# NEURONAL CALCIUM SENSORS IN HEALTH AND DISEASE

EDITED BY : Karl-Wilhelm Koch, Jose R. Naranjo, Beat Schwaller, Daniele Dell'Orco and Michael R. Kreutz PUBLISHED IN : Frontiers in Molecular Neuroscience

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-342-5 DOI 10.3389/978-2-88963-342-5

#### 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

# NEURONAL CALCIUM SENSORS IN HEALTH AND DISEASE

Topic Editors:

Karl-Wilhelm Koch, University of Oldenburg, Germany Jose R. Naranjo, Spanish National Research Council (CSIC), Spain Beat Schwaller, Université de Fribourg, Switzerland Daniele Dell'Orco, University of Verona, Italy Michael R. Kreutz, Leibniz Institute for Neurobiology (LG), Germany

Citation: Koch, K.-W., Naranjo, J. R., Schwaller, B., Dell'Orco, D., Kreutz, M. R., eds. (2020). Neuronal Calcium Sensors in Health and Disease. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-342-5

# Table of Contents


Hanna Wimberg, Dorit Lev, Keren Yosovich, Prasanthi Namburi, Eyal Banin, Dror Sharon and Karl-Wilhelm Koch

*62 Zebrafish Recoverin Isoforms Display Differences in Calcium Switch Mechanisms*

Dana Elbers, Alexander Scholten and Karl-Wilhelm Koch


Rocío Naranjo, Paz González, Alejandro Lopez-Hurtado, Xosé M. Dopazo, Britt Mellström and José R. Naranjo

*120 Functional Status of Neuronal Calcium Sensor-1 is Modulated by Zinc Binding*

Philipp O. Tsvetkov, Andrei Yu. Roman, Viktoriia E. Baksheeva, Aliya A. Nazipova, Marina P. Shevelyova, Vasiliy I. Vladimirov, Michelle F. Buyanova, Dmitry V. Zinchenko, Andrey A. Zamyatnin Jr., François Devred, Andrey V. Golovin, Sergei E. Permyakov and Evgeni Yu. Zernii

*141 The Complex Conformational Dynamics of Neuronal Calcium Sensor-1: A Single Molecule Perspective*

Dhawal Choudhary, Birthe B. Kragelund, Pétur O. Heidarsson and Ciro Cecconi

*149 Ca2+-Dependent Transcriptional Repressors KCNIP and Regulation of Prognosis Genes in Glioblastoma* Isabelle Néant, Jacques Haiech, Marie-Claude Kilhoffer, Francisco J. Aulestia, Marc Moreau and Catherine Leclerc *158 The Binding Properties and Physiological Functions of Recoverin* Jingjing Zang and Stephan C. F. Neuhauss *168 Light-Induced Thiol Oxidation of Recoverin Affects Rhodopsin Desensitization* Evgeni Yu. Zernii, Aliya A. Nazipova, Ekaterina L. Nemashkalova, Alexey S. Kazakov, Olga S. Gancharova, Marina V. Serebryakova, Natalya K. Tikhomirova, Viktoriia E. Baksheeva, Vasiliy I. Vladimirov, Dmitry V. Zinchenko, Pavel P. Philippov, Ivan I. Senin and Sergei E. Permyakov *187 Disruption of Otoferlin Alters the Mode of Exocytosis at the Mouse Inner Hair Cell Ribbon Synapse* Hideki Takago, Tomoko Oshima-Takago and Tobias Moser *195 Synaptotagmin Ca2+ Sensors and Their Spatial Coupling to Presynaptic Cav Channels in Central Cortical Synapses* Grit Bornschein and Hartmut Schmidt *210 Global Gene Knockout of* Kcnip3 *Enhances Pain Sensitivity and Exacerbates Negative Emotions in Rats* Yu-Peng Guo, Yu-Ru Zhi, Ting-Ting Liu, Yun Wang and Ying Zhang *223 Identification of IQM-266, a Novel DREAM Ligand That Modulates KV4 Currents* Diego A. Peraza, Pilar Cercós, Pablo Miaja, Yaiza G. Merinero, Laura Lagartera, Paula G. Socuéllamos, Carolina Izquierdo García, Sara A. Sánchez, Alejandro López-Hurtado, Mercedes Martín-Martínez, Luis A. Olivos-Oré, José R. Naranjo, Antonio R. Artalejo, Marta Gutiérrez-Rodríguez and Carmen Valenzuela *237 Caldendrin and Calneurons—EF-Hand CaM-Like Calcium Sensors With Unique Features and Specialized Neuronal Functions* Jennifer Mundhenk, Camilla Fusi and Michael R. Kreutz *246 20 Years of Secretagogin: Exocytosis and Beyond* Magdalena Maj, Ludwig Wagner and Verena Tretter *256 Emerging Roles of Neuronal Ca2+ Sensor-1 in Cardiac and Neuronal Tissues: A Mini Review* Tomoe Y. Nakamura, Shu Nakao and Shigeo Wakabayashi *264 Evolutionary-Conserved Allosteric Properties of Three Neuronal Calcium Sensor Proteins*

Valerio Marino and Daniele Dell'Orco

*276 Calcium, Dopamine and Neuronal Calcium Sensor 1: Their Contribution to Parkinson's Disease*

Cristina Catoni, Tito Calì and Marisa Brini


Carsten Simons, Julia Benkert, Nora Deuter, Christina Poetschke, Olaf Pongs, Toni Schneider, Johanna Duda and Birgit Liss

# Editorial: Neuronal Calcium Sensors in Health and Disease

Daniele Dell'Orco<sup>1</sup> \*, Karl-Wilhelm Koch<sup>2</sup> , Michael R. Kreutz <sup>3</sup> , Jose R. Naranjo4,5 and Beat Schwaller <sup>6</sup>

<sup>1</sup> Section of Biological Chemistry, Department of Neurosciences, Biomedicine and Movement Sciences, University of Verona, Verona, Italy, <sup>2</sup> Department of Neuroscience, School of Medicine and Health Sciences, University of Oldenburg, Oldenburg, Germany, <sup>3</sup> RG Neuroplasticity, Leibniz Institute for Neurobiology, Magdeburg, Germany, <sup>4</sup> Spanish Network for Biomedical Research in Neurodegenerative Diseases (CIBERNED), Instituto de Salud Carlos III, Madrid, Spain, <sup>5</sup> National Biotechnology Center, CNB-CSIC, Madrid, Spain, <sup>6</sup> Département des neurosciences et des sciences du mouvement (NMS), Université de Fribourg, Fribourg, Switzerland

#### Keywords: neuronal Ca2<sup>+</sup> sensors, calcium signaling, molecular basis of disease, calcium bindind proteins, neurological disorders

#### **Editorial on the Research Topic**

#### **Neuronal Calcium Sensors in Health and Disease**

The precise detection and regulation of free intracellular Ca2<sup>+</sup> is a prerequisite for keeping normal cell function. Several physiological processes such as fertilization, apoptosis, muscle contraction, neuronal activity, and sensory perception are based on the spatial and temporal variation in intracellular Ca2<sup>+</sup> and rely on a class of proteins that specifically respond to these highly dynamic stimuli. Neuronal calcium sensor (NCS) proteins are exclusively expressed or enriched in neurons, and their structural and biochemical diversity reflects the multiplicity of their biological roles, which include control of gene transcription, neuronal growth and survival, channel and receptor regulation, neurotransmitter release, synaptic plasticity, and regulation of enzymatic activity. Neurological disorders and neurodegenerative diseases are increasingly associated with altered functions of specific NCS proteins. This Research Topic includes original articles and reviews that provide an interdisciplinary collection of up-to-date biochemical and biophysical research on NCS proteins and their established and novel biological roles in normal and altered (pathological) conditions.

Neuronal calcium sensor-1 (NCS-1) is highly conserved from yeast to human, and it has been

#### Universitätsklinikum Bonn, Germany implicated in a number of psychiatric conditions including autism, bipolar disorder, schizophrenia, and X-linked mental retardation. At odds with other members of the NCS family, NCS-1 interacts with several cellular targets, which is reflected by a variety of roles. Ratai et al. show that NCS-1 plays an important role in adipocyte function and its deficiency gives rise to obesity and diabetes type 2 in adult mice, thus suggesting a potential genetic link between psychiatric disorders and the risk of being obese. The progressive degeneration of dopaminergic neurons within the Substantia nigra is the hallmark of Parkinson's disease and causes its motor symptoms. The link between dopamine release and NCS-1 and its possible implications in Parkinson's disease has been reviewed by Catoni et al., who summarize the role of the interplay between Ca2<sup>+</sup> and dopamine signaling in neuronal activity and cell death. Simons et al. provide novel data at the transcript level that link

NCS-1 deficiency to impaired ATP-production and elevated metabolic stress in Substantia nigra dopaminergic neurons in mice. An overview of other novel roles of NCS-1 is provided by the review of Nakamura et al., which focuses on both the neuronal system and the heart, presenting NCS-1 as a regulator of voltage-gated Ca2<sup>+</sup> channels, ionotropic dopamine receptors, and inositol 1,4,5-trisphosphate receptors. A complex interplay between Mg2+, Ca2<sup>+</sup> and Zn2<sup>+</sup> binding to NCS-1 leads to the appearance of multiple protein conformations and modulate its functional status as suggested by Tsvetkov et al., who demonstrate that NCS-1 binds Zn2<sup>+</sup> with differential affinities favoring either the interaction with targets or protein aggregation. Choudhary et al., who

Edited and reviewed by: Ildikó Rácz,

\*Correspondence:

Daniele Dell'Orco daniele.dellorco@univr.it

Received: 04 October 2019 Accepted: 31 October 2019 Published: 15 November 2019

#### Citation:

Dell'Orco D, Koch K-W, Kreutz MR, Naranjo JR and Schwaller B (2019) Editorial: Neuronal Calcium Sensors in Health and Disease. Front. Mol. Neurosci. 12:278. doi: 10.3389/fnmol.2019.00278 focus on optical tweezers investigations to reveal a complex folding mechanism underlaid by a rugged and multidimensional energy landscape, provide insight into the structural and mechanistic details of the folding and misfolding processes of NCS-1 at the single molecule level.

The importance of allosteric interactions between Ca2+ binding motifs and amino acids involved in target recognition has been investigated by Marino and Dell'Orco, who suggest for NCS-1, recoverin and GCAP1 an evolution-driven correlation between the amino acids mediating many persistent interactions and their conservation. An inter-domain interaction triggered by Mg2+-binding is essential for the ubiquitous Ca2<sup>+</sup> sensor CIB2 to reach a fully functional conformation, as shown by Vallone et al., who found that the apparently conservative E64D mutation associated with Usher Syndrome 1J and non-syndromic hearing loss prevents this long-range allosteric mechanism. Mutations in calmodulin, another ubiquitous Ca2<sup>+</sup> sensor, were long thought to be incompatible with life due to the completely conserved amino acid sequence across all vertebrates. The review by Jensen et al. provides an overview of the human missense mutations found in patients with severe cardiac arrhythmias.

Many NCS proteins form dimers, a process that is often Ca2<sup>+</sup> dependent. The review by Ames summarizes the results of recent studies on GCAPs, VILIP1 and recoverin dimerization. A paradigmatic example of the myristoyl-switch protein, recoverin is involved in the regulation of the phototransduction cascade in rods and cones as reviewed by Zang and Neuhauss. Four different recoverin isoforms exist in zebrafish photoreceptors. Their specific Ca2+-sensing properties and conformational changes have been investigated by Elbers et al., who found that binding of Ca2<sup>+</sup> leads to less pronounced structural rearrangements compared to the bovine ortholog indicating either a modified Ca2+-myristoyl switch or no switch at all. Novel roles for recoverin in health and disease-associated conditions have been found by Zernii et al. who provide in vitro and in vivo evidence that illumination of the mammalian retina leads to the accumulation of disulfide dimers of recoverin, which are thought to favor light-induced oxidative stress and photoreceptor apoptosis.

The complex between GCAPs and the retinal guanlylate cyclases is a crucial component of the vertebrate phototranduction machinery as it regulates the interplay between the second messengers cGMP and Ca2+. Rehkamp et al. present a chemical cross-link/mass spectrometry investigation on the interaction between GCAP2 and GC-E, while Wimberg et al. investigate five recently identified GC-E mutants associated with Leber Congenital Amaurosis, a cone-rod dystrophy, finding severe alteration of the cGMP synthesis. Another prototypical guanylate cyclase, the atrial natriuretic factor receptor guanylate cyclase is found to be regulated by the Ca2<sup>+</sup> sensor neurocalcin δ and hormone ANF via two distinct and non-overlapping transduction modes, as elucidated by Duda et al.

Ca2+-binding proteins have been found to be involved in the complex etiology of psychiatric disorders. By combining a stereology-based approach and molecular analyses Lauber et al. investigated the involvement of parvalbumin in autism spectrum disorder, finding a dysregulation of its expression in Cntnap2 knockout mouse. The review by Mundhenk et al. presents a detailed structural and biophysical characterization of the Ca2+-sensor proteins caldendrin and calneuron-1 and−2, focusing also on their cellular function and their role in neuropsychiatric disorders.

A role for the multi C2-domain protein otoferlin in modulating the Ca2+-triggered exocytosis at the ribbon synapse in mouse inner hair cells is proposed by Takago et al. by a combination of electrophysiology and biochemical analyses. The importance of Ca2+-binding proteins in regulating the fundamental process of exocytosis and synaptic coupling is emphasized in the review by Maj et al., focusing on secretagogin and its novel roles in developing and adult neuronal cells and Bornschein and Schmidt, focused on synaptotagmin 1 and 2 and their role in presynaptic voltage-gated Ca2<sup>+</sup> channel regulation.

Downstream Regulatory Element Antagonist Modulator (DREAM)/KChIP3 exerts multiple functions, including the regulation of A-type outward potassium currents. The contribution by Peraza et al. identifies by biochemical and biophysical investigations a novel ligand of DREAM that modulates Kv4 potassium channels currents, with consequences that are relevant for physiology and disease. The role of DREAM/KChIP3 in pain transmission and its possible involvement in emotional processing was studied by Guo et al., who assess the pain sensitivity and negative emotional behaviors of Kcnip3−/<sup>−</sup> rats and find a possible role for the protein in central nociceptive processing. Novel tools to regulate the role of DREAM in the endoproteolysis of endogenous presenilin-2 in mouse brain are presented by Naranjo et al., who suggest that the interaction between the two proteins may have a therapeutic potential in Alzheimer's disease. Finally, the review by Néant et al. summarizes the current knowledge regarding Ca2<sup>+</sup> signaling in quiescent glioblastoma stem-like cells and discussed how Ca2<sup>+</sup> via KCNIP proteins may affect gene expression in glioblastoma.

Taken together, this Research Topic delivers new visions to our knowledge on NCS proteins and will stimulate future research. We wish to thank all the authors for having submitted papers of high quality and all the reviewers that contributed with constructive and fruitful suggestions.

#### AUTHOR CONTRIBUTIONS

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

**Conflict of Interest:** 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 © 2019 Dell'Orco, Koch, Kreutz, Naranjo and Schwaller. 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) and the copyright owner(s) 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.

# Dysregulation of Parvalbumin Expression in the Cntnap2−/− Mouse Model of Autism Spectrum Disorder

#### Emanuel Lauber, Federica Filice and Beat Schwaller\*

Anatomy Unit, Section of Medicine, University of Fribourg, Fribourg, Switzerland

Due to the complex and heterogeneous etiology of autism spectrum disorder (ASD), identification of convergent pathways and/or common molecular endpoints in the pathophysiological processes of ASD development are highly needed in order to facilitate treatment approaches targeted at the core symptoms. We recently reported on decreased expression of the Ca2+-binding protein parvalbumin (PV) in three well-characterized ASD mouse models, Shank1−/−, Shank3B−/− and in utero VPAexposed mice. Moreover, PV-deficient mice (PV+/− and PV−/−) were found to show behavioral impairments and neuroanatomical changes closely resembling those frequently found in human ASD individuals. Here, we combined a stereology-based approach with molecular biology methods to assess changes in the subpopulation of PV-expressing (Pvalb) interneurons in the recently characterized contactin-associated protein-like 2 (Cntnap2−/−) knockout mouse model of ASD. The CNTNAP2 gene codes for a synaptic cell adhesion molecule involved in neurodevelopmental processes; mutations affecting the human CNTNAP2 locus are associated with human ASD core symptoms, in particular speech and language problems. We demonstrate that in Cntnap2−/− mice, no loss of Pvalb neurons is evident in ASD-associated brain regions including the striatum, somatosensory cortex (SSC) and medial prefrontal cortex (mPFC), shown by the unaltered number of Pvalb neurons ensheathed by VVA-positive perineuronal nets. However, the number of PV-immunoreactive (PV+) neurons and also PV protein levels were decreased in the striatum of Cntnap2−/− mice indicating that PV expression levels in some striatal Pvalb neurons dropped below the detection limit, yet without a loss of Pvalb neurons. No changes in PV<sup>+</sup> neuron numbers were detected in the cortical regions investigated and also cortical PV expression levels were unaltered. Considering that Cntnap2 shows high expression levels in the striatum during human and mouse embryonic development and that the cortico-striato-thalamic circuitry is important for speech and language development, alterations in striatal PV expression and associated (homeostatic) adaptations are likely to play an important role in Cntnap2−/− mice and, assumingly, in human ASD patients with known Cntnap2 mutations.

Keywords: parvalbumin interneurons, calcium-binding proteins, contactin-associated protein-like 2, autism spectrum disorder, hyperpolarization-activated cyclic nucleotide-gated channels, perineuronal nets

#### Edited by:

Oliver Stork, Universitätsklinikum Magdeburg, Germany

#### Reviewed by:

Michael J. Schmeisser, Universitätsklinikum Magdeburg, Germany Yuri Bozzi, University of Trento, Italy Laurence Goutebroze, Institut National de la Santé et de la Recherche Médicale (INSERM), France

> \*Correspondence: Beat Schwaller Beat.Schwaller@unifr.ch

Received: 08 May 2018 Accepted: 12 July 2018 Published: 02 August 2018

#### Citation:

Lauber E, Filice F and Schwaller B (2018) Dysregulation of Parvalbumin Expression in the Cntnap2−/− Mouse Model of Autism Spectrum Disorder. Front. Mol. Neurosci. 11:262. doi: 10.3389/fnmol.2018.00262

#### INTRODUCTION

fnmol-11-00262 July 30, 2018 Time: 16:58 # 2

The etiology of autism spectrum disorder (ASD) is highly complex and diverse. A combination of genetic, epigenetic and environmental factors is assumed to ultimately lead to the ASD phenotype characterized by impairments in social interaction and language/communication, and the presence of restricted or stereotyped patterns of behavior (Kleijer et al., 2014). The genetic architecture of ASD is extremely heterogeneous and difficult to apprehend. Common, rare, and de novo mutations are known to contribute to the genetic risk of ASD and the number of risk genes is steadily increasing (Bourgeron, 2015). Amongst the identified ASD risk genes, contactinassociated protein-like 2 (Cntnap2), coding for a synaptic cell adhesion molecule, is listed as a "strong candidate (2S)" in the Simons Foundation Autism Research Initiative database (SFARI)<sup>1</sup> . The CNTNAP2 gene is one of the largest genes in humans spanning 2.3 Mb of genomic DNA, thus with an increased probability of structural rearrangements such as copy number variations (CNVs), insertions or deletions, single nucleotide polymorphisms (SNPs), altered transcription factor binding and epigenetic modifications affecting the CNTNAP2 locus (Poot, 2015). Several studies identified CNVs, common and rare SNPs and/or polymorphisms in the CNTNAP2 locus associated with syndromic neurodevelopmental disorders such as cortical dysplasia-focal epilepsy (CDFE) syndrome (Strauss et al., 2006), Pitt-Hopkins syndrome (Zweier et al., 2009), Gilles de la Tourette syndrome (Belloso et al., 2007), intellectual disability (ID) (Mikhail et al., 2011), obsessive compulsive disorder (Verkerk et al., 2003), attention deficit hyperactivity disorder (ADHD) (Elia et al., 2010), schizophrenia (Friedman et al., 2008) and autism (Alarcon et al., 2008; Arking et al., 2008). Although patients with CNTNAP2 mutations display a complex phenotype, most of these patients manifest autistic characteristics; language impairments such as dysarthric language, language delay or absent speech/language are particularly frequent (Rodenas-Cuadrado et al., 2014; Penagarikano et al., 2015). Accordingly, Cntnap2−/− mice show impairments in social behavior and reduced vocalizations (Penagarikano et al., 2011; Brunner et al., 2015; Liska et al., 2017), deficits in learning and memory (Rendall et al., 2016) and epileptiform activity (Penagarikano et al., 2011; Thomas et al., 2016).

Due to the complex and heterogeneous etiology of ASD, major efforts are aimed to possibly identifying convergent pathways across different ASD mouse models in order to facilitate treatment approaches. One cell type, whose function is consistently impaired in different ASD mouse models (Wöhr et al., 2015) and post-mortem brains of human ASD patients (Hashemi et al., 2016) and thus representing a promising point-of-convergence with respect to ASD heterogeneity is a subset of GABAergic interneurons expressing parvalbumin (PV); hereafter termed Pvalb neurons. PV is a slow-onset Ca2<sup>+</sup> buffer (more precisely described as a neuronal Ca2<sup>+</sup> signal modulator) and a characteristic marker for fast-spiking interneurons (FSI). The number of Pvalb neurons (solely based on PV immunohistochemistry) was reported to be decreased in the cortex, striatum and hippocampus of Cntnap2−/− mice compared to WT mice (Penagarikano et al., 2011). However, we have shown in several validated ASD mouse models that the previously described "loss" of Pvalb neurons is caused by the down-regulation of PV in some neurons below a detectable threshold, while the number of Pvalb neurons is unchanged (Filice et al., 2016; Lauber et al., 2016). This is of great functional importance and was consistently found in PV+/−, PV−/−, Shank1−/−, Shank3B−/− and in utero VPA-exposed mice (VPA) mice; the genetic ASD mouse models also listed in the SFARI database. Here we show that the number of Pvalb neurons, identified by Vicia Villosa Agglutinin-positive (VVA+) perineuronal nets (PNN), is unchanged in Cntnap2−/− compared to WT mice in the somatosensory cortex, medial prefrontal cortex and striatum, three brain regions often implicated in ASD pathophysiology. However, decreased protein expression of PV was detected exclusively (selectively) in the striatum of Cntnap2−/− mice, as before in Shank3B−/− and VPA mice (Filice et al., 2016; Lauber et al., 2016), further supporting the hypothesis that PV down-regulation represents a common molecular endpoint across different ASD models. These findings thus closely link the Cntnap2−/− model with other previously validated ASD mouse models such as Shank3B−/− and VPA mice.

#### MATERIALS AND METHODS

#### Animals

Mice were group-housed at the University of Fribourg, Switzerland, in temperature-controlled animal facilities (24◦C, 12:12 h light/dark cycle). Mice had free access to water and were fed ad libitum. C57Bl/6J (WT), B6.129(Cg)- Cntnap2tm1Pele/J (Cntnap2−/−), B6.129-Shank3tm2Gfng/J (Shank3B−/−) mice were purchased from Jackson Laboratory (Bar Harbor, ME, United States). B6.Pvalbtm1Swal (PV−/−) mice were generated by homologous recombination as described before (Schwaller et al., 1999). All strains were backcrossed to C57Bl/6J for more than ten generations and therefore considered congenic with C57Bl/6J. Thus, alterations at the level of morphology, biochemistry and molecular biology are assumed to be the result of genotypic differences and not related to mouse strain. Pups obtained from at least 3 different litters per group were used in order to exclude a litter bias. Mice were euthanized either with Esconarkon (300 mg/kg body weight; Streuli Pharma AG, Uznach, Switzerland) for subsequent perfusion, or by cerebral dislocation, if followed by dissection and isolation of selected brain regions. Only male mice participated in the experiments. All experiments were executed with the permission of the local animal care committee (Canton of Fribourg, Switzerland) and according to the present Swiss law and the European Communities Council Directive of 24 November 1986 (86/609/EEC).

<sup>1</sup>https://gene.sfari.org

# Tissue Preparation and Immunohistochemistry

fnmol-11-00262 July 30, 2018 Time: 16:58 # 3

Mice were anesthetized and perfused using 0.9% NaCl solution followed by 4% paraformaldehyde (PFA). Brains were removed and post-fixed in 4% PFA at RT for 24 h before they were transferred to 30% sucrose-TBS 0.1 M, pH 7.3 solution for cryopreservation at 4◦C. After cryopreservation, brains were cut coronally in rostro-caudal direction using a freezing microtome (Frigomobil, Reichert-Jung, Vienna, Austria). Following stereological systematic random sampling principles, every sixth section was collected. Floating sections were then blocked for 1 h at RT in TBS 0.1 M, pH 7.3 containing 10% bovine serum albumin (BSA) and 0.4% Triton X-100. Afterward, sections were rinsed 3 times in TBS before being incubated using a PV antibody (anti-rabbit PV25, Swant, Marly, Switzerland) diluted 1:1000 and Vicia Villosa Agglutinin (biotinylated-VVA, Reactolab, Servion, Switzerland) at 10 µg/ml in TBS containing MgCl2, MnCl2, CaCl<sup>2</sup> (final salt concentration: 0.1 mM each) for 16 h at 4◦C. Next, sections were washed once in TBS and twice in Tris-HCl 0.1 M, pH 8.2 before incubation with the following antibodies: anti-rabbit Cy3 conjugated (diluted 1:200) and Cy2 streptavidin-conjugated (diluted 1:200, Milan Analytic AG, Switzerland) in Tris-HCl for 2 h at RT protected from light. After rinsing, sections were stained with DAPI (diluted 1:1000; LuBio Science GmbH, Luzern, Switzerland) in PBS 0.1 M, pH 7.4. After final rinsing, sections were mounted and cover-slipped using Hydromount mounting medium (National Diagnostics, Atlanta, GA, United States).

# Stereological Quantification and Counting Criteria

A Stereo Investigator system (Version 11, MicroBrightField, Williston, VT, United States) coupled to a Zeiss Axioplan microscope carrying a motorized x-y stage (Ludl Electronic Products, Ltd, Hawthorne, NY, United States) and connected to a Hamamtsu Orca Camera was used. Estimation of the total number of PV-positive (PV+) and Vicia Villosa Agglutinin-binding (VVA+) cells in pre-defined brain regions of interest (ROIs) was performed using the optical fractionator method (West et al., 1991) as described previously (Lauber et al., 2016). Stereotactic coordinates provided by the Paxinos and Franklin atlas (Paxinos, 2001) were used to define ROIs. The striatum (caudate-putamen) was defined at 1.10 to −0.82 mm from bregma, the SSC at 1.94 to −1.82 mm from bregma and the mPFC at 1.94 to 1.10 mm from bregma. Cell counting was performed on images obtained with the following objective lenses: 100x; NA = 1.40, oil immersion for the SSC; and 63x; NA = 1.30, oil immersion for the mPFC and striatum. The Cavalieri estimator (Gundersen et al., 1988) was used to estimate the total volume of the analyzed ROI. 6 WT mice and 5 Cntnap2−/− mice including pups obtained from at least three different litters per group were analyzed. All stereological parameters and results are reported in **Tables 1**, **2**, respectively.

PV<sup>+</sup> and VVA<sup>+</sup> cells were quantified independently and according to the following criteria: (1) presence of a DAPI-stained nucleus; (2) well-defined and roundish perineuronal net (PNN) for VVA<sup>+</sup> cells; (3) nucleus-surrounding PV staining for PV<sup>+</sup>


TABLE 2 | Mean total number of PV<sup>+</sup> and VVA<sup>+</sup> cells in the striatum, SSC and mPFC of WT and Cntnap2−/− mice.


neurons. Examples are shown in **Figure 1B**. At every fifth sampling location, the section thickness was measured and the mean of all measurements was used for computations. The total number of cells (N) in a defined ROI was estimated as proposed by West et al. (1991, 1996) using the formula:

<sup>−</sup> × (1/ssf) × (1/asf) × (1/tsf)

where Q<sup>−</sup> stands for "tops" (counts), and asf, tsf, and ssf for area sampling fraction, thickness sampling fraction and section sampling fraction, respectively. The coefficient of error CE was used to evaluate the precision of the estimates (Gundersen et al., 1999). CEs (m = 1 and m = 0) are provided in **Table 2**. To ensure independent and unbiased cell estimates, PV<sup>+</sup> or VVA<sup>+</sup> cells were quantified without cross-checking the other channel.

Frontiers in Molecular Neuroscience | www.frontiersin.org

N =

XQ

#### TABLE 3 | RT-qPCR primers.

fnmol-11-00262 July 30, 2018 Time: 16:58 # 5


# RT-qPCR

After euthanasia by cervical dislocation, the whole brain of mice was quickly removed and put into ice-cold 0.9% NaCl solution for dissection. After cutting the brain in half along the midline, the cerebellum was removed. Next, the hippocampus followed by the striatum were dissected by carefully removing (pulling) the tissue pieces as described previously (see Figure S2 in Lauber et al., 2016). Unlike in our previous studies, exclusively cortical tissue was obtained by removing the remaining parts of the brain such as thalamus and fiber tracts from the cortex. Tissue samples were separately snap-frozen using liquid nitrogen before storing them at −80◦C. Since RNA is more prone to degradation, the left hemisphere of the brain was dissected first and afterward used for RT-qPCR analysis. RNA was extracted from tissue pieces with the peqGold TRIzol reagent (Peqlab, VWR International GmbH, Erlangen, Germany). Promega's Reverse Transcription Kit (Promega, Dübendorf, Switzerland) was used to synthesize cDNA. RT-qPCR to examine mRNA expression levels for Gapdh, Rn18S, Pvalb, Gad67, Hcn1, Hcn2, Hcn4, Kcnc1, Kcnc2, and Kcns3 genes was performed using the universal 2X KAPA SYBR FAST qPCR Master Mix (Axonlab AG, Mont-sur-Lausanne, Switzerland) and a DNA thermal cycler (Corbett Rotor gene 6000, QIAGEN Instruments AG, Hombrechtikon, Switzerland). Primer sequences and PCR products are reported in **Table 3**. After initial denaturation at 95◦C for 3 min, a two-step protocol was run: 40 cycles comprising denaturation at 95◦C for 3 s and annealing/extension/data acquisition between 54 and 60◦C for 20 s. The housekeeping genes Gapdh (striatum) and Rn18S (18S rRNA) (cortex) were used to normalize the mRNA content for each sample. Finally, the 2<sup>−</sup> <sup>11</sup>Ct method was used to quantify mRNA levels and genes of interest were normalized to I) the housekeeping gene and II) the control, i.e., wild-type group as previously described (Livak and Schmittgen, 2001).

### Western Blot Analysis

Tissue samples from the right hemispheres were sonicated and soluble proteins extracted for Western blotting. Proteins (30 µg/sample) were separated under denaturing conditions by SDS-PAGE (12.5%). Following gel electrophoresis, semi-dry transfer was performed to transfer proteins onto nitrocellulose membranes (MS solution, Chemie Brunschwig, Basel, Switzerland). The membranes were initially blocked at RT for 1 h in 5% BSA in TBS 0.1 M, pH 7.6 followed by incubation with primary antibodies: rabbit anti-PV25 (Swant, Marly, Switzerland), rabbit anti-GAPDH (Sigma-Aldrich, Buchs, Switzerland) both diluted 1:10,000 in TBS with 0.1% Tween (TBS-T) and 2% BSA overnight at 4◦C. On the next day, membranes were rinsed 3 times with TBS-T before being incubated for 1 h at RT in the secondary antibody containing solution: goat anti-rabbit IgG HRP-conjugated (Sigma–Aldrich, Buchs, Switzerland) diluted 1:10,000 in TBS-T. Finally, membranes were washed 3 times with TBS before development using ECL (Merck Millipore, Schaffhausen, Switzerland). Image Studio Light Version 5.0 software was used to quantify visualized bands. GAPDH bands served as loading control on all membranes.

# Statistical Analysis and Cell Number Estimates

A two-tailed, unpaired t-test was used to compare stereological data, mRNA and protein levels between the two groups. Data were analyzed using the R version 3.3.3 software. Stereological data from the two hemispheres of the same

mouse were pooled and analyzed together. A p-value < 0.05 was considered to be statistically significant.

#### RESULTS

#### Unaltered Numbers of Pvalb Neurons, but Decreased PV Expression Levels in the Striatum of Cntnap2−/− Mice

Unbiased stereological analysis of brain sections obtained from postnatal day (PND) 25 ± 1 control C57Bl/6J (WT) and Cntnap2−/− mice was aimed to obtain robust and precise cell number and volume estimates in different brain regions. The optical fractionator method and the Cavalieri estimator were used to quantify cell numbers and volumes in the striatum, somatosensory cortex (SSC) and medial prefrontal cortex (mPFC), three brain regions frequently implicated in ASD pathology. Coefficient of error (CE) values ranged from 0.05 to 0.11 for the WT group and from 0.05 to 0.13 for the Cntnap2−/− group; all stereological parameters and results are reported in **Tables 1**, **2**, respectively. We independently quantified the number of parvalbumin-positive (PV+) and Vicia Villosa Agglutinin-positive (VVA+) cells in the same regions of interest as described before (Filice et al., 2016; Lauber et al., 2016). VVA recognizes N-acetylgalactosamine residues on the surface of perineuronal nets (PNN), which preferentially surround Pvalb neurons (Hartig et al., 1992; Haunso et al., 2000). Thus, Pvalb neurons with strongly decreased PV expression levels (i.e., below the detectable threshold) or even completely devoid of PV, may still be detected by the VVA<sup>+</sup> staining as previously shown in PV+/− and PV−/− mice, respectively (Filice et al., 2016). In line with previous findings in Cntnap2−/− mice that reported a ∼20– 25% decrease in the number of PV<sup>+</sup> neurons in the striatum, cortex and hippocampus at PND14 (Penagarikano et al., 2011), we found a significant decrease in the order of 20% (p = 0.0001) in the total number of PV<sup>+</sup> cells in the striatum of PND25 Cntnap2−/− mice compared to WT mice. Yet the total number of VVA<sup>+</sup> cells was unchanged between the two genotypes, indicating that there was no loss of Pvalb neurons in the striatum of Cntnap2−/− mice (**Figure 1A**). This conclusion was confirmed by the percentage of double-positive cells. As observed in WT mice, ∼90% of PV<sup>+</sup> cells were also VVA<sup>+</sup> in Cntnap2−/− mice; however, only ∼58% of cells out of the VVA pool were detected as PV<sup>+</sup> cells in Cntnap2−/− mice (∼72% in WT), indicating that ∼15% (p = 0.0002) of VVA<sup>+</sup> cells had "lost" their PV signal compared to WT mice (**Figure 1A**). Representative immunofluorescence images of VVA+, PV<sup>+</sup> and double-positive cells in the striatum are shown in **Figures 1B,C**, respectively. The VVA<sup>+</sup> immunofluorescence images corresponding to **Figure 1C** are shown in Supplementary Figure 1. It is evident that while almost all VVA<sup>+</sup> cells co-localized with PV<sup>+</sup> cells in the WT group resulting in red-fluorescent cells with a "yellow rim," several VVA<sup>+</sup> cells were PV-immuno-negative ("green only") in the striatum of Cntnap2−/− mice (**Figure 1B**). The mean volume of the striatum, as determined using the Cavalieri estimator, was unchanged between the two genotypes (**Figure 1A**).

RT-qPCR and Western blot analysis of striatal tissue revealed changes in PV protein expression levels between WT and Cntnap2−/− mice. The trend toward a decrease in Pvalb mRNA levels in Cntnap2−/− mice measured by RT-qPCR did not reach statistical significance compared to WT samples (∼10%, p = 0.3190) (**Figure 2A**). But more importantly at the functional level and consistent with the stereological analysis, Western blot analysis of Cntnap2−/− striatal lysates revealed a ∼22% (p = 0.0219) decrease in PV protein levels compared to WT (**Figure 2B**; uncropped images of all Western blots are shown in Supplementary Figure 2). Representative protein Western blot signals for PV and GAPDH are shown in **Figure 2B**. Taken together, these results confirm that Cntnap2−/− mice exhibit striatal PV down-regulation during early post-natal development without any indication for Pvalb neuronal loss. The non-significant decrease in Pvalb transcript levels hints toward a possible post-transcriptional regulation, unlike in the previously investigated ASD mouse models Shank3B−/− and VPA mice, where both Pvalb transcript and PV protein levels were decreased to a similar extent (Filice et al., 2016; Lauber et al., 2016).

To further support that the total number of striatal Pvalb neurons was unchanged between the two groups, mRNA levels for typical universal GABAergic interneuron markers and Pvalb neuron-specific markers were investigated. Glutamate decarboxylase isoform 67 (Gad67) mRNA levels were not significantly different between WT and Cntnap2−/− mice (**Figure 2C**). Of note, Gad67 is not exclusively expressed in Pvalb neurons, but also in other interneurons. Thus, additional Pvalb neuron-specific markers were analyzed including the potassium voltage-gated channels subfamily C member 1 (Kcnc1), the protein also known as Kv3.1 (Chow et al., 1999) and subfamily S member 3 (Kcns3) coding for Kv9.3 (Georgiev et al., 2012). Both potassium channel transcripts were found to be unaltered between WT and Cntnap2−/− mice (**Figure 2D**). The same held true for Kcnc2 (Kv3.2) (**Figure 2D**), the next closest relative of Kcnc1, which also co-localizes with Pvalb neurons but to a lesser extent than Kcnc1 (Chow et al., 1999).

In an exploratory approach, we quantified mRNA levels for hyperpolarization-activated cyclic nucleotide-gated (HCN) channel isoforms 1, 2, and 4. HCN channel expression and/or the currents mediated by these channels (I<sup>h</sup> current) were previously reported to be altered in cultured hippocampal neurons derived from Shank3−/− mice (Yi et al., 2016) and in VPA mice (Lauber et al., 2016), two validated mouse ASD models with a similar striatal PV deficit as the Cntnap2−/− mice. While Hcn1 and Hcn2 mRNA expression levels in Cntnap2−/− mice did not differ from transcript levels in WT mice in the striatum, Hcn4 mRNA levels were significantly decreased by ∼40% (p = 0.0418) in Cntnap2−/− mice (**Figure 2E**). The conceivable functional implications of these results are discussed below.

#### Lack of Cntnap2 Has No Effect on the Number of Pvalb Neurons and PV Expression in SSC and mPFC

Unlike previously reported by Penagarikano et al. (2011) and by Vogt et al. (2017) in the SSC of PND14 and PND30 Cntnap2−/−

mice, respectively, we did not find Pvalb neuron-specific alterations in the cortex of the same mice at PND25. Neither were PV<sup>+</sup> cell numbers altered in the SSC and mPFC, nor did we observe differences in the number of VVA<sup>+</sup> cells in the same brain regions (**Figures 3A**, **4A**). Accordingly, the percentage of double-positive cells, in both directions (PV pool and VVA pool), was not significantly altered between WT and Cntnap2−/− mice in the SSC and mPFC (**Figures 3A**, **4A**). The volumes of the SSC and mPFC were also similar between groups (**Figures 3A**, **4A**). Representative immunofluorescence images of the SSC and mPFC are depicted in **Figures 3B**, **4B**, respectively. In agreement with the unchanged numbers of Pvalb neurons, subsequent mRNA and protein quantification showed that Pvalb transcript and PV protein levels were essentially identical between WT and Cntnap2−/− mice in cortical tissue obtained from male PND25 mice (**Figures 5A,B**). Similarly, PV protein levels in the hippocampus and cerebellum were not different between genotypes at PND25 (Supplementary Figure 3). RT-qPCR analysis of cortical tissue confirmed the stereological findings. No alterations in the mRNA levels for Gad67, Kcnc1, Kcnc2, and Kcns3 were observed in cortical extracts from Cntnap2−/− mice at PND25 (**Figures 5C,D**). Concomitantly, mRNA levels for the three HCN family members were unchanged in the cortex of Cntnap2−/− mice (**Figure 5E**). We had previously reported about increased Hcn1 mRNA levels in the cortex of VPA mice, which might thus explain some of the electrophysiological deficits that have previously been found in this ASD model (Rinaldi et al., 2007, 2008; Lauber et al., 2016).

# Unchanged Transcript Levels of Pvalb Neuron Markers Kcnc1 and Kcns3, as Well as Hcn Family Members in Adult (PND70) Cntnap2−/− Mice

The role of Cntnap2 in the regulation and electrophysiological properties of Pvalb neurons had been investigated previously. Vogt et al. (2017) had used a cell transplantation approach, where neurons from the medial ganglionic eminence (MGE)

of E13.5 Cntnap2+/+, Cntnap2+/− and Cntnap2−/− mice were transplanted into the cortex of PND1 WT mice. These tdTomato-tagged neurons gave rise also to Pvalb neurons evidenced by their fast, non-accommodating firing properties. After 6–8 weeks in vivo, several electrophysiological parameters were found to be different, depending on whether the MGE-derived Pvalb neurons were from Cntnap2+/+ or Cntnap2−/− mice. In neurons from the latter, the resting membrane potential was more positive, the AP spike half-width increased, the maximum AP slope decreased, and also the input resistance was slightly increased; the authors had hypothesized that these changes might be caused by alterations in K<sup>+</sup> and/or Na<sup>+</sup> channels. Since changes related to K<sup>+</sup> conductances had been observed in striatal neurons with reduced or absent PV expression (e.g., higher firing frequency, increased excitability, slower AHP recovery, more regular inter spike interval) (Bischop et al., 2012; Orduz et al., 2013), we performed Western blot analysis of cortical extracts obtained from adult mice (PND70 ± 1). However, we found no evidence for reduced PV levels in the cortex of PND70 Cntnap2−/− mice (Supplementary Figures 4A,B), as we had also observed in PND25 mice (**Figures 3**–**5**). Moreover, no changes in mRNA levels for Pvalb, Kcnc1 Kcns3, Hcn1 and Hcn4 were detected between WT and Cntnap2−/− mice in the cortex at PND70 (Supplementary Figures 4C,D). Of note, the previously reported differences were observed in the MGE-derived implanted neurons from WT and Cntnap2−/− mice and not between endogenous Pvalb neurons from the two genotypes and might thus explain the apparent discrepancies (see also Discussion).

#### "Analogous" Changes in the Striatum of Mouse ASD Models (Cntnap2−/−, PV−/−, Shank3B−/−, VPA) May Hint Toward Points of Convergence

Based on the hypothesis that various genetic and environmental ASD mouse models might entail similar alterations in the expression levels of proteins present in Pvalb neurons and/or in Pvalb neuron networks, transcript levels of the selective Pvalb neuron markers Kcnc1 and Kcns3, and HCN family members (Hcn1, Hcn2, and Hcn4) in the striatum and cortex were quantified in PV−/− and Shank3B−/− mice and compared to results obtained in Cntnap2−/− mice, the latter also shown in **Figures 2D,E**). While Kcns3 transcript levels were unchanged in both brain regions of PV−/−, Shank3B−/− and Cntnap2−/− mice compared to WT mice (**Figures 6A,C**) and thus supporting the previously drawn conclusion that there is no loss of Pvalb neurons in these ASD mouse models (Filice et al., 2016), there was a trend toward decreased expression of Kcnc1 in the cortex of PV−/− mice (p = 0.0885; **Figure 6C**). A similar down-regulation of Kcnc1 transcript and KV3.1 protein had been observed before in the cortex of VPA-exposed mice (Lauber et al., 2016). mRNA levels for Hcn1, Hcn2 and Hcn4 were unchanged in the cortex of all 3 ASD models (**Figure 6D**). Levels for Hcn2

were also similar in the striatum of all investigated models; however Hcn1 expression levels were prominently decreased in Shank3B−/− mice (∼50%, p = 0.0113). Although Hcn1 signals were rather weak in the striatum, the finding of altered Hcn1 levels in Shank3B−/− mice is of interest, since Hcn-mediated Ih-channelopathy was previously suggested to be a main driver for the ASD phenotype observed in Shank3−/− mice (Yi et al., 2016) (see Discussion). In addition, Hcn4 transcript levels were significantly lower in the striatum of Cntnap2−/− compared to WT mice (**Figure 6B**). While decreases in transcript levels of Pvalb and Kcnc1 are in most cases also mirrored in lower protein expression levels (Filice et al., 2016; Lauber et al., 2016), it is currently unknown whether such a correlation also holds true for Hcn channels.

In summary, our results suggest that Pvalb-neuron numbers are not altered in the investigated cortical regions and moreover that global cortical PV expression levels in PND25 ± 1 Cntnap2−/− mice are unchanged. Yet in the striatum of Cntnap2-deficient mice, we found decreased PV protein expression levels. The implications of striatal PV down-regulation in the various ASD mouse models are discussed below.

#### DISCUSSION

The interest in PV-expressing (Pvalb) interneurons linked to neurodevelopmental disorders such as ASD and schizophrenia

normalized PV/GAPDH signals in the WT group. Right: Representative Western blot signals for PV and GAPDH are shown. All data are expressed as mean ± SEM.

has steadily grown over the past years. The number of PV<sup>+</sup> neurons was reported to be decreased in human ASD patients (Hashemi et al., 2016) and several ASD mouse models (see Table 1 in Wöhr et al., 2015). In most cases, this decrease was assumed to be the result of partial Pvalb neuron loss. Noteworthy, RNA-seq and RT-qPCR analyses on post-mortem tissue samples of temporal and frontal cortex as well as cerebellum from 48 ASD patients and 49 healthy controls showed that the PVALB gene was in the group of most strongly down-regulated genes in the ASD group (Parikshak et al., 2016). So far, mutations in the PVALB gene have not been reported in human ASD patients. Nonetheless, mice with decreased or absent PV levels (PV+/− and PV−/−) show a strong behavioral ASD phenotype and moreover, ASD-associated neuroanatomical changes (Wöhr et al., 2015). These findings support the hypothesis that PV may represent a promising molecular target providing a common endpoint for at least some forms of ASD.

Here, we showed that the number of PV<sup>+</sup> cells is reduced in the striatum of Cntnap2−/− mice at PND25 ± 1. The time point was chosen, since PV expression and the electrophysiological maturation of Pvalb neurons reach adult levels at this age (de Lecea et al., 1995; Okaty et al., 2009). Additionally, at this age, mice with reduced or absent PV levels (PV+/−, PV−/−) manifest an ASD-like behavioral phenotype (Wöhr et al., 2015; Filice et al., 2018). The decrease in PV<sup>+</sup> cells is the result of PV down-regulation and not due to a partial loss of Pvalb neurons, since the number of VVA<sup>+</sup> cells serving as a proxy measure of striatal Pvalb neurons (Hartig et al., 1992) was unchanged between WT and Cntnap2−/− mice. Also transcript levels for specific and well-described Pvalb neuron markers such as KV3.1 (Kcnc1) and KV9.3 (Kcns3) were unaltered in the striatum of Cntnap2−/− mice. In order to understand the mechanisms and the functional consequences of Cntnap2-deficiency resulting in a behavioral ASD-like phenotype in mice, the distinction between Pvalb neuron loss vs. PV-downregulation is of utmost

relevance. A "simple" loss of inhibitory interneurons such as the Pvalb subtype is expected to result in a decrease of the inhibitory tone in the brain. In contrast, Pvalb neurons with decreased or absent PV expression were consistently found to exhibit enhanced short-term facilitation (Schwaller, 2012; Orduz et al., 2013), hence increasing the inhibitory tone. Recently, by using fiber photometry, unperturbed Pvalb neuronal activity in the mPFC of freely moving Cntnap2−/− and WT mice was recorded, when mice were confronted with different social and non-social stimuli (Selimbeyoglu et al., 2017). Distinct activity patterns were observed in the two groups in response to the different stimuli, suggesting that the underlying neural dynamics of Pvalb neurons in the mPFC differed between genotypes. Additionally, expression of stabilized step-function opsin (SSFO)

in Pvalb neurons allowed to optogenetically increasing Pvalb neuron excitability in the mPFC, which rescued the social deficits prevailing in Cntnap2−/− mice and was associated with reduced activity of cortico-striatal projections. This finding is in line with the hypothesis that reduced PV levels are another mean of enhancing the output of PV neurons in order to increase the inhibitory tone in the brain, according to the notion of homeostatic plasticity (Turrigiano, 2011). Interestingly, human patients carrying mutations affecting the CNTNAP2 locus and adult (>6 month) Cntnap2−/− mice display frequent and severe seizures (Strauss et al., 2006; Penagarikano et al., 2011), a state generally linked to an altered excitation/inhibition balance. Optogenetically enhancing Pvalb neuron excitability (i.e., enhancing inhibition) shown to rescue social deficits in Cntnap2−/− mice (Selimbeyoglu et al., 2017), might also have an effect on seizure severity, yet this remains to be proven experimentally. Of note, in PND25 Cntnap2−/− mice, we found no indication of cortical PV down-regulation, thus cortical alterations in Pvalb neuron excitability –if present– may not be linked to PV expression; whereas the reduction of PV levels in striatal Pvalb neurons likely increases excitability as shown before in the same neuron population in PV−/− mice (Orduz et al., 2013). Striatal PV downregulation was found to be a common feature in several prototypical and well-established mouse ASD models. In addition to the present findings in Cntnap2−/− mice, also in Shank3B−/− and in utero VPA-exposed mice, reduced PV levels (in the order of ∼25–50%) and unaltered numbers of Pvalb neurons were reported (Filice et al., 2016; Lauber et al., 2016). This suggests that in these ASD models, the striatum is a particularly susceptible region and that it might play an important role in the development/manifestation of ASD core symptoms in mice. The basal ganglia, where the striatum serves as input structure, have developed to a complex circuitry in higher vertebrates not only mediating motor output control, but also higher cognitive functions such as sensory processing, learning, and memory tasks as well as the generation of intentional and motivated behaviors in a reward-guided manner (Fuccillo, 2016). There is substantial clinical evidence that striatal dysfunction is associated with ASD. Several MRI studies reported about alterations in the volume of the caudate nucleus or putamen (Langen et al., 2007, 2014; Estes et al., 2011) and enhanced striatal connectivity in ASD patients compared to healthy controls (Di Martino et al., 2011). Abnormalities in striatal structure and function have moreover been described for several ASD mouse or rat models, including Shank3−/− mice (Peca et al., 2011; Jaramillo et al., 2016) and in utero VPA-exposed rats (Schneider et al., 2007). Recently, it was demonstrated that targeted depletion of PV-expressing fast-spiking interneurons (FSIs) and large cholinergic interneurons (CINs) specifically in the dorsal striatum of mice leads to spontaneous stereotypy and marked impairments in social interactions (Rapanelli et al., 2017).

Currently, reports about Cntnap2−/− mice in relation to ASD are still sparse. Cntnap2 is a member of the neurexin family of synaptic cell adhesion molecules involved in the clustering of K+ channels in myelinated axons (Poliak et al., 1999, 2003). However, since myelination takes place postnatally and Cntnap2 is strongly expressed during embryonic development, Cntnap2 is expected to fulfill additional functions in early brain development. This assumption is reinforced by the rising number of reports linking mutations in the CNTNAP2 locus with ASD (Rodenas-Cuadrado et al., 2014). Three independent studies reported about reduced spine density in neurons derived from Cntnap2-deficient mice relative to controls, associated with a decrease in synaptic transmission (Anderson et al., 2012; Gdalyahu et al., 2015; Varea et al., 2015). Also perisomatic evoked IPSCs were found to be significantly reduced in hippocampal slices of Cntnap2−/− mice (Jurgensen and Castillo, 2015).

During human and mouse embryonic development, Cntnap2 shows enriched expression in the frontal lobe, striatum and thalamus (Abrahams et al., 2007; Alarcon et al., 2008). Within the cortex, Cntnap2 is particularly abundant in certain interneuron subpopulations: in chandelier cells and PV<sup>+</sup> neurons, as well as in VIP (CR<sup>+</sup> and CCK+) cells (Vogt et al., 2017). The lack of Cntnap2 leads to a decrease in the number of PV<sup>+</sup> neurons, which was interpreted as a defect in the differentiation and/or activity of Pvalb neurons resulting secondarily in decreased PV expression (Vogt et al., 2017) evidenced at PND14. In line, a delayed maturation of Pvalb neurons associated with impaired PV circuit function in ASD has been reported before, e.g., in Shank3−/− mice (Gogolla et al., 2014). As mentioned before, the cortico-striato-thalamic circuitry is crucially involved in higher order cognitive functions and Cntnap2 appears to be particularly important for language development, since most human patients with Cntnap2 mutations manifest some sort of language impairment (see Table 1 in Penagarikano et al., 2015). Compared to WT mice, Cntnap2−/− mice emit less ultrasonic vocalizations in the isolation-induced pup vocalization test (Penagarikano et al., 2011), as well as in response to the female urine stimulus (Brunner et al., 2015). Interestingly, Cntnap2 is also highly expressed within the cortico-striato-thalamic circuitry of the zebra finch songbird, a popular animal model for vocal learning (Panaitof et al., 2010). Lesioning of the lateral magnocellular nucleus of the anterior neostriatum in songbirds results in songs with monotonous repetitions of a single note complex (Scharff and Nottebohm, 1991), thus providing an additional connection between striatal Cntnap2 expression deficits and language-related impairments. Taken together, decreased/absent Cntnap2 expression and/or striatal impairments seem to be potent inducers of ASD core symptoms, with language dysfunction being particularly frequent in patients affected by mutations in the CNTNAP2 gene. The observed down-regulation of PV in the striatum of Cntnap2−/− mice may thus represent an adaptive mechanism to counteract reduced inhibition, as observed in the hippocampus of these mice (Jurgensen and Castillo, 2015), and thus keeping the excitation/inhibition balance within a correct physiological window.

Striatal PV down-regulation during early postnatal development does not seem to be the only similarity between Cntnap2−/− mice, Shank3B−/− and VPA-exposed mice. Using human and mouse neurons, Yi et al. (2016) demonstrated that mutations in Shank3 predispose to ASD possibly by inducing a severe Ih-channelopathy. Hyperpolarization-activated cation (Ih) currents, which are mediated by hyperpolarization-activated

cyclic nucleotide-gated (HCN) channels, regulate membrane resting potentials, input resistance, neuronal excitability and synaptic transmission (Biel et al., 2009). Neurons derived from Shank3−/− mice were found to exhibit severely decreased I<sup>h</sup> currents and furthermore, decreased Hcn4 protein levels in vitro (Yi et al., 2016). When quantifying mRNA levels of the Hcn isoforms 1, 2 and 4 in the cortex and striatum of Cntnap2−/− mice, we found reduced mRNA levels of Hcn4 exclusively in the striatum of Cntnap2−/− mice compared to WT animals, while mRNA levels of the other isoforms were unchanged in both brain regions. Therefore, putative impairments in Hcn-mediated I<sup>h</sup> currents might also contribute to the phenotype of Cntnap2−/− mice. Of note, the observed changes in striatal Hcn4 transcript levels are but a first hint toward possible alterations in the striatum of these mice. Whether Hcn4 protein levels are altered, in which neuron population and to which extend I<sup>h</sup> currents are functionally altered in Cntnap2−/− mice, remains to be thoroughly investigated.

In summary, we show that Cntnap2−/− mice display down-regulation of PV expression in the striatum, a brain structure with enriched Cntnap2 expression during early development and functional importance for higher cognitive functions in later life. Striatal PV down-regulation moreover seems to occur in various ASD mouse models including Cntnap2−/−, Shank3B−/−, in utero VPA-exposed (and evidently PV−/−) mice. Moreover, striatal PV down-regulation was associated with a decrease in Hcn isoform transcripts in Cntnap2−/− and Shank3B−/− mice, representing another point of possible convergence in ASD-implicated mechanisms.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

BS devised the study and took part in data analysis and manuscript writing. EL carried out all the experiments, did the statistical analysis and took part in writing the manuscript. FF participated in setting up the experiments and writing the manuscript. All of the authors read and endorsed the final version of the manuscript.

#### FUNDING

This study was funded by the Swiss National Science Foundation (SNF grant: 310030\_155952/1) and Novartis Foundation (16C172) to BS.

#### ACKNOWLEDGMENTS

The authors appreciate the help of Simone Eichenberger and Martine Steinauer, University of Fribourg, in the maintenance of the mouse facility and technical assistance, respectively; and Markus Wöhr, University of Marburg, for helpful discussions and advice.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00262/full#supplementary-material


GABA interneuron subpopulation in genetic parvalbumin and shank mouse models of autism. Mol. Brain 9:10. doi: 10.1186/s13041-016-0192-8


deletions encompassing critical neurodevelopmental genes in patients with developmental delay, mental retardation, and/or autism spectrum disorders. Am. J. Med. Genet. A 155a, 2386–2396. doi: 10.1002/ajmg.a.34177



**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 MJS and handling Editor declared their shared affiliation at time of review.

Copyright © 2018 Lauber, Filice and Schwaller. 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) and the copyright owner(s) 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.

# Preferential Binding of Mg2<sup>+</sup> Over Ca2<sup>+</sup> to CIB2 Triggers an Allosteric Switch Impaired in Usher Syndrome Type 1J

Rosario Vallone1†, Giuditta Dal Cortivo1†, Mariapina D'Onofrio<sup>2</sup> and Daniele Dell'Orco<sup>1</sup> \*

<sup>1</sup> Section of Biological Chemistry, Department of Neurosciences, Biomedicine and Movement Sciences, University of Verona, Verona, Italy, <sup>2</sup> Department of Biotechnology, University of Verona, Verona, Italy

Calcium and integrin binding protein 2 (CIB2) shares with the other members of the CIB family the ability to bind Ca2<sup>+</sup> and Mg2<sup>+</sup> via two functional EF-hand motifs, namely EF3 and EF4. As a cation sensor, CIB2 is able to switch to a conformation likely associated with specific biological functions yet to be clarified. Recent findings demonstrate the involvement of CIB2 in hearing physiology and a single, conservative point mutation (p.E64D) has been related to Usher Syndrome type 1J (USH1J) and non-syndromic hearing loss. We present an exhaustive biochemical and biophysical characterization of human wild type (WT) and E64D CIB2. We found that CIB2 does not possibly work as a calcium sensor under physiological conditions, its affinity for Ca2<sup>+</sup> (K<sup>d</sup> app = 0.5 mM) being too low for detecting normal intracellular levels. Instead, CIB2 displays a significantly high affinity for Mg2<sup>+</sup> (K<sup>d</sup> app = 290µM), and it is probably Mg2<sup>+</sup> -bound under physiological conditions. At odds with the homologous protein CIB1, CIB2 forms a non-covalent dimer under conditions that mimic the physiological ones, and as such it interacts with its physiological target α7B integrin. NMR spectroscopy revealed a long-range allosteric communication between the residue E64, located at the N-terminal domain, and the metal cation binding site EF3, located at the C-terminal domain. The conservative E64D mutation breaks up such inter-domain communication resulting in the impaired ability of CIB2 to switch to its Mg2+-bound form. The ability to bind the target integrin peptide was substantially conserved for E64D CIB2, thus suggesting that the molecular defect associated with USH1J resides in its inability to sense Mg2<sup>+</sup> and adopt the required conformation.

Keywords: calcium sensor, Usher syndrome 1J, DFNB48, nuclear magnetic resonance, hearing loss, calcium and integrin binding protein, allostery, magnesium

#### INTRODUCTION

Calcium and integrin binding protein 2 (CIB2) is a 21.6 kDa protein sharing with the other members of the CIB family the ability to bind Ca2<sup>+</sup> and Mg2<sup>+</sup> via two functional EF-hand motifs, namely EF3 and EF4, therefore switching to a specific conformation likely associated with specific biological functions (Huang et al., 2012). Since its discovery as a ubiquitously expressed DNA-dependent protein kinase interacting protein (Seki et al., 1999), CIB2 has been found to be

#### Edited by:

Teresa Duda, Salus University, United States

#### Reviewed by:

Marina Mikhaylova, Universitätsklinikum Hamburg-Eppendorf, Germany Andrea Sinz, Martin Luther University of Halle-Wittenberg, Germany

> \*Correspondence: Daniele Dell'Orco daniele.dellorco@univr.it

†These authors have contributed equally to this work

> Received: 26 April 2018 Accepted: 20 July 2018 Published: 17 August 2018

#### Citation:

Vallone R, Dal Cortivo G, D'Onofrio M and Dell'Orco D (2018) Preferential Binding of Mg2<sup>+</sup> Over Ca2<sup>+</sup> to CIB2 Triggers an Allosteric Switch Impaired in Usher Syndrome Type 1J. Front. Mol. Neurosci. 11:274. doi: 10.3389/fnmol.2018.00274 expressed in a variety of tissues but its physiological role remains largely unknown. It has been established that CIB2 binds specifically to the integrin α7B cytoplasmic domain (Häger et al., 2008; Huang et al., 2012), however the protein interacts also with the αIIb integrin (Huang et al., 2012), thus broadening its potential involvement in a variety of signal transduction processes.

Recent lines of evidence demonstrate the direct involvement of CIB2 in hearing physiology, as CIB2 knockout mice showed abolished mechanoelectrical transduction in auditory cells leading to profound hearing loss (Wang et al., 2017). Interestingly, four missense mutations in the gene encoding for CIB2 [p.F91S, p.C99W, p.I123T (Riazuddin et al., 2012) and more recently p. R186W (Patel et al., 2015)] have been found to be associated with non-syndromic deafness (DFNB48) while a single, conservative point mutation (p.E64D) (Riazuddin et al., 2012) has been related to Usher Syndrome type 1J (USH1J, OMIM entry: 614869), a genetic disorder characterized by hearing loss and progressive vision loss due to retinitis pigmentosa. Altogether, these recent findings suggest that CIB2 is an essential component for the normal development of both hair cells and photoreceptor cells.

Among the members of the CIB family, CIB1 is the protein that has been characterized in deeper biochemical and structural detail (Yamniuk et al., 2004, 2006, 2007; Gentry et al., 2005; Yamniuk and Vogel, 2005; Huang et al., 2011). CIB1 and CIB2 are homologous proteins, but both the sequence identity (37.4%) and the overall similarity (60%) are not extremely high. Differences are found throughout the primary structure and importantly, key residues are substituted in the metal-ion binding sites EF3 and EF4 (**Figure 1**). These differences may be reflected in an overall distinct structure/function behavior of CIB2 as compared to CIB1. To date, only few studies have focused on the biochemical and biophysical characterization of CIB2 (Häger et al., 2008; Blazejczyk et al., 2009; Huang et al., 2012) and a comprehensive picture that allows a molecular-level understanding of its biological properties under physiological conditions as well as their alteration in USH1J is currently missing.

In this work, we present an exhaustive characterization of human wild type (WT) and E64D CIB2 by using an integrated biochemical and biophysical approach to highlight the molecular defects of the variant associated with USH1J. Interestingly, we found that CIB2 does not possibly work as a calcium sensor under physiological conditions, because its affinity for Ca2<sup>+</sup> is too low for normal intracellular levels. Instead, CIB2 has a fairly high affinity for Mg2<sup>+</sup> and it is probably Mg2+-bound under physiological conditions. At odds with CIB1, which is monomeric both when isolated and when interacting with its target (Gentry et al., 2005), we found that CIB2 is a non-covalent dimer under conditions that mimic the physiological ones, and as such it interacts with its physiological target α7B integrin. NMR spectroscopy revealed a long range allosteric communication between the residue E64, located at the N-terminal domain, and the metal cation binding site EF3, located at the C-terminal domain (**Figure 1**). The E64D mutation associated with USH1J, although conservative, apparently breaks up such inter-domain communication resulting in the impaired ability of CIB2 to switch to its Mg2<sup>+</sup> (and Ca2+)-bound form, thus suggesting that the molecular defect associated with CIB2 and causing USH1J resides in its inability to sense Mg2<sup>+</sup> and adopt the required conformation.

# MATERIALS AND METHODS

#### Materials

QuikChange II Site-Directed Mutagenesis kit was purchased from Agilent. The Bradford reagent was purchased from Bio-Rad. Chromatographic columns were purchased from GE Healthcare, and synthetic oligonucleotides were from Eurofins. All other chemicals were purchased from Sigma Aldrich. All purchased chemicals were of the highest commercially available purity grade.

#### Expression and Purification of Human Wild Type and E64D CIB2

The cDNA of human CIB2 isoform 1 (Uniprot entry 075838- 1) was cloned into a pET24a(+) vector (Genscript) containing a 6xHis-tag at the N-terminal, followed by Tobacco Etch Virus (TEV) cleavage site. The plasmid was used to transform BL21 DE3 cells. Cells were grown in LB medium or in M9 minimal medium supplemented with <sup>15</sup>NH4Cl (1 g L−<sup>1</sup> ) as a sole nitrogen source for NMR studies, at 37◦C until the OD<sup>600</sup> reached a value around 0.4. Flasks were then cooled down and, after induction by 0.5 mM ITPG at OD<sup>600</sup> = 0.6, bacteria were grown at 15◦C for 20 h. After centrifugation at 5,500 g (20 min at 4◦C) the obtained pellets were suspended in lysis buffer (20 mM TRIS pH = 7.5, 0.5 M NaCl, 20 mM imidazole, 1 mM DTT, 5 U/mL DNAse, 0.1 mg/mL lysozyme, 1 mM PMSF, 2.5 mM MgCl2) and incubated at 25◦C for 30 min. In addition, 10–12 sonication cycles on ice, 10 s each, were performed. Soluble and insoluble fractions were separated by centrifugation at 16,000 g, 4◦C for 30 min. WT CIB2 was found in high amount in the soluble phase and it was directly leaded in a 5 mL His-trap FF Crude column (GE Healthcare) previously equilibrated with loading buffer (20 mM TRIS, pH = 7.5, 0.5 M NaCl, 1 mM DTT, 20 mM imidazole). A one-step elution by 500 mM imidazole was chosen after initial tests with a gradient. In order to remove the imidazole excess to allow for TEV-protease activity, His-CIB2 was dialyzed against 50 mM TRIS pH 8, 150 mM NaCl, 1 mM DTT and then incubated with a previously prepared His-tagged TEV-protease (Dal Cortivo et al., 2018) (ratio 1:30) overnight at 8◦C. Tagfree CIB2 was purified from His-TEV and cleaved His-tails by reloading into a His-trap column and collecting the flowthrough. Protein concentration was measured by a Bradford assay optimized for homolog calcium sensor proteins or by using the predicted molar extinction coefficient (ε<sup>280</sup> = 6,400 M−<sup>1</sup> cm−<sup>1</sup> , http://protcalc.sourceforge.net/) and the purity was verified by SDS-PAGE to be at least 90%. Purified WT CIB2 was washed in 20 mM TRIS pH 7.5, 150 mM KCl, 1 mM DTT, using an Amicon concentrator. Protein aliquots were then flash-frozen and stored at −80◦C until use.

The E64D point mutation was obtained by site-directed mutagenesis on the complete cDNA of CIB2 using a forward

primer (5′ -ATCATTCAAATGCCGGACCTGCGTGAGAAC CCGTT-3′ ) and a reverse primer (5′ -AACGGGTTCTCA CGCAGGTCCGGCATTTGAATGAT-3′ ). Protein expression was performed as for the WT but the mutant protein was found to concentrate in the insoluble fraction, thus requiring purification from the inclusion bodies. After cell lysis, the insoluble pellets were suspended in the unfolding buffer (20 mM TRIS pH 7.5, 0.5 M NaCl, 6M guanidine hydrochloride, 20 mM imidazole, 1 mM DTT) and incubated overnight at 4◦C. Unfolded CIB2 was loaded into a His-trap column and then renatured by a gradient from 0 to 100% of refolding buffer (unfolding buffer guanidine hydrochloride-free) setting the flow rate to 1 mL/min (100 mL total volume). After elution with 500 mM imidazole E64D CIB2 was treated as the WT protein.

position and labeled in the zoomed-in protein cartoon on the right.

#### Peptides

The peptide corresponding to the membrane proximal segment of the cytoplasmic domain of α7B integrin (Uniprot entry: Q13683) comprised between residues 1101–1116 (α7B\_M, Ac-LLLWKMGFFKRAKHPE-NH2) and a scrambled peptide obtained by shuffling the α7B\_M sequence (Scrb, Ac-KEFWGLHAKPRLKLMF-NH2) were synthesized by GenScript USA Inc. (New Jersey, 144 USA). The purity of peptides, estimated by HPLC, was ≥95% and concentration was determined using the predicted molar extinction coefficient (ε<sup>280</sup> = 5,690 M−<sup>1</sup> cm−<sup>1</sup> , http://protcalc.sourceforge.net/).

# Circular Dichroism Spectroscopy and Thermal Denaturation Profiles

Secondary and tertiary structures of WT and E64D CIB2 and thermal denaturation profiles were investigated by using a Jasco J-710 spectropolarimeter equipped with a Peltier type cell holder, using protocols previously described (Astegno et al., 2014; Marino et al., 2014, 2015a,b; Vocke et al., 2017). Briefly, near UV (320–250 nm) and far UV (250–200 nm) spectra of ∼30µM and 12µM CIB2 respectively were collected at 37◦C after consecutive additions of 0.5 mM EDTA, 1 mM Mg2<sup>+</sup> and 1 mM Ca2+. Quartz cuvettes were used both for near UV (1 cm) and far UV (0.1 cm). Solvent spectra were recorded and considered as a blank.

Thermal denaturation profiles were collected in the same conditions as for far UV spectra by monitoring ellipticity signal at 222 nm in a temperature range between 4 and 70◦C (scan rate 90◦C/h).

Titration experiments were designed starting from the apo WT and E64D CIB2 in order to estimate an apparent K<sup>d</sup> value (K<sup>d</sup> app) for calcium and magnesium binding, similar to what was done previously for calmodulin (Maune et al., 1992). The dichroic signal (in terms of molar ellipticity per residue (MRE) at 2 = 222 nm, indicative of typical acquirement of secondary structure, was followed as a function of the concentration of free Mg2<sup>+</sup> or Ca2+. In order to obtain a controlled free ion concentration under well-defined pH and salt conditions, the MaxChelator software (http://maxchelator.stanford.edu/) was used. Each titration point represents an independent sample where 0.5 µL of a Ca2<sup>+</sup> or Mg2<sup>+</sup> stock solution at the appropriate concentration was added to the fixed volume (200 µL) of 12µM CIB2 aliquots in the presence of 1 mM EGTA. After 3 min incubation at 25◦C, three replicas of each spectrum were collected.

#### Fluorescence Spectroscopy

Fluorescence spectra were obtained with a Jasco FP750 spectrofluorimeter. The interaction of WT CIB2 with α7B\_M and Scrb peptides and that of E64D with α7B\_M was studied by monitoring the peptide intrinsic fluorescence. The only Trp residue of both peptides was selectively excited at 295 nm and fluorescence emission was recorded from 300 to 400 nm, setting 5 nm excitation and emission bandwidths. Two scan averaged spectra were recorded. α7B\_M (4µM) was titrated with increasing concentrations of WT or E64D CIB2 in 20 mM Hepes, 150 mM KCl, 1 mM DTT pH 7.5 at 37◦C in the presence of 1 mM Mg2<sup>+</sup> and 1 mM Ca2+. Titration experiments were performed by monitoring the change (blue-shift) in wavelength (λ) of the peptide emission peak on the fluorescence spectrum. The apparent equilibrium dissociation constant (Kd) was calculated by using the following equation:

$$\wp = \wp\_0 + a\chi / (K\_d + \varkappa) \tag{1}$$

where y<sup>0</sup> is the wavelength of the peptide emission peak in the absence of WT/E64D CIB2, a is the difference between the maximum and minimum (1/λ) × 10<sup>5</sup> values of the peptide emission peak as a function of x, the concentration of CIB2.The intrinsic fluorescence emission of the single Trp of the Scrb peptide (4µM) was measured in the same buffer at 37◦C, in the presence of 2, 4, and 8µM WT CIB2.

8-Anilinonaphthalene-1-sulfonic acid (ANS) fluorescence was used to probe the changes in hydrophobicity of WT and E64D CIB2 upon Mg2<sup>+</sup> and Ca2<sup>+</sup> binding. Two micromolar of WT or E64D CIB2 in 20 mM Hepes pH 7.5, 150 mM KCl, 1 mM DTT, was incubated with 30µM ANS and fluorescence was measured after the addition of 0.5 mM EDTA, 1 mM Mg2<sup>+</sup> and 1 mM Ca2+. ANS fluorescence spectra were recorded at 37◦C in the 400– 650 nm range after excitation at 380 nm, with 5 nm bandwidths. Three scan averaged spectra were recorded.

#### Size Exclusion Chromatography

The molecular weight (MW) of the Ca2+-free, Mg2+-bound and Mg2+/Ca2<sup>+</sup> -bound states of WT and E64D CIB2 was determined by size exclusion chromatography (SEC) in an ÄKTA FPLC system using a Superose 12 column (10/300GL, GE Healthcare). Standard proteins for calibration were: carbonic anhydrase (29 kDa), alcohol dehydrogenase (150 KDa), βamylase (200 kDa), and cytocrome c (12.4 kDa). The column was equilibrated with a buffer containing 20 mM Tris pH 7.5, 150 mM KCl, 1 mM DTT with either 3 mM EGTA or 2 mM EGTA + 3 mM Mg2<sup>+</sup> or 3 mM Mg2<sup>+</sup> + 2 mM Ca2<sup>+</sup> added. WT (100µM) or E64D (70µM) CIB2 were incubated with 3 mM EGTA or 2 mM EGTA + 3 mM Mg2<sup>+</sup> or 3 mM Mg2<sup>+</sup> + 2 mM Ca2<sup>+</sup> at 25◦C for 5 min before being applied to the column. The protein elution profile was recorded at 280 nm; elution volumes V<sup>e</sup> were determined and the distribution coefficient K<sup>d</sup> was calculated according to the equation:

$$K\_d = (V\_\varepsilon - V\_0) / (V\_t - V\_0) \tag{2}$$

in which V<sup>t</sup> is the total column volume and V<sup>0</sup> is the void volume. Molecular weights were determined from a calibration plot of log(MW) vs. Kd.

#### Native Page

In order to investigate CIB2 oligomeric state with another approach, the Ferguson plot technique (Ferguson, 1964) was used. Three continuous gels (lacking a stacking phase) under non-denaturing conditions were polymerized at increasing acrylamide concentration (10%, 12%, 15%) using two different BSA concentrations as standards (0.25, 0.41 mgmL−<sup>1</sup> ). Twenty micromolar of CIB2 was incubated at room temperature with EGTA (4.5 mM), Mg2<sup>+</sup> (3 mM EGTA + 4.5 mM Mg2+) or both Ca2<sup>+</sup> and Mg2<sup>+</sup> (3 mM Ca2+, 4.5 mM Mg2+) in the presence of 1 mM DTT for 20 min. Samples were loaded in each gel and let run as in a normal electrophoresis experiment for 40 min, 200 V at room temperature. Bands were visualized by Comassie Blue staining. Data analysis was performed as explained in (Ferguson, 1964).

#### Dynamic Light Scattering

Dynamic light scattering (DLS) measurements were performed with a Zetasizer Nano-S (Malvern Instruments) and polystyrene low volume disposable sizing cuvettes (ZEN0112) using a general setup optimized previously (Sulmann et al., 2014; Marino et al., 2015a, 2017; Vocke et al., 2017).Viscosity and refractive index were set to 0.6864 cP and 1.33 (default values for water), respectively; the temperature was set to 37◦C, with 2 min equilibration time. The measurement angle was 173◦ backscatter, and the analysis model was set to multiple narrow modes. For each measurement, 12 determinations were performed, each consisting of 14–16 repetitions. DLS measurements were performed on the samples of dimeric WT or E64D CIB2 in 20 mM Tris–HCl pH 7.5, 150 mM KCl, 1 mM DTT with 3 mM EGTA or 2 mM EGTA + 3 mM Mg2<sup>+</sup> or 3 mM Mg2<sup>+</sup> + 2 mM Ca2+, immediately after their purification by SEC. Each measurement was run for 5 h. The samples were filtered through an Anotop 10 filter (Whatman, 0.02µm) before each measurement.

#### Nuclear Magnetic Resonance Experiments and Data Analysis

NMR spectra were acquired on a Bruker Avance III spectrometer (Bruker, Karlsruhe, Germany) operating at 600.13 MHz proton Larmor frequency, and equipped with a cryogenic probe. The spectra were recorded at 25◦C, the samples were at protein concentration of 320µM (unless otherwise specified) in 20 mM Hepes, 100 mM KCl, 1 mM DTT, pH 7.5 and 7% D2O.

A standard <sup>1</sup>H-15N heteronuclear single-quantum coherence (HSQC) pulse sequence was used, with pulsed field gradients for suppression of the solvent signal and cancellation of artifacts. <sup>1</sup>H-<sup>15</sup>N HSQC spectra were acquired with a data matrix consisting of 2K (F2, <sup>1</sup>H) × 256 (F1, <sup>15</sup>N) complex points, spectral windows of 8417.509 Hz (1H) × 2189.44 Hz (15N), 8 transients, and 1.5 s relaxation delay.

NMR titration experiments were run on 320µM <sup>15</sup>N-WT CIB2 with Ca2<sup>+</sup> ion added stepwise from a concentrated stock solution. The following protein/ligand ratios were analyzed by <sup>1</sup>H-15N HSQC spectra: 1:1, 1:3, 1:5, 1:7, 1:10, 1:15, 1:20. Intensity perturbations were computed as: I/Imax, where I is the signal intensity at titration step analyzed, and Imax is the maxium signal intensity at the last titration point.

The K<sup>d</sup> values were obtained by fitting the NMR isotherms to a singlestep one-site binding model (Equation 3: **Figures 5D,F**) or to a single step with Hill slope binding model (Equation 4: **Figure 5E**), using GraphPad software according to the following equations:

$$\begin{split} I/I\_{\max} &= \langle (\mathcal{K}\_d + [L]\_t + [P]\_t) - \langle (\mathcal{K}\_d + [L]\_t \\ &\quad + [P]\_t)^2 - 4[L]\_t [P]\_t)^{0.5} \rangle / 2 [P]\_t; \end{split} \tag{3}$$

$$I/I\_{\text{max}} = B\_{\text{max}} [L]^h / (K\_d^{\text{approx}} + [L]^h) \tag{4}$$

where I/Imax is the relative intensity observed at each titration point, [P]<sup>t</sup> and [L]<sup>t</sup> represent the total protein and ligand concentration, respectively, [L] is approximated by [L]<sup>t</sup> and K<sup>d</sup> is the dissociation constant of the complex.

For Equation (4) K<sup>d</sup> app is the apparent equilibrium dissociation constant, h is the Hill slope, and Bmax is the maximum intensity observable. All data were processed and analyzed using TOPSPIN 3.2 (Bruker, Karlsruhe, Germany) and CARA software.

#### Building of WT CIB2 Homology Model

The homology model of WT CIB2 monomer was built using the MODWEB-MODBASE server version r189 (Pieper et al., 2014). Briefly, 4 out of 189 structural models were selected based on the MPQS, TSVMOD, LONGEST\_DOPE and DOPE criteria. The most reliable model, covering the 13–187 region of the full protein sequence, was built based on the X-ray structure of human CIB1 [PDB entry: 1XO5 (Gentry et al., 2005) chain A], which shares 39% sequence identity with CIB2. Ca2<sup>+</sup> ions were manually positioned in EF3 and EF4 binding sites based on the experimental coordinates of the Ca2+-loaded CIB1 structure (1XO5.pdb) (Gentry et al., 2005). The structure was energyminimized in two steps, first with the steepest descent and then with the conjugate gradients algorithm, keeping in both cases the position of the backbone atoms restricted, according to a previous protocol used for other Ca2<sup>+</sup> sensor proteins (Marino et al., 2015b; Marino and Dell'Orco, 2016).

# RESULTS

### Wild-Type and E64D CIB2 Form Non-covalent Dimers With Different Colloidal Properties

We investigated the oligomeric state of both WT and E64D CIB2 by three different approaches, namely PolyAcrylamide Gel Electrophoresis under non-denaturing conditions (native-PAGE), analytical SEC and DLS. **Figure S1a** shows that, under denaturing conditions (SDS-PAGE), the electrophoretic mobility of CIB2 is compatible with that of a 21.6 kDa protein, although the band is shifted to a slightly higher molecular weight, as previously observed in other studies (Blazejczyk et al., 2009; Huang et al., 2012). Gels obtained under non-denaturing conditions show that in the presence of a reducing agent (1 mM DTT) both apo (lane 2) and Ca2+-bound (lane 4) CIB2 run as single bands (**Figure S1b**). However, multiple bands were observed in the absence of DTT independent of the presence of Ca2<sup>+</sup> (lanes 1 and 3). This is compatible with the formation of covalent oligomers due to disulfide bridges resulting from the oxidation of thiol groups in either of the four Cys residues.

In order to investigate the nature of the single bands observed in the native-PAGE experiments, we performed analytical SEC of both WT and E64D CIB2 in the apo form as well as in the presence of Mg2<sup>+</sup> and Ca2+/Mg2<sup>+</sup> and determined the molecular weight (MW) by using a calibration curve shown in **Figure S2**. As previously observed for other Ca2+/Mg2<sup>+</sup> sensor proteins (Sulmann et al., 2014; Marino et al., 2015a; Astegno et al., 2016, 2017; Vallone et al., 2016) the elution profile was sensitive to the metal ion loading state (**Figure S2**) and resulted in an apparent lower MW for the Ca2+/Mg2<sup>+</sup> bound form of CIB2 compared to the apo-form (**Table S1**). Under reducing conditions, the elution profiles for both WT and E64D CIB2 variants in the Ca2+/Mg2<sup>+</sup> form (MW = 37–39 kDa, **Table S1**) were compatible with a dimer and incompatible with a monomer (MW∼22 kDa). The dimeric nature of CIB2 under all the tested conditions was further confirmed by Ferguson plots, which estimated a MW in the 51– 53 kDa range for both WT and E64D variants independent on the cation loading state (**Table S1**).

Samples from SEC experiments were further analyzed by DLS immediately after elution. Results are reported in **Figure 2**. The DLS intensity profile of WT CIB2 in the absence of Ca2<sup>+</sup> and Mg2<sup>+</sup> showed multiple peaks and a generally high polydispersity (**Figure 2A**), with two not-well separated prevailing peaks. However, the addition of Mg2<sup>+</sup> (**Figure 2B**) or Ca2<sup>+</sup> (**Figure 2C**) led to a general improvement of the colloidal properties and a single prominent peak was distinguished in both cases, which allowed the determination of the hydrodynamic diameter (d Mg = 8.43 ± 0.12 nm and d Ca = 8.18 ± 0.01 nm, respectively; see **Figure 2D**). The E64D CIB2 variant instead showed less satisfactory colloidal properties, as under no tested condition, apo (**Figure 2E**), Mg2<sup>+</sup> (**Figure 2F**), or Ca2<sup>+</sup> (**Figure 2G**) could

FIGURE 2 | Dynamic light scattering spectroscopy. Comparison between hydrodynamic diameters measured (T = 37◦C) for WT (A–D) and E64D (E–H) CIB2 in the presence of: 3 mM EGTA (A,E), 3 mM Mg2<sup>+</sup> + 2 mM EGTA (B,F), 3 mM Mg2<sup>+</sup> + 2 mM Ca2<sup>+</sup> (C,G). (D,H) Superimposition of the average curves obtained by (A–C) for the WT (D) and (E–G) for the E64D CIB2 (H).

a single, prevailing peak be observed in the intensity profile. The constant presence of higher-size aggregates and the overlapping of peaks (**Figure 2H**) prevented an estimate of the hydrodynamic diameter to be made for this CIB2 variant.

The differences observed in the DLS profiles of WT and E64D CIB2 under the tested conditions prompted us to analyze the time-dependent properties of the dispersions. The mean count rate (MCR) of the samples, which can be indicative of timedependent protein aggregation, was thus followed over time for 5 h (**Figure S3**). Interestingly, for WT CIB2 (**Figure S3a**) no trend was observed under the investigated conditions, but significant fluctuations in the MCR were observed especially in the apo conditions (150–600 kcps), in line with a partly reversible protein aggregation process. Less prominent but still significant MCR fluctuations were observed in the presence of Ca2<sup>+</sup> or Mg2<sup>+</sup> (150–300 kcps). A somewhat different pattern was detected for E64D CIB2 (**Figure S3b**). Both in apo conditions and in the presence of Ca2<sup>+</sup> and Mg2<sup>+</sup> a slow, constantly increasing trend in MCR was observed. In the sole presence of Mg2+, significantly broad fluctuations of MCR (150–400 kcps) were detected, which also showed a slowly increasing trend.

### Apo, Mg2+- and Ca2+-Loaded WT and E64D CIB2 Show Different Levels of Folding

One-dimensional (1D) <sup>1</sup>H NMR spectroscopy and far/near UV CD spectroscopy were used to monitor the folding state of WT and E64D CIB2 under different conditions (**Figure 3**).

1D <sup>1</sup>H NMR spectroscopy is a fast and powerful technique that can provide information on the global fold of a protein. In 1D <sup>1</sup>H NMR spectra, the signal dispersion in the regions of the amide (6–10 ppm), and methyl (−0.5 to 1.5 ppm) protons provides indications on the folded globular state of the proteins. Moreover calcium binding proteins show typical downfieldshifted NMR peaks at ∼10.5 ppm belonging to residues of the EF-hands upon binding of divalent metals (Huang et al., 2012). The 1D <sup>1</sup>H NMR spectrum of WT CIB2 in the absence of metal ions displayed evidence of chemical shift dispersion but also line broadening in the amide region (**Figure 3A**, black line). Addition of 1 mM Mg2<sup>+</sup> promoted a large change in the WT CIB2 spectrum (**Figure 3A**, blue line); the signals appeared more disperse and sharp indicating the ability of the protein to assume a globular folded structure upon binding of the metal cation. Further addition of 1 mM Ca2<sup>+</sup> (**Figure 3A**, red line) caused slight changes in the 1D <sup>1</sup>H NMR spectrum, indicating a low degree of structural rearrangement upon binding of the second ion.

The NMR data recorded on samples of E64D CIB2 drive to different conclusions. The 1D <sup>1</sup>H NMR spectrum of the E64D CIB2 in the absence of metal ions displayed narrow signal dispersion throughout the spectrum and especially in the region of amide protons (**Figure 3D**, black line). The addition of 1 mM Mg2<sup>+</sup> (**Figure 3D**, blue line) did not promote changes in the NMR spectrum clearly indicating the inability of the E64D CIB2 to bind Mg2+. Upon subsequent addition of Ca2<sup>+</sup> the NMR signals appeared more disperse as a consequence of the binding of the metal ion, however the 1D spectrum suggests that the protein still retains a certain degree of flexibility and it is not characterized by a rigid tertiary structure (**Figure 3D**, red line).

Near UV (250–320 nm) CD spectroscopy provides information as to the microenvironment of the aromatic amino acids Phe, Tyr, and Trp, which contribute to the stabilization of protein tertiary structure. CIB2 lacks Trp, therefore near UV CD spectra represent a fingerprint of the possible variations in tertiary structure in the Tyr (5 residues) and Phe (16 residues) microenvironments upon metal cation binding. Monitoring the CD signal in the far UV region (200–250 nm) provides instead information as to

(black dotted), 1 mM Mg2<sup>+</sup> (solid blue lines) and 1 mM Ca2<sup>+</sup> (red solid lines). Temperature was fixed at 37◦C, each spectrum represents the mean of 5 accumulations.

variations of the protein secondary structure. The near UV CD spectrum of WT CIB2 in the absence of metal cations was almost flat (**Figure 3B**, black dotted line), nevertheless some helical content was clearly observed in the far UV region (**Figure 3C**), thus suggesting that apo CIB2 forms a molten globule state, in line with NMR findings. Addition of 1 mM Mg2<sup>+</sup> led to a significant response both in the Phe and Tyr bands (**Figure 3B**, blue dashed line) and to a remarkable increase in the helical content as observed in the far UV region (**Figure 3C**). Notably, upon addition of physiological concentrations of Mg2<sup>+</sup> the far UV spectrum acquired the typical α-helix minima at 208 and 222 nm, while the first minimum was shifted to 206 nm in the apo form (black dotted line). Further addition of 1 mM Ca2<sup>+</sup> refined the near UV CD spectrum especially in the Tyr region (**Figure 3B**, solid red line) and further increased the intensity of the far UV spectrum (**Figure 3C**).

The behavior of E64D CIB2 was substantially different. When exposed to the same Mg2<sup>+</sup> and Ca2<sup>+</sup> conditions, E64D CIB2 showed only minor variations in the near UV region, in line with a substantial conservation of the molten globule conformation independent of the metal cation (**Figure 3E**). Only a slight response to Ca2<sup>+</sup> was observed in the far UV region (**Figure 3F**), however the first minimum at 206 nm did not shift to 208 nm upon addition of Mg2<sup>+</sup> or Ca2+, at odds with the WT variant.

#### Hydrophobicity and Thermal Denaturation Profiles of WT and E64D CIB2

In their apo form, both WT and E64D CIB2 present a partially folded, molten globule conformation, thus suggesting a significant solvent-exposition of hydrophobic patches. We investigated the surface hydrophobicity of CIB2 by using the fluorescent probe ANS, whose fluorescent properties will change as it binds to hydrophobic regions on the protein surface. Results are shown in **Figure S4**. Both ANS fluorescence spectra of WT and E64D CIB2 highlight a significant hydrophobicity of the protein surface under all the tested conditions, as clearly displayed by the remarkable blue-shift of the fluorescence emission maximum (27 nm for WT CIB2, 31 nm for E64D CIB2, in the apo state, **Table S2**) and the relative increase in fluorescence intensity as compared to the emission of ANS alone (2.6- to 2.8-fold, in the apo state, **Table S2**). Addition of 1 mM Mg2<sup>+</sup> or 1 mM Ca2<sup>+</sup> slightly reduced the blue-shift (1– 2 nm, depending on the CIB2 variant, **Table S2**), however the change in relative fluorescence (Fmax/Fref) was higher for WT (2.1–2.2 vs. 2.6) than for E64D CIB2 (2.5–2.6 vs. 2.8). Overall, fluorescence data confirm that both WT and E64D CIB2 are highly hydrophobic and the pathogenic variant maintains higher hydrophobicity in the presence of Mg2<sup>+</sup> or Ca2<sup>+</sup> compared to the WT.

The thermal stability of WT and E64D CIB2 in the 4–70◦C range was investigated by monitoring the dichroic signal at

FIGURE 4 | Thermal denaturation of WT and E64D CIB2. Thermal denaturation profiles were recorded between 4 and 70◦C with 12µM WT CIB2 (A) and E64D CIB2 (B) in the presence of 0.5 mM EDTA (gray circles), 1 mM Ca2<sup>+</sup> (red squares), 1 mM Mg2<sup>+</sup> (light blue diamonds) and 1 mM Ca2<sup>+</sup> + 1 mM Mg2<sup>+</sup> (green triangles). Data fitting was performed using a Hill 4 parameter function, results are shown by solid lines and parameters are reported in Table S2.

FIGURE 5 | <sup>1</sup>H-15N HSQC NMR spectra of WT CIB2 in its apo form and in the presence of Mg2<sup>+</sup> and Ca2<sup>+</sup> highlight an allosteric communication between E64 and N121. (A) Superimposition of the two-dimensional <sup>1</sup>H-15N HSQC NMR spectra of the apo- (black), and Ca2+-bound (green) <sup>15</sup>N-WT CIB2. (B) Superimposition of the HSQC spectra of apo (black), and Mg2+-bound (blue) <sup>15</sup>N-WT CIB2. In the insets, zoom of the HSQC spectra of the downfield peaks of the metal-bound forms of <sup>15</sup>N-WT CIB2. Metal ions were present at a protein:metal ratio of 1:15. (C) Superimposition of downfield region of the <sup>1</sup>H-15N HSQC NMR spectra recorded on <sup>15</sup>N-WT CIB2 containing 15 equivalents of Mg2<sup>+</sup> (blue), 15 eq Mg2<sup>+</sup> + 15 eq Ca2<sup>+</sup> (red), and 15 eq Ca2<sup>+</sup> (green). (D–F) Variation of <sup>1</sup>H-15N HSQC peak intensities of WT CIB2 as a function of Ca2<sup>+</sup> concentration. The peak intensities were normalized with respect to the maximum value. The continuous lines represent the data fitted against equations as indicated in section Materials and Methods. The plots refer to the amide peaks of residues E64 (D), G162 (E), and N121 (F). All the spectra were recorded at 600 MHz and 25◦C. All samples were at protein concentration of 320µM in 20 mM Hepes, 100 mM KCl, 1 mM DTT, pH 7.5.

222 nm, where a minimum was observed in the far UV CD spectrum (**Figure 3**). Thermal denaturation profiles are reported in **Figure 4**. Apo WT CIB2 was found to be rather unstable, having a melting temperature of 35.1◦C (**Table S2**). Addition of Mg2<sup>+</sup> increased the thermal stability of ∼11◦C, and a similar effect was observed for addition of 1 mM Ca2+, although the stabilization was lower (∼8 ◦C). In the presence of both cations the thermal stability resembled that of Mg2<sup>+</sup> (T<sup>m</sup> = 45.9◦C, **Table S2**). The transition was faster in the metal-bound states compared to the apo-state (H<sup>c</sup> = 11–12.5 vs. 7.5, **Table S2**). The persistent CD signal at 222 nm (**Figure 4A**) suggests that the transition ended in a still partially folded structure, independent of the presence of Mg2<sup>+</sup> or Ca2+. Interestingly, the thermal profile of apo E64D CIB2 was unperturbed in the scanned range of temperature, as no transition was observed (**Figure 4B**). Addition of Mg2<sup>+</sup> resulted in a Tm-value ∼ 11.5◦C lower than that of WT CIB2, and a very similar effect was observed after the addition of 1 mM Ca2<sup>+</sup> (**Table S2**). Only the co-presence of both cations slightly increased the thermal stability of E64D CIB2, which however was ∼9 ◦C lower than the respective WT case (**Table S2**). All the transitions observed for E64D CIB2 were significantly slower compared to the respective cases in WT CIB2 (compare the H<sup>c</sup> values, **Table S2**).

# Mg2<sup>+</sup> and Ca2<sup>+</sup> Binding to WT CIB2 Explored by NMR Spectroscopy Reveals an Inter-domain Allosteric Switch

Two-dimensional <sup>1</sup>H-15N HSQC NMR spectra are often employed to investigate protein structural changes. The NMR chemical shift of the signals belonging to all HN groups of the protein is a sensitive reporter of the local and global structure rearrangements occurring upon ligand binding events. The NMR signals of the <sup>1</sup>H-15N HSQC spectrum of the apo <sup>15</sup>N WT CIB2 were broad and poorly dispersed in line with the previous observation that apo CIB2 forms a molten globule state.

In line with previously reported data (Huang et al., 2012), the <sup>1</sup>H-15N HSQC spectrum changed dramatically upon addition of either Ca2<sup>+</sup> or Mg2<sup>+</sup> ions; the NMR signals became more disperse and sharp and new downfield peaks appeared, indicative of metal binding (**Figures 5A–C**).

In order to investigate at a deeper level the structural mechanisms associated with Ca2<sup>+</sup> binding, NMR titration experiments of Ca2<sup>+</sup> into <sup>15</sup>N-WT CIB2 have been performed collecting a series of <sup>1</sup>H-15N HSQC spectra. Notably, the variation of the intensity of the peaks belonging to residues E64, which belongs to the kinked H3b helix in the non-functional EF1 motif and N121 of EF3 loop appeared to be correlated (**Figures 5D,F**), thus indicating that these amino acid residues belong to the same allosteric network. The NMR titration data were fitted assuming a simple one-site binding model which better describes a hyperbolic shape of the curve and K<sup>d</sup> values of 0.55 ± 0.13 and 0.48 ± 0.15 mM were obtained for E64 and N121, respectively.

Moreover, we were able to follow the peculiar behavior of the G162 belonging to the EF4 loop, upon addition of increasing amount of Ca2<sup>+</sup> (**Figure 5E**). Interestingly, the intensity variation upon addition of Ca2<sup>+</sup> had a sigmoidal shape, indicative of positive cooperativity of the binding mechanism. From the fitting of the data a K<sup>d</sup> app value of 2.22 ± 0.25 mM and a Hill coefficient h = 2.27 ± 0.38 were obtained. The data of the two sites with different affinities for the Ca2<sup>+</sup> ion are consistent with the analysis of the <sup>1</sup>H-15N HSQC spectra of the protein in the early steps of titration (data not shown). When only the peaks of E64 and N121 are visible, the protein already adopts a well-folded structure, further addition of Ca2<sup>+</sup> promotes only small changes in the spectrum, thus confirming that the binding of Ca2<sup>+</sup> into the first site triggers the structural rearrangement of WT CIB2.

NMR spectroscopy was also employed to investigate whether WT CIB2 had a preferential binding capability toward Ca2<sup>+</sup> or Mg2+. To this aim we recorded a <sup>1</sup>H-15N HSQC spectrum of <sup>15</sup>N-WT CIB2 after addition of a solution containing equal concentration of the two cations (**Figure 5C**) and we analyzed the downfield peaks as indicators of the binding site occupancy. Interestingly, WT CIB2 displayed a preferential binding site for Mg2<sup>+</sup> in the EF3 loop and for Ca2<sup>+</sup> in the EF4 loop. This observation was confirmed by analyzing the <sup>1</sup>H-15N HSQC spectra of the protein in the presence of different ratios of Ca2<sup>+</sup> or Mg2<sup>+</sup> (**Figure S5**). As expected, when Mg2<sup>+</sup> was in excess we could observe the peak of N121 at the chemical shift of the Mg2<sup>+</sup> bound form, and the peak of G162, although with low intensity, as a reporter of Ca2<sup>+</sup> bound to the EF4 loop. Notably, when Ca2<sup>+</sup> was in excess the protein still retained its capability to bind Mg2<sup>+</sup> in EF3 and we could still observe the peak of N121, typical of the Mg2<sup>+</sup> bound form, while EF4 became occupied by Ca2+.

### The Affinity of WT and E64D CIB2 for Ca2<sup>+</sup> Is Incompatible With a Role as a Physiological Ca2<sup>+</sup> Sensor

NMR experiments suggest that the affinity of WT CIB2 for Ca2<sup>+</sup> is in the submillimolar range. We sought to quantify the affinity for Ca2<sup>+</sup> and Mg2<sup>+</sup> of both WT and E64D CIB2 in a comparative fashion in order to assess the potential role of CIB2 as a sensor protein under physiological and USH1J-related conditions. Titration experiments were performed by monitoring the CD signal at 222 nm, under very carefully determined pH and free cation (Ca2<sup>+</sup> or Mg2+) conditions spanning over their known physiological range. Although the method does not allow attributing the macroscopic association constant to each individual EF-hand, it provides an estimate of the cation concentration at which the conformational change, starting from the apo-state, is half maximal (K<sup>d</sup> app). Therefore, it is a useful approach for comparisons of the two CIB2 variants over a physiological range of cation stimuli. Results are reported in **Figure 6** and **Table 1**.

In line with the data from NMR titrations, the measured apparent affinity for Ca2<sup>+</sup> of both WT (500µM) and E64D (2 mM) CIB2 was extremely low, thus excluding a possible role of CIB2 as a Ca2<sup>+</sup> sensor under physiological conditions (see the section Discussion for further details). An alternative well-established spectroscopic method based on the competition with the chromophoric chelator 5,5′ Br2-BAPTA was applied for WT CIB2, but it failed to detect any individual macroscopic binding constant in the low µM range (results not shown), thus confirming the overall low affinity for Ca2+. Interestingly, WT CIB2 showed a relatively high affinity for Mg2<sup>+</sup> (290µM), compatible with a fully loaded state under physiological conditions. On the contrary, E64D CIB2 is likely incapable of detecting Mg2<sup>+</sup> under physiological conditions due to the low affinity (K<sup>d</sup> app = 1.5 mM).

with Ca2<sup>+</sup> (A) and Mg2<sup>+</sup> (B) starting from the apo-form (1 mM free EGTA). (C,D) Ca2<sup>+</sup> (C) and Mg2<sup>+</sup> (D) titrations of WT (diamonds) and E64D (downwards triangles) CIB2 in the presence of α7B\_M peptide. Ions concentration ranged between 1µM and 10 mM, the obtained data were fitted using the Hill 4 parameters function. Data were normalized on protein concentration (MRE). Titrations were performed at 25◦C, each point represents the mean of 3 accumulations.

All the titration curves could be fitted to a Hill sigmoid function (**Figure 6**), thus suggesting in some cases a cooperative effect of the cation binding on the structural transition. Interestingly, while Ca2<sup>+</sup> binding to WT CIB2 was substantially non-cooperative (H<sup>c</sup> = 1.1, **Table 1**), binding of Mg2<sup>+</sup> showed evidence of positive cooperativity (H<sup>c</sup> = 2.3, **Table 1**). As for E64D CIB2, data suggest positive cooperativity both in the case of Mg2<sup>+</sup> (H<sup>c</sup> = 1.7, **Table 1**) and Ca2<sup>+</sup> binding (H<sup>c</sup> = 2.2, **Table 1**).

The fact that we detected for E64D CIB2 a significantly low affinity for both Ca2<sup>+</sup> and Mg2<sup>+</sup> (**Table 1**) made us wonder if higher concentration of cations could trigger a WT-like conformation. Near UV-CD spectra were thus recorded following sequential additions of increasing Ca2<sup>+</sup> or Mg2<sup>+</sup> (**Figure S6**). In line with the results from far UV-CD spectroscopy and titration experiments, our data show that at a Ca2<sup>+</sup> concentration up to 5 mM E64D CIB2 did not switch to a WT-like three-dimensional conformation; however, at 10 mM Ca2<sup>+</sup> the near UV spectra became similar (compare **Figure S6a** with **Figure 3B**). Interestingly, the same finding did not apply to Mg2+. Even at 10 mM Mg2<sup>+</sup> the near UV CD spectrum did not reach the shape and the intensity observed for the WT case (compare **Figure S6b** with **Figure 3E**). Therefore, our data are overall consistent with the inability of E64D CIB2 to sense Mg2<sup>+</sup> under physiological conditions, and even under conditions that exceed the intracellular levels.

TABLE 1 | Apparent affinity for Ca2<sup>+</sup> and Mg2<sup>+</sup> of WT and E64D CIB2 assessed by CD titrations.


# α7B Integrin Is a Specific Target of Both WT and E64D CIB2

We asked whether E64D CIB2 was still capable of interacting with specific targets of WT CIB2. Based on the results of previous work (Huang et al., 2012), we focused on a peptide (α7B\_M) covering the membrane-proximal CIB2-specific sequence of recognition of the integrin α7B cytosolic domain. As a negative control, we generated a scrambled peptide (Scrb) by shuffling the α7B\_M sequence, thus conserving general physicochemical properties such as net charge and hydrophilicity while losing biological specificity. Both peptides have a single Trp residue, which allowed us to exploit their fluorescence for studying the interaction with both WT and E64D CIB2, which lack Trp residues (see section Materials and Methods). Results from fluorescence experiments are reported in **Figure 7**.

. spectra in 1 mM Mg2<sup>+</sup> and 1 mM Ca2+of 4µM Scrb peptide alone (solid black line) and in the presence of 2µM CIB2 WT (dotted red line), 4µM CIB2 WT (short dashed blue line) and 8µM CIB2 WT (dash dotted green line). (C) Fluorescence titration of 4µM α7B\_M peptide with WT (black circles) and E64D (red triangles) CIB2 in the presence of 1 mM Mg2<sup>+</sup> and 1 mM Ca2+.

The interaction between WT CIB2 and the α7B\_M peptide was apparent, as assessed from the 1.7-fold increase in the maximal fluorescence emission and the 12 nm blues shift (**Figure 7A**) indicative of an augmented hydrophobicity of the peptide Trp residue upon interaction with the protein. On the contrary, no significant change in either the fluorescence emission intensity and the relative maximum wavelength was observed when the experiments were performed with the Scrb peptide, even when the concentration of WT CIB2 was brought up to 8µM, thus indicating the lack of specific binding (**Figure 7B**). Similar results were obtained for the E64D CIB2 variant (results not shown).

Titration experiments were performed to estimate the stoichiometry of interaction between WT/E64D CIB2 and α7B\_M peptide and to assess the apparent affinity. Results are reported in **Figure 7C**. The curves showed that both WT and E64D CIB2 interact with the target peptide with a 2:1 stoichiometry, that is, a CIB2 dimer binds a single peptide. The estimated affinities are similar (K<sup>d</sup> app = 4.99 ± 1.01µM for WT CIB2; K<sup>d</sup> app = 3.1 ± 0.2µM for E64D CIB2; mean ± s.d. of 4 and 3 repetitions, respectively), therefore the USH1J-related variant is still capable of binding the α7B\_M target peptide, with even higher affinity compared to the WT case.

In order to assess if the binding of the target peptide could influence the sensing of Ca2<sup>+</sup> or Mg2<sup>+</sup> and the protein conformation, far UV CD spectra were recorded and titration experiments performed in the same conditions as with the protein alone. **Figure S7** clearly shows that the peptide does not possess any secondary structure and that its incubation with WT CIB2 led to the same spectral properties observed in response to additions of Ca2<sup>+</sup> for the protein alone (compare with **Figure 3C**). Therefore, we conclude that the interaction with the α7B\_M peptide does not induce any appreciable structural change in WT CIB2. Moreover, the interaction with the α7B\_M peptide had a relatively small effect on the Ca2<sup>+</sup> or Mg2<sup>+</sup> sensing abilities of CIB2. While a 1.6-fold increase in the K<sup>d</sup> app was observed for Mg2<sup>+</sup> binding to WT CIB2 (**Table 1** and **Figure 6**), a slightly increased affinity for Ca2<sup>+</sup> was detected, although the Kd app was still quite high for physiological relevance (0.2 mM, **Table 1**). Minor differences were observed in the variation of Kd app in the presence of α7B\_M peptide for E64D CIB2 (1.2-fold increase for Mg2<sup>+</sup> and 1.2-fold decrease for Ca2+; **Figure 6** and **Table 1**).

#### DISCUSSION

The ubiquitous expression of CIB2 in various tissues suggests that it may exert yet unknown biological functions in a broad range of biochemical processes. Besides being involved in hearing physiology and pathology (Riazuddin et al., 2012; Jan, 2013; Patel et al., 2015; Seco et al., 2016; Wang et al., 2017), CIB2 has been indeed found to play a role in congenital muscular dystrophy type 1A (Häger et al., 2008), in the N-methyl-D-aspartate receptor-mediated Ca2<sup>+</sup> signaling in cultured hippocampal neurons (Blazejczyk et al., 2009), in the promotion of HIVviral infection (Godinho-Santos et al., 2016), and very recently it was found to act as a negative regulator of sphingosine kinase 1-mediated oncogenic signaling in ovarian cancer (Zhu et al., 2017). Available mechanistic studies focusing on the Ca2<sup>+</sup> and Mg2<sup>+</sup> sensing properties of CIB2 are just a few, and a comprehensive characterization of the protein in comparison to its disease-associate variants was missing. Indeed, so far much of the molecular interpretation of the processes in which CIB2 is involved has been based on the significantly better explored structure-function properties of the homologous protein CIB1 (Leisner et al., 2016), although the relatively low sequence identity and similarity call for particular caution when inferring common functions for the two proteins.

In this work, we present a thorough characterization of two variants of human CIB2, namely the WT form and the E64D mutant associated with USH1J. It should be reported that a recent study disqualified CIB2 as a USH1J-related gene, however the E64D variant was found to be associated with autosomal recessive non-syndromic hearing loss (Booth et al., 2018). Our biochemical and biophysical study highlights a number of clear structural and functional differences with CIB1, which may thus pose the molecular basis for understanding the malfunctioning of CIB2 in USH1J and possibly other genetic diseases causing hearing loss.

While the general topology of CIB2 is similar to that of CIB1 (**Figure 1**), a first clear difference between CIB1 and CIB2 resides in their oligomeric states. While analytical SEC experiments performed with CIB1 detected a monomeric protein independently on the presence of Ca2<sup>+</sup> and target peptide (Gentry et al., 2005), our SEC data, electrophoresis experiments under non-denaturing conditions (**Figure S2** and **Table S1**) and DLS experiments (**Figure 2**) all converge to CIB2 forming non-covalent dimers both in the apo and in Ca2+/Mg2+-bound conditions. The oligomeric state of CIB2 is particularly relevant for its interaction with biological targets. Although we cannot exclude different situations with different targets, fluorescence titration experiments (**Figure 7**) point to a 1:1 stoichiometry for a CIB2 dimer:α7B\_M peptide complex, at odds with the results observed for CIB1:αIIb peptide complex, where the 1:1 stoichiometry involved a monomeric protein (Gentry et al., 2005).

The stability of the oligomeric state of WT CIB2 was found to be significantly affected by the presence of metal cations and by the presence of the E64D point mutation. DLS spectroscopy highlighted how, in order to achieve a substantially monodisperse protein solution, the saturation with Ca2<sup>+</sup> or Mg2<sup>+</sup> was necessary, as the apo-form was observed to dynamically fluctuate between oligomers of different size (**Figure 2** and **Figure S3**). Surprisingly, the USH1J-associated E64D mutation, that does not change the physicochemical properties of the substituted amino acid, still leads to a dimeric protein (**Figure S2**), which is however more prone to form heterogeneous aggregates over time independent of the presence of Ca2<sup>+</sup> or Mg2<sup>+</sup> (**Figure 2** and **Figure S3**).

Important differences between WT and E64D CIB2 were found in their cation-dependent folding state. A general agreement between <sup>1</sup>H NMR and near and far UV-CD spectra was obtained for both protein variants (**Figure 3**). Indeed, WT CIB2 was found to respond to both Ca2<sup>+</sup> and Mg2<sup>+</sup> by adopting a similar secondary (**Figure 3C**) and tertiary structure (**Figures 3A**,**B**), at odds with the E64D variant, for which 1 mM Mg2<sup>+</sup> was not enough to trigger any detectable switch (**Figures 3D–F**). Further addition of 1 mM Ca2<sup>+</sup> led to a detectable increase in E64D CIB2's tertiary structure (**Figures 3D,E**) although the change was significantly lower compared to that observed for the WT. Besides showing a lower structural responsiveness to Ca2<sup>+</sup> and Mg2<sup>+</sup> compared to the WT, E64D CIB2 was found to have a significantly lower thermal stability under all the tested conditions (**Figure 4** and **Table S2**), and its apo-form apparently maintains a more hydrophobic surface that persists upon exposition to Ca2<sup>+</sup> and Mg2<sup>+</sup> (**Figure S4**). This could also explain while no transition was observed for apo E64D CIB2 upon thermal denaturation in the 4–70◦C range (**Figure 4B**), this form being particularly unstable and unstructured.

2D HSQC NMR experiments shed light on the mechanisms related to Ca2<sup>+</sup> and Mg2<sup>+</sup> binding to WT CIB2 (**Figure 5**). While confirming the molten globule conformation of the apo form, NMR highlighted that the Ca2+- and Mg2+-bound states of WT CIB2 have a rather similar three-dimensional structure. The analysis of the downfield regions permits the distinction of specific Ca2+- or Mg2+-related fingerprints in the observed pattern. In particular, by performing Ca2<sup>+</sup> titrations we observed that the intensity of the peak attributed to N121, which is located in the sixth position of the EF3 metal binding loop (**Figure 1**) shows a very similar trend compared to that of E64, the residue substituted by Asp in USH1J, which is located in the N-terminal domain, far from the metal binding loops (**Figure 1**). This surprising finding suggests that an inter-domain allosteric communication occurs between the EF3 binding loop and E64, which according to the homology model based on the structure of CIB1, forms an electrostatic interaction with R33 (**Figure 1**) and is therefore likely contributing to the stability of the EF1 subdomain. The titration patterns observed by NMR (**Figures 5D–F**) further confirm that EF3 is the first EF-hand to be occupied by Ca2+, followed by EF4, whose structural probe is the G162 residue in the sixth position of the loop (**Figure 5E**). Our data support a model, in which under physiological conditions EF3 is never occupied by Ca2<sup>+</sup> but is instead always Mg2+-bound (**Figure S5**). Ca2<sup>+</sup> will however bind to the EF4 loop under conditions of particularly high Ca2<sup>+</sup> concentration. Moreover, no replacement of Mg2<sup>+</sup> was observed in EF3 following additions of equal amounts of Ca2<sup>+</sup> into Mg2+ bound WT CIB2 (**Figure 5C**).

