanding its acute modulatory effect on functional connectivity in the rat brain using rs-fMRI is of significant interest. Ketamine provokes a dose dependent increase in prefrontal connectivity (between the posterior hippocampus, retrosplenial cortex, and prefrontal regions) and these changes are highly concordant and directionally consistent with network reconfigurations observed in humans (Becker et al., 2016). A recent investigation of network reconfiguration induced by ketamine in anaesthetized monkeys assessed differences in functional networks 18 h after drug administration. Here, a down regulation most prominently in the orbital prefrontal cortex, the subgenual and posterior cingulate cortices and the nucleus accumbens were reported (Lv et al., 2016). The changes were reported to oppose the maladaptive alterations characteristic in the depressed brain and further suggested to reflect alterations in local synaptic plasticity triggered by blockade of NMDA receptors leading to network reconfiguration within cortico-limbic circuits. The technique may therefore serve as a novel approach for the identification of translational imaging biomarkers in drug development.

#### LIMITATIONS OF PRECLINICAL IMAGING

A limitation in the interpretation of animal MRI results is that animals are usually anaesthetized when undergoing MRI scans which may alter resting state brain function and perfusion. This is particularly important when considering the interpretation of resting state functional connectivity studies, however the choice of anaesthetic is important in this regard with α2-adrenoceptor agonist medetomidine shown to be better suited to resting state functional connectivity studies than inhalational volatile anaesthetics, such as isoflurane (Grandjean et al., 2014; Nasrallah et al., 2014). For most other imaging acquisitions isoflurane is still the most commonly used anaesthetic (for review see Hanusch et al., 2007; Hoyer et al., 2014). As always the size of the rodent brain makes high resolution imaging, and obtaining adequate signal-to-noise ratios challenging, particularly in relation to MRS. However, with the increasing field strength of preclinical scanners and also advanced coil systems these issues are being overcome.

In addition, an advantage of preclinical MRI is the ability to couple translational neuroimaging techniques with invasive or post-mortem investigations of molecular or cellular alterations associated with the phenotypes. Currently very few preclinical MRI studies incorporate significant post-mortem investigations and including invasive or ex vivo techniques such as immunohistochemistry for cell markers, or measurement of analytes in brain regions would advance interpretation of MRI findings.

Furthermore, the most common analysis methods for preclinical data rely upon human software packages which are not designed for the size and composition of the rodent brain therefore much of the pre-processing needs to be done manually, which whilst being labor intensive also introduces some investigator bias. There are ways to make rodent data compatible with human packages, with in-house scripts and some preclinical packages being developed and made freely

#### REFERENCES

Abbott, C. C., Jones, T., Lemke, N. T., Gallegos, P., McClintock, S. M., Mayer, A. R., et al. (2014). Hippocampal structural and functional changes available (e.g., DBAPI, SPM Mouse), and although progress is being made (Ioanas et al., 2017), a custom built, user friendly analysis package for both structural and functional preclinical MRI analysis would enable preclinical neuroimaging to be more accessible, therefore increasing the scope of the field.

Finally, it is worth highlighting that one of the benefits of animal models is the ability to use invasive techniques and given the limitations described above, preclinical MRI may not always be the best tool for delineating certain aspects of the pathophysiology. Tools such as optogenetics (Touriño et al., 2013) and positon emission tomography (PET) have allowed dissection of the neural circuits involved and alterations in neurochemistry related to depressive symptoms. Perhaps using these techniques in combination with translational MRI will further elucidate the functional alterations in depression and mechanism of antidepressant action.

# CONCLUSIONS

This review describes the recent findings in preclinical neuroimaging in animal models of stress and depression and highlights modalities that are particularly promising in translating results from animal models to clinical studies. Advances in preclinical models of depression with MRI-related changes reflecting the human condition, will further our understanding of the pathophysiology of MDD, and potentially uncover novel opportunities for drug development. Furthermore, this review emphasizes the potential for multi-modal MRI as a tool to highlight detectable physiological markers which may be useful indicators or susceptibility and lead to the development of clinical strategies that allow for patient stratification and tailored therapeutic strategies.

#### AUTHOR CONTRIBUTIONS

AM and SG prepared an initial draft of the animal section and LT prepared an initial draft of the clinical section of the manuscript. TF and AH contributed to writing all sections, and oversaw revisions and preparation of the final manuscript.

# FUNDING

AM, LT, TF, and AH are funded by the Marie Curie Brain Imaging Return to Health (r'Birth) consortium. SG was funded by a postgraduate scholarship awarded by Trinity College Dublin.

#### ACKNOWLEDGMENTS

The authors would like to acknowledge support from the Marie Curie Brain Imaging Return to Health (r'Birth) consortium.

associated with electroconvulsive therapy response. Transl. Psychiatry 4:e483. doi: 10.1038/tp.2014.124

Alcaro, A., Panksepp, J., Witczak, J., Hayes, D. J., and Northoff, G. (2010). Is subcortical–cortical midline activity in depression mediated by glutamate and GABA? A cross-species translational approach. Neurosci. Biobehav. Rev. 34, 592–605. doi: 10.1016/j.neubiorev.2009.11.023


perfusion-weighted functional MRI. Psychiatry Res. Neuroimaging 146, 213–222. doi: 10.1016/j.pscychresns.2005.12.008


behavior and increased GABA levels in the anterior cingulate: an 11.7T 1H-MRS study in rats. Prog. Neuro-Psychopharmacol. Biol. Psychiatry 51, 9–15. doi: 10.1016/j.pnpbp.2013.11.003


medial prefrontal cortex in late-life depression. Psychiatry Res. 172, 49–54. doi: 10.1016/j.pscychresns.2008.07.001


connectivity in a rat model of depression. World J. Biol. Psychiatry 1–9. doi: 10.1080/15622975.2016.1190866


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer BS and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review

Copyright © 2017 McIntosh, Gormley, Tozzi, Frodl and Harkin. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Richness in Functional Connectivity Depends on the Neuronal Integrity within the Posterior Cingulate Cortex

Anton R. Lord1, 2, 3 \*, Meng Li 2, 4, Liliana R. Demenescu1, 2, Johan van den Meer 2, 4, 5 , Viola Borchardt <sup>2</sup> , Anna Linda Krause2, 6, Hans-Jochen Heinze1, 4, 7, Michael Breakspear 3, 8 and Martin Walter 1, 2, 6, 7, 9 \*

*<sup>1</sup> Department of Behavioral Neurology, Leibniz Institute for Neurobiology, Magdeburg, Germany, <sup>2</sup> Clinical Affective Neuroimaging Laboratory, Otto-von-Guericke University, Magdeburg, Germany, <sup>3</sup> QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia, <sup>4</sup> Department of Neurology, Otto-von-Guericke University, Magdeburg, Germany, <sup>5</sup> Department of Cognition and Emotion, Netherlands Institute for Neuroscience, An Institute of the Royal Academy of Arts and Sciences, Amsterdam, Netherlands, <sup>6</sup> Department of Psychiatry and Psychotherapy, Otto-von-Guericke University, Magdeburg, Germany, <sup>7</sup> Center of Behavioral Brain Sciences, Otto-von-Guericke University, Magdeburg, Germany, <sup>8</sup> Metro North Mental Health Service, Royal Brisbane and Women's Hospital, Brisbane, QLD, Australia, <sup>9</sup> Department of Psychiatry, Eberhad Karls University Tuebingen, Tuebingen, Germany*

#### Edited by:

*Andrew Harkin, Trinity College, Dublin, Ireland*

#### Reviewed by:

*Andreas Hahn, Medical University of Vienna, Austria Vesa J. Kiviniemi, University of Oulu, Finland*

#### \*Correspondence:

*Martin Walter Martin.walter@uni-tuebingen.de Anton R. Lord Anton.Lord@qimrberghofer.edu.au*

#### Specialty section:

*This article was submitted to Systems Biology, a section of the journal Frontiers in Neuroscience*

Received: *26 October 2016* Accepted: *20 March 2017* Published: *07 April 2017*

#### Citation:

*Lord AR, Li M, Demenescu LR, van den Meer J, Borchardt V, Krause AL, Heinze H-J, Breakspear M and Walter M (2017) Richness in Functional Connectivity Depends on the Neuronal Integrity within the Posterior Cingulate Cortex. Front. Neurosci. 11:184. doi: 10.3389/fnins.2017.00184* The brain's connectivity skeleton—a rich club of strongly interconnected members—was initially shown to exist in human structural networks, but recent evidence suggests a functional counterpart. This rich club typically includes key regions (or hubs) from multiple canonical networks, reducing the cost of inter-network communication. The posterior cingulate cortex (PCC), a hub node embedded within the default mode network, is known to facilitate communication between brain networks and is a key member of the "rich club." Here, we assessed how metabolic signatures of neuronal integrity and cortical thickness influence the global extent of a functional rich club as measured using the functional rich club coefficient (fRCC). Rich club estimation was performed on functional connectivity of resting state brain signals acquired at 3T in 48 healthy adult subjects. Magnetic resonance spectroscopy was measured in the same session using a point resolved spectroscopy sequence. We confirmed convergence of functional rich club with a previously established structural rich club. N-acetyl aspartate (NAA) in the PCC is significantly correlated with age (*p* = 0.001), while the rich club coefficient showed no effect of age *(p* = 0.106). In addition, we found a significant quadratic relationship between fRCC and NAA concentration in PCC (*p* = 0.009). Furthermore, cortical thinning in the PCC was correlated with a reduced rich club coefficient after accounting for age and NAA. In conclusion, we found that the fRCC is related to a marker of neuronal integrity in a key region of the cingulate cortex. Furthermore, cortical thinning in the same area was observed, suggesting that both cortical thinning and neuronal integrity in the hub regions influence functional integration of at a whole brain level.

Keywords: fMRI, MRS, graph metric, aged, cortical thickness

#### INTRODUCTION

Functionally, human brains are known to exhibit a highly organized structure (Meunier et al., 2010). Sub-networks within the brain have been identified which activate while performing various tasks related to visual stimuli (Calhoun et al., 2001), action (Karni et al., 1995), cognitive tasks (Buckner et al., 1996), or at rest (Damoiseaux et al., 2006). Brain regions involved in each of these tasks organize into distinct communities (Meunier et al., 2009). Underlying these segregated communities is a set of highly inter-connected regions facilitating functional connectivity between communities (Sporns et al., 2007). Typically highly influential regions are both structurally (van den Heuvel and Sporns, 2011) and functionally (van den Heuvel et al., 2009) interconnected, forming the so called rich club (Colizza et al., 2006). Despite advancements in imaging analysis techniques regarding graph theoretical analysis for functional MRI, there is currently little knowledge of the biochemical and biological neurotransmitter underpinnings of graph outcome metrics.

The rich club in human brains was first characterized structurally using diffusion tensor imaging (DTI) (van den Heuvel et al., 2013; Collin et al., 2014a), highlighting a consistent set of key regions. High resolution DTI studies have identified a myriad of rich club regions including frontal-temporal, anterior and posterior midline regions, and posterior cingulate cortex (van den Heuvel and Sporns, 2011). Functionally, there are few studies exploring the rich club in human brains at rest. One such study that found a significant enrichment of the fRCC in healthy adults when compared to adolescents (Grayson et al., 2014). While rich club like properties have been observed in both structural and functional modalities of brain imaging, fundamental underlying differences in imaging techniques and their subsequent analysis preclude meaningful direct comparison of the two. Most importantly, resting state functional connectivity is susceptible to the transitive property, particularly between highly correlated regions (Langford et al., 2001), an effect that structural imaging does not suffer from.

Recent studies have identified a number of regions which typically enter their active state during periods of rest with eyes closed (Zhang and Raichle, 2010), commonly known as the default mode network (DMN). Within the DMN the posterior cingulate cortex (PCC) plays an important role, being directly structurally connected to both the medial temporal lobe (MTL) and the medial prefrontal cortex (MPFC) (Greicius et al., 2009). Functionally, PCC is predominantly known as a core component of the default mode network (Greicius et al., 2003). However, it is also involved in other networks, such as the dorsal attention network and the executive control network (Leech and Sharp, 2014). Due to the extensive functional importance of the PCC across multiple networks, we hypothesize that it is a key hub node facilitating communication between multiple canonical networks. While the importance of the PCC in widespread communication between numerous brain networks is well documented, little is currently known about the impact of biochemical and structural alterations within the PCC on its functional connectivity with other key areas. Hence, we sought to elicit the impact of local biochemical and structural factors in the PCC on the fRCC in a healthy population.

Proton magnetic resonance spectroscopy (MRS) enables the identification of regionally specific metabolic signatures noninvasively (Duncan, 1996). Of particular importance to the present study, one particular metabolite, N-acetyl Aspartate (NAA), can be identified using this technique. NAA is thought to be present predominantly in neuronal cells, which has been shown to be a biomarker for neuronal and axonal integrity (Dautry et al., 2000; Demougeot et al., 2001). Another measure, which can be acquired using T1-weighted structural images allows for the quantification of neuronal and glial density (Li et al., 2014), known as cortical thickness (CTh).

Cortical thinning has been shown to occur due to aging in a healthy population (Salat et al., 2004). Furthermore, decreased cortical thickness within the pregenual cingulate cortex has been observed in patients suffering from late onset depression (Mayberg, 2003; Lim et al., 2012). One recent study identified a positive correlation between CTh and NAA in the dorsal anterior cingulate cortex (Li et al., 2014). However, the relationship between CTh, NAA, and the rich club coefficient are to our knowledge unexplored.

Due to the highly influential nature of the PCC at a functional level, we hypothesized that it exerts a substantial influence on the functional backbone of the network. Furthermore, we conjectured that both local metabolic and structural properties impact on the extent of its influence. Hence, we sought to elicit the impact of local metabolic and structural properties of the PCC on the overall importance of the rich club.

#### MATERIALS AND METHODS

Forty eight healthy volunteers (33.08 ± 8.68 years) were recruited in Magdeburg, Germany. The sample consisted of 35 males and 13 females. Participants were excluded based on major medical illness, pregnancy, history of seizures, current psychiatric disorder or general MRI contradictions. The study was approved by the institutional review board of the University of Magdeburg and all subjects gave written informed consent before inclusion.

#### Data Acquisition

The fMRI data were acquired on a 3 Tesla Siemens MAGNETOM Trio scanner (Siemens, Erlangen, Germany) with an eightchannel phased-array head coil. Subjects were requested to lie still with their eyes closed for the duration of the resting state scan. A total of 488 volumes were acquired with an echo-planar imaging sequence. The following acquisition parameters were used: echo time = 25 ms, field of view = 22 cm, acquisition matrix = 44 × 44, isometric voxel size = 5 × 5 × 5 mm<sup>3</sup> . Whole brain coverage was achieved with 26 contiguous axial slices, using a repetition time of 1,250 ms and flip angle of 70◦ . The first 10 volumes were discarded to allow for magnetic field homogenization. High resolution T1-weighted structural MRI scans of the brain were acquired for structural reference using a 3D-MPRAGE sequence (TE = 4.77 ms, TR = 2,500 ms, T1 = 1,100 ms, flip angle = 7 ◦ , bandwidth = 140 Hz/pixel, acquisition matrix = 256 × 256 × 192, isometric voxel size = 1.0 mm<sup>3</sup> ).

#### Data Preprocessing

Functional data were corrected for differences in slice time acquisition, motion-corrected using a least squares approach and a six-parameter (rigid body) linear transformation and spatially normalized. The data were linearly detrended. All subjects had less than 1 mm head motion in any direction during the scanning session. An additional regression of nuisance covariates was applied during which the functional data was corrected for global mean signal as well as for white matter and cerebrospinal fluid signal and bandpass filtering between the frequencies of 0.01 and 0.1 Hz. Data were preprocessed using spm8 (Wellcome Trust Center for Neuroimaging, London, England) using the data processing assistant for resting-state fMRI (DPARSF version 2.3, Yan and Zang, 2010).

The resulting volumes were parcellated into 102 nodes using a modified version of the automatic anatomic labeling (AAL) atlas (Tzourio-Mazoyer et al., 2002) containing a fine grained parcellation for the cingulate and insular cortices (Horn et al., 2010; Lord et al., 2012). To compute the resting state functional connectivity (rsFC) of the ROIs, the mean time course of every voxel within each ROI was extracted and Pearsons correlation coefficients were calculated pair-wise for all pairs of ROI's, resulting in 102 by 102 symmetric matrices.

#### MRS Data Acquisition and Analysis

Proton MRS data were acquired from a single voxel (10 × 20 × 20 mm<sup>3</sup> ), located in the PCC for each subject at rest (**Figure 1**). MRS data were acquired using a point resolved spectroscopy (PRESS) sequence with an echo time of 80 ms, repetition time of 2,000 ms, 256 averages, 1,200 Hz bandwidth, 853 ms acquisition time and water suppression. Shimming was performed automatically, with manual fine tuning to improve field homogeneity. To correct for eddy currents, four pulses without water suppression (TR = 10 s) were averaged. Voxel acquisition time was 8 min and 40 s. Spectra were analyzed using LCModel version 6.3.0 (Provencher, 2001). Sixteen different metabolites (Creatine, Glutamate, myo-Inositole, Lactate, NAA, Phosphocholine, Taurine, Aspartate, GABA, Glutamine, Glucose, Alanine, NAAG, Phosphocreatine, Guanine, and Glycerophosphocholine) were fitted using a basis set including all these substances. The concentration of NAA was calculated as its ratio to creatine (Cr) as it is known to be an appropriate reference (Yildiz-Yesiloglu and Ankerst, 2006). Model fit error was estimated using the Cramer-Rao lower bounds (CRLB) (Cavassila et al., 2001) measure, with spectra deemed to be reliable if CRLB was less than 20%. Other minimum quality criteria included a full width half maximum value of smaller than 12 Hz and a signal to noise ratio of greater than eight.

#### Cortical Thickness

Cortical thickness (CTh) was measured across the whole brain using the CIVET pipeline (Zijdenbos et al., 2002). MR-images were linearly transformed (rigid body, 6 degrees of freedom) to standard Montreal Neurological Institute (MNI) space with intensity and non-uniformity also corrected (Sled et al., 1998). These images were then segmented into gray matter (GM), white matter (WM), cerebral spinal fluid using a neural net classifier. Next cortical surfaces were extracted for both the pial-cortical surface and GM-WM boundary for each hemisphere separately using a deformable mesh model implemented inside the Constrained Laplacian Anatomic Segmentation using Proximity algorithm (June et al., 2005). Both surfaces were non-linearly registered to a high resolution standard template to ensure the co-registration of all subjects to a common space (Lyttelton et al., 2007). To minimize morphological distortion, the inverse of the aforementioned transformation was applied for each subject and CTh was measured in native space. To calculate the GM percentage in the MRS voxel, the voxel was co-registered to the subjects T1 image using SPM8. MRS voxels for each subject were projected into template space using each subject's affine matrix calculated to transform each hemisphere to standard space earlier.

#### Rich Club Coefficient

These matrices were rendered sparse by recursively removing connections, starting with the lowest correlation until only 10% of connections remain. The sparsity of 10% was selected as networks with more edges tends to introduce weaker, noisy effects obscuring between-group effects (Rubinov et al., 2009; Lord et al., 2012; Borchardt et al., 2015). After thresholding

no negative edges remained. To ensure that the graph didn't disconnect, edges which would cause a disconnection if removed were retained even if their correlation was below the cutoff threshold. These sparse networks were then binarized, and the degree of each node is calculated.

The fRCC was then calculated across a range of degree thresholds. This was done by removing nodes with a degree of less than the chosen threshold and dividing the number of connections in the remaining subnetwork with the maximum possible number of connections. This ratio in and of itself is known not to be a good measure of interconnectedness (van den Heuvel and Sporns, 2011), thus we reference this information against what we would expect by random chance. To this end, we randomly move connections in the thresholded connectivity matrix, creating a randomized matrix with the same degree distribution (Maslov and Sneppen, 2002), preserving the degree distribution of the network while destroying the connection pattern. The rich club coefficient for each randomized graph was then calculated and the ratio of the raw RCC to the mean RCC of all randomized graphs at each degree threshold was computed. As such, a normalized RCC of 1 speaks to the connectivity pattern of the rich club being equal to a random graph, that is, high degree nodes are no more or less likely to connect to each other than by random chance. A value greater than 1 shows an affinity of rich nodes to cluster more tightly than expected together, while a value below 1 shows rich nodes predominantly connect with non-rich nodes. Since all analysis using the rich club refers to the normalized rich club and not the raw rich club, the normalized rich club will be henceforth described solely as the rich club. The regions included as rich club members were also extracted.

Rich club membership was defined on an individual level. That is, for each individual the degree threshold giving the highest RCC was identified, then the regions identified as members at this threshold were selected. At the group level, rich club members were identified as those regions appearing in at least 50% of subjects.

#### RESULTS

A positive rich club coefficient was observed for all subjects, with the largest fRCC across the group at a degree threshold of 9 (Figure S1). Twenty-one rich club members were consistent in at least half of the participants (**Figure 2**), with seven out of 21 regions from within the cingulate cortex, including the PCC (**Table 1**). No correlation between global signal and fRCC was observed (Figure S2).

NAA in the PCC negatively correlated with age (**Figure 3A**, r = −0.456, p = 0.001, t = −3.475, df = 46), while no significant correlation exists between rich club coefficient and age (r = 0.2356, p = 0.107, t = 1.644, df = 46, **Figure 3B**). Furthermore, a significant quadratic relationship was identified between RCC and NAA/Cr in the PCC (Equation 1, **Figure 4**, p = 0.009). This relationship was not observed using a linear model (r = −0.041, p = 0.093, t = −1.708, df = 46). The model fit was formally tested using the bayesian information criterion (BIC), which revealed that a quadratic model fit the data better than a linear model (delta BIC = −3.16) or any higher order polynomial (Figure S3).

TABLE 1 | Rich club members.


Linear correlates between low and high RCC values were assessed post hoc. The data were split into two groups, below the RCC value predicted to give the lowest concentration of NAA/Cr (2.44), and above this value. A negative linear correlation was observed in the low RCC values (r = −0.365, p = 0.036, t = −2.185, df = 31), while a trending positive linear correlation was observed in the high RCC values (r = 0.473, p = 0.075, t = 1.935, df = 13). Using a Fisher R-to-Z transformation, these two correlations were significantly different (z = −2.62, p = 0.009, Figure S4). In contrast this did not translate into dependency of global metrics (SWI and CC), which did not correlate with PCC-NAA.

Nonlinear relationship between NAA/Cr in the PCC and the RCC. F = 5.206, p = 0.009254. y = 1.933 − 0.621x + 0.127x<sup>2</sup> (1)

Cortical thickness in the PCC positively correlated with the rich club coefficient after accounting for age and NAA/Cr as covariates (**Figure 5**, FDR corrected). Although extensive parts of the cortex showed reduced CT as a factor of age (Figure S5), this did not overlap with the previous finding. CT was also positively correlated with NAA/Cr, predominantly in the mid cingulate cortex, but did not overlap with the fRCC correlation with CT in the PCC (Figure S6). After accounting for age as a covariate, fRCC and mean CTh in the PCC was positively correlated (r = 0.339, p = 0.015, t = 2.519, **Figure 6**). When considering only age as a covariate, a cluster of vertices in the same location is observed at an uncorrected threshold only (Figure S7).

#### DISCUSSION

Although the rich club coefficient has become increasingly studied in both functional and structural networks in human brains (van den Heuvel et al., 2012; Collin et al., 2014b; Grayson et al., 2014), little is currently known about the impact of local metabolic and structural properties of key rich club regions on the integrity of the club as a whole. Hence, we sought to identify effects of neuronal integrity and cortical thinning in the PCC on the density of connections within this functional backbone of the human brain.

Regional distribution of rich club members in this study were in line with previous work (van den Heuvel and Sporns, 2011; van den Heuvel et al., 2013), showing most of the members of the rich

club surrounding the inter-hemispheric fissure. Furthermore, we observe many areas of the cingulate cortex as members of the rich club, suggesting a high level of communication both within the cingulate cortex as well as to other regions of the brain. This result speaks to the central role of the cingulate cortex in information transfer at a whole brain level. Indeed, one third of all rich club members were subregions of the cingulate cortex, including pregenual and rostal parts of the ACC, posterior MCC, dorsal and ventral PCC and Brodman area 23d. During preprocessing we regressed out global signal from out resting state data. This procedure has met with criticism as it can introduce spurious relationships within the data. We observed no relationship between fRCC and GSR, suggesting our results are not driven by the global signal.

The quadratic relationship between the functional rich club and NAA/Cr initially declines linearly, then increases at higher RCC levels. This suggests that other factors, in addition to the NAA/Cr concentration in the PCC, influence the functional rich club coefficient. Confirming this, we found a significant reduction of CTh in the PCC related to the RCC, after accounting for age and NAA/Cr. These results show that both the neuronal integrity and cortical thickness affect the richness of the network.

The directionality of correlations will require further in depth analysis, at best including suitable behavioral parameters. Lower NAA is normally considered indicative of reduced neuronal integrity while its relevance to functional rich club alterations is less clear. Age related whole network changes for example in Alzheimers disease found effects rather in peripheral functional connections leaving the core network rather undisturbed (Daianu et al., 2015) at least for functional connectivity properties. In clinical context, schizophrenia was associated with reduced rich club connections, speaking in favor of a beneficial role of higher functional rich club routing and routing efficiency.

Nodes of the structural rich-club are especially resilient to targeted attacks and enhance global information flow by acting as a "highway" system (van den Heuvel and Sporns, 2011; van den Heuvel et al., 2012; Xia et al., 2016). Nevertheless, the structural rich-club is a relatively high-cost component of brain networks because the wiring cost is greater between members of the rich club than between less well-connected nodes in the periphery (Collin et al., 2014b). Also, structural rich-club nodes have higher levels of metabolic energy consumption than peripheral nodes (Collin et al., 2014b). Structural rich clubs were shown to connect early and are maintained through in maturation ("rich-get-richer"), suggesting that these nodes may have a key developmental role (Schroeter et al., 2015). Thus, it has been argued that maintenance of such a costly network component should offer advantages to the brain's computational performance. In line with this thought, mature rich-clubs were indeed shown to be of great importance for routing of spontaneous activity flow in the network frequently acting as brokers for spontaneous multi-unit activity, suggesting a role of rich-clubs for orchestrating coordinated activity in the network, for example switching between different network states (Crossley et al., 2013; Leech and Sharp, 2014; Senden et al., 2014; Schroeter et al., 2015).

In our case we found lower fRCC associated with higher NAA, thus a direct interpretation in terms of functional relevance will be difficult, particularly since we only investigated healthy subjects.

Our results are in line with previous accounts for the influence of local metabolism within a node on brain network configuration. Horn et al. reported that functional resting state connectivity between pgACC and anterior insula was correlated with Glx concentrations in pgACC (Horn et al., 2010). This effect was localized in that insula MRS did not show similar influences on this connectivity. More recently, Demenescu et al. (2016) show that anterior insula MRS predicted long range resting state connectivity, however toward temporo-parietal and visual cortices. Importantly, these findings of glutamatergic modulations were found explicitly for depressed patients, while healthy controls did not show the same correlation (Horn et al., 2010). This was interpreted as a reflection of a primary glutamatergic deficit in MDD, as supported by respective meta-analyses (Taylor, 2014) while in the absence of a direct

metabolic deficit variance of the metabolite was considered to be too low.

In contrast, work on PCC MRS in healthy subjects reported an interrelation of both glutamate and GABA on PCC functional connectivity, at least when investigating the default mode network as a whole via independent component analysis (Kapogiannis et al., 2013). Therefore, it is plausible that also in healthy conditions, variance of both functional connectivity and brain metabolites as acquired in MRS is large enough to detect covariations in healthy cohorts. Such covariations should then reflect a common biological mechanism and the most prominent mechanism in healthy cohorts would be that of aging.

Age related brain changes have been investigated in greater extent and prominent effect have ben postulated for NAA (Block et al., 2002). Indeed we found NAA to be correlated with age and RCC however there was no direct correlation between RCC and age. Furthermore, the relationship of PCC cortical thickness was controlled for potential age effects, thus pointing toward a second, age independent mechanism, which relates PCC functional integrity as evident in cortical thickness and NAA levels toward RCC. It must also be stated that the sample included in our study is of rather young age and thus neurodegenerative effects, previously investigated may not be the main driver.

Although Grayson et al. (2014) found an enriched rich club in adolescents, we did not see a significant correlation between age and fRCC (p = 0.1068), which may be explained by the vast developmental changes in the brain during the adolescent period (Giedd, 2004) in comparison the putative effects in our sample.

An alternative interpretation would consider the mild apparent, if not statistically significant increase of fRCC with age as a subtle counterpart to the significant age dependent NAA decrease. As such MRS would be considered a very sensitive age marker while secondary effects of age on whole brain organization would be more subtle and not yet detectable in the observed age range between early twenties and mid-fifties. This would then be reflected by the linear component fRCC increase with decreasing NAA, and may, in a far stretch, be interpreted as a sign of age related "hyper-efficiency."

However, such interpretation would not work for the observed quadratic relationship, which for brains with very high RCC's rather found the inverse relationship of increasing fRCC with higher NAA. This suggests that other factors, in addition to the NAA/Cr concentration in the PCC, influence the rich club coefficient. In support of this interpretation, one may add that we found linear negative relationship of CTh in the PCC and fRCC. However, we found this after accounting for age and NAA/Cr, therefore results at this stage should best be appreciated in that we found evidence that both the neuronal integrity and cortical thickness affect the richness of the network and the parallel, potentially driving, or counteracting effects of age and other sources of inter-individual variation will be subject to future investigations in larger cohorts.

Within the context of these considerations, our results in a comparably large, young adult population point toward an effect of local neuronal integrity or at least neurobiological constitution on resting state network configuration.

The new advance in our finding is first the extension to NAA, after previous reports mainly focused on neurotransmitters GABA and Glutamate. Secondly after previous research as investigated metabolite influence on direct connections of a region, either in terms of seed based edges or in terms of the independent component which hosts the location of the single

MRS voxel, we extend the view toward a much more global effect on network constitution. This effect is very likely due to the central role of the PCC within the rich club. While we cannot test this explicitly at this stage, we would assume that the relationship found here is mainly representative of the central role of the PCC within a backbone of small network organization rather than due to the specific impact of NAA in comparison to GABA or glutamate. Ideally, to test such claim, one would request MRS sequences which allow assessment of all mentioned metabolites within the scope of a single session along a number of regions and further leave time for additional resting state assessments. Such methods are available for higher field strengths, making explicit use of the improve line separation (Dou et al., 2013; Li et al, 2016).

#### LIMITATIONS

Only 13 out of 48 subjects included in this study were female. Due to the small sample of females included, gender effects have not been discussed. However, future work would benefit from investigating gender specific variation in the context of NAA, CTh and RCC.

# CONCLUSION

The rich club in fMRI is an area of growing interest, with implications in neurological development (Grayson et al., 2014) and psychiatric disease (van den Heuvel et al., 2013; Collin et al., 2014a). Our findings add evidence to this growing avenue of research, progressing our understanding of the impact of

metabolic and structural factors on the rich club. To our knowledge, this is the first study identifying the impact of these

#### REFERENCES


factors targeted to the PCC, showing a relationship between CT and NAA/Cr neurotransmitter levels and its effects on this core subnetwork of resting state functional connectivity via the PCC.

#### AUTHOR CONTRIBUTIONS

MW and HH designed the experiment. AK, MW, and ML acquired the data. AL, ML, MW, Jv, LRD, and VB analyzed the data. AL, VB, MW, and MB interpreted the results. AL drafted the manuscript. AL, VB, MW, ML, Jv, and LRD revised the manuscript. All authors approved the final version of the manuscript.

#### FUNDING

This study was supported by funding from the following grants: SFB 779/A6 (German Research Foundation) to MW and FP7 MC-ITN R'Birth (Marie Curie) to AL and MW.

## ACKNOWLEDGMENTS

The authors would like to thank Claus Tempelmann, Denise Scheermann, Renate Blobel, Carina Födisch, Berit Wiegmann, and Maya Nathan for their help and assistance during data collection and subject testing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnins. 2017.00184/full#supplementary-material


loss: its relevance to studies of acute brain injury. J. Neurochem. 77, 408–415. doi: 10.1046/j.1471-4159.2001.00285.x


networks. Front. Neuroinformatics 3:37. doi: 10.3389/neuro.11.0 37.2009


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Lord, Li, Demenescu, van den Meer, Borchardt, Krause, Heinze, Breakspear and Walter. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Investigating the Role of Glutamate and GABA in the Modulation of Transthalamic Activity: A Combined fMRI-fMRS Study

Nathalie Just 1, 2 \* and Sarah Sonnay <sup>3</sup>

<sup>1</sup> CIBM-AIT core, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, <sup>2</sup> University Hospital Münster, Münster, Germany, <sup>3</sup> LIFMET, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

The Excitatory-Inhibitory balance (EIB) between glutamatergic and GABAergic neurons is known to regulate the function of thalamocortical neurocircuits. The thalamus is known as an important relay for glutamatergic and GABAergic signals ascending/descending to/from the somatosensory cortex in rodents. However, new investigations attribute a larger role to thalamic nuclei as modulators of information processing within the cortex. In this study, functional Magnetic Resonance Spectroscopy (fMRS) was used to measure glutamate (Glu) and GABA associations with BOLD responses during activation of the thalamus to barrel cortex (S1BF) pathway at 9.4T. In line with previous studies in humans, resting GABA and Glu correlated negatively and positively respectively with BOLD responses in S1BF. Moreover, a significant negative correlation (R = −0.68, p = 0.0024) between BOLD responses in the thalamus and the barrel cortex was found. Rats with low Glu levels and high resting GABA levels in S1BF demonstrated lower BOLD responses in S1BF and high amplitude BOLD responses in the thalamus themselves linked to the release of high GABA levels during stimulation. In addition, early analysis of resting state functional connectivity suggested EIB controlled thalamocortical neuronal synchrony. We propose that the presented approach may be useful for further characterization of diseases affecting thalamocortical neurotransmission.

#### Keywords: fMRS, BOLD, resting state, barrel cortex, thalamus, glutamate, GABA

#### HIGHLIGHTS


# INTRODUCTION

There is a growing body of evidence showing that both excitatory and inhibitory neurons govern the hemodynamic response to increased neuronal activity (Kocharyan et al., 2008; Enager et al., 2009; Logothetis et al., 2010; Lecrux et al., 2011) at the same time as they need "fuel" to work. As a consequence of this activity, neurons release glutamate (Glu), and γ-amminobutyric acid (GABA) neurotransmitters. Glu and GABA as well as other metabolites (Lactate, Aspartate, Glucose..) play a significant role in modulating brain activity during both stimulus-induced activity and "intrinsic"

#### Edited by:

Andrew Harkin, Trinity College Dublin, Ireland

#### Reviewed by:

Annette Horstmann, Max Planck Institute for Human Cognitive and Brain Sciences, Germany Anton Lord, QIMR Berghofer Medical Research Institute, Australia

> \*Correspondence: Nathalie Just nathalie.just@ukmuenster.de

#### Specialty section:

This article was submitted to Systems Biology, a section of the journal Frontiers in Physiology

Received: 12 October 2016 Accepted: 11 January 2017 Published: 31 January 2017

#### Citation:

Just N and Sonnay S (2017) Investigating the Role of Glutamate and GABA in the Modulation of Transthalamic Activity: A Combined fMRI-fMRS Study. Front. Physiol. 8:30. doi: 10.3389/fphys.2017.00030 ongoing activity (Duncan et al., 2014). More than 10 years ago, Chen et al. (2005) demonstrated that GABA enhancement can decrease Blood Oxygen Level dependent (BOLD) signal amplitudes in the rat forepaw cortex. It is now well-admitted that resting state GABA concentrations correlate negatively with Blood Oxygen Level dependent responses in various regions of the human brain (Northoff et al., 2007; Donahue et al., 2010; Muthukumaraswamy et al., 2012; Bednaˇrík et al., 2015). Morover, the modulation of neuronal activity by resting glutamate concentrations was also shown (Kapogiannis et al., 2013) although studies examining the relationship were sparser.

Another interesting approach for an improved understanding of neuroimaging signals may be to investigate how neuronal activity in a specific region of the brain influences the rest of its interconnected network. In particular, derivation of metabolic relationships between different regions in a neural network may be informative. Duncan et al. (2011) assessed functional connectivity between regions of the human anterior cingulate cortex (ACC) and related BOLD responses to [Glu] in one region but not the other thus demonstrating the influence of deactivation of one region on activation of the other through a glutamatergic pathway. Although the number of studies investigating how excitatory and inhibitory neurotransmitters mediate BOLD responses is growing, the potential of this relatively new approach remains to be explored.

In rodents, the thalamus is recognized as one of the most important brain areas for driving cortical processing (Devor et al., 2005; Poulet et al., 2012; Feldmeyer et al., 2013). Thalamic nuclei such as the ventral posteromedial thalamic nucleus (VPM) were often described as important relays for glutamatergic and GABAergic functional information processing of the ascending (and descending) response to whisker or forepaw stimulations. Notably, in models of absence epilepsy, the thalamus revealed to be more than a passive resonator for spike-wave-discharges maintenance while directional nonlinear couplings between cortex and thalamus were suggested in conjunction with several glutamatergic and GABAergic receptor dysfunction (Lüttjohann and van Luijtelaar, 2015). Other pathologies such as schizophrenia could also originate from thalamic dysregulation and functional disconnectivity again linked to excitatory-inhibitory balance (EIB) dysfunction (Behrendt, 2006). In particular, several optogenetics studies suggested that glutamatergic and GABAergic modulations of neural activity could benefit the investigation of diseases affecting thalamo-cortical neurotransmission. In order to examine this proposition under pathological conditions, normal biochemical modulatory activities between cortex and thalamus must be characterized.

In the present study, the neurochemical profiles within the barrel cortex (S1BF) and the thalamus of rats before and during stimulation of the trigeminal nerve as well as BOLD responses were measured. A correlation study was then conducted to characterize associations between BOLD, [Lac], [Glu], and [GABA] in both thalamic and cortical structures and between them. We expected to verify that BOLD responses and [GABA] were negatively correlated within cortical and subcortical areas while positive correlations between BOLD and stimulation-induced changes in glutamate levels (1[Glu]) as well as lactate (1[Lac]) were envisaged in cortex. The influence of glutamate and GABA on thalamocortical neurotransmission were discussed.

#### MATERIALS AND METHODS

#### Animals

All studies were performed following the approval of Service de la consommation et des affaires vétérinaires du canton de Vaud (Switzerland) and according to the federal guidelines of the Animal Care and approved by the local authority (EXPANIM-SCAV). Male Sprague-Dawley rats (n = 15, 350 ± 40 g; Charles River, L'Arbresle, France) under isoflurane anesthesia (2–3%) vaporized in 30% O<sup>2</sup> in air were intubated, and mechanically ventilated. Two femoral arteries and one femoral vein were catheterized for blood gas sampling and blood pressure measurements as well as α-chloralose (an initial intravenous dose of 80 mg/kg was administered followed by a continuous intravenous infusion of 27 mg/kg/h at a rate of 2 ml/h) and pancuronium administrations. Respiration rate was monitored through a pillow (SA Instruments, Stony Brook, NY, USA) placed underneath each rat. Temperature was measured using a rectal sensor and regulated via control of the temperature of water flowing through tubing covering the body of each rat and linked to a temperature-regulated bain-marie. Less than 300 µl of arterial blood were sampled every 30 min and blood parameters directly measured using an AVL blood gas analyzer (Dotmed, USA). Mean Arterial blood pressure (MABP) was measured continuously using a transducer attached to the femoral artery catheter. Body temperature and blood parameters were maintained at physiological levels (T = 37.5◦C ± 0.5◦C; pH = 7.4 ± 0.05, pCO2 = 39.7 ± 7 mmHg and MABP = 148.9 ± 11 mmHg) throughout each experiment. An intravenous femoral injection of Pancuronium Bromide (Sigma, Switzerland) of 0.7 ml per hour was performed to minimize tremors. Rats were positioned in a dedicated stereotactic holder equipped with ear and bite bars which was tilted in the magnet (30–45◦ ) for a better positioning of voxels for fMRS over the barrel cortex. fMRS measurements were conducted sequentially in thalamus and S1BF followed by BOLD fMRI in 15 rats. The resting-state fMRI analysis is presented as supplementary data in the present study.

#### Trigeminal Nerve Stimulation (TGN)

Electrodes were percutaneously inserted in the left infraorbital nerve. Electrical stimulation of the left trigeminal nerve (TGN) was performed using an external stimulator (WPI, Stevenage, UK) as described in Just et al. (2010). The paradigm of stimulation (1 minOFF–1 minON...) for both fMRI and fMRS

**Abbreviations:** fMRS, Functional Magnetic Resonance Spectroscopy; BOLD, Blood Oxygen Level Dependent; TGN, trigeminal Nerve; S1BF, Primary somatosensory Barrel Field cortex; [Glu], glutamate concentration; [GABA], γ-amminobutyric acid concentration; [Lac], Lactate; [Gln]: Glutamine. VPM, ventral posteromedial thalamic nucleus POm, Posteromedian Thalamic Nucleus; VOI, Volume of Interest; VASO, vascular space occupancy; CBF, cerebral blood flow; fWE, family wise error.

was repeated for 32 min with pulse duration of 0.5 ms, stimulation frequency of 1 Hz and stimulation current amplitude of 2 mA (Sonnay et al., 2015).

#### Magnetic Resonance Experiments

Experiments were described schematically in **Figure 1**. Experiments were performed on an actively shielded 9.4T/31 cm bore magnet (Agilent, USA) with 12 cm gradients and a surface coil. Shims were adjusted using FAST(EST)MAP (Gruetter and Tkác, 2000).

#### Functional MR Spectroscopy

Localized proton spectroscopy was performed using Spin Echo Full Intensity Acquired Localized Sequence (SPECIAL, TE = 2.8 ms, TR = 4 s) (Mlynárik et al., 2006) in 15 rats. The voxel of interest for the thalamus was chosen by reference to the Paxinos and Watson atlas (Paxinos and Watson, 1998) so that it encompasses the VPM and POm structures (n = 12). Its size was 3 × 3 × 4 mm<sup>3</sup> and it was shimmed down to a linewidth of 12 ± 2 Hz. In the barrel cortex, the VOI size was 1.5 × 3 × 5 mm<sup>3</sup> and was shimmed down to a linewidth of 10 ± 2 Hz (n = 15). For each VOI, <sup>1</sup>H spectra were acquired during 32-min of rest followed by 32 min of TGN stimulation corresponding to 480 scans (30 × 16) per period. The water signal was suppressed using the VAPOR module containing a series of seven 25 ms asymmetric variable power RF pulses with optimized relaxation delays. To improve the signal localization, three modules of outer volume saturation (OVS) were interleaved with the water suppression pulses. The raw <sup>1</sup>HMRS spectra corrected for frequency drift and summed were used for LCModel analysis with a basis set of 21 simulated metabolites (Macromolecules: Mac; Scyllo-inositol: Scyllo; Ala: Alanine; Ascorbate: Asc; Aspartate: Asp; β-hydroxybutyrate: bHb; Glycerophosphocholine: GPC, Phosphocholine: PCho; Creatine: Cr; Phosphocreatine; γaminobutyric acid: GABA; Glucose: Glc; Glutamine: Gln; Glutamate: Glu; Glutathione: GSH; myo-inositol: Ins; Lactate: Lac; N-acetylaspartate: NAA; N-Acetylaspartylglutamic acid:NAAG; Phosphatidylethanolamines: PE; Taurine: Tau). Absolute metabolite concentrations were obtained using unsuppressed water signal (8 averages) as an internal reference. The Cramer–Rao lower bounds were used as a reliability measure of the metabolite concentration estimates. Higher concentration metabolites (Glu, NAA, Ins, Tau, PCr, Cr) with Cramer-Rao lower bounds (CRLB) under 10% were considered to be reliably quantified whereas for Glc, GABA, Lac, Asp CRLB under 30% were considered acceptable and were kept for further analysis.

Neurochemical profiles in thalamus and S1BF of the same rat were acquired serially for 12 rats starting with the thalamic nuclei and S1BF at rest followed by successive 32-min TGN stimulation periods for S1BF and thalamus. A 15-min period of time was left before acquiring in S1BF at rest, which was used to tilt the holder and perform further adjustments (morphological T2 imaging and Shim) as well as to return the holder to initial position between acquisitions in S1BF and thalamus. A one-way ANOVA test with Bonferroni correction was used to compare metabolite concentrations at rest and during stimulation. The significance level was set at 0.05. All the results are presented as Mean ± SEM. A two-way ANOVA test with Bonferroni correction was performed to compare Glu, GABA, Lac, Glc, and Gln between thalamus and S1BF brain structures during ON (stimulation) and OFF (rest) measurement times.

## BOLD fMRI

BOLD responses in the barrel cortex and thalamus were measured post-fMRS measurements using single shot gradient echo EPI in 15 rats (7 extra rats underwent BOLD fMRI only, a total of 22 rats underwent BOLD–fMRI). **First** and second order shims were adjusted using FAST(EST)MAP (Gruetter and Tkác, 2000) resulting in water linewidths of 12 ± 3 Hz in a 216 µl (6 × 8 × 4.5 mm<sup>3</sup> ) volume. After echo re-alignment using a reference scan, BOLD responses were assessed using single shot gradient echo EPI (TR/TE = 2500–2000/25 ms; FOV = 20 × 20 mm; matrix = 64 × 64; slice thickness = 1 mm; 8 slices, Bandwidth = 325 KHz, 900 volumes).

#### Data Analysis

Images were analyzed with SPM8 (Matlab; The Mathworks; Natick, USA; Statistical Parametric Mapping, www.fil.ion.ucl. ac.uk/spm/) using the general Linear model (GLM) analyzing each voxel independently and creating a parametric map of statistical significance (Friston et al., 1995). Gradient echo (GRE-EPI) time series were realigned, motion corrected, slice time corrected, normalized to each rat's anatomical images and spatially smoothed with a 3D Gaussian kernel (0.6 × 0.6 × 1 mm<sup>3</sup> ). Within each analysis, the mean global intensities were mean scaled to an arbitrary value calculated within SPM8. The design model tested was a comparison between "off " and "on" conditions within each TGN stimulation paradigm. The paradigm was convolved with SPM's haemodynamic response function defined as a gamma-variate function and high passfiltered. Residuals for the realigned rat movement were taken into account by submitting the realignment parameters (translations and rotations) as regressors. T-maps were calculated on a voxel by voxel basis. Thresholding criteria of 5 adjacent voxels, each with a T-score > 3.0 were used to identify regions of interest. Only clusters comprising at least 5 voxels were considered significant (p < 0.0001, corrected for multiple comparisons.

With STIMULATE (University of Minnesota, Minneapolis, USA) (Strupp, 1996), regions of interest (ROIs) over the activated primary somatosensory barrel field cortex (S1BF) were drawn with respect to the Paxinos and Watson's Atlas (Paxinos and Watson, 1998). ROIs were delineated from the thresholded tmap of each rat and had the same size as VOIs for 1H-MRS. A representative average time-course was recorded for each animal. When needed baseline correction was performed. T-maps were overlaid on single shot gradient echo EPI images.

#### Correlations, Robustness, and Statistical Analysis

Correlations were performed between BOLD responses in thalamus and S1BF, and [Glu] and [GABA] measured at rest and during stimulation in thalamus and S1BF. In addition BOLD responses were also correlated to changes in [Glu] and in [Lac] in S1BF only. Only metabolites with Cramer-Rao lower bounds

(CRLB) under 30% were used for statistical analysis. To examine the association between BOLD responses and metabolites and between metabolites, the Shapiro-wilk test in SPSS 22 was performed for all the variables demonstrating p-values above 0.05 and therefore normality of each variable distribution.

The correlation analysis was performed within Origin (OriginLab version 9, Massachusetts, USA). Data were tabulated and scatter plots were obtained. Subsequently, a linear regression analysis was performed allowing to calculate the Pearson's r coefficient of correlation. Linear regression was peformed concomitantly with an ANOVA test with a threshold defined as 0.05. However, appropriate statistical analysis is required to perform multiple comparisons. With the conservative Bonferroni correction (Bretz et al., 2011) one would assume that all variables were independent which was not necessarily the case. Using a principal component analysis method within Origin, the number of effective comparisons (or eigenvalues) was determined as decribed by Cheverud et al. (1983). The adjusted threshold after Bonferroni correction was therefore p = α/4 with α = 0.05. Uncorrected p-values were denoted as P and were reported as correlating positively or negatively for P < 0.05. These values were comparable to values reported in the literature for a similar sample size (**Table 3**).

In order to examine the robustness of the correlation analysis performed in the present study, we further performed a Spearman- Rho correlation analysis within SPSS 22 since for small sample size data sets, a permulation analysis is automatically performed (Bretz et al., 2011). Finally, multiple linear regression analyses were performed in SPSS 22 for a sample size of 15 animals and for a sample size of 22 animals where the animal number was increased by adding 7 animals with CRLBs below 40% for Glu and GABA, since this type of regression analysis is valid for sample sizes of at least 20 subjects. Using a Kolmonorov-Smirnov test, the multivariate normality was checked again. To avoid over-fitting, a stepwise model analysis was used together with F statistics. The validity of the regression analysis was assessed and a Durbin-Watson test was included to check for autocorrelation and included a colinearity diagnostic. Homoscedasticity and normality of residuals were tested using standardized plots. To test the robustness of the multiple linear regression analysis Cook's and leverage tests were performed.

#### RESULTS

#### Metabolic Responses to Prolonged TGN Stimulations

The neurochemical profiles of the thalamus and the barrel cortex before and during TGN stimulation were obtained successively for each rat. High spectra SNR levels were reproducibly measured in S1BF (91 ± 8) and in the thalamus (62 ± 6).GABA, Glutamate, Glutamine, Glucose, and Lactate levels were measured at rest and during stimulation periods both in the thalamus (n = 12) and in S1BF (n = 15) in VOI represented in **Figure 2A**. In order to increase SNR levels for thalamic MRS acquisitions, the thalamic voxel encompassed several thalamic nuclei of interest [VPM, Reticular Thalamic nucleus (RTN), and Posteromedian Thalamic Nucleus (POm) that can be recognized by reference to the labeled map. Examples of labeled spectra (mean spectra averaged over rat population and over time)] in both structures acquired during stimulation and rest periods are shown (**Figure 2B**). Increased and decreased levels of Glu and Lac levels during TGN stimulations can be visualized in each structure respectively. **Figures 2C,D** illustrate positive correlations between BOLD changes and the percent change in amplitude of NAA and PCr+ Cr peaks respectively. As a consequence of BOLD effect, the linewidths of these peaks should decrease while the peak height should increase. No relationship was found in the thalamus. The neurochemical profiles (± standard error of the mean) are depicted (**Figure 3**) for each structure and each condition. In the present work, Glutamate and GABA concentrations were compared as well as Glucose and Glutamine in 15 rats (1 rat demonstrated lipid contamination during stimulation and was discarded from further analysis). In the barrel cortex, Lac, Glu, GABA levels were significantly higher during stimulation (p = 0.0015, p = 0.015, and p = 0.021 respectively, one-way ANOVA and Bonferroni correction). In the thalamus, Gln was significantly higher during TGN stimulation (p < 0.009, one-way ANOVA and Bonferroni correction). Although not significantly due to high variability across the rat population, glucose levels decreased in S1BF and thalamus. Regional differences were demonstrated between thalamus and S1BF metabolites at rest and during stimulation using a two-way ANOVA test with Bonferroni correction. Significant differences between brain structures were found for Glu (p < 0.0001), Lac (p = 0.02), and GABA (p =

in the barrel cortex and the thalamus. GABA, Glutamate, Glutamine, Glucose, and Lactate levels were measured at rest and during long stimulation periods both in the thalamus (n = 11) and in S1BF (n = 15) in the represented voxels of interest. The thalamic voxel of interest (36 µl) for fMRS in rats is depicted. In order to increase SNR levels for thalamic MRS acquisitions, the thalamic voxel encompassed several thalamic nuclei of interest [VPM, Reticular thalamic nucleus (RTN) and Posteromedian thalamic nucleus (POm)] that can be recognized by reference to the labeled map. Cx, Cortex; Th, Thalamus; S1BF, Primary somatosensory barrel field; Hb, Habenula; Hip, Hippocampus; Cpu, Caudate Putamen; Amy, Amygdala; (B) Examples of labeled spectra (mean spectra averaged over rat population and over time) in both structures acquired during stimulation and rest periods are shown. Increased and decreased levels of Glu and Lac levels during TGN stimulations can be visualized in each structure respectively. (C) and (D) BOLD effects result in a decrease of NAA and tCr (PCr+Cr) peak linewidths (increase in T2\*). Changes in linewidths can also be reflected by a change in peak height reported here. Positive correlations (r = 0.68, p = 0.007 and R = 0.54, P = 0.046) were found between BOLD responses in S1BF and percent change of NAA and PCr+Cr peak height respectively. These changes may represent surrogate markers of BOLD responses but remain to be validated.

0.005) during rest and stimulation periods. Changes in [Glu] and [GABA] in both S1BF and thalamus were significantly related to changes related to activation (p = 0.02).

#### Relationships between Bold Responses, Glutamate and GABA in S1BF, and Thalamus

Although rats underwent sequential prolonged MRS measurements in S1BF and thalamus successively during rest and stimulation periods, BOLD responses were still detected in S1BF and in the thalamus respectively. Unexpectedly, a negative correlation (Pearson's r = −0.68, p = 0.0024) was found as shown in **Figure 4A**, between BOLD responses in S1BF and thalamus. In order to investigate further this relationship, correlations between BOLD responses, [Glu] and [GABA] were performed within each structure and between structures. BOLD changes in the barrel cortex were negatively correlated to resting GABA levels (Pearson's coefficient = −0.72, p = 0.003) (**Figure 4B**). Resting Glu levels correlated positively with BOLD changes (Pearson's r = 0.72; P < 0.03, **Figure 4C**) and changes in glutamate (1Glu) were negatively correlated to BOLD (Pearson's coefficient = −0.70; P = 0.024) (**Figure 4D**). A trend toward positive correlation between BOLD changes and Lac changes (1Lac) was also found but not significant (Pearson's r = 0.47; P < 0.08) (**Figure 4E**).

In the thalamus, resting or stimulated GABA levels were not correlated to thalamic BOLD responses. Resting S1BF GABA levels demonstrated a positive interaction with thalamic BOLD (Pearson's r = 0.67 and P = 0.016, **Figure 4F**). Resting thalamic Glu levels were not correlated with thalamic BOLD changes (Pearson's r = 0.57; P < 0.09) and with BOLD changes in the barrel cortex (Pearson's r = −0.55, P = 0.09). A negative correlation between S1BF BOLD and [Glu] during stimulation was also suggested (Pearson's r = 0.62 and P = 0.055, **Figure 4G)**.

Since the metabolic status of a structure may be predictive of its functional status including connectivity (Kapogiannis et al., 2013; Castro-Alamancos and Gulati, 2014), the relationships

between [Glu] and [GABA] within and between structures were also investigated. As the energetic status of the whole brain is tightly controlled at rest, changes due to neuronal activity also involve changes between related structures. As reported in **Table 1**, during TGN stimulation, there was no correlation between [GABA] and [Glu] whereas at rest, GABA and Glu levels in the barrel cortex were positively correlated (Pearson's r = 0.74; P = 0.014, **Figure 5A**) and resting and stimulated [Glu] in S1BF were positively correlated (**Figure 5B)**. On the other hand, resting thalamic GABA levels were negatively correlated to barrel cortex GABA levels measured during stimulation (Pearson's r = −0.69; P = 0.024) (**Figure 5D)** but not at rest (**Figure 5C)**. However, Glu levels in S1BF at rest strongly influenced stimulated thalamic GABA levels (**Figure 5D**; **Table 1**). In S1BF, resting and activated Glu levels were correlated (Pearson's r = 0.67; P < 0.05, **Figure 5E**).

#### Multiple Linear Regression Analysis and Robustness of the Analysis

**Table 2** presents the results of the Spearman-Rho correlation analysis. Results were in agreement with previous Pearson's correlation (p-values under 0.01 and 0.05).

Stepwise non-parameteric multiple linear regression analyis in a sample size of 15 animals revealed that resting Glutamate and GABA levels added significant explanatory power to the regression model. In addition, resting GABA and glutamate levels in the barrel cortex (BC) explained between 52.3 and 89% of the variance. The Durbin-watson test (D = 1.631) allowed to assume that there was no significant first order linear autocorrelation in the multiplelinear regression data. The F-Test is the test of significance of the multiple linear regression. The F-test of was highly significant (F = 19.063, p = 0.012 for Glurest and p = 0.001 for GABArest,) thus it can be assumed that there is a linear relationship between the variables in our model. Since we have multiple independent variables in the analysis the Beta weights compared the relative importance of each independent variable in standardized terms. We found that a regression model encompassing BOLD and resting Glutamate and GABA levels had the highest impact on the regression (β = 0.955 and 0.383). Multicollinearity in our multiple linear regression model was higher than MC = 0.811 showing that there was no suspiscion of multicollinearity. Q-Q plots indicated that in our multiple linear regression analysis there was no tendency in the error terms. Finally, Cook's and centered leverage mean values were 0.223 ± 0.4 and 0.273 ± 0.2 showing that exclusion of data points wouldn't have changed the regression statistics substantially while the centered leverage value demonstrated that there was no significant influence of specific datapoints. The same analysis was performed with a sample size of 22 animals showing that resting glutamate levels only added significant explanatory power to the regression model explaining 80% of the variance confirmed by highest Beta weights (β = 0.896). Again, there were no significant influence of autocorrelations and multicollinearity (D = 2.3, MC = 0.655). F-tests were again highly significant (F = 44.837, p <

0.0001). Finally due to the slight increase in sample size, Cook#s and centered leverage values were further decreased 0.077 ± 0.258 and 0.048 ± 0.045.

#### DISCUSSION

In the present work, using fMRS and BOLD fMRI measurements in the thalamus and barrel cortex of rats, the modulation of neuronal activity- represented by BOLD responses- within and between these connected structures by GABA and Glu neurotransmitters was addressed with correlations. To the best of our knowledge, the present study is the first attempt to measure and compare functional hemodynamic and metabolic responses upon sensory stimulation of related brain structures in rats using MR-based methods. Although as discussed below, a number of technical challenges remain to be solved to validate the methodology proposed, the approach is potentially powerful for multi-regional investigations during brain activity and a better understanding of the underlying relationships between neurotransmitters and functional neuroimaging signals as well as functional connectivity.

#### Confirmation of Cortical BOLD-Neurotransmitter Relations in the Activated Region

Our study confirmed the cortical BOLD responses dependence on resting GABA levels as well as changes in Glu and Lac in response to stimulation but revealed that the direction of the correlations was specific to the system under analysis. In agreement with previous results obtained in the visual human cortex (Northoff et al., 2007; Donahue et al., 2010; Muthukumaraswamy et al., 2012; Bednaˇrík et al., 2015), positive S1BF BOLD signals were negatively correlated to baseline S1BF GABA levels. In the human visual cortex, concomitant positive, and negative correlations with Cerebral Blood Flow (CBF) and


#### TABLE 1 | Glutamate and GABA concentrations correlations in S1BF and Thalamus at rest and during TGN Stimulations.


FIGURE 5 | Correlations between metabolites within and between the thalamus and the cortex. (A) At rest, GABA and Glu levels in the barrel cortex were positively correlated (Pearson's r = 0.74; P = 0.014). (B) In S1BF, Glu levels at rest and during stimulation were correlated (Pearson's r = 0.67; P < 0.05) (C,D). Thalamic GABA levels at rest were negatively correlated to barrel cortex GABA levels during stimulation (Pearson's r = −0.69; P = 0.024) but not at rest (Figure 4C). (E) Glu levels in S1BF at rest were strongly influenced by stimulated thalamic GABA levels.

#### TABLE 2 | Spearman–Rho correlation and statistics (\*\* p < 0.01; \*p < 0.05).


The Spearman-Rho coefficient of correlation between BOLD BC and 1Lac was 0.461 with p = 0.084 (n = 15).

vascular space occupancy (VASO) signals respectively, were interpreted as a greater influx of blood to the visual cortex region in participants with higher GABA levels in order to fuel the greater amount of excitatory activity required for overcoming the higher resulting inhibition. In S1BF, low [Glu]rest predicted lower BOLD responses whereas lower 1[Glu] predicted higher BOLD percent changes during TGN stimulation. Furthermore, 1[Lac] suggested a positive correlation with BOLD fMRI responses (Pearson's r = 0.47 P = 0.07). These findings were in line with the study by Bednaˇrík et al. (2015) where higher Glu changes (1[Glu]) as well as higher Lac changes (1[Lac]) predicted higher BOLD percent changes in the human visual cortex, interpreted as a response to increased energy demands of neuronal activation. Although, this study confirmed the existence of relationships between BOLD responses and both GABA at rest and Glu changes, the negative relationship with the latter was unexpected and difficult to interpret: If higher 1Glu levels not compensating for the inherently higher GABA levels in S1BF of some rats and therefore leading to lower BOLD responses could represent a reasonable interpretation, null 1Glu (at intercept) resulting in highest BOLD responses appeared contradictory to previous results. To date, the majority of fMRS studies reported weak mean 1[Glu] ranging between 2 and 4% due to cortical stimulation (Mangia et al., 2007; Lin et al., 2012; Schaller et al., 2014). In these studies, fMRS was conducted after carefully positioning the VOI over the core activated area in BOLD maps for which fMRI acquisitions were optimized. In the present study, BOLD responses were measured after fMRS acquisitions. The position of the VOI in S1BF relied on previous work (Just et al., 2013) possibly missing the S1BF activated core and thus shifting the amplitudes of vascular and metabolic responses. However, positive changes in amplitudes of NAA and tCr peaks as a consequence of BOLD changes were detected as in Just et al. (2013) suggesting that the choice of VOI was adequate. Moreover, averaging the voxels yielding the highest statistically significant S1BF BOLD responses also revealed a negative correlation (not shown) with 1[Glu] thus discarding effects related to thresholding. Weak changes in [Gln] in S1BF also corresponded to rats with highest BOLD responses and could not serve as an explanation. Bednaˇrík et al. (2015) also found a non-zero 1[Glu] intercept for BOLD = 0%. Although they point at different sensitivities of fMRI and fMRS signals and agree that larger populations need to be examined, their study and the present one point at other effects. In particular, the effects of other neuromodulators such as norepinephrine, noradrenaline, dopamine etc... also need to be taken into account in future studies and may explain these discrepancies (Fontanez and Porter, 2006; Qiao et al., 2007; Castro-Alamancos and Gulati, 2014). Moreover, the literature reported either positive (Enzi et al., 2012) or negative correlations between BOLD and baseline Glu concentrations, which depended on the health status of patients (Falkenberg et al., 2014) and were mostly inter-regional (Duncan et al., 2014). Therefore, the direction of correlation may be specific to the network under investigation and its physiological status under rest and stimulation. Present results also showed that BOLD changes were strongly dependent on resting state metabolic conditions in S1BF.

#### Thalamocortical Relationships between BOLD Responses [Glu] and [GABA]

During TGN stimulation, GABA and Glu levels were uncorrelated within each structure and between structures. On the other hand, resting [Glu] and [GABA] in S1BF were correlated suggesting a strong interplay between these neurotransmitters in S1BF to maintain the EIB at stable levels as would be expected without specific brain activity. Such a relationship was not found in the thalamus and attributed to a lack of specificity since the region of interest taken into account encompassed several thalamic nuclei in order to increase SNR levels for single voxel MR <sup>1</sup>H spectroscopy. Notably, the VOI emcompassed the reticular thalamic nucleus described as the only source of thalamic GABAergic inhibition. In addition, it receives inputs from the cerebral cortex but does not project to the cerebral cortex. The inclusion of this nucleus may explain differences between thalamic and cortical neurochemical profiles but also uncorrelated thalamic BOLD responses and resting thalamic GABA levels. Many studies suggested that RTN plays a significant role in thalamic gating, facilitating relevant stimuli and inhibiting others by increased inhibition of the information transfer via other relay nuclei. The shape and size of RTN make this important nucleus very difficult to avoid. Other techniques such as proton MR spectroscopic imaging (MRSI) (Seuwen et al., 2015) may be more adequate for future investigations while fMRS techniques coupled to optogenetic stimulation may help unraveling the exact contribution of this structure (Jurgens et al., 2012).

A significant negative correlation between BOLD responses measured in both S1BF and thalamus was found. In the literature, weak or absent thalamic BOLD responses with regards to forepaw or barrel cortex activations were reported in the past (Esaki et al., 2002; Keilholz et al., 2004; Zhao et al., 2008; Mishra et al., 2011; Devonshire et al., 2012). Thalamic BOLD responses are usually expected to be weaker than cortical ones due to generally higher GABA levels linked to higher densities of GABAergic neurons (Mishra et al., 2011; Devonshire et al., 2012; Tiwari et al., 2013). Here, accordingly, thalamic GABA levels were higher than in S1BF at rest and during stimulation but there was no significant increase of mean GABA levels in the thalamus upon TGN stimulation. Moreover, thalamic BOLD responses and GABA levels were uncorrelated. While these findings may be attributed to the inclusion of RTN, rats with higher resting S1BF GABA levels had high amplitude BOLD responses in the thalamus correlated to low amplitude BOLD responses in S1BF. Moreover, [Glu]rest in S1BF were negatively correlated to [GABA]stim in the thalamus. Therefore, rats with low Glu levels and high resting GABA levels in S1BF demonstrated lower BOLD responses in S1BF. Consequently, the release of high GABA concentrations in the thalamus could be interpreted as a way to overcome the higher excitatory activity supported by significantly increased thalamic Gln levels during TGN stimulation and thus balancing the decreased Glu levels and inducing strong BOLD thalamic activation. In the thalamus, [Gln] increased significantly during TGN stimulation while [Glu] decreased without significance. Interestingly, identical changes were reported by Xu et al. (2005) in the forepaw cortex of rats and were attributed to an augmented release of Glu and a rapid conversion into Gln through the Glu-Gln cycle between glutamatergic neurons and astrocytes. (Hirata and Castro-Alamancos, 2010) suggested that neuromodulators can control the synaptic activity of thalamocortical and corticothalamic cells and therefore control the state of their respective targets without being released there directly, thereby ensuring that both structures are under specific conditions essential for a proper connectivity. This is furthermore supported by studies suggesting the control of the synchrony of neuronal activity by regional EIB (Kapogiannis et al., 2013) and its correlation with functional connectivity. Therefore, the present study confirmed the interest in investigating relationships between neurotransmitters within the thalamus and the barrel cortex and between these two regions as potential predictors of functional connectivity, which could be of interest for studying pathologies such as absence epilepsy or schizophrenia. Duncan et al. (2014) in their review pointed at the fact that most studies looking at the roles of GABA and glutamate in brain function remain at the correlational stage while most of them include small numbers of subjects thus limiting the validation of findings. In the present work, functional connectivity was conducted on 8 rats only and showed high variability across this population (**Figures S1**, **S2**). At this stage, findings remain only indicative and should be further validated with more robust resting-state connectivity analysis. **Figure 6** illustrates a preliminary schematical overview of the Glutamate and GABA modulatory effects and functional connectivity on cortico-thalamo cortical loop in a healthy rat.

# LIMITATIONS AND PERSPECTIVES Statistical Analysis and Multiple Comparisons

In the text, correlations between either BOLD responses and metabolites or between metabolites were reported as being significant for uncorrected p-values. Upon application of corrected p-values as stated in Materials and Methods most correlations did not pass significance level which was attributed to the small number of animals included in the study. As a consequence of this, the correlations reported here point at interesting effects that remain to be validated. Nevertheless, to date, most of the correlation studies involving BOLD-metabolite relationships were not corrected for multiple comparisons. **Table 3** summarizes various results from these studies with Pearson's correlation coefficients, p values and subject numbers. The column "statistics" mentions the statical test applied to the correlation. Unfortunately, these studies were mainly performed in humans. The number of subjects taken into account was in general similar to the number of rats used in the present study while significant negative correlation of brain activity measures with resting GABA were consistently

found across studies including this one but not correlations with Glu levels. Although significance levels were not always adjusted, all these studies demonstrate that BOLD variability can be explained by GABA levels at least in cortical areas and the negative relationship between BOLD-resting GABA is a hallmark of neuronal activity. Here, the robustness of the correlation study was further confirmed by the Spearman-Rho correlation analysis and the multiple linear regression analysis. Although we delibaretely added Glutamate and GABA levels estimates with lower reliability (CRLB <=40%) from seven animals, we found that F statistics were significant while calculation of distance values (Cook's and leverage) increased the robustness of the analysis compared to a lower sample size of 15 animals. In addition, these analyses emphasized the roles of S1BF resting glutamate and GABA levels. The multiple linear regression analysis further demonstrated the need for higher sample sizes to validate metabolic correlations in rats but also pointed at the need for improved methods for associating MRS values as the interpretation of multiple linear regression models with an increased number of variables becomes extremely complicated.

### Study Design

One of the main pitfalls of the present study was its sequential design that unfortunately did not allow simultaneous assessment of BOLD, thalamic and S1BF metabolites. Using magnetic MRSI techniques such as FIDLOVs (Seuwen et al., 2015) which would at the same time contribute to preserve SNR levels and time may help while there is definitely a need for increased targeting at smaller structures with higher SNR levels for an increased specificity of studies as mentioned earlier for RTN or in pathologies (Pan et al., 2015). Finally, BOLD responses

#### TABLE 3 | BOLD-metabolites relationships in various human studies.


may be measured directly from fMRS acquisitions using changes in linewidths of NAA or total Creatine peaks as a result of activation which would also decrease scanning time. Again these measurements require high SNR levels per subject and require validation against standard BOLD acquisitions. In order to increase SNR levels for the temporal assessment of metabolite changes during TGN stimulation (not presented here), more fids were acquired increasing significantly the scanning time. Adaptation/habituation effects may therefore have been induced in the present study and will certainly need to be taken into account in future investigations.

#### Impact of Anesthesia

In the present study, fMRS findings in the cortex of rats under α-chloralose anesthesia were similar to findings obtained in humans (Schaller et al., 2014; Bednaˇrík et al., 2015). BOLD-GABA and BOLD-1Lac correlations had similar directions. Nevertheless, the impact of anesthesia on the modulatory effects of Glu and GABA on thalamic and S1BF BOLD responses as well as resting-state functional connectivity cannot be neglected (Williams et al., 2010).

# CONCLUSION

In rats under α-chloralose anesthesia, barrel cortex fMRS findings and association of these findings with BOLD responses were in accordance with recent results obtained in the human visual cortex. These results suggested consistent regulatory roles of both glutamate and GABA on cortical neural activity across species. Within the thalamus, neurochemical profiles during TGN stimulation differed from those obtained in the barrel cortex. The correlation study conducted within and between cortical and subcortical structures suggested a complex interplay between glutamatergic and GABAergic modulations of the cortico-thalamo-cortical loop. In addition, preliminary results suggested regulation of functional connectivity between thalamus and cortex by excitatoryinhibitory neurotransmitters. These results will need to be followed up with larger population sizes to demonstrate their validity.

To the best of our knowledge, this is the first study investigating two interconnected brain regions during both rest and activation periods using fMRS in rodents. Although in its early stage, the proposed methodology allowed investigating the different contributions of excitatory and inhibitory neurtransmitters on neuronal activity indirectly measured with BOLD in the thalamus and barrel cortex, which could benefit investigations of diseases affecting thalamo-cortical neurotransmission (Autism, Parkinson's disease, Gilles de la Tourette syndrome, cardiac dysfunction...).

#### AUTHOR CONTRIBUTIONS

NJ conceived and designed the project. NJ performed experiments and analyzed the data. NJ wrote the manuscript. SS contributed to the discussion and revised the manuscript.

#### FUNDING

This study was supported by the Centre d'Imagerie BioMédicale (CIBM) of Ecole Polytechnique Fédérale de Lausanne (EPFL), the University of Lausanne (UNIL) and the Foundations Leenards et Jeantet. SS was funded by a National Competence Center Biomedical Imaging grant (NCCBI).

#### REFERENCES


#### ACKNOWLEDGMENTS

The author would like to thank Prof. Rolf Gruetter for providing the necessary tools for this study.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fphys. 2017.00030/full#supplementary-material

Figure S1 | Resting-state BOLD fMRI. (A,B) BOLD T map overlaid over gradient echo EPI and showing bilateral resting-state BOLD both in the barrel cortex (A) and the thalamus (B) during long stimulation periods (2 min OFF–10 minON–2 minOFF) and short stimulation periods (30 s OFF–30 s ON...). (C). Correlations between left and right hemisphere time courses for bilateral seeds were increased after long stimulations in individual animals but did not reach significance at the population level.

Figure S2 | Influence of [Glu] and [GABA] during rest and stimulation on cortico-cortical and thalamo-thalamic functional connectivities. (A,B) Highest positive Pearson's correlation coefficients in both contralateral and ipsilateral S1BF (r = 0.57–0.67), (A) were found between stimulated Glu levels and T-values while cluster numbers were mainly correlated to GABA levels at rest (B). (C) Thalamo-thalamic FC was negatively correlated to stimulated GABA levels (D–G) At rest, thalamic GABA levels were positively correlated to cluster numbers (r = +0.57. (D) A positive relationship between stimulated thalamic Glu levels and S1BF T-values (r = 0.69). (E) was observed but a negative one was observed between stimulated thalamic GABA levels and both T-values and Cluster numbers (r = −0.57, −0.56 respectively) (F,G). (H) Negative correlation between cortico-thalamic FC and glutamate at rest.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Just and Sonnay. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Neonatal Hypoxia Ischaemia: Mechanisms, Models, and Therapeutic Challenges

#### Lancelot J. Millar<sup>1</sup> \*, Lei Shi1,2, Anna Hoerder-Suabedissen<sup>1</sup> and Zoltán Molnár<sup>1</sup>

<sup>1</sup> Molnár Group, Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, UK, <sup>2</sup> JNU-HKUST Joint Laboratory for Neuroscience and Innovative Drug Research, College of Pharmacy, Jinan University, Guangzhou, China

Neonatal hypoxia-ischaemia (HI) is the most common cause of death and disability in human neonates, and is often associated with persistent motor, sensory, and cognitive impairment. Improved intensive care technology has increased survival without preventing neurological disorder, increasing morbidity throughout the adult population. Early preventative or neuroprotective interventions have the potential to rescue brain development in neonates, yet only one therapeutic intervention is currently licensed for use in developed countries. Recent investigations of the transient cortical layer known as subplate, especially regarding subplate's secretory role, opens up a novel set of potential molecular modulators of neonatal HI injury. This review examines the biological mechanisms of human neonatal HI, discusses evidence for the relevance of subplate-secreted molecules to this condition, and evaluates available animal models. Neuroserpin, a neuronally released neuroprotective factor, is discussed as a case study for developing new potential pharmacological interventions for use post-ischaemic injury.

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Alla B. Salmina, Krasnoyarsk State Medical University, Russia Gunnar Naulaers, KU Leuven, Belgium

#### \*Correspondence:

Lancelot J. Millar lancelot.millar@univ.ox.ac.uk

Received: 31 October 2016 Accepted: 07 March 2017 Published: 08 May 2017

#### Citation:

Millar LJ, Shi L, Hoerder-Suabedissen A and Molnár Z (2017) Neonatal Hypoxia Ischaemia: Mechanisms, Models, and Therapeutic Challenges. Front. Cell. Neurosci. 11:78. doi: 10.3389/fncel.2017.00078 Keywords: neonatal, hypoxia-ischemia, encephalopathy, subplate, neurodevelopment, neuroserpin, neuroprotection

## INTRODUCTION: GLOBAL CLINICAL IMPACT OF NEONATAL HYPOXIA ISCHAEMIA

The clinical definition of neonatal HI injury is "asphyxia of the umbilical blood supply to the human fetus occurring at 36 gestational weeks or later" (Perlman, 1997, 2006; Volpe, 2001, 2012; Shah P.S. et al., 2006). Neonatal HI is synonymous with hypoxic-ischaemic encephalopathy (HIE) occurring in the term infant, where term is defined as 36 gestational weeks or later. This review addresses neonatal HI or HIE, any results concerning perinatal hypoxic-ischaemia injury will be clearly indicated in the text. This disorder encompasses a large range of physiological origins

**Abbreviations:** AMPA, alpha-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid; BBB, blood brain barrier; CPR, cardiopulmonary resuscitation; CNS, central nervous system; CSF, cortico-spinal fluid; CSPG, chondroitin sulfate proteoglycan; CTGF, connective tissue growth factor; E(18), embryonic day (18); EM, electron microscopy; GABA, γ-aminobutyric acid; HI, hypoxia-ischaemia; HIF-1, hypoxia inducible factor 1; IL-1ß, interleukin-1ß; IL-6, interleukin-6; IL-9, interleukin-9; LPS, lipopolysaccharide; MRI, magnetic resonance imaging; MRS, magnetic resonance spectroscopy; NMDA, N-methyl-D-aspartate; OGD, oxygen glucose deprivation; P(8), postnatal day (8); poly(I:C), polyinosinicpolycytidylic acid; rER, rough endoplasmic reticulum; TNF-α, tumour-necrosis factor-α; tPA, tissue plasminogen activator; VEGF, vascular endothelial growth factor.

and clinical outcomes (Volpe, 1995). The diagnostic criteria for neonatal HI are based on a set of markers demonstrated to correlate with clinical outcome (Finer et al., 1981; Perlman, 1997; Richer et al., 2001). These include: 5-min Apgar score of less than 5 (Levene et al., 1986; Ruth and Raivio, 1988; Laptook et al., 2009); need for delivery room intubation or CPR (Richardson B.S. et al., 1996; Salhab et al., 2004a; Shalak and Perlman, 2004; Kattwinkel et al., 2010); umbilical cord arterial pH less than 7.00 (Ruth and Raivio, 1988; Perlman and Risser, 1993; Robertson N.J. et al., 2002; Salhab et al., 2004b); and abnormal neurological signs, such as hypotonic muscles or lack of sucking reflex (Levene et al., 1986; Robertson et al., 1989; Richer et al., 2001). Electroencephalography (EEG) has also proved helpful as a predictor of clinical outcome (reviewed in Walsh et al., 2011; van Laerhoven et al., 2013). Amplitude-integrated EEG (aEEG) in particular, a filtered time-compressed continuous one- or two-channel read-out, has been demonstrated a reliable predictor in meta-analyses up to 5 years after birth (al Naqeeb et al., 1999; Sinclair et al., 1999; Toet et al., 1999; Biagioni et al., 2001; Shah D.K. et al., 2006; Nash et al., 2011; Murray et al., 2016; Weeke et al., 2016), however, some report that aEEG remains less reliable than MRI, especially following hypothermia treatment (Doyle et al., 2010; Weeke et al., 2016). The Thompson score, an EEG measure of predictive neurodevelopment, is likely to remain useful to clinicians (Thompson et al., 1997; Horn et al., 2016). This is by no means an exhaustive list of risk factors and signs of clinical concern that can occur during the early postnatal period (Shalak and Perlman, 2004; Hagberg et al., 2015).

Neonatal HI is the most common cause of death and disability in human neonates (Grow and Barks, 2002; Ferriero, 2004; Shalak and Perlman, 2004), accounting for 23% of infant mortality worldwide, and affecting 0.7–1.2 million infants annually (Lawn et al., 2005; see **Figure 1**). In developed countries, incidence of HI injury has not decreased in the past two decades (Himmelmann et al., 2005; Vincer et al., 2006), remaining a significant cause of fatality and disability. The frequency of motor and cognitive disorders linked to perinatal and early postnatal brain injury actually increased during the 1990s, and currently remains stable (Vincer et al., 2006; Robertson and Iwata, 2007; Wilson-Costello et al., 2007). Progress in assisted respiratory and intensive care technology has led to greater than 90% survival of infants born after gestational week 23 (Larroque et al., 2004; Fellman et al., 2010), perhaps accounting for the increased burden of disability within the population as mortality rates have decreased substantially.

Amongst those who survive the initial injury, rates of disability remain high throughout life. Of patients surviving neonatal HI, 5–10% of infants demonstrate persistent motor deficits, and 20– 50% display sensory or cognitive abnormalities that persist to adolescence (Hack et al., 1992; Vohr et al., 2000; Volpe, 2001, 2012; Lee et al., 2013). A meta-analysis of seven studies including 386 infant patients investigated the average incidence of mortality and morbidity: 5.9% of all patients across all studies died, 16.3% suffered neonatal seizures, and 17.2% experienced neurological deficits, with 14.2% qualifying for a diagnosis of cerebral palsy (Graham et al., 2008). Long-term outcomes in neonatal asphyxia infants have also been investigated. One meta-analysis found that 1–18% of patients were identified as having severe sensorimotor or learning disorders by the age of 2–5 years, with only 50–60% of patients reported as developmentally normal (Dilenge et al., 2001). This study covered a wide range of injury severities and follow-up ages. Disorders included seizures, hearing and vision loss (Robertson and Finer, 1985), language disorders, microcephaly, and muscle spasticity (Shankaran et al., 1991, 2012). Studies also report more severe neurological signs in patients suffering severe HI compared to those with milder HI injury (Robertson and Finer, 1985). Yet, many individuals who showed abnormal neurological signs at birth were normal at 2 year follow-up (De Souza and Richards, 1978). Clinical features and outcomes of neonatal HI are summarized in **Figure 1**.

Despite the high disability burden associated with surviving neonatal HI patients, there are very few preventative or protective treatments available for infants suspected to have suffered an HI event. The only licensed treatment currently available is hypothermia. This treatment involves subjecting either the infant's whole body or head-only to temperatures of around 33◦C (Choi et al., 2012; Tagin et al., 2012). Hypothermia was first demonstrated to improve survival in cases of cardiac arrest (Bernard et al., 2002; Nikolov and Cunningham, 2003), and has since been applied as a neuroprotective treatment in acute neonatal HI injury patients (Gunn et al., 1997, 2005; Gunn and Gunn, 1998; Shankaran et al., 2002, 2005; Eicher et al., 2005; Gluckman et al., 2005; Jacobs et al., 2005, 2013; Shah et al., 2007; Azzopardi et al., 2009; Shah, 2010). A recent meta-analysis found that hypothermia carried out in over 1,200 infants reduced the rate of death and neurological handicaps at 18 months followup across all severity categories of neonatal HI injury (Tagin et al., 2012). However, hypothermia alone is not sufficient to prevent all brain injury or neurological symptoms, highlighting the need for additional therapies to use in conjunction. Xenon gas administration is currently being trialed as an additive therapy alongside hypothermia (Hobbs et al., 2008; Thoresen et al., 2009; Johnston et al., 2011). There are currently no other licensed treatments available for neonatal HI.

2008; Lee et al., 2013).

Despite the efficiency of hypothermia (Tagin et al., 2012; Jacobs et al., 2013; Pauliah et al., 2013; Srinivasakumar et al., 2013; Kracer et al., 2014; Kali et al., 2015), death and disability remain a common feature of neonatal HI prognosis. Observations concerning global prevalence and poor long-term outcome reiterate the urgency of finding novel neuroprotective treatments for use during and directly following HI injury. This review examines the neuropathology resulting from neonatal HI injury in humans. The review then examines currently available animal models of neonatal HI and summarizes the strengths and weakness of such models for research into this complex human condition. Finally, the review will detail a potential approach toward identifying new pharmacological targets for neonatal HI therapies, focusing on the protein neuroserpin.

## NEUROBIOLOGY OF NEONATAL HYPOXIA ISCHAEMIA IN HUMANS

There is ample evidence that brain damage occurs in human neonatal HI patients with poor clinical outcomes, documented in both imaging and histopathological studies (Mortola, 1999; Volpe, 2008, 2012; Northington et al., 2011). Neuropathology has been characterized in human post-mortem studies, concluding that most areas of the brain are vulnerable to some extent to neonatal HI injury (Miller et al., 2005; Triulzi et al., 2006; Billiards et al., 2008; Cohen and Scheimberg, 2008; de Vries and Groenendaal, 2010; Volpe, 2012). Gray and white matter lesions have been described at term after HI (Eken et al., 1994; Marin-Padilla, 1996, 1997, 1999). Localization and extent of neuropathology has been shown to be associated with neurodevelopmental symptoms, giving insights into the nature of disability the patients may present with.

#### Neuroanatomy: Structural Imaging

Infants who survive the initial HI insult display cerebral damage visible with structural imaging. Magnetic resonance imagining (MRI) studies of term infants with neurological signs and combinations of fetal distress, cord acidemia, and depressed Apgar scores have been carried out in over 1,000 infants (reviewed in Volpe, 2012). These studies report great variation in the anatomical areas involved between individual patients, yet most samples described either patients demonstrating predominant or substantial injury to cerebral cortex (Barkovich et al., 1995; Rutherford et al., 2004; Miller et al., 2005; Chau et al., 2008, 2012, 2013; Li et al., 2009), or basal ganglia and thalamus (Barkovich et al., 1995; Rutherford et al., 1998, 2004; Cowan et al., 2003; Kaufman et al., 2003; Miller et al., 2005; Chau et al., 2008) in partially overlapping subpopulations. These two patterns of injury are shown in **Figure 2**. Cerebral white matter has been described as selectively sensitive to term HI injury (Inder et al., 1999; Craig et al., 2003). Although less common, severe selective involvement of subcortical white matter has been documented (Neil et al., 2002; Vermeulen et al., 2003). One review described the literature concerning structural MRI scans in the acute phase (within 2 weeks after birth): approximately 15–30% of scans were normal, lesions in basal ganglia and thalamus are present in 40– 80% of cases, with abnormalities of watershed white matter and cortex present in 40–60% of patients (Volpe, 2012). MRI anatomy has been shown to agree well with post-mortem studies (Cowan et al., 2003). Therefore, no single area of the brain is specifically damaged following neonatal HI. Any future treatments should take this diversity into account and provide neuroprotection to neurons throughout the brain.

The MRI scan is currently the method of choice for investigation of neonatal anatomy in both clinical and experimental circumstances (Perrin et al., 1997; Ment et al., 2002; Li et al., 2009). Diffusion-weighted MRI imaging has greatly improved identification of the time of onset of brain lesions (L'Abee et al., 2005; Chau et al., 2009). A reduced diffusion coefficient can be calculated, showing restricted diffusion during the first few days after the insult, with pseudonormalization by the end of the first week (McKinstry et al., 2002; Malik et al., 2006; Liauw et al., 2009). Sequential imaging has shown that lesions in the basal ganglia may increase in size during the first week after birth (Soul et al., 2001; Barkovich et al., 2006), and asymmetric diffusion within white matter has been correlated with clinical severity of hemiparesis (Glenn et al., 2007). Cranial ultrasound also remains a valuable clinical tool (Daneman et al., 2006). Another technique promising to add to understanding of neonatal HI is magnetic resonance spectroscopy (MRS), which allows brain metabolism to be imaged in real time (Kemp and Radda, 1994; Soares and Law, 2009). Full-term neonates with perinatal asphyxia have been studied, indicating that brain metabolism becomes abnormal after 6 to 12 h only to decrease even further after 24 h (Wyatt et al., 1989; Moorcraft et al., 1991; Roth et al., 1992). This coincided with clinical deterioration such as development of seizures. The concept of a delayed metabolic abnormality or 'secondary energy failure' has been elaborated in animal models (Lorek et al., 1994; Penrice et al., 1997; Groenendaal et al., 2006). Using MRS in these animal models, neuroprotective strategies could be tested. By using MRS data as real-time measurements of decreased brain metabolism in brain injury. <sup>1</sup>H-MRS and <sup>31</sup>P-MRS studied have demonstrated metabolic abnormalities following HI insult, which may persist for weeks (Groenendaal et al., 1994; Robertson et al., 1999). Unfortunately, with the magnetic field strength of current clinical systems, only large brain areas can be examined, limiting the use of this technique. Infants who have suffered neonatal HI often exhibit abnormal EEG activity (reviewed in Walsh et al., 2011; van Laerhoven et al., 2013). A range of abnormalities have been described, including: low voltage in isoelectric EEG (Finer et al., 1983; Legido et al., 1991), mild voltage depression (Watanabe et al., 1980; Toet et al., 2002; Murray et al., 2010), and asymmetry of trace (Aso et al., 1989; Zeinstra et al., 2001; Murray et al., 2009), although all of these criteria have been differently defined by different analysts (Rennie et al., 2004; Shellhaas et al., 2007). However, imaging techniques are constantly improving.

Lesions occur in many clinical patients, yet the effect on cognitive function is as diverse as the neuroanatomy. The area of cortex and basal ganglia damaged during the initial HI injury is directly predictive of language and motor outcome in childhood (Steinman et al., 2009; Martinez-Biarge et al., 2011). Examination

of diffusion-tensor imaging of neonates was predictive of survival and motor outcome (Hunt et al., 2004; Ward et al., 2006). Children with basal-ganglia-thalamus pattern of injury tend to be severely disabled due to dyskinetic cerebral palsy and epilepsy (Himmelmann et al., 2007). Infants with predominant watershed white matter and cortex lesions have more prominent cognitive than motor deficits (Miller et al., 2005; Gonzalez and Miller, 2006; Steinman et al., 2009). Severe motor impairment is uncommon, and this group is often considered to have a normal outcome when seen at 12–18 months, although suboptimal head growth, behavioral problems, epilepsy, and a delay in language emerge during late childhood (Mercuri et al., 2000; Miller et al., 2002; Oguni et al., 2008; Sato et al., 2008; Steinman et al., 2009). Therefore, cortical damage appears to be relevant to functional outcome in surviving neonatal HI patients. This information could be used to predict future susceptibility to disability before it manifests, allowing social and educational supports to be put in place early.

#### Molecular Mechanisms of Cell Death

Anatomical studies describe loss of brain volume following moderate and severe neonatal HI. However, the underlying molecular mechanisms responsible for cell death are debated (McLean and Ferriero, 2004; Fatemi et al., 2009; Northington et al., 2011; Baburamani et al., 2012). Many pathways have been implicated in HI injury in the term brain, primarily: excitotoxicity, oxidative stress, and inflammation. Molecular studies have drawn attention to a fact essential for the development of successful new therapies that the neonatal brain and its injury is fundamentally different from that seen in adult HI stroke injury (McLean and Ferriero, 2004; Johnston et al., 2011; Baburamani et al., 2012; Semple et al., 2013).

There are many important differences between neonatal HI and adult ischaemic stroke. For example, severe HI events in the infant brain can lead to liquifactive disintegration, not seen after adult stroke (Larroche, 1977; Rorke, 1992). Newly formed blood vessels are fragile and prone to rupture (Trommer et al., 1987; Volpe, 1989; Ment et al., 1991; Jones et al., 2002), and surrounded by fewer astrocyte end-feet (El-Khoury et al., 2006). Another key site of difference is the BBB. Studies in rodents indicate that the BBB is compromised as a result of neonatal HI (Muramatsu et al., 1997; Svedin et al., 2007; Ferrari et al., 2010; Tu et al., 2011; Yang et al., 2012). Yet the common belief that the neonate BBB is less effective has recently come under revision (Saunders et al., 1999, 2012, 2014; McLean and Ferriero, 2004; Baburamani et al., 2012; Stolp et al., 2016). Tight junctions, the occlusive element of the BBB, are present as soon as embryonic vessels invade the brain (Schulze and Firth, 1992; Bauer et al., 1993; Stewart and Hayakawa, 1994; Kniesel et al., 1996), and are functional (Ek et al., 2003, 2006; Daneman et al., 2010). In a model of hypoxia in newborn piglet, BBB integrity was maintained (Stonestreet et al., 1992), yet other experiments have demonstrated damage to the BBB following neonatal HI (Alvarez-Diaz et al., 2007; Leonardo and Pennypacker, 2009).

Cerebrovascular autoregulation is another factor which must be considered in neonates. The concept that preterm infants have a 'pressure passive' cerebral circulation is widely accepted. However, sick term infants demonstrate impaired autoregulation (Pryds et al., 1990; Hardy et al., 1999; Boylan et al., 2000) and the range of blood pressure over which cerebrovascular autoregulation functions expands with maturity (Tuor and Grewal, 1994; Verma et al., 2000). Also, the concentrations and actions of various signaling molecules is different in the developing brain including; caspase-3 (Cheng et al., 1998), VEGF (Carmeliet and Storkebaum, 2002), and HIF-1 (Iyer et al., 1998), among others (reviewed in Baburamani et al., 2012).

One surprising difference is sexual dimorphism in response to neonatal HI. Male babies are at higher risk of cerebral

palsy than females (Jarvis et al., 2005). Cognitive and motor outcomes are worse in male than in female low birth weight infants (Johnston and Hagberg, 2007). Quantitative imaging shows that male premature infants are more vulnerable to white matter injury, whereas females are more vulnerable to gray matter injury (Thompson et al., 2007). This sex difference has also been replicated in rodent in vitro models of hypoxic cell death (Zhu et al., 2006; Nijboer et al., 2007; Du et al., 2009). Although many molecular mechanisms are currently under investigation, this sexual dimorphism remains largely unexplained (Hill and Fitch, 2012; Chavez-Valdez et al., 2014; Demarest et al., 2016a,b; Waddell et al., 2016). Therefore, the unique state of the developmental brain should always be at the forefront of the researcher's minds.

Neonatal HI injury evolves over time (McLean and Ferriero, 2004). Injuries seen with MRI scans within the first few hours after asphyxia are subtle, restricted diffusion typically starting as small lesions in the putamen and thalami, progressing over the next 3 to 4 days to involve more extensive areas of the brain (Takeoka et al., 2002). Within the first few hours, regionally specific fluctuations in blood flow trigger excitotoxicity, free radical generation, and edema (Wigglesworth and Pape, 1978; Bennet et al., 1998; Jensen et al., 1999; Shalak and Perlman, 2004; Ferrari et al., 2010). A secondary phase of injury occurs during the following hours and days, resulting in neuroinflammation, mitochondrial permeabilization, and loss of cerebral autoregulation (Inder and Volpe, 2000; Hamrick and Ferriero, 2003; Scheepens et al., 2003; Hagberg et al., 2009; Leonardo and Pennypacker, 2009). A tertiary phase of brain injury has been proposed, which may exacerbate injury through persistent inflammation (Fleiss and Gressens, 2012).

The balance between molecular cell-death processes which cause this damage in neonatal HI remains debated. Early evidence indicates that the majority of cell death in neonatal HI is necrotic, however, all regions also undergo increased apoptotic death (Edwards and Mehmet, 1996; Edwards et al., 1997; Northington et al., 2001). Some studies suggest a more prominent role for apoptosis (Hill et al., 1995; Sidhu et al., 1997; Pulera et al., 1998; Hu et al., 2000; McLean and Ferriero, 2004). Immature neurons in vitro are more susceptible to apoptotic death than mature neurons (McDonald et al., 1997). Others report that necrosis is the major cellular pathology in humans and animals (Adamsons and Myers, 1973; Myers, 1975; Towfighi et al., 1995; Northington et al., 2001, 2005, 2011; Folkerth, 2005; Carloni et al., 2007; Stridh et al., 2013). Yet others recognize that both occur. Some report that necrosis predominates in severe cases, whereas apoptosis occurs in milder injury (Stroemer and Rothwell, 1998; Daval and Vert, 2004; Fatemi et al., 2009). Neurons often display morphologic features along an apoptosis-necrosis continuum (Portera-Cailliau et al., 1997a,b; Nakajima et al., 2000; Northington et al., 2007, 2011). In addition to apoptosis and necrosis, some neurons in the neonatal HI brain undergo autophagy (reviewed in Klionsky and Emr, 2000; Northington et al., 2011; Balduini et al., 2012). Neuronal autophagy occurs in rodent neonatal HI models (Lockshin and Zakeri, 1994; Carloni et al., 2008; Ginet et al., 2009). However, there is conflicting evidence as to whether the occurrence of autophagy augments brain damage (Koike et al., 2008; Puyal et al., 2009), or prevents the spread of necrotic cell death (Carloni et al., 2008). Artificially exclusive classification of cell death may hinder research and therapy development.

To add to this complexity, neonatal HI injury appears to activate several interacting molecular cascades. A simple schematic of the three major cascades is shown in **Figure 3**. The first is excitotoxicity, through which physiological glutamate neurotransmission leads to overactivation of postsynaptic receptors and cell death (reviewed in Hagberg et al., 1987; Choi, 1988, 1992; Hattori and Wasterlain, 1990; Danbolt, 2001). The N-methyl-D-aspartate (NMDA) receptor is relatively overexpressed in the developing brain (McDonald et al., 1989a; Represa et al., 1989; Fox et al., 1996). In P6 rats, the NMDA receptor is expressed at 150–200% of adult levels (Tremblay et al., 1988). The predominating combination of NMDA receptor subunits in the perinatal period seems to favor prolonged calcium influx for a given excitation (Danysz and Parsons, 1998). The same NMDA receptor that promotes plasticity can lead to massive Na<sup>+</sup> and water influx, cellular swelling, pathologically elevated intracellular calcium, and energy failure, leading to a 'spiral of death' (Choi, 1988). Oxygen glucose deprivation (OGD) in rat hippocampal neurons leads to a marked reduction in glutamate removal from the synapse (Jabaudon et al., 2000; Tao et al., 2001). Injection of NMDA into rat brain produces more extensive cell death in the neonate than in the adult (McDonald et al., 1988). Elevated glutamate has been documented in the cerebrospinal fluid (CSF) of infants who have suffered severe HI injury (Riikonen et al., 1992; Hagberg et al., 1993; Pu et al., 2008). The neonatal brain is much more prone to seizure activity than the mature brain (Holmes, 1991; Holmes and Ben-Ari, 2001), suggesting a prominent role for neuronal hyperexcitability and excitotoxicity although the molecular mechanisms behind this have not been fully elucidated (reviewed in Rakhade and Jensen, 2009). However, seizure activity could also be explained by paradoxical excitatory activity of the neurotransmitter gamma-amino butyric acid (GABA) in the developing brain (Staley et al., 1995). Drugs that block NMDA receptors are protective against HI injury in neonatal rodent models (McDonald et al., 1989a,b, 1990). Activation of AMPA receptors also contribute to injury (McDonald and Johnston, 1992; Deng et al., 2003; Talos et al., 2006), however, AMPA antagonists are not as protective (Ikonomidou et al., 1999; Noh et al., 2006). These findings are yet to be exploited in human clinical trials, as the integral role of glutamate receptors in healthy neuronal plasticity (Ikonomidou et al., 1999; Failor et al., 2010; Rocha-Ferreira and Hristova, 2015) could be damaged by the use of NMDA and AMPA antagonists at such a sensitive developmental stage.

An integrally linked cascade is that of oxidative stress. Excitotoxicity causes energy depletion, mitochondrial dysfunction, and cytosolic calcium accumulation, which in turn leads to generation of free radicals (Ferriero et al., 1996; Ferriero, 2001). Free radicals alter membrane pump function, allowing more glutamate release and NMDA receptor activation, leading to more excitotoxicity (Schanne et al., 1979; Robertson

J.D. et al., 2002; Starkov et al., 2004). Oxidative stress is a general term for the increase in free radical production as a result of oxidative metabolism under pathologic conditions (Inder and Volpe, 2000; Ferriero, 2001). When oxygen floods the microenvironment of cells damaged by hypoxia, mitochondrial oxidative phosphorylation is overwhelmed and reactive oxygen species accumulate (Ferriero, 2001). Fetal life elapses in a low oxygen environment (East et al., 1998). In the first minutes of life, an abrupt increase in O<sup>2</sup> partial pressure occurs, which creates a pro-oxidant condition (Stiller et al., 2002). During birth asphyxia, excess calcium influx and other factors lead to severe oxidative stress (Forder and Tymianski, 2009). There is accumulation of hydrogen peroxide after HI in neonatal mice but not in adults (Lafemina et al., 2006). Because of its high lipid content, the brain is particularly susceptible to free radical attack (O'Brien and Sampson, 1965; Northington et al., 2001). The polyunsaturated fatty acid content of the brain increases during gestation (Crawford and Sinclair, 1971; Mishra and Delivoria-Papadopoulos, 1989). Lipid peroxidation may be a major factor in the white matter damage (Back et al., 1998; Baud et al., 2004). The developing brain's immature antioxidant defense also contributes to sensitivity to oxidative stress (Li et al., 1997; Mishra and Delivoria-Papadopoulos, 1999; Li and Jackson, 2002; Felderhoff-Mueser et al., 2002; Vannucci and Hagberg, 2004; Blomgren and Hagberg, 2006; Ikonomidou and Kaindl, 2011; Miller et al., 2012). Adequate stores of antioxidants are necessary to protect against oxidative injury. Specifically, depletion of neuronal reduced glutathione exacerbates oxidative injury (Chen and Liao, 2003; White and Cappai, 2003; Brongholi et al., 2006).

Finally, inflammation is a major component of neonatal HI injury. Low-dose treatment with intrauterine LPS dramatically increases severity of HI injury in neonatal mice, but protects against HI in adult rodents (Wang et al., 2007b). Intracerebral injection of NMDA receptor agonist produces a pattern of white matter injury, in which pretreatment with systemic IL-1ß, IL-6, IL-9, or TNF-α leads to a significant increase in lesion size (Marret et al., 1995; Dommergues et al., 2000). There is now substantial experimental evidence that intrauterine inflammation can exacerbate neonatal HI (Lehnardt et al., 2003; Eklind et al., 2005; Marini et al., 2007), which some have referred to as the "double-hit hypothesis" (reviewed in Agrawal and Hirsch, 2012; Hagberg et al., 2012; Dammann and Leviton, 2014). Microglia, the resident macrophages of the CNS, are among the first cells to become activated after HI (Fujimoto et al., 1989; Tahraoui et al., 2001; Kaur et al., 2007). Activated microglia migrate to damaged regions (Leonardo and Pennypacker, 2009) and produce inflammatory cytokines, glutamate, nitric oxide, and free radicals (Wood, 1995; Kaur and Ling, 2009). Drugs that block microglial activation protect the neonatal brain (Dommergues et al., 2003). Following hypoxia-ischemia, compromise of the BBB allows the entry of macrophages (Alvarez-Diaz et al., 2007; Leonardo and Pennypacker, 2009). Astrocytes also play a role in inflammation (Wang et al., 2003; Girard et al., 2008, 2009). CSF cytokines are elevated in term infants who later develop cerebral palsy (Savman et al., 1998; Dammann and O'Shea, 2008). The diverse network of interacting mechanisms demonstrate the molecular complexity of neonatal HI injury. Potential protective treatments should strive to tackle common mediators of these cascades relevant

to all three pathways, otherwise full protection will not be achievable.

## EVALUATION OF AVAILABLE ANIMAL MODELS OF NEONATAL HYPOXIA ISCHAEMIA

Neonatal hypoxia ischaemia has been modeled extensively in mice and rats (reviewed in Hagberg et al., 2002; van der Worp et al., 2007; Yager and Ashwal, 2009; Dean et al., 2015), with a minority of researchers also studying larger animals such as pigs, sheep (reviewed in Roohey et al., 1997; Dean et al., 2015) or primates (Fahn et al., 1979; Volpe, 2012). Models intending to replicate the clinical symptoms of neonatal human HI can be roughly divided into the four categories discussed below. All models have distinct advantages and disadvantages. A summary of available rodent models is shown in **Figure 4**.

## Rice–Vannucci Model of Term Hypoxia Ischaemia

Most published studies modeling neonatal HI in animals have employed the Rice–Vannucci model (Rice et al., 1981). This model comprises unilateral carotid artery ligation, recovery with the dam for approximately 1 h, followed by exposure to 8% oxygen for 1–3 h at 37◦C. Although, the model was initially described in rat (Rice et al., 1981), it has been successfully adapted for mouse with similar anatomical and behavioral effects (Ditelberg et al., 1996; Ferriero et al., 1996; Sheldon et al., 1998).

Rice' and Vannucci's model replicates anatomical damage seen in human neonates. Their initial study showed selective graymatter sensitivity to neuronal necrosis, with gray matter injury observed in cortex, hippocampus, thalamus, and basal ganglia (Rice et al., 1981; Andine et al., 1990; Towfighi et al., 1995; Vannucci et al., 1999), encompassing the sites damaged in human neonatal HI. Histologically, there is a gradation of injury that correlates with the duration or severity of insult (Towfighi et al., 1991, 1995). White matter lesions have also been described in this model (de Torres et al., 1997; Ness et al., 2001; Liu et al., 2002; Drobyshevsky et al., 2005, 2007b), the extent of which correlate with the duration of exposure to hypoxia (Liu et al., 2002). However, bilateral common carotid artery ligation appears a stronger model of white matter damage (Jelinski et al., 1999; Uehara et al., 1999; Cai et al., 2001). Metabolic alterations in the Rice–Vannucci model include decreased cerebral blood flow (Sakurada et al., 1978; Vannucci et al., 1988), brain acidosis (Welsh et al., 1982; Yager et al., 1991), and decreased cerebral glucose uptake (Vannucci et al., 1989; Sokoloff et al., 1977). An inflammatory response has also been demonstrated (Bona et al., 1998).

Another convincing aspect of the Rice–Vannucci model is its ability to predict the therapeutic effect of hypothermia following the neonatal HI event. Mice treated with hypothermia showed smaller lesion volumes, in addition to better performance on the Morris water maze and circling tests (Yager et al., 1993; Lee et al., 2010; Kida et al., 2013; Lin et al., 2014). Many papers have investigated the behavioral outcomes of Rice–Vannucci injury in adult rodents. This model gives rise to well documented behavioral phenotypes including: impaired spatial learning and memory (Balduini et al., 2000, 2001; Ikeda et al., 2001; Wang et al., 2002; Arteni et al., 2003; Pereira et al., 2007; Cai et al., 2009; Greggio et al., 2011; Hill et al., 2012; Zheng and Weiss, 2013; Alexander et al., 2014; Gillani et al., 2015); impaired motor function as assessed by rotarod test, open field and motor reflexes (Barth and Stanfield, 1990; Jansen and Low, 1996a,b; Jansen et al., 1997; Balduini et al., 2000, 2001; Tomimatsu et al., 2002; Ådén et al., 2003; Lubics et al., 2005; Pazaiti et al., 2009; Im et al., 2010; Nijboer et al., 2010; Karalis et al., 2011; Chen et al., 2012; Ruiz et al., 2012; Sanches et al., 2012; Zheng et al., 2012; Xiong et al., 2013; Alexander et al., 2014; Gillani et al., 2014, 2015; Kim et al., 2014; Zhang Q. et al., 2014; Park D. et al., 2015; Park W.S. et al., 2015); sensory processing abnormalities (Alexander et al., 2014); and other cognitive phenotypes, such as reduced attention (Buwalda et al., 1995; Martin et al., 1997; Sanches et al., 2013; Perera et al., 2014; Miguel et al., 2015). Despite this broad range of documented effects, there are some contradictions between individual investigators (reviewed in Lubics et al., 2005), which suggest that different genetic backgrounds, severity, or experimenters can significantly affect the outcome of Rice–Vannucci model.

The Rice–Vannucci model of neonatal hypoxia ischaemia has several advantages. One is its prevalence, allowing direct comparisons with many other published papers (Vannucci et al., 1993, 2005; Yager and Ashwal, 2009; Dean et al., 2015). Another is that the contralateral hemisphere, exposed to hypoxia in the absence of ischemia, appears normal (Yager et al., 1991, 1992, 1996; Vannucci and Yager, 1992), providing a control hemisphere within the experimental brain. Thorough behavioral characterization (Arteni et al., 2003; Lubics et al., 2005) support the long term consequences of this model mimicking neonatal HI. One significant drawback of this model is the high variability in size and severity of infarct between animals, making comparisons between experimenters difficult (Vannucci and Hagberg, 2004; Vannucci and Vannucci, 2005). Additionally, the invasive nature of severing the common carotid artery does not replicate human injury; such severe vascular abnormalities occur rarely, if at all (Ment et al., 1984; Hill, 1991).

#### Hypoxia-Only Models

Some experimenters induce hypoxia in rodents exclusively using an oxygen deprivation chamber, without a preceding ischaemic procedure. These models are not as widely used as the Rice–Vannucci method, but have the potential to describe milder injuries and avoid the unphysiological occlusion of the common carotid artery. There are currently no published reviews or meta-analyses of hypoxia-only investigations, Supplementary Table 1 contains a brief summary of 122 published papers using hypoxia-only methodology compiled from Pubmed search results.

Historically, these methods have been used to investigate hypoxic brain biochemistry. Several studies have documented altered levels of neurotransmitters (Hedner et al., 1980; Kaneko et al., 1985; Yamamoto et al., 1985; Yamamoto and Kato,

1986; Hadjiconstantinou et al., 1990; Seidler and Slotkin, 1990; Dell'Anna et al., 1993; Tanaka et al., 1995; Anju et al., 2010a,b; Anju and Paulose, 2011, 2013). However, there is little consensus over the direction or magnitude of changes (Decker M.J. et al., 2003; Decker et al., 2005). Hypoxia-only models have become an established model used to generate seizures in neonatal rats (Jensen et al., 1995; Applegate et al., 1996; Rodríguez-Alvárez et al., 2015; Sampath et al., 2015). In one such model, P6 rat pups were placed in chambers at 9% O<sup>2</sup> partial pressure and 20% CO<sup>2</sup> partial pressure for 60 min. Some pups were immediately restored to room air, whereas others underwent gradual reduction of CO<sup>2</sup> (Helmy et al., 2011; Tolner et al., 2011). Pups which underwent hypoxia with immediate restoration of CO<sup>2</sup> had a greater mortality rate and higher seizure frequency. However, subsequent anatomical analysis of these brains at P8 (Boss et al., 2005; Wang et al., unpublished), failed to show any differences in expression of cell death markers or layer-specific markers of healthy cortical neurons. Therefore, the hypoxiaonly insult resulting in seizures appears to generate only subtle injury to the brain, far short of that seen in some human patients.

A range of behavioral phenotypes have been reported in hypoxia-only models, including hyperactivity (Shimomura and Ohta, 1988; Dell'Anna et al., 1991; Speiser et al., 1991; Decker et al., 2005); increased aggression (Mikati et al., 2005; Tang et al., 2006), altered ultrasonic vocalization (Venerosi et al., 2006), and disturbed sleep (Decker M.B. et al., 2003). Relatively few investigators have pursued standard tests of spatial memory and

locomotor behavior, and have obtained mixed results (Dell'Anna et al., 1991; Speiser et al., 1998; Rotstein et al., 2006; Coq et al., 2008; Raveendran and Skaria, 2013; Wang S. et al., 2015). The majority of hypoxia-only phenotypes are based on custom behavioral tests, making them difficult to compare to other established animal models. Additionally, some studies report no behavioral deficit following neonatal hypoxia alone (Buwalda et al., 1995; Iuvone et al., 1996; Casolini et al., 2005; Blaise et al., 2009; Mikhailenko et al., 2009; Anju et al., 2010c; Wang S. et al., 2015).

Although hypoxia-only models offer the potential to replicate the mechanism of hypoxia without major ischaemia seen in human neonatal HI patients, the models currently available are not ideal. One significant problem is the lack of methodological unity between different experimenters. There is little consensus on age of animal, background strain, oxygen partial pressure, time of exposure to hypoxia, or body temperature (see Supplementary Table 1). One example demonstrating the relevance of close control of these variables is temperature. P0 rat pups exposed to anoxia exhibit behavioral defects when anoxia was conducted at 39◦C, yet not at 33 and 36◦C (Rogalska et al., 2004, 2009; Caputa et al., 2005; reviewed in Rogalska et al., 2006). Therefore, far more care is needed to justify the design of hypoxia-only experiments before these models can be considered dependable models of human neonatal HI.

## Inflammatory Models of Perinatal Brain Injury

Intrauterine infection is strongly associated with preterm birth and brain injury (Stoll et al., 2004; Mitha et al., 2013; Strunk et al., 2014; Dean et al., 2015). Many models have been described which introduce different inflammation-inducing molecules at different ages (Dean et al., 2015), many of which cause cerebral inflammation and white matter damage seen in human patients.

Administration of live E. coli into the uterus of pregnant rats can result in neutrophil infiltration in the fetal brain, increased fetal reabsorption and stillbirth, while surviving pups exhibit increased brain chemokines, cytokines, white matter injury, and behavioral phenotypes (Debillon et al., 2003; Rodts-Palenik et al., 2004; Pang et al., 2005; Yuan et al., 2005; Girard et al., 2009; Bergeron et al., 2013). The effects of bacterial mimetics such as the cell wall component lipopolysaccharide (LPS), have also been investigated. Intracervical injection in embryonic day 15 (E15) mice was associated with mild white matter injury but no behavioral deficits (Bell and Hallenbeck, 2002; Poggi et al., 2005; Wang et al., 2007a), whereas repeated intracervical LPS was associated with delayed neurosensory development (Toso et al., 2005; Rousset et al., 2006, 2013). Other inflammatory models include viral infection simulated by injection of poly(I:C), a synthetic double stranded viral RNA, injection of which is associated with long-term behavioral deficits (Shi et al., 2009; Richetto et al., 2013). Postnatal administration of inflammatory agents is widely used in rodents to model postnatal infection. Subcutaneous injection of live E. coli to P3 mouse pups was associated with microgliosis, loss of oligodendrocytes, and impaired motor coordination (Lieblein-Boff et al., 2013). However, most techniques which employ live bacterial injection have very high mortality rates (Rodewald et al., 1992; Tran and Weisman, 2004; Placencia et al., 2009; Loron et al., 2011). Postnatal intraperitoneal injection of LPS can also cause white matter damage and cerebral cytokine response (Brochu et al., 2011; Brehmer et al., 2012; Nobuta et al., 2012; Smith et al., 2014). Repeated daily injection of LPS in mice resulted in elevated serum IL-6, reduced gray matter volume, decreased oligodendrocyte numbers, and decreased myelin staining (Wang et al., 2009; Malaeb et al., 2014). Similarly, repeated IL-1β injection in

P1–P5 mice has been associated with impaired oligodendrocyte progenitor maturation, and severe memory deficits (Favrais et al., 2011).

One of the strengths of the inflammation model is that it reflects the exposure to infectious or inflammatory agents present outside of the highly sterile individually ventilated cages where many academic institutions keep experimental animals. Most of the inflammatory risk factors for increased severity of HI injury in human patients, such as maternal infection (Stoll et al., 2004; Girard et al., 2012; Dean et al., 2015), will result in systemic inflammation in addition to CNS-specific recruitment of microglia. However, there is currently debate concerning the relevance of maternal inflammation to fetal brain damage (Leviton et al., 1999; Redline and O'Riordan, 2000; Neufeld et al., 2005). The debate intensifies when fetal systemic inflammation is contrasted with neuroinflammation. Although the majority of publications cited above administer pro-inflammatory agents by intracerebral injection, some studies have administered LPS by intravenous injection (reviewed in Wang et al., 2006). In fetal sheep, intravenous LPS causes white-matter specific damage (Garnier et al., 2001, 2006; Duncan et al., 2002; Mallard et al., 2003; Peebles et al., 2003; Svedin et al., 2005), suggesting that systemic inflammation may be sufficient to trigger brain injury. However, the results of these experiments could be explained by a secondary neuroinflammatory response to systemic inflammation causing the white matter injury. Separating these parameters in future experiments will prove demanding.

Data suggest that gene expression in mouse models of inflammation closely corresponds to that seen in humans (Takao and Miyakawa, 2015), however, postnatal murine response to inflammation is likely different to that experienced by a human fetus. An important limitation of inflammatory studies is impact on cardiovascular function and the potential for secondary cerebral HI (Eklind et al., 2001; Favrais et al., 2011; Wang C.T. et al., 2015). Another is that varying results have been reported. This may be due to differences in LPS used between groups, purification method (Westphal, 1965; Lam et al., 2014), or dose (Fujihara et al., 2003). Alternatively, there is increasing evidence that the considerable individual variability in response to clinical sepsis between patients is related to host genetics (Christaki and Giamarellos-Bourboulis, 2014). Therefore, although inflammation is doubtless an important factor in neonatal brain injury, the variation between different models makes these methods difficult to evaluate.

#### Alternative Models

In many ways, non-rodent models are more representative of neonatal HI as it affects human patients. Here follows a brief flavor of techniques. Relatively few studies have investigated neonatal HI in primates. Classical studies asphyxiated term monkey fetuses by covering their heads with a rubber sac and clamping the umbilical cord (Fahn et al., 1979). Fetuses resuscitated after 20 min had extremely high mortality. However, 12 min of asphyxia were required to produce any neuropathologic injury. The fetuses displayed damage predominantly within the brainstem. In a model of partial ischemia (Fahn et al., 1979), pregnant females were rendered hypotensive during the third trimester. When blood oxygen saturation was reduced to 10% for 5 h, fetuses became profoundly acidemic. At birth, these displayed opisthotonus, decerebrate posturing, and convulsions. There is one model of whitematter injury based on baboons delivered prematurely by hysterotomy (Inder et al., 2004, 2005a,b). Premature baboons were treated in an intensive care setting. Approximately half displayed white-matter injury. Analysis of behavior has not been undertaken.

More work has been carried out in fetal sheep. Umbilical cord occlusion and term asphyxia has been carried out (Gunn et al., 1992; Williams et al., 1992; Mallard et al., 1993; Richardson B. et al., 1996; de Haan et al., 1997a,b,c; Richardson and Bocking, 1998; Dalitz et al., 2003), repeated asphyxia at 5-h intervals resulted in injury to the striatum almost exclusively, whereas episodes of asphyxia repeated every 5 min caused diffuse and extensive damage to cortex, thalamus, and cerebellum in 40% of animals, and selective neuronal necrosis in the remainder. White matter injury has also been investigated (Ting et al., 1983), only those fetuses in which the mean arterial blood pressure fell below 30 mmHg exhibit brain damage, irrespective of hypoxia. Neuropathologic injury after carotid artery occlusion for 30 min caused both gray and white matter involvement, with cortical damage and selective neuronal necrosis in thalamus and striatum (Reddy et al., 1998; Petersson et al., 2002). Several laboratories have developed models in sheep after systemic endotoxemia (Duncan et al., 2002; Mallard et al., 2003). Unfortunately, no behavioral outcomes are currently available in sheep.

Finally, other small laboratory animals have been investigated. Preterm rabbit fetuses exposed to sustained placental insufficiency via intrauterine occlusion of the descending aorta displayed significant alterations in motor responses to olfactory stimuli, coordination of suck and swallow, and marked hypertonia, reminiscent of spastic quadriplegia (Yoon et al., 1997; Derrick et al., 2004, 2007; Saadani-Makki et al., 2008). Diffusion-weighted imaging detected a threshold in white matter loss below which all rabbit kits developed hypertonia, (Drobyshevsky et al., 2005, 2007a). Another promising model is that of the spiny mouse, a rodent which shows a similar level of brain development to a human neonate at birth (Brunjes, 1985, 1989, 1990; Gozzo et al., 1985; D'Udine and Alleva, 1988; Brunjes et al., 1989), which also shows varied motor deficits (Ireland et al., 2008, 2009, 2010) and neuroanatomical pathology (Hutton et al., 2009; O'Connell et al., 2013) following neonatal HI.

In brief, large animals can better replicate the conditions of a single human fetus exposed to a non-sterile environment. However, none of these models have access to the same extent of transgenic manipulation or validated behavioral tests that are possible in rodent studies. These models are invaluable for aiding comparison of brain development at birth which occur between species (Clancy et al., 2007), whereas the relative immaturity of mouse and rat brains at birth introduce an unwelcome variable into these experiments. For the foreseeable future, insights from

both rodent work and larger animals will be important for better understanding neonatal HI.

## SUBPLATE: A REGION OF HYPER-SENSITIVITY TO NEONATAL HI?

One brain region exemplifying how much remains unknown about the developing brain's response to HI, and the uncertainties involved in interpreting results from animal studies, is subplate. The subplate is an early-born transitory neuronal layer of cerebral cortex which serves an important role in the establishment of thalamocortical connectivity during development (reviewed in Judas et al., 2010; Kanold and Luhmann, 2010; Kostovic and Judas, 2010; Hoerder-Suabedissen and Molnár, 2015) (see **Figure 4**). Thalamocortical axons associate with and grow alongside subplate neurons in the developing cortex (Herrmann et al., 1994; Molnár et al., 1998; Kostovic and Judas, 2006), and selective excitotoxic ablation of subplate disrupts thalamocortical connectivity in animal models (Ghosh et al., 1990; Ghosh and Shatz, 1992; Lein et al., 1999; Kanold et al., 2003; Kanold and Shatz, 2006; Magnani et al., 2013).

Subplate has been identified in a range of mammals by conserved molecular markers (Hoerder-Suabedissen et al., 2009; Hoerder-Suabedissen and Molnár, 2012; Oeschger et al., 2012; Pedraza et al., 2014). Neuronal death occurs in infant subplate during normal development (Chun et al., 1987; Al-Ghoul and Miller, 1989; Hamre et al., 1989; Kostovic and Rakic, 1990; Woo et al., 1991; Price et al., 1997). However, the molecular mechanisms behind this programmed cell-death remain unknown (McQuillen and Ferriero, 2005; Hoerder-Suabedissen and Molnár, 2015). Functional roles for subplate in thalamocortical (Allendoerfer and Shatz, 1994; Kanold and Luhmann, 2010) and corticothalamic (Grant et al., 2016) development have been described, demonstrating lasting significance of this transient population.

The susceptibility of subplate neurons to neonatal HI remains little understood. Early studies claimed subplate is selectively vulnerable. Cultured rat subplate neurons were more vulnerable to OGD compared to mixed cortical neurons (Nguyen and McQuillen, 2010). In rats exposed to systemic HI at E18, a significant decrease in Nurr1-expressing subplate neurons was documented by P2 (Jantzie et al., 2014, 2015a,b,c) and similarly in fetal sheep (Dean et al., 2011). However, Nurr1-expressing Layer VI neurons were also decreased. It is also possible that HI induced differential marker expression in subplate neurons, rather than cell death. In vitro electrophysiological recordings in neocortical slices from newborn rats have demonstrated a pronounced functional impairment of subplate neurons following OGD (Albrecht et al., 2005). Another influentiential study that claimed selective vulnerability of subplate to hypoxia used immunohistochemistry to detect cell-death in neurons labeled with BrdU at E10.5 following a modified Rice–Vannucci model in P1 rats (McQuillen et al., 2003; McQuillen and Ferriero, 2004, 2005). However, Layer VI neurons born at this date also expressed cell-death markers, suggesting subplate may not be selectively susceptible to HI injury. Conversely, a systematic quantification of cell death markers throughout subplate compared with other cortical layers was conducted using immunohistochemistry (Okusa et al., 2014). Three subplate-specific marker positive populations showed little costaining with caspase-3 in brains with mild to moderate HI lesions. In severely damaged cases, caspase-3 staining was found throughout the cortex. Therefore, layer-specific sensitivity of cortical neurons to HI remains debated. A double birthdating study, targeting subplate and layer V or VI combined with HI could resolve this issue. However, such an experiment has not yet been performed.

Human literature supports the proposition that interstitial white matter, the equivalent of subplate in the postnatal human brain (Clancy et al., 2009; Suarez-Sola et al., 2009; Garcia-Marin et al., 2010; Hoerder-Suabedissen and Molnár, 2015), shows structural damage as a result of neonatal injury. Tissue from preterm human infants with periventricular leukomalacia showed a deficit in the number of MAP-2 expressing neurons throughout the interstitial white matter (Kinney et al., 2012). Scarring has been observed in interstitial white matter, alongside the expression of cell death markers, in brain tissue from human infants (Pogledic et al., 2014). The number of interstitial neurons present in white matter is known to peak at the developmental stage most sensitive to white matter injury (Kostovic et al., 2002; Kostovic and Judas, 2010). Immunohistochemical analysis on neonatal telencephalon samples obtained post-mortem from infants born at 25–32 weeks gestation showed a significant loss of GABA-ergic subplate neurons (Robinson et al., 2006). Many reviews have highlighted the interstitial white matter as a site which should be more thoroughly investigated in neonatal brain injury (Volpe et al., 1996; Leviton and Gressens, 2007; Volpe, 2009; Kostovic et al., 2011). Several of these reviews propose a role for subplate in resultant cognitive and behavioral deficits (Volpe, 2009, 1977; Kostovic et al., 2011), based on the established role of subplate in thalamocortical development. Unfortunately, post-mortem studies are unable to convey whether subplate pathology is responsible for any of the clinical outcomes in patients. The response of both human and rodent subplate to neonatal HI remains unclear, despite assertions of interest from the field. A systematic study of the subplate in vitro and in vivo following HI injury would provide valuable information. As subplate transiently integrates within the developing cortical circuitry (see **Figure 5**) interactions between subplate neurons and other cortical neurons, such as key interneuron populations, should also be further investigated (Marques-Smith et al., 2016).

There are several reasons for believing that subplate could be critical in neonatal HI. One is the proposed transient secretory function. Subplate is enriched in CSPGs, whereas the adjacent intermediate zone is not (Kostovic et al., 2014). Chondroitin sulfate proteoglycans (CSPGs) are generally secreted from cells, and the subplate transcriptome catalogs a plethora of genes involved in production of extracellular matrix and proteoglycans (Belgard et al., 2011; Hoerder-Suabedissen et al., 2013). Additional genes with subplate-restricted expression in the cortex encode secreted proteins, including neuroserpin (Serpini1), neuronal pentraxin 1 (Nptx1), and insulin-like

growth factor binding protein 5 (Igfbp5) (Hoerder-Suabedissen et al., 2013), several of which have been validated by immunohistochemistry (Kondo et al., 2015). Connective tissue growth factor (CTGF), secreted extracellular matrix-associated protein involved in regulation of cellular adhesion, migration, mitogenesis, differentiation and survival (Brigstock, 1999; Stritt et al., 2009), is also detectable in the subplate at E18, increasing in intensity at P3 and P8 (Hoerder-Suabedissen et al., 2009, 2013). It is unknown whether subplate-secreted proteins serve a protective function in the developing brain or not. Related to this secretory function, subplate neurons express relatively mature rER (Kondo et al., 2015), an organelle essential for production of secreted proteins (reviewed in Novikoff, 1976; Pfeffer and Rothman, 1987; Lodish, 1988; Hurtley and Helenius, 1989; Pelham, 1989). As a result of the cellular pressures of high protein production, cells activate a series of mechanisms referred to as the ER stress response (Schroder and Kaufman, 2005; Wu and Kaufman, 2006; Kondo et al., 2011). Nissl staining demonstrates enriched protein production in subplate at P8 in mouse, although staining is much fainter in adult (Kondo et al., 2015). Morphology of subplate neurons is similar to that of rER-rich plasma cells under EM (Bloom, 1968; Kondo et al., 2015), and immunohistochemistry for ER stress marker binding immunoglobulin protein (BiP) confirmed that ER stress occurs in developing subplate (Okiyoneda et al., 2004; Kondo et al., 2005, 2012, 2015). BiP protein synthesis is up-regulated in whole brain under stress conditions, such as glucose deprivation, hypoxia, or the presence of toxins (Lee, 1987, 2001, 2005). It is not known exactly which proteins are secreted throughout development by the subplate, but this enhanced metabolic stress during hypoxia and the potential for neuroprotective secretion reinforce the value of further study of subplate in neonatal HI models and human patients.

#### NEUROSERPIN: A CASE FOR A NOVEL NEUROPROTECTIVE TREATMENT

The development of novel treatments to supplement the sole currently licensed therapy, hypothermia (Jacobs et al., 2005, 2013; Shah et al., 2007; Tagin et al., 2012), is imperative. Since the discovery of this groundbreaking treatment, little progress has been made in identifying additive pharmacological therapies. Few potential treatments have translated to human clinical trials. Resuscitation at room temperature (Vento et al., 2001a,b; Rabi et al., 2007) and xenon gas administration alongside hypothermia (Hobbs et al., 2008; Thoresen et al., 2009) are the only therapies shown to have any additive effect. However, more recent publications have questioned the efficacy of xenon as an additive therapy alongside hypothermia. For example, randomized clinical trials have demonstrated that although xenon gas is a safe treatment, there is little or no therapeutic effect of combined hypothermia and xenon gas in moderate and severe cases of neonatal HI at 18 months follow-up (Azzopardi et al., 2013, 2015; Dingley et al., 2014). A similar experiment in rats found that xenon treatment made no difference to lesion size or neuronal cell numbers in cases of severe HI (Sabir et al., 2016). Barbiturate anticonvulsants have no effect on long-term neurological development when given following neonatal HI (reviewed in Goldberg et al., 1986; Hall et al., 1998; Singh et al., 2004; Vargas-Origel et al., 2004; Evans et al., 2007). Recent clinical studies suggest that high dose erythropoietin (EPo) treatment in term HI neonates reduces disability (Strunk et al., 2008; Zhu et al., 2009; McPherson and Juul, 2010). However, even proponents of this potential treatment advise caution in interpreting these early results, and the therapeutic effect by no means completely prevents disability. Combination therapy of N-acetylcysteine, a free radical scavenger, and systemic hypothermia reduces infarct volume after focal HI injury (Jatana et al., 2006). Another free radical scavenger, allopurinol, reduces cerebral edema and neuropathological damage (Palmer et al., 1990). However, these treatments have only been trialed in animals, and the field is still awaiting a candidate neuroprotectice molecule which is both safe and effective in neonatal humans (Lai and Yang, 2011; Pazos et al., 2012).

Another therapeutic approach under development for neonatal HI is stem cell therapy (reviewed in Parolini et al., 2010; Zhang X. et al., 2014; Douglas-Escobar and Weiss, 2015; González-Portillo et al., 2015; Ruiz et al., 2017). These therapies make use of evidence that transplantation of human bone-marrow derived stem cells into the lesion can assist brain plasticity (Bonifacio et al., 2011; Tajiri et al., 2013). Results of this therapy have been promising in animal models (Jendelová et al., 2004; Daadi et al., 2009a,b), however, the technique awaits validation in a clinical setting. Additionally, the use of stem cell therapy for cerebral palsy in human patients has produced mixed results (Li et al., 2012; Liao et al., 2013; Sharma, 2014).

Therefore, new approaches are required to identify potential neuroprotective molecules as treatments for neonatal HI (Tuor et al., 1995; Ferriero, 2004; Fan et al., 2010). Drawing on our detailed yet incomplete knowledge of the physiology of neonatal HI, several factors must be satisfied in a new potential therapy. All potential treatments should be safe for vulnerable neonates and not interfere with essential developmental milestones. This challenges the NMDA-inhibition strategies (Meldrum, 1990; Johnston, 2001, 2005), as the glutamate system is essential for setting up normal synaptic plasticity within the developing brain (Hattori and Wasterlain, 1990; Ikonomidou et al., 1999; Johnston, 2009). Treatments should also be specific, to avoid extreme adverse effects in these vulnerable infants. The ideal treatment would target molecules common to the excitotoxicity, oxidative stress and inflammation pathways (Chaudhari et al., 2014; Fischer and Maier, 2015; Chamorro et al., 2016; Burd et al., 2016). Targeting these common mediators would allow a single therapy to be efficacious against multiple mediators of brain damage, instead of merely eliciting a reshuffle of pathways to favor a different method of cell death. Here follows an outline of one potential therapy under development which may satisfy these theoretical restraints as a model to incite debate concerning how future therapies are identified. Enter: neuroserpin.

# Neuroserpin's Molecular Pedigree

Neuroserpin is a neuronally secreted serine-protease inhibitor enzyme with roles in cell death, neural plasticity, and microglial activation. Neuroserpin was first identified in the secreted product of cultured neuronal axons (Osterwalder et al., 1996). Protein expression is primarily localized to neurons (Osterwalder et al., 1996; Hastings et al., 1997; Lee et al., 2015b). The protein neuroserpin is encoded in mouse and human by the gene Serpini1. Serpini1 mRNA expression is specific to the CNS, both in mouse (Krueger et al., 1997) and human (Teesalu et al., 2004). In situ hybridization studies have demonstrated expression of Serpini1 throughout cortex, hippocampus, and olfactory bulbs, with scattered expression in cerebellum, pons, and thalamus from birth until adulthood (Krueger et al., 1997). High Serpini1 expression has been identified in subplate relative to Layer VI of cortex in P8 mouse brain by microarray (Hoerder-Suabedissen et al., 2009), confirmed by immunohistochemistry (Kondo et al., 2015). Although Serpini1 expression has been demonstrated in human brain tissue (Teesalu et al., 2004), detailed maps of expression are currently missing. This highly specific localisation suggests that Serpini1/neuroserpin modulatory treatments may not create dangerous side effects in other vulnerable organs such as the cardiovascular and pulmonary systems.

Neuroserpin is an inhibitory enzyme. Its primary target is the adult human stroke treatment, tissue plasminogen activator (tPA) (Osterwalder et al., 1996, 1998; Barker-Carlson et al., 2002; Yepes and Lawrence, 2004a; Miranda and Lomas, 2006; Ricagno et al., 2009, 2010). Neuroserpin has also been shown to exhibit weak inhibition of plasmin and urokinase plasminogen activator (Osterwalder et al., 1998). This function is conserved across species (Schrimpf et al., 1997; Hill et al., 2001; Ellisdon et al., 2014). Through inhibition of tPA, neuroserpin has the potential to influence many distinct molecular cascades (Yepes and Lawrence, 2004b,c; Benarroch, 2007; Yepes, 2015). tPA, also secreted by neurons (Krystosek and Seeds, 1981a,b; Lochner et al., 2006), has been shown to interact with a large number of pathways; activating NMDA receptors (Qian et al., 1993; Nicole et al., 2001), activating microglia (Benchenane et al., 2004), and recruitment of cell death related cascades (Yepes and Lawrence, 2004a; Zhao et al., 2007), in addition to its function of cleaving vascular thrombosis (Rijken et al., 1979, 1980). The known components of the neuroserpin/tPA molecular cascade are summarized in **Figure 6**.

Other, tPA-independent enzymatic pathways for neuroserpin have also been described (Ma et al., 2012a). Cell lines over-expressing neuroserpin demonstrate an increase in N-cadherin protein expression and related cell adhesion, maintained when the tPA binding site of neuroserpin is mutated (Lee et al., 2008). In vitro, neuroserpin has been shown to prevent excitotoxic neuronal death induced by plasmin and kainic acid (Wu et al., 2010), and the 20 methionine residues present within neuroserpin have been claimed to convey protection against oxidative stress (Mohsenifar et al., 2007), and exogenous neuroserpin has been shown to possess anti-inflammatory properties (Munuswamy-Ramanujam et al., 2010). The molecular cascades behind these tPA independent functions are not well understood (Ma et al., 2012a). Therefore, neuroserpin has the pedigree to target multiple cell death pathways occurring in the brain during neonatal HI.

# Cell Death, Inflammation, and Plasticity

Neuroserpin's physiological actions can be broadly categorized into three major pathways: cell death, neuronal plasticity, and immune cell activation. These closely follow the major classes of molecular change proposed by pathological theories of neonatal HI, suggesting that neuroserpin offers the potential to target multiple brain-damage pathways.

Neuroserpin has been implicated in protection against neuronal death (Galliciotti and Sonderegger, 2006). Exogenous neuroserpin administration in vitro decreases apoptosis caused by tPA, NMDA, kainic acid, and OGD (Lebeurrier et al., 2005a,b; Mohsenifar et al., 2007; Lee et al., 2008; Wu et al., 2010; Rodríguez-González et al., 2011a; Ma et al., 2012b). In addition, aberrant initiation of neuroserpin expression in cancer cells preserves the tumor and has been linked with increased treatment-resistance in prostate cancer (Chang et al., 2000; Hasumi et al., 2005; Valiente et al., 2014). Exogenous neuroserpin reduces spread of kainate-induced seizures in mouse, and decreases the expression of cell death markers (Yepes et al., 2002).

Of the three classes of neuroserpin function, its role in inflammation is the least well understood. When applied to vascular plaques in vitro, recombinant neuroserpin reduced T-Cell lymphocyte invasion (Munuswamy-Ramanujam et al., 2010). Administration of intracerebral neuroserpin directly following adult HI stroke in mouse, showed a qualitative decrease in the volume of infarct infiltrated by activated microglia (Yepes et al., 2000), and lower microglial inflammatory marker expression (Gelderblom et al., 2013). In human adult stroke patients, higher levels of neuroserpin in blood samples correlate with lower levels of immune marker proteins (Rodríguez-González et al., 2011c). There is considerable debate whether microglial activation is beneficial or detrimental in neonatal HI (McRae et al., 1995; Mallard et al., 2013; Kaur et al., 2013), so understanding the role of neuroserpin in balancing this cascade could provide important data. The lack of a complete molecular network explaining these anti-inflammatory effects highlights the complexity of the pathways involved in cell death and inflammation. Understanding the underlying explanation for these changes will be an important preliminary to administration in human infants who do not live in the highly sterile environments maintained for experimental animals.

In addition, neuroserpin is involved in neuronal plasticity. This could be problematic in the development of neuroserpin as a neonatal HI treatment for human patients, as the neonatal human brain circuitry is undergoing a great deal of developmental plasticity modifications (reviewed in Aoki and Siekevitz, 1988; Partanen et al., 2013; Simpson et al., 2014). The first evidence that neuroserpin may have a role in neuronal plasticity originated from studies of endocrine cell lines, which grow neurite-like processes when neuroserpin is administered to their media (Hill et al., 2001; Jacovina et al., 2001; Borges et al., 2010). At the molecular level, neuroserpin secretion from dense core vesicles is enhanced by depolarisation in cultured neurons (Berger et al., 1999; Ishigami et al., 2007). Monocular

enucleation at P11 in mouse led to decreased Serpini1 mRNA expression in contralateral primary visual cortex compared to the control hemisphere (Wannier-Morino et al., 2003). The relative importance of this molecular cascade for human plasticity is as yet unclear.

# Neuroserpin in In Vitro and Adult Hypoxia Ischaemia

Neuroserpin has been studied extensively in cell culture models of HI stroke. Administration of neuroserpin protects mouse cortical neurons in vitro against NMDA-induced excitotoxicity, but not AMPA-induced excitotoxicity (Lebeurrier et al., 2005a,b; Ma et al., 2012b). Further, neuroserpin is protective in an in vitro model of HI, OGD (Wu et al., 2010; Rodríguez-González et al., 2011a; Ma et al., 2012b). Both neurons and astrocytes cultured in OGD undergo less apoptosis and do not exhibit damaged neural processes when exogenous neuroserpin is administered following OGD (Rodríguez-González et al., 2011a; Ma et al., 2012b). Hippocampal neurons from both wild-type and tPA knockout mice were protected from OGD, plasmin-mediated cell death and kainic acid by neuroserpin administration (Wu et al., 2010), demonstrating that neuroserpin's neuroprotective role is not exclusively dependent on tPA inhibition.

Neuroserpin is also neuroprotective in vivo. The major model used to simulate hypoxic-ischaemic stroke in adult rodents is analogous to the Rice–Vannucci model described above for neonates (Eklöf and Siesjö, 1973; Rice et al., 1981; Bederson et al., 1986). Adult rats injected with neuroserpin before left common carotid artery ligation followed by acute hypoxia showed a decrease in infarct volume compared to controls, in addition to a decrease in expression of apoptosis markers (Yepes et al., 2000). Neuroserpin knockout mice show increased lesion volumes following adult HI (Cinelli et al., 2001; Gelderblom et al., 2013), whereas transgenic mice expressing six-times normal neuroserpin expression demonstrate smaller infarct volumes (Cinelli et al., 2001). Exogenous neuroserpin injection is also protective against NMDA injection in live adult rat (Lebeurrier et al., 2005a). Currently, the only evidence for this has come from adults models of HI stroke, far from conclusive evidence of neonatal protection due to the differences in neuroanatomy and neurochemistry between the developing and adult brain. However, this work remains of utility, as at the very least investigating the neuroserpin-tPA system in neonatal models will add to understanding of the differences between neonatal and adult response to HI.

Neuroserpin administration in rats increases the window for effective tPA treatment in a model of adult stroke (Zhang et al., 2002). Increased tPA expression is found in neuroserpin knockout mice undergoing adult HI injury compared to controls (Cinelli et al., 2001; Gelderblom et al., 2013). However, exogenous neuroserpin is neuroprotective in tPA knockout mice (Wu et al., 2010). Ischaemic-reperfusion-induced injury in the retina of tPA knockout mice demonstrated rescued retinal apoptotic marker expression when injected with neuroserpin (Gu et al., 2015). Therefore, the neuroprotective properties of neuroserpin in adult HI appear to involve both neuroserpin's tPA-dependent and tPA-independent molecular cascades.

A role for neuroserpin in human adult HI stroke has also been described. Correlative studies have shown that high concentrations of neuroserpin in blood samples from human patients are associated with better functional outcome (Rodríguez-González et al., 2011b) and reduced inflammation (Rodríguez-González et al., 2011c). However, attempts to find a relationship between neuroserpin polymorphisms in humans and likelihood of stroke have produced little evidence of protective neuroserpin variants (Cole et al., 2007; Tjärnlund-Wolf et al., 2011). Mutations in the neuroserpin gene have been associated with two rare hereditary disorders (Galliciotti and Sonderegger, 2006); individuals present either with epilepsy (Yepes et al., 2002; Coutelier et al., 2008) or dementia with neuroserpin inclusion bodies (Davis et al., 1999a,b, 2002; Yazaki et al., 2001; Ricagno et al., 2010). It is not currently known whether these disorders are caused by novel functions of the mutated protein or by loss of physiological functions (Lee et al., 2015a).

# Neuroserpin in Neonatal Hypoxia-Ischaemia

fncel-11-00078 May 4, 2017 Time: 20:34 # 15

The evidence above from adult models of HI stroke is far from conclusive evidence of neonatal protection due to differences between the developing and adult brain. However, this work remains of utility, as at the very least investigating neuroserpin in neonatal models will add to understanding of the differences between neonatal and adult response to HI.

Evidence for a role of neuroserpin in neonatal HI is not entirely absent. A patent application published online claims that neuroserpin injection 4 h after neonatal HI in rat pups reduced infarct volume, but had no effect on LPS-sensitized HI (Kuan et al., 2010). However, the small number of published rodent microarrays following neonatal HI have failed to detect significant changes in Serpini1 expression (Carmel et al., 2004; Hedtjarn et al., 2004a,b; Juul et al., 2009; Nagel et al., 2012; Rognlien et al., 2014). This could be explained by the delay between HI injury and sample collection, 6 h to 7 days post injury in this sample, which may be too early or too late to expect to see genomic changes in Serpini1 expression. Also, microarray studies cannot reveal whether neonatal HI causes differences in neuroserpin translation or secretion, neither of which have yet been studied in neonatal HI.

Despite the lack of evidence for neuroserpin's direct role in neonatal HI, substantially more has been published documenting effects of its primary molecular target, tPA, in neonatal HI. tPA is broadly considered to be deleterious to cerebral recovery from neonatal HI (Adhami et al., 2008; Omouendze et al., 2013). In P7 rats which underwent unilateral common carotid artery ligation, tPA treatment was found to impair blood flow throughout the brain, an effect which was rescued by injection of the tPA inhibitor α2-antiplasmin (Adhami et al., 2008). Release of tPA by the cerebral microvasculature, as seen following neonatal HI, has also been shown to compromise the glucose content of the extracellular medium in neonatal mouse cortical culture (Henry et al., 2013). Administration of tPA inhibitor Plasminogen-activator-inhibitor 1 (PAI1) reduces locomotor disorder and white matter damage in LPS-sensitized neonatal HI rats (Yang et al., 2013a,b; Yang and Kuan, 2015). This pathway is relatively understudied in neonatal HI research, despite the potential to target a multitude of injury-relevant molecules. Drawing attention to molecules like neuroserpin and tPA, potential common moderators of injury pathways, will be essential to the development of new efficacious pharmacological treatments.

# CONCLUSION

Neonatal HI remains the most common cause of infant death and disability globally. Reducing the burden of morbidity should be a high priority for biomedical research. A dearth of treatments are currently available postnatally for these vulnerable infants, restricted to only hypothermia, a breakthrough treatment which is not fully effective in all cases, with no currently licensed additive pharmacological treatments. A complex network of interacting molecular cascades, including excitotoxicity, oxidative stress, and inflammation contribute to the gradually developing pattern of neuronal cell death seen in asphyxiated neonates. Targeting common mediators of these pathways with specific targets linked to far-reaching effects offers a potential approach to generating new pharmacological treatments to exact neuroprotection in these neonatal patients.

# AUTHOR CONTRIBUTIONS

All authors listed have made substantial, direct, and intellectual contribution to this work, and approved its final version for publication. LM, LS, AH-S, and ZM: conceived the review focus, conducted literature review, evaluated the literature. LM: reviewed literature, wrote first draft, critically revised the first draft, and finalized the manuscript. LS, AH-S, and ZM: gave feedback, and finalized the manuscript. All authors approved final version of manuscript.

# FUNDING

This work was supported by the Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford UK. LM is funded by the Clarendon Fund, Oxford, UK; University College, University of Oxford, UK; and the Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, UK. ZM is funded by the Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, UK. LS, AH-S and ZM are funded by the Medical Research Council Grant numbers: G00700311 and G00900901. LS and ZM are funded by The Academy of Medical Sciences Newton Advanced Fellowship in partnership with The Royal Society and The National Natural Science Foundation of China for collaborative projects (UK/China grant numbers: NA160314/8161101585).

# ACKNOWLEDGMENTS

We thank Prof. David Edwards, Prof. Pierre Gressens, Dr. Bobbi Fleiss, Dr. Claire Thornton, and Dr. Ana Baburamani for informative discussions on neonatal hypoxia ischaemia and its models.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fncel. 2017.00078/full#supplementary-material

# REFERENCES

fncel-11-00078 May 4, 2017 Time: 20:34 # 16


of glucose, oxygen and epinephrine resuscitation. Neurol. Sci. 34, 287–296. doi: 10.1007/s10072-012-0989-x




neonatal encephalopathy. Lancet 361, 736–742. doi: 10.1016/S0140-6736(03) 12658-X


developing brain of the opossum (Monodelphis domestica). J. Comp. Neurol. 460, 451–464. doi: 10.1002/cne.10661







resonance spectroscopy. Pediatr. Res. 36, 699–706. doi: 10.1203/00006450- 199412000-00003



hypoxic-ischaemic encephalopathy and neurodevelopmental outcome at 24 months. Dev. Med. Child Neurol. 52, e55–e59. doi: 10.1111/j.1469-8749.2009. 03550.x


cells for neonatal hypoxic-ischemic encephalopathy. PLoS ONE 10:e0120893. doi: 10.1371/journal.pone.0120893


in neonates. Neurosurgery 40, 1190-1199; discussion 1199–1200. doi: 10.1097/ 00006123-199706000-00016


of glucose, oxygen, and epinephrine. Int. J. Neurosci. 123, 80–88. doi: 10.3109/ 00207454.2012.731457


months after neonatal encephalopathy measured by magnetic resonance spectroscopy. Pediatr. Res. 46, 287–296. doi: 10.1203/00006450-199909000- 00007


subjected to hypoxia. J. Pharmacol. Exp. Ther. 319, 332–339. doi: 10.1124/jpet. 106.106898


asphyxia at term. Eur. J. Paediatr. Neurol. Soc. 12, 359–365. doi: 10.1016/j.ejpn. 2007.10.003


asphyxia in term infants. Early Hum. Dev. 25, 135–148. doi: 10.1016/0378- 3782(91)90191-5


decarboxylase activity and protein synthesis. Neonatology 66, 254–266. doi: 10.1159/000244115



MRI in predicting neurodevelopmental outcome in full-term infants with hypoxic-ischaemic encephalopathy in the era of therapeutic hypothermia. Eur. J. Paediatr. Neurol. 20, 855–864. doi: 10.1016/j.ejpn.2016.06.003


Inflammation. CNS Neurosci. Ther. 21, 367–373. doi: 10.1111/cns. 12365



cerebral palsy rats. Behav. Brain Res. 230, 108–115. doi: 10.1016/j.bbr.2012. 01.026


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Millar, Shi, Hoerder-Suabedissen and Molnár. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Phospholipase A2 of Peroxiredoxin 6 Plays a Critical Role in Cerebral Ischemia/Reperfusion Inflammatory Injury

Yu Shanshan1,2,3,4 , Jiang Beibei 1,2,3,4 , Tan Li 1,2,3,4 , Gao Minna1,2,3,4 , Lei Shipeng<sup>5</sup> , Peng Li 1,2,3,4 and Zhao Yong1,2,3,4 \*

<sup>1</sup>Department of Pathology, Chongqing Medical University, Chongqing, China, <sup>2</sup>Molecular Medical Laboratory, Chongqing Medical University, Chongqing, China, <sup>3</sup> Institute of Neuroscience, Chongqing Medical University, Chongqing, China, <sup>4</sup>Key Laboratory of Neurobiology, Chongqing Medical University, Chongqing, China, <sup>5</sup>Department of Respiratory Medicine, Jiangjin Center Hospital, Chongqing, China

Microglia-mediated inflammation is an important step in the progression of cerebral ischemia/reperfusion injury and the associated production of receptors of immunomoudulation, including Toll-like receptors (TLRs). Peroxiredoxin 6 (Prdx6) has been demonstrated as the endogenous antioxidant protein for its peroxidase properties. However, the role of the independent phospholipase A2 (iPLA2) activity of Prdx6 in stroke has not been well studied. In this study, we evaluated whether blocking the calcium-iPLA2 activity of Prdx6 using siRNA and inhibitors (1-hexadecyl-3-(trifluoroethgl)-sn-glycerol-2 phosphomethanol, MJ33) would have a critical effect on inflammatory brain damage. We conducted oxygen-glucose deprivation (OGD)/recovery (R) in vitro and middle cerebral artery occlusion (MCAO) in vivo in a microglia/neuron co-culture system and in rats. In vitro, we found that Prdx6-iPLA2 activity was associated with the secretion of neurotoxic inflammatory mediators interleukin1β (IL-1β), interleukin-17 (IL-17) and interleukin-23 (IL-23) and elevated expression of Toll-like receptor 2/4 (TLR2/4), leading to the formation of nuclear factor-kappa B (NF-κB), inducible nitric oxide synthase (iNOS) and cyclooxygenase-2 (COX-2) in microglial cells. In vivo, combined treatment with Prdx6-iPLA2 activity inhibitor MJ33 showed a greater diminution in neurologic deficits, cerebral infarction, brain water content and inflammatory molecules than Prdx6-siRNA treatment alone. Our findings provide new insight into Prdx6-iPLA2 function in the brain. Inhibition of Prdx6-iPLA2 activity by gene therapy and/or pharmacology may constitute a promising new therapeutic approach to the treatment of stroke.

Keywords: Prdx6-iPLA2 activity, cerebral ischemia/reperfusion, OGD/R, microglia/neuron co-culture system, MJ33

#### INTRODUCTION

Despite the complex pathogenesis of ischemic stroke, emerging evidence suggests that inflammation is an important step in the primary and secondary progression of cerebral ischemia/reperfusion injury (Shichita et al., 2012a,b; Kuang et al., 2014; Yu et al., 2015). Microglia in the brain plays a prominent role in initiating, sustaining and resolving post-ischemic inflammation

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Daniela Tropea, Trinity College, Dublin, Ireland Ertugrul Kilic, Istanbul Medipol University, Turkey

> \*Correspondence: Zhao Yong zhaoyong668@cqmu.edu.cn

Received: 20 September 2016 Accepted: 23 March 2017 Published: 05 April 2017

#### Citation:

Shanshan Y, Beibei J, Li T, Minna G, Shipeng L, Li P and Yong Z (2017) Phospholipase A2 of Peroxiredoxin 6 Plays a Critical Role in Cerebral Ischemia/Reperfusion Inflammatory Injury. Front. Cell. Neurosci. 11:99. doi: 10.3389/fncel.2017.00099 (Chen et al., 2011; Benakis et al., 2014). It has previously been recognized that inhibiting the activation of immunomoudulation receptors, including Toll-like receptors (TLRs) caused by microglial phagocytosis could prevent inflammatory neuronal death (Neher et al., 2011; Kuang et al., 2014). Among the TLRs, TLR2 and TLR4 are mainly expressed in the microglia of the brain and have been found to be more important than others in the immune response of the microglia (Wang et al., 2013; Lv et al., 2016). Therefore, the mechanisms that underlie microglia-neuron crosstalk are under extensive investigation for the development of innovative neuroprotective therapies for ischemic stroke.

Peroxiredoxin 6 (Prdx6) is the uniquely 1-Cys member of the peroxiredoxin family with both GSH peroxidase and calciumindependent phospholipase A2 (iPLA2) activities (Manevich and Fisher, 2005). Although the GSH peroxidase activity of Prdx6 has been widely studied in cell and animal models for its antioxidant function, the function of the iPLA2 activity of Prdx6 remains unclear. With loss- and gain-of-function mutations of Prdx6, Kim et al. (2011) found that the restoration of Prdx6 without iPLA2-mutant resulted in dramatic recovery of tumor necrosis factor-induced apoptosis. Blocking Prdx6-iPLA2 activity by its specific inhibitor 1-hexadecyl-3-(trifluoroethgl)-sn-glycerol-2 phosphomethanol (MJ33) could significantly protect the lung against damage from hyperoxia (Benipal et al., 2015). Recently, Prdx6-iPLA2 activity has been reported as giving off potent proinflammatory signals (Garcia-Bonilla and Iadecola, 2012; Shichita et al., 2012a,b). In human bronchial epithelial cells (BEAS2B), the production of interleukin-1β (IL-1β) was dependent on the iPLA2 activity of Prdx6 (Kim et al., 2011). Lee et al. (2013) reported that Prdx6-iPLA2 activity was associated with lung inflammation through activation of NADPH oxidase (NOX2). In primary cultured astrocytes, iPLA2 activity of Prdx6 induced astrocytic activation followed by increased proinflammatory cytokines (tumor necrosis factor-α (TNF-α) and IL-1β (Yun et al., 2015). All of these results strongly suggest that the iPLA2 activity of Prdx6 may be involved in the proinflammatory stimuli. However, until now, no study has described the iPLA2 activity of the Prdx6 function in microgliamediated neuroinflammation and cerebral ischemia/reperfusion.

Based on previous research, we hypothesized that the iPLA2 activity of Prdx6 has a critical role in cerebral ischemia/reperfusion inflammatory injury. To explore the molecular mechanism that underlines the critical effect of Prdx6-iPLA2 activity, we conducted oxygen-glucose deprivation/recovery (OGD/R) in vitro and middle cerebral artery occlusion (MCAO) in vivo, in a microgila/neuron co-culture system and in rats, respectively. siRNA and inhibitors (MJ33) were used to inhibit Prdx6-iPLA2 enzymatic activity.

### MATERIALS AND METHODS

## Rat Primary Microglia/Neuron Co-Culture System

Microglia and neurons were derived from 1-day-old Sprague-Dawley rats. All experimental protocols were approved by the Institutional Animal Care and Use Committee of Chongqing Medical University, China. All efforts were made to minimize suffering. Microglia-enriched cells were obtained by the method of Suzumura et al. (1987). After maintenance of mixed glial cultures for 9–12 days, microglial cells were separated by shaking flasks for 1 h using a rotary shaker at 200 rpm. The isolated cells were grown in DMEM/F12 supplemented with 10% FBS and 1% penicillin/streptomycin. When the microglia's growth density reached 55%–60%, cortex neuron-enriched cultures were obtained according to our previously published methods (Chen et al., 2016; Li et al., 2016). The purity of the primary neuron cultures was over 90%, as determined by neuron-specific marker NeuN.

The rat primary microglia/neuron co-culture system was performed as described previously with some changes (Correa et al., 2013; Bi et al., 2014). Briefly, a Transwell co-culture system utilized the non-contact Transwell inserts. Then the primary microglia and neurons were co-cultured for a period of 24 h, sharing the same culture medium containing 2% B27 and 1% penicillin/streptomycin through a 0.4-µm transmembrane that prevented cell migration but allowed small molecule exchange.

#### Oxygen Glucose Deprivation and Regeneration (OGD/R) and Groups

To simulate ischemic conditions, cultures were exposed to OGD/R as we previously described (Chen et al., 2016; Li et al., 2016). The medium was replaced by Neurobasal media without glucose and the cells were placed in an incubator with 1% O2, 94% N<sup>2</sup> and 5% CO<sup>2</sup> concentration. After a 6 h OGD, reoxygenation was carried out by transferring the cells to the normal cell culture incubator for 24 h. The co-culture systems were divided into five groups: (1) control group (Control): microglia without any treatment; (2) OGD/R group (OGD/R); (3) scramble group (Scramble): microglia treated with a scramble of siRNA and subjected to OGD/R; (4) Prdx6 iPLA2 activity siRNA group (Prdx6-iPLA2 siRNA): microglia transfected with Prdx6-iPLA2 siRNA and subjected to OGD/R; and (5) MJ33 group (MJ33): microglia given MJ33 treatment (50 µmol/L) and subjected to OGD/R.

#### Prdx6-iPLA2 siRNA

Lentivirus was supplied by Neuron Biotech (Shanghai, China). In order to create siRNA in the iPLA2 activity of Prdx6, we knocked down Prdx6 with lentivirus (NCBI accession no. NM\_053576.2 → NP-446028.1; shRNA, 5<sup>0</sup> -ACAGCCCGTGTGGTATTCAT-3 0 ). At the same time, reintroduction of over-expressed Prdx6 without iPLA2 activity was performed by Ser32 gene mutation into the knockdown cells. The iPLA2 assay kit (Xinyu Biological Technology Co, Shanghai, China) and GSH assay kit (Nanjin Jiancheng Bioengineering Institute, Nanjing, China) was used to detect the lentivirus interference efficiency by measuring the iPLA2 activity and GSH activity of Prdx6.

#### MTS Assay

Neuron viability was measured by an MTS assay kit (Nanjin Jiancheng Bioengineering Institute, Nanjing, China) to detect any remaining dehydrogenase activity in living cells. After the co-culture system was treated with OGD/R, a total of 10 uL MTS was directly added to the neuron, then incubated for 2 h to allow MTS to metabolize to formazan. The result was analyzed at 490 nm using a microplate reader (Bio-Rad). The neuron viability was expressed as relative percentage compared with control cells. Number of experiments: 9.

#### Lactate Dehydrogenase (LDH) Assay

The release of Lactate dehydrogenase (LDH) by microglia was measured using the LDH assay kit according to manufacturer's instructions (Nanjin Jiancheng Bioengineering Institute, Nanjing, China). Following treatment, the medium of the cell co-culture system from each group was collected and transferred to new 96-well plates, then mixed with the reaction solution provided in the kit. The optical density was measured at 450 nm using a microplate reader (Bio-Rad). Number of experiments: 9.

#### Animals and Groups

Adult male Sprague-Dawley rats weighing 270–310 g were bred at and obtained from the Laboratory Animal Center of the Chongqing Medical University. All rats were housed in a colony room with food and water available before the operation under optimal conditions (12/12 h light/dark cycle with humidity 60 ± 5%, 22 ± 3 ◦C). Experimental animals were randomly allocated to the following groups: (1) sham group (Sham); (2) MCAO group (MCAO): underwent MCAO; (3) scramble group (Scramble): treated with scramble of Prdx6 siRNA and subjected to MCAO; (4) Prdx6 siRNA group (Prdx6 siRNA): treated with Prdx6 siRNA and subjected to MCAO; and (5) Prdx6 siRNA + MJ33 group (Prdx6 siRNA + MJ33): treated with MJ33 (0.5 µmol/kg, by tail vein, at 24 h before MCAO) and Prdx6-siRNA and subjected to MCAO.

#### Administration of Prdx6 siRNA

Prdx6 siRNA (sense primer 5-CUUCCACGAUUUCCUAGG ATT-3 and antisense primer 5-UCCUAGGAAAUCGUGGAA GTT-3) were designed and chemically synthesized by GenePharma Corporation, Shanghai, China. A scramble of Prdx6 siRNA, which has the same nucleotide composition of the target gene siRNA with no sequence homology to any known rat genes, was used as the control. Rats were anesthetized, and 6 µL

siRNA was injected into the left lateral cerebral ventricle, siRNA was slowly injected into the left lateral ventricle over a 5 min duration using a Hamilton microsyringe with the coordinates of 1.0 mm posterior to the bregma, 2.0 mm lateral to the midline, and 3.5 mm ventral to the surface of the skull under the guidance of a stereotaxic instrument. After injection, the needle was held in place for 5 min and then removed slowly.

#### Middle Cerebral Artery Occlusion Model

The MCAO technique was used to induce transient focal cerebral ischemia according to our previously published methods (Chen et al., 2016; Li et al., 2016). Briefly, ischemia was produced by advancing the tip of a rounded 0.32 mm monofilament nylon suture (Beijing Sunbio Biotech Co. Ltd, Beijing, China) into the left internal carotid artery through the common carotid artery stump and gently advanced to occlude the MCA. After 1 h of occlusion, the thread was withdrawn to restore blood flow. Sham-operated animals received the same procedure, with the left common carotid artery isolated, but not occluded. Rats that did not show neurological deficits after reperfusion (neurological score <2) were excluded from the study, as were animals that died after ischemia induction. Rats that showed neurological deficits immediately after reperfusion (neurological score =2) but were found to be experiencing skull-base or subarachnoid hemorrhage were also excluded from the study.

#### Evaluation of Neurological Deficit Score

Neurological deficit scores were assessed by an examiner blinded to the experimental groups after 24 h of reperfusion. The deficits were scored on a modified scoring system developed by Longa et al. (1989) as follows: 0, no neurological deficits; 1, failure to extend right forepaw fully; 2, circling to right; 3, falling to right; and 4, does not walk spontaneously and has depressed levels of consciousness. The higher the neurological deficit score, the more severe the impairment of motor motion injury. Nine animals were taken for each group. Brains from these rats were analyzed for water content, infarct volume, ELISA analysis, Western blot, and real-time qPCR.

#### Measurement of Brain Edema

Brain water content was measured using the standard wet-dry method. The brains were rapidly removed and dissected into the ipsilateral and contralateral hemispheres. The two hemispheres were immediately weighed on an electronic balance to obtain the wet weight and then heated in an oven at 100◦C for 48 h to obtain the dry weight. Brain water content was calculated using the following formula: brain water content (%) = (1 − dry weight/wet weight) × 100%.

(IL-1β), interleukin-17 (IL-17) and interleukin-23 (IL-23) in culture medium in response to OGD/R (A–C). Number of experiments: 9. Values are mean ± SEM, <sup>∗</sup>p < 0.05 vs. Control; ∗∗p < 0.01 vs. Control; #p < 0.05 vs. Scramble; ##p < 0.01 vs. Scramble.

#### Infarct Volume Analysis

The brain infarct area was evaluated using 2,3,5 triphenyltetrazolium chloride (TTC) staining. TTC was converted to red formazan product in the presence of a functioning mitochondrial electron transport chain. The infarct tissue areas of all coronal sections of each brain were measured using Image-Pro Plus software as described previously. The corrected volume was calculated according to the formula: percentage of infarct volume (%) = [total infarct volume − (left hemisphere volume − right hemisphere volume)]/right hemisphere volume × 100% (Chen et al., 2012).

#### ELISA Assay for IL-1β, IL-17 and IL-23

To investigate the level of interleukin cytokine expression, a rat IL-1β, interleukin-17 (IL-17) and interleukin-23 (IL-23) enzymelinked immunosorbent assay (ELISA) kit (Nanjing Jiancheng; R&D Systems) was used according to the manufacturer's instructions. Absorbance was determined at 450 nm by spectrometry.

#### Real-Time qPCR Analysis

For real-time RT-PCR, the total RNA from primary cultured microglia cells and brain tissues of rats was extracted and purified as previously reported (Yamada et al., 2002). Then the RNA was converted to complementary DNA (cDNA) by reverse transcription (Bio-Rad). Then, cDNA was performed by a PrimeScriptTM RT regent kit (TaKaRa Biotechnology, Dalian). The primer sequences (Sangon Biotech, Shanghai, China) were as follows: Prdx6 (forward primer, 5<sup>0</sup> -ACAgCCCgTgTggTA TTCAT-3<sup>0</sup> , reverse primer, 5<sup>0</sup> -CTCTCTCCCTTCTTCCAGTC AA-3<sup>0</sup> ), inducible nitric oxide synthase (iNOS; F, 5<sup>0</sup> -gTgCTAA TgCggAAggTCAT-3<sup>0</sup> , R, 5<sup>0</sup> -gAAggCgTAgCTgAACAAgg-3<sup>0</sup> ),

TLR2 (F, 5<sup>0</sup> -gAgTCTgCTgTgCCCTTCTC-3<sup>0</sup> , R, 5<sup>0</sup> -gCTTTCTT gggCTTCCTCTT-3<sup>0</sup> ), TLR4 (F, 5<sup>0</sup> -TggCATCATCTTCATTgTC C-3<sup>0</sup> , R, 5<sup>0</sup> -CAgAgCATTgTCCTCCCACT-3<sup>0</sup> ), cyclooxygen ase-2 (COX-2; F, 5<sup>0</sup> -GTTGCTGGGGGAAGGAATGT-3, R, 5<sup>0</sup> -A GAAGCGTTTGCGGTACTCA-3<sup>0</sup> ), nuclear factor-kappa B (NF-κB; F, 5<sup>0</sup> -ACGACGATCCTTTCGGAACT-3<sup>0</sup> , R, 5<sup>0</sup> - TGTTGACAGTGGTATATCTGTTGAA-3<sup>0</sup> ), β-actin (F, 5<sup>0</sup> - CACCCGCGAGTACAACCTTC-3<sup>0</sup> , R, 5<sup>0</sup> -CCCATACCCACCA TCACACC-3<sup>0</sup> ). The data were analyzed using a Bio-Rad CFX96 Connect Real-Time system. The mRNA levels were standardized to β-actin, and the values were expressed as the fold change of the threshold cycle value for the control by the 2−11Ct method. Three independent experiments were performed.

#### Western Blot Analysis

Proteins were obtained and homogenized in the RIPA buffer. Equivalent amounts of protein (30 µg) were loaded and separated by 10% SDS-PAGE gels, then transferred to polyvinylidene difluoride (PVDF) membranes (0.45 mm). The membranes were washed in Tris-buffered saline containing 0.05% Tween-20 (TBST) followed by blocking for 1 h using 5% nonfat milk in TBST at room temperature, then incubated at 4◦C overnight with the following primary antibodies: monoclonaln antibody to Prdx6 (1:1000; ab133348, Abcam, Cambridge, MA, USA), rabbit polyclonal antibody to TLR2 and TLR4 (1:400, BA1716/BA1717, Boster Biological Technology, China), rabbit polyclonal antibody to NF-κB (1:1000; sc-109, Santacruz, CA, USA), rabbit monoclonal antibody to iNOS (1:500, bs0162R, Bioss, China), rabbit monoclonal antibody to COX-2 (1:500, 55070-1-AP, Proteintech, China,) and mouse monoclonal antibody to β-actin (1:10,000, D110007-0100, Songon Biotech, China). The next day, the blots were washed and incubated for 1 h with respective HRP-conjugated secondary antibodies (ZSGB-BIO, Beijing, China; dilution 1:3000) at room temperature. Finally, the protein was visualized using the ECL kit (Millipore, Temecula, CA, USA). The results were quantified using Quantity One 1-D analysis software, and β-actin was used as the internal control. Three independent experiments were taken.

#### Statistical Analysis

All data were expressed as mean ± SEM and analyzed using Statistical Package for Social Sciences (SPSS) 16.0 software. One-way analysis of variance (ANOVA) were performed, followed by post hoc tests with the least significant difference (LSD; under variance homogeneity). Moreover, Kruskal-Wallis and Welch's ANOVA were both used for heteoscedastic data. A value of P < 0.05 was considered statistically significant.

#### RESULTS

## The Efficiency of Prdx6-iPLA2 siRNA

As mentioned in "Materials and Methods" Section, we used an siRNA and, following functional recovery, performed mediated knockdown of Prdx6-iPLA2 activity. Previous research has shown that Prdx6 siRNA can reduce both iPLA2 and GSH activity (Jo et al., 2013), In order to detect the siRNA efficiency, we added a Prdx6 siRNA group as a control. First, qPCR and Western blotting were used to ascertain whether the Prdx6 mRNA and protein levels were reduced in the microglia (three independent experiments were performed). **Figures 1A,B** show that considerable reduction in Prdx6 was observed in the Prdx6 siRNA group (P < 0.05). No statistical difference in Prdx6 expression was observed between the OGD/R group, Scramble group, Prdx6-iPLA2 siRNA group and MJ33 group. Next, an iPLA2 ELISA kit was used to measure the iPLA2 activity in microglia (**Figure 1C**). Compared with the Sham group, the iPLA2 activity was increased in the OGD/R group and Scramble group. Treatment with Prdx6 iPLA2 activity siRNA or iPLA2 inhibitors (MJ33) resulted in a significant decrease of iPLA2 activity compared with the Scramble group. Additionally, we detected GSH peroxidase activity (**Figure 1D**). Prdx 6 siRNA suppressed GSH activity. Both Prdx6-iPLA2 siRNA and MJ33 had no effect on GSH peroxidase activity (P > 0.05). All of these results suggest that our methods had interference efficiency of Prdx6-iPLA2 activity in microglial cells.

#### Effect of Prdx6-iPLA2 Activity on Neuron Viability and Damage in Response to OGD/R

The MTS assay was used to measure the effect of Prdx6 iPLA2 activity on neuron viability after OGD/R exposure (**Figure 2A**). Cell viability was significantly decreased in the OGD/R group compared with the untreated group (P < 0.01). The total number of viable neurons increased to 65 ± 6.4% and 58.8 ± 7% in the Prdx6-iPLA2 siRNA and MJ33 groups, respectively (**Figure 2A**). In parallel, the release of LDH in neurons was measured (**Figure 2B**). The siRNA of Prdx6 iPLA2 could decrease the LDH release from neurons compared with the control group (n = 9, P < 0.05). MJ33 is a fluorinated phospholipid analog that shows relatively tight binding to Prdx6 (Manevich and Fisher, 2005). MJ33 treatment had similar results. These results suggest that the Prdx6-iPLA2 siRNA could reduce neuron damage after OGD/R.

#### Effect of Prdx6-iPLA2 Activity on the Release of IL-1β, IL-17 and IL-23 in Culture Medium in Response to OGD/R

In order to measure the effects of Prdx6-iPLA2 activity on the expression of inflammatory mediators, ELISA assays was performed. As shown in **Figures 3A–C**, Prdx6 iPLA2 siRNA significantly decreased the levels of IL-1β, IL-17 and IL-23–35.25 ± 4.2 (pg/ml; **Figure 3A**, n = 9, P < 0.05),

FIGURE 6 | Effect of Prdx6-siRNA and MJ33 on the expression of Prdx6 and iPLA2 activity following middle cerebral artery occlusion (MCAO). qPCR (A) and Western blot (B) results showed that the expression of Prdx6 was reduced in the Prdx6-siRNA group. ELISA (C) results also showed that a significant decrease of iPLA2 activity was found in the Prdx6 + MJ33 group. Results are expressed as mean ± SEM of three independent experiments. &p < 0.01 vs. Sham; ∗∗p < 0.01 vs. MCAO; ##p < 0.01 vs. Prdx6-siRNA.

53 ± 4.5 (pg/ml; **Figure 3B**, P < 0.01) and 49 ± 5.4 (pg/ml; **Figure 3C**, n = 9, P < 0.01), respectively, compared to the Scramble group. MJ33 treatment also decreased these mediators. These results suggest that Prdx6-iPLA2 activity may affect the release of some inflammatory cytokines.

#### Effect of Prdx6-iPLA2 Activity on TLR2 and TLR4 Signaling Pathway in Response to OGD/R

We next investigated whether the TLR2/4 signaling pathway was involved in the Prdx6-iPLA2-mediated activation of inflammatory mediators. After OGD 6 h/R 24 h, microglial cells were harvested for Western blot analysis. As shown in **Figure 4**, the Scramble group showed no changes in expression levels of TLR2 or TLR4, while Prdx6-iPLA2 siRNA markedly reduced both the mRNA and protein levels of TLR2 and TLR4, relative to the Scramble group. Similar results were obtained with MJ33 treatment.

#### Effect of Prdx6-iPLA2 Activity on the Expression of NF-κB, iNOS and COX- 2 in Response to OGD/R

The expression of molecules involved in the intracellular TLR2 and TLR4 signaling pathways was detected. The mRNA expression levels of NF-κB, iNOS and COX-2 were decreased by Prdx6-iPLA2 siRNA treatment (**Figures 5A–C**). In **Figures 5D–G**, the Western blot results and quantification showed that the protein expression levels of NF-κB, iNOS and COX-2 in the Prdx6-iPLA2 siRNA group decreased by 50.27% (P < 0.01), 66.67% (P < 0.05) and 58.44%

#### FIGURE 7 | Effect of Prdx6 siRNA and MJ33 on cerebral infarct volume, neurologic deficit scores and brain water content following transient MCAO. (A) Representative images of TTC-stained sections 24 h after MCAO. (B) Prdx6 siRNA significantly increased cerebral infarct volume compared with the MCAO group. Combine treatment with Prdx6 siRNA and MJ33 attenuated the injury. Neurological function (C) and brain water content (D) were tested after 24 h of MCAO in rats. Values are mean ± SEM of nine animals in each group. &p < 0.01 vs. Sham; ∗∗p < 0.01 vs. Scramble; #p < 0.05, vs. Prdx6 siRNA; ##p < 0.01 vs. Prdx6 siRNA.

(P < 0.01), respectively, relative to the Scramble group. Consistent with previous studies, blocking Prdx6 iPLA2 activity by MJ33 significantly decreased the injury associated with inflammation (Kuang et al., 2014).

## Effect of Prdx6 siRNA and MJ33 on the Expression of Prdx6 and iPLA2 Activity Following MCAO

We next investigated the role of iPLA2 activity of Prdx6 in transient cerebral ischemia and the potential mechanism underlying the observations. Western blotting and real-time PCR were performed to measure the expression of Prdx6 at mRNA and protein levels (**Figures 6A,B**). Then, the iPLA2 activity of Prdx6 was detected by ELISA (**Figure 6C**). Significant decreases in Prdx6 mRNA and proteins levels were observed in the Prdx6 siRNA group. Although additional treatment with MJ33 did not significantly influence the expression of Prdx6 stimulated by Prdx6 siRNA (**Figures 6A,B**, P > 0.05), additional injection with MJ33 resulted in a significant decrease of iPLA2 activity (**Figure 6C**, P < 0.01).

#### Effect of Prdx6 siRNA and MJ33 on Cerebral Infarct Volume, Neurologic Deficit Scores and Brain Water Content Following MCAO

As shown in **Figure 7**, compared with the MCAO group, Prdx6 siRNA was associated with significantly aggravated infarct volume (**Figures 7A,B**, P < 0.01). However, this increase was mitigated by additional treatment of the rat with MJ33, an inhibitor of iPLA2 activity. Compared with the Prdx6 siRNA group, similar attenuation was observed in the neurological deficit analysis (**Figure 7C**, n = 9, P < 0.01) and brain water content (**Figure 7D**, n = 9, P < 0.01) in the Prdx6 siRNA + MJ33 group.

### Effect of Prdx6 siRNA and MJ33 on the Producion of IL-1β, IL-17 and IL-23 in Cortex Following MCAO

To detect the production of IL-1β, IL-17 and IL-23 in brain ischemia/reperfusion, ELISA were performed. As shown in **Figure 8**, Prdx6 siRNA caused increased release and expression of IL-1β(A), IL-17 (B) and IL-23 (C). However, additional treatment with MJ33 significantly blocked the increase in these inflammatory cytokines in rats stimulated by Prdx6 siRNA treatment.

## Effect of Prdx6 siRNA and MJ33 on TLR2 and TLR4 Expression in the Cortex Following MCAO

To determine the underlying molecular effects of iPLA2 activation in vivo, rat cortical tissue from MCA regions was harvested and analyzed by real-time PCR and Western blotting. As shown in **Figure 9**, the expression of TLR2 and TLR4 in cortex was significantly increased after

FIGURE 8 | The influence of Prdx6-siRNA and MJ33 on the production of inflammatory cytokines IL-1β, IL-17 and IL-23. The ELISA results indicated that IL-1β (A), IL-17 (B) and IL-23 (C) were upregulated by Prdx6 siRNA after MCAO. These increase were attenuated by additional treatment of rats with MJ33. The results are expressed as the mean ± SEM of nine animals in each group. &p < 0.01 vs. Sham; ∗∗p < 0.01 vs. Scramble; #p < 0.05, vs. Prdx6 siRNA.

MCAO. Prdx6 siRNA significantly increased the expression of TLR2 and TLR4 in comparison to the Sham group. Compared with the Prdx6 siRNA group, both the mRNA (**Figures 9A,B**) and protein levels (**Figures 9C–E**) of TLR2 and TLR4 were significantly decreased in the same brain areas by combined exposure to Prdx6 siRNA and MJ33.

### Effect of Prdx6 siRNA and MJ33 on the Expression of NF-κB, iNOS and COX- 2 in the Cortex Following MCAO

We next investigated changes in the mRNA and protein expression of molecules involved in the intracellular TLR2 and TLR4 signaling pathways. MCAO caused increased protein expression of NF-κB, iNOS and COX-2 in the cortex compared with the Sham group (**Figures 10A–C**). Prdx6 siRNA upregulated the expression of these proteins compared with the control groups (**Figures 10D–G**). Compared with the Prdx6 siRNA group, combined exposure to Prdx6 siRNA and MJ33 significantly downregulated the mRNA and protein expression of NF-κB, iNOS and COX-2.

experiments. &p < 0.01 vs. Sham; ∗∗p < 0.01 vs. Scramble; #p < 0.05, vs. Prdx6 siRNA; ##p < 0.01 vs. Prdx6 siRNA.

# DISCUSSION

In the present study, we explored the function of Prdx6 iPLA2 in cerebral ischemia/reperfusion injury in both OGD/R microglia/neuron co-culture and MCAO models. In the co-culture system, we employed siRNA knockdown and inhibition of Prdx6-iPLA2 activity in the microglia; both resulted in increased neuron survival. We further found that Prdx6-iPLA2 was associated with the secretion of neurotoxic inflammatory mediators (IL-1β, IL-17 and IL-23) and the levels of the TLR2/4 signaling pathway in microglial cells. in vivo, consistent with previous reports, the knockdown of Prdx6 by antisense (Prdx6-siRNA) revealed that loss of Prdx6 contributed to increased cerebral infarct volume and poor functional outcomes (Chhunchha et al., 2013; Pan et al., 2014). However, combined treatment with Prdx6 siRNA and Prdx6-iPLA2 inhibitor MJ33 showed a greater diminution in neurologic deficits, cerebral infarction, brain water content and inflammatory molecules than Prdx6-siRNA treatment alone. Thus, the iPLA2 activity of Prdx6 may play a critical role in cerebral ischemia/reperfusion injury, which might target TLR2/4 inflammatory cascades.

Emerging evidence suggests that macrophage/microgliainduced neuroinflammation often precedes and triggers

neuronal death in cerebral ischemic disorders. In comparison to monolayer cultures, we established a Transwell co-culture model to better explore microglia-neuron crosstalk. In this system, the two kinds of cells share the same medium and interact through diffusible molecules only. As reported by others, we adopted the Neurobasal medium to meet higher nutrition requirements for neurons (Xing et al., 2013). Before OGD/R treatment, microglia and neurons were co-cultured 24 h for utmost microglial activation. The factors for the harmful inflammatory secreted by microglia have not been clearly defined, but activation of receptors of the innate immune system, including TLR2, TLR4 and their downstream inflammatory cytokines (NF-κB, COX-2 and NOS2) has emerged as a key step in the signaling cascade (Tu et al., 2011; Wang et al., 2013; Lv et al., 2016). Specific molecules blocking TLR2/4 signaling or its endogenous danger-associated molecular patterns (DAMPs) may be a novel therapeutic strategy for post-stroke neuronal inflammation and brain injury (Tu et al., 2010; Kuang et al., 2014; Yu et al., 2015).

Prdx6 is a bifunctional enzyme that harbors iPLA2 (Ca[2+] independent phospholipase A2) activity in addition to its GSH peroxidase function. Prdx6-iPLA2, rather than other iPLA2 enzymes, was found to be harmful to the cells (Ellison et al., 2012; Krishnaiah et al., 2013). iPLA2 activity of Prdx6 involved in the TNF-induced apoptosis and the proinflammatory response (Kim et al., 2011). In the pulmonary endothelium and alveolar macrophages, Prdx6-iPLA2 activity modulates NOX2 activation via lysophosphatidic acid receptor signaling (Vázquez-Medina et al., 2016). In this study, our focus was the function of Prdx6-iPLA2 activity in the ischemic brain, which is especially at risk for microglia-mediated neuroinflammatory injury associated with TLR2/4 activation.

Single-point deficient mutation by lentivirus gene transduction in primary cerebral cells seemed difficult under our experimental conditions. Here, we described an siRNA and, following functional recovery, performed a mediated knockdown of Prdx6-iPLA2 activity. Finally, we detected successful iPLA2 siRNA with iPLA2 assays. The Prdx6 iPLA2 siRNA showed the same results as MJ33: both had no effect on GSH peroxidase activity (Fisher et al., 1992; Benipal et al., 2015). Previous reports have shown that the iPLA2 activity of Prdx6 is involved in the proinflammatory response (Kim et al., 2011; Shichita et al., 2012a,b). In our study, we found that knockdown of Prdx6-iPLA2 in microglia significantly moderated OGD/R injury to co-culture neurons, which may be associated with decreases in both the RNA and protein levels of TLR2/4 expression. Like other Prdx proteins, Prdx6 proteins have an active region of TLR2/4 activation and are extracellularly released over 12 h after stroke onset, which coincides with the timing of leukocyte infiltration (Shichita et al., 2012a,b). iPLA2 activity of Prdx6 induced the release of arachidonic acid (AA), which is involved in the inflammatory response (Kim et al., 2011). In addition, AA's upregulation of TLR4 and downregulation of PPARγ may be associated with acute pancreatitis (Mateu et al., 2015). All of these findings suggest that TLR signaling is effected directly or indirectly by Prdx6 iPLA2 activity. Recently, two opposing functions of Prdx6 were shown in brain cells: first, intracellular Prdx6 is thought to be neuroprotective; second, when Prdx6 is released from necrotic cells, it functions as a strong TLR2 and TLR4 stimulator (Shichita et al., 2012a,b; Kuang et al., 2014). It is interesting to note that Prdx6-iPLA2 may be associated with the same inflammation pathway in our present experiment. Because the stimulation of cells may result in Prdx6 binding to the cell membrane, that may activate its iPLA2 activity (Ambruso et al., 2012; Ellison et al., 2012; Krishnaiah et al., 2013). This suggests that there may be a relationship between the Prdx6-iPLA2 and its extracellular danger signal, which requires further study.

#### REFERENCES


In conclusion, our study demonstrated that the iPLA2 of Prdx6 plays important physiological roles in cerebral inflammatory injury both in vitro and in vivo. siRNA of Prdx6 iPLA2 was accompanied by reduced expression of inflammation cytokines IL-1β, IL-17 and IL-23.The iPLA2 activity of Prdx6 may contribute to neuroinflammation by regulating TLR2/4, leading to the formation of neurotoxic levels of NF-κB, iNOS and COX-2. Our findings provide new insight into Prdx6 iPLA2 function in the brain. Inhibition of Prdx6 iPLA2 activity by gene therapy and/or pharmacological means may constitute a promising new therapeutic approach to the treatment of stroke.

#### AUTHOR CONTRIBUTIONS

YS, JB, TL and ZY conceived and designed the experiments; YS, JB, TL and GM conducted the experiments; YS, JB, TL, LS and PL analyzed the results; GM, LS and PL contributed materials and analysis tools. YS wrote the article; all authors reviewed the manuscript.

#### ACKNOWLEDGMENTS

This work was supported by National Natural Science Foundation of Chongqing (No.cstc2016jcyjA0135), Natural Science Foundation of Chongqing Education Committee, China (No. KJ1500230).


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer DT and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Shanshan, Beibei, Li, Minna, Shipeng, Li and Yong. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Calpastatin Overexpression Preserves Cognitive Function Following Seizures, While Maintaining Post-Injury Neurogenesis

Vanessa M. Machado1,2,3 , Ana Sofia Lourenço1,2,3 , Cláudia Florindo2,3 , Raquel Fernandes <sup>3</sup> , Caetana M. Carvalho<sup>1</sup> and Inês M. Araújo2,3,4 \*

<sup>1</sup>Center for Neuroscience and Cell Biology, University of Coimbra, Coimbra, Portugal, <sup>2</sup>Center for Biomedical Research, CBMR, University of Algarve, Faro, Portugal, <sup>3</sup>Regenerative Medicine Program, Department of Biomedical Sciences and Medicine, University of Algarve, Faro, Portugal, <sup>4</sup>Algarve Biomedical Center, University of Algarve, Faro, Portugal

In the adult mammalian brain, new neurons continue to be produced throughout life in two main regions in the brain, the subgranular zone (SGZ) in the hippocampus and the subventricular zone in the walls of the lateral ventricles. Neural stem cells (NSCs) proliferate in these niches, and migrate as neuroblasts, to further differentiate in locations where new neurons are needed, either in normal or pathological conditions. However, the endogenous attempt of brain repair is not very efficient. Calpains are proteases known to be involved in neuronal damage and in cell proliferation, migration and differentiation of several cell types, though their effects on neurogenesis are not well known. Previous work by our group has shown that the absence of calpastatin (CAST), the endogenous inhibitor of calpains, impairs early stages of neurogenesis. Since the hippocampus is highly associated with learning and memory, we aimed to evaluate whether calpain inhibition would help improve cognitive recovery after lesion and efficiency of post-injury neurogenesis in this region. For that purpose, we used the kainic acid (KA) model of seizure-induced hippocampal lesion and mice overexpressing CAST. Selected cognitive tests were performed on the 3rd and 8th week after KA-induced lesion, and cell proliferation, migration and differentiation in the dentate gyrus (DG) of the hippocampus of adult mice were analyzed using specific markers. Cognitive recovery was evaluated by testing the animals for recognition, spatial and associative learning and memory. Cognitive function was preserved by CAST overexpression following seizures, while modulation of post-injury neurogenesis was similar to wild type (WT) mice. Calpain inhibition could still be potentially able to prevent the impairment in the formation of new neurons, given that the levels of calpain activity could be reduced under a certain threshold and other harmful effects from the pathological environment could also be controlled.

#### Keywords: calpains, calpastatin, cognition, hippocampus, post-injury neurogenesis

**Abbreviations:** BrdU, 5-bromo-2<sup>0</sup> -deoxyuridine; CAST, calpastatin; DCX, doublecortin; DG, dentate gyrus; EdU, 5-ethynyl-2<sup>0</sup> -deoxyuridine; FC, fear conditioning; GZ, granular zone; hCAST, mice overexpressing calpastatin; KA, kainic acid; NeuN, neuronal nuclei; NSC, neural stem cell; OR, object recognition; SAL, saline; SGZ, subgranular zone; WM, water maze; WT, wild type.

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Daniela Tropea, Trinity College, Dublin, Ireland Xiaochu Lou, University of Wisconsin-Madison, USA

> \*Correspondence: Inês M. Araújo imaraujo@ualg.pt

Received: 11 October 2016 Accepted: 22 February 2017 Published: 23 March 2017

#### Citation:

Machado VM, Lourenço AS, Florindo C, Fernandes R, Carvalho CM and Araújo IM (2017) Calpastatin Overexpression Preserves Cognitive Function Following Seizures, While Maintaining Post-Injury Neurogenesis. Front. Mol. Neurosci. 10:60. doi: 10.3389/fnmol.2017.00060

#### Machado et al. Calpastatin Overexpression, Seizures and Cognition

# INTRODUCTION

New neurons in the adult mammalian brain are known to originate from mainly two regions, the subventricular zone, in the walls of the lateral ventricles, and the subgranular zone (SGZ), in the dentate gyrus (DG) of the hippocampus (Bond et al., 2015). Neural stem cells (NSCs) from the subventricular zone migrate long distances, through a rostral migratory stream, into the olfactory bulb, differentiating into interneurons. NSCs from the SGZ, in turn, migrate shorter distances into the granular zone (GZ), where they fully mature into granule neurons after 2 months (Aimone et al., 2014; Jin, 2016). Functions of these newborn cells from both regions have been consistently associated with learning and memory (Deng et al., 2010; Lazarini and Lledo, 2011).

When a brain lesion occurs, behavioral traits and adult neurogenesis can become altered. In neurodegenerative diseases, neurogenesis is mostly hampered, but in acute disorders, such as temporal lobe epilepsy, NSC proliferation can increase, in a possible attempt to repair the damage. However, this repair is limited by impaired cell migration or decreased survival of new neurons (Kaneko and Sawamoto, 2009; Ma et al., 2009).

Another feature of brain damage is excitotoxicity, which leads to an increase of cellular levels of calcium, activating several proteases, including calpains (Neumar et al., 2001). Calpains are ubiquitously expressed calcium-activated proteases, with only one known natural endogenous inhibitor, a protein called calpastatin (CAST; Murachi, 1984). In the CNS, calpain inhibition has been shown to afford neuroprotection, showing improved neuronal function and limiting neuronal damage in several brain disorders (Saez et al., 2006). Although little is known about their contribution to neurogenesis, calpains were seen to be involved in cell proliferation, migration and differentiation in other systems (Rock et al., 2000; Yajima and Kawashima, 2002; Lokuta et al., 2003; Parnaud et al., 2005; Qiu et al., 2006; Shimada et al., 2008; Kashiwagi et al., 2010; Kuchay et al., 2012).

Previous studies by our group show that increased calpain activity, through lack of CAST, can impair early stages of neurogenesis in the hippocampus, the main structure involved in learning and memory (Machado et al., 2015). Moreover, the kainic acid (KA) model of seizure-induced hippocampal lesion presents excitotoxic damage mediated by calpains (Araújo et al., 2008) and is characterized by pathologic alterations in hippocampal neurogenesis (Carreira et al., 2010). With this work, we proposed to assess whether overexpression of CAST (hCAST mice) could improve cognitive recovery and neurogenesis after KA-induced hippocampal lesion.

# MATERIALS AND METHODS

#### Animals

Three-month old male and female hCAST mice (Rao et al., 2008), in a C57BL/6 background, and their wild type (WT) littermates, were used in this study. hCAST mice present CAST overexpression, due to the cloning of a human CAST construct (Hitomi et al., 1998) into a Thymocyte differentiation antigen 1.1 expression cassette (Thy1). The animals were kept in our animal facilities, in a room with controlled temperature (21 ± 1 ◦C) and humidity (55%), with food and water ad libitum in a 12-h dark:light cycle. All experiments were performed in accordance with institutional and European guidelines (2010/63/EU) for the care and use of laboratory animals. The procedures performed in mice described in this work have been reviewed and approved by the Animal Welfare Body of the Center for Neuroscience and Cell Biology and have been approved by the Direcção Geral de Alimentação e Veterinária (reference 0421/000/000/2013).

### KA Treatment

Hippocampal damage was induced by administering KA subcutaneously (25 mg/kg, in a concentration of 5 mg/ml, in a saline (SAL) solution of 0.9% NaCl), as previously described by our group (Carreira et al., 2010). After KA administration, the animals went through several stages, according to a well-defined scale (Schauwecker and Steward, 1997): immobility (I), tail/forelimb extension/rigid posture (II), repetitive movements/head bobbing (III), rearing and falling (IV), continuous rearing and falling (V), severe tonic-clonic seizures (VI). Only mice that reached stage V or higher were used in this study. SAL-treated animals were used as controls.

# Behavior Analysis and Neuronal Differentiation

For the studies of short-term behavioral recovery, three different behavioral tests were performed in WT and hCAST mice, on the 3rd week after KA or SAL treatment (**Figure 1A**): object recognition (OR; **Figure 1C**), water maze (WM; **Figure 1D**) and fear conditioning (FC; **Figure 1E**), as explained further in more detail. To assess long-term behavioral recovery and neuronal differentiation after lesion, WT and hCAST mice were treated as shown in **Figure 1B**. The thymidine analog 5-ethynyl-2<sup>0</sup> deoxyuridine (EdU) was intraperitoneally administered to WT and hCAST mice on days 3, 4 and 5 after KA or SAL treatment, and the animals were sacrificed by transcardial perfusion after 8 weeks, when new neurons are fully mature (Aimone et al., 2014). Behavioral analysis was performed during the last week.

#### Object Recognition

For the study of recognition memory, the OR test (Bevins and Besheer, 2006) was used. Mice were presented to two similar objects, following habituation to the apparatus (40 × 40 × 40 cm<sup>3</sup> acrylic box), after which one of the objects was replaced by a novel one. Each stage of the test had the duration of 5 min, 2 h apart, and the apparatus and objects were cleaned with 10% ethanol between animals. In the presence of the objects, the percentage of time spent with each object (totalizing 100%) was measured. The analysis was done in videos acquired with a GoPro Hero (GoPro, Inc., San Mateo, CA, USA) during testing of 3–5 animals per group (for short-term recovery) and 12–15 animals per group (for long-term recovery), using the Any-maze software (version 4.99, Stoelting Co., Wood Dale, IL, USA).

FIGURE 1 | Experimental procedures for the study of behavioral recovery and neurogenesis after kainic acid (KA) or saline (SAL) treatment. Wild type (WT) and hCAST mice were treated with either SAL or KA (25 mg/kg, s.c.). (A) To study behavioral recovery in the short-term group, the cognitive behavior of the animals was tested on the 3rd week after treatment. (B) For the behavioral recovery in the long-term group and the study of neuronal differentiation, mice were given 5-ethynyl-2<sup>0</sup> -deoxyuridine (EdU; 50 mg/kg, i.p.) on days 3, 4 and 5, 8 weeks before sacrifice, and the cognitive behavior was tested during the last week. (C) On the first day, object recognition (OR) was tested. (D) Water maze (WM) trials were performed on the next 3 days and the final test, without the platform, was on day 4. (E) On the next day, fear conditioning (FC) trials were performed. One day later, the animals were at first left on the same context and, after 3 h, the context was completely modified and the cued test performed. The animals were sacrificed on the next day. (F) To study neural stem cell (NSC) proliferation, mice were given 5-bromo-2<sup>0</sup> -deoxyuridine (BrdU; four doses of 50 mg/kg, i.p.) on day 5, up to 12 h before sacrifice. (G) To study neuroblast migration, mice were sacrificed 14 days after treatment.

To evaluate spatial learning and memory, the WM test (Morris et al., 1982) was performed. During the first 3 days, the animals were trained to find a platform (10 cm diameter, carved to increase grip) hidden 0.5 cm below the water surface level in a circular pool (1.52 m diameter), in a fixed location. Visual cues on the walls around the pool were provided to help the mice locating the platform and the water temperature was maintained at 22◦C ± 1 ◦C. The pool was divided into four quadrants and, on each day of training, the animals were placed in a different quadrant, always facing the border of the pool, under the course of four 60 s trials, 30 min apart. The latency to escape was measured as the time each animal took to find and climb onto the platform. If the animal did not reach the platform within the course of the trial, the latency to escape was considered 60 s and the animal was left on the platform for 10 s, in order to learn the platform location. On the last day of testing, the platform was removed and the mice were placed in the quadrant opposite to the target quadrant (where the platform was previously located), facing the border of the pool. The time spent on each quadrant and the number of crossings over the previous platform location were measured for 60 s, to evaluate if the animals remembered where the platform was located during training. All animals were dried under an infrared light after each trial. The analysis was done in videos acquired with a GoPro Hero (GoPro, Inc., San Mateo, CA, USA) during testing of 4–5 animals per group (for short-term recovery) and 11–16 animals per group (for long-term recovery), using the Any-maze software (version 4.99, Stoelting Co., Wood Dale, IL, USA).

# Fear Conditioning

Associative learning and memory were tested by performing the FC test (Wehner and Radcliffe, 2004). The first day of testing consisted of learning to associate a cue tone (80 dB) with a footshock (0.7 mA). For this part of the test, mice were placed in a chamber (17 × 17 × 25 cm<sup>3</sup> ) with a grid floor inside a soundproof box (Stoelting Co., Wood Dale, IL, USA), with a background white noise, 70% ethanol scent and 100% lux lighting (context A). After 2 min (acquisition/habituation), the tone was played for 30 s and the footshock given during the last 2 s of tone. This was repeated two more times, after a 60 s rest. On the next day, the association of the context with the footshock was tested (context test), by placing the mice inside the chamber in the same conditions as during learning for 5 min, without presentation of tone or footshock. After 3 h, the association of the footshock with the tone (cued test) was tested, by placing the animals in a completely different environment for 3 min (habituation to new context) and then playing the tone for the last 3 min. The new context consisted of a background fan noise, vanilla scent, 10% lux lighting, and the previously used chamber was altered by using striped black and white walls and a white floor covering the grid (context B). During the whole FC testing, the chamber was cleaned with 10% ethanol between animals, and the fear behavior measured by calculating the percentage of freezing time, for 3–5 animals per group (for short-term recovery) and 9–16 animals per group (for long-term recovery), using the Any-maze software (version 4.99, Stoelting Co., Wood Dale, IL, USA).

# NSC Proliferation

Cell proliferation after KA or SAL treatment was assessed by the incorporation of the thymidine analog 5-bromo-2<sup>0</sup> -deoxyuridine (BrdU), as shown in **Figure 1F**. WT and hCAST mice were given either KA or SAL (9–14 animals per group) and BrdU was administered intraperitoneally (four doses of 50 mg/kg, 2 h apart) on day 5, when a peak of proliferation is known to occur in the DG after KA treatment (Carreira et al., 2010). The animals were sacrificed after at least 12 h, by transcardial perfusion.

#### Neuroblast Migration

In order to investigate neuroblast migration after lesion, WT and hCAST mice were given either KA or SAL (6–9 animals per group) and sacrificed by transcardial perfusion, after 14 days (**Figure 1G**), when an increase in cell migration is known to occur in the DG after KA treatment (Carreira et al., 2015). Cell migration in the DG was then assessed by analyzing neuroblast staining with doublecortin (DCX), as described further in more detail.

#### EdU Detection and Immunohistochemistry

Brains used for immunohistochemistry and EdU detection were obtained after transcardial perfusion of the mice with 0.9% NaCl and 4% paraformaldehyde. The brains were removed and kept in 4% paraformaldehyde overnight for further fixation and then dehydrated in 20% sucrose in phosphate buffer for at least one day, at 4◦C. Coronal sections from the hippocampal region were cryosectioned (30 µm thick, in 6-series) using a CryoStar NX50 cryostat (Thermo Fisher Scientific, Waltham, MA, USA) and stored in an antifreeze solution (30% ethylene glycol and 30% glycerol in phosphate buffer), at 4◦C. Free-floating brain sections from one of the series were labeled against BrdU, DCX, EdU or EdU/NeuN (neuronal nuclei), as previously described (Morte et al., 2013; Machado et al., 2015). EdU labeling was performed using a commercially available kit (Click-iT<sup>r</sup> EdU Alexa Fluor<sup>r</sup> 488 HCS Assay, Thermo Fisher Scientific, Waltham, MA, USA). After rinsing with 3% bovine serum albumin, the sections were permeabilized with 0.5% triton X-100 for 45 min, rinsed again and then incubated with a reaction cocktail (reaction buffer, CuSO4, Alexa Fluor 488 and reaction buffer additive) for 1 h, protected from light, at room temperature. For immunohistochemistry, sections were rinsed with PBS and then blocked for 1 h, at room temperature, in 5% blocking solution in 0.25% triton X-100, using normal horse serum. After blocking, the sections were kept with the primary antibodies—rat anti-BrdU 1:50 (AbD Serotec, Oxford, UK), goat anti-DCX 1:400 (Santa Cruz Biotechnology, Inc., Dallas, TX, USA), mouse anti-NeuN 1:200 (Merck Millipore, Billerica, MA, USA)—for 48 h, at 4◦C. Sections were then rinsed in 2% blocking solution and incubated for 2 h, at room temperature and protected from light, with the correspondent Alexa Fluor-conjugated secondary antibodies—donkey anti-rat 488, donkey anti-goat 594 and donkey anti-mouse 594; 1:200 (Thermo Fisher Scientific, Waltham, MA, USA). For the labeling of BrdU, a DNA denaturation step with 1 M HCl for 20 min, at 65◦C, was performed in the beginning. In the cases where NeuN was not labeled, nuclei were stained with 2 µg/ml Hoechst 33342 for 10 min, at room temperature. For the combination of EdU and NeuN, EdU labeling was performed, followed by the immunohistochemistry, starting with the blocking step. The entire procedure was done using an orbital shaker. After a final rinsing step, the sections were mounted in gelatin-coated slides, with DAKO fluorescence mounting medium.

#### Analysis of Incorporation of BrdU and EdU

For the analysis of incorportation of the thymidine analogs, BrdU-positive and EdU-positive cells in the SGZ, the first layer of cells adjacent to the hilus, in the GZ or in the hilus of five central coronal sections of the hippocampal region were counted for each animal (Salazar-Colocho et al., 2008; Machado et al., 2015), directly under an epifluorescence microscope (Axio Imager Z2 microscope, Zeiss, Oberkochen, Germany). Cell counting was carried out in both upper and lower blades of the DG.

#### DCX Immunoreactivity

DCX immunoreactivity in the DG was determined in images acquired in a laser scanning microscope (LSM710, Zeiss, Jena, Germany). The quantification of the DCX-positive area was performed in ImageJ (version 1.47v, National Institutes of Health, Bethesda, MD, USA), using a threshold analysis in five central coronal sections of the hippocampal region of each animal. This consisted in defining the optimal staining threshold and calculating the area labeled with DCX (Komitova et al., 2005; Machado et al., 2015).

#### Neuronal Differentiation Analysis

For the quantification of newborn neurons, the percentage of cells labeled for both EdU and NeuN was determined in a total of up to 50 EdU-positive cells (Carreira et al., 2015) in the DG of eight animals per group, in images (orthogonal reconstructions of projections from 0.73 µm z-stacks) acquired in a LSM710 (Zeiss, Jena, Germany).

#### FluoroJade C Staining

For FluoroJade C staining, coronal sections from the hippocampal region of WT and hCAST mice (3–4 animals per group) treated with KA for 24 h were mounted in gelatincoated slides. After air-drying for at least 24 h, the sections were rinsed twice in distilled water, and then immersed for 5 min in 0.1% sodium hydroxide prepared in 80% ethanol, 1 min in 70% ethanol and rinsed twice for 2 min in distilled water. The slides with the sections were then transferred to a 0.06% potassium permanganate solution and left in an orbital shaker for 10 min. After another rinsing step, the slides were left with agitation and protected from light for 10 min in 0.0001% FluoroJade C prepared in 0.1% acetic acid. The sections were rinsed three more times for 1 min in distilled water and dried for 15 min at 60◦C, after which they were immersed in xylene and coverslipped with DPX mounting medium. Images were acquired in an epifluorescence microscope (Axio Imager Z2 microscope, Zeiss, Oberkochen, Germany) and the percentage of staining in the GZ and the hilus measured using a threshold analysis in ImageJ (version 1.47v, National Institutes of Health, Bethesda, MD, USA; Carreira et al., 2010; Machado et al., 2015).

#### Statistical Analysis

Data are expressed as means ± SEM. Statistical significance was determined using the Kruskal-Wallis test, with Dunn's post-test, or the Mann-Whitney test, as indicated in the figure legends, using the GraphPad Prism 5 software (GraphPad Software, Inc., La Jolla, CA, USA). Differences were considered significant when p < 0.05.

#### RESULTS

## Short-Term Impairment in Novel Object Exploration after KA Treatment Is Prevented in Mice Overexpressing CAST

To evaluate whether CAST overexpression would allow for a better and/or faster recovery after brain injury, we performed several cognitive behavioral tests on WT and hCAST mice, on the 3rd (short-term) or 8th (long-term) week after treatment with either KA or SAL. We first tested the animals for OR memory (3–5 animals per group for short-term recovery and 12–15 animals per group for long-term recovery). We observed that, for both time periods, WT mice treated with SAL spent significantly more time exploring the novel object (short-term recovery: 58.3 ± 3.7%, long-term recovery: 62.9 ± 3.3%) than the familiar object, that they had already explored previously (short-term recovery: 41.7 ± 3.7%, p < 0.05, long-term recovery: 37.1 ± 3.3%, p < 0.001). However, after KA treatment, only WT mice from the long-term recovery group were able to distinguish the novel object (58.9 ± 4.5%), from the familiar object (41.1 ± 4.5%, p < 0.05; **Figure 2**, left panels), indicating impairment in recognition memory in the short-term group. On the other hand, all mice overexpressing CAST spent more time exploring the novel object, in both short-term recovery (SAL, novel object: 61.5 ± 5.0%, familiar object: 38.5 ± 5.0%, p > 0.05; KA, novel object: 61.1 ± 1.3%, familiar object: 34.9 ± 1.3%, p < 0.01) and long-term recovery (SAL, novel object: 62.9 ± 3.8%, familiar object: 37.1 ± 3.8%, p < 0.001; KA, novel object: 59.7 ± 3.6%, familiar object: 40.3 ± 3.6%, p < 0.001; **Figure 2**, right panels).

#### Long-Term Impairment in Spatial Memory for an Exact Location during the Water Maze Test after KA Treatment Is Prevented in Mice Overexpressing CAST

In order to study spatial learning and memory, the Morris WM test was performed (4–5 animals per group for short-term recovery and 11–16 animals per group for long-term recovery). During trials (**Figure 3A**), we observed that KA-treated mice

in the short-term recovery period learned the platform location similarly to SAL on all days (Day 1, WT: 49.3 ± 7.3 s, hCAST: 39.4 ± 4.1 s; Day 2, WT: 36.3 ± 10.0 s, hCAST: 15.5 ± 3.6 s; Day 3, WT: 24.2 ± 10.9 s, hCAST: 14.2 ± 4.7 s, p > 0.05), as well as all KA-treated mice in the long-term recovery period (Day 1, WT: 38.4 ± 2.7 s, hCAST: 47.9 ± 2.7 s; Day 2, WT: 26.4 ± 3.9 s, hCAST: 30.9 ± 5.3 s; Day 3, WT: 17.4 ± 4.9 s, hCAST: 20.7 ± 2.1 s, p > 0.05). All mice showed improved spatial learning over the course of the trials.

After learning the platform location, the platform was removed and the spatial memory evaluated by calculating the percentage of time spent in the target quadrant (where the platform previously was) and by counting the number of crossings through the exact previous platform location. Time spent in the target quadrant was maintained with KA treatment in both time periods, by comparison with SAL (short-term recovery, WT: 60.4 ± 3.7%, hCAST: 49.0 ± 5.7%, p > 0.05; long-term recovery, WT: 56.4 ± 4.1%, hCAST: 49.0 ± 4.3%, p > 0.05; **Figure 3B**, left panels), indicating that all animals remembered the relative position of the platform. The number of platform crossings (**Figure 3B**, right panels), in turn, also seemed to be maintained with KA treatment in the short-term recovery period when comparing to SAL (WT: 4.3 ± 1.6, hCAST: 3.8 ± 0.9, p > 0.05). However, in the long-term recovery period, the number of platform crossings was significantly impaired in KA-treated WT mice (2.4 ± 0.4) by comparison with SAL treatment (4.5 ± 0.7, p < 0.05), indicating increased difficulty in remembering the exact platform location, which is not observed in mice overexpressing CAST (SAL: 3.0 ± 0.6, p > 0.05).

#### Associative Fear Memory Is Preserved after KA Treatment

Lastly, associative learning and memory was evaluated, by performing the FC test (3–5 animals per group for short-term recovery and 9–16 animals per group for long-term recovery). By the end of the trials, on the first day, the animals successfully learned to associate the cue tone with a footshock. All animals spent nearly half the time of the last hearing of the tone freezing, both in the short-term recovery (WT SAL: 42.7 ± 5.4%, hCAST SAL: 51.5 ± 5.3%, p > 0.05) and in the long-term recovery (WT SAL: 48.9 ± 6.8%, hCAST SAL: 47.1 ± 4.5%, p > 0.05; **Figure 4A**).

<sup>∗</sup>p < 0.05, significantly different from SAL.

FIGURE 4 | Associative fear memory is preserved after KA treatment. WT and hCAST mice behavior was tested on the 3rd (short-term) or 8th (long-term) week after treatment with either SAL or KA. (A) Associative learning was assessed by measuring the percentage of freezing time during FC trials, with the delivery of a footshock by the end of each cue tone. Acqui. (acquisition). (B) Associative memory to the context and cue tone were assessed by measuring the percentage of freezing time during the FC context test (context A) and cued test (cued stimulus), in a new context (context B). Data are presented as means ± SEM of 3–5 animals per group (for short-term recovery) and 9–16 animals per group (for long-term recovery). Statistical significance was determined using the Kruskal-Wallis test (Dunn's post-test), p > 0.05 between treatments.

FIGURE 5 | KA treatment enhances NSC proliferation and neuroblast migration in the dentate gyrus (DG). (A) Representative image of a section of the DG, showing the division into granular zone (GZ), subgranular zone (SGZ) and hilus. EdU in green and nuclei, stained for Hoechst 33342, in gray. Scale bar: 50 µm. (B) WT and hCAST mice were treated with either SAL or KA and were given BrdU on the day before sacrifice, to assess cell proliferation. Representative images from hippocampal brain sections for each group (left panels), showing BrdU-positive cells in white. Scale bar: 200 µm. BrdU-positive cells were counted in the SGZ of five central coronal sections of the hippocampal region for each animal (right panel). (C) WT and hCAST mice were treated with either SAL or KA and sacrificed after 14 days, to assess cell migration. Representative images from hippocampal brain sections for each group (left panels), showing migrating neuroblasts, labeled for doublecortin (DCX), in white. Scale bar: 100 µm. Percentage of DCX-positive area was determined in the DG of five central coronal sections of the hippocampal region for each animal (right panel). Data are presented as means ± SEM of 9–14 animals per group (for cell proliferation) and 6–9 animals per group (for cell migration). Statistical significance was determined using the Kruskal-Wallis test (Dunn's post-test), <sup>∗</sup>p < 0.05, ∗∗p < 0.01 and ∗∗∗p < 0.001, significantly different from SAL.

When tested for associative memory to context (context A; **Figure 4B**, left panels), all KA-treated mice seemed to recognize the context similarly to SAL, both in the short-term recovery (WT: 48.9 ± 6.8%, hCAST: 64.2 ± 10.8%, p > 0.05) and in the long-term recovery (WT: 51.5 ± 6.0%, hCAST: 69.4 ± 2.6%, p > 0.05). When in the context meant for the cued test (context B; **Figure 4B**, middle panels), all hCAST mice from the short-term recovery group treated with KA spent over half the time freezing (SAL: 37.4 ± 7.2%, KA: 63.4 ± 9.8%, p > 0.05), while the percentage of freezing time was maintained low in WT mice (SAL: 23.1 ± 4.5%, KA: 23.5 ± 5.7%, p > 0.05). In the long-term recovery group, the percentage of freezing time was more similar to SAL in both WT and hCAST mice (WT: 25.4 ± 4.4%, hCAST: 33.8 ± 4.4%, p > 0.05). Moreover, all KA-treated mice seemed to maintain the associative memory to the cued stimulus by comparison with SAL, both in the short-term recovery (WT: 56.7 ± 3.9%, hCAST: 75.5 ± 5.2%, p > 0.05) and in the long-term recovery (WT: 69.1 ± 6.5%, hCAST: 75.1 ± 3.2%, p > 0.05; **Figure 4B**, right panels).

## CAST Overexpression Maintains Enhancement of Early Hippocampal Neurogenesis after KA Treatment

In order to study the effects of CAST overexpression on the early stages of hippocampal neurogenesis after a brain insult, NSC proliferation and neuroblast migration were analyzed in the DG of adult mice (**Figure 5A**), after treatment with either SAL or KA (9–14 animals per group for cell proliferation and 6–9 animals per group for cell migration). Cell proliferation was assessed by incorporation of the thymidine analog BrdU into the DNA of dividing NSCs from the SGZ of the hippocampus. We observed that cell proliferation was greatly enhanced in mice overexpressing CAST after KA treatment (73.1 ± 8.0 cells/section), when compared to SAL-treated mice (19.2 ± 1.5 cells/section, p < 0.001), similarly to what was seen in WT mice (KA: 75.6 ± 9.7 cells/section, SAL: 25.0 ± 2.5 cells/section, p < 0.01; **Figure 5B**).

With regard to neuroblast migration, as assessed by DCX staining in the DG, we also observed a significant increase in KA-treated mice overexpressing CAST (6.1 ± 0.6%), when compared to SAL-treated mice (1.9 ± 0.2%, p < 0.05). Although cell migration in the DG of KA-treated WT mice (8.5 ± 0.3%) was also enhanced in comparison with SAL (2.1 ± 0.2%, p < 0.001), CAST overexpression seemed to attenuate this effect (**Figure 5C**).

To evaluate whether differences in neuronal death after lesion could be observed, WT and hCAST mice (3–4 animals per group) were sacrificed 24 h after KA treatment and FluoroJade C staining was performed. We observed that hCAST mice presented an amount of dying neurons similar to WT mice in the DG, both in the GZ (WT: 0.1 ± 0.1%, hCAST: 0.2 ± 0.0%, p > 0.05) and in the hilus (WT: 0.3 ± 0.2%, hCAST: 0.5 ± 0.1%, p > 0.05; Supplementary Figure 1).

#### CAST Overexpression Affects Cell Differentiation in the DG Similarly to WT after KA Treatment

Besides cell proliferation and migration, we were also interested in investigating differentiation of newborn cells after treatment with KA. For that purpose, mice were treated with the thymidine analog EdU (eight animals per group), and the number of EdU-positive cells remaining in the DG after 8 weeks was counted. We observed that the number of newborn cells still surviving in the DG was generally enhanced with KA-treatment in WT (SGZ + GZ: 9.7 ± 1.8 cells/section, hilus: 1.6 ± 0.3 cells/section) and hCAST (SGZ + GZ: 9.2 ± 1.1 cells/section, hilus: 1.6 ± 0.3 cells/section) mice, when compared to SAL treatment (SGZ + GZ, WT: 1.8 ± 0.1 cells/section, p < 0.01, hCAST: 1.6 ± 0.3 cells/section, p < 0.001; hilus, WT: 0.3 ± 0.1 cells/section, p < 0.01, hCAST: 0.6 ± 0.2 cells/section, p > 0.05; **Figure 6A**). When assessing neuronal differentiation specifically, by determining the percentage of those cells that colocalized with a neuronal marker (NeuN), we observed that treatment with KA seemed to reduce the amount of new neurons in the DG in WT mice (52.5 ± 8.0%) and in mice overexpressing CAST (47.0 ± 4.1%), when compared to SAL-treated animals (WT: 71.6 ± 4.4%, p > 0.05; hCAST: 67.2 ± 4.2%, p < 0.05; **Figure 6B**), suggesting decreased neuronal survival.

### DISCUSSION

The realization that the adult mammalian brain is not an immutable organ and that new neurons are able to form and thrive in the neuronal circuitry has opened new doors for the putative treatment of neurodegenerative disorders and brain lesions. NSCs not only give origin to new neurons in the healthy brain, but are also known to increase the formation of new cells after lesion. NSCs can travel as neuroblasts into regions where neurons were lost as a consequence of brain damage, in what seems like an endogenous attempt to replace those lost neurons. However, this process is not very efficient, as most of the new cells die, fail to integrate the networks, or do it erroneously (Kaneko and Sawamoto, 2009; Ma et al., 2009). While cell transplantation into the sites of injury, for example, also seems promising, modulating endogenous neurogenesis would offer a less invasive approach for the treatment of brain damage. Calpains are proteases involved in neurodegeneration and their inhibition has been shown to be neuroprotective in several diseases (Saez et al., 2006), though their influence in neurogenesis is still not very clear. One of the approaches for the study of calpains is the modulation of their selective endogenous inhibitor, CAST. We had previously seen that the absence of CAST impairs NSC proliferation and neuroblast migration in the adult mouse brain (Machado et al., 2015), namely hippocampal neurogenesis, which has been implicated in learning and memory. Thus, we were interested in investigating whether overexpressing CAST would improve cognitive recovery after hippocampal injury, induced with KA, and its effects on post-injury neurogenesis. Finding a way to modulate calpain

Frontiers in Molecular Neuroscience | www.frontiersin.org March 2017 | Volume 10 | Article 60 |

(NeuN)-positive and EdU-positive/NeuN-negative cells in the DG (top panels), showing cells labeled for EdU in green and cells labeled for NeuN in red. Scale bar: 5 µm. The percentage of cells labeled for both EdU and NeuN in the DG (bottom panel) was determined in a total of up to 50 EdU-positive cells. Data are presented as means ± SEM of eight animals per group. Statistical significance was determined using the Kruskal-Wallis test (Dunn's post-test), <sup>∗</sup>p < 0.05, ∗∗p < 0.01 and ∗∗∗p < 0.001, significantly different from SAL.

activity in order to help improve the outcome of brain injury would highly benefit patients with neurodegenerative diseases or other types of brain insults.

KA induces excitotoxicity and consequent cell death, by activating ionotropic receptors highly abundant in the hippocampus (Wang et al., 2005; Carta et al., 2014). The KA model of brain lesion is thus often used for research involving hippocampal damage, especially in studies of temporal lobe epilepsy (Levesque and Avoli, 2013). It is also a good model for investigating hippocampal injury and its consequences (Araújo et al., 2008; Carreira et al., 2010, 2015). Since the hippocampus is the main structure involved in learning and memory, it is not surprising that treatment with KA in rodents can induce cognitive impairment (Stubley-Weatherly et al., 1996), which has also been reported in patients with temporal lobe epilepsy (Hattiangady and Shetty, 2008). With this in mind, we were especially interested in verifying whether mice overexpressing CAST would show differences in cognitive recovery on the 3rd (short-term) or 8th week (long-term) after KA treatment. In the short-term group, neurons born a few days after the brain insult are still immature, while in the long-term group those neurons are already fully mature and integrated (Deng et al., 2010). We tested the mice for three different types of cognitive behavior: recognition memory (OR), spatial learning and memory (WM) and associative learning and memory (FC).

The OR test makes use of the natural exploratory behavior of rodents. In the presence of a familiar and a novel object, rats and mice will normally spend relatively more time exploring the novel one, in a process shown by several studies to implicate the hippocampus (Cohen and Stackman, 2015). It was thus not a surprise that KA treatment impaired recognition of the previously explored object in WT mice from the short-term recovery group. In the long-term group, however, this impairment seemed to be overcome, suggesting that the newly formed cells ultimately succeeded in restoring the networks necessary for OR. Interestingly, CAST overexpression prevented impairment after KA treatment even in the short-term group. Having in mind that in this group the newly formed neurons are still immature, it is unlikely that they were already able to restore lost connections. This might thus mean that the regions involved in recognition memory were less affected when CAST was overexpressed.

Spatial memory, in turn, has long been known to require the hippocampus, being one of the most affected by hippocampal lesions (Best et al., 2001; Martin and Clark, 2007). In a fundamental work by Richard Morris (Morris et al., 1982), for example, rats with hippocampi removed took longer to learn to reach an escape platform hidden in a pool, but not a visible platform, and also had more difficulty in distinguishing the exact previous escape location when the platform was removed from the pool. This indicates the importance of the hippocampal integrity in navigating through space using only distal environmental cues. In our study, we used this Morris WM test to investigate whether impairment in spatial memory was also seen in our model and if CAST overexpression could be beneficial. By the end of the trials, all animals were able to learn to reach the platform, and also spent a high percentage of time in the target quadrant when platform was removed. Moreover, in the short-term group, there seemed to be no differences in the number of crossings over the exact previous platform location. In the long-term group, however, the number of crossings was impaired in KA-treated WT mice in comparison with SAL, which was not seen in mice overexpressing CAST. Interestingly, certain neurons in the hippocampus, the place cells, have the peculiarity of firing only when the animal enters a specific position in space. Different place cells respond to different locations of the same environment, allowing the brain to create a spatial map of that particular environment (O'Keefe and Dostrovsky, 1971). Additionally, a study on the pattern of place-cell activity during a WM test in rats has shown that the percentage of place cells firing maximally was more than doubled near the platform when compared to other locations in the pool (Hollup et al., 2001). This indicates a great contribution of these cells to the correct mental positioning of the platform in this particular task. Our results may suggest a loss of place cells or correct place-cell activity in WT mice treated with KA, since they seemed to have only a general idea of the platform whereabouts, but failed to determine exactly where it was, as compared to SAL-treated mice. This notion is further corroborated by a study showing that seizure-induced hippocampal lesion can indeed affect the activity of place cells several weeks later, making them less precise and less stable (Liu et al., 2003). The existing place cells in mice overexpressing CAST seem therefore to be less affected by KA treatment, suggesting that calpain inhibition may also limit the seizure-induced damage of these cells.

The FC test was the last to be performed, since it was the most stressful for the animals. This test is based on the association of an aversive stimulus with a tone and with the context where it occurred, which is measured by the time the animal spends freezing, i.e., when the only observed movement is from breathing, a known indicative of fear in rodents (Blanchard and Blanchard, 1969; Wehner and Radcliffe, 2004). As with the other performed cognitive tests, the FC test also has the involvement of the hippocampus, but only in relation to context, being the response to the cued stimulus dependent on the integrity of the amygdala (Phillips and LeDoux, 1992). For that reason, during learning, we were expecting all mice to associate the cued tone with the aversive stimulus, with progressively more time spent freezing with further tone presentations. Indeed, by the last time the tone was played, all animals spent around half of the time freezing, in contrast with virtually no freezing that was observed when they were first placed in the chamber. On the following day, the animals were placed in the exact same context, without presentation of tone or footshock, and SAL-treated animals presented associative memory to this context, by presenting freezing behavior for most of the time spent inside the chamber, as expected. However, KA-treatment failed to impair the time spent freezing, which may be linked with the relative resistance of the used strain to KA-induced cell death (McKhann et al., 2003). Even though differences in associative memory might be too subtle for detection, freezing to the modified context seemed to be increased in KA-treated mice overexpressing CAST from the short-term recovery group, since they spent over half the time freezing. Unlike the other mice, that seemed to understand the context was different, these mice seemed to assimilate that they were in a similar situation as before. One interesting feature of the aforementioned place cells is that they can be involved in the mechanisms of pattern separation, the ability to discriminate between very similar memories, and pattern completion, the ability to recall a previous memory in full based only on some of the components that composed that memory (Sahay et al., 2011). If these mechanisms are not balanced, there is an alteration in the perception of the environment. For instance, if pattern completion is relatively enhanced, there is generalization of the current environment, which means that several contexts may be linked together, even if they present reduced similarities (Sahay et al., 2011). This imbalance may thus be what caused the tendency of KA-treated hCAST mice from the short-term group to present higher freezing in the modified context, since the small similarities between both contexts could have been enough for them to recall the aversive stimulus. A possible explanation for this may relate to the findings that young granule cells (immature neurons) seem to be involved with pattern separation and older granule cells with pattern completion (Nakashiba et al., 2012). Assuming that old neurons can be more resistant, and that this might be the case in KA-treated mice overexpressing CAST, relatively more younger ones would die (Spalding et al., 2013). Since KA enhances the early stages of neurogenesis, an increased number of younger granule cells is expected to exist in the hippocampus of mice from the short-term group. At first, the new cells do not appear to be enough to compensate for the relative higher number of older cells, though later on they seem to regain balance, since the differential behavioral response to the new context of hCAST mice treated with KA from the short-term recovery group is lost in the long-term group. On the other hand, assuming that in WT mice younger and older neurons are lost more or less equally in KA-treated WT mice, and since the balance between younger and older granule cells seems to be maintained, by observation of memory association to the new context, we suggest that the new young neurons may not be fully functional. Electrophysiology studies would help to evaluate if this is in fact the case. Lastly, like with learning, we were expecting, and observed, associative memory to the cue tone to be maintained in all animals, since the hippocampus does not seem to be involved in this process.

Furthermore, we wanted to evaluate whether CAST overexpression affected the modulation of post-injury hippocampal neurogenesis. Like with other types of brain damage, an enhancement of endogenous neurogenesis has also been reported after seizures. Independently of the model chosen to induce seizures in rodents, NSC proliferation is specially shown to be largely increased in the DG after just a few days (Parent et al., 1997; Gray and Sundstrom, 1998; Nakagawa et al., 2000; Hüttmann et al., 2003; Mohapel et al., 2004; Jessberger et al., 2007; Sierra et al., 2015). Interestingly, evidence of increased cell proliferation has also been observed in the hippocampus of patients with temporal lobe epilepsy (Blümcke et al., 2001; Crespel et al., 2005). In the model used by our group, the peak of cell proliferation in the DG was shown to occur on day 5 after seizure induction, with a significant increase in neuroblast migration after 14 days (Carreira et al., 2015), results that we were able to replicate in the present work. This heavy enhancement of early neurogenesis indicates that the microenvironment in the hippocampus not long after seizures is extremely pro-neurogenic, which makes it difficult for changes due to altered levels of calpain activity to be observed. Nonetheless, since increased calpain activity can impair NSC proliferation and migration of neuroblasts, it is important to exclude whether, in this case, calpain inhibition, by overexpression of CAST, increases cell proliferation and migration to alarming levels. An even more extreme rise in cell proliferation could originate unwanted cellular masses of undifferentiated cells, and more migrating cells could also mean more cells migrating into ectopic regions, interfering with the normal neuronal networks (Parent et al., 2006).

Regarding cell proliferation, the enhancement observed after KA treatment was similar in WT mice and in mice overexpressing CAST. Curiously, however, while the increase in neuroblast migration in KA-treated WT mice was around 4-fold, it dropped to 3-fold with the overexpression of CAST. Brain damage is closely associated with an increase in calpain activity, and it is known that calpain inhibition can be neuroprotective (Saez et al., 2006). This means that the fact that CAST was already overexpressed when seizures were induced may have attenuated the damage. Even though the amount of dying neurons in the DG 24 h after KA treatment seems similar to WT in a preliminary study by our group, the surrounding circuitry may still be more preserved with CAST overexpression, or different types of neurons may be affected. Less damage could therefore also mean a less dramatic increase in neurogenesis. In the long term this could be favorable, proven that more cells would integrate correctly in the existing networks. Cell proliferation in the DG was shown by another group to substantially rise even after less severe seizures, with no further increase with seizure severity (Mohapel et al., 2004). This might explain why we observed a similar increase in cell proliferation even with the overexpression of CAST.

This increase in neurogenesis after lesion, however, is only efficient if the new cells ultimately differentiate, survive and integrate the neuronal circuitry correctly. After seizures, this is usually not the case. Despite the enhancement in the early stages of neurogenesis, most of the new cells end up differentiating into astrocytes, with relatively less neurons surviving 1–2 months after seizures (Hattiangady and Shetty, 2010; Carreira et al., 2015; Sierra et al., 2015). Moreover, several newly formed neurons resembling granule cells are found ectopically in the hilus, and have also been reported to occur in patients with temporal lobe epilepsy (Parent et al., 1997, 2006). These ectopic granule-like hilar cells are hyperexcitable, contributing to a disruption of the existing neuronal networks and possibly aggravating the outcome of the disease (Dashtipour et al., 2001; Cameron et al., 2011). Our findings with WT mice are consistent with what has been reported. After a 2-month period, there were significantly more new cells found in the hilus, SGZ and GZ of the DG of mice treated with KA. Though not significant, the percentage of new neurons seemed to be decreased in these mice, which is in accordance with the aforementioned reduced survival of new neurons after seizures. The overall increase in the number of new cells may thus be justifiable by the increased differentiation into astrocytes that occurs after seizures. The results obtained with hCAST mice were very similar, indicating that suppression of calpain activity provided by CAST overexpression is not enough to compensate for the impairment of neuronal differentiation in the DG. Although calpain inhibition was shown to increase neuronal differentiation in vitro, in a cell line from fetal NSCs (Santos et al., 2012), we cannot exclude that in vivo and in the developed brain other mechanisms may compensate for the decrease in calpain activity under a certain threshold, or even help in maintaining levels of calpain activity too elevated, especially in a pathologic environment. In this case, additional factors, such as nitric oxide from inflammatory origin, may also contribute for neuronal impairment (Carreira et al., 2015). This multifactorial feature of brain pathology must thus be taken into account when developing new strategies for brain repair. Nonetheless, even though the percentage of neurons formed after seizures is not ameliorated by CAST overexpression, the ones that thrive may still be more functional and better integrated in the hippocampal neuronal networks.

Overall, the fact that CAST was already overexpressed prior to lesion may have masked potential benefits of calpain inhibition on post-injury neurogenesis. Moreover, the fact that the absence of CAST and, consequently, increased calpain activity, can impair hippocampal neurogenesis, does not necessarily mean that reduced or lack of calpain activity can enhance it. In conclusion, even though reduced calpain activity does not seem to be able to enhance hippocampal neurogenesis, it could potentially be able to prevent the impairment in the formation of new neurons after injury, given that the levels of calpain activity could be reduced under a certain threshold. Strategies to reduce the harmful effects of the pathologic environment, such as controlling

#### REFERENCES


neuroinflammation, could also help boost the outcome of calpain inhibition in the survival of new neurons. Moreover, these observations are in regard to hippocampal neurogenesis. Cells derived from the subventricular zone may hold more promise for the enhancement of endogenous neurogenesis with calpain inhibition, encouraging further research involving these cells and calpain activity in the field of brain damage.

#### AUTHOR CONTRIBUTIONS

VMM: conception and design of the work, acquisition, analysis, interpretation of data, drafting of manuscript; ASL: design and acquisition; CF and RF: acquisition; CMC: conception and design of the work; IMA: conception and design of the work, interpretation of data, drafting of manuscript. All authors revised and approved the final version of the manuscript.

#### FUNDING

This work was supported by the Foundation for Science and Technology (FCT, Portugal), COMPETE and FEDER (grants PTDC/SAU-NMC/112183/2009, UID/NEU/04539/2013 and UID/BIM/04773/2013). VMM and ASL were supported by FCT (fellowships SFRH/BD/78050/2011 and SFRH/BD/79308/2011).

#### ACKNOWLEDGMENTS

We thank Dr. Ralph Nixon and Dr. Mala Rao for providing us the animals used in this study. VMM and ASL are PhD students of the PhD program in Biomedical Sciences of the University of Algarve.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00060/full#supplementary-material

dentate gyrus of pediatric patients with early-onset temporal lobe epilepsy. Hippocampus 11, 311–321. doi: 10.1002/hipo.1045


implications for stroke. Front. Cell. Neurosci. 9:22. doi: 10.3389/fncel.2015. 00022


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer DT and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Machado, Lourenço, Florindo, Fernandes, Carvalho and Araújo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Neuroplastic Changes Following Brain Ischemia and their Contribution to Stroke Recovery: Novel Approaches in Neurorehabilitation

Claudia Alia1,2 \* † , Cristina Spalletti <sup>1</sup> \* † , Stefano Lai <sup>3</sup> , Alessandro Panarese<sup>3</sup> , Giuseppe Lamola<sup>4</sup> , Federica Bertolucci <sup>4</sup> , Fabio Vallone3,5,6 , Angelo Di Garbo<sup>5</sup> , Carmelo Chisari <sup>4</sup> , Silvestro Micera3,7 and Matteo Caleo<sup>1</sup>

<sup>1</sup>CNR Neuroscience Institute, National Research Council (CNR), Pisa, Italy, <sup>2</sup>Laboratory of Biology, Scuola Normale Superiore, Pisa, Italy, <sup>3</sup>Translational Neural Engineering Area, The BioRobotics Institute, Scuola Superiore Sant'Anna, Pontedera, Italy, <sup>4</sup>Department of Neuroscience, Unit of Neurorehabilitation—University Hospital of Pisa, Pisa, Italy, <sup>5</sup>CNR Biophysics Institute, National Research Council (CNR), Pisa, Italy, <sup>6</sup>Neural Computation Laboratory, Center for Neuroscience and Cognitive Systems @UniTn, Italian institute of Technology (IIT), Rovereto, Italy, <sup>7</sup>Ecole Polytechnique Federale de Lausanne (EPFL), Bertarelli Foundation Chair in Translational NeuroEngineering Laboratory, Center for Neuroprosthetics and Institute of Bioengineering, Lausanne, Switzerland

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Mathias Hoehn, Max Planck Institute for Neurological Research, Germany Jennifer Grau-Sánchez, University of Barcelona, Spain Ertugrul Kilic, Istanbul Medipol University, Turkey

\*Correspondence:

Claudia Alia alia@in.cnr.it Cristina Spalletti spalletti@in.cnr.it

†These authors have contributed equally to this work.

Received: 05 December 2016 Accepted: 03 March 2017 Published: 16 March 2017

#### Citation:

Alia C, Spalletti C, Lai S, Panarese A, Lamola G, Bertolucci F, Vallone F, Di Garbo A, Chisari C, Micera S and Caleo M (2017) Neuroplastic Changes Following Brain Ischemia and their Contribution to Stroke Recovery: Novel Approaches in Neurorehabilitation. Front. Cell. Neurosci. 11:76. doi: 10.3389/fncel.2017.00076 Ischemic damage to the brain triggers substantial reorganization of spared areas and pathways, which is associated with limited, spontaneous restoration of function. A better understanding of this plastic remodeling is crucial to develop more effective strategies for stroke rehabilitation. In this review article, we discuss advances in the comprehension of post-stroke network reorganization in patients and animal models. We first focus on rodent studies that have shed light on the mechanisms underlying neuronal remodeling in the perilesional area and contralesional hemisphere after motor cortex infarcts. Analysis of electrophysiological data has demonstrated brain-wide alterations in functional connectivity in both hemispheres, well beyond the infarcted area. We then illustrate the potential use of non-invasive brain stimulation (NIBS) techniques to boost recovery. We finally discuss rehabilitative protocols based on robotic devices as a tool to promote endogenous plasticity and functional restoration.

Keywords: stroke, motor cortex, plasticity, callosal connections, non-invasive brain stimulation, local field potentials, rehabilitation, robotics

#### INTRODUCTION

Following an ischemic insult within the motor cortex, one or more body parts contralateral to the infarct result impaired or paretic. The degree of the motor impairment depends on many factors, such as the extent of the infarct, the identity of the damaged region(s) and the effectiveness of the early medical care. Substantial functional recovery can occur in the first weeks after stroke, mainly due to spontaneous mechanisms (Kwakkel et al., 2004; Cramer, 2008; Darling et al., 2011; Ward, 2011; Grefkes and Fink, 2014). About 26% of stroke survivors are able to carry on everyday activities (Activity of Daily Living or ADLs, i.e., eating, drinking, walking, dressing, bathing, cooking, writing) without any help, but another 26% is forced to shelter in a nursing home (Carmichael, 2005). Impairments of upper and lower limbs are particularly disabling as they impact on the degree of independence in ADLs. Overall, a significant percentage of the patients exhibit persistent disability following ischemic attacks. Therefore, it is critical to increase our knowledge of post-stroke neuroplasticity for implementing novel rehabilitative strategies. In this review we summarize data about plastic reorganizations after injury, both in the ipsilesional and contralesional hemisphere. We also describe non-invasive brain stimulation (NIBS) techniques and robotic devices for stimulating functional recovery in humans and rodent stroke models.

## NEUROPLASTICITY AFTER STROKE

The term brain plasticity defines all the modifications in the organization of neural components occurring in the central nervous system during the entire life span of an individual (Sale et al., 2009). Such changes are thought to be highly involved in mechanisms of aging, adaptation to environment and learning. Moreover, neuronal plastic phenomena are likely to be at the basis of adaptive modifications in response to anatomical or functional deficit or brain damage (Nudo, 2006). Ischemic damage causes a dramatic alteration of the entire complex neural network within the affected area. It has been amply demonstrated, by many studies, that the cerebral cortex exhibits spontaneous phenomena of brain plasticity in response to damage (Gerloff et al., 2006; Nudo, 2007). The destruction of neural networks indeed stimulates a reorganization of the connections and this rewiring is highly sensitive to the experience following the damage (Stroemer et al., 1993; Li and Carmichael, 2006). Such plastic phenomena involve particularly the perilesional tissue in the injured hemisphere, but also the contralateral hemisphere, subcortical and spinal regions.

# Neuroplasticity in Perilesional Area: Map Reorganization

The most convincing evidence of post-stroke spontaneous plasticity in the perilesional area is the observation of topographical map reorganization (Harrison et al., 2013). Motor cortices show in fact a topographical organization, so that sites evoking movements of specific body parts cluster together. Maps are shaped during early life and remain quite stable in adulthood. Interestingly, they can change even in the adult by experiencedependent plasticity (such as after an intensive training) or after brain injury.

Remapping of the motor cortical areas has been observed in stroke patients via either functional Magnetic Resonance Imaging (fMRI) or Transcranial Magnetic Stimulation (TMS; Cicinelli et al., 1997, 2003; Traversa et al., 1997; Liepert et al., 1998; Rossini et al., 2001). In animal models, reorganization of motor maps has been observed using intracortical microstimulation (ICMS; Nudo and Milliken, 1996; Nishibe et al., 2010; Alia et al., 2016) or optogenetic techniques (Harrison et al., 2013).

Studies on primates have demonstrated that following an ischemic injury to the hand area of primary motor cortex (M1) there is a significant reduction of hand representation if no rehabilitative training is applied (Nudo, 2007). However, if the monkey undergoes rehabilitative exercises, the area of the hand is preserved; it is possible that training encourages reacquisition of motor skills in the impaired hand, maintaining the efficacy of corticospinal cells in driving hand motoneurons (Nudo, 2007). Other studies confirmed these results in primates and rats (Nudo, 2013; Nishibe et al., 2015; Combs et al., 2016).

Learning and post-stroke remapping seem to follow different mechanisms, even though they probably share many effectors (Krakauer, 2006; Ramanathan et al., 2006). A proof of these two different mechanisms, has been provided by Ramanathan et al. (2006) who found that complex movements evoked with ICMS do not show plasticity during learning, but they exhibit remapping during post-stroke recovery. Moreover, it has been shown that the cholinergic system plays a crucial role in remapping after stroke. In fact, Conner et al. (2005) showed that immunolesioning the cholinergic system abolished post-stroke recovery and related remapping. The cholinergic system is a component of the ascending neuromodulatory systems. Many studies reported the role of specific neuromodulators such as dopamine, norepinephrine, or serotonin in recovery from stroke also in humans (for a systematic review see Berends et al., 2009). In rodents, it has been shown that activation of modulatory neurotransmitters (via vagus nerve stimulation) in phase with motor exercise (lever pulling) improves post-stroke motor function (Hays et al., 2016).

It is well established that after a small subtotal cortical lesion, peri-infarct areas could actually vicariate lost or damaged functions (Murphy and Corbett, 2009; Dancause and Nudo, 2011). For example, following an ischemic injury in M1, premotor areas can remain functional and contribute to recovery. The ventral premotor area, which receives most of its inputs from M1, produces and releases Vascular Endothelial Growth Factor (VEGF), which has angiogenic and neuroprotective properties, in the early phase after the infarct (Nudo, 2007). In rodents, the Rostral Forelimb Area (RFA) represents a pre-motor cortex involved in the planning and execution of forelimb movements (Rouiller et al., 1993; Saiki et al., 2014; Vallone et al., 2016). The RFA shows a sustained reorganization of the motor map after stroke (Tennant et al., 2015; Touvykine et al., 2016), and preventing RFA reorganization after stroke hinders a long-lasting motor recovery even after rehabilitation (Conner et al., 2005). Consistently, inducing a second lesion in RFA after rehabilitation-induced motor recovery leads to a reappearance of the motor deficit (Okabe et al., 2016).

### Cellular and Molecular Substrates of Post-Stroke Plasticity

However, many issues regarding mechanisms underlying network reorganization and regain of motor function remain still incompletely understood. These mechanisms could involve unmasking of subthreshold pre-existing connections or sprouting of new fibers (Murphy and Corbett, 2009). In this context, the GABAergic system and the extracellular matrix could have an important role in controlling these plastic phenomena. For example, Perineuronal Nets (PNNs), specialized extracellular matrix structures made of condensed chondroitin sulfate proteoglycans (CSPGs), have been correlated with brain plasticity and repair, and preferentially surround the soma of GABAergic neurons, in particular fast-spiking parvalbumin-positive interneurons (Fawcett, 2015). The role of PNNs has been extensively investigated during the maturation of the visual system in relation to the opening and closure of the critical period (Pizzorusso et al., 2002; Deidda et al., 2015). PNNs are thought to stabilize mature connections and downregulate spine motility and functional plasticity. Following CNS injury, the degradation of PNNs, by means of injections of the bacterial enzyme chondroitinase ABC, promotes sensory-motor recovery (Bradbury et al., 2002; Soleman et al., 2012; Gherardini et al., 2015). Moreover, a recent study found a spontaneous decrease in the number of PNNs in the perilesional cortex, suggesting an enhanced plasticity (Alia et al., 2016).

The GABAergic system has also been studied in relation to the opening and closure of early ''critical periods'' in sensory cortices (Hensch, 2005) and in post-stroke motor recovery. Previous works showed that enhancing GABAergic signaling after stroke does not improve post-stroke performance (Madden et al., 2003), but rather induces an acute reappearance of the motor deficit in stroke patients (Lazar et al., 2010). Moreover, a correlation study in humans, showed that a reduced GABAergic inhibition is associated with functional recovery (Kim et al., 2014).

The inhibitory effect in the brain is mainly mediated by GABA signaling through a vast family of GABA<sup>A</sup> receptors (Farrant and Nusser, 2005; Fritschy and Panzanelli, 2014). These ionotropic receptors are composed of different subunits and the resulting molecular assembly determines the localization in different cell districts (i.e., synaptic vs. extra-synaptic) and consequently the biological action of the receptor (phasic vs. tonic signaling; Cherubini, 2012). After a focal stroke, a substantial reorganization of these GABA<sup>A</sup> receptor complexes occurs (Schiene et al., 1996). In a study from Clarkson et al. (2010), tonic GABAergic signaling appears to be increased after stroke. In fact, recordings from brain slices showed an increase in GABAA-receptor mediated tonic inhibition in layer 2/3 pyramidal neurons. Experimental reduction of this heightened inhibition in the first weeks post-stroke using a benzodiazepine inverse agonist, produces significant improvements of forelimb function in several behavioral tasks. Consistently, transgenic mice lacking α5- or δ-GABA<sup>A</sup> receptors (mediating tonic GABA current) showed a lower functional deficit after stroke (Clarkson et al., 2010; Lake et al., 2015). In a recent study, phasic GABA was enhanced in the first week after stroke, specifically in the layer 5 of the perilesional cortex and a chronic and treatment with a positive modulator of α1 containing GABA<sup>A</sup> receptors ameliorated motor outcome during the period of treatment (Hiu et al., 2016). However, these latter findings are difficult to reconcile with the clinical observation that administration of midazolam reinstates stroke deficits in hemiparetic subjects (Lazar et al., 2010). In our recent article we found a downregulation of GABAergic inhibitory presynaptic terminals in the peri-lesional area after photothrombotic stroke in mice (Alia et al., 2016). Interestingly, reducing GABA signaling in the first week post-stroke, using DMCM, an inverse agonist of GABA<sup>A</sup> receptors with an high preference for α1-enriched receptors (Lüddens and Wisden, 1991; Fritschy et al., 1998), strongly improved general motor outcome, and the effects persisted well after the end of the treatment (**Figure 1**; Alia et al., 2016). Overall, these findings demonstrate that the GABA system offers different opportunities for therapeutic intervention and further studies are needed to better delineate the proper timing and target (phasic vs. tonic) of therapeutic treatments.

The role of excitatory neurotransmission has also been studied in relation with post-stroke recovery. Pharmacological activation of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors improves motor outcome by inducing release of the neurotrophin Brain Derived Neurotrophic factor (BDNF) and phosphorylation of TrkB receptors (Clarkson et al., 2011). Moreover, increased BDNF levels and TrkB activation have been also detected after blocking NMDA receptors using memantine, a selective antagonist. As in the previous study, activation of the BDNF signaling pathway was associated with an improved motor performance and increased area of forepaw sensory maps (López-Valdés et al., 2014). Thus, the activity-dependent release of BDNF appears to be essential for the motor recovery (Berretta et al., 2014). In fact, BDNF modulation has also been suggested to mediate the therapeutic effect of treatments increasing perilesional cortex excitability, such as transcranial direct current stimulation (tDCS; Clarkson and Carmichael, 2009; Fritsch et al., 2010).

Stroke induces the production of various inhibitors of neural regeneration, sprouting and plasticity, such as myelin components (Nogo-A, myelin-associated glycoprotein), and guidance molecules (ephrins, semaphorins). The application of drugs able to neutralize the effect of anti-plastic agents, such as Nogo-A antibodies has been seen to encourage axon regeneration, sprouting and functional recovery in a variety of animal models of cortical and spinal injuries (Freund et al., 2006; García-Alías et al., 2009; Maier et al., 2009; Alilain et al., 2011). Particularly, anti-Nogo-A antibodies treatment delivered before motor training strongly improve motor recovery (Wahl and Schwab, 2014). In this study, timing of the treatments was found to be critical, indeed delivering anti-Nogo-A during motor training is not effective to improve motor deficits (Wahl and Schwab, 2014).

After stroke, it has been reported a consistent change in terms of ''sprouting markers'', with an increase of classical axonal growth markers such as, growth-associated protein of 43 KDa (GAP43), CAP23, c-Jun, in the peri-infarct region but also a parallel increase of growth inhibitory genes such as ephrin-A5, CSPGs and others, at later time points (Carmichael et al., 2001, 2005; Caleo, 2015).

#### Post-Stroke Changes in Contralateral Hemisphere and Interhemispheric Coupling

A critical point in literature about stroke evolution and functional recovery is the role of the uninjured hemisphere. It is well established both from animal and patient studies that contralesional neuronal connections appear to be altered as a

result of a unilateral cortical damage (Jones, 1999; Witte et al., 2000; Papadopoulos et al., 2006; Jones and Jefferson, 2011; Dancause et al., 2015). Human functional imaging studies on post-stroke patients, using PET and functional MRI (fMRI), have identified a role of the healthy hemisphere in recovery. An enhanced activity in the contralesional hemisphere has been reported in patients in the first 10 days post injury, followed by an increase in the ipsilesional one (3–6 months). This sequential activation was related to improvements in motor performance (Marshall et al., 2000; Ward et al., 2003). In preclinical studies on rodent models, the activity of the contralesional hemisphere was found enhanced in the very acute stage after stroke, when the deficit was more pronounced, and was followed by perilesional activation at later stages during the recovery phase (Dijkhuizen et al., 2001). Focal stroke targeted in the somatosensory cortex (SSC) induced a transient but consistent increase of basal metabolism and field potentials in the healthy hemisphere, in terms of baseline activity and response-related activity after somatosensory stimulation of the unaffected forelimb (Takatsuru et al., 2009). Other preclinical studies in SSC showed lesioninduced changes in the contralesional cortex sensory map, with an increase in dendritic branches of layer V pyramidal cells. These changes appear to be increased if the animal is subjected to early exercises that enhance motor skills (Biernaskie et al., 2004; Gonzalez et al., 2004). Moreover, two-photon imaging in vivo studies have highlighted phenomena of structural rearrangements of neurons in the healthy hemisphere, both at the level of individual cells and of whole circuits. In particular, data have shown a transient, localized increase in the turn-over of dendritic fungiform (mushroom) spines, which are usually known to be highly stable in a healthy brain, in a time period limited to 1 week post-stroke. Overall, the number of dendritic spines remained unchanged probably because of a balanced number of new-born and degenerated contacts (Takatsuru et al., 2009). This could be a primary difference in mechanisms of post-stroke modifications between the intact and the injured hemisphere. Indeed in the ipsilesional cortex a net gain of dendritic spines has been observed (Brown et al., 2007). The loss of stability of dendritic spines in the healthy hemisphere can be explained by increased baseline and sensory input from the periphery (Takatsuru et al., 2009). Finally, experimental silencing of the healthy hemisphere with the GABA<sup>A</sup> agonist muscimol within hours after stroke can improve functional recovery and the duration of the inactivation is directly correlated with improvement (Mansoori et al., 2014).

These results indicate an involvement of the healthy hemisphere in functional alterations following a unilateral ischemic injury. However, whether the healthy hemisphere has a positive or negative impact on recovery is still controversial (Murase et al., 2004; Hummel et al., 2008; Di Pino et al., 2014). In fact, there are many evidences that in some cases the activity of the healthy hemisphere can worsen motor recovery. For example, a recent quantitative electroencephalography (EEG) study in stroke survivors showed that the increase of contralesional hemisphere activity, during the acute phase, is related to negative final outcome. In fact, the increase of the contralesional power associates to an interhemispheric communication breakdown (Assenza et al., 2013). A possible hypothesis is that the lesion volume and the amount of spared tissue in the injured hemisphere could influence the role of the healthy cortex. In particular, when the lesion is sufficiently small to allow the reorganization of spared adjacent motor areas, the contralesional hemisphere activity would have a negative impact on the recovery. Conversely, when the lesion extent is so large to involve most of motor areas, the healthy hemisphere could be important to vicariate lost functions (Di Pino et al., 2014). In line with this theory, acute pharmacological inactivation of the healthy hemisphere (via lidocaine injection), induced different effects in ischemic rats, depending on the lesion size. Animals with large lesions were dramatically affected by lidocaine administration and showed a strongly impaired performance in a reaching task (Biernaskie et al., 2005).

Changes in the interaction between the two brain hemispheres after stroke are also a widely investigated topic. The neural activity in the brain motor areas is functionally coupled between the two hemispheres (Kinsbourne and McMurray, 1974; Vallone et al., 2016) and the lateralization of neural activity during movements is likely to be related to interhemispheric inhibition between motor areas exerted via transcallosal connections (Bütefisch et al., 2008). After a cortical injury, the subjects recovering from stroke showed changes in these interhemispheric influences (Kobayashi and Pascual-Leone, 2003; Mohajerani et al., 2011) which are thought to be caused by an imbalance in the mutual interhemispheric inhibition between the two motor cortices (Murase et al., 2004; Vallone et al., 2016) that could be an obstacle for motor recovery.

It has been proposed that after unilateral stroke, transcallosal connections could transmit an excessive inter-hemispheric inhibition onto the unaffected hemisphere. Despite the scarce knowledge about the mechanisms mediating these phenomena, the involvement of transcallosal glutamatergic connections acting on pyramidal tract neurons via local GABAergic interneurons is widely accepted (Reis et al., 2008). The role of the corpus callosum in inhibitory interhemispheric mechanisms have been strongly demonstrated through transcranial stimulation studies on patients with agenesis of the callosum (Meyer et al., 1995). FMRI studies have shown an increased bihemispheric activation during movements of the affected limb in early post-stroke patients (Loubinoux et al., 2003) and suggest persistent alterations in intracortical and transcallosal connections, despite a good degree of functional recovery of patients (Nair et al., 2007). This is probably due to a decrease of the ipsilesional neuronal activity and an increase of the contralesional one (Murase et al., 2004; Fregni and Pascual-Leone, 2007), so that the imbalanced activation of the healthy hemisphere causes an increased inhibitory transcallosal signal to the affected side. In such a scenario, low-frequency inhibitory repetitive TMS (rTMS) could be applied over the intact side as a therapeutic strategy to increase rehabilitation-induced motor performances (Nowak et al., 2009; Kim et al., 2010; Silasi and Murphy, 2014; Caleo, 2015). This is consistent with the theory that a balanced, mutual inhibition exists between the two hemispheres and that a unilateral lesion can destroy this equilibrium with the healthy hemisphere taking control and interfering with the activity of the perilesional, spared tissue (Murase et al., 2004; Vallone et al., 2016). Evidences from animal models suggest that experimental inactivation of contralesional hemisphere could actually increase motor recovery, especially when the treatment is prolonged (Barry et al., 2014; Mansoori et al., 2014; Dancause et al., 2015). In humans, application of inhibitory rTMS to the non-lesioned hemisphere improved paretic hand reach-to-grasp performance, movement time and coordination (Tretriluxana et al., 2013). However, other studies have failed to support this idea and the role of the healthy hemisphere in post-stroke recovery remains controversial (see below). Thus, the significance of contralesional activation during execution of a motor task with the affected limb is still uncertain: it could represent an epiphenomenon of recovery, an adaptive neuroplastic process or even a sign of maladaptive modifications that might interfere with the recovery process.

# STUDIES OF FUNCTIONAL CONNECTIVITY: INTER-HEMISPHERIC BREAKDOWN IN STROKE

## Computational Analyses of Post-Stroke Functional Connectivity

Recently, thanks to the advancement of new technology and theoretical/computational analyses, researches on stroke have focused on the effects produced on distant brain areas by a local ischemic injury (see for detailed reviews, Grefkes and Fink, 2011, 2014; Carter et al., 2012; Silasi and Murphy, 2014). This type of approach has been called connectivity-based and is related to the concept of connectome, which is defined by the connections between neurons (Sporns et al., 2005). Three major different spatial scales can be considered in studies of the connectome: microscopic (synapses), mesoscopic (regional interactions, for instance connections from homotopic brain areas) and macroscopic (e.g., thalamus-cortex interactions). The human brain is an extremely complex system containing a huge number of neurons (on the order of 1011) highly and specifically interconnected (a neuron typically receives 10<sup>4</sup> inputs from other cells) and nowadays a precise connectome is missing (see Silasi and Murphy, 2014 and references therein). The development of new theoretical and computational tools for the analysis and modeling of neural signals can be an important step forward to dissect the structure of neural circuits. As an example, in the framework of the statistical data assimilation problem (Abarbanel, 2013), we should ask ourselves which kind of experimental data we need to infer the exact geometry of a large neural network, composed by thousands of interacting neurons. Usually, due to the sparseness of data, this missing information is replaced by several assumptions that can be crucial in building such large scale network models.

Mainly due to these limitations, in many studies on human stroke, the definition of connected brain areas is based on the estimation of the interdependence level (and directional prevalence of the coupling) between the neural signals recorded in the corresponding regions. To get this, specific measures quantification, borrowed from methods of linear and nonlinear time series analysis, can be employed. Well known examples of these measures are: the cross-correlation, the mutual information, the Granger causality, the transfer entropies (Abarbanel, 1996; Kantz and Schreiber, 2004; Pereda et al., 2005) and Dynamic Causal Modeling (DCM; Friston et al., 2013).

These signal processing techniques aim at finding common dynamical indices to be used as indicators of functional coupling between different areas (e.g., intact and injured motor cortex) in humans and animal models of stroke. Longitudinal variations in the values of such indicators can be employed to quantify the level of synaptic reorganization and plasticity during the rehabilitation process, medical therapy and for identifying suitable prognostic indicators.

From a theoretical perspective, the purpose of these studies is to understand how the dynamics of coupled neural networks (brain areas) is modulated by the local modifications of inhibitory and excitatory neurotransmissions in one of them (here represented by the ischemic area). This system level view could provide additional insight into basic fundamental questions about the mechanism of functional segregation and integration in the nervous system (Tononi et al., 1994) and neuroplastic phenomena (see Silasi and Murphy, 2014 and references therein).

In human studies of stroke, the typical techniques suitable to record the neural activity of the brain are fMRI (Logothetis, 2008; James et al., 2009), EEG and Magnetoencephalography (MEG; see for detailed review Grefkes and Fink, 2011, 2014; Carter et al., 2012).

All of the above techniques have the advantage to be non-invasive and therefore suitable for studies on human subjects. However they are difficult to relay to the underlying neural activity (Buzsáki et al., 2012, see for fMRI signals Logothetis, 2008). For example, the direct relationship between neuronal activity and EEG recordings it is not well defined because the electrical activity, generated by thousands of neurons, must diffuse through various media (cerebrospinal fluid, dura mater, cranium, muscle and skin) that distort and attenuate the signals.

To avoid distortion effects, intra-cortical recordings called Local Field Potentials (LFPs) can be employed in animal models. LFP signals represent the low frequency part (<500 Hz) of the extracellular potential generated by the flow of trasmembrane currents in neuronal populations located near the recording electrode (Buzsáki et al., 2012). In contrast to the high-frequency part (which reflects spiking activities), the LFP carries information on the collective synaptic activity of thousands of interconnected neurons. Therefore, the interpretation of the LFP is challenging, and even knowing that the main contribution to the LFP recording is due to the synaptic currents (since they are slow events that can be more easily synchronized than fast events), several factors such as fast (Na+) action potentials, calcium spikes, spikes after hyperpolarization, gap junctions, neuron-glia interactions and ephaptic effects can influence to the LFP signal. Nevertheless, it is believed that LFP recordings are the most informative signals of the underlying neural activity generated by an ensemble of coupled neurons (Buzsáki et al., 2012).

## Changes in Interhemispheric Interactions After Stroke

Recent studies (see below) demonstrated that connectivity measures can improve our ability to correlate behavioral deficits to clinical indices of dysfunction (both structural and functional). In this context, many studies have investigated the interactions between brain hemispheres before and after stroke.

In clinical studies, using MEG signals recorded from stroke patients in resting state condition, changes in alpha-band functional connectivity, both in the peri-lesional and contralesional cortex, have been related to improvements of functional outcomes of upper extremities (Westlake et al., 2012). Moreover, previous fMRI data have demonstrated a correlation between post-stroke loss of sensorimotor function and deterioration of inter-hemispheric functional connectivity in animals (van Meer et al., 2010). In humans, inter-hemispheric interactions have been investigated mainly by resting state fMRI signals (Grefkes and Fink, 2011, 2014; Carter et al., 2012) reporting loss of coherence between inter-hemispheric communication that predicts post-stroke behavioral deficits. Accordingly, the results obtained by analyzing EEG signals from stroke patients support the idea that a reduction of the interhemispheric coupling is correlated to functional deficit and a promising target for rehabilitation is the restoring of the inter-hemispheric communication (Wu et al., 2011; Assenza et al., 2013; Finnigan and van Putten, 2013 and references therein).

In a recent report (Vallone et al., 2016), our group employed a photothrombotic mouse model of stroke in caudal forelimb area (CFA) to study the plastic reorganization in the spared circuits of the adjacent premotor area (RFA). Our aim was to understand the potential role of RFAs in recovery of forelimb function after an ischemic episode in CFA (Rouiller et al., 1993; Guggenmos et al., 2013).

To investigate the electrophysiological changes within and among Pre-Motor Areas, LFPs were recorded from both RFAs in freely moving mice after a cortical lesion in CFA (i.e., 9–16–23 days after surgery, see **Figure 2A**). These time intervals were chosen since they roughly correspond to highest period of circuit plasticity in humans (i.e., 3 months after stroke, see Zeiler and Krakauer, 2013). Quantitative methods of time series analysis were used to assess longitudinal changes in electrical neural activity (Vallone et al., 2016).

Cross correlation and mutual information analyses were employed as linear and nonlinear measures of functional coupling between the two hemispheres, respectively (**Figures 2B,C**). We found a dampening of the functional connectivity between RFAs in ischemic animals (with respect to control group) at day 16 and 23 after surgery (Vallone et al., 2016). These data are consistent with resting-state fMRI in humans (Grefkes and Fink, 2011, 2014; Carter et al., 2012) and rats (van Meer et al., 2010). These studies reported a significant loss of functional connectivity between the two hemispheres following stroke that slowly recovered in parallel with spontaneous behavioral improvements.

Furthermore, in this study (Vallone et al., 2016), a significant stroke-dependent reduction in inter-hemispheric cross-correlation and mutual information for gamma band (30–50 Hz) at day 9 after surgery was observed. The values of the mutual information for delta band (0.5–4 Hz) also indicated a reduction in stroke mice at day 9. Thus, changes in interhemispheric coupling for the gamma and delta range preceded the variations observed using the whole LFP signal. Interestingly, our results confirm a key role of gamma oscillations in cross-

Bregma position. Cross correlation (B) and Mutual Information (C) measures between the two hemispheres in control (black) and ischemic (red) animals are shown. Data are means ± standard errors. Modified from Vallone et al. (2016). <sup>∗</sup>P < 0.05, ∗∗P < 0.01.

talking between brain regions (Buzsáki and Schomburg, 2015), especially in combination with slower frequencies (Lisman and Jensen, 2013; Buzsáki and Schomburg, 2015) and support the idea that stroke impairs brain functions because it disrupts communication in large brain networks (see Carter et al., 2012).

Using DCM on fMRI signals (see Friston et al., 2013 and references therein), stroke patients have shown a reduced excitatory influence from the contra-lesional to the ipsi-lesional PMAs and also inhibitory effects are significantly reduced from the ipsi-lesional to contra-lesional M1 cortices (see Grefkes and Fink, 2014 and references therein). Notably, in stroke patients the inter-hemispheric coupling parameters increased with recovery predicting better functional outcome after 3–6 months (see Grefkes and Fink, 2014 and references therein).

#### NON-INVASIVE MODULATION OF POST-STROKE PLASTICITY IN HUMANS

Thanks to neuroplasticity, the CNS compensates for the functional impairment after stroke. In particular, adaptive plasticity allows the acquisition of new skills, learning, memory, adaptation to new environments throughout the life span (Rossi et al., 1998; Hosp and Luft, 2011). Moreover in the last decade it has been described a different type of neural plasticity due to the injury and to excessive training (Quartarone et al., 2006), named ''maladaptive plasticity''. Clinically, this phenomenon is relevant in several functions such as the vicariation of the upper limb movements with compensatory or substitutive movements and the delayed onset involuntary abnormal movements (Takeuchi and Izumi, 2012). Furthermore, several studies have reported that maladaptive plasticity weakens motor function and limits motor recovery after stroke (Murase et al., 2004; Rijntjes, 2006; Takeuchi et al., 2007). Moreover, it is thought to contribute to the pathogenesis of phantom pain and dystonia (Quartarone et al., 2006; Flor, 2008).

The idea to stimulate plasticity in the injured CNS to improve motor recovery is referred as ''top-down'' approach, and comprises the use of promising tools like plasticizing drugs or NIBS techniques (Chisari, 2015).

NIBS paradigms are applied in human stroke subjects in various ways (Dayan and Cohen, 2011; Dayan et al., 2013; Sandrini and Cohen, 2013; Wessel et al., 2015), both in experimental evaluative protocols and for therapeutic applications (Ziemann et al., 1996; Huang et al., 2005; Hummel and Cohen, 2006; Reis et al., 2008) as a possible technical adjuvant to customarily used neurorehabilitative treatments to enhance motor recovery (Liew et al., 2014; Chisari et al., 2015).

TMS is a NIBS technique that allows to study and to modulate the cortical excitability. The biophysical mechanisms induced by magnetic stimulation are still not completely understood. Given that the axons are the most effective conductors in the CNS, for their higher density of ion channels, the prevailing hypothesis is that they are preferentially affected by the TMS pulse, which may activate both inhibitory and excitatory neurons (Huerta and Volpe, 2009).

The use of particular protocols of rTMS enabled to produce a prolonged modification in cortical excitability with long-term potentiation (LTP)—and long-term depression (LTD)—like changes. In the motor system, low-frequency (1 Hz) rTMS inhibits cortical excitability, creating a transient ''virtual lesion'' (Chen et al., 1997). Instead, high-frequency (5–20 Hz) rTMS produces an increase in cortical excitability (Pascual-Leone et al., 1994), which can facilitate learning of motor sequences (Kim et al., 2004), though the effects may vary (Agostino et al., 2007).

A way to induce longer-lasting effects than conventional rTMS paradigms (Dayan et al., 2013) is theta-burst stimulation (TBS), which involves the application of a burst of three 50-Hz pulses in trains repeated at 200-ms intervals. Continuous TBS (cTBS) consists of the application of burst trains for 20–40 s and has an inhibitory effect on corticospinal excitability. Instead, for intermittent TBS (iTBS), burst trains with a duration of 2 s are applied over a total of 190 s, with the trains repeating every 10 s (Huang et al., 2005). iTBS can induce LTP-like changes in the stimulated hemisphere and LTD-like changes in the opposite hemisphere.

Another stimulation protocol widely used for demonstrating LTP-like and LTD-like phenomena is paired associative stimulation (PAS). PAS takes advantage of the principles of associative plasticity by repeatedly coupling a low-frequency peripheral stimulation from the median nerve with a cortical TMS pulse applied over contralateral motor cortex, with an inter-stimulus interval (ISI) of 10–25 ms (Stefan et al., 2000). An ISI of 10 ms induces a depression of TMS-evoked MEPs, while enhancement of cortical excitability is consequent to the use of 25 ms of ISI, with effects of at least 1 h of duration and resembling LTP-like and LTD-like mechanisms. Protocols using PAS are particularly relevant because they demonstrate some characteristics of spike timing-dependent plasticity (Wolters et al., 2003): the order and precise temporal interval between presynaptic and postsynaptic spikes determine the sign and magnitude of LTP-like or LTD-like synaptic changes.

Transcranial electric stimulation (tES) is a method that has attracted significant attention because its application is thought to induce neuromodulation, as shown by improvements in behavioral and cognitive performance in normal and pathological subjects (Miniussi and Vallar, 2011). Different types of tES are differentiated by specific modalities of current discharge (e.g., direct vs. alternating) that might have different neuromodulatory effects on cortical networks.

Among tES techniques, tDCS offers the possibility to change cortical excitability in a polarity-specific manner (anodal vs. cathodal; Nitsche and Paulus, 2000) through the application of electrodes with different polarity to different locations on the surface of the skull to excite the underlying neural tissue (Utz et al., 2010). tDCS effects are most likely induced by membrane polarization, altering the firing rates of neurons (Fritsch et al., 2010). Anodal tDCS induces depolarization, while cathodal tDCS induces hyperpolarization, so that anodal stimulation produces excitation and cathodal stimulation produces inhibition (Liebetanz et al., 2002).

In part, the use of NIBS techniques is based on the interhemispheric competition model, based on the concept that motor deficits in stroke patients relate to reduced output from the affected hemisphere and excessive interhemispheric inhibition from the unaffected hemisphere to the affected hemisphere (Kinsbourne, 1977, 1980; Murase et al., 2004; Takeuchi et al., 2005). Therefore, improvement in motor deficits can be achieved by increasing the excitability of the affected hemisphere or decreasing the excitability of the unaffected hemisphere (Ward and Cohen, 2004; Nowak et al., 2008). This model has been recently brought into question by Di Lazzaro et al. (2013): they used inhibitory TBS of affected hemisphere in chronic stroke patients to verify if this intervention had the potential to enhance recovery, possibly via a homeostatic increase in learning capacity. Results showed clinical improvements for up to 3 months post-treatment, suggesting the possibility to design protocols of inhibition of affected hemisphere for chronic stroke patients. It is conceivable that either upregulation or downregulation of activity in the affected hemisphere may promote recovery depending on different factors like magnitude of baseline motor function (Fridman et al., 2004) as highlighted in a protocol using iTBS followed by rehabilitative motor training (Volz et al., 2016).

Excitability enhancement in the motor cortex appears to be required for motor learning (Pascual-Leone et al., 1998; Muellbacher et al., 2002; Reis et al., 2009; Censor et al., 2010; Schambra et al., 2011). Therefore, NIBS can facilitate motor learning and induce motor recovery by directly or indirectly increasing the excitability in the ipsilesional motor cortex. In fact, compared to motor training or rTMS alone, pairing motor training with rTMS results in prolonged performance improvements and functional neural plasticity in the ipsilesional motor cortex (Nowak et al., 2008; Takeuchi et al., 2009).

NIBS-induced metabolic changes may also promote neural plasticity and motor recovery after stroke (Conchou et al., 2009). Furthermore, excitatory NIBS over the affected hemisphere can induce LTP-like changes in the affected hemisphere and promote motor recovery after stroke (Di Lazzaro et al., 2010). Therefore, NIBS may resolve impairment of experiencedependent plasticity in the affected hemisphere after stroke (Carmichael, 2006; Di Filippo et al., 2008; Takeuchi and Izumi, 2012). In addition fMRI and EEG studies proved that NIBS is able to modulate neural networks also in brain regions far from the stimulated area (Grefkes et al., 2010; Takeuchi et al., 2010). Excitatory rTMS over the affected hemisphere has been shown to reduce neural activity in the contralesional motor cortex, in addition to facilitation of the ipsilesional motor cortex (Ameli et al., 2009). Moreover, inhibitory rTMS over the unaffected hemisphere reduced the connectivity of both hemispheres and enhanced coupling between the primary and non-primary motor cortices in the affected hemisphere (Grefkes et al., 2010; Takeuchi et al., 2010). Enhanced excitability in the unaffected hemisphere inhibits the affected hemisphere via excessive interhemispheric inhibition and weakens motor function of the paretic side (Murase et al., 2004). Although the change in neural coupling after excitatory NIBS remains still unclear, normalized excitability of both hemispheres and reconstruction of effective connectivity between the primary and non-primary motor cortices in the affected hemisphere after NIBS may contribute to motor recovery in stroke patients (Takeuchi and Izumi, 2012).

The pattern of neural network activation in both hemispheres has important influences on the effect of NIBS therapy for stroke patients (Nowak et al., 2008; Ameli et al., 2009). Therefore, it seems to be important to develop predictors of NIBS response. In fact the neural impact of a NIBS therapy is not determined only by the properties of the stimulus but also on the activation state of the brain. This ''statedependency'' is a general feature of cortical neural processing and it plays an important role on the efficacy of TMS protocols (Silvanto and Pascual-Leone, 2008). To address these issues, more recently EEG combined with TMS has open the new chapter of ''closed-loop NIBS'' (Raco et al., 2016), which with millisecond precision enables selective interference with ongoing brain activity with high temporal, spatial and spectral precision. This approach has the important advantage to take into account not only inter- individual differences in the excitability and connectivity of brain networks but also the time-course of dynamic changes of network reorganization during stroke rehabilitation. Zrenner et al. (2016) argued that two different closed-loop interactions can be differentiated: a ''brain-state dynamics'' loop, used to couple with and modulate the trajectory of neuronal activity patterns, and a ''task dynamics'' loop, that is the bidirectional motor-sensory interaction between brain and (simulated) environment, and which enables goal-directed behavioral tasks to be incorporated. Both loops need to be considered and combined to realize the full experimental and therapeutic potential of closed-loop neuroscience and to interactively optimize neuromodulatory efficacy.

#### ROBOT-ASSISTED REHABILITATION FOLLOWING STROKE

#### Features and Advantages of Robotic Devices

The use of robotic devices aimed at improving the recovery of upper limb motor function in post-stroke therapy was first introduced in the 1990s with the MIT-Manus system, a mechanized device to assist planar reaching movements (Aisen et al., 1997), and with the ARM Guide, a robotic device to assist reaching movements in a range of directions in space, but restricted to follow a linear trajectory (Reinkensmeyer et al., 2000). Lum and colleagues (Burgar et al., 2000; Lum et al., 2002) have then demonstrated the efficacy of robot-assisted whole arm exercises in 3D with the Mirror Image Movement Enabler (MIME). Since then, several other studies have been carried out with the common goal to design and control robotic devices able to monitor and administer exercises to the patient, by eliciting motor brain plasticity and therefore improving motor recovery (Amirabdollahian et al., 2003; Reinkensmeyer et al., 2004; Micera et al., 2005; Schmidt et al., 2005).

Devices for upper limb rehabilitation can be broadly classified into three types, based on the different type of motion assistance they can provide: active devices that provide an active motion assistance and need at least one actuator, able to produce movement of the upper-extremity along a defined trajectory; passive devices that offer non-powered support of the limb during movement attempts (elastic bands or springs); and interactive devices that combine actuators and control strategies allowing for the correction of ''wrong'' motor exercise but also for the modification of the control parameters based on ongoing participant performance during the training (Marchal-Crespo and Reinkensmeyer, 2009).

A further categorization of robotic devices for upper limb motor rehabilitation can be based on their mechanical structure: end-effector-based and exoskeleton-based systems (Maciejasz et al., 2014). In end-effector-based devices, only the most distal part of the robot (i.e., end effector) is attached to patient's upper limb extremity (hand or wrist). Movements of the end effector change the position of the upper-limb extremity, but also indirectly affect the position of the other segments of the patient's upper-limb. Exoskeleton-based systems have a more complex mechanical structure that mimics the structure of patient's limb. They allow for independent and concurrent control of many robot joints, which directly affect the position of correspondent joints of patient's arm. The mechanical and control algorithm complexity of such devices is usually higher than the endeffector-based devices, because of the need to adjust lengths of particular device's segments to the lengths of the patient arm's segments or to manage the high number of movements (high number of degrees of freedom, DOF) allowed. The number of DOF, i.e., all independent movements (i.e., displacements or rotations) that can be performed in all the joints of a robotic device, generally depends on the target of the rehabilitation process: for example, the devices for the rehabilitation of the whole upper limb can have up to 10 DOF (Ren et al., 2009) whereas hand exoskeleton can reach even 11 DOF (Hasegawa et al., 2008). However many devices with a limited number of DOF have been developed. Planar robots have only 3 DOF, 2 translational and 1 rotational, allowing movements only on a specified plane (Aisen et al., 1997; Colombo et al., 2005). Planar devices reduce the range of upper limb movements that can be trained, but also cuts down the cost of the system, leading to an improved cost-effectiveness of robot assisted therapy (Wagner et al., 2011). However, when the working plane is appropriately selected, this range of training motion may be sufficient in most of therapeutic scenarios (Maciejasz et al., 2014).

Robotic systems have many properties, as high repeatability, possibility to perform a great amount of exercises in a single session and high intensity of task-oriented training (Posteraro et al., 2009). Many of these devices can be also adapted for the needs of the patients, allowing for a customization of the therapy. For example, exoskeleton-based systems can be tailored based on the length of the patient arm's segments whereas end-effector-based devices can provide different types of motor exercises in agreement with the spared motor patient's ability (Maciejasz et al., 2014). Furthermore, recent studies have also attempted to improve motivation in stroke rehabilitation using these robotic devices coupled with elements of virtual reality (e.g., audiovisual elements, score displays and cognitive challenges; Novak et al., 2014). Indeed, motivation could promote exercising for longer periods, increasing the total amount of training?

These devices incorporate several sensory components (e.g., encoders, accelerometers, load cells, etc.) allowing for a complete feedback and monitoring of the therapy progress over the time, even adjusting the degree of robot assistance based on the progress of the patient. Thus, it is possible to obtain an evaluation of patient motor performance that is extremely accurate and objective (Hidler et al., 2005; Prange et al., 2006). Several kinematic and kinetic measures can be recorded offering a complementary evaluation of the motor performance with respect to traditional methods, i.e., clinical scales (Bosecker et al., 2010). A number of parameters have been defined quantifying, for example, the smoothness of the movement, the mean speed, the movement accuracy, the mean arrest period and also the details of the constituent sub-movements (Rohrer et al., 2002; Rohrer and Hogan, 2006; Panarese et al., 2012, 2016). Upper limb weakness (i.e., lack of strength) is another common consequence of stroke and its time evolution is an important clinical parameter. Robotic systems can continuously monitor the force signal exerted by the patient by means of force sensors (i.e., pressure sensors or load cells) and extract important parameters as average force amplitude, direction and number of force peaks performed during the motor task (Reinkensmeyer et al., 2000; Colombo et al., 2010).

#### Clinical Aspects of Robot-Mediated Motor Recovery in Humans

Up to now, a wide range of strategies and devices have been developed for promoting motor recovery after stroke by taking advantage from the brain's ability to reorganize its neural networks after the injury (Lamola et al., 2014). Traditional approaches towards rehabilitation can be qualified as ''bottom-up'' approaches as they operate at the peripheral effectors and expect for central nervous modifications (Chisari, 2015). Currently, robotic technologies and mechatronic devices represent the modern version of bottom-up treatments providing a high dosage of task-oriented training to patients affected by different degree of functional impairment (Fasoli et al., 2004). Robotic training can increase the intensity of therapy, and bring down the requirement of assistance for rehabilitation, with a consequent decrease of costs for the health care system (Barbeau and Visintin, 2003), especially in case of gait impairment.

Robotic systems for gait recovery can be essentially divided in two main categories: end-effectors or electromechanical exoskeletons. Examples of end-effector devices are the ''G-EO-System'' (Hesse et al., 2010), the ''Lokohelp'' (Freivogel et al., 2009) and the ''Gait Trainer GT 1'' (Hesse et al., 2010). End-effector systems are characterized by the absence of any constraint at the hip and knee joints and the presence of foot platforms which movement simulates the phases of the gait cycle (Hesse et al., 2010). Among the exoskeleton systems we find the ''LOPES'' (Veneman et al., 2005) and the ''Lokomat'' (Colombo et al., 2000). This kind of devices can be considered as ''fixed'' robotic gait orthosis that move the patient's lower limbs miming the gait kinematics by acting at hip and knee level (Hesse et al., 2010). The comparison of end-effector and exoskeleton devices in a systematic review (Mehrholz and Pohl, 2012) suggested that post-stroke walking recovery may depend on the type of robotic device, even if the lack or a direct comparison between the two typologies of device make difficult to base a final clinical recommendation.

Although a set of data does not support a clear benefit of robotic gait training when compared with therapist-assisted one (Hidler et al., 2005; Hornby et al., 2008) and the matter is still disputed (Husemann et al., 2007), several publications (Colombo et al., 2001; Werner et al., 2002; Schwartz et al., 2009; Westlake and Patten, 2009) highlighted that robotic gait training is at least equivalent to therapistassisted treadmill training in terms of efficacy and that electromechanical driven gait rehabilitation leads to a more symmetric gait pattern, to a lower spasticity and a more physiologic gait kinematics (Mayr et al., 2007). Nevertheless, despite the fact that robotic gait rehabilitation is paving the way for a substantial improvement of rehabilitation deliver, the way by which they determine restoration of function has not been yet clarified and the neurophysiological mechanisms underlying the recovery is still undefined. In a recent case series study (Chisari et al., 2014) the efficacy of Lokomat in gait rehabilitation and its capability to act on motor control has been tested in a group of stroke patients, resulting in a significant improvement after the training in clinical scales. Strength and Motor Unit firing rate of Vastus Medialis were also recorded and analyzed: no increase of force was observed whereas a significant increase of firing rate of Vastus Medialis was recorded, suggesting an effect of training on motorneuronal firing rate that may contribute to the improvement of motor control. In any case large effect size and robust effects of robotic treatment have not yet been fully demonstrated.

Concerning the upper extremity, impaired arm and hand function contributes considerably to limitations in the ability to perform activities of daily living (ADL). One of the goals of post stroke rehabilitation is to regain arm and hand function, since this is essential to perform ADL independently. Most of the robotic devices applied in clinical practice offer the possibility of choosing among four modalities for training: active, active-assisted, passive and resistive (Chang and Kim, 2013). These terms refer to subject's status during interaction. In active mode, performance arises from subject contribution only, whereas in passive mode the movement is performed by the robot regardless of subject's response. In active-assisted mode, the user performs active movement at the beginning and the robot acts only in particular conditions (i.e., if the target has not been reached in the requested time), systematically leading to success. Finally, resistive mode consists of resisting the movement received from the subject, so the robot makes the movement quite more difficult (Basteris et al., 2014). On this background, the ''assistance-as-needed'' control was conceived to encourage patients' active motion. In this approach, the robotic device is able to either assist or correct the movements of the subject, with the aim to manage simultaneous activation of efferent motor outputs and afferent sensory inputs during training (Belda-Lois et al., 2011). Current assist-as-needed strategies aim to provide the suitable definition of the desired upper limb trajectory in space and time the robot must generate to assist the subject during the task (Frisoli et al., 2011).

Robotic devices for upper limb rehabilitation are also divided in end-effectors and exoskeletons. End effectors act by applying mechanical forces to the distal joints of the arm. An example of devices which act as end-effector is the ''MIT Manus'' (Lo et al., 2010). The advantage of end-effector devices is to guarantee an easy setup; the disadvantage is due to the lack of control of the proximal segments of the arm, and this could result in undesired, compensatory patterns of movement. Conversely, exoskeleton robotic devices are designed so that robot axes are aligned with anatomical segments of the subject's arm. Exoskeletons allow direct control of individual joints, which can minimize abnormal posture or movement, but their construction is more complex and more expensive than that of the end-effectors. An example of exoskeleton device is the ''ARMEO'' (Taveggia et al., 2016).

For what concerns end-effectors, there are several studies comparing these devices with conventional therapy. A comparison between conventional therapy alone and robotic training combined with conventional therapy have been done in a sample of 56 subacute stroke patients (Fasoli et al., 2004), and the latter approach showed a greater improvement. This datum stands for a positive impact of end-effector devices on upper-limb recovery in patients in the subacute phase of stroke, thus recommending their use (Chang and Kim, 2013). Differently, another study conducted on a large sample of chronic stroke patients (Lo et al., 2010) showed that robotassisted upper limb training with end-effector device and intensive conventional therapy determined the same degree of clinical improvement after 12 weeks of rehabilitation, although after further 24 weeks of treatment the robotic approach resulted in further improvement. Another study involving chronic patients (Hsieh et al., 2012) compared high-intensity robot-assisted training with control treatment group and low-intensity robot-assisted training with control treatment group and found significantly greater improvement after robotic training only when performed in the high-intensity modality. These insights indicate that the intensity is the most relevant parameter in the rehabilitation of upper limb with robotic devices in chronic stroke patients, when end-effector are used. An effect of robot-assisted therapy on ADL function is observed only in patients with subacute stroke (Chang and Kim, 2013).

As regards Exoskeleton-type robot devices for upper limb recovery, almost all trials performed up to now comprised patients in the chronic stage of stroke. Among them, one study reported a significantly better effect on spasticity in the robot-assisted therapy group than in the conventional therapy group (Fazekas et al., 2007). In contrast, ADL function improved more markedly in the conventional therapy group that received the same amount of treatment. Other reports demonstrate no significant difference between robot-assisted therapy with exoskeleton devices and conventional therapies (Kahn et al., 2006; Mayr et al., 2008; Housman et al., 2009). In addition, there are no randomized controlled trials that investigate robot-assisted therapy with exoskeleton devices in patients with subacute stroke. Therefore, at this time there is insufficient evidence to draw a definite assertion as to what is the real effectiveness of exoskeleton systems for the rehabilitation of upper limb after stroke.

It has also been demonstrated that performing a task that mimics an ADL and that requires the coordination of more than one joint, a progressive complexity of the exercise and a greater number of DOF is more effective than standard two-dimensional reaching task (Schaefer and Hengge, 2016). Therefore, novel devices that combine the effect of end-effectors and exoskeletons have been proposed. One example is a new system that has been designed to combine functional grasping and active reaching in the same task; it also allows to perform bimanual exercises and consists of a system for weight support for the proximal joints, a robotic hand exoskeleton and a virtual reality interface (**Figure 3**; Barsotti et al., 2017). This integrated device showed improvement in motor performance and kinematic and strength metrics in a group of chronic stroke patients after an intensive rehabilitative treatment protocol (Sgherri et al., 2017).

To draw a conclusion, the impact of robot-assisted training is still debated: it guarantees intensive, repeatable, task-specific training, but up to now it seems an implementation to rather than a replacement for conventional physical therapy. Randomized controlled trials with large sample of patients are needed to draw more defined conclusions. A precise analysis of the economic burden and of the effectiveness of robotic rehabilitation is required.

Nevertheless, some authors argue that task-specific training alone is more likely to enhance behavioral compensation than effective recovery, highlighting the importance of providing the human equivalent of the ''enriched environment'' to augment the generalizing effect of spontaneous biological recovery instead of promoting compensation strategies (Zeiler and Krakauer, 2013). From this point of view, task-specific training is suggested to be delivered only involving ecologic and global tasks (i.e., reaching and grasping) and if accompanied by other approaches able to augment, prolong, or mimic the post-stroke sensitive period, such as NIBS techniques (see above, Section ''Non-Invasive Modulation of Post-Stroke Plasticity in Humans'').

In this perspective, a better understanding of the central mechanisms underlying both spontaneous and training-guided recovery becomes mandatory in order to maximally take advantage from the brain capacity to reorganize its neural networks after damage. The neural mechanisms underlying the

possible improvement led by robot-based therapeutic approach are still unclear, although recent studies have begun to shed light on this topic (Takahashi et al., 2008; Edwards, 2009; Posteraro et al., 2009; Pellegrino et al., 2012; Várkuti et al., 2013). An improved knowledge of these neural processes could help to ameliorate the effectiveness of the robotic therapy, and design combined protocols, i.e., robot-based and plasticizing therapy, able to boost the functional motor recovery. Further studies will allow a better understanding of the effect of rehabilitation on neural plasticity in order to adapt treatment resources to meet the needs of each patient and optimize the recovery process. In this context, animal models can be a suitable solution to investigate neural and motor changes after stroke and possible rehabilitative strategies, allowing the integration of behavioral, molecular and electrophysiological data.

#### Robotic Devices for Rodents

Recently, some robotic devices designed for interaction with rodents, in particular rats, have been developed to study a particular task or training, demonstrating the successful integration of mechatronics and robotics with such animal models. The idea to exploit robotics in combination with animal models is not addressed to developing new devices for animals and then translating them to new systems for human, that already have reached a high level of technology and complexity. Rather, the robot developed to interact with animals tend to mimic human robot devices to try to investigate mechanisms at the basis of robot-based rehabilitation as well as providing quantitative, reliable and accurate data where qualitative and experimenterbiased tests are still intensively used.

Most of these studies have focused on the development of technological-advanced devices aimed at to assessing and training gait function of spinal cord injured (SCI) rodents.

Reinkensmeyer' group (Nessler et al., 2005) proposed the first example of exoskeleton for rodent, the Rat-Stepper, a device consisting of two lightweight robotic arms connected to the rat hindlimbs, a body weight support system (BWS), and a motorized treadmill. This device can deliver precise amounts of weight support as well as perturbing and assisting the injured animals during the stepping. The authors claimed that the Rat-Stepper was able to accurately and easily capture step trajectories, provide sensitive and valid measures (e.g., length of the step or velocity) of locomotor recovery following contusion injury and even discriminate among different SCI levels (Nessler et al., 2006). This capacity to quantify the animal motor performance, without being affected by the experimenter's bias, makes robotic devices extremely appealing to the researchers. Moreover the same group designed a smaller but similar device for mice, the Mice-Stepper, used to provide an assisted-as-needed (AAN) paradigm for rehabilitate SCI animals. The results showed the highest level of recovery in AAN trained mice, measured through the number, reliability and frequency of steps during the testing sessions (Cai et al., 2006). The Mouse-Stepper was also used in combination with serotonin agonist, quipazine, and demonstrated significant recovery of locomotor function in SCI mice (Fong et al., 2005).

Another example of robotic apparatus specifically-targeted to research into the field of SCI is the IronRat system, recently developed in the laboratories of the MIT (Song and Hogan, 2015). The IronRat permits the rats to move freely in an open space (i.e., arena) and perform a wide range of voluntary movements. It is composed of a BWS, a monitoring-control system and the Rat Backpack. This latter component was designed as an exoskeletal module for the rat's hindlimbs, mounted on the lower back of the animal and coupled with hindlimb ankle joints. It allows hindlimbs to have two actuated DOF along the sagittal plane of ankle motion, and provides force feedback with the animal's hindlimb to compensate friction not compromising the back-driveability. Although this system appears quite bulky, after an initial habituation, healthy animals attached to the active Rat Backpack performed locomotion with stride length, stride duration, and duty cycle analogous to unconstrained overground movements (evaluated with BBB Locomotor Rating Scale; Basso et al., 1995). The IronRat is a promising tool in the field of SCI research, even though it has been tested only in healthy animals so far.

In the treatment of SCI on human, a key point is that the therapists can optimally adapt training to the requirements of each patient based on their own expertise and the severity of the injury. However both Rat-Stepper and IronRat cannot replicate the degree of adaptation of human therapists because of the limits imposed by their mechanical constraints. Florez et al. (2016) tried to overcome this issue by developing an exoskeleton for rats equipped with soft pneumatic actuators (SPA) attached to the hindlimbs. This system is composed of a BWS, a body structure customized to fit onto different sized subjects and an active part moved by the SPA. SPA are made in polymeric elastomers, can be tailored to any specific embodiment (Holland et al., 2014), are highly compliant and can produce high power-to-weight ratios (over 10 N with a 100 g actuator; Florez et al., 2016). The softness introduced by this actuation method makes the device suitable for interaction with fragile environments, as animal body, and also control fine mechanical stimulation that should allow SCI rats to produce a broad range of foot trajectories, improving the physical interaction between rodent and robot, as shown in their preliminary experiments.

A device that have been already proficiently validated and tested in the SCI treatment is the robotic exoskeletons developed by the Courtine' group (Dominici et al., 2012; van den Brand et al., 2012). Dominici et al. (2012) designed a versatile robotic neuroprosthesis consisting of three translational axes frame (x, y, z), as well as one rotational axis (ϕ). This device is provided with a suspension support (i.e., BWS) and a multidirectional elastic decoupling system that allows high-fidelity force control in each of the four DOF of the structure. The authors tested the efficacy of this device in combination with epidural electrical stimulation and tailored cocktail of serotonine and dopamine agonists (Dominici et al., 2012; van den Brand et al., 2012). To enable stepping in complete SCI rats: their results showed the capacity of this system to assess pattern generation and dynamic equilibrium, as well as promoting advanced locomotor capacities as walking and even stair climbing in SCI rats. This robotic postural interface was used to force, the rats to actively use their paralyzed hindlimbs in order to locomote bipedally (van den Brand et al., 2012) by encouraging an active participation of the animal.

Although so far most of the robot-based devices proposed for rodents have been developed to investigate gait function after SCI in rats, some robotic systems have been designed also to train and study forelimb function. These systems generally tend to automate already existing systems as lever pulling or similar. One of the first example of sensorized system to evaluate forelimb performance was presented by Fowler et al. (1994) where a simple force-sensing operandum was used to study the effect of a specific drug during a continuous pressure task performed by healthy rats. This system was thus used as means to quantify the performance in an unbiased-way.

Francis and Chapin (2004) developed a 1 DOF lever arm for rats. It was implemented to investigate feedforward and feedback control mechanisms in rodent forelimb motor tasks: indeed animals were trained to grasp the end-operandum and move it to some fixed targets. Different force field perturbations, implemented following different paradigms (i.e., viscous, constant torque, spring and isometric force fields) were opposed to the animal movement but, as author claimed, animals were able to adapt themselves to face all of these paradigms.

The application of force fields to effect on forelimb movements was studied and implemented also by Gassert's group who developed a robotic platform, ETH Pattus, designed to be used in motor learning experiments with rats (Vigaru et al., 2013). This compact device is highly transparent, has 3 DOF manipulandum (consisting of a pantograph frame provided with a further rotational DOF) and is capable to provide forces up to 2 N to guide or perturb rat forelimb movements, in accordance to different and possible force field implementations. Their preliminary experiments with healthy rats showed that ETH Pattus is able to collect data and quantitatively describe the dynamic interaction with the animal's paw. Despite its potentiality as a valuable tool to assess recovery after brain lesions (e.g., stroke), it was initially thought to investigate planar reaching and pronosupination movements, such as required when performing skilled reaching movements.

Skilled reaching is a natural behavior in rodents and its modeling has currently assumed a great importance and interest. In fact, although rodents display behavioral specializations quite different from humans, skilled reaching shares many similarities with the homologous behavior in humans. For example, velocity profiles in rat movements (Vigaru et al., 2013) exhibited bell-shaped profiles similar to humans (Morasso, 1981). However, traditional analyses assess motor performance using end-point measures of success/failure and only recently more sophisticated kinematic measures of movement execution (Dominici et al., 2012; Lai et al., 2015) or assessment based on mechatronic devices have been carried out.

Hays et al. (2013) developed an automatic system to train rats to perform a isometric pull task, described as novel technique to precisely measure the strength and the function of the forelimb during a skilled reaching task. The animals were taught to reach for a handle linked to a stationary force transducer and pull it isometrically until getting a threshold force level followed by the delivery of a food reward. Force, success rate, pull attempts as well as latency to maximal force are all quantitative parameters monitored by this device, through the commercial MotoTrak software (Vulintus, USA). Moreover, the authors tested the capacity of this system to detect deficits in rats undergone an ischemic stroke in M1 in comparison with other traditional tests (i.e., pasta matrix and end-points skilled reaching): the animals showed evident impairments in performing the task as described by all the three tests but only the isometric pull task was sensitive enough to detect significant deficit by 6 week post-lesion (Sloan et al., 2015). The same group from the UT Dallas also designed and tested a similar device for mice and observed long-term impairments in pulled force (Becker et al., 2016).

Similarly to the Hays's work, Reinkensmeyer's group recently developed another robotic interface, the Robotic Rehabilitator of the Rodent Upper Extremity (RUE), This 1 DOF system allows rats to voluntarily perform a task of reaching followed by the pulling of a bar in order to retrieve a food reward (Sharp et al., 2016). The bar is connected to an interface by means of which the force required to accomplish the pulling task can be changed, simply varying the stiffness of a voice coil actuator used to generate the resistance force. The authors used the RUE as a tool to assess the upper limb force production of healthy rats in comparison with a well-known method for measuring strength, i.e., the Grip Strength Meter (GSM; Anderson et al., 2004). They observed that the rats performed a pulling force higher compared to the maximum strength exerted through the GSM, showing that a more advanced system as the RUE can unveil more detailed behavioral information (e.g., rat's capacity to perform higher force). To date, RUE has been used as assessment tool but its utility even as rehabilitation device in SCI and stroke models seem to be clear.

Indeed robotic devices could be used to investigate the robotmediated recovery after neural injury and try to boost recovery. In this context, mouse models could seem more appealing for the possibility to exploit advanced techniques (e.g., optogenetics, knock-out technology) in combination with motor training. However, in spite of the advanced several methods and models, to date there is an evident lack of robotic systems to train mouse and in particular the functionality of its forelimbs. Spalletti et al. (2014) proposed the M-Platform (**Figure 4**), a 1 DOF robotic device for mice that mimics a human robot system for upper limb stroke rehabilitation (Arm-Guide; Reinkensmeyer et al., 2000). Head-restrained mice can be trained to perform intensive and highly repeatable exercises by retracting their forelimb previously extended by a linear actuator. Forces exerted during the task, time required for task execution (t-target), number of submovements and attempts (i.e., force peaks not overcoming a static friction force) can be quantified for each trial. The M-Platform is able to detect motor deficit in ischemic mice (i.e., local damage in M1) and to train animals to reach pre-injury performance. Moreover this system is currently being combined with advanced techniques, such as optogenetics and mesoscale brain imaging, to study plasticity mechanisms after stroke as well as pharmacological treatments aimed to boost recovery.

In conclusion, these devices try to implement similar control strategies or to extract the measures used in clinical cases, as trajectories, velocities and forces, with the obvious limitations related to the model and the task (Vigaru et al., 2013). However, these first examples of devices for robot-mediated neurorehabilitation in rodent models are paving the way to increase the understanding of the mechanisms underlying clinical improvements in patients affected by neural diseases (Grimaldi and Manto, 2013).

## CONCLUSIONS

In this review article, we analyzed recent advances in the understanding of the mechanisms of post-stroke network reorganization and discussed the use of the state-of-the-art therapeutic techniques, such as NIBS and robot-based protocols.

It is now well known that an ischemic damage leads to spontaneous neuroplasticity in perilesional tissue, promoting map reorganization observable both in human and animal models. Many researchers have focused on the study of different neurotransmitter systems which have an important role in this remapping process, such as the glutamatergic and the GABAergic networks. In fact, whereas pharmacological activation of AMPA receptors improves motor outcomes after stroke, a reduced inhibition is correlated to an enhanced plasticity. After stroke, an augmented activity in the contralesional hemisphere has been also reported in patients and animal models, particularly in acute stage, and even the contralesional neuronal connections appear to be altered. However the role of the healthy hemisphere in recovery is still controversial and debated.

A reasonable hypothesis is related to the lesion volume and the amount of spared tissue: in the case of sufficiently small injuries allowing the reorganization of spared adjacent motor areas, the contralesional hemisphere activity would have a negative impact on the recovery process. Indeed, subjects recovering from stroke showed changes in interhemispheric influences between the two hemispheres, probably due to a decrease of the ipsilesional neuronal activity and an increase of the contralesional one. For this reason, many studies have tried to shed light on the dynamics of the coupled brain areas and the local modifications of inhibitory

and excitatory neurotransmission after stroke. A loss of sensorimotor function has been correlated to deterioration of inter-hemispheric functional connectivity in animals. In human strokes, a loss of coherence and connectivity between the hemispheres has been documented, that slowly recovers in parallel with spontaneous behavioral improvements.

In such a scenario, NIBS can have the potential to foster recovery. Indeed several studies observed improvements in motor deficits after exciting the affected hemisphere or inhibiting the healthy one by implementing ''top-down'' protocols of rTMS or tDCS. Although a general consensus about the best protocols has been not achieved yet, it is believed that the general ''statedependency'' is a critical feature of the cortical neural processing and it plays a crucial role on the efficacy of NIBS protocols.

Another important approach we discussed to promote network plasticity and functional recovery is the robot-based rehabilitation. These devices guarantee intensive, repeatable, task-specific training, but up to now they represent just an adjunct to rather than a replacement for conventional rehabilitation therapy. Robot-assisted therapy is successful on improving upper limb motor function in stroke patients and the possibility of delivering high intensity treatment is one of the most important features of robotic technologies. However there are insufficient evidences to draw a definite conclusion regarding its effectiveness. The possibility to study plastic mechanisms of functional recovery has recently led to the introduction of robot-based paradigms even in animal models, in particular rodents. Many groups have demonstrated the successful integration of robotics with such animal models and laid the foundation to study neural mechanisms at the basis of robot-based rehabilitation as well as providing quantitative and accurate information about the recovery after neural injury.

# AUTHOR CONTRIBUTIONS

All authors contributed to writing and discussion of this review article.

# FUNDING

We gratefully acknowledge financial support from Fondazione Pisa (Project 158/2011). Work in the authors' laboratory is also supported by Regione Toscana (RONDA Project, ''Programma Attuativo Regionale'' financed by FAS—now FSC) and ERC Advanced Grant ''BrainBit''.

# REFERENCES


upper-limb impairment after stroke. N. Engl. J. Med. 362, 1772–1783. doi: 10.1056/NEJMoa0911341


of forelimb retraction in mice. Neurorehabil. Neural Repair 28, 188–196. doi: 10.1177/1545968313506520


hemispheric dominance in mouse pre-motor cortex. PLoS One 11:e0146858. doi: 10.1371/journal.pone.0146858


gait trainer for restoration of gait in subacute stroke patients: a randomized crossover study. Stroke 33, 2895–2901. doi: 10.1161/01.str.0000035734. 61539.f6


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Alia, Spalletti, Lai, Panarese, Lamola, Bertolucci, Vallone, Di Garbo, Chisari, Micera and Caleo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Influence of Aerobic Training and Combinations of Interventions on Cognition and Neuroplasticity after Stroke

Annabelle Constans 1† , Caroline Pin-barre1,2† , Jean-Jacques Temprado<sup>1</sup> , Patrick Decherchi <sup>1</sup> and Jérôme Laurin<sup>1</sup> \*

<sup>1</sup> Aix-Marseille Université, CNRS, ISM, UMR 7287, Marseille, France, <sup>2</sup> Université Nice Sophia Antipolis, LAMHESS, UPRES EA 6309, Nice, France

Stroke often aggravated age-related cognitive impairments that strongly affect several aspects of quality of life. However, few studies are, to date, focused on rehabilitation strategies that could improve cognition. Among possible interventions, aerobic training is well known to enhance cardiovascular and motor functions but may also induce beneficial effects on cognitive functions. To assess the effectiveness of aerobic training on cognition, it seems necessary to know whether training promotes the neuroplasticity in brain areas involved in cognitive functions. In the present review, we first explore in both human and animal how aerobic training could improve cognition after stroke by highlighting the neuroplasticity mechanisms. Then, we address the potential effect of combinations between aerobic training with other interventions, including resistance exercises and pharmacological treatments. In addition, we postulate that classic recommendations for aerobic training need to be reconsidered to target both cognition and motor recovery because the current guidelines are only focused on cardiovascular and motor recovery. Finally, methodological limitations of training programs and cognitive function assessment are also developed in this review to clarify their effectiveness in stroke patients.

Keywords: aging neuroscience, stroke rehabilitation, cerebral ischemia, angiogenesis, cognitive disorders, exercise intensity, neurotrophic factors, rat and human model

# INTRODUCTION

Sedentary older adults are prone to cardiovascular diseases, such as stroke (Bherer et al., 2013), which occurs when blood flow is interrupted to a part of the brain. This trauma leads to severe motor dysfunctions and it may also aggravate cognitive impairments resulting from normal aging (Rafnsson et al., 2007, 2009; Deary et al., 2009; Waldstein and Wendell, 2010). Indeed, stroke survivors have more than twice the risk of subsequently developing dementia compared with people who have never had a stroke (Tatemichi et al., 1992; Patel et al., 2002). For instance, a stroke situated on the left hemisphere might disturb language and comprehension, which reduce the ability to communicate (Karbe et al., 1990; Pirmoradi et al., 2016). When the right hemisphere is affected, the intuitive thinking, reasoning, solving problems as well as the perception, judgment and the visualspatial functions could be impaired (Tatemichi et al., 1994; Patel et al., 2002; Cumming et al., 2012; Sun et al., 2014b; Harris et al., 2015; Tiozzo et al., 2015; Save-Pédebos et al., 2016). It makes thus

#### Edited by:

Daniela Tropea, Trinity College Dublin, Ireland

#### Reviewed by:

In Koo Hwang, Seoul National University, South Korea Ramesh Kandimalla, Emory University, USA

#### \*Correspondence:

Jérôme Laurin jerome.laurin@univ-amu.fr

†These authors have equally contributed to this work and should be considered both as first authors.

> Received: 30 May 2016 Accepted: 21 June 2016 Published: 30 June 2016

#### Citation:

Constans A, Pin-Barre C, Temprado J-J, Decherchi P and Laurin J (2016) Influence of Aerobic Training and Combinations of Interventions on Cognition and Neuroplasticity after Stroke. Front. Aging Neurosci. 8:164. doi: 10.3389/fnagi.2016.00164 difficult for patients to locate objects, walk up or down stairs or get dressed. Consequently, cognitive disorders are one of the strongest predictor of the inability to return to work, that contribute to the socioeconomic burden of stroke (Kauranen et al., 2013). However, stroke-induced cognitive disorders are often underestimated relative to motor impairments because they are confused with pre-existing symptoms of age-related mild cognitive impairments or Alzheimer's Disease (AD; **Figure 1**; Sun et al., 2014b; Corriveau et al., 2016). Furthermore, cognitive impairments are frequently associated with poor motor recovery (Patel et al., 2002; Le´sniak et al., 2008; Rand et al., 2010). It suggests that stroke-induced cognitive dysfunctions and brain plasticity might also affect the stability, flexibility and learning of complex movements (e.g., locomotion, unimanual aiming, bimanual coordination), in which cognitive resources are highly involved as it was already observed in older adults (Temprado et al., 1999, 2013; Sleimen-Malkoun et al., 2012, 2013; Cohen et al., 2016).

It is thus of great importance to find effective interventions to induce both motor and cognitive improvement after stroke. In this respect, it is now widely established that aerobic training enhances cardiorespiratory fitness, muscular endurance and functional recovery of stroke patients resulting in a higher quality of life (Ivey et al., 2005; Macko et al., 2005; Marsden et al., 2013). Over the past few years, few studies have also shown that aerobic training could improve cognitive functions and promotes neuroplasticity in stroke patients (Quaney et al., 2009;

et al., 2011; Cumming et al., 2012). However, the outcomes on brain plasticity

Rand et al., 2010; Wogensen et al., 2015). In support of these observations, animal studies have revealed that such training effects on cognitive functions might be partially mediated by the release of neurotrophic factors that promotes angiogenesis, neurogenesis, synaptogenesis and synthesis of neurotransmitters that could not be directly investigated at central level in humans (Churchill et al., 2002; Cotman et al., 2007; Lövdén et al., 2010). However, despite firm evidence supporting the use of aerobic exercise for stroke patients, the mechanisms underlying neuroplasticity that is at origin of cognitive and motor recovery in stroke patients remain unknown.

In the present review, we first examine how aerobic training might play a beneficial role on cognition after stroke. In particular, we highlight the influence of aerobic training on neuroplasticity in both human and animal. We also discuss whether additional rehabilitation strategies and/or pharmacological treatments could accentuate neuroplasticity and consequently, the recovery of cognitive functions. In addition, we present the different exercise parameters that should be taken into account in aerobic training, namely: intensity, duration, frequency (number of session per week) and types of exercise as well as timing of training initiation after stroke onset. Indeed, they might differ in their effectiveness to improve cognition and motor functions. Therefore, classic recommendations for aerobic exercise are reconsidered to target cognition as well as motor and cardiovascular functions. Finally, we discuss about the methodological limitations that may hamper the understanding of aerobic training effects on cognition both in human and animal studies.

# INFLUENCE OF AEROBIC TRAINING ON NEUROPLASTICITY AND COGNITIVE FUNCTIONS

Cognitive disorders (i.e., executive function and/or information processing speed), motor dysfunctions (i.e., locomotion, balance and strength impairments) and cardiorespiratory fitness weakness are frequently observed in both elderly and stroke people. However, the severity of these impairments is exacerbated due to ischemia in older stroke patients (Gordon et al., 2008; Billinger et al., 2012; Cumming et al., 2013; Harada et al., 2013). Moreover, some neural mechanisms involved in cognitive disorders appear to be close between aging and cerebral ischemia. Specifically, a decrease of neurotrophic factor release (Ang et al., 2003; Silhol et al., 2005; Chae and Kim, 2009) or an abnormal level of oxidative stress and inflammation in hippocampus have been observed in both population (Joseph et al., 2005; Wang et al., 2007). In addition, cerebral blood flow decreases in ischemic region (Cupini et al., 2001; Zhang et al., 2013a) while the reduced vessel density in aging brain leads to an overall reduced blood flow and oxygenation into the brain (Petcu et al., 2010). Therefore, promoting angiogenesis might be critical for these two populations.

In numerous studies, it was demonstrated that aerobic training could induce beneficial effects on brain plasticity and associated cognitive functions as well as motor and

is specific to each brain disorders (Sopova et al., 2014).

cardiorespiratory functions in aging population (Patel et al., 2002; McAuley et al., 2004; Kramer and Erickson, 2007; Le´sniak et al., 2008; Erickson and Kramer, 2009). Thus, it can be hypothesized that such training could lead to similar positive effects on cognitive functions in stroke patients. In the following, we present evidence supporting this hypothesis in both older adults and animal studies. Then, on the basis of the few available studies in the literature, we made an overview of the effects of aerobic training on cognitive functions in people and animal with cerebral ischemia.

#### Studies on Healthy Older Adults

It is now recognized that aerobic training stimulates the positive plasticity of the aging brain. Such exercise-related plasticity mediates the maintenance, or even the increase, of cognitive performance as indicated by the improvement of executive functions and long-term memory. Such enhancements were observed in spite of the heterogeneity of methods, subject characteristics, training parameters or cognitive tasks used in the related studies (Colcombe and Kramer, 2003). Thanks to the use of sophisticated brain imaging technologies, cerebral modifications induced by aerobic training has been observed at both structural (i.e., increase in white and gray matter volumes and changes in synaptic connections) and functional (i.e., changes in brain activation patterns) levels (Churchill et al., 2002; Cotman and Berchtold, 2002; Colcombe and Kramer, 2003; Colcombe et al., 2004a,b, 2006; McAuley et al., 2004; Erickson and Kramer, 2009; Greenwood and Parasuraman, 2010; Voss et al., 2010a,b). For instance, using functional magnetic resonance imagery (fMRI), it was shown that elderly people aerobically trained displayed a higher activation of brain areas involved in attentional control and inhibitory functioning, while a reduced activation is observed in brain areas involved in behavioral conflict (Colcombe et al., 2004a). Additionally, 12-week of both bike and treadmill aerobic training increases the hippocampal and anterior cingulate cortex cerebral blood flow and also an immediate and delayed verbal memory. Such increase is closely related to neuroregeneration (Petcu et al., 2010; Chapman et al., 2013; Duzel et al., 2016). This result was recently reinforced by showing with the use of gadoliniumbased perfusion imaging (3 Tesla MRI) that aerobic fitness improvement in healthy older adults is correlated with changes in hippocampal perfusion and volume that were positively related to changes in recognition memory and early recall for complex spatial objects (Maass et al., 2015).

Circulating neurotrophic factor measurements such as brainderived neurotrophic factor (BDNF), insulin-like growth factor-1 (IGF-1) and vascular endothelial growth factor (VEGF; see below for details) might also explain the influence of aerobic training on neuroplasticity and cognition in Human but this point remains controversial (Cotman et al., 2007). In the one hand, few studies have observed an increase in serum BDNF level in older adults after 1-year of aerobic training that was correlated with an increase in hippocampal volume and improvement in spatial memory performance as well as in executive function (Erickson et al., 2012; Leckie et al., 2014). Specifically, Leckie et al. (2014), revealed that the oldest subjects exhibited the greatest increase in circulating BDNF levels and improvement in task performance after training. This observation suggests that age affects BDNF serum levels (Leckie et al., 2014). On the other hand, numerous studies did not provide robust evidence for enhancement of these neurotrophic factors after several weeks of training in aging people (Voss et al., 2013; Maass et al., 2015). For instance, serum BDNF, VEGF and IGF-1 level did not increase after 12-month of aerobic or non-aerobic (stretching exercises) training despite an increase of connectivity in the temporal lobe between the bilateral parahippocampus and the bilateral middle temporal gyrus (Voss et al., 2013). Recently, Maass et al. (2015), confirmed these results by showing that no difference in changes of circulating BDNF, VEGF and IGF-1 was observed after 3-month of treadmill (training group) or progressive-muscle relaxation/stretching (control group) despite an increase of hippocampal perfusion and volume (Maass et al., 2015). On the basis of these results, it was suggested that the training-related improvement of cerebral perfusion could lead to beneficial effects on neural function without any contribution of growth factors.

These contradictory findings might result from limitations associated with the circulating serum growth factor measurements and training protocols. For instance, nutritional status and changes in energy balance inducing by exercise interventions might affect growth factor levels (Monteleone et al., 2004). Moreover, as it was previously mentioned (Coelho et al., 2013), studies presented different training durations and exercise intensity and heterogeneous samples. Authors also included a different proportion of females and/or males in their studies. Such imbalance in gender composition of experimental groups could have an effect of baseline serum BDNF that might hide possible influence of aerobic training (Komulainen et al., 2008; Driscoll et al., 2012). It has also been suggested that the variable time-windows of circulating growth factor measurements and the analysis techniques used for quantifying blood growth factor levels could also affect the results and need to be clarified (Maass et al., 2015). Finally, data from circulating growth factor assessment should be interpreted with caution since no direct relationships between circulating and brain growth factor levels has been found in human. Studies on human only demonstrated that peripheral BDNF was associated with cognitive performance and cerebral structure integrity (Ventriglia et al., 2013). However, it was also established in animal model that BDNF, VEGF and IGF-1 might have the potential to cross the blood-brain barrier (BBB) in both directions in the central nervous system (CNS; Pan et al., 1998; Cotman et al., 2007). As an illustration, Karege et al. (2002), demonstrated that serum BDNF was positively correlated with cortical BDNF in rat model. In addition, blocking the peripheral IGF-1 or VEGF access to the brain impeded exercise-induced plasticity in the hippocampus (Karege et al., 2002; Cotman et al., 2007). However, other studies have shown in neurological disorders such as stroke and depression that changes in regional brain BDNF levels are not associated with changes of BDNF levels in peripheral blood (Elfving et al., 2010; Béjot et al., 2011). To date, it thus remains difficult to clearly determine the real role of circulating growth factors on cognitive function in older subjects.

#### Studies on Healthy Animals

Animal experiments could provide information about neuroplasticity mechanisms at cellular level in brain areas involved in cognitive functions. To date, three main neurotrophic factors have been identified to contribute to increase neuroplasticity after aerobic training: (i) the VEGF, which is a protein whose main role is to stimulate angiogenesis; (ii) the BDNF, which is a critical protein involved in brain plasticity. Indeed, this neurotrophin mediates numerous proteins expressions and molecular pathways, i.e., synapsin-I and synpatophysin, both involved in synaptic transmission and neurotransmitter release, C-AMP response element-binding protein (CREB) and beta calcium/calmodulinedependent kinase II (β-CaMKII) both contributing in long-term potentiation (LTP). BDNF thus promotes synaptic plasticity, neurogenesis and neuronal survival within the hippocampus (Vaynman et al., 2004; Cassilhas et al., 2012); and (iii) the IGF-1, which is a peptide that crosses the BBB and stimulates neurogenesis and angiogenesis. These growth factors act together to promote training-related benefits in brain plasticity and associated cognitive functions.

It was found in adult rats that aerobic training inducedangiogenesis is associated with an increase in brain VEGF mRNA and protein (Ding et al., 2006; Cotman et al., 2007). Moreover, Swain et al. (2003), have proved that prolonged voluntary exercise induced an increase of blood volume in the motor cortex of 19% greater than control animals (Swain et al., 2003). Angiogenesis needs to be promoted because some animal studies have suggested that angiogenesis was closely linked to adult neurogenesis and other neuroplasticity mechanisms (Pereira et al., 2007).

It was demonstrated that 8-week of aerobic training on treadmill induced an improvement in learning speed and spatial learning. Authors suggested that these results could be explained by an increase of hippocampal BDNF and IGF-1 and more precisely by the activation of BDNF/TrkB/β-CaMKII pathway and to a lesser extent to IGF-1/IGF-1R/Akt (also known as protein kinase B) pathway (Cassilhas et al., 2012). Radák et al. (2001, 2006) indicated that short- and long-term memory was improved after 8 weeks of regular swimming exercise in middleaged rats (14-month-old). BDNF and nerve growth factor (NGF) expressions were up-regulated as well as a reduction of accumulation of reactive carbonyl derivatives that are known to damage proteins, nucleic acids ad lipids (Radák et al., 2001, 2006). Similarly, one study on aging rat model, induced by D-galactose injection, has shown that 8-week of moderate aerobic training treadmill could importantly increase both level of NGF and his receptor, the tyrosine kinase A (TrkA) in the hippocampus. NGF/TrkA actives the phosphoinositide 3-kinase (PI3-K)/Akt pathway that decreases apoptosis (Wiesmann and de Vos, 2001; Chae and Kim, 2009). Excessive reactive oxygen species (ROS) production can also trigger apoptosis in brain areas such as the hippocampus that contribute to neurodegenerative disorders and cognitive function impairments in aging people. However, moderate intensity training on treadmill (18 m/min, 30 min/day during 15 weeks) in middle-age female rats (12-month-old) could enhance antioxidant defense system and thus induced a neuroprotective effect in the hippocampus (Marosi et al., 2012). Training also facilitated the release of metabolic proteins in the hippocampus such as mitochondrial precursor of ornithine aminotransferase and isocitrate deshydrogenase subunit alpha. However, these results remain difficult to interpret because authors suggested that alteration of mitochondrial proteins may be either reflecting mitochondrial damage or adaptation to mitochondrial dysfunction (Kirchner et al., 2008).

#### Studies on Stroke Patients

Only three studies investigated the role of aerobic training on cognition in stroke patients (**Table 1**). Two of them indicated that aerobic training might reduce cognitive disorders by improving functional outcomes as well as motor learning by the increase of information processing speed and implicit memory while executive functions remain affected (Quaney et al., 2009; El-Tamawy et al., 2014). However, patients did not preserve long-term beneficial effects 8 weeks after the end of aerobic training (Quaney et al., 2009). Contrary to what was observed in healthy subject, an acute and short aerobic exercise on treadmill (20 min; 70% of the heart rate reserve) did not induce cognitive improvement in stroke patient while a slightly improvement of the movement of upper-extremity was observed (Ploughman et al., 2008).

#### Studies on Animal with Cerebral Ischemia

Consistent with healthy older animals, neuroplasticity can be partially promoted by aerobic training through up-regulation of the expression of BDNF, synapsin-I and post-synpatic density protein 95 (PSD-95), also involved in LTP within hippocampus (dentate gyrus, CA1 and CA3 areas) and by increase of both CREB phosphorylation and newborn cell survival (Luo et al., 2007; Shih et al., 2013). The beneficial effect of training was reinforced when the activation of the high affinity BDNF receptor, the TrkB, was blocked because the function of BDNF was reduced and the exercise-induced spatial learning enhancement was impeded (Griesbach et al., 2009). Thus, consistent with data collected within aging rodents, it is supposed that BDNF-mediated pathway contributed to explain the improvement of spatial memory performances after cerebral ischemia. Furthermore, spatial memory performances were positively correlated with an increase of both newborn cell survival in bilateral dentate gyrus and restored microtubuleassociated protein 2 (MAP2) density after cerebral ischemia (Luo et al., 2007; Shimada et al., 2013). Treadmill training might also protect against cognitive impairments in rats with bilateral common carotid artery occlusion (CCAO) by reducing the lipoperoxidation in the hippocampus through an increase of antioxidant capacity and by improving the acquisition of a spatial task as well as the performance for both retention and working memory (**Figure 2**; Cechetti et al., 2012). In addition, few weeks of treadmill training in rat with cerebral ischemia increases the VEGF, and its regulatory protein, caveolin-1, and improves the cerebral blood flow in ischemic region (Zhang et al., 2013a; Gao et al., 2014). However, no study has directly shown that vascular changes could contribute to explaining the cognitive disorders.



Stepper; MoCA: Montreal Cognitive Assessment

 ; PM-47: Raven's Coloured Progressive

 Matrices 47; FAB: Frontal Assessment

 Battery; BDNF: Brain Derived Neurotrophic

 Factor.

#### IS THE COMBINATION OF AEROBIC TRAINING WITH OTHER TYPES OF THERAPIES EFFECTIVE FOR IMPROVING COGNITION RECOVERY AFTER STROKE?

### Combination with Other Rehabilitation Exercises

The question arises of whether different types of exercise could be combined to improve the effects that they could have separately. As reviewed by Chang et al. (2012), resistance exercise alone could positively impact selective cognitive functions in older adult such as information-processing speed, attention and several aspects of memory and executive functions (Chang et al., 2012). It was postulated that the main potential underlying mechanism for these benefits is based on resistance training-induced IGF-1 both in brain and blood circulation. Moreover, growing evidence have shown that the combination of aerobic and resistance training induced superior beneficial effect on cognition than aerobic exercise alone in aging population (Colcombe and Kramer, 2003). In this way, the association between resistance and aerobic training might be a potential strategy to further facilitate the cognitive function recovery in stroke population. However, few studies have been conducted to test this attractive hypothesis (Rand et al., 2010; Kluding et al., 2011; Marzolini et al., 2013). Results indicated that combination of aerobic training and lower extremity muscle strengthening improved executive functions, attention and voluntary motor control (Kluding et al., 2011). This is consistent with the study carried out by Marzolini et al. (2013), in which 6 months of both aerobic training at moderate intensity associated with resistance exercises induced improvements in the Montreal Cognitive Assessment (MoCA) scores in subdomain of attention/concentration and visuospatial/executive functioning (Marzolini et al., 2013). Another study on chronic stroke patients combining aerobic training with stretching, balance and task-specific exercises as well as sessions of recreation time has shown an improvement in verbal memory and cognitive flexibility but did not promote executive function (Rand et al., 2010). However, since the isolated effects of resistance and aerobic training were not compared with the combination of both of them, it remains difficult to ensure that such combination induces superior beneficial effects on cognition than aerobic intervention alone after stroke. It should also be noticed that the neuroplasticity associated with cognition recovery after combination of aerobic and other rehabilitation strategies has not been investigated in both human and animal stroke studies.

## Combination of Aerobic Training with Pharmacological Treatments

Currently, both aerobic and behavioral training have been shown to improve cognition on stroke patient but it remains insufficient to induce total recovery (Quaney et al., 2009; Cherry et al., 2014). Thus, it seems crucial to find new strategy combinations including aerobic exercise to amplify the effects on cognition. Many pharmacological treatments are possible candidates to improve cognitive function after stroke by stimulating brain plasticity. For instance, the drug memantine, used for the treatment of moderate to severe dementia of the Alzheimer's type, could increase synaptic plasticity and LTP and decrease overactivation of N-methyl-D-aspartate-type glutamate receptors (NMDA-r) in order to limit excitotoxic neuronal death (Martin and Wang, 2010; Berthier et al., 2011). Other substances, like citicoline, an intermediate in the generation of phosphatidylcholine from choline, could increase the level of acetylcholine and dopamine as well as decrease infarct volume in animal model with cerebral ischemia (Alvarez-Sabín and Román, 2011).

To date, only the intravenous (IV) recombinant tissue plasminogen activator (rt-PA) was approval by the USA Food and Drug Administration for the treatment of acute ischemic stroke in human (Marshall, 2015). The rt-PA is focused on thrombolytic events and could reduce the infarct volume (Nys et al., 2006) by restoring cerebral blood flow and oxygen supply to ischemic brain tissue (Thompson and Ronaldson, 2014). When rt-PA is administrated within the first 3 h after stroke, it could improve clinical outcomes and activities of daily life from 3, 6 and 12 months (Kwiatkowski et al., 1999; Nys et al., 2006; Murao et al., 2014). However, the effects of rt-PA administration during the first 24 h on cognitive functions are limited to aphasia that is explained by the reperfusion of language areas (only for one-third of patient; Nys et al., 2006; Hajjar et al., 2013; Kremer et al., 2013). Only one study used rt-PA in combination with physical activity during leisure-time. Results showed that combination of the 2 strategies did not accentuate the cognitive and physical improvements compared to rt-PA alone after 3 months (Decourcelle et al., 2015).

In preclinical study, L-deprenyl is a pharmacological treatment used to decrease ROS known to contribute to neuronal cell death in the ischemic core following permanent middle cerebral artery occlusion in mice (Unal et al., 2001). This substance has already shown the clinical effectiveness on cognitive functions in the elderly human with AD (Wilcock et al., 2002). L-deprenyl generates an amphetamine-like effect by acting on the release of dopamine and hampering its reuptake that contribute to the modulation of both attention and executive functions (**Table 1;** Bartolo et al., 2015). After stroke, combination of both L-deprenyl and standard rehabilitation including physiotherapy, cycle arm-ergometer training and occupational therapy during 6 weeks showed an improvement of logical memory, visual selective attention and non-verbal reasoning (Bartolo et al., 2015). However, this study did not compare the effect of such combination with the L-deprenyl treatment effect or the standard rehabilitation alone. Therefore, it remains difficult to postulate that the combination of these two strategies reinforced the cognitive recovery. In rat model, D-amphetamine administration combined to locomotor training enhances locomotor recovery after cortical injury (Feeney et al., 1982; Ramic et al., 2006). Moreover, other study assessed the effect of D-amphetamine administration combined with functional rehabilitation and physical therapy alone on cognitive memory performance and motor recovery after embolic stroke. Results showed that D-amphetamine with functional rehabilitation was more effective to improve cognitive performance than training alone, especially for memory, but the latter could improve fine motor performance (Rasmussen et al., 2006). By contrast, it was found in stroke patient that the combination of unique training session of—lower-extremity stability platform task, upper-extremity simulated feeding task and declarative learning—with D-cycloserine treatment did not accentuate the effect of motor and cognitive training (Cherry et al., 2014). This agent is already known to improve both motor and cognitive functions through increasing LTP in the hippocampus by regulating NMDA-r (Yaka et al., 2007). Cherry et al. (2014) postulated that an unique training session is not enough to amplify pharmacological treatment and that D-cycloserine could not act on the reduced NMDA-r function, and thus, on the motor task learning after stroke.

Finally, the review of literature suggests that no efficient combination between pharmacological treatments and training has yet found to improve recovery on cognitive functions. However, several combinations of pharmacological treatments and aerobic training have proven effectiveness on motor recovery after cerebral ischemia (Pin-Barre and Laurin, 2015). For example, the combination of both S-nitrosoglutathione and training accentuated neuroplasticity (reduce excitotoxicity, inflammation and protect BBB integrity) and motor recovery (Sakakima et al., 2012). Future studies should investigate the effects of these types of associations on cognitive recovery.

#### MOTOR VS. COGNITIVE FUNCTION RECOVERY: CAN AEROBIC TRAINING SIMILARLY FACILITATE BOTH?

Currently, the recommendations for the use of aerobic training after stroke were only established according to the effectiveness of endurance programs on cardiovascular and motor functions (Pang et al., 2013; Billinger et al., 2014). The cognitive outcomes are thus not taken into account for prescribing physical exercise. However, several findings arising from rodent studies indicate that the most appropriate intensity, the optimal timing to initiate training and the exercise mode (type of exercise) for improving motor performance might be different to improve cognitive function and motor recovery.

# Cognitive Performance Improvement by Aerobic Training

#### Exercise Intensity

Low-intensity aerobic training seems to be more effective than high-intensity training to promote cognitive health. Indeed, animal studies have shown that low-intensity training induced superior positive effects on spatial learning task in a water maze test and on both object recognition and location tasks than intense training (Shih et al., 2013; Shimada et al., 2013). Such cognitive task improvements were concomitant with an increase in plasticity-related proteins such as the increase of hippocampal BDNF level, synapsin-I as well as the number of dendritic spines and the number of neurons in the ipsilateral dentate gyrus (Shen et al., 2013; Shih et al., 2013; Shimada et al., 2013). It was mainly explained by the fact that, compared to low-intensity exercise, the high-intensity exercise induced higher levels of stress-hormone, which might downregulate BDNF level in hippocampus (Schaaf et al., 1999; Soya et al., 2007). Nevertheless, caution is required concerning the chosen running intensity. Indeed, the speed, fixed to 20–22 m/min, was considered as ''high'' intensity in Shih and Shimada's studies (Shih et al., 2013; Shimada et al., 2013) while others have classified such range of speed as moderate intensity (Sun et al., 2008; Ni et al., 2012). Therefore, the real influence of exercise intensity based on physiological parameters (for example, lactate threshold or VO2peak) on cognitive recovery remains to be further investigated (see ''Methodological Considerations'' Section).

#### Timing of Training Initiation

Late exercise, beginning after the first week post-brain injury, is more effective than early exercise (starting during the first week) to improve cognitive functions. Indeed, spatial learning and retention were better improved with the late training and was associated with the upregulation of BDNF (Griesbach et al., 2004; Clark et al., 2008). Voluntary running wheel training, starting 1 week after transient focal cerebral ischemia in mice, promoted neurogenesis in the adult dentate gyrus and spatial memory rebuilding (Luo et al., 2007). Interestingly, the Schmidt meta-analysis indicated that a start of skilled forelimb training from days 1 to 5 post-injury was more effective to improve cognitive function compared to late training (Bland et al., 2000; Wurm et al., 2007; Schmidt et al., 2014). It might be suggested that motor skilled training might start before aerobic training to improve cognitive functions.

#### Forced vs. Voluntary Training

The training mode might also influence the neuroplasticity underlying cognitive performance. Some studies have recently demonstrated that involuntary exercise (functional electrical stimulation), forced (treadmill) and voluntary (running wheel) training have similar beneficial effects on cognitive function after cerebral ischemia as indicated by improvement of both object recognition and location tests (Lin et al., 2015). Moreover, these three training regimens enhanced the levels of synapsin I, synaptophysin, PSD-95, MAP-2 and Tau protein in the hippocampus. It was confirmed by a previous study on healthy rats where 6 weeks of forced swimming or voluntary running resulted in similar increase of hippocampal BDNF level and in similar effect on learning capabilities and short/long term memories (Alomari et al., 2013). Nevertheless, other studies indicated divergent results. On one side, Hayes et al. (2008), demonstrated that, after 2-h of middle cerebral artery occlusion, forced treadmill training reduced infarct volume and increased the gene expression of heat shock proteins (Hsp), in particular the 27 kDa Hsp and the 70 kDa Hsp mRNA than voluntary exercise despite higher corticosterone level. The Hsp acts in the brain as molecular chaper ones with neuroprotective activities (Hayes et al., 2008). Other authors indicated that 12 weeks of forced treadmill training could protect against cognitive and biochemical impairments caused by CCAO in rat (Cechetti et al., 2012). Similar results were found after whole-brain irradiation where forced running training reduced the neurocognitive deficits but also the hippocampal neurogenesis impairments, i.e., the down-regulation of BDNFmediated pathway (including TrkB receptors, Akt and CREB, for example; Ji et al., 2014). On other side, some authors indicated that voluntary exercise is the most effective training in up-regulating the hippocampal BDNF level (Luo et al., 2007; Ke et al., 2011). Indeed, Ke et al. (2011), compared the effect of voluntary, involuntary and forced training after cerebral ischemia in rat. Results indicated that 7-day intervention of voluntary training induced higher level of BDNF in the hippocampus than the other modes (Ke et al., 2011). Such divergent results might be attributable to variable experimental designs.

## Motor Performance Improvement by Aerobic Training

#### Exercise Intensity

To date, post-stroke guidelines recommend moderate-intensity continuous aerobic training to improve aerobic capacity and motor recovery (40–80% of the maximum heart rate reserve; 3–5 sessions/week; 20–60 min/session). However, it was demonstrated that high-intensity exercise could improve aerobic fitness by increasing the peak oxygen uptake (VO2peak) and 6-min walk performances, that remained higher 1 year after the end of training compared with baseline value (Globas et al., 2012). In addition, it has been found that the high-intensity interval training (HIT) could promote superior beneficial cardiovascular and muscular adaptations among persons with different cardiorespiratory disorders (Rognmo et al., 2004; Wisløff et al., 2007). HIT is defined as repeated series of brief and intense exercise separated by active or passive rest periods. This type of intense training is also well known to be a time-efficient strategy to promote metabolic adaptations because the total session duration is strongly reduced compared to traditional moderate-intensity continuous aerobic training (Sun et al., 2014b). Such finding is important given that ''lack of time'' remains the most cited barrier to regular aerobic exercise participation.

It was observed that HIT is well accepted for ambulatory chronic stroke and could induce encouraging improvements of quality of life as observed by improvements of VO2peak and work economy (Gjellesvik et al., 2012; Boyne et al., 2013; Mattlage et al., 2013). Recently, only one study has compared HIT (series of 30 s at maximum tolerated treadmill speed separated by 30–60 s rest periods) and traditional moderate-intensity continuous aerobic training (Boyne et al., 2016). Authors indicated that no clear difference between HIT and moderate-intensity continuous aerobic training was observed because of a small sample size. Nevertheless, this type of training seems to be feasible and safe because no adverse events occurred.

#### Timing of Training Initiation

It was postulated in human studies that starting rehabilitation program in the acute-subacute phase after stroke could prevent complications relating to prolonged inactivity (i.e., deconditioning period) and presented a low relative risk for adverse effects (Musicco et al., 2003). Numerous studies in rodents with cerebral ischemia have also observed that an early starting aerobic exercise (from days 1 to 5) was more effective to improve the running performance and to reduce the infarct volume compared to late training (Bland et al., 2000; Luo et al., 2007; Schmidt et al., 2014). For example, Park et al. (2010), indicated that early treadmill training could better improve the motor performance, using the Rotarod test, than late treadmill training after hemorrhagic stroke (Park et al., 2010). Such early training did not increase the infarct volume or brain edema in accordance to other studies (Matsuda et al., 2011; Zhang et al., 2013b). Likewise, an early treadmill exercise increased the cellular expression levels of some neurotrophic factors, promoted cell growth and reduced the expression of apoptosis markers (Mizutani et al., 2010; Matsuda et al., 2011). Moreover, an early endurance exercise improved blood flow in the ischemic region and promoted angiogenesis (Zhang et al., 2013b). We may also add that sensorimotor deficits and cortical infarct volume were aggravated on a longer-term when training started too soon i.e., before 24 h post-ischemia (Kozlowski et al., 1996; Risedal et al., 1999; Bland et al., 2000; DeBow et al., 2004; Schmidt et al., 2014).

#### Forced vs. Voluntary Training

Numerous studies highlighted that forced treadmill training is more effective than all the other types of exercise, included voluntary exercise, to improve running function, aerobic fitness and to reduce infarct volume (Takamatsu et al., 2010, 2016; Schmidt et al., 2014). However, some authors indicated that voluntary exercise is more effective to improve motor recovery using the De Ryck's behavioral test (Ke et al., 2011). These controversial results might be explained by the use of different motor behavioral tests between studies as well as by a different training protocol (variable speed and timing of training initiation).

#### Concomitant Improvement of Cognitive and Motor Functions Induced by Aerobic Exercise

Interestingly, Sun et al. (2014), might find a compromise in rats by proving that training with gradually increased intensity on treadmill (from 5 to 26 m/min) could better improve motor function and produce higher hippocampal BDNF with lower stress compared to both stably low and high intensity training (Sun et al., 2014a). These results were in accordance with Zhang et al. (2012), study in which both motor performance (forelimb placing, stepping coordination) and spatial memory in rats with middle cerebral artery occlusion-reperfusion were improved after progressive intensity aerobic training (Zhang et al., 2012).

On the basis of these findings, it appears that training might alternate between high- and low-intensity sessions or might progressively reach high-intensity to accentuate improvement of either cognitive or motor performance. It also suggested that treadmill training might be appropriate for improving these two functions. However, it is more difficult to find a compromise for the timing of training initiation. Indeed, an early training seems to be more appropriate to promote motor recovery while cognitive performances were improved when aerobic training started later. Therefore, the influence of aerobic training on cognitive deficits might be considered to complete the actual exercise recommendations.

#### METHODOLOGICAL CONSIDERATIONS

Although it is currently admitted that aerobic training positively affects neuro-cognitive impairments, available studies reveal a great heterogeneity in the methods used and for some, weaknesses, which make results difficult to compare (Cumming et al., 2012). These methodological limitations, which are either specific or common to human and animal models, need to be considered before interpreting results.

#### Methodological Considerations Concerning Exercise Parameters in both Animal and Human Studies

Available studies strongly differ in the parameters related to exercise during aerobic training: duration, intensity, frequency, mode and timing of rehabilitation initiation. In particular, exercise intensity, which is a critical parameter of aerobic training effectiveness, deserves to be questioned (Pin-Barre and Laurin, 2015). For both human and animal models, intensity is frequently based on empirical speed/power (Ploughman et al., 2008; Kluding et al., 2011; Påhlman et al., 2012; Shih et al., 2013; Shimada et al., 2013). In some human studies, the intensity was determined from subjective parameters such as level of exertion perceived by the patient (Ploughman et al., 2008; Kluding et al., 2011; Påhlman et al., 2012). In these conditions, exercise intensity (moderate, intense and severe) could not be precisely determined because no physiological markers were recorded (Xu and Rhodes, 1999). Therefore, a given absolute intensity was considered as moderate for some authors but as severe for others. When training intensity was based on physiological markers, percentages of maximal heart rate or the maximal oxygen uptake (VO2peak) were the most frequently used parameters (Ploughman et al., 2008; Quaney et al., 2009; Kluding et al., 2011; El-Tamawy et al., 2014). However, these methods are insufficiently reliable to distinguish between high and moderate training intensities. Indeed, patients barely reach their maximal aerobic capabilities during an incremental test. Recently, submaximal parameters such as ventilatory or lactic threshold have been recognized to be more suitable than VO2peak to induce a higher interindividual reproducibility in physiological response to exercise (Faude et al., 2009; Marzolini et al., 2013; Bosch et al., 2015). It is noticeable however that there is no consensus on the methods used to measure these physiological parameters from an incremental exercise test (Bentley et al., 2007). Indeed, depending on the type of incremental test, the performance and the related physiological parameters could be altered. The chosen ergometer (ergocycle or treadmill), stage duration as well as the magnitude of intensity increment between each stage are known to affect the performance (for review see, Bentley et al., 2007). Moreover, the progressive increase of exercise intensity on treadmill could be induced either by an elevation of speed (m/s) and/or inclination (percentage). The speed increment is not systematically reported in literature but some authors indicated that the grade increment was increased of 2% every 2 min with a constant (high) walking speed (Voss et al., 2013). Increase in the slope of the treadmill needs to be considered to improve the validity and the relevance of the chosen incremental test on treadmill for aging subjects. Indeed, the increment of treadmill inclination seems to be more appropriate for aging people and/or for the individuals for whom running is impossible or difficult. For instance, a lower running/walking speed on treadmill could reduce the perceived exertion of the exercise for some individuals and thus might reach highest intensities (Ehlen et al., 2011). Moreover, it was reported on obese persons that faster walking speeds might increase the risk of musculoskeletal injuries because of higher reaction forces and loading rates (Ehlen et al., 2011) in lower extremities tendons, joints and ligaments (Puga et al., 2012). Finally, most authors frequently used a stationary cycle ergometer for aging people (Maass et al., 2015) because measurements of physiological parameters during the test are more stable using this device compared to treadmill. The risk of falls is also lower on cycle ergometer. For the rodent model, some studies have proposed different treadmill protocol in order to reach the highest VO2peak by modifying the treadmill inclination. It has been found in both rats and mice that the highest VO2peak was reached at 25◦ because a distinct leveling-off of VO<sup>2</sup> was mainly observed at this inclination (Wisløff et al., 2001; Kemi et al., 2002).

In addition, among different studies, intensities are rarely individualized, especially in rodents, while training individualization is one of the most important recommendations of stroke rehabilitation (Pang et al., 2013; Schmidt et al., 2014). This limitation might attenuate the ''real'' effectiveness of aerobic training. Finally, it is commonly considered in exercise physiology studies that energy expenditure needs to be equivalent between exercise types in order to compare the different effects of a specific training parameter (such as intensity or duration). In this way, all the experimental groups have the same energy expenditure and thus only the influence of a tested exercise parameter is assessed (Rognmo et al., 2004; Wisløff et al., 2007). However, it has never been applied in animal as well as in human stroke studies.

#### Specific Methodological Considerations in Human

Inter-individual differences in cerebral ischemia location and/or aerobic fitness level may affect, positively or negatively, cognitive impairments. However, they remain difficult to counteract (Tang et al., 2013; Sun et al., 2014b). For example, it was found that patients with infarction located within cortical regions, middle cerebral artery territory and/or on left hemisphere were more prone to cognitive impairments (Sun et al., 2014b).

Otherwise, using a cognitive test that did not detect the specific cognitive impairments of a patient might hide some potential effect of a training intervention (McDonnell et al., 2011; Cumming et al., 2012). For instance, cognitive measurements are frequently limited to clinical tests, as functional independence measures (FIM), that are not enough sensitive. Likewise, mini-mental state examination (MMSE) may underestimate stroke-related cognitive deficits because it presents a lack of sensitivity for identifying disorders of visual perception and of high-order executive functions (Nys et al., 2006; Pendlebury et al., 2010; Cumming et al., 2012). This might be problematic given that these latter cognitive functions are frequently affected by stroke (Sun et al., 2014b; Tiozzo et al., 2015). In this respect, the MoCA can assess numerous cognitive impairments such as executive function, attention and delayed recall disorders that appear to be more suitable for stroke patient (Pendlebury et al., 2010). Some studies have also used specific neuropsychological tests such as Trail-making part A and B, Symbol digit test, Stroop test, Digit backward test, which allow to better detect cognitive deficits induced by stroke (Ploughman et al., 2008; Quaney et al., 2009; Rand et al., 2010; Kluding et al., 2011).

To ensure that cognitive performance improvements are related to aerobic training effectiveness, an increase of aerobic fitness needs to be observed at the end of the intervention. However, change in cardiorespiratory fitness after aerobic training is not systematically reported in the different studies. Thus, caution is often required when it is claimed that cognitive improvements are associated with aerobic training rather than other interventions or environmental factors.

The issue of cognitive-motor interactions in stroke patients also deserves to be considered. Indeed, cognitive and motor processes are classically considered as functionally independent and then, explored separately in the literature. However, the control and learning of complex goal-directed movements require a close cooperation between sensorimotor control processes and higher cognitive functions. This is even more marked in older adults, for which cognitive and motor processes become less differentiated by virtue of functional reorganization of brain activation patterns. Thus, change in cognitive-motor interplay expresses an important facet of age-related intrinsic plasticity of brain and cognition. Strategic variations might be thus analyzed to assess behavioral adaptability in cognitive (Lemaire and Hinault, 2014) and sensorimotor tasks (Poletti et al., 2015, 2016).

#### Specific Methodological Considerations in Animal Models

Animal experiments can provide information about underlying mechanisms of neuroplasticity that could not be investigated in human. However, several drawbacks are often observed in some studies. For instance, it is impossible to investigate the large range of cognitive functions identified in human, such as verbal learning and memory for example (Voss et al., 2013). Cognitive dysfunctions after cerebral ischemia are limited to spatial and working memory, recognition and motor learning skills (Morris water maze, passive avoidance test, object recognition or location test; Luo et al., 2007; Griesbach et al., 2009; Cechetti et al., 2012; Shih et al., 2013; Shimada et al., 2013).

In addition, exercise-induced neuroplasticity are mainly explored within hippocampus (CA1 and CA3 areas and dentate gyrus), which is related to memory and learning (Vaynman et al., 2003). However, other areas such as basal ganglia, prefrontal cortex, thalamus and cerebellum are also involved in learning and memory processes, executive functions and motor control (Graybiel, 1995; Doya, 2000; Johnson and Ojemann, 2000). Except the hippocampus, other regions, remote away from the lesion zone, are connected to the affected structures and thus might also be disturbed after stroke (i.e., diaschisis effect and synaptic inhibition). For example, inflammatory responses could be observed within thalamus or substantia nigra after cortical brain injury that might partially contribute to explain the cognitive deficits (Block et al., 2005). It could be relevant to investigate the effect of aerobic training on the cognitive functions of these cerebral areas (Carmichael et al., 2004).

#### CONCLUSION

This present article provides an overview of the positive effect of aerobic training on cognitive functions. It seems that training could increase the release of the same neurotrophic

#### REFERENCES


factors (BDNF and VEGF) in both elderly and stroke people that mediate beneficial neuroplasticity in brain areas involved in cognitive functions. We also identify some methodological limitations in both human and animal studies such as the standardization procedure of aerobic exercise intensity and the chosen cognitive tests, depending on the target population, that remain one of the most important concerns. Moreover, our review article suggests that the combination of aerobic training with other exercises/therapies or treatments represent a promising strategy with strong clinical perspectives. Importantly, this review highlights the lack of a firm guideline for exercise recommendations targeting recovery of cognition in stroke patient. Therefore, no standard aerobic protocol has yet been established as a commonly accepted reference regarding intensity, timing of training initiation and exercise type. It also appears that investigating changes in cognitive-motor interplay are critical to develop appropriate rehabilitation to improve both cognition and motor control after stroke.

#### AUTHOR CONTRIBUTIONS

All authors listed, have made substantial, direct, and intellectual contribution to the work, and approved its final version for publication. JL and J-JT: conceived the review focus, conducted literature review, summarized, and finalized the manuscript. PD: summarized, and finalized the manuscript. AC, CP-B and JL: reviewed literature, wrote first draft, and finalized the manuscript. All authors approved final version of manuscript.

#### FUNDING

This work was supported by Aix-Marseille Université (AMU) and Centre National de la Recherche Scientifique (CNRS).


for exercise prescription. Arch. Phys. Med. Rehabil. 96, 1779–1784. doi: 10. 1016/j.apmr.2015.04.023


a pilot study with one year follow-up. Top. Stroke Rehabil. 19, 353–360. doi: 10. 1310/tsr1904-353


treadmill exercise rats. Exp. Neurol. 212, 145–151. doi: 10.1016/j.expneurol. 2008.03.014


localization and lateralization during magnetoencephalography. Epilepsy Res. 119, 1–9. doi: 10.1016/j.eplepsyres.2015.11.015


to functional abilities. J. Neurol. Neurosurg. Psychiatry 57, 202–207. doi: 10. 1136/jnnp.57.2.202


and reinstates long-term potentiation (LTP) in a mouse model of closed head injury. FASEB J. 21, 2033–2041. doi: 10.1096/fj.06-7856com


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Constans, Pin-Barre, Temprado, Decherchi and Laurin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# C1q/Tumor Necrosis Factor-Related Protein-3 Attenuates Brain Injury after Intracerebral Hemorrhage via AMPK-Dependent Pathway in Rat

Shaohua Wang1,2† , Yang Zhou1,2† , Yang Bo1,2 , Lingyu Li 1,2 , Shanshan Yu1,2 , Yanlin Chen1,2 , Jin Zhu1,2 and Yong Zhao1,2 \*

<sup>1</sup> Department of Pathology, Chongqing Medical University, Chongqing, China, <sup>2</sup> Key Laboratory of Neurobiology, Chongqing Medical University, Chongqing, China

C1q/tumor necrosis factor (TNF)-related protein-3 (CTRP3) is a recently discovered adiponectin paralog with established metabolic regulatory properties. However, the role of CTRP3 in intracerebral hemorrhage (ICH) is still mostly unresolved. The aim of the present report was to explore the possible neuroprotective effect of CTRP3 in an ICH rat model and to elucidate the fundamental mechanisms. ICH was induced in rats by intracerebral infusion of autologous arterial blood. The effects of exogenous CTRP3 (recombinant or lentivirus CTRP3) on brain injury were explored on day 7. Treatment with CTRP3 reduced brain edema, protected against disruption of the blood-brain barrier (BBB), improved neurological functions and promoted angiogenesis. Furthermore, CTRP3 greatly intensified phosphorylation of AMP-activated protein kinase (AMPK) in addition to expression of hypoxia inducing factor-1α (HIF-1α) and vascular endothelial growth factor (VEGF). Finally, the protective effects of CTRP3 could be blocked by either AMPK or VEGF inhibitors. Our findings give the first evidence that CTRP3 is a new proangiogenic and neuroprotective adipokine, which may exert its protective effects at least partly through an AMPK/HIF-1α/ VEGF-dependent pathway, and suggest that CTRP3 may provide a new therapeutic strategy for ICH.

#### Edited by:

Daniela Tropea, Trinity College Dublin, Ireland

#### Reviewed by:

Dennis Qing Wang, Third Affiliated Hospital of Sun Yat-Sen University, China Tatsuro Mutoh, Fujita Health University, Japan

> \*Correspondence: Yong Zhao blys01@163.com

†These authors have contributed equally to this work.

Received: 26 July 2016 Accepted: 29 September 2016 Published: 19 October 2016

#### Citation:

Wang S, Zhou Y, Yang B, Li L, Yu S, Chen Y, Zhu J and Zhao Y (2016) C1q/Tumor Necrosis Factor-Related Protein-3 Attenuates Brain Injury after Intracerebral Hemorrhage via AMPK-Dependent Pathway in Rat. Front. Cell. Neurosci. 10:237. doi: 10.3389/fncel.2016.00237 Keywords: CTRP3, intracerebral hemorrhage, brain edema, blood-brain barrier disruption, neuroprotection, angiogenesis

# INTRODUCTION

Intracerebral hemorrhage (ICH) is responsible for about 10–15% of stroke cases. Nevertheless, it is the most alarming type due to its large fatality rate and poor functional outcome (Qureshi et al., 2009; Manaenko et al., 2013). ICH results from rupture of blood vessels in the brain. The quick release of blood into the parenchyma leads to substantial mechanical damage that may only be somewhat ameliorated by limiting hematoma volume (Mayer et al., 2008). Meanwhile, secondary damage ensues due to toxic effects of released blood components, such as thrombin, as well as erythrocyte rupture (for example, iron-catalyzed free radical reactions). These events, including disruption of the blood-brain barrier (BBB) as well as growth of edema and inflammation, are currently therapeutic targets for ICH (Fingas et al., 2009; Keep et al., 2012).

Therapies such as angiogenesis have been suggested for ICH. Growing evidence suggests that after ICH, angiogenesis is upregulated in damaged brain tissue of the peri-hematoma area, leading to compensatory cerebral vascular network remodeling (Teng et al., 2008; Lei et al., 2015). Past reports showed that many angiogenic aspects could ease ischemic brain injury, elevate focal blood flow and enhance neurological results, which indicate that recently formed microvessels are indeed functional (Hao et al., 2011; Shen et al., 2013). Blood vessels are a significant scaffolding factor that assist with the migration of neurons to damaged brain areas and supply trophic material to new neurons (Kojima et al., 2010; Lei et al., 2013). The molecules that encourage neurogenesis and angiogenesis after brain injuries (e.g., ICH) are still unidentified.

Recently, a highly conserved family of adiponectin paralogs, C1q/tumor necrosis factor(TNF)-related proteins (CTRPs), was identified. Each of the 15 known members (CTRP1–CTRP15) is made up of four separate domains including an N-terminal signal peptide, a short variable domain, a collagen-like domain, and a C-terminal C1q-like globular domain (Ahima et al., 2006). Both CTRPs and adiponectin are a part of the C1q/TNF protein superfamily, which proceeds to increase in size as more C1q domain proteins are identified (Yi et al., 2012). The CTRP family members exhibit broadly diverse physiological functions, including regulation of metabolism, protection against endothelial dysfunction and angiogenesis.

CTRP3 is ubiquitously expressed in adipocytes, cartilagocytes, monocytes, fibroblasts, placenta, osteosarcoma, chondroblastoma, giant cell tumor, colon, small intestine, pancreas, kidney, thymus, ovary and in brain (Schaffler and Buechler, 2012). Most importantly, CTRP3 is the only one whose biological functions have been identified (Peterson et al., 2010). It was found that CTRP3 can encourage in vitro endothelial cell proliferation and migration (Akiyama et al., 2007). But, the role of CTRP3 in promoting angiogenesis in ICH-induced brain injury is not yet known. Further, whether or not CTRP3, an important member of the most recently discovered adipokine family, works as a mediator or inhibitor of ICH has not been studied previously. Therefore, the goals of this research were: (1) to investigate the effects of exogenous CTRP3 in an ICH rat model; (2) to determine whether CTRP3 administration promotes angiogenesis after ICH; and (3) to elucidate the role of CTRP3 in pathogenesis of ICH.

#### MATERIALS AND METHODS

#### Experimental Animals

All animal studies were given approval by the Chongqing Medical University Biomedical Ethics Committee. All experimental procedures were done in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. All efforts were made to minimize the number of animals used and their suffering. A total of 115 adult male Sprague-Dawley rats (60–80 d old, 240–300 g) were used for the in vivo study.

# Establishment of Intracerebral Hemorrhage Model

ICH was induced by an intrastriatal blood infusion method as described previously (Ni et al., 2015). Briefly, rats were deeply anesthetized with chloral hydrate (350 mg/kg, intraperitoneal injection) and placed in a stereotaxic frame (Kopf Instruments, Tujunga, CA, USA). After removing the hair and cleaning the scalp, the skin was incised. A burr hole was drilled 0.2 mm anterior and 3.0 mm lateral right of bregma. Whole blood (50 µL), which was drawn from the femoral artery, was infused manually over 10 min via a 26 G needle inserted into the striatum at a depth of 5.8 mm below the surface of the skull. After 10 min, the needle was steadily taken out for 5 min followed by the sealing of the burr hole with a sterilized medical bone wax. The wound was cleaned, and the scalp was sutured. The animals were given time to heal in their cages. During the recovery period, the animals had unlimited access to nourishment.

#### In vivo Experiments

Rats were given free access to food and water in an optimal environment preceding the operation. Three in vivo experiments were performed as described below.

#### Experiment 1

Adult rats were split at random into the following four groups: sham-operated (sham) group, ICH group, ICH + vehicle group and ICH + recombinant CTRP3 (rCTRP3, Chimerigen, USA) group. rCTRP3 was injected intracerebroventricularly (80 µg/kg) 30 min after ICH and then three times per week until the animals were killed. Neurological deficits (assessed by a modified Garcia test, beam walking test and wire hanging test), hematoma volume, BBB integrity and brain edema were measured 7 days after ICH, and samples for western blot, qRT-PCR and immunohistochemistry were collected.

#### Experiment 2

Adult rats were split at random into the following four groups: sham-operated (sham) group, ICH group, ICH + null vector control (Lenti.null) group, ICH + lentivirus overexpression of CTRP3 (Lenti.CTRP3) group. Fourteen days after Lenti.CTRP3 intracerebroventricular injection, the rats underwent ICH. Neurological deficits, hematoma volume, BBB integrity and brain edema were measured 7 days after ICH, and samples for western blot, qRT-PCR and immunohistochemistry were collected.

#### Experiment 3

Adult rats were split at random into the following four groups: ICH group, ICH + rCTRP3 group, ICH + rCTRP3 + compound C (Com.C) group (AMP-activated protein kinase (AMPK) axis inhibitor, 20 µg/kg, intracerebroventricular injection, 3 times per week), and ICH + rCTRP3 + CBO-P11 (CBO) group (vascular endothelial growth factor (VEGF) inhibitor 40 µg/kg, intracerebroventricular injection, 3 times per week). Neurological deficits and BBB integrity were measured 7 days after ICH, and samples for western blot and immunohistochemistry were collected.

#### Lentivirus-CTRP3 Gene Transfer in the Rat Brain

Adult rats were anesthetized with chloral hydrate (350 mg/kg intraperitoneal injection) and then placed in a Kopf stereotactic frame. A burr hole was bored in the pericranium 0.9 mm lateral to the sagittal suture and 1.9 mm posterior to the coronal suture. A 10 µL microinjection pump (WPI Inc., Sarasota, FL, USA) was stereotactically inserted 3.5 mm deeper into the cortex. A 5 µL viral suspension consisting of 1 × 10<sup>9</sup> genomic copies of the lentivirus-CTRP3 (Lenti-CTRP3) gene, which was injected ipsilaterally into the right lateral cerebral ventricle at a rate of 0.2 µL/min. The needle was taken out after 15 min of injection. The animals then were allowed to heal and brought back to their cages.

#### Behavioral Assessment

Behavioral functions were measured using a modified Garcia test, beam walking test, and wire hanging test 7 days after ICH in a blind fashion (Chen et al., 2013). In the modified Garcia test, rats were given a score of 0–18. The scoring system consisted of six tests, with possible scores of 0–3 (0 = worst; 3 = best). The minimum score was 0 and the maximum was 18. Beam walking and wire hanging tests utilized bridges (550 cm wire or 590 cm beam) between two platforms on which the rats were placed in the center. Rats were assessed according to six criteria that described if the animal could reach the platform and use its limbs in a symmetrical manner and were then given a score of 0–5 (normal). The average of three trials per test for each animal was calculated.

#### BBB Permeability

Seven days after ICH, rats were intravenously injected with 2% Evans blue dye (4 mL/kg; Sigma-Aldrich, St. Louis, MO, USA). Three hours later, the amount of extravasated Evans blue dye in the hemorrhagic brain hemispheres was evaluated by spectrophotometry (Thermo Scientific, MA, USA) at 620 nm (Cai et al., 2015).

#### Brain Water Content

Seven days after ICH, the cerebral hemisphere was cut into 4-mm thick blocks around the needle track. Brain tissues were immediately weighed using an analytical balance and heated at 100◦C for 24 h to obtain the dry weight. The water content was calculated using the following formula: (wet weight– dry weight)/wet weight × 100%.

#### Hematoma Volume

Hematoma volume was evaluated using a spectrophotometric hemoglobin assay 7 days after the ICH operation (Ma et al., 2014).

#### Western Blot Analysis

Total protein was extracted from the peri-hematoma area of the rat striatum using cell lysis buffer supplemented with proteinase and phosphatase inhibitors. Cell lysates were split by 10% SDS-PAGE and transferred to polyvinylidene fluoride membranes. Then, the membranes were blocked in 5% non-fat milk TBST buffer for 1.5 h at room temperature. The membranes were incubated in primary antibody overnight at 4◦C and in secondary antibody for 1 h at room temperature. Dilutions for primary antibodies were the following: anti-vascular endothelial growth factor (VEGFA; 1:1000, Abcam, Cambridge, MA, USA), antihypoxia inducing factor-1α (HIF-1α; 1:500, Abcam, Cambridge, MA, USA), and anti-AMPK (phospho-thr172; 1:500, Bioworld, Dublin, OH, USA). The secondary antibody was diluted 1:5000 (Sangon Biotech, Shanghai, Co., Ltd.). The density of bands was detected using an imaging densitometer (Bio-Rad, Foster City, CA, USA), and the gray value of bands was quantified using Quantity One 1-D analysis software (Bio-Rad).

#### qRT-PCR

Total RNA was removed with RNAiso Plus (TaKaRa Biotechnology, Dalian, China) using the manufacturer's instructions. Reverse transcription was done with a cDNA

FIGURE 1 | Effects of CTRP3 treatment on brain water content 7 days after ICH. (A) Brain water content in ipsilateral (Ipsi) and contralateral (Contra) hemispheres 7 days after ICH in rats treated with rCTRP3. (B) Brain water content in ipsilateral (Ipsi) and contralateral (Contra) hemispheres 7 days after ICH in rats treated with Lenti-CTRP3. Values are mean ± SEM. n = 8 per group. <sup>∗</sup>p < 0.05 vs. sham; #p < 0.05 vs. vehicle or Lenti.null. CTRP3, C1q/tumor necrosis factor-related protein-3; rCTRP3, recombinant CTRP3; Lenti.CTRP3, Lentivirus overexpression of CTRP3; ICH, intracerebral hemorrhage.

#### TABLE 1 | Primers used in qRT-PCR.


synthesis kit (TaKaRa Biotechnology). Real-time PCR reactions were performed with TaKaRa SYBR Premix Ex Taq II (TliRnaseH Plus, TaKaRa Biotechnology) on a PCR amplifier (CFX-96 Content Real-time System). Primers (Sangon Biotech) are recorded in **Table 1**.

#### Immunohistochemistry

Rats were killed 7 days after ICH induction by intraperitoneal injection of chloral hydrate. Immunohistochemistry was performed as previously described (Lei et al., 2013; Lapi et al., 2015). The primary antibody used was rabbit anti-rat CD31 (1:100, Abcam, USA). Total vessel densities were computed by counting 4 areas in 3 sections through the stroke region for each animal. Sections were stained with antibody against new vessel marker CD31; positive staining appeared brown. Standard quantitation was done as percent CD31-positive in the region bordering the hematoma.

FIGURE 2 | Effects of CTRP3 treatment on neurological functions 7 days after ICH. All animals after ICH showed significant neurological deficits based on performance on the modified Garcia (A), beam balance (B) and wire hanging (C) tests. Rats treated with rCTRP3 showed reduced neurological deficits in all three tests. Treatment with Lenti-CTRP3 starting 2 weeks before induction of ICH showed a tendency to ameliorate neurological deficits (D–F). Values are mean ± SEM. n = 9 per group. <sup>∗</sup>p < 0.05 vs. sham; ∗∗p < 0.01 vs. sham; #p < 0.05 vs. vehicle or Lenti.null. CTRP3, C1q/tumor necrosis factor-related protein-3; rCTRP3, recombinant CTRP3; Lenti-CTRP3, Lentivirus overexpression of CTRP3; ICH, intracerebral hemorrhage.

#### Statistical Analysis

All data are given as mean ± S.E.M. One-way analysis of variance (ANOVA) followed by Student's t test was utilized to collate outcomes among all groups. The SPSS 17.0 software package was utilized to do all statistics. P < 0.05 was considered statistically significant.

#### RESULTS

#### CTRP3 Reduced Brain Edema and Improved Functional Outcomes after ICH

Quantification of brain water content showed that ICH rats had significantly greater edema in the ipsilateral hemisphere than sham-operated rats (**Figure 1A**). Brain edema in the ipsilateral hemisphere was significantly less in rCTRP3 treated rats than in vehicle-treated rats (**Figure 1A**). In the contralateral hemisphere, rCTRP3 failed to affect brain water content (**Figure 1A**). Subsequently, we tested functional outcomes using a battery of behavioral tests in rats treated with rCTRP3 or vehicle. Neurological deficits were significantly more severe in all ICH vs. sham animals 7 days after ICH as tested by the modified Garcia test (**Figure 2A**), wire hanging test (**Figure 2B**), and beam balance test (**Figure 2C**). A statistically significant advancement was seen in all three neurobehavioral tests after rCTRP3 treatment.

Similarly, Lenti.CTRP3 treatment decreased brain water content in ICH rats (**Figure 1B**). Statistically significant neurological shortfalls were observed in all ICH vs. sham animals 7 days after ICH. Treatment with Lenti-CTRP3 tended to ameliorate neurological deficits 7 days after ICH (**Figures 2D–F**). Both the rCTRP3 and Lenti.CTRP3 treatments were effective.

FIGURE 3 | Effects of CTRP3 treatment on ICH-induced blood-brain barrier (BBB) damage 7 days after ICH. (A) Quantification of Evans blue dye extravasation (blue staining) in the ipsilateral (Ipsi) and contralateral (Contra) hemispheres 7 days after ICH in rats treated with rCTRP3. (B) Quantification of Evans blue dye extravasation (blue staining) in ipsilateral (Ipsi) and contralateral (Contra) hemispheres 7 days after ICH in rats treated with Lenti-CTRP3. Values are mean ± SEM. n = 6 per group. ∗∗p < 0.01 vs. sham; #p < 0.05 vs. vehicle or Lenti.null. CTRP3, C1q/tumor necrosis factor-related protein-3; rCTRP3, recombinant CTRP3; Lenti-CTRP3, Lentivirus overexpression of CTRP3; ICH, intracerebral hemorrhage.

## CTRP3 Reduced Disruption of the BBB

We measured BBB permeability by Evans blue extravasation in ICH rats. Significant accumulation of Evans Blue stain was seen in the ipsilateral hemisphere of ICH in comparison to sham animals (**Figure 3A**). Treatment with rCTRP3 significantly reduced the quantity of stain in the ipsilateral hemisphere in comparison to the vehicle 7 days after ICH (**Figure 3A**). Rats were similarly protected when treatment with Lenti-CTRP3 also caused a significant reduction in BBB permeability (**Figure 3B**).

### Effect of CTRP3 on the Hematoma Volume-Hemoglobin Assay

Neither recombinant nor lentivirus CTRP3 treatment had any obvious effect on hemorrhage volume at 7 days (data not shown).

### CTRP3 Promotes Angiogenesis and Activates the AMPK/HIF-1α/VEGF Axis in the ICH Brain

Angiogenesis is a vital part in beginning brain recovery after ICH. Having shown CTRP3 as an innovative adipokine involved in recovery following ICH, we assessed angiogenic effects of CTRP3. As illustrated in **Figures 4A,B**, CTRP3 treatment significantly enhanced the amount of CD31 positive capillary vessels in the zone bordering the hematoma 7 days after ICH, pointing to the observation that CTRP3 encourages vessel formation following ICH.

To further figure out the signaling pathways in charge of increased angiogenesis after CTRP3 treatment, many critical mediators and cytokines required for angiogenesis were tested. rCTRP3 significantly enhanced AMPK phosphorylation and increased HIF-1α and VEGF expression (**Figures 5A–E**). Similar changes in mediator and cytokine expression were found when animals were treated with lentivirus CTRP3 (**Figures 5F–J**). The results from qRT-PCR were in agreement with those from Western blots (**Figure 6**). These outcomes imply that CTRP3 promotes angiogenesis after ICH possibly through the AMPK/HIF-1α/VEGF axis.

#### Inhibiting AMKP or VEGF Activation Attenuated Effects of Recombinant CTRP3

To determine whether the AMPK/HIF-1α/VEGF signaling pathway is accountable for CTRP3 protection seen in vivo, Com.C (AMPK axis inhibitor) or CBO-P11 (VEGF inhibitor) was given together with rCTRP3 via the right lateral cerebral ventricle, and neurological function, BBB integrity and capillary density were assessed 7 days after ICH. As shown in **Figures 7A–D**, blocking AMPK activation abolished rCTRP3-induced pAMPK, HIF-1α and VEGF upregulation, and blocking VEGF activation only abolished rCTRP3-induced VEGF upregulation. In addition, CTRP3-induced capillary formation was entirely terminated when either AMPK or VEGF was inhibited (**Figures 7G,H**). But, the neurological

FIGURE 6 | Quantitative RT-PCR analysis of HIF-1α and VEGF-A mRNA levels in the hematoma border zone 7 days after ICH. (A–C) Quantitative RT-PCR analysis of the effect of rCTRP3 on CTRP3, HIF-1α and VEGF-A mRNA levels. (D–F) Quantitative RT-PCR analysis of the effect of rCTRP3 on CTRP3, HIF-1α and VEGF-A mRNA levels. Values are mean ± SEM. n = 4–6 per group. <sup>∗</sup>p < 0.05 vs. sham; ∗∗p < 0.01 vs. sham; ##p < 0.01 vs. vehicle or Lenti.null. CTRP3, C1q/tumor necrosis factor-related protein-3; rCTRP3, recombinant CTRP3; Lenti.CTRP3, Lentivirus overexpression of CTRP3; ICH, intracerebral hemorrhage.

protective aspect of CTRP3 was completely blocked by Com.C and partially blocked by CBO-P11 (**Figures 7E,F**). These results imply that, though the neurological protective effect of CTRP3 is controlled significantly by AMPK/HIF-1α/VEGF signaling-induced angiogenesis, other mechanisms also provide neurological protection to CTRP3 against ICH.

## DISCUSSION

In the present study, we investigated the ability of CTRP3 to ameliorate secondary brain injury after ICH in rats. Based on our knowledge, this is the first report that investigated the effects of CTRP3 therapy in angiogenesis and brain injury after ICH. The main discoveries of the study are as follows: (1) exogenous rCTRP3 or Lenti-CTRP3 in ICH rats demonstrates the same tendency to attenuate ICH-induced brain injury; (2) CTRP3 promotes focal angiogenesis and attenuates ICH-induced brain edema and breakdown of the BBB; and (3) CTRP3 may exert its angiogenic effect through AMPK/HIF-1α/VEGF signaling. These findings imply that CTRP3 is a novel

angiogenic factor that might perform a key part in encouraging angiogenesis by activating the AMPK signaling pathway during ICH.

The prognosis of ICH is affected by multiple factors (Brown et al., 1996). Neurotrophin family, anti-oxidative mediators, anti-mitochondrial impairment or anti-inflammatory drugs contribute to functional recovery and promote neuronal survival in the central nervous system (Ip et al., 1993; Chung et al., 2013; Xu et al., 2013; Chen et al., 2015; Wei et al., 2015). Angiogenesis induced in the ischemic penumbra (Risau, 1997) as well as rapid recovery of reperfusion and oxygen supply in injured brain tissues are critical prognostic factors (Mayer et al., 1998). In this study, we provide new evidence that CTRP3 has strong angiogenic and neuroprotective aspects, implying that CTRP3 could be an innovative therapeutic target of ICH.

It is known that compensatory angiogenesis can happen in the peri-hematoma region after ICH. Angiogenesis is a stepwise procedure that includes an increase in vascular permeability, degradation of the surrounding matrix, proliferation and migration of endothelial cells, and stabilization of freshly created microvessels (Conway et al., 2001). Concerted actions of many angiogenic molecules are necessary in this procedure, and VEGF is the most vital factor during each step of angiogenesis (Rosenstein et al., 1998; Yancopoulos et al., 2000). A number of animal experimental studies have shown that VEGF and fibroblast growth factor treatment encourages angiogenesis with ideal efficacy and increases capillary numbers (Lavu et al., 2011; Ye, 2016). However, endogenous angiogenesis following a stroke is insufficient to reverse brain injury. Our study shows that CTRP3 successfully encourages angiogenesis and upregulates VEGF expression in the striatum ipsilateral to the hemorrhage, which leads to increased vessel density. Because CTRP3 has been established to directly encourage endothelial cell proliferation and migration but not increase tube formation (a procedure that requires complexity surpassing proliferation and migration), angiogenic factors other than VEGF are likely involved.

These findings contribute to the increasing literature on the vital part of VEGF in brain injury. In fact, VEGF levels are increased during a plethora of pathological events in the brain, implicating its essential role in brain repair processes (Cristofaro and Emanueli, 2009). VEGF binds to two receptors, VEGF receptor-1 (VEGFR-1) and VEGF receptor-2 (VEGFR-2), through which it encourages revascularization and the mending of the BBB and re-establishes metabolic and trophic assistance to injured tissue (Krum et al., 2008; Shimotake et al., 2010). Future work in this field should aim to illuminate whether CTRP3 interacts with one or both receptors during ICH.

Because ICH-induced stress can change the composition, structure and distribution of the extracellular matrix, which has a vital role in creating normal brain tissue structures and is closely linked with brain injury-induced brain edema formation (Keep et al., 2012; Chung et al., 2013; Turner and Sharp, 2016), we investigated the effect of CTRP3 on BBB integrity by measuring brain water content and extravasation of Evans blue dye. We showed that angiogenesis via CTRP3 treatment resulted in preservation of the BBB. This observation in ICH extends past discoveries from other models of brain injury. In a model of cerebral ischemia, VEGF bound to VEGFR-2, which is expressed predominantly on activated astrocytes in the central nervous system, and encouraged revascularization and repair of the BBB by giving metabolic and trophic assistance to injured tissue (Krum et al., 2008; Shimotake et al., 2010; Hao et al., 2011). In a rat model of stroke, VEGF enhanced angiogenesis in the ischemic brain and reduced neurological deficits during recovery (Zhang et al., 2000). But, we were not able to demonstrate any effect of CTRP3 treatment on hematoma volume. We speculate that this may be due to use of a bloodinduced ICH model.

It is important to note that the CTRP3-treated rats manifested not only greatly strengthened angiogenesis and enhanced BBB preservation but also diminished neurological deficits. There are several possibilities for this improvement. First, the great angiogenic effect of CTRP3 in the border zone may help in restoring the blood flow, thereby rescuing dying neurons around the hematoma. Second, angiogenesis may assist new neurons migrating to damaged brain regions and give trophic material to these cells (Kojima et al., 2010; Lei et al., 2013). Finally, CTRP3 may promote differentiation of neural stem cells into neurons. Such intriguing possibilities all merit express study.

Around the hematoma, ischemia and hypoxia are apparent in ICH and this stress activates the AMPK and HIF1 signaling pathway, which induces the AMPK phosphorylation in human umbilical vein endothelial cells and promotes the recruitment, migration, proliferation and differentiation of endothelial cells (Nagata et al., 2003). In ischemic mice hindlimbs, AICAR, an AMPK antagonist induces expression of endogenous HIF target gene VEGF, but dominant-negative AMPK abolishes this expression at both steady state mRNA and protein levels (Ouchi et al., 2005). These data suggest that AMPK signaling is likely to regulate the expression of VEGF and promote angiogenesis in response to ischemic injury.

CTRP3 is a paralog of adiponectin, and it is well received that adiponectin encourages angiogenesis through activation of AMPK signaling (Shimano et al., 2010). Yet, whether CTRP3 encourages angiogenesis in cerebral tissue through the same pathway is not yet known. More important, inhibiting AMPK phosphorylation by Com.C greatly eradicated CTRP3-induced HIF-1α and VEGF expression and blocked the angiogenic effect of CTRP3. These findings defend the existence of an AMPKdependent mechanism for angiogenic effects of CTRP3.

# CONCLUSION

We conclude that CTRP3, a key member of a newly identified adipokine family, upregulates expression of angiogenic cytokines and induces robust angiogenesis, which led to enhanced preservation of the BBB and reduction of neurological deficits after ICH. The effect is mediated by the AMPK signaling pathway. These findings suggest that CTRP3 plays a positive role during ICH and has therapeutic potential.

# AUTHOR CONTRIBUTIONS

SW, YaZ, YB, LL and YoZ: conceived and designed the experiments. SW, YaZ, YB and SY: conducted the experiments. SW and YaZ: analyzed the results. SW, YC and JZ: contributed materials and analysis tools. SW and YaZ: wrote the article. SW and YaZ: contributed equally to this study. All authors reviewed the manuscript.

## FUNDING

This work was supported by The National Natural Science Foundation of China (81271460 and 81671158), Natural Science Youth Foundation of China (No. 81301125), the Medical scientific research projects of Chongqing (20120221) and the Natural Science Foundation of Chongqing Science and Technology Committee, China (No. cstc2015jcyjA10048).

# ACKNOWLEDGMENTS

We are grateful to Key Laboratory of Neurobiology (Chongqing, China) for technical guidance and partial sponsorship.

304, 243–248. doi: 10.1007/s11010-007-9506-6 Brown, R. D., Whisnant, J. P., Sicks, J. D., O'Fallon, W. M., and Wiebers, D. O. (1996). Stroke incidence, prevalence and survival: secular trends in Rochester,

Ahima, R. S., Qi, Y., Singhal, N. S., Jackson, M. B., and Scherer, P. E. (2006). Brain adipocytokine action and metabolic regulation. Diabetes 55, S145–S154. doi: 10.

Akiyama, H., Furukawa, S., Wakisaka, S., and Maeda, T. (2007). CTRP3/cartducin promotes proliferation and migration of endothelial cells. Mol. Cell. Biochem.


REFERENCES

2337/db06-s018


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Wang, Zhou, Bo, Li, Yu, Chen, Zhu and Zhao. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Intracellular Fibroblast Growth Factor 14: Emerging Risk Factor for Brain Disorders

Jessica Di Re1,2 , Paul A. Wadsworth<sup>3</sup> and Fernanda Laezza2,4,5 \*

<sup>1</sup>Neuroscience Graduate Program, University of Texas Medical Branch, Galveston, TX, USA, <sup>2</sup>Department of Pharmacology and Toxicology, University of Texas Medical Branch, Galveston, TX, USA, <sup>3</sup>Biochemistry and Molecular Biology Graduate Program, The University of Texas Medical Branch, Galveston, TX, USA, <sup>4</sup>Mitchell Center for Neurodegenerative Diseases, The University of Texas Medical Branch, Galveston, TX, USA, <sup>5</sup>Center for Addiction Research, The University of Texas Medical Branch, Galveston, TX, USA

The finely tuned regulation of neuronal firing relies on the integrity of ion channel macromolecular complexes. Minimal disturbances of these tightly regulated networks can lead to persistent maladaptive plasticity of brain circuitry. The intracellular fibroblast growth factor 14 (FGF14) belongs to the nexus of proteins interacting with voltagegated Na+ (Nav) channels at the axonal initial segment. Through isoform-specific interactions with the intracellular C-terminal tail of neuronal Na<sup>v</sup> channels (Nav1.1, Nav1.2, Nav1.6), FGF14 controls channel gating, axonal targeting and phosphorylation in neurons effecting excitability. FGF14 has been also involved in synaptic transmission, plasticity and neurogenesis in the cortico-mesolimbic circuit with cognitive and affective behavioral outcomes. In translational studies, interest in FGF14 continues to rise with a growing list of associative links to diseases of the cognitive and affective domains such as neurodegeneration, depression, anxiety, addictive behaviors and recently schizophrenia, suggesting its role as a converging node in the etiology of complex brain disorders. Yet, a full understanding of FGF14 function in neurons is far from being complete and likely to involve other functions unrelated to the direct regulation of Na<sup>v</sup> channels. The goal of this Mini Review article is to provide a summary of studies on the emerging role of FGF14 in complex brain disorders.

Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Alexander Dityatev, German Center for Neurodegenerative Diseases (DZNE), Germany Elsa Fabbretti, University of Nova Gorica, Slovenia

\*Correspondence:

Fernanda Laezza felaezza@utmb.edu

Received: 21 December 2016 Accepted: 28 March 2017 Published: 19 April 2017

#### Citation:

Di Re J, Wadsworth PA and Laezza F (2017) Intracellular Fibroblast Growth Factor 14: Emerging Risk Factor for Brain Disorders. Front. Cell. Neurosci. 11:103. doi: 10.3389/fncel.2017.00103 Keywords: neuronal excitability, intracellular signaling, protein-protein interactions, biological psychiatry

# INTRODUCTION

In 2014, nearly one in four adults in the United States was diagnosed with a mental illness (National Institutes of Mental Health<sup>1</sup> ). Treatment for many of these illnesses is hampered by limited efficacy of the medications and patient noncompliance due to intolerable side effects. To address this need, the National Institutes of Mental Health has launched an initiative to research these illnesses from all levels, ranging from genomic to behavioral. Dubbed the Research Domain Criteria (RDoC), this initiative proposes to complement top-down understanding of these diseases, beginning with human behavior, with bottom-up research by understanding the molecular and cellular causes of these disorders.

<sup>1</sup>https://www.nimh.nih.gov/health/statistics/index.shtml

Di Re et al. FGF14 in Brain Disorders

Because many neuropsychiatric disorders are associated with maladaptive plasticity and excitability, one area of importance within neurons is the axon initial segment (AIS), which serves as the action potential initiation site (Palmer and Stuart, 2006). This highly complex subcellular region contains a nexus of scaffolding and regulatory proteins that ensure proper targeting, clustering and function of the ion channels underlying the action potential (Ogawa and Rasband, 2008; Hsu et al., 2014). Neuropsychiatric disorders have been associated with many of these proteins, including ankyrin-G, α-spectrins, β-spectrins, neurofascin, contactin and intracellular fibroblast growth factors (iFGFs; Hsu et al., 2014). One such protein, fibroblast growth factor 14 (FGF14, also known as fibroblast homologous factor 4 or FHF4), is an iFGF that binds to voltage-gated Na<sup>+</sup> (Nav) channels and promotes their localization to the proximal region of the axon, providing the fine-tuned regulation necessary for normal functioning (Lou et al., 2005; Laezza et al., 2007, 2009; Goetz et al., 2009; Wang et al., 2011, 2012; Wildburger et al., 2015; Ali et al., 2016; Bosch et al., 2016; Hsu et al., 2016; Pablo et al., 2016). Loss of functional FGF14 may change the biophysical properties of Na<sup>v</sup> channels or alter their localization to the AIS, leading to changes in neuronal excitability (Goldfarb et al., 2007; Shakkottai et al., 2009; Bosch et al., 2015; Hsu et al., 2016). Recent findings also show that FGF14 regulates the function of voltage-gated K<sup>+</sup> and Ca2<sup>+</sup> channels, however none of these interactions are direct, and therefore might represent a different type of regulation from what has been described for Na<sup>v</sup> channels (Yan et al., 2013; Pablo and Pitt, 2017).

Initially cloned on the basis of sequence similarity with other FGF members, FGF14 was first associated to a human disease with the F145S mutation causing spinocerebellar ataxia 27 (SCA27), a naturally occurring complex neurodegenerative disorder characterized by onset of ataxia in early adulthood and deficits in cognition, memory and behavior (Smallwood et al., 1996; van Swieten et al., 2003; Brusse et al., 2006). Genetic deletion of FGF14 in mice recapitulates some of these symptoms at the behavior and circuitry level (Wang et al., 2002; Wozniak et al., 2007).

Since then, FGF14 has been indicated by several linkage and genome wide association studies (GWAS) to be a putative risk factor for other neuropsychiatric diseases including depression, addiction and schizophrenia, as well as neurodegenerative diseases, such as Alzheimer's Disease (Detera-Wadleigh et al., 1999; Park et al., 2004; Mulle et al., 2005; Need et al., 2009; Johnson et al., 2011; Verbeek et al., 2012; Singh and Rajeswari, 2015; Yang et al., 2015). These recent associations clearly indicate that the role of this gene in the CNS is yet to be fully understood. In the next paragraphs, we will summarize some of the most recent studies on FGF14 in animal models and human tissue.

#### HISTORICAL PROSPECTIVE: FGF14 AS VOLTAGE-GATED Na<sup>+</sup> (Nav) CHANNEL INTERACTING PROTEIN

While iFGFs share a conserved core β-trefoil region with other FGFs their functions and distributions are distinct from canonical FGFs (Itoh and Ornitz, 2008). Canonical FGFs are normally secreted to activate FGF receptors on the cell surface, however iFGFs lack a secretory sequence, fail to activate or antagonize FGF receptors and are primarily found in the cytoplasm, nucleus or the AIS (Smallwood et al., 1996; Olsen et al., 2003; Ornitz and Itoh, 2015; Pablo et al., 2016). Initial discoveries using yeast-two-hybrid screening identified

FIGURE 1 | Known functions of amino acids in fibroblast 14 (FGF14). (A) Alignment of wild-type FGF14-1a and FGF14-1b protein sequences with highlighted residues representing surface hot-spots for FGF14:Nav1.6 interactions (green, Ali et al., 2016), casein kinase II (CK-2) phosphorylation sites (yellow, Hsu et al., 2016), and the spinocerebellar ataxia 27 (SCA27) F145S mutation site (red, Laezza et al., 2007). (B) An FGF14 homology model generated using the FGF13:Nav1.5 crystal structure (Protein Data Bank ID: 4DCK) as template, visualized using the visual molecular dynamic (VMD) software package (Humphrey et al., 1996). Surface hot-spots important for protein:protein interactions (green) and the SCA27 F145S mutation (red) are shown.

Di Re et al. FGF14 in Brain Disorders

FGF12 and FGF13 as direct interactors of Na<sup>v</sup> channels (Liu et al., 2003; Wittmack et al., 2004; Rush et al., 2006) and subsequent studies resulted in similar discoveries for the two isoforms of FGF14 (Lou et al., 2005; Laezza et al., 2007, 2009), aligned in the illustration of **Figure 1A**. To date, the evidence for direct interaction and functional modulation of Na<sup>v</sup> channels by FGF14 ranges from crystal structure to biochemical in cell assays to animal models and includes recent identification of critical amino acid residues at the FGF14: Nav1.6 channel complex illustrated in **Figure 1B** (Lou et al., 2005; Laezza et al., 2007, 2009; Goetz et al., 2009; Ali et al., 2014, 2016; Hsu et al., 2016; Pablo et al., 2016).

The N-terminus of FGF14 is alternatively spliced into two isoforms: FGF14-1a and FGF14-1b. FGF14-1a shares sequence homology with FGF12-1a and FGF13-1a, while the amino terminus of FGF14-1b contains a unique 69 amino-acid sequence and is the more prevalent isoform in the CNS (Munoz-Sanjuan et al., 2000). Importantly, the interaction of FGF14 in cells that heterologously express individual Na<sup>v</sup> isoforms shows that FGF14 is unique, as it results in very distinct isoform-specific Na+ current phenotypes that are dictated by the two FGF14 splice variants (Lou et al., 2005; Laezza et al., 2009; Ali et al., 2016). In primary hippocampal neurons, overexpression of FGF14-1b increases Na+ current density, causes a hyperpolarizing shift in the voltage-dependence of activation and a depolarizing shift in the voltage-dependence of inactivation, while the F145S, SCA27 dominant negative loss-of-function mutation causes opposite phenotypes, possibly by disrupting the function of wild-type FGF14 (Laezza et al., 2007). Neurons from Fgf14−/<sup>−</sup> mice exhibit impaired excitability in both the hippocampus and cerebellum (Goldfarb et al., 2007; Shakkottai et al., 2009; Bosch et al., 2015; Hsu et al., 2016). The effect of FGF14 on Na<sup>v</sup> channels and excitability is summarized in **Table 1**.

## FGF14 AS SCAFFOLD FOR KINASES

Recent studies have added a new dimension to FGF14, showing that its interaction with the Na<sup>v</sup> channel is controlled by selective kinases (Shavkunov et al., 2012, 2013; Hsu et al., 2015, 2016). Initial studies using the luciferase complementation assay demonstrated that the FGF14:Nav1.6 complex formation is controlled by glycogen synthase kinase 3 (GSK; Shavkunov et al., 2013) and more recently by the GSK3 priming kinase, casein kinase II (CK2) which phosphorylates FGF14 at S228 and S230 (Hsu et al., 2016; **Figure 1B**). Inhibition of either GSK3 or CK2 is sufficient to disrupt the FGF14:Na<sup>v</sup> channel complex formation with consequences for targeting of the two proteins to the AIS and for intrinsic excitability (Shavkunov et al., 2013; Hsu et al., 2016). It is possible that phosphorylation at these kinase specific sites that confers functional specificity to FGF14 contributing to regulation of other ion channels (i.e., voltage-gated K<sup>+</sup> and Ca2<sup>+</sup> channels).

More is known about the specific phosphorylation of FGF14 by GSK3 and CK2, however other kinases have been shown to affect FGF14:Nav1.6 interactions (Shavkunov et al., 2012; Hsu et al., 2015). Importantly, many


kinases involved in tyrosine receptor kinase signaling are implicated in this regulation, including the mitogen activated protein kinase (MAPK), C-Jun N-terminal kinase (JNK; Hsu et al., 2015). JNK signaling is disrupted in insulin resistance associated with type-II diabetes and Alzheimer's disease (Najem et al., 2016). Functional enrichment of single nucleotide polymorphisms (SNPs) in patients with type-II diabetes and Alzheimer's disease shows that FGF14 is significantly overrepresented in these two diseases because of its phosphorylation by JNK (Hao et al., 2015). Changes in the mRNA expression of MAPK/JNK signaling proteins, including FGF14, are also significantly overrepresented in early-onset Alzheimer's disease patients (Antonell et al., 2013). Taken together, these results indicate that the interaction between JNK and FGF14 might be an important area for future research in Alzheimer's disease.

#### FGF14 AS REGULATOR OF EXCITATORY AND INHIBITORY SYNAPTIC TRANSMISSION

Numerous are the reports of the effect of genetic deletion of Fgf14 on synaptic transmission. Studies in the cerebellum of Fgf14−/<sup>−</sup> mice revealed decreased excitatory transmission from granule cells to Purkinje cells (parallel fibers, PF), a phenotype that is accompanied by reduced AMPA receptor-mediated excitatory postsynaptic currents and decreased expression of vesicular glutamate transporter 1, a specific presynaptic marker at PF-Purkinje neuron synapses (Tempia et al., 2015). Presynaptic changes in neurotransmitter release have been also reported at the Schaffer's collaterals to CA1 synapses where deletion of Fgf14 results in reduction in the readyreleasable pool of presynaptic glutamate and diminished expression of synaptobrevin, synaptophysin, syntaxin I (Xiao et al., 2007). Other changes in presynaptic function have been reported at inhibitory GABAergic terminals onto CA1 pyramidal cells of Fgf14−/<sup>−</sup> mice, which exhibit reduced expression of glutamic acid decarboxylase 67 (GAD67) and vesicular GABA transporter (vGAT), presumably deriving from fast-spiking parvalbumin (PV) interneurons synapses (Alshammari T. K. et al., 2016). Additional studies in the same animal model identified selective loss of PV interneurons, reduced γ frequency oscillations and deficits in working memory (Alshammari T. K. et al., 2016). Collectively, these results recapitulate some endophenotypes of schizophrenia and are supported by human studies finding significant reduction and co-variation of FGF14, PV, vGAT and GAD67 in post-mortem samples from schizophrenic patients compared to healthy control individuals (Alshammari T. K. et al., 2016). Whether all these changes at presynaptic glutamatergic and GABAergic terminals result from neuroadaptive responses to impaired firing or represent disruption of a separate function of the FGF14 protein remains to be determined. However, the evidence for genetic links between vGAT, GAD67 and FGF14 might argue for a ''separate function''

# FGF14 IS REQUIRED FOR SYNAPTIC PLASTICITY

Studies have also supported a role of FGF14 in synaptic plasticity in the hippocampus. Fgf14−/<sup>−</sup> mice show impaired long-term potentiation (LTP) at the Schaffer's collaterals to CA1 synapses, which is accompanied by decreased expression of synaptic vesicles docked at the active zone, and fewer miniature excitatory postsynaptic currents in primary hippocampal neurons (Xiao et al., 2007). Short-term plasticity is also impaired at these Fgf14−/<sup>−</sup> terminals, at which repetitive stimuli causes significant synaptic fatigue, consistent with impaired presynaptic function (Xiao et al., 2007).

#### FGF14 AS FACTOR REQUIRED FOR NEUROGENESIS

Adult neurogenesis, or the proliferation, differentiation and integration of new neurons into existing brain circuitry has become an area of research interest in part due to its implication in the cognitive pathophysiology of several neuropsychiatric disorders, including Alzheimer's disease, depression and schizophrenia (Ming and Song, 2005; Taupin, 2005, 2008; Reif et al., 2007; Johnson et al., 2009; Sun et al., 2011; Jun et al., 2012; Walton et al., 2012; Ouchi et al., 2013). It was recently found that FGF14 is required for the maturation of progenitor cells in the dentate gyrus of the hippocampus. Fgf14−/<sup>−</sup> mice show impaired transition from late immature neuronal progenitor cells to mature neurons, which is accompanied by reduced paired-pulse facilitation at the perforant path to granule neurons in the dentate gyrus (Alshammari M. A. et al., 2016). Overall, deletion of FGF14 results in an immature dentate gyrus, an endophenotype that corroborates a link between the gene and schizophrenia (Hagihara et al., 2013).

TABLE 2 | Single nucleotide polymorphisms (SNPs) in FGF14 introns associated with neuropsychiatric disease.


#### FGF14 AS AN ASSOCIATED FACTOR FOR NEUROPSYCHIATRIC DISEASE

As many neuropsychiatric disorders are heterogeneous and complex, GWAS have become an important tool for sorting relevant genetic information from large patient populations. Numerous GWAS have reported SNPs in FGF14 in the context of neuropsychiatric disorders (**Table 2**). Although all these SNPs are in the FGF14 intronic region and thus their role on the protein expression and function are unclear, they might provide guidance for future investigations. A Brazilian pilot study on early onset/familial schizophrenia found a link between earlyonset schizophrenia and FGF14 (Gadelha et al., 2012). GWAS in German cohort found an association between FGF14 and schizophrenia, which is corroborated by a linkage study of familial schizophrenia in Canadian families of Celtic or German descent (Brzustowicz et al., 1999; Need et al., 2009). Additionally, SNPs in FGF14 have been associated with dependence on alcohol and illegal substances in humans, and a fine-mapping study found several SNPs to be associated with major depressive disorder in a study of Dutch twins (Drgon et al., 2011; Johnson et al., 2011; Verbeek et al., 2012). Furthermore, an FGF14 SNP is associated with volumetric changes in the entorhinal cortex in AD patients (Yang et al., 2015). Overall, genetic variations in FGF14 are linked to the pathophysiology of several neuropsychiatric disorders, a promising area for further research that is supported by studies in Fgf14−/<sup>−</sup> preclinical models

# REFERENCES


(Alshammari T. K. et al., 2016; Alshammari M. A. et al., 2016).

# CONCLUSION

FGF14 plays a role in all fundamental properties of neurons: intrinsic firing, synaptic transmission of excitatory and inhibitory neurons and plasticity, while deletion of the gene leads to disruptive motor and cognitive behaviors. The role of FGF14 in humans is yet to be fully understood, but the emerging technologies for genome sequencing and protein characterization will provide potential opportunities for identifying new disease signatures associated with FGF14.

#### AUTHOR CONTRIBUTIONS

All of the authors have contributed substantially to the work. JDR and FL contributed to writing and editing the manuscript. PAW created the image and legend for **Figure 1**.

# FUNDING

This research was funded by National Institute of Mental Health: grant no. 1R01MH111107-01A1, R01 MH095995-A1 and the National Institutes of Health: grant no 5T32AG051131-02.


in a mouse model of spinocerebellar ataxia type 27. Front. Cell. Neurosci. 9:205. doi: 10.3389/fncel.2015.00205


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Di Re, Wadsworth and Laezza. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Early Origin and Evolution of the Angelman Syndrome Ubiquitin Ligase Gene Ube3a

#### Masaaki Sato1,2 \*

<sup>1</sup>Graduate School of Science and Engineering and Brain and Body System Science Institute, Saitama University, Saitama, Japan, <sup>2</sup>RIKEN Brain Science Institute, Wako, Japan

The human Ube3a gene encodes an E3 ubiquitin ligase and exhibits brain-specific genomic imprinting. Genetic abnormalities that affect the maternal copy of this gene cause the neurodevelopmental disorder Angelman syndrome (AS), which is characterized by severe mental retardation, speech impairment, seizure, ataxia and some unique behavioral phenotypes. In this review article, I highlight the evolution of the Ube3a gene and its imprinting to provide evolutionary insights into AS. Recent comparative genomic studies have revealed that Ube3a is most phylogenetically similar to HECTD2 among the human HECT (homologous to the E6AP carboxyl terminus) family of E3 ubiquitin ligases, and its distant evolutionary origin can be traced to common ancestors of fungi and animals. Moreover, a gene more similar to Ube3a than HECTD2 is found in a range of eukaryotes from amoebozoans to basal metazoans, but is lost in later lineages. Unlike in mice and humans, Ube3a expression is biallelic in birds, monotremes, marsupials and insects. The imprinting domain that governs maternal expression of Ube3a was formed from non-imprinted elements following multiple chromosomal rearrangements after diversification of marsupials and placental mammals. Hence, the evolutionary origins of Ube3a date from long before the emergence of the nervous system, although its imprinted expression was acquired relatively recently. These observations suggest that exogenous expression and functional analyses of ancient Ube3a orthologs in mammalian neurons will facilitate the evolutionary understanding of AS.

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Daniela Tropea, Trinity College, Dublin, Ireland Robert Weissert, University of Regensburg, Germany Edgar Richard Kramer, Medical Research Council Harwell (MRC), UK

#### \*Correspondence:

Masaaki Sato masasato@mail.saitama-u.ac.jp; msato@brain.riken.jp

Received: 13 December 2016 Accepted: 22 February 2017 Published: 07 March 2017

#### Citation:

Sato M (2017) Early Origin and Evolution of the Angelman Syndrome Ubiquitin Ligase Gene Ube3a. Front. Cell. Neurosci. 11:62. doi: 10.3389/fncel.2017.00062 Keywords: brain evolution, genomic imprinting, developmental disorder, autism, intellectual disability, HECT domain, synapse

#### INTRODUCTION

The evolution of the brain in mammals is characterized by dramatic increases in size and complexity, especially in the neocortex (Striedter, 2005). Previous advances in comparative genomics have revealed emerging principles of the genetic basis of brain evolution (Khaitovich et al., 2006; Vallender et al., 2008). Changes in protein-coding sequences and regulatory elements as well as emergence of new genes and loss of existing ones likely had profound phenotypic impacts on brain development and ultimately led to significant alterations in brain structure and function. Thus, understanding how genes that play essential roles in human brain development and cognition evolved is of great importance and interest.

The ubiquitin ligase gene Ube3a (also known as E6-associated protein; E6AP) provides an excellent model for studies of gene evolution because of its brain-specific imprinting and implication in the neurodevelopmental disorder Angelman syndrome (AS). Ube3a is a homologous to the E6AP carboxyl terminus (HECT) domain-containing E3 ubiquitin ligase that was initially discovered as the protein involved in human papillomavirus E6-mediated p53 degradation (Huibregtse et al., 1993). It is expressed monoallelically from the maternal allele in the brain in a parent-of-origin specific manner (Albrecht et al., 1997; Rougeulle et al., 1997; Vu and Hoffman, 1997). The imprinting of Ube3a and its neighboring genes is coordinated by a regulatory region known as the Prader-Willi syndrome (PWS)-AS imprinting center (IC), which is located upstream of the adjacent SNURF (SNRPN upstream reading frame)–SNRPN (small nuclear ribonucleoprotein-associated protein N) gene on the human 15q11-q13 chromosome region (Buiting et al., 1999; Ohta et al., 1999; Perk et al., 2002). Genetic abnormalities that affect the maternal copy of Ube3a are known to cause AS, which is characterized by a wide variety of symptoms such as severe mental retardation, speech impairment, seizure, ataxia and unique behavioral phenotypes such as frequent laughter (Angelman, 1965; Williams et al., 1995; Kishino et al., 1997; Matsuura et al., 1997; Clayton-Smith and Laan, 2003; Mabb et al., 2011; Buiting et al., 2016), whereas duplication or increased expression of this gene is linked to autism spectrum disorders (Bolton et al., 2001; Glessner et al., 2009; Smith et al., 2011; Urraca et al., 2013). Accordingly, Ube3a is essential for neural circuit maturation and experience-dependent plasticity in the mammalian cerebral cortex (Yashiro et al., 2009; Sato and Stryker, 2010). In this review article, I highlight the evolution of the Ube3a gene and its imprinting to gain evolutionary insights into AS.

#### ANCIENT ORIGIN OF THE Ube3a GENE

Ube3a contains a single HECT domain at the C-terminal and no discernible functional domain at its N-terminal side. Recent analyses revealed that the Ube3a protein is phylogenetically closest to HECTD2 among 28 human HECT domain-containing ubiquitin ligases, and a group of proteins called small HERCs (HERC3-6) that possess a single N-terminal RCC1-like domain (RLD) and a C-terminal HECT domain are also similar to Ube3a (Marín, 2010; Grau-Bové et al., 2013; Scheffner and Kumar, 2014; **Figure 1A**). HECTD2 and small HERCs are expressed in the brain, although their functions are not well understood (Sánchez-Tena et al., 2016). HECTD2 has been associated with susceptibility to neurological diseases (Lloyd et al., 2009a,b). Two other HECT E3 ligases, Ube3b and Ube3c, were named after Ube3a (Gong et al., 2003), but carry a calmodulin-binding IQ domain in addition to a HECT domain and are categorized as a distinct class of HECT E3 ligases (**Figure 1A**). Notably, Ube3b has been strongly implicated in the human developmental disorder blepharophimosis-ptosis-intellectual-disability syndrome (Basel-Vanagaite et al., 2012).

Orthologs of Ube3a are commonly found in bilaterians including vertebrates and Drosophila (Reiter et al., 2006; Wu et al., 2008; Hope et al., 2016) but not in Caenorhabditis elegans, implying that Ube3a was lost in some nematode lineages (Marín, 2010). The evolutionary origins of Ube3a can be traced to the genomes of basal metazoans, including the cnidarian Nematostella vectensis, the sponge Amphimedon queenslandica, and the placozoan Trichoplax adherens (Marín, 2010), and to fungi such as Mortierella verticillata (Grau-Bové et al., 2013; **Figures 1B,C**). However, no Ube3a orthologs are present in choanoflagellates, suggesting secondary loss in this lineage (Grau-Bové et al., 2013). Notably, the E3 ligase gene HECTX is more similar to Ube3a than HECTD2 and is found in the genomes of amoebozoans, fungi, choanoflagellates and early metazoans, but not in those of bilaterians (Marín, 2010; Grau-Bové et al., 2013; **Figures 1A–C**). These findings suggest that an ancient E3 ligase gene that was more similar to Ube3a than the extant HECTD2 was secondarily lost in the bilaterian lineage.

The evolution of Ube3a substrate specificity remains unclear. Several neuronal proteins have been identified to date as possible direct ubiquitination targets of Ube3a (Sell and Margolis, 2015; Sun et al., 2015). Although the precise modes of these protein interactions have not been characterized, specific substrate recognition by Ube3a is thought to be mediated by its non-catalytic N-terminal region (Cooper et al., 2004; Scheffner and Kumar, 2014). Whereas Ube3a orthologs share the conserved HECT domain at their C-termini, their N-terminal regions are more variable among lineages, suggesting that ancient Ube3a orthologs recognized differing sets of target proteins to those ubiquitinated by the present human Ube3a. New substrate specificity was likely acquired during evolution by changes in substrate binding regions and encounters of Ube3a with potential novel substrates, the latter of which were probably brought by changes in expression and subcellular localization, and the emergence of new proteins.

#### Ube3a AND THE EVOLUTION OF THE NERVOUS SYSTEM

The early origin of Ube3a indicates that it predates the origins of nerve cells and synapses. The expression and function of the Ube3a protein in primitive organisms are currently unclear. Whether the nervous system evolved from single or multiple independent origins remains controversial (Miller, 2009; Ryan et al., 2013; Moroz et al., 2014; Liebeskind et al., 2016). At the base of the metazoan tree, sponges and placozoans lack nerve and muscle cells, but exhibit coordinated behaviors such as feeding and contraction (Ellwanger et al., 2007; Smith et al., 2015). On the other hand, cnidarians and ctenophores have diffuse nervous systems called nerve nets, which communicate by synapses (Anderson and Spencer, 1989; Tamm and Tamm, 1995; Marlow et al., 2009). Centralized nervous systems evolved in the bilaterian lineage (Arendt et al., 2016).

Searches for orthologs of specific postsynaptic density (PSD) proteins demonstrate that the genomes of nerve-less basal metazoans and unicellular choanoflagellates contain core sets

of scaffold protein orthologs, and these are co-expressed in a distinct cell type of Amphimedon larvae (Sakarya et al., 2007; Alié and Manuel, 2010). Shank postsynaptic scaffold proteins have been implicated in autism spectrum disorders in humans (Durand et al., 2007; Berkel et al., 2010; Sato et al., 2012) and are also found in the choanoflagellate genome (Alié and Manuel, 2010), providing another remarkable example of the ancient origins of genes that are involved in human developmental disorders. More recent studies showed that human PSD proteins that are essential for basic cellular processes, such as amino acid biosynthesis and energy generation, are conserved between prokaryotes and eukaryotes, whereas the majority of structural and signaling molecules, including those involved in ubiquitination, are specific to eukaryotes (Emes and Grant, 2011). The ancient eukaryotic origin of Ube3a is thus consistent with the early origins of human postsynaptic proteins, many of which are linked to neurogenetic disorders (Bayés et al., 2011).

#### ASSEMBLY OF THE PWS-AS IMPRINTED DOMAIN

Ube3a expression is imprinted in the brain but not in peripheral tissues in humans and mice (Albrecht et al., 1997; Rougeulle et al., 1997; Vu and Hoffman, 1997). Moreover, Ube3a expression is imprinted in neurons but not in glial cells of the brain (Yamasaki et al., 2003; Judson et al., 2014). Interestingly, imprinting of Ube3a is not fully established in the postnatal mouse brain and paternal Ube3a expression decreases as neurons mature (Sato and Stryker, 2010; Judson et al., 2014). Hence, imprinted expression of Ube3a is tissue- and cell type-specific and is developmentally regulated.

Genomic imprinting, or parent-of-origin specific epigenetic gene silencing, is widespread in placental mammals and also occurs in marsupials, suggesting evolution from common ancestors of marsupials and eutherians (Renfree et al., 2009). Genome-wide characterization of imprinted genes revealed parent-of-origin allelic effects in over 1300 loci in embryonic and adult mouse brains (Gregg et al., 2010). However, hypotheses regarding the origins and evolutionary advantages of genomic imprinting are few. Among these, the host defense hypothesis proposes that genomic imprinting evolved from the cellular mechanisms that mediate methylation and silencing of foreign DNA elements (Barlow, 1993). Alternatively, the kinship theory suggests fitness advantages of genomic imprinting. Specifically, paternally and maternally expressed genes have been shown to increase and decrease the transfer of maternal nutrients to the fetus during pregnancy, respectively, as observed for the paternally expressed IGF2 growth factor and the maternally expressed IGF2 receptor (IGF2R) growth repressor (Haig, 2004). Another hypothesis, the coadaptation theory, proposes that genomic imprinting coordinates placental and hypothalamic functions of the fetus and mother to optimize growth, postnatal suckling and maternal care, as demonstrated by the paternal Peg3 transcription factor that is expressed in these tissues (Li et al., 1999; Curley et al., 2004).

A comparative genomic study revealed an unexpected picture of the assembly of the PWS-AS imprinted domain during evolution (Rapkins et al., 2006). In human chromosome 15q and homologous mouse chromosome 7C regions, Ube3a is located downstream of SNURF–SNRPN, which forms a bicistronic transcript and is expressed from the paternal allele. Maternal expression of Ube3a and paternal expression of SNURF–SNRPN are controlled by the IC that lies upstream of SNURF–SNRPN (**Figure 2A**). This arrangement is conserved in eutherians including mice and humans, but is not present in marsupials such as the gray shorttailed opossum Monodelphis domestica and other animals of greater evolutionary age. In these animals, the gene CNGA3 is present downstream of Ube3a instead of SNURF–SNRPN (**Figure 2A**). Accordingly, expression of Ube3a is biallelic in the marsupial tammar wallaby, the montreme platypus, and in chickens and Drosophila (Colosi et al., 2006; Rapkins et al., 2006; Hope et al., 2016). The searches for the marsupial ortholog of SNRPN revealed that it resides beside the closely related SNRPB gene in the Monodelphis domestica genome (**Figure 2B**). Furthermore, the genomes of evolutionarily older animals including monotremes have SNRPB but no SNRPN orthologs. These findings suggest that SNRPN was formed by tandem duplication of the evolutionally older SNRPB gene in marsupials.

The PWS-AS imprinted domain encompasses a genomic region of about 2 Mb in humans and comprises a smaller AS subdomain that includes two maternally expressed genes (Ube3a and ATP10A) and a larger PWS subdomain that contains six paternally expressed genes (MKRN3, MAGEL2, NDN, NPAP1, SNURF and SNRPN) and two major clusters of the paternally expressed small nucleolar RNAs (snoRNAs) SNORD115 (HBII-52) and SNORD116 (HBII-85). Numbers of snoRNA genes in this region are highly variable across eutherian lineages (Zhang et al., 2014). In addition, the paternally expressed NPAP1 is primate specific and lacks an ortholog in other eutherians (Neumann et al., 2014), suggesting that this imprinted domain is a highly dynamic genomic region. Currently, it is held that the PWS-AS imprinted domain formed from non-imprinted components by genome rearrangement in an eutherian ancestor after divergence from marsupials. The following ordered or concurrent steps have been proposed (Hore et al., 2007; Renfree et al., 2009): (1) fission of Ube3a-CNGA3 border; (2) translocation of SNRPN next to Ube3a; (3) generation of SNURF and IC; (4) insertion and expansion of snoRNA repeats; (5) insertion of the three retroposed genes MKRN3, MAGEL2, and NDN, followed by the integration of NPAP1 in the primate lineage. A few key questions remain unsolved regarding the assembly of the PWS-AS imprinted domain. In particular, it is unclear why Ube3a was fused to SNRPN and became a part of the PWS-AS imprinting domain, and no marsupial progenitors of SNURF and IC have yet been identified (Renfree et al., 2009).

# PSEUDOGENES OF Ube3a

The two processed pseudogenes Ube3ap1 and Ube3ap2 have been identified in the human genome, although there is no evidence of their expression (Kishino and Wagstaff, 1998). These pseudogenes are also found in chimpanzees but not in mice and macaques, indicating that they formed in a common ancestor of chimpanzees and humans. Ube3ap1 and Ube3ap2 are located on chromosome 2 and 21, respectively, in the human and chimpanzee genomes.