These findings appear particularly relevant for their physiological implications when considered together with the estimated affinities for Ca2<sup>+</sup> and Mg2<sup>+</sup> of WT and E64D CIB2 (**Figure 6** and **Table 1**). The intracellular concentration of free Ca2<sup>+</sup> oscillates in the 0.1–10µM range (Berridge et al., 1998, 2000) and it is even lower in the outer segments of photoreceptor cells, where a fine regulation of the phototransduction cascade by Ca2<sup>+</sup> and cGMP operates (Koch and Dell'Orco, 2013, 2015). The level of free Mg2<sup>+</sup> in most cells, however, is relatively constant and ranges in the 0.5–1 mM interval (Romani and Scarpa, 1992, 2000). While a 290µM apparent affinity for Mg2<sup>+</sup> (**Table 1**) is consistent with the binding of Mg2<sup>+</sup> to WT CIB2 under physiological conditions, the affinity measured for the E64D variant (1.5 mM) is too low for ensuring sensing capabilities under normal conditions. Moreover, neither WT nor E64D CIB2 could possibly work as Ca2<sup>+</sup> sensors with the apparent affinities detected in our study (500µM and 2 mM, respectively). It should be noticed that other authors (Blazejczyk et al., 2009) previously determined a much higher affinity for Ca2<sup>+</sup> for GST-fused CIB2 by using a TNS fluorescence assay (apparent K<sup>d</sup> = 0.l4µM). Such a high affinity is in contrast with our data based on three different experimental approaches, namely NMR and CD spectroscopic titrations and competition experiments with the 5,5′ Br2-BAPTA chromophoric chelator. This latter approach excluded apparent K<sup>d</sup> values below 6µM (results not shown) and was instead able to detect binding of Ca2<sup>+</sup> to CIB1 (Yamniuk et al., 2008). We don't have an explanation for such discrepancy, except for pointing out that all our experiments were performed with unlabeled and untagged human proteins, while those in Blazejczyk et al. used a TNS-labeled rat CIB2 fused with GST, which might introduce artifacts when probing the protein sensing capabilities.

The lower affinity for Ca2<sup>+</sup> is one of the elements distinguishing CIB2 from CIB1, which binds Ca2<sup>+</sup> with high affinity in EF4 (K<sup>d</sup> = 0.5µM) and with lower affinity in EF3 (K<sup>d</sup> = 1.9µM); binding of Mg2<sup>+</sup> is instead limited to EF3 (K<sup>d</sup> = 120µM) (Yamniuk et al., 2004, 2007). A closer look at the sequence alignment of the EF3 and EF4 Ca2+-binding motifs (**Figure 1**) explains, at least in part, such difference. The high affinity of the EF4 loop for Ca2<sup>+</sup> in CIB1 can be attributed to the optimal pentagonal bipyramid geometry of the Ca2+ coordinating oxygens, also due to the presence of an Asn residue (N169) in the -X position and especially to a Glu residue in the -Z position (E172). This latter constitutes a bidentate ligand providing the highly conserved coordination via the two γcarboxyl groups to the Ca2<sup>+</sup> ion (Gagné et al., 1997; Grabarek, 2011). In CIB2, positions -X and -Z are occupied respectively by G165, lacking contributions from the side chain, and D168, which does not possibly act as a bidentate ligand (**Figure 1**). Moreover, the position occupied by the side chain of D159 in the structural model of CIB2 does not seem optimal for contributing to Ca2<sup>+</sup> coordination even after energy-minimization, at odds with that of the aligned residue D163 in CIB1 (**Figure 1**). Overall, the geometry of the Ca2+-coordinating oxygens in the EF4 loop of CIB2 is thus likely distorted with respect to the canonical one, hence leading to a low affinity for Ca2+. Differences in the sequence of the EF3 loop also distinguish CIB2 from CIB1, which may explain the lower affinity for Ca2<sup>+</sup> shown by the first protein. Position Y in CIB1 is occupied by the negatively charged D118, which is substituted by N118 in CIB2, moreover the -X position, again occupied by an Asn (N124) in CIB1 is occupied by a Cys (C124) in CIB2. Interestingly, CIB1 like CIB2 does not have a Glu at position -Z, but has an Asp instead (D127), thus explaining the lower affinity for Ca2+-compared to EF4.

The fact that all four CIBs have an Asp instead of a Glu residue at position -Z (**Figure 1**) suggests that EF3 can serve as a Mg2+-binding specific motif. Previous mutagenesis studies showed that the replacement of the Asp residue in the 12th position (-Z) of an EF-hand metal binding loop with a Glu increases the affinity for Ca2+and abolishes binding of Mg2+, rendering the site calcium-specific, probably due to the reduced ability of side chains to change conformation (da Silva et al., 1995). We are therefore tempted to generalize that EF3 is the Mg2+-specific binding motif among the CIB family, while Ca2<sup>+</sup> might bind to EF4 under physiological conditions, although this is clearly not the case for CIB2. The conformational switch from a molten globule to a well-defined tertiary structure is likely governed by the acidic residue in the 12th position (- Z) of an EF-hand (Gifford et al., 2007), therefore Mg2<sup>+</sup> seems to be the initiator of the functional switch among the CIB family.

We have also studied the potential effect of a CIB2 specific target on the protein metal cation-sensing ability. Although our investigation has been limited to one of the many possible binding targets of CIB2, namely a peptide covering the membrane-proximal sequence of the integrin α7B cytosolic domain, our data show that the interaction with the target doubles the apparent affinity of CIB2 for Ca2<sup>+</sup> (**Table 1**), however the detected K<sup>d</sup> app is still incompatible with a physiological capability of Ca2<sup>+</sup> sensing. We cannot exclude, however, that in specific cell compartments and/or under specific conditions related to disease and/or cell death, the increased intracellular Ca2<sup>+</sup> and the concomitant presence of a specific target would render CIB2 capable of Ca2<sup>+</sup> sensing, therefore providing specific functions. This is however not possibly the case of E64D CIB2, which showed a mM affinity for Ca2<sup>+</sup> under all the tested conditions, including the presence of the target peptide (**Table 1** and **Figure 6**).

Both WT and E64D CIB2 were shown to bind specifically the α7B\_M target peptide with a low µM affinity comparable to that shown for CIB1-αIIb interaction (K<sup>d</sup> = 1.41–1.02µM, in the presence of Ca2<sup>+</sup> and Mg2+, respectively) (Shock et al., 1999; Yamniuk and Vogel, 2005). Future studies will be necessary to further elucidate the binding thermodynamics of CIB2 to its putative targets, including those belonging to the Usher interactome. A broader set of conditions including different cation concentrations and models that account for the dimeric nature of CIB2 shall be specifically tested. Nevertheless, our data seem sufficient to exclude that the principal dysfunction of the USH1J-associated E64D CIB2 be related to the lack of recognition of specific targets. Instead, our data point clearly to the incapability of this CIB2 mutant to switch to its native, Mg2+-bound conformation (**Figures 3**, **5**, **6**). E64D CIB2 was indeed observed to maintain, under physiological levels of Mg2+, a partially unfolded conformation that makes it significantly less stable and prone to aggregation compared to the WT (**Figure 2** and **Figure S3**).

The switch that allows WT CIB2 to acquire a functional conformation at physiological Mg2<sup>+</sup> appears to be finely regulated by an allosteric, long-range communication connecting EF1 with EF3. Our data are inconsistent with mutations in CIB2 disrupting auditory hair cell calcium homeostasis (Jan, 2013) as with such a low affinity not even the WT protein is expected to be involved in Ca2<sup>+</sup> sensing under physiological conditions. Instead, we propose that the inability to bind Mg2<sup>+</sup> of E64D CIB2 prevents the allosteric regulation that makes the protein switch to the native conformation required for its normal function.

#### AUTHOR CONTRIBUTIONS

RV, GD, MD, and DD planned the experiments and analyzed the results. RV, GD, and MD performed the experiments. DD wrote the manuscript with contributions from all the authors.

#### FUNDING

This work was supported by the research grant Ricerca di Base 2015—project UMBUSH from the University of Verona (to DD).

# ACKNOWLEDGMENTS

The Centro Piattaforme Tecnologiche (CPT) of the University of Verona is acknowledged for providing facilities and technical assistance.

#### REFERENCES


# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00274/full#supplementary-material


domain involves a novel C-terminal displacement mechanism. J. Biol. Chem. 281, 26455–26464. doi: 10.1074/jbc.M603963200


**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 © 2018 Vallone, Dal Cortivo, D'Onofrio and Dell'Orco. 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) and the copyright owner(s) 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 Details of Retinal Guanylyl Cyclase 1/GCAP-2 Interaction

Anne Rehkamp<sup>1</sup> , Dirk Tänzler <sup>1</sup> , Claudio Iacobucci <sup>1</sup> , Ralph P. Golbik <sup>2</sup> , Christian H. Ihling<sup>1</sup> and Andrea Sinz <sup>1</sup> \*

<sup>1</sup> Department of Pharmaceutical Chemistry and Bioanalytics, Charles Tanford Protein Center, Institute of Pharmacy, Martin Luther University Halle-Wittenberg, Halle, Germany, <sup>2</sup> Department of Microbial Biotechnology, Charles Tanford Protein Center, Institute of Biochemistry and Biotechnology, Martin Luther University Halle-Wittenberg, Halle, Germany

The rod outer segment guanylyl cyclase 1 (ROS-GC1) is an essential component of photo-transduction in the retina. In the light-induced signal cascade, membrane-bound ROS-GC1 restores cGMP levels in the dark in a calcium-dependent manner. With decreasing calcium concentration in the intracellular compartment, ROS-GC1 is activated via the intracellular site by guanylyl cyclase-activating proteins (GCAP-1/-2). Presently, the exact activation mechanism is elusive. To obtain structural insights into the ROS-GC1 regulation by GCAP-2, chemical cross-linking/mass spectrometry studies using GCAP-2 and three ROS-GC1 peptides were performed in the presence and absence of calcium. The majority of cross-links were identified with the C-terminal lobe of GCAP-2 and a peptide comprising parts of ROS-GC1's catalytic domain and C-terminal extension. Consistently with the cross-linking results, surface plasmon resonance and fluorescence measurements confirmed specific binding of this ROS-GC peptide to GCAP-2 with a dissociation constant in the low micromolar range. These results imply that a region of the catalytic domain of ROS-GC1 can participate in the interaction with GCAP-2. Additional binding surfaces upstream of the catalytic domain, in particular the juxtamembrane domain, can currently not be excluded.

#### Edited by:

Daniele Dell'Orco, Università degli Studi di Verona, Italy

#### Reviewed by:

Teresa Duda, Salus University, United States Karl-Wilhelm Koch, University of Oldenburg, Germany Ana Méndez, Institut d'Investigació Biomèdica de Bellvitge (IDIBELL), Spain

#### \*Correspondence:

Andrea Sinz andrea.sinz@pharmazie.uni-halle.de

Received: 26 June 2018 Accepted: 27 August 2018 Published: 19 September 2018

#### Citation:

Rehkamp A, Tänzler D, Iacobucci C, Golbik RP, Ihling CH and Sinz A (2018) Molecular Details of Retinal Guanylyl Cyclase 1/GCAP-2 Interaction. Front. Mol. Neurosci. 11:330. doi: 10.3389/fnmol.2018.00330 Keywords: cross-linking, guanylyl cyclase-activating protein (GCAP), interaction site, mass spectrometry, phototransduction, rod outer segment guanylyl cyclase (ROS-GC1)

#### INTRODUCTION

The retinal guanylyl cyclase 1 (ROS-GC1) is a transmembrane protein enabling the light adaption process in the eye's rods and cones (Koch, 1991; Dizhoor et al., 1994; Goraczniak et al., 1994). Dimeric ROS-GC catalyzes the conversion of GTP into cGMP in order to restore cGMP levels in the dark (Koch and Stryer, 1988). After light excitation of rhodopsin, a signal cascade induces the decrease of the intracellular cGMP concentration, resulting in the reduction of the intracellular calcium concentration to ∼50 nM (Gray-Keller and Detwiler, 1994). Low calcium concentrations result in the activation of ROS-GC1 by the guanylyl cyclase-activating proteins 1/2 (GCAP-1/-2) that bind to the 68-kDa intracellular ROS-GC1 domain (Koch et al., 2002). This process is known as phototransduction (Schwartz, 1985; Pugh and Cobbs, 1986). The review by Schwartz (1985) addresses biophysical and electrophysiological properties of the, at that time so-called, "light-sensitive current."

The GCAPs proteins (GCAP-1 and GCAP-2) belong to the neuronal calcium sensor (NCS) proteins that have four characteristic EF-hands. Three of them bind calcium, depending on the intracellular calcium concentration (Palczewski et al., 1994; Dizhoor et al., 1995; Burgoyne and Weiss, 2001). GCAP proteins are active in their calcium-free states and exist as N-terminally myristoylated forms (Hwang and Koch, 2002a). For GCAP-1, the N-terminal myristoyl group affects the protein's calcium sensitivity and activity, although a typical calcium-myristoyl switch, as observed for recoverin, has not been reported for GCAPs (Otto-Bruc et al., 1997; Hwang and Koch, 2002b). In case of GCAP-2, it has been suggested that myristoylation has little influence on the calcium-dependent activation of ROS-GC (Olshevskaya et al., 1997; Hwang and Koch, 2002b). Structures of myristoylated GCAP-1 (Stephen et al., 2007) and non myristoylated GCAP-2 (Ames et al., 1999) have been solved, however to date, no high-resolution structure is available for ROS-GC1.

The exact mechanisms of how GCAP-1 and GCAP-2 activate their target proteins are currently not understood. In particular, the surfaces within ROS-GC1 that interact with GCAP-1/-2 are controversially discussed (Laura and Hurley, 1998; Lange et al., 1999; Sokal et al., 1999; Duda et al., 2005; Peshenko et al., 2015a,b). Several studies indicate that both GCAP proteins share the same interaction sites in the kinase homology domain of ROS-GC1 and implicate that GCAP-1 and GCAP-2 compete for binding (Laura and Hurley, 1998; Peshenko et al., 2015b). In addition, Peshenko et al. showed that the ROS-GC dimerization domain participates in GCAP binding and thus the regulation of the human ROS-GC1 (Peshenko et al., 2015a). Another report suggests that the dimerization domain is not of vital importance for the activator binding, but may be involved in the calcium-dependent signal transduction (Zägel et al., 2013). An alternative point of view is that the juxtamembrane and kinase homology domain (KHD) of ROS-GC's intracellular domain represent the binding site of GCAP-1, while GCAP-2 interacts at the C-terminal region of the catalytic domain (Lange et al., 1999; Duda et al., 2005).

Moreover, it has been shown that dimerization of the catalytic domain can occur in the absence of the signal helix domain (SHD, also termed dimerization domain) as GCAP-2 can activate a ROS-GC1 mutant lacking the SHD (Duda et al., 2012). The <sup>657</sup>WTAPELL<sup>663</sup> motif in the C-terminal part of the KHD of ROS-GC1 engages in signal transfer processes to activate the catalytic domain, but not in the binding reaction to GCAPs (Duda et al., 2011). The hypothesis of separate interaction sites for GCAP-1 and GCAP-2 at ROS-GC1 is supported by affinity determinations via backscattering interferometry using different constructs of the intracellular domain (Sulmann et al., 2017).

To characterize protein interactions, complementary strategies, such as the cross-linking/mass spectrometry (MS) approach can be of advantage (Sinz, 2006, 2018; Rappsilber, 2011; Leitner et al., 2016). Cross-linking reagents covalently connect functional groups of amino acids located at a specific distance that can be bridged by the crosslinker. The identification of cross-linked products can be performed in a "bottom-up" approach, in which the crosslinked sample is enzymatically digested and subsequently analyzed by liquid chromatography coupled with tandem mass spectrometry (LC/MS/MS). Previously, cross-linking/MS studies using GCAP-2 and a ROS-GC peptide, derived from the C-terminal extension of the catalytic domain, allowed to define the structure of a GCAP-2/ROS-GC peptide complex in its Ca2+-bound state (Pettelkau et al., 2012).

To gain further insights into ROS-GC1 activation via GCAP-2, we extended these initial cross-linking studies with the aim to clarify whether GCAP-1 and -2 possess overlapping or separate binding sites in the intracellular region of ROS-GC1. To this end, three ROS-GC peptides were employed (**Figure 1**): Peptide 1 comprises the GCAP-2 binding motif (aa 965-981) (Duda et al., 2005), peptide 2 resembles peptide 1 with an N-terminal extension (aa 942-981), and peptide 3 represents the putative binding motif of GCAP-1 (aa 503-522) (Lange et al., 1999). For our cross-linking studies, we used the in-house developed MS/MS cleavable urea-based cross-linker disuccinimidyl dibutylurea (DSBU) (Müller et al., 2010) as well as the "zero-length" cross-linker 1, 1 ′ -carbonyldiimidazole (CDI) (Hage et al., 2017). The spacer arms of DSBU and CDI are 12.5 and 2.6 Å, respectively. DSBU mainly reacts with amine groups of lysines, while CDI reacts with both amine groups as well as hydroxyl groups of serines, threonines and tyrosines. The 1,3-diallylurea (DAU) cross-linker with a spacer length of ∼10 Å connects exclusively cysteines (Iacobucci et al., 2018b). In addition, the artificial, diazirinecontaining amino acid photo-methionine was incorporated into GCAP-2 to gain complementary structural information on the ROS-GC1 interaction by UV-induced cross-linking (Suchanek et al., 2005; Piotrowski et al., 2015). In order to quantify the interaction between GCAP-2 and ROS-GC1 peptides surface plasmon resonance (SPR) and fluorescence measurements were employed.

In this work, we provide insight into the GCAP-2 binding sites within ROS-GC1 by chemical cross-linking/MS. The majority of cross-links with ROS-GC1 peptide 2 were obtained with the C-terminal lobe of GCAP-2. Via SPR and fluorescence measurements, a dissociation constant in the micromolar range was determined.

**Abbreviations:** CCD, Core catalytic domain; cGMP, Cyclic guanosine monophosphate; CDI, 1,1′ -Carbonyldiimidazole; CMD, Carboxymethyldextran; CTE, C-terminal extension; DAU, 1,3-Diallylurea; DMSO, Dimethyl sulfoxide; DSBU, Disuccinimidyl dibutylurea; DTT, Dithiothreitol; EDC, 1-Ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride; EGTA, Ethylene glycol-bis(2-aminoethylether)-N,N,N ′ ,N ′ -tetraacetic acid; ESI, Electrospray ionization; ExtD, Extracellular domain; FA, Formic acid; GCAP, Guanylyl cyclase-activating protein; GTP, Guanosine-5′ -triphosphate; HCD, Higherenergy collision-induced dissociation; HEPES, 4-(2-Hydroxyethyl)-1-piperazine ethanesulfonic acid; IAA, Iodacetamide; JmD, Juxtamembrane domain; KHD, Kinase homology domain; LC/MS/MS, Liquid chromatography/tandem mass spectrometry; LS, Leader sequence; MS, Mass spectrometry; NCS, Neuronal calcium sensor; NHS, N-Hydroxysuccinimide ester; Photo-Met, Photomethionine; ROS-GC, Rod outer segment guanylyl cyclase; RP, Reversed phase; SDS-PAGE, Sodium dodecyl sulfate-polyacrylamide gel electrophoresis; SHD, Signal helix domain; SPR, Surface plasmon resonance; TFA, Trifluoroacetic acid; TmD, Transmembrane domain; TRIS, Tris(hydroxymethyl)aminomethane.

Peptide 1 is the postulated binding motif of GCAP-2 (aa 965-981) used in previous studies (Pettelkau et al., 2012). Peptide 2 (aa 942-981) is an N-terminal extension of peptide 1 comprising a part of the catalytic domain. Peptide 3 (aa 503-522) is the potential binding motif of GCAP-1 (Lange et al., 1999). Amino acid sequences of all three peptides are displayed.

# MATERIALS AND METHODS

# Expression and Purification of GCAP-2

For the expression and purification of bovine GCAP-2, an existing protocol was modified (Schröder et al., 2011). GCAP-2 was expressed in E. coli BL21 (DE3) cells from vector pET-11a. For in vivo myristoylation, plasmid pBB131 encoding the yeast N-myristoyltransferase I was co-expressed. 60µg/ml myristate (Thermo Fisher Scientific) was added to the culture at an OD<sup>600</sup> of 0.4 (Duronio et al., 1989). At OD<sup>600</sup> = 0.6 and 37◦C, recombinant gene expression was induced with 1 mM IPTG for 4 h. For photo-methionine (photo-Met, Thermo Fisher Scientific) labeling, the compound was added (30 mg/l) at the time of induction. Further incubation was performed in the dark. The procedure of photo-Met incorporation was performed with slight adaptations according to an established protocol (Piotrowski et al., 2015). After harvesting the cells, inclusion bodies were isolated and solubilized. Briefly, the cell pellet was dissolved in 0.1 M TRIS-HCl and 1 mM EDTA (pH 7.0) and cell disruption was performed via French press. The disrupted cells were diluted in half the volume of buffer, consisting of 60 mM EDTA, 6% triton X-100, 1.5 M NaCl (pH 7.0), and incubated on ice for 30 min. Inclusion bodies were centrifuged, the pellet was washed four times with 0.1 M TRIS-HCl and 20 mM EDTA (pH 8.0) and solubilized in 6 M urea. Protein refolding was performed by dialysis against 50 mM TRIS-HCl, 1 mM CaCl2, and 1 mM TCEP (pH 8.0). GCAP-2 was purified by anion-exchange chromatography (HiTrap Q Sepharose HP, GE Healthcare), via a gradient from 0 to 1 M NaCl in 50 mM TRIS-HCl, 1 mM TCEP (pH 8.0), followed by size exclusion chromatography (Superdex 75 pg, 16/600, GE Healthcare) applying 10 mM HEPES, 150 mM NaCl, 2.5 mM TCEP, 10% glycerol (pH 7.5). To separate non-myristoylated from myristoylated GCAP-2, a reverse phase Agilent Eclipse XDB-C8 column (4.6 × 150 mm, 5µm, and 1 ml/min) was used. Recombinant GCAP-2 was eluted by a gradient from 0 to 100% acetonitrile in 0.1% TFA. After drying and resuspensing in 6 M urea, the refolding process was repeated via dialysis in 20 mM HEPES, 1 mM TCEP (pH 7.5) (Hwang and Koch, 2002a). The purity of the protein was confirmed by ESI-MS.

### Cross-Linking Experiments

For all experiments, the myristoylated form of GCAP-2 and peptides (**Figure 1**) derived from bovine ROS-GC1 were used. For the cross-linking reactions with DSBU, CDI (Carbolution Chemicals), GCAP-2 and ROS-GC peptides were used at final concentrations of 10µM in 20 mM HEPES buffer (pH 7.5). First, GCAP-2 was incubated at room temperature for 10 min in the presence of 1 mM CaCl<sup>2</sup> or 10 mM EGTA to obtain the calcium-loaded, non-activating or the calcium-free, activating forms. Before the cross-linking reactions were initiated by adding the cross-linker (stock solution freshly prepared in DMSO, DSBU with a 100-fold and CDI with a 20-fold molar excess over GCAP-2), the samples were incubated with ROS-GC peptides (Thermo Fisher Scientific) for further 30 min. After 30 min at room temperature, the cross-linking reactions were stopped by addition of 20 mM ammonium bicarbonate (DSBU samples) and 0.5 M TRIS/HCl (pH 8.0) (CDI samples). For cross-linking with DAU, GCAP-2 was incubated with the ROS-GC peptide 2 as described for cross-linking with DSBU and CDI. A 100-fold excess of DAU (Iacobucci et al., 2018b) and a 20-fold excess of the photo-radical inducer benzophenone (freshly prepared in DMSO) were added. The cross-linking reaction was performed by UV-A exposure (8,000 mJ/cm<sup>2</sup> ) on ice and quenched with 5 mM dithiothreitol (DTT). For photo-Met cross-linking, labeled GCAP-2 was used at a concentration of 10µM in 20 mM HEPES-buffer (pH 7.5). After 10 min of incubating GCAP-2 at room temperature in the presence of CaCl<sup>2</sup> (final concentration 1 mM) or EGTA (final concentration 10 mM), ROS-GC peptides were added at a 10-fold molar excess (final concentration 100µM) followed by further 30 min of incubation. The crosslinking reaction on ice was induced by irradiation with UV-A light (365 nm, 8,000 mJ/cm<sup>2</sup> ). In case the samples were not immediately applied, they were stored at −20◦C.

#### SDS-PAGE Analysis and Proteolysis

To separate cross-linked species from non-reacted proteins, 4–20% gradient gels (Mini-PROTEAN TGX Gel, Biorad) were applied. The DSBU cross-linked and predicted GCAP-2 peptide (1:1) complexes were excised from the gel, reduced with dithiothreitol (DTT) and carbamidomethylated with iodacetamide (IAA). For in-solution digestion, after denaturation by sodium deoxycholate, CDI and photo-Met cross-linked samples were prepared according to an existing protocol with slight adaptations (Lössl and Sinz, 2016). DSBU-, CDI-, or DAUcross-linked samples, were incubated overnight with GluC (1:20, enzyme:protein ratio) at 37◦C. Photo-Met-cross-linked samples were digested with AspN (1:50 ratio). Subsequently, the samples were digested for 4 h with 250 ng trypsin (all proteases from Promega).

# LC/MS/MS

Separation of the proteolytic peptide mixtures was performed via the Ultimate 3000 RSLC Nano system (Thermo Fisher Scientific). As precolumn, a C8 reversed phase (RP) (Acclaim PepMap, 300µM <sup>∗</sup> 5 mm, 5µm, 100 Å, Thermo Fisher Scientific) was employed. As separation columns, C18 RP (Acclaim PepMap, 75µm <sup>∗</sup> 250 mm, 2µm, 100 Å, Thermo Fisher Scientific) or PicoFrit C18 nanospray columns (75µm ID, 10µm tip, New Objective, packed with ReproSil-Pur 120 C18-AQ, 1.9µm) were employed. The nano-HPLC system was coupled to the nano-ESI source of the Orbitrap Fusion Tribrid or the Orbitrap Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific). The samples were desalted on the pre-column by 0.1% trifluoroacetic acid (TFA) for 15 min. Solvent A was H2O (LC-MS grade, VWR) with 0.1% formic acid (FA), Solvent B consisted of 80% acetonitrile (LC-MS grade, VWR) and 0.08% FA. For peptide separation, an elution gradient (flow rate 300 nl/min) was set to 35% solvent B in 90 min, according to a previous study (Iacobucci et al., 2018b). The data-dependent MS/MS mode was applied for data acquisition. For MS/MS, the most intense signals within 5 s of the previous full MS scan were isolated (isolation window 2 u) and fragmented by higher-energy collision-induced dissociation (HCD) (normalized collision energy 30 ± 3%). The fragment ions were analyzed in the orbitrap. Xcalibur 4.0.27 (Thermo Fisher Scientific) was used to control the data acquisition.

#### Identification of Cross-Linked Products

Cross-linked products were identified with the in-house developed software StavroX (version 3.6.0.1) and MeroX (version 1.6.0.1) (Götze et al., 2012, 2015). The software employs Mascot generic format (mgf) files for data analysis. The maximum mass deviations for precursor and fragment ions were fixed at 3 and 10 ppm. The signal-to-noise ratio was ≥2.0. The following settings were applied to define enzymatic cleavage sites: C-terminal to K and R (for trypsin), C-terminal to D and E (for GluC), N-terminal to D and E (for AspN). Three missing cleavage sites were allowed for amino acid residues K, R, D, and E. For the DSBU and CDI cross-linkers, one cross-linking site was defined for K, the second one for K, S, T, Y and N-termini. Photo-Met can react with all amino acids, while DAU only connects C. All cross-links suggested by the software were manually validated. 3D-protein structures were visualized by PyMOL (0.99rc6, Schrödinger LLC) and cross-links were presented as circular plots by Circos (0.67-7) (Krzywinski et al., 2009).

### Surface Plasmon Resonance (SPR) Measurements

For SPR measurements, the MP-SPR Navi 200 OTSO system (BioNavis) was employed in order to determine dissociation constants for the GCAP-2/ROS-GC peptide interactions. For all measurements 2D (planar) carboxymethyldextran (CMD) hydrogel-coated sensor slides (SPR102-CMD-2D, BioNavis) were applied. All buffers were degassed. The running buffer for the immobilization and peptide measurements consisted of 20 mM HEPES, 0.05% Tween (pH 7.5). The regeneration step occurred via addition of 10 mM glycine, 0.05% Tween (pH 2.0). The flow rate was set to 30 µl/min and the temperature was set to 22◦C. GCAP-2 was immobilized by amine coupling. To this end, the chip surface was activated by the 0.4 M EDC/0.1 M NHS (freshly prepared). Activation was repeated twice with 250 µl solution. 5µM GCAP-2, which had been diluted in 10 mM sodium acetate buffer (pH 3.9), was injected twice for immobilization on the CMD sensor surface. Non-reacted NHS ester groups were deactivated with 1 M ethanolamine injection. For binding measurements, ROS-GC peptides [final concentrations 0.5, 1, 5, 10, 20, and 40 µM (peptides 1 and 2); 5, 10, 20, and 40 µM (for peptide 3)] were diluted in running buffer. The injection time was 420 s at a flow rate of 30 µl/min. SPR Navi control and SPR Navi Data Viewer (BioNavis) were used to control the SPR measurements. Data analysis was performed via the kinetic evaluation tool by the software TraceDrawer (version: 1.8, Ridgeview Instruments AB). The graphs presented in **Figure 4** were generated with OriginPro 2018 (Northampton, MA). To describe the binding behavior between GCAP-2 and ROS-GC peptides 1 and 2, a two-state model with equimolar binding was applied. This interaction model represents an initial interaction event, which then changes into an alternative interaction. While the initial binding is assumed to be weak, the primary complex will rearrange into a secondary complex with stronger interaction.

$$\frac{d\,\mathrm{[B]}}{dt} = \,k\_{\mathrm{d1}} \cdot \mathrm{[AB]} \, - \, k\_{\mathrm{a1}} \cdot \mathrm{[A]} \cdot \mathrm{[B]}\tag{1}$$

$$\frac{\text{d}\,\text{[AB]}}{\text{dt}} = k\_{\text{a1}} \cdot \text{[A]} \cdot \text{[B]} - k\_{\text{d1}} \cdot \text{[AB]} - k\_{\text{a2}} \cdot \text{[AB]} + k\_{\text{d2}} \cdot \text{[AB']} \text{ \{2\}}$$

$$\frac{\text{d}\left[\text{AB}'\right]}{\text{dt}} = \left|k\_{\text{d2}} \cdot \text{[AB]} - k\_{\text{d1}} \cdot \left[\text{AB}'\right] \right.\tag{3}$$

$$\text{Y Y} \sim \text{ [AB]} \; + \; \left[ \text{AB}' \right] \tag{4}$$

The signal amplitude is proportional to the sum of the concentrations of the two complexes AB and AB′ .

$$\text{At } \mathbf{t} = \mathbf{0} \rightarrow \text{ [AB] } = \begin{bmatrix} \mathbf{AB} \end{bmatrix} = \begin{bmatrix} \mathbf{AB} \end{bmatrix} \\ \mathbf{0} \text{ and } \begin{bmatrix} \mathbf{B} \end{bmatrix} = \begin{bmatrix} \mathbf{B} \end{bmatrix}\_{\text{max}} \quad \text{(5)}$$

Y recorded signal

Ymax maximum of the recorded signal

[AB] concentration of the primary complex

[AB′ ] concentration of the secondary complex

[B] concentration of the unbound target

[A] concentration of the ligand

ka1 bimolecular association rate constant for primary complex formation

k h a2 monomolecular rate constant of the interconversion [AB] → AB′ i

(secondary complex formation)

kd1 monomolecular dissociation rate constant

kd2 monomolecular rate constant of the interconversion - AB′ → [AB]

The equation is solved by numerical integration by the TraceDrawer software. Since the formation of the complex AB′ from AB does not cause any signal change, it is difficult to estimate the microscopic kinetic constants. The changes in the signal amplitude can be investigated according to the following equation to obtain the bimolecular association and the monomolecular dissociation rate constants.

$$\ln\left(\frac{\mathbf{Y}\_0}{\mathbf{Y}\_t}\right) = k\_\mathbf{d} \cdot (\mathbf{t} - \mathbf{t}\_o) \tag{8}$$

$$Y\_t = Y\_0 \cdot \exp\left[-k\_d \cdot (\mathbf{t} - \mathbf{t}\_0)\right] \tag{9}$$

Y<sup>0</sup> recorded signal at time t<sup>0</sup>

Y<sup>t</sup> recorded signal at time t

The K<sup>D</sup> value can be calculated according to

$$\mathcal{K}\_{\rm D} = \frac{k\_{\rm d}}{k\_{\rm a}} \tag{10}$$

#### Fluorescence Measurements

Fluorescence measurements were carried out with a Jasco Spectrofluorometer FP-8200, equipped with a Jasco MCB-100 Mini Circulation Bath (20◦C, xenon lamp, data interval 0.5 nm, accumulation number 3). The excitation wavelength was 280 nm, fluorescence emission was recorded at 350 nm to exclude self-fluorescence of non-bound peptide. For the measurements, GCAP-2 was mixed with increasing concentrations of ROS-GC peptides in 20 mM HEPES (pH 7.5). As a negative control, separate recordings were performed with N-acetyl-L-tryptophanamide or N-acetyl-L-tyrosinamide, respectively. The composition of the controls correlates with the peptides' composition of tryptophan and tyrosine residues to exclude that a change in fluorescence intensity results from self-fluorescence of the peptides.

#### RESULTS

To investigate the binding site of GCAP-2 at the intracellular domains of ROS-GC1, cross-linking reactions of myristoylated GCAP-2 and three ROS-GC1 peptides (**Figure 1**) were performed. Peptide 1 had been identified in previous crosslinking studies as a GCAP-2 binding segment, which formed the basis for deriving a model of the GCAP-2/peptide complex in the presence of calcium (Pettelkau et al., 2012). This peptide had been earlier suggested as a core binding site for GCAP-2 (Duda et al., 2005). In this work, we also studied an N-terminally extended version of peptide 1 (peptide 2) comprising a part of the catalytic domain in order to gain more detailed insights into the GCAP-2 binding site at ROS-GC1. In addition, peptide 3, representing a potential interaction site of GCAP-1 (Lange et al., 1999), was investigated to identify possible overlapping or additional binding sites of GCAP-1 and-2 in the juxtamembrane domain (Peshenko et al., 2015b).

#### Chemical Cross-Linking Strategies

Cross-linkers possessing diverse reactivities and spanning varying distances were used to covalently fix a

$$\begin{aligned} \frac{dY}{dt} &= \begin{array}{c} k\_{\text{il}} \cdot [\text{A}] \cdot \text{Y}\_{\text{max}} - \left(k\_{\text{il}} \cdot [\text{A}] + k\_{\text{il}}\right) \cdot Y = k\_{\text{il}} \cdot [\text{A}] \cdot (\text{Y}\_{\text{max}} - \text{Y}) - k\_{\text{il}} \cdot Y\\ k\_{\text{s}} &= \, k\_{\text{al}} \cdot [\text{A}] + k\_{\text{il}} \end{array} \tag{7}$$

The plot of dY/dt against Y will yield the term k<sup>s</sup> as slope. The plot of k<sup>s</sup> against [A] will yield a straight line with ka1 as slope and kd1 as intercept. Furthermore, the value kd1 can be determined from the dissociation curve according to the following equation:

GCAP-2/ROS-GC1 peptide complex in the presence (+Ca2+) and absence of calcium (−Ca2+). The DSBU cross-linker targets mainly primary amine groups, while CDI reacts with both amine and hydroxy groups. DAU reacts with sulfhydryl groups and incorporated photo-Met interacts with any amino acid, with a preference for the acidic amino acids, glutamic and aspartic acid (Iacobucci et al., 2018a). For DSBU cross-linking, the different cross-linked species were analyzed by SDS-PAGE (**Supplementary Figure 1**). In the presence of calcium, no clear signal was detected for the GCAP-2 control sample in the absence of cross-linker and peptide, which is caused by the presence of different calcium-loaded states. Consequently, an SDS-PAGE separation of GCAP/ peptide (1:1) complex and GCAP monomer bands was not completely possible, and for the in-gel digestion both signals (Ia/Ib) had to be excised en bloc. For CDI, DAU, and photo-Met, in-solution digestion was performed. It should be noted in this context that all four cross-linking reagents are MS/MS-cleavable and generate characteristic reporter ions to facilitate the identification of cross-linking reaction products. As such, the fragmentation patterns of the cross-linkers used herein prevent that isobaric species—originating from partially hydrolyzed cross-linker with consecutive peptide sequences—can be mistaken for "true" cross-links (Iacobucci and Sinz, 2017). For our studies, the automated cross-link identification performed by the MeroX software proved highly beneficial for a correct assignment of cross-linked products (Götze et al., 2015; Hage et al., 2017; Iacobucci et al., 2018a,b).

#### Cross-Linked Products Between GCAP-2 and ROS-GC1 Peptides

All unique cross-links identified between the GC-peptides and myristoylated GCAP-2 in its calcium-loaded, non-activating (+Ca2+, **Figure 2**, Upper) and calcium-free, activating state (–Ca2+, **Figure 2**, Lower) are visualized as circular plots to contrast calcium-dependent differences. The unique crosslinking sites are summarized in **Table 1**, all cross-links identified are listed in the **Supplementary Tables 1**–**4**. For peptide 1, the cross-linking sites between the N-terminus of the peptide and GCAP-2 are, with one exception, identical in the presence and absence of calcium. A single additional cross-link with photo-Met was identified in the absence of calcium. Peptide 2 has two main reactions sites, the first site comprising residues 1-9, the second one comprising residues 18-24. The latter site overlaps with the N-terminus of peptide 1 to confirm the crosslinking site of peptide 1. A DSBU cross-link between GCAP-2 and the N-terminus of peptide 2 and a CDI cross-link with

Rehkamp et al. ROS-GC1/GCAP-2 Interaction

TABLE 1 | Unique cross-linking sites between GCAP-2 and ROS-GC peptides 1-3 in the presence (1 mM CaCl2) and absence (10 mM EGTA) of calcium using DSBU, CDI, DAU, and photo-Met (PM) as cross-linkers.


(Continued)


S19/T20 are shown as representative examples of both reactions sites in **Figures 3A,B**. For peptide 2, the number of crosslinks was similar in the non-activating (+Ca2+) and activating state (–Ca2+) of GCAP-2 (**Table 1**). DSBU cross-links (**Figure 2**, colored in blue) that can bridge up to 30 Å and thereby yield longer distances than CDI were identified at a higher frequency in the calcium-free than in the calcium-loaded state. These observations are in agreement with the previously published data using the BS2G cross-linker (Pettelkau et al., 2012). BS2G has comparable distance properties and can capture flexible structures, as does DSBU.

In contrast, CDI (**Figure 2**, colored in red), an ultra-short cross-linker bridging distances up to maximally ∼16 Å, was employed to obtain complementary structural information. Cross-links between lysines in the C-terminal part of GCAP-2 (K129-K200) and peptide 2 were mainly identified in the calcium-loaded state of GCAP-2. Calcium-independent CDI cross-links with peptide 2 were observed for all lysine residues in the amino acid sequence stretch K96-K126 of GCAP-2. We speculate that residues K96-K126 in GCAP-2 represent a major interaction site with ROS-GC. This is underlined by the fact that the cross-links with photo-Met and DAU point to the same region (**Figure 5**). The DAU cross-link (C131 in GCAP-2 – C5 in peptide 2) was only found in the presence of calcium.

The number of cross-links with peptide 3 was comparable in the presence and absence of calcium (**Figure 2**). Strikingly, DSBU and CDI reacted exclusively with the N-terminus and Y2 of peptide 3 although it is the only peptide of this study containing two lysine residues that should be preferentially targeted by amine-reactive cross-linkers. No photo-Met cross-links were observed between GCAP-2 and peptide 3.

#### Affinities of GCAP-2 and ROS-GC1 Peptides

We performed surface plasmon resonance (SPR) measurements with GCAP-2 and the three ROS-GC1 peptides to determine K<sup>D</sup> values of the protein/peptide interactions. GCAP-2 was immobilized on 2D-CMD chips, and peptides were injected at different concentrations. The kinetic evaluation was conducted according to a 1:1 binding model including two binding states and a conformational change upon binding (**Scheme 1**).

For peptides 1 and 2, K<sup>D</sup> values of 2.36µM and 1.99µM, respectively, were determined. These values are derived from two separate SPR measurements (**Figure 4**, **Supplementary Figure 3**). For peptide 3, the SPR curves could not be fitted suggesting that the interaction between GCAP-2 and peptide 3 is too weak to be detected with SPR.

Additionally, we conducted fluorescence spectroscopy by titrating peptides 1-3 to GCAP-2. A wavelength of 350 nm was chosen to rule out that changes in fluorescence originate from an increased concentration of non-bound peptide. Only peptides 1 and 2 delivered changes in fluorescence, while peptide 3 did not exhibit any changes compared to the negative control samples, N-acetyl-L-tryptophanamide or N-acetyl-L-tyrosinamide. We chose these compounds as controls as they resemble the aromatic

FIGURE 4 | One set of SPR measurements between GCAP-2 and ROS-GC (A) peptide 1 and (B) peptide 2. The other SPR dataset is shown in (Supplementary Figure 3). The following peptide concentrations were applied: 0.5µM (dotted line), 1µM (dashed line), 5µM (solid line), 10µM (long-dashed line), 20µM (long-dotted line), and 40µM (dashed/dotted line). Curve fittings are shown as thin, solid lines. The KD value of the interaction of GCAP-2 and peptide 1 is 1.12 ± 0.74µM, while that of peptide 2 is 3.00 ± 0.59µM.

composition of the ROS-GC1 peptide and as such, possess similar spectroscopic properties.

According to the fluorescence measurements, GCAP-2 binding only takes place for peptides 1 and 2. The K<sup>D</sup> values between GCAP-2 and ROS-GC peptides 1 and 2 derived from fluorescence titration were in the same range as the values determined by SPR. For peptide 1, a KD1 value of 4.2µM was observed (**Supplementary Figure 2**). The data were investigated by fitting according to an optimum behavior. Due to the susceptibility of the fitting routine in the first part, the value KD1 of peptide 2 was set to 1µM by extrapolation of the starting point.

During titration of peptide 1 to GCAP-2, an initial increase and subsequent decrease in fluorescence were recorded, describing an optimal behavior. Both signal changes are related to peptide binding at different GCAP-2 sites, orginating from dequenching by the first binding event and quenching by a second event. The K<sup>D</sup> values of the first and second binding events differ by two orders of magnitude with KD2 values of 244µM (peptide 1) and 326µM (peptide 2).

Finally, it has to be noted that for SPR and fluorescence measurements, calcium concentrations were not controlled by a chelator system. The calcium concentrations of the buffer solutions used for both measurements were determined by ICP-MS to be ∼2µM.

#### DISCUSSION

The calcium-dependent molecular regulation process of ROS-GC1 has so far remained elusive. In particular, the binding regions between ROS-GC1 and GCAP-2 are under debate: A mutational study predicted that ROC-GC1 activation by GCAP-2 requires the N-terminal part of the intracellular domain (residues M443-S746) of ROS-GC1 (Laura and Hurley, 1998). Binding of GCAP-2 upstream the catalytic domain was confirmed by in vivo studies (Peshenko et al., 2015b). These investigations involved ROS-GC1 mutants consisting solely of this region or with a deletion of a potential GCAP-2 binding motif in the C-terminal extension of the catalytic domain. This alternative binding region for GCAP-2 in the C-terminal extension of ROS-GC1 has been proposed upon activity and SPR measurements of further ROS-GC mutants (Duda et al., 2005). In addition, Sharma and co-workers observed that GCAP-2 could activate a ROS-GC mutant lacking the juxtamembrane and kinase homology domains (Duda et al., 2012).

In this work, the calcium-dependent binding of ROS-GC1 peptides to GCAP-2 was investigated by cross-linking/MS and biophysical methods. To resolve the debate of whether GCAP-2 binds to the membrane close or at the C-terminal part of ROS-GC1, three peptides were employed (**Figure 1**): Peptides 1 and 2 represent the C-terminal region beyond residue 942, while peptide 3 comprises residues 503-522 of the juxtamembrane part. Peptide 2 is an N-terminally extended version of peptide 1 that is thought to comprise two α-helices and two β-strands, according to homology models with a catalytic domain of rat type II adenylyl cyclase (PDB 1AWL) and green algae

guanylyl cyclase (PDB 3ET6) (Liu et al., 1997; Winger et al., 2008). Peptide 2 comprises conserved motifs of the catalytic domain of mammalian guanylyl cyclases (Tucker et al., 1998; Winger et al., 2008; Ravichandran et al., 2017) and C946, N953, and R957 may be involved in GTP recognition and binding.

# Affinity Between GCAP-2/ROS-GC1 Peptides

In fact, affinities with K<sup>D</sup> values in the low micromolar range were determined between GCAP-2 and peptides 1 and 2 via SPR and fluorescence measurements (**Figure 4**, **Supplementary Figures 2**, **3**). In agreement with these findings, previous binding studies employing recombinant ROS-GC1 fragments spanning residues 733-1054 and 965-1054 had suggested K<sup>D</sup> values of approximately 2µM (Duda et al., 2005). For both SPR and fluorescence studies, 1:1 binding between GCAP-2 and ROS-GC1 peptides was assumed (**Scheme 1**). A 1: 1 binding model is consistent with MS analyses of GCAP-2 and peptide 1 (Pettelkau et al., 2012). Interestingly, under the settings used in the fluorescence measurements, second binding sites were recorded between GCAP-2 and each of the two ROS-GC1 peptides 1 and 2 with high K<sup>D</sup> values (peptide 1: 244µM and peptide 2: 326µM), representing weak secondary interaction sites. We cannot completely rule out at the moment that these secondary binding events originate from possible ternary complexes between ROS-GC1 peptides and GCAP-2.

In comparison, K<sup>D</sup> values determined for GCAP-2 and the human, full-length guanylyl cyclase (418 ± 135 nM) via backscattering interferometry, are consistent with the results shown herein (Sulmann et al., 2017). However, no interaction between GCAP-2 and the isolated domain of human ROS-GC-1 residues 496-806 (comprising the juxtamembrane and kinase homology domains) could be demonstrated in those experiments. This is consistent with our observation that no interaction was detected between peptide 3 (aa 503-522) and GCAP-2, neither in SPR nor in fluorescence experiments. Taken together, the affinities determined between GCAP-2 and peptides 1 and 2 are in a moderate range. Currently, additional binding surfaces that may arise from the flexible 3Dstructure of the intracellular domain of ROS-GC1 cannot be excluded.

#### Structural Details of GCAP-2/ROS-GC1 Peptide Interaction

Chemical cross-linking in combination with MS represents a powerful technique to define interaction sites in protein complexes. In this work, ROS-GC1 peptides 1-3 were crosslinked with GCAP-2 via DSBU, CDI, DAU and photo-Met, reagents which all possess differing reactivities and bridge various distances (from ∼8 to 30 Å). The reactivity of specific amino acids is however not only determined by the topology of the protein/peptide complex, but also by local pK<sup>a</sup> values. In our case, the majority of cross-links were identified between GCAP-2 and peptide 2. Especially for peptide 2, distinct regions were targeted by the different cross-linkers. DSBU cross-links between GCAP-2 and peptide 2 were more frequently observed in the calcium-free than in the calcium-loaded state (**Figure 2**). As GCAP-2 is more flexible in the calcium-free state, DSBU with a spacer length of 12.5 Å may be able to capture flexible segments, in particular the C-terminal part of GCAP-2. In the activating, calcium-free state, CDI mainly connected peptide 2 with the N-terminal part of GCAP-2 (G2-K126). Probably, the CDI cross-linker is too short for interactions with the C-terminal, flexible region of GCAP-2. CDI exhibited more reactions in the non-activating, tighter, calcium-loaded conformation. However, independent of the calcium-bound state, CDI reacted with four lysines (K96, K98, K102, and K106) in the first α-helix of EF-hand motif 4 (Ames et al., 1999; **Figure 5**). A single cross-link (C131) with DAU was identified only in the presence of calcium. Residue C131 is located at the beginning of the loop between EF-motifs 3 and 4 and due to a calcium-induced conformational change in this region, cross-linking with DAU seems to be hindered in the absence of calcium. In **Figure 5**, all residues cross-linked with peptide 2 are highlighted, revealing that several cross-linked amino acids (K46/50, K96/98/102/106, Y125/K126/129, photo-Met181/186, C131) flank a cleft in GCAP-2, which might represent a preferred interaction site for ROS-GC1.

One additional point to be mentioned is the involvement of the peptides' N-termini in all cross-linking reactions (**Figure 2**). This is a drawback of performing cross-linking experiments with short peptides as their N-termini do not reflect the true situation in the protein by creating a novel, artificial reaction site for the cross-linker. For peptide 3, only the N-terminus was found to be cross-linked to GCAP-2 with DSBU and CDI. Strikingly, both lysine residues in peptide 3, which should preferentially be targeted by amine-reactive cross-linkers, were not crosslinked to GCAP-2 at all. This finding is consistent with SPR and fluorescence measurements and implies a very weak or even missing interaction between GCAP-2 and the ROS-GC region comprising peptide 3.

# CONCLUSION

In this work, the interaction between GCAP-2 and three ROS-GC1 peptides (**Figure 1**) was investigated by crosslinking/MS and biophysical analyses. The majority of cross-links were obtained with ROS-GC1 peptide 2 (aa 942-981), in which two segments (residues 1-9, 18-24) reacted preferably. Peptide 2 exhibits the highest affinity for GCAP-2 in the low micromolar range. This result indicates that one GCAP-2 interaction site in ROS-GC1 is located in the region comprising parts of the catalytic domain and the C-terminal extension. A few crosslinks were identified with peptide 3 (aa 503-522), derived from the juxtamembrane area of ROS-GC1, but no interaction was detected between GCAP-2 and peptide 3 in SPR and fluorescence measurements. Considering conformational transitions of the GCAP-2/ROS-GC1 complex, additional binding regions cannot be ruled out at present, but this issue will be addressed in further investigations using domains as well as full-length ROS-GC1.

#### AUTHOR CONTRIBUTIONS

AR, DT, RG, CI, and CHI performed the experiments. AR and AS planned the experiments. AR and AS wrote the manuscript.

# ACKNOWLEDGMENTS

AS acknowledges funding from the DFG (project Si 867/15- 2) and the region of Sachsen-Anhalt. The authors are indebted to Elisabeth Schwarz for her contribution in writing this manuscript and to Martin Herzberg for performing ICP-MS measurements. Michael Götze is acknowledged for continuous improvements of the StavroX and MeroX software, Stephan Theisgen is acknowledged for providing the GCAP-2 plasmid.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00330/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 © 2018 Rehkamp, Tänzler, Iacobucci, Golbik, Ihling and Sinz. 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) and the copyright owner(s) 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.

# Photoreceptor Guanylate Cyclase (GUCY2D) Mutations Cause Retinal Dystrophies by Severe Malfunction of Ca2+-Dependent Cyclic GMP Synthesis

Hanna Wimberg<sup>1</sup> , Dorit Lev2,3, Keren Yosovich2,3, Prasanthi Namburi<sup>4</sup> , Eyal Banin<sup>4</sup> , Dror Sharon<sup>4</sup> and Karl-Wilhelm Koch<sup>1</sup> \*

<sup>1</sup> Department of Neuroscience, Biochemistry Group, University of Oldenburg, Oldenburg, Germany, <sup>2</sup> The Rina Mor Institute of Medical Genetics, Wolfson Medical Center, Holon, Israel, <sup>3</sup> Sackler School of Medicine, Tel Aviv University, Tel Aviv, Israel, <sup>4</sup> Department of Ophthalmology, Hadassah-Hebrew University Medical Center, Jerusalem, Israel

#### Edited by:

Vsevolod V. Gurevich, Vanderbilt University, United States

#### Reviewed by:

Maxim Sokolov, West Virginia University, United States Alecia K. Gross, The University of Alabama at Birmingham, United States Marie E. Burns, University of California, Davis, United States Ching-Kang Jason Chen, Baylor College of Medicine, United States

\*Correspondence:

Karl-Wilhelm Koch karl.w.koch@uni-oldenburg.de

Received: 29 June 2018 Accepted: 06 September 2018 Published: 25 September 2018

#### Citation:

Wimberg H, Lev D, Yosovich K, Namburi P, Banin E, Sharon D and Koch K-W (2018) Photoreceptor Guanylate Cyclase (GUCY2D) Mutations Cause Retinal Dystrophies by Severe Malfunction of Ca2+-Dependent Cyclic GMP Synthesis. Front. Mol. Neurosci. 11:348. doi: 10.3389/fnmol.2018.00348 Over 100 mutations in GUCY2D that encodes the photoreceptor guanylate cyclase GC-E are known to cause two major diseases: autosomal recessive Leber congenital amaurosis (arLCA) or autosomal dominant cone-rod dystrophy (adCRD) with a poorly understood mechanism at the molecular level in most cases. Only few mutations were further characterized for their enzymatic and molecular properties. GC-E activity is under control of neuronal Ca2+-sensor proteins, which is often a possible route to dysfunction. We investigated five recently-identified GC-E mutants that have been reported in patients suffering from arLCA (one large family) and adCRD/maculopathy (four families). Microsatellite analysis revealed that one of the mutations, c.2538G > C (p.K846N), occurred de novo. To better understand the mechanism by which mutations that are located in different GC-E domains develop different phenotypes, we investigated the molecular consequences of these mutations by expressing wildtype and mutant GC-E variants in HEK293 cells. Analyzing their general enzymatic behavior, their regulation by Ca2<sup>+</sup> sensor proteins and retinal degeneration protein 3 (RD3) dimerization domain mutants (p.E841K and p.K846N) showed a shift in Ca2+-sensitive regulation by guanylate cyclase-activating proteins (GCAPs). Mutations in the cyclase catalytic domain led to a loss of enzyme function in the mutant p.P873R, but not in p.V902L. Instead, the p.V902L mutation increased the guanylate cyclase activity more than 20 fold showing a high GCAP independent activity and leading to a constitutively active mutant. This is the first mutation to be described affecting the GC-E catalytic core in a complete opposite way.

Keywords: GUCY2D mutation, Leber congenital amaurosis, cone-rod dystrophy, guanylate cyclase, RD3 protein, GCAP

# INTRODUCTION

Signal transduction in vertebrate rod and cone photoreceptor cells is characterized by an interplay between the two second messengers Ca2<sup>+</sup> and cGMP. Cyclic nucleotide-gated (CNG)-channels in the cell membranes of rod and cones are kept open by cGMP and close, when cGMP is hydrolyzed upon illumination leading to the hyperpolarization of the cell. A second consequence of

illumination is the decrease of the cytoplasmic Ca2<sup>+</sup> level providing negative feedback regulation (Arshavsky and Burns, 2012; Koch and Dell'Orco, 2015). The photoreceptor guanylate cyclase GC-E (alternatively dubbed retGC1 or ROS-GC1) represents a key enzyme in phototransduction, important for the restoration of cytoplasmic cGMP and return to the dark state of the cell. Synthesis of cGMP by GC-E is regulated by guanylate cyclase-activating proteins (GCAPs), which are activated by decreasing Ca2<sup>+</sup> concentrations in the cell (Palczewski et al., 2004; Dizhoor et al., 2010; Koch and Dell'Orco, 2013).

Mutations in the GUCY2D gene coding for GC-E lead to severe retinal diseases in humans and mainly autosomal dominant cone-rod dystrophy (adCRD) or autosomal recessive Leber congenital amaurosis type 1 (arLCA1; Duda and Koch, 2002). For adCRD, GUCY2D mutations are the major cause (Sharon et al., 2018). In CRD, degeneration starts in the cones and leads to loss of the central visual field due to the high presence of cones in the macula of a non-affected retina. CRD can lead to complete blindness, when degeneration of rods follows those of cones (Hamel, 2007; Berger et al., 2010). The LCA1 phenotype appears even more severe, with photoreceptor function loss and blindness emerging very early in life (den Hollander et al., 2008; Boye, 2014a,b). Another gene that is involved in the pathogenesis of LCA (type 12) is rd3 coding for the retinal degeneration 3 (RD3) protein, which is an effective inhibitor of GCAP-mediated activation of GC-E and is involved in trafficking of GC-E from the inner to the outer segment in photoreceptors (Lavorgna et al., 2003; Friedman et al., 2006; Azadi et al., 2010; Peshenko et al., 2011).

While more than a hundred mutations in the GUCY2D gene were described, a link to functional consequences in the enzyme was set just for a small number, compared to the large number of known mutations. Most previous functional studies focused on mutations in the dimerization domain (DD) of the GC-E, which harbors a so-called "mutation hot spot region" (Wilkie et al., 2000; Kitiratschky et al., 2008; Zägel et al., 2013; Dizhoor et al., 2016).

In this work, we attempt to biochemically characterize some recently identified mutations and relate the phenotype to functional impairments of the enzyme. While two mutations are positioned in the DD in close vicinity to the hot spot region (p.E841K and pK846N; Lazar et al., 2014), three other mutations are located in other GC-E domains. For example, the mutation p.A710V leading to arLCA (Gradstein et al., 2016) is located in the kinase homology domain of the enzyme and two further mutations in the catalytic domain of GC-E (p.P873R) cause either adCRD or are found in a heterozygous state in an isolated case with CRD (p.V902L; both are not published so far).

Our functional analysis using recombinant proteins in heterologous expression systems showed different effects on enzyme activity due to localization in the various regions of the GC-E. Mutations in the DD are known to cause CRD and often lead to a change in Ca2+-sensitive regulation of the protein, which we also observed for the mutants E841K and K846N. Thus, both GC-E mutant forms needed higher Ca2<sup>+</sup> concentrations to shut off enzyme activity. In contrast, the A710V and P873R mutations showed no enzyme activity at all (basal or GCAP-activated). However, a strong increase in enzyme activity was found for the V902L mutant by directly affecting the catalytic mechanism of the enzyme. This was rather unexpected, because other described mutations in the GC-E catalytic domain drastically decrease GC-E activity causing a LCA1 phenotype.

These results provide a route for better understanding the negative effects of GUCY2D mutations in photoreceptor cell physiology. Differences in biochemical key properties of GC-E mutants might help us to understand why some GC-E mutations lead to a LCA phenotype while others result in CRD.

#### MATERIALS AND METHODS

#### Clinical Analysis, Mutation Detection, Cloning of GC-E Mutants With Site-Directed Mutagenesis

The study protocols adhered to the tenets of the Declaration of Helsinki and received approval from the local Ethics Committee of Hadassah Medical Center. Prior to donation of a blood sample, a written informed consent was obtained from all individuals who participated in this study, after explanation of the nature and possible consequences of the study. Ocular evaluation included a comprehensive ophthalmologic exam, Goldmann perimetry, electroretinography (ffERG), electro-oculography (EOG), color vision testing, color and infrared fundus photos, optical coherence tomography (OCT), and fundus autofluorescence (FAF) imaging were performed.

Sanger sequencing of PCR products was used to screen all exons of GUCY2D for mutations. The primers are listed in **Supplementary Table S1**. In addition, we used four sets of microsatellite markers flanking the GUCY2D gene (**Supplementary Table S1**). For each set, the forward primer was labeled with a FAM fluorescence dye.

To create the five desired GC-E mutants, the wildtype (WT) sequence was cloned into a pIRES2-eGFP vector and used as a template (Zägel et al., 2013). The Q5 <sup>R</sup> Site-Directed Mutagenesis Kit (New England Biolabs, Ipswich, MA, United States) was used to introduce point mutations in the GC-E sequence. Instructions according to the manufacturer's protocol were followed. The primers used to produce the mutants are listed in **Supplementary Table S1**.

The obtained clones were verified by full-length sequencing of the GC-E coding region.

#### Stable Transfection and Expression of GC-E Mutants in HEK-293T Cells

HEK 293T cells were used for the expression of GC-E WT protein and the five mutants. For each clone, a stable cell line was created. Cells were transfected with PolyFect (Qiagen, Hilden, Germany) and stable clones were selected via G418 antibiotic resistance. Positive clones were recognized by GFP fluorescence and were validated by western blotting with a GC-E specific antibody (following the protocol as described recently; Zägel et al., 2013). Confluent HEK cells were harvested. The cells of one

10 cm plate were transferred into a 15 ml tube and centrifuged for 5 min at 1000 × g. Cell pellets were washed with PBS, transferred into a 1.5 ml tube, and centrifuged again for 5 min at 12,000 × g. The pellets were frozen at −80◦C until further use. Determination of protein concentration in the presence of lipids was performed according to a standard Amido Black assay (Kaplan and Pedersen, 1985).

#### Expression and Purification of GCAP1, GCAP2, and RD3 in E. coli

Bovine myristoylated GCAP1 and GCAP2 were expressed in Escherichia coli and purified via size-exclusion and anionexchange chromatography. The detailed procedure was described earlier (Hwang et al., 2003; Koch and Helten, 2008). Human rd3 was cloned into a petM11 vector, creating a His6-tagged construct. RD3 was expressed and purified from E. coli. Niaffinity chromatography was used for purification. The protein was stored in 10% glycerol at –80◦C. The detailed purification protocol was described recently (Wimberg et al., 2018).

### GC-E Activity Assays

To analyze the effect of point mutations on GC-E function, the enzymatic activity of GC-E mutants was measured in comparison to the WT. HEK cell pellets were resuspended in 500 µl of 10 mM Hepes/KOH pH 7.4, 1 mM DTT, and protease inhibitor cocktail. The suspension was incubated for 30 min on ice. Cell lysis was performed using a syringe with a 0.6 mm tip. After centrifugation (5 min, 13,000 × g), the cell pellet was resuspended in 100 µl of 50 mM Hepes/KOH pH 7.4, 50 mM KCl, 20 mM NaCl, and 1 mM DTT. For each sample, 10 µl of these membrane suspensions were used. They were mixed with 20 µl of a GCAP1 or GCAP2 solution (5 µM) that was previously adjusted to different free Ca2<sup>+</sup> concentrations using a Ca2+/EGTA buffer system exactly as described before (Hwang et al., 2003; Koch and Helten, 2008; Zägel et al., 2013). Samples were pre-incubated for 5 min at room temperature. Reaction was started by adding 20 µl of 2.5× GC-buffer (75 mM Mops/KOH pH 7.2, 150 mM KCl, 10 mM NaCl, 2.5 mM DTT, 8.75 mM MgCl2, 2.5 mM GTP, 0.75 mM, and 0.4 mM Zaprinast). The reaction mixtures were incubated for 5 min at 30◦C. Reaction was stopped by adding 50 µl 0.1 M EDTA and 5 min of incubation at 95◦C. Samples were centrifuged for 10 min at 13,000 × g. Supernatants were analyzed for the amount of produced cGMP by RP-HPLC using a LiChrospher <sup>R</sup> 100 RP-18 (5 µm) column (Merck, Darmstadt, Germany) exactly as described (Hwang et al., 2003; Koch and Helten, 2008; Zägel et al., 2013). Inhibition of GCAP-mediated activation of GC-E variants by RD3 was tested by adding increasing RD3 concentrations (0–500 nM) to the reaction mixture. Further, we tested whether the V902L mutant shows any GCAP-dependent change in activity by varying GCAP concentrations in the range from 0.25 to 10 µM (free Ca2<sup>+</sup> buffered to 1.7 nM).

# GC-E Localization in HEK-293T Cells

HEK cells were grown on coverslips in a 24-well plate. Transfection was performed using polyethylenimine (PEI) at 80% cell confluence; 0.5 µg DNA were mixed with 2 µg PEI in DMEM without supplements and incubated for 10 min at 20◦C. Subsequently, samples were added to the cells and incubated for 48 h at 37◦C, 5% CO2. Cells were washed with PBS, fixed with 4% paraformaldehyde (PFA) in PBS for 10 min and again washed three times with PBS. Cells were incubated with 5% NGS (normal goat serum) in PBS pH 7.4 with 0.1% Triton X-100 for 1 h at room temperature. Primary antibodies were added and incubated for 24 h at 4◦C. GC-E was detected by an anti-GC-E antibody (1:100, rabbit polyclonal H-225 named anti-ROS-GC1, Santa Cruz Biotechnology, Dallas, TX, United States). For staining of the endoplasmic reticulum (ER), an anti-Na+/K+-ATPase (1:200, mouse monoclonal H-3, Santa Cruz Biotechnology, Dallas, TX, United States) antibody was used. Cells were washed three times with PBS and incubated with secondary antibodies in PBS pH 7.4 with 0.1% Triton-X100 for 2 h at room temperature [donkey anti rabbit conjugated to Fura350, 1:200, Thermo Fisher Scientific (Invitrogen), Waltham, MA, United States; goat antimouse conjugated to Dylight594, 1:500, Thermo Fisher Scientific, Waltham, MA, United States]. Again, cells were washed with PBS and sealed with Fluoromount-G (Southern Biotech, Birmingham, AL, United States). The staining was analyzed using a Zeiss Axiophot fluorescence microscope.

# RESULTS

# Genetic Screening and Clinical Assessment

Family MOL0064 includes seven individuals affected with adCRD (**Figure 1A**), four of whom participated in the study and suffered from early-onset retinal degeneration (**Table 1**). ERG performed in four affected individuals showed extinguished or severely reduced cone responses (**Table 1**) and extinguished rod responses. All patients suffered from nystagmus and low visual acuity. Sanger sequencing of genes involved in inherited retinal diseases including the GUCY2D region encoding GC-E DD, revealed a novel heterozygous missense variant (c.2618C > G, p.Pro873Arg; **Figure 1B**) in the four affected individuals who participated in the study. This variant is absent from databases (gnomAD, ExAC), is predicted to be pathogenic according to a variety of prediction tools (MutationTaster<sup>1</sup> ; PolyPhen-2<sup>2</sup> ; and SIFT<sup>3</sup> ), and is highly conserved (**Figure 1C**). The mutation was validated in our clinical lab and it was not found in the healthy brother.

MOL0308 (**Figure 1A**) includes an affected child with earlyonset CRD. Following negative analysis for the known CRD mutations in the relevant population, we performed whole exome sequencing on his DNA sample. The analysis revealed a novel heterozygous missense variant in GUCY2D: c.2704G > T; p.V902L (**Figure 1B**). This variant is absent from databases (gnomAD, ExAC), is predicted to be pathogenic according to a variety of prediction tools, and is highly conserved (**Figure 1C**).

<sup>1</sup>http://www.mutationtaster.org

<sup>2</sup>http://genetics.bwh.harvard.edu/pph2/

<sup>3</sup>http://sift.jcvi.org/seq\_submit.php

The mutation was validated in our clinical lab and it was not found in the healthy mother. Unfortunately, the father's DNA

(C) Sequence alignment of GUCY2D proteins in the variants regions.

sample was not available. MOL0508 includes two affected individuals, an index case and her mother, with macular degeneration and cone dystrophy. We have previously reported (Lazar et al., 2014) that we identified a heterozygous variant (c.2521G > A, p.E481K) in GUCY2D.

MOL0430 includes an affected male with CRD who was found to be heterozygous for the c.2538G > C (p.K846N) variant in GUCY2D as we previously reported (Lazar et al., 2014). For the current study, we were able to recruit additional family members, including both parents and three siblings, all are unaffected and none carries the variant in GUCY2D. Haplotype analysis using four microsatellite markers flanking GUCY2D revealed that the index case (individual II:1 in **Supplementary Figure S1**) was the only sibling to inherit the paternal haplotype 141-205-251-272 while his three siblings inherited the counter allele. However, the index case shares the maternal haplotype 161-199-255-277 with two of his unaffected siblings (II:3 and II:4), indicating a paternal de novo mutation.

#### Cloning and Stable Expression of GC-E Mutants in HEK-293T Cells

Point mutants of GC-E WT were successfully created by site-directed mutagenesis as proven by full-length sequencing. Immunohistochemistry and immunoblot analyses confirmed expression of WT and mutant GC-E in stable cell lines as shown in **Supplementary Figure S2**. Protein expression levels of GC-E variants were similar, when the same amounts of total protein

(10 µg of cell homogenates) were loaded on a gel. Samples shown in **Supplementary Figure S2** were used for the experiment displayed in **Figure 2**, which compares activity levels of WT and mutant GC-E forms.

Cellular localization in HEK cells was the same for GC-E mutants and GC-E WT and was visible in the ER in agreement with previous observations (Peshenko et al., 2008; Zägel and Koch, 2014). Cellular localization was analyzed by immunostaining using an anti-Na+/K+-ATPase antibody as ER marker (red, **Supplementary Figure S3B**). Localization of GC-E was detected with a specific antibody (blue, **Supplementary Figure S3C**). GFP signal (green, **Supplementary Figure S3A**) indicated successfully transfected cells. Localization of GC-E in the ER became visible in the overlay (magenta, **Supplementary Figure S3D**).

#### Activity of GC-E Wildtype and Mutants

In order to gain insight into the structure-function relationship of retinal disease-causing mutations in GC-E, we investigated three critical parameters of the enzymatic activity profile of GC-E: (1) increase of GC-E activity in the presence of Ca2+-free/Mg2+ bound GCAP1 and GCAP2; (2) inhibitory effect of RD3, when GC-E is in the activated state in complex with GCAP1 or GCAP2; and (3) Ca2+-sensitive activation profile of GC-E.

All mutants showed a severe impairment of normal GC-E activity (**Figure 2**), but type and impact of the disturbance differed among all mutants. GC-E A710V and P873R had no measurable activity. The expression of A710V in these samples was much lower than that of WT GC-E and the other mutants. To exclude that the lack of measurable activity of A710V is due to low expression levels, we created a new stable cell line for A710V that showed higher expression levels (**Supplementary Figure S2B**). But even with the highest expression level of A710V, we did not detect any activity of this mutant. The two mutations in the DD near the hot spot region exhibited drastically decreased activity in the presence and absence of GCAP1 and GCAP2, but still were able to switch from a basal enzymatic state to the GCAP-mediated activation state. Most surprisingly, the V902L mutation resulted in high basal activity that did not increase in the presence of GCAP2 and increased only to a small extent in the presence of GCAP1 (**Figure 2**). Thus, the V to L exchange in position 902 in the cyclase catalytic domain transformed the enzyme into a constitutively active conformation.

#### Inhibitory Effect of RD3

Retinal degeneration protein 3 is a strong inhibitor of GCAPmediated activation of GC-E (Peshenko et al., 2011) showing half-maximal inhibition in the lower nanomolar range reaching complete inhibition > 100 nM (**Figure 3A**). The inhibitory profiles of RD3 inhibition were nearly identical for GCAP1 and GCAP2 mediated activation of GC-E. In comparison to WT GC-E, we tested the three mutants that have residual (E841K and K846N) or constitutive activity (V902L) by setting up the same titration series with purified RD3 (**Figures 3B–D**). Inhibition by RD3 differed in all mutant cases from inhibition of the WT, except for E841K in the presence of GCAP2. When E841K was tested with GCAP1 and increasing concentrations of RD3, halfmaximal inhibition is shifted to higher concentrations of RD3 (**Figure 3B**). A shift in half-maximal inhibition was also observed for the mutant K846N (**Figure 3C**), but the GC-E activity was only suppressed to 50% at around 500 nM RD3 in the presence of GCAP1 and to more than 90% in the presence of GCAP2.

Most interestingly, the constitutively active mutant V902L stayed active even in the presence of high concentrations of RD3, which are sufficient for completely suppressing GCAP-mediated activity of the WT (**Figures 3A,D**). The presence or absence of


NA, not available; IT, implicit time. <sup>∗</sup>Age at ERG testing is indicated. If measurements were performed at an age that is different from the ERG testing age, the age is indicated in parentheses. Best corrected visual acuity is presented in decimal values as an average of the two eyes. #Published previously.

GCAPs did not lead to a significant difference in the inhibitory profiles.

# Ca2+-Sensitive Activation of GC-E Mutants

The constitutive activation of the V902L mutant seems to mimic the activation of GCAPs. GCAP1, but not GCAP2, caused a slight increase on top of the activity without GCAPs (see above and **Figure 2**). Since the experiment in **Figure 3** did not show any possible distortion of the Ca2+-sensitive regulation, we tested for an effect in the presence of GCAP1, GCAP2, and without GCAPs present. The activation profile of the V902L mutant shifted about 100-fold from an IC<sup>50</sup> of 0.26 (WT) to an IC<sup>50</sup> of 22.03 µM free Ca2<sup>+</sup> (**Figure 4A**, upper panel), which further adds to the severe dysregulation of this GC-E mutant. An IC<sup>50</sup> shift for GCAP2 was also observed, but to a lesser extent. Here the value shifted from 0.16 to 8.56 µM free Ca2<sup>+</sup> (**Figure 4A**, middle panel). Interestingly, when no GCAPs are present, the V902L mutant was inhibited by increasing free Ca2<sup>+</sup> resulting in an IC<sup>50</sup> of about 61 µM free Ca2<sup>+</sup> (**Figure 4A**, lower panel). Apparently, this inhibitory effect was not mediated by GCAPs. Instead, it could originate from a competition of Ca2<sup>+</sup> with Mg2<sup>+</sup> that is a necessary co-factor of GTP in the catalytic site. However, since Mg2<sup>+</sup> concentrations in the assay medium are relatively high with 1 mM free Mg2+, this effect is normally visible at millimolar Ca2<sup>+</sup> concentrations.

The mutants E841K and K846N shared some characteristic features in their Ca2+-sensitive regulation in the presence of GCAPs (**Figure 4B**). They showed no complete inhibition at high free Ca2<sup>+</sup> concentrations of 50–100 µM, a shift in the IC<sup>50</sup> for K846N with GCAP1 and for E841K and K846N with GCAP2. No shift was observed for E841K in the presence of GCAP1. **Table 2** summarizes data on biochemical properties of the GC-E mutants.

# Effect of Increasing GCAP Concentration on the V902L Mutant

Experiments in **Figure 2** indicated a GCAP independent high activity for the V902L mutant. We assayed the activity of the V902L mutant with increasing GCAP1 and GCAP2 concentrations (0–10 µM; **Figure 5**). No effect of GCAP1 or GCAP2 on the GC-E V902L mutant activity was observed, not even at high GCAP concentrations of 10 µM that are saturating concentrations for WT GC-E. Probably GCAPs have no stimulating effect on the V902L mutant, but still exhibit an inhibitory effect, as seen in the IC<sup>50</sup> measurements.

# DISCUSSION

Guanylate cyclases are expressed in two forms in human photoreceptor cells: GC-E and GC-F (Dizhoor et al., 1994; Lowe et al., 1995). They play a central role in phototransduction and mutations in the GUCY2D gene coding for human GC-E can lead to severe retinal diseases in humans. Inherited retinal diseases display a very heterogeneous group of disorders and the number of causative genes heads toward 300. A total number of 144 different GUCY2D mutations have been described so far (see Sharon et al., 2018 for a recent update). The majority (127

mutations) result in a LCA phenotype in the affected patients. While LCA-related mutations are usually recessive and null (mainly frameshift, non-sense, and splicing mutations) and can affect all domains of the GC-E enzyme, CRD mutations are mainly dominant missense clustered in a "hot-spot region" in the DD, at positions between E837 and T849 (Wilkie et al., 2000; Kitiratschky et al., 2008; Sharon et al., 2018). To answer the question why some GUCY2D mutations lead to a CRD phenotype and others to LCA, it is important to understand how these mutations influence the enzyme properties. Different studies already investigated the effect of GUCY2D point mutations. Some general findings are summarized in **Figure 6** (Rozet et al., 2001; Tucker et al., 2004; Peshenko et al., 2010; Jacobson et al., 2013; Zägel and Koch, 2014).

In this study, we characterized five different GC-E mutants. Patients with the homozygous missense mutation p.A710V showed extinct ERG responses and nystagmus. All exhibited poor vision and nyctalopia before 1 year of age. Position A710 is located in the KHD and highly conserved between species. Molecular modeling approaches implied that the mutation probably leads to a loss of GC-E helical structure, which might affect the catalytic center (Gradstein et al., 2016). Mutations affecting the JMD and KHD of the GC-E typically show no or very low basal activity and cannot be activated by GCAPs (Duda et al., 1999; Jacobson et al., 2013). These mutations may change the overall structure of the protein preventing activation of GC-E by GCAPs. Direct binding of GCAP1 could be impaired, since either its binding site or an important activity control site is located in this domain (Sulmann et al., 2017). Further, the KHD harbors a putative Mg2<sup>+</sup> binding site that is part of the nucleotide (ATP) binding site and is suggested to stabilize the active conformation of the catalytic domain by multiple hydrogen bonds (Bereta et al., 2010). The mutation p.A710V is located within a structural motif of the GC-E, called <sup>708</sup>WTAPELL<sup>714</sup> motif, which is critical for the regulatory catalytic activity of GC-E and conserved in all membrane GCs (Duda et al., 2011). Our experiments showed that the A710V mutant lacks any enzymatic activity. This may explain the LCA phenotype in these patients, because LCA1 in general is related to a loss of GC-E function or proper expression. Interestingly, the complete deletion of the WTAPELL motif affects the GC-E to a lesser extent. The basal GC-activity was normal, but activation by GCAP1 and GCAP2 was reduced. Single point mutations W→A, T→A, P→A, and E→A did not affect the basal activity and activation by

FIGURE 4 | IC<sup>50</sup> measurements for GC-E wildtype and mutants. (A) Comparison of GC-E wildtype and mutant V902L, activated by GCAP1 (upper panel) and GCAP2 (middle panel). An additional Ca2<sup>+</sup> titration of the mutant V902L (lower panel) without GCAPs was performed. (B) Comparison of GC-E wildtype with the mutants E841K and K846N. GC-E from HEK cells and GCAPs (5 µM) were incubated at different [Ca2+] ranging from 1.7 nM to 1 mM. The relative GC-E activity was calculated via the amount of produced cGMP. The GC-E activity at the lowest [Ca2+] was set to 1. IC<sup>50</sup> values are indicated in the graph and represent the half-maximal inhibition by Ca2+. Each data point represents the mean value of three replicates with the standard deviation. The experiment was repeated for GC-E wildtype and V902L with similar results. The data were fitted in SigmaPlot 11.0 with a Hill 3 parameter fit. The inset shows IC<sup>50</sup> values with the standard error of the fitting.

GCAPs yielded 25–50% compared to the WT (Duda et al., 2011). These less dramatic effects would probably not lead to a LCA phenotype assuming that corresponding mutant proteins are still transported to outer segments and not degraded. In contrast, the A710V exchange seems to influence the overall GC-E structure more dramatically and completely abolished its activity in our enzymatic assays.

The two DD mutants E841K and K846N exhibited typical hallmarks of activity changes in comparison to the WT guanylate cyclase as summarized in Sharon et al. (2018). These are mainly reduced basal activity and a drastically reduced activation by GCAP1 and GCAP2. The low, but remaining activation by GCAPs was shifted to higher Ca2<sup>+</sup> concentrations keeping the GC-GCAP complex constitutively active (**Figures 2**, **4B**). The consequence of permanent cGMP production under conditions of high cytoplasmic Ca2<sup>+</sup> would open CNG-channels and increase Ca2<sup>+</sup> influx. Accumulation of cGMP and disturbance in the Ca2<sup>+</sup> homeostasis of the cell can have neurotoxic effects (Iribarne and Masai, 2017). This may explain the progressive CRD phenotype in patients. In line with the biochemical analysis, patients with the p.E841K or p.K846N mutations were diagnosed with maculopathy and CRD (Lazar et al., 2014). We demonstrated here that the c.2538G > C (p.K846N) mutation is a de novo paternal mutation, and therefore verifying the pathogenicity of this sequence variant. Only one previous GUCY2D de-novo mutation has been reported thus far (Mukherjee et al., 2014).


standard deviation. The experiment was repeated with similar results.

Usually, CCD mutations are associated with a complete loss of GC-E function and a LCA phenotype. The newly identified CCD mutations p.P873R and p.V902L differ in these aspects and are the first described CRD-related mutations that are found in the catalytic domain of the GC-E. Four individuals of the MOL0064 family participated in this study, showing early onset retinal degeneration and extinguished or severely reduced ERG responses. The p.P873R mutation abolished the activity of the enzyme completely and is therefore expected to be associated with a recessive LCA phenotype. The patients, however, were heterozygous for the p.P873R variant, which cosegregated perfectly in the family and no possible disease-causing mutation has been identified on the counter allele. Earlier studies showed that CCD mutants can exhibit dominant negative effects, so that the disease is also prominent in heterozygotes with milder effects (Tucker et al., 2004). This means that in these patients also GC-E WT protein is present, but to a much lesser extent.

Furthermore, RD3 mediated trafficking can be effected in LCA-related CCD mutations due to less efficient binding of RD3 to GC-E (Zulliger et al., 2015). Interestingly, all tested mutants (E841K, K846N, and V902L) were less inhibited by RD3 than WT GC-E. This indicates a possible role for RD3 in the disease development in case of GUCY2D mutations. Lower GC-E inhibition in photoreceptor inner segments would lead to nonbalanced cGMP production in inner segments and uncontrolled activation of cGMP target proteins (e.g., protein kinase G, CNGchannels). A disturbance of the inner to outer segment trafficking (Azadi et al., 2010) would lead to lower expression levels of GC-E in the outer segment and an imbalance of cGMP levels. Future studies need to address, which consequences develop from such distortions.

The V902L variant displayed a unique biochemical phenotype. The point mutation resulted in a GCAP independent

Frontiers in Molecular Neuroscience | www.frontiersin.org

fnmol-11-00348 September 21, 2018 Time: 14:45 # 9

permanently active GC-E. It turned out that CRD mutations can also appear in the catalytic domain of the GC-E and they do not always lead to a loss of function. At first glance, the amino acid exchange from valine to leucine is not dramatic. Both are non-polar aliphatic amino acids with a branched chain. With the change to leucine only the side chain is prolonged from an isopropyl to an isobutyl group. Somehow, this exchange leads to structural changes in the catalytic core of the enzyme, that turns the GC-E in a GCAP independent, permanent active form. To date, no complete structure of an active membrane bound mammalian guanylate cyclase was resolved (Potter, 2011). Therefore, good predictions on the conformational changes due to the mutation cannot been made, but one can assume that the mutation causes a stabilization of the enzymatic transition state that in the WT is achieved by the binding of GCAPs. Furthermore, the constitutive activity of the V902L mutant is similar to those seen with DD mutations and high synthesis rates of cGMP would result in the permanent opening of CNGchannels and increased Ca2<sup>+</sup> levels. The effect may be even more severe, because this mutant showed a very high basal activity compared to the other mutants (E841K and K846N) investigated in this study.

We here show that mutations in GUCY2D result in multiple effects on guanylate cyclase function and may provide a basis to develop specific therapies for patients suffering from GUCY2D mutations. Currently, cGMP analogs targeting protein kinase G and CNG-channels are under investigation for the treatment of retinal diseases. Compounds with sufficient efficacy could counteract photoreceptor degeneration, while interfering with photoreceptor death pathways (Vighi et al., 2018). Abnormal cGMP levels are a common feature in retinal diseases and probably also in CRD-related GUCY2D mutations, due to the constitute activation of guanylate cyclase. Therefore, drug treatment approaches may be possible in CRD cases caused by GUCY2D mutations and could include application of cGMP analogs (Vighi et al., 2018) or CNG-channel blockers (Koch and Kaupp, 1985) to counteract high cGMP levels.

Recently, the idea for gene augmentation therapy in LCA1 cases has been suggested (Aguirre et al., 2017), in light of the lack of basal activity of GUCY2D mutants that cause LCA, for example, the mutation p.A710V. This makes it a suitable target for gene replacement therapy, because no interfering native protein will be present. Additionally, most LCA1 patients show apparently normal fundus and some photoreceptors showing normal structure, which is required to restore vision by gene therapy. Although reports on photoreceptor degeneration in LCA1 patients are inconsistent, a recent study with patients aged from 6 months to 37 years described some rod photoreceptors with normal architecture (Perrault et al., 1999; Jacobson et al., 2013; Boye, 2014b). Adeno associated virus-based gene therapy for LCA1 gave promising results in mice leading to structural and functional improvement for at least one year (Boye et al., 2010, 2013; Boye, 2014a). For other LCA types, for example, RPE65 mutations, already promising studies employing gene therapy were performed (MacLaren et al., 2016) and gene augmentation therapy for this gene has been recently approved by the FDA. However, gene augmentation therapy might not be effective for dominant GUCY2D mutations that cause CRD, since the mutant allele produces a mutant protein that affects retinal function even in the presence of a normal protein that is expressed at a similar level. Other negative side effects of introducing exogenous GC-E could arise from the complexation of the GC-E target RD3. Therefore, other approaches, and mainly those abolishing the expression of the mutant allele should be considered. Recent successful in vivo experiments with CRISPR-Cas9 on dominant mutations, including those causing retinal diseases, bring hope for using this technique for GUCY2D dominant mutations as well (DiCarlo et al., 2018). Therefore, depending on the type of mutation in GUCY2D, different therapeutic modalities should be applied.

#### AUTHOR CONTRIBUTIONS

fnmol-11-00348 September 21, 2018 Time: 14:45 # 11

HW designed the study, planned and carried out the experiments concerning the biochemical characterization of GC-E mutants, analyzed the data, wrote the manuscript, and prepared **Figures 2–6** and **Supplementary Figures S2**, **S3**. HW, DS, and K-WK discussed and structured the manuscript. DL, KY, PN, and DS did the genetic analysis of the data and generated **Figure 1** and **Supplementary Figure S1**. EB performed the clinical analysis. K-WK formulated the research question, designed the study, participated in data analysis, and wrote the final paper. All authors revised the manuscript.

#### REFERENCES


#### FUNDING

This study was funded in part by the Chief Scientist Office of the Israeli Ministry of Health (Grant No. 3-12583) and by the Yedidut Research Grant. This work was supported by grants from the Deutsche Forschungsgemeinschaft, KO948/10-2.

#### ACKNOWLEDGMENTS

The authors would like to thank the individuals who participated in the study for their cooperation and Dr. Anat Blumenfeld for her assistance with microsatellite analysis.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00348/full#supplementary-material



GCAP1: stoichiometry of binding and effect of new LCA-related mutations. Biochemistry 49, 709–717. doi: 10.1021/bi901495y


**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 © 2018 Wimberg, Lev, Yosovich, Namburi, Banin, Sharon and Koch. 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) and the copyright owner(s) 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.

# Zebrafish Recoverin Isoforms Display Differences in Calcium Switch Mechanisms

#### Dana Elbers, Alexander Scholten and Karl-Wilhelm Koch\*

Department of Neuroscience, Biochemistry, University of Oldenburg, Oldenburg, Germany

Primary steps in vertebrate vision occur in rod and cone cells of the retina and require precise molecular switches in excitation, recovery, and adaptation. In particular, recovery of the photoresponse and light adaptation processes are under control of neuronal Ca2<sup>+</sup> sensor (NCS) proteins. Among them, the Ca2<sup>+</sup> sensor recoverin undergoes a pronounced Ca2+-dependent conformational change, a prototypical so-called Ca2+ myristoyl switch, which allows selective targeting of G protein-coupled receptor kinase. Zebrafish (Danio rerio) has gained attention as a model organism in vision research. It expresses four different recoverin isoforms (zRec1a, zRec1b, zRec2a, and zRec2b) that are orthologs to the one known mammalian variant. The expression pattern of the four isoforms cover both rod and cone cells, but the differential distribution in cones points to versatile functions of recoverin in these cell types. Initial functional studies on zebrafish larvae indicate different Ca2+-sensitive working modes for zebrafish recoverins, but experimental evidence is lacking so far. The aims of the present study are (1) to measure specific Ca2+-sensing properties of the different recoverin isoforms, (2) to ask whether switch mechanisms triggered by Ca2<sup>+</sup> resemble that one observed with mammalian recoverin, and (3) to investigate a possible impact of an attached myristoyl moiety. For addressing these questions, we employ fluorescence spectroscopy, surface plasmon resonance (SPR), dynamic light scattering, and equilibrium centrifugation. Exposure of hydrophobic amino acids, due to the myristoyl switch, differed among isoforms and depended also on the myristoylation state of the particular recoverin. Ca2+-induced rearrangement of the protein-water shell was for all variants less pronounced than for the bovine ortholog indicating either a modified Ca2+-myristoyl switch or no switch. Our results have implications for a step-by-step response of recoverin isoforms to changing intracellular Ca2<sup>+</sup> during illumination.

Keywords: recoverin, photoreceptor, calcium-switch, conformational change, zebrafish

# INTRODUCTION

Light-dependent changes in the second messenger concentration of cGMP and Ca2+-ions control the photoresponse characteristics of vertebrate rod and cone cells (Arshavsky and Burns, 2012; Koch and Dell'Orco, 2013). Feedback control of phototransduction in rod and cone cells crucially depends on cytoplasmic Ca2<sup>+</sup> that is detected by photoreceptor specific neuronal Ca2+-sensor (NCS) proteins. Recoverin is a NCS protein and inhibits G protein-coupled receptor kinase GRK1,

#### Edited by:

Clint L. Makino, Boston University, United States

#### Reviewed by:

Sadaharu Miyazono, Asahikawa Medical University, Japan Polina Geva, Boston University, United States Ching-Kang Jason Chen, Baylor College of Medicine, United States

#### \*Correspondence:

Karl-Wilhelm Koch karl.w.koch@uni-oldenburg.de

Received: 24 July 2018 Accepted: 11 September 2018 Published: 28 September 2018

#### Citation:

Elbers D, Scholten A and Koch K-W (2018) Zebrafish Recoverin Isoforms Display Differences in Calcium Switch Mechanisms. Front. Mol. Neurosci. 11:355. doi: 10.3389/fnmol.2018.00355

also known as rhodopsin kinase, at high levels of free Ca2<sup>+</sup> concentration. Biochemical in vitro data and results on recoverin knockout mice support such a role of recoverin (Kawamura, 1993; Gorodovikova et al., 1994; Chen et al., 1995; Klenchin et al., 1995; Senin et al., 1995; Makino et al., 2004). However, several questions concerning the physiological role of recoverin are still under debate (Morshedian et al., 2018). Strongly coupled with recoverin's function is its so-called Ca2+ myristoyl switch (Zozulya and Stryer, 1992; Dizhoor et al., 1993). Recoverin is posttranslationally modified with a myristoyl group at its N-terminus. In the Ca2+-free state, this acyl moiety is buried inside a hydrophobic cleft. Under saturating Ca2<sup>+</sup> concentrations, the two functional Ca2+-binding sites in recoverin are occupied and the myristoyl group is exposed and can thereby interact with phospholipid membranes (Ames et al., 1995; Senin et al., 2002).

The pronounced conformational change and subsequent membrane translocation in response to changes in Ca2<sup>+</sup> have made recoverin a working model for myristoyl switch mechanisms (Ames et al., 1997; Lange and Koch, 1997). Recent studies address the influence of lipid composition and membrane fluidity upon interaction of recoverin with membranes (Calvez et al., 2016; Potvin-Fournier et al., 2016, 2017; Yang et al., 2016), the impact on the structural organization of phospholipids (Brand and Koch, 2018), the orientation of membrane anchored recoverin (Timr et al., 2017), the intermediate states during Ca2+-dependent conformational transitions investigated by molecular dynamics simulations (Timr et al., 2018), and the complex interactions of recoverin with liposomes and CaF<sup>2</sup> nanoparticles (Marino et al., 2014).

The zebrafish (Danio rerio) retina expresses four recoverin genes, rcv1a, rcv1b, rcv2a, and rcv2b coding for proteins zRec1a, zRec1b, zRec2a, and zRec2b, respectively (Zang et al., 2015). These recoverin isoforms differ in their expression pattern in the adult retina as zRec1a was only found in rods and UV cones, whereas all other zRec forms are present in all cone photoreceptors, but immunohistochemical staining of zRec2a is also seen in bipolar cells. The in vivo function of zRec proteins was studied by a morpholino-based targeted gene knockdown approach and normal and spectrum electroretinography (ERG). Target binding and regulation is suggested to occur with orthologs of mammalian GRK1 and GRK7 (Wada et al., 2006; genes annotated as grk1a, grk1b, grk7a, and grk7b) yielding pairs GRK1a-Rec1a (in rods) and GRK7a-Rec2a (in UV cones), but regulation of GRK1b and GRK7b by Rec2b is feasible as well (Rinner et al., 2005; Zang et al., 2015). According to the flash response data, zRec2a and zRec2b operate under different light regimes indicating different Ca2+-sensitive properties. Except for the studies of Zang et al. (2015), no molecular properties of zRec variants are known so far.

Guanylate cyclase-activating proteins (GCAPs) are related to recoverin and are activator/inhibitor proteins that regulate membrane bound sensory guanylate cyclases (Palczewski et al., 2004; Dizhoor et al., 2010; Koch and Dell'Orco, 2013). The variety of zRec isoforms is reminiscent of zebrafish GCAPs, of which six isoforms are expressed in the zebrafish retina (Imanishi et al., 2004; Rätscho et al., 2009). Detailed studies on zGCAP expression profiles in larval and adult animals, their regulatory properties, Ca2<sup>+</sup> sensitivities, and conformational dynamics revealed a differential action mode for each protein supporting a Ca2<sup>+</sup> relay mode of Ca2+-dependent negative feedback regulation (Scholten and Koch, 2011; Fries et al., 2012; Sulmann et al., 2015; Lim et al., 2017). We suggest a similar regulatory mode for zRec forms. A first step to unravel the mechanism of differential response properties of zRec to oscillating changes in cytoplasmic Ca2<sup>+</sup> concentration is to investigate possible Ca2+-myristoyl switches, to determine their Ca2+-binding properties and conformational dynamics.

#### MATERIALS AND METHODS

#### Protein Expression and Purification

All recoverin isoforms (zRec1a, zRec1b, zRec2a, and zRec2b) were heterologously expressed in Escherichia coli as described previously for bovine recoverin (bRec) (Lange and Koch, 1997; Senin et al., 2002). The zRec cDNA were provided by Prof. Dr. Stephan Neuhauss (University of Zurich, Switzerland) and cloned into plasmids pET21a(+) (zRec2b) or pET11d(+) (zRec1a, zRec1b, and zRec2a) by standard cloning techniques. To obtain myristoylated isoforms, E. coli cells were cotransformed with the plasmid pBB131 containing a gene for the yeast N-myristoyltransferase. After cell lysis, the isoforms were extracted from the insoluble fraction by homogenization in 6 M guanidine-hydrochloride and following refolding by dialysis against Tris buffer (20 mM Tris, 150 mM NaCl, 0.1 mM DTT, pH 8.0). The soluble and insoluble fractions of all recoverin forms were purified by a hydrophobic interaction chromatography except of zRec1b, which was purified by an ammonium sulfate precipitation followed by a size exclusion chromatography. After purification, recoverin containing fractions were combined and dialyzed against 50 mM ammonium hydrogen carbonate to remove residual EGTA that is present from the hydrophobic interaction chromatography step, followed by a buffer exchange against decalcified 50 mM ammonium hydrogen carbonate (purified recoverin samples are shown in **Supplementary Figure S1A**). Degree of myristoylation was determined by reversed phase high-performance liquid chromatography as previously described (Lange and Koch, 1997; Senin et al., 2002) yielding 89% (zRec1a), 63% (zRec1b), 96% (zRec2a), 73% (zRec2b), and 96% (bRec).

#### Antibodies

Recombinant non-myristoylated zrecoverin isoforms were used as antigens by a company (Pineda, Berlin, Germany) for producing polyclonal antibodies in rabbits. The obtained sera were further purified with an affinity chromatography column. For this purpose, each recombinant non-myristoylated recoverin (zRec1a, zRec1b, zRec2a, and zRec2b) was immobilized on a CNBr-activated sepharose column and the corresponding serum was passed over the column to remove unspecific antibodies. Due to the elution, all antibodies were diluted 1:10. Anti-zRec2a crossreacted also with zRec2b; therefore, two purification steps were necessary. The first step was catching

cross-reactive antibodies by immobilizing zRec2b to the column and collecting the non-bound fraction, which was passed over the zRec2a affinity column in the second step. The specificities of all antibodies were tested via western blot (**Supplementary Figure S1B**).

# Ca2+-Dependent Membrane Binding

An equilibrium centrifugation assay was performed to study the membrane binding of all recoverin isoforms in a Ca2+ dependent manner (Senin et al., 2002). For this assay, 1 mg/ml recoverin in HEPES buffer (10 mM HEPES/KOH, 150 mM NaCl, 20 mM MgCl2, 1 mM DTT, pH 7.4) were incubated with 2 mg/ml urea-washed rod outer segment (ROS) membranes and 2 mM CaCl<sup>2</sup> or 2 mM EGTA. After 30 min of incubation (25◦C, 700 rpm), the samples were centrifuged (30 min, 13,000 rpm), supernatants were discarded, and pellets were resuspended with 2 mM CaCl<sup>2</sup> or 2 mM EGTA in HEPES buffer to remove unbound recoverin. After additional 30 min of centrifugation, supernatants were discarded and pellets were resolved in sample buffer. The samples were separated by SDS-PAGE, followed by a protein transfer to a nitrocellulose membrane by a semi-dry blotting using Towbin buffer (25 mM Tris, 192 µM glycine, 20% methanol p.a.). After the transfer, all blots were blocked with 1% milk powder (1 h, RT) in TBS-T (20 mM Tris, 154 mM NaCl, 0.05 % Tween-20). After blocking, incubation followed with primary specific antibodies against zRec isoforms (see below) and against bRec (rabbit anti-Rec k/2+; Lambrecht and Koch, 1992). Primary antibodies in 1% milk powder in TBS-T were incubated for 1 h at room temperature using the following dilutions: for zRec1b (1:20,000), zRec2b (1:40,000), and bRec (1:30,000). Anti-zRec1a and Anti-zRec2a antibodies were incubated over night at 4◦C (dilution: zRec1a at 1:20,000, zRec2a at 1:10,000). Incubation with primary antibodies was followed by incubation with the secondary antibody that was a goat anti-rabbit IgG coupled to peroxidase (Dianova) for 1 h at room temperature. Dilution was for all zRecs 1:20,000, and for bRec 1:30,000. Every incubation step was followed by different washing steps with TBS and TBS-T. For visualization, membranes were incubated for 1 min in WesternBright reagents (Avansta) and then exposed to LucentBlue X-ray films (Advansta). Signal intensity was determined by a densitometric analysis of antibody reactive bands with an AlphaImager (Biozym). For each band, the integral density value (IDV) was determined by measuring the density in a certain area and subtracting the background of the blot.

# Determination of Ca2+-Binding Constants

A chelator assay was employed for the determination of macroscopic Ca2+-binding constants (Linse, 2002; Dell'Orco et al., 2010a) with the following modifications. Oregon GreenTM 488 BAPTA-5N (Invitrogen) was used instead of Dibromo-BAPTA as the competing chelator. It has a K<sup>D</sup> value of 20 µM in MOPS buffer (10 mM MOPS, 100 mM KCl, pH 7.2) (Agronskaia et al., 2004), thus matching the Ca2+-binding constants for myristoylated and non-myristoylated bRec (17 µM for myristoylated bRec, 0.11 and 6.9 µM for non-myristoylated bRec, respectively; Ames et al., 1995). Using non-myristoylated bRec as internal standard, we validated the method for our purpose yielding a pattern of similar apparent K<sup>D</sup> values (K 1 <sup>D</sup> = 0.79µM and K 2 <sup>D</sup> = 13.7µM). Recoverin concentrations of stock solutions were determined by using a recoverin-specific Bradford calibration curve. Titrations were performed with 15 µM recoverin and 0.5 µM BAPTA-5N in decalcified MOPS buffer. To decalcify the buffer, the buffer passed several times over a Chelex column (Chelex 100 <sup>R</sup> sodium form, Sigma). The initial [Ca2+] in the titration was determined by BAPTA and ranged between 15 and 60 nM. Chelator fluorescence spectra were measured with a fluorescence spectrophotometer from Photon Technology International using excitation at 494 nm and recording emission spectra between 500 and 560 nm. The Ca2<sup>+</sup> titration was performed by using calcium stocks of 3 mM and 30 mM CaCl<sup>2</sup> in decalcified MOPS-buffer. In a total volume of 500 µl protein solution, 15 injections of 0.5 µl of 3 mM CaCl2, followed by seven injections of 0.5 µl of 30 mM CaCl<sup>2</sup> were done. For each [Ca2+], the peak amplitude at 524 nm was determined, normalized, plotted as a function of free calcium concentration and fitted by the CaLigator software (Andre and Linse, 2002).

#### Surface Plasmon Resonance

A Biacore 2000 and Biacore 3000 surface plasmon resonance (SPR) instrument (GE Healthcare) was used for detecting Ca2+-induced conformational changes of recoverin variants as described before for other NCS proteins (Dell'Orco et al., 2010b, 2012; Sulmann et al., 2014). Immobilization at high densities on a commercially available CM5 sensor chip (GE Healthcare) is a prerequisite for the detection of conformational transitions by SPR as previously outlined in detail (Sulmann et al., 2014). All myristoylated (myr) recoverin variants were immobilized by thiol coupling yielding sufficient immobilization densities between 4 and 13 ng × mm−<sup>2</sup> (1000 RU correspond to 1 ng protein per mm<sup>2</sup> ; myr bRec = 3.6-5.4 ng × mm−<sup>2</sup> , myr zRec1a = 7.9 ng × mm−<sup>2</sup> , myr zRec2a = 12.6-13.5 ng × mm−<sup>2</sup> , myr zRec2b = 11.1 ng × mm−<sup>2</sup> ). After immobilization, increasing [Ca2+] in a decalcified Tris buffer (5 mM Tris, 100 mM KCl, pH 7.5) were injected and flushed over the protein surface. Control injections over an empty flow cell were performed and subtracted from the sample flow cell (Dell'Orco et al., 2010b, 2012; Sulmann et al., 2014). By plotting the amplitude of the RU signal as a function of free [Ca2+], the half maximal change of response amplitude was determined after normalization by SigmaPlot 13 with a dynamic curve fitting with sigmoidal equation of each experiment, followed by averaging all K1/<sup>2</sup> values and determining the standard deviation. The experiment was only performed with myristoylated isoforms; it was not possible to immobilize non-myristoylated isoforms in a sufficient amount. Furthermore, immobilization and subsequent titrations also failed with zRec1b.

#### Fluorescence Studies

Changes in fluorescence emission of 8-anilinonaphthalene-1 sulfonic acid (ANS) caused by interaction with exposed protein regions (Hughes et al., 1995; Gorczyca et al., 2003) were recorded with a fluorescence spectrophotometer from Photon Technology

International. Lyophilized protein was dissolved in HEPES buffer (80 mM HEPES/KOH, 40 mM KCl, 1 mM DTT, pH 7.5) and protein concentration was determined by a Bradford assay (Bradford, 1976). For every experiment, 2 µM recoverin variant, 2.5 µM ANS, and a certain [Ca2+] in HEPES buffer were incubated for 20 min on ice. For every [Ca2+] above 1 µM, a CaCl<sup>2</sup> stock solution in HEPES buffer was prepared, every [Ca2+] under 1 µM was adjusted by a mixture of K2H2EGTA and K2CaEGTA (Tsien and Pozzan, 1989). ANS fluorescence excitation was performed at 380 nm, and emission spectra were recorded between 400 and 550 nm. After every measurement, the cuvette was cleaned by 70% ethanol, 5 mM EGTA, 100% acetone, and several steps of water in between. Data recording and processing were done with the software Felix32 (Photon Technology International). Maximal fluorescence emission for every [Ca2+] was determined, normalized and plotted as a function of free [Ca2+]. In SigmaPlot 13, a dynamic curve fitting with sigmoidal equation was used to determine a half maximal value (K1/2) of each experiment, followed by averaging all K1/<sup>2</sup> values and determining the standard deviation.

#### Dynamic Light Scattering

Dynamic light scattering (DLS) measurements were performed with a Zetasizer Nano-S (Malvern Instruments). For the experiment a polystyrene, disposable, semi-micro cuvette (Ratiolab) was used. Refractive index and viscosity were set to 1.330 and 0.8872 cP (values for water), and temperature was set to 25◦C with 2 min equilibration time. The measurement angle was 173◦ backscatter, and the analysis model was set to multiple narrow models (high resolution). For each measurement, a minimum of 11 runs with 30 repetitions were performed. The used Tris buffer (5 mM Tris, 100 mM KCl, pH 7.5) was filtered through a Rotilabo <sup>R</sup> syringe filter (Carl Roth, 0.22 µM PDVF). Lyophilized protein was dissolved in decalcified Tris-buffer, protein concentration was determined by a Bradford assay and adjusted to a final concentration of 10 µM. After adjusting either a calcium concentration or EGTA concentration of 1 mM, the protein solution was filtered by an AnotopTM 10 filter (Whatman, 0.02 µm) and the measurement was started. After recording, mean and standard deviation for each recoverin variant was determined.

### RESULTS

# Ca2+-Myristoyl Switch and Ca2+Affinity of zRec Variants

Bovine recoverin shows a reversible binding to membranes triggered by changes in free Ca2<sup>+</sup> concentration. This translocation process is connected to the functional role of mammalian recoverin controlling the activity of mammalian GRK1 in a Ca2+-dependent manner, inhibiting GRK1 at high Ca2+, and relieving the inhibition at low Ca2+. All zRec forms contain a consensus sequence for myristoylation and could in principle undergo a Ca2+-myristoyl switch and membrane translocation process. Therefore, we tested purified myristoylated and non-myristoylated zRec variants in a membrane binding assay using an equilibrium centrifugation assay. Surprisingly, only zRec1a, and to a lesser extent zRec2a, interacted with membranes in a Ca2+-dependent way like it is known from bRec (**Figure 1A**). bRec served as a benchmark control, since its Ca2+-dependent association with membranes or lipid mixtures is well described in test tube experiments that show 20-25% of total applied recoverin (Zozulya and Stryer, 1992; Senin et al., 2002). The other myristoylated zRec forms did not show a pronounced Ca2<sup>+</sup> dependency during membrane binding. Instead, the non-myristoylated zRec forms exhibited an untypical binding pattern (**Figure 1B**) as all isoforms except zRec2b bound stronger to the membrane in their Ca2+-free state. By contrast, zRec2b showed no difference between its Ca2+-free and Ca2+-bound state.

Furthermore, we determined the Ca2+-binding affinities of zRec forms by measuring macroscopic Ca2+-binding constants with a modified chelator method (**Figure 2** and **Table 1**). A representative example of a titration is displayed in **Figure 2** for myristoylated zRec2a revealing an apparent K<sup>D</sup> of 11.9 µM (**Table 1**), which is close to the apparent K<sup>D</sup> values of 14-17 µM obtained previously with bRec (Ames et al., 1995; Senin et al., 2002; Weiergräber et al., 2006). Myristoylated zRec1a with an apparent K<sup>D</sup> of 15.4 µM fell also in this range. Larger differences became visible with myristoylated zRec1b and zRec2b revealing 9.2 and 23.4 µM, respectively (**Table 1**). Titrations with nonmyristoylated zRec isoforms allowed the determination of two apparent K<sup>D</sup> values. Fitting of binding curves gave for all zRec isoforms one apparent K<sup>D</sup> of higher affinity and one of lower affinity (**Table 1**) resembling the Ca2+-binding studies with nonmyristoylated bRec (Section "Materials and Methods"; Ames et al., 1995; Senin et al., 2002; Weiergräber et al., 2006). Previous <sup>45</sup>Ca2+-binding studies assigned the K<sup>D</sup> values of different affinities to the functional EF hands 2 and 3 in bRec (Senin et al., 2002). Although our data seem to reproduce the general pattern of high and low affinities, they indicate in all cases lower affinity of non-myristoylated zrecoverin variants for Ca2+.

#### Conformational Changes

Although all myristoylated zRec forms bound Ca2<sup>+</sup> with moderate, but different affinity, they differed in their Ca2+ dependent membrane attachment. A lack of Ca2+-dependent attachment to the membrane could indicate that these zRec forms do not exhibit a large or distinct conformational change leading to the typical Ca2+-myristoyl switch. Therefore, we tested for conformational transitions in zRec variants by three different independent methods, SPR, ANS fluorescence spectroscopy, and dynamic light scattering (DLS), which allow the investigation of conformational changes from different perspectives.

#### Surface Plasmon Resonance

Ca2+-sensor proteins such as recoverin are well suited for monitoring conformational changes by SPR devices under precisely defined experimental conditions (Dell'Orco et al., 2010b, 2012; Sulmann et al., 2014). For this purpose, we immobilized zRec variants on hydrophilic dextran-coated sensor chips and injected increasing concentrations of CaCl<sup>2</sup> in the nanomolar to micromolar range (**Figure 3**). All titrations with

and binding of myristoylated (A) and non-myristoylated (B) recoverin forms was probed in the presence (2 mM Ca2+) or absence (2 mM EGTA) of Ca2+. Variants of zRec and bRec were detected by isoform specific antibodies via western blotting, followed by a densitometric analysis of protein bands. IDV, integral density value; myr, myristoylated; nm, non-myristoylated. Shown is the mean ± s.d. N = 4, (EGTA) N = 5; nm zRec2a (Ca2+) N = 5, (EGTA) N = 4; myr zRec2b (Ca2+) N = 5, (EGTA) N = 7; nm zRec2b (Ca2<sup>+</sup> and EGTA) N = 5.

immobilized myristoylated zRec variants showed an increase in the maximal amplitudes at defined Ca2<sup>+</sup> concentrations resembling those reversible changes in resonance units (RU) that were previously reported for bRec. **Table 2** gives a summary of the determined K1/<sup>2</sup> values (Ca2<sup>+</sup> concentration at which the increase in amplitudes is half-maximal). K1/<sup>2</sup> values of zRec forms are around 9-10 µM and are thus slightly higher than 6.1 µM obtained for bRec and those reported in the literature (Dell'Orco et al., 2010b, 2012; Sulmann et al., 2014). These titrations demonstrate that zRec forms undergo distinct conformational changes reflecting changes in the hydrodynamic properties of the protein and that these changes seem to be triggered at similar [Ca2+]. However, they do not provide information about Ca2+-dependent exposition of domains or regions, about Ca2+-sensitive changes in hydrodynamic parameters or about the movement of the myristoyl group.

#### ANS Fluorescence Spectroscopy

ANS interacts non-covalently with hydrophobic regions in proteins. Ca2<sup>+</sup> binding to zRec variants could trigger the exposure or burying of hydrophobic parts (the myristoyl group or hydrophobic amino acid residues) and therefore, can monitor conformational changes of the whole protein by increasing (or decreasing) fluorescence emission. ANS fluorescence emission of myristoylated zRec2a was half-maximal at a K1/<sup>2</sup> of 22.6 µM [Ca2+] (**Figure 4**) similar to the bovine control variant (24.6 µM, **Supplementary Table S1**). All other zRec forms showed significantly different K1/<sup>2</sup> values being either fivefold lower (myristoylated zRec1b) or two- to-threefold higher (**Supplementary Table S1**). However, more unexpected differences became visible except for zRec1a, when we compared myristoylated and non-myristoylated variants (**Supplementary Table S1**) indicating a more specialized role of the myristoyl moiety for each zRec (see Section "Discussion" below).

We further analyzed difference fluorescence spectra providing information about the net fluorescence change between the Ca2+-saturating and the Ca2+-free state (**Figures 4B,C**). Most similar to the myristoylated bRec control was zRec2a, all other zRec variants exhibited lower net changes leading to the following sequence of 1F: bRec > zRec2a > zRec1a > zRec2b > zRec1b. This pattern changed completely, when the myristoyl group was absent leading to a large net change in zRec1b, a lower change



Determination of apparent K<sup>D</sup> values (AppK 1 D and AppK 2 D ; mean ± s.d.) by using the chromophoric chelator BAPTA-5N conjugated with the fluorescence dye Oregon Green 488 as described in Section "Materials and Methods" and in the legend of Figure 2. All curves were fitted with two-site binding model yielding two constants for non-myristoylated recoverin forms. For myristoylated recoverin variants, we calculated one macroscopic binding constant according to K<sup>D</sup> = 10−(logK1 + logK2)/2 as previously described for GCAP1 variants (Dell'Orco et al., 2010a) taking into account that myristoylated bRec shows one Ca2+ affinity constant in other Ca2+-binding assays.

for other zRec forms that was still higher than the comparison with bRec (1F: zRec1b > zRec2a > zRec2b > zRec1a/bRec; **Figure 4C**). Maximal amplitudes of ANS fluorescence emission in the presence of Ca2<sup>+</sup> gave an identical sequence (**Supplementary Table S2**).