#### CONCLUSION AND PERSPECTIVE

Ube3a is an ancient gene that emerged prior to nervous systems, and its imprinted expression was acquired much later (**Figure 2C**). These findings accord with the current view that genes involved in human neurogenetic disorders are not necessarily evolutionarily new. To deepen the understanding of the evolution of Ube3a, comparison with the evolution of genes involved in brain disorders other than neurodevelopmental disorders, such as neurological disorders, is instructive. For example, presenilins (encoded by PSEN1 and PSEN2 in humans) form the catalytic center of γ-secretase that processes amyloid precursor protein (APP) to produce amyloid-β (Aβ) peptide, and mutations in PSEN1, PSEN2 and APP are found in early-onset familial Alzheimer's disease (Bertram et al., 2010). Presenilin orthologs are widespread among eukaryotes, including amoebozoans, metazoans and plants, suggesting that their ancestral gene was already present in the last common eukaryotic ancestor (Gazave et al., 2009). On the other hand, orthologs of the APP gene family (amyloid precursor-like protein 1 (APLP1), amyloid precursor-like protein 2 (APLP2), and APP) have been identified only in multicellular metazoans, including Nematostella vectensis, and the amyloidogenic Aβ motif and γ-secretase cleavage sites are conserved only across APP orthologs from jawed vertebrates (Tharp and Sarkar, 2013; Moore et al., 2014). Although phylogenetic studies of the proposed Ube3a substrates are yet to be conducted and searches for additional candidates of AS-relevant substrates should be continued, the evolutions of Ube3a and presenilin suggest that the ancient emergence of diseaserelated enzymes and more recent appearance of their relevant substrates could be a common evolutionary scheme of the key signal transduction components across different brain disorders.

Recent studies suggest that diverse symptoms of AS are mediated by distinct circuits, cell types, substrates and downstream pathways that act at different developmental stages (Mandel-Brehm et al., 2015; Silva-Santos et al., 2015; Judson et al., 2016). From an evolutionary point of view, it can be suggested that the key events in the evolutionary history of Ube3a led to the current etiology of AS. These

SNRPN from SNRPB in the opossum genome is enclosed by a red dashed box. (C) Early origin and evolution of Ube3a in relation to the evolution of the nervous

system. LECA, last eukaryotic common ancestor.

events likely include: (1) expression of the Ube3a protein in nerve cells and its localization at functionally important subcellular compartments such as synapses; (2) colocalization and interaction with substrates that play essential roles in neuronal development and function; and (3) acquisition of genomic imprinting, leading to increased vulnerability of Ube3a to genetic damage. Thus, further studies of the expression and localization of Ube3a orthologs in primitive extant organisms, and exogenous expression and functional analyses of these orthologs in mammalian neurons, will broaden the evolutionary perspective of AS, as described for a few other synaptic proteins (Burkhardt et al., 2014; Yang et al., 2015).

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

MS wrote the manuscript.

#### FUNDING

This work was supported in part by a grant from the Angelman Syndrome Foundation.

#### ACKNOWLEDGMENTS

I thank Toru Takumi and Junichi Nakai for support and Takeshi Kawashima for helpful discussion.


human papillomavirus E6 oncoprotein with p53. Mol. Cell. Biol. 13, 775–784. doi: 10.1128/mcb.13.2.775


**Conflict of Interest Statement**: The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer DT and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Sato. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Genetic Determinants of Parkinson's Disease: Can They Help to Stratify the Patients Based on the Underlying Molecular Defect?

Sara Redenšek <sup>1</sup> , Maja Trošt <sup>2</sup> and Vita Dolžan<sup>1</sup> \*

<sup>1</sup> Pharmacogenetics Laboratory, Institute of Biochemistry, Faculty of Medicine, University of Ljubljana, Ljubljana, Slovenia, <sup>2</sup> Department of Neurology, University Medical Centre Ljubljana, Ljubljana, Slovenia

Parkinson's disease (PD) is a sporadic progressive neurodegenerative brain disorder with a relatively strong genetic background. We have reviewed the current literature about the genetic factors that could be indicative of pathophysiological pathways of PD and their applications in everyday clinical practice. Information on novel risk genes is coming from several genome-wide association studies (GWASs) and their meta-analyses. GWASs that have been performed so far enabled the identification of 24 loci as PD risk factors. These loci take part in numerous cellular processes that may contribute to PD pathology: protein aggregation, protein, and membrane trafficking, lysosomal autophagy, immune response, synaptic function, endocytosis, inflammation, and metabolic pathways are among the most important ones. The identified single nucleotide polymorphisms are usually located in the non-coding regions and their functionality remains to be determined, although they presumably influence gene expression. It is important to be aware of a very low contribution of a single genetic risk factor to PD development; therefore, novel prognostic indices need to account for the cumulative nature of genetic risk factors. A better understanding of PD pathophysiology and its genetic background will help to elucidate the underlying pathological processes. Such knowledge may help physicians to recognize subjects with the highest risk for the development of PD, and provide an opportunity for the identification of novel potential targets for neuroprotective treatment. Moreover, it may enable stratification of the PD patients according to their genetic fingerprint to properly personalize their treatment as well as supportive measures.

Keywords: Parkinson's disease, genetic risk factors, genome-wide association studies, single nucleotide polymorphisms, personalized medicine

# INTRODUCTION

Parkinson's disease (PD) is the second most common neurodegenerative brain disease after Alzheimer's disease. It is believed that both genetic and environmental factors contribute to the development of PD. It is a chronic, progressive and incurable disease (Pringsheim et al., 2014). Two pathological hallmarks characterize PD: the formation of cytoplasmic inclusions termed Lewy bodies and Lewy neurites and the loss of dopaminergic neurons in the substantia nigra pars compacta (Gallegos et al., 2015). The disease manifests as bradykinesia, muscular rigidity, rest

#### Edited by:

Daniela Tropea, Trinity College, Ireland

Reviewed by: Robert Petersen, Case Western Reserve University, USA Anelyssa D'Abreu, Brown University, USA

> \*Correspondence: Vita Dolžan vita.dolzan@mf.uni-lj.si

Received: 26 October 2016 Accepted: 25 January 2017 Published: 10 February 2017

#### Citation:

Redenšek S, Trošt M and Dolžan V (2017) Genetic Determinants of Parkinson's Disease: Can They Help to Stratify the Patients Based on the Underlying Molecular Defect? Front. Aging Neurosci. 9:20. doi: 10.3389/fnagi.2017.00020 tremor and postural and gait impairment (Postuma et al., 2015), which can be accompanied by several other nonmotor symptoms (Gallegos et al., 2015). At the present time, there is unfortunately no cure for PD except for the symptomatic treatment and supportive measures (Connolly and Lang, 2014).

Two different forms of PD have been identified, i.e., sporadic and familial. The familial form starts at an earlier stage of life (<50 years), is usually more severe and progresses more quickly (Klein and Westenberger, 2012). At early stages of research on genetics of PD, only linkage studies were performed, which means that families with PD history were included in the study. These studies identified mutations in six genes that conclusively cause monogenic PD—SNCA, LRRK2, Parkin, PINK1, DJ-1, ATP13A2 (Klein and Westenberger, 2012).

In this review, we focus mainly on the sporadic form of the disease. However, there is some overlap between genes associated with familial and sporadic disease; in particular, SNCA and LRRK2 are involved in both forms (Verstraeten et al., 2015; van der Brug et al., 2015). Recently, genome-wide association studies (GWASs) that compared single nucleotide polymorphisms' (SNPs) frequencies between sporadic PD patients and healthy individuals have identified several loci as Parkinson's disease susceptibility genes (Klein and Westenberger, 2012; Nalls et al., 2014). A few of these loci, 24 in particular, harboring SNPs that were found to be associated with PD were validated in the replication phase of the latest and largest meta-analysis of GWASs performed by Nalls et al. in 2014. In the discovery phase, they performed a meta-analysis of GWASs including 13,708 cases and 95,282 controls chosen from the populations of European descent (the USA, France, Germany, Iceland, the Netherlands, the UK; 23 and Me), while the replication phase of the study included 5,353 cases and 5,551 controls (Nalls et al., 2014). The identified loci segregate with numerous cellular pathways that may contribute to PD pathology: protein aggregation, protein, and membrane trafficking, lysosomal autophagy, immune response, synaptic function, endocytosis, inflammation, and metabolic pathways being among the most important ones (Kumaran and Cookson, 2015). Each locus identified as a risk factor has a rather low contribution to PD development; therefore, a combination of molecular defects rather than a single event probably plays a role in PD risk, hence the idea of a cumulative nature of genetic risk factors (Pihlstrom et al., 2016).

This review summarizes the latest knowledge on genetics and genomics of PD susceptibility, obtained by GWASs and their meta-analyses. We focus on the largest meta-analysis of GWASs on PD risk so far (Nalls et al., 2014), but we also searched the PubMed database and GWAS catalog (Welter et al., 2014) for studies that pointed out the same loci as the above-mentioned meta-analysis. We identified 13 GWASs and their meta-analyses (summarized in **Table 1**), which we included in this review. We compiled the available data on the 24 loci that were found to be associated with the risk of sporadic form of PD. To obtain the information about gene functions, we searched the PubMed database and collected the available data on genes' functions with the help of the following words: "Parkinson's disease and gene name" or "Parkinson's disease and polymorphisms and gene name." We divided the susceptibility genes into seven groups according to their physiological functions: protein aggregation; protein, and membrane trafficking; lysosomal autophagy; immune system; neurodevelopment, neuron cell differentiation, and survival; mitochondrial homeostasis; and genes involved in other processes. This review highlights the main functions of these genes' products and their role in the PD pathogenesis.

#### GENETIC DETERMINANTS OF RISK FOR SPORADIC PD AND THEIR IMPLICATED PATHWAYS

#### Protein Aggregation

The products of the genes listed below, i.e., SNCA and tau, are both constituents of the protein aggregates typical of PD called Lewy bodies. PD is often thought to be a prion-like disease because of the presence of these bodies (Hasegawa et al., 2016).

#### SNCA

SNCA codes for α-synuclein (SNCA), which is a small, acidic protein of 14.5 kDA and 140 amino acids (Gallegos et al., 2015). It is the main component of aggregates called Lewy bodies, a hallmark of PD pathology. Lewy bodies are formed because a mutated protein usually adopts the β-sheet structure, which is harder to degrade than α-helixes, the main conformation of native proteins (Gallegos et al., 2015; Inoshita and Imai, 2015). Different variants of the gene have been connected to both familial and sporadic forms of PD. Several studies on families with positive PD history have reported the association of SNCA with PD, the first one being published in 1997 by Polymeropoulos et al. (1997). The association with sporadic PD was first described by Kruger et al. who compared the allelic frequences of REP1 polymorphism in the promoter region of SNCA between cases and controls (Kruger et al., 1999). This association was later confirmed on a larger set of data (Maraganore et al., 2006). The majority of GWASs performed also confirmed SNCA as a susceptibility gene (Satake et al., 2009; Simon-Sanchez et al., 2009; Edwards et al., 2010; Hamza et al., 2010; Do et al., 2011; Nalls et al., 2011, 2014; Saad et al., 2011; Spencer et al., 2011; Lill et al., 2012; Pankratz et al., 2012; Hill-Burns et al., 2014). To date, over 800 SNPs within SNCA have been reported, with nearly half of them showing a positive association with PD (Kumaran and Cookson, 2015). SNPs within SNCA recognized as PD susceptibility factors by GWASs are listed in **Table 2**. The most replicated SNP rs256220 was confirmed as a risk factor in six studies, whereas several others were found in only one or two studies. Some SNCA mutations can cause both sporadic and familial forms of PD. These are usually more penetrant (Singleton et al., 2013). Besides point mutations—Ala53Thr, Ala30Pro, and Glu46Lys in the amino-terminal sequence, whole locus multiplications (duplications and triplications) were also found in both forms (Bisaglia et al., 2009).

SNCA is widely expressed in the central nervous system, especially in the presynaptic terminals of neurons (Inoshita and Imai, 2015). It has two main physiological roles, the first being

#### TABLE 1 | GWAS studies and their meta-analyses that identified 24 risk loci included in this review.


\*country of recruitment, if specified.

#meta-analysis.

Data compiled from the GWAS catalog database and PDgene (Lill et al., 2012; Nalls et al., 2014; Welter et al., 2014).

the control of synaptic membrane processes and the second being the control of neurotransmitter release via interactions with members of the SNARE family (Bellucci et al., 2012; Tsigelny et al., 2012). It promotes SNARE-complex assembly through a non-enzymatic mechanism, binding to phospholipids via its N-terminal and to synaptobrevin-2 via its C-terminal (Burre et al., 2010). SNCA is deeply involved in the synaptic vesicle cycle, including trafficking, docking, fusion, and recycling after exocytosis. It was also suggested that SNCA is a negative regulator of neurotransmitter release, including dopamine, with traffic restriction of synaptic vesicles from the resting pool to the sites of release (Emanuele and Chieregatti, 2015).

Wild-type SNCA is translocated to the lysosomes for degradation via chaperone-mediated autophagy via the lysosomal membrane receptor LAMP2A (Gan-Or et al., 2015). Lysosomal enzyme β-glucocerebrosidase (GBA) then modulates SNCA levels (Klein and Westenberger, 2012). Mutant SNCA somehow inhibits chaperone-mediated autophagy, presumably by attaching to the LAMP2A and preventing its internalization of wild-type and mutant SNCA. Recently, it has been shown that GBA and SNCA form a positive feedback loop that leads to the accumulation of SNCA. The loss of GBA function results in the accumulation of SNCA, whereas SNCA inhibits the lysosomal activity of GBA (Mazzulli et al., 2011).

SNCA is presumably also involved in dopamine release and in the synaptic vesicle dynamics, especially in the recycling pathway. The excess of SNCA reduces the recycling of synaptic vesicles and their motility (Inoshita and Imai, 2015). Furthermore, it has been confirmed that dopamine can interact with the SNCA molecule via the C-terminal residues which can induce and/or modulate its structure and oligomerization. As a result dopamine blocks chaperone-mediated autophagy of SNCA. This may explain why dopaminergic neurons are more prone to SNCA accumulation (Bellucci et al., 2012; Lashuel et al., 2013; Gan-Or et al., 2015).

SNCA also modulates dopamine vesicle trafficking by other mechanisms. According to Ahn et al. SNCA binds and consequently inhibits phospholipase D2 (PLD2), which is involved in vesicle trafficking in terms of endo- and exocytosis (Ahn et al., 2002). SNCA's interacting partner is also actin which is essential for synaptic vesicle mobilization between different functional pools. SNCA binding modulates actin's polymerization in a manner dependent on Ca <sup>2</sup><sup>+</sup> concentration (Bellani et al., 2010). Intracellular Ca2<sup>+</sup> concentration may also change due to the formation of pore-like structures in the cell membranes by mutant SNCA (Lashuel et al., 2002). Furthermore, SNCA's interacting partner is also a rate-limiting enzyme in the dopamine synthesis, tyrosine hydroxylase (TH), responsible for the conversion of tyrosine to L-3,4-dihydroxyphenylalanine



Ch, Chromosome. Data compiled from the GWAS catalog database, the dbSNP and the e!Ensembl (Sherry et al., 2001; Welter et al., 2014; Yates et al., 2016). \*Denotes SNPs in 3'-UTR.

(L-DOPA). This interaction inhibits the phosphorylation of TH and thus also its action (Venda et al., 2010).

SNCA is also linked to neuroinflammation. In mice with aggregated SNCA, MHCII expression is induced in microglia. Moreover, SNCA as a danger-associated molecular pattern (DAMP) stimulates toll-like receptors (TLR). Misfolded and fibrillar SNCA activates microglia via TLR2 and TLR4 and activation of NLRP3 inflammasome (Gustot et al., 2015). These processes contribute to dopaminergic neuronal death (Dzamko et al., 2015; Kumaran and Cookson, 2015).

In addition, SNCA mutations can lead to dysfunctions in several other cellular pathways, such as mitochondrial function, oxidative stress, cytochrome c release, ER stress, apoptosis (Gallegos et al., 2015).

#### MAPT

MAPT encodes a microtubule-associated protein tau, which has a role in stabilizing microtubules in the neurons. The gene was first recognized as a PD risk factor by Pastor et al. (2000). It has been repeatedly confirmed as a risk factor for sporadic PD by GWASs and their meta-analyses (Simon-Sanchez et al., 2009; Edwards et al., 2010; Hamza et al., 2010; Do et al., 2011; Nalls et al., 2011, 2014; Spencer et al., 2011; Lill et al., 2012; Hill-Burns et al., 2014; Vacic et al., 2014), but each variant has only been pointed out once and has not been replicated in other studies. More than 40 mutations have been discovered to date. They are either coding mutations, that affect the binding activity of the tau or its susceptibility to aggregation, or mutations that alter the splicing of exon 10 which encodes the fourth microtubule binding repeat (4R tau isoforms) (Wray and Lewis, 2010). Mutations within MAPT locus recognized as PD susceptibility factors by GWASs are listed in **Table 2**. The MAPT locus is the longest region of linkage disequilibrium in Caucasians. There are two variants of the locus resulting from an inversion of a 900 kb long portion of the sequence, which leads to two haplotypes— H1 and H2. MAPT H1 haplotype increases the transcription of the gene, whereas MAPT H2 haplotype decreases it (Golpich et al., 2015). The inversion is present in the H2 haplotype, which has a 20% frequency in the Caucasians, but is almost not present in the Asian population (Wider et al., 2010; Schulte and Gasser, 2011; Soto-Ortolaza and Ross, 2016). MAPT association was not replicated in the Japanese population probably due to inter-population heterogeneity at this locus (Satake et al., 2009; Labbé and Ross, 2014). H1 haplotype is more dynamic and can be subdivided into subhaplotypes. Homozygous MAPT H1/H1 genotype is a biomarker of dementia in PD (Lin and Wu, 2015). As H1 haplotype is associated with a greater risk of PD and other neurodegenerative diseases, it may be assumed that H2 haplotype is under positive selection (Schulte and Gasser, 2011). Polymorphisms in the MAPT gene are also involved in the pathologies of Alzheimer's disease (AD), progressive supranuclear palsy (PSP), and corticobasal degeneration (CBD) (Vandrovcova et al., 2009).

MAPT encoded protein tau is a potential component of Lewy bodies. In PD patients with a more pronounced cognitive decline, neurofibrillary tangles, more typical of AD, can also be found besides Lewy bodies. These tangles consist of hyperphosphorylated tau (Horvath et al., 2013; Lin and Wu, 2015). Physiologically, phosphorylation of the tau protein regulates its propensity to interact with microtubules (Wray and Lewis, 2010). According to some studies, SNCA also promotes tau fibrillization and influences tau phosphorylation (Giasson et al., 2003). The interplay between the two genes was also confirmed by the study of Goris et al., reporting nearly a double risk of PD in carriers of the combined MAPT H1/H1 and SNCA rs356219 G/G genotype (Goris et al., 2007). Most of the pathological changes in the MAPT sequence disrupt the ability of tau to interact with microtubules (Wray and Lewis, 2010). Tau is also fundamental for the maintenance of cytoskeletal network via the regulation of axonal transport by interactions with kinesin and dynein (Kumaran and Cookson, 2015).

Mutated MAPT may also affect the function of lysosomes and its role in autophagy, since it is degraded with this process (Gan-Or et al., 2015).

MAPT gene lies in a block of nearly complete linkage disequilibrium that extends over nearly 2 Mb, therefore it is possible that other genes from this region are also associated with PD risk (Charlesworth et al., 2012). For example, a variant in the CRHR1 gene coding for the corticotropin-releasing hormone receptor has recently been recognized as a factor that decreases the PD susceptibility (Davis et al., 2016).

#### Protein and Membrane Trafficking

SNPs within loci associated with protein and membrane trafficking recognized as PD susceptibility factors by GWASs are listed in **Table 3**. The genes within this group are involved in processes, such as clathrin-mediated vesicular transport, trafficking, and fusing of synaptic vesicles, clearance of Golgiderived vesicles, endo- and exocytosis and protein sorting.

#### TMEM175/GAK/DGKQ

A haplotype block TMEM175/GAK/DGKQ (transmembrane protein 175/cyclin G associated kinase/theta diacylglicerol kinase) was first associated with sporadic PD in the GWAS performed by Hamza et al. after it was already confirmed to be associated with the familial form of the disease (Pankratz et al., 2009; Hamza et al., 2010). This locus was reevaluated several times by different research groups and also showed positive results in other GWASs and meta-analyses of GWASs. The most replicated SNP is an intron variant rs11248060, others were found in individual studies only (Edwards et al., 2010; Do et al., 2011; Nalls et al., 2011, 2014; Lill et al., 2012; Pankratz et al., 2012).

GAK is a particularly promising candidate for a risk gene, because it is differentially expressed in the substantia nigra pars compacta of PD patients as compared to controls (Pankratz et al., 2009). GAK gene product participates in multiple steps of clathrin-mediated vesicular transport. For example, together with the heat-shock cognate protein 70 (Hsc70), it promotes the uncoating of endocytosed clathrin-coated vesicles. Probably it has also some other functions related to its serine/threonine kinase domain. It also directly interacts with LRRK2 and participates in the clearance of Golgi-derived vesicles via the lysosomal autophagy pathway as a part of LRRK2/RAB7L1/GAK trans-Golgi complex (Kumaran and Cookson, 2015; Perrett et al., 2015). Recently, depletion of GAK has been shown to influence lysosomal sorting of cathepsin D, the main lysosomal enzyme involved in SNCA degradation (Latourelle et al., 2009). Knockdown of GAK in primary rat neurons increased SNCA toxicity (Perrett et al., 2015). GAK also takes part in cell cycle regulation (Labbé and Ross, 2014).

Less is known about DGKQ; however, it is expressed in the brain where it may be involved in the phosphatidylinositol and lipid signaling (Pankratz et al., 2009). DGK proteins are also known to affect Ca2<sup>+</sup> signaling as well as the trafficking and fusing of synaptic vesicles at nerve terminals in the central nervous system (Ran and Belin, 2014).

#### NUCKS1/RAB7L1

A haplotype block NUCKS1/RAB7L1 (nuclear casein kinase and cyclin-dependent kinase substrate 1/RAB7, member RAS oncogene family-like 1) is in linkage disequilibrium and is also known by the name PARK16 locus. This locus contains five genes—SLC45A3, NUCKS1, RAB7L1, SLC41A1, and PM20D1. The association between PD and NUCKS1 was first described in the GWAS conducted by Simon-Sanchez et al. (2009), whereas the association with RAB7L1 was first shown in Satake et al. GWAS (Satake et al., 2009). The implication of this locus in PD was then again confirmed in meta-analysis of GWASs by Nalls et al. and in the GWAS by Vacic et al. (Nalls et al., 2014; Vacic et al., 2014). Four SNPs were pointed out by genome-wide studies in this locus but none of them was replicated.

RAB7L1, also named RAB29, is a cytoplasmic GTP-binding protein, which plays an important role in endo- and exocytosis. Exocytosis is relevant in PD because dopamine is released via exocytotic vesicles and because it has been shown that vesicle abnormalities occur in SNCA knock-out mice (Plagnol et al., 2011). RAB7L1 is one of the LRRK2 interacting partners in the process of removal of Golgi derived vesicles by autophagy-dependent mechanisms. RAB7L1 deficiency caused neurodegeneration in mammalian or Drosophila dopaminergic neurons having a human LRRK2 mutation, while RAB7L1 overexpression restored the function of LRRK2 mutant neurons. MacLeod et al. showed that expressing mutant LRRK2 or reducing the expression of RAB7L1 in a cell line led to the loss of components of the retromer complex, which regulates protein sorting from the lysosome to the Golgi apparatus.


#### TABLE 3 | SNPs within loci associated with protein and membrane trafficking recognized as PD susceptibility factors by GWASs.

NA, data not available; Ch, Chromosome. Data compiled from the GWAS catalog database, the dbSNP and the e!Ensembl (Sherry et al., 2001; Welter et al., 2014; Yates et al., 2016). \*Denotes SNPs in 3'-UTR.

Secondary to these changes, the swelling of lysosomes and the reduction of lysosomal mannose 6-phosphate receptor (M6PR) accumulation occur (MacLeod et al., 2013; Perrett et al., 2015). M6PRs are needed for the recruitment and transport of lysosomal hydrolases to the lysosome; therefore, the disruption of this function may lead to lysosomal dysfunction (Gan-Or et al., 2015). RAB7L1 thus functions downstream of GAK in the retromer sorting process (Hoang, 2014). The authors also suggested an epistatic effect, with nonlinear increase in PD susceptibility when both risk alleles are present in the patient's DNA sequence (MacLeod et al., 2013).

Less is known about NUCKS1 and its involvement in PD pathogenesis. NUCKS1 is a nuclear protein containing several consensus phosphorylation sites for casein kinase II and cyclindependent kinases of unknown function (Satake et al., 2009). According to some studies, NUCKS1 might also be involved in inflammation and immunity (Dzamko et al., 2015). The recent GWAS meta-analysis has reported that rs823118 was tagged an expression quantitative trait locus and also a methylation quantitative trait locus for both RAB7L1 and NUCKS1 (Wang et al., 2016).

#### LRRK2

LRRK2 (leucine-rich repeat kinase 2) gene was first described as a risk factor for the familial form of PD, but was later also confirmed as sporadic PD susceptibility gene by GWASs. It encompasses rare highly penetrant variants as well as more common variants with a lower effect (van der Brug et al., 2015). LRRK2 mutations are the most common gene defect in sporadic PD (Perrett et al., 2015). The first GWAS that confirmed LRRK2 as a risk factor for sporadic PD was performed by Satake et al. (2009). Later, its association with PD was also confirmed in other GWASs and their meta-analyses (Do et al., 2011; Nalls et al., 2011, 2014; Lill et al., 2012; Vacic et al., 2014). Six different SNPs were found by GWASs as risk factors, but only one (rs1491942) was also replicated in a later study (Lill et al., 2012). LRRK2 accounts for a large part of risk for sporadic PD in some populations— Ashkenazi Jews, Arabs, East Asians. Coding variability at the LRRK2 locus explains 10% of PD risk in these populations. In the outbred Europeans, LRRK2 and GBA account for around 8% of risk. Carriers of LRRK2 mutations usually have Lewy bodies, the minority has also tangles, but some of them have neither (Hardy, 2010).

LRRK2 is a multidomain protein that has a GTPase and kinase function as well as many protein/protein interaction motifs. It is expressed in axons and dendrites in the striatum and the cortex, but expression is low in nigral cell bodies. LRRK2 seems to regulate actin complexes, vesicle trafficking, endosome maturation, cytoskeletal dynamics, and protein translation (Volta et al., 2015). It was detected as a binding partner of the late-endosomal marker Rab7 and also as a binding partner of the lysosomal marker LAMP2A, which generated the idea that it has an important function in the endo-lysosomal pathway. Mutant LRRK2-RAB7 colocalization can lead to a reduced RAB7 function and impaired late endosomal trafficking events. LRRK2 also interacts with RAB7L1 to modify protein sorting. It also binds the PD susceptibility factor GAK in a complex that promotes removal of Golgi derived vesicles by autophagydependent mechanisms (Perrett et al., 2015). It is tightly linked to endocytotic and exocytotic processes required for rapid synaptic vesicles and receptor recycling via interaction with clathrin (Volta et al., 2015). It regulates endophilin association to clathrin-coated vesicles through phosphorylation (Drouet and Lesage, 2014). Furthermore, it influences the action of EndoA via phosphorylation, a protein important for endocytic synaptic vesicles recycling (Inoshita and Imai, 2015). Moreover, LRRK2 is involved in mitochondrial membrane maintenance via fusion/fission processes and in lysosomal autophagy and recycling of V-ATPase required for lysosomal acidification (Ryan et al., 2015; Volta et al., 2015). The dysregulation of mitochondrial function can also be caused by the inhibition of the endogenous peroxidase phosphorylation by mutant LRRK2 (Saiki et al., 2012). LRRK2 also has a role in neuroinflammation by increasing the cytokine release from activated primary microglial cells, which results in neurotoxicitiy (Russo et al., 2014; Blesa et al., 2015). Moreover, it interferes with the translation machinery by phosphorylation of several proteins, for example translation repressor protein 4E-BP (Taymans et al., 2015).

Among PD patients having the LRRK2 mutation, the most common one is p.Gly2019Ser (rs34637584). The penetrance is high—50% at the age of 50 and 74% at the age of 79. This mutation is located in the kinase domain resulting in an increase in the activity (Wallings et al., 2015). 1.6% of sporadic PD patients possess this mutation (Satake et al., 2009). Mutated LRRK2 (p.Gly2019Ser) actually binds to outer mitochondrial membrane, which leads to a decrease in mitochondrial membrane potential and to a decrease in the intracellular ATP level. On the contrary, mitochondrial elongation and interconnectivity were elevated (Subramaniam and Chesselet, 2013).

#### INPP5F

Inositol polyphosphate-5-phosphatase F (INPP5F), one of the polyphosphoinositide phosphatases, was first described as a PD risk factor in a meta-analysis carried out by Nalls et al. in 2014 (Nalls et al., 2014), which means that only one SNP was found to be associated with PD susceptibility so far. The gene's product is supposed to be involved in the endocytic pathways. It contains a Sac domain, which is involved in the endocytosis of synaptic vesicles (Inoshita and Imai, 2015; Nakatsu et al., 2015). The knowledge about the gene's function is limited and on the basis of only one study pointing out the association, we cannot reach any conclusions as to the main role of the gene in PD pathology and the actual association.

#### BCKDK/STX1B

BCKDK/STX1B (branched chain ketoacid dehydrogenase kinase/syntaxin 1B) locus was established as a PD risk factor in GWAS meta-analysis by Nalls et al. Only one SNP (rs14235) is known to be associated with PD, thus replication studies are needed to confirm the actual association (Nalls et al., 2014).

BCKDK is located in the mitochondrial matrix and plays a major role in valine, leucine and isoleucine catabolism. Its function is phosphorylation and thus inactivation of the BCKD. BCKD concentration is the same in all tissues, whereas BCKDK concentration varies. BCKD's level of function is thus regulated by BCKDK (García-Cazorla et al., 2014).

STX1B codes for syntaxin 1B, which functions as a synaptic receptor for vesicle transport. It was previously shown to be directly implicated in the process of calcium-dependent synaptic transmission in rat brain, having been suggested to play a role in the excitatory pathway of synaptic transmission (Plagnol et al., 2011). It plays a role in dopamine neurotransmission (Kalia and Lang, 2015).

#### VPS13C

Vacuolar protein sorting 13 homolog C (VPS13C) was shown to be associated with PD as a risk factor only in the metaanalysis by Nalls et al. (2014). First, it was considered only as a risk factor for sporadic PD, but it has recently been discovered that it could also cause the familial autosomal-recessive earlyonset form of PD with a very severe course of the disease (Lesage et al., 2016). The gene has two variants, variant 2 being specific to brain tissue (Velayos-Baeza et al., 2004). It functions in the vesicular transport, more specifically endosomal transport pathway (Inoshita and Imai, 2015). Functional studies of the protein are lacking, but a very recent one also suggests a protein quality control function (Yang et al., 2016). More association and functional studies will have to be conducted to be completely confident of the gene's role in PD susceptibility and pathology.

#### Lysosomal Autophagy

SNPs within loci associated with lysosomal autophagy recognized as PD susceptibility factors by GWASs are listed in **Table 4**. These genes are playing their part in general lysosomal functions, especially lysosomal autophagy.

#### GBA/SYT11

GBA (beta acid glucosidase) is associated with PD susceptibility according to several GWASs and their meta-analyses, but the first association was found in 2011 by Do et al. (Do et al., 2011;


TABLE 4 | SNPs within loci associated with lysosomal autophagy recognized as PD susceptibility factors by GWASs.

Ch, Chromosome. Data compiled from the GWAS catalog database, the dbSNP and the e!Ensembl (Sherry et al., 2001; Welter et al., 2014; Yates et al., 2016).

Pankratz et al., 2012; Nalls et al., 2014; Vacic et al., 2014). On the other hand, SYT11 (synaptotagmin XI) was confirmed as a PD risk gene by Nalls et al. in 2011 (Nalls et al., 2011, 2014). No SNP has been replicated in a later study.

GBA is located within the inner lysosomal membrane, where it cleaves membrane glucocerebrosides into ceramide and glucose. Functional studies have shown that any kind of reduction in GBA enzymatic activity leads to the accumulation of SNCA which, in turn, inhibits the function of normal GBA, causing the additional aggregation of SNCA. The authors hypothesize that the reduction of GBA function, either caused by GBA mutations or impaired GBA trafficking from ER and Golgi to lysosomes, increases the glucocerebrosides concentration in lipid rafts. This change in membrane composition may lead to the reduced formation of LAMP2A (chaperone-mediated autophagy receptor) protein-complexes, which could result in diminished autophagy of SNCA and its accumulation. GBA dysfunction may also lead to general disruption in lysosomal function and autophagy. Furthermore, it can also be the reason for defective mitophagy and mitochondrial damage. It was also shown that PD patients with no mutation in the GBA may have a lower enzyme activity. This implies that other factors may also affect the GBA function, for example environmental factors (Gan-Or et al., 2015).

SYT11 encodes a calcium-sensing protein involved in membrane trafficking in synaptic transmision and is also a substrate of Parkin, which is a risk factor for the familial form of PD. Due to its ability of calcium-dependent phospholipid binding, it is important in the regulation of vesicle fusion and endocytosis at synaptic terminals (Ran and Belin, 2014). More studies will have to be performed in order to properly determine the function of the gene.

#### SCARB2/FAM47E

SCARB2/FAM47E (scavenger receptor class B member 2) locus was first found to be associated with PD in 2009 by Simon-Sanchez et al.; this was later confirmed by two other GWASs—Do et al. and Nalls et al. The same SNP (rs6812193) was pointed out in all three studies (Simon-Sanchez et al., 2009; Do et al., 2011; Nalls et al., 2014).

SCARB2 encodes the lysosomal membrane protein type 2 (LIMP-2), which is a GCase receptor. It directs GBA to lysosomes. The reduced function of SCARB2 can lead to a reduced GBA activity and decreased SNCA degradation (Gan-Or et al., 2015). The receptor, being an enterovirus 71 receptor, also plays a role in the neuroinflammation process (Dzamko et al., 2015). FAM47E lies in close proximity to SCARB2 and the most frequently associated SNP (rs6812193) is situated between the two genes. However, among the two, SCARB2 seems to be a more promising candidate for a PD risk factor (Do et al., 2011; Ran and Belin, 2014).

#### Immune System

SNPs within loci associated with the immune system recognized as PD susceptibility factors by GWASs are listed in **Table 5**. The genes' function in PD susceptibility and pathology should be tested in a more targeted way with the help of cell and animal models. There are some indications that genes mentioned below play a role in the immune system, but we do not know to what extent and if this is their main purpose, except for HLA-DQB1 whose function is quite well determined. There is also a need for more association studies especially for STK39, HLA-DQB1 and DDRGK1.

#### BST1

BST1 (bone marrow stromal cell antigen-1) was recognized as a PD risk factor in the Satake et al. GWAS in 2009 and after that in several other GWASs. Three intron variants were found to be associated with PD risk. Two of them (rs11724635 and rs4698412) were replicated in a later study (Satake et al., 2009; Nalls et al., 2011, 2014; Saad et al., 2011; Pankratz et al., 2012). BST1, a member of the CD38 gene family, is a cell surface protein bound to the membrane by glycosylphosphatidylinositol linkage; it possesses both ADP ribosyl cyclase and cyclic ADP ribose hydrolase enzymatic activities. It produces cyclic ADP-ribose, which as a second messenger releases calcium from intracellular



Ch, Chromosome. Data compiled from the GWAS catalog database, the dbSNP and the e!Ensembl (Sherry et al., 2001; Welter et al., 2014; Yates et al., 2016). \*Denotes SNPs in 3'-UTR.

ER stores, leading to calcium influx. Calcium is a great regulator of neurotransmitter release from presynaptic terminals; therefore, a disruption in neuronal calcium homeostasis could lead to selective death of dopaminergic neurons. Interestingly, BST1 seems to be an important factor in the immune system, since it is highly expressed in bone marrow cells in patients with rheumatoid arthritis and facilitates immature B-cell proliferation and growth (Chen et al., 2014; Ran and Belin, 2014).

#### STK39

STK39 (serine threonine kinase 39) is highly expressed in the brain and pancreas (Wang et al., 2014). It codes for a SPAK protein (Ste20-related proline/alanine rich-kinase) (Kumaran and Cookson, 2015). As a PD risk factor, it was first described by Nalls et al. in 2011 and then again confirmed three years later by the same group. Two different intergenic variants were pointed out in these studies (Nalls et al., 2011, 2014). It is a serine/threonine kinase, which is activated under cellular stress and plays a role in stress signals, ion homeostasis and inflammation (Li et al., 2013; Ran and Belin, 2014). It has been shown that its overexpression alters intestinal inflammatory levels in mouse models of colitis, whereas its knockout attenuates intestinal inflammation (Dzamko et al., 2015; Kumaran and Cookson, 2015). Apart from PD, it has been associated with hypertension, autism and early-stage non-small-cell lung cancer (Li et al., 2013).

#### HLA-DQB1

HLA-DQB1 (major histocompatibility complex, class II, DQ beta 1) was shown to be associated with the risk for PD in the Nalls et al. study in 2014 (Nalls et al., 2014). Only one intergenic variant was shown to be associated with a PD risk, which means that more association studies confirming the association need to be conducted. It plays a key role in the immune system by presenting peptides to the antigen presenting cells. The HLA region is one of the most complex regions in the genome. It encompasses a high number of closely packed genes, with numerous polymorphisms and complicated patterns of linkage disequilibrium. As part of immune response HLA genes, are involved in many pathologies, such as autoimmune diseases, infections and malignant and neurological disorders. The HLA region is divided into classes, I and II, but they both code for proteins that present antigens to T cell receptors as part of the adaptive immune system. HLA class II molecules are expressed in antigen presenting cells—macrophages, B lymphocytes and dendritic cells. They consist of two chains, alpha (DQA) and beta (DQB) (Lampe et al., 2003; Wissemann et al., 2013; Chang et al., 2015). Several HLA genes (Hill-Burns et al., 2011) are thought to be associated with PD, besides HLA-DQB1 also HLA-DRA (Hamza et al., 2010; Pankratz et al., 2012; Hill-Burns et al., 2014) and HLA-DRB (Ahmed et al., 2012).

#### DDRGK1

DDRGK1 (DDRGK domain containing 1) gene was found to be associated with PD in the meta-analysis of GWASs performed by Nalls et al. (2014), but the association has not been confirmed yet. DDRGK1 regulates the nuclear factor-κB (NF-κB) activity (Dzamko et al., 2015). NF-κB is a group of inducible nuclear transcription factors; it is largely expressed in a wide variety of cells and regulates the expression of a lot of genes in response to different stimuli. It plays an important role in inflammation, immune system as well as in tumorigenesis (Mitchell et al., 2016). DDRGK1 is localized in the endoplasmic reticulum (ER) and its expression is induced by ER stress (Xi et al., 2013).

#### Neurodevelopment, Neuron Cell Differentiation and Survival

SNPs within loci associated with neurodevelopment, neuron cell differentiation and survival recognized as PD susceptibility factors by GWASs are listed in **Table 6**.

#### CCDC62

CCDC62 (coiled-coil domain containing 62) was found to be related to PD susceptibility in the Nalls et al. studies conducted in 2011 and 2014 (Nalls et al., 2011, 2014) pointing out two different intron variants. The gene has a role in cell growth, estrogen receptor transactivation, cyclin D1 expression in prostate cancer cells as well as in other varieties of cancer, since its antibodies are


TABLE 6 | SNPs within loci associated with neurodevelopment, neuron cell differentiation and survival recognized as PD susceptibility factors by GWASs.

Ch, Chromosome. Data compiled from the GWAS catalog database, the dbSNP and the e!Ensembl (Sherry et al., 2001; Welter et al., 2014; Yates et al., 2016).

often produced and thus detected (Li et al., 2013; Lu et al., 2016). Until recently, it was mainly reported as related to cancer (Yu et al., 2015). In order to define the gene's function more precisely, more functional studies are required.

#### RIT2

RIT2 (ras-like without CAAX 2) was first described as a PD susceptibility factor in the GWAS performed by Do et al. in 2011 and after that confirmed in two meta-analyses (Pankratz et al., 2012; Nalls et al., 2014). Two variants within the gene were found to be associated with the PD risk. One of them (rs12456492) was replicated, while the other (rs4130047) was significant in only one study (Do et al., 2011; Pankratz et al., 2012; Nalls et al., 2014). RIT2 is a small GTPase of the Ras family. It is neuron-specific and preferentially expressed in dopaminergic neurons in substantia nigra (Ran and Belin, 2014). Accumulating reports suggest its important role in neuronal differentiation and function. It promotes neurite outgrowth through Rac/Cdc42 activation and calmodulin association (Zhang et al., 2013). It binds calmodulin 1, a phosphorylase kinase, in a calciumdependent manner and regulates certain signaling pathways and cellular processes (Do et al., 2011). It interacts with SNCA and tau via calmodulin (Lu et al., 2015). RIT2 is also a specific interacting partner of the dopamine transporter (DAT). DAT is a transmembrane protein which can be internalized by protein kinase C-mediated endocytosis and thus downregulated. It was proposed that extracellular dopamine concentrations and halflife are regulated in this manner. This process presumably depends on RIT2 GTPase activity (Zhang et al., 2013; Ran and Belin, 2014). A reduced production of RIT2 was detected in postmortem samples of PD patients when compared to controls (Bossers et al., 2009). In another postmortem study, a possible regulation of INF-γ signaling by RIT2 was discovered (Liscovitch and French, 2014).

#### FGF20

FGF20 (fibroblast growth factor 20) was proved to be associated with PD in the Nalls et al. meta-analysis (Nalls et al., 2014), in which one intron variant within the gene with a positive association was found. More studies will have to be done to confirm this association. FGF20 is a neurotrophic factor preferentially expressed in substantia nigra pars compacta. It acts in an autocrine/paracrine manner. FGF20 regulates central nervous development and function (Plagnol et al., 2011). It plays a major role in dopaminergic neurons differentiation and survival (Itoh and Ohta, 2013). According to some studies, FGF20 also increases SNCA levels in dopaminergic neurons, but there is a huge discrepancy among studies (Wang et al., 2008; Wider et al., 2009; Sekiyama et al., 2014; Tarazi et al., 2014).

#### GCH1

GCH1 (GTP cyclohydrolase 1) was found to be associated with PD pathology only in the Nalls et al. meta-analysis of GWASs (Nalls et al., 2014), which means that more studies whose results will show a positive association are required to claim that the association is in fact true. GCH1 is an essential enzyme in dopamine synthesis in the nigrostriatal nervous cells. Mutations in this gene can result in the degeneration of nigral neurons. Loss of function leads to severe depletion of dopamine levels and is the most frequent cause of DOPA-responsive dystonia (DRD), a rare disease that classically occurs during the childhood and is manifested as generalized dystonia and an excellent sustained response to low doses of levodopa, usually without motor fluctuations. DRD is often associated with PD. Mutations in the gene might cause striatal cell death and thus evolve into PD. The enzyme controls the first and rate-limiting step in the biosynthesis of tetrahydrobiopterin (BH4), which is a cofactor of tyrosine hydroxylase, which converts tyrosine to levodopa (Mencacci et al., 2014; Rengmark et al., 2016). It may also have some role in inflammation (Dzamko et al., 2015). The debate on whether this gene is related to the familial or sporadic form is still ongoing. Some say that PD-like symptoms in adulthood could also be a different phenotype of DRD.

#### GPNMB

GPNMB (glycoprotein nonmetastatic melanoma protein B) was recognized as associated with a PD risk in the Nalls et al. meta-analysis in 2014. As the association was found in one study only, we cannot conclude that the association is definite (Nalls et al., 2014). It presumably plays an important role in


#### TABLE 7 | SNPs within loci associated with mitochondrial homeostasis recognized as PD susceptibility factors by GWASs.

Ch, Chromosome. Data compiled from the GWAS catalog database, the dbSNP and the e!Ensembl (Sherry et al., 2001; Welter et al., 2014; Yates et al., 2016). \*Denotes SNPs in 3'-UTR.

neuronal survival and neuroprotection (Xu et al., 2016). It is also known as osteoactivin, dendritic cell-heparin integrin ligand or haematopoietic growth factor inducible neurokinin-1 type. It is important for the differentiation and functioning of osteoclasts and osteoblasts, the impairment T-cell activation, the invasion, and metastasis of many cancers and the regulation of degeneration/regeneration of extracellular matrix in skeletal muscles. The gene was also associated with amyotrophic lateral sclerosis as another type of a neurodegenerative disease (Tanaka et al., 2012). The protein mostly localizes in lysosomes. It is also involved in phagocytosis and helps to recruit an autophagy protein LC3-II to the phagosome. It is essential for the fusion of phagosome and lysosome to degrade the phagosome content (Gan-Or et al., 2015). GPNMB is somehow involved in innate and adaptive immunity along with many PD susceptibility factors. It may play a role in regulating microglial inflammation downstream of LPS activation (Dzamko et al., 2015; Herrero et al., 2015).

#### Mitochondrial Homeostasis

SNPs within loci associated with mitochondrial homeostasis recognized as PD susceptibility factors by GWASs are listed in **Table 7**.

#### SREBF1/RAI1

SREBF1/RAI1 (sterol regulatory element binding transcription factor 1/retinoic acid induced 1) locus was associated with PD susceptibility for the first time by Do et al. GWAS (Do et al., 2011). In 2014, it was again confirmed by the meta-analysis of GWASs (Nalls et al., 2014). The same SNP (rs11868035) was pointed out in both studies.

SREBF1 encodes SREBP-1 (sterol regulatory element-binding protein 1), a transcriptional activator (Do et al., 2011). The connection of mitophagy pathway and familial PD has been firmly established via two genes—PINK1 and Parkin, but also GWASs and functional studies found the association between this pathway and sporadic PD through SREBF1. It is involved in the mitophagy as well as in the regulation of lysosomal lipid accumulation. Mitochondria and lipid metabolism have complementary functions via Krebs cycle. Knockdown of SREBF1 blocks the translocation of Parkin into the mitochondria and consequently decreases mitophagy (Gan-Or et al., 2015). This process can be restored by the addition of exogenous lipids, both fatty acids and cholesterol. Downregulation of SREBF1 may also decrease mitophagy by blocking the stabilization of PINK1 on the outer mitochondrial membrane of injured mitochondria, which can also be repaired by additional exogenous lipids (Ivatt and Whitworth, 2014). It regulates cholesterol synthesis and its cellular uptake from plasma LDL. A reduced expression of SREBF1 also downregulates the NPC1 gene, which may lead to the accumulation of cholesterol within late endosomes and lysosomes (Gan-Or et al., 2015). Autophagic turnover of mitochondria needs to be balanced with its biogenesis, which heavily relies upon SREBP pathway for membrane synthesis (Ivatt et al., 2014). Some studies also suggest that SREBP-1 is an important mediator of NMDA-induced excitotoxicity (Do et al., 2011). It may also be involved in the innate immune response via lipid metabolism (Jeon and Osborne, 2012).

RAI1 is a part of transcription regulation mechanism, beacuse it has the ability to remodel chromatin and interacts with the basic transcriptional machinery (Do et al., 2011).

#### MCCC1

MCCC1 [methylcrotonyl-CoA carboxylase 1 (alpha)] was shown to be associated with sporadic PD in 2011 with two GWASs and after that again confirmed with the meta-analysis. In all three studies, a different intron variant was implicated as a risk factor (Do et al., 2011; Nalls et al., 2011, 2014). Its product, a biotin-requiring enzyme, functions within the mitochondria as a part of leucine catabolism pathway (Wang et al., 2014). It is currently not known if this enzyme has a role in the mitophagy or mitochondrial quality control. Researchers are also debating on whether this gene or rather LAMP3 located on the same locus is involved in PD pathogenesis (Gan-Or et al., 2015).

#### Other Processes

SNPs within loci associated with other processes recognized as PD susceptibility factors by GWASs are listed in **Table 8**. More association and functional studies will have to be performed to discover a clear connection between a PD risk and genes mentioned below.

#### ACMSD/TMEM163

ACMSD/TMEM 163 (aminocarboxymuconate semialdehyde decarboxylase-transmembrane protein 163) was shown to be associated with PD by Nalls et al. studies as well as by Pankratz et al. meta-analysis of GWASs. Two variants were found in these three studies, which means that one of them (rs6430538) was also replicated (Nalls et al., 2011, 2014; Pankratz et al., 2012). This gene is expressed at very low but significant levels


#### TABLE 8 | SNPs within loci associated with other processes recognized as PD susceptibility factors by GWASs.

NA, data not available; Ch, Chromosome. Data compiled from the GWAS catalog database, the dbSNP and the e!Ensembl (Sherry et al., 2001; Welter et al., 2014; Yates et al., 2016).

in the brain and encodes a cytosolic enzyme (Martí-Massó et al., 2013). ACMSD is involved in the picolinic and quinolinic acid homeostasis and is thus a possible therapeutic target for many central nervous system disorders (Plagnol et al., 2011). ACMSD is also categorized as a detoxifying enzyme because of its ability to prevent the accumulation of neurotoxic metabolite quinolinate, which is one of the tryptophan catabolites (Ran and Belin, 2014). It also participates in the kynurenine pathway of tryptophan catabolism. Mutations in the gene presumably result in a decreased enzymatic activity, which means that the conversion of tryptophan to picolinic acid is blocked. Consequently, quinolinic acid levels in the brain are elevated. Quinolinat is also an intermediate of the de novo synthesis of NAD from tryptophan (Martí-Massó et al., 2013).

#### MIR4697

MIR4697 is a non-protein coding sequence that was shown to be associated with PD in the Nalls et al. meta-analysis in 2014 (Nalls et al., 2014) in which one SNP with a positive association was found. The function of the microRNA in general is not known, much less its function in PD pathology (Chen et al., 2016).

#### SIPA1L2

SIPA1L2 (signal-induced proliferation-associated 1 like 2) is a gene related to PD pathogenesis according to the Nalls et al. study conducted in 2014, but the function of the gene has to be determined yet. Only one SNP has been pointed out as a risk factor so far (Nalls et al., 2014).

#### APPLICABILITY OF THE KNOWLEDGE ABOUT GENETIC SUSCEPTIBILITY FACTORS FOR PD

The knowledge about the genetic factors of PD risk or pathogenesis related to a specific pathway could be applied in the clinical setting to support early diagnosis and to predict disease prognosis as suggested by the study of Nalls et al. (2015). Furthermore, the information on susceptibility genes could be easily integrated in composite diagnostic and/or prognostic algorithms that at present include clinical characteristics and imaging data (Siderowf et al., 2012; Gaenslen et al., 2014; Noyce et al., 2014). These genetic factors could also help to stratify the patients on the basis of underlying molecular defect into respective groups that would benefit from targeting a specific pathway with novel or existing treatment approaches, which would enable personalized treatment planning. The latter also indicates an opportunity to identify novel potential targets for innovative treatment approaches and supportive measures.

First, described genetic factors could support early diagnosis. This is an important goal for clinicians dealing with PD, because it is important to start treatment early in the disease course (Olanow and Schapira, 2013). It would be ideal to identify patients in the prodromal phase of the disease (Berg et al., 2015). By the time motor symptoms occur, more than half of dopaminergic neurons are already lost. The time of the prediagnostic phase, which can last for up to 20 years, thus presents the window of opportunity for searching for and for the application of neuroprotective treatment (Salat et al., 2016). For the purpose of early diagnosis, a panel of genes should be tested for different mutations and polymorphisms to check the cumulative effect of the genetic defects as a single mutation or polymorphism has a very low contribution to disease risk. The combined assessment of early non-motor symptoms with a genetic predisposition could be an ideal method to identify the patients in the earliest phases of the disease. Only genetic testing for diagnosis of sporadic PD or for the prediction of developing sporadic PD is not specific and sensitive enough at the moment. There is no known genetic factor or combination of genetic factors that could conclusively predict the development of sporadic PD. Therefore, genetic testing would probably be more suitable for high risk families with positive PD history or for people with manifestations of early non-motor symptoms and not for general population. Furthermore, a cumulative score and not individual genetic factors should be considered, jointly with clinical and imaging data. Nalls et al. developed an algorithm for early diagnosis of PD on the basis of clinical and genetic classification. Their model included several clinical factors, such as age, gender, olfactory function and family history of PD. They combined clinical factors with a genetic risk score, that accounted for 28 genetic variants identified and replicated in the most recent large-scale meta-analysis of PD GWASs data plus two other rare risk variants, a GBA (rs76763715) and a LRRK2 (rs34637584) mutation. The model correctly distinguished between patients with PD and controls with an area under curve of 0.923 with both high sensitivity (0.834) and specificity (0.903). They also validated the model on five other cohorts and AUC was never lower than 0.894 (Nalls et al., 2015). There are a few prospective longitudinal studies running (PARS, TREND, PREDICT-PD), which are looking for biomarkers to identify PD and predict the way of disease progression before a diagnosis based on motor symptoms can be made, and assessing combinations of risk factors and early features of PD. Among these studies, PARS included a cohort of elderly patients with olfactory dysfunction, TREND included people older than 50 years with one of prodromal symptoms of PD (olfactory dysfunction, REM sleep behavior disorder or depression), while PREDICT-PD included general population from 60 to 80 years of age irrespective of PD prodromal symptoms (Salat et al., 2016).

Furthermore, the impact of genetic susceptibility factors on the disease prognosis could be evaluated. It would be interesting to explore the association between the course of the disease and the genetic defect in genes involved in certain pathways. For example, GBA mutations or MAPT H1 allele status might be independent risk factors for cognitive impairment in PD patients and the knowledge of these allele statuses in patients could have an impact on the treatment plan (Clarimón and Kulisevsky, 2013).

Moreover, genetic susceptibility factors segregating in different pathways could help to stratify PD patients into distinct groups according to the differences in the main molecular cause of the disease. So far, the stratification of patients has only been made based on different phenotypes (van Rooden et al., 2010, 2011; Fereshtehnejad et al., 2015), but our main idea is to group them on the basis of the main molecular cause of the disease rather than on the basis of different ways of manifestation. The stratification should be based on cumulative effects of genetic susceptibility factors within a particular pathway as well as across pathways, so different combinations of genetic defects should be tested to find a way to stratify the PD patients. This approach was already addressed in a review written by Schapira et al. (Schapira, 2013), where the problem of aetiologically heterogeneous cohorts of PD patients evaluated in different studies was exposed. The latter makes it difficult to seek for different genetic biomarkers as there may be different ones in distinct groups. The stratification of PD patients based on underlying genetic defects could also be beneficial in a clinical setting as patients with different genetic defects may need different treatment plans. This kind of personalized treatment could become the treatment of choice in PD in the future, as we could stratify patients into groups according to their compromised pathway and treat them according to their underlying pathogenic processes. The treatment could be adjusted for each group to obtain the best possible response.

Last but not least, the disease genes or the products of the genes involved in the aethiopathogenesis could serve as novel targets for neuroprotective or disease-modifying drugs to delay or prevent disease progression. However, the problem of a very modest impact of the above-mentioned mutations and polymorphisms on the PD pathology occurs. If the impact of a single mutation is very low, there is probably no point in developing a new active pharmaceutical agent specific for that target, because the effect on the disease course would probably not be satisfactory. A more feasible way of incorporating means of personalized medicine and patient startification into PD management is to target a corrupted pathway as a whole instead of focusing only on the compromised gene and its product. The goal would be to strengthen or improve the function of a compromised pathway, such as protein aggregation; protein and membrane trafficking; lysosomal autophagy; immune system; neurodevelopment, neuron cell differentiation and survival; mitochondrial homeostasis; and genes involved in other processes. Although this is a less specific approach, it focuses on the putative main cause of the disease occurrence in a particular patient. In the light of this, there is a possibility of the repurposing of drugs already used for other indications, for example ursodeoxycholic acid, nicotine, caffeine, isradipine, exenatide, statins (Salat et al., 2016). In order to establish such a pathway-oriented approach, clinical trials must focus on PD patients with the same underlying molecular defect. As long as heterogenous cohorts are evaluated in these studies, no such treatment will be available (Korczyn and Hassin-Baer, 2015).

#### FUTURE PERSPECTIVES

With the emergence of GWASs, researchers believed that the discovery of the exact genetic background of PD is at their fingertips, but more questions arose afterwards. GWASs are a type of studies with no hypothesis at the time of their performance, so the results must be critically evaluated and supported by functional studies of genes recognized to be involved in the pathogenesis of a certain disease. Apart from functional studies of not yet known genes' functions, animal disease models should also be investigated to check and validate the results of GWASs. We also have to be aware of the fact, that GWAS usually detect common SNPs that most probably have a modest impact only. Rare but highly penetrant SNPs are often overlooked. Hence, in terms of susceptibility prediction or early diagnosis, physicians should check several genetic biomarkers (SNPs) and their cumulative impact on the prognosis along with clinical and imaging data. Additionally, SNPs associated with a certain disease are often not located within the disease gene, but are rather in linkage disequlibrium with it.

The main goal of searching for genetic biomarkers of PD susceptibility is early diagnosis or more optimistically, a chance to slow down or even prevent the development of the disease. Furthermore, with better characterization of risk genes' functions, we could stratify the PD patients into groups according to the main route of pathogenesis. Prospective monitoring is needed to compare the symptomatology as well as the rate of progression between the groups in order to get a chance of tailoring the therapy to each group. To take this idea to the next level, one could also find new drug targets based on the susceptibility genes or their products if the impact of the susceptibility gene is high enough. A personalized approach would allow to treat each group of patients with the most effective and safest drug and adopt optimal supportive measures, e.g., antiinflammatory therapy for patients with an increased inflammatory response.

#### CONCLUSIONS

In conclusion, the newest GWASs have identified top hit susceptibility genes with emerging information on their physiological functions and involvement in PD pathology. These susceptibility genes belong to specific pathways that are already known to be compromised in PD and could thus serve as a genetic tool for the stratification of patients. In our opinion, this would allow the treatment of patients according to the underlying cause of their clinical signs to choose the most beneficial treatment with minimal side effects. This knowledge gives the opportunity to personalize the treatment of PD patients, but more studies need to be carried out on cell models, animal models and patients before new knowledge can be translated into the everyday clinical practice of PD treatment.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

All authors have made a substantial intellectual contribution to this work and approved its final version for submission. VD and SR formed the review focus. SR conducted the literature review and summarized and wrote the first draft of the manuscript under the supervision of VD. VD and MT evaluated the manuscript and contributed to the final version.

#### FUNDING

This work was partially funded by the Horizon 2020 Artemida 664536 Grant and the Slovenian Research Agency (ARRS) Grant P1-0170.


for Parkinson's disease: a meta-analysis of genome-wide association studies. Lancet 377, 641–649. doi: 10.1016/S0140-6736(10)62345-8


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Redenšek, Trošt and Dolžan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Novel Missense Mutation of the DDHD1 Gene Associated with Juvenile Amyotrophic Lateral Sclerosis

Chujun Wu and Dongsheng Fan\*

*Department of Neurology, Peking University Third Hospital, Beijing, China*

Background: Juvenile amyotrophic lateral sclerosis (jALS) is a rare form of ALS with an onset age of less than 25 years and is frequently thought to be genetic in origin. *DDHD1* gene mutations have been reported to be associated with the SPG28 subtype of autosomal recessive HSP but have never been reported in jALS patients.

Methods: Gene screens for the causative genes of ALS, HSP and CMT using next-generation sequencing (NGS) technologies were performed on a jALS patient. Sanger sequencing was used to validate identified variants and perform segregation analysis.

Results: We identified a novel c.1483A>G (p.Met495Val) homozygous missense mutation of the *DDHD1* gene in the jALS patient. All of his parents and young bother were heterozygous for this mutation. The mutation was not found in 800 Chinese control subjects or the database of dbSNP, ExAC and 1000G.

Edited by:

*Daniela Tropea, Trinity College Dublin, Ireland*

Reviewed by:

*Pawan Gupta, University of Illinois at Chicago, USA Mihai Moldovan, University of Copenhagen, Denmark*

\*Correspondence:

*Dongsheng Fan dsfan2010@aliyun.com*

Received: *08 September 2016* Accepted: *17 November 2016* Published: *06 December 2016*

#### Citation:

*Wu C and Fan D (2016) A Novel Missense Mutation of the DDHD1 Gene Associated with Juvenile Amyotrophic Lateral Sclerosis. Front. Aging Neurosci. 8:291. doi: 10.3389/fnagi.2016.00291* Conclusion: The novel c.1483A>G (p.Met495Val) missense mutation of the *DDHD1* gene could be a causative mutation of autosomal recessive jALS.

Keywords: amyotrophic lateral sclerosis, DDHD1 gene, hereditary spastic paraplegia, juvenile, mutation

# INTRODUCTION

Juvenile amyotrophic lateral sclerosis (jALS) is a rare form of ALS with an onset age of less than 25 years and is thought to more frequently have a genetic origin than the adult-onset forms. Juvenile ALS is a clinically and genetically heterogeneous disease. Mutations in ALS2, SETX and SPG11 are known to cause familial jALS with slow disease progression (Orban et al., 2007). In recent literatures, SIGMAR1 have been reported to be a new causative gene of autosomal recessive jALS (Al-Saif et al., 2011; Ullah et al., 2015). However, in sporadic juvenile ALS patients, mutations in Fus are the most frequent genetic cause (Hübers et al., 2015; Zou et al., 2016). On the contrary to slow progression in familial jALS, sporadic jALS patients with FUS or SOD1 mutations experienced aggressive progression and short survival times (Zou et al., 2016). Some of jALS causative genes have also been reported in other diseases, such as ALS2 and SPG11 causing hereditary spastic paraplegia (HSP) as well (Eymard-Pierre et al., 2002; Klebe et al., 2015). In this report, we describe a jALS patient with a novel missense mutation in the DDHD1 gene, which is a member of the intracellular phospholipase A1 gene family and involved in the regulation of mitochondrial function. Associations have been reported between mutations in the DDHD1 gene and the SPG28 subtype of autosomal recessive HSP but have never been reported in jALS patients.

# MATERIALS AND METHODS

#### Subjects

The pedigree for the family is presented in the **Figure 1A**. The clinical characteristics of the patient will be discussed in results section.

#### Ethics Statement

The institutional ethics committee of Peking University Third Hospital approved this study (IRB00006761). Written, informed consent was obtained from each participant.

#### Genetic Analysis

We obtained blood sample from affected and unaffected subjects in the pedigree (III-7, III-8, IV-1, IV-2). Next generation sequencing (NGS)was performed on an Illumina GAIIx platform to screen for variations in the patient, which covers the coding exons and flanking intronic sequences of 190 causative genes of ALS, HSP, and Charcot-Marie-Tooth disease (CMT) (Supplementary file). Identified variants in NGS were validated by Sanger sequencing. In order to perform segregation analysis, all of the patients' parents and young brother were screened for the identified variants using Sanger sequencing.

# RESULTS

## Clinical Features

The patient (IV-1) was a 24-year-old male who developed walking difficulties due to leg weakness beginning at 16 years of age and exhibited atrophy of the bilateral first interosseous

FIGURE 1 | Family pedigree and genomic sequence electropherograms. (A) Family pedigree. The jALS patient was born to consanguineous Chinese parents. All of his parents and young brother were healthy. Males and Females are represented as squares and circles, respectively. The filled symbol represents the affected patient while unaffected individuals are represented by clear symbols. Crossed circles or squares represent deceased individuals. A double line indicates consanguineous mating. (B) Genomic sequence electropherograms. The jALS patient carried a novel homozygous c.1483A>G (p.Met495Val) missense mutation of the *DDHD1* gene, which was not detected in 800 healthy unrelated Chinese individuals. All of his parents and young brother were heterozygous for this mutation. Pedigree analysis suggested that the disease is consistent with autosomal recessive inheritance.

muscles accompanied by a mild weakness in the hands 1 year later. These symptoms have progressed slowly. Besides, he didn't complain of any sensory abnormality. He was born to consanguineous Chinese parents. His parents and younger brother were clinically healthy. All of them didn't present any weakness or atrophy of muscle.

Neurologic examinations were performed for the patient (IV-1), his parents (III-7, III-8) and young brother (IV-2). In the patient, examinations revealed a steppage gait, atrophy of the bilateral interosseous and thenar muscles with a split-hand sign, mild weakness in the hands and lower limbs, hyperreflexia in all limbs with positive bilateral Babinski signs and Hoffmann signs, and disappearance of abdominal reflexes. Sensation and coordination were normal. As for the patients' parents and young brother, neurological examinations showed no muscle atrophy, normal tendon reflex and negative pathological reflex.

Needle EMG was performed on the patient. The results showed neurogenic changes, including fibrillation potentials and positive sharp waves, in four regions (brainstem, cervical, thoracic, and lumbosacral spinal cord). Nerve conduction studies revealed decreases in the amplitude of compound motor and sensory action potentials with an almost normal nerve conduction velocity. The structure brain and cervical MRIs were normal. Upon <sup>1</sup>H-MRS examination, no pathological lactate accumulation was found in the cerebrospinal fluid of lateral ventricles.

#### Genetic Results

Gene screens for the causative genes of ALS, HSP, and CMT using NGS technologies identified a novel homozygous c.1483A>G (p.Met495Val)(RefSeq NM\_001160147.1) missense mutation of the DDHD1 gene in the patient (**Figure 1B**), which was verified by Sanger sequencing. Meanwhile, Sanger sequencing revealed all of his parents and young brother were heterozygous for this mutation (**Figure 1B**). The mutation was not detected in 800 healthy unrelated Chinese individuals by whole exome sequencing or NGS.

# DISCUSSION

DDHD1, also known as SPG28 or PA-PLA1, is a member of the intracellular phospholipase A1 gene family. The protein encoded by the DDHD1 gene is a cytosolic protein with some mitochondrial localization and is involved in the regulation of mitochondrial dynamics via phosphatidic acid (PA) (Baba et al., 2014). The PA on the surface of mitochondria is known to regulate mitochondrial fusion. DDHD1 is the first identified intracellular phospholipase A1 and preferentially hydrolyzes PA in vitro. The DDHD1 pathogenic mutations cause reduced PA-PLA1 activity, and the resultant increased PA content on the surface of mitochondria might cause the impairment of mitochondrial fusion and lead to the dysfunction of mitochondria (Tesson et al., 2012). The previous study has showed the ectopic expression of DDHD1 in HeLa cells induced mitochondrial fragmentation, whereas its depletion caused mitochondrial elongation. Gene disruption of DDHD1 in mice caused sperm malformation due to mitochondrial organization defects (Baba et al., 2014). In patients harboring pathogenic DDHD1 gene mutations, histochemical analyses in muscle showed mitochondrial alterations, and multiple mitochondrial DNA (mtDNA) deletions were evident (Mignarri et al., 2016). Besides, mitochondrial respiration rate, total cellular and mitochondrial ATP content were found to be significantly lower in lymphoblast from SPG28 patients (Tesson et al., 2012). In brain <sup>1</sup>H-MRS analysis, a mild pathological accumulation of lactate in the cerebrospinal fluid was detected in a SPG28 patient with 20 years of disease duration, (Liguori et al., 2014) but not other three patients with shorter disease duration.(Liguori et al., 2014; Mignarri et al., 2016)

In previous studies, 8 mutations of the DDHD1 gene have been reported in autosomal recessive HSP patients, including three nonsense mutations, three frame-shift mutations and two mutations affecting the mRNA splicing site (Tesson et al., 2012; Liguori et al., 2014; Mignarri et al., 2016; Miura et al., 2016). All of these mutations predict changes in the protein translation of the DDHD domain, which is responsible for the phospholipase activity of the DDHD1 protein, thereby leading to a loss of function in the protein and finally mitochondrial dysfunction.

In our report, the patient carried a novel homozygous c.1483A>G (p.Met495Val) missense mutation located in exon 7 of the DDHD1 gene (**Figure 2**). His parents and younger brother were heterozygous for this mutation. Pedigree analysis suggested that the disease is consistent with autosomal recessive inheritance. The mutation was not detected in 800 healthy unrelated Chinese individuals, suggesting that the mutation is not present in the normal Chinese population.

Furthermore, this new mutation, located in a conserved domain (**Figure 3**), was not found in dbSNP, ExAC or 1000G database and was predicted to be disease causing by MutationTaster, possibly damaging by Polyphen-2 and affecting protein function by SIFT. So we concluded this variant was a pathogenic mutation rather than a polymorphism.

Mitochondrial dysfunction is one of the important pathophysiological mechanisms in ALS (Mancuso and Navarro, 2015). In our report, the novel homozygous missense DDHD1 mutation might cause jALS via mitochondrial dysfunction. Unlike the previously reported mutations, the new mutation may not change the protein primary structure but rather the three-dimensional structure of the DDHD domain, ultimately causing a loss of function in the protein. However, further functional studies are needed to confirm mitochondrial dysfunction in ALS patients with DDHD1 gene mutations.

Patients with DDHD1 gene mutations have presented pure or mildly complicated HSP in previous studies (Bouslam et al., 2005; Tesson et al., 2012; Liguori et al., 2014; Mignarri et al., 2016; Miura et al., 2016). Tongue fasciculation with wasting have been reported previously, (Liguori et al., 2014) indicating that DDHD1 gene mutations may influence lower motor neuron (LMN) function. In this patient, muscle atrophy of hands with a splithand sign was apparent, and needle EMG confirmed neurogenic damage in four regions. We believe that the decreased amplitude of compound motor action potentials was the result of severe LMN damage. Apart from motor system involvement, subclinical sensory axonal neuropathy was detected in this patient. In previous reports, clinical and subclinical sensory defects had been reported in HSP patients with DDHD1 gene mutations,

which reflected DDHD1 gene mutations might cause sensory defects (Bouslam et al., 2005; Liguori et al., 2014). Moreover, subclinical sensory abnormalities, peripheral as well as central levels, could be found in ALS patients who didn't have other known potential causes of polyneuropathy, such as diabetes mellitus (Pugdahl et al., 2007; Iglesias et al., 2015; Isak et al., 2016). A multicenter study reported 22.7% patients with ALS had sensory nerve action potentials (SNAPs) abnormalities in at least one nerve (Pugdahl et al., 2007). Although pathophysiological mechanisms underlying sensory abnormalities remains to be further clarified, these findings reflected that ALS might be a multi-systemic disorder involving other systems than motor. In the revision of the EI Escorial criteria 2015, sensory impairment does not exclude ALS diagnosis. Besides, UMN and LMN impairment, rather than subclinical sensory impairment, was the dominant presentation in this patient. Finally, the patient was diagnosed with lab-supported probable jALS. Considering the clinical history of this patient and previously reported cases, we believe that jALS patients with DDHD1 mutations could have slow disease progression, similar to that of other familial jALS patients.

In summary, this is the first case report about the relationship between DDHD1 gene mutations and jALS. Besides of dominant motor system involvement, subclinical sensory impairment was detected in this jALS patient. All of these findings reflected the clinically and genetically heterogeneity of jALS and further proved the genetic overlap between HSP and jALS. Meanwhile, further functional studies are needed to confirm the effects of DDHD1 gene mutations in ALS pathogenesis.

# CONCLUSION

Juvenile ALS is a clinically and genetically heterogeneous disease. As described in our report, we discovered a novel c.1483A>G (p.Met495Val) missense mutation of the DDHD1 gene that could be a causative mutation of autosomal recessive jALS.

# AUTHOR CONTRIBUTIONS

DF conceived this study and provided financial support; CW and DF performed the experiments, analyzed the data, and wrote the manuscript.

# FUNDING

This study was supported by the National Natural Science Foundation of China (81030019).

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnagi. 2016.00291/full#supplementary-material

#### REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Wu and Fan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Slitrk Missense Mutations Associated with Neuropsychiatric Disorders Distinctively Impair Slitrk Trafficking and Synapse Formation

Hyeyeon Kang<sup>1</sup> , Kyung Ah Han<sup>1</sup> , Seoung Youn Won<sup>2</sup> , Ho Min Kim<sup>3</sup> , Young-Ho Lee<sup>1</sup> , Jaewon Ko<sup>4</sup> and Ji Won Um<sup>1</sup> \*

<sup>1</sup> Department of Physiology and BK21 PLUS Project for Medical Science, Yonsei University College of Medicine, Seoul, Korea, <sup>2</sup> Department of Chemistry, Korea Advanced Institute of Science and Technology, Daejeon, Korea, <sup>3</sup> Graduate School of Medical Science and Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea, <sup>4</sup> Department of Biochemistry, College of Life Science and Biotechnology, Yonsei University, Seoul, Korea

Slit- and Trk-like (Slitrks) are a six-member family of synapse organizers that control excitatory and inhibitory synapse formation by forming trans-synaptic adhesions with LAR receptor protein tyrosine phosphatases (PTPs). Intriguingly, genetic mutations of Slitrks have been associated with a multitude of neuropsychiatric disorders. However, nothing is known about the neuronal and synaptic consequences of these mutations. Here, we report the structural and functional effects on synapses of various rare de novo mutations identified in patients with schizophrenia or Tourette syndrome. A number of single amino acid substitutions in Slitrk1 (N400I or T418S) or Slitrk4 (V206I or I578V) reduced their surface expression levels. These substitutions impaired glycosylation of Slitrks expressed in HEK293T cells, caused retention of Slitrks in the endoplasmic reticulum and cis-Golgi compartment in COS-7 cells and neurons, and abolished Slitrk binding to PTPδ. Furthermore, these substitutions eliminated the synapse-inducing activity of Slitrks, abolishing their functional effects on synapse density in cultured neurons. Strikingly, a valine-to-methionine mutation in Slitrk2 (V89M) compromised synapse formation activity in cultured neuron, without affecting surface transport, expression, or synapse-inducing activity in coculture assays. Similar deleterious effects were observed upon introduction of the corresponding valine-to-methionine mutation into Slitrk1 (V85M), suggesting that this conserved valine residue plays a key role in maintaining the synaptic functions of Slitrks. Collectively, these data indicate that inactivation of distinct cellular mechanisms caused by specific Slitrk dysfunctions may underlie Slitrk-associated neuropsychiatric disorders in humans, and provide a robust cellular readout for the development of knowledge-based therapies.

#### Keywords: Slitrks, schizophrenia, Tourette's Syndrome, synapse formation, transmembrane protein

## INTRODUCTION

Neuropsychiatric disorders such as schizophrenia, autism spectrum disorders (ASDs), and Tourette syndrome usually comprise heterogeneous and complex clinical syndromes with largely unknown etiologies (State, 2011; State and Levitt, 2011). Although epidemiological and descriptive studies have attempted to formulate etiological hypotheses to account for a subset of

#### Edited by:

Daniela Tropea, Trinity College, Ireland

#### Reviewed by:

Chiara Verpelli, Istituto di Neuroscienze (CNR), Italy Andreas Vlachos, University of Düsseldorf, Germany

> \*Correspondence: Ji Won Um jiwonum@yuhs.ac

Received: 25 June 2016 Accepted: 04 October 2016 Published: 20 October 2016

#### Citation:

Kang H, Han KA, Won SY, Kim HM, Lee YH, Ko J and Um JW (2016) Slitrk Missense Mutations Associated with Neuropsychiatric Disorders Distinctively Impair Slitrk Trafficking and Synapse Formation. Front. Mol. Neurosci. 9:104. doi: 10.3389/fnmol.2016.00104

neuropsychiatric disorders, genetic studies have clearly demonstrated the high heritability of these disorders, albeit with complex inheritance pattern (Heinzen et al., 2015). To further complicate matters, genetic causation can range from a simple point mutation in a single gene to polygenetic causes that enlist an unknown mode of inheritance, incomplete penetrance, variable expressivity, epistasis, and/or etiological heterogeneity (Heinzen et al., 2015). Faced with this daunting challenge, researchers seeking to attain the conceptual advances necessary to design effective and precise therapeutics, critically require a more detailed comprehension of brain function (Insel and Landis, 2013). Rapid advances in human genome sequencing techniques have contributed significantly to the elucidation of candidate genes that are highly associated with a variety of neuropsychiatric disorders (Medland et al., 2014; Heinzen et al., 2015). In particular, large-scale, genome-wide association studies of patients suffering from various neuropsychiatric diseases have identified copy number variants, singlenucleotide polymorphisms and a variety of point mutations (State, 2011; State and Levitt, 2011; Heinzen et al., 2015). Intriguingly, a number of synaptic genes have frequently been identified as susceptibility factors, supporting the well-accepted 'synaptopathy' hypothesis, which posits that distinct synaptic dysfunctions constitute core features of various neuropsychiatric disorders (Brose et al., 2010). Not surprisingly, a host of synaptic adhesion molecules (e.g., neurexins and neuroligins) have been recognized as candidate contributors to various neuropsychiatric disorders, given their central importance in synaptic functions (Sudhof, 2008; Missler et al., 2012; Ko J. et al., 2015). However, how dysfunctions of most synaptic proteins lead to these devastating disorders remains poorly understood. In particular, why identical genetic factors are commonly found in clinically separable neuropsychiatric disorders, a finding that could account for the high comorbidity of a subset of brain disorders (Keezer and Sander, 2016; Keezer et al., 2016), has remained puzzling.

Slit- and Trk-like (Slitrk) proteins constitute a family of leucine-rich repeat (LRR)-containing synaptic adhesion molecules that are highly expressed in the central nervous system (Aruga and Mikoshiba, 2003; Aruga et al., 2003; Beaubien and Cloutier, 2009; de Wit et al., 2011; Ko, 2012). They possess a common structural architecture composed of tandem LRR domains (LRR1 and LRR2), a single transmembrane domain, and a divergent cytoplasmic region (Aruga and Mikoshiba, 2003). Functionally, they control excitatory and inhibitory synapse formation by interacting with LAR-RPTPs (leukocyte common antigen-related receptor protein tyrosine phosphatases, PTP) through the LRR1 domain (Takahashi et al., 2012; Um and Ko, 2013; Yim et al., 2013; Han et al., 2016a; see Um et al., 2014a, for complex structure). Slitrk3 specifically regulates inhibitory synapse development through binding to PTPδ, whereas other Slitrks organize excitatory synapse development through binding to PTPσ (Takahashi et al., 2012; Yim et al., 2013). Like other synaptic adhesion proteins, a subset of Slitrk point mutants have been linked to schizophrenia, ASDs, Tourette syndrome, or obsessive−compulsive disorder (OCD; reviewed in Ko, 2012; Um and Ko, 2013). For example, Slitrk1 has been postulated to be causative for Tourette syndrome (Abelson et al., 2005; Grados and Walkup, 2006), although subsequent studies (Keen-Kim and Freimer, 2006; Chou et al., 2007; Fabbrini et al., 2007; Scharf et al., 2008) have questioned the validity of conclusions reached by these earlier studies (Keen-Kim and Freimer, 2006; Chou et al., 2007; Fabbrini et al., 2007; Scharf et al., 2008). Moreover, several non-synonymous, rare variants of Slitrk1, Slitrk2, and Slitrk4 were recently reported to be associated with schizophrenia or OCD spectrum disorders (Zuchner et al., 2006; Piton et al., 2011; Ozomaro et al., 2013). However, it remains to be determined whether these mutations are functionally significant or represent benign polymorphisms.

Here, we systematically investigated the effects of four Slitrk1, five Slitrk2, and two Slitrk4 missense mutations on biochemical properties, surface transport, ligand-binding activity, and synaptogenic activities in cultured hippocampal neurons. We show that a subset of Slitrk1 (N400I and T418S) and Slitrk4 (V206I and I578V) mutations impair the biochemical and cell-biological properties of the corresponding wild-type (WT) Slitrks, through a common loss-of-function mechanism that basically traps them intracellularly, blocking their surface transport and ligand binding, and abolishing their synapsepromoting activity. By comparison, a Slitrk2 V89M mutation did not alter any of these parameters; instead it acted through a gain-of-function mechanism to compromise the ability of WT Slitrk2 to restore deficits in synapse density observed in Slitrk2-deficient neurons. Intriguingly, the analogous Slitrk1 point mutant (Slitrk1[V85M]) also displayed an impaired ability to rescue synaptogenic deficits in Slitrk1-deficient neurons. Our data reveal the underlying mechanisms by which a subset of Slitrk missense mutations observed in patients with neuropsychiatric disorders induce distinct gain- and loss-of-function phenotypes, and may provide a partial correlation with the clinical phenotypes shown in Slitrk dysfunction-associated disorders.

#### MATERIALS AND METHODS

#### In silico Analysis

Previously identified Slitrk missense mutations implicated in neuropsychiatric disorders were analyzed using four different prediction programs. (1) PolyPhen-2, available via the Web server<sup>1</sup> , predicts the functional significance of an allele replacement from its individual features employing a Naïve Bayes classifier, trained using supervised machine learning. Mutations whose posterior probability scores are associated with estimated false-positive rates (FPR) at or below the first (lower) FPR are predicted to be 'probably damaging'. Mutations with posterior probabilities associated with FPR at or below the second (higher) FPR are predicted to be 'possibly damaging'. Mutations with an estimated FPR above the second (higher) FPR value are classified as 'benign'. (2) PROVEAN (Protein Variation Effect Analyzer), available via the Web server<sup>2</sup> , clusters BLAST hits using its CD-HIT module based on a global

<sup>1</sup>http://genetics.bwh.harvard.edu/pph2/

<sup>2</sup>http://provean.jcvi.org/index.php

(Continued)

#### FIGURE 1 | Continued

fnmol-09-00104 October 18, 2016 Time: 14:46 # 4

human patients with neuropsychiatric disorders. The target mutated resides are indicated in bold. Schematic drawings of the entire domain organization of human Slitrk1 (A), Slitrk2 (B), and Slitrk4 (C) are shown. (D−F) Similarity or identity of mutated residues investigated in the current study was determined by analyzing the amino acid sequences of human Slitrk1 (D), Slitrk2 (E), and Slitrk4 (F) deposited in the National Center for Biotechnology Information (NCBI) database. Identical residues across various species are indicated in yellow letters on a black background. The following GenBank accession numbers were utilized for sequence alignment: Slitrk1/human, NP\_443142; Slitrk1/gorilla, XP\_004054683; Slitrk1/macaque, NP\_001247716; Slitrk1/bovine, XP\_002691955; Slitrk1/sheep, XP\_004012247; Slitrk1/mouse, EDL00537; Slitrk1/rat, NP\_001100753; Slitrk1/dog, XP\_542628; Slitrk1/zebrafish, XP\_687093; Slitrk1/chicken, XP\_416993; Slitrk2/human, NP\_001137482; Slitrk2/gorilla, XP\_004065021; Slitrk2/macaque, NP\_001248149; Slitrk2/bovine, XP\_015325702; Slitrk2/sheep, XP\_004022343; Slitrk2/pig, XP\_013841934; Slitrk2/rat, NP\_001101057; Slitrk2/mouse, AAI12407; Slitrk2/dog, XP\_013967295; Slitrk2/bird, XP\_012428589; Slitrk2/chicken, XP\_420364; Slitrk4/human, NP\_001171679; Slitrk4/gorilla, XP\_004065019; Slitrk4/macaque, XP\_001086308; Slitrk4/bovine, XP\_005227618; Slitrk4/sheep, XP\_004022346; Slitrk4/rat, NP\_001100417; Slitrk4/mouse, AAI17892; Slitrk4/dog, XP\_005641950; Slitrk4/bird, XP\_012428616; and Slitrk4/chicken, XP\_015134020.



ASD, Autism spectrum disorder; n.a., not applicable; n.d., not determined; PANTHER, Protein Analysis Through Evolutionary Relationships; PolyPhen-2, Polymorphism Phenotyping v2; PROVEAN, Protein Variation Effect Analyzer; SCZ, Schizophrenia; SIFT, Sorting Intolerant From Tolerant; TS, Tourette's syndrome; TTM, Trichotillomania.

sequence identity parameter of 75%. The supporting sequence set, representing the top 30 clusters of closely related sequences, is used to generate predictions. For each supporting sequence, a delta alignment score is computed and used to determine the PROVEAN score. If the PROVEAN score is less than or equal to −2.5 (predefined threshold), the protein variant is considered 'deleterious'; variants with scores greater than −2.5 are considered 'neutral'. (3) MutationAssessor was automatically run using the Web server<sup>3</sup> , as previously described (Reva et al., 2011). (4) PANTHER (Protein Analysis Through Evolutionary Relationships), available through the Web server<sup>4</sup> , generates SubPSEC scores, which are continuous values from 0 (neutral) to about −10 (most likely deleterious). A Pdeleterious value of 0.5

<sup>3</sup>http://mutationassessor.org/r3/

<sup>4</sup>http://www.pantherdb.org/tools/

corresponds to a SubPSEC score of −3. The probability that a given variant will cause a deleterious effect on protein function is estimated from Pdeleterious such that a SubPSEC score of −3 corresponds to a Pdeleterious of 0.5 (Mi et al., 2016). 'n.d.' in PANTHER indicates that the position does not align to Hidden Markov Model (HMM) libraries.

#### Structural Modeling

Structures of human Slitrk2 (LRR1, aa D27–P264; and LRR2, aa S342–P579) and human Slitrk4 (LRR1, aa N28–P64; and LRR2, aa P343–P581) were modeled through the SWISS-MODEL server using human Sltirk1 LRR1 (PDB ID: 4RCA), mouse Slitrk2 LRR1 (PDB ID: 4Y61), and human Slitrk1 LRR2 (PDB ID: 4RCW) as templates. All drawings depicting molecular structures were prepared using PyMOL (PyMOL Molecular Graphics System) (Um et al., 2014a).

Slitrk2, and Slitrk4 are colored gray, cyan and purple, respectively. Yellow dotted regions in the LRR1 domains of Slitrks indicate the LAR-RPTP binding surfaces based on the 3D structure of the human Slitrk1 LRR1/PTPδ Ig1-3 complex (PDB ID: 4RCA), and black dotted lines represent flexible linkers between LRR1 and LRR2 domains (left). Close-up views of mutated and neighboring residues in LRR domains (right).

#### Construction of Expression Vectors

pDisplay-Slitrk1 mutants (V85M, N400I, T418S, R584K, S593G), pDisplay-Slitrk2 mutants (R32L, V89M, S549F, S601P, L626F), and pDisplay-Slitrk4 mutants (V206I, I578V) were generated with a site-directed mutagenesis kit (Stratagene) using the corresponding WT pDisplay constructs as templates. The following constructs were previously described: pDisplay-Slitrk1 WT, pDisplay-Slitrk2 WT, pDisplay-Slitrk4 WT, L-315 sh-Slitrk1, L-315 sh-Slitrk2, and L-315 sh-Slitrk4 (Yim et al., 2013); and pVL1393-PTPδ (Um et al., 2014a). The BFP-KDEL vector was purchased from Addgene (construct #49150).

# Antibodies

The following antibodies were obtained from the indicated commercial sources: mouse monoclonal anti-HA (clone HA-7; Covance), rabbit polyclonal anti-HA (H6908; Sigma), mouse monoclonal anti-α-tubulin (clone DM1A; Sigma), rabbit polyclonal anti-actin (A2066; Sigma), goat polyclonal anti-EGFP (Rockland), and mouse monoclonal anti-GM130 (clone 35/GM130; BD Transduction Laboratories). The antisynapsin antibody was previously described (Han et al., 2016b).

# Heterologous Synapse-Formation Assay

Heterologous synapse-formation assays were performed using HEK293T cells as previously described (Ko et al., 2009; Um et al., 2014b). In brief, HEK293T cells were cotransfected with EGFP or HA-Slitrk1 WT, HA-Slitrk1 point mutants, HA-Slitrk2 WT, HA-Slitrk2 point mutants, HA-Slitrk4 WT, or HA-Slitrk4 point mutants (described in 'Construction of expression vectors') using the FuGene reagent (Roche). After 48 h, transfected HEK293T cells were trypsinized and seeded onto DIV9 hippocampal neuron cultures. Cells were further cocultured for 72 h and then double-immunostained with anti-HA and anti-synapsin antibodies at DIV12 as described previously (Ko J.S. et al., 2015). All images were acquired with a confocal microscope. For quantifications, the contours of transfected HEK293T cells were chosen as the region of interest. The fluorescence intensity of synapsin puncta normalized to the area of each HEK293T cell was quantified for both red and green channels using MetaMorph Software (Molecular Devices).

#### Glycosylation Assay

Solubilized proteins from HEK293T cells transfected with expression plasmids for WT or mutant Slitrks, were first denatured by adding 10x denaturing buffer (New England

SDS-PAGE and immunoblotted using anti-HA antibodies; α-tubulin was used for normalization. Molecular mass markers are labeled in kilodaltons. (B) Immunoblot analysis of Endo-H and PNGase F enzyme-digested lysates of HEK293T cells expressing WT HA-tagged Slitrk; α-tubulin or actin was used for normalization. Molecular mass markers are labeled in kilodaltons. (C−E) Surface expression analysis of HEK293T cells expressing WT or point mutant forms of HA-tagged Slitrk1 (Continued)

#### FIGURE 3 | Continued

fnmol-09-00104 October 18, 2016 Time: 14:46 # 7

(C), Slitrk2 (D), or Slitrk4 (E). Transfected cells were immunostained with mouse anti-HA antibodies (red) and detected with Cy3-conjugated anti-mouse secondary antibodies under non-permeabilized conditions, followed by permeabilization of cells. Cells were then stained first with rabbit anti-HA antibodies (green) and then with FITC-conjugated anti-rabbit secondary antibodies. Scale bar, 10 µm (applies to all images). (F) Quantification of the proportion of cells exhibiting surface expression of Slitrks. All data are shown as means ± SEMs (2∗p < 0.01; <sup>3</sup>∗p < 0.001; ANOVA with post hoc Tukey's test; p-value of WT Slitrk1 vs Slitrk1[N400I] = 0.000033; p-value of WT Slitrk1 vs Slitrk1[T418S] = 0.000065; p-value of WT Slitrk1 vs. Slitrk1[R584K] = 0.259; p-value of WT Slitrk1 vs. Slitrk1[S593G] = 0.34; p-value of WT Slitrk2 vs. Slitrk2[R32L] = 0.941; p-value of WT Slitrk2 vs. Slitrk2[V89M] = 0.948; p-value of WT Slitrk2 vs. Slitrk2[S549F] = 0.988; p-value of WT Slitrk2 vs. Slitrk2[S601P] = 0.997; p-value of WT Slitrk2 vs. Slitrk2[L626F] = 0.996; p-value of WT Slitrk4 vs. Slitrk4[V206I] = 0.000677; p-value of WT Slitrk4 vs. Slitrk4[I578V] = 0.003946). The numbers of cells counted (n) were as follows: WT Slitrk1, n = 360; Slitrk1[N400I], n = 327; Slitrk1[T418S], n = 412; Slitrk1[R584K], n = 496; Slitrk1[S593G], n = 333; WT Slitrk2, n = 345; Slitrk2[R32L], n = 321; Slitrk2[V89M], n = 371; Slitrk2[S549F], n = 360; Slitrk2[S601P], n = 313; Slitrk2[L626F], n = 319; WT Slitrk4, n = 256; Slitrk4[V206I], n = 181; and Slitrk4[I578V], n = 161. (G) Slitrk surface exposure on transfected HEK293T cells, analyzed by immunoblotting of affinity-purified, surface-biotinylated Slitrk proteins. Biotinylated cell surface proteins (S) and total lysate proteins (I) were assessed by immunoblot with anti-HA antibodies. Input, 5% of total lysates used in biotinylation experiments. Note that mutants of Slitrk1 (N400I and T418S) and Slitrk4 (V206I and I578V) exhibit impaired surface expression, as similarly observed in (C−F).

Biolabs) and heating to 100◦C for 10 min. For endoglycosidase H (Endo H) treatment, denatured protein was treated with 1 µl of enzyme and incubated at 37◦C for 1 h. For PNGase F treatment, denatured protein was incubated with 1 µl of enzyme in the presence of 2 µl of NP-40 (100%) at 37◦C for 1 h. Thereafter, enzyme-treated proteins, together with an equal amount of untreated proteins, were analyzed by immunoblotting with the indicated antibodies followed by enhanced chemiluminescence (ECL) detection.

## Biotinylation Assay

Biotinylation experiments in HEK293T cells were performed as previously described (Han et al., 2016b). Briefly, HEK293T cells transfected with the indicated Slitrk plasmids were washed three times with ice-cold PBS containing 1 mM MgCl<sup>2</sup> and 1 mM CaCl2. Sulfo-NHS-LC-biotin (1 mg/ml) was then added, and cells were kept at 4◦C for 30 min. After incubation, cells were washed three times with PBS plus 100 mM glycine to quench and remove excess biotin. Purified membrane proteins were then incubated with Neutravidin (Thermo Scientific) overnight at 4◦C. After three washes with PBS, proteins were eluted with sample buffer and analyzed by immunoblotting.

### Primary Neuronal Culture, Transfection, and Immunocytochemistry

Rat hippocampal cultures were prepared from embryonic day 18 (E18) embryos as described previously (Ko et al., 2011; Um et al., 2016). All experimental protocols using pregnant rats were approved by the Institutional Animal Care and Use Committee of Yonsei University College of Medicine. For overexpression, hippocampal neurons were transfected with various Slitrks vector using CalPhos Kit (Clontech) at DIV10 and immunostained at DIV14. For knockdown of Slitrk1, Slitrk2, or Slitrk4, hippocampal neurons were co-transfected with the corresponding L-315 sh-Slitrk vector at DIV8 and immunostained at DIV14, as indicated in the figure legends. For immunocytochemistry, cultured neurons were first fixed with 4% paraformaldehyde/4% sucrose for 10 minutes at room temperature and then permeabilized with 0.2% Triton X-100 in PBS for 5 min at 4◦C. Fixed, permeabilized neurons were then blocked by incubating with 3% horse serum/0.1% crystalline grade bovine serum albumin (BSA) in PBS for 30 min at room temperature, and incubated with the indicated primary and secondary antibodies in blocking solution for 1 h each at room temperature.

## Neurite Outgrowth Assay

Hippocampal neurons prepared from E18 rat embryo were transfected with WT Slitrk, Slitrk[N400I] or Slitrk[T418S] together with pEGFP-N1 vector at DIV3 and immunostained with anti-EGFP antibodies at DIV6. Fluorescent images of neurons were randomly captured and neurite length was analyzed using MetaMorph software (Molecular Devices).

# Confocal Microscopy Image Acquisition and Analysis

Transfected neurons were randomly chosen and acquired at constant imaging settings using a confocal microscope (LSM700; Carl Zeiss) with a 63× objective lens. Z-stack images obtained at 0.1µm intervals by confocal microscopy were converted to maximal projections, and the size and density of presynaptic terminals were analyzed using MetaMorph software. All images were separated into different color channels (red and green), and red-colored images were transformed into an image in grayscale mode using Photoshop (Adobe). After selecting one or two primary dendrites from neurons in a single image frame, dendrite lengths were recorded and dendritic regions of interest were manually traced in MetaMorph software and saved for puncta measurements (in rgn file format). A constant intensity threshold that excluded diffuse nonsynaptic signals but included synaptic signals (90; range, 0–255) was applied to all gray images. The saved dendritic regions were loaded, calibrated, and measured using the 'integrated morphometry analysis' option. The linear density of synapsin clusters was determined from calculated total puncta numbers, normalized to 10 µm length of dendrite. For puncta size and intensity measurements, normalized puncta areas and averaged puncta intensities were calculated and exported automatically to the Excel program (Microsoft). All quantitative analyses were performed in a blinded manner.

#### Statistics

All data are expressed as means ± SEM. All statistical analyses were performed using SPSS Statistics 23 (IBM, Armonk, NY,

USA). The normal distribution of the data was investigated by p-values from Kolmogorov-Smirnov test (obtained p-values > 0.05, except the data in **Figures 6C** and **10B**). Thus, these data were statistically evaluated using one-way analysis of variance (ANOVA), using cell numbers (>10) or the number of experiments (>3) as the basis for 'n'. For the **Figures 6C** and **10B,** the data were statistically assessed using a non-parametric Kruskal−Wallis test.

#### RESULTS

A variety of rare Slitrk point mutations have been found to be associated with schizophrenia or OCD spectrum disorders (Proenca et al., 2011; Ko, 2012; Um and Ko, 2013). Because one of the aims of this study was to evaluate the functional consequences of SLITRK gene mutations for neuropsychiatric disorders, we focused on only non-synonymous, missense mutations in this

stained with anti-HA (green) and anti-GM130 (red) antibodies. Scale bar, 10 µm (applies to all images). (D) Quantification of the average intensity of GM130-positive Slitrks. All data are shown as means ± SEMs (3∗p < 0.001; ANOVA with post hoc Tukey's test; p-value of WT Slitrk1 vs Slitrk1[N400I] = 0.00053; p-value of WT Slitrk1 vs. Slitrk1[T418S] = 0.000383; p-value of WT Slitrk2 vs. Slitrk2[V89M] = 1.83943E-7; p-value of WT Slitrk4 vs. Slitrk4[V206I] = 8.0637E-9; p-value of WT (Continued)

#### FIGURE 5 | Continued

fnmol-09-00104 October 18, 2016 Time: 14:46 # 10

Slitrk4 vs. Slitrk4[I578V] = 5.1018E-9). The numbers of cells counted were as follows: WT Slitrk1, n = 26; Slitrk1[N400I], n = 22; Slitrk1[T418S], n = 24; WT Slitrk2, n = 24; Slitrk2[V89M], n = 27; WT Slitrk4, n = 19; Slitrk4[V206I], n = 22; and Slitrk4[I578V], n = 23. (E−G) Immunofluorescence staining of COS-7 cells demonstrates ER retention of mutants. COS-7 cells were transfected with WT or mutant forms of HA-tagged Slitrk1 (E), Slitrk2 (F), or Slitrk4 (G), together with BFP-KDEL (red; pseudo-colored). The cells were then stained with anti-HA (green). Scale bar, 10 µm (applies to all images). (H) Quantification of the average intensity of KDEL-positive Slitrks. All data are shown as means ± SEMs (∗p < 0.05, <sup>3</sup>∗p < 0.001; ANOVA with post hoc Tukey's test; p-value of WT Slitrk1 vs. Slitrk1[N400I] = 0.000093; p-value of WT Slitrk1 vs. Slitrk1[T418S] = 0.036; p-value of WT Slitrk2 vs. Slitrk2[V89M] = 0.000712; p-value of WT Slitrk4 vs. Slitrk4[V206I] = 0.000257; p-value of WT Slitrk4 vs. Slitrk4[I578V] = 0.000458). The numbers of cells counted were as follows: WT Slitrk1, n = 26; Slitrk1[N400I], n = 24; Slitrk1[T418S], n = 22; WT Slitrk2, n = 17; Slitrk2[V89M], n = 23; WT Slitrk4, n = 22; Slitrk4[V206I], n = 22; and Slitrk4[I578V], n = 21.

study. Eleven missense mutations have previously been reported to be linked to neuropsychiatric disorders (Zuchner et al., 2006; Piton et al., 2011; Ozomaro et al., 2013). These include N400I, T418S, R584K, and S593G in human Slitrk1; R32L, V89M, S549F, S601P, and L626F in human Slitrk2; and V206I and I578V in human Slitrk4 (**Figures 1A–C**). The L626F mutation in human Slitrk2 also exists in other human Slitrks at equivalent positions (**Figure 1B**), but none of the other residues exhibit complete sequence identity across the six Slitrk members (**Figures 1A–C**). Notably, all four mutated residues identified in human Slitrk1 are unique to Slitrk1 (**Figure 1A**). However, most of these residues are quite evolutionarily conserved among various species, implying their possible functional significance (**Figures 1D–F**). To draw inferences regarding the structural and functional importance of these single amino acid substitutions, we employed the widely used PolyPhen2 (Kumar et al., 2009), PANTHER (Thomas and Kejariwal, 2004), SIFT (Adzhubei et al., 2010), and MutationAssessor (Reva et al., 2011) software packages (**Table 1**). Interestingly, none of the Slitrk missense mutations that are the focus of this study were consistently predicted to be either benign or have deleterious impacts on the stability and function of human Slitrks by the four different in silico prediction tools (**Table 1**).

## Prediction of Structural Phenotypes Produced by Slitrk Missense Mutations, As Reflected in Protein Folding and Three-Dimensional (3D) Structure

Notably, two Slitrk1 mutations (N400I and T418S) are located in the LRR2 domain (**Figure 1A**). Crystal structure of human Slitrk1 LRR2 indicated that the residue N400 of human Slitrk1 forms a weak hydrogen bond with an amino group of the main chain of S375 and a hydroxyl group of the side chain of N376 on the neighboring loop (**Figure 2**). Therefore, mutating N400 to a nonpolar isoleucine (Ile) residue is likely to disrupt these interactions, possibly causing misfolding and aberrant protein trafficking (see below). The side chain of T418 in human Slitrk1 forms a hydrogen bond with a carboxyl group of the main chain of E415 and is involved in hydrophobic interactions with I390, F395, and F419 (**Figure 2**). Thus, a point mutation of T418 to serine (T418S) is also expected to disrupt these hydrophobic interactions. The other Slitrk1 mutations (R584K and S593G) and Slitrk2 mutations (S601P and L626F) are located outside major structural domains, consistent with the results of in silico analyses (**Figure 1A**; **Table 1**). R32L in human Slitrk2 is located immediately preceding the LRR1 domain (i.e., the terminal residue of the signal peptide; **Figure 1A**). Point mutations in human Slitrk2 (V89M or S549F) and human Slitrk4 (V206I or I578V) were predicted to have little effect on the 3D structures of individual Slitrks (**Figure 2**) (Um et al., 2014a).

# Biochemical and Ligand-Binding Phenotypes of Disease-Associated Slitrk Missense Mutants

We next investigated the expression levels and intracellular trafficking of Slitrk mutants in non-neuronal cells (**Figure 3**). As is typically observed for numerous glycoproteins (Yim et al., 2013), immunoblot analyses of HEK293T cells transfected with expression vectors for HA-tagged, full-length Slitrks showed that WT Slitrk proteins were detectable as two discrete bands: approximately 75–100 kDa for Slitrk1 and Slitrk2, and 100–110 kDa for Slitrk4 (**Figure 3A**). Total protein expression levels of Slitrk point mutants were comparable to those of the corresponding WT Slitrks (**Figure 3A**). Strikingly, the mature protein levels of a subset of Slitrk1 mutants (N400I and T418S) and Slitrk4 mutants (V206I and I578V) were significantly decreased (**Figure 3A**). The upper band observed in lysates of Slitrk-expressing HEK293T cells represents fully glycosylated mature protein species that are resistant to Endo-H (which cleaves only immature sugars attached in the endoplasmic reticulum, ER) and are presumably presented to the cell surface, whereas the lower band represents glycosylated immature protein species that are sensitive to Endo-H and are undergoing processing in the ER compartment (**Figure 3B**). Immunoblot analyses of WT Slitrks treated with PNGase F, which removes all attached N-linked oligosaccharides, showed a band shift (**Figure 3B**), consistent with a previous report that Slitrks are highly glycosylated (Yim et al., 2013). We next examined the surface and intracellular protein levels of individual WT Slitrks and Slitrk point mutants in HEK293T cells (**Figures 3C–E**). In line with the biochemical data, two Slitrk1 mutants (N400I and T418S) and two Slitrk4 mutants (V206I and I578V), but not the other Slitrk mutants, displayed significantly reduced surface expression levels with complete trapping in intracellular compartment (**Figures 3C–F**). To reaffirm findings of immunofluorescence analyses in HEK293T cells, we performed biotinylation experiments. We found that the same mutations in Slitrk1 (N400I and T418S) and Slitrk4 (V206I and I578V) impaired the surface expression of the corresponding protein (**Figure 3G**). We then performed binding assays between recombinant Ig-fusion proteins of PTPδ (IgC-PTPδ)

immunofluorescence in primary dendrites. The average intensity of Slitrk WT and mutants in soma region of the transfected neurons was also quantified. All data are shown as means ± SEMs (2∗p < 0.01; <sup>3</sup>∗p < 0.001; a non-parametric Kruskal−Wallis test; p-value of WT Slitrk1 vs. Slitrk1[N400I] = 0.001; p-value of WT Slitrk1 vs. Slitrk1[T418S] < 0.001; p-value of WT Slitrk2 vs. Slitrk2[R32L] = 0.052; p-value of WT Slitrk2 vs. Slitrk2[V89M] = 0.000; p-value of WT Slitrk2 vs. Slitrk2[S549F] = 0.178; p-value of WT Slitrk2 vs. Slitrk2[S601P] = 0.169; p-value of WT Slitrk2 vs. Slitrk2[L626F] = 0.284). 'n' denotes the number of neurons as follows: WT Slitrk1, n = 16; Slitrk1[N400I], n = 19; Slitrk1[T418S], n = 16; WT Slitrk2, n = 16; Slitrk2[R32L], n = 14; Slitrk2[V89M], n = 14; Slitrk2[S549F], n = 13; Slitrk2[S601P], n = 14; and Slitrk2[L626F], n = 13.

and HEK293T cells expressing HA-tagged Slitrks (**Figure 4**). Recombinant IgC-PTPδ robustly bound to HEK293T cells expressing various Slitrk mutants, except those expressing the Slitrk1/4 mutants with impaired surface expression (**Figures 4** and **3C–G**). These data suggest that a subset of Slitrk point mutations observed in neuropsychiatric patients causes improper biochemical processing and abnormal cellular trafficking in nonneuronal cells.

Frontiers in Molecular Neuroscience | www.frontiersin.org October 2016 | Volume 9 | Article 104 |

Kang et al. Slitrk Mutations Implicated in Brain Disorders

#### FIGURE 7 | Continued

staining intensity (red) to HA/EGFP intensity (blue). All data are shown as means ± SEMs (∗p < 0.05; <sup>2</sup>∗p < 0.01; <sup>3</sup>∗p < 0.001; ANOVA with post hoc Tukey's test; p-value of WT Slitrk1 vs. Slitrk1[N400I] = 0.00176; p-value of WT Slitrk1 vs. Slitrk1[T418S] = 0.0497; p-value of WT Slitrk1 vs. Slitrk1[R584K] = 0.959; p-value of WT Slitrk1 vs. Slitrk1[S593G] = 0.434; p-value of WT Slitrk2 vs. Slitrk2[R32L] = 0.818; p-value of WT Slitrk2 vs. Slitrk2[V89M] = 0.957; p-value of WT Slitrk2 vs. Slitrk2[S549F] = 0.956; p-value of WT Slitrk2 vs. Slitrk2[S601P] = 0.984; p-value of WT Slitrk2 vs. Slitrk2[L626F] = 0.987; p-value of WT Slitrk4 vs. Slitrk4[V206I] = 0.00062; and p-value of WT Slitrk4 vs. Slitrk4[I578V] = 1.5112E-7). 'n' denotes the number of HEK293T cells as follows: Control, n = 14; WT Slitrk1, n = 14; Slitrk1[N400I], n = 12; Slitrk1[T418S], n = 14; Slitrk1[R584K], n = 14; Slitrk1[S593G], n = 21; WT Slitrk2, n = 12; Slitrk2[R32L], n = 10; Slitrk2[V89M], n = 10; Slitrk2[S549F], n = 15; Slitrk2[S601P], n = 15; Slitrk2[L626F], n = 15; WT Slitrk4, n = 17; Slitrk4[V206I], n = 13; and Slitrk4[I578V], n = 15.

# Cellular Phenotypes of Disease-Associated Slitrk Missense Mutants

The immunocytochemical results clearly indicate that a subset of Slitrk point mutants exhibits abnormal trafficking and decreased surface expression (**Figure 3**). To independently corroborate this interpretation, we analyzed the subcellular distribution of Slitrk mutants using immunofluorescence microscopy (**Figure 5**). First, we transfected COS-7 cells with WT or mutant Slitrks, and stained them with antibodies against HA and GM130 (a cis-Golgi marker) (**Figures 5A–C**). Consistent with the biochemical data, a subset of Slitrk mutant proteins showed distinct patterns of colocalization with GM130 (**Figures 5A–C**). The average fluorescence intensity of GM130-positive Slitrk mutant proteins was significantly increased than that of WT (**Figure 5D**). We also cotransfected COS-7 cells with WT and mutant Slitrks, together with a blue fluorescent protein-fused KDEL construct to visualize the ER compartment (**Figures 5E–G**). Again, we found that Slitrk mutant proteins showed overlapping colocalization with BFP-KDEL (**Figure 5H**). These data indicate that Slitrk mutants are in general intracellularly trapped in the cis-Golgi and/or ER in non-neuronal cells. To ensure that these Slitrk mutations had a similar effect on the transport of Slitrks out of the ER in neurons, we transfected cultured hippocampal neurons with WT or mutant Slitrks and analyzed them by immunocytochemistry (**Figure 6**). HA-tagged WT Slitrk1 and Slitrk2 were efficiently transported into dendrites, whereas HA-Slitrk1[N400I], HA-Slitrk1[T418S] or HA-Slitrk2[V89M] were either completely retained in the cell body of transfected neurons or displayed markedly impaired trafficking into dendrites (**Figures 6A,B**), as judged by the comparable immunofluorescence signals in the soma of transfected neurons (**Figure 6C**). Notably, other Slitrk2 mutants were targeted to dendrites similarly to Slitrk2 WT (**Figures 6B,C**). These observations suggest that the surface transport deficiency observed in heterologous cells (except for HA-Slitrk2[V89M]) is similarly recapitulated in cultured hippocampal neurons. We were unable to examine the distribution of WT or mutant forms of HA-Slitrk4 because the expression levels of these constructs were too low to be evaluated in this type of analysis.

were then immunostained with antibodies against EGFP or HA (blue) and synapsin (red). Scale bar, 10 µm (applies to all images). (D) Synapse-formation activity in panels (A–C) was quantified by measuring the ratio of synapsin

(Continued)

puncta size (C). More than three dendrites per transfected neuron were analyzed and group-averaged. All data shown are means ± SEMs (∗p < 0.05; <sup>2</sup>∗p < 0.01; <sup>3</sup>∗p < 0.001; ANOVA with post hoc Tukey's test; p-value of sh-Control vs. sh-Slitrk1 = 0.029; p-value of sh-Control vs. sh-Slitrk1+WT Slitrk1 = 0.99; p-value of sh-Control vs. sh-Slitrk1+Slitrk1[N400I] = 0.004; p-value of sh-Control vs. sh-Slitrk1+Slitrk1[T418S] = 0.049; p-value of sh-Control vs. sh-Slitrk2 = 0.039; p-value of sh-Control vs. sh-Slitrk2+WT Slitrk2 = 0.99; p-value of sh-Control vs. sh-Slitrk2+Slitrk2[V89M] = 0.0093; p-value of sh-Control vs. sh-Slitrk4 = 0.00075; p-value of sh-Control vs. sh-Slitrk4+WT Slitrk4 = 0.877; p-value of sh-Control vs. sh-Slitrk4+Slitrk4[V206I] = 0.003; and p-value of sh-Control vs. sh-Slitrk4+Slitrk4[I578V] = 0.008). 'n' denotes the number of neurons as follows: sh-Control, n = 16, sh-Slitrk1, n = 14, sh-Slitrk1+WT Slitrk1, n = 14; sh-Slitrk1+Slitrk1[N400I], n = 14; sh-Slitrk1+Slitrk1[T418S], n = 14; sh-Slitrk2, n = 13; sh-Slitrk2+WT Slitrk2, n = 13; sh-Slitrk2+Slitrk2[V89M], n = 13; sh-Slitrk4, n = 17; sh-Slitrk4+WT Slitrk4, n = 15; sh-Slitrk4+Slitrk4[V206I], n = 15; and sh-Slitrk4+Slitrk4[I578V], n = 15.

# Neuronal Phenotypes of Disease-Associated Slitrk Missense Mutants

Slitrks were previously shown to trigger presynaptic differentiation when expressed in heterologous cells and cocultured in contact with axons (Takahashi et al., 2012; Yim et al., 2013). To test whether the surface transport-deficient Slitrk mutants showed differences in presynaptic differentiationinducing behavior compared with WT Slitrks, we performed heterologous synapse-formation assays with HEK293T cells expressing WT or mutant Slitrks (**Figure 7**). We found that, whereas WT Slitrk1, Slitrk2, and Slitrk4 robustly recruited synapsin clustering into the corresponding transfected HEK293T cells, Slitrk1[N400I] and Slitrk1[T418S] as well as Slitrk4[V206I] and Slitrk4[I578V] mutants were functionally inactive, as expected by their lack of surface transport (**Figures 7A,C**; see quantification in **Figure 7D**). None of the tested Slitrk2 mutants exhibited altered synapse-inducing activity (**Figures 7B,D**). To determine whether the surface transport-impairing Slitrk mutations also compromised the ability of WT Slitrk proteins to promote synapse formation in cultured neurons, we introduced Slitrk isoform-specific knockdown (KD) vectors into cultured hippocampal neurons at 8 days in vitro (DIV8), and stained

anti-HA antibodies (green) and then with FITC-conjugated anti-rabbit secondary antibodies. Scale bar, 10 µm (applies to all images). (B,C) Representative images (B) and summary graph (C) of heterologous synapse-formation activities of WT Slitrk1 and Slitrk1[V85M]. Neurons were stained with antibodies against HA (blue; pseudo-colored) and synapsin (red). Scale bar, 10 µm (applies to all images). 'n' denotes the number of HEK293T cells analyzed: WT Slitrk1, n = 13; and Slitrk1[V85M], n = 12. (D) Fluorescence images of hippocampal neurons transfected with WT HA-Slitrk1 or HA-Slitrk1[V85M], at DIV8. After 48 h (DIV10), the transfected neurons were double-immunostained for antibodies against the somatodendritic marker MAP2 (green) and HA (red). Note that Slitrk1[V85M] exhibits decreased dendritic targeting. Scale bar, 50 µm (applies to all images). (E) Dendritic targeting of WT HA-Slitrk1 or HA-Slitrk1[V85M] in hippocampal neurons was quantified by measuring fluorescent intensity (HA) in primary dendrites. All data are shown as means ± SEM (2∗p < 0.01; Student's t-test; p-value = 0.00215). 'n' denotes the number of neurons as follows: WT Slitrk1, n = 15 and Slitrk1[V85M], n = 13. (F,G) Representative images (F) and summary graph (G) of neuron transfection assays with a lentiviral vector expressing sh-Control, sh-Slitrk1, or coexpressing sh-Slitrk1 and the indicated shRNA-resistant Slitrk1 vectors at DIV8 and analyzed at DIV14 by double-immunofluorescence staining with antibodies to EGFP (green) and synapsin (red). Scale bar, 5 µm (applies to all images). All data are shown as means ± SEMs (3∗p < 0.001; ANOVA with post hoc Tukey's test; p-value of sh-Control vs. sh-Slitrk1 = 0.000411; p-value of sh-Control vs. sh-Slitrk1+WT Slitrk1 = 0.991; and p-value of sh-Control vs sh-Slitrk1+Slitrk1[V85M] = 2.74E−<sup>7</sup> ). 'n' denotes the number of neurons quantified as follows: sh-Control, n = 15, sh-Slitrk1, n = 15, sh-Slitrk1+WT Slitrk1, n = 15; and sh-Slitrk1+Slitrk1[V85M], n = 17.

the transfected neurons with anti-EGFP and anti-synapsin antibodies at DIV14 (**Figure 8**). As previously reported (Yim et al., 2013), single KD of Slitrk1, Slitrk2, or Slitrk4 significantly decreased the linear density of synapsin clusters (**Figures 8A,B**). Expression of short hairpin RNA (shRNA)-resistant forms of WT Slitrk1, Slitrk2, or Slitrk4 completely rescued these deficits in the numbers of synapsin clusters (**Figures 8A,B**). Similar to the results of heterologous synapse-formation assays, expression of shRNA-resistant Slitrk1 mutants (N400I or T418S) or Slitrk4 mutants (V206I or I578V) failed to reverse the reduction in the density of synapsin clusters induced by knockdown of Slitrk1 or Slitrk4 (**Figures 8A,B**). Strikingly, expression of the Slitrk2[V89M] mutant, which showed total expression levels, surface transport, and synapse-inducing properties comparable to those of WT Slitrk2, did not rescue the Slitrk2 KD-induced deficit in the numbers of synapsin clusters, likely due to impaired dendritic targeting in neurons (**Figures 8A,B** and **3**). Since the valine residue at position 89 in human Slitrk2 is also found Slitrk1, -3 and -4 (but not Slitrk5 and- 6; **Figure 1B**), we introduced a similar point mutation into human Slitrk1 (Slitrk1[V85M]) and examined whether this artificially generated Slitrk1 mutant behaves similarly to Slitrk2[V89M] (**Figure 9**). We found that Slitrk1[V85M] showed surface expression levels and synaptogenic activity comparable to those of WT Slitrk1 (**Figures 9A–C**), but exhibited decreased dendritic targeting in hippocampal neurons (**Figures 9D,E**). Moreover, Slitrk1[V85M] showed impaired synapse-promoting activity when expressed in Slitrk1 deficient neurons (**Figures 9F,G**). These data suggest that different Slitrk variants found in patients with neuropsychiatric disorders may manifest distinct cellular readouts, compounding the enormous complexity at both cellular and behavioral levels.

#### DISCUSSION

The 'synaptopathy' hypothesis has been dominant in the neuroscience field since the term was coined, aided in part by the rapid advent of high-resolution human genetic sequencing technologies (Brose et al., 2010). Various classes of synaptic genes have been associated with a range of neuropsychiatric and neurodevelopmental disorders, particularly in the case of ASDs and schizophrenia, although most copy number variants are rarely found in patients with these disorders (Glessner et al., 2012; Malhotra and Sebat, 2012). Surprisingly, similar genetic pathways have often been found to be linked to phenotypically distinct outcomes, confounding the interpretation of genetic approaches (Marshall and Scherer, 2012). Specifically, synaptic adhesion molecules and their associated scaffold proteins, such as neurexin-1α, neuroligin-4 and Shank3, are among the few synaptic genes that are frequently identified as causative factors for ASDs and schizophrenia (Sudhof, 2008; Won et al., 2013; de la Torre-Ubieta et al., 2016). However, how these diseasesusceptible genes cause the associated disorders has only recently begun to be understood.

In the present study, we employed a series of functional approaches to ask whether Slitrk mutations identified in schizophrenia, trichotillomania or OCDs alter the biochemical, cellular and synaptic processes that are mediated by WT Slitrk proteins. We found that a subset of Slitrk1, Slitrk2, and Slitrk4 missense mutants exhibit common and distinct phenotypes in a variety of assays, results that often differed from those predicted by algorithm-based in silico analyses (**Table 1**). Indeed, many of these mutations did not induce any prominent alterations in the properties of Slitrk proteins, although it is possible that the functional assays employed in the current study were unable to capture all aspects of Slitrk function. Thus, our data underscore

the importance of experimentally testing whether candidate mutants identified based on human genetics are physiologically significant. In exploring whether candidate human mutations have distinctive phenotypes at molecular and cellular levels, we focused on two Slitrk1 missense mutations (N400I and T418S), a single Slitrk2 mutation (V89M), and two Slitrk4 mutations (V206I and I578V).

Our data suggest that the respective disorder phenotypes observed in patients with Slitrk mutations are at least partly caused by Slitrk dysfunctions, based on the following evidence: (1) most of the mutated residues in Slitrk1, Slitrk2, and Slitrk4 are evolutionarily conserved across various species, although only a few are also found across all six Slitrk family members; (2) most of the substitutions caused retention in intracellular compartments, such as the ER and cis-Golgi, with an accompanying loss of normal Slitrk glycosylation patterns; (3) none of the substitutions perturbed ligand-binding properties, judging from previously determined complex structures; and (4) all but one of the substitutions abolished the effects of Slitrks on synapse formation, with Slitrk2[V89M] uniquely exhibiting the ability to trigger presynaptic differentiation. Notably, none of the Slitrk mutations described here abolished binding to LAR-RPTPs per se. However, a majority of Slitrk mutations are positioned in the LRR2 domain of Slitrks (**Figure 1**); thus, the possibility that LRR2 domain-mediated molecular interaction(s) might be influenced by Slitrk mutations cannot be excluded at this point. At present, precisely how Slitrks promote distinct types of synapse development in an isoform-dependent manner remains to be elucidated (Yim et al., 2013), but it would be worthwhile investigating the relationship between altered synapse numbers induced by Slitrk missense mutations and the development of associated neuropsychiatric disorders in follow-up studies. Intriguingly, Slitrk family proteins resemble neuroligin family proteins in many ways (Brose, 2013). For example, the action of Slitrk3 at inhibitory synapses is phenomenologically analogous to that of neuroligin-2, whereas the effects of the other Slitrk family members are related to those of neuroligin-1. However, more rigorous analyses should be undertaken to provide a complete understanding of how synaptogenic adhesion molecules operate cooperatively, competitively, or both.

Most Slitrk substitutions represent loss-of-function mutations that perturb the normal folding and glycosylation of Slitrks, seemingly similar to the situation previously described for the R87W substitution in human neuroligin-4 and D1129H substitution in human CNTNAP2 (Zhang et al., 2009; Falivelli et al., 2012). The V89M substitution is somewhat analogous to previous descriptions of the R451C substitution in human neuroligin-3 (Comoletti et al., 2004; De Jaco et al., 2010) in that Slitrk2[V89M] protein retained WT-like presynapseinducing activity (**Figure 7**). However, this substitution acted as dominant-negative mutation, abolishing the ability of Slitrk2 to promote synapse formation in transfected neurons because of its ability to alter the synaptic properties of neurons (**Figure 8**). Intriguingly, our preliminary analyses indicate a weak correlation between dendritic spine density and synapsin puncta density in Slitrk-deficient neurons (Data not shown). Further investigation using membrane-anchored GFP plasmid to more accurately visualize the dendritic spines would be required to solve the seemingly discrepancy between postsynaptic spine density and presynaptic marker density.

Currently, we have no insight into how the V89M substitution causes inactivation of Slitrk2 function, but a similar valine-to-methionine mutation was previously described for BDNF (brain-derived neurotrophic factor) (Bath and Lee, 2006; Chen et al., 2006). The human BDNF[V66M] polymorphism, which is associated with altered dendritic trafficking of BDNF mRNA, changes in hippocampal volume, impaired hippocampal-dependent memory and NMDA-receptor dependent synaptic plasticity, and extinction of conditioned aversive memory, has been implicated in anxiety disorders (Chiaruttini et al., 2009; Yu et al., 2009; Ninan et al., 2010). The pathophysiological mechanism postulated for the effect of this BDNF mutation is that sortilin directly interacts with WT BDNF, but not BDNF[V66M], in secretory granules and controls the pathway that regulates its secretion in neurons (Chen et al., 2005). However, Slitrk2[V89M] does not block the transport of Slitrk2 protein to the cell surface; thus, it is conceivable that other mechanism(s) may operate in this case. Slitrk1[N400I] was previously shown to be unable to stimulate neurite outgrowth (Ozomaro et al., 2013), and the current study also found that its glycosylation, surface transport, and synaptogenic activities were impaired (**Figures 3–8**). In contrast, Slitrk1[T418S] was not previously examined, because it was frequently detected in individuals without OCDs and thus was considered to be functionally tolerated in the general population (Ozomaro et al., 2013). However, our data clearly suggested that the T418S mutation also abolishes the surface transport of Slitrk1 proteins, possibly by impairing glycosylation patterns, similar to the biochemical and cellular phenotypes of the N400I mutation. More intriguingly, the Slitrk1[T418S] mutation also compromised the enhanced neurite outgrowth activity shown by Slitrk1 WT (**Figure 10**).

Apart from the unique action of the Slitrk2 V89M mutation, the other Slitrk1, Slitrk2, and Slitrk4 mutations described in the current study have features in common that are typical of loss-of-function mutations, consistent with the phenotypes of Slitrk KD in cultured hippocampal neurons (Yim et al., 2013). Remarkably, Slitrk1-knockout (KO) mice exhibit elevated anxiety-like behaviors that are attributable to increased norepinephrine levels, and administration of the centrally acting adrenergic agonist clonidine has been shown to normalize these behaviors (Katayama et al., 2010). However, the core symptoms of Tourette syndrome and trichotillomania, such as self-grooming, were not reported in Slitrk1-KO mice (Katayama et al., 2010). In the case of neuroligin-3, the R451C mutation exhibits distinct gain-of-function synaptic phenotypes compared with neuroligin-3-KO mice, but both neuroligin-3 R451C knock-in (KI) and KO also impair tonic endocannabinoid signaling in specific interneuron-type synapses (Foldy et al., 2013). Thus, a worthwhile task would be to compare the behavioral, physiological, and circuit phenotypes of Slitrk-KI mice harboring some of the point mutations with altered Slitrk functions described in the current study with those obtained from Slitrk-KO mice. These approaches hold the potential of unveiling more detailed pathophysiological mechanisms underlying Slitrkassociated neuropsychiatric disorders.

#### AUTHOR CONTRIBUTIONS

fnmol-09-00104 October 18, 2016 Time: 14:46 # 17

JK and JWU conceived and supervised the project; HK, KAH, SYW, and JWU performed the experiments; HK, KAH, HMK, YHL, JK, and JWU analyzed the data; JK and JWU wrote the paper. All authors were involved in drafting the paper and provided final approval of the version to be submitted.

#### REFERENCES


#### FUNDING

This study was supported by a grant from the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT and Future Planning (2016R1A2B2006821 to JK), a faculty research grant from Yonsei University College of Medicine (6-2016-0032 to JWU); and a grant from the Korean Healthcare Technology R&D Project, Ministry for Health & Welfare Affairs, Republic of Korea (HI15C3026 to JWU).

## ACKNOWLEDGMENTS

We thank the members of Ko and Um laboratories for critical comments on the manuscript.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Kang, Han, Won, Kim, Lee, Ko and Um. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Bioenergetic Failure in Rat Oligodendrocyte Progenitor Cells Treated with Cerebrospinal Fluid Derived from Multiple Sclerosis Patients

Deepali Mathur1,2, Angela L. Riffo-Campos2,3, Josefa Castillo<sup>2</sup> , Jeffery D. Haines<sup>4</sup> , Oscar G. Vidaurre<sup>4</sup> , Fan Zhang<sup>4</sup> , Francisco Coret-Ferrer<sup>5</sup> , Patrizia Casaccia<sup>4</sup> , Bonaventura Casanova<sup>6</sup> and Gerardo Lopez-Rodas<sup>2</sup> \*

<sup>1</sup> Department of Functional Biology, University of Valencia, Valencia, Spain, <sup>2</sup> Department of Biochemistry and Molecular Biology, INCLIVA Biomedical Research Institute, University of Valencia, Valencia, Spain, <sup>3</sup> Laboratory of Molecular Pathology, Faculty of Medicine, University of La Frontera, Temuco, Chile, <sup>4</sup> Department of Neuroscience, Genetics and Genomics, Icahn School of Medicine at Mount Sinai, New York, NY, United States, <sup>5</sup> Hospital Clínico Universitario de Valencia, Valencia, Spain, <sup>6</sup> CSUR-Esclerosi Múltiple, Unitat Mixta d'Esclerosi Múltiple i Neurorregeneració del'IIS-La Fe, Hospital Universitari i Politécnic La Fe, Valencia, Spain

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Stefania Ceruti, Università degli Studi di Milano, Italy Robert Weissert, University of Regensburg, Germany

> \*Correspondence: Gerardo Lopez-Rodas gerardo.lopez@uv.es

Received: 23 March 2017 Accepted: 03 July 2017 Published: 18 July 2017

#### Citation:

Mathur D, Riffo-Campos AL, Castillo J, Haines JD, Vidaurre OG, Zhang F, Coret-Ferrer F, Casaccia P, Casanova B and Lopez-Rodas G (2017) Bioenergetic Failure in Rat Oligodendrocyte Progenitor Cells Treated with Cerebrospinal Fluid Derived from Multiple Sclerosis Patients. Front. Cell. Neurosci. 11:209. doi: 10.3389/fncel.2017.00209 In relapsing-remitting multiple sclerosis (RRMS) subtype, the patient's brain itself is capable of repairing the damage, remyelinating the axon and recovering the neurological function. Cerebrospinal fluid (CSF) is in close proximity with brain parenchyma and contains a host of proteins and other molecules, which influence the cellular physiology, that may balance damage and repair of neurons and glial cells. The purpose of this study was to determine the pathophysiological mechanisms underpinning myelin repair in distinct clinical forms of MS and neuromyelitis optica (NMO) patients by studying the effect of diseased CSF on glucose metabolism and ATP synthesis. A cellular model with primary cultures of oligodendrocyte progenitor cells (OPCs) from rat cerebrum was employed, and cells were treated with CSF from distinct clinical forms of MS, NMO patients and neurological controls. Prior to comprehending mechanisms underlying myelin repair, we determine the best stably expressed reference genes in our experimental condition to accurately normalize our target mRNA transcripts. The GeNorm and NormFinder algorithms showed that mitochondrial ribosomal protein (Mrpl19), hypoxanthine guanine phosphoribosyl transferase (Hprt), microglobulin β2 (B2m), and transferrin receptor (Tfrc) were identified as the best reference genes in OPCs treated with MS subjects and were used for normalizing gene transcripts. The main findings on microarray gene expression profiling analysis on CSF treated OPCs cells revealed a disturbed carbohydrate metabolism and ATP synthesis in MS and NMO derived CSF treated OPCs. In addition, using STRING program, we investigate whether gene–gene interaction affected the whole network in our experimental conditions. Our findings revealed downregulated expression of genes involved in carbohydrate metabolism, and that glucose metabolism impairment and reduced ATP availability for cellular damage repair clearly differentiate more benign forms from the most aggressive forms and worst prognosis in MS patients.

Keywords: multiple sclerosis, neuromyelitis optica, myelin repair, glucose metabolism, gene expression, cerebrospinal fluid, oligodendrocyte progenitor cells

#### INTRODUCTION

fncel-11-00209 July 17, 2017 Time: 14:13 # 2

Multiple sclerosis (MS) is a debilitating disease of the central nervous system (CNS) affecting the quality of life in mainly young people. Inflammatory demyelination, axonal degeneration and gliosis constitute major pathological hallmarks of MS. Although there are a variety of pharmaceutical drugs available to prevent relapses and slow down progression in MS, lack of proper understanding of its pathogenesis poses a great challenge for a complete cure of disease. During the relapsing-remitting MS subtype (RRMS), the patient's brain itself is capable of repairing the damage, remyelinating the axon and recovering the neurological function. Furthermore, CSF is in close proximity with brain parenchyma and contains proteins and other factors, which may influence the cellular physiology of brain cells. Indeed, CSF is a promising biofluid in the search for biomarkers and disease associated proteins in MS, both with respect to inflammatory and neurodegenerative processes. Studies revealed that exposure of human CSF in xenogeneic models cause neurotoxicity in culture, although the molecular mechanisms remained poorly understood (Xiao et al., 1996; Alcázar et al., 2000). Our group found that ceramides present in CSF derived from MS patients impair neuronal bioenergetics in rat neuronal cultures (Vidaurre et al., 2014). Increased energy requirement following Na+/K+ ATPase redistribution in demyelinated axons (Waxman, 2006) and disturbed metabolic or trophic support due to damaged oligodendrocytes (Lee et al., 2012) is imputable to axonal injury which causes irreversible permanent neurological deficits in MS patients.

Several lines of evidence have indicated that metabolic disturbances contribute to the pathogenesis of neurodegenerative diseases including Alzheimer's and Huntington's disease (Mazzola and Sirover, 2001; Senatorov et al., 2003; Soucek et al., 2003; Blalock et al., 2004; Brooks et al., 2007; Lin et al., 2009) or MS (Jones et al., 1950; Smith and Lassmann, 2002; Mahad et al., 2008; Iñarrea et al., 2013; Broadwater et al., 2011; Safavizadeh et al., 2013; Mathur et al., 2014). Furthermore, a number of gene expression studies have been conducted in peripheral mononuclear white blood cells (Der et al., 1998; Wandinger et al., 2001; Bomprezzi et al., 2003; Koike et al., 2003; Hong et al., 2004), in MS brain tissues (Becker et al., 1997; Whitney et al., 1999; Chabas et al., 2001) and in CSF (Brynedal et al., 2010). In amyotrophic lateral sclerosis (ALS), a neurodegenerative disorder, mitochondrial deformities have been reported by several studies (Afifi et al., 1966; Atsumi, 1981; Sasaki and Iwata, 1996; Siklós et al., 1996). Furthermore, several investigations have revealed changes in the mitochondrial electron transport chain of ALS patients. These include decreased complex I activity and cytochrome c oxidase activity in skeletal muscle, spinal cords and motor cortex of ALS patients (Bowling et al., 1993; Fujita et al., 1996; Wiedemann et al., 1998; Borthwick et al., 1999). ALS patients who are in their benign stage showed an elevated oxidative phosphorylation capacity of skeletal muscle mitochondria whereas patients in their progressive stage showed a significantly reduced activity of muscular mitochondrial respiratory complex IV (Echaniz-Laguna et al., 2006). These studies suggest perturbation in mitochondrial-associated pathways and eventually a variation in energy generation process. Gonzalez de Aguilar et al. (2005) have shown that the rate of glucose uptake, oxygen consumption, and lactate formation is increased significantly in skeletal muscles of ALS patients indicating disturbance in glucose metabolic pathways. Moreover, the levels of glucose within the CNS are significantly decreased in ALS patients (Dalakas et al., 1987). Apart from identifying defects in glucose metabolism in ALS, Dodge et al. (2013) found abnormalities in glycogen metabolism in spinal cords of ALS patients. All of these case studies reveal significant variations in carbohydrate metabolism in ALS, which may be related to the underlying pathology of ALS.

In support of human studies, animal mouse models of ALS have also displayed metabolic dysfunctions. In particular, mutant SOD1 mice model contains a modified structure of mitochondria, as observed in motor neurons (Dal Canto and Gurney, 1995; Wong et al., 1995; Sasaki and Iwata, 1996; Kong and Xu, 1998). Furthermore, the activity of complex I of electron transport chain is significantly reduced in SOD1G93A mice suggesting mitochondrial abnormality (Jung et al., 2002; Mattiazzi et al., 2002). Other investigations have also revealed impaired electron transport chain and defects in ATP synthesis in spinal cords of mutant SOD1 mice (Mattiazzi et al., 2002). It was found that glucose levels and ATP generation declined to a great extent within CNS tissues in SOD1G93A mice (Browne et al., 2006). Other neurodegenerative disorders, such as Alzheimer's disease, have also shown aberrations in mitochondrial DNA and in associated enzymes (Zhu et al., 2006). In this study, mitochondrial DNA and cytochrome oxidase-1 levels are elevated in hippocampal neurons, compared to control brains, even though the number of mitochondria per neuron is declined. Altogether, evidence from these studies indicates that glucose metabolism, oxidative balance, and ATP production are extensively impaired in ALS and other neurodegenerative diseases.

Findings from our previous work demonstrated that Gapdh, a commonly used reference gene, showed downregulated expression when cerebellar granule neurons were treated with CSF obtained from distinct clinical types of MS and NMO

patients (Mathur et al., 2015). Gapdh is a key glycolytic enzyme involved mainly in the production of ATP. Standard reference genes such as Gapdh and b-actin have been demonstrated to show variable expression in different experimental conditions (Zhong and Simons, 1999; Torres et al., 2003; Toegel et al., 2007; Gubern et al., 2009; Nelissen et al., 2010). We therefore, investigated the expression stability of six reference genes in OPCs treated with CSF from MS and NMO patients, using geNorm and NormFinder algorithms. Same reference genes that we studied previously in neuronal analysis (Mathur et al., 2015) were selected. geNorm program defines the gene stability as the average pairwise variation of a particular gene with all other reference genes and ranks the genes according to their average expression stability (M). The gene with minimum M-value is considered to be highly stable whereas the gene with highest M-value is least stable. The geNorm algorithm is based on the assumption that the reference genes selected for analysis are not co-regulated. Keeping this in mind, we selected genes that were regulated differently in order to elude unbiased results. Another program, NormFinder, ranks the candidate reference genes based on the combined estimates of both intra- and intergroup variations. In general, more than one reference genes are recommended to use for the correct normalization of gene expression data (Zhong and Simons, 1999; Hamalainen et al., 2001; Tricarico et al., 2002; Vandesompele et al., 2002; Ohl et al., 2005).

The goal of the present study was to investigate the effect of CSF on OPCs, which could contain factors that damage OPCs during attempts at brain repair. Additionally, we wanted to assess the expression stability of various commonly used reference genes when rat OPCs were treated with CSF from MS and NMO patients. This step would enable us to accurately normalize target mRNA transcripts in gene expression experiments.

#### MATERIALS AND METHODS

#### Study Approval

All procedures were approved by the Committee of Animal Care of Prince Felipe Research Center (CIPF, Valencia,) in accordance with the regulations of the European Union and Spanish legislations, within the expedient PS09/00976 of the Institute of Health Carlos III. Written informed consent was obtained from all the patients and authorized by the Ethical Committee of the Hospital Universitario y Politecnico La Fe and Hospital Clínico Universitario de Valencia for this research (Mathur et al., 2015).

#### Patient Cohort

A total of 59 patients were recruited and CSF samples were obtained from the Department of Neurology, Hospital Universitario y Politécnico La Fe and Hospital Clínico Universitario de Valencia. Out of 59 patients, 21 had inflammatory MS (11 G+/M+ and 10 G+/M−), 8 had medullary subtype (Med), 11 had primary progressive MS (PPMS), 9 had NMO, and 10 were non-inflammatory neurological controls (NIND patients) (**Table 1**). In CSF, apart from factors related to MS or NMO, there are factors from other diseases that produce their action. This must be considered as "background noise" as average population. Mixing of total CSF samples in all clinical forms may potentiate the factors related to MS. Therefore, samples from patients suffering from the same form of MS (e.g., G+/M−, G+/M+, medullary, PPMS or controls) were pooled together, then several OPCs cultures were treated with the CSF mix, and the RNA were extracted from the culture (see below).

Multiple sclerosis patients were defined and grouped in different clinical courses, according to the current criteria (Lublin and Reingold, 1996) and diagnosed according to McDonald criteria. They all met the following characteristics: oligoclonal IgG bands (OCGB) present, not in a phase of relapse, and have spent more than a month after the last dose of steroids. Wingerchuk et al. (2006) criteria were used to diagnose patients with NMO disease. Patients suffered relapses of optic neuritis and myelitis, and two of the three criteria, normal MRI or that did not accomplish the Patty criteria for MRI diagnosis of MS.

#### Patient Characteristics

Inflammatory MS (RRMS and SPMS forms): MS is categorized into: (1) Relapsing remitting MS (RRMS) that later develops into secondary progressive stage (SPMS); and (2) primary progressive MS (PPMS). Over 95% of patients with MS show oligoclonal bands (OCBs) of IgG in CSF (G+) (Kostulas et al., 1987) and 40% show IgM OCBs in CSF (M+) related to a more aggressive course of disease (Sharief and Thompson, 1991). In this study we also classified and named inflammatory MS into "G+/M−" and "G+/M+" subtypes (see below) on the basis of aggressivity and prognosis that is more complete than just RRMS or PPMS. In addition, we have studied separately a set of patients with MS but with a predominant affectation of the spinal cord, because these patients have some peculiarities, and we wanted to explore if they have some differences in light of our experiments. The most aggressive cases termed as "medullary" have more spinal injuries.

Relapsing-remitting multiple sclerosis (RRMS) IgG+/IgM− clinical form of MS: Patients named as "G+/M− subtype" had IgG antibodies (+) but no IgM (−) oligoclonal antibodies detected in the CSF of brain. RRMS IgG+/IgM+ clinical form of MS: Patients named as "G+/M+ subtype" had both IgG antibodies (+) and IgM (+) oligoclonal antibodies detected in the CSF of brain. Medullary (Med) clinical form of MS: All these patients were positive for OCGBs and negative (most frequent) or not for oligoclonal IgM bands (OCMBs) in CSF as well as presence of diffuse hyperintensity signal in the spinal cord and with mainly relapses from this location. The patients accomplished also Swanton's criteria for dissemination in time. Primary progressive MS (PPMS): These patients are characterized by progressive decline in neurological disability. Neuromyelitis optica (NMO) patients: Individuals that met at least two of the following three features. (1) Long extensive transverse myelitis (>3 vestibule bodies); (2) Antibodies against aquaporin-4; (3) Normal brain at the first event Controls [Non-Inflammatory Neurological Diseases (NIND)]: Individuals who were suspected to have MS but after protocolized analysis were not diagnosed with MS.

#### TABLE 1 | Clinical characteristics of the patients.

fncel-11-00209 July 17, 2017 Time: 14:13 # 4


(Continued)

#### TABLE 1 | Continued

fncel-11-00209 July 17, 2017 Time: 14:13 # 5


Case, patient number; G, gender; F, female; M, male; Age, patient age in years; E.T., evolution time in years; Actual EDSS, Expanded disability status scale; OCGB IgG, oligoclonal IgG bands in CSF; OCMB IgM, oligoclonal IgM bands in CSF; Ac−AQ4, Antiaquaporin 4 antibodies; T, Treatment. Notations: RRMS, relapsing-remitting multiple sclerosis; SPMS, secondary progressive multiple sclerosis; CIS, clinically isolated syndrome; MedMS, medullary MS; PPMS, primary progressive multiple sclerosis; NMO, neuromyelitis optica patients; G+/M−, presence of IgG but no IgM antibodies in the CSF; G+/M+, presence of both IgG and IgM antibodies in the CSF; N.A., not applicable; N.T., no treatment at the time of collection of the samples; P, presence of Ac-AQ4; N., non-detected Ac-AQ4; IFN, interferon; FGM, fingolimod; MTZ, mitozantrone; U., unavailable; CPX, copaxone; NTZ, natalizumab; IV-IgG, bi-monthly pulsed intravenous immunoglobulin; PE, plasma-exchange 2 months before lumbar puncture; ASCT, autologous stem cell transplant 1 year before LP; Cy, Cyclophosphamide; RTX, rituximab 5 months before lumbar puncture.

# Cerebrospinal Fluid Samples of Patients

Cerebrospinal fluid samples were obtained by lumbar puncture at the time of diagnosis. No patient had received treatment with immunosuppressive drugs, immunomodulators or corticosteroids for at least 1 month prior to the extraction of CSF. The routine clinical extraction of CSF from patients was 10 mL obtained by lumbar puncture in subarachnoid space under sterile conditions every 3–6 months. The samples were centrifuged for 10 min at 700 × g and aliquots were frozen and stored at −80◦C in 1 ml aliquots until use. To preserve the integrity of the samples, the aliquots were used just once in an experiment without re-frozen.

# Primary Cultures of OPCs from the Neonatal Rat Brain

The OPCs were isolated from the cortex of postnatal day 1 Wistar rats (Harlan Iberica) and grown for 1 week as mixed glial cultures according to a modified McCarthy and deVellis procedure (McCarthy and de Vellis, 1980). Differential shaking, followed by magnetic beads immunoselection, was used to isolate progenitors (Haines et al., 2015). Briefly, cells were incubated with the A2B5 antibody, purified using magnetic beads (Miltenyi Biotec), and then plated at a density of 2 × 10<sup>5</sup> cells/ml on poly-Dlysine coated 10 cm plates. OPC cells were allowed to proliferate for 48 h in chemically defined media (ODM) supplemented with growth factors (GFs) [20 ng/mL basic fibroblast growth factor (bFGF) and 10 ng/mL platelet-derived growth factor AA (PDGF-AA)] prior to treatments indicated below. This procedure led to a 95% pure population of A2B5+ cells that expressed Myelin and GalC.

# CSF Treatment and RNA Extraction

Treatment of OPC was conducted by culturing the cells in a 1:1 dilution of CSF and chemically defined medium supplemented with growth factors to allow a total concentration of 20 ng/ml bFGF and 10 ng/ml PDGF-AA. In these conditions OPC remain proliferative and this does not interfere with their survival or differentiation. We performed an MTT cellular toxicity assay to determine whether CSF-exposed OPCs were viable, and no difference was detected amongst the CSF groups tested (p > 0.05) (Haines et al., 2015).

The RNA was extracted from OPCs treated cells using the Qiagen RNeasy RNA extraction kit according to the

manufacturer's instructions. The RNA concentration was determined spectrophotometrically at 260 nm using the Nanodrop 1000 spectrophotometer (V3.7 software) and the quality of every RNA sample was measured by means of the absorbance ratio at 260/280 nm and by capillary electrophoresis using 2100 Bioanalyzer instrument (Agilent). For microarray assays (see below), we used samples with OD (A260/A280) ratio value > 1.8 and an OD (A260/A230) ratio > 1.9 and a minimum RNA Integrity Number (RIN) score of 9.8 (mainly 10) in the Total RNA Nano Series.

#### Selection of Reference Genes

We selected a panel of reference genes from microarray data that we used in neuronal analysis (Mathur et al., 2015), to normalize expression of target mRNA transcripts. From the microarray data we determined the changes of gene expression in the different MS and NMO patients, and evaluated their expression stability using geNorm and NormFinder algorithms. Candidate reference genes, included β-actin (ActB), hypoxanthine guanine phosphoribosyl-transferase (Hprt), mitochondrial ribosomal protein L19 (Mrpl19), transferrin receptor (Tfrc), microglobulin β2 (B2m), and glyceraldehyde-3-phosphate-dehydrogenase (Gapdh), were selected for evaluation. The function and references of the genes are listed in **Table 2**. Primer sequences and amplification summary are listed in **Table 3**.

#### Determination of Reference Gene Expression Stability

To determine the stability of these genes, we employed publicly available software tools named geNorm and NormFinder. A statistical test was applied to look for significant differences between experimental conditions for each candidate reference gene. A one-way analysis of variance (ANOVA) was conducted to determine the significantly variable genes. A p-value < 0.05 was considered statistically significant.

#### Gene Microarray, Data Normalization and Gene Validation

Isolated RNA from OPC treated cells were subjected to one color microarray-based gene expression analysis (Agilent Technologies). The labeled cRNA was hybridized to the Agilent SurePrint G3 Rat GE 8x60K Microarray (GEO-GPL13521, in situ oligonucleotide), according to the manufacturer's protocol. We used four microarrays per MS type of patients. Briefly, the mRNA was reverse transcribed in the presence of T7-oligo-dT primer to produce cDNA. cDNA was then in vitro transcribed with T7 RNA polymerase in the presence of Cy3-CTP to produce labeled cRNA. The labeled cRNA was hybridized to the Agilent Sure Print G3 Rat GE 8x60K Microarray according to the manufacturer's protocol. The arrays were washed, and scanned on an Agilent G2565CA microarray scanner at 100% PMT and 3 µm resolution. The intensity data was extracted using the Feature Extraction Software (Agilent). 75th percentile signal value was used to normalize Agilent one-color microarray signals for inter-array comparisons. After normalization, the data was filtered in order to exclude probesets with low expression and/or affected by differences between the laboratories.

Differentially expressed genes were identified by comparing average expression levels in MS and NMO cases and controls. mRNA expression (in terms of absolute fold change) in OPCs treated with CSF of MS and NMO individual patients was compared with gene expression in OPC exposed to CSF from neurological control patients. Fold change cut off was considered as 2. The microarrays data correspond to at least 3–4 independent assay per group of individual MS or NMO patients and controls.

#### Statistical Analysis

Statistical analysis was conducted after background noise correction using NormExp method. Differential expression analysis was carried out on non-control probes with an empirical Bayes approach on linear models (Smyth, 2004). Results were corrected for multiple testing hypothesis using false discovery rate (FDR), and all statistical analyses were performed with the Bioconductor project<sup>1</sup> in the R statistical environment<sup>2</sup> . To filter out low expressed features, the 30% quartile of the whole array was calculated, and probe sets falling below threshold were filtered out. After merging the probes corresponding to the same gene on the microarray, the statistical significance of difference in gene expression were assessed using a standard one-way ANOVA followed by Tukey's HSD post hoc analysis (cut-off p < 0.01 and FDR <0.1). All data processing and analysis including PCA plot was carried out using R functions.

#### Analysis of Gene–Gene Interaction Networks Using String v10 Software

We used STRING v10 software and correlated the gene interaction in different disease subtypes in OPCs (Szklarczyk et al., 2015; STRING database)<sup>3</sup> . We defined a parameter to "integrate" our data as "Cumulative Flux Index or CFI" within the network to compare our experimental MS conditions.

The values mean that the reduction of local flux due to the inhibition of an enzymatic activity in a specific gene affected synergically to the whole metabolic flux network. It means that as more genes were down regulated, the total flux was reduced as a cumulative factor that we integrated as the total CFI of the network.

### RESULTS

#### Demographic and Clinical Profiles of MS, NMO, and NIND Groups

Baseline characteristics of the study population are described in **Table 4**. Prevalence of MS was found more in women (75%) than in men. Mean age of MS patients was 30.7 ± 9.7 years whereas 25.6 ± 15 years for NMO patients. According to the classical clinical classification, that only takes care of the general

<sup>1</sup>http://www.bioconductor.org/

<sup>2</sup>http://cran.r-project.org/

<sup>3</sup>http://string-db.org



TABLE 3 | Primer sequences and amplification summary.

TABLE 5 | Characteristics of MS patients according to clinical classification.


F, forward primer; R, reverse primer. Annealing temperature was 60◦ in whole amplicons.



characteristics of MS patients, and classified as RRMS, SPMS, and PPMS, there were significant differences observed in age at the beginning of PPMS and the other two MS forms (p < 0.003); Expanded Disability Status Scale (EDSS) of RRMS and the two other MS forms (p < 0.001); evolution time from the first to the second episode between PPMS and RRMS (p = 0.043) (**Table 5**). In patients experiencing a progressive course, evolution time was similar in secondary progressive cases and in cases that were progressive from onset (13.5 vs. 13.8) (**Table 5**).

In the **Table 6** we organize the MS patients paying attention to data we obtained in which the presence of CSF-restricted IgM OCB was associated with an active inflammatory disease


phenotype in PPMS patients with more active inflammatory disease (Villar et al., 2014). With this working classification, we found significant differences in the age at beginning between PPMS and the other two MS forms (RRMS and SPMS) (p < 0.003) (**Table 6**). People with PPMS are usually older at the time of diagnosis, with an average age of 40. Furthermore, different subtypes of MS help to predict disease severity and response to treatment hence their categorization is important. Moreover, we found significant differences in EDSS between RRMS and the two other MS forms (SPMS and PPMS) (p < 0.001) (**Table 6**). Although nerve injury always occurs, the pattern is specific for each individual with MS. Disease severity and disability increases from RRMS to SPMS course and in PPMS subtype, symptoms continually worsen from the time of diagnosis rather than having well-defined attacks and recovery. PPMS usually results in disability earlier than relapsing-remitting MS.

#### Identification of Stably Expressed Reference Genes in Treated OPCs

According to geNorm algorithm, Mrpl19 and Hprt1 were identified in the microarray analysis as the best reference genes, followed by B2m (average M-value: 0.102 for Mrpl19 and Hprt1 genes and 0.147 for B2m gene). Surprisingly, ActB and Gapdh, mostly used as standard housekeeping gene for normalization, showed the most unstable gene expression in OPCs that were exposed to the CSF of our experimental conditions. The use of NormFinder software with the data showed that Tfrc and B2m were identified as the best reference genes followed by ActB (average M-value: 0.033 for Tfrc and B2m genes and 0.087 for ActB gene). Mrpl19 and Gapdh showed the least stable



TABLE 7 | Candidate reference genes ranked in OPCs exposed to the CSF of RRMS and PPMS patients according to their expression stability by geNorm and NormFinder methods.


M: Average expression stability. Lower M-value indicates more stable expression while the highest M-value indicates variable expression.

gene expression in OPCs that were exposed to the CSF of our experimental conditions. **Table 7** depicts candidate reference genes ranked in OPCs exposed to the CSF of RRMS and PPMS patients according to their expression stability by geNorm and NormFinder methods.

The expression of β-actin (ActB) gene was downregulated significantly, showing 37 and 42% gene expression in OPCs exposed to the CSF of G+/M+ and medullary MS as compared to control. The expression was downregulated by 37% in OPCs exposed to the CSF of NMO patients as compared to control (**Figure 1A**). A marked fluctuation in ActB gene expression was seen in OPCs exposed to the CSF of various experimental conditions. geNorm also identified ActB as highly variable gene in these experimental conditions. We conclude that this gene is not suitable to normalize gene transcripts in treated OPCs.

The expression of β-2 microglobulin (B2m) gene was downregulated significantly, showing 50, 48, and 58 % in OPCs exposed to the CSF of G+/M−, G+/M+ and medullary MS as compared to control. Similarly, significantly reduced expression with 43% in OPCs exposed to the CSF of NMO patients was observed as compared to control (**Figure 1B**). geNorm identified B2m as the third most stable gene hence this gene can be used for normalization purpose in treated OPCs.

The expression of hypoxanthine phosphoribosyltransferase (Hprt1) gene was downregulated significantly by 49, 38, and 49% in OPCs exposed to the CSF of G+/M−, G+/M+ and medullary MS as compared to control. Similarly the expression was around 52% significantly reduced in OPCs exposed to the CSF of PPMS and NMO patients as compared to control (**Figure 1C**). geNorm identified Hprt1 as the most stable gene hence this gene can be used for normalization purpose in treated OPCs.

The expression of mitochondrial ribosomal protein L19 (Mrpl19) gene was downregulated significantly by 52, 46, and 48% in OPCs exposed to the CSF of G+/M−, G+/M+ and medullary MS as compared to control. Similarly the expression was 52 and 45% lower in OPCs exposed to the CSF of PPMS and NMO patients as compared to control (**Figure 1D**). geNorm identified Mrpl19 as the most stable gene hence this gene can be used for normalization purpose in treated OPCs.

Similarly, reduced variation was observed in transferrin receptor (Tfrc) gene expression across all experimental conditions (**Figure 1E**).

The data indicates that the expression of glyceraldehyde 3 phosphate dehydrogenase (Gapdh) gene was downregulated by 44, 54, and 49% in OPCs treated with G+/M−, G+/M+ and medullary clinical form of MS as compared to OPCs exposed to the CSF of neurological controls. Similarly, the expression was reduced by 55 and 44% in OPCs treated with the CSF of PPMS and NMO patients as compared to OPCs exposed to the CSF of non-inflammatory neurological controls (NIND) (**Figure 1F**). According to geNorm and NormFinder algorithms, Gapdh was ranked as an unstable gene for normalizing mRNA transcripts.

We conclude from this data that Mrpl19, Hprt, Tfrc, and B2m should be used to normalize the gene transcripts in experiments related to the current one, without much differences between them, and the use of ActB and Gapdh should be avoided.

#### Differential Expression of Genes Involved in Glucose Metabolism in OPCs by Microarray Gene Expression Profiling

Our findings in the microarray analysis revealed that genes involved in carbohydrate metabolism were differentially expressed in our experimental conditions. **Figures 2**–**4** shows expression of genes involved in glycolytic pathway; TCA cycle; and oxidative phosphorylation in OPCs treated with CSF derived from MS and NMO patients normalized to gene expression in OPCs treated with CSF derived from non-inflammatory neurological controls. We have plotted the expression of genes in our experimental conditions, normalized with respect to neurological control, calculated from the absolute fold change values from microarray data.

When OPCs were exposed to CSF of G+/M− RRMS patients, the expression of several glycolytic genes including Hk (hexokinase), Gpi (Glucose-6-phosphate isomerase), Gapdh (Glyceraldehyde 3-phosphate dehydrogenase), Tpi (Triosephosphate isomerase), Pgk1 (Phosphoglycerate kinase 1), Eno1 (Enolase 1), Eno2 (Enolase 2), Pk (Pyruvate kinase), and Pkm2 (pyruvate kinase M2) was found to be decreased (**Figures 2A,B,D–F,H–K**). Furthermore, genes implicated in TCA cycle including Pdha1 (pyruvate dehydrogenase alpha 1), Aco2 (Aconitate hydratase), Idh (Isocitrate dehydrogenase), and Ogdh (Oxoglutarate dehydrogenase) were downregulated (**Figures 3A–D**). Similarly, genes involved in oxidative phosphorylation, namely ATP5a1 (ATP synthase subunit alpha 1), ATP5b (ATP synthase subunit alpha 5b), MT-ND2 (mitochondrial encoded NADH dehydrogenase 2), and Cyc1 (cytochrome c1), showed decreased expression as compared to neurological controls (**Figures 4A,B,D,E**).

When OPCs were exposed to CSF of G+/M+ RRMS patients, it was found that most of the enzymes involved in glycolysis including Hk, Gpi, Gapdh, Tpi, Pgk1, Eno1, and Pk, (**Figures 2A,B,D–H,J**); the related TCA cycle enzymes including Pdha1, Aco2, Idh, Ogdh, Sdh (Succinate dehydrogenase), and Mdh2 (Malate dehydrogenase 2) (**Figures 3A–F**); and the mitochondrial electron chain enzymes [MT-ND2, Cyc1, Cox (Cytochrome c oxidase), ATP5a1 and ATP5b] were strongly reduced in gene expression as compared to neurological controls (**Figures 4A–E**).

of patients.

When OPCs were exposed to CSF of medullary patients, it was found that most of the enzymes involved in glycolysis including Hk, Gpi, Gapdh, Tpi, Pgk1, Eno1, Eno2, Pk, and Pkm2 (**Figures 2A,B,D–F,H–K**), the related TCA cycle enzymes including Pdha1, Aco2, Idh, Ogdh, Sdh, and Mdh2 (**Figures 3A–E**); and the mitochondrial electron chain enzymes (MT-ND2, Cyc1, Cox, ATP5a1, and ATP5b) were strongly reduced in gene expression as compared to neurological controls (**Figures 4A–E**).

In case of OPCs treated with CSF from PPMS patients, we found that most of the glycolytic genes, including Hk1, Gpi, Gapdh, Tpi, Pgk1, Eno2, Pk, and Pkm2, were reduced in gene expression (**Figures 2A,B,D–F,I–K**). Our findings also revealed down regulation of genes implicated in TCA cycle, including Pdh, Aco2, Idh, Ogdh, and Sdh (**Figures 3A–E**). Genes of ETC, including ComplexI/NADH:ubiquinone oxidoreductase, Cyc1, Complex IV/COX and ATP synthase (both alpha and beta subunits) showed down regulated gene expression (**Figures 4A–E**).

When OPCs were exposed to CSF of NMO patients, it was found that most of the enzymes involved in glycolysis including Gpi, Aldoc (Fructose 1,6-bisphosphate aldolase), Gapdh, Tpi, Pgk1, Eno1, Pgam, Eno1, Eno2, Pk, and Pkm2 (**Figures 2B–F,H–K**); the related TCA cycle enzymes including Pdha1, Aco2, Idh, Ogdh, Sdh, and Mdh2 (**Figures 3A–F**); and the mitochondrial electron chain enzymes (MT-ND2, Cyc1, Cox, ATP5a1, and ATP5b) were strongly reduced in gene expression as compared to neurological controls (**Figures 4A–E**).

Overall, the microarray data demonstrate that the genes involved in carbohydrate metabolism were differentially expressed in OPCs treated with the CSF from MS and NMO patients as compared to OPCs exposed to the CSF of neurological controls. We conclude that CSF exposure to OPCs altered the carbohydrate metabolism and may have altered the capacity of these cells to repair axonal damage in different clinical forms of MS.

## Analysis of Gene–Gene Interaction Networks Using String v10 Software

**Figure 5** illustrates a general metabolic network including glycolytic pathway, TCA cycle and electron transport chain with cumulative flux indexes. Gene–gene interaction network was visualized in OPCs exposed to CSF from G+/M− MS (**Figure 6A**); G+/M+ MS (**Figure 6B**); medullary MS (**Figure 6C**); PPMS (**Figure 6D**); and NMO patients (**Figure 6E**) generated by STRING v10. Significantly downregulated genes were indicated by blue color, and significantly upregulated genes were indicated by red color, in the STRING figure. The variation of metabolic flux, estimated with their values of CFI, in the different treatments with CSF of MS and NMO patients were integrated. CFI values were calculated as a parameter to integrate the reduced activity of the different enzymes in a specific network (glycolysis, TCA cycle and ATP generation, or together) as the expected total flux. This parameter is a simplified linear cumulative form of the flux control coefficients in the metabolic control analysis. This value roughly compares the different fluxes that may occur

in the MS patients according to the number of enzymes down-regulated (after normalizing them by housekeeping gene expression) and the enzymatic activity level of each one. **Table 8** depicts upregulated and downregulated genes in distinct MS clinical forms and NMO and their related cumulative flux index values.

With the calculation of CFI parameters, we may say that whole carbohydrate metabolic flux and ATP synthesis decreased in OPCs when exposed to CSF derived from MS and NMO patients. Our findings suggest a significant downregulation of genes involved in carbohydrate metabolism suggesting that factors present in the CSF, in our model, perturb the metabolism of OPCs.

#### DISCUSSION

In the present study, we found a downregulated expression of genes involved in carbohydrate metabolism in OPCs exposed to CSF from MS and NMO patients as compared to controls. We found that total cumulative flux index associated with glycolysis, TCA cycle and ATP generation declined to a great extent in G+/M+ RRMS patients (CFI: 1.9E-07) as compared to G+/M− RRMS subtype (CFI: 1.3E-04), where lower energy level may have reduced the repair process by the OPCs. G+/M− RRMS is the less severe and less aggressive form displaying oligoclonal bands of IgG antibodies (+) but no IgM (−) oligoclonal antibodies in the CSF of brain. On the other hand, G+/M+

clinical form of RRMS displayed oligoclonal bands of both IgG antibodies (+) and IgM (+) antibodies in the CSF of brain. This form was more aggressive with worse prognosis than G+/M− RRMS. In G+/M− patients, metabolic genes showed a downregulated expression leading to a reduction in the overall metabolic flux. This caused a decrease in ATP synthesis and overall ability of OPCs to repair the damaged neuronal cells, which can be related to poor prognosis and development of pathology in these patients. Our results are consistent with previous findings, which showed a reduction in ATP synthase expression in MS lesions (Smith and Lassmann, 2002). It may be indicated that genes involved in metabolism of carbohydrates and ATP synthesis were strongly affected in G+/M+ subtype in OPCs as compared to G+/M− subtype. The energy required to repair damaged neurons degenerated in MS lesions could be much lower in G+/M− RRMS less severe clinical form as compared to more aggressive G+/M+ RRMS. To combat oxidative stress generated in neurological diseases, the cells need to produce large quantities of reducing equivalents for energy production. However, insufficient energy production severely impaired the ability to repair nerve damage. We conclude that the differential expression of metabolic genes reduced the repairing potential by OPCs which could be related to worse prognosis in patients with G+/M+ RRMS type as compared to G+/M− RRMS.

In medullary MS, our results showed a significant reduction of several metabolic genes as compared to OPCs exposed to the CSF from neurological controls. Compared with non-MS patients (controls), total cumulative flux index in medullary MS dropped to a very low value (CFI: 7.9E-07) in all three metabolic pathways, similar to what occurred in G+/M+ RRMS. The most aggressive form was classified as medullary MS with a predominant affectation of the spinal cord. All these patients were positive for oligoclonal IgG bands (OCGB) and negative for oligoclonal IgM bands (OCMB) in CSF of spinal region. The drastic reduction of metabolic genes expression in both G+/M+ RRMS and medullary MS would result in bioenergetic failure eventually causing OPCs to reduce the repair of axonal damage which is correlated with worst prognosis compared with G+/M− RRMS.

Moreover, our findings demonstrated a reduced expression of genes involved in glucose metabolism in OPCs treated with CSF from PPMS patients. Altogether, total cumulative flux index in carbohydrates and ATP production was decreased strongly in OPCs (CFI: 1.3E-04) in this subtype of patients. Relapsing remitting MS (RRMS) is the most common form of MS affecting 80–90% of the patients. On the contrary, patients with primary progressive MS (PPMS) are characterized by a steady worsening of neurological symptoms with no relapse or remission affecting 10–15% of the patients. Our findings are consistent with the fact that PPMS patients are less aggressive, slower and have progressive course and a better prognosis than RRMS subtypes G+/M+ or medullary MS.

Finally, we observed a reduced carbohydrate metabolism and ATP synthesis in OPCs exposed to CSF of NMO patients compared to non-MS (controls). In addition, we found a decrease in the cumulative flux index accumulated in all three pathways.

The results indicate that the flux indexes accumulated in OPCs exposed to CSF of NMO patients decreased significantly (CFI: 1.1E-07). It has been documented that anti-NMO IgG does not


gene.

have a direct effect over OPCs, and it has high affinity for the astroglial receptors (Marignier et al., 2010). Despite purified cultures of OPCs were used, the 1% presence of astrocytes could explain, although unlikely, the downregulation of metabolic genes in OPCs treated with NMO derived CSF. Another, more plausible explanation (not tested), is that some NMO patients could have presented anti-MOG, a recent antibody against myelin associated oligodendrocyte, that in cases of NMO patients caused a profound oligodendropathy non-associated to astrocyte damage (Ikeda et al., 2015). Because anti-MOG is present in 40% of seronegative-NMO, and in our series we had four NMO patients with no anti-NMO antibodies, we cannot rule out the possibility that some patients presented anti-MOG.

Several studies have demonstrated the unmet need of energy requirement in MS, which include mitochondrial impairment in cultured neurons (Kim et al., 2010), animal models of MS (Nikic et al., 2011 ´ ) and in MS samples (Dutta et al., 2006). Metabolic abnormalities are implicated in the pathogenesis of neurodegenerative diseases (Henneman et al., 1954; McArdle et al., 1960; Lu et al., 2000; Dutta et al., 2006). In MS literature indicating any association between perturbed glucose metabolism with its pathogenesis is meager. Royds et al. (1981), observed an increased activity of metabolic enzymes including enolase, pyruvate kinase, lactate dehydrogenase and aldolase in the CSF of patients with disseminated sclerosis. Kölln et al. (2006) demonstrated that B cells and antibodies reactive with Tpi and Gapdh are produced intrathecally in CSF and lesions of MS. Both TPI and GAPDH are essential metabolic enzymes involved in ATP production. Another investigation by the same group showed that these antibodies bind with TPI and GADPH, and inhibit the glycolytic activity of GAPDH but not TPI in MS patients (Kölln et al., 2010).

The main findings of this study revealed a disturbed carbohydrate metabolism in OPCs treated with the CSF derived from MS and NMO subjects. Factors present in the CSF, in our model, affected the metabolism of OPCs and clearly differentiate more benign forms from the most aggressive forms in MS. The effect of CSF was different in MS aggressivity of RRMS and PPMS clinical form (G+/M−, G+/M+, Med, PPMS). G+/M+ RRMS and medullary derived CSF treated OPCs were strongly affected by reducing carbohydrate metabolism as evidenced by down regulation of most of the genes which is suggestive of least ATP synthesis. This indicates blockage of myelin repair by OPCs and correlated with worst prognosis. OPCs treated with CSF from G+/M− RRMS patients demonstrates slightly reduced carbohydrate metabolism correlated with poor prognosis.

Finally, our geNorm analysis revealed Mrpl19 and Hprt as the most stably expressed genes followed by B2m, whereas β-actin and Gapdh were least stably expressed genes when OPCs were treated with diseased CSF, as revealed by geNorm analysis. Our findings are consistent with previous findings in which Gapdh and β-actin have been demonstrated to show variable expression in different experimental conditions (Hamalainen et al., 2001; Deindl et al., 2002; Glare et al., 2002; Radonic et al., 2004; Mathur et al., 2015). The results allowed us to differentiate different clinical forms and aggressivity in MS and MS from NMO. However, it is still elusive whether these alterations in metabolic gene expression cause MS and NMO or are a consequence of the disease. These findings open new avenues of study and allow the development of therapeutic agents targeted to restore the metabolic function and hence repair and/or prevent axonal damage responsible for functional disability in the patient. A greater understanding of these impaired metabolic pathways may offer new insights into more efficacious treatments for MS and NMO.

#### ETHICS STATEMENT

All subjects gave written informed consent in accordance with the Declaration of Helsinki. The study protocol was approved by the Institutional Ethical Committee.

# AUTHOR CONTRIBUTIONS

DM conducted experiments, analyzed and wrote the manuscript. AR-C, JC, JH, OV, and FZ conducted experiments. FC-F and PC designed the study. BC and GL-R designed the study and edited and revised the manuscript.

## FUNDING

Grant support: Consolider-Ingenio CSD2006-49 (Network Group Valencia) from Ministerio de Ciencia y Tecnología, (Spain) to GL-R; Instituto Carlos III del Ministerio de Economia y Competitividad PI13/00663 to BC; and NIH-NINDS grants R01-NS69385 and R37-NS42925 to PC.

#### ACKNOWLEDGMENTS

The authors thank Dr. Sanjib K. Agarwalla and Institute of Physics, India for providing required software to make graphics. DM would also like to acknowledge the support from SERB National postdoctoral fellowship received (file no. PDF/2016/001639).

# REFERENCES

fncel-11-00209 July 17, 2017 Time: 14:13 # 16



analysis of gene expression in CNS tissues and disease models. J. Neurosci. Methods 98, 9–20. doi: 10.1016/S0165-0270(00)00178-3


data by geometric averaging of multiple internal control genes. Genome Biol. 3:RESEARCH0034. doi: 10.1186/gb-2002-3-7-research0034


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Mathur, Riffo-Campos, Castillo, Haines, Vidaurre, Zhang, Coret-Ferrer, Casaccia, Casanova and Lopez-Rodas. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Lactate Shuttles in Neuroenergetics—Homeostasis, Allostasis and Beyond

#### Shayne Mason\*

*Centre for Human Metabolomics, North-West University, Potchefstroom, South Africa*

Understanding brain energy metabolism—neuroenergetics—is becoming increasingly important as it can be identified repeatedly as the source of neurological perturbations. Within the scientific community we are seeing a shift in paradigms from the traditional neurocentric view to that of a more dynamic, integrated one where astrocytes are no longer considered as being just supportive, and activated microglia have a profound influence. Lactate is emerging as the "good guy," contrasting its classical "bad guy" position in the now superseded medical literature. This review begins with the evolution of the concept of "lactate shuttles"; goes on to the recent shift in ideas regarding normal neuroenergetics (homeostasis)—specifically, the astrocyte–neuron lactate shuttle; and progresses to covering the metabolic implications whereby homeostasis is lost—a state of allostasis, and the function of microglia. The role of lactate, as a substrate and shuttle, is reviewed in light of allostatic stress, and beyond—in an acute state of allostatic stress in terms of physical brain trauma, and reflected upon with respect to persistent stress as allostatic overload—neurodegenerative diseases. Finally, the recently proposed astrocyte–microglia lactate shuttle is discussed in terms of chronic neuroinflammatory infectious diseases, using tuberculous meningitis as an example. The novelty extended by this review is that the directionality of lactate, as shuttles in the brain, in neuropathophysiological states is emerging as crucial in neuroenergetics.

Keywords: neuroenergetics, lactate, astrocyte-neuron lactate shuttle (ANLS), astrocyte-microglia lactate shuttle (AMLS), neuropathology, traumatic brain injury (TBI), neurodegenerative disease, infectious neuroinflammatory disease

#### INTRODUCTION

The human brain represents approximately 2% of total body weight and receives up to 15% of total blood flow, consuming up to 20% of oxygen and 25% of circulating glucose under normal conditions (Pellerin, 2010). The metabolism of this high energy consuming organ involves complex intercellular trafficking of metabolites and compartmentalization of numerous processes. Tight coupling exists between the supply and demand of energy in brain metabolism with changes in cerebral blood flow and glucose utilization in response to neuronal activity (i.e., neurovascular and neurometabolic coupling) (Bélanger et al., 2011) under normal homeostatic conditions. Mechanisms involved in brain energy metabolism adapt during periods of perturbation and trauma. The important role of lactate, and its shuttles, has been overlooked and merits acknowledgement.

Edited by:

*Andrew Harkin, Trinity College, Dublin, Ireland*

#### Reviewed by:

*Claude Messier, University of Ottawa, Canada Fahmeed Hyder, Yale University, USA*

> \*Correspondence: *Shayne Mason nmr.nwu@gmail.com*

#### Specialty section:

*This article was submitted to Neuroenergetics, Nutrition and Brain Health, a section of the journal Frontiers in Neuroscience*

> Received: *31 October 2016* Accepted: *20 January 2017* Published: *02 February 2017*

#### Citation:

*Mason S (2017) Lactate Shuttles in Neuroenergetics—Homeostasis, Allostasis and Beyond. Front. Neurosci. 11:43. doi: 10.3389/fnins.2017.00043*

The interconversion of lactate and pyruvate occurs via lactate dehydrogenase, with increased lactate typically being associated with anaerobic respiration. Thus, one could postulate that elevated levels of lactate present in the cerebrospinal fluid (CSF) due to neuroinflammation could be from hypoxia caused by ischemia—increased anaerobic respiration (Rossi et al., 2007), or by raised glucose levels and hence elevated flow through the glycolysis pathway. However, in most neuroinflammatory cases there are typically periods of low levels of glucose in the CSF. Furthermore, several studies have shown no correlation between CSF lactate levels and cerebral blood flow (i.e., they are unrelated to ischemia) (Brodersen and Jorgensen, 1974; DeSalles et al., 1986; De Salles et al., 1987). Thus, an alternative postulate is that the elevated levels of lactate in the CSF of neuroinflammatory cases is unlikely to be due to anaerobic respiration but instead is possibly a product of temporarily increased flux in the glycolysis pathway—using glycogen as a supplementary source of glucose.

The notion that lactate is not simply just a by-product of glycolysis but also a shuttle system was pioneered by Brooks (1985). In June the following year, Brooks provided more evidence for a "shuttling" system for lactate (Brooks, 1986a); by December Brooks (1986b) formally coined the term "lactate shuttle" in skeletal muscle during rest and exercise under fully aerobic conditions. Brooks further defined lactate shuttles as being either intracellular (cytosolic to mitochondrial) or cell-to-cell (extracellular) (Brooks, 2000, 2002). Later, Brooks (2009) provided evidence that glycolytic and oxidative pathways should be viewed as linked, as opposed to alternative processes, because lactate, the product of one pathway, is the substrate for the other (Brooks, 2009). The concept of intracellular lactate shuttles was challenged (Sahlin et al., 2002), and continues to be so (Laureys et al., 2014). A study by Cruz et al. (2012) discussed these challenges but ultimately reiterated Brooks' sentiment regarding interaction between energy systems—the product of one is the substrate of another. Soon after Brooks first introduced the concept of lactate shuttles in 1986, several studies by Schurr et al. (1988, 1997a,b,c, 1999a,b) provided experimental evidence that astrocytic lactate is even used preferentially over glucose by neurons after episodes of cerebral ischemia.

Thus, lactate plays an important role as a shuttle, even beyond the brain—in numerous systems, for example, in pregnancy (Zuo et al., 2015), in reproduction (Kuchiiwa et al., 2011), and, notably, within the human heart (Cruz et al., 2012; Rakus et al., 2016). This paper provides an overview of brain neuroenergetics and the crucial roles of lactate shuttles. These roles will be discussed: (1) with regard to the normal physiological function and relationship between astrocytes and neurons (homeostasis)—the astrocyte– neuron lactate shuttle (ANLS) model; (2) the dynamic role of microglia and their preferential utilization of lactate under perturbed conditions (allostasis); (3) the activity of lactate and its shuttles in neuropathological states (in particular in traumatic brain injury and neurodegenerative diseases); and (4) the brain in crisis—in response to a neuroinflammatory infectious disease—according to the astrocyte–microglia lactate shuttle (AMLS) model, using tuberculous meningitis (TBM) as an example.

#### HOMEOSTATIC NEUROENERGETICS

#### Shifting Neuroenergetics Paradigms

The classical view of neuroenergetics is that the blood supplies oxygen and glucose to the brain (Sokoloff, 1989). Glucose is the primary source of energy utilized by both neurons and astrocytes. It undergoes complete oxidation via glycolysis, the Krebs cycle and oxidative phosphorylation, which ultimately produces adenosine triphosphate (ATP) for energy-dependent reactions. Thus, glucose is used in the same way by all cell types. Since neurons consume the greatest quantity of energy of all brain elements, metabolic intermediates (e.g., in the Krebs cycle) are diverted toward neurons. Some of the pyruvate produced by glycolysis is converted to lactate and released into the extracellular space. In this classical view, lactate is considered a by-product with deleterious effects when in excess (Norenberg et al., 1987; Siesjö, 1988; Bender et al., 1997). Astrocytes have long been thought to play a passive role in supporting neuronal function, with the neuron being the star of the show. However, the dynamic involvement of astrocytes (Ranjbar and Amiri, 2015) in the forefront of neuroenergetics is now being recognized, shifting paradigms (Haydon and Carmignoto, 2006; Giaume et al., 2010). The neurocentric view of neuroenergetics is evolving into a more integrated one of complementary and co-operative metabolic interactions between astrocytes and neurons.

#### Astrocytes–More than a Supporting Role?

At the cellular level the human brain consists of the allimportant neurons and up to 10 times more glial cells than neurons (Kimelberg and Norenberg, 1989). There are four types of glial cells: ependymal cells, oligodendrocytes, astrocytes, and microglia (resident macrophages in the brain). Astrocytes constitute about 50% of the total human brain volume and are classically divided into three types based on morphology and spatial organization. These three types are: (1) radial—orientated perpendicular to ventricular surfaces with long, unbranched processes (end-feet); (2) protoplasmic—displaying bushy morphology with numerous, highly branched, short processes; and (3) fibrous—manifesting stellate shapes with smooth, long processes that are less branched. The cytoarchitectural organization of astrocytes is such that, according to Pellerin (2010) particular sections cover 99% of the surface area of cerebral blood vessels; although, due to tissue shrinkage with chemical fixation (Korogod et al., 2015), this 99% value is likely an inflated estimate. These astrocytes are, however, the preferential site for glucose uptake from the blood, as well as having projections in peri-synaptic areas of neurons, providing close interaction with neuronal elements and acting as a cellular barrier between blood and neurons. The unique morphological and phenotypic characteristics of astrocytes ideally position them to sense and respond dynamically to changes in neuronal activity (Pellerin, 2010; Bélanger et al., 2011), lending them to conduct numerous critical functions (Chen and Swanson, 2003; Steele and Robinson, 2012), such as glutamate homeostasis (in the glutamate–glutamine cycle), maintaining brain ionic equilibrium (K<sup>+</sup> and H<sup>+</sup> buffering), the maintenance of reactive oxygen species (ROS) (in glutathione recycling) and osmotic regulation. Astrocytes therefore support neuronal activity via structural, trophic and metabolic means, suggesting a critical role in regulating neuroenergetics and homeostatic functions (Pellerin, 2010). Notably, neurons rely on astrocytes to supply precursors of the Krebs cycle intermediates, or their derivatives, as the enzyme pyruvate carboxylase is present in only astrocytes but not in neurons (Hertz et al., 1999).

Astrocytes exhibit a higher capacity for glucose utilization, as well as greater metabolic plasticity, than neurons; these characteristics are important for homeostatic and neuroprotective functions. The calculated energy needs of astrocytes only represent about 10–15% of the total brain energy needs (Attwell and Laughlin, 2001; Gjedde et al., 2002; Rothman et al., 2003; Shulman et al., 2004). Hence, approximately 85% of the glucose in the brain is used in the expenditure of energy in neurons via the glycolytic pathway and the Krebs cycle leading to the synthesis of ATP (Jueptner and Weiller, 1995; Attwell and Laughlin, 2001). The high glycolytic rate of astrocytes suggests a preference for the production and release of lactate. The neuroprotective role of lactate has been experimentally demonstrated by studies (Cater et al., 2001, 2003). Neuroprotection here is defined as an intervention that prevents the death of vulnerable neurons and slows disease progression. Hence, evidence that has emerged over the past two decades has begun to highlight lactate as a supplementary substrate for neurons, resulting in the (re)emergence of a dynamic nursing role for astrocytes (Bouzier-Sore et al., 2002).

#### Astrocyte–Neuron Lactate Shuttle (ANLS) Hypothesis

Magistretti and Pellerin (1996) presented the framework of a hypothesis that they have since developed and refined to become one of the prevailing contemporary viewpoints of neuroenergetics—the ANLS hypothesis. The hypothesis states that astrocytes respond to intensified neuron activity by increasing their rate of glucose uptake, glycolysis and the release of lactate into the extracellular space, as shown schematically in **Figure 1**. At the metabolic level, it begins with glutamatergic activity, a process whereby increased neuronal activity results in the release of glutamate, the main excitatory neurotransmitter in the brain, into the extracellular space along the glutamate transporter EAAT3, which is exclusively located in neurons. Astrocytes sense increased activity at the glutamatergic synapses, followed by glutamate uptake via the glia-specific glutamate transporters EAAT1 and EAAT2. The transport of glutamate is driven by a sodium gradient (i.e., by a Na+-dependent mechanism), with a stoichiometry of three Na<sup>+</sup> ions cotransported with one glutamate, resulting in a significant increase in intracellular Na<sup>+</sup> concentrations in astrocytes (Magistretti and Pellerin, 1999; Pellerin and Magistretti, 2004; Bélanger et al., 2011). Glutamate taken up by astrocytes is converted to glutamine through an ATP-dependent reaction catalyzed by astrocyte-specific glutamine synthetase. Glutamine is released back into the extracellular space and taken up by neurons, where it is converted to glutamate by glutaminase (Magistretti and Pellerin, 1999; Bélanger et al., 2011). This reaction thereby replenishes the neurotransmitter pool of glutamate and completes the glutamate–glutamine cycle. Glutamate uptake by astrocytes stimulates glucose uptake with a stoichiometric relationship of 1:1 between uptake of glutamate and glucose. Increased concentrations of Na<sup>+</sup> in astrocytes activate the enzyme Na+-K+-ATPase, particularly the α<sup>2</sup> subunit. The result is the triggering of glycolysis, leading to the production and release of lactate into the extracellular space; the lactate is then taken up as an energy substrate by neurons for oxidativederived ATP production (Debernardi et al., 1999; Magistretti and Pellerin, 1999; Pellerin, 2003; Pellerin and Magistretti, 2004; Pellerin et al., 2007; Bélanger et al., 2011). This demonstrates the presence of open astroglial metabolic networks—an intercellular route, allowing the trafficking of energy substrates through astrocytes from their source, which is blood vessels, to the site of high energy demand and use, the neurons (Giaume et al., 2010).

Fox et al. (1988) were the first to show that focal, physiological increase in neuronal activity induced by visual stimulation in humans was associated with increased glucose uptake and blood flow in the human visual cortex. This suggested that increased neuronal functional activity stimulates glycolysis. Prichard et al. (1991) and Sappey-Marinier et al. (1992) further developed this postulate by demonstrating increased levels of lactate in the human visual cortex using nuclear magnetic resonance (NMR) technology, following visual stimulation. Thus, a tight coupling was demonstrated between neuronal activation and glucose utilization in astrocytes (Magistretti, 2006), an intrinsic feature of astrocytes not specifically linked to culture conditions or cell origin (Pellerin et al., 2007). It is interesting to note that glucose uptake by astrocytes is disproportionately high compared to their energy requirements. This suggests that sustained astrocytic glycolysis occurs in order to maintain the extracellular lactate pool to meet the energy requirements of neurons. Most recently, data from Angamo et al. (2016) suggest that astrocytic neuronal lactate shuttles contribute to the regulation of ion homeostasis and synaptic signaling in the presence of ample glucose.

In addition to blood-borne glucose, the brain also makes use of an important energy reserve in the form of glycogen stores located almost exclusively in astrocytes. Mobilization of glycogen occurs without ATP requirement during extended periods of limited energy supply (e.g., in hypoglycemia) and leads to enhanced lactate production and release, thereby maintaining energetic homeostasis and preserving neuronal function and viability (Wender et al., 2000; Pellerin, 2003; Pellerin and Magistretti, 2004). Glycogen mobilization can thus be viewed as an extension of the ANLS concept instead of a competing hypothesis (Pellerin et al., 2007). But, as stated by Dienel and Cruz (2015), glycogen is very difficult to study in the human brain. Roles for glycogen in brain function continue to emerge, revealing increasing complexity of the signaling and regulatory mechanisms that integrate glycogen mobilization with physiological activities.

It is important to note that the ANLS hypothesis does not preclude glucose as a source of energy for the brain. Glucose remains an important energy substrate with concentration gradients facilitating the transport of glucose across the blood– brain barrier into the brain via glucose transporters (GLUTs).

This gradient exists since concentrations of glucose in plasma and CSF are approximately 5 mM (Psychogios et al., 2011) and 3 mM (Wishart et al., 2008) respectively, with a CSF:plasma glucose ratio typically being approximately 0.4–0.8 (Leen et al., 2012). GLUT1 is specific to astrocytes and cerebral blood vessels, GLUT3 is exclusively localized at neurons, and GLUT5, primarily a fructose transporter, is found in microglia (Vannucci et al., 1997).

Other important regulating proteins in neuroenergetics are: pyruvate dehydrogenase, the rate-limiting enzyme that catalyzes lactate oxidation in astrocytes, with the inactive form being greater in astrocytes than in neurons (Itoh et al., 2003); lactate dehydrogenase (LDH), of which isoenzymes are found to be specific to astrocytes (LDH–5: associated with glycolysis) and neurons (LDH–1: associated with oxidative metabolism) (Bittar et al., 1996; Laughton et al., 2000); and monocarboxylate transporters (MCTs), proton-linked transporters of lactate with localization of subtypes MCT2 to neurons (with high affinity for lactate) and MCT1 and MCT4 to blood vessels and astrocytes, respectively (with lower affinity for lactate) (Bergersen, 2007). The kinetics of these regulating proteins supports the concept of a flux of lactate from astrocyte to neuron. Most recently in literature there has been renewed vigor in the research in the role of MCTs (Bourgeois et al., 2016; Chaumeil et al., 2016; Karagiannis et al., 2016; Kolko et al., 2016; Lee and Kang, 2016; Pérez-Escuredo et al., 2016; Rosafio et al., 2016). Hence, we are entering an era in science whereby we are recognizing that understanding not only shuttles, but the means by which shuttles operate—transporters (e.g., MCTs), are integral toward unraveling new information of the complex dynamic processes found in neuroenergetics.

Astrocytes are therefore viewed as "lactate sources" supplying the extracellular lactate pool, and neurons as "lactate sinks" consuming lactate oxidatively in response to energy demands. The transfer of lactate from astrocyte to neuron can be viewed as a spatially and temporally independent process (Pellerin and Magistretti, 2004). Thus, the metabolic plasticity of astrocytes is likely to be associated with synaptic plasticity (Magistretti, 2006). Magistretti (2000, 2009)sums up the ANLS model in terms of neurometabolic coupling in which sodium-coupled uptake of glutamate by astrocytes activates Na+-K+-ATPase, which triggers glucose uptake and its glycolytic processing. This results in the production and release of lactate that is used by neurons for activity-dependent energy demands. It has become generally accepted that lactate is a pivotal component in neuronal brain energy homeostasis (Schurr et al., 1999a; Pellerin, 2003, 2010; Gladden, 2004; Schurr, 2006; van Hall et al., 2009).

#### Controversy Surrounding the ANLS Hypothesis–Is this the New Paradigm?

Since its introduction, the ANLS has received much attention, some of which criticizes and refutes the concept; however, substantial evidence has accumulated from independent sources that supports the model. Based on in vitro and in vivo studies (Kuhr and Korf, 1988; Pellerin et al., 2002, 2007; Itoh et al., 2003; Smith et al., 2003; Pellerin and Magistretti, 2004; Hashimoto et al., 2008; Sampol et al., 2013), there exists experimental evidence that neurons use lactate derived from astroglial metabolism efficiently as energy substrates to preserve normal neuronal function. The lactate shuttle concept at the core of the ANLS was further supported by studies on the critical role of glutamate transporters (Pierre et al., 2000; Voutsinos-Porche et al., 2003) and distribution of MCTs in the brain (Pierre et al., 2000; Chiry et al., 2008; Erlichman et al., 2008; Robinet and Pellerin, 2011), particularly for long-term memory formation (Suzuki et al., 2011).

Molecular mechanisms provide further in vitro evidence that supports the ANLS hypothesis (Bliss et al., 2004; Porras et al., 2004), showing that upregulation of glycolysis in neurons decreased oxidation of glucose through the pentose phosphate pathway, resulting in impaired regeneration of reduced glutathione, and subsequently oxidative stress and apoptotic death. Thus, neurons downregulate glycolysis in order to use available glucose to maintain antioxidant status (a neuroprotective mechanism) at the expense of its use for bioenergetics purposes. Neuroenergetic demands can be met by other sources, such as lactate (Tabernero et al., 1996; Herrero-Mendez et al., 2009).

Rouach et al. (2008) demonstrated in vitro the change in character of astrocyte metabolic networks in response to local energy demand and trafficking of energetic metabolites from blood vessels, through astrocytes, to distal neurons. Using <sup>13</sup>C-NMR, this metabolic flux of lactate from astrocyte to neuron can be measured noninvasively in the human brain (Bouzier-Sore et al., 2003; Gallagher et al., 2009; Boumezbeur et al., 2010). Plasma lactate supports up to 10% of brain energy metabolism under physiological conditions, and up to 60% under supra-physiological conditions. Mathematical modeling studies have also described these dynamics, in particular brain lactate kinetics (Aubert et al., 2005) and compartmentalization of brain energy metabolism (Aubert and Costalat, 2005). However, as stated by Pellerin et al. (2007): "[mathematical modeling] ... can give a coherent, quantitative framework for the discussion, suggest possible mechanisms or, conversely, emphasize the contradictions or implausibility or some hypotheses." An example of the argument about the possible ambiguity of mathematical modeling is given by Genc et al. (2011). This sentiment is further expressed by Jolivet et al. (2010): "mathematical models are very powerful tools but they are ultimately only partial replicas of the system they model, not the system itself." Thus, mathematical models are useful guides but one should be mindful to draw definitive conclusions based on modeling studies alone. Jolivet et al. (2010) commented on two mathematical studies of hypotheses of opposing lactate flux in neuroenergetics—they compared the mathematical model of ANLS by Aubert et al. (2007) and the neuron-to-astrocyte lactate shuttle (NALS) by Mangia et al. (2009). Jolivet et al. (2010) concluded that the Aubert (ANLS) model currently remains the best biophysical representation of neuron–astrocyte metabolic interactions. Further multi-timescale mathematical modeling by Jolivet et al. (2015) has since provided further support for the ANLS hypothesis.

While less vocal, there remains strong opposition to the ANLS hypothesis, with various studies expressing divergent views. Dienel and Hertz (2001) emphasized the paradox of intense production of lactate by the brain and the slow rate of its uptake, thus disagreeing that lactate is used as a major neuronal fuel. Dienel and Cruz (2003, 2004) further argued that cerebral metabolic rates of oxygen consumption do not equal that of glucose plus glycogen and that this disproportionate consumption is strong evidence against stoichiometric transfer of lactate from astrocytes to neighboring neurons for oxidation. Gjedde et al. (2002) averred that, based on their experimental evidence, changes in metabolism in afferent phase (activity involving presynaptic terminals and astrocytes) and efferent phase (activity involving neurons) are additive. This view opposed the idea that changes in metabolism are characterized by significant transfer of lactate from astrocyte to neurons. Gjedde et al. (2002) conclude that there is no suggestion that astrocytic glycolysis supports oxidative metabolism of neurons in the baseline condition. Instead, neurons increase their oxidative metabolism in parallel with a rise in pyruvate generated by neuronal rather than astrocytic glycolysis (Gjedde and Marrett, 2001). Mangia et al. (2003a) discussed various ambiguities from other studies and conclude that astrocyte activation supports only the glutamate–glutamine cycle and that neurons primarily metabolize directly absorbed glucose to support neuronal activity. This behavior is demonstrated by an initial dip in lactate concentrations following visual stimulation (Mangia et al., 2003b). As mentioned above, Mangia et al. (2009) also proposed a model of opposite flux to the ANLS, namely the NALS hypothesis which utilized the mathematical model introduced by Simpson et al. (2007), suggesting that depending on the thermodynamic and kinetic status of the cytosolic and mitochondrial redox states, lactate transfers from neurons to astrocytes. The directionality of the ANLS hypothesis (i.e., lactate from astrocytes to neurons), has been covered in many reviews. However, the other perspective of shuttling direction (i.e., from neurons to astrocytes) is an intriguing one, albeit without substantiated experimental proof of the NALS, beyond modeling. Other divergent views come from studies by, for example, Bak et al. (2006), who suggested that synaptic activity does not induce corresponding upregulation of lactate metabolism in neurons; DiNuzzo et al. (2010a), who suggested that carbon recruitment by neurons relies upon glucose uptake rather than that of a lactate shuttle, and that glycogenolysis in astrocytes preserves glucose availability for neurons (DiNuzzo et al., 2010b).

Others have critically reviewed the ANLS hypothesis in defense of the traditional role of neuroenergetics (Chih et al., 2001; Chih and Roberts, 2003). Pellerin and Magistretti (2011) not only addressed all these divergent views by reviewing experimental evidence supporting the main tenets of the ANLS model, but also were able to confute several unfounded criticisms. With overwhelming support for the ANLS hypothesis, it was becoming clear that lactate is a crucial element in neuronal brain energy homeostasis (Schurr et al., 1999b; Pellerin, 2003, 2010; Gladden, 2004; Schurr, 2006; van Hall et al., 2009). However, as is the case in science, refuting evidence continues to challenge the foundations of the ANLS hypothesis, leading to it being redefined and strengthened over time. The tenets of this hypothesis continue to hold, begging the question—is this the new paradigm for homeostatic neuroenergetics?

# NEUROPATHOLOGY AND THE HUMAN RESPONSE

#### Beyond Homeostasis

Biological processes maintain stability, by detecting environmental (external) and physiological (internal) changes, and activating specialized adaptive responses. The dynamic metabolic characteristics of astrocytes lend them to being particularly adept at such a task within the neuronal framework of homeostasis. Beyond homeostasis lies the comparatively new concept of allostasis. Allostasis is an extension of the concept of homeostasis and refers to the maintenance of stability by means of robust, energy-demanding adaptive mechanisms in response to severe physical, psychosocial or environmental challenges. Homeostasis involves maintaining a stable internal environment of an organism (i.e., by means of a feedback mechanism), whereas allostasis is more dynamic in that it involves the continuous response to physiological needs, resulting in biochemical adaptive mechanisms. These two concepts seem similar and are not intended to operate independently; instead, allostasis supports homeostasis, placing emphasis on flexible adaptation with the ultimate goal of maintaining a stable internal environment (McEwen, 2006, 2008; Logan and Barksdale, 2008; McEwen and Gianaros, 2011; Danese and McEwen, 2012). Frequent or chronic challenges (e.g., neuroinflammation) produce dysregulation of several major physiological systems by triggering chemical mediators of adaption that operate in a nonlinear network. The cumulative "wear and tear" associated with the inability to disengage these physiological systems is referred to as allostatic load. In vulnerable biological systems an allostatic overload prevails that results in the development of disease.

Since the term was first coined, by Sterling and Eyer (1988), the concept of allostasis has primarily been used in the literature to describe mild perturbations that are stress related, psychosomatic and/or psychopathological (Tannenbaum et al., 2002; Stewart, 2006; Shannon et al., 2007; Logan and Barksdale, 2008; Blair et al., 2011; Danese and McEwen, 2012; Tomiyama et al., 2012). As a concept allostasis is still being developed and so is gradually being applied to more diverse fields, such as metabolic diseases and lipidomics (Oresic et al., 2008) and metabolomics (Ramautar et al., 2013). Allostasis has also been used to explain extreme glucose fluctuations as reflecting the body's inability to cope with allostatic load in the case of chronic illness, predisposing the individual to serious harm as manifested by heightened mortality (Stumvoll et al., 2003, 2004; Rake et al., 2010).

Herein this review the typology, as given by Peters and McEwen (2015) to describe cardiovascular disease, has been adapted to define terms in neuroenergetics. Stress occurs in a state of increased cerebral energy demand to safeguard an individual's physical, mental and social well-being. A reversible stress response distinguishes allostasis from homeostasis and can be indicated by a relief result (restored homeostasis) in which the individual returns toward physiological normality. An allostatic load is indicated by an acute stress challenge—an irreversible or nearly irreversible state of allostatic stress from homeostasis—as seen in severely elevated CSF lactate in acute, physical damage to the brain (Gallagher et al., 2009). Chronic cases of persistently highly elevated CSF lactate and lactate:pyruvate ratios would be parameters to define the transition of allostatic load toward allostatic overload (disease).

Thus, allostasis encompasses typical dynamic adaptations to transient stress mediators; allostatic load pertains to an acute state of persistent perturbation and, ultimately, circumstances of allostatic overload are associated with disease and the body's inability to cope. These terms are henceforth discussed in terms of neuropathological conditions, with the focus on lactate and its shuttles.

#### Microglia–Addressing Neuronal Allostasis

Under pathological conditions, such as caused by infection, the brain enters a state of allostasis and microglia become activated. Microglia are immunocompetent cells that act as the intrinsic macrophages of the brain, bearing multiple similarities to those of macrophages in peripheral tissues. Some of the shared functions include phagocytosis, antigen presentation, effector inflammatory response and production of various cytotoxins and cytokines. Thus, microglia are an important component of both the innate and adaptive immune response to central nervous system (CNS) pathogens (Olson and Miller, 2004). They populate the entire brain parenchyma with a homogeneous distributed network with territorial organization and act as the first and only line of defense in the brain.

Under physiological conditions microglia exist in a resting ("ramified") state characterized by a small soma and numerous thin, branched processes (Giaume et al., 2007). Microglia remain quiescent until activated upon by brain insult (i.e., injury, disease or infection), characterized by proliferation and immunophenotypical (expression of various surface markers and ion channels) and morphological (transformation to amoeboid morphology) changes (Eder, 1998). The response of microglia to brain insult is first to detect the site of assault by constant dynamic monitoring of the surrounding micro-environment, and then to send out process extensions toward the lesion site where process tips reorganize to confine, control and eliminate the source of the disturbance.

Neuroinflammation, the reaction of surrounding brain tissue to brain insult, is characterized by synthesis of various inflammatory mediators and reactive gliosis, associated with phenotypic changes and proliferation of glia (both astrocytes and microglia), in response to a dynamically changing environment (Giaume et al., 2007). The modified phenotype of astrocytes, from basal to reactive state, results in them abandoning their neuroprotective role, allowing excessive oxidative stress and production of ROS, and subsequently ROS-induced neuronal damage (Pellerin, 2003). Chronic neuroinflammation, the response to sustained and widespread stimuli, typically occurs due to either disease (e.g., Alzheimer's and multiple sclerosis) or persistent infection by viruses (Olson et al., 2001; Ovanesov et al., 2008) or bacteria (e.g., chronic meningitis as in TBM). The role of microglia in CNS infections was extensively reviewed by Rock et al. (2004). Hence, metabolic coupling of microglia is essential during neuronal allostasis, especially in response to pathogens.

#### THE BRAIN IN CRISIS

#### Allostatic Load–Physical Brain Trauma

The topic of traumatic brain injury (TBI)—acute, physical damage to the brain—gained notoriety in 2009 when Jeanne Marie Laskas wrote an article titled "Game brain" in GQ magazine (Laskas, 2009), exposing the hidden, negative medical implications within some popular, professional contact sports. In 2015 this was expanded into a book (Laskas, 2015) and a film called "Concussion," about forensic pathologist Bennet Omalu and his attempts to publicize his scientific findings regarding physical contact sports (such as in the National Football League, the apex of American football). Omalu postulated that persistent mild TBI leads to chronic traumatic encephalopathy. Consequently, there has been renewed research into concussion and TBI (Sone et al., 2016; Sussman et al., 2016), making it another hot topic in neuroscience.

It is well recognized that there is an elevation of brain extracellular lactate and the lactate:pyruvate ratio in TBI cases (Gallagher et al., 2009). The group of Gallagher et al. (2009) and Carpenter et al. (2014) presented direct evidence of brain utilization of lactate in TBI cases by administration of <sup>13</sup>C-labeled lactate via a microdialysis catheter and using <sup>13</sup>C-NMR analysis. Based on these studies, Carpenter et al. (2015) suggested that where neurons are too damaged to use the lactate produced from glucose by astrocytes (i.e., by uncoupling of neuronal and glial metabolism), high extracellular levels of lactate would accumulate—explaining the association between high lactate and poor outcome. Thus, therapeutic intervention before metabolic uncoupling is imperative. Glucose administration after TBI has been shown to be beneficial but there is a correlation between hyperglycemia and increased infection or mortality; hence, alternative sources of energy, such as lactate, may improve outcome (Moro et al., 2013). However, as with the ANLS model, there are opposing opinions of the notion that lactate is a preferential fuel in TBI cases (Dienel, 2014).

Acute administration of exogenous fuels, such as lactate, after experimentally induced TBI in rats has been shown to attenuate histopathology and improve outcome (Chen et al., 2000). In two companion reports by Glenn et al. (2015a,b), evidence was provided that central venous tracer infusion of both glucose and lactate showed massive mobilization of mainly lactate systematic lactate was preferentially being directly consumed and used by TBI cases. Thus, there is a high production and clearance rate of lactate in TBI cases. The role of this lactate in cerebral metabolism following TBI and the advantages of treatment by exogenous lactate infusions was reviewed by Brooks and Martin (2015). In recent clinical trials of TBI, and other neurocritical care cases, the beneficial properties of hypertonic lactate have been observed with respect to cerebral blood flow and intracranial pressure (Bouzat and Oddo, 2014; Patet et al., 2016). Thus, accumulating evidence points to lactate being an important component used during allostatic load caused by acute, physical damage to the brain—a therapeutic target.

# Allostatic Overload–Neurodegenerative Diseases

A common trait among neurodegenerative diseases is perturbed brain energy metabolism (Magistretti and Pellerin, 1996; Beal, 2000; Bélanger et al., 2011; Albanese et al., 2016). Huntington's disease and multiple sclerosis have been linked to mitochondrial dysfunction, and so are susceptible to oxidative stress and energy deficits (Dutta et al., 2006; Regenold et al., 2008; Sack, 2010; Gouarné et al., 2013); with implications of perturbed lactate levels. In contradiction, two <sup>1</sup>H-NMR studies found conflicting results regarding lactate in Huntington's disease. An <sup>1</sup>H-NMR investigation by Gårseth et al. (2000) reported decreased levels of lactate, which they attributed to neuronal loss, but based upon only a small sample (n = 7). Another <sup>1</sup>H-NMR study, by Verwaest et al. (2011), found lactate to be significantly increased, but this result was also based on a small sample (n = 10). In a more comprehensive study, Gouarné et al. (2013) used cultured neuronal subpopulations from transgenic mice and provided evidence that neurons use lactate, along with pyruvate, as an energy source to support respiration. The importance of lactate was corroborated by a study (Covarrubias-Pinto et al., 2015) on Huntington's disease (HD) using ascorbic acid to inhibit use of neuronal glucose—showing the favoring of lactate uptake to sustain brain activity. Hence, experimental evidence is emerging that lactate is important and used preferentially as an energy source in HD cases.

Two reviews on neurodegenerative diseases support the ANLS hypothesis and its role in neuropathology—namely, lactate is neuroprotective and a therapeutic agent. A review by Newington et al. (2013) refers to their previous scientific work, which showed that increased lactate production proved to be protective against Aβ-induced neuronal toxicity that isinherently associated with Alzheimer's disease. A review by Finsterwald et al. (2015) summarizes numerous studies involving astrocytic lactate in treatments for multiple neurodegenerative diseases [Alzheimer's, Parkinson's disease and amyotrophic lateral sclerosis (ALS)]. Finsterwald et al. (2015) concluded that astrocytic function elicits intrinsic neuroprotective properties through, amongst others, the ANLS mechanism. One could go one step further and speculate that a "specialized" type of ANLS mechanism is necessary.

In multiple sclerosis (MS) there have been reports of the importance of CSF lactate. A negative correlation between the presence of lactate and the presence of inflammatory plaques and MS severity has been experimentally shown (Lutz et al., 2007; Albanese et al., 2016)—increased lactate in MS cases without plaques and vice versa, suggesting close metabolic coupling between plaque activity and lactate production. A gene expression study by Zeis et al. (2015) revealed alterations (downregulation) in the ANLS mechanism in MS, such that alternative lactate shuttle systems may be at play. Indeed, other studies have proposed such shuttle systems to supply demyelinated axons with lactate as an important energy source namely, the astrocyte-axon lactate shuttle (Cambron et al., 2012; Nijland et al., 2015) and the oligodendrocyte-axon lactate shuttle (Campbell et al., 2014).

A common thread seen in these studies of neurodegenerative diseases is the perturbation of lactate. Evidence in neuropathophysiological cases indicates that lactate acts as an important, sometimes preferential, source of energy, and also has neuroprotective properties. The role of lactate shuttle systems is also emerging as being vital—imperative as supply and demand needs fluctuate to extremes in situations of allostatic overload. This is when normal physiological systems can no longer provide the required resources and/or effects. It is, however, important to note that neuropathologically activated astrocytes, as well as other glial cells, such as microglia, may support mechanisms, such as neuroinflammation, aggravating neuronal degeneration (Li et al., 2011). The dynamics behind addressing neuroinflammation caused by an invading pathogen involves the use of microglia.

In the remainder of this review an adaptation of the ANLS model—the AMLS model (Mason et al., 2015)—is discussed in terms of TBM; a chronic, infectious neuroinflammatory disease.

# Pathogen-Induced Chronic Neuroinflammation

In the case of a chronic infection in the brain by a persistent pathogen, such as Mycobacterium tuberculosis (Mtb), the bacillus responsible for TBM, there is sustained allostatic overload. Here, the microglia strive to eradicate the scourge but unwittingly become the habitat of the persisting pathogen. Similar to the ANLS hypothesis, the AMLS model proposes that when the brain is in crisis due to infection, energy flow in brain metabolism is shifted away from the neurons, and shunted toward the microglia (see **Figure 2**). The initial mechanics of the AMLS remain the same as the ANLS, the only difference being the directionality of the lactate upon exiting the astrocyte. The AMLS hypothesis postulates that in neuroinflammatory infectious diseases, such as TBM, lactate produced through glycolysis in astrocytes participates in the activated immune response and, in association with ketones and gluconeogenic amino acids, is collectively directed from the neurons preferentially into microglia. Within the microglia, lactate is expected to enter the mitochondrial Krebs cycle, contributing to oxidative phosphorylation and hence producing high levels of ATP and forms of ROS, such as hydrogen peroxide, required for degradation of the invading pathogen. Thus, the AMLS hypothesis uses the same line of logic of the ANLS model, but instead the microglia are the prime focus under conditions of neuroinflammatory infectious diseases; increased astrocytic lactate is directed toward the microglia.

Using the same CSF samples from the study by Mason et al. (2015), it was shown in a targeted analysis (Mason et al., 2016a) that the highly elevated amounts of lactate observed in the CSF of confirmed TBM cases was only in the L-form (L-lactate is produced by the host, whereas D-lactate is of bacterial origin). Thus, elevated lactate found in the CSF of TBM cases is solely a response by the host to the infection. These results from this follow-up study by Mason et al. (2016a) provide experimental evidence to support the proposed AMLS model and highlight the fact that lactate plays an important role in neuroinflammatory diseases, such as TBM. Activation of microglia, as required in the AMLS hypothesis, does not, however, present a uniform process and involves intricate interactions and feedback loops between the microglia, astrocytes and neurons that hamper attempts to construct basic and linear cascades of cause and effect; TBM involves a complex integration of the responses from the various cell types present within the CNS, with microglia and the astrocytes as main players.

During the course of a neuroinflammatory infectious disease, such as TBM, the ANLS is not only inadequate but, indeed, a liability as activated neurons will only become vulnerable to neurodestructive components. The bypassing of metabolic energy intermediates from the neurons effectively inactivates them to protect them from neurodestructive agents in response to chronic neuroinflammation. This neuron inactivation is evident from a clinical viewpoint as the progression of TBM is reflected by reduced scoring in the Glasgow Coma Score—specifically, decreased consciousness and awareness. The AMLS hypothesis can thus be viewed as invoking a persistent allostatic overload state that becomes activated when allostatic mechanisms fail and the body enters into a diseased state. The regulation of dynamic systems that function within the brain, in terms of homeostasis, allostasis and disease states, is delicate. Chronic activation of microglia can lead to neuropathological sequelae (Streit et al., 2004), and is also linked to neuropathic pain (Raghavendra et al., 2003). The persistent activation of microglia can be viewed as the proverbial double-edged sword—simultaneously exhibiting neuroprotective and neurodestructive properties in an attempt to save the whole at the expense of the part (Hall et al., 1998; Giaume et al., 2007).

Using TBM as an example of an infectious neuroinflammatory disease—an extreme scenario of allostatic overload—it can be speculated that the Mtb-induced host and microbe markers can reflect: (1) the disease state on admission to hospital, and (2) differentiation in the restoration toward a new condition of homeostasis following treatment. Upon admission to hospital, most TBM patients present with moderate to severe symptoms of ketosis. Metabolomics analysis of the urine (Mason et al., 2016b) substantiates this clinical picture through a ketosis urinary biosignature—cases admitted to hospital with very high ketosis biomarkers tended to have a poor prognosis, leading to severe neurological complications irrespective of treatment. A similar profile was also reflected in the CSF biochemistry (Mason et al., 2016a); as the TBM disease progressed into its later stages there was a deterioration in patient prognosis. Hence, in more advanced TBM stages there is less CSF lactate and therefore fewer energy substrates available for microglia activity, and decreased protection of neurons. During this period, irreversible neurological damage can begin to occur—illustrating the importance of lactate levels during periods of chronic neuroinflammation.

The concept of lactate shuttle systems extends to other types of neuronal cells, such as the oligodendrocyte–axon lactate shuttle (Baltan, 2015) and the Bergmann glia (BG) Purkinje cell (PC) lactate shuttle (Sawada et al., 2016). Lactate shuttles extend beyond the brain—they are of particular interest in cancer research (Pizzuto et al., 2012). Several cancer studies have suggested a tumor-to-stroma coupling via a lactate shuttle in

cancer cells (Pértega-Gomes et al., 2014; Sanità et al., 2014). These studies indicate that over-propagation of cells is strongly dependent upon the anaplerotic role of increased lactate uptake (Whitaker-Menezes et al., 2011; Fiaschi et al., 2012). This concept is supported by a large-scale in silico metabolic model (Capuani et al., 2015). Interestingly, Somashekar et al. (2011) stated that their most relevant metabolic observations in studying tuberculosis are similar to metabolic changes seen in cancer during tumor development—corroborated by Zhou et al. (2015). Conceptually, the "Reverse Warburg Effect" (Pavlides et al., 2009), proposed as operating in tumor cells, has been interpreted to be analogous to the astrocyte–neuron metabolic coupling (Pavlides et al., 2010), where the astrocytes resemble cancerassociated fibroblasts and the neuron the epithelial tumor cells, resulting in their higher proliferative capacity.

The regulation of lactate shuttles has been discussed as being a potential therapeutic target in cancer research (Draoui and Feron, 2011; Sanità et al., 2014). While disruption of transport of lactate into cancerous cells (i.e., reducing lactate shuttling into cancerous cells) is considered beneficial, upregulation in neuropathological cases shows merit for research. Several studies have demonstrated that intracerebroventricular or intravenous injection of lactate yields a neuroprotective effect during experimentally induced hypoglycemia or cerebral ischemia (Berthet et al., 2009, 2012; Wyss et al., 2011). In addition, other experimental work (Yamanishi et al., 2006; Gordon et al., 2008) has shown that lactate increases cerebral blood flow (vasodilation)—increasing the supply of metabolic (e.g., energy) substrates during allostasis. Thus, lactate has a potentially beneficial role in the clinical management of several neurological disorders (Taher et al., 2016). This review, like the others cited here, therefore shows that further research into the role of lactate and its shuttling mechanism(s) are certainly warranted to address neuroenergetics and neuropathology.

#### CONCLUDING REMARKS

In summary, we are seeing the prevailing school of thought moving away from the neurocentric paradigm, albeit with divergent views, toward that of one that is more integrated. Within this emerging new paradigm, an unraveling of the intricacies of the complex and diverse dynamic interactions of the human brain at the metabolic and cellular levels is occurring. Our current knowledge has led us to the realization that lactate shuttle systems, both intracellular and extracellular, play a far more pivotal role than originally considered. Indeed, there exist numerous other shuttles in neuroenergetics—for example: the lactate–alanine shuttle (Zwingman et al., 2000); the astrocyte– neuron ketone body shuttle (Guzmán and Blázquez, 2001); the glycerol phosphate shuttle (McKenna et al., 2006); and the malate–aspartate shuttle (McKenna et al., 2006; Moffett et al., 2013). Each in its own right is essential in terms of both homeostasis and allostasis. It is evident that transitions between different states of homeostasis and allostasis, and disease, are dynamic—they are often difficult to distinguish and/or control. However, the role of neuroenergetics is undeniable in the brain's constant objective of protecting itself and trying to maintain an optimum state. Here, we have discussed, in terms of neuroenergetics, the role of lactate in a shuttle system during homeostasis (ANLS) and in the infected state (AMLS). Other studies on neurodegenerative diseases have described complex pathophysiological changes that highlighted the importance of lactate (Newington et al., 2013; Finsterwald et al., 2015). There is thus overwhelming evidence that astrocytes, lactate and the associated shuttle systems so far recognized in combination are of crucial importance in neuroenergetics—in both the healthy and diseased state. Perhaps not surprisingly, the latest technology has revealed that neuro(patho)physiological metabolism is far more complex than originally thought and that further knowledge

#### REFERENCES


of the lactate shuttle systems apparently operating, as outlined in this article, is key to our understanding of the mechanisms involved.

#### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

#### FUNDING

Research funding was provided by the Technological Innovation Agency (TIA) of the Department of Science and Technology of South Africa. Opinions expressed and conclusions arrived at, are those of the author and are not necessarily to be attributed to the funding body TIA.


Kolko, M., Vosborg, F., Henriksen, U. L., Hasan-Olive, M. M., Diget, E. H., Vohra, R., et al. (2016). Lactate transport and receptor actions in retina: potential roles in retinal function and disease. Neurochem. Res. 41, 1229–1236. doi: 10.1007/s11064-015-1792-x


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Mason. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# ProNGF Drives Localized and Cell Selective Parvalbumin Interneuron and Perineuronal Net Depletion in the Dentate Gyrus of Transgenic Mice

Luisa Fasulo1, 2 †, Rossella Brandi 2 †, Ivan Arisi <sup>2</sup> , Federico La Regina<sup>2</sup> , Nicola Berretta<sup>3</sup> , Simona Capsoni <sup>1</sup> , Mara D'Onofrio<sup>2</sup> and Antonino Cattaneo1, 2 \*

<sup>1</sup> Bio@SNS Laboratory, Scuola Normale Superiore, Pisa, Italy, <sup>2</sup> European Brain Research Institute Rita Levi-Montalcini, Rome, Italy, <sup>3</sup> Department of Experimental Neurology, Fondazione Santa Lucia IRCCS, Rome, Italy

ProNGF, the precursor of mature Nerve Growth Factor (NGF), is the most abundant NGF form in the brain and increases markedly in the cortex in Alzheimer's Disease (AD), relative to mature NGF. A large body of evidence shows that the actions of ProNGF and mature NGF are often conflicting, depending on the receptors expressed in target cells. TgproNGF#3 mice, expressing furin-cleavage resistant proNGF in CNS neurons, directly reveal consequences of increased proNGF levels on brain homeostasis. Their phenotype clearly indicates that proNGF can be a driver of neurodegeneration, including severe learning and memory behavioral deficits, cholinergic deficits, and diffuse immunoreactivity for A-beta and A-beta-oligomers. In aged TgproNGF#3 mice spontaneous epileptic-like events are detected in entorhinal cortex-hippocampal slices, suggesting occurrence of excitatory/inhibitory (E/I) imbalance. In this paper, we investigate the molecular events linking increased proNGF levels to the epileptiform activity detected in hippocampal slices. The occurrence of spontaneous epileptiform discharges in the hippocampal network in TgproNGF#3 mice suggests an impaired inhibitory interneuron homeostasis. In the present study, we detect the onset of hippocampal epileptiform events at 1-month of age. Later, we observe a regional- and cellular-selective Parvalbumin interneuron and perineuronal net (PNN) depletion in the dentate gyrus (DG), but not in other hippocampal regions of TgproNGF#3 mice. These results demonstrate that, in the hippocampus, the DG is selectively vulnerable to altered proNGF/NGF signaling. Parvalbumin interneuron depletion is also observed in the amygdala, a region strongly connected to the hippocampus and likewise receiving cholinergic afferences. Transcriptome analysis of TgproNGF#3 hippocampus reveals a proNGF signature with broad down-regulation of transcription. The most affected mRNAs modulated at early times belong to synaptic transmission and plasticity and extracellular matrix (ECM) gene families. Moreover, alterations in the expression of selected BDNF splice variants were observed. Our results provide further mechanistic insights into the vicious negative cycle linking proNGF and neurodegeneration, confirming the regulation of E/I homeostasis as a crucial mediating mechanism.

Keywords: proNGF, interneurons, parvalbumin, extracellular matrix, dentate gyrus, E/I imbalance, transgenic mice, expression profile

#### Edited by:

Daniela Tropea, Trinity College Dublin, Ireland

#### Reviewed by:

María Llorens-Martín, Spanish National Research Council, Spain Petra Wahle, Ruhr University Bochum, Germany

> \*Correspondence: Antonino Cattaneo antonino.cattaneo@sns.it † Joint first authors.

Received: 20 October 2016 Accepted: 16 January 2017 Published: 09 February 2017

#### Citation:

Fasulo L, Brandi R, Arisi I, La Regina F, Berretta N, Capsoni S, D'Onofrio M and Cattaneo A (2017) ProNGF Drives Localized and Cell Selective Parvalbumin Interneuron and Perineuronal Net Depletion in the Dentate Gyrus of Transgenic Mice. Front. Mol. Neurosci. 10:20. doi: 10.3389/fnmol.2017.00020

# INTRODUCTION

ProNGF, the precursor of mature Nerve Growth Factor (NGF), is the most abundant NGF form in the brain and increases markedly in the cortex in Alzheimer's Disease (AD), relative to mature NGF (Francke et al., 1983; Scott et al., 1983; Fahnestock et al., 2001; Peng et al., 2004). Interestingly, signs of activated proNGF-signaling are detected both in AD and in preclinical mild cognitive impairment (MCI) (Mufson et al., 2012; Counts et al., 2016). A large body of evidence shows that the actions of ProNGF and mature NGF are often conflicting, depending on the receptors expressed in target cells: trkA, the preferred receptor for mature NGF; p75NTR, the pan-neurotrophin receptor, and, in association with its co-receptor sortilin (belonging to the VPS10 receptor family), a high affinity receptor for proNGF (Kaplan et al., 1991; Klein et al., 1991; Lee et al., 2001; Nykjaer et al., 2004; Masoudi et al., 2009). ProNGF can induce apoptosis in cells expressing p75NTR and sortilin, regardless of the presence of trkA (Chao and Bothwell, 2002), and in different lesion models, generally characterized by a higher expression of p75NTR (Beattie et al., 2002; Harrington et al., 2004).

In AD neurodegeneration, the activation of the amyloidogenic pathway has been demonstrated to promote proNGF/NGF dysmetabolism, shifting the balance of the processing reaction in favor of the precursor (Bruno et al., 2009; Iulita and Cuello, 2014).

Conversely, proNGF/NGF imbalance is itself a driver of neurodegeneration (Capsoni and Cattaneo, 2006), as shown, respectively, with indirect and direct evidence in AD11 (Ruberti et al., 2000) and TgproNGF#3 transgenic mice (Tiveron et al., 2013). AD11 mice, which express an antibody that neutralizes selectively mature NGF, with respect to proNGF, develop a progressive and comprehensive neurodegeneration (Capsoni et al., 2000) that can be fully rescued by NGF itself (De Rosa et al., 2005) or, in part, by p75NTR gene ablation (Capsoni et al., 2010). TgproNGF#3 mice, expressing furin-cleavage resistant proNGF in CNS neurons, directly reveal consequences of increased proNGF levels on brain homeostasis. Their phenotype clearly indicates that proNGF can be a driver of neurodegeneration, including severe learning and memory behavioral deficits, cholinergic deficits, and diffuse immunoreactivity for A-beta and A-beta-oligomers (Tiveron et al., 2013). Interestingly, in aged TgproNGF#3 mice spontaneous epileptic-like events are detected in entorhinal cortex-hippocampal slices, suggesting the occurrence of excitatory/inhibitory (E/I) imbalance, whereas in vivo no spontaneous seizures were observed (Tiveron et al., 2013).

In such a view, the molecular events linking increased proNGF levels to the observed in vitro epileptiform activity in hippocampal slices deserve further investigation.

The occurrence of spontaneous epileptiform discharges in the hippocampal network in TgproNGF#3 mice suggests an impaired inhibitory interneuron homeostasis.

NGF/proNGF balance was demonstrated to deeply affect cholinergic phenotype (Capsoni and Cattaneo, 2006; Capsoni et al., 2010). Pharmacologically-induced chronic failure in extracellular NGF maturation leads to a reduction in mNGF levels, proNGF accumulation, cholinergic degeneration, and cognitive impairment in rats (Allard et al., 2012). TgproNGF#3 mice directly reveal the impact of increased proNGF levels and NGF/proNGF imbalance on the cholinergic phenotype, presenting a marked cholinergic deficit in BFCN projecting to cortex, and hippocampus in 3-month-old mice (Tiveron et al., 2013). No deficit is observed earlier, at 1-month.

Cholinergic modulatory activity has been demonstrated to affect hippocampal Parvalbumin interneurons subpopulation essential in determining the oscillatory activity (Gulyás et al., 2010; Lawrence et al., 2015). We thus propose that pathological conditions characterized by increased levels of proNGF in the brain, might lead to a reduced cholinergic drive to interneurons, with the ensuing E/I imbalance. In the present study we therefore evaluated the interneuron subpopulations in the hippocampus of TgproNGF#3 mice, at different ages. Parvalbumin+ (Parv+) interneurons are ensheathed by an aggregation of proteoglycan extracellular matrix (ECM) components, the perineuronal net (PNN), that reaches full maturation only during post-natal development and plays an essential role in regulating neuronal firing, stabilizing synapses and regulating synaptic plasticity (Pizzorusso et al., 2002; Wang and Fawcett, 2012). Impairment in PNN structures is described in condition of enhanced neuronal activity (McRae et al., 2012). We therefore also analyzed the PNN system. The robust transcriptional effects of NGF are wellcharacterized (Dijkmans et al., 2009). More recent data show that proNGF activates a largely distinct transcriptional program, and is a less potent transcriptional activator, compared to NGF, in target cells in vitro (D'Onofrio et al., 2011); moreover the extent of NGF/proNGF imbalance, affects the transcriptional outcome (Arisi et al., 2014). Therefore, a parallel analysis of early transcriptional changes in transgenic mice hippocampus was performed, in order to gain further insights into the mechanism starting and sustaining TgproNGF#3 mice phenotype.

# MATERIALS AND METHODS

#### Animal Handling and Experiments

All experiments with mice were performed according to the national and international laws for laboratory animal welfare and experimentation (EEC council directive 86/609, 12 December 1987, and Leg. Decree n◦ 26, Implementation of the UE Directive 2010/63/UE, 4 March 2014). The experiments were performed according to a protocol approved by the Italian Ministry of Health with the authorization n. 3/2012 issued on February 21st 2012, valid until February 21st 2015, in accordance with the guidelines and regulations of the Italian Law (DLGs n.116, 27/1/1992), and carried out in that time interval. Mice were kept under a 12 h dark to light cycle, with food and water ad libitum. Genotyping was performed as described in Tiveron et al. (2013).

**Abbreviations:** NGF, Nerve Growth Factor; AD, Alzheimer's Disease; wt, wildtype; HP, hippocampus; BF, basal forebrain; BFCN, basal forebrain cholinergic neurons; EC, entorhinal cortex; E/I, excitatory/inhibitory; MEA, multi electrode array; ECM, extracellular matrix proteins; PNN, perineuronal net; Parv, Parvalbumin; CB, calbindin; MCI, mild cognitive impairment; DG, dentate gyrus; FS, fast-spiking; BDNF, brain-derived neurotrophic factor; LTP, Long Term Potentiation; LTD, Long Term Depression.

#### Transgenic Mice

As we previously reported in Tiveron et al. (2013), the mouse pre-proNGF cDNA was mutagenized at the furin cleavage site (Beattie et al., 2002) and placed under the transcriptional control of the mouse Thy1.2 promoter, driving brain-specific expression in post-natal and adult transgenic mice (Tiveron et al., 2013). The furin-resistant pre-proNGF cDNA, plus IRES sequence and EGFP, were cloned into the Thy1.2 promoter vector (Caroni, 1997), containing 6.5 kb of the murine thy1.2 gene, driving brain-specific expression in adult transgenic mice. The fragment includes exons I, II, and IV of the mouse Thy 1.2 gene (Vidal et al., 1990), responsible for the selective expression in neuronal cells, whereas exons containing the Thy 1.2 coding region or those responsible for expression in the thymus are absent. Transgenic mice were generated by pronuclear DNA injection of zygotes C57Bl/6xDBA/2 F2 generation using standard procedures (Ciana et al., 1998). Microinjected zygotes were reimplanted into pseudopregnant C57Bl/6xDBA/2 F1 foster mothers to complete their development. For genotyping genomic DNA was extracted from tail biopsies.

#### Primers for Genotyping

IRES-EGFP 4F: (s) gga cgt ggt ttt cct ttg aa; IRES-EGFP 5R: (as) gtc ctc ctt gaa gtc gat gc.

#### Histological Analysis: Immunofluorescence and Image Analysis

ProNGF and WT mice were anesthetized with an excess of 2,2,2 tribromethanol (400 mg/kg) and intracardially perfused with a saline solution and following a 4% solution of paraformaldehyde in phosphate buffer saline (PBS). Brains were post-fixed for 24 h then transferred in 30% sucrose/PBS solution and sectioned on a sliding freezing microtome (Leica, Wetzlar, Germany). Forty micrometers coronal sections were collected in 0.05% sodium azide/PBS in 1.5 ml tubes and stored at 4◦C until usage. Immunofluorescence (IF) stainings were performed using the following primary antibodies: rabbit anti-proNGF antibody (dilution 1:100, Millipore Merkgroup, Italy), rabbit antiaggrecan antibody (1:100 dilution, Millipore, Merkgroup, Italy), mouse anti-Parvalbumin antibody (1:800 dilution, Millipore Merkgroup, Italy) rabbit anti-GAD65-67 antibody (1:1000 dilution, Millipore Merkgroup, Italy), rabbit anti-calbindin 28k antibody (1:1000 dilution, Swant, Switzerland), PNN was stained using biotinylated lectin from Wisteria Floribunda (1:800 dilution, SIGMA, Italia), followed by FITC-, Texas Red-, or AMCA-Streptavidin (Vector Labs, UK).

Double-IF was performed using Alexa 594 and Alexa 488 secondary antibodies (1:200, Invitrogen). Immunofluorescence was examined under a confocal laser-scanning microscope (Leica SP5, Leica Microsystems, Wetzlar, Germany). Confocal acquisition settings were consistent between wild type and transgenic cases. Negative controls for secondary antibodies are shown in Figure S1.

Images were acquired with 5X and 20X dry objective and 40/63X oil immersion objectives at 1024 × 1024 pixel resolution. Settings for laser intensity, gain, offset, and pinhole were optimized initially and held constant through the experiment.

Final figures were assembled using Adobe Photoshop 7 and Adobe Illustrator 10 and Omnigraffle Professional. Image analysis was performed under visual control to determine thresholds that subtracts background noise and take into account neuronal structures with ImageJ software (US NIH) or Imaris Suite 7.4 <sup>R</sup> (Bitplane A.G., Zurich, Switzerland). During image processing, the images were compared with the original raw data to make sure that no structures were introduced that were not seen in the original data series or that structures present in the original data series were not removed.

Cell counting was performed for each region on two squares of 450 × 450 µm per brain section, at least 4–6 sections per animal, females, and males (n = 4 per each animal group). Cell counting was performed on the hippocampus and DG subregion on both side or on the lateral region of the amygdala on both sides. Cells positive for each marker or double stained were counted manually and their density was calculated. Stereology counting was applied (Capsoni et al., 2010). Cell counting was expressed as cell density (Mainardi et al., 2009; Cardoso et al., 2013) per volume (cells/mm3). All analyses were done using a blind procedure. Statistical analysis: all data are expressed as mean ± S.E.M. Statistical analyses were performed using Student's t-test (twotailed distribution) Differences were considered significant at p < 0.05.

#### Microarray Analysis: RNA Isolation, Hybridization, and Analysis

Hippocampus (HP) of the right hemisphere was dissected from the brains of freshly sacrificed mice. All the tissue samples were stored in RNAlater (QIAGEN, UK). Total RNA was isolated from this brain area, using Trizol (Invitrogen S.R.L., Life Technologies, Italy) and DNAse, by Qiagen columns. RNA quantity was determined on a NanoDrop UV-VIS. Only samples with an absorbance ratio in the range 1.8<OD260/OD280<2.0 were selected. Quality of RNA samples was checked for integrity with the Agilent BioAnalyzer 2100 (Agilent RNA 6000 nano kit, Agilent Technologies, Inc., Santa Clara, CA, USA): samples with a RIN index lower than 8.0 were discarded.

The gene expression profiling was performed using the One-Color Microarray Agilent Platform according to the Agilent protocol (Agilent 8X60K whole mouse genome oligonucleotide microarrays, GRID ID 028005). Data extraction from the Agilent scanner images was accomplished by Feature Extraction software. Data filtering and analysis were performed using Agilent GeneSpring GX 11.0, Microsoft Excel and R-Bioconductor (Limma package). All the features too close to background (with the flag gIsWellAboveBG=0 in raw data files generated by Feature Extraction software) were filtered out and excluded from the following analysis. Filtered data were normalized by aligning samples to the 75th percentile in Log2 scale. Differentially expressed genes were selected by a combination of fold change and moderated T-test thresholds (R Limma-test p < 0.05; fold change ratio transgenic/WT >2.0 OR <0.5 in linear scale). The analysis of over- and under-represented functional gene categories was performed using the DAVID web tool (https://david.ncifcrf.gov) and Gene Set Enrichment

Analysis (GSEA) (http://software.broadinstitute.org/gsea/index. jsp) (Subramanian et al., 2005; Huang et al., 2009).

The following link provides information about how to interpret GSEA data http://software.broadinstitute.org/gsea/ doc/GSEAUserGuideFrame.html?Interpreting\_GSEA. All gene expression microarray data are publicly available in the Gene Expression Omnibus database at the following link: https://www. ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE70757.

## Real-Time qRT-PCR

RNA was isolated, quality controlled as described above, and subjected to quantitative real-time RT-PCR (qRT-PCR) using the two-step iCycler iQ5 Real-Time Detection system (Bio-Rad, USA). For quantification of gene expression changes, the 11Ct method was used to calculate relative fold changes normalized against the housekeeping gene Peptidylprolyl Isomerase A (Ppia). Each data point was obtained from four biological replicates (four mice TgproNGF#3 and four age-matched control mice), each of them in duplicate. Error bars were computed according to the standard error of the mean and the error propagation. The one-tail T-test, assuming equal variances, was used to select significant expression values.

Primers for real time (qRT-PCR) analysis were selected from Harvard Primer database and are listed below.

#### Primers for Real-Time qRT-PCR Analysis

**Kcc2** (**Slc12a5**) (**s**)gggcagagagtacgatggc; (**as**)tggggtaggttggtg tagttg;

**Nkcc1** (**Slc12a2**) (**s**) ttccgcgtgaacttcgtgg; (**as**) ttggtgtgggtgtca tagtagt;

**Calm3** (**s**) tctccctcttcgacaaggatg; (**as**) ggttctgtcccagcgatctc; **Camk2a**(**s**) tggagactttgagtcctacacg; (**as**) ccgggaccacaggttttca; **Dlg4**(**s**) tccgggaggtgacccattc; (**as**) tttccggcgcatgacgtag; **Eif2s1**(**s**) atgccggggctaagttgtaga; (**as**) aacggatacgtcgtctggata; **H3f3a** (**s**) tgtggccctccgtgaaatc; (**as**) ggcataattgttacacgtttggc; **Hist2h4** (**s**) ggtggaaagggtctaggcaag; (**as**) cctggatgttgtcacgca aga;

**Aph1b**(**s**) tcactggaatcagttggctct;(**as**) catccgggaagatgatcagta; **Slc6a13**(**s**) acctgtgagcctggctgt; (**as**) ccaccacagaggggtagttc; **Lama1**(**s**) cagcgccaatgctacctgt; (**as**)ggattcgtactgttaccgtcaca; **Col6a2**(**s**) gatctgttagaccgccatgc; (**as**) cagggctagggtcctattagc; **Col3a1**(**s**) ctgtaacatggaaactggggaaa; (**as**) ccatagctgaactgaaaa ccacc.

#### Specific Primers for BDNF

**BDNF1** (**s**)agtctccaggacagcaaagc (**as**)tgcaaccgaagtatgaaataacc **BDNF2A (s)**gatcccggagagcagagtc**(as)**tctcacctggtggaactgg **BDNF2B (s)**gcggtgtaggctggaataga**(as)**aaggatggtcatcactcttctca **BDNF2C (s)**gtggtgtaagccgcaaaga**(as)**aaccatagtaaggaaaaggatggtc **BDNF3 (s)**gagactgcgctccactcc**(as)**aaggatggtcatcactcttctca **BDNF4 (s)**gctgccttgatgtttactttga**(as)**aaggatggtcatcactcttctca **BDNF5 (s)**gatccgagagctttgtgtgg**(as)**aaggatggtcatcactcttctca **BDNF6A (s)**ctcctgaggaagtgaaagttttg**(as)**aaggatggtcatcactcttctca **BDNF6B (s)**ccgaacaaactgattgctga**(as)**aaggatggtcatcactcttctca **Exon 8** (shared between long and short BDNF mRNAs) **(s)**gcctttggagcctcctctac**(as)**gcggcatccaggtaatttt

**Exon 8 IIPolyA**(upstream the second polyA, amplifies long BDNF mRNAs)

**(s)**gctctcttacccactaagatacatca **(as)**ttttaacaaataaatctcaggtcaaca (according to Liu et al., 2006; Juan et al., 2008).

#### Electrophysiology

For multisite recordings, combined entorhinal cortexhippocampal (EC-hippocampal) slices (350 µm) were cut according to the method described by Jones and Heinemann (1988) and then incubated in ACSF. To evaluate epileptiform activity, EC-hippocampal slices were placed over an 8 × 8 multi-electrode array (MEA) of planar electrodes, each 50 × 50 µm in size, with an interpolar distance of 300 µm (MED-P5305; Alpha MED Sciences, Kadoma, Japan) under visual control, so that signals from each electrode could be assigned according to their relative position within the hippocampal-entorhinal area (Berretta et al., 2012; Tiveron et al., 2013). Slices were kept submerged in ACSF with a nylon mesh glued to a platinum ring and continually perfused in ACSF (6 ml/min) at 34◦C. Voltage signals were acquired using the MED64 System (Alpha MED Sciences, Kadoma, Japan), digitized at 20 kHz and filtered (0.1–1 Hz) with a 6071E Data Acquisition Card (National Instruments, Austin, USA), using Mobius software (Alpha MED Sciences, Kadoma, Japan).

# RESULTS

#### Parvalbumin Interneurons Are Selectively Reduced in the Dentate Gyrus (DG) and Amygdala of TgproNGF#3 Mice

The occurrence of spontaneous epileptiform discharges in the hippocampal network in aged TgproNGF#3 mice (Tiveron et al., 2013) suggests impaired inhibitory interneuron homeostasis. We therefore evaluated different interneuron inhibitory subpopulations in TgproNGF#3 mice hippocampus. A pilot survey in the hippocampus demonstrated that while no major changes were observed in the number of calretinin and calbindin interneurons (Figures S2, S3), the number of Parv+ interneurons appeared to be strikingly affected in the DG.

In wt with the same genetic background of TgproNGF#3 mice, the Parv+ neuron density in the DG (**Figures 1A–D**) was in accordance to the literature (Takahashi et al., 2010; Cardoso et al., 2013). In 3-month-old TgproNGF#3 mice hippocampus, no reduction in the total number of Parv+ cells in the DG was observed (**Figure 1A**). Later, in 6- and 12-months-old TgproNGF#3 mice, the number of Parv+ interneurons was markedly reduced in the DG (47 and 53%, respectively, of the number in age-matched controls (**Figures 1B–D**). Of note, the number of Parv+ neuron was not affected in other hippocampal regions, such as CA1 (Figure S4), of TgproNGF#3 mice, showing that the effect is regional-selective.

Since Active Avoidance behavioral response is severely impaired in TgproNGF#3 mice (Tiveron et al., 2013), the Parvalbuminergic interneuron subpopulation was evaluated also

FIGURE 1 | (A–D) Parvalbumin+ interneuron depletion in the DG of TgproNGF#3 mice. (A–C) Quantification of Parvalbumin+ interneurons (Parv+ interneuron density, expressed as cells/mm<sup>3</sup> ) in the DG of wt and TgproNGF#3 mice (3, 6, and 12 ms-old). Cell counting was performed on at least 4–6 sections per animal, males and females (n = 4 per each animal group). Bars and lines are representative of mean ± SEM (T-test, \*\*P < 0.005; \*P < 0.05). (D) Confocal micrographs of Parvalbumin immunofluorescence in 12-month-old wt and TgproNGF#3 mice hippocampus (DG region). Scale bar: 75 micron. (E,F) Parvalbumin+ interneuron depletion in the amygdala of 12-month-old TgproNGF#3 mice. (E) Quantification of Parv+ interneurons in the basolateral amygdala of 6 and 12-month-old wt and TgproNGF#3 mice (Parv+ interneuron density, expressed as cells/mm<sup>3</sup> ). Cell counting was performed on the lateral region of the amygdala on both sides on at least 4–6 sections per animal, males and females (n = 4 per each animal group) (T-test, \*P < 0.05). (F) Confocal micrographs of Parvalbumin immunofluorescence labeling in the amygdala of 12-month-old wt and TgproNGF#3 mice (Parv+ neurons are located in the lateral region of the amygdala). Scale bar: 75 micron.

in the basolateral amygdala, a region involved in avoidance responses (Choi et al., 2010; Tiveron et al., 2013) and strongly connected to the hippocampus. Interestingly, also in the lateral amygdala a marked reduction of Parv+ interneurons was observed in 6–12 months TgproNGF#3 mice (40 and 44%, respectively, of the number of age-matched control **Figures 1E,F**), whereas no change was observed at 3-months (not shown).

Double labeling with antibodies to Parvalbumin and GAD65- 67 (a marker for GABAergic interneurons that does not strongly label DG principal granule cells, **Figure 2B**, upper panels) showed, in the DG of 12-month-old TgproNGF#3 mice, a reduction in the number of Parv/GAD65-67 double-labeled neurons (**Figures 2A,B**), a sign of neuron loss at this age or of reduced GAD protein expression. However, the change was selectively restricted to the DG, therefore biochemically undetectable. No significant decrease of Parv/GAD65-67 doublepositive interneurons was observed, instead, at 3-months (**Figure 2A**).

However, selective vulnerability of Parv+ interneurons in the DG cannot be ascribed to a higher level of proNGF expression in this hippocampal subregion, as revealed by the very low proNGF immunoreactivity (Figure S5). Determination of transgenic furin-resistant proNGF transcript (vs. the endogenous one) and proNGF/NGF protein levels in the hippocampus and other brain regions were previously reported (Tiveron et al., 2013).

TgproNGF#3 mice show therefore an impaired inhibitory interneuron homeostasis, as revealed by a regional- and cell typeselective Parv interneuron depletion in the DG that might be the cause of the epileptiform discharges observed in the hippocampal network.

### CB Immunoreactivity Is Selectively Reduced in DG Granule Cells and Mossy Fibers of Aged TgproNGF#3 Mice

Calbindin (CB) is another cytosolic Ca++ buffer protein expressed in inhibitory interneurons and in granule cells. We evaluated whether the CB+ interneurons and /or the CB levels were also reduced in the hippocampus of TgproNGF#3 mice. No significant change was observed in the density of CBpositive interneurons in the hippocampus of TgproNGF#3 mice, as revealed by hippocampal calbindin immunoreactivity (Figure S3). Moreover, no significant reduction of CB-immunoreactivity was observed in granule cells in the DG in 3-month-old TgproNGF#3 mice (**Figure 3B,** quantification in **A**). A striking depletion was instead observed in granule cells in the DG of 12-month-old TgproNGF#3 mice (**Figure 3E,** upper panels, quantification in **C**). In the same region, granule neurons number (Figure S8) as well as the GCL area were unaffected, thus demonstrating that the reduced CB-immunoreactivity indicates depletion of CB protein, also observed in CB+ mossy fiber axons projecting to CA3 (**Figure 3E**, lower panels, quantification in **D**). The reduction of CB-immunoreactivity was region-specific, as it was not observed in other hippocampal regions.

### Perineuronal Nets Are Progressively and Selectively Depleted in the DG of TgproNGF#3 Mice

Since Parv+ interneurons are ensheathed by the PNN, playing an essential role in synaptic maturation, we evaluated PNN in the DG of TgproNGF#3 mice (**Figures 4A,B**). The lectin from Wisteria Floribunda (an agglutinin commonly used to map PNN and ECM) only faintly decorates the DG in 12-month-old TgproNGF#3 mice, compared to the much stronger labeling of WT mice (**Figure 4B**), providing evidence for a broad downregulation of ECM protein components, specifically at these important extracellular structures. The deficit appeared to be specific for the DG, as other hippocampal sub-regions appeared unaffected.

Aggrecan is an ECM component of the PNN, expressed in the hippocampus primarily around Parv+ interneurons (McRae et al., 2010). In order to confirm the reduced PNN immunoreactivity, aggrecan expression in the DG was evaluated. Reduced aggrecan immunoreactivity was also observed (**Figure 4C**), confirming the PNN depletion.

Interestingly, the percentage of Parv+ neurons ensheathed by lectin+ PNNs in the DG is significantly and progressively reduced from 6- to 12-months of age in TgproNGF#3 mice (**Figure 4A**), closely paralleling the reduction of total Parv+ interneurons (100% refers to the total amount of Parv+ cells in the DG; the histogram in red indicates the fraction of the total Parv+ neurons ensheathed by the PNN).

#### Transcriptomic Analysis in the Hippocampus of TgproNGF#3 Mice

In order to evaluate whether Parvalbumin interneuron/PNN depletion in the DG of TgproNGF#3 mice is sustained by a modulation of the expression of specific transcripts and gene families, we analyzed the hippocampal transcriptome of TgproNGF#3 mice at different ages (1-, 3-, and 12-months) by microarray analysis. Transcriptional changes are usually known to anticipate the onset of phenotypic alterations, therefore the expression profiling study was performed also at early age (1 month), when neurodegenerative traits and behavioral deficit are not established yet (Tiveron et al., 2013). In 1-monthold TgproNGF#3 mice, only a few mRNAs were differentially regulated (107 transcripts), with 34% up-regulated and 66% down-regulated (**Figure 5A**). Three-months-old TgproNGF#3 mice show a striking global and massive down-regulation of transcription (1645 transcripts modulated, only 15% upregulated and 85% down-regulated) (**Figure 5A**). Similarly, in 12-month-old TgproNGF#3 mice (377 transcripts) 25.2% of differentially expressed mRNAs were found to be up-regulated and 74.8% mRNAs down-regulated (**Figure 5A**). Complete list of differentially expressed mRNAs at 1, 3 and 12 months is included in Supplementary Information Files as Table S1.

Clustering of the differentially expressed genes into functional categories, by gene-ontology tools including DAVID and GSEA analysis (**Table 1** and **Figure 6**), shows that:

**(i)** TgproNGF#3 hippocampus has a clear proNGF-induced transcriptional signature (D'Onofrio et al., 2011; Arisi

et al., 2014), markedly different from the response typically evoked by NGF, which is characterized by a strong induction of transcripts; no NGF-response genes were induced in the hippocampus of TgproNGF#3 mice (see list in Table S2).

**(ii)** At 1-month of age, mRNAs encoding proteins involved in synaptic transmission, such as synaptotagmin X (Syt10), and an ECM component proteoglycan (Prg4) were downregulated, early representatives of two major gene families globally regulated at later ages (**Table 1**).

**(iii)** In 3-month-old TgproNGF#3 mice hippocampus, the most significantly down-regulated categories of mRNAs include

– transcription, chromatin packaging and remodeling – synaptic transmission, LTP, glutamatergic transmission (**Table 1**, validation of selected individual mRNAs in **Figure 5B**)

#### Conversely,

– the up-regulation of ECM genes stands out in contrast to this general scenario of down-regulated transcription.

FIGURE 4 | (A,B) The perineuronal nets (PNN) are markedly reduced in the dentate gyrus of TgproNGF#3 mice. Confocal microscopy. (A) Quantification of Parvalbumin+ interneurons of wt and TgproNGF#3 mice (3-, 6-, and 12-month-old). % total Parv+ neurons − blue bar − and Parv+ neurons ensheathed by the PNN − red bar. The reduction in Parv+ neurons ensheathed by PNN is significant in 6- and 12-month TgproNGF#3 mice. Analysis was performed on at least 4–6 sections per animal, males and females (n = 4 per each animal group) (T-test, \*P < 0.005). (B) Double immunofluorescence for the perineuronal net and for Parvalbumin in the DG of 12-months-old wt and TgproNGF#3 mice. Brain sections were labeled with anti-Parvalbumin antibody, with the lectin Wisteria Floribunda (an agglutinin commonly used to map PNN and ECM, since it binds the glycan component of PNN proteoglycans) and Dapi. Scale bar: 50 micron. (C) Double immunofluorescence for aggrecan (a component of the PNN surrounding Parv+ interneurons) and for Parvalbumin in the DG of 12-months old wt and TgproNGF#3 mice. Scale bar: 50 micron.

**(iv)** Interesting findings derive from GSEA analysis (**Figure 6**). At 1-month functional analysis reveals down-regulation of the Long Term Depression mRNA system, while the ECM gene sets are up-regulated. At 3-months the GSEA plots highlight a general down-regulation of the Long Term Potentiation and Transcription systems, while the Collagen and ECM categories are up-regulated at 3 and down-regulated at 12-months. GSEA analytical tool allows identifying with high sensitivity global up- or downregulation trends affecting large sets of functionally related genes. GSEA plots represent, for each functional category, a kind of summation integral of expression changes for all single genes in the category. The methodology is sensitive even for small changes involving large gene sets. Such findings were confirmed also using a second well-established tool, DAVID, recently updated, for gene Ontology and pathway analysis. These methodologies are

shown. The error bar is the S.E.M. and (\*) indicates statistical significance by

T-test (\*P < 0.05).

also robust to a reasonable rate of false positive differential genes. Both tools reach very similar biological conclusions, though being based on different algorithms, which further support the findings of the study.

Such global analysis provides therefore interesting and significant results.

At 12-months, down-regulation of aggrecan transcript, a PNN component, is detected by microarray and highlighted by GSEA analysis (**Figure 6**), in line with the reduced aggrecan protein expression observed by immunofluorescence (**Figure 4C**).


#### Validation by Real Time QRT-PCR of a Selected Panel of Transcripts

Chloride transporters Nkcc1 and Kcc2, known determinants of E/I homeostasis, essential in establishing GABAergic system maturation, were not found to be modulated in microarray analysis. In order to confirm this data, the mRNA expression of chloride transporters Nkcc1 and Kcc2 in TgproNGF#3 mice hippocampus was evaluated by real-time RT-PCR at 1.5, 3, and 12-months of age. No significant change in the expression pattern was observed (Figure S6). Maturation and function of parvalbuminergic Fast Spiking interneurons critically depend on BDNF (Berghuis et al., 2004; Cancedda et al., 2007). However, microarray analysis showed that global BDNF expression was unchanged in the hippocampus of TgproNGF#3 mice. In order to confirm the lack of variation, the global level of BDNF-encoding mRNA was evaluated by real time qRT-PCR. No major changes were observed (not shown). Several differentially regulated transcripts highlighted by gene-ontology analysis were individually validated by real time qRT-PCR, confirming the down-regulated pattern (**Figures 5B,C**): at 3 months, calmodulin3 (Calm3), calcium/calmodulin-dependent protein kinaseII alpha (CAMK2a), PSD95 synaptic protein [discs large homolog 4 (Drosophila) or Dlg4] and the translation initiation factor eIF2, at 12-months, collagen, type III alpha 1 (Col3a1), collagen, type VI alpha 2 (Col6a2), laminin, alpha 1 (Lama1) (Chung et al., 2005; Cheng et al., 2009; Dityatev et al., 2010; Mercier and Arikawa-Hirasawa, 2012). Finally, at 12-months some components of the ion channels/transporters gene family are modulated, such as the GABA transporter solute carrier family 6 (Slc6a13, also known as GAT2) (Zhou and Danbolt, 2013). Interestingly, the trend observed in array analysis was confirmed by real time qRT-PCR, often in a significant pattern. A few of the modulated transcripts could not be significantly validated by real time qRT-PCR. However, bioinformatic tools for gene Ontology and pathway analysis


TABLE 1 | Comparative functional analysis of differentially expressed mRNAs in the hippocampus of 3- vs. 12-month-old TgproNGF#3 mice.

The analysis of over-represented mRNA categories was performed using the DAVID tool. Only differential genes corresponding to each time points were used. P-values are shown for up and down-regulated genes in separate columns. Categories referring to genes related to collagen (yellow) and extracellular matrix (light blue) have opposite trends at 3-months compared to 12-months of age. Categories referring to similar genes have the same color.

(GSEA and DAVID) highlighted the significant involvement of ECM among the modulated categories.

#### Expression of BDNF-Encoding mRNAs: Selective Modulation of BDNF Transcript Variants

Having established that the parvalbuminergic interneuron population is selectively reduced in the DG of TgproNGF#3 mice, we investigated the possible role of the BDNF neurotrophin. The role of BDNF in interneuron maturation and function is wellestablished (Huang et al., 1999; Cancedda et al., 2007; Sakata et al., 2009), however in TgproNGF#3 mice the global level of BDNF-encoding mRNAs in the hippocampus was unaffected, as revealed by microarray analysis.

The rodent BDNF gene produces several different splicing variants (Figure S7), each composed of one alternatively spliced 5 ′UTR exon linked to a common downstream exon containing the coding region with two possible (either short or long) 3 ′UTRs (Aid et al., 2007). Therefore, we further analyzed the expression of individual BDNF transcript variants by qRT-PCR.

In 1-month-old TgproNGF#3 mice, a significant down-regulation of BDNF transcript I was observed (**Figure 7Aa**)

Conversely, at 3-months of age a significant up-regulation of BDNF splice variants (III and V) was detected in TgproNGF#3 mice (**Figure 7Ba**).

At 12-months of age a significant down-regulation of BDNF splice variants 1 and 2b was detected in TgproNGF#3 mice (**Figure 7Ca)**.

At 1-, 3-, and 12-months the relative proportion of short and long isoforms, regardless of their 5′ splicing pattern, was, on the other hand, unchanged in the two mouse strains (wt and transgenic, **Figures 7A–C**). The distribution of the 3′ short form, evaluated using primers encompassing the common coding exonVIII, confirmed that global BDNF levels were unaffected, as revealed by microarray (**Figures 7A–C**).

Therefore, the observed subtle specific changes in the expression of selected BDNF transcripts might underlie differential BDNF influence on interneuron maturation and homeostasis in TgproNGF#3 mice hippocampus.

FIGURE 6 | Functional analysis of specific gene categories. The whole dataset was analyzed by the tool GSEA (Gene Set Enrichment Analysis). FDR q < 0.05 (statistically significant) is indicated by (\*) (obtained by gene set permutation). The GSEA analytical tool allows to identify small changes affecting a large set of functionally related genes compared to a per gene statistics. GSEA analytical tool allows to identify with high sensitivity global upor down-regulation trends affecting large sets of functionally related genes. GSEA plots represent, for each functional category, a kind of summation integral of expression change for all single genes in the category. The methodology is sensitive even for small changes involving large gene sets. The GSEA plots highlight a general down-regulation of the Long Term Potentiation and Transcription systems at 3-months, while the Collagen and Extracellular Matrix categories are up-regulated at 3-months and down-regulated at 12-months. In each panel, the green line is the GSEA score plot, while the gray plot on the bottom of each panel is the global distribution of Log2 fold change for all genes, with up-regulated genes on the left and down-regulated ones on the right. The black thin bars on the bottom of each sub-panel indicate the genes belonging to the functional category, with each bar position corresponding to the specific score and Log2 fold change levels in the two plots. The following link provides information about how to interpret GSEA data and plot: http://software.broadinstitute.org/gsea/doc/ GSEAUserGuideFrame.html?Interpreting\_GSEA.

# Early Spontaneous Epileptic-Like Events in TgproNGF#3 Mice

Aged TgproNGF#3 mice were previously shown to display spontaneous epileptiform interictal-like discharges in the entorhinal cortex-hippocampal (EC-HP) network, suggesting a role of proNGF/NGF balance in the E/I homeostasis (Tiveron et al., 2013). The age-dependency of this phenomenon was investigated by multielectrode analysis with field recordings, using a 64-channels MEA device, revealing the presence of spontaneous epileptic-like seizures in 1-, 3-, and 12-months-old TgproNGF#3 mice, in combined EC-HP slices. Repetitive spontaneous interictal-like events restricted to the hippocampal region were detected in every slice (n = 10) from 1-month-old TgproNGF#3 mice, while no spontaneous events were detected in the entorhinal area at this age (**Figure 8A**). Similarly, in slices (n = 5) of 3-months-old TgproNGF#3 mice, spontaneous interictal-like events were never observed in the EC, as opposed to a pronounced hippocampal spontaneous hyperactivity (**Figure 8B**). Conversely, slices from 12-month-old TgproNGF#3 mice displayed spontaneous interictal-like events not only in the hippocampus, but also in the entorhinal area (**Figure 8C**), in line with previous reports (Tiveron et al., 2013). Thus, TgproNGF#3 mice display spontaneous epileptic-like events in EC-HP slices starting from 1-month of age, well before the onset of behavioral and neurodegenerative changes.

#### DISCUSSION

The presence of spontaneous epileptic-like events in the HP/EC network, previously described in aged TgproNGF#3 mice (Tiveron et al., 2013), suggested the occurrence of an E/I imbalance, as a consequence of proNGF increase. In order to gain further insights into the mechanisms triggering and sustaining TgproNGF#3 mice phenotype, we investigated early events driven by proNGF/NGF imbalance in these mice.

Our results show that proNGF overexpression alters the interneuron-PNN-system, in a region- and interneuron subpopulation-selective manner. The DG appears to be particularly sensitive to the effects of proNGF.

## Parvalbuminergic Interneurons Are Selectively Depleted in the DG

Selective depletion of parvalbuminergic interneurons was detected in the DG, starting from 6-months of age, despite of no detectable enrichment in proNGF expression in this hippocampal sub-region. Interestingly, other hippocampal areas were preserved.

Selective depletion of Parv+ neurons was observed also in the basolateral amygdala, a region involved in the active avoidance response (Li and Richter-Levin, 2012), severely impaired in TgproNGF#3 mice (Tiveron et al., 2013). The amygdala, a brain region involved in memory processing, particularly in encoding emotional significance of environmental stimuli, strongly connected to other brain structures, receives, in analogy to the pyramidal cells and GABAergic interneurons of the hippocampus, cholinergic inputs from basal forebrain synapses that modulate both excitatory and inhibitory synaptic transmission (Feduccia et al., 2012). The cholinergic deficit previously described in TgproNGF#3 mice (Tiveron et al., 2013), and confirmed in the present study by transcriptomic data, may therefore affect both regions.

3-months of age. (a) Differential expression of main Bdnf isoforms, measured ad Log2 fold change ratio TgproNGF#3 vs. Wild Type; (\*) indicates a statistical significant variation by 1-tail T-test, for Bdnf3 and Bdnf5, while the two rightmost bars indicates the global relative expression of short and long isoforms. (b) Relative distribution of short and long Bdnf forms (left) and of the different isoforms (right) in the TgproNGF and Wild Type mice: the relative proportion of short and long isoforms is comparable in the two mouse strains. (C) Expression level of Bdnf gene isoforms, by qRT-PCR, at 12-months of age. (a) Differential expression of main Bdnf isoforms, measured ad Log2 fold change ratio TgproNGF#3 vs. Wild Type; (\*) indicates a statistical significant variation by 1-tail T-test, for Bdnf1 and Bdnf2B, while the two rightmost bars indicates the global relative expression of short and long isoforms. (b) Relative distribution of short and long Bdnf forms (left) and of the different isoforms (right) in the TgproNGF#3 and Wild Type mice: the relative proportion of short and long isoforms is comparable in the two mouse strains.

3-, and 12-months old TgproNGF#3 mouse, respectively, placed over an array of 64 planar multi-electrodes, detectable in transparency through the slice. Colored dots indicate position of the electrodes, whose voltage signals appear in the corresponding colored area. (A–C right panels) Simultaneous recordings from each one of the 64 electrodes. (A–C insets) Trace records from the HP (red) and EC (green) detected by the colored electrodes in the corresponding photograph, showing no spontaneous activity in the EC of 1- and 3-months old mice, as opposed to spontaneous cortical events at 12-month, while the hippocampal area displays spontaneous events at all 3 ages.

In both DG and amygdala alterations of parv interneurons are reported as consequence of stress, with opposite changes in the two regions affecting excitatory and inhibitory components in a homeostatic balance (Seidel et al., 2008).

Moreover, a number of studies have shown the role of amygdala Parv+ interneurons in fear conditioning and in active avoidance. Parvalbumin interneurons play a default-silencing role in the amygdala during fear memory encoding. When animals acquire a fear memory, the suppressive influence of Parv+ interneurons is relieved, allowing the fear system to respond promptly (Lucas et al., 2016). Parv+ interneuron inhibition shapes the size of neuronal memory ensembles (engram that allows patterns of activity present during learning to be reactivated in the future, Morrison et al., 2016). These findings suggest that, in case of Parv depletion, the basal default silencing activity of Parv+ cells might be disrupted and the aberrant lack of suppression of basal activity might, conversely, affect the encoding of fear memory.

In addition, during acquisition and extinction of an active avoidance task, neural activity in amygdala involves Parv+ interneurons, as revealed by induction of the immediate early gene product c-Fos (Jiao et al., 2015).

Therefore, the observed Parv+ depletion in basolateral amygdala is likely to contribute to the behavioral deficit in avoidance response observed in TgproNGF#3 mice.

The GABAergic Parv+ neurons are surrounded by the PNN (Pizzorusso et al., 2002). In post-natal development, the PNN restricts plasticity at the end of the critical period in the visual cortex and regulates firing of Parv+ neurons, influencing their maturation (Pizzorusso et al., 2002).In the DG of TgproNGF#3 mice, also the percentage of Parv+ neurons ensheathed by PNNs is significantly and progressively reduced. Enhanced neuronal activity is known to cause a decrease in the expression of components of the PNN (McRae et al., 2012), therefore such finding might be related to the aberrant spontaneous epileptiform activity in the EC-HP network. A deteriorated PNN surrounding Parv+ interneurons lays the ground for extrasynaptic movement of receptors and neurotransmitters into the extrasynaptic space (McRae and Porter, 2012), leaving inhibitory interneurons susceptible to increased synaptic reorganization (McRae et al., 2012).

Possible impairment of adult neurogenesis must be taken into account, since intracerebral injection of proNGF inhibits SGZ neurogenesis (Guo et al., 2013). Interestingly, among modulated RNAs there is laminin, alpha 1 (Lama1), an ECM protein component of PNN and structural components of basal laminae found in the fractones (Mercier and Arikawa-Hirasawa, 2012) of the neurogenic niche and contributing to synapse formation (Dityatev et al., 2010).

Moreover, it will be interesting to evaluate, in future analysis, a possible impairment in electrical oscillation patterns in these mice, since Parv+ interneurons are the fast-spiking (FS) population driving oscillation rhythms, including gamma oscillations (Buzsáki and Wang, 2012).

It is known that BDNF crucially controls the functional differentiation and maturation of FS-interneurons (Berghuis et al., 2004); surprisingly the overall expression of BDNF transcript was unchanged in TgproNGF#3 mice hippocampus. However, more subtle changes in selective splicing isoforms of BDNF have been detected in TgproNGF#3 mice (see below).

The selective involvement of the dentate gyrus (DG) deserves some comments. In neurodegenerative models the DG is often selectively compromised (Palop et al., 2005). The DG represents a crossroad, receiving the perforant path as main excitatory input, which funnels distinctly unidirectional progression of excitatory activity arriving from other brain regions to the trisynaptic hippocampal circuit. These excitatory synaptic inputs are complemented by cholinergic, GABAergic, noradrenergic, dopaminergic, and serotonergic projections (Perederiy and Westbrook, 2013). The adult brain is in a continuous state of remodeling. This is particularly true in the DG, where competing forces, such as neurodegeneration and neurogenesis, dynamically modify neuronal connectivity and can occur simultaneously (Perederiy and Westbrook, 2013). Altered or aberrant activity in such a critical node might alter the excitation/inhibition balance in the DG (as it is known to occur after lesion of the perforant path, Clusmann et al., 1994).

#### Calbindin Depletion in DG Granule Cells

In neurodegeneration models Ca++ buffer proteins are often altered in the DG: hAPP mice (a mouse model of AD) develop AD-like abnormalities, including depletions of calcium-related proteins in the DG, spontaneous non-convulsive seizure activity in cortical and hippocampal networks, associated with synaptic plasticity deficits in the DG (Palop et al., 2007). Moreover, epileptiform activity can also lead to depletion of calciumdependent proteins over time (Palop et al., 2011). Calbindin+ neurons were therefore evaluated in the same region.

In aged TgproNGF#3 mice, a marked depletion of calbindin (CB)-protein in granule cells and in their axons projecting to CA3 (Mossy fibers) was observed. Recent evidence that removal of CB from amyloid precursor protein/presenilin transgenic mice aggravates AD pathology, suggests a critical role of CB (Kook et al., 2014). Moreover, in late-stage AD, a higher ratio of CBnegative granule cells is detected in the DG (Stefanits et al., 2014). In TgproNGF#3 mice we previously described a learning and memory deficit starting from 3-months of age (Tiveron et al., 2013). The late calbindin deficit observed in the DG may contribute to aberrant neuronal activity. Evaluation of DG-LTP will be necessary in order to investigate a selective plasticity deficit. No LTP deficit was previously detected in CA1 (Tiveron et al., 2013).

#### Hippocampal Transcriptional Profiling Reveals a Specific proNGF-Induced Signature, Lacking NGF-Response Genes Induction, with Broad Early Down-Regulation of Transcripts

The most remarkable finding of transcriptome analysis is the global down-regulation of mRNAs expression in the hippocampus of TgproNGF#3 mice in early neurodegeneration (at 3-months), a trend that already begins at 1-month. Consistently, a broad reduction in the expression of genes involved in regulation of transcription and chromatin remodeling is observed. Notably, the expression of mRNAs known to be heavily regulated by NGF (Egr1, Egr2, Egr4, Fos, Jun, Arc, Myc, Vgf) (whose induction represents, on the contrary, a typical "NGF signature," see Dijkmans et al., 2009), is not changed in TgproNGF#3 mice hippocampus, despite the presence of mature NGF, consequent to cleavage by extracellular proteases, as previously measured (Tiveron et al., 2013). This confirms that proNGF signaling predominates over that of NGF in these mice, confirming what had been shown in cellular systems exposed to either NGF or proNGF added singularly or in various combinations (D'Onofrio et al., 2011; Arisi et al., 2014).

Notably, hippocampal transcriptional profiling reveals, a clear dominant "proNGF signature," with broad down-regulation of transcription, whereas lacks completely the classical "NGF signature," characterized typically by IEG induction, followed by an up-regulation of their target genes (D'Onofrio et al., 2011).

### Gene Ontology Analysis Highlights Marked Down-Regulation of Synaptic Transmission-Related Genes

Down-regulated transcript categories include those related to synaptic transmission and synaptic plasticity, such as Calm3 (principal mediator of Ca++ signal, essential for CAMK activation), CAMKIIa, Dlc4 (better known as PSD95, main constituent of the post-synaptic compartment, essential for spine stability) and EIF2, involved in translational control of synaptic plasticity, acting on local protein synthesis (Toutenhoofd and Strehler, 2002; Bingol et al., 2010; Borck et al., 2012; Cane et al., 2014). Interestingly functional analysis of differential categories reveals down-regulation of Long Term Depression system at 1 month and of Long Term Potentiation at 3-months. It would be interesting to evaluate in future electrophysiological studies both LTD and LTP in this model.

Specific effects of proNGF on such crucial protein targets, essential for synaptic plasticity and for the establishment of longterm memory, may well sustain the behavioral deficits previously described in this model (Tiveron et al., 2013).

In TgproNGF#3 mice the observed spontaneous epileptic-like discharges could be due, in principle, to changes in post-natal maturation of Cl<sup>−</sup> homeostasis, known to determine GABAergic system post-natal maturation. However, the expression of the cation-chloride cotransporters Nkcc1 and Kcc2 mRNA was not significantly altered in TgproNGF#3 mice HP. The most interesting insights of expression profiling derive from the analysis of the globally modulated categories, more than of single transcripts: LTP—related or synaptic transmission related transcripts and ECM, as described below.

## ECM Components Show Initial Up-Regulation Followed by Late Down-Regulation

In aged transgenic mice fewer genes were differentially modulated. The most striking transcriptional change involves ECM transcripts: interestingly, the trend is opposite in early and late neurodegeneration. Initial up-regulation of ECM transcripts at 3-months, possibly contributing to impaired plasticity, stands out in the context of a general mRNA down-regulation scenario, and is followed by their severe down-regulation at 12-months. Human genetic studies and analysis of transgenic mice deficient in ECM molecules link ECM molecules to epileptogenesis (Suzuki et al., 2002; Dityatev, 2010; Geissler et al., 2013). Our results place proNGF as a regulator of this link.

Interestingly, a consistent down-regulation of a number of collagen mRNAs, alongside the regulation of a number of ECM related mRNAs was observed in TgproNGF#3 mice, a finding which could be mechanistically related to structural alterations in the GABAergic inhibitory network and to a disruption of the excitatory/inhibitory balance. Indeed, the ECM is at the crossroad of circuit development, reshaping synaptic plasticity and excitatory/inhibitory balance in the nervous system. It is tempting to suggest that the ECM dysregulation uncovered by this transcriptional profile in TgproNGF#3 mice might be related to the neurological aspects of matrix diseases. Further studies will be required to investigate this aspect further.

#### Alterations in the Expression Of Selected BDNF Splice Variants

Taking into account that BDNF crucially controls the functional differentiation and maturation of FS-interneurons (Berghuis et al., 2004; Cancedda et al., 2007) we evaluated changes in the expression of BDNF splice-variants pattern.

BDNF splice variants are characterized by differential distribution in brain regions and in neuronal subcellular compartments (somatic vs. dendritic); their pattern is modulated in response to various stimuli and activity-dependent BDNF mRNA localization in dendrites is observed. Splice variant patterns constitute therefore a "spatial and temporal code" directing BDNF expression locally. Such regulation explains the contradictory effects of BDNF, which may oppose or promote epileptogenesis (Tongiorgi et al., 2006; Chiaruttini et al., 2008; Sakata et al., 2009). In the hippocampus, in basal conditions, the main BDNF dendritic variant is BDNF6 (although other variants are expressed in smaller amounts: 7 in CA1; 1, 6, and 9a in CA3; and 5, 6, 7, and 8 in DG). Stimuli such epileptogenesis prompt dendritic accumulation of variants 4 and 6 (and also, in small amounts, in selected subregions, of BDNF2, 3, and 9a) (Chiaruttini et al., 2008; Baj et al., 2013). Interestingly, in prefrontal cortex BDNF4 is known to play an essential role in GABA interneuron homeostasis (Sakata et al., 2009). Notably, proNGF selectively alters the complex pattern of BDNF transcripts, up-regulating BDNF3 and BDNF5 splice variants in 3-month-old TgproNGF#3 mice hippocampus. In the brain, BDNF3 is the only splice variant also expressed by microglial cells (Kruse et al., 2007) and is known to participate in macrophages activation process in an autocrine manner (Asami et al., 2006). The up-regulation of isoform BDNF3 at 3-months, suggests therefore an interesting involvement of microglial signaling. Future characterization of microglial phenotype is therefore necessary to address their hypothesized role. It is known that pilocarpine-treatment inducing status epilepticus changes the pattern of hippocampal BDNF mRNA variants in rat, selectively increasing transcript encoding exon3 in the DG (Baj et al., 2013). Notably, BDNF3 up-regulation is also induced by endothelin1 (ET1) (Böhm and Pernow, 2007), whose intraventricular injection evokes epileptic seizures, apparently mediated by its vasoconstrictor effect (Koyama et al., 2005). The hypothesis that up-regulation of BDNF3 in microglial cells, as a consequence of proNGF/NGF imbalance, contributes to

Frontiers in Molecular Neuroscience | www.frontiersin.org February 2017 | Volume 10 | Article 20 |

E/I imbalance is intriguing and need to be addressed in future studies.

Consequences of up-regulation of BDNF5 (at 3-months of age) and down-regulation of BDNF1 (at 1- and 12 months) and BDNF2b (at 12-month), also observed in TgproNGF#3 mice hippocampus, require further investigation.

#### Spontaneous Epileptiform Discharges in the HP Network Are a Very Early Event

We observe the presence of spontaneous epileptic-like events in the HP network, previously described in aged TgproNGF#3 mice, in the hippocampus as early as at 1-month of age, when (as reported in Tiveron et al., 2013), the cholinergic deficit is not established yet, but proNGF is already accumulating. These spontaneous discharges spread also to the EC network at later ages (12-months). Therefore, their onset precedes the learning and memory deficit observed at 3-months of age, suggesting that E/I imbalance is a primary and direct consequence of proNGF/NGF unbalanced signaling. However, TgproNGF#3 mice do not exhibit frank spontaneous seizures; possible increased susceptibility to seizures will be evaluated in the future.

#### Cellular Targets of proNGF Action

The known cellular targets of proNGF action are cells expressing p75NTR. Hippocampal expression of p75NTR in principal neurons is weak or absent (Dougherty and Milner, 1999).

Parv+ interneurons in the DG do not express p75NTR , whereas they are physiologically known to express NGF (Holm et al., 2009; Biane et al., 2014) and to feed the NGF-dependent BF cholinergic fibers, expressing both TrkA and p75NTR, that make synaptic contacts with them in the DG (**Figure 9**), predominantly with processes (Dougherty and Milner, 1999; Ludkiewicz et al., 2002). Parv+ interneurons do not express trkA, but are known to express trkB, being responsive to BDNF (Holm et al., 2009). However, the overall expression of BDNF is not affected in TgproNGF#3 mice hippocampus. Interestingly, some changes in BDNF splicing variants were observed: a different pattern in BDNF splicing variants may account, at least in part, for alterations in Parv+ interneurons homeostasis. A cholinergic deficit starting from 3-months was detected in TgproNGF#3 mice (Tiveron et al., 2013) as confirmed also by hippocampal transcriptomic data (not shown). Pathological conditions characterized by increased levels of proNGF in the brain, might lead to a reduced cholinergic drive to Parv+ interneuron, with the ensuing E/I imbalance (**Figure 9B**). How

FIGURE 9 | Schematic representation of hypothesized targets of proNGF/NGF imbalance. (A) We hypothesize that proNGF/NGF imbalance primarily impacts Parv+ interneurons that make inhibitory synaptic contact with DG granule cells that project their axons (the so-called Mossy fibers) to CA3 neurons. These interneurons normally produce NGF and feed cholinergic fibers deriving from NGF-dependent BFN expressing p75NTR, making synaptic contact with Parv interneuron dendrites in the DG. In such a way, cholinergic input tunes Parv+ interneuron inhibitory output onto granule cells, affecting their excitatory activity to CA3 neurons (mediated by Mossy fibers). (B) Overexpression of furin-resistant proNGF, producing an NGF/proNGF imbalance, severely impairs cholinergic fibers, disrupting this feed-back control mechanism influencing hippocampal memory encoding and retrieval and contributing to E/I imbalance. I, interneuron; Parv+, parvalbuminergic interneuron; GC, granule cell; DG, dentate gyrus; BFN, Basal Forebrain Neuron.

astrocytes, which are also a potential target of proNGF actions, contribute to this mechanism is being currently investigated.

Taking into account the time-scale of the observed cholinergic deficit (measured as reduced counting of ChAT+ neurons, starting from 3-months of age, Tiveron et al., 2013), it precedes the selective depletion of Parv+ interneurons. However, it seems difficult to explain the early onset of epileptiform discharges in the hippocampus (at 1-month of age), as a consequence of an interneuron dysfunction occurring much later, according to the timeline of progression of TgproNGF#3 phenotype. Moreover, we previously showed (Tiveron et al., 2013) that cholinergic impairment is not present at 1-month of age, therefore it cannot represent the "primum movens" responsible of the supposed hyperexcitability suggested by MEA results, although we cannot exclude that a reduced functional cholinergic drive begins even earlier.

To better understand the hypothesized cycle linking proNGF, neurodegeneration and E/I homeostasis, further electrophysiological evaluation at early stages of proNGF/NGF imbalance is needed. A functional study directly addressing the effective E/I ratio is therefore expected to provide more clues about the real upstream driver.

Moreover, the selective involvement of the DG in alterations of Parv+ neurons/PNN system, jointly with transcriptomic data, showing down-regulation of synaptic transmission-related transcripts, suggest future analysis of DG–LTP, frequently altered in neurodegeneration models (Houeland et al., 2010).

#### CONCLUSIONS

Neuronal network hyperexcitability and cognitive dysfunction have been detected in mouse models of AD and associated to depletion of Ca ++ dependent proteins and to inhibitory interneuron deficits or hippocampal remodeling (Palop et al., 2011; Verret et al., 2012). Reported experimental evidence favors a view of E/I imbalance as a consequence of amyloidogenic process activation (Palop and Mucke, 2009; Palop et al., 2007; Palop and Mucke, 2010; Harris et al., 2010), but also as upstream driver of neurodegeneration and cognitive impact (Sanchez et al., 2012), since antiepileptic treatment was shown to slow down the cognitive decline in transgenic models of neurodegeneration and, recently, in MCI subjects (Sanchez et al., 2012; Bakker et al., 2015).

Our results provide further mechanistic insights into the negative cycle linking proNGF, neurodegeneration and E/I homeostasis (Tiveron et al., 2013).

#### REFERENCES


In summary, we observed a regional- and cellular-selective Parvalbumin interneuron and PNN depletion in the DG, but not in other hippocampal regions. These results demonstrate that in the hippocampus the DG is selectively vulnerable to altered proNGF/NGF signaling. Parvalbumin interneuron depletion is also observed in the amygdala, a region strongly connected to the hippocampus and likewise receiving cholinergic afferences.

However, the onset of spontaneous discharges in the hippocampal network precede those changes, suggesting E/I imbalance represents a primary event, immediately following proNGF accumulation. The observed alterations in the expression of selected BDNF splice variants might be directly involved; the impact of those changes on E/I balance and how they relate to proNGF increase are intriguing questions that need to be addressed further.

#### AUTHOR CONTRIBUTIONS

LF and RB are joint first authors. LF, RB, and AC: designed experiments. LF, RB, FL, IA, NB, MD, and AC: performed experiments. LF, RB, IA, FL, NB, SC, MD, and AC: analyzed data. LF and AC wrote the manuscript. All authors read, reviewed, and commented the manuscript. LF and IA prepared the figures.

#### FUNDING

This work was funded by the following grants: FIRB RBAP10L8TY from the Italian Ministry of Higher Education and Scientific Research; Fondazione Roma; PAINCAGE FP7 Collaborative Project number 603191; Fondazione Italiana Sclerosi Multipla FISM 2013/R/6; Italian Research Council (Framework Agreement EBRI-CNR 2015–2017) and European grant Horizon H2020-ICT-2016, "MADIA," number 1732678.

#### ACKNOWLEDGMENTS

We gratefully acknowledge Prof. Enrico Cherubini for fruitful discussions, Dr. Silvia Marinelli and Dr. Simone Pacioni for kind advice. We thank Dr. Sabrina Turturro and Dr. Andrea Ennio Storti for technical help in gene expression experiments and Dr. Gianluca Amato for genotyping.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00020/full#supplementary-material

phenotype. J. Neurosci. 32, 2002–2012. doi: 10.1523/JNEUROSCI.1144- 11.2012


neurotrophic factor. Biochem. Biophys. Res Commun. 344, 941–947. doi: 10.1016/j.bbrc.2006.03.228


cells generate the field potential oscillations induced by cholinergic receptor activation in the hippocampus. J. Neurosci. 30, 15134–15145. doi: 10.1523/jneurosci.4104-10.2010


mutant mouse and in Alzheimer's disease. Brain Struct. Funct. 214, 145–160. doi: 10.1007/s00429-010-0242-4


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Fasulo, Brandi, Arisi, La Regina, Berretta, Capsoni, D'Onofrio and Cattaneo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Insulin-Like Growth Factor 1: At the Crossroads of Brain Development and Aging

Sarah Wrigley <sup>1</sup> , Donia Arafa<sup>1</sup> and Daniela Tropea2,3 \*

<sup>1</sup>School of Medicine, Trinity College Dublin, Dublin, Ireland, <sup>2</sup>Neuropsychiatric Genetics, Trinity Translational Medicine Institute St. James Hospital, Dublin, Ireland, <sup>3</sup> Institute of Neuroscience, Trinity College Dublin, Dublin, Ireland

Insulin-like growth factor 1 (IGF1) is a polypeptide hormone structurally similar to insulin. It is central to the somatotropic axis, acting downstream of growth hormone (GH). It activates both the mitogen-activated protein (MAP) kinase and PI3K signaling pathways, acting in almost every tissue in the body to promote tissue growth and maturation through upregulation of anabolic processes. Overall GH and IGF1 signaling falls with age, suggesting that it is this reduced IGF1 activity that leads to age-related changes in organisms. However, mutations that reduce IGF1-signaling activity can dramatically extend the lifespan of organisms. Therefore, the role of IGF1 in the overall aging process is unclear. This review article will focus on the role of IGF1 in brain development and aging. The evidence points towards a role for IGF1 in neurodevelopment both prenatally and in the early post-natal period, and in plasticity and remodeling throughout life. This review article will then discuss the hallmarks of aging and cognitive decline associated with falls in IGF1 levels towards the end of life. Finally, the role of IGF1 will be discussed within the context of both neuropsychiatric disorders caused by impaired development of the nervous system, and neurodegenerative disorders associated with aging. IGF1 and its derivatives are shown to improve the symptoms of certain neuropsychiatric disorders caused by deranged neurodevelopment and these effects have been correlated with changes in the underlying biology in both in vitro and in vivo studies. On the other hand, studies looking at IGF1 in neurodegenerative diseases have been conflicting, supporting both a role for increased and decreased IGF1 signaling in the underlying pathogenesis of these diseases.

#### Edited by:

Hansen Wang, University of Toronto, Canada

#### Reviewed by:

Enrique Cadenas, University of Southern California, USA Carlos Vicario-Abejón, Spanish National Research Council, Spain Derek LeRoith, Icahn School of Medicine at Mount Sinai, USA

> \*Correspondence: Daniela Tropea tropead@tcd.ie

Received: 23 November 2016 Accepted: 16 January 2017 Published: 01 February 2017

#### Citation:

Wrigley S, Arafa D and Tropea D (2017) Insulin-Like Growth Factor 1: At the Crossroads of Brain Development and Aging. Front. Cell. Neurosci. 11:14. doi: 10.3389/fncel.2017.00014 Keywords: insulin-like growth factor 1, aging, neurodevelopment, growth factors

Insulin-like growth factor 1 (IGF1) signaling is an essential factor for early brain development, but its role in the aging brain remains unclear. In fact, while reduced IGF1 signaling with age has been historically observed as a causative factor in the aging process, correlation does not imply causation, and falling IGF1 signaling with age may in fact attenuate the effects of aging. Further work must be done in order to truly discern the role and effects of IGF1 signaling in the aging brain.

#### INSULIN-LIKE GROWTH FACTOR-1: DOWNSTREAM SIGNALING CASCADES AND CELLULAR EFFECTS

IGF-1 is synthesized primarily in the liver, where its synthesis is regulated by pituitary secretion of growth hormone (GH). It is central to the somatrotropic axis, acting downstream of GH to promote anabolic processes and tissue growth throughout life. IGF1 is also synthesized locally in many organs, including the brain. Both in circulation and in tissues, IGF1 is bound to high affinity IGF1-binding proteins (IGFBPs), which modulate interactions between IGF1 and its receptor.

The biological actions of IGF1 are mediated through IGF1R, a membrane-bound receptor tyrosine kinase (RTK). Binding of IGF1 to its receptor causes autophosphorylation of the intracellular component, leading to enzymatic activation and subsequent phopshorylation of the insulin receptor substrate-1 (IRS1) protein on multiple tyrosine sites. These phosphotyrosine sites then serve as docking sites for numerous intracellular signaling proteins. By bringing these interacting proteins together, complex signaling pathways are begun, including the canonical PI3-kinase and mitogen-activated protein (MAP) kinase pathways.

IGF-1R activation triggers the PI3kinase-Akt signaling pathway, which promotes cell growth and maturation. IRS1 binds PI3kinase which phosphorylates PIP2 to PIP3. PIP3 binds two protein kinases; Akt and PDK1, leading to activation of Akt, which acts on numerous proteins throughout the cell to promote cell growth and survival. Downstream substrates of Akt include mammalian target of rapamycin (mTOR), which stimulates ribosome production and protein synthesis, and Bad, a pro-apoptopic protein that is inhibited by Akt phosphorylation. Another effect of IGF1R activation through PI3kinase-Akt signaling is inhibitory phosphorylation of pro-apoptopic glycogen synthase 3β (GSK3β), which is associated with increased glycogen storage in projection neurons in the post-natal brain, as well as reduced tau hyperphosphorylation which causes neuronal death (Bondy and Cheng, 2004). IGF1-induced PI3K-Akt signaling is also linked to production of GLUT4 glucose transporters and translocation to neuronal cell membranes, promoting glucose uptake into neurons (Bondy and Cheng, 2004). This corresponds to studies demonstrating increased glucose utilization in areas of higher IGF1 and IGF1R expression in the developing brain, and reduced glucose utilization in IGF1 null brains (Cheng et al., 2000). This pathway also leads to inactivation of FOXO1, preventing FOXO-driven transcription of pro-apoptopic genes (Yin et al., 2013). Hence, PI3K signaling directly inhibits the pro-apoptopic machinery via multiple pathways.

IGF1R also phosphorylates the Sch protein, which recruits the GDP2/SOS complex, leading to activation of Ras, thereby triggering the MAP kinase pathway central to growth-related gene transcription and mitogenesis (**Figure 1**).

# IGF1-SIGNALING, AGING AND LIFESPAN IN MODEL ORGANISMS

IGF1 signaling is central to pathways that promote cell growth and survival, maturation and proliferation, allowing for tissue growth and renewal. Furthermore, the activity of the GH-IGF1 somatotrophic axis decreases with age, and is almost undetectable in people over 60 years (Junnila et al., 2013). This has led to many theories that upregulation of the GH/IGF1 pathway may delay aging. However, studies of IGF1 signaling on nematodes and many other species have suggested that, on the contrary, downregulation of IGF1 signaling delays aging and increases lifespan.

The IGF1 signaling pathway is an evolutionarily ancient pathway, conserved from C. elegans through to modern humans. The C. elegans IR-IGF1 receptor homolog is Daf-2. Mutations that lower the level of Daf-2 double the lifespan of the C. elegans model (Kenyon et al., 1993). Exactly how Daf-2 mutations increase C. elegans lifespan in unclear. Some daf-2 mutants adopt a quiescent state of reduced movement and fertility known as the dauer state, whereas other mutants are shown to have a lower metabolic rate, and other mutants again demonstrate a metabolic shift to fat production. However, these findings are not consistent among all daf-2 mutants, and can therefore not be coupled to lifespan extension (Kenyon, 2010). Another theory is that daf-2 mutants are better able to withstand oxidative stress (Holzenberger et al., 2003). Reduced Daf-2 activity downregulates Akt-mediated inhibition of Daf-16 (FOXO homolog), allowing Daf-16 translocation to the nucleus for target gene activation. The transcriptional targets of Daf-16/FOXO may be at least in part responsible for the stress resistance and longevity associated with Daf-2 mutants (Lin et al., 1997; Gami and Wolkow, 2006).

The impact of altered Daf-2 activity on lifespan varies between different cell lineages. While present in ectodermal, mesodermal and endodermal lineages, it is reduction in ectodermal (neuronal, skin tissue) daf-2 activity that induces a dauer state (Guarente and Kenyon, 2000). Restoration of insulin-like signaling alone in Daf-2 mutants reverts these mutants back to a wild-type lifespan, whereas restoration of insulin-like signaling in muscle or adipose tissue had no effect on lifespan (Wolkow et al., 2000). This study provides persuasive evidence that insulin/IGF1 signaling in neurons regulates lifespan. Interestingly, reduction in sensory input to the olfactory system by mutations in sensory cilia or olfactory support cell ablation can increase lifespan by up to 50% without affecting development or reproduction. This suggests that sensory neurons influence lifespan and this is at least partly mediated through Daf-2 signaling (Apfeld and Kenyon, 1999). Downstream of Daf-2 (IGF1R), mutations in age-1 (PI3K homolog) also increase lifespan, suggesting that reduction in factors downstream of IGF1 signaling are sufficient to extend lifespan (Morris et al., 1996).

The effect on lifespan by inhibition of insulin/IGF1 signaling observed in the C. elegans model is conserved in other species,

including the Drosophila fly, whereby inhibiting IGF1 signaling or increasing the activity of FOXO (the Daf-16 homolog) in adipose tissue increases lifespan (Kenyon, 2010). Heterozygous mutation of the insulin receptor (IR) in Drosophila extends lifespan by up to 85% (Tatar et al., 2001). Furthermore, mutation of CHICO, the IRS homolog, extends lifespan by 48% in homozygotes and 36% in heterozygotes (Clancy et al., 2001). A striking inverse correlation between IGF1 levels and lifespan is also observed in mice (Kenyon, 2010). While IGF1R null mice die shortly after birth (Liu et al., 1993), heterozygous knockout (KO) of IGF1R extends lifespan by 26% compared to wild-type littermates (Holzenberger et al., 2003). This mouse model of reduced IGF1 signaling were normal in size, and displayed normal energy metabolism, but showed greater resistance to oxidative stress (Holzenberger et al., 2003).

Interestingly, this is in contrast to mice with IR or IRS KO, which show severe insulin resistance and die earlier due to hyperglycemia. From invertebrates to mammals, IGF1 signaling and insulin signaling became distinct cellular pathways with different downstream effects. Therefore, IGF1 signaling in mammals can be manipulated without interfering with systemic glucose metabolism.

In humans, lowered IGF signaling is also shown to improve longevity. Mutations known to impair IGFR function are observed in Ashkenazi Jewish centenarians (Suh et al., 2008). Additionally, Akt, FOXO3A and FOXO1A mutations are linked to longevity in numerous patient cohorts (Willcox et al., 2008; Flachsbart et al., 2009; Kenyon, 2010).

The findings that reductions in Daf-2/IGF1R signaling can radically increase lifespan would suggest that IGF1 signaling is directly linked to the aging of organisms, which is in contradiction to the theory that a fall in the activity of GH-IGF1 somatotropic axis underlies the mechanism of aging.

Although in the whole organism a general decrease of IGF signaling delays aging, this review will focus on the role of IGF1 signaling in the developing and aging brain, where IGF1 signaling in the brain promotes development and, in some cases, appears to attenuate age-related changes. Numerous studies outlined below demonstrate the neurotrophic effects of IGF1 signaling, giving evidence for promotion of neurogenesis, development and maturation, myelination, prolonged survival and resistance to injury.

## IGF1 AND THE DEVELOPING BRAIN: EXPRESSION PATTERNS OF IGF1 AND IGF1R IN THE DEVELOPING BRAIN

IGF1 is produced by all major cell types in the CNS. IGF1 expression peaks perinatally and falls throughout life, though it persists in discrete brain regions associated with continual renewal and remodeling. This is in contrast to the IGF1 receptor, which is shown to be widely expressed throughout the brain, and concentrated in neuron-rich areas including the granule cell layers of olfactory bulb, dentate gyrus and cerebellar cortex, with little hybridization in white matter regions (Bondy et al., 1992a). IGF1R is expressed in all neuroepithelial cell types, and shows a relatively stable pattern of expression from early development to maturity (Bondy et al., 1992b). There is, however, a period of increased IGF1R expression coinciding with increased IGF1 expression in specific sensory and cerebellar projecting neurons during late post-natal development (Bondy et al., 1992b). Early post-natal IGF1 expression was identified in brain regions where neurogenesis persisted after birth, including the cerebellum, olfactory bulb and hippocampus, and was shown to fall after this period of neuronal proliferation (Bach et al., 1991; Bartlett et al., 1991).

IGF1 expression persists in these areas in adult brains, but at levels much lower than that of early neonatal animals (García-Segura et al., 1991). These expression patterns highlight the association between regions of increased neurogenesis and local IGF1 and IGF1R expression during early development.

While local IGF1 expression falls shortly after birth, there exists throughout life an active transport mechanism that allows peripheral circulating IGF1 to cross the blood brain barrier (Fernandez and Torres-Alemán, 2012). This finding, combined with the fact that IGF1 receptor expression in the brain persists throughout life, suggests a role for peripherally produced IGF1 in adult brain function. Therefore, while locally produced IGF1 appears to play a role in brain function during prenatal and early post-natal development, peripherally produced IGF1 may have a continued role in the adult brain.

## IGF1 AND THE DEVELOPING BRAIN: EVIDENCE FROM IN VITRO AND IN VIVO STUDIES ON THE EFFECT OF IGF1 ON BRAIN SIZE, NEURONAL CELL NUMBER, AXONAL GROWTH AND MYELINATION DURING EARLY DEVELOPMENT

IGF1 has been shown to have pleiotropic actions in all neural cells, including neurons, oligodendrocytes and astrocytes, by increasing cell number and promoting maturation and myelination. This has been proven through in vitro culture studies, and through both overexpression and under-expression studies in transgenic mice.

# IGF1 and the Developing Brain: Effects of IGF1 on Neural Stem Cells

In vitro studies examining the effect of IGF1 in neural stem cells report increased neural progenitor cell proliferation and maintenance in cell culture following treatment with IGF1 (Drago et al., 1990; Supeno et al., 2013). IGF1 was found to be more potent than insulin in stimulating mitosis of sympathetic neuroblasts (DiCicco-Bloom and Black, 1988). Furthermore, numbers of neurons produced from neural stem cell clones was increased following administration of IGF1 or heparin or withdrawal of FGF2 from cell culture, the effects of which were negated by administration of IGF1 and IGF1BP antibodies (Brooker et al., 2000). Examination of adult rat hippocampal progenitor cells showed uniform IGF1R expression, and the combined addition of IGF1 and FGF2 to these cells in culture increased DNA synthesis and cell division, without significant changes in the rate of cell death (Åberg et al., 2003).

# IGF1 and the Developing Brain: Evidence from Prenatal Overexpression and Underexpression Studies

Popken et al. (2004) demonstrated through nestin-driven overexpression of IGF1 early in the embryonic development of transgenic mice a 6% increase in brain weight by embryonic day 16, associated with a cortical plate volume 42% greater and a total cell number 54% greater in the transgenic mice compared to controls. This increase in total cell number was attributed to a 15% increase in proliferating cells in the ventricular and subventricular zones of the embryonic cerebral cortex, which give rise to the neuronal and glial cell types respectively. At post-natal day 12, a 27% increase in overall brain weight was observed, with significant increases in volume and total cell number in certain brain regions including the cerebral cortex, subcortical white matter, caudate-putamen, hippocampus, dentate gyrus and habernacular complex (Popken et al., 2004). This enhancement in neuroepithelial cell proliferation by IGF1 during embryonic neurogenesis has been shown to result from acceleration through the cell cycle (Hodge et al., 2004). Further studies in the same mouse model reported that, in ddition to enhanced proliferation of neural progenitors, there are reduced numbers of apoptopic cells throughout the cerebral cortex both prenatally and post-natally in the mice showing IGF1 overexpression, correlating with overall increased brain weight 9 months post-natally (Hodge et al., 2007).

Conversely, examination of homozygous IGF−/<sup>−</sup> KO mice at 2 months demonstrates a grossly structurally normal brain, but a 38% reduction in brain weight. A reduction in parvalbumincontaining neurons in the hippocampus and striatum was shown, as well as a significant reduction in the thickness of the dentate gyrus granule cell layer (Beck et al., 1995). IGF1−/<sup>−</sup> mice also show reduced dendritic length and complexity, and a 16% reduction in synaptotagmin levels, suggesting a reduction in the number of synapses, in the frontoparietal cortex (Cheng et al., 2003). A study by Liu et al. (2009) showed that brainspecific IGF1R heterozygous KO mice had brain weights 56% lower than controls at birth and 60% lower at post-natal day 90. The hippocampus was particularly affected, where the rate of growth post-natally was much lower in IGF1R heterozygous KOs compared to wild-type controls, with a greater fall in the numbers of neurons in the CA1–3 region and a smaller rise in the neuronal cell number in the dentate gyrus postnatally. This was attributed to a higher rate of cell death in IGF1R heterozygous KO mice. In all experiments performed, the phenotype was more severe in the two IGF1R homozygous KO mice that survived to adulthood, suggesting that degree of brain growth retardation was related to the level of IGF1R expression (Liu et al., 2009).

### IGF1 and the Developing Brain: Evidence from Post-Natal Overexpression and Underexpression Studies

While the above studies examined the downstream effect of prenatal IGF1 over-and under-expression on subsequent brain development, other studies had focused on the role of IGF1 in the post-natal period, given that IGF1 mRNA expression in the developing rodent brain peaks in the first two post-natal weeks of life. Using a line of transgenic mice that begin to express the transgene at birth, causing brain-restricted overexpression of IGF1, studies have shown that IGF1 overexpression causes increases in brain weight after day 10, with enlargement of the brainstem, cerebellum, cerebral cortex and hippocampus (Ye et al., 1996; Dentremont et al., 1999; O'Kusky et al., 2000).

Cerebellar weight was increased by 90%, with concomitant increases in granule and Purkinje cell numbers by 82% and 20% respectively (Ye et al., 1996).

Specifically, given the persistence of IGF1 expression in the hippocampal subventricular zone and dentate gyrus postnatally, the in vivo actions of IGF1 on growth and development of the hippocampal dentate gyrus up to 130 post-natal days have been investigated. Transgenic IGF1 overexpression was shown to increase granule cell layer and molecular cell layer by 27%–69%, total number of neurons by 29%–61%, and total number of synapses in the molecular layer by 42%–105% compared to control littermates (O'Kusky et al., 2000). Increased neuronal numbers in the dentate gyrus in the post-natal period is thought to be due not only to increased neurogenesis but also reduced neuronal death.

Numerous studies have reported that IGF1 promotes cell survival in the post-natal brain. Transgenic mice overexpressing IGF1 have significantly fewer apoptopic cerebellar neurons at post-natal day 7 compared to wild-type controls, with increased expression of anti-apoptopic proteins and reduced expression of pro-apoptopic proteins (Chrysis et al., 2001). Similarly, in the dentate gyrus of post-natal brains, IGF1 null mice had higher rates of granule cell proliferation but lower numbers of mature granule cells, suggesting the IGF1 promotes survival of granule cells in the post-natal period (Cheng et al., 2001). For a focused review on the function of IGF1 in brain development and plasticity see Dyer et al. (2016). The role of IGF1 in improving neuronal survival and reorganization following injury in both the developing and aging brain are outlined further on in this article.

#### IGF1 and the Developing Brain: Oligodendrocyte Development and Myelination

In addition to the effect of IGF1 on neuronal proliferation and survival in the developing brain, the effect on glial cells, especially oligodendrocytes, has been investigated through in vitro and in vivo studies.

In vitro studies show that administration of IGF1 to oligodendrocyte cells in culture promotes oligodendrocyte proliferation, differentiation, myelin production and survival (McMorris and Dubois-Dalcq, 1988; Mozell and McMorris, 1991; Barres et al., 1992; Ye and D'Ercole, 1999).

In vivo studies comparing transgenic mice with increased IGF-1 expression to those with IGFBP-1 expression (an inhibitor of IGF1) have shown that the mice with increased IGF-1 expression demonstrate higher numbers of oligodendrocytes, increased percentage of myelinated axons and increased thickness of myelin sheath compared to the transgenic mice with increased IGFBP-1 expression (Ye et al., 1995). In another study, prenatal overexpression of IGF1 in transgenic mice produced a mouse brain 55% larger, owing to an overall increase in total cell number, and a total myelin brain content 130% higher than that found in their non-transgenic littermates, which was not associated with a higher percentage increase in oligodendrocyte number, suggesting that increased meylin content was due to increased myelin production per oligodendrocyte (Carson et al., 1993).

Conversely, examination of IGF1 KO mice reported reduced cerebral and spinal cord white matter volume due to fewer myelinated axons and oligodendrocytes at post-natal day 55 (Beck et al., 1995). Further detailed study of IGF-1 KO mice showed consistently reduced overall brain weight 1 week post-natally compared to wild-type littermates, affecting all brain regions, as well as reduced myelination in all brain regions during the first 3 weeks of life, though this normalized and became similar to that of wild-type mice thereafter. Ths coincided with reduced levels of myelin-binding protein (MPB) and proteolipid protein (PLP) in IGF-1 KO mice during the first 3 weeks, which also normalized and became similar to controls by 10 weeks, while percentage oligodendrocyte number was persistently reduced in IGF-1 KO mice at weeks 1, 3 and 10. Additionally, reduced levels of the median subunit of neurofilament (a neuron-specific intermediate filament which acts a s a marker of axon growth) was reduced in IGF-1 KO mice and did not recover as the brain matured (Ye et al., 2002).

These studies suggest the importance of prenatal IGF1 expression in the developing brain for axon growth and CNS myelination in the early post-natal brain.

#### IGF1 AND THE DEVELOPING BRAIN: EVIDENCE FROM STUDIES OF EXTERNAL STIMULI KNOWN TO PROMOTE PAST-NATAL CORTICAL MATURATION

Certain external stimuli have been shown to promote neurodevelopment, and in many cases these studies show that the effect is mediated by IGF1 signaling. In particular, the visual cortex is studied in the context of how brain development is affected by external sensory input. For instance, environmental enrichment (EE), defined as ''a complex of inanimate and social stimulation,'' is known to promote hippocampal neurogenesis and increase dendritic branching and synaptogenesis. It accelerates development of the visual cortex, causing enhanced visual acuity and increased expression of BDNF, which is known to act through the GABAergic system to promote neuroplasticity (Cancedda et al., 2004). A further study showed that this effect of EE on post-natal visual cortical development is inhibited by treatment of enriched pups with IGF1 antagonists and mimicked by treatment of non-EE pups with IGF1 infusion (Ciucci et al., 2007). Similarly, maturation of the visual system in preterm infants was accelerated by massage therapy, as evidenced both through electrophysoiogical studies and through visual acuity testing performed 3 months later. This correlated with higher serum levels of IGF1 and IGFBP3 in massaged infants. This was similarly shown in rat pups undergoing tactile stimulation, who showed faster maturation of electrophysiological tracings and higher levels of IGF1 in the brain compared to controls (Guzzetta et al., 2009).

While the above studies use the EE model to investigate the role of IGF1 in post-natal cortical maturation, other studies have used the model of monocular deprivation (MD) to assess the role of IGF1 in neuroplasticity and reorganization, whereby blocking visual input from one eye leads to structural organization of the visual cortex, allowing for ocular dominance of the intact eye. Expression of regulatory IGFBP5 gene is highly upregulated after MD, and the effects of MD on ocular dominance plasticity are negated by exogenous application of IGF1 (Tropea et al., 2006). Given that the capacity of ocular dominance plasticity in response to short deprivation is a marker of circuit immaturity, this study further supports the theory that IGF1 signaling promotes brain maturation in juvenile animals.

#### IGF1 AND THE ADULT BRAIN: ADULT NEUROGENESIS

As well as the role of IGF1 in early brain development, much research has been done into the role of IGF1 in ongoing neurogenesis and CNS plasticity in the adult brain. In most regions of the brain, neurogenesis ceases after birth. IGF1 mRNA expression correlates both temporally and spatially with these periods of rapid neurogenesis in the perinatal brain. Similarly, there are certain brain regions, namely the dentate gyrus and subventricular zone of the hippocampus, where neurogenesis persists into adulthood, and this correlates with the finding that IGF1 expression in the brain is diffuse during antenatal development, but persists only in the subventricular zone and dentate gyrus of the hippocampus after birth (Anderson et al., 2002). IGF1 expression levels decrease again later in life, again at a time that corresponds with a decrease in hippocampal neurogenesis. Therefore, much research has been done to investigate how manipulation of IGF1 signaling affects adult hippocampal neurogenesis.

It has already been noted in in vitro studies that adult hippocampal neural progenitor cells express IGF1R, and that administration of IGF1 with FGF2 increased progenitor cell proliferation (Åberg et al., 2003). This group further suggested that low-dose IGF1 treatment triggered a small increase in the differentiation of neuronal progenitors into neurons. A further in vivo study used hypophysectomied rats, as this model would have low levels of circulating IGF1. Six days of peripheral subcutaneous IGF1 administration increased proliferation of neuronal progenitors in the hippocampal dentate gyrus, as evidenced by BrdU uptake. After 20 days of subcutaneous IGF1 administration, these new cells expressed neuronal-specific proteins, suggesting stimulation of neurogenesis (Aberg et al., 2000). This suggests that, while local IGF1 mRNA expression levels fall after birth, peripheral IGF1 plays a role in adult hippocampal neurogenesis.

Another study looked at the effect of local IGF1 administration on neuronal cell numbers in the dentate gyrus. Mice aged 5, 18 and 28 months were examined. The dentate gyrus was divided for analysis into the proliferative subgranular zone (PZ), the granular cell layer (GCL) and the hilus, and the numbers of new cells were quantified using BrdU labeling. This study showed that the number of BrdU labeled cells decreases with age, with 80% fewer BrdU labeled cells in the PZ of the 18 month old rats than the 5 month old rats, with a much smaller but significant decrease in BrdU labeled cells in the hilus, though no age-related changes were observed in the GCL. Intracerebroventricular infusion of IGF-I maintained an approximately three-fold increase in the number of BrdU-labeled cells in the PZ, GCL and hilus in old rats examined 31 days after BrdU injection. IGF-I did not, however, selectively induce a neuronal fate since the percentage of BrdU-labeled cells in IGF-I-treated animals that colocalized NeuN was identical to that observed in age-matched controls (Lichtenwalder et al., 2001). Transgenic overexpression of IGF1 was similarly shown to increase the proliferation of neural stem cells in the subgranular and subventricular zones of adult mice brains (Yuan et al., 2015). In this study, however, IGF1 overexpression also led to an increase in the differentiation of neuronal stem cells into neurons, in contrast to the previous study (Yuan et al., 2015).

A recent study has looked in detail at the effects of both global and brain-specific KO of IGF1 in adult hippocampal neurogenesis. Global IGF1 KO mice had both low brain IGF1 expression and low serum IGF1 levels, and showed a 2.4-fold reduction in hippocampal volume compared to controls. Using extensive immunohistochemisty studies, this group demonstrated higher staining for markers of immature differentiation from neural progenitor cells to neurons in IGF−/<sup>−</sup> mice compared to controls, reduced staining for markers of the later stages of differentiation, and disorganized staining for markers of differentiated granule cells in the GCL of the dentate gyrus compared to controls (Nieto-Estévez et al., 2016b). IGF1−/<sup>−</sup> cells in the GCL showed greater proliferative capacity but more immature morphology compared to controls (Nieto-Estévez et al., 2016b). The group then studied the effects of brainspecific IGF1 KO on adult hippocampal neurogenesis. These mice had normal body size and total brain volume compared to controls, but reduced volume of the GCL of the dentate gyrus. As with the global IGF1 KO mice, these mice demonstrated greater immunostaining for immature differentiation markers and more disorganized distribution of mature differentiation markers in the GCL of the dentate gyrus (Nieto-Estévez et al., 2016b). In all, this study of two IGF1−/<sup>−</sup> mouse models demonstrates that a lack of IGF1 in the brain is associated with accumulation of neuronal progenitor cells, impaired transition from neural progenitor to mature granule cell neurons, a reduction in mature morphology of granule cells and disorganization of the GCL (Nieto-Estévez et al., 2016b), all of which supports the idea that IGF1 signaling is key in promoting organized adult hippocampal neurogenesis. Thus, IGF1 not only promotes adult neurogenesis through increased stem cell proliferation, but also through organized cell migration. This is similarly demonstrated in the study of IGF1 KO mice who showed reduced neuroblast migration from the subventricular zone to the olfactory bulb and poor organization of immature neurons in the olfactory bulb compared to normal mice (Hurtado-Chong et al., 2009).

Multiple processes are thought to stimulate adult neurogenesis, the best studied of which is exercise, and recent studies show that this effect is mediated through IGF1 signaling. This was shown through administration of an antibody that blocked systemic IGF1 uptake into the brain parenchyma, which reversed the exercise-induced effects on hippocampal neurogenesis (Trejo et al., 2001). Blocking IGF1R reverses exercise-induced increases in BDNF, suggesting that the downstream effects of IGF1 signaling are at least in part medicated though upregulation of brain-derived neurotrophic factor (Ding et al., 2006).

Exercise promotes functional recovery of spatial memory acquisition in rats who have undergone hippocampal injury, and attenuated the loss of motor coordination in rats following brainstem injury or Purkinje cell degeneration (Carro et al., 2001). The neuroprotective effects of exercise demonstrated in this study were reduced in rats who received subcutaneous IGF1 antibody treatment (Carro et al., 2001). Memory deficits are frequently reported in depression, and imaging studies have shown decreased hippocampal volume in patients with depression. Antidepressant therapy, including SSRIs and ECT, is known to cause upregulation of BDNF, and this has recently been shown to be IGF1-dependent (Chen and Russo-Neustadt, 2007). For a focused review on IGF1 and neurogenesis please see Nieto-Estévez et al. (2016a).

#### IGF1 AND THE ADULT BRAIN: PROLONGED SURVIVAL, REDUCED CELL DEATH, RESISTANCE TO INJURY, REPARATION AND NEUROPLASTICITY IN RESPONSE TO ENVIRONMENTAL CUES

In addition to its role in neurogenesis in the adult brain, IGF1 has been studied for its effect on CNS reparation and plasticity using injury and sensory deprivation models. These studies may indicate that falling levels of IGF1 in the aging brain may indirectly lead to aging through reduced reparation, remodeling and resistance to stress.

One model of injury is the cerebellar deafferentiation model, whereby the olivocerebellar pathway is transected. The remaining olivocerebellar fibers reinnervate the hemicerebellum, but only in the early post-natal period (days 7–10). However, injection of IGF1 into the cerebellum of rats aged 11–30 days allowed for reinnervation of this pathway, indicating that this period of neuroplasticity is extended with use of IGF1 (Sherrard and Bower, 2003). IGF1-mediated reinnervation of the olivocerebellar pathway caused full motor recovery in rats rendered ataxic after 3-acetylpyridine-induced cerebellar injury (Fernandez et al., 1998).

More evidence for its role in reparative neuroplasticity is shown through the use of excitotoxicity models. Chronic intracerebral administration of IGF1 following an excitotoxic lesion in the dentate gyrus caused increased dendritic formation in young neurons in the dentate gyrus compared to untreated controls, and recovery of contextual fear memory, which is a dentate gyrus-dependent function (Liquitaya-Montiel et al., 2012). Another model of dentate gyrus injury is injection of trimethyltin, which is shown to cause elevated IGF1 mRNA levels in the hippocampus. Mice deficient in IGF1 had a significant level of CA1 hippocampal cell death, implying a role for IGF1 in CA1 hippocampal cell survival (Wine et al., 2009).

IGF1 is also thought to be protective in hypoxic-ischemic injury. Intracerebroventricular infusion of IGF1 during perinatal asphyxia in near-term foetal sheep was linked to reduced loss of striatal cholinergic and GABAergic neurons compared to controls on post-mortem examination (Guan et al., 2000). A single dose of intracerebroventricular IGF1 2 h hypoxic-ischemic injury reduced somatosensory deficits for up to 20 days after the initial insult (Guan et al., 2001). The effects of malnourishment in the post-natal period are also attenuated by IGF1 administration. Malnourished mice treated with IGF1 comparable brain weights and cell numbers compared to nourished controls, and higher numbers of oligodendrocytes and expression of myelin markers MPB and PLP, suggesting that this protective effect is through increased myelination (Ye et al., 2000).

The effects of IGF1 signaling following traumatic brain injury (TBI) have also been studied. Overexpression of IGF1 was shown to increase the density of hippocampal immature neurons following TBI. This was shown to be the result of post-traumatic proliferation and differentiation of neural stem cells into immature neurons, rather than through protection against initial insult. IGF1 overexpression also enhanced dendritic arborization of immature neurons compared to normal controls (Carlson et al., 2014). Thus, through enhanced neurogenesis and maturation, IGF1 overexpression accelerated recovery.

The role of IGF1 in activity-dependent neuroplasticity is also demonstrated through the use of sensory deprivation models. MD is a model of reduced activity-dependent activity, whereby one eye is deprived of visual stimuli, which leads to neuronal network reorganization with subsequent ocular dominance of the normal eye. IGFBP-5 is highly upregulated after MD, and administration of IGF1 promotes recovery of normal visual function in these models (Tropea et al., 2006) through upregulation of BDNF (Landi et al., 2009). Based on these findings, studies were performed to investigate whether administration of IGF1 in adult brains restores neuroplasticity. Adult rats underwent MD by eyelid suturing, and were then treated with IGF1. This was shown to trigger ocular dominance plasticity compared to MD rats untreated with IGF1 (Maya-Vetencourt et al., 2012). Furthermore, in rats rendered amblyopic by long-term sensory deprivation, those treated with IGF1 prior to restoration of sensory input showed full recovery of visual acuity compared to rats untreated with IGF1 (Maya-Vetencourt et al., 2012). Downregulation of intracortical inhibitory GABA activity, increased utilization of glucose, and interaction with 5-HT were offered as potential mechanisms through which IGF1 mediates this restoration of plasticity. An alternative sensory deprivation model is that of hindlimb unloading (HU). In adult rats, 14 days of HU decreases IGF1 levels in the somatosensory cortex (Mysoet et al., 2014) and shrinks the somatotopic representation of the hindpaw. Administration of IGF1 prevents this change in somatotopic representation (Papadakis et al., 1996; Mysoet et al., 2015). These interventional studies have demonstrated that IGF1 administration alters neuronal plasticity in response to sensory deprivation in adult animals.

The above studies outline the role of IGF1 in adult neurogenesis, reparation and reorganization in response to stressors. It is therefore possible that loss of these IGF1-driven mechanisms with age leads to age-related cognitive changes. However, numerous studies outlined later in this article report the contrary theory that it is a reduction IGF1 signaling that is neuroprotective following CNS insult.

#### IGF1 AND THE AGING BRAIN: EVIDENCE FROM MOLECULAR BIOLOGY

The GH-IGF1 signaling pathway is the best characterized hormonal pathway in the process of aging. Pulsatile pituitary secretion of GH in response to stimuli such as induced hypoglycemia or arginine administration declines with age (Laron et al., 1969). This is associated with concomitant declines of circulating IGF1 levels (Johanson and Blizzard, 1981). Numerous age-related changes in the brain have been identified that suggest changes in IGF1 signaling. The density of GH receptors decreases with age, while reports are conflicting regarding age-related changes in IGF1 receptor density, with one group reporting increased density of IGF1R expression in the CA3 region of the hippocampus (Chung et al., 2002) and others reporting reduced decreased hippocampal and cortical IGF1R density in aging rats (D'Costa et al., 1993; Sonntag et al., 1999). Expression of IGF1 mRNA was reported to be reduced in the cerebellum of aging rats (Pañeda et al., 2003). These findings would suggest that IGF1 signaling is reduced in the aging brain. How this is linked to functional changes in the aging brain is unclear.

#### IGF1 AND THE AGING BRAIN: EVIDENCE FROM COGNITIVE TESTING

Therefore, studies have been done to examine whether reduced IGF1 signaling is linked to cognitive dysfunction. Studies in humans found a significant correlation between better perceptual motor performance, information processing speed and fluid intelligence and higher circulating IGF1 levels (Aleman et al., 1999, 2001). Others have found a correlation between higher IGF1 levels and higher MMSE scores (Paolisso et al., 1997; Rollero et al., 1998). The MMSE is a well-validated test in terms of repeat-test reliability and tracking of cognitive function over time. In a 2 year prospective study, higher levels of circulating IGF1 levels were associated with reduced cognitive decline over the study period. However, these findings are inconsistent, with many studies also reporting no correlation between IGF1 and attention, fluid intelligence, memory or cognitive decline (Papadakis et al., 1995; Aleman et al., 1999, 2001).

#### IGF1 AND THE AGING BRAIN: EVIDENCE FROM INTERVENTIONAL STUDIES

In any case, correlation does not imply causation. Therefore interventional studies in human were performed to assess if GH or IGF1 levels improved cognitive function, but the results were conflicting and ultimately inconclusive (Papadakis et al., 1996; Friedlander et al., 2001). In mice, intracerebroventricular infusion of IGF1 attenuated age-related deficits in working and reference memory as assessed by Morris water maze and object recognition tasks (Markowska et al., 1998).

Further work has been done in mice studies to assess the downstream molecular effects of IGF1 administration in an aging brain. Intracerebroventricular infusion of IGF1 has been shown to increase microvascular density in aged animals (Sonntag et al., 2000b). Furthermore, it increases hippocampal NMDAR2A/B subunit expression (Sonntag et al., 2000a), which is relevant because NMDAR2B subunit ablation has been shown to impair spatial learning (Clayton et al., 2002). Local IGF1 increases local glucose utilization in the anterior cingulate cortex of aged rats (Lynch et al., 2001). Finally administration of IGF1 attenuates the age-related decline in neurogenesis in aged rats (Lichtenwalder et al., 2001).

### IGF1 AND DISEASE: EVIDENCE FROM DISORDERS OF NEURODEVELOPMENT AND NEURODEGENERATION

# IGF1 and Disorders of Impaired Neurodevelopment

In order to better understand the physiological roles of IGF1 in normal neurodevelopment and aging, one can look at IGF1 activity in the context of known disorders of neurodevelopment and neurodegeneration. For instance, Rett Syndrome is an X-linked neurological disorder characterized by seemingly normal post-natal development initially, followed by a sudden deterioration in function, with loss of acquired functional and motor skills at 12–18 months of age. Rather than being a neurodegenerative process, the underlying pathology is thought to be a stagnation in neuronal maturation. It is caused by a mutation in the MECP2 gene, which codes a transcriptional modulator. It is abundant in neuronal tissue and its expression correlates with that of synaptic maturation. A downstream factor of MECP2 is BDNF, which activates the same PI3K and MAPK pathways that are activated by IGF1 signaling. Therefore, subcutaneous injections if IGF1 have been administered to both mouse models and human subjects to assess if this intervention Wrigley et al. IGF1 in Development and Aging

can reverse the Rett Syndrome phenotype. IGF1 has been shown to increase brain weight, dendritic spine density and levels of PSD-95, a post-synaptic scaffold protein that promotes synaptic maturation, in MECP2 null mice, as well as partially reverse the reduction in amplitude of excitatory post-synaptic current in MECP2 mice (Tropea et al., 2009). Interestingly, persistence of ocular dominance plasticity following MD, a marker of neuronal immaturity, is a feature of MECP2 mouse models of Rett syndrome. It is prevented by pre-treatment with IGF1, giving further evidence for the role of IGF1 in neuronal circuit maturation and reduction in neuroplasticity (Tropea et al., 2009; Castro et al., 2014). These studies indicate the role of IGF1 signaling in neuronal circuit maturation. Similarly, work in SHANK3 deficient models of autism have showed that treatment with IGF1 promotes maturation of excitatory synapses (Shcheglovitov et al., 2013), and reverses deficits in LTP, AMPA signaling and motor function (Bozdagi et al., 2013). Therefore, through the study of IGF1 in the context of disorders caused by poor brain development, it appears that IGF1 promotes neuronal development and brain maturation. For a review focused on IGF1 function in neurodevelopmental disorders, see Vahdatpour et al. (2016).

#### IGF1 and Disorders Associated with Brain Aging

Findings through the study of IGF1 in age-related neurodegenerative disorders, however, have been contradictory, with some studies reporting that reduced IGF1 signaling is neuroprotective, while others claim that reduced IGF1 signaling with age contributes to brain aging. For instance, Alzheimer's disease (AD) is a neurodegenerative disease associated with aging, and one group have shown that a reduction in IGF1 signaling rescues mice from AD-like pathology. An AD mouse model with reduced IGF1 signaling was created, and was reported as having reduced neuronal loss and behavioral deficits compared to the control AD mouse model with normal levels of IGF1 signaling (Cohen et al., 2009). This was attributed to tighter aggregation of Aβ plaques leading to reduced proteotoxicity. These mice also show greater resistance to oxidative stress than mice with intact IGF1 signaling (Holzenberger et al., 2003), suggesting that these findings may be due to an enhanced capacity to protect against the inflammatory effects of Aβ plaques. Similarly, another group demonstrated protection of the aging brain from amyloid pathology by knocking out neuronal IGF1R activity in the brains of adult rats (Gontier et al., 2015). In this study, IGF1R KO in adult neurons led to reduced Aβ pathology and neuroinflammation, along with preservation of spatial memory. Another study reported reduction in Aβ plaques and improved learning and memory following IRS2 KO compared to controls in APP transgenic mice (Killick et al., 2009).

This is in contrast to the prevailing theory that AD is a disorder of insulin and IGF1 resistance. AD has recently been termed ''Type 3 Diabetes,'' given the observation that the spectrum of Mild Cognitive Impairment—AD is associated with global reductions in glucose uptake and utilization occurring early in the course of the disease (Steen et al., 2005; Mosconi et al., 2013). In particular, hippocampal hypometabolism has been observed to correlate with faster progression to dementia (Mosconi et al., 2013). AD could therefore represent a form of CNS insulin resistance. Expression patterns of IR, IGF1 receptor (IGF1R), the intracellular substrate proteins IRS1 and IRS2, and the regulatory IGFBP-2, have been studied in brains affected by AD. One study reported increased IGF1R and decreased IGFBP-2 expression in AD brains, with higher IGF1R expression levels concentrated around amyloid plaques and in neurons with neurofibrillary tangles. These AD neurons showed decreased intracellular levels of IRS1 and IRS2, in association with greater levels of the phosphorylated inactivated forms of these proteins. These findings would suggest that AD neurons show resistance to IGF1 signaling (Moloney et al., 2010). Similarly, another study showed that cerebral neurons in AD brains demonstrate reduced responses to insulin and IGF1 signaling, mainly through phosphorylation and subsequent inactivation of IRS1 (Talbot et al., 2012). Another group that found that mutant rats with lower circulating levels of IGF1 have higher levels of Aβ plaques in the brain, and that levels of Aβ can be reduced in aging rats to levels similar to that in young rats by increasing serum levels of IGF1 (Carro et al., 2002). This is thought to be due to increased clearance of Aβ by albumin and transthyretin carrier proteins due to increased choroid plexus permeability to these proteins (Carro et al., 2002). Furthermore, IGF1R blockade in the choroid plexus worsens AD like pathology, causing amyloidosis, tau hyperphosphorylation and cognitive disturbance (Carro et al., 2006). These studies would therefore suggest that decreased insulin-IGF1 signaling in the brain at least correlates with the development of AD.

The effect of IGF1 signaling in the pathogenesis of Huntington's disease (HD) has also been investigated. This is a triple repeat disorder caused by mutation of the HTT protein, whereby elongation of the CAG triple repeat leads to a resultant HTT protein with a prolonged polyglutamine tract. This prolongated protein is cut into toxic fragments that aggregate, causing neurotoxicity and degeneration. Reduced IGF1 signaling is linked to pathological and symptomalogical improvements in mouse models of HD. The R6/2 mouse model of HD showed more rapid neurodegeneration following increased expression of IRS2, while decreasing IRS2 expression is associated with a longer lifespan in this model (Sadagurski et al., 2011). This is attributed to fewer polyQ-HTT aggregates in the brain. Conversely, another study showed that treatment of neurons transfected with the HTT mutation with IGF1 reduced polyQ-htt aggregation through Akt-mediated huntingtin phosphorylation (Humbert et al., 2002).

#### IGF1 AND METABOLISM: TARGETING IGF1

IGF1 signaling has many effects on metabolism, at the tissue and cellular levels. This signaling is not limited to glucose and lipid homeostasis but also influences protein turnover (Sharples et al., 2015).

Manipulating IGF1 depends on our understanding of the metabolic trade-offs, especially in the brain, adipose tissue and skeletal muscle, that are associated with it. IGF and its related molecules are important in protein metabolism and the regulation of skeletal muscle mass. KO of IGFI, IGFII or the IGFI receptor causes neonatal lethality and decreases in skeletal muscle mass in rodents (Sharples et al., 2015). The role of IGF1 and its related molecules in the maintenance of skeletal muscle mass in humans is especially important for elderly individuals whose IGF1 levels decrease with age and are at risk of frailty (Maggio et al., 2013) and sarcopenia (Sharples et al., 2015). Manipulating metabolism by using dietary restriction offers a similar mechanism to reduced IGF1 signaling in that it results in the inhibition of mTOR and it has also been linked to reduced IGF1 levels. It has the potential to be used in combination with an increase in protein or amino acid intake to counteract the losses in skeletal muscle that accompany a reduction in IGF1 (Sharples et al., 2015).

Choosing which pathway to target poses an obstacle to the modification of IGF1 signaling. mTOR, which functions through complexes mTORC1 and mTORC2, might be a promising target. mTORC1 is responsive to nutrients, energy and growth factors and its inhibition has been shown to decrease aging rate and age-related weight gain in mice (Hu and Liu, 2014). KO of RAPTOR, an mTORC1-specific accessory protein, in adipose tissue preserves the lifespan extension seen in other models reducing Insulin/IGF1 signaling while also improving metabolic markers such as glucose tolerance and insulin sensitivity (Hu and Liu, 2014). This, taken with the detrimental effects of KO on tissues such as skeletal muscle and the importance of IGF1 in metabolism in the developing brain, suggests that tissue-specific reductions in signaling may be one way of overcoming this obstacle. In addition to approaches which take into account specific pathways and tissues, it may also be beneficial to investigate the differences in the effects of IGF1 reductions at different time points. Human population studies point to a positive impact of IGF1 reductions at a young age and elevations at an old age (Sharples et al., 2015).

It has been questioned whether the extended longevity afforded by reduced IGF-1 signaling and its effects on metabolism are mediated in the same way. In fact, it has recently been suggested that extended lifespan in the Ames and Snell dwarf mice is not due to IGF1 levels but rather the decrease in GH (Brown-Borg and Bartke, 2012). Furthermore, it is thought that distinct sets of neurons mediate the functions of insulin/IGF1 in the brain. KO of IRS2 in the entire brain in mice does result in lifespan extension but at 22 months these mice are also overweight, hyperinsulinemic and glucose-intolerant, demonstrating the centrality of this pathway in metabolic homeostasis. Nutrient-sensing which is central to the regulation of glucose and lipid homeostasis is carried out by the leptin-sensitive neurons of the arcuate nucleus which contain IRS2 and the IR. While it is in these neurons that insulin/IGF1 exert profound effects on metabolism, IRS2 is not required for leptin action. So then, IRS2 in these neurons mediates the functions of insulin/IGF1 ad not those of leptin, once again highlighting the importance of insulin-like signaling in metabolism. Interestingly, the decrease of IRS2 on leptin receptor-expressing neurons did not result in the increase in lifespan seen with overall IRS2 decrease. These results indicate that the neurons which act as the metabolic mediators of insulin/IGF1 signaling in the CNS are distinct from those which underlie mammalian lifespan extension due to a reduction in insulin/IGF1 signaling (Sadagurski and White, 2013; White, 2014). This provides a powerful starting point for the potential development of strategies to manipulate IGF1 function without causing metabolic dysfunction.

There is a paradox presented by the Insulin/IGF1 signaling pathway in metabolism which warrants further research. In the CNS, this paradox could well be due to the existence of neuronal subsets which mediate the effects of IGF1/Insulin and in the periphery could be owed to the differing actions of IGF1 on various tissues. The importance of IGF1 and its related molecules on metabolism is undeniable and if manipulated correctly could prove to be viable therapeutically.

## THE DEBATE CONTINUES

In conclusion, the above studies largely support a role for IGF1 signaling in brain development, and adult neuroplasticity and neurogenesis. However, while numerous studies report that IGF1 signaling serves to delay brain aging, and that the known fall in IGF1 signaling with age acts as a causative factor in age-related brain changes, there remain as many studies that stand in contradiction, and suggest that a reduction in IGF1 signaling delays age-related changes and diseases.

IGF1 levels are higher in the developing brain, and this is shown, through the studies outlined above, to promote neuronal development. It activates the PI3K pathway, which promotes survival by directly inactivating pro-apoptopic machinery (van der Heide et al., 2006), and increases glucose uptake by neurons (Bondy and Cheng, 2004). Post-natally, IGF1 promotes neuronal maturation, and has been shown to partially correct the phenotype of certain neurodevelopmental disorders. IGF1 is associated, in the adult brain, with regions of continued neurogenesis. These findings would suggest that IGF1 signaling exerts an overall neuroprotective effect, and that falling IGF1 levels with age contribute to the effects of aging in the brain.

However, there also exists a body of evidence suggesting that reduced IGF1 signaling attenuates the effects of aging, both in the brain and in the whole organism. A reduction in IGF1 signaling increases the life span of C. elegans, as DAF-2 mutants with a lower level of DAF-2 signaling have a lifespan double that of normal controls (Kenyon et al., 1993). With regards to brain function, age-related decline in axonal regeneration in C.elegans was shown to be regulated by DAF-2 signaling. While the rate of axonal regeneration following injury was 65% in day1 C.elegans worms and 28% in day 5 worms, the same experiment in DAF2−/<sup>−</sup> worms had no reduction in axonal degeneration. Reduced DAF-2 signaling allowed for increased DAF-16 activity, stimulating neuronal regeneration in response to injury (Byrne et al., 2014). However, in the C. elegans, the insulin and IGF1 pathways are not diverged, and therefore the effects of IGF1 on aging cannot be studied in isolation. Studies of IGF1 signaling and lifespan in mammals have also been done, and an IGF1R+/<sup>−</sup> transgenic mouse model with a reduced level of IGF1 signaling activity has been created that has a longer lifespan than control mice, and demonstrated greater resistance to oxidative stress (Holzenberger et al., 2003). These findings would therefore suggest that the fall in IGF1 signaling with age is not in fact the cause of aging, but is perhaps a protective mechanism that occurs as to attenuate the effects of aging.

These contradictions may arise partly because of the differential activity of IGF1 signaling in the brain compared to the whole body in different experimental models. One particular study characterizes the distinction in Ames mice, which have a primary deficiency in GH, leading to low levels of circulating IGF1. These mice demonstrate a longer lifespan, which has been attributed to absence of GH-IGF1 signaling, thereby affording evidence that IGF1 signaling contributes to aging. However, Sun et al. (2005) demonstrated that while these Ames mice show lower levels of GH and IGF1 peripherally, they have elevated levels of IGF1 in the hippocampus compared to normal mice. Furthermore, this correlated with higher levels of neurogenesis in the dentate gyrus, compared to the controls (Sun et al., 2005). This elevated level of neurogenesis in the Ames mice may underlie the observation that these mice showed less age-related cognitive deficits. Therefore, while globally reduced IGF1 signaling appears to extend the lifespan of these organisms, one should not assume that brain-specific IGF1 signaling is also reduced, or that the effect of IGF1 activity in the brain compared to the rest of the organism on the process of aging is necessarily the same.

Contradictions again arise when studying the neuroprotective effect of IGF1 signaling in hypoxic-ischemic injury. While above studies describe reduced neuronal loss following intracerebroventricular infusion of IGF1 (see above), other studies report that following Cre-LoxP-mediated inactivation of IGF1R in forebrain neurons, there was reduced neuronal damage, inflammation and edema in response to hypoxicischemic insult (De Maghalaes Filho et al., 2016).

These contradictions between different studies may be due to the different approaches taken: decreasing IGF1 or the receptor, or the IRS receptor, or the targeting of the modulators IGFBPs.

#### REFERENCES


In fact, the systems are highly regulated and the changes in each factor may have a different effect. For example, a moderate decrease in IGF1R increases life span, contrary to what happens to decreases in IGF1. Another factor to take into account is the integration insulin-IGF1 in the brain. In neurons, IRSBPs proteins and IGF1R form a complex which also binds insulin and may activate different intracellular signals. This interaction should be taken into account in therapies which use IGF1 to improve brain function, because the variability of the results may be due to different levels of insulin in the body. This theory is in line with the homeostatic function of IGF1 as a connector between body and brain. In addition, as demonstrated by Sun et al. (2005), inconsistencies also arise when the differential activity and effect of IGF1 in different organ systems is not taken into account, for it may be the case that peripheral and brain IGF1 signaling have opposing effects, with the former leading to overall acceleration of aging in the body while the latter continues to promote renewal and reparation. Furthermore, it is possible that while IGF1 signaling continues to promote neuronal development and plasticity throughout life, through its effects on cellular apoptotic machinery, glucose utilization and other neurotrophic factors, such anabolic process may simultaneously contribute to aging through accumulation of reactive oxygen species and resultant prolonged oxidative stress over time.

In all, there remains much to be done in elucidating the role of IGF1 signaling in the brain as it develops, matures and ages. IGF1 appears to act in concert with BDNF and other neurotrophic factors to promote neurogenesis and remodeling in the brain. However, its overall effects on energy metabolism and cellular oxidation may contribute to aging in all organs. What is obvious is that it is not simply a matter of high IGF1 signaling early in life promoting development and falling levels thereafter underlying the process of aging. An evolutionarily ancient pathway, IGF1 signaling has likely taken on numerous differential roles in different body tissues in health and disease (Forbes, 2016), and its complex effects on cellular maturation, tissue development and energy metabolism may contribute to organismal development and aging simultaneously.

#### AUTHOR CONTRIBUTIONS

SW wrote a consistent part of the manuscript. DA wrote part of the manuscript and contributed to the figures. DT designed the structure of the review, wrote part of the manuscript and contributed to the figures.


ataxia. Proc. Natl. Acad. Sci. U S A 95, 1253–1258. doi: 10.1073/pnas.95. 3.1253


signalling. Neurobiol. Aging 31, 224–243. doi: 10.1016/j.neurobiolaging.2008. 04.002


in neurons from 22q13 deletion syndrome patients. Nature 503, 267–271. doi: 10.1038/nature12618


**Conflict of Interest Statement**: DT has a patent for the potential use of IGF1 in neurodevelopmental disorders.

The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Wrigley, Arafa and Tropea. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# NPAS3 Regulates Transcription and Expression of VGF: Implications for Neurogenesis and Psychiatric Disorders

Dongxue Yang1† , Wenbo Zhang1† , Arshad Padhiar <sup>1</sup> , Yao Yue<sup>1</sup> , Yonghui Shi <sup>1</sup> , Tiezheng Zheng<sup>1</sup> , Kaspar Davis <sup>2</sup> , Yu Zhang<sup>2</sup> , Min Huang<sup>1</sup> , Yuyuan Li <sup>1</sup> and Li Sha<sup>1</sup> \*

<sup>1</sup> College of Basic Medicine, Dalian Medical University, Dalian, China, <sup>2</sup> Department of Physical Education, Dalian University of Technology, Dalian, China

Neuronal PAS domain protein 3 (NPAS3) and VGF (VGF Nerve Growth Factor (NGF) Inducible) are important for neurogenesis and psychiatric disorders. Previously, we have demonstrated that NPAS3 regulates VGF at the transcriptional level. In this study, VGF (non-acronymic) was found regulated by NPAS3 in neuronal stem cells. However, the underlying mechanism of this regulation remains unclear. The aim of this study was to explore the correlation of NPAS3 and VGF, and their roles in neural cell proliferation, in the context of psychiatric illnesses. First, we focused on the structure of NPAS3, to identify the functional domain of NPAS3. Truncated NPAS3 lacking transactivation domain was also found to activate VGF, which suggested that not only transactivation domain but other structural motifs were also involved in the regulation. Second, Mutated enhancer box (E-box) of VGF promoter showed a significant response to this basic helix-loop-helix (bHLH) transcription factor, which suggested an indirect regulatory mechanism for controlling VGF expression by NPAS3. κB site within VGF promoter was identified for VGF activation induced by NPAS3, apart from direct binding to E-box. Furthermore, ectopically expressed NPAS3 in PC12 cells produced parallel responses for nuclear factor kappa-light-chain-enhancer of activated B cells [NF-κB (P65)] expression, which specifies that NPAS3 regulates VGF through the NF-κB signaling pathway. Over-expression of NPAS3 also enhances the cell proliferation, which can be blocked by knockdown of VGF. Finally, NPAS3 was found to influence proliferation of neural cells through VGF. Therefore, downstream signaling pathways that are responsible for NPAS3-VGF induced proliferation via glutamate receptors were explored. Combining this work and published literature, a potential network composed by NPAS3, NF-κB, Brain-Derived Neurotrophic Factor (BDNF), NGF and VGF, was proposed. This network collectively detailed how NPAS3 connects with VGF and intersected neural cell proliferation, synaptic activity and psychiatric disorders.

Keywords: NPAS3, VGF, cell proliferation, transcription regulation, neurogenesis, psychiatric disorders, glutamatergic signaling pathways

#### Edited by:

Andrew Harkin, Trinity College Dublin, Ireland

#### Reviewed by:

Gianluca Serafini, University of Genoa, Italy Sung Jun Jung, Hanyang University, South Korea

\*Correspondence: Li Sha shali-nk@hotmail.com † These authors have contributed equally to this work.

Received: 04 August 2016 Accepted: 12 October 2016 Published: 08 November 2016

#### Citation:

Yang D, Zhang W, Padhiar A, Yue Y, Shi Y, Zheng T, Davis K, Zhang Y, Huang M, Li Y and Sha L (2016) NPAS3 Regulates Transcription and Expression of VGF: Implications for Neurogenesis and Psychiatric Disorders. Front. Mol. Neurosci. 9:109. doi: 10.3389/fnmol.2016.00109

# INTRODUCTION

Neuronal PAS domain protein 3 (NPAS3) encodes a member of the basic helix-loop-helix (bHLH) PAS domain transcription factor family that typically integrates environmental signals and binds with heterodimeric partner to generate a transcriptional response (Brunskill et al., 1999; Gilles-Gonzalez and Gonzalez, 2004). Disruption of the NPAS3 gene carried by a Scottish mother and daughter diagnosed with schizophrenia and mild learning disability provided the first indication of the role of this gene in psychiatric illness (Kamnasaran et al., 2003). Various gene specific and genome-wide case-control association studies have linked single nucleotide polymorphisms in the NPAS3 gene with increased risk of schizophrenia, major depression and bipolar disorder (Pickard et al., 2009; Huang et al., 2010; Yu et al., 2014). NPAS3 protein is also involved in other processes in the brain such as neurogenesis, circadian rhythm and cell proliferation. Npas3 knockout mice display a range of behavioral phenotypes consistent with it being a representative model of human psychiatric disorders (Sha et al., 2012). NPAS3 knockout mice also display an additional deficit in adult hippocampal neurogenesis and aberrations in synaptic transmissions (Pieper et al., 2005, 2010), but the underlying mechanism of dysfunctions of neurogenesis and synaptic activity in relation to the pathology of psychiatric disorder is currently unknown.

Similar to NPAS3, VGF plays a major role in depression (Cattaneo et al., 2010), bipolar disorder (Thakker-Varia et al., 2010) and schizophrenia (Busse et al., 2012; Ramos et al., 2014). Studies in mice have produced compelling evidence that VGF mediates the connection between physical activities, amplified adult neurogenesis and attenuation of depression-like phenotypes. VGF and other neuropeptides such as secretogranin II and neuropeptide Y (NPY) are regulated in the hippocampus when voluntary exercise was employed as a mood stimulator. It was also reported that VGF peptide can produce a robust antidepressant response in a dose dependent manner in follistatin (FST) and tail suspension trials (Hunsberger et al., 2007). VGF was reported as an important factor in the central nervous system in a peptidomic analysis on normal cerebrospinal fluid (CSF; Yuan and Desiderio, 2005; Spellman et al., 2015). In addition, TLQP-62, one of the mature peptides of VGF, was found to specifically enhance the generation of early progenitor cells in mice (Thakker-Varia et al., 2014). On the whole, these studies suggest a positive role of VGF in neurogenesis and connections of VGF in psychiatric disorders.

VGF enhances proliferation of neurogenesis in the adult hippocampus, and this process requires synaptic activity (Thakker-Varia et al., 2014). The specific reporters they identified for VGF-induced neurogenesis, includes N-methyl D-aspartate receptor (NMDA receptor) and metabotropic glutamate receptor 5 (mGluR5). mGluR5 mutant mice have reduced proliferation and differentiation of neuronal progenitors (Xiao et al., 2013). Activation of mGluR5 induced the phosphorylation of Protein kinase D (PKD) in hippocampus neurons (Krueger et al., 2010) and in neural progenitor cells (Thakker-Varia et al., 2014). PKD modulates DNA synthesis and cell proliferation in various cell lines (Wong and Jin, 2005) and has anti-apoptic properties in tumor cells (Trauzold et al., 2003).

The role of NMDA in neurogenesis has also been documented. Elevated levels of NMDA or its subunit enhanced neurogenesis of animal models (Kalev-Zylinska et al., 2009; Marx et al., 2011; Sharma et al., 2012; Ren et al., 2013). Furthermore, reduced expression of NMDA produced impaired neurogenesis of the dentate gyrus in mice (Sha et al., 2013). Finally, deletion of NMDA subunit (NR2B) impairs a neurogenesis-dependent form of LTP (Kheirbek et al., 2012). Ca2+/calmodulin-dependent protein kinase II (CaMKII) was found to be associated with NMDA receptors and obtains its activation form by phosphorylation (Bayer et al., 2001). Mice with mutant CaMKII have immature granule cells in the dentate gyrus (Yamasaki et al., 2008). Moreover, administration Dehydroepiandrosterone (DHEA), a steroid receptor agonist, to olfactory bulbectomized mice increases synaptic efficacy and neurogenesis in the hippocampal dentate gyrus by activating CaMKII.

Our previous work investigated the function of NPAS3 at the transcriptional level, and VGF was found to be the most up-regulated gene by NPAS3 (Sha et al., 2012). Although VGF is highly relevant to NPAS3, the way these two factors correlate has not been well explored.

In this study, we found that the transcription and expression of VGF is regulated by NPAS3 in neural stem cells (NSCs). However, the underlying mechanism remains unclear. Therefore, the correlation of NPAS3 and VGF and the roles of these two proteins during neural cell proliferation were the key questions addressed in this study. For that, we focused on three aspects: first, we focused on the structure of NPAS3, to identify the functional domain of NPAS3 by comparing the reputational activities of different NPAS3 domains on VGF promoter. Regulation of VGF by dNPAS3, a truncated form NPAS3 lacking transactivation domain, was compared with wild type NPAS3. Second, several mutated VGF promoters were compared to identify the specific regulatory elements involved in transactivation of NPAS3. Finally, we suspect that NPAS3 might influence proliferation of neural cells through VGF. Therefore, proliferation of neural cells was examined. In addition, in order to understand the relationship of NPAS3- VGF-prolifaration, one must observe the key changes that occur during proliferation of neural cells, which are the changes in glutamate receptors. Downstream signaling pathways that are responsible for NPAS3-VGF induced proliferation via glutamate receptors were also explored.

# MATERIALS AND METHODS

#### Plasmids Used in this Study

The NPAS3 open reading frame (acc. NM\_001164749) and the truncated form, dNPAS3, cloned into pcDNA 3.1 expression plasmid were gifted by Dr. Ben Pickard (University of Strathclyde, UK). dNPAS3 was generated by deleting the second PAS domain and the putative transactivation domain.


TABLE 1 | Four suitable small hairpin RNA (shRNA) target sequences for Neuronal PAS domain protein 3 (NPAS3).

Four suitable small hairpin RNA (shRNA) target sequences were selected in the human NPAS3 sequence (**Table 1**). All shRNA plasmids were constructed and sequenced by GenePharma Co., Ltd. (Shanghai, China). Each plasmid was constructed by inserting a shRNA of human NPAS3 into a pGPU-GFP-neo vector. Negative controls (NCs) were also provided. The most efficient shRNA plasmid was selected by quantitative real-time PCR (qPCR).

#### Cell Culture and Transient Transfection of SH-SY5Y, 293T Cells and PC-12 Cells

The HEK-293T embryionic kidney cell line was cultured in DMEM medium (Gibco) supplemented with 10% fetal bovine serum (FBS). The SH-SY5Y human neuroblastoma cells (SH-SY5Y cells) were cultured in RPMI medium 1640 (Gibco) with 10% FBS. The rat PC-12 pheochromocytoma cells were cultured in RPMI medium 1640 (Gibco) with 5%FBS and 10% horse serum. Cells were transfected with corresponding plasmids using Lipofectamine 3000 (Invitrogen) according to the manufacturer's instructions.

#### Dual Luciferase Assay

Human VGF promoters of different sizes (2029 bp, 1237 bp and 990 bp) were amplified by PCR and cloned into digested pGL3 reporter vectors. Site-directed mutagenesis was carried out using Fast Mutagenesis System (TransGen Biotech Co., Ltd., Beijing, China). Sequences were confirmed by sequencing analysis. 293T cells were harvested at 24 h after transfection for luciferase assay. The activities of Firefly luciferase (expressed from all pGL3 reporter vectors) and Renilla luciferase (expressed from co-transfected pRL-TK vector) were examined sequentially from each sample by using the Dual luciferase Assay kit (Promega, Madison, WI, USA) with the Plate-reader (Synergy HT, BioTek, Potton, UK). For each sample, the Firefly luciferase activity was normalized by the Renilla luciferase activity. The results were statistically analyzed using SigmaPlot© (Bruxton, Seattle, WA, USA).

#### Western Blotting

Cells were solubilized in RIPA lysis buffer containing protease and phosphatase inhibitors. Protein concentrations were determined using the BCA Protein Quantitation Assay Kit (KeyGEN, China). Proteins were separated on a 10% SDS-polyacrylamide gels and transferred to a polyvinylidene difluoride (PVDF) membrane. Membranes were probed with anti-NPAS3 antibody (1:1000, Abcam, Cambridge, MA, USA), anti-VGF antibody (1:1000, Abcam, Cambridge, MA, USA), anti-NF-κB (p65) antibody (1:1000, Cell Signaling, Boston, MA, USA), anti-NF-κB (p52) antibody (1:1000, Cell Signaling, Boston, MA, USA), anti-β-Actin antibody (1:1000, Cell Signaling, Boston, MA, USA), anti- GAPDH antibody (1:2000, ZSGB-Bio, China), anti-PKD (1:500, Cell Signaling, Boston, MA, USA), anti-phospho-PKD (1:250, Cell Signaling, Boston, MA, USA), anti-CaMKII (1:500, Cell Signaling, Boston, MA, USA), anti-phospho- CaMKII (1:250, Cell Signaling, Boston, MA, USA) overnight, respectively. Membranes were washed, followed by incubation with goat anti-rabbit horseradish peroxidase-conjugated IgG (1:5000, abbkine) at room temperature for 1 h. Proteins were detected using Chemiluminescent HRP Substrate (Advansta) and visualized with the ECL detection system (Bio-Rad, Berkeley, CA, USA). The bands were measured by Gel-Pro Analyzer software (Media Cybernetics, Rockville, MD, USA).

#### Quantitative Real-Time PCR

Total RNA was isolated using E.Z.N.A.rTotal RNA Kit I (OMEGA) and analyzed on the NanoDrop2000c analyzer (Thermo Scientific, Wilmington, DE, USA). Reverse transcriptase reactions were performed using Revert Aid First Strand cDNA Synthesis Kit (Thermo Scientific, Wilmington, DE, USA). The primer sequences were synthesized by Invitrogen and were designed as previously described (Sha et al., 2012). 18s RNA was used as an internal control. qPCR was performed using TransStart Top Green qPCR SuperMix (TransGen Biotech, China) on the Stratagene Mx3000p PCR machine (Agilent Technologies, Santa Clara, CA, USA).

#### Culture of Rat Neural Stem Cell Line

Rat fetal NSCs were purchased from Gibco Co. The cells were plated in growth medium consisting of KnockOutTM D-MEM/F-12 with StemPro<sup>r</sup> Neural Supplement, bFGF, EGF and GlutaMAXTM-I and incubated at 37◦C, 5% CO<sup>2</sup> and 90% humidity. Neurospheres were observed within the first week and the medium was changed every 3–4 days. NSCs were passaged when spheres reached the size of 3.5 mm. NSCs were transiently transfected with corresponding plasmids using Lipofectamine 3000 for 48 h. The transfection efficiency was evaluated by flow cytometry.

#### Immunofluorescence

The wistar rats were perfused with 4% paraformaldehyde in PBS and brains were postfixed in the same fixative for 24 h and cryoprotected in 30% sucrose before vibratome sectioning (40 µm, Leica CM1950). All experiments were performed in accordance with the 1996 National Institutes of Health Guide for the Care and use of Laboratory Animals, and the experimental procedures were approved by the Local Committee of Animal Use and Protection. Free-floating sections were permeabilized with 0.2% TritonX-100 in PBS for 15 min, blocked in PBS containing 5% donkey serum for 1 h at room temperature and then incubated with anti-NPAS3 (1:100, Santa Cruz, CA, USA) and anti-VGF (1:200, Abcam, Cambridge, MA, USA) overnight at 4◦C. the sections were washed in the PBS and incubated with fiuorescent secondary antibodies, donkey anti rabbit conjugated with Alexa Fluor 488 (1:1000; Invitrogen) and donkey anti goat conjugated with Alexa Fluor 594 (1:1000; Invitrogen), for 1 h at room temperature. DNA (nuclei) was stained with Dapi for 15 min, and mounted onto slides and coverslipped with ProLong (Invitrogen). The images were captured using the Olympus system and analyzed with the ImageJ software.

### Cell Proliferation Assay

The SH-SY5Y and PC12 cells were plated in 96-well plates in 1640 medium containing 10% FBS or 10% horse serum and 5% FBS at a density of 1 × 10<sup>5</sup> cells each well and then transfected with NPAS3, shRNA for NPAS3, siRNA for VGF, NPAS3 plus siRNA for VGF, mock siRNA for VGF for 48 h. A 10-µL volume of the Vita-Orange (WSTr-8; Biotool, USA) was then added to each well and the cells were cultured for another 2 h. The plates were then examined in a spectrometer at 450 nm to obtain the absorbance (OD value) of each well.

#### Statistical Analysis

SPSS (IBM, Armonk, NY, USA) software was used for analysis. Prism 5 software (GraphPad, San Diego, CA, USA) was used to create the graphs. All experiments were repeated at least three times. Data were presented as means ± standard deviation and analyzed using two-tailed Student t-test or one way analysis of variance (ANOVA) followed by Turkey or Dunnett T3 post hoc tests for multiple comparisons.

#### RESULTS

#### Correlation Between NPAS3 and VGF in Rat Neural Stem Cells

Our previous microarray analysis suggested that VGF is a highly up-regulated gene by NPAS3 in 293T cells (Sha et al., 2012). Expression of VGF was examined at transcriptional (qPCR assay) and at translational level (Western blotting) by manipulating the expression of NPAS3 in NSCs. The mRNA (**Figure 1A**) of VGF was significantly increased in NPAS3 overexpressing NSCs (P = 0.007) and decreased in NPAS3 knockdown NSCs (P = 0.003). Results of Western blots showed that VGF exhibited a parallel response to NPAS3 in NSCs (**Figures 1B,C**; pcDNA3.1 vs. NPAS3 P = 0.000, non-specific shRNA vs. shRNA P = 0.004). The expression of Npas3 and Vgf in the rat hippocampus was investigated using immunofluorescence microscopy. Npas3 is expressed in the subgranular zone of the dentate gyrus with processes radiating into the granule cell proper (**Figure 1D**). Vgf was co-stained with Npas3 in the rat hippocampus. Vgf was found to colocalize with Npas3, which suggests that Npas3 might exert its function at the same process in the hippocampus.

#### VGF Promoter can be Regulated by NPAS3 in Various Cell Lines

In the light of our findings that VGF exhibited a parallel response to NPAS3 in various cell lines, we hypothesized that NPAS3 regulates VGF at the promoter level. 293T (**Figure 2A**), SH-SY5Y (**Figure 2B**) and U251 (**Figure 2C**) cell lines were used to examine the VGF promoter activity. The reason for choosing 293T cell line is that the behavior of the cell itself is not of interest and 293T is a very good model for analyzing transcription of a specific gene and widely used for neuronal study (Thomas and Smart, 2005). Dual luciferase assay results showed that activation of VGF promoter was significantly increased due to over expression of NPAS3 (293T: pcDNA3.1 vs. NPAS3 P = 0.000; SH-SY5Y: pcDNA3.1 vs. NPAS3 P = 0.001; U251: pcDNA3.1 vs. NPAS3 P = 0.000) and markedly reduced in NPAS3 down-regulated cells (293T: non-specific shRNA vs. shRNA P = 0.000; SH-SY5Y: non-specific shRNA vs. shRNA P = 0.000; U251: non-specific shRNA vs. shRNA P = 0.000). These findings suggested that VGF could be regulated by NPAS3 at the promoter level.

#### Identification of Potential Regulatory Domains Within NPAS3 Protein

We first compared the regulation of VGF by full-length (NPAS3) and truncated NPAS3 (dNPAS3; **Figure 3A**), lacking the second PAS domain and the putative transactivation domain, using qPCR. As shown in **Figure 3B**, VGF gene showed a much more robust up-regulated response in NPAS3 over expressed SH-SY5Y cells (one way ANOVA P = 0.000, post hoc analysis: control vs. dNPAS3 P = 0.000; control vs. NPAS3: P = 0.000; dNPAS3 vs. NPAS3: P = 0.000). Then, Dual luciferase assays were carried out to examine the regulation of VGF by NPAS3 at the promoter level (**Figure 3C**). We observed that the luciferase activity of VGF promoter was significantly increased by NPAS3 or dNPAS3 (one way ANOVA P = 0.000, post hoc analysis: control vs. dNPAS3 P = 0.000; control vs. NPAS3: P = 0.000). However, NPAS3 shows a stronger activation on VGF promoter than dNPAS3 (P = 0.000; **Figure 3C**). Taken together, each motif in NPAS3 might be involved in the regulation of VGF.

#### E-Box Within VGF Promoter is Not the Only Regulatory Sequence for Activation by NPAS3

The VGF promoter region contains several consensus motifs for transcriptional regulators, such as a CCAAT box, various SP-1

and AP-2 consensus binding sites, a cAMP-response element (Possenti et al., 1992), a CREB binding sites (Bozdagi et al., 2008), a putative silencer element and an enhancer box (E-box) near the transcription start site (D'Arcangelo et al., 1996). As we know that bHLH domain transcription factors (such as NPAS3) bind specifically to E-box, it is highly possible that NPAS3 might regulate VGF through binding to its E-box. To test this hypothesis, a VGF promoter with a mutated E-box (TACGTG; **Figure 4B**) was compared with wild type VGF promoter (**Figure 4A**). As the mutated promoter without E-box sequence could not be recognized by bHLH domain, it should not be activated by NPAS3. Luciferase activity of VGF promoter with the site-mutated E-box was found lower than wild type VGF promoter (P = 0.000). This indicates that the E-box is partially

NPAS3 open reading frame. (dNPAS3) refers to truncated form of NPAS3 cloned in pcDNA3.1, while (NPAS3) refers ORF of NPAS3 cloned in pcDNA3.1. (A) Structures of NPAS3 and dNPAS3 protein. (B) Transcription of VGF in the pcDNA3.1/NPAS3/dNPAS3 over expressed SH-SY5Y cells by triplicated qPCR. 18s RNA was used to normalize data from each sample and values were expressed as fold-change in gene expression in comparison to the control samples. The VGF value of control cells is supposed to be 1. (C) Luciferase activity of VGF promoter in the pcDNA3.1/NPAS3/dNPAS3 vector over-expressed SH-SY5Y cells. The values represent the percentage of the relative activities to the pRL-TK vector (mean ± SD). All experiments were repeated three times. Asterisks mean a significant difference at ∗∗∗P ≤ 0.001 (Analysis of variance [ANOVA] followed by Turkey post hoc test).

responsible for mediating the NPAS3 activation of the promoter. However, that fact that NPAS3 could still significantly activate the mutated E-box promoter suggests the additional route of VGF activation by NPAS3.

# NF-κB is Involved in the Transcriptional Regulation of VGF

In order to identify the potential indirect regulation of VGF by NPAS3, a shorter VGF promoter lacking κB site (990 bp;

5 0 -GGGRNW YYCC-3<sup>0</sup> , R = purines, N = any nucleotide, W = adenine or thymine and Y = pyrimidine) was constructed. Luciferase activity of this promoter was compared with wild type promoter (1237 bp) in 293T cells with over expressed or down regulated NPAS3. Interestingly, our results showed that the NPAS3 dependent activation of VGF promoter was decreased in the absence of the κB site compared to wild-type promoter (P = 0.001; **Figure 5A**). This finding suggests that the κB site also contributes to NPAS3 activation of the VGF promoter.

To further reveal the role of NF-κB in this process, expression of VGF (non-acronymic) and molecules involved in NF-κB pathway were examined by western blotting in PC12 cells. Expression of VGF is significantly up-regulated by over-expression NPAS3 (P = 0.000) and slightly down-regulated in knock-down NPAS3 PC12 cells (P = 0.04; **Figures 5B,C**). Caffeic acid phenethyl ester (CAPE) is a well documented NF-κB inhibitor at high concentrations (Natarajan et al., 1996). The CAPE treatment significantly reduced VGF expression (DMSO vs. CAPE P = 0.000) and blocked the effect of NPAS3 on VGF (NPAS3 + DMSO vs. NPAS3 + CAPE P = 0.000). The positive control, NPAS3 + DMSO, significantly enhances expression of VGF relative to samples treated by DMSO or inhibitor. P65 and p52 are members of NF-κB superfamily and have been examined for transcriptional regulation of genes involved in cell proliferation. P65 showed parallel changes with NPAS3 in PC12 cells (**Figure 5D**). NPAS3 has no effect on p52 expression in PC12 cells (**Figure 5E**). The altered level of p65 induced by NPAS3 implies that NF-κB might be involved in the transcriptional regulation of VGF by NPAS3.

#### The Effects of NPAS3 and VGF on Cell Proliferation

VGF was reported to induce proliferation of neural progenitor cells (Thakker-Varia et al., 2014). Based on the correlation of VGF and NPAS3 described in this study, we supposed that NPAS3 might play a role in cell proliferation. Thus, effects of NPAS3 on cell proliferation in vitro were subsequently examined (**Figure 6**). Cell number quantifications (cell validations) of SH-SY5Y and PC12 cells with different treatments were examined by Vita-Orange (WSTr-8). The effects of NPAS3 on cell proliferation in PC12 (**Figure 6B**) and SH-SY5Y (**Figure 6C**) cells are similar.

First, cell proliferation was significantly increased in NPAS3 over expressed cells (SH-SY5Y: pcDNA3.1 vs. NPAS3 P = 0.018; PC-12: pcDNA3.1 vs. NPAS3 P = 0.036) and reduced in NPAS3 knockdown cells (SH-SY5Y: non-specific shRNA vs. shRNA for NPAS3 P = 0.033; PC-12: non-specific shRNA vs. shRNA for NPAS3 P = 0.05). This reveals the proliferative effects of NPAS3.Secondly, siRNA for VGF was used to detect, whether NPAS3 exerts it proliferative function though VGF. The validation of cells transfected by siRNA for VGF was decreased when compared to cells transfected by non-specific siRNA, but not enough to be considered as significant statistically. There was no significant difference between the values of cells transfected by non-specific siRNA and cells transfected by NPAS3 plus siRNA for VGF. The reason for this could be the increased cell proliferation induced by NPAS3 was corrected by knocking down VGF. This indicates that VGF is involved in the proliferation induced by NPAS3. Furthermore, our results showed that cell proliferation was significantly increased by transfecting NPAS3 plus siRNA for VGF compared with neural cells transfected alone by VGF (siRNA; SH-SY5Y: P = 0.012; PC-12: P = 0.044). This phenomenon may be due to two possibilities. First, other NPAS3 targets, which are important for cellular growth and proliferation, may be involved in the proliferative effect of NPAS3 (Sha et al., 2012). The other possibility is that the knocking down method is not effective enough to eliminate VGF and so, NPAS3 still could enhance cell proliferation through VGF.

#### Signaling Pathways Downstream of NPAS3-Induced Cell Proliferation

As it was reported that VGF enhances the proliferation of neurogenesis through glutamatergic pathways, we supposed that NPAS3 might contribute proliferation of neural cells through the same route. Therefore, we first determined if NPAS3 activates downstream signaling molecules of two glutamate receptors, NMDA receptors and mGluR5 in PC12 cells. Inhibitors of NMDA receptors and mGluR5 were used to exploit the association of NPAS3 with glutamatergic pathways. Previous research demonstrated dizocilpine (MK801, 50 µM)'s potential to exert its function as an antagonist to NMDA receptors (Wong et al., 1986). 2-methyl-6-(phenylethynyl)-pyridine (MPEP, 100 µM) was used as an antagonist to mGluR5 (Gasparini et al., 1999). Activation of NMDA receptors induces phosphorylation of Ca2+/CaMKII (Krueger et al., 2010), whereas PKD is regulated by mGluR5 and sustains its phosphorylation form (Bayer et al., 2001).

As shown in **Figure 7A** the phosphorylation of PKD was significantly enhanced upon ectopic expression of NPAS3

(P = 0.000), whereas the NPAS3 over-expression could not enhance the phosphorylation of PKD when mGluR5 was inhibited by MPEP (P = 0.415). To further confirm the participation of VGF in regulation of PKD via NPAS3, p-PKD was observed on blots after treating PC12 cells with VGF (siRNA) alone or along with overexpression of NPAS3 (**Figure 7B**). As can be seen, up-regulated expression of NPAS3 could not facilitate the phosphorylation of PKD when VGF expression was reduced (P = 0.306), which further validated that NPAS3 can activate PKD via mGluR5 by targeting VGF. Similarly phosphorylation of CaMKII was also increased significantly on overexpression of NPAS3 (P = 0.000), whereas the increased expression of NPAS3 failed to augment the phosphorylation of CaMKII in the presence of NMDA inhibitor MK801 (P = 0.166; **Figure 7C**). Likewise phosphorylation of PKD, CaMKII activation was also induced by VGF, which was confirmed by using VGF (siRNA) alone or with ectopically expressed NPAS3 (P = 0.147; **Figure 7D**).

#### DISCUSSION

This study demonstrates that NPAS3 regulates transcription and expression of VGF and delineates that NF-κB signaling pathway is involved in this activation. In addition, this work clarifies that NPAS3 enhances proliferation of neural cells through VGF. This process requires synaptic activity. The signaling molecules required for NPAS3-induced cell proliferation include PKD and CaMKII through glutamate receptors, which were reported to be involved in VGF-induced proliferation of NSCs (Thakker-Varia et al., 2014). Revealing the regulation of NPAS3 on VGF and the precise mechanism of how NPAS3 influences neural cell proliferation will increase our understanding of pathophysiological mechanisms of psychiatric disorders.

To date, adult hippocampal neurogenesis has gained a great deal attention for its potential implication in various psychiatric disorders. Disruption of neurogenesis may reflect the latest stages of a subtle misregulation of brain development and result in

ANOVA followed by Turkey post hoc test).

a particular set of hippocampal symptoms in schizophrenia patients (Kempermann et al., 2008; Hill et al., 2015). However, correlation of these two processes still remains unclear. This study details how several pivotal factors involved in the process of neurogenesis and also in the development of psychiatric illnesses.

VGF was identified as one of the NPAS3 targets in our previous study (Sha et al., 2012). In addition, VGF shares a number of key features with NPAS3: regulation by circadian rhythm (Cirelli and Tononi, 2000); involvement in metabolic control (Altshuler and Hirschhorn, 1999; Salton et al., 2000; Jethwa et al., 2007; Sadahiro et al., 2015); contribution to activity-related adult neurogenesis (Thakker-Varia et al., 2007), and association with neurological diseases (Ruetschi et al., 2005; Selle et al., 2005; Pasinetti et al., 2006; Altar et al., 2009) and psychiatric diseases, including schizophrenia (Huang et al., 2006) and depression (Huang et al., 2006; Malberg and Monteggia, 2008; Thakker-Varia and Alder, 2009).

NPAS3 is a brain-enriched transcription factor containing a bHLH motif at the amino terminus, followed by two PAS domains. The carboxyl-terminal end contains a putative transactivation domain. The bHLH domain contains the DNA binding region, which typically binds to a consensus DNA sequence (E-box, CANNTG; Chaudhary and Skinner, 1999). The PAS domain includes two conserved regions (PAS-A and PAS-B; Crews et al., 1988). Many PAS-domain proteins exert their function by dimerization with another subunit, such as NPAS3-BMAL1 complex. In this project, the correlation of VGF by dNPAS3, a truncated form of NPAS3, was compared with wild type NPAS3. dNPAS3 is a very good model to study the function of individual domains of NPAS3, as it is lacking the second PAS domain and the putative transactivation domain. The transactivation activity of VGF by dNPAS3 is much weaker than the wild type NPAS3. This implicated that each motifs in NPAS3 is important for regulation of VGF.

Our previous studies found location of NPAS3 in hippocampus of adult mice and proposed that it might have a role during neurogenesis (Sha et al., 2012). Furthermore, in the current study, members of NF-κB family might be regulated by NPAS3. The NF-κB family of transcription factors, including p50, p52, RelA/p65, cRel and RelB, have been shown to play important roles in hippocampal neurogenesis and psychiatric disorders (Meffert et al., 2003; Crampton and O'Keeffe, 2013; Bortolotto et al., 2014; Aloor et al., 2015; Malki et al., 2015).

Combining our work and previously published literature, a potential network intersecting the processes of neurogenesis and psychiatric disorders was proposed (**Figure 8**). NPAS3 might regulate VGF not only through E-box, but also through κB site. Besides the κB site, other elements in VGF promoter might be indirectly involved in the regulation of NPAS3 as well. As known, Various NF-κB transcriptional targets, such as Brain-Derived Neurotrophic Factor (BDNF) and Nerve growth factor (NGF; Zaheer et al., 2001; Krock et al., 2016), could regulate VGF (Thakker-Varia and Alder, 2009). Thus, NF-κB (p65) may regulate VGF in part by binding to κB site as well as by modulating BDNF and NGF. Thus, NPAS3, NF-κB, BDNF and NGF are upstream molecules of VGF. Phosphorylation of CaMKII and PKD could be activated by NPAS3-VGF through NMDA receptors and mGluR5, respectively. PKD and CaMKII are downstream of mGluR5 and NMDA receptors. PKD, a serine threonine kinase, is involved in promoting DNA synthesis and cell proliferation in different cells (Wong and Jin, 2005). CaMKII has an important role during maturation of granule cells in the dentate gyrus (Yamasaki et al., 2008). Activation of CaMKII produces increased neurogenesis and LTP in olfactory bulbectomized mice (Moriguchi et al., 2013).

In this network, BDNF and NGF are upstream of VGF and highly associated with psychiatric disorders and neurogenesis. Neuropeptide VGF and BDNF are intrinsic factors that induce neurogenesis and enhance synaptic activity of hippocampal neurons (Alder et al., 2003). VGF is identified as an NGF-responsive gene and they are involved in many processes in the brain (Levi et al., 2004; Ferri et al., 2011). Therefore, NPAS3 and VGF might function to orchestrate the molecular response during the processes of neurogenesis and psychiatric disorders. This work is consistent with previous reports (Sha et al., 2012) and intersects psychiatric candidate genes and molecules involved in neurogenesis and synaptic activity. However, although we have verified the connection of NPAS3 and VGF in neural cell lines, how NPAS3 and VGF influence neurogenesis at different phases need to be further studied in animal models and NSCs.

Understanding the correlation of these factors and how they contribute to the neurogenesis and synaptic activities might shed a light on antidepressant therapies. The neurotrophic/plasticity hypothesis of depression has been proposed more than a decade ago (Duman et al., 1997) and has been supported by multiple basic and clinical studies (Duman et al., 1999; Czeh et al., 2001; Madsen et al., 2003; Sahay and Hen, 2007; Mostany et al., 2008; Li et al., 2010). Recent antidepressant compounds, aiming at new targets and cascades associated with neurotrophic mechanism were developed. Agomelatine, a melatonergic analog drug acting as melatonin agonist and a 5-hydroxytryptamine (5-HT)2C antagonist (Millan et al., 2003), is of particular interest due to its efficacy, safety and tolerability. It has been demonstrated to enhance proliferation and neurogenesis in the ventral dentate gyrus under basal conditions (Banasr et al., 2006; Soumier et al., 2009). Another neuroprotective compound, P7C3, exerts its antidepressant effect through enhancing hippocampal neurogenesis (Walker et al., 2015). Prolonged administration of P7C3 to NPAS3 −/− mice corrected deficits in hippocampal neurogenesis and malformation and dysfunction of dentate gyrus (Pieper et al., 2010).

The antidepressant properties of Agomelatine have been reviewed by Pompili et al. (2013). Agomelatine has a rapid antidepressant actions and good tolerability in most clinical trials. It facilitates all stage of neurogenesis and promotes cell survival in the ventral hippocampus through the combination of activation of MT1/MT2 melatonergic receptor and the blockade of 5-HT2C receptors after chronic administration (Banasr et al., 2006; Soumier et al., 2009). Furthermore, BDNF, ERK1/2, Akt and GSK3β have been reported to participate in the induction of hippocampal neurogenesis by agomelatine (Banasr et al., 2006; Conboy et al., 2009; Soumier et al., 2009; Molteni et al., 2010). Therefore, agomelatine may be considered as an interesting and valid treatment option given its potential in neuroplasticity mechanisms.

Although schizophrenia and major depression are not specifically hippocampal disorders, adult neurogenesis might be involved in hippocampal aspects of psychiatric disorders (Kempermann et al., 2008). Adult neurogenesis can be activated by several physical and chronic antidepressant treatments (Warner-Schmidt and Duman, 2006). Decreased progenitor cell proliferation in adult dentate gyrus in schizophrenia was reported (Reif et al., 2006). A better understanding of the regulation of neurogenesis by psychiatric candidate genes may yield insights into the discovering of more selective targets. Several molecules relevant to pathophysiology of depression, including BDNF, ERK1/2, Akt and GSK3β, could be modulated by agomelatine. As known, VGF could be regulated by BDNF and PI3K/AKT/mTOR signaling (Lu et al., 2014). Whether agomelatine will be efficient in major depressive and schizophrenic patients with abnormal expression of VGF needs to be further examined. In addition, NPAS3 knockout mice also display an additional deficit in adult hippocampal neurogenesis and aberrations in synaptic transmissions (Pieper et al., 2005, 2010). Results of this study suggest that NPAS3 enhances proliferation of neural cells through VGF. It remains to be discovered whether NPAS3−/− mice have disturbances

#### REFERENCES


at the signaling pathways which are agomelatine targets. Furthermore, identification of neuroprotective compounds which could overcome the deficits in hippocampus of NPAS3−/− mice will surely be of value in the development of treatments for psychiatric patients with abnormal expression of NPAS3.

In conclusion, the discovery and understanding of the psychiatric candidate genes will provide novel cellular targets for development of safer, more efficient treatments. The findings in this study indicate a correlation between NPAS3 and VGF, and propose a potential network composed by NPAS3, VGF and several other pivotal factors relevant to neurogenesis and psychiatric disorders. Furthermore, this study demonstrates that NPAS3 enhances cell proliferation through glutamatergic pathways via VGF.

#### AUTHOR CONTRIBUTIONS

LS conceived and supervised the study. LS and DY designed experiments. WZ, DY, YZ, YS and YY performed the experiments. TZ, MH and YL provided facilities. LS, DY, WZ, YL and YY analyzed and interpreted the data. LS, DY and AP wrote and revised the manuscript. KD reviewed the manuscript.

#### FUNDING

This study was supported by National Natural Science Foundation of China (No. 81201044).

#### ACKNOWLEDGMENTS

The authors are grateful to Benjamin Pickard for his encouragement. pcDNA3.1—NPAS3 and pcDNA3.1—dNPAS3 were provided by Dr. Benjamin Pickard (University of Strathclyde, Scotland).


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Yang, Zhang, Padhiar, Yue, Shi, Zheng, Davis, Zhang, Huang, Li and Sha. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Inhibition of Histone Deacetylase 3 (HDAC3) Mediates Ischemic Preconditioning and Protects Cortical Neurons against Ischemia in Rats

#### Xiaoyu Yang† , Qimei Wu† , Lei Zhang and Linyin Feng\*

CAS Key Laboratory of Receptor Research, Shanghai Institute of Materia Medica, Chinese Academy of Sciences, Shanghai, China

Brain ischemic preconditioning (PC) provides vital insights into the endogenous protection against stroke. Genomic and epigenetic responses to PC condition the brain into a state of ischemic tolerance. Notably, PC induces the elevation of histone acetylation, consistent with evidence that histone deacetylase (HDAC) inhibitors protect the brain from ischemic injury. However, less is known about the specific roles of HDACs in this process. HDAC3 has been implicated in several neurodegenerative conditions. Deletion of HDAC3 confers protection against neurotoxicity and neuronal injury. Here, we hypothesized that inhibition of HDAC3 may contribute to the neuronal survival elicited by PC. To address this notion, PC and transient middle cerebral artery occlusion (MCAO) were conducted in Sprague-Dawley rats. Additionally, primary cultured cortical neurons were used to identify the modulators and effectors of HDAC3 involved in PC. We found that nuclear localization of HDAC3 was significantly reduced following PC in vivo and in vitro. Treatment with the HDAC3-specific inhibitor, RGFP966, mimicked the neuroprotective effects of PC 24 h and 7 days after MCAO, causing a reduced infarct volume and less Fluoro-Jade C staining. Improved functional outcomes were observed in the neurological score and rotarod test. We further showed that attenuated recruitment of HDAC3 to promoter regions following PC potentiates transcriptional initiation of genes including Hspa1a, Bcl2l1, and Prdx2, which may underlie the mechanism of protection. In addition, PC-activated calpains were implicated in the cleavage of HDAC3. Pretreatment with calpeptin blockaded the attenuated nuclear distribution of HDAC3 and the protective effect of PC in vivo. Collectively, these results demonstrate that the inhibition of HDAC3 preconditions the brain against ischemic insults, indicating a new approach to evoke endogenous protection against stroke.

#### Keywords: preconditioning (PC), HDAC3, MCAO, RGFP966, calpain

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Daniela Christiane Dieterich, Leibniz Institute for Neurobiology, Germany Natalia N. Nalivaeva, University of Leeds, UK

> \*Correspondence: Linyin Feng lyfeng@simm.ac.cn

†These authors have contributed equally to this work.

Received: 27 August 2016 Accepted: 11 November 2016 Published: 28 November 2016

#### Citation:

Yang X, Wu Q, Zhang L and Feng L (2016) Inhibition of Histone Deacetylase 3 (HDAC3) Mediates Ischemic Preconditioning and Protects Cortical Neurons against Ischemia in Rats. Front. Mol. Neurosci. 9:131. doi: 10.3389/fnmol.2016.00131

**Abbreviations:** ChIP, chromatin immunoprecipetition; DIV, days in vitro; HAT, histone transacetylase; HDAC, histone deacetylase; MCAO, middle cerebral artery occlusion; MTT, 3-(4,5-dimethylthiazolyl)-2,5-diphenyltetrazolium bromide; OGD/R, oxygen-glucose deprivation and reoxygenation and restored energy supply; PC, preconditioning; TTC, 2,3,5- Triphenyltetrazolium chloride.

# INTRODUCTION

fnmol-09-00131 November 24, 2016 Time: 16:29 # 2

Brain ischemia, induced by transient or permanent interruption of the blood supply, is a major cause of mortality and morbidity worldwide (Johnston et al., 2009; Feigin et al., 2014). The mechanisms of neuronal death in stroke are complex; involving cell membrane depolarization, free radical generation, excitotoxicity, and neuroinflammation (Dirnagl et al., 1999; Moskowitz et al., 2010). Despite extensive efforts, therapeutic strategies focusing on the pathological signaling cascades have not been successful (Moretti et al., 2015). Effective treatments for ischemic brain damage remain one of the major unfulfilled medical needs of clinical care (Fisher and Saver, 2015). Ischemic PC is an approach in which a sub-lethal ischemic exposure evokes endogenous protection against a subsequent, more severe ischemic insult. Not only can this therapy be applied in individuals, induced ischemic tolerance is also a strategy to obtain insights into neuroprotective mechanisms (Dirnagl et al., 2009; Stevens et al., 2014). PC stimuli may be identified by diverse sensors and transducers, and consequently initiate protection such as temporal gene profiles resistant to typically lethal ischemic insults by effectors.

Different genomic profiles and epigenetic reprogramming of the brain have been reported between ischemic PC and ischemic injury (Stenzel-Poore et al., 2003; Thompson et al., 2013). Thereby identification of epigenetic determinants of PC is the key for pharmacological manipulation to evoke conditioning. In vivo studies have shown that pan-HDAC inhibitors protected the brain from ischemic injury, by elevating the severe decrease of histone acetylation (Chuang et al., 2009; Langley et al., 2009). The treatment of myocardial ischemia with HDAC inhibitors triggers the PC effects against ischemia/reperfusion injury. Likewise, studies in retina and brain ischemia revealed an elevation of histone acetylation following PC, which may be associated with regulation of the deacetylase activity of HDAC or HAT recruitment (Yildirim et al., 2014; Fan et al., 2016). These studies have raised the hypotheses that HDACs might converge in the conditioning signaling pathways. However, considering the significant effects of HDAC inhibitors against ischemia, less is known about the specific role of HDACs in brain PC.

HDAC3, a homologue of Rpd3 from budding yeast, has been linked to neurotoxicity in several neuropathological conditions (Butler and Bates, 2006; Yang and Seto, 2008). In Caenorhabditis elegans knock-down of the homolog of HDAC3, HDA-3, suppressed Htn-Q150 toxicity in a model of Huntington's disease (Bates et al., 2006). In rat cerebellar granule neurons, mutant Htt disrupted the sequestration of HDAC3 and the liberation of HDAC3 resulted in neurotoxic activity (Bardai et al., 2013). Additionally, suppression by HDAC3 shRNA protected cerebellar granule neurons against a low-potassium insult, while overexpression of HDAC3 promoted the death of neurons (Bardai and D'Mello, 2011). Conditional knock-out of Hdac3 in retinal ganglion cells displayed a significant amelioration of nuclear atrophy and reduction in cell death induced by optic nerve injury (Schmitt et al., 2014).

Given the properties of HDAC3 in neurodegeneration, we speculate whether inhibition of HDAC3 contributes to the neuronal survival elicited by PC. We first investigated histone acetylation and class I HDAC subcellular localization following PC. We found that PC-induced acetyl-histone 3 Lysine 9 (H3K9ac) elevation was accompanied by reduced HDAC3 nuclear localization in cortical neurons. Based on this, efficient and specific pharmacological inhibition of HDAC3 in vivo and knock-down of HDAC3 in vitro in models of ischemia were performed. The results showed that specific inhibition of HDAC3 could precondition the brain against ischemic injury 24 h and 7 days after MCAO in vivo, via the initiation of a gene-expressing program associated with neuroprotection. Furthermore, we identified that calpains were implicated in the cleavage of HDAC3, which blocked the nuclear distribution of HDAC3. Overall, the results demonstrate the importance of HDAC3 in the process of PC, providing a new pharmacological approach to evoke protection due to ischemic conditioning.

## MATERIALS AND METHODS

#### Animals

Adult male Sprague-Dawley rats weighing 250–280 g were used in this study (Laboratory Animal Center, Chinese Academy of Sciences, Shanghai, China). All experimental procedures were performed in accordance with the Guide for the Care and Use of Laboratory Animals (National Institutes of Health) and approved by the Institutional Animal Care and Use Committee of Shanghai Institute of Materia Medica.

## MCAO and Ischemic Preconditioning

All animals were acclimatized for 2 weeks before pharmacological treatment or surgery. The induction of transient focal cerebral ischemia was performed as previously described (Longa et al., 1989), with minor modifications. Briefly, the animals were anesthetized with chloral hydrate (400 mg/kg, i.p.). A 4-0 monofilament nylon surgical suture (Sunbio Biotech Co. Ltd., Beijing, China) with a rounded tip was introduced through the external carotid artery to the internal carotid artery and advanced up to 18–20 mm to block the middle cerebral artery. After a period of occlusion, 5 min for ischemic PC and 90 min for injurious ischemia, the suture was withdrawn to allow reperfusion. Regional cerebral blood flow was monitored throughout surgery using a laser Doppler (moorVMS-LDF2, Axminster, UK) to exclude rats that showed a cerebral blood flow (CBF) reduction of less than 80%. Rectal temperature was monitored and the core body temperature was maintained at 37 ± 0.5◦C by a feed-back-controlled heat pad (Hugo Sachs Elektronik, March-Hugstetten, Germany). Rats were transferred back to their cages after recovery from anesthesia and rested in home cages before sacrifice. Sham operation were conducted using an identical procedure but without the insertion of a filament.

#### Experimental Design

Rats were randomly grouped (n = 8–14 per group) and subjected to different MCAO and pharmaceutical treatments as follows (Supplementary Figures S2A,B):

Sham group: Sham surgery was performed on day 1 and at the same time on day 2. Rats received vehicle injections at the same time as the RGFP966 group or the calpeptin group below.

MCAO group: Rats underwent sham operation on day 1 and 90 min of MCAO on day 2. Rats received vehicle injections at the same time as RGFP966 group or calpeptin group below.

PC group: Rats were subjected to 5 min of MCAO to induce ischemic PC on day 1, and 90 min of MCAO on day 2.

RGFP966 group: Rats were treated with RGFP966 (7.5 mg/kg, i.p) when subjected to a sham operation. The second RGFP966 was injected 6 h prior to the MCAO of 90 min on day 2.

Calpeptin group: Rats were treated with calpeptin (125 µg/kg, i.p) 6 h before surgery, and then underwent the same procedure as the PC group.

Rats were sacrificed 24 h or 7 days after injurious MCAO.

#### Pharmacological Treatment

Drugs were administrated at a volume of 5 ml/kg. RGFP966 (S7229, Selleckchem, Houston, TX, USA) was dissolved in DMSO, and diluted in a vehicle of 100 mM sodium acetate (PH 5.4) and 30% (wt/vol) hydroxypropyl-beta-cyclodextrin, with the final DMSO less than 10% (vol/vol). Calpeptin (C8999, Sigma–Aldrich, St Louis, MO, USA) was dissolved in DMSO and diluted in saline. Vehicles were obtained by identical procedures without drugs. The doses were determined based on pervious report to obtain proper concentrations in the brain (Malvaez et al., 2013; Samantaray et al., 2015).

#### Infarct Volume Quantification

The cerebral infract volumes were measured using TTC (T8877, Sigma–Aldrich) staining. Rat brains were harvested and sectioned into 2 mm thick coronal sections 24 h or 7 days after MCAO. Then sections were stained with 1% TTC at 37 ◦C for 5 min, and fixed in 4% paraformaldehyde (PFA) solution for 48 h. The images were captured and analyzed using Image-Pro Plus 7.0 (Media Cybernetics, Silver Spring, MD, USA). The infarct volume (%) for the brain was calculated with the following formula: (the volume of the contralateral hemisphere – the volume of the nonlesioned ipsilateral hemisphere)/(the volume of the contralateral hemisphere × 2). The ipsilateral hemisphere underwent occlusive treatment in the MCA, while the contralateral hemisphere did not. Investigators were kept blind to the treatment assignments.

#### Neurological Function Analysis

The neurological deficits score after surgery was evaluated according to validated scoring analyses with minor modifications (Longa et al., 1989): 0, no observable deficit; 1, unable to fully stretch the forelimb of the right side (the left MCA was occluded in this study); 2, decreased resistance to lateral push, circling to the right side but normal gesture at rest; 3, unilateral rolling to the right side; 4, occasionally leaning to the right side, declined locomotor activity; 5, unable to sustain normal gesture at rest, unresponsive. Investigators were kept blind to treatment assignments.

# Rotarod Test

Rats were trained and tested on rotarod (IITC Life Science, Woodland Hills, CA, USA) which gradually accelerated from 4 to 40 rpm over 5 min (Hunter et al., 2000). The latency to fall was recorded as the time before rats fell off the rod or gripped and spun around for two successive revolutions. Rats underwent three training trails a day for 3 days. The base line control was recorded 1 day before the surgery. The mean latency was obtained from 3 trials with 30 min interval on the testing day. Investigators were kept blind to treatment assignments.

## Primary Culture of Rat Cortical Neurons and Treatment

Primary cortical neurons were prepared from embryonic E17 Sprague-Dawley rats and cultured with Neurobasal medium and B27 supplement (Thermo Fisher Scientific, USA) as described previously (Xie et al., 2009). Neurons were cultured at 37◦C in a humidified 5% CO<sup>2</sup> atmosphere and the medium was replaced by fresh every 3 days. Neurons at 7 DIV were used for experimentation. In the OGD treatment, the glucose-free Dulbecco's Modified Eagle Medium (DMEM) (Thermo Fisher Scientific) was bubbled with 95% N2/5% CO2. The chamber was pre-warmed, humidified, and flushed with 95% N2/5% CO<sup>2</sup> at 3 liter/min for 5 min. The oxygen content was under 0.50% (v/v) throughout the experimental period detected with oxygen monitor (Jiande Analytical Instrument, Hangzhou, China). For PC treatment, the culture medium was replaced with and cells were placed glucose-free DMEM in an experimental hypoxia chamber for 45 min for neurons or 1.5 h for PC12. The cells were then removed from the chamber and cultured under normal conditions for the indicated periods (see figure legends). For the OGD/R model, neurons were cultured in glucose-free DMEM in a hypoxia chamber for 90 min and returned to normal culture conditions for 24 h. The control cells were cultured in a medium with glucose in a normal oxygen-condition incubator for the same time periods.

# Tissue Preparation

After anesthetization, brains were immediately removed. Brain tissue were obtained by coronal sectioning and harvested along the core region of occlusion underlying the middle cerebral artery. The tissues were immediately frozen in liquid nitrogen, and stored at −80◦C.

# Western Blot

The brain tissue or cultured cells were lysed in a RIPA buffer with proteinase and phosphatase inhibitor cocktail (Sigma–Aldrich) for whole cell protein. The nuclear and cytoplasmic fractions were extracted with NE-PER Nuclear and Cytoplasmic Extraction Reagents Kit (Pierce, Rockford, IL, USA) following the manufacturer's instruction. The protein concentration was quantified using a BCA assay kit (Thermo Fisher Scientific). Samples were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to PVDF membranes (Merck-Millipore, Bedford, MA, USA). Blots were then probed with specific antibodies:

anti-H3K9ac (ab12179, Abcam, Cambridge, MA, USA), anti-HDAC1 (WH0003065M2, Sigma–Aldrich), anti-HDAC2 (ab16032, Abcam), anti-HDAC3 (ab47237, Abcam), anti-lamin A/C (2032, Cell Signaling Technology, Beverly, MA, USA), anti-β-actin (A8481, Sigma–Aldrich), anti-Bcl-xL (2764, Cell Signaling Technology), anti-Prdx2 (10545-2-AP, Proteintech Group, Chicago, IL, USA), anti-HSP70 (sc-32239, Santa Cruz, Dallas, TX). After incubation with HRP-conjugated secondary antibodies (Santa Cruz), chemiluminescence signals were detected with ECL reagents (GE Health, Little Chalfont, UK).

### Histological and Immunohistochemical Assessment

After anesthesia with chloral hydrate, rats were injected transcardially with ice-cold 4% PFA in PBS. Brains were post-fixed in PFA for 24 h and dehydrated with 30% (wt/vol) sucrose for 48 h. Next, 25 µm coronal slices were obtained with a Leica freezing microtome. Primary antibodies used in immunofluorescence staining were: anti-H3K9ac (ab12179, Abcam), anti-HDAC3 (ab47237, Abcam) and anti-MAP2 (ab11267, Abcam). The sections were incubated with 300 nM 4,6-diamidino-2-phenylindole (DAPI, Sigma), Alexa Fluor 488 and Alexa Fluor 555 secondary antibodies (Thermo Fisher Scientific) for 45 min at 25◦C. An Olympus FV1000 confocal laser scanning microscope was applied for acquisition of fluorescence images. 3,3<sup>0</sup> -diaminobenzidine immunostaining was performed with primary antibodies against: HSP70 (sc-32239, Santa Cruz), following with SignalStain Boost IHC Detection Reagent and visualized with SignalStain DAB Substrate Kit (Cell Signaling Technology). Images were acquired using a Leica microscope. Fluoro-Jade C (FJC) staining was performed according to the protocol provided by the manufacturer (Chemicon, Merck-Millipore). Images were obtained using an Olympus FV1000 confocal microscope. For histological quantification of H3K9ac intensity and FJC positive cells, images were obtained from two random 20× fields of the cortical area in three coronal sections (Interaural 10.60 mm/Bregma 1.60 mm, Interaural 9.20 mm/Bregma 0.20 mm, Interaural 7.12 mm/Bregma −1.88 mm) (Supplementary Figure S2C). Average intensities or cell counts were calculated from 5 rats per group with Image Pro Plus 7.0. Investigators were kept blind to treatment assignments.

#### Co-immunoprecipitation (Co-IP) Assay

Cell lysates were generated by incubating cells with an NP-40 lysis buffer with proteinase inhibitor cocktail on ice and then centrifuged at 10000 g for 10 min. Immunoprecipitation was carried out by incubating the lysate with the corresponding antibodies: anti-HDAC3 (ab47237, Abcam), anti-calpain1 (ab28258, Abcam), anti-calpain2 (ab39165, Abcam) and anti-HA (3724, Cell Signaling Technology) with slow agitation overnight at 4◦C. Next, 20 µl PureProteome Protein A/G Mix Magnetic Beads (Merck-Millipore) was added and incubated with slow agitation for 4 h at 4◦C. Immune complexes were washed extensively, boiled in SDS-sample buffer and analyzed using SDS-PAGE.

# Quantitative Reverse Transcription Polymerase Chain Reaction (qRT-PCR)

Total RNA was extracted from cortical neurons after treatment using TRIzol reagent (Thermo Fisher Scientific). Next, 1 µg total RNA was converted to cDNA using a reverse transcription kit (Takara, Cat# RR047A). After the RT reaction, cDNA was used for subsequent real time PCR (SYBR Premix Ex TaqII Kit) following the manufacturer's protocol. The qPCR and data collection were carried out on ABI 7500 real-time PCR system (Applied Biosystems, Foster City, CA, USA). The results relative to the quantitation value for each target gene were expressed as 2 <sup>−</sup>11Ct. All PCR primers are shown in Supplementary Table S1.

# Chromatin Immunoprecipetition and Real-Time PCR

Chromatin immunoprecipetition was carried out according to the protocol of the EZ-Magna ChIP kit (17-10086, Millipore). Briefly, the nuclear content of cortical neurons was extracted and the chromatin inside was sonicated into fragments. Fragmented chromatin was incubated and immunoprecipitated using anti-H3K9ac antibody (17-609, Millipore) and anti-HDAC3 antibody (17-10238, Millipore). After the de-crosslink, immunoprecipitated DNA and whole-cell extract DNA were then eluted and subjected to real-time PCR analysis using SYBR Premix Ex TaqII Kit (Takara, Shiga, Japan). All PCR primers are shown in Supplementary Table S2.

# Plasmid Constructs and Transfection

Full-length human HDAC3 cDNA was synthesized by GENEWIZ (South Plainfield, NJ, USA) and ligated into the expression vector, HA-pcDNA3.0. C-terminus-truncated construct HDAC3 mutant (1C-HDAC3, 1-312) was amplified with primers: 50GATATCGATGGCCAAGACCGTGGCGTATT TC3<sup>0</sup> and 50CTCGAGTTAAGATGTTTCATATGTCCAGCAC CG3<sup>0</sup> using full length HDAC3 cDNA as the template. Transient transfection was conducted using the FuGENE HD transfection reagent (Promega, Madison, WI, USA) according to the manufacturer's instructions. siRNA-HDAC3 (s136736, Dharmacon) was transfected into cortical neurons by the FuGENE HD transfection reagent. The silencing effect of protein expression was confirmed by western blot analysis.

# Cell Viability

Neuron viability was assessed using the MTT assay as described previously. Briefly, MTT solution (0.5 mg/ml) was added to each well and incubated with cells for an additional 3 h at 37◦C. The formazan was then dissolved with DMSO and the absorbance at 595 nm was read by a microplate reader (NOVOstar, BMG LABTECH, Offenburg, Germany).

#### Statistical Analysis

Quantifications and statistical analysis were carried out using GraphPad Prism 6 (Graphpad Software, San Diego, CA, USA). All values are presented as means ± SEM. Differences between groups were compared with a one-way ANOVA followed by

group comparisons using a post hoc Bonferroni test or twotailed Student's t-test. Neurological deficit scores were analyzed using the Kruskal–Wallis non-parametric test followed by Dunn's multiple comparisons test.

# RESULTS

#### Acetylation of H3K9 Increased in Cortical Neurons in Preconditioning, Accompanied by Reduction of Nuclear Localization of HDAC3

Histone acetylation was shown to be differentially regulated depending on the duration of ischemia. While injurious ischemia led to a severe decline of histone acetylation in stroke models (Supplementary Figure S1A) (Ren et al., 2004; Faraco et al., 2006; Kim et al., 2007), PC for 5 min followed by 2 h reperfusion in rats induced increased levels of acetyl-H3K9 in the cortex, as assessed with both immunoblotting (p < 0.05, n = 5) and immunohistochemistry (p < 0.001, n = 5) (**Figure 1A**). We next measured levels of expression of the three class I HDAC subtypes with which the deacetylation of histone lysine residuals is largely associated. Western blot assay on nuclear lysates revealed that the levels of HDAC3, but not HDAC1 or HDAC2, were reduced in the nuclei following PC (p < 0.05, n = 4–6) (**Figure 1B**). To further determine the involvement of HDAC3 in PC, the subcellular localization of HDAC3 was assessed with immunofluorescence. The results showed that HDAC3 was distributed both in the nucleus and cytoplasm in cortical neurons before PC, while it was localized mainly in the cytoplasm following PC treatment (**Figure 1C**). Meanwhile, the modulation of histone acetylation (**Figure 1D**) and the nuclear reduction of HDAC3 following PC were verified with primary cortical neuron cultures (**Figure 1E**, Supplementary Figures S1C,D). These results indicated that a diminished recruitment of HDAC3 to the genomic arena might contribute to the modulation of the histone acetylation profile following PC.

# Inhibition of HDAC3 Mimicked the Neuroprotective Effects of PC In vivo

Given the changes in redistribution of HDAC3 following PC, we wondered whether inhibition of HDAC3 mimics the neuroprotective effect of PC. Rats with different treatments were subjected to MCAO (Supplementary Figure S2A). RGFP966, a known HDAC3-specific inhibitor, was injected 24 and 6 h prior to MCAO. The RGFP966-treated group exhibited significantly reduced infarct volume compared with the vehicle-treated group 24 h (7.68 ± 2.90% versus 19.09 ± 2.91%, p < 0.05, n = 7–9) and 7 days (8.62 ± 2.00% versus 18.25 ± 3.75%, p < 0.05, n = 8–10) after reperfusion. Furthermore, the anterior-posterior analysis of infarct distribution revealed that the infarction mostly occupied the striatum and temporal cortex, which was decreased in slices 2, 3, 4 in the PC and RGFP966 group (**Figures 2A,B**). Brain coronal sections obtained 24 h and 7 days after MCAO were stained with FJC to label degenerating neurons (**Figures 2C,D**). Higher levels of FJC staining were observed in the MCAO group compared to with the Sham group, and significantly decreased labeling with FJC in the RGFP966-treated group compared with the MCAO group in both 24 h and 7 days assessments (n = 5). Furthermore, we analyzed the counterstain of FJC with GFAP or Iba with brain sections obtained 24 h after MCAO. The schematic brain section and outlined box showed the area where images were acquired in the ipsilateral hemisphere (Supplementary Figure S5A). Activated astrocytes and microglia were observed in MCAO group, while less activation in PC and RGFP966 group (Supplementary Figures S5B,C). Meanwhile, neurological function was evaluated using the neurological deficit score and rotarod test. Rats treated with RGFP966 displayed lower neurological deficit scores (n = 8–10) and improved rotarod performance (n = 8–10) throughout 7-day test compared with the Sham group (**Figures 2E,F**).

# Attenuated Recruitment of HDAC3 to Promoter Regions Following PC Potentiated Transcriptional Initiation of Oxidation Relative Genes

Concomitant with the data in vivo, the reduction in cell viability elicited by OGD/R was blocked by both RGFP966 pretreatment and HDAC3 knock-down in primary cultured cortical neurons, as determined by the MTT assay (**Figure 3A**). To further elaborate the underlying role of HDAC3 in PC, data mining linking HDAC3 targeting genes (Feng et al., 2011) and ischemia was performed and determined a group of associated genes. ChIP analysis revealed that promoter occupancy of HDAC3 on Hspa1a, Prdx2 and Bcl2l1 were markedly reduced in response to PC, and concomitantly H3K9 acetylation levels were significantly elevated. Furthermore, RGFP966 pretreatment or HDAC3 knock-down also induced increases in H3K9 acetylation near the HDAC3 binding sites of indicated genes (**Figures 3B,C**). Consistently the expression of indicated genes increased significantly in PC, RGFP966 treatment or HDAC3 knock-down (**Figure 3D**). We next exposed primary neurons to injurious OGD/R with the same procedure in **Figure 3A**. Western blot analysis showed that OGD/R slightly induced HSP70 expression, but suppressed Bcl-xL and Prdx-2 expression, while RGFP966 treatment or HDAC3 knock-down, consistent with PC, enhanced expression of all indicated genes compared with OGD/R (**Figure 3E**, Supplementary Figure S6). Taken together, we have demonstrated that the diminished recruitment of HDAC3 observed in PC led to up-regulation of HDAC3 related genes associated with neuroprotection.

# Inhibition of Calpain1/Calpain2 Blocked the Reduction of HDAC3 Nuclear Localization Following PC

Subcellular localization of HDAC3 was shown to be controlled by protein-protein interaction or proteolytic cleavage (Karagianni and Wong, 2007). The unique C-terminus of HDAC3 is essential for nuclear localization and assembly into co-repression complexes (Yang et al., 2002). Several cleavages of HDAC3 by different proteases, which are activated with temporal specificity

in vitro. Rats underwent 5 min PC treatment followed by 2 h reperfusion (A–C). (A) Acetyl-H3K9 levels in the cortex by immunoblotting (Left, n = 5) and immunohistochemistry (Right, n = 5). <sup>∗</sup>p < 0.05, ∗∗∗p < 0.001 Student's t-test. Scale bar: 100 µm. (B) Quantification of class I HDAC subtypes in nuclear lysates (n = 4–6). <sup>∗</sup>p < 0.05, Student's t-test. (C) Representative images of HDAC3 subcellular localization following sham or PC treatment in vivo. Scale bar: 50 µm. (D) Primary cultured cortical neurons underwent PC for 45 min followed by reoxygenation and restored energy supply for indicated period. Levels of H3K9 acetylation were assessed with whole cell lysates. Data are mean ± SEM. ∗∗p < 0.01 versus control, ANOVA. (E) Representative images of HDAC3 subcellular localization in rat cortical neurons after PC. Scale bar: 25 µm.

∗∗p < 0.01 versus MCAO group, ANOVA.

in the process of cerebral ischemia, are involved in regulating the cellular distribution of HDAC3. We pretreated neurons with three major proteolytic pathway inhibitors including bortezomib (proteasome inhibitor), Z-VAD-FMK (pan-caspase inhibitor), and calpeptin (calpain inhibitor). PC was then conducted. Immunofluorescence results showed that calpeptin pretreatment, but not bortezomib or Z-VAD-FMK rescued the reduction of nuclear HDAC3 (**Figure 4A**). Western blot analysis with whole cell lysates showed a lower band of HDAC3 in PC treated neurons, suggesting a cleavage of HDAC3 induced by PC (**Figure 4B**). Furthermore, immunoprecipitation was performed with anti-HDAC3 or anti-calpain1/calpain2

genes. (A) For non-PC groups, cortical neurons at 7 DIV were pretreated with RGFP966 (2 µM) for 2 h, or transfected with siRNA targeting HDAC3 at 5.5 DIV 36 h in advance. For PC group, PC was induced at 6 DIV followed by reoxygenation. All groups underwent OGD/R at the same time at 7 DIV. At the end of the OGD/R period, cell viability was assessed using an MTT assay and calculated as the percentage of the non-treatment control. ###p < 0.001 versus control; ∗∗p < 0.01, ∗∗∗p < 0.001 versus OGD/R, ANOVA. Primary cultured neurons were subjected to PC or pretreated with RGFP966 for 2 h at 7 DIV, or transfected with siRNA targeting HDAC3 at 5.5 DIV 36 h in advance. Samples for ChIP were collected right after PC or RGFP966 and siHDAC3 treatment (B,C). For mRNA test samples were collected 6 h after PC followed by reoxygenation (D). (B) ChIP with HDAC3 antibodies was conducted followed by qPCR analysis using primers for HDAC3 binding sites near the promoter of the indicated genes. (C) Acetylation of H3K9 near the HDAC3-targeting gene promoters were analyzed by ChIP-qPCR. (D) RT-qPCR analysis for indicated genes. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 versus control, Student's t-test and ANOVA. (E) Neurons were induced with PC, pretreated with RGFP966 or siHDAC3 and then exposed to OGD/R (the same procedure with A). The expression of oxidative relative genes was analyzed at the protein levels. Cell lysates were analyzed with specific antibodies. Data are presented as means ± SEM from three independent experiments. #p < 0.05, ##p < 0.01, ###p < 0.001 versus control; <sup>∗</sup>p < 0.05, ∗∗p < 0.01 versus OGD/R, ANOVA.

antibodies. We found that direct interactions between HDAC3 and calpain1/calpain2 were strengthened after PC (**Figure 4C**). To further identify the interaction domain between HDAC3 and calpain, we constructed HA-tagged WT HDAC3 and truncated HDAC3 (aa1-312, HDAC3 1C) plasmids. Consistent with previous reports (Yang et al., 2002), HA-HDAC3 WT displayed both cytoplasmic and nuclear distributions, while HA-HDAC3 1C specifically localized in the cytoplasm of neurons (Supplementary Figure S3). A PC12 cell line was transfected with HDAC3 constructs and underwent PC. As expected, PC significantly increased the interaction of HA-HDAC3 WT with calpains, while truncated mutation, HA-HDAC3 1C, reduced

Frontiers in Molecular Neuroscience | www.frontiersin.org November 2016 | Volume 9 | Article 131 |

the interaction with calpains after PC compared with control group (**Figure 4D**), indicating that the interaction domain may lie in the C-terminal region of HDAC3. From these results we concluded that blockade of HDAC3 nuclear localization following PC treatment resulted from the C-terminal cleavage of HDAC3 mediated by calpain1/calpain2.

## Calpeptin Blocked the Attenuated Nuclear Localization of HDAC3 and the Neuroprotective Effect of PC In vivo

Rats were treated with calpeptin 6 h prior to surgery. To determine the subcellular localization of HDAC3, brain sections were obtained 2 h after PC. Immunofluorescence results showed that calpeptin pretreatment blocked the reduction of nuclear HDAC3 induced by PC (**Figure 5A**). For pathological assessment, rats subjected to MCAO were sacrificed 24 h or 7 d later (Supplementary Figure S2B). TTC staining (**Figure 5B**) and FJC immunoreactivity (**Figure 5C**) revealed that calpeptin had significantly attenuated the protective effect of PC 24 h after MCAO, consistent with results observed 7 days after MCAO (Supplementary Figure S4). Activated astrocytes and microglia were observed in MCAO and calpeptin group compared with PC group (Supplementary Figures S5D,E). Furthermore, we compared the expression level of indicated genes in all experimental groups, including the RGFP966 treatment group. Specimens were obtained 24 h after MCAO. Consistent with results in vitro, the levels of Prdx-2, Bcl-xL and HSP70 were significantly elevated in PC and RGFP966 groups compared with the MCAO group, while pretreatment with calpeptin attenuated the expression enhancement of PC (**Figure 5D**). HSP70 regulation was further assessed by immunohistochemistry, which revealed that HSP70 expression was enhanced by PC and RGFP966 treatment, whereas its expression was blocked by calpeptin pretreatment (**Figure 5E**).

# DISCUSSION

Ischemic PC has been studied in order to develop new therapeutic approaches to benefit patients with stroke (Dirnagl et al., 2003; Gidday, 2006; Stetler et al., 2014). The present study provides evidence that HDAC3 inhibition may have a role in PC the brain against ischemia. We found that the reduction of HDAC3 nuclear localization following PC resulted in elevated acetylation of H3K9. Acting as one of the leading causes of PC, release of the HDAC3 co-repressor complex facilitated the transcription of relevant genes including Hspa1a, Bcl2l1, and Prdx2, which contributed to the endogenous protection elicited by PC. Further, we showed that C-terminal cleavage of HDAC3 by calpain underlies the failure of HDAC3 nuclear localization, and inhibition of calpain attenuated the PC protection exerted by HDAC3 inhibition.

There has been increasing interest in targeting epigenetic process as potential treatment for stroke (Langley et al., 2009; Aune et al., 2015). Several HDAC inhibitors such as VPA (Kim et al., 2007; Wang et al., 2012b), TSA (Yildirim et al., 2008; Wang et al., 2012a) and SAHA (Faraco et al., 2006) have previously shown protective effects against cerebral ischemia. Meanwhile, epigenetic modification is believed to play a fundamental role in ischemic tolerance. Yet the precise roles of HDACs in PC remains less clearly understood. Notably, it has been reported that total histone-H3 acetylation was down-regulated in cerebral ischemic injury but up-regulated following PC both in vitro and in vivo (Yildirim et al., 2014). Another study revealed a significant increase in retinal acetyl histone-H3 labeling after PC but a pronounced decrease after ischemia (Fan et al., 2016). In our study we further illustrated the time course of histone modification in both PC and OGD/R in vitro. Acetyl-H3K9 was elevated within 2 h following PC and returned to normal levels 6 h later (**Figure 1C**), while it was decreased profoundly 12 h after OGD followed by reoxygenation (Supplementary Figures S1A,B). Different mechanisms underlying the opposite regulation have been proposed. In the retina the deacetylase activity of HDAC1/2 displayed an increase in ischemia but a decrease following PC, which was closely correlated with histone acetylation modification (Fan et al., 2016). Whereas in the brain, a decrease in histone H3 acetylation was observed for 6 h after MCAO, strikingly without concomitant change in HAT or HDAC activity, which might be associated with decreased acetyl-CoA contents after ischemia (Faraco et al., 2006). In the brain following PC, increased histone acetylation was reported to result from enhanced CBP recruitment at the promoter of the neuroprotective gene gelsolin (Yildirim et al., 2014). Here, we reported that additional regulation is involved in this process, whereby cleaved HDAC3 is distributed in the cytoplasm and reducing the recruitment to targeted genes following PC, which consequently elevated the level of acetyl-H3K9. All these finding present a new perspective in explaining the protective role of HDAC inhibitors that reinstatement of the decreased histone acetylation level may exert the PC protection following ischemic injury.

Numerous studies have identified the mechanisms of PC against injurious ischemia. Gene profile analysis demonstrated that PC induced an extensive down-regulation of genes responsible for metabolic pathways and ion-channel activity, indicating lowered cellular activity (Stenzel-Poore et al., 2003, 2007). Similar regulation is implicated in the response to resveratrol treatment, which functions via the activation of SIRT1 (NAD-dependent deacetylase sirtuin-1; Raval et al., 2006; Koronowski et al., 2015). HIF-1-alpha also plays a fundamental role in adapting the metabolism to hypoxic conditions via transcriptional regulation (Prass et al., 2003; Tang et al., 2006). Apart from the known effectors that suppress energy turnover, cumulative evidence has suggested that epigenetic modulators may be involved in PC paradigms. The volume of ischemic tissue was reduced by the inhibition of DNA methylation and histone acetylation, which elicited a reprogramming of ischemic tolerance (Thompson et al., 2013). In our study, ChIP assays have revealed that the induction of PC, HDAC3 knock-down and HDAC3 inhibition elevated the histone acetylation level of related gene including Hspa1a, Bcl2l1, and Prdx2, and were further permissive to

transcriptional initiation. Notably, over 12-fold up-regulation of HSP70 transcription was observed 12 h in PC, siHDAC3 and RGFP966 treated cortical neurons 6 h after PC induction or mimicking. It has been reported that HSP70 was induced in the penumbra, but not core, of the ischemic region, as well as in neurons which survive PC (Masada et al., 2001). Delaying administration of HSP70 by 2.25 and 3 h after the onset of ischemia significantly decreased infarct volume by 68% (Zhan et al., 2010). Bcl-xL is required for neurite outgrowth and potentially inhibits programmed cell death through changing the

Frontiers in Molecular Neuroscience | www.frontiersin.org November 2016 | Volume 9 | Article 131 |

topology of the mitochondrial membrane (Park et al., 2015). Peroxiredoxin, an antioxidant enzyme, protects cells from the insults due to free radical accumulation (Finkel and Holbrook, 2000). These observations suggest that like other methods of manipulating PC, HDAC3 inhibition triggers signaling cascades in neurons leading to increased levels of free radical species and superoxide clearance, inhibition of the endogenous cell apoptosis program, and improved overall protein integrity, which shield the brain from ischemic injury.