#### Dynamic Light Scattering

For dynamic light scattering, all recoverin isoforms were investigated in the absence and presence of Ca2<sup>+</sup> to determine a potential difference in the conformation based on Ca2<sup>+</sup> binding. Every variant was tested at least 40 times in a myristoylated and non-myristoylated form. To determine the hydrodynamic



Values of K1/2 are the free [Ca2+] at which SPR amplitudes are half maximal. Maximal amplitude in RU at saturating [Ca2+] and normalized per ng of immobilized protein; N, number of repetition (mean ± s.d.). Values for myristoylated zRec1b are missing due to difficulties of protein immobilization and long-term protein stability.

radius, an intensity plot of DLS data was used. The intensity plot showed one higher peak and one lower peak (**Figure 5A**). Concerning the absence of the second peak in a number plot (**Figure 5B**), the second peak was negligible and only the first peak was used for the further analysis. This peak reflected the size of the investigated protein revealing the hydrodynamic radius (**Figure 5C**). Myristoylated bRec showed the largest difference of the hydrodynamic radius between the Ca2+-bound and Ca2+-free state among all recoverin forms (**Table 3**) revealing for the Ca2+-bound state, a larger hydrodynamic radius of 7.54 ± 0.17 nm than for the Ca2+-free state, 6.15 ± 0.14 nm (**Table 3**). Nearly all other myristoylated zRec forms also showed a larger hydrodynamic radius in the Ca2+-bound state but the difference was not as prominent as for the bovine form. Only myristoylated zRec2a had a slightly larger hydrodynamic radius in the Ca2+-free state (**Table 3**).

All non-myristoylated zRec forms except zRec2b have a larger hydrodynamic radius in the Ca2+-free than in the Ca2+ bound state (**Table 3**), which was opposite to non-myristoylated bRec, whose hydrodynamic radius was nearly the same in the Ca2+-bound (6.4 ± 0.11 nm) than in the Ca2+-free state (6.34 ± 0.13 nm).

#### DISCUSSION

Zebrafish recoverin isoforms share a high sequence homology to the bovine (or mammalian) ortholog (**Supplementary Figure S2**). However, our comparative analysis of the molecular properties of zebrafish recoverin forms demonstrate a pattern of significant differences, which we will discuss for each of the experimental approaches.

Mammalian recoverin, in particular bRec, is the prototype of those NCS proteins that undergo a Ca2+-myristoyl switch, which has been probed experimentally by measuring the Ca2+ dependent binding of myristoylated Rec forms to biological membranes (Zozulya and Stryer, 1992; Dizhoor et al., 1993; Senin et al., 2002). Other NCS proteins undergoing a Ca2+-myristoyl switch are, for example, hippocalcin, neurocalcin δ, and visininlike proteins 1 and 3 (Kobayashi et al., 1993; Ladant, 1995; Spilker et al., 2000, 2002). In the Ca2+-free state, the myristoyl group is buried in a protein cleft consisting of hydrophobic amino acid side chains. All zRec variants have identical (or

highly conserved) amino acids at the corresponding positions (**Figure 6A**) indicating that all Ca2+-free zRec isoforms keep the myristoyl group buried in a hydrophobic pocket. Critical residues for the Ca2+-triggered transition (and extrusion of the myristoyl group) are Gly42 and Gly96, around which rotational movements occur, and His68 and Leu108, which are involved in the changing domain interactions of EFhand 2 and 3 (Ames et al., 2002). All these amino acids are identical in zRec variants and bRec. Differences in Ca2+ switch mechanisms as we see with zRec1b and zRec2b must therefore have other causes (see Section "Discussion" further below).

Unexpected was the higher binding of Ca2+-free nonmyristoylated zRec1a, 1b, and 2a to membranes than binding of their Ca2+-saturated forms to membranes. Binding of proteins to membranes can be the result of hydrophobic and electrostatic interactions. Amino acid sequences of zRec isoforms differ in a few segments from those of bRec. Most of these short stretches of two to three amino acids show no consensus to corresponding amino acids in bRec. Furthermore, they are located on the protein surface (**Supplementary Figure S3**) and could account for the observed differences.

Ca2<sup>+</sup> affinities of zRec isoforms should determine Ca2+sensitive properties such as the Ca2+-dependent membrane association. In mammalian and zebrafish recoverin forms, the two functional EF hands 2 and 3 are highly conserved in all critical positions with oxygen containing side chains (**Supplementary Figure S2**). However, in EF-hand 3 of zRec2b, a Pro is at the N-terminal part of the exiting helix and located next to the Glu that is essential for Ca2<sup>+</sup> coordination. **Figure 6B** illustrates the position of Glu next to Pro that is substituted for Asn. Pro interfering with the helix structure could slightly disturb or shift the position of the complexing Glu leading to the lower Ca2<sup>+</sup> affinity observed with zRec2b. Furthermore, the C-terminus in bRec was previously identified as an internal modulator of Ca2<sup>+</sup> sensitivity. Truncation

FIGURE 5 | Dynamic light scattering measurement of zRec2a. Exemplary intensity (A) and number (B) plot of zRec2a in presence of Ca2+. DLS data reported the intensity of scattering light as % of the total area. The second peak in the intensity versus size plot originates from oligomeric or aggregated protein. However, the number plot refers to the relative percentage of sizes showing that only a very low percentage of the total protein is in oligomeric or aggregated form. (C) Comparison of the hydrodynamic diameter d (in nm) of zRec2a in the presence (2 mM Ca2+) and absence (2 mM EGTA) of Ca2+. For statistical analysis, see Table 3.


Determination of the hydrodynamic diameter d ± s.d. in the presence (dCa) and absence (dEGTA) of Ca2+. Nm, nanometer; N, number of repetition; 1d, difference dCa − dEGTA. Statistical analysis was done by Student's t-test (nm bRec, nm zRec1a, myr zRec2a, myr zRec2b, and myr zRec2b) or Mann–Whitney U test (myr bRec, myr zRec1a, myr zRec1b, nm zRec1b, and nm zRec2a). Recoverin isoforms with significant P-values of dCa/dEGTA are highlighted with a gray background; P ≥ 0.05, not significant.

of C-terminal amino acids causes a shift to lower Ca2<sup>+</sup> affinity in bRec (Weiergräber et al., 2006), which very likely determines the differences in Ca2<sup>+</sup> affinity of zRec forms. In particular, not only zRec2b but also zRec2a lack the 12 or 10 amino acids, which are present in bRec. However, this C-terminal stretch is present in zRec1a and 1b with only some minor amino acid differences. Thus, the lack of the C-terminus cannot account alone for differences in Ca2<sup>+</sup> sensitivities.

Binding of Ca2<sup>+</sup> to bRec occurs in a sequential order with binding first to EF-hand 3 and second to EF-hand 2 (Permyakov et al., 2000; Senin et al., 2002), thereby triggering the conformational change. Binding of Ca2<sup>+</sup> to zRec isoforms triggers in all variants conformational changes, which we could show by different experimental approaches. Changes in conformation detected by SPR spectroscopy was half maximal in the lower micromolar range in agreement with the Ca2<sup>+</sup> affinity constants determined with the chelator assay, although differences in affinity constants are not directly mirrored by the SPR K1/<sup>2</sup> values. We have observed and discussed this apparent mismatch in previous contributions concluding that the empirical parameter K1/<sup>2</sup> reflects a concerted binding-conformational process (Dell'Orco et al., 2010b, 2012; Sulmann et al., 2014), which occur in different NCS proteins.

ANS fluorescence emission is a tool to estimate the increase of solvent accessible hydrophobic surfaces depending on experimental conditions, for example, Ca2+-binding

FIGURE 6 | Protein regions that are important for recoverin function. (A) Highly conserved amino acid positions forming the hydrophobic myristoyl pocket. (B) EF hands 2 and 3 of bRec with Ca2+-bound (magenta). The essential Glu (E) is indicated in EF-hand 3. An Asn was replaced in silico by Pro (P) to illustrate a possible interference with the position of Glu. Images were created by Pymol using coordinates of Ca2+-free and Ca2+-bound bRec (PDB codes: 1IKU and 1JSA, Tanaka et al., 1995; Ames et al., 1997).

(Hughes et al., 1995; Gorczyca et al., 2003). It has previously been noticed that myristoylated and non-myristoylated bRec expose a large hydrophobic patch on Ca2<sup>+</sup> binding (yellow labeled amino acids in **Figure 7**; Weiergräber et al., 2003). Aromatic and aliphatic amino acids constituting this hydrophobic patch are Leu28, Phe23, Trp31, Phe35, Ile44, Phe56, Phe57, Leu81, Phe83, and Leu90 and are identical in all zRec isoforms except for zRec1a and zRec1b, where Phe23 is replaced by Tyr. The increase in polarity by the presence of Tyr might decrease the general hydrophobicity of the patch (**Figure 7**) and could partially account for the lower 1F value determined for zRec1a and even more for zRec1b, where the surface of the patch decreases.

Two other hydrophobic patches of lower size are visible on the surface of bRec, one made of Leu167, Phe172, Ile173, Leu187, and Ala188 and the second one made of Phe158, Phe159, Ile182, Leu183, Leu185, and Ile186. The latter one forms a kind of canyon-like structure, in which Ile182 in bRec is replaced by Ala and Gly in zRec2a and zRec2b, respectively (orange labeled, **Figure 7**). This hydrophobic groove might be another target of ANS. The Gly in zRec2b decreases the degree of hydrophobicity thereby lowering binding of ANS, whereas Ala in zRec2a has a side chain of lower hydrophobicity than Ile, but probably sufficient for ANS binding (**Supplementary Tables S1**, **S2**). In non-myristoylated bRec, the large hydrophobic patch is solvent accessible in both forms of bRec, with one or two Ca2<sup>+</sup> bound (**Supplementary Figure S4** Weiergräber et al., 2003; Kumar et al., 2015). The hydrophobic canyon-like cleft seems to be exposed increasing the general hydrophobicity, although ANS binding to bRec occurred to nearly the same extent independent on the presence of the myristoyl group. Since we lack the threedimensional structure of any zRec, we have no structure based explanation for the large increase in ANS fluorescence for the zRec isoforms, in particular for non-myristoylated zRec1b. We hypothesize that Ca2+-triggered conformational changes in zRec forms differ from those of bRec, which is already indicated by the differences in membrane association, but became even more apparent in our DLS data. All non-myristoylated zRec isoforms except zRec2b had a larger hydrodynamic radius in the Ca2+-free than in the Ca2+-bound state, which was opposite to bRec. Among all zRec forms, non-myristoylated zRec1b showed the largest difference, which correlates with the high 1F value

in **Figure 4C** and might even explain the unusual binding of zRec1b to membranes in the absence of Ca2+. A similar correlation of DLS data and the 1F value was visible for non-myristoylated zRec2a, and this variant showed also higher binding to membranes in the absence of Ca2<sup>+</sup> than in the presence, but to a general lower degree (**Figure 1B**). However, other zRec variants did not exhibit similar correlations.

We conclude from these observations that all zRec isoforms undergo Ca2+-triggered transitions, but some or all attain different conformations with consequences for the biochemical properties. Recent work by Zang et al. (2015) investigating recoverin deficient larvae suggested that zRec1a could be replaced by zRec2a. This is in agreement with both proteins sharing a similar Ca2<sup>+</sup> sensitivity, which can be seen in their similar K<sup>D</sup> values (**Table 1**). Zang et al. further showed that cone photoresponse recovery differs in zRec2a and zRec2b morphants depending on illumination that triggers transient changes in cytoplasmic Ca2<sup>+</sup> concentration (Brockerhoff et al., 2003; Cilluffo et al., 2004; Leung et al., 2007). These findings are in agreement with the about twofold different Ca2+ binding constants that we determined for zRec2a and zRec2b (**Table 1**). Membrane binding and Ca2+-binding constants of zRec1a were the most similar to those of bRec. Due to the expression of zRec1a in rods (and UV cones) and the relationship of their targets zGRK1a and bovine GRK1, our results are in agreement with a functional pair of zRec1a and zGRK1a. Finally, we cannot exclude that one or more zRec forms target a different protein as this was observed for bRec interacting with caldendrin (Fries et al., 2010) and suggested for mammalian recoverin that is involved in the signal transmission between rod and bipolar cells (Sampath et al., 2005). For example, the high ability of non-myristoylated zRec

#### REFERENCES


forms to associate with membranes in low Ca2<sup>+</sup> concentration in combination with the DLS data could point to conformational differences that are suitable for interacting with different targets.

#### AUTHOR CONTRIBUTIONS

DE, AS, and K-WK designed the study. DE and AS performed experiments. DE and K-WK analyzed data. K-WK wrote the first draft of the manuscript. All authors corrected and approved the final version of the manuscript.

#### FUNDING

This work was supported by a grant from the Science Ministry of Lower Saxony (Lichtenberg stipend to DE) and by the Deutsche Forschungsgemeinschaft (DFG; KO948 15-1).

#### ACKNOWLEDGMENTS

The authors thank Werner Säftel for excellent technical assistance.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00355/full#supplementary-material




Zozulya, S., and Stryer, L. (1992). Calcium-myristoyl protein switch. Proc. Natl. Acad. Sci. U.S.A. 89, 11569–11573. doi: 10.1073/pnas.89.23. 11569

**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 PG and handling editor declared their shared affiliation at the time of the review.

Copyright © 2018 Elbers, Scholten and Koch. 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) and the copyright owner(s) 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.

# Dimerization of Neuronal Calcium Sensor Proteins

James B. Ames\*

INTRODUCTION

Department of Chemistry, University of California, Davis, Davis, CA, United States

Neuronal calcium sensor (NCS) proteins are EF-hand containing Ca<sup>2</sup><sup>+</sup> binding proteins that regulate sensory signal transduction. Many NCS proteins (recoverin, GCAPs, neurocalcin and visinin-like protein 1 (VILIP1)) form functional dimers under physiological conditions. The dimeric NCS proteins have similar amino acid sequences (50% homology) but each bind to and regulate very different physiological targets. Retinal recoverin binds to rhodopsin kinase and promotes Ca<sup>2</sup><sup>+</sup>-dependent desensitization of light-excited rhodopsin during visual phototransduction. The guanylyl cyclase activating proteins (GCAP1–5) each bind and activate retinal guanylyl cyclases (RetGCs) in lightadapted photoreceptors. VILIP1 binds to membrane targets that modulate neuronal secretion. Here, I review atomic-level structures of dimeric forms of recoverin, GCAPs and VILIP1. The distinct dimeric structures in each case suggest that NCS dimerization may play a role in modulating specific target recognition. The dimerization of recoverin and VILIP1 is Ca<sup>2</sup><sup>+</sup>-dependent and enhances their membrane-targeting Ca<sup>2</sup><sup>+</sup>-myristoyl switch function. The dimerization of GCAP1 and GCAP2 facilitate their binding to dimeric RetGCs and may allosterically control the Ca<sup>2</sup><sup>+</sup>-dependent activation of RetGCs.

Keywords: calcium, dimer, GCAP1, GCAP2, GCAP5, recoverin, VILIP1, NCS protein

#### Edited by:

Daniele Dell'Orco, Università degli Studi di Verona, Italy

#### Reviewed by:

Eugene Anatolievich Permyakov, Institute for Biological Instrumentation (RAS), Russia Rameshwar K. Sharma, Salus University, United States

> \*Correspondence: James B. Ames jbames@ucdavis.edu

Received: 31 August 2018 Accepted: 11 October 2018 Published: 02 November 2018

Ames JB (2018) Dimerization of Neuronal Calcium Sensor Proteins. Front. Mol. Neurosci. 11:397. doi: 10.3389/fnmol.2018.00397

Recoverin (Dizhoor et al., 1991; Kawamura and Murakami, 1991) is expressed exclusively in retinal rod and cone cells, where it promotes the desensitization of light-excited rhodopsin (Kawamura, 1993; Erickson et al., 1998; Makino et al., 2004) by inhibiting rhodopsin kinase activity in dark-adapted photoreceptors (Calvert et al., 1995; Chen et al., 1995; Klenchin et al., 1995; Komolov et al., 2009). The Ca2+-bound form of recoverin forms a dimer in solution

Intracellular calcium ion (Ca2+) is a second messenger in the brain and retina that modulates sensory signal transduction processes (Berridge et al., 2000; Augustine et al., 2003). Sensory stimuli cause changes in cytosolic Ca2<sup>+</sup> levels that are detected by a family of Ca2+-binding proteins called, neuronal calcium sensor (NCS) proteins (Ames et al., 1996; Braunewell and Gundelfinger, 1999; Burgoyne and Weiss, 2001; Burgoyne et al., 2004; Weiss et al., 2010; Ames and Lim, 2012). More than 20 different NCS proteins have been identified so far (Weiss and Burgoyne, 2002; Haynes et al., 2012), including recoverin (Dizhoor et al., 1991) and guanylyl cyclase activating proteins (GCAP1–5; Dizhoor et al., 1994; Palczewski et al., 1994) that regulate visual phototransduction in retinal photoreceptor cells (Palczewski et al., 2000; Ames and Ikura, 2002; Stephen et al., 2008; Ames and Lim, 2012). NCS homologs are also expressed in the brain and spinal cord, such as neurocalcin (Hidaka and Okazaki, 1993), frequenin (NCS-1; Pongs et al., 1993; McFerran et al., 1998), visinin-like proteins (VILIPs; Bernstein et al., 1999; Braunewell and Klein-Szanto, 2009) and

hippocalcin (Kobayashi et al., 1992, 1993; Tzingounis et al., 2007).

(Myers et al., 2013) that binds to rhodopsin kinase (Chen et al., 1995; Klenchin et al., 1995). Recoverin dimerization has been suggested to facilitate the binding of rhodopsin kinase with dimeric rhodopsin (Myers et al., 2013). Recoverin dimerization may also regulate light-dependent activation of phosphodiesterase (Chen et al., 2012) and light-induced disulfide dimerization at Cys39 (Permyakov et al., 2007, 2012; Zernii et al., 2015). Lastly, recoverin appears to have alternative functions in the rod inner segment (Strissel et al., 2005) that are implicated in cancer-associated retinopathy (Polans et al., 1991; Subramanian and Polans, 2004).

GCAP1–5 bind to and activate retinal guanylyl cyclases (RetGCs1 and RetGC2; Dizhoor et al., 1994; Palczewski et al., 1994, 2004). The GCAP1-modulated RetGC1 transduction system also exists in the olfactory bulb (Duda et al., 2001). GCAP1, GCAP2 and GCAP5 each form a dimer in solution (Ermilov et al., 2001; Lim et al., 2017, 2018) that binds to dimeric RetGC1 (Liu et al., 1997; Ramamurthy et al., 2001). The GCAPs activate RetGCs at low Ca2<sup>+</sup> levels in light activated photoreceptors (Peshenko and Dizhoor, 2006; Lim et al., 2009), whereas Ca2+-bound GCAPs inhibit RetGCs at high Ca2<sup>+</sup> levels in dark-adapted photoreceptors (Dizhoor and Hurley, 1996; Dizhoor et al., 1998). Surprisingly, Ca2+-bound GCAP1 can stimulate the odorant surface receptor ONE-GC (Duda et al., 2012a), which raises the question about how GCAP1 dimeric sites can recognize two different target sites existing in RetGC1 and ONE-GC. The Ca2<sup>+</sup> sensitive activation of RetGCs by GCAPs in the retina promotes the recovery phase of visual excitation, and particular GCAP1 mutants that disrupt the cyclase activation are linked to retinal degenerative diseases (Semple-Rowland et al., 1996; Sokal et al., 1998; Baehr and Palczewski, 2007; Bondarenko et al., 2010; Jiang and Baehr, 2010).

The VILIP1–3; (Braunewell and Klein-Szanto, 2009) are dimeric NCS proteins (Li et al., 2009) that are expressed exclusively in the brain and spinal cord. VILIP1 is localized primarily in the rat hippocampus (Paterlini et al., 2000; Zhao and Braunewell, 2008), where it controls neuronal excitability important for learning and memory (Braunewell et al., 2003; Brackmann et al., 2004). In particular, VILIP1 binds to the α–subunit of the α4β<sup>2</sup> nicotinic acetylcholine receptor (nAChR), which promotes its surface expression and trafficking in oocytes (Lin et al., 2002) and hippocampal neurons (Zhao et al., 2009b). The Ca2+-induced surface expression of nAChR promoted by VILIP1 therefore modulates neuronal excitability in hippocampal neurons (Gierke et al., 2008; Zhao et al., 2009a,b) and regulates synaptic plasticity (Braunewell, 2005; Braunewell and Klein-Szanto, 2009).

All NCS proteins contain four EF-hand Ca2+-binding motifs (Moncrief et al., 1990; Ikura, 1996), a covalently attached N-terminal myristoyl group, and conserved amino acid residues in the EF-hand motifs, particularly in the Ca2+-binding loops (**Figure 1**). The first EF-hand (EF1) contains a Cys followed by


FIGURE 1 | Amino acid sequence alignment of selected neuronal calcium sensor (NCS) proteins. Residues at the domain interfaced are highlighted in bold and red. Swiss Protein Database accession numbers are P46065 (bovine GCAP1), 51177 (bovine GCAP2), Q5MAC8 (zebrafish GCAP5), P21457 (bovine recoverin), P61602 (bovine neurocalcin-δ) and P62760 (human visinin-like protein 1 (VILIP1)).

Pro in the binding loop that disables Ca2<sup>+</sup> binding at this site in all NCS proteins. The second and third EF-hands (EF2 and EF3) both bind Ca2<sup>+</sup> with high affinity (Cox et al., 1994; Ames et al., 1995). The fourth EF-hand sequence is variable, and Ca2<sup>+</sup> is able to bind to EF4 in neurocalcin-δ (Ladant, 1995) and GCAPs (Peshenko and Dizhoor, 2007; Stephen et al., 2007) but Ca2<sup>+</sup> does not bind to EF4 in recoverin (Ames et al., 1995) and VILIPs (Cox et al., 1994; Li et al., 2011). Ca2+-binding to EF4 in GCAP1 controls whether GCAP1 can activate or inhibit guanylyl cyclase (Peshenko and Dizhoor, 2007). Residues outside the EF-hand motifs are generally not conserved and may play a role in target recognition (Zernii et al., 2011).

N-terminal myristoylation anchors some NCS proteins to cellular membranes by a mechanism termed, Ca2+-myristoyl switch (Dizhoor et al., 1992; Kobayashi et al., 1993; Ladant, 1995). Myristoylated recoverin binds to retinal disc membranes at high Ca2<sup>+</sup> levels in dark-adapted photoreceptors (Zozulya and Stryer, 1992; Dizhoor et al., 1993; Dell'Orco et al., 2012), whereas unmyristoylated recoverin is localized in the cytosol (Zozulya and Stryer, 1992; Dizhoor et al., 1993). Likewise, myristoylated forms of neurocalcin (Ladant, 1995), hippocalcin (Kobayashi et al., 1993) and VILIPs (Li et al., 2011) each exhibit Ca2+ induced localization at the plasma membrane in neurons. The attached fatty acyl group interacts quite differently with each NCS protein as seen in the structures for Ca2+-free recoverin (Tanaka et al., 1995), NCS1 (Lim et al., 2011), GCAP1 (Lim et al., 2016) and VILIP3 (Li et al., 2016). Thus, N-terminal myristoylation serves to fine tune the tertiary structure of each NCS protein in a unique way to promote functional diversity (Ames and Lim, 2012). Recoverin's Ca2+-myristoyl switch may control its light-induced movement into the rod inner segment (Strissel et al., 2005). GCAP proteins are also myristoylated (Palczewski et al., 1994; Frins et al., 1996; Olshevskaya et al., 1997), but do not possess a functional Ca2+-myristoyl switch (Olshevskaya et al., 1997; Hwang and Koch, 2002). Instead the N-terminal myristoyl group remains sequestered inside GCAP1 in both Ca2+-free and Ca2+-bound states (Hughes et al., 1998; Lim et al., 2009) as demonstrated in the crystal structure of Ca2+-bound GCAP1 (Stephen et al., 2007) and NMR structure of the Ca2+-free activator state (Lim et al., 2016).

In this review article, I discuss the recent atomic-resolution structures of dimeric forms of recoverin (Myers et al., 2013), GCAP1 (Lim et al., 2018), GCAP2 (Pettelkau et al., 2012), GCAP5 (Lim et al., 2017) and VILIP1 (Li et al., 2011) that each adopt very different quaternary structures (**Figure 2**). While the tertiary structures of each monomeric subunit are somewhat similar, the distinct quaternary structures and unique subunit packing arrangement at each dimer interface may play a role in facilitating target recognition and specificity.

#### NCS PROTEINS HAVE DISTINCT DIMERIC STRUCTURES

#### GCAP1 Forms a Symmetric and Functional Dimer

Recent NMR (Lim et al., 2009, 2013, 2016) and EPR double electron-electron resonance (DEER; Lim et al., 2018) studies reveal that GCAP1 exists as a dimer in solution. The GCAP1 dimerization is Ca2+-independent and Ca2+-binding to GCAP1 does not appear to cause large changes in the dimer quaternary structure (Lim et al., 2018). A structural model of the GCAP1 dimer was determined recently by DEER (Lim et al., 2018; **Figure 2A**). The GCAP1 dimer is

bound to GCAP5 is orange, and N-terminal myristoyl group in GCAP1 is magenta.

**76**

cyan. Hydrophobic amino acids at the dimer interface are highlighted red.

symmetric (**Figure 2A**), and is stabilized by hydrophobic contacts at the dimer interface (**Figure 3A**). The most noteworthy intermolecular contacts involve hydrophobic residues, H19, Y22, V77 and W94 (**Figure 3A**). In particular, the methyl side-chain atoms of V77 each contact one another at the dimer interface and therefore explain why the V77E mutation dramatically weakens GCAP1 dimerization (Lim et al., 2016). The GCAP1 dimer is further stabilized by aromatic side chains of F73 and W94 that form intermolecular contacts at the dimer interface (Lim et al., 2018). Individual point mutations at the dimer interface in GCAP1 (H19A, Y22A, F73A, V77E and W94A) each weaken the dimerization dissociation constant by more than 10-fold and completely abolish the activation of RetGC by GCAP1 (Lim et al., 2018). Thus, the hydrophobic contacts at the GCAP1 dimer interface (**Figure 3A**) are essential for both its dimerization and activation of RetGCs. This implies that GCAP1 dimerization may be important for activating RetGC1, which itself is a dimer (Liu et al., 1997; Ramamurthy et al., 2001). Therefore, it is tempting to speculate that the GCAP1 dimer (**Figure 2A**) may bind to dimeric RetGC1 to form a 2:2 complex. This binding of the GCAP1 dimer is proposed to induce an allosteric conformational change in the RetGC1 dimer in order to modulate the cyclase activity. The allosteric regulation of RetGC activity may involve quaternary structural changes in the 2:2 complex akin to quaternary structural changes that regulate O<sup>2</sup> binding to hemoglobin (Monod et al., 1965).

#### GCAP2 Forms an Asymmetric Dimer With a Polar Dimer Interface

GCAP2 forms a stable dimer in solution (Ames et al., 1999; Ermilov et al., 2001), and the original NMR structure of GCAP2 (Ames et al., 1999) was determined in the presence of detergent that dissociated the dimer into a stably folded monomer. There is some dispute about the Ca2+-dependence of the GCAP2 dimerization. The original work by Ermilov et al. (2001) determined that GCAP2 forms a dimer only in the Ca2+-free state, and the Ca2+-bound GCAP2 was shown to be monomeric. However, a more recent study suggested that GCAP2 can form a dimer in both the Ca2+-free and Ca2+-bound states (Pettelkau et al., 2013). A structural model of the GCAP2 dimer (in both Ca2+-free and Ca2+ bound states) was determined recently by mass spectrometry (Pettelkau et al., 2013). Ca2+-binding to GCAP2 does not affect the dimer quaternary structure (Pettelkau et al., 2013), similar to the Ca2+-independent dimer quaternary structure of GCAP1 (Lim et al., 2018). However unlike GCAP1, the GCAP2 dimer is asymmetric (**Figure 2B**). The residues at the GCAP2 dimer interface (residues K98, L167, V171, R175, K183, Q186, D188 highlighted red in **Figure 2B**) are not conserved and are completely unrelated to the residues at the GCAP1 interface (highlighted red in **Figure 3A**). Surprisingly, the GCAP2 interface involves mainly polar and charged residues in contrast to the exclusively hydrophobic interface in GCAP1. The positively charged side-chain atoms of R175 in GCAP2 (yellow colored chain in **Figure 2B**) are within 5 Å of the polar side-chain atoms of Q186 in the opposite chain (cyan in **Figure 2B**), and the positively charged side-chain atoms of K98 (yellow chain in **Figure 2B**) are less than 4 Å from the side chain carboxylate atoms of D188 in the opposite chain (cyan in **Figure 2B**). These intermolecular contacts at the GCAP2 dimer interface are not conserved in GCAP1 and may explain why the GCAP2 dimer (**Figure 2B**) is structurally quite different from the GCAP1 dimer (**Figure 2A**). The dramatically different quaternary structures and dimerization interface for GCAP2 compared to GCAP1 might also explain their functional differences (Duda et al., 2012b; Peshenko et al., 2015).

#### GCAP5 Dimerization Is Bridged by Fe2+

Zebrafish photoreceptors contain specific GCAP homologs (GCAP3–5; Imanishi et al., 2004; Rätscho et al., 2009) that are not expressed in mammals. The amino acid sequence of the zebrafish homolog called GCAP5 is the most divergent compared to the amino acid sequences of mammalian GCAP1 and GCAP2 (**Figure 1**). Two non-conserved Cys residues in GCAP5 (Cys 15 and Cys17) were shown recently to ligate Fe2<sup>+</sup> (Lim et al., 2017). Fe2+-binding to GCAP5 serves as a potent inhibitor and the Fe2+-bound GCAP5 is unable to activate RetGC at low Ca2<sup>+</sup> levels in light-adapted photoreceptors, suggesting that Fe2<sup>+</sup> binding to GCAP5 may serve as a redox sensor for phototransduction in zebrafish photoreceptors (Lim et al., 2017). Structurally, the Fe2<sup>+</sup> binding by Cys15 and Cys17 bridges two GCAP5 molecules into a [Fe(SCys)4] dimeric complex (Lim et al., 2017) like that observed previously in two-iron superoxide reductases (deMaré et al., 1996; Min et al., 2001; Emerson et al., 2003). The GCAP5 dimer has a symmetric structure (**Figure 2C**) somewhat similar to that of GCAP1 (**Figure 2A**). The GCAP5 dimer interface contains hydrophobic residues (H18, Y21 and V76 in **Figure 3B**) that are conserved in the GCAP1 dimer (**Figure 3A**). However unlike GCAP1, the GCAP5 dimer contains a single Fe2<sup>+</sup> bound at the dimer interface that is ligated by Cys15 and Cys17 in both dimer subunits (colored yellow and cyan in **Figure 2C**) of the symmetric GCAP5 dimer. The four cysteinyl thiolate groups that ligate the bound Fe2<sup>+</sup> (**Figure 2C**) are similar in structure to the four Cys residues found in the Cys4 zinc finger motif that binds to Zn2<sup>+</sup> (Tang et al., 2014). The structural similarity to the Cys4 zinc finger suggests that GCAP5 may also bind to Zn2<sup>+</sup> in place of Fe2+. Zn2<sup>+</sup> is transported into retinal photoreceptor cells and has been suggested to play a role in phototransduction (Redenti et al., 2007). It is tempting to speculate that GCAP5 might serve as a Zn2<sup>+</sup> sensor in the zebrafish photoreceptor. Future studies are needed to probe whether Zn2<sup>+</sup> can bind to GCAP5 and test whether Zn2<sup>+</sup> binding to GCAP5 (like Fe2<sup>+</sup> binding) can also regulate zebrafish RetGCs during visual phototransduction.

#### Ca2+-Induced Dimerization of Ca2+-Myristoyl Switch Proteins

The Ca2+-myristoyl switch proteins, recoverin (Myers et al., 2013) and VILIP1 (Li et al., 2011) both exhibit Ca2+-induced dimerization that enhances their membrane anchoring. The dimeric structure of Ca2+-bound recoverin (**Figure 2D**) places both of its exposed N-terminal myristoyl groups (highlighted magenta in **Figure 2D**) pointing in the same direction to serve as a dual pronged myristate membrane anchor (**Figure 4**). The recoverin dimer is stabilized mostly by hydrophobic intermolecular contacts (**Figure 3C**). In essence, the dimer

and lipid bilayer in blue.

interface is formed by the exposed hydrophobic residues in the exiting helix of EF4 (residues I173 and L177) that fit snuggly into the exposed hydrophobic groove between EF1 and EF2. Aliphatic side chain atoms of I173 and L177 from EF4 (yellow chain in **Figure 3C**) make intermolecular contacts with aromatic side chain atoms of F23 and W31 (cyan chain in **Figure 3C**). Additional intermolecular contacts are formed by side chain atoms of L90 and L177. The N-terminal domain residues (F23, W31, V87 and L90) at the dimer interface are the same residues that interact with the sequestered myristoyl group in Ca2+-free recoverin (Tanaka et al., 1995; Ames et al., 1997; Ames and Lim, 2012). The Ca2+-induced extrusion of the myristoyl group causes the exposure of these residues (F23, W31, V87 and L90), making them accessible to promote dimerization of the Ca2+-bound protein. The Ca2+-induced dimerization of recoverin enhances membrane binding by creating a dual pronged myristoyl anchor (**Figure 4**). The membrane anchored recoverin dimer bound to two rhodopsin kinase molecules in the dark may serve to place the two kinase molecules in close proximity of dimeric rhodopsin and therefore facilitate their rapid binding upon light activation (Ames et al., 2006; Myers et al., 2013).

The dimeric structure of VILIP-1 contains two protein subunits attached at their exposed C-terminal ends, forming an elongated structure (see yellow and cyan chains in **Figure 2D**). The exposed helices of EF4 are packed against each other at the dimer interface, forming an intermolecular four helix bundle. The VILIP1 dimer interface is almost entirely hydrophobic (**Figure 3D**). Exposed residues on the entering helix of EF4 (V151 and I154) in the yellow chain of **Figure 2D** make intermolecular contacts with exposed hydrophobic residues (F155 and F171) in the cyan colored chain. Additional intermolecular hydrophobic contacts are formed by I136 and M140 located in the region between EF3 and EF4. The intermolecular hydrophobic contacts are essential for VILIP1 dimerization as demonstrated by mutants (I136G, V151G and F155G) that each weaken the dimerization affinity (Li et al., 2011). The VILIP1 dimer structure has its N-terminal myristoyl group from each dimer subunit pointing upward in the same direction to serve as a dual-pronged anchor for targeting VILIP-1 to membranes (**Figure 4**). The opposite end of the VILIP1 dimer contains an exposed hydrophobic crevice in the N-terminal domain (residues F22, W30, L43, F48, I51, Y52, F55, F56, F72, F82, F85, I86, A88, L89) that are suggested to interact with hydrophobic segments of target proteins (Li et al., 2011).

#### FUNCTIONAL IMPLICATIONS OF NCS DIMERIZATION

#### Recoverin and VILIP1 Dimers Enhance Membrane Targeting Affinity

The Ca2+-bound dimeric structures of recoverin (**Figure 2C**) and VILIP1 (**Figure 2D**) have their N-terminal myristoyl groups pointing in the same direction toward the membrane surface (**Figure 4**). The juxta positioning of the two myristoyl groups creates a dual pronged membrane anchor that entropically enhances membrane binding. Since the effect of dimerization is multiplicative, a dimeric myristoyl switch protein is predicted to bind to lipid bilayer membranes with 10<sup>4</sup> -fold higher affinity (K<sup>d</sup> = 10–8 M) compared to the affinity of a monomeric Ca2+ myristoyl switch (K<sup>d</sup> = 10–4 M; Peitzsch and McLaughlin, 1993; Kim et al., 1994; Dell'Orco et al., 2012). In other words, the membrane binding dissociation constant of the dimer (Kd(dimer)) is equal to the square of the dissociation constant of the monomer: Kd(dimer) = Kd(monomer)<sup>2</sup> = (10–4 M)<sup>2</sup> = 10–8 M. Thus, the membrane binding affinity of Ca2+ myristoyl switch proteins is predicted here to be dramatically enhanced by the combined effect of both Ca2+-binding and protein dimerization. Dimerization of Ca2+-myristoyl switch proteins may also entropically enhance its binding to dimeric protein targets, as was suggested for the binding of dimeric Ca2+-bound recoverin to dimeric rhodopsin (Myers et al., 2013).

#### GCAP1 Dimerization in Retinal Photoreceptors

An important unresolved question in visual phototransduction is how the Ca2+-free GCAP proteins are able to specifically bind and activate RetGC in light-adapted photoreceptors, and conversely how Ca2+-bound GCAPs inhibit RetGC in dark-adapted photoreceptors. The crystal structure of the Ca2+-bound GCAP1 inhibitory state (Stephen et al., 2007) is overall similar to the recent NMR structure of the Ca2+ free GCAP1 activator state (Lim et al., 2016). Although, the Ca2+-induced changes in tertiary structure for GCAP1 appear moderately small, these small tertiary structural changes may promote a functional change in the quaternary structure of the GCAP1 dimer that in turn could modulate the quaternary structure of the RetGC1 dimer in order to allosterically regulate cyclase activity. In other words, small changes in tertiary structure may result in much larger changes in quaternary structure in order to amplify the response. Consistent with this prediction, mutations in GCAP1 (H19E, Y22E, F73E, V77E and W94E) that each weaken dimerization also abolish activation of the cyclase (Lim et al., 2018). These mutants indicate that GCAP1 dimerization is necessary and sufficient to activate RetGC, and furthermore suggests that a pre-formed GCAP1 dimer may facilitate its binding to the dimeric RetGC and thus stabilize a high affinity 2:2 target complex.

An alternative view is that the GCAP1 dimer that has been detected in solution and in the absence of RetGC may not necessarily exist in the presence of RetGC, because residues in the GCAP1 dimer interface (**Figure 3A**) appear to overlap with residues that interact with RetGC (Peshenko et al., 2014). In this scenario, the residues at the GCAP1 dimerization site may prefer to interact with RetGC (rather than itself) in the presence of saturating RetGC, and the binding of RetGC to GCAP1 in this case would be expected to prevent GCAP1 dimerization. To distinguish whether GCAP1 dimerization facilitates or opposes RetGC binding, future studies are needed to probe whether or not the structure of the GCAP1 dimer (**Figure 2A**) will remain intact when GCAP1 is bound to RetGC.

#### CONCLUSION

The dimerization of NCS proteins could help explain how these highly conserved proteins adopt distinctive structures that recognize many different targets. Recent structures of dimeric forms of GCAP1, GCAP2, GCAP5, recoverin and VILIP1 each reveal a unique quaternary structure at the dimer interface. GCAP1 forms a symmetric dimer that consolidates key residues for interacting with RetGCs, whereas GCAP2 forms an asymmetric dimer. The dimerization of GCAPs may facilitate allosteric regulation of its dimeric target protein (RetGC), which

#### REFERENCES


may help explain the steep Ca2+-dependent regulation of RetGC. Dimerization of both recoverin and VILIP1 creates a dual pronged myristate membrane anchor that enhances membrane targeting and may facilitate recognition of dimeric membranebound targets.

#### AUTHOR CONTRIBUTIONS

JA wrote and conceived the entire manuscript.

#### FUNDING

This work was supported by grants to JA (EY012347) from the National Institutes of Health (NIH), and Eye Institute.

protein VILIP-1 in vitro and in vivo: implications for mGlu receptor-dependent hippocampal plasticity? Neuropharmacology 44, 707–715. doi: 10.1016/s0028- 3908(03)00051-0


with cancer-associated retinopathy. J. Cell Biol. 112, 981–989. doi: 10.1083/jcb. 112.5.981


**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 © 2018 Ames. 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) and the copyright owner(s) 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.

# Human Calmodulin Mutations

Helene H. Jensen<sup>1</sup> , Malene Brohus<sup>1</sup> , Mette Nyegaard<sup>2</sup> and Michael T. Overgaard<sup>1</sup> \*

<sup>1</sup> Section for Biotechnology, Department of Chemistry and Bioscience, Aalborg University, Aalborg, Denmark, <sup>2</sup> Department of Biomedicine, Aarhus University, Aarhus, Denmark

Fluxes of calcium (Ca2+) across cell membranes enable fast cellular responses. Calmodulin (CaM) senses local changes in Ca2<sup>+</sup> concentration and relays the information to numerous interaction partners. The critical role of accurate Ca2<sup>+</sup> signaling on cellular function is underscored by the fact that there are three independent CaM genes (CALM1-3) in the human genome. All three genes are functional and encode the exact same CaM protein. Moreover, CaM has a completely conserved amino acid sequence across all vertebrates. Given this degree of conservation, it was long thought that mutations in CaM were incompatible with life. It was therefore a big surprise when the first CaM mutations in humans were identified six years ago. Today, more than a dozen human CaM missense mutations have been described, all found in patients with severe cardiac arrhythmias. Biochemical studies have demonstrated differential effects on Ca2<sup>+</sup> binding affinities for these CaM variants. Moreover, CaM regulation of central cardiac ion channels is impaired, including the voltage-gated Ca2<sup>+</sup> channel, CaV1.2, and the sarcoplasmic reticulum Ca2<sup>+</sup> release channel, ryanodine receptor isoform 2, RyR2. Currently, no non-cardiac phenotypes have been described for CaM variant carriers. However, sequencing of large human cohorts reveals a cumulative frequency of additional rare CaM mutations that raise the possibility of CaM variants not exclusively causing severe cardiac arrhythmias. Here, we provide an overview of the identified CaM variants and their known consequences for target regulation and cardiac disease phenotype. We discuss experimental data, patient genotypes and phenotypes as well as which questions remain open to understand this complexity.

#### Edited by:

Daniele Dell'Orco, Università degli Studi di Verona, Italy

#### Reviewed by:

David James O'Connell, University College Dublin, Ireland Stephan C. F. Neuhauss, Universität Zürich, Switzerland

#### \*Correspondence:

Michael T. Overgaard mto@bio.aau.dk

Received: 02 September 2018 Accepted: 11 October 2018 Published: 13 November 2018

#### Citation:

Jensen HH, Brohus M, Nyegaard M and Overgaard MT (2018) Human Calmodulin Mutations. Front. Mol. Neurosci. 11:396. doi: 10.3389/fnmol.2018.00396 Keywords: calmodulin, cardiac arrhythmia, calmodulinopathy, CPVT, LQTS, CALM1, CALM2, CALM3

#### INTRODUCTION

For generations, a large Swedish family presented with repeated episodes of syncope and cardiac arrest in response to exercise or emotional stress. Several family members were diagnosed with the inherited disorder catecholaminergic polymorphic ventricular tachycardia (CPVT), which is often fatal due to a high risk of ventricular fibrillation and sudden cardiac death (SCD). Indeed, two of the 13 affected individuals died from SCD, both at a young age. In 2012, we linked the disease to a mutation in the gene CALM1, which encodes the calcium (Ca2+) sensor calmodulin (CaM) (Nyegaard et al., 2012). The identification of a human CaM missense mutation came as a dramatic surprise to the CaM research field; CaM is exceptionally conserved across species with all vertebrate

**Abbreviations:** AP, action potential; CaM, calmodulin; CaV, voltage-gated calcium channel; CDI, calcium-dependent inactivation; CPVT, catecholaminergic polymorphic ventricular tachycardia; IVF, idiopathic ventricular fibrillation; LQTS, long QT syndrome; RyR2, ryanodine receptor isoform 2; SCD, sudden cardiac death; SR, sarcoplasmic reticulum.

CALM genes encoding identical proteins, and human mutations had not previously been reported. The slow evolution of CaM emphasizes the strong selection pressure against even minor changes in the protein sequence (Halling et al., 2016). Further, CaM regulates more than 300 intracellular targets, each interaction with unique facets of binding sites, Ca2+-dependency, target affinity, and functionality (Shen et al., 2005; O'Connell et al., 2010). With this versatility in mind, it was believed that mutations in CaM could not be tolerated.

After our initial finding, a number of CaM mutations have been identified in patients with severe cardiac arrhythmia disorders involving recurrent syncope, ventricular fibrillation, and in some instances SCD under adrenergic stimulation (**Table 1**). The vast majority of these mutations are de novo and carriers present with disease phenotypes early or very early in childhood, in some cases even before birth. In addition to CPVT, carriers suffer from long QT syndrome (LQTS), and one individual was diagnosed with idiopathic ventricular fibrillation (IVF). The link between CaM mutations and these arrhythmias has primarily been attributed to impaired regulation of the cardiac ryanodine receptor isoform 2 (RyR2), and the cardiac L-type voltage gated Ca2<sup>+</sup> channel isoform 1.2 (CaV1.2) (**Table 1**).

Since CaM is encoded by three active genes and expressed in all cells, the CaM field is faced with intriguing questions and paradoxes at the genetic and phenotypic level. First, how can a single mutation in one of six CaM-encoding alleles dominantly cause SCD? Second, how can identical missense mutations cause LQTS in one patient and CPVT in another? Third, with an increasing number of new rare CaM missense mutations identified in sequencing databases of large human cohorts, could there be other phenotypes associated with CaM mutations? Improved understanding of the functional impact of CaM mutations may enable predictions of the genotype–phenotype relationship for variants in any of the three CALM genes.

In this review, we summarize and discuss the current knowledge on CaM mutations and their impact on the regulation of CaV1.2 and RyR2, and address the few studies that suggest an involvement of other targets. Finally, we discuss the special genetic context of CaM and the implications for future studies.

# CaM, THE CELLULAR Ca2<sup>+</sup> SENSING PROTEIN

Fast and compound changes in cytosolic Ca2<sup>+</sup> concentration is the foundation for a wide number of cellular responses, including muscle contraction and neuronal firing (Clapham, 2007). Thus, at rest, the cytosolic Ca2<sup>+</sup> concentration is maintained at ∼100 nM, but can rapidly increase to more than 100 µM, when Ca2<sup>+</sup> channels open in the plasma membrane or in internal stores such as the sarcoplasmic reticulum (SR). Detection of this steep change in Ca2<sup>+</sup> concentration depends on Ca2<sup>+</sup> binding proteins. CaM is one of the major Ca2<sup>+</sup> sensors that relay information on Ca2<sup>+</sup> concentration to functionally modulate target proteins (known as calmodulation). CaM is synthesized as a 149 amino acid protein, however, the initiator Met residue is removed upon translation, leaving 148 amino acid residues in the mature protein (Sasagawa et al., 1982). This has led to some confusion in the numbering of CaM variant positions. The Human Genome Variant Sequence (HGVS) nomenclature (den Dunnen et al., 2016) recommends to count the initiator Met as residue number 1, while the CaM protein community tends to leave the residue out as it has no functional role in the mature protein. Throughout this paper we will use the HGVS nomenclature.

The structure of CaM reflects its refined Ca2<sup>+</sup> sensing abilities (Kretsinger et al., 1986; Chattopadhyaya et al., 1992). CaM is a 16.7 kDa protein consisting of two lobes connected by a flexible and unstructured or α-helical linker (**Figures 1A,B**). Each lobe has two EF-hands, which can each coordinate one Ca2<sup>+</sup> ion (**Figures 1A,B**, gray spheres). The C-terminal lobe of CaM binds Ca2<sup>+</sup> with six times higher affinity (K<sup>D</sup> 2.5 µM) than the N-terminal lobe (K<sup>D</sup> 16 µM), allowing CaM to sense Ca2<sup>+</sup> across a wide concentration range (Linse et al., 1991; Søndergaard et al., 2015a). Hydrophobic patches on the inside of each lobe recognize binding motifs on interaction partners, and thereby facilitate CaM binding and target regulation (Tidow and Nissen, 2013). Ca2<sup>+</sup> binding to CaM and CaM binding to target proteins allosterically affect the affinity of each other, and targets specifically modulate the conformation of CaM. In this way, the small CaM protein displays a plethora of binding and regulation properties.

#### CaM CONSERVATION AND MUTATIONS

Despite the remarkable conservation of CaM, 26 cases of arrhythmogenic mutations have now been identified in humans. Their positions in CaM are indicated on a CaM structure in **Figure 1A** and on the CaM sequence in **Figure 1C**. Strikingly, the mutations are primarily found in the C-terminal lobe and most affect residues involved in Ca2<sup>+</sup> coordination, dramatically reducing Ca2<sup>+</sup> affinity (**Table 1**). One interesting exemption is the mutation initially identified in the large Swedish family, CaM-N54I. This mutation is unique since it (1) resides in the N-terminal lobe and (2) neither coordinates Ca2<sup>+</sup> nor is part of the hydrophobic target binding patches. Biochemical and cellular experiments have been employed to model and explain how CaM mutations lead to arrhythmic phenotypes. The results from these studies are discussed in the following.

#### CaM Mutations Disturb Heart Rhythm

The composite effect of CaM mutations on heart function has been investigated using different experimental model systems. In zebrafish, the CaM-N54I and -N98S mutations caused increased heart rate upon β-adrenergic stimulation, which is in line with the CPVT phenotype observed for patients with these mutations (Søndergaard et al., 2015a). Similarly, the LQTS-mutation D130G increased zebrafish heart rate (Berchtold et al., 2016). In cultures of ventricular cardiomyocytes, expression of LQTS-associated CaM mutations leads to prolonged action potentials (APs), in some cases spilling over to the next stimulation and causing alternans (Limpitikul et al., 2014, 2017;


Frontiers in Molecular Neuroscience | www.frontiersin.org


Gomez-Hurtado et al., 2016; Yamamoto et al., 2017). Moreover, CaM-D96V was associated with early after-depolarizations (Gomez-Hurtado et al., 2016). Imaging of Ca2<sup>+</sup> fluxes during electrical pacing of cardiomyocytes demonstrated dysregulated Ca2<sup>+</sup> concentration in cells expressing LQTS-associated CaM-D96V, -D130G, -F142L, and in particular the CPVT-mutation N54I showed Ca2<sup>+</sup> overload, Ca2<sup>+</sup> reuptake errors, or alternans (Limpitikul et al., 2014, 2017; Yin et al., 2014).

CaV1.2 and RyR2 are the two major Ca2<sup>+</sup> channels involved in Ca2<sup>+</sup> handling in the heart. Both channels are essential for stimulating – and importantly, terminating – heart contraction. Briefly, APs stimulate the opening of CaV1.2 in the sarcolemma. This allows Ca2<sup>+</sup> to enter the cell. The resulting increase in cytosolic Ca2<sup>+</sup> stimulates opening of RyR2, thereby releasing large amounts of Ca2<sup>+</sup> from the SR which ultimately leads to heart contraction (Sorensen et al., 2013). CaM binds to both CaV1.2 and RyR2 and is important for the precise and timely gating of the channels in response to changes in Ca2<sup>+</sup> concentration. Generally, CPVT is an SR Ca2<sup>+</sup> handling disease, most often caused by RyR2 mutations, whereas LQTS involves dysregulation of ion-fluxes across the sarcolemma, e.g., Ca2<sup>+</sup> flux disturbances caused by mutations in CaV1.2 (Landstrom et al., 2017).

#### Impaired Regulation of CaV1.2

CaM is tethered to the intracellular C-terminal tail of CaV1.2 and functions as a Ca2<sup>+</sup> sensor to stimulate channel closure when Ca2<sup>+</sup> has entered the cell, a process termed calcium-dependent inactivation (CDI) (Brehm and Eckert, 1978; Peterson et al., 1999). Reduction of CDI was observed for several LQTS-causing CaM mutations, including D132H, D132V, and E141G (Boczek et al., 2016; Pipilas et al., 2016), whereas CDI was completely absent for LQTS-associated CaM mutations D96V, D130G, and F142L in HEK293 cells and adult guinea pig cardiomyocytes (Limpitikul et al., 2014). The CaM-N98S mutation is special in the sense that carriers display either CPVT or LQTS arrhythmias or both. Expression of this mutant slightly reduces CDI of CaV1.2, whereas the strictly CPVT-causing mutation CaM-N54I showed no effect (Limpitikul et al., 2014; Yamamoto et al., 2017).

#### Effects of CaM Mutations on RyR2

RyR2 opening is normally stimulated by the increase in Ca2<sup>+</sup> upon CaV1.2 opening. In response to the dramatic increase in cytoplasmic Ca2<sup>+</sup> concentration (and the drop in SR luminal Ca2<sup>+</sup> concentration), RyR2 closes. CaM acts as a gatekeeper, by modulating the open state probability of RyR2 in response to changes in the Ca2<sup>+</sup> concentration (Fabiato, 1985; Xu and Meissner, 2004). Here, we have compiled the diverse experimental approaches used to evaluate whether CaM mutations affect RyR2 regulation under different Ca2<sup>+</sup> concentrations (**Table 1**). The results are mixed and in some cases contradictory. The CPVT-associated CaM mutations, N54I, N98S, and A103V, all showed an increased level of RyR2 opening, that is, decreased inhibition by CaM (Hwang et al., 2014; Søndergaard et al., 2015b; Gomez-Hurtado et al., 2016). These results suggest that the molecular disease mechanism for

fnmol-11-00396 November 10, 2018 Time: 13:42 # 4

intermediate

receptor isoform 2.

iBoczek et al. (2016),

(2014),

rYin et al. (2014),

sLimpitikul et al. (2014),

aNyegaard

jReed et al. (2015),

 et al. (2012),

 (1 µM), or high (10–200 µM) free Ca

2+

bMarsman

kPipilas et al. (2016), tYamamoto

 et al. (2017),

 et al. (2014),

concentrations.

∗Under cChaix et al. (2016),

lSøndergaard

 et al. (2015a), uLimpitikul et al. (2017),

β-adrenergic

 stimulation. dCrotti et al. (2013),

mSøndergaard

 et al. (2015b), vRocchetti et al. (2017),

nSøndergaard

wBerchtold

 et al. (2016).

 et al. (2017),

 CaV1.2, voltage-gated eMakita et al. (2014),

2+

channel isoform 1.2; HGVS, human genome variation society; RyR2, ryanodine

fJiménez-Jáimez

 et al. (2016), oNomikos et al. (2014),

gGomez-Hurtado

 et al. (2016),

pVassilakopoulou

 et al. (2015),

hTakahashi

 et al. (2017), qHwang et al.

 Ca

the CPVT-causing CaM mutations is dysregulation of cardiac SR Ca2<sup>+</sup> release, in line with CPVT-causing mutations in RyR2.

Surprisingly, LQTS-associated CaM mutations also caused differences in RyR2 binding and regulation (Hwang et al., 2014; Søndergaard et al., 2015b; Vassilakopoulou et al., 2015). However, in particular at low Ca2<sup>+</sup> concentrations, the data are somewhat contradictory (**Table 1**). Curiously, CaM-F142L showed an increased binding affinity toward RyR2 in the apo-form, as well as an increased inhibitory effect on RyR2 in some assays, an effect opposite of all other CaM mutants analyzed (Søndergaard et al., 2017).

#### Other Suggested and Potential Targets

Although mutations in CaV1.2 can lead to LQTS, most cases of LQTS can be attributed to mutations in genes encoding the voltage-gated K<sup>+</sup> channels KV7.1 (KCNQ1) and KV11.1 (KCNH2), as well as the voltage-gated Na<sup>+</sup> channel NaV1.5 (SCN5A) (Modell and Lehmann, 2006). Although regulated by CaM, no clear effects of CaM mutations have been observed on NaV1.5 (Yin et al., 2014; Boczek et al., 2016; Rocchetti et al., 2017). KV7.1 utilizes CaM as a sensor of Ca2<sup>+</sup> to stimulate opening (Ghosh et al., 2006). Whereas CaM-F142L did not show any effect on KV7.1 current (IKs) (Rocchetti et al., 2017), CaM-N98S significantly shifted the half-activation of KV7.1 (Sun and MacKinnon, 2017). Further, the small-conductance Ca2+-activated K<sup>+</sup> (SK) channel was decreased by several CaM mutations (Yu et al., 2016). Although SK channels play a minor role in ventricular myocytes, they are expressed in atrial myocytes, and interestingly, widely expressed in the nervous system where they play a major role in synaptic transmission (Adelman et al., 2012).

Ca2+/CaM-dependent kinase II (CaMKII) regulates a wide number of pathways and protein targets, but no significant effects were observed on CaMKII with either of the CaM variants N54I, D96V, N98S, D130G, or F142L (Hwang et al., 2014; Berchtold et al., 2016).

Although current literature suggests that Ca2<sup>+</sup> channels are the main targets affected by CaM mutations, we hypothesize that other ion channels and potentially other signaling proteins may also be dysregulated if tested in greater detail. But how can one predict which targets are likely affected by CaM mutations? We believe that the Ca2+-dependency of target binding plays an important role. That is, proteins that bind to both the apo- and the Ca2+-form of CaM may be more sensitive to mutations in CaM than targets that only bind the Ca2+-form. Thus, CaM mutations may exert a dominant effect in cases where CaM remains associated with its target when the cell is at rest.

#### IMPLICATIONS OF THE GENETIC ARCHITECTURE OF CALM GENES

In humans, CaM is encoded by three different and independent loci; on chromosome 2 (CALM2), 14 (CALM1), and 19 (CALM3). Although there are differences in the genomic sequence, the three different transcripts are translated into the exact same protein (Fischer et al., 1988). During the last six years (2012–2018), 26 cases of pathogenic mutations in CaM have been reported, and all three CALM genes are now established major genes for both CPVT and LQTS (**Table 1**). All identified pathogenic CaM mutations cluster in the C-lobe, except the CPVT-causing variant N54I. Interestingly, this is the mutation with the mildest effect on biophysical parameters of CaM, including Ca2<sup>+</sup> binding affinity. It is also the only variant found in a large family. Further, all LQTS-causing CaM mutations strongly reduce the C-lobe Ca2<sup>+</sup> affinity, and, except for the F142L mutation, are all located in Ca2+-coordinating residues. This suggests that CaM mutations that strongly affect C-lobe Ca2<sup>+</sup> affinity lead to LQTS (**Table 1**).

To date, there are more published cases of pathogenic mutations in CALM1 (11) and CALM2 (11) compared to CALM3 (4). This could either reflect that the first published mutations were found in CALM1 and CALM2 and thus these two genes were included in genetic screening panels before CALM3. Or, it may be due to some subtle functional differences between the three genes. Quantitative PCR have shown that the three genes are not expressed at equal levels in cardiomyocytes, but the relative levels are not clear: whereas one study found higher levels of CALM3 transcripts in human hearts (Crotti et al., 2013), the CALM1 transcript was the most abundant in human stem cell-derived cardiomyocytes (Rocchetti et al., 2017).

As the number of published pathogenic CaM mutations has increased, several conclusions about genotype–phenotype relationships begin to form. Of particular interest is the CaM-D130G mutation, which has been identified in four unrelated individuals; two carrying the mutation in CALM1, one in CALM2, and one in CALM3, and all four suffering from LQTS. Similarly, the CaM-F142L mutation was found in both CALM1 and CALM3, and all carriers suffered from LQTS. These observations imply that the amino acid position and type of change is important for the phenotype, and not the genetic origin of the transcript (CALM1, −2, or −3). One intriguing observation does, however, challenge this simple genotype– phenotype conclusion. The CaM-N98S mutation was found in CALM1 in one individual, and in three other individuals in CALM2. Interestingly, these four patients present with different phenotypes – either CPVT or LQTS or both – suggesting that we still do not fully understand the underlying mechanisms determining the disease phenotype. These may involve other genomic variants able to shape the phenotype, complex protein regulatory effects, or environmental factors.

Since the protein products from all CALM genes are identical, it is tempting to speculate if a deletion of one allele (equivalent to a loss-of-function mutation) is less pathogenic than missense mutations. This idea immediately poses a therapeutic solution to patients carrying a CaM missense mutation, for example using the CRISPR/Cas9 technology to delete the pathogenic allele. Two studies specifically silenced the mutated CaM allele in patient-derived pluripotent stem cells differentiated into cardiomyocytes. Here, the CaM-D130G and - N98S mutations were silenced in CALM2 with a partial or almost full restoration of CaV1.2 regulation (Limpitikul et al., 2017;

Yamamoto et al., 2017). These experiments are proof-of-principle that removal of the diseased allele may be a therapeutic solution. Also, these studies suggest that potential frameshift mutations causing premature stop codons or protein degradation may not be as detrimental as missense mutations. It still needs to be determined, however, if individuals carrying a loss-of-function mutation in a CALM gene are in fact unaffected from disease. Interestingly, a large exome sequencing study [Exome Aggregate Consortium (ExAC)] (Lek et al., 2016), found that CALM1 and CALM2 are intolerant to loss-of-function mutations (pLI = 0.89 and 0.86 respectively).

#### DIGGING DEEPER MAY REVEAL A BROADER IMPACT

Given the ubiquitous role of CaM, it is striking that all mutations identified are associated with a strong cardiac phenotype. We speculate whether these cases were discovered because of their unusual severity and because cardiologists and geneticists have specifically screened for CALM mutations in populations with cardiac disorders. Looking for CaM mutations in other patient groups may reveal new aspects and consequences of these mutations.

In a database containing variants from a large sequencing effort of almost 140,000 individuals (GnomAD, Lek et al., 2016), additional rare CaM missense mutations are reported (**Figures 1B,C**, gray residues and circles). The number of coding variants for all three genes is much lower than expected by chance. However, the cumulative frequency of additional rare CaM mutations suggests that CaM variants do not exclusively cause severe cardiac arrhythmias. At present, there is no overlap between variants identified in GnomAD (database variants) and the published pathogenic mutations. Further, the GnomAD missense variants are distributed throughout the entire protein, and more evenly distributed on the three CALM genes (9, 9, and 12 mutations in CALM1, −2, and −3, respectively), compared to the arrhythmogenic CaM variants. Also, all GnomAD variants except two, fall outside Ca2+-coordinating residues. Taken together, we therefore speculate that some of these uncharacterized variants are associated with unknown traits not involving cardiac arrhythmia. Sequencing results from large cohorts with known phenotypes are required to confirm this hypothesis.

We propose that studies of tissues other than cardiac are warranted for future research on the effects of CaM mutations. In particular, CaM expression is high in excitable neuronal cells. Also CaV1.2 is widely expressed in neuronal tissues. Here, CaV1.2 plays a role in cellular firing as well as in gene regulation, and

#### REFERENCES


mutations in CaV1.2 have been attributed to psychiatric diseases (Nyegaard et al., 2010; Nanou and Catterall, 2018). Neurons express a number of other Ca<sup>V</sup> channel isoforms, including CaV1.3 and CaV2 variants, which are also regulated by CaM. RyR2 plays a less prominent role in neurons, where the inositol triphosphate receptor (IP3R), which is also regulated by CaM, is the dominating intracellular Ca2<sup>+</sup> release channel. Mild neuronal defects have been observed in some patients with CaM mutations, but these effects were suggested to be secondary, resulting from the frequent and severe episodes of syncope or cardiac arrest (Crotti et al., 2013; Boczek et al., 2016; Pipilas et al., 2016).

# CONCLUSION AND OUTLOOK

Within the last six years, CaM mutations have emerged as a novel cause of human diseases, the calmodulinopathies. All described pathogenic mutations have been identified in patients suffering from severe arrhythmic disorders, and biochemical as well as cellular studies have demonstrated that particularly the regulation of the Ca2<sup>+</sup> channels CaV1.2 and RyR2 are affected by these mutations. Currently, there is a strong correlation between LQTS-causing CaM mutations and Cav1.2 dysregulation, whereas all mutations affect RyR2 function. Given the ubiquitous role of CaM in a vast number of cellular processes, we predict that yet other targets may be affected. Our database search revealed a number of uncharacterized CaM missense mutations with unknown phenotypic consequences present in the population. Future studies will reveal whether other protein targets as well as other disease phenotypes can be assigned to mutations in CaM.

#### AUTHOR CONTRIBUTIONS

HJ wrote the first draft of the manuscript. HJ, MB, MN, and MO contributed to the content and writing. HJ and MB prepared the figure and table. All authors have read and approved the manuscript.

#### FUNDING

This study was supported by research grants from the Novo Nordic Foundation (NNF15OC0012345 and NNF16OC0023344), the Lundbeck Foundation (R151-2013- 14432), and the Danish Council for Independent Research (DFF-4181-00447) to MO, and by a postdoctoral fellowship from the Lundbeck Foundation (R2017-134) to HJ.

cell growth, calmodulin-dependent kinase II activity, and cardiac function in zebrafish. J. Biol. Chem. 291, 26636–26646. doi: 10.1074/jbc.M116. 758680

Boczek, N. J., Gomez-Hurtado, N., Ye, D., Calvert, M. L., Tester, D. J., Kryshtal, D. O., et al. (2016). Spectrum and prevalence of CALM1-, CALM2- , and CALM3-encoded calmodulin variants in long QT syndrome and functional characterization of a novel long QT syndrome-associated calmodulin

missense variant, E141G. Circ. Cardiovasc. Genet. 9, 136–146. doi: 10.1161/ CIRCGENETICS.115.001323


arrhythmia susceptibility. Circ. Cardiovasc. Genet. 7, 466–474. doi: 10.1161/ CIRCGENETICS.113.000459



cardiomyocyte Ca2+ regulation by distinct mechanisms. J. Am. Heart Assoc. 3:e000996. doi: 10.1161/JAHA.114.000996

Yu, C. C., Ko, J. S., Ai, T., Tsai, W. C., Chen, Z., Rubart, M., et al. (2016). Arrhythmogenic calmodulin mutations impede activation of smallconductance calcium-activated potassium current. Hear. Rhythm 13, 1716–1723. doi: 10.1016/j.hrthm.2016. 05.009

**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 © 2018 Jensen, Brohus, Nyegaard and Overgaard. 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) and the copyright owner(s) 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.

# Ca2+-Sensor Neurocalcin δ and Hormone ANF Modulate ANF-RGC Activity by Diverse Pathways: Role of the Signaling Helix Domain

Teresa Duda<sup>1</sup> \*, Alexandre Pertzev<sup>1</sup> , Sarangan Ravichandran<sup>2</sup> and Rameshwar K. Sharma<sup>1</sup>

<sup>1</sup> Research Divisions of Biochemistry and Molecular Biology, The Unit of Regulatory and Molecular Biology, Salus University, Elkins Park, PA, United States, <sup>2</sup> Advanced Biomedical Computational Sciences Group, Frederick National Laboratory for Cancer Research Sponsored by the National Cancer Institute, Leidos Biomedical Research Inc., Fredrick, MD, United States

Prototype member of the membrane guanylate cyclase family, ANF-RGC (Atrial Natriuretic Factor Receptor Guanylate Cyclase), is the physiological signal transducer of two most hypotensive hormones ANF and BNP, and of the intracellular free Ca2+. Both the hormonal and the Ca2+-modulated signals operate through a common second messenger, cyclic GMP; yet, their operational modes are divergent. The hormonal pathways originate at the extracellular domain of the guanylate cyclase; and through a cascade of structural changes in its successive domains activate the C-terminal catalytic domain (CCD). In contrast, the Ca2<sup>+</sup> signal operating via its sensor, myristoylated neurocalcin δ both originates and is translated directly at the CCD. Through a detailed sequential deletion and expression analyses, the present study examines the role of the signaling helix domain (SHD) in these two transduction pathways. SHD is a conserved 35-amino acid helical region of the guanylate cyclase, composed of five heptads, each meant to tune and transmit the hormonal signals to the CCD for their translation and generation of cyclic GMP. Its structure is homo-dimeric and the molecular docking analyses point out to the possibility of antiparallel arrangement of the helices. Contrary to the hormonal signaling, SHD has no role in regulation of the Ca2+- modulated pathway. The findings establish and define in molecular terms the presence of two distinct nonoverlapping transduction modes of ANF-RGC, and for the first time demonstrate how differently they operate, and, yet generate cyclic GMP utilizing common CCD machinery.

Keywords: membrane guanylate cyclase, cyclic GMP, atrial natriuretic factor receptor guanylate cyclase, neurocalcin delta, ANF

# INTRODUCTION

Kick started by historic discovery of the prototype ANF-RGC (Atrial Natriuretic Factor Receptor Guanylate Cyclase) (Paul, 1986; Paul et al., 1987), the membrane guanylate cyclase (MGC) field has expanded to constitute a family of seven - -ANF-RGC, CNP-RGC, STa-RGC, ROS-GC1, ROS-GC2, ONE-GC and GC-G [reviewed (including) in terminology: (Sharma et al., 2016)].

#### Edited by:

Karl-Wilhelm Koch, University of Oldenburg, Germany

#### Reviewed by:

Yogendra Sharma, Centre for Cellular and Molecular Biology (CSIR), India Evgeni Yurievich Zernii, Lomonosov Moscow State University, Russia

#### \*Correspondence:

Teresa Duda tduda@salus.edu; rsharma@salus.edu

Received: 30 August 2018 Accepted: 05 November 2018 Published: 27 November 2018

#### Citation:

Duda T, Pertzev A, Ravichandran S and Sharma RK (2018) Ca2+-Sensor Neurocalcin δ and Hormone ANF Modulate ANF-RGC Activity by Diverse Pathways: Role of the Signaling Helix Domain. Front. Mol. Neurosci. 11:430. doi: 10.3389/fnmol.2018.00430

By synthesizing cellular second messenger cyclic GMP the family is interlocked with the diverse physiological processes ranging from cardiac vasculature to cellular growth, sensory transductions, neural plasticity, temperature sensing and pineallinked brain functions (reviewed and Figure 10 in Sharma et al., 2016). These processes are compromised by natural mutations incurred in their key genes (discussed in Duda et al., 2018); a few examples are: CNP-RGC-R819C causes acromesomelic dysplasia (Vasques et al., 2013; Nakao et al., 2015); and more than 100 ROS-GC1 mutations result in retinal dystrophies, mainly Leber's congenital amaurosis (LCA1) and cone-rod degeneration (CORD6) (Perrault et al., 1996, 1999; Kelsell et al., 1998; Wilkie et al., 2000; Rozet et al., 2001; Hunt et al., 2010; Gradstein et al., 2016). A common molecular lesion of these abnormalities is loss of the basal catalytic activity to generate cyclic GMP (Duda et al., 1999a,b, 2000a; Wilkie et al., 2000).

The MGC signal transduction system is structurally and functionally different from the other two major signaling pathways, cyclic AMP and IP3, (Paul et al., 1987; Figure 2 in Sharma and Duda, 2014). Instead of three, it is a one component transduction system. A single transmembrane-spanning protein is crafted with a multi-modular design. This design allows it to be elastic, to exist in multiple forms, and to be multifunctional. MGC family consists of three subfamilies. (1) The original, Surface Receptor. It inherits a unique trait of being both a hormone receptor and a catalyst, guanylate cyclase. It contains three members - - ANF-RGC (Paul et al., 1987; Chinkers et al., 1989; Lowe et al., 1989; Pandey and Singh, 1990; Duda et al., 1991), the receptor for hormones ANF and BNP; CNP-RGC (Chang et al., 1989; Schulz et al., 1989; Duda et al., 1993), the receptor for hormone CNP; and STa-RGC (de Sauvage et al., 1991; Singh et al., 1991), the receptor for heat-stable enterotoxin (STa) and also for hormones guanylin and uroguanylin. (2) The [Ca2+]imodulated ROS-GC, with two members, ROS-GC1 (Margulis et al., 1993; Duda et al., 1994) and ROS-GC2 (Lowe et al., 1995; Goraczniak et al., 1997; reviewed in Sharma and Duda, 2012). Its unique characteristic is that as of yet the function of their extracellular domains is unknown and their activities are under control of intracellular Ca2<sup>+</sup> through specific calcium sensors. (3) The Ca2+-modulated and odorant receptor, with one member, ONE-GC (Duda et al., 2001, 2004; Hu et al., 2007; Leinders-Zufall et al., 2007; Duda and Sharma, 2008; Sharma and Duda, 2010). Its signature characteristic is that it is a quadric-modal transducer of: uroguanylin, [Ca2+]i-modulated myristoylated neurocalcin δ, GCAP1, and gaseous CO2. Notably, the gene encoding ONE-GC is rodent-specific, not present in the humans (Torrents et al., 2003; Caenepeel et al., 2004; Young et al., 2007).

Despite these distinctive features, the general topography of the subfamilies is similar. Each is composed of three domains: Extracellular (ExtD), Transmembrane (TMD) and Intracellular (ICD) (reviewed in Sharma et al., 2016). ICD is further subdivided into two vaguely defined domains, N-terminal KHD (Kinase Homology Domain) and C-terminal catalytic domain (CCD). Because KHD terminology was imprecise- - it has recently been redefined (discussed in Ravichandran et al., 2017; Duda et al., 2018). The 43-residue α-helical region located at its C-terminus is no longer considered to be a part of it but is now recognized as an individual module, termed the signaling helix domain (SHD) (Anantharaman et al., 2006). Its structure is conserved in MGC family, and it serves as a critical transmission switch to the catalytic domain for the downstream signaling pathways (Venkataraman et al., 2008; Saha et al., 2009; Zagel et al., 2013; Ravichandran et al., 2017; Duda et al., 2018). With in-depth definitions of the CCD structural and functional boundaries (Ravichandran et al., 2017), a new signal transduction paradigm of the MGC family members has been defined. It negates the old notion that trajectory of all signals from their origin to the catalytic domain is only downstream (reviewed in Sharma et al., 2016). And, demonstrates, that with exception of the surface receptor subfamily, the signal trajectory in other subfamilies is bi-directional, downstream and upstream (Ravichandran et al., 2017; Duda et al., 2018). This happens because the ROS-GC and ONE-GC subfamilies contain an extra C-Terminal Extension (CTE) domain beyond its catalytic domain. This domain wedges the catalytic domain between SHD and CTE, and CTE attains an independent status of being a modular domain. Consequently, three modes of signal transduction pathways can, and, do occur- -(1) downstream, (2) upstream, (3) mainstream or direct (the signals originate and are translated at the catalytic domain); the last situation in the case of ANF-RGC (Duda et al., 2012a), as explained below.

When discovered, ANF-RGC was declared to be uniquely regulated by the hormone ANF (Paul, 1986; Paul et al., 1987) and its function was linked with the physiological processes of diuresis and blood pressure regulation (de Bold, 2011). Implicit in this conclusion was that ANF-RGC machinery is solely structured to generate and transduce the extracellular hormonal signal. This conclusion, however, was recently revised. In this revelation, ANF-RGC also transduced an intracellularly generated Ca2<sup>+</sup> signal; this signal is modulated through Ca2+-sensor myristoylated neurocalcin δ (myr-NCδ) and is physiologically linked with blood pressure regulation (Duda et al., 2012b). Aberration in this signaling pathway causes hypertension and cardiac myopathy in the mice (Duda et al., 2012b). Thus, ANF-RGC is a bimodal signal transducer, yet both modes have identical physiological outcome.

A unified structural theme of the MGC family is that it is homo-dimeric (reviewed in Sharma, 2010; Sharma et al., 2016). Biochemical details on MGC structure are limited, derived only from crystal structure of the extracellular domain of ANF-RGC (Ogawa et al., 2004) and from bimolecular fluorescence complementation (Duda et al., 2012c), modeling, and mutational analyses of the isolated catalytic module of ROS-GC1 (Liu et al., 1997; Venkataraman et al., 2008; Ravichandran et al., 2017). The crystal and single particle electron microscopy data demonstrate that the homo-dimeric structure of the extracellular domain is parallel (N-terminus to N-terminus and C-terminus to C-terminus) head-to-head (van den Akker et al., 2000; Ogawa et al., 2004; Ogawa et al., 2009). The bimolecular fluorescence complementation analysis shows that the monomers of the CCD form antiparallel dimers (Duda et al., 2012b) and the modeling analyses demonstrate that the two monomers assume a two-fold symmetry axis with a central gap of a wreath- or circlet-like shape (Liu et al., 1997; Rauch et al., 2008; Ravichandran et al., 2017).

dimer model that satisfies the Knobs-into-Holes mode of packing between alpha-helices – a unique feature of coiled-coils. Protein chains are displayed in solid ribbon. The interface residues, 3 Å from each chain, are identified with asterisk in the linear sequence. They are the same for both chains and are shown in stick form and labeled except for two residues, E<sup>797</sup> and A813, shown in stick-mode but are not labeled for clarity of the figure. For the same reason, only the interface residues from the top chain are labeled.

Notably, the plasma membrane anchoring is obligatory for the catalytic domain to manifest its full activity; without it, it is about one-order of magnitude lower (Ravichandran et al., 2017). Thus, the individual MGC modules do not show uniformity of existing in parallel or antiparallel orientation; and markedly, except for these two modules, status of the others is unknown.

Obligatory for ANF signaling of ANF-RGC catalytic activity is ATP (Chinkers et al., 1991; Marala et al., 1991). Mechanistically, ANF binding to the extracellular domain signals ATP binding to the intracellular ATP Regulated Domain (ARM), causing its allosteric modification and through a cascade of structural modifications involving <sup>669</sup>WTAPELL<sup>675</sup> motif activating the

catalytic domain and accelerating the production of cyclic GMP (Goraczniak et al., 1992; Burczynska et al., 2007; Duda et al., 2009; reviewed in Sharma, 2010).

All signals, including the downstream, upstream and direct are translated at the common catalytic center consisting of 7 residues from each monomer of the MGC existing in the form of a wreath or circlet (Liu et al., 1997; Ravichandran et al., 2017). In resting state, the wreath is in an open conformation but in response to a regulatory signal it acquires a closed conformation optimal for accelerated formation of cyclic GMP from GTP.

The present study was designed to deepen our knowledge on the basic principles that make the hormonal and the Ca2+ dependent mechanisms of ANF-RGC activation distinct. The study establishes a critical role of SHD in maintaining the basal catalytic activity of ANF-RGC and in its hormonal but not Ca2+-dependent modulation.

### MATERIALS AND METHODS

#### Construction of Heptad Deletion and Site-Specific Mutants

The ANF-RGC deletion and single substitution mutants were created by PCR using the cDNA of the wild type (wt) rat ANF-RGC (Duda et al., 1991) as template and the following mutagenic primers: for deletion of the first heptad (1H1 mutant): FW: 5<sup>0</sup> - AACATCCTGGACAACCTGGCTAACAACCTGGAGGAA-3<sup>0</sup> ; REV: 5<sup>0</sup> -TTCCTCCAGGTTGTTAGCCAGGGTTGTCCAGG ATGTT-3<sup>0</sup> ; for deletion of the second heptad (1H2 mutant), FW: 5 0 -TCAAGCATGGAGCAGTATGTAGAGGAGAGAACACAA-3 0 ; REV: 5<sup>0</sup> -TTGTGTTCTCTCCTCTACATACTGCTCCCATG CTTGA-3<sup>0</sup> ; for deletion of the third heptad (1H3 mutant), FW: 5 0 -AACAACCTGGAGGAACTGTATCTGGAGGAGAAGCGC-3 0 ; REV: 5<sup>0</sup> -GCGCTTCTCCTCCATACAGTTCCTCCAGGTT GTT-3<sup>0</sup> ; for deletion of the fourth heptad (1H4 mutant), FW: 5 0 -GAGGAGAGAACACAAGCTGCTGAGGCCTTGCTTTAC-3 0 ; REV: 5<sup>0</sup> -GTAAAGCAAGGCCTCAGCAGCTTGTGTTCTCT CCTC-3<sup>0</sup> ; for deletion of the fifth heptad (1H5 mutant), FW: 5<sup>0</sup> - CTGGAGGAGAAGCGCAAAATTCTGCCTCACTCCGTG-3<sup>0</sup> ; REV: 5<sup>0</sup> -CACGGAGTGAGGCAGAATTTGCGCTTCTCCTC CAG-3<sup>0</sup> ; and for substituting the R<sup>802</sup> residue with C (R802C mutant), FW: 5<sup>0</sup> -GAACTGGTAGAGGAGTGTACACAAGCT TATCTG-3<sup>0</sup> ; REV: 5<sup>0</sup> -CAGATAAGCTTGTGTACACTCCTCTA CCAGTTC-3<sup>0</sup> .

To delete the entire SHD, amino acid residues L785-Q<sup>819</sup> , from ANF-RGC (1SHD mutant) two HpaI restriction sites were introduced in ANF-RGC cDNA; the first site was between codons for L<sup>784</sup> and L785, the second between Q<sup>819</sup> and I820. The mutated ANF-RGC cDNA was digested with HpaI and re-ligated.

All mutations were verified by sequencing.

### Expression in a Heterologous Cell System

COS-7 cells were induced to express ANF-RGC or its mutants using Lipofectamine (Thermo Fisher Scientific). Seventy hours after transfection, the cells were harvested and their membranes prepared (Duda et al., 2018).

#### ANF-RGC Guanylate Cyclase Catalytic Activity Assays

Membrane samples of transfected COS cells were incubated individually without or with 10−<sup>7</sup> M ANF and varying concentrations of ATP or varying concentrations of recombinant myristoylated neurocalcin d (myrNCδ). MyrNCδ was purified as described in Duda et al. (2012b). The assay mixture (25 ml) consisted of (mM): 10 theophylline, 15 phosphocreatine, and 50 Tris-HCl; pH 7.5, and 20 µg creatine kinase (Sigma). In experiments with myr-NCδ, 1 µM Ca2<sup>+</sup> was present in the reaction mixture. The reaction was initiated by addition of the substrate solution 4 µM MgCl<sup>2</sup> and 1 µM GTP (final concentrations) and maintained by incubation at 37◦C for 10 min. In the experiments for [GTP] dependency of guanylate cyclase activity, the GTP concentrations varied from 0 to 2 µM and the MgCl<sup>2</sup> concentration was constant, 4 µM. The reaction was terminated by the addition of 225 ml of 50 µM sodium acetate buffer, pH 6.2, followed by heating on a boiling water bath for 3 min. The amount of cyclic GMP formed was determined by radioimmunoassay (Nambi et al., 1982). All assays were done in triplicate and except where stated otherwise, were performed three times. Guanylate cyclase activity is presented as average ± SD of these experiments. The EC<sup>50</sup> values were determined from the experimental values by GraphPad PRISM program. The effect of different conditions on ANF-RGC activity was evaluated by performing a one-way ANOVA; p-values ≤ 0.05 were considered to be significant.

To correlate the catalytic changes brought about by the mutations, the activities of the mutants were compared with wild type recombinant ANF-RGC through Michaelis plots for the ligand used by fitting the data with the Hill equation, v = Vmax <sup>X</sup> (S)<sup>n</sup> / (KM) <sup>n</sup> + (S)<sup>n</sup> . Vmax is the maximal activity, S is the concentration of the ligand, K<sup>M</sup> is the ligand concentration at which half-maximal velocity is achieved, and n is the Hill coefficient.

#### Western Blot

fnmol-11-00430 November 27, 2018 Time: 16:18 # 5

After boiling in a gel-loading buffer (62.5 µM Tris-HCl, pH 7.5, 2% SDS, 5% glycerol, 1 µM β-mercaptoethanol, and 0.005% bromophenol blue) 50 µg of membrane protein was subjected to SDS-polyacrylamide gel electrophoresis in a buffer (pH 8.3) containing 0.025 M Tris, 0.192 M glycine, and 0.1% SDS. The proteins were transferred to immobilon membranes (Millipore) in the same buffer but containing 5% methanol. The blot was incubated in Tris-buffered saline (TBS, pH 7.5) containing 100 µM Tris-HCl, 0.9% NaCl, and 0.05% Tween-20 (TBS-T) with 5% bovine serum albumin (blocking buffer) overnight at 4 ◦C and rinsed with TBS-T. The anti-ANF-RGC antibody (raised in rabbit; Santa Cruz Biotechnology, Inc.) was added at 1: 2000 dilution in the blocking buffer and the incubation was continued overnight at 4◦C. After the blot was rinsed with TBS-T, the incubation was continued with peroxidase-conjugated AffiniPure goat anti-rabbit IgG (Jackson ImmunoResearch Laboratories, Inc.) diluted 1: 50,000 for 1 h at room temperature. Finally, the blot was treated with SuperSignal West Dura Extended Duration Substrate (Thermo Scientific) for 5 min according to the manufacturer's protocol. The immunoreactive band was detected by exposing the blot to CL-XPosure film (Thermo Scientific).

#### ANF-RGC Catalytic Efficiency

The protocol is described in Duda et al. (2012b). Briefly, aliquots of 10 to 0.1 ng of the antigen for the ANF-RGC antibody (ANF-RGC fragment aa 486–661) were diluted in Laemmli sodium dodecyl sulfate sample buffer and loaded next to 50 µg (total protein) of COS cell membranes expressing ANF-RGC. After electrophoresis the proteins were transferred to immobilon membranes (Millipore) and immunostained with anti-ANF-RGC antibody as described above and exposed to X-ray film. The amount of ANF-RGC in COS cell membranes was quantified from the calibration curve produced by the antigen standards. COS cell membranes from the same transfection experiment were assayed for guanylate cyclase activity.

The catalytic efficiencies of the mutants were calculated by comparing their expression levels in COS cells and basal catalytic activities with wt-ANF-RGC.

#### Molecular Modeling

ANF-RGC (Gene: Npr1) is a homo-dimer protein, the amino acid sequence segment N783- P<sup>822</sup> was analyzed for the features characteristic of SHD domain. Note that the amino acid numbering shown above corresponds to the mature ANF-RGC protein (Duda et al., 1991) which is identical to the UniProt P18910-1 full-length native sequence segment, 811– 850. Due to lack of experimental structure for SHD domain, three-dimensional structural models were built using I-TASSER (Iterative Threading and ASSEmbly Refinement; web server version)<sup>1</sup> software. I-TASSER is an automated protein 3D structure prediction program. It employs a hierarchical protein modeling approach that uses secondary structural data, template identification, profile-profile threading and iterative threading assembly refinement steps to build 3D protein models. I-TASSER identified five unique 3D-structural templates (PDB IDs), 1zxaA, 3hlsA, 4pkyB, 5svaV, and 1y11 for the SHD domain. Structure summary of the templates indicated that they all belong to a coiled-coiled domain family. Based on the secondary structure predictions (**Supplementary Figure S1**) and the I-TASSER C-score, the top-ranked monomer was selected as the representative structure and hereafter referred to as the default model for SHD sequence.

The dimer models for the SHD signaling helix were built and analyzed using protein-protein docking programs, M-ZDOCK (Radially Symmetric Multimer Docking software)<sup>2</sup> and Z-DOCK<sup>3</sup> . The SOCKET software<sup>4</sup> was used to search for the most energy favorable packing structures from the top ten M-ZDOCK dimer models (the details are provided in the **Supplementary Figures S2, S3**).

#### RESULTS

#### Structure-Focused View of the SHD

According to the previously defined characteristics of the individual residues (Anantharaman et al., 2006), the ANF-RGC segment, N783- P<sup>822</sup> [mature protein numbering (Duda et al., 1991)], meets the criteria of being the SHD. To verify this, in absence of any crystal structure information, a three-dimensional model of the N783- P<sup>822</sup> region was built employing a hierarchical approach of structural template identification, threading and iterative template fragment assembly simulations. The secondary structure prediction (PSSpred)<sup>5</sup> and I-TASSER models demonstrate that this segment exists as a single α-helix (**Figure 1A** and **Supplementary Figure S1**).

Since ANF-RGC, like other membrane guanylate cyclases, exists as a homodimer with its catalytic domain in antiparallel, wreath-like orientation (Ravichandran et al., 2017), the question was: what is the configurational positioning of the monomers within the SHD dimer? Two molecular docking programs (M-ZDOCK<sup>2</sup> and Z-DOCK<sup>3</sup> ) were used for the analyses. M-ZDOCK (a symmetric dimer prediction software) (top-10) predicted an antiparallel arrangement (**Figure 1B** and **Supplementary Figure S3**). Up to 30 top ranked M-ZDOCK predicted conformations were analyzed and all of them showed anti-parallel orientations. We also repeated the modeling using a related docking software, Z-DOCK. Unlike with M-ZDOCK, we found some parallel docked conformations (4 out of top-10

<sup>1</sup>http://zhanglab.ccmb.med.umich.edu/I-TASSER/

<sup>2</sup>http://zdock.umassmed.edu/m-zdock

<sup>3</sup>http://zdock.umassmed.edu/

<sup>4</sup>http://coiledcoils.chm.bris.ac.uk/socket

<sup>5</sup>http://zhanglab.ccmb.med.umich.edu/PSSpred/

ranked complexes). Altogether, based on the modeling results and the topology of the ANF-RGC intracellular domains (TM, ARM and SHD), we believe that antiparallel orientation could be a possible packing form for SHD. Notably, the conserved signature <sup>801</sup>ERT<sup>803</sup> motif (shown in **Figure 1B** in stick form and marked with blue, green, and red colors for the E, R, and T residues, respectively) is centrally located what constitutes to be one of the distinctive features of the SHD (Anantharaman et al., 2006). The antiparallel homodimers were then analyzed for the existence of heptad arrangement because it has been proposed to be critical for the SHD coiled-coil structure and function (Anantharaman et al., 2006). The SOCKET software (Walshaw and Woolfson, 2001) was utilized to search for the most energy favorable packing structures between the two alpha-helices. Out of the top ten M-ZDOCK predicted complexes (**Supplementary Figure S3**), one (4th rank) was identified to satisfy the threedimensional arrangement of the heptad residues in a coiled-coil, for example: i and i+7 residues face the same direction and the coiled coil dimer interface is created using interactions between "a" and "d" heptad residues. The other complexes (shown in the **Supplementary Figure S3**) did not satisfy this arrangement. Detailed SOCKET analysis of top-ranked M-ZDOCK structure that satisfies the coiled coil helix conditions is provided in the **Supplementary Figure S4**. Thus, this homo-dimer structure comprising five repeating heptads was chosen as the best possible model for the analyzed region. The residues that could most likely (at 3 Å cut-off) form the dimer interface were: S786, E789, N<sup>793</sup> , E <sup>796</sup>, E797, E800, T803, L807, K810, A813, and E814. These residues and their locations within the respective heptads are shown in the three-dimensional homo-dimer model (**Figure 1C**) and their possible interactions are listed in **Table 1**.

Since the modeling studies demonstrated that the ANF-RGC sequence segment N783-P<sup>822</sup> satisfy all structural requirements of being a SHD, the next logical step was to scrutinize its functional significance in ANF-RGC signal transduction. Through sequential deletion and expression analyses, the roles of the entire SHD and each individual heptad in modulation of the ANF/ATP-dependent and the Ca2+-dependent signaling pathways were investigated.

#### Heptad-Deletion Analysis on the Basal Catalytic Activity of CCD

Six deletion mutants - -1SHD, 1H (Heptad)1, 1H2, 1H3, 1H4, 1H5- - were constructed. The wt-ANF-RGC and the deletion mutants were expressed in COS cells and their BASAL catalytic activities were determined in the cells particulate fractions (**Figure 2**). Each activity is denoted as [pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> ]. The basal catalytic activity of wt-ANF-RGC was 56 ± 9 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , K<sup>M</sup> for the substrate GTP, 0.63 µM and Hill coefficient 2.1 (**Table 2**). These values served as a control. Notably, divergence in a mutant's basal catalytic characteristics from the control values will measure the deviation from the optimal positioning of one or more of the seven catalytic elements formed at the core catalytic domain homodimer interface: D<sup>849</sup> and D<sup>893</sup> (Mg2<sup>+</sup> binding), N<sup>968</sup> (ribose positioning), E<sup>879</sup> and C<sup>951</sup> (guanine recognition), and/or TABLE 1 | Key interactions between amino acid residues forming the interface of ANF-RGC SHD as identified by M-ZDOCK program.


TABLE 2 | Deletion of the individual heptads or the entire SHD from ANF-RGC signaling helix does not affect of the cyclases' catalytic characteristics.


COS cells were induced to express ANF-RGC or its mutants. Membranes of these cells were individually analyzed for guanylate cyclase activity in the presence of increasing concentrations of GTP (0–2 µM) and constant 4 µM MgCl2. The experiment was done in triplicate and repeated four times. Presented values of K<sup>M</sup> ± SD were determined by GraphPad PRISM 4 program, of the Hill's coefficients ± SD and Kcat ± SD were calculated as described in Section "Materials and Methods."

R <sup>940</sup> and R<sup>972</sup> (triphosphate positioning) (Ravichandran et al., 2017).

#### 1SHD

The mutant exhibited basal catalytic activity of 61 ± 10 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 2**). Its activity was dependent on GTP concentration with a K<sup>M</sup> for GTP equal to 0.6 µM and the Hill coefficient 2.1 ± 0.3. These values are almost identical to the control ones. In conclusion, the whole SHD structure has no role in controlling the basal catalytic activity of CCD.

#### 1H1, 1H2, 1H3

Deletion of these heptads resulted in lowering of the basal activities to 20 ± 1, 19 ± 4, and 22 ± 5 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 2**). These respective values demonstrate that in the deletion mutants, the alignment of the 7-residue catalytic element in the CCD homodimer structure has been disturbed and suggest that these three heptads in their native states cause about three-fold upregulation of the basal catalytic activity.

#### 1H5

Deletion of the 5th heptad caused lowering of the basal activity to 10 ± 3 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 2**). This value indicates that the 5th heptad controls more than 80% of the basal

catalytic activity of CCD, suggesting that its deletion severely compromises alignment of the 7-residue catalytic element.

#### 1H4

Deletion of the 4th heptad resulted in basal catalytic activity of 105 ± 18 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 2**). This activity, significantly higher than that of the wt-ANF-RGC indicates that this heptad, in contrast to the other four, functions as a suppressor of ANF-RGC activity. It prevents the 7-residue catalytic element from acquiring the optimal for catalysis conformation but does not interfere with the substrate GTP binding as the K<sup>M</sup> for GTP remains comparable to that of wt-ANF-RGC.

To verify that these differences in basal catalytic activities truly reflect the differences of enzymatic properties but not different levels of the proteins expression, the membranes were analyzed by Western blot (**Figure 2**: the intensities of the immunoreactive bands are shown at the top of the figure).

When the heptad-deletion mutants' catalytic velocities were analyzed, they showed identical relationships between their catalytic activities and the substrate, GTP, concentration. The GTP concentration at which the activities of the wt-ANF-RGC and the deletion mutants were equal one-half of the maximal activity (KM) was ∼0.65 µM and the Hill coefficient values were similar, averaging at ∼2 (**Table 2**).

We therefore, conclude that in ANF-RGC the SHD tunes the basal catalytic activity of the CCD module. In basal (unstimulated) state, SHD tightly regulates configuration of the wreath- or circlet-like structure of the CCD formed by the 7 residue catalytic element at the interface of its homo-dimer. The tuning is critical, because its individual heptad regions in isolation cause disharmony in functioning of the CCD and generation of cyclic GMP in its resting state.

#### <sup>801</sup>E <sup>802</sup>R <sup>803</sup>T in H3 Is a Critical Conserved Motif in the MGC Family

The original signature characteristic defined for the SHD was that its third heptad contained a strongly conserved "ERT" motif (**Figure 1B**). The motif is also present in all MGC family members. In ROS-GC1 the third heptad is the "hot-spot" for naturally occurring mutations which ultimately lead to blindness with the R<sup>787</sup> being the most mutation-prone residue as four single mutations and two complex mutations are known to affect it (reviewed in Sharon et al., 2017). R787C mutation disrupts H3's structural integrity, disables the basal ROS-GC1's catalytic activity and causes a cone-rod degeneration (Kelsell et al., 1998; Duda et al., 2000a).

In ANF-RGC the R<sup>802</sup> residue corresponds to R<sup>787</sup> in ROS-GC1. No naturally occurring mutation at this position as yet has been recorded in ANF-RGC. However, to test if the concept on the critical significance of the R residue in the function of MGC is broadly applicable, ANF-RGC R802C mutant was constructed and analyzed.

To assure that the introduced mutation did not impair proper membrane targeting, the wt-ANF-RGC and the mutant were expressed in COS cells and analyzed by Western blots using anti-ANF-RGC antibody (**Figure 2**). In both cases, immunoreactivity was found in the plasma membranes and in almost quantitatively equivalent amounts.

In contrast to its parent wt-ANF-RGC's basic catalytic activity of 56 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , the mutant's activity was only 24 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , the value almost identical to that for 1H3 mutant (**Figure 2**). The K<sup>M</sup> for GTP and Hill coefficient values were 0.61 ± 0.13 µM and 1.9 ± 0.3, respectively, very close to those for the wt-ANF-RGC (**Table 2**). These results demonstrate that like 1H3, R <sup>802</sup>C mutation lowers the basal catalytic activity of ANF-RGC and concomitantly destroys the structural integrity of H3 and SHD. At the sub-molecular level, R<sup>802</sup> residue in H3 appears to be the real determinant of the modulation of CCD's 7 residue catalytic element. This interpretation is supported by the finding where the analogous mutation in ROS-GC1 changes its structural configuration from the active homodimer to the inactive monomer form (Duda et al., 2000a).

# ANF-RGC Regulatory Activity

Signaling helix domain was predicted to always exist between two signaling domains where it separated the N-terminal sensory domain from the C-terminal catalytic signaling domain (Anantharaman et al., 2006). This theoretical prediction was supported experimentally with the ROS-GC1 transduction system (Duda et al., 2012b). Deletion of SHD blocked the downstream flowing (N-terminal-origin) GCAP1-modulated Ca2<sup>+</sup> signal, yet not the upstream flowing GCAP2-modulated Ca2<sup>+</sup> signal that did not pass through the SHD. Here we test whether this prediction is also true for two ANF-RGC regulatory systems, ANF-modulated and Ca2+-modulated.

#### ANF/ATP-Dependent ANF-RGC Regulatory Activity

Binding of ANF to the extracellular receptor domain of ANF-RGC triggers a chain of structural changes which are carried through the transmembrane domain to the intracellular portion of the cyclase where, by ATP-dependent changes within the ARM domain, they are processed further, sensed by the SHD and finally translated into the production of cyclic GMP at the catalytic domain.

#### **Intact SHD structure is obligatory for signaling**

When the membranes of COS cells expressing 1SHD mutant were exposed to 10−<sup>7</sup> M ANF and increasing concentrations of ATP, the activity of the mutant did not increase but stayed constant at ∼ 61–63 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> within the 0–1 µM range of ATP concentrations tested (**Figure 3**). Under identical conditions, membranes of COS cells expressing wt-ANF-RGC showed 10−<sup>7</sup> M ANF and ATP dose-dependent increase of guanylate cyclase catalytic activity (**Figure 3**). Since deletion of the total SHD renders the cyclase completely unresponsive to ANF/ATP it is concluded that the intact SHD is obligatory for the transduction of the signal originating at the cyclase's extracellular domain.

#### **Each heptad differentially impacts signaling**

Wt-ANF-RGC and the deletion mutants were expressed in COS cells and challenged with 10−<sup>7</sup> M ANF and varying concentrations of ATP. The results were analyzed in terms of

the maximal catalytic activity achieved and the x-fold stimulation exhibited by the cyclases in response to the ligands. The EC<sup>50</sup> values for ATP are provided in **Table 3**.

As expected, the activity of wt-ANF-RGC was stimulated in an ATP-dose-dependent fashion with half-maximal stimulation (EC50) at ∼0.25 µM ATP (**Figure 4A**). The maximal stimulation achieved was about 420 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , what represents ∼7.5-fold increase above the basal level (**Figure 4B**).

The 1H1 mutant responded to the ANF/ATP stimulation by dose-dependent increase in activity, with an EC<sup>50</sup> at 0.25 µM ATP (**Figure 4C**). The maximal activity reached was ∼100 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , what with the mutant's basal activity of 20 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , corresponded to 5-fold increase (**Figure 4D**).

TABLE 3 | The EC<sup>50</sup> (±SD) values for ATP in the presence of 10−7M ANF and for myr-NCδ in the presence of 1 µM Ca2+, in modulating the catalytic activity of wt-ANF-RGC and its mutants.


The experiments were performed as described in Section "Materials and Methods" as well as in the respective figure legends.

The 1H2 mutant responded to ANF/ATP with only marginally increased catalytic activity. The increase was from 19 to 27 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 4E**) what amounted to no more then 1.4-fold stimulation (**Figure 4F**). These values demonstrate critical role of this heptad in both basal and ANF/ATP-regulated ANF-RGC activity.

The 1H3 mutant in addition to losing 60% of the wildtype cyclase's activity (**Figure 2**) lost also most of its ability to be stimulated by ANF/ATP (**Figure 5A**). The stimulation of approximately 2.3-fold above the basal level (**Figure 5B**) reached plateau at ∼0.75 µM ATP with the Vmax averaging at ∼50 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> and the half-maximal stimulation occurred at ATP concentration of 0.25 µM.

When the R802C mutant was exposed to 10−<sup>7</sup> M ANF and increasing concentrations of ATP, its activity rose in a dosedependent fashion (**Figure 5C**), but the stimulation was limited, not exceeding 3-fold (**Figure 5D**). Thus, deletion of the third heptad or R802C mutation, similar to R787C mutation in ROS-GC1, significantly disables the ANF-RGC's basic catalytic activity [lowers it from 56 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> for the wt-ANF-RC to 22 and 24 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> for the 1H3 and R802C mutants, respectively]. The picture is different however, when analyzing the ligand-dependent activity. The R802C mutation in ANF-RGC disables the cyclase's responsiveness to ANF/ATP thus, makes it hyporesponsive to the physiological ligand, whereas in ROS-GC1 the R787C mutation makes the cyclase hyper-responsive to its physiological ligand GCAP1 (Duda et al., 2000a, 2018).

1H4 mutant, which basal activity exceeds that of the parental ANF-RGC, in the presence of 10−<sup>7</sup> M ANF and 0–1 µM ATP was further stimulated, in a dose-dependent fashion, in its catalytic activity and ultimately reaches the point of saturated activity of ANF-RGC (compare **Figures 4A**, **6A**). Despite achieving the same Vmax as the wt-ANF-RGC, due to its elevated basal activity, the 1H4 mutant was stimulated only 4.5-fold above the basal activity (**Figure 6B**).

1H5 mutant, characterized by the lowest of all heptaddeletion mutants basal activity of 10 pmol cyclic GMP min−<sup>1</sup> (mg prot)−<sup>1</sup> was responsive to ANF/ATP. When challenged with 10−<sup>7</sup> M ANF and increasing concentrations of ATP its activity rose in a dose-dependent fashion (**Figure 6C**). Although the x-fold stimulation was significant, 5-fold (**Figure 6D**) the Vmax achieved amounted to no more than the basal activity of the wt-ANF-RGC.

#### Ca2+/Neurocalcin δ Dependent ANF-RGC Regulatory Activity

Given that SHD is the gateway module for the downstream hormonal (ANF/ATP) signaling pathway, does it have any role in the mainstream Ca2+/myristoylated neurocalcin δ (myr-NCδ) modulated signaling pathway? This question is critical because these two pathways originate in the modules at the opposite ends of the SHD, the hormonal in the extracellular domain and the Ca2<sup>+</sup> in CCD [Figure 11 in Duda et al., 2012a; Figure 7; in Review: (Sharma et al., 2016)].

To answer this question, first, the recombinant wt-ANF-RGC (control) and 1SHD mutant expressed in COS cells

were individually reconstituted with myr-NCδ and the guanylate cyclase activity was measured. The reaction mixtures contained 1 µM of Ca2<sup>+</sup> because in the absence of Ca2+, myr-NCδ is ineffective in modulating ANF-RGC activity (Duda et al., 2012b).

In accordance with previous observations (Duda et al., 2012a,b), Ca2+-bound myr-NCδ stimulated the catalytic activity of the wt-ANF-RGC in a dose-dependent fashion from 56 to 278 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 7A**); this equaled to 5-fold amplification of the basal activity (**Figure 7B**). The half

were assayed for guanylate cyclase activity in the presence of 10−<sup>7</sup> M ANF and increasing concentrations of ATP. (A,C) Show respectively, the specific guanylate cyclase activities of the 1H3 and R802C mutants, while figures (B,D), their x-fold stimulations. The experiment was done in triplicate and repeated three times. The results shown are mean ± SD of these experiments. The error bars for (A,C) are shown whereas for (B,D) they are within the size of the symbols.

maximal stimulation was achieved at 0.5 µM myr-NCδ (EC50) and the calculated Hill coefficient was 1.2. These values are very similar to those obtained previously (Duda et al., 2012b) and served as control for the analyses of ANF-RGC mutants' responses. Because ANF-RGC like other membrane guanylate cyclases exists as a homodimer and only dimer of Ca2+-bound myr-NCδ activates it (Duda et al., 2012b), the Hill coefficient of 1.2 demonstrates that 1 dimer of Ca2+-bound myr-NCδ binds and activates 1 dimer of ANF-RGC.

1SHD mutant, in response to Ca2+-bound myr-NCδ behaved almost identically to the control wt- ANF-RGC. The mutant was stimulated in its catalytic activity up to 348 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> with a half maximal stimulation at 0.4 µM myr-NCδ (**Figure 7C**). The x-fold stimulation was 5.7 (**Figure 7D**) and the Hill coefficient for myr-NCδ, 1.08 ± 0.1. These results demonstrate that the SHD has no role in Ca2+ modulated myr-NCδ signaling of CCD.

This being the case, it was predicted that individual building blocks of SHD- -H1, H2, H3, H4, H5- -will also have no regulatory role in Ca2+-modulated signaling, meaning that deletion of any individual heptad will not affect the x-fold stimulation which will remain approximately the same for all deletion mutants. To test this prediction the individual heptad deletion mutants were analyzed for their Ca2+-bound myr-NCδ-dependent activity. Because the results were very similar (**Figure 8**) only the essentials are summarized below and the EC<sup>50</sup> values for myr-NCδ in modulating the guanylate cyclase activities of the mutants are provided in **Table 3**.

1H1 mutant: basal activity 20 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , myr-NCδ-modulated activity, 105 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 8A**), 5-fold stimulation (**Figure 8B**), Hill coefficient 1.40 ± 0.09.

1H2 mutant: basal activity 19 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , myr-NCδ-modulated activity, 78 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 8A**), ∼4-fold stimulation (**Figure 8B**), Hill coefficient 0.87 ± 0.12.

1H3 mutant: basal activity 22 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , myr- NCδ-modulated activity, 109 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 8A**), ∼5-fold stimulation (**Figure 8B**), Hill coefficient 1.23 ± 0.08.

R <sup>802</sup>C mutant: basal activity 24 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , myr-NCδ-modulated activity, 124 pmol cGMP min−<sup>1</sup>

guanylate cyclase activities of the 1H4 and 1H5 mutants, respectively while figures (B,D), their x-fold stimulations. The experiment was done in triplicate and repeated three times. The results shown are mean ± SD of these experiments. The error bars for (A,C) are shown whereas for (B,D) they are within the size of the symbols.

(mg prot)−<sup>1</sup> (**Figure 8A**), ∼5-fold stimulation (**Figure 8B**), Hill coefficient 1.12 ± 0.11.

1H4 mutant: basal activity 105 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , myr-NCδ-modulated activity, 346 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 8A**), ∼3.5-fold stimulation (**Figure 8B**), Hill coefficient 0.91 ± 0.09. Note: Despite lower x-fold stimulation this mutant's Vmax was comparable to that of the wt-ANF-RGC.

1H5 mutant: basal activity 9 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> , myr-NCδ-modulated activity, 40 pmol cGMP min−<sup>1</sup> (mg prot)−<sup>1</sup> (**Figure 8A**), ∼4.5-fold stimulation (**Figure 8B**), Hill coefficient 0.94 ± 0.07.

The above results demonstrate that Ca2+/myr-NCδdependent activation of ANF-RGC is independent of SHD.

#### DISCUSSION

One of the seven members, ANF-RGC is the first discovered membrane guanylate cyclase (Paul, 1986; Paul et al., 1987; Sharma, 1988). This discovery is historic because it established cyclic GMP as the third hormonal second messenger and the membrane guanylate cyclase as a bona fide signal transduction system (Reviewed in Sharma et al., 2016). Structurally, in contrast to the other two, cyclic AMP and IP3, the system is neither G-protein driven nor a three-component signaling system. Instead, and uniquely, it is a single component hormonal (ANF) signal transducer that generates cyclic GMP and also embeds the hormonal receptor. After directly binding the hormone, it accelerates the production of cyclic GMP. Hydropathic analysis of its cloned form reveals ANF-RGC to be a multi-modular protein (Chang et al., 1989; Chinkers et al., 1989; Pandey and Singh, 1990), and the prediction was made that the interplay of these modules is required for the ANF signal transduction (Review Sharma et al., 2016; Sharma and Duda, 1997).

More than four decades-research bears out this prediction. Studies with the reconstructed ANF-RGC gene demonstrated that the ANF-receptor binding site resided in the guanylate cyclase's ExtD (Duda et al., 1991), and the deletion analysis disclosed that the its ATP-regulated and catalytic site resided in ICD (Marala et al., 1992). These studies began to decode the function-based identity of these modules, define their biochemistry and link them with multiple physiological processes. Thereby, it is now established that ANF-RGC is not

only the sole transducer of the hormonal signals, ANF and BNP, but also of the free [Ca2+] (Duda et al., 2014).

Empowered with the bimodal characteristic, utilizing entirely different modes, extraordinarily, ANF-RGC is directly linked with the physiological control of blood pressure regulation, hypertension and cardiac hypertrophy (Duda et al., 2012b, 2013; Review: Duda et al., 2014). And, equally amazingly, two signaling pathways, hormonal- and Ca2+-modulated never overlap, yet they are translated at a common CCD site. The present study focuses on these issues; identifies the SHD, discloses its three-dimensional model, demonstrates that the two signaling pathways different in their dependencies of on SHD, and then moves on to provide their advanced molecular descriptions.

#### Structure and Biochemistry of SHD

The original definition of the SHD was "a long helix consisting of multiple copies of a heptad (7-amino-acid) repeating unit . . .. with each heptad containing a similar configuration of residues. . . .The resulting coiled coil may be either parallel or anti-parallel depending on the orientation of the dimerization partners." (Anantharaman et al., 2006).

Except for STa-RGC (Saha et al., 2009) and ROS-GC1 (Duda et al., 2012b; Zagel et al., 2013), the SHD boundary of any MGC member has not been precisely mapped. The present report demonstrates that it is represented by the amino acid region <sup>783</sup>N-P<sup>822</sup> in ANF-RGC. Within it resides its core 35-amino acid α-helical region, <sup>786</sup>S-I<sup>820</sup> (**Figures 1A–C**). Based on the theoretical criterion set forth in Anantharaman et al. (2006), it is composed of five helical heptads, each heptad exhibiting its unique regulatory role in transmission of the ANF/ATPdependent down-stream signaling pathway for its translation into generation of the cyclic GMP.

Our 3D-model of the SHD is based on the docking results obtained through two programs, M-ZDOCK and Z-DOCK. Up to 30 top ranked M-ZDOCK predicted conformations and 6 out of top-10 ranked Z-DOCK predicted conformations show that in ANF-RGC the most energy-favorable dimeric arrangement of the two helical monomers is antiparallel (**Figure 1B**). These data form the base for our cautious proposal of ANF-RGC SHD conformation. This type of arrangement is in agreement with the crystallographic results of Ma et al., on the signaling helix coiled-coil domain of rat beta1 subunit of the soluble guanylyl cyclase (Ma et al., 2010) but in disagreement with the electron microscopic results on rat soluble guanylate cyclase coiled-coil (Campbell et al., 2014) and crystallographic, on Mycobacterium intracellulare adenylate cyclase (Vercellino et al.,

2017). In addition, studies on serine chemotaxis receptor identified the signaling helix consisting of "long coiled coil stretches in both parallel and antiparallel configurations" (Kim et al., 1999).

Is it feasible that the SHD monomers within the ANF-RGC molecule form an antiparallel dimer? We compared the protein sequences of Cya from Mycobacterium intracellulare (Vercellino et al., 2017) with rat ANF-RGC (**Supplementary Figures S5, S6**). Unlike Cya, ANF-RGC is not a multipass transmembrane protein. In ANF-RGC there are more than 300 amino acids between the TM and the SHD domains, while there are only 2 amino acids between TM and SHD regions in Cya. Thus, it makes sense that the parallel arrangement of the Cya SHD monomers is the only possibility. In ANF-RGC, however, considering the spacing between the TM and SHD segments, the dimerizing monomers have the possibility to acquire either a parallel or antiparallel arrangement. Although we appreciate the limitations of the modeling approach, we also realize that the experimental (crystallography or electron microscopy) structure of the ANF-RGC SHD is not known yet, and that the differences in the Spatial arrangement of Secondary structure Elements (SSE) - especially the distance between the transmembrane and SHD segment – between ANF-RGC and adenylate cyclases are significant (Mycobacterium intracellulare adenylate cyclase), we regard our results a trigger for further research on the ANF-RGC signal transduction mechanism.

Twelve residues (**Figure 1C**, indicated by asterisks) form the dimer interphase. Disclosure of this antiparallel feature assumes a significant structural importance because it represents a mode of communication between three-dimensional fold of SHD with the three-dimensional antiparallel fold of the CCD dimer (Liu et al., 1997; Duda et al., 2012b; Ravichandran et al., 2017). Together with the prior evidence that configuration of the ExtD dimer is head-to-head (Ogawa et al., 2004) and that it is the point of origin of the ANF/ATP signal (Duda et al., 1991), it is now concluded that this down-stream signal flows from the parallel-oriented receptor module to the terminal antiparallel-oriented modules of SHD and CCD for translation into the generation of cyclic GMP.

Through site-directed mutational and expression analysis of the detergent-exposed mutants of STa-RGC (Saha et al., 2009), a common functional theme of the SHD structure in the downstream signaling path has been proposed. Here, SHD interacts with "regions on the guanylate cyclase domain, thereby acting as a clamp to ensure low levels of cyclic GMP production." This means that SHD is a constitutive switch; it suppresses the basal catalytic activity of CCD in its native state and releases the suppression to the downward flowing hormonal signal, causing amplification of the CCD activity and generating accelerated production of the cyclic GMP.

Our previous study with ROS-GC1 (Duda et al., 2012c) and the present one with ANF-RGC supports the general conclusion that SHD is a critical transmission switch of the down-stream signaling pathways, yet it disputes the concept that it is a general suppressor of the basal CCD activity (Saha et al., 2009). We conclude this through analysis of the deleted SHD module of ANF-RGC. This mutant ( 1SHD) and the wt-ANF-RGC have almost identical basal catalytic activities. Thus, in its native state SHD is not the suppressor of the CCD's activity. Rather, in accord with the previous findings (Hirayama et al., 1993; van den Akker et al., 2000; Venkataraman et al., 2008; Saha et al., 2009; Zagel et al., 2013; Ravichandran et al., 2017), the present study demonstrates that the SHD and CCD are operationally independent modules, the latter controlling its own intrinsic catalytic activity.

Importantly, these results further challenge the alternate concept that the SHD in ANF-RGC (termed DD by the investigators) (Garbers, 1992; Wilson and Chinkers, 1995) and ROS-GC1 (Tucker et al., 1999; Ramamurthy et al., 2001; Peshenko et al., 2004; Dizhoor et al., 2016) is indispensable for dimerization of the CCD. The CCD exists as a functional homo-dimer in the absence of SHD (Venkataraman et al., 2008; Duda et al., 2012c). Additionally, circular dichroism experiments demonstrate that the isolated SHD in STa-RGC and in ROS-GC1 does not form a typical coiled-coil (Saha et al., 2009; Zagel et al., 2013).

#### Five-Heptad Intra-Helical-Regions Control Basal Catalytic Activity of ANF-RGC Through Differential Modes

Although intact SHD has no role in controlling the basal catalytic activity of the CCD, analysis of the individual heptads demonstrates surprising results. Every heptad individually controls the CCD's catalytic activity. And, except for H4, the pattern is that they all amplify it, H5 being the most effective. H4 suppresses it. Additionally, like its parent H3, the arginine, R <sup>802</sup>, residue in it also amplifies the basal catalytic activity of ANF-RGC.

To explain these mechanisms, important clues are provided by the kinetic properties of these deleted heptads (**Table 2**). The deletions or point mutation in H3 do not affect K<sup>M</sup> values for the substrate GTP, yet affect catalytic efficiency (**Table 2**). Thus, individually the heptads are connected with the 7-residue catalytic element of the CCD. In this manner, they control the wreath- or circlet-like configuration of the element, and degree of its catalytic activity (Ravichandran et al., 2017). However, collectively, under native conditions they compensate each other's affect to keep the configuration of the catalytic center optimal for its normal catalytic operation. Yet, if their structural alignment in the core SHD is disturbed, e.g., by mutation, they concomitantly destabilize the CCD and affect its basal catalytic activity. This is illustrated by the fact that site-directed mutations in H4 of the ROS-GC1 cause significant rise in the basal catalytic activities of the mutants (Zagel et al., 2013), consistent with the present results established for the properties of 1H4 mutant.

#### Heptad 3 and Its Conserved R Residue

In ANF-RGC similar to ROS-GC1, heptad 3 of the SHD and its node "R" residue are critical in tuning total modular activity of the cyclase. The R residue is the central part of the "ERT" conserved signature motif in the entire membrane guanylate cyclase family (Anantharaman et al., 2006) and its naturally occurring mutations cause a loss of function and lead to a serious abnormalities (Vasques et al., 2013; Nakao et al., 2015; Sharon et al., 2017; Duda et al., 2018). The most frequently observed is the R to C substitution. Why it has so severe consequences? They can be explained by the profound differences in biochemical properties of these two aminoacids (data from Amino Acid Explorer)<sup>6</sup> . The R residue has high side-chain flexibility, has the ability to form ionic and up to 7 hydrogen bonds, its isoelectric point is 10.8, and it is hydrophilic; in contrast, the C residue has low side chain flexibility, forms mostly covalent disulphide bonds, and only under certain conditions can form hydrogen bonds (Mazmanian et al., 2016; van Bergen et al., 2016) but never ionic bonds; its isoelectric point is 5, and hydrophobicity is 0.721. Most importantly, the R to C substitution is extremely rare as reflected by its BLOSUM62 score (−3).

#### SHD Is the ANF/ATP Signal Transmission Switch to CCD

Early studies demonstrate that ROS-GC1 transduces two Ca2+ sensor-modulated, GCAP1 and GCAP2, pathways (Duda et al., 2012c). The trajectories of these pathways are opposite, GCAP1's down-stream and GCAP2's upstream. Only in the downstream pathway SHD constitutes the pathway TRANSMISSION SWITCH to the CCD. Does SHD also constitute the signaling switch for the down-stream ANF/ATP signaling pathway?

The answer is yes. The SHD-deleted mutant of ANF-RGC does not respond to the ANF/ATP signaling. Thus, the SHD switch is the gateway for the down-stream ANF/ATP signaling pathway. We propose that this may constitute a common theme of all the down-stream signaling pathways.

The conclusions on the SHD role in ANF-RGC signaling are in a general agreement with the available results on the role of the signaling helix in activation of the receptor/adenylyl cyclase reporter Artrospira maxima (Winkler et al., 2012). Deletion of various fragments of the SHD resulted in an increase or decrease of the enzyme basal activity and reversal or lack of response to its ligand, serine, signaling.

Incorporating this new feature, we now advance our existing model (Duda et al., 2013) of "ANF/ATP Signaling of ANF-RGC, Cyclic GMP Production Events." Barring this addition, the model narration remains unchanged.

MODEL (**Figure 9**): The signal originates by binding of one molecule of ANF to the ExtD of the ANF-RGC dimer (Duda et al., 1991; van den Akker et al., 2000). The binding modifies the juxtamembrane domain where the disulfide <sup>423</sup>C-C<sup>432</sup> structural motif is a key element (Ogawa et al., 2004; Duda and Sharma, 2005). The signal twists the TMD, induces a structural change in the ARM domain, and prepares it for the ATP binding (Burczynska et al., 2007). Step 1, the ARM domain binds ATP, which leads to a cascade of temporal and spatial changes (Duda et al., 2000b). They involve a shift in the ATP binding pocket position by 3– 4 Å and rotation of its floor by 15◦ (G<sup>505</sup> acts as a critical pivot for both the shift and the rotation), (2) movement by 2– 7 Å but not the rotation of its β4 and β5 strands and its loop, and (3) movement of its αEF helix by 2–5 Å. This movement exposes the hydrophobic motif, <sup>669</sup>WTAPELL675, which through SHD facilitates its direct interaction with the CCD resulting in its partial, ∼50%, activation (Duda et al., 2009). Step 2, the six phosphorylation sites are brought from their buried state to their exposed state (Duda et al., 2011b). ATP, through a hypothetical protein kinase, phosphorylates the residues, and full activation (additional 50%) of ANF-RGC is achieved. ANF-dependent cyclic GMP is generated and functions as the second messenger of blood pressure regulation. Concomitantly, phosphorylation converts the ATP binding site from high to low affinity; ATP dissociates, and ANF-RGC returns to its ground state 55 (Duda et al., 2011b).

### SHD Has No Role in Transmission of the Ca2+-Modulated ANF-RGC Signaling Pathway

In contrast to the ANF/ATP-modulated ANF-RGC signaling pathway, the myr-NCδ-modulated Ca2<sup>+</sup> signaling pathway is main-stream, it originates and is translated in the CCD (Duda et al., 2014). In addition to the SHD, its other N-terminal modules - - ExtD, JMD and ARM- - have no role in transmission of the

<sup>6</sup>https://www.ncbi.nlm.nih.gov/Class/Structure/aa/aa\_explorer.cgi

CCD. (B) ANF/ATP. The trajectory of ANF signaling pathway is shown with a red dashed arrow. From its origin at the ExtD, it passes through the structural domains of TM, ARM (ATP binds to ARM in response to ANF signal), and SHD on its way to CCD. The figure was modified from (Duda et al., 2012b).

Ca2+-modulated pathway to CCD (**Figures 7**, **8**). In this manner, this pathway does not overlap with the ANF/ATP-modulated pathway; consequently the two pathways operate independently. This transduction model "Ca2+/myr-NCd signaling of ANF-RGC" is illustrated below. [Note, studies with ROS-GC1 have demonstrated that the conserved WTAPELL motif constitutes hinge region of the membrane guanylate cyclase and is critical for both downstream and upstream signaling pathways (Duda et al., 2011a); it is therefore possible that this may be a general property of the MGC family].

MODEL (**Figure 9**): The CCD is composed of three subdomains. Between the two flanking, N-terminal and C-terminal, resides the third <sup>849</sup>DIVGFTALSAESTPMQVV<sup>866</sup> conserved subdomain of the MGC family (Ravichandran et al., 2017). Notably, it contains none of the residues forming the core catalytic element; all seven residues constituting the element reside in the two flanking domains. This domain represents the universal NCδ binding site. The active form of NCδ is its myristoylated dimeric form (Duda et al., 2012b). The domain becomes functional upon binding Ca2+-bound myr-NCδ; and it functions by activating the seven-residue catalytic element that spans the two flanking CCD subdomains (Ravichandran et al., 2017).

In the basal state, one dimer of myr-NCδ is bound to its target site in the CCD (Duda et al., 2012b). The catalytic activity of ANF-RGC is at the threshold level in the production of cyclic GMP from GTP. In the activated state, an increase in free [Ca2+]<sup>i</sup> with K1/<sup>2</sup> of 0.5 µM is captured by the myr-NCδ; It undergoes Ca2+-dependent configurational change/s, this, in turn, causes a concomitant structural change in the CCD's catalytic center. The residues, seven from each chain, align to form a wreath-like structure, it enhances its catalytic efficiency (kcat) and generates the saturated production of cyclic GMP.

It needs to be stressed again that the outstanding feature of this bimodal operation is that both operations occur independently, their operational modes are totally different, their signaling pathways never overlap, they energize the CCD through different mechanisms, yet they converge to control the common processes of blood and cardio-vasculature regulation (Duda et al., 2012a, 2013).

#### CONCLUSION

fnmol-11-00430 November 27, 2018 Time: 16:18 # 16

This study represents a continuation of the saga on the development of the membrane guanylate cyclase (MGC) field and is a part of the overall goal of decoding the molecular principles governing the transduction machineries of various MGCs. About five-decades ago the existence of hormonally regulated MGC was negated. Studies by the authors' group on the epinephrine-sensitive rat adrenal 494 carcinoma cells, challenged this notion (Perchellet and Sharma, 1980; Shanker and Sharma, 1980; reviewed in Sharma, 2002). Meticulous dissection of the MGC system elevated its stature from anonymity to being a preeminent signal transducer that generates intracellular second messenger, cyclic GMP and controls countless physiological processes.

Now, we focus on two signals that modulate the catalytic activity of ANF-RGC in the physiology of blood pressure regulation: one, hormonal ANF/ATP and the other, nonhormonal Ca2<sup>+</sup> , signaling through its sensor myr-NCδ. These signals employ different modes of regulating ANF-RGC catalytic activity. The ANF/ATP signal originating at the extracellular portion of the cyclase is critically dependent on the integrity of the signaling helix, whereas the Ca2<sup>+</sup> signal is independent of it. Finally, using the appropriate molecular modeling tools we show

#### REFERENCES


that the most plausible configuration of the ANF-RGC signaling helix is an antiparallel or head-to-tail dimer.

#### AUTHOR CONTRIBUTIONS

TD designed, carried out the experiments, and analyzed the results. SR created the 3D models. AP created and expressed all the mutants. RS conceptually planned and coordinated the study. All authors contributed to the writing of the manuscript.

# FUNDING

This project has been funded by the National Eye Institute: EY 023980 and with federal funds from the National Cancer Institute, National Institutes of Health, under contract HHSN 261200800001E. The authors are solely responsible for the contents of this study, which may not represent the official views of the National Institutes of Health or policies of the Department of Health and Human Services. There is no mention of trade names, commercial products, or organizations implying endorsement by the United States Government.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00430/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 © 2018 Duda, Pertzev, Ravichandran and Sharma. 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) and the copyright owner(s) 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 the Neuronal Calcium Sensor DREAM Modulates Presenilin-2 Endoproteolysis

Rocío Naranjo1,2 , Paz González 1,2 , Alejandro Lopez-Hurtado1,2 , Xosé M. Dopazo1,2 , Britt Mellström1,2 and José R. Naranjo1,2 \*

<sup>1</sup>Spanish Network for Biomedical Research in Neurodegenerative Diseases (CIBERNED), ISCIII, Madrid, Spain, <sup>2</sup>National Biotechnology Center (CNB), CSIC, Madrid, Spain

Deregulated intracellular Ca<sup>2</sup><sup>+</sup> and protein homeostasis underlie synaptic dysfunction and are common features in neurodegenerative diseases. DREAM, also known as calsenilin or KChIP-3, is a multifunctional Ca<sup>2</sup><sup>+</sup> binding protein of the neuronal calcium sensor superfamily with specific functions through protein-DNA and protein-protein interactions. Small-molecules able to bind DREAM, like the anti-diabetic drug repaglinide, disrupt some of the interactions with other proteins and modulate DREAM activity on Kv4 channels or on the processing of activating transcription factor 6 (ATF6). Here, we show the interaction of endogenous DREAM and presenilin-2 (PS2) in mouse brain and, using DREAM deficient mice or transgenic mice overexpressing a dominant active DREAM (daDREAM) mutant in the brain, we provide genetic evidence of the role of DREAM in the endoproteolysis of endogenous PS2. We show that repaglinide disrupts the interaction between DREAM and the C-terminal PS2 fragment (Ct-PS2) by coimmunoprecipitation assays. Exposure to sub-micromolar concentrations of repaglinide reduces the levels of Ct-PS2 fragment in N2a neuroblastoma cells. These results suggest that the interaction between DREAM and PS2 may represent a new target for modulation of PS2 processing, which could have therapeutic potential in Alzheimer's disease (AD) treatment.

Keywords: presenilins, calcium, DREAM, neuronal calcium sensors, repaglinide

#### INTRODUCTION

The neuronal calcium sensor DREAM, also known as calsenilin or KChIP-3, is a multifunctional Ca2<sup>+</sup> binding protein with specific roles in different subcellular compartments through protein-DNA and/or protein-protein interactions. Thus, in the nucleus DREAM binds to DRE sites in the DNA and represses transcription of target genes (Carrión et al., 1999). In addition, DREAM interacts with other nucleoproteins, like CREB, CtBP1, nuclear receptors or TTF-1 and modifies their transcriptional function (Ledo et al., 2002; Rivas et al., 2004; Scsucova et al., 2005; Zaidi et al., 2006). Outside the nucleus, the DREAM interactome comprises a heterogeneous set of proteins encompassing ion channels, membrane receptors, GRK kinases and presenilins (PSs), among others (reviewed in Rivas et al., 2011; Burgoyne and Haynes, 2012). Binding of DREAM to these proteins modulates channel gating, response to agonist stimulation (An et al., 2000; Savignac et al., 2005; Rivas et al., 2009; Wu et al., 2010) or regulates DREAM activity through post-translational modifications including phosphorylation or sumoylation (Ruiz-Gomez et al., 2007; Palczewska et al., 2011). Binding to Ca2<sup>+</sup> induces conformational changes in DREAM that prevents DREAM binding to DRE sites in the DNA (Carrión et al., 1999) and distinctly modifies the interaction

#### Edited by:

Jaewon Ko, Daegu Gyeongbuk Institute of Science and Technology (DGIST), South Korea

#### Reviewed by:

Homira Behbahani, Karolinska Institutet (KI), Sweden Se-Young Choi, Seoul National University, South Korea

\*Correspondence:

José R. Naranjo naranjo@cnb.csic.es

Received: 17 October 2018 Accepted: 21 November 2018 Published: 03 December 2018

#### Citation:

Naranjo R, González P, Lopez-Hurtado A, Dopazo XM, Mellström B and Naranjo JR (2018) Inhibition of the Neuronal Calcium Sensor DREAM Modulates Presenilin-2 Endoproteolysis. Front. Mol. Neurosci. 11:449. doi: 10.3389/fnmol.2018.00449 with other proteins (Holmqvist et al., 2001; Naranjo et al., 2016). Furthermore, binding to arachidonic acid and other smallmolecules, like repaglinide or Cl-888, also induces changes in DREAM that affect its activity on Kv4 channel gating or on the processing of activating transcription factor 6 (ATF6; Holmqvist et al., 2001; Okada et al., 2003; Bowlby et al., 2005; Naranjo et al., 2016).

PSs (PS1 and PS2) are the catalytic core of the γ-secretase complex, an enzymatic activity linked to the plasma membrane that cleaves multiple intramembrane substrates including the amyloid precursor protein (APP), cell surface receptors and adhesion molecules like Notch, E-cadherin and ErbB4 (McCarthy et al., 2009; Haapasalo and Kovacs, 2011). In addition, PS proteins participate in other γ-secretase independent cellular functions including Wnt/β-catenin signaling, calcium homeostasis, apoptosis, protein trafficking, lysosomal function and autophagy (reviewed in Jurisch-Yaksi et al., 2013; Duggan and McCarthy, 2016). Importantly, mutations in the PS genes are associated with inherited familial Alzheimer's disease (AD), and they influence both γ-secretase dependent and independent cellular functions (Sherrington et al., 1995; Zhang et al., 1998; Lee et al., 2010; Supnet and Bezprozvanny, 2011).

Posttranslational modifications of the PSs include endoproteolysis, caspase cleavage, phosphorylation and ubiquitination (Duggan and McCarthy, 2016). PS endoproteolysis occurs in the large cytosolic loop of the protein generating N- and a C-terminal fragments that remain stably associated as a heterodimer, which constitutes the active core of the γ-secretase holoenzyme (Wolfe et al., 1999; Esler et al., 2000; Li et al., 2000; Spasic and Annaert, 2008; Lu et al., 2014; Sun et al., 2015). PS proteins are highly homologous, however, fragments generated from either PS1 or PS2 do not interact with each other, suggesting that the two PSs form independent complexes (Capell et al., 1998; Saura et al., 1999). Further supporting the idea of independent functions for PS proteins, PS1 knockout (KO) mice die shortly after birth displaying aberrant defects in the central nervous system and spinal ganglia (Shen et al., 1997; Wong et al., 1997). To date, the protease activity that mediates PS proteolysis is unknown, though the acidic protease inhibitor pepstatin A is the best known PS endoproteolysis inhibitor (Campbell et al., 2002). The involvement of other protease activities in PS cleavage like the proteasome (Liu et al., 2003; Kopan and Ilagan, 2004; Massey et al., 2005) or the γ-secretase complex (Takasugi et al., 2003; Xie et al., 2005; Fukumori et al., 2010; Honarnejad et al., 2013) has been suggested. Importantly, PS endoproteolysis is developmentally regulated (Hartmann et al., 1997) and the amount of PS fragments in the cell is tightly controlled (Steiner et al., 1998; Saura et al., 1999), though the mechanisms that regulate these activities are not well understood. Other posttranslational changes are also important for PS function. Thus, PS1 phosphorylation occurs at multiple sites by a variety of kinases and modulates γ-secretase activity (Kuo et al., 2008), γ-secretase independent functions (Walter et al., 1999; Uemura et al., 2007) and PS stability (Lau et al., 2002). Moreover, caspase cleavage of the C-terminal fragment of PS2 (Ct-PS2) has been associated with the role of PS2 in apoptosis (Alves da Costa et al., 2002, 2003, 2006) and PS ubiquitination has been related to the stability of the full-length holoproteins and the degradation of the PS fragments (Massey et al., 2005).

Yeast two-hybrid assays identified the interaction between Ct-PS2 and DREAM, and coimmunoprecipitation experiments after heterologous overexpression in COS-7 cells confirmed the binding of DREAM to PS1 and PS2 C-terminal fragments (Buxbaum et al., 1998). Recently, the structural determinants of the DREAM-Ct-PS1 interaction have been elegantly analyzed (Pham and Miksovska, 2016). Upon binding, DREAM regulates PS activity at different levels. First, overexpression of DREAM in human neuroglioma H4 cells increases PS2 cleavage and accumulation of Ct-PS2 (Buxbaum et al., 1998). Since mutations in the PS genes are associated with the development of familial type of AD (FAD), this finding linked DREAM/calsenilin and neurodegeneration in AD (Buxbaum et al., 1998). This mechanism, however, was questioned in a more recent study that showed no change or a decrease in Ct-PS2 in human neuroblastoma SH-5YSY cells overexpressing DREAM or Ca2+ insensitive DREAM mutants, respectively (Fedrizzi et al., 2008). Second, binding of PS2 to DREAM increases γ-secretase activity and the levels of Aβ peptides, one of the end-products of APP processing. In this regard, genetic ablation of DREAM leads to reduced APP processing and lower brain levels of Aβ peptides (Lilliehook et al., 2003). Third, PSs regulate calcium homeostasis through interaction with the Sarco/ER/Ca2+-ATPase (SERCA) pump, InsP<sup>3</sup> and ryanodine receptors and/or forming leak channels for calcium ions in the endoplasmic reticulum membrane (Tu et al., 2006; Cheung et al., 2008; Green et al., 2008; Hayrapetyan et al., 2008; Nelson et al., 2011). DREAM overexpression enhances the depletion of calcium from the ER store (Fedrizzi et al., 2008) and reduction of PS endoproteolysis reduces calcium release from this subcellular compartment (Honarnejad et al., 2013). Finally, PSs have been associated with apoptosis through γ-secretase-dependent and -independent pathways (Alves da Costa et al., 2002, 2003, 2006; Wang et al., 2005; Zeng et al., 2015) and DREAM overexpression have been associated with cell death (Jo et al., 2001). Whether DREAMinduced Ct-PS2 fragment accumulation mediates this effect is currently not established.

Here, we show that endogenous DREAM interacts with PS2 and regulates PS2 endoproteolysis in the brain. Moreover, repaglinide, a DREAM binding molecule, blocks the DREAM/PS2 interaction and reduces PS2 processing in N2a mouse neuroblastoma cells.

# MATERIALS AND METHODS

#### Mice

Homozygous DREAM KO mice (Cheng et al., 2002) and transgenic mice expressing a dominant active mutant DREAM insensitive to calcium (Dierssen et al., 2012; Mellström et al., 2014) were used to assess the role of DREAM in PS2 endoproteolysis in the brain.

#### Cells

N2a mouse neuroblastoma and HEK293T cells were from ATCC. All cells were cultured in DMEM (with 10% FBS, penicillin/streptomycin, Glutamax; all from Invitrogen). Cell cultures were routinely checked for mycoplasma infection. To evaluate the effect of DREAM binding molecules on PS processing, neuroblastoma cells were exposed to increasing concentrations of repaglinide (30 nM to 3 µM, Sigma) and harvested 36 h later.

#### Coimmunoprecipitation

Endogenous DREAM and PS2 proteins were coimmunoprecipitated from whole cell extracts (200 µg) from mouse brain or N2a cells, using 5 µg affinity-purified mouse monoclonal anti-DREAM 1B1 (Ledo et al., 2002) Tagged proteins were coimmunoprecipitated from whole cell extracts (200 µg) from HEK293T cells cotransfected with plasmids encoding Myc-DREAM71–256 and Flag-Ct-PS2335–448, using 1 µg affinity-purified rabbit anti-Myc (ab9106, Abcam) in the presence of vehicle (DMSO) or repaglinide. Immunoprecipitated PS2 was detected by western blot with anti-PS2 (Cell Signaling Technology, #2192; endogenous) or mouse anti-Flag antibody (Sigma; overexpressed).

### Western Blot

Mouse hippocampal and cortical extracts were prepared as described (López-Hurtado et al., 2018). Briefly, brain tissue was homogenized on ice in lysis buffer (RIPA, Cell Signaling Technology, #9806) supplemented with 1 mM PMSF. After 20 min incubation on ice, extracts were cleared by centrifugation (14,000× g, 20 min) and protein concentration assessed (Bradford, BioRad). For the analysis of the effect by DREAM binding molecules on PS2 processing, N2a cells were treated with repaglinide, or vehicle (DMSO) at the indicated concentrations for 36 h. Cells were pelleted and incubated on ice in lysis buffer for 45 min, with occasional vortexing. Samples (20 µg) were analyzed by SDS-PAGE and immunoblot. PVDF membranes were incubated overnight at 4◦C with rabbit anti-PS2 (Cell Signaling Technology, #2192 or Abcam, EP1515Y, ab51249). Secondary antibody used was HRP-conjugated donkey anti-rabbit IgG (Jackson) and signals were detected with ECL Select (GE Healthcare). Control for equal loading were with total protein measured by Coomassie staining of the membrane after immunoblotting or HRP-conjugated β-actin (Sigma, A3854 clone AC-15). Lane and band intensity were quantified with ImageLab software (BioRad). We obtained similar results when we used as a primary antibody a rabbit monoclonal antibody raised against a C-terminal peptide of PS2 (Abcam Ab51249). This confirmed the specificity of the Ct-PS2 band detected by western blot.

#### Real-Time Quantitative PCR

RNA was isolated from tissue samples using TRIzol, treated with DNase (Ambion) and reverse transcribed using hexamer primer and Moloney murine leukemia virus reverse transcriptase. To confirm the absence of genomic DNA, each sample was processed in parallel without reverse transcriptase. Real-time quantitative PCR (qPCR) for endogenous PS2 was performed using the assay from Applied Biosystems (Mm00448405\_m1) and expression was normalized using HPRT as reference with primers 5<sup>0</sup> TTG GAT ACA GGC CAG ACT TTG TT 3<sup>0</sup> and 5 <sup>0</sup> CTG AAG TAC TCA TTA TAG TCA AGG GCA TA 3<sup>0</sup> and the MGB probe 5<sup>0</sup> TTG AAA TTC CAG ACA AGT TT 3<sup>0</sup> .

### Statistical Analysis

All data values shown are mean ± SEM. Differences were considered significant at P < 0.05. When possible, ordinary one-way ANOVA was used to analyze statistical differences among groups. For small or unequal sample size or non-Gaussian distribution, comparisons between groups were analyzed using t-test or the nonparametric ANOVA, Kruskal-Wallis test with Dunn's multiple comparison between groups. Animal experiments were randomized. Sample size was not predetermined by statistical method. Prism 6.0 software (GraphPad) was used for statistical analysis.

#### Study Approval

Experiments with mice were conducted in accordance with standard ethical guidelines (European Communities Directive 86/609 EEC; National Institutes of Health 1995). The CNB-CSIC and Madrid Community ethical committees approved the experiments with mice (PROEX 28/05).

# RESULTS

### DREAM Regulates the Processing of Presenilin-2 in Mouse Brain

Overexpression of DREAM increased PS2 processing and the levels of the Ct-PS2 in human H4 neuroglioma but not in SH-SY5Y neuroblastoma cells (Buxbaum et al., 1998; Fedrizzi et al., 2008). Overexpression conditions could be part of

the controversy between these results. Thus, we aimed to investigate whether or not endogenous DREAM regulates PS2 processing in vivo in basal conditions in mouse brain. Using total extracts from mouse hippocampus or cerebral cortex, coimmunoprecipitation experiments showed the interaction of endogenous DREAM and Ct-PS2 both in the hippocampus and the cerebral cortex (**Figure 1**). Moreover, endogenous DREAM was shown to interact not only with the Ct-PS2 processed fragment but also with the full length PS2 holoprotein (**Figure 1**). Similar results were observed using whole cell extracts from mouse N2a neuroblastoma cells (**Figure 1**).

We then investigated whether PS2 processing is regulated by DREAM in mouse brain. For this, we compared Ct-PS2 basal brain levels in wild type (WT), DREAM deficient and transgenic mice overexpressing a dominant active DREAM (daDREAM) mutant in the brain. Absence of DREAM in DREAM deficient mice significantly reduced Ct-PS2 levels both in the cerebral cortex and the hippocampus (**Figure 2A**). Correspondingly, increased levels of DREAM protein in daDREAM transgenic mouse brain resulted in significant accumulation of Ct-PS2 in both brain areas (**Figure 2A**). Changes in Ct-PS2 levels in the cerebral cortex were not accompanied by significant changes in PS2 gene expression (**Figure 2B**). However, a slight but significant increase in PS2 mRNA level was observed in the hippocampus of daDREAM mice (**Figure 2B**). Taken together, these results indicate that endogenous DREAM interacts with endogenous PS2 and regulates PS2 endoproteolysis, most likely, without directly affecting PS2 gene expression.

#### Repaglinide Blocks the DREAM-PS2 Interaction

We next analyzed whether DREAM-binding molecules could affect the DREAM/PS2 interaction. For this, we reproduced previous coimmunoprecipitation experiments (Buxbaum et al., 1998) using HEK293T cells overexpressing tagged proteins. Combined overexpression of Myc-DREAM and Flag-Ct-PS2 resulted in coimmunoprecipitation of Flag-Ct-PS2 with an anti-Myc antibody (**Figure 3A**). Detection of the Flag-Ct-PS2 fragment with an anti-Flag antibody or an anti-Ct-PS2 antibody rendered identical results. As negative controls, non-transfected cells or cells transfected with only one of the expression plasmids did not show immunoprecipitation (**Figure 3A**). Inclusion in the immunoprecipitation reaction of 1 mM CaCl<sup>2</sup> or up to 2 mM EGTA did not affect the interaction and did not modify the intensity of the immunoprecipitated

band (**Figure 3B**). These results confirm the Ca2+-independent nature of this interaction, as previously described (Buxbaum et al., 1998; Fedrizzi et al., 2008). Then, we proceeded to investigate a potential blockage of the DREAM-PS2 interaction by the DREAM binding molecule repaglinide, known to affect the interaction of DREAM with other proteins including Kv4 potassium channels and ATF6 (Naranjo et al., 2016; López-Hurtado et al., 2018). In vitro exposure to repaglinide significantly reduced the Myc-DREAM/Flag-Ct-PS2 coimmunoprecipitation (**Figure 3C**). Inclusion of DMSO, the vehicle for repaglidine, as a negative control did not change the extent of the DREAM-PS2 interaction (**Figure 3B**).

#### Repaglinide Blocks PS2 Endoproteolysis in N2a Cells

Finally, we analyzed whether partial block of the DREAM-PS2 interaction by repaglinide was translated in a noticeable effect on PS2 processing in an in vivo cell model. For this, we used N2a neuroblastoma cells, a cell model widely used to analyze PS2 endoproteolysis (Thinakaran et al., 1996). Exposure to repaglinide produced a concentration-dependent decrease in the levels of the Ct-PS2 fragment in N2a neuroblastoma cells with an estimated IC50 value of 0.76 µM (**Figure 4**).

# DISCUSSION

In previous work, we have shown that reduction in DREAM protein levels or blockade of DREAM activity, using repaglinide, activates ATF6 processing which results in a neuroprotective effect in murine models of Huntington's disease, delaying the onset and the progression of motor and cognitive decline in these mice (Naranjo et al., 2016; López-Hurtado et al., 2018). This finding opened a new avenue toward the search for effective HD treatments. Whether small-molecules binding to DREAM could have an effect on modifying the onset and/or the progression of other neurodegenerative pathologies is presently uncharacterized. DREAM has been

associated with the progression of AD through its interaction with PSs (Buxbaum et al., 1998; Lilliehook et al., 2003). Therefore, we analyzed whether repaglinide has an effect on this interaction. Here, we show that blockade of the DREAM-PS2 interaction with repaglinide is directly translated in a decrease in PS2 endoproteolysis and in reduced levels of the Ct-PS2 fragment in N2a neuroblastoma cells. Whether this effect also occurs in vivo in mouse brain and the potential consequences for the onset and progression of AD remains to be investigated.

Repaglinide was developed as a potent insulinotropic agent for treatment of type-2 diabetes (Malaisse, 1995). Repaglinide binding to neuronal calcium sensors (NCS) was first reported in bovine brain and retinal extracts, which showed Ca2+ dependent binding respectively to neurocalcin and VILIP-1, or to recoverin (Okada et al., 2003). Repaglinide also binds to members of the DREAM/KChIP subfamily (Naranjo et al., 2016) which indicates that repaglinide binding is a characteristic of all proteins of the NCS superfamily but not to other Ca2+-binding proteins, including calmodulin or proteins of the S-100 superfamily (Okada et al., 2003). Nonetheless, using tagged proteins in the coimmunoprecipitation experiments here we show that repaglinide specifically blocks the DREAM-PS2 interaction. Furthermore, the interaction of PS2 with other NCS except DREAM has not been reported.

After binding, repaglinide interferes with the biological activity of the Ca2<sup>+</sup> sensor, e.g., blockage of recoverin-mediated inhibition of rhodopsin kinase activity (Okada et al., 2003) DREAM-induced suppression of ATF6 processing (Naranjo et al., 2016) or, in this case, DREAM-induced activation of PS2 endoproteolysis. Work in progress, using single site mutations in DREAM aims to identify the binding pocket for repaglinide in the DREAM protein and whether these point mutations affect the interaction with PS2 and/or the induction of PS2 proteolysis.

The mechanisms that control PS endoproteolysis and finely keep a stoichiometry approximately 1:1 between N- and C-terminal fragments are poorly understood. To further complicate the picture, posttranslational mechanisms that potentially participate in these processes might be different for PS1 and PS2. Phosphorylation of PS1, but not of PS2, is known to occur at multiple sites within the cytosolic loop involving multiple kinases (Duggan and McCarthy, 2016). Early in vitro experiments described the interaction between DREAM and the C-terminal fragments of PS1 and PS2 as well as the increase in Ct-PS2 levels upon DREAM overexpression (Buxbaum et al., 1998). Our results confirm the DREAM-PS2 interaction in vivo and show the effect of DREAM inhibition on Ct-PS2 levels in N2a neuroblastoma cells. The functional consequences of the DREAM-PS1 interaction in terms of PS1 endoproteolysis and, the potential effect of DREAM interacting molecules in PS1 processing remains to be analyzed.

Overexpression of PS2 or DREAM has been associated with cell death (Janicki and Monteiro, 1997; Jo et al., 2001) and this effect is further enhanced by coexpression of DREAM and PS2 but not by a truncated form of PS2 (PS2/411stop) that lacks the C-terminal part (Jo et al., 2001). A parallel increase in Aβ1–42 levels after coexpression of DREAM with PS2, but not with PS2/411stop, indicated the involvement of γ-secretase activation in the cell death process, however, γ-secretase independent mechanisms can not be excluded. Interestingly, interaction between the calcium binding protein calmyrin and the cytosolic loop of PS2, and with lower affinity for the same domain in PS1, has been reported (Stabler et al., 1999). Like in the case of DREAM, coexpression of calmyrin and PS2 in HeLa cells modified the subcellular distribution of these proteins and increased cell death (Stabler et al., 1999). It was not analyzed, however, whether calmyrin modifies PS2 endoproteolysis, neither whether coexpression with PS1 also increases apoptosis (Stabler et al., 1999). The calmyrin-PS2 interaction requires and is regulated by Ca2+, suggesting that changes in cellular calcium homeostasis might control the functional output of this interaction (Stabler et al., 1999; Zhu et al., 2004). In contrast, the DREAM-PS2 interaction has been shown to be Ca2+-independent (Buxbaum et al., 1998; Fedrizzi et al., 2008). Our results support this view, however, titration curves for the binding of Apo-DREAM or Ca2+DREAM to peptides derived from the cytosolic loop of PS1 indicated that the interaction is stronger in the presence of Ca2<sup>+</sup> (Pham and Miksovska, 2016). The different experimental conditions may account for this discrepancy.

With respect to the mechanism by which DREAM regulates PS2 endoproteolysis nothing has been reported. One possibility, among others, is that DREAM could compete with ubiquilin proteins for binding to the cytosolic loop region and/or to the C-terminal region of PS2, increasing the endoproteolysis of PS2 and/or the half-life of the Ct-PS2 fragment. It has been shown that ubiquilin-1 and -2 proteins bind to PSs and regulate PS stability by two different mechanisms (Mah et al., 2000). One, it has been shown that ubiquilin-1 and -2 regulate PS endoproteolysis and overexpression of ubiquilin proteins increases the amount of full length PS2 holoprotein and reduces the formation of Nt- and Ct-PS2 fragments (Mah et al., 2000; Massey et al., 2005). Though, a more recent study from the same group suggests that ubiquilin rather increases PSs biosynthesis (Ford and Monteiro, 2007). Two, binding of ubiquilin to the PS2 was initially associated with the ubiquitin-mediated degradation of PS2 through the proteosome pathway (Massey et al., 2004, 2005), however, binding of ubiquilin to Ct-PS2 does not require ubiquitination of critical lysine residues in this domain (Ford and Monteiro, 2007). Whether DREAM increases Ct-PS2 levels in vivo by competing the ubiquilin/PS2 interaction will need future experimental analysis.

Taken together, the effect of repaglinide blocking the DREAM-PS2 interaction and reducing PS2 endoproteolysis suggests that the interaction between DREAM and PS2 may represent a new target for modulation of PS2 processing, which could have therapeutic potential in AD treatment. Future studies, however, should analyze whether the reduction in PS2 endoproteolysis by DREAM binding molecules efficiently translates in a decrease in γ-secretase activity, in a reduction in β-amyloid accumulation and amyloid plaque formation and, finally, in an improvement in cognition.

#### REFERENCES


#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript.

#### AUTHOR CONTRIBUTIONS

RN, PG, AL-H and XD performed the experiments and analyzed the data. BM and JRN conceived the study, analyzed the data and wrote the article.

### FUNDING

This work was funded by the Spanish Ministry of Economy, Industry and Competitivity (AEI-FEDER, EU grants): SAF2014- 53412-R and SAF2017-89554-R (to JRN), the Instituto de Salud Carlos III biomedical research in neurodegenerative diseases (CIBERNED) program (to JRN) and the Madrid regional government/Neurodegmodels (to JRN).

repressor for pain modulation. Cell 108, 31–43. doi: 10.1016/s0092-8674(01) 00629-8


by arachidonic acid is dependent on potassium channel interacting proteins. J. Neurosci. 21, 4154–4161. doi: 10.1523/JNEUROSCI.21-12-04154.2001


**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 © 2018 Naranjo, González, Lopez-Hurtado, Dopazo, Mellström and Naranjo. 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) and the copyright owner(s) 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.

# Functional Status of Neuronal Calcium Sensor-1 Is Modulated by Zinc Binding

Philipp O. Tsvetkov<sup>1</sup> , Andrei Yu. Roman<sup>2</sup> , Viktoriia E. Baksheeva<sup>3</sup> , Aliya A. Nazipova<sup>4</sup> , Marina P. Shevelyova<sup>4</sup> , Vasiliy I. Vladimirov<sup>5</sup> , Michelle F. Buyanova<sup>6</sup> , Dmitry V. Zinchenko<sup>5</sup> , Andrey A. Zamyatnin Jr.3,7, François Devred<sup>1</sup> , Andrey V. Golovin3,6,7,8, Sergei E. Permyakov<sup>4</sup> and Evgeni Yu. Zernii3,7 \*

<sup>1</sup> Aix-Marseille University, CNRS, INP, Institute of Neurophysiopathology, Faculty of Pharmacy, Marseille, France, <sup>2</sup> Institute of Physiologically Active Compounds (RAS), Chernogolovka, Russia, <sup>3</sup> Belozersky Institute of Physico-Chemical Biology, Lomonosov Moscow State University, Moscow, Russia, <sup>4</sup> Institute for Biological Instrumentation of the Russian Academy of Sciences, Pushchino, Russia, <sup>5</sup> Shemyakin and Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, Pushchino, Russia, <sup>6</sup> Faculty of Bioengineering and Bioinformatics, Lomonosov Moscow State University, Moscow, Russia, <sup>7</sup> Institute of Molecular Medicine, Sechenov First Moscow State Medical University, Moscow, Russia, <sup>8</sup> Faculty of Computer Science, Higher School of Economics, Moscow, Russia

#### Edited by:

Jose R. Naranjo, Spanish National Research Council (CSIC), Spain

#### Reviewed by:

Yogendra Sharma, Centre for Cellular and Molecular Biology (CSIR), India Alexander Dizhoor, Salus University, United States

> \*Correspondence: Evgeni Yu. Zernii zerni@belozersky.msu.ru

Received: 21 July 2018 Accepted: 28 November 2018 Published: 14 December 2018

#### Citation:

Tsvetkov PO, Roman AYu, Baksheeva VE, Nazipova AA, Shevelyova MP, Vladimirov VI, Buyanova MF, Zinchenko DV, Zamyatnin AA Jr, Devred F, Golovin AV, Permyakov SE and Zernii EYu (2018) Functional Status of Neuronal Calcium Sensor-1 Is Modulated by Zinc Binding. Front. Mol. Neurosci. 11:459. doi: 10.3389/fnmol.2018.00459 Neuronal calcium sensor-1 (NCS-1) protein is abundantly expressed in the central nervous system and retinal neurons, where it regulates many vital processes such as synaptic transmission. It coordinates three calcium ions by EF-hands 2-4, thereby transducing Ca2<sup>+</sup> signals to a wide range of protein targets, including G protein-coupled receptors and their kinases. Here, we demonstrate that NCS-1 also has Zn2+-binding sites, which affect its structural and functional properties upon filling. Fluorescence and circular dichroism experiments reveal the impact of Zn2<sup>+</sup> binding on NCS-1 secondary and tertiary structure. According to atomic absorption spectroscopy and isothermal titration calorimetry studies, apo-NCS-1 has two high-affinity (4 × 10<sup>6</sup> M−<sup>1</sup> ) and one low-affinity (2 × 10<sup>5</sup> M−<sup>1</sup> ) Zn2+-binding sites, whereas Mg2+-loaded and Ca2+-loaded forms (which dominate under physiological conditions) bind two zinc ions with submicromolar affinity. Metal competition analysis and circular dichroism studies suggest that Zn2+-binding sites of apo- and Mg2+-loaded NCS-1 overlap with functional EF-hands of the protein. Consistently, high Zn2<sup>+</sup> concentrations displace Mg2<sup>+</sup> from the EF-hands and decrease the stoichiometry of Ca2<sup>+</sup> binding. Meanwhile, one of the EF-hands of Zn2+-saturated NCS-1 exhibits a 14-fold higher calcium affinity, which increases the overall calcium sensitivity of the protein. Based on QM/MM molecular dynamics simulations, Zn2<sup>+</sup> binding to Ca2+-loaded NCS-1 could occur at EF-hands 2 and 4. The high-affinity zinc binding increases the thermal stability of Ca2+ free NCS-1 and favours the interaction of its Ca2+-loaded form with target proteins, such as dopamine receptor D2R and GRK1. In contrast, low-affinity zinc binding

promotes NCS-1 aggregation accompanied by the formation of twisted rope-like structures. Altogether, our findings suggest a complex interplay between magnesium, calcium and zinc binding to NCS-1, leading to the appearance of multiple conformations of the protein, in turn modulating its functional status.

Keywords: neuronal calcium sensor-1, zinc, calcium, magnesium, EF-hand motif, dopamine receptor D2R, GRK1, protein aggregation

#### INTRODUCTION

Divalent metal ions play a vital role in the vast majority of cellular processes. Among them, two alkaline earth metals, magnesium and calcium, as well as the transition metal zinc, are particularly important, since they are the most abundant ones in the human body. These ions are significantly different in intracellular levels and play different physiological roles. While intracellular concentration of free magnesium is high (0.5–2 mM), the free concentration of calcium and zinc is much lower (40–100 nM and 100 pM, respectively) (Romani and Scarpa, 1992;Sabatini et al., 2002;Krezel and Maret, 2006). In addition, free calcium concentration drastically increases in signaling waves, reaching > 100 µM in magnitude (Augustine et al., 2003). Consistently, magnesium binds to proteins with much lower affinity (equilibrium association constant, Ka, below 10<sup>5</sup> M−<sup>1</sup> ) than calcium (K<sup>a</sup> > 10<sup>5</sup> M−<sup>1</sup> ) and zinc (K<sup>a</sup> > 10<sup>7</sup> M−<sup>1</sup> ) (for review, see (Dudev and Lim, 2003)). Magnesium binding to proteins is less specific compared to Ca2<sup>+</sup> and Zn2<sup>+</sup> binding; and, although the common structural motif for Mg2<sup>+</sup> binding has not been described yet, magnesium is able to occupy calcium and zinc-binding sites. Yet, at high concentrations, zinc can compete with other metals or bind to nonspecific sites in proteins, which may interfere with their structural integrity and normal function, thus contributing to pathology (Choi and Koh, 1998;Barwinska-Sendra and Waldron, 2017).

The great majority of calcium-binding proteins (CaBPs) contain the same helix-loop-helix Ca2+-binding motif (referred to as an "EF-hand"), wherein calcium ion is coordinated by six oxygen atoms. Generally, CaBPs comprise several EF-hands, which bind calcium with micromolar dissociation constant. Commonly, this interaction induces a significant rearrangement in the CaBP structure, inducing the exposure of its hydrophobic surface, which is responsible for interaction with target(s), resulting in their activation/deactivation. All CaBPs have specific tissue, cellular and subcellular distribution profiles (for review, see (Yanez et al., 2012)). Along with other tissues, CaBPs are abundantly expressed in the nervous system where they play an essential role in decoding calcium signals and regulating many processes crucial for the viability and functioning of neurons. The calcium signals, emanating from photoreceptor neurons in response to light stimuli, regulate rhodopsin desensitization and cGMP synthesis through interaction with recoverin and guanylate cyclase activating proteins (GCAPs), two members of the neuronal calcium sensor (NCS) family [reviewed in (Burgoyne, 2007;Burgoyne and Haynes, 2012;Koch and Dell'Orco, 2015)]. Neuronal calcium sensor-1 (NCS-1) is another NCS protein, which was found in photoreceptors and other retinal neurons (De Raad et al., 1995;Baksheeva et al., 2015). In contrast to recoverin and GCAPs, NCS-1 is expressed throughout the nervous system. Its N-terminus contains a myristoyl group, which participates in the interaction of the protein with cellular membranes. According to in vitro and in vivo data, NCS-1 regulates more than 20 target proteins, including G protein-coupled receptors and their kinases (GRKs). As such, NCS-1 participates in neuronal growth and survival, reception, neurotransmission, synaptic plasticity and other cellular mechanisms [for review, see (Burgoyne and Haynes, 2012)]. NCS-1 contains four EF-hand motifs, but only three of them (EF2, EF3 and EF4) are able to bind calcium with nanomolar to micromolar dissociation constants (Jeromin et al., 2004). In the absence of calcium, two EF-hands (EF2 and EF3) could be occupied by magnesium ions (Warren et al., 2007;Aravind et al., 2008).

Zinc is long known to be a key element in neuronal growth and activity necessary for the normal development and functioning of the brain (Frederickson et al., 2005). Zinc deficiency results, for instance, in lowered glutamate receptor expression and decreased cognitive and motor performance in children (Penland et al., 1997;Gardner et al., 2005). It is also critical for the development, viability and specific function of the retinal neurons (Ugarte and Osborne, 2014). The levels of Zn2<sup>+</sup> in the retina and retinal pigment epithelium are decreased in the elderly, which may contribute to the pathogenesis of age-related macular degeneration (Wills et al., 2008;Lyubartseva and Lovell, 2012). Expression of retinal proteins involved in zinc homeostasis also becomes downregulated with age (Leung et al., 2012). In general, zinc serves to maintain the structure and function of hundreds of proteins, including enzymes of all known classes, transcription factors, receptors and signaling proteins. While, in many cases, zinc is tightly bound to proteins, it could also interact with their Zn2+-binding sites transiently in order to conduct biochemical stimuli. The most abundant Zn2+-binding motif in proteins is the Z-finger, which chelates zinc ion with nanomolar to picomolar affinity [for review, see (Maret and Li, 2009)]. Nevertheless, zinc ions can bind to EF-hands [as reported for calmodulin (Warren et al., 2007)] or between two EF-hand motifs, or even in-between two protein subunits in dimer, as observed in some members of the S100 family (Tsvetkov et al., 2010;Moroz et al., 2011). Considering that some members of the NCS family, such as recoverin (Permyakov et al., 2003), are able to bind zinc ions, we hypothesized that NCS-1 is also sensitive to Zn2+.

In this study, we demonstrate that apo-NCS-1 has two highaffinity zinc-specific sites and one low-affinity zinc-binding site. Zinc binding to NCS-1 reduces stoichiometry and increases the affinity of Ca2<sup>+</sup> binding to the protein. In contrast, physiologically relevant Mg2+- and Ca2+-loaded NCS-1 forms only bind two zinc ions with high affinity, which stabilizes their structure and may be required for maintaining the functional status of these forms. In addition, our findings suggest that the elevated concentration of free zinc, characteristic of some neurodegenerative and neuro-ophthalmological disorders, may lead to the formation of unstable, prone-to-aggregation pathological NCS-1 forms.

### MATERIALS AND METHODS

fnmol-11-00459 December 12, 2018 Time: 14:48 # 3

#### Purification of Proteins and Membranes

NCS-1 was obtained according to the protocol previously developed for recoverin with some modifications. To obtain recombinant myristoylated protein, NCS-1 gene was coexpressed in Escherichia coli strain BL21(DE3) Codon Plus RP with N-myristoyl transferase 1 from Saccharomyces cerevisiae (4 h, 37◦C) in the presence of 200 µg/ml myristic acid. The cells were harvested by centrifugation and lysed by freezing/thawing in extraction buffer (50 mM Tris-HCl (pH 8.0), 100 mM NaCl, 1 mM EDTA, 1 mM PMSF, 1 mM DTT) and subsequent incubation in the presence of 50 µg/ml of egg white lysozyme in the same buffer for 30 min. The lysate was clarified by centrifugation, loaded onto Phenyl Sepharose column (GE Lifesciences) equilibrated with 20 mM Tris-HCl buffer (pH 8.0), 2 mM CaCl2, 1 mM DTT and NCS-1 was eluted using the same buffer containing 2 mM EGTA. The obtained protein was loaded on HiTrap Q FF anion exchange column (GE Lifesciences) equilibrated with 20 mM Tris-HCl buffer (pH 8.0), 1 mM DTT and eluted by linear gradient of 0-1 M NaCl in the same buffer. NCS-1 (> 90% purity) was present in the fractions eluted at 380-500 mM NaCl. The obtained protein was dialysed overnight against 20 mM Tris (pH 8.0), 1 mM DTT and stored at −70◦C. Alternatively, to remove residual calcium NCS-1 sample obtained immediately after anion exchange chromatography was subjected to dialysis against 50 mM Tris-HCl (pH 8.0), 5 mM EDTA (3 h), followed by dialysis against deionized water (3 h), and dialysis against 20 mM Tris-HCl (pH 8.0), 1 mM DTT (Blachford et al., 2009). The degree of NCS-1 myristoylation was determined by analytical HPLC using a reversed-phase column [Phenomenex Luna C18(2)] and was more than 97%. NCS-1 concentration was measured with bicinchoninic acid assay kit (Thermo Fisher Scientific) or spectrophotometrically using previously determined molar extinction coefficient at 280 nm of 21,430 M−<sup>1</sup> (Kazakov et al., 2017).

N-terminal fragment of GRK1 (M1-G183) was obtained as GST-fusion construct (GST-N-GRK1) following the previously developed procedure (Komolov et al., 2009).

Dopamine receptor D2 (D2R) peptide (N430-R443) was produced using Fmoc/But solid-phase peptide synthesis.

Polyclonal (monospecific) antibodies against NCS-1 were generated by rabbit immunization and purified from immune serum on a column with immobilized antigen according to the previously published procedure (Zernii et al., 2003).

Photoreceptor membranes were prepared from frozen bovine retinas following the standard protocol with some modifications described in (Grigoriev et al., 2012).

#### Equilibrium Dialysis Experiments

Ca2+/Zn2<sup>+</sup> binding to NCS-1 was studied by equilibrium dialysis method using 96-well micro-equilibrium dialysis system (HTDialysis, LLC) (Banker et al., 2003; Waters et al., 2008). Each well (500 µL) of the teflon block was separated by dialysis membrane (regenerated cellulose, 3.5 kDa MWCO). One half of each well was filled with 180 µL of (4.4–6.2) µM solution of NCS-1 in a buffer (10 mM Hepes-KOH, pH 7.6), whereas the other half contained 180 µL of the same buffer with (2–150) µM Zn(NO3)<sup>2</sup> or (2–50) µM Ca(NO3)<sup>2</sup> without NCS-1. The wells were tightly sealed and equilibrated by continuous shaking (130 rpm) of the block at (25.0 ± 0.5)◦C for 17–20 h. Total concentrations of Ca2+/Zn2<sup>+</sup> in the equilibrated solutions were measured by electrothermal atomization atomic absorption spectrometer iCE 3000 (Thermo Scientific), using argon as an inert gas. Zinc content was evaluated using the absorption bands at 213.9 or 307.6 nm and deuterium background correction. For calcium content estimates, the band at 422.7 nm and Zeeman background correction were used. The analytical signal was calibrated using AAS standard solutions for Ca2<sup>+</sup> (Sigma-Aldrich #69349) and Zn2<sup>+</sup> (Sigma-Aldrich #18827). Concentration of Ca2+/Zn2<sup>+</sup> bound to NCS-1 was estimated for each well as a difference between the total metal concentrations measured for both halves of the well, assuming that free Ca2+/Zn2<sup>+</sup> concentrations do not differ between the two halves of the well.

#### Analytical Gel-Filtration

Analytical gel-filtration of NCS-1 forms was carried out using fast protein liquid chromatography instrument as described for GCAP2 in (Olshevskaya et al., 1999) with modifications. NCS-1 (180 µM) was pre-incubated for 30 min at 37◦C in 50 mM Tris-HCl buffer (pH 8.0), 150 mM NaCl, 1 mM DTT containing either 1 mM EGTA or 1 mM Ca2<sup>+</sup> or 100 µM Zn2+. The obtained protein sample (200 µl) was loaded onto Superdex 200 10/300 GL column (GE Lifesciences) pre-equilibrated with the same buffer and eluted at 0.5 ml/min. Alternatively, gel-filtration was performed using high performance liquid chromatography instrument on Ultropack TSK G 2000 SW column (Pharmacia) at 1 ml/min.

#### Isothermal Titration Calorimetry (ITC)

Binding of divalent ions (Zn2+, Ca2+, and Mg2+) to NCS-1 was analyzed by ITC using MicroCal iTC200 instrument as described previously (Tsvetkov et al., 2012, 2013). Experiments were performed at 25◦C in 50 mM Tris-HCl buffer (pH 7.5) in the presence of 1 mM TCEP. Protein concentration in the calorimetric cell was 25 µM, whereas the concentration of ions in the syringe varied from 375 to 750 µM. In competition experiments the concentration of competitive ions (Zn2+, Ca2+, and Mg2+) in cell and syringe were 250 µM, 1 mM and 5 mM, respectively. NCS-1 was titrated by repeated injections of 2 µL aliquots of ions solution. If necessarily syringe was refilled with the same solution without cell refilling and the titration was

continued. Each resulting titration peak was integrated and plotted as a function of the NCS-1/ion molar ratio. The baseline was measured by injecting titrant into the protein-free buffer solution. Data were analyzed using Origin software and were fitted with "sequential binding," "one set of sites" or "two set of sites" models via a non-linear least squares minimization method and led to the determination of affinity constants (Ka), enthalpy changes (1H) and stoichiometry. Thermodynamic values are an average of at least three different experiments.

Binding of NCS-1 to D2R peptide was registered using MicroCal VP-ITC instrument, according to previously developed protocol (Pandalaneni et al., 2015) with modifications described in (Vladimirov et al., 2018). Experiments were performed at 25◦C in 20 mM Tris-HCl buffer (pH 8.0), 150 mM NaCl, 1 mM EDTA. Alternatively, the buffer contained 5 mM CaCl<sup>2</sup> instead of EDTA, with or without addition of 100 µM ZnCl2. Recombinant NCS-1 was dialyzed against the same buffer, and protein concentration of the stock solution was adjusted to 1 mM. Calorimetric cell contained 50 µM peptide, which was titrated by thirty 10 µl injections of NCS-1. Each injection was followed by 5 min stabilization phase. The resulting titration peaks were integrated and plotted as a function of the NCS-1/ion molar ratio. The baseline was measured by injecting the protein into the working buffer solution. Data were analyzed using Origin software and were fitted with "one set of sites" model. Thermodynamic parameters were determined as an average of at least three different experiments.

# Fluorimetry and Light Scattering (LS)

Fluorescence emission spectra of NCS-1 and bis-ANS were measured using Cary Eclipse spectrofluorimeter (Varian Inc.), equipped with a Peltier-controlled cell holder essentially as previously described (Baksheeva et al., 2015; Zernii et al., 2015). Tryptophan fluorescence of NCS-1 (14 µM) was excited at 280 nm and measured at 25◦C in 10 mM Hepes-KOH, 100 mM KCl, pH 7.6 buffer under various content of metal ions: either metal-free conditions (1 mM EDTA) or in the presence of Mg2<sup>+</sup> (1 mM MgCl2), Ca2<sup>+</sup> (100 µM CaCl2) or Zn2<sup>+</sup> (100 µM ZnCl2), or their combinations. Fluorescence of bis-ANS (1 µM) complexes with NCS-1 (5 µM) in the same buffer at 20oC was excited at 385 nm. All spectra were corrected for spectral sensitivity of the instrument and fitted to log-normal curves (Burstein and Emelyanenko, 1996) using LogNormal software (IBI RAS, Pushchino, Russia). Spectrofluorimetric temperature scans were performed at the average heating rate of 0.5◦C/min. The mid-transition temperatures for conversion from the native to the thermally denatured protein state were estimated from fits of the temperature dependencies of λmax by Boltzmann function using OriginPro 9.0 software (OriginLab Corporation, United States).

Alternatively, tryptophan fluorescence and thermal stability of NCS-1 in the presence of different concentrations of divalent ions were measured in 50 mM Tris-HCl, 1 mM TCEP, pH 7.5 buffer using differential scanning fluorimetry (DSF) instrument Prometheus NT.Plex (NanoTemper Technologies) equipped with LS module. NanoDSF grade capillaries were filled with 25 µM NCS-1 solution. Concentrations of Zn2+, Ca2<sup>+</sup> or Mg2<sup>+</sup> varied from 25 to 500 µM. The capillaries were loaded into the Prometheus NT.Plex instrument and the ratio of NCS-1 fluorescence emission intensities at 330 nm (I330) and 350 nm (I350) was registered at 25◦C at low detector sensitivity and excitation power of 10% (excitation wavelength of 280 nm). Then capillaries were heated from 15◦C to 95-110◦C at rate of 1 K/min. The unfolding mid-transition temperature (Tm) was determined from first derivative of the temperature dependence of the ratio, as implemented in Prometheus NT.Plex software. The temperatures of protein aggregation (Tagg) were determined from temperature dependences of the LS at 350 nm.

#### Circular Dichroism (CD)

Circular dichroism measurements were carried out with a JASCO J-810 spectropolarimeter (JASCO Inc., Japan), equipped with a Peltier-controlled cell holder as described in ref. (Permyakov et al., 2012). Briefly, CD spectra of NCS-1 (8 µM) were recorded at 25oC in 10 mM Hepes-KOH, 100 mM KCl, pH 7.6 buffer, either under metal-free conditions (1 mM EDTA) or in the presence of Mg2<sup>+</sup> (1 mM MgCl2), Ca2<sup>+</sup> (100 µM CaCl2) or Zn2<sup>+</sup> (100 µM ZnCl2), or their combinations. The secondary structure contents were estimated using CDPro software package (Sreerama and Woody, 2000).

#### Modeling and QM/MM Molecular Dynamics

To predict zinc binding sites in NCS-1, the major parameters for Zn2<sup>+</sup> coordination (distance and angle between the cation, coordinator and one of following atoms) were analyzed in 6327 structures available in PDB. The possible range of coordinators was defined as a list of the following types of atoms: SG, ND1, NE2, OD1, OD2, OE1, OE2, OG, OG1, OH and backbone oxygen (O). The maximal distance from the cation to chelator was limited at 3Å (Laitaoja et al., 2013). Based on these data, a search for possible Zn2+-binding sites was performed in X-ray structure of NCS-1 (PDB 5AEQ, Pandalaneni et al., 2015) starting from identification of tightly interconnected (can be defined as cliques in undirected graph) zinc coordinators, namely at least 3 atoms at distances less than 6Å. For every found coordinator, all possible positions of zinc were predicted yielding local density areas. The positions that fall within VdW radii of neighboring atoms were subtracted. The resulting putative Zn2+-binding areas were ranged according to maximum density, which was visualized as volumetric data in PyMol (DeLano, 2002).

Putative coordination of zinc in the identified areas of Ca2+-occupied EF-hands was assessed by evaluating cation coordination stability using QM/MM molecular dynamics simulations. Two best score metal binding sites at distance of more than 3Å from each other were selected in Ca2+-binding loops of all three EF-hands. The best site was loaded with Ca2+, whereas the second site was loaded with Zn2+. The resulting three systems were filled with TIP3P water with 0.1 M NaCl and the net charge was neutralized with additional ions. The solvated protein, Ca2<sup>+</sup> and Zn2<sup>+</sup> were positionally restrained and water together with sodium and chloride ions were equilibrated with molecular dynamics simulation for 100 ps. On the next step, each

simulation system was divided in molecular mechanics (MM) and quantum mechanics (QM) subsystems. MM subsystem was described with parameters from the parm99sb-ildn force field with corrections (Lindorff−Larsen et al., 2010). The QM subsystem was described utilizing DFTB approach (Grimme et al., 2010; Gaus et al., 2012) and defined as any atom including waters at distance less than 5Å from Zn2<sup>+</sup> or Ca2+. The coupling between MM and QM subsystems was performed using ONIOM approach (Dapprich et al., 1999). Linking atoms were introduced in single C-C bonds to preserve unsaturated structures in QM systems. The accordingly prepared systems were subjected to the QM/MM molecular dynamics simulation in NVT ensemble. The time step was set to 0.2 fs. Temperature coupling was performed with Velocity Rescale scheme (Bussi et al., 2007) allowing observing behavior of the systems at 300K. The total length of each simulation was set to 30 ps. All simulations were performed with GROMACS/DFTB package (Abraham et al., 2015; Kubaˇr et al., 2015).

#### Equilibrium Centrifugation Assay

The binding of NCS-1 to urea-washed photoreceptor membranes was performed according to the previously described procedure (Weiergräber et al., 2006; Senin et al., 2011) with some modifications. Briefly, 25 µM NCS-1 in buffer containing 20 mM Tris (pH 8.0), 150 mM NaCl and saturating concentration of MgCl<sup>2</sup> (20 mM), was mixed with the membranes in the absence and in the presence of 1 mM CaCl<sup>2</sup> and 0–100 µM ZnCl2, agitated in a thermostatic shaker for 15 min (37◦C, 1000 rpm) and centrifuged (24000 × g, 15 min). The pellet was dissolved in SDS-PAGE buffer and the rate of NCS-1 binding to membranes was measured by densitometric analysis of bands in polyacrylamide gel, using GelAnalyzer software<sup>1</sup> .

### Pull-Down Assay

Interaction of Zn2+-bound NCS-1 with GST-tagged N-terminal fragment of GRK1 (M1-G183) was monitored using analytical affinity chromatography (pull-down assay) (Zernii et al., 2011). Briefly, 50 µg of the fusion protein was immobilized on Glutathione Sepharose resin in 20 mM Tris (pH 8.0), 150 mM NaCl, 1 mM DTT. Next, 25 µM of NCS-1 was applied to the pellet. This suspension was incubated in a thermostatic shaker (1000 rpm) for 1 h at 4◦C in the presence of 1 mM CaCl<sup>2</sup> and 0–100 µM ZnCl2. After each incubation step non-bound protein was removed by washing the resin with the working buffer containing 0.05% Tween 20. Bound NCS-1 was eluted by SDS-PAGE sample buffer and analyzed by western blotting.

#### Precipitation Assay

Precipitation of NCS-1 was monitored in the mixture containing 25 µM protein, 20 mM Tris (pH 8.0), 150 mM NaCl, 1 mM DTT, and 0–500 µM ZnCl<sup>2</sup> with or without addition of 1 mM CaCl2. NCS-1 was incubated for 30 min at 37◦C with mild agitation, then precipitated protein was collected by centrifugation (24000 × g, 15 min) and the pellet was dissolved in SDS-PAGE sample buffer and analyzed by SDS-PAGE. The ratio of precipitated NCS-1 was estimated by densitometric analysis.

#### Transmission Electron Microscopy (TEM)

Four microliters of NCS-1 samples (10 µM) obtained in the presence of 1 mM ZnCl<sup>2</sup> were placed on carbon-coated copper grids (300 mesh) during 1 min. After having been blotted, grids were washed with distilled water, blotted again, negatively stained for 30 s with 2% (wt/vol) uranyl acetate. The grids were then dried and observed with a JEOL 2200FS transmission electron microscope (Tokyo, Japan) operating at 200 kV. Images were recorded using a 4k × 4k slow-scan CCD camera (Gatan, Inc., Pleasanton, United States).

# RESULTS

#### Stoichiometry of Zinc Binding to NCS-1

To test the hypothesis of the zinc interaction with NCS-1, we directly assessed the amount of zinc ions that can be bound per protein molecule using a micro-equilibrium dialysis system. To this end, a sample of recombinant myristoylated NCS-1 was prepared by Ca2+-dependent hydrophobic and ion exchange chromatographies and subjected, at 25◦C, to 20-h dialysis (MWCO of 3.5 kDa) against buffer containing different Zn2<sup>+</sup> concentrations. Zinc concentrations on both sides of the dialysis membrane were then measured by electrothermal atomization atomic absorption spectroscopy (AAS). The approximation of the resulting experimental data using the Hill equation (**Figure 1**) revealed half-maximal binding at 4.7 µM [Zn2+]free. Meanwhile, maximal stoichiometry of the zinc binding reached 1.5. The fractional stoichiometry may have been due to either the manifestation of an intermolecular zinc-binding site (which suggests NCS-1 multimerization) or the inaccessibility to zinc ions for some fraction of the protein molecules. Analytical gel filtration experiments did not reveal NCS-1 multimers in the presence of Zn2<sup>+</sup> (data not shown). Moreover, under these solution conditions, NCS-1 exhibited an even lower Stokes radius than in the presence of Ca2<sup>+</sup> or EGTA (which also confirms zincbinding to NCS-1). Therefore, we supposed that some fraction of the zinc-binding sites of NCS-1 remain shielded from zinc. Since AAS analysis of the NCS-1 sample revealed a calciumto-protein molar ratio of 0.42, we suggest that at least one of zinc-binding sites of NCS-1 overlapped with its active EF-hands. Therefore, we further decalcified NCS-1 samples by stepwise dialysis against EDTA, deionized water and a reaction buffer, as described earlier for the non-myristoylated protein (Blachford et al., 2009). As evidenced by AAS, this procedure decreased the fraction of residual calcium in NCS-1 sample down to 0.17, which means that the protein remains Ca2+-bound only by 5.6% of the saturation. Although the resulting protein sample (apo-NCS-1) bound Ca2<sup>+</sup> with a stoichiometry of 3 (data not shown), we failed to detect zinc binding to apo-NCS-1 by AAS, as its long-term incubation during equilibrium dialysis in the presence of Zn2<sup>+</sup> was accompanied by NCS-1 aggregation and accumulation on the dialysis membrane. Given this observation,

<sup>1</sup>http://www.gelanalyzer.com/

further characterization of the cation-binding properties of NCS-1 was performed by ITC. Yet, the AAS data represent direct evidence of zinc binding to NCS-1, revealing its dependence on calcium binding.

#### Thermodynamics of Calcium and Magnesium Binding to NCS-1

We employed ITC to determine Ca2+/Mg2+-binding parameters of the decalcified myristoylated NCS-1 (apo-NCS-1) sample, given that previous data on these properties were contradictory. Apo-NCS-1 (25 µM) was titrated by CaCl<sup>2</sup> in 50 mM Tris-HCl pH 7.5 buffer in the presence of 1 mM TCEP (**Figure 2A**, top panel). The use of "one set of sites" or "two sets of sites" models did not allow for a correct fit of the titration curve. Meanwhile, the experimental data were well fitted using the "sequential binding" model assuming three calcium sites (**Figure 2A**, bottom panel): the respective equilibrium association constants are 4.3 × 10<sup>6</sup> M−<sup>1</sup> , 2.0 × 10<sup>5</sup> M−<sup>1</sup> , and 3.5 × 10<sup>6</sup> M−<sup>1</sup> (**Table 1**). It should be noted that calcium binding to the two high-affinity sites was enthalpy-driven (**Table 1**), while calcium binding to the lower affinity site had an unfavorable enthalpy of 1.4 kcal/mol, indicating significant rearrangement of hydrophobic amino acids upon calcium binding to this site.

The ITC data on magnesium binding to apo-NCS-1 were well described by the "one set of sites" model (**Figure 2C**), revealing 2.7 Mg2<sup>+</sup> bound per protein molecule with an equilibrium association constant of 5.2 × 10<sup>5</sup> M−<sup>1</sup> (**Table 1**). Considering that apo-NCS-1 contained small fraction of residual calcium (0.17, see previous section), one can suppose that actual stoichiometry of Mg2<sup>+</sup> binding tends to 3. Indeed, magnesium ions compete with calcium for the same sites as no Mg2<sup>+</sup> binding was observed for NCS-1 saturated with Ca2<sup>+</sup> (1 mM). Consistent with this suggestion, in the presence of 5 mM Mg2<sup>+</sup> NCS-1 exhibited decreased affinity with calcium (**Table 1**). At the same time, the number of Ca2<sup>+</sup> bound per NCS-1 molecule in excess of magnesium reached 2.7, indicating that calcium completely replaced the bound magnesium ions.

#### Thermodynamics of Zinc Binding to NCS-1

The apo-form of myristoylated NCS-1 (25 µM) in 50 mM Tris-HCl pH 7.5 buffer, in the presence of 1 mM TCEP, was titrated by ZnCl<sup>2</sup> using ITC (**Figure 2B**). The fitting of the resulting titration curve using the "two sets of sites" model indicated that NCS1 bound Zn2<sup>+</sup> in two equal high-affinity sites and one low-affinity site. The corresponding equilibrium association constants were calculated as 9.2 × 10<sup>6</sup> M−<sup>1</sup> and 2.3 × 10<sup>5</sup> M−<sup>1</sup> , respectively (**Table 1**). The saturation of NCS-1 with Ca2<sup>+</sup> or Mg2<sup>+</sup> abolished the low-affinity Zn2<sup>+</sup> binding, while zinc affinity of the other two sites was almost unaffected. Meanwhile, the enthalpy changes (1H) accompanying zinc interaction with these forms differed: 1H was negative for Zn2<sup>+</sup> binding to apo- and Mg2+-loaded NCS-1, and positive for Zn2<sup>+</sup> binding to Ca2+-loaded NCS-1, thereby reflecting significant conformational differences between Mg2+- and Ca2+-loaded NCS-1 states. Finally, Zn2+-loaded NCS-1 was unable to bind magnesium, but coordinated 1.5 Ca2<sup>+</sup> ions per protein molecule with increased affinity of the first site (K<sup>a</sup> = 5.9 × 10<sup>7</sup> M−<sup>1</sup> ).

FIGURE 2 | Thermodynamics of calcium, zinc and magnesium binding to NCS-1. Typical ITC curves (upper panels) and binding isotherms (lower panels) representing titration of NCS-1 (25 µM) by different cations. (A) Binding of Ca2+. (B) Binding of Zn2+. (C) Binding of Mg2+. (D) Binding of Ca2<sup>+</sup> in the presence of 250 µM Zn2+. (E) Binding of Zn2<sup>+</sup> in the presence of 1 mM Ca2+. (F) Binding of Mg2<sup>+</sup> in the presence of 250 µM Zn2+. (G) Binding of Ca2<sup>+</sup> in the presence of 5 mM Mg2+. (H) Binding of Zn2<sup>+</sup> in the presence of 5 mM Mg2+. (I) Binding of Mg2<sup>+</sup> in the presence of 1 mM Ca2+. Best fits are shown as solid curves (see Table 1).

Overall, the different modes of zinc binding to NCS-1, as revealed by our data, indicate that structural and functional consequences of this interaction depend on NCS-1 conformation. Thus, the ITC data argue for the existence of Mg2+-bound, Ca2+-bound and Zn2+-bound conformers of NCS-1 as well as its Zn2+(Mg2+)-bound, Zn2+(Ca2+)-bound and Ca2+(Zn2+) bound forms, where Zn2<sup>+</sup> or Ca2<sup>+</sup> are bound to the background of the excess of Mg2+, Ca2<sup>+</sup> and Zn2+, respectively (hereinafter the cation that is taken in excess is indicated in parentheses).

#### Conformational Properties of Zinc-Bound NCS-1

The ITC experiments suggested the existence of several distinct states of NCS-1 with two different metal ions bound simultaneously. To explore structural differences between these NCS-1 states, we measured the intrinsic fluorescence spectra of NCS-1 in the presence of various combinations of the metals studied, which enabled examination of the mobility and polarity of the microenvironment of Trp30 and Trp103 residues located in N- and C-terminal domains of the protein.

The fluorescence spectra of 15 µM NCS-1 were measured either under metal-free conditions (1 mM EDTA) or in the presence of 1 mM Mg2+, 0.1 mM Ca2<sup>+</sup> or 0.1 mM Zn2+, or their combinations. Apo-NCS-1 exhibited a characteristic tryptophan fluorescence emission spectrum with a maximum position (λmax) at 338 nm (**Figures 3A,B**). Mg2<sup>+</sup> binding significantly increased the maximal intensity (Imax) of the fluorescence emission spectrum of NCS-1 without affecting the maximum position. Calcium binding to NCS-1 increased its Imax value and shifted its λmax to 334 nm, indicating movement of the emitting Trp residue(s) to a less polar and/or mobile environment. Zinc binding to apo-protein increased Imax without affecting λmax resembling the effect of magnesium in this respect (**Figures 3A,B**). Meanwhile, the presence of zinc only moderately affected the fluorescence spectra of Mg2+- and Ca2+-saturated NCS-1, indicating minor structural rearrangements near to the emitting Trp residue(s) under these experimental conditions.

Apo-NCS-1 (8 µM) represented far-UV circular dichroism (CD) spectra typical for an α-helical fold with characteristic minima at 208 nm and 222 nm (**Figure 3C**). Binding of all examined cations, including zinc, to NCS-1 was accompanied by a similar increase in its α-helical content and a decrease in the content of β-sheets and unordered regions (**Figure 3D** and **Supplementary Table S1**). Importantly, NCS-1 contains only two short antiparallel β-sheets, which connect the Ca2+ binding loops of EF1-EF2 and EF3-EF4 pairs of EF-hand motifs (Heidarsson et al., 2012; Pandalaneni et al., 2015). Therefore, the revealed similar changes in the β-structure content of NCS-1 upon binding of Ca2+/Mg2<sup>+</sup> and Zn2<sup>+</sup> suggest that zinc binds to sites overlapping with the EF-hand loops. It should be noted that single and double ion-bound NCS-1 forms also exhibited certain differences in their secondary structure (**Figure 3D** and **Supplementary Table S1**). Thus, Ca2+-bound and Zn2+(Ca2+)-bound forms, as well as Mg2+-bound and Zn2+(Mg2+)-bound forms, were identical in α-helical content, but differed in β-structure content 1.46-fold and 1.27-fold, respectively (**Supplementary Table S1**). These data suggest that zinc binding to Ca2+-saturated or Mg2+-saturated NCS1 does not significantly alter its overall secondary structure, but still affects its EF-hands.

To gain further insight into conformational differences between single and double ion-bound forms of NCS-1, we measured concentration dependencies of the ratio of its fluorescence intensities at 350 nm and 330 nm (I350/I330) for apo-NCS-1 or NCS-1, saturated by Ca2+, Mg2<sup>+</sup> or Zn2+. In the presence of increasing concentrations of Mg2+, Ca2<sup>+</sup> or Zn2+, a gradual reduction in the I350/I<sup>330</sup> ratio of NCS-1 was observed, confirming interaction of the protein with these cations (**Figure 4A**). The decrease in the ratio was most pronounced for Ca2<sup>+</sup> ions, suggesting that Ca2+-loaded NCS-1 is structurally different from its Mg2+-bound or Zn2+-bound states (**Figure 4A**). In the case of Mg2+-saturated NCS-1, low concentrations of Zn2<sup>+</sup> decreased the I350/I<sup>330</sup> ratio, while, at a more than threefold molar excess of zinc, the ratio increased toward the level of the Zn2+-bound form (**Figure 4B**), probably


TABLE 1 | Thermodynamic parameters of zinc, calcium and magnesium binding to NCS-1 in 50 mM Tris-HCl buffer (pH 7.5) in the presence of 1 mM TCEP at 25◦C, estimated from ITC data (see Figure 2).

<sup>∗</sup>Data were fitted using "sequential binding" model. ∗∗Data were fitted using "two sets of sites" model. The concentrations of competitive ions Zn2+, Ca2+, and Mg2<sup>+</sup> were 250 µM, 1 mM, and 5 mM, respectively.

reflecting the replacement of magnesium by zinc (see previous section). In contrast, the binding of Zn2<sup>+</sup> to Ca2+-NCS-1 produced a highly moderate increasing effect on the ratio (**Figure 4C**). Monitoring of the I350/I<sup>330</sup> ratio for Zn2+-saturated NCS-1 in the presence of increasing calcium levels revealed signs of the Ca2+-bound-like conformation of the protein at lower calcium concentrations than in the case of Ca2<sup>+</sup> binding to apo-NCS-1 (**Figure 4D**), which agreed with the increased Ca2<sup>+</sup> affinity of Zn2+-saturated protein (see **Table 1**). It should be noted that differences in conformational changes induced by zinc binding to apo, Mg2+-saturated and Ca2+-saturated NCS-1 were the most striking at low zinc levels, when they were likely correlated with the stoichiometry of the metals' binding. Thus, the I350/I<sup>330</sup> ratio for apo-NCS-1 decreased even in the case of a onefold excess of zinc (one Zn2<sup>+</sup> bound), while the same value for Ca2+-NCS-1 exhibited a moderate increase only when Zn2<sup>+</sup> concentration exceeded the protein concentration by three times (two Zn2<sup>+</sup> bound). In contrast, the fluorescence of Mg2+-NCS-1 remained unchanged until reaching a 2.5-fold excess of zinc, whereas the further elevation of Zn2<sup>+</sup> concentration resulted in a sequential decrease (presumably one Zn2<sup>+</sup> and two Mg2<sup>+</sup> bound to the protein) and an increase (presumably two Zn2<sup>+</sup> and one Mg2<sup>+</sup> bound to the protein) in the ratio.

Summing up, the spectral measurements reveal certain structural differences between apo, Mg2+-bound, Ca2+-bound, Zn2+-bound, Zn2+(Mg2+)-bound, Zn2+(Ca2+)-bound and Ca2+(Zn2+)-bound conformers of NCS-1.

#### Thermal Stability of NCS-1 in the Presence of Zinc

Thermal unfolding of NCS-1 is accompanied by a red shift in its tryptophan fluorescence spectrum, implying that λmax can be used for monitoring thermal denaturation of the protein (Baksheeva et al., 2015). We compared the thermal unfolding profiles of NCS-1 (15 µM) in the presence of 1 mM EDTA, 1 mM Mg2+, 100 µM Ca2+, 100 µM Zn2<sup>+</sup> or their combinations. Inspection of the experimental curves revealed that apo-NCS-1 was relatively unstable with a mid-transition temperature (Tm) of 40◦C, whereas, in the presence of magnesium, T<sup>m</sup> increased up to 70◦C (**Figure 5A**). In the presence of Zn2+, or Mg2<sup>+</sup> and Zn2+, NCS-1 exhibited similar temperature profiles without a clear transition over the experimental temperature range (**Figure 5A**). In both cases, the dispersion of λmax values observed at temperatures above 60oC indicated protein aggregation. Ca2+ saturated NCS-1 demonstrated blue-shifted emission spectra and a T<sup>m</sup> value exceeding 80◦C (**Figure 5B**). Meanwhile, zinc binding

to calcium-loaded NCS-1 shifted the thermal transition of the protein toward lower temperatures, thereby reflecting structural differences between Ca2+-bound and Zn2+(Ca2+)-bound NCS1.

More information was obtained upon monitoring NCS-1 (25 µM) denaturation by registering temperature dependences of the I350/I<sup>330</sup> ratio at different excesses of the cations. A nanoDSF instrument was used since it allows us to monitor, in parallel, the aggregation of the protein by measuring the LS of the sample at 350 nm upon heating. The binding of any of the three cations to apo-NCS-1 increased the stability of the protein, but with different efficacy and within different concentration ranges (**Figure 6A**). Indeed, in the presence of the fourfold excess of calcium (100 µM), the T<sup>m</sup> of the protein increased to > 80◦C, while, in the case of the same Mg2<sup>+</sup> and Zn2<sup>+</sup> concentrations, the increase was moderate (48 and 42◦C, respectively). Interestingly, the use of higher calcium or magnesium concentrations further improved protein stability without affecting the aggregative state, whereas zinc, at more than a fourfold excess, increased susceptibility of the protein to aggregation as indicated by LS (**Figure 6C**). The binding of zinc to Mg2+-NCS-1 had no effect on its stability until a 2.5 fold excess of Zn2<sup>+</sup> was used (**Figure 6B**). At this point, the denaturation temperature increased to 80◦C and then began to drop, apparently reflecting the formation of Zn2+/2Mg2<sup>+</sup> NCS-1 intermediate and 2Zn2+/Mg2<sup>+</sup> NCS-1 conformer, respectively (see above). The drop was associated with a reduction in aggregation temperature (Tagg), indicating increased propensity of the protein to aggregation (**Figure 6D**). We were technically unable to monitor the impact of low zinc concentrations (one- to fivefold excess) on Ca2+-saturated NCS-1 (1 mM Ca2+) as T<sup>m</sup> of the latter exceeded 90oC, which is beyond the detection limit of the method (**Figure 6E**). Yet, at higher levels, zinc produced a gradual destabilizing effect on NCS-1 and enhanced its susceptibility to aggregation (**Figures 6E,G**, see also **Supplementary Figure S2A**). Finally, the presence of Ca2<sup>+</sup> inhibited aggregation of Zn2+-saturated NCS-1 and increased its thermal stability as soon as the protein bound the first calcium ion (onefold excess of Ca2+). However, at high calcium concentrations, T<sup>m</sup> did not exceed 78oC, indicating that the resulting NCS-1 conformer represents a Ca2+(Zn2+) bound form rather than a Ca2+-saturated form of the protein (T<sup>m</sup> > 90oC) (**Figures 6F,H**).

Overall, at low zinc concentrations, corresponding to full saturation of Zn2+-binding sites in each of the NCS-1 forms, the binding of the cation slightly destabilizes Ca2+-loaded NCS1 and enhances the thermal stability of Ca2+-free protein. Meanwhile, upon elevation of Zn2<sup>+</sup> levels, all NCS-1 forms become gradually destabilized and prone to aggregation. Ca2+-NCS-1 is the most resistant to the destabilizing effects of high zinc. Consistently, the binding of calcium to Zn2+-saturated NCS-1 improves its structure by forming a relatively stable Ca2+(Zn2+)-bound conformer.

# Putative Zinc-Binding Sites in Calcium-Saturated NCS-1

The results of spectroscopic and thermal stability studies revealed the existence of a Zn2+(Ca2+)-bound form of NCS-1, which structurally differs from the well-recognized Ca2+-saturated conformer of the protein. Therefore, we next attempted to predict Zn2+-binding site locations in Ca2+-NCS-1 in silico, based on the available crystal structure of this form of the protein [PDB 5AEQ (Pandalaneni et al., 2015)]. Considering the averaged geometry of zinc coordination in all Zn2+-binding proteins presented in PDB, the density of Zn2+-binding probability in the NCS-1 structure was built in grid with a step of 0.1 Å (**Figure 7A**). It was found that areas with the required number of chelating groups for Zn2<sup>+</sup> was located only in the loops of three functional EF-hand sites, namely EF2 (the highest score), EF3 and EF4. Interestingly, the size of these areas was around 4.5 Å, suggesting that they could simultaneously accommodate calcium and zinc ions. Furthermore, such a configuration would compensate for a negative charge (-2 in EF2, -1 in EF3 and EF4), which remained in EF hand loops upon binding of single Ca2+. In order to check this suggestion, we performed QM/MM simulations of molecular dynamics associated with Zn2<sup>+</sup> binding in each Ca2+-occupied EF-hand motif. It was found that, in EF2, the number of coordinators around calcium ions in the presence of zinc decreased from seven to six, but most of the metalchelating residues of the loop (Asp 73, Asn 75, Asp77, Arg79, Glu 81), as well as a water molecule, remained involved in the binding. In this case, the coordination of zinc was maintained by four oxygen atoms from Asp 73 (two atoms, from α-carbonyl and β-carboxyl groups), Asn 75 and Glu 84 (**Figure 7B**). In EF3, calcium lost three chelators coordinated by Asp109, Aps 111, Glu120 and a water molecule, whereas zinc possessed less favorable coordination due to three oxygen atoms from Tyr115, Asp 109 and Asp 111 (**Figure 7C**). As for EF4, both cations bound simultaneously in relatively optimal coordination. Thus, calcium was chelated by Glu168, Asp157, Asn159, Asp161, Lys163 (backbone) and a water molecule, whereas zinc was bound to Met156 (backbone), Asp157, Glu168 and a water molecule (**Figure 7D**). Taken together, these data indirectly support our suggestion of zinc coordination in EF2, EF3 and EF4 in apo and Mg2+-bound NCS-1, as well as provide a rationale for the prediction of Zn2+-binding sites in the second and fourth EF-hands of the Ca2+-saturated form of the protein.

#### Functional Properties of NCS-1 in the Presence of Zinc

Previous in vitro and in vivo studies revealed that NCS-1 can regulate a number of targets including membrane-associated proteins. Consistently, an important feature of NCS-1 is its Ca2+ induced interaction with cellular membranes via the N-terminal myristoyl group of the protein (Baksheeva et al., 2015). Thus, we next explored whether zinc binding affects the affinity of NCS-1 to photoreceptor membranes. Among the detected forms of protein, we focused on the Zn2+(Mg2+)-bound and Zn2+(Ca2+)-bound NCS-1 conformers as they might dominate under physiological conditions. According to the data from the modified equilibrium centrifugation assay, Ca2+-saturated NCS-1 (25 µM) at 25oC in 20 mM Tris-HCl pH 8.0 buffer, bound to urea-washed photoreceptor membranes, and the binding decreased approximately twofold in the case of Ca2+-free/Mg2+ saturated NCS-1. Meanwhile, the presence of 0–100 µM Zn2<sup>+</sup> did not affect the membrane association of both NCS-1 forms (**Supplementary Figure S1**).

In order to further address the possible effects of zinc on the functional activity of NCS-1, we monitored the interaction of the protein, with D2R and GRK1 representing its well-established Ca2+-dependent targets (Pandalaneni et al., 2015). Firstly, the binding of NCS-1 to the complementary D2R peptide N430-R443 (50 µM) was monitored at 25oC in 20 mM Tris-HCl pH 8.0 buffer using ITC. Without calcium, no interaction between NCS-1 and the peptide was registered, regardless of the presence of zinc (data not shown). Meanwhile, Ca2+-loaded NCS-1 bound two moles of D2R peptide with a dissociation constant of 30.12 µM (**Figure 8A**). Remarkably, in the presence of zinc, Ca2+-NCS-1

interacted with D2R peptide with the same stoichiometry, but with a 3.5-fold increase in affinity (**Figure 8B** and **Table 2**). Similar observations were made upon monitoring the interaction of NCS-1 (25 µM) with N-terminal domain of GRK1 (M1-G183), fused with glutathione-S-transferase (GST-N-GRK1) at 25oC in 20 mM of Tris-HCl pH 8.0 buffer, by means of a pull-down assay. Thus, 1 mM Ca2<sup>+</sup> GST-N-GRK1 was bound to NCS-1 and the binding was enhanced twofold in the presence of 25 µM of Zn2<sup>+</sup> (**Figure 8C**). Interestingly, a further increase in zinc concentration resulted in the gradual destabilization of the NCS-1-GRK1 complex.

NCS-1 is known to interact with D2R and GRK1 via hydrophobic sites, which become available in response to

TABLE 2 | Thermodynamic parameters of binding of D2R peptide to NCS-1 in 20 mM Tris-HCl buffer (pH 8.0), 150 mM NaCl, 5 mM CaCl<sup>2</sup> in the presence or in the absence of 100 µM ZnCl2.


Ca2<sup>+</sup> binding (Pandalaneni et al., 2015). Therefore, we next investigated effects of zinc on the accessibility of such sites in Mg2+-saturated and Ca2+-saturated NCS-1, using fluorescent dye bis-ANS. The interaction of bis-ANS with hydrophobic cavities of a protein is accompanied by increased intensity and a shift in the λmax of the fluorescence of the dye. It was found that, in the presence of zinc, bis-ANS binding to both Mg2+-saturated and Ca2+-saturated NCS-1 was moderately enhanced, suggesting increased surface hydrophobicity of these forms (**Figure 8D**). Thus, it is this effect that may partially account for the increased affinity of Zn2+(Ca2+)-bound NCS-1 to D2R and GRK1.

Taken together, our data demonstrate that, at low physiological concentrations, zinc cannot substitute calcium in relation to NCS-1 activation; rather, it affects the structure and stability of Ca2+-saturated protein, thereby improving its normal functionality.

#### Abnormal Behavior of NCS-1 in the Presence of Excessive Zinc Concentrations

Although the estimated intracellular concentration of free zinc is considerably low, it is entirely possible that, under certain pathological conditions, it can abnormally increase. As such, we further analyzed behavior of different forms of NCS-1 in the presence of "pathological" amounts of zinc. According to

titration of 50 µM D2R peptide with 5–150 µM NCS-1 in the presence of 5 mM Ca2<sup>+</sup> (A) or 5 mM Ca2<sup>+</sup> and 100 µM Zn2<sup>+</sup> (B). (C) Binding of 25 µM NCS-1 to GST-N-GRK1 at 1 mM Ca2<sup>+</sup> in the presence of 0, 25, 50, or 75 µM Zn2<sup>+</sup> (i.e., at [Zn2+]/[NCS-1] ratio of 0-3), monitored by pull-down assay. (D) Representative fluorescence spectra of bis-ANS (1.2 µM) and NCS-1 (5 µM) complexes formed in the presence of either 1 mM Mg2<sup>+</sup> or 100 µM Ca2<sup>+</sup> with or without addition of 100 µM Zn2+.

LS data, the susceptibility of the protein to aggregation, in the presence of high zinc concentrations, decreased in the following order: apo-NCS-1 > Mg2+-bound NCS-1 > Ca2+ bound NCS-1 (**Supplementary Figure S2A**). For apo and Mg2+ bound NCS-1, the decrease in the temperature of aggregation started when Zn2<sup>+</sup> concentration exceeded the concentration required for full saturation of the protein by 25 µM. In contrast, Ca2+-bound NCS-1 can sustain up to 100 µM free Zn2+. At physiological temperatures, the signs of aggregation of apo and Mg2+-bound NCS-1 were observed at 300–500 µM free Zn2<sup>+</sup> (data not shown). Meanwhile, in the presence of calcium, no NCS-1 aggregation was detected under these conditions. Since aggregation includes multimeric associations, which can produce insoluble precipitates of the protein, we also monitored the Zn2+-induced precipitation of Mg2+-loaded and Ca2+-loaded NCS-1 (25 µM) at 25◦C (**Supplementary Figure S2B**). The formation of NCS-1 precipitates was initiated at 200 and 325 µM of free zinc for Mg2+-loaded and Ca2+-loaded NCS-1 forms, respectively. To visualize the shape and arrangement of the insoluble NCS-1 conglomerates formed in the presence of zinc, we further examined the respective protein precipitates by means of TEM. As can be seen from TEM data (**Supplementary Figures S2C,D**), Zn2+-bound NCS-1 constitutes fibrilic twisted ropelike structures resembling the aggregates of another neuronal protein, TDP-43, found in the presence of zinc ions (Garnier et al., 2017).

We concluded that, at high concentrations, zinc might bind to NCS-1 non-specifically, thereby deteriorating the structure of the

protein and promoting its aggregation and precipitation, which are most prominent in the absence of calcium.

#### DISCUSSION

Previous in vitro studies reported the existence of three major forms of NCS-1 in terms of metal binding, namely, apo, Mg2+-bound and Ca2+-bound. Meanwhile, the data concerning the stoichiometry and affinity of calcium binding to the protein are contradictory. Thus, according to flow dialysis, nonmyristoylated NCS-1 (nNCS-1) cooperatively binds two calcium ions with nanomolar and micromolar affinities (Cox et al., 1994). Subsequent ITC experiments also suggested the binding of two Ca2+, but in a non-cooperative manner and with a dissociation constant of 1.8 µM for both sites (Jeromin et al., 2004). Meanwhile, refinement of the data using NMR studies revealed that nNCS-1 actually coordinates three calcium ions in EF2-EF4 (Chandra et al., 2011; Heidarsson et al., 2012). Myristoylated NCS-1 (mNCS-1) was reported to bind three calcium ions. However, two sets of ITC studies conducted by the same authors report different modes of calcium binding, which likely depend on the preparation of the protein samples and the model applied for fitting of the ITC data (Jeromin et al., 2004; Aravind et al., 2008). Thus, in the first study, the use of a "three sequential binding sites" model revealed binding constants of a micromolar, nanomolar and submicromolar order (Jeromin et al., 2004), whereas, in the second study, the "two sets of sites" model was applied, which allowed for identifying two similar sites with submicromolar affinity and one site with nanomolar affinity (Aravind et al., 2008). The ITC data on Ca2<sup>+</sup> binding to mNCS-1, obtained in our current study, are generally in agreement with the data reported by Jeromin et al. (2004) including the revealed positive enthalpy of Ca2<sup>+</sup> binding to the low-affinity site. Thus, we confirmed different calcium affinities of three EF-hands of mNCS-1 and the sequential mode of their filling (**Figure 2** and **Table 1**). Such a mechanism agrees with previous NMR studies, according to which Ca2+-binding sites become occupied in the following order EF2→EF3→EF4 (Chandra et al., 2011). Considering the evaluations of the Ca2<sup>+</sup> affinity of individual EF hands reported by Chandra et al., we can attribute calciumbinding constants K<sup>A</sup> 1 , K<sup>A</sup> 2 and K<sup>A</sup> 3 , as calculated in the current work (**Table 1**), to EF3, EF2 and EF4, respectively.

In early magnesium binding experiments, nNCS-1 exhibited the non-cooperative coordination of two Mg2<sup>+</sup> with a dissociation constant of 12 µM (Cox et al., 1994). Similar findings were reported for mNCS-1, based on ITC, NMR and mutagenesis studies (Aravind et al., 2008). According to our ITC data, the amount of Mg2<sup>+</sup> bound to mNCS-1 tends toward three (**Table 1**). We speculate that such stoichiometry is a specific feature of myristoylated protein, where Mg2<sup>+</sup> binds to EF2-EF4. It should be noted that the actual amount of protein-associated magnesium, which binds with low affinity, might be highly sensitive to the quality of the protein sample (i.e., the content of the nNCS-1 admixture or residual calcium) and may therefore be differently evaluated. Yet, all three studies including ours agree that Mg2<sup>+</sup> antagonizes Ca2<sup>+</sup> binding by reducing the affinity of the respective sites of the protein. These data confirm competition between the ions for the same binding sites with a preference for calcium (Cox et al., 1994; Aravind et al., 2008).

Our brand-new finding is that myristoylated NCS-1 is capable of coordinating up to three zinc ions. The mechanism of zinc binding to the protein and the exact Zn2+-binding sites are yet to be determined. For the moment, based on our metal competition analysis, CD studies and molecular modeling, we can hypothesize that zinc binds to functional EF-hands of the protein. Indeed, the ability of EF-hands to coordinate Zn2<sup>+</sup> was previously reported for another ubiquitous Ca2+-binding protein calmodulin by X-ray crystallographic studies (Warren et al., 2007). Based on the analysis of all Zn2+-binding proteins presented in PDB, we found that, in NCS-1, the density of chelating groups required for Zn2<sup>+</sup> binding is located only in the loops of EF2 (the highest score), EF3 and EF4 (**Figure 7A**). According to our CD measurements, the interaction of zinc with apo-NCS-1 induces a decrease in the content of β-sheets and an increase in α-helical content, exactly as in the case of the binding of Ca2<sup>+</sup> or Mg2<sup>+</sup> to EF-hands (**Figure 3D** and **Supplementary Table S1**). Since NCS-1 contains only two short antiparallel β-sheets, which connect Ca2+-binding loops of EF1-EF2 and EF3-EF4 (Heidarsson et al., 2012; Pandalaneni et al., 2015), one can suggest that zinc binds to EF-hands of the protein. This conclusion is further supported by our ITC data, indicating that Zn2+-saturated NCS-1 does not bind magnesium and exhibits reduced stoichiometry of Ca2+-binding (**Figure 2** and **Table 1**). Interestingly, Ca2<sup>+</sup> binding to one of the sites in Zn2+-saturated NCS-1 is one order of magnitude higher in affinity than any of the sites in apo-NCS-1 (**Table 1**). Given the proposed model for sequential filling of the mNCS-1 by calcium in the order EF2→EF3→EF4 (Chandra et al., 2011), we hypothesize that Zn2+-bound EF2 may adopt a conformation that facilitates the binding of calcium to the remaining two sites. Consistently, Zn2+-bound EF-hands of calmodulin resembled an intermediate state in the chain of conformational transitions induced by Ca2+-binding (Warren et al., 2007).

The unique mode of zinc binding to Ca2+-saturated NCS-1 is predicted by QM/MM simulations of the associated molecular dynamics, based on the crystal structure of the respective NCS-1 form [PDB 5AEQ (Pandalaneni et al., 2015)]. In the absence of zinc, EF3 possesses the most favourable environment for the coordination of calcium among EF-hands of the protein, which agrees with its maximal affinity for Ca2<sup>+</sup> (Chandra et al., 2011). At the same time, coordination of both cations in this site seems unlikely due to the absence of the required number of chelating groups (**Figure 7C**). Therefore, EF3 can bind strictly to one ion with a preference for calcium and the latter can replace zinc from the site but not vice versa. In contrast, EF2 and EF4 can accommodate both Ca2<sup>+</sup> and Zn2+, at least under our in silico conditions. In both cases, calcium loses one chelator, in turn becoming coordinated by six oxygen atoms (**Figures 7B,D**). Yet, such a configuration is common for proteins (Pidcock and Moore, 2001). Furthermore, such a configuration would completely compensate for the high negative

charge in the EF2 (-2 in Ca2+-bound NCS-1). Thus, based on these observations, we suggest that Ca2+-saturated NCS-1 can accommodate up to two zinc ions, one in EF4 and the other one in EF2.

The proposed binding modes for Zn2<sup>+</sup> and Ca2<sup>+</sup> are generally in agreement with our ITC and spectroscopic data. Thus, Ca2+ saturated NCS-1 coordinates one or two zinc ions (**Table 1**), apparently in terms of EF4/EF2 yielding the Zn2+(Ca2+)-bound protein form, which does not significantly differ from the "open" Ca2+-bound conformer in the overall protein fold (**Figure 4C**) but possesses enhanced thermal stability (**Figures 5B**, **6E–H**). In contrast, Zn2+-saturated NCS-1 coordinates two calcium ions (**Table 1**) yielding a Ca2+(Zn2+)-bound conformer. In this case, EF2 likely remains occupied by zinc, which could facilitate calcium binding to EF3 (and consequently to EF4) as suggested by the absence of a low-affinity Ca2+-binding site and an increased binding constant for the high-affinity site in the ITC data (**Table 1**). At the same time, calcium replaces zinc from EF3 and could replace or temporary co-reside with zinc in EF4. It cannot be ruled out, however, that one of the EFhands, being occupied with zinc, might adopt a conformation that is favorable for calcium binding, thereby exhibiting increased Ca2<sup>+</sup> affinity as seen in our ITC studies. In any case, the resulting Ca2+(Zn2+)-bound conformer possesses only a small difference with the Zn2+(Ca2+)-bound from of the protein in the I350/I<sup>330</sup> ratio (**Figures 4C,D**), but significantly differs from it in thermal stability (> 20◦C, **Figures 6E,F**). It should be emphasized that, despite being highly consistent with the experimental and literature data, the above mechanisms of Zn2+/Ca2<sup>+</sup> binding are mostly speculative and require additional confirmations.

In the aggregate, our in vitro studies suggest the existence of Zn2+-bound, Zn2+(Mg2+)-bound, Zn2+(Ca2+)-bound and Ca2+(Zn2+)-bound conformers of NCS-1 in addition to previously recognized apo, Mg2+-bound and Ca2+-bound forms of the protein. It should be mentioned that structural differences between the two latter forms, as observed in this study, are generally in accord with the reported data. Thus, the binding of both Mg2<sup>+</sup> and Ca2<sup>+</sup> increases the α-helical content of NCS-1, whereas only Ca2<sup>+</sup> binding notably increases its surface hydrophobicity, as originally described by Jeromin et al. (2004). In addition, Mg2<sup>+</sup> binding induced a more pronounced increase in the intensity of intrinsic fluorescence of the protein than Ca2<sup>+</sup> binding, in agreement with previous observations (Aravind et al., 2008). It has been suggested that the unique mode of Ca2+/Mg2<sup>+</sup> binding and resulting structural alterations govern the target recognition by NCS-1. Indeed, the NCS-1, preloaded with Mg2+, binds D2R in response to Ca2<sup>+</sup> elevation more efficiently when compared to apo-protein, indicating that magnesium can serve as a physiological co-factor with calcium in this interaction (Woll et al., 2011).

Alongside NCS-1, the Ca2+/Mg2<sup>+</sup> interplay was shown to regulate the structure and function of the other NCS proteins belonging to all five classes of the NCS family. Interestingly, the mechanisms of this regulation are quite distinct. Thus, magnesium and calcium bind to different EF-hand motifs of these proteins and the binding differently affects their functional specificity. In recoverin, Mg2<sup>+</sup> binds to functional EF2 and EF3, which reduces the Ca2<sup>+</sup> affinity of the protein (at high magnesium concentration), but only slightly affects its secondary and tertiary structure, does not lead to activation of its myristoyl switch and is not required for its interaction with GRK1 (Ozawa et al., 2000; Ames et al., 2006; Marino et al., 2015). The cooperative sequential binding of calcium to the EF3 and EF2 of recoverin increases its thermal stability and α-helical content, as well as leads to exposure of its myristoyl group and hydrophobic pocket residues, thereby providing the protein with a capability to interact with membranes and GRK1 (Zozulya and Stryer, 1992; Ames et al., 1997, 2006; Permyakov et al., 2000; Zernii et al., 2015). A similar mechanism of Ca2<sup>+</sup> binding (cooperative binding to EF2 and EF3), structural alterations (increase in α-helical content in the presence of Ca2<sup>+</sup> but not Mg2+) and a Ca2+-myristoyl switch were recognized in the case of another NCS protein, VILIP1. However, unlike recoverin, VILIP1 only coordinates magnesium in EF3 with a relatively high affinity (K<sup>D</sup> = 20 µM), suggesting the functional significance of this complex. Furthermore, VILIP-1 forms a stable dimer, which is not dependent on Ca2<sup>+</sup> or Mg2+, but seems to be required for proper target recognition (Jheng et al., 2006; Li et al., 2011). In the proteins belonging to another class of the NCS family, GCAPs, Mg2<sup>+</sup> and Ca2<sup>+</sup> play a crucial role in tuning their activity toward target enzymes, i.e., retinal guanylate cyclases (GCs). For instance, in GCAP1, Mg2<sup>+</sup> binds to EF2 with micromolar affinity (EF3 and EF4 exhibit only low affinity with the cation) and the binding stabilizes a tertiary structure of the protein, which otherwise represents a molten globule incapable of regulating GCs (Lim et al., 2009; Dell'Orco et al., 2010). Thus, the presence of magnesium in EF2 is necessary for maintaining a GC-activator state of GCAP1 (Peshenko and Dizhoor, 2004; Lim et al., 2016). Calcium binds to EF2, EF3 and EF4 of GCAP1 in a non-cooperative manner, which drastically increases the thermal stability of the protein without altering its secondary structure and triggering exposure of its myristoyl group (Lim et al., 2009; Marino et al., 2015). Instead, the binding converts GCAP1 into a GC-inhibitor state by inducing local conformational changes via the Ca2+-myristoyl tug mechanism (Peshenko et al., 2012; Lim et al., 2016). Finally, a rather different mechanism for Ca2+/Mg2+-dependent regulation was reported for NCS protein of the KChIP class, i.e., KChIP3, also known as the transcriptional repressor DREAM. In the absence of magnesium, this protein binds Ca2<sup>+</sup> non-cooperatively in the following sequence EF3→EF4→EF2. Interestingly, the apo-form of KChIP3 coordinates Mg2<sup>+</sup> with high affinity (K<sup>D</sup> = 13 µM) in EF2 (EF3 and EF4 bind Mg2<sup>+</sup> in the millimolar range), and this bound magnesium cannot be replaced by calcium, suggesting that, under cellular conditions, the protein will exist in either Mg2+-bound, or 2Ca2+/Mg2+-bound forms. Consistently, Mg2+-bound KChIP3 exists as a monomer and can specifically recognize target DNA elements, whereas Ca2<sup>+</sup> binding to EF3 and/or EF4 induces dimerization of the protein and suppresses DNA binding. Similar to GCAP1, apo-KChIP3 represents a molten globule and Ca2+/Mg2<sup>+</sup> binding enhances its stability (Osawa et al., 2005).

Overall, Ca2+/Mg2<sup>+</sup> interplay governs the structural and functional properties of the majority of NCS proteins, although they exhibit different modes of regulation. Meanwhile, the involvement of zinc ions in this regulation so far has only been determined for recoverin. Similar to NCS-1, recoverin binds Zn2+, regardless of the presence of calcium, while the binding only slightly affects the secondary structure of the protein and destabilizes its Ca2+-saturated form. In recoverin, Zn2<sup>+</sup> was proposed to be coordinated outside EF-hands since it binds to "inactivated" mutant with E→Q substitutions in the 12th position of the loop of functional EF2 and EF3 (E85Q/E121Q). However, this conclusion does not seem to be strict, as our current calculations indicate that E→Q mutation in such a position does not necessarily prevent the four-chelator coordination of Zn2+, which becomes bound by the other chelators in the loop. Consistently, such mutation does not prevent the six-chelator coordination of Mg2<sup>+</sup> in EF-hands (Cates et al., 1999). Thus, it cannot be excluded that, similar to NCS1, Ca2+-loaded recoverin binds Zn2<sup>+</sup> in one of the functional EF-hands. In this case, the reduced stoichiometry of zinc binding to NCS1 (2 Zn2<sup>+</sup> per protein), compared to recoverin (1 Zn2<sup>+</sup> per protein), can be explained by the fact that the latter contains a smaller amount of functional EF-hands: its EF4 is naturally non-functional due to substitutions of the metal coordinating residues in the first and third positions of the EF-hand loop. It should be noted that, unlike NCS-1, recoverin exhibits an increased affinity with photoreceptor membranes in the presence of zinc (Permyakov et al., 2003). Thus, although the coordination of Zn2<sup>+</sup> may be a common property of NCS proteins, it produces somewhat different effects concerning their function, which are similar to those observed in the case of Ca2<sup>+</sup> and Mg2+. Zinc binding may therefore additionally diversify specific regulation of NCS proteins.

It still remains an open question as to which of the discovered Zn2+-bound conformers of NCS-1 (see above) dominate under physiological conditions. In contrast to the well-recognized physiological role of calcium in cell signaling, zinc has long been considered as a solely structural component of proteins. Thus, being bound with picomolar to nanomolar affinities, presumably to sulfur- and nitrogen-containing ligands in tetrahedral coordination, zinc normally serves to maintain the structure and function of enzymes, transcription factors, receptors and signaling proteins (Maret and Li, 2009). According to our data, NCS-1 binds zinc transiently with a much lower affinity and likely to the sites in EF-hands. Assuming that, in neurons, the binding will occur against the background of a constantly high [up to 1-2 mM (Romani and Scarpa, 1992)] magnesium concentration and recurring elevations [up to 1 – 2 µM (Sabatini et al., 2002)] of calcium concentration, one might suggest physiological relevance only for Zn2+(Mg2+)-bound and Zn2+(Ca2+)-bound conformers of NCS-1 in addition to the well-known Mg2+ bound and Ca2+-bound forms. Nevertheless, even the formation of two additional forms might extend the functional repertoire of the protein.