In our study we demonstrated that calpain-mediated C-terminus cleavage of HDAC3 was responsible for the absence of HDAC3 in nuclei following PC. HDAC3 is a member of the class I HDAC family (HDAC1, HDAC2, HDAC3, and HDAC8), which share homology with Rpd3 from budding yeast (Yang and Seto, 2008). The unique C-terminus of HDAC3 is essential for its deacetylase activity and nuclear localization (Yang et al., 2002). Several ways of HDAC3 cleavage have been reported to affect its distribution. Etoposide treatment resulted in the cleavage of HDAC3 by caspase-3 and its subsequent cytoplasmic accumulation (Choi et al., 2015). Following treatment with FasL or ultraviolet irradiation, HDAC3 was cleaved in an indirect caspase-dependent manner and subsequently distributed in the cytoplasm (Escaffit et al., 2007). In the present study, minimum amount of interaction between HDAC3 and calpains were observed under normal conditions. While following PC and pretreatment with different protease inhibitors, we confirmed HDAC3 was cleaved by calpains, but not caspase or proteasome. From our results we cannot determine the difference in cleavage participation between calpain1 and calpain2. Through truncation research, we confirmed that the non-conserved C-terminal region of HDAC3 is required for the interaction of calpain with HDAC3, but the specific sequence remains to be illustrated. Calpains are a family of calcium-dependent proteases, which are profoundly and causally linked to post-ischemic insults (Bevers and Neumar, 2008), and calpain inhibitors have been investigated as potential therapeutics for cerebral ischemia (Koumura et al., 2008). The sustained calcium overload in stroke led to pathological activation of calpains, which resulted in extensive degradation of structural proteins and enzymes and neuronal cell death. The spectrin cleavages generated by calpains, SBDP150 and SBDP145, appeared 60 min after ischemia, and followed by a second expansion between 24 and 48 h after reperfusion (Neumar et al., 2001; Lin et al., 2013). These two steps of calpain activation may indicate that the controlled activation of calpain was induced at the beginning of ischemia, whereas sustained calcium overload resulted in pathological hyper-activation of calpains, which played predominant pathologic protease activity in vulnerable neurons following ischemia (Zhang et al., 2002; Liu et al., 2008). Regulated activation of calpain in the brain are involved in synaptic function and memory formation, but the mechanism underlying the controlled activation of calpain in PC has not been illustrated. We propose that firstly, PC is not an injurious stimulus, whereby transient signaling cascades may not lead to loss of Ca2<sup>+</sup> homeostasis and pronounced activation of proteases. Secondly, other signaling pathways involved in PC may regulate and stabilize the homeostasis of Ca2+. For example, A1 adenosine receptors are reported to mediate the protection against infarction afforded by PC (Liu et al., 1991). And adenosine release from astrocytes down-regulated the synaptic activity level in transient hypoxia by negatively modulating the external or internal Ca2<sup>+</sup> concentrations (Martín et al., 2007).

Here, we demonstrated that HDAC3 inhibition increased the resistance of the CNS to ischemic insult. Controlled activation of calpains in PC was involved in the cleavage of HDAC3, which reduced nuclear localization of HDAC3 and led to elevated acetylation of H3K9 and initiation of neuroprotective genes. Consistent with the role of HDAC inhibitors in neuroprotection, this study illustrates the role of specific HDAC subtypes in ischemic PC, and proposes new connections between PC and epigenetic effectors.

#### ETHICS STATEMENT

The using of animals and experimental procedures were performed following the rules of the Association for Assessment and Accreditation of Laboratory Animal Care International and approved by the Institutional Animal Care and Use Committee of the Shanghai Institute of Material Medica, Chinese Academy of Sciences. No endangered animals were applied in this research.

#### AUTHOR CONTRIBUTIONS

XY and LF designed experiments. XY and QW performed most of the experiments. XY, QW, and LZ contributed to the animal behavior studies. XY, QW, and LZ analyzed the data. XY and LF wrote the paper.

# ACKNOWLEDGMENTS

This work was supported by grants from "Key New Drug Creation and Manufacturing Program" of the National Science and Technology Major Project (2014ZX09102-001-05), and the "Personalized Medicines—Molecular Signature-based Drug Discovery and Development," Strategic Priority Research Program of the Chinese Academy of Sciences (XDA12040304). All experiments were conducted in compliance with the ARRIVE guidelines.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2016.00131/full#supplementary-material

# REFERENCES

fnmol-09-00131 November 24, 2016 Time: 16:29 # 13


Finkel, T., and Holbrook, N. J. (2000). Oxidants, oxidative stress and the biology of ageing. Nature 408, 239–247. doi: 10.1038/35041687


Karagianni, P., and Wong, J. (2007). HDAC3: taking the SMRT-N-CoRrect road to repression. Oncogene 26, 5439–5449. doi: 10.1038/sj.onc.1210612



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Yang, Wu, Zhang and Feng. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Aberrant Expression of Histone Deacetylases 4 in Cognitive Disorders: Molecular Mechanisms and a Potential Target

#### Yili Wu1,2, Fei Hou<sup>3</sup> , Xin Wang<sup>1</sup> , Qingsheng Kong2,4 \*, Xiaolin Han<sup>2</sup> \* and Bo Bai<sup>2</sup> \*

<sup>1</sup> Department of Psychiatry, Jining Medical University, Jining, China, <sup>2</sup> Collaborative Innovation Center for Birth Defect Research and Transformation of Shandong Province, Jining Medical University, Jining, China, <sup>3</sup> College of Science, Qufu Normal University, Jining, China, <sup>4</sup> Department of Biochemistry, Jining Medical University, Jining, China

Histone acetylation is a major mechanism of chromatin remodeling, contributing to epigenetic regulation of gene transcription. Histone deacetylases (HDACs) are involved in both physiological and pathological conditions by regulating the status of histone acetylation. Although histone deacetylase 4 (HDAC4), a member of the HDAC family, may lack HDAC activity, it is actively involved in regulating the transcription of genes involved in synaptic plasticity, neuronal survival, and neurodevelopment by interacting with transcription factors, signal transduction molecules and HDAC3, another member of the HDAC family. HDAC4 is highly expressed in brain and its homeostasis is crucial for the maintenance of cognitive function. Accumulated evidence shows that HDAC4 expression is dysregulated in several brain disorders, including neurodegenerative diseases and mental disorders. Moreover, cognitive impairment is a characteristic feature of these diseases. It indicates that aberrant HDAC4 expression plays a pivotal role in cognitive impairment of these disorders. This review aims to describe the current understanding of HDAC4's role in the maintenance of cognitive function and its dysregulation in neurodegenerative diseases and mental disorders, discuss underlying molecular mechanisms, and provide an outlook into targeting HDAC4 as a potential therapeutic approach to rescue cognitive impairment in these diseases.

#### Edited by:

Andrew Harkin, Trinity College Dublin, Ireland

#### Reviewed by:

Elizabeth A. Thomas, Scripps Research Institute, USA Shuchi Mittal, Harvard Medical School, USA

#### \*Correspondence:

Qingsheng Kong qingshengkong2016@126.com; jnyxykqs@163.com Xiaolin Han sdjnhxll@sohu.com Bo Bai bbai@mail.jnmc.edu.cn

Received: 12 August 2016 Accepted: 17 October 2016 Published: 01 November 2016

#### Citation:

Wu Y, Hou F, Wang X, Kong Q, Han X and Bai B (2016) Aberrant Expression of Histone Deacetylases 4 in Cognitive Disorders: Molecular Mechanisms and a Potential Target. Front. Mol. Neurosci. 9:114. doi: 10.3389/fnmol.2016.00114 Keywords: HDAC4, cognitive function, cognitive impairment, neurodegenerative diseases, mental disorders

# INTRODUCTION

Histone deacetylases (HDACs), accompanying with histone acetyltransferases (HATs), are implicated in chromatin remodeling and subsequent transcription regulation by controlling the status of histone acetylation. Histone acetylation makes chromatin conformation more relaxed, facilitating gene transcription, whereas histone deacetylation induces a condensed chromatin conformation repressing gene transcription. By controlling the status of histone acetylation, HDACs are involved in diverse physiological and pathological processes. Moreover, the function of HDACs is not limited to the histone deacetylation. Recent evidences suggest that HDACs may also contribute to the deacetylation of non-histone proteins (Lardenoije et al., 2015). In addition, HDACs also have deacetylase-independent functions, such as histone deacetylase 4 (HDAC4) (Lardenoije et al., 2015; Han et al., 2016).

HDAC4 is highly expressed in brain (Grozinger et al., 1999; Bolger and Yao, 2005; Darcy et al., 2010). It plays a key role in the maintenance of cognitive function and its alteration is associated with cognitive impairment in both age-related neurodegenerative diseases (e.g., Alzheimer's disease, AD) and development-related mental disorders (e.g., autism). Therefore, the role of HDAC4 in cognitive function, its dysregulation in cognitive impairmentrelated neurodegenerative diseases and mental disorders, and underlying mechanisms are discussed in this review.

#### HDAC4 and HDACs

#### HDACs Classification

fnmol-09-00114 October 28, 2016 Time: 12:48 # 2

Eighteen human HDACs are identified and classed into four groups based on their homology to yeast HDACs (Didonna and Opal, 2015). Class I HDACs, consisting of HDAC1, 2, 3, and 8, are homologous to yeast RPD3 while class II HDACs have high identity to yeast HDA1, consisting of HDAC4, 5, 6, 7, 9, and 10. According to the protein structure and motif organization, class II HDACs are further divided into two subclasses, class IIa with HDAC4, 5, 7, and 9, and class IIb with HDAC6 and 10. Class III HDACs, named sirtuins, including SIRT1-7, are homologous to yeast SIR2. Compared with zinc-dependent HDACs of class I and class II, class III HDACs are nicotinamide-adenine-dinucleotide (NAD) dependent. HDAC11 is the only member of Class IV, which is also a Zn-dependent HDAC.

#### The HDAC4 Gene and Protein

The human HDAC4 gene, located on chromosome 2q37.3, spans approximately 353,480 bp encoding HDAC4 protein with 1084 amino acids. HDAC4 shuttles between cytoplasm and nucleus depending on signal transduction-related phosphorylation status of HDAC4 (Mielcarek et al., 2013). Normally, phosphorylated HDAC4 retains in the cytoplasm, while dephosphorylated HDAC4 is imported into the nucleus (Nishino et al., 2008).

Histone deacetylase 4 protein consists of a long N-terminal domain and a highly conserved C-terminal catalytic domain. The deacetylase activity of HDAC4 is almost undetectable although it has a conserved C-terminal catalytic domain, which might be caused by a substitution of tyrosine to histidine in the enzyme active site (Lahm et al., 2007). However, HDAC4 does play an important role in the regulation of gene transcription via different ways (**Figure 1**). First, HDAC4 interacts with multiple transcriptional factors [e.g., myocyte enhancer 2 (MEF2), runt related transcription factor 2 (Runx2), serum response factor (SRF), heterochromatin protein 1(HP1), nuclear factor kappa B (NF-κB)] regulating gene transcription (Sando et al., 2012; Ronan et al., 2013). Although HDAC4 per se lacks deacetylase activity, it may be involved in histone deacetylation-mediated transcriptional regulation via interacting with HDAC3, another member of the HDAC family with deacetylase activity (Grozinger et al., 1999; Lee et al., 2015). For example, Lee et al. (2015) showed that HDAC4 is crucial for HDAC3-mediated deacetylation of mineralocorticoid receptor, which could be inhibited by class I HDAC inhibitor but not class II HDAC inhibitor, indicating that HDAC4 is implicated in protein deacetylation via the deacetylase activity of HDAC3. Moreover, the deacetylase activity of HDAC4 needs to be further investigated by multiple approaches as it is not convincing by the in vitro assay from one study (Lahm et al., 2007). As the nuclear localization of HDAC4 is regulated by its interaction with14-3-3, it is possible that the alteration of nuclear HDAC4 mediated by tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein (14-3-3) is involved in transcriptional regulation by its deacetylase activity (Nishino et al., 2008). A recent study suggests that HDAC4 may function to regulate protein SUMOylation via interacting with SUMOconjugating enzyme Ubc9 (Ubc9), a SUMO E2-conjugating enzyme, contributing to memory formation (**Figure 1**) (Schwartz et al., 2016).

### HDAC4 IN COGNITIVE FUNCTION AND MOLECULAR MECHANISMS

A growing body of evidence indicates that the homeostasis of HDAC4is crucial for the maintenance of cognitive function by regulating genes involved in synaptic plasticity, neuronal survival and neurodevelopment (**Figure 1**) (Schwartz et al., 2016).

## HDAC4 and Synaptic Plasticity

fnmol-09-00114 October 28, 2016 Time: 12:48 # 3

Histone deacetylase 4 interacts with multiple transcription factors (e.g., MEF2, Runx2, SRF, HP1), 14-3-3, HDAC3 etc. regulating the transcription of genes involved in synaptogenesis, synaptic plasticity and neurodevelopment, such as activity regulated cytoskeleton associated protein (Arc) and protocadherin (Pcdh10) (**Figure 1**) (Grozinger et al., 1999; Sando et al., 2012; Ronan et al., 2013; Rashid et al., 2014; Lee et al., 2015; Sharma et al., 2015; Krishna et al., 2016; Nott et al., 2016). First, nuclear HDAC4 represses the expression of constituents of synapses leading to the impairment of synaptic architecture and strength in mice (Sando et al., 2012). In addition, mice carrying a gain-of-function nuclear HDAC4 mutant exhibit deficits in neurotransmission, learning and memory (Sando et al., 2012). On the other hand, silencing HDAC4 expression does result in the impairment of synaptic plasticity, and learning and memory deficits in both mice and Drosophila (Kim et al., 2012; Fitzsimons et al., 2013). A proteomics analysis indicates that HDAC4 is a regulator of proteins involved in neuronal excitability and synaptic plasticity, which are differentially expressed in normal aging subjects and AD patients and associated with memory status (Neuner et al., 2016). A recent study showed that HDAC4 interacts with Ubc9 during memory formation, while the reduction of Ubc9 in adult brain of Drosophila impairs longterm memory, suggesting that the role of HDAC4 in memory formation may be associated with the regulation of protein SUMOylation (**Figure 1**) (Schwartz et al., 2016).

Above evidence suggests that HDAC4 homeostasis is crucial for the maintenance of synaptic plasticity and cognitive function, i.e., both HDAC4 elevation and reduction lead to cognitive deficits. It is not surprised that both up-regulation and downregulation of HDAC4 impairs synaptic plasticity and memory function as previous studies have demonstrated that a number of molecules play a dual role in synaptic plasticity and memory function. For example, both overexpression and disruption of regulator of calcineurin 1 (RCAN1) leads to synaptic impairment and memory deficits in Drosophila and mice (Chang et al., 2003; Hoeffer et al., 2007; Chang and Min, 2009; Martin et al., 2012). Moreover, the bidirectional alterations of HDAC4 may differentially disrupt the balance between HDAC4 and its interacting partners leading to synaptic impairment and memory deficits as HDAC4 is implicated in multiple signaling pathways by interacting with many functional proteins (**Figure 1**). However, the underlying mechanisms need to be further investigated.

#### HDAC4 and Apoptosis

Neuronal apoptosis is a major mechanism linking to cognitive deficits. In addition to synaptic plasticity, HDAC4 is also involved in neuronal apoptosis. For example, HDAC4 interacts with NF-κB repressing proapoptotic gene expression, and it also inhibit ER stress-induced apoptosis by interacting with activating transcription factor 4 (ATF4), a key transcriptional factor in ER stress response (**Figure 1**) (Zhang et al., 2014; Vallabhapurapu et al., 2015). Moreover, Majdzadeh et al. (2008) showed that HDAC4 overexpression protects mouse cerebellar granule neurons (CGNs) from apoptosis by inhibiting cyclin dependent kinase 1 (CDK1) activity. Consistently, upregulation of HDAC4 by a NMDAR antagonist protects mouse hippocampal neurons from naturally occurring neuronal death, whereas HDAC4 reduction promotes neuronal apoptosis during development (Chen and Cepko, 2009). Sando et al. (2012) further demonstrated that HDAC4-C-terminal is crucial for rescuing HDAC4 knockdown-induced cell death and reduction of synaptic strength in mouse brains. However, Bolger and Yao (2005) showed that increased expression of nuclear-localized HDAC4 promotes neuronal apoptosis in mouse CGNs, while down-regulation of HDAC4 protects neurons from stressinduced apoptosis. The conflicting results may be caused by different cell types and culture conditions. Majdzadeh et al. (2008) cultured CGNs for 4−5 days before transfection, while Bolger and Yao (2005) transfected CGNs immediately after cell isolation. Although CGNs were used in both studies, the maturation status of neurons when they were transfected may have significant effects on the conflict results. Moreover, short-term and longterm protein overexpression may have opposite effects. For example, a previous study showed that RCAN1 plays an opposite role in neuronal apoptosis at different culture stages, which may be associated with aging or maturation processes (Wu and Song, 2013).

#### HDAC4 and Brain Development

Histone deacetylase 4interacts with multiple transcriptional factors, repressing the transcription of genes involved in neurodevelopment (**Figure 1**) (Sando et al., 2012; Ronan et al., 2013). Moreover, HDAC4 may be implicated in neurodevelopment via interacting with HDAC3 which is necessary for brain development (Norwood et al., 2014). In human, both HDAC4 deletion and duplication lead to mental retardation and intelligence disability, suggesting that HDAC4 plays an important role in neurodevelopment which directly links to cognitive function (Shim et al., 2014).

# ABERRANT HDAC4 EXPRESSION/LOCALIZATION IN NEURODEGENERATIVE DISEASES

Cognitive decline, in particular, learning and memory deficits, is a characteristic of neurodegenerative diseases [e.g., Alzheimer's disease (AD), Huntington's disease (HD), and Parkinson's disease (PD)] which are associated with synaptic dysfunction and synaptic and neuronal loss. A large body of evidence indicates that aberrant HDAC4 expression and subcellular distribution may contribute to the cognitive decline in patients with neurodegenerative diseases (**Table 1**). First, increased HDAC4



tHDAC4, total HDAC4; nHDAC4, nuclear HDAC4; cHDAC4, cytoplasmic HDAC4; no, no change; <sup>∗</sup> , mRNA.

expression was observed in prefrontal cortex of aged individuals, while aging is the major risk factor of neurodegenerative disorders (Sharma et al., 2008). In addition, a more recent study showed that HDAC4 is a global regulator of memory deficits with age (Neuner et al., 2016). Moreover, HDAC4 is involved in the regulation of SIRT1 which is implicated in both aging and memory process in rats (Sasaki et al., 2006; Sommer et al., 2006; Quintas et al., 2012; Han et al., 2016). Furthermore, as mentioned in the section 2.1, HDAC4 homeostasis is crucial for the maintenance of cognitive function, i.e., both HDAC4 elevation and reduction lead to cognitive deficits.

#### HDAC4 in Alzheimer's Disease

Alzheimer's disease is the most common form of neurodegenerative disorders in the elderly leading to dementia. Progressive memory loss is the clinical characteristics of AD. Neuritic plaques, neurofibrillary tangles and neuronal loss are the neuropathological hallmarks of AD. Amyloid β (Aβ) and phosphorylated microtubule associated protein tau (Tau) are the major components of neuritic plaques and neurofibrillary tangles, respectively, while apoptosis is a major mechanism of neuronal loss (Wu et al., 2014).

The nuclear expression of HDAC4 is markedly increased in brains of AD patients, while the alteration of total HDAC4, including both cytoplasmic and nuclear HDAC4, is not conclusive (Shen et al., 2016). However, the expression of HDAC4 was significantly increased in AD model mice (Anderson et al., 2015). Moreover, ApoE4, the only confirmed genetic risk factor of late onset AD, increases nuclear HDAC4 levels compared with the ApoE3 in transgenic mice (Sen et al., 2015). It suggests that increased HDAC4 expression or its nuclear localization may contribute to learning and memory deficits in patients with AD.

## HDAC4 in Frontotemporal Lobar Degeneration

Frontotemporal lobar degeneration (FTLD) is a heterogeneous neurodegenerative process resulting in frontotemporal dementia. Progressive difficulties in planning, organizing and language are the major characteristics of FTLD. The atrophy of frontal and temporal lobe and inclusions containing abnormal accumulation of Tau, TAR DNA binding protein (TDP-43) or FUS RNA bindind protein (FUS) are the characteristic pathological features of FTLD.

In FTLD patients, cytoplasmic HDAC4 is increased in granule cells of the dentate gyrus, while HDAC5, the other member of class IIa HDACs, is not altered, suggesting that HDAC4 may have a specific role in the pathology of FTLD (Whitehouse et al., 2015).

#### HDAC4 in Huntington's disease

Huntington's disease is a common autosomal dominant neurodegenerative disease, which is caused by the expansion of polyglutamine repeats in huntingtin (HTT) protein, named as mutant HTT (mHTT) (MacDonald et al., 1993; Myers et al., 1993; Kremer et al., 1994). The characteristic clinical features are chorea, progressive cognitive decline, and psychiatric symptoms, while the cognitive problem is often the earliest symptom in patients with HD (Walker, 2007). mHTT impairs fast axonal transport, disrupts mitochondrial function and inflammatory response and promotes apoptosis, which may contribute to the cognitive decline (Szebenyi et al., 2003; Beal and Ferrante, 2004; Trushina et al., 2004).

Growing evidence suggests that increased HDAC4 is implicated in HD pathology, such that reducing HDAC4 expression has beneficial effects. First, overexpression of miR-22 has a protective effect on mHTT model cells, which may be mediated by HDAC4 reduction as HDAC4 is a target gene of miR-22 (Jovicic et al., 2013). Second, HDAC4 interacts with microtubule associated protein 1S (MAP1S) resulting in MAP1S destabilization and reduction, subsequently suppressing the clearance of mHTT aggregates and potentiating the toxicity of mHTT to cultured cells (Yue et al., 2015). Moreover, HDAC4 is associated with HTT in a polyglutamine-lengthdependent manner and co-localized with cytoplasmic aggregates. However, reducing HDAC4 expression delays the formation of

cytoplasmic aggregates, restores BDNF expression, and rescues synaptic dysfunction in HD mouse models (Mielcarek et al., 2013). In addition, suberoylanilide hydroxamic acid (SAHA) promotes HDAC4 degradation, suggesting that reducing HDAC4 expression may contribute to SAHA's rescue effects on HD model mice via multiple HDAC4-associated pathways (**Figure 1**) (Mielcarek et al., 2011). However, SAHA is also an inhibitor of class I HDACs and HDAC6, suggesting that its rescue effects may also be mediated by inhibiting the deacetylase activity of class I HDACs and HDAC6. Although Quinti et al. (2010) showed that the reduction of HDAC4 is associated with the progression of HD in HD model mice, Mielcarek et al. (2013) did not observe the reduction in same HD model mice. However, they found that reducing HDAC4 expression has beneficial effects on HD mice (Mielcarek et al., 2013). As the alteration of HDAC4 in HD model mice remains inconclusive and it still lacks the evidence from HD patients, the alteration of HDAC4 in HD and its role in the pathology of HD need to be further investigated.

#### HDAC4 in Parkinson Disease

Parkinson disease is the second most common neurodegenerative disease in the elderly. In addition to tremor, rigidity, gait disturbances etc. motor dysfunctions, PD patients also have cognitive impairments (Jankovic, 2008). The major pathological hallmark of PD is the Lewy bodies which mainly consist of protein aggregates of a-synuclein, parkin, and ubiquitin (Jellinger, 2009). In addition, same pathological features were observed in patients with Lewy body (LB) dementia (Jellinger, 2009).

A couple of studies indicate that HDAC4 is associated with the pathology of PD. First, mutations in the Parkin gene cause early onset familial PD and the dysregualtion of parkin has also been observed in sporadic PD. Second, parkin controls the levels of sumoylated HDAC4 (Kirsh et al., 2002; Um et al., 2006). Moreover, HDAC4 co-localized with α-synuclein in the LB (Takahashi-Fujigasaki and Fujigasaki, 2006). In addition, paraquat, a widely used herbicide, implicated in the induction of the pathology of PD, reduces the expression of HDAC4 in culture cells (Song et al., 2011). Furthermore, previous studies showed that aberrant HDAC4 expression results in learning and memory deficits in both mice and Drosophila (Kim et al., 2012; Fitzsimons et al., 2013). Above evidence suggests that alteration of HDAC4 may contribute to cognitive decline in patients with PD. However, no direct evidence shows that HDAC4 is implicated in the pathology of PD.

# HDAC4 in Ataxia-Telangiectasia

Ataxia-telangiectasia (A-T), a rare neurodegenerative disease, is caused by mutations in the ATM gene. A-T patients showed many premature aging components, characterized by difficulty in movement and coordination, and early cognitive impairment including learning and memory deficits (Vinck et al., 2011; Shiloh and Lederman, 2016).

In ATM deficient mice, nuclear HDAC4 is increased, which is mediated by the reduction of ATM-dependent phosphorylation of protein phosphatase 2A (PP2A). Reduced phosphorylation of PP2A results in increased HDAC4 dephosphorylation by enhancing PP2A-HDAC4 interaction (Li et al., 2012). HDAC4 dephosphorylation promotes its nuclear import and subsequent dysregulation of genes involved in synaptic plasticity, neuronal survival and neurodevelopment, which may contribute to the cognitive deficits in ATM mice. Consistently, reduced ATM accompanying with the increase of nuclear HDAC4 has been observed in brains of AD patients (Shen et al., 2016).

# ABERRANT HDAC4 EXPRESSION/FUNCTION IN MENTAL DISORDERS

Many mental disorders, including autism spectrum disorders (ASDs), depression, and schizophrenia, are associated with neurodevelopment defects and cognitive impairment is a core feature of mental disorders (Ronan et al., 2013). Increased evidence indicates that aberrant HDAC4 expression or function plays an important role in cognitive deficits of mental disorders (**Table 1**).

# HDAC4 in ASD and BMDR Syndrome

Autism spectrum disorder is characterized by the impairment of social and communication ability, as well as cognitive defects. Several lines of evidence suggests that dysregulation of HDAC4 is implicated in ASD (Pinto et al., 2014; Fisch et al., 2016). First, HDAC4 mRNA was significantly increased in autistic brains (Nardone et al., 2014). Moreover, ASD, intellectual disability, developmental delay etc. are the characteristics of Brachydactylymental-retardation (BDMR) syndrome which is caused by 2q37 microdeletion. Importantly, the HDAC4 gene is located in this small region (Doherty and Lacbawan, 1993–2016). A rare case of BMDR syndrome carries an inactive mutant of HDAC4, suggesting that HDAC4 deficiency may be the cause of BMDR syndrome (Doherty and Lacbawan, 1993–2016; Williams et al., 2010). Moreover, in patients with BMDR syndrome, HDAC4 modulates the severity of symptoms in a dosage dependent manner, which further confirms the role of HDAC4 in ASD and other BMDR features (Morris et al., 2012).

# HDAC4 in Depressive Disorders

Depressive disorders are the most common mood disorder leading to disability, which is characterized by the presence of sad, empty, or irritable mood and cognitive impairment (Rock et al., 2014). Recent studies highly suggest that HDAC4 is implicated in the pathology of depressive disorders. First, aberrant expression of HDAC4 mRNA has been detected in patients with depression (Otsuki et al., 2012). Consistently, antidepressant reduces the recruitment of HDAC4 to the glial cell-derived neurotrophic factor (GDNF) promoter, consequently increasing the expression of GDNF which is reduced in patients with depression (Otsuki et al., 2012; Lin and Tseng, 2015). In patients with bipolar disorder (BPD), HDAC4 mRNA is significantly increased in a depressive state, while its expression is marked decreased in a remissive state (Hobara et al., 2010). In addition, HDAC4 mRNA is significantly increased in brains of forced-swim stress-induced- and postnatal fluoxetine-induced depression model mice (Sailaja et al., 2012; Sarkar et al., 2014). Intriguingly, adult fluoxetine application does not induce depression-like behavior in mice which is associated with unchanged HDAC4 expression. Ectopic overexpression of HDAC4 in hippocampus is sufficient to induce depression-like behavior in adult mice, indicating that HDAC4 elevation is the key to induce depression-like behavior (Sarkar et al., 2014). Furthermore, depression is a common feature in AD, which may be associated with the increase of HDAC4 expression in AD patients.

#### HDAC4 in Schizophrenia

fnmol-09-00114 October 28, 2016 Time: 12:48 # 6

Schizophrenia is a complex psychiatric disorder, characterized by impairments in behavior, thought, and emotion. Cognitive impairment is common in patients with schizophrenia, in particular, learning and memory deficits. A couple of evidence suggests that HDAC4 might be associated with the pathology of schizophrenia. First, one SNP (rs1063639) in the HDAC4 gene associates with the development of schizophrenia in a Korea population (Kim et al., 2010). Moreover, in patients with schizophrenia, HDAC4 mRNA is negatively associated with the expression of GAD67, a candidate gene of schizophrenia (Sharma et al., 2008). However, the exact role of HDAC4 in the cognitive deficits of schizophrenia needs to be further investigated.

#### HDAC4, A SPECIFIC TARGET FOR COGNITIVE IMPAIRMENT

Growing evidence indicates that HDAC4 is a specific target for the treatment of cognitive impairment in multiple disorders, which is different from other HDACs. First, HDAC4 is highly enriched in brain compared with other HDACs. Second, HDAC4 has no or weak HDAC activity, suggesting that global HDAC inhibitors, targeting the catalytic sites of HDACs, may have no effect on HDAC4's function (Sando et al., 2012). Consistently, HDAC4 has a different effect on cognitive function compared with other HDACs. For example, conditional deletion of HDAC4 leads to learning and memory deficits, while global HDACs inhibition or HDAC2 deficiency significantly improves learning and memory in mice (Vecsey et al., 2007; Guan et al., 2009; Kim et al., 2012). Moreover, the maintenance of HDAC4 homeostasis is crucial for the disease treatment as either increased or

#### REFERENCES


decreased HDAC4 expression is detrimental to the cognitive function. It suggests that HDAC4 is a potential target for the treatment of cognitive impairment. However, only one selective HDAC4 inhibitor, tasquinimod, is commercially available, and its effect on cognitive function has not been explored. Therefore, specific HDAC4 modulators should be developed and their roles in cognitive disorders need to be investigated.

#### CONCLUSION

Although HDAC4 belongs to the family of HDAC, its deacetylase activity is weak or undetectable. Thus, it remains elusive whether HDAC4 per se could repress gene transcription by its HDAC activity (**Figure 1**). However, HDAC4 could regulate the transcription of genes involved in synaptic plasticity, neuronal survival, and neurodevelopment by interacting with multiple proteins, which is essential for the maintenance of normal cognitive function (**Figure 1**). Moreover, HDAC4 may function to regulate protein SUMOylation via interacting with Ubc9 contributing to the maintenance of cognitive function (**Figure 1**). Moreover, aberrant expression of HDAC4 may be implicated in the cognitive impairment of neurodegenerative diseases and mental disorders. Therefore, HDAC4 is a potential therapeutic target to rescue cognitive deficits in above disorders.

#### AUTHOR CONTRIBUTIONS

YW: Formulated the study, wrote the manuscript, and designed the figure. FH, XW: Formulated the study and wrote the manuscript. QK, XH, BB: Provided intellectual thoughts, revised the manuscript, and project leaders.

#### FUNDING

This work was supported by grants from the Natural Science Foundation of Shandong Province (No.ZR2013CM031 to QK), the Jining Science and Technology Program for Public Wellbeing (No.2014kjhm-10 to QK), the Jining Science and Technology Development Plan (No.2013jnwk75 to BB).

Down syndrome. Proc. Natl. Acad. Sci. U.S.A. 106, 17117–17122. doi: 10.1073/pnas.0904397106



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Wu, Hou, Wang, Kong, Han and Bai. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fnmol-09-00114 October 28, 2016 Time: 12:48 # 8

# Changes in the Transcriptome of Human Astrocytes Accompanying Oxidative Stress-Induced Senescence

Elizabeth P. Crowe<sup>1</sup> , Ferit Tuzer<sup>1</sup> , Brian D. Gregory<sup>2</sup> , Greg Donahue<sup>3</sup> , Sager J. Gosai<sup>2</sup> , Justin Cohen<sup>1</sup> , Yuk Y. Leung<sup>4</sup> , Emre Yetkin<sup>1</sup> , Raffaella Nativio<sup>3</sup> , Li-San Wang<sup>4</sup> , Christian Sell<sup>1</sup> , Nancy M. Bonini<sup>2</sup> , Shelley L. Berger<sup>3</sup> , F. Brad Johnson<sup>4</sup> and Claudio Torres<sup>1</sup> \*

<sup>1</sup> Department of Pathology and Laboratory Medicine, Drexel University College of Medicine, Philadelphia, PA, USA, <sup>2</sup> Department of Biology, Penn Genome Frontiers Institute, University of Pennsylvania, Philadelphia, PA, USA, <sup>3</sup> Epigenetics Program, Department of Cell and Developmental Biology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA, <sup>4</sup> Department of Pathology and Laboratory Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA

Aging is a major risk factor for many neurodegenerative disorders. A key feature of aging biology that may underlie these diseases is cellular senescence. Senescent cells accumulate in tissues with age, undergo widespread changes in gene expression, and typically demonstrate altered, pro-inflammatory profiles. Astrocyte senescence has been implicated in neurodegenerative disease, and to better understand senescence-associated changes in astrocytes, we investigated changes in their transcriptome using RNA sequencing. Senescence was induced in human fetal astrocytes by transient oxidative stress. Brain-expressed genes, including those involved in neuronal development and differentiation, were downregulated in senescent astrocytes. Remarkably, several genes indicative of astrocytic responses to injury were also downregulated, including glial fibrillary acidic protein and genes involved in the processing and presentation of antigens by major histocompatibility complex class II proteins, while pro-inflammatory genes were upregulated. Overall, our findings suggest that senescence-related changes in the function of astrocytes may impact the pathogenesis of age-related brain disorders.

Keywords: astrocyte senescence, astrocyte function, brain aging, RNA sequencing, brain oxidative stress

# INTRODUCTION

Astrocytes are the most abundant population of cells within the central nervous system (CNS) and the structural diversity and functional complexity of cortical astrocytes is a distinguishing feature of the primate brain (Oberheim et al., 2006). Astrocytes form a functionally coupled network through a series of gap junctions and have pleiotropic roles in maintaining the blood–brain barrier and controlling cerebral blood flow (Abbott et al., 2006); regulating ion, water and neurotransmitter homeostasis (Simard and Nedergaard, 2004); and modulating synaptic transmission as part of the tripartite synapse (Perea et al., 2009). Astrocytes can respond to CNS insults through the

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

James C. Vickers, University of Tasmania, Australia David Morgan, University of South Florida, USA

\*Correspondence: Claudio Torres claudio.torres@drexelmed.edu

Received: 17 June 2016 Accepted: 15 August 2016 Published: 31 August 2016

#### Citation:

Crowe EP, Tuzer F, Gregory BD, Donahue G, Gosai SJ, Cohen J, Leung YY, Yetkin E, Nativio R, Wang L-S, Sell C, Bonini NM, Berger SL, Johnson FB and Torres C (2016) Changes in the Transcriptome of Human Astrocytes Accompanying Oxidative Stress-Induced Senescence. Front. Aging Neurosci. 8:208. doi: 10.3389/fnagi.2016.00208

acquisition of immune cell features (Jensen et al., 2013), and during repair, astrocytes undergo a spectrum of molecular and functional changes termed reactive astrogliosis (Sofroniew, 2009).

Recently, there has been a paradigm shift toward recognizing the integral role of glial cells in the pathogenesis of age-related cognitive decline and neurodegeneration (Nagelhus et al., 2013; Phatnani and Maniatis, 2015; Pekny et al., 2016). Secreted factors from astrocytes exacerbate the neurotoxicity of amyloid beta (Aβ) in primary culture (Garwood et al., 2011), and contribute to the decline in hippocampal neurogenesis in aged brains (Miranda et al., 2012). Altered astrocyte physiology has also been linked to aging and to the most common age-related neurodegenerative disorder, Alzheimer's disease (AD), by transcriptome profiling of gene expression changes in astrocytes from aged mouse cortex (Orre et al., 2014) and in glial fibrillary acidic protein (GFAP)-positive cells isolated by laser-capture microdissection from postmortem tissues of subjects with AD (Simpson et al., 2011; Sekar et al., 2015). Therefore, a greater understanding of how aging impacts astrocytes should provide new insight into age-related diseases of the brain.

Aging is the greatest risk factor for cognitive decline and neurodegenerative disease, and a key feature of aging biology that may underlie age-related diseases is cellular senescence. In support of this idea, senescent cells accumulate in tissues with age, including the brain, and at sites of aging-related pathology (Price et al., 2002; Krishnamurthy et al., 2004; Herbig et al., 2006; Bhat et al., 2012; Jurk et al., 2012, 2014; Zhu et al., 2014), undergo widespread changes in gene expression, and demonstrate a pro-inflammatory secretion pattern (Coppé et al., 2008). The induction of senescence in astrocytes has been implicated in neurodegenerative disease (Bhat et al., 2012; Chinta et al., 2013). Either a cell-intrinsic loss of function or the acquisition of detrimental neuroinflammatory function in astrocytes could have profound consequences for the aging CNS. While senescenceassociated gene expression changes have been described in celltypes from the periphery that were senescent in situ or induced to senesce in vitro (Shelton et al., 1999; Gruber et al., 2010), they remain largely understudied in the context of the CNS.

Treatment with sublethal concentrations of hydrogen peroxide (H2O2) induces senescence in a variety of cell types (Chen et al., 1998; Kim et al., 2011). Our previous studies characterized this type of stress-induced senescence in human astrocytes as determined by changes in cell morphology (enlarged and flattened shape), cessation of division, increased senescence-associated β-galactosidase activity (85% of positive cells compared to 5% of controls), increased expression of p53 and the cyclin-dependent kinase inhibitors p21 and p16INK4a , and a p38MAPK-dependent increase in interleukin-6 secretion (Bitto et al., 2010; Bhat et al., 2012). Astrocytes are sensitive to oxidative stress and low doses of H2O<sup>2</sup> are enough to induce the senescence program compared to other cell types (Bitto et al., 2010; Bhat et al., 2012; Aravinthan et al., 2014). This is physiologically relevant because the CNS is particularly exposed to elevated levels of oxidative stress due to several factors including a high metabolic rate with an elevated oxygen consumption compared to its relatively small weight, low antioxidant capacity, and high concentration of lipids and pro-oxidant metals. The generation of this robust oxidative environment disturbs cells and results in oxidative damage to macromolecules, which is a common underlying feature of both aging and diseased brains (Smith et al., 1991; Esiri, 2007; Radak et al., 2011). Levels of mitochondrial H2O<sup>2</sup> and defects in protective mechanisms that reduce it are implicated in cognitive defects in AD mouse models and also in inflammation (Yin et al., 2016).

In order to better understand how astrocyte senescence relates to changes in astrocyte physiology during aging, we investigated global changes in the astrocyte transcriptome using RNA Sequencing (RNA-Seq) following the induction of oxidative stress-induced senescence using H2O2. From this analysis, we confirmed that senescent astrocytes acquire an inflammatory phenotype indicative of the senescence-associated secretory phenotype (SASP) and downregulate the expression of brainexpressed genes. In keeping with the myriad of complex functions that astrocytes perform in the healthy brain, senescent astrocytes could affect tissue dysfunction during aging and neurodegenerative disease via multiple mechanisms.

#### MATERIALS AND METHODS

#### Cell Culture and Senescence Induction

Human fetal astrocytes (passage 1) were obtained from ScienCell Research Laboratories (Carlsbad, CA, USA) and cultured in ambient O<sup>2</sup> and 5% CO<sup>2</sup> as previously described (Bitto et al., 2010; Bhat et al., 2012). In order to induce premature senescence via oxidative stress, cells were seeded at standard density (1 × 10<sup>4</sup> cells/cm<sup>2</sup> ) and the following day treated with 200 µM hydrogen peroxide (H2O2) for 2 h. Cells were considered senescent at least 5 days after the initiation of treatment, as verified previously, (Bitto et al., 2010) and in subsequent quantitative real-time PCR (qRT-PCR) experiments by increases in senescence marker p21, flattened and enlarged morphology, and cessation of division, and were harvested 7 days after treatment. Viability of senescent astrocytes was not significantly different than the controls (92% ± 1 vs. 95% ± 2.7; p = 0.08) as measured by the Guava ViaCount assay (EMD Millipore).

#### RNA Preparation and Sequencing

Total RNA was isolated using the RNeasy Mini Kit (Qiagen; Valencia, CA, USA) according to the manufacturer's instructions and the concentration was determined using a NanoDrop ND-1000 spectrophotometer (NanoDrop; Rockland, DE, USA). RNA-Seq libraries were prepared as previously described (Elliott et al., 2013). RNA-Seq libraries were prepared from two replicate cDNA libraries per condition. We used two biological replicates from one donor, where a biological replicate is defined as an independent growth of cells and subsequent analysis, based on the "Standards, Guidelines and Best Practices for RNA-Seq" published by The ENCODE consortium<sup>1</sup> recommending

<sup>1</sup> genome.ucsc.edu/ENCODE/protocols/dataStandards/ENCODE\_RNAseq\_ Standards\_V1.0.pdf

the use of a minimum of two biological replicates in RNA-Seq experiments, where a biological replicate is defined as an independent growth of cells and subsequent analysis. The two replicate cDNA libraries per condition (four libraries in total) were submitted to the Next Generation Sequencing Core (NGSC) at the Perelman School of Medicine, University of Pennsylvania, for sequencing. The Illumina HiSeq sequencing platform was used to generate 50 bp single-end sequencing reads. Analysis of RNA-Seq data, including read mapping and differential gene expression analysis using the DESeq package with a Benjamini–Hochberg correction, was performed as previously described by Elliott et al. (2013). The RNA-Seq dataset was deposited in the Gene Expression Omnibus (GEO) at the National Center for Biotechnology under the accession number GSE58910.

# Gene Ontology and Gene Set Intersection Analysis

Gene Ontology (GO) analysis was performed on transcripts that were significantly differentially expressed in senescent astrocytes with a greater than 1.5-fold change and a p-value ≤ 0.05 (Benjamini–Hochberg adjusted). The functional annotation clustering tool of the online bioinformatics resource Database for Annotation, Visualization and Integrated Discovery (DAVID) version 6.7 (Huang da et al., 2009) was used to perform GO analysis limited to biological process terms (BP\_FAT). By satisfying a false discovery rate (FDR) of 10%, GO terms were considered to be enriched. In addition, all GO categories have a gene count of 10 or greater and a fold-enrichment of 2 or greater. Enriched GO biological process terms were collapsed if they shared 25 or more differentially expressed transcripts and thus considered functionally synonymous. Enrichment (N, B, n, b) is defined as (b/n)/(B/N) where N is the total number of transcripts in the experiment, B is the total number of transcripts within a GO term, n is the number of total transcripts in the intersection of the two gene sets in comparison, b is the number of transcripts in the intersection that belong to that GO term. "Count" is the number of genes in the single or collapsed GO term. "Weight" is the summed weight of all genes in the GO term, where each gene is given a weight inversely proportional to the total number of GO terms it appears in. "FDR," i.e., false discovery rate is the percent likelihood of that GO term coming up by the same number of random genes by chance, as calculated by Benjamini–Yekutieli correction of the p-value obtained by Fisher's Exact test. "Genes" represents the genes constituting the GO term. GO analysis on the intersection of senescence and AD downregulated transcripts was done similarly. The enrichment and statistical significance of gene set overlaps between astrocyte and hepatocyte senescence was performed on http://nemates.org/MA/progs/overlap\_stats.html, with the number of detected transcripts from the RNA-Seq, 19580, as the total number of genes. In comparison of gene expression in senescent astrocytes to those from AD patient brains, we only compared genes with expression levels in our control set matching those in the Stanford Brain database (Zhang et al., 2014). This criterion was expression level of greater than or less than 100 in both datasets. For this comparison, "Percentage" is "Count" as a percentage of the total number of input genes. P-value is a modified Fisher's exact p-value corrected for the representation of the gene set in the whole genome. List total is the total number of genes in the input that are part of any ontology.

## Identification of Transcription Factor Motifs on Differentially Expressed Genes

The chromosomal coordinates of all promoter regions 1000 bp upstream of the transcription start site were obtained using the RefSeq genes track, refGene table and the hg19 human genome assembly on UCSC Genome Browser – Table Browser tool<sup>2</sup> . For all genes up- or downregulated 1.5 fold or more which enriched GO categories, promoter coordinates were submitted to the Cistrome Analysis Pipeline<sup>3</sup> SeqPos tool. Public motif databases Transfac and JASPAR were searched for motifs enriched in the promoter sequences. Additionally, a de novo motif analysis was performed to find motifs with no correlate in the public databases. Results were filtered by human and mouse speciesspecificity, using a 1000 bp scan length.

#### qRT-PCR Validation

Candidate genes were chosen based upon pathways of interest for validation by qRT-PCR. Total RNA was independently isolated as described. Primers were designed using the PrimerQuest design tool to span an exon–exon junction and were supplied by Integrated DNA Technologies (IDT, Coralville, IA, USA). The NCBI Basic Local Alignment Search Tool (BLAST) was used to confirm the specificity of primer sequences. The primers used in qRT-PCR assays are listed in Supplementary Table S1. SYBR Green-based RT-PCR was performed with Verso 1-Step RT-qPCR reagents (Thermo Fisher Scientific; Pittsburgh, PA, USA) on an Applied Biosystems 7500 Real-Time PCR System (Life Technologies, Grand Island, NY, USA). Dissociation curve analysis was performed to verify single products for each reaction. The absence of product in reactions without reverse transcriptase (no RT) was also verified. Data analysis was performed using DataAssist software v3.01 (Life Technologies, Grand Island, NY, USA). The data were glyceraldehyde-3-phosphate dehydrogenase (GAPDH)-normalized and expressed as fold change (RQ) relative to pre-senescent astrocytes.

#### Cell Cycle Analysis

Pre-senescent astrocytes (60–70% confluent) and astrocytes treated with H2O<sup>2</sup> to undergo stress-induced premature senescence were grown in complete Astrocyte Medium (AM, ScienCell) as described (Bitto et al., 2010; Bhat et al., 2012), harvested by trypsinization, washed in phosphate buffered saline (PBS), and fixed with ice cold 70% ethanol overnight at 4◦C. Fixed cells were centrifuged to remove ethanol, washed with PBS, and stained with Guava Cell Cycle reagent (EMD Millipore; Billerica,

<sup>2</sup>https://genome.ucsc.edu/

<sup>3</sup>http://cistrome.org/ap/root

MA, USA) containing the nuclear DNA stain propidium iodide (PI) for 30 min at room temperature in the dark. Guava Cell Cycle data were acquired using Guava EasyCyte Mini flow cytometer using the Guava Cell Cycle program (Guava Technologies, Hayward, CA, USA). The percentage of cells in cell debris, G1-, S-, and G2/M-phase of the cell cycle was determined using the ModFit LT curve fitting algorithm, version 4.0.5 (Verity Software House, Topsham, ME, USA).

### Bromodeoxyuridine (BrdU) Incorporation Assay

Pre-senescent astrocytes in log phase of growth and astrocytes that were treated with H2O<sup>2</sup> to undergo stress-induced premature senescence 7 days prior were treated with 10 µM BrdU (5-bromo-2<sup>0</sup> -deoxyuridine; BD Pharmingen; San Diego, CA, USA) in complete astrocyte medium for 30 min. After this incubation, cells were harvested by trypsinization, washed in PBS, and fixed with ice cold 70% ethanol. Fixed cells were centrifuged to remove ethanol, resuspended in 2N HCl and incubated for 30 min at room temperature for DNA denaturation, neutralized with 0.1 M Na2B4O<sup>7</sup> (pH 8.5), and washed two times in PBS containing 5% fetal bovine serum (FBS). Anti-BrdU monoclonal antibody (eBioscience; San Diego, CA, USA) diluted 1:100 in PBS containing 0.5% Tween-20 was applied for 30 min at room temperature, after which cells were washed, and resuspended in goat anti-mouse-Alexa Fluor 488 (Molecular Probes, Life Technologies; Grand Island, NY, USA) diluted 1:100 in 1x PBS containing 0.5% Tween-20 for 20 min at room temperature in the dark and then washed twice with PBS-5% FBS. Cells were stained with Guava Cell Cycle solution as described previously and analyzed using the Guava EasyCyte Mini flow cytometer using the Guava ExpressPlus program and the percent of cells labeled with BrdU was quantified. The percent of BrdU positive cells was quantified using FlowJo software v10 (Tree Star; Ashland, OR, USA).

#### Immunofluorescence

Cells were seeded on coverslips and fixed with 4% paraformaldehyde in PBS, permeabilized with PBS containing 0.1% Triton-X-100, and blocked in PBS containing 0.1% bovine serum albumin (BSA) and 5% normal donkey serum for 2 h at room temperature. Coverslips were incubated with rabbit anti-phosphorylated histone H3 (Ser10) (Upstate Biotechnology; Lake Placid, NY, USA) diluted 1:500 in PBS containing 0.1% BSA overnight at room temperature. Following washes with PBS, coverslips were incubated with Alexa-Fluor Donkey 555 anti-Rabbit (Life Technologies; Carlsbad, CA, USA) diluted 1:500 in PBS 0.1% BSA for 1 h at room temperature protected from light. Coverslips were then washed, stained with DAPI, and mounted on slides with Vectashield fluorescence mounting medium (Vector Laboratories; Burlingame, CA, USA). Cells were visualized using an Olympus BX61 fluorescence microscope coupled with a Hamamatsu ORCA-ER camera and using SlideBook software (Intelligent Innovations, Inc., Denver, CO, USA). The percent of cells positive for phosphorylated histone H3 was quantified.

# RESULTS

# RNA-Seq Broad Picture of Differentially-Expressed (DE) Genes

We sequenced two biological replicates each of pre-senescent astrocytes and astrocytes induced to senesce by oxidative stress, and obtained approximately 12 to 35 million reads per sample. From these datasets, ∼97.5% of all reads mapped to the reference human genome (hg 19; see Supplementary Table S2 for all mapped transcripts). We also found by a principle component analysis (PCA) that these samples were tightly clustered based on cellular treatment (**Figure 1A**). Expression levels of genes between the pre-senescent and senescent repeats were highly correlated, with r 2 -values of 0.986 and 0.998, respectively. In total, these results suggest that these high-throughput sequencing libraries were highly reproducible and were differentiated from one another based on the biological differences of pre- and post-senescent astrocytes.

We then performed a differential expression analysis that revealed significant senescence-associated changes in the transcriptome. Overall, there were 3569 significantly differentially expressed transcripts (padj < 0.05), which represents 18.3% of the total number of detected transcripts, with 1772 transcripts being downregulated in senescence and 1797 transcripts upregulated in senescence (**Figure 1B**). These results demonstrate that there are significant changes to the astrocyte transcriptome during oxidative stress-induced senescence.

## Gene Ontology Term Enrichment Analysis and Tissue Expression of DE Genes

To identify functional categories of differentially expressed transcripts in senescent astrocytes, we performed GO enrichment analysis using biological process terms with the functional annotation-clustering tool in the DAVID, using a cut-off of 1.5-fold differential expression to define up- or downregulated genes. 1510 downregulated and 1258 upregulated transcripts satisfied this criterion with padj < 0.05, (Supplementary Table S2). Genes involved in cell division, major histocompatibility complex (MHC) class II antigen processing and presentation, metabolism, and CNS development and differentiation were enriched among the downregulated transcripts in senescent astrocytes (**Figures 2A,B**; Supplementary Table S3). MHC Class II presentation and gliogenesis were the two most enriched non-cell division related processes formed by the senescence downregulated transcripts, with enrichment scores of 4 and 3.5, respectively. Among the upregulated gene GO categories, several have known associations with senescence (**Figures 2C,D**; Supplementary Table S4). Inflammation, modification of the extracellular matrix and resistance to apoptosis are known senescence-associated changes (Yoon et al., 2004; Hampel et al., 2006; Freund et al., 2010; Childs et al., 2014) and are represented by the GO terms regulation of I-kappaB kinase/NF-kappaB cascade, positive regulation of cytokine production, extracellular structure organization, vasculature development, and resistance to apoptosis. Cellular adhesion (regulation of cell adhesion,

normalized counts" are the raw read counts, normalized by the total library size, and averaged for each group.

regulation of cell motion, positive regulation of binding) and cytoskeleton (actin cytoskeleton organization) related genes were also previously seen to be upregulated in in vitro senescence of human dermal fibroblasts (Yoon et al., 2004). The upregulation of inflammatory genes suggests a mechanism by which astrocyte senescence may be causing further damage in the brain.

In order to identify possible regulators of the senescenceassociated genes, we analyzed the promoter regions of differentially regulated genes for over-represented transcription factor binding motifs. GO categories formed by downregulated genes yielded a total of 13 motifs, while those formed by upregulated genes yielded only 1 motif, that for p53, formed

FIGURE 2 | Gene Ontology (GO) term enrichment analysis for senescent astrocyte down- and upregulated transcripts. (A) The top overrepresented GO biological process classes in astrocyte downregulated transcripts (padj < 0.05 and fold change of 1.5 or greater; Supplementary Table S3) are shown in the pie chart. The weighted number of genes in each category corresponds to the width of the wedges in the pie chart. (B) GO terms that are not cell division related are expanded below in a bar graph. Developmental categories are coded in gray, immune function is in green, sterol metabolism in blue, and development of the nervous system is in black. (C) The top overrepresented GO biological process classes by fold enrichment within astrocyte upregulated transcripts (padj < 0.05 and fold change of 1.5 or greater; Supplementary Table S4). (D) All GO terms overrepresented by the senescence upregulated genes, ranked by fold enrichment, color coded as in (C). \*Full names of GO terms: antigen processing and presentation of peptide or polysaccharide antigen via major histocompatibility complex (MHC) class II, regulation of Wnt receptor signaling pathway, morphogenesis of a branching structure, the combined term of "cell projection morphogenesis, cell morphogenesis involved in differentiation, neuron differentiation," regulation of smooth muscle cell proliferation, embryonic skeletal system development, regulation of I-kappaB kinase/NF-kappaB cascade, positive regulation of cytokine production, positive regulation of protein kinase cascade.

by the genes in the 'extracellular structure organization' GO category (Supplementary Table S5).

To determine whether genes with differential expression in oxidative stress-induced astrocyte senescence are brainexpressed, we analyzed differentially expressed transcripts (padj < 0.05) that also had a ≥1.5-fold change, for tissue expression using the UniProt tissue expression database (DAVID:UP\_TISSUE). Of all the non-exclusive expression sites found (FDR <10%) for downregulated transcripts, genes belonging to CNS sites comprise the vast majority (762 transcripts or 94.8%), with expressed tissue definition of brain and hippocampus (**Figure 3A**). Therefore, upon the induction of senescence in astrocytes, we see a loss of brain-expressed genes. In contrast, none of the genes upregulated in senescence was CNS-enriched (not shown). CNS enrichment for all detected transcripts was 27%, which is the ratio of the total size of the CNS expression gene sets (7858) to all the defined tissue expression gene sets (29348, FDR <10%, **Figure 3B**). Thus, H2O<sup>2</sup> induced astrocyte senescence specifically downregulates genes that are CNS-enriched.

#### Validation of RNA-Seq by qRT-PCR Astrocyte-Enriched Genes

To determine whether our astrocyte transcriptome data is similar to previously published astrocyte gene expression data, we compared our list of differentially expressed transcripts with celltype specific markers of astrocytes as described in a previous microarray study (Cahoy et al., 2008). The expression levels

of selected astrocyte-enriched genes GFAP, S100B, ALDH1L1, FGFR3, CNS enriched Synapse Differentiation Induced Gene 1 (SynDIG1) and a non-CNS enriched gene, KLF3 were validated by qRT-PCR. The expression of astrocyte and CNS enriched genes was lost or diminished in senescent astrocytes, while that of KLF3 did not change (**Figure 4B**). We verified the association of this decrease in GFAP with senescence by measuring the levels of this protein in pre-senescent astrocytes [cumulative population doubling (cPD) 8.1] and astrocytes that reached replicative senescence, as verified by cessation of growth (cPD 12.2). The GFAP expression in pre-senescent astrocytes was greatly reduced in replicative senescence, confirming our findings with oxidative stress-induced senescence (Supplementary Figure S1). When comparing the log2-fold changes in transcript levels between pre-senescent and senescent astrocytes using qRT-PCR and RNA-Seq, we observed a significant positive correlation (r <sup>2</sup> = 0.656, n = 17) between the results from the two distinct methodologies (**Figure 4A**; Supplementary Table S6). Thus, the loss of astrocyte-enriched genes, combined with the GO analysis demonstrating reduced expression of genes involved in glial and neuronal development suggests loss of normal function in these cells upon undergoing oxidative stress-induced senescence.

#### Senescence-Enriched Genes

We validated by qRT-PCR the levels of senescence-related transcripts that were differentially expressed by RNA-Seq. The levels of senescence-related transcripts CCND1 (Cyclin D1), IL8, IGFBP5, and ICAM-1 were significantly increased (**Figure 4C**) and correlated with changes observed with RNA-Seq (**Figure 4A**; Supplementary Table S6).

Treatment with H2O<sup>2</sup> to induce senescence robustly induces p21 expression in human diploid fibroblasts (Chen et al., 1998). Surprisingly, the expression of CDKN1A, which encodes for the cyclin-dependent kinase inhibitor p21, was not called as significantly differentially expressed in our dataset, although a trend toward increased expression in senescent astrocytes was apparent (RNA-Seq, fold change = 5.84, padj = 0.11). One potential reason for this is low levels of read coverage for this transcript; therefore, we determined the mRNA expression level of p21 using qRT-PCR (**Figure 4C**). We confirmed an almost fourfold increase in p21 mRNA in senescent astrocytes compared with pre-senescent controls.

### Cell Cycle Analysis

Gene Ontology analysis revealed that genes involved in cell cycle, cell division, and mitosis were over-represented among the downregulated genes in senescent astrocytes consistent with the lost proliferative potential of these cells ("cell division" category, **Figure 2A**). In order to examine the cell cycle distribution of senescent astrocytes, cells were stained for DNA content 7 days after H2O<sup>2</sup> treatment and flow cytometric analysis was performed. Pre-senescent astrocytes that were serum-starved for 24 h arrested predominantly in G0/G1, while in senescent astrocyte cultures, we observed an increase in the fraction of cells with 4N DNA content and a concomitant loss of cells in G0/G1 compared with pre-senescent controls cultured in complete growth medium (**Figures 5A,B**).

The proliferative arrest associated with the onset of cellular senescence has often been presumed to occur solely in G1; however, replicatively senescent cells retain the capacity to synthesize DNA under certain conditions and accumulate in both G1 and G2/M (Mao et al., 2012). A multi-phase cell cycle arrest is also a feature of many cell types exposed to oxidative stress and DNA damage (Baus et al., 2003; Oyama et al., 2011). In order to address the possibility that senescent cells with G2 DNA content are progressing to mitosis, we stained astrocytes for phosphorylated histone H3 (Ser10), which is a marker of mitotic chromosome condensation (Hendzel et al., 1997). Compared with pre-senescent controls, H2O2 treated astrocytes exhibited few phospho-H3-positive cells (**Figure 5D**). In pre-senescent and senescent astrocyte cultures, we observed a similar proportion of cells with DNA content between 2N and 4N; therefore, we pulsed the cells with BrdU to determine whether they were actively synthesizing DNA. The BrdU-positive population was significantly reduced in senescent astrocytes compared with pre-senescent controls (**Figure 5C**). Overall, these results support a multi-phase cell cycle arrest in H2O2-induced senescence in human astrocytes.

#### DISCUSSION

In order to better understand how astrocyte senescence is linked to aging-related decline in cognition and neurodegeneration, an unbiased interrogation of the changes that occur at the molecular level is essential. Here, we report a comprehensive analysis of the astrocyte transcriptome following the induction of senescence by oxidative stress. Although gene expression changes have been profiled extensively in brain tissue homogenates from different brain regions during aging (Wood et al., 2013) and in Alzheimer's disease (Twine et al., 2011), fewer studies have addressed cell-type specific changes in these contexts (Simpson et al., 2011; Orre et al., 2014; Sekar et al., 2015). To our knowledge, this is the first report of senescence-associated gene expression changes in a CNSderived cell type using a whole transcriptome sequencing method (RNA-Seq), which is an accurate and quantitative measurement of transcript abundance.

As expected from the cessation of cell cycle in senescence, the majority of genes downregulated in astrocyte senescence following oxidative stress were related to the cell cycle. Several upregulated genes were also related to senescence-associated phenotypes, such as chronic inflammation (comprising NFkB activation and cytokine production), extracellular remodeling, and changes in cell morphology (actin cytoskeleton organization).

growth medium using one-way ANOVA followed by Bonferroni post hoc testing. (C) Pre-senescent and senescent astrocytes were stained for DNA content and BrdU incorporation. The BrdU-positive population was significantly reduced in senescent astrocytes compared with pre-senescent controls <sup>∗</sup>p < 0.01, Student's t-test. (D) Representative images of immunofluorescence staining for mitosis marker phosphorylated histone H3 (pH3) (red) and DAPI (blue), with percent of cells staining positive for pH3 shown in the bar graph. <sup>∗</sup>p < 0.01, student's t-test.

We found that oxidative stress-induced astrocyte senescence is accompanied by a loss of brain-expressed transcripts involved in neuronal and glial differentiation and development, axonogenesis, and axon guidance. These results are supported by studies of in vitro aging in astrocytes where prolonged culture of astrocytes results in a decline in their functional properties including a loss of neuroprotective capacity (Pertusa et al., 2007); and in impaired synaptic transmission in co-culture with neurons (Kawano et al., 2012). The loss of differentiated function upon senescence is also a feature of human ocular keratocytes (Kipling et al., 2009).

The expression of classical markers of astrocyte reactivity glial fibrillary acidic protein (GFAP) and S100β— is downregulated with oxidative stress-induced astrocyte senescence in our study. Interestingly, this finding correlates with recent transcriptome analyses showing a decrease in GFAP expression in astrocytes isolated from the brains of aged mice (Orre et al., 2014) and in aged rat cortical tissue homogenates (Wood et al., 2013). Although aging in astrocytes has traditionally been synonymous with an increase in GFAP expression (Pertusa et al., 2007), recent studies have highlighted the heterogeneity of astrocyte expression of stereotypical markers, including GFAP and S100β, in different brain regions during aging (Rodríguez et al., 2014). Furthermore, the response of astrocytes to different CNS insults, in a process termed reactive astrogliosis, is also more heterogeneous than was once thought (Anderson et al., 2014). Although astrocyte senescence shares some features of reactive astrogliosis including cell hypertrophy and the production of inflammatory mediators, whether astrocyte senescence and reactive astrogliosis are distinct phenomena or part of a continuum of changes will require a more comprehensive analysis of these two phenotypes. It is possible that downregulation of certain markers of astrogliosis helps limit the damaging effects of gliosis, or, alternatively, the downregulation may reflect an inability of senescent astrocytes to respond properly to injury. The upregulation of several cytokines and pro-inflammatory genes, on the other hand, suggests that while astrocyte function is decreased in oxidative stress-induced senescence, the cells may be inducing a more general proinflammatory environment. The upregulation of Golgi vesicle transport related genes in senescence (**Figure 2**) suggests an increase in the rate of vesicle secretion, which, together with the above categories, would contribute to the SASP. Ablation of reactive astrocytes with upregulated GFAP and vimentin expression, or deletion of these proteins in knockout models have resulted in increased neurodegeneration and immune cell infiltration in models of spinal cord injury and infantile neuronal ceroid lipofuscinosis, respectively (Faulkner et al., 2004; Macauley et al., 2011), supporting a protective role for reactive astrocytes. The decreased GFAP and S100β expression seen in senescent astrocytes may be a contributing factor to neurodegenerative conditions that arise with age.

Astrocyte senescence may be downregulating certain astrocyte immune functions, and in this sense it would be different from astrogliosis. Senescence induced by oxidative stress in astrocytes downregulates the expression of genes involved in antigen processing and presentation on MHC class II proteins. These results are in concordance with a recent RNA-Seq dataset from rat cerebral cortex, which demonstrates a significant downregulation of MHC class II genes (Cd74, RT1-ba, RT1-Da, and RT1-Db1) during aging (Wood et al., 2013). Human homologs of these genes were also downregulated significantly in oxidative stressinduced astrocyte senescence (Supplementary Figure S2). In contrast, mRNA levels of MHC class II genes are elevated in the rat hippocampus with normal aging, suggesting regional differences (Frank et al., 2006). MHC class I and II genes are upregulated in astrocytes isolated from aged mouse cortex (Orre et al., 2014), however, this trend is reversed for MHC class II in the microglial population, suggesting that overall gene expression changes seen in whole brain regions may not be representative of every cell type. In the human brain, a decrease in both GFAP and MHC class II receptors was also observed by immunostaining in the temporal cortex of aged AD subjects (>80 years) compared with younger AD subjects (<80 years; Hoozemans et al., 2010). Furthermore, SNPs in the MHC class II region have been strongly associated with AD in a recent metaanalysis of GWAS studies (Alperovitch et al., 2013), suggesting potentially important functional links to AD pathology.

Although human astrocytes undergo inducible expression of MHC class II antigens, their role as functional antigen presenting cells is controversial (Jensen et al., 2013); therefore, the functional significance of a loss of MHC class II gene expression in senescent astrocytes is unclear. In professional antigen-presenting cells, activation of p38MAPK has been shown to negatively regulate CIITA, the master regulator of MHC class II gene expression (Yao et al., 2006). Because p38 MAPK activation is a key pathway driving senescence, this suggests a possible convergence between the senescence program (Iwasa et al., 2003; Bhat et al., 2012) and dysregulation of immune function during aging or immunosenescence. Consistent with this idea, inducible MHC class II expression is impaired during aging in murine macrophages (Herrero et al., 2001).

There are also parallels between gene expression changes in astrocytes with AD and our senescence RNA-Seq data, as expected from the increase of senescent astrocytes in AD brain (Bhat et al., 2012). We compared senescence gene expression changes in vitro to those in astrocytes captured by laser capture microdissection from brains of deceased subjects with early or late stage AD, as analyzed by microarray in a previously published study (Simpson et al., 2011). Thirty-one genes showed a decrease greater than 1.5-fold in both astrocyte senescence in vitro and in astrocytes in AD (Supplementary Table S7). Seven GO terms were significantly represented (FDR < 10%) by the genes downregulated in senescence and AD, out of which four were related to development of non-CNS organs. The remaining three GO categories were neuron development, cell–cell signaling, and neuron differentiation (Supplementary Table S8). The fold changes for the genes in these GO categories are shown in Supplementary Figure S3.

We thus observe that genes involved in generation and differentiation of neural cell types were commonly downregulated in astrocytes in oxidative stress induced senescence and in Alzheimer's disease. Among these genes, the neurotrophic tyrosine kinase 2 receptor (NTRK2) gene codes for the tyrosine kinase B receptor (TrkB). TrkB's primary ligand is brain-derived

neurotrophic factor (BDNF) and its phosphorylation activates pathways involved in neuronal survival, growth, differentiation, transmission, and synaptic plasticity (Boulle et al., 2012). Expression of NRTK2 was also lower in neurons from the anterior cingulate cortex of brains from patients with autism spectrum disorder (Chandley et al., 2015). Another gene with known CNS function that is represented in these GO terms is FGF9. Knockdown of FGF9 downregulates astrogenesis in the developing rat brain and when added to ex vivo cultures, FGF9 upregulates this process (Falcone et al., 2015). FGF9 conditional knockdown caused movement and growth defects in mice, with defects in Bergmann glia formation and Purkinje cell alignment possibly due to a lack of signals from the Bergmann glia. Moreover, extracellular FGF9 was shown to be necessary for glia to form radial morphology. (Lin et al., 2009). Other genes related to neural regeneration and development that are not included in these GO terms were also commonly downregulated between oxidative stressinduced astrocyte senescence and AD (Supplementary Table S7). One such gene, teneurin transmembrane protein 4 (TENM4), encodes for the teneurin-4 (Ten-4) transmembrane protein. An insertion into this gene was responsible for tremors in mice and caused defects in myelination of small diameter axons. The cause was shown to be inhibited oligodendrocyte differentiation, growth and process formation, due to defective FAK signaling by Ten-4 (Suzuki et al., 2012). Ten-4 overexpression and knockdown experiments have shown that this protein is necessary for filopodia formation and neurite outgrowth in neurons via FAK and N-WASP signaling (Suzuki et al., 2014). An intronic variant of this gene was also significantly overrepresented in genomes of bipolar disorder patients (Witt et al., 2014). The protein product (γ-1-syntrophin) of another gene downregulated in senescence and AD, SNTG1, binds and localizes the neurotrophic peptide γ–enolase to the plasma membrane and neurite growth cones of neuroblastoma cells. Knockdown of γ-1-syntrophin disrupts this localization and inhibits the neurite outgrowth and cell proliferation induced by exogenous γ–enolase peptide (Hafner et al., 2010; Falcone et al., 2015). Furthermore, numerous observations of senescence markers in mammalian development may explain the abundance of GO terms related to development of other tissues in the senescence up- and downregulated genes (Meisler and Paigen, 1972; Barral et al., 2014). These gene classes are also an important part of the total down regulated transcriptome in oxidative stress-induced astrocyte senescence. These findings suggest that senescence may be contributing to AD through slowing down of regeneration and differentiation of astrocytes and neurons. Changes in neurogenesis rates were indeed observed in multiple animal and in vitro models of AD (Winner and Winkler, 2015).

We define the transcriptional response of human astrocytes to H2O<sup>2</sup> induced senescence, which has unique characteristics compared to that of other cell types. Whereas H2O<sup>2</sup> induced senescence led to three times as many upregulated genes as downregulated genes in a human hepatocyte cell line (Aravinthan et al., 2014), the number of downregulated genes was slightly higher for astrocyte senescence (Supplementary Figure S4). There were significantly more genes regulated in the same direction by senescence in both cell types than would be expected by chance, however, the differentially regulated gene sets from the two cell types are clearly distinct. These findings suggest cell-type specific responses to oxidative stress induced senescence, with shared mechanisms.

Aging is a major risk factor for chronic diseases in a host of organ systems. The clearance of senescent cells alleviates several signs of pathology associated with aging (Baker et al., 2011, 2016), suggesting that the presence of senescent cells may be deleterious for tissue and organism homeostasis. There is now strong evidence that senescent cells accumulate in tissues, including brain, during aging and in the setting of pathology. We propose that oxidative stress-induced astrocyte senescence is a model for understanding how the basic processes of aging may lead to a decline in cognition and neurodegeneration, and for identification of potential targets for therapeutic intervention.

#### AUTHOR CONTRIBUTIONS

Conceived and designed experiments EC, FT, BG, GD, SG, CS, FJ, and CT. Perform the experiments EC, FT, BG, SG, and CS. Analyzed the data EC, FT, BG, GD, SG, YL, EY, JC, RN, L-SW, NB, SB, FJ, and CT. Contributed reagents/materials/analysis tools BG, GD, SG, YL, EY, JC, RN, L-SW, CS, NB, SB, FJ, and CT. Wrote the manuscript EC, FT, BG, GD, YL, FJ, and CT.

### FUNDING

Research reported in this publication was supported by grants NIH/NINDS 1RO1NS078283, NIH/NIA F30AG043307, and NIH/NIA R21AG046943.

#### ACKNOWLEDGMENTS

The authors thank Dr. Gregg Johannes for providing assistance with qRT-PCR assays. We would like to thank Drs. Elizabeth Powell and Katharine Irvine for their guidance about hepatocyte datasets and generosity in sharing their data.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnagi. 2016.00208

#### REFERENCES

fnagi-08-00208 August 29, 2016 Time: 12:37 # 12


preserve function after spinal cord injury. J. Neurosci. 24, 2143–2155. doi: 10.1523/JNEUROSCI.3547-03.2004



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Crowe, Tuzer, Gregory, Donahue, Gosai, Cohen, Leung, Yetkin, Nativio, Wang, Sell, Bonini, Berger, Johnson and Torres. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# An Insight into the Increasing Role of LncRNAs in the Pathogenesis of Gliomas

#### Yuanliang Yan1,2† , Zhijie Xu<sup>3</sup>† , Zhi Li<sup>4</sup> , Lunquan Sun<sup>4</sup> and Zhicheng Gong1,2 \*

<sup>1</sup> Department of Pharmacy, Xiangya Hospital, Central South University, Changsha, China, <sup>2</sup> Institute of Hospital Pharmacy, Central South University, Changsha, China, <sup>3</sup> Department of Pathology, Xiangya Hospital, Central South University, Changsha, China, <sup>4</sup> Center for Molecular Medicine, Xiangya Hospital, Key Laboratory of Molecular Radiation Oncology of Hunan Province, Central South University, Changsha, China

Long non-coding RNAs (LncRNAs) are essential epigenetic regulators with critical roles in tumor initiation and malignant progression. However, the roles and mechanisms of aberrantly expressed lncRNAs in the pathogenesis of gliomas are not fully understood. With the development of deep sequencing analyses, an extensive amount of functional non-coding RNAs has been discovered in glioma tissues and cell lines. Additionally, the contributions of several lncRNAs, such as Hox transcript antisense intergenic RNA, H19 and Colorectal neoplasia differentially expressed, previously reported to be involved in other pathogenesis and processes to the oncogenesis of glioblastoma are currently addressed. Thus, lncRNAs detected in tumor tissues could serve as candidate diagnostic biomarkers and therapeutic targets for gliomas. To understand the potential function of lncRNAs in gliomas, in this review, we briefly describe the profile of lncRNAs in human glioma research and therapy. Then, we discuss the individual lncRNA that has been under intensive investigation in glioma research, and the focus is its mechanism and clinical implication.

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Sushmita Jha, Indian Institute of Technology Jodhpur, India Yingjie Sun, Harvard University, USA Amul J. Sakharkar, Savitribai Phule Pune University, India

#### \*Correspondence:

Zhicheng Gong gongzhicheng2013@163.com

†These authors have contributed equally to this work.

Received: 05 July 2016 Accepted: 15 February 2017 Published: 28 February 2017

#### Citation:

Yan Y, Xu Z, Li Z, Sun L and Gong Z (2017) An Insight into the Increasing Role of LncRNAs in the Pathogenesis of Gliomas. Front. Mol. Neurosci. 10:53. doi: 10.3389/fnmol.2017.00053 Keywords: glioma, lncRNAs, diagnostic biomarkers, therapeutic targets, therapy

# INTRODUCTION

Glioblastoma (GBM) is the most common primary intracranial tumor, with varying malignancy grades and histological subtypes. Although relatively rare in occurrence, GBM frequently causes mortality and morbidity (Ostrom et al., 2014; Bian et al., 2015), and its median survival time is only 12–14 months after initial diagnosis (Stetson et al., 2016). The current standard therapy for GBM is concomitant radiochemotherapy following maximal surgical tumor resection. However, aggressive growth and recurrence frequently follows after the optimal treatment (Penaranda Fajardo et al., 2016). It is conceivable that complicated signaling pathways and related molecular events underlie the development of gliomas. Consequently, investigations exploring the accurate molecular mechanisms and reliable therapeutic targets for GBM have drawn extensive attention and provided a hopeful prospect for GBM treatment (Kitambi et al., 2014; Furnari et al., 2015).

**Abbreviations:** ADAMTS, a disintegrin and metalloproteinase with thrombospondin motif; BTB, blood-tumor barrier; CASC2, cancer susceptibility candidate 2; CRNDE, colorectal neoplasia differentially expressed; CSCs, cancer stem cells; GAS5, growth arrest-specific 5; GBM, glioblastoma; HOTAIR, Hox transcript antisense intergenic RNA; HULC, highly up-regulated in liver cancer; lncRNAs, long non-coding RNAs; NEAT1/2, nuclear enriched abundant transcript <sup>1</sup>/2; TMZ, temozolomide; TUG1, taurine up-regulated gene 1; XIST, X-inactive specific transcript.

Recently, epigenetic regulation has also drawn remarkable attention, particularly in terms of lncRNAs, which are indispensable for the regulation of cellular processes. LncRNAs are transcripts of more than 200 nucleotides without functional protein-coding ability in a conventional way (Quinn and Chang, 2016). Intriguingly, their coding and translation potential have been reported; they may act as a repository for the synthesis of small polypeptides with interesting biological activity (Cohen, 2014; Ruiz-Orera et al., 2014). LncRNAs can be grouped into five non-exclusive categories according to their genomic location. The subcellular localization is a good indication of the putative function of a lncRNA (Schmitz et al., 2016) (**Figure 1**). For the past three decades, lncRNAs have been found to regulate gene expression during both biological and pathological processes (Fatica and Bozzoni, 2014). For instance, lncRNAs can work as cellular "address codes," which allows protein complexes to be transferred to the appropriate locations on chromosomes and results in consequent activation or deactivation (Batista and Chang, 2013). Mechanistically, in contrast to small interfering RNAs (siRNAs) and microRNAs, lncRNAs can fold into higher order structures to provide much greater potential for target recognition, which facilitates chromatin remodeling as well as transcriptional and post-transcriptional regulation (Mercer and Mattick, 2013; Sahu et al., 2015).

In accordance with their significant roles in normal biological processes, lncRNAs have been implicated in the oncogenesis of gliomas and are increasingly being considered potential therapeutic targets (Ma et al., 2016; Schmitt and Chang, 2016). For example, the well-studied HOTAIR, a lncRNA highly expressed in breast cancer that participates mainly in the chromatin remodeling process, was found to be associated with the biogenesis, development and differentiation of gliomas (Bian et al., 2016). Furthermore, some newly discovered lncRNAs have been found in glioma tissue and cell lines, such as lncRNA ASLNC22381and KIAA0495 (Trojan et al., 2003; Zhang X.Q. et al., 2013, 2015). Through investigating the lncRNAs in tissue specimens for their expression stability in human gliomas and normal brain, Kraus et al. (2015) identified four lncRNAs (HOXA6as, H19 upstream conserved 1 and 2, Zfhx2as and BC200) with stable expression levels in gliomas compared with normal brain. Collectively, these lncRNAs have gained value for clinical purposes as novel biomarkers, but despite this great potential, many issues remain in this rapidly growing field. Here, we summarize the most up-to-date findings regarding how lncRNAs are regulated at the molecular level and their implications in the areas of glioma research and therapy.

#### PROFILE OF LNCRNAS IN HUMAN GLIOMA RESEARCH AND THERAPY

Recent studies in the large-scale analyses of full-length cDNA sequences have discovered many lncRNAs as key players of cell differentiation, immune responses, tumorigenesis, and other biological processes (Wakamatsu et al., 2009; Fujimoto et al., 2016; Wang J. et al., 2016). The Cancer Genome Atlas (TCGA), an ambitious and successful cancer genomics project, generates large-scale multi-dimensional genomic datasets covering over 20 malignancies, providing valuable insights into the underlying genetic and genomic alteration of cancer (Wang Z. et al., 2016). Deep sequencing studies, including large consortia, such as TCGA, have identified numerous tumor-specific mutations not only in protein-coding sequences, but also in noncoding sequences, which have proven to be an important component hidden in the "dark matter" of the genome. These cancer-associated mutations within non-coding RNA, including lncRNAs, can affect gene regulation in the pathogenesis and development of gliomas (Ramos et al., 2016; Diederichs et al., 2016). Differentially expressed lncRNAs in gliomas have been widely analyzed using human glioma tissues and cell lines (**Table 1**). These studies indicated that abnormal lncRNAs plays critical roles in the development and progression of gliomas.

The lncRNA profile in clinical specimens reveals their potential roles in GBM pathogenesis. Using microarrays to analyze the tissues of GBM patients and age-matched normal donors, Han et al. (2012) found the lncRNA expression profile in GBM tissue is significantly altered. In GBM tissue, 654 lncRNAs are up-regulated (fold change ≥4.0), and 654 are down-regulated (fold change ≤0.25). Among the up-regulated lncRNAs, ASLNC22381 and ASLNC2081 are likely to serve as the key elements in the regulation of glioma signaling pathways. Target gene-related pathway analysis indicated that ASLNC22381 and ASLNC20819 may play important roles via their target insulin-like growth factor 1 (IGF-1) genes, which has been thought to be a positive risk factor for human glioma development (Rohrmann et al., 2011). In addition, applying the Affymetrix HG-U133 Plus 2.0 array, Zhang X. et al. (2012) revealed that in tumors relative to normal brain tissues, lncRNA C21orf131-B, MEG3, and RFPL1S are down-regulated, while HOTAIRM1 (HOX antisense intergenic RNA myeloid 1) and CRNDE are comparably up-regulated. Of note, these lncRNA expression patterns show a close correlation with malignancy grade and histological differentiation in human gliomas (Zhang X. et al., 2012). The same group later identified a set of six lncRNAs in 107 GBM patients, including KIAA0495, PART1, MGC21881, MIAT, GAS5, and PAR5, that are significantly associated with overall survival. The prognostic value of this six-lncRNA signature is independent of the methylation status of O-6-methylguanine-DNA methyltransferase (MGMT) promoter, which can promote the treatment resistance of glioma cells to alkylating agent chemotherapy (Zhang X.Q. et al., 2013; Wick et al., 2014). Moreover, based on the lncRNA expression profiles, Li R. et al. (2014) identified three novel molecular subtypes (named LncR1, LncR2 and LncR3) in gliomas. Survival analysis indicated that the LncR1 subtype has the poorest prognosis, while the LncR3 subtype shows the best overall survival rate (Li R. et al., 2014). Another study on lncRNA and mRNA interactions revealed that lncRNAs, such as Hox cluster-associated lncRNAs, can modulate a list of genes participating in the pathogenesis of GBM (Yan et al., 2015). In addition, the expression profiles analysis in recurrent gliomas compared with primary gliomas identified abundant differentially expressed lncRNAs, such as

H19, CRNDE, and HOTAIRM1. These results imply that the future studies of specific expressed lncRNAs would help elucidate the mechanism of glioma recurrence at the genetic level and identify effective therapeutic targets for glioma patients (Chen et al., 2015).

Additionally, in vitro studies have strongly suggested that the altered expression of lncRNAs during genome mutation or genotoxic stress is involved in multiple neuro-oncological disorder-associated cellular processes. Isocitrate dehydrogenase 1 (IDH1) mutations have been shown to be an important prognostic marker for patients with gliomas (Cai et al., 2016; Wang P.F. et al., 2016). LncRNA profiling between gliomas with or without IDH1 mutations show significantly altered gene expressions in astrocytic and oligodendroglial tumors. Among the differential lncRNAs, KIAA0495, LOC254559 and LOC255130 have a close correlation with clinical outcomes in IDH1-mutant patients. Moreover, these three IDH1 mutation-associated lncRNAs participate in multiple tumor-associated cellular biological behaviors, including cell proliferation, apoptosis and metastasis (Zhang X.Q. et al., 2015). In addition, after treatment with DNA damaging reagents, such as doxorubicin and resveratrol, specific candidate lncRNAs (MEG3, ST7OT1, TUG1, BC200 and MIR155HG) are detected in human glioma cell lines (U251 and U87). During apoptosis induced by both reagents, MEG3 and ST7OT1 are up-regulated in both cell lines. Instead, when necrosis is induced with a high dose of doxorubicin, TUG1, BC200 and MIR155HG are significantly down-regulated (Liu Q. et al., 2015). As NEAT2 (nuclear-enriched abundant transcript 2), also known as MALAT1 (metastasis-associated lung adenocarcinoma transcript 1), is a highly conserved lncRNA associated with the metastatic potential of tumor cells, Han et al. (2016a) found that the knockdown of NEAT2 by RNA interference could promote the invasion and proliferation of glioma cells. Concomitantly, the apoptosis rate of the glioma cell lines is shown to dramatically increase (Han et al., 2016a; Xiang et al., 2016). Over all, these results indicate that an investigation into the abnormal expression profiles of lncRNAs may help in the understanding of oncogenesis and identify novel potential treatment targets in glioma research and therapy.

Accumulating evidence indicates that a rare population of self-renewing cells, called tumorigenic CSCs, is responsible for tumor formation and therapeutic resistance in gliomas (Lathia et al., 2015). Studies have indicated that lncRNAs are involved in several biological processes in CSCs (Li Y. et al., 2015). A large-scale expression study of functional ultra-conserved (uc) ncRNAs showed that the uc.283 lncRNA, a 277 nucleotide-long sequence located at ultra-conserved regions (UCRs) of human genes, is highly specific for pluripotent stem cells, as well as some solid cancers, particularly gliomas (Galasso et al., 2014). Moreover, Han et al. (2016b) found that the down-regulation of NEAT2 suppresses the expression of stemness markers Sox2 and Nestin, and further promotes cell proliferation by regulating the ERK/MAPK (extracellular signal-regulated kinase/mitogenactivated protein kinase) signaling axis in the glioma stem cell line SHG139. Furthermore, the knockdown of the lncRNA XIST could exert tumor-suppressive effects in human GBM stem cells by up-regulating miR-152 (Yao et al., 2015). In addition, as the gene enhancer of zeste homolog 2 (EZH2) serves as an oncogene and is required for cancer stem cell maintenance, the inhibition of EZH2 by lncRNAs can effectively promote the therapeutic sensitivity in gliomas (van Vlerken et al., 2013; Yin et al., 2016). Based on these observations, much more attention should be paid on the regulation of lncRNAs in the maintenance of glioma stem cells (GSCs), a decisive event occurring in the development of gliomas.

### ABERRANTLY EXPRESSED LNCRNAS AND THEIR IMPLICATIONS IN HUMAN GLIOMAS

The differential expression patterns of lncRNAs between tumor and normal tissues, along with the expression discrepancies in tumors with different clinical features, provide the possibility that lncRNAs act as diagnostic, prognostic biomarkers and pharmaceutical targets in gliomas. Although an increasing


TABLE1|Longnon-codingRNAs(lncRNAs)expressionprofileingliomatissuesandcell

DOX, doxorubicin; Rsv, resveratrol; NA, not available.

number of lncRNAs are being characterized, their detailed mechanisms are still not completely elucidated. In this regard, recent studies have demonstrated that lncRNAs in gliomas can serve as molecular decoys, which move proteins or RNAs away from a specific location, like a "sponge" to miRNAs (e.g., HOTAIR/miR-326, CASC2/miR-21, XIST/miR-152, and Gas5/miR-222). Additionally, other investigations demonstrate lncRNAs can function as molecular signaling mediators, which modulate the expression of a certain set of genes (e.g., H19/CD133 and NEAT2/MMP2) (**Figure 2**). To provide an exhaustive description of the rapid development in this field, the molecular mechanisms and potential functions of several representative lncRNAs in gliomas will be discussed in the following sections.

#### LncRNA HOTAIR

Long non-coding RNA HOTAIR, transcribed from the antisense strand of the homeobox C (HOXC) gene locus in chromosome 12, is involved in the regulation of specific gene transcription. A study by Tsai et al. (2010) demonstrated that HOTAIR regulates gene expression by interacting with polycomb repressive complex 2 (PRC2) and lysine-specific demethylase 1A (LSD1). The 5<sup>0</sup> - and 3<sup>0</sup> -domains of HOTAIR can bind to the PRC2 and LSD1/CoREST/REST complex, respectively. Serving as a scaffold, HOTAIR can tether two distinct complexes together and recruit specific histone modification enzymes, thereby resulting in H3K27 methylation and H3K4 demethylation and ultimately gene silencing (Tsai et al., 2010). In addition, HOTAIR could also serve as an inducer of ubiquitin-mediated proteolysis to control protein levels. HOTAIR facilitates the ubiquitination of Ataxin-1 through E3 ubiquitin ligases Dzip3, Snurportin-1, and Mex3b to further accelerate their degradation. Through the rapid decay of targets Ataxin-1 and Snurportin-1, HOTAIR can prevent cellular premature senescence (Yoon et al., 2013). In addition, aberrant HOTAIR expression has been extensively revealed to correlate with cancer metastasis and is characterized as a negative prognostic factor for cancer patients (Cai et al., 2014; Wu et al., 2014).

Hox transcript antisense intergenic RNA expression is upregulated in glioma tissues and cell lines, and can serve as a potential biomarker or therapeutic target for human gliomas (Kiang et al., 2015; Zhou et al., 2015) (**Figure 3**). Recent studies have indicated that HOTAIR expression is a critical regulator of cell cycle progression in gliomas (Zhang J.X. et al., 2013). HOTAIR regulates cell cycle progression predominantly via the HOTAIR 5<sup>0</sup> -domain-PRC2 axis, which is EZH2 (predominant PRC2 complex component)-dependent in GBM cells (Zhang K. et al., 2015). In addition, bromodomain and extraterminal (BET) domain proteins are required for GBM cell proliferation. BET protein inhibitors can reduce the proliferation of gliomas, in part, through the induction of the cyclin-dependent kinase inhibitor p21Cip1 in vitro and in vivo (Pastori et al., 2014). Pastori et al. (2014) found that the bromodomain protein BRD4 could directly control HOTAIR expression by binding to its promoter. The overexpression of HOTAIR in conjunction with the BET

protein inhibitor I-BET151 abolishes the anti-proliferative activity of the BET bromodomain inhibitor (Pastori et al., 2015). Furthermore, the HOTAIR-miRNA axis has an important role in malignant biological behaviors of human glioma. Ke et al. (2015) found that fibroblast growth factor 1 (FGF1) mediates oncogenic effects by activating the PI3K/AKT and MEK 1/2 pathways. HOTAIR, one target of miR-326, has been confirmed to down-regulatemiR-326; then, it exerts its tumor-suppressive activities by reducing the expression of FGF1 (Ke et al., 2015). Similarly, HOTAIR can act as an endogenous "sponge" of miR-141, thereby promoting the promoter methylation of miR-141 by DNA methyltransferase 1 (DNMT1) in glioma cells. Then, the hypermethylated miR-141 can repress the expression of the spindle and kinetochore associated complex subunit 2 (SKA2), which results in a significant increase in tumor growth (Bian et al., 2016). Collectively, these results suggest that HOTAIR may potentiate glioma development in many facets; thus, it is worthy of further investigation.

#### LncRNA H19

Long non-coding RNA H19, produced from the imprinted gene H19, is one of the most highly conserved transcripts involved in mammalian development. Studies have also demonstrated that H19 could potentially serve as an oncogenic lncRNA in different types of cancers, including gliomas (Kiang et al., 2015; Chen et al., 2016). Mechanistically, the product of the

MYC oncogene, c-Myc, induces the expression of the H19 non-coding RNA, thereby potentiating gliomagenesis (Barsyte-Lovejoy et al., 2006). Furthermore, serving as a miRNA precursor, H19 could modulate glioma progression by generating miR-675. The oncogenic function of H19/miR-675 is dependent on the expression of cancer-associated cadherin 13 (CDH13), which is the direct target of miR-675 (Shi et al., 2014). Li C. et al. (2015) found that miRNA-675, which is derived from the first exon of H19, could regulate the immoderate proliferation and migration of glioma cell lines by inhibiting the expression of CDK6, which is a pivotal regulator of the cell cycle and involved in glioma development (Rader et al., 2013; Sherr et al., 2016). These findings agree with another study showing that H19 overexpression can promote the cell-cycle progression of cancer cells (Berteaux et al., 2005). Moreover, the knockdown of H19 by siRNA displays higher therapy efficiency when induced by the chemotherapy drug TMZ in GBM cells (Li W. et al., 2016). Thus, LncRNA H19 could be increasingly recognized as a potential target for glioma treatment.

Accumulating evidence has identified that tumorigenic CSCs, with self-renewing capability, contribute to tumor initiation and therapeutic resistance (Lathia et al., 2015). Intriguingly, H19 overexpression could maintain the stem cell properties of GBM cells. Li W. et al. (2016) found that the markers of CSCs, including CD133, NANOG, Oct4, and Sox2, are significantly down-regulated in H19-deficient cells. This conclusion was further confirmed by Jiang et al. (2016), who found that the increased level of H19 promotes invasion, angiogenesis, and stemness of GBM cells. H19 is significantly overexpressed in CD133-positive GBM cells, and higher H19 expression levels are associated with increased tumor growth (Jiang et al., 2016). In spite of the critical role of H19 in the maintenance of glioma stemness, its exact mechanism is still unclear and needs to be further investigated.

#### LncRNA CRNDE

Colorectal neoplasia differentially expressed was initially identified by Derrien et al. (2012) as a putative non-coding RNA; it is highly expressed in developmental neurobiology and neuropathology. Studies have found that CRNDE expression is also elevated in many colorectal cancers and brain cancers, such as GBM, astroblastomas, and astrocytomas (Ellis et al., 2012; Kiang et al., 2015). Of note, among the 129 lncRNAs differentially expressed in glioma tissues, CRNDE is consistently identified as the most up-regulated lncRNA by 32-fold up (Zhang X. et al., 2012).

Colorectal neoplasia differentially expressed potentiates glioma development possibly by maintaining the stemness of the tumor cells, as it functions in neural precursors (Ellis et al., 2012; Watkins and Sontheimer, 2012). In support of this notion, a previous study by Zheng J. et al. (2015) demonstrated a direct link between the overexpression of CRNDE and GSCs. Mechanistically, CRNDE could negatively regulate miR-186 and depress the expression of the downstream target genes XIAP (X-linked inhibitor of apoptosis) and PAK7 [p21 protein (Cdc42/Rac)-activated kinase 7], thus contributing to the malignant characteristics of human GSCs (Zheng J. et al., 2015). In addition to these observations, Wang Y. et al. (2015) showed that the overexpression of the CRNDE transcript promotes glioma cell growth in vitro and in vivo through mammalian target of rapamycin (mTOR) signaling. Intriguingly, epigenetic modifications, including histone acetylation in the promoter region, can also promote CRNDE expression (Wang Y. et al., 2015). More recently, CRNDE was reported to promote malignant behavior by attenuating the miR-384/PIWIL4 (piwilike RNA-mediated gene silencing 4) axis. Briefly, CRNDE knockdown can decrease the protein level of PIWIL4, a target of miR-384, which leads to glioma regression in vivo (Zheng J. et al., 2016). Overall, these results revealed that CRNDE could potentiate glioma via multiple signaling pathways and may be a promising novel therapeutic target for glioma therapy.

#### LncRNA CASC2

Cancer susceptibility candidate 2, located at chromosome 10q26, is a lncRNA originally identified as a tumor suppressor gene in endometrial cancer. CASC2 consists of three alternatively

spliced transcript isoforms, CASC2a, CASC2b and CASC2c, which contain identical first three exons and diverse downstream exons (Zhao et al., 2014). CASC2a expression is down-regulated at the transcription level in endometrial cancer. Baldinu et al. (2007) revealed that the exogenous expression of CASC2a in undifferentiated endometrial cancer cells significantly inhibits the clonal growth. Using a positional candidate approach, 7% CASC2a mutations in tumor DNA from 44 endometrial cancer patients were identified (Baldinu et al., 2004), suggesting that inactivation of CASC2a might probably be due to mechanisms

different from genetic alterations. In non-small cell lung cancer (NSCLC) tissues and cell lines, He X. et al. (2016)) reported that CASC2 expression is involved in the development and progression of NSCLC. However, little is known about the role and function of CASC2 in human gliomas.

Recently, Wang P. et al. (2015) reported that CASC2 expression is decreased in glioma tissues as well as glioma cell lines (U251 and U87). Consistent with previous studies in other tumors, the overexpression of CASC2 could inhibit the malignancy of glioma cells through an arrest of proliferation and migration, correspondingly promoting cellular apoptosis. RIP and RNA pull-down assays confirmed that the tumor suppressive role of CASC2 is mainly mediated via the down-regulation of miR-21, one potential direct target of CASC2, in a sequencespecific manner (Wang P. et al., 2015). A growing body of literature has shown that miR-21 serves as an oncogene, and the inhibition of miR-21 is a novel therapeutic strategy for specific and effective action against gliomas (Harmalkar et al., 2015; Belter et al., 2016). Mechanistically, miR-21 promotes gliomagenesis by regulating multiple oncogenesis-related processes, including proliferation, apoptosis, migration and invasion. Therefore, targeting the CASC2-miR-21 axis may be an effective strategy for the treatment of malignant gliomas.

### LncRNA XIST

X-chromosome inactivation (XCI) ensures dosage compensation between the sexes in mammals and is a paradigm for allelespecific gene expression on a chromosome-wide scale. The lncRNA XIST, a product of the XIST gene, is located within the 500 kb stretch of XCI DNA at Xq13, which is known as the X-inactivation center (XIC); XIST is the master regulator of X chromosome inactivation in mammals (Furlan and Rougeulle, 2016; Maduro et al., 2016). The current model proposes that XIST induces epigenetic silencing of multiple genes by recruiting the chromatin modifier, the PRC2 complex, to the XIC (Goodrich et al., 2016). With the help of the highaffinity RNA-binding protein ATRX (alpha thalassemia/mental retardation syndrome X-linked), a growing number of XIST RNAs accumulate and are tethered to the X chromosome. Afterward, the XIST RNA spreads and forms a RNA "cloud" coating the XIC in cis. To recruit PRC2, the XIST RNA first associates with approximately 150 intense PRC2 binding sites (CpG islands), followed by its association with 3,000–4,000 moderate-strength binding sites of PRC2. Finally, XIST RNA spreads to both gene-rich and poor regions in distinct stagespecific forms on the X chromosome (Simon et al., 2013; Sarma et al., 2014).

X-inactive specific transcript has been found to be dysregulated in a variety of human cancers (Yildirim et al., 2013; Tantai et al., 2015). Specifically, a recent study showed that XIST expression is abnormally up-regulated in glioma tissues and GSCs. The knockdown of XIST by short-hairpin RNA exerts a tumor suppressive function in GSCs. Furthermore, as XIST and miR-152 may form a reciprocal repression feedback loop and are located in the same RNA induced silencing complex (RISC), miR-152 can mediate the promotion of GSCs by XIST (Yao et al., 2015). In addition, XIST can inhibit hepatoma cell proliferation and metastasis by targeting miR-92b (Zhuang et al., 2016). Moreover, XIST has been identified to directly bind tomiR-210 (Fasanaro et al., 2009). Consistently, other miRNAs, such as miR-92b and miR-210, may also regulate the expression of XIST in gliomas. Altogether, further studies should focus on the XIST-miRNA axis in glioma research and treatment.

#### LncRNA TUG1

Taurine up-regulated gene 1, a 7.1 kb lncRNA located at chromosome 22q12, is a cancer-related lncRNA in some tumors, including NSCLC (Zhang et al., 2014), bladder cancer (Tan et al., 2015) and gliomas (Li J. et al., 2016). TUG1 was first identified in a genomic screen for genes differentially regulated by taurine in developing mouse retinal cells. Furthermore, TUG1 is found to play crucial roles in the formation of photoreceptors and retinal development (Young et al., 2005).

Recent investigations have reported that in human glioma cell lines, TUG1 is down-regulated, in response to necrosis induced by a high dose of DOX (Liu Q. et al., 2015). Li J. et al. (2016) showed that TUG1 acts as a tumor suppressor in glioma tumorigenesis, and is negative correlated with glioma grade, tumor size, and overall survival. Further studies via gainand loss-of-function assays revealed that TUG1 induces glioma cell apoptosis through caspases-mediated intrinsic pathways, rather than the Bcl-2-mediated anti-apoptotic pathway (Li J. et al., 2016). However, the precise mechanism of TUG1 in cell proliferation, as well as invasion, in glioma development is still unclear. The BTB limits the effect of conventional chemotherapy by restricting drug delivery to brain tumor tissues (Hendricks et al., 2015). Using a co-culture assay with glioma and endothelial cells, Cai et al. (2015) revealed that the knockdown of TUG1 could reduce tight junction protein expression in endothelial cells by down-regulating heat shock transcription factor 2 (HSF2), the target of miR-144, increasing BTB permeability of chemotherapeutic agents. Thus, there may be potential role of TUG1 in anti-glioma therapy, and BTB function may represent a useful therapeutic intervention strategy in the future.

#### LncRNA NEAT1/2

Nuclear enriched abundant transcript 1 (NEAT1) is an essential lncRNA for the formation of paraspeckles, which are nuclear bodies named for their close proximity to nuclear speckles (Yu and Shan, 2016). NEAT1 is an unusual RNA polymerase II (pol II) transcript that lacks introns, and it is widely expressed in many types of mammalian cells (Naganuma and Hirose, 2013). NEAT2/MALAT1 is a highly conserved lncRNA associated with

tumorigenesis and plays a prognostic role in various cancers (Wei and Niu, 2015).

Up to date, lncRNAs have been demonstrated to be involved in the DNA damage response, thus contributing to the process of cellular defense against genotoxic agents (Zhang and Peng, 2015). Upon treatment with the DNA damage-inducing agent resveratrol, NEAT1 is up-regulated in the glioma cell lines U251 and U87 (Liu Q. et al., 2015). An increase in NEAT1 expression has also been reported in human glioma tissues compared with non-cancerous brain tissues. NEAT1 promotes glioma pathogenesis by regulating glioma cell proliferation, invasion, and migration. Zhen et al. (2016) demonstrated that functioning as a molecular sponge for miR-449b-5p, NEAT1 could up-regulate the expression of c-Met, a direct target of miR-449b-5p, thus promoting glioma oncogenesis. Furthermore, clinical investigations revealed that aberrant NEAT1 expression is negatively associated with clinical outcome in high-grade glioma patients (He C. et al., 2016).

Recent works have illustrated the tumor-suppressive role of NEAT2 in the development of glioma cells. NEAT2 expression is lower in glioma tissues than in normal brain tissues. Mechanistically, NEAT2 inhibits the proliferation and invasion of glioma cells (U87 and U251) by inactivating ERK/MAPK signaling and down-regulating MMP2 (matrix metalloproteinase 2; Han et al., 2016a). In contrast, Xiang et al. (2016) showed an opposite role of NEAT2 in gliomas. According to their observations, NEAT2 expression is significantly increased in glioma tissues, as well as in U87 and U251 cells (Xiang et al., 2016). Remarkably, GSCs of the U87, SHG44 and SHG139 cell lines expressed higher levels of NEAT2 than their parental lines (Han et al., 2016a). In addition, Han et al. (2016b) found that the down-regulation of NEAT2 suppresses the expression of stemness markers Sox2 and Nestin in SHG139S cells, while NEAT2 down-regulation promotes the proliferation of SHG139S cells. Therefore, NEAT2 plays a complex role in gliomagenesis as both a positive and a negative regulator, possibly based on its specific cellular context.

# LncRNA GAS5

Growth arrest-specific 5, localized at chromosome 1q25.1, could transcribe a tumor-suppressive lncRNA in human cancers. To date, GAS5 has been considered to act as a "riborepressor" or "miRNA sponge" that modulates the transcriptional activity of cancer-associated genes (Kino et al., 2010; Zhang Z. et al., 2013). Recent studies have reported that GAS5 negatively regulates the growth of cancer cell lines in vitro and in vivo, including gliomas (Pickard and Williams, 2015). GAS5 exerts complementary effects on cell proliferation (inhibitory) and apoptosis (stimulatory), and taken together, these cellular mechanisms likely form the basis of its tumor-suppression action (Yin et al., 2014; Shi X. et al., 2015). Mechanistically, the up-regulation of Gas5 increases the expression of tumor suppressor bmf (Bcl-2-modifying factor) and Plexin C1 via directly reducing the expression of miR-222 (Zhao X. et al., 2015). In addition, the overexpression of GAS5 could enhance the cellular response to erlotinib, a tyrosine kinase inhibitor used as a second line treatment for glioma (Garcia-Claver et al., 2013). The induction of GAS5 is apparently detected during DOXinduced apoptosis in human glioma cell lines (Liu Q. et al., 2015). The above examples suggest that GAS5 may be used as diagnostic markers or therapeutic targets for gliomas, but much work needs to be done before such applications become clinically practical.

#### LncRNA ADAMTS9-AS2

The ADAMTS family has been implicated in essential physiological processes, such as angiogenesis and organ development (Ho et al., 2016). ADAMTS9-AS2 is the antisense transcript of ADAMTS9, a member of the ADAMTS family. Walsh et al. (2016) pointed that ADAMTS9-AS2 plays a critical role in epigenetic regulation, affecting early stage digit development. Recently, the ADAMTS9-AS2 locus has been revealed as a potential therapeutic target and prognostic marker in gliomas. ADAMTS9-AS2 serves as a tumor suppressor, which is significantly down-regulated in glioma tissues, and its expression is negatively correlated with tumor grade and prognosis. Meanwhile, DNMT1 knockdown remarkably enhances ADAMTS9-AS2 expression, inhibiting cell migration in gliomas (Yao et al., 2014).

# LncRNA SPRY4-IT1

SPRY4-IT1, a 708 bp intron-retained lncRNA localized at chromosome 5q31.3, is found to be significantly expressed in breast cancer (Shi Y. et al., 2015), osteosarcoma (Ru et al., 2016) and bladder cancer (Zhao X.L. et al., 2015), and its suppression can inhibit proliferation and induce apoptosis in cancer cells. SPRY4-IT1 was originally reported by Khaitan et al. (2011) to play an important role in the molecular etiology, modulation of cell apoptosis and invasion of human melanoma. Recently, the expression of SPRY4-IT1 is shown to be significantly expressed in glioma tissues and glioma cell lines compared with normal donors (Liu H. et al., 2015). The epithelial-to-mesenchymal transition (EMT), as a relevant molecular event in malignant gliomas, is an essential process in tumor dissemination and metastatic behavior (Kahlert et al., 2013). Liu H. et al. (2015) showed that the knockdown of SPRY4-IT1 by siRNA could suppress the EMT phenotype in glioma cells (U251 and SF295). However, the exact mechanism underlying the role of SPRY4-IT1 in glioma pathology still remains to be elucidated.

# LncRNA HULC

Highly up-regulated in liver cancer has pro-oncogenic activity in many human malignancies, such as B-cell lymphoma (Peng et al., 2016), hepatocellular carcinoma (Huang et al., 2016), and osteosarcoma (Sun et al., 2015). Recently, Zhu's et al. (2016) reported that HULC has important biological function in human gliomas. HULC can promote the angiogenesis, one hallmark of malignant gliomas, by inhibiting the expression of angiogenesisrelated molecule ESM-1 (endothelial cell specific molecule 1). In addition, the PI3K/AKT/mTOR signaling pathway is involved in the response induced by HULC (Zhu et al., 2016). These intriguing findings will help pave the way for exciting functional studies of HULC in gliomagenesis.

#### CONCLUSION AND REMARKS

fnmol-10-00053 February 25, 2017 Time: 15:46 # 9

Long non-coding RNA-based mechanisms alter cell fate during development, and their dysregulation underscores many human disorders, including gliomas. LncRNAs play indispensable roles in the onset and progression of this malignancy, including the proliferation, metastasis and EMT of glioma cells. Though previously considered "junk sequences" in our genomes, the epigenetic role of lncRNA should promise to be another exciting marker for glioma research and therapy. In addition, extracellular vesicles (EVs), like exosomes, isolated from blood, cerebrospinal fluid (CSF), and other biofluids of GBM patients could offer new insight into cancer biology with both diagnostic and therapeutic implications. These exosomes have been found to harbor gliomaderived specific lncRNAs that are significantly different in cancer patients compared with normal controls (Chistiakov and Chekhonin, 2014). Moreover, exosome-transmitted lncRNAs could promote chemotherapeutic resistance in cancer by acting as a competing endogenous RNA (ceRNA; Qu et al., 2016). They can act as sponges for competitively binding miRNAs through their miRNA-recognizing elements (MREs) and further regulate the expression of miRNAs (Denzler et al., 2014; Yang et al., 2016). Strikingly, these MRE elements implicated in the ceRNA networks are also able to regulate the mRNA expression playing critical roles in tumorigenesis (Guo et al., 2015). Understanding the key roles of "lncRNA-miRNA" and "lncRNA-mRNA" interactions in the pathogenesis of gliomas will lead to the identification of new targets for GBM treatment.

In addition, TMZ, an alkylating agent, is the most widely used and effective first-line chemotherapeutic drug for treating primary and recurrent high-grade gliomas (Messaoudi et al., 2015). TMZ could activate autophagy in tumor cells. Autophagic modulators could lead to either cell survival or cell death, depending on the cellular context, which further affects the therapeutic sensitivity of TMZ in GBM (Yan et al., 2016). Recently, it has been proposed that serving as factors in gene regulation, lncRNAs could control cellular processes such as autophagy in disease conditions (Choudhry et al., 2016). The oncogene lncRNA HNF1A-AS1 could promote tumor growth by sponging tumor-suppressive hsa-miR-30b-5p in hepatocellular carcinoma. Meanwhile, the HNF1A-AS1-miR-30b axis could significantly up-regulate cell autophagy during starvation by enhancing the expression of ATG5, the target of miR-30b (Liu Z. et al., 2016). However, upon energy stress, lncRNA NBR2 (neighbor of BRCA1 gene 2) could promote AMP-activated protein kinase (AMPK) activity through interacting with AMPK, leading to a depressed autophagy response and increased tumor

#### REFERENCES


development (Liu X. et al., 2016). Thus, further investigation of lncRNAs in autophagy regulation would be able to identify novel strategies to enhance the benefits of TMZ chemosensitivity and chemoprotection in the treatment of gliomas.

In the last decade, lncRNAs have been regarded as molecular targets for the treatment of many cancers, including gliomas (Lavorgna et al., 2016). Furthermore, recent advancements in deep sequencing are now providing new tools to functionally annotate disease-associated lncRNAs, facilitating the identification of these new transcripts for cancer therapy (Huarte, 2015; Zheng L.L. et al., 2016). However, their biological effects are easily influenced by many factors, such as delivery strategies to cross the BTB. A better understanding of the real efficacy and mechanisms of lncRNAs, particularly in human patients, represents a matter of great interest for possible clinical application in future. Ma et al. (2016) found that the knockdown of the lncRNA NEAT2 in gliomas could result in the significantly increased permeability of BTB, which might contribute to enhancing potential therapeutic strategies for human gliomas. Meanwhile, the results from Liu's group indicated that the lncRNA TUG1, which is highly expressed in vascular endothelial cells from glioma tissues, could influence BTB permeability via binding to miR-144, further reducing the expression of tight junction proteins in endothelial cells, such as ZO-1, occludin, and claudin-5 (Cai et al., 2015). Thus, extensive work should focus on the role of lncRNAs in BTB permeability, which may represent a useful therapeutic target for human glioma treatment.

#### AUTHOR CONTRIBUTIONS

YY, ZX and ZL wrote this review article. LS and ZG designed the study and contributed in manuscript preparation.

#### FUNDING

This work was supported by the National Natural Science Foundation of China (No. 81572946), the Changsha Science and Technology Project (No. k1508024-31), and the Clinical and Rehabilitation Research Foundation of Xiangya hospital – Beidaweiming.

#### ACKNOWLEDGMENT

ZX is right now a Postdoctoral Fellow in Department of Pharmacy, Xiangya Hospital, Central South University.

loss at chromosome 10q26 in human endometrial cancer. Hum. Mutat. 23, 318–326. doi: 10.1002/humu.20015

Barsyte-Lovejoy, D., Lau, S. K., Boutros, P. C., Khosravi, F., Jurisica, I., Andrulis, I. L., et al. (2006). The c-Myc oncogene directly induces the H19 noncoding RNA by allele-specific binding to potentiate tumorigenesis. Cancer Res. 66, 5330–5337. doi: 10.1158/0008-5472.CAN-06-0037



genes in the pathogenesis of human glioblastoma multiforme. J. Cancer Res. Clin. Oncol. 141, 827–838. doi: 10.1007/s00432-014-1861-6



via the PI3K/Akt/mTOR signaling pathway in human gliomas. Oncotarget 7, 14429–14440. doi: 10.18632/oncotarget.7418

Zhuang, L. K., Yang, Y. T., Ma, X., Han, B., Wang, Z. S., Zhao, Q. Y., et al. (2016). MicroRNA-92b promotes hepatocellular carcinoma progression by targeting Smad7 and is mediated by long non-coding RNA XIST. Cell Death Dis. 7, e2203. doi: 10.1038/cddis.2016.100

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Yan, Xu, Li, Sun and Gong. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Genomic Editing of Non-Coding RNA Genes with CRISPR/Cas9 Ushers in a Potential Novel Approach to Study and Treat Schizophrenia

Chuanjun Zhuo1,2,3 \* † , Weihong Hou4† , Lirong Hu<sup>1</sup> , Chongguang Lin<sup>1</sup> , Ce Chen<sup>1</sup> and Xiaodong Lin<sup>1</sup>

<sup>1</sup>Department of Psychiatry, Wenzhou Seventh People's Hospital, Wenzhou, China, <sup>2</sup>Department of Psychiatry, Tianjin Mental Health Center, Tianjin Anding Hospital, Tianjin, China, <sup>3</sup>Department of Psychiatry, Tianjin Anning Hospital, Tianjin, China, <sup>4</sup>Department of Biology, University of North Carolina at Charlotte, Charlotte, NC, USA

Schizophrenia is a genetically related mental illness, in which the majority of genetic alterations occur in the non-coding regions of the human genome. In the past decade, a growing number of regulatory non-coding RNAs (ncRNAs) including microRNAs (miRNAs) and long non-coding RNAs (lncRNAs) have been identified to be strongly associated with schizophrenia. However, the studies of these ncRNAs in the pathophysiology of schizophrenia and the reverting of their genetic defects in restoration of the normal phenotype have been hampered by insufficient technology to manipulate these ncRNA genes effectively as well as a lack of appropriate animal models. Most recently, a revolutionary gene editing technology known as Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)/CRISPR-associated nuclease 9 (Cas9; CRISPR/Cas9) has been developed that enable researchers to overcome these challenges. In this review article, we mainly focus on the schizophrenia-related ncRNAs and the use of CRISPR/Cas9-mediated editing on the non-coding regions of the genomic DNA in proving causal relationship between the genetic defects and the pathophysiology of schizophrenia. We subsequently discuss the potential of translating this advanced technology into a clinical therapy for schizophrenia, although the CRISPR/Cas9 technology is currently still in its infancy and immature to put into use in the treatment of diseases. Furthermore, we suggest strategies to accelerate the pace from the bench to the bedside. This review describes the application of the powerful and feasible CRISPR/Cas9 technology to manipulate schizophrenia-associated ncRNA genes. This technology could help researchers tackle this complex health problem and perhaps other genetically related mental disorders due to the overlapping genetic alterations of schizophrenia with other mental illnesses.

Keywords: schizophrenia, non-coding RNAs, lncRNAs, miRNAs, CRISPR/Cas9, gene editing

# INTRODUCTION

Schizophrenia is a severe mental illness, in which patients exhibit both mental and behavioral dysfunction mainly associated with neurodevelopmental abnormality. Indeed, schizophrenia is considered as a neurodevelopmental disorder (Marenco and Weinberger, 2000). To date, the pathological causes underlying the onset and progression of schizophrenia remain unclear.

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Daniela Tropea, Trinity College, Dublin, Ireland Hyunsoo Shawn J. E., Duke NUS Graduate Medical School, Singapore Yingjie Sun, Harvard University, USA

> \*Correspondence: Chuanjun Zhuo chuanjunzhuotjmh@163.com

†These authors have contributed equally to this work.

Received: 07 October 2016 Accepted: 23 January 2017 Published: 03 February 2017

#### Citation:

Zhuo C, Hou W, Hu L, Lin C, Chen C and Lin X (2017) Genomic Editing of Non-Coding RNA Genes with CRISPR/Cas9 Ushers in a Potential Novel Approach to Study and Treat Schizophrenia. Front. Mol. Neurosci. 10:28. doi: 10.3389/fnmol.2017.00028

A wealth of science-based evidence has indicated that schizophrenia is a genetically related mental illness that is inherited in approximately 80% of cases (Cardno et al., 1999) and that certain alterations in the genomic DNA inherited from the parents are likely to contribute to disease onset in late adolescence or early adulthood (International Schizophrenia Consortium et al., 2009; Costain et al., 2013; Human Molecular Genetics). Decades of research have shown that the interplay between genes and environmental factors largely contribute to the development of schizophrenia (Owen et al., 2016). It has been found that the majority of genetic alterations occur in the non-coding regions of human genome, from which regulatory non-coding RNAs (ncRNAs), mainly microRNAs (miRNAs) and long non-coding RNAs (lncRNAs), are transcribed. miRNAs and lncRNAs are two major regulatory ncRNAs with little capacity to encode proteins and have differences in size, location and biological function. miRNAs play a vital role in the modulation of gene expression, mainly via post-transcriptional destabilization of target mRNAs that harbor one or more complementary sites with miRNAs, translational repression, or a combination of the two molecular mechanisms (Bartel, 2004, 2009). While lncRNAs are defined as regulatory ncRNAs with a size longer than 200 nt in length. In contrast to miRNAs, our knowledge with respect to the molecular mechanism whereby lncRNAs function remains largely unexplored (Ponting et al., 2009; Wilusz et al., 2009). Recently, a growing number of lncRNAs have been reported to be associated with a broad range of diseases, including mental illnesses such as schizophrenia (Barry et al., 2014; Merelo et al., 2015; Rao et al., 2015). miRNAs and lncRNAs as two large groups of regulatory ncRNAs are highly expressed in the brain where they act as important modulators of genes essential for maintaining proper brain development and function. This is of particular importance because it has been well documented that abnormal brain development, maturation and plasticity have been strongly linked to schizophrenia. For instance, manipulation of schizophrenia-associated miRNAs such as miR-132/miR-121 and miR-219 has been reported to alter an activity-dependent neuronal maturation and plasticity at synapse (Nudelman et al., 2010; Tognini et al., 2011; Mellios and Sur, 2012). Further studies have shown that these miRNAs function through either directly regulating synthesis of proteins which are essential for plasticity at the synapse or interacting with transcription factors which are likely to modulate more enduring neuroplastic changes (Smalheiser and Lugli, 2009; Mellios and Sur, 2012). These recent findings of schizophreniaassociated miRNAs suggest that the biological implication of dysregulation of the miRNAs and their targeted genes are profound for schizophrenia. Considering the magnitude of miRNA alterations and their broad effects on target genes in schizophrenia, the schizophrenia-associated miRNAs might be important for the pathogenesis of schizophrenia. Over past years, an increasing number of schizophrenia-associated ncRNAs such as miR-137, lncRNA Gomafu, etc., has been identified, and the correlation of their genetic alterations with the pathogenesis of schizophrenia has been characterized (Sone et al., 2007; International Schizophrenia Consortium et al., 2009; Schizophrenia Psychiatric Genome-Wide Association Study (GWAS) Consortium, 2011; Barry et al., 2014; Ma et al., 2014). However, the exact functional roles of these genetic variations in the development and progression of schizophrenia are less understood. Indeed, the functioning of these ncRNAs in the development of schizophrenia and the correction of their genetic changes in the genomic DNA, the key to develop novel therapies for schizophrenia, have been hampered by insufficient technology to manipulate these ncRNA genes effectively and a lack of appropriate animal models. Apparently, understanding these changes in the genomic DNA, including the coding and non-coding regions in schizophrenia, in combination with the development of novel genetic tools, will undoubtedly enable the mutations to be corrected, which is the key to developing novel therapies for schizophrenia.

In recent years, with a range of advanced technologies, including next-generation sequencing, high-resolution microarrays and genotyping, a great number of miRNAs and lncRNAs associated with schizophrenia have been identified and characterized. The understanding of these ncRNAs and their genetic variations in the development of schizophrenia has relied on the novel tools of advanced technologies in biology and beyond. Clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated nuclease 9 (Cas9) is a recently developed revolutionary gene editing technology that is able to manipulate effectively the non-coding regions of the genomic DNA in human cell lines and animal models, to create or correct the mutations more easily than ever in living cells and animals, and therefore to overcome obstacles in schizophrenia research and therapy (Bhaya et al., 2011; Jinek et al., 2012; Cho et al., 2013; Cong et al., 2013; Jiang et al., 2013; Mali et al., 2013; Doudna and Charpentier, 2014).

In this review article, we bring together schizophreniaassociated regulatory ncRNAs, mainly miRNAs and lncRNAs, summarize CRISPR/Cas9-mediated editing on the non-coding regions of genomic DNA in human cell lines and animal models, and propose applications of the revolutionary gene editing CRISPR/Cas9 tool in schizophrenia research, especially on the functional study of ncRNAs genes and on the development of animal models to accelerate schizophrenia research, thus proving the causal relationship between genetic defects and the pathophysiology of schizophrenia, and eventually translating the scientific findings into a clinical cure for schizophrenia. The information described in the present review may help researchers to tackle this complex health problem and perhaps other genetically related mental disorders due to the overlapping genetic alterations of schizophrenia with other mental illnesses.

#### METHODS

A systematic search and review of the recent literature was performed using the main databases, including PubMed, Google Scholar and Wanfang Med Online, by using the following keywords: schizophrenia, mental disorder, gene editing, CRISPR/Cas9, ncRNAs, miRNAs and lncRNAs.

### ncRNAs INCLUDING miRNAs AND lncRNAs THAT ARE ASSOCIATED WITH SCHIZOPHRENIA

The involvement of complex genetic components in the etiology of schizophrenia has been well established by a wealth of scientific evidence. Indeed, the heritability of schizophrenia is approximately 80% (Cardno et al., 1999); therefore, novel genomic tools offer hope for insight into the profound etiology of this complicated, genetically related mental illness. Genome-wide association studies have revealed a great number of strong susceptibility loci for schizophrenia, most of which are located in the non-coding regions of the genome, including genetic fragments for the transcription of miRNAs and lncRNAs. In fact, numerous studies have found that the majority of miRNAs and lncRNAs are highly expressed in the brain, where their genetic mutations, abnormal expression and impact on target gene levels may lead to defective development, suppression of synaptic efficacy, and mGluR-dependent synaptic plasticity in the hippocampus (Olde Loohuis et al., 2015; Siegert et al., 2015) as well as other neurological impairment that is largely correlated with schizophrenia and may contribute to its pathogenesis and progression (Krichevsky et al., 2003; Sempere et al., 2004; Giraldez et al., 2005; Vo et al., 2005; Schratt et al., 2006). In the past decade or two, with the help of high-throughput microarray techniques, comparison of differential expression of miRNAs in schizophrenia vs. controls has identified numerous miRNAs that are dysregulated in schizophrenia, which have been summarized previously (Beveridge and Cairns, 2012; Wang J. et al., 2014). In contrast to the above studies with microarray analysis, a genome-wide association study performed in a large-scale population revealed schizophrenia-associated non-coding genes, including miRNA genes, among which miRNA137 stands out (International Schizophrenia Consortium et al., 2009; Schizophrenia Psychiatric Genome-Wide Association Study (GWAS) Consortium, 2011). International Schizophrenia Consortium et al. (2009) reported that the genetic elements for a small ncRNA, the microRNA137 (MIR137) gene with chromosomal location 1p21.3, have been the schizophrenia-associated loci, conferring susceptibility to schizophrenia and the discovery has received great attention within the field. In the following large-scale genome-wide association study conducted by Franke et al. (2016) using more than 40,000 recruited participants, the rs1625579 polymorphism in the miRNA137 gene was identified to be strongly linked to schizophrenia (Schizophrenia Psychiatric Genome-Wide Association Study (GWAS) Consortium, 2011; Ma et al., 2014). To date, many further investigations have supported the association between miR137 and schizophrenia as well as the miR-137 variant rs1625579 as a predictor of an earlier age-at-onset of psychosis in schizophrenia (Lett et al., 2013). In a Han Chinese population, analysis of the single nucleotide polymorphism rs1625579 of the miRNA137 gene in a cohort of 1430 individuals with schizophrenia compared with 1570 healthy individuals was performed; rs1625579 showed significant differences in the allele frequencies between those who suffered from schizophrenia and healthy control subjects (Guan et al., 2014), while the single nucleotide polymorphism rs66642155 allelic variant was likely to have an impact on the age-at-onset and the degree of positive symptoms (Wang S. et al., 2014). The molecular mechanism whereby the genetic variants of miR-137 confer a high risk of schizophrenia also has been shown to be related to a declined fronto-striatal brain white and gray matter structure but without changes in the volume of the local brain (Rose et al., 2014; Wright et al., 2016), which may cause symptoms such as poor concentration, low processing speed, cognitive impairment, etc. (Kuswanto et al., 2015); the impact of miR137 genetic alterations on the expression of a set of target neuronal transmission genes and synaptogenesis (Kos et al., 2015; Strazisar et al., 2015; Wright et al., 2015); and involvement of miR137 on a glucocorticoid receptor-dependent signaling pathway (Vallès et al., 2014). A very recent in-depth study in a zebrafish model system by Giacomotto et al. (2016) has reported that a slight change in synaptic function or a defect in the axonal network mediated by schizophrenia-associated suppression of miR-137 is likely responsible for the behavioral phenotype.

Apart from miRNAs, lnRNAs have emerged as one of most important classes of ncRNAs in the regulation of gene expression, and their genetic alterations and dysregulation play an important role in the pathogenesis of various diseases. In contrast to a larger number of schizophrenia-associated miRNAs being studied, only several lncRNAs have been identified to be correlated with this disease, among which myocardial infraction-associated transcript (MIAT), also known as retinal ncRNAs 2, was most recently found to be strongly correlated with schizophrenia. MIAT was originally identified as a new member of the lncRNA family in Blackshaw et al. (2004), and it was initially found to play a role in regulating retinal cell fate differentiation and to be associated with myocardial infraction (Ishii et al., 2006; Rapicavoli et al., 2010; Liao et al., 2016). It is also abundantly expressed in the nucleus of neurons during development through adulthood. In addition, MIAT is known as lncRNA Gomafu (Sone et al., 2007), a Japanese word reflecting its speckled distribution in the nucleoplasm. Despite its chromosomal location of 22q12.1, which has been documented to be associated with schizophrenia, and its expression in the nervous system, the link between this lncRNA and schizophrenia was unveiled in 2014. The Australian research team of Barry et al. (2014) has demonstrated for the first time the strong association of MIAT with schizophrenia, including downregulation of Gomafu in schizophrenic patients, and has provided insight into the mechanisms: acutely declined levels of Gomafu in response to neuronal activation and Gomafumediated impaired alternative splicing through its binding directly to the two splicing factors QK1 and serine/argininerich splicing factor 1, which ultimately leads to the abnormal regulation of the two schizophrenia genes DISC-1 and ErbB4 (Barry et al., 2014; Liao et al., 2016), causing a decrease in the activity of parvalbumin interneurons (Chung et al., 2016). Soon after the identification of the association between lncRNA Gomafu and schizophrenia, the link of its genetic alterations with schizophrenia in different populations has been subsequently reported. Rao et al. (2015) have conducted an analysis on the genetic variants of lncRNA Gomafu in 1255 patients diagnosed with paranoid schizophrenia, compared to 1209 healthy individuals in a Chinese Han population, and found that the rs1894720 locus was significantly associated with paranoid schizophrenia (Rao et al., 2015). Aside from lncRNA Gomafu, there are other mental disease-related lncRNAs such as Evf2, BDNF-AS and DISC-2 that have been found to be associated with schizophrenia and have been reviewed previously (Merelo et al., 2015).

As the number of new schizophrenia-associated ncRNAs continues to grow and the variants within the ncRNA genes may directly correlate with schizophrenia, there is an urgent need for novel genetic tools that can be used to understand the functional role of these ncRNAs in schizophrenia and to correct the mutations as a novel approach for schizophrenia research and treatment. CRISPR/Cas9 is such a revolutionary technology that can be applied to work especially for this purpose.

### CRISPR/Cas9 FOR EDITING GENOMIC ELEMENTS TARGETING NON-CODING RNAs

The CRISPR/Cas9 system, originally well-documented as an adaptive immune defense mechanism in bacteria, is an emerging revolutionary yet feasible technique for precise genome editing of various organisms, including plants, animals, and even humans, so that specific changes in the genomic DNA stretches can be easily and precisely made (Bhaya et al., 2011; Jinek et al., 2012; Cho et al., 2013; Cong et al., 2013; Jiang et al., 2013; Mali et al., 2013; Doudna and Charpentier, 2014). The CRISPR/Cas9 system generally consists of two components: the Cas9 protein and the guide RNA (gRNA). Being guided by the gRNA, the Cas9 protein is recruited to the target site and is able to cut the genomic DNA at a specific site. The past years have witnessed the emergence of this innovative technology as well as dramatic progress in genomic DNA editing with the CRISPR/Cas9 system (Doudna and Charpentier, 2014). The technology has been applied to manipulate the genomic DNA elements targeting not only the coding regions but also the non-coding ones, from which small and long coding ncRNAs such as miRNAs and lncRNAs are encoded. It is particularly important to edit the genomic DNA elements targeting non-coding regions, since silencing these ncRNA genes with the current, conventional RNA interference (RNAi) technology has turned out to be challenging due to their resistance to the RNAi techniques (Zamore et al., 2000; Gilbert et al., 2013; Fatica and Bozzoni, 2014; Hilton et al., 2015). Recently, the obstacles have been overcome by the application of a modified CRISPR/Cas9 system, in which a lentiviral vector expressing two gRNAs are used simultaneously, with which the genomic DNA fragments ranging from 100 to 3000 bp in length have been manipulated and edited (Han et al., 2014; Aparicio-Prat et al., 2015). To date, a number of ncRNA genes in the genomic DNA have been successfully silenced, including miRNAs (miR21, miR29a) and lncRNAs (UCA1, MALAT1), with MALAT1 reduced by up to 98% inside human HCT116, Hela, and HEK293T cell lines after the promoter region of the MALAT1 gene was edited by the Double Excision CRISPR Knockout method (Aparicio-Prat et al., 2015; Ho et al., 2015; Chang et al., 2016). More detailed information about the miRNA and lncRNA genes successfully edited by the CRIPSR/Cas9 system in cell lines and entire animals were summarized in **Tables 1**, **2**. This powerful genome-editing tool has brought a wealth of innovative applications and perspectives for both biological research and the treatment of diseases, including the complicated disease of schizophrenia. In the following sections, we offer a review of a number of successful examples of CRISPR/Cas9-mediated editing on the non-coding regions of genomic DNA in human



Abbreviations: UCA1, urothelial cancer associated 1; MALAT1, metastasis associated lung adenocarcinoma transcript 1; NEAT2, noncoding nuclear-enriched abundant transcript 2.

cell lines and animal models as well as comment on the perspectives for both schizophrenia research and an eventual cure for schizophrenia. We also discuss the challenges of translating this advanced technology into a clinical therapy for schizophrenia and suggest strategies for removing the obstacles to accelerate the pace along the path from the bench to the bedside.

#### PERSPECTIVE OF EDITING ncRNA GENES WITH CRISPR/Cas9 FOR SCHIZOPHRENIA RESEARCH

The application of the novel CRISPR/Cas9 technology in schizophrenia-associated ncRNAs will usher in a new perspective for schizophrenia research to advance our understanding about the biological function for ncRNAs and to facilitate creating animal models with specific mutations or with the original form restored. Silencing or enhancing gene expression to lose or gain function has been well accepted in the study of a gene of interest. RNAi with specific small-interfering RNAs (siRNAs) is a commonly used approach to silence a desired gene encoded by the coding regions of genomic DNA, which usually occurs in the cytoplasm (Zamore et al., 2000). However, siRNAs designed to target ncRNA genes, including lncRNAs and miRNA genes, have been found to be inefficient mainly because many lncRNAs are located in the nucleus (Fatica and Bozzoni, 2014). In fact, it has been difficult to achieve successful knockdown of a desired lncRNA gene. Recently, scientists from independent research groups have applied a modified CRISPR/Cas9 system to target the ncRNA genes in the nucleus that has resulted in robust knockdown of a number of ncRNAs, including miRNAs in the zebrafish genome (Xiao et al., 2013), miRNAs in human cell lines, lncRNAs in a mouse model (Han et al., 2014) and lncRNAs in human cell lines (Ho et al., 2015), allowing the study of the biological roles of schizophrenia-associated ncRNAs in the pathogenesis of schizophrenia. The key feature of the modified CRISPR/Cas9 system is the use of dual gRNAs, which produce two cuts in specific sites and allow deletion of a larger fragment (Ho et al., 2015). With the modified CRISPR/Cas9 system, there is the possibility to manipulate any exon fragment of the ncRNAs and to explore the biological function of these ncRNA genes in schizophrenia. Since schizophrenia is a complex disorder that involves multiple genetic alterations of ncRNAs, the modified CRISPR/Cas9 approach will enable the disruption of these ncRNAs and will test if disruption of these ncRNA genes can cause schizophrenia. Apart from the silencing of ncRNA genes, the CRISPR/Cas9 system can also deliver regulatory components to the target genes and activate or upregulate target gene expression. Together with its silencing of the ncRNA gene, activation of the gene at the transcriptional level also empowers researchers to understand the biological role for a gene in the development and progression of schizophrenia. Apparent knockdown or activation of an ncRNA gene using the modified CRISPR/Cas9 system has couple of advantages: it is more effective than RNAi, and it is able to target multiple genes simultaneously. When it comes to the schizophrenia-associated ncRNAs miR137 and Gomafu, it is possible to target these two ncRNAs simultaneously and investigate if miRNA137 and Gomafu each alone or in combination affect the development and progression of schizophrenia.

The main bottleneck of schizophrenia research has been lack of translatable animal model system to prove causal relationship between genetic defects and pathophysiology of schizophrenia owing to not only difficulties in reproducing its prominent symptoms but also the tedious work required for creating animal models with specific genomic mutations. The CRISPR/Cas9 system brings a novel perspective for developing animal models for schizophrenia research due to a number of advantages. With the help of the CRISPR/Cas9 system, specific mutations of the target ncRNA gene can be introduced into the embryo and the normal form of the ncRNA gene can be restored in a rat or mouse embryo. The rat or mouse as well as its offspring will contain the mutation or restoration of the original form, which allows researchers to compare the symptoms directly, to determine how disruption of these genes affects the development and progression of schizophrenia, and to pinpoint the underlying molecular pathway in those animal models. The generated animal models can also be used to test the efficacy of medications or other potential therapeutic approaches in the treatment of schizophrenia. In fact, the animal model generated by traditional procedures usually takes up to 2 years to get the specific mutations in the offspring because multiple breeding steps are needed, while creating the animal model using the CRISPR/Cas9 approach only requires about 2 months and costs less. Furthermore, in combination with template DNAs and using multiple gRNAs, the CRISPR/Cas9 system is able to introduce a number of desired mutations into the embryo of an animal or their offspring, which is not likely to work out with other previous approaches.

#### PERSPECTIVE OF EDITING ncRNA GENES WITH CRISPR/Cas9 FOR THE TREATMENT OF SCHIZOPHRENIA

Medications have been the cornerstone therapy for schizophrenia, and medications in combination with psychosocial interventions are widely used to manage schizophrenia. However, medications for schizophrenia usually cause serious adverse side effects; therefore, a large portion of patients with schizophrenia fail to take medications, leading to this devastating mental disease being uncontrolled. As a strong genetic component is involved in the pathogenesis of schizophrenia, multiple alterations in the genomic DNA of neurons have been implicated as causative factors, and fortunately great progress has been recently made in both identifying critical genomic regions and developing advanced genetic technologies, multiple alterations may be manipulated in the genomic DNA of neurons. As we discussed earlier, the genetic mutations in both miRNA and lncRNA genes have been directly linked to schizophrenia. The CRISPR/Cas9 system has provided a valuable tool to correct mutations not only in inheritable genetic diseases but also gene mutations in DNA genomes related to diseases, including ncRNAs genes. The CRISPR/Cas9 system, together with a donor DNA template, is required and delivered together with the CRISPR/Cas9 system into the recipients, and it has been successfully used to replace the mutations with substitutions in cells, plants, and entire animals to correct certain genomic DNA mutations. For example, a research group from Duke University has recently explored the application of the CRISPR/Cas9 system to treat Duchenne muscular dystrophy (DMD), a genetic disease with a debilitating mutation within one of the exons of the dystrophin gene (Nelson C. E. et al., 2016). For the first time, researchers have successfully treated a human disease in a living mouse model (Nelson C. E. et al., 2016) with the CRISPR/Cas9 gene editing technology. Meanwhile, similar results from two other research groups from Harvard University and the University of Texas are exciting as well (Long et al., 2016; Tabebordbar et al., 2016). These three independent research groups have demonstrated that correction of the dystropin gene, the consequential restoration of functional dystrophin, and the enhancement of muscle strength can be achieved after one of the exons in the dystrophin gene was reverted with the CRISPR/Cas9 technique by using different methods to deliver the CRISPR/Cas9 components (Long et al., 2016; Nelson C. E. et al., 2016; Tabebordbar et al., 2016).

In addition to DMD, a variety of other genetic diseases such as sickle-cell anemia and Alzheimer's disease have been treated with the CRISPR/Cas9 technique by correcting the causative mutations and reverting the defect to the original form in the genomic DNA, and functional restoration has been gained (Huang et al., 2015; Sankaran and Weiss, 2015; Paquet et al., 2016). Recently, a research team from Sun-Yatsen University in Guangzhou, China, has explored, for the first time, the utilization of this novel CRISPR/Cas9 system to edit the thalassemia-causing gene in human embryos (Liang et al., 2015; Callaway, 2016). As a result, a number of human embryos have successfully gained the corrected form of the target gene. More recently, the successful application of the CRISPR/Cas9 gene editing technique to human induced pluripotent stem cells (IPSCs) have been reported to generate disease model (Horii et al., 2013) and to treat effectively a number of diseases, including epi-dermolysis bullosa (Osborn et al., 2013), b-thalassemia (Ma et al., 2013), a1-antitrypsin deficiency (Choi et al., 2013), AIDS (Ye et al., 2014), and Niemann-Pick Type C (Maetzel et al., 2014), DMD (Li et al., 2014). Therefore, this novel approach has provided enormous promise for gene therapy of many other forms of diseases including schizophrenia.

Despite the use of genome editing technologies including zinc finger nucleases, TALENS, and CRISPR/Cas9 has opened up the possibility of in vivo genome editing therapy. Unlike peripheral nerve disorder such as DMD, brain and mental disorders (i.e., Parkinson's disease and schizophrenia) have a unique challenge in clinical settings. AAVs vectors, which have been recently approved by FDA for a clinical use, have effectively delivered the nucleases and other components to a variety of tissues including brain. Compared to the other genome editing approaches, the gRNA sequence used in the CRISPR/Cas9 system is more easily to be altered than for TALEs, and a much shorter gRNA and multiple gRNAs can be used to result in multiple DNA cuts (Hsu et al., 2014; Cox et al., 2015) suggesting more promising for the use of CRISPR/Cas9 as a potential therapeutic intent. Most recently, Deverman et al. (2016) reported that they developed a harmless virus, namely AAV-PHP.B, to across the blood brain barrier to successfully deliver treatment to the brain.

These exciting results in living mammals and human cells show great potential for treating human diseases by translating this CRISPR/Cas9 system to a novel therapeutic approach to treat genetically related diseases in humans, including schizophrenia and perhaps other genetically related mental disorders due to the overlapping genetic alterations of schizophrenia with other mental illnesses.

#### CHALLENGES AND FUTURE DIRECTIONS

Despite the remarkable progress, the CRISPR/Cas9 system is still immature and has some limitations and challenges to overcome. Technical improvement is needed, while ethical issues require resolution as well. There are a couple of technical challenges, including specificity, efficiency of the system and penetration of the blood-brain barrier for delivery of the components in the CRISPR/Cas9 system. When a potential therapeutic approach to treat brain and mental disorders is discovered, the blood-brain barrier always poses an obstacle for the drugs to reach the brain cells. The components themselves in the CRISPR/Cas9 system cannot pass through the blood-brain barrier. In an effort to deliver the CRISPR/Cas9 components, a few inactive genetically engineered viruses have been successfully used as a carrier that can be packed with the CRISPR/Cas9 components, cross the blood-brain barrier, and finally unload those molecules in the brain cells. In spite of having no capacity to cause disease, further improvement of the efficiency and the evaluation of long-term safety are needed. In addition to the abovementioned technical concerns, there are ethical issues as well for manipulating genomic DNA in human sperm, eggs and embryos, mainly because of limited knowledge regarding the CRISPR/Cas9 system and the long-term impact of genetic disruption and off-targeting on future generations (Krishan et al., 2016). To date, ncRNAs have not been reported to express proteins; Nelson B. R. et al. (2016) have recently reported that a lnRNA encodes a peptide with the ability to enhance SERCA activity in muscle. With the possibility of unidentified proteins

### REFERENCES


being translated by ncRNAs, it is uncertain whether the proteins may exert any unfavorable, even disease-related activity. All of these new questions require answers; therefore, along with other issues, there is still a long way to go before this revolutionary technology is translated into a clinical cure for schizophrenia and possibly other mental diseases as well due to significant overlaps of the risk genes shared by different forms of psychiatric disorders.

#### AUTHOR CONTRIBUTIONS

CZ: conceptional design and writing of the draft manuscript. WH: conceptional design and writing of the final manuscript. LH, CL, CC and XL collected and examined the enrolled articles in this review.

#### FUNDING

This review was funded by Jiangsu Haosen pharmaceutical Limited by Share Ltd (2016-Young scholar support project to CZ).


integration-free b-thalassemia induced pluripotent stem cells. J. Biol. Chem. 288, 34671–34679. doi: 10.1074/jbc.M113.496174


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer DT and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Zhuo, Hou, Hu, Lin, Chen and Lin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Transcriptomics of Environmental Enrichment Reveals a Role for Retinoic Acid Signaling in Addiction

Yafang Zhang1,2,3, Fanping Kong4,5, Elizabeth J. Crofton1,2,3, Steven N. Dragosljvich1,2,3 , Mala Sinha4,5,6, Dingge Li1,2,3, Xiuzhen Fan1,2,3, Shyny Koshy1,2,3, Jonathan D. Hommel1,2 , Heidi M. Spratt4,5,6,7, Bruce A. Luxon4,5,6 and Thomas A. Green1,2,3 \*

<sup>1</sup> Center for Addiction Research, The University of Texas Medical Branch, Galveston, TX, USA, <sup>2</sup> Department of Pharmacology and Toxicology, The University of Texas Medical Branch, Galveston, TX, USA, <sup>3</sup> Mitchell Center for Neurodegenerative Diseases, The University of Texas Medical Branch, Galveston, TX, USA, <sup>4</sup> Department of Biochemistry and Molecular Biology, The University of Texas Medical Branch, Galveston, TX, USA, <sup>5</sup> Biomedical Informatics Program, The University of Texas Medical Branch, Galveston, TX, USA, <sup>6</sup> Sealy Center for Molecular Medicine, Institute for Translational Science, The University of Texas Medical Branch, Galveston, TX, USA, <sup>7</sup> Department of Preventive Medicine and Community Health, The University of Texas Medical Branch, Galveston, TX, USA

There exists much variability in susceptibility/resilience to addiction in humans. The environmental enrichment paradigm is a rat model of resilience to addiction-like behavior, and understanding the molecular mechanisms underlying this protective phenotype may lead to novel targets for pharmacotherapeutics to treat cocaine addiction. We investigated the differential regulation of transcript levels using RNA sequencing of the rat nucleus accumbens after environmental enrichment/isolation and cocaine/saline self-administration. Ingenuity Pathways Analysis and Gene Set Enrichment Analysis of 14,309 transcripts demonstrated that many biofunctions and pathways were differentially regulated. New functional pathways were also identified for cocaine modulation (e.g., Rho GTPase signaling) and environmental enrichment (e.g., signaling of EIF2, mTOR, ephrin). However, one novel pathway stood out above the others, the retinoic acid (RA) signaling pathway. The RA signaling pathway was identified as one likely mediator of the protective enrichment addiction phenotype, an interesting result given that nine RA signaling-related genes are expressed selectively and at high levels in the nucleus accumbens shell (NAcSh). Subsequent knockdown of Cyp26b1 (an RA degradation enzyme) in the NAcSh of rats confirmed this role by increasing cocaine self-administration as well as cocaine seeking. These results provide a comprehensive account of enrichment effects on the transcriptome and identify RA signaling as a contributing factor for cocaine addiction.

Keywords: drug dependence, stimulants, RNA-seq, differential rearing, self-administration, regionally enhanced gene expression

# INTRODUCTION

Some people experimenting with cocaine become addicted at first exposure, while others are resistant, even after many exposures. These important individual differences in vulnerability to addiction are a function of the interaction between genes and the environment (McGue et al., 1996). Genetic factors that play an essential role in individual differences in susceptibility to drugs of abuse are well studied; however, environmental influences on gene expression is an area in need of further study (Thiriet et al., 2008).

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Rupert W. Overall, Technische Universität Dresden, Germany Elena Martín-García, Pompeu Fabra University, Spain

> \*Correspondence: Thomas A. Green tom.green@utmb.edu

Received: 15 August 2016 Accepted: 25 October 2016 Published: 16 November 2016

#### Citation:

Zhang Y, Kong F, Crofton EJ, Dragosljvich SN, Sinha M, Li D, Fan X, Koshy S, Hommel JD, Spratt HM, Luxon BA and Green TA (2016) Transcriptomics of Environmental Enrichment Reveals a Role for Retinoic Acid Signaling in Addiction. Front. Mol. Neurosci. 9:119. doi: 10.3389/fnmol.2016.00119

Environmental enrichment is a non-drug, non-surgical, non-genetic manipulation producing a protective addiction phenotype in rodent models (Bardo et al., 2001; Green et al., 2002, 2010; Chauvet et al., 2009; Thiel et al., 2009, 2011; Solinas et al., 2010; Alvers et al., 2012; Chauvet et al., 2012; Nader et al., 2014). In the enriched condition (EC), animals are group-housed with access to children's plastic toys which are changed and rearranged daily, while those in the isolated condition (IC) are single-housed without disturbance. In drug self-administration studies, EC rats self-administer less cocaine than IC rats in the acquisition, maintenance, extinction, and reinstatement phases of cocaine self-administration (Green et al., 2010). Understanding the mechanisms underlying the protective phenotype of environmental enrichment may help uncover novel pharmacotherapeutic targets for prevention and treatment of addiction.

The nucleus accumbens (NAc) is an essential brain region for reward (among several regions) and altering the expression of genes in the NAc shell affects cocaine self-administration in rats (Green et al., 2010; Larson et al., 2011; Zhang et al., 2014). First we utilize quantitative RNA sequencing to analyze expression of 14,309 transcripts in the NAc of EC and IC rats selfadministering cocaine or saline. Next, a transcriptomic analysis of topographical gene expression was performed to identify genes expressed selectively in the mouse NAc shell, using tools and data from the Allen Brain Atlas<sup>1</sup> . The convergence of RNA-seq and topographical gene expression analyses pointed clearly to retinoic acid (RA) signaling as the most promising pathway to target.

As the active metabolite of vitamin A, RA acts as an essential molecule in multiple biological processes, such as embryonic development (Rhinn and Dolle, 2012), immune response (Mielke et al., 2013), cell proliferation and differentiation (Chen et al., 2014), and maintenance of the nervous system (Maden, 2007). There is increasing evidence that RA plays an important role in the adult brain (Maden, 2007). Although no study has yet reported on the role of RA in addiction-related behavior per se, there is some evidence that double null mutants of the RA receptor β (RARβ) with the retinoid X receptor RXRβ or RXRγ decrease dopamine D2 receptor expression selectively in the shell of the NAc (Krezel et al., 1998). These rats displayed decreased cocainestimulated locomotor activity; however, they also had severe decrements in the rotarod task and spontaneous locomotor activity.

Retinoic acid is highly concentrated in the brain, including the striatum (Kane and Napoli, 2010). RA is synthesized in the cytoplasm in two steps: first, retinol is oxidized to retinaldehyde via retinol dehydrogenase (Radh a.k.a. Adh). Then, retinaldehyde is irreversibly converted to RA through retinaldehyde dehydrogenase (Raldh a.k.a. Aldh1a1-3). Excess RA is degraded by Cyp26b1 into polar metabolites (Maden, 2007; Chen et al., 2014). We show here that knockdown of Cyp26b1 via a novel adeno-associated viral vector increases cocaine selfadministration.

#### MATERIALS AND METHODS

#### Animals

For the RNA-seq study, male Sprague-Dawley rats (Harlan, Houston, TX, USA) arrived at 21 days of age and were enriched or isolated for 30 days before behavioral testing. EC rats (n = 20) were group-housed in a large metal cage (70 cm × cm 70 × 70 cm) with 14 hard children's plastic toys changed and rearranged daily. This density (245 cm<sup>2</sup> /rat) was higher than a majority of environmental enrichment studies, but was still more than double that required by NIH. EC rats were split into two cages after 50 days of age. IC rats (n = 20) were single housed in standard polycarbonate cages. These conditions produce a resistant (i.e., EC) and susceptible (i.e., IC) behavioral addiction phenotype (Green et al., 2002, 2003, 2010). While it is true that short term isolation is a stressor, enriched rats (even at lower density) show greater signs of chronic stress, even though these rats do not show outward signs of stress (Crofton et al., 2015). Rats remained in these homecage conditions throughout all behavioral tests, except during testing. For vector injection and behavioral tests, male Sprague-Dawley rats were obtained at 225– 249 g. Rats were pair-housed and maintained in a controlled environment (temperature, 22◦C; relative humidity, 50%; and 12 h light/dark cycle, lights on 0600 h) in an Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) approved colony and procedures were approved by the UTMB Institutional Animal Care and Use Committee and conform to the NIH Guide for the Care and Use of Laboratory Animals.

#### Intravenous Cocaine Self-Administration with Environmental Enrichment

Rats were anesthetized with ketamine (100 mg/kg, IP) and xylazine (10 mg/kg, IP), and implanted with an indwelling Silastic catheter (0.2 mm I.D.; Fisher Scientific, Pittsburgh, PA, USA) into the jugular vein. The catheter passed under the skin to exit on the rat's back. The catheters were infused with 0.1 ml of a sterile saline solution containing heparin (30.0 U/ml), ticarcillin (250,000 U/ml) and streptokinase (8000 IU/ml) daily, to prevent infection and maintain catheter patency throughout the duration of experiments.

One week after catheter surgery, rats were allowed to selfadminister 0.5 mg/kg/infusion cocaine (National Institute on Drug Abuse, Bethesda, MD, USA) or saline under a fixed ratio 1 (FR1) schedule for 2 h/day for 14 days. The session terminated when the rat received 30 infusions to eliminate cocaine intake differences between EC and IC rats. Tissue was harvested 3 h after the beginning of the last session (**Figure 1A**).

#### Quantitative RNA Sequencing

Rat brains were harvested 3 h after the beginning of the last self-administration session and the left side of the NAc was dissected on an ice-cold platform for mRNA analysis. The right side was used for protein quantification (Lichti et al., 2014).

<sup>1</sup>http://www.brain-map.org

The RNA was extracted and purified with the RNeasy mini kit (Qiagen, Valencia, CA, USA). cDNA libraries were created by reverse transcribing the RNA and creating the second strand. Blunt ends were phosphorylated and "a-tailed" so that adapters could be ligated to both ends. Adapters were individually "bar coded" and thus samples were not pooled despite having 4 samples per flow cell, yielding n = 7–8 for each condition (total N = 30). RNA was sequenced with a HiSeq 1000 system from Illumina. cDNA was amplified using "bridge" amplification. Base calls were made using fluorescently labeled nucleotides. More than 100 million reads with 50 bp (paired-end reads) were mapped for each rat and quality was checked with FastQC (v0.9.1) (Andrew, 2010). Reads were mapped to the rat reference genome (RN4) using Tophat2 (v2.0.4) (Kim et al., 2013) and Bowtie2 (v2.0.0.6) (Langmead and Salzberg, 2012) software packages. The R package EdgeR (v.3.0.8) (Anders and Huber, 2010; Robinson et al., 2010) was then used for analysis using the log-transformed "trimmed mean for M-values" (TMM) method for normalization and tag-wise dispersion using "count" data. A likelihood ratio F-test was used for generating p-values to compare EC vs. IC rats and cocaine vs. saline. Cross-validation of RNA-seq results was achieved by looking at correlation of expression between qPCR and RNA-seq for Fabp5 (Forward: 5<sup>0</sup> CTTGCACCTTGGGAGAGAAG 3<sup>0</sup> , Reverse: 5<sup>0</sup> CATCTTCCCGTCCTTCAGTT 3<sup>0</sup> ) and Hspa5 (Forward: 5<sup>0</sup> AACCAAGGATGCTGGCACTA 3<sup>0</sup> ; Reverse: 5<sup>0</sup> ATGACCCGC TGATCAAAGTC 3<sup>0</sup> ), with the qPCR being normalized to Reep5 (Forward: 5<sup>0</sup> GGTTCCTGCACGAGAAGAACT 3<sup>0</sup> ; Reverse: 5 <sup>0</sup> GAGAGAGGCTCCATAACCGAA 3<sup>0</sup> ) (Supplementary Figure S6). The Fabp5 was chosen because it is in the RA pathway and selectively expressed in the NAc shell, both aspects central to this paper. Hspa5 (Bip) was chosen because it has extremely high expression levels, and because we have studied this transcript with cocaine in the past (Pavlovsky et al., 2013). Using qPCR to cross-validate the RNA-seq data is somewhat problematic. Our conclusion is that using a less accurate and less precise method to validate a more accurate and precise method is difficult at best. The problem encountered is that qPCR needs to be normalized to a control or "housekeeping" transcript. The variance from the transcript of interest gets compounded with the variance inherent in the housekeeping gene, increasing variance. Multiple transcripts were assessed for normalization and the end results were quite different depending solely upon which normalization gene was chosen. Many traditional "housekeeping" transcripts were regulated (e.g., gapdh) or trended toward regulation (e.g., β-actin) by enrichment and/or cocaine. In the end, Reep5 was chosen because it was a housekeeping gene not regulated by cocaine or enrichment. However, the variability of the Reep5 expression among samples washed out main effects.

The quantitative transcriptomic data have been deposited in NCBI's Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE88736<sup>2</sup> .

#### <sup>2</sup>http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE88736

#### Ingenuity Pathway Analysis (IPA)

In order to study the biological functions and pathways regulated by cocaine and environmental enrichment, transcripts significantly regulated (p < 0.05) were analyzed with IPA. Canonical Pathways, Upstream Regulators, Diseases and Biological Functions, and Networks were used to identify significantly regulated transcript sets. Because regulation of any given gene could be a statistical anomaly (i.e., false positive), bioinformatic analyses have been developed under the assumption that regulation important for function will occur in a coordinated fashion at multiple targets within a given pathway. Thus, the IPA analysis assesses over-representation of multiple targets within known pathways. For this analysis, the list of significantly regulated genes is further analyzed for significantly regulated pathways, thus minimizing the effect of individual false positives. Preliminary analysis of ribosomal proteins, tyrosine phosphatases, de-ubiquitinating enzymes and proteasomal proteins, all of which show highly coordinated regulation (see Supplementary Figures S4A,B) demonstrate that the p < 0.05 cutoff does **not** produce an overabundance of false positives.

#### Gene Set Enrichment Analyses (GSEA)

To complement the IPA analysis and avoid the problems of Type I vs. II error, normalized expression intensity (normalized by EdgeR using TMM) of all identified transcripts were analyzed by GSEA. The normalized enrichment score (NES) indicates the degree to which this gene set is overrepresented at the top or bottom of the ranked list of genes in the expression dataset (Subramanian et al., 2005). In IPA analyses, a p-value cutoff is required to decide whether transcription is significantly regulated or not. The problem of setting a cutoff is that high stringency introduces false negative results (i.e., high Type II error), whereas low stringent cutoffs introduce false positive results (i.e., high Type I error). To complement the IPA analysis and avoid the problem of Type I vs. II error, normalized expression intensity of all identified transcripts were analyzed by GSEA. In the GSEA analysis, all transcripts are ranked by a signal2noise metric (the difference of means scaled by the standard deviation), and the significance of a given set of genes is determined using a runningsum statistic to determine rank-order over-representation. Thus, this analysis produces a statistic at the gene set level without the need for a p-value cutoff. Gene sets were from the Broad Institute's set (v.4.0) of curated gene sets [C2.all], gene ontology gene sets [C5.all] and transcription factor target gene sets [C3.tft]. In GSEA, the Enrichment Score (ES) for the gene set is the degree to which this gene set is overrepresented (i.e., "enriched") at the top or bottom of the ranked list of genes in the expression dataset. NES indicates the enrichment score for the gene set after it has been normalized to adjust for size of the gene set.

#### Quantitative Proteomics

A secondary analysis of our previously published liquid chromatography tandem mass spectrometry (LC-MS/MS) protein data from these same rats (Lichti et al., 2014) was used to corroborate RNA results where appropriate.

#### Topographic Transcriptomic Analysis

fnmol-09-00119 November 14, 2016 Time: 18:7 # 5

The regional specificity of gene expression within the brain strongly suggests that genes are not expressed where they are not needed. Thus, in situ visualization of regional expression patterns in brain slices can provide a topographical map of enhanced gene expression for each specific brain region. One could then hypothesize that genes with regionally enhanced expression patterns in the NAc shell would likely be important for addiction-related behavior. Accordingly, the Allen Brain Institute's Anatomic Gene Expression Atlas Gene Finder<sup>3</sup> (Lein et al., 2007) algorithm was used with coronal image sets, seed coordinates of 4200, 5400, 6400 (right side) and 4200, 5600, 5000 (left side) with an expression threshold of 1. This Atlas defines mouse brain expression patterns; however, there is good concordance in expression patterns among mice, rats and humans (Zetterstrom et al., 1999; Strand et al., 2007). The genes with a regional enrichment score of 1.25 fold or higher (>250 genes) were then culled by individual manual inspection to remove genes whose enrichment fold change was clearly driven by experimental artifact such as bubbles, debris, misalignment and high expression in nearby regions (e.g., Islands of Calleja, NAc core, olfactory tubercles, etc.) to yield 178 genes. These 178 genes were submitted to IPA analysis. Once the RA signaling pathway was identified, manual curation of sagittal sets for RA signaling genes without coronal sets was used to identify additional targets (Rbp1 and Rdh10) with shell-specific expression. Images for **Figures 6A–D** were taken from the Allen Brain Atlas:

**STRA6**: http://mouse.brain-map.org/experiment/siv?id=750414 92&imageId=74948838&initImage=ish&coordSystem=pixel&x= 5288.5&y=3704.5&z=1

**FABP5**: http://mouse.brain-map.org/experiment/siv?id=706343 96&imageId=70561807&initImage=ish&coordSystem=pixel&x= 4104.5&y=3104.5&z=0

**RBP1**: http://mouse.brain-map.org/experiment/siv?id=680768 99&imageId=68005433&initImage=ish&coordSyste&imgeId=68 005433&initImage=ish&coordSystem=pixel&x=7448.5&y=4096. 5&z=1

**RDH10**: http://mouse.brain-map.org/experiment/siv?id=75831 736&imageId=75791261&initImage=ish&coordSystem=pixel&x =7472.5&y=4040.5&z=1

**ALDH1A3:** http://mouse.brain-map.org/experiment/siv?id=75 861799&imageId=75806893&initImage=ish&coordSystem=pixel &x=6872.5&y=3496.5&z=1

**CYP26B1:** http://mouse.brain-map.org/experiment/siv?id=7208 1548&imageId=71940737&initImage=ish&coordSystem=pixel &x=4280.5&y=3160.5&z=1

**RAR**β**:** http://mouse.brain-map.org/experiment/siv?id=7503 8442&imageId=74930074&initImage=expression&colormap=0.5,1, 0,256,4&coordSystem=pixel&x=4416.5&y=2952.5&z=1

Finally, a literature search of published papers with in situ and immunohistochemical images was used to confirm Allen Brain Atlas expression and identify additional shell-specific expression components of the RA signaling pathway (McCaffery and Drager, 1994; Zetterstrom et al., 1999).

#### Adeno-Associated Virus knockdown of Cyp26b1

In order to knock down CYP26b1 expression, five 24-nucleotide sequences were identified within the CYP26b1 mRNA sequence (Ensembl transcript ID: ENSRNOT00000020505) using the criteria previously described (Hommel et al., 2003; Benzon et al., 2014) (Supplementary Table S1). The oligonucleotide sequences were synthesized and the annealed hairpin oligonucleotides were cloned into pAAV-shRNA plasmids (pAAV-Cyp26b1shRNA). Hairpin expression from these plasmids was driven by the mouse U6 promoter using a pol-III mechanism. In addition to the hairpin, enhanced green fluorescent protein (eGFP) was expressed from a separate expression cassette driven by a pol-II promoter (CMV).

In order to determine the most effective hairpin, all five hairpins were screened in vitro. Since HEK293 does not express rat Cyp26b1, a Cyp26b1 overexpression plasmid was constructed. To create the Cyp26b1 overexpression plasmid to test knockdown efficiency, the rat CYP26b1 gene sequence was amplified from rat genome cDNA using polymerase chain reaction (forward primer: TAGGAATTCCTCCTGGGTTTCTTCGAGGG; Reverse: TAG GTCGACATCCAAGAGGGTGGGAGTCA) and cloned into the pAAV-IRES-hrGFP plasmid (Agilent Technologies, Santa Clara, CA, USA). The various pAAV-Cyp26b1**-s**hRNA plasmids or pAAV-Control shRNA plasmid was co-transfected with pAAV-Cyp26b1-IRES-hrGFP plasmid into HEK-293 cells using FuGENE <sup>R</sup> 6 Transfection Reagent (Promega)/Lipofectamine 2000 (Life Technologies, Grand Island, NY, USA).

The cells were harvested 24–48 h later, followed by RNA extraction and reverse transcription to cDNA. The RNA was extracted using RNeasy Mini Kit (Cat No. 74104). Contaminating DNA was removed (TURBO DNA- Free, Life Technologies, Carlsbad, CA, USA) and 5 µg total RNA was reverse transcribed into cDNA (SuperScript III First Strand Synthesis: Invitrogen catalog # 18080051). Relative knockdown was measured with Real-time PCR (SYBR Green: Applied Biosystems, Foster City, CA, USA) on an Applied Biosystems 7500 fast thermocycler with Cyp26b1 qPCR primers (forward: CCAGCAGTTTGTGGAGAATG; Reverse: GTCCAGGGCGTCTGAGTAGT). The results were normalized to Gapdh (forward: AACGACCCCTTCATTGAC; reverse: TCCACGACATACTCAGCAC). All primers were validated and analyzed for specificity and linearity prior to experiments (Alibhai et al., 2007).

Efficiency of hairpin was further validated by western blot. HEK293 cells were homogenized in a buffer containing sucrose, Hepes buffer, sodium fluoride, 10% SDS, and protease and phosphatase inhibitors (Sigma–Aldrich: P-8340, P-2850, P-5726). Protein concentration was assessed using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA). Protein samples were denatured at 95◦ for

<sup>3</sup>http://mouse.brain-map.org/agea

10 min and run on a 12% gel (Criterion TGX, Bio-Rad Laboratories, Hercules, CA, USA) then transferred to a polyvinylidene fluoride (PVDF) membrane (Millipore, Billerica, MA, USA). The membrane was blocked with blotting-grade blocker (non-fat dry milk), incubated with Cyp26b1 primary antibody (rabbit, 1:500, 21555-1-AP, Proteintech, Rosemont, IL, USA) and GAPDH primary antibody (mouse, 1:10000, Abcam, Cambridge, MA, USA), washed with TBST and then incubated with fluorescent secondary antibodies (donkey antirabbit (780 nm), donkey anti-mouse (680 nm), 1:15000, Li-Cor Biosciences, Lincoln, NE, USA). Western blots were then imaged (Odyssey, Li-Cor Biosciences, Lincoln, NE, USA) and protein levels quantified with the Odyssey software. The hairpin plasmid with the highest knockdown efficacy (50 .AGTTCTTTGGTCTAGACTCCAATC.3<sup>0</sup> ) was packaged into Adeno-Associated Virus2 (AAV2) by the University of North Carolina Gene Therapy Core Facility and used in subsequent behavioral tests. Control vector expressed a previously validated control hairpin not targeted to any gene (Hommel et al., 2003; Benzon et al., 2014).

#### In vivo Knock Down of Cyp26b1

An AAV2-based vector that expresses Cyp26b1 shRNA and eGFP or a non-targeted hairpin control vector (n = 10–11) was injected bilaterally into the rat NAc shell (1 µl/side over 10 min) using coordinates AP = 1.7, L = 2.2, D = −6.7. To validate the knockdown efficiency of AAV in vivo, AAV-control shRNA or AAV-Cyp26b1 shRNA (n = 6) was injected in the nucleus accumbens shell (NAcSh) in rats. Two microliter of AAV was injected per side of the NAc to increase the number of infected neurons. NAc regions with eGFP fluorescence were collected and tissues from two rats were pooled together to increase the yield of protein concentration. The expression of CYP26B1 protein level was detected using western blot described above. For behavioral tests, an AAV2-based vector that expresses Cyp26b1 shRNA and eGFP, or control vector (n = 10–11 each) was injected bilaterally with 1 µl/side into the rat NAc. Pair-housed rats were used instead of isolated rats for this study to increase relevance to the scientific community by demonstrating the effects of Cyp26b1 independent of the EC/IC procedure. Behavioral tests started 5 weeks after stereotaxic surgery (**Figure 7**). Accurate placement was verified immunohistochemically after the conclusion of behavioral testing.

#### Cocaine Self-Administration Acquisition

One week after catheter surgery, all rats were placed in operant chambers (30 cm × 24 cm × 21 cm; Med-Associates, St. Albans, VT, USA) and allowed to self-administer 0.2 mg/kg/infusion unit dose of cocaine for 2 h per session for 5 days; then 0.5 mg/kg/infusion for 3 days on a fixed ratio (FR1) schedule. Each infusion was delivered intravenously in a volume of 0.1 ml over 5.8 s. The infusion was signaled by illumination of two cue lights for 20 s, which signaled a timeout period during which no further infusions could be attained. **Fixed ratio dose response**: Each rat was allowed to self-administer 0.5, 0.25, 0.125, 0.06, 0.03, 0.015, 0.0075, 0.00325 mg/kg/infusion cocaine in descending order on an FR1 schedule each day for five consecutive days. Rats self-administered each dose of cocaine for 30 min. **Cue responding**: Rats were subjected to forced abstinence in their home cages for 7 days. On the 8th day, rats were placed in the operant chamber and allowed to self-administer **saline** under an FR1 schedule for 1 h with cue light presentation contingent on bar pressing. **Extinction**: Stably responding rats underwent a within-session extinction procedure for 3 days. All rats were allowed to self-administer 0.5 mg/kg/infusion cocaine under an FR1 schedule for 1 h followed by extinction for 3 h. During the extinction period, lever responding resulted in cuelight illumination under an FR1 schedule, but the infusion pump did not deliver cocaine. **Reinstatement**: All rats received 0.5 mg/kg/infusion unit dose of cocaine under an FR1 schedule for 1 h followed by 3 h of extinction. Next, all rats received an IP injection of cocaine of one of five doses (0, 2.5, 5, 10, 20 mg/kg) in a random order for each rat across the five sessions, followed by 3 h reinstatement responding session.

#### Immunohistochemistry

For **Figure 7D**, the placement of AAV-Cyp26b1shRNA expression in vivo was validated by immunofluorescence staining with eGFP. The brains were extracted, post fixed, cryoprotected and sectioned into 40 µm slices containing the NAc on a sliding freezing microtome (Leica Biosystems, Richmond, IL, USA). The slices remained floating and were rinsed with 1xPBS prior to blocking with 3% normal donkey serum (Jackson ImmunoResearch, West Grove, PA, USA) with 0.3% triton. NAc slices were incubated with eGFP primary antibody overnight (1:500, chicken, Aves labs, Tigard, OR, USA) with 3% donkey serum, 0.3% triton in 1xPBS. After washing, slices were incubated with secondary Alexa 488 donkey anti chicken antibody (Jackson ImmunoResearch, West Grove, PA, USA) in 1xPBS. Finally, slices were mounted, dehydrated using ethanol and CitriSolv (Fischer Scientific, Waltham, MA, USA) and coverslipped with DPX (Fisher Scientific).

# Statistical Analysis for Behavior

Two-factor analyses of variance (ANOVAs) and two-factor repeated-measures ANOVAs were performed to compare four treatment groups. Significance between only two conditions was analyzed using a Student's t-test. All t-test data passed the Shapiro–Wilk test of normality. All data are expressed as mean ± SEM. Statistical significance was set at p < 0.05.

# RESULTS

## Descriptive Statistics

Raw and processed RNA sequencing data can be found in the Gene Expression Omnibus database with the project number GSE88736.

After the primary data alignment and analysis, 14,309 transcripts were quantified as the result of RNA-seq. The first step of our analysis was to investigate the regulation of individual transcripts. Based on the likelihood F-tests, the Venn diagram shows 106 transcripts significantly regulated (p < 0.001) by

cocaine, 683 transcripts significantly regulated by environmental enrichment and 64 transcripts significant for the interaction (**Figure 1B**). In addition, the Venn diagram also displays the number of transcripts common among the effects, which indicates the overlapping effects of cocaine and environmental enrichment. Note that there are more transcripts upregulated than downregulated by cocaine, while there are many more downregulated transcripts than upregulated by environmental enrichment (**Figure 1C**).

In the significantly regulated gene lists, we first looked at the top 50 upregulated and downregulated transcripts from cocaine and environmental enrichment. These transcripts give the highest confidence of regulation. For cocaine (**Figure 1D**), immediate early genes such as EGR4, NR4A3, FOS, EGR2 were induced, which agrees with previous publications (Hope et al., 1992; Berke et al., 1998; Werme et al., 2000; Guez-Barber et al., 2011). For enrichment, the top regulated transcripts include transcription factors, such as NR4A1, EGR3, ARC, EGR2 and NR4A3, etc. (**Figure 1E**). In our search for a therapeutic target, we moved beyond individual transcripts to explore molecular pathways.

### Cocaine Effects on Transcription in the NAc

To investigate which biological functions and cellular pathways are regulated by cocaine, significantly regulated transcripts were analyzed with IPA. **Figure 2A** lists some top-ranked diseases and biological functions of interest for the cocaine main effect. Complete IPA data can be found in the supplemental information.

#### Data Validation

In large data sets such as RNA sequencing, some orthogonal data cross-validation is important to provide confidence in the validity of the results. One approach of validating the RNA-seq data is to compare results with previous cocaine studies. In the top-ranked canonical pathways (**Figure 2B**), Signaling by Rho GTPases is represented in Supplementary Figure S1 and was previously shown to be repressed by cocaine in the NAc (Kim et al., 2009; Gourley et al., 2011). The Endoplasmic Reticulum Stress pathway identified in the current results was confirmed previously (Pavlovsky et al., 2013). In the Upstream Regulator analysis, cocaine and Creb1 are predicted to be upstream regulators by the IPA analysis (Supplementary Figures S1B,C). Full results of the upstream regulator analysis are presented in table form in the supplemental information. Additionally, Depressive Disorder and Anxiety Disorders, which show comorbidity with cocaine abuse (Sonne et al., 1994; Morton, 1999), were highlighted (Supplementary Figure S1D).

#### Retinoic Acid Pathway

One novel pathway highlighted in the cocaine main effect is RA signaling. Specifically, Retinoic acid receptor (RAR) activation was identified as a regulated canonical pathway (p = 0.023). Further, in the upstream regulator analysis, RA (p = 9.78E-10, activation Z-score = −1.553; **Figure 2C**), RA receptor α (RARA; p = 1.69E-7, Z-score = 0.723; **Figure 2D**), and the RARγ agonist CD437 (p = 4.25E-4, Z-score = −3.15, data not shown) were predicted to be upstream regulators of the cocaine main effect, all suggesting that RA signaling is important for the effects of cocaine. Full results of the upstream regulator analysis are presented in table form in the supplemental information. A secondary proteomic analysis from our previously-published report (Lichti et al., 2014) confirms CD437 as an upstream regulator at the protein level (p = 7.70E-4, Z-score = −2.23; **Figure 5**).

#### Other Functions and Pathways

In our results, one of the most regulated Diseases and Biological Functions from the IPA analysis is Neurodegeneration-related Disorders. In the Upstream Regulator analysis, 4 top predicted upstream regulators from the cocaine main effect are involved in neurodegeneration (Supplementary Figure S2), including amyloid precursor protein (APP; p = 2.66E-13), microtubuleassociated protein tau (MAPT; p = 4.36E-7), presenilin 1 (PSEN1; p = 4.11E-7) and huntingtin (HTT; p = 1.16E-13). Additionally, the Canonical Pathway Huntington's Disease signaling (p = 0.000282) is significantly regulated.

#### Environmental Enrichment Effects on Transcription in the NAc

To explore the molecular mechanism of environmental enrichment, IPA and GSEA were used to analyze the biological functions and pathways. For the environmental enrichment main effect, many mental disorder-related functions and diseases were significantly regulated (**Figure 3A**). The top-ranked canonical pathways involved Protein Translation-Related EIF2 Signaling, PKA Signaling, Mitochondrial Dysfunction, Kinase Signaling, etc. (**Figure 3B**). Complete IPA data can be found in the supplemental information.

#### Retinoic Acid Signaling Pathway

The RA signaling pathway was also significantly regulated by environmental enrichment (p = 9.77E-06) (**Figure 3B**). Within the RA signaling pathway, the transcripts involved in RA synthesis and translocation are upregulated by environmental enrichment, such as retinol binding proteins 1 and 4 (Rbp1 and Rbp4), retinol dehydrogenase 10 (Rdh10), and cellular RA binding protein 2 (Crabp2), while the repressors of this pathway, such as the kinases Akt and Pkc are mainly downregulated (**Figure 3C**). Additionally, 215 RA target genes are regulated by environmental enrichment, most being downregulated (p = 1.84E-5, activation Z-score = −2.655; **Figure 3E**). Further, the agonist of the RA receptor γ, CD-437, was predicted to be inhibited as an upstream regulator of enrichment (p = 3.63E-18, activation Z-score = −7.541; **Figure 3D**). A secondary analysis of protein data from these rats (Lichti et al., 2014) confirms CD437 as an upstream regulator (p = 4.83E-15, Z-score = −3.13; **Figure 5**), highlighting the importance of RA signaling for further study.

#### Other Functions and Pathways

One of the most striking environmental enrichment effects was the regulation of transcription. The Gene Ontology

gene set containing 217 transcription factor-related genes suggests a decrease in EC rats (Supplementary Figure S3A; NES = 1.67, p = 0.004). At the top of this gene set, the transcription of Egr transcription factors was highly repressed by environmental enrichment. In support of a functional effect on EGR transcription factors, Egr1/2/3 target genes were also significantly repressed by environmental enrichment, as shown by a GSEA analysis of the Transcription Factor Target gene set V\$EGR\_Q6 (Supplementary Figure S3B; NES = 1.64, p < 0.001). Complete GSEA results can be found in the Supplementary Information. IPA also highlighted the roles of transcription factors. In the Upstream Regulator analysis, Fos, Esrra and Srf were identified as likely upstream regulators of the transcripts regulated by environmental enrichment. Differentially regulated Fos target genes revealed that Fos activity was enhanced in environmental enrichment despite the mRNA for Fos being down at 3 h (Supplementary Figure S3C; p = 1.06E-07, activation z-score = 1.420). In addition, Esrra (Supplementary Figure S3D; p = 2.11E-3, activation z-score = 2.668) was also identified as an activated upstream

FIGURE 3 | Transcripts regulated by environmental enrichment. (A) Selection of significantly regulated Biological Functions and Diseases by environmental enrichment as determined by IPA. The y-axis represents –log (p-value). Dotted line represents the threshold p = 0.05. (B) Selection of significantly regulated canonical pathways by environmental enrichment as determined by IPA. Orange line represents the ratio of regulated transcripts to all transcripts in the pathway. (C) Canonical pathway showing regulation of transcripts for retinoic acid signaling by environmental enrichment. (D,E) Transcripts from Upstream Regulator analysis whose expression predicts repressed signaling of CD437 (D) and retinoic acid (E) by environmental enrichment. Red represents upregulation; green represents downregulation.

regulator. This result agrees with our prior research that Esrra was identified as an upstream regulator with energy metabolism proteins regulated in the proteomic study from tissue from these same rats (Lichti et al., 2014). Another transcription factor that was identified as upstream regulator was Srf (Supplementary Figure S3E; p = 2.34E-5, activation z-score = 0.101).

Beyond transcription, EIF2 signaling (p = 5.25E-17) was the top regulated canonical pathway (**Figure 3B**). Transcription of Eif3, Eif4a, and 39 different 40S and 60S ribosomal subunits were increased by environmental enrichment while transcription of the Eif2 inhibitor, Gsk3β, and the phosphatase of Eif2b activator, Erk, were downregulated (Supplementary Figure S4A).

In addition to protein synthesis, mRNA for the Protein Ubiquitination pathway (p = 3.63E-09) was also significantly differentially regulated by environmental enrichment (Supplementary Figure S4B). These results were confirmed by the GSEA of the Gene Ontology gene set for Proteasome Complex (NES = 1.50, p = 0.024). In the protein ubiquitination process, target polyubiquitinated proteins undergo either degradation by the proteasome or de-ubiquitination by de-ubiquitinating enzymes (DUBs). Our results show an increase in transcription of ubiquitin C (UBC) and many proteasomal subunits with a coordinated decrease in transcription of DUBs (Supplementary Figures S4B,C), indicating that the enrichment condition likely enhances protein degradation through ubiquitination, an effect in agreement with our prior investigations of the proteomics of environmental enrichment (Fan et al., 2013b; Lichti et al., 2014). In addition to the pathway analysis, the role of UBC is highlighted in the Network analysis by IPA (Supplementary Figure S4D, network score = 35). Transcription of 21 out of 35 UBC target genes was upregulated by environmental enrichment. Sumoylation and ubiquitination have an important crosstalk in determining protein fate (Gill, 2004; Ulrich, 2012; Sriramachandran and Dohmen, 2014). SUMO proteins 1, 2, and 3 are the major hubs in another network determined by IPA (Supplementary Figure S4D), with 29 of 44 transcripts downregulated.

## Cocaine X Environment Interaction

The transcription regulated by the interaction indicates that EC and IC rats respond differently to cocaine (**Figure 4**). Studying this differential response helps to identify the molecular mechanisms of the protective EC phenotype. For the interaction, Drug Dependence (p = 3.38E-6; Supplementary Figure S5A) and Release of Dopamine (p = 5.41E-7; Supplementary Figure S5B) were top-regulated diseases and biological functions in IPA. Regulated transcripts in Drug Dependence were dominated by ion channels and G-protein coupled receptors (GPCRs). Release of Dopamine was also dominated by GPCRs.

#### Retinoic Acid Signaling Pathway

Retinoic acid receptor (RAR) activation was also identified in the Canonical Pathway analysis (p = 5.13E-05) in the interaction of Cocaine and Enrichment (**Figure 4B**). Some essential genes in this pathway showed significant interaction at the mRNA level, such as retinol binding protein (Rbp4), retinol dehydrogenase (Rdh10), retinoid X receptor (Rxr), etc. (**Figure 4D**). Additionally, RA was identified as an upstream regulator (p = 1.6E-2; **Figure 4C**). These results indicate that the activity of the RA signaling pathway and the transcripts of RA target genes are differentially regulated by cocaine in EC and IC rats, and are therefore a promising avenue for developing novel addiction therapeutics.

#### Other Functions and Pathways

Protein kinases play a significant role in post-translational modification to activate or inhibit target proteins through phosphorylation. For the interaction, Activation of Protein Kinases was also identified as differently regulated (p = 1.81E-6; Supplementary Figure S5C). This prediction is based on the regulation of kinase and kinase related transcripts, including 6 different mitogen-activated protein kinases (MAPKs; Supplementary Figure S5D). In addition to the expression of kinases in general, Protein Kinase A Signaling (p = 1.6E-05) ranked 12th among the regulated canonical pathways. Our results also show that corticosterone was identified as an upstream regulator in the NAc (p = 8.66E-4; Supplementary Figure S5F). This result is not surprising because it has been found that EC rats have blunted induction of corticosterone induced by psychostimulants (Stairs et al., 2011; Crofton et al., 2015). One important function in the NAc that responded differently to cocaine in EC and IC rats was Transport of Ca2<sup>+</sup> (p = 3.67E-5; Supplementary Figure S5G). Angiotensinogen (AGT), ATPase (ATP2B4) and voltage-dependent calcium channel (CACNA1G) lead to activation of transport of Ca2+, while parathyroid hormone-like hormone (PTHLH) and arginine vasopressin (AVP) lead to inhibition (Supplementary Figure S5G). In addition to the above functions, NMDA receptor downstream transcripts (p = 2.72E-6; Supplementary Figure S5E) also responded differently to cocaine in EC and IC rats.

### Validation of Quantitative RNA Sequencing

To confirm the validity of the quantitative RNA sequencing technique, real-time PCR was used to quantify the mRNA expression of Fabp5 and Hspa5 from the same rats. mRNA fold change results from RNA sequencing and qPCR of Fabp5 (R <sup>2</sup> = 0.5155, p < 0.0001) and Hspa5 (R <sup>2</sup> = 0.4301, p = 0.0003) was compared for every rat. These results indicate qPCR and RNA sequencing results are well correlated in addition to the orthogonal validation of comparing the current results against previous cocaine and enrichment findings.

#### Regional Enhancement (NAc shell) of Retinoic Acid-Related Genes

A topographic transcriptomic analysis for genes with selectively enhanced gene expression in the NAc shell from the Allen Brain Atlas identified 178 transcripts with selective expression in the NAc shell ≥ 1.25 fold over surrounding regions. An IPA analysis of these 178 genes revealed the central RA pathway as being significant with Stimulated by RA 6 (Stra6), Retinoic acid receptor β (Rarb), Fatty

represents –log (p-value). Dotted line represents the threshold p = 0.05. (B) Selection of significantly regulated canonical pathways. Orange line represents a ratio of regulated genes to all genes in the pathway. (C) Transcripts from Upstream Regulator analysis of the Environmental Enrichment X Cocaine interaction whose expression predicts regulation by retinoic acid. (D) Canonical pathway showing regulation of transcripts for retinoic acid signaling. Yellow represents differential regulation in EC and IC rats in response to cocaine.

Acid Binding Protein 5 (Fabp5) and Cytochrome P450, Family 26, Subfamily B, Polypeptide 1 (Cyp26b1) as shellselective genes (p = 1.67E-5; **Figure 6**). A manual search of NAc shell-selective RA gene expression in sagittal sections identified Rbp1, Rdh10 and Aldh1a3 as additional selective genes. A search of the literature confirmed Rbp1 and Rarb selective expression, along with Aldh1a1 and Rxrg (McCaffery and Drager, 1994; Zetterstrom, 1999). These nine NAc shell-enhanced genes are depicted in the pathway for **Figure 6H**.

# RA Signaling in NAc Shell Increases Cocaine Self-Administration

Two strategies for changing concentrations of RA are to either alter synthesis or degradation. However, because there are several different subtypes of Rdhs and Raldhs, knocking down any one could be compensated by the other RA synthases. Therefore, we decided to alter RA concentration by knocking down the degradation enzyme, Cyp26b1, since RA signaling was regulated by cocaine, enrichment, and their interaction,

and many genes showed selectively enhanced expression in the NAc shell. With decreased expression of Cyp26b1, RA subsequently builds up in neurons, enhancing RA downstream signaling (Kim et al., 2014). To knock down the expression of Cyp26b1, an shRNA targeting the Cyp26b1 coding sequence was designed and knockdown efficiency was examined in vitro and in vivo. Compared with a non-targeted control shRNA, Cyp26b1 shRNA significantly decreased expression of Cyp26b1 in HEK293 cells in both mRNA and protein (**Figure 7A**) and in rat NAc shell at the protein level (**Figure 7B**). **Figure 7C** shows the schematic diagram of the experimental timeline for behavioral tests. For behavioral testing, AAV-Cyp26b1shRNA or control vector was injected into the NAc shell of rats (**Figure 7D**, atlas comparison **Figure 7E**). For acquisition of cocaine self-administration, results demonstrate significant acquisition across sessions [F(4,80) = 3.855, p = 0.006; **Figure 7F**] with a trend for increased acquisition in Cyp26b1 shRNA rats [interaction F(4,80) = 2.142, p = 0.083]. A ttest showed that Cyp26b1 shRNA rats responded significantly more for Sessions 3 and 4. For maintenance responding (**Figure 7G**), knocking down Cyp26b1 significantly increased responding for cocaine at low unit doses, resulting in significant main effects of Dose [F(7,26) = 39.239, p < 0.001], Vector [F(1,18) = 7.938, p = 0.011] and a significant interaction [F(7,126) = 4.871, p < 0.001]. Knockdown significantly elevated the responding for cues in a cue responding test [**Figure 7H**; t(20) = −2.572, p = 0.018]. Finally, in a withinsession extinction procedure (**Figure 7I**), rats with Cyp26b1 knockdown exhibited increased responding compared to control rats, with significant main effects of Session [F(2,36) = 15.317, p < 0.001] and the Vector main effect is at the threshold of the p-value cutoff for significance [F(1,18) = 4.433, p = 0.05], indicating that knockdown of Cyp26b1 in the NAc shell enhances drug-seeking behavior. In cocaine-induced reinstatement, high variance prevented detection of a difference in reinstatement between the two groups of rats at any dose.

# DISCUSSION

These studies highlight mechanisms of the protective addiction phenotype of environmental enrichment and identify novel targets that play a role in regulating addiction-related behavior. Among the novel molecules and pathways identified, the RA signaling pathway was predicted to play an important role in the differential response to cocaine in EC and IC rats. Separately, a topographic transcriptomic analysis identified RArelated genes and RA target genes as being selectively expressed in the NAc shell, further highlighting the likely importance of RA in addiction-related behavior. These results generated a hypothesis-driven experiment that confirmed the role of RA in addiction-related behavior.

# Cocaine Transcriptomic Effects

In the upstream analysis of cocaine-regulated transcripts, cocaine itself and CREB1, an important mediator of the effects of psychostimulants (Carlezon et al., 1998; Pliakas et al., 2001), ranked at the top of the list as upstream regulators, strongly supporting that cocaine-regulated transcription seen here agrees with previous studies. Our prior research demonstrated that enriched rats have less phospho-CREB in the NAc and that decreasing CREB function in the accumbens shell produces a behavioral phenotype identical to that of environmental enrichment (Bowling and Bardo, 1994; Green et al., 2002, 2010), an interesting behavioral phenotype marked by increased sensitivity to the rewarding effects of stimulants (as measured by CPP) coupled with decreased self-administration (Pliakas et al., 2001; Larson et al., 2011).

## Environmental Enrichment and Transcription

Compared to the cocaine main effect, there were approximately 5X more transcripts significantly regulated by environmental enrichment, revealing that environment has a much more extensive impact on gene expression than cocaine exposure. Transcription factors were the most regulated gene sets by environmental enrichment. Another impressive difference between EC and IC rats is the regulation of transcripts involved with EIF2 signaling. Even though Eif2 itself is not regulated at the mRNA level, decreased upstream inhibitors of Eif2 and upregulated downstream ribosomal subunits suggests regulation of the protein translation process. In addition to protein synthesis, the protein degradation system is also altered by enrichment. Prior research from this laboratory demonstrated that expression of ubiquitin target proteins is different in EC and IC rats (Fan et al., 2013a,b). Ubiquitination is also important in differential expression of proteins from the current rats (Lichti et al., 2014). The current mRNA data revealed increased transcription of ubiquitin and proteasomal subunits, but reduced mRNA expression for deubiquitinating enzymes in the NAc, possibly suggesting enhanced protein degradation in EC rats. Taken together, enhanced protein translation and degradation likely indicate more rapid protein turnover in EC rats compared with IC rats.

Rbp1 in situ mRNA expression. Signal image is inset. (C) Coronal section of Fabp5 expression. (D) Sagittal section of Rdh10 expression. (E) Coronal section of Cyp26b1 expression. (F) Sagittal section of Aldh1a3. (G) Coronal section of Rarβ expression. (A–G Images are from Allen Brain Atlas) (H). Red symbols denote retinoic acid-related genes with NAc shell-specific expression.

pAAV-Cyp26b1shRNA and Cyp26b1 overexpression plasmid. (B) Fold change of Cyp26b1 protein (±SEM) in rat NAc shell injected with AAV- non-targeted control shRNA or AAV-Cyp26b1shRNA. (C) Schematic diagram of experimental procedure for vector behavioral tests. (D) Representative in vivo titer from AAV-Cyp26b1 shRNA vector co-expressing eGFP. Bar represents 75 µm. (E) Stereotaxic atlas (Paxinos and Watson, 2005) defining region targeted. (F–I) Knocking down Cyp26b1 in NAc increases response rate for acquisition of cocaine self-administration at 0.2 mg/kg/infusion in 2-h sessions (F), maintenance response rate at low unit doses (30min/dose) (G), cue response rate (1 h) (H) and increases in extinction responding (1-h FR1 session followed by 3-h extinction) (I). (∗p < 0.05; n = 9–10/group).

#### Retinoic Acid Signaling

fnmol-09-00119 November 14, 2016 Time: 18:7 # 15

Retinoic acid genes are selectively expressed in the NAc shell, as shown by a topographic transcriptomic analysis of the Allen Brain Atlas and the published literature (McCaffery and Drager, 1994; Zetterstrom, 1999). Given the importance of the NAc shell to addiction, the RA signaling pathway offers promising targets for novel therapeutic development for cocaine addiction. One previous report found that constitutive whole-body RARβ/RXRβ or RARβ/RXRγ double null mice had a selective decrease of dopaminergic D2 receptors in the shell of the NAc, with concomitant decrements in locomotor and rotarod performance (Krezel et al., 1998). Thus far, however, there have not been any other systematic studies aimed at understanding RA signaling and addiction.

The current report provides converging evidence from the Upstream Regulator analysis (predicting function) of environmental enrichment and the vector knockdown of Cyp26b1 suggesting that RA signaling activity in the NAc shell increases susceptibility to drug taking. Given that every core component of the RA signaling pathway involves protein interactions with small molecules (i.e., retinoids) this pathway is a prime candidate for the development of selective small molecule inhibitors as possible pharmacotherapeutics for cocaine addiction. One advantage of choosing targets in this pathway is that nine components of this pathway have some enhancement of expression in the NAc shell (**Figure 6**), providing some degree of regional selectivity and thereby decreasing the likelihood of unwanted side effects. These regionally-enhanced components include the binding proteins Stra6, Rbp1, and Fabp5, the synthesis enzymes Adh10, Aldh1a1, and Aldh1a3, the degradation enzyme Cyp26b1, and the RA receptors Rarβ and Rxrγ. Ongoing experiments are investigating which of these targets would be most suitable for pharmacotherapeutic development.

#### CONCLUSION

Environmental factors play a significant role in individual differences in responses to drugs of abuse. Although some transcription factors, such as 1FosB and CREB, have been

#### REFERENCES


reported to mediate the protective addiction phenotype of environmental enrichment (Green et al., 2010; Zhang et al., 2014), RNA sequencing technology has produced a broader view of transcriptomic responses of the NAc in EC and IC rats after cocaine. Taken together, the discovery-based transcriptomic analyses and hypothesis-driven behavioral tests have revealed RA signaling as a novel mechanism involved in regulating the responses to both cocaine and environmental enrichment, revealing a novel pharmacotherapeutic target for the effective treatment of drug addictions.

#### AUTHOR CONTRIBUTIONS

TG, YZ, EC, and FK participated in the design of the work. TG, YZ, EC, FK, MS, DL, and XF participated in the acquisition, analysis, or interpretation of data for the work. All authors participated in the final approval of the manuscript and the review.

# FUNDING

This research has been funded by NIDA grant DA029091 (TG) and T32 DA007287 (Cunningham, EC), NINDS grant NS081121 (TG), and supported by the Clinical and Translational Science Award (CTSA) UL1TR01439.

#### ACKNOWLEDGMENTS

We would like to thank the Molecular Genomics Core at UTMB for performing the next generation RNA sequencing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2016.00119/full#supplementary-material




of retinoid binding proteins and receptors and evidence for presence of retinoic acid. Eur. J. Neurosci. 11, 407–416. doi: 10.1046/j.1460-9568.1999.00444.x

Zhang, Y., Crofton, E. J., Li, D., Lobo, M. K., Fan, X., Nestler, E. J., et al. (2014). Overexpression of DeltaFosB in nucleus accumbens mimics the protective addiction phenotype, but not the protective depression phenotype of environmental enrichment. Front. Behav. Neurosci. 8:297. doi: 10.3389/fnbeh.2014.00297

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Zhang, Kong, Crofton, Dragosljvich, Sinha, Li, Fan, Koshy, Hommel, Spratt, Luxon and Green. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Many Faces of Elongator in Neurodevelopment and Disease

#### Marija Kojic\* and Brandon Wainwright

Genomics of Development and Disease Division, Institute for Molecular Bioscience, The University of Queensland, Brisbane, QLD, Australia

Development of the nervous system requires a variety of cellular activities, such as proliferation, migration, axonal outgrowth and guidance and synapse formation during the differentiation of neural precursors into mature neurons. Malfunction of these highly regulated and coordinated events results in various neurological diseases. The Elongator complex is a multi-subunit complex highly conserved in eukaryotes whose function has been implicated in the majority of cellular activities underlying neurodevelopment. These activities include cell motility, actin cytoskeleton organization, exocytosis, polarized secretion, intracellular trafficking and the maintenance of neural function. Several studies have associated mutations in Elongator subunits with the neurological disorders familial dysautonomia (FD), intellectual disability (ID), amyotrophic lateral sclerosis (ALS) and rolandic epilepsy (RE). Here, we review the various cellular activities assigned to this complex and discuss the implications for neural development and disease. Further research in this area has the potential to generate new diagnostic tools, better prevention strategies and more effective treatment options for a wide variety of neurological disorders.

Keywords: Elongator complex, neurodevelopment, neurological disorders, tRNA modifications, translation

#### Edited by:

INTRODUCTION

Daniela Tropea, Trinity College Dublin, Ireland

#### Reviewed by:

Gregg Stanwood, Florida State University, USA Hansen Wang, University of Toronto, Canada

> \*Correspondence: Marija Kojic m.kojic@imb.uq.edu.au

Received: 28 August 2016 Accepted: 18 October 2016 Published: 01 November 2016

#### Citation:

Kojic M and Wainwright B (2016) The Many Faces of Elongator in Neurodevelopment and Disease. Front. Mol. Neurosci. 9:115. doi: 10.3389/fnmol.2016.00115 There are over 600 known neurological disorders and we are still far from fully understanding and finding treatment options for the majority of them. Many neurological disorders are inherited diseases, arising from defects in the development of the nervous system. Neurodevelopment is a complex dynamic process regulated by various genetic and environmental factors. The process is comprised of a series of intricate and coordinated events, required for converting neural precursor cells into functional neurons. Deregulation of these complex processes leads to abnormal neurodevelopment and profound neurological dysfunction. Genetic analysis of various neurodevelopmental disorders, including familial dysautonomia (FD; Anderson et al., 2001; Slaugenhaupt et al., 2001; Cuajungco et al., 2003), intellectual disability (ID; Najmabadi et al., 2011; Cohen et al., 2015), amyotrophic lateral sclerosis (ALS; Simpson et al., 2009) and rolandic epilepsy (RE; Strug et al., 2009), have identified mutations in Elongator complex subunits, suggesting that this highly evolutionarily conserved complex plays an important role in regulating neurodevelopment. In eukaryotes, Elongator is associated with diverse cellular activities including transcriptional elongation (Hawkes et al., 2002; Kim et al., 2002), cytoplasmic kinase signaling (Holmberg et al., 2002; Close et al., 2006), exocytosis (Rahl et al., 2005), cytoskeletal organization (Johansen et al., 2008), tubulin acetylation (Creppe et al., 2009) and translation (Huang et al., 2005; Esberg et al., 2006; Johansson et al., 2008; Bauer et al., 2012). In this review, we discuss the growing experimental evidence supporting the importance of Elongator in cellular processes known to be crucially important for neurodevelopment and nervous system function.

#### THE ELONGATOR COMPLEX

The Elongator complex consists of six subunits (Elp1–Elp6), which are organized into two three-subunit sub-complexes: the core sub-complex Elp123 (Elp1–Elp3), and the accessory sub-complex Elp456 (Elp4–Elp6; Otero et al., 1999; Li et al., 2001; Winkler et al., 2001). Each Elongator subunit is structurally well characterized in yeast (**Figure 1A**). Elp1 is the largest of the six subunits and acts as a scaffold for other Elongator proteins. This subunit harbors several WD40 repeats within two WD40 propeller domains, and one tetratricopeptide repeat (TRP) domain that binds specific peptide ligands and mediates protein–protein interactions (Cortajarena and Regan, 2006). An additional Elp1 domain has been recently identified: the C-terminus-localized dimerization domain (Xu et al., 2015; **Figure 1B**). Elp2 is the second largest subunit of Elongator complex with two WD40 propeller domains (**Figure 1C**; Fellows et al., 2000). Together with Elp1, Elp2 contributes to the stability of Elp123 sub-complex and integrates signals from different factors that regulate Elongator activity. Elp3 functions as the enzymatic core of Elongator, harboring two domains essential for Elongator function. These include: the S-adenosyl-L-methionine (SAM) binding domain required to catalyze a variety of radical reactions (Paraskevopoulou et al., 2006), and the histone acetyl-transferase (HAT) domain (**Figure 1D**; Wittschieben et al., 1999). Elongator subunits Elp4–6 share a RecA-like fold (**Figure 1E**) and assemble into a heterohexameric, ring-like structure. Glatt et al. (2012) showed that Elp4, 5 and 6 specifically bind to the anti-codon loop of transfer RNAs (tRNAs) and preserve ATPase activity, likely as means to control tRNA binding and release. The interaction of Elongator subunits and complex assembly has been reported by two separate studies, both proposing that the Elp456 heterohexamer bridges two peripherally attached Elp123 sub-complexes. These data indicate that Elongator is a dodecameric complex containing two copies of each of the six Elongator subunits (**Figure 1F**; Glatt et al., 2012; Lin et al., 2012). Elongator subunits are evolutionarily highly conserved from yeast to humans both in their sequence and interaction with other subunits. Conserved function across all species has been clearly demonstrated using a variety of different cross-species rescue experiments (Li et al., 2005; Chen et al., 2006, 2009). Deletion of any of the genes encoding the six subunits confers almost identical biochemical phenotypes in yeast (Fellows et al., 2000; Winkler et al., 2001; Frohloff et al., 2003), suggesting that there is a tight functional association between the proteins comprising the Elongator complex, and that the functional integrity of Elongator is compromised in the absence of any of its subunits.

FIGURE 1 | The Elongator complex architecture. (A) Schematic view of Elongator subunits (Elp1–6) and their domain structure highlighted by different colors. Structural model of: (B) Elp1 harboring two WD40 propeller domains, tetratricopeptide repeat (TRP) and DD domain; (C) Elp2 containing two WD40 propeller domains; (D) Elp3 with histone acetyl-transferase (HAT) and S-adenosyl-L-methionine (SAM) domain (E) and Elp4, 5 and 6 subunits that share a RecA fold. (F) The Elongator complex assembly in dodecamer with two Elp123 trimeric sub-complexes peripherally attached to the ring-like hexameric Elp456 sub-complex. Structural models of Elp1–6 were prepared using program Phyre2 (Kelley and Sternberg, 2009).

The Elongator complex has been reported to orchestrate multiple functions across diverse organisms. Several loss-offunction studies have illustrated a key role for this complex in development by regulating a variety of different cellular activities. For example, yeast Elp mutants are hypersensitive to high temperature and osmotic conditions, presenting with defects in exocytosis, telomeric gene silencing, DNA damage response and adaption to new growth medium (Wittschieben et al., 1999; Rahl et al., 2005; Li et al., 2009; Chen et al., 2011). In Arabidopsis thaliana, mutations in Elp subunits resulted in impaired root growth (Nelissen et al., 2005), whilst deletion of Elp3 in Drosophila melanogaster was shown to be lethal at the larval stage (Walker et al., 2011). Elongator-deficient Caenorhabditis elegans exhibit defects in neurodevelopment (Solinger et al., 2010). In mice, a transgenic Elp1 knockout resulted in embryonic lethality due to failed neurulation and vascular system formation (Chen et al., 2009). Furthermore, several human neurological disorders have been linked to a deficiency of the Elongator, which will be discussed in more detail below.

The substrate specificity of Elongator remains controversial, and the definite number of roles this complex plays in different cellular activities is still to be confirmed (also reviewed in Svejstrup, 2007; Versées et al., 2010; Glatt and Müller, 2013). The complex was initially identified in yeast as the major component of RNA polymerase II (RNAPII) holo-enzyme (Otero et al., 1999; Wittschieben et al., 1999). In vitro studies using the HeLa cell line further confirmed that Elongator directly interacts with RNAPII and facilitates transcription in a chromatin- and acetyl-CoA-dependent manner (Hawkes et al., 2002; Kim et al., 2002). The Elongator complex has also been reported to play two other distinct nuclear roles. Knockdown of Elp3 has been shown to impair paternal DNA demethylation in mouse zygotes, a process that requires the Elp3 SAM domain (Okada et al., 2010). The complex was also demonstrated to be involved in microRNA (miRNA) biogenesis in Arabidopsis, whereby Elongator is believed to play a role in coupling the transcription of primary miRNAs and their subsequent processing (Fang et al., 2015). The majority of Elongator has been found to be located in the cytoplasm, consistent with the various cellular processes assigned to this complex that take place in the cytosol. Two studies have reported that Elongator regulates cytoplasmic kinase signaling through its interaction with c-Jun N-terminal kinase (JNK; Holmberg et al., 2002; Close et al., 2006). Holmberg et al. (2002) showed that Elongator is involved in the assembly of JNK-MAPK module through the association of Elp1 with JNK, resulting in JNK activation. Rahl et al. (2005) proposed an active requirement for the Elongator complex in the establishment and maintenance of yeast cell polarity, and in exocytosis, through its interaction with Rab GTPase Sec2p. As suggested by the authors, Elongator negatively regulates Sec2p-dependent, polarized secretion through a transcription-independent pathway. Johansen et al. (2008) proposed a model in which Elp1-assisted localization of filamin A into membrane ruffles regulates neuron migration in rats. Another study linked Elongator to the process of cytoskeletal organization and cell motility by demonstrating acetylation of α-tubulin by this complex in murine cortical neurons (Creppe et al., 2009). Although the Elongator complex has been implicated in the various cellular processes described here thus far, there is accumulating evidence in the last decade to indicate that the main role of this complex is to maintain translational fidelity via regulation of tRNA modifications. In eukaryotes, U<sup>34</sup> in the anticodons of tRNALys UUU, tRNAGlu UUC and tRNAGln UUG are modified to 5-carbamoyl-methyl-uridine (ncm5U), 5-methoxy-carbonyl-methyl-uridine (mcm5U), or 5-methoxy-carbonyl-methyl-2-thio-uridine (mcm<sup>5</sup> s <sup>2</sup>U). A number of studies have reported that these modifications require the Elongator complex (Huang et al., 2005; Esberg et al., 2006; Johansson et al., 2008; Bauer et al., 2012). The methyl-group transfer to tRNA U<sup>34</sup> by Elongator likely involves a SAM-mediated mechanism in conjunction with an electron transfer from a cofactor complex Kti11/Kti13 (Boal et al., 2011; Kolaj-Robin et al., 2015). However, the precise molecular mechanism that underlies the tRNA modification by this complex is yet to be elucidated.