Our results suggest that the binding of zinc to NCS-1 required micromolar concentration of the free cation. However, it has been generally accepted that, in contrast to magnesium and calcium, both extracellular and intracellular free zinc concentration is low. Indeed, cytosolic zinc levels are regulated by a complex Zn2+-buffering system and the so-called "muffling reactions", involving buffer proteins of metallothionein (picomolar affinity with zinc) class, as well as transporters, such as ZnTs, ZIPs and DCTs, which shuttle Zn2<sup>+</sup> outside the cell or into subcellular stores including mitochondria, Golgi apparatus and lysosomes (Cousins et al., 2006; Colvin et al., 2010) [for review, see (Colvin et al., 2000)]. As a result, although the total concentration of zinc in cells reaches 0.2 mM (Colvin et al., 2008), the levels of free zinc in the cytoplasm were estimated as picomolar to low micromolar (Krezel and Maret, 2006). In this case, what are the physiological conditions in which the binding of zinc to NCS-1 can occur? The growing evidence indicates that, under certain conditions, the intracellular zinc levels can transiently increase, while zinc can perform signaling functions by playing complementary signaling roles with calcium (Maret, 2001). This is especially valid for the nervous system, as it is characterized by the highest extracellular zinc concentration and Zn2<sup>+</sup> is known to be specifically accumulated in neurons (Frederickson et al., 2005). The hallmark of neuronal Zn2<sup>+</sup> is its neurotransmitter function, along with glutamate in so-called "gluzinergic" neurons of forebrain. In presynaptic terminals, the cation is accumulated in ZnT3-loaded synaptic vesicles and undergoes a Ca2+-induced release into a synaptic cleft, where it can modulate various ionotropic and metabotropic receptors. The resulting high zinc concentration in the cleft (raised from 0.5 to 300 µM) can be pumped back to the presynaptic cell by ZnT3, or permeate into the postsynaptic neurons through calcium channels, thereby increasing the local cytosolic level of the cation (Frederickson and Bush, 2001). Furthermore, under certain conditions, zinc can be released from intracellular sources. For instance, in the hippocampal neurons exposure to glutamate-induced Ca2<sup>+</sup> influx triggers cytosolic acidification and intracellular Zn2<sup>+</sup> release (Kiedrowski, 2012). The resulting zinc signals could be recognized by specialized Zn2+-binding sites in neuronal proteins in order to conduct biochemical stimuli (Maret, 2006; Maret and Li, 2009). The levels of zinc are also high in neuroretina, where the maximal amount of total zinc was found in inner segments and synaptic terminals of photoreceptors cells, suggesting that it may participate in photoreceptor metabolism and neurotransmission (Ugarte et al., 2012; Ugarte and Osborne, 2014). Similar to gluzinergic neurons, photoreceptors have been suggested as releasing zinc together with glutamate in synapses. Indeed, both neurotransmitters were found in synaptic vesicles of the outer plexiform layer, which also contained ZnT3 (Akagi et al., 2001; Ugarte and Osborne, 2014). Importantly, photoreceptors contain considerable amounts of free zinc (or so-called "loosely bound zinc"), while its concentration varies depending on the light conditions, which, in turn, are likely to generate zinc signals (Ugarte and Osborne, 2014). Our data suggest that such signals in CNS and retinal neurons can be detected and transmitted by NCS-1. In ITC studies, the presence of zinc enhances the binding of Ca2+-NCS-1 to D2R (**Figures 8A,B** and **Table 2**), a process known to suppress the desensitization of the receptor. Therefore, upon receiving a joint Ca2+/Zn2<sup>+</sup> stimulus, NCS-1 can modulate dopamine signaling in a specific

enhanced manner. In addition, Zn2+(Ca2+)-bound NCS-1 can specifically regulate the desensitization of D2R or other homologous receptors by GRKs. Indeed, the presence of low zinc levels improved the binding of Ca2+-NCS-1 to GRK1, whereas the subsequent elevations in zinc concentration produced the opposite effect (**Figure 8C**). Finally, Zn2<sup>+</sup> binding can affect the well-recognized function of NCS-1 in the Ca2+-dependent regulation of neurotransmission and synaptic plasticity, as the most pronounced increase in intracellular zinc is expected to be in synaptic terminals, while NCS-1 is known to be specifically accumulated in this part of the neurons including photoreceptors (De Raad et al., 1995; Tsujimoto et al., 2002; Sippy et al., 2003; Negyessy and Goldman-Rakic, 2005).

On the other side, uncontrolled elevations of zinc in neurons and the consequential impairment of Zn2<sup>+</sup> and Ca2<sup>+</sup> ion interplay can produce pathological effects. One of the principal causes of these elevations is thought to be oxidative stress (Wood and Osborne, 2001; Sensi et al., 2003; Sheline et al., 2010b). Thus, in CNS neurons, zinc becomes released from metallothioneins in response to the oxidation or nitrosylation of their cysteine residues (Bossy-Wetzel et al., 2004). In photoreceptors, the major source for pathological free zinc is rhodopsin, which coordinates seven zinc ions per dimer and can lose them in response to light-induced oxidative stress (Sheline et al., 2010b). All these events have close ties to neurological and neuro-ophthalmological disorders. Thus, increased zinc concentration in postsynaptic neurons was shown to promote excitotoxic cell death after seizures and mechanical brain trauma, while zinc chelators were found to be neuroprotective (Suh et al., 2000). Consistently, exposure of neurons to high zinc concentrations or induction of intracellular zinc release promotes their apoptosis (Manev et al., 1997; Bossy-Wetzel et al., 2004). Elevations of retinal zinc are associated with ischemia, trophic deprivation or hypoglycemia leading to neuronal death (Yoo et al., 2004; Suh et al., 2008; Sheline et al., 2010a). The mechanisms of zinc toxicity generally involve massive Zn2+-induced aggregation of neuronal proteins (Cuajungco and Faget, 2003). For instance, an increase in extracellular zinc is a key factor in the aggregation of amyloid plaques in Alzheimer's disease. Furthermore, overexpression of zinc transporter proteins was observed in patients in the early stages of this disorder (Lyubartseva and Lovell, 2012). In the current study, we have demonstrated that non-specific lowaffinity binding of excessive zinc promotes the aggregation and precipitation of NCS-1, which are associated with the formation of fibrilic twisted rope-like structures of the protein (**Supplementary Figure S2**). These structures resemble the Zn2+-induced aggregates of another neuronal protein, TDP-43, the aggregation of which is associated with amyotrophic lateral sclerosis and frontotemporal lobar degeneration (Garnier et al., 2017). There are no direct indications concerning the involvement of NCS-1 in the pathogenesis of neurodegenerative diseases. Meanwhile, its transcription levels were found to be altered in Alzheimer's disease, which is known to be associated with altered zinc homeostasis (Karim et al., 2014). Furthermore, a large body of evidence supports the neuroprotective role of NCS-1 (Nakamura et al., 2006; Yip et al., 2010), which might be suppressed upon the Zn2+-induced loss of the protein structure. In addition, it has been proposed that NCS-1 misfolding, together with calcium dysregulation, contributes to neurodegeneration (Heidarsson et al., 2014). These observations are in accord with the fact that the partial proteolytic degradation of NCS-1 and the loss of intracellular calcium signaling induce peripheral neuropathy associated with chemotherapy by paclitaxel (Boeckel and Ehrlich, 2018).

In summary, our study suggests that the complex interplay between magnesium, calcium and zinc ions results in the appearance of multiple conformations of NCS-1, thereby modulating its functional status. It also indicates that the extreme elevation of zinc levels peculiar to some neurodegenerative and neuro-ophthalmological disorders may cause the formation of unstable Zn2+-bound conformers of NCS-1 and promote its aggregation. Further studies are required for unraveling the molecular mechanism and exact sites of zinc binding to NCS-1 and firmly establishing of physiological and pathological roles of this phenomenon.

# AUTHOR CONTRIBUTIONS

PT, AR, and FD performed ITC, DSF, and TEM studies. VB, AZ, and EZ performed functional assays and ITC studies. AN, MS, and SP performed fluorimetric, CD, contributed to equilibrium dialysis, and AAS experiments. VV and DZ performed expression and purification of the proteins and analytical gel-filtration. MB and AG performed molecular modeling and QM/MM molecular dynamics simulations. PT, SP, and EZ wrote the article.

# FUNDING

This work was supported by grants from the Russian Foundation for Basic Research (15-04-07963 and 18-04-01250) to EZ.

# ACKNOWLEDGMENTS

The research is carried out using the equipment of the shared research facilities of HPC computing resources at Lomonosov Moscow State University supported by the project RFMEFI62117X0011. ITC and nanoDSF measurements were performed in Plateau Microcalorimetry Timone.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol.2018. 00459/full#supplementary-material

FIGURE S1 | Binding of NCS-1 to photoreceptor membranes in the presence of zinc. Weight fractions of NCS-1 (25 µM) bound to urea-washed bovine photoreceptor membranes in the presence of 0, 25, 50, 75, or 100 µM Zn2<sup>+</sup> [i.e., at (Zn2+)/(NCS-1) ratio of 0-4] under Ca2+-free conditions (−Ca2+) or on the background of 1 mM Ca2<sup>+</sup> (+Ca2+) according to the data from equilibrium centrifugation assay.

FIGURE S2 | Structural properties of NCS-1 in the presence of excessive zinc concentrations. (A) Mid-transition temperatures of NCS-1 (25 µM) aggregation in the presence of 0.1–2000 µM Zn2<sup>+</sup> on the background of either 1 mM Mg2<sup>+</sup> or 1 mM Ca2<sup>+</sup> determined form light scattering at 350 nm. (B) Weight fractions of NCS-1 (25 µM) precipitated in the presence of 0–500 µM Zn2<sup>+</sup> on the background of either 1 mM Mg2<sup>+</sup> or 1 mM Ca2+. (C,D) Electron microphotographs

#### REFERENCES


of NCS-1 aggregates, obtained in the presence of 5 mM Zn2+; scale: 0.5 µm (C), 0.2 µm (D).

TABLE S1 | Secondary structure content (in %) of NCS-1 estimated from CD data obtained at 25◦C (10 mM Hepes-KOH buffer (pH 7.6), 100 mM KCl) under metal-free conditions or in the presence of Ca2<sup>+</sup> (100 µM CaCl2), Mg2<sup>+</sup> (1 mM MgCl2), Zn2<sup>+</sup> (100 µM ZnCl2) or their combinations.


regulation of rhodopsin kinase. Front. Mol. Neurosci. 5:28. doi: 10.3389/fnmol. 2012.00028



**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 © 2018 Tsvetkov, Roman, Baksheeva, Nazipova, Shevelyova, Vladimirov, Buyanova, Zinchenko, Zamyatnin, Devred, Golovin, Permyakov and Zernii. 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) and the copyright owner(s) 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 Complex Conformational Dynamics of Neuronal Calcium Sensor-1: A Single Molecule Perspective

Dhawal Choudhary 1,2 , Birthe B. Kragelund<sup>3</sup> \*, Pétur O. Heidarsson<sup>4</sup> \* and Ciro Cecconi 1,2 \*

<sup>1</sup>Department of Physics, Informatics and Mathematics, University of Modena and Reggio Emilia, Modena, Italy, <sup>2</sup>Center S3, CNR Institute Nanoscience, Modena, Italy, <sup>3</sup>Structural Biology and NMR Laboratory, Department of Biology, University of Copenhagen, Copenhagen, Denmark, <sup>4</sup>Department of Biochemistry, University of Zurich, Zurich, Switzerland

The human neuronal calcium sensor-1 (NCS-1) is a multispecific two-domain EF-hand protein expressed predominantly in neurons and is a member of the NCS protein family. Structure-function relationships of NCS-1 have been extensively studied showing that conformational dynamics linked to diverse ion-binding is important to its function. NCS-1 transduces Ca<sup>2</sup><sup>+</sup> changes in neurons and is linked to a wide range of neuronal functions such as regulation of neurotransmitter release, voltage-gated Ca<sup>2</sup><sup>+</sup> channels and neuronal outgrowth. Defective NCS-1 can be deleterious to cells and has been linked to serious neuronal disorders like autism. Here, we review recent studies describing at the single molecule level the structural and mechanistic details of the folding and misfolding processes of the non-myristoylated NCS-1. By manipulating one molecule at a time with optical tweezers, the conformational equilibria of the Ca<sup>2</sup><sup>+</sup>-bound, Mg<sup>2</sup><sup>+</sup>-bound and apo states of NCS-1 were investigated revealing a complex folding mechanism underlain by a rugged and multidimensional energy landscape. The molecular rearrangements that NCS-1 undergoes to transit from one conformation to another and the energetics of these reactions are tightly regulated by the binding of divalent ions (Ca<sup>2</sup><sup>+</sup> and Mg<sup>2</sup><sup>+</sup>) to its EF-hands. At pathologically high Ca<sup>2</sup><sup>+</sup> concentrations the protein sometimes follows non-productive misfolding pathways leading to kinetically trapped and potentially harmful misfolded conformations. We discuss the significance of these misfolding events as well as the role of inter-domain interactions in shaping the energy landscape and ultimately the biological function of NCS-1. The conformational equilibria of NCS-1 are also compared to those of calmodulin (CaM) and differences and similarities in the behavior of these proteins are rationalized in terms of structural properties.

Keywords: NCS-1, protein folding and misfolding, calcium binding, optical tweezers, single molecule studies

### NCS-1 IS A MULTI-FUNCTIONAL, TWO-DOMAIN PROTEIN

Calcium ion (Ca2+) signaling is crucial for neurotransmitter release and is intrinsic for neuronal functions. Calcium signaling is mediated by calcium binding proteins that sense changes in cellular concentration of Ca2<sup>+</sup> ions and respond by interacting with downstream regulatory targets, further cascading the signal. One example of such calcium sensor proteins is the family

#### Edited by:

Daniele Dell'Orco, Università degli Studi di Verona, Italy

#### Reviewed by:

Yogendra Sharma, Centre for Cellular and Molecular Biology (CSIR), India Hans Vogel, University of Calgary, Canada

#### \*Correspondence:

Birthe B. Kragelund bbk@bio.ku.dk Pétur O. Heidarsson p.heidarsson@bioc.uzh.ch Ciro Cecconi ciro.cecconi@gmail.com

Received: 29 September 2018 Accepted: 03 December 2018 Published: 17 December 2018

#### Citation:

Choudhary D, Kragelund BB, Heidarsson PO and Cecconi C (2018) The Complex Conformational Dynamics of Neuronal Calcium Sensor-1: A Single Molecule Perspective. Front. Mol. Neurosci. 11:468. doi: 10.3389/fnmol.2018.00468 of neuronal calcium sensors (NCSs), which are expressed primarily in neurons and photoreceptor cells (Weiss et al., 2010; Reyes-Bermudez et al., 2012). These proteins respond to changes in Ca2<sup>+</sup> concentration through conformational changes that allow them to bind diverse protein partners (Burgoyne and Haynes, 2012). The NCS family has 15 highly-conserved members in mammals and these proteins have numerous functions (Burgoyne, 2007). NCS-1 is the primordial member of the family and is the most widely expressed. It has been shown to have roles in an array of cellular processes, such as regulation of Ca2+- (N and P/Q type) and K<sup>+</sup> (Kv4)-channels (Weiss et al., 2000; Nakamura et al., 2001; Guo et al., 2002; Tsujimoto et al., 2002), phosphodiesterase activity (Burgoyne et al., 2004), membrane trafficking (McFerran et al., 1999), and the direct regulation of the dopamine D2 receptor and the GRK2 kinase (Kabbani et al., 2002). A connection between defective NCS-1 and neurodegenerative disorders like schizophrenia, autism and bipolar disorder has also been suggested (Kabbani et al., 2002; Koh et al., 2003; Bai et al., 2004; Handley et al., 2010; Pavlowsky et al., 2010). It was observed that in comparison to healthy individuals, patients suffering from schizophrenia had upregulated levels of NCS-1 in the prefrontal cortex, which may explain their reduced prefrontal cortex activity (Kabbani et al., 2002; Koh et al., 2003). Also, NCS-1 has been attributed to the regulation of synaptic activity as it directly interacts with D2 receptors and it is plausible that defective NCS-1 may lead to impairment of cognitive functions and mental retardation (Burgoyne, 2007).

All NCS proteins are composed of ∼200 residues organized in a compact and globular structure, have either an N-terminal myristoylation or palmitoylation site (Pongs et al., 1993; Rivosecchi et al., 1994; Tsujimoto et al., 2002; Burgoyne, 2007), and bind calcium through EF-hand motifs. NCS-1 is an all-helical 190 residue protein organized in two ∼100 a.a. domains, each containing a pair of EF-hands; the N-domain EF1 and EF2, and the C-domain EF3 and EF4 (**Figure 1A**; Heidarsson et al., 2012a). Due to a conserved Cys/Pro substitution in EF1, NCS-1 is only capable of binding three calcium ions (Aravind et al., 2008; Grabarek, 2011), a trait of all NCS family members except for recoverin and KChIP1 who have only two active EF-hands (Scannevin et al., 2004; Zhou et al., 2004). In NCS-1, EF2 and EF3 are structural sites as they can bind to both Ca2<sup>+</sup> and Mg2+, whereas EF4 is a regulatory site only binding Ca2<sup>+</sup> and with lower affinity (Aravind et al., 2008). A few recent high-resolution

FIGURE 1 | Mechanical manipulation of a single non-myristoylated neuronal calcium sensor-1 (NCS-1) molecule. (A) Schematic representation of the optical tweezers setup. A NCS-1 molecule is tethered to two polystyrene beads by means of molecular handles (∼500 bp DNA molecules) that function as spacers to avoid unspecific interaction between the tethering surfaces (Cecconi et al., 2005). One bead is held in an optical trap, while the other is held at the end of a pipette by suction. During the experiment the protein is stretched and relaxed by moving the pipette relative to the optical trap, and the applied force and the molecular extension are measured as described in Smith et al. (2003). The inset shows the NMR structure of NCS-1 (PDB code 2LCP), where the C- and N-domains are shown in blue and green, respectively. (B) Force vs. extension cycle obtained by stretching (red trace) and relaxing (blue trace) NCS-1 in the absence of divalent ions. (C) Force vs. extension cycle obtained by stretching (red trace) and relaxing (blue trace) NCS-1 in the presence of 10 mM Mg2+. (D) Mechanical manipulation of the Ca2+-bound state of NCS-1. During stretching (red trace) the N-domain unfolds at ∼13 pN and the C-domain at ∼16 pN. During relaxation (black trace), U folds into N through a four-state process (U > I2 > I1 > N) coordinated by calcium binding. (E) Extension vs. time traces acquired at different constant forces showing the Ca2+-bound state of NCS-1 fluctuating at equilibrium between N, I1, I2 and U. The population probability of the different states can be modulated by force. Analysis of these experimental data with the Hidden Markov Model allows for the characterization of the energy landscape of the protein in terms of activation energy barriers, separating the different molecular states, and positions of the transition states along the reaction coordinate (Rabiner, 1989; Chodera et al., 2011). (F) Energy landscape of the Ca2+-bound state of NCS-1 at zero force. The transition states of the different folding reactions are indicated with the letter B. The activation barriers separating the different molecular states were calculated using a pre-exponential factor of 1.2 × 10−<sup>4</sup> Hz (Gebhardt et al., 2010). Panels adapted from Heidarsson et al. (2013b) and Naqvi et al. (2015).

structures are available for NCS-1, which all show excellent agreement regarding the general tertiary structure, the relative orientation of the two domains, and the locations of α-helices (Ames et al., 2000; Bourne et al., 2001; Strahl et al., 2007; Heidarsson et al., 2012a). An NMR study on NCS-1 suggested that independent movements of the N- and C-domains can occur, which may help accommodating different ligands in the binding pocket, and that the last 15 residues of the protein are disordered yet crucial for conformational stability, undergoing conformational exchange that may regulate ligand binding (Heidarsson et al., 2012a). The equilibrium unfolding of NCS-1 has been well studied using guanidinium chloride (GdmCl) denaturation and shows that the protein unfolds through two separate unfolding transitions, corresponding to the sequential unfolding of the N- (unfolding at 3.1 M GdmCl with ∆G ∼35 KJ mol−<sup>1</sup> ) and C-domain (unfolding at 4.6 M GdmCl with ∆G ∼38 KJ mol−<sup>1</sup> ; Aravind et al., 2008; Heidarsson et al., 2012a). While the structure, thermodynamics, and biological functions of NCS-1 have been elucidated in detail with different techniques, the folding and misfolding mechanisms of this protein have long remained elusive. This information gap has recently been bridged by single-molecule optical tweezers studies.

### A SINGLE MOLECULE PERSPECTIVE ON PROTEIN FOLDING USING DIRECT MECHANICAL MANIPULATION

Protein folding, especially in the case of large multi-domain proteins, is a complex process encompassing many concerted events such as breaking of non-native contacts, formation of native interactions and changes in conformation and dynamics (Batey et al., 2008; Borgia et al., 2015; Jahn et al., 2016). In large multi-domain proteins different domains can fold and unfold via different pathways either independently or dependently and intermediate states can be populated on the way to a native conformation. Their folding and unfolding processes are also dependent on the crosstalk between the domains and on internal friction, brought about by a frustrated search for inter-domain contacts between partially formed domains. Traditional methods such as X-ray crystallography, NMR and optical spectroscopies have been extraordinarily powerful to decipher the structure and folding characteristics of proteins. However, the ensemble nature of the signal provided by such techniques can blur the diversity of folding trajectories and short lived populations, making it difficult to gauge the complicated sequence of events that characterize the folding mechanism of proteins and other biomolecules.

Single molecule force spectroscopy using optical tweezers enables direct mechanical manipulation of individual molecules and the detailed characterization of their conformational equilibria under tension (Cecconi et al., 2005; Shank et al., 2010; Yu et al., 2012; Xu and Springer, 2013; Alemany et al., 2016; Zhang, 2017; Jahn et al., 2018; Wruck et al., 2018). In these experiments, a single molecule is tethered between two polystyrene beads using DNA molecular handles and then is stretched and relaxed by changing the distance between the tethering surfaces (**Figure 1A**; Heidarsson et al., 2013a). Under tension, the unfolding and refolding of a protein is accompanied by changes in the extension of the molecule, giving rise to discontinuities (rips) in the recorded force traces. A molecule can be stretched and relaxed at constant speed (Heidarsson et al., 2012b; Caldarini et al., 2014), or it can be kept at a specific force through a force feedback mechanism and observed to fluctuate between different molecular conformations (Heidarsson et al., 2013b). A careful analysis of the experimental data and the use of advanced statistical methods allow a detailed characterization of the folding and, possibly, misfolding processes of the molecule and the reconstruction of its energy landscape (Stigler et al., 2011). An important limitation of optical tweezers to keep in mind is that they describe three-dimensional molecular processes, such as the folding of a protein, along a single reaction coordinate defined by the points of force application, namely the residues to which the DNA handles are attached. As a consequence, during the mechanical manipulation of a molecule, all structural variations that do not produce measurable changes of the molecular extension along the pulling axis are not detected, and hence the energy landscape emerging from these studies is necessarily one-dimensional (Avdoshenko and Makarov, 2016). However, to circumvent this limitation an increasing number of solutions are emerging. For example, the attachment points of the DNA handles on the protein surface can be changed at will to pull the molecule along various reaction coordinates and assess different regions of its energy landscape (Elms et al., 2012; Heidarsson et al., 2012b). Alternatively, novel instrumentation designs with multiple optical traps can be employed that allow application of force along different pulling axes simultaneously (Dame et al., 2006). Otherwise, it is nowadays possible to use hybrid systems combining optical tweezers with single molecule Förster Resonance Energy Transfer (FRET) where the additional possibility of monitoring the distance between two or more residues suitably marked with fluorophores allows a multidimensional description of the molecular process under study (Sirinakis et al., 2012; Lee and Hohng, 2013).

#### SIMPLE FOLDING MECHANISM UNDER RESTING CONDITIONS

The cellular environment contains a vast amount of different ions and most EF-hand proteins can bind other divalent ions besides calcium, most prominently Mg2+, which is kept relatively constant at ∼5 mM in cells (Gifford et al., 2007; Grabarek, 2011). The Mg2+-bound state of EF hand proteins has many specific roles besides modulating Ca2<sup>+</sup> binding and its properties are therefore of high interest (Zot and Potter, 1982; Wingard et al., 2005; Peshenko and Dizhoor, 2006; Aravind et al., 2008). In the case of NCS-1, the Mg2+-bound state seems to be the primary interacting conformation for many of its targets, such as the dopamine receptor D2 and P14Kb (Kabbani et al., 2002; Burgoyne et al., 2004) existing in equilibria with the apo and Ca2+-bound states. Elucidating the conformational dynamics of all these states (Burgoyne and Haynes, 2012) will help us understand better the molecular mechanisms mediating the biological functions of NCS-1.

At resting conditions in the cell, the two dominant forms of NCS-1 are the apo and Mg2+-bound states. In a recent publication, Naqvi et al. (2015) used optical tweezers to characterize at the single molecule level the conformational equilibria of these states for the non-myristoylated form of NCS-1. When they stretched and relaxed individual NCS-1 molecules in the absence of any divalent ions, they obtained force vs. extension traces characterized by reversible two-state fluctuations around 6.5 pN that originate from the folding and unfolding of the C-domain (**Figure 1B**). In fact, the N-domain of the apo form of NCS-1, as shown using deletion variants (Naqvi et al., 2015), is unstructured or loosely folded, and under tension it does not give rise to detectable transitions in the recorded traces. In contrast, under the same experimental conditions the C-domain is collapsed in a folded conformation that under tension displays a behavior that resembles that of molecular structures mainly stabilized by secondary interactions and weak tertiary contacts (Cecconi et al., 2005; Elms et al., 2012). Indeed, these data are also consistent with previous NMR studies showing little tertiary contacts in the apo form of NCS-1, and lack of stable globular structure (Cox et al., 1994; Aravind et al., 2008).

In the presence of 10 mM Mg2<sup>+</sup> the behavior of NCS-1 changes drastically. Both the N- and C-domain acquire compact and mechanically resistant conformations that under tension loose and gain structure as separate units, through fully cooperative and sequential two-state transitions (**Figure 1C**; Naqvi et al., 2015). The C-domain is mechanically more resistant than the N-domain and unfolds at higher forces mirroring the observation from equilibrium chemical denaturation experiments (Aravind et al., 2008; Heidarsson et al., 2012a). After full mechanical denaturation of the protein and during relaxation of the applied force, the C-domain of NCS-1 is the first to fold, followed by the N-domain folding at lower forces. These sequential folding events are coordinated by Mg2<sup>+</sup> binding to EF3 and EF2, respectively. The Mg2+-state of NCS-1 thus folds into its native state through a three-state process involving an intermediate state comprising a folded C-domain and an unstructured N-domain with ion occupancy only in EF3. The dimensions of the observed rips in the force vs. extension traces reveal more structuring of NCS-1 upon Mg2<sup>+</sup> binding than previously suggested by NMR studies (Aravind et al., 2008), although the dynamical nature of the folded state is not revealed from the single molecule data.

#### INCREASING FOLDING COMPLEXITY UNDER CALCIUM ACTIVATING CONDITIONS

In the presence of activating concentrations of calcium the folding process of non-myristoylated NCS-1 gets considerably more complex (Heidarsson et al., 2014). Under these experimental conditions, unfolded NCS-1 molecules fold back into their native states through a strict sequence of events coordinated by calcium binding (**Figure 1D**). Ca2<sup>+</sup> first binds to EF3, triggering the collapse of the polypeptide chain (U state) into an on-pathway intermediate state I1; then EF4 binds a calcium ion, to induce the full folding of the C-domain (intermediate state I2). The last calcium ion finally binds to EF2 to fold the N-domain and thus the entire protein reaches its native state (N state). In this folding process, the C-domain always folds first and acts as ''internal chaperone'' for the correct folding of the N-domain. If the C-domain does not reach its native state, as in the knockout mutant NCS1EF4, the N-domain fails to fold correctly. This asymmetrical folding process that always starts with the folding of the C-domain probably reflects the asymmetrical structure of the protein, as further discussed below.

Each folding transition of NCS-1 is associated with a different structural change and energy cost. By monitoring in real time individual NCS-1 molecules fluctuating between different molecular conformations in constant force experiments, Heidarsson et al. (2013b) were able to characterize the kinetics and thermodynamics of the folding reactions of non-myristoylated NCS-1 and reconstruct the salient features of its energy landscape (**Figures 1E,F**). The results of these studies show how the folding transitions from U to I2 and from I1 to N are associated with large changes in molecular extensions but, at the same time, they are downhill reactions with no activation energy barriers. On the other hand, the transition from the intermediate state I2 to the intermediate state I1, which leads to the complete folding of the C-domain, involves a small compaction of the protein but is a barrier-limited reaction. Thus, folding of the C-domain into its native structure upon calcium binding to EF4 is the rate-limiting step of the entire folding process of NCS-1 at activating concentration of Ca2+. Interestingly, this step is affected by calcium concentration, which can open access to alternative and off-pathway folding trajectories.

#### PATHOLOGICALLY HIGH CALCIUM CONCENTRATIONS OPEN ACCESS TO MISFOLDING PATHWAYS

Once an unfolded NCS-1 molecule has collapsed into the on-pathway intermediate state I2, it can either transit into I1 to eventually reach the native state, or take alternative pathways that lead to misfolded states (Heidarsson et al., 2014). The probability that the protein takes one pathway or the other depends on the calcium concentration. At physiologically relevant calcium concentration (0.5 µM) only 5% of the NCS-1 molecules misfold, but as the [Ca2+] increase, so does the misfolded population and at [Ca2+] of 10 mM one molecule out of two populates a pathway that leads to non-native structures. Two main misfolded states (M1 and M2) are populated by NCS-1, differing in extension and occupancy probability, with M1 being significantly populated only at high [Ca2+]. **Figure 2** summarizes the folding and misfolding pathways of NCS-1 under different environmental conditions. Strikingly, misfolding reactions were observed only with wild-type NCS-1. Variants carrying disabled ion-binding sites never populate M1 or M2. These intriguing

results suggest that the misfolding reactions observed in Heidarsson et al. (2014), stem from specific interactions that take place between properly folded EF hands. Inter-domain crosstalk thus plays a crucial role for the folding and misfolding of NCS-1, determining the final fate of the protein.

# COMPARISON TO THE FOLDING NETWORK OF CALMODULIN

The folding and misfolding mechanisms of the other NCS family members and of other EF hand calcium binding proteins are still largely uncharacterized and the question remains: is it likely that the same principles of folding and misfolding can be inferred for these similar proteins? Other related proteins have been studied with regards to folding and stability using ensemble methods with varying degrees of details (Yamniuk et al., 2007; Suarez et al., 2008). However, to the best of our knowledge, the only EF hand protein whose conformational equilibria have been characterized in great detail at the single molecule level with optical tweezers is calmodulin (CaM; Stigler et al., 2011; Stigler and Rief, 2012), allowing for a direct comparison with NCS-1. The archetypical calcium binding protein CaM is characterized by a symmetrical structure with two almost identical domains, each binding two calcium ions, separated by a long and flexible α helix (Chattopadhyaya et al., 1992). The comparison between the folding network of NCS-1 and that of CaM indeed reveals many differences yet some similarities. The folding process of NCS-1 always starts with the U to I2 transition, indicating that only a single pathway is available on its energy landscape to initiate its journey to the native state. In contrast, unfolded CaM can initiate its folding process following three distinct pathways; one leading to an off-pathway intermediate (F23) with the EF hands 2 and 3 mispaired (16% probability), and the other two leading to either the on-pathway intermediate F<sup>12</sup> (42% probability) or the on-pathway intermediate F<sup>34</sup> (42% probability), with either the N- or the C-domain, respectively, fully folded. The folding process of CaM can thus start with equal probability from the N- or the C-domain, while that of NCS-1 always start from the C-domain and always through the same U to I2 transition. For CaM, F<sup>12</sup> and F<sup>34</sup> can then both transit into the native state, although with different probabilities as in 50% of the cases F<sup>12</sup> takes a misfolding pathway leading to the off-pathway intermediate state F123, where EF3 is collapsed onto the folded N-domain in a non-native conformation. The native state of CaM can thus be reached from two distinct intermediate states through two different folding pathways, unlike the native state of NCS-1 that can be accessed only from I1. Moreover, the native unidimensional folding process of NCS-1 is characterized by a bottleneck where the protein can either proceed towards its native state by overcoming the large activation barrier separating I2 from I1 or take misfolding pathways in a calcium dependent manner. No bottleneck characterizes the folding network of CaM where the interconversions between the different molecular states are controlled by similar activation barriers. In addition, despite CaM folding rates and folding mechanism having been shown to be highly dependent on calcium concentration (Stigler and Rief, 2012), the population of CaM misfolding states have not yet been demonstrated to display such calcium dependence.

Where do the differences in the folding networks of CaM and NCS-1 originate from? The answer may lie in inter-domain interactions and the coupling free energies that are more developed in NCS-1. Proteins such as CaM have a flexible linker connecting their two domains, allowing a high degree of conformational variation between them and at the same time imposes separation of their folding funnels (Gifford et al., 2007; Kiran et al., 2017). For NCS-1, instead, a very short U-shaped linker of four residues (Ames et al., 2000) considerably limits the orientations of the two domains to face-to-face conformations, and the interdomain interaction involves evolutionarily conserved residues pointing to a functional relevance of this interface. Additionally, NCS-1 is an asymmetrical protein as its N-domain has one active and one inactive EF hand, resulting in it being thermodynamically less stable than the C-domain (Heidarsson et al., 2012a), very unlike CaM where all EF-hands are active with similar calcium binding affinities (Masino et al., 2000). Moreover, NCS-1 carries a C-terminal tail that has an N-domain stabilizing effect (Heidarsson et al., 2012a) and this effect may only become functional, once the C-domain is folded. Thus, the sequential folding observed for NCS-1 could be the result of its intricate structural features and its biological function as a sensor. More studies on related NCS proteins will undoubtedly help clarify these issues. Nevertheless, given the complex folding networks of CaM and NCS-1, and the frequent occurrence of misfolded states in multi-domain proteins (Han et al., 2007; Borgia et al., 2011, 2015) we predict that misfolding is likely to occur in similar calcium binding proteins.

#### CONCLUSIONS AND FUTURE PERSPECTIVES

We have outlined how the Ca2+-bound, Mg2+-bound and apo forms of the non-myristoylated NCS-1 have been investigated at the single molecule level using optical tweezers. We have illustrated the rugged and multistate energy landscape arising from these studies, which underlays the calcium-dependent folding and misfolding trajectories of NCS-1. We have also discussed the importance that the conformational sensitivity of NCS-1 might have for its calcium sensing actions. Finally, we have made comparison to the very similar calcium binding EF hand protein CaM and have highlighted differences and similarities between the folding networks of the two proteins, which are likely to have arisen due to their different structural and functional features across the domains.

For NCS-1, the calcium-dependent modulation of misfolding pathways is highly interesting and suggests an important link between the conformational space, calcium dysregulation, and neurodegeneration. Interestingly, sustained elevated levels of free Ca2<sup>+</sup> have been associated with aging cells and linked to the development of major neurological diseases such as bipolar disorder and Alzheimer's (Toescu and Vreugdenhil, 2010; Berridge, 2012). In some neurological diseases, such as bipolar disorder and schizophrenia, NCS-1 is up-regulated in the pre-frontal cortex of patients (Koh et al., 2003). A compelling hypothesis would be that the upregulation is a response to a loss of protein function caused by calcium-induced misfolding. Studies in cell, focusing directly on the calcium effects on NCS-1 expression levels might reveal whether there indeed is a specific phenotype associated with NCS-1 misfolding, and potentially also of other members of the family.

The behavior of NCS-1 outlined in this review article may also help rationalize how NCS-1 is able to have such promiscuous interactions with over 20 reported binding partners (Burgoyne, 2007) despite a highly folded and globular structure. A certain degree of flexibility must thus be required to accommodate this list of diverse ligands such as dopamine receptors, various ionchannels, glial cell line-derived neurotrophic factor as well as membranes and their constituents. The many conformational states that were uncovered in the articles discussed above may reflect a significant flexibility of NCS-1, which is needed for it to act as a molecular hub (Heidarsson et al., 2013b, 2014; Naqvi et al., 2015). These properties may be even further expanded including its N-terminal myristoylation, which leads to localization and broader range of interactors. How this lipidic modification changes the molecular landscape of NCS-1 remains to be addressed in detail. The single molecule approach provided by optical tweezers can potentially be used to probe the interaction of NCS-1 with different ligands as well as its modulation by posttranslational modification, and help uncover the connection between conformational states, ligand recognition and disease states. Indeed, the two EF hand proteins for which detailed insight into their single-molecule folding networks exist (Stigler et al., 2011; Heidarsson et al., 2014) have revealed that single-molecule approaches are key to their studies and enable the extraction of important insight into structures as well as function and dysfunction.

#### AUTHOR CONTRIBUTIONS

DC designed and wrote the article; carried out bibliographic research. BK assisted in structural details and biological aspects, and contributed in the final revisions of the manuscript. PH designed and wrote the article. CC designed and wrote the article; prepared the figures. All authors have provided final approval of the version to be published.

#### FUNDING

This work was supported by the Italian Ministry of Education, University and Research-MIUR and by the grant 020145\_17\_FDA\_CARRAFAR2016INTER (for DC and CC) and by grants from the Novo Nordisk Foundation (to PH and to BK). The University of Modena and Reggio Emilia

#### REFERENCES


(Dr. Tiziana Covili, tiziana.covili@unimore.it) will pay for the publication of this article using funds allocated to CC (FAR2016-CARRA).


(NCS-1) in the prefrontal cortex of schizophrenic and bipolar patients. Proc. Natl. Acad. Sci. U S A 100, 313–317. doi: 10.1073/pnas.2326 93499


**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 © 2018 Choudhary, Kragelund, Heidarsson and Cecconi. 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) and the copyright owner(s) 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.

# Ca2+-Dependent Transcriptional Repressors KCNIP and Regulation of Prognosis Genes in Glioblastoma

Isabelle Néant<sup>1</sup> , Jacques Haiech<sup>2</sup> , Marie-Claude Kilhoffer<sup>2</sup> , Francisco J. Aulestia<sup>3</sup> , Marc Moreau<sup>1</sup> and Catherine Leclerc<sup>1</sup> \*

<sup>1</sup> Centre de Biologie du Développement (CBD), Centre de Biologie Intégrative (CBI), CNRS, UPS, Université de Toulouse, Toulouse, France, <sup>2</sup> Laboratoire d'Excellence Medalis, CNRS, LIT UMR 7200, Université de Strasbourg, Strasbourg, France, <sup>3</sup> Department of Basic Science and Craniofacial Biology, NYU College of Dentistry, New York, NY, United States

#### Edited by:

Jose R. Naranjo, Spanish National Research Council (CSIC), Spain

#### Reviewed by:

Ana Cristina Calvo, Universidad de Zaragoza, Spain Marisa Brini, Università degli Studi di Padova, Italy Mario Vallejo, Instituto de Investigaciones Biomédicas Alberto Sols (IIBM), Spain

> \*Correspondence: Catherine Leclerc catherine.leclerc@univ-tlse3.fr

Received: 03 October 2018 Accepted: 04 December 2018 Published: 18 December 2018

#### Citation:

Néant I, Haiech J, Kilhoffer M-C, Aulestia FJ, Moreau M and Leclerc C (2018) Ca2+-Dependent Transcriptional Repressors KCNIP and Regulation of Prognosis Genes in Glioblastoma. Front. Mol. Neurosci. 11:472. doi: 10.3389/fnmol.2018.00472 Glioblastomas (GBMs) are the most aggressive and lethal primary astrocytic tumors in adults, with very poor prognosis. Recurrence in GBM is attributed to glioblastoma stem-like cells (GSLCs). The behavior of the tumor, including proliferation, progression, invasion, and significant resistance to therapies, is a consequence of the self-renewing properties of the GSLCs, and their high resistance to chemotherapies have been attributed to their capacity to enter quiescence. Thus, targeting GSLCs may constitute one of the possible therapeutic challenges to significantly improve anti-cancer treatment regimens for GBM. Ca2<sup>+</sup> signaling is an important regulator of tumorigenesis in GBM, and the transition from proliferation to quiescence involves the modification of the kinetics of Ca2<sup>+</sup> influx through store-operated channels due to an increased capacity of the mitochondria of quiescent GSLC to capture Ca2+. Therefore, the identification of new therapeutic targets requires the analysis of the calcium-regulated elements at transcriptional levels. In this review, we focus onto the direct regulation of gene expression by KCNIP proteins (KCNIP1–4). These proteins constitute the class E of Ca2<sup>+</sup> sensor family with four EF-hand Ca2+-binding motifs and control gene transcription directly by binding, via a Ca2+-dependent mechanism, to specific DNA sites on target genes, called downstream regulatory element (DRE). The presence of putative DRE sites on genes associated with unfavorable outcome for GBM patients suggests that KCNIP proteins may contribute to the alteration of the expression of these prognosis genes. Indeed, in GBM, KCNIP2 expression appears to be significantly linked to the overall survival of patients. In this review, we summarize the current knowledge regarding the quiescent GSLCs with respect to Ca2<sup>+</sup> signaling and discuss how Ca2<sup>+</sup> via KCNIP proteins may affect prognosis genes expression in GBM. This original mechanism may constitute the basis of the development of new therapeutic strategies.

Keywords: Ca2<sup>+</sup> signaling, neuronal Ca2<sup>+</sup> sensors, KCNIP, glioblastoma multiform, cancer stem cells (CSC), quiescence

# INTRODUCTION

fnmol-11-00472 December 16, 2018 Time: 13:6 # 2

Among tumors of the central nervous system, glioblastomas (GBMs) are the most aggressive and lethal primary astrocytic tumors in adults, with very poor prognosis (Louis et al., 2016; Lapointe et al., 2018). More than 90% of the patients show recurrence after therapies combining surgical resection, radiotherapy, and temozolomide (TMZ)-based chemotherapy, and the mean survival period rarely exceeds 2 years (Stupp et al., 2005). According to the cancer stem cell model, recurrence in GBM is attributed to a small sub-population of tumor cells called glioblastoma stem-like cells (GSLCs). These GSLCs have stemlike properties and are responsible for the initiation and the growth of the tumors (Visvader and Lindeman, 2008). Indeed, the GSLCs provide all the subtypes of cells that comprise the tumor including some pseudo-endothelial cells (Ricci-Vitiani et al., 2010). GSLCs are characterized by a molecular signature which combines markers of neural and/or embryonic stem cells and of mesenchymal cells. Numerous studies support the proposal that the behavior of the tumor, including proliferation, progression, invasion, and significant resistance to therapies, is determined by the self-renewing properties of the GSLCs (Stupp et al., 2005; Bao et al., 2006; Hegi et al., 2006; Stupp and Hegi, 2007; Murat et al., 2008). More importantly, this high resistance capacity to TMZ treatment have been attributed to slow cycling or relatively quiescent GSLCs (Pistollato et al., 2010; Deleyrolle et al., 2011). Quiescent GSLCs have been identified in vivo in a mouse model of GBM (Chen et al., 2012) and in human GBM tumors (Ishii et al., 2016). Thus, targeting GSLCs and their stem cell-like properties may constitute one of the possible therapeutic challenges to significantly improve anti-cancer treatment regimens for GBM.

Ca2<sup>+</sup> is a crucial second messenger (Carafoli and Krebs, 2016) that controls a wide variety of cell functions from cell proliferation and apoptosis to organogenesis (Berridge et al., 2000; Machaca, 2011; Moreau et al., 2016). Thus, the intracellular Ca2<sup>+</sup> concentration ([Ca2+]i) is tightly regulated and involves Ca2<sup>+</sup> channels, pumps, and exchangers both at the plasma membrane and at the membrane of endoplasmic reticulum, mitochondria, or Golgi apparatus (Bootman, 2012; Humeau et al., 2018). In addition, changes in [Ca2+]i do not proceed in a stereotypical manner. The Ca2<sup>+</sup> signal can be described by its amplitude (variations of [Ca2+]i levels) and by its spatial (sources of Ca2+; organelles where changes occur) and timedependent (duration, frequency) components (Berridge, 1992; Haiech et al., 2011; Smedler and Uhlén, 2014; Monteith et al., 2017). The remodeling of Ca2<sup>+</sup> signaling contributes also to cancer hallmarks such as excessive proliferation, survival, or resistance to cell death (Roderick and Cook, 2008; Prevarskaya et al., 2014) and accumulating evidence suggests that Ca2<sup>+</sup> is also an important positive regulator of tumorigenesis in GBM (Robil et al., 2015; Leclerc et al., 2016). Interestingly, screening of the Prestwick Chemical library identified bisacodyl, an organic compound used as a stimulant laxative drug, with cytotoxic effect on quiescent GSLCs (Zeniou et al., 2015). Bisacodyl inhibits Ca2<sup>+</sup> release from inositol 1,4,5-triphosphate-dependent Ca2<sup>+</sup> stores without affecting the store-operated Ca2<sup>+</sup> entry

(SOCE) (Dong et al., 2017). These data exemplify the fact that Ca2<sup>+</sup> channels, pumps, and exchangers may represent potential therapeutic targets. In this review, we will summarize the current knowledge regarding the quiescent GSLCs with respect to Ca2<sup>+</sup> signaling and describe an original mechanism by which Ca2<sup>+</sup> can activate some genes involved in the prognosis of GBM in order to propose new strategies to explore the molecular basis of GBM development for therapeutic issues.

### TRANSITION FROM PROLIFERATION TO QUIESCENCE AND Ca2<sup>+</sup> SIGNALING

Quiescent cells are non-proliferative cells, arrested in a specific phase of the cell cycle called G0 (Coller et al., 2006). Quiescence is not a prolonged G1 phase and in contrary to the cellcycle arrest observed in differentiation or senescence, it is reversible. Transcriptional profiling data reveal that quiescent stem cells are characterized by a common set of genes which are either downregulated, these are genes associated with cell-cycle progression (i.e., CCNA2, CCNB1, and CCNE2), or upregulated and classified as tumor suppressors, including the cyclin-dependent kinase inhibitor p21 (CDKN1A) and the G0/G1 switch gene 2 (G0S2) (Yamada et al., 2012; Cheung and Rando, 2013). Quiescence represents a strategy for GSLCs to evade killing. It is thus vital to better characterize the quiescent GSLCs and to understand the mechanisms involved in the transition from a proliferative to a quiescence state. Quiescence is actively regulated by signals provided by the stem cell microenvironment. In GBM, quiescent cells are found close to necrotic tissues, in specific niches characterized by a hypoxic (Pistollato et al., 2010; Persano et al., 2011; Ishii et al., 2016) and acidic microenvironment (Garcia-Martin et al., 2006; Honasoge et al., 2014).

A recent study suggests that Ca2<sup>+</sup> is an important regulator of the balance between quiescence and proliferation in hematopoietic stem cell (HSC) (Umemoto et al., 2018). In HSCs, re-entry into cell-cycle requires Ca2<sup>+</sup> influx through Cav1 voltage-dependent Ca2<sup>+</sup> channel and the resultant activation of mitochondria. Recent findings in our group showed that Ca2<sup>+</sup> signaling is also required for GBM stem cells quiescence. On GSLCs lines, established from surgical resections of primary GBMs, we showed that change in Ca2<sup>+</sup> homeostasis is an important actor of the transition from proliferation to quiescence. In order to analyze the signals underlying this switch, we modified the culture condition by lowering the extracellular pH from pH 7.5 to 6.5. GSLCs kept in such conditions for 5 days enter G0. This simple protocol allowed to reversibly maintain GSLCs in a proliferating or in quiescent state (Zeniou et al., 2015; Aulestia et al., 2018). A RNAseq analysis, focusing on the Ca2<sup>+</sup> toolkit genes (Robil et al., 2015), established the transcriptional profiles of these proliferative and quiescent GSLCs and revealed that genes regulating plasma membrane Ca2<sup>+</sup> channels (CACNA2D1 and ORAI2) and mitochondrial Ca2+-uptake (MCU, MICU1, MICU2, and VDAC1) are downregulated in quiescence while others are upregulated (CACNB1, CAPS, and SLC8B1). A functional

analysis through a bioluminescent Ca2<sup>+</sup> imaging approach showed that quiescence in GSLCs does not involve Cav1 channels like in HSCs, but is rather due to the modification of the kinetics of the store-operated Ca2<sup>+</sup> entry (SOCE), mediated by plasma membrane ORAI channels associated with the ER membrane protein STIM1. The inhibition of store-operated channels (SOC) by SKF96365 triggers quiescence, further supporting the crucial role of SOC in quiescence in GSLCs. Interestingly, the use of bioluminescent Ca2<sup>+</sup> reporter targeted to mitochondria revealed that this change in SOCE kinetics is due to an increased capacity of quiescent GSLCs' mitochondria to capture Ca2<sup>+</sup> and not to the modification of the SOCE mechanism itself (Aulestia et al., 2018). These data highlight the importance of mitochondria as regulator of Ca2<sup>+</sup> homeostasis.

Over the past decade, many studies have identified changes in the expression levels of proteins involved in Ca2<sup>+</sup> homeostasis such as Ca2<sup>+</sup> channels, pumps, and exchangers and established that some of these proteins contribute to tumorigenesis through regulation of proliferation, migration, or apoptosis (Monteith et al., 2012; Leclerc et al., 2016). As a second messenger, Ca2<sup>+</sup> is also an important regulator of gene expression. This occurs either indirectly, via changes in the transactivating properties of transcription factors following the activation of Ca2+-dependent kinases and/or phosphatases (Dolmetsch, 2001; West et al., 2001; Kornhauser et al., 2002; Spotts et al., 2002), or directly via EF hand Ca2+-binding proteins which belongs to a group of four proteins (KCNIP1–4) (Mellström et al., 2008). The identification of new therapeutic targets now requires not only to target the identified proteins but also to analyze the molecular mechanisms responsible for the changes in gene expression observed in cancer cells. In this review, we choose to focus on the direct mode of action of Ca2<sup>+</sup> on transcription with the implication of KCNIPs in GBM.

# THE FAMILY OF NEURONAL Ca2<sup>+</sup> SENSORS: KCNIPs

Potassium channel-interacting proteins (KCNIPs), which constitute the class E of Ca2<sup>+</sup> sensor family, are globular proteins of 217–270 amino acids in size, with variable N- and C-termini and a conserved core domain containing four EF-hand Ca2+ binding motifs (EF-1, EF-2, EF-3, and EF-4). Among the four EF hands, EF-1 is not able to bind Ca2<sup>+</sup> (Buxbaum et al., 1998; Carrión et al., 1999; An et al., 2000). Drosophila melanogaster has a single Kcnip, whereas mammals have four KCNIPs (KCNIP1–4) and several alternatively spliced variants (Burgoyne, 2007). In mammals, the four KCNIPs are predominantly expressed in adult brain, with specific or overlapping patterns according to the tissues (Rhodes, 2004; Xiong et al., 2004; Pruunsild and Timmusk, 2005). KCNIP3, also called calsenilin, KChIP3, and DREAM [i.e., Downstream Regulatory Element (DRE) Antagonist Modulator] is also found in the thyroid gland (Dandrea et al., 2005; Rivas et al., 2009) and in the hematopoietic progenitor cells (Sanz, 2001). KCNIP2 and KCNIP3 are found in T and B lymphocytes (Savignac et al., 2005, 2010). During mouse development, Kcnip3 transcript first occurs at E10.5 (Spreafico et al., 2001) and Kcnip1, 2, and 4 are not detected before E13 (Pruunsild and Timmusk, 2005). In the fish Danio rerio, the embryonic expressions of kcnip1b and kcnip3 are not detectable before somitogenesis (Stetsyuk et al., 2007) and in the amphibian Xenopus laevis among the four kcnips, only kcnip1 is expressed at all developmental stages, from fertilized egg to the tadpole stages. By contrast, the transcripts for kcnip2, kcnip3, and kcnip4 are expressed at later stages, after the specification of neural territories (Néant et al., 2015).

KCNIP proteins are known to control gene transcription directly by binding, via a Ca2+-dependent mechanism, to specific DNA sites, called DRE, of target genes. DRE sites are localized in the proximal 5<sup>0</sup> sequence of the gene, downstream of the TATA box and upstream of the start codon, with the sequence GTCA forming the central core of the DRE site (Carrión et al., 1999; Ledo et al., 2000). This mechanism has been particularly well described for KCNIP3 (DREAM). When the intracellular Ca2<sup>+</sup> level is low, KCNIP3 is bound as tetramer to the DRE sites, acting mainly as a transcriptional repressor. An increase in intracellular Ca2<sup>+</sup> leads to dissociation of the KCNIP3 tetramer from its DRE site, thus allowing transcription (Carrión et al., 1999). KCNIP3 can affect transcription by acting either as a transcriptional repressor (Carrión et al., 1999; Link, 2004) or activator (Scsucova, 2005; Cebolla et al., 2008). In a more recent study, KCNIP3 has been shown to be required for human embryonic stem cells (hESCs) survival and to maintain hESCs pluripotency (Fontán-Lozano et al., 2016). KCNIP3 was initially the only Ca2<sup>+</sup> sensor known to bind to DRE sites and to directly regulate transcription in a Ca2+-dependent manner (Mellström and Naranjo, 2001). However, all the four KCNIPs exhibit DRE-binding site affinity as homo or heterotetramers and act as Ca2+-dependent transcriptional regulators (Osawa et al., 2001; Craig et al., 2002; Link, 2004), allowing functional redundancy. KCNIP2 and KCNIP3 interactions are indirectly evidenced by two-hybrid and immunoprecipitation experiments (Savignac et al., 2005) and by the fact that KCNIP3 and KCNIP2 are both able to physically interact with EF-hand mutated KCNIP3 and that such associations still inhibit DRE-dependent gene expression (Gomez-Villafuertes, 2005; Savignac et al., 2005). In vivo studies also argue for the existence of compensatory mechanisms and the formation of functional KCNIP heterotetramers. Particularly, while in cortico-hippocampal neurons from Kcnip3 knockdown mice, the expression levels of KCNIP3 target genes such as Npas4 and cfos are not significantly modified, the additional invalidation of Kcnip2 with an antisense lentiviral vector (in this Kcnip3 KO context) results in a significant increase in the expression of these KCNIP3-dependent target genes (Mellström et al., 2014). In amphibian embryos, we demonstrated that Kcnip1 binds DRE sites in a Ca2+-dependent manner. Kcnip1 is the earliest kcnip gene expressed in X. laevis embryo. Its transcripts are timely and spatially present in the presumptive neural territories. In this in vivo model, loss of function experiments indicate that Kcnip1 is a Ca2+-dependent transcriptional repressor that controls the size of the neural plate by regulating the proliferation of neural progenitors (Néant et al., 2015).

### KCNIP PROTEINS IN GLIOBLASTOMA

To the best of our knowledge, no published work has analyzed the expression of KCNIPs in GSLCs or more generally in cancer stem cells. Using the UALCAN server (Chandrashekar et al., 2017), it was possible to compare gene expression in normal brain tissues versus GBM multiform. KCNIP1–4 are expressed in normal tissues at comparable levels. Interestingly, in GBM tissues while KCNIP1 is significantly upregulated compared to its levels in normal brain tissues, KCNIP2 and KCNIP3 are strongly downregulated (**Table 1**). Although KCNIP4 expression appears downregulated in GBM, the results are not statistically significant. This is probably due to large variability of KCNIP4 expression in normal brain tissues and the small number of samples analyzed. In terms of survival, only KCNIP2 expression is relevant. Among GBM patients, those with high KCNIP2 expression appear to have a significant reduction in their overall survival time (UALCAN analysis). A recent study incidentally provides additional information on KCNIP expression in BT189 GSLC (Wang et al., 2018). Wang and coworkers analyzed the function of ING5, an epigenetic regulator overexpressed in GBM, and showed that ING5 promotes GSLCs self-renewal capabilities. Using the fluorescent Ca2<sup>+</sup> probe fluo3, these authors showed that [Ca2+]i increases in cells overexpressing ING5. This increase


Gene expression is presented as a number of transcripts for each KCNIP genes per million of total transcripts. Data extracted using the UALCAN server (http://ualcan.path.uab.edu/index.html). Genes and proteins symbols are formatted according to the specific conventions particular to each organism (www.biosciencewriters.com).


<sup>a</sup>Position upstream from the ATG; S, sense; AS, anti-sense; <sup>b</sup>putative DRE sites in close proximity. NF, not found, in accordance with the criteria in proximal 5<sup>0</sup> upstream sequence between the tata box and the start codon (see details in text).

in the resting Ca2<sup>+</sup> level is required to maintain GSLCs' self-renewal. Conversely, ING5 knockdown results in a strong reduction of the resting [Ca2+]i. To decipher further how ING5 is acting, they performed the transcriptomic analysis of GSLC cells where ING5 is knockdown. Among the differentially expressed genes, several Ca2<sup>+</sup> channels were identified as upregulated by ING5, including some subunits of L and P/Q types of voltage-gated Ca2<sup>+</sup> channels (CACNA1F, 1S, 1D, and 1C and CACN1A, respectively) and of transient receptor potential cation channels (TRPC3, C5, C4, and M1). Of note, close examination of this list revealed that KCNIP1–4 are indeed expressed in the BT189 GSLCs, although with different expression levels, and that KCNIP2 is upregulated by ING5 in this GLSC (see Supplementary Table S1 in Wang et al., 2018).

These data suggest a role of KCNIP proteins in stemness maintenance and dormant status of the GSLCs. The importance of KCNIPs in GBM is further emphasized by the presence of potential DRE sites within the proximal promoter of MCU and MICU2, two genes downregulated in quiescent GSLCs (Aulestia et al., 2018) and within the proximal promoters of TRPC5, TRPC4, and TRPM1, genes from the TRP family upregulated by the epigenetic factor ING5 in BT189 GSLC (Wang et al., 2018; **Table 2**).

#### REGULATION OF GBM PROGNOSIS GENES BY KCNIP PROTEINS

Ion channels are now considered as important actors in cancers. Recent studies using microarray datasets of glioma samples obtained from the CGGA (Chinese Glioma Genome Atlas) and the TCGA (The Cancer Genome Atlas) identified genes belonging to the Ca2<sup>+</sup> signaling machinery as new candidate genes that can predict GBM patients with high risk of unfavorable outcome (Wang et al., 2016; Zhang et al., 2017, 2018). These genes, listed in **Table 3**, are ion channels


<sup>a</sup>Position upstream from the ATG; S, sense; AS, anti-sense; <sup>b</sup>putative DRE sites in close proximity; <sup>c</sup> two putative DRE sites in tandem. NF, not found, in accordance with the criteria in proximal 5<sup>0</sup> upstream sequence between the tata box and the start codon (see details in text).

genes namely CACNA1H, a T-type Ca2<sup>+</sup> channel (Cav3.2); KCNN4, a potassium Ca2+-activated channel (KCa3.1); KCNB1, a voltage-gated potassium channel (Kv2.1); KCNJ10, a potassium Ca2+-activated channel (Kir4.1); and classified as Ca2+-binding protein; PRKCG, Ca2+-dependent serine/threonine protein kinase Cγ (PKCγ); PRKCB, Ca2+-dependent serine/threonine protein kinase Cβ (PKCβ); and CAMK2A, the Ca2+-calmodulindependent protein kinase IIα. KCNIP proteins are known to control gene transcription directly by binding to DRE sites. To test whether KCNIP proteins may be involved in the regulation of the expression of these selected prognosis genes, we searched for the presence of DRE sites within their proximal promoters. The CACNA1H and PRKCB genes present both one DRE potential site in their proximal promoter and KCNB1 presents two DRE-binding sites (**Table 3**). More exciting are the three and four putative DRE sequences exhibited by CAMK2A and KCNN4, respectively, ideally positioned between the TATA box and the start codon, within the highly conserved sequence of proximal promoter in primates (**Figure 1**). The CAMK2A proximal promoter is also particularly conserved in mouse compared to human (87%), their DRE sites respect orientation and repartition, even for tandem organization. This promising observation has to be tested for KCNIP binding efficiency. Recent evidence argues for the existence of functional DRE sites within the CAMK2A proximal promoter. KCNIP3 mutants with two amino acids substitution in the EF-hands two, three, and four are unable to respond to Ca2<sup>+</sup> and function as a constitutively dominant active (daDREAM) transcriptional repressor (Savignac et al., 2005). In transgenic mice with neuronal expression of this daDREAM, the CAMK2A mRNA level is reduced by 1.7-fold compared to wild type (Benedet et al., 2017). Mouse promoter for KCNN4 is conserved (79%), but in a lesser extend concerning DRE sequences. These sequence alignments for proximal promoters let guess a putative regulatory role of KNCIPs in the expression of some prognosis genes in GBM. Of note, not all of these prognosis genes exhibit DRE-like sites, as no DRE putative sequence was detected for KCNJ10 or PRKCG (**Table 3**), suggesting that KCNIPs are not the only transcriptional regulators directly implicated in the regulation of these prognosis genes, but the hypothesis of their contribution remains attractive. It is noteworthy that the previous results were obtained using transcriptomic data issued from DNA chips. When using the portal UALCAN (Chandrashekar et al., 2017) interfaced with the TCGA data base of transcriptomic cancer profiles obtained by RNA-seq techniques, only CACNA1H and KCNN4 expression levels are correlated with significant differences in survival curves. It is noticeable that these two genes present one and three DRE sites, respectively. Anyhow, the presence of these putative DRE sites on prognosis genes, suggests that remodeling of Ca2<sup>+</sup> homeostasis in GBM stem cells may contribute to the alteration of the expression of these prognosis genes. These preliminary observations urge for a more complete analysis

taking into account the high level of false negatives when using the transcriptomic signatures built from DNA chip data.

#### PERSPECTIVES/PROSPECT ON KCNIPS IN GBM

Although no specific data are available for KCNIPs' function in GBM or even cancers, one can speculate taking into account published functions of KCNIP in other cell types. KCNIPs are in fact multifunctional EF hand Ca2+-binding proteins and according to their interaction partners and subcellular localization one can discriminate at least three main functions: (1) regulation of cellular excitability, (2) regulation of intracellular signaling, and (3) control of transcription.

#### Control of Cellular Excitability

The control of cellular excitability which involves the formation of a macromolecular signaling complex between KCNIP1 or 2, the A-type Kv4 potassium channel, and the T-type Ca2<sup>+</sup> channel Cav3 (Anderson et al., 2010a,b) is unlikely to occur in GSLCs. Indeed, investigation of the electrophysiological properties of glioma cells revealed the absence of A-type potassium channels in these cells (Bordey and Sontheimer, 1998). Therefore, only the two other functions of KCNIP may be relevant to GBM physiology.

### Regulation of Intracellular Ca2<sup>+</sup> Signaling

In cardiomyocytes, KCNIP2 participates in the modulation of Ca2<sup>+</sup> release through ryanodine receptors (RyR) by interacting with the ryanodine modulator, presenilin (Nassal et al., 2017). The presenilin/KCNIP3 complex has also been shown to modulate IP3-mediated Ca2<sup>+</sup> release (Leissring et al., 2000). We have already shown that the unique drug able to kill quiescent GSLCs acts through a modulation of IP3 signaling (Dong et al., 2017).

#### Control of Transcription

As mentioned above, all KCNIPs can bind to DRE sites on DNA and directly control transcription. KCNIP3 (DREAM)

#### REFERENCES


can also interact with other transcription factors such as CREB and therefore affects transcription of genes that do not contain DRE sites (review in Rivas et al., 2011). Interestingly, in cardiomyocytes, it has been shown that the complex Ca2+/CAMK2 regulates nuclear translocation of KCNIP3 (Ronkainen et al., 2011). As CAMK2A has been identified as a prognosis gene in GBM (**Table 3**), such a mechanism is likely to occur in GBM.

In conclusion, since no experimental data exists for the moment in the literature concerning the function of KCNIP family in GBM, this opens a new field of research. In other models, KCNIPs have pleiotropic effects. Their well-known role as transcriptional repressors, and the presence of DRE sites in the promoter region of some GBM prognosis genes argue for a transcriptional function of KCNIPs in GBM. However, nontranscriptional roles have also to be considered more closely in the future.

#### AUTHOR CONTRIBUTIONS

IN, JH, M-CK, FA, MM, and CL designed the experiments. IN, FA, and CL performed and analyzed the experiments. IN, JH, M-CK, MM, and CL wrote the manuscript. JH, MM, and CL analyzed the data, provided financial support, and the final approval of manuscript. All authors reviewed the manuscript.

### FUNDING

This work was supported by the Centre National de la Recherche Scientifique (CNRS), Université de Strasbourg, Université Toulouse 3, by a joint grant from the Agence Nationale de la Recherche (ANR) given between France and Hong Kong to CL, JH, and MM (CalciumGlioStem ANR-13-ISV1-0004 and A-HKUST601/13), SATT Conectus (M-CK), and has been performed within the LABEX ANR-10-LABX-0034\_Medalis and received a financial support from French Government managed by "Agence Nationale de la Recherche" under "Programme d'investissement d'avenir." FA was supported by a grant from the ANR CalciumGlioStem.




**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 © 2018 Néant, Haiech, Kilhoffer, Aulestia, Moreau and Leclerc. 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) and the copyright owner(s) 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 Binding Properties and Physiological Functions of Recoverin

Jingjing Zang and Stephan C. F. Neuhauss\*

Institute of Molecular Life Sciences, University of Zurich, Zurich, Switzerland

Recoverin (Rcv) is a low molecular-weight, neuronal calcium sensor (NCS) primarily located in photoreceptor outer segments of the vertebrate retina. Calcium ions (Ca2+)-bound Rcv has been proposed to inhibit G-protein-coupled receptor kinase (GRKs) in darkness. During the light response, the Ca2+-free Rcv releases GRK, which in turn phosphorylates visual pigment, ultimately leading to the cessation of the visual transduction cascade. Technological advances over the last decade have contributed significantly to a deeper understanding of Rcv function. These include both biophysical and biochemical approaches that will be discussed in this review article. Furthermore, electrophysiological experiments uncovered additional functions of Rcv, such as regulation of the lifetime of Phosphodiesterase-Transducin complex. Recently, attention has been drawn to different roles in rod and cone photoreceptors. This review article focuses on Rcv binding properties to Ca2+, disc membrane and GRK, and its physiological functions in phototransduction and signal transmission.

Keywords: recoverin, phototransduction cascade, G protein-coupled receptor kinase, visual pigment phosphorylation, Ca2+ myristoyl switch

# INTRODUCTION

#### Edited by:

Karl-Wilhelm Koch, University of Oldenburg, Germany

#### Reviewed by:

Ching-Kang Jason Chen, Baylor College of Medicine, United States Lorenzo Cangiano, University of Pisa, Italy

\*Correspondence: Stephan C. F. Neuhauss stephan.neuhauss@imls.uzh.ch

Received: 31 August 2018 Accepted: 04 December 2018 Published: 20 December 2018

#### Citation:

Zang J and Neuhauss SCF (2018) The Binding Properties and Physiological Functions of Recoverin. Front. Mol. Neurosci. 11:473. doi: 10.3389/fnmol.2018.00473 Calcium ions (Ca2+) are multifaceted second messenger that regulate a large variety of signaling pathways in all cell types, including those of the brain and the retina. Specialized Ca2+-binding proteins, belonging to the neuronal calcium sensor (NCS) family, detect neuronal Ca2+ (Burgoyne et al., 2004; Weiss et al., 2010). One family member, the approximately 23-kDa protein Recoverin (Rcv), is mainly expressed in retinal photoreceptors. Like other family members, it contains four EF hand motifs and a myristoyl chain on the N-terminus. Only two of the EF hands (EF2 and EF3) bind Ca2+ across species (Senin et al., 2002; Lamb and Hunt, 2018). Rcv undergoes a conformational change upon Ca2+ binding in darkness and subsequently attaches to the disc membrane in photoreceptor outer segment (Tanaka et al., 1995; Ames et al., 1997). This in turn allows binding of G protein-coupled receptor kinases (GRK). Ca2+-bound Rcv inhibits GRK and thereby prevents visual pigment phosphorylation. Under bright light condition, Ca2+ dissociates from Rcv and the protein translocates to the inner segment. The released GRK is now able to phosphorylate and eventually deactivate the light activated visual pigment (R<sup>∗</sup> ; Strissel et al., 2005). This mechanism was postulated to be part of the negative Ca2+ feedback during light adaption. Recent studies demonstrate its additional roles in regulating PDE<sup>∗</sup> lifetime and synaptic transmission (Sampath et al., 2005; Chen et al., 2012, 2015).

**Abbreviations:** Ca2+, calcium ions; Rcv, Recoverin; GRK, G protein-coupled receptor kinase; [Ca2+]o, extracellular Ca2+ concentration; [Ca2+]<sup>i</sup> , intracellular Ca2+ concentration; GC, guanylyl cyclase; GCAP, guanylyl cyclase activating protein; R ∗ , light-activated visual pigment; PDE<sup>∗</sup> , PDE-Transducin complex; GAPs, GTPase-activating proteins; cGMP, cyclic guanosine monophosphate; CNG channels, cyclic nucleotide-gated ion channels; NMR, nuclear magnetic resonance spectroscopy; SPR, surface plasmon resonance; DEER, double electron−electron resonance; NBD, 4-chloro-7-nitro-1, 2,3-benzoxadiazole.

Here, we review recent insights into the mechanism of Rcv binding to Ca2+, phospholipid membranes and GRK, before focusing on the physiological importance of Rcv in regulating the phototransduction cascade and synaptic transmission.

# Ca2+ BINDING INDUCED CONFORMATIONAL TRANSITION

The three-dimensional structures of myristoylated Rcv in solution with no or 2 Ca2+ ions bound have been solved by nuclear magnetic resonance (NMR) spectroscopy (Tanaka et al., 1995; Ames et al., 1997). In the absence of Ca2+ binding, the myristoyl group is hidden in a deep hydrophobic pocket in the N-terminal domain, forming a compact structure termed as the T (tense) state (**Figure 1**). A protein conformational change is induced by the binding of two Ca2+ ions, which leads to the extrusion of the myristoyl group, forming an elongated structure termed as the R (relaxed) state. The Ca2+ induced exposure of the myristoyl group, known as Ca2+ myristoyl switch, allows Rcv binding to the phospholipid membranes only when intracellular Ca2+ concentration ([Ca2+]i) is high (Zozulya and Stryer, 1992; Dizhoor et al., 1993; Ames et al., 1995).

Recently, an intermediate conformational state, named as the I state, has been proposed following NMR relaxation dispersion and chemical shift analyses on <sup>15</sup>N-labeled Rcv (Xu et al., 2011). The I state coexists with the structurally different T state

Society.

are labeled alphabetically according to their order in the protein sequence. Adapted with permission from Timr et al. (2018). Copyright © 2018 American Chemical

([I]/[T]>1%) on a millisecond time scale with its myristoyl group still buried inside the protein structure. NMR cannot demonstrate the exact structure because of its short lifetime, but molecular dynamics (MD) simulations describe a more detailed process underlining the importance of the I state (Timr et al., 2018). When [Ca2+]<sup>i</sup> is low, Rcv fluctuates between the T and the I state. Stronger Ca2+ binding to the EF3 loop of I state stabilizes itself in response to the elevated [Ca2+]<sup>i</sup> environment. Following the second Ca2+ binding to the EF2 loop, Rcv moves into the R state by performing the myristoyl switch.

Many different approaches have been applied to study the conformational change of Rcv in the past decade. Surface plasmon resonance (SPR) is known as a powerful tool to measure interaction modes of proteins, peptides, and lipids (Koch, 2012). The subtle conformational dynamics of Rcv induced by 100 nM to 600 nM of Ca2+ have been demonstrated by this method (Dell'Orco et al., 2010). This concentration is in the range of the estimated physiological [Ca2+]<sup>i</sup> in intact photoreceptors, although some studies measure slightly lower values (Sampath et al., 1998, 1999; Woodruff et al., 2002). More importantly, SPR allows the simultaneous measurement of different Ca2+ binding proteins of the phototransduction cascade in response to the identical Ca2+ stimulus (Dell'Orco et al., 2012), mimicking the physiological condition during the light response. Double electron−electron resonance (DEER) analysis has also been used to demonstrate increased distance between two spin-labels at engineered cysteine residues on the Rcv surface upon Ca2+ binding, which indicates that Rcv forms dimer when [Ca2+]<sup>i</sup> is high (Myers et al., 2013). Myers and colleagues proposed that Ca2+-induced dimerization of Rcv may interact with Rhodopsin which also forms a dimer in the membrane (Fotiadis et al., 2006; Knepp et al., 2012). Meanwhile Rcv dimer binds to two GRKs, which brings these molecules close to Rhodopsin dimer. When [Ca2+]<sup>i</sup> drops upon light stimulation, Ca2+ free Rcv releases GRK, resulting in a rapid phosphorylation for Rhodopsin.

#### Rcv-DISC MEMBRANE INTERACTION

Many studies have demonstrated that the binding of Rcv to the membrane depends primarily on Ca2+ and the presence of the myristoyl group (Lange and Koch, 1997; Desmeules et al., 2002, 2007). When Ca2+ concentration is high, the extruded myristoyl group inserts into the disc membrane bilayer and allows the recruitment of Rcv at the membrane surface. However, the direct demonstration of membranebound Rcv is not available, because its three-dimensional structure is very hard to be determined by NMR in solution due to its slow reorientation rate. The membrane-bound Rcv structure as predicted by two-dimensional solid-state NMR suggests that the long molecular axis of Rcv is oriented at a 45◦ angle with respect to the membrane normal (Valentine et al., 2003). Several positively charged basic residues at the Rcv N-terminal are exposed closely towards the membrane surface. Although it is clear that the membrane insertion depends on both the presence of Ca2+ and the myristoyl switch, the detailed mechanism of how this is achieved is still elusive.

Numerous factors may regulate the Rcv binding process. Electrostatic interactions have become the focus of current investigation. Although neutral phosphatidylethanolamine and phosphatidylcholine constitute about 80% of the total phospholipid content in the rod outer segment disc membranes (Wu and Hubbell, 1993), Rcv preferentially binds to the negatively charged phosphatidylserine monolayers (Calvez et al., 2016). This preference disappears when the charges of the phosphatidylserine monolayer are shielded or the positively charged N-terminal residues of Rcv are mutated. This argues for a role of these residues in electrostatic interactions with the negatively charged phosphatidylserine in the presence of Ca2+. MD simulations support this notion, proposing that positively charged residues frequently contact the bilayer and that negatively charged phospholipids regulate the orientation of Rcv toward the membrane (Timr et al., 2017). On the other hand, deletion of 13 amino acids of the positively charged C-terminus does not change the binding property, although these residues have been proposed to be part of the electrostatic interactions (Senin et al., 2007).

The phospholipids in the rod outer segment disc membrane contain a high percentage of unsaturated acyl chains, suggesting that the disc membranes are highly fluid (Fliesler and Schroepfer, 1983). In order to assess the importance of membrane fluidity, Potvin-Fournier and colleagues compared membranes composed of the most abundant lipid phosphatidylcholine with either saturated or unsaturated acyl side chains (Potvin-Fournier et al., 2017). Solid-state NMR indicates that phosphatidylcholine with monounsaturated acyl chain (DOPC) produces the strongest Rcv immobilization, implying that Rcv binding favors the membrane with greater fluidity.

Another useful technique to study the Rcv-membrane interaction is site-specific fluorescent labeling (Yang et al., 2016). Traditional non-specific labeling for multiple lysines or specific single labeling for cysteine may damage Rcv structure and function. Especially Cys39, which is highly conserved in neuronal NCS proteins, has been shown to play an important role in regulating both Ca2+ and membrane binding of Rcv (Permyakov et al., 2012; Ranaghan et al., 2013). In this method, a p-azido-L-phenylalanine (AZF) group, which is a phenylalanine analog with an azide substituted in para position, is genetically engineered to replace either phenylalanine or tyrosine. In the case of Rcv, two phenylalanines are carefully selected because they have good accessibility to the solvent but do not affect membrane binding. Rcv is attached to a fluorescent probe 4-chloro-7-nitro-1,2,3-benzoxadiazole (NBD) via strainpromoted azide-alkyne cycloaddition. In this way, the membrane translocation of Rcv is visualized in a Ca2+ dependent manner and the membrane with negative spontaneous curvature or higher fluidity shows stronger Rcv binding in the presence of Ca2+ .

Little is known about how the insertion of the myristoyl moiety affects the lipid membrane. Although the insertion does not seem to disturb membrane integrity (Potvin-Fournier et al., 2018), it greatly impacts the electrochemical properties of the bilayer (Brand and Koch, 2018). Both the surface charge density and the surface pressure of membranes are reduced upon Rcv binding, supporting the importance of electrostatic interactions.

#### Ca2+-INDUCED INHIBITION OF GRK BY Rcv

The extrusion of the myristoyl group induced by Ca2+ exposes a hydrophobic groove, which consists of residues highly conserved in all NCS proteins (Tanaka et al., 1995; Ames et al., 1997). NMR demonstrates that Rcv binds with a functional N-terminal fragment consisting of the first 25 amino acid residues of GRK (RK25) in the presence of Ca2+, indicating an interaction between the hydrophobic surface of RK25 amphipathic helix and Rcv hydrophobic groove (Ames et al., 2006). The same study also shows that Rcv-bound RK25, but not RK25 alone binds to immobilized Rhodopsin. Moreover, N-terminal deletion mutants of GRK still contain a functional catalytic domain but lose the ability to phosphorylate Rhodopsin (Higgins et al., 2006). These observations suggest that Ca2+-Rcv is positioned between GRK and Rhodopsin in a ternary complex to sterically obstruct Rhodopsin recognition without affecting GRK catalytic activity. On the other hand, the C-terminal region of Rcv (residues 190–202) has been implicated as an internal modulator of Ca2+ sensitivity and therefore may also affect GRK binding (Weiergräber et al., 2006). Indeed, Rcv mutants with various deletions of the C-terminal region show a lower affinity to the N-terminus of GRK and a weaker inhibition on GRK activity (Zernii et al., 2011). Moreover, mutating phenylalanine of the GRK N-terminus, which is predicted to contact the C-terminus of Rcv, leads to a strong reduction of binding to Rcv as demonstrated by SPR spectroscopy. These results indicate a direct involvement of the C-terminus in interaction between Rcv and GRK, providing another mechanism to regulate the inhibitory effect of Rcv.

Most studies investigated the biochemical properties of rod specific Rcv. There is very little information of Rcv variants present in cone photoreceptors. Such cone specific Rcv variants are found in all vertebrates with the exception of mammals (Yamagata et al., 1990; Kawamura et al., 1996; Arinobu et al., 2010; Zang et al., 2015).

Cone opsin in the zebrafish retina is phosphorylated in a Ca2+-dependent manner, suggesting a similar role of cone Rcv in the regulation of visual pigment phosphorylation during light response (Kennedy et al., 2004). In carp, cone Rcv binds to N-terminus of both rod and cone GRKs with similar affinity (Arinobu et al., 2010). Although the concentration of cone Rcv in the cone outer segment is estimated to be 20 times higher than that of rod Rcv in the rod outer segment, their inhibitory activity on both GRKs and their Ca2+ dependency in vitro is very similar. Judging on [Ca2+]<sup>i</sup> measurements in darkness and determination of Rcv concentrations, the inhibition by cone Rcv is calculated to be 2.5 times higher than that by rods Rcv.

Interestingly, GRK is not the only binding partner of Rcv in the outer segment disc membrane. Caveolin-1, a major integral component of cholesterol-rich detergent-resistant lipid rafts of the rod disc membrane, has been shown to bind to Rcv in the absence of Ca2+ (Vladimirov et al., 2018). Although Caveolin-1 is not directly involved in phototransduction, its binding seems to increase Ca2+ affinity of Rcv, thereby allowing inhibition by Rcv even under low Ca2+ concentrations (Zernii et al., 2014). Therefore, Caveolin-1 binding may provide a mechanism which reserves small amounts of Rcv during bright light response or during light adaptation in outer segment membrane rafts and facilitates GRK inhibition upon Ca2+ change with high temporal resolution.

#### PHYSIOLOGICAL FUNCTIONS OF Rcv IN PHOTOTRANSDUCTION CASCADE

The phototransduction cascade starts with the absorption of a photon by the visual pigment rhodopsin (**Figure 2**), which is a member of the G protein-coupled receptor family (Burns and Baylor, 2001; Fain et al., 2001; Lamb and Pugh, 2006; Fu and Yau, 2007). Activated rhodopsin (R<sup>∗</sup> ) interacts with the trimeric G protein transducin, which in turn binds to its target effector enzyme, phosphodiesterase (PDE), resulting in cyclic guanosine monophosphate (cGMP) hydrolysis. The decrease in cGMP concentration leads to the closure of cyclic nucleotidegated ion (CNG) channels in the outer segment membrane, producing the electrical response to light. In darkness, CNG channels are partially open and there is a steady Ca2+ (and Na+) influx across the plasma membrane into the outer segment. This Ca2+ influx is reduced by the closure of the CNG channels during light response, while Ca2+ efflux via Na+/Ca2+ K + exchanger continues, leading to a decrease in [Ca2+]<sup>i</sup> in the outer segment. This decline in [Ca2+]<sup>i</sup> modulates the quenching of the phototransduction cascade, which requires the shutoff of both active intermediates (R<sup>∗</sup> and PDE<sup>∗</sup> ) and the resynthesis of cGMP. Under steady background illumination, the [Ca2+]<sup>i</sup> decline is proportional to the reduction in the circulating current and is crucial for lower photosensitivity and faster response kinetics during this process. What are the targets of Ca2+? The light-induced decrease in [Ca2+]<sup>i</sup> is postulated to accelerate phosphorylation of R<sup>∗</sup> via Rcv, speed up cGMP synthesis via guanylyl cyclase activating protein (GCAP), and increase cGMP affinity of the CNG channel via calmodulin (Koch and Stryer, 1988; Hsu and Molday, 1993).

Despite the fact that Rcv has been extensively shown to inhibit GRK in a Ca2+ dependent manner to regulate the phosphorylation of R<sup>∗</sup> in vitro (Kawamura, 1993; Chen et al., 1995, 2012, 2015; Klenchin et al., 1995; Kennedy et al., 2004; Chen C.-K. et al., 2010; Sakurai et al., 2011b; Invergo et al., 2013), its physiological function in the phototransduction cascade has remained controversial. Electrophysiological recordings on mouse rods lacking Rcv showed accelerated photoresponse decay, which is consistent with an inhibition of GRK by Rcv at high [Ca2+]<sup>i</sup> in darkness and a reduced lifetime of R<sup>∗</sup> in the mutants during light response (e.g., Makino et al., 2004; Chen C.-K. et al., 2010). However, other authors proposed

Recoverin; CNGC, cyclic nucleotide–gated ion channels; P, Phosphorylation; Figure was drawn using Inkscape. Inkscape http://www.inkscape.org/.

that both Rcv and GRK participate in the regulation of PDE activity (e.g., Chen et al., 2012, 2015). In addition, Rcv accumulates largely in the inner segment, cell body and synapse, indicating that Rcv may have greater importance in some processes other than phototransduction, such as signal transmission (Sampath et al., 2005; Zang et al., 2015). Moreover, the affinity of Ca2+ to Rcv is rather low with an EC<sup>50</sup> of about 3,000 nM and the light induced Ca2+ decline in the outer segment, which is in the range of 200 nM to 600 nM, may not even affect Rcv activity (Chen et al., 1995; Sampath et al., 1999; Woodruff et al., 2002), while a more recent study demonstrated a much lower EC<sup>50</sup> (400–550 nM; Arinobu et al., 2010). Furthermore, cone homologs of Rcv have been identified in many animals and have been shown to work in a similar way as rod Rcv (Yamagata et al., 1990; Gray-Keller et al., 1993; Kawamura et al., 1996; Arinobu et al., 2010; Zang et al., 2015; Lamb and Hunt, 2018). However, the physiological importance of Rcv may very well differ between photoreceptor types or even among species, considering the difference between rods and cones (e.g., Rcv concentration and [Ca2+]<sup>i</sup> dynamics).

# Rcv IN ROD PHOTOTRANSDUCTION

Rod photoreceptors operate at low light intensities and are able to be activated by the absorption of a single photon. It has been suggested that PDE<sup>∗</sup> deactivation (around 250 ms) is slower than R<sup>∗</sup> decay (around 50 ms) and dominates the overall photoresponse recovery in mouse rods (Krispel et al., 2006; Tsang et al., 2006; Doan et al., 2009; Chen C.-K. et al., 2010; Gross and Burns, 2010), but see (Doan et al., 2009). If the major role of Rcv is to inhibit GRK, preventing the Zang and Neuhauss Recoverin Functions

phosphorylation of R<sup>∗</sup> when Ca2+ is high, its deletion should result in a shortened lifetime of R<sup>∗</sup> and decrease the gain of signaling pathway. This in turn should lead to lower light sensitivity, without strongly affecting the time course of response decay. However, electrophysiological recordings from Rcv−/<sup>−</sup> mice showed an accelerated dark-adapted flash response, but only a minor effect on response sensitivity (Makino et al., 2004; Sampath et al., 2005; Chen J. et al., 2010). Although phosphorylation is essential for R<sup>∗</sup> shutoff (Chen et al., 1999), the phosphorylation regulated by Rcv seems to have a limited impact on photosensitivity, which corresponds to little effect of GRK expression levels on dim light amplitude in mice. Interestingly, absence of Rcv reduces around 50% of single photon response amplitude in GCAPs−/<sup>−</sup> background, which indicates that the sensitivity regulation mediated by Rcv may be masked by a more powerful feedback mediated by GCAPs (Makino et al., 2004; Vinberg et al., 2015). However, both the dominant time constant (τD) and the exponential decay constant of the response (τREC) are significantly accelerated in Rcv−/<sup>−</sup> rods (Chen et al., 2012, 2015). In order to reduce τD, the effect of Rcv depletion must be mediated via shortening the lifetime of the rate-limiting step. PDE<sup>∗</sup> decay has been proven to still remain rate-limiting for response recovery both in Rcv−/<sup>−</sup> rods in WT background and in Rcv−/<sup>−</sup> rods with delayed PDE<sup>∗</sup> quenching. In this case, a novel role of Rcv regulating PDE<sup>∗</sup> deactivation is the most parsimonious explanation.

During light adaptation, rod sensitivity is reduced and response recovery is accelerated when background illumination increases, which extends the working range of rods to brighter light levels (Fain et al., 2001). Rcv, GCAP and Calmodulin are all Ca2+ binding proteins, which have been proposed to contribute to the light adaptation. Surprisingly, deletion of Rcv or Calmodulin binding site on CNG channels has little impact on sensitivity and removal of GCAPs affects only some but not all the changes in sensitivity during background adaption. Furthermore, no impact of Calmodulin binding site deletion in GCAPs−/<sup>−</sup> rods is observed (Chen J. et al., 2010). Perfusing rods with low Ca2+ solution in darkness can mimic the [Ca2+]<sup>i</sup> reduction during light adaptation (Vinberg et al., 2015). This method can decrease the single photon response amplitude of GCAPs−/<sup>−</sup> rods and GCAPs−/<sup>−</sup> Rcv−/<sup>−</sup> rods to the same level, about 25% of the value in GCAPs−/<sup>−</sup> rods in normal solution. This may suggest other Ca2+ dependent mechanism(s).

In the case of time course, an increase in the background intensity produces a speed up in light response recovery and a progressive decrease in τ<sup>D</sup> nearly in proportion to background intensities (Fain et al., 2001; Woodruff et al., 2008; Chen et al., 2015). Deleting PDE<sup>Υ</sup> subunit eliminates this gradual reduction in τD, suggesting that the accelerated response recovery during light adaptation primarily depends on PDE<sup>∗</sup> turn off. Removal of Rcv from the genome yields the same effect. Moreover, τ<sup>D</sup> produced by different background intensities in Rcv−/<sup>−</sup> rods is very close but slightly longer than Rcv+/<sup>+</sup> rods under the brightest ambient intensity, indicating the regulation of Rcv on PDE<sup>∗</sup> contributes the majority of the acceleration during light adaption, but some other mechanism(s) may still be involved. Is this mechanism Ca2+ dependent? Background light can still accelerate the response in Rcv−/<sup>−</sup> rods recorded in low extracellular Ca2+ concentration ([Ca2+]o) which in theory can clamp [Ca2+]<sup>i</sup> to its maximal light-adapted level, indicating a possible Ca2+-independent process (Vinberg et al., 2015). Furthermore, during background light exposure, photoresponse integration time declines while the circulating current gradually increases (Morshedian et al., 2018). Both effects work to increase sensitivity at dim light condition near the threshold and improve the temporal resolution under bright light condition, but these effects disappear in Rcv−/<sup>−</sup> rods. An effect on spontaneous and light activated PDE lifetime may contribute to these observations.

In short, Rcv can only regulate photosensitivity in the absence of Ca2+ feedback mediated via GCAPs in mouse rods. Another Ca2+ sensitive mechanism, other than mediated by Calmodulin, must be involved in this regulation. Rcv modulation on PDE<sup>∗</sup> contributes the majority of the regulation on response kinetics.

How exactly Rcv may work on PDE<sup>∗</sup> is still unknown. Most likely its regulation works still via GRK, because the change in photoresponse kinetics is very similar in rods overexpressing GRK and in rods without Rcv (Sakurai et al., 2011b; Chen et al., 2012, 2015). Most likely Rcv binds to Ca2+ and inhibits GRK in darkness, while during light response, Ca2+-free Rcv releases from GRK, enabling phosphorylation of target protein(s) and the deactivation of PDE<sup>∗</sup> . GTPase-activating proteins (GAPs) accelerate the turn off rate of Transducin. Deletion of GRK in GAPs deficient rods produces little effect on the time course of response, suggesting that the target protein(s) may be involved in the PDE-Transducin pathway (Chen et al., 2015). However, there is no biochemical evidence to support this novel function of GRK and the electrophysiological recordings from rods overexpressing GRK show contradictory results in different studies. For example, a threefold overexpression of GRK is able to speed up the response kinetics of dim light (indicated by τREC) but not saturating light (indicated by τD; Sakurai et al., 2011b). Because τ<sup>D</sup> is unaffected, GRK most likely regulates R <sup>∗</sup> decay which is certainly shorter than the lifetime of the rate-liming step PDE<sup>∗</sup> , but close enough to affect the overall dim light response recovery. In contrast, when GRK is 12-fold overexpressed, both τREC and τ<sup>D</sup> are reduced, suggesting a direct modulation of GRK on PDE<sup>∗</sup> (Chen et al., 2012). Furthermore, a mathematical model of phototransduction partially predicts the experimental effects of GRK downregulation and overexpression on response kinetics without consideration of an interaction between GRK and PDE<sup>∗</sup> (Invergo et al., 2013). Therefore, the possible mechanism underlining GRK regulation needs to be further investigated.

Rcv has indeed been shown to regulate the lifetime of R<sup>∗</sup> in intact rods. When GRK1 is overexpressed and R<sup>∗</sup> turnoff becomes rate-limiting for light response termination (Chen C.-K. et al., 2010), response acceleration during light adaptation disappears in Rcv−/<sup>−</sup> rods. However, together with other studies mentioned above, the regulation of Rcv on PDE<sup>∗</sup> may play a more significant role.

Interestingly, the situation in amphibian rods appears somewhat different. The overall response recovery or the rate-limiting step is insensitive to [Ca2+]<sup>i</sup> (Lyubarsky et al., 1996; Matthews, 1996). However, there is a Ca2+ sensitive step early in the light response, which can be prolonged to dominate photoresponse kinetics by prolonging the lifetime of R ∗ (Matthews et al., 2001). Therefore, unless there is some novel mechanism, this Ca2+ sensitive step most likely represents the Rcv mediating phosphorylation of R<sup>∗</sup> , while the lifetime of PDE<sup>∗</sup> which seems to dominate the response recovery in amphibian rods may not regulated by Ca2+ (Nikonov et al., 1998). This proposal is also supported by the fact that τ<sup>D</sup> is insensitive to different background lights in salamander rods (Pepperberg et al., 1992; Nikonov et al., 2000), in stark contrast to the Ca2+ sensitive τ<sup>D</sup> and background sensitive τ<sup>D</sup> in mice (Woodruff et al., 2008; Chen et al., 2015; Vinberg et al., 2015).

#### Rcv IN CONE PHOTOTRANSDUCTION

Cone photoreceptors are mainly responsible for daytime vision and mediate color vision (Fu and Yau, 2007). They share a similar G-protein signaling pathway with rods, but their response is characterized by lower sensitivity and faster kinetics. Cones are capable to operate over a nine-order of magnitude intensity range, which is much larger than the range typically observed in rods. In addition, cones do not saturate under bright background illumination, suggesting a more powerful light adaptation mechanism (Baylor and Hodgkin, 1974; Matthews et al., 1990; Schneeweis and Schnapf, 1999). Indeed, light induced Ca2+ decline is much faster and the Ca2+ dynamic range is three times wider in cones than in rods (Sampath et al., 1999, 1998; Woodruff et al., 2002). Ca2+ regulation on GC via GCAPs is much weaker in mouse cones than rods (Sakurai et al., 2011a). Furthermore, the elevated cGMP turnover observed in background light, which accounts for many adaptive changes in salamander rod response, is estimated to be much higher in rods than in cones (Cornwall and Fain, 1994; Cornwall et al., 1995; Hodgkin and Nunn, 1988; Nikonov et al., 2000). Therefore, the Ca2+ feedback on GCAPs seems not to contribute to the difference between photoreceptor types during light adaptation. In contrast, the range of Ca2+-dependent regulation of cone homolog of GRK (GRK7) activity in cones is more than 100 times greater than the range of regulation of rod homolog of GRK (GRK1) activity in rods, suggesting that the phosphorylation mediated by Rcv may contribute more during light adaption in cones (Arinobu et al., 2010). Overall, the details of Rcv modulation in cone photoreceptors are not fully understood.

Mammalian photoreceptors share the same Rcv. Mouse photoreceptors even share the same GRK (GRK1), therefore, they presumably work together in a similar way in both cell types. Electroretinography (ERG) on Rcv deficient cones shows not only an accelerated response recovery, but also an around twofold decrease in photosensitivity in darkness, which differs from rods (Sakurai et al., 2015). This result is consistent with in vitro evidence for light-dependent phosphorylation of cone opsin by GRK1 (Zhu et al., 2003). During light adaptation, photosensitivity is clearly lower in Rcv−/<sup>−</sup> cones than in controls in dim background light and becomes identical under bright light conditions (Sakurai et al., 2015). On the one hand, it is likely that the bright background light reduces [Ca2+]<sup>i</sup> to a certain level that all the Rcv becomes Ca2+ free in control and has no ability to inhibit GRK. On the other hand, it is also possible that the spontaneous decay primarily contributes to quench the cone opsin when the light intensity is very high. This possibility is consistent with the independency of response kinetics on the expression levels of GRK when high percentage of cone pigment is bleached, which is demonstrated in the same study. Nevertheless, the response sensitivity continues to decrease in both mutant and control when background intensity increases further. These observations suggest that additional mechanism(s) must be at work to regulate the photoresponse, for example via GCAPs (Sakurai et al., 2011a). Surprisingly, the effect of Rcv removal is similar to GRK1 knockdown but opposite to GRK1 overexpression in cones. GRK1 knockout slows down the photoresponse as expected. The turning point between GRK1 knockdown and knockout is unknown, making it hard to estimate the molecular mechanism underlining this unexpected response kinetics. In other mammalian cones, the situation may be even more complicated with an additional cone homolog of GRK (GRK7) being present.

In salamanders, single cell recording indicates that cone photoresponse decay is dominated by the quenching of cone opsin and it is sensitive to [Ca2+]<sup>i</sup> (Matthews and Sampath, 2010; Zang and Matthews, 2012). Although this Ca2+ sensitive process is not specified, cone opsin phosphorylation mediated by Rcv seems probable and this process allows Ca2+ directly controlling the lifetime of rate-limiting step, in contrast to salamander rods.

In zebrafish, cone opsin phosphorylation shows strong light and Ca2+ dependence (Kennedy et al., 2004). Four Rcvs have been identified in zebrafish (Zang et al., 2015). All four of them are expressed in cones with only one of them existing in rods as well. ERG response recovery accelerates in Rcv(s) knockdown animals and some Rcvs operate at varying light intensities, suggesting different Ca2+ sensitivity and (or) GRK affinity among Rcvs. Indeed, a recent biochemical work shows different Ca2+ affinity and Ca2+-induced conformational changes among zebrafish Rcv isoforms and bovine Rcv (Elbers et al., 2018). Interestingly, the amino acids at the critical positions of Ca2+ binding sites are completely conserved across species with the exception of one zebrafish Rcv, where only three out of four positions are conserved in EF2 (Lamb and Hunt, 2018). This may at least partially contribute to their different molecular properties. Downregulation of GRK7 largely delays the response and Rcv knockdown in this background has no effect, consistent with the function of Rcv to inhibit GRK in vitro (Rinner et al., 2005). Notably, the effect of GRK7 knockdown in zebrafish seems to work the opposite way when compared to the situation in mouse cones, but the downregulation level here is around 95% and may represent the situation of GRK1−/<sup>−</sup> in mice. Therefore, whether GRKs function the same way in mouse and zebrafish cones is still unknown.

In summary, Rcv certainly plays a role in regulating the cone phototransduction decay. However, as the popular animal models of mouse and bovine have rod-dominant retina, our knowledge about the function of Rcv in cones is very limited. Different roles of Rcv in rods and cones may very well underlie overall differences in their photoresponse.

#### ROLE OF Rcv IN SIGNAL TRANSMISSION

Rcv is expressed in all photoreceptors and also in some subtypes of bipolar cells (Dizhoor et al., 1991; Milam et al., 1993; Haverkamp and Wässle, 2000; Strissel et al., 2005; Zang et al., 2015). Most of Rcv has been quantitatively shown to accumulate in rod inner segment in both light and dark adapted retina in mice (Strissel et al., 2005). Light induces a remarkable translocation of Rcv from the outer and inner segments towards the synaptic terminals. Interestingly, GRK remains in the outer segments independent of illumination. Those observations are not only consistent with the well-documented role of Rcv releasing GRK upon [Ca2+]<sup>i</sup> reduction in the outer segment, but also indicate some other function(s) of Rcv which is independent from GRK in other cellular compartments of photoreceptors.

Indeed, although the light sensitivity is not affected in Rcv−/<sup>−</sup> rods, the sensitivity of rod-mediated vision is reduced in the behavior assay (Sampath et al., 2005). The same study also demonstrate that the dim light-evoked response of rod bipolar cells and ganglion cells, which do not express Rcv, is shortened in Rcv knockout mice. This decreased response time is produced by the reduced signal transfer from rods to rod bipolar cells instead of rod phototransduction itself.

#### CONCLUSION

Here, we reviewed the molecular process underlining Ca2+ induced membrane and GRK binding of Rcv. Moreover we

#### REFERENCES


summarized the physiological function of Rcv in regulating phototransduction cascade and in signal transmission. Rcv is clearly capable of regulating phototransduction decay in dark-adapted as well as light-adapted response, but the molecular mechanisms are still not entirely clear. Biochemical evidence supporting or excluding the modulation of Rcv on PDE<sup>∗</sup> is required. Moreover, animal models favored by both biochemists and electrophysiologists have traditionally been roddominant. Hence our knowledge of cone phototransduction regulation lags behind. Therefore, the importance of Rcv cone homolog(s) and the potential function difference between rod and cone Rcv will need to become a focus of future research.

#### AUTHOR CONTRIBUTIONS

JZ and SN drafted and revised the review article.

#### FUNDING

This work was supported by the Swiss National Science foundation (31003A\_173083).

#### ACKNOWLEDGMENTS

We would like to thank Dr. Matthias Gesemann for comments on the manuscript.

rhodopsin kinase. Proc. Natl. Acad. Sci. U S A 96, 3718–3722. doi: 10.1073/pnas. 96.7.3718


myristoyl switch protein, and lipid bilayers by AFM-based force spectroscopy. Biophys. J. 82, 3343–3350. doi: 10.1016/S0006-3495(02)75674-9


phototransduction. J. Biol. Chem. 270, 16147–16152. doi: 10.1074/jbc.270.27. 16147


Koch, K. W., and Stryer, L. (1988). Highly cooperative feedback control of retinal rod guanylate cyclase by calcium ions. Nature 334, 64–66. doi: 10.1038/334064a0


immobilization in zwitterionic lipids. Biochem. Biophys. Res. Commun. 490, 1268–1273. doi: 10.1016/j.bbrc.2017.07.005


of mammalian rods to brighter light. J. Gen. Physiol. 146, 307–321. doi: 10.1085/jgp.201511412


**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 © 2018 Zang and Neuhauss. 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) and the copyright owner(s) 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.

# Light-Induced Thiol Oxidation of Recoverin Affects Rhodopsin Desensitization

Evgeni Yu. Zernii1,2, Aliya A. Nazipova<sup>3</sup> , Ekaterina L. Nemashkalova<sup>3</sup> , Alexey S. Kazakov<sup>3</sup> , Olga S. Gancharova1,4, Marina V. Serebryakova<sup>1</sup> , Natalya K. Tikhomirova<sup>1</sup> , Viktoriia E. Baksheeva<sup>1</sup> , Vasiliy I. Vladimirov<sup>5</sup> , Dmitry V. Zinchenko<sup>5</sup> , Pavel P. Philippov<sup>1</sup> , Ivan I. Senin<sup>1</sup> and Sergei E. Permyakov<sup>3</sup> \*

<sup>1</sup> Belozersky Institute of Physico-Chemical Biology, Lomonosov Moscow State University, Moscow, Russia, <sup>2</sup> Institute of Molecular Medicine, Sechenov First Moscow State Medical University, Moscow, Russia, <sup>3</sup> Institute for Biological Instrumentation of the Russian Academy of Sciences, Pushchino, Russia, <sup>4</sup> Institute for Regenerative Medicine, Sechenov First Moscow State Medical University, Moscow, Russia, <sup>5</sup> Shemyakin and Ovchinnikov Institute of Bioorganic Chemistry of the Russian Academy of Sciences, Pushchino, Russia

Edited by:

Daniele Dell'Orco, Università degli Studi di Verona, Italy

#### Reviewed by:

James Ames, University of California, Davis, United States Pere Garriga, Universitat Politecnica de Catalunya, Spain

> \*Correspondence: Sergei E. Permyakov

permyakov.s@gmail.com

Received: 27 September 2018 Accepted: 05 December 2018 Published: 07 January 2019

#### Citation:

Zernii EY, Nazipova AA, Nemashkalova EL, Kazakov AS, Gancharova OS, Serebryakova MV, Tikhomirova NK, Baksheeva VE, Vladimirov VI, Zinchenko DV, Philippov PP, Senin II and Permyakov SE (2019) Light-Induced Thiol Oxidation of Recoverin Affects Rhodopsin Desensitization. Front. Mol. Neurosci. 11:474. doi: 10.3389/fnmol.2018.00474 The excessive light illumination of mammalian retina is known to induce oxidative stress and photoreceptor cell death linked to progression of age-related macular degeneration. The photochemical damage of photoreceptors is suggested to occur via two apoptotic pathways that involve either excessive rhodopsin activation or constitutive phototransduction, depending on the light intensity. Both pathways are dramatically activated in the absence of rhodopsin desensitization by GRK1. Previously, we have shown that moderate illumination (halogen lamp, 1,500 lx, 1–5 h) of mammalian eyes provokes disulfide dimerization of recoverin, a calcium-dependent regulator of GRK1. Here, we demonstrate under in vivo conditions that both moderate long-term (metal halide lamp, 2,500 lx, 14 h, rat model) and intense short-term (halogen lamp, 30,000 lx for 3 h, rabbit model) illumination of the mammalian retina are accompanied by accumulation of disulfide dimer of recoverin. Furthermore, in the second case we reveal alternatively oxidized derivatives of the protein, apparently including its monomer with sulfinic group. Histological data indicate that thiol oxidation of recoverin precedes apoptosis of photoreceptors. Both disulfide dimer and oxidized monomer (or oxidation mimicking C39D mutant) of recoverin exhibit lowered α-helical content and thermal stability of their apo-forms, as well as increased Ca2<sup>+</sup> affinity. Meanwhile, the oxidized monomer and C39D mutant of recoverin demonstrate impaired ability to bind photoreceptor membranes and regulate GRK1, whereas disulfide dimer exhibits notably improved membrane binding and GRK1 inhibition in absence of Ca2+. The latter effect is expected to slow down rhodopsin desensitization in the light, thereby favoring support of the light-induced oxidative stress, ultimately leading to photoreceptor apoptosis. Overall, the intensity and duration of illumination of the retina affect thiol oxidation of recoverin likely contributing to propagation of the oxidative stress and photoreceptor damage.

Keywords: light-induced retinal damage, photoreceptor, apoptosis, neuronal calcium sensor, recoverin, thiol oxidation, disulfide dimerization, GRK1

# INTRODUCTION

fnmol-11-00474 December 24, 2018 Time: 17:25 # 2

Photochemical damage is the most common form of the light-induced retinal damage. The prolonged eye exposure to sunlight or artificial light sources may cause PD of the retina, associated with progression of light-induced retinopathies (light maculopathy) (Organisciak and Vaughan, 2010; van Norren and Vos, 2016; Zernii et al., 2016). Being accumulated over years, photochemical injuries of the retina were suggested to provoke AMD, the major cause of blindness in the elderly worldwide (Loeffler et al., 2001; Plestina-Borjan and Klinger-Lasic, 2007 ´ ; Vojnikovic et al., 2007 ´ ). Furthermore, recent advances in ophthalmology and ophthalmic surgery increased incidence of iatrogenic PD of the retina caused by illumination of slit lamps, indirect ophthalmoscopes, fiberoptic endoilluminators and light sources of operative microscopes (Hunter et al., 2012; Wolffe, 2016). Thus, the establishment of molecular mechanisms underlying PD is an urgent task, as it will allow identifying approaches to the prevention and treatment of both iatrogenic and age-related retinal diseases.

Photoreceptor cells were recognized as primary targets of retinal PD. The damage is initiated in distal tips of their outer segments eventually engaging the entire cell (Organisciak and Vaughan, 2010; Hunter et al., 2012). There are several hypotheses of light-induced damage to photoreceptors, but they all consider rhodopsin as a key trigger for cell death pathways (Grimm et al., 2000; Organisciak and Vaughan, 2010). The main pathogenic factor is thought to be the reactive oxygen species, generated by the bleaching of rhodopsin. The light-induced oxidative stress results in oxidation of various molecules, including lipids and proteins, and ultimately leads to apoptosis of photoreceptors (Organisciak and Vaughan, 2010). Two major apoptotic pathways were suggested to become activated, depending on intensity of the light illumination. Low-intensity light induces constitutive transducin activation and excessive phototransduction that may activate PERK pathway of the unfolded protein response (Hao et al., 2002; Wang and Chen, 2014). By contrast, bright light induces apoptosis independently of transducin, but involves activation of transcription factor AP-1 (Hao et al., 2002). The specific mechanisms that trigger photoreceptor apoptosis pathways remain obscure, although they were suggested to depend on rhodopsin desensitization by GRK1 and arrestin (Chen et al., 1999a,b) and may include thiol oxidation of photoreceptor proteins (Lieven et al., 2012; Zernii et al., 2015b). Indeed, specific sulfur reductant protected photoreceptors from light-induced degeneration in vivo (Lieven et al., 2012).

The redox proteomic screening of retinal extracts identified appearance of disulfide homodimers of visual arrestin in response to toxic levels of light (Lieven et al., 2012). Visual arrestin terminates the phototransduction cascade by binding to phosphorylated light-activated rhodopsin (Gurevich et al., 2011). Hanson et al. revealed that oligomeric form of arrestin looses ability to bind the light-activated phosphorhodopsin (Hanson et al., 2007). Consistently, Lieven et al. (2012) demonstrated that disulfide dimers of visual arrestin generated in the retina during photic injury did not form stable complexes with its physiological binding partners, rhodopsin and enolase 1. Thus, disulfide dimerization of arrestin could contribute to light/oxidative stress-induced cell death pathways via affecting rhodopsin desensitization or activity of other arrestin targets (Song et al., 2006).

Recently, we reported that recoverin is one more photoreceptor protein undergoing light-induced disulfide dimerization under ex vivo and in vivo conditions (Zernii et al., 2015b). Recoverin is a Ca2+-sensor membrane-binding protein that serves as a Ca2+-dependent inhibitor of GRK1 in retinal rod cells. It coordinates two calcium ions by the second and the third EF-hand motifs thereby exposing its GRK1-recognizing site and N-terminal myristoyl group according to the mechanism known as a 'Ca2+-myristoyl switch' (for reviews, see Ames and Lim, 2012; Philippov and Zernii, 2012). Recoverin belongs to NCS proteins, which are expressed in the brain and retina where they transduce calcium signals in a wide range of signaling pathways and they are genetically linked to degenerative diseases (for reviews, see Burgoyne, 2007; Ames and Lim, 2012; Koch and Dell'Orco, 2015). A characteristic feature of the NCS family members is a highly conserved cysteine residue located in the third position of their first non-functional EF-hand motif (C39 in recoverin), which is sensitive to redox conditions. Recoverin was the first NCS protein that was shown to exhibit redox sensitivity of the thiol (Permyakov et al., 2007). Some of recoverin orthologs contain only one conservative cysteine (in mice, rats and cows), whereas the others include one more cysteine (in rabbits and humans). The conservative cysteine Cys39 of bovine recoverin undergoes oxidation under mild oxidizing in vitro conditions, forming a disulfide dimer and a thiol oxidized monomeric form (Permyakov et al., 2007). Recoverin mutant C39D that mimics oxidative conversion of Cys39 into sulfenic, sulfinic or sulfonic acids, exhibits decreased α-helicity and thermal stability, as well as suppressed affinity to photoreceptor membranes and GRK1 (Permyakov et al., 2012; Ranaghan et al., 2013). Disulfide dimer of recoverin as well as its multimeric/aggregated forms are accumulated in retina of the experimental animals subjected to moderate illumination of eyes with visible light (1,500 lx, 1–5 h), while monomeric recoverin remains mostly reduced (Zernii et al., 2015b). Histologic study demonstrates that the light-induced oxidation of recoverin occurs in intact retina and precedes damage of the photoreceptor layer. Taken together, these data suggest involvement of thiol oxidation of recoverin in triggering of photoreceptor cell damage/death mechanisms.

**Abbreviations:** AMD, age-related macular degeneration; bis-ANS, 4,4'-dianilino-1,1'-binaphthyl-5,5'-disulfonic acid; CD, circular dichroism; C39D, recoverin mutant with C39 replaced by aspartic acid; DTT, dithiothreitol; dRec, disulfide dimer of recombinant wild type recoverin; EDTA, ethylenediaminetetraacetic acid; EGTA, ethylene glycol-bis(2-aminoethylether)-N,N,N',N'-tetraacetic acid; 1-25GRK1, N-terminal fragment of GRK1 (amino acids M1-S25) fused with glutathione-S-transferase; HEPES, N-(2-hydroxyethyl)piperazine-N0-(2 ethanesulfonic acid); λmax, position of fluorescence spectrum maximum; NCS, neuronal calcium sensor; N-GRK1, N-terminal domain of GRK1 (amino acids Met1-Gly183) fused with glutathione-S-transferase; nNOS, neuronal nitric oxide synthase; OmRec, oxidized monomer of recombinant wild type recoverin; PD, photochemical damage; RmRec, reduced monomer of recombinant wild type recoverin; ROS, rod outer segments; RPE, retinal pigment epithelium; SPR, surface plasmon resonance; T1/2, mid-transition temperature; Tris, tris(hydroxymethyl) aminomethane.

Here, we investigated the oxidative state of recoverin in mammalian retina exposed to various doses of visual light illumination such as long-term moderate illumination (2,500 lx for 14 h, rats without anesthesia) or short-term moderate/intense illumination (2,200/30,000 lx for 3 h, rabbits under general anesthesia). We found that both types of illumination promoted disulfide dimerization of recoverin. Meanwhile, in the rabbit model the oxidized derivatives of the protein included monomer with intramolecular disulfide bond (due to presence of the second Cys in rabbit recoverin) and monomer with sulfinic group. In both models, the oxidation of recoverin preceded apoptosis of photoreceptors. Of interest, disulfide dimerization and oxidation of monomeric recoverin caused similar changes in the protein secondary structure, its overall stability and Ca2<sup>+</sup> affinity, but differently affected its functional properties. The oxidized monomer (or oxidation mimicking C39D mutant) of recoverin demonstrated impaired ability to bind photoreceptor membranes and regulate GRK1, whereas disulfide dimer of recoverin exhibited improved membrane binding and GRK1 inhibition in absence of Ca2+. We proposed that the latter effect would slow down rhodopsin desensitization in the light, which may favor retention of the light-induced oxidative stress and induction of photoreceptor apoptosis.

# MATERIALS AND METHODS

### Illumination of Mammalian Eyes Under in vivo Conditions

Rat eye illumination was performed exactly as described in Novikova et al. (2014). Forty 2-month-old Wistar rats were used. Prior to the experiment, the animals were housed under normal vivarium conditions (12 h light/12 h dark cycle, 22–25◦C, 55–60% humidity). Before illumination, rats were kept in the dark for 14 h for dark adaptation of their eyes. In experimental group, 16 unrestrained animals were exposed to visual light for 14 h using an NC-DE 70W/DW RX7s metal halide lamp (NARVA, Germany) with the following specifications: power 70 W, luminous flux 5,000–5,500 lumens, color temperature 4,000 K. The lamp was placed 2 m from the animal cages, yielding intensity of retinal illumination of 2,500 lx (0.003 W/cm<sup>2</sup> ) and total dose of 151 J/cm<sup>2</sup> . In control group, eight animals were kept dark-adapted for the same time interval. The rats were decapitated under intraperitoneal anesthesia immediately after the illumination or at the lapse of 7 days during which the animals were kept under normal conditions described above.

Illumination of the rabbit eyes was performed according to the previously described procedure (Zernii et al., 2015b) with modifications. Thirty six healthy pigmented rabbits (6 months old, weight of 2.3–3 kg) were kept for 2 weeks at a 12 h lightdark cycle at temperature of 22–25◦C and humidity of 55–60% with free access to food and water. Prior to experiment, all animals were dark-adapted for 12 h and then anesthetized by intramuscular injection of a commercial preparation containing 50 mg/ml tiletamine and 50 mg/ml zolazepam (15 mg of preparation per kg of body weight). One eye of each animal was exposed to visual light for 3 h using fiber-optic illuminator equipped with 150 W halogen lamp (Euromex Microscopen). The device settings were adjusted to ensure retina illuminance of 2,200 lx (0.011 W/cm<sup>2</sup> ; illumination scheme 1) or 30,000 lx (0.15 W/cm<sup>2</sup> ; illumination scheme 2) yielding total dose of 118 J/cm<sup>2</sup> or 1620 J/cm<sup>2</sup> , respectively. Another eye was kept in the dark to be used as a control. The pupils of the light-exposed eyes were dilated using 25 mg/ml solution of phenylephrine hydrochloride. The animals were euthanized with an overdose of the anesthetic either immediately after the experiment, or after 3 days of normal husbandry.

Animal handling was performed according to the guidelines of the Association for Research in Vision and Ophthalmology for use of animals in ophthalmic and vision research (ARVO). The protocol was approved by the Belozersky Institute of Physico-chemical Biology Animal Care and Use Committee (Protocol number 1/2016).

### Identification of Recoverin Forms in Retinal Extracts

Detection and identification of monomeric and multimeric oxidized forms of recoverin in retinal extracts were performed following the protocols developed in Zernii et al. (2015b) with modifications. The retinas of the enucleated rat and rabbit eyes were isolated in the dim red light at 4◦C and homogenized in 50 mM Tris-HCl buffer (pH 7.5), 100 mM NaCl on ice. The protein extracts were centrifuged at 39,000 × g (20 min, 4◦C) and the supernatants were incubated for 1 h with rabbit polyclonal anti-recoverin antibodies, immobilized on CNBr-activated Sepharose 4B (GE Healthcare) in 0.1 M potassium phosphate buffer (pH 8.2), 150 mM NaCl at 4◦C, with gentle shaking. The matrix was washed with 300 mM NaCl in the same buffer and recoverin forms were eluted with 200 mM sodium citrate buffer (pH 2.5), 150 mM NaCl, 5% glycerol. The resulting protein samples were adjusted to pH 6.2, concentrated and analyzed by non-reducing or reducing Western blotting using affinity-purified rabbit polyclonal antirecoverin antibodies (Senin et al., 2011) or mouse monoclonal anti-recoverin antibodies (Santa Cruz Biotechnology), for rat and rabbit retinal extracts, respectively. The antigen-antibody complexes were visualized in the ChemiDocTM MP System (Bio-Rad) using Enhanced chemiluminescence (ECL) kit (Bio-Rad).

Alternatively, 5 µl aliquots of the immunoaffinity purified recoverin were desalted using P2 ZipTip C18 pipette tips, the protein was eluted by 50% acetonitrile with 0.1% TFA, and 0.5 µl of the resulting solution were mixed on a steel target with 0.5 µl of 20 mg/ml 2,5-dihydroxybenzoic acid in 30% (v/v) acetonitrile with 0.5% (v/v) TFA, and analyzed by mass spectrometry. For peptide mass fingerprinting, recoverin was digested by incubation with 5 µl of 15 µg/ml sequencing grade modified trypsin (Promega) in 0.05 M NH4HCO<sup>3</sup> at 37◦C for 1 h. The reaction mixture was adjusted by 10 µl of 0.5% TFA and 1 µl of the resulting solution was mixed on a steel target with 0.5 µl of 20 mg/ml 2,5-dihydroxybenzoic acid in 20% (v/v) acetonitrile with 0.5% (v/v) TFA, and subjected to mass spectrometry analysis. Mass spectra were recorded using an ultrafleXtreme MALDI-TOF/TOF mass spectrometer (Bruker Daltonics) equipped with a Smartbeam-II laser (Nd:YAG, 355 nm). The [MH]<sup>+</sup> molecular ions of full-length proteins were analyzed in linear mode; the values of average m/z ratios were accurate to 10 Da. Peptide mass fingerprints were obtained in reflector mode. Monoisotopic [MH]<sup>+</sup> molecular ions were measured in the 600–5,000 m/z range with a peptide tolerance of 30 ppm. MS/MS spectra of selected peptides were obtained using Lift mode with accuracy of 1 Da for daughter ion measurements. Mass spectra were analyzed using flexAnalysis 3.3 software (Bruker Daltonics). Protein analysis was carried out by MS and MS/MS ions searches using Mascot software (Matrix Science) and the NCBI protein database.

#### Histologic Examination

fnmol-11-00474 December 24, 2018 Time: 17:25 # 4

For histological analysis, the enucleated rat and rabbit eyes were fixed in a Bouin's (rats) or Carnoy's (rabbits) solutions. After the fixation, the posterior segments containing retina, choroid, and sclera were isolated. The preparations were washed in ethanol and subjected to routine histological processing including dehydration by sevenfold incubation in absolute isopropanol for 5 h, and embedding of properly oriented samples into Histomix paraffin medium. Sagittal sections (4–7-µm thick each) were obtained from each paraffin block. The sections were mounted on slides, deparaffinized, rehydrated, stained with Carazzi's hematoxylin and 0.5% eosin Y and examined using LEICA DM 4000B (Leica) and AxioScope A.1 (Carl Zeiss) microscopes. Microphotographs were obtained using AxioCam MRc5 high-resolution digital camera (Carl Zeiss) and Leica DFC400 digital camera (Leica). Processing of the microphotographs was performed using the AxioVision 8.0 (Carl Zeiss) and Adobe Photoshop CS6 Extended (Adobe Systems, United States) software.

#### Preparation of Recoverin Forms

Recombinant myristoylated bovine recoverin (reduced monomer, RmRec) was prepared according to previously published procedure (Zernii et al., 2011). Recoverin mutant C39D was obtained as described in previous study (Permyakov et al., 2012). The degree of myristoylation of RmRec and C39D was determined by analytical HPLC using C18 "Symmetry" 3.9 × 150 mm reversed-phase column (Waters) and was more than 95%. Disulfide dimer (dRec) and oxidized monomer (OmRec) of recoverin were prepared from RmRec (Zernii et al., 2015b). Briefly, RmRec was incubated with 0.005% H2O<sup>2</sup> at 30◦C for 5 h and the resulting sample was subjected to gel-filtration using SuperdexTM 75 HR 10/30 HPLC column (GE Healthcare Life Sciences). The fractions containing dRec or OmRec were collected and dialyzed against ultrapure water at 4 ◦C, freeze-dried and stored at −70◦C.

For identification of oxidized derivatives of recoverin forming under mild oxidizing in vitro conditions, the recoverin sample was prepared mainly as described in previous study (Permyakov et al., 2007). Briefly, recombinant bovine recoverin was dialyzed overnight under non-reducing conditions against 50 mM Tris-HCl buffer (pH 7.5), 100 mM NaCl, followed by MS analysis.

Concentration of the recoverin forms was measured spectrophotometrically using a molar extinction coefficient at 280 nm of 24,075 M−<sup>1</sup> cm−<sup>1</sup> , calculated according to Pace et al. (1995). For comparative studies, the concentration of dRec was calculated per mole of recoverin monomer.

### Calcium Binding Assay

Ca2<sup>+</sup> binding to RmRec/dRec (40 µM/20 µM) was examined using <sup>45</sup>Ca2+-binding assay (Weiergraber et al., 2006; Zernii et al., 2013). Briefly, recoverin was incubated with increasing <sup>45</sup>Ca2<sup>+</sup> concentrations, followed by the protein separation using ultrafiltration. Ca2<sup>+</sup> concentrations were determined by radioactivity counting. Stoichiometry of Ca2<sup>+</sup> binding was determined from the excess Ca2<sup>+</sup> in the protein sample over that present in the ultrafiltrate. The resulting values were plotted versus [Ca2+]free and the plots were fitted to the 4-parameter Hill equation using SigmaPlot 11 software (Systat Software). The apparent equilibrium dissociation constant (KD) and Hill coefficient values characterizing Ca2<sup>+</sup> affinity of recoverin were obtained from these fits.

#### Fluorescence Measurements

Fluorescence emission spectra were measured with a Cary Eclipse spectrofluorimeter (Varian Inc.), equipped with a Peltier-controlled cell holder essentially as described before (Baksheeva et al., 2015). Protein concentration was 6–14 µM. Measurements were carried out in 10 mM HEPES-KOH, 150 mM KCl, pH 7.3 buffer in the presence of either 1 mM CaCl<sup>2</sup> or 1 mM EGTA and 1 mM DTT for RmRec and C39D recoverin. Concentrations of bis-ANS and protein in these experiments were 1 µM and 6 µM, respectively. The concentration of water stock solution of bis-ANS was evaluated using molar extinction coefficient ε385nm of 16,790 M−<sup>1</sup> cm−<sup>1</sup> (Farris et al., 1978). Fluorescence of bis-ANS/recoverin was excited at 385/280 nm; emission slit width was 5 nm. All spectra were fitted to log-normal curves (Burstein and Emelyanenko, 1996) using LogNormal software (IBI RAS, Pushchino, Russia), implementing non-linear regression algorithm by Marquardt (Marquardt, 1963). The fluorescence spectrum maximum positions (λmax) were obtained from these fits. Spectrofluorimetric temperature scans were performed stepwise, allowing the sample to equilibrate at each temperature. Temperature was monitored inside the sample cell. The average heating rate was 0.5◦C/min. The T1/<sup>2</sup> were calculated from temperature dependence of λmax using Boltzmann sigmoid as implemented in OriginPro 8.0 (OriginLab Corporation) software for Ca2+-loaded recoverin, and as described in Permyakov and Burstein (1984) for apo-forms.

#### Circular Dichroism Measurements

Circular dichroism measurements were carried out with a JASCO J-810 spectropolarimeter (JASCO Inc., Japan), equipped with a Peltier-controlled cell holder as previously described (Zernii et al., 2015a). Protein concentration was 4 µM for RmRec and OmRec, or 2 µM for dRec. The contribution of buffer [pH 8.2, 10 mM H3BO3–KOH, 1 mM CaCl<sup>2</sup> or 1 mM EDTA, 20 µM DTT (for RmRec)] was subtracted from the experimental spectra. Estimations of the secondary structure fractions were made using the CDPro software package (Sreerama et al., 2000).

# The Equilibrium Centrifugation Assay of Membrane Binding

Bovine ROS and urea-washed photoreceptor membranes were prepared from frozen retinas according to a method described in Grigoriev et al. (2012). The binding of recoverin forms to photoreceptor membranes was performed using equilibrium centrifugation assay (Permyakov et al., 2003), with modification described in Zernii et al. (2003). Samples of RmRec (30 µM) or dRec (15 µM) were mixed with bleached urea-washed photoreceptor membranes containing 0–100 µM rhodopsin in 20 mM Tris–HCl (pH 8.0), 150 mM NaCl, 20 mM MgCl2, 1 mM DTT (for RmRec), with addition of Ca2+/5,50Br2-BAPTA buffer yielding 0.11–500 µM free [Ca2+] in total volume of 50 µl. The probes were incubated at 37◦C for 20 min and centrifuged (15 min, 14,000 rpm). The supernatants were discarded and the pellets were dissolved in 50 µl of sample buffer [125 mM Tris-HCl, pH 6.8, 4% (w/v) SDS, 20 (v/v) glycerol, 10% (v/v) b-mercaptoethanol, 0.004% (w/v) bromophenol blue] and analyzed by SDS-PAGE. The amounts of recoverin forms bound to photoreceptor membranes were evaluated by densitometric scanning of the corresponding bands in polyacrylamide gel.

#### Rhodopsin Phosphorylation Assay

GRK1 (rhodopsin kinase) was purified from ROS as described elsewhere (Senin et al., 2011). GRK1 assay was performed as described in Vladimirov et al. (2018). Briefly, 40 µM RmRec or C39D mutant, or 20 µM dRec were mixed with 10 µM rhodopsin (urea-washed photoreceptor membranes) and 0.3–0.5 units of GRK1 in 20 mM Tris-HCl (pH 8.0), 100 mM NaCl, 1 mM [γ-<sup>32</sup>P]ATP, 1 mM DTT, 3 mM MgCl2, with addition of Ca2+/5,50Br2-BAPTA buffer yielding either 200 µM or 0.01 µM free [Ca2+]. Free Ca2+-concentration in the buffers was calculated using Webmaxc Standard software (Stanford University). The reaction (15 min) was initiated by light illumination and terminated by addition of sample buffer for SDS-PAGE. The proteins were separated by polyacrylamide gel electrophoresis and <sup>32</sup>P emission was registered by phosphorimaging radioautography.

#### Surface Plasmon Resonance Studies

N-terminal domain of rhodopsin kinase (N-GRK1) and its fragment M1-S25 (1-25GRK1) fused with glutathione-S-transferase were obtained according to previously published procedure (Grigoriev et al., 2012). SPR measurements were performed at 25◦C using Bio-Rad ProteOnTM XPR36 protein interaction array system (Kazakov et al., 2017). Ligand (40 µg/mL N-GRK1 or 1-25GRK1 in 10 mM sodium acetate, pH 4.5 buffer) was immobilized on ProteOn GLH sensor chip surface (up to 10,000–15,000 resonance units, RUs) by amine coupling, according to the manufacturer's instructions. The remaining activated amine groups on the chip surface were blocked by 1 M ethanolamine solution. Analyte (RmRec, OmRec, C39D and dRec, 1 µM to 40 µM) in a running buffer (10 mM HEPES, 150 mM NaCl, 0.05% TWEEN 20, pH 7.4 buffer with 1 mM CaCl<sup>2</sup> and 2 mM MgCl2) was passed over the chip at a rate of 30 µl/min for 350 s, followed by flushing the chip with the running buffer for 1,500 s. The double-referenced SPR sensograms were globally fitted according to a heterogeneous ligand model, which assumes existence of two populations of the ligand (L1 and L2) that bind a single analyte molecule (A):

$$\begin{array}{ccccc} \text{Kd1} & & & \text{Kd2} \\\\ L1 & + & A \Leftrightarrow L1A; L2 & + & A \Leftrightarrow L2A \\\\ & & & & \text{k}d2 \\\\ & & & & \text{k}d2 \\ \end{array} \tag{1}$$

where Kd and kd refer to equilibrium and kinetic dissociation constants, respectively. Kd, kd and Rmax (maximum response) values were evaluated using Bio-Rad ProteOn ManagerTM v.3.1 software. The sensor chip surface was regenerated by passage of 0.5% SDS water solution for 50 s.

# RESULTS

#### Illumination of Mammalian Retina Promotes Accumulation of Disulfide Dimer of Recoverin

To explore redox state of recoverin (both for the protein containing one cysteine and the one with two cysteines) in mammalian retina exposed to different doses of illumination, two animal models were used. In the first model, unrestrained rats were exposed to long-term (14 h) illumination by visual light of moderate intensity (metal halide lamp, 2,500 lx, 0.003 W/cm<sup>2</sup> ). In the referent group, rats were kept dark-adapted for the same time intervals. The second model employed pigmented rabbits under general anesthesia that were exposed to short-term (3 h) illumination by visual light of either moderate (halogen lamp, 2,200 lx, 0.011 W/cm<sup>2</sup> ; scheme 1) or high intensity (halogen lamp, 30,000 lx; 0.15 W/cm<sup>2</sup> ; scheme 2). In this case, the illumination doses were chosen to model iatrogenic retinal damage induced by common operative microscopes used in surgical ophthalmology (see the section "Discussion"). One eye of each animal was illuminated, whereas another eye was kept in the dark for use as a reference. Immediately after the illumination/dark adaptation the animals were euthanized, the retinas were isolated, and soluble proteins were extracted from the isolated retinas. To improve identification of recoverin forms and their mass-spectrometry analysis (see below), the protein was quickly purified from the retinal extracts using immunoaffinity chromatography (Zernii et al., 2015b).

Non-reducing Western blotting revealed recoverin dimer in the protein fractions extracted from all illuminated eyes of the both species, as indicated by appearance of the band of ca 45 kDa in addition to the band of recoverin monomer (ca 23 kDa, **Figure 1**). The dimeric recoverin was also found in control samples, but the dimer to monomer ratio was lower. In rabbit model illumination dose-dependent accumulation of the dimeric recoverin was observed (**Figure 1B**). Thus, weight fraction of the dimer was ∼20% in control animals and it increased up to ∼30% upon illumination following scheme 1 and up to ∼40% upon illumination following scheme 2 (**Figure 1C**). In the latter

case, an additional band corresponding to multimeric/aggregated forms of the protein (>170 kDa) was identified, in accord with our previous findings (Zernii et al., 2015b) (**Figure 1B**). Meanwhile, this band was absent in the samples derived from the eyes illuminated according to the scheme 1, as well as in the control samples. Furthermore, such band was absent in the rat recoverin samples (**Figure 1A**). Since Western blotting of the samples from the both species under reducing conditions revealed disappearance of the dimer and multimeric bands, they were stabilized by disulfide bonds. It should be noted, that some bands of monomeric/dimeric recoverin exhibited bifurcation, which can be attributed to partial proteolysis of the protein or presence of its non-myristoylated form.

#### Illumination of Rabbit Retina Induces Formation of Monomeric Thiol Oxidized Forms of Recoverin

The oxidative modifications of monomeric recoverin, extracted from the illuminated retinas, were explored by MALDI-TOF mass-spectrometry. The analysis was first performed using linear mode of the spectrometer for detection of [MH]<sup>+</sup> molecular ions of full-size protein, enabling mass determination with accuracy of 10 Da. In MS spectra of recoverin samples from the illuminated rat eyes only one peak with m/z ratio of 23,483 Da was observed within the 23,000–24,000 m/z range (data not shown). Despite a slight difference from expected mass of the intact recoverin (23,472 Da), this peak corresponded to rat recoverin, as confirmed by peptide mass fingerprinting (**Supplementary Table S1**). Two C39-containing peptides (molecular masses of 2507.1666 Da and 3817.8752 Da) were identified in trypsin hydrolysate of the protein. Since C39 residue of the both peptides remains reduced, the illumination of rat eyes by visual light of moderate intensity is accompanied by accumulation of disulfide dimer of recoverin, whereas the conservative thiol of the monomeric protein remains intact.

Mass-spectrometry of recoverin derived from the referent (dark-adapted) rabbit eyes revealed a single peak with m/z ratio of 23,470–23,475 Da that corresponded to the fully reduced protein (**Figure 2**). Meanwhile, MS analysis of recoverin samples

obtained upon the high-intensity illumination of rabbit eyes (scheme 2) additionally revealed a minor peak centered at m/z ratio of 23,504 Da (**Figure 2B**). This peak can be attributed to rabbit recoverin with two extra oxygen atoms. One of the possibilities is Cys39 oxidation with formation of sulfinic acid. Although we failed to detect this derivative by direct MS/MS evidences, we confirmed its formation in recombinant bovine recoverin under mild oxidizing in vitro conditions using mass fingerprinting of its trypsin hydrolysate peptides. Two cysteine-containing peptides of this protein, E38-R43 and F23-R43, were found in reduced (648.3026 and 2523.1568 Da), singly oxidized (664.2945 and 2539.1372) and doubly oxidized (680.2717 and 2555.1170) states, thereby indicating conversion of Cys39 into sulfenic/sulfinic acid. The structure of these peptides was confirmed by MS/MS analysis (**Supplementary Figure S1**).

Interestingly, the peak at 23,504 Da was absent in the MS spectra of recoverin extracted from the animals exposed to lower dose of illumination within the scheme 1 (**Figure 2A**). Instead, mass fingerprinting of trypsin hydrolysate peptides revealed (MS and MS/MS analysis) recoverin form with an intramolecular disulfide bridge (opposite to rat recoverin, its rabbit ortholog contains two cysteine residues, C29 and C39). Among 20 identified peptides that cover 87% of the protein sequence (**Supplementary Table S2**) two peptides (molecular masses of 648.2951 Da and 1923.8839 Da) contained thiols of C29 and C39 residues in a reduced form. Meanwhile, the MS peak with m/z ratio of 2551.1380 corresponded to the peptide F23-R43, in which C29 and C39 residues lacked hydrogen atoms, i.e., formed an intramolecular disulfide bridge. Notably, neither of other redox-sensitive amino acids (M, W or H) were found in oxidized state, despite their presence in the identified peptides. Hence, C39 has the highest redox sensitivity among all recoverin residues, in line with our previous findings (Permyakov et al., 2007).

Overall, the accumulation of disulfide dimer of recoverin in the light-exposed rabbit eyes was accompanied by additional formation of oxidized monomer of the protein, which contained either an intramolecular disulfide bond or sulfinic acid, depending on illumination dose.

### Thiol Oxidation of Recoverin in Illuminated Mammalian Retina Precedes Apoptosis of Photoreceptors

To examine if the light-induced thiol oxidation of recoverin was accompanied by PD of mammalian retina, the illuminated/dark-adapted in vivo rat and rabbit eyes were analyzed by histological techniques. Posterior segments (containing RPE and neural retina) of the eyes were obtained upon decapitation of the animals either immediately after exposure to light or 3–7 days later. The samples were embedded in paraffin and stained using hematoxylin and eosin (**Figures 3**, **4**).

layer; INL, inner nuclear layer; OPL, outer plexiform layer; ONL, outer nuclear retinal layer; Ph, photoreceptor layer. (A) The retina of a control dark-adapted animal. (B) The retina isolated immediately after illumination for 14 h with metal halide lamp (2,500 lx). ONL migration in the vitreal direction is indicated by black arrow. (C,D) The retina isolated 7 days after the light exposure. ONL disorganization and thinning are indicated with white arrows. INL migration and thinning of OPL and INL are marked on (C,D) with gray and black arrows, respectively.

Posterior segments of the rat eyes enucleated immediately after the prolonged light exposure exhibited multifocal changes in the retina affecting mainly photoreceptor layer without significant alterations in RPE (**Figure 3**). Although the overall structure of the retina remained intact, in some regions the signs of advanced stage of apoptosis such as karyopyknosis and formation of apoptotic bodies were observed. In addition, the retina exhibited thinning of the outer plexiform layer formed by photoreceptor axons, and shifting of the photoreceptor cell bodies in the vitreal direction (**Figure 3B**). Consistently, in 7 days after the illumination, the outer nuclear layer and inner nuclear layer of the rat retinas drastically decreased in thickness. Destruction of these layers together with deterioration of the outer plexiform layer resulted in severe changes in retina cytoarchitecture representing symptoms of irreversible retinal damage through apoptosis (**Figures 3C,D**).

In the rabbit samples prepared immediately after the light illumination following scheme 1 (data not shown) and scheme 2 (**Figure 4B**), the tissues of the eye fundus exhibited no changes as compared to referent eyes (**Figure 4A**). In the first case, the normal morphology of the retina was maintained even on the third and the seventh day of post-exposure (data not shown). Meanwhile, in the second case, the retina exhibited destructive changes on the third day of post-exposure and these changes involved both photoreceptor layer and RPE (**Figures 4C,D**). In the most affected regions of this tissue both outer and inner segments of photoreceptors were absent and multifocal death of photoreceptor cells was evident (**Figure 4C**, inset). Outer nuclear layer was sparse and photoreceptor cells showed morphological signs of apoptotic type cell death such as karyopyknosis (**Figure 4C**, inset, red arrows). These changes in outer nuclear layer were accompanied by cell death in inner nuclear layer of the retina resulting in destruction and thinning of these layers as well as outer plexiform layer lying between them. Furthermore, the damage involved RPE cells that were vacuolized, activated and increased in size

(demonstrated «tombstone» morphology) (**Figure 4C**, inset, yellow arrows) and some of them migrated to the dying retinal cells (**Figure 4C**). Finally, the retina exhibited detachments from RPE, which were noticeable not only in the most photosensitive regions (**Figure 4D**), but also adjacent to the damaged areas (**Figures 4C,D**, asterisks), where the photoreceptors were not directly affected. In the newly formed wide space between RPE and outer nuclear layer, the residuals of photoreceptor segments became engulfed by macrophages-like cells (**Figure 4C**, inset, white arrows). These cells had pigment granules suggesting that they migrated from RPE. The presence of the macrophage-like cells in the space between RPE and the retina (**Figure 4D**, white arrows) reflected development of compensatory processes in the retina.

Thus, despite some differences in time scales, both rat and rabbit models of the light-induced retinal damage were characterized by death of photoreceptor cells via apoptosis, which developed in the post-exposure period. Since recoverin was isolated from the retinas immediately after the illumination, one can conclude that the revealed formation of disulfide dimer and oxidized monomeric forms of the protein precedes the apoptosis of photoreceptors.

#### Disulfide Dimerization Increases Sensitivity of Recoverin to Calcium

To examine functional consequences of the light-induced thiol oxidation of recoverin, we studied in vitro Ca2+ dependent structural and functional properties of the following derivatives of the protein: reduced monomer (RmRec), disulfide dimer (dRec) and C39D mutant imitating conversion of C39 thiol into the negatively charged sulfinic acid. In some experiments, oxidized monomer of recoverin (OmRec) prepared by incubation of RmRec with 0.005% H2O<sup>2</sup> was employed for comparison with C39D mutant that had been partially characterized in previous studies (Permyakov et al., 2012). The monomer of rabbit recoverin with intramolecular disulfide bond was left beyond the scope of our studies, as C29 residue of the protein is not conservative.

Since recoverin is known to serve as both Ca2<sup>+</sup> sensor and Ca2<sup>+</sup> buffer in the cell (Makino et al., 2004), we explored the oxidation-induced changes in its Ca2+-binding parameters. According to <sup>45</sup>Ca2+-binding assay, RmRec coordinated two Ca2<sup>+</sup> ions in a cooperative manner with Hill coefficient of 1.5 and half-saturation at 24 µM (**Figure 5** and **Table 1**), in accord with previous estimates (Zernii et al., 2011). Meanwhile, dRec exhibited markedly less cooperative Ca2<sup>+</sup> binding (Hill coefficient of 1.1), although its stoichiometry remained unchanged. Furthermore, disulfide dimerization increased overall Ca2<sup>+</sup> affinity of recoverin up to K<sup>D</sup> value of 11 µM (**Figure 5** and **Table 1**). Since the C39D mutant also demonstrated slightly increased affinity to Ca2<sup>+</sup> (Permyakov et al., 2012), we concluded that both disulfide dimerization and conversion of Cys39 into sulfinic acid enhanced calcium sensitivity of recoverin, thereby affecting its Ca2+-dependent functions.

FIGURE 5 | Calcium binding to RmRec/dRec. 40 µM RmRec (open circles) or 20 µM dRec (open squares) were incubated with increasing Ca2<sup>+</sup> concentrations in the presence of <sup>45</sup>Ca2<sup>+</sup> and the Ca2+-bound protein was separated by ultrafiltration. Stoichiometry of Ca2<sup>+</sup> binding determined from the excess Ca2<sup>+</sup> in the protein sample over that present in the ultrafiltrate was plotted versus [Ca2+]free and the plots were fitted to the 4-parameter Hill equation. Solid and dashed curves represent the best fits for RmRec and dRec, respectively.

#### Thiol Oxidation Reduces Structural Stability of Recoverin

To reveal an impact of thiol oxidation on recoverin structure, we examined physicochemical properties of RmRec, dRec, and OmRec and analyzed them with the respective data obtained earlier for C39D mutant (Permyakov et al., 2012). In absence of calcium dRec and OmRec exhibited decrease in negative molar ellipticity at 205–240 nm, as compared to RmRec (**Figure 6A**). Analysis of the CD spectra revealed decreased α-helicity of apo-dRec/OmRec, compensated by increase in the content of other elements of their secondary structure (**Table 2**). Calcium binding does not induce marked changes in secondary structure of RmRec, but shifts secondary structure fractions of OmRec and dRec toward those for RmRec (**Figure 6B**). Overall, α-helical content of recoverin decreased regardless of Ca2<sup>+</sup> level in the order RmRec > C39D > OmRec > dRec. Thus, thiol oxidation disorganized secondary structure of recoverin, especially in the case of disulfide dimer.

Thermal unfolding of recoverin was monitored by changes in maximum position of tryptophan fluorescence spectrum (λmax), which reflects solvent accessibility of its emitting Trp residues: protein unfolding induces red shift of its fluorescence spectrum maximum (**Figure 6C**). Since at temperatures below the thermal transition the λmax values of Ca2+-bound forms of dRec and OmRec were lowered with respect to RmRec, the oxidized recoverin derivatives exhibited lower solvent exposure of emitting Trp residues. Calcium removal further protected them from the solvent, as observed for RmRec. In the absence of calcium the recoverin derivatives exhibited lowered thermal stability. Their T1/<sup>2</sup> values decreased in the order observed for their α-helices fractions:


<sup>∗</sup>According to <sup>45</sup>Ca2+-binding assay data fitted to the Hill model. ∗∗Estimated from spectrofluometric measurements within the model of sequential filling of Ca2+-binding sites as described in Permyakov et al. (2012). ∗∗∗The data from Permyakov et al. (2012). ∗∗∗∗According to equilibrium centrifugation assay data fitted to the Hill model.

symbols) and Ca2+-loaded (filled symbols) RmRec (14 µM, squares), OmRec (14 µM, circles) and dRec (7 µM, triangles) samples at pH 7.3. (D) The binding of bis-ANS (1 µM) to Ca2+-free (dashed curves) or Ca2+-loaded (solid curves) RmRec (6 µM, medium curves), C39D mutant (6 µM, thin curves) and dRec (3 µM, thick curves) monitored by fluorescence emission spectrum of the dye at 20◦C, pH 7.3.

RmRec > C39D > OmRec > dRec (**Table 1**). Calcium binding markedly enhanced thermal stability of all recoverin variants studied.

Finally, the oxidation-induced alterations in recoverin structure were examined by spectrofluorimetric monitoring of its interaction with bis-ANS, a hydrophobic probe recognizing surface-accessible non-polar protein cavities (**Figure 6D**) (Hawe et al., 2008). Fluorescence intensity of bis-ANS in the presence of RmRec was threefold to fivefold higher for the Ca2+-loaded protein, apparently reflecting Ca2+-induced exposure of the hydrophobic pocket of recoverin responsible for its interaction with GRK1 (Ames et al., 2006; Zernii et al., 2015a). In the presence of C39D mutant, bis-ANS exhibited fluorescence spectra similar to those observed for RmRec, which indicate equivalent surface hydrophobicities of these forms. Meanwhile, in the presence of disulfide dimer of recoverin bis-ANS showed notably increased fluorescence intensity as compared to RmRec. Thus, Ca2+-bound dRec exhibited highest surface hydrophobicity among the recoverin derivatives. Furthermore, surface hydrophobicity of Ca2+-free dRec resembled that of Ca2+-bound RmRec, which suggest that at low intracellular calcium levels dRec might expose the non-polar residues interacting with GRK1, and thereby could be aberrantly active.



<sup>∗</sup>The data from Permyakov et al. (2012).

#### Disulfide Dimer of Recoverin Interacts With Photoreceptor Membranes and Inhibits Rhodopsin Kinase at Low Calcium

Given the observed changes in Ca2<sup>+</sup> affinity and surface hydrophobicity of recoverin induced by its disulfide dimerization, we suggested that membrane-binding properties of the thiol-oxidized protein could be affected. To probe this suggestion, we monitored association of dRec with photoreceptor membranes using equilibrium centrifugation assay. At saturating Ca2<sup>+</sup> levels, the photoreceptor membrane-bound protein fraction for dRec was lowered with regard to RmRec by 50% (**Figure 7A**). This effect was accompanied by the respective shift in Ca2<sup>+</sup> dependence of the membrane binding toward higher calcium concentrations (**Figure 7A**): the membrane binding of dRec is half-maximal at 5.1 µM free Ca2<sup>+</sup> versus 3.4 µM free Ca2<sup>+</sup> for RmRec (**Table 1**). In this respect, dRec resembled the C39D mutant that exhibited the half-maximal effect at 11 µM free Ca2<sup>+</sup> (**Table 1**) (Permyakov et al., 2012). However, in the absence of Ca2<sup>+</sup> dRec and C39D mutant demonstrated different behavior. Whereas the mutant exhibited low affinity to membranes similarly to RmRec (Permyakov et al., 2012), membrane binding of dRec increased about twofold as compared to RmRec (**Figure 7A**, inset). Taken together with the results of bis-ANS fluorescence studies (see **Figure 6D**), these data suggested that disulfide dimerization of recoverin abolishes Ca2+-myristoyl switch leading to constitutive exposure of its hydrophobic surfaces that anchor the protein on membrane at low calcium. It should be added that treatment of dRec with excess of DTT yielded monomeric recoverin, which bound to the membranes in Ca2+-dependent manner similarly to RmRec (∼26% and 5% of the control at high and low calcium, respectively).

Since Ca2+-myristoyl switch of recoverin involves exposure of its hydrophobic residues responsible for interaction with GRK1, we supposed that disulfide dimerization may affect regulatory activity of recoverin toward the enzyme. Therefore, we compared the effect of various recoverin forms on GRK1 activity

30 µM RmRec (open circles) and 15 µM dRec (open squares) to photoreceptor membranes. Recoverin was mixed with urea-washed photoreceptor membranes in the presence of 0.11–500 µM [Ca2+]free at 37◦C, pH 8.0 and the membranes were separated by ultracentrifugation. The fractions of membrane-bound protein evaluated by SDS-PAGE were plotted versus [Ca2+]free and the plots were fitted to the 4-parameter Hill equation. Solid and dashed curves represent best fits for RmRec and dRec, respectively. The inset: fractions of the Ca2+-free and Ca2+-saturated protein bound to the membranes. (B) Inhibition of GRK1 by RmRec or C39D mutant (40 µM), or dRec (20 µM). Rhodopsin phosphorylation by GRK1 in the presence of [γ − <sup>32</sup>P]ATP was monitored at high [Ca2+]free (200 µM Ca2+, filled bars) or low [Ca2+]free (0.01 µM, open bars) by phosphorimaging radioautography.

in rhodopsin phosphorylation assay employing urea-washed photoreceptor membranes harboring the receptor. At high calcium concentrations, C39D mutant exhibited about twofold decline in inhibitory efficiency compared to RmRec (**Figure 7B**). Under these conditions, dRec inhibited GRK1 by almost 70%, which is close to the effect of RmRec. Meanwhile, the differences between the behavior of dRec and RmRec were absent only at their saturating levels (40 µM recoverin monomer), whereas at fourfold lower concentration disulfide dimer inhibited GRK1 one third less efficiently than reduced monomer (data not shown). In the absence of Ca2+, neither RmRec nor C39D exhibited significant influence on GRK1 activity. Meanwhile, dRec inhibited the enzyme by 40%, thereby demonstrating constitutive Ca2+-independent activity toward the enzyme (**Figure 7B**). Notably, the incubation of dRec with excess of

DTT almost completely restored normal regulatory activity of recoverin toward GRK1 (data not shown) indicating that disulfide dimerization of recoverin is functionally reversible.

To examine whether the altered GRK1 regulation observed in the case of dRec and C39D mutant resulted from changes in recoverin affinity to GRK1, we employed SPR spectroscopy to monitor recoverin interaction with N-terminal domain of the enzyme (N-GRK1) or its fragment M1-S25 (1-25GRK1), which both contain specific recoverin-binding site (Ames et al., 2006). To this end, N-GRK1 or 1-25GRK1 were immobilized on the surface of SPR sensor chip by amine coupling and a set of injections of RmRec, dRec, OmRec and C39D mutant was carried out. The SPR sensograms were approximated by the heterogeneous ligand model (1) yielding respective equilibrium constants (**Table 3**). In the presence of calcium, the affinity of N-GRK1 to the three recoverin variants was 6- to 10 fold lower than to RmRec. This effect was less noticeable for their binding to 1-25GRK1. The binding constants for dRec interaction with 1-25RK and N-GRK1 are close to each other, whereas the affinity of OmRec and C39D to 1- 25GRK1 is about 1.7- to 2-fold higher than to N-GRK1. Meanwhile, in the absence of calcium, none of the recoverin forms shows noticeable binding to the GRK1 fragments (data not shown). Thus, the revealed ability of Ca2+-free dRec to inhibit GRK1 may be due to its interaction with a site located outside the N-terminal domain of the enzyme or can be related to increased membrane affinity of the apodimer.

#### DISCUSSION

Recoverin is one of a few photoreceptor proteins demonstrating thiol oxidation in response to moderate doses of light illumination. Since intravitreous administration of anti-recoverin antibodies induces in rats photoreceptor dysfunction similar to that produced by excessive light illumination (Maeda et al., 2001; Ohguro et al., 2001), recoverin is likely involved in PD mechanisms. Therefore, we aimed at identification of the oxidized forms of recoverin accumulating in mammalian retina exposed to different doses of light illumination and exploration of structural and functional properties of these forms. The first model of retinal PD employed pigmented rabbits, since they share with humans many features of ocular anatomy, including eyeball size, its internal structure, and optical system along with biochemical features (Zernii et al., 2016). We developed rabbit model of iatrogenic light-induced retinal damage: the restrained animals were exposed to short-term (3 h) illuminations by visual light of either moderate (halogen lamp, 2,200 lx, 0.011 W/cm<sup>2</sup> ) or high intensity (halogen lamp, 30,000 lx, 0.15 W/cm<sup>2</sup> ). The latter irradiation dose corresponded to that received by the retina from common operative microscopes during 0.5–1 h in the course of, for instance, keratoplasty or vitrectomy [(Irvine et al., 1984), our own calculations]. In the second model of retinal PD, the long-term low-intensity illumination was applied to unrestrained rats: the rats were exposed to 14 h illuminations by visual light of moderate intensity (metal halide lamp, 2,500 lx, 0.003 W/cm<sup>2</sup> ), similarly to conditions described in a number of previous studies (van Norren and Gorgels, 2011; Novikova et al., 2014).

Analysis of recoverin fractions extracted from the light-exposed eyes of the both species revealed disulfide dimerization of recoverin. Fraction of the dimer found in the dark-adapted retinas was substantially lower. In the rabbit model, illumination dose-dependent accumulation of the dimeric recoverin was observed, in accord with our previous findings (Zernii et al., 2015b). The disulfide dimerization of recoverin in rat retinas was discovered for the first time. Unexpectedly, the retina of control rats contained relatively high basic level of recoverin dimer (∼40%) and we revealed only moderate increase of the dimer content upon illumination. The enhanced oxidation of recoverin in the control group can be related to higher basic redox potential in the retina of albino rats as compared to pigmented rabbits (Sohal and Weindruch, 1996). It is possible that the content of recoverin dimer transiently increased even above this basic level, but decayed by the completion of the long-term illumination, characteristic to the rat model. Previous proteomic study of rats identified arrestin as the only retinal protein exhibiting the light-dependent disulfide dimerization (Lieven et al., 2012). The failure to detect dRec in that work may be due to about two orders of magnitude lower amount of recoverin in retina compared to arrestin (Klenchin et al., 1995; Vishnivetskiy et al., 2007) along with lower sensitivity of 2D electrophoresis relative to Western blotting with ECL detection used here. In contrast to rats, the light-induced disulfide dimerization of recoverin in the rabbit retina was accompanied by accumulation of its monomer with an intramolecular disulfide bond and disulfide multimers/aggregates, depending on intensity of illumination. Indeed, in contrast to rat recoverin with single Cys residue, the rabbit protein contains two Cys residues (C29 and C39) enabling formation of an intramolecular disulfide bridge. The second Cys residue seems to be involved into accumulation of the disulfide-linked recoverin multimers/aggregates.

TABLE 3 | The affinity of Ca2+-bound recoverin variants to N-terminal domain of GRK1 (N-GRK1) and its fragment corresponding to the residues M1-S25 (1-25GRK1), estimated using SPR spectroscopy according to the heterogeneous ligand model (1).


Thus, formation of intramolecular disulfides and disulfide multimerization of recoverin might be inherent to its orthologs containing the second Cys residue (rabbit, human, etc.). Notably, MALDI mass-spectrometry analysis of rabbit recoverin samples obtained under the high-intensity light illumination (0.15 W/cm<sup>2</sup> ) revealed another monomer of the protein with two extra oxygen atoms. Despite lack of strict MS evidences, we attributed this recoverin form to its derivative with C39 converted into sulfinic acid. Firstly, peptide mass fingerprinting of the in vivo oxidized rabbit recoverin demonstrated that neither of other redox-sensitive residues (M, W, H) were oxidized. Thus, C39 is the most redox sensitive residue of recoverin, due to its low pK<sup>a</sup> value favoring formation at physiological pH of the highly reactive thiolate anion (Permyakov et al., 2007). Secondly, MS and MS/MS analysis of in vitro oxidized recombinant bovine recoverin demonstrated stepwise oxidation of C39 with formation of sulfenic and sulfinic acids. Finally, the conversion of C39 of recoverin into sulfenic acid under mild oxidizing conditions was previously confirmed by X-ray crystallography (Ranaghan et al., 2013). Overall, we propose that the oxidation state of recoverin depends on regime of the visual light illumination: lowintensity exposure yields accumulation of recoverin forms with an intermolecular or intramolecular disulfide bond, whereas high-intensity illumination apparently results in further oxidation of the thiol with accumulation of sulfinic acid.

The histological studies of both rat and rabbit models indicate that thiol oxidation of recoverin preceded or accompanied photoreceptor cells death via apoptosis, but kinetic parameters of the apoptotic changes were species-dependent. In the rat model, despite the lower light intensity (0.011 W/cm<sup>2</sup> ) the signs of acute apoptosis such as karyopyknosis and formation of apoptotic bodies (Noell et al., 1966; Novikova et al., 2014) manifested in some regions of the retina immediately after the exposure. Furthermore, they were accompanied by shift of the photoreceptor cell bodies in the vitreal direction that is commonly observed in the final step of photoreceptor apoptosis (Marc et al., 2008). By contrast, rabbit retina after illumination by the high-intensity light (0.15 W/cm<sup>2</sup> ) exhibited no signs of photoreceptor death. The photo-induced lesions could develop in the retina regions beyond the scope of our analysis, since at early stages the damage is of focal nature (Mckechnie and Foulds, 1981). Nevertheless, the multifocal histological picture of acute photoreceptor apoptosis [loss of outer and inner segments, karyopyknosis (Hoppeler et al., 1988)], accompanied by general changes in the retina cytoarchitecture, was observed only on the third day post-exposure. The revealed discrepancies between rat and rabbit models could be related to higher susceptibility of albino rats to retinal PD (van Norren and Gorgels, 2011; Hunter et al., 2012). In addition, they could be related to differences in the time intervals between the illumination onset and histological analysis, used in the rat and rabbit models (14 h versus 3 h, respectively). It seems that 3 h had been insufficient for development of photoreceptor damage in rabbits, since we observed no early signs of photoreceptor deterioration, such as swelling and shortening of their outer segments, but these effects were previously observed is response to low-intensity light for 5 h (Zernii et al., 2015b). Thus, rabbit retina required at least 5 h to develop welldefined response to visual light irradiation, regardless of its intensity.

The low-intensity light illumination (0.011 W/cm<sup>2</sup> ) did not cause any morphological alterations in the rabbit retina neither immediately after the exposure nor after 14 subsequent days. This is in agreement with the previous report that 4-h exposure of the rabbit retina to light illumination of at least 0.045 W/cm<sup>2</sup> is required for induction of noticeable retinal damage (Lawwill, 1973). Thus, visible light causes PD of photoreceptors only upon reaching of a certain threshold illumination dose (van Norren and Gorgels, 2011), and that the light-induced destructive processes in photoreceptors may be delayed, often developing in the cells that exhibit normal morphology by the time of completion of the light exposure. This suggests existence in photoreceptor cells of certain mechanisms for making life-death decisions, which depend on severity of the developing lightinduced oxidative stress and likely involve specific redoxsensitive proteins, such as recoverin.

The low-intensity illumination of rabbit retina induced only moderate increase in the content of the disulfide dimer, which likely did not promote photoreceptor apoptosis. Meanwhile, the high-intensity illumination of rabbit eyes induced accumulation of the recoverin dimer and its aggregates apparently along with C39 oxidation up to sulfinic acid, accompanied by photoreceptor death via apoptosis. In the rat model, the dimer to monomer ratio also increased upon illumination, but to a lesser extent, despite the fact that photoreceptors had more time for response to the light-induced stress (14 h versus 3 h in the rabbit model). We suppose that excess of the thioloxidized proteins, such as recoverin, was utilized in proteasome (Shang and Taylor, 2012), as confirmed by formation of mixed disulfides of recoverin with proteasome subunits upon the prolonged low-intensity illumination (Zernii et al., 2015b). Thus, photoreceptor apoptosis seems to have been induced under pronounced accumulation of disulfide dimer of recoverin and C39 oxidation to sulfinic acid. Given the existence of several mechanisms for photoreceptor death induction (Hao et al., 2002; Chen et al., 2016), its plausible that different combinations of recoverin thiol oxidation could trigger different mechanisms of cell death.

Our in vitro studies indicate that disulfide dimerization of recoverin and formation of its oxidized monomer differently affect secondary and tertiary structures of the protein, especially in the absence of calcium. α-Helicity and thermal stability of the Ca2+-free recoverin forms decreased in the following order: reduced monomer – oxidized monomer – disulfide dimer. Notably, T1/<sup>2</sup> for thermal unfolding of apo-OmRec was similar to that for the C39D mutant that mimics conversion of C39 into sulfenic/sulfinic acid (Permyakov et al., 2012), indicating importance of electrostatic charge in 39th position of the protein for maintenance of its structural stability. Meanwhile, calcium binding stabilized secondary and tertiary structures of OmRec and C39D mutant, making them closer to Ca2+-bound RmRec. Consistently, in the presence of calcium tertiary structures of the

sulfenate-containing monomer of recoverin and its C39D mutant were equivalent to that of the reduced protein (Ranaghan et al., 2013). Despite partial structural destabilization of the Ca2+-free forms of C39D and dRec, both of them exhibited common thermal denaturation profile (**Figure 6**) and fluorescence spectrum (data not shown) and remained responsive to Ca2<sup>+</sup> (**Figures 5**, **6**) indicating preservation of general fold of recoverin. Furthermore, they even demonstrated slightly elevated Ca2<sup>+</sup> affinity (**Table 1**). The thiol oxidation-induced destabilization of Ca2+-free recoverin with myristoyl group sequestered in the hydrophobic pocket seemed to facilitate the conversion into the Ca2+-bound state, in which myristoyl group and non-polar residues of the pocket are solvent-exposed (for review, see Ames and Lim, 2012). This effect was most pronounced in the case of dRec suggesting that myristate and hydrophobic pocket may have been partially solvent exposed in the Ca2+-free dimer. Consistently, our studies employing hydrophobic probe bis-ANS showed markedly increased surface hydrophobicity of dRec, apo-form of which was reminiscent of Ca2+-saturated reduced recoverin in this respect.

Since myristoyl group and the hydrophobic pocket residues of recoverin are responsible for its binding to photoreceptor membranes and GRK1, respectively (for review, see Ames and Lim, 2012), disulfide dimer of recoverin was expected to exhibit these functional activities in the absence of calcium. Indeed, Ca2+-free dRec demonstrated approximately twofold increase in the membrane-bound fraction and inhibition of the enzyme by 40%, showing its partial constitutive activity. The activity of Ca2+-free dRec toward GRK1 could be due to their interaction via the GRK1 site located outside of its N-terminal domain, or could be related to increased membrane affinity of the dimer. Similar effect was recently observed in the case of recoverin chimera with C-terminal segment of GCAP2, which in Ca2+-free conformation exhibited elevated membrane and GRK1 binding (Zernii et al., 2015a).

The increased hydrophobicity of Ca2+-bound dRec (**Figure 6D**) might have accounted for its high susceptibility to aggregation (Zernii et al., 2015b) that would reduce fraction of the functional protein capable of binding to membranes (**Figure 7A**). In addition, this reduction could be associated with alterations in recoverin-membrane interface in the case of dRec due to modification of C39 by the second recoverin molecule. The structure of Ca2+-saturated dRec also seemed to be not optimal for specific binding of GRK1 (**Table 3**). In contrast to dRec, OmRec and C39D mutant underwent more limited structural changes and their apo-forms remained non-functional. However, in the presence of calcium the membrane binding and regulatory activity of these forms were strongly suppressed (**Figure 7B**) (Permyakov et al., 2012; Ranaghan et al., 2013). We attribute this phenomenon to direct effect of the negative charge introduced into 39th position of recoverin. In the Ca2+-loaded protein bound to membrane C39 residue is positioned in close proximity to the negatively charged surface of phospholipid bilayer (**Figure 8A**), whereas in the recoverin-GRK1 complex this residue flanks amphipathic N-terminal α-helix of the enzyme (**Figure 8B**). Thus, the introduction of negative charge into position 39 of recoverin may have perturbed these interactions. Consistently, our SPR measurements and previous pull-down studies revealed that affinity of Ca2+-bound OmRec and C39D mutant to GRK1 was substantially decreased as compared to reduced recoverin (**Table 3**) (Ranaghan et al., 2013).

Overall, the functional activities of recoverin were clearly modulated by redox state of its conservative cysteine residue in a Ca2+-dependent manner. The thiol-oxidized monomer of recoverin demonstrated impaired ability to bind photoreceptor membranes and inhibit GRK1 regardless of calcium level. In contrast, disulfide dimer of the protein exhibited lowered functional activity in the Ca2+-saturated form, but acquired partial activity in the absence of calcium. It should be added that the incubation of dRec with excess of DTT restored normal membrane binding and regulatory activity of recoverin indicating that its disulfide dimerization is functionally reversible.

The revealed oxidation-induced alterations in functional properties of recoverin may affect physiological activity of the protein contributing to the light-induced retinal damage, triggered via two mechanisms depending on intensity of the light illumination (**Figure 9**) (Hao et al., 2002; Wang and Chen, 2014). We propose that disulfide dimer and thiol-oxidized monomer of recoverin play specific roles in progression of these mechanisms. The retinal damage caused by the short-term high-intensity illumination (mechanism 1, **Figure 9**) is associated with significantly reduced kinetics of rhodopsin dephosphorylation (Ishikawa et al., 2006). The excessive phosphorylation of the receptor would induce formation of stable arrestin-rhodopsin complexes (Vishnivetskiy et al., 2007), accumulation of which is known to enhance retinal degeneration (Chen et al., 2006). Under the oxidative stress induced by intense illumination some fraction of recoverin is expected to be converted into the thiol-oxidized monomer (**Figure 2**) that exhibits diminished binding to photoreceptor membranes and GRK1 inhibition at high calcium level (**Figure 7**). Besides, marked fraction of recoverin would form disulfide dimer with reduced affinity to membranes and GRK1 in the presence of calcium (**Figure 7A** and **Table 3**). Notably, one of the early events accompanying high-intensity light-induced oxidative stress and preceding photoreceptor apoptosis is a pronounced increase in intracellular calcium level through activation of nNOS and NO-sensitive guanylate cyclase (Donovan et al., 2001). Thus, the C39 oxidation of recoverin should suppress its activity, thereby promoting excessive phosphorylation of the receptor, favoring formation of arrestin-rhodopsin complexes [arrestin is accumulated in light-exposed ROS (Broekhuyse et al., 1985)] and development of the respective retinal damage. Besides, the malfunction of the oxidized Ca2+-bound recoverin could contribute to photoreceptor degeneration in the post-exposure period, when the retina is reared under dark or cyclic light conditions (Organisciak and Vaughan, 2010). These effects are similar to the photoreceptor degeneration in the rat model of cancerassociated retinopathy produced by intravitreous administration of anti-recoverin antibodies, which suppresses recoverin functioning and thereby promotes rhodopsin phosphorylation (Ohguro et al., 1999, 2001; Maeda et al., 2001). Furthermore, the mouse phenotypes overexpressing GRK1 (GRK1 upregulation is analogous to recoverin inactivation) exhibit an increased

susceptibility to high-intensity illumination (Whitcomb et al., 2010). The mechanism underlying the induction of photoreceptor apoptosis by arrestin-rhodopsin complexes remains poorly understood. We speculate that it may involve one of the protein targets of visual arrestin, JNK3 (Song et al., 2006), which is able to upregulate pro-apoptotic factor AP-1 (Coffey et al., 2002). Yet, AP-1 induction by high-intensity light was previously observed not only in wild-type animals, but also in the phenotypes lacking GRK1 or arrestin (Hao et al., 2002). AP-1 can activate downstream pro-apoptotic proteins, such as nNOS that can be upregulated by c-jun, a component of AP-1 complex (Cheng et al., 2014). Upregulation of nNOS and NO-sensitive guanylate cyclase would further increase calcium influx into the photoreceptors through cGMP-gated channels and promote their apoptosis through mitochondrial depolarization and, likely, activation of Ca2+-dependent endonucleases (Donovan et al., 2001).

Opposite to the thiol-oxidized monomer of recoverin signs of which were found only in the retinas exposed to high-intensity illumination, noticeable accumulation of disulfide dimer of the protein occurs during both high-intensity and low-intensity light illumination (**Figure 1**). At low calcium level, corresponding to the light-adapted photoreceptor cells, significant fraction of dRec could remain on the membranes and inhibit GRK1 (**Figure 7**). This would prevent rhodopsin from desensitization, thereby supporting the light-induced oxidative stress. These considerations are especially relevant to the retinal damage induced by long-term low-intensity light illumination (mechanism 2, **Figure 9**). The GRK1 downregulation observed under these conditions (Hajkova et al., 2010) should synergistically support the inhibitory action of Ca2+-free dRec. The resulting activation of Gt by rhodopsin and operation of the phototransduction cascade involve constantly active PDE6 and lowered intracellular calcium level (Wang and Chen, 2014). The latter could be supported by dRec due to the Ca2<sup>+</sup> buffering function of recoverin (Makino et al., 2004) and increased Ca2<sup>+</sup> affinity of dRec (**Figure 5**). The subsequent excessive cGMP utilization is suggested to induce metabolic stress, followed by multiple protein ubiquitination, proteasome overload and induction of apoptosis via PERK pathway of the unfolded protein response and activation of caspase 3 (Wang and Chen, 2014). Consistently, recoverin forms disulfide complexes with proteasome subunits upon prolonged exposure to low-intensity light (Zernii et al., 2015b). The multiple ubiquitination of photoreceptor proteins could be ensured by photoreceptor MDM2, an ubiquitin ligase that was recognized as a protein target of visual arrestin (Song et al., 2006). Interestingly, arrestin preferentially binds

MDM2 being in the basal (receptor-free) conformation (Song et al., 2006). Free arrestin might be accumulated in the outer segments only after translocation from the inner segments in response to the low-intensity light illumination (Hajkova et al., 2010), and when rhodopsin phosphorylation is suppressed as in the case of GRK1 inhibition by Ca2+-free dRec. Such hypothetic regulation of MDM2 by arrestin may be affected by the light-induced disulfide dimerization of the latter (Lieven et al., 2012; Zernii et al., 2015b). Notably, the disulfide dimerization would also reduce affinity of recoverin to membranes and GRK1 at high calcium levels (**Figure 7A** and **Table 3**), corresponding to the dark-adapted photoreceptor cells. This effect could contribute to photoreceptor degeneration in post-exposure period, when the retina is reared under dark or cyclic light conditions (Organisciak and Vaughan, 2010).

The conservative thiols are functionally important for some other proteins of the NCS family. For instance, substitutions of the Cys residue in photoreceptor proteins GCAP1 (C29) and GCAP2 (C35) suppress their activity (Ermilov et al., 2001; Hwang et al., 2004). Oxidative stress triggers disulfide dimerization of a non-photoreceptor NCS protein, visinin-like protein 1 (VILIP-1) (Chen et al., 2009; Liebl et al., 2014). Interestingly, VILIP-1 oxidation involves not its conservative Cys (C38) corresponding to C39 of recoverin, but the C-terminal residue C187, which is responsible for Ca2+-dependent disulfide dimerization of the protein (Chen et al., 2009; Liebl et al., 2014). Functional significance of this process is unclear, but it affects regulatory activity of the protein toward guanylate cyclase B (Chen et al., 2009). Furthermore, VILIP-1 oxidation is associated with amyotrophic lateral sclerosis (ALS), since VILIP-1 is a part of the ALS-specific protein aggregates, whereas its soluble dimers are enriched in spinal cord of the animals with experimental ALS and their formation generally correlates with the disease progression (Liebl et al., 2014).

Overall, considering conservative nature of the cysteine residues in NCS proteins, one can expect that redox sensitivity is inherent to members of this family. The thiol-oxidized forms of these proteins and functional significance of such modifications remain to be established. Since alterations in redox state of NCS proteins may contribute to progression of the oxidative stress-related neurological and neuro-ophthalmological diseases, the use of broad spectrum antioxidants or specific sulfur reductants might be regarded as a promising approach to their treatment (Organisciak and Vaughan, 2010; Lieven et al., 2012).

# AUTHOR CONTRIBUTIONS

EZ, PP, and SP participated in the design of the study and performed literature screening and analysis. OG, NT, and EZ developed animal models of light-induced retinal degeneration. NT and EZ performed immunoaffinity chromatography and immunoblotting. MS conducted mass-spectrometry studies. OG performed histological examinations. AN, EN, VV, and DZ prepared recoverin forms. IS performed calcium binding assay. AN and EN conducted fluorescence and CD measurements.

VB and EZ performed membrane binding (equilibrium centrifugation) and rhodopsin phosphorylation assays. AK performed SPR analysis. EN, EZ, and SP wrote the manuscript.

#### FUNDING

The development of rabbit models of iatrogenic light-induced retinal damage, histological characterization of pathomorphological changes in the retina as well as immunochemical and mass-spectrometric identification of alterations in redox state of recoverin were performed with support of the Russian Science Foundation (Grant No. 16-15-00255). The structural and functional in vitro studies of oxidized forms of recoverin were supported by the Russian Foundation for Basic Research (Grant No. 18-04-01250).

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00474/full#supplementary-material

FIGURE S1 | MS/MS spectra of recoverin peptides E38-R43 (A) and F23-R43 (B) containing C39 in reduced form (upper panel) or oxidized with formation of sulfinic acid (lower panel). The identified amino acid sequences of the peptides are indicated in red.

TABLE S1 | The results of peptide mass fingerprinting of recoverin samples extracted from the light-illuminated (metal halide lamp, 2,500 lx, 14 h) in vivo rat retinas.

TABLE S2 | The results of peptide mass fingerprinting (MS and MS/MS) of recoverin samples extracted from the rabbit retinas illuminated in vivo according to the scheme 1 (halogen lamp, 2,200 lx, 3 h).

stress, mediating c-Jun activation, in the presence of constitutive JNK1 activity in cerebellar neurons. J. Neurosci. 22, 4335–4345. doi: 10.1523/JNEUROSCI.22- 11-04335.2002




**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 © 2019 Zernii, Nazipova, Nemashkalova, Kazakov, Gancharova, Serebryakova, Tikhomirova, Baksheeva, Vladimirov, Zinchenko, Philippov, Senin and Permyakov. 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) and the copyright owner(s) 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.

# Disruption of Otoferlin Alters the Mode of Exocytosis at the Mouse Inner Hair Cell Ribbon Synapse

Hideki Takago1,2,3 \*, Tomoko Oshima-Takago1,2,3,4 and Tobias Moser1,3,4,5,6

1 Institute for Auditory Neuroscience and InnerEarLab, University Medical Center Göttingen, Göttingen, Germany, <sup>2</sup> Department of Rehabilitation for Sensory Functions, Research Institute, National Rehabilitation Center for Persons with Disabilities, Saitama, Japan, <sup>3</sup> Collaborative Research Center 889 Cellular Mechanisms of Sensory Processing, Göttingen, Germany, <sup>4</sup> Göttingen Graduate School for Neurosciences and Molecular Biosciences, University of Göttingen, Göttingen, Germany, <sup>5</sup> Auditory Neuroscience Group, Max Planck Institute for Experimental Medicine, Göttingen, Germany, <sup>6</sup> Synaptic Nanophysiology Group, Max Planck Institute for Biophysical Chemistry, Göttingen, Germany

Sound encoding relies on Ca2+-mediated exocytosis at the ribbon synapse between cochlear inner hair cells (IHCs) and type I spiral ganglion neurons (SGNs). Otoferlin, a multi-C<sup>2</sup> domain protein, is proposed to regulate Ca2+-triggered exocytosis at this synapse, but the precise mechanisms of otoferlin function remain to be elucidated. Here, performing whole-cell voltage-clamp recordings of excitatory postsynaptic currents (EPSCs) from SGNs in otoferlin mutant mice, we investigated the impact of Otof disruption at individual synapses with single release event resolution. Otof deletion decreased the spontaneous release rate and abolished the stimulus-secretion coupling. This was evident from failure of potassium-induced IHC depolarization to stimulate release and supports the proposed role of otoferlin in Ca2<sup>+</sup> sensing for fusion. A missense mutation in the Otof gene (pachanga), in which otoferlin level at the IHC plasma membrane was lowered without changing its Ca2<sup>+</sup> binding, also reduced the spontaneous release rate but spared the stimulus-secretion coupling. The slowed stimulated release rate supports the hypothesis that a sufficient abundance of otoferlin at the plasma membrane is crucial for the vesicle supply. Large-sized monophasic EPSCs remained present upon Otof deletion despite the drastic reduction of the rate of exocytosis. However, EPSC amplitude, on average, was modestly decreased. Moreover, a reduced contribution of multiphasic EPSC was observed in both Otof mutants. We argue that the presence of large monophasic EPSCs despite the exocytic defect upon Otof deletion supports the uniquantal hypothesis of transmitter release at the IHC ribbon synapse. Based upon the reduced contribution of multiphasic EPSC, we propose a role of otoferlin in regulating the mode of exocytosis in IHCs.

Keywords: auditory, cochlea, hair cell, spiral ganglion neuron, ribbon synapse, otoferlin, calcium, EPSC

# INTRODUCTION

Ca2<sup>+</sup> influx and subsequent neurotransmitter release at inner hair cell (IHC) active zones govern sound encoding at the first auditory synapse. Excitatory postsynaptic currents (EPSCs) recorded from type I spiral ganglion neurons (SGNs) show remarkable variability in amplitude and shape. This EPSC heterogeneity led to the hypothesis of synchronized multiquantal release (MQR) at hair cell synapses (Glowatzki and Fuchs, 2002; Keen and Hudspeth, 2006; Goutman and Glowatzki, 2007; Neef et al., 2007; Li et al., 2009; Grant et al., 2010; Graydon et al., 2011; Schnee et al., 2013).

#### Edited by:

Michael R. Kreutz, Leibniz Institute for Neurobiology (LG), Germany

#### Reviewed by:

Martin Heine, Johannes Gutenberg University Mainz, Germany Kirill Volynski, University College London, United Kingdom

#### \*Correspondence:

Hideki Takago takago-hideki@rehab.go.jp

Received: 29 September 2018 Accepted: 19 December 2018 Published: 09 January 2019

#### Citation:

Takago H, Oshima-Takago T and Moser T (2019) Disruption of Otoferlin Alters the Mode of Exocytosis at the Mouse Inner Hair Cell Ribbon Synapse. Front. Mol. Neurosci. 11:492. doi: 10.3389/fnmol.2018.00492

MQR would cause monophasic (temporally compact) EPSCs when the exocytosis of vesicles comprising a MQR event is highly synchronized or multiphasic (temporally non-compact) EPSCs when synchronization of MQR is poor. However, whereas Ca2<sup>+</sup> influx by presynaptic depolarization increases the EPSC amplitude at the frog hair cell synapse (Li et al., 2009; Graydon et al., 2011), neither changes in voltage-gated Ca2<sup>+</sup> influx (Glowatzki and Fuchs, 2002; Grant et al., 2010) nor those in presynaptic Ca2<sup>+</sup> buffering (Goutman and Glowatzki, 2007) affects the EPSC amplitude distribution at the rat IHC ribbon synapse. Strikingly, even when the presynaptic Ca2<sup>+</sup> influx was abolished in mouse IHCs, EPSC size remained heterogeneous and the charge distribution of mono- and multiphasic EPSCs were unchanged (Chapochnikov et al., 2014). This, together with other experimental observations and modeling, led to the proposal that uniquantal release (UQR) is a candidate mechanism for exocytosis at the mammalian IHC ribbon synapse, whereby a combination of large clusters of postsynaptic AMPA receptors and presynaptic fusion pore dynamics would generate large and variably shaped EPSCs (Chapochnikov et al., 2014; for review, see Takago and Oshima-Takago, 2018). The UQR hypothesis of IHC exocytosis has received further support by recent studies manipulating presynaptic IHC Ca2<sup>+</sup> influx (Huang and Moser, 2018) or directly measuring membrane fusion steps via cellattached membrane capacitance recordings from IHCs (Grabner and Moser, 2018).

Disruption of OTOF, coding for otoferlin, was identified to cause hereditary deafness DFNB9 (Yasunaga et al., 1999), while missense mutations of OTOF can lead to less profound hearing impairment (Varga et al., 2003; Marlin et al., 2010; Vogl et al., 2016; for review, see Pangršic et al., 2012 ˇ ; Moser and Starr, 2016). As a multi-C<sup>2</sup> domain protein, otoferlin, in analogy to synaptotagmins, was initially proposed to serve as a Ca2<sup>+</sup> sensor for fusion at the IHC ribbon synapse based on a functional analysis of a mouse line with a null mutation in otoferlin (Otof−/−, Roux et al., 2006) and biochemical studies (Johnson and Chapman, 2010). Moreover, a role of otoferlin in efficient vesicle priming is proposed based upon an analysis of a mouse line called pachanga that carries a missense mutation in otoferlin C2F domain (OtofD1767G/D1767<sup>G</sup> or OtofPga/Pga , Pangršic et al., ˇ 2010). Whereas such a mutation does not affect Ca2<sup>+</sup> binding, OtofPga/Pga IHCs show reduced membrane-bound otoferlin level but unaltered fusion, thus suggesting that the hearing impairment arises from strongly reduced vesicle replenishment (Pangršicˇ et al., 2010). This hypothesis is further supported by analyses of mouse lines carrying a missense mutation in otoferlin C2C domain that again lowers membrane-bound otoferlin level but unaltered fusion (OfofI515T/I515<sup>T</sup> , Strenzke et al., 2016) and a null mutation in transmembrane recognition complex40 receptor tryptophan-rich basic protein that is essential for the insertion of otoferlin into the endoplasmic reticulum in IHCs (Vogl et al., 2016). Thus, the abundance of otoferlin in IHCs is critical for maintaining the vesicle resupply to the ribbon-type active zone. Otoferlin's function in vesicle replenishment appears to involve the regulation of short filamentous tethering formation between synaptic vesicles and the active zone membrane (Vogl et al., 2015) and the facilitation of clearance of vesicular release sites from previously exocytosed membranes (exocytosisendocytosis coupling) via an interaction with the endocytic adaptor protein 2 (Duncker et al., 2013; Jung et al., 2015). On the other hand, a recent study utilizing mice with double missense mutations in otoferlin C2C domain that affect Ca2<sup>+</sup> sensing (OtofD515A,D517A/D515A,D517A, Michalski et al., 2017) has reported that fast and sustained components of release are reduced despite unaltered otoferlin level, probably due to dysfunctional Ca2<sup>+</sup> binding of otoferlin.

In order to further examine the otoferlin's function we performed postsynaptic voltage-clamp recordings from afferent boutons of type I SGNs in wild-type (Otof <sup>+</sup>/+), OtofPga/Pga and Otof−/<sup>−</sup> mice and investigated exocytosis at the levels of single synapses and single release events. Analyzing the spontaneous and stimulated release rates as well as the amplitude and shape of EPSCs, we find evidence for roles of otoferlin in Ca2+-dependent fusion and replenishment of vesicles. Moreover, we propose an additional role of otoferlin in regulating the exocytic mode of IHCs to facilitate multiphasic EPSCs, potentially by controlling the vesicle fusion pore during uniquantal release at the IHC ribbon synapse.

#### MATERIALS AND METHODS

#### Ethics Statement

All experiments complied with national animal care guidelines in Germany and Japan and were approved by the University of Göttingen board for animal welfare together with the animal welfare office of the state of Lower Saxony (Germany) as well as National Rehabilitation Center for Persons with Disabilities animal experimentation committee (Japan).

#### Animals and Preparations

Postnatal day (P) 8–11 mice of either sex were used. Generation and general description of OtofPga/Pga (Schwander et al., 2007) and Otof−/<sup>−</sup> (Reisinger et al., 2011) mice were previously provided. As wild-type controls, C57BL6 mice (Otof <sup>+</sup>/+), which were not littermates of OtofPga/Pga or Otof−/<sup>−</sup> mice, were employed. In total, 33 mice (13 Otof <sup>+</sup>/+, 7 OtofPga/Pga and 13 Otof−/<sup>−</sup> mice, respectively) were used in the present study. After decapitation under deep carbon dioxide inhalation anesthesia, the apical coils of organ of Corti were harvested out of the cochlea.

#### Electrophysiology

Whole-cell voltage-clamp recordings from postsynaptic boutons of mouse type I spiral ganglion neurons in apical coils of freshly dissected organ of Corti were performed as previously described for rats (Glowatzki and Fuchs, 2002; Rutherford et al., 2012) and for mice (Pangršic et al., 2010 ˇ ; Jing et al., 2013; Chapochnikov et al., 2014). The recording pipette resistance was 8–15 M after pressure polishing (Goodman and Lockery, 2000). The intracellular solution contained (in mM): 150 CsCl (or 150 KCl in some recordings), 3.5 MgCl2, 0.1 CaCl2, 5 EGTA, 5 HEPES, and 2.5 Na2ATP, pH 7.2. The extracellular solution (artificial perilymph, aPL) for both dissection and recording

contained (in mM): 5.8 KCl, 155 NaCl, 0.9 MgCl2, 1.3 CaCl2, 0.7 NaH2PO4, 5.6 D-glucose, and 10 HEPES, pH 7.4. In the high K<sup>+</sup> extracellular solution to depolarize presynaptic IHCs, 40 out of 155 mM NaCl were replaced with equimolar KCl. In most recordings, tetrodotoxin (1–2 µM) was added to block voltage-gated Na<sup>+</sup> channels. Currents were low-pass filtered at 5–10 kHz and sampled at 20–50 kHz. EPSCs were recorded at a holding potential of −90 mV (∼4 mV liquid junction potential not corrected) at room temperature (21–24◦C).

#### Chemicals and Equipment

All chemicals were purchased from Sigma-Aldrich (St. Louis, MO, United States) except for tetrodotoxin (Tocris Bioscience, Bristol, United Kingdom or Wako Pure Chemical Industries, Ltd., Osaka, Japan). The EPC-10 amplifier controlled by Patchmaster software (HEKA Elektronik, Lambrecht, Germany) as well as an upright microscope with differential interference contrast optics (Axioskop FS2, Carl Zeiss, Oberkochen, Germany or BX51WI, Olympus, Tokyo, Japan) was used.

#### Data Analysis and Statistics

For detection and analysis of EPSCs, MiniAnalysis software (Synaptosoft, Decatur, GA, United States) was used with a detection threshold set at 3–5 times greater than the root mean square (rms) of the baseline noise. To classify EPSCs into monoor multiphasic, the methods introduced by Grant et al. (2010) as well as Chapochnikov et al. (2014) were employed. For plotting, IGOR Pro (Wavemetrics, Lake Oswego, OR, United States), Sigmaplot (Systat Software Inc., San Jose, CA, United States) were used. The passive membrane properties such as series resistance (Rs, 40 ± 2 M for 13 Otof <sup>+</sup>/<sup>+</sup> SGNs, 41 ± 3 M for 7 OtofPga/Pga SGNs, 40 ± 2 M for 13 Otof−/<sup>−</sup> SGNs), membrane capacitance (Cm, 1.9 ± 0.2 pF for Otof <sup>+</sup>/+, 1.9 ± 0.1 pF for OtofPga/Pga, 1.8 ± 0.2 pF for Otof−/−) and membrane input resistance (Rm, 498 ± 141 M for Otof <sup>+</sup>/+, 831 ± 225 M for OtofPga/Pga, 651 ± 106 M for Otof−/−) were calculated as previously described (Chapochnikov et al., 2014). Recordings with R<sup>s</sup> > 50 M or less than 20 EPSCs were excluded from the EPSC amplitude/charge or kinetics analysis, but included in the EPSC frequency analysis (except for recordings with less than 5 EPSCs) not to ignore very low-frequency auditory nerve fibers. Data is shown as mean ± SEM. Statistical significance was evaluated by Student's t-test or one way ANOVA followed by post-hoc Tukey's test.

#### RESULTS

#### Missense Mutation of Otoferlin Decreases the Rate of Spontaneous Release, Loss of Otoferlin Abolishes Stimulus-Secretion Coupling in IHCs

EPSCs were recorded from afferent boutons of type I SGNs in P8–11 Otof <sup>+</sup>/+, OtofPga/Pga (missense mutation with reduced abundance of otoferlin) and Otof−/<sup>−</sup> mice. At rest (1.3 mM Ca2<sup>+</sup> and 5.8 mM K<sup>+</sup> in aPL), SGNs of all three genotypes

FIGURE 1 | Otoferlin regulates spontaneous and high K+-stimulated release at the mouse IHC ribbon synapse. (A) Sample traces of EPSCs recorded from exemplar boutons of type I SGNs from a P9 Otof <sup>+</sup>/<sup>+</sup> (top), a P8 OtofPga/Pga (middle) and a P9 Otof <sup>−</sup>/<sup>−</sup> (bottom) mouse. Each recording started in the extracellular solution containing 5.8 mM KCl, and 40 mM KCl was bath-applied during the time indicated by horizontal bars to depolarize presynaptic IHCs. (B) Summary of EPSC frequency in 5.8 mM (5K, spontaneous rate) and 40 mM (40 K, stimulated rate) [K+]e. Circles and diamonds indicate individual and mean ± SEM values for Otof <sup>+</sup>/<sup>+</sup> (WT, black), OtofPga/Pga (Pga, dark gray) and Otof <sup>−</sup>/<sup>−</sup> (KO, light gray) SGNs. Asterisks at the bottom (spontaneous rate) and top (stimulated rate) show significant differences (∗p < 0.05). Note that the vertical axes for this and the next panels are shown on logarithmic scale. In some recordings, where 40 mM KCl was not applied, data points for 5.8 mM [K+]<sup>e</sup> alone were plotted.

steadily exhibited spontaneous EPSCs (**Figure 1A**), but their rate was reduced in SGNs of both mutants. The frequencies of spontaneous EPSCs were 0.71 ± 0.19 Hz for Otof <sup>+</sup>/<sup>+</sup> (n = 13, **Figure 1B**), 0.18 ± 0.07 Hz for OtofPga/Pga (n = 7, p = 0.048 compared to Otof <sup>+</sup>/+) and 0.14 ± 0.04 Hz for Otof−/<sup>−</sup> SGNs (n = 13, p = 0.010 and 0.985 compared to Otof <sup>+</sup>/<sup>+</sup> and OtofPga/Pga, respectively). Hence, unlike disruption of synaptotagmin I or II at the central synapses, which is thought to unclamp spontaneous release (Nishiki and Augustine, 2004; Pang et al., 2006; Xu et al., 2009; Kochubey and Schneggenburger, 2011; Lee et al., 2013; Liu et al., 2014), Otof deletion did not increase the spontaneous EPSC frequency. This argues against a clamping function of otoferlin for spontaneous release and highlights the importance of otoferlin in vesicle replenishment, since the spontaneous rate was reduced to a similar extent in Otof−/<sup>−</sup> and OtofPga/Pga mice, which in a previous study showed intact vesicle fusion but impaired replenishment in IHCs (Pangršicˇ et al., 2010).

Next, we examined the stimulus-secretion coupling in IHCs. High K<sup>+</sup> solution (40 mM in aPL) was bath-applied onto IHCs to increase the open probability of their CaV1.3 L-type Ca2<sup>+</sup> channels. The frequency of EPSCs in Otof <sup>+</sup>/<sup>+</sup> SGNs was increased from 0.25 ± 0.11 (5.8 mM, control) to 9.67 ± 2.90 Hz (first 10 s during high K<sup>+</sup> stimulation, n = 6, p = 0.021, Student's paired t-test, **Figures 1A,B**). Here, the spontaneous EPSC frequency for this control (5.8 mM K+) was undervalued due

to our preferential application of high K+-containing aPL onto low-frequency synapses, since remaining more active synapses (1.10 ± 0.28 Hz, n = 7) provided a sufficient number of EPSCs for the analyses of EPSC size and shape even in the normal K <sup>+</sup>-containing aPL. In OtofPga/Pga SGNs, high K<sup>+</sup> stimulation increased the frequency of EPSCs in 4 out of 5 OtofPga/Pga SGNs from 0.20 ± 0.10 (control) to 1.86 ± 0.57 Hz (high K+, n = 5, p = 0.047), indicating that the missense Otof mutation in OtofPga/Pga mice spares stimulus-secretion coupling in IHCs. The reduced rate of stimulated release in OtofPga/Pga IHCs (p = 0.032 compared to Otof <sup>+</sup>/+) is compatible with impaired vesicle replenishment (Pangršic et al., 2010 ˇ ). In contrast, the high K<sup>+</sup> stimulation failed to increase the EPSC frequency in Otof−/<sup>−</sup> SGNs (control: 0.12 ± 0.06 Hz vs. high K+: 0.10 ± 0.05 Hz, n = 5, p = 0.302), indicating that Otof deletion abolishes stimulus-secretion coupling in IHCs. The difference in stimulated release rate between Otof <sup>+</sup>/<sup>+</sup> and Otof−/<sup>−</sup> SGNs again showed significance (p = 0.010).

#### Otoferlin Disruption Decreases EPSC Amplitude and the Fraction of Multiphasic EPSCs

We studied the effects of Otof disruption on the size and shape of EPSCs. As reported in previous studies using the organ of Corti of rats (Glowatzki and Fuchs, 2002; Grant et al., 2010), high K <sup>+</sup> stimulation of mouse Otof <sup>+</sup>/<sup>+</sup> organ of Corti increased the frequency of EPSCs without affecting the EPSC size (monophasic EPSC amplitude: 141 ± 31 pA for control vs. 157 ± 24 pA for high K+, multiphasic EPSC amplitude: 101 ± 13 pA for control vs. 109 ± 16 pA for high K+) or the fraction of multiphasic EPSCs (41.9 ± 3.8 % for control vs. 41.1 ± 1.7 % for high K+). Moreover, EPSC size and kinetics distributions were unaltered by high K<sup>+</sup> stimulation in each phenotype SGNs (**Supplementary Figures S1, S2**). Therefore, EPSCs recorded in control and high K<sup>+</sup> were pooled for subsequent analyses in this study.

Surprisingly, despite the reduction in release rate and the lack of stimulus-secretion coupling, Otof−/<sup>−</sup> synapses showed variable EPSC amplitudes from 11 up to several hundred pA (**Figure 2**). As shown in **Figure 3** and **Table 1**, the average amplitudes of mono- and multiphasic EPSCs as well as the average charge transfer of mixed EPSCs were reduced in Otof−/<sup>−</sup> SGNs (p = 0.013 for monophasic EPSC amplitude, p = 0.015 for multiphasic EPSC amplitude, p = 0.011 for EPSC charge). There was a trend to smaller EPSC amplitudes also for OtofPga/Pga SGNs, which, however, did not reach significance. EPSC kinetics (i.e., rise time and decay time for monophasic as well as time to peak and half width for multiphasic), on the other hand, was not different among those three groups (**Figures 2**, **4**).

Notably, the scatter plot of EPSC amplitude versus charge shows the predominance of monophasic EPSCs clustering around the unity line in SGNs of both Otof mutants, while a substantial group of multiphasic EPSCs with smaller amplitude but similar charge exists in Otof <sup>+</sup>/<sup>+</sup> SGNs (**Figure 5A**). The fraction of multiphasic EPSC in normal K+-containing aPL was significantly reduced in SGNs of both mutants (41.3 ± 2.6%

for Otof <sup>+</sup>/+, 24.0 ± 5.2% for OtofPga/Pga and 13.2 ± 3.5% for Otof−/−, p < 0.01 between Otof <sup>+</sup>/<sup>+</sup> and OtofPga/Pga , p < 0.001 between Otof <sup>+</sup>/<sup>+</sup> and Otof−/−, p = 0.145 between OtofPga/Pga and Otof−/−, **Figure 5B**). Also, the fraction of multiphasic EPSC in high K+-containing aPL was also significantly reduced in both mutant**s** (40.3 ± 2.3% for Otof <sup>+</sup>/+, 18.7 ± 2.9% for OtofPga/Pga and 11.8 ± 5.9% for Otof−/−, p < 0.01 between Otof <sup>+</sup>/<sup>+</sup> and OtofPga/Pga , p < 0.01 between Otof <sup>+</sup>/<sup>+</sup> and Otof−/−, p = 0.408 between OtofPga/Pga and Otof−/−). Thus, monophasic EPSCs dominate transmission in the absence of otoferlin, suggesting that otoferlin regulates the mode of spontaneous and stimulated release at the first auditory synapse.

#### DISCUSSION

#### Roles of Otoferlin for Stimulus-Secretion Coupling and Vesicle Replenishment

In the present study, we recorded EPSCs from the postsynaptic afferent boutons of type I SGNs in wild-type and otoferlin mutant mice, and tested the function of otoferlin in exocytosis at the IHC ribbon synapse. We found a complete disruption of stimulussecretion coupling in Otof−/<sup>−</sup> SGNs. As Ca2<sup>+</sup> influx and vesicle availability on a morphological level are maintained at the AZs of Otof−/<sup>−</sup> IHCs (Roux et al., 2006), we argue that Ca2<sup>+</sup> no longer efficiently drives fusion in the absence of otoferlin. This supports the role of otoferlin as a Ca2<sup>+</sup> sensor of fusion (Roux et al., 2006; Johnson and Chapman, 2010; Michalski et al., 2017). But how is otoferlin-independent spontaneous release regulated? Possible mechanisms include (1) an additional high affinity Ca2<sup>+</sup> sensor whose capacity is saturated at resting [Ca2+]<sup>i</sup> and (2) a Ca2+ independent release process. For (1), the activation range of the remaining Ca2<sup>+</sup> sensor should be below the resting [Ca2+]<sup>i</sup> of a few tens of nanomolar (Beutner and Moser, 2001). However, to date, there has been no report about such a molecule that meets this requirement. The closest candidate Ca2<sup>+</sup> sensor for fusion is

the C<sup>2</sup> domain protein Doc2b that promotes membrane fusion at submicromolar Ca2<sup>+</sup> concentrations (but higher than 100 nM) (Groffen et al., 2010).

Interestingly, the absence of otoferlin did not increase the spontaneous EPSC frequency, which is in contrast to the effect of ablation of synaptotagmin I (Xu et al., 2009) or II (Pang et al., 2006) in the mouse central neurons, further highlighting differences among otoferlin and synaptotagmin I (Reisinger et al., 2011). We note here that the "spontaneous" release from resting IHCs likely includes Ca2+-evoked release that is triggered by rare openings of

TABLE 1 | EPSC size in SGN afferent boutons of wild-type and Otof mutant mice.


CaV1.3 channels. We consider it likely that the reduction of spontaneous release reflects a combination of disrupted Ca2+-triggered spontaneous release of IHCs and reduced vesicle replenishment.

Previous studies showed that otoferlin abundance at the IHC plasma membrane correlates with the presynaptic function and sound encoding (Strenzke et al., 2016). OtofPga/Pga mice, wherein otoferlin membrane abundance is attenuated down to 3% of Otof <sup>+</sup>/<sup>+</sup> mice (Pangršic et al., 2010 ˇ ; Strenzke et al., 2016), displays lower EPSC frequency upon high K+-induced IHC depolarization. Although our combination of postsynaptic recordings and high K<sup>+</sup> stimulation cannot track fast stimulussecretion coupling, the robust EPSC rates in OtofPga/Pga afferents (**Figure 1A**) suggests that Ca2+-triggered membrane fusion is intact despite the mutation in the C2F domain (i.e., D1767G, Pangršic et al., 2010 ˇ ). On the other hand, the dual mutations in the C2C domain (i.e., D515A and D517A), which alter its Ca2+-binding (Johnson and Chapman, 2010) but preserves otoferlin level (Michalski et al., 2017), impairs both vesicle fusion and replenishment functions of otoferlin, suggesting that Ca2+-sensing of otoferlin is critical for both steps of exocytosis (Michalski et al., 2017). Thus, not only Ca2+-sensing capacity but also plasma membrane level of otoferlin is essential for IHC exocytosis.

#### Multiquantal Versus Uniquantal Release: Can Otoferlin Disruption Provide Insight?

The heterogeneity of EPSC amplitude and shape is a hallmark of ribbon synapses, but the underlying mechanisms are not well understood. Notably, the mammalian IHC ribbon synapse differs from other ribbon synapses such as the amphibian and reptile hair cell synapses (Li et al., 2009; Graydon et al., 2011; Schnee et al., 2013) as well as the mammalian retinal bipolar-AII cell synapse (Singer et al., 2004; Mehta et al., 2013), where low [Ca2+]<sup>i</sup> conditions break EPSCs down into unitary events. In contrast, at the rodent IHC ribbon synapse the EPSC amplitude remains sizable despite the massive buffering of presynaptic Ca2<sup>+</sup> (Goutman and Glowatzki, 2007) or abolition of Ca2<sup>+</sup> influx (Chapochnikov et al., 2014). This and other findings have led us to consider an alternative hypothesis for explaining EPSC amplitude and shape heterogeneity at the IHC synapse: the large EPSC amplitude may reflect activation of a large number of postsynaptic AMPA receptors (Saito, 1990; Meyer et al., 2009) activated by glutamate liberated from a single synaptic vesicle. Post-fusion regulation of such uniquantal release

by a dynamic fusion pore may explain multiphasic EPSCs (successive bouts of release through a flickering pore) and small monophasic EPSCs (incomplete release of a vesicle's glutamate content) (Chapochnikov et al., 2014; Huang and Moser, 2018; for review, see Moser and Vogl, 2016; Takago and Oshima-Takago, 2018).

Given the presence of large monophasic EPSCs at the Otof−/<sup>−</sup> IHC afferent synapse (**Figure 2**), i.e., under conditions of strongly reduced rate of release that seem not permissive for synchronized fusion of multiple vesicles, we favor the interpretation that uniquantal release prevails in IHCs. Interestingly, the fraction of multiphasic EPSCs was smaller when otoferlin was disrupted. Within the framework of the uniquantal release hypothesis of IHC exocytosis this can be explained as favoring full-collapse fusion and/or singular fusion pore openings. Accordingly, otoferlin would then enhance promote fusion pore flickering. Alternatively, within the framework of multiquantal release hypothesis, which the present study cannot exclude, our observation might be explained as highly synchronized fusion of a smaller number of vesicles upon Otof disruption. Besides, either hypothetic mechanism might be modulated by altered Ca2<sup>+</sup> signaling at the release site as might have resulted from the scaffolding function of Otoferlin for Ca2<sup>+</sup> channels (Hams et al., 2017) or attenuated fast-inactivating Ca2<sup>+</sup> currents in Otoferlin-lacking IHCs (Vincent et al., 2017). However, we note that we had previously found normal Ca2<sup>+</sup> signals at the individual IHC AZs in OtofPga/Pga mice (Pangršic et al., ˇ 2010). Further testing these hypotheses and elucidating the

underlying molecular events remain important goals for future studies.

A smaller fraction of multiphasic EPSCs was also found in OtofPga/Pga SGNs (**Figure 5**) and we suggest that the reduction of otoferlin levels at the plasma membrane suffices to alter the IHC exocytic mode. Tens of missense mutations have been described for patients with otoferlin-related deafness (for review, see Pangršic et al., 2012 ˇ ; Santarelli et al., 2015). It is noteworthy that even a missense mutation of otoferlin such as D1767G (pachanga) drastically decrease the heterogeneity of EPSC shape (this study). Although the functional significance of multiphasic EPSCs at the IHC ribbon synapse remains to be clarified, a wide range of spike jitter in the SGN afferent boutons caused by variable EPSP waveforms (Rutherford et al., 2012) may contribute to the heterogeneity of auditory nerve fiber responses in response to sound (Liberman, 1982; Grant et al., 2010). By promoting the shift in IHC exocytic mode, otoferlin may serve to endow the diversity in sound encoding at this ribbon-type synapse.

#### AUTHOR CONTRIBUTIONS

HT and TM designed the study. HT and TO-T performed experiments and analyzed the data. HT, TO-T, and TM wrote the manuscript. The experiments were performed at the University Medical Center Göttingen and Research Institute of National Rehabilitation Center for Persons with Disabilities.

#### FUNDING

This study was supported by a fellowship of MED-EL company (to HT) as well as grants from the German Research Foundation through the Collaborative Research Center 889 (Project A2 to TM) and JSPS KAKENHI (Grant Nos. 25670722 and 16K11204 to HT).

#### ACKNOWLEDGMENTS

fnmol-11-00492 January 5, 2019 Time: 10:57 # 7

We would like to thank Dr. Ulrich Müller and Dr. Ellen Reisinger for providing otoferlin mutant mice as well as Dr. Elisabeth Glowatzki and Dr. Darina Khimich for teaching postsynaptic patch clamp technique. We also would like to thank Drs. Masao Tachibana, Tomoyuki Takahashi, Takeshi Sakaba, Jeong-Seop Rhee, Nikolai Chapochnikov, Chao-Hua Huang, Yukihiro Nakamura, Yoshinori Sahara, Shinichi Iwasaki, Kensuke Watanabe, Ken Kitamura, Yuko Seko, and Koichi Mori for comments and helps.

# REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00492/full#supplementary-material

FIGURE S1 | High K<sup>+</sup> stimulation does not alter the EPSC size distribution in wild-type or Otoferlin mutant SGNs (A–C) Cummulative histograms of monophasic EPSC amplitude (A), multiphasic EPSC amplitude (B) and EPSC charge transfer (C) derived from exemplar Otof <sup>+</sup>/+, OtofPga/Pga and Otof <sup>−</sup>/<sup>−</sup> SGNs before (5.8 mM [K+]e, dotted lines) and during high K<sup>+</sup> (40 mM [K+]e, solid lines) stimulation. No significant differences in each phenotype.

FIGURE S2 | High K<sup>+</sup> stimulation does not alter the EPSC kinetics distribution in wild-type or Otoferlin mutant SGNs (A,B) Cummulative histograms of 10–90 % rise time (A1) and dacay time constant (A2) of monophasic EPSCs as well as time to rise (B1) and half-width (B2) of multiphasic EPSCs derived from exemplar Otof <sup>+</sup>/+, OtofPga/Pga and Otof <sup>−</sup>/<sup>−</sup> SGNs before (5.8 mM [K+]e, dotted lines) and during high K<sup>+</sup> (40 mM [K+]e, solid lines) stimulation. No significant differences in each phenotype.



**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 © 2019 Takago, Oshima-Takago and Moser. 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) and the copyright owner(s) 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.

# Synaptotagmin Ca2<sup>+</sup> Sensors and Their Spatial Coupling to Presynaptic Ca<sup>v</sup> Channels in Central Cortical Synapses

Grit Bornschein and Hartmut Schmidt\*

Carl-Ludwig Institute for Physiology, Medical Faculty, University of Leipzig, Leipzig, Germany

Ca2<sup>+</sup> concentrations drop rapidly over a distance of a few tens of nanometers from an open voltage-gated Ca2<sup>+</sup> channel (Cav), thereby, generating a spatially steep and temporally short-lived Ca2<sup>+</sup> gradient that triggers exocytosis of a neurotransmitter filled synaptic vesicle. These non-steady state conditions make the Ca2+-binding kinetics of the Ca2<sup>+</sup> sensors for release and their spatial coupling to the Cavs important parameters of synaptic efficacy. In the mammalian central nervous system, the main release sensors linking action potential mediated Ca2<sup>+</sup> influx to synchronous release are Synaptotagmin (Syt) 1 and 2. We review here quantitative work focusing on the Ca2<sup>+</sup> kinetics of Syt2-mediated release. At present similar quantitative detail is lacking for Syt1-mediated release. In addition to triggering release, Ca2<sup>+</sup> remaining bound to Syt after the first of two successive high-frequency activations was found to be capable of facilitating release during the second activation. More recently, the Ca2<sup>+</sup> sensor Syt7 was identified as additional facilitation sensor. We further review how several recent functional studies provided quantitative insights into the spatial topographical relationships between Syts and Cavs and identified mechanisms regulating the sensor-to-channel coupling distances at presynaptic active zones. Most synapses analyzed in matured cortical structures were found to operate at tight, nanodomain coupling. For fast signaling synapses a developmental switch from loose, microdomain to tight, nanodomain coupling was found. The protein Septin5 has been known for some time as a developmentally down-regulated "inhibitor" of tight coupling, while Munc13-3 was found only recently to function as a developmentally up-regulated mediator of tight coupling. On the other hand, a highly plastic synapse was found to operate at loose coupling in the matured hippocampus. Together these findings suggest that the coupling topography and its regulation is a specificity of the type of synapse. However, to definitely draw such conclusion our knowledge of functional active zone topographies of different types of synapses in different areas of the mammalian brain is too incomplete.

#### Edited by:

Beat Schwaller, Université de Fribourg, Switzerland

#### Reviewed by:

Kirill Volynski, University College London, United Kingdom Lu-Yang Wang, University of Toronto, Canada

#### \*Correspondence:

Hartmut Schmidt hartmut.schmidt@ medizin.uni-leipzig.de

Received: 11 October 2018 Accepted: 21 December 2018 Published: 15 January 2019

#### Citation:

Bornschein G and Schmidt H (2019) Synaptotagmin Ca2+ Sensors and Their Spatial Coupling to Presynaptic Cav Channels in Central Cortical Synapses. Front. Mol. Neurosci. 11:494. doi: 10.3389/fnmol.2018.00494

Keywords: Synaptotagmin, release sensor, Ca2<sup>+</sup> channel, coupling distance, nanodomain, microdomain

# INTRODUCTION

The release of neurotransmitter from presynaptic terminals and its modulation via synaptic plasticity are the bedrocks of directed information flow within neuronal circuits of the central nervous system (CNS). An action potential (AP) triggers the release of a neurotransmitter filled synaptic vesicle (SV) by opening voltagegated Ca2<sup>+</sup> channels (Cavs) in the presynaptic active zone (AZ). The inflowing Ca2<sup>+</sup> diffuses toward the SV, which bears the primary Ca2<sup>+</sup> sensor proteins Synaptotagmin (Syt) 1 or 2 on its surface that are required for triggering its fusion with the presynaptic plasma membrane. Ca2<sup>+</sup> binding to Syt changes its conformation and the resulting interaction with proteins of the core release machinery, the soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) proteins, and other proteins at the AZ ultimately triggers the fusion of the SV with the presynaptic plasma membrane (Südhof, 2013; Kaeser and Regehr, 2014). Thus, although aspects of this process are still not understood, it can be noted that Syts link Ca2<sup>+</sup> influx to SNARE mediated SV fusion.

The process of transmitter release is probabilistic, i.e., not every AP leads to exocytosis; rather it triggers the release of a SV only with a certain probability. The average vesicular release probability (pr) can be quantified by way of analyzing fluctuations in postsynaptic current amplitudes (PSCs) under conditions of different p<sup>r</sup> , e.g., at different concentrations of extracellular Ca2<sup>+</sup> ([Ca2+]e; Clements and Silver, 2000). Instead of recording PSCs, recently it became also feasible to more directly monitor glutamate release from individual boutons by imaging the fluorescence of a genetically encoded, intensitybased glutamate-sensing fluorescent reporter (iGlusnFr; Jensen et al., 2017; Helassa et al., 2018; Marvin et al., 2018).

The initial p<sup>r</sup> (pr1) is an important factor not only in determining the release fidelity for a single AP but also in setting the properties of short-term plasticity of a synapse (Zucker and Regehr, 2002; Abbott and Regehr, 2004). This can be illustrated by a simple example of paired-pulse ratio (PPR) in the absence of replenishment of SVs between the two APs of a paired-pulse experiment. In this case PPR = pr2/pr1 (1-pr1), where pr2 is the release probability of the second release process, which may differ from pr1. In general, it can be noted that if pr1 > 0.5 the synapse will depress, i.e., PPR < 1, and only the magnitude of the depression will depend on pr2. However, if pr1 < 0.5 the synapse will show facilitation or depression depending on pr2.

The p<sup>r</sup> depends on the Ca2+-binding kinetics of the release machinery, i.e., the Ca2+-binding kinetics of Syt in the context of the SNARE and other proteins at the AZ, and on the amplitude and time course of the Ca2<sup>+</sup> signal "seen" by Syt. The latter depends on different factors, including the number and types of Cavs, their diffusional distance to Syt, and the characteristics of other Ca2+-binding proteins present in the terminal. Ca2<sup>+</sup> entering the presynaptic terminal builds a steep, short-lasting concentration gradient around the mouth of the open Cavs that rapidly diminishes with increasing distance from the channel. Due to the steepness and short duration of this Ca2<sup>+</sup> gradient a chemical equilibrium is never established in this process. This makes the intracellular Ca2+-binding kinetics of the release sensor, rather than its affinity alone, as well as its diffusional distance to the Cavs crucial to the control of speed and reliability of transmitter release (Bollmann et al., 2000; Schneggenburger and Neher, 2000; Eggermann and Jonas, 2012). In this review we will focus on these two prominent factors in the regulation of pr , the synaptic Ca2+-binding kinetics of Syt proteins and their topographical relationships to Cavs. We will put an emphasis to more recent findings at small synapses in cortical structures of the mammalian brain.

#### PROPERTIES OF RELEASE SENSORS FOR SYNCHRONOUS RELEASE

Synaptotagmin–1,−2, and −9 (Syt1, 2, 9) are the known Ca2<sup>+</sup> sensors for fast, synchronous transmitter release in the millisecond time window following an AP (Südhof, 2014). Syt1 and Syt2 are the dominating Syt isoforms for synchronous release in the mammalian brain while Syt9 expression appears to be restricted to the limbic system and the striatum (Berton et al., 1997; Fox and Sanes, 2007; Xu et al., 2007). In addition to fast synchronous release, a second, slow and asynchronous component of transmitter release has been described (Geppert et al., 1994; Goda and Stevens, 1994). Asynchronous release is primarily activated during and following repetitive stimulation and operates via sensors different from those for synchronous release (Sun et al., 2007; Kochubey et al., 2011). Due to their dominating role for rapid neuronal communication, we will focus here on Syt1 and Syt2 triggered release processes. Molecular and structural aspects of Syt1, 2 proteins and their interactions with SNARE- and scaffold proteins were covered by several comprehensive recent reviews (Südhof and Rothman, 2009; Südhof, 2012, 2013; Kaeser and Regehr, 2014; Brunger et al., 2018; Park and Ryu, 2018).

Briefly, a synaptic vesicle bears approximately 15 copies of Syt on its surface (Takamori et al., 2006). Each Syt has two C2 domains that constitute Ca2+-binding and in addition might mediate protein-protein interactions with SNAREs and other Syt proteins or interactions with the membrane. One of the C2 domains is a C2A domain that binds three Ca2<sup>+</sup> ions, while the other one is a C2B domain that binds two Ca2<sup>+</sup> ions (Südhof, 2013). Upon Ca2<sup>+</sup> binding Syt triggers rapid synchronous vesicle fusion but the detailed molecular mechanisms are complex and still controversial (Brunger et al., 2018; Park and Ryu, 2018). Some of the proposed models discuss the role of Syt in at least two processes: First, prior to Ca2<sup>+</sup> influx spontaneous fusion of synaptic vesicles has to be prevented by inhibiting the constititively active SNARE complex from full zippering (SNARE clamping). Second, upon Ca2<sup>+</sup> influx fusion is triggered by relieving SNAREs from the clamp (SNARE unclamping). Among the proposed models it is under debate if SNARE clamping is mediated directly by Syt or if and in as much it involves a second protein called Complexin (Cpx), which is discussed to also have a SNARE clamping function (Südhof, 2013; Trimbuch and Rosenmund, 2016), and can form a protein complex with SNAREs and Syt (Zhou et al., 2017). Hence, according to these models, Syt either has a dual function by first clamping SNARE

corresponding [Ca2+]

i

reaction between vesicular sensor (V) and Ca2+.

zippering and an uncalmping function by relieving the clamp upon Ca <sup>2</sup><sup>+</sup> binding or only by relieving a Cpx-mediated SNARE clamp upon Ca2<sup>+</sup> binding. Other models emphasize the membrane binding properties of Syt and suggest that membrane insertion of Ca2+-bound Syt could cross-link vesicle and plasma membrane or lower the energy barrier for fusion by either regulating the vesicle to plasma membrane distance or by locally curving the plasma membrane. In addition, there is evidence that Syt also directly binds to Ca2<sup>+</sup> channels (Sheng et al., 1997). Since a detailed discussion of the molecular mechanisms of the fusion process is beyond the scope of this review, we refer the reader to most recent reviews (Trimbuch and Rosenmund, 2016; Brunger et al., 2018; Park and Ryu, 2018). We will focus here on the kinetic aspects of the interaction between Ca2<sup>+</sup> ions and Syt1, 2.

# Synaptic Ca2+-Binding Kinetics of Synaptotagmins

It has been known for half a century that transmitter release has a non-linear, approximately power of 4 dependency on [Ca2+]<sup>e</sup> (Dodge and Rahamimoff, 1967). However, a quantification of the intracellular presynaptic Ca2+-binding kinetics of a CNS release process became available only more recently (Bollmann et al., 2000; Schneggenburger and Neher, 2000), resulting in a detailed kinetic model of Ca2+-binding and release for the young calyx of Held synapse in the auditory brainstem (**Figure 1**), which expresses the Syt2 isoform as prime release sensor (Kochubey et al., 2016). The model was established based on an elegant combination of presynaptic Ca2<sup>+</sup> uncaging and Ca2<sup>+</sup> imaging with pre- and postsynaptic patch-clamp recordings (**Box 1**). The established model covers five cooperative, lowaffinity Ca2+-binding sites with fast kinetic rate constants for Ca2+-binding and -unbinding (kon ∼10<sup>8</sup> M−<sup>1</sup> s −1 , koff ∼5000 s−<sup>1</sup> , respectively) and accounted for the experimental, cooperative power of 4 dependency of the release rate onto the intracellular Ca2<sup>+</sup> concentration ([Ca2+]i) as well as brief synaptic delays (**Figure 1**, **Table 1**). It should be noted that this model does not reflect the Ca2+-binding kinetics of Syt2 alone but rather the kinetics of Syt2 embedded in its natural synaptic environment. For simplicity we will refer to it as the Syt2 model.

In following work the Syt2 model has been extended (**Scheme 1**) to also account for release at low [Ca2+]<sup>i</sup> (Lou et al., 2005), for phasic and tonic release (Millar et al., 2005; Pan and Zucker, 2009), for asynchronous release (Sun et al., 2007), and to address mechanisms of synaptic plasticity (Felmy et al., 2003; Sakaba, 2008; Pan and Zucker, 2009; Bornschein et al., 2013; Brachtendorf et al., 2015). In addition, it has been shown that the intracellular Ca2<sup>+</sup> sensitivity of Syt2-driven release is slightly reduced between postnatal day (P) 8-9 and P12-15 at the calyx of Held (Wang et al., 2008; Kochubey et al., 2009). Currently, the established Syt2 models are widely used in

BOX 1 | Quantication of the Ca2+-binding kinetics of Syt in presynaptic terminals via Ca2<sup>+</sup> uncaging.

. Insets show the kinetic binding schemes for the

Syt is integrated in the supra-molecular protein complex of the release machinery, which will influence its Ca2+-binding kinetics in a non-predictable manner similar to other Ca2<sup>+</sup> sensor proteins (Xia and Storm, 2005). Consequently, Syt2 has been analyzed in Syt2 expressing synapses (cf. above). Toward this end, it was required to first define the local [Ca2+] i at the release sensor and second, to quantify corresponding release rates.

At present it is difficult or even impossible to directly quantify the local [Ca2+] <sup>i</sup> at the release sensor. Even if it were possible to measure [Ca2+] i at areas as small as ∼0.5 µm<sup>2</sup> as performed at squid giant synapses (Llinás et al., 1992), the local [Ca2+] i at the sensor would remain rather ill-defined due to the steep spatial gradient of synaptic [Ca2+] <sup>i</sup> elevations, the unknown distance to the sensor and uncertainties about endogenous Ca2<sup>+</sup> buffers (Neher, 1998a; Bucurenciu et al., 2008, 2010; Bornschein et al., 2013; Schmidt et al., 2013).

Ca2<sup>+</sup> uncaging has been shown to be a useful method to resolve this problem (Heidelberger et al., 1994). Ca2<sup>+</sup> uncaging elevates [Ca2+] i uniformly in a presynaptic terminal. Due to this uniform [Ca2+] <sup>i</sup> elevation throughout the terminal, local [Ca2+] <sup>i</sup> will be identical to global [Ca2+] i , which in turn is quantified by concomitant Ca2<sup>+</sup> imaging. Uniform elevations of synaptic [Ca2+] i to different levels by flash photolysis of caged Ca2<sup>+</sup> have been employed for establishing the relationship between [Ca2+] i and release and permitted the construction of the above described Syt2-based release models. This method was applied at the giant calyx of Held (Bollmann et al., 2000; Schneggenburger and Neher, 2000; Lou et al., 2005; Sun et al., 2007; Kochubey and Schneggenburger, 2011), which permits direct whole-cell patchclamp equilibration with caged Ca2<sup>+</sup> compounds and Ca2<sup>+</sup> indicator dyes, Ca2<sup>+</sup> uncaging at the presynaptic terminal and concomitant patch-clamp recordings from the postsynaptic site. Thus, differences in PSC amplitudes and synaptic delays recorded at the postsynaptic site can be directly correlated to differences in [Ca2+] i at the presynaptic release sensor. Finally, recording of quantal PSCs ("minis") allows for calculating the release rates by deconvolution analysis (Van der Kloot, 1988; Diamond and Jahr, 1995; Bollmann et al., 2000; Schneggenburger and Neher, 2000; Neher and Sakaba, 2001). Deconvolution decomposes the PSC into the times of release of individual quanta, thereby, giving the release rate in quanta/s during the PSC. The recorded mini serves as elementary quantal event for the deconvolution. Deconvolution assumes that there are no quantal interactions at the synapse, i.e., the PSC arises from linearly summing minis.

Bornschein and Schmidt Synaptotagmins and Coupling


TABLE 1 | Parameters of release sensors.

quantitative descriptions of transmitter release (Eggermann et al., 2012; Stanley, 2016).

For mammalian CNS synapses, the Syt2-based models were originally constructed at the young (1–2 weeks old) calyx of Held but subsequently also at a small CNS synapse, the immature cerebellar basket cell to Purkinje cell (PC) synapse (Sakaba, 2008), at which Syt2 also represents the dominant Syt isoform (Chen et al., 2017). Notably, Syt2 is the dominating fast release sensor in hindbrain structures, while in most forebrain structures, including the neocortex, Syt1 is the sensor mediating fast synchronous release (Berton et al., 1997; Fox and Sanes, 2007; Xu et al., 2007). Importantly, a detailed kinetic model similar to that of Syt2 is at present not available for Ca2+-binding to Syt1 in mammalian CNS synapses. At very young, P5-6 pre-calyx synapses in the brainstem a fast release component has been reported to be mediated via Syt1 but no kinetic model has been constructed (Kochubey et al., 2016). This Syt1 triggered release process had a less than power of 2 dependency on [Ca2+]<sup>i</sup> , i.e., its [Ca2+]<sup>i</sup> dependency was substantially shallower than that of Syt2 triggered release at the young calyx of Held. A kinetic model for Syt1-mediated release has been constructed for fusion of dense core vesicles at chromaffin cells of the adrenal gland (Voets, 2000; Sørensen et al., 2003). In this model three to four Ca2+-binding sites with rate constants of about two orders of magnitude smaller than those for the synaptic Syt2 model were found to be suitable to describe secretion from chromaffin cells, which is much slower than rapid synaptic release (**Table 1**). Consequently, the resulting dependency of the release rate onto [Ca2+]<sup>i</sup> was again much shallower than for synaptic Syt2 (**Figure 1**). Also, a quantitative comparison of the dependency on [Ca2+]<sup>e</sup> of release processes mediated by Sy1 and Syt2 in cultured neurons revealed differences between the two proteins. Finally, differences were found for the kinetics of Syt1 and Syt2 mediated postsynaptic currents (PSCs), indicating differences in the kinetics of Syt1 vs. Syt2 triggered release. Consequently, it has been suggested that the expression of a particular Syt isoform dictates the properties of release at its synapse (Xu et al., 2007). Thus, it will be interesting to see, whether Syt1-triggered release at mature synapses in the mammalian brain indeed has a dependency on [Ca2+]<sup>i</sup> different from Syt2-triggered release.

# Kinetics of Ca2+-Unbinding From Syt, Active Ca2+, and Facilitation Sensors

Paired pulse facilitation (PPF) is a form of short-term synaptic plasticity important for synaptic computation (Abbott and Regehr, 2004). PPF is a use-dependent enhancement of transmitter release following the second of two successive APs separated by a millisecond time interval. Although PPF was discovered more than 70 years ago, its mechanisms remain controversial and may differ between synapses (Zucker and Regehr, 2002). Originally, it has been suggested that "Ca2<sup>+</sup> remaining attached to specific sites on the inner axon membrane" causes facilitation. For this Ca2<sup>+</sup> the term "active Ca2+" was coined (Katz and Miledi, 1968). In a simpler form of the "residual Ca2<sup>+</sup> hypothesis" a residue of free Ca2<sup>+</sup> ([Ca2+]res) from the first AP summates with free Ca2<sup>+</sup> ([Ca2+]i) from the second AP, thereby, causing amplified release. However, it has been recognized early that the decay of [Ca2+]res deviates from the time course of facilitation, such that [Ca2+]res cannot fully account for facilitation (Blundon et al., 1993). Additionally, due to the large amplitude difference between [Ca2+]res (∼100 nM) and nano- or microdomain [Ca2+]<sup>i</sup> at the release site during the second AP (∼20–100µM) simple Ca2<sup>+</sup> summation is unlikely to be the exclusive source of facilitation (Zucker and Regehr, 2002). Consequently, at different synapses different conceptions were developed to account for facilitation. These include slow Ca2<sup>+</sup> relaxation of the bound sensor (Yamada and Zucker, 1992; Bertram et al., 1996; Matveev et al., 2002), separate sites for release and facilitation (Atluri and Regehr, 1996), elevated release site [Ca2+]<sup>i</sup> during the second pulse (Geiger and Jonas, 2000; Felmy et al., 2003; Bollmann and Sakmann, 2005), buffer effects (Neher, 1998a; Rozov et al., 2001), or activity dependent recruitment of additional release sites (Valera et al., 2012; Miki et al., 2016; Doussau et al., 2017). For a recent comprehensive review on mechanisms of PPF (see Jackman and Regehr, 2017).

Here, we focus on Syt-related mechanisms of PPF. Ca2+ unbinding from the release sensor has been suggested as one mechanism of PPF (Yamada and Zucker, 1992; Bertram et al., 1996; Matveev et al., 2002). Young cerebellar PCs are connected among each other via recurrent GABAergic synapses that show PPF during high-frequency activation. Although PCs strongly express the "slow" and "fast" native Ca2<sup>+</sup> buffers Parvalbumin (PV) and Calbindin-D28k (CB), respectively, PPF was not affected by loss of either of the buffers (Bornschein et al., 2013). Rather the results indicated that a residue of Ca2<sup>+</sup> remaining bound to the release sensor Syt2 (Schneggenburger and Neher, 2000; Sakaba, 2008) after the first AP is the probable main cause of PPF at PC to PC synapses, a mechanism highly reminiscent of the original "active Ca2+" mechanism (Katz and Miledi, 1968).