It is still unclear whether Elongator has many distinct functions in a cell or it regulates one process that leads to different downstream effects, via altered translation. Interestingly, Esberg et al. (2006) found that elevated levels of two tRNA species bypass all the in vivo requirements of Elongator in transcription and exocytosis. All the phenotypes of Elongator-deficient yeast cells can be suppressed by overexpression of tRNALys UUU and tRNAGlu UUC (Esberg et al., 2006). A recent study by Bauer et al. (2012) demonstrated that translation of a large number of proteins is regulated by Elongator and that cell division is under translational control of this complex. The most recent finding in Caenorhabditis elegans shows that Elongator is not a direct tubulin acetytransferase, but it rather regulates the expression of α-tubulin acetyltransferase at translational level, through its elevated AAA codon content and tRNA modification (Bauer and Hermand, 2012). The other cell activities regulated by Elongator might also be explained by its tRNA modification role and codon-dependent regulation of translation, which future experiments will elucidate.

#### ROLE OF ELONGATOR IN NEURODEVELOPMENT

Neural development is a complex process that requires neural induction, migration, differentiation, axon guidance and synapse formation. Both cell motility and the actin cytoskeleton play a central role in regulating how neuronal precursors proliferate and migrate to different parts of the developing brain. Once neural precursors have reached their final destination they undergo the process of differentiation, which involves the production and extension of axons and dendrites to form synapses, resulting in the establishment of functional neural circuits. Synapses are specialized sites of cell–cell contact where electrical signals trigger the release of neurotransmitters, which in turn activates postsynaptic receptors (Haucke et al., 2011). This highly regulated process is based on cytoskeletal organization, vesicular trafficking and polarized exocytosis. Here, we review the roles of Elongator during neurodevelopment, from transcription to translation (**Figure 2**).

Neurodevelopmental processes are dependent upon a broad number of factors regulating the expression of hundreds of genes controlling the terminal differentiation of neurons. The majority of genes activated during developmental processes are regulated at the level of transcriptional elongation (Muse et al., 2007). Elongator has been linked to the transcriptional regulation of several genes critical to various neurodevelopmental processes. Studies in human 293T (Han et al., 2007) and HeLa (Li et al., 2011) cell lines clearly demonstrate a role for the Elongator complex in the transcriptional elongation of heat shock protein 70 (HSP70), a gene that plays an important role in protecting of cells from apoptotic stimuli and in the stabilization of protein structures (Jäättelä et al., 1992; Mosser et al., 1997; Han et al., 2007). HSPs are developmentally regulated in the nervous system. HSP70 has been suggested to play an important role in both developing and adult mammalian brain and its expression within the nervous system is significantly higher compared to non-neuronal tissue (Manzerra et al., 1997). HSP70 has a neuroprotective role (attenuation of toxicity in a variety of neurodegenerative disease models (Arawaka et al., 2010)), and plays a role in axonal transport and neuronal signaling (de Waegh and Brady, 1989; Houenou et al., 1996; Thekkuveettil and Lakhotia, 1996). Elongator is also required for the activation of several genes involved in cell migration, such as those coding for the integrin receptor CD61, tenascin-C, and actin cytoskeleton modulators (Close et al., 2006). Integrins and tenascin-C are widely expressed in neuronal extracellular matrix during CNS development and they are shown to enhance neuronal precursors proliferation, migration and differentiation (Garcion et al., 2001; Flanagan et al., 2006). Taken together, these data show that the Elongator complex is responsible for the transcriptional regulation of a number of proteins that each plays a crucial role in various steps of development and maintenance of the nervous system.

The Elongator complex regulates a broad number of neurodevelopmental transcription-independent processes. Elongator activates JNK (Holmberg et al., 2002; Close et al., 2006), a stress-activated protein kinase that modulates the activity of a vast number of pathways. JNK signaling has been reported to be crucially important for neurodevelopment. JNK knockout studies in mice revealed its important role in brain morphogenesis, axonal specification and axon growth and guidance. In addition, JNK has been shown to govern synapse and memory formation (reviewed in Coffey, 2014).

Elongator is linked to synaptogenesis based on its role in vesicular trafficking and exocytosis via interacting with Rab proteins (Rahl et al., 2005). Rab proteins regulate membrane trafficking, which include vesicle formation, vesicle movement, and membrane fusion (Pfeffer, 2001). The yeast Rab protein Sec4p regulates exocytosis of post-Golgi secretory vesicles (Salminen and Novick, 1987). Sec2p is an essential protein that is recruited to sites of exocytosis, it targets the Sec4p activation event and facilitates polarized exocytosis (Walch-Solimena et al., 1997). Rahl et al. (2005) propose that the Elongator complex function in a cytosolic signal transduction pathway to regulate the localization of Sec2p and thereby the Rab activation event critical for polarized secretion. Rab proteins play a central role in neurodevelopment, by regulating the polarized neurite growth, axonal trafficking, and formation and maintenance of synapses (specific functions in synaptic vesicle exocytosis, reviewed in Ng and Tang, 2008).

The Elongator complex has also been shown to regulate migration of neural precursors through its interaction with filamin A, whereby Elongator is involved in the recruitment of filamin A in the membrane ruffles upon cell migration (Johansen et al., 2008). Elp3 was shown to localize to actin-rich domains at the edges of spreading HeLa cells (Barton et al., 2009). Filamin A organizes cortical actin filaments and dynamic three-dimensional networks in the leading edges of migrating cells and is essential for regulating the polarity of neocortical neurons during radial migration through the subventricular zone (SVZ) and intermediate zone (IZ) of the cerebrum (Nagano et al., 2004). Loss-of-function mutations in filamin A give rise to human periventricular heterotopia, a neurodevelopmental disorder caused by a failure of neurons to migrate to the cortex (Fox et al., 1998).

Acetylation of α-tubulin by the Elongator complex is yet another Elongator function that may play role in neural migration and branching (Creppe et al., 2009). In neurons the majority of cellular α-tubulin is acetylated. Creppe et al. (2009) demonstrated that lowering α-tubulin acetylation levels in microtubules through expression of α-tubulin K40A (a dominant-negative α-tubulin form that cannot be acetylated) recapitulated the migratory defects induced by Elp1/Elp3 silencing. The reduced acetylated α-tubulin levels seen upon Elongator deficiency in this study, suggest that this complex does not exclusively regulate cell motility via its association with filamin A, as migratory defects did not affect cell transition through neocortical SVZ and IZ or promote the formation of periventricular nodular heterotopia, nor via transcriptional elongation of key genes coding for proteins involved in cell migration, as the identity of these genes was cell-specific. The relationship between α-tubulin acetylation, neuron migration and branching is not yet clear but could rely on intracellular trafficking as α-tubulin acetylation is known to increase binding of motor proteins that regulate bidirectional molecular transport in axons and dendrites (Reed et al., 2006). The growing experimental and clinical evidence suggest that defective intracellular transport of specific proteins or organelles might be the hallmarks of several neurodegenerative processes, such as ALS, Parkinson's disease, Huntington disease and Alzheimer's disease (reviewed in Nguyen et al., 2010).

Elongator regulates neurotransmitter release and synapse formation, as demonstrated by one study showing that in Drosophila neurons, this complex is necessary and sufficient to acetylate Bruchpilot (BRP), an integral component of the presynaptic density where neurotransmitters are released (Mi´skiewicz et al., 2011). BRP is a large cytoskeletal-like protein with its individual strands having their N termini facing the plasma membrane, contacting Ca2<sup>+</sup> channels, and their C termini extending into cytoplasm capturing synaptic vesicles (Fouquet et al., 2009; Hallermann et al., 2010). Thus, BRP acts by concentrating synaptic vesicles at active zones, and facilitating synaptic transmission by establishing proximity between Ca2<sup>+</sup> channels and vesicles to allow efficient transmitter release (Kittel et al., 2006; Hallermann et al., 2010). Mi´skiewicz et al. (2011) suggest a model where Elongator is a BRP acetyltransferase and acetylation of BRP reorganizes its cytoplasmic tentacles, thereby regulating vesicle capturing by the C-terminal end of BRP and transport of vesicles at dense bodies. The local regulation of Elongator may enable single active zones to control neurotransmitter release and may hold the key to synaptic transmission regulation. The importance of protein acetylation in regulation of synapse composition and functionality in neurons has been demonstrated by recent study (Catarino et al., 2013). The possible role of Elongator in vesicular trafficking, exocytosis and synaptic transmission using above mentioned mechanisms remains to be tested in humans, yet evolutionary conservation of Elongator implies that it may play a similar role.

Recent studies have focused largely upon the pivotal role of Elongator in tRNA modification and these can be related to neurodevelopment in multiple ways. One recent report shows that Elongator is indeed linked to the translation of a variety of cell proteins, including those implicated in cell division (Bauer and Hermand, 2012). tRNA are more than simple adaptor molecules and have a surprising range of functions in the cell, such as tuning translation and protein expression in a tissue-specific manner and stress signaling, whereby these functions are all dependent on its modifications (Giegé, 2008; Thompson and Parker, 2009; Kirchner and Ignatova, 2015). Klassen et al. (2015) recently showed that the loss of wobble uridine modifications in the Elongator deficient yeast strain affects tRNALys UUU function and results in a reduced total protein level. Moreover, the loss of these modifications in a subset of tRNAs was shown to lead to ribosome pausing at their cognate codons in both, yeast and C. elegans (Nedialkova and Leidel, 2015). Hence, upon Elongator malfunction, the kinetics of translation is perturbed, leading to the disruption of protein homeostasis in a cell and aggregate formation, as suggested by Nedialkova and associates. Neurons are particularly sensitive to the toxicity of misfolded proteins, hence, this could be the basis of neurodegenerative pathologies. In accordance with this, it can be postulated that Elongator-dependent tRNA modification regulate critical steps in biosynthesis and homeostasis of proteins required for the development and survival of specific neurons.

#### ELONGATOR DEFECTS IN NEUROLOGICAL DISORDERS

Mutations in Elongator complex subunits have been linked to different neuronal diseases (**Table 1**). Elp1 mutations have been linked to an autosomal recessive disorder, FD (Anderson et al., 2001; Slaugenhaupt et al., 2001; Cuajungco et al., 2003). FD is among the most frequent hereditary sensory and autonomic neuropathies (Axelrod and Abularrage, 1982; Axelrod, 2002). The progressive degeneration of the sensory and autonomic nervous system in FD patients results in the following symptoms: cardiovascular dysfunction, pain insensitivity, gastrointestinal dysfunction, scoliosis, vomiting, defective lacrimation, extensive sweating and postural hypotension (Aguayo et al., 1971; Axelrod and Abularrage, 1982). Mortality in FD patients is high, and only 40% of patients survive beyond age 20 (Axelrod, 2004). FD patients have a mutation in the donor splice site of intron 20 of the Elp1 gene, resulting in aberrant splicing (Anderson et al., 2001; Slaugenhaupt et al., 2001; Cuajungco et al., 2003). This mis-splicing results in tissue-specific exon skipping and consequently reduced levels of the Elp1 full-length protein. The FD mutation is incompletely penetrant, since the full length Elp1 is still synthesized albeit at lower levels in other cell types, whereas, in the central and peripheral nervous system only the truncated product is made (Slaugenhaupt et al., 2001; Cuajungco et al., 2003). This Elp1 truncated form cannot be detected in patients due to its degradation by the nonsense-mediated decay pathway (Cuajungco et al., 2003; Slaugenhaupt et al., 2004). Hence, Elp1 levels are very low in neuronal tissue from FD patients and unchanged in lymphoblasts for instance (Slaugenhaupt et al., 2001; Cuajungco et al., 2003). Heterozygous carriers also show reduced Elp1 expression but do not develop FD, suggesting the existence of an Elp1 tissue-specific threshold for the appropriate nervous system development and functioning. Experiments in HeLa, neuronal derived Elp1 RNAi cells and FD fibroblasts, showed that the Elp1-deficient cells exhibit defects in cell motility in vitro (Close et al., 2006). More recently, one group demonstrated that the cerebrum and fibroblasts from FD patients have reduced levels of mcm<sup>5</sup> s <sup>2</sup> modifications at tRNA U<sup>34</sup> (Karlsborn et al., 2014), which was further confirmed by Yoshida et al. (2015). This suggests that reduced levels of Elp1 due to aberrant splicing result in inefficient translation in FD, once again supporting the hypothesis that the predominant role of Elongator is the regulation of the rate of translation.

ID is a disorder characterized by incomplete mental development, which results in limitations in both intellectual functioning and adaptive behavior (Schalock et al., 2010). A large-scale sequencing study identified missense mutations in the Elp2 gene to be associated with ID or related neurological disabilities. Homozygous mutations in the Elp2 gene were found in two families, each with three children suffering from moderate or severe IDs (Najmabadi et al., 2011). Cohen et al. (2015) recently reported on one more family with two brothers being affected by severe ID, spastic diplegia and self-injury. In both brothers, sequencing analysis found Elp2 missense mutations to be linked with the inheritance of this disorder. One of the two Elp2 gene mutations identified in this family was shown to have the same amino acid position as the recessive missense mutation in one of the two previously reported families (Najmabadi et al., 2011). The mechanism of this neurodevelopmental dysfunction may be related to the compromised function of the Elongator complex, due to the presence of a dysfunctional Elp2 subunit that normally acts as a signal-transducing platform. Thus, Elp2 is likely to be a novel gene that has an important role in the development of recessive cognitive disorders, such as ID.

ALS, commonly called motor neuron disease, has been associated to allelic variants of Elp3 gene (Simpson et al., 2009). ALS is a neurodegenerative disease characterized by progressive muscle weakness and atrophy due to degeneration of motor neurones in the primary motor cortex, corticospinal tracts, brainstem and spinal cord (Rowland and Shneider, 2001). Within 3 years of onset of the disease, respiratory muscle weakness generally results in death. The causative molecular pathway underlying ALS remains unknown and it is considered to be a complex disease caused by interplay between multiple mechanisms (Rowland and Shneider, 2001). Two independent studies performed by Simpson et al. (2009): a microsatellite-based genetic association study of ALS in humans and a mutagenesis screen in Drosophila melanogaster, identified allelic variants of Elp3 gene as crucial for axonal biology. In the genetic association study, Elp3 allelic variants were associated with ALS in three different populations. A mutagenesis screen in Drosophila identified Elp3 mutations that conferred abnormal photoreceptor axonal targeting and neurodegeneration. In addition to this, Elp3 knockdown in zebrafish via morpholino technology resulted in dose-dependent shortening and abnormal branching of motor neurons. Therefore, understanding


Elongator function is a promising route whereby we might gain insights into the mechanism of motor neuron degeneration in ALS.

Elp4 mutations have been linked to RE, the most common human epilepsy, with onset at 7–10 years of age (Gomez and Klass, 1983). A neurodevelopmental disorder with epileptic focus typically located in the lower motor and/or somatosensory cortex (rolandic area; Koutroumanidis, 2007), RE is characterized by centro-temporal spikes (CTS) on an electroencephalogram. The common form of RE appears to have complex genetic inheritance, where the genetic contribution to the disease is yet to be elucidated. A recent study suggests that Elp4 mutations are associated with CTS in RE families (Strug et al., 2009). A genome-wide association study in RE was conducted and fine mapping evidence pointed to the association of CTS with SNP markers in Elp4 in both, discovery and replication data sets. Beside RE, there are other neurodevelopmental disorders associated with the CTS trait, such as speech disorder (Echenne et al., 1992) developmental coordination disorder (Scabar et al., 2006) and attention deficit-hyperactivity disorder (Holtmann et al., 2003) that could be also linked to Elp4, as in the case of RE. Strug et al. (2009) hypothesized that a non-coding mutation in Elp4 gene impairs Elongator-mediated interaction of genes important for brain development, which leads to the susceptibility to seizures in neurodevelopmental disorders.

Given that Elongator has been shown to be involved in a variety of neurodevelopmental processes, this complex might be associated with a broad range of neurological disorders. Neurodegenerative diseases are mostly accompanied by transcriptional dysfunction, leading to neuronal death and in many of these cases, chromatin acetylation status is impaired by a mechanism related to the loss of HAT activity (Selvi et al., 2010). On the other hand, the Elongator association with neurological disorders could be through its tRNA modification role. Growing experimental evidence supports the link between tRNAs and neurological disorders (Kirchner and Ignatova, 2015). Several reports have associated ID and mutations in genes that encode for tRNA modification enzymes (Torres et al., 2014). A defect in Elongator complex-dependent tRNA modification could perturb translation in two different ways, by reducing the protein synthesis due to general slow-down of translation of lysine rich proteins that are found predominantly in the ribosomal machinery, or by leading to translational inaccuracy and protein misfolding. Neurons are known to be vulnerable to misfolded proteins, and the prion-like spread of pathogenic misfolded proteins probably holds a key as a general mechanism underlying

#### REFERENCES


neurodegeneration (Raj et al., 2012; Zhou et al., 2012). Warren et al. (2013) proposed the term ''molecular nexopathy'' to link accumulation of toxic protein aggregates with neural network disintegration, and argued that this can be a new paradigm of neurodegenerative diseases. Elongator-dependent regulation of translation could be an important factor contributing to this new paradigm.

#### CONCLUDING REMARKS

The Elongator complex is a dodecamer composed of six subunits (Elp1–6) with two Elp123 trimeric sub-complexes peripherally attached to the ring-like hexameric Elp456 sub-complex (Glatt et al., 2012). It is a highly conserved complex among eukaryotes and its function is dependent on the integrity of all its subunits. Although a large number of cellular functions have been attributed to this complex, the predominant role of Elongator is in regulating tRNA modification. Yet a large amount of data exists to support a possibility that Elongator acts as a transcriptional regulator. Future high-resolution structural and functional studies aimed to identifying Elongator interactors, regulators, the molecular mechanism underlying its activity and substrate specificity will shed light on the true nature of this multi-subunit complex.

Elongator activity has been linked to a host of cellular processes crucial for neurodevelopment, including cytoskeletal organization, neuritogenesis, axon growth, axonal transport, neuronal signaling and cell motility. A role for Elongator in a number of distinct neurological disorders is emerging yet. The mechanism by which the perturbation of the complex leads to the specific neuropathogenic effects is yet to be defined. Once this is elucidated, exploration of methods to complement for Elongator dysfunction offers an approach to developing effective therapies for a variety of neurological disorders.

#### AUTHOR CONTRIBUTIONS

MK: main conception and design of the work, drafting the manuscript. BW: substantial contribution to the conception and design of the manuscript, critical revision of the work presented here. Both authors approve this manuscript to be published.

#### ACKNOWLEDGMENTS

We thank Dr. Laura Genovesi and Dr. Christelle Adolphe for critical reading of the manuscript.

neurodegenerative diseases. Curr. Pharm. Biotechnol. 11, 158–166. doi: 10. 2174/138920110790909713


integral subunit of elongating RNA polymerase II holoenzyme. Mol. Cell 4, 123–128. doi: 10.1016/s1097-2765(00)80194-x


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Kojic and Wainwright. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The GABAergic Hypothesis for Cognitive Disabilities in Down Syndrome

#### Andrea Contestabile<sup>1</sup> \*, Salvatore Magara<sup>1</sup> and Laura Cancedda1,2 \*

<sup>1</sup>Department of Neuroscience and Brain Technologies, Istituto Italiano di Tecnologia (IIT), Genova, Italy, <sup>2</sup>Dulbecco Telethon Institute, Genova, Italy

Down syndrome (DS) is a genetic disorder caused by the presence of a third copy of chromosome 21. DS affects multiple organs, but it invariably results in altered brain development and diverse degrees of intellectual disability. A large body of evidence has shown that synaptic deficits and memory impairment are largely determined by altered GABAergic signaling in trisomic mouse models of DS. These alterations arise during brain development while extending into adulthood, and include genesis of GABAergic neurons, variation of the inhibitory drive and modifications in the control of neural-network excitability. Accordingly, different pharmacological interventions targeting GABAergic signaling have proven promising preclinical approaches to rescue cognitive impairment in DS mouse models. In this review, we will discuss recent data regarding the complex scenario of GABAergic dysfunctions in the trisomic brain of DS mice and patients, and we will evaluate the state of current clinical research targeting GABAergic signaling in individuals with DS.

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Enrico Cherubini, Scuola Internazionale di Studi Superiori Avanzati (SISSA), Italy Nicola Berretta, Fondazione Santa Lucia (IRCCS), Italy

#### \*Correspondence:

Andrea Contestabile andrea.contestabile@iit.it Laura Cancedda laura.cancedda@iit.it

Received: 21 December 2016 Accepted: 14 February 2017 Published: 07 March 2017

#### Citation:

Contestabile A, Magara S and Cancedda L (2017) The GABAergic Hypothesis for Cognitive Disabilities in Down Syndrome. Front. Cell. Neurosci. 11:54. doi: 10.3389/fncel.2017.00054 Keywords: GABA, down syndrome, cognitive impairment, gaba receptors, chloride homeostasis

#### INTRODUCTION

Down syndrome (DS) or trisomy 21 is the leading cause of genetically-defined intellectual disability and congenital birth defects. DS is characterized by many phenotypical features affecting almost all body systems, including developmental defects and growth delay (Nadel, 2003; Antonarakis and Epstein, 2006). In particular, brains of individuals with DS show decreased volume and reduced neuronal density in diverse brain areas (e.g., cortex, hippocampus and cerebellum; Sylvester, 1983; Coyle et al., 1986; Aylward et al., 1997, 1999; Shapiro, 2001). These alterations originate early during development (Schmidt-Sidor et al., 1990; Winter et al., 2000; Pinter et al., 2001; Larsen et al., 2008) and are possibly due to defective neuronal precursor proliferation during gestation (Contestabile et al., 2007; Guidi et al., 2008). Accordingly, the development of individuals with DS is characterized by delayed cognitive progress in infancy and childhood, leading to mild-tomoderate mental retardation with an Intelligence Quotient (IQ) ranging from 30 to 70 (Vicari et al., 2000, 2005; Pennington et al., 2003; Vicari, 2004). This scenario is additionally worsened during adulthood by further loss of cognitive abilities and the development of Alzheimer's disease (AD) by the fourth decade of life (Wisniewski et al., 1985; Mann and Esiri, 1989; Leverenz and Raskind, 1998; Teipel and Hampel, 2006).

Although cognitive impairment is the most common and severe feature of DS, other neurological and psychiatric manifestations of the disease highly impinge on the quality of life of individuals with DS and their families. In particular, the DS population shows increased frequency of anxiety (Vicari et al., 2013), clinically relevant sleep disturbance (Carter et al., 2009; Breslin et al., 2011; Angriman et al., 2015; Edgin et al., 2015; Konstantinopoulou et al., 2016; Maris et al., 2016), and hyperactivity or movement disorders (Pueschel et al., 1991; Haw et al., 1996). Finally, DS patients demonstrate increased incidence of epileptic episodes with seizure onset mainly concentrated during early life and aging (Stafstrom et al., 1991; Lott and Dierssen, 2010; Robertson et al., 2015).

The identification of possible mechanisms leading to cognitive impairment in DS has been largely conducted thanks to the analysis of different DS genetic mouse models (Dierssen, 2012). Interestingly, compelling studies in these animals indicate that altered signaling of the neurotransmitter GABA is one of the main determinants in reducing cognitive and memory functions. In particular, GABAergic dysfunctions impair synaptic plasticity and learning and memory in DS by altering optimal excitatory/inhibitory synaptic balance (Kleschevnikov et al., 2004, 2012a,b; Costa and Grybko, 2005; Fernandez et al., 2007; Deidda et al., 2014). In this review article, we will survey available data on the alteration of GABAergic signaling in DS trisomic animal models and individuals with DS. Additionally, since no effective pharmacological treatment for ameliorating cognitive deficits in DS has been found yet, we will also evaluate the current status of pre-clinical and clinical trials with GABAergic drugs in DS.

#### GABA SIGNALING IN THE BRAIN

GABA is the main inhibitory neurotransmitter in the adult and healthy brain and it acts through ionotropic and metabotropic receptors (GABAARs and GABABRs, respectively).

# GABA**<sup>A</sup>** Receptors

GABAARs are chloride-permeable ion channels. In the adult brain, the chloride gradient across the neuronal cell membrane is sustained by the exporter K-Cl cotransporter (KCC2), which maintains low intracellular chloride concentration ([Cl−]i). Thus, GABAAR opening generates an influx of negative chloride ions that hyperpolarizes the cell membrane potential and inhibits neuronal activity. Moreover, opening of GABAAR ion channels may also shunt concurrent excitatory currents (e.g., driven by the neurotransmitter glutamate), thus preventing them from bringing the membrane potential to the action potential threshold. Indeed, GABAAR opening short-circuits depolarizing synaptic currents by locally reducing the input resistance (Ben-Ari, 2002; Jonas and Buzsaki, 2007; Silver, 2010; Khazipov et al., 2015). Thus, adult GABAA-mediated transmission is physiologically hyperpolarizing and inhibitory. Conversely, in early neurodevelopment, KCC2 expression is low, and associated with high expression of the chloride importer Na-K-Cl cotransporter (NKCC1), which generates high [Cl−]<sup>i</sup> . In this condition, the chloride flux through GABAARs is outward and depolarizes the cell membrane potential (Cherubini et al., 1990; Rivera et al., 1999; Ben-Ari, 2002; Fiumelli et al., 2005). Although GABAAR-mediated depolarizing postsynaptic potentials (PSPs) will be in most cases not sufficient to reach the threshold for action-potential generation, they will be able to activate voltage-gated calcium channels (VGCCs), remove the voltage-dependent magnesium block from NMDA receptors (Leinekugel et al., 1995, 1997; Ben-Ari, 2002), and possibly add up to concurrent excitatory inputs (Gulledge and Stuart, 2003). Consequently, even an initially mild depolarizing effect of GABA can contribute to a more pronounced excitation and eventually reach the threshold for action potential firing. Thus, GABAA-mediated transmission is generally depolarizing and possibly excitatory in the immature brain, although depolarizing GABA<sup>A</sup> currents may still be inhibitory by shunting concurrent excitatory inputs. Notably, a depolarizing (rather than hyperpolarizing) GABAAR response may also occur in mature neurons depending on timing and location of GABAARs, upon relatively small difference in [Cl−]<sup>i</sup> at different cell compartments, more hyperpolarized resting membrane potentials, sustained activity or even pathological conditions (Lamsa and Taira, 2003; Khirug et al., 2008; Chiang et al., 2012; Cellot and Cherubini, 2014). Indeed, as the reversal potential for GABAAR-mediated chloride currents (ECl) sits very near to the resting membrane potential (VREST), relatively small changes in [Cl−]<sup>i</sup> are sufficient to change the polarity of GABAergic responses. In particular, if ECl is more negative than VREST, the chloride flow will be inward (hyperpolarizing). Instead, if ECl is less negative than VREST, the chloride flow will be in the opposite direction, generating depolarizing PSPs.

GABAARs are composed of five subunits, and the many subunit variants identified in mammals (six α, three β, three γ, one δ, three ρ, one ε, one π and one θ) confer different pharmacological and kinetic properties to the receptor (Olsen and Sieghart, 2008; Fritschy and Panzanelli, 2014). In the CNS, the largest population of GABAARs is composed of 2 α, 2 β and 1 γ subunits, but isoform expression and receptor composition change with subcellular localization, brain area, and development, thus assuring receptor properties set to fulfil specific neuronal needs. For example, the α1–3, β2–3 and γ<sup>2</sup> isoforms are the most represented in the adult brain, constituting synaptic GABAARs. In particular, thanks to their fast kinetics with rapid onset and desensitization, they generate phasic currents following GABA release from presynaptic vesicles with spatial and temporal accuracy. Conversely, α5- or δ-containing GABAARs are mainly extra-synaptic. The presence of the α<sup>5</sup> or δ subunits confers high GABA affinity and slow desensitization kinetics to GABAARs. Thus, α<sup>5</sup> or δ–containing receptors are able to detect ambient GABA spilled out of the synaptic cleft and generate long-lasting (tonic/extrasynaptic) currents (Böhme et al., 2004; Caraiscos et al., 2004; Farrant and Nusser, 2005; Zheleznova et al., 2009).

Phasic vs. tonic GABAA-mediated currents provide for different regulation of neuronal physiology. Phasic GABA transmission plays a relevant role in synaptic-input integration. Indeed, fast inhibition acts as coincidence detector of excitatory inputs by feed-forward inhibition: this mechanism ensures that a second input desynchronized with a first input is canceled out; thus, excitatory inputs sum up only when they are perfectly synchronized (Pouille and Scanziani, 2001). A similar although opposite integration occurs when GABA is depolarizing. GABAAR opening may generate small excitatory potentials that last after the channel closure and sum up with concurrent excitatory inputs, thus facilitating the achievement of action potential threshold (Gulledge and Stuart, 2003). While phasic inhibition participates in synaptic integration, tonic inhibition generates relatively persistent changes of input conductance, thus rightward shifting the input-output relationship. Therefore, tonic inhibition reduces the magnitude, duration and propagation length of excitatory PSPs, limiting their temporal/spatial summation possibilities and inhibiting neuronal excitability. Such effect is mainly mediated by shunting which is particularly relevant for the inhibitory action of tonic GABA transmission and less for the phasic inhibition, given the longer opening of tonic GABAARs (Ben-Ari, 2002; Farrant and Nusser, 2005; Jonas and Buzsaki, 2007; Silver, 2010; Khazipov et al., 2015).

Integration processes of excitatory inputs depend from PSPs generated from GABAARs, shunting inhibition, and also from the localization of GABAARs. For example, on the dendrites of CA1 pyramidal neurons, GABAergic synapses are mostly located on the dendritic shaft and on spines (Megías et al., 2001), thus participating to synaptic integration of two different sets of inputs: the ones already partially processed along the way from their synaptic origin to the dendritic shaft, and the ones originated in neighborhood synapses in the same spine, respectively. Notably, different cellular compartments show different [Cl−]<sup>i</sup> that critically determine GABAAR participation to input integration in a compartmentspecific fashion. Indeed, [Cl−]<sup>i</sup> is maximum in the axon and axonal initial segment, lower in the soma, and it continues to decrease gradually along dendrites (Szabadics et al., 2006; Khirug et al., 2008; Waseem et al., 2010). In summary, phasic inhibition plays as a coincidence detector and undergoes to tight spatial and temporal integration with excitatory inputs in a compartment-specific manner; tonic inhibition sets the neuronal excitability background, by affecting excitatory inputoutput relation.

# GABA<sup>B</sup> Receptors

GABABRs are Gi/o-protein-coupled receptors (GPCRs) mostly localized extra-synaptically. Their high affinity for GABA ensures that ambient GABA spilled out of synapses can activate GABABRs despite of their distance. In particular, the GABABR Gαi/<sup>o</sup> subunit can lead to adenylate cyclase inhibition and consequent reduction of cAMP levels and PKA pathway activity (Bettler et al., 2004; Padgett and Slesinger, 2010); Gβγi/<sup>o</sup> subunit, instead, inhibits voltage-gated calcium channels and opens G protein-coupled inwardly-rectifying potassium channels (GIRK/Kir3), tetramers formed by different compositions of GIRK1–4 subunits (Koyrakh et al., 2005). In particular, presynaptic GABABRs reduce vesicle release by inhibition of the voltage-gated calcium channels and by a calcium independent mechanism (Rost et al., 2011), whereas GABABR subunits found on dendrite and spine necks regulate neuronal excitability by coupling to the GIRK channels (Nicoll, 2004; Koyrakh et al., 2005). Interestingly, all GIRK subunits virtually form protein complexes with GABABR subunits in heterologous systems (David et al., 2006; Fowler et al., 2007; Ciruela et al., 2010). Nevertheless, GABABR-triggered potassium currents seem to be mostly mediated by GIRK2 containing homo- and hetero-tetramers (Koyrakh et al., 2005). GIRK channel opening elicits an outward hyperpolarizing K<sup>+</sup> current and a decrease in the input resistance (Lüscher et al., 1997; Koyrakh et al., 2005). Thus, through GIRK channels, GABABRs inhibit neuronal excitability by shunting excitatory currents, hyperpolarizing the membrane with slow inhibitory PSPs, and contributing in maintaining VREST (Lüscher et al., 1997; Koyrakh et al., 2005; Gassmann and Bettler, 2012). These effects also prevent action potential back-propagation in dendrites (Leung and Peloquin, 2006). Finally, through inhibition of postsynaptic voltage-gated calcium channels, GABABR activation also prevents dendritic calcium spikes (Chalifoux and Carter, 2011).

# GABA<sup>A</sup> and GABA<sup>B</sup> Receptor Functions Across Neurodevelopment and Adulthood

Given the variety of potential signaling dynamics by GABA<sup>A</sup> and GABA<sup>B</sup> receptors, it is not surprising that GABA may play a relevant role in different processes during healthy neurodevelopment and adulthood, as well as under pathological conditions. In particular, depolarizing and possibly excitatory GABAA-mediated transmission in early life plays key roles in promoting neurodevelopment. α5- or δ-containing extrasynaptic GABAARs are predominant at this stage when synapses are not yet formed and ambient GABA released from growth cones and astrocytes is detected by high affinity subunits (Cellot and Cherubini, 2013). Depolarizing tonic currents facilitate spiking, thus increasing the probability of firing coincidence from different cells and therefore of synaptic wiring important for synaptogenesis. Moreover, these tonic currents drive neuronal migration and maturation, axon growth, and synaptic plasticity (Ben-Ari et al., 2007; Wang and Kriegstein, 2009; Kilb et al., 2013; Luhmann et al., 2015). Also GABABRs regulate neuronal migration, maturation of pyramidal neurons, synapse formation and circuit development (Fiorentino et al., 2009; Gaiarsa and Porcher, 2013). Notably, GABABR activation is not able to elicit GIRK-mediated (hyperpolarizing) currents in early neurodevelopment, although maintaining the pre-synaptic inhibitory function on neurotransmitter release. Indeed, coupling of postsynaptic GABABRs to Kir channels is delayed during development (Fukuda et al., 1993; Gaiarsa et al., 1995; McLean et al., 1996).

Finally, in neurodevelopment, both GABAA- and GABABmediated transmission control synaptic plasticity. Indeed, GABA<sup>A</sup> receptors interfere both with the opening and the closure of the visual cortex critical period, i.e., the time window when brain plasticity can be evoked by environmental stimuli and experimental paradigms in slices and in vivo (i.e., long-term potentiation (LTP) and induction and monocular deprivation, respectively; Levelt and Hübener, 2012). Eventually, in the adult brain, GABAARs usually exert a negative regulation on hippocampal plasticity and cognition. Indeed, GABAAR-mediated inhibition suppresses LTP both in vitro and in vivo (Wigström and Gustafsson, 1986; Grover and Yan, 1999; Matsuyama et al., 2008), and benzodiazepines (positive regulators of GABAARs) or GABAAR activation impairs memory (Roth et al., 1984; Zarrindast et al., 2002; Raccuglia and Mueller, 2013). While LTP is enhanced by GABAAR blockade (Wigström and Gustafsson, 1986), long term depression (LTD) is facilitated by GABAAR activation in vitro (Steele and Mauk, 1999), thus suggesting that GABAA-mediated inhibition balances the ratio between LTP and LTD. On the other hand, GABABRmediated membrane hyperpolarization, inhibition of voltagegated calcium channels and of back-propagating spikes, as well as reduction of the cAMP/PKA generally contribute to prevent synaptic plasticity. Indeed, GABABRs generally suppress LTP and memory performance, however a dual role in plasticity regulation is mediated by auto- and hetero-GABABRs (Davies et al., 1991; Stäubli et al., 1999). Finally, GABAergic regulation of brain plasticity also involves adult neurogenesis. Indeed, tonic depolarizing GABA<sup>A</sup> responses by GABAergic Parvalbumin interneurons negatively regulate adult neurogenesis in the dentate gyrus (DG) of the hippocampus (Song et al., 2012; Pontes et al., 2013; Pallotto and Deprez, 2014).

Phasic GABAA-mediated transmission also ensures simultaneous and temporally limited inhibition, able to synchronize network activity and generate network oscillations, according to both computational models and in vitro slice experiments from juvenile (P14–27) and adult mice (Wang and Buzsáki, 1996; Mann and Mody, 2010). Moreover, GABABRs synchronize hippocampal network activity at low oscillation frequency (Scanziani, 2000; Kohl and Paulsen, 2010) and are activated during cortical up-states, contributing to their termination (Mann et al., 2009). Finally, GABA also exerts key roles in pathological conditions such as a number of neurodevelopmental disorders (Ramamoorthi and Lin, 2011; Deidda et al., 2014), epilepsy (Kaila et al., 2014a), anxiety (Nuss, 2015), and neurodegenerative diseases (e.g., AD; Li et al., 2016).

#### MOUSE MODELS OF DS

Most of the current knowledge regarding alterations of the GABAergic signaling in the DS brain has come from the study of mouse models of DS. According to a recent study utilizing the Vertebrate Genome Annotation (VEGA) database<sup>1</sup> , the human chromosome 21 (Hsa21) contains a total of 222 proteincoding genes (of which 218 map to the long arm 21q), including two large clusters of 49 keratin-associated proteins (KRTAPs; Gupta et al., 2016). The mouse genes orthologue of those mapping to the long arm of Hsa21 are distributed on three syntenic regions present on mouse chromosomes 10, 16 and 17. In particular, the distal portion of mouse chromosome 16 (Mmu16) encompasses a large (∼28 Mb) region that contains ∼55% of Hsa21 orthologous protein-coding genes (Antonarakis et al., 2004; Gupta et al., 2016). Therefore, many of the available DS mouse models have been created by genetic manipulation of this Mmu16 region. Specifically, the Ts65Dn mouse (Reeves et al., 1995) is the most widely used murine model of DS and carries an extra translocation chromosome composed of the Mmu16 syntenic region fused to the centromeric portion of Mmu17. This freely-segregating extra chromosome contains 90 non-KRTAP, Hsa21 proteincoding orthologues, plus 35 protein-coding genes (deriving from Mmu17) that are not triplicated in DS (Duchon et al., 2011; Gupta et al., 2016). Additional DS mouse models carrying a smaller triplication of the Mmu16 syntenic region are the Ts1Cje and the Ts1Rhr. Ts1Cje mice are characterized by the genomic duplication of a Mmu16 segment containing 71 Hsa21 proteincoding orthologues and translocated to the distal portion of Mmu12 (Sago et al., 1998; Gupta et al., 2016). However, the translocation resulted in the deletion of seven genes in the most telomeric segment of Mmu12 (Duchon et al., 2011). Ts1Rhr mice (Olson et al., 2004) were generated by Cre/lox chromosome engineering and carry a tandem duplication of an even smaller Mmu16 region comprising 29 Hsa21 protein-coding orthologues from the so-called ''DS critical region'' (DSCR; Delabar et al., 1993; Korenberg et al., 1994).

The vast majority of the studies on DS-related cognitive and electrophysiological abnormalities have been performed on the Ts65Dn mouse. Indeed, although the Ts65Dn model still presents issues from a genetic point of view (Gardiner et al., 2003), it recapitulates many of the phenotypic features of the human syndrome (Dierssen, 2012; Rueda et al., 2012), and it is currently the only mouse model used for preclinical identification of pharmacological interventions targeting DS cognitive impairment (Gardiner, 2014). Moreover, phenotypic comparison of different DS mouse models has suggested that the genes triplicated in the Ts65Dn mouse are major responsible for DS-related cognitive abnormalities (Rueda et al., 2012). In particular, Ts65Dn mice show severe behavioral deficits in different learning and memory tasks, including fear conditioning, T-maze spontaneous alternation, Morris water maze and object recognition tests (Reeves et al., 1995; Costa et al., 2007; Fernandez et al., 2007; Contestabile et al., 2013), and electrophysiological alterations in both synaptic transmission (Kleschevnikov et al., 2004, 2012b; Best et al., 2007, 2012; Hanson et al., 2007; Mitra et al., 2012) and hippocampal synaptic plasticity (Siarey et al., 1997, 1999; Kleschevnikov et al., 2004; Costa and Grybko, 2005; Contestabile et al., 2013). In addition, Ts65Dn mice display alterations in dendritic spine morphology (Belichenko et al., 2004; Guidi et al., 2013) and impaired neurogenesis both in the developing brain (Baxter et al., 2000; Roper et al., 2006; Chakrabarti et al., 2007, 2010; Contestabile et al., 2007, 2009) and in neurogenic niches of the adult brain (Clark et al., 2006; Bianchi et al., 2010a; Contestabile et al., 2013). Additionally, Ts65Dn mice are not spontaneously epileptic, but show increased seizures incidence in some experimental epilepsy paradigms (Cortez et al., 2009; Westmark et al., 2010; Joshi et al., 2016). Finally, similarly to DS patients, Ts65Dn mice also exhibit some sleep alterations and hyperactivity in locomotor behavior (Escorihuela et al., 1995; Reeves et al., 1995; Sago et al., 2000; Colas et al., 2008).

<sup>1</sup>http://vega.sanger.ac.uk

Although Ts1Cje and Ts1Rhr mice show some DS-related phenotypes, the extent of these features is somewhat milder and with some distinct characteristics (Rueda et al., 2012). In particular, Ts1Cje mice displayed deficits in synaptic plasticity and behavior in the T-maze test comparable to Ts65Dn mice, whereas memory performances were less severely affected in the Morris water maze and substantially spared in the novel object recognition (Sago et al., 1998; Siarey et al., 2005; Belichenko et al., 2007; Fernandez and Garner, 2007). On the other hand, Ts1Rhr mice showed impaired synaptic plasticity in the hippocampal DG, but unaffected plasticity in the CA1 hippocampal region. At the behavior level, deficits were detected in the T-maze and novel object recognition tests, but not in the Morris water maze test (Olson et al., 2007; Belichenko P. N. et al., 2009). Finally, Ts1Cje mice were found hypoactive (Sago et al., 1998), and Ts1Rhr mice mostly not different compared to WT animals for locomotor behavior (Belichenko P. N. et al., 2009).

In more recent years, systematic application of Cre/loxmediated chromosome engineering has permitted the creation of three new murine lines (Dp10, Dp16 and Dp17) individually trisomic (through a tandem duplication on the corresponding chromosome) for the three complete syntenic regions on Mmu10, 16 and 17, respectively (Li et al., 2007; Yu et al., 2010a,b). Nevertheless, although these mouse lines represent ideal models in terms of genetic triplication, they lack the extra freely-segregating chromosome that is found in most individuals with DS and in Ts65Dn mice. In fact, the presence of an extra unpaired chromosome could also have a role per se in the phenotypic consequences of trisomy by impacting on global gene expression and/or chromatin structure (Reinholdt et al., 2009; Dierssen, 2012). On the other hand, mice fullytrisomic for all Hsa21 orthologue genes can also be created by successively crossing the Dp10, Dp16 and Dp17 lines, although decreased viability and poor breading have limited the experimental studies on these triple trisomic mice (Yu et al., 2010a; Belichenko et al., 2015). Moreover, the Dp10, Dp16 and Dp17 mice have also permitted the dissection of the relative contribution of the different triplicated regions to the diverse disease phenotypes. Interestingly, while Dp16 and triple trisomic mice (Dp10/Dp16/Dp17) show behavioral and synaptic plasticity deficits comparable to the ones found in Ts65Dn mice (Yu et al., 2010a,b; Belichenko et al., 2015), and the single trisomic mice Dp10 as well as Dp17 show normal (or even enhanced) performances (Yu et al., 2010b), highlighting the importance of the Mmu16 syntenic region in DS-related phenotype.

#### GABA SIGNALING IN DS MOUSE MODELS

## GABA<sup>A</sup> Signaling and Trisomy

A first main finding regarding GABA-related phenotypes in DS came from the observation that the number of GABAergic interneurons is increased in both the cortex and the hippocampus of Ts65Dn mice (Chakrabarti et al., 2010; Pérez-Cremades et al., 2010; Hernández et al., 2012; Hernández-González et al., 2015). Specifically, this increase arises from amplified neurogenesis during embryonic development of neural progenitor cells in the medial ganglionic eminence (MGE, where most interneurons originate during development; Marín and Müller, 2014), and it is more prominent for Parvalbumin and Somatostatin-positive GABAergic interneurons (Chakrabarti et al., 2010). In line with the increase in GABAergic interneurons, the same authors reported also an increase in spontaneous GABAergic postsynaptic events in CA1 pyramidal neurons. Nevertheless, further functional analysis of GABAergic transmission by electrophysiology did not find evidence for alterations in the frequency of miniature inhibitory postsynaptic currents (mIPSC, which represent activity-independent quantal release of GABA), release probability at GABAergic synapses and evoked GABA<sup>A</sup> transmission in the hippocampal CA1 region of adult Ts65Dn mice (Chakrabarti et al., 2010; Best et al., 2012). Therefore, the increase of spontaneous GABAergic events observed in Ts65Dn mice may be due to a general enhancement in interneuron excitability rather than a specific increase in the number of GABAergic synapses. Indeed, further electron microscopy studies in the temporal cortex and hippocampus of adult Ts65Dn mice found the density of symmetric synapses (putative GABAergic) to be unaffected (Kurt et al., 2000, 2004; Belichenko P. V. et al., 2009). Similarly, immunohistochemical evaluation of GABAergic terminals confirmed comparable density between WT and Ts65Dn mice in the DG, even if the distribution of GABAergic synapses appeared altered, with a selective redistribution of GABAergic synapses from the dendrite shaft and the spine heads to the spine neck of trisomic neurons (Belichenko et al., 2004; Belichenko P. V. et al., 2009; Kleschevnikov et al., 2012b). Such selective retribution of GABAergic synapses to the spine neck was later also confirmed in the DG of Ts1Cje mice (Belichenko et al., 2007). The altered location of GABAergic synapses in DS mice may impair synaptic input integration. Indeed, GABAergic contacts on the spine neck may account for integration of spine-converging inputs, in replacement of contacts on the spine head and on the dendritic shaft that integrate inputs at local synaptic and dendritic levels, respectively. Anyhow, the increased number of GABAergic neurons seems not to be directly associated with increased synaptic contacts in the DS brain. The reason is not fully understood, but GABAergic synaptic density may normalize in adulthood for compensatory mechanisms, after defective brain development. Indeed, differently than in adult Ts65Dn mice, both mIPSC frequency and evoked GABA<sup>A</sup> transmission (but not release probability) were found larger in the CA1 region of 2 weeks old Ts65Dn mice (Mitra et al., 2012). Further support to the compensation hypothesis is the observation that an increased GABAergic synaptic density was in fact found specifically in the inner molecular layer and granular layer of the hippocampal DG of adult Ts65Dn mice (Martínez-Cué et al., 2013; García-Cerro et al., 2014; Mojabi et al., 2016). Indeed, this brain area is highly innervated by GABAergic fibers, and one could expect that developmental compensatory mechanisms may be more difficult to put in place. On the other hand, all these seemingly conflicting results may simply indicate that GABAA-mediated dysfunctions are not uniform in all areas of the trisomic brain and/or may also be simply due to age-related changes. Indeed, mIPSC frequency and evoked GABAergic transmission were found increased in the DG of adult Ts65Dn mice, but with this alteration primarily attributed to increased release probability at GABAergic terminals, rather than increased synapse number (Kleschevnikov et al., 2004, 2012b), indicating possible sub-regional differences in the density of GABAergic synapses and its functional consequences (see **Figure 1**). Additionally, mIPSC frequency was found decreased (rather than increased) in the hippocampal CA3 region of Ts65Dn mice (Hanson et al., 2007; Stagni et al., 2013). Finally, outside the hippocampus, increased excitability, enhanced release probability and decreased tonic

GABA inhibition were found in cerebellar granule cells of Ts65Dn mice (Usowicz and Garden, 2012; Das et al., 2013; Szemes et al., 2013).

Phasic and tonic GABA transmission is necessary to synchronize neuronal firing and maintain physiological network oscillation. Interestingly, gamma oscillation power is reduced in Ts65Dn hippocampal slices, while duration and frequency of up-states are reduced in Ts65Dn layer 4 of somatosensory cortex slices (Hanson et al., 2013; Cramer et al., 2015).

Regarding the expression of the different GABA receptors, no statistical difference was found in the hippocampus of Ts65Dn mice compared to WT for α1, α2, α3, α5, γ<sup>2</sup> subunits of GABAAR (Belichenko P. V. et al., 2009; Best et al., 2012; Kleschevnikov et al., 2012b). However, decreased immunostaining for β2/<sup>3</sup> subunits of GABAAR have been reported in the DG of 3 month-old Ts65Dn mice, but not at later time points, suggesting possible age-related changes (Belichenko P. V. et al., 2009). Moreover, the binding of the selective α<sup>5</sup> subtype radioligand [3H]RO0154513 was unchanged in vivo in Ts65Dn mice compared to WT, suggesting similar expression of GABAARs containing the α<sup>5</sup> subunit (Martínez-Cué et al., 2013).

As detailed above, GABAAR-mediated transmission undergoes a developmental switch from depolarizing (excitatory) to hyperpolarizing (inhibitory) during neuronal maturation and it can shift between inhibition and excitation under some circumstances and/or in particular cellular compartments (Andersen et al., 1980; Staley et al., 1995; Gulledge and Stuart, 2003; Viitanen et al., 2010; Ruusuvuori et al., 2013). Consequently, the complex relationship of ionic movements determined by GABAAR signaling and chloride transporters must be highly regulated to avoid deleterious consequences on neuronal physiology (Deidda et al., 2014). In this regard, we have recently found that NKCC1 is upregulated in the brain of both Ts65Dn mice and individuals with DS (Deidda et al., 2015b). Accordingly, we found that [Cl−]<sup>i</sup> was increased in CA1 pyramidal neurons from Ts65Dn mice and ECl was depolarized by 7.7 mV compared to WT neurons (Deidda et al., 2015b). As a result, ECl was less negative than VREST in trisomic neurons, thus predictive of outward Cl<sup>−</sup> depolarizing currents upon GABAAR activation. Accordingly, the mean firing frequency of individual CA1 neurons was increased by exogenous application of GABA in trisomic hippocampal slices, whereas it was decreased by blockade of GABAAR-mediated transmission with bicuculline, the opposite of what physiologically observed in WT neurons (Deidda et al., 2015b). Therefore, despite the complex dual excitatory and shunting/inhibitory effect of GABA following ECl depolarization, GABA<sup>A</sup> signaling was mainly excitatory in Ts65Dn neurons in our experimental conditions of prolonged bath-application of the drugs in acute brain slices. In this regard, the contribution of evoked, synaptically released or tonic GABA signaling to the depolarizing effect of GABAARs has not been studied yet in Ts65Dn slices. Similarly, experiments are needed to evaluate in vivo the strength and direction of GABA<sup>A</sup> transmission in DS. Moreover, little is known about the contribution of GABAergic signaling on network dynamics in trisomic neurons. However, one calcium imaging study on hippocampal cultures showed decreased response to bicuculline application on network burst-amplitude and duration in Ts65Dn mice, as well as decreased network bursts upon GABA application in both WT and Ts65Dn culture (Stern et al., 2015).

Despite of whether GABAAR-induced depolarization may reach the threshold for neuronal excitation or not, GABAARdriven depolarization may still work in favor of excitation more efficiently than GABAAR-driven shunting would work in favor of inhibition. Indeed, in Ts65Dn neurons the shunting effect of GABA<sup>A</sup> opening may be very mild on dendritic signal processing, since most synapses are located on the spine neck. Instead, GABA<sup>A</sup> depolarizing PSPs may generate in the spine neck for the depolarized ECl and propagate to spine heads and dendritic shaft, consequently affecting NMDA receptor and voltage-gated channel openings, and adding up to concurrent excitatory PSPs. Notably, the concomitant hyperpolarized shift of VREST in Ts65Dn neurons, due to GIRK2 triplication (Best et al., 2012), may also contribute to the depolarizing action of GABA. Anyhow, pharmacological inhibition of NKCC1 transport activity with the specific antagonist Bumetanide completely restored the hyperpolarizing and inhibitory action of GABA in Ts65Dn neurons (Deidda et al., 2015b).

# GABA<sup>B</sup> Signaling and Trisomy

Another direct link between DS genetic triplication and GABA signaling came from the discovery that the KCNJ6 gene, which encodes the subunit 2 of the GIRK channel (GIRK2/Kir3.2), maps to Hsa21 (Ohira et al., 1997; Hattori et al., 2000). The presence of an extra KCNJ6 copy in Ts65Dn mice leads to overexpression of GIRK2 mRNA and protein in hippocampus, cortex and midbrain (Harashima et al., 2006). Interestingly, also the GIRK1 protein (which is not triplicated in DS) is overexpressed in Ts65Dn brains, with normal levels of mRNA. Thus, the increased GIRK1 protein expression is most likely due to enhanced hetero-dimerization with GIRK2 subunit (which drives GIRK subunit trafficking) and downstream decreased protein turnover (Harashima et al., 2006). As a consequence, GABAB/GIRK currents are increased in cultured primary hippocampal neurons (Best et al., 2007), as well as in CA1 pyramidal neurons and in DG granule cells from acute slices of the Ts65Dn hippocampus (Best et al., 2012; Kleschevnikov et al., 2012b). This increase of GABAB/GIRK signaling in Ts65Dn neurons could strongly affect neuronal excitability and plasticity by enhancement of GIRK-mediated shunting inhibition, reduction of excitatory PSPs, back-propagating action potentials, and change of neuronal passive properties (Lüscher et al., 1997; Koyrakh et al., 2005). Anyhow, genetically restoring GIRK2 gene dosage to disomy in Ts65Dn mice (by crossing to GIRK2+/<sup>−</sup> mice) rescued the observed increase in GABABtriggered currents in CA1 pyramidal neurons (Joshi et al., 2016). In agreement with triplication of KCNJ6 and the increase of GABAB/GIRK signaling, VREST is hyperpolarized by 2.3 mV in CA1 pyramidal neurons and by 6.2 mV in DG granule cells (DGGC; Best et al., 2012; Kleschevnikov et al., 2012b). Of note, whereas no statistical difference was found in the hippocampus of Ts65Dn mice compared to WT for GABABR subunit 1, GABABR subunit 2 was slightly decreased, most likely due to a compensatory adaptation mechanism in response to the increase of GIRK2 levels (Best et al., 2012; Kleschevnikov et al., 2012b).

Apart from GABABRs, GIRKs channels may also mediate signaling from many other neurotransmitters including acetylcholine, dopamine, opioid, serotonin, somatostatin and adenosine through their metabotropic GPCRs (Lüscher et al., 1997). Therefore, GIRK2 triplication in DS may account for a large plethora of different effects depending on the neurotransmitter system involved. For example, GIRK2 channels also mediate the hypothermic response induced by administration of the serotonin 5-HT1A and 5-HT<sup>7</sup> receptor agonist 8-OH-DPAT (Costa et al., 2005). Indeed, 8-OH-DPAT hypothermic response is increased in Ts65Dn mice (Stasko et al., 2006).

## Synaptic Plasticity and GABA Signaling in DS

A large body of evidence linking altered GABA physiology to DS-related cognitive phenotypes came from the study of hippocampal synaptic plasticity in trisomic mice. LTP and LTD are widely accepted models of synaptic plasticity, characterized by the strengthening or weakening (respectively) of synaptic efficiency following specific stimulation protocols. In particular, LTP represents the cellular correlate of memory and it is necessary for different cognitive processes, including learning (Lynch, 2004; Whitlock et al., 2006; Nabavi et al., 2014). In line with the impairment in cognitive abilities characteristic of persons with DS and of DS mouse models, synaptic plasticity paradigms (both LTP and LTD) were found altered in Ts65Dn hippocampal slices (Siarey et al., 1997, 1999, 2006).

Given the already described pivotal role of GABAergic inhibition in regulating synaptic plasticity and given the defective GABAergic transmission in DS animals, researchers have evaluated the effect of GABAergic drugs on different forms of hippocampal LTP in DS mice. Interestingly, LTP occurring at Shaffer collateral-CA1 synapses (CA3-CA1 LTP) was rescued in Ts65Dn slices by acute application in the recording bath of the GABAAR antagonist picrotoxin (PTX). Similarly, potentiation at perforant path-DGGC synapses (DG-LTP) was restored by application of either PTX or the GABABR antagonist CGP55845 (Kleschevnikov et al., 2004, 2012a; Costa and Grybko, 2005), establishing a causal link between GABAergic transmission and plasticity deficits in DS. Remarkably, CA3-CA1 LTP was also rescued by application of the NKCC1 inhibitor Bumetanide, indicating the involvement of depolarizing GABAergic signaling in the impairment of synaptic plasticity in DS (Deidda et al., 2015b). An even more striking evidence came about when it was shown that DG-LTP was rescued in acute slices from Ts65Dn mice that had been chronically treated with the GABAAR blocker pentylenetetrazole (PTZ). The effect was evident for up to 1–3 months after treatment cessation (Fernandez et al., 2007). Accordingly, chronic in vivo treatment with RO4938581, a selective negative allosteric modulator of α5-containing GABAARs, rescued in vitro CA3-CA1 LTP in Ts65Dn mice, although it was not assessed after drug withdrawal (Martínez-Cué et al., 2013).

Finally, comparative evaluation of synaptic plasticity deficits in the other DS mouse models has shown that DG-LTP was similarly impaired and rescued by PTX application in Ts1Cje, Ts1Rhr, and triple trisomic (Dp10/Dp16/Dp17) mice (Belichenko et al., 2007; Belichenko P. N. et al., 2009; Belichenko et al., 2015). Conversely, CA3-CA1 LTP was decreased in Ts1Cje, Dp16 and triple trisomic mice, unchanged in Ts1Rhr and Dp10 mice, and even significantly increased in Dp17 mice, but none of these later studies evaluated the possible contribution of GABAergic signaling (Siarey et al., 2005; Olson et al., 2007; Yu et al., 2010a,b).

Of note, beside the two studies performed in primary neuronal cultures (Best et al., 2007; Stern et al., 2015), all other electrophysiological investigations of GABAergic signaling alteration in DS mouse model have been performed on acute brain slices. Therefore, further studies are needed to assess whether GABAergic dysfunctions can be fully reproduced on in vitro neuronal cultures and, most importantly, if they are also occurring in vivo in the intact brain.

### PHARMACOLOGICAL INTERVENTIONS TARGETING GABA TRANSMISSION TO RESCUE COGNITIVE DEFICITS IN DS MOUSE MODELS

Given the large body of evidence showing the involvement of GABA signaling in neurophysiology, cognition and synaptic plasticity in DS, many studies have evaluated learning and memory processes in DS mice after pharmacological interventions targeting GABAergic transmission. Although therapy with GABAAR antagonists may be hampered by possible pro-epileptic and anxiogenic side effects in patients already at increased risk for these conditions, a first seminal study evaluated the efficacy of different GABAAR antagonists (PTX, PTZ and Bilobalide) at non-epileptic doses on long-term declarative memory in the novel object recognition test in Ts65Dn mice: chronic (but not acute) blockade of GABA<sup>A</sup> transmission rescued object recognition memory after 2-weeks of treatment (Fernandez et al., 2007). Strikingly, the positive effect of PTZ (a drug previously used in the clinic) was maintained after an additional 2 months of drug withdrawal, indicating that the treatment likely induced long-term neuronal-circuit rearrangements able to sustain cognitive performance in trisomic mice. The positive effect of chronic PTZ treatment on learning and memory was further confirmed in the Morris water maze by a later study (Rueda et al., 2008). However, the same study showed a worsening effect of PTZ in Ts65Dn mice in a test for equilibrium, indicating a possible side effect of the drug. Interestingly, a follow-up study found that the effective dose of PTZ could be reduced by 10 times without compromising its efficacy, thus defining a potential safer therapeutic window with respect to the possible pro-epileptic and anxiogenic side effects of the drug (Colas et al., 2013).

Prompted by the effectiveness of PTZ treatment and by the observation that mice knock-out for the GABAAR α<sup>5</sup> subunit—which is highly expressed in the hippocampus (Wisden et al., 1992)—show increased learning and memory performance (Collinson et al., 2002), a second series of studies has evaluated the efficacy of two inverse agonists selective for the α<sup>5</sup> subunit of the GABAAR and acting as negative allosteric modulators. These two drugs had been developed by different pharmaceutical companies as cognitive enhancers: α5IA (Chambers et al., 2003) and RO4938581 (Ballard et al., 2009). Both acute and chronic treatments with these drugs were proven effective in ameliorating cognitive performance of Ts65Dn mice in both the novel object recognition and Morris water maze tests without showing pro-epileptic or anxiogenic side effects that may be associated with GABAAR antagonism (Braudeau et al., 2011a,b; Martínez-Cué et al., 2013).

Moreover, following the observation that the chloride importer NKCC1 is upregulated in DS and that GABA<sup>A</sup> transmission is depolarizing in adult Ts65Dn mice, we have recently assessed the efficacy of the NKCC1 inhibitor Bumetanide (a FDA-approved loop diuretic) on learning and memory in trisomic mice. We found that chronic NKCC1 inhibition in Ts65Dn mice was able to rescue discriminative memory in the object recognition test, spatial memory in the object location test and associative memory in the contextual fear conditioning task (Deidda et al., 2015b). Interestingly, the effect of Bumetanide was evident also after acute treatment and quickly lost upon treatment cessation, indicating that the effect of Bumetanide relied on direct NKCC1 inhibition rather than on neuronal-circuit rearrangements. These findings are in line with previous studies on the positive effect of GABAAR antagonists on learning and memory in DS mice. Indeed, GABAAR antagonists will reduce aberrant depolarizing GABA<sup>A</sup> signaling in Ts65Dn mice regardless of whether GABAAR signaling is increased in DS. On the other hand, since modulation of [Cl−]<sup>i</sup> is predicted to have only little effect on shunting inhibition, Bumetanide would preserve the ability of GABA<sup>A</sup> currents to shunt concurrent excitatory inputs, hence possibly reducing potential pro-epileptic side effects. Conversely, GABAAR antagonists together with reducing receptor transmission will also reduce the shunting inhibition, thus possibly increasing the risk of seizures and profoundly altering neuronal input integration.

With respect to increased GABABR signaling in DS, both acute and chronic treatment with the specific GABABR antagonist CGP55845 restored cognitive performance in the novel object recognition test and associative memory in the contextual fear conditioning test in Ts65Dn mice (Kleschevnikov et al., 2012a), indicating also the possible involvement of metabotropic GABA signaling in DS cognitive impairment.

#### GABA SIGNALING IN DS PATIENTS AND PATIENT-DERIVED iPSC

Less detailed information is obviously available regarding the GABAergic system in individuals with DS. Reduced brain size and decreased density of neurons are hallmarks of DS and mainly arise from reduced neurogenesis during brain development (Colon, 1972; Sylvester, 1983; Wisniewski et al., 1984; Wisniewski and Schmidt-Sidor, 1989; Wisniewski, 1990; Kesslak et al., 1994; Wisniewski and Kida, 1994; Raz et al., 1995; Aylward et al., 1997, 1999; Teipel et al., 2003; Teipel and Hampel, 2006; Contestabile et al., 2007; Guidi et al., 2008). Nevertheless, only two histological studies on DS autoptic brain samples have selectively evaluated cell counts of bona fide GABAergic interneurons. One first study found decreased number of Golgi-stained aspinous stellate (putative GABAergic) cells in the somatosensory, visual and auditory cortices (Ross et al., 1984). Accordingly, the number of Parvalbumin and Calbindin-positive non-pyramidal neurons was also reduced in the frontal and temporal cortices of DS patients (Kobayashi et al., 1990). Although such decrease of GABAergic neurons may come from a secondary effect due to Alzheimer-like degeneration in DS, brain GABA concentration has been shown to be specifically decreased in AD, but not in aging DS patients (Seidl et al., 2001). Moreover, microarray studies to evaluate changes in gene expression on human DS cortical neuronal progenitor cells (hNPCs) in culture have shown gene changes indicative of decreased GABAergic interneuron genesis (Bhattacharyya et al., 2009). Moreover, the same study showed increased expression of the α<sup>2</sup> subunit of the GABAAR, and downregulation of the α<sup>3</sup> and α<sup>5</sup> subunits in trisomic cells compared to controls (Bhattacharyya et al., 2009). This GABAAR composition favoring the α<sup>2</sup> over the α<sup>5</sup> subunits may be indicative that GABAAR opening is fastened, possibly reducing the opportunity of shunting inhibition and generating quick PSPs, a condition that may be of particular relevance if GABAAmediated transmission is depolarizing in humans. Indeed, we have shown that NKCC1 is overexpressed also in the brains of DS patients, establishing a direct parallel with the Ts65Dn model (Deidda et al., 2015b), and possibly suggesting depolarized ECl in humans as in animal models. Finally, GABA levels were found decreased or unchanged in neurochemical (Reynolds and Warner, 1988; Seidl et al., 2001; Whittle et al., 2007) and <sup>1</sup>H MRS studies (S´migielska-Kuzia and Sobaniec, 2007; S´migielska-Kuzia et al., 2010) on human trisomic brains.

The recently developed technique for reprogramming somatic cells has opened the possibility of studying patient-derived neurons as a valuable tool for modeling neurological diseases (Takahashi et al., 2007; Park et al., 2008). This approach has been also applied to DS and has permitted the generation of different induced pluripotent stem cells (iPSCs) lines. In particular, two studies have used trisomic iPSCs to assess GABAergic neurogenesis and synaptogenesis upon induction of neuronal differentiation. The results from these studies have highlighted a general impairment in synaptogenesis of DS iPSC-derived neurons that was mirrored by a decrease in the frequency of both inhibitory and excitatory spontaneous postsynaptic currents (sPSCs). However, the percentage of neurons expressing GABAergic markers, the fraction of GABAergic synapses, or the ratio of glutamatergic to GABAergic sPSCs were substantially unaffected (Weick et al., 2013; Hibaoui et al., 2014).

Overall, although the available data from human studies—while limited—seem not to support the data derived from animal research of increased GABA-mediated transmission due to the overproduction of GABAergic interneurons, further electrophysiological studies on iPSCs-derived neurons are needed. On the other hand, no study has yet assessed the occurrence of depolarizing GABA<sup>A</sup> signaling or increased GABAB-mediated transmission in human DS neurons.

# GABAergic DRUGS IN DS CLINICAL TRIALS

Due to the encouraging preclinical data highlighting a strong link between GABA signaling and DS cognitive impairment, some of the pharmacological interventions effective in DS mouse models have been translated into clinical trials on individuals with DS. In particular, although the use of PTZ on DS patients has been questioned due to the potential pro-epileptic side effects, the dosage effective in rescuing learning and memory in Ts65Dn mice is well below the epileptic dose (Colas et al., 2013). Indeed, Balance Therapeutics is conducting a clinical trial in Australia to evaluate the efficacy of PTZ (BTD-001) on individuals with DS. This COMPOSE trial (Cognition and Memory in People with DS, registered in the Australian-New Zealand clinical trial registry as ACTRN12612000652875) is a phase IB study aimed at evaluating safety, tolerability, preliminary efficacy and pharmacodynamics of BTD-001 at two different doses in adults and adolescents with DS. Recruitment for this study has been completed<sup>2</sup> and the results are eagerly awaited.

Moreover, Hoffmann-La Roche has conducted two clinical trials for assessing the efficacy of the selective negative allosteric modulator of the α5-containig GABAAR Basmisanil (RG1662/RO5186582), a derivative of RO4938581, previously shown to rescue learning and memory in Ts65Dn mice (Martínez-Cué et al., 2013). The first trial (CLEMATIS) was designed as a phase II placebo-controlled study (NCT02024789) aimed at evaluating the efficacy and safety of RG1662 at two different doses in adults and adolescents with DS. Disappointingly, although the complete results of the CLEMATIS trial have not been disclosed yet, a media release from Roche later this June has announced that the study did not meet its primary and secondary endpoints on improving cognitive functions in DS patients, and that there was no significant difference between the placebo and the treated groups<sup>3</sup> . The lack of efficacy seen in the CLEMATIS trial induced the discontinuation of the second ongoing placebo-controlled dose-investigating pediatric study (NCT02484703), aimed at evaluating pharmacokinetics, pharmacodynamics, efficacy, and safety of RG1662 in children with DS. The interruption was not decided for safety reasons, as the drug appeared to be well-tolerated and no relevant side effects were observed.

# SPECULATION FOR FUTURE DIRECTIONS IN THE RESEARCH ON DS AND GABAergic TRANSMISSION

## Possible Convergence of Different Drug Treatments on GABAergic Signaling in DS

Alternative mechanisms, apart from GABA signaling, have been shown to underline LTP and/or cognitive deficits in Ts65Dn

<sup>2</sup>http://www.anzctr.org.au

<sup>3</sup>http://www.roche.com/media/store/statements.htm

mice. Indeed, CA3-CA1 LTP and/or behavioral performances were rescued in Ts65Dn mice by a variety of manipulations including: acute application of the GluN2B-selective antagonist Ro25–6981 (Hanson et al., 2013) or of the green tea polyphenolic compound epigallocatechin-3-gallate (EGCG, an inhibitor of the DS triplicated kinase Dyrk1A; Xie et al., 2008), chronic treatments with polyphenolic green tea extracts enriched in EGCG (De la Torre et al., 2014; Catuara-Solarz et al., 2015), the monoacylglycerol lipase inhibitor JZL184 (Lysenko et al., 2014), the neuro-hormone Melatonin (Corrales et al., 2013), the Sonic Hedgehog agonist SAG1.1 (Das et al., 2013), the serotonin reuptake inhibitor Fluoxetine (Bianchi et al., 2010b; Begenisic et al., 2014; Guidi et al., 2014), or also exposure to an enriched environment (EE; Begenisic et al., 2011, 2015). Nevertheless, most of these treatments are also known to directly or indirectly modulate the GABAergic system, while for others such link has not been established yet. For example, Ro25–6981 selectively reduced the activation of GABAergic interneurons in the Stratum Radiatum of the hippocampus (Hanson et al., 2013), whereas JZL184 decreased GABAergic transmission by likely modulating presynaptic cannabinoid receptors (Katona et al., 1999; Zhang et al., 2009; Lee et al., 2015). Instead, both treatment with Fluoxetine and exposure of Ts65Dn mice to EE showed beneficial effects on LTP and memory in Ts65Dn mice, possibly by reducing release from GABAergic terminals (Begenisic et al., 2011, 2014, 2015). Indeed, Fluoxetine has been previously found to reduce GABAergic neurotransmission in the hippocampus independently from the inhibition of serotonin reuptake (Méndez et al., 2012; Caiati and Cherubini, 2013), and to decrease the levels of extracellular GABA in vivo (Maya Vetencourt et al., 2008). Additionally, Fluoxetine may also inhibit GIRK channels (Kobayashi et al., 2003; Cornelisse et al., 2007), therefore possibly normalizing enhanced GABABR signaling in Ts65Dn mice. In this regard, it is important to note that GIRK channels are also coupled to serotonin receptors 5-HT1A (Williams et al., 1988; Llamosas et al., 2015; Montalbano et al., 2015) and GIRK2 triplication can therefore impact on serotoninergic signaling in DS. Indeed, stimulation of 5-HT1A receptors in the hippocampus can reduce neuronal firing frequency and gamma oscillations through GIRK channels activation (Johnston et al., 2014). Since fluoxetine treatment reduces GIRK-mediated 5-HT1A and GABA<sup>B</sup> receptor signaling in the dorsal Raphe (Cornelisse et al., 2007), this mechanism may be involved in the therapeutic effect of fluoxetine in Ts65Dn mice. However, a possible link between changes in serotonin signaling due to GIRK2 overexpression and the effects of fluoxetine on learning and memory has not been assessed in DS mice. On the other hand, chronic treatment with polyphenolic green tea extracts enriched in EGCG decreased the expression of the GABAergic synaptic markers GAD67 and VGAT in the cortex of Ts65Dn mice (Souchet et al., 2015). Conversely, EGCG positively modulated GABAA-mediated transmission when administered acutely (Vignes et al., 2006; Park et al., 2011). Of note, a recently concluded clinical trial (TESDAD, NCT01699711) evaluating the efficacy of long-term green tea extract treatment on DS patients showed some behavioral improvements, although—as stated by the authors—below the threshold for clinical relevance in 2 out of 15 measured tests (de la Torre et al., 2016). Finally, the positive effect of Melatonin is unlikely to depend on decreased GABAergic signaling because it acts as a positive allosteric modulator of GABA<sup>A</sup> receptors (Wang et al., 2003; Scott et al., 2010; Cheng et al., 2012). Altogether, these data indicate that a number of the large plethora of effective drug treatments able to restore LTP and/or memory in Ts65Dn mice may rely on a common ground of action through the modulation of the GABAergic system.

#### Excitatory Deficits in DS

Altogether, the literature reported above indicates a key role for GABAergic signaling in neuronal network deficits in trisomic mice. However, deficits in excitatory inputs and glutamatergic transmission could also contribute to the imbalance in excitatory/inhibitory transmission in the trisomic brain. Indeed, several lines of evidence indicate delayed development and decreased production of excitatory glutamatergic neurons in the cortex (Chakrabarti et al., 2007, 2010; Tyler and Haydar, 2013; Guidi et al., 2014), DG (Lorenzi and Reeves, 2006; Contestabile et al., 2007; Bianchi et al., 2010b) and cerebellum (Baxter et al., 2000; Roper et al., 2006; Contestabile et al., 2009) of Ts65Dn mice. Accordingly, a decreased density of glutamatergic synapses was found in the cortex and hippocampus of Ts65Dn mice by electron microscopy and immunohistochemistry (Kurt et al., 2000, 2004; Chakrabarti et al., 2007; Rueda et al., 2010; Guidi et al., 2013; Stagni et al., 2013; García-Cerro et al., 2014), as well as in human DS iPSCs-derived neurons (Weick et al., 2013; Hibaoui et al., 2014). However, in vivo <sup>1</sup>H MRS evaluation of different metabolites and neurotransmitters in the Ts65Dn hippocampus showed no difference in the concentration of either GABA or glutamate (Santin et al., 2015). Despite of the evidence supporting a possible decrease of excitatory inputs, few electrophysiological studies have functionally evaluated glutamatergic signaling in trisomic mice. Decreased frequency of miniature excitatory postsynaptic currents (mEPSC) was found in the CA3 hippocampal region (Hanson et al., 2007; Stagni et al., 2013), whereas a decreased ratio of postsynaptic NMDA/AMPA-evoked responses was found in the CA1 region of Ts65Dn mice (Das et al., 2013). Interestingly, Ts65Dn mice show increased electrophysiological and behavioral response to pharmacological manipulations of NMDA receptors (Costa et al., 2007; Scott-McKean and Costa, 2011). Although a complete mechanistic explanation behind such effect will need further investigations (Costa, 2014), inhibition of NMDA transmission with the noncompetitive antagonist Memantine rescued learning and memory performance in different behavioral tests both after acute and chronic administration in Ts65Dn mice (Costa et al., 2007; Rueda et al., 2010; Lockrow et al., 2011). Following these studies, two clinical trials evaluated the efficacy of Memantine on improving cognitive functions in DS patients. Although the drug was well-tolerated, the results of the first study (NCT00240760) did not show any improvement on cognitive functions (Hanney et al., 2012). However, a pitfall of this study may be represented by the advanced age of participants. Indeed, since DS patients are at increased risk of developing Alzheimer degeneration early in life, irreversible pathological and/or degenerative changes may have been already in place by the time of treatment (Costa, 2014). A second trial (NCT01112683) on a relatively small number of young-adults with DS showed no significant difference in the primary outcome. However, some encouraging improvements were detected in a secondary measure of verbal memory (Boada et al., 2012). Overall, the lack of a comprehensive assessment of glutamatergic functions in DS mouse models will require a more detailed investigation to clearly evaluate the involvement of glutamatergic signaling in excitatory/inhibitory imbalance in the trisomic brain.

#### GABA in Neurodevelopment and Critical Period Plasticity in DS

DS is widely recognized as a neurodevelopmental disorder since many (but not all) brain deficits originate during the embryonic and early life. Since activation of both GABAAR and GABABR plays a key role in brain development (Gaiarsa and Porcher, 2013; Le Magueresse and Monyer, 2013), changes in ambient GABA (also due to increased number of GABAergic interneurons) may underline at least some of the brain alterations that originate during DS fetal life and persist into adulthood. Nevertheless, no study has so far addressed the role of aberrant GABAergic signaling in neural circuit formation in DS. In particular, it would be interesting to evaluate the timing of the depolarizing/hyperpolarizing GABA switch (Ben-Ari, 2002) and related developmental changes in GABAARsubunit expression (Succol et al., 2012) in trisomic mice. On the other hand, GABABR signaling may affect DS brain development by modulating adenylate cyclase and calcium channels. Triplication of GIRK2 is instead unlikely to directly contribute to GABAB-mediated early developmental DS brain alterations, as coupling of GABABRs with GIRK channels does not occur until the second postnatal week of life in rats (Fukuda et al., 1993; López-Bendito et al., 2003; Bony et al., 2013). Nevertheless, the possibility of premature coupling of GABABRs to GIRK2 due its overexpression in DS may still exist.

Anyhow, future studies are also needed to investigate whether modulating GABAergic signaling and/or intracellular Cl<sup>−</sup> accumulation during specific developmental periods, when brain circuits are possibly more prone to plastic changes, may result in beneficial effects in learning and memory that persist into adulthood. In this regard, it would be interesting to test the long-term effects of GABAAR inhibition (i.e., PTZ or α<sup>5</sup> negative allosteric modulator treatments) or of lowering [Cl−]<sup>i</sup> with Bumetanide during brain development. The importance of an early intervention during a likely essential period of brain development in DS is highlighted by the observation that different treatments (e.g., Fluoxetine, SAG1.1 and Choline) administered during gestation or in the early postnatal period rescued memory performances in Ts65Dn mice later in life (Bianchi et al., 2010b; Moon et al., 2010; Das et al., 2013; Guidi et al., 2013; Velazquez et al., 2013; Ash et al., 2014). Nevertheless, one important and apparently neglected aspect of early pharmacological interventions is related to possible side effects of drug treatment during development, which should in fact be carefully considered. Indeed, similarly to the beneficial effect of the drugs, adverse effects may persist into adulthood and may impact on the functionality of different organs or systems. A recent example came from the NEMO trail (NCT01434225) regarding the use of Bumetanide on newborns with hypoxic ischemic encephalopathy. Indeed, the trial was interrupted because of hearing loss adverse effect (Pressler et al., 2015). Moreover, in light of the encouraging results obtained on Ts65Dn offsprings prenatally and perinatally treated with Fluoxetine (Bianchi et al., 2010b; Guidi et al., 2014), a clinical trial has been recently announced<sup>4</sup> , aimed at assessing the efficacy of Fluoxetine administration during the prenatal (2nd trimester) and postnatal (up to 2 years of age) periods in ameliorating the developmental abilities of children with DS. However, the dosage and timing of administration should be carefully considered in the light of the fact that Fluoxetine administration during pregnancy has been associated to birth defects (Reefhuis et al., 2015), and treatment in adult patients increased seizure susceptibility (Pisani et al., 1999). Moreover, Fluoxetine treatment in rodents during embryonic and early postnatal life modifies the migration of cortical GABAergic interneurons, and—later in life—increases aggression in males as well as changes emotional and social behaviors (Kiryanova et al., 2013; Ko et al., 2014; Frazer et al., 2015; Svirsky et al., 2016).

Aberrant GABAergic transmission may also affect the criticalperiod plasticity in the visual cortex. Interestingly, this criticalperiod plasticity depends on the depolarizing action of GABA during early development and its length can be extended by reducing depolarizing GABA<sup>A</sup> signaling by treatment with Bumetanide (Deidda et al., 2015a). Therefore, an early intervention with Bumetanide may provide an extended window for neuronal plasticity in DS. Although no study has assessed critical-period plasticity in visual cortical circuits of trisomic animals, Ts65Dn mice show deficits in cortical visual evoked potentials (VEPs; Scott-McKean et al., 2010). Interestingly, exposure to EE either during development or adulthood restored cortical VEP responses in Ts65Dn mice, possibly through modulation of GABAergic transmission (Begenisic et al., 2011, 2015).

#### GABA in Other DS Symptoms, Possibly Affecting Cognition

Besides cognitive impairments, individuals with DS present a number of other symptoms (i.e., epilepsy, sleep disorders and anxiety) that affect the quality of their lives and may in turn also impinge on their cognitive abilities (Pueschel et al., 1991; Stafstrom et al., 1991; Haw et al., 1996; Carter et al., 2009; Lott and Dierssen, 2010; Breslin et al., 2011; Rissman and Mobley, 2011; Vicari et al., 2013; Angriman et al., 2015; Edgin et al., 2015; Robertson et al., 2015; Konstantinopoulou et al., 2016; Maris et al., 2016). Interestingly, epilepsy, sleep disorders and anxiety have all been associated to defective GABAergic transmission (Wagner et al., 1997; Choi et al., 2008; Rudolph and Knoflach, 2011; Möhler, 2012; Kaila et al., 2014b).

DS patients and Ts65Dn mice demonstrate increased incidence of epileptic seizures (Stafstrom et al., 1991; Westmark et al., 2010; Rissman and Mobley, 2011; Lott, 2012; Robertson et al., 2015). These observations appeared contradictory when considering that increased GABA signaling in Ts65Dn mice was expected to overall decrease neuronal network activity, and therefore reduce incidence of seizures. However, the increased incidence of seizures is in line with depolarizing GABA<sup>A</sup> signaling in DS, because the shift in GABAAR-mediated responses will also clearly impact on the excitatory/inhibitory balance and promote brain neuronal circuit hyperexcitability. Interestingly, administration of γ-butyrolactone (GBL, a prodrug for the GABABR agonist γ-hydroxybutyrate: GHB) induced epileptiform activity in Ts65Dn mice that was rescued by genetically restoring GIRK2 gene dosage to disomy (Cortez et al., 2009; Joshi et al., 2016). Although the general molecular mechanism of GHB action is still matter of debate (Bay et al., 2014; Venzi et al., 2015), it is intriguing to speculate that, in the scenario of GABA dysregulation in DS, both the GABABR-GIRK2 signaling and depolarizing GABA<sup>A</sup> signaling may play a role in GBL-induced epileptic phenotype seen in Ts65Dn mice. Indeed, GHB would further increase the already enhanced GIRK2-mediated signaling, thus abnormally drifting VREST towards hyperpolarization. This condition would emphasize the depolarizing GABA<sup>A</sup> signaling, with amplification of all depolarizing inputs and consequent trigger of epileptic activity. On the other hand, Ts65Dn mice also show increased incidence of audiogenic seizure that can be reduced by treatment with the metabotropic glutamate receptor subtype mGluR<sup>5</sup> antagonist Fenobam (Westmark et al., 2010), but not by inhibition of NKCC1 with Bumetanide (Deidda et al., 2015b). Moreover, also the characteristic hyperactivity of Ts65Dn mice (Escorihuela et al., 1995; Reeves et al., 1995; Sago et al., 2000) was reduced by the α5-containing GABAAR negative modulator RO4938581 (Martínez-Cué et al., 2013), but not by Bumetanide treatment (Deidda et al., 2015b). These observations indicate that more complex mechanisms in addition to altered GABA signaling may underline these increased seizure susceptibility and hyperactive phenotypes in Ts65Dn mice.

DS patients have also higher incidence of sleep disturbance in relation to the general population (Carter et al., 2009; Breslin et al., 2011; Edgin et al., 2015; Konstantinopoulou et al., 2016; Maris et al., 2016), and Ts65Dn mice show some sleep alterations mainly consisting in increased awaking and higher theta power in sleep EEG (Colas et al., 2008). On the other hand, Ts65Dn mice show little or no differences in circadian rhythms, which are not altered by PTZ treatment (Stewart et al., 2007; Ruby et al., 2010). Nevertheless, it is striking the observation that PTZ was effective in restoring memory performances in Ts65Dn mice only when it was administrated during the light phase of the day, but not during the dark phase, indicating a possible different circadian contribution of the GABAergic system on learning and memory in DS (Colas et al., 2013). Given the recently

<sup>4</sup>http://www.utsouthwestern.edu/research/fact/detail.html?studyid=STU%20 032014-006

identified connection between GABA signaling and memory in circadian arrhythmic animals (Ruby et al., 2008, 2013), more studies are needed to uncover this possible relationship in DS. Since GABA<sup>A</sup> signaling (and possibly Cl<sup>−</sup> homeostasis) has been repeatedly described as a key regulator of sleep and circadian rhythms (Wagner et al., 1997; Choi et al., 2008), it is possible that alterations of GABA signaling and/or expression of NKCC1 may play a role in sleep disorders in DS.

Finally, the strong involvement of GABA<sup>A</sup> signaling in anxiety disorders (Rudolph and Knoflach, 2011; Möhler, 2012) and the observation that DS patients show increased anxiety (Vicari et al., 2013) may indicate that also this aspect possibly originates, at least in part, from the alteration of GABA<sup>A</sup> signaling. However, this issue has not been addressed yet in DS mice. On the other hand, since GIRK2 knockout mice show reduced anxiety-related behavior (Pravetoni and Wickman, 2008) but GABABR knockout mice show an anxious phenotype (Mombereau et al., 2005), coupling of GIRK channels to metabotropic neurotransmitter receptor systems other than GABABRs may play a role in regulating anxiety-related behaviors in DS. For example the overexpression of GIRK2 in the midbrain of Ts65Dn mice (Harashima et al., 2006) may modulate signal transduction of dopamine receptors (Perez et al., 2006; Podda et al., 2010; Marcott et al., 2014; Zhao et al., 2016) and impact on different DS phenotypes including anxiety-related symptoms (Sim et al., 2013). Finally, a possible role for altered GABAergic transmission in impaired adult neurogenesis (Song et al., 2012; Pallotto and Deprez, 2014) and AD (Li et al., 2016)-which both eventually result in impaired cognition- is still an unexplored field of research in DS (Rissman and Mobley, 2011).

# CONCLUDING REMARKS

The genetic cause of DS has been unequivocally identified in the triplication of genes located on the human chromosome 21, although the exact group of genes and the pathological mechanisms underlying DS intellectual disability are still unclear. Despite of the limitations in reproducing in mice the exact genetic condition characterizing DS human pathology, the creation of DS mouse models with construct and face validity has been giving nevertheless a wide contribution in understanding gene-phenotype association, DS pathological processes, and in extending therapeutic prospects. Here, we evaluated the evidence pointing at a role for abnormal signaling from GABA<sup>A</sup> and GABA<sup>B</sup> receptors in the neuronal defects associated with DS, and we considered particularly the Ts65Dn mouse model of DS, one of the most largely used. The origin of defective GABAergic transmission tracks back into early neurodevelopment, with possible excitatory/inhibitory unbalance and neuronal and plasticity impairments that persist into adulthood. However, the weight of the GABAergic inhibitory vs. the glutamatergic excitatory transmission in this unbalance is still unclear. Indeed, depolarizing GABA<sup>A</sup> signaling coexists with glutamatergic excitatory alterations in Ts65Dn mice, and excitatory/inhibitory unbalance appears to be brain-region specific in DS mouse models. Moreover, possible decreased GABAergic transmission in human DS patients may seem to be discordant with the increased GABAergic transmission in Ts65Dn mice, although to date, human data are still scarce and inconclusive. Fortunately, the recent advances in the use of human-derived iPSCs may give a large contribution in the future understanding of DS neuropathology, with higher translational viability. Indeed, whether the failure of the CLEMATIS trial and limited positive results from other clinical studies can be ascribed to inefficacy of the drugs (despite the strong preclinical data), a low predictive potential of the Ts65Dn mouse model may also have played a role. For example, changes in GABAAR subunit composition specific for human DS (Bhattacharyya et al., 2009), but lacking in Ts65Dn mice, could account for the failure of pharmacological treatments with a α<sup>5</sup> selective antagonist in individuals with DS. Since the ''perfect'' mouse model does not exist yet, it will be worth for the future to assess the efficacy of each new pharmacological treatment in more than one DS mouse model, and promote parallel human and animal studies. On the other hand, the failures of clinical trials may also rest in the inadequacy of current neuropsychological tests in measuring cognitive improvements in DS (Gardiner, 2010, 2014; Fernandez and Edgin, 2016). Indeed, cognitive performance measured by tests are mostly dependent by the integration of different cognitive domains, thus the selective improvement of one of them may not be detected; on the other hand, an improvement in neuropsychological tests may not imply a perceived substantial improvement in the daily life.

In conclusion, the data reported above clearly highlight the multifaceted nature of DS brain abnormalities. These alterations represent the sum of different molecular mechanisms that most likely include impaired GABAergic transmission. Possibly, these abnormalities originate (at least in part) during development and lead to complex synaptic, physiological and circuit changes, ultimately causing cognitive deficits and other neurological manifestations. As a consequence, the still unmet need of identifying effective pharmacological interventions to alleviate DS-related cognitive impairment represents an incredible complex challenge for the future. Surely, the insights about GABA-related impairments in DS models may be of great relevance also for other neurodevelopmental disorders where defective GABAergic transmission (including depolarizing GABA<sup>A</sup> signaling) may play a pathological role (e.g., Autism, Fragile X syndrome, epilepsy, and possibly Rett syndrome; Cellot and Cherubini, 2014; Deidda et al., 2014; He et al., 2014; Khazipov et al., 2015).

# AUTHOR CONTRIBUTIONS

AC, SM and LC contributed to the conception and writing of the manuscript.

# FUNDING

This work was supported by Jérôme Lejeune Foundation (grants 254-CA2014A to AC, and 1266\_CL2014A to LC) and Telethon Foundation (grants GGP15043 to AC and GGP13187, TCP15021 to LC).

#### REFERENCES


learning deficits in a mouse model of Down syndrome. Front. Behav. Neurosci. 9:330. doi: 10.3389/fnbeh.2015.00330


deficits in Down syndrome mouse models and in humans. Mol. Nutr. Food Res. 58, 278–288. doi: 10.1002/mnfr.201300325


human chromosome 21 with orthologous mouse genomic regions. Gene 318, 137–147. doi: 10.1016/s0378-1119(03)00769-8


(Ts65Dn) mice. Brain Res. 858, 191–197. doi: 10.1016/s0006-8993(00) 01984-3


mechanism. Neuroscience 167, 678–690. doi: 10.1016/j.neuroscience.2010.02. 075


impaired by circadian arrhythmia and restored by the GABAA antagonist pentylenetetrazole. PLoS One 8:e72433. doi: 10.1371/journal.pone.0072433


**Conflict of Interest Statement**: AC and LC are named as co-inventors on International Patent Application PCT/EP2014/078561, filed on December 18, 2014, and connected US, EP, JP National Phase Applications, claiming priority to US Provisional Application US 61/919,195, priority date December 20, 2013.

The other author SM declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Contestabile, Magara and Cancedda. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# MPTP Impairs Dopamine D1 Receptor-Mediated Survival of Newborn Neurons in Ventral Hippocampus to Cause Depressive-Like Behaviors in Adult Mice

#### Tingting Zhang1,2 , Juan Hong1,2 , Tingting Di <sup>2</sup> and Ling Chen1,2 \*

<sup>1</sup> State Key Lab of Reproductive Medicine, Nanjing Medical University, Nanjing, China, <sup>2</sup> Department of Physiology, Nanjing Medical University, Nanjing, China

Parkinson's disease (PD) is characterized by motor symptoms with depression. We evaluated the influence of dopaminergic depletion on hippocampal neurogenesis process to explore mechanisms of depression production. Five consecutive days of 1-Methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) injection in mice (MPTP-mice) reduced dopaminergic fibers in hippocampal dentate gyrus (DG). MPTP-mice exhibited depressive-like behaviors later for 2–3 weeks. BrdU was injected 4 h after last-injection of MPTP. BrdU-positive (BrdU<sup>+</sup>) cells in dorsal (d-DG) and ventral (v-DG) DG were examined on day 1 (D1), 7 (D7), 14 (D14) and 21 (D21) after BrdU injection. Fewer D7-, D14- and D21-BrdU<sup>+</sup> cells or BrdU<sup>+</sup>/NeuN<sup>+</sup> cells, but not D1-BrdU<sup>+</sup> cells, were found in v-DG of MPTP-mice than in controls. However, the number of BrdU<sup>+</sup> cells in d-DG did not differ between the both. Loss of doublecortin-positive (DCX<sup>+</sup>) cells was observed in v-DG of MPTP-mice. Protein kinase A (PKA) and Ca<sup>2</sup><sup>+</sup>/cAMP-response element binding protein (CREB) phosphorylation were reduced in v-DG of MPTP-mice, which were reversed by D1-like receptor (D1R) agonist SKF38393, but not D2R agonist quinpirole. The treatment of MPTP-mice with SKF38393 on days 2–7 after BrdUinjection reduced the loss of D7- and D21-BrdU<sup>+</sup> cells in v-DG and improved the depressive-like behaviors; these changes were sensitive to PKA inhibitor H89. Moreover, the v-DG injection of SKF38393 in MPTP-mice could reduce the loss of D21-BrdU<sup>+</sup> cells and relieve the depressive-like behaviors. In control mice, the blockade of D1R by SCH23390 caused the reduction of D21-BrdU<sup>+</sup> cells in v-DG and the depressive-like behaviors. Our results indicate that MPTP-reduced dopaminergic depletion impairs the D1R-mediated early survival of newborn neurons in v-DG, producing depressive-like behaviors.

Keywords: Parkinson's disease (PD), dopaminergic receptor (DR), depression, neurogenesis, hippocampal dentate gyrus (DG)

#### Edited by:

Andrew Harkin, Trinity College Dublin, Ireland

> Reviewed by: Subhabrata Sanyal, Biogen Idec, USA Sungjin Park, University of Utah, USA

\*Correspondence: Ling Chen lingchen@njmu.edu.cn

Received: 09 July 2016 Accepted: 28 September 2016 Published: 13 October 2016

#### Citation:

Zhang T, Hong J, Di T and Chen L (2016) MPTP Impairs Dopamine D1 Receptor-Mediated Survival of Newborn Neurons in Ventral Hippocampus to Cause Depressive-Like Behaviors in Adult Mice. Front. Mol. Neurosci. 9:101. doi: 10.3389/fnmol.2016.00101

# INTRODUCTION

Parkinson's disease (PD) is a neurodegenerative disorder characterized by motor symptoms and a progressive loss of dopaminergic neurons (Rodriguez-Oroz et al., 2009). Major depression is present in approximately 30–50% of PD patients (Bower et al., 2010; Nègre-Pagès et al., 2010). However, the underlying mechanisms remain unclear.

The hippocampal volume is smaller in patients with depression than in normal subjects of the same-age (Videbech and Ravnkilde, 2015). Electroconvulsive therapy in refractory depression can increase hippocampal volume. Growing evidence suggests that neurogenesis continues throughout adulthood within the hippocampal dentate gyrus (DG; Toni et al., 2008). The newly generated neurons can integrate into hippocampal circuitry, thereby maintaining a functional structure, which is required for mood control and antidepressant efficacy (Petrik et al., 2012). The deficits in adult neurogenesis in the postmortem brains of PD patients have recently attracted much attention. For example, Höglinger et al. (2004) have reported the deficits in the hippocampal neurogenesis of PD patients. The hippocampal cell proliferation capacity is decreased in PD patients (Guiard et al., 2009). A reduction in neuronal precursor cells has been observed in the subgranular zone (SGZ) of patients with PD (Borta and Höglinger, 2007).

The hippocampus and the midbrain dopaminergic neurons of the ventral tegmental area (VTA) form a functional loop (Lisman and Grace, 2005). Dopaminergic fibers, originating mainly from VTA (Gasbarri et al., 1997), directly contact the newborn cells in DG, but are sparse in the granule cell layer or adjacent hilus (Höglinger et al., 2004). Neuronal precursor cells in the hippocampal DG of adult mammals express all dopaminergic receptors (DRs), which receive dopaminergic afferents (Takamura et al., 2014). Dopamine (DA) has been reported to play an important role in the regulation of endogenous neurogenesis in the adult mammalian brain. The activation of D1-like receptors (D1R) promotes the survival of newborn cells in the adult hippocampus (Takamura et al., 2014). Winner et al. (2009) have observed that a D2-like receptor (D2R) agonist in DA-depleted rats increases the proliferation of neural precursor cells in the subventricular zone (SVZ) but not in SGZ. A large body of evidence has established that the DA depletion in rodents decreases the cell proliferation and survival of neuronal precursor cells in DG (Höglinger et al., 2004; Khaindrava et al., 2011). However, other data have revealed that the DA depletion increases or has no effect on cell proliferation (Oizumi et al., 2008; Park and Enikolopov, 2010).

In rodents and primates, the changes in afferent and efferent connectivity along the longitudinal axis of the hippocampus suggest distinct functions of the dorsal and ventral hippocampus. The dorsal DG (d-DG) receives projections arising in the lateral and caudomedial portion of entorhinal cortex, whereas the ventral DG (v-DG) receives inputs from the rostromedial region of the entorhinal cortex (Dolorfo and Amaral, 1998). The innervation density of dopaminergic plexuses is very high in the ventral hippocampus, but is low in the dorsal portion (Bjarkam et al., 2003). 1-Methyl-4-phenyl-1,2,3,6 tetrahydropyridine (MPTP) is commonly used to prepare an animal model of PD as it selectively induces cell death of dopaminergic neurons (Schober, 2004). The MPTP lesion can be used for behavioral studies of affective disorders in C57BL/6 mice (Gorton et al., 2010). To investigate whether the DA depletion in PD brain impairs the hippocampal neurogenesis to cause the depression, we in this study examined the affective behaviors and the neurogenesis process of d-DG and v-DG (including the proliferation of stem cells, the differentiation of progenitor cells and the survival of newborn neurons) in mice treated with MPTP (MPTP-mice), and explored the correlation between hippocampal neurogenesis deficits and depressive-like behaviors. The present study provides evidence that MPTP-reduced dopaminergic afferents impairs the D1Rmediated early survival of newly generated neurons in v-DG, which may be responsible for the production of depressive-like behaviors.

# MATERIALS AND METHODS

#### Mice

The use of animals was approved by Institutional Animal Care and Use Committee of Nanjing Medical University and was performed in accordance with the experimental animal guidelines of Laboratory Animal Research Institute. Eight-weekold male C57BL/6 mice (24–26 g; Oriental Bio Service Inc., Nanjing, China) were used at the beginning of the experiment. The mice were maintained under constant environmental conditions (temperature 23 ± 2 ◦C, humidity 55 ± 5% and 12:12 h light/dark cycle) in Animal Research Center of Nanjing Medical University with free access to food and water.

#### Drug Administration

The mice received an intraperitoneal (i.p.) injection of MPTP (25 mg/kg, measured as free base; Sigma-Aldrich, St. Louis, MO, USA) once a day for five consecutive days (Crocker et al., 2003). BrdU (Sigma-Aldrich) was dissolved freshly in 0.9% saline to make 10 mg/ml solution just before injection. Mice were given three injections of BrdU (50 mg/kg, i.p.) at intervals of 6 h. SKF38393 (Tocris, UK), SCH23390 and quinpirole (Sigma-Aldrich) were dissolved in sterile saline; L-sulpiride and H89 (Sigma-Aldrich) were dissolved in 1.0% DMSO (Sutton and Caron, 2015). The mice were treated daily with the injection (i.p.) of SKF38393 (10 mg/kg), H89 (1 mg/kg; Seyedi et al., 2014), quinpirole (2 mg/kg), SCH23390 (0.5 mg/kg) or L-sulpiride (15 mg/kg). For the v-DG injection of drugs, the mice were anesthetized with chloral hydrate (400 mg/kg, i.p.) and placed in a stereotaxic instrument (Stoelting, Wood Dale, IL, USA). The scalp was incised and a small hole (2 mm diameter) was drilled in the skull using a dental drill. Guide cannulas (26-Gauge, Plastics One, Roanoke, VA, USA) were implanted into the bilateral v-DG (2.3 mm posterior, 1.3 mm lateral and 2.0 mm ventral to Bregma; Zhou et al., 2011b). On day 2 after surgery, the dummy cannulas were removed from the guide cannula, and then replaced by infusion cannulas (30 Gauge). The infusion cannula was connected by polyethylene tubing (PE 10; Becton Dickinson, Sparks, MD, USA) with a stepper-motorized micro-syringe (Stoelting, Wood Dale, IL, USA). SKF38393 (4.8 nmol), SCH23390 (0.6 nmol) and H89 (0.8 nmol) were diluted with ACSF and infused daily in a volume of 0.2 µl/side (Lai et al., 2008; Nasehi et al., 2011). After 2% Evans-blue (2.5 µl) was injected, the mice were killed by an overdose of chloral hydrate, and coronal sections (100 µm) were cut using a cryostat to validate the injection-site. Control mice were given an equal volume of vehicle.

### Behavioral Examination

A single cohort of animals was used for the following test sequence: open-field test (OFT) → Forced swim test (FST) → Tail suspension test (TST; Zhou et al., 2014). All behavioral data were captured by a video-monitor and analyzed using TopScan Lite 2.0 (Clever Sys., Reston, VA, USA).

OFT was performed in a cuboid Plexiglass box (60 cm × 60 cm × 40 cm). Total distance traveled (mm/6 min) was recorded.

FST was performed in a glass cylinder (300 mm high, 280 mm in diameter) that was filled with water (25 ± 1 ◦C) to a height of 20 cm. Total immobility time during a 6 min test was scored. Mice were considered to be immobile when they stopped struggling and moved only to remain floating in the water, while keeping their heads above water.

TST was performed by using adhesive tape to attach the tail to a rod that was 60 cm above the floor. Trials were conducted for a period of 6 min, during which the immobility time was recorded.

Morris water maze task (MWM) was performed in a blackcolored plastic pool (diameter = 120 cm) at 20 ± 1 ◦C. A cylindrical platform (diameter = 7 cm) was placed 0.5 cm below the surface of water. Each mouse was randomly released from four different quadrants and allowed to swim for 90 s. Four trials were conducted each day with an intertrial interval of 30 min. Average swimming speed (m/s) and latency (s) to reach the platform were scored for all trials. If a mouse could not reach the platform within 90 s, the experimenter gently assisted the mouse onto the platform and allowed it to remain there for 15 s.

# Immuno-Staining and Quantification

#### BrdU Immuno-Staining

Mice were anesthetized with chloral hydrate (400 mg/kg, i.p.) and perfused transcardially with 4% paraformaldehyde on day 1 (D1), day 7 (D7) , day 14 (D14) and day 21 (D21) after the lastinjection of BrdU. Coronal hippocampal slices (40 µm) were cut using a vibrating microtome (Microslicer DTK 1500; Dousaka EM Co, Japan). The d-DG (AP: −0.94 to −2.30) and the v-DG (AP: −2.46 to AP: −3.80) were each harvested on the basis of the coordinates of the Paxinos and Franklin atlas of the mouse brain (O'Leary et al., 2012). The free-floating sections were treated with 3% normal goat serum, and then incubated with mouse anti-BrdU antibody (1:1000, Millipore, Billerica, MA, USA) at 4 ◦C overnight. The sections were incubated in biotin-labeled goat anti-mouse IgG antibody (1:500, Bioworld Technology, Inc., St. Louis Park, MN, USA) for 2 h. Immunoreactivities were visualized using an avidin-biotin horseradish peroxidase complex (Vector Laboratories, Inc., Burlingame, CA, USA). BrdU-positive (BrdU+) cells in SGZ and granule cell layer of every 5th section (200 µm apart) were counted using a conventional light microscope (DP70, Olympus Optical, Tokyo, Japan). The number of BrdU<sup>+</sup> cells per section was multiplied by five to obtain the total number per DG (Sha et al., 2015).

#### BrdU and NeuN or GFAP Double Immuno-Staining

The sections were incubated with rat anti-BrdU antibody (1:200, Abcam, Cambridge, UK), which was detected using CY3 labeled anti-rat IgG antibody (1:200, Millipore) and mouse antineuronal nuclei (NeuN) antibody (1:500, Millipore), which was detected using fluorescein-labeled anti-mouse antibody (1:50, Millipore) or mouse anti-GFAP antibody (1:200, Millipore), which was detected using a FITC-labeled anti-mouse antibody (1:50, Millipore). BrdU+/NeuN<sup>+</sup> or GFAP<sup>+</sup> cells were observed using a confocal laser-scanning microscope (Leica, Heidelberg, Germany). The number of BrdU+/NeuN<sup>+</sup> or GFAP<sup>+</sup> cells per section (200 µm apart) was multiplied by 5 to obtain the total number.

#### Doublecortin (DCX) or Tyrosine Hydroxylase (TH) Immuno-Staining

The free-floating sections (40 µm) were incubated with goat anti-DCX antibody (1:500, Santa Cruz, CA, USA) or chicken anti-TH (1:1000, Abcam) at 4◦C overnight, and then in biotinlabeled rabbit anti-goat IgG antibody (1:500, Bioworld) or goat anti-chicken IgG antibody (1:500, Santa Cruz) for 2 h at room temperature. Density of doublecortin positive (DCX+) cells was expressed as the number per mm length along SGZ (Sha et al., 2015). Tyrosine hydroxylase positive (TH+) cells were counted using a stereological system, which consisted of a light microscope with a CCD camera (DP70), a motorized specimen stage for automatic sampling and a computer running Microbrightfield Stereo Investigator software (Microbrightfield, Williston, VT, USA; Hong et al., 2015).

# Western Blot Analysis

After the dorsal and ventral hippocampal slices were harvested, the DG regions were micro-dissected and stored at −80◦C until assayed. Protein was extracted and the concentration of protein was determined using a bicinchoninic acid (BCA) protein assay kit (Pierce, IL, USA). Protein (40 µg) was separated by 10% acrylamide denaturing gels (SDS-PAGE) and transferred to membranes. The membranes were incubated with rabbit anti-protein kinase A (PKA) phosphorylation (1:1000, Millipore) and rabbit anti-cyclic AMP-response element binding protein (CREB) phosphorylation (1:1000, Santa Cruz, CA, USA). Then, the membranes were incubated with horseradish peroxidase-labeled goat anti-rabbit antibody (1:5000, Santa Cruz, CA, USA), and developed using an enhanced chemiluminescence detection kit (Millipore). After visualization, the blots were stripped by incubation in stripping buffer for 15 min, and then incubated with rabbit anti-PKA antibody (1:1000, Millipore) and rabbit anti-CREB antibody (1:1000, Santa Cruz, CA, USA). An internal control was performed using mouse anti-β-actin antibody (1:2000, Cell Signaling, Danvers, MA, USA). Western blot bands were scanned and analyzed with the ImageJ analysis software package (NIH).

#### Reverse Transcription-Polymerase Chain Reaction (RT-PCR)

Total RNA of v-DG and d-DG was isolated using TRIzol reagent (Invitrogen, Camarillo, CA, USA) and reverse-transcribed into cDNA using a PrimeScript RT reagent kit (Takara, China) for quantitative PCR (ABI Step One Plus, Foster City, CA, USA) in the presence of a fluorescent dye (SYBR Green I; Takara). The relative expression of genes was determined using the 2−11ct method with normalization to the GAPDH expression. The primer sequences of D1R and D2R mRNA were designed as described in a previous publication (Kim et al., 2010).

#### Statistical Analysis

The group data were expressed as the means ± standard error (SEM). All statistical analyses were performed using SPSS software, version 16.0 (SPSS Inc., Chicago, IL, USA). Differences among means were analyzed using the Student's t-tests or one/two-factor analysis of variance (ANOVA) or repeatedmeasures ANOVA, followed by a post hoc Bonferroni test. Differences of p < 0.05 were considered statistically significant.

#### RESULTS

#### MPTP Causes Depression-Like Behaviors in Mice

MPTP injection had a marked effect on the dopaminergic neurons in VTA (F(4,49) = 6.767, p < 0.001; **Figure 1A**). In comparison with the control values, the number of TH<sup>+</sup> cells on day 1 (D1, p < 0.05), day 7 (D7, p < 0.01), day 14 (D14, p < 0.01) and day 21 (D21, p < 0.01) after the last-injection of MPTP (post-MPTP) was significantly reduced.

The affective behaviors were examined on days 5–7, days 12–14 and days 19–21 post-MPTP (**Figure 1B**). MPTP-mice had a tendency to travel a shorter total distance in the OFT (**Figure 1C**), but the group comparison with control mice failed to reach statistical difference (F(1,54) = 0.825, p = 0.368). Depression-like behaviors were examined by FST and TST. There was a main effect of MPTP-injection for the immobility time in FST (F(1,54) = 16.130, p < 0.001; **Figure 1D**) and TST (F(1,54) = 15.360, p < 0.001; **Figure 1E**). Notably, MPTPmice showed a progressive prolongation of immobility time in FST and TST from day 13–14 (FST: p < 0.05; TST: p < 0.05) to day 20–21 (FST: p < 0.01; TST: p < 0.01) post-MPTP.

The spatial memory was further examined using the place learning of the MWM on days 22–26 post-MPTP. Repeatedmeasures ANOVA revealed that the escape latency of hidden platform progressively decreased with training days in all groups (F(4,72) = 59.308, p < 0.001; **Figure 1F**), which was not affected by the MPTP-injection (F(1,18) = 0.049, p = 0.828).

### MPTP Impairs the Survival of Newborn Neurons in Ventral DG

To investigate the mechanisms underlying the MPTP-induced depression-like behaviors, we examined the dopaminergic innervations and the hippocampal neurogenesis in d-DG and v-DG (**Figure 2A**). In control mice, the density of TH positive (TH+) fibers in v-DG was higher than that in d-DG (**Figure 2B**). In comparison with those of controls, the TH<sup>+</sup> fibers in v-DG were clearly decreased on day 7 post-MPTP. The BrdU was injected starting from 4 h post-MPTP. BrdU<sup>+</sup> cells were examined on day 1 (D1), day 7 (D7), day 14 (D14) and day 21 (D21) after the BrdU-injection (**Figure 2C**). The number of BrdU<sup>+</sup> cells (D1→D7→D14→D21) in either d-DG (F(3,72) = 57.836, p < 0.001; **Figure 2D**) or v-DG (F(3,72) = 64.856, p < 0.001; **Figure 2E**) were decreased progressively in all groups. MPTP injection markedly reduced the number of BrdU<sup>+</sup> cells in v-DG (F(1,72) = 7.800, p = 0.007), but not in d-DG (F(1,72) = 3.497, p = 0.066). In comparison with controls, the MPTP-mice had significantly fewer D7-BrdU<sup>+</sup> cells (p < 0.01), D14-BrdU<sup>+</sup> cells (p < 0.01) and D21-BrdU<sup>+</sup> cells (p < 0.01), but not D1-BrdU<sup>+</sup> cells (p > 0.05), in v-DG. However, the numbers of D1-BrdU<sup>+</sup> cells (p > 0.05), D7-BrdU<sup>+</sup> cells (p > 0.05), D14-BrdU<sup>+</sup> cells (p > 0.05) and D21-BrdU<sup>+</sup> cells (p > 0.05) in d-DG had no significant difference between MPTP-mice and control mice.

Doublecortin (DCX), a microtubule-associated protein, is specifically expressed in immature neuroblasts and progenitor cells (Francis et al., 1999). To evaluate the survival of young newborn neurons, we examined the number of DCX<sup>+</sup> cells in the hippocampal DG on day 7 post-MPTP. The results showed that the number of DCX<sup>+</sup> cells was significantly reduced in v-DG (p < 0.01, n = 10; **Figure 2F**), but not in d-DG (p > 0.05, n = 10), compared with those in control mice.

To examine the differentiation of precursor cells, we further examined the number and relative proportions of D21-BrdU<sup>+</sup> cells expressing NeuN, a mature neuron marker, or glial fibrillary acidic protein (GFAP), a glial marker. The number of D21- BrdU+/NeuN<sup>+</sup> cells in v-DG of MPTP-mice was less than that in control mice (p < 0.01, n = 10; **Figure 2G**), whereas there was no difference in d-DG between the both (p > 0.05, n = 10). In contrast, the number of D21-BrdU+/GFAP<sup>+</sup> cells in v-DG (p > 0.05, n = 10; **Figure 2H**) or d-DG (p > 0.05, n = 10) did not differ between MPTP-mice and control mice. In v-DG of MPTP-mice, the ratio of BrdU+/NeuN<sup>+</sup> cells to total D21-BrdU<sup>+</sup> cells (53.79 ± 5.02%) was lower than that in control mice (76.60 ± 5.48%, p < 0.01, n = 10), whereas the ratio of BrdU+/GFAP<sup>+</sup> cells (15.06 ± 1.09%) was higher than the ratio in control mice (10.54 ± 0.74%, p < 0.01, n = 10).

# MPTP Reduces D1R-Activated PKA-CREB Signaling

The RT-PCR analysis showed that the levels of D1R (v-DG: p > 0.05; d-DG: p > 0.05, n = 8; **Figure 3A**) and D2R mRNA (v-DG: p > 0.05; d-DG: p > 0.05, n = 8; **Figure 3B**) were

(A) Representative photomicrographs of tyrosine hydroxylase (TH) immuno-staining in ventral tegmental area (VTA; a circular area is indicated by white line). Scale bars = 40 µm. Stereological counts of TH<sup>+</sup> cells throughout VTA on day 1 (D1), 7 (D7), 14 (D14) and 21 (D21) after the last-injection of MPTP (post-MPTP). <sup>∗</sup>p < 0.05 and ∗∗p < 0.01 vs. control mice (one-way ANOVA). (B) Time chart of the experimental procedure. Horizontal open arrow indicates the time post-MPTP (day). Vertical open arrows indicate the time of open-field test (OFT), forced swim test (FST) and tail suspension test (TST) and Morris water maze task (MWM). (C) Each point represents the total distance traveled within 6 min in OFT on day 5, 12 and 19 post-MPTP, respectively. (D,E) The points indicate the immobility time in FST and TST on days 6–7, days 13–14 and days 20–21 post-MPTP, respectively. <sup>∗</sup>p < 0.05 and ∗∗p < 0.01 vs. control mice (two-way ANOVA). (F) Each point represents the mean latencies (sec) to reach the hidden platforms on days 22–26 post-MPTP.

not altered on day 7 post-MPTP. Compared with the control levels, the phosphorylation of PKA (p-PKA: p < 0.01; **Figure 3C**) or CREB (p-CREB: p < 0.01, n = 8; **Figure 3D**) in v-DG was reduced on day 7 post-MPTP, which was reversed by the administration of D1R agonist SKF38393 for 2 days (p < 0.01, n = 8). The PKA inhibitor H89 could block the D1R-reversed CREB phosphorylation in v-DG of MPTP-mice (p < 0.05, n = 8). The administration of D1R antagonist SCH23390 for 2 days to control mice decreased the phosphorylation of PKA (p < 0.01) and CREB (p < 0.01, n = 8) in v-DG. Consistently with the results of a report by Zhao et al. (2013), the activation of D2R by quinpirole led to a decline in the phosphorylation of PKA (p < 0.05) and CREB (p < 0.05, n = 8) in v-DG of control mice, whereas the D2R antagonist L-sulpiride had no effects (p > 0.05, n = 8). The treatment of MPTP-mice with quinpirole showed a tendency to decrease the phosphorylation of PKA and CREB in v-DG, but the difference had no statistical significance (p > 0.05, n = 8). In contrast, MPTP-mice did not show the changes in the levels of the PKA and CREB phosphorylation (p > 0.05, n = 8) in d-DG compared with those of controls, which was not affected by SKF38393 (p > 0.05, n = 8) or quinpirole (p > 0.05, n = 8). Furthermore, the phosphorylation of PKA or CREB in d-DG of control mice was not altered by the application of SCH23390 (p > 0.05, n = 8) or L-sulpiride (p > 0.05, n = 8).

#### MPTP Reduces the D1R-Mediated Survival of Newborn Neurons

The newborn cells in the adult hippocampal DG express functional D1R and D2R (Takamura et al., 2014). To test whether the down-regulation of D1R-activated PKA-CREB signaling affects the survival of young newborn neurons, MPTP-

mice were treated with SKF38393 or quinpirole daily on days 2–7 after the BrdU-injection, and control mice received the treatment with SCH23390, L-sulpiride or quinpirole. Then, we examined the number of D7- and D21-BrdU<sup>+</sup> cells, respectively (**Figure 4A**). As shown in **Figures 4B,C**, the treatment with SKF38393 in MPTP-mice not only rescued the loss of D7-BrdU<sup>+</sup> cells (p < 0.01, n = 10), but also corrected the reduction of D21-BrdU<sup>+</sup> cells (p < 0.01, n = 10) in v-DG, which was sensitive to H89 (p < 0.05, n = 10). However, the administration of quinpirole to MPTP-mice had no effects on the reduced D7- or D21-BrdU<sup>+</sup> cells in v-DG (p > 0.05, n = 10). Interestingly, the treatment of control mice with either SCH23390 (p < 0.01, n = 10) or quinpirole (p < 0.05, n = 10), but not L-sulpiride (p > 0.05, n = 10), caused a decrease in the number of D7-BrdU<sup>+</sup> cells in v-DG, which was associated with the reduction of D21-BrdU<sup>+</sup> cells (SCH23390: p < 0.01, n = 10; quinpirole: p < 0.05, n = 10). In contrast, the number of D7- or D21- BrdU<sup>+</sup> cells in d-DG of control mice was not affected by SCH23390, quinpirole or L-sulpiride (p > 0.05, n = 10). However, the application of SKF38393 on days 16–21 after the BrdU-injection had no effect on the reduced D21-BrdU<sup>+</sup> cells in v-DG of MPTP-mice (p > 0.05, n = 10; **Figure 4D**). At the same time, the treatment of control mice with SCH23390 did not change the number of D21-BrdU<sup>+</sup> cells in v-DG (p > 0.05, n = 10). Additionally, the administration of SKF38393 for 6 days post-MPTP could increase the number of DCX<sup>+</sup> cells in v-DG (p < 0.01, n = 10; **Figure 4E**), while the activation of D2R by quinpirole could not (p > 0.05, n = 10). Similarly, the treatment of control mice with SCH23390 (p < 0.01, n = 10), but not L-sulpiride (p > 0.05, n = 10), caused a reduction in the number of DCX<sup>+</sup> cells in v-DG.

#### Association of MPTP-Impaired Neurogenesis With Depressive-Like Behaviors

To explore whether the impaired neurogenesis in MPTP-mice is related to their depressive-like behaviors, MPTP-mice were treated with SKF38393 or quinpirole for 18 days starting from the first-injection of MPTP, and control mice received SCH23390 and quinpirole for 18 days (**Figure 5A**). To avoid the direct effects of drugs, the affective behavioral tests were performed on days 6–7 after the end of the drugs administration. The results showed that the treatment of MPTP-mice with SKF38393, but not quinpirole, corrected the prolongation of the immobility time in FST (p < 0.05, n = 10; **Figure 5B**) and TST (p < 0.01, n = 10; **Figure 5C**), which was blocked by H89 (FST: p < 0.05; TST: p < 0.05, n = 10). In addition, the administration of SCH23390 (FST: p < 0.01; TST: p < 0.01, n = 10) or quinpirole (FST: p < 0.05; TST: p < 0.05, n = 10) to control mice caused an increase in the immobility time of FST and TST.

To determine further the relationship between the MPTPimpaired neurogenesis and the depressive-like behaviors, MPTPmice were treated with the v-DG injection of SKF38393 (**Figure 5E**), and control mice received the v-DG injection of SCH23390 and H89 for 18 days (**Figure 5D**). Similarly, the v-DG injection of SKF38393 in MPTP-mice could prevent the prolongation of the immobility time in FST (p < 0.05, n = 10; **Figure 5F**) and TST (p < 0.01, n = 10; **Figure 5G**) and the reduction of D21-BrdU<sup>+</sup> cells in v-DG (p < 0.01, n = 10; **Figure 5H**). The v-DG injection of SCH23390 and H89 in control mice caused the prolongation of the immobility time in FST (SCH23390: p < 0.01, n = 10; H89: p < 0.05, n = 10) and TST (SCH23390: p < 0.05, n = 10; H89: p < 0.05, n = 10) and the loss of D21-BrdU<sup>+</sup> cells in v-DG (SCH23390: p < 0.01, n = 10; H89: p < 0.05, n = 10).

#### DISCUSSION

The present study provides evidence that the MPTP-induced decline of dopaminergic afferents in v-DG impairs the D1Rmediated early survival of newly generated neurons, and this deficit in hippocampal neurogenesis is associated with depressive-like behaviors.

The adult DG contains at least two types of proliferating immature cells: GFAP<sup>+</sup> radial glia-like cell (B-cell) and GFAP<sup>−</sup> cell (C-cells). The B-cells, as stem cells, generate frequently dividing transit-amplifying C-cells. Asymmetrically divided C-cells are more likely to be progenitor cells and can differentiate into neural precursors or DCX<sup>+</sup> neuroblasts (A-cells; Zhao et al., 2008). D2R is predominantly expressed in C-cells, while A-cells express both D1R and D2R (Höglinger et al., 2004). The activation of D2R has been reported to stimulate C-cell proliferation (Yang et al., 2008). The number of D1- BrdU<sup>+</sup> cells in v-DG or d-DG of MPTP-mice was not altered, although the dopaminergic neurons in VTA were reduced by approximately 25% on day 1 post-MPTP. The microtubuleassociated protein DCX expression is specific to newly generated neurons, but not to glial cells and undifferentiated cells, which reaches a peak during the second week and is downregulated concomitantly with the appearance of NeuN (Rao and Shetty, 2004). DCX, as an immature neuron marker, is used to evaluate the survival of young newborn neurons (Kim et al., 2009). MPTP-mice had a significant decrease in the DCX<sup>+</sup> cells in v-DG, which could be rescued by the activation of D1R rather than D2R. In addition, the blockade of D1R caused a decrease of the DCX<sup>+</sup> cells in v-DG of control mice. The number of D7-BrdU<sup>+</sup> cells in v-DG of MPTP-mice was reduced. Although fewer D14- or D21-BrdU<sup>+</sup> cells were found in v-DG of MPTP-mice than those in control mice, the percentage of reduced BrdU<sup>+</sup> cells (D7 → −25% D14 → −57% D21) in MPTP-mice did not differ from control mice (D7 → −20% D14; D14→ −52% D21). Therefore, it is likely that the MPTP-induced DA depletion mainly impairs the early survival of newborn neurons in v-DG. Khaindrava et al. (2011) have reported that DA depletion causes deficits in the survival of newborn cells in DG. Our pharmacological experiments support this idea by showing that in MPTP-mice, the D1R agonist when administered on days 2–7 after BrdU-injection rescued the loss of D7- or D21-BrdU<sup>+</sup> cells, but when administered on days 16–21 after BrdU-injection, it failed to alter the reduced D21-BrdU<sup>+</sup> cells in MPTP-mice. Moreover, the blockade of D1R on days 2–7 after BrdU-injection reduced the number of D7- and D21-BrdU<sup>+</sup> cells in control mice. The absolute number of D21-BrdU+/NeuN<sup>+</sup> cells and proportion of these

FIGURE 4 | Involvement of D1R and D2R in MPTP-impaired survival of newborn neurons. (A) Time chart of the experimental procedure. The drugs were administered on days 2–7 after the BrdU-injection (black part). (B,C) Bar graph shows the mean number of D7-BrdU<sup>+</sup> cells and D21-BrdU<sup>+</sup> cells in d-DG and v-DG of MPTP-mice treated with SKF38393 (SKF), H89 or quinpirole (Quin) or control mice treated with SCH23390 (SCH), L-sulpiride (L-s) or Quin. <sup>∗</sup>p < 0.05 and ∗∗p < 0.01 vs. control mice; ##p < 0.01 vs. MPTP-mice; <sup>+</sup>p < 0.05 vs. MPTP-mice treated with SKF. (D) Bar graph shows the mean number of D21-BrdU<sup>+</sup> cells in v-DG when the drugs were administered on days 16–21 after the BrdU-injection. ∗∗p < 0.01 vs. control mice. (E) Bar graph shows the mean number of DCX<sup>+</sup> cells in v-DG when the drugs were administered on days 2–7 post-MPTP. ∗∗p < 0.01 vs. control mice; ##p < 0.01 vs. MPTP-mice.

cells relative to the total D21-BrdU<sup>+</sup> cells were reduced in v-DG of MPTP-mice. The number of BrdU+/GFAP<sup>+</sup> cells in v-DG of MPTP-mice did not differ from control mice, and the ratio of BrdU+/GFAP<sup>+</sup> cells was higher than the ratio in control mice. Thus, it is conceivable that the MPTP-induced DA depletion does not affect the differentiation of precursor cells. On the other hand, the number of BrdU<sup>+</sup> cells, DCX<sup>+</sup> cells, BrdU+/NeuN<sup>+</sup> cells or BrdU+/GFAP<sup>+</sup> cells in d-DG did not differ between control and MPTP-mice, which was not affected by the activation or blockade of D1R and D2R. Consistently with the report by Bjarkam et al. (2003), there was few dopaminergic projections in d-DG, indicating that the neurogenesis process in d-DG may be insensitive to the dopaminergic regulation.

The administration of D1R agonist to MPTP-mice could rescue the loss of D7-BrdU<sup>+</sup> cells in a PKA-dependent manner, but the D2R agonist did not. The major signaling cascaded by D1R activation is the cAMP-PKA pathway in the hippocampus (Arnsten and Dudley, 2005). PKA-CREB signaling is one of the most important pathways for promoting neuron regeneration in DG (Miyamoto et al., 2009). The activation of PKA induces the phosphorylation of CREB at Ser-133, thereby facilitating the gene transcription of key molecules such as c-Fos, Jun-B, Bcl-2, GDNF and neurotrophins to regulate neuronal survival and regeneration. Inhibition of phosphodiesterase-4, an enzyme that catalyzes the hydrolysis of cAMP, stimulates the activation of CREB and increases the survival of newborn cells (Nakagawa et al., 2002). In this study, we observed the decline of PKA and CREB phosphorylation in v-DG of MPTP-mice without changes in the expression of D1R. In v-DG of MPTP-mice, the decrease of PKA and CREB phosphorylation could be corrected by the D1R agonist and the protection of D1R agonist on the D7-BrdU<sup>+</sup> cells was blocked by the inhibition of PKA. The blockade of D1R caused the loss of D7-BrdU<sup>+</sup> cells and decline of PKA and CREB phosphorylation in v-DG of control mice. The D2R antagonist haloperidol has been shown to decrease (Wakade et al., 2002), increase (Kippin et al., 2005) or not affect (Malberg et al., 2000) hippocampal neurogenesis. Our results showed that the activation of D2R reduced the PKA and CREB phosphorylation and the D7-BrdU<sup>+</sup> cells in v-DG of control mice. Therefore, it is proposed that the MPTP-induced DA depletion through the down-regulation of D1R-induced PKA-CREB signaling impairs the early survival of newborn neurons in v-DG.

It is widely accepted that the hippocampal neurogenesis is required for mood control (Petrik et al., 2012) and cognitive performance (Rola et al., 2004). Selectively impairing the adult neurogenesis by telomerase inhibitor could cause the depressionlike behaviors in mice (Zhou et al., 2011a). Irradiation of young animals impairs hippocampal neurogenesis that is associated with cognitive deficits (Rola et al., 2004). The newly generated neurons are integrated into the neuronal circuitry within 3–4 weeks after birth (Zhao et al., 2006) to regulate hippocampal output (Ming and Song, 2011). Importantly, we have observed the depressive-like behaviors in MPTP-mice for 2–3 weeks later after DA depletion. The timing of depressive-like behaviors in MPTP-mice seems to be coincident with the loss of mature newborn neurons. Interestingly, MPTP-mice did not appear affected by the spatial cognitive deficits, although they showed depressive-like behaviors. One possible explanation may be a difference in the regions of impaired neurogenesis in hippocampal DG. The idea may be supported by the facts that the lesions in the dorsal hippocampus affect spatial learning and memory (Moser et al., 1995), while lesions of the ventral hippocampus lead to anxiety and depressive-like behaviors (McHugh et al., 2004). The antidepressants may exert their behavioral effects by increasing neurogenesis in v-DG (Banasr et al., 2006). In particular, the v-DG injection of SKF38393 in MPTP-mice not only reduced the loss of D21-BrdU<sup>+</sup> cells but also relieved the depressive-like behaviors. In control mice, the v-DG injection of either SCH23390 or H89 caused the reduction of D21-BrdU<sup>+</sup> cells in v-DG, which was companied by the production of depressive-like behaviors. These results indicate a possible cause-and-effect relationship between the MPTP-impaired neurogenesis and the MPTP-induced affective disorder. In addition, the ventral hippocampus is involved in regulation of the hypothalamic-pituitary-adrenal axis (Herman et al., 1995), buffering the stress response (Snyder et al., 2011). The basolateral amygdaloid complex receives the hippocampal afferents (Supcun et al., 2012). Hippocampal dysfunction affects the synaptic plasticity of the basolateral amygdaloid complex, which is important for the acquisition and consolidation of fear memories (Goosens and Maren, 2002) and the extinction of learned fear (Dalton et al., 2012).

Forebrain DA circuitry has been studied by two largely independent researchers: a nigrostriatal DA system that originates in the substantia nigra (SN) and a mesolimbic DA system that originate in the VTA. MPTP selectively destroys the dopaminergic neurons in the SN and VTA (Lu et al., 2014; Hong et al., 2015; the present study). Target neurons in the primary terminal field of the SN are striatal medium spiny neurons. It is well known that the MPTP-induced loss of dopaminergic neurons in SN pars compacta through the striatal DA depletion impairs the motor functions including slowness, rigidity, resting tremor and postural instability (Fahn, 2003). The projection fields of VTA DA neurons are the hippocampus or medial prefrontal cortex, which contributes to the emotional regulation (Russo and Nestler, 2013). Tye et al. (2013) have reported that the selective inhibition of VTA DA neurons acutely induces multiple distinct depression-like behaviors. Furthermore, the dopaminergic fibers from VTA dopaminergic neurons have been reported to contact directly with the newborn cells in DG (Höglinger et al., 2004). Taken together, the results in the present study give an indication that the MPTP-impaired VTA dopaminergic neurons leading to the dopaminergic deficiency in v-DG, impairs the D1R-mediated early survival of newborn neurons.

The present study provides evidence that the MPTP-induced dopaminergic depletion impairs the D1R-mediated early survival of newly generated neurons in v-DG; this impairment of hippocampal neurogenesis is associated with the production of the depressive-like behaviors. Thus, D1R agonists may be candidate substrates for treating PD depression.

#### AUTHOR CONTRIBUTIONS

In this study, TZ performed the immuno-staining, western blotting and all statistical analysis. JH finished the behavioral examination and RT-PCR analysis. TD carried out the animal care. LC designed the experiment and finished the manuscript.

# FUNDING

This study was supported by the National 973 Basic Research Program of China (2014CB943303) and the National Natural Science Foundation of China (81471157; 81671253).

#### REFERENCES


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Zhang, Hong, Di and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Synaptic Dysfunction in Alzheimer's Disease and Glaucoma: From Common Degenerative Mechanisms Toward Neuroprotection

Chiara Criscuolo<sup>1</sup> , Carlotta Fabiani <sup>1</sup> , Elisa Cerri <sup>1</sup> and Luciano Domenici 1,2 \*

<sup>1</sup>Neuroscience Institute of the National Council of Research (CNR), Pisa, Italy, <sup>2</sup>Department of Biotechnological and Applied Clinical Sciences (DISCAB), University of L'Aquila, L'Aquila, Italy

Alzheimer's disease (AD) and glaucoma are two distinct multifactorial neurodegenerative diseases, primarily affecting the elderly. Common pathophysiological mechanisms have been elucidated in the past decades. First of all both diseases are progressive, with AD leading to dementia and glaucoma inducing blindness. Pathologically, they all feature synaptic dysfunction with changes of neuronal circuitry, progressive accumulation of protein aggregates such as the beta amyloid (Aβ) and intracellular microtubule inclusions containing hyperphosphorylated tau, which belongs to microtubule associated protein family. During an early phase of degeneration, both diseases are characterized by synaptic dysfunction and changes of mitogen-activated protein kinases (MAPK). Common degenerative mechanisms underlying both diseases are discussed here, along with recent results on the potential use of the visual system as a biomarker for diagnosis and progression of AD. Common neuropathological changes and mechanisms in AD and glaucoma have facilitated the transfer of therapeutic strategies between diseases. In particular, we discuss past and present evidence for neuroprotective effects of brainderived neurotrophic factor (BDNF).

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Mario Buffelli, University of Verona, Italy Aline Stephan, University of Strasbourg, France Claudia Perez-Cruz, Centro de Investigaciones y Estudios Avanzados (CINVESTAV), Mexico

\*Correspondence:

Luciano Domenici domenici@in.cnr.it

Received: 04 October 2016 Accepted: 14 February 2017 Published: 27 February 2017

#### Citation:

Criscuolo C, Fabiani C, Cerri E and Domenici L (2017) Synaptic Dysfunction in Alzheimer's Disease and Glaucoma: From Common Degenerative Mechanisms Toward Neuroprotection. Front. Cell. Neurosci. 11:53. doi: 10.3389/fncel.2017.00053 Keywords: beta amyloid, tau, synaptic dysfunction, visual system impairment, brain-derived neurotrophic factor

# INTRODUCTION

Alzheimer's disease (AD) is a neurodegenerative progressive disease of the elderly leading to dementia. The world Alzheimer report (Alzheimer's Disease International) of 2015 indicated that 46.8 million people worldwide are living with dementia; this number is expected to double every 20 years (Reitz and Mayeux, 2014). There are two forms of AD.


Vascular diseases such as hypertension and brain ischemia (Pluta et al., 2013; Origlia et al., 2014), diabetes (Zhao et al., 2008; Adzovic and Domenici, 2014), traumatic brain injury (Van Den Heuvel et al., 2007) and mood disorders (Tsuno and Homma, 2009) represent risk factors for SAD; the most important risk factor for SAD is aging. Neuropathological changes of AD include classical hallmarks such as the senile plaques formed by beta amyloid (Aβ), neurofibrillary tangles (NFTs) and dystrophic neurites containing hyperphosphorylated tau (Serrano-Pozo et al., 2011). Histopathological findings showed that the entorhinal cortex and hippocampus are affected during the earliest phases of AD (Braak and Braak, 1991, 1993). Recent studies have highlighted an apparent dichotomy between the progress of histopathological findings in different brain areas and the occurrence of visual dysfunctions in AD (Moschos et al., 2012; Sivak, 2013; Yamasaki et al., 2016). More than 60% of people with AD have a decline in one or more visual function(s). AD causes vision impairment by affecting the eye (Graw, 2015) and by deterioration of visual functions, from retina to visual cortex. In particular, clinical studies support a link between cognitive performance and visual dysfunction even at an early stage of AD; the gradual loss of memory is frequently accompanied by alteration of visuospatial function in animal models and AD patients (Rizzo et al., 2000; Crow et al., 2003). Recently, we showed an impairment of the visual responses arising from the magnocellular streams of visual processing (Sartucci et al., 2010), suggesting that large retinal ganglion cells (RGCs) are primarily affected in AD. Often, retinal involvement is an early occurrence in AD (Sivak, 2013) as also suggested by results obtained by the use of ocular imaging techniques such as the optical coherence tomography (Moschos et al., 2012).

Glaucoma is a group of eye disorders, currently recognized to be multifactorial and progressive, leading to reduction in vision and eventual blindness. Usually glaucoma affects the older population. Over 60 million people worldwide were estimated to be affected with glaucoma, and bilateral blindness from the disease was estimated to be present in 4.5 million people with glaucoma (Quigley and Broman, 2006). Glaucoma is characterized by the progressive degeneration of RGCs till cell death, optic nerve (ON) atrophy, impairment of visual function with visual field defects and finally, loss of neurons in the lateral geniculate nucleus and primary visual cortex. A generally accepted theory suggests an initial insult to the axons of RGCs in the ON head region (Quigley, 2011). Several types of glaucoma are known; these can be divided in primary and secondary. Primary open-angle glaucoma (POAG) is considered to be the most common subtype of glaucoma. There are two main types of POAG: one that occurs with an intraocular pressure (IOP) higher than normal and represents 60%–70% of the total POAG (Coleman and Brigatti, 2001; Quigley and Broman, 2006) and the other occurring with a normal or lower IOP. Thus, ocular hypertension represents the major risk factor for glaucoma onset and progression. In the presence of ocular hypertension, there is no obvious damage to the RGCs and ON or evidence of visual field changes. However, we recently showed that retinal responses to patterned visual stimuli (pattern electroretinogram, P-ERG) together with Brn3 (POU-domain transcription factor) expressed in RGCs are altered during ocular hypertension in a murine model of glaucoma (Domenici et al., 2014). It is reasonable to think that ocular hypertension applies some stress to RGCs and their circuitry during a phase preceding the degeneration of RGCs and ON atrophy.

# COMMON FEATURES BETWEEN AD AND GLAUCOMA

Both diseases affect older populations and are neurodegenerative, chronic and progressive leading to irreversible cell death. AD and glaucoma are thought to share, at least in part, some common features such as the Aβ accumulation/aggregation, tau aggregation and hyperphosphorylation. Both diseases are characterized by early changes of neuronal circuitry and phosphorylation of mitogen-activated protein kinases (MAPK) followed by inflammatory process, glial reaction, reactive oxygen species production, oxidative stress and mitochondrial abnormalities, propagation of neurodegenerative processes leading to cell death. Both diseases are characterized by common features such as synaptic dysfunction and neuronal cell death at the level of the inner retina (Sartucci et al., 2010; Sivak, 2013). Taken together, all these observations suggest similar degenerative mechanisms between AD and glaucoma. Glaucoma is recognized as a disease frequently associated with AD and aging (Martinez et al., 1982; Chandra et al., 1986; Tamura et al., 2006; Tsolaki et al., 2011; Jefferis et al., 2013; Elyashiv et al., 2014). Conflicting data have been reported among the different studies carried out to compare AD frequency in glaucoma patients (Kessing et al., 2007; Yochim et al., 2012). Thus, a clear epidemiologic relationship between glaucoma and AD remains elusive.

#### AMYLOID-DEPENDENT MECHANISMS IN AD AND GLAUCOMA

Senile plaques in AD comprise aggregates of Aβ filaments, dystrophic neurites and mitochondrial abnormalities (Hirai et al., 2001; Serrano-Pozo et al., 2011). Aβ peptides start to be generated in considerable amounts by the cleavage of APP due to sequential activation of β- and presenilin catalytic site of γ-secretases. Aβ can be found in different compositions of monomers, oligomers, or fibrils (Stromer and Serpell, 2005); in particular, increasing Aβ tends to form monomers which aggregate into oligomers, prefibrillar assemblies (protofibrils) and amyloid fibrils in a concentration-dependent manner. Toxic Aβ peptides are formed by 36–43 amino acids; the 42 amino acid peptide (Aβ42) is one of the most neurotoxic amyloidogenic fragment and represents the chief component of senile plaques. Increasing Aβ level tends to form oligomers of different length and composition (Stromer and Serpell, 2005), which are toxic for neuronal cells (Origlia et al., 2008). In particular, Aβ oligomeric extracts from cerebral cortex of AD patients (Shankar et al., 2008) and synthetic Aβ formed by dimers and trimers (Origlia et al., 2008, 2009) are capable of inhibiting long term synaptic plasticity in the form of long term potentiation (LTP), which is involved in learning/memory in hippocampus and parahippocampal cortices (Nabavi et al., 2014); this represents an early step in the disease progression. Oligomeric Aβ inhibits LTP through phosphorylation of p38 MAPK (Criscuolo et al., 2014). Increasing synthetic Aβ concentration affects synaptic transmission, AMPA current (Origlia et al., 2010). Thus, accumulation of extracellular Aβ is likely to result in progressive synaptic dysfunctions and cognitive impairment (Selkoe, 2002). Increasing Aβ induces phosphorylation of MAPKs in neuronal and non-neuronal cells along with the induction of pro-inflammatory cytokines, such as the IL-β (Origlia et al., 2010). Activation of receptors such as the receptor for advanced glycation end products (RAGE) by Aβ accounts for progress of synaptic dysfunction, development of inflammatory and, possibly, oxidative processes, leading cells to degenerate (Origlia et al., 2010).

The death of the RGCs in glaucoma is preceded by a remodeling of retinal circuitry, RGC dendritic arbor and axonal atrophy (Jakobs et al., 2005). Although it is not yet clear what initiates the death of RGCs in glaucoma, recent experimental evidence indicates that functional alterations caused by impairment of synapses (Della Santina et al., 2013) precede the degeneration of RGCs. A possible interpretation would be that the impairment of synapses is followed by consequences for RGC viability. Similarly to AD, enhanced retinal levels of soluble Aβ may act by impairing the synaptic circuitry and retrograde trafficking of neurotrophic factors in the ON axons (Poon et al., 2011; Gupta et al., 2014). Interestingly, high IOP, which characterizes an early reversible phase of retinal degeneration, leads to Aβ induction (McKinnon et al., 2002). The hypothesis can be advanced that stressor stimuli such as the high IOP in glaucoma may cause accumulation of Aβ in the retina, contributing to synaptic progressive dysfunction in the inner retina and impairment of visual responses. However, whether low amounts of Aβ in the form of oligomers result in synaptic toxicity with detriment of visual retinal responses during ocular hypertension is still lacking. Concerning cell death, progressive degeneration of RGCs is associated with increased production of Aβ (McKinnon et al., 2002; Guo et al., 2007). Hence, it is assumed that Aβ leads not only to neuronal cell impairment in AD but also to retinal cell impairment and degeneration in glaucoma in general. Yan et al. (1996) showed that the RAGE activated by Aβ is able to induce neuronal toxicity; it is known that oligomeric Aβ induces phosphorylation of p38 MAPK through RAGE activation during an early phase of degeneration in AD (Origlia et al., 2008). Interestingly, hyperphosphorylation of p38 MAPK characterizes the degeneration of RGCs in glaucoma, mainly during an early phase with high IOP and synaptic dysfunction (Fabiani et al., 2016), whether RAGE is involved in the mechanisms of glaucoma onset and progression is still unanswered.

#### TAU-DEPENDENT MECHANISMS IN AD AND GLAUCOMA

The tau protein is expressed from the gene known as microtubule associated protein tau (MAPT) on chromosome 17. Tau is highly expressed in neurons and is abundant in axons (Lee et al., 2001). Tau facilitates assembly and the stabilization of microtubule polymers (Cleveland et al., 1977; Caceres and Kosik, 1990), modulating microtubule dynamics. Thus, under physiological conditions tau is mainly expressed within neurons. Hyperphosphorylated, insoluble and filamentous tau proteins were shown to be the main component of NFTs, a pathological hallmark of AD and other tauopathies (Lee et al., 2001). NFTs accumulate inside the cells, disrupting the intracellular transport system. Cytoskeletal changes are visible as dystrophic neurites, pre-tangles, NFTs in the cell bodies of affected neurons in AD (Iqbal et al., 1984). Interestingly, phosphorylation of tau potentiates MAPK activation similarly to Aβ and tau is one of p38 MAPK substrates (Corrêa and Eales, 2012). Studies on cell viability have shown that misfolding of tau leads to the aggregation of tau and the appearance of toxic tau species in the extracellular space (Gómez-Ramos et al., 2006, 2008). The endogenous intracellular tau may be released as aggregates to the extracellular space upon neuron degeneration (Gómez-Ramos et al., 2006). Extracellular tau could be toxic by increasing intracellular calcium into neighboring neurons (Gómez-Ramos et al., 2008). The presence of extracellular tau can be due to other causes, for example exocytosis; the N-terminal region of tau seems to be required for its secretion (Kim et al., 2010). Tau can also be released into the extracellular space, as oligomers (Saman et al., 2012). Indeed, neuronal toxicity may be caused by tau aggregates, even small and soluble aggregates in the form of oligomers, which have been identified in AD brains (Sahara et al., 2008). Recently, it has been shown that oligomeric extracellular tau is able to interact with cell receptors resulting in synaptic dysfunction and signaling propagation that could contribute to onset of neurodegeneration (Fá et al., 2016). These observations point to the involvement of extracellular tau aggregates as one of the main agent in the neuron-to-neuron propagation of neurofibrillary pathology and progression of toxicity in AD.

Tau was found to be expressed in RGCs and it is involved in RGC axon development and survival (Lieven et al., 2007). In an aged retina there is an increase in the number of RGCs and photoreceptors expressing phosphorylated tau (Leger et al., 2011). In POAG with ocular hypertension, decreased total tau and increased phosphorylated tau was reported (Gupta et al., 2008). Hyperphosphorylation and aggregation of tau were associated in vivo with reduced axonal transport in the ON of transgenic mice line expressing human P301S tau transgene (Gasparini et al., 2011; Bull et al., 2012). However, whether intracellular and/or extracellular tau plays a role in glaucoma onset and progression is still an open question.

## NEW PERSPECTIVES ON THERAPEUTIC APPROACH

Increasing lines of evidence suggest that aggregation and accumulation of Aβ and Tau eventually leading to MAPK phosphorylation represent common degenerative mechanisms in both AD and glaucomatous retinal degeneration. Determining the degenerative mechanisms is crucial for development of new therapeutics. The retina, which is affected in both diseases, can be an important brain area where to investigate common mechanisms also in view of new therapeutics. On this ground, a neuroprotective approach based on neurotrophic factors can be considered promising. Neuroprotection by neurotrophic factors was initially investigated for neurodegenerative diseases such as the AD. Evidence suggests that treatments with neurotrophic factors such as the brain-derived neurotrophic factor (BDNF), ciliary neurotrophic factor (CNTF), glial cell line-derived neurotrophic factor (GDNF) increase the survival of neurons in animal models of injury and disease (Alqawlaq et al., 2012). BDNF, together with its receptor tropomyosin receptor kinase B (TrkB), is highly expressed in several brain areas under the control of neuronal activity (Castrén et al., 1992; Pattabiraman et al., 2005). BDNF acts by binding to the TrkB, activating downstream pathways including the MAPK, phosphatidylinositol kinase (PI3K) and phospholipase C-γ (PLC-γ) signaling cascades (Kaplan and Miller, 2000). BDNF controls synaptic plasticity and cell survival in the visual system (Liu et al., 2007; Schwartz et al., 2011; Kimura et al., 2016). BDNF appears to provide the highest level of protection by supporting both protective and regenerative functions of RGCs in various models of ON injury and disease (Peinado-Ramón et al., 1996; Weber et al., 2008; Parrilla-Reverter et al., 2009). BDNF is locally produced by retinal cells in the ganglion cell and inner nuclear layers (Perez and Caminos, 1995); its TrkB receptor is expressed in RGCs, amacrine and Müller cells (Perez and Caminos, 1995; Cellerino and Kohler, 1997; Wahlin et al., 2001).

BDNF level is reduced in the glaucomatous retina (Pease et al., 2000; Quigley et al., 2000; Fabiani et al., 2016) as well as in several brain areas of AD (Connor et al., 1997; Michalski and Fahnestock, 2003; Peng et al., 2005), thus contributing to advancing the hypothesis that a scarce availability of BDNF renders neurons more vulnerable. In addition, BDNF is considered a peripheral marker of neurodegeneration. BDNF in the tears and/or blood is used for detection and assessment of neurodegenerative processes in POAG (Ghaffariyeh et al., 2009, 2011). The interpretation of the results obtained on BDNF level in blood (serum/plasma) of AD patients is more complex and contradictory (Komulainen et al., 2008; Faria et al., 2014). The idea is that cognitive deficits in AD are related to change of BDNF blood level as well as that of other neurotrophic factors such as the nerve growth factor and GDNF in blood (Budni et al., 2015). Interestingly, Yasutake et al. (2006) showed that there is a decline in blood BDNF once the disease has progressed to severe level. Thus, BDNF level in blood represents a marker of cognitive dysfunction and progress of neurodegeneration in AD and other neurodegenerative diseases such as Parkinson's disease (Scalzo et al., 2010) and vascular dementia (Yasutake et al., 2006).

BDNF is important for survival and plasticity of the RGCs in models of ON injury and disease (Peinado-Ramón et al., 1996; Weber et al., 2008; Parrilla-Reverter et al., 2009). BDNF delivery into the entorhinal cortex and hippocampus is able to ameliorate cognitive deficits in aging and experimental AD models (Nagahara et al., 2009). Thus, the reported results make BDNF a good candidate to drive full neuroprotective and repair strategies in neurodegenerative diseases, including glaucoma and AD. However, the therapeutic approach based on BDNF is promising if the limits imposed by complex pharmacokinetic of high molecular weight proteins (for example BDNF low propensity to pass blood-brain and blood-ocular barriers following systemic treatment) are definitely defeated. Recently, we showed that BDNF topical eye treatment in the form of collyrium was able to increase the retinal level of BDNF up to rescue visual responses in a murine model of glaucoma (Domenici et al., 2014). Moreover, the intranasal delivery of proteins has recently emerged as a non-invasive and effective method to target high molecular weight proteins such as the BDNF to several brain areas (Alcalá-Barraza et al., 2010; Dhuria et al., 2010). Thus, BDNF non-invasive treatments represent a promising feasible therapeutic strategy to preserve neuronal function and diminish cell vulnerability in neurodegenerative diseases such as the AD and glaucoma. Although at present it is unclear how BDNF concentrations vary in vivo the use of high BDNF doses should be avoided to circumvent potential undesired effects such as the proconvulsant effects (Heinrich et al., 2011; Gu et al., 2015) and neovascularization (Lam et al., 2011). Additional therapeutic strategies based on BDNF have been used; these consist of gene delivery (Nagahara et al., 2013), transplantation of BDNF-expressing cell grafts (Kurozumi et al., 2005), TrkB agonists (Hu et al., 2010; Devi and Ohno, 2012; Gu et al., 2015).

# CONCLUSION

We reported that several common disease features appear in the AD and glaucoma. Based on common disease features that have been described here, several opportunities exist to develop common therapeutic strategies. A successful example involves neuroprotection by BDNF.

# AUTHOR CONTRIBUTIONS

All authors listed, have made substantial, direct and intellectual contribution to the work, and approved its final version of the manuscript. LD conceived the review focus, conducted literature review and finalized the manuscript. CC summarized and finalized the manuscript. CF conducted literature review and finalized the manuscript. EC conducted literature review and finalized the manuscript.

# ACKNOWLEDGMENTS

Supported by the Department of Biotechnological and Applied Clinical Sciences (DISCAB), University of L'Aquila, and the scientific consortium IN-BDNF. We thank Ms. S. Wilson for revising the English style.

# REFERENCES


specific in DBA/2J mice. J. Cell Biol. 171, 313–325. doi: 10.1083/jcb.2005 06099


entorhinal cortex of APP transgenic mice. J. Neurosci. 33, 15596–15602. doi: 10.1523/JNEUROSCI.5195-12.2013


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Criscuolo, Fabiani, Cerri and Domenici. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# ROCK1 Is Associated with Alzheimer's Disease-Specific Plaques, as well as Enhances Autophagosome Formation But not Autophagic Aβ Clearance

Yong-Bo Hu1† , Yang Zou1† , Yue Huang2† , Yong-Fang Zhang<sup>3</sup> , Guinevere F. Lourenco<sup>2</sup> , Sheng-Di Chen<sup>1</sup> , Glenda M. Halliday <sup>2</sup> \*, Gang Wang<sup>1</sup> \* and Ru-Jing Ren<sup>1</sup> \*

<sup>1</sup> Department of Neurology and Neuroscience Institute, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China, <sup>2</sup> Neuroscience Research Australia and Faculty of Medicine, University of New South Wales (UNSW), Sydney, NSW, Australia, <sup>3</sup> Research Laboratory of Cell Regulation, School of Medicine, Shanghai Jiao Tong University, Shanghai, China

#### Edited by:

Andrew Harkin, Trinity College Dublin, Ireland

#### Reviewed by:

Kah-Leong Lim, National Neuroscience Institute, Singapore Lei Liu, University of Florida, USA

#### \*Correspondence:

Glenda M. Halliday g.halliday@neura.edu.au Gang Wang wgneuron@hotmail.com Ru-Jing Ren docotorren2001@126.com

†These authors have contributed equally to this work.

Received: 27 July 2016 Accepted: 14 October 2016 Published: 02 November 2016

#### Citation:

Hu Y-B, Zou Y, Huang Y, Zhang Y-F, Lourenco GF, Chen S-D, Halliday GM, Wang G and Ren R-J (2016) ROCK1 Is Associated with Alzheimer's Disease-Specific Plaques, as well as Enhances Autophagosome Formation But not Autophagic Aβ Clearance. Front. Cell. Neurosci. 10:253. doi: 10.3389/fncel.2016.00253 Alzheimer's disease (AD) is the most prevalent form of late-life dementia in the population, characterized by amyloid plaque formation and increased tau deposition, which is modulated by Rho-associated coiled-coil kinase 1 (ROCK1). In this study, we further analyze whether ROCK1 regulates the metabolism of amyloid precursor protein (APP). We show that ROCK1 is colocalized with mature amyloid-β (Aβ) plaques in patients with AD, in that ROCK1 enhances the amyloidogenic pathway, and that ROCK1 mediated autophagy enhances the intracellular buildup of Aβ in a cell model of AD (confirmed by increased ROCK1 and decreased Beclin 1 protein levels, with neuronal autophagosome accumulation in prefrontal cortex of AD APP/PS1 mouse model). In vitro over-expression of ROCK1 leads to a decrease in Aβ secretion and an increase in the expression of autophagy-related molecules. ROCK1 interacts with Beclin1, an autophagy initiator, and enhances the intracellular accumulation of Aβ. Reciprocally, overexpression of APP/Aβ promotes ROCK1 expression. Our data suggest ROCK1 participates in regulating Aβ secretion, APP shedding and autophagosome accumulation, and that ROCK1, rather than other kinases, is more likely to be a targetable enzyme for AD therapy.

Keywords: Alzheimer's disease, ROCK1, autophagy, Aβ, Beclin1

#### INTRODUCTION

Alzheimer's disease (AD) is the leading cause of dementia, characterized by the accumulation of extracellular and vascular amyloid in the brain. Amyloid plaques are composed of amyloid-β (Aβ) peptides, and they are derived from the sequential cleavage of the amyloid precursor protein (APP) by a set of proteases. The cleavage of APP occurs through two different pathways:

**Abbreviations:** AD, Alzheimer's disease; APP, amyloid precursor protein; Aβ, amyloid-β; CoIP, co-immunoprecipitation; PtdIns3KC3, phosphatidylinositol 3-kinase class III; ROCK1, Rho-associated coiledcoil kinase 1; 3-MA, 3-methyladenine.

a non-amyloidogenic pathway mediated by α-secretase and an amyloidogenic pathway mediated by β-secretase (Tang, 2005). α-Secretase is a metalloprotease and disintegrin enzyme that cleaves within the Aβ sequence to generate a secreted APP fragment (sAPPα). sAPPα has neurotrophic properties and promotes neurite outgrowth modulating the efficacy of synaptic neurotransmission (Ma et al., 2009). Conversely, APP cleavage by β-secretase occurs at the N-terminus with the C-terminal APP fragment subsequently cleaved by γ-secretase leading to the production of Aβ peptides.

Rho-associated coiled-coil kinases (ROCKs) are Ser/Thr kinases and the first identified effectors of Rho GTPase. They are involved in several aspects of cell behavior, such as cell motility, cell proliferation and apoptosis (Chang et al., 2006; Yang et al., 2013; Loirand, 2015; Swanger et al., 2015). Research indicates that ROCKs are potential therapeutic targets for Aβ metabolism modulation (Weggen et al., 2001; Jacobs et al., 2006; Salminen et al., 2008). Two constitutively-expressed ROCK enzymes have been identified—ROCK1 and ROCK2 that share 92% homology in their kinase domain and differ most in their regulatory domains and subcellular localization (Ishizaki et al., 1997; Shi et al., 2013; Chong et al., 2016). Recent studies show that ROCK1 modulates the shedding of the α-secretase-cleaved soluble APP ectodomain (sAPPα) from cultured cells, and that ROCK1 depletion reduces the levels of Aβ (Herskowitz et al., 2013; Henderson et al., 2016). The buildup of internal Aβ rather than its external deposition is associated with neuronal loss in APP/PS1 knock-in mice suggesting externalization of cellular Aβ is protective (Christensen et al., 2008). However, the mechanism by which ROCK1 externalizes Aβ or regulates Aβ production remains to be determined.

In AD, autophagy directly affects the secretion of Aβ and amyloid plaque formation (Nilsson et al., 2013), and long-term rapamycin treatment reduces Aβ plaque load by the induction of autophagy in an AD mouse model (Nilsson et al., 2013). The pivotal role for autophagy in the clearance of aggregate-prone proteins also manifests with an accumulation of autophagic vacuoles (autophagosomes and autolysosomes) in several neurodegenerative diseases (Zhang et al., 2016). Beclin1-mediated autophagy is the major cellular pathway for degradation of aggregated proteins and is required for the formation of double-membrane vesicles, cytoplasmic organelles called an autophagosome (McLeland et al., 2011). This dynamic process involves autophagosome formation, autophagosome-lysosome fusion and the degradation of autophagosomal contents by all kinds of hydrolases (Hong et al., 2016). During the process of autophagosome biogensis, Beclin1 plays a central role and governs the autophagic process by binding to phosphatidylinositol 3-kinase class III (PtdIns3KC3), which recruits additional autophagic proteins for autophagosome formation (Miki et al., 2016). ROCK1 phosphorylation of Beclin1 is the critical regulator for stress-induced autophagy (Gurkar et al., 2013). A reduction in ROCK1 is thought to diminish Aβ levels by enhancing lysosomal degradation of APP (Henderson et al., 2016).

To investigate the involvement of ROCK1 in the Aβ pathology diagnostic features for AD, the levels of and location of ROCK1 in brain tissue from preclinical and clinical AD patients were examined. Aβ production and autophagosome formation following increases or decreases in ROCK1 levels were tested using cell culture methods. We showed that ROCK1 is deposited with highly aggregated Aβ in human brain tissue, and that activated ROCK1 inhibits autophagic Aβ clearance via interaction with Beclin1, increasing Aβ burden during AD progression.

## MATERIALS AND METHODS

## Pathological Cohort

The human ethics committee of the University of New South Wales approved this tissue study. Following institutional approvals, 10 µm formalin-fixed paraffin-embedded temporal cortex sections from 17 cases with different severities of AD pathology (female: male = 13:4, age = 87 ± 9.5 (years old), post mortem delay = 15 ± 11 (h), low severity, Braak stages 0–2 and not demented: intermediate severity and Braak stages 3–4: high severity, Braak stages 5–6 and dementia = 5:3:9) were obtained from the Sydney Brain Bank which collects brain tissue with informed consent from longitudinally followed research participants. Subjects were characterized according to recent criteria for the pathological diagnosis of AD. One of the cases with an intermediate severity had a diagnosis of AD for 3 years, while all those with a high probability of AD had dementia for between 5 and 17 years. Exclusion criteria were alternate neurodegenerative disorders or a dominant family history of neurodegenerative disorder.

#### Immunofluorescence

For immunofluorescence, the 10 µm formalin-fixed, paraffin-embedded sections underwent heat-induced antigen retrieval with Citrate Buffer pH 6 for 3 min, followed by formic acid pre-treatment for 3 min. Tissue sections were double-labeled with rabbit anti-ROCK1 (1:50, Abcam) and mouse monoclonal anti-human Aβ1–42 (1:200, M0872, Dako, Glostrup, Denmark), and visualized with goat anti-rabbit Alexa Fluor 488 (1:500, A-11008, Life Technologies) and goat anti-mouse Alexa Fluor<sup>r</sup> 594 (1:500, A-11005, Life Technologies), respectively. Sections were coverslipped with Vectashield HardSet Antifade Mounting Medium with DAPI (H-1500, Vector Laboratories, Burlingame, CA, USA) and examined on confocal microscope (Nikon Eclipse E400).

## Quantitative Analysis of Aβ and ROCK1 Colocalization

For quantification, 10 images were made for each case where plaques were present. All diffuse and cored plaques on each image was identified using the Aβ1–42 immunofluorescence, and the proportion that had ROCK1 co-localization recorded separately for both diffuse and core plaques. In total, an average of 40 ± 17 plaques were examined per case, with 66% on average being diffuse and 36% being cored. Multivariate statistics were performed in SPSS (IBM, v23) to assess differences between the percentages of plaques colocalizing ROCK1 immunoreactivity and the severity of cortical plaques observed, as assessed by the CERAD score (0 = none, 1 = mild or 1–5/image, 2 = moderate or 6–15/image and 3 = severe or more than 15/image), as well as to case diagnosis based on pathological load of different plaque types and neurofibrillary tangles (control, preclinical AD and AD).

#### Cell Lines and Cell Culture

HEK293 cells stably transfected with human APP695 harboring the ''Swedish'' mutation (HEK293 APP695sw) and SH-SY5Y human neuroblastoma were maintained in Eagle's minimal essential media or DMEM (Gibco), respectively, with 10% fetal bovine serum, and 1% penicillin/streptomycin. Cells were plated at 100,000 cells/cm<sup>2</sup> density in 6-well dishes that were coated with 100 µg/ml poly-lysine. On day 2 post plating, cells were transduced with indicated lentivirus with a multiplicity of infection of 1. For all studies, cells were treated with drugs at 72 h post-transduction in conditioned media for 16 h. For RNAi knockdown of ROCK1, cells were harvested 96 h post-transduction. For transductions or transfections, equivalent amounts of cells were plated, and transfections were performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. At 72 h post-transfection, cells were treated as indicated and then processed for western blot analysis or fluorescence imaging.

#### Cell Proliferation Assay

Cell proliferation was measured by the CCK-8 Kit (Beyotime) according to the manufacturer's instructions. Cells in 100 µl media were treated with 10 µl CCK-8 reagent at 37◦C for 2 h, followed by measuring the absorbance at 450 nm on a microplate reader.

### Plasmid Construct, RNA Interface and Transfection

ROCK1 plasmid (pCAG-myc-ROCK1myc-<sup>727</sup> ∆3, ROCK1 CA) was constructed as previously reported (Ishizaki et al., 1997). The siRNA sequence against ROCK1 (sense 5 0 -GGCAGAGGAAGAAUAUAAATT-3<sup>0</sup> ; antisense 5 0 -UUUAUAUUCUUCCUCUGCCTT-3<sup>0</sup> ) and scrambled siRNA was synthesized by GenePharma (Shanghai, China). Cells were transfected with ROCK1 plasmid, ROCK1 siRNA or scrambled siRNA using Lipofectamine 2000 reagent according to the manufacturer's instructions. At 72 h post-transfection, cells and supernatant were harvested and then processed for western blot analysis.

#### Aβ Measurements

For all cells, conditioned media were used for 48 h, and then the cells and media collected separately for biochemical analyses. Aβ40 was detected using a sandwich ELISA for human Aβ40 following the manufacturer's instructions. Plates were read at 450 nm on a Synergy MX plate reader (BioTeck, Winooski, VT, USA).

#### Immunoblotting

Cells were lysed with 1% RIPA Lysis Buffer (Beyotime, China). Cell lysate was subjected to a 13,000 rpm spin to remove nuclei and debris. Cleared lysate was used for the indicated biochemical assay. Protein concentrations were determined using the Enhanced BCA Protein Assay Reagent (Beyotime). Equal amounts of cell lysate were loaded onto SDS-PAGE gels and then transferred to PVDF membranes. Membranes were blocked with 5% fat-free dry milk in Tris buffered saline (TBS), containing 0.05% Tween-20, and incubated with primary antibodies. Protein bands were detected by horseradish peroxidase-conjugated species-specific secondary antibodies. Actin was used as loading control. Images were captured using an Odyssey Image Station (LI-COR), and band intensities were quantified using Scion Image.

#### mCherry-GFP-LC3B Assay

In order to evaluate the effects of ROCK1 plasmid on GFP–LC3B puncta formation, SH-SY5Y cells were plated on poly-lysinecoated 96-well plates and transduced with adenovirus expressing mCherry-GFP-LC3B fusion protein. After 24 h and 48 h, cell images were obtained by EVOS fl Auto (Life, USA).

#### Co-Immunoprecipitation

Co-immunoprecipitation (CoIP) was performed as previously described (Rijal Upadhaya et al., 2014). HEK293 cells in 60-mm dishes were co-transfected with pCAG-myc-ROCK1myc-<sup>727</sup> ∆3 plasmid. Forty eight hours later, the cells were lysed with 1% RIPA Lysis Buffer (Beyotime). After clarification by centrifugation at 4◦C for 30 min at 12,000 rpm, 500 µl cell lysates were incubated with 20 µl of protein A + G agarose beads (Beyotime, China) for 4 h with gentle rotation at 4◦C. The beads were washed four times with the cell lysis buffer and precipitates were eluted with 2× SDS-PAGE sample buffer, and then analyzed by western blot for anti-Myc and anti-Beclin1 immunoreactivity, respectively.

#### Primary Neuron Culture, Transfection and Immunofluorescence

Primary cortical neurons were prepared from E16 to E18 days of Sprague-Dawley rat brain tissues. Briefly, cerebral cortex was removed aseptically, and then digested and dispersed into single cells; then neurons were re-suspended in neuro-basal medium with 2% B27 supplement (Invitrogen), 2 mM glutamine and 1% penicillin and streptomycin (Invitrogen, Carlsbad, CA, USA) and then seeded at 1.25 × 10<sup>5</sup> cells per cm<sup>2</sup> on 12-well plates (Corning Inc., Midland, NC, USA) coated with 100µg/ml polyl-lysine and incubated in humidified atmosphere with 5% CO<sup>2</sup> at 37 ◦C. Transfection using Lipofectamine 2000 (Invitrogen) was carried out on acutely dissociated neurons before plating, and 4 h later, the whole medium was replaced with NB/B27 and immunofluorescence experiment was performed after 48 h transfection.

For immunofluorescence, primary neurons were cultured on a cover glass and fixed with 4% paraformaldehyde in PBS for 10 min at room temperature. Permeabilization was performed in PBS with 0.3% Triton X-100 for 10 min at room temperature. After blocking for 30 min with 5% normal donkey serum, neurons were incubated with rabbit anti-LC3 (1:100, CST) antibody or mouse anti-ROCK1 (1:500, Abcam) antibody overnight at 4◦C. After washing three times with 0.1 M PBS, the sections were incubated with AlexaFluor 488-conjugated donkey anti-rabbit or anti-mouse IgG secondary antibodies (Invitrogen), respectively, and the sections were visualized with a confocal microscope (FV-1000).

#### Immunohistochemistry and Histology

APP/PS1 mice were provided by Model Animal Research Center of Nanjing University. APP/PS1 mice have been intensively characterized for Aβ plaques load and AD behavioral phenotypes. In this experiment, 10-month-old APP/PS1 transgenic mice and age-matched C57BL/6J were used. Mice were deeply anesthetized and perfused with 10% formalin, and their brains were removed and fixed in 4% paraformaldehyde in PBS overnight at 4◦C. Whole brains were cryoprotected in 25% sucrose in 0.1M PBS at 4◦C until sectioning. The brain tissues were cut in to coronal 30 µm sections. Immunohistochemistry (IHC) for APP and Beclin1 were performed using primary antibodies of 6E10 (1:100, Novus) and rabbit anti-Beclin1 (1:100, Novus) respectively (Sugrue et al., 2010). APP and Beclin1 immunopositive neurons were quantified according to the previously established method (Dlugos and Pentney, 2002).

#### Antibodies and Reagents

Mouse anti-ROCK1 antibody (Abcam, 1:500 dilution), rabbit anti-Beclin1 (Novus, 1:10,000), rabbit anti-APP-CTF (Sigma-Aldrich, 1:5000), anti-mouse sAPPα (IBL, 1:50), anti-mouse myc (Abcam, 1:1000), anti-rabbit LC3 (Cell Signaling, 1:100) and mouse anti-β-actin (Sigma-Aldrich, 1:1000) were used. Loading controls (β-actin) were used for Western blot standardization. Lipofectamine 2000 was sourced from Invitrogen. 3-methyladenine (3-MA) was purchased from Sigma Aldrich (St. Louis, MO, USA) and 3-MA was dissolved in double distilled water and titrated to 1 mM.

#### Statistical Analysis of the Cell Culture Experiments

Data are reported as mean ± SEM. Comparisons between experimental groups were analyzed by Student's t-test or ANOVA as appropriate, with p < 0.05 considered as significant.

#### RESULTS

#### ROCK1 is Associated with Aβ Plaques in the Brains of AD Patients

To determine whether Aβ associates with ROCK1 in the brains of patients with AD, co-localization studies were performed using specific antibodies in healthy aged controls (n = 5 aged 89 ± 4 years, Braak neuritic stages 0–2), preclinical AD (n = 4 aged 85 ± 3 years, Braak neuritic stages 3–4) and clinical end-stage AD (n = 8 aged 86 ± 4 years, Braak neuritic stages 5–6). Consistent with this finding, ROCK1 immunoreactivity was identified in more cored than diffuse plaques (**Figure 1**). Quantitation showed that the majority of cored plaques contained ROCK1 immunoreactivity, and that colocalization increased with increasing severity of cortical Aβ deposition as assessed by CERAD plaque score (**Figure 1J**, Wilks' Lambda p < 0.0001 covarying for age and brain size, post hoc Bonferroni p = < 0.002). Nearly half of all Aβ plaques in preclinical AD contained ROCK1 immunoreactivity, with significantly higher numbers of ROCK1-immunopositive Aβ plaques found in AD patients (**Figure 1K**, Wilks' Lambda p = 0.03 covarying for age and brain size, post hoc Bonferroni p = 0.003 for AD diagnosis). These results demonstrate a close relationship between ROCK1 localization and pathological Aβ plaque in both preclinical and clinical AD.

#### ROCK1 Regulates Aβ Secretion

To further explore the effects of ROCK1 cleavage on the production of Aβ, we constructed a plasmid expressing this constitutively active ROCK1 fragment (ROCK1 CA), as previously reported (Ishizaki et al., 1997). In HEK 293T cells stably transfected with the APP695 Swedish mutation (APPsw), ROCK1 levels were manipulated by transfection with ROCK1 CA plasmid, ROCK1 siRNA or scrambled siRNA. Forty eight hours after transfections, the levels of secreted Aβ40 in the cell culture medium were measured by ELISA. As shown in **Figure 2**, secreted Aβ40 levels were increased by 80% compared to the control after depletion of ROCK1 (**Figures 2A,B**), which is consistent with a recent study. In contrast, secreted Aβ40 levels were decreased following ROCK1 CA over-expression in HEK293T APPsw cells (**Figures 2C,D**). The Aβ42 level was too low to be detected by ELISA. ROCK1 CA plasmid transfection did not change cell proliferation (**Figure 2E**). These results confirm that ROCK1 influences Aβ secretion.

#### ROCK1 Modulates APP Shedding

Given that there is an established relationship between ROCK1 and Aβ secretion, we next investigated the effects of ROCK1 on APP cleavage. After transfection of ROCK1-targeted siRNA or ROCK1 CA plasmid to HEK293T APPsw cells, cells were harvested 48 h later and lysates were analyzed. The holo-APP and sAPPα were examined by Western Blot with the mAb 6E10, specific for residues 1–17 of the Aβ sequence (**Figure 3A**). Electrophoresis revealed no significant difference of full-length APP (**Figure 3C**), however the amount of sAPPα was higher after ROCK1 depletion (**Figure 3D**). In cells with increased ROCK1 CA levels, sAPPα was significantly decreased (**Figure 3D**). These results confirm that ROCK1 modulates the metabolism of APP. To further explore the interaction between ROCK1 and APP/Aβ, we performed normal HEK293T cell culture with Aβ40 treatment or APP695 plasmid transfection. Addition of Aβ40 or APP695 over-expression promoted endogenous ROCK1 expression in normal HEK293T cells (**Figures 3B,E**).

plaque deposition (n = 14/17). Multivariate analysis identified a significant increase (∗p < 0.02) in ROCK1 and Aβ colocalization in plaques with increasing cortical deposition of Aβ plaques, as assessed by the CERAD plaque score (J). With increasing CERAD plaque severity, more cored plaques contained ROCK1 immunoreactivity, whereas colocalization of ROCK1 and Aβ in diffuse plaques was mainly observed when there was a high cortical load of Aβ (J). Multivariate analysis revealed that the degree of colocalization of ROCK1 and Aβ in cored plaques was highest in patients with clinical Alzheimer's disease (AD; <sup>∗</sup>p = 0.003), although significant numbers of Aβ plaques contained ROCK1 immunoreactivity in preclinical AD (K).

#### Identification of ROCK1 as an Enhancer of Beclin1 Dependent Autophagy

In further experiments, we explored the mechanism of ROCK1 as an enhancer of Beclin 1 dependent autophagy. When ROCK1 was depleted, the expression of Beclin1 was decreased, and when ROCK1 CA was over-expressed by plasmid transfection, the level of Beclin1 was also increased. To determine whether autophagy is involved in the mechanism

by which ROCK1 depletion increases Aβ secretion, SH-SY5Y cells were co-transfected with mCherry-LC3B adenovirus and ROCK1 CA plasmid and the patterns of LC3 puncta were analyzed by immunoflouresence. **Figure 4A** shows an increase in LC3 puncta after ROCK1 CA plasmid transfection. LC3 was also assessed in HEK293T cells and, as the results show, the LC3II/I ratio was increased in cells when ROCK1 CA was over-expressed and decreased following ROCK1 depletion (**Figures 4E,G**). We also explored whether ROCK1 interacted with Beclin1 during autophagy (**Figures 4B,C**). HEK293T cells transfected with myc-tagged human ROCK1 were lysed and the proteins subjected to co-immunoprecipitation with anti-ROCK1 and anti-myc antibody. Our results showed that endogenous ROCK1 and Beclin1 interacted with each other. Moreover, when ROCK1 CA was over-expressed, this interaction was enhanced (**Figures 4D,F**). These results showed that an interaction between ROCK1 and Beclin1 triggered the autophagic build-up of intracellular Aβ and a decrease in Aβ secretion.

To further elucidate ROCK1 influence on autophagy, 3-MA was used as an inhibitor of autophagy. After HEK293T APPsw cells were treated with 3-MA for 6 h, they were transfected with ROCK1 CA plasmid for 48 h. To test whether ROCK1 CA over-expression rescued 3-MA-induced autophagy inhibition, the cells were lysed and proteins analyzed by ELISA and Western blot. Inhibiting autophagy increased intracellular Aβ40 levels (**Figure 4J**), and over-expression of ROCK1 by transfection with ROCK1 CA plasmid reversed this effect even though 3-MA treatment decreased Beclin1 levels (**Figures 4H,I**).

# APP/PS1 Mice have Increased ROCK1 and Neuronal Autophagosome Accumulation

After confirming the relationship between ROCK1 and autophagosome formation in vitro, we next assessed whether ROCK1 and autophagosome proteins are upregulated with Aβ in vivo by conducting immunofluorescence and IHC staining of brain sections from 10 months old APP/PS1 transgenic C57BL/6 mice. Our data showed that ROCK1 was increased significantly with neuronal autophagosomes accumulation (**Figures 5A,C**) and that Beclin1 was decreased in the AD mouse model (**Figures 5B,D**). These observations suggested that autophagy was impaired in neurons and increased ROCK1 inhibited Aβ secretion, resulting in intracellular Aβ accumulation and neurotoxicity. Taken together, these data demonstrated the same associations between ROCK1, intracellular Aβ and autophagosomes accumulation as we identified in our in vitro experiments.

#### DISCUSSION

The critical events that trigger or cause increased Aβ levels and post-translational modifications that increase Aβ insolubility are still under debate. However, modulation of the production, metabolism and fibrilization of the protein are key pathogenic aspects. Our results show that ROCK1 levels regulate Aβ clearance by modulating the autophagy pathway via an interaction with its substrate, Beclin1, that initiates autophagosome formation that then accumulate

intracellularly. This is consistent with previous studies showing that ROCK1 regulates autophagosome formation (Gurkar et al., 2013), that autophagosomes accumulate in AD and that ROCK1 inhibition reduces Aβ levels in brain (Nixon and Yang, 2011). Our data also show that ROCK1 collocates with fibrilized Aβ. It has recently been shown that different post-translational modifications of Aβ occur in different types of plaques (Rijal Upadhaya et al., 2014). The cases in this study have been previously analyzed using Western blotting for ROCK1 with no reduction in membrane associated SDS-soluble ROCK1 protein levels by end-stage AD (Wang et al., 2016). Taken together, our data indicate that ROCK1 may be involved in Aβ modification.

Phosphorylation of Aβ at serine 8 at the cell surface has recently been identified to dramatically increase the rate and amount of Aβ oligomerization and amyloid fibril formation (Rezaei-Ghaleh et al., 2016). In fact, AD patients are distinguished from those without clinical AD by selectively accumulating this phosphorylated form of Aβ (Rijal Upadhaya et al., 2014). The kinases that are involved in phosphorylating Aβ are poorly understood, but are thought to be of the AGC kinase family of serine-threonine kinases (Rezaei-Ghaleh et al., 2016), which includes ROCK1. Unlike ROCK2, ROCK1 interacts with a number of proteins in the plasma membrane (Stroeken et al., 2006) and we show that ROCK1 co-localizes with highly fibrilized Aβ in human brain. The localization and association of ROCK1-associated Aβ with clinical AD mirrors the data observed for serine 8 phosporylated Aβ (Rijal Upadhaya et al., 2014) and suggests that ROCK1 may be the kinase involved in this disease-specific Aβ phosphorylation.

We provide further evidence that ROCK1 has a role in the plasma membrane in its influence on APP shedding. APP shedding is directly mediated by α- and β-secretase generating either non-amyloidogenic (sAPPα) or amyloidogenic fragments (Tang, 2005). Our data show that ROCK1 activation enhances the amyloidogenic pathway, consistent with recent data (Henderson et al., 2016). However, we demonstrate that ROCK1 decreased sAPPα shedding in cultured cells, but also reduced rather than increased Aβ secretion from the cells, suggesting that ROCK1 may also play a role in Aβ secretion at the plasma membrane (potentially associated with the aberrant phosphorylation of Aβ, see above). Together these results are also consistent with ROCK1 activation increasing intracellular Aβ.

We show that activated ROCK1 inhibits the autophagic clearance of Aβ via its interaction with Beclin1, increasing Aβ burden during AD progression. Autophagy is a critical clearance system for cellular toxic protein aggregates and it is crucial for the physiological regulation of normal cell function, which can be induced by extracellular and intracellular signals, including oxidative stress, starvation and endoplasmic reticulum stress (Wang and Mao, 2014). In the brain, the complex axonal and dendritic structures are highly dependent on efficient autophagic turnover (Wong and Holzbaur, 2015; Hong et al., 2016). In AD, autophagy is upregulated at early disease stages, as evidenced by swollen endosomes and the accumulation of autophagosomes (Wong and Holzbaur, 2015). At the transcriptional level, autophagy-activating factors are also upregulated in AD (Jiang et al., 2013). ROCK1 phosphorylation of Beclin1 is known to be the critical regulator for stress-induced autophagy (Gurkar et al., 2013). We confirm that Beclin1 is a

substrate of ROCK1 and that the interaction of ROCK1 and Beclin1 results in the inhibition of autophagic Aβ clearance. It is well known that Beclin1 is a critical regulator of autophagy and several lines of investigation have demonstrated that autophagy plays an important role in Aβ clearance (also known as ''autophagic Aβ clearance''; Wirawan et al., 2012). Although ROCK1 phosphorylation of Beclin1 activates autophagosome production, our data show that these organelles accumulate intracellularly with enhanced intracellular Aβ and a reduction in secreted Aβ. This is consistent with previous reports showing that autophagosome maturation to autolysosomes is the main problem with autophagy in AD and results in accumulating intracellular Aβ (Nixon and Yang, 2011).

Frontiers in Cellular Neuroscience | www.frontiersin.org November 2016 | Volume 10 | Article 253 |

Overall our data suggest a role for ROCK1 in regulating Aβ production, secretion and post-translational modification at the plasma membrane, as well as its important role in intracellular autophagosome induction. The location of ROCK1 to the plasma membrane (Stroeken et al., 2006), and also in extracellular plaques (as shown in this study), makes this kinase potentially more easily targetable and relevant for therapeutic manipulations than others. If as suggested, Aβ phosphorylation at serine 8 enhances plaque maturation to progress presymptomatic pathology to symptomatic AD (Rijal Upadhaya et al., 2014), then inhibition of ROCK1 and/or Aβ dephosphorylating enzymes, may provide the most relevant treatments. Taken together, our findings outline a critical role of ROCK1 in the progression of Aβ pathology, highlighting its potential as a therapeutic target.

#### AUTHOR CONTRIBUTIONS

Y-BH and YZ performed most experiments, collected and analyzed data, and wrote the manuscript. YH and GFL performed pathological experiments. GW, R-JR, S-DC and GMH participated in the design of the study. Manuscript was written by Y-BH and YH, and critically reviewed by GW, R-JR, Y-FZ, S-DC and GMH. All authors read and approved the final manuscript.

#### ACKNOWLEDGMENTS

This study was supported by the National Basic Research Development Program of China (Nos. 2010CB945200, 2011CB504104), the National Natural Science Foundation

#### REFERENCES


of China (Nos. 81001426, 81171027, 81200842, 91332107), Shanghai Pujiang Program (No. 15PJ1405400) and the National Health and Medical Research Council of Australia (program grant #1037746 and senior principal research fellowship grant #1079679). Human brain tissue was received from the Sydney Brain Bank at Neuroscience Research Australia that is supported by the University of New South Wales and Neuroscience Research Australia.

immunoblot. Methods Mol. Biol. 697, 199–206. doi: 10.1007/978-1-60327- 198-1\_21


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Hu, Zou, Huang, Zhang, Lourenco, Chen, Halliday, Wang and Ren. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Regional and Gender Study of Neuronal Density in Brain during Aging and in Alzheimer's Disease

#### Eva Martínez-Pinilla, Cristina Ordóñez, Eva del Valle, Ana Navarro\* and Jorge Tolivia\*

Departamento de Morfología y Biología Celular, Facultad de Medicina, Instituto de Neurociencias del Principado de Asturias, Universidad de Oviedo, Oviedo, Spain

Background: Learning processes or language development are only some of the cognitive functions that differ qualitatively between men and women. Gender differences in the brain structure seem to be behind these variations. Indeed, this sexual dimorphism at neuroanatomical level is accompanied unequivocally by differences in the way that aging and neurodegenerative diseases affect men and women brains.

Objective: The aim of this study is the analysis of neuronal density in four areas of the hippocampus, and entorhinal and frontal cortices to analyze the possible gender influence during normal aging and in Alzheimer's disease (AD).

Methods: Human brain tissues of different age and from both sexes, without neurological pathology and with different Braak's stages of AD, were studied. Neuronal density was quantified using the optical dissector.

#### Edited by:

Daniela Tropea, Trinity College, Ireland

Reviewed by: Ramesh Kandimalla, Texas Tech University, USA Linda Ann Bean, Rush University Medical Center, USA

#### \*Correspondence:

Ana Navarro anavarro@uniovi.es Jorge Tolivia jtolivia@uniovi.es

Received: 13 July 2016 Accepted: 23 August 2016 Published: 13 September 2016

#### Citation:

Martínez-Pinilla E, Ordóñez C, del Valle E, Navarro A and Tolivia J (2016) Regional and Gender Study of Neuronal Density in Brain during Aging and in Alzheimer's Disease. Front. Aging Neurosci. 8:213. doi: 10.3389/fnagi.2016.00213 Results: Our results showed the absence of a significant neuronal loss during aging in non-pathological brains in both sexes. However, we have demonstrated specific punctual significant variations in neuronal density related with the age and gender in some regions of these brains. In fact, we observed a higher neuronal density in CA3 and CA4 hippocampal areas of non-pathological brains of young men compared to women. During AD, we observed a negative correlation between Braak's stages and neuronal density in hippocampus, specifically in CA1 for women and CA3 for men, and in frontal cortex for both, men and women.

Conclusion: Our data demonstrated a sexual dimorphism in the neuronal vulnerability to degeneration suggesting the need to consider the gender of the individuals in future studies, regarding neuronal loss in aging and AD, in order to avoid problems in interpreting data.

Keywords: age, Alzheimer's disease, sexual dimorphism, human, hippocampus, entorhinal cortex, frontal cortex

# INTRODUCTION

Compelling evidences of changes in the human brain, at the anatomical and molecular level, related with aging and during Alzheimer's disease (AD) have been described by different authors. Anatomically, it has found a clear decrease in the volume and weight of the brain and histologically an increase in the size of astrocytes and microglia, as well in the neuronal lipofuscin content

**Abbreviations:** AD, Alzheimer's disease; CNS, central nervous system; NFT, neurofibrillary tangles; SD, standard deviation; SP, senile plaques.

(Mrak et al., 1997; Sheffield and Berman, 1998; Schultz et al., 2004; Dorszewska, 2013). However, there is a great controversy about the occurrence and extent of neuronal loss processes in these situations. Some influential papers from as early as the 1950s, looking for changes in neuronal density in twodimensional space concluded that a substantial loss of neurons occurs with age. The percentage of this loss varies from 10 to 60% depending on the methodology employed and the neuronal population examined. Moreover, it has been shown that this death is cell type-dependent since some populations do not exhibit signs of degeneration and others, as cerebral cortex and hippocampus, are particularly affected (Brody, 1955; Colon, 1972). In AD, neuronal loss constitutes one of the major pathological markers that extensively affect different brain areas as entorhinal or prefrontal cortex. This decrease in the neuronal population that reaches 90% correlates well with the severity of the disease (Terry, 2006; Zilkova et al., 2006; Padurariu et al., 2012). The development of more accurate procedures for counting neurons over the last years confirmed these previous observations and open new avenues to better understand how the brain changing with age and with the progression of different neurodegenerative diseases.

The application of stereological techniques to several species, including humans, has led to the conclusion that the decline in neuronal number during aging is not significant in brain regions such as neocortex or hippocampus (West and Gundersen, 1990; Pakkenberg and Gundersen, 1997; Hof and Morrison, 2004) or in some vestibular nuclei (Alvarez et al., 2000). However, it has been reported a highly significant correlation between loss of neurons and age in the entorhinal cortex (Simic et al., 2005) and in the human medial vestibular nucleus (Alvarez et al., 1998). In AD patients, the loss of neurons takes place primarily in the neocortex, hippocampus and entorhinal region (Terry, 2006). Interestingly, the characteristic degenerative processes of AD do not affect equally all cell types. As an example, the pyramidal cells in the entorhinal cortex and the CA1 and subiculum regions of the hippocampus seem to be more vulnerable to neurofibrillary tangles (NFT) formation and neurodegeneration than cells of other hippocampal areas (Adachi et al., 2003; Kerchner et al., 2010; Padurariu et al., 2012). The same phenomenon has been observed in the neocortex. This specific vulnerability to degeneration assumes particular importance in the pyramidal cells that furnish long cortico-cortical projections, leading to a global disruption of interconnections among association cortices (Desikan et al., 2010). In contrast, primary sensory and motor areas exhibit minimal loss of neurons.

Several authors have described evidences for sexual dimorphism in the number and cell density in the central nervous system (CNS). In fact, studies on cerebral cortex show that men have 15% more cortical neurons and 13% greater total neuronal density than women, without variations in thickness (Pakkenberg and Gundersen, 1997; Rabinowicz et al., 2002). Complementary studies achieved by magnetic resonance, confirmed that some brain areas such as human cerebellar cortex or gray matter of the left amygdala (Raz et al., 1997) exhibit larger volumes in men than in women (Raz et al., 2004; Sowell et al., 2007). On the contrary, women show larger right striatal and bilateral hippocampal gray matter volumes than men (Neufang et al., 2009). Another sexual dimorphic brain structure in humans is the corpus callosum, which is higher in women than in men (DeLisi et al., 1989). All these gender differences may underline gender-specific abilities and susceptibilities to disease, probably influenced by gonadal hormones i.e., estrogen, testosterone and progesterone, in certain areas of the brain. The sexual differentiation of a particular brain region is related to its hormonal environment that comprises local and circulating hormones of the CNS (Neufang et al., 2009). Thus, it seems reasonable to expect that the loss of neuronal cells that occurs in the aging processes as well as in several pathologies of the CNS as AD is influenced by these sexual differences.

In recent years, some studies have been interested in the potential impact of sex on age-related brain changes and in the development of different neurodegenerative diseases. As a rule, these studies concluded that men exhibited greater agerelated brain atrophy than women over the entire life expectancy (Gur et al., 1999); this effect is enhanced in the frontal and temporal lobes (Raz et al., 1997; Gur et al., 2002). Meanwhile, a significant reduction of gray matter in women has been reported in the parietal lobes and hippocampus (Raz et al., 2004; Sowell et al., 2007). According with different authors, there is a higher prevalence and incidence of AD in women than in men (Breitner et al., 1988; Fratiglioni et al., 1997; Viña and Lloret, 2010; Alzheimer's Association, 2015). It has been postulated that estrogen deficiency, following menopause, may contribute to the etiology of the disease. In a 12 case-control and cohort studies conducted in the 1990s, it was suggested that estrogen therapy could delay the onset or contribute to the prevention and/or significantly attenuation of AD (Sundermann et al., 2006), but this question is still under discussion (Asthana et al., 1999; Henderson et al., 2000; Mulnard et al., 2000; Henderson, 2014).

The bulk of evidence suggests that sexual dimorphism in the neuronal content may determine the way that aging and neurodegenerative processes affect men and women brains. The aim of this research work is to study the changes in neuronal density in hippocampus and entorhinal and frontal cortices of both, men and women, in order to analyze the possible gender influence during normal aging and in AD.

#### MATERIALS AND METHODS

#### Human Tissues

Use of human brain tissues were approved by "Comité Ético de Investigación Clínica Regional del Principado de Asturias" as follows. These studies were granted waivers of consent on the following bases: (1) samples were gathered retrospectively from pathology archives of necropsies performed for diagnostic purposes; (2) patient identities were anonymized and completely delinked from unique identifiers; and (3) there was no risk to the participants.

Human brain tissues were provided by The Pathologic Anatomy Service of the University Central Hospital of Asturias and the Bank of Neurologic Tissues of the Clinic Hospital of Barcelona. This material was the same used in recent studies of our group (Ordóñez et al., 2012; Martínez et al., 2013; Navarro et al., 2013). Seventy-two cases were employed. Thirty-six individuals with not known neurological, psychiatric, or neuropathological disorders (18 men and 18 women) were divided in three groups according to their age: the first group includes individuals in their 30's and 40's; the second from those in their 50's to the ones in their 70's; and the last one those older than 80. Other thirty-six cases of AD (18 men and 18 women) were divided in three groups based on their AD neuropathological stage, according to Braak's criteria (Braak and Braak, 1991). Postmortem intervals ranged between 2 and 6 h. The pieces from human frontal cortex (Brodmann's area 9), hippocampus and adjacent entorhinal cortex were fixed by immersion in 10% buffered formalin. After fixation, they were washed in distilled water, dehydrated through successive alcohols, cleared in two baths of butyl acetate, embedded in paraffin, and placed in a suitable mold. Transverse sections about 10µm thick were obtained and attached to gelatin-covered slides, deparaffined in xylene, and rehydrated.

#### Neurons, Senile Plaques, Neurofibrillar Tangles, and Amyloid-Beta Staining

Alternated sections were stained using a Nissl-like method developed in our laboratory to counterstaining paraffin sections stained with alcoholic Congo Red to show amyloid. This method allows a clear discrimination between neurons and the rest of nervous cells (Navarro et al., 1999). To visualize the typical cerebral markers of the AD neuropathology, the silver technique of Reusche (1991) and a modification of the Congo red method developed in our laboratory (Navarro et al., 1999, 2013) were used.

#### Neuronal Counting

The estimation of the total number of neurons was carried out using the optical fractionator principle. Stereological analysis was performed using an Olympus BX-51 microscope with Olympus CAST system version 2.0 (Olympus, Denmark A/S, Albertslund, Denmark). The analyzed area (field included by a 20x lens), was delimited. From a random start position, a counting frame was superimposed on the image and neurons were systematically sampled using a 60x lens (Plan Apo N 60x /1.42 oil, Olympus) and the nucleolus as the sampling unit. The sampling frequency was chosen by adjusting the xy-axis step length so that up to 200 cells were counted in each specimen.

Neuronal counting was undertaken on five sections at different levels (separated by 100µm). For each section, the neuronal density was calculated by dividing the total number of neurons by the area of the sections surveyed. The mean value of each brain region studied was used for the statistical analysis. Hippocampal variations in neuronal density was assessed separately in each anatomical region (CA1, CA2, CA3, CA4; **Supplementary Figure 1**).

#### Statistical Analysis

The data in the graphs are presented as the mean ± SD. All statistical calculations were conducted using SPSS 15.0 for Windows. The test of Kolmogorov–Smirnov with the correction of Lilliefors was used to evaluate the fit of the data to a normal distribution. One-way ANOVA followed by multiple comparisons Tukey's test was performed to analyze changes in neuronal density between the different groups of age or the Braak's stages of AD. The Student's t-test (two-tailed) was used to analyze possible gender differences. Finally, a Pearson Correlation analysis was used to evaluate the relationship between neuronal density and age or Braak's stage. p < 0.05 was defined as statistically significant.

#### RESULTS

#### Changes in Neuronal Density during Aging

The quantitative study showed that there were no significant differences regarding neuronal density in CA1, CA2, and entorhinal cortex between women and men, and between groups of age (**Figures 1A,B,E**). However, the CA3 and CA4 hippocampal areas of the youngest group showed a higher value in men than in women (**Figures 1C,D**, **2A–D**). These differences disappeared with age since the men group tends to suffer a decrease in neuronal density. A correlation between neuronal density and the age of subjects was not found in these areas; only a statistical significant diminution in neuronal density in men of median age was observed respect to the youngest group. In frontal cortex, both sexes showed a decrease in neuronal density in the 50's–70's and a significant increase in elderly with respect to median age subjects (**Figures 1F**, **2E–H**). However, no correlation between the number of neurons per area and age was found in this brain area.

#### Morphological Changes of Hippocampus and Entorhinal Cortex during Aging

When we analyzed the morphological brain changes during aging in both sexes, we first observed some shrinking neurons in the different areas of the hippocampus in aged individuals. These cells undergo a wide range of modifications in their somas and nuclei that made them look larger and thinner, even in some cases, due to the intensity of the nuclear staining, it was impossible to distinguish the nucleolus (**Figure 3A**). Moreover, we often observed lipofuscin granules in non-degenerating neurons (**Figure 3B**).

The number of astrocytes increased with age in the entorhinal cortex, CA2, CA3, and CA4 while it was always scarce in the CA1 and subiculum. These glial cells also accumulated lipofuscin granules with age and their nuclei turned rounder and bigger than in the youngest subjects (**Figure 3C**). We also observed, mainly in the oldest individuals, corpora amylacea in contact to the pial surface, and in the CA4 close to the dentate gyrus (**Figure 3D**). The entorhinal cortex of some individuals of the over 80 group showed diffuse senile plaques (SP) and few isolated mature SP (**Figure 3E**), whereas the presence of NFT was not a common feature. In the subpial area, we found some vessels with a great extracellular deposit of amyloid protein (**Figure 3F**). It is important to note that these changes are not related with the gender of the subjects.

In the frontal cortex, we found the same morphological changes described previously for the hippocampus and the entorhinal cortex. In fact, we also found an increased neuronal

content of lipofuscin and a higher number of corpora amylacea in contact with the pial surface in aged individuals. SP (mainly diffuse SP) were only observed in the oldest subjects who also showed inconstant perivascular amyloid deposits (data not shown). Once again, all these changes are related with age and not with gender.

#### Neuronal Density in Alzheimer's Disease

The analysis of cellular counting demonstrated that neuronal density in AD is not related with the gender or the degree of disease progression in the CA2, CA4, and entorhinal cortex, since significant differences were not found in these areas between sexes and Braak's stages (**Figures 4B,D,E**). In the CA1 hippocampal area, women displayed a decrease in the number of neurons per mm<sup>2</sup> when we compared the initial with the advance stages of AD (**Figures 4A, 5A,B**), with a negative correlation between neuronal density and Braak's stage (r = −0.662, p < 0.01; **Table 1**). Meanwhile, men did not show significant differences in this area. In contrast, in the CA3 hippocampal area men but not women showed a statistical significant decrease in the neuronal density in the later stages of AD (**Figures 4C, 5C,D**), and a negative correlation between these two parameters (r = −0.662, p < 0.01; **Table 1**).

In the frontal cortex, we found that both men and women showed a statistical significant decrease in the neuronal density in the later stages of AD (**Figure 4F**). Moreover, a clear negative correlation between the number of neurons per area analyzed and the progression of the pathology in both, women (r = −0.537, p < 0.01; **Table 1**) and men (r = −0.514, p < 0.01; **Table 1**), was detected.

FIGURE 2 | Changes in neuronal density in hippocampus and frontal cortex between men and women during aging. Representative microphotographs of human brain sections of non-pathological individuals contrasted with a Nissl method modification. (A) Hippocampus (CA3) of a 37 years old man. (B) Hippocampus (CA3) of a 35 years old woman. (C) Hippocampus (CA4) of a 37 years old man. (D) Hippocampus (CA4) of a 35 years old woman. (E) Frontal cortex of a 40 years old man. (F) Frontal cortex of a 65 years old man. (G) Frontal cortex of a 37 years old woman. (H) Frontal cortex of a 69 years old woman. Bar, 60µm.

# Morphological Changes of Hippocampus and Entorhinal Cortex in Alzheimer's Disease

The morphological brain changes observed in individuals with AD were similar to those previously described for the nonpathological cases over 80's (i.e., changes in neuronal features). As expected, the number of NFT and SP (as well as their level of development) was higher in AD subjects, increasing with the Braak's stage. Thus, we found diffuse isolated SP and some NFT in the entorhinal cortex of patients diagnosed with Braak's stage I-II. In advanced stages, the number of diffuse and mature SP with amyloid core increased notably in this brain area (**Figures 6A,C**).

FIGURE 3 | Morphological changes in hippocampus and entorhinal cortex during aging in both sexes. Representative microphotographs of human brain sections of non-pathological individuals contrasted with a Nissl method modification (A–D), silver technique of Reusche (E) and a modification of Congo Red method (F). (A) Shrinking neurons in the hippocampus of an 80 years old man. (B) Frontal cortex of a 75 years old woman, lipofuscin in neurons can be observed. (C) Hippocampus (CA3) of an 80 years old man, several astrocytes with lipofuscin granules are showed (arrowhead). (D) Hippocampus (CA4) of a 75 years old woman, several corpora amylacea can be observed (arrows). (E) Diffuse senile plaques in the entorhinal cortex of an 85 years old man. (F) Vessels with amyloid (red fluorescence) in frontal cortex of an 80 years old woman. Bars: (A,C), 60 µm; (B,D), 10µm; (E), 40µm; (F), 50µm.

Finally, these characteristic AD hallmarks appeared in CA1 and CA2 (**Figure 6B**) and in a great number in the entorhinal cortex (**Figure 6D**) in the last stages of the disease.

Respect to the glia, we observed the presence of microglial cells, close to neuronal debris, in advanced stages of AD. Likewise, reactive astrocytes with pale big round nuclei and an important accumulation of lipofuscin granules were detected in all areas under analysis (except CA1 and subiculum) and in all Braak's stages. The number of corpora amylacea was higher in AD cases than in non-pathological ones.

In the frontal cortex of AD brains, we observed the same morphological changes as we did in the oldest subjects without pathology, but once again, lipofuscin accumulation as well as the presence of corpora amylacea and microglia seems to be higher in the AD samples. In the same way, the amount of amyloid protein and the number of SP and NFT are considerably higher in AD individuals than in those older subjects without pathology (data not shown). All the changes described here are common and similar to both sexes.

# DISCUSSION

In this work, we look for changes in neuronal density during aging and in AD, in men and women, in some particularly vulnerable brain areas as hippocampus and entorhinal and frontal cortices.

Classical reports analyzing changes in neuronal density in two-dimensional space concluded that the percentage of substantial loss, which occurs with age, varies between 10 and

60 %, depending on the study and the neuronal population examined. Some studies, employing stereological-based sampling to derive estimates of cell number, reported that neuronal loss during aging is either undetectable or relative mild (West and Gundersen, 1990; Hof and Morrison, 2004). Our results are in concordance with these studies since we did not find a significantly neuronal loss with age. Importantly, we demonstrated a negative correlation between Braak's stage and neuronal density in some areas of AD brains. In this sense, West et al. concluded that neurodegenerative processes associated with normal aging and with AD are qualitatively different (West et al., 1994). A possible explanation could be that individuals without neurodegenerative pathology are able to mitigate the constant increased production of free radicals that naturally occurs with age, avoiding the neuronal loss. Interestingly, in several areas as frontal cortex middle-aged individuals showed less neurons per mm<sup>2</sup> than older subjects did. We have to take into account that during human life exists a critical period around 55–70 years old where there is an important incidence of stroke, carcinogenic processes, and neurodegenerative diseases. In this sense, individuals who reach 80 years old are the ones that were able to throw this "bottleneck" and therefore could be the best adapted, known as "SuperAgers" (Harrison et al., 2012; Gefen et al., 2015). In this respect, some authors showed that these nonagenarians have immunological and antioxidant defenses and even episodic memory function equal or better than middleaged individuals (Moroni et al., 2005; Harrison et al., 2012; Gefen et al., 2015).

TABLE 1 | Pearson coefficient of correlation between neuronal density and Braak's staging.


CA1–4, Hippocampal areas; EC, entorhinal cortex; FC, frontal cortex; \*\*, correlation exists with p < 0.001; 0, no correlation; r, correlation coefficient.

Unlike aging, the loss of neurons in AD pathology may be due to a number of features that could facilitate or prone neurodegeneration. In this sense, the presence of SP, resulted from the abnormal extracellular accumulation and deposition of the amyloid-β peptide (40 or 42 amino acids), seems to promote neuronal death (Serrano-Pozo et al., 2011a). Several studies have demonstrated that mature SP are associated with deleterious effects on the surrounding neuropil i.e., increases of dystrophic neurites, recruitment and activation of astrocytes and microglial cells, synaptic loss, and neuronal death (Itagaki et al., 1989; Masliah et al., 1990, 1994; Pike et al., 1995; Knowles et al., 1999; Urbanc et al., 2002; Vehmas et al., 2003). Moreover, it is shown that the number of neurons per area is lower near to the mature SP than in distal areas in human and primate brains. The findings described by Shah et al. (2010) unequivocally reflect that while diffuse SP are commonly found in the brain of cognitive intact elderly people, dense-core plaques are most often in patients with AD dementia (Shah et al., 2010). In this work, we have observed that SP content increases as the disease progresses, which would explain the reduced neuronal density of these individuals.

Besides the SP, the presence of NFT, described as intraneural filamentous inclusions whose major constituent is aberrantly misfolded and abnormally hyperphosphorylated protein tau, correlates with the severity of the pathology and consequently with the neuronal loss in AD (Bierer et al., 1995; Gómez-Isla et al., 1997; Giannakopoulos et al., 2003), as we described in the Results section. However, whether NFT formation is a necessary precursor of the neuronal death or represents a protective response of damaged neurons in AD is still controversial.

Reactive astrocytes and activated microglial cells are commonly associated to dense-core SP, indicating, according with some authors, that amyloid-β peptide accumulation is the major trigger of this glial response (Itagaki et al., 1989; Pike et al., 1995; Vehmas et al., 2003). Our findings demonstrated an increase in this reactive cells with aging and with the Braak's stages. However, it has been recently described that the linear increase in reactive astrocytes and activated microglial cells through the entire disease course does not correlate with the amyloid deposition in the temporal associative isocortex (Serrano-Pozo et al., 2011b). Indeed, it has found a highly significant positive correlation between astrocytosis or microgliosis and NFT burden but not with

FIGURE 6 | Morphological brain changes in men and women during AD progression. Representative microphotographs of human brain sections of individuals with AD pathology, contrasted with a silver technique of Reusche (A,B) and a modification of Congo Red method (C,D). (A) Diffuse and mature plaques in the entorhinal cortex of an 80 years old man (Braak's stage II). (B) Senile plaques and neurofibrillary tangles in the hippocampus (CA2) of an 85 years old woman (Braak's stage VI). (C) Diffuse and mature plaques (red fluorescence) in the entorhinal cortex of an 80 years old man (Braak's stage II). (D) Senile plaques, in different maturation stages, and neurofibrillary tangles (red fluorescence) in the entorhinal cortex of an 80 years old man (Braak's stage VI). Bars: (A,C), 60µm; (B), 10µm; (D), 60µm.

amyloid concentration, suggesting that glial responses are related to neurofibrillary degeneration (Serrano-Pozo et al., 2011b).

As we have shown, there are little differences in neuronal density along aging in contrast to a clear loss of neurons in the progression of AD. According to our data, these differences not only depend on brain area but also on gender. In fact, in the CA3 field we found a decrease in the number of neurons per area in the non-pathological middle-aged men with respect to women. One possible explanation for this difference is the estrogen-mediated neuroprotection. Several authors pointed out that the high plasma levels of estradiol that pre-menopausal women show could exert a protective effect, preventing neuronal degeneration (Ishunina et al., 2007; Henderson, 2014). In this sense, in the first years after cessation of gonadal function women may show less accumulated damage. After menopause, women would become as vulnerable to neuronal degeneration as men, coinciding with the reduction in the estrogen production. In addition, sex differences were also found in the CA3 and CA4 hippocampal areas of the younger group. In these individuals, the neuronal density was higher in men than in women. This is in agreement with previous studies that show a higher neuronal content but a lower synaptic density in men than women in some brain areas (Allen and Gorski, 1991; Rabinowicz et al., 1999; Zaidi, 2010). It seems that the greatest amount of estrogens in brains of younger women could facilitate synaptic activity and optimize neuronal function, thus less neurons would be necessary to perform the same task (Aloisi et al., 1997; Brandt et al., 2013; Fester and Rune, 2015).

In AD subjects, a decrease in the density of neurons in relation to the stage of disease has been observed in the CA1 field in women but not in men. On the contrary, in the CA3 field, are men who lost more neurons. These observations are consistent with a region-specific neuronal loss in AD, which depends on the sex of the individuals and the stage of the pathology (Zaidi, 2010; Padurariu et al., 2012). The decrease in the neuronal population in AD may reflect the fact that the activation of antioxidant or compensatory mechanisms in both, women and men, cannot prevent neuronal loss in the brain. This phenomenon, together with the proven existence of sexual dimorphism in certain brain regions (Aloisi et al., 1997), could explain our results regarding gender differences in the pattern of neuronal degeneration in AD hippocampus; women seem to be more vulnerable in the CA1 area, while men in the CA3. So far, it has been described that the region-specific neuronal loss of this pathology occurs mainly in the CA1 brain area in both sexes (Adachi et al., 2003; Kerchner et al., 2010).

In summary, our findings clearly demonstrate that the neuronal loss that takes place during aging and to a greater degree in AD depends on the brain region and also varies between women and men. Therefore, our study highlights the need to consider the gender of the individuals in future studies to discriminate sexual variations in the pathological progression.

#### AUTHOR CONTRIBUTIONS

EM-P contributed to writing the manuscript, performed and analyzed the neuronal density quantification. CO contributed to writing the manuscript and assisted in analysing the results and images processing. AN supervised the project and contributed to image processing. EdV performed part of staining procedures.

#### REFERENCES


JT contributed by formulating the initial hypothesis, directed the experiments performed and contributed to scientific discussions and manuscript writing.

#### FUNDING

This work was supported by FISS, Instituto de Salud Carlos III and FEDER (Fondo Europeo de Desarrollo Regional) (PI15/00601) and Principado de Asturias (SV-PA-13-ECOEMP-80) grants.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnagi. 2016.00213

Supplementary Figure 1 | (A) Schematic representation of hippocampal areas. (B) Representative microphotography of human hippocampus. DG, dentate gyrus.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Martínez-Pinilla, Ordóñez, del Valle, Navarro and Tolivia. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Six Years of Research on the National Institute of Mental Health's Research Domain Criteria (RDoC) Initiative: A Systematic Review

Dean Carcone and Anthony C. Ruocco\*

Departments of Psychology and Psychological Clinical Science, University of Toronto, Toronto, ON, Canada

Six years have passed since the National Institute of Mental Health (NIMH) in the United States launched the Research Domain Criteria (RDoC) initiative. The RDoC introduces a framework for research on the biology of mental illness that integrates research findings across multiple levels of information. The framework outlines constructs that represent specific quantifiable dimensions of behavior (e.g., responses to acute threat, cognitive control) and corresponding units of analysis that can be used to study the constructs, beginning at the levels of genes, molecules, cells, circuits and physiology, and moving up to behaviors and self-reports. In this systematic review, a literature search was conducted to synthesize empirical research published since the proposal of the framework that incorporated the RDoC. Forty-eight peer-reviewed scholarly articles met eligibility criteria for the review. Studies differed according to whether they analyzed RDoC constructs and units of analysis within vs. between clinicallydiagnosed and non-psychiatric samples. The most commonly studied constructs were subsumed within the domains of Negative Valence Systems, Positive Valence Systems and Cognitive Systems, providing initial results which primarily connected genetics, brain circuits and physiology research findings with behavior and self-reports. Prospects for future research adopting the RDoC matrix and utilizing a dimensional approach to studying the biology of mental illness are discussed.

Edited by: Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Wen-Jun Gao, Drexel University College of Medicine, USA Vladeta Ajdacic-Gross, University of Zurich, Switzerland

> \*Correspondence: Anthony C. Ruocco anthony.ruocco@utoronto.ca

Received: 25 October 2016 Accepted: 13 February 2017 Published: 03 March 2017

#### Citation:

Carcone D and Ruocco AC (2017) Six Years of Research on the National Institute of Mental Health's Research Domain Criteria (RDoC) Initiative: A Systematic Review. Front. Cell. Neurosci. 11:46. doi: 10.3389/fncel.2017.00046 Keywords: research domain criteria (RDoC), mental disorders, cells, circuits and systems, physiology, brain, neuroscience

#### INTRODUCTION

The Research Domain Criteria (RDoC) framework was unveiled by the National Institute of Mental Health (NIMH) in a commentary published in the American Journal of Psychiatry (Insel et al., 2010). In response to concerns over the validity of the diagnostic criteria espoused in the Diagnostic and Statistical Manual (DSM) and what some consider inadequate efforts to address these concerns in the process of revising the manual in the DSM-5, the RDoC was proposed as an alternate framework to conceptually organize and direct biological research on mental disorders (Cuthbert and Insel, 2013; Insel, 2013). This framework encourages research to be structured around five major domains: Negative Valence Systems, driving reactions to aversive stimuli; Positive Valence Systems, driving reactions to positive stimuli; Cognitive Systems, including various mental processes; Social Processes, responsible for interpersonal behavior and cognition; and Arousal and Regulatory Systems, involved in context-based and homeostatic regulation or neural systems. Research on these systems and processes is organized around a dimensional approach incorporating and integrating the following levels or units of analysis: genes, molecules, cells, circuits, physiology, behavior and self-report. By re-orienting research away from DSM categories and toward a multimodal dimensional framework based on empirically validated constructs, the aim of the NIMH's RDoC initiative has been to progress further understanding of these domains such that a new diagnostic nosology can be developed.

The initial structure and five domains of the RDoC Matrix were the product of a series of workshops arranged by the NIMH (see Development of the RDoC Framework (2016) on the NIMH website). Each of the domains is subdivided into constructs representing specific functional dimensions of behavior. For example, the Cognitive Systems domain encompasses the constructs Attention, Perception, Declarative Memory, Language, Cognitive Control and Working Memory. Some of these constructs are further divided into subconstructs, such as the division of Perception into Visual Perception, Auditory Perception, and Olfactory/Somatosensory/Multimodal Perception. These constructs and subconstructs are crossed with the seven RDoC units of analysis to form the skeleton of the RDoC Matrix. Cells of this matrix are filled with relevant empirically supported elements, describing relevant research topics within a construct which can be investigated at a given unit of analysis. For example, dopamine and serotonin are molecular elements of the Reward Valuation subconstruct of Approach Motivation, under the Positive Valence Systems domain. For an illustration of all divisions within the matrix, please refer to The RDoC Matrix (2016) on the NIHM website. The organization of the RDoC Matrix remains flexible, having undergone additions, modifications, and a recent redesign since its launch (RDoC Launches User-Friendly Matrix Format, 2016).

Six years have passed since the proposal of the RDoC framework, conceivably allowing sufficient time for researchers to design, conduct, and publish initial studies that adopt this innovative new framework. The purpose of this brief systematic review is to summarize all peer-reviewed studies published since the proposal of the framework that explicitly purport to incorporate at least one domain, construct, or element of the RDoC. The aim of the review article is to provide an overview of research on the RDoC to identify patterns in the domains, constructs, and units of analysis most commonly studied and the research designs that have been employed.

#### METHOD

#### Database Search

The ProQuest search engine was used to access the PsycINFO and Medline databases to return all studies indexed using the term ''Research Domain Criteria'' and/or the acronym ''RDoC''. The results were restricted to peer-reviewed work published in a scholarly journal after April 2013, the date that the RDoC was officially proposed. This search was conducted on April 4, 2016, and returned 330 unique results. Abstracts for these 330 studies were initially screened to remove reviews, commentaries, and results erroneously returned by the search (e.g., those pertaining to Refractory Dissolved Organic Carbon). Following this screen, 61 potentially relevant articles were selected to pass through a second screen following review of the full text of each manuscript. Please refer to **Figure 1** for a visual representation of article screening and selection.

#### Article Inclusion and Exclusion

In order for an article to be selected for inclusion in this review, it must have met the following inclusion criteria: (1) empirically examined one or more identified RDoC domains, constructs, or elements; and (2) made direct reference to the RDoC framework as part of the study rationale, interpretation of results, or implications and future directions. In order to be retained, an article must also not have met any of the following exclusion criteria: (1) the article represented a proposal for RDoC-related research that had not yet been conducted; (2) the focus of an article was the empirical validation of a psychometric measure related to an RDoC construct, rather than the examination of the construct itself; and (3) the article centered around a case study. Of the 61 articles selected for full-text review, the majority (48 articles) were found to be relevant primary empirical studies which explored themes consistent with the RDoC framework. These articles have been included in this review. A summary of the constructs and units of analysis explored, as well as the key results of these studies, have been included in a Supplementary Table provided with this review article. It should be noted that additional research closely related to RDoC constructs, although not explicitly identifying use of the RDoC framework, has been conducted during the previous 6 years. These studies have not been included in this review article.

#### RESULTS

## Common Experimental Designs

Many of the studies reviewed shared similar experimental designs and approaches to applying the RDoC framework.

The most common approach was to explore an explicitly defined RDoC construct, subconstruct, or element using two or more units of analysis. This often resulted in describing relationships between higher level behaviors or traits and lower level biological underpinnings. Additionally, many of these studies overtly employed a transdiagnostic approach by examining a given construct across more than one group defined by clinical diagnosis. Another common approach was to explore associations between constructs or subconstructs within or across RDoC domains using one or more units of analysis. Alternatively, a number of studies employed a dimensional approach to study psychological or behavioral phenotypes related to established RDoC domains but focusing on constructs not explicitly defined within the RDoC matrix. Finally, two studies explored the application of the RDoC to the design of clinical interventions. Please refer to the Supplementary Table for additional information on the experimental designs and results of all studies reviewed.

#### Arousal and Regulatory Systems

Since the proposal of the RDoC framework, domains have not been equally represented in the literature. With the exception of six studies not specifying a particular domain of focus, almost all the research summarized in this review examined at least one construct in either the Cognitive Systems, Positive Valence Systems, or Negative Valence Systems domains. Only two studies (Tegeler et al., 2015; Olbrich et al., 2016) focused exclusively on the domain of Arousal and Regularity Systems, both targeting the Arousal construct. These studies also both used similar physiological measures to study clinical samples. Tegeler et al. (2015) demonstrated that hemispheric dominance in high-frequency brain activity was related to baroreflex sensitivity and higher resting heart rate. Olbrich et al. (2016) showed that pre-treatment physiological measures predict remission from major depressive disorder (MDD) and response to SSRI treatment. Both of these studies highlight the potential of the RDoC for studying arousal systems relevant to psychopathology.

#### Social Processes

Similar to the Arousal and Regulatory Systems, only two studies exclusively examined the Social Processes Domain (Fang et al., 2014; Lindberg et al., 2015). Both of these studies explored the Affiliation and Attachment construct, although the experimental paradigms employed by each were quite distinct. Fang et al. (2014) recorded behavior on a social exclusion task in males with social anxiety disorder following administration of oxytocin. For males with lower self-reported attachment avoidance, oxytocin promoted social affiliation and cooperation. In contrast, Lindberg et al. (2015) described how similar scales of the Attachment and Clinical Issues Questionnaire (Lindberg and Thomas, 2011) are associated with both current alcohol dependence in adults and risk of future alcohol dependence in high school students. Although research has now examined the Affiliation and Attachment construct within the context of the RDoC, the remaining constructs of Social Communication, Perception and Understanding of Self, and Perception and Understanding of Others have yet to be independently explored.

#### Positive Valence Systems

Twelve publications included in this review article, examined Positive Valence Systems, although none of these explored the constructs of Habit or Initial/Sustained Responsiveness to Reward Attainment. Rather, research has primarily focused on reward-related constructs such as approach motivation (Karalunas et al., 2014) and reward learning (Webb et al., 2016). For example, Sharp et al. (2014) used fMRI to compare brain activation patterns during a card-guessing reward task between three female groups: depressed adolescents with a maternal history of depression, never-depressed adolescents with a maternal history of depression, and never-depressed adolescents with no maternal history of depression. Lower right ventral striatal activity was observed for both currently depressed and at-risk adolescents. Additionally, activity in this region was inversely correlated with maternal Beck Depression Inventory (BDI) scores across all groups. These results highlight the potential for disruptions in Positive Valence Systems to be used as identifiable vulnerability factors for the development of MDD.

Arrondo et al. (2015) employed a similar transdiagnostic approach in the examination of circuitry related to the Expectancy construct. Patients with schizophrenia, MDD, and controls completed a task to induce anticipation of monetary reward while undergoing fMRI. Reduced activation associated with reward anticipation was observed in the bilateral ventral striatum for both patient groups compared to controls. This reduction in activation was related to self-reported anhedonia and overall depression symptoms for patients with schizophrenia but not MDD. Through their application of the RDoC, these two studies showcase the potential for transdiagnostic and dimensional examination of psychological constructs.

#### Cognitive Systems

Research on Cognitive Systems has been considerable and diverse, with eighteen studies published on four of the six constructs within the domain (e.g., Cognitive Control: Lopez-Garcia et al. (2016); Perception: Silverstein et al. (2015); Attention: Kleinman et al. (2015); and Working Memory: Francazio and Flessner (2015)). To elaborate on one example, Chan et al. (2015) demonstrated how a standard performancemonitoring event-related potential paradigm can be used within an RDoC framework to examine Cognitive Control and Performance Monitoring systems. Adults with a history of psychosis and a control group completed a flanker task while undergoing EEG. Reduced error-related negativity and error positivity amplitudes were observed for individuals with a history of psychosis compared to controls. Additionally, reduced error positivity amplitude was associated with schizotypal personality traits across both groups, suggesting that these disruptions in neural circuitry may represent a transdiagnostic phenotype.

As an additional example of a study employing multiple units of analysis to explore a Cognitive System, Newman et al. (2016) focused on the Response Selection construct by examining the relationship between cortical thickness and performance on a go/no-go task. In a sample of 114 adults, 46% of which exhibited symptoms of Attention Deficit Hyperactivity Disorder (ADHD) persisting into adulthood, thickness of the caudal inferior frontal gyrus was associated with poorer response inhibition (i.e., more commission errors) regardless of ADHD symptoms or history of substance abuse. Of note, cortical thickness of this area was also inversely correlated with frequency of cannabis use and persistence of ADHD symptoms.

A transdiagnostic approach to the study of the Visual Perception Cognitive System was adopted by Silverstein et al. (2015). This work investigated disturbances in perceptual organization across patients with body dysmorphic disorder (BDD) and schizophrenia, which are two patient groups defined in part by perceptual distortions. Performance on tasks of perceptual organization was compared between these groups, non-patients, and an obsessive-compulsive disorder (OCD) group. Only patients with schizophrenia performed worse on these measures than comparison groups, while patients with BDD did not differ from non-patients or OCD patients. Consequently, it was suggested that the disturbances in visual perception found in BDD may be unrelated to problems in perceptual organization. Furthermore, these results are consistent with the conclusion that disturbances in perceptual organization may be more specific to schizophrenia and other neurodevelopmental disorders.

Rather than focusing on an RDoC construct, one study examined a single element of the Perception construct within the Cognitive Systems matrix. Chen et al. (2016) explored the relationship between the expression of the dysbindin protein, found in neural tissue, and disturbances in the regulation of lipid synthesis and synaptic plasticity. A reduction of sterol regulatory element binding protein-1 (SREBP1), a transcriptional regulator for lipid synthesis, was found in both deceased patients with schizophrenia and dysbindin-1 knockout mice. Additional results suggest that this disturbance in SREBP1 maturation may lead to a disruption in synaptic plasticity, and these disruptions may be corrected in knockout mice with the administration of the anti-psychotic drug clozapine. This multimodal approach illustrates the goals of the RDoC framework, using multiple levels of analysis to provide a comprehensive empirical characterization of clinically relevant phenotypic variation.

#### Negative Valence Systems

The domain most commonly explored by the research collected in this systematic review was Negative Valence Systems, with 20 publications examining at least one construct within this domain. With the exception of Frustrative Non-Reward, four of the five constructs within this domain were investigated by at least one study (e.g., Acute Threat: Yancey et al. (2016); Potential Threat: Latzman et al. (2016); Sustained Threat: Weinberg et al. (2016); and Loss: Woody et al. (2014)), with Acute Threat the most commonly studied construct. As an example, Bauer et al. (2013) explored Acute Threat using both physiological and self-report measures. The Clinician Administered Posttraumatic Stress Disorder (PTSD) Scale (Blake et al., 1995) was administered to 36 adults reporting prior exposure to a traumatic event, a sample in which the Acute Threat system may be compromised. Additionally, these adults underwent physiological recordings, including heart rate, skin conductance, and eye-blinks, while performing a script-driven emotional imagery task. Measures of physiological reactivity, thought to be sensitive to disturbances in the Acute Threat system, correlated significantly with severity of PTSD symptoms at an initial visit and at a six-month follow-up assessment. This was taken to suggest that physiological reactivity is a stable and valid measure of disturbances in the Acute Threat construct associated with PTSD. Furthermore, Pineles et al. (2013) demonstrated that these physiological measures are a better predictor of PTSD diagnosis than self-report measures of emotional response.

An additional noteworthy study used a population of 76 captive chimpanzees to investigate the Potential Threat construct as it relates to genes, physiology and anxious behavior (Latzman et al., 2016). It was shown that scratching behavior, an indicator of negative arousal, exhibited a sex-specific association with both AVPR1A genotype, a gene related to mammalian social behavior, and brain morphometry in regions associated with this gene. Although only three studies (Chen et al., 2016; Kondo et al., 2016; Latzman et al., 2016) employed non-human mammals in their research, they highlight how animal research can contribute to the understanding of species-general elements within the RDoC matrix.

A number of studies explored topics related to established domains, but focused on constructs not explicitly defined as a construct in the RDoC matrix, such as neuroticism (i.e., the tendency to experience negative affect; Bedwell et al., 2014) and anhedonia (i.e., the inability to feel pleasure; Webb et al., 2016). One of these studies, Østergaard et al. (2014), overtly proposed that melancholia should be included as an additional construct within the Negative Valence Systems domain. Analysis of self-report measures of depression symptoms in a large treatment study of depression suggested a common and unidimensional construct reflecting melancholia is sensitive to pharmacological intervention. According to the authors, the results of this and prior work highlight how melancholia meets the criteria for inclusion as an RDoC construct, as stipulated in Cuthbert and Insel (2013). As the RDoC matrix is a work in progress, it will be interesting to see how proposals like this can influence its development.

#### Research Exploring More than One Domain

Ten of the studies reviewed examined more than one RDoC domain, commonly exploring associations between constructs or subconstructs across multiple domains. Verona and Bresin (2015), for example, examined the relationship between threatrelated Negative Valence Systems and the Cognitive System subconstruct of Response Selection. Adults with a history of violent and criminal offences completed the Buss Perry Aggression Questionnaire (Buss and Perry, 1992), which was taken as an index of proneness to aggression associated with Sustained and Acute Threat responses. Participants then underwent EEG while completing an emotional-linguistic go/no-go task. During blocks presenting threatening words, self-reported anger and aggressive behavior each were associated with smaller P3 amplitude recorded during response inhibition, an electrophysiological marker that is taken to represent reduced inhibitory control processing. It is suggested that this relationship may indicate that sensitivity to acute or sustained threat may deteriorate cognitive control processes.

An additional example of research spanning across multiple domains is work which explored the application of the RDoC to the design of clinical interventions. ''Engage'', first described in Alexopoulos and Arean (2014), is a psychological intervention designed to specifically target disturbances in RDoC-defined Positive and Negative Valence Systems through reward exposure, the facilitation of meaningful and rewarding activities. A sample of older adults with MDD received 9 weeks of Engage (Alexopoulos et al., 2015). Symptom improvements within this group were compared to a patient group from a previous study (Alexopoulos et al., 2011), in which participants received problem-solving therapy (PST). The Engage intervention demonstrated comparable efficacy to PST in reducing symptoms and functional impairment associated with MDD. In order to explore the mechanism underlying this change, Alexopoulos et al. (2016) examined how MDD symptoms and a self-reported rating of behavioral activation influenced each other throughout Engage. In a sample of 48 older adults with MDD completing 9 weeks of Engage, greater increases in behavioral activation, taken as a broad index of Positive Valence Systems function, predicted change in depression severity during treatment and at a 36-week follow-up assessment. These investigations may mark the first of many treatment studies based on targeting dimensions included in the RDoC Matrix.

# DISCUSSION

With 6 years having elapsed since the RDoC framework was introduced, it is timely to review findings from the first wave of research which adopted the framework. This systematic review identified 48 primary empirical articles published in the last 4 years (see Supplementary Table) which either directly examined an RDoC-defined construct, or interpreted their results in a manner consistent with the RDoC framework. As described above, the majority of these studies fall into common prototypes regarding their goals and their approach to applying the RDoC framework. At present, it seems that an ''RDoC study'' describes: (1) research on a single defined RDoC construct using multiple levels of analysis; (2) research which explores associations between separate RDoC constructs and/or elements; and/or (3) research which adopts a transdiagnostic perspective to investigate an RDoC construct by examining multiple categories of a disorder or symptom dimensions. Some additional studies purport to use approaches consistent with the RDoC framework but incorporate phenotypes which are not (or may not yet have been) specified within the RDoC matrix. We do not know if RDoC studies will continue to fall within these rough categories, but the approaches that they represent are informative for the design of subsequent research. The topics of focus for these future studies will likely be influenced by the growth and modification of the RDoC matrix and by significant advances in research made within this framework. Whether this pattern of research reflects the way future studies will be designed likely depends on subsequent modifications to the RDoC matrix and the progression of research during these formative years of the framework.

This systematic review highlights a notable scarcity of research explicitly intended to examine the RDoC domains of Social Processes and Arousal and Regulatory Systems. This is of course not to say that research in these and related areas has not been carried out but that the RDoC framework specifically has not been adopted in publications of this work. Conversely, a remarkable number of studies have been published which explore the domains of Cognitive Systems, Negative Valence Systems, and Positive Valence Systems. It is unlikely that this disparity is an artifact of the database search used in this review, as no domain-specific terms were included as search criteria. It may be speculated that constructs related to the more well-represented domains may currently be easier to cast within the RDoC framework, prompting a greater number of researchers to incorporate this new perspective. For example, research on attention may lend itself more easily to be examined from a multi-level biological and behavioral perspective, while work of this type may be less common when examining topics such as self-perception or non-verbal communication. Different fields of study may also have been quicker or slower to adopt the RDoC framework depending at least in part on the established theories in the field that emphasize or deemphasize biological contributions to the constructs of interest. This may be reflected in the relative number of unoccupied cells in the lower-level units of analysis present within the RDoC matrix at the time of this review (e.g., Cognitive Systems: 17% unoccupied vs. Social Processes: 34% unoccupied; see The RDoC Matrix (2016)). However, this does not explain the lack of research using the RDoC framework to investigate Arousal and Regulatory Systems, a domain that is well populated with respect to RDoC units of analysis. Rather, it can be argued that this domain is historically better understood than other domains (although perhaps not in relation to self-report constructs relevant to mental and neurological illness), and consequently has been the focus of fewer research studies since the introduction of the RDoC Framework only 6 years ago.

Although the transition to a dimensional and empiricallydriven research framework on mental illness has been anticipated for many years (Krueger and Piasecki, 2002), the RDoC proposal has been met with a mixture of support, resistance and controversy. For a discussion of the challenges facing the RDoC framework, please refer to Patrick and Hajcak (2016). It has been suggested that all authoritative systems of nosology (including both the DSM and the RDoC) likely impeded the development of scientific theory by constraining competitive discourse (Markon, 2013). Additionally, having more than one active system of nosology may promote fractures in research continuity between funding bodies and geographic regions. Although researchers in the United States are in part incentivized to adopt an RDoC framework by receiving funding from the NIMH, researchers outside the region may not yet have as strong of an incentive to do so and therefore may be less likely to make the transition. Conversely, many researchers are optimistic about the promise of dimensional approaches to studying mental illness, components of which are inherent to many aspects of the RDoC framework. Indeed, there is potential for the framework to provide a useful alternative to structuring clinical research on mental illness (Casey et al., 2013), although the existing body of research incorporating this framework is relatively small and requires further advancement to refine the model and realize the potential of this transformative new approach.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

DC and ACR conceptualized the topic, reviewed studies for eligibility in the review, and wrote the manuscript. DC conducted the literature search.

#### FUNDING

ACR is supported by a New Investigator Salary Award (MSH-130177) from the Canadian Institutes of Health Research and an Early Researcher Award (ER14-10-185) from the Ministry of Research and Innovation, Province of Ontario. DC is supported by an Ontario Graduate Scholarship.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fncel. 2017.00046/full#supplementary-material


depressed patients completing an alternate milieu inpatient program. Gen. Hosp. Psychiatry 37, 601–605. doi: 10.1016/j.genhosppsych.2015. 06.006


a function of mild depressive symptoms: a voxel-based morphometric analysis. Psychol. Med. 44, 2833–2843. doi: 10.1017/S0033291714000348


moderating impact of depression. Int. J. Psychophysiol. 98, 262–269. doi: 10.1016/j.ijpsycho.2014.10.014

Yancey, J. R., Venables, N. C., and Patrick, C. J. (2016). Psychoneurometric operationalization of threat sensitivity: relations with clinical symptom and physiological response criteria. Psychophysiology 53, 393–405. doi: 10.1111/psyp.12512

**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Carcone and Ruocco. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Molecular Mechanisms of Bipolar Disorder: Progress Made and Future Challenges

#### Yeni Kim1,2† , Renata Santos 1,3† , Fred H. Gage<sup>1</sup> and Maria C. Marchetto<sup>1</sup> \*

<sup>1</sup>Laboratory of Genetics, The Salk Institute for Biological Studies, La Jolla, CA, USA, <sup>2</sup>Department of Child and Adolescent Psychiatry, National Center for Mental Health, Seoul, South Korea, <sup>3</sup>Ecole Normale Supérieure, PSL Research University, Centre National de la Recherche Scientifique (CNRS), Institut National de la Santé et de la Recherche Médicale (INSERM), Institut de Biologie de l'Ecole Normale Supérieure (IBENS), Paris, France

Bipolar disorder (BD) is a chronic and progressive psychiatric illness characterized by mood oscillations, with episodes of mania and depression. The impact of BD on patients can be devastating, with up to 15% of patients committing suicide. This disorder is associated with psychiatric and medical comorbidities and patients with a high risk of drug abuse, metabolic and endocrine disorders and vascular disease. Current knowledge of the pathophysiology and molecular mechanisms causing BD is still modest. With no clear biological markers available, early diagnosis is a great challenge to clinicians without previous knowledge of the longitudinal progress of illness. Moreover, despite recommendations from evidence-based guidelines, polypharmacy is still common in clinical treatment of BD, reflecting the gap between research and clinical practice. A major challenge in BD is the development of effective drugs with low toxicity for the patients. In this review article, we focus on the progress made and future challenges we face in determining the pathophysiology and molecular pathways involved in BD, such as circadian and metabolic perturbations, mitochondrial and endoplasmic reticulum (ER) dysfunction, autophagy and glutamatergic neurotransmission; which may lead to the development of new drugs.

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Haim Einat, Tel Aviv-Yaffo Academic College, Israel James P. Kesby, University of Queensland, Australia

#### \*Correspondence:

Maria C. Marchetto marchetto@salk.edu †These authors have contributed equally to this work.

Received: 28 November 2016 Accepted: 01 February 2017 Published: 14 February 2017

#### Citation:

Kim Y, Santos R, Gage FH and Marchetto MC (2017) Molecular Mechanisms of Bipolar Disorder: Progress Made and Future Challenges. Front. Cell. Neurosci. 11:30. doi: 10.3389/fncel.2017.00030 Keywords: bipolar disorder, mitochondrial dysfunction, endoplasmic reticulum stress, oxidative stress, glutamate, hyperexcitability, disease modeling

#### INTRODUCTION

In his 1889 lecture ''Zur Diagnose und Prognose der Dementia praecox'', Emil Kraepelin proposed separating psychiatric disorders with psychotic features into two major categories. Based on the observations of the long-term outcome and the nosological principles of Kahlbaum (1863), the famous diagnostic dichotomy was born: the ''Manisch-depressives Irresein'' which was later reclassified to bipolar disorder (BD) and major depression, and the ''Dementia-praecox-Gruppe'' which became schizophrenia. BD is a complex syndrome with 2% prevalence worldwide (Merikangas et al., 2011). The impact of BD on patients can be devastating; 9%–15% of patients commit suicide (Rihmer and Kiss, 2002; Medici et al., 2015). This disorder is associated with psychiatric comorbidities including personality disorder, anxiety disorder and substance abuse disorder and medical comorbidities such as diabetes, obesity and hyperlipidemia (Leboyer et al., 2012; Blanco et al., 2017).

Even with typical symptoms of BD, the disease is difficult to diagnose accurately and promptly in clinical practice. Both BD types I and II patients spend most of the duration of their illness in a depressive phase (Hirschfeld et al., 2003) and they often fail to recognize hypomanic or manic symptoms as pathological, which results in a mean delay of 5–10 years between the onset of illness and diagnosis (Baldessarini et al., 2007). There are also subtle differences between the two major diagnostic criteria used throughout the world today, the DSM and the International Classification of Disease (ICD). According to DSM-5 criteria, BD type I is diagnosed when there has been at least one episode of full-blown mania, with or without one or more major depressive or hypomanic episodes (American Psychiatric Association, 2013). A diagnosis of BD II is based on several protracted episodes and at least one hypomanic episode but no manic episodes. The ICD-10 does not discriminate between BD types I and II (WHO, 1993). While one episode of mania or one episode of hypomania plus major depressive episodes is sufficient for diagnosis according to DSM-5, the ICD-10 requires at least two distinct mood episodes, one of which must be manic or hypomanic for the diagnosis of BD. With no clear clinically relevant biological markers available, early diagnosis is a great challenge for clinicians without knowledge of the longitudinal progress of illness (Phillips and Kupfer, 2013).

Treatment of BD usually consists of two stages: acute stabilization and relapse prevention. Acute stabilization entails the conversion of a manic or depressive phase to an euthymic state; relapse prevention consists of maintaining the euthymic status while minimizing subthreshold symptoms and enhancing general function (Geddes and Miklowitz, 2013; Goodwin et al., 2016). Acute treatment of BD is complex, as one mood phase may spill over into the opposite mood phase before the euthymic status can be achieved, complicating the various aspects of clinical decisions e.g., the choice of psychotropics and the dosage. Also teasing out the therapeutic effects from the possible adverse effects such as somnolence, psychomotor retardation and akathisia, all of which may mimic a change in the mood status, is one of the biggest challenges during the initial phase of BD drug treatment. A meta analysis of short-term randomized control trials of medications showed that aripiprazole, asenapine, carbamazepine, cariprazine, haloperidol, lithium, olanzapine, paliperdone, quetiapine, risperidone, tamoxifen, valproate and ziprasidone were effective as acute anti-manic agents (Yildiz et al., 2011). The same study also showed that responses to various antipsychotics were somewhat greater or more rapid than lithium, valproate, or carbamazepine and that lithium did not differ from valproate in a direct comparison between the drugs (Yildiz et al., 2011). Relapse prevention usually necessitates long-term treatment that calls for drugs that have minimal long-term side effects. Lithium is the best-established long-term treatment compound for BD, reducing both relapse and suicide (Geddes et al., 2004; Nivoli et al., 2010; Rybakowski, 2014). However, the incidence of adverse effects and a low therapeutic index restrict its benefits. Despite recommendations from evidence-based guidelines, polypharmacy is still common in clinical practice of BD, reflecting the gap between research and routine clinical practice (Fornaro et al., 2016).

The reduced understanding of the underlying pathophysiology and neurobiology of the disorder hampered the development of effective drugs. Neuroimaging studies have consistently revealed structural changes in the brain of BD patients (Maletic and Raison, 2014). In addition, observation of post-mortem tissue showed histopathologic features in neurons and glia in BD (Rajkowska, 2000; Uranova et al., 2004). These findings encouraged moving the research from monoamine neurotransmission to the synaptic and neural plasticity and the cellular processes that control the physiology of brain cells. In this review article, we will focus on how alterations in the energetic metabolism and mitochondrial dysfunction contribute to the vulnerability of BD cells, which we expect may lead to future therapies.

#### CIRCADIAN AND METABOLIC PERTURBATIONS IN BD

The clinical manifestations and the pathogenesis of BD are linked to circadian rhythm alteration (Melo et al., 2016a,b; **Figure 1**). Circadian disruptions and sleep complaints can be both precipitating factors and consequences of mood disorders (Bechtel, 2015; Cretu et al., 2016; Grierson et al., 2016). One of the main characteristics of manic episodes is the reduced need for sleep, whereas depressive episodes are frequently characterized by insomnia and hypersomnia (American Psychiatric Association, 2013). Circadian disruption and ''eveningness'' (being more active during the evening) have been associated with mood episodes, functional impairment, poor quality of life and treatment resistance (Duarte Faria et al., 2015; Pinho et al., 2015; Cudney et al., 2016; Ng et al., 2016). Moreover, sleep deprivation and light therapy are therapeutic approaches that have been used effectively as adjuncts to the more standard pharmacological therapies (Lewy et al., 1982; Benedetti et al., 2014; Tseng et al., 2016).

Existing hypotheses for the biological mechanism underlying dysregulation of circadian rhythm in BD include changes in melatonin levels, in expression of melatonin receptors in the central nervous system and in daily cortisol profiles (Wu et al., 2013). Genetic evidence also links circadian rhythm dysregulation with BD. Two polymorphisms on the CLOCK gene that control circadian rhythm—aryl hydrocarbon receptor nuclear translocator-like (ARNTL) and timeless circadian clock (TIMELESS)—have been linked to lithium responsiveness in BD (Rybakowski et al., 2014). In addition, Per2, Cry1 and Rev-Erbα expression, all components of the circadian clock, increased the individual susceptibility to the therapeutic effects of lithium (Schnell et al., 2015).

It is interesting to note that circadian rhythm dysregulation and molecular clock mechanism are observed across psychiatric diagnoses, including schizophrenia and depression (Lamont et al., 2007). In a genome-wide association analysis of UK biobank, genetic correlation between longer sleep duration and schizophrenia risk was observed (Lane et al., 2017). In addition, a SNP analysis showed CLOCK gene T3111C polymorphism in Japanese schizophrenia patients compared to healthy controls (Takao et al., 2007), although the same polymorphism was not observed in patients with major depressive disorder (MDD)

or BD in another study of the Japanese population (Kishi et al., 2011). Recently, it was observed in primary fibroblasts from schizophrenic patients with poor sleep, a loss of rhythmic expression of CRY1 and PER2 when compared to healthy controls (Johansson et al., 2016). Increased sleep latency, poor sleep quality and reduced latency to first rapid eye movement sleep are well documented in MDD, but there is no data supporting the role of circadian rhythm genes in the disorder (Thase, 2006).

Metabolic cues contribute to the regulation of circadian clocks and circadian rhythm impact the cardiovascular and metabolic systems (Morris et al., 2012; Gamble and Young, 2013; Bailey et al., 2014). Indeed it has long been known that BD patients present energetic metabolism changes (Altschule et al., 1956; Kato and Kato, 2000) and that they have a higher risk of obesity (Boudebesse et al., 2015) and type 2 diabetes mellitus compared to the general population (McElroy et al., 2002; Keck and McElroy, 2003; McIntyre et al., 2005). Systemic analysis have shown that natural causes like cardiovascular illness contribute significantly to the decrease in the life expectancy of BD patients compared to the general population (Kessing et al., 2015). Circadian disturbance appears to be independently associated with increased lipid peroxidation in BD patients but not in controls (Cudney et al., 2014). An association between evening chronotype and a higher percentage of body fat composition among patients with BD has been suggested (Soreca et al., 2009). Many cellular metabolic sensors act directly on core components of the clock, adjusting biological timing with metabolic status. Leptin, which is produced by adipocytes, regulates appetite and modulates sleep duration. Increased levels of leptin have been described previously in overweight patients with BD as compared with overweight controls (Barbosa et al., 2012). Adipose tissue-derived hormones, or adipokines, regulate appetite and metabolism and have activity in limbic brain regions; mood episodes and medication treatment both contribute to adipokine abnormalities in BD and adipokines influence the course of psychiatric illness and changes in BMI (Bond et al., 2016).

The mechanisms underlying weight gain and metabolic imbalance in BD patients are poorly understood. Genetic susceptibility, recurrent depressive episodes, low activity levels, poor dietary habits, poor medical care, and side effects of antipsychotics/mood-stabilizers medication have been suggested (McElroy et al., 2002). Sleep disturbance is a core symptom of BD and may contribute to the association of BD with metabolic disturbances. Evidence indicates that shorter sleep duration is associated with low HDL cholesterol (Soreca et al., 2012) and increased risk of coronary events (Ayas et al., 2003). However, other studies suggest that comorbid medical illnesses of BD may not only be due to poor health behaviors and psychotropic medications, but manifestations of common biological pathways between the BD and the comorbid illnesses (Leboyer et al., 2012). The close associations between metabolic and psychiatric disorders have introduced the ''metabolic mood syndrome'' hypothesis, which speculates the existence of common biological mechanisms underlying both conditions.

The innate energetic glucose-dependent brain metabolism may be one of the factors that contribute to this phenomena. Our brain has a very high-energy requirement and will disturb other parts of the body to acquire its energetic need, which is the core of the ''selfish brain'' hypothesis (Peters et al., 2004). We can hypothesize that some of the metabolic changes observed in BD may be a compensatory mechanism of the body trying to offset the pathological energy imbalance during the initial stages of BD characterized by loss of appetite, increased energy and lack of sleep, all of which would deplete the body and brain of energy sources. Maybe it is not a coincidence that many of the initial side effects of antipsychotics and mood stabilizers (appetite increase and somnolence) used to treat manic episodes, point toward the body trying to tip the scale toward anabolism instead of catabolism. The previous findings have suggested that high BMI impacts negatively on clinical and functional outcomes in BD (Kolotkin et al., 2008), adversely influencing treatment response to mood stabilizers and remission rate (Kemp et al., 2010). However, higher weight gain may be a compensatory response to more severe pathological process than the cause of the negative clinical outcome and poor drug response. Although a shared risk and overlapping pathophysiology implicate either shared biological mechanisms or causal interactions for circadian rhythm, metabolic disturbances and BD, more research is needed to study the specific mechanisms in place.

## MITOCHONDRIAL DYSFUNCTION AND ENERGY METABOLISM

Accumulating evidence from imaging, biochemical and genetic studies support the view that mitochondrial dysfunction is a central feature in BD (Kato and Kato, 2000), characterized by impaired oxidative phosphorylation and changes in mitochondrial morphology and number and in calcium signaling (**Figure 1**). Several mitochondrial DNA polymorphisms have been described (Kato et al., 2001; Munakata et al., 2004), providing additional support for the association between BD and mitochondrial impairment.

Magnetic resonance spectroscopy (MRS) studies were the first to show perturbations in several pathways involved in energy metabolism in BD patients. In a pioneering study using phosphorous <sup>31</sup>P-MRS, Kato et al. (1994) identified reduced phosphocreatinine in the frontal cortex of BD patients regardless of mood phase; this finding was later confirmed by other studies using a different cohort (Deicken et al., 1995; Frey et al., 2007). Phosphocreatinine is a cellular reservoir for ATP synthesis in periods of intense metabolic demand, and a chronic decrease in phosphocreatinine levels is an indication of mitochondrial dysfunction and deficient ATP synthesis. Inorganic phosphate regulates oxidative phosphorylation and ATP synthesis (Brown, 1992; Bose et al., 2003). Based on early pioneering studies, Stork and Renshaw (2005) proposed a model of mitochondrial dysfunction in which a metabolic shift towards glycolysis occurs in the brain of BD individuals (**Figure 2**). A recent <sup>31</sup>P-MRS studies in adolescents showed that inorganic phosphate was decreased in medication-free patients compared to medicated patients and controls (Shi et al., 2012), again emphasizing the low energy status of BD brain cells.

The tricarboxylic acid cycle (TCA) is a fundamental component of aerobic respiration and is also disturbed in BD (**Figure 2**). Metabolomic analysis showed that the serum levels of pyruvate and α-ketoglutarate were significantly higher in BD patients than in healthy controls (Yoshimi et al., 2016b). Pyruvate is an end product of glycolysis and is used to fuel the TCA cycle in the mitochondria after conversion into acetyl-CoA. α-Ketoglutarate is a TCA intermediate that results from the oxidative decarboxylation of isocitrate catalyzed by isocitrate dehydrogenase. The levels of this enzyme were found to be significantly higher in the cerebrospinal fluid (CSF) of BD patients compared to neurotypical controls, possibly accounting for the α-ketoglutarate increase (Yoshimi et al., 2016a). Studies using proton <sup>1</sup>H-MRS detected decreased intracellular pH and increased lactate in several brain regions of BD patients compared to healthy individuals (Kato et al., 1994; Dager et al., 2004; Chu et al., 2013). Accordingly, the lactate level in the CSF is higher in BD patients compared to healthy controls (Regenold et al., 2009). At the transcriptional level, different studies reported post-mortem decreased expression of genes encoding numerous subunits of complexes I, III, IV and V of the electron transport chain in the hippocampus (Konradi et al., 2004) and prefrontal cortex (Iwamoto et al., 2005; Sun et al., 2006; Andreazza et al., 2010) of BD patients. All of these observations converge to support the hypothesis that a metabolic shift occurs from oxidative phosphorylation to the less-efficient pathway glycolysis in the brain of BD individuals (**Figure 2**).

Mitochondria accomplish other cellular functions, such as the regulation of calcium homeostasis and of cell death (McBride et al., 2006; Giacomello et al., 2007; Suen et al., 2008; Bhosale et al., 2015). At the same time, mitochondria actively participate in the intracellular regulation of calcium signaling by buffering the calcium waves. Lethal challenges stimulate calcium release by the endoplasmic reticulum (ER) and uptake by mitochondria, which are early steps in the apoptotic cascade, and the capacity of mitochondria to handle calcium fluxes will determine survival or death (Giacomello et al., 2007; Bhosale et al., 2015; Raffaello et al., 2016). In general, an intensification in cellular energy demand is associated with increased calcium (Bhosale et al., 2015). Since BD patient cells have impaired oxidative phosphorylation, it is likely that they also have disturbed calcium homeostasis.

However, only a few studies have addressed these aspects of mitochondrial function in the context of BD. As expected, markers of apoptosis and an increase in the intracellular calcium concentration were found in blood cells from BD patients compared to healthy individuals (Perova et al., 2008; Dubovsky et al., 2014; Fries et al., 2014). Underscoring the potential role for calcium homeostasis in BD pathogenesis is the repeated identification of CACNA1C, which encodes the α-subunit of the L-type voltage-gated Ca2<sup>+</sup> channel, as a risk gene (Maletic and Raison, 2014).

Mitochondria undergo continuous fusion and fission events in physiological conditions. The imbalance of these two processes has dramatic effects on the morphology, physiology and distribution of mitochondria in the cells (Detmer and Chan, 2007; Ramos et al., 2016; Schrepfer and Scorrano, 2016). When the equilibrium is directed towards fusion, mitochondria are interconnected, net-like or aggregated in small regions of the cell. When the equilibrium is directed towards fission, mitochondria are fragmented, respiration-incompetent and tend to lose mitochondrial DNA. The data from Cataldo et al. (2010) suggest alterations in mitochondrial morphology, number and distribution in post mortem prefrontal cortex samples and primary fibroblasts and lymphocytes from BD individuals compared to controls. The mitochondria were also smaller in neurons differentiated from induced pluripotent stem cells (iPSC) from BD patients compared to healthy controls (Mertens et al., 2015). Considering that BD is characterized by low energy status, it is tempting to speculate that these studies are reporting altered mitochondrial dynamics. In both studies, treatment with lithium did not cause any change in mitochondria (Cataldo et al., 2010; Mertens et al., 2015). There is no treatment strategy targeting metabolism in BD, most studies aim at finding biomarkers or at better understanding the pathology. However, drugs commonly used for treatment of metabolic disease may have beneficial effects on BD metabolism; for example, quetiapine reduces lactate (Kim et al., 2007) and lithium increases oxidative phosphorylation (Maurer et al., 2009).

#### BD AND OXIDATIVE STRESS

One outcome of oxidative phosphorylation decline is the increase in the generation of superoxide as a result of electron leak from the electron transport chain, which may lead to oxidative stress. A cell is in an oxidative stress state when an imbalance between the production of reactive species (RS) and antioxidant activities occurs (Halliwell, 2007). Increasing evidence suggests the involvement of oxidative stress in the pathology and progression of BD (Scaini et al., 2016; Data-Franco et al., 2017).

Two meta-analysis studies have shown that lipid peroxidation and nitric oxide level were significantly increased in red blood cells or serum from BD patients compared to healthy controls (Andreazza et al., 2008a; Brown et al., 2014). Oxidative damage of nucleic acids was also repeatedly observed and was found to be increased in peripheral and post-mortem patient brain samples (Andreazza et al., 2008a; Che et al., 2010; Soeiro-de-Souza et al., 2013; Brown et al., 2014). However, numerous studies have reported contradictory data on the antioxidant enzymatic activities (e.g., superoxide dismutase, catalase, glutathione peroxidase) in BD patients (Brown et al., 2014). In addition, using <sup>1</sup>H-MRS, no change was observed in the levels of glutathione, a major antioxidant in the brain, in the anterior cingulate cortex of BD patients and healthy controls (Chitty et al., 2013; Lagopoulos et al., 2013). On the other hand, a biochemical study of blood samples from patients with different ages of disease onset showed that glutathione levels are lower in BD patients and that a negative correlation was observed with the age at onset (Rosa et al., 2014). The discrepancies reported could be related to the heterogeneity of the studies in terms of type of tissue analyzed, age at onset, illness duration, phase of the disorder, number of manic/depression episodes and treatment. The cellular effects of oxidative stress are cumulative and it is predictable to worsen with time and number of manic episodes. Hatch et al. (2015) observed that protein carbonyl and lipid hydroperoxide content is higher in adults compared to adolescents with BD. Another study showed that, indeed, antioxidant defenses might oscillate according to the phase of the disorder; superoxide dismutase activity was higher in manic and depressed patients compared to euthymic patients and controls (Andreazza et al., 2007). Notably, numerous reports have shown the antioxidant properties of mood stabilizers (Cui et al., 2007; Andreazza et al., 2008b; Bakare et al., 2009; Jornada et al., 2011; Banerjee et al., 2012; de Sousa et al., 2014).

The interest of researchers on the effects of oxidative stress on the pathophysiology of BD is recent; therefore the available data is limited. Over the years a number of trials with antioxidants have failed to provide the expected benefits for patients with various diseases (Casetta et al., 2005; Steinhubl, 2008; Halliwell, 2009; Persson et al., 2014). One of the reasons is because oxidative stress is frequently a secondary phenotype of mitochondrial dysfunction, as it is likely the case in BD patients. Further research is needed to evaluate the therapeutic potential of antioxidants and it's efficacy when given as adjunctive treatments.

### LINK BETWEEN MITOCHONDRIAL DYSFUNCTION, ENDOPLASMIC RETICULUM STRESS AND AUTOPHAGY

The principal functions of the ER are protein synthesis, folding and post-translational modifications, but it also interacts functionally with mitochondria to control calcium signaling and apoptosis (Pizzo and Pozzan, 2007; Halperin et al., 2014; Senft and Ronai, 2015; Raffaello et al., 2016). Accumulation of unfolded proteins, which may be triggered by defaults in protein folding or post-translational modifications, calcium changes and by redox imbalance, causes ER stress. The cellular response to ER stress involves other organelles, such as the mitochondria, which leads to restoring cell homeostasis or to committing cells to death. The pathways activated by ER stress are the unfolded protein response (UPR), ER-associated degradation, autophagy, hypoxic signaling and mitochondrial biogenesis (Raffaello et al., 2016). The UPR is mediated by three stress sensors—the inositolrequiring enzyme 1 (IRE1), the activating transcription factor 6 (ATF6) and protein kinase RNA-like ER kinase (PERK)—that activate a complex transcriptional cascade that leads to multiple adaptive responses or cell death (Hetz, 2012; Senft and Ronai, 2015; **Figure 3**).

Several studies using BD patient-derived lymphoblastoid cell lines or blood cells showed an impaired response to ER stress (So et al., 2007; Hayashi et al., 2009; Pfaffenseller et al., 2014). So et al. (2007) were the first to report decreased induction of expression of the X-box binding protein 1 (XBP1) and C/EBP homologous protein (CHOP) genes in response to ER stress stimulated by thapsigargin and tunicamycin in B-lymphocytes from BD patients compared to controls. In agreement with these observations, another study showed an increase in the amount of several proteins implicated in the UPR (phosphorylated eukaryotic initiation factor 2 (eIF2α-P), chaperone GRP78, XBP1 and CHOP) in leucocytes treated with tunicamycin from controls but not in those from BD patients (Pfaffenseller et al., 2014). Hayashi et al. (2009) also reported an attenuation in the expression of XBP1 and GRP94 in BD patient-derived lymphoblastoid cell lines treated with thapsigargin compared to the control cell lines. A single nucleotide polymorphism (SNP; −116C/G; rs2269577) in the promoter of the XBP1 gene impairs the feedback loop regulation

FIGURE 3 | Cellular response to ER stress. Accumulation of unfolded proteins in the ER lumen signal the unfolded protein response (UPR). The activated stress sensor proteins—protein kinase RNA-like ER kinase (PERK), inositol-requiring enzyme 1 (IRE1) and activating transcription factor 6 (ATF6) –signal different transduction pathways aiming at restoring cell homeostasis or committing the cell to death. Abbreviations: ERAD, endoplasmic reticulum-associated protein degradation; ER, endoplasmic reticulum.

in the ER response and is associated with BD (Kakiuchi et al., 2003; Cheng et al., 2014). However, the attenuated XBP1 induction in patient B-lymphocytes was independent of the promoter polymorphism (So et al., 2007). Differential gene analysis of data obtained by RNA sequencing from blood cells from healthy controls, lithium-responsive patients and lithiumnon-responsive patients identified the response to ER stress as a lithium-regulated gene network (Breen et al., 2016). Overall, these data suggest that the adaptive response of BD cells to ER stress is compromised, which may decrease the resilience of cells to stress conditions (**Figure 3**). Interestingly, the most frequently used mood stabilizers, lithium and valproate, seem to also have beneficial effects and increase the adaptive response to ER stress.

Autophagy is a cellular response aiming at restoring homeostasis or committing cells to death under nutrient starvation conditions (Klionsky and Emr, 2000; Baehrecke, 2005). During autophagy, protein aggregates, cytoplasmic components and organelles are degraded and the released molecules are recycled in biosynthetic pathways. The molecular mechanisms that regulate autophagy are the activation of ATG (autophagy-related) genes by phosphatidylinoditol 3-kinase (PI3K) pathway and the repression of the mTOR (mammalian target of rapamycin) kinase (Klionsky and Emr, 2000). A complex interaction exists between autophagy, ER stress and mitochondria. For instance, the UPR mediators activate autophagy (**Figure 3**; Senft and Ronai, 2015; Raffaello et al., 2016) and affect mitochondrial function by regulating Parkin (Bouman et al., 2011). On the other hand, Parkin is a regulator of mitochondrial dynamics and is necessary to target damaged mitochondria to mitophagy (Narenda et al., 2008; Poole et al., 2008). Recently, it was shown that autophagy is down regulated in schizophrenia and MDD (Jia and Le, 2015; Merenlender-Wagner et al., 2015). No data is available hitherto for patient cells or animal models of BD. However, since there is evidence that mitochondrial and ER functions are disturbed in BD, it is possible that autophagy is also altered and contributes to the pathophysiology of the disorder. Toker and Agam (2015) suggested an original hypothesis; they proposed that in psychiatric disorders mitochondrial dysfunction results from autophagy impairment.

## GLUTAMATERGIC NEUROTRANSMISSION AND HYPEREXCITABILITY

Glutamate and γ-aminobutyric acid (GABA) are the major excitatory and inhibitory neurotransmitters in the brain, respectively. The most abundant neurons in the cortex are the excitatory pyramidal cells, the inhibitory interneurons account for only ≈20% of the total neurons (Markram et al., 2004; Molyneaux et al., 2007). A chronic accumulation of glutamate in the synaptic cleft causes excitotoxicity and neuronal death due to excessive stimulation of the postsynaptic glutamate receptors (Wang and Qin, 2010). A metabolic glutamate-glutamine cycle between neurons and astrocytes maintain glutamate levels below toxicity. Neuronal impulses trigger the release of glutamate into the synaptic cleft, generating postsynaptic currents. Glutamate is then taken up by astrocytes and converted into the non-toxic glutamine, which is transported back by the neurons and converted to glutamate. It is thus predictable that changes in the astrocyte ability to transport glutamate and synthesize glutamine will affect neurotransmission and neuronal survival in the brain.

The question of brain glutamate levels in BD patients has been addressed in numerous studies using MRS. The magnetic field strengths and signal-to-noise ratio in most studies using human individuals do not allow a fine resolution of the peak into glutamate and glutamine, so it is a composite peak of two metabolites that is quantified (generally named Glx; Stork and Renshaw, 2005). Since glutamate is in large supply compared to glutamine, it is assumed that the changes observed in the Glx signal are correlated with glutamate (Stork and Renshaw, 2005; Gigante et al., 2012). Adult BD patients show a consistent increase in glutamate levels in the frontal brain areas compared to healthy controls; these increases are independent of the mood phase (Castillo et al., 2000; Michael et al., 2003; Dager et al., 2004; Yildiz-Yesiloglu and Ankerst, 2006; Hashimoto et al., 2007; Moore et al., 2007; Eastwood and Harrison, 2010; Gigante et al., 2012; Kondo et al., 2014; Ehrlich et al., 2015). Treatment of patients with the multi-target mood stabilizers lithium and valproate restores the Glx levels to normal (Friedman et al., 2004; Strawn et al., 2012). Findings in post-mortem brain tissue from BD patients confirmed changes in glutamatergic neurotransmission. Hashimoto et al. (2007) reported an increase in the levels of glutamate in BD patient samples from frontal cortices. Proteomics and transcription studies showed alterations in N-methyl-D-aspartate (NMDA) receptors and other intermediates of glutamatergic signaling (Hashimoto et al., 2007; Rao et al., 2009; Eastwood and Harrison, 2010; Gottschalk et al., 2015). Accordingly, the SLC1A2 gene that encodes the astrocytic excitatory amino acid transporter 2 (EAAT2, responsible for majority of glutamate re-uptake in the brain) is a susceptibility locus to BD (Fiorentino et al., 2014).

A relationship was found between glucose metabolism and glutamatergic neuronal function in vivo in the rat cortex by measuring the rates of TCA cycle and glutamate synthesis using <sup>14</sup>C-MRS (Sibson et al., 1998). A stoichiometry close to 1:1 was calculated between glucose metabolism and glutamate cycling, suggesting that the majority of the glucose consumed and energy produced in the cortex supports the glutamatergic synaptic activity (Sibson et al., 1998). In inhibitory neurons, GABAergic transmission also imposes high energy expense (Patel et al., 2005). Therefore, the increased levels of excitatory neurotransmitter glutamate in the brain of BD patients imply a higher energy demand on the neurons. Dager et al. (2004) suggested that the increased rate of glycolysis observed in MRS studies is the metabolic response of BD cells to the increased energy requirements and to the deficient oxidative metabolism. Consistent with increased glutamate levels and the pressure on energy metabolism, Rao et al. (2009) found excitotoxicity and neuroinflammation in post-mortem frontal cortex from BD patients.

Studies using transcranial magnetic stimulation paradigms showed a significant deficit in cortical inhibition in BD patients compared to healthy controls (Levinson et al., 2007; Chroni et al., 2008), which is in agreement with the data showing increased glutamatergic neurotransmission in BD patients. The hippocampus, another brain region affected in BD, also is the site of adult neurogenesis (Vadodaria and Gage, 2014). New excitatory granule cells are continuously generated in the dentate gyrus. After maturation and integration in the neural circuit, the new neurons are indistinguishable from those generated during embryonic development, but it is their hyperexcitable nature during maturation that gives the hippocampus its plasticity and particular cognitive functions (Kempermann et al., 2015). Studies using several mouse models of psychiatric disorders, such as the Ca2+/calmodulin-dependent protein kinase II (α-CaMKII) heterozygous knockout, showed that the dentate gyrus granule cells were arrested at a stage with similar molecular and physiological properties to those of the immature neurons (Yamasaki et al., 2008; Hagihara et al., 2013). This phenotype was named ''immature dentate gyrus'' and was proposed to be an endophenotype of BD and other psychiatric disorders (Hagihara et al., 2013).

# MODELS OF BD AND CLINICAL TRANSLATION TO DRUG TARGET

The development of novel treatments for psychiatric disorders has been hindered by the slow progress in our understanding of the underlying neurobiology, which results from the difficulty of developing faithful animal and cellular models. The complexity of psychiatric disorders and the still unknown relationships between the diagnosis and the etiology, neurobiology, genetics or response to medication led to the endophenotype concept. Endophenotypes are simple measurable heritable components that can be neurophysiological, biochemical, endocrine, neuroanatomical, cognitive or neuropsychological (Gould and Gottesman, 2006). Endophenotypes are useful in the construction of animal models and help dissect genetics and biological mechanisms of specific features of the disorders. Using this strategy, numerous rodent models of BD depression and mania have been constructed using approaches as diverse as genetic, pharmacological, nutritional and environmental (Nestler and Hyman, 2010; Kato et al., 2016; Logan and McClung, 2016). The use of animal models of human disease in research and drug testing should meet three criteria: construct validity, face validity and predictive validity (Nestler and Hyman, 2010). These criteria are useful in the evaluation of how similar is the animal model to the human disease in terms of shared genetics and mechanisms (construct validity), symptoms (face validity) and efficiency of medications on the animal phenotypes (predictive validity). Mice with a loss of function mutation in the CLOCK gene (Clock∆19 mutant mice) exhibit mania symptoms similar to those observed in patients, including hyperactivity, decreased sleep, lowered depression-like behavior, lower anxiety and an increase in the reward value for cocaine, sucrose, and medial forebrain bundle stimulation (Roybal et al., 2007). Interestingly, these symptoms were also reversed by chronic lithium administration. However, there is no evidence for circadian gene mutations in the majority of BD patients. To date, none of the BD models have fulfilled the requirements needed for their use in drug development, but they contributed to the understanding of the pathophysiology of the disorder (Nestler and Hyman, 2010; Kato et al., 2016; Logan and McClung, 2016).

Transgenic mice with overexpression of glycogen synthase kinase-3β (GSK-3β) show hyperactivity as observed in the manic phase of BD (Prickaerts et al., 2006). Lithium inhibits GSK-3β and this effect has been suggested as one possible mechanism of action in BD patients (Stambolic et al., 1996; Li et al., 2010). Lithium and valproate also act on the GSK-3β signaling pathway to reverse the manic-like behavior in an animal model of mania induced by ouabain (Valvassori et al., 2016). Ouabain inhibits Na+/K+-ATPase activity, which induces hyperactivity. Both lithium and GSK-3β knockdown act on circadian rhythm by producing a lengthening of mPer2 period in mouse fibroblasts (Kaladchibachi et al., 2007). In addition, GSK-3β also phosphorylates PER2 (Iitaka et al., 2005) and REV-ERBα (Yin et al., 2006) regulating localization and stability of these proteins.

Lithium has numerous molecular targets, such as inositol monophosphatase (Agam et al., 2002; Harwood, 2005), protein kinase C pathway (Newberg et al., 2008) and calcium channels (Andreazza and Young, 2014) just to name a few. It is still an open question as to which one is responsible for the anti-manic effect in humans (Shaldubina et al., 2001; Beaulieu et al., 2008); nevertheless, these molecular pathways are used as targets to develop novel drugs or repurpose old ones. Long-term treatment of rats with lithium carbonate decreased membrane associated PKC in hippocampal structures (Manji et al., 1993) and treatment with valproate sodium increased cytosol/membrane ratio of PKC activity (Chen et al., 1994). Tamoxifen, a centrally acting PKC inhibitor has been shown to demonstrate anti-manic properties (Yildiz et al., 2008) in a randomized control trials of humans. More recently, the antioxidant ebselen, which inhibits inositol monophosphatase and induces lithium-like effects on mouse behavior, was suggested as a safe alternative to lithium (Singh et al., 2013). In anterior cingulate region of the brain, ebselen was shown to reduce glutamate and inositol levels possibly by inhibiting glutaminase (Masaki et al., 2016).

The advent of cellular reprogramming technology has allowed for the generation of iPSCs from somatic tissues (e.g., skin and blood) from patients with neuropsychiatric disorders (Takahashi et al., 2007; Brennand et al., 2011; Mertens et al., 2015). Significant methodological advances in human iPSC differentiation protocols has enabled iPSC-disease modeling using specific neuronal subtypes (Maroof et al., 2013; Nicholas et al., 2013; Zhang et al., 2013; Yu et al., 2014; Vadodaria et al., 2016a,b).

Chen et al. (2014) reported differences in transcriptional profiles in iPSC-derived neurons from controls and BD patients. The expression of transcripts for membrane-bound receptors and ion channels was significantly increased in BD neurons (Chen et al., 2014). Control neurons expressed transcripts that confer dorsal telencephalic fate, whereas BD neurons expressed transcripts that are involved in the differentiation of ventral regions (e.g., medial ganglionic eminence). iPSC technology allows for the interrogation of cellular phenotypes that can be detected during neuronal development and are not directly evident in post-mortem studies. Specifically for BD, neurodevelopmental deficits have been suggested (O'Shea and McInnis, 2016), but the lack of access to embryonic tissues has hindered the confirmation of a neurodevelopmental hypothesis. Evidence for altered neuronal development in BD has also been suggested using iPSC-derived neural progenitor cells (NPCs) from BD patients (Madison et al., 2015). In this study, the authors derived and characterized a set of 12 iPSC lines from a quartet of two BD-affected brothers and their two unaffected parents and found significant differences in neurogenesis and in expression of genes involved in WNT signaling and ion channel subunits (Madison et al., 2015). Subsequent treatment of the NPCs with a pharmacological inhibitor of GSK-3β (CHIR99021), a known regulator of WNT signaling, rescued the progenitor proliferation deficit in the BD patient NPCs. The role of micro RNAs (miRNAs) was also investigated in BD neuronal tissue and cultures. Bavamian et al. (2015) showed increased levels of miR-34a in post-mortem cerebellar tissue from BD patients, as well as in BD patient iPSC-derived neuronal cultures. miR-34a is predicted to target multiple genes implicated as genetic risk factors for BD, and the authors have validated a number of predicted mir-34a direct targets in the BD cultured neurons (ankyrin-3, ANK3 and voltage-dependent L-type calcium channel subunit beta-3, CACNB3; Hunsberger et al., 2013; Bavamian et al., 2015). In addition, overexpression of miR-34a was shown to result in abnormalities in neuronal differentiation and morphology as well as in the expression of synaptic proteins in control cells (Bavamian et al., 2015). The authors propose that miR-34a regulates a molecular network essential for neuronal development and synaptogenesis that is implicated in BD neuropathology.

A study examined hippocampal DG granule cell neurons (Prox1 positive) differentiated from six BD patients and four healthy controls (Mertens et al., 2015). Gene expression studies were performed and suggested mitochondrial abnormalities in DG granule cell neurons from BD patients. Interestingly, electrophysiological functional studies revealed hyperexcitability in BD neurons that was selectively decreased after lithium treatment in neurons from lithium-responsive patients, and not in neurons from the non-responders (Mertens et al., 2015). This work suggests that clinical information and drug response patterns can be used to test the validity of cellular phenotypes in culture. That realization is very powerful since it could open new avenues to find new drugs and therapies that ameliorate phenotypes in cultured neurons and could potentially be translated into patient treatments.

# CONCLUSION AND FUTURE CHALLENGES

The conventional drug development approach for psychotropics has been through the manipulation of receptor profiles of existing drugs or purely by an empirical approach. Neuroscience-based treatment development for psychiatric disorders has stagnated over the last four decades, with molecular and neuroscience research findings often not mapping onto clinical phenomenological approaches. One of the reasons for the slow progress is the lack of BD accurate animal and cellular models for drug testing and pathophysiology investigation. The expansion of studies using iPSC-derived technologies hopefully will allow for a better understanding of the affected molecular pathways and provide an initial platform for drug development. Despite the fact that it seems impossible to have animal models of psychiatric illness that fully reproduce the complex neurologic symptoms and associated comorbidities, animal testing for new drugs before clinical trials is an obligatory step. The construction of valid animal disease models is thus a foremost challenge.

The integration of other symptoms observed in BD patients, besides the neuropsychiatric, and medical comorbidities led to the exploration of essential cellular functions, not specific to neurons but shared by multiple cell types. The early observation of altered brain energetic metabolism encouraged the search for mitochondrial dysfunction. Mitochondria are central organelles in a cell and even minor dysfunction can lead to a cascade of changes and damage. To ensure survival, the cells adapt to chronic mitochondrial dysfunction coordinating responses with other organelles, such as the ER. However, survival of BD cells has a cost on physiology and ultimately causes perturbations in different tissues and organs. In addition, BD is a neurodevelopmental disorder and mitochondrial metabolism modifications are essential during neurogenesis (Zheng et al., 2016). Following this line of thinking, drugs that target these pathways are potentially interesting for BD treatment, as primary or adjuvant medicine. For example the use of minocycline, which is an antibiotic that can modulate glutamate-induced excitotoxicity and has antioxidant and anti-inflammatory properties, showed promising results in clinical trials (Dean et al., 2012). We believe that a better understanding of the molecular mechanisms that result in mitochondrial impairment and oxidative stress together with the regulation of adaptive UPR and autophagy responses will provide the key pieces of information that will unlock novel drug treatments for BD beyond mood stabilizers and antipsychotics.

# AUTHOR CONTRIBUTIONS

YK, RS and MCM designed and wrote the article. FHG contributed with discussions and revision of the text.

# ACKNOWLEDGMENTS

The authors would like to acknowledge financial support from Janssen Pharmaceuticals. This work was also supported by the Paul G. Allen Family Foundation, Bob and Mary Jane Engman, The Leona M. and Harry B. Helmsley Charitable Trust grant #2012-PG-MED002, Annette C. Merle-Smith, National Institutes of Health (NIH; grant no. R01 MH095741 (FHG), U19MH106434 (FHG)) and by The G. Harold and Leila Y. Mathers Foundation. The authors would also like to thank M. L. Gage for editorial comments.

#### REFERENCES


with bipolar disorder. Psychiatry Res. 213, 230–234. doi: 10.1016/j.pscychresns. 2013.03.004


in bipolar disorder. Biol. Psychiatry 56, 340–348. doi: 10.1016/j.biopsych.2004. 06.012


with bipolar disorder. J. Clin. Psychiatry 64, 161–174. doi: 10.4088/jcp. v64n0209


and energy metabolism in the rat cortex in vivo. Proc. Natl. Acad. Sci. U S A 102, 5588–5593. doi: 10.1073/pnas.0501703102


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Kim, Santos, Gage and Marchetto. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# New Insights into the Crosstalk between NMDARs and Iron: Implications for Understanding Pathology of Neurological Diseases

Huamin Xu1,2\*, Hong Jiang1,2 and Junxia Xie1,2 \*

<sup>1</sup>Collaborative Innovation Center for Brain Science, Department of Physiology, Shandong Provincial Key Laboratory of Pathogenesis and Prevention of Neurological Disorders and State Key Disciplines: Physiology, Medical College of Qingdao University, Qingdao, China, <sup>2</sup>Shandong Provincial Collaborative Innovation Center for Neurodegenerative Disorders, Qingdao University, Qingdao, China

Both iron dyshomeostasis and N-methyl-D-aspartate receptors (NMDARs)-mediated neurotoxicity have been shown to have an important role in neurological diseases such as Parkinson's disease (PD) and Alzheimer's disease (AD). Evidence proved that activation of NMDARs could promote iron overload and iron-induced neurotoxicity by enhancing iron importer divalent metal transporter 1 (DMT1)-mediated iron uptake and iron releasing from lysosome. Also, iron overload could regulate NMDARs-mediated synaptic transmission. This indicates that there might be a possible relationship between iron and activation of NMDARs in neurological diseases. Understanding this interaction between iron and activation of NMDARs may provide new therapeutic avenues for a more targeted neurotherapeutic strategy for these diseases. Therefore, in this review article, we will describe the dysfunction of iron metabolism and NMDARs in neurological diseases including PD and AD, and summarize the new insight into the mechanisms underlying the interaction between iron and activation of NMDARs.

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Christian Gonzalez-Billault, University of Chile, Chile Daniela Tropea, Trinity College, Dublin, Ireland Fabrizio Gardoni, University of Milan, Italy

#### \*Correspondence:

Huamin Xu huamin102@163.com Junxia Xie jxiaxie@public.qd.sd.cn

Received: 13 September 2016 Accepted: 01 March 2017 Published: 16 March 2017

#### Citation:

Xu H, Jiang H and Xie J (2017) New Insights into the Crosstalk between NMDARs and Iron: Implications for Understanding Pathology of Neurological Diseases. Front. Mol. Neurosci. 10:71. doi: 10.3389/fnmol.2017.00071 Keywords: Parkinson's disease, Alzheimer's disease, iron, NMDA receptor, divalent metal transporter 1

#### INTRODUCTION

Iron is an important cofactor in many proteins such as heme-containing proteins and iron-containing enzymes. It is required for many physiological processes for life (Zucca et al., 2015). In the central nervous system (CNS), iron also participates in myelin synthesis, development of dendritic spines in the hippocampus and synthesis of neurotransmitters including monoamine transmitters and gamma-aminobutyric acid (GABA; Li, 1998; Jorgenson et al., 2003; Todorich et al., 2009; Zucca et al., 2015; Bastian et al., 2016). However, excess iron is toxic due to its ability to produce cytotoxic hydroxyl radicals, which could cause damages to proteins, nucleic acids and cell membranes (Whitnall and Richardson, 2006). Therefore, intracellular iron metabolism is tightly regulated. There are two pathways for iron uptake: the classical transferrin (Tf)-mediated iron uptake pathway and the non-Tf bound iron (NTBI) uptake pathway (Qian and Shen, 2001). Traditionally, the Tf-transferrin receptor 1 (Tf-TfR1) pathway is considered as major pathway for cellular iron uptake in the brain. Divalent metal transporter 1 (DMT1) is identified as the main NTBI pathway, responsible for ferrous iron uptake. There are four DMT1 mRNA isoforms: N-terminus (1A, 1B) generated from alternative promoters with subsequent, exclusive splicing of the respective first exon to exon 2 and C-terminus splice variants (+iron-responsive element (IRE), −IRE) due to alternative splice mechanisms (Hubert and Hentze, 2002). DMT1 + IRE contains an IRE in the 3<sup>0</sup> -untranslated region (UTR) and DMT1 − IRE has no IRE. Additionally, 1A/DMT1 is predominantly expressed in kidney and duodenum and 1B/DMT1 is ubiquitously expressed in the peripheral organs and brain (Hubert and Hentze, 2002). Ferroportin 1 (FPN1), also known as metal transport protein1 (Abboud and Haile, 2000) or iron-regulated transporter 1 (IREG1) is the only known iron transporter responsible for cellular iron export (Ganz, 2005). Disturbance in iron metabolism plays a key role in the pathogenesis of neurodegenerative disorders such as Parkinson's disease (PD) and Alzheimer's disease (AD) (LaVaute et al., 2001; Lieu et al., 2001; Perez et al., 2008; Zhu et al., 2009). Elevated iron levels were found in the substantia nigra (SN) in PD patients (Sofic et al., 1988). Studies also showed that iron accumulation was observed in the hippocampus and cerebral cortex in AD patients (Piñero et al., 2000, 2001; Altamura and Muckenthaler, 2009). Increased import, decreased export or redistribution of intracellular iron might be responsible for iron metabolism disturbance in these diseases, which may increase the vulnerability of neurons to iron.

Glutamate is the major excitatory neurotransmitter in the CNS, exerting its functions by binding to different receptors. Among them, N-methyl-D-aspartate receptors (NMDARs) are cation channels that mediate entry of Na<sup>+</sup> and Ca2<sup>+</sup> ions and are activated by the co-agonists glutamate (or NMDA) and glycine or D-serine. There are several subunits of NMDARs: NR1, NR2A, NR2B, NR2C and NR2D and NR3. Among them, NR1 is a fundamental subunit of the receptor. Appropriate activation of NMDARs plays a critical role in physiological functions such as excitatory neurotransmission, synaptic plasticity (Carvajal et al., 2016). However, excitotoxicity induced by excessive activation of NMDARs contributes to pathological changes in the CNS (Ambrosi et al., 2014; Gonzalez et al., 2015). Influx of Ca2<sup>+</sup> through NMDARs might lead to neuronal loss in several neurodegenerative diseases such as PD and AD (Hynd et al., 2004). NMDARs have both synaptic and extrasynaptic locations. One hypothesis posits that synaptic NMDARs were neuroprotective and extrasynaptic NMDARs were neurotoxic (Hardingham and Bading, 2010). The death signaling induced by extrasynaptic NMDARs or relocalization of NMDARs to extrasynaptic sites has been shown to contribute to pathology of neurodegenerative diseases (Bading, 2017). Fine-tuning might provide a promising operation to optimize the activity of the glutamatergic system in order to maintain normal function of neurons (Köles et al., 2016). Activation of NMDARs as well as iron deposits in neurological diseases including PD and AD suggest that there might be a correlation between iron deposition and activation of NMDARs. Therefore, in this article, we reviewed the studies about the involvement of iron dyshomeostasis and NMDARs-mediated neurotoxicity in PD and AD. We then described the advanced knowledge on interaction of iron and NMDARs activation. This will provide implications for understanding pathology of these neurological diseases.

# DYSFUNCTION OF IRON METABOLISM AND NMDARs IN PD

PD is a common neurodegenerative disorder characterized by loss of dopamine (DA) neurons in the SN pars compacta (SNpc), resulting in depletion of DA in the striatum (Parkinson, 2002; Hornykiewicz, 2008). Although the etiology of PD has not been clarified until now, iron accumulation and excitatory neurotoxicity are considered as contributing factors to the etiology of PD. It has become increasingly evident that elevated iron levels in the SNpc play a key role in the degeneration of DA neurons in PD (Jiang et al., 2016; Muñoz et al., 2016). Nigral iron levels increased with age, and PD patients showed an even greater increase, which correlates with clinical PD status (Wu et al., 2014; Du G. et al., 2016). Experimental evidence also showed that iron was involved in the degeneration of SNpc dopaminergic neurons in 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) and 6-hydroxydopamine (6-OHDA)-induced animal models of PD (Salazar et al., 2008; Jiang et al., 2010). Mechanisms underlying iron-induced neurodegeneration of DA neurons have been reported: First, iron participates in Fenton reaction to generate ·OH, which could damage proteins, nucleic acids and cell membranes (Mohanakumar et al., 1994). Second, iron-induced α-synuclein oligomers can form ion-permeable pores in lipid bilayers and give rise to α-synuclein-dependent toxicity in neuronal cells (Kostka et al., 2008). Our previous studies also showed iron could promote α-synuclein aggregation (He et al., 2013). Third, iron overload induced mitochondrial fragmentation via increasing intracellular calcium (Ca2+) and activated calcineurin via Ca2+/calmodulin and Ca2+/calpain pathways. This mitochondrial fragmentation and neuronal cell death could be rescued by chelation of intracellular Ca2<sup>+</sup> (Lee et al., 2016). Recently, it was reported that ferroptosis, an iron-dependent regulated cell death process, is an important cell death pathway for DA neurons in PD (Do Van et al., 2016).

A recent research showed that increase in DMT1 expression, rather than TfR1 and Fpn1 expression might be partly responsible for age-dependent increase in brain iron (Lu et al., 2016). This indicated the importance of DMT1 in age-induced iron overload. Our previous studies showed that increased expression of DMT1 + IRE and decreased expression of FPN1 were responsible for nigral-specific iron accumulation in PD models (Wang et al., 2007; Jiang et al., 2010; Song et al., 2010). DMT1 was found in the SNpc and associated predominantly with neuromelanin-containing DA neurons. It was consistently less intense in DA neurons of the ventral tegmental area (VTA) than in those of the SNpc in the same tissue sections of human brain older than 60 years (Salazar et al., 2008). In addition, postmortem studies have shown an increase of DMT1 in the SN of PD patients (Salazar et al., 2008). And a mutation in DMT1 that impaired iron transport protected rodents against parkinsonism-inducing neurotoxins MPTP and 6-OHDA (Salazar et al., 2008). Recent study demonstrated that DMT1 polymorphisms might be a risk factor for PD (Saadat et al., 2015). This indicated that higher DMT1 expression and consequently higher iron levels in nigral DA neurons might contribute to increased vulnerability of nigral DA neurons to PD-related neurotoxins (Huang E. et al., 2004). The mechanisms underlying the expression of iron transporters DMT1 + IRE and FPN1 are IRE/iron regulatory proteins (IRPs) dependent in 6-OHDA-induced PD models and IRE/IRPsindependent in MPP+-treated dopaminegic cells (Zhang et al., 2009; Jiang et al., 2010). In addition, ceruloplasmin (CP) is the strongest ferroxidase to stabilize iron exporter FPN1. It was reported that the activity of CP in PD brains was reduced in SN (Ayton et al., 2013). In CP-knockout mice, iron overload was found in several tissues, including the brain (Kaneko et al., 2008). In our previous report, we also showed that decreased expression of CP in the SN was involved in the nigral iron accumulation of 6-OHDA-induced PD rats (Wang et al., 2015). These results suggested that CP might also play an important role in iron deposit in PD.

In addition, NMDARs-mediated excitotoxicity also contributed to the progressive degeneration of nigral DA neurons in PD (Ambrosi et al., 2014). Research has showed that ligand-gated ion channel NMDARs were vital in the glutamate-induced excitotoxicity in primary dopaminergic cell culture (Oster et al., 2014). Furthermore, glutamate exposure induced Parkin accumulation at mitochondria in a calciumand NMDARs-dependent manner (Van Laar et al., 2015). It has been confirmed that SNpc DA neurons express functional triheteromeric NMDARs composed of NR1, NR2B and NR2D subunits (Jones and Gibb, 2005). A selective decrease of NR1 pan mRNA levels in layer IV of frontal cortex was found in PD patients (Meoni et al., 1999). However, NR1 expression increased in surviving SN DA neurons from PD brains compared with neurons from controls (Schiemann et al., 2012). Meanwhile, intracellular Ca2<sup>+</sup> overload triggered by activation of NMDARs is believed to be responsible for inducing of nitric oxide synthases (NOS) activity, affecting mitochondrial integrity and functions (Yamauchi et al., 1998; Sattler and Tymianski, 2000). It has been reported that NMDARs antagonists were considered as potential therapies for patients with PD (Little and Brown, 2014). Among the NMDARs antagonists, memantine has been reported to preferentially blocks extrasynaptic NMDARs of SNpc DA neurons in slices of rat midbrain using whole-cell patch-clamp recordings (Wu and Johnson, 2015). This selective effect on NMDARs might be protective due to the neurotoxic effect of extrasynaptic NMDARs.

#### DYSFUNCTION OF IRON METABOLISM AND NMDARs IN AD

AD is a progressive neurodegenerative disorder, clinically characterized by a progressive loss of cognitive abilities and dementia, which is closely related to a degree of neuronal and synaptic loss (Selkoe, 1996; Hardy, 1997; Mattson, 1997). The key features of the disease are the accumulation of extracellular amyloid-β (Aβ) plaques and neurofibrillary tangles inside neurons. Although genetic and non-genetic factors are involved in the etiology of AD, the exact reason is still unknown. Increasing evidence suggests that iron might play an important role in the development or progression of AD (Mandel et al., 2007; Ward et al., 2014; Belaidi and Bush, 2016; Van Bergen et al., 2016). The first evidence is that iron accumulates in the same brain regions which are characterized by Aβ deposition such as hippocampus, parietal cortex and motor cortex in AD patients (Dedman et al., 1992; Good et al., 1992). High levels of iron have also been reported in the amyloid plaques in PS/amyloid precursor protein (APP) and APP [V717I] transgenic mice (Falangola et al., 2005), resembling those seen in the brains of AD patients. In addition, higher cortical iron was associated with increased Aβ-plaque-load in mild cognitive impairment (MCI; Van Bergen et al., 2016). Another important link between iron and AD is based on the observation that APP expression is iron-regulated. APP is a single transmembrane metalloprotein that is cleaved to generate the 40–42-amino-acid Aβs by β- and γ-secretases. The existence of a functional IRE in the 5<sup>0</sup> - UTR of APP mRNA makes it possible to be controlled at the level of mRNA translation by the action of IRE/IRP's response to iron (Rogers et al., 2002). This implied that high intracellular iron levels could cause increased APP translation and Aβ formation via this mechanism. In addition, iron could induce Aβ precipitation (Mantyh et al., 1993; Huang X. et al., 2004) and significantly enhance the toxicity of Aβ in cultured neuronal cells, whereas iron chelators protect the neurons from Aβ toxicity (Schubert and Chevion, 1995). This provided a direct link between excessive iron and loss of neuronal function seen in AD patients.

However, the underlying mechanisms involved in disturbance of iron homeostasis in AD brain remain unclear. It is reported that Aβ is a metalloprotein that binds transition metal ions through three histidine residues (His6, His13 and His14) located in the N-terminal domain (Nakamura et al., 2007; Altamura and Muckenthaler, 2009). And other protein that accumulates in AD such as tau protein also possesses metal-binding sites (Perry et al., 2003). These might account for the iron accumulation in the affected brain regions in AD. Furthermore, it is reported that DMT1 was colocalized with Aβ in the plaques of postmortem AD brain and the levels of DMT1 was significantly increased in the cortex and hippocampus in APP/PS1 transgenic mouse model compared with wild type-control (Zheng et al., 2009). And the mean serum p97 concentration was elevated 3- to 4-fold in patients with AD as compared to non-AD dementia and normal controls (Kim et al., 2001). In addition, HO-1 is increased in neurofibrillary tangles, senile plaque neurites, granulovacuolar degeneration and neuropil threads in human AD brains (Smith et al., 1994; Perry et al., 2003), suggesting the redistribution of iron due to release from heme proteins in affected areas of the AD brain.

NMDARs–mediated excitotoxicity also contributed to the pathology of AD. Evidence proved that impaired glutamate uptake in astrocytes and neurons in AD can lead to increased concentrations of glutamate at the synapse, which can subsequently trigger NMDARs–mediated excitotoxicity via increase of intracellular Ca2<sup>+</sup> concentrations (Greenamyre et al., 1988; Tong et al., 2017). It has been shown that extrasynaptic NMDARs were largely associated with NMDARs-mediated excitotoxicity in AD (Hardingham and Bading, 2010). Prolonged activation of extrasynaptic NMDARs increases APP processing, leads to neuronal Aβ release, and ultimately results in AD Xu et al. Crosstalk between NMDARs and Iron

pathology (Bordji et al., 2010). It has been reported that an open channel blocker memantine could preferentially antagonize excessively activated NMDARs (Lipton, 2004). In vitro and in vivo studies have demonstrated the neuroprotective effect of memantine in AD models. Memantine could prevent oligomeric Aβ-induced oxidative stress in mature hippocampal neurons (De Felice et al., 2007). More interestingly, memantine decreased the levels of secreted APP and Aβ peptide in human neuroblastoma cells (Ray et al., 2010) and lowered cortical levels of Aβ1–42 in APP/PS1 transgenic mice (Alley et al., 2010). This indicated the important role of excessively activated NMDARs in AD.

#### INTERACTION OF IRON AND NMDARs ACTIVATION

#### NMDARs Activation Promoted Iron Accumulation

Results revealed that activation of NMDARs significantly promoted Fe2<sup>+</sup> entry into cells (Cheah et al., 2006). It was reported that NMDARs activation might promote iron accumulation by accelerating DMT1-mediated iron influx (Cheah et al., 2006), enhancing iron releasing from lysosome and regulation the expression of DMT1 (White et al., 2016). Elevated intracellular iron then aggravated iron-induced cell damage.

#### NMDARs Activation Increased NTBI Influx

It has been demonstrated that ferrous iron could block the influx of Ca2<sup>+</sup> across NMDARs channels in cultured neurons (Nakamichi et al., 2004). This indicated that Fe2<sup>+</sup> competed with Ca2<sup>+</sup> for NMDARs to enter primary neurons. This iron entry can be harmful for neurons during aging due to increased NTBI levels. An investigation with interest to the involvement of NMDARs in NTBI influx pathways showed that activation of NMDARs significantly promoted Fe2<sup>+</sup> entry into cells using fluorescence-based single cell analysis in rat hippocampal primary cultures (Pelizzoni et al., 2011). This elevation of iron was accompanied by a corresponding increase in reactive oxygen species (ROS) production and higher susceptibility of neurons to death.

Another investigation demonstrated a novel signaling cascade for glutamate in regulating iron uptake in the brain (Cheah et al., 2006). It is increasingly appreciated that glutamate via NMDARs triggers calcium influx, then activates neuronal NOS (nNOS) to produce NO, which causes excitability toxicity (Guix et al., 2005; Cheah et al., 2006; Chen et al., 2013; Courtney et al., 2014). Research has identified a signaling cascade whereby NMDARs regulated iron homeostasis (Cheah et al., 2006). Their results showed that the activation of NMDARs could increase intracellular iron levels in PC12 cells via NO-Dexras1-peripheral benzodiazepine receptor-associated protein 7 (PAP7)-DMT1 signaling cascade to enhance iron uptake (Cheah et al., 2006). Additionally, chelation of intracellular iron blocked formation of free radicals in brain cultures and also markedly attenuated NMDA neurotoxicity (Cheah et al., 2006). Furthermore, subsequent investigation showed that Dexras1 was required for NMDA-elicited neuronal toxicity via NO and iron influx (Chen et al., 2013). This implies that NMDA-NO activation-induced iron uptake might play an important role in neurotoxicity and misregulation of this pathway might participate in iron accumulation in PD. It has been shown that nNOS binding to CAPON leading to NO delivery to Dexras1 and S-nitrosylation of Dexras1. Then S-nitrosylation of Dexras1 could enhance iron uptake by regulating the function of iron importer DMT1. It was also reported that Dexras1 could be phosphorylated by protein kinase A (PKA) on serine 253. This PKA activation reduced Dexras1 S-nitrosylation, leading to an inhibition of iron influx (Chen et al., 2015). This indicated the important role of Dexras1 S-nitrosylation in iron uptake. In addition, PAP7 might presumably serve as a scaffold delivering Dexras1 into proximity to DMT1 (Cheah et al., 2006). However, another research showed that PAP7 played a role in cellular iron metabolism. They found that PAP7 was internalized in parallel with the internalization of DMT1 following iron feeding. And downregulated PAP7 expression in K562 cells with small interfering RNA led to downregulation of DMT1 (IRE) protein but not DMT1 (−IRE) mRNA. However, they did not measure iron uptake after downregulation of PAP7 expression. The levels of TfR1 and ferritin were not affected following transfection with siPAP7, indicating that the intracellular iron level might not change (Okazaki et al., 2012). Moreover, they also reported that overexpression of PAP7 had no effect on DMT1 (IRE) expression which is consist with the results of Cheah et al. (2006). Further research should be conducted to figure out the mechanisms underlying the effect of PAP7 on DMT1 and DMT1-induced iron uptake.

Rhes (Dexras2), a homolog of Dexras1, is a highly conserved small GTP binding protein belonging to the Ras superfamily. Rhes shares 67% identity with Dexras1. Although the physiological role of Rhes is not fully understood, results showed that Rhes physiologically interacted with PAP7 and participated in DMT1-induced iron uptake via pathway similar to Dexras1 (Choi et al., 2013). However, the mechanisms underlying the effect of Rhes on DMT1-induced iron uptake are different from Dexras1. They showed that Rhes was not S-nitrosylated after NO-treatment, however was phosphorylated by PKA at serine-239 (Choi et al., 2013). Rhes is selectively localized to the corpus striatum (Errico et al., 2008), so might involved in NMDARs-induced iron entry into the striatal neurons and maintain striatal iron homeostasis via PKA-Rhes-DMT1 pathway. In addition, striatum received nigral dopaminergic synapses projection and is responsible for movement balance. Imbalance of iron metabolism in striatum might contribute to the degeneration of striatal neurons and dysregulation of dopaminergic function in the striatum due to iron-induced oxidative stress, then affecting motor function.

#### NMDARs Activation Enhanced Iron Releasing from Lysosome

In fact, Dexras1 also could regulate TfR-mediated iron uptake as well as NTBI uptake (DMT1-mediated iron uptake) (Cheah et al., 2006). However, the exact mechanisms underlying this effect are not elucidated. Subcellular localization experiments showed that DMT1 + IRE had higher surface expression and was internalized from the plasma membrane with slower kinetics than DMT1 − IRE. And it was not efficiently recycled and was targeted to lysosomes. While DMT1 − IRE is efficiently sorted to recycling endosomes upon internalization (Lam-Yuk-Tseung and Gros, 2006). This implicated that different isoforms of DMT1 might have different functions on regulation of the subcellular localization of Fe2<sup>+</sup> transport. As lysosomal iron serves as a main source for intracellular iron and DMT1 plays a role in iron recycling from lysosome to cytoplasm, DMT1-mediated iron release from lysosome might be responsible for increased intracellular iron levels. Findings from patch-clamping of individual lysosomes showed that they could transmit iron through their membrane (Dong et al., 2008). Recently, an investigation confirmed that the Dexras1/ACBD3(PAP7)/DMT1 complex was located on the lysosomes (White et al., 2016). Collapsing the proton gradient or blocking DMT1 channel both could reduce cytosolic iron pool (White et al., 2016), indicating that Dexras1/ACBD3/DMT1 complex played roles in iron release from lysosome. Therefore, NMDA activation not only led to Dexras1/DMT1 mediated increase in iron uptake, but also enhanced Dexras1-dependent iron release from lysosome.

#### NMDARs Activation Increased DMT1 Expression

Activation of NMDARs not only could affect the iron uptake function of DMT1, but also could regulate its expression. It has been reported that mRNA expression of DMT1-1B and DMT1 + IRE increased after 5 min exposure of 50 µM NMDA to primary hippocampal cultures, but not mRNA expression of DMT1 − IRE (Haeger et al., 2010). NMDA also enhanced DMT1 protein expression, which was abolished by the transcription inhibitor actinomycin D and NMDARs antagonist MK-801 (Haeger et al., 2010). This stimulation on iron entry pathway via DMT1 could ensure an adequate iron supply. This is of potential importance because iron deficiency hinders learning processes and impairs cognitive performance (Carlson et al., 2009; Muñoz et al., 2011; Estrada et al., 2014). This implicated that NMDARs-activation stimulated expression of the iron transporter DMT1-1B + IRE, which presumably played a significant role in hippocampal spatial memory formation. Interestingly, both excessive activation of NMDARs and upregulation of DMT1 were contributing factors in degeneration of DA neurons in PD, indicating the link between activation of NMDARs and upregulation of DMT1 in PD. Further studied should be conducted to elucidate this possibility and the underlying mechanisms.

#### NMDARs-Induced Co-Activation of ATP-Sensitive Potassium (KATP) Channels Might Promote Iron Influx via DMT1

KATP channel was first discovered by Akinori Noma in cardiac myocytes and was demonstrated to be important for regulation of cellular energy metabolism in the control of membrane excitability (Noma, 1983). Liss et al. (2005) provided evidence that selective activation of KATP channels of DA midbrain neurons in the SN was a potential mechanism for the selective degeneration of DA neurons in PD. Increased activity of KATP channels due to metabolic stress leads to membrane potential hyperpolarization and reduced SN DA activity (Liss et al., 2005). Although in adult mice KATP channels in DA neurons of both SN and VTA are formed by the same subunits (four Kir6.2 subunits and four regulatory SUR1 subunits), the activation of nigral KATP channels of DA neurons was responsible for the selective degeneration of DA neurons in PD (Liss et al., 2005). Moreover, genetic inactivation of the pore-forming subunit Kir6.2 of KATP channels induced a selective rescue of SN DA (but not VTA DA) neurons in MPTP-induced PD models (Liss et al., 2005). This indicated that activation of nigral KATP channels contributed to the selective degeneration of DA neurons in the SN.

It is reported that the activation of KATP channels induced hyperpolarization of the membrane potential of DA neurons in the SN following MPP<sup>+</sup> treatment (Liss et al., 2005). Furthermore, it was reported DMT1-mediated iron transport was driven at higher rates of hyperpolarized potentials (Gunshin et al., 1997). This makes the possibility that the hyperpolarization of cell membrane induced by activation of KATP channels might increase DMT1-mediated iron transport, and enhance iron influx into the DA neurons. Results in our experiments demonstrated that activation of the KATP channels in the DA cells caused hyperpolarization of the cell membrane and subsequently enhanced ferrous iron uptake function (Du X. et al., 2016). Results showed that activation of KATP channels resulted in increased free iron levels in the SK-N-SH cells and this was partially blocked by DMT1 knockdown. Further studies showed that activation of KATP channels by diazoxide prolonged Fe2+-evoked currents in DMT1-transfected HEK293 cells using whole cell patch clamp recordings (Du X. et al., 2016). And inhibition of the KATP channels protected the DA neurons from the ferrous iron insult (Du X. et al., 2016). These results suggest that the activation of KATP channels could enhance DMT1-mediated iron uptake, resulting in an increased intracellular iron contents and oxidative stress, and ultimately cell damage.

It has been proved that there is a relationship between NMDARs and KATP channels. Elevated mRNA expressions of KATP channels and NMDARs subunits were found in human SN DA neurons from PD patients (Schiemann et al., 2012). NMDARs stimulation in vitro induced the co-activation of KATP channels in subthalamic neurons (Shen and Johnson, 2010). Activity of KATP channels in medial DA neurons of the SN enabled NMDA-mediated bursting in vitro and in vivo in anesthetized mice (Schiemann et al., 2012). Furthermore, activation of KATP channels in already metabolically challenged SN DA neurons could promote excitotoxicity and increase NMDARs-mediated calcium loading (Schiemann et al., 2012). Calcium-triggered ROS production from mitochondria in turn activated KATP channels in highly vulnerable SN DA neurons (Liss et al., 2005). In addition, NMDARs-mediated NO production might activate KATP channels in large DRG neurons via direct S-nitrosylation of SUR1 subunit (Kawano et al., 2009). The link of KATP channels and DMT1 indicated that co-activation of NMDARs and KATP

FIGURE 1 | Possible interaction between N-methyl-D-aspartate receptors (NMDARs) activation and iron overload. NMDARs activation might promote divalent metal transporter 1 (DMT1)-mediated iron influx via protein kinase A (PKA)-Rhes-DMT1, NO-Dexras1-peripheral benzodiazepine receptor-associated protein 7 (PAP7)-DMT1 signaling cascade pathway and enhance iron releasing from lysosome, then aggravated iron-induced cell damage. In addition, iron overload might stimulate calcium release from endoplasmic reticulum (ER). This indicated that glutamate-induced neurotoxicity and iron participated to a vicious cycle of neuronal death.

channels could affect iron uptake function of DMT1, which might contribute to iron accumulation and the degeneration of neurons in neurological diseases. Further investigation should be conducted in the future to reveal the underlying mechanisms.

#### Iron Affected NMDARs-Mediated Synaptic Transmission

It was reported that iron has an effect on NMDARsmediated synaptic plasticity. Investigation demonstrated that glutamatergic neurotransmission pathways were regulated by dietary iron (Han and Kim, 2015). They showed that NMDARs were significantly elevated in the prefrontal cortex and hippocampus of iron-loaded rats. These findings indicate the important role of iron in learning and memory via regulation NMDARs-mediated neurotransmission. In addition, study showed that intracellular iron signaling could modulate neuronal excitability in the hippocampus (White et al., 2016). They showed that iron was released in neurons from lysosome can regulate NMDARs mediated glutamatergic excitability via PKC/Src/NR2A pathway (White et al., 2016). This indicates that iron can act in a feedback manner to modulate NMDA function and thus maintain the excitability of neurons in normal condition. In addition, NMDARs activation induce rapid opening of Ca2<sup>+</sup> channel across cell membranes, followed by an increase in free Ca2<sup>+</sup> ions in the cytoplasm and subsequent signaling cascade in the CNS (Paul and Connor, 2010). Hippocampal neurons require iron to generate RyR-mediated calcium signals after NMDARs stimulation, which in turn promotes ERK1/2 activation, an essential step of sustained LTP (Muñoz et al., 2011). It was reported that iron stimulated RyR-mediated Ca2<sup>+</sup> release from endoplasmic reticulum (ER) in PC12 cells (Múñoz et al., 2006). Further study showed that iron-induced ROS generation is required for RyR-mediated Ca2<sup>+</sup> release for long-term potentiation in primary hippocampal neurons (Muñoz et al., 2011).

However, evidence has also proved that iron was involved in glutamate excitotoxicity (Yu et al., 2009). As is known that glutamate excitotoxicity could be induced by threohydroxyaspartate (THA), which could inhibit glutamate uptake and lead to accumulation of synaptic glutamate and over stimulation of the postsynaptic receptors. It was found that increased iron level was involved in THA-induced glutamate excitotoxicity in rat spinal cord tissue (Yu et al., 2009). And iron chelator deferoxamine (DFO) could completely prevent THA-induced motor neuron degeneration. Importantly, NMDARs inhibitor MK-801 could inhibit glutamate and hypoxia/reperfusion-mediated damage, but not the kainate/AMPA antagonist CNQX (Yu et al., 2009). This

# REFERENCES

Abboud, S., and Haile, D. J. (2000). A novel mammalian iron-regulated protein involved in intracellular iron metabolism. J. Biol. Chem. 275, 19906–19912. doi: 10.1074/jbc.m000713200

supports the idea that iron contributed to glutamate-induced excitotoxicity via NMDARs and regulation of iron level might be an effective strategy for protection against glutamate-induced excitotoxicity. Therefore, iron is required for NMDARsmediated synaptic plasticity in physiologic conditions. However, iron overload might promote NMDARs-mediated excitotoxicity possibly by enhancing RyR-mediated Ca2<sup>+</sup> release and calciuminduced cell damage. Further studies should be conducted to investigate this possibility and the underlying mechanisms.

# CONCLUSION AND PERSPECTIVES

In this review article, we describe the possible relationship between NMDARs activation and iron deposit, which both contributed to the pathogenesis of neurological diseases including PD and AD. NMDARs activation might promote iron accumulation by enhancing DMT1-mediated iron influx, stimulating iron releasing from lysosome, and regulating DMT1 expression, thus aggravating iron-induced cell damage. On the other hand, iron overload might aggravate NMDARsmediated glutamate excitotoxicity. This suggests that glutamateinduced neurotoxicity and iron participate to a vicious cycle of neuronal death (**Figure 1**). Therefore, regulation of iron levels might be effective for protection against glutamate-induced excitotoxicity and NMDARs might be a potential target for the treatment of iron-induced neurodegenerative processes. The drug discovery strategy is being oriented toward the development of new molecules targeting both iron overload and activation of NMDARs. However, NMDARs-mediated excitotoxicity and iron deposit were also reported in other neurological diseases such as stroke, Huntington' disease (Carvajal et al., 2016; Li and Reichmann, 2016; Petrova et al., 2016). The exact mechanisms underlying interaction between NMDARs activation and iron accumulation in these diseases should be conducted in the future. The future of this research might have significant impact on the clinical treatment of these neurological diseases.

# AUTHOR CONTRIBUTIONS

HX wrote the manuscript. HJ and JX revised the manuscript.

# ACKNOWLEDGMENTS

This work was supported by grants from the National Foundation of Natural Science of China (31371081, 81430024 and 31471114), the Department of Science and Technology of Shandong Province (ZR2015JL011) and the Bureau of Science and Technology of Qingdao (15-9-1-20-jch), Taishan Scholarship and Program for New Century Excellent Talents in University.

Alley, G. M., Bailey, J. A., Chen, D., Ray, B., Puli, L. K., Tanila, H., et al. (2010). Memantine lowers amyloid-β peptide levels in neuronal cultures and in APP/PS1 transgenic mice. J. Neurosci. Res. 88, 143–154. doi: 10.1002/jnr.22172

Altamura, S., and Muckenthaler, M. U. (2009). Iron toxicity in diseases of aging: Alzheimer's disease, Parkinson's disease and atherosclerosis. J. Alzheimers Dis. 16, 879–895. doi: 10.3233/JAD-20 09-1010


striatal medium spiny neurons. Mol. Cell. Neurosci. 37, 335–345. doi: 10.1016/j. mcn.2007.10.007


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer DT and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Xu, Jiang and Xie. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Efficient Binding of the NOS1AP C-Terminus to the nNOS PDZ Pocket Requires the Concerted Action of the PDZ Ligand Motif, the Internal ExF Site and Structural Integrity of an Independent Element

#### Li-Li Li1,2† , Katryna Cisek<sup>1</sup> and Michael J. Courtney1,2 \* †

<sup>1</sup> Molecular Signalling Laboratory, Department of Neurobiology, A. I. Virtanen Institute, University of Eastern Finland, Kuopio, Finland, <sup>2</sup> Neuronal Signalling Laboratory, Turku Centre for Biotechnology, University of Turku, Turku, Finland

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Frantisek Jursky, Slovak Academy of Sciences, Slovakia Florian Freudenberg, University Hospital Frankfurt, Germany

> \*Correspondence: Michael J. Courtney michael.courtney@utu.fi

#### †Present address:

Li-Li Li and Michael J. Courtney, Neuronal Signalling Laboratory, Turku Centre for Biotechnology, University of Turku, Turku, Finland

> Received: 22 December 2016 Accepted: 20 February 2017 Published: 15 March 2017

#### Citation:

Li L-L, Cisek K and Courtney MJ (2017) Efficient Binding of the NOS1AP C-Terminus to the nNOS PDZ Pocket Requires the Concerted Action of the PDZ Ligand Motif, the Internal ExF Site and Structural Integrity of an Independent Element. Front. Mol. Neurosci. 10:58. doi: 10.3389/fnmol.2017.00058 Neuronal nitric oxide synthase is widely regarded as an important contributor to a number of disorders of excitable tissues. Recently the adaptor protein NOS1AP has emerged as a contributor to several nNOS-linked conditions. As a consequence, the unexpectedly complex mechanisms of interaction between nNOS and its effector NOS1AP have become a particularly interesting topic from the point of view of both basic research and the potential for therapeutic applications. Here we demonstrate that the concerted action of two previously described motif regions contributing to the interaction of nNOS with NOS1AP, the ExF region and the PDZ ligand motif, efficiently excludes an alternate ligand from the nNOS-PDZ ligand-binding pocket. Moreover, we identify an additional element with a denaturable structure that contributes to interaction of NOS1AP with nNOS. Denaturation does not affect the functions of the individual motifs and results in a relatively mild drop, ∼3-fold, of overall binding affinity of the C-terminal region of NOS1AP for nNOS. However, denaturation selectively prevents the concerted action of the two motifs that normally results in efficient occlusion of the PDZ ligand-binding pocket, and results in 30-fold reduction of competition between NOS1AP and an alternate PDZ ligand.

Keywords: nos1ap, nNOS, PDZ domain, fluorescence polarization, dissociation constant, molecular dynamics simulation, multi-site binding, ZLc-002-1

# INTRODUCTION

NMDA receptor mediated signaling through nNOS is increasingly recognized as a contributor to a number of neurological conditions including stroke, depression and neuropathic pain (Florio et al., 2009; Hill et al., 2012; Doucet et al., 2013; Lee et al., 2015). Experimental models suggest that NOS1AP may mediate the actions of NMDA-driven nNOS function for example in excitotoxic conditions such as neonatal hypoxia and anxiety induced by chronic mild stress (Li et al., 2013; Zhu et al., 2014). It has been specifically suggested that nNOS-NOS1AP interaction could be a therapeutic target in schizophrenia (Weber et al., 2014; Freudenberg et al., 2015),

while NOS1AP has emerged as a highly significant associated gene in cardiovascular conditions (Arking et al., 2006; Kapoor et al., 2014). For these reasons the interaction between nNOS and NOS1AP is emerging as a potential target for drug development. Both peptide and small molecule inhibitors of the interaction have already been reported (Li et al., 2013, 2015; Zhu et al., 2014), and improved understanding of the interaction between nNOS and NOS1AP may assist in future drug development.

The interaction had originally been considered to be a canonical PDZ interaction (Jaffrey et al., 1998; discussed in Courtney et al., 2014), which involves the C-terminal 7–9 residues of a ligand (NOS1AP in this case) docking into a well-defined pocket in a PDZ domain (between residues 14–98 of nNOS in this case, Tochio et al., 1999). Supporting this, deletion of the C-terminus of NOS1AP eliminates binding to nNOS by conventional pulldown or immunoprecipitation methods (Jaffrey et al., 1998; Li et al., 2015; Candemir et al., 2016). Conflicting with this, is that the last 9–20 residues of NOS1AP are not sufficient for interaction to be detected by either pulldown or immunoprecipitation methods (Li et al., 2015; Candemir et al., 2016). The explanation for this finding is that the C-terminal PDZ ligand motif alone has exceedingly low affinity for nNOS (Li et al., 2015) but it nevertheless contributes stabilization to a primary interaction with an internal NOS1AP sequence containing an ExF motif. This results in ∼5-fold increased affinity and a ∼5 fold slower off-rate and, in the context of multiple wash steps of a conventional pulldown assay this amounts to the difference between strong binding and no binding (Li et al., 2015).

Thus the concerted action of these two interaction sites leads to a longer-lived protein complex and this appears to be important for the efficiency of mediation of signal transduction pathways, specifically the activation of the p38 pathway and excitotoxic neuronal cell death (Li et al., 2015). Nevertheless, it has not been shown whether docking of NOS1AP to nNOS via the ExF motif alone has any influence on the competence of the nNOS PDZ pocket to bind ligands, nor even the apparent affinity with which NOS1AP occupies the actual PDZ ligand-binding pocket when NOS1AP is bound to nNOS. The independence and interdependence of the two nNOS-interacting motifs has not been directly addressed.

Here we report that denaturation of the C-terminal nNOSbinding region (which contains both the ExF motif and the PDZ ligand) does not prevent interaction via the ExF motif, suggesting the latter does not require a structure that can be denatured. Surprisingly, however, we find the denatured binding region interacts with nNOS N-terminus at a similar affinity whether or not the PDZ ligand is present, in sharp contrast to previous findings using the native binding region (Li et al., 2015) in which ExF and PDZ ligand regions cooperate to generate high-affinity stable binding of nNOS. As a result, the reported increase in affinity of the C-terminal NOS1AP region containing both nNOS binding motifs is largely lost in the denatured peptide. This suggests PDZ pocket occupancy depends on a native structure. Indeed we show directly that, whereas the two-site binding region of NOS1AP (residues 400–506, human numbering) in native form exhibits high affinity competition against exogenous ligands for the nNOS PDZ-binding pocket (∼0.8 µM), denaturation of this NOS1AP fragment reduces competition at the pocket 30-fold. This shows that a relatively small change in overall affinity of the two-site binding region, NOS1AP[400–506], can mask a considerable drop in binding of one of the sites, and demonstrates the importance of investigating interactions at each site independently where feasible. The PDZ-binding C-terminal ligand of NOS1AP is a short peptide, and a molecular dynamic simulation suggest that no rigid structure is required for docking of the PDZ ligand to nNOS, which appears to only involve the last 3–4 amino acids. However, secondary structure prediction algorithms reveal potential structural elements that form between the ExF motif region and the PDZ ligand. Our data is consistent with a requirement for a structural element outside the ExF and PDZ motifs to be in a native conformation to allow concerted action of the ExF motif and PDZ ligand interactions. In denatured state, even if the two interaction sites can bind nNOS independently of one another, they cannot cooperate with one another to generate the increased apparent affinity of interaction otherwise seen, and this is most likely for steric reasons. The potential relevance of this additional requirement for nNOS-NOS1AP to the development of inhibitory strategies is discussed.

#### MATERIALS AND METHODS

#### Peptides

Peptide "GDLV" refers to NH2-RRRRWGDLV-COOH, whereas F-GDLV refers to the N-terminally fluoresceinated variant. Both were obtained from Genic Bio (Shanghai, China) and were of >92 and >97% purity respectively. Peptide "EIAV" refers to NH2- DSLDDEIAV-COOH, corresponding to the last 9 amino acids of rat/mouse NOS1AP, was synthesized as described Li et al. (2013) by Xigen AG (Lausanne, Switzerland). The corresponding human sequence is NH2-DGLDDEIAV-COOH, i.e., a S > G substitution at position −7 from the C-terminal valine. We previously aligned the C-terminal region from multiple species (Li et al., 2015) which showed that birds, reptiles and terrestrial and marine mammals typically have a serine at this position, fish typically have a cysteine, while the primates we aligned have a glycine. As we do not find evidence for the contribution of the residues distal to the C-terminus to binding, and this sequence alone binds nNOS extremely weakly (Li et al., 2015), serine does not appear to contribute to binding. As glycine has no side chain (only two hydrogens on the alpha-carbon), its contribution to binding is unlikely to be greater than serine. Notably, rat, mouse and human nNOS-PDZ domains are identical in the region from residues 6– 126 except for D/G substitution at amino acid 69. This residue is not close to the PDZ-ligand binding pocket and faces solvent in solved structures 1B8Q (Tochio et al., 1999) and 1QAV/1QAU (Hillier et al., 1999).

#### Antibodies

Antibody against NOS1AP (rabbit polyclonal, R-300, sc-9138, RRID:AB\_2251417) was from Santa Cruz Biotech, and Dy-Light coupled secondary antibody was from Cell Signaling Technologies.

# Plasmid constructs

fnmol-10-00058 March 13, 2017 Time: 16:51 # 3

pET28a-TAT-NOS1AP[400–506], pET28a-TAT-NOS1AP[400– 503], pET28a-TAT-NOS1AP[400–503] E429A, pET28a-TAT-NOS1AP[400–503] F431A, pET28a-nNOS[1–155], pGEX-nN OS[1–155], pGEX-6P-NOS1AP[400–503], pGEX-6P-NOS1AP [400–506], pGEX-6P-NOS1AP[432–506], pGEX-6P-NOS1AP [400–503] F431A and pGEX-6P-NOS1AP[400–506] F431A, encoding human protein sequences (amino acids as specified) fused to GST, His and His-TAT tags, and pGEX-HRV3C protease expression vector were previously described Li et al. (2015).

# Recombinant Protein Expression and Purification

His-TAT-NOS1AP fusions were purified under native conditions using lysozyme-based lysis conditions (Courtney and Coffey, 1999) or under denaturing conditions as described Becker-Hapak et al. (2001). His-nNOS(1–155) was generated as described Li et al. (2015) as was GST-nNOS(1–155) and GST. All HisnNOS[1–155] used for fluorescence polarization (FP) assays was polished by Superdex200 size-exclusion chromatography as described Li et al. (2015). NOS1AP[400–506], NOS1AP[400– 503], NOS1AP[432–503] and NOS1AP[400–506] F431A were obtained from GST fused versions by on-column cleavage using GST-HRV3C protease as described Li et al. (2015).

#### Solid-Phase No-Wash Binding Assay

This was carried out as previously described Li et al. (2015). Briefly, 5 µl beads bound to 1 µg GST-fused nNOS protein (or GST for background measurement) were rotated with concentrations of recombinant His-TAT-NOS1AP peptides as shown for 1 h at 4◦C and transferred to multiscreen filter plates (Millipore). Unbound peptides were removed by centrifugation, beads were resuspended in LSB (Cao et al., 2005) and protein eluted with Laemmli buffer. The affinity of each peptide was determined using fluorescent immunoblotting quantified with the Odyssey infrared imaging system (LI-COR) and ImageJ as follows. Bound peptide, together with the background binding of each TAT peptide to GST beads, was quantified in each blot using a set of standard concentrations of the same TAT fragment on the same blot. Background binding to GST, barely detected except at the highest fragment concentration, was subtracted in each case. Fitting was performed for each replicate with Excel using the quadratic solution formula to account for bound ligand, fb = B/L<sup>0</sup> = {+(K<sup>d</sup> + P<sup>0</sup> + L0)− √ [(K<sup>d</sup> + P<sup>0</sup> + L0) <sup>2</sup> – 4L0P0]}/2L0.

# Fluorescence Polarization and Competition Fluorescence Polarization

FP assays were carried using a BMG Polarstar OPTIMA reader as previously described Li et al. (2015). Briefly, synthetic peptide corresponding to the optimal nonamer ligand for the PDZ ligand-binding pocket of nNOS, based on the peptide scan of Stricker et al. (1997), was labeled with fluorescein at the N-terminus and used as a fluorescent ligand. This is referred to as F-GDLV. Fluorescence titration was performed by adding increasing amounts of His-tagged nNOS(1–155) to a constant amount of the fluoresceinated peptide (1 µM) in FP buffer (50 mM Tris buffer, pH 7.4, containing 100 mM NaCl, 1 mM EDTA, 0.1% BSA). The dissociation constant (Kd) was obtained by fitting the titration curves with the classical one-site binding model with MS Excel Solver the quadratic solution formula to account for bound ligand, as shown above in the section "solidphase no wash assay." The K<sup>d</sup> values were obtained in triplicate from each of two different batches of His-nNOS, curve fitting was performed on each replicate. The mean ± SEM shown in **Figure 2** and **Table 2** (10.7 ± 0.8 µM) corresponds to the results of curve fittings to the 6 replicates. The individual nNOS batches gave values of 10.2 ± 1.7 µM and 11.2 ± 0.2 µM.

Competition FP was carried out using 1 µM F-GDLV peptide 4 µM His-nNOS(1–155) and increasing amounts of competitor peptide as indicated. Approximate values for Kc, the dissociation constant of the competitor, were obtained using the formula for single site competition as described in Harris et al. (2001). For each experiment the data were fitted with MS Excel Solver according to the measured dissociation constant for the specific nNOS batch used in the experiment (see above).

#### Molecular Dynamics Simulation

All simulations were carried out using the Desmond Molecular Dynamics System v2.2 (D. E. Shaw Research, New York, NY, USA) and Schrodinger suite tools (Schrodinger, LLC, Portland, OR, USA) on the supercomputing clusters of the Centre for Scientific Computing (Espoo, Finland). NMR structures (pdb ID 1B8Q) were imported using Schrodinger's Maestro to assign proper bond orders. Next, the structures were prepared using Schrodinger's Protein Preparation Wizard, following a standard solvation box with 10 Å buffering, 150 mM Na<sup>+</sup> and Cl<sup>−</sup> ion buffering, and SPC water model. The OPLS-AA/2005 force field was used for all simulations with a multi-step minimization procedure using default settings to relax the system prior to simulation. For the molecular dynamics production runs, the NPT ensemble was used and a trajectory for 10 ns generated and visualized using the Maestro Trajectory Player. Three poses are shown at 0.15 ns intervals to show the rapid movement of the non-docked ligand residues compared with the docked ones.

# RESULTS

We previously demonstrated that it is possible to use the ExF site-bearing peptides alone as neuroprotectants in models of excitotoxic neuronal stress. This was achieved using recombinantly expressed and purified TAT-fused peptides that pass the plasma membrane (Li et al., 2015). In this previous report, we found using a quantitative solid-phase binding assay that TAT-fused peptides of NOS1AP (incorporating residues from 400 onward, human usage) exhibited ∼5-fold lower affinity when the PDZ ligand (residues 504–506) was lacking (**Table 1**, column 1; Li et al., 2015). This was consistent with a ∼5-fold lowering of the off-rate of interaction of the nNOS-NOS1AP complex when the PDZ ligand motif was present that we observed using an independent assay system. Additional

TABLE 1 | Comparison of K<sup>d</sup> values for native and denatured forms of NOS1AP for nNOS[1–155] previously reported and shown here.


Affinity constants were obtained by fitting each replicate for each peptide. Means ± SEM K<sup>d</sup> values are indicated. Native His-TAT protein data is taken from Li et al., 2015 whereas the denatured protein data carried out in parallel is shown here in Figure 1. Note that denaturation selectively reduces the affinity of NOS1AP[400– 506] ∼3-fold. The increased affinity of the essentially non-binding ExF point mutants lacking of PDZ ligand is an expected consequence of denaturation, but these proteins still have very low affinity for nNOS and this effect is unlikely to influence the specific binding of NOS1AP fragments with functional ExF motifs.

quantitative and qualitative protein interactions supported this interpretation and suggested that the PDZ motif, while having little affinity itself, nevertheless contributed to the formation of a stable interaction of nNOS with NOS1AP (Li et al., 2015).

# A Native Conformation of NOS1AP Is Required for the Concerted Action of the PDZ Ligand and ExF Motif Region in Binding nNOS

It is now well established that regions of proteins, domains of proteins and even entire proteins can exist in natively unfolded or intrinsically unstructured states (called IDP or intrinsically disordered proteins; for review see Latysheva et al., 2015; Wright and Dyson, 2015). It is thought that this allows these proteins to adopt confirmations required for target-binding without imposing such conformational constraints in the absence of binding, thereby permitting the binding to different targets. These regions and proteins are resistant to denaturation. In contrast, other regions or often entire proteins are structured. In this case, they form specific structures, and these are denaturable by heat, chemical denaturants or other conditions to which the proteins are not adapted. Thus, the comparison of protein function before and after denaturation is an investigative approach that can be used to determine the importance of structural elements for the specific functions of a protein. This does not necessarily mean that denaturation or loss of structure (or the opposite) is a physiological mechanism of regulation, although in some cases this is case (reviewed in Mitrea and Kriwacki, 2013; Csizmok et al., 2016). The light-induced switch of the J-alpha sequence of plant photoprotein Lov2 domains from alpha-helical to unfolded state (Harper et al., 2003) is one prominent example, which has been exploited by synthetic biologists to generate optogenetic regulators of cell signaling (Strickland et al., 2008; Wu et al., 2009; Lungu et al., 2012; Melero-Fernandez de Mera et al., 2013, 2017; Guntas et al., 2015).

We previously reported two-site affinity measurements for NOS1AP-nNOS interaction using TAT-fused NOS1AP protein expressed and isolated from bacterial expression systems under native conditions (Li et al., 2015). Uptake of recombinant TAT-fusions into cells is promoted by denaturation and they are presumed to be refolded in the cytoplasm by intracellular chaperone systems (Becker-Hapak et al., 2001). Considering the prevalence of functional intrinsically disordered regions in the proteome and the lack of a predicted structured domain in the C-terminal region of NOS1AP, we decided to measure affinities of the denatured TAT-proteins in parallel with the native versions, for the same sites using the same no-wash pulldown assay as we previously used for the native proteins (Li et al., 2015). We intentionally used a no-wash pulldown assay in this study because we previously reported that the conventional washing steps, which have no physiological counterpart, greatly exaggerate the impact of PDZ motif on overall binding affinity (Li et al., 2015). **Figure 1** shows that the affinity of nNOS for denatured NOS1AP[400–503], the NOS1AP form that lacks the PDZ ligand motif, is similar to that of the native protein (Li et al., 2015; **Table 1**). However, in contrast to the results obtained with native protein, the affinity difference between NOS1AP[400–506] and NOS1AP[400–503] was almost completely lost, both having a similar affinity to that previously described for native NOS1AP[400–503] (**Table 1**). This suggests that, although the interaction of the internal ExF motif region (present in NOS1AP[400–503]) with nNOS does not require a pre-existing natively folded conformation, the increased affinity obtained by concerted action of the two interaction motifs in NOS1AP[400–506], in contrast does require a native structure.

# Fluorescence Polarization Can Be Used to Directly Probe the Occupancy of the PDZ Ligand-Binding Pocket of nNOS

This sensitivity of concerted NOS1AP binding to nNOS to denaturation has considerable potential impact on the ability of NOS1AP to compete with ligands for the nNOS PDZ pocket. To investigate this in more detail, we set up a solution-phase FP competition assay to selectively monitor the occupancy of the nNOS-PDZ ligand pocket independent of overall protein interaction, by use of a fluorescent ligand of the pocket (**Figure 2A**). Based on a previously published random peptide screen (Stricker et al., 1997) we developed the 9 amino acid FP ligand fluorescein-RRRRWGDLV, termed here 'GDLV.' This interacts with nNOS[1–155] with an affinity 10.7 ± 0.8 µM (**Figure 2B**), and unlabelled GDLV competes with a comparable affinity (10.7 ± 2.0 µM, **Figure 2C**). This is in sharp contrast to the C-terminal 9 amino acid PDZ ligand motif from NOS1AP, DSLDDEIAV, which was reported to have extremely low affinity (>600 µM in fluoresceinated form, Li et al., 2015; **Table 2**). Consistent with this, we cannot detect any competition for F-GDLV/nNOS[1–155] interaction by up to 300 µM unlabelled NOS1AP-9C peptide ("EIAV," **Figure 2D**).

# The nNOS-NOS1AP Interaction Inhibitor ZLc-002-1 Shows Weak Affinity for the nNOS PDZ Ligand-Binding Pocket

Recently N-(2-carboxyacetyl)-D-valine-methyl ester, also known as ZLc-002-1, a valine-based analog of the C-terminus of NOS1AP (**Figure 3A**), has been proposed as a competitive

inhibitor for nNOS PDZ pocket ligands and has shown efficacy against anxiety induced by chronic mild stress (Zhu et al., 2014). However, no actual binding or competition data has been reported for this molecule. Using the FP competition assay we find that ZLc-002-1 does compete with the PDZ pocket ligand F-GDLV (**Figure 3B**). The affinity is rather weak (>100 µM, Kc, which represents the dissociation constant of the competitor, i.e., ZLc-002-1 in this case), but it should be noted

fnmol-10-00058 March 13, 2017 Time: 16:51 # 5


#### TABLE 2 | Comparison of K<sup>d</sup> values for binding of NOS1AP forms to nNOS[1–155] and Kc values for competing with F-GDLV ligand as measured by fluorescence polarization (FP).

Means ± SEM are indicated. "n.d.", several proteins cleaved from GST fusion were not measured by a comparable assay and therefore there are no K<sup>d</sup> estimates in these cases. However by more qualitative assays NOS1AP[400-506]E429A, F431A and NOS1AP[432-506] exhibit no or reduced interaction with nNOS[1-155] (Li et al., 2015).

that this charged (and probably cell-impermeant) carboxylic acid species is thought to be generated in cells exposed to the cell-permeant ester prodrug ZL002 by cleavage with esterases. Ester-loading is a well-known strategy used for decades for example in loading calcium dyes into cells. In such cases, incubation with 2.5 µM uncharged ester form has been reported to lead to 200–800 µM cytoplasmic concentrations after 15 min of exposure (Hagen et al., 2012). Our observed Kc for ZLc-002-1 is therefore well within the intracellular concentration range expected from the reported 72 h incubation with 10 µM ester to obtain an effect in cell-based experiments (Zhu et al., 2014).

# The ExF Site of NOS1AP Is Required to Compete With Docking of PDZ Ligand to nNOS But Is Not Sufficient

In contrast to ZLc-002-1, native NOS1AP[400–506] which contains two sites of interaction (Li et al., 2015), shows a much higher affinity in this assay (0.52 ± 0.21 µM; **Figure 4B**). NOS1AP[432–506] completely lacks the ExF motif and, like the NOS1AP-9C peptide (**Figure 2D**), also fails to compete (**Figure 4D**) even though the C-terminal PDZ ligand motif is present. Native NOS1AP[400–503], which as a His-TAT fusion binds nNOS almost as well as NOS1AP[400–506] in native state

(Li et al., 2015) and equally well when denatured (**Figure 1**), does not compete with PDZ ligand in this assay (**Figure 4C**). This is the expected outcome if the ExF and PDZ ligand sites are non-overlapping and independent, because 400–503 does not have a PDZ ligand and should not be able to compete with the PDZ probe F-GDLV. Finally, NOS1AP[400–506] F431A which contains a PDZ ligand but bears an inactivated ExF motif, and NOS1AP[400–503] F431A which has neither functional ExF nor PDZ motif, each fail to compete in the FP assay (**Figures 4E,F**). In conclusion, these data show the NOS1AP PDZ ligand motif alone does not compete effectively at the PDZ ligand pocket, and nor is the functional ExF region alone sufficient. But a NOS1AP sequence containing both functional ExF region and PDZ ligand together is able to occupy the PDZ ligand pocket. This data is consistent with and provides more direct evidence than previously available in support of an earlier model of the nNOS-NOS1AP interaction (Figure 3Biii of Li et al., 2015).

### The Native State of a Structural Element in NOS1AP Greatly Increases the Affinity for the PDZ Ligand Binding Pocket of nNOS

We applied this FP competition assay to the native and denatured His-TAT-fused NOS1AP[400–506] peptides that we had already characterized by the solid phase assay (**Figure 1**; Li et al., 2015). We find that native His-TAT- NOS1AP[400–506] competes with F-GDLV, whereas denatured His-TAT-NOS1AP[400–506] interaction is 30-fold weaker (**Figures 5A,B**; significant difference at P < 0.0001, two-tailed unpaired t-test, n = 4). This difference greatly exceeds the ∼3-fold impact of denaturation on overall binding affinity by no-wash pulldown assay (**Figure 1**; K<sup>d</sup> and Kc data are compared in **Table 2**). In the competition-FP assay we only monitor the state of the PDZ ligand pocket so this data alone does not distinguish between denaturation destroying the affinity of the ExF region or disrupting the concerted action of the ExF region and PDZ motif. However, considering the proteininteraction results shown in **Figure 1** and **Table 1** (also Li et al., 2015), we can conclude that the interaction by the ExF region is resistant to denaturation and it is the concerted action of the two sites to occlude the pocket that fails upon denaturation. Furthermore it is clear that the denaturation procedure used here does not merely result in loss of functional protein by aggregation as this would affect all binding events similarly but the data shows a 30-fold effect on PDZ ligand occupancy and no effect on nNOS binding the via ExF motif region (compare Kc for NOS1AP[400–506] with and without denaturation and K<sup>d</sup> for NOS1AP[400–503] with and without denaturation, **Table 2**).

# Molecular Dynamics Simulation of nNOS-NOS1AP PDZ Ligand Interaction

Two possible explanations are that either the PDZ motif itself is denaturable or that there is an additional denaturable element required for facilitating the concerted action of the two interaction sites. The term "denaturable" refers here to an element the properties of which are lost after denaturation of its structure, and excludes intrinsically disordered sequences. This does not necessarily imply that denaturability is itself a functional property that is modulated in vivo although in some cases this has been demonstrated (Mitrea and Kriwacki, 2013; Csizmok et al., 2016).

Canonical C-terminal PDZ ligands are short peptides, typically too short to have a stable structure, and are thus considered to represent a class of intrinsically disordered regions

(Ivarsson, 2012). The NOS1AP C-terminal PDZ ligand itself, residues 504–506 and potentially those preceding it, is therefore not by itself expected to form a stable structure. Any stable structure it might adopt in free solution could lead to an energy barrier impeding the acquisition of the bound state in the PDZ ligand binding pocket. We wished to more precisely predict how this might influence the interpretation of the binding data. Therefore, we carried out a molecular dynamic simulation based on the existing NMR structure for nNOS docked with the melatonin receptor peptide (1B8Q.pdb). This process calculates the positions of all atoms from a starting point after each of a number of specified small time steps. This models a relevant atomic-level representation of the binding event between the peptide and protein in the predicted binding pocket and reveals the biophysics of this binding event. This can provide support for a specific hypothesis or it can contradict it. For example if the simulation shows a candidate ligand leaves a binding pocket, the binding event is predicted to be non-existent.

Comparing the docking of MelA peptide in the published NMR structure (Tochio et al., 1999) and NOS1AP PDZ ligand (**Figure 6** and Supplementary Movie) reveals that interaction of the peptides with the PDZ pocket is predominantly based on backbone–backbone interactions rather than any specific rigid conformation of side chains. Specifically, the hydrogen bond interactions between the peptide and protein backbones are the key driving force of the interaction rather than other forces, (i.e., hydrophobicity, van der Waals forces). The simulation (**Figure 6** and Supplementary Movie) specifically shows that the last three residues remain docked in the pocket whereas the remaining six residues move rapidly over the protein on a picosecond timescale and do not remain docked with any specific surface region of nNOS. The results of the simulation are therefore consistent with (i) a lack of requirement for a pre-formed structure of the PDZ ligand motif for binding to the PDZ pocket and (ii) a participation of only the 3 most C-terminal amino acids of the ligand (isoleucine-alanine-valine) in the interaction with nNOS.

We conclude that a pre-existing structure of the PDZ ligand in NOS1AP is unlikely to be required. For these reasons the disruption of the concerted action of the two binding sites by denaturation of the NOS1AP C-terminal nNOS binding domain cannot be ascribed to either motif alone - neither the internal ExF motif, as it interacts even if denatured (**Figure 1**), nor the PDZ ligand as it interacts without adopting a specific fold (**Figure 6**).

## Secondary Structure Prediction of the C-Terminal nNOS Binding Region of NOS1AP

We have found that denaturation does not affect the affinity of the ExF-containing region of NOS1AP for nNOS (residues 400– 503, **Table 1**). MD simulation suggests the C-terminal PDZ motif interaction involves only 3 amino acids (**Figure 6**), which is too short to be sensitive to denaturation. Therefore, the denaturationinduced sharp decline in competition at the PDZ pocket (**Figure 5** and **Table 2**) as well as the reduced interaction with nNOS of denatured NOS1AP containing both these sites (residues 400–506, **Table 1**) indicates that denaturation targets a third, previously unidentified, element in the C-terminal region of NOS1AP (residues 400–506) that is required for optimal binding to nNOS. We specifically infer a folded structural element most likely residing between the ExF region and the PDZ ligand at the extreme C-terminus. As the PDZ ligand of NOS1AP interacts with the PDZ domain pocket and deletion of only three amino

acids from the C-terminus of NOS1AP eliminates occlusion of the PDZ ligand binding pocket (**Figure 4C**), the ExF motif region must interact at another site on the compact nNOS N-terminal domain. The concerted simultaneous binding of both sites to the nNOS domain must impose steric requirements on their relative orientations, as was previously demonstrated for efficient binding of dimeric ligands to the tandem PDZ domains of PSD95-PDZ1-2 (Long et al., 2003). In the case of nNOS-NOS1AP interaction it seems likely that the steric requirements could be fulfilled by the inferred folded structural element of NOS1AP. To explore the possible presence of folded elements, we applied secondary structure prediction methods to the C-terminal nNOS binding region, i.e., NOS1AP residues 400–506 (**Figure 7A**).

The JPRED3 engine (Cole et al., 2008) predicts two main structural elements within the C-terminal nNOS binding region. The first is a β-sheet region around the ExF motif, from residues 425 to 432. The requirements for residues 429 and 431 but not 430 is consistent with a beta sheet structure bearing residues with alternating orientations, but the equal affinities of denatured 400–503 and 400–506 (**Figure 1**) suggests the ExF motif readily adopts a binding conformation even after denaturing conditions. The second structural element is a putative α-helix from 452 to 460, with a third predicted, again α-helical, around residues 484–489. These putative helical regions may potentially represent a denaturable element that could serve a role in orientation of ExF and PDZ ligand motifs and conversely denaturing them might then lead to steric incompatibility of the two sites to bind a single nNOS molecule, which is what we observe.

#### DISCUSSION

Our data show that not only are two binding elements, the ExF motif and PDZ ligand required for the optimal binding of NOS1AP to nNOS as described before (Li et al., 2015), but that there is in addition a requirement for a denaturable structural element for the concerted action of these two sites (**Figures 7Bi,ii**). This is presumably important to provide appropriate steric conditions for the two sites to simultaneously access their respective binding sites on nNOS. Although it was previously shown (Li et al., 2015; and shown again here in **Figures 1**, **4**, **5**) that the NOS1AP C-terminal PDZ ligand motif was dispensable for binding of NOS1AP to nNOS – the ExF motif region was sufficient for binding – it was not previously clear whether or not the ExF-interaction also occluded the PDZ motif site of the bound nNOS molecule by either steric or allosteric mechanisms, or left it vacant and able to interact. Here we have demonstrated directly that the interaction of NOS1AP to nNOS via the ExF motif region and PDZ ligand interaction are separable events, and thus binding of nNOS by NOS1AP[400– 503] via the ExF motif alone has no detectable occluding effect on the PDZ pocket which remains equally able to bind ligand (shown schematically in **Figure 7Biii**; corresponding data in **Figure 4C**).

Conversely, we provide here an estimate of the apparent competitor affinity at the PDZ ligand binding pocket by a NOS1AP sequence containing the C-terminal PDZ ligand motif. The latter motif of NOS1AP in isolation is known to have extremely weak affinity for nNOS (**Figure 2**; Li et al., 2015; **Table 2**). It is also known that the presence of this ligand together with the ExF motif within the C-terminal 107 amino acids of NOS1AP, (i.e., residues 400–506) generates a high affinity interaction with nNOS as previously described (Li et al., 2015) but this does not mean the PDZ pocket would be occupied with the same apparent affinity as the overall dissociation constant for the NOS1AP-nNOS complex. We show directly that the two motifs together lead to a high affinity occupancy of the PDZ ligand binding pocket (apparent Kc ∼0.5 µM, **Figure 4B**) that efficiently competes with an alternate model ligand (F-GDLV). It is this efficient competition at the pocket that requires three

fnmol-10-00058 March 13, 2017 Time: 16:51 # 9

components – the ExF motif, the C-terminal PDZ ligand and native conformation of a denaturable element within residues 400-506 of NOS1AP (**Figure 7Bi**) - not the overall NOS1APnNOS interaction for which the ExF motif site is sufficient (e.g., **Figure 7Biii**). Thus, without a native structure, the ExF motif-containing region still interacts with nNOS (**Figure 1** and **Table 1**) but the contribution of the PDZ ligand, which is believed to be critical for signaling (Li et al., 2013, 2015), is lost (**Figure 7Bii**; cf. **Figure 5B**). The impact in terms of competition at the PDZ pocket is similar to deletion of the C-terminal PDZ ligand (**Figures 7Biii,iv**; cf. **Figure 4C** and **Table 2**).

We previously reported that it is possible to pharmacologically inhibit the cellular impact of nNOS-NOS1AP interaction using a competing PDZ ligand (Li et al., 2013) and more recently reported a complementary approach using the ExF motif

bearing peptide devoid of PDZ ligand (Li et al., 2015). The distinct third requirement for a structural component that we report here, is potentially a source of additional inhibitors. Structure-based drug design may not easily target this element, particularly in the complete absence of structural coordinates of the molecule. The precise location of the third element is not known, but it is likely to reside between ExF motif (residues 429–431) and PDZ ligand (residues 504–506). The predicted alpha-helical regions between residues 452–460 and/or 484–489 are therefore potential candidates. Even while the structure of NOS1AP remains unsolved, the contributions of these two candidate regions could nevertheless be evaluated by mutational approaches. However, even without actually locating the residues making up this denaturable element, future library screens could reveal HTS hits that act at this site. It is not uncommon for the actual binding sites of HTSderived protein–protein interaction inhibitors to be unclear. For example, IC87201 inhibits the interaction of nNOS with PSD95 and it is believed to interact with nNOS (Florio et al., 2009) but that actual site of interaction of the nNOS molecule has not been mapped. A similar molecule, ZL006, was in fact described as having been designed precisely to disrupt the interaction between two secondary structure elements (Lai and Wang, 2010; Zhou et al., 2010), though once again the actual interaction site has yet to be determined (Bach et al., 2015). Therefore, it is important to identify the precise requirements for protein–protein interaction, not only for the design of new inhibitor molecules but also to interpret and fully understand the mechanism of inhibition by HTS hits so that their properties can be rationally explained and hits can be optimized. For instance, a screen for modulators targeting the denaturable element we have reported here would most easily be addressed using an assay that reports its function, i.e., allowing the concerted action of PDZ ligand and ExF motif for interaction of NOS1AP binding to nNOS. Such a screen would therefore inevitably also report inhibitors of both the PDZ ligand interaction and the ExF motif interaction. The actual site of action of screen hits would therefore have to be deconvolved by additional site-specific assays once screen hits have been revealed and the knowledge that three distinct elements are involved in the interaction is required to successfully achieve this.

We previously demonstrated that the PDZ ligand contributes an increased stability of the interacting complex and that competition at the PDZ ligand site alone inhibits signaling, presumably due to a reduced lifetime of this complex impacting on NMDA receptor/nNOS-dependent signaling outputs. The additive impact of the PDZ ligand on affinity is shown here to be largely lost after denaturation. This suggests that any successful targeting of the novel denaturable structural element revealed in this report would likely reduce the affinity of NOS1AP for nNOS and lifetime of the nNOS-NOS1AP complex in a manner similar to that we have observed by competition at the PDZ ligand-binding site (Li et al., 2015). Structural elements in proteins are not completely rigid and there is no evidence that the C-terminal region of NOS1AP acts like a highly stable globular protein. Therefore, it is reasonable to expect that the required structural element has sufficient dynamic 'breathing' movement to allow its targeting by small molecules in analogous way to that proposed for ZL006 (Lai and Wang, 2010; Zhou et al., 2010). Moreover, given the requirement for this element for optimal interaction of NOS1AP with nNOS and for occlusion of the PDZ ligand-binding pocket, any dynamic reorganization of this element could potentially form the basis of additional cellular regulation of nNOS-NOS1AP interaction. However, such cellular regulation is speculative as there is currently no evidence to support an endogenous regulation of this element at this time. Nevertheless, it should be noted that NOS1AP is not the only protein reported to interact with the PDZ ligandbinding pocket of nNOS. Other candidates for interaction at this site include phosphofructokinase-M (Firestein and Bredt, 1999), the melatonin A receptor (MelA, Tochio et al., 1999), alpha-adrenergic receptors (Pupo and Minneman, 2002), the transcriptional corepressor carboxyl-terminal binding protein 1 (CtBP1, Lin et al., 2003), and DHHC domain palmitoyl acetyl transferases such as ZDHHC23/NIDD (Saitoh et al., 2004). Therefore, the ability of NOS1AP to occlude the nNOS PDZbinding pocket, the requirement of the NOS1AP ExF motif and additional structural element in this occlusion and the possible pharmacological manipulation or even endogenous regulation of these determinants would be expected to determine whether nNOS is competent to recruit these additional candidate interacting proteins. This in turn could reshape nNOS-dependent signaling profiles and thus influence disease mechanisms involving nNOS.

# CONCLUSION

Our findings extend our understanding of nNOS-NOS1AP interaction by revealing the contribution of three separate elements. In addition to (i) the recently described internal NOS1AP-ExF motif region which is sufficient for interaction (Li et al., 2015), and (ii) the initially described NOS1AP-PDZ ligand which is not sufficient alone but increases stability and affinity in the presence of the ExF motif (Jaffrey et al., 1998; Li et al., 2015), we provide evidence here for (iii) a third NOS1AP element, the structure and therefore function of which is sensitive to denaturation, that orients the other two NOS1AP elements to allow them to form a stable interaction with nNOS and occlude alternate ligands from the nNOS ligand binding pocket. It would be important to determine if this third element is merely encoded in a robust and static manner by the amino acid sequence of NOS1AP or whether it is dynamically regulated by cellular signaling pathways and whether it is amenable to pharmacological modulation. Our findings therefore have implications for drug discovery opportunities for the numerous diseases that have been linked to nNOS and NOS1AP as well as the manner in which the small molecule inhibitor compounds for the different sites could be identified. It is important to consider that NOS1AP has been reported to mediate the activation of

p38MAPK (Li et al., 2013, 2015), a kinase which is in turn linked to excitotoxicity (Cao et al., 2004, 2005; Semenova et al., 2007), neurodegeneration (Dau et al., 2014; Ittner et al., 2014; Roy et al., 2015), neuropsychiatric conditions (Bruchas et al., 2011; Ehrich et al., 2015) and neuropathic pain (Berta et al., 2016). As the neurological side effects of direct p38MAPK inhibitors have been addressed in clinical trials by reducing brain penetrance (Krementsov et al., 2013), inhibiting p38MAPK in the CNS may be difficult. Targeting disease-related upstream mechanisms such as NOS1AP could in this case be an attractive alternative approach. Moreover, as NOS1AP has itself been directly associated with a range of disorders and diseases, it would be valuable to develop mathematical models to describe and predict its function and regulation under different conditions. As more details emerge of the regulatory mechanisms of NOS1AP, such as those we present here, a model to predict the interaction of NOS1AP with its targets and the impact of inhibitors at specific sites on these interaction surfaces becomes increasingly feasible.

# AUTHOR CONTRIBUTIONS

L-LL carried out the experimental work, its analysis, interpretation and presentation, acquired necessary resources and participated in writing the manuscript. KC carried out, analyzed and described the molecular dynamics simulation. MC planned and supervised the project, participated in the analysis, interpretation and presentation of the data, acquired and provided the resources required and wrote the manuscript.

# REFERENCES


# FUNDING

This work was funded by the University of Turku and National Institutes of Health National Cancer Institute Grant No. R01CA200417 (MC and L-LL), the European Union 7th Framework Programme Initial Training Networks FP7- PEOPLE-2013-ITN Project Number 608346 Project 'Brain Imaging Return To Health' r'Birth (KC and MC), the Magnus Ehrnrooth Foundation (MC) and the North Savo fund of the Finnish Cultural Foundation and the Instrumentarium Foundation (L-LL).

## ACKNOWLEDGMENTS

CSC – Scientific Computing, Ltd. is gratefully acknowledged for software licenses and computational resources, Tuomo Laitinen and Antti Poso at the Biocentre Finland Drug Discovery and Chemical Biology/Translational Technologies platform for access to these resources, and the Turku Bioimaging Screening Unit for access to facilities. We thank Yvonne Lai (Indiana University, Bloomington, IN, USA) for providing ZLc-002-1.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00058/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Li, Cisek and Courtney. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fnmol-10-00058 March 13, 2017 Time: 16:51 # 13

# Communication Impairment in Ultrasonic Vocal Repertoire during the Suckling Period of Cd157 Knockout Mice: Transient Improvement by Oxytocin

Olga L. Lopatina1, 2 \*, Kazumi Furuhara<sup>1</sup> , Katsuhiko Ishihara<sup>3</sup> , Alla B. Salmina<sup>2</sup> and Haruhiro Higashida<sup>1</sup>

*<sup>1</sup> Department of Basic Research on Social Recognition and Memory, Research Center for Child Mental Development, Kanazawa University, Kanazawa, Japan, <sup>2</sup> Department of Biochemistry, Medical, Pharmaceutical, and Toxicological Chemistry, Krasnoyarsk State Medical University Named after Prof. V.F. Voino-Yasenetsky, Krasnoyarsk, Russia, <sup>3</sup> Department of Immunology and Molecular Genetics, Kawasaki Medical School, Kurashiki, Japan*

#### Edited by:

*Andrew Harkin, Trinity College, Dublin, Ireland*

#### Reviewed by:

*Claudio V. Mello, Oregon Health and Science University, USA Daniela Tropea, Trinity College, Dublin, Ireland*

> \*Correspondence: *Olga L. Lopatina ol.lopatina@gmail.com*

#### Specialty section:

*This article was submitted to Neuroendocrine Science, a section of the journal Frontiers in Neuroscience*

Received: *31 October 2016* Accepted: *24 April 2017* Published: *17 May 2017*

#### Citation:

*Lopatina OL, Furuhara K, Ishihara K, Salmina AB and Higashida H (2017) Communication Impairment in Ultrasonic Vocal Repertoire during the Suckling Period of Cd157 Knockout Mice: Transient Improvement by Oxytocin. Front. Neurosci. 11:266. doi: 10.3389/fnins.2017.00266* Communication consists of social interaction, recognition, and information transmission. Communication ability is the most affected component in children with autism spectrum disorder (ASD). Recently, we reported that the *CD157/BST1* gene is associated with ASD, and that CD157 knockout (*Cd157*−/−) mice display severe impairments in social behavior that are improved by oxytocin (OXT) treatment. Here, we sought to determine whether *Cd157*−/<sup>−</sup> mice can be used as a suitable model for communication deficits by measuring ultrasonic vocalizations (USVs), especially in the early developmental stage. Call number produced in pups due to isolation from dams was higher at postnatal day (PND) 3 in knockout pups than wild-type mice, but was lower at PNDs 7 and 10. Pups of both genotypes had similarly limited voice repertoires at PND 3. Later on, at PNDs 7 and 10, while wild-type pups emitted USVs consisting of six different syllable types, knockout pups vocalized with only two types. This developmental impairment in USV emission was rescued within 30 min by intraperitoneal OXT treatment, but quickly returned to control levels after 120 min, showing a transient effect of OXT. USV impairment was partially observed in *Cd157*+/<sup>−</sup> heterozygous mice, but not in *Cd157*−/<sup>−</sup> adult male mice examined while under courtship. These results demonstrate that *CD157* gene deletion results in social communication insufficiencies, and suggests that CD157 is likely involved in acoustic communication. This unique OXT-sensitive developmental delay in *Cd157*−/<sup>−</sup> pups may be a useful model of communicative interaction impairment in ASD.

Keywords: CD157, Bst-1, communication, development, ultrasonic vocalization, oxytocin, autism

# INTRODUCTION

There is an increasing number of patients with syndromes of multiple etiologies that are covered by the umbrella of autism spectrum disorder (ASD) (Mychasiuk and Rho, 2016) Communication ability and emotional expression are major issues in children with ASD (Eigsti et al., 2011; DiStefano et al., 2016). Language communication from an early life stage is an essential tool for bidirectional information transmission (Fitch et al., 2010; Kuhl, 2011), and is clearly delayed in certain ASD subtypes.

Interdisciplinary research platforms (including non-human models) are interested in finding relevant pharmacological treatments (Shen et al., 2016; Wei et al., 2016; Zheng et al., 2016). In this regards, oxytocin (OXT) is a promising therapy. OXT is a neuropeptide with potent and profound effects on many physiological processes in the brain, cardiovascular system, and reproductive system (Gimpl and Fahrenholz, 2001). Specific pattern of OXT function (Zik and Roberts, 2015) directly relates to characteristics of behavior (Veenema and Neumann, 2008). Lower endogenous OXT levels are associated with impaired social cognition in various mental diseases including ASD, schizophrenia, and anxiety (Hoge et al., 2008; Zhang et al., 2012; Eapen et al., 2014; Strauss et al., 2015; Husarova et al., 2016; Massey et al., 2016). Further, OXT treatment rescues social behavioral deficits in animals (Jin et al., 2007; Freeman et al., 2016; Lawson et al., 2016; Teng et al., 2016), and patients with ASD during clinical trials to test its beneficial effects (Munesue et al., 2010; Althaus et al., 2016; Kosaka et al., 2016; Yatawara et al., 2016). We recently reported that under a social context, mutual interaction is significantly increased during the OXT arm of nasal administration, compared with the placebo arm, for ASD patients with intellectual disabilities (Munesue et al., 2016).

Behavioral changes are likely to be associated with brain OXT levels. OXT release from the soma, dendrites, and axon terminals of hypothalamic neurons occurs in response to intracellular Ca2<sup>+</sup> mobilization (Amina et al., 2010; Lopatina et al., 2010; Leng et al., 2015; Higashida, 2016). This mechanism is regulated by activity of ADP-ribosyl cyclase CD38, a multifunctional molecule that combines enzymatic and receptor properties, and plays a key role in OXT secretion, critically regulating maternal and social behavior in mice (Jin et al., 2007; Salmina et al., 2010; Lee, 2011; Mushtaq et al., 2011; Schmid et al., 2011). Moreover, this mechanism is suggested to affect human behavior during development and in adulthood (Munesue et al., 2010).

In a context of CD38 as a social behavior regulator, the same gene family, bone marrow stromal cell antigen-1 (BST-1), also attracts attention as a similar social behavior regulator. BST-1 was first isolated from bone marrow stromal cell lines (Kaisho et al., 1994), and BST1 was identified as CD157 by gene cloning (Itoh et al., 1994). Interestingly, despite the important role of CD157 in the immune system (Shimaoka et al., 1998; Lo Buono, 2014), the CD157/BST1 gene was identified as a risk-factor for neurodegeneration, particularly for Parkinson's disease (PD), or at least one of a variety of PD symptoms (Satake et al., 2009; Simón-Sánchez et al., 2011; Sharma et al., 2012; Zhu et al., 2012; Liu et al., 2013b) . In addition, a new role for CD157 has been reported in stem cells, which is that CD157 induces the catalysis of cyclic ADP-ribose in paneth cells, which promotes intestinal stem cell self-renewal in mice that are on a calorie-restricted diet (Yilmaz et al., 2012), and cyclic ADP-ribose/CD157 promotes the proliferation of lung stem/progenitor cells (Wu et al., 2013). However, though very recently CD157 immunoreactivity has been shown to colocalize with nestin-positive cells in the ventricular zone in the brain (Higashida et al., 2017), little is known about the role of CD157 in brain function or in the brain degeneration deficits of PD.

Recently, there have been reports that the CD157/BST1 gene is associated with other diseases, including ASD (Ceroni et al., 2014; Yokoyama et al., 2015). In addition, we have reported that CD157 knockout (Cd157−/−) mice display severe anxiety-related and depression-like behaviors or social avoidance that were reversed upon treatment with anti-depression drugs and OXT (Lopatina et al., 2014; Mizuno et al., 2015; Higashida et al., 2017). These findings prompted us to examine if the Cd157−/<sup>−</sup> mice could be a suitable model of ASD or autistic-like behavior with impaired social behavior in the absence of motor dysfunction, especially in an early stage of development that has not been previously studied in these knockout mice.

Thus, the aim of our present study was to investigate the effect of Cd157/Bst1 gene deletion on developmental aspects of vocal communication ability to determine if ASD-related communication deficits are due to a general impairment or developmental delay. Additionally, Cd157 knockout mice are a useful ASD mouse model from the viewpoint of ASD-related communication impairment or delay. Accordingly, we show that Cd157 knockout pups display a poor vocal repertoire (less variety in vocal pattern) during neonatal stages (postnatal day (PND) 3–10). Vocal syllable repertoire was reversely increased after 30–60 min of intraperitoneal injection of OXT in knockout pups at PND 7 and 10, although OXT did not advance the variety of vocal pattern. Finally, we show that the number of ultrasonic vocalization (USV) emissions is positively correlated with ADP-ribosyl cyclase activity and plasma OXT concentrations in wild-type but not knockout mice.

# MATERIALS AND METHODS

#### Animals

Cd157/Bst1 knockout (Cd157−/−) mice (C57BL/6J background) were previously described (Itoh et al., 1998) and maintained by crossbreeding homozygous mutant mice (Lopatina et al., 2014). Wild-type (Cd157+/+) and Cd157−/<sup>−</sup> mice were kept in the animal center under standard conditions (24◦C; 12-h light/dark cycle, lights on at 8:45 a.m.) in mouse cages (300 × 160 × 110 mm) with sawdust as bedding, and they received food and water ad libitum. Breeding pairs were separately maintained (1 pair per cage). At 21 days of age, offspring were removed, and housed in same-sex sibling pairs. Pups from postnatal days (PNDs) 3, 7, and 10 were used in this study.

Heterozygote CD157+/<sup>−</sup> mouse pups were maintained by backcrossing Cd157−/<sup>−</sup> with C57BL/6 mice. Offspring were genotyped as previously described (Itoh et al., 1998).

All animal experiments were conducted in accordance with the Fundamental Guidelines for the Proper Conduct of Animal Experiment and Related Activities in Academic Research Institutions under the jurisdiction of the Ministry of Education, Culture, Sports, Science, and Technology of Japan, and they were approved by the Committee on Animal Experimentation of Kanazawa University.

#### Behavioral Tests

The experimental animals were subjected to a series of behavioral tests performed 4 h before the dark phase. The mice were habituated to the room for 60 min before testing. The procedure for each behavioral test is described below. Dimensions of experimental chambers are represented as the (width × length × height). After each trial, the test chambers were cleaned with 80% alcohol and damp towels.

#### Ultrasonic Vocalization Recording

#### **Isolation-induced USV recording in a mouse pup**

Male C57BL/6, Cd157+/−, and Cd157−/<sup>−</sup> pups (PND 3–10) from breeding pairs were tested as previously described (Liu et al., 2008, 2013a). Before testing, the cage with pups and their parents were transferred from the animal room to the test room for approximately 60 min for adaptaion. For pup isolation USVs, call recordings were performed in an anechoic box (700 × 600 × 600 mm) using the method described by Shu et al. (2005). Each pup was placed in a 500-ml glass beaker in the anechoic box. The microphone was located 5 cm above the pup for the 2-min recording period. Naïve mouse pups (n = 10–15) were used at every step of experiments to exclude pup handling and tiredness of infant mice.

#### **Context-specific (courtship) USV recoding in adult mice**

Male C57BL/6 and Cd157−/<sup>−</sup> mice (10 weeks old, n = 13) were individually habituated to the testing environment for 30 min with a subsequent 5-min session pairing with an individual female. Vocalizations were recorded over a 5-min period.

#### **Analysis of USVs**

Recordings were performed blind to mouse genotype. Number of calls, and peak frequency (maximum syllable frequency point) and duration (difference between syllable start and end points) of emitted calls were measured using a USV monitor (SpectraLAB; Sound Technology Inc., Stage College, PA, USA; Liu et al., 2013a). Each syllable was classified as one of the following seven waveform categories: upward (up), downward (down), chevron, complex, harmonic, plate, or "V"-call. Classification was determined by internal pitch change, length, and shape, and was based on and adapted from previously described methods (Scattoni et al., 2008; Grimsley et al., 2011; Liu et al., 2013a; Zampieri et al., 2014). All syllable types exhibited typical structure from USV recorded spectrogram. Statistical analysis of peak frequency demonstrated a normal distribution pattern across data with relevant SD (data not shown). Thus, peak frequency and call duration were averaged across different syllable types. Click-like sounds of ≤40 ms in duration were filtered out of pup vocalizations (Liu et al., 2013a).

#### Blood and Tissue Collection

All blood and tissue collection was performed independently of behavioral tests. Mice at different ages were anesthetized by intraperitoneal injection of pentobarbital (20 mg/kg body weight). Blood samples of 0.1–0.2 ml were collected by cardiac puncture and centrifuged at 1,600 × g for 15 min at 4◦C.

The whole hypothalamus was removed according to the stereotaxic coordinates (Franklin and Paxinos, 2008), and it was then homogenized in 10 mM Tris-base (pH 7.4) using a 1-ml Teflon/glass homogenizer. The fresh homogenates were used to determine ADP-ribosyl cyclase activity as previously described (Graeff et al., 1994; Liu et al., 2008; Lopatina et al., 2014). Protein content was determined using a Bio-Rad protein assay kit with bovine serum albumin as a standard (Bio-Rad, Hercules, CA, USA).

#### Enzyme Immunoassay for Oxytocin

To determine the concentrations of OXT in the plasma, an OXT immunoassay kit was used according to the manufacturer's protocol (Assay Designs, Ann Arbor, MI) and as previously described (Jin et al., 2007; Zhong et al., 2016).

#### ADP-Ribosyl Cyclase Activity

Assessment of the ADP-ribosyl cyclase activity in the murine hypothalamus was performed in whole homogenates using the nicotinamide guanine dinucleotide technique as previously described (Graeff et al., 1994). Briefly, 2 ml of reaction mixtures containing 60µM NGD+, 50 mM Tris-HCl (pH 6.6), 100 mM KCl, and 10µM CaCl<sup>2</sup> were maintained at 37◦C with constant stirring. The samples were then excited at 300 nm, and fluorescence emission was continuously monitored at 410 nm with a Shimadzu RF-5300 PC spectrofluorometer (Kyoto, Japan). Activity was calculated from the linear portion of the 10 min time course by fitting a linear function to the data points recorded every 15 s. The specific ADP-ribosyl cyclase activity was calculated using cyclic GDP-ribose (cGDPR) standard, and the results are presented as nanomoles cGDPR per min per mg of protein (Jin et al., 2007; Lopatina et al., 2010).

#### OXT Treatment

OXT was injected (i.p.) in male pups (PNDs 7–10). Individual and group locomotor activities as well as isolation-induced USVs were recorded after 30, 60, 120, min of OXT administration, which was followed by blood plasma sampling.

#### Statistical Analysis

Data are expressed as mean ± SEM. The Kolmogorov-Smirnov test was used to examine cumulative sample distribution. Comparisons were made between two groups (Cd157+/<sup>+</sup> and Cd157−/−) using two-tailed Student's t-test (for normal distributions) or Mann–Whitney U-test (absence of normal distribution). Two-way analysis of variance (ANOVA) was used to determine Genotype × Age or Treatment × Call type interactions. Subsequently, post-hoc Turkey's or Sidak's multiple comparison tests were used for group comparisons. Categorical variables were compared by χ 2 test. In all analyses, P < 0.05 indicated statistical significance.

#### RESULTS

When isolated from their mothers, mouse pups produce USVs until PND 12 (Scattoni et al., 2009). Consequently, we examined isolation-induced USVs in Cd157+/<sup>+</sup> and Cd157−/<sup>−</sup> pups at PNDs 3, 7, and 10. Two-way ANOVA demonstrated a significant Genotype × Age interaction [P < 0.0001, F(2, 14) = 17.54, n = 10–15]. Analysis of USVs by post-hoc Turkey's multiple comparison test showed a significantly higher call number at PND 3 in Cd157−/<sup>−</sup> mice compared with Cd157+/<sup>+</sup> controls (17 ± 2 calls/min vs. 5 ± 1 calls/min per 3-min session, respectively; P < 0.001, n = 10–15; **Figure 1**). During the first postnatal week, the number of USVs in Cd157−/<sup>−</sup> mice dramatically decreased to 4 ± 2 calls/min at PND 10 (P = 0.02, χ <sup>2</sup> = 0.974, n = 10–15). By contrast, the number of USVs in Cd157+/<sup>+</sup> mice increased at PNDs 7 and 10: 12 ± 3 and 10 ± 2 calls per 3-min session, respectively, from 5 ± 1 calls at PND 3 (**Figure 1**). The number of USVs in Cd157−/<sup>−</sup> mice at PNDs 7 and 10 was higher than in Cd157−/<sup>−</sup> mice (P < 0.0001, n = 10–15). **Supplementary Table 1** summarizes call type and number produced by Cd157+/<sup>+</sup> and Cd157−/<sup>−</sup> mice at PNDs 3, 7, and 10.

Frequency of USVs decreased equally in both genotypes during development from PNDs 3–10, with no significant Genotype × Age interaction [two-way ANOVA: P < 0.1893, F(2, 74) = 1.681, n = 10–1]. Indeed, there were no differences between Cd157+/<sup>+</sup> and Cd157−/<sup>−</sup> mice, except for lower USV frequency at PND 7 in Cd157+/<sup>+</sup> mice (61.1 ± 1.1 kHz) compared with Cd157−/<sup>−</sup> mice (65.0 ± 1.2 kHz; post-hoc Turkey's multiple comparison test: P = 0.02, χ <sup>2</sup> = 0.7778; **Supplementary Figure 1A**). Duration of USVs was essentially unchanged in Cd157+/<sup>+</sup> mice during development (**Supplementary Figure 1B**), although two-way ANOVA

showed a significant Genotype × Age interaction [P < 0.0001, F(2, 84) = 45.10, n = 10–15; Genotype, P < 0.0001, F(1, 84) = 25.4; Age, P < 0.0001, F(2, 84) = 34.76]. At PND 3, duration of USVs was extremely high in Cd157−/<sup>−</sup> mice (86.2 ± 4.3 ms) compared with Cd157+/<sup>+</sup> mice (52.8 ± 1.8 ms) (post-hoc Turkey's multiple comparison test: P < 0.0001). During the first postnatal week, USV duration in Cd157−/<sup>−</sup> mice reduced to 49.8 ± 1.5 ms (PND 7). Consistent with a previous report (Scattoni et al., 2009), neither genotype emitted isolation-induced USVs after PND 10.

Voice patterns of USVs from Cd157+/<sup>+</sup> and Cd157−/<sup>−</sup> mice in the first two postnatal weeks are shown in **Figure 2A**. Threeday-old Cd157−/<sup>−</sup> pups emitted USV calls consisting of three categories (72% downwards, 22% upwards, and 6% chevron calls), whereas 3-day-old Cd157+/<sup>+</sup> pups demonstrated similar waveform patterns but consisting of only two call types (15% upward and 85% downward). Two-way ANOVA demonstrated a significant Genotype × Call type interaction at PND 7 [P < 0.0001, F(2, 114) = 22.39, n = 10–15]. However, interestingly, at PND 7, Cd157+/<sup>+</sup> mice emitted a broader, multi-faceted repertoire of USVs, which already contained six call types (44% downward, 23% chevron, 13% complex, 8% upward, 6% plate, and 6% "V" calls). In contrast, Cd157−/<sup>−</sup> pups displayed no complexity in USV categories, and only emitted two USV call types (44% downward and 58% chevron). Two-way ANOVA detected a significant Genotype × Call type interaction at PND 7 [P < 0.0001, F(6, 266) = 45.18, n = 10–15]. This USV pattern of a rich repertoire in Cd157+/<sup>+</sup> mice and poor repertoire in Cd157−/<sup>−</sup> pups continued from PND 7 until 10, with twoway ANOVA confirming a significant Genotype × Call type interaction at PND 10 [P < 0.0001, F(6, 266) = 44.69, n = 10–15]. Detailed inspection found that 7-day-old Cd157−/<sup>−</sup> pups emitted a higher percentage of chevron calls [Cd157+/<sup>+</sup> (23%) vs. Cd157−/<sup>−</sup> (56%), P < 0.001, n = 10–15], but the same percentage of downward voice events (44 and 43% for both genotypes, respectively). Thus, while USV repertoire developed in wildtype pups, there was no developmental progress in knockout pups, and instead "retardation" was apparent. To investigate gene-dosage effects on USV repertoire, we examined isolationinduced USVs in heterozygote (Cd157+/−) infant males at PNDs 3, 7, and 10 (**Supplementary Table 1**). Cd157+/<sup>−</sup> pups displayed intermediate patterns of USVs, in which no clear development of complex patterns with varied voice types was observed, although 3–4 voice types were detected (**Figure 2B**).

Next, we sought to determine whether OXT facilitates USV development and can influence the observed retardation at PNDs 7 and 10. Within 30 min, intraperitoneal OXT injection (10 ng/mouse) triggered a dramatically richer USV repertoire, with two-way ANOVA confirming a significant Treatment × Call type interaction [P < 0.001, F(2, 266) = 73.53, n = 10–15] (**Figure 3**, **Supplementary Tables 2**, **3**). At 60 min after OXT injection, USV pattern reduced to 3 or 4 voice types (from six types at 30 min). At 120 min, the repertoire returned to the initial pattern of two voice types. Thus, OXT clearly influences the normal wild-type pattern, but the effect is transient, and OXT cannot facilitate development nor fundamentally alter voice repertoire.

PBS injection had no effect on USV repertoire in Cd157−/<sup>−</sup> pups at PNDs 7 or 10 (**Figure 3**, **Supplementary Tables 2**, **3**). Intraperitoneal OXT injection (10 ng/mouse) in Cd157+/<sup>−</sup> at PNDs 7 and 10 strikingly altered the rich USV repertoire within 30 min, which returned to initial levels (two voice types in 120 min), similar to Cd157−/<sup>−</sup> mice (data not shown).

Adult male mice do not produce USVs under social isolation. Consequently, to assess vocalization ability in adult mice, we measured USV parameters during courtship, because many USVs are emitted in the context of sexual behavior from male mice to females. With these USV recordings, Cd157−/<sup>−</sup> mice produced more calls in comparison to Cd157+/<sup>+</sup> mice (45 ± 12 vs. 17 ± 6 respectively, P < 0.05, n = 10–15; **Figure 4**). However, Cd157−/<sup>−</sup> mice emitted USVs with a rich repertoire (seven different voice types), similar to Cd157+/<sup>+</sup> mice (also seven types). We did not detect any difference in peak frequency or call duration between genotypes (data not shown). These results show that Cd157−/<sup>−</sup> mice can emit "normal" USVs, and therefore do not possess a genetic defect in voice emissions. Our results also show that Cd157−/<sup>−</sup> pups with restricted USV patterns show "a" developmental delay that is "corrected" in adult ages when seeking a partner.

To determine whether the observed alterations in USVs might be linked to impaired OXT secretion (because of Cd157 deletion), we examined ADP-ribosyl cyclase activity, which is required for central OXT secretion (Jin et al., 2007) in the hypothalamus, as well as OXT plasma levels in Cd157 wild-type and knockout mice during the first 10 days of life (**Figure 5**). Two-way ANOVA detected a non-significant Genotype × Age interaction in ADP-ribosyl cyclase activity [P = 0.0600, F(2, 24) = 3.17, n = 10–15]. However, significant differences separately influenced by Genotype [P = 0.0025, F(1, 24) = 11.37] and Age [P = 0.0001, F(2, 24) = 123.4] were detected. Post-hoc Sidak's multiple comparison test confirmed significantly lower hypothalamic ADP-ribosyl cyclase activity at PND 7 (P = 0.018, n = 10–15) and PND 10 (P = 0.003, n = 10–15) in Cd157−/<sup>−</sup> mice compared with Cd157+/<sup>+</sup> mice. Two-way ANOVA also showed a significant Genotype × Age interaction in plasma OXT levels [P = 0.045, F(2, 35) = 3.386, n = 10–15], with lower plasma OXT levels in knockout mice at PND 7 [P = 0.033, n = 10–15] and PND 10 (P = 0.04, n = 10–15] (post-hoc Sidak's multiple comparison test). ADP-ribosyl cyclase activity and plasma OXT levels reversed, and starting from PND 7, we observed lower ADP-ribosyl cyclase activity and plasma OXT levels in Cd157−/<sup>−</sup> mice compared with wild-type mice (P < 0.05). In Cd157+/<sup>+</sup> mice, call number was positively correlated with higher ADP-ribosyl cyclase activity (R <sup>2</sup> = 0.3228; **Figure 5A**) and plasma OXT levels (R <sup>2</sup> = 0.75; **Figure 5B**), according to age. In contrast, Cd157−/<sup>−</sup> mice demonstrated negative correlation between emitted call number and ADPribosyl cyclase activity (R <sup>2</sup> = 0.4648; **Figure 5A**) and plasma OXT (R <sup>2</sup> = 0.5428; **Figure 5B**) over PNDs 3–10.

USV characterization and plasma OXT levels after OXT administration in Cd157−/<sup>−</sup> mice at PND 7 are listed in **Table 1**.

At 30 and 60 min after OXT application, USV features were altered and showed a significant difference compared with PBS application. Hence, it is likely that changes in USV reflect shifting plasma OXT levels.

#### DISCUSSION

Using a mouse model of neurodevelopmental disease to study social communication and interaction is an important strategy for shedding light on the black box of molecular processes that underlie formation and implementation of social behavior, as well as aiding identification of new molecular targets for pharmacological treatment of communication impairments. Here, we demonstrated that Cd157 knockout mice have developmental abnormalities, and do not exhibit the normal, full range of USVs during the lactating period.

Mice emit USVs under different social conditions throughout their lifespan (Scattoni et al., 2009). Pups separated from the nest and their mother emit vocalization signals with clear communicative value, which correlate with social approach or exploratory behavior (D'amato, 1991; Scattoni et al., 2008). Since the first description by Zippelius and Schleidt (1956), neonatal USVs have been interpreted as a communicative behavior (Wöhr and Schwarting, 2007; Hammerschmidt et al., 2009; Scattoni et al., 2009; Shepard and Liu, 2011). Neonatal USVs also map onto later development of adult anxiety profiles (Dichter et al., 1996; Scattoni et al., 2009). Consequently, loss of USV complexity after PND 3, as observed in Cd157−/<sup>−</sup> mice, may reflect delayed development of communication skills. Moreover, it may be partially associated with the pre-existing autistic- (anxietyand avoidance-like) behavior that has already been reported in Cd157−/<sup>−</sup> mice (Lopatina et al., 2014).

It is particularly intriguing that adult knockout mice show a full, complex syllable repertoire, whereas juveniles do not, unless under exogenous OXT administration. This begs the question of when exactly juveniles develop the ability to produce these other vocalizations in a more natural context. As the motherpup social interaction is most important in mice, isolation from dams initiates USV emission (Scattoni et al., 2009). Nonetheless, it is also well-known that isolation from dams to a cold location causes pups to intensively emit very distinct USVs of the complex type (Branchi et al., 2001). This type of alarm call was indistinguishable between pups of the different genotypes (data not shown). Therefore, a delay in variety of syllable patterns relates more to social interaction-based maturation, because lifethreatening-based USVs are the same.

Vocal delay was restored at the adult age when seeking a partner, suggesting that any delay is likely corrected in adults. Estimation that this correction is undertaken during the juvenile period later than PND 10 is possible, but was not tested in the current experiments. This question "cannot be tested" because it is known that isolation-induced USVs sharply decrease after

PND 10, during the late suckling period (Branchi et al., 2001). At best, we can state that this delay is not apparent in adults in the context of partner seeking.

The OXT neurotransmitter system regulates USV signaling in rodents, and has been previously reviewed (Branchi et al., 2001). We found that communicative ability of wild-type pups corresponded to higher plasma OXT levels and elevated ADPribosyl cyclase activity in the hypothalamus, while preserved communicative ability of Cd157−/<sup>−</sup> pups was not positively associated with either variable with age. In fact, both values were apparently strain-specific. At PND 7, OXT administration restored communication skills and plasma OXT levels in CD157−/<sup>−</sup> mice, but had no effect in wild-type mice. Further, these strain differences in ADP-ribosyl cyclase activity and OXT levels are indirect, and only show a developmental profile of the two variables. Accordingly, our findings do not provide evidence that these two variables are related in a causative way, or to vocal behavior.

Exogenous OXT administration in isolated rat pups reduces the rate of USVs, although OXT antagonist treatment does not change USVs (Insel and Winslow, 1991), and OXT null mutant mouse pups display fewer USVs than wild-type controls (Winslow et al., 2000). This counterintuitive finding (given that exogenous OXT also decreases calling) has been interpreted as evidence that social separation is not perceived as distress, and does not induce USVs in the absence of OXT (Winslow et al., 2000; Winslow and Insel, 2002; Scattoni et al., 2009). Takayanagi et al. (2005) reported that OXT receptor knockout pups exhibited fewer USVs at 7-days-old, which is consistent with the hypothesis that OXT neurotransmission is necessary for perception of social separation and subsequent USV response. In our current experiments, ADP-ribosyl cyclase-mediated OXT release was altered by PND 7 in Cd157−/<sup>−</sup> mice, resulting in aberrant USVs. Interestingly, this effect was prevented by exogenous OXT administration. Notably, our previous observations (Liu et al., 2008) demonstrated that Cd38 knockout mice had abnormal vocalizations during the neonatal period, while 7-day-old Cd38 knockout pups (compared with Cd38 wild-type mice) expressed significantly lower levels of hypothalamic and neurohypophysial ADP-ribosyl cyclase activity.

CD157 and CD38 have similar amino acid sequences and ADP-ribosyl cyclase activity (catalyzing cyclic ADP ribose), but distinct NAD base exchange activity (producing nicotinic acid adenine dinucleotide phosphate; Higashida et al., 2017). Further, both are involved in NAD<sup>+</sup> metabolism (Lee, 2012; Higashida et al., 2017). Additionally, the human CD157 gene shares unique genomic organization with the human CD38 gene (Ortolan et al., 2002). Taking into consideration that

and 10. ADP-ribosyl cyclase activity was measured as rate of cyclic GDP-ribose formation in whole-cell homogenates isolated from the hypothalamus. (B) Plasma oxytocin (OXT) concentration correlates with call number in *Cd157*+/<sup>+</sup> (blue round) and *Cd157*−/<sup>−</sup> (red squares) mice at PNDs 3, 7, and 10. Data were obtained from pups at the indicated ages (arabic numbers). (C) Recorded ADP-ribosyl cyclase activity, plasma concentration of OXT, and call number in *Cd157*+/<sup>+</sup> and *Cd157*−/<sup>−</sup> mice at PNDs 3, 7, and 10. \**P* < 0.05 from *Cd157*+/<sup>+</sup> pups at the corresponding measuring day.

#### TABLE 1 | USV characterization and plasma OXT level after OXT administration in Cd157−/<sup>−</sup> pups.


\**p* < *0.001.*

CD38 and CD157 belong to the same family of NAD+ glycohydrolases (Lee, 2001; Guse, 2005), it is tempting to speculate that CD157 might be involved in the mechanism of OXT release in a similar manner as CD38. However, this needs

to be clarified in relation to developmental delay in future experiments.

Previously, we found that CD157 and CD38 and their signaling pathways are shared in anxiety, autism, early atypical motor function, speech and language disorders, and social avoidance (Lopatina et al., 2014). Although we demonstrated differences between Cd38 and Cd157 knockout mouse behavior (even on different mouse backgrounds; Jin et al., 2007; Lopatina et al., 2014; Higashida et al., 2017), the exact role of CD157 in OXT secretion and regulation of social behavior requires further investigation. One suggested approach would be to generate Cd38/Cd157 double knockout mice and confirm the compensatory role of Cd157 in Cd38 knockouts.

In summary, this is the first study to demonstrate association of CD157 with neonatal developmental delay in communicative ability, which could be rescued by exogenous OXT administration. Additionally, the CD157 gene may be a valuable candidate for involvement in increased risk for anxiety or social avoidance. Our data may have important implications for future studies of anxiety disorders, social avoidance, and communication delay in children with ASD, which can slowly develop with age.

#### AUTHOR CONTRIBUTIONS

OL and HH conceived and designed the research. All performed experiments on mice. OL and KF analyzed data and prepared figures. OL and AS prepared the initial draft; OL, AS, and HH revised the manuscript. All authors reviewed the final manuscript and approved its publication.

#### REFERENCES


#### ACKNOWLEDGMENTS

This work was supported by grant-in-aid from "Integrated research on neuropsychiatric disorders" carried out under the Strategic Research Program for Brain Sciences by the Ministry of Education, Culture, Sports, Science and Technology of Japan (MEXT) and from the Japan Agency for Medical Research and Development (AMED) and also by the industry - Academia Collaborative R&D Programs (COI) from MEXT.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnins. 2017.00266/full#supplementary-material

Supplementary Figure 1 | Development of isolation-induced USV production in Cd157+/<sup>+</sup> and Cd157−/<sup>−</sup> mouse pups. Frequency (A) (*<sup>n</sup>* <sup>=</sup> 18) and duration (B) (*n* = 18) of ultrasonic calls were measured in wild-type (C57BL/6) and *Cd157*−/<sup>−</sup> pups at post-natal days (PNDs) 3–10. Data are shown as mean ± SEM. Two-way ANOVA with *post-hoc* Tukey's multiple comparison test was performed. <sup>∗</sup>*P* < 0.05 from *Cd157*+/<sup>+</sup> pups.

Supplementary Table 1 | Call types and call numbers C57BL/6, Cd157−/−, Cd157+/<sup>−</sup> male mice at PND3,7,10.

Supplementary Table 2 | Call types and call numbers after PBS or oxytocin (OXT) treatment in Cd157−/<sup>−</sup> male mice at PND7.

Supplementary Table 3 | Call types and call numbers after PBS or oxytocin (OXT) treatment in Cd157−/<sup>−</sup> male mice at PND10.


study of CD157/BST-1 as a neuroregulator. BMC Neurosci. 18:35. doi: 10.1186/s12868-017-0350-7


extant PD risk loci among the Dutch. Eur. J. Hum. Genet. 19, 655–661. doi: 10.1038/ejhg.2010.254


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer DT and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Lopatina, Furuhara, Ishihara, Salmina and Higashida. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Hypothesis for Using Pathway Genetic Load Analysis for Understanding Complex Outcomes in Bilirubin Encephalopathy

#### Sean M. Riordan1, 2, Douglas C. Bittel 3, 4, Jean-Baptiste Le Pichon1, 2, 4, 5, Silvia Gazzin<sup>6</sup> , Claudio Tiribelli 6, 7, Jon F. Watchko<sup>8</sup> , Richard P. Wennberg<sup>9</sup> and Steven M. Shapiro1, 2, 4, 5 \*

*<sup>1</sup> Division of Child Neurology, Department of Pediatrics, Children's Mercy Hospital, Kansas City, MO, USA, <sup>2</sup> Department of Neurology, University of Kansas Medical Center, Kansas City, KS, USA, <sup>3</sup> Ward Family Heart Center, Children's Mercy Hospital, Kansas City, MO, USA, <sup>4</sup> Department of Pediatrics, University of Missouri-Kansas City School of Medicine, Kansas City, MO, USA, <sup>5</sup> Department of Pediatrics, University of Kansas Medical Center, Kansas City, KS, USA, <sup>6</sup> Italian Liver Foundation, Centro Studi Fegato (CSF), Trieste, Italy, <sup>7</sup> Department of Medical Sciences, University of Trieste, Trieste, Italy, <sup>8</sup> Division of Newborn Medicine, Department of Pediatrics, University of Pittsburgh School of Medicine, Pittsburgh, PA, USA, <sup>9</sup> Department of Pediatrics, University of Washington, Seattle, WA, USA*

#### Edited by:

*Andrew Harkin, Trinity College, Dublin, Ireland*

#### Reviewed by:

*Adelaide Fernandes, Universidade de Lisboa, Portugal Jun Zhang, Texas Tech University Health Sciences Center, USA*

> \*Correspondence: *Steven M. Shapiro sshapiro@cmh.edu*

#### Specialty section:

*This article was submitted to Neurogenomics, a section of the journal Frontiers in Neuroscience*

Received: *07 June 2016* Accepted: *02 August 2016* Published: *18 August 2016*

#### Citation:

*Riordan SM, Bittel DC, Le Pichon J-B, Gazzin S, Tiribelli C, Watchko JF, Wennberg RP and Shapiro SM (2016) A Hypothesis for Using Pathway Genetic Load Analysis for Understanding Complex Outcomes in Bilirubin Encephalopathy. Front. Neurosci. 10:376. doi: 10.3389/fnins.2016.00376* Genetic-based susceptibility to bilirubin neurotoxicity and chronic bilirubin encephalopathy (kernicterus) is still poorly understood. Neonatal jaundice affects 60–80% of newborns, and considerable effort goes into preventing this relatively benign condition from escalating into the development of kernicterus making the incidence of this potentially devastating condition very rare in more developed countries. The current understanding of the genetic background of kernicterus is largely comprised of mutations related to alterations of bilirubin production, elimination, or both. Less is known about mutations that may predispose or protect against CNS bilirubin neurotoxicity. The lack of a monogenetic source for this risk of bilirubin neurotoxicity suggests that disease progression is dependent upon an overall decrease in the functionality of one or more essential genetically controlled metabolic pathways. In other words, a "load" is placed on key pathways in the form of multiple genetic variants that combine to create a vulnerable phenotype. The idea of epistatic interactions creating a pathway genetic load (PGL) that affects the response to a specific insult has been previously reported as a PGL score. We hypothesize that the PGL score can be used to investigate whether increased susceptibility to bilirubin-induced CNS damage in neonates is due to a mutational load being placed on key genetic pathways important to the central nervous system's response to bilirubin neurotoxicity. We propose a modification of the PGL score method that replaces the use of a canonical pathway with custom gene lists organized into three tiers with descending levels of evidence combined with the utilization of single nucleotide polymorphism (SNP) causality prediction methods. The PGL score has the potential to explain the genetic background of complex bilirubin induced neurological disorders (BIND) such as kernicterus and could be the key to understanding ranges of outcome severity in complex diseases. We anticipate that this method could be useful for improving the care of jaundiced newborns through its use as an at-risk screen. Importantly, this method would also be useful in uncovering basic knowledge about this and other polygenetic diseases whose genetic source is difficult to discern through traditional means such as a genome-wide association study.

Keywords: bilirubin, kernicterus, pathway genetic load score, bilirubin encephalopathy, GWAS

# INTRODUCTION

Chronic bilirubin encephalopathy (CBE—also known as kernicterus) is caused by exposure to toxic levels of unconjugated bilirubin (UCB) in severely jaundiced newborns resulting in deposition of bilirubin in the brain (Shapiro, 2010). Neonatal jaundice, present in 60–80% of newborns, is commonly due to unconjugated hyperbilirubinemia and is assessed clinically by measuring total serum bilirubin (TSB) and UCB. Effective monitoring and treatment has helped make the development of CBE a rare event in the majority of countries despite the prevalence of neonatal jaundice. The concentration of albumin, the major binding protein of bilirubin in the blood, and its ability to bind bilirubin are important factors in the transport of bilirubin to the liver for clearance. To prevent excessive hyperbilirubinemia infants are commonly treated with phototherapy to convert bilirubin into water-soluble photoisomers that can be excreted by the liver and kidneys. When hyperbilirubinemia exceeds a specific level defined in clinical guidelines (e.g., American Academy of Pediatrics Subcommittee on Hyperbilirubinemia, 2004) exchange transfusions may be performed. While these treatments have significantly decreased the overall incidence of CBE in most industrialized countries, CBE with lifelong neurological sequelae still occurs, and is a major problem in many areas of the world, especially in lowand middle-income countries (Olusanya et al., 2014; Greco et al., 2016).

Despite the widespread use of TSB levels as a guide for intervention in jaundiced newborns, TSB alone has been shown repeatedly to be a poor predictor of neurological impairment (Ahlfors et al., 2009; Gamaleldin et al., 2011). TSB measures bilirubin levels in blood, where more than 99% of bilirubin is bound to proteins, especially albumin (90% or more) (Hulzebos and Dijk, 2014), but does not measure bilirubin in brain tissue. The toxic form of bilirubin, i.e., free, unbound bilirubin (Bf) correlates better with both acute and chronic bilirubin encephalopathy (Nakamura et al., 1985; Wennberg et al., 2006; Ahlfors et al., 2009; Amin et al., 2016). The inability to easily and reliably measure Bf levels, in addition to the complex nature of unbound bilirubin deposition and toxicity, has impaired investigations of predictive accuracy of Bf for CBE. Furthermore, subtle chronic bilirubin encephalopathy, part of a spectrum of kernicterus or bilirubin-induced neurological dysfunction (BIND) (Shapiro, 2005, 2010; Bhutani and Johnson-Hamerman, 2015), has been difficult to link retrospectively to TSB and UCB levels.

We hypothesize that the production, transport, and elimination of bilirubin, as well as the cellular response to bilirubin, have an important genetic component. If this genetic component could be better understood, it would enhance identification of patients at risk and allow for an individualized therapeutic approach to hyperbilirubinemia. In this paper, we first review genetic methods commonly used to study poly-genetic diseases in large populations and show that these methods cannot be used in a disease such as kernicterus. We then propose a modified use of the pathway genetic load (PGL) score method to decipher the genetic factors that contribute to increased susceptibility to bilirubin neurotoxicity. Finally, we summarize our proposal and describe the current status of this ongoing project.

# THE PROMISE OF GENETICS

#### Genetic Analysis and Disease

The study of molecular genetics has identified more than 6000 different monogenetic diseases, mostly through familial linkage analysis. Autosomal dominant, autosomal recessive, and X-linked genetic diseases with complete penetrance were the first and easiest to identify by linkage analysis. Later, genome-wide association studies (GWAS) were developed to identify genes contributing to incompletely penetrant and polygenetic diseases. Both methods have successfully identified key genetic contributors to various diseases. However, these genetic analyses have also had many failures. Studies of thousands of individuals have failed to uncover significant genetic heritability for numerous diseases (the "missing heritability" problem), leading to a re-evaluation of the basic tenants of GWAS (reviewed in Marjoram et al., 2014).

#### Current Strategies: Successes and Shortcomings

Family linkage analysis is very good at identifying rare mutations with large effects that confer susceptibility to common diseases. Examples include the BRCA1 and BRCA2 genes in breast cancer (Easton et al., 1995; Ford et al., 1998). By comparison, GWAS studies excel at analyzing complex diseases and disease genes with a weak effect (Risch and Merikangas, 1996). The most common target in GWAS studies has been the single nucleotide polymorphism (SNP) (Palmer and Cardon, 2005). SNPs identified in GWAS studies have largely been shown to be representative of loci of interest and thus have an indirect association with the actual causal variant(s). For example, the FTO gene has been shown through GWAS studies to associate with obesity. However, its effect on obesity was eventually shown to be through an enhancer mechanism on downstream genes, including IRX3 and IRX5 (Smemo et al., 2014; Claussnitzer et al., 2015). A major advantage of GWAS studies is the unbiased inclusion of the entire genome, thus allowing for the identification of novel loci of interest and the development of original hypotheses to be tested. Once statistically significant associations have been established with one or more loci, they can be confirmed with data from other GWAS populations or through determination of the causal relationship by experimental means (Sekar et al., 2016).

While GWAS studies have produced important discoveries, their experimental designs impose hurdles that can present significant challenges. The first hurdle is population size. GWAS studies are, by definition, very large, requiring scans of large numbers of SNPs. In order to reach statistical significance, sample sizes in the thousands are required. For common diseases, very large sample sizes are attainable; however, for rarer diseases, sample size becomes problematic and is prohibitive in some cases. The second hurdle is that GWAS studies focus on identifying one or sometimes a small group of mutations in loci that are associated with a disease phenotype. It is then left to interpretation what, if any, relationship there is between these loci and the condition being studied. The third hurdle is that GWAS studies largely operate on the common disease– common variant hypothesis. This hypothesis states that the genetic contributions to the susceptibility of common diseases are attributable to a small number of variants present in more than 1–5% of the population (Manolio, 2010). While this hypothesis has proven useful for common diseases, it is not as applicable to rare diseases and more complex diseases whose mutations may not reach the 1% threshold. The fourth and final hurdle is that GWAS is not well suited to diseases in which the genetic impact on a disease state is largely dependent upon epistatic interactions. Epistatic interactions are described as the interaction between two loci whereby the phenotype of one locus depends on the genotype of the second locus (Carlborg and Haley, 2004). For quantitative traits, epistasis describes a situation where the sum of the single locus effects cannot be used to predict the phenotype of a particular genotype (Carlborg and Haley, 2004). When considering the effect of genetic interaction at the pathway level, one or two small effect mutations may not cause a noticeable phenotype, but 10 or 20 mutations could combine to create a significant "load" on the pathway and cause deterioration in functionality, leading to a diseased state. Epistatic interactions could also occur within a single gene where one SNP functionally affects another. In this model, neither mutation alone causes an effect, but together they combine to act synergistically to cause an effect greater than would be predicted by an additive model.

#### AN ALTERNATIVE APPROACH: THE PATHWAY GENETIC LOAD (PGL) SCORE

#### Description of the Pathway Genetic Load Score

The GWAS method has identified a number of genetic associations that are important to numerous complex diseases. Within these predictions, often a large portion of the heritability remains to be uncovered. Understanding this "missing heritability" has become one of the major goals of clinical genetic research. Some of the tactics that have been proposed include combining linkage and association analysis (Ott et al., 2011), correcting an overestimation in heritability (Zuk et al., 2012), and considering gene–gene interactions (Liu et al., 2012).

A particularly intriguing approach to resolving the missing heritability problem was recently presented by Huebinger et al., who used a PGL risk score (Huebinger et al., 2010) to predict the outcome of patients exposed to severe burn trauma. The PGL risk score was defined as the "sum total of mutant alleles present at risk loci within a common biological pathway." The PGL risk score is calculated using a number of assumptions that require certain concessions to be made in order to be both useful and accurate. Others have used similar pathway analysis methods to study genetic risks for coronary infarction (Yiannakouris et al., 2006; Trichopoulou et al., 2008) and Parkinson's disease (Lesnick et al., 2007). The PGL method has the potential to offer an alternative method to GWAS for identifying relevant genetic contributors, especially for lower incidence diseases.

# APPLYING THE PGL RISK SCORE TO BILIRUBIN ENCEPHALOPATHY

#### Description of Bilirubin Encephalopathy

Kernicterus or CBE is a rare disease in affluent countries because of preventive measures. However, neonatal jaundice is very common, occurring in 60–80% of full term (≥37 weeks' gestational age) otherwise healthy newborns. About 7% of jaundiced infants receive phototherapy (Bhutani et al., 2004a,b, 2013; Keren et al., 2009). Classical kernicterus is a well-described clinical tetrad of (i) abnormal motor control, movements, and muscle tone, (ii) an auditory processing disturbance with or without hearing loss, (iii) oculomotor impairments, especially impairment of upward vertical gaze, and (iv) dysplasia of the enamel of deciduous (baby) teeth, an abnormality leading to pitting, flaking, and chipping of the enamel (Shapiro et al., 2006). In addition to classical kernicterus, subclasses of kernicterus are characterized as auditory predominant, motor predominant, and subtle kernicterus (BIND) (Shapiro, 2010).

While efforts to prevent dangerous hyperbilirubinemia have been successful in the developed world, a high incidence of glucose-6-phosphate dehydrogenase deficiency, delays in seeking care, lack of TSB screening resources, and ineffective treatments in low and middle income countries ensure that bilirubin encephalopathy remains a serious world health problem (Olusanya et al., 2014; Slusher et al., 2015). Ogunlesi et al. (2007) reported a 2.7% incidence of bilirubin encephalopathy among neonates admitted to two hospitals in Nigeria between 2002 and 2005. This high prevalence of kernicterus, in conjunction with new reports of the more subtle predominantly auditory and predominantly motor forms of kernicterus (Shapiro and Popelka, 2011), underscores the need to better understand why hyperbilirubinemia causes permanent brain damage in some infants and not in others. The relative rarity of kernicterus in the West and, to a lesser extent, milder forms of bilirubin encephalopathy, makes it exceedingly difficult to perform traditional genetic analysis through large population GWAS studies. In addition, the apparent wide range of susceptibility to a common bilirubin exposure points toward polygenetic determination of risk for neurotoxicity. The current understanding of the genetics of bilirubin encephalopathy (described below) also points toward a polygenetic cause for increased risks to bilirubin neurotoxicity.

Together, these factors indicate a need for an alternative method, such as the PGL risk score, to understand genetic factors contributing to risk for CBE. The PGL method is a good fit for bilirubin encephalopathy because bilirubin encephalopathy results from a one-time insult, similar to the severe burns examined in Huebinger et al.'s study. For bilirubin encephalopathy that insult is in the form of moderate to extreme hyperbilirubinemia at birth. Also, similar to the total body surface area measure for a serious burn, the body's response to bilirubin toxicity is variable and not well explained by its key diagnostic test, peak TSB levels.

While significant progress has been made in understanding the various causes of neonatal hyperbilirubinemia, few studies have been undertaken to understand the mechanism of bilirubin toxicity. Multiple cellular and molecular cascades likely underlie bilirubin-induced neuronal injury, including plasma membrane perturbations, excitotoxicity, neuroinflammation, oxidative stress, and cell cycle arrest. It is likely that variable effectiveness of the cellular response to bilirubin exposure contributes to the range of observed clinical outcomes. To explain part of this variability, we hypothesize that a combination of SNPs in key pathways are responsible for creating a vulnerable state to bilirubin neurotoxicity. The development of this vulnerable state likely occurs through a combination of increased CNS bilirubin levels and impaired cellular response to bilirubin.

On the basis of the current understanding of bilirubin encephalopathy, we suspect that the genetic impact is the result of the combined effect of mutations with relatively small effects that, if present alone, would otherwise be clinically benign. Alternative hypotheses that could also explain the TSB's poor predictive value for kernicterus include disrupted albumin/bilirubin binding capacity, reductions in available albumin, and use of TSB levels as the sole diagnostic test for bilirubin exposure. A number of studies have been published that examine the issue of TSB's reliability in both predicting and improving outcomes. These studies have compared the effectiveness of alternatives to TSB such as albumin/bilirubin binding capacity (Kapitulnik et al., 1974), unbound (free) bilirubin levels (Cashore and Oh, 1982; Ahlfors et al., 2009; Morioka et al., 2015), and bilirubin-albumin ratios either in place of or in addition to TSB testing (Ahlfors, 1994; Amin et al., 2001, 2016).

While Nakamura et al. showed over 30 years ago that free bilirubin levels predict neurotoxicity better than TSB levels do (Nakamura et al., 1985), technical issues, reliability, and lack of availability of these tests have called their usefulness into question (McDonagh and Maisels, 2006; Hulzebos and Dijk, 2014). The use of the bilirubin/albumin ratio has also been shown to be as strong a predictor of neurotoxicity as TSB (Iskander et al., 2014) but has not been shown to offer an advantage in predicting or improving outcomes over TSB alone (Iskander et al., 2014) or when used in addition to TSB (Hulzebos et al., 2014).

## Known Genetics of Bilirubin Encephalopathy

Bilirubin is largely a breakdown product of hemoglobin. Bilirubin rises after birth in all mammals, and in humans usually peaks at 3–5 days after birth and then falls to very low values through the rest of life (Kivlahan and James, 1984; Maisels et al., 2014). It is bound in the blood to proteins, especially albumin. Very little bilirubin normally reaches brain tissue. There is evidence that at low levels bilirubin is beneficial, acting as a natural antioxidant (Adin et al., 2005; Bakrania et al., 2016), but very high levels can exceed the buffering capacity of the blood and allow bilirubin to move into the brain, causing a unique pattern of brain injury (Shapiro et al., 2006; Shapiro, 2010).

Hyperbilirubinemia results from the excessive breakdown of red blood cells (RBC) (hemolysis) or impaired elimination of bilirubin. Most genetic abnormalities linked to the presentation of bilirubin encephalopathy are related to production and elimination. Hemolysis may occur as a result of immune mechanisms (e.g., Rh disease, ABO blood group incompatibilities), increased fragility of RBC [Glucose-6 phosphate dehydrogenase (G6PD) deficiency], or from bleeding (e.g., cephalohematoma). Decreased elimination is enhanced in infants with genetic variations in the enzyme or the promotor regions of UDP-glucuronosyltransferase 1A1 (UGT1A1), (Watchko and Tiribelli, 2013). Other factors that contribute to hyperbilirubinemia include decreased oral intake and stooling. It is believed that hyperbilirubinemia due to decreased oral intake may account for a substantial number of readmissions of newborns to hospitals after discharge (Seidman et al., 1995; Farhat and Rajab, 2011; Maisels et al., 2014).

## Outcome Variability at the Same or Very Similar TSB Levels

Possible sources of outcome variability include factors such as the source of hyperbilirubinemia, the gestational age of the infant, the duration of exposure, the kinetics of bilirubin's movement between miscible compartments of the body (e.g., blood, tissue, CNS, extracellular, and intracellular), the type of treatment administered and other known risk factors such as acidosis or sepsis (Shapiro, 2010; Hulzebos and Dijk, 2014). Furthermore, a significant genetic contribution to the susceptibility of an individual to neurological damage due to hyperbilirubinemia may be related to the disruption of mitochondria, which may in turn affect cellular and molecular cascades including plasma membrane perturbations, excitotoxicity, neuroinflammation, oxidative stress, and cell cycle arrest referred to previously (Rodrigues et al., 2002a,b; Vaz et al., 2010; Brites, 2011; Barateiro et al., 2012).

Clinical observations have shown that in cases where TSB reaches extremely high levels (>40 mg/dL or >680µM<sup>1</sup> ), the likelihood of developing at least some form of kernicterus approaches 100%, although exceptions to this generalization have been published (Newman et al., 2003, 2006). These clinical observations likely indicate that for most neonates, there is a

<sup>1</sup>Conversion from international units of bilirubin (µM) to mg/dL (a.k.a. mg%) is 17.1, i.e., 1 mg/dL = 17.1µM.

maximum level of bilirubin or "tipping point" where the body's natural defense mechanisms for clearing unbound bilirubin from susceptible neurons are overwhelmed and deposition occurs. However, because kernicterus has been known to develop at lower TSB levels, it is clear that the maximum level of bilirubin exposure varies from person to person (Ritter et al., 1982; Odutolu and Emmerson, 2013). Even allowing for external factors and the inexact nature of TSB as a measure of exposure to bilirubin in brain tissue, there is still a strong likelihood that genetic variability contributes to the overall susceptibility of a child to bilirubin neurotoxicity and subsequent chronic adverse neurodevelopmental sequelae.

## Selecting a "Pathway" for Bilirubin Encephalopathy

Here we propose using the PGL method in order to determine the extent to which a multifactorial genetic response contributes to the variable susceptibility to bilirubin toxicity and the resultant variability in disease outcome. Due to the lack of a CBE GWAS to guide our pathway selection, we propose creating three custom "pathways" of genes that are organized into "tiers" based on the mechanism of physiological contributions and the relative strength of evidence that these genes contribute to bilirubin susceptibility and sensitivity. Each of these three gene tiers will be analyzed individually and independently using the PGL method.

The PGL method differs from traditional association analysis in a number of important ways. First, the PGL score uses a biased approach by leveraging what is already known about a disease to focus on a specific set of genes (pathway) and SNPs. Second, unlike GWAS and linkage analysis, the PGL score method involves the calculation of a risk score. Third, the method can be performed using a smaller sample size than in GWAS studies because the number of variables is dependent on candidate pathway size, thus reducing the required sample size to achieve adequate statistical power. Finally, the PGL method allows detection of combinatorial epistatic interactions that would otherwise have been missed by both GWAS and linkage analysis.

The decision to use a biased approach comes with the understanding that the available knowledge will very likely present only a partial picture of the entire genetic heritability. However, the hope is that if enough of the biological background is known, then the analysis will prove to be useful and actionable. By accepting a higher rate of false negatives, we will find it is then possible to drastically reduce the number of genes and thus variants being surveyed. Reducing the number of SNPs being investigated allows for a corresponding reduction in sample size, opening up analysis to rarer diseases. Huebinger et al. used their knowledge of injury and infection to select the Tolllike receptor signaling pathway to assess whether an increased PGL risk score within this pathway could predict the outcome following burn trauma (Huebinger et al., 2010). Similarly, using the PGL method, we will perform a targeted genetic analysis on a small number of samples and will determine whether a common mutational "load" exists in our affected individuals that points toward the development of a state of bilirubin hypersensitivity.

#### Tier 1: Known Genetic Risk Factors for Increased Hyperbilirubinemia: Genes for Increased Production or Decreased Elimination of Bilirubin

The tier 1 "pathway" is composed of genes known as contributing risk factors for hyperbilirubinemia and bilirubin encephalopathy (i.e., classically recognized) (Watchko and Lin, 2010, 2012; Watchko, 2013). This pathway includes all of the known mechanisms contributing to the production and elimination of bilirubin.

The production pathway includes genes that predispose to hemolysis or reduce RBC lifespan, leading to increased bilirubin production in neonates. G6PD and Pyruvate Kinase (PK) mutations can cause defects in RBC metabolism (Watchko and Lin, 2012). Several conditions lead to reduced RBC lifespan such as hereditary spherocytosis, elliptocytosis, stomatocytosis, and infantile pyknocytosis (Delaunay, 2007; Kraus et al., 2010; Kaplan et al., 2014).

Genes that affect elimination of bilirubin include allelic variation in the regulatory region of the UGT1A1 gene (affecting bilirubin conjugation in the liver, Gilbert's syndrome) and the solute transporter SLCO1B1 that facilitates UCB uptake and influences the ability to clear bilirubin. Poorly functioning variants of either gene can contribute to hyperbilirubinemia (Johnson et al., 2009; Watchko and Lin, 2010). We consider these genes to be the first tier of primary candidate genes that are known to influence bilirubin levels. It is reasonable to presume that a combination of subtle, seemingly benign mutations in these genes could result in increasing bilirubin to toxic levels. The complete tier 1 list can be seen in (**Supplemental Table 1**).

#### Tier 2: In vitro—Genetic Responses to Bilirubin in Neuronal and Glial Cell Cultures

The tier 2 "pathway" is composed of genes identified in recent studies as genes whose expression levels are modulated by bilirubin exposure. These genes were identified by in vitro experiments in human SH-SY5Y neuroblastoma cells (Calligaris et al., 2009). In these experiments, the cells were exposed to UCB for 24 h and transcriptome changes were examined by high-density oligonucleotide microarrays. Selected genes were then validated by RT-PCR. These genes represent neuronal genes that had altered expression in response to bilirubin exposure. In addition, we have also included those genes that are related to the immunoreactive response shown to occur in primary rodent neuronal and glial cultures exposed to UCB (Brites, 2012). Primary glial cell lines including rat astrocytes (Fernandes et al., 2004, 2006, 2007b) rat microglia (Gordo et al., 2006; Silva et al., 2010), and rat oligodendrocytes (Barateiro et al., 2012, 2013, 2014) have been shown to be sensitive to exposure of UCB leading to cell death in all cases. In addition, rat astrocytes and microglia, but not oligodendrocytes release pro-inflammatory cytokines upon exposure to UCB (Gordo et al., 2006; Fernandes et al., 2007a; Barateiro et al., 2012). It is thought that this inflammatory response is a crucial element in the UCB response and plays a major role in the cell death (Vaz et al., 2011; Silva et al., 2012). Unlike the list of genes in tier 1, this list represents a response to bilirubin exposure rather than the production of a hyperbilirubinemic state. We hypothesize that this list contains a

core network of genes that are important at the cellular level to respond to toxic levels of bilirubin. The complete tier 2 list can be seen in **Supplemental Table 2**.