Another suggested mechanism reminding on the original "active Ca2+" mechanism was that a facilitation sensor separate from the release sensor could exist (Atluri and Regehr, 1996). The molecular identity of the facilitation sensor, however, remained elusive until recently Syt7 has been identified to operate as a facilitation sensor (Jackman et al., 2016). Syt7 is abundantly found in presynaptic plasma membranes (Li et al., 2017), while Syt1 and Syt2 rather locate to SV membranes. The intrinsic Ca2<sup>+</sup> affinities of Syt1 and Syt7 are comparably low in solution (K<sup>D</sup> ≥ 100µM; Radhakrishnan et al., 2009; Voleti et al., 2017). In the presence of lipids the apparent Ca2<sup>+</sup> affinity of both proteins increases, albeit for Syt7 stronger than for Syt1, such that the apparent Ca2<sup>+</sup> affinity of Syt7 is ∼10fold higher than that of Syt1 (Sugita et al., 2002). The apparent Ca2+-sensing properties of Syt1 and Syt7 correlate with their specific functions, such that Syt1 is activated only by high Ca2<sup>+</sup> concentrations (∼10– 100µM) typical for AP-evoked [Ca2+]<sup>i</sup> elevations in the vicinity of Ca2<sup>+</sup> channels, while Syt7 can also operate during longer lasting increases in residual Ca2<sup>+</sup> in the low micromolar range (Volynski and Krishnakumar, 2018). These characteristics made Syt7 a promising candidate for the proposed facilitation sensor. Indeed, it was found that Syt7 contributes to PPF at different types of facilitating synapses in the hippocampus and at corticothalamic synapses. Mechanistically, Ca2+-binding to the C2A domain of Syt7 was required for facilitation (Jackman et al., 2016; Jackman and Regehr, 2017; Turecek et al., 2017).

For Syt7 at least two other functions were reported (Volynski and Krishnakumar, 2018): It was found to act as a Ca2<sup>+</sup> sensor for SV replenishment (Liu et al., 2014) and to mediate slow, asynchronous transmitter release (Bacaj et al., 2013; Luo and Südhof, 2017). Interestingly, the different proposed functions of Syt7 need not be mutually exclusive (Chen and Jonas, 2017). Consistently, it was found at cerebellar PF to PC synapses that Syt7 is involved in mediating both, PPF and asynchronous release (Turecek and Regehr, 2018). PPF at PF synapses has further been reported to depend on rapid SV replenishment with recruitment of additional release sites that resulted in an activity dependent, transient increase in the RRP (Valera et al., 2012; Brachtendorf et al., 2015; Miki et al., 2016; Doussau et al., 2017). It is tempting to speculate that the "overfilling" of an RRP by additional release sites could involve Syt7.

### SPATIAL COUPLING BETWEEN SYNAPTOTAGMIN AND CAVS

Besides the Ca2+-binding kinetics of Syt, its spatial relationship to the presynaptic Cavs is crucial for setting fundamental synaptic properties, including p<sup>r</sup> , synchronicity of release and synaptic delays (Bucurenciu et al., 2008). The distance between Syt and the Cavs is frequently referred to as the coupling distance. In general it may be asserted that for AP evoked release a tight coupling favors high p<sup>r</sup> (Bucurenciu et al., 2008; Baur et al., 2015; Kusch et al., 2018), short synaptic delays (Bucurenciu et al., 2008), energy efficacy (Eggermann et al., 2012; Lu et al., 2016) and renders the release process less modifiable by Ca2<sup>+</sup> buffers (Adler et al., 1991; Eggermann and Jonas, 2012; Bornschein et al., 2013; Schmidt et al., 2013; Brachtendorf et al., 2015). Loose coupling, on the other hand, offers more options for regulation and plasticity (Nadkarni et al., 2012; Vyleta and Jonas, 2014). There has been a comprehensive review on influx-release coupling at mammalian synapses of the peripheral NS (PNS) and the CNS (Eggermann et al., 2012). However, since then, a large body of work at AZs focused directly or indirectly on coupling distances and greatly advanced our knowledge about coupling and its regulation at different synapses. Some of these insights stem from classical model synapses, like the calyx of Held in the auditory brainstem, the frog neuromuscular junction, the squid giant synapse, or chick ciliary ganglion cells, which offer favorable conditions for electrophysiological analysis, in particular due to their large size. Insights from these synapses were covered by two recent comprehensive reviews (Wang and Augustine, 2015; Stanley, 2016) to which we refer the reader here. We will review advances in understanding AP-mediated Ca2<sup>+</sup> influx-evoked transmitter release coupling at mammalian cortical AZs as well as their regulation during postnatal development and emerging roles of specific proteins in this regulatory process.

### Coupling Topographies

We will start this chapter with a brief note on nomenclature. The border between "tight" and "loose" coupling is not clearly defined (**Box 2**). A border line in the range of 70–100 nm has been suggested previously to distinguish between the two coupling regimes (Eggermann et al., 2012; Vyleta and Jonas, 2014). In light of the most recent quantitative estimates of coupling distances and domain topographies at mammalian CNS synapses (**Table 2**), we suggest an even narrower line of demarcation of no larger than 50 nm. At this coupling distance a given open Ca<sup>v</sup> will be essentially ineffective in triggering release of a SV (**Figure 3**). Throughout this review we will use "loose coupling" if the coupling distance is ≥ 50 nm and "tight coupling" otherwise. In addition, we will use "single domain topography" (SDT) if only a single open Ca<sup>v</sup> triggers release, and "domain overlap topography" (DOT) if a cluster of open Cavs with overlapping Ca2<sup>+</sup> signaling domains controls release. Finally, we will use "nanodomain" as synonym for tight coupling plus SDT and "microdomain" as synonym for loose coupling plus DOT (Fedchyshyn and Wang, 2005; **Table 3**; **Box 3**).

Tight influx-release coupling has been reported for giant synapses specialized for escape reflexes in the squid (Adler et al., 1991), bipolar cells in the goldfish retina (von Gersdorff and Matthews, 1994; Burrone et al., 2002) and for the frog neuromuscular junction (Harlow et al., 2001). The first descriptions of nanodomain coupling came from the squid giant synapse and chick ciliary ganglia cells (Adler et al., 1991; Stanley, 1993). In the mammalian CNS, inhibitory synapses in the hippocampus and cerebellum were found to operate at tight coupling and at least in part with SDT, i.e., with nanodomains (Bucurenciu et al., 2008, 2010; Eggermann and Jonas, 2012; Bornschein et al., 2013). Surprisingly, cortical glutamatergic synapses seemed to forgo the benefits of tight coupling. Experimental studies performed on young pyramidal neurons (PNs; Ohana and Sakmann, 1998; Rozov et al., 2001) and in hippocampal cell cultures (Ermolyuk et al., 2013) as well as computational models of hippocampal CA3 – CA1 PN synapses (Nadkarni et al., 2012) showed loose coupling and established the view that small glutamatergic synapses in the brain, in particular excitatory cortical synapses, operate at microdomain coupling (Eggermann et al., 2012).

#### BOX 2 | Deriving quantitative estimates of coupling distances.

We are not aware of any report of a direct quantification of the coupling distance between Syt and Cavs at AZs by microscopic techniques. In particular this appears to be due to the non-availability of appropriately sized labels. Hence, information about the average coupling distance is classically obtained by dialyzing a presynaptic terminus with exogenous Ca2<sup>+</sup> chelators of similar affinity (KD) but different Ca2+-binding kinetics, i.e., different on-rates (kon; Adler et al., 1991; Augustine et al., 1991; Neher, 1998b; Eggermann et al., 2012). Typically the Ca2<sup>+</sup> chelators EGTA (ethylene glycol-bis(2-aminoethylether)-N,N,N',N'-tetraacetic acid; <sup>K</sup><sup>D</sup> <sup>=</sup> 70 nM, <sup>k</sup>on <sup>=</sup> <sup>10</sup><sup>7</sup> <sup>M</sup>−<sup>1</sup> s −1 ; Nägerl et al., 2000) and BAPTA (1,2-bis(2-aminophenoxy)ethane-N,N,N',N'-tetraacetic acid; <sup>K</sup><sup>D</sup> <sup>=</sup> <sup>220</sup> nM, <sup>k</sup>on <sup>=</sup> <sup>4</sup>∗10<sup>8</sup> <sup>M</sup>−<sup>1</sup> s −1 ; Naraghi and Neher, 1997) are used for this approach since they have similar <sup>K</sup><sup>D</sup> values but BAPTA is <sup>∼</sup>40 times faster than EGTA. Ca2<sup>+</sup> chelators suppress synaptic transmission by reducing the amount of Ca2<sup>+</sup> that binds to Syt (Figure 2). The exact amount of interference depends on four factors: the average coupling distance, the mobility of the chelator, its <sup>k</sup>on, and its concentration. If influx-release coupling is tight, only a chelator with a rapid <sup>k</sup>on like BAPTA is able to capture Ca2<sup>+</sup> in the nanodomain in the immediate vicinity of the channel before it reaches Syt, while chelators with slow <sup>k</sup>on like EGTA fail to influence the nanodomain Ca2+. Thus, at moderate concentrations only the fast BAPTA will reduce the amount of transmitter released in a tight coupling regime. In a loose coupling regime, on the other hand, both, BAPTA and EGTA will interfere with transmitter release since Ca2<sup>+</sup> has to diffuse a larger distance from the Cavs to reach the sensor. This allows also the slower EGTA to capture Ca2<sup>+</sup> in the microdomain before the ions reach the release sensor. Using this exogenous chelator dialysis approach, most of the estimates of coupling distances reviewed here were derived. It should be noted that the degree of interference actually not only depends on the kon but also on the concentration of the buffer, i.e., a large concentration of EGTA interferes with release similar to a much smaller concentration of BAPTA (Figure 3). In order to obtain quantitative values of the coupling distance, additional information about the magnitude and duration of the Ca2<sup>+</sup> influx and potential Ca2<sup>+</sup> sensor saturation is required (e.g., Bucurenciu et al., 2008; Schmidt et al., 2013; Nakamura et al., 2015, 2018; Kusch et al., 2018). Finally, by combining all results in experimentally constrained computer simulations quantitative estimates of the average coupling distance can be obtained (Bucurenciu et al., 2008; Bornschein et al., 2013; Schmidt et al., 2013; Vyleta and Jonas, 2014; Nakamura et al., 2015; Kusch et al., 2018).

The exogenous chelator dialysis approach was applied to large synapses that can be directly infused with chelator containing solution (Adler et al., 1991; Borst and Sakmann, 1996) and to large neurons that permit dialyzing the distant presynaptic sites by prolonged somatic whole-cell patch-clamp recordings (Ohana and Sakmann, 1998; Bucurenciu et al., 2008; Bornschein et al., 2013). The advantage of this approach is that the intracellular concentrations of the Ca2<sup>+</sup> chelators are well-defined.

Another way of loading neurons with exogenous chelators is by application of membrane permeant acetoxymethyl ester variants of the Ca2<sup>+</sup> chelators (EGTA-AM or BAPTA-AM) to the extracellular bath solution (Atluri and Regehr, 1996; Matsui and Jahr, 2003; Hefft and Jonas, 2005). The AM-chelator compound passes the lipophilic plasma membrane and enters the presynaptic cytosol. There, the ester group is cleaved by enzymes, which makes the chelator membrane-impermeable. Depending on loading time, its intracellular concentration can substantially exceed its bath concentration due to continuous intracellular accumulation of the chelator as long as its AM-form is present in the bath. The advantage of this approach is its relative experimental ease and that it is well-tolerated also by small neurons. It has the disadvantage that the intracellular chelator concentration remains rather ill defined. Thus, it permits a rapid initial assessment of relative differences in coupling e.g., between age groups, if differently aged synapses are compared under otherwise identical experimental conditions (Matsui and Jahr, 2003; Hefft and Jonas, 2005; Baur et al., 2015).

Neurons express endogenous Ca2<sup>+</sup> buffers with quantified Ca2+-binding kinetics (Lee et al., 2000; Faas et al., 2007). Knowledge about the expression of specific native Ca2<sup>+</sup> buffers and there Ca2+-binding kinetics offers an alternative route to deriving quantitative estimates of coupling distances by comparing transmitter-release from wild-type terminals to release from mutant terminals lacking a specific native buffer (Bornschein et al., 2013; Schmidt et al., 2013).

Initial experimental evidence against the generality of this view was available from the CA3 – CA1 PN synapse (Scimemi and Diamond, 2012), showing SDT and results that were more consistent with a tighter coupling at this synapse rather than with DOT and a very large number of Cavs loosely coupled to the release sensor as suggested by the above mentioned study of Nadkarni et al. (2012). In the cerebellar cortex of 3 to 4 weeks old mice, subsequent work quantified the coupling distance at the parallel fiber (PF) to PC synapse, an excitatory, glutamatergic synapse in the cerebellar cortex and probably the most abundant synapse in the mammalian brain. It was found that this synapse operates at tight coupling of ∼24 nm (Schmidt et al., 2013). In successional work it was found that at this age coupling is not only tight but that these synapses operate with a nanodomain topography (Baur et al., 2015; Kusch et al., 2018) and that also further excitatory synapses in the cerebellar cortex operate with tight coupling (Ritzau-Jost et al., 2014; Delvendahl et al., 2015). Together these studies clearly contradicted the generality of microdomain coupling at excitatory synapses in mammalian cortical structures.

#### Regulation of Coupling

An interesting difference between the experiments suggesting microdomain coupling in glutamatergic cortical synapses (Ohana and Sakmann, 1998; Rozov et al., 2001) and the experiments showing nanodomain coupling (Schmidt et al., 2013; Ritzau-Jost et al., 2014) was the age of the experimental animals: While the former studies were performed in young rats (∼2 weeks old), in the latter studies, the coupling distance was assessed in more matured mice (>3 weeks old). Considering that the postnatal development of rats likely proceeds slower than that of mice, the age difference most likely corresponds to an even larger difference in brain maturation. This raised the possibility that the coupling distance could be regulated developmentally.

Support for the idea of a developmental regulation of coupling came from experiments performed at the calyx of Held. Experiments performed on young (∼10 days old) and matured (∼3 weeks old) calyces indicated a substantial developmental tightening of the coupling distance during postnatal development (Taschenberger et al., 2002; Fedchyshyn and Wang, 2005; Wang et al., 2008; Kochubey et al., 2009). Simulations quantified that the experimental results are explained by a developmental tightening of the spatial coupling between Ca2<sup>+</sup> channels and Syt from ∼60 to ∼20 nm at the calyx of Held (Wang et al., 2009).

Coupling distances and domain topographies were quantified more recently in a developmental context (**Table 2**), again at the calyx of Held (Nakamura et al., 2015) and at the PF to PC synapse (Baur et al., 2015). At the calyx of Held, a moderate developmental tightening of the coupling distance between Syt and the closest Ca<sup>v</sup> of a cluster from ∼30 nm to ∼20 nm between P7 and P14 was found, while the number of Cavs within a cluster controlling a given release site remained relatively constant with an average in the range of 25 to 30 (Nakamura et al., 2015). Thus, although a developmental shortening of the coupling distance was found at the calyx of Held, it operated at fairly tight coupling with DOT, independent of age in the range of P7 to P14. By contrast, at the PF to PC synapse a switch from DOT to SDT was found. At ∼P9, PF terminals operated with a DOT with a distance of ∼60 nm between the closest Ca<sup>v</sup> within a cluster and Syt, while at ∼P23 a coupling distance of ∼20 nm and SDT were

FIGURE 2 | Ca2<sup>+</sup> signaling domains. (A) General Ca2<sup>+</sup> dynamics: Ca2<sup>+</sup> enters a presynaptic terminal through a voltage-gated Ca2<sup>+</sup> channel. Due to rapid diffusion (indicated by red gradient and the arrows) Ca2<sup>+</sup> forms a steep, short-lived spatio-temporal gradient around the mouth of the open channel. It binds to mobile or fix Ca2+-binding proteins (CaBPs); some CaBP are pure buffers, others have an additional Ca2<sup>+</sup> sensor function. Ultimately Ca2<sup>+</sup> is cleared from the cytosol via Ca2+-ATPases (white circles with arrows) that either pump Ca2<sup>+</sup> into the extracellular space or sequester it into organelles. (B) In a tight coupling regime a Syt bearing SV is located very close to the site of Ca2<sup>+</sup> entry (<50 nm). If coupling is tight, at moderate concentrations only a buffer with rapid Ca2<sup>+</sup> binding kinetics like BAPTA (red) can interfere with Ca2<sup>+</sup> binding to Syt and prevent release. In a loose coupling regime, on the other hand, the SV is further away from the site of Ca2<sup>+</sup> entry and also a slow buffer like EGTA (blue) can bind Ca2<sup>+</sup> before it reaches the release sensor.

FIGURE 3 | (A) Simulated [Ca2+] i transients at increasing distances between 5 and 50 nm (5 nm increments) from a single Cav2.1 channel (inset; Li et al., 2007) opening during an AP in the absence of Ca2<sup>+</sup> buffers. (B) Release rates were simulated at increasing distances from the Cav2.1 channel (1 nm increments) using the Syt2 sensor model from Figure 1. Release probabilities were calculated by integrating the release rates over time and plotted against the corresponding sensor-to-Cav coupling distances. Note the steep decline in pr between 6 and 20 nm coupling distance and that pr at 50 nm is almost 0. (C) Simulated relative reduction of pr for different concentrations of EGTA (blue) and BAPTA (red). Moderate concentrations of EGTA are not very effective in blocking release close to a channel, while moderate concentrations of BAPTA are highly effective already at coupling distances of 10–20 nm. Higher concentration of EGTA mimic the effects of lower concentrations of BAPTA (concentrations are indicated). Note that in native boutons the concentrations of EGTA and BAPTA that yield corresponding effects on pr will be different due to the presence of native Ca2+-binding proteins, which were not included in the simulations.

TABLE 2 | Quantitative estimates of coupling topographies at mammalian CNS synapses.


\*Coupling distances need not be homogeneous (Scimemi and Diamond, 2012; Ermolyuk et al., 2013; Ritzau-Jost et al., 2018).

BC, basket cell; CH, calyx of Held; GC, granule cell; MF, mossy fiber; P, postnatal day; PN, pyramidal neuron; PC, Purkinje cell; PF, parallel fiber; n.d., not determined.

#### BOX 3 | Estimating the functional domain topography.

How many Cavs open during an AP and do their Ca2<sup>+</sup> signaling domains overlap? Immunolabelling techniques combined with electron microscopy provide highly valuable insights into the structural organization of Cav clusters and Cav subtypes at the AZ. Yet, in order to answer the above question for functional SDT or DOT they need to be combined with physiological studies at the synapse (Holderith et al., 2012; Baur et al., 2015; Nakamura et al., 2015; Kusch et al., 2018). Specifically, the use of the unspecific Ca<sup>v</sup> blocker Cd2<sup>+</sup> and/or a combination of Ca<sup>v</sup> subtype specific blockers were shown to yield valuable insights into the functional domain topography (Table 3; Augustine et al., 1991; Mintz et al., 1995; Scimemi and Diamond, 2012).

At physiological temperature Cd2<sup>+</sup> dissociates slowly from a Ca<sup>v</sup> (Chow, 1991), thus, blocking a channel in an all-or-none fashion on the time-scale of an AP. The shape of Cd2<sup>+</sup> dose-effect curves onto EPSC amplitudes as read-out for release will depend on the domain topography. If a presynaptic terminal harbors release sites with DOT, the curve will be non-linear whereas it will be linear if the release sites operate with SDT (Augustine et al., 1991). The construction of full dose-effect curves may be circumvented by analyzing the effects of a subsaturating concentration of Cd2<sup>+</sup> onto the PPR (Scimemi and Diamond, 2012).

Application of a subsaturating concentration of Cd2<sup>+</sup> reduces the amplitude of the first postsynaptic current (PSC) of a paired pulse experiment irrespective of the domain topography. However, its impact on the PPR markedly depends on whether the release sites operate with SDT or DOT. At a subsaturating concentration of Cd2<sup>+</sup> some but not all Cavs will be blocked during an AP. In a DOT, blocking some of the Cavs controlling a synaptic vesicle will have effects similar to reducing [Ca2+]e, i.e., the initial <sup>p</sup><sup>r</sup> will be reduced while the PPR will increase. On the other hand, if a release site is controlled by a single Ca<sup>v</sup> (SDT), release of synaptic vesicles encountering no Ca2<sup>+</sup> would be blocked while release or facilitation of vesicles encountering Ca2<sup>+</sup> would be the same as in the absence of Cd2+. In consequence, application of Cd2<sup>+</sup> will increase PPR in a DOT but leave it unaltered in a SDT (Scimemi and Diamond, 2012).

These results can further be substantiated by using Cav subtype specific blockers, if more than one channel subtype contributes to release. For a SDT in which a given vesicle is linked to either of the Cav subtypes, the sum of the toxin-sensitive release fractions will not exceed release measured in the absence of toxin, i.e., the toxin sensitive release fractions will sum linearly. Contrariwise, for an AZ at which release of a vesicle is controlled in a DOT composed of different Cav subtypes, the sum of the blocked release fractions can exceed the control value ("supralinear" summation) because of the non-linear dependency of release on Ca2<sup>+</sup> (Mintz et al., 1995; Scimemi and Diamond, 2012).

found (Baur et al., 2015; Kusch et al., 2018). The DOT at young PF terminals was composed of Cav2.1 (P/Q-type) and Cav2.2 (Ntype) concomitantly controlling a release site, with likely 2 Cav2.1 and 1 Cav2.2 triggering release during an AP. The nanodomain at more matured PF terminals comprised only a Cav2.1 (Kusch et al., 2018).

These results may suggest that developmental tightening of the coupling distance is a common phenomenon in the mammalian brain, which could be accompanied by a switch from DOT to SDT in small, but not in large synapses. However, an alternative is that coupling distances, domain topographies and their regulation are synapse specific properties. At the glutamatergic MF to CA3 PN synapse a loose coupling distance of ∼75 nm has been quantified in the matured hippocampus (**Table 2**; Vyleta and Jonas, 2014). However, this finding does not necessarily exclude a developmental tightening of the coupling distance at the MF to CA3 synapse. It remains possible that the young synapse operates at an even larger coupling distance. Thus, while developmental tightening and loose coupling in mature brain are not mutually exclusive, the finding of loose

#### TABLE 3 | Active zone topographies.


pr, release probability of a vesicle; N, number of vesicles (or release sites that can release a max. of one vesicle), pr,avg, average release probabilities across vesicles; Psyn, synaptic release probability, Fsyn, synaptic failure rate; DE, EGTA dose effect curve; PRR, paired-pulse ratio; RR, release rate.

coupling at the mature MF – CA3 synapse clearly suggests that the coupling distance is a synapse specific property in the context of its developmental stage.

Are there other forms of regulation of the coupling distance besides developmental regulation? An intriguing possibility would be a regulation of the coupling distance depending on the activity of a synapse, i.e., as a mechanism of synaptic plasticity. Evidence for such use-dependent regulation of the coupling distance came from a recent study at hippocampal mossy fiber boutons (Midorikawa and Sakaba, 2017). It was found that increasing the level of cAMP in the boutons, which is a crucial step in the induction of long term potentiation, results in increased release from the bouton, while not increasing the number of synaptic vesicles in the RRP nor altering the Ca2<sup>+</sup> influx. Based on the differential action of EGTA prior and following the induction of cAMP-mediated plasticity the study provides evidence for a tightening of the coupling distance following cAMP application (Midorikawa and Sakaba, 2017).

#### Functional Considerations

The MF – CA3 PN synapse, which was found to operate at loose coupling in the matured hippocampus (Vyleta and Jonas, 2014), is highly plastic and expresses several forms of presynaptic plasticity (Salin et al., 1996). It has been suggested that loose coupling provides a molecular framework for high plasticity (Vyleta and Jonas, 2014). Consistent with this idea, synapses with tight coupling are mostly fast-signaling synapses in neuronal circuits specialized for high-frequency coding of sensory information or in motor control (**Table 2**). However, some of these synapses also show pronounced presynaptic plasticity. For example, the PF to PC synapse exhibits lowfrequency depression and high-frequency facilitation (Doussau et al., 2017).

Loose coupling offers more possibilities for regulating transmitter release and plasticity, e.g., via the action of Ca2<sup>+</sup> buffers, since in loose coupling also slow Ca2<sup>+</sup> buffers can intercept sizable amounts of Ca2<sup>+</sup> before it reaches Syt (Adler et al., 1991). In tight coupling regimes, on the other hand, only rapid buffers like BAPTA (Adler et al., 1991), Calretinin (Schmidt et al., 2013; Brachtendorf et al., 2015), or Calbindin (Bornschein et al., 2013) were found to be regulators of p<sup>r</sup> , while the "slow" buffer Parvalbumin (PV) did not affect p<sup>r</sup> (Bornschein et al., 2013). At high concentrations, however, even PV becomes effective in affecting [Ca2+]<sup>i</sup> and release in tight coupling regimes (Eggermann and Jonas, 2012). This is because PV actually is a rapid, high-affinity Ca2<sup>+</sup> buffer but its Ca2+-binding sites also have a medium affinity for Mg2<sup>+</sup> such that most binding sites are occupied by Mg2<sup>+</sup> under physiological resting conditions and only a small amount of binding sites (∼5%) are metal free (Lee et al., 2000). Thus, Ca2+-binding has to be preceded by Mg2+-unbinding, which proceeds with slow kinetics, i.e., the slow Mg2+-unbinding kinetics makes PV a slow Ca2<sup>+</sup> buffer (Lee et al., 2000). However, if PV is expressed strongly in a synapse the small relative fraction of Mg2+-free binding sites can constitute a sufficiently large absolute concentration of rapidly Ca2+-binding PV to significantly affect [Ca2+]<sup>i</sup> even in the nanodomain around a Ca<sup>v</sup> channel. Metal free binding sites are then continuously replenished efficiently from the large pool of Mg2+-bound sites (Eggermann and Jonas, 2012). It should be noted that this action of PV is different from the effects of large concentrations of EGTA in tight coupling regimes. PV was already effective at concentrations ∼500µM due to rapid Ca2+-binding and replenishment via Mg2+-unbinding, while slow buffering by EGTA requires concentrations >10 mM to intercept [Ca2+]<sup>i</sup> in the nanodomain.

Tight coupling increases speed and efficacy of synaptic transmission (Eggermann et al., 2012). In addition, it can provide an energy efficient design compared to loose coupling. To obtain a certain [Ca2+]<sup>i</sup> level at the release sensor less Cavs have to open in a tight than in a loose coupling regime (Eggermann et al., 2012). As the ATP cost of Ca2<sup>+</sup> removal is a significant post of the presynaptic energy consumption (Kim et al., 2005), tight coupling can save energy. This, however, requires that Ca2<sup>+</sup> influx would indeed be different between terminals with tight or loose coupling. Indeed results from the calyx of Held conform to this requirement, showing that concomitant with developmental coupling distance tightening the amplitudes of presynaptic Ca2<sup>+</sup> transients decreased (Nakamura et al., 2015). On the other hand, at the PF – PC synapse presynaptic Ca2<sup>+</sup> transients did not change developmentally despite the developmental switch from loose to tight coupling (Baur et al., 2015). Several Cavs opening during the presynaptic AP no longer contributed to driving release at later developmental stages (Kusch et al., 2018). Their primary function remains speculative but could be in Ca2+-driven replenishment of synaptic vesicles into the readily releasable pool (Brachtendorf et al., 2015; Miki et al., 2016; Doussau et al., 2017).

# Molecular Regulators of the Coupling Distance

Ca2<sup>+</sup> influx-transmitter release coupling is mediated via proteins of the AZ scaffold, albeit, this process is still not well-understood at present and a detailed review of the AZ scaffold is far beyond the scope of this paper. We briefly focus on some recent advances directly related to establishing influx—release coupling. RIMs (Rab3-interacting molecules) are known as central organizer of the AZ (Südhof, 2012). Specifically, they are required for recruiting Cav2.1 and Cav2.2 channels to the AZ (Kaeser et al., 2011), which can be considered as a first step in coupling Ca <sup>2</sup><sup>+</sup> influx to transmitter release, in particular since these channel subtypes are the most important ones for APmediated fusion (**Table 2**). The protein Septin5 was identified as a negative regulator of tight coupling during development, i.e., its downregulation was permissive to the establishment of tight coupling (Yang et al., 2010). Proteins involved in mediating tight coupling were identified more recently, suggesting RIM-BPs (RIM-binding proteins; Acuna et al., 2015; Grauel et al., 2016) and Munc13-3 (Ishiyama et al., 2014; Kusch et al., 2018), as positive regulators of the coupling distance. Thereby, Munc13- 3 was found to be a developmental mediator of tight coupling (Kusch et al., 2018). However, details of the interplay between identified regulators of the coupling distance, their relationships to other regulatory proteins at the AZ, and details of their interaction with the exocytotic core complex remain essentially unclear.

#### CONCLUDING REMARKS

More than 30 years after the steep non-linear dependency of transmitter release onto [Ca2+]<sup>e</sup> has been established (Dodge and Rahamimoff, 1967), detailed kinetic five-site models of the [Ca <sup>2</sup>+]<sup>i</sup> dependency of Syt2-triggered transmitter release were developed (Bollmann et al., 2000; Schneggenburger and Neher, 2000) and subsequently elaborated to cover sub-modes and subtleties of release (Lou et al., 2005; Sun et al., 2007; Pan and Zucker, 2009) and to capture developmental aspects (Kochubey et al., 2009). These models are widely applied in functional quantitative studies of transmitter release and AZ topography.

During the past decade several functional studies focused directly or indirectly on the coupling distance between Syts and Cavs at mammalian cortical synapses. Initially it was thought that only GABAergic synapses in cortical structures make use of tight coupling, while cortical glutamatergic synapses seemed to operate with loose coupling (Ohana and Sakmann, 1998; Rozov et al., 2001; Eggermann et al., 2012; Nadkarni et al., 2012; Stanley, 2016). However, results from a glutamatergic synapse in the mature cerebellar cortex falsified the generality of this hypothesis (Schmidt et al., 2013). From subsequent work (**Table 2**) it became evident that synapses in the matured mammalian brain, including synapses in cortical structures of hippocampus and cerebellum, indeed make widespread use of tight coupling and, furthermore, that release of a SV was frequently triggered by opening of only a few or even a single Cav. At synapses investigated in a developmental context, it was found that tight coupling at matured synapses develops from an initially loose coupling at younger synapses. This latter result provides an explanation why previous studies predominantly found microdomain coupling at glutamatergic cortical synapses. These earlier studies were performed at synapses of very young animals (Eggermann et al., 2012; Stanley, 2016). The concept of nanodomain coupling was developed 20 years ago at the squid giant synapse (Adler et al., 1991) and calyx-type synapses in the chick ciliary ganglion (Stanley, 1993) and now experiences a revival at matured mammalian central synapses.

However, tight coupling is not universal for synapses of the mature mammalian brain (Vyleta and Jonas, 2014). As suggested by Vyleta and Jonas, the present state of knowledge indicates that coupling distances are specific adaptations to the function of a synapse. GABAergic central synapses appear to operate at tight coupling, most probably irrespective of age (**Table 2**). For glutamatergic synapses the situation is more complex. While excitatory synapses specialized for rapid signaling develop a tight, nanodomain coupling topography, synapses highly adaptive via plasticity make use of loose coupling even in matured brain. To learn more about the rules that regulate coupling distances will require to investigate further types of synapses in different brain regions. For example, a particularly striking lack of quantitative data on coupling distances and AZ topographies exists for neocortical synapses (Eggermann et al., 2012; Stanley, 2016; **Table 2**). To our knowledge, a coupling distance has never been quantified at a neocortical synapse.

For understanding the rules regulating coupling, it will be also required to identify the proteins that link Syt bearing SVs to Cavs. Recent studies indicated RIM-BPs (Acuna et al., 2015; Grauel et al., 2016) and Munc13-3 (Kusch et al., 2018)

#### REFERENCES


to be involved in organizing Ca<sup>v</sup> clusters at the AZ and in narrowing the coupling distance. Munc13-3 was identified as a specific developmental mediator of nanodomain coupling at a glutamatergic synapse in cerebellar cortex (Kusch et al., 2018). Interestingly, Munc13-3 protein is expressed strongly in the cerebellar cortex, more weekly in the brainstem and is essentially absent from the hippocampus and cerebral cortex (Augustin et al., 1999). Does this indicate that developmental tightening of the coupling distance is a specificity of glutamatergic synapses in the cerebellum and brainstem? To answer this question, it will be required to quantify coupling distances in a developmental context also at neocortical synapses. Since coupling distances are key parameters of synaptic function, understanding the rules regulating this distance will advance our general understanding of the rules regulating synaptic transmission, which is the basic substrate of information flow in neuronal networks.

#### AUTHOR CONTRIBUTIONS

HS wrote the first draft of the manuscript. GB and HS prepared figures and tables. GB and HS contributed to manuscript revision, read and approved the submitted version.

#### FUNDING

This work was supported by the German Research Foundation (DFG SCHM1838/2) and DFG and University of Leipzig within the program of open access publishing.

#### ACKNOWLEDGMENTS

We thank Stefan Hallermann for critical discussion of the manuscript.


synapses is independent of calbindin and parvalbumin during high-frequency activation. J. Physiol. 591, 3355–3370. doi: 10.1113/jphysiol.2013.254128


active zones and the nanopositioning of Cav2.1 near release sensors. Cell Rep. 22, 1965–1973. doi: 10.1016/j.celrep.2018.02.010


**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 © 2019 Bornschein and Schmidt. 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) and the copyright owner(s) 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.

# Global Gene Knockout of Kcnip3 Enhances Pain Sensitivity and Exacerbates Negative Emotions in Rats

Yu-Peng Guo<sup>1</sup>† , Yu-Ru Zhi<sup>1</sup>† , Ting-Ting Liu<sup>1</sup> , Yun Wang1,2 and Ying Zhang<sup>1</sup> \*

<sup>1</sup> Department of Neurobiology, School of Basic Medical Sciences and Neuroscience Research Institute, Key Laboratory for Neuroscience, Ministry of Education and Ministry of National Health, Peking University, Beijing, China, <sup>2</sup> PKU-IDG/McGovern Institute for Brain Research, Peking University, Beijing, China

The Ca2+-binding protein Kv channel interacting protein 3 (KChIP3) or downstream regulatory element antagonist modulator (DREAM), a member of the neuronal calcium sensor (NCS) family, shows remarkable multifunctional properties. It acts as a transcriptional repressor in the nucleus and a modulator of ion channels or receptors, such as Kv4, NMDA receptors and TRPV1 channels on the cytomembrane. Previous studies of Kcnip3−/<sup>−</sup> mice have indicated that KChIP3 facilitates pain hypersensitivity by repressing Pdyn expression in the spinal cord. Conversely, studies from transgenic daDREAM (dominant active DREAM) mice indicated that KChIP3 contributes to analgesia by repressing Bdnf expression and attenuating the development of central sensitization. To further determine the role of KChIP3 in pain transmission and its possible involvement in emotional processing, we assessed the pain sensitivity and negative emotional behaviors of Kcnip3−/<sup>−</sup> rats. The knockout rats showed higher pain sensitivity compared to the wild-type rats both in the acute nociceptive pain model and in the late phase (i.e., 2, 4 and 6 days post complete Freund's adjuvant injection) of the chronic inflammatory pain model. Importantly, Kcnip3−/<sup>−</sup> rats displayed stronger aversion to the pain-associated compartment, higher anxiety level and aggravated depression-like behavior. Furthermore, RNA-Seq transcriptional profiling of the forebrain cortex were compared between wild-type and Kcnip3−/<sup>−</sup> rats. Among the 68 upregulated genes, 19 genes (including Nr4a2, Ret, Cplx3, Rgs9, and Itgad) are associated with neural development or synaptic transmission, particularly dopamine neurotransmission. Among the 79 downregulated genes, 16 genes (including Col3a1, Itm2a, Pcdhb3, Pcdhb22, Pcdhb20, Ddc, and Sncaip) are associated with neural development or dopaminergic transmission. Transcriptional upregulation of Nr4a2, Ret, Cplx3 and Rgs9, and downregulation of Col3a1, Itm2a, Pcdhb3 and Ddc, were validated by qPCR analysis. In summary, our studies showed that Kcnip3−/<sup>−</sup> rats displayed higher pain sensitivity and stronger negative emotions, suggesting an involvement of KChIP3 in negative emotions and possible role in central nociceptive processing.

Keywords: KChIP3, nociceptive pain, inflammatory pain, conditioned place aversion, anxiety, depression, negative emotions, RNA-Seq analysis

#### Edited by:

Karl-Wilhelm Koch, University of Oldenburg, Germany

#### Reviewed by:

Jose R. Naranjo, Spanish National Research Council (CSIC), Spain Tim Hucho, Universität zu Köln, Germany

\*Correspondence: Ying Zhang zhangyingnri@bjmu.edu.cn †These authors have contributed equally to this work

Received: 30 August 2018 Accepted: 09 January 2019 Published: 25 January 2019

#### Citation:

Guo Y-P, Zhi Y-R, Liu T-T, Wang Y and Zhang Y (2019) Global Gene Knockout of Kcnip3 Enhances Pain Sensitivity and Exacerbates Negative Emotions in Rats. Front. Mol. Neurosci. 12:5. doi: 10.3389/fnmol.2019.00005

# INTRODUCTION

fnmol-12-00005 January 23, 2019 Time: 17:10 # 2

Pain is defined as a distressing experience associated with actual or potential tissue damage with sensory, emotional, cognitive, and social components (Williams and Craig, 2016). Clinically, chronic pain induced by various factors, including tissue inflammation, nerve damage, viral infection, and metabolic disorders, causes patients to suffer spontaneous pain, hyperalgesia and allodynia. At the same time, chronic pain induces a strong emotional response, making it aversive and commonly comorbid with anxiety or depressive disorders. Furthermore, a reciprocal facilitatory effect exists between pain sensitivity and negative emotions. However, chronic pain complaints are generally poorly served by existing therapies (Yekkirala et al., 2017). These patients often do not receive adequate and effective treatment due to limited efficacy or dose-limiting side effects of the current analgesics. Therefore, an in-depth understanding of the molecular mechanisms underlying the development of chronic pain is of significance for the development of innovative analgesics.

Downstream regulator element antagonist modulator (DREAM), a member of the neuronal calcium sensor (NCS) family that contains four Ca2+-binding EF-hand motifs, was shown to be a critical transcriptional repressor for pain modulation (Cheng et al., 2002). It represses gene expression as a tetramer via direct binding to the downstream regulatory element (DRE) site containing the central core sequence GTCA (Ledo et al., 2000). Increased intracellular Ca2<sup>+</sup> concentration can prevent binding of DREAM to the DRE site and derepress DRE-dependent gene expression. Functional expression of DREAM in cortex, hippocampus, cerebellum, spinal cord, dorsal root ganglion (DRG), pineal gland, thyroid and blood cells was validated, and numerous target genes of DREAM were identified, including Fos, Pdyn (prodynorphin), Slc8a3 (solute carrier family 8 member A3), Npas4 (neuronal PAS domain protein 4) and Bdnf (brain-derived neurotrophic factor) (**Table 1**) (Carrion et al., 1998, 1999; Sanz et al., 2001, 2002; Link et al., 2004; Rivas et al., 2004; D'Andrea et al., 2005; Gomez-Villafuertes et al., 2005; Savignac et al., 2005; Venn et al., 2008; Dierssen et al., 2012; Mellstrom et al., 2014; Benedet et al., 2017). Actually, DREAM is identical to KChIP3 (Kv4 channel interacting protein 3), a member of the KChIPs family, which is composed of KChIP1, 2, 3 and 4. KChIPs interact with the Kv4 channels and modulate A-type potassium currents in a Ca2+-dependent manner (An et al., 2000; Anderson et al., 2010). Our recent work revealed the regulation of NMDA receptors and TRPV1 channels by KChIP3/DREAM (Zhang et al., 2010; Tian et al., 2018). Herein, we will refer to the protein KChIP3 for consistency with its gene name Kcnip3.

The role of KChIP3 in pain modulation was studied by both genetic deletion and transgene-mediated overexpression methods. Kcnip3−/<sup>−</sup> mice displayed markedly reduced responses in models of acute thermal, mechanical, and visceral pain and in models of chronic neuropathic and inflammatory pain (Cheng et al., 2002). Elevated levels of Pdyn mRNA and dynorphin A peptide in the spinal cord contributed to the reduction of pain responses. Subsequently, transgenic mice expressing a TABLE 1 | Target genes modulated by transcriptional repressor KChIP3/DREAM.


constitutively active DREAM (daDREAM) mutant displayed a biphasic pain response. The thermal pain reaction was enhanced under basal conditions, while hypoalgesia was observed under inflammatory pain conditions (Rivera-Arconada et al., 2010). Sustained repression of the Bdnf gene impaired the development of central sensitization during inflammatory pain. In addition, recent studies from our group indicated that Kcnip3−/<sup>−</sup> rats showed aggravated thermal hyperalgesia in the complete Freund's adjuvant (CFA)-induced inflammatory pain model (Tian et al., 2018). Taken together, these findings demonstrated the complex regulatory roles of KChIP3 in nociceptive processing.

In the current studies, we performed a series of behavioral tests to investigate the changes in acute and chronic pain responses and negative emotions caused by Kcnip3 gene deletion. Our results showed that Kcnip3−/<sup>−</sup> rats displayed an enhanced response to acute and chronic pain stimuli and stronger paininduced aversion and negative emotions. Finally, the potential novel target genes of KChIP3 were revealed by RNA-Seq analysis and validated by quantitative real-time polymerase chain reaction (qPCR).

#### MATERIALS AND METHODS

#### Antibodies

For Western blot, rabbit anti-KChIP3 (N-terminal 1–20 amino acids) polyclonal antibody was custom-made by GL Biochem Ltd. (Shanghai, China). Mouse anti-pan KChIP (1/2/4) monoclonal antibody (clone K55/82) was purchased from Millipore (Temecula, CA, United States). Mouse anti-β-actin monoclonal antibody was purchased from Zhongshan Jinqiao Biotechnology Ltd. (Beijing, China). Horseradish peroxidase (HRP)-conjugated secondary antibodies, including goat anti-rabbit IgG and goat anti-mouse IgG, were purchased from Santa Cruz Biotechnology (Dallas, TX, United States).

#### Animals

fnmol-12-00005 January 23, 2019 Time: 17:10 # 3

Male Sprague–Dawley rats weighing 250–280 g at the start of the experiments were used. Kcnip3−/<sup>−</sup> rats were generated by the Nanjing Biomedical Research Institute of Nanjing University (Nanjing, China). Deletion of exon 2 of the Kcnip3 gene using the CRISPR/Cas9 system induces a frameshift mutation. Null Kcnip3 gene expression in the knockout rats was verified by qPCR analysis (**Figure 1A**) and Western blot (**Figure 1B**). Targeted deletion of Kcnip3 gene and possible off-target effects of CRISPR/Cas9 were examined by PCR analysis (**Supplementary Figure S1**). Animals were housed under controlled conditions (22 ± 2 ◦C temperature, 55 ± 5% humidity and a 12:12 light/dark cycle) and had free access to food and water.

The animals were handled 3 days before all the experiments. The experiments were carried out in accordance with the recommendations of the Guidelines of the International Association for the Study of Pain. The protocol was approved by the Animal Care and Use Committee of Peking University (permit number: LA2015074). The behavioral tests were performed in a double-blinded manner by two experimenters. One experimenter was responsible for grouping and numbering the rats. The other one who took charge of the behavioral tests was unaware of the genotypes in the whole experiment.

#### qPCR

Total RNA was extracted from the forebrain cortex and purified using the EASYspin Kit (Aidlab, Beijing, China). RNA concentration and purity were measured using a NanoDrop 2000c spectrophotometer (Thermo Scientific, Waltham, MA, United States). SuperScript III reverse transcriptase (Invitrogen, Carlsbad, CA, United States) was used for reverse transcription of RNA into cDNA according to the manufacturer's instructions.

Quantitative PCR was performed using the ABI 7500 instrument (Applied Biosystems, Foster City, CA, United States). SYBR Green 2× PCR Master Mix (Toyobo, Osaka, Japan) was used for the PCR reaction. The primers are listed in **Supplementary Table S3**. The reaction conditions were set as follows: incubation at 95◦C for 1 min, 40 cycles of 95◦C for 15 s, 60◦C for 15 s, 72◦C for 30 s. Lastly, melting curve analysis was performed with 95◦C for 15 s, 60◦C for 1 min and 95◦C for 15 s. Ct values were defined as the number of PCR cycles at which the fluorescence signals were detected. The relative expression levels of Kcnip3 were calculated using the 2−11Ct method and were normalized by Gapdh. Each sample was measured in triplicate.

#### Western Blot Analysis

Naïve wild-type rats or Kcnip3−/<sup>−</sup> rats were deeply anesthetized with 1% sodium pentobarbital. The cortex, hippocampus, spinal cord and bilateral L4–L5 DRG were quickly removed and immediately homogenized in ice-cold lysis buffer (Tiangen Biotech, Beijing, China). The homogenates were centrifuged at 12,000 × g for 5 min at 4◦C and the supernatants were analyzed. Protein concentrations were measured using a BCA assay kit (Thermo Scientific). Next, 50 µg of each sample was boiled for 5 min with SDS-PAGE sample buffer, subjected to SDS-PAGE using 12% running gels, and transferred onto

FIGURE 1 | Validation of Kcnip3 gene knockout in Kcnip3−/<sup>−</sup> rats by quantitative real-time PCR (qPCR) and Western blot. (A) qPCR analysis of Kcnip3 gene expression in the cerebral cortex of wild-type (WT) and Kcnip3−/<sup>−</sup> rats. ∗∗p < 0.01, paired t-test. (B) Western blot analysis of KChIP3 protein expression in the central nervous system of wild-type and Kcnip3−/<sup>−</sup> rats. KO, knockout; Hippo, hippocampus. (C) Western blot analysis of KChIP1, 2 and 4 expression in the peripheral and central nervous system of wild-type and Kcnip3−/<sup>−</sup> rats under normal condition. DRG, dorsal root ganglion. (D) Relative quantification of KChIP1, 2 and 4 protein levels in the peripheral and central nervous system of wild-type and Kcnip3−/<sup>−</sup> rats. n = 3 for both groups. <sup>∗</sup>p < 0.05, paired t-test.

nitrocellulose membranes. The membranes were blocked with 5% non-fat milk in TBST (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.05% Tween 20) for 1 h at room temperature and then incubated overnight at 4◦C with the appropriate primary antibody [anti-KChIP3, 1:100; anti-pan KChIP(1/2/4) antibody, 1:500; β-actin, 1:1,000]. The blots were then washed with TBST three times for 10 min each time. Next, the membranes were incubated with the appropriate horseradish peroxidase-conjugated secondary antibody for 1 h at room temperature. Finally, the blots were developed with a lightening chemiluminescence kit (Santa Cruz Biotechnology).

#### RNA-Sequencing and Data Analysis

RNA isolation was performed as described above. Genomic DNA was removed by gDNA removal column provided in the kit. An Agilent RNA Nano Kit and an Agilent 2100 Bioanalyzer (Santa Clara, CA, United States) were used for RNA integrity and concentration detection. For each sample, 5 µg of total RNA was used to construct the Illumina sequencing libraries according to the manufacturer's instructions. The libraries were sequenced using the Illumina HiSeq X Ten platform to generate high-quality paired-end reads of 150 nt.

Rattus norvegicus genome sequences and annotated gene models were downloaded from ENSEMBL (Rnor6). Raw sequencing reads were first processed to remove adaptors and low-quality bases using Fastqc and Trimmomatic (Bolger et al., 2014) and then aligned to reference genome sequences using STAR (2.5.2b) with gene annotation indexed (Dobin and Gingeras, 2015). The mapping quality and saturation analysis were performed using RSeQC (Wang et al., 2016). Differentially expressed genes were identified using DESeq2 (Love et al., 2014) with absolute log<sup>2</sup> transformed fold change (FC) value >0.58 and multiple-testing adjusted p-value (also known as false discovery rate, FDR) <0.05. Comparison analysis and plots were performed using in-house transcripts and online plot tools<sup>1</sup> based on Python and R.

The sequencing data have been uploaded in the website of BIG Data Center in Beijing Institute of Genomics of Chinese Academy of Sciences (Beijing, China). The assigned accession number of the submission is CRA001181.

#### CFA-Induced Inflammatory Pain Model

Paw inflammation was induced by intraplantar injection of 100 µl CFA (Sigma-Aldrich, St. Louis, MO, United States) into the left hindpaw.

#### Formalin Test

Before testing, the rats were allowed to acclimatize to the experimental environment for 30 min. Then, the planta of left hindpaw received subcutaneous injection of 100 µl 5% formalin, which was prepared by dilution with 0.9% saline. Then the number of flinches and the time spent licking the injected paw were recorded for 60 min by a digital camera.

#### <sup>1</sup>www.ehbio.com/ImageGP

#### Hot Plate Test

The rats were placed on a hot plate (Bioseb, United States) to adapt to the environment for 15 min before testing. During the test, the temperature of the hot plate was stabilized at 52◦C. The latency to lick the hindpaw or jumping behavior was recorded by a digital camera. To avoid tissue injury, a cut-off time was set at 30 s.

#### Cold Plate Test

Adaptation to the environment is performed according to the process mentioned above. During the test, the temperature of the cold plate (Bioseb, United States) was stabilized at 4◦C. The number of paw elevations within 1 min was recorded by a digital camera. To avoid tissue injury, a cut-off time was set at 60 s.

#### Elevated Plus Maze Test

The elevated plus maze apparatus consisted of four arms (50 cm × 10 cm) made of black Plexiglas, two comprising 40 cm-high walls (closed) and two comprising 1.5-cm-high borders (open) and was elevated 73 cm above the floor. Rats were placed at the center of the maze in a room with dim light for 5 min. A video camera fixed above the maze was used to record the movement of each animal. The number of entries and time spent in each arm were scored. The maze was cleaned with absolute ethanol between each rat.

#### Open Field Test

The open field apparatus consisted of a clear Plexiglas box (100 cm<sup>3</sup> × 100 cm<sup>3</sup> × 40 cm<sup>3</sup> ). Each rat was gently placed in the center of the arena and was allowed to explore the area in a room with dim light for 5 min. The cumulative distance traveled and the time spent in the center (60 cm × 60 cm) were recorded using a digital camera above the arena. The arena was cleaned with absolute ethanol between each rat.

#### Forced Swimming Test

Rats were gently placed in the open transparent vertical cylindrical container (diameter 20 cm, height 50 cm) containing water (25◦C) and allowed to swim for 6 min under normal light. Water depth prevented rats from hitting the bottom of cylinder with their tails or hind limbs. Rat behaviors were videotaped from the side. The immobile time when rats remained floating or motionless with only movements necessary for maintaining balance in the water during the last 4 min of the test was scored. Each rat received a pretest 24 h before the test, during which rats were placed to the cylinder of water for 15 min.

#### Sucrose Preference Test

Rats were housed individually and acclimated for 2 days with two bottles of water, followed by two bottles of 2% sucrose for 2 days. Rats were then deprived of water for 24 h and then exposed to a bottle of 2% sucrose and a bottle of water for 2 h in the dark room. Bottle positions were switched after 1 h (for 2 h test). The total consumption of each liquid was recorded, and the sucrose preference was defined as the average sucrose

Frontiers in Molecular Neuroscience | www.frontiersin.org

consumption ratio, which was calculated by dividing the total volume of sucrose intake by the total volume of water and sucrose intake.

# Conditioned Place Aversion Test (CPA)

The CPA test was performed in a shuttle box that consisted of two equal-sized compartments with distinct tactile and visual cues (one had four lamplets that formed a square on the wall and a stainless-steel mesh floor, and the other had four lamplets that formed a triangle on the wall and a stainless-steel rod floor) under dim light.

After habituation for 1 day, the experiment was conducted for consecutive 6 days. On day 1 (preconditioning session), rats freely explored the two compartments for 900 s, and the time spent in each compartment was recorded. Rats that spent more than 600 s or that spent more than 80% of the total time (>720 s) in one compartment were eliminated in the following experiments. In later experiments, we chose the compartment in which the rat spent more than 50% of the total time (>450 s) as the painpaired compartment. On day 2–day 5 (conditioning session), each rat was confined in the non-pain-paired compartment for 1 h following an intraplantar injection of saline (100 µl) into the right hindpaw. Then the rat was given an intraplantar injection of 5% formalin (100 µl) into the left hindpaw and confined in the pain-paired compartment for 1 h. On day 6 (test session), each rat was allowed to explore the two compartments freely, and the time spent in each compartment during the 900-s session was measured. Two compartments were equipped with an overhead camera to track the rat position. The percentage of preference was determined via AnyMaze software.

# Statistical Analysis

All of the data are represented as the mean ± SEM. Comparisons between two groups were performed using Student's unpaired or paired t-test. Comparisons between two groups at different time points were performed using two-way ANOVA with Sidak's multiple comparisons test. The criterion for statistical significance was p < 0.05, and differences were calculated using GraphPad Prism 7.0.

# RESULTS

#### Validation of Kcnip3 Gene Deletion in the Kcnip3−/<sup>−</sup> Rats

Kcnip3−/<sup>−</sup> rats were generated by CRISPR/CAS9-mediated deletion of exon 2 of rat Kcnip3 gene as described previously (Tian et al., 2018). qPCR experiments using primers spanning exon 1 and exon 2 of the rat Kcnip3 gene could barely detect the predicted PCR product in the forebrain cortex of knockout rats (**Figure 1A**), suggesting efficient deletion of the targeted sequence. At the same time, RNA-Seq analysis in the following studies could not detect the presence of transcripts from exon 2 (data not shown). Although transcripts of exon 3–exon 9 did not show difference between the wild-type and knockout rats, their translation is blocked by frameshift mutation in the knockout animal. In addition, Western blot analysis in the cortex, hippocampus and spinal cord tissues indicated the absence of KChIP3 protein, which was detected using the custom-made anti-KChIP3 antibody against KChIP3 N-terminus (**Figure 1B**). These data validated the efficient targeted deletion of the Kcnip3 gene.

Compensatory upregulation among family members commonly occurred in the global knockout animal. Therefore, the cortex, hippocampus, spinal cord, and DRG tissues were collected from naïve wild-type or Kcnip3−/<sup>−</sup> rats. Anti-pan KChIP antibody (Pruunsild and Timmusk, 2012), which can recognize KChIP1, 2 and 4, but not KChIP3 (**Supplementary Figure S2**), was used for the Western blot analysis (**Figure 1C**). Quantification analysis showed significant upregulation of KChIP1, 2 and 4 protein in the cortex and hippocampal tissues of knockout rats (p < 0.05, paired t-test). And an increased trend of expression was observed in the spinal cord and DRG tissues. In addition, qPCR analysis was performed to detect the transcripts of Kcnip1, 2 and 4 (**Supplementary Figure S3A**). However, no significant difference was observed between the groups. Therefore, compensatory upregulation of KChIP1, 2 and 4 occurs in the post-transcriptional level.

### Kcnip3−/<sup>−</sup> Rats Display Increased Pain Sensitivity in Both Acute Nociceptive and Chronic Pain Models

In general, Kcnip3−/<sup>−</sup> rats appear healthy and normal and no obvious abnormality in their motor activity and behaviors were observed. Although the knockout rats would be slightly smaller in the late phase of growth, no obvious weight difference was observed around 8 weeks, when the behavioral tests were performed.

To observe the influence of Kcnip3 gene knockout on the pain responses of rats, we established acute nociceptive and chronic pain models by intraplantar injection of formalin and CFA, respectively (**Figure 2A**). The formalin test is a reliable and widely used model of continuous pain with a biphasic response consisting of the first transient phase lasting for the first 10 min and the second phase from 10 to 60 min. The first phase is thought to result from direct activation of primary sensory neurons, whereas the second phase has been proposed to reflect the combined effects of afferent input and central sensitization occurring in the dorsal horn (McNamara et al., 2007). We found that both wild-type and Kcnip3−/<sup>−</sup> rats produced a typical biphasic pain response after formalin injection, whereas Kcnip3−/<sup>−</sup> rats exhibited increased flinching behavior (time = 25 min, wild-type: 41.00 ± 3.11, Kcnip3−/−: 56.88 ± 4.88, p < 0.05, two-way ANOVA followed by Sidak's multiple comparisons test; two groups, p < 0.05, two-way repeated measures ANOVA; **Figure 2B**) and longer duration to lick (time = 25 min, wild-type: 76.38 ± 4.66, Kcnip3−/−: 97.63 ± 5.42, p < 0.05; time = 30 min, wildtype: 61.88 ± 5.76, Kcnip3−/−: 89.75 ± 3.79, p < 0.001; two groups, p < 0.001, two-way ANOVA followed by Sidak's multiple comparisons test; **Figure 2C**) compared to the wildtype rats in the second phase of the test. Cumulative analysis

FIGURE 2 | Increased pain sensitivity after intraplantar injection of formalin or complete Freund's adjuvant (CFA) in Kcnip3−/<sup>−</sup> rats. (A) Schematic diagram of the establishment of formalin or CFA-induced pain models. (B,C) Flinching counts (B) and licking duration (C) of hindpaw at 5-min intervals for 1 h in the formalin pain model of wild-type and Kcnip3−/<sup>−</sup> rats. n = 8 for both groups. <sup>∗</sup>p < 0.05, ∗∗∗p < 0.001, two-way repeated-measures ANOVA followed by Sidak's multiple comparisons test. (D) Analysis of the cumulative number of flinching behaviors (left) and licking time (right) of hindpaw in the second phase (20–50 min post injection) of formalin pain model of wild-type and Kcnip3−/<sup>−</sup> rats. ∗∗p < 0.01, ∗∗∗p < 0.001, unpaired t-test. (E) Time course of hindpaw licking latency in the 52◦C hot plate test in the CFA pain model of wild-type and Kcnip3−/<sup>−</sup> rats. n = 8 for both groups. ∗∗p < 0.01, comparison between the genotypes at the indicated time points; ###p < 0.001, comparison between the two curves, two-way repeated-measures ANOVA followed by Sidak's multiple comparisons test. (F) Time course of lifting counts of hindpaw in the 4◦C cold plate test within 1 min post CFA injection in wild-type and Kcnip3−/<sup>−</sup> rats. n = 8 for both groups. <sup>∗</sup>p < 0.05, comparison between the genotypes at the indicated time points; ###p < 0.001, comparison between the two curves, two-way repeated-measures ANOVA followed by Sidak's multiple comparisons test.

of the number of flinches and licking time during 20– 50 min showed a stronger pain response in Kcnip3−/<sup>−</sup> rats (flinching, wild-type: 208.8 ± 6.886, Kcnip3−/−: 255.4 ± 11.24, p < 0.01, unpaired t-test; licking, wild-type: 302.4 ± 9.464, Kcnip3−/−: 402.4 ± 12.59, p < 0.001, unpaired t-test; **Figure 2D**).

We also induced a persistent inflammatory pain model via injection of CFA in the hindpaw and measured the pain responses using the 52◦C hot plate and 4◦C cold plate tests. There was no significant difference in the basal pain responses between Kcnip3−/<sup>−</sup> rats and the wild-type control. However, after CFA injection the Kcnip3−/<sup>−</sup> rats exhibited significantly decreased licking latency in the hot plate test (time = 2nd day, wild-type: 10.28 ± 0.40, Kcnip3−/−: 8.09 ± 0.47, p < 0.01; time = 4th day, wild-type: 10.39 ± 0.48, Kcnip3−/−: 8.13 ± 0.40, p < 0.01; time = 6th day, wild-type: 9.88 ± 0.40, Kcnip3−/−: 7.69 ± 0.38, p < 0.01; two groups, p < 0.001, two-way ANOVA followed by Sidak's multiple comparisons test; **Figure 2E**). Similarly, the Kcnip3−/<sup>−</sup> rats showed an increased number of hindpaw lifting within 1 min in the cold plate test (time = 2nd day, wild-type: 10.50 ± 0.63, Kcnip3−/−: 12.88 ± 0.23, p < 0.05; time = 4th day, wild-type: 9.88 ± 0.97, Kcnip3−/−: 12.63 ± 0.63, p < 0.05; time = 6th day, wild-type: 10.00 ± 0.60, Kcnip3−/−: 12.63 ± 0.68, p < 0.05; two groups, p < 0.001, two-way ANOVA followed by Sidak's multiple comparisons test; **Figure 2F**). Taken together, these data suggested increased pain sensitivity of Kcnip3−/<sup>−</sup> rats in the late phase of inflammatory pain.

### Kcnip3−/<sup>−</sup> Rats Show a Stronger Aversive Response to the Nociceptive Stimuli

Pain consists of sensory-discriminative and negativeaffective components, including aversion to pain-associated

FIGURE 4 | Enhanced anxiety-like behavior in Kcnip3−/<sup>−</sup> rats during inflammatory pain. (A,B) The elevated plus-maze test. Kcnip3−/<sup>−</sup> rats made fewer visits to the open arms (A) and spent less time in the open arms (B) in a 5-min test 1 day after CFA injection compared to the wild-type rats. n = 8 for both groups. <sup>∗</sup>p < 0.05, unpaired t-test. (C,D) The open field test. (C) Kcnip3−/<sup>−</sup> rats displayed less time spent in the center of the open field in a 5-min test 1 day after CFA injection compared to the wild-type rats. n = 8 for both groups. ∗∗p < 0.01, unpaired t-test. (D) Wild-type and Kcnip3−/<sup>−</sup> rats exhibit similar locomotor activity in the open field test 1 day after CFA injection. n = 8 for both groups, unpaired t-test.

environments, anxiety and depression. In the following studies, we explored how Kcnip3 gene knockout affected the emotional responses of pain. First, we performed the CPA test associated with formalin injection (**Figure 3A**). During the 4-day training session, the rat is confined to one compartment following formalin injection and confined to the opposite compartment following saline injection each day. In the test session, both wild-type and Kcnip3−/<sup>−</sup> rats showed avoidance for the compartment paired with formalin injection, as demonstrated in **Figure 3B** (wild-type, pre: 54.02 ± 1.21, post: 34.01 ± 1.10, p < 0.001; Kcnip3−/−, pre: 54.31 ± 1.36, post: 25.52 ± 2.30, p < 0.001, two-way ANOVA followed by Sidak's multiple comparisons test). However, Kcnip3−/<sup>−</sup> rats displayed stronger aversion to the formalinpaired compartment (two groups, p < 0.05, two-way repeated measures ANOVA; **Figure 3B**) and exhibited a more profound decrease in preference for the formalin-paired compartment (wild-type: −20.01 ± 1.37%, Kcnip3−/−: −28.80 ± 2.49%, p < 0.01, unpaired t-test; **Figure 3C**). As a control, the amount of locomotor activity between the two groups showed no significant difference before and after the conditioning procedure (**Figure 3D**). In summary, these data suggest that Kcnip3 knockout aggravated the aversive response to nociceptive stimuli in rats.

### Kcnip3−/<sup>−</sup> Rats Exhibited Increased Anxiety-Like Behavior Post CFA Injection

Elevated plus maze and open field tests are routinely used for the assessment of anxiety-related behavior. These tests are based on the naturalistic conflict between the tendency to explore a novel environment and the aversive properties of a brightly lit, open area. Our previous studies indicated that Kcnip3−/<sup>−</sup> rats displayed high basal anxiety levels in the elevated plus maze test (Li et al., 2018). Herein, we examined the anxiety level of Kcnip3−/<sup>−</sup> rats during inflammatory pain. These rats showed decreased open-arm entries (wild-type: 2.75 ± 0.367, Kcnip3−/−: 1.38 ± 0.32, p < 0.05, unpaired t-test; **Figure 4A**) and spent less time in the open arms (wild-type: 24.76 ± 2.04, Kcnip3−/−: 17.71 ± 1.48, p < 0.05, unpaired t-test; **Figure 4B**) in the elevated plus maze test 1 day after CFA injection. In the open field test, Kcnip3−/<sup>−</sup> rats displayed a significant decrease in the time spent in the open center area (wild-type: 12.18 ± 0.58, Kcnip3−/−: 9.42 ± 0.62, p < 0.05, unpaired t-test; **Figure 4C**). In contrast, analysis of the locomotor ability during the open field test showed no difference between the two groups (**Figure 4D**). Altogether, these results suggest that Kcnip3 knockout exacerbates inflammatory pain-induced anxiety-like behavior in rats.

### Kcnip3−/<sup>−</sup> Rats Exhibited Increased Depressive-Like Behavior Under Both Normal and Inflammatory Pain Conditions

The core symptoms of depression include behavioral despair and anhedonia, a reduced sensitivity to reward. The forced swimming test (**Figure 5A**) measures the coping strategy of

FIGURE 6 | (A) Volcano plots illustrating log10(adjusted p-value) in relation to the log2(fold change) for the differentially expressed genes in Kcnip3−/<sup>−</sup> rats compared to wild-type rats. Genes that passed the significance threshold (FDR < 0.05) and the expression cut-off log2(fold change) > 0.58 are colored red (upregulated, fold change > 1.5) or green (downregulated, fold change < 0.67), while genes outside this range are colored blue. (B) qPCR analysis of upregulated genes in the forebrain cortex of wild-type (WT) and Kcnip3−/<sup>−</sup> rats. n = 4–5, <sup>∗</sup>p < 0.05, paired t-test. (C) qPCR analysis of downregulated genes in the forebrain cortex of WT and Kcnip3−/<sup>−</sup> rats. n = 3, <sup>∗</sup>p < 0.05, ∗∗p < 0.01, paired t-test.

the animal to an acute inescapable stress, and the physical immobility is thought to be an indication of behavioral despair. Kcnip3−/<sup>−</sup> rats showed significantly increased immobility time both under normal conditions (wild-type: 50.86 ± 6.18, Kcnip3−/−: 78.14 ± 7.98, p < 0.05, unpaired t-test) and 1 day post CFA injection (wild-type: 68.57 ± 6.48; Kcnip3−/−:

108.00 ± 7.06, p < 0.01, unpaired t-test; **Figure 5B**). In addition, the wild-type showed an increasing trend in the immobility time (pre: 50.86 ± 6.18, post: 68.57 ± 6.48, p > 0.05, paired t-test) post CFA injection compared to that prior to CFA injection, whereas Kcnip3−/<sup>−</sup> rats showed significantly increased immobility time (pre: 78.14 ± 7.98, post: 108.00 ± 7.06, p < 0.05, paired t-test), suggesting that Kcnip3−/<sup>−</sup> rats are more vulnerable to depression than wild-type rats.

The sucrose preference test (**Figure 5C**) represents the anhedonia-like behavioral change. Kcnip3−/<sup>−</sup> rats exhibited lower sucrose preference to 2% solution both under normal conditions (wild-type: 81.57 ± 4.8, Kcnip3−/−: 63.86 ± 4.11, p < 0.05, unpaired t-test) and 1 day post CFA injection (wildtype: 67.43 ± 3.88, Kcnip3−/−: 44.43 ± 5.47, p < 0.05, unpaired t-test; **Figure 5D**). At the same time, the wild-type rats showed a decreasing trend in sucrose preference post CFA injection (pre: 81.57 ± 4.83, post: 67.43 ± 3.88, p > 0.05, paired t-test), whereas the Kcnip3−/<sup>−</sup> rats showed significantly decreased sucrose preference (pre: 63.86 ± 4.11, post: 44.43 ± 5.47, p < 0.05, paired t-test) post CFA injection compared to that prior to CFA injection, suggesting that Kcnip3−/<sup>−</sup> rats are more susceptible to depression than wild-type rats. Taken together, the above data indicate that Kcnip3 knockout aggravates anxiety- and depressive-like behaviors in rats. KChIP3 might play an anxiolytic and antidepressant action in vivo.

# RNA-Seq Analysis Revealed the Differentially Expressed Genes in Kcnip3−/<sup>−</sup> Rats

To search for differentially expressed genes in Kcnip3−/<sup>−</sup> rats, RNA-Seq transcriptional profiling was performed. The cerebral cortex in the forebrain was collected from four Kcnip3−/<sup>−</sup> rats and four wild-type rats at 8 weeks of age. Using the criteria of FDR < 0.05 and abs(log2FC) > 0.58 (upregulated: FC > 1.5; downregulated: FC < 0.67), 68 upregulated genes were identified, and 79 downregulated genes were identified (**Figure 6A** and **Supplementary Tables S1, S2**).

Among the upregulated genes, 15 genes are associated with neural development, including Nr4a2, Ret, Egr3, Rgs9, Bcl11b, Rtn4rl2, Rspo1, Auts2, Scrt2, Itgad, Bag 3, Pcdhgb6, Hmox1, Rfx4 and Hbegf (**Table 2**). Five genes are involved in transcriptional regulation, including Nr4a2, Egr3, Bcl11b, Scrt2 and Rfx4. Genes Rps27a and Sgms25 are associated with myelin sheath function. In addition, Cplx3 is involved in synaptic vesicle exocytosis and neurotransmitter release. Genes Nr4a2, Rgs9 and Itgad are related to dopamine neurotransmission. The Nr4a2 encoded protein acts as a transcriptional activator and is involved in projection neuron axonogenesis, neuron maturation and migration. In particular, it is associated with dopaminergic neuron differentiation and dopamine biosynthetic processes. Rgs9 is involved in nervous system development and the dopamine receptor signaling pathway. Itgad is related to the negative regulation of dopamine metabolic processes. Consistently, expression of Crem and Fosb, the known target genes of KChIP3 (Link et al., 2004; Ruiz-DeDiego et al., 2015), were found to be upregulated in Kcnip3−/<sup>−</sup> rats.

Among the downregulated genes, six genes are associated with neural development, including AABR070317, Col3a1, Itm2a, Aqp1, Pcdhgb8 and Pcdhgb7 (**Table 3**). Notably, the cell adhesion molecule genes Pcdhb3, Pcdhb22 and Pcdhb20 are associated with chemical synaptic transmission and synapse assembly. The Ddc and Sncaip genes are related to dopamine and serotonin biosynthetic processes and dopamine metabolic processes, respectively. Hba-a2 and Ubc are associated with myelin sheath function. In addition, the Nqo2-encoded protein is involved in the positive regulation of the neuronal apoptotic process and memory deficit.

Further, we used qPCR experiments to check the upregulated or downregulated genes caused by Kcnip3 gene deletion. Gene expression of Nr4a2, Ret, Rgs9 and Cplx3 were significantly increased (Nr4a2, 2.24 ± 0.37 fold of WT control, p < 0.05; Ret, 1.67 ± 0.17 fold of WT control, p < 0.05; Rgs9, 2.59 ± 0.52, p < 0.05; Cplx3, 1.63 ± 0.19 fold of WT control, p < 0.05, paired t-test; **Figure 6B**). The expression of Nr4a1 and Itgad showed an increased trend. Considering the transcriptional repressor activity of KChIP3, these upregulated genes might be the new candidate target genes repressed by KChIP3. However, expression of Pdyn and Bdnf, the known target genes of KChIP3, did not show upregulation in our analysis (**Supplementary Figure S3B**). On the other hand, gene expression of Col3a1, Pcdhb3, Itm2a and Ddc were significantly decreased (Col3a1, 0.69 ± 0.07 fold of WT control, p < 0.05; Pcdhb3, 0.32 ± 0.08 fold of WT control, p < 0.05; Itm2a, 0.21 ± 0.05, p < 0.01; Ddc, 0.21 ± 0.11 fold of WT control, p < 0.05 or p < 0.01, paired t-test; **Figure 6C**). Altogether, results from qPCR analysis validated the upregulated and downregulated genes revealed by RNA-Seq analysis in Kcnip3−/<sup>−</sup> rats.

# DISCUSSION

In the current studies, we performed a series of behavioral tests to observe the changes in pain sensitivity and negative emotions in Kcnip3−/<sup>−</sup> rats. The knockout rats showed increased spontaneous behaviors in the formalin test and enhanced heat hyperalgesia and cold hyperalgesia in the CFA test. Notably, Kcnip3−/<sup>−</sup> rats displayed stronger aversion to the pain-paired compartment in the CPA test and showed higher levels of anxiety and depression post CFA injection. At the same time, the knockout rats are more depressed than the wild-type rats under the basal condition. As negative emotions might aggravate the pain responses, the higher anxiety and depression level in Kcnip3−/<sup>−</sup> rats might contribute to the increased sensitivity to pain. Altogether, our studies provide evidence for the involvement of KChIP3 in negative emotions and possible role in central nociceptive processing.

With respect to the mechanisms of the behavioral changes of Kcnip3−/<sup>−</sup> rats, changes in gene expression associated with neural development and synaptic transmission, particularly dopaminergic neurotransmission, were revealed by RNA-Seq analysis in the forebrain cortex. Further studies are needed to address whether structural changes in the brain during development and abnormalities in dopaminergic



TABLE 3 | Downregulated genes associated with neural development, neurotransmission or myelin sheath function in the forebrain cortex of Kcnip3−/<sup>−</sup> rat compared to that of wild-type rats.


neurotransmission occurring in Kcnip3−/<sup>−</sup> rats affect central nociceptive and emotional processing.

#### Involvement of KChIP3 in Pain Modulation

The involvement of KChIP3 in pain modulation was first described in Kcnip3−/<sup>−</sup> mice (Cheng et al., 2002). The knockout mice displayed markedly reduced pain behaviors both in models of acute thermal, mechanical, and visceral pain and in models of chronic neuropathic and inflammatory pain. The attenuation of the pain response was ascribed to the elevated level of Pdyn expression in the spinal cord. Later, studies in transgenic daDREAM mice (with high expression of the dominant active DREAM in DRG and spinal cord in addition to telencephalon) showed that these mice displayed a biphasic pain response, basal hyperalgesia and reduced hyperalgesic response following peripheral inflammation (Rivera-Arconada et al., 2010). In detail, the daDREAM mice showed an enhanced response to thermal and visceral noxious stimuli in basal conditions. However, they

displayed an impaired response to inflammatory pain with milder and shorter-lasting hyperalgesia compared to the wildtype mice. Attenuation of central sensitization due to reduced Bdnf gene expression contributed to the hypoalgesic behavior of the transgenic mice. Recently, it was reported that another line of daDREAM mice (with daDREAM expression in trigeminal ganglia) showed a significant increase in the rubbing response in the first and second phases of the formalin test, which might be correlated with decreased expression of Pdyn (Benedet et al., 2017). Conversely, the nocifensive response to 4.5% formalin injection in the snoot in Kcnip3−/<sup>−</sup> mice was milder than that in normal mice.

Recent studies from our lab demonstrated that Kcnip3−/<sup>−</sup> rats exhibited aggravated heat hyperalgesia behavior following CFA injection, which was measured by a radiant heat test (Tian et al., 2018). Consistently, current studies using formalin and CFA model of pain showed enhanced pain responses in Kcnip3−/<sup>−</sup> rats. The exact reason for the discrepancy between Kcnip3−/<sup>−</sup> mice and Kcnip3−/<sup>−</sup> rats remains unknown. In addition to the species difference, potential unwanted off-target effects of the CRISPR-Cas9 technology or the compensatory upregulation of KChIP1, 2 and 4 protein following Kcnip3 gene deletion (**Figures 1C,D**) might also influence the behavioral phenotype. The region- and time-specific gene deletion method needs to be used to further elucidate the role of KChIP3 in pain transmission.

# Potential Anxiolytic and Antidepressant Effects of KChIP3

Previously, lines of evidence indicated the participation of central KChIP3 in learning and memory in mice. For example, Kcnip3−/<sup>−</sup> mice exhibited remarkably increased short-term memory as well as significantly enhanced long-term memory (Fontan-Lozano et al., 2009). Conversely, contextual fear memory, but not auditory fear memory, was significantly impaired in daDREAM mice (Wu et al., 2010). In addition, the daDREAM mice showed a clear defect in spatial memory and associative learning (Mellstrom et al., 2014). However, the involvement of KChIP3 in emotional processing has attracted less attention.

In fact, previous studies in Kcnip3−/<sup>−</sup> mice demonstrated that they have slightly increased anxiety levels compared to the wild-type controls, and no significant difference was observed compared to the wild-type control (Alexander et al., 2009). Another study in female Kcnip3−/<sup>−</sup> mice also showed that the Kcnip3 gene deletion did not affect the anxiety level in the ovariectomized mice receiving or not receiving estradiol injections (Tunur et al., 2013). However, both our studies from elevated plus maze and open field tests indicated that Kcnip3−/<sup>−</sup> rats had a higher anxiety level compared to the wild-type rats post CFA injection. Combined with our previous results showing the higher basal anxiety level of Kcnip3−/<sup>−</sup> rats (Li et al., 2018), the potential anxiolytic action of KChIP3 protein was supposed.

In addition to increased anxiety-like behavior, Kcnip3−/<sup>−</sup> rats showed stronger aversion mood in the CPA test and showed more depression-like behavior in the forced swimming test and sucrose preference test both under basal and inflammatory pain conditions. All these data support the relief effect of KChIP3 on negative emotions. However, the target genes or ion channels of regulated by KChIP3 in the above processes need to be investigated in future studies.

# The Emerging Role of KChIP3 in Neural Development

The RNA-Seq analysis in the forebrain cortex revealed that 15 upregulated genes and 6 downregulated genes in Kcnip3−/<sup>−</sup> rats are associated with neural development, implying the potential involvement of KChIP3 in neural development. In detail, Nr4a, Ret and Bcl11b are related to neuron axonogenesis. Ret and Bcl11b are associated with neuron differentiation. Ret, Auts2, Scrt2, Col3a1 and Aqp1 are involved in neuron migration. Bcl11b, Rtn4rl2, Bag3 and Aqp1 contribute to neuron projection. In particular, Nr4a2, Egr3, Rtn4rl2, Bag3 and Col3a1 are associated with habenular, peripheral nervous system, corpus callosum, brain and spinal cord, and cerebral cortex development, respectively. Rfx4 is involved in forebrain, midbrain and dorsal spinal cord development. Altogether, these data support the potential role of KChIP3 in neural development. However, many other genes, such as those related to oxygen transport, iron ion binding and blood cell differentiation, showed changes in their expression after Kcnip3 deletion. Possible impact of these genes on development could not be excluded.

Previous studies indicated that expression of midline 1 (Mid1), a ubiquitin ligase specific for the protein phosphatase 2A, is repressed in daDREAM mice (Dierssen et al., 2012). Related to this, daDREAM mice exhibit a significant shortening of the rostro-caudal axis of the cerebellum and a severe delay in neuromotor development early after birth, suggesting a role of DREAM in cerebellar development. In addition, daDREAM mice showed reduced dendritic basal arborization and spine density in CA1 pyramidal neurons but increased spine density in dendrites in dentate gyrus granule cells in hippocampus (Mellstrom et al., 2016). Recent in vitro studies indicated that KChIP3 contributes to neuritogenesis through RhoA inactivation. PC12 cells expressing KChIP3 had increased neurite outgrowth (Kim et al., 2018). Therefore, although general brain morphology is not remarkably altered in Kcnip3−/<sup>−</sup> rats, structural and morphological changes occurring in the central nervous system of knockout rats might affect central nociceptive and emotional processing, and detailed studies are needed to elucidate this issue.

# The Potential Role of KChIP3 in Dopamine Neurotransmission

RNA-Seq analysis revealed genes involved in synaptic transmission, including Cplx3, Pcdhb3, Pcdhb22 and Pcdhb20. In particular, dopaminergic neurotransmission might be affected by Kcnip3 gene deletion. Nr4a2 and Ddc are associated with dopamine biosynthetic processes. Itgad and Sncaip are associated with dopamine metabolic processes. Rgs9 is related to the modulation of dopamine receptor signaling. Notably, downregulation of Ddc might decrease the biosynthesis of

dopamine and serotonin, which play key roles in motivation, reward and emotional processing. Therefore, decreased dopaminergic and serotonergic neurotransmission might contribute to the enhanced pain-induced aversion, anxiety- and depression-like behaviors in Kcnip3−/<sup>−</sup> rats under both basal and inflammatory pain conditions.

Previous studies demonstrated that the KChIP3 protein was localized in the cell bodies and processes of dopaminergic neurons in the midbrain (Duncan et al., 2009). Moreover, regulation of KChIP3, particularly in mesocortical dopamine neurons, may be part of the action of antipsychotic drugs, such as haloperidol. In addition, L-DOPA-induced dyskinesia was decreased in daDREAM mice, while genetic deletion of Kcnip3 potentiated the intensity of dyskinesia. The KChIP3 protein plays a protective role in L-DOPA-induced dyskinesia in mice (Ruiz-DeDiego et al., 2015). In addition to its transcriptional regulatory function, KChIP3 participates in the modulation of ion channels. A-type potassium channel complex formed by Kv4.3 and KChIP3 plays a key role in the pacemaker control of firing rates of dopaminergic substantia nigra neurons, which correlates with dopamine release (Liss et al., 2001). Taken together, dopaminergic neurotransmission might be affected by Kcnip3 gene deletion, which might cause changes in emotional processing in the brain.

#### A Comparison Between the Previous Study and the Current Study

Previously, Mellstrom et al. (2014) performed cDNA microarray analysis of gene expression profiles in the hippocampus of daDREAM mice. They found that gene expression of Ctgf and Tshz2 was decreased while gene expression of Col3a1, Pcdhb3 and Ddc was increased in the transgenic mice. Consistent with these findings, our RNA-Seq analysis revealed the upregulation of Ctgf and Tshz2 gene expression and downregulation of Col3a1, Pcdhb3 and Ddc gene expression in Kcnip3−/<sup>−</sup> rats. In addition, changes in the expression profiles of Nr4a, Egr, Rgs, Rps, Adamts, Slc and Rbm gene family members were detected in both studies (**Supplementary Table S4**).

#### CONCLUSION

The behavioral tests in Kcnip3−/<sup>−</sup> rats provide the evidence for the involvement of KChIP3 in negative emotions and possible role in central nociceptive processing. RNA-seq analysis in the forebrain cortex revealed novel potential target genes of KChIP3, which are associated with neural development, synaptic transmission, and particularly, dopaminergic neurotransmission. Further studies are needed to elucidate the molecular mechanism of the emotional changes in Kcnip3−/<sup>−</sup> rats and the possible contribution of these target genes.

#### REFERENCES

Alexander, J. C., McDermott, C. M., Tunur, T., Rands, V., Stelly, C., Karhson, D., et al. (2009). The role of calsenilin/DREAM/KChIP3 in contextual fear conditioning. Learn. Mem. 16, 167–177. doi: 10.1101/lm.126 1709

#### AUTHOR CONTRIBUTIONS

Y-PG and YZ designed the experiments. Y-PG performed the behavioral test. Y-RZ and T-TL performed the biochemical studies. Y-PG, Y-RZ, YW, and YZ analyzed the data. Y-PG, Y-RZ, YW, and YZ wrote the manuscript.

#### FUNDING

This work was supported by the National Natural Science Foundation of China (Grants 31771295, 31371143, 31530028, 31720103908, and 81521063) and the Ministry of Science and Technology of China (973 Program Grant 2017YFA0701300).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2019.00005/full#supplementary-material

FIGURE S1 | PCR analysis of CRISPR/Cas9-mediated gene deletion of Kcnip3 (A) and possible off-target effects (B). Arrows indicate the size of PCR products. The sequence of target primers are: forward, ATGAACAAGGCAGGGCTCACT; reverse ATGTTCAAAATAGCTCTGCGGGT; The sequence of off-target primers are: forward, TGGGTGAGCCACCAGGATGAT; reverse, TCTCCCACTGACTGGATGTGG. Touch down PCR procedure was used as follows: incubation at 95◦C for 5 min, 20 cycles of 98◦C for 30 s, 65◦C for 30 s, 72◦C for 45 s followed by 20 cycles of 98◦C for 30 s, 55◦C for 30 s, 72◦C for 45 s, and lastly 72◦C for 5 min. WT, wild-type. KO, knockout.

FIGURE S2 | Western blot analysis ruled out the recognition of KChIP3 by anti-pan KChIP antibody. (A) Western blot analysis in N2a cells transfected with GFP or GFP-KChIP3 plasmid (as described in our previous studies performed by Na-Xi Tian et al., 2018). Expression of KChIP3 protein can be detected by anti-KChIP3 antibody, but not anti-pan KChIP antibody, in the GFP-KChIP3 transfected group. 1 µg GFP or GFP-KChIP3 plasmid was transfected with jetPRIME reagent (Polyplus, NY, United States) into N2a cells and the cells were harvested 24 h later. (B) Expression of KChIP3 in the spinal cord of wild-type rats could be detected by anti-KChIP3 antibody, but not anti-pan KChIP antibody. Kcnip3 gene deletion leads to absence of KChIP3 protein in the knockout group.

FIGURE S3 | qPCR analysis of Kcnip1, Kcnip2 and Kcnip4 (A), Pdyn and Bdnf (B) expression in the forebrain cortex of wild-type (WT) and Kcnip3−/<sup>−</sup> rats. n = 6 for both groups (A). n = 5 for both groups (B).

TABLE S1 | Upregulated genes in the forebrain cortex of Kcnip3−/<sup>−</sup> rat compared to that of wild-type rats.

TABLE S2 | Downregulated genes in the forebrain cortex of Kcnip3−/<sup>−</sup> rat compared to that of wild-type rats.

TABLE S3 | Sequence of primers used for real-time quantitative PCR.

TABLE S4 | A comparison of the differentially expressed genes in Kcnip3−/<sup>−</sup> rats and that in daDREAM mice.


**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 handling Editor and reviewer JN declared their involvement as co-editors in the Research Topic, and confirm the absence of any other collaboration.

Copyright © 2019 Guo, Zhi, Liu, Wang and Zhang. 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) and the copyright owner(s) 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-12-00005 January 23, 2019 Time: 17:10 # 13

# Identification of IQM-266, a Novel DREAM Ligand That Modulates KV4 Currents

Diego A. Peraza1,2† , Pilar Cercós 3† , Pablo Miaja<sup>1</sup> , Yaiza G. Merinero<sup>1</sup> , Laura Lagartera<sup>3</sup> , Paula G. Socuéllamos 1,4 , Carolina Izquierdo García<sup>3</sup> , Sara A. Sánchez <sup>1</sup> , Alejandro López-Hurtado5,6 , Mercedes Martín-Martínez <sup>3</sup> , Luis A. Olivos-Oré<sup>4</sup> , José R. Naranjo5,6 , Antonio R. Artalejo<sup>4</sup> , Marta Gutiérrez-Rodríguez <sup>3</sup> \* and Carmen Valenzuela1,2 \*

1 Instituto de Investigaciones Biomédicas Alberto Sols (IIBM), CSIC-UAM, Madrid, Spain, <sup>2</sup>Spanish Network for Biomedical Research in Cardiovascular Research (CIBERCV), Instituto de Salud Carlos III, Madrid, Spain, <sup>3</sup> Instituto de Química Médica (IQM), IQM-CSIC, Madrid, Spain, <sup>4</sup> Instituto Universitario de Investigación en Neuroquímica & Departamento de Farmacología y Toxicología, Facultad de Veterinaria, UCM, Madrid, Spain, <sup>5</sup>Centro Nacional de Biotecnología (CNB), CNB-CSIC, Madrid, Spain, <sup>6</sup>Spanish Network for Biomedical Research in Neurodegenerative Diseases (CIBERNED), Instituto de Salud Carlos III, Madrid, Spain

#### Edited by:

Lane Brown, Washington State University, United States

#### Reviewed by:

Ying Zhang, Dalhousie University, Canada Thomas Budde, University of Münster, Germany

#### \*Correspondence:

Marta Gutiérrez-Rodríguez mgutierrez@iqm.csic.es Carmen Valenzuela cvalenzuela@iib.uam.es

†These authors have contributed equally to this work

Received: 03 October 2018 Accepted: 14 January 2019 Published: 04 February 2019

#### Citation:

Peraza DA, Cercos P, Miaja P, Merinero YG, Lagartera L, Socuéllamos PG, Izquierdo García C, Sánchez SA, López-Hurtado A, Martín-Martínez M, Olivos-Oré LA, Naranjo JR, Artalejo AR, Gutiérrez-Rodríguez M and Valenzuela C (2019) Identification of IQM-266, a Novel DREAM Ligand That Modulates KV4 Currents. Front. Mol. Neurosci. 12:11. doi: 10.3389/fnmol.2019.00011 Downstream Regulatory Element Antagonist Modulator (DREAM)/KChIP3/calsenilin is a neuronal calcium sensor (NCS) with multiple functions, including the regulation of A-type outward potassium currents (IA). This effect is mediated by the interaction between DREAM and KV4 potassium channels and it has been shown that small molecules that bind to DREAM modify channel function. A-type outward potassium current (IA) is responsible of the fast repolarization of neuron action potentials and frequency of firing. Using surface plasmon resonance (SPR) assays and electrophysiological recordings of KV4.3/DREAM channels, we have identified IQM-266 as a DREAM ligand. IQM-266 inhibited the KV4.3/DREAM current in a concentration-, voltage-, and time-dependent-manner. By decreasing the peak current and slowing the inactivation kinetics, IQM-266 led to an increase in the transmembrane charge (Q<sup>K</sup>V4.3/DREAM) at a certain range of concentrations. The slowing of the recovery process and the increase of the inactivation from the closed-state inactivation degree are consistent with a preferential binding of IQM-266 to a pre-activated closed state of KV4.3/DREAM channels. Finally, in rat dorsal root ganglion neurons, IQM-266 inhibited the peak amplitude and slowed the inactivation of IA. Overall, the results presented here identify IQM-266 as a new chemical tool that might allow a better understanding of DREAM physiological role as well as modulation of neuronal I<sup>A</sup> in pathological processes.

Keywords: KV4.3 channels, DREAM, DREAM ligands, KChIP, A-type current, Alzheimer

# INTRODUCTION

The Downstream Regulatory Element Antagonist Modulator (DREAM; Carrion et al., 1999), also known as KChIP3 (An et al., 2000) or calsenilin (Buxbaum et al., 1998), is a member of the K<sup>V</sup> channel interacting proteins (KChIPs) belonging to the neuronal calcium sensor (NCS) family (Burgoyne, 2007). DREAM is a 29 kDa protein with four EF-hand motif (EF1–4), conserved among other NCS members, in which the EF2 mediates low affinity Ca2<sup>+</sup> binding and is occupied by Mg2<sup>+</sup> under physiological conditions, whereas EF3 and EF4 mediate high affinity Ca2<sup>+</sup> binding (Bahring, 2018). The physiological roles of DREAM have been gradually revealed. In the nucleus, DREAM binds to a specific DRE to repress transcription of target genes (Carrion et al., 1999; Cheng et al., 2002; Ruiz-Gomez et al., 2007; Wu et al., 2010). Outside the nucleus, DREAM interacts with presenilins to modulate calcium release from the endoplasmic reticulum (Lilliehook et al., 2002). Additionally, the role of the downregulation of DREAM as part of an endogenous neuroprotective mechanism that improves ATF6 processing, neuronal survival in the striatum, and motor coordination in R6/2 mice, a model of Huntington's disease (HD), has been recently described (Naranjo et al., 2016; López-Hurtado et al., 2018). Besides, DREAM acts as a regulatory subunit of KV4.3 channels by inducing: (i) increased traffic of KV4.3 channels to the membrane; (ii) delayed inactivation kinetics; and (iii) accelerated activation and recovery kinetics from inactivation of KV4.3 channels (An et al., 2000). KV4 α-subunits, KChIPs and dipeptidyl aminopeptidase-like proteins (DPPs) form ternary complexes, which regulate the A-type outward potassium currents in neurons (IA; Nadal et al., 2003; Maffie and Rudy, 2008). I<sup>A</sup> is responsible for the fast repolarization of neuron action potentials and the frequency of firing, and thereby controls neuronal excitability (Birnbaum et al., 2004; Johnston et al., 2010). Among the KV4 α-subunits, KV4.2 and KV4.3 underlie the somatodendritic I<sup>A</sup> in the central nervous system (CNS; Huang et al., 2005), whereas, KV4.1 mRNA levels are lower than KV4.2 or KV4.3 (Serôdio and Rudy, 1998). Among other KChIPs, DREAM binding to KV4 channels regulates potassium currents, and hence neuronal excitability, in response to changes in intracellular calcium. Alterations in the function of the complexes KV4/KChIP and/or DREAM are implicated in different neuronal pathologies such as Alzheimer's (Hall et al., 2015) and HD (Naranjo et al., 2016), spinocerebellar ataxia (Smets et al., 2015) or epilepsy (Villa and Combi, 2016). Additionally, small molecules that bind to DREAM also modify channel function (Gonzalez et al., 2014; Naranjo et al., 2016). In this regard, repaglinide and CL-888 showed an inhibition of the I<sup>A</sup> (Naranjo et al., 2016), whereas NS5806 is the only described DREAM ligand showing a potentiation of I<sup>A</sup> under certain conditions (Witzel et al., 2012). Hence, it would be of great interest to have broader range of chemical tools that might allow a better understanding of the physiological role of DREAM and the modulation of neuron I<sup>A</sup> in pathological processes.

In this work, using surface plasmon resonance (SPR) assays and electrophysiological recordings of KV4.3/DREAM channels, we described IQM-266 as a new DREAM ligand able to inhibit the KV4.3/DREAM current in a concentration- and voltage-dependent manner, and to slow the activation and the inactivation kinetics. Blocking the peak current and slowing the inactivation kinetics led to an increase in the transmembrane charge (QKV4.3/DREAM) at a certain range of concentrations, identifying IQM-266 as a new activator of the DREAM-mediated KV4.3 currents. Importantly, I<sup>A</sup> recording from rat dorsal root ganglia (DRG) neurons revealed IQM-266 effects reminiscent of those observed in KV4.3/DREAM channels. Our findings offer new possibilities to control neuronal hyperexcitability by modulating the potassium outward current through KV4 channel complexes.

#### MATERIALS AND METHODS

All experiments shown in the present study were performed through the NIH rules (Guide for the care and use of laboratory animals; NIH publications number 23-80) revised in 2011; as well as the European Parliament 2010/63/EU and the rules of the Helsinki Declaration.

### IQM-266 Chemical Synthesis

3-(2-(3-Phenoxyphenyl)acetamido)-2-naphthoic acid: 2-(3 phenoxyphenyl)acetic acid (1.5 equiv.) in SOCl<sup>2</sup> (2 mL/mmol) was refluxed for 6 h, and the excess of thionyl chloride was evaporated to dryness. The residue was then dissolved in anhydrous THF (2 mL/mmol), and 3-amino-2-naphthoic acid (1.0 equiv.) and propylene oxide (15.0 equiv.) were added to the solution. After stirring overnight at room temperature, the solvent was evaporated to dryness and the crude residue was dissolved in EtOAc (3 × 10 mL), washed with brine (30 mL) and dried over Na2SO4. After removal of the solvent to dryness, the residue was triturated with Et2O, and the resultant solid subsequently triturated with CH3CN. The obtained solid was lyophilized to give a brown pale solid. m.p. 198.1–196.3◦C. <sup>1</sup>**H-NMR** (400 MHz, dimethyl sulfoxide (DMSO)-d6) δ (ppm): 3.81 (s, 2H, CH2CO), 6.94 (ddd, J = 8.2, 2.5, 0.9 Hz, 1H, H<sup>4</sup> 0), 7.03 (dt, J = 7.7, 1.1 Hz, 2H, H<sup>2</sup> <sup>00</sup>,6<sup>00</sup>), 7.07 (t, J = 2.5 Hz, 1H, H<sup>2</sup> <sup>0</sup>), 7.11 (tt, J = 7.7, 1.1 Hz, 1H, H<sup>4</sup> <sup>00</sup>), 7.17 (dt, J = 8.2, 0.9 Hz, 1H, H<sup>6</sup> <sup>0</sup>), 7.33–7.41 (m, 3H, H<sup>5</sup> 0 ,300,5<sup>00</sup>), 7.47 (ddd, J = 8.1, 6.9, 1.1 Hz, 1H, H6), 7.60 (ddd, J = 8.1, 6.9, 1.1 Hz, 1H, H5), 7.86 (d, J = 8.1 Hz, 1H, H7), 8.01 (d, J = 8.1 Hz, 1H, H4), 8.67 (s, 1H, H8), 8.94 (s, 1H, H3), 11.12 (s, 1H, NH). <sup>13</sup>**C-NMR** (100 MHz, DMSO-d6) δ (ppm): 44.3 (CH2CO), 116.6 (C3), 117.3 (C1), 117.4 (C<sup>4</sup> <sup>0</sup>), 118.5 (C<sup>2</sup> <sup>00</sup>,6<sup>00</sup>), 120.0 (C<sup>2</sup> <sup>0</sup>), 123.4 (C<sup>4</sup> <sup>00</sup>), 124.9, 125.6 (C<sup>6</sup> 0), 127.1, 128.2, 129.1, 129.3, 130.0 (C<sup>3</sup> <sup>00</sup>,5<sup>00</sup>), 130.2 (C<sup>5</sup> <sup>0</sup>), 133.1 (C7a), 135.5 (C3a), 136.0 (C2), 137.0 (C<sup>1</sup> <sup>0</sup>), 156.7 (C<sup>3</sup> 0 ,1<sup>00</sup>), 168.6 (CO2H), 169.4 (CH2CO). **HPLC** (Sunfire C18, gradient 50%–95% of acetonitrile in water, 10 min): t<sup>R</sup> = 7.04 min. **LC-MS**: 398.2 ([M + H]+). **HRMS (EI**+**)** m/z found 397.1306 ([M]<sup>+</sup> C25H19NO<sup>4</sup> calculated 397.1314).

#### Surface Plasmon Resonance (SPR): Binding Experiments

SPR experiments were performed at room temperature (20◦C) with a Biacore X-100 apparatus (Biacore, GE Healthcare Life Sciences) in running buffer (50 mM Tris pH 7.5, 50 mM NaCl, 2 mM CaCl<sup>2</sup> with 2% DMSO and 0.05% Tween 20). The protein GST-DREAM was immobilized on a CM5 sensor chip (Biacore, GE) following a standard amine coupling method (Johnsson et al., 1991). The carboxymethyl dextran surface of the experimental flow cell was activated with a 7-min injection of a 1:1 ratio of 0.4 M 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride (EDC) and 0.1 M N-hydroxysuccinimide. The protein was coupled to the surface with a 7-min injection at several dilutions at 10–100 µg/ml in 10 mM sodium acetate, pH 4.0. Unreacted N-hydroxysuccinimide esters were quenched by a 7-min injection of 0.1 M ethanolamine-HCl (pH 8.0). Immobilization levels were in the 7,000–8,000 RUs range. Reference flow cell was treated as experimental flow cell (amine coupling procedure) but without protein. Prior to use, 10 mM stock solutions of IQM-266 compound were diluted several times to a 1–7 µM final concentration in running buffer. Affinity measurements were made by a series of different concentrations injected onto the sensor chip at a 90 µl/min flow rate for 1 min, followed by a 1 min dissociation period. After dissociation, an extra wash was done over the flow cells with 50% DMSO. No regeneration was needed.

Sensograms data were double-referenced and solventcorrected using the BIAevaluation X-100 software (Biacore, GE Healthcare Life Sciences). Experimental data for affinity measurements were adjusted to a one site-specific model binding with Hill slope, using the equation: Req = Rmax[A] n / K n <sup>D</sup> + [A] n where Req is the equilibrium response at each concentration, Rmax is the maximum specific binding, [A] is the analyte concentration, K<sup>D</sup> the equilibrium dissociation constant and n the Hill slope.

#### Cellular Cultures and Transient Transfection

All experiments were performed in CHO-K1 (Chinese Hamster Ovary, CHO) cells obtained from the American Type Culture Collection (Rockville, MD, USA) and cultured at 37◦C in Iscove's modified Eagle's medium supplemented with 10% (v/v) fetal bovine serum (FBS), 1% (v/v) L-Glutamine (Gibco), and antibiotics (100 IU/ml penicillin and 100 mg/ml streptomycin; all from Gibco, Paisley, UK) in a 5% CO<sup>2</sup> atmosphere.

Cells were transiently cotransfected with KV4.3 (cloned into pEGFPn1, gently given by Dr. D.J. Synders, University of Antwerpen, Belgium) and DREAM (cloned into pcDNA3.1). In both cases, cells were cotransfected with EBO-pcDLeu2 as a reporter gene, codifying CD8. Transfection was performed using Fugene-6 (Promega) following manufacturer's instructions as previously described (Moreno et al., 2017; López-Hurtado et al., 2018). After 48 h transfection, cells were removed from culture plates using TrypLETM Express (Life Technologies), after exposing them to polystyrene microspheres bound to anti-CD8 (Dynabeads M-450, Thermo Fisher Scientific; Franqueza et al., 1999; Naranjo et al., 2016). Because the level of expression of DREAM can be crucial for the effects of IQM-266, only cells cotransfected with KV4.3 and DREAM that exhibit a recovery kinetics from inactivation between 20 and 45 ms were selected for electrophysiological recording.

#### Electrophysiology

Potassium currents elicited by the activation of KV4.3/DREAM channels expressed in CHO cells were recorded at room temperature (20–25◦C), at a frequency of 0.1 Hz using the whole-cell patch-clamp technique with an Axopatch 200B patchclamp amplifier (Molecular Devices) connected to an analogicdigital conversor (Digidata 1322A). Micropipettes were pulled from borosilicate glass capillary tubes (Narishige GD-1) on a programmable horizontal puller (Sutter Instrument Co.) and heat-polished with a microforge (Narishige, Japan). Micropipette resistance was 2–4 M. Data acquisition and genesis of experimental protocols were performed by the CLAMPEX utility of the PCLAMP 9.0.1 program (Molecular Devices). Currents were filtered at 2 kHz and sampled at 4 kHz (Bessel filter of 4 poles). Capacitance and series resistance compensation were optimized, with 80% compensation of the effective access resistance usually obtained. The intracellular pipette filling solution contained (in mM): 80 K-aspartate, 42 KCl, 3 phosphocreatine, 10 KH2PO4, 3 MgATP, 5 HEPES-K, 5 EGTA-K and it was adjusted to pH 7.25 with KOH. The bath solution contained (in mM): 136 NaCl, 4 KCl, 1.8 CaCl2, 1 MgCl2, 10 HEPES-Na and 10 glucose and it was adjusted to pH 7.40 with NaOH. IQM-266 was dissolved in DMSO at a stock concentration of 5 mM and added to the external solution at the desired concentration in each experiment. The currents were stored in a computer and analyzed with the CLAMPFIT utility of the PCLAMP 9.0.1 program and Origin 2018 (OriginLab Co.). Origin 2018 (OriginLab Co.) and Clampfit 10 programs were used to perform least-squares fitting and for data presentation (Valenzuela et al., 1994; Longobardo et al., 1998; Naranjo et al., 2016).

In order to obtain the concentration-response curve, block produced by IQM-266 was measured at the maximum peak and under the area of the current after applying different concentrations of the compound (0.01–100 µM) and thus, %block = 1 − IDrug IControl <sup>×</sup> 100. From the fitting of these values to a Hill equation, concentration-effect curves were generated, obtaining the values of the IC<sup>50</sup> and the Hill coefficient (n).

Activation and inactivation were fitted to a monoexponential process with an equation of the form y = Ae(<sup>−</sup> t τ ) + C, where τ represents the system time constant, A represents the amplitude of the exponential, and C is the baseline value. The voltage dependence of the activation curves was fitted with a Boltzmann equation: y = 1/ 1 + e (−(V−Vh)/s) , where s represents the slope factor, V represents the membrane potential, and V<sup>h</sup> represents the voltage at which 50% of the channels are open. Recovery from inactivativation was analyzed by applying a two pulse protocol consisting in a prepulse from −80 to +60 mV of 1 s in duration, followed by a test pulse to +60 mV of 250 ms in duration after different recovery time. The current measured at the maximum peak and the current in the test pulse were normalized vs. the first prepulse and they were plotted against the recovery time and then fitted to a monoexponential equation in order to obtain the τre (view ''Results'' section). In all cases, the control and the experimental condition was the same cell before and after being exposed to IQM-266.

### Isolation of DRG Neurons and Recording of Transient Potassium Currents (IA)

DRG neurons were isolated from male Sprague-Dawley rats (200–220 g/6–8 weeks old). Rats were sacrificed by cervical dislocation followed by decapitation, and lumbar segments of the spinal column were removed and placed in a cold Ca2+, Mg2+ free Hank's solution (Sigma-Aldrich). The bone surrounding the spinal cord was removed and L4, L5 and L6 DRG were exposed and pulled out. After removing the roots, DRG were chopped in half and incubated for 60 min at 37◦C in Dulbecco's modified Eagle's Medium-low glucose (DMEM; Sigma-Aldrich) containing 5 mg/mL collagenase XI (Worthington Biochemical, Lakewood, NJ, USA), 100 U/ml penicillin (Sigma-Aldrich), and 0.1 mg/ml streptomycin (Sigma-Aldrich). The cell suspension was then washed with DMEM by centrifugation (300 g, 5 min at 4◦C), filtered through a 100 µm mesh and washed again by centrifugation. The cell pellet was resuspended in DMEM and 40 µl were dropped onto 10 mm diameter glass coverslips treated with poly-D-lysine (1 mg/ml, 30 min; Sigma-Aldrich) and placed in 35 mm diameter Petri dishes. Finally, plated cells were flooded with 2.5 ml of DMEM and supplemented with 10% fetal calf serum (BioWhittaker, UK), 100 U/ml penicillin and 0.1 mg/ml streptomycin, stored in an incubator (Hera Cell, Heraeus, Germany) at 37◦C under a 5% CO2/95% air atmosphere. This protocol yields spherical cell bodies without neurites, from which only medium DRG neurons (30–40 µm diameter; 30–50 pF) were chosen for recording within 12–24 h of plating.

Current recordings were performed at room temperature (21–24◦C) in the perforated-patch variant of the whole-cell configuration of the patch-clamp technique, with an EPC10 amplifier using PatchMaster software (HEKA Electronic, Lambrecht, Germany; Carabelli et al., 2003). Patch pipettes were pulled from borosilicate glass to have a final resistance of 5.5–6.5 MΩ when filled with internal solution (see below). Membrane currents were filtered at 3 kHz and sampled at 10 kHz from cells held at a voltage of −80 mV. Series resistance (<20 MΩ) was compensated by 80% and monitored together with the cell membrane capacitance throughout the experiment. The perforated-patch configuration was obtained using amphotericin B (Sigma-Aldrich) dissolved in DMSO and stored at −20◦C in aliquots of 50 mg/ml. The pipette-filling solution contained (mM) 90 K2SO4, 55 KCl, 8 NaCl, 1 MgCl2, 15 HEPES (pH 7.2 with KOH; ≈280 mOsm). Fresh pipette solution was prepared every 2 h. The bath solution containing (mM) 145 NaCl, 2.8 KCl, 2 CaCl2, 1 MgCl2, 10 HEPES, and 10 glucose (pH 7.4 adjusted with NaOH; ≈300 mOsm) was constantly superfused at a rate of approximately 1 ml/min. IQM-266, at the desired concentration (3 or 10 µM), was directly applied onto the cell under investigation by gravity from a multibarrel concentration-clamp device coupled to electronically driven miniature solenoid valves. The common outlet of this was placed within 100 µm of the cell to be patched.

Transient potassium current (IA) was isolated by using voltage protocols, taking advantage of the distinct voltage-dependent inactivation of the underlying potassium channels (Phuket and Covarrubias, 2009). First, total voltage-activated potassium current was measured in cells held at −80 mV, in which a 1-s conditioning pulse to −100 mV was delivered prior to 250-ms step depolarizations, ranging from −20 to +20 mV in 10 mV increments. Then, a depolarizing 1-s conditioning pulse to −20 mV was applied, sufficient to inactivate IA, such that the outward current evoked by subsequent step depolarizations was mostly comprised of a delayed rectifying potassium current. I<sup>A</sup> was finally revealed by subtracting the delayed rectifying current from total current. Peak amplitude of I<sup>A</sup> was used to determine percent block by IQM-266. I<sup>A</sup> inactivation was fitted to a biexponential process with an equation of the form:

$$\gamma = \left. \gamma\_0 + A\_f \exp\left(\frac{-\left(\mathbf{x} - \mathbf{x}\_0\right)}{\mathbf{r}\_f}\right) + A\_s \exp\left(\frac{-\left(\mathbf{x} - \mathbf{x}\_0\right)}{\mathbf{r}\_s}\right) \right|$$

where τ<sup>f</sup> and τ<sup>s</sup> represent the fast and slow time constants, respectively, and A<sup>f</sup> and A<sup>s</sup> represent the amplitudes of the corresponding kinetic components.

#### Statistics

Data are expressed as the mean ± SEM. Direct comparisons between mean values in control conditions and in the presence of drug for a single variable were performed by paired Student's t-test. Differences were considered significant if the p value was less than 0.05.

#### RESULTS

#### Identification of IQM-266 as DREAM Ligand Modulator of KV4.3/DREAM Channel Activity

As part of our program of medicinal chemistry searching for DREAM ligands, we performed a high throughput screening of our chemical libraries to evaluate their binding to DREAM using SPR. We selected the novel compound 3-(2-(3- Phenoxyphenyl)acetamido)-2-naphthoic acid, IQM-266, which showed a K<sup>D</sup> = 4.63 ± 0.73 µM, and we have investigated its effect on the KV4/DREAM channels currents (**Figure 1**; for synthesis and full characterization of IQM-266 see ''Materials and Methods'' section).

**Figure 2A** shows the concentration-dependence of block produced by IQM-266 on KV4.3/DREAM channels when measured at the maximum peak current. By fitting these data to a Hill equation, the concentration that inhibited 50% of the channels (IC50) and the Hill coefficient (n) were calculated (8.6 µM and 0.75, respectively, n = 34). **Figure 2A** also shows the Hill fit in which we assume that this curve exhibits a basal value of 0 and a maximum percentage block of 100 (dashed line). The IC<sup>50</sup> and the n obtained with this fit were very similar to those obtained without any constriction (9.0 µM and 1.0, respectively). The n value obtained led us to conclude that binding of IQM-266 to KV4.3/DREAM channels does not exhibit cooperativity. **Figure 2B** exhibits a bar graph in which the effects of this compound on the maximum peak current and on the charge (measured as the integral of the current recordings) are shown. At all concentrations tested, IQM-266 decreased the maximum peak current and, to a lesser extent, the charge through the membrane, these differences being very marked between 1 and 10 µM. Indeed, at 3 µM, IQM-266 decreased the maximum peak current (33.7 ± 5.3%, n = 6), but increased the charge (13.0 ± 4.1%, n = 6), which can be explained by the slowing

effect on the inactivation kinetics produced by this compound (**Figure 2C**). IQM-266 slows down both the activation and the inactivation kinetics of the current in a concentration-dependent manner. Therefore, the equilibrium between the decrease in the maximum peak current and the slowing of the inactivation process lead, either to an increase (at 3 µM) or to a decrease in the charge (concentrations >3 µM; **Figure 2D**). In order to characterize both effects: the increase and the inhibition of the charge through KV4.3/DREAM channels, two different concentrations of IQM-266 were used, 3 and 10 µM (close to the IC50), respectively.

# Time Dependent Effects of IQM-266 on KV4.3/DREAM Channels

The activation kinetics of KV4.3/DREAM current in the absence and in the presence of IQM-266 was analyzed by fitting the traces to a monoexponential equation, from which the activation time constant (τAct) was obtained. The inactivation process was also fitted to a monoexponential curve after applying a 250 ms depolarizing pulse from −80 mV to +60 mV, from which the time constant of inactivation (τInac) of the KV4.3/DREAM, in the absence and in the presence of IQM-266, were obtained (**Figure 3**). **Figures 3A,B** show the first 25 ms of the normalized currents after applying a depolarizing pulse from −100 to +60 mV in the absence and in the presence of 3 or 10 µM IQM-266. In order to analyze the concentration-dependence of this slowing in the activation kinetics, the τAc,IQM-266/τAc,Control ratio vs. IQM-266 concentration was plotted (**Figure 3C**). As it can be observed, the slowing effect induced by this compound was concentration-dependent. **Figure 3D** shows the absolute values under control and in the presence of IQM-266 at 3 and 10 µM, respectively. This slowing effect on the activation kinetics was observed at membrane potentials positive to 0 mV (p < 0.05). This compound also slows the inactivation process at membrane potentials positive to 0 mV (p < 0.05), therefore, this slowing is concentration-dependent (**Figures 3F–J**).