#### Tier 3: In silico—Genes Identified by Bioinformatics Analysis

We hypothesize that gene expression differences in susceptible vs. resistant regions of the brain may provide some insight into additional secondary mediators of bilirubin neurotoxicity. The tier 3 "pathway" is composed of genes identified to be differentially expressed in regions of the brain that are susceptible to bilirubin compared to similar regions in close proximity that are resistant. We chose two pairs of susceptible vs. resistant brain regions for comparison based on clinical knowledge (Shapiro, 2005, 2010) and animal studies (Gazzin et al., 2012): (1) the inferior colliculus (IC) is susceptible to bilirubin toxicity compared to, superior colliculus (SC), the resistant control, and (2) the globus pallidus (GP) vs. putamen (PU). The IC is in part responsible for the auditory abnormalities and the GP for the motor abnormalities in kernicterus and BIND. Furthermore, in addition to showing a marked difference in response to bilirubin toxicity, these regions of the brain are in close proximity to each other, thus minimizing the potential for variability in blood flow and other factors.

The Allen Brain Atlas is a collection of freely available online resources integrating extensive gene expression and neuroanatomical data from developing and adult human brains (Website: © 2015 Allen Institute for Brain Science. Allen Human Brain Atlas [Internet]. Available from: http://human.brain-map. org) (Hawrylycz et al., 2012). The atlas allows exploration of the expression of known genes in carefully defined anatomical regions of the human brain at high resolution. We used the brain atlas to compare gene expression between regions that are susceptible to bilirubin toxicity and those that are resistant.

For the creation of Tier 3 A and B we identified those genes whose expression was significantly different (p < 0.05) between the IC (susceptible) compared to SC (resistant) brain regions in normal adult brains using the resources of the Allen Brain Atlas. From these genes we selected 100 genes with the largest increase in expression level in the IC vs. SC (3A) and 100 genes with the largest increase in expression level in the SC vs. IC (3B), as measured by fold change, for a total of 200 genes. For the creation of Tier 3 C and D we identified those genes whose expression was significantly different (p < 0.05) between GP (susceptible) and PU (resistant) brain regions. And as with 3A and 3B we selected 100 genes with the largest increase in expression level in the GP vs. PU, (3C) and 100 genes with the largest increase in expression level in the PU vs. GP (3D), as measured by fold change, for a total of 200 genes.

We chose to limit our gene lists to 200 genes per comparison because we believe that this strategy provides a fair representation of those genes that show the most dramatic change between regions and thus potentially reflect the genetic components underlying the physiological differences that contribute to region specific bilirubin toxicity resistance or susceptibility. While we recognize that this strategy will not encompass all the genes responsible for the varied response to bilirubin we believe that this abbreviated strategy will allow for the identification of major contributors and their associated pathways thus opening up new avenues for future research. We hypothesize that analyzing the expression of the genes in these differentially affected regions of the brain will reveal susceptible and protected genetic profiles within the brain that could then be extrapolated to better understand the variable susceptibility of individuals to bilirubin toxicity. The complete tier 3 list can be seen in **Supplemental Table 3**.

# Selecting Key SNPs for PGL Risk Score Calculation and Analysis

After the pathway of interest is selected, the next step is to select the candidate SNPs from within the pathway. For this crucial step, Huebinger et al. chose to create their candidate SNP list from variants that had previously published associations with burn trauma. Once these SNPs are determined, the PGL risk score can be calculated by summing the numerical scores for genotypes across all SNPs that are homozygous protective (assigned a score of "0"), heterozygous (score of "1") and homozygous detrimental (score of "2"). The initial calculation results in the raw risk score where each SNP is treated equally. The risk score is then weighted to better represent the impact of each of the SNPs. For this step Huebinger et al. chose to use the adjusted odds ratio from previously reported studies for each SNP between each individual locus and either death or sepsis. For loci without previously published data, the value was left unweighted. The data could then be analyzed further by Mann-Whitney U Test to investigate whether the PGL risk score alone can predict developing sepsis. In addition, the multivariate logistic regression test could be used to investigate the simultaneous effect of multiple variables as risk factors for various outcomes. A synopsis of the PGL workflow can be seen in **Figure 1**.

For their analysis Huebinger et al. chose to adjust for age, gender, race, total body surface area burned, and inhalation injury. They were able to show that each unit increase in weighted PGL risk scores was associated with increased probability of both complicated sepsis and death. By analyzing data from only 155 burn trauma patients and using only 6 SNPs at 6 different loci, they showed that a larger PGL risk score indicated increased odds of both sepsis and mortality in response to burn trauma. This work, along with other similar pathway load work (Lesnick et al., 2007), demonstrates that by using a targeted approach it is possible to show that a genetic load placed on a crucial pathway can have predictive capabilities even with a relatively small sample size.

To develop a SNP set for each tier, we will collect whole exome sequence data from patients with moderate to high TSB levels (10–25 mg/dL and >25 mg/dL respectively) both with and without evidence of classic auditory or motor dysfunction. We will identify positive "hits" through a pipeline utilizing the Partek Genomics Suite and Flow (Partek Inc., St. Louis, MO, USA). Causal variants in this study will then be identified through the use of QIAGEN's Ingenuity <sup>R</sup> Variant AnalysisTM software (www.qiagen.com/ingenuity) from QIAGEN Redwood

developing and testing a PGL risk score according to Huebinger et al. (2010). The raw and weighted scores are analyzed separately to assess the impact of relevant disease association data on the PGL risk score calculation. Abbreviations: SNP, Single Nucleotide Polymorphism; PGL, Pathway Genetic Load; aOR, adjusted odds ratio gathered from a relevant GWAS study.

City. Ingenuity <sup>R</sup> Variant AnalysisTM allows for multiple types of SNP filtering to create a list of screened causal variants that are present in a significant number of the case group vs. controls. The causal variant filter is based on the American College of Medical Genetics and Genomics (ACMG) categorization guidelines (Richards et al., 2015). Alternatively, or in combination with ACMG categorization, it is possible to select variants that cause a gain of function or a loss of function based on information in the Ingenuity <sup>R</sup> Knowledge Base. The Ingenuity <sup>R</sup> Knowledge Base is a "repository of expertly curated biological interactions and functional annotations created from millions of individually modeled relationships between proteins, genes, complexes, cells, tissues, drugs, and diseases" manually created from the literature (http://www.ingenuity.com/science/knowledge-base). The causal variant list can also be filtered to include only those genes present in each of the gene set tiers.

Once the key SNPs predicted to be deleterious are identified, the PGL risk score can be calculated. The raw score will be calculated from the individual genotypes as described above as homozygous protective = 0, heterozygous = 1 and compound heterozygote/homozygous deleterious = 2. The raw score will then be summed and the adjusted odds ratio, confidence interval, and p-value will be calculated. Multivariate logistic regression will then be used to evaluate the simultaneous effects of multiple variables as risk factors for the development of one of three outcomes. Possible outcomes include predominantly auditory dysfunction, predominantly motor dysfunction, and auditory plus motor dysfunction (classic kernicterus) (Shapiro, 2010). Outcomes will be further analyzed with adjustments for TSB, albumin, gender, race, sepsis, and gestational age at birth.

For weighted analysis, the SNPs will be weighted based on the Sorting Intolerant from Tolerant (SIFT) score (Ng, 2003) or the Combined Annotation-Dependent Deletion (CADD) score (Kircher et al., 2014). By utilizing the ACMG score for initial selection and the SIFT or CADD scores for weighting, we are able to evaluate the direct effect of a SNP, a crucial difference from the GWAS based method. A synopsis of the proposed PGL workflow to be used for bilirubin encephalopathy can be seen in **Figure 2**.

### SHORTCOMINGS OF THE PATHWAY GENETIC LOAD SCORE

While the PGL method presented has been successful at producing significant and meaningful results, there are drawbacks to using PGL risk score analysis. These drawbacks include the previously described risk for false negatives, the possibility that major contributing factors are outside the pathway chosen to be studied, and that considerable cross-talk is occurring between overlapping pathways, leading to confusing and difficult to confirm results. In addition, the selections of the target pathway and the set of SNPs are based on the information available at the time of the study which may not be complete enough to make accurate selections. To address this issue, we have used three different methods to develop our tiers aimed at maximizing the currently available knowledge base for CBE. Finally, the weighting method described by Huebinger et al. utilizes an adjusted odds ratio based on previously performed GWAS studies (Huebinger et al., 2010). This type of GWAS data may not be available for rare diseases. To address this issue, we have proposed a method that will weigh the PGL risk score of each SNP using pathogenicity scores that are available from various bioinformatics sources as an attempt to investigate a direct impact of each SNP on the functionality of the pathway of interest.

## MOVING FORWARD

To test the method described above, we plan on analyzing whole exome sequence data from patients exposed to moderate (15–25 mg/dL) and high (>25 mg/dL) peak TSB levels at birth. We acknowledge that it would be preferable to also investigate the free unbound bilirubin levels for these patients, but those data are rarely available. Because of the rarity of kernicterus in developed countries, 1.15:100,000 live births in an average of three studies from Denmark (Ebbesen, 2000; Ebbesen et al., 2005; Bjerre et al., 2008) and 0.93:100,000 live births in the United Kingdom and Ireland (Manning et al., 2007), it will likely be necessary to obtain biological samples from international locations such as Africa and southeast Asia where kernicterus, unfortunately, continues to be a significant problem. The inclusion of such disparate populations has the potential to add additional complicating factors, including differences in genetic background and differences in common treatment methods. We will address these issues by comparing patients of similar ethnic backgrounds whenever possible. The collection of biological material for this project has received institutional review board approval and is ongoing. Collaborating with other laboratories internationally is also in progress.

# SUMMARY

We present a method for addressing the problems associated with identifying the genetic factors that influence both level of bilirubin and susceptibility to bilirubin encephalopathy (kernicterus). Because rare diseases by definition affect small numbers of people, it is impractical to use unbiased populationbased GWAS studies. In this paper, we present an evidencebased candidate pathway approach in the form of the PGL risk score that will reduce the number of genes and variants investigated and thus reduce the need for overly large sample sizes. We propose new alternatives for selecting pathways and variants through the creation of custom gene lists and use of functional scoring methods. The logistic regression analysis proposed in Huebinger et al. will be used to determine whether the PGL score for patients with moderate and high TSB levels could successfully predict auditory or motor deficit outcomes. As this method is tested and refined, we believe it will lead to the creation of new hypotheses that will aid the investigation of the molecular mechanisms of bilirubin toxicity in newborns. This work could also potentially be used to develop a screening process for jaundiced infants that could inform physicians about whether more aggressive treatment is needed in infants with moderate TSB levels. Finally, this method could be adapted to other rare or complex diseases to address the challenge of missing heritability that exists despite GWAS.

#### AUTHOR CONTRIBUTIONS

SR drafted the manuscript. SR, DB, JL, SG, CT, JW, RW, and SS contributed to the conception of the work and provided critical revision for important intellectual content.

#### FUNDING

This work was supported by institutional startup funds from Children's Mercy Hospital.

#### ACKNOWLEDGMENTS

The authors would like to thank Dan Heruth for helpful suggestions in revising the manuscript. The authors would also like to thank Nataliya Kibiryeva for her help with the analysis development. Finally, the authors would like to thank the

#### REFERENCES


Children's Mercy Hospital medical writing center for reviewing and editing the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnins. 2016.00376

Supplemental Table 1 | Tier 1: Known genetic risk factors for increased hyperbilirubinemia. This list represents the classical list of genes associated with hyperbilirubinemia, jaundice and kernicterus. A reference for each gene is listed in the final column.

Supplemental Table 2 | Tier 2: In vitro—genetic responses to bilirubin in neuronal and glial cell cultures. This list represents the genes reported to be affected in either expression level or protein response in cultured cells exposed. Cell types include human neuronal (SH-SY5Y cells), primary rat astrocytes, primary rat oligodendrocytes and primary rat microglia. A reference for each gene is listed in the final column.

Supplemental Table 3 | Tier 3: In silico—comparative analysis of gene expression in susceptible and resistant brain regions in data from the Allen Brain Atlas. Gene expression data from the Allen Brain Atlas (*n* = 6) was collected from two brain region comparisons. 3A shows significantly different (*p* < 0.05) genes with the 100 largest fold change differences that were upregulated in the IC vs. SC. 3B shows significantly different (*p* < 0.05) genes with the 100 largest fold change differences that were upregulated in the SC vs. IC. 3C shows significantly different (*p* < 0.05) genes with the 100 largest fold change differences that were upregulated in the GP vs. PU. 3D shows significantly different (*p* < 0.05) genes with the 100 largest fold change differences that were upregulated in the PU vs. GP. This list was last updated on 7/14/16.


outcome: a randomized controlled trial–BARTrial. PLoS ONE 9:e99466. doi: 10.1371/journal.pone.0099466


filtered sunlight in African Neonates. N. Engl. J. Med. 373, 1115–1124. doi: 10.1056/NEJMoa1501074


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Riordan, Bittel, Le Pichon, Gazzin, Tiribelli, Watchko, Wennberg and Shapiro. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Association of Cell Adhesion Molecules Contactin-6 and Latrophilin-1 Regulates Neuronal Apoptosis

Amila Zuko<sup>1</sup>† , Asami Oguro-Ando<sup>1</sup>† , Harm Post2,3, Renske L. R. E. Taggenbrock<sup>1</sup> , Roland E. van Dijk<sup>1</sup> , A. F. Maarten Altelaar2,3, Albert J. R. Heck2,3 , Alexander G. Petrenko<sup>4</sup> , Bert van der Zwaag<sup>5</sup> , Yasushi Shimoda<sup>6</sup> , R. Jeroen Pasterkamp<sup>1</sup> and J. Peter H. Burbach<sup>1</sup> \*

<sup>1</sup> Department of Translational Neuroscience, Brain Center Rudolf Magnus, University Medical Center Utrecht, Utrecht, Netherlands, <sup>2</sup> Biomolecular Mass Spectrometry and Proteomics, Bijvoet Center for Biomolecular Research and Utrecht Institute for Pharmaceutical Sciences, Utrecht University, Utrecht, Netherlands, <sup>3</sup> Netherlands Proteomics Centre, Utrecht, Netherlands, <sup>4</sup> Laboratory of Receptor Cell Biology, Shemyakin-Ovchinnikov Institute of Bioorganic Chemistry Russian Academy of Sciences, Moscow, Russia, <sup>5</sup> Department of Genetics, University Medical Center Utrecht, Utrecht, Netherlands, <sup>6</sup> Department of Bioengineering, Nagaoka University of Technology, Nagaoka, Japan

#### Edited by:

Daniela Tropea, Trinity College, Dublin, Ireland

#### Reviewed by:

Michael J. Schmeisser, University of Ulm, Germany Subhrangshu Guhathakurta, University of Central Florida, USA

#### \*Correspondence: J. Peter H. Burbach

j.p.h.burbach@umcutrecht.nl

#### †Present address:

Amila Zuko, RIKEN Brain Science Institute, Wako, Saitama, Japan Asami Oguro-Ando, Wellcome Wolfson Centre for Medical Research, University of Exeter Medical School, Exeter, UK

> Received: 05 October 2016 Accepted: 28 November 2016 Published: 15 December 2016

#### Citation:

Zuko A, Oguro-Ando A, Post H, Taggenbrock RLRE, van Dijk RE, Altelaar AFM, Heck AJR, Petrenko AG, van der Zwaag B, Shimoda Y, Pasterkamp RJ and Burbach JPH (2016) Association of Cell Adhesion Molecules Contactin-6 and Latrophilin-1 Regulates Neuronal Apoptosis. Front. Mol. Neurosci. 9:143. doi: 10.3389/fnmol.2016.00143 In view of important neurobiological functions of the cell adhesion molecule contactin-6 (Cntn6) that have emerged from studies on null-mutant mice and autism spectrum disorders patients, we set out to examine pathways underlying functions of Cntn6 using a proteomics approach. We identified the cell adhesion GPCR latrophilin-1 (Lphn1, a.k.a. CIRL1/CL, ADGRL1) as a binding partner for Cntn6 forming together a heteromeric cis-complex. Lphn1 expression in cultured neurons caused reduction in neurite outgrowth and increase in apoptosis, which was rescued by coexpression of Cntn6. In cultured neurons derived from Cntn6−/<sup>−</sup> mice, Lphn1 knockdown reduced apoptosis, suggesting that the observed apoptosis was Lphn1-dependent. In line with these data, the number of apoptotic cells was increased in the cortex of Cntn6−/<sup>−</sup> mice compared to wild-type littermate controls. These results show that Cntn6 can modulate the activity of Lphn1 by direct binding and suggests that Cntn6 may prevent apoptosis thereby impinging on neurodevelopment.

Keywords: autism, ASD, neurodevelopmental disorders, cell adhesion molecules, neuronal outgrowth, Cntn6, Lphn1, apoptosis

# INTRODUCTION

The six members of the contactin family of immunoglobulin cell adhesion molecules (IgCAMs) play diverse roles in the nervous system (Shimoda and Watanabe, 2009; Stoeckli, 2010; Zuko et al., 2013). Contactin-1 (Cntn, a.k.a. F3) and contactin-2 (Cntn2 a.k.a. Tag-1) have been well characterized for their specialized functions in neuron–glia interaction, specifically in the paranode and juxtapararanode of the nodes of Ranvier (Peles and Salzer, 2000; Scherer and Arroyo, 2002; Poliak and Peles, 2003). The contactins act through homophilic and heterophilic interactions with various classes of proteins and form codes for specified connectivity (Stoeckli, 2010). For example, all contactin members, except Cntn6 (a.k.a. NB-3) have been proposed to participate in an IgCAM code to guide lamina-specific neurite targeting (Yamagata and Sanes, 2012). Increasing evidence

suggests that Cntn6 is involved in brain development, since mice deficient for Cntn6 display a delay in the development of the corticospinal tract, a misorientation of apical dendrites in layer V of the visual cortex, and an increase in neuronal cell death during development (Ye et al., 2008; Sakurai et al., 2009; Pinto et al., 2010; Huang et al., 2011b). A significant reduction in glutamatergic synapses was found in the hippocampus and in the cerebellum of Cntn6 null-mutants (Sakurai et al., 2009, 2010), implicating Cntn6 in the regulation of synaptogenesis. In addition, behavioral studies have shown that Cntn6-deficient mice display impaired motor coordination (Takeda et al., 2003). These data indicate that Cntn6 plays a pivotal role in brain development.

A role of Cntn6 in brain development is further emphasized by genetic findings of copy number variations (CNVs) in the human CNTN6 gene in rare cases with autism spectrum disorder (ASD) (Pinto et al., 2010; van Daalen et al., 2011; Hu et al., 2015). Furthermore, point mutations and shared CNVs between the CNTN4 and CNTN6 genes have also been implicated in the pathogenesis of bipolar disorder and anorexia nervosa (Pinto et al., 2010; Kerner et al., 2011; van Daalen et al., 2011; Wang et al., 2011). Finally, deletion of the tip of the short arm of chromosome 3, that harbors the CNTN6, CNTN4 and CHL1 genes, causes a mental retardation syndrome with ASD comorbidity, called 3pdeletion syndrome (Shuib et al., 2009). This further underscores the importance of CNTN6 for appropriate neural development. However, it is still unknown what the molecular pathways are through which CNTN6 acts and how the loss of function of this protein contributes to disease.

The mode of action of Cntn1 and Cntn2, the best studied members of the contactin family involves the formation of multiple homo- and heterodimers in both cis and transconfigurations that are essential for the structure of the paranode and juxtaparanode (Peles et al., 1997; Rios et al., 2000; Poliak et al., 2003). Cntn4 and Cntn5 also appear to be engaged in cis- and trans-interactions with their homologs or with other proteins (Traka et al., 2003; Cui et al., 2004; Osterfield et al., 2008; Ye et al., 2008, 2011; Bouyain and Watkins, 2010; Shimoda et al., 2012; Yamagata and Sanes, 2012; Ashrafi et al., 2014; Osterhout et al., 2015). Cntn6 is known to complex with other membrane proteins as well, including Ptpra, Ptprg, PTPσ, Notch, and Chl1 (Cui et al., 2004; Hu et al., 2006; Ye et al., 2008; Bouyain and Watkins, 2010; Zuko et al., 2011; Huang et al., 2016) Thus, interaction with other protein partners appears as a common theme in the mode of action of contactins. Therefore, we further examined protein networks in which Cntn6 participates. We demonstrate here that Cntn6 binds to the cell adhesion G-protein-coupled receptor (adhesion GPCR) latrophilin-1 (Lphn1, ADGRL1, a.k.a. CIRL1/CL) forming a silenced cis-complex. Loss of Cntn6 results in inhibition of neurite outgrowth and an increased neuronal cell death due to unoccupied Lphn1. This study indicates that Cntn6 serves as an endogenous ligand for Lphn1 thereby controlling apoptosis. This conclusion was supported by in vivo analyses of the Cntn6−/<sup>−</sup> mouse brain displaying increased apoptosis which links Cntn6 to one of the pathogenic pathways of autism (Wei et al., 2014).

# MATERIALS AND METHODS

## Animals and Tissue Treatment

B57BL/6 and Cntn6−/<sup>−</sup> mice were obtained from Charles River and Nagaoka University (Takeda et al., 2003), respectively. Mice were maintained on a 12-h light/dark cycle with ad libitum food and water in an animal facility at Brain Center Rudolf Magnus, Utrecht University. For immunohistochemistry, P14 mouse pups were anesthetized with an overdose of sodium pentobarbital (19.4 µl/gr) and were perfused intracardially with 0.9% saline, followed by 4% PFA in PBS, pH 7.5. Brains were post fixed in 4% PFA before transferred to 30% sucrose for cryopreservation. Tissue was sectioned at 40 µm sections and free-floating sections were stored in PBS with 0.02% sodium azide until immunohistochemistry was performed. For in situ hybridization, P7 mouse pups were killed by decapitation and their brains were quickly dissected and flash-frozen in 2 methylbutane. Brains were sliced into 16 µm sections using a cryostat and mounted onto Superfrost slides (VWR).

# Cell Adhesion Assay

Cell adhesion assays were performed with HEK293 cells as previously described Ko et al. (2009). HEK293 cells were cotransfected either with pCMV-EGFP-N1 or pCAG-DsRed and full-length pcDNA3.1-Cntn6, pcDNA3.1-Lphn1, pCAG-HA-Nlgn1 and pCAG-HA-Nrxn1β <sup>−</sup> (latter two were gifts from Dr. Scheiffele) expression constructs. After 48 h, the cells were detached using 1 mM EDTA in PBS, pH 7.4, and centrifuged at 1000 rpm for 5 min. The pellets were resuspended in suspension medium (10% HIFCS, 50 mM Hepes-NaOH, pH 7.4, 10 mM CaCl<sup>2</sup> and 10 mM MgCl2) and combined to a total of 5x10<sup>6</sup> (1:1) in 0.3 ml total volume of 0.5 ml eppendorf tubes. The cell mixtures were incubated at RT under gentle agitation. The extent of cell aggregation was assessed at 90 min by removing aliquots, spotting them onto culture slides (BD Falcon), and imaging by a Zeiss Axiosop A1 microscope. The resulting images were then analyzed by counting the number and size of particles using ImageJ. An arbitrary value for particle size was then set as a threshold based on negative control values. The aggregation index was calculated by expressing the number of particles participating in aggregation as a percentage of the total particles in 10 to 5 fields of 1.509 mm<sup>2</sup> per cell suspension combination of each independent experiment (n = 3). Statistical analysis was carried out using unpaired Student's t-test.

# Cell Culture and Transfection

HEK293 cells were maintained in high glucose Dulbecco's modified Eagle's medium 5 g/L glucose (DMEM; Gibco). Cell culture media were supplemented with 10% (v/v) heatinactivated fetal bovine serum (FBS, Lonza, BioWhittaker), 2 mM L-glutamine (PAA) and 1x penicillin/streptomycin (pen/strep; PAA) and cultured in a humidified atmosphere with 5% CO<sup>2</sup> at 37◦C. HEK293 cells were transfected using polyethylenimine (PEI; Polysciences) (Reed et al., 2006) or Lipofectamine LTX (Invitrogen, according to manufacturers manual). For the examination of Cntn6 effects on Lphn1 autoproteolysis, pcDNA3.1-Cntn6 and pcDNA3.1-HA-Lphn1 were cotransfected in HEK293 cells in the ratios 6:3 and 6:1 respectively. The transfected cells were lysed and analyzed on Western blot.

#### Cell Surface Binding Assay

fnmol-09-00143 January 18, 2017 Time: 16:35 # 3

To investigate whether Cntn6 interacts with Lphn1, a cell surface binding assay was used with slight modifications (Shimoda et al., 2012). Transfection of HEK293 cells with pIGplus-RGMa-Fc or pCr/TEV-ectoLphn1-Fc was performed. Forty eight hours after transfection the medium with soluble RGMa-Fc or Lphn1-Fc was concentrated through a 50,000 kDa column (YM-50, Milipore). The concentrated proteins were supplemented with Dulbecco's modified Eagle's medium 1 g/L glucose (DMEM; Gibco, Invitrogen) with 2 mM L-glutamine (PAA) and 1x penicillin/streptomycin (pen/strep; PAA) and distributed in 6-well plates with HEK293 cells transfected with pcDNA3.1-neogenin-myc or pcDNA3.1-HA-Cntn6 constructs. Binding between the proteins was allowed overnight in a humidified atmosphere with 5% CO<sup>2</sup> at 37◦C. Cells were fixed with 4% PFA in PBS, pH 7.4, and 0.01% sodium azide until immunocytochemistry was performed. For the cell surface binding analysis, images from the Zeiss Axioscop A1 were used. Analyses were performed of about 400 transfected cells per condition of each independent experiment (n = 3). The images were analyzed by quantification of the number of double labeled cells as a percentage of the total amount of transfected cells in ImageJ. Statistical analysis was carried out using unpaired Student's t.

#### Construction of Expression Vectors

A biotin- and GFP-tagged extracellular rat Cntn6 (Cntn6- TMGFPBio) fusion protein was generated by subcloning the coding sequence of the extracellular Cntn6 domains (NM\_013225.1: nt 248–3169), excluding the coding sequence of the GPI anchor. This was amplified from wild-type Cntn6 cDNA (pcDNA3.1-Cntn6) and ligated to the sequence of plexin-A1 transmembrane domain coding sequence (NM\_008881.2: nt 3962–4123). The coding sequences of a five glycine linker and intracellular GFP and biotin tags followed and were inserted in a pcDNA3.1(-)/myc-His (Invitrogen) vector backbone. The control vector (TMGFPBio) is identical but it is truncated beyond the transmembrane domain.

# Construction of Lphn1 shRNA Vectors

The pSUPER vector backbone (OligoEngine) was used to synthesize short hairpin RNA (shRNA) designed against Lphn1. This vector backbone carries the polymerase-III H1-RNA promotor, which produces small RNA transcripts lacking a polyadenosine tail and has a well-defined start of transcription and termination signal (Brummelkamp et al., 2002a,b). The following sequences were used: Lphn1 shRNA1: GCAACACC ATCCACAAGAA, Lphn1 shRNA2: CAAGGGAACTCGAGGA ATT, Lphn1 shRNA3: TCTCAGAGCTGGTGCACAA, Lphn1 shRNA4: GGGCAAATGCAGTTGGTCA. A non-targeting shRNA with a fully scrambled targeting sequence was designed as a control with the following sequence: GCTCTT AATCGCAAATACA. To examine the efficiency of Lphn1 knockdown, HEK293 cells were cotransfected with pcDNA3.1- HA-Lphn1-GFP and the Lphn1 shRNA constructs in a 1:2 ratio. The Lphn1 knockdown was examined in one experiment (n = 1) 3 days after transfection by quantification of fluorescence of about 1500 cells from five fields of 0.4 mm<sup>2</sup> per cell suspension of each transfection condition. The lysates were used in Western blot experiments and blot quantification was done by ImageJ.

# Ethics Statement

The experiments performed in this study were approved by the Experimental Animal Committee (DEC) of Utrecht (2010.I.06.073). All animal experiments were conducted in agreement with Dutch law (Wet op de Dierproeven, 1996) and European regulations (Guideline 86/609/EEC) related to the protection of vertebrate animals used for experimental and other scientific purposes.

## Immunostaining

Immunocytochemistry was performed after HEK293 cells were fixed with 4% PFA for 15 min at room temperature (RT) and washed in PBS (pH 7.4). The HEK293 cells were incubated in goat blocking buffer [PBS, 1% bovine serum albumin (BSA), 2% normal goat serum, 0.3% Triton X-100] for 1 h at RT. HEK293 cells were incubated with primary antibodies in goat blocking buffer overnight at 4◦C. Cells were washed in PBS and incubated with species-specific secondary antibodies conjugated to Alexa Fluor (Invitrogen) 1:2000 for 1 h at RT. Cells were washed in PBS and incubated with 4<sup>0</sup> ,6 diamidino-2-phenylindole (DAPI) (Sigma) before embedding. Images were captured by epifluorescence illumination on a Zeiss Axioscop A1. The primary antibodies that were used: rabbit anti-Cntn6-45 (antiserum produced by Harlan) 1:1000; rat anti-GFP (Chromotek) 1:500; rabbit anti-myc (Abcam) 1:500; and mouse anti-Fc-HRP (Bioconnect) 1:2500. Immunocytochemistry on primary cultures was performed as described before, with the following primary antibodies: rat anti-GFP (Chromotek) 1:500; rabbit anti-Caspase-3 (Cell Signaling) 1:1000; rabbit anti-Flag (Sigma) 1:250; rat anti-HA (Roche) 1:500; sheep anti-Cntn6 (R&D systems) 1:100; rabbit anti-Lphn1-p85 1:1000. Images were captured by confocal laser scanning microscopy (Olympus FV1000) by a Zeiss Axiosop A1.

For immunohistochemistry, the visual cortex was identified using standard stereotaxic coordinates (−2.80 mm to bregma). The sections were washed with PBS and incubated for 45 min in blocking buffer [1% BSA, 0.2% fish skin gelatin (Sigma), 0.1% Triton X-100 in PBS] and washed again. Sections incubated for 10 min in permeabilization buffer (0.3% Triton X-100 in PBS) before 2 h incubation with primary antibody in blocking buffer at 4◦C. The sections were washed in PBS and pre-incubated with blocking buffer before incubating with species-specific secondary antibodies conjugated to Alexa Fluor (Invitrogen) 1:500 for 2 h at RT. A 10 min DAPI incubation was performed after the sections were washed in PBS. The sections were embedded with Polyvinyl alcohol mounting medium with DABCO antifading (Fluka) onto glass slides after additional PBS wash steps. Primary antibodies that were used: sheep anti-Cntn6 (R&D systems) 1: 100; chick anti-Lphn1-p85 1: 500; rabbit anti-Caspase-3 (Cell Signaling) 1:400. Images were captured by confocal laser scanning microscopy (Olympus FV1000). Quantifications of caspase-3 immunoreactivity in the visual cortex were performed under a Zeiss Axioscop A1. At least three sections were analyzed from of Cntn6+/<sup>+</sup> and Cntn6−/<sup>−</sup> P14 animals (n = 5). Statistical analysis was carried out using unpaired Student's t.

#### Immunoprecipitation

fnmol-09-00143 January 18, 2017 Time: 16:35 # 4

Immunoprecipitation (IP) experiments were performed using GFP-Trap-A agarose beads (Chromotek, according to manufacturers manual). For proteomics, HEK293 cells expressing the indicated GFP-tagged fusion proteins were collected in ice-cold PBS and centrifuged at 1000 rpm in a precooled centrifuge at 4◦C for 5 min. Cell pellets were lysed in lysis buffer [10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 0.5% NP40, 1 mM PMSF and Complete protease inhibitor cocktail (Roche)], incubated on ice for 30 min and centrifuged at 13,200 rpm at 4◦C for 10 min. Cleared supernatant containing roughly 5.4 – 6.6 mg of protein was mixed with 50 µl GFP-Trap-A agarose beads (Chromotek), which had been equilibrated in dilution buffer [10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 1 mM PMSF and Complete protease inhibitor cocktail (Roche)] at 4◦C. After 1.5 h incubation at 4◦C, beads were washed two times in dilution buffer. Precipitated proteins were eluted by boiling the pull-down samples in NuPAGE LDS sample buffer (Invitrogen) containing 2% β-mercaptoethanol at 95◦C for 10 min.

For in vitro coIP, HEK293 cells were cotransfected with the following constructs: pcDNA3.1-Cntn6-TMGFPBio, pcDNA3.1- Lphn1-GFP or pCMV-EGFP-N1 with either pcDNA3.1-HA-Lphn1 or pcDNA3.1-HA-Cntn6. The pull-down experiments were performed using 1.8 – 2.2 mg total protein and 25 µl GFP-Trap-A agarose beads, as previously described. For endogenous coIP, P14 mouse cortex was lysed in lysis/washing buffer [20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 5% Glycerol, 1% CHAPS, 1 mM PMSF, Complete protease inhibitor cocktail (Roche) and phosphatase inhibitor cocktail (Sigma)], incubated for 30 min on ice and centrifuged at 13,200 rpm at 4◦C for 10 min. Cleared supernatants containing roughly 4 mg of protein were incubated at RT for 30 min with 50 µl paramagnetic beads with coupled recombinant protein G (Dynabeads Protein-G, Invitrogen), which were preincubated with 10 µg sheep anti-Cntn6 or normal sheep IgG (Milipore) antibodies in PBS and 0.02% Tween. Pull-down samples were washed three times in lysis/washing buffer and precipitated proteins were eluted by boiling in NuPAGE LDS sample buffer containing 2% β-mercaptoethanol at 70◦C for 10 min or by 5 min incubation with elution buffer (0.1 M Glycine, pH 2.5) before adding 5 µl Tris-HCl (1 M, pH 8.5).

#### In situ Hybridization

Non-radioactive in situ hybridization was performed as previously described Pasterkamp et al. (1998). In brief, probe sequences for Cntn6 (NM\_017383.3: nt 283–876) or Lphn1 (NM\_181039.2: nt 5203–5585) were polymerase chain reaction (PCR)-amplified from cDNA. Digoxigenin (DIG)-labeled RNA probes were generated by a RNA polymerase reaction using 10x DIG RNA labeling mix (ENZO). Tissue sections were post-fixed in 4% PFA in PBS, pH 7.40 for 20 min at RT. To enhance tissue penetration and decrease a specific background staining, sections were acetylated with 0.25% acetic anhydride in 0.1 M triethanolamine and 0.06% HCl for 10 min at RT. Sections were prehybridized for 2 hrs at RT in hybridization buffer (50% formamide, 5x Denhardt's solution, 5x SSC, 250 µg/ml baker's yeast tRNA and 500 µg/ml sonicated salmon sperm DNA). Hybridization was performed for 15 h at 68◦C, using 400 ng/ml denatured DIG-labeled probe diluted in hybridization buffer. After hybridization, sections were first washed briefly in 2x SSC followed by incubation in 0.2x SCC for 2 hrs at 68◦C. Sections were adjusted to RT in 0.2x SSC for 5 min. DIG-labeled RNA hybrids were detected with anti-DIG Fab fragments conjugated to AP (Boehringer) diluted in 1:2500 in TBS (pH 7.4) overnight at 4◦C. Binding of AP-labeled antibody was visualized by incubating the sections in detection buffer (100 mM Tris-HCl, pH 9.5, 100 mM NaCl and 50 mM MgCl2) containing 240 µg/ml levamisole and nitroblue tetrazolium chloride/5-bromo-4-chloro-3-indolyl-phosphatase (NBT/BCIP, Roche) for 14 h at RT. Sections subjected to the entire in situ hybridization procedure, but with no probe or sense probe added, did not exhibit specific hybridization signals. The specificity of the in situ hybridization procedure was also inferred from the clearly distinct gene expression patterns observed. Staining was visualized using a Zeiss Axioskop 2 microscope.

#### Mass Spectrometry: RP-NanoLC-MS/MS

The data was acquired using an LTQ-Orbitrap coupled to an Agilent 1200 system or an Orbitrap Q Exactive mass spectrometer connected to an Agilent 1290 system. In case of the LTQ-Orbitrap, peptides were first trapped ((Dr Maisch GmbH) Reprosil C18, 3 µm, 2 cm × 100 µm) before being separated on an analytical column (50 µm × 400 mm, 3 µm, 120 Å Reprosil C18-AQ). Trapping was performed at 5 µl/min for 10 min in solvent A (0.1 M acetic acid in water), and the gradient was as follows; 10 – 37% solvent B (0.1 M acetic acid in 80% acetonitrile) in 30 min, 37–100% B in 2 min, 100% B for 3 min, and finally solvent A for 15 min. Flow was passively split to 100 nl min−<sup>1</sup> . Data was acquired in a data-dependent manner, to automatically switch between MS and MS/MS. Full scan MS spectra from m/z 350 to 1500 were acquired in the Orbitrap at a target value of 5e5 with a resolution of 60,000 at m/z 400 in case of the LTQ-Orbitrap XL and 30,000 for the LTQ-Orbitrap Discovery. The five most intense ions were selected for fragmentation in the linear ion trap at normalized collision energy of 35% after the accumulation of a target value of 10,000. In case of the Q Exactive samples were first trapped [(Dr Maisch GmbH) Reprosil C18, 3 µm, 2 cm × 100 µm) before being separated on an analytical column (Agilent Poroshell EC-C18, 2.7 µm, 40 cm × 50 µm)]. Trapping was performed for 10 min in solvent A and the gradient was as follows; 13–41% solvent B in 35 min, 41–100% in 3 min and finally solvent A for 10 min. Flow was passively split to 100 nl min−<sup>1</sup> . The mass spectrometer was operated in data-dependent mode. Full scan MS spectra from m/z 350 – 1500 were acquired at a resolution of 35,000 at m/z 400 after accumulation to a target value of 3e6. Up to ten most intense precursor ions were selected for fragmentation. HCD fragmentation was performed at normalized collision energy of 25% after the accumulation to a target value of 5e4. MS/MS was acquired at a resolution of 17.500. In all cases nanoelectrospray was performed at 1.7 kV using an in-house made gold-coated fused silica capillary (o.d. 360 µm; i.d. 20 µm; tip i.d. 10 µm).

#### Neuronal Culture

fnmol-09-00143 January 18, 2017 Time: 16:35 # 5

P0–P1 mouse cerebral cortices were dissected and washed three times in L15 medium (Gibco) with 7 mM HEPES (L15- HEPES) and once in L15-HEPES with 0.5 M EDTA. For dissociation, the tissue was incubated in 0.25% trypsin (PAA) in L15-HEPES for 20 min at 37◦C, followed by trituration in complete Neurobasal [Neurobasal medium supplemented with 2% B27 (Invitrogen), 25 µM β-mercaptoethanol, 0.5 mM L-glutamine (PAA) and 1x penicillin/streptomycin (pen/strep; PAA)] with 20 µg/ml DNase I (Roche) using a fire-polished Pasteur pipette. Dissociated cortical neurons were run through a 100 µm cell strainer (BD Falcon) and plated in complete Neurobasal medium at 150 K/well of 12-well plates onto PDL (20 µg/ml) and laminin (40 µg/ml), Cntn6.6His (R&D Systems, 10 µg/ml) or BSA (Sigma–Aldrich, 10 µg/ml) coated glass coverslips.

#### Neuronal Transfection and Analysis

At DIV2, neurons in culture were cotransfected by Lipofectamine LTX (Invitrogen, according to manufacturers protocol), with pCAG-aGFP and either full-length pcDNA3.1-Cntn6, pcDNA3.1-Lphn1, combination of both or empty pcDNA3.1 control vector. For colocalization experiments, neurons were cotransfected with pcDNA3.1-HA-Lphn1, pCMV-Flag-Cntn6, or a combination of both. For Lphn1 knockdown experiments, neurons were cotransfected with pCAG-aGFP and a pSUPER vector carrying a scrambled sequence or a sequence designed against Lphn1 (the aforementioned shRNA3 and shRNA4) in a 1:12.5 ratio.

Neuronal medium was replaced with complete Neurobasal without antibiotics. A total of 0.5 µg DNA was incubated with 1.68 µl PLUS reagent in 200 µl Optimem (Invitrogen) for 10 min. One microliter Lipofectamine LTX was added to the DNA mix and was incubated for 30 min before addition to the neurons. At DIV5, neurons were washed with PBS, fixed with 4% PFA and 4% Sucrose in PBS, pH 7.4 for 20 min at 37◦C before washing 3 more times with PBS. After immunostaining, images from the Zeiss Axioscop A1 were taken. For analysis of neuronal morphological parameters, about 110 transfected neurons were examined per condition of each independent experiment (n = 3). WIS-Neuromath (Weizmann Institute) software was used for determining morphological parameters (Rishal et al., 2013), which included total branch number, soma size, total outgrowth and maximal process length. For analysis of neuronal apoptosis in the protein overexpression experiments, the immunoreactivity of caspase-3 in about 60 transfected neurons was quantified per condition of each independent experiment (n = 3). For analysis of neuronal apoptosis in the Lphn1 knockdown experiments, the immunoreactivity of caspase-3 in about 60 transfected neurons was quantified per condition of each independent experiment (n = 5 for Cntn6+/<sup>+</sup> and n = 4 for Cntn6−/<sup>−</sup> cultures). Positive neurons were analyzed by quantification of the number of double-labeled cells as a percentage of the total amount of transfected cells in ImageJ. Statistical analyses were carried out using unpaired Student's t.

#### Protein Separation and Digestion

Thirty microliter of each sample ran on a 12% Bis-Tris 1D SDS-PAGE gel (Biorad) either for 2–3 cm or ran completely and stained with colloidal coomassie dye G-250 (Gel Code Blue Stain Reagent, Thermo Scientific). The lane was cut into bands, which were treated with 6.5 mM dithiothreitol (DTT) for 1 h at 60◦C for reduction and 54 mM iodoacetamide for 30 min for alkylation. The proteins were digested overnight with trypsin (Promega) at 37◦C. The peptides were extracted with acetonitrile (ACN) and dried in a vacuum concentrator.

### Proteomics Data Analysis

Raw files were processed using Proteome Discoverer 1.3 (version 1.3.0.339, Thermo Scientific, Bremen, Germany). The database search was performed against the Swissprot database (version August 2014) using Mascot (version 2.4.1, Matrix Science, UK) as search engine. Carbamidomethylation of cysteines was set as a fixed modification and oxidation of methionine was set as a variable modification. Trypsin was specified as enzyme and up to two miss cleavages were allowed. Data filtering was performed using percolator, resulting in 1% false discovery rate (FDR). Additional filter was Mascot ion score >20. Raw files corresponding to one sample were merged into one result file.

Data was further analyzed with Saint (Choi et al., 2011) using the Crapome web interface<sup>1</sup> in order to identify interacting proteins. Default settings were used for calculating the FC-A and FC-B score. The probability score was calculated using Saint Express performing 20,000 iterations.

# PSD Preparation

To isolate postsynaptic densities (PSDs) from rat cortex or hippocampus, a modification of the method of Gardoni et al. (1998) was used. In brief, whole brain from 1 adult rat rapidly dissected and frozen on dry ice within 2 min to avoid postmortem intracellular protein trafficking (Suzuki et al., 1994). Homogenization was carried out by 20 strokes in a Teflon glass homogenizer (700 rpm) in 10 ml/g of cold 0.32 M sucrose containing 1 mM HEPES, 1 mM MgCl2, 1 mM NaHCO2, and 0.1 mM phenylmethylsulfonyl fluoride (PMSF) (pH 7.4) in the presence of a complete set of protease and phosphatase inhibitors (Sigma). The homogenized tissue (H) was centrifuged at 1,000 g for 10 min at 4◦C in a Sorvall centrifuge with SM24 inner rotor. The resulting supernatant was centrifuged at 13,000 g for 15 min at 4◦C with the Sorvall centrifuge to obtain the crude membrane fraction (P2). The pellet was resuspended in 5 ml/g

<sup>1</sup>www.crapome.org

of 0.32 M sucrose containing 1 mM HEPES, 1 mM NaHCO2, and 0.1 mM PMSF (pH 7.4), overlaid on a sucrose gradient (0.85—1.0—1.2 M), and centrifuged at 82.500 g for 2 h at 4◦C in a Beckman ultracentrifuge with SW41 swingout rotors. The synaptosome fraction (F1) between 1.0 M and 1.2 M sucrose was removed and diluted in 75 ml total volume in 0.5% Triton X-100, 0.32 M sucrose, 1 mM HEPES. This solution was spun down at 82,500 g for 30 min at 4◦C in the Beckman ultracentrifuge. The pellet was collected, resuspended and pottered by 20 strokes in a Teflon-glass homogenizer in 3 ml 0.32 M sucrose, 1 mM HEPES. The Triton insoluble postsynaptic fraction (P3) was removed and stored while the rest was layered on a sucrose gradient (1.0—1.5—2.1 M), and centrifuged at 100,000 g at 4◦C for 2 h the Beckman ultracentrifuge. The fraction between 1.5 M and 2.1 M was removed and diluted in total volume of 13 ml with 0.5% Triton X-100 and 75 mM KCl. The enriched PSD fraction (P4) was finally collected by centrifugation at 100,000 g at 4◦C for 30 min by the Beckman ultracentrifuge, and stored at −80◦C.

# Real-Time PCR

Mouse brain RNA was isolated from wild-type embryos at developmental stages E12.5, E14.5, E16.5, E18.5 and postnatal stages P7 and adult. To determine levels of Cntn6 expression, one-step qPCR was performed using a Quantifast SYBR Green RT PCR kit (Qiagen) and a LightCycler (Roche, Mannheim, Germany), according to manufacturers instruction. GAPDH primers: FW: CATCAAGAAGGTGGTGA AGC, RT: ACCACCCTGTTGCTGTAG, Cntn6 primers: FW: CCCAAGTTCCAACAAGAGGA, RV: GCCACGTGTA CGAAGGATT.

# Western Blotting

Cells were collected with a cell scraper in ice-cold PBS (pH 7.4) and centrifuged at 1000 rpm for 5 min in a precooled centrifuge at 4◦C. The cell pellet was resuspended in ice-cold lysis buffer (20 mM Tris-HCl, pH 8, 150 mM KCl, 1% Triton X-100, 1 mM PMSF and Complete protease inhibitor cocktail (Roche)), incubated on ice for 10 min, followed by centrifugation at 13,200 rpm for 10 min at 4◦C. The supernatant was collected, NuPAGE LDS sample buffer (Invitrogen) containing 2% β-mercaptoethanol was added and samples were boiled for 5 min at 90◦C. Proteins were separated in 8% SDS-PAGE gels and transferred onto nitrocellulose membrane (Hybond-C Extra; Amersham). Membranes were incubated in blocking buffer [PBS, 0.05% (v/v) Tween 20 and 5% milk powder] for 30 min at RT. Membranes were incubated with corresponding primary antibodies in blocking buffer overnight at 4◦C. Antibodies used: rat anti-GFP (Chromotek) 1:1000; rat anti-HA (Roche) 1:100, rabbit anti-Cntn6-45 (Harlan) 1:2000; chick anti-Lphn1-p120 and rabbit anti-Lphn1-p85 1:2500, rabbit anti-SynapsinI (Sigma) 1:2000, mouse anti-PSD95 (Milipore) 1:250, mouse anti-βActin (Sigma) 1:1000. Blots were incubated with SuperSignal West Dura Extended Duration Substrate (Pierce) and exposed to ECL films (Pierce) or imaged by FluorchemE Digital Darkroom (Cell Biosciences). ImageJ was used for blot quantification.

# RESULTS

# Identification of Lphn1 As Interacting Partner of Cntn6

An unbiased proteomics approach using Cntn6 protein fused to the Plexin A2 transmembrane domain and GFP (Cntn6- TMGFPBio and control TMGFPBio) was used to analyze immunoprecipitates (IP) of this protein expressed in HEK293 cells (**Figures 1A,B**). These cells are known to express a large range of proteins (Geiger et al., 2012). Following IP experiments, Cntn6 fusion and control proteins were detected at 148 and 40 kDa, respectively, by Western blotting and Coomassie blue staining (**Figures 1C,D**). Raw mass spectrometry data was analyzed with the Mascot search engine and scores were assigned to identify peptides. In comparison to control experiments, confidence scores using Saint scoring (Choi et al., 2011) were assigned to the identified proteins. A ranked list of putative interacting proteins was obtained representing proteins that were significantly higher or exclusively present in the Cntn6 fusion protein pull-down samples (**Table 1**). The highest scoring transmembrane protein was the adhesion G protein-coupled receptor Lphn1. We therefore further examined the potential of Cntn6 to interact and function with Lphn1.

# Cntn6 Interacts With Lphn1 In vitro and In vivo

To validate the association of Cntn6 and Lphn1, HEK293 cells were cotransfected with Cntn6-TMGFPBio and HA-Lphn1 expression plasmids, and appropriate controls. IP analysis by Western blotting demonstrated coprecipitation of HA-Lphn1 with Cntn6-TMGFPBio, but not with controls GFP (**Figures 1E,F**). Similarly, HA-Cntn6 and Lphn1-GFP coprecipitated in cotransfected HEK293 cells (**Figure 1G**). HEK293 cells coexpressing GFP and HA-Cntn6 were used as a control experiment, and IP did not result in HA-Cntn6 coprecipitation with control GFP (**Figure 1H**).

Next we investigated the association of endogenous Cntn6 and Lphn1 in P14 wild-type mouse cerebral cortex. These experiments resulted in the coprecipitation of both the p85 and p120-fragments of Lphn1 with Cntn6 indicating that an endogenous interaction of Lphn1 and Cntn6 exists in the brain (**Figure 1I**).

# Cntn6 and Lphn1 Interact in a Cis-Complex

The configuration of the Cntn6-Lphn1 interaction was determined through cell assays. Firstly, cell surface binding assays confirmed the biochemical interaction between Lphn1 and Cntn6: the soluble, tagged ectodomain Lphn1-Fc bound to membrane-bound HA-Cntn6 (**Figures 2A,B**). The wellcharacterized trans-interacting proteins neogenin and RGMa were used as positive controls in this assay (Yamashita et al., 2007; Itokazu et al., 2012). Secondly, to study whether Cntn6 and Lphn1 interacted in cis or trans, cell adhesion assays were performed involving separate populations of HEK293 cells cotransfected with either native Cntn6 and EGFP or with

Lphn1-GFP, (H) but not with control GFP. (I) Proteins were IPed from wild-type P14 mouse cortex lysates using an anti-Cntn6 antibody. Blots stained with antibodies against Cntn6 and both p85- and p120-fragments of Lphn1 revealed interaction between Cntn6 and Lphn1. No coprecipitation was found in normal IgG control IP. Molecular weights are as follows: Cntn6-TMGFPBio = 147.8 kDa; TMGFPBio = 39.6 kDA; Lphn1-GFP = 125 kDa; GFP = 27 kDa; HA-Cntn6 = 141 kDa; HA-Lphn1 = 131 kDa; Cntn6 = 130 kDa; Lphn1-p85 = 85 kDa; Lphn1-p120 = 120 kDa. Ig-like, immunoglobulin-like; FNIII, fibronectin type III; TM, transmembrane domain; GFP, green fluorescent protein; Bio, biotin.

Lphn1 and DsRed expression plasmids (**Figures 2C,D**). As a positive control, cells were cotransfected with neuroligin-1 (Nlgn1) and DsRed or with neurexin-1β <sup>−</sup> (Nrxn1β <sup>−</sup>) and EGFP expression plasmids (Ichtchenko et al., 1995). As negative controls, cells were transfected with either DsRed or EGFP expression plasmids only. Since Nlgn1 and Lphn1 both individually interact with Nrxn1β <sup>−</sup> (Nguyen and Südhof, 1997; Boucard et al., 2005, 2012) these proteins were used as positive controls. Indeed, a significant increase in the number of adhering cell clumps was observed when Nrxn1β <sup>−</sup>-expressing cells were mixed and incubated either with Nlgn1- or Lphn1 expressing cells (**Figures 2D,E**), demonstrating the validity of the assay. A significantly smaller degree of cell-aggregation was observed in the mixture of Nlgn1- (red) with EGFPexpressing (green) cells (**Figure 2C**). However, these aggregates comprised only red cells, which indicated Nlgn1's capability to homodimerize in trans. No cell-aggregation was found when Cntn6-expressing cells were mixed with Lphn1-expressing cells. These data show that the binding of Lphn1 to Cntn6 cannot occur when the proteins are expressed on opposing cells in trans-configuration. Together with the results of the cell surface binding experiments and coIP, the data indicate that Cntn6 and Lphn1 bind each other in cis-configuration and may form a heterodimer.


#### TABLE 1 | Top Cntn6 interacting proteins.

fnmol-09-00143 January 18, 2017 Time: 16:35 # 8

The proteins that were coimmunoprecipitated with Cntn6-TMGFPBio and detected by mass spectrometry are listed according their SAINT score. Data from triplicate IPs were compared to control using CRAPome, which employs SAINT and FC statistical analyses. This table lists the protein identification code (PROTID) and its related gene name, together with FC-A, FC-B, and SAINT scoring. The standard CRAPome Fold Change calculations (FC-A) estimates the background by averaging the spectral counts across the selected controls. Another more stringent Fold Change (FC-B) calculation estimates the background by combining the top 3 values for each identified interacting protein, while SAINT reports a probability of true interaction. Cellular localization was determined by searching Uniprot (UniProt Consortium, 2015). See the Supplementary Proteomics Datasheet for the complete raw data.

In view of the Cntn6-Lphn1 cis-interaction and the proposed localization of Lphn1 and Cntn6 in the synapse (Sakurai et al., 2009; Silva et al., 2011), we examined the distribution of Lphn1 and Cntn6 across different synaptic fractions. Both proteins were present in the postsynaptic density fractions (Supplementary Figure 1). This supports the conclusion that Cntn6 and Lphn1 interact in cis and suggests that they form a postsynaptic complex.

#### Coexpression of Cntn6 and Lphn1 in the Mouse Brain

An endogenous Cntn6-Lphn1 heteromeric protein complex in the post-synapse can only be present in neurons that coexpress both proteins. We therefore determined and compared the expression patterns of Cntn6 and Lphn1. First, the temporal expression of Cntn6 was determined in the developing mouse brain by real time PCR. These experiments showed that Cntn6 expression was highest at P7 (Supplementary Figure 2A), when Lphn1 expression was also appreciable (Boucard et al., 2014). Comparison of Cntn6 and Lphn1 expression by in situ hybridization at P7 showed that Cntn6 expression was more restricted as compared to Lphn1 expression. Overlapping expression of both genes was particularly prominent in layer V of the cerebral cortex, the anterodorsal (AD) and anteroventral (AV) nuclei of the thalamus and the internal and external granular layers of the cerebellum (IGL and EGL respectively) (Supplementary Figure 2B). Additional overlapping expression was observed in layer V of the cerebral cortex, the subiculum and the CA1 region of the hippocampus. Immunostaining for Cntn6 and Lphn1 proteins confirmed the presence of both proteins in layer V of the cerebral cortex (**Figure 3A**) and in the AD and AV nuclei of the thalamus in P14 mice (Supplementary Figure 2C). No Cntn6 immunoreactivity was observed in Cntn6−/<sup>−</sup> mice, confirming the specificity of the immunohistochemical staining. Primary neuronal cultures immunostained for endogenous Cntn6 and Lphn1 proteins revealed colocalization at the cell surface and intracellular sites in neurons (**Figure 3B**). Staining along neurites was especially pronounced. Similar results were obtained by cotransfection with Flag-Cntn6 and HA-Lphn1 expression plasmids (**Figure 3C**).

Taken together, these data show that specific regions in the brain contain neurons that coexpress Cntn6 and Lphn1, suggesting that interaction of these proteins may serve biologically relevant functions. The coexpression in cortical neurons provided us with the opportunity to further examine these putative functions.

## Cntn6 Reverses Morphological Defects Induced by Lphn1

To explore functions of the cis-interaction of Cntn6 and Lphn1, we investigated the effects of Cntn6 and Lphn1 expression alone and in combination, on the cellular morphology of cultured cortical neurons (**Figure 4A**). Cellular analysis showed that expression of Cntn6 did not affect neuronal morphology, including length and branching of neurites and soma size (**Figure 4B**). However, Lphn1-expressing neurons displayed a significant reduction in the number of branching points, soma size, total neurite length and longest branch length per neuron (**Figure 4B**). A progressive decrease of these parameters was observed with increasing plasmid concentrations (Supplementary Figure 3). Total neurite length and the longest branch length were most affected. Staining for the apoptosis marker caspase-3 showed that the level of apoptosis significantly increased upon expression of Lphn1 (**Figures 4C,D**). Markedly fewer transfected cells were observed in neuronal cultures transfected with Lphn1 overexpression plasmids (data not shown). This suggested that cultured neurons were severely compromised by Lphn1 expression and underwent apoptosis, pointing to a neurotoxic effect of Lphn1.

Notably, in these experiments coexpression of Cntn6 together with Lphn1 significantly reversed morphological parameters affected by Lphn1 alone (**Figure 4B**). When Cntn6 was coexpressed with Lphn1, morphological parameters reached or increased toward the levels in controls and neurons transfected with Cntn6 alone. Furthermore, the increased number of apoptotic neurons induced by Lphn1 expression was reduced by coexpression of Cntn6 (**Figures 4C,D**) Comparing Lphn1-expressing neurons cultured on Cntn6-coated substrate, providing Cntn6 in trans, or control BSA-coated substrate showed that Cntn6 in trans did not affect the neurotoxic effect

of Lphn1 (Supplementary Figure 4). From these experiments we concluded that Lphn1 expression in cultured cortical neurons confers a neurotoxic activity resulting in altered overall neuronal morphology, reduced survival, which is rescued when Cntn6 interacts in cis with Lphn1.

To further validate this interaction, Lphn1 knockdown was performed in primary cultures obtained from Cntn6−/<sup>−</sup> mice. Several short hairpin RNA (shRNA) expressing plasmids directed at different sequences of Lphn1 coding region were designed. Of these, shRNA3 and -4 were most efficient in reducing Lphn1 protein expression (**Figure 5A**). ShRNA3, -4, and the scrambled shRNA plasmids were separately transfected, together with EGFP, in primary cortical cultures derived from wild-type and Cntn6−/<sup>−</sup> mice. Immunostaining for EGFP and caspase-3 revealed that Cntn6−/<sup>−</sup> cortical cultures displayed an increase in apoptosis compared to wild-type cultures when transfected with the scrambled shRNA plasmid. When cortical cultures were transfected with the Lphn1 shRNA plasmids, apoptosis was significantly decreased in the Cntn6−/<sup>−</sup> cultures whereas this remained unchanged in the wild-type cultures (**Figures 5B,C**). These data indicated that endogenous Lphn1 was responsible for the increased level of apoptosis in Cntn6−/<sup>−</sup> neurons, and consequently that Cntn6 plays an endogenous role in inhibiting Lphn1-incuded apoptosis in these cultured neurons.

To critically test this indication in vivo apoptosis was assessed in the visual cortex of 14-day-old wild-type and Cntn6−/<sup>−</sup>

mice. A significant increase of apoptosis was observed in the visual cortex of Cntn6−/<sup>−</sup> mice, compared to wild-type mice (**Figures 5D,E**). These results are in line with previous reports showing increased apoptosis in the internal granule cell layer of the cerebellum and in primary cultured cortical neurons in Cntn6−/<sup>−</sup> mice (Sakurai et al., 2009; Huang et al., 2011a). Taken together, the in vitro and in vivo data indicate that in the absence of Cntn6 Lphn1-induced apoptosis occurs. These data also indicate that the Cntn6-Lphn1 cis-complex is functional in controlling apoptosis.

#### DISCUSSION

Since CNTN6 has been implicated in neurodevelopmental disorders we set out to examine pathways of action of this contactin member. Current insights in the mode of action of Cntn6 were too limited to fully explain its involvement in developmental mechanisms, despite the description of phenotypes in null mutant mice, such as biochemical interactors (Zuko et al., 2013). These phenotypes included a developmental delay of the corticospinal tract, a misorientation of apical

dendrites in the cortex, altered numbers of subtype specific projection neurons and interneurons in the cortex, and an increase in neuronal cell death during cerebellar development (Ye et al., 2008; Sakurai et al., 2009; Pinto et al., 2010; Huang et al., 2011b; Zuko et al., 2016). Also, significant reduction in glutamatergic synapses was found in the hippocampus and in the cerebellum of Cntn6 null-mutants (Sakurai et al., 2009, 2010). These data suggested that loss of Cntn6 impairs development. Here we show that null-mutation of Cntn6 increases apoptosis in vitro and in vivo and that this effect involves a pathway that is dependent on the presence of the adhesion GPCR Lphn1. Expression of Lphn1 in primary cultures resulted in smaller neurons, shorter neurites and increased apoptosis. These phenotypes were rescued by coexpression with Cntn6, indicating that Cntn6 inhibits these adverse effects of Lphn1. Moreover, Lphn1 knockdown in cultured neurons from Cntn6−/<sup>−</sup> mice reduced the amount of caspase-3-positive neurons, suggesting that Lphn1 indeed contributed to the increase of apoptosis in absence of Cntn6. Essentially this conclusion was supported by our in vivo findings of increased apoptosis in the visual cortex, a region coexpressing the Cntn6 and Lphn1 genes.

In line with this, it was previously reported that cortical cultures from Cntn6−/<sup>−</sup> animals displayed a significant decrease

difference compared to the scrambled control plasmid. However, apoptosis in Cntn6−/<sup>−</sup> primary cultures was significantly reduced after treatment with Lphn1 shRNA3 and -4 compared to the scrambled control plasmid. About 60 transfected neurons were analyzed for caspase-3 immunoreactivity per condition of each independent experiment (n = 5 for Cntn6+/<sup>+</sup> cultures and n = 4 for Cntn6−/<sup>−</sup> cultures). (D,E) In vivo analysis of caspase-3 expression (green) in the visual cortex of P14 wild-type and Cntn6−/<sup>−</sup> animals revealed a significant increase of apoptosis in the absence of Cntn6. DAPI staining is in blue. Scale bar represents 150 µm. Analysis was performed on at least three sections per brain of wild-type and Cntn6−/<sup>−</sup> mice (n = 5 per genotype). Statistical analyses were performed using unpaired Student's t. The graph bars are presented as mean ± SEM. <sup>∗</sup>p < 0.05, ∗∗p < 0.01.

in cell survival (Huang et al., 2011a). Additionally, Cntn6−/<sup>−</sup> mice showed aggravated brain damage after trauma compared to wild-type mice, due to impaired neuronal survival and neurite growth (Huang et al., 2011a). Also the delay of corticospinal tract formation in Cntn6−/<sup>−</sup> mice might be attributed to the loss of a protective influence exerted by Cntn6 (Sakurai et al., 2009; Huang

et al., 2012). These observations agree with a role of Cntn6 in preventing neurotoxicity.

The finding that Cntn6 reverses the effects of Lphn1 on neuronal morphological parameters and survival, only when coexpressed with Lphn1 and not when provided in trans, underscores the functional role for a Cntn6-Lphn1 heterodimeric cis-complex. This mechanism bears similarities to the concept of dependence receptors (Mehlen and Bredesen, 2004; Goldschneider and Mehlen, 2010; Mehlen and Tauszig-Delamasure, 2014). Cells expressing dependence receptors require the presence of a ligand to survive (Goldschneider and Mehlen, 2010; Mehlen and Tauszig-Delamasure, 2014). We show that Cntn6 is an endogenous ligand for Lphn1 and prevents this receptor to confer apoptosis when Cntn6 is present. As such Lphn1 may be considered a dependence receptor. Dependence receptors generally trigger two opposite signaling pathways depending on the occupation by their ligands. Classical signaling pathways are activated when bound to their ligands, supporting cell survival, migration and differentiation, and apoptotic signaling is conferred in unbound state (Goldschneider and Mehlen, 2010). As yet about 20 dependence receptors have been labeled as such, none of them belonging to the class of adhesion GPCRs (Mehlen and Tauszig-Delamasure, 2014). Adhesion GPCRs are a specific subfamily of receptors (Langenhan et al., 2013; Hamann et al., 2015; Krishnan and Schiöth, 2015) that display multiple signaling properties depending on structural conformation and state (Kenakin, 2011; Kishore et al., 2016). Indeed it has been reported that Lphn and Lat1, the Caenorhabditis elegans ortholog of mammalian Lphns, can activate Ca2<sup>+</sup> and cAMP and bind to multiple G-proteins, suggesting that Lphns can activate multiple signal transduction cascades (Silva et al., 2011; Boucard et al., 2012; Müller et al., 2015).

The complex cell biology of adhesion GPCRs at the level of intracellular transport, proteolysis, reassociation, and dimerization and (self)activation (Langenhan et al., 2013; Hamann et al., 2015) leaves multiple possibilities for inhibition of Lphn1 by Cntn6 in cis. In our experiments Cntn6 appeared to be bound to full length Lphn1, since the tryptic peptides found in mass spectrometry were derived from N- as well as C-terminal parts of Lphn1 (Supplementary Figures 5A,B), and in our coIP experiments both Lphn1 N- and C-terminal domains were coprecipitated together with Cntn6 (**Figure 1**). Since we found that no difference in the relative quantities of the intact protein and ectodomain of Lphn1 was present in cells with or without coexpression of Cntn6 (Supplementary Figure 5C), it is suggested that the expression of Cntn6 does not affect the autoproteolysis of Lphn1, and that other mechanisms are involved which are subject of future experiments.

Since the apoptotic activity of Lphn1 is regulated in a ciscomplex with Cntn6, coexpression of Cntn6 and Lphn1 is required for this mode of action. We found an increase of caspase-3-positive cells in the visual cortex of Cntn6−/<sup>−</sup> animals. The cortex indeed appeared one of the brain regions that contain neurons coexpressing Cntn6 and Lphn1. Another region with marked co-expression of Cntn6 and Lphn1 mRNA was the cerebellum, in particular the IGL (Supplementary Figure 2B). A significant increase of cell death has also been reported previously in the IGL of the cerebellum of Cntn6−/<sup>−</sup> animals (Sakurai et al., 2009). However, Lphn1 is more widely expressed in the brain than Cntn6. Therefore, we speculate that other ligands may regulate Lphn1 activity in brain regions where Cntn6 is absent. Such ligands potentially include Lasso/teneurin-2, FLRT3, and Nrxn1. These proteins all form high-affinity trans-synaptic ligand-receptor pairs with Lphn1 with signaling capabilities, shaping synapse structure and regulating synaptic development and function (Silva et al., 2011; Boucard et al., 2012; O'Sullivan et al., 2012). On the other hand, Cntn6 is known to complex with other membrane proteins as well, including Ptpra, Ptprg, PTPσ, Notch, and Chl1 (Cui et al., 2004; Hu et al., 2006; Ye et al., 2008; Bouyain and Watkins, 2010; Zuko et al., 2011). A case in which cell adhesion proteins form supercomplexes with competing components has recently been made for Lphn3 association with Flrt and UncD members at the structural and functional level (Jackson et al., 2016). In the protein interaction repertoire of Lphn1 other ASD gene products are known to be present, in particular, Nrxn1 which is a major ASD gene interacting with Nlgn1 and LRRTMs (Ichtchenko et al., 1995; de Wit et al., 2009). Thus, Cntn6 may also link to the Nrxn1-Nlgn1 pathway of autism through interaction with Lphn1.

Present and previous data have indicated that loss of function of Cntn6 can result in increased apoptosis. Apoptosis plays a crucial role during development of organisms and organs, including the brain (Meier et al., 2000; Hipfner and Cohen, 2004; Mehlen and Bredesen, 2004). Apoptosis has also been implicated in neurodevelopmental disorders, particularly schizophrenia, in which cases with severe reduction of neuron numbers have been found (Margolis et al., 1994; Jarskog et al., 2005). It has been suggested that neurodevelopmental disorders with complex genetic etiology and large numbers of risk genes, including ASD, may include cases that share apoptosis as a pathogenic pathway (Wei et al., 2014). Neuropathology on brain tissues of autistic subjects has revealed an increase in apoptosis in several brain areas (Sheikh et al., 2010a,b). Furthermore, changes in apoptotic and anti-apoptotic proteins in post-mortem brain tissue have been found, and interpreted as derangements in regulation of apoptosis in autism (Fatemi and Halt, 2001; Araghi-Niknam and Fatemi, 2003; Mahfouz et al., 2015). This suggestion is further emphasized by several genetic animal models of ASD which display increased caspase-3 activity (Yochum et al., 2008; Sakurai et al., 2009; Olczak et al., 2010; El-Ansary et al., 2012). The Cntn6-deficient mouse shares this phenotype. Together, our data define Cntn6 as a ligand for Lphn1 modulating its apoptotic activity by direct binding thereby impinging on neurodevelopment.

#### AUTHOR CONTRIBUTIONS

AZ, AO-A, AA, AH, AP, YS, RP, and JB designed experiments; AZ, HP, and RvD performed experiments; AZ, RvD, and RT

analyzed results; AZ, BvdZ, RP, and JB wrote the paper; all authors edited and approved the manuscript.

#### FUNDING

This study was supported by a Fellowship from the Dutch Brain Foundation nr. F2008(1)-08 (BvdZ), by a JSPS Fellowship (AO-A), by the Russian Science Foundation grant No. 14-14-01195 (AP), and by Stichting Parkinsonfonds (RP). AZ is supported by a JSPS Fellowship. This work is part of the project Proteins At Work, financed by the Netherlands Organisation for Scientific Research (NWO) as part of the National Roadmap Largescale Research Facilities of the Netherlands (project number 184.032.201).

## REFERENCES


## ACKNOWLEDGMENTS

We would like to thank Inma Luque Molina for performing the subcellular fractionation and Henk Spierenburg for performing the real time PCR experiments and for genotyping of the animals. We are grateful for the gift of expression plasmids by Peter Scheiffele and Joris de Wit, to Kazutada Watanabe for continuous support.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2016.00143/full#supplementary-material


neural guidance receptors. Nat. Commun. 7: 11184. doi: 10.1038/ncomms 11184



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Zuko, Oguro-Ando, Post, Taggenbrock, van Dijk, Altelaar, Heck, Petrenko, van der Zwaag, Shimoda, Pasterkamp and Burbach. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Ulises Gómez-Pinedo1 , Rocio N. Villar-Quiles2 , Lucia Galán2 , Jordi A. Matías-Guiu2 , Maria S. Benito-Martin1 , Antonio Guerrero-Sola2 , Teresa Moreno-Ramos2 and Jorge Matías-Guiu1,2\**

*1Neurobiology Laboratory, Faculty of Medicine, Neuroscience Institute, IdISSC, Hospital Clínico San Carlos, Universidad Complutense de Madrid, Madrid, Spain, 2Neurology Department, Faculty of Medicine, Neuroscience Institute, IdISSC, Hospital Clínico San Carlos, Universidad Complutense de Madrid, Madrid, Spain*

#### *Edited by:*

*Daniela Tropea, Trinity College Dublin, Ireland*

#### *Reviewed by:*

*Ian Paul Johnson, University of Adelaide, Australia Cintia Roodveldt, Andalusian Molecular Biology and Regenerative Medicine Centre (CSIC), Spain*

#### *\*Correspondence:*

*Jorge Matías-Guiu inc@hcsc.salud.madrid.org*

#### *Specialty section:*

*This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neurology*

*Received: 26 July 2016 Accepted: 24 October 2016 Published: 08 November 2016*

#### *Citation:*

*Gómez-Pinedo U, Villar-Quiles RN, Galán L, Matías-Guiu JA, Benito-Martin MS, Guerrero-Sola A, Moreno-Ramos T and Matías-Guiu J (2016) Immununochemical Markers of the Amyloid Cascade in the Hippocampus in Motor Neuron Diseases. Front. Neurol. 7:195. doi: 10.3389/fneur.2016.00195*

Background: Several findings suggest that the amyloid precursor protein (APP) and the amyloid cascade may play a role in motor neuron disease (MND).

Objective: Considering that dementia is one of the most frequent non-motor symptoms in amyotrophic lateral sclerosis (ALS) and that hippocampus is one of the brain areas with greater presence of amyloid-related changes in neurodegenerative diseases, our aim was to analyze the molecular markers of the amyloid cascade of APP in pathology studies of the hippocampus of autopsied patients with ALS and ALS–frontotemporal dementia (FTD).

Methods: We included nine patients with MND and four controls. Immunohistochemical studies and confocal microscopy were used to analyze the expression of APP, TDP-43, pho-TDP-43, Aβ, APP intracellular cytoplasmatic domain (AICD) peptide, Fe65 protein, and pho-TAU in the hippocampus of seven patients with ALS, two patients with ALS– FTD, and four controls. These findings were correlated with clinical data.

results: Patients displayed increased expression of APP and Aβ peptide. The latter was correlated with cytoplasmic pho-TDP-43 expression. We also found decreased Fe65 expression. A parallel increase in AICD expression was not found. Patients showed increased expression of pho-TAU in the hippocampus. Findings were similar in patients with ALS and those with ALS–FTD, though more marked in the latter group.

conclusion: Post-mortem analyses showed that the amyloid cascade is activated in the hippocampus of patients with MND and correlated with cytoplasmic pho-TDP-43 expression. The number of intracellular or extracellular aggregates of Aβ peptides was not significant.

Keywords: amyotrophic lateral sclerosis, frontotemporal dementia, amyloid precursor protein, A**β** peptide, AICD peptide, TDP-43, TAU protein, Fe65 protein

## INTRODUCTION

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease affecting motor neurons in the brain, brainstem, and spinal cord. Its prevalence rate is estimated at 4–6 cases per 100,000 population (1). It has a variable clinical course, and onset is usually focal (more frequently spinal than bulbar). Life expectancy is estimated at 3–5 years on average. Although the cause is unknown, it has been linked to mutations in different genes even in the sporadic forms (2). Pathology studies in patients with ALS have shown that degeneration not only affects motor areas but also the dorsolateral prefrontal cortex, anterior cingulate, hippocampus, dentate gyrus (DG), parietal lobe, substantia nigra, cerebellum, amygdala, and basal ganglia (3).

The copresence of several neurodegenerative diseases in some patients has led to defining new groups of diseases, some of which are linked to certain genetic or molecular markers. For example, the association between ALS and certain forms of frontotemporal dementia (FTD) has been linked to TDP-43 positive cytoplasmic inclusions as well as to mutations in the gene encoding this protein or other related proteins, in both familial and sporadic forms whether these molecular changes are a cause or a consequence of neuronal degeneration is unclear (4–7). Although the association between ALS and Alzheimer disease (AD) has not been well studied, some authors have suggested that these entities may co-occur. Although the association between ALS and AD has not been well studied, some authors have suggested that these entities may co-occur (8–17).

Although patients with ALS may also present memory impairment and even dementia (18), only two pathology studies have analyzed the presence of changes compatible with AD. In a series of 30 patients with ALS, Hamilton and Bowser (19) found a prevalence of dementia of 24.1%. Around 30% of the patients displayed pathological changes associated with AD, especially amyloid and neuritic plaques in the hippocampus, which were more common in the DG and neocortex. These authors also found patients without dementia who showed AD lesions; Aβ deposition was present in up to 50% of cases. In that study, pathological changes of AD, especially neuritic plaques, were associated with older age and shorter survival time (20). In another study, Coan and Mitchell (20) analyzed 46 autopsies of patients with ALS and found that 22% of the cases met criteria for AD and 26% for FTD, 78% displayed neurofibrillary tangles, and 35% showed moderate increases in Aβ expression, especially in the amygdala, hippocampus, and the entorhinal and insular cortices. Likewise, ALS onset in patients who met the criteria for AD was more frequently bulbar; onset type had no impact on survival time. All patients with Aβ peptide accumulation showed greater prevalence of neurofibrillary tangles in the hippocampus and amygdala. Furthermore, the form of ALS reported in Guam residents is associated with abundant neurofibrillary tangles in addition to the classic neuropathological findings of the disease (21, 22).

Amyloid cascade is defined by the consequences of amyloid precursor protein (APP) cleavage after two successive proteolysis, which has served as the basis for the amyloid hypothesis of AD (23), and finally produces Aβ. While β- and γ-secretases promote Aβ formation, α-secretase has the opposite effect, which results in two pathways: the amyloidogenic and the non-amyloidogenic APP pathways, respectively. The last proteolysis in the amyloidogenic pathway produces two substrates: the peptides Aβ and the APP intracellular cytoplasmatic domain (AICD), which plays an important role in the transcriptional regulation of certain genes. The idea that certain pathological changes of AD may be present in patients with ALS or ALS–FTD suggests the possibility that they have common mechanisms (24), but it may also indicate that these changes are not primary (25). Patients with ALS do not generally present clinical changes as intense amnestic symptoms suggesting that the hippocampus could be significantly affected, contrary to what occurs in AD. Taking into consideration that the hippocampus is one of the areas of the brain most affected by molecular changes associated with AD, we analyzed the expression of proteins linked to the amyloid cascade in the hippocampus of autopsied patients with ALS or ALS–FTD.

#### MATERIALS AND METHODS

#### Selection and Study of Patients and Controls

We analyzed the autopsies of nine patients that were included in the ALS registry of our neurology department. This registry follows a protocol for diagnosing and treating ALS that includes clinical assessment, electromyographic and neuroimaging studies, and blood and CSF analyses, as well as criteria for assessing FVC and using BiPAP, and indicating gastrostomy and riluzole (26). These patients, who died between 2006 and 2012, met the El Escorial revised and Ludolph criteria for ALS (27, 28). The following clinical variables were included in our analysis: age, sex, disease duration, clinical form, whether riluzole was administered, and clinical data related to potential cognitive impairment. Data were gathered from the ALS registry by thoroughly reviewing clinical histories and conducting telephone interviews with the patients' relatives. Eight of the patients included in our study had died of respiratory failure at the terminal stage of ALS. The remaining patient died due to cardiac arrest upon admission after attending the hospital due to progressive dysphagia and moderate bulbar involvement and being diagnosed with ALS. Our controls were four patients who died at a hospital due to a non-neurological disease. They were included because their families had consented to donation of the body. None of them had a history of neurodegenerative diseases. We reviewed their medical records to rule out a history of neurological disease. One of our controls was donated by another center; we were therefore unable to review the medical history but confirmed that the patient had not died from a neurological cause and had no history of neurodegenerative disease.

#### Autopsy Procedure and Preparation and Storing of Biological Material

Autopsies were performed within 2–6 h after death following our hospital's standard protocol and in compliance with Spanish regulations for this procedure. We followed the standard method: opening the cranial cavity and severing the upper end of the spinal cord at the foramen magnum. To separate the hemispheres, we cut right along the midline of the corpus callosum and prepared them for later sectioning. Tissue samples were fixed in 10% buffered formalin [phosphate-buffered saline (PBS); 0.1M, pH 7.35]. The hemisphere allocated to histological and immunohistochemical analyses was sectioned into coronal slices (maximum slice thickness = 1 cm), which were placed on a flat surface in order from the frontal pole to the occipital pole. The nervous system analysis included weighing the brain, examining its macroscopic morphology, and conducting a microscopic study, using conventional techniques. In all cases, the study included hematoxylin-eosin and Nissl staining. Additional techniques were used at the discretion of the pathologist, including Congo red and silver staining to determine the presence of such macroscopic and microscopic alterations as atrophy, senile plaques, eosinophilia, or neuronal loss. Thioflavin-S staining was used for all cases and controls.

#### Study of the Hippocampus

Once the tissue for microscopic analysis had been adequately prepared, paraffin blocks were sectioned to obtain the hippocampus, which should include at least the following areas: CA1, CA2, CA3, and DG. All tissue samples were embedded in paraffin following the protocol established by the pathology department at our hospital and subsequently sectioned to study the cytoarchitecture of the hippocampus. We used Braak staging, which was developed by Braak and Braak (29), to assess pathological findings.

#### Immunohistochemical Study

Tissue was sectioned into 6-μm slices using a microtome (Leica). Slices were deparaffinized and thoroughly washed with PBS 0.1M. Epitopes were unmasked in a 10-mM sodium citrate buffer with a pH of 6 at 96°C for 30 min. Samples destined for APP immunostaining were additionally incubated for 20 min in formic acid v/v. Following this, all samples were incubated in a blocking solution (PBS, 0.2 Triton X-100, 10% normal goat serum) for 1 h. After that, tissues were incubated in primary antibodies diluted in PBS for 24 h. After incubation with primary antibodies, sections were washed with PBS and incubated in the appropriate HRP or Alexa-Fluor secondary antibody (see Table S1 in Supplementary Material). For APP and phospho TAU antibodies, goat anti-rabbit HRP was used, the sections were stained brown with a DAB-peroxidase solution of 0.03% diaminobenzidine and 0.01% H2O2, and counterstained with Nissl staining. They were subsequently mounted in DPX and observed under a ZEISS microscope. For the immunofluorescence study of the other markers, tissue sections were thoroughly washed and incubated with the appropriate Alexa-Fluor antibody for 24 h after incubation with primary antibodies. After thoroughly washing the sections, they were mounted in ProLong Gold reagent with DAPI (Molecular Probes, Invitrogen) and observed in an Olympus confocal microscope AF2000. Five hippocampal slides were used, those containing CA1–3 and DG (separated by at least 200 μm each, to avoid double quantifications), and the quantitative study included the analysis of 10 different fields for each of the analyzed antibodies; the result was the mean of the 10 measurements. In the cases where the unit of measurement was the amount of labeling per field [optical density (OD)], we used Image J software version 1.46r, developed by the National Institutes of Health. In the cases where inclusions were assessed, these were quantified based on the number of stained inclusions that were found in neurons divided by the mean number of neurons per field (percentage of cells in 500 μ<sup>2</sup> ).

#### Statistical Analysis

Statistical analysis was performed using SPSS statistical software version 20.0. Results were represented graphically using GraphPad Prism version 5.0. Data are expressed as mean ± SD. Means were compared using the non-parametric Mann–Whitney *U* test due to the small sample size. Graphs were created using the program mentioned above. Statistical significance was set at *p* < 0.05.

#### RESULTS

#### Description of the Study Sample

Our study included nine patients (seven with ALS and two with ALS–FTD) and four controls. Five patients were men (56%), and four were women (44%). Mean age at diagnosis was 64.56 ± 16.8 years, and mean age at death was 65.33 ± 15.7 years. Regarding the controls, there were three men (75%) and one woman (25%), with a mean age at death of 68.75 ± 14.86 years. Six of the nine patients had an additional disease: one had a history of stroke, one had ulcerative colitis, two had a history of depression, one presented essential tremor, and the remaining one had REM sleep behavior disorder. Only one of the patients had a first-degree relative with a history of dementia of unknown origin. Likewise, only one patient presented a familial form of ALS. This patient, who had a non-pathological SOD1 mutation, had been included in a previous study (30). Mean time elapsed from onset of motor/bulbar symptoms to diagnosis of ALS was 6.44 ± 3.9 months (not including the two cases with ALS–FTD). Disease onset was bulbar in five patients (56%) and spinal in the remaining four (44%). Seven patients (78%) had received riluzole and six (67%) required permanent mechanical ventilation; mean time elapsed from diagnosis to indication of mechanical ventilation was 9 months. Five patients (56%) were being fed by gastrostomy. Mean time elapsed from onset of motor/bulbar symptoms to death was 17.6 ± 16.5 months (we did not count the time the two patients with FTD had presented impairment linked to FTD before ALS onset) and 11.11 ± 15.6 months from diagnosis to death. Cognitive disorders were only reported in the patients with ALS–FTD. Patient histories and clinical characteristics are shown in Table S2 in Supplementary Material.

## Description of General Pathological Findings in the Autopsies

Table S3 in Supplementary Material summarizes the results of the neuropathology study and shows the fixed brain weight for each subject. No significant differences were found in brain weight between patients with ALS (1220 ± 147 g) and controls (1290 ± 72.73 g; *p* = 0.244). Neuropathological findings resulted in a diagnosis of motor neuron disease (MND) in all patients. From a macroscopic point of view, precentral gyrus atrophy was seen in patients eight and nine and none of the controls. Likewise, two patients, and none of the controls, displayed thinning of the ventral roots of the spinal cord; all patients showed considerable neuronal loss in the anterior horn of the spinal cord. One of the patients exhibited loss of motor neurons in the medulla oblongata. One patient displayed numerous neurofibrillary tangles and amyloid plaques, while another showed moderate extracellular amyloid deposition. Ubiquitinated inclusions were found in the hippocampus and spinal cord of five patients. The mean number of neurons with ubiquitin-positive inclusions was 0.5 ± 0.58 in controls and 7.43 ± 2.7 in patients (*p* = 0.007), or 7.67 ± 0.8 if we exclude the patients with FTD (*p* = 0.01) (data expressed in 500 μm2 field).

## Cytoplasmic Phosphorylated TDP-43 Expression Is Increased in the Hippocampus of Patients with ALS and ALS–FTD

TDP-43 is a transcription factor that is located in the nucleus of cells. In MNDs, this protein is cleaved and translocated to the cytoplasm, where it may be phosphorylated. TDP-43, like ubiquitin, can be observed in cytoplasmic inclusions, leading to the idea of "TDP-43 pathology," which encompasses such diseases as ALS and ALS–FTD. This protein is also found in the cytoplasm of neurons in patients with AD. In our sample, the mean percentage of expression of cytoplasmic phosphorylated TDP-43 (pho-TDP-43) in the hippocampus was higher in patients than in controls, which confirms that this area is affected. The area of the hippocampus showing the greatest TDP-43 expression was CA1, followed in decreasing order by GD, CA3, and CA2 (data not shown). The mean percentage of hippocampal cells with TDP-43-positive inclusions was 20.15 ± 10.46 (range 6.5–33.34) in the patient group and 1.62 ± 1.45 (range 0–2.67) in the control group. All patients displayed increased pho-TDP-43 expression, compared to controls; differences between patients and controls were statistically significant (*p* = 0.0028). Comparing ALS patients with ALS–FTD patients is not feasible since our sample included only two patients with ALS–FTD; however, higher pho-TDP-43 expression was observed in the ALS–FTD group. These data can be seen in Figure S1 in Supplementary Material.

#### APP Expression Is Increased in the Hippocampus of ALS and ALS–FTD Patients but Is Not Correlated with TDP-43 Expression

Amyloid precursor protein is a ubiquitous transmembrane, type-1, integral glycoprotein of 110–130 kDa that is extensively expressed in human tissues. In the CNS, some functions attributed to APP are neurite outgrowth and synaptogenesis, protein trafficking along axons, cell adhesion, calcium metabolism, and signal transduction (31). Mean APP expression was higher in the hippocampus of our patients [4345 ± 1975 (range 3832–8204) OD] compared to controls [1925 ± 309 (range 1761–2458) OD]. APP expression was increased in all patients; differences between patients and controls were statistically significant (*p* = 0.0028). APP expression was greater in patients with ALS–FTD. These data are summarized in **Figure 1**. APP expression in the hippocampus was not correlated with phospo TDP-43 expression (*r* = 0.20), as shown in **Figure 2A**, and APP did not co-localize with TDP-43 in the cytoplasm (data not shown). Considering that APP plays an important role in cell survival, increased APP expression may be a cell response to neurodegeneration. However, the hippocampus is not clinically affected in ALS; the correlation found between APP and TDP-43 supports this hypothesis. The amyloid cascade of APP is active in the hippocampus in patients with MND and correlates with pho-TDP-43 expression.

The successive activity of two proteolytic processes, involving β- and γ-secretase, respectively, produces Aβ peptides, mainly Aβ40 and Aβ42. These may form oligomers and insoluble fibrils that accumulate both intracellularly and in extracellular amyloid plaques. This latter represents the pathological substrate for AD since Aβ peptide secretion plays a role in regulating neurotransmitter release in the synapses (32). Increased intracellular Aβ expression indicates greater amyloid cascade activity. Mean intraneuronal Aβ expression in the hippocampus was higher in patients than in controls: 32,261 ± 25,720 (range 7431–79,676) OD vs. 5393 ± 2048 (range 3542–8270) OD. All but one patient showed increased Aβ expression; differences between patients and controls were statistically significant (*p* = 0.0056). Greater APP expression was observed in one of the patients with ALS–FTD. These data are shown in **Figure 3**. The areas of the hippocampus showing increased intraneuronal Aβ expression were, in descending order, CA1, GD, CA3, and CA2 (data not shown). Although scarce, we observed Aβ deposits in the form of dense, diffuse plaques. Similar to what occurred with APP and cytoplasmic pho-TDP-43, Aβ expression and cytoplasmic pho-TDP-43 expression were moderately correlated (*r* = 0.530) (**Figure 2B**) but did not co-localize (data not shown). This correlation suggests that there is a link between cytoplasmatic pho-TDP-43 and activation of the amyloid cascade of APP in the hippocampus of patients with MND.

#### Expression of AICD Is Variable in MND

Amyloid precursor protein intracellular cytoplasmatic domain, the substrate common to both APP pathways, results from the activity of γ-secretase on sAPPα or sAPPβ; these soluble intermediate fragments are generated by α- and β-secretase, respectively (33, 34). This peptide forms a complex with adaptor protein Fe65 and histone acetyltransferase Tip60, and this complex plays an important role in regulating transcription of the genes encoding such proteins as APP, β-secretase, KAI1 (CD82), neprilysin, and p53 (35). In our sample, mean AICD expression in the hippocampus of patients was significantly raised compared to controls (*p* = 0.02); however, the expression was not elevated in three patients (**Figure 4**). The mean number of hippocampal cells showing AICD expression per area analyzed was 15 ± 3.18 (range 9–18) in patients and 8 ± 2.58 (range 5–11) in controls; patients with ALS–FTD displayed even higher numbers. These data are shown in Figure S2 in Supplementary Material. Wang et al. (36)

demonstrated that AICD binds to and co-localizes with TDP-43 in the nucleus of cultured HEK293 cells. In our study, we neither found a significant correlation between AICD expression and cytoplasmic pho-TDP-43 expression nor did we observe co-localization between the two (data not shown).

#### Fe65 Expression Is Lower in Patients with MND

Highly expressed in the hippocampus (37), Fe65 is an adaptor protein that is thought to play a crucial role in modulating the amyloid cascade of APP (38–42). Fe65 promotes the APP cascade by increasing Aβ production; this mechanism is attenuated by Fe65 phosphorylation (43). Mean Fe65 expression in hippocampal cells was significantly lower in patients [14 ± 3.37 (range 8–19)] than in controls [22.5 ± 3.37 (range 21–29)] (*p* = 0.018). None of the patients presented levels similar to those found in controls. Within the patient group, Fe65 expression was higher in patients with ALS–FTD. These findings (**Figure 5**) suggest that decreased Fe65 expression is probably due to phosphorylation in an attempt to decrease Aβ production. Cell labeling of Fe65 was at times weak or diffuses in the nucleus. The areas with greater positive immunoreactivity were, in descending order, DG, CA1, CA3, and CA2. Fe65 expression is inversely correlated with that of APP, AICD, and TDP-43 (Figure S3 in Supplementary Material), which supports the hypothesis that the amyloid cascade is activated in the hippocampus of patients with MND.

#### Pho-TAU Is Expressed to a More Marked Degree in the Hippocampus of Patients with ALS

TAU is overexpressed in patients with AD and other neurodegenerative diseases. TAU phosphorylation and aggregation is the molecular basis of neuritic plaques (44, 45). In our study, mean cell expression of hippocampal pho-TAU s396 was significantly increased in patients compared to controls (*p* = 0.002); none of

the patients showed similar levels to those displayed by controls. Patients had a mean OD of 21,431 ± 21,455 (range 7431–73,844) and controls, 3278 ± 1378 (2481–5534). Pho-TAU s396 expression, which was slightly greater in patients with ALS–FTD (**Figure 6**), was shown to be inversely correlated with AICD expression (*r* = 0.570) (Figure S4 in Supplementary Material). A possible explanation for this correlation is provided by a study in which AICD-overexpressing transgenic mice showing no increase in Aβ levels displayed greater TAU expression and protein aggregation (46). The explanation for this is that AICD peptide upregulates GSK-3β expression, GSK-3β activation, and consequently TAU phosphorylation in rat neuronal cultures (47). This suggests that AICD induces TAU expression.

#### Correlations with Clinical Profile and Progression

Despite the small number of patients included in our study, we correlated immunopathological findings with clinical characteristics and progression. These data are shown in Figure S5 in Supplementary Material. There is a statistically significant inverse correlation between total TDP-43 levels and time elapsed from diagnosis to death (*r* = 0.75), and between the percentage of cytoplasmic TDP-43 and time to indication of mechanical ventilation (*r* = 0.69). According to this latter correlation, greater levels of cytoplasmic TDP-43 are associated with a poorer prognosis. Likewise, increased expression of pho-TDP-43 and AICD are associated with bulbar onset (pho-TDP-43: 16.20 vs. 11.35% of cells with TDP-43-positive inclusions; AICD: 15.8 vs. 11.5, mean number of cells per analyzed area). Three patients showed increased pho-TAU s396 expression, which was associated with increases in AICD. Pho-TAU s396 overexpression was associated with more advanced Braak stages and greater quantities of thioflavin-S stained fibers.

Comparing immunopathology findings with the clinical forms of the disease showed that bulbar forms were associated with greater AICD (*p* = 0.031) and TDP-43 (*p* = 0.020) expression. We observed no significant differences in the expression of the remaining markers. Despite the small number of patients with ALS–FTD, the percentage of cytoplasmic TDP-43 was found to be greater in that subgroup (*p* = 0.0475).

The expression profile of markers showed no significant differences between patients with survival times shorter than 12 months and those with longer survival times. Higher levels of TDP-43 were correlated with shorter survival times (*r* = 0.753), as shown in Figure S5 in Supplementary Material.

#### DISCUSSION

There are many studies in the literature suggesting that neurodegenerative diseases share molecular characteristics; however, few studies have addressed this hypothesis with regard to AD and ALS (8, 19, 20). Findings from different studies point to certain common mechanisms. For example, Aβ accumulation has been found in the spinal cords of patients with both the familial and sporadic forms of ALS (48) as well as in the skin and muscles of ALS patients (49, 50), and biomarkers linked to the amyloid cascade have been found in the CSF of patients with ALS and FTD (49). In addition, increased APP expression has been observed in spinal cord motor neurons in experimental models in the early stages of ALS (51–54), and some experimental models of AD have displayed extracellular Aβ plaques in motor neurons, similar to those found in humans with AD (55). Likewise, increased Aβ expression has been found in affected motor neurons and the surrounding glial cells in SOD1G93A mouse models; genetic ablation of APP in these mice reduces motor neuron degeneration (56). These findings are the reason for our interest in understanding

The graphs display mean optical density (arbitrary units).

biomarker expression in the amyloid cascade of APP and the connection with TDP-43 in patients with ALS.

Some clinical data, especially cognitive and neuroimaging findings, suggest that the hippocampus is affected in ALS (57, 58). Some authors have even proposed that hippocampal involvement in ALS has a different, less marked, pattern to that associated with AD degeneration (59, 60). Our aim was to study molecular alterations in the hippocampus, since this area shows no clinical changes in ALS but is greatly affected in AD patients. We found increased expression of cytoplasmic TDP-43 and pho-TDP-43, which confirms that ALS affects the hippocampus at a molecular level. Increased levels of TDP-43 in the hippocampus have been reported in previous studies, although they were linked to long disease progression times (61, 62). This is not applicable to our patients, who presented a short mean survival time (17.6 months from symptom onset). In any case, the presence of both cytoplasmic TDP-43 and pho-TDP-43 indicates that the mechanisms of neurodegeneration are active in the hippocampus of patients with MND.

Our study shows that the amyloid cascade of APP is activated and expressed in the hippocampus *via* its molecular markers in both ALS and ALS–FTD, as we found increased expression of APP and Aβ peptides, and even pho-TAU s396 overexpression. This increase in Aβ peptides is correlated with the expression of cytoplasmic pho-TDP-43 peptides. These data may suggest that APP expression and amyloid cascade activation are a response to molecular changes caused by MND: increased APP expression may be a mechanism of cell survival.

Our patients showed reduced Fe65 expression, probably due to the fact that Fe65 binds to AICD to downregulate APP expression. The hypothesis that Fe65 forms a complex with AICD is reinforced by the fact that, in some patients, increases in Fe65 expression were not present as would be expected considering that Fe65 production is simultaneous to that of Aβ after γ-secretase

activity. We also found increased expression of pho-TAU in the hippocampus in all cases. Although it has been suggested that increased TAU expression is linked to AICD production (63), we found no correlation between the two. It is therefore most likely due to increased intraneuronal Aβ expression.

Our findings appear to confirm the hypothesis that the amyloid cascade of APP is activated in the hippocampus of patients with ALS and ALS–FTD. However, thioflavin-S staining in some of the patients revealed few intracellular or extracellular Aβ aggregates. This is consistent with *in vivo* findings from a previous study using PET with 18F-florbetaben, which reported low tracer uptake in the hippocampus of patients with ALS (64). This may be explained by the fact that this tracer binds to Aβ deposits and fibers and does not detect increases in peptide expression (65), as shown in our pathology study. However, some studies on ALS have shown amyloid tracer uptake (66–68). For instance, one study found tracer uptake in elderly patients (69). We hypothesize that a cohort of patients with longer survival times may present longer cascade activation periods, which may in turn lead to a greater presence of aggregates that PET imaging or thioflavin staining would detect.

Molecular changes found in the hippocampus are not linked to a specific clinical profile and progression pattern. Increased expression of AICD was only found in bulbar ALS; however, the significance of this finding is difficult to interpret. Greater expression of total or cytoplasmic TDP-43, which occurs in bulbar onset ALS and ALS associated with FTD, is correlated with poorer prognosis, meaning shorter survival time from diagnosis or shorter time elapsed to indication of mechanical ventilation. This idea is in line with studies suggesting that the measurement of TDP-43 may be a prognostic biomarker of the disease (70), although several other studies show elevations in CSF cannot be considered a sensitive diagnostic marker at that moment (71, 72).

Our study has a number of limitations. First, the included patients are representative of the most severe forms of the disease (high percentage of bulbar onset ALS and short survival time),

since all of them were bodies that had been donated. We therefore cannot rule out the premise that the degree to which the amyloid cascade is activated depends on the intensity of neurodegeneration. This hypothesis is also supported by the fact that changes seemed to be more intense in patients with ALS–FTD than in those with ALS. Studies with greater sample sizes are necessary to confirm this idea. Furthermore, this limitation added to a sample of patients with short survival times does not allow us to evaluate the extent to which the amyloid cascade affects the hippocampus over longer time periods. The convenience of increase in the sample is particularly necessary in the subgroup of patients with FTD/ALS, where the small number of cases does not allow statistical analysis in any molecular marker, such as cytoplasmatic TDP-43. Second, autopsies were performed 2–6 h after death since these patients rarely die in hospital and must therefore be transported from their homes or palliative care centers. This time window may have had some impact compared with those from animal research where the times were shorter. In addition, none of the patients without FTD had cognitive alterations according to the data from medical records and clinical histories, which was confirmed by telephone interviews with the patients' relatives. Therefore, we cannot establish a link between cognitive alterations and molecular changes in the amyloid cascade of APP in the hippocampus. This should be studied in a patient cohort with longer survival times.

In conclusion, our post-mortem analyses showed that the amyloid cascade of APP is activated in the hippocampus of patients with ALS and ALS–FTD and correlates with TDP-43 expression. Immunohistochemical analyses revealed no significant intracellular or extracellular Aβ aggregates.

#### ETHICAL STANDARDS

The present study complies with the ethical standards of the research committee at our center and the 1964 Declaration of Helsinki and its subsequent amendments.

Pho-TAU s396 expression is greater in patients. Scale bar: 50 μm. The dotted line indicates the significance threshold. The graphs display mean optical density

AUTHOR CONTRIBUTIONS

(arbitrary units).

Study design: UG-P and JM-G; patient evaluation: RV-Q, LG, and AG-S; coordination of autopsy studies: LG; microscopy and molecular study: UG-P and MB-M; database: UG-P and RV-Q; statistical analysis: UG-P, RV-Q, and JAM-G; analysis of results: UG-P, JAM-G, LG, TM-R, and JM-G; figures and tables: UG-P and RV-Q; manuscript draft: JM-G; manuscript revision and approval: all the authors.

# ACKNOWLEDGMENTS

The authors would like to thank Prof. Armando Martínez of the Pathology Department for his help as advisor of the work. They also wish to thank to translator advisory service of the Spanish Society of Neurology's Research office for helping in the preparation of the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fneur.2016.00195/ full#supplementary-material.

TABLE S1 | Antibodies and techniques used in the study.

#### TABLE S2 | Clinical characteristics of the patient sample. ALS,

amyotrophic lateral sclerosis; M, male; F, female; ARF, acute respiratory failure; CRA, cardiorespiratory arrest; FTD, frontotemporal dementia; PPA, primary progressive aphasia.

TABLE S3 | General data from the pathology study of the MND patients (numbers 4 and 7 were ALS/FTD cases). Semi-quantitative assessments. The scale used is 0: no alterations, +: alteration present, this alteration ++: moderately, this alteration +++: markedly.

FIGURE S1 | Expression of TDP-43 (A) and cytoplasmic pho-TDP-43 in mean values **±** SD and individual values for controls and patients [(B,E),

#### respectively] and fluorescence images from controls (C) and patients

(D). Although there are no differences in total TDP-43 expression (A), we found statistically significant differences (*p* = 0.0061) in the expression of cytoplasmic pho-TDP-43 (B). Photomicrographs (C,D) show the differences in labeling between controls and patients; as can be observed, cytoplasmic pho-TDP-43 expression (arrows) is greater in patients (D). (E) Shows the individual data corresponding to patients and controls; expression is greater in patients than in controls, and that increase is even more marked in patients with ALS–FTD. Scale bar: 50 μm. The dotted line indicates the significance threshold. The graphs present the percentage of cells with immunopositive inclusions in 500 μ<sup>2</sup> .

#### FIGURE S2 | The directly proportional relationship between AICD and the expression of pho-TDP-43 (A), APP (B), and A**β** (C) is displayed.

FIGURE S3 | The inversely proportional connection between Fe65 and the markers of the amyloid cascade is demonstrated: APP (A), AICD (B), and

#### REFERENCES


A**β** peptide (C); this suggests that Fe65 levels may be downregulated by mechanisms linked to AICD and the amyloid cascade.

#### FIGURE S4 | A direct correlation between AICD peptide and pho-TAU s396 in the hippocampus of patients with ALS (*r* **=** 0.570) is shown. This is

probably due to the direct link between activation of the amyloid cascade and TAU phosphorylation, which is closely related to neuronal transport. This would confirm that both AICD and pho-TAU play a major role in the pathogenesis of ALS.

#### FIGURE S5 | Correlations between total TDP-43/cytoplasmic pho-TDP-43 expression and such clinical variables as time to indication of mechanical ventilation (A), time elapsed from diagnosis to death (B), and

age at death (C). The graph shows a weak correlation between time to mechanical ventilation and expression of pho-TDP-43 (*r* = 0.02) (A). We found a linear correlation between pho-TDP-43 expression and time (in months) elapsed from diagnosis to death (*r* = 0.45) (B). The correlation between total TDP-43 and age at death (C) was *r* = 0.47.


associated with the C9ORF72 repeat expansion. *Neurocase* (2014) 20:150–7. doi:10.1080/13554794.2012.741252


a possible prognostic value. *Amyotroph Lateral Scler* (2011) 12:140–3. doi:10.3109/17482968.2010.541263

72. Kasai T, Tokuda T, Ishigami N, Sasayama H, Foulds P, Mitchell DJ, et al. Increased TDP-43 protein in cerebrospinal fluid of patients with amyotrophic lateral sclerosis. *Acta Neuropathol* (2009) 117:55–62. doi:10.1007/ s00401-008-0456-1

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2016 Gómez-Pinedo, Villar-Quiles, Galán, Matías-Guiu, Benito-Martin, Guerrero-Sola, Moreno-Ramos and Matías-Guiu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# A Novel Approach for Amplification and Purification of Mouse Oligodendrocyte Progenitor Cells

Junlin Yang<sup>1</sup> , Xuejun Cheng<sup>1</sup> , Jiaxi Shen<sup>1</sup> , Binghua Xie<sup>1</sup> , Xiaofeng Zhao<sup>1</sup> , Zunyi Zhang<sup>1</sup> , Qilin Cao<sup>2</sup> , Ying Shen<sup>3</sup> and Mengsheng Qiu1,4 \*

<sup>1</sup> Zhejiang Key Laboratory of Organ Development and Regeneration, The Institute of Developmental and Regenerative Biology, College of Life and Environment Sciences, Hangzhou Normal University, Hangzhou, China, <sup>2</sup> The Vivian L Smith Department of Neurosurgery, University of Texas Medical School at Houston, Houston, TX, USA, <sup>3</sup> Department of Neurobiology, Key Laboratory of Medical Neurobiology of the Ministry of Health, Zhejiang Province Key Laboratory of Neurobiology, Zhejiang University School of Medicine, Hangzhou, China, <sup>4</sup> Department of Anatomical Sciences and Neurobiology, University of Louisville, Louisville, KY, USA

Although transgenic and knockout mice are widely used to study the specification and differentiation of oligodendrocyte precursor cells (OPCs), mouse primary OPCs are difficult to be purified and maintained, and many in vitro studies have to resort to rat OPCs as substitutes. In this study, we reported that mouse O4 negative earlystage OPCs can be obtained by culturing cortical tissue blocks, and the simultaneous treatment of OPCs with Platelet Derived Growth Factor-AA (PDGFaa), basic fibroblast growth factor (bFGF), and epidermal growth factor (EGF) is the key for the propagation of mouse OPCs in culture. EGF was found to be a potent mitogen for OPCs and cooperate with PDGFaa to extend cell division and inhibit their differentiation. EGF also collaborates with PDGFaa and bFGF to convert bipolar or tripolar OPCs to more vital fibroblast-like OPCs without compromising their oligodendrocyte differentiation potential. In addition, EGF promoted the survival and proliferation of glial progenitor cells (GPCs) derived from primary OPC cultures, and a mixture of GPCs and OPCs can be obtained and propagated in the presence of EGF, bFGF, and PDGFaa. Once EGF is withdrawn, GPC population decreased sharply and fibroblast-like OPCs changed into typical OPCs morphology, then homogeneous OPCs were obtained subsequently.

Keywords: OPCs, GPCs, coculture, EGF, synergistic effect, PDGFaa, stratified culture

# INTRODUCTION

In the vertebrate, central nervous system (CNS), oligodendrocytes (OLs) produce myelin sheaths wrapping around axons to facilitate the rapid conduction of nerve impulses and support axonal survival (Marinelli et al., 2016). Oligodendrocyte precursor cells (OPCs) that give rise to OLs were first identified in cultures of postnatal rat optic nerve cells. They are bipotential GPCs, which can differentiate into OLs under defined culture conditions, or into type-2 astrocytes (A2B5+/GFAP+) in the presence of fetal bovine serum (FBS; Wang et al., 2013). As a result, they were previously termed oligodendrocyte – type 2 astrocyte (O-2A) progenitor cells (Tanner et al., 2011). OPCs

#### Edited by:

Daniela Tropea, Trinity College, Ireland

#### Reviewed by:

Adelaide Fernandes, Universidade de Lisboa, Portugal Ruben Lopez-Vales, Autonomous University of Barcelona, Spain

#### \*Correspondence: Mengsheng Qiu

m0qiu001@yahoo.com

Received: 20 May 2016 Accepted: 08 August 2016 Published: 22 August 2016

#### Citation:

Yang J, Cheng X, Shen J, Xie B, Zhao X, Zhang Z, Cao Q, Shen Y and Qiu M (2016) A Novel Approach for Amplification and Purification of Mouse Oligodendrocyte Progenitor Cells. Front. Cell. Neurosci. 10:203. doi: 10.3389/fncel.2016.00203

were initially characterized in cultures by their antigenic phenotype of A2B5 and their bipolar or tripolar morphology, but more specific molecular markers such as platelet-derived growth factor receptor alpha (PDGFRα) and basic helix-loophelix transcription factor Olig2 were later identified (Lu et al., 2002; McKinnon et al., 2005). While early-stage OPCs are mostly O4− negative, late-stage OPCs start to acquire O4 antigen with multipolar morphology as they progress along the oligodendroglial lineage (Yang et al., 2011).

Oligodendrocyte precursor cells are widely used as a model system to explore the molecular pathways controlling oligodendrocyte differentiation and axonal myelination in vitro, and the pathogenic mechanisms underlying certain demyelinating diseases (Clemente et al., 2013). Establishment of culture conditions for primary OPCs can provide a large number of purified cells for study of development/pathology mechanisms and transplantation-based myelin repair research (Windrem et al., 2004; Cao et al., 2010). To date, the success in isolation and amplification of primary OPCs remains to be restricted to rat tissues. Two different methods have been developed to isolate rat OPCs from brain tissues. One is the cell sorting approach based on cell surface antigen such as immunopanning (Yang et al., 2013) and fluorescence-activated cell sorting (FACS; Sohn et al., 2006), with the cell purity depending on the specific surface antigen. However, many surface antigens of OPCs such as A2B5 and NG2 can be found in other cell types in the CNS (Richardson et al., 2011). The other is the shake method based on the differential adherent properties of glia; however, this method is largely limited to rat tissues due to the difficulty in obtaining stratified culture of OPCs and astrocytes in mouse tissues (Chen et al., 2007). Several studies described methods to obtain mouse OPCs through formation of "oligospheres" from multipotent cortical progenitor cells (Chen et al., 2007; Pedraza et al., 2008), but the scale is limited and their genetic, epigenetic or molecular identity may depart from their in vivo counterparts. As mice are widely used in transgenic and knockout studies, it becomes increasingly important to study the molecular or signaling mechanisms underlying the phenotypic chances in oligodendrocyte differentiation or axonal myelination with purified OPC cells in vitro (Clemente et al., 2013; White and Krämer-Albers, 2014).

In this study, we developed a new method to obtain stratified cultures of mouse cortical OPCs and astrocytes, and most OPCs separated from the astroglial cell layer by shaking were O4− with bipolar or tripolar morphology. Epidermal growth factor (EGF) was found to be a potent mitogen for OPCs and synergized with Platelet Derived Growth Factor-AA (PDGFaa) to promote cell division and inhibit their differentiation into O4+ cells. More importantly, EGF promotes morphological change of bipolar or tripolar OPCs to fibroblast-like GPCs, maintains the committed oligodendrocyte differentiation potential and functions as a potent mitogen. EGF-dependent tripotential GPCs derived from primary OPC cultures were capable of divisions as well, and provided nourishments for OPCs in the mixed cultures. Upon EGF withdrawal from culture medium, fibroblastlike OPCs reversed to bipolar or tripolar morphology and GPCs reduce gradually, then homogeneous young OPCs were obtained.

#### MATERIALS AND METHODS

#### Cortical Tissue Dissociation

Mouse cortices were isolated from postnatal day 1 mouse pups (Chen et al., 2007) in accordance with the NIH Guide for the Care and Use of Laboratory Animals. Cortical tissues were diced into ∼1 mm<sup>3</sup> pieces in a 60 mm dish with a sterilized razor blade, and the minced tissues were resuspended with freshly prepared D/F20S medium: Dulbecco's Modified Eagle Medium: Nutrient Mixture F-12 (DMEM/F-12; Gibco, Grand Island, NY, USA) supplemented with 20% v/v FBS (HyClone, Logan, UT, USA) and 1% v/v Penicillin/streptomycin (P/S; Gibco; 5 ml D/F20S medium added for each brain). Tissues were then pipetted up and down until homogenized, transferred into flasks (material from one brain to one T25 flask; Corning) and incubated with 5% CO<sup>2</sup> at 37◦C. Seventy-two hours after plating medium was replaced, most cortical pieces attached onto flasks and some flat cells started to migrate out of tissues. Afterward, medium was replaced every other day.

## OPC Isolation and Culture

With continuous migration of flat cells out of the tissue pieces, some process-bearing cells appeared on the top of bed layer of flat cells. Around the 10th day after plating, mixed glial cultures became confluent. Floating cells were removed by gently rinsing the cell cultures twice with D/F20S medium, and culture flasks were shaken for 15–18 h (37◦C, 250 rpm) with tightened caps on an orbital shaker model 420, Orbital Size 1.0 (Thermo Fisher, Waltham, MA, USA). Cell suspension was collected and filtered through Cell Strainer of 40 µm pore size (BD, Franklin Lakes, NJ, USA) to remove small clumps of astrocytes, and then pelleted by centrifugation at 100 × g for 5 min at RT. Cells were resuspended in OPC basal media: DMEM/F12 (Gibco) supplemented with 1 × N2 (Gibco), 1 × B27 (Gibco), 1 × P/S (Gibco), and 0.1% w/v BSA (Sigma) plus 10 ng/ml PDGFaa (PeproTech, Rocky Hill, NJ, USA), and then transferred to untreated Petri dishes (Corning) and incubated for 30 min, allowing astrocytes and microglia to attach to the surface, while OPCs remained suspended. Cells in suspension were then collected and counted using a hemocytometer before plated into Poly-D-lysine (PDL; Sigma, 100 µg/mL) and Laminin (LN; Sigma, 20 µg/mL; PDL/LN) coated dishes or plates (Corning) as needed.

## OPC Differentiation

For the directed differentiation of OLs, mouse OPCs were cultured in PDL/LN-coated chamber slides (Millipore, Temecula, CA, USA) with OPC basal media without any growth factor for 3 days, then examined for O4 and MBP antigen expression. For astrocyte differentiation, OPCs were cultured in PDL/LN-coated chamber slides with DMEM/F12 (Gibco) containing 10% v/v FBS (HyClone) and 1 × P/S for 6 days, and then immunostained with A2B5 and GFAP antibody.

# Effects of Growth Factors on OPC Proliferation and Differentiation

For the analysis of cell proliferation and differentiation, 1 × 10<sup>4</sup> OPCs freshly prepared from the astroglial cell layer were plated to each PDL/LN-coated 24-well before different growth factors or combinations were added. Cell proliferation was analyzed by adding EDU (Life Technologies, Waltham, MA, USA) to a final concentration of 10 µM. Following 24 h of incorporation, cells were fixed in 4% paraformaldehyde at RT for 10 min, and EDU positive cells were detected by Click-iT <sup>R</sup> EdU Alexa Fluor <sup>R</sup> 488 Imaging Kit (Life Technologies), followed by immunostaining with O4 or MBP antibody. Double positive cells were counted from three different areas of each well under fluorescence microscopy. The results were expressed as mean values and standard deviation.

# Preparation of Mixed Cultures of OPCs and GPCs

After detached from the bed layer of flat cells, mouse cortical OPCs were suspended in OPC basal media supplemented 10 ng ml−<sup>1</sup> EGF (PeproTech), bFGF (PeproTech), and PDGFaa (PeproTech) and plated at a density of 1 × 10<sup>3</sup> cells/cm<sup>2</sup> into PDL/LN-coated 100 mm dishes. Cells were incubated in 5% CO<sup>2</sup> at 37◦C, and medium was changed every other day. Earlystage OPCs and few GPCs among the primary OPC cultures divided rapidly under the stimulation of EGF + bFGF + PDGFaa. OPCs and GPCs are more firmly attached to dishes and can only detach from dishes by enzymatic treatment instead of by shaking. Following several passages with the aid of trypsinization, the fibroblast-like OPCs and GPCs increased to about 20–40% based on their morphology, and the trypsinization did not appear to reduce cell replication significantly for the first several passages.

## OPC Amplification from Mixed Cultures of OPCs and GPCs

To grow mouse OPCs in large numbers, we plated the mixed culture of OPCs and GPCs at a low density of 1 × 10<sup>3</sup> cells/cm<sup>2</sup> on PDL/LN-coated 100 mm dishes, in culture medium supplemented with bFGF (PeproTech) and PDGFaa (PeproTech), and replaced half the medium every other day. In the absence of EGF, GPCs division slowed down, then followed by apoptosis, whereas fibroblast-like OPCs reversed to bipolar or tripolar morphology and continued to proliferate. The mixed cells were split (1:5 ratio) every 5 days for two passages, and high-purity OPCs were obtained.

# Dorsal Root Ganglion Neurons (DRG) Co-culture for Myelination Assays

Similar to OPCs in mice, DRGs were also difficult to prepare, so we have to resort to their rat counterpart as substitutes. DRGs were isolated from E15 Sprague-Dawley rat embryos as previously described (Dadsetan et al., 2009), and cultured for neurite outgrowth as described by Dincman et al. (2012). OPCs/DRGs co-cultures were maintained 8–12 days in OPC basal media supplemented with 30 ng/ml Triiodothyronine (T3; Sigma), and fixed for immunostaining with anti-MBP and antineurofilament (axons) antibodies.

# Immunofluorescence Staining Analysis

Immunostaining analysis was performed as previously described (Yang et al., 2009; Dincman et al., 2012). Anti-mouse A2B5 IgM and anti-mouse O4 IgM (50%, v/v) were produced by hybridoma culture. Anti-mouse Olig2 (1:1000), anti-mouse GFAP (1:1000), and anti-rabbit neurofilament (1:1000) were purchased from Millipore. Anti-rat MBP (1:500) was obtained from Abcam, anti-rabbit β-tubulin (1:1000) from Sigma, and anti-rabbit PDGFRα (1:500) from Santa Cruz. The Alexa-488 or Alexa-594 conjugated secondary antibodies were purchased from Invitrogen. The nucleic acid dye 4<sup>0</sup> ,6-diamidino-2-phenylindole (DAPI) was obtained from Roche. All quantitative data are presented as means ± SD. Statistical significance of the difference was evaluated by Student's t-test. P < 0.05 was considered statistically significant.

# RESULTS

# Preparation of Mouse Primary OPCs from Cortical Tissues

When mouse cortical tissues were dissociated with trypsin and plated in PDL/LN-coated flasks with D/F20S medium, stratified cultures of OPCs and astrocytes were not formed as described in rat OPC preparations (**Figure 1B**) (Chen et al., 2007). However, when mouse cortical tissues were diced and plated directly without trypsinization, flat cells migrated out of tissue chunks and formed a bed layer, on top of which small surface cells appeared (**Figure 1A**). The vast majority of small surface cells displayed two or three fine cell processes (**Figure 1A**), characteristic of OPC morphology (Marinelli et al., 2016).

The surface OPCs can be physically separated by shake force from the bed layer of GFAP+flat cells (**Figure 1D**). After replated onto the PDL/LN-coated dishes in OPC basal media plus 10 ng/ml PDGFaa, OPCs displayed typical morphological features with fine processes (bipolar 42.75 ± 2.54%, tripolar 36.34 ± 7.54%, multipolar 20.09 ± 3.91%; **Figures 1C,E**). Immunofluorescent staining revealed a high percentage of these cells immunoreactive to A2B5, PDGFRα, and Olig2 (96.74 ± 5.77%, 95 ± 7.4%, and 93.08 ± 6.69%, respectively; **Figures 1F,G**), and less than 0.5% of them were GFAP+ (Supplementary Figure S1C; **Figure 1G**), indicating that the vast majority of them possessed the antigenic properties of OPCs (Dincman et al., 2012; Marinelli et al., 2016). However, a small percentage of OPCs expressed O4 antigen (11.05 ± 1.46%; Supplementary Figure S1A; **Figure 1G**) or mature OL marker MBP (∼1%; Nicolay et al., 2007) with multipolar morphology (Supplementary Figure S1B; **Figure 1G**). Neurofilament+ neurons were not detected in the culture (Supplementary Figure S1D) (Shaw et al., 1981). When OPCs were cultured in basal media without PDGFaa for 3 days, they appeared to differentiate into mature OLs rapidly, and O4 expression almost reached the peak around day 3, however, MBP expression is slower than O4 for about 1 day (**Figure 2A**;

Supplementary Figure S2). In contrast, when OPCs were cultured in the DMEM/F12 medium supplemented with 10% FBS, the vast majority of them differentiated into A2B5+/GFAP+ type-2 astrocytes by d6 (**Figure 2B**; Supplementary Figure S2). Together, these results indicated that mouse OPCs prepared from cortical tissues displayed typical O-2A antigenic characteristics and differentiation potentials (Dincman et al., 2012).

OPCs cultures, n = 5. Statistical analyses are presented as mean ± SD. Scale bars (A–C): 100 µm, (D,F): 50 µm.

# EGF Cooperates with PDGFaa to Promote Extended Division and Self-Renewal of Mouse Cortical OPCs

Although PDGFaa potently promotes proliferation of rat OPCs (Noble et al., 1988; Raff et al., 1988; Hart et al., 1989; Gard and Pfeiffer, 1993; Kleinsimlinghaus et al., 2013), it could not maintain the self-renewal and proliferation of mouse OPCs for a long time period even in the presence of bFGF or NT-3 (Supplementary Figure S3), suggesting that other important factors are also required for the sustained proliferation of mouse OPCs. The previous observation that EGFR overexpression expanded OPC pool in vivo (Aguirre et al., 2007) prompted us to examine the effect of EGF on the proliferation and maintenance of mouse cortical OPCs. Following 1-day EGF

treatment (10 ng/ml), more than half of cells (52.49 ± 2.06%) were found EDU-positive in the cultures, which is comparable to cultures treated with PDGFaa (56.03 ± 3.91%). However, the percentage of EDU-positive cells in EGF-treated cultures

decreased quickly thereafter, followed by an increase in MBP+ cells, indicating that EGF alone can stimulate the proliferation of mouse cortical OPCs, but can not prevent their differentiation. We next investigated whether EGF could work synergistically with PDGFaa to promote OPC proliferation and repress differentiation. When mouse cortical OPCs were exposed to PDGFaa alone or EGF + PDGFaa for 4 days, EGF was found to enhance the response of mouse cortical OPCs to PDGFaa to divide (**Table 1**). The cultures exposed to EGF + PDGFaa contained a higher percentage of EDU-positive cells than that treated with PDGFaa alone (P < 0.05). EGF also synergized with PDGFaa to inhibit expression of O4 antigen in OPCs, and the percentage of O4+ cells in EGF + PDGFaa treated cultures was reduced by half compared with that exposed to PDGFaa only. At the same time, EGF cooperated with PDGFaa to inhibit OPC differentiation into mature OLs, as the percentage of MBP+ cells in the presence of EGF + PDGFaa was about 4%, as compared to 10.7% for PDGFaa treatment (P < 0.05).

Basic fibroblast growth factor was previously reported to amplify the effect of PDGFaa in promoting DNA synthesis and inhibiting differentiation of rat OPCs (Tang et al., 2000). We found that bFGF and EGF had similar effects in synergizing with PDGFaa to promote mouse OPC proliferation and inhibit their differentiation in parallel experiments. More importantly, EGF, bFGF, and PDGFaa had additive effects in enhancing cell proliferation and reducing differentiation (**Table 1**), suggesting that the proliferation and differentiation of mouse cortical OPCs are regulated by multiple signaling pathways.

# A Mixture of OPCs and GPCs Were Induced by Stimulation of EGF + bFGF + PDGFaa

Although bFGF + PDGFaa and EGF + PDGFaa promoted extended division and inhibited differentiation of mouse cortical OPCs as described above, primary mouse OPCs can only be maintained in vitro for a limited period of time. With time, cell divisions slowed down and ceased, and eventually all OPCs underwent apoptosis. However, when the isolated OPC cells were exposed to these three growth factors simultaneously, the earlystage OPCs with bipolar or tripolar morphology maintained strong proliferation ability, while the late-stage multipolar ones gradually died and then disappeared (**Figure 3A**). Unexpectedly, a few EGF-responsive fibroblast-like cells started to appear in primary OPC cultures (**Figure 3A**) and EGF has strong synergistic effect with bFGF + PDGFaa tostimulate division of the mixed culture of OPCs and GPCs (**Figures 3B,C**). Clonal analysis showed that EGF + bFGF + PDGFaa can change the morphology of OPCs, a mixture of bipolar or tripolar OPCs and fibroblast-like cells can be derived from single OPC cell in the presence of these three factors (**Figure 4**). In contrast, bFGF + PDGFaa kept OPC the typical bipolar or tripolar morphology (**Figure 4**).

TABLE 1 | Effects of Platelet Derived Growth Factor-AA (PDGFaa), basic fibroblast growth factor (bFGF), or/and epidermal growth factor (EGF) on oligodendrocyte precursor cell (OPC) proliferation and differentiation in vitro.


Data are presented as mean ± SD, n = 3.

FIGURE 3 | Epidermal growth factor (EGF) + basic fibroblast growth factor (bFGF) + Platelet Derived Growth Factor-AA (PDGFaa) promoted the proliferation of mixed culture of OPCs and glial progenitor cells (GPCs). (A) A mixed cultures of GPCs (white arrow) and early-stage OPCs derived from EGF + bFGF + PDGFaa stimulation, while late-stage multipolar OPCs gradually died (black arrow). (B) EGF has strong synergistic effect with bFGF + PDGFaa to stimulate division of the mixed culture of OPCs and GPCs. (C) The percentage of EDU+ cells in bFGF + PDGFaa and EGF + bFGF + PDGFaa treated mixed culture of OPCs and GPCs, n = 6. Abbreviation: F+P, PDGFaa + bFGF; E+F+P, EGF + bFGF + PDGFaa. Statistical analyses are presented as mean ± SD, n = 3. ∗∗P < 0.01, Scale bars (A) 100 µm, (B) 50 µm.

To determine the identity of the induced flat cells, we analyzed 50 single clones derived from fibroblast-like cells by immunostaining and directed differentiation. It was found that these fibroblast-like cells were immunoreactive to OPC markers including Olig2, A2B5, and PDGFRα (**Figure 5B**), but negative for GFAP, O1, O4, and neurofilament (Supplementary Figure S4). In the presence of FBS, 29 clones produced A2B5+/GFAP+ type-2 astrocytes; in the parallel cultures, T3 promoted them to differentiate into MBP+ OLs. The other 21 clones (**Figure 5A**) mainly yielded A2B5-GFAP+ and A2B5+GFAP+ astrocytes in the presence of FBS and CNTF + bFGF, respectively, and MBP+ OLs were detected in T3 culture (**Figure 5C**), indicating that these clones were less committed to oligodendrocyte lineage and displayed the same developmental potentials as the tripotential GPCs (Gregori et al., 2002; Dadsetan et al., 2009; Haas et al., 2012). Intriguingly, GPC-conditioned medium significantly promoted the proliferation of OPCs and reduced apoptosis of them (**Figure 5D**), suggesting that GPCs may provide some unknown nourishments to OPCs to enhance their proliferation and self-renewal in the mixed cultures.

#### OPCs Can Be Amplified from the Mixture of OPCs and GPCs

EGF signaling promoted the emergence of fibroblast-like GPCs in primary OPC culture; however, these EGF-responsive GPCs were difficult to survive the absence of EGF. When EGF was withdrawn, the division of GPCs slowed down and underwent apoptosis despite the presence of bFGF and PDGFaa (**Figures 6A–C**), resulting in a continuous decline of the fibroblast-like cell population in the mixed culture. Interestingly, fibroblast-like OPCs derived from EGF + bFGF + PDGFaa treatment reversed to bipolar or tripolar morphology after EGF withdrawal (**Figure 7A**). One possible reason for the disappearance of fibroblast-like cells in the mixed cultures following EGF withdrawal could be due to the apoptosis of GPCs and the morphological changes of OPCs.

OPCs continued to proliferate for a period of time even when fibroblast-like cells were difficult to find, and they continued to display typical bipolar or tripolar morphology and immunoreactivity to Olig2, A2B5, PDGFRα (**Figure 7B**). O4+ cells were not observed in the culture, indicating that they remained as early-stage OPCs. In addition, these OPCs were more dependent on T3 for oligodendrocyte differentiation than the initially isolated OPCs. When OPCs were cultured in basal media plus T3 for 3 days, multiple and highly branched processes grew out rapidly until typical oligodendrocyte morphology

emerged (**Figure 8A**). In the absence of T3, the processes were much shorter or less branched (**Figure 8A**). Immunological analysis confirmed the higher percentage of O4+ and MBP+ cells (80.89 ± 17.81% and 58.93 ± 11.69%, respectively) in T3 differentiation culture, as compared to the culture without T3 (29.59 ± 8.24% and 4.65 ± 2.07%, respectively; **Figures 8B,C**).

To test whether mouse OPCs derived from the mixed culture retained their capacity to myelinate axons, we co-cultured them with embryonic DRGs (Chan et al., 2004) for 2 weeks. Double immunostaining experiments demonstrated that these OPCs gave rise to MBP+ OLs, which formed myelin sheaths around the neurofilament+ axons (**Figure 9**). These results indicated that mouse cortical OPCs maintained their biological characteristics after amplification in vitro.

#### DISCUSSION

#### Species Difference in OPC Isolation and Amplification

Purification and culture of mouse primary OPCs have been a technical challenge (Chen et al., 2007), as they cannot be obtained and maintained by the simple dissociation culture of brain tissues used for preparation of rat OPCs. As a result, many studies are forced to use rat OPCs as the alternative sources to complement their genetic studies in transgenic and mutant

(B) 50 µm.

mouse models (Chew et al., 2011; Liu et al., 2012; Chen et al., 2013; Zhu et al., 2014). The mechanisms underlying the observed species difference in OPC isolation are currently unknown. One possibility is that mouse OPCs are particularly sensitive to enzymatic digestion and fail to survive the treatments. Use of small cortical tissues can avoid the damage from enzyme digestions. Moreover, small tissue blocks may provide OPCs an environment mimicking their in vivo niche, and help them to adapt to the culture conditions and maintain self-renewal. In addition, astrocytes that migrate out of cortical tissues could serve as feeder cells for OPCs and provide favorable environment for OPC migration and proliferation. Astrocyte layers may actively secrete some cytokines that promote OPC survival and proliferation. Interestingly, although rat cortical OPCs can be prepared by dissociation culture of nerve cells, the majority of them are O4+ with multipolar morphology (Bansal et al., 1992). It is plausible that cytokines secreted by rat astrocytes may be sufficient to support the survival of rat OPCs, but ineffective to inhibit their differentiation.

#### EGF Signaling Promotes the Proliferation of Mouse OPCs

Epidermal growth factor was previously found to be a mitogen for neural stem cells (NSCs; Deleyrolle et al., 2006; Kanakasabai et al., 2012). In this study, we observed EGF responsiveness in committed mouse glial precursors including OPCs and GPCs. Although EGF is effective in stimulating OPC proliferation, it did not inhibit cell differentiation and OPCs gradually differentiated into MBP+ OLs in EGF treatment alone. PDGFaa inhibits OPC differentiation more effectively than EGF, but PDGFaa

FIGURE 9 | Confocal image of myelinated axons in OPC/DRG co-culture. Purified OPCs matured into MBP+ oligodendrocytes (OLs) in the co-culture, and aligned with axons marked by neurofilament (NF1) staining. Scale bars: 10 µm.

alone was not sufficient to sustain self-renewal of mouse O4− early-stage OPCs. Interestingly, EGF and PDGFaa exhibited a synergistic effect in preventing the differentiation of O4− OPCs, analogous to the effect of bFGF and PDGFaa in rat OPCs (Tang et al., 2000). Thus, the influence of EGF on OPCs is not restricted to proliferation, it also regulates OPC differentiation by collaborating with other factors. This is in keeping with the recent observation that EGFR signaling promotes oligodendrocyte proliferation after injuries (Scafidi et al., 2014).

# EGF Signaling Promotes GPC Development

In addition, the EGF-dependent proliferation and survival were observed for mouse postnatal cortical GPCs in primary OPC culture, which means EGF signaling plays an important role in GPC development. Certainly, it can not be excluded that the degree of EGF-dependence may be different for GPCs from different regions and developmental stages. Formerly it was a challenge to maintain a homogeneous culture of GPCs for prolonged periods in culture, and immortalization by constitutive expression of oncogene was forced to be adopted for obtaining a clonal cell line of GPC (Wu et al., 2002). We demonstrated that the single cell derived GPC clones can maintain an undifferentiated dividing state for a long time in presence of EGF + bFGF + PDGFaa in vitro. To our surprise, GPC conditioned medium contributed to the dividing progenitor state of mouse cortical OPCs, indicating that some unknown cytokines secreted by GPCs are beneficial for the survival and self-renewal of OPCs in vitro. Since OPCs can exist in vivo throughout life span, similar factors may be secreted by the surrounding cells in the nervous tissue (El Waly et al., 2014). Elucidation of these cytokines shall contribute to our further understanding of molecular mechanisms governing the survival, division, and differentiation of mouse OPCs.

## EGF Signaling Promotes a Juvenile Stage of Mouse OPCs

The cooperation of EGF with bFGF + PDGFaa promoted morphological change of early-stage OPCs into fibroblastlike cells resembling GPCs in shape. These GPC-like OPCs

proliferated quickly with a low apoptosis rate, and they seem to represent OPCs at an early or young stage. Despite their morphological similarities to GPCs, they still retained the oligodendrocyte differentiation potential, and were not fully converted to GPCs. Based on its function in maintaining proliferation and self-renewal of NSCs, GPCs, and earlystage OPCs (Deleyrolle et al., 2006; Kanakasabai et al., 2012), it is postulated that EGF plays an important role in stem cell maintenance and oligodendrocyte development. High concentration of EGF maintains the proliferation and selfrenewal of stem cells or progenitor cells which possess potential to differentiate into OLs, and low concentration promotes their differentiation (Scafidi et al., 2014).

In summary, we reported a novel approach for the isolation, amplification, and purification of mouse OPCs by utilizing the EGF dependence of GPCs and the nourishing effects of GPCs on OPCs. High purity of OPCs can be obtained by withdrawing EGF from the mixed cultures of GPCs and OPCs. In addition, GPCs/OPCs mixtures can tolerate cryopreservation, and they can be stored after amplification for later use, avoiding repeated preparations of OPCs from central nervous tissues.

#### REFERENCES


# AUTHOR CONTRIBUTIONS

JY designed, performed experiments, collected, analyzed the data, and contributed to writing the manuscript; XC, JS, BX, and XZ performed experiments; QC, ZZ, and YS analyzed data; MQ designed, supervised the experiments, collected, analyzed, and discussed data, and wrote the manuscript.

#### ACKNOWLEDGMENTS

This work is supported by the National Key Basic Research Program of China (2013CB531303; 2012CB910402), and National Natural Science Foundation of China (81200961; 31372150; 31572224).

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fncel. 2016.00203



precursor cells generate different tumor phenotypes in response to the identical oncogenes. J. Neurosci. 33, 16805–16817. doi: 10.1523/JNEUROSCI.0546- 13.2013


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Yang, Cheng, Shen, Xie, Zhao, Zhang, Cao, Shen and Qiu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Sustained HSP25 Expression Induces Clasmatodendrosis via ER Stress in the Rat Hippocampus

Ji-Eun Kim , Hye-Won Hyun , Su-Ji Min and Tae-Cheon Kang \*

Department of Anatomy and Neurobiology, Institute of Epilepsy Research, College of Medicine, Hallym University, Chuncheon, South Korea

Heat shock protein (HSP) 25 (murine/rodent 25 kDa, human 27 kDa) is one of the major astroglial HSP families, which has a potent anti-apoptotic factor contributing to a higher resistance of astrocytes to the stressful condition. However, impaired removals of HSP25 decrease astroglial viability. In the present study, we investigated whether HSP25 is involved in astroglial apoptosis or clasmatodendrosis (autophagic astroglial death) in the rat hippocampus induced by status epilepticus (SE). Following SE, HSP25 expression was transiently increased in astrocytes within the dentate gyrus (DG), while it was sustained in CA1 astrocytes until 4 weeks after SE. HSP25 knockdown exacerbated SE-induced apoptotic astroglial degeneration, but mitigated clasmatodendrosis accompanied by abrogation of endoplasmic reticulum (ER) stress without changed seizure susceptibility or severity. These findings suggest that sustained HSP25 induction itself may result in clasmatodendrosis via prolonged ER stress. To the best of our knowledge, the present study demonstrates for the first time the double-edge properties of HSP25 in astroglial death induced by SE.

# Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Rafael Linden, Federal University of Rio de Janeiro, Brazil Daniela Tropea, Trinity College, Dublin, Ireland

#### \*Correspondence: Tae-Cheon Kang

tckang@hallym.ac.kr

Received: 06 July 2016 Accepted: 13 February 2017 Published: 22 February 2017

#### Citation:

Kim J-E, Hyun H-W, Min S-J and Kang T-C (2017) Sustained HSP25 Expression Induces Clasmatodendrosis via ER Stress in the Rat Hippocampus. Front. Cell. Neurosci. 11:47. doi: 10.3389/fncel.2017.00047 Keywords: heat shock protein B1, status epilepticus, astrocyte, autophagy, apoptosis, ER stress

# INTRODUCTION

Astrocyte plays an important role in the maintenance of extracellular ion homeostasis and neuronal functionality in the brain. Since brain injury induces reactive astrogliosis representing astroglial hypertrophy, proliferation and migration, it is a general concept that astrocytes are invulnerable to various harmful stresses. However, a growing body of evidence indicates astroglial damage before or after neuronal damage and reactive astrogliosis (Sugawara et al., 2002; Borges et al., 2006; Kang et al., 2006; Kim D. S. et al., 2008; Kim J. E. et al., 2008; Kim et al., 2014a).

Status epilepticus (SE) is defined as continuous unremitting seizure activity longer than 5 min. SE is the most extreme form of seizure with significant mortality, and causes severe brain damage that may involve epileptogenesis (Kang et al., 2006; Kim J. E. et al., 2008; Trinka et al., 2012). Interestingly, SE evokes three different patterns of astroglial death in the rat brain independent of hemodynamics (Kang et al., 2006; Kim J. E. et al., 2008; Kim et al., 2010a,b, 2014a; Ryu et al., 2011a,b). One is focal necrotic astroglial degeneration detected in the piriform cortex accompanied by severe vasogenic edema 1–2 days after SE (Kim et al., 2013, 2014b), similar to other brain regions (Ingvar et al., 1994; Schmidt-Kastner and Ingvar, 1994; Gualtieri et al., 2012). Another is terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) positive astroglial apoptosis, which is observed in the molecular layer (not the hilus) of the dentate gyrus (DG; Kang et al., 2006; Kim D. S. et al., 2008; Kim et al., 2010b, 2015). The other is clasmatodendrosis that is observed in the CA1 region (Kang et al., 2006; Kim D. S. et al., 2008; Kim et al., 2010b; Ko et al., 2016). Clasmatodendrosis is characterized by extensive swollen vacuolized cell bodies and disintegrated/beaded processes, which is first reported by Alzheimer (1910), and termed ''clasmatodendrosis'' by Cajal (Penfield, 1928). In addition, clasmatodendritic astrocyte has round-shaped edematous cell body, short blunt processes, GFAP tangles in the cytoplasm and nuclear dissolution. This irreversible astroglial change indicates impairment of astroglial functions. Vacuolization and eosinophilic cytoplasm in clasmatodendritic astrocytes make clasmatodendrosis be considered as coagulative necrotic astroglial death (Sugawara et al., 2002; Kim D. S. et al., 2008; Kim et al., 2009). However, we have reported that vacuolization in clasmatodendritic astrocytes is relevant to up-regulation of lysosome-associated membrane protein 1 (LAMP1), which is required for the essential activation of autophagy. Thus, clasmatodendrosis is an autophagic astroglial death (Ryu et al., 2011a,b). However, less defined are the molecular information available to clasmatodendrosis, although the distinct mechanisms or the heterogeneous properties of astrocytes seem to be relevant to clasmatodendrosis in the CA1 region.

Heat shock protein (HSP) 25 (murine/rodent 25 kDa, human 27 kDa) constitutes one of the major astroglial HSP families and plays a protective role against cell death (Cuesta et al., 2000; Bidmon et al., 2008; Lively and Brown, 2008). Therefore, HSP25 induction provides insight into the astroglial invulnerability in response to harmful insults. Furthermore, HSP25 induction is an indicative of the early astroglial responses including energy-consuming protein synthesis (Cuesta et al., 2000; Kirschstein et al., 2012). Thus, HSP25 is one of the highly sensitive and reliable molecules of full development of SE (Kirschstein et al., 2012). Recently, we have reported that endoplasmic reticulum (ER) stress induced by accumulation of unfolded or misfolded protein aggregates (Bernales et al., 2006) is relevant to astroglial apoptosis and clasmatodendrosis following SE (Ko et al., 2015). Because HSP25 binds to misfolded proteins and inhibits protein aggregation (Haslbeck, 2002; Richter-Landsberg, 2007), these findings had encouraged us to speculate that HSP25 induction would protect astrocytes from SE-induced ER stress. Unexpectedly, in the present study we found that SE-induced prolonged HSP25 expression resulted in ER stress and subsequently provoked clasmatodendrosis in the CA1 region in vivo. Thus, we demonstrate for the first time that sustained HSP25 induction itself may play a pro-autophagic factor in astrocytes via prolonged ER stress.

# MATERIALS AND METHODS

#### Experimental Animals and Chemicals

This study utilized the progeny of male Sprague-Dawley (SD) rats (7 weeks old) obtained from Experimental Animal Center, Hallym University (Chuncheon, South Korea). The animals were provided with a commercial diet and water ad libitum under controlled temperature, humidity and lighting conditions (22 ± 2 ◦C, 55 ± 5% and a 12:12 light/dark cycle with lights). Experimental procedures were approved by the Institutional Animal Care and Use Committee of the Hallym university (Chuncheon, South Korea). The number of animals used and their suffering was minimized in all cases. All reagents were obtained from Sigma-Aldrich (USA), unless otherwise noted.

# Surgery and HSP25 Knockdown

Animals were anesthetized with 1–2% Isoflurane in O<sup>2</sup> and placed in a stereotaxic frame. Animals were then implanted with a brain infusion kit 1 (Alzet, Cupertino, CA, USA) into the right lateral ventricle (1 mm posterior; 1.5 mm lateral; −3.5 mm depth). The infusion kit was sealed with dental cement and connected to an osmotic pump (1007D, Alzet, Cupertino, CA, USA) containing control siRNA or HSP25 siRNA. For knockdown of HSP25, we applied a set of four on-target HSP25 rat siRNAs and a non-targeting control was used in the preliminary study. Of four on-target siRNAs, a siRNA sequence corresponding to coding region (sense 5- GGAACAGUCUGGAGCCAAGUU-3; antisense 5-CUUGGCU CCAGACUGUUCCUU-3) was selected as the best probe (reduction efficiency: 43.2%, Genolution Pharmaceuticals, Inc., South Korea), and used for the final experiments. A siRNA sequence coding region 5-GCAACUAACUUCGUUAGAA UCGUUAUU-3 was used as a non-targeting control siRNA. The pump was placed in a subcutaneous pocket in the dorsal region. Animals received 0.5 µl/h of saline or (100 µM in saline) for 1 week (Kim et al., 2011, 2014b). Some animals were also implanted with monopolar stainless steel electrode (Plastics One, Roanoke, VA, USA) into the left dorsal hippocampus (−3.8 mm posterior; 2.0 mm lateral; −2.6 mm depth). Connecting wire and electrode socket were then inserted in an electrode pedestal (Plastics One, Roanoke, VA, USA), secured to the exposed skull with dental acrylic. Three days after surgery, rats were induced SE by LiCl-pilocarpine.

# SE Induction and EEG Analysis

Animals were given LiCl (3 mEq/kg, i.p) 24 h before the pilocarpine treatment. To reduce the peripheral effects of pilocarpine, atropine methylbromide (5 mg/kg, i.p.) was also received 20 min before the pilocarpine treatment. To induce SE, animals were treated with pilocarpine (30 mg/kg, i.p.). Control animals received saline in place of pilocarpine. For evaluation of the effect of siRNA knockdown on seizure susceptibility in response to LiCl-pilocarpine, animals were recorded EEG signals with a DAM 80 differential amplifier (0.1–3000 Hz bandpass; World Precision Instruments, Sarasota, FL, USA). EEG activity was measured during the 2 h recording session from each animal. The data were digitized (400 Hz) and analyzed using LabChart Pro v7 (AD Instruments, Bella Vista, NSW, Australia). Time of seizure onset was defined as the time point showing paroxysmal depolarizing shift, which lasted more than 3 s and consisted of a rhythmic discharge between 4 and 10 Hz with amplitude of at least two times higher than the baseline EEG (Kim and Kang, 2011). EEG activity was measured during the 2 h recording session from each animal. Spectrograms were also automatically calculated using a Hanning sliding window with 50% overlap. Two hours after SE onset, diazepam (Valium; Roche, France; 10 mg/kg, i.p.) was administered and repeated, as needed.

#### Tissue Processing

Animals were perfused transcardially with phosphate-buffered saline (PBS, pH 7.4) followed by 4% paraformaldehyde in 0.1 M phosphate buffer (PB, pH 7.4) under urethane anesthesia (1.5 g/kg i.p.). The brains were removed, postfixed in the same fixative for 4 h and rinsed in PB containing 30% sucrose at 4◦C for 2 days. Thereafter the tissues were frozen and sectioned with a cryostat at 30 µm and consecutive sections were collected in six-well plates containing PBS. For western blot (WB), animals were decapitated under urethane anesthesia. The hippocampus was rapidly removed and homogenized in lysis buffer. The protein concentration in the supernatant was determined using a Micro BCA Protein Assay Kit (Pierce Chemical, Rockford, IL USA).

#### Double Immunofluorescence Study

**Table 1** is a list of the primary antibodies used in the present study. Sections were incubated in a mixture of antisera in PBS containing 0.3% Triton X-100 overnight at room temperature. After washing, sections were incubated in a mixture of FITC- and Cy3-conjugated IgG (or streptavidin, Jackson Immunoresearch Laboratories Inc., West Grove, PA, USA; diluted 1:250) for 2 h at room temperature. Images were captured using an AxioImage M2 microscope. Fluorescent intensity was measured using computer-based image analysis program (AxioVision Rel. 4.8 software, Germany). Fluorescent intensity was then standardized by setting the threshold level (mean background intensity obtained from five image input). Manipulation of the images was restricted to threshold and brightness adjustments to the whole image.

#### TUNEL Staining

TUNEL staining was performed with the TUNEL apoptosis detection kit (Merck Millipore, Billerica, MA, USA) according to the manufacturer's instructions. Following TUNEL reaction, GFAP immunofluorescence (IF) staining was performed. For nuclei counterstaining, we used Vectashield mounting medium with DAPI (Vector, Torrance, CA, USA).

# Cell Counts

For quantification of immunohistochemical data, cells in 2–4 regions (1 × 10<sup>5</sup> µm<sup>2</sup> ) from each section were counted on 20× images. Results are presented as means ± SD of 15–24 regions from seven animals. All immunoreactive cells were counted regardless of the intensity of labeling. Cell counts were performed by two different investigators who were blind to the classification of tissues.

#### Western Blot

Aliquots containing 20 µg total protein were loaded into a polyacrylamide gel. After electrophoresis, gels were transferred to nitrocellulose transfer membranes (Schleicher and Schuell BioScience Inc., Keene, NH, USA). To reduce background staining, membranes were incubated with 5% nonfat dry milk in Tris-buffered saline containing 0.1% Tween 20 for 45 min, followed by incubation with the primary antibody (**Table 1**), and subsequently with an HRP-conjugated secondary antibody. Immunobands were detected with an ECL Western Blotting Detection Kit (Amersham, Piscataway, NJ, USA). Intensity measurements were represented as the mean gray-scale value on a 256 gray-level scale (Kim and Kang, 2011).

## Quantification of Data and Statistical Analysis

All data obtained from the quantitative measurements were analyzed using Mann-Whitney or Kruskal-Wallis test to determine statistical significance. Bonferroni's test was used for post hoc comparisons. A p-value below 0.05 was considered statistically significant (Kim and Kang, 2011; Kim et al., 2014a).

# RESULTS

#### Regional Specific HSP25 Induction in the Hippocampus Following SE

**Figure 1A** shows that HSP25 band was detected from 12 h post-SE animals (p < 0.05 vs. non-SE animal). HSP25 protein expression increased until 3 days after SE (p < 0.05 vs. non-SE animal), and gradually decreased 1 and 4 weeks after SE (p < 0.05 vs. 3 days after SE; **Figures 1A,B**). Immunohistochemical study revealed that HSP25 expression was observed in astrocytes within the all hippocampal region 3 days after SE. One week after SE, HSP25 expression was reduced in the CA3 field and the DG (p < 0.05 vs. 3 days after SE; **Figures 1C,D**). Thus, HSP25 expression was mainly observed in astrocytes within the stratum radiatum of the CA1 field. Four weeks after SE, HSP25 expression was also declined in the CA1 field (p < 0.05 vs. 3 days after SE; **Figures 1C,D**). These findings indicate that HSP25 expression showed the distinct spatio-temporal specific pattern in the hippocampus following SE.

# Astroglial Death and HSP25 Induction in the Hippocampus Following SE

In the molecular layer of the DG, 50% of astrocytes showed HSP25 expression 3 days after SE (**Figures 2A,B**). One week after


IF, Immunofluorescence; WB, Western blot.

Four weeks after SE, HSP25 is also decreased in the CA1 region. Bar = 300 µm. Arrows indicate the hippocampal fissure (the upper border of the molecular layer of the DG). (D) Quantification of HSP25 fluorescent intensity in the hippocampus following SE (mean ± SEM; n = 7, respectively). <sup>∗</sup>p < 0.05 vs. 3 days after SE.

SE, total number of astrocytes was reduced as compared to 3 days after SE (p < 0.05, respectively; **Figures 2A,B**). The fraction of HSP25 positive astrocytes in total astrocytes was also declined to 9.3% in this region (p < 0.05, respectively; **Figures 2A,B**). Four weeks after SE, the fraction of HSP25 positive astrocytes in total astrocytes was decreased to 4.3%, since the total number of

astrocytes was increased as compared to 3 days after SE (p < 0.05; **Figures 2A,B**).

In the CA1 field, about 70% of astrocytes showed HSP25 expression 3 days after SE (**Figures 3A,B**). One week after SE, the fraction of HSP25 positive astrocytes in total astrocytes was 48% of astrocytes (**Figures 3A,B**). However, total number of astrocytes was unaltered. Four weeks after SE, total number of astrocytes was reduced in this region, as compared to 3 days after SE (p < 0.05; **Figures 3A,B**). The fraction of HSP25 positive astrocytes in total astrocytes was also decreased to 19% (p < 0.05 vs. 3 days after SE; **Figures 3A,B**). Interestingly, HSP25 was accumulated in clasmatodendritic astrocytes, which had round-shaped edematous cell body, short blunt processes, loss of distal processes, vacuoles, GFAP tangles (aggregation) and nuclear dissolution (absence of nucleus or watery pale nuclear staining), more than that in reactive astrocytes (**Figure 3A**). These findings indicate that the distinct HSP25 induction may be relevant to the different patterns of SE-induced astroglial loss in the hippocampus.

## Effect of HSP25 Knockdown on Seizure Susceptibility in Response to Pilocarpine

To investigate the role of HSP25 in SE-induced astroglial responses, we applied HSP25 knockdown before SE induction. Control- or HSP25 siRNA-infusion could not induce neurotoxicity (hind-limb paralysis, vocalization, food intake, seizure or neuroanatomical damage), indicating that both siRNA did not affect brain activity under normal condition. After SE induction, the first behavioral seizure occurred 32.1 and 34.3 min in control siRNA- and HSP25 siRNAinfused animals, respectively. In addition, both control siRNAand HSP25 siRNA-infused animals showed episodes of high-amplitude and high-frequency discharges representing typical SE. An EEG analysis revealed no difference in the latency of seizure on-set and total EEG power during SE between control siRNA- and HSP25 siRNA-infused animals (**Figures 4A–C**). As shown in **Figure 4D**, HSP25 siRNA resulted in an approximate 43.2% reduction of HSP25 protein expression level following SE, as compared with control siRNA (p < 0.05). These findings indicated that HSP25 siRNA effectively inhibited SE-induced HSP25 induction without alterations in seizure susceptibility or severity in response to pilocarpine.

#### Protective Role of HSP25 in Astroglial Apoptosis in the Molecular Layer of the Dentate Gyrus

Next, we investigated the effect of HSP25 knockdown on SE-induced astroglial apoptosis in the DG using TUNEL staining. Consistent with our previous studies (Kang et al., 2006; Kim et al., 2010b), control siRNA-infused animals showed apoptotic astroglial degeneration in the DG 1 week after SE (**Figures 5A–C**). HSP25 siRNA exacerbated astroglial loss in this region, and increased the number of TUNEL positive astrocytes (p < 0.05 vs. control siRNA infusion; **Figures 5A–C**). Thus, it is likely that HSP25 may play an anti-apoptotic role in astrocytes within the DG.

## Protective Effect of HSP25 Knockdown on Clasmatodendritic Astroglial Loss in the CA1 Region

Since up-regulations of LAMP1 and LC3-II represent clasmatodendrosis (Ryu et al., 2011a,b), we applied double immunofluorescent studies for GFAP and LAMP1/LC3-II to evaluate the effect of HSP25 siRNA on autophagic astroglial death in the CA1 region. Four weeks after SE, control siRNAinfused animals showed clasmatodendrosis in the stratum radiatum of the CA1 region (**Figure 6A**). Consistent with our previous studies (Ryu et al., 2011a,b), clasmatodendritic astrocytes showed LAMP1-positive vacuolization and up-regulated LC3-II expression (**Figures 6B,C**). HSP25 siRNA effectively abrogated LAMP1-positive vacuolization and LC3-II over-expression. In addition, HSP25 siRNA prevented the decrease in the total number of astrocytes (p < 0.05 vs. control siRNA; **Figures 6A–E**). These findings indicate that sustained HSP25 expression may contribute to SE-induced autophagic astroglial death in the CA1 region.

## Reduced SE-Induced Astroglial ER Stress by HSP25 Knockdown

Protein kinase RNA (PKR)-like ER kinase (PERK) activation mediates the initial ER stress response by phosphorylating eukaryotic initiation factor 2α (eIF2A). In turn, the activation of the PERK/elF2A leads to ER stress-induced cell death via CCAAT/enhancer-binding protein homologous protein (CHOP)-dependent apoptosis (Moreno et al., 2012; Moreno and Tiffany-Castiglioni, 2015) or CHOP-independent autophagy (Bernales et al., 2006; Ito et al., 2009; Matsumoto et al., 2013). Recently, we have reported that clasmatodendritic astrocytes contained phospho-PERK (pPERK) and phospho-eIF2A (peIF2A) immunoreactivities without CHOP expression, and that up-regulated calnexin (CNX) expression is associated to the activation of autophagy leading to clasmatodendrosis (Ryu et al., 2011a,b; Ko et al., 2015). Therefore, it is likely that sustained HSP25 expression in CA1 astrocytes may be relevant to clasmatodendrosis via ER stress-mediated autophagy. To confirm this hypothesis, we investigated the effect of HSP25 knockdown on astroglial ER stress induced by SE. In non-SE animals, pPERK expression was rarely observed in CA1 astrocytes (**Figure 7A**). Following SE, clasmatodendritic astrocytes showed pPERK and peIF2A immunoreactivity, which was attenuated by HSP25 siRNA infusion (p < 0.05 vs. control siRNA; **Figures 7A,B**). In addition, most of clasmatodendritic astrocytes showed strong CNX expression without CHOP, protein-disulfide isomerase (PDI) or glucose-regulated protein 78 (GRP78) expressions (**Figures 7C**, **8A**). HSP25 knockdown effectively decreased CNX expression in astrocytes as well as clasmatodendritic astroglial death (p < 0.05 vs. control siRNA infusion; **Figures 8A,B**). These findings indicate that accumulation of HSP25 protein may induce autophagic astroglial death via prolonged ER stress.

#### DISCUSSION

The major novel findings in the present study are that HSP25 regulates regional specific astroglial death in the hippocampus following SE. Briefly, HSP25 induction played an anti-apoptotic role in astrocytes within the DG, but initiated ER stress-mediated autophagic astroglial death in the CA1 region.

CCAAT/enhancer-binding protein homologous protein (CHOP), protein-disulfide isomerase (PDI), glucose-regulated protein 78 (GRP78) and GFAP in the CA1 region. Clasmatodendritic astrocytes do not show CHOP, PDI and GRP78 expressions Bar = 12.5 µm.

HSP25 is the main chaperone in astrocytes (FitzGerald et al., 2007), which plays a role in blockade of mitochondrial damage and apoptosis in naïve astrocytes (Bruey et al., 2000; Parcellier et al., 2003). In the present study, HSP25 induction was observed in the hippocampus 3 days after SE. These findings are consistent with previous studies demonstrating astroglial HSP25 induction in various SE models and temporal lobe epilepsy patients (Kato et al., 1999; Erdamar et al., 2000; Bidmon et al., 2004, 2008; Kirschstein et al., 2012). Unexpectedly, we found that HSP25 induction showed regional specific patterns in the distinct hippocampal region. Astroglial HSP25 expression was transiently induced in the DG, while it was sustained in the CA1 region. With respect to the distinctive properties of astrocytes independently of hemodynamics (Kim et al., 2014a), these findings indicate that the distinct capability of HSP25 induction may represent the differential properties of astrocytes in the hippocampus.

In the present study, 50% of astrocytes showed HSP25 expression in the DG 3 days after SE. One week after SE, apoptotic astroglial degeneration was accompanied by reduction in HSP25 expression in this region, which was exacerbated by HSP25 siRNA infusion. These findings indicate that HSP25 induction may contribute to astroglial invulnerability to apoptosis-inducing stresses. In contrast to the DG, the present study demonstrates that clasmatodendritic astrocytes showed sustained HSP25 expression 4 weeks after SE. Clasmatodendrosis is the result from energy failure and acidosis coupled to mitochondrial inhibition in Alzheimer disease and brain ischemia (Friede and van Houten, 1961; Kraig and Chesler, 1990; Hulse et al., 2001). In epilepsy, we have speculated that clasmatodendrosis might be relevant to extracellular acidosis during seizure activity (Kim et al., 2009, 2011). Indeed, conventional anti-epileptic drugs inhibit clasmatodendrosis in the CA1 region (Kim D. S. et al., 2008). Thus, it is likely that clasmatodendrosis may be a consequence of prolonged recurrent seizures causing sustained HSP25 induction. Since astroglial loss increases neuronal excitability in the hippocampus (Kang et al., 2006; Kim D. S. et al., 2008), clasmatodendrosis may increase excitatory output from the CA1 region and subsequently drive synchronous epileptiform discharges to the other brain regions.

Autophagy, originally described as a stress response to nutrient deprivation, is an important catabolic route for bulk degradation of aberrant organelles and protein aggregates by the formation of double-membrane vesicles known as autophagosomes, which ultimately fuse with lysosomes (autolysosomes) for degradation of contents by lysosomal proteases. However, excessive or unquenched autophagic process leads to non-apoptotic programmed cell death (type II programmed cell death) independent of caspase activity (Bursch et al., 1996; Tsujimoto and Shimizu, 2005; Gozuacik and Kimchi, 2007). In the present study, both LC3-II and LAMP1 intensities in HSP25-positive clasmatodendritic astrocytes were higher than those in reactive astrocytes, which were effectively abrogated by HSP25 siRNA infusion. Since up-regulations of LAMP1 and LC3-II expression represent the formations of autophagosomes and autolysosomes (Ryu et al., 2011a,b), these findings indicate that sustained HSP25 expression may lead to clasmatodendrosis in the CA1 region. Therefore, our findings indicate that HSP25 may be one of the pro-autophagic factors in CA1 astrocytes.

How does sustained HSP25 expression induce autophagic astroglial death? Recently, we have reported that ER stress is closely relevant to clasmatodendrosis. During ER stress, aggregated proteins segregate unfolded molecules and inhibit their interaction with other cellular components (Sherman and Goldberg, 2001). In addition, accumulation of aggregated proteins induces HSP25 to prevent further aggregation (Lee and Goldberg, 1998; Haslbeck, 2002; Goldbaum and Richter-Landsberg, 2004; Richter-Landsberg, 2007). Therefore, HSP25 protects astrocytes against harmful stress by non-native proteins (Goldbaum et al., 2009). However, impaired clearance of HSP25 reduces cell viability in astrocytes (Jänen et al., 2010). In the present study, clasmatodendritic astrocytes showed GFAP tangles and fibrous GFAP-positive fragments in the cytoplasm (GFAP aggregation) with over-expressions of CXN as well as HSP25. CNX is one of the ER chaperones, which retains unfolded or unassembled N-linked glycoproteins

#### REFERENCES

Alzheimer, A. (1910). ''Beiträge zur kenntnis der pathologischen neuroglia und ihrer beziehungen zu den abbauvorgängen im nervengewebe,'' in Histologische und Histopathologische Arbeiten über die Grosshirnrinde mit besonderer Berücksichtigung der Pathologischen Anatomie der Geisteskrankheiten, eds F. Nissl and A. Alzheimer (Jena: Verlag von Gustav Fischer), 401–562.

in the ER with ATP and Ca2<sup>+</sup> (Ou et al., 1995; Williams, 2006). Thus, up-regulation of CXN is an indicative of ER stress that activates autophagy modulated by PERK/eIF2A activation independent of CHOP (Bernales et al., 2006; Granell et al., 2008; Ito et al., 2009; Korkhov, 2009; Matsumoto et al., 2013; Ko et al., 2015; Men et al., 2015). Interestingly, the present data demonstrate that HSP25 siRNA-infusion effectively abrogated increases in pPERK, peIF2A and CNX expression induced by SE. Thus, our findings suggest that sustained HSP25 expression/accumulation may trigger ER stress in CA1 astrocytes. Alternatively, HSP25-mediated nuclear factor-kappa B (NF-κB) activation may also involve clasmatodendrosis, since HSP25 over-expression increases NF-κB nuclear relocalization, DNA binding and transcriptional activity (Parcellier et al., 2003). Indeed, clasmatodendritic astrocytes show increase in NF-κB activity (Ryu et al., 2011b), which leads to ER stress and autophagy (Nivon et al., 2009; Prell et al., 2014). Therefore, prolonged HSP25 induction may switch autophagy concomitantly with ER stress via activation of NF-κB signaling pathway. Taken together, the present data suggest that the extensive clasmatodendrosis may be autophagic astroglial degeneration per se induced by sustained HSP25-mediated prolonged ER stress.

In conclusion, we provide novel evidence that HSP25 induction involves regional specific astroglial death induced by SE. In the DG, the early HSP25 induction may play an anti-apoptotic role in astrocytes. In the CA1 region, sustained HSP25 expression may result in ER stress, which would induce clasmatodendrosis. To the best of our knowledge, these findings propose for the first time the double-edge profiles of HSP25 in astroglial death following SE. Therefore, we suggest that the HSP25 may be one of the modulators for autophagic astroglial death as well as apoptosis.

# AUTHOR CONTRIBUTIONS

J-EK and T-CK designed and supervised the project. T-CK performed the experiments described in the manuscript with J-EK, H-WH and S-JM. J-EK and T-CK analyzed the data and wrote the manuscript.

#### ACKNOWLEDGMENTS

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (No. 2013R1A6A3A04058272 and 2015R1A2A2A01003539). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer DT and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Kim, Hyun, Min and Kang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Alterations in CD200-CD200R1 System during EAE Already Manifest at Presymptomatic Stages

Tony Valente<sup>1</sup> , Joan Serratosa<sup>1</sup> , Unai Perpiñá<sup>1</sup> , Josep Saura<sup>2</sup> and Carme Solà<sup>1</sup> \*

<sup>1</sup> Department of Cerebral Ischemia and Neurodegeneration, Institut D'Investigacions Biomèdiques de Barcelona-Consejo Superior de Investigaciones Científicas (CSIC), Institut D'Investigacions Biomèdiques August-Pi i Sunyer (IDIBAPS), Barcelona, Spain, <sup>2</sup> Biochemistry and Molecular Biology Unit, School of Medicine, Institut D'Investigacions Biomèdiques August-Pi i Sunyer (IDIBAPS), University of Barcelona, Barcelona, Spain

In the brain of patients with multiple sclerosis, activated microglia/macrophages appear in active lesions and in normal appearing white matter. However, whether they play a beneficial or a detrimental role in the development of the pathology remains a controversial issue. The production of pro-inflammatory molecules by chronically activated microglial cells is suggested to contribute to the progression of neurodegenerative processes in neurological disease. In the healthy brain, neurons control glial activation through several inhibitory mechanisms, such as the CD200- CD200R1 interaction. Therefore, we studied whether alterations in the CD200- CD200R1 system might underlie the neuroinflammation in an experimental autoimmune encephalomyelitis (EAE) model of multiple sclerosis. We determined the time course of CD200 and CD200R1 expression in the brain and spinal cord of an EAE mouse model from presymptomatic to late symptomatic stages. We also assessed the correlation with associated glial activation, inflammatory response and EAE severity. Alterations in CD200 and CD200R1 expression were mainly observed in spinal cord regions in the EAE model, mostly a decrease in CD200 and an increase in CD200R1 expression. A decrease in the expression of the mRNA encoding a full CD200 protein was detected before the onset of clinical signs, and remained thereafter. A decrease in CD200 protein expression was observed from the onset of clinical signs. By contrast, CD200R1 expression increased at EAE onset, when a glial reaction associated with the production of pro- and anti-inflammatory markers occurred, and continued to be elevated during the pathology. Moreover, the magnitude of the alterations correlated with severity of the EAE mainly in spinal cord. These results suggest that neuronal-microglial communication through CD200-CD200R1 interaction is compromised in EAE. The early decreases in CD200 expression in EAE suggest that this downregulation might also occur in the initial phases of multiple sclerosis, and that this early neuronal dysfunction might facilitate the development of neuroinflammation. The increased CD200R1 expression in the EAE model highlights the potential use of targeted agonist molecules as therapeutic tools to control neuroinflammation. In summary, the CD200-CD200R1 system is a potential therapeutic target in multiple sclerosis, and CD200R1 agonists are molecules that may be worth developing in this context.

Keywords: CD200-CD200R1, EAE, multiple sclerosis, neuroinflammation, glial activation, neuron-glia communication, neurological disease, microglia

#### Edited by:

Andrew Harkin, Trinity College, Dublin, Ireland

#### Reviewed by:

Adelaide Fernandes, Faculdade de Farmácia, Universidade de Lisboa, Portugal Daniela Tropea, Trinity College, Dublin, Ireland Eric J. Downer, Trinity College Dublin, Ireland

> \*Correspondence: Carme Solà carme.sola@iibb.csic.es

Received: 26 October 2016 Accepted: 18 April 2017 Published: 04 May 2017

#### Citation:

Valente T, Serratosa J, Perpiñá U, Saura J and Solà C (2017) Alterations in CD200-CD200R1 System during EAE Already Manifest at Presymptomatic Stages. Front. Cell. Neurosci. 11:129. doi: 10.3389/fncel.2017.00129

# INTRODUCTION

fncel-11-00129 May 2, 2017 Time: 15:18 # 2

Neuroinflammation plays a role in the pathogenesis and progression of multiple sclerosis (MS), involving immune cells from both the peripheral immune system and the central nervous system (CNS) (González et al., 2014). Because the modulation of neuroinflammation is considered a possible therapeutic strategy in MS, a thorough knowledge of the cellular and molecular mechanisms is necessary to identify candidate therapeutic targets. To what extent microglia, the main innate immune cells of the CNS, contribute to the development of MS is still a matter of debate (Bogie et al., 2014; Correale, 2014; Gertig and Hanisch, 2014).

Activated microglia/macrophages are not only found in active lesions but also in normal appearing white matter in the brains of MS patients, which suggests that microglia are involved from the initial stages of pathology (Allen et al., 2001; Zeis et al., 2008; van Horssen et al., 2012; Melief et al., 2013). Microglia are crucial in the maintenance of brain homeostasis, and activated microglia show a wide range of molecular and functional phenotypes, from the classically activated M1 phenotype, associated with a proinflammatory effect, to the alternatively activated M2 phenotype, associated with a beneficial effect (Napoli and Neumann, 2010; Gertig and Hanisch, 2014). The production of pro-inflammatory molecules by chronically activated microglial cells may contribute to the progress of neurodegenerative processes in neurological disease, and the inhibition of this response could be a therapeutic target. However, microglial functions also include promotion of neuronal survival, control of the inflammatory response, induction of phagocytosis of cellular debris and stimulation of tissue repair. Consequently, the precise role of microglia in the development of MS is still controversial. An optimal therapeutic approach targeting microglial cells could focus on modulating microglial activation to suppress their deleterious effects and promote their beneficial ones. In fact minocycline, a tetracycline antibiotic with anti-inflammatory properties, attenuates the development of pathology in the experimental autoimmune encephalomyelitis (EAE) animal model of MS through its action on microglia (Popovic et al., 2002). However, beneficial microglial responses are activated during the development of EAE, such as the induction of TREM-2 expression, which controls excess inflammation and stimulates phagocytosis of myelin debris (Takahashi et al., 2007).

Several pharmacological strategies in MS focus on the inhibition of the peripheral immune response, and their effect on microglial cells have been poorly investigated (Giunti et al., 2013). In fact, therapeutic strategies designed to modulate the innate immune response through an action on microglial cells are underexplored (Goldmann and Prinz, 2013). Potential candidates include the inhibitory mechanisms involved in the control of the microglial inflammatory response. Under physiological conditions, inhibitory mechanisms involved in neuron-glia communication participate in the control of the microglial inflammatory response, including CX3CL1-CXRCR1, CD172a-CD45, CD47-CD172, and CD200-CD200R1 ligand-receptor pairs (Tian et al., 2009; Jurgens and Johnson, 2012; Kierdorf and Prinz, 2013). In neurological diseases, the persistence of glial activation over time suggests a possible impairment of these inhibitory mechanisms. In the present study we focus on the CD200-CD200R1 system (Gorczynski, 2012; Walker and Lue, 2013). The CD200 molecule is a transmembrane glycoprotein mainly expressed by neurons and endothelial cells in the CNS, but that is also thought to be expressed by oligodendrocytes and astrocytes. Its receptor CD200R1 is mainly expressed by myeloid cells, like microglia or macrophages. CD200-CD200R1 interaction in myeloid cells results in the phosphorylation of a NPxY motif in the cytoplasmic domain of CD200R1 (Wright et al., 2000), which binds the phosphotyrosine-binding (PTB) domain in the adaptor molecules downstream of tyrosine kinase 1 (Dok1) and Dok2 upon tyrosine phosphorylation. This results in the recruitment and activation by phosphorylation of the Ras GTPase effector enzyme (RasGAP) (Zhang et al., 2004; Mihrshahi and Brown, 2010) and the inhibition of Ras and downstream MAPKs (mitogen-activated protein kinases) PI3K and ERK activation, leading to the inhibition of the production of inflammatory cytokines.

In both the mouse and the human brain, two CD200 isoforms are expressed by alternative mRNA splicing, a full form of the protein (CD200full) -the most abundant isoform- and a truncated form (CD200tr) (Borriello et al., 1998; Chen et al., 2008, 2010). CD200tr also binds to CD200R1, but does not activate the signal transduction pathway, acting as a physiological CD200 antagonist (Chen et al., 2008, 2010). In regard to CD200R1, a single mRNA variant has been described in the mouse brain (Wright et al., 2000), but four mRNA variants resulting from alternative splicing are described in the human brain (Vieites et al., 2003). While the mouse CD200R1 mRNA encodes a transmembrane protein, only two of the human CD200R1 mRNA variants encode transmembrane proteins whereas the other two encode shorter soluble proteins lacking the transmembrane and cytoplasmic domains. Little is known of the changes occurring in CD200 and CD200R1 expression, or in the mechanisms that regulate this expression under physiological and pathological conditions in the CNS, but it has been suggested that the CD200- CD200R1 system could be a candidate therapeutic target in MS (Koning et al., 2009b). The expression of molecules, such as CD200, involved in the control of the inflammatory response by microglia/macrophage, is modified in the postmortem brain tissue of patients with MS (Koning et al., 2007, 2009a). Decreased expression of CD200 and CD200R has been described in postmortem brain tissue of patients with Alzheimer's disease (Walker et al., 2009). Several experimental approaches using the EAE model of MS have shown that reducing the CD200- CD200R1 interaction can aggravate the pathology (Hoek et al., 2000; Meuth et al., 2008), while facilitating CD200R1 activation can improve outcomes (Chitnis et al., 2007; Liu et al., 2010). Although studies in postmortem human tissue have allowed the characterization of the CD200-CD200R1 system at the final stages of the pathology, data are missing on the changes occurring in this system over time, and on their possible involvement in the development of MS. Finally, although manipulation of the CD200-CD200R1 interaction can modify the course of EAE, the extent to which the CD200-CD200R1 system is modified during EAE has not been studied to date.

We have recently shown that CD200R1 expression is inhibited in vitro in mouse reactive microglial cells, and that transcription factors involved in the control of the inflammatory response in these reactive microglia modulate CD200R1 expression (Dentesano et al., 2012, 2014). In the present study, we aimed to determine the dynamics of the CD200-CD200R1 system in EAE by looking at the changes in CD200 and CD200R1 expression in mouse CNS during the development of pathology in the context of associated glial activation and inflammation.

#### MATERIALS AND METHODS

#### Animals

All animal experiments were performed in accordance with the Guidelines of the European Union Council (Directive 2010/63/EU) and Spanish Government (BOE 67/8509-12) and were approved by the Ethics and Scientific Committees of the Spanish National Research Council (CSIC) and the University of Barcelona. All protocols were registered at the "Departament d'Agricultura, Ramaderia, Pesca, Alimentació i Medi Natural de la Generalitat de Catalunya" (DARP 7065). Mice were maintained under regulated light and temperature conditions at the animal facilities of the Faculty of Medicine, University of Barcelona. All efforts were made to minimize animal suffering and discomfort and to reduce the number of animals used.

#### EAE Model

The EAE model used female 6 to 8-week-old C57BL/6 mice (Harlan UK Ltd., Blackthom, UK), as previously described (Mannara et al., 2012). Briefly, mice were immunized under isoflurane anesthesia with a subcutaneous injection of an encephalitogenic emulsion containing 100 µg/mouse of myelin oligodendrocyte glycoprotein (MOG) peptide 35–55 (MOG35−55, Espikem, Italy) and 1 mg/mouse of H37R Mycobacterium tuberculosis (Difco, USA) in 200 µl of complete Freund's adjuvant (CFA) (Sigma–Aldrich, St. Louis, MO, USA). These were then called MOG-EAE mice. Sham-treated mice were injected with a similar emulsion but without MOG35−55, and were used as controls. These were then called CFA mice. All mice were injected intraperitoneally with pertussis toxin from Bordetella pertussis (500 ng/mouse, Sigma–Aldrich) at 1 and 48 h after immunization. Body weight was checked daily from 7 days post-immunization (DPI). At the same time, clinical EAE symptoms (mobility loss and hind limb paralysis) were evaluated according to the following score: 0 = no symptoms; 0.5 = tail weakness; 1 = tail completely flaccid; 1.5 = low difficulty in righting; 2 = high difficulty in righting; 2.5 = unsteady gait and paraparesis (mild paralysis of one or two hind limbs); 3 = complete paralysis of one hind limb; 3.5 = complete paralysis of one hind limb and mild paralysis of the other hind limb; 4 = paraplegia (complete paralysis of two hind limbs) and incontinence; 4.5 = paraplegia and mild paralysis of one or two forelimbs; and 5 = moribund or dead.

For evaluation of clinical EAE symptoms, we used 11 CFA and 26 MOG-EAE mice in three independent experiments. MOG-EAE mice were killed at 28 DPI. In subsequent experiments, CFA and MOG-EAE mice were killed at the following points to determine cellular and molecular alterations in EAE: 9 DPI (presymptomatic phase), 14 DPI (symptomatic phase), 21 DPI (EAE peak) and 28 DPI (chronic phase). For mRNA and protein expression studies, 6 CFA and 19 MOG-EAE mice were considered, CFA mice were processed at 14 (n = 3) and 28 (n = 3) DPI, whereas MOG-EAE mice were processed at 9 (n = 4), 14 (n = 4), 21 (n = 5), and 28 (n = 6) DPI. Spinal cords and brains were carefully removed and longitudinally divided into equal left and right halves, to be processed for RNA or protein extraction, respectively. The spinal cord was dissected into cervical, thoracic, and lumbar regions, while the brain was dissected into mesencephalon plus diencephalon, the rhombencephalon, and the telencephalon regions. All samples were quickly frozen in dry ice. For histology, in situ hybridization and immunohistochemistry 5 CFA and 11 MOG-EAE mice were considered. MOG-EAE mice were perfused in 4% paraformaldehyde at 14 (n = 5) and 21 (n = 6) DPI as previously described (Mannara et al., 2012), while CFA mice were processed at 21 DPI (n = 5). Spinal cords and brains were carefully removed, post-fixed, and cryoprotected in a solution of 30% sucrose and 4% paraformaldehyde, and spinal cords were also divided into cervical, thoracic and lumbar regions. All samples were frozen in dry ice and 20 µm-thick cryostat coronal sections were obtained and stored at −20◦C.

## Histology and Immunohistochemistry

Sequential sections from CFA and MOG-EAE mice (15 and 25 µm-thick) were used for histological analysis, as described previously (Mannara et al., 2012). Hematoxylin and eosin staining was performed to evaluate the general status of the tissue and the presence of cell infiltration. Luxol fast blue method was used for staining myelin projections. Mouse sections were processed for immunohistochemistry as described previously (Valente et al., 2012). The primary and secondary antibodies used are shown in **Table 1**. Microscopy images were obtained with an Eclipse 1000 Nikon microscope (Nikon, Tokyo, Japan) and a digital camera (Olympus DP72, Tokyo, Japan).

#### In Situ Hybridization

Digoxigenin-d-UTP (Boehringer-Mannheim, Mannheim, Germany)-labeled antisense and sense riboprobes were obtained using mouse CD200 cDNA (a kind gift from Prof. R. Gorczynski, Toronto General Research Institute, Canada) and commercial CD200R1 cDNA (OriGene, Rockville, MD, USA). In situ hybridization was performed on mouse brain and spinal cord sections, as described previously (Valente et al., 2005).

## Quantitative Real-Time Polymerase Chain Reaction

Total RNA was isolated from frozen tissue samples, using the Trizol method (Tri <sup>R</sup> Reagent, Sigma–Aldrich). One microgram of RNA was reverse transcribed with random primers using Transcriptor Reverse Transcriptase (Roche Diagnostics Scheiwz AG, Rotkreuz, Switzerland). Then, cDNA was diluted 1/30 to



CD200R1, CD200 receptor 1; GFAP, glial fibrillary acidic protein; Iba1, ionized calcium binding adaptor molecule 1; MBP, myelin basic protein; NeuN, Neuronal nuclei. Companies: Sigma–Aldrich, St. Louis, MO, USA; R&D Systems Inc., Minneapolis, MN, USA; Santa Cruz Biotechnology, Inc., Dallas, Texas, USA; BD Pharmingen: BD Bioscencies, San José, CA, USA; Abcam plc, Cambridge, United Kingdom; Bioss antibodies, Woburn Massachusetts, USA; Dako, Carpinteria, CA, USA; Wako Pure Chemical Industries Ltd, Tokyo, Japan; Millipore, Bedford, MA, USA; Invitrogen Molecular Probes, Eugene, OR, USA; Vector Laboratories Inc., Burlingame, CA, USA.

perform quantitative real-time polymerase chain reaction (qRT-PCR) with IQ SYBRGREEN SuperMix (Bio-Rad Laboratories, Hercules, CA, USA) as previously described (Valente et al., 2012). The primers (Integrated DNA Technologies, Leuven, Belgium) used to amplify mouse mRNA are shown in **Table 2**. Relative gene expression was calculated by the comparative Ct or 11Ct method (Livak and Schmittgen, 2001) using CFX 2.1 software (Bio-Rad Laboratories).

#### Western Blot

Mouse total protein extracts were obtained as previously described (Valente et al., 2012). Protein quantification was determined by Bradford assay (Bio-Rad Laboratories). Western blot analysis of 30 µg extracts of total protein was performed using appropriate primary and secondary antibodies (**Table 1**), as previously described (Valente et al., 2012). Membranes were developed with ECL-Plus (Amersham GE Healthcare

#### TABLE 2 | Primers used for qRT-PCR.


#Chen et al., 2010; ∗ reference gene; Arg1, arginase1; CD200full, full-length CD200; CD200tr, truncated CD200; CD200R1, CD200 receptor 1; CD200R1-Large, CD200R1 mRNA variants 1 and 4 (large mRNA variants); CD200R1-Short, CD200R1 mRNA variants 2 and 3 (short mRNA variants); CD200R1-V1, CD200R1 mRNA variant 1; CD200R1-V2, CD200R1 mRNA variant 2; CD200R1-V3, CD200R1 mRNA variant 3; CD200R1-V4, CD200R1 mRNA variant 4; COX2, cyclooxygenase-2; Gfap, glial fibrillary acidic protein; Iba1, ionized calcium binding adaptor molecule 1; Il1b, interleukin 1 beta; Il10, interleukin 10; Mrc1, mannose receptor 1; Nos2, nitric oxide synthase 2; Socs3, suppressor of cytokine signaling 3; Tgfb, transforming growth factor beta; Tnfa, tumor necrosis factor alpha; Rn18s, 18S ribosomal RNA; RPS18, ribosomal protein S18.

Life Sciences Europe GmbH, Barcelona, Spain) and images were obtained using a VersaDoc System camera (Bio-Rad Laboratories). Data were expressed as the ratio between the band intensity of the protein of interest and that of the loading control (β-actin).

# Data Presentation and Statistical Analysis

Normality of data was determined by D'Agostino and Pearson omnibus normality test. Statistical analyses were performed using one-way analysis of variance (ANOVA) and the Dunnett post hoc test. The Spearman correlation coefficient (r) was calculated to measure the linear correlation between the clinical EAE score and mRNA expression. Statistical analyses were performed using GraphPad Prism 4.02 (GraphPad Software, Inc., La Jolla, CA, USA). All results are presented as mean ± SEM values, unless otherwise stated. Values of p < 0.05 were considered statistically significant.

# RESULTS

#### EAE Model and Glial Activation

To determine the kinetics of CD200-CD200R axis dysregulation during EAE, we first characterized EAE in this study, because this is known to be dependent on the facility/environment. MOG-EAE mice showed the first clinical signs of EAE at 10 DPI, with a mean onset at 12.9 ± 0.5 DPI (**Figure 1A**). The incidence of EAE was 26.1, 78.3, and 100%, at 10, 14, and 18 DPI, respectively (**Figure 1B**). At 18 DPI, 13.0% of the animals showed limp tails (score 0.5–1.0), 21.7% showed dysfunctional motor coordination (score 1.5–2.0) and 65.2% showed mild-to-moderate paraparesis (score 2.5–4.0). The EAE clinical data correlated with the demyelination observed in the spinal cord of MOG-EAE mice at 14 and 21 DPI. Thus, some small patches of demyelination were observed in the white matter of the dorsolateral spinal cord at 14 DPI, which became more evident at 21 DPI (**Figure 1C**). Infiltrating cells were detected by hematoxylin and eosin staining, which were evaluated by specific lymphocyte markers in the white matter of MOG-EAE mice, and were mainly localized around the marginal zone of demyelinated areas (**Figure 1C**), close to the spinal cord surface. Some cells were observed at 14 and 21 DPI showing B- and T-lymphocyte markers (B220 and CD3) (**Figure 1D**).

Regarding glial cells, we detected significant increases in the mRNA expressions of both the microglia/macrophage marker Iba1 and the astroglial marker GFAP in spinal cord regions of MOG-EAE mice at different post-immunization times (**Figure 2A**), albeit with some exceptions. Larger increases were observed in the lumbar spinal cord. Curiously, in the presymptomatic EAE phase (9 DPI), Gfap mRNA upregulation was observed in the cervical region. In MOG-EAE mice, strong Iba1 immunostaining was observed in spinal cord white matter at 14 and 21 DPI, suggesting microglia/macrophage activation. In addition, an increase in GFAP immunostaining was observed at 14 and 21 DPI in spinal cord white matter, which suggested astroglial

eosin staining (HEM). Scale bars: 250 µm in Luxol fast blue and MBP images, and 200 µm in HEM staining. (D) Lymphocyte infiltrates are evident in the lumbar spinal cord (arrows). B220-positive B-cell (green) and CD3-positive T-cell (red) infiltrates were found at 14 and 21 DPI. Scale bar: 100 µm.

activation. **Figure 2B** shows images from lumbar spinal cord as representative region showing the most apparent glial reactivity in the spinal cord. We also determined Iba1 and Gfap mRNA expression in all brain regions (**Figure 2C**) of MOG-EAE mice at different post-immunization times. Interestingly, upregulation of Iba1 mRNA could be observed in the presymptomatic phase in mesencephalon/diencephalon (**Figure 2C**). Microglial/macrophage activation was confirmed

Frontiers in Cellular Neuroscience | www.frontiersin.org May 2017 | Volume 11 | Article 129 |

by Iba1 immunostaining in coronal brain sections (**Figure 2D**), where reactive microglia/macrophages were detected in the hippocampus at 14 and 21 DPI. GFAP-labeled reactive astrocytes were also observed in hippocampal and cortical areas (**Figure 2D**).

### Time Course of CD200 Expression in EAE

We studied whether the immune reaction associated to MOG injection produced changes in the expression of the inhibitory immune receptor, CD200R1, and its ligand, CD200, in the CNS. To this end, we determined their mRNAs and/or protein expression in the spinal cords and brains of CFA and MOG-EAE mice after several DPI. In MOG-EAE mice, downregulation of Cd200full mRNA expression was observed in the presymptomatic phase of EAE (9 DPI) in lumbar and thoracic regions, but was observed in all spinal cord regions during symptomatic phases, except at 28 DPI in the thoracic region (**Figure 3A**). The magnitude of changes was similar among the spinal cord regions. Less-marked alterations were observed in the brain than in the spinal cord, with a decrease in Cd200full mRNA expression in the mesencephalon/diencephalon, and punctual increases in the rhombencephalon and telencephalon (**Figure 3A**). This contrasted with Cd200tr mRNA expression, which was increased in the thoracic and cervical spinal cord in the symptomatic phase and in the mesencephalon/diencephalon and telencephalon in the presymptomatic phase of MOG-EAE mice (**Figure 3A**). Cd200full mRNA expression negatively correlated with EAE clinical score in thoracic and cervical spinal cord areas, while there was a positive correlation between Cd200tr mRNA expression and the EAE clinical score in the cervical spinal cord (**Table 3**). By in situ hybridization, we observed cells positive for Cd200full mRNA with neuronal morphology only in the ventral part of the gray matter of the spinal cords of CFA mice, where the somas of motor neurons are localized (**Figure 3B**). However, cellular Cd200full mRNA expression was noticeably decreased in MOG-EAE mice at 14 and 21 DPI (**Figure 3B**). To determine whether CD200 transcript modulation during EAE translated to protein changes, we analyzed spinal cord CD200 expression by western blot. The anti-CD200 antibody used recognized amino acids 31–232 in the CD200 protein, and consequently the full-length and truncated isoforms (45– 48 and 35–40 kDa, respectively) (Gorczynski, 2002). A main dense band was observed at approximately 52 kDa. From 14 DPI, there was a significant decrease in CD200 protein in the MOG-EAE mice (**Figure 3C**). The decrease observed by western blot was confirmed by CD200 immunofluorescence on spinal cord sections (**Figure 3D**), where CD200 was found to co-localize with the neuronal marker NeuN in the ventral part of the gray matter of the spinal cord.

## Time Course of CD200R1 Expression in EAE

In MOG-EAE mice, Cd200r1 mRNA expression was strongly upregulated in all the spinal cord regions at all symptomatic phases, showing a clear peak at 14 DPI (**Figure 4A**). Induction of Cd200r1 mRNA was also observed in the brain, but to a lesser extent than in the spinal cord. In addition, the EAE clinical score was positively correlated with Cd200r1 mRNA expression, mainly in spinal cord areas (**Table 3**). This increased expression was corroborated by in situ hybridization using spinal cord sections (**Figure 4B**). Thus, although no Cd200r1 positive cells were detected in the CFA mice, they were clearly observed in MOG-EAE mice at both 14 and 21 DPI. Cells positive for Cd200r1 mRNA were mainly localized to the white matter, mostly in areas of demyelination and infiltration in the lateral part of the dorsal spinal cord. Nevertheless, scattered CD200R1 and Iba1 positive cells were also observed in more distal areas. Western blotting of spinal cord total protein showed an increase in CD200R1 in protein extracts of MOG-EAE mice at 14 and 21 DPI (**Figure 4C**). The cellular localization of CD200R1 protein was then analyzed by immunohistochemistry (**Figure 4D**). Few CD200R1-positive cells were observed in the spinal cords of CFA mice, and this number increased markedly in MOG-EAE mice in the areas where cells positive for Cd200r1 mRNA-were found. In most cells, CD200R1 co-localized with the microglial/macrophage marker Iba1.

### Inflammatory Response in EAE

As the CD200-CD200R1 interaction is involved in the control of the inflammatory response in microglial cells, we analyzed the inflammatory context in which changes in CD200 and CD200R1 expression were observed. We studied the time course of pro- and anti-inflammatory molecule expression during the development of EAE using qRT-PCR. For classical pro-inflammatory M1 markers, we analyzed Nos2, Il1b, and Tnfa (**Figures 5A,B**). In the spinal cords of MOG-EAE mice, the respective genes showed mRNA expression profiles characterized by markedly peaked expression at 14 DPI (**Figure 5A**). In the brain, an increase in Nos2 mRNA was also observed at 14 DPI, particularly in the mesencephalon/diencephalon, while Il1b mRNA was significantly upregulated in the rhombencephalon and mesencephalon/diencephalon (**Figure 5B**). By contrast, Tnfa mRNA upregulation was observed in all brain areas at several times after immunization (**Figure 5B**), and notably in both the mesencephalon/diencephalon and telencephalon in the presymptomatic phase (9 DPI). We also analyzed COX2 mRNA expression, an M1/M2 marker, and found its levels significantly upregulated in all spinal cord areas during the symptomatic phases in MOG-EAE mice (i.e., 14–28 DPI). A significant increase in COX2 mRNA expression was also observed in all brain areas, but to a lesser extent than in the spinal cord (**Figure 5C**). In addition, COX2 mRNA upregulation was significantly increased in the telencephalon during the presymptomatic phase of EAE (9 DPI) (**Figure 5C**).

We analyzed the following M2 genes as anti-inflammatory M2 markers: Arg1, Il10, Mrc1, Socs3, and Tgfb (**Figure 6**). Arg1 mRNA was strongly upregulated in spinal cord regions at 14 DPI (**Figure 6A**), while more moderate increases were found in all brain areas (**Figure 6B**). However, we also observed Arg1 mRNA induction during the presymptomatic phase in the telencephalon. For Il10 mRNA expression, we observed expression induction in spinal cord and brain regions between 14 and 28 DPI

FIGURE 3 | Time course of CD200 expression in EAE. (A) Full-length Cd200 (Cd200full) and truncated Cd200 (Cd200tr) mRNAs were evaluated in spinal cords and brains by qRT-PCR during EAE, using Rn18s as the housekeeping gene. Bars are the means ± SEM of 4–6 animals. <sup>∗</sup>p < 0.05, ∗∗p < 0.01 vs. respective CFA; one-way ANOVA and Dunnett post-test. (B) In situ hybridization of Cd200full mRNA in the spinal cord showing Cd200full-positive cells (arrowheads) in the ventral gray matter. (C) Expression of CD200 protein in the spinal cord by western blot using β-actin as the loading control. Bars are the means ± SEM of 4–6 animals. <sup>∗</sup>p < 0.05, ∗∗p < 0.01 vs. respective CFA; one-way ANOVA and Dunnett post-test. A representative western blot is shown. (D) Double immunofluorescence of CD200 (green) and NeuN (red) in the gray matter of the spinal cord (arrows). Cells were counterstained with DAPI (blue). Results from lumbar spinal cord, as representative of all spinal cord areas studied, are shown in (B–D). RE, rhombencephalon; ME/DE, mesencephalon/diencephalon; TE, telencephalon. Scale bars: 100 µm.


TABLE 3 | Correlation between EAE severity and Cd200full, Cd200tr or Cd200r1 mRNA expression in symptomatic MOG-EAE mice.

<sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

fncel-11-00129 May 2, 2017 Time: 15:18 # 9

(**Figure 6**). The largest increase was observed in lumbar spinal cord at 14 DPI. In addition, Il10 mRNA expression was observed in the cervical spinal cord, mesencephalon/diencephalon, and telencephalon during the presymptomatic phase. Mrc1 mRNA expression was strongly upregulated in lumbar spinal cord, but significant increases were observed in the other spinal cord areas and brain tissue at several symptomatic phases (**Figure 6**). Socs 3 expression was mainly upregulated by 14 DPI in spinal cord and brain areas, but also during the presymptomatic phase in the brain (**Figure 6B**). Tgfb mRNA expression was significantly upregulated in the spinal cord during the symptomatic phase, but only became significant in the brain at 14 DPI (**Figure 6**).

Finally, we analyzed the correlation between EAE severity and pro- and anti-inflammatory mRNA expression in symptomatic MOG-EAE mice, which showed that the EAE clinical score was positively correlated with the expression levels of both pro- and anti-inflammatory genes in spinal cord, especially in the lumbar region, and brain areas, mainly in the rhombencephalon and mesencephalon/diencephalon (**Table 4**).

#### DISCUSSION

Alterations in neuronal-glial and/or glial–glial crosstalk may contribute to glial activation in MS, and changes in either neuronal or glial cells could trigger this process. In turn, the resulting functional phenotype will determine the beneficial or detrimental role of microglial cells. Various inhibitory mechanisms, involving both soluble signals and ligand-receptor pairs, coexist to control the pro-inflammatory response of microglial cells under physiological conditions. The presence of glial activation in MS suggests that these mechanisms have been overloaded, implicating them as potential therapeutic targets. In the present study of the dynamics of CD200 and CD200R1 expression in the mouse CNS of an EAE model, we show that changes in the inhibitory CD200-CD200R1 system are already observed at presymptomatic EAE stages, in association with glial activation, with these changes preceding the inflammatory response that accompanies the onset of clinical signs.

The interaction between CD200 and the microglial inhibitory receptor CD200R1 is postulated to be a mechanism involved in controlling the microglial inflammatory response in healthy brain tissue (Tian et al., 2009; Jurgens and Johnson, 2012; Kierdorf and Prinz, 2013). However, CD200 and CD200R1 expression were altered at presymptomatic and symptomatic phases in the CNS of mice that developed EAE, suggesting compromised control of the microglial inflammatory response from early pathological stages. Regarding the CD200 ligand, we looked at the expression of the two Cd200 mRNA variants, Cd200full and Cd200tr, in the mouse brain and spinal cord. Cd200tr mRNA encodes a protein lacking part of the N-terminal domain that is critical for CD200-CD200R1 interaction and CD200R1 stimulation (Hatherley and Barclay, 2004), so acts as an endogenous antagonist of CD200R1 (Chen et al., 2008, 2010). Consequently, the ratio of Cd200full to Cd200tr expression in a tissue may regulate the CD200-CD200R1 interaction. Cd200full mRNA expression was clearly reduced in the spinal cords of MOG-EAE mice during EAE, even during presymptomatic stage, while Cd200tr mRNA expression increased during the symptomatic phase. These changes may decrease the ratio of CD200full to CD200tr, reducing the inhibitory input to microglial/macrophage cells. A sustained decrease in CD200 protein levels was also detected after onset of EAE in the spinal cord. Cd200full mRNA and CD200 protein expression were localized in the cell bodies of motor neurons in the ventral gray matter by in situ hybridization and immunohistochemistry, respectively. Interestingly, no CD200 expression was detected in the dorsal gray matter, where interneurons are located. In brain areas, mRNA expression of both CD200 isoforms were altered, but to a lesser extent than in the spinal cord. Curiously, though, there was a clear increase in Cd200tr mRNA expression in brain areas before EAE onset. Together, these results show that the immune reaction induced by MOG35−<sup>55</sup> administration changed the expression of CD200, suggesting that early motor neuron dysfunction promotes the inflammatory response.

Stimulation of CD200R1 by CD200 activates a signal transduction pathway that inhibits pro-inflammatory gene expression (Zhang et al., 2004). The presence of altered CD200 expression in EAE therefore suggests that mechanisms controlling the inflammatory response are compromised. Previous studies have shown that the manipulation of CD200 expression or function modifies the development of EAE. In this sense, EAE is more severe in CD200−/− than in wild-type mice (Hoek et al., 2000) and the clinical signs of EAE are attenuated in mice that overexpress CD200 (Chitnis et al., 2007). In addition, mice administered anti-CD200 blocking antibody develop a more severe EAE (Meuth et al., 2008). Thus, it is plausible that CD200 levels influence the course of pathology in EAE, and the decreased CD200 expression we observed in MOG-EAE mice, which was already detected at the presymptomatic phase, will probably facilitate the development of pathology. However, it

(arrowheads). Results from lumbar spinal cord, as representative of all spinal cord areas studied, are shown in (B–D). RE, rhombencephalon; ME/DE,

must also be true that the remaining CD200 expression retains a significant functional effect because EAE is more severe when CD200 is completely absent (Hoek et al., 2000) or blocked (Meuth et al., 2008).

mesencephalon/diencephalon; TE, telencephalon. Scale bars: 100 µm.

Regarding CD200R1, the spinal cords of MOG-EAE mice showed a strong induction of Cd200r1 mRNA expression after the onset of EAE, and this remained elevated during EAE. Cd200r1 mRNA expression also increased in the brain after

EAE induction, though to a lesser extent than in the spinal cord. Accordingly, while CD200R1 protein levels were very low in the CNS of control mice, they were markedly increased in the spinal cords of MOG-EAE mice. We clearly localized the CD200R1 mRNA and immunosignal in an increased population of Iba1-positive microglia/macrophages to parenchyma in the dorsal spinal cord white matter of MOG-EAE mice, in and around the areas of demyelination and infiltration. Liu et al. (2010) showed that giving a CD200R1 agonist reduced disease severity, demyelination, and axonal damage in a mouse EAE model, suggesting that CD200R1 stimulation and the resulting inhibition of the inflammatory response positively affect the outcome of EAE. The increase in CD200R1 expression observed in EAE may, therefore, act to compensate for the loss of function in the CD200-CD200R1 system that follows decreased CD200 expression.

CD200R1 agonist treatment has been shown to inhibit the production of pro-inflammatory cytokines in activated glial cells in culture (Liu et al., 2010; Hernangómez et al., 2012; Lyons et al., 2012). Therefore, the increased CD200R1 expression observed in the MOG-EAE mice in our model could be a compensatory response to limit the expression of pro-inflammatory markers.


TABLE 4 | Correlation between EAE severity and pro- and anti-inflammatory mRNA expression in symptomatic MOG-EAE mice.

<sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

To asses this, we correlated the time course of changes in Cd200 and Cd200r1 mRNA expression in the CNS of MOG-EAE mice with the evolution of the inflammatory response during EAE, and evaluated the RNA expressions of pro-inflammatory M1 (Nos2, Il1b, Tnfa) and anti-inflammatory M2 (Arg1, Il10, Mrc1, Socs3, and Tgfb) markers. In spinal cord regions, the mRNA expression of pro- and anti-inflammatory genes was strongly induced after EAE onset, but only anti-inflammatory gene expression remained elevated thereafter (except for Arg1). A decrease in Cd200full mRNA expression in the spinal cord preceded the inflammatory response, which was attenuated in the presence of a maintained increase in Cd200r1 mRNA expression. However, an inflammatory response also developed in brain areas, though they did not show overall alterations in Cd200full mRNA and Cd200R1 mRNA as in the spinal cord. Curiously, we observed a significant induction of anti-inflammatory genes in brain areas during the presymptomatic phase. The expression pattern of COX2, which is classified as a marker of M1 and M2 phenotypes (David and Kroner, 2011; Chhor et al., 2013; Franco and Fernández-Suárez, 2015), has shown a similar expression pattern to the M2 markers in our MOG-EAE mice. A sustained increase in the expressions of CD200R1 and antiinflammatory markers could reflect an attempted compensatory response aimed at resolving the inflammatory process (Colton, 2009).

Regarding the clinical score for EAE, a negative correlation was noted with Cd200full mRNA expression in thoracic, cervical but not lumbar spinal cord areas in symptomatic MOG-EAE mice, while a positive correlation was found with Cd200r1 mRNA expression in all spinal cord areas. In addition, a positive correlation was also observed between the clinical score for EAE and Cd200tr mRNA expression in cervical spinal cord. These results suggest that the magnitude of changes in CD200R1 expression in spinal cord reflects the severity of EAE. The magnitude of changes in the mRNA expression of both pro- and anti-inflammatory genes also positively correlated to the severity of EAE spinal cord (mainly lumbar) and brain (mainly rhombencephalon and mesencephalon/diencephalon), and suggests the coexistence of M1 and M2 phenotypes in microglia/macrophages in EAE.

#### CONCLUSION

Our results show that the expressions of CD200 and CD200R1 in the CNS are modified during EAE. Postmortem samples from patients with MS are usually only able to show the final stages of disease, and cannot illuminate the changes that occur early in the pathogenesis. In this study, we demonstrated that there was a decrease in CD200 expression before the onset of clinical symptoms in EAE. This suggests that alterations in CD200 expression might also occur in the early stages of MS, which may be responsible for downregulated control of microglial/macrophage activation, thereby stimulating the inflammatory response and contributing to the development of the pathology. By contrast, there was a subsequent increase in CD200R1 expression that possibly represented a compensatory response to re-establish control of the inflammation. The fact that CD200R1 expression was increased points to this receptor as a potential therapeutic target for the regulation of neuroinflammation. Indeed, we consider that CD200R1 agonists are promising molecules that should be developed to modulate neuroinflammation and the resulting neurotoxicity in neurological disease.

#### AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: TV, JSa, and CS. Performed the experiments: TV, JSe, UP, JSa, and CS. Analyzed

the data: TV, JSe, JSa, and CS. Wrote the paper: TV and CS. All authors critically revised and approved the final manuscript.

#### FUNDING

This study was supported by grants PI10/378, PI12/00709 and PI15/00033 from the Instituto de Salud Carlos III (Spain) with joint financing by FEDER funds from the European Union, and a grant from La Marató de TV3 foundation (20110530). TV received a JAE-Doc contract and UP received a JAE-Tec contract

#### REFERENCES


("Junta para la Ampliación de Estudios") from CSIC, with joint financing by European Social Fund.

#### ACKNOWLEDGMENTS

The authors thank Dr. Reginald Gorczynski (Toronto Hospital, Toronto, Canada) for providing mouse CD200 cDNA. The authors also thank Guido Dentesano and Luca Ricci for their technical assistance, and Dr. Marco Straccia for critical reading of the manuscript.

neurodegenerative diseases. J. Immunol. 274, 1–13. doi: 10.1016/j.jneuroim. 2014.07.012


impaired CD200-CD200R-mediated macrophage silencing. J. Neuroimmunol. 194, 62–69. doi: 10.1016/j.jneuroim.2007.11.013


potential mechanism leading to chronic inflammation. Exp. Neurol. 215, 5–19. doi: 10.1016/j.expneurol.2008.09.003


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewers DT, EJD and the handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Valente, Serratosa, Perpiñá, Saura and Solà. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.