#### Effects of IQM-266 on the Recovery Kinetics of Inactivation of KV4.3/DREAM Channels

In order to analyze the recovery from inactivation of KV4.3/DREAM channels, a double pulse protocol was applied (**Figures 4A,B**, upper panel), consisting in a conditioning prepulse from −80 to +60 mV of 1 s in duration (I0) in order to inactivate most of the channels, followed by a test pulse applied after a variable interpulse (between 10 and 800 ms at −90 mV) to +60 mV (It). This pulse protocol was applied before and after perfusing the cells with 3 µM or 10 µM IQM-266. The ratio It/I<sup>0</sup> measured at the maximum peak was plotted vs. the time interpulse between the end of I<sup>0</sup> and the application of I<sup>t</sup> . In all experimental conditions, data were fitted to a monoexponential function, from which the time constant of recovery (τre) was obtained. Under control conditions, the τre arose a mean value of 42.7 ± 4.5 ms (n = 8). IQM-266 slowed the recovery process of the KV4.3/DREAM current in a concentration-dependent manner (from 46.8 ± 5.3 to 109.2 ± 13.0 ms in control and in the presence of IQM-266 3 µM, n = 5, p < 0.01 and from 37.7 ± 6.9 to 153.6 ± 26.3 ms in control and in the presence of IQM-266 10 µM n = 4 p < 0.05; **Figures 4C,D**). Interestingly, both concentrations of IQM-266 induced an overshoot in the recovery process similar to that observed in cardiac Ito, and that has been attributed to KCNE2 effects (Wettwer et al., 1993; Zhang et al., 2001; Radicke et al., 2008).

# Voltage Dependence Effects of IQM-266 on KV4.3/DREAM Channels

**Figure 5A** shows superimposed current traces obtained in the absence and in the presence of IQM-266 at 3 or 10 µM. After

plotting the maximum peak current obtained under control and after perfusion with external solution containing 3 µM or 10 µM IQM-266, the current-voltage (I-V) relationships were obtained (**Figure 5B**). These two plots also show the ratio IIQM-266/IControl at both concentrations (blue triangles), together with the activation curve (dashed line). A block of the maximum peak current produced by IQM-266 3 µM increased in the range of activation of KV4.3/DREAM channels but remained constant at membrane potentials positive to +10 mV. At higher concentrations, the maximum block was obtained at +20 mV (63.1 ± 3.6%, n = 4) and it decreased only when the more positive potential was applied (58.8 ± 3.0% at +60 mV, n = 4, p < 0.05). **Figure 5C** shows the charge-voltage relationships (Q-V) obtained after plotting the charge under the current obtained in control and in the presence of IQM-266 3 or 10 µM. These plots also show the relative charge (blue triangles) at each membrane potential in order to analyze the voltage dependence of block. IQM-266 at 3 µM increased the charge in

a voltage independent manner. The maximum block (measured at the charge) produced by IQM-266 at 10 µM was observed at −20 mV and this block decreased in a voltage dependent manner (45.8 ± 6.7% vs. 23.5 ± 4.5% at −20 mV and +60 mV, respectively, n = 4, p < 0.05). Importantly; both the increase of the charge and the decrease in the maximum peak current were observed at all membrane potentials.

The activation curves were obtained from the I-V relationships. Data were plotted against membrane potential to which each current record was generated and fitted to a Boltzmann equation, in order to obtain the V<sup>h</sup> and s values. IQM-266, at 3 µM, did not shift the activation curve (V<sup>h</sup> = +4.3 ± 2.6 mV and +7.8 ± 3.9 mV in the absence and in the presence of IQM-266, n = 5, p > 0.05; s = 16.1 ± 0.3 mV and 19.6 ± 0.6 mV, n = 5, p > 0.05).

In order to study the voltage dependence of inactivation of KV4.3/DREAM channels, a double pulse protocol consisting in a 250 ms conditioning pulse to different potentials between

FIGURE 3 | Activation and inactivation kinetics of KV4.3/DREAM currents in the absence and in the presence of IQM-266. (A) Normalized first 20 ms of original current recordings in control and in the presence of IQM-266 (3 µM). The current records were fitted to a monoexponential equation in order to obtain the activation time constant (τAc). (B) Normalized first 20 ms of original current recordings in control and in the presence of IQM-266 (10 µM). The current records were fitted to a monoexponential equation in order to obtain the τAc values. (C) Concentration-dependence of the τAc,IQM-266/τAc,Control. (D) Histogram representing the τAc in the different experimental conditions. (E) Voltage-dependent effects of IQM-266 (3 µM) on the time constant of activation (τAct). (F) Normalized first 250 ms of original current recordings in control and in the presence of IQM-266 (3 µM). The current records were fitted to a monoexponential equation in order to obtain the inactivation time constant (τInac). (G) Normalized first 250 ms of original current recordings in control and in the presence of IQM-266 (10 µM). The current records were fitted to a monoexponential equation in order to obtain the τInac values. (H) Concentration-dependence of the τInac, IQM-266/τInac,Control. (I) Histogram representing the τInac in the different experimental conditions. (J) Voltage-dependent effects of IQM-266 (3 µM) on the time constant of inactivation (τInact). <sup>∗</sup>p < 0.05, ∗∗p < 0.01 and ∗∗∗p < 0.001 when comparing data in presence of IQM-266 with control; #p < 0.05 when comparing data obtained in the presence of IQM-266 (3 µM) and IQM-266 (10 µM).

−110 and +60 mV, followed by a test pulse to +50 mV of 250 ms in duration, was applied (**Figure 6A**). The maximum peak currents measured at the test pulse were plotted against the membrane potential of the previous conditioning pulse, and the data were fitted to a Boltzmann equation. IQM-266, at 3 µM, shifted the inactivation curve to negative potentials (V<sup>h</sup> = −30.1 ± 0.7 mV and −38.0 ± 0.9 mV in the absence and in the presence of IQM-266, n = 4, p < 0.01; s = 4.5 ± 0.2 mV and 6.6 ± 0.7 mV, n = 4, p > 0.05; **Figure 6C**).

Since KV4.3 channels inactivate predominantly from the closed state (Campbell et al., 1993; Beck and Covarrubias, 2001), the negative shift induced by IQM-266 of the steadystate inactivation curve is indicative of an acceleration of the closed-state inactivation. In order to elucidate this issue, a double pulse protocol was applied. A pre-pulse test to +60 mV during 100 ms was immediately preceded by a pulse to −40 mV of variable duration (**Figure 6B**). As it can be observed, IQM-266 increased the degree of closed-state inactivated channels, thus suggesting that IQM-266 promotes inactivation from the closedstate (**Figure 6D**).

# Effects of IQM-266 on KV4.3 Channels

In order to analyze the selectivity of IQM-266 for KV4.3/DREAM over KV4.3 channels, the effects of IQM-266 were studied on KV4.3 channels expressed in CHO cells in the absence of DREAM. As it is shown in the **Figure 7**, the IC<sup>50</sup> value obtained when measured at the maximum peak current was very close to that observed in KV4.3/DREAM channels (7.1 µM vs. 8.6 µM, n = 21–23). Also, IQM-266 slowed the activation kinetics (τ = 0.73 ± 0.09 ms vs. 1.92 ± 0.32 ms, in control and in the presence of IQM-266, respectively, n = 8,

channels in the absence and in the presence of IQM-266 after applying the pulse protocols shown in the upper part of the figure. (B) I-V relationship of the currents generated by KV4.3/DREAM channels in the absence and in the presence of IQM-266 at 3 µM (left) and 10 µM (right) when measured at the maximum peak current. It is also shown the relative current (IIQM-266/IControl) vs. membrane potential. Dotted and dashed lines show the activation and the inactivation curves, respectively (n = 4–6). (C) Q-V relationship of the charge through KV4.3/DREAM channels in the absence and in the presence of IQM-266 at 3 µM (left) and 10 µM (right) when measured at the area under the current during the application of depolarizing pulse protocol shown in the upper part of the figure. It is also shown the relative charge (QIQM-266/QControl) vs. membrane potential. Dotted and dashed lines show the activation and the inactivation curves, respectively (n = 4–6).

p < 0.01), as well as the inactivation kinetics. In fact, the latter process that exhibits a biexponential decay under control conditions (τ<sup>f</sup> = 19.6 ± 2.4 ms and τ<sup>s</sup> = 76.2 ± 9.6 ms, n = 14) becomes monoexponential in the presence of IQM-266 (τ = 62.3 ± 4.7 ms, n = 14, p > 0.05 vs. τ<sup>s</sup> value in control conditions; **Figures 7C–E**). However, IQM-266 did not increase the charge at any concentration tested (**Figure 7B**). In contrast to what occurs in the presence of DREAM, IQM-266 did not modify the recovery kinetics of inactivation (**Figure 7F**). **Figure 7G** shows the two I-V relationships obtained in the absence and in the presence of IQM-266, together with the ratio IIQM-266/IControl (blue triangles) and the activation curve (dashed line). A block of the maximum peak current produced by IQM-266 increased in the range of activation of KV4.3 channels but remained constant at membrane potentials positive to +10 mV. IQM-266 did not shift the inactivation curve to more negative potentials (**Figure 7H**), and it did not modify the closed-state inactivation (**Figures 7H,I**). All these results indicate that, although this compound also binds to KV4.3 channels, the increase in the charge observed in KV4.3/DREAM channels and induced by IQM-266 is due to its specific interaction with DREAM. Moreover, this interaction seems to prevent the effect of DREAM on the recovery from inactivation.

# Effect of IQM-266 on I<sup>A</sup> From DRG Neurons

DRG neurons are known to express DREAM as well as KV4.3 channels, which contribute to I<sup>A</sup> (Phuket and Covarrubias, 2009; Tsantoulas and McMahon, 2014; Tian et al., 2018). Hence, we decided to record I<sup>A</sup> from rat DRG neurons in order to evaluate the effect of IQM-266 on native potassium channels. By using a voltage protocol designed to isolate IA, we recorded transient, fast activating and inactivating potassium currents in the voltage-range in which I<sup>A</sup> makes a substantial contribution to voltage-dependent potassium currents (−20 mV to +20 mV) (**Figure 8A**). Isolated currents were sensitive to 4-aminopyridine 5 mM (data not shown) and displayed inactivation kinetics that required the sum of two exponential terms for an adequate description. Time constant values and relative amplitude of the fast and slow kinetic components at 0 mV were τ<sup>f</sup> = 5.1 ± 0.6 ms (45%) and τ<sup>s</sup> = 86.1 ± 7.5 ms (55%, n = 13 cells). As previously reported, time constants exhibited weak voltage-dependence (**Figure 8B**), and the relative weight of the two components barely changed in the voltage range that we studied (data not shown; Phuket and Covarrubias, 2009).

Interestingly, IQM-266 at 3 and 10 µM reduced the peak amplitude of I<sup>A</sup> dose dependently. Percentage block showed slight voltage dependence, increasing with the depolarization (**Figures 8A,B**). Likewise, IQM-266 at 10 µM slowed current inactivation by increasing both τ<sup>f</sup> and τ<sup>s</sup> . This effect reached statistical significance on τ<sup>s</sup> and developed in a voltagedependent manner (at potentials equal and positive to 0 mV; **Figures 8C,D**).

#### DISCUSSION

In the present study the effects of IQM-266 on recombinant KV4.3 and KV4.3/DREAM channels expressed in mammalian cells, as well as on I<sup>A</sup> from DRG neurons, have been analyzed. We demonstrate that this new compound: (1) binds to KV4.3/DREAM channels in a concentration-, time- and voltagedependent manner; (2) inhibits the maximum peak current and, to a lesser extent, the charge crossing the cell membrane during depolarization; (3) at certain concentrations (3 µM), IQM-266 increases the charge through the cell membrane during the application of depolarizing pulses; and (4) inhibits peak I<sup>A</sup> amplitude while slowing its inactivation. Overall, the results presented here are consistent with a preferential binding of IQM-266 to a pre-activated closed state of KV4.3/DREAM channels.

In the last decade, there small molecules have been developed that bind to DREAM and modify KV4 channel function (Gonzalez et al., 2014; Naranjo et al., 2016). Among them, repaglinide and CL-888 were shown to inhibit IA, whereas NS5806 would be the only DREAM ligand able to potentiate this

with control.

conduct. Note that IQM-266 does not modify the closed-state inactivation. <sup>∗</sup>p < 0.05, ∗∗p < 0.01 and ∗∗∗p < 0.001 when comparing data in presence of IQM-266

sort of potassium current (Witzel et al., 2012; Gonzalez et al., 2014). Like IQM-266, NS5806 slowed down the inactivation decay of neuronal I<sup>A</sup> and slightly decreased the maximum peak current.

The two main reported effects of DREAM on KV4.3 channels are: (i) an increased traffic of KV4.3 channels to the membrane; and (ii) an acceleration of the activation and recovery kinetics from inactivation (An et al., 2000; Naranjo et al., 2016). In the present study, we show that IQM-266 produces the opposite effects to those induced by DREAM: slowing of the activation and recovery from inactivation kinetics, which might be attributed to IQM-266 binding to DREAM, hence supporting the results of the SPR experiments. IQM-266 interacts with KV4.3/DREAM channels in a concentration-, time- and voltage-dependent manner, consistent with binding preferentially to a pre-activated closed state of the channels and with very low or no affinity for the open state. There are several pieces of evidence supporting this mechanism of action: (1) the maximum degree of block (produced by 10 µM IQM-266, a concentration close to its IC50) measured at the maximum peak and at the charge, was obtained at −10 mV, a potential at which most of the channels are closed or in pre-activated, closed states; (2) block steeply increased in the activation range of KV4.3/DREAM channels, achieving a maximum and plateau level at potentials positive to +20 mV, where the probability of channels opening increases; (3) IQM-266 slows the activation of the KV4.3/DREAM current. Also, IQM-266 slows the recovery process from inactivation, suggesting that its binding to KV4.3/DREAM channels promotes inactivation; and (4) in fact, IQM-266 negatively shifted the inactivation curve, as well as the closed-state inactivation.

percent block of peak I<sup>A</sup> by IQM-266 (3 µM or 10 µM) at different potentials. Data were obtained from eight cells for IQM-266 (3 µM), and five cells for IQM-266 (10 µM). (C) Effect of IQM-266 (3 µM; upper panels) and (10 µM; lower panels) on inactivation kinetics of IA. Original recordings (left panels) have been normalized to peak I<sup>A</sup> in the absence of IQM-2666 (right panels) to better appreciate the change in inactivation kinetics. (D) Bar graph showing the effect of IQM-266 at 3 µM (upper graph) or 10 µM on time constants of inactivation at different potentials. The current records were fitted to a biexponential equation to obtain the τ<sup>f</sup> and τ<sup>s</sup> values. Data are from eight cells for IQM-266 (3 µM), and five cells for IQM-266 (10 µM). <sup>∗</sup>p < 0.05 with regard to Control.

Since KV4.3 channels mostly inactivate from the closed states (Beck and Covarrubias, 2001), this result is consistent with the interaction of IQM-266 with a closed or pre-activated closed state of the channels (Snyders et al., 1992; Longobardo et al., 1998).

Furthermore, IQM-266 inhibits KV4.3 current in the absence of DREAM. However, the KV4.3 IQM-266 interaction exhibits differential features. In the presence of DREAM, IQM-266 prevents the effect of this regulatory subunit on the channel (acceleration of: (i) recovery from inactivation and activation; (ii) slower decay kinetics; and (iii) less prominent closed-state inactivation), yet in the absence of DREAM, IQM-266 decreased the peak potassium current and slowed the activation and the inactivation kinetics.

However, the more striking effect produced by IQM-266 is the slowing of inactivation kinetics. Indeed, this effect may explain why at concentrations lower than the IC50, IQM-266 augments the efflux of potassium ions resulting in an increase in charge (activating effect). Importantly, this increase in the charge is observed at membrane potentials positive to +10 mV. This effect is more evident at concentrations at which the inhibition of the maximum peak current is negligible, but still capable of slowing the inactivation decay. This effect could be the basis of a promising therapeutic strategy for the treatment of certain pathologies affecting cardiac (cardiac arrhythmias) or neuronal (epilepsy, Alzheimer disease or ataxia) cells, in which a downregulation of KV4.3 or DREAM has been demonstrated (Huo et al., 2014; Hall et al., 2015; Smets et al., 2015; Villa and Combi, 2016). IQM-266 also modulated I<sup>A</sup> from rat DRG neurons. At 10 µM, IQM-266 effects on I<sup>A</sup> were reminiscent of those observed on heterologously expressed KV4.3/DREAM channels. Hence, IQM-266 10 µM inhibited peak IA, and this effect increased with the depolarization in the physiological range of activation of the current. Likewise, IQM-266 10 µM slowed inactivation kinetics at potentials positive to 0 mV. In contrast, no facilitation of I<sup>A</sup> could be observed with IQM-266 3 µM. At present, we do not have an explanation for this result except the fact that DRG neurons express other potassium channel regulatory proteins in addition to DREAM (David et al., 2012; Cheng et al., 2016; Tian et al., 2018), which may prevent the potentiating effect seen at 3 µM IQM-266 on KV4.3/DREAM channels. Notwithstanding, our results in DRG neurons suggest that IQM-266 constitutes a small, novel chemical molecule suitable to modulate KV4.3 channels in native systems.

Different neuronal (Hall et al., 2015; Gross et al., 2016) or cardiac pathologies are related to abnormalities in the function of different ion channels and/or regulatory subunits, such as KV4.3 and the regulatory subunit DREAM. Thus, KChIPs starts

#### REFERENCES


to emerge as a realistic drug target, and IQM-266 could be considered as a new chemical tool that might allow a better understanding of: (i) DREAM physiological role; and (ii) the modulation of I<sup>A</sup> in pathological conditions.

#### AUTHOR CONTRIBUTIONS

DP, PC, LL, PM, YM, PS, CIG, SS and AL-H performed the experiments and analyzed the data. MM-M, LO-O, AA, JN, MG-R and CV conceived the study, analyzed the data and wrote the article.

#### FUNDING

PC was the recipient of a postgraduate FPI fellowship from the Spanish Ministry of Economy, Industry and Competitivity (MINECO). This work was funded by the Spanish Ministry of Economy, Industry and Competitivity (Ministerio de Economía y Competitividad; AEI-FEDER, EU grants): SAF2012-32209 and BFU2015-67284-R (to MG-R), SAF2014-53412-R and SAF2017- 89554-R (to JN), SAF2013-45800-R, SAF2016-75021-R (to CV) and SAF2015-66275-C2-2-R (to MM-M); Universidad Complutense de Madrid (UCM) grant: PR75/18-21593 (to AA); the Instituto de Salud Carlos III CIBERNED and CIBERCV programs (to JN and to CV, respectively) and the Madrid regional government/Neurodegmodels (to JN); Consejo Superior de Investigaciones Científicas (CSIC) grants: PIE 201820E104 (to CV) and 201880E109 (to MG-R and MM-M). We acknowledge support of the publication fee by the CSIC Open Access Publication Support Initiative through its Unit of Information Resources for Research (URICI).

#### ACKNOWLEDGMENTS

We want to express our thanks to Dr. Snyders for kindly providing us with the cloned pEGFPn1, University of Antwerpen (Belgium) and the Technical Services of the Instituto de Investigaciones Biomedicas Alberto Sols (CSIC-UAM), Madrid, Spain.

presenilins and regulates the levels of a presenilin fragment. Nat. Med. 4, 1177–1181. doi: 10.1038/2673


channel Kv4-positive nociceptors and pain-modulating spinal interneurons. J. Comp. Neurol. 524, 846–873. doi: 10.1002/cne.23876


**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 © 2019 Peraza, Cercos, Miaja, Merinero, Lagartera, Socuéllamos, Izquierdo García, Sánchez, López-Hurtado, Martín-Martínez, Olivos-Oré, Naranjo, Artalejo, Gutiérrez-Rodríguez and Valenzuela. 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) and the copyright owner(s) 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.

# Caldendrin and Calneurons— EF-Hand CaM-Like Calcium Sensors With Unique Features and Specialized Neuronal Functions

Jennifer Mundhenk <sup>1</sup> , Camilla Fusi <sup>1</sup> and Michael R. Kreutz 1,2 \*

<sup>1</sup>RG Neuroplasticity, Leibniz Institute for Neurobiology, Magdeburg, Germany, <sup>2</sup>Leibniz Group "Dendritic Organelles and Synaptic Function", Center for Molecular Neurobiology, ZMNH, Hamburg, Germany

The calmodulin (CaM)-like Ca<sup>2</sup><sup>+</sup>-sensor proteins caldendrin, calneuron-1 and -2 are members of the neuronal calcium-binding protein (nCaBP)-family, a family that evolved relatively late during vertebrate evolution. All three proteins are abundant in brain but show a strikingly different subcellular localization. Whereas caldendrin is enriched in the postsynaptic density (PSD), calneuron-1 and -2 accumulate at the trans-Golgi-network (TGN). Caldendrin exhibit a unique bipartite structure with a basic and proline-rich N-terminus while calneurons are the only EF-Hand CaM-like transmembrane proteins. These uncommon structural features come along with highly specialized functions of calneurons in Golgi-to-plasma-membrane trafficking and for caldendrin in actinremodeling in dendritic spine synapses. In this review article, we will provide a synthesis of available data on the structure and biophysical properties of all three proteins. We will then discuss their cellular function with special emphasis on synaptic neurotransmission. Finally, we will summarize the evidence for a role of these proteins in neuropsychiatric disorders.

#### Edited by:

Teresa Duda, Salus University, United States

#### Reviewed by:

Ivan Ivanovich Senin, Lomonosov Moscow State University, Russia Maria E. Rubio, University of Pittsburgh, United States

#### \*Correspondence:

Michael R. Kreutz kreutz@ifn-magdeburg.de

Received: 16 October 2018 Accepted: 17 January 2019 Published: 06 February 2019

#### Citation:

Mundhenk J, Fusi C and Kreutz MR (2019) Caldendrin and Calneurons— EF-Hand CaM-Like Calcium Sensors With Unique Features and Specialized Neuronal Functions. Front. Mol. Neurosci. 12:16. doi: 10.3389/fnmol.2019.00016 Keywords: golgi, trafficking, dendritic spines, F-actin, structural plasticity, CaV1.2, InsP3R

#### INTRODUCTION

Neuronal Ca2+-signaling is based on spatio-temporal gradients called Ca2<sup>+</sup> waves, spikes, transients and puffs. The strict segmentation of such gradients allows for complex signaling events at the micro- and even nanoscale (Berridge et al., 2003). Consequently, the huge variety of Ca2+ evoked processes require a highly specialized machinery leading to alterations in cellular functions and Ca2<sup>+</sup> sensor proteins from the calmodulin (CaM) family are instrumental in this regard.

### THE nCaBP-FAMILY OF EF-HAND CaM-LIKE CALCIUM SENSOR PROTEINS

The predominant neuronally expressed Ca2+-binding proteins of the neuronal calcium sensor (NCS)- and neuronal calcium-binding protein (nCaBP)-family share the same EF-hand organization as their ancestor CaM (**Figure 1**). In contrast to CaM, however, these proteins exhibit a more restricted and cell type specific expression. They are frequently targeted to subcellular compartments, contain at least one cryptic EF-hand incapable of Ca2+-binding, while many of

FIGURE 1 | Schematic diagram depicting the neuronal calcium sensor (NCS) and neuronal calcium-binding proteins (nCaBPs) families. Analogous to their ancestral protein calmodulin (CaM), all nCaBPs and NCS-proteins harbor four EF-hands out of which at least one is not capable of binding Ca2<sup>+</sup> and is therefore referred to as non-functional (green). All NCS family members

different conformational response upon Ca2+-binding than CaM. It is nowadays widely believed that NCS and nCaBP family members are involved in distinct signaling processes and thereby increase the versatility of the Ca2+-signaling tool kit (Mikhaylova et al., 2011). NCS proteins are evolutionarily related to Frequenin and, on the basis of their sequences, they have been grouped into five subfamilies (Burgoyne, 2007; Mikhaylova et al., 2011), comprising Frequenin/NCS-1, Recoverin, VILIPs 1–3/Hippocalcin/Neurocalcin delta, GCAP1–3 and the voltage-gated K<sup>+</sup> (Kv) channel-interacting proteins (KChiPs) 1–4/DREAM (**Figure 1**).

The nCaBP family consists of six family members, caldendrin, L-CaBP1, S-CaBP1, short- and long-calneuron-1, calneuron-2 (**Figure 1**). Caldendrin harbors a cryptic EF-hand 2, whereas in calneuron-1 and -2 EF-hands 3 and 4 are non-functional (**Figure 1**; Mikhaylova et al., 2006; McCue et al., 2010). This excludes interdomain cooperativity in Ca2+-binding like it was reported for other CaM-like EF-hand Ca2+-sensor proteins (Fefeu et al., 2000; Heidarsson et al., 2013; Ranaghan et al., 2013) including caldendrin (Kiran et al., 2017). Two shorter isoforms arise from alternative splicing of the caldendrin/CaBP1 gene that are denoted L-CaBP1 and S-CaBP1, respectively. Similar to caldendrin, L-CaBP1 and S-CaBP1 harbor a non-functional EF-hand 2 (Haeseleer et al., 2000; Laube et al., 2002). Caldendrin is prominently expressed in brain, whereas S- and L-CaBP1 like all other CaBPs are mainly abundant in retina and the cochlea (Seidenbecher et al., 1998; Haeseleer et al., 2000; Laube et al., 2002; Landwehr et al., 2003; Kim et al., 2014). Thus, in addition to caldendrin, from the nCaBP family only calneuron-1 and -2 are abundant in brain (Hradsky et al., 2015).

# STRUCTURAL FEATURES AND BIOPHYSICAL PROPERTIES OF CALDENDRIN

Caldendrin exhibits a unique bipartite structure with a N-terminal half that contains seven phosphorylation sites with an unusual high proline content (13%) and multiple PxxP motifs (**Figure 1**). The N-terminus shows no sequence homology to other family members. The shorter splice isoforms share the C-terminus with caldendrin (**Figure 1**) that resembles the structure of CaM (**Figure 1**). In contrast to CaM, however, the first EF-hand motif will bind both Mg2<sup>+</sup> and Ca2<sup>+</sup> while the second EF-hand is not capable of Ca2<sup>+</sup> binding. The global affinity of Ca2<sup>+</sup> and Mg2<sup>+</sup> binding was found to be in the low µM range (Wingard et al., 2005; Reddy et al., 2014). Since the Mg2+-binding affinity is clearly higher than in S-CaBP1 it is likely that the extended N-terminus will impact Mg2+-binding to the first EF-hand (Wingard et al., 2005; Reddy et al., 2014). Several lines of evidence suggest complex intra- and intermolecular interactions of caldendrin (Reddy et al., 2014). However, the protein shows very little change conformational change as evidenced by alterations in surface hydrophobicity and secondary as well as tertiary structure upon Ca2+-binding to the Mg2+-saturated protein (Reddy et al., 2014). Caldendrin is in the C-terminal part the closest homolog of CaM in brain and shares with its ancestor a flexible linker region between both EF-hand domains. But in case of caldendrin insertion of four additional amino acids will allow for a more flexible orientation of both domains that might allow for binding to sterically more complex target interactions (Seidenbecher et al., 1998; Haeseleer et al., 2000; Laube et al., 2002). It was shown that the folding of the EF-hand domains can occur independently and in any order in CaM and probably also all other nCaBP. Unfortunately, still very little is known about the structure-function relationships in relation to ion binding properties of this sensor. Although NMR and crystal structures have been reported for the shorter isoforms (Li et al., 2009; Findeisen and Minor, 2010), the unique N-terminal region and reports pointing towards functional differences between caldendrin and the shorter

splice isoforms asks for an independent analysis of their function (Tippens and Lee, 2007; Findeisen and Minor, 2010). Along these lines the rather complex inter- and intramolecular interactions in Caldendrin that are regulated by Ca2+-binding, the exceptionally high Mg2+- and the relatively low Ca2+ binding affinity, the rigid first EF-hand domain and the rather modest changes in confirmation upon binding of Ca2<sup>+</sup> make it plausible that the association with binding partners might follow different molecular principles in comparison to other neuronal Ca2+-sensors.

#### THE CALDENDRIN INTERACTOME AND CELLULAR FUNCTIONS OF CALDENDRIN

Caldendrin is enriched at synapses and in the somatodendritic compartment of pyramidal neurons in cortex and hippocampus (Seidenbecher et al., 1998; Bernstein et al., 2007). Synaptic [Ca2+]<sup>i</sup> levels can be increased by activation of L-type voltage activated Ca2<sup>+</sup> channels like CaV1.2, N-methyl-D-aspartate receptors (NMDARs) or Inositol trisphosphate receptors (InsP3Rs; Sala and Segal, 2014). It is well established that the activity of InsP3Rs is regulated by CaM (Kasri et al., 2002), however, several reports also indicate that CaBP1, can modulate the activity of InsP3Rs (Haynes et al., 2004; Kasri et al., 2004; Li et al., 2009, 2013). Of note, two binding sites for CaBP1 were identified in InsP3Rs and a high affinity interaction requires self-association and a close proximity of both binding domains (Li et al., 2009). Although binding was originally described in contrast to those of CaM to be Ca2+-independent, it turned out that Ca2+-binding to EF-hands-3 and -4 clearly enhances binding affinity and strengthens the interaction (Li et al., 2009). In addition, it was reported that phosphorylation of CaBP1 by casein kinase II promotes the association (Kasri et al., 2004). In functional terms CaBP1 binding reduces InsP3-triggered [Ca2+]<sup>i</sup> release (Haynes et al., 2004; Kasri et al., 2004). Based on NMR studies, it was postulated that a cluster of hydrophobic residues in the C-terminal domain of CaBP1 interact with a complementary cluster of hydrophobic residues in the βtrefold domain of InsP3R, trapping the channel in a closed conformation (Li et al., 2013). Upon binding of Ca2<sup>+</sup> the affinity of CaBP1 was found to be increased for InsP3R, locking the channel in the closed state (Li et al., 2009, 2013). Importantly, Li et al. (2009) could show that binding of CaBP1 to InsP3R is much stronger than those of CaM to InsP3Rs, indicating that the former interaction will easily dominate and is more significant for InsP3R function and given the abundance of Caldendrin in spines it will clearly outcompete CaM. Furthermore, the low expression of CaBP1 in the hippocampus makes it likely that caldendrin is the Ca2<sup>+</sup> sensor for regulating hippocampal InsP3Rs and might thereby play an important role in synaptodendritic Ca2<sup>+</sup> signaling as a prominent inhibitor of InsP3Rs in the hippocampus, a prediction that can be easily tested now due to the existence of caldendrin/CaBP1 knockout mice models (Kim et al., 2014; Mikhaylova et al., 2018; Yang et al., 2018).

Several reports indicate that the association the association of caldendrin/CaBP1 with the α1-C subunit of L-type voltagedependent Ca2<sup>+</sup> channels (CaV1.2) is important for its physiological function (Zhou et al., 2004, 2005; Tippens and Lee, 2007; Findeisen and Minor, 2010; Oz et al., 2011, 2013). Caldendrin/CaBP1 reduce the Ca2<sup>+</sup> dependent inactivation (CDI) of CaV1.2 L-type voltage gated Ca2<sup>+</sup> channels and thereby facilitate Ca2<sup>+</sup> currents (Zhou et al., 2004, 2005; Tippens and Lee, 2007; Oz et al., 2011, 2013). In contrast, binding of CaM reportedly results in an inactivation of Ca2<sup>+</sup> influx and competes with caldendrin/CaBP1 for the same interaction site (Zhou et al., 2004, 2005; Tippens and Lee, 2007). It is plausible that the interaction takes place in the postsynaptic membrane since co-immunoprecipitation of caldendrin and CaV1.2 Ca2<sup>+</sup> channels was shown with synaptic protein preparations (Zhou et al., 2004, 2005; Tippens and Lee, 2007). The emerging scenario from several studies shows that the regulation of CaV1.2 Ca2<sup>+</sup> channels by caldendrin, CaBP1 and CaM is complex and that differential binding will allow for a high degree of fine-tuning of synapto-dendritic Ca2<sup>+</sup> signaling. All three sensors interact with an IQ-domain that is part of the C-terminal cytoplasmic domain of the α1-subunit of CaV1.2. In addition, CaBP1 binds to a second region in the N-terminus and this interaction is in contrast to those of CaM Ca2+ independent (Zhou et al., 2005; Dick et al., 2008). Of note, the presence of this domain is essential for inactivation of CaV1.2 Ca2<sup>+</sup> currents by Caldendrin, but not the CDI mediated by CaM (Zhou et al., 2004). Interestingly, CaBP1 seems to more efficiently inhibit inactivation of CaV1.2 channels than caldendrin (Zhou et al., 2005; Tippens and Lee, 2007). Whereas caldendrin displaces CaM and CaBP1 from the C-terminal IQ-domain it does not interact with the N-terminus of CaV1.2 (Tippens and Lee, 2007). It is therefore likely that the functionally different roles of CaM, CaBP1 and caldendrin are mediated by association to two different domains with different Ca2<sup>+</sup> requirements.

Other interactions have been studied in much less detail. In functional terms a role in synapse-to-nucleus communication has been ascribed to caldendrin by preventing the nuclear import of the synapto-nuclear messenger Jacob (Dieterich et al., 2008). Caldendrin/CaBP1 knockout mice show a rapid depression at inhibitory presynaptic sites that is related to binding and inactivation of CaV2.1 calcium channels in control of short-term synaptic plasticity (Lee et al., 2002; Few et al., 2011; Leal et al., 2012; Nanou et al., 2018). Other potential binding partners where a cellular and in particular synaptic function is less well investigated include light chain 3 (Seidenbecher et al., 2004), myo1c (Tang et al., 2007), recoverin (Fries et al., 2010), metabotropic glutamate receptors (Nakajima, 2011), AKAP79/150 (Seeger et al., 2012) and CaV1.3 calcium channels (Findeisen et al., 2013; Yang et al., 2018).

#### A SPECIFIC SYNAPTIC FUNCTION OF CALDENDRIN

Interestingly, caldendrin is highly enriched in the postsynaptic density (PSD) of spine synapses, which is very uncommon for a CaM-like Ca2+-sensor protein (Seidenbecher et al., 1998; Laube et al., 2002). Dendritic spines are considered as microcompartments of Ca2<sup>+</sup> signaling (Raghuram et al., 2012) with exceptionally fast Ca2<sup>+</sup> decay times, much faster than in dendrites (Cornelisse et al., 2007). The presence of ''fast'' Ca2<sup>+</sup> buffers are thought to mediate the fast decay of Ca2+-transients in spines immediately following influx through Ca2<sup>+</sup> channels or release from Ca2<sup>+</sup> stores (Keller et al., 2008). Fast buffers are the first to intercept free Ca2<sup>+</sup> in the spine. Surprisingly, it was found utilizing UV flash photolysis of DM-nitrophen caged Ca2<sup>+</sup> that CaM binds Ca2<sup>+</sup> with a very high on-rate at the N-terminal lobe, even faster than those of classical known calcium buffers like calbindin (Faas et al., 2011). In conclusion, CaM, due to high abundance, its ubiquitous expression and the fast Ca2+ binding capacity, has to be considered as the principal buffer in pyramidal neurons. In addition numerous Ca2+-dependent targets have been identified which in regulate different aspects of cellular function. In light of these arguments the simple question arises how members of the NCS and nCaBP family operate as calcium sensors in the presence of CaM? Or in other terms will CaM-like sensors even have a chance to compete with CaM for Ca2<sup>+</sup> binding in spines?

In a recently published study we addressed this question systematically (Mikhaylova et al., 2018). We first found that caldendrin has a similar on rate like CaM for Ca2+-binding and can therefore by abundance and properties easily compete with CaM in spines (Mikhaylova et al., 2018). Alterations in synaptic strength require an intimate link between functional and structural plasticity. The latter one is based on the unique cytoskeletal organization of differentially arranged actin filaments. Compartmentalization of calcium-dependent plasticity allows for rapid actin remodeling and in a landmark study, Bosch et al. (2014) addressed the question of structural and molecular remodeling of dendritic spines following the induction of long-term potentiation (LTP). LTP induction results in rapid cofilin-dependent severing of filamentous actin and a massive increase in actin remodeling proteins, which is then followed by a stabilization phase where different F-actin stabilizing and capping proteins gradually increase and only at this later stage reorganization of the PSD comes into play. What is still unclear is how a minimal stable pool of branched actin that is essential for remodeling of F-actin is maintained in spines undergoing plasticity. In fact, molecular machineries coupling local and rapid synaptic elevation of [Ca2+]<sup>i</sup> to actin remodeling in the initial reorganization phase are still unknown. Finally, the role of spatially segregated spinous F-actin pools in transition from short-term to long-term synaptic potentiation was unclear.

We could show that caldendrin activates the actin-binding protein cortactin in dendritic spines and thereby stabilizes a synaptic pool of branched F-actin that is essential for the maintenance of LTP (Mikhaylova et al., 2018). We found that steep elevations in spinous [Ca2+]<sup>i</sup> disrupt an intramolecular interaction of caldendrin that hinders access to a series of PxxPmotifs. Opening of the intramolecular interaction results in a rapid association with the SH3 domain of cortactin (**Figure 2**). A fast on and a slow off rate of binding keep cortactin in an active, F-actin-stabilizing conformation. In this conformation the protein protects a minimal stable spinous F-actin pool against cofilin-induced severing and primes cortactin for sequential binding to N-WASP/Arp2/3 complex in vicinity of the PSD. Caldendrin gene knockout or protein knockdown result in higher actin turnover caused by loss of stable pool of actin filaments and a disordered topology of spinous F-actin which lead to a loss of spatial segregation of F-actin nanodomains, defects in structural spine plasticity, LTP and hippocampus dependent learning (Mikhaylova et al., 2018). We think that caldendrin directly couples elevation of [Ca2+]<sup>i</sup> to the stabilization of actin branches in a very early step of temporary gating for F-actin remodeling in dendritic spines and thus controls the life time of functionally different F-actin pools during the reorganization phase (**Figure 2**). This mechanism is an essential component of structural plasticity. Accordingly, it was reported in a recent study (Yang et al., 2018) that caldendrin knock-out mice exhibit deficits in spatial learning and memory and also fear-related memories. Surprisingly, it was also found that adult neurogenesis in the hippocampus is severely impaired in knock-out animals by yet unknown mechanisms (Yang et al., 2018). It is likely that this impairment will also contribute to learning and memory deficits in these mice.

#### THE TRANSMEMBRANE nCaBP FAMILY MEMBERS CALNEURON-1 AND -2

Calneuron-1 and -2 (also called CaBP8 and CaBP7; McCue et al., 2010) configure a separate subfamily within the nCaBP family (Mikhaylova et al., 2006, 2011). In contrast to other NCSand nCaBP proteins they harbor a non-functional C-terminal EF-hand domain (**Figure 1**; Wu et al., 2001; Mikhaylova et al., 2006; McCue et al., 2010). Both calneurons show 64% identity at the amino acid level when compared to each other whereas less than 30% identity exists in comparison to CaBPs and caldendrin (Mikhaylova et al., 2006; McCue et al., 2010). The loops of the functional EF-hands 1 and 2 are almost identical (Mikhaylova et al., 2006) and also the loops of EF-hand 3 exhibit high similarity. Calneurons have the highest Ca2+-affinity of all identified CaM-like EF-hand Ca2+-sensors with a global dissociation constant of 230 and 180 nM for calneuron-1 and -2, respectively (Mikhaylova et al., 2009). However, these affinities were measured for proteins without membrane insertion, which might have obscured the outcome of these measurements. Apart from differences within the EF-hand domains calneurons harbor in addition an extension of the C-terminus compared to other nCaBPs and NCS-proteins, which is crucial for the targeting of these proteins to the Golgi-membrane (McCue et al., 2010; Hradsky et al., 2011), where they are involved in the regulation of Golgi to plasma membrane trafficking (Mikhaylova et al., 2009). Being almost identical in both proteins the C-terminus contains a highly hydrophobic stretch, which configures a transmembrane domain (TMD; Hradsky et al., 2011; McCue et al., 2011). Since both calneurons do not harbor a N-terminal myristoylation motif, transmembrane insertion of these proteins would explain their strong membranal association as compared to other neuronal Ca2+-sensors like caldendrin. Of note they

are the only transmembrane CaM-like calcium sensors and this points to a probably unique role in the neuronal Ca2+-sensing toolkit.

### CALNEURON-1 AND -2 ARE TAIL ANCHORED PROTEINS THAT REGULATE GOLGI-TO-PLASMA MEMBRANE TRAFFICKING

The C-terminal position of the transmembrane segment indicates that calneurons belong to the group of tail-anchored proteins, which have to undergo post-translational insertion (Borgese et al., 2003, 2007). It turned out that calneurons are indeed non-classical type II tail-anchored proteins and their posttranslational insertion into the ER membrane via an association of the TMD with the TRC40/Asna1 chaperone complex was demonstrated (Hradsky et al., 2011). Their tight association with the trans-Golgi-network (TGN) might be explained by the length of the TMD and phosphatidylinositol 4-phosphate (PI(4)P) lipid binding (Hradsky et al., 2011). Self-association in vitro and in vivo occurs via the TMD and EF-hand containing N-terminus. Despite the fact that dimerization will hinder TRC40/Asna1 binding and in consequence membrane insertion, evidence for the existence of a cytosolic non-membrane associated pool of calneurons is currently lacking and dimerization was only found for membrane inserted protein (Hradsky et al., 2011). This almost exclusive and unique association with membranes of the secretory pathway indicates a probably highly specific function with a limited number of target interactions.

In functional terms calneurons play an important role at the Golgi-apparatus where they control TGN to plasma membrane trafficking by regulating the activity of phosphatidylinositol 4-OH kinase IIIβ (PI-4KIIIβ; Mikhaylova et al., 2009). PI-4KIIIβ catalyzes local synthesis of phosphoinositides necessary for vesicle budding at the TGN. Calneurons directly bind to PI-4KIIIβ and inhibit the enzyme at low [Ca2+] levels (**Figure 3**; Mikhaylova et al., 2009). With increased [Ca2+] levels the inhibition is released and PI4KIIIβ is activated via a preferential association with NCS-1. Taken together that data suggest that calneurons establish a [Ca2+] threshold for activation of the enzyme (**Figure 3**; Mikhaylova et al., 2009). Of note, the tight association of calneurons to the Golgi can be even used to target proteins to the TGN (Bera et al., 2016). The TMD of calneuron-2 was employed to develop a plasmid-based expression system called pGolt that has the advantage to fuse other proteins to the extraluminal part. This in turn makes it possible to study protein-protein interactions outside of the Golgi lumen (Bera et al., 2016). An obvious and particularly promising application in neuroscience is to use this Golgi-tracker system for the visualization of Golgi

outposts (GOs). In non-neuronal cells the organelles of the secretory pathway have a highly restricted spatial organization. In stark contrast in neuronal cells along with the localization of secretory organelles in the cell soma, numerous discrete and discontinuous structures resembling Golgi cisternae are present along dendrites, which are known as GOs. We next showed with pGolt the presence of Golgi-related organelles in all dendrites of pyramidal neurons in close proximity to endoplasmic reticulum-Golgi intermediate compartment and retromer (Mikhaylova et al., 2016). We found that this Golgi-Satellite secretory system (GS) in dendrites is much more widespread than previously described GOs. Most importantly, this GS contains at least part of the cellular glycosylation machinery but as opposed to GOs lacks many protein components for sorting and organization of Golgi cisternae. Moreover, we realized that a broad spectrum of synaptic transmembrane proteins (including GluA1, GluN1, GluN2B, NCAM and Neuroligin-1) might pass and even recycle through these organelles and that also calneuron-1 is present at GS (Mikhaylova et al., 2016). Collectively the study suggest that GS will enable local glycosylation of proteins, and that these proteins will be subsequently recruited to membranes in spatially confined dendritic segments. It will be interesting to investigate in the future whether calneuron-1 has a similar role at GS like at the TGN. Of note, another report (Rajamanoharan et al., 2015) indicated that calneuron-2 mediated inhibition of PI4KIIIβ is instrumental for cytokinesis in HeLa cells. In this study, it was reported that calneuron-2 associates with lysosomes and regulates their clustering and that PI4KIIIβ plays an important role for normal cytokinesis (Rajamanoharan et al., 2015).

# OTHER POSSIBLE CELLULAR FUNCTIONS OF CALNEURONS IN NEURONAL Ca2+-SIGNALING

Compared to the multitude of CaM protein interactions, the interactome of nCaBPs and here in particular of calneurons is relatively small. In several studies an interaction with G-protein coupled receptors (GPCRs) was reported (Navarro et al., 2014; Angelats et al., 2018; Franco et al., 2018). It is well established in striatal neurons that [Ca2+]<sup>i</sup> impact modulation of A2AR-D2R heteromers that is mediated adenylylcyclase and MAP-kinases. Differential modulation is based on the interaction of the heteromer with NCS-1 and calneuron-1 at low and high [Ca2+]<sup>i</sup> , respectively (Navarro et al., 2016). The association with both Ca2+-binding proteins appears to differentially regulate subsequent allosteric interactions within the A2AR-D2R heteromer. Thus, binding of both proteins configures a unique cellular mechanism to integrate extracellular (adenosine and dopamine) and [Ca2+]<sup>i</sup> signals in order to elicit a specific downstream signaling event (Navarro et al., 2016). In the past years further interactions with either GPCR heteromers or heteromers of GPCR with glutamate receptors were reported (Angelats et al., 2018; Franco et al., 2018) and the emerging picture suggests that: (i) [Ca2+]i determines the binding affinity of different sensors; and that (ii) in case of cooperative binding differential intracellular responses can be elicited depending upon binding. What is clearly missing is an understanding of the cellular function of these interactions. This also holds true for work that shows an interaction between calneuron-1 and N-type calcium channels (Shih et al., 2009), an interaction that could play a role in the modulation of action potential firing or neurotransmitter release. Finally, calneuron-1 is highly expressed in aldosterone producing adenoma cells where it regulates the storage of Ca2<sup>+</sup> into the ER (Kobuke et al., 2017). Overexpression of calneuron-1 leads to an increased Ca2<sup>+</sup> level in the ER and aldosterone overproduction (Kobuke et al., 2017). A mechanistic explanation for this phenotype, however, is lacking.

#### ON A PUTATIVE ROLE OF CALNEURON-1 AND CALDENDRIN IN MENTAL DISORDERS

The onset of schizophrenia usually takes place in young adulthood between the age of 20–30 (Thompson et al., 2004). It is widely believed that subtle developmental alterations in brain structure and connectivity in the interaction with environmental factors cause schizophrenia in early adulthood. The underlying processes have been also related to modifications of the epigenome. Three relatively recent studies from different groups propose that the human CALN1 is a candidate schizophrenia gene (Li et al., 2015; Xia et al., 2015; Roussos et al., 2016) but it is at present unclear how calneuron-1 function can be related to psychotic behavior. However, it has been shown that alteration in methylation levels on the CALN1 and the AUTS2 gene occur in schizophrenia patients (Wockner et al., 2014). A recently published study (Engmann et al., 2017) shows that the exposure to drugs of abuse, like cocaine, alters epigenetic patterns of certain brain regions due to chromatin modifications. Engmann et al. (2017) found that the methylation of the CALN1 promoter increases the expression of AUTS2, which is another known risk gene for schizophrenia (Zhang et al., 2014), and vice versa. They show that methylation of the AUTS2 or

#### REFERENCES


the CALN1 promoter enhances calneuron-1 expression although both genes are not in close proximity to each other. The distance of around 1542 Kb between both genes is bypassed by an epigenetic modification called chromatin looping, which enables regulatory elements such as enhancers, which are located relatively far from the transcription start site, to interact with the associated promoter regions using CCCTC-binding-factors (Sexton et al., 2009; Engmann et al., 2017). Interestingly, a meta-analysis of the genetic association studies came to the conclusion that calneuron-1 mRNA levels might be up-regulated in schizophrenia in the dorsolateral prefrontal cortex (Ohi et al., 2016). A potential mechanism might have been proposed by Engmann et al. (2017), whose results are indicating that an overexpression of CALN1 might lead to an overexpression of AUTS2, a well-known risk gene for schizophrenia, via chromatin looping. Alternatively, higher calneuron-1 protein levels might have independently an effect on cellular signaling that is related to psychotic behavior.

Interestingly, caldendrin protein levels are also regulated in human schizophrenia (Bernstein et al., 2007) and in mouse models of drug-induced psychosis (Smalla et al., 2009). Fewer caldendrin-immunopositive neurons are found in the left dorsolateral prefrontal cortex in schizophrenia patients, and it is tempting to speculate that synaptic and/or dendritic Ca2+ signaling is altered in schizophrenia due to a redistribution of the protein. Thus, it appears that the remaining pyramidal neurons expressing caldendrin exhibit higher protein levels. In summary, although it is at present unclear how exactly both proteins contribute to the pathophysiology of schizophrenia, the basic characterization of their unique cellular role of calneuron-1 might pave the way to understand how they are involved in schizophrenia.

#### AUTHOR CONTRIBUTIONS

JM, CF, MK wrote the review article. CF designed the figures.

#### FUNDING

The work in the lab of MK is supported by grants from the Deutsche Forschungsgemeinschaft (DFG; Kr1879/5-1/6- 1/SFB 779 TPB8), BMBF ''Energi'' FKZ: 01GQ1421B, The EU Joint Programme—Neurodegenerative Disease Research (JPND) project STAD and the Leibniz Foundation.


phosphorylation of the receptors. Biochem. Biophys. Res. Commun. 412, 602–605. doi: 10.1016/j.bbrc.2011.08.006


**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 © 2019 Mundhenk, Fusi and Kreutz. 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) and the copyright owner(s) 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.

# 20 Years of Secretagogin: Exocytosis and Beyond

#### Magdalena Maj <sup>1</sup> , Ludwig Wagner <sup>2</sup> and Verena Tretter <sup>3</sup> \*

<sup>1</sup>Department of Biological Sciences, California Polytechnic State University, San Luis Obispo, CA, United States, <sup>2</sup>Department of Internal Medicine III, Division of Nephrology and Dialysis, Medizinische Universität Wien, Vienna, Austria, <sup>3</sup>Department of Anesthesia and General Intensive Care, Clinical Department of Anesthesia, Medizinische Universität Wien, Vienna, Austria

Calcium is one of the most important signaling factors in mammalian cells. Specific temporal and spatial calcium signals underlie fundamental processes such as cell growth, development, circadian rhythms, neurotransmission, hormonal actions and apoptosis. In order to translate calcium signals into cellular processes a vast number of proteins bind this ion with affinities from the nanomolar to millimolar range. Using classical biochemical methods an impressing number of calcium binding proteins (CBPs) have been discovered since the late 1960s, some of which are expressed ubiquitously, others are more restricted to specific cell types. In the nervous system expression patterns of different CBPs have been used to discern different neuronal cell populations, especially before advanced methods like single-cell transcriptomics and activity recording were available to define neuronal identity. However, understanding CBPs and their interacting proteins is still of central interest. The post-genomic era has coined the term "calciomics," to describe a whole new research field, that engages in the identification and characterization of CBPs and their interactome. Secretagogin is a CBP, that was discovered 20 years ago in the pancreas. Consecutively it was found also in other organs including the nervous system, with characteristic expression patterns mostly forming cell clusters. Its regional expression and subcellular location together with the identification of protein interaction partners implicated, that secretagogin has a central role in hormone secretion. Meanwhile, with the help of modern proteomics a large number of actual and putative interacting proteins has been identified, that allow to anticipate a much more complex role of secretagogin in developing and adult neuronal cells. Here, we review recent findings that appear like puzzle stones of a greater picture.

#### Edited by:

Beat Schwaller, Université de Fribourg, Switzerland

#### Reviewed by:

Joern R. Steinert, University of Leicester, United Kingdom Valentin Stein, Universität Bonn, Germany

#### \*Correspondence:

Verena Tretter verena.tretter@meduniwien.ac.at

Received: 29 September 2018 Accepted: 23 January 2019 Published: 12 February 2019

#### Citation:

Maj MA, Wagner L and Tretter V (2019) 20 Years of Secretagogin: Exocytosis and Beyond. Front. Mol. Neurosci. 12:29. doi: 10.3389/fnmol.2019.00029 Keywords: secretagogin, calcium binding protein, calcium sensor, insulin secretion, hormone secretion

# INTRODUCTION

Calcium ions (Ca2+) are one of the major messengers used by cells to regulate their metabolism, drive gene expression and activate specific cellular functions like exo- and endocytosis or contraction. In order to fulfill these tasks resting cells actively expel calcium ions from the cytosol with the help of Ca2<sup>+</sup> transport ATPase (PMCA) and Na+/Ca2<sup>+</sup> exchanger (NCX) that are localized in the plasma membrane and transport Ca2<sup>+</sup> against a massive concentration gradient into the extracellular space (Rosado et al., 2006). Calcium is also stored intracellularly in the endoplasmatic reticulum (ER) or, as in specialized muscle cells, in the sarcoplasmic reticulum (SR) as well as in the mitochondria and is taken up into these compartments by the sarcoendoplasmic reticulum Ca2<sup>+</sup> - ATPase (SERCA) or the mitochondrial Ca2<sup>+</sup> uniporter. Activation of Ca2+-dependent signaling pathways can be initiated by membrane depolarization or extracellular signaling molecules that activate voltage- or ligand-activated calcium channels in the plasma membrane or via intracellular messengers that cause the release of Ca2<sup>+</sup> from intracellular stores, mainly via the 1, 4, 5- triphosphate receptor (IP3R) or the ryanodine receptor (RyR) from the ER or SR (Supnet and Bezprozvanny, 2010). Beside these transporters and channels, cells can draw from a huge number of calcium-dependent interactions and signaling components, that Berridge et al. (2000) call the ''Ca2<sup>+</sup> signaling toolkit.'' Parts of this universal toolkit are combined by individual cells to fulfill their special tasks. As Ca2<sup>+</sup> flows into the cells due to so-called ''ON mechanisms'' it switches on multiple signaling cascades. The amplitude and duration of the signal is shaped by cytosolic Ca2<sup>+</sup> buffers like parvalbumin, calbindin-D28K and calretinin that transiently take up Ca2<sup>+</sup> and contribute to the spatial restriction of the signal. This is especially important in neurons where compartmentalization of signaling is mandatory for neuronal transmission (Sudhof, 2004). In contrast to high-affinity Ca2<sup>+</sup> buffers, lower affinity Ca2<sup>+</sup> sensors respond to significant elevations of intracellular Ca2<sup>+</sup> concentrations with conformational changes thereby facilitating interaction with downstream targets and initiating major cellular responses (Amici et al., 2009). Depending on the Ca2<sup>+</sup> binding motifs there are three classes of Ca2<sup>+</sup> binding proteins: the EF-hand proteins, the annexins and the C2 domain proteins (Bagur and Hajnoczky, 2017). The EF-hand is frequently found in buffers and sensors alike (structural vs. regulatory EF-hand) and contains Ca2<sup>+</sup> in a complex with seven ligands in a loop flanked by two perpendicular α-helices (da Silva et al., 1995). Prominent members of this class are calmodulin, parvalbumin, troponin C, calretinin, calcineurin and secretagogin.

While the other representatives have been known for a long time and their function is well characterized, secretagogin's precise role in specialized cell types has been beginning to emerge only recently. Here, we will review recent findings with regard to secretagogin's cellular localization, interactions and functions as a novel Ca2<sup>+</sup> sensor after a brief glance at the beginnings.

#### DISCOVERY OF SECRETAGOGIN

Secretagogin (SCGN) was discovered in an ''expression dictates function'' manner. Wagner et al. (1998) generated the murine monoclonal antibody D24 (mAb D24) against an unknown human insulinoma-specific antigen, that binds to all cells present in the pancreatic islets of Langerhans and to all tested insulinomas but does not stain the exocrine pancreatic tissue. Screening of a human pancreatic beta cell cDNA library with D24 mAb identified a unique mRNA sequence, which encoded a novel Ca2+-binding protein related to human and murine calbindin and calretinin. A bioinformatic sequence analysis suggested that this novel protein has six tandem repeats of an EF-hand and a molecular weight of 32kDa. The newly discovered protein was named ''Secretagogin'' likely because it was shown to facilitate insulin release. Initial studies in Rin-5F rat insulinoma cells identified secretagogin mainly in the cytosolic fraction with some scant presence in the nuclei. Overexpression of secretagogin resulted in a greater calcium flux in response to KCl along with increased insulin secretion rates compared to vector controls (Wagner et al., 2000). Interestingly, these SCGN overexpressing Rin-5F clones had significantly lower cell growth rates. This fact correlated with down-regulation of substance-P (a putative β-cell trophic factor), known to be largely restricted to proliferating fetal and neonatal islet cells and to pancreatic beta cell lines. Therefore, it is possible that SCGN-induced suppression of substance-P transcription is responsible for the observed reduction in cell growth rates (Maj, 2012).

#### CHARACTERISTICS OF SECRETAGOGIN AND INTERACTION NETWORK

Secretagogin is a hexa EF-hand protein that binds Ca2<sup>+</sup> with an overall half maximal affinity of approximately 25 µM, a relatively low affinity compared to calbindin [with a (Ca2+) 0.5 of approx. 1 µM] or calretinin [with a (Ca2+) 0.5 of approx. 1.5 µM; Rogstam et al., 2007]. The secretagogin gene and protein structure has been reviewed in detail previously (Alpár et al., 2012).

The tertiary structure of secretagogin changes significantly upon Ca2<sup>+</sup> binding, but not upon Mg2<sup>+</sup> binding, suggesting that SCGN belongs to the ''sensor'' family of Ca2<sup>+</sup> -binding proteins (Rogstam et al., 2007).

Expression of secretagogin can be influenced by insulin and glucose. In a cell culture study an insulin bolus was shown to increase secretagogin mRNA in hippocampal neurons after 24 h, while addition of glucose down-regulated SCGN mRNA levels already after 1 h (Maj et al., 2012). It has further been established, that SCGN mRNA expression in pancreatic islets negatively correlates with the incidence of diabetes and the levels of glycated hemoglobin in blood (Malenczyk et al., 2017). As a mechanism to activate secretagogin expression Malenczyk et al. (2017) could pinpoint the family of Ca2+permeable TRP(V) channels, that, when activated, induce nuclear translocation of the Sp1 transcription factor to initiate secretagogin transcription. Sp1 has previously been shown to also mediate upregulation of the CBP calmodulin in response to an insulin stimulus (Solomon et al., 1997).

A Ca2<sup>+</sup> sensor responds to a strong Ca2<sup>+</sup> signal with changes in protein conformation that can only be translated in cellular processes via interaction with other proteins. Therefore identification of protein interaction partners is crucial to understand functional mechanisms.

Rogstam et al. (2007) identified the first Ca2<sup>+</sup> dependent interaction partner of SCGN by affinity purification from mouse and cow brain and the rat insulinoma cell line RIN-5F, which was identified as 25kDa synaptosome-associated protein (SNAP-25) in LC-MS/MS. SNAP-25 is one of the components of the soluble N-ethylmaleimide-sensitive-factor attachment receptor (SNARE; soluble N-ethylmaleimide-sensitive fusion protein attachment protein receptor) complex, which is the main complex for membrane fusion events (Chen and Scheller, 2001). As SNAP-25 is a protein involved in Ca2+-induced exocytosis in neurons and neuroendocrine cells, it was anticipated that secretagogin might be involved in the exocytosis processes of neurotransmitters and insulin granules.

Three years later nine other proteins were identified from screening a human protein array for Ca2+-dependent secretagogin interactors: SNAP-23, DOC2alpha, ARFGAP2, rootletin, KIF5B, β-tubulin, DDAH-2, ATP-synthase and myeloid leukemia factor 2 (Bauer et al., 2011). All of these interaction partners are involved in secretion events and vesicle trafficking. For example, SNAP-23 is structurally and functionally similiar to SNAP-25, binds to the SNAREs syntaxin and synaptobrevin and is an important regulator of transport vesicle docking and fusion. DOC2a itself is a Ca2+sensor, that promotes among others the spontaneous release of neurotransmitters glutamate and GABA (Courtney et al., 2018). ARFGAP2 is involved in specific cargo sorting to the plasma membrane. Kinesin 5B in neurons has been shown to be essential as motor protein in the anterograde transport of active zone proteins (Cai et al., 2007). Rootletin functions as a linker of the centrosome, which is the central microtubule organization center. Dimethylarginine dimethylaminohydrolase (DDAH) 2 is an enzyme, that metabolizes the NO-synthesis antagonist ADMA (asymmetric dimethylarginine). ATP synthase is an enzyme, that produces the cell's main energy currency, ATP, which fires most cellular processes including intracellular transport. Tubulin as constituent of microtubuli is involved in major transport processes in the cell. Further, Ca2+-dependent interaction of secretagogin with the microtubule-associated protein (MAP) Tau was initially demonstrated by GST-pull downs from pancreatic cell lines and later confirmed from rat brains (Maj et al., 2010, 2012).

These SCGN interaction partners already implicate the possibility that secretagogin might link Ca2<sup>+</sup> signaling to exocytotic processes.

In the last 3 years the discovered secretagogin interactome has grown extensively due to huge proteomic efforts. In **Supplementary Table S1**, we summarize to the best of our knowledge all published confirmed and putative interactors of secretagogin. Beside the pioneering studies, there are results from large scale proteomics essentially from two major groups available (Huttlin et al., 2015, 2017; Romanov et al., 2015; Hanics et al., 2017; Malenczyk et al., 2017, 2018). Without a special focus on secretagogin, Huttlin et al. (2015, 2017) found 16 putative interactions in their ''BioPlex Network,'' a high throughput combination of affinity purification with thousands of baits from a lentiviral library and mass spectrometry. Some of these hits were consistent with previously known proteins, indicating the validity of results.

Malenczyk et al. (2017, 2018) were specifically interested in SCGN interaction partners and used pull-down assays from pancreas and the INS-1E beta cell line combined with LC-MS/MS analysis. Hanics et al. (2017) focused on migrating SCGN<sup>+</sup> neurons in the rostral migratory stream (RMS) and used micropunches from that area in co-immunoprecipitations. Similarily, Romanov et al. (2015) identified Ca2+-dependent and independent interactors by co-immunoprecipitations from hypothalamus using MALDI-TOF analysis.

The resulting interactome is extensive and can be functionally grouped into proteins involved in vesicle-mediated transport and exocytosis, protein folding, ubiquitination and proteasomal degradation, cytoskeleton-related proteins and their organization, carbohydrate metabolism, lipid metabolism, mitochondrial organization and processes, gene expressionrelated processes from DNA-replication to protein translation, metabolism of nucleotides, kinases and phosphatases and some other proteins (**Supplementary Table S1**, **Figure 1**). Certainly, not all interactions have been characterized in detail yet, but in many cases it can be anticipated, that these interactions are important for the function of secretory cells.

In their study Malenczyk et al. (2018) focused on proteins, that might have protective roles in supporting protein folding, stabilization and in preventing ER stress. Proteins like ubiquitin carboxyl-terminal hydrolases USP9X and USP7, that de-ubiquitinate their substrates and thereby stabilize them, are kept in an active form by secretagogin, in order to promote β-cell proliferation. Another example for such an interaction partner is T-complex protein 1 (CCT), a chaperonin, or ADP-ribosylation factor (Arf)4, involved in vesicle trafficking and a marker of ER/Golgi stress (Reiling et al., 2013), that is inversly regulated to secretagogin levels (Malenczyk et al., 2018). The cytosolic chaperonin CCT has an important role in the biogenesis of the actin cytoskeleton (Grantham et al., 2002) and was previously found to bind to chromaffin granules possibly involved in exocytosis (Creutz et al., 1994).

In order to elucidate more of secretagogin's cellular functions, it is essential to note, that it is not only a pancreatic protein, but occurs in several other tissues, as it seems, frequently locally restricted to larger groups of cells. In this regard, secretagogin's distribution in the CNS is of major interest.

#### CELL SPECIFIC DISTRIBUTION OF SECRETAGOGIN IN THE NERVOUS SYSTEM

Secretagogin has been studied in various organisms as it is expressed from zebrafish to human. Initial studies only roughly identified secretagogin positive areas in various species. Beside the pancreas, it was detected in the gastrointestinal tract, thyroid, adrenal medulla, adrenal gland and brain (Wagner et al., 2000). Especially prominent is SCGN in neuroendocrine cells such as the islet of Langerhans and subpopulations of developing or adult neurons (Gartner et al., 2010; Mulder et al., 2009, 2010).

However, secretagogin's expression pattern is not conserved from rodents to humans and also significant differences exist between mice and rats (Garas et al., 2016; Raju et al., 2018). At the mRNA and protein level secretagogin has the highest expression in human cerebellum (Maj et al., 2012), while in mouse and rat it is found predominantly in the olfactory bulb (Mulder et al., 2009; Maj et al., 2012). Interestingly, SCGN is also highly expressed in the hippocampus of both, men and rodents. Immunohistochemical analysis identified secretagogin positive neurons mostly as interneurons and pyramidal cells (Gartner et al., 2010); SCGN is highly expressed in the molecular layer of human cerebellum, while in rat olfactory bulb it is found mainly in cell bodies of the granular layer and neurites and some cell bodies of the glomerular and external plexiform layer (Maj et al., 2012).

As mentioned before, it is apparent, that expression of secretagogin frequently occurs in cell clusters, shown for rat brain by Maj et al. (2012), implicating that these neurons either share common functional characteristics or their developmental origin. For example, SCGN<sup>+</sup> clusters are seen in peripheral cell layers of the olfactory bulb, lining the surface above the optic nerve, in the supraoptic nuclei on both sides of the optic chiasm, in the suprachiasmatic nuclei of the hypothalamus, in the paraventricular nucleus (PVN), in the CA1-CA3 of the hippocampus, and outer cortex around the posterolateral cortical amygdaloid area. This specific expression of SCGN anticipates an important role in the mechanisms of olfaction, vision, activity of the hypothalamus, and memory formation in hippocampus.

Early immunocytochemical localization studies revealed that in addition to cytoplasmatic staining, SCGN seemed to be accumulated on specific subcellular structures in rat neurons. Specifically, SCGN<sup>+</sup> clusters were often found in direct contact with GM130 clusters, a marker of the cis-Golgi network. Adding to the fact, ultracentrifugation experiments using sucrose density centrifugation of rat hippocampal tissue extracts revealed that SCGN immunoreactivity was detectable in the membrane-fraction as well as in the cytosolic fraction (Maj et al., 2012). In semiquantitative electron microscopy analyses of the PVN of the hypothalamus, gold particlecoupled secretagogin antibodies bound to the plasma membrane and to endomembranes even outnumbered cytosolic particles (Romanov et al., 2015).

More detailed studies aimed to characterize SCGN-positive cell populations with regard to co-expression of other cell markers are required. Gyengesi et al. (2013) investigated the SCGN expression pattern in neurons of the basal forebrain, the major cholinergic output of the central nervous system. SCGN is expressed in the forebrain, where Ca2+-dependent signaling plays an important role in neuronal plasticity and synaptic function. Disturbance of Ca2<sup>+</sup> buffering capacity in the basal forebrain is observed to be increased in aged, cognitively impaired rats (Murchison et al., 2009). Hence, the study focused on characterizing cholinergic corticopedal neurons in the basal forebrain that play important roles in cortical activation, sensory processing, and attention. They demonstrated that SCGN is expressed in cell bodies of the medial and lateral septum, vertical and horizontal diagonal band nuclei and in the extension of the amygdala, but it is almost absent in the ventral pallidum. SCGN is co-localized with choline acetyltransferase (ChAT) in neurons of the bed nucleus of the stria terminalis, which is considered an important regulator site of the hypothalamicpituitary-adrenal (HPA) axis, the extension of the amygdala, and the interstitial nucleus of the posterior limb of the anterior commissure. SCGN is frequently co-localized with calretinin, but not with parvalbumin or neuropeptide Y (Gyengesi et al., 2013).

A detailed characterization of SCGN-positive cells in zebrafish retina has been provided by Dudczig et al. (2017). SCGN starts to be expressed at day-3 postfertilization coinciding with the occurrence of Ca2+-waves initiated by retinal bipolar cells at a time, when larvae show first visual responses. SCGN stains a population of interneurons (approximately 60% GABAergic) in the inner nuclear layer with neurites extending into the inner plexiform layer. The expression of SCGN shows significant overlap with calbindin and calretinin, but not parvalbumin expression.

Secretagogin has been demonstrated to label several subtypes of cone bipolar cell in the mouse, rat, and rabbit retina (Puthussery et al., 2010). This finding was supported by a second study showing that SCGN is strongly expressed in the bipolar cell type DB1 in the macaque retina (Puthussery et al., 2011). Utilizing SCGN as a DB1 cell-specific marker allowed to investigate the cone connectivity as well as characterization of neurotransmitter receptors. This showed that the DB1 cells make synaptic contact with both L/M as well as S-cone photoreceptors and only minimal contact with rod photoreceptors (Puthussery et al., 2011).

In the human hippocampus, secretagogin occurs exclusively in CA1-CA4 and subiculum pyramidal neurons (Attems et al., 2007). Interneurons targeting striatal projection neurons were regarded as a relatively homogenous cell population due to their common expression of parvalbumin. However, co-staining for secretagogin has defined two novel subpopulations that preferentially innervate different pathways in the striatum (Garas et al., 2016). A very distinct population of SCGN-positive neurons with regard to other CBPs like calbindin or calretinin occurs in the habenula (Maj et al., 2012). Here, large groups of SCGN<sup>+</sup> cells are flanked by similarily large groups of neurons expressing the other CBPs.

Secretagogin has also been found in a population of nociceptive dorsal root ganglia (DRG) neurons in the dorsal horn of mouse and human that coexpress calcitonin gene-related peptide (CGRP) and in rat DRGs not CGRP. SCGN-positive neurons were further found in the mouse, rat and human dorsal horn implicating a role of these neurons in processing of sensory information including pain (Shi et al., 2012).

An interesting fact is, that secretagogin is frequently expressed in young, not yet terminally differentiated neurons, like the migrating neuroblasts in the RMS and newly born dentate granule cells, possibly stimulating the release of factors necessary for the survival of these young neurons (Mulder et al., 2009). In the RMS of rodents, a large amount of newly born neuroblasts migrate towards the olfactory bulb to mediate sensory plasticity. Within the RMS a scaffold of wired secretagogin-positive neurons release on demand matrix-metalloprotease-2 (MMP-2) in order to loosen the extracellular matrix and to facilitate migration of the neuroblasts. MMP2-release is known to be mediated by annexin V, that has been identified as secretagogin interaction partner (see **Supplementary Table S1**).

Based on the unique SCGN expression and localization in the nervous system, this CBP broadens the repertoire of available cell-specific marker, which can be used in anatomical and neurophysiological studies, albeit with caution, as this population still might not be homogenous.

Meanwhile several studies have put effort in elucidating the mechanisms, how secretagogin actually functions in the cell. These have investigated its role in insulin release and have identified a neuronal correlate in hormone release.

#### SECRETAGOGIN'S ROLE IN INSULIN SECRETION

Early work already implicated that secretagogin influences insulin release (Wagner et al., 2000), while the detailed mechanism has only started to be understood recently. Glucose-stimulated insulin secretion (GSIS) is a multifactorial process that centrally involves an intracellular rise in Ca2<sup>+</sup> concentration after the opening of voltage-dependent Ca2<sup>+</sup> channels (Komatsu et al., 2013). The redox sensitive Ca2<sup>+</sup> sensor secretagogin has been postulated to dimerize by forming a disulfide bridge in response to the concomitantly elevated ROS, thereby strengthening its association with the actin cytoskeleton (Yang et al., 2016; Lee et al., 2017). Secretagogin dissociates from the syntaxin1A-inhibitor tomosyn after its de-SUMOylation by the protease SENP1 and involves in insulin granule trafficking and exocytosis, via its other identified interaction partners, the t-SNAREs (Ferdaoussi et al., 2017). Secretagogin then assumingly gets involved in a complex of syntaxin and the v-SNARE VAMP2 in order to recruit insulin-laden vesicles to the plasma membrane. Finally, insulin release from the dense-core vesicles is regulated by synaptotagmin, a N-terminal transmembrane C2-domain Ca2<sup>+</sup> sensor and v-SNARE, that also binds syntaxin and SNAP23/25 (MacDougall et al., 2018). However, secretagogin seems to have an additional function in restructuring the cytoskeleton during insulin release: it influences F-actin dynamics, in order to facilitate vesicle transport to the periphery, as well as focal adhesion remodeling (Yang et al., 2016). Proteomic studies have identified multiple SCGN-interacting proteins, that are either actin-binding proteins or have a regulatory function towards the actin cytoskeleton (see **Supplementary Table S1**).

#### CONTROL OF STRESS HORMONE RELEASE

A major finding of the last years has been the involvement of SCGN in the control of stress hormone release (Romanov et al., 2015). It has previously been described, that secretagogin occurs in groups of neurons in the hypothalamus, especially the PVN (Mulder et al., 2009; Maj et al., 2012). The PVN essentially contains two sorts of neurosecretory neurons: the magnocellular neurons secreting oxytocin and vasopressin in the posterior pituitary gland, and the parvocellular neurons secreting corticotropin-releasing hormone (CRH), vasopressin and thyrotropin-releasing hormone (TRH). Additionally, the PVN contains many neurons expressing neuropeptides. Calbindin-D28k and calretinin mainly occur in magnocellular neurosecretory cells (Sánchez et al., 1992; Arai et al., 1994), while secretagogin antibodies predominately stain a subpopulation of parvocellular neurons in the dorsolateral PVN. In their study secretagogin mRNA mainly occured together with transcripts for CRH, somatostatin, tachykinin 1, amphetamin-regulated transcript, neuromedin B, neuromedin U, neuropeptide Y and neuropeptide B, but not with vasopressin, oxytocin, thyrotropin-releasing hormone (TRH), galanin, cholecystokinin, neurotensin S, calcitonin, neuromedin S, natriuretic peptide C and adenylate cyclase activating polypetide 1. Axons of the parvocellular neurosecretory neurons project to the median eminence where they release their hormones to the hypophyseal portal system. On the ultrastructural level they found, that secretagogin not only occurs in the cytoplasm and close to the ER, but also in axonal terminals, frequently associated with the plasma membrane and the outer membrane of dense-core vesicles containing CRH implicating that it might be involved in CRH release. Indeed, knock-down of secretagogin reduced CRH release from PVN neurons, similiary like insulin release from pancreatic β-cells. in vivo knock-down of secretagogin by injection of siRNA into the lateral ventricle led to hypothalamic SCGN downregulation by 20–30% concomitant with CRH retention in the soma of parvocellular neurons. These results gain even more significance, as the authors were able to prove, that secretagogin plays a significant role in the overall stress response. CRH function is to stimulate the release of adenocorticotropic hormone (ACTH) from the anterior pituitary gland and ACTH then activates the adrenal glands to release steroid hormones like cortisol, the primary stress hormone. Romanov et al. (2015) could show in an in vivo stress model, that the acute stress led to the activation of the immediate early gene c-fos mostly in secretagogin-positive neurons of the PVN and the corresponding ACTH response including corticosterone levels in the blood could be blunted by secretagogin siRNA-pretreated animals.

Authors from the same group went on to find out, that CRH neurons not only mediate HPA-axis activation, but also influence cortical excitability in a secretagogin-dependent fashion. The prefrontal cortex (PFC) coordinates vigilance, alertness and behavioral responses to acute stress facilitated by noradrenergic afferents originating in the locus coeruleus. These norepinephrinergic (NE+) cells on one hand are directly innervated by CRH<sup>+</sup> cells (Zhang et al., 2017), but on the other hand inter-neuronal communication is bypassed by volume transmission from ependymal cells via the cerebrospinal fluid in order to extend responses over longer periods. Specifically, CRH<sup>+</sup> cells from PVN also innervate ependymal cells of the 3rd ventricle inducing ciliary neurotrophic factor (CNTF) release into the cerebrospinal fluid. In the locus coeruleus, CNTF binds to its cognate Trk receptor on NE<sup>+</sup> neurons initiating a signaling cascade of ERK1 activation (a secretagogin-interactor) and subsequent tyrosine hydroxylase (TH) phosphorylation thereby stimulating its enzymatic activity. Alpár et al. (2018) proved a central role of secretagogin in mediating TH-phosphorylation in knock-down experiments and observed a blunted behavioral stress response in secretagogin knock-out animals. A detailed mechanism of secretagogin action in this context is still missing, however the necessity of secretagogin presence for proper signal transduction in the hypothalamus-locus coeruleus-prefrontal cortex circuit of stress response is obvious.

The data presented by Romanov et al. (2015) and Alpár et al. (2018) define a critical role of secretagogin in the molecular axis of stress responsiveness. Their results let the study of Gyengesi et al. (2013) appear in a new light: the bed nucleus of the stria terminalis functions as a regulator of the HPA-axis and many of the SCGN-positive neurons identified in this area are CRH-positive. Similarily, extrahypothalamic CRH also affects neuronal firing in the lateral habenula, a brain region that negatively regulates the dopamine-dependent reward circuit and mediates emotional responses to stress (Authement et al., 2018). Secretagogin is highly expressed in this area, as described above. However further investigations into SCGN function in this area of the brain have not been performed yet.

Dysfunction of the HPA-axis is associated with pathological developments as observed in Alzheimer's disease (AD), major depression, and serious metabolic consequences including hypoglycemia.

#### SECRETAGOGIN IN DISORDERS AND DISEASES OF THE CNS

The crucial roles of secretagogin in cellular function as defined by today already put it into an important position in the context of pathological developments.

Clinical data have suggested links between Type II diabetes (T2D) and AD, which some researchers have proposed to call a form of ''Type III diabetes.'' Animal models have shown that T2D promotes the formation of amyloid-β plaques, tau phosphorylation and neurofibrillary lesions, however in humans this evidence is scarce or ambigous. What has been shown, is that brain insulin resistance with an impaired signaling response to insulin is also a pathomechanism in AD (Arnold et al., 2018).

It is interesting to note that secretagogin's main expression in the brain occurs in regions with the highest density of insulin receptor, like the olfactory bulb, hypothalamus, hippocampus, cortex, striatum and cerebellum. De novo insulin production in the brain has been proven, but still needs further investigation. However, insulin—either secreted from neurons or entered into the brain via the blood-brain barrier—has enormous impact on neuronal function like trafficking of neurotransmitter receptors, synaptic plasticity, synapse formation and neuronal survival. It can only be speculated that secretagogin could have a role in neuronal insulin secretion.

In addition to its role in insulin secretion it has been discussed whether secretagogin expressing neurons could be protected from pathological manifestations and cell death in diseases like AD. It has been shown, that CBPs like parvalbumin, calbindin-D28K and calretinin (i.e., Ca2<sup>+</sup> buffers) protect neurons from detrimental Ca2<sup>+</sup> overload as it occurs for instance in ischemic conditions (Turovsky et al., 2018). Although secretagogin is considered to be mainly a Ca2<sup>+</sup> sensor, it also has one high-affinity Ca2<sup>+</sup> binding site and could also have some buffering capacity (Rogstam et al., 2007). With this regard, Khandelwal et al. (2017) have shown, that the Ca2<sup>+</sup> affinity of secretagogin also depends on the redox-status of the protein. Under reducing conditions the overall affinity for Ca2<sup>+</sup> significantly increases to the nM range.

Dysregulated Ca2<sup>+</sup> signaling seems also to underlie neuronal dysfunction in the development of AD (Magi et al., 2016). Key manifestations of the disease are amyloid plaques and neurofibrillary tangles from hyperphosphorylated tau protein. Post mortem studies found that colocalization of secretagogin and hyperphosphorylated tau was negligable in brains from patients with AD of Braak stage III or higher and that the number of secretagogin-positive cells was not significantly changed in brains with different tau burden (Attems et al., 2008). These findings supported the hypothesis that secretagogin-positive neurons are possibly resistant to neurodegeneration at least in this area of the brain. However, Attems et al. (2011) in a later study investigated secretagogin expression in the brains of P301L tau transgenic mice, a mouse model for tau pathology in AD. They found significant reduction of secretagogin expression in P301L homozygous mice.

A special situation obviously exists in the olfactory system of humans and primates. Here, Attems et al. (2012) identified a population of not terminally differentiated, yet synaptically integrated secretagogin-positive neurons in the olfactory tract of humans, that are selectively affected by neurofibrillary pathology in AD and seem to be lost in advanced stages of the disease. Perturbed olfaction is an early clinical sign in AD and this might involve the loss of these special shell cells. However, secretagogin-positive cells in the periglomerular layer seemed not to be affected.

A possible beneficial role of secretagogin was postulated in multiple sclerosis therapy. Patients with multiple sclerosis are recommended to take a dietary supplementation of vitamin D. However, the exact molecular mechanism behind the beneficial role of vitamin D are not well-known. Oveland et al., 2018 have recently shown that a high dose of the hormonally active 1, 25-dihydroxyvitamin-D3 (1, 25D) promotes myelin repair in the cuprizone model for de- and remyelination. They studied the brain proteome in the cuprizone model for de- and remyelination and found that 125 proteins were differentially regulated in brain tissue from 1, 25D-treated mice during remyelination, compared to placebo. Proteins upregulated in the early remyelination phase were involved in calcium binding, e.g., calretinin, S10A5 and secretagogin. Oveland et al., 2018 concluded that vitamin D may influence remyelination by mechanisms involving an increase in CBPs including secretagogin.

Although secretagogin is not co-released with glucosestimulated insulin, it is found in plasma and concentrations are elevated in T2D patients (Hansson et al., 2018). In their in vitro studies in human β-cells they found increased secretagogin release due to ER stress inducers or due to inflammatory cytokines, implicating higher plasma levels as a biomarker for islet dysfunction.

Secretagogin has also been proposed as a biomarker for tumors and in ischemic brain insults, i.e., stroke. However, its plasma concentrations were either found lower or higher than in controls, dependent on the time of blood sampling (Gartner et al., 2010; Montaner et al., 2011).

Another study demonstrated that SCGN plasma levels were significantly lower in autistic children as compared to the healthy controls (Alhowikan et al., 2017). Children with severe and mild to moderate autism had significantly lower SCGN levels than healthy controls, however, there was no significant difference between the severity of autism and SCGN levels. In schizophrenia, decreased levels of SCGN mRNA were measured in post-mortem pituitaries, as well as in serum from a cohort of living first-onset schizophrenia patients (Krishnamurthy et al., 2013).

These studies convey an insight in the complex roles of SCGN in the neuronal network, specifically in aging, neurodegeneration and neuro-psychiatric disorders. Future studies will have to pave the way for translation of novel diagnostics and to investigate, if SCGN can serve as a biomarker especially in neuro-psychiatric onsets.

#### CONCLUSION AND OUTLOOK

Secretagogin's role as a Ca2<sup>+</sup> sensor is now well established from the available experimental data. Its location and interaction with proteins involved with vesicle trafficking and in the exocytotic machinery not only implicates involvement in the release of signaling molecules, but this function has already been proven for insulin release from pancreatic βcells and the stress response in the brain (CRH-release from parvocellular neurons of the PVN in the hypothalamus and activation of norepinephrinergic neurons in the locus coeruleus). Future research will prove, if secretagogin is engaged in the release of further hormones, or eventually even neuropeptides or other neurotransmitters. With this aim in mind, single-cell transcriptomics can be a suitable methodology for expression profiling of SCGN<sup>+</sup> cells. However secretagogin's protein interaction network, as known by today, is vast. Therefore it can be assumed, that in those selected cell types, where it is expressed, it might function as a kind of Ca2+-dependent switch or moderator for more cellular processes including cytoskeletal rearrangements and modulation of cellular stress. In order to delineate individual mechanisms, further functional investigations with known interactors will have to be undertaken.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

All authors were involved in the conception of the article, MM and VT wrote the final version of the manuscript.

#### ACKNOWLEDGMENTS

We thank Marie-Louise Zach for professional graphical design of **Figure 1**.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2019.00029/full#supplementary-material


telencephalon. Eur. J. Neurosci. 31, 2166–2177. doi: 10.1111/j.1460-9568.2010. 07275.x


tumor cell- and human pancreatic islet cell-specific monoclonal antibody. J. Endocrinol. 156, 469–476. doi: 10.1677/joe.0.1560469


**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 © 2019 Maj, Wagner and Tretter. 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) and the copyright owner(s) 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.

# Emerging Roles of Neuronal Ca2+ Sensor-1 in Cardiac and Neuronal Tissues: A Mini Review

Tomoe Y. Nakamura<sup>1</sup> \*, Shu Nakao1,2 and Shigeo Wakabayashi 1,3

<sup>1</sup>Department of Molecular Physiology, National Cerebral and Cardiovascular Center Research Institute, Suita, Japan, <sup>2</sup>Department of Biomedical Sciences, College of Life Sciences, Ritsumeikan University, Kusatsu, Japan, <sup>3</sup>Department of Pharmacology, Osaka Medical College, Takatsuki, Japan

The EF-hand calcium (Ca<sup>2</sup><sup>+</sup>)-binding protein, neuronal Ca<sup>2</sup><sup>+</sup> sensor-1 (NCS-1/frequenin), is predominantly expressed in neuronal tissues and plays a crucial role in neuronal functions, including synaptic transmission and plasticity. NCS-1 has diverse functional roles, as elucidated in the past 15 years, which include the regulation of phosphatidylinositol 4-kinase IIIβ (PI-4K-β) and several ion channels such as voltagegated K<sup>+</sup> and Ca<sup>2</sup><sup>+</sup> channels, the D2 dopamine receptors, and inositol 1,4,5 trisphosphate receptors (InsP3Rs). Functional analyses demonstrated that NCS-1 enhances exocytosis and neuronal survival after injury, as well as promotes learning and memory in mice. NCS-1 is also expressed in the heart including the Purkinje fibers (PFs) of the conduction system. NCS-1 interacts with KV4 K<sup>+</sup> channels together with dipeptidyl peptidase-like protein-6 (DPP-6), and this macromolecule then composes the transient outward current in PFs and contributes to the repolarization of PF action potential, thus being responsible for idiopathic arrhythmia. Moreover, NCS-1 expression was reported to be significantly high at the immature stage and at hypertrophy in adults. That report demonstrated that NCS-1 positively regulates cardiac contraction in immature hearts by increasing intracellular Ca<sup>2</sup><sup>+</sup> signals through interaction with InsP3Rs. With the related signals, NCS-1 activates nuclear Ca<sup>2</sup><sup>+</sup> signals, which would be a mechanism underlying hormone-induced cardiac hypertrophy. Furthermore, NCS-1 contributes to stress tolerance in cardiomyocytes by activating mitochondrial detoxification pathways, with a key role in Ca<sup>2</sup><sup>+</sup>-dependent pathways. In this review, we will discuss recent findings supporting the functional significance of NCS-1 in the brain and heart and will address possible underlying molecular mechanisms.

Keywords: neuronal calcium sensor-1, frequenin, ion channel, survival, immature heart contraction, hypertrophy, nuclear Ca2+ signaling, learning and memory

# NCS-1 AND ITS INTERACTING PROTEINS

Intracellular calcium (Ca2+) is a versatile second messenger that regulates diverse cellular processes, including neurotransmission, muscle contraction, and signal transduction. Changes in intracellular Ca2<sup>+</sup> are transduced by multiple proteins, with a key role of a large family of EF-hand Ca2+-binding proteins that act as Ca2<sup>+</sup> sensors or Ca2<sup>+</sup> buffers. Ca2+-buffer proteins chelate Ca2<sup>+</sup> and

#### Edited by:

Daniele Dell'Orco, University of Verona, Italy

#### Reviewed by:

Jeanne M. Nerbonne, Washington University in St. Louis, United States William A. Coetzee, New York University, United States

> \*Correspondence: Tomoe Y. Nakamura tomoen@ncvc.go.jp

Received: 18 October 2018 Accepted: 15 February 2019 Published: 04 March 2019

#### Citation:

Nakamura TY, Nakao S and Wakabayashi S (2019) Emerging Roles of Neuronal Ca2+ Sensor-1 in Cardiac and Neuronal Tissues: A Mini Review. Front. Mol. Neurosci. 12:56. doi: 10.3389/fnmol.2019.00056 often terminate Ca2<sup>+</sup> signals (Ikura, 1996). In contrast, the binding of Ca2<sup>+</sup> to Ca2+-sensor proteins causes a large conformational change, which consequently transduces the Ca2<sup>+</sup> signal into various cellular functional changes, by regulating specific target proteins. Calmodulin is one of the best-characterized Ca2+-sensor proteins and is involved in many aspects of Ca2<sup>+</sup> signaling. Neuronal calcium sensor-1 (NCS-1) is the mammalian homolog of the Drosophila frequenin protein, which belongs to the larger NCS protein family that includes NCS-1, visinin-like proteins, recoverin, guanylate cyclase-activating proteins, and potassium channel-interacting proteins (KChIPs). The structure of NCS-1 is shown in the **Figure 1A**, indicating that it is a small (22 kDa) Ca2+-binding protein containing 4 EF-hand motifs; of these, 3 (EF2-4) bind to Ca2+. Unlike ubiquitously expressed calmodulin, NCS-1 is predominantly expressed in the brain and cardiac tissues, suggesting its specialized roles in these tissues. The Ca2+-binding affinity of NCS-1 is significantly higher than that of calmodulin (K<sup>d</sup> values of ∼300 nM vs. ∼10 µM, respectively). Although both Ca2+-binding proteins can operate within the physiological range of Ca2<sup>+</sup> levels (∼100 nM to ∼1–5 µM), the above data suggest that NCS-1 may be more sensitive to small changes in intracellular Ca2+.

The functional roles of NCS-1 are still being elucidated. Currently, known functions include the regulation of diverse target proteins, including phosphatidylinositol 4-kinase (PI-4K), voltage- and ligand-gated ion channels, and interleukin-1 receptor accessory protein like-1 (IL1RAPL1). Here, we summarize the current understanding of NCS-1 regulation of some of these target proteins and how it affects brain and cardiac functions (**Table 1**). Even though NCS-1 has a well-characterized function in these organ systems, it should be pointed out that NCS-1 may have more diverse functions in human physiology and disease, such as a potential role in oncogenesis (Jerng et al., 2004). Such roles should be the focus of future studies.

#### Phosphatidylinositol 4-Kinase

#### Role in Exocytosis and Secretion

PI-4K IIIβ (PI-4K-β) catalyzes the synthesis of phosphatidylinositol 4-phosphate, which is a late limiting step in the synthesis of phosphatidylinositol 4,5-bisphosphate, an important lipid regulator of many cellular functions including exocytosis. A yeast homolog of NCS-1 and PI-4-K interact, and NCS-1-induced activation of PI-4-K is required for yeast survival (Hendricks et al., 1999). Structural support for interaction was obtained from a recent NMR structure of Ca2+-bound yeast NCS-1 (Ncs1) in complex with an N-terminal yeast PI-4-K (Pik1) fragment (Strahl et al., 2007). This interaction was also detected in neuroendocrine cells (Koizumi et al., 2002; Scalettar et al., 2002; Rajebhosale et al., 2003; de Barry et al., 2006), neurons (Taverna et al., 2002; Zheng et al., 2005) and other cell types including pancreatic beta cells (Gromada et al., 2005) and mast cells (Kapp-Barnea et al., 2003), and was shown to facilitate exocytosis and secretion in these cells (**Table 1**). However, previous reports have suggested no direct interaction in neurons (Bartlett et al., 2000). This contradiction may be explained by the presence of newly discovered PI-4-Kβ regulators, calneurons. While calneurons interact with PI-4K-β at low Ca2<sup>+</sup> levels to inhibit its enzyme activity, NCS-1 binds to PI-4K-β at high Ca2<sup>+</sup> levels to activate it (Mikhaylova et al., 2009), suggesting that calneurons and NCS-1 compete for PI-4-K-β interaction depending on intracellular Ca2<sup>+</sup> levels. Thus, when the intracellular Ca2<sup>+</sup> level is low, its interaction might be difficult to be detected.

# Voltage-Gated KV4 K<sup>+</sup> Channels

#### Regulation of Excitability in the Brain

The V7 Drosophila mutant that overexpresses NCS-1 has a phenotype of hyperactivity, which results in the proposal that NCS-1 facilitates neurotransmission, possibly by regulating the activities of ion channels (Pongs et al., 1993; Poulain et al., 1994). Indeed, we found that NCS-1 is a Ca2+-sensitive regulatory component of a native K<sup>+</sup> current (Nakamura et al., 2001; **Table 1**). In the brain and heart, rapidly inactivating (Atype) voltage-gated K<sup>+</sup> currents control cellular excitability. Although the pore-forming alpha-subunits of these channels are considered to be KV4 channels (Serôdio et al., 1994; Fiset et al., 1997; Nakamura et al., 1997), the kinetic properties of KV4 channels differ from native A-type currents, suggesting the presence of regulatory subunits. KChIPs, a member of the NCS protein subfamily, were initially reported as a specific KV4 regulatory subunit (An et al., 2000). Because NCS-1 modulates KV4 currents similar to KChIPs, by increasing current amplitude and slowing the inactivation time course, and NCS-1 physically interacts with KV4.2 in mouse brain, it was identified as a regulator of A-type K<sup>+</sup> currents in neurons (Nakamura et al., 2001; **Table 1**).

#### Involvement in Cardiac Arrhythmia

This interaction and activation also occurs in adult mouse cardiomyocytes (Guo et al., 2002) and in zebrafish heart (Nakamura and Coetzee, 2008), which lacks KChIPs. The differential regulation of KV4 channels by NCS-1 and KChIPs in specific tissues and cell types was an unaddressed topic, and this was clearly demonstrated in the report by Nattel's group (Xiao et al., 2013). Purkinje fibers (PFs) show an unusual form of transient outward K<sup>+</sup> current Ito with slow recovery kinetics and TEA sensitivity compared with ventricular Ito, suggesting a distinct molecular composition. This group found that NCS-1 and DPP6, which were also reported to be auxiliary subunits of KV4 K<sup>+</sup> channels (Jerng et al., 2004), are preferentially enriched in PFs, while KChIP2, an essential subunit of ventricular KV4.3 is weakly expressed. Moreover, NCS-1 slowed inactivation kinetics of KV4.3, while DPP6 increased its current amplitude, thus increasing the Itomediated K<sup>+</sup> efflux (**Figure 1Ba**), which would accelerate PF repolarization and shortening of action potentials (**Figure 1Bb**; similar computer simulation was reported by Xiao et al., 2013). Thus, overexpression of KV4 auxiliary subunits may result in steep transmural repolarization gradients in PFs with adjacent ventricular tissues that induces coupled ectopic activity, and potentially leads to lethal arrhythmias (**Figure 1Bc**). NCS-1 also interacts with the anti-cancer drug taxol (Boehmerle et al., 2006), and is involved in the regulation of taxol-induced cardiac arrhythmia (Zhang et al., 2010). Thus, NCS-1 can be a potential target for anti-arrhythmic therapy.

# Voltage-Gated Ca2+ Channels

#### Neurotransmitter Release and Neurite Elongation

NCS-1 regulates voltage-gated Ca2<sup>+</sup> channels. Published data, however, are somewhat inconsistent with reports demonstrating

auxiliary subunits of KV4 channels in Purkinje fiber (PF), slows inactivation kinetics of KV4 current and increases the current amplitude, respectively, thus increase Ito-mediated K<sup>+</sup> efflux (Ba). This would accelerate PF repolarization and shortening of APD (Bb), and may lead to cardiac arrhythmias (Bc). The same concepts of the traces in (B) were originally reported by Xiao et al. (2013). (C) NCS-1 also interacts with IP3Rs on the SR and increases local Ca2+. This activates CaMKII, followed by CaMKII-dependent phosphorylation of PLB that enhances the Ca2+-pump activity of SERCa2, resulting in the increase in SR Ca2<sup>+</sup> content (Ca). This increases the global Ca2<sup>+</sup> transient and contraction in the immature heart. NCS-1 deficiency results in a smaller Ca2+-transient and contraction (Cb; the composite figure of echocardiograms and Ca2<sup>+</sup> transients are based on data from Nakamura et al., 2011). NCS-1 also increases nuclear Ca2<sup>+</sup> levels because the SR and the nuclear envelope are interconnected (Ca). NCS-1-mediated increase in nuclear Ca2<sup>+</sup> signal can promote hormone-induced cardiac hypertrophy, whereas NCS-1 deficiency prevents progression of hypertrophy (Cc; adapted from Nakamura et al., 2011). Phenylephrine is an agonist of α1-adrenergic receptor. For further details, please refer to the text. APD, action potential duration; CaMKII, calcium/calmodulin-dependent protein kinase II; DPP6, dipeptidyl peptidase-like protein 6; EF, EF-hand; InsP3R, inositol 3,4,5-trisphosphate receptor; Ito, transient outward K<sup>+</sup> current; KO, knock-out; KV, voltage-dependent potassium channel; NCS-1, neuronal Ca2<sup>+</sup> sensor-1; PLB, phospholamban; SERCa2, sarcoplasmic/endoplasmic reticulum calcium ATPase 2; SR, sarcoplasmic reticulum; WT, wild type.

both positive and negative effects (**Table 1**). NCS-1 was described to inhibit P/Q-type Ca2<sup>+</sup> channels an regulates autocrine pathways in adrenal chromaffin cells (Weiss et al., 2000; Weiss and Burgoyne, 2001) and N-type Ca2<sup>+</sup> channels in PC12 cells, which reduces neurite elongation (Gambino et al., 2007). Other studies, in contrast, have demonstrated positive regulation of N-type Ca2<sup>+</sup> channels, causing glial cell line-derived neurotrophic factor (GDNF)-induced enhancement of neurotransmitter release in motoneurons (Wang et al., 2001). Activation of P/Q-type Ca2<sup>+</sup> channels by NCS-1 causes


activity-dependent synaptic facilitation in nerve terminals (Tsujimoto et al., 2002). In Drosophila, NCS-1 enhances neurotransmission and nerve terminal growth, by functionally interacting with the α1 subunit of the voltage-gated Ca2<sup>+</sup> channel (Dason et al., 2009). The possible reason for the apparent contradictory findings is that the effects may be cell type-specific and/or mediated by accessory proteins, such as a βsubunit (Rousset et al., 2003), or dependent on other interacting proteins, such as IL1RAPL1 that cooperatively regulates the N-type Ca2<sup>+</sup> channel via NCS-1 (Gambino et al., 2007). In addition, regulation of the Ca2<sup>+</sup> channel by Ca2<sup>+</sup> influx through the channel should be considered. Well-characterized examples are Ca2+-dependent inactivation (Standen and Stanfield, 1982) and facilitation (Dolphin, 1996) of Ca2<sup>+</sup> channels regulated by other Ca2+-binding proteins, such as calmodulin (Budde et al., 2002; Christel and Lee, 2012). Future research should aim to understand the regulatory mechanisms of Ca2<sup>+</sup> channels that involve NCS-1.

#### D2 Dopamine Receptor

#### Role in Synaptic Plasticity and Psychiatric Illness

Dopamine plays an important role in the reward system of the brain. Disorders of the dopamine system result in several psychiatric and neurological conditions. Dopamine transmission is regulated by dopamine receptor-interacting proteins (DRIP), including NCS-1, calcyon, and DARPP-32. NCS-1 directly interacts with the D2 dopamine receptor, inhibits D2 receptor phosphorylation, and reduces the agonist-mediated internalization of the receptor (Kabbani et al., 2002), indicating that NCS-1 preserves dopamine signaling (**Table 1**). Indeed, modest NCS-1 overexpression in the dentate gyrus in mice promotes exploration, synaptic plasticity, and rapid acquisition of spatial memory (Saab et al., 2009). NCS-1 is upregulated in the prefrontal cortex of patients with schizophrenia and bipolar disorder (Koh et al., 2003; Bai et al., 2004). Because the levels of other DRIPs were also changed in patients with schizophrenia (Bai et al., 2004; Souza et al., 2008), DRIP signaling is possibly involved in psychiatric disorders. Furthermore, recent findings indicate the N-terminal 60 residues of NCS-1 are responsible for binding to the D2 receptor (Woll et al., 2011). Such knowledge would provide an opportunity to screen for drugs that can specifically interrupt the NCS-1-D2 dopamine receptor interaction and thus prevent psychiatric disorders.

#### Role in Learning and Memory and Possible Mechanism

Several studies have demonstrated that NCS-1 modulates learning and memory. For example, deletion or reduction of NCS-1 resulted in dysfunction of learning and memory in Caenorhabditis elegans (Gomez et al., 2001), as well as in mice (Mun et al., 2015), whereas mice overexpressing NCS-1 rapidly acquire spatial memory (Saab et al., 2009). Thus, NCS-1 affects neurophysiology, possibly through various interacting proteins. A mechanism underlying NCS-1-mediated learning and memory was further investigated (Nakamura et al., 2017). Ncs1−/<sup>−</sup> mice exhibited impaired spatial learning and memory function in the Morris Water Maze test, with slight changes in their exercise activity or a structural change in the hippocampus. However, the levels of brain-derived neurotrophic factor (BDNF), a key regulator of memory function, and dopamine were decreased. Furthermore, phosphorylation of Ca2+/calmodulin-dependent protein kinase II-α (CaMKII-α), which regulates long-term potentiation, and BDNF levels were decreased, suggesting that CaMKII-α signaling that increases BDNF production is at least partly involved in NCS-1-mediated learning and memory function.

# Inositol 1,4,5-Trisphosphate Receptors

#### Role in Neuronal Pathogenesis

Ca2<sup>+</sup> signaling via inositol 1,4,5-trisphosphate receptors (InsP3Rs) regulates cellular function and is involved in pathogenesis (**Table 1**). NCS-1 physically interacts with InsP3R1 and enhances InsP3R-mediated Ca2<sup>+</sup> signaling in rat brains. Indeed, physical/functional interaction of these proteins was directly demonstrated in an in vitro experiment showing that the addition of NCS-1 to InsP3R1 in the lipid bilayer increased InsP3R channel activity (Schlecker et al., 2006). This interaction was also detected at the growth cone region of neurites in cultured neurons, and indicated to be crucial for neurite outgrowth (Iketani et al., 2009). Metabotropic glutamate receptor-mediated cis also mediated by NCS-1/InsP3R interaction (Jo et al., 2008). Pathologically, NCS-1/InsP3R1 interaction is believed to be involved in bipolar disorder (Schlecker et al., 2006) because lithium, a medical drug for bipolar disorder, inhibited the NCS-1-induced enhancement of InsP3R function. NCS-1/InsP3R interaction is also considered to mediate neuropathy (Boehmerle et al., 2006, 2007; Blachford et al., 2009), as paclitaxel (taxol), a chemotherapeutic agent used for the treatment of solid cancers, modulates the expression/function of NCS-1, and hence InsP3R1-mediated Ca2<sup>+</sup> signaling.

#### Enhancement of Immature Heart Contraction and Hypertrophy

NCS-1/InsP3R interaction is also detected in the heart and is crucial for contraction at the immature stage and cardiac hypertrophy in adult (Nakamura et al., 2011; **Table 1** and **Figure 1C**). A high expression of NCS-1 was found in the immature heart (Nakamura et al., 2003, 2011), but its function at this stage was unknown. Using Ncs1−/<sup>−</sup> mice, Nakamura et al demonstrated that NCS-1 contributes to an increase in contraction and Ca2<sup>+</sup> signaling, specifically at the immature stage (**Figure 1Cb**). Intracellular Ca2<sup>+</sup> levels and the sarcoplasmic reticulum (SR) Ca2<sup>+</sup> content were significantly lower in Ncs1−/<sup>−</sup> myocytes at the neonatal stage than in wild-type cells. Mechanistically, the interaction of NCS-1 with InsP3R increases InsP3R-dependent Ca2<sup>+</sup> signaling, followed by the activation of CaMKII-dependent pathways, and promotes SR Ca2<sup>+</sup> pump via the phosphorylation of phospholamban (PLB), which ultimately induce increase in the SR Ca2<sup>+</sup> content and global Ca2<sup>+</sup> transient, thus cardiomyocyte contraction (**Figures 1Ca,b**). The importance of crosstalk among NCS-1, InsP3Rs, and CaMKII in the immature hearts was evident by the high expression of all three proteins in immature hearts (Nakamura et al., 2011). In the neonatal mouse heart, the structure and function of SR are immature. Nonetheless, it is considered a primary source of Ca2<sup>+</sup> necessary for muscle contractions, suggesting the existence of factors missing during development and promoting SR-dependent excitationcontraction (E-C) coupling in the postnatal stages. NCS-1 may act as one of these missing factors. Numbers of molecules which levels are high at the immature stage are often up-regulated in the disease conditions, such as cardiac hypertrophy. NCS-1 is also highly expressed during the early stages of hypertrophy in the adult heart and promotes the progression of hypertrophy, at least in part, through InsP3R activation (Nakamura et al., 2011; **Figure 1Cc**). A possible molecular mechanism is suggested in the next section.

#### Regulation of Nuclear Ca2+ Signals

The aforementioned data indicate that NCS-1 can discretely regulate different types of Ca2<sup>+</sup> signaling pathways in the heart (i.e., regulation of E-C coupling in immature heart and changes in gene expression in the adult heart). Recent evidence suggests that gene transcription is regulated by nuclear Ca2<sup>+</sup> signals. However, the mechanisms underlying nuclear Ca2<sup>+</sup> regulation and its relationship to cytoplasmic Ca2<sup>+</sup> regulation have not been completely solved. Using a subcellular-specific, fluorescent protein-based Ca2<sup>+</sup> indicator GECO (Zhao et al., 2011; Nakao et al., 2015) confirmed the following: (1) nuclear Ca2<sup>+</sup> transients were elicited by both electrical and receptor stimulations (with insulin-like growth factor-1, IGF-1) in neonatal mouse ventricular myocytes; and (2) receptor stimulation-elicited nuclear Ca2<sup>+</sup> transients were mainly mediated by InsP3Rs. Furthermore, based on the evidence that IGF-1-elicited nuclear Ca2<sup>+</sup> transient was significantly diminished in Ncs1−/<sup>−</sup> cardiomyocytes, NCS-1 is involved in the receptor stimulation-induced nuclear Ca2<sup>+</sup> regulation through interaction with InsP3Rs (Nakao et al., 2015; **Table 1**). This may contribute to NCS-1-mediated hypertrophy, which was described above. A possible mechanism underlying a dual effect of NCS-1 on nuclear Ca2<sup>+</sup> signals and E-C coupling is that NCS-1 increases the Ca2<sup>+</sup> content of SR (Nakamura et al., 2011) that is interconnected to the nuclear envelope (Wu and Bers, 2006) and consequently may increase nuclear Ca2<sup>+</sup> (Nakao et al., 2015; **Figure 1Ca**).

#### Other Functions of NCS-1 With Unknown Interacting Proteins

#### Enhancement of Neuronal Survival After Injury

Physical or chemical injury and genetic abnormalities can result in neuronal degeneration, which may underlie human neurodegenerative disorders, such as Alzheimer's disease and Parkinson's disease. Both intrinsic and extrinsic factors, including neurotrophic factors, can activate the anti-apoptotic process to rescue neuronal cell death. The signaling pathway leading to cell survival remains unresolved. In this regard, NCS-1 was found to be a novel Ca2+-dependent survival-promoting factor upregulated in injured neurons (Nakamura et al., 2006), based on the following observations. (1) NCS-1 expression increases in injured neurons; (2) NCS-1 overexpression diminished various stress-induced neuronal cell death in culture; and (3) the dominant negative EF-hand NCS-1 mutant (E120Q) accelerated cell death. Mechanistically, the expression level of NCS-1 in neuron is increased by GDNF, a neurotrophic factor upregulated by neuronal injury, and NCS-1 mediates GDNF survival signaling via the activation of the Akt pathway.

#### Role in Stress Tolerance in Cardiomyocytes

Not only in neurons, NCS-1 also plays a key role in protecting cardiomyocytes against stress through the activation of mitochondrial detoxification pathways (Nakamura et al., 2016). Excessive stress induces cytosolic Ca2<sup>+</sup> overload and cell death. In contrast, mild forms of stress lead to physiologically relevant changes in Ca2+, which activate Ca2+-dependent survival pathways by binding to Ca2+-sensor proteins. As one such protein, NCS-1 was found to play important roles in Ca2+-dependent survival signaling. Ncs1−/<sup>−</sup> myocytes were more susceptible to oxidative and metabolic stress, and cellular ATP levels, mitochondrial respiration and biosynthesis were significantly reduced in these cells. In wild-type myocytes, mild oxidative stress increased the mitochondrial proton leak, which exerted a protective effect by inhibiting the production of reactive oxygen species. However, this response was diminished in Ncs1−/<sup>−</sup> cardiomyocytes, thus resulting in cell death. Similar susceptibility was also observed in Ncs1−/<sup>−</sup> hearts subjected to ischemia-reperfusion injury. In these hearts, molecules regulating Ca2+-dependent survival pathways, such as Akt and PGC-1α, which promote mitochondrial biogenesis and function, were significantly downregulated compared to wild-type hearts. These data demonstrate a novel role of NCS-1 that contributes to stress tolerance in cardiomyocytes, partly by the activation of Ca2+-dependent survival pathways. NCS-1 may also participate in cardioprotection by mediating receptor-signaling pathways. For example, NCS-1 associates with, and modulates, adenosine receptor activity (Navarro et al., 2012). Given the central role of adenosine in mediating the protective effects of ischemic preconditioning (Cohen and Downey, 2008), it is entirely possible that the cardioprotective effects of NCS-1 is partially mediated by this pathway.

# CONCLUSION

Recently, studies have elucidated new roles of NCS-1 in physiology and pathophysiology. In this review, we have mainly focused on NCS-1 in the neuronal system and heart. Our particular interest is the emerging theme that NCS-1 directly regulates the function of several ion channels that permeate Ca2<sup>+</sup> (e.g., several types of voltage-gated Ca2<sup>+</sup> channels, ionotropic dopamine receptors, and InsP3Rs), suggesting a general role of Ca2<sup>+</sup> influx via the channel that binds to NCS-1 and consequently regulates channel functions and/or downstream Ca2+-dependent signaling, which affect various neuronal and cardiac functions. Furthermore, many established roles of NCS-1 are related to protective responses of cells against exogenous stress that leads to mild increases in cytosolic Ca2+. This suggest that intracellular Ca2<sup>+</sup> as a determinant of cell survival and cell death, and Ca2+-sensor proteins, such as NCS-1, may serve as a switch to proceed the signal. We believe that this review provides intriguing observations and compels researchers to conduct detailed investigations and extend their studies on NCS-1 and its important regulatory proteins.

### AUTHOR CONTRIBUTIONS

TN conducted most part of research, wrote, organized, and finalized the article. SN did some experiments and wrote some part of the article. SW contributed to the discussion on all part

#### REFERENCES


of the study and wrote some part of the manuscript. SN and SW wrote and edited some parts of the article.

#### FUNDING

This work was supported by Grant Number JP24590293, 15K08200 JSPS KAKENHI, The Naito Science and Engineering Foundation #J004, and the Salt Science Research Foundation (Grant No. 1547 and 1643), and Intramural Research Fund for Cardiovascular Disease of the National Cerebral and Cardiovascular Center #22-2-3 obtained by TN; Grant Number JP26460312, JSPS KAKENHI obtained by SW; Grant Number JP25860174, JSPS KAKENHI obtained by SN.

learning and memory in C. elegans. Neuron 30, 241–248. doi: 10.1016/s0896- 6273(01)00276-8


**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 © 2019 Nakamura, Nakao and Wakabayashi. 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) and the copyright owner(s) 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.

# Evolutionary-Conserved Allosteric Properties of Three Neuronal Calcium Sensor Proteins

Valerio Marino1,2 \* and Daniele Dell'Orco<sup>1</sup>

<sup>1</sup> Section of Biological Chemistry, Department of Neurosciences, Biomedicine, and Movement Sciences, University of Verona, Verona, Italy, <sup>2</sup> Department of Translational Research and New Technologies in Medicine and Surgery, University of Pisa, Pisa, Italy

Neuronal Calcium Sensors (NCS) are highly conserved proteins specifically expressed in neurons. Calcium (Ca2+)-binding to their EF-hand motifs results in a conformational change, which is crucial for the recognition of a specific target and the downstream biological process. Here we present a comprehensive analysis of the allosteric communication between Ca2+-binding sites and the target interfaces of three NCS, namely NCS1, recoverin (Rec), and GCAP1. In particular, Rec was investigated in different Ca2+-loading states and in complex with a peptide from the Rhodopsin Kinase (GRK1) while NCS1 was studied in a Ca2+-loaded state in complex with either the same GRK1 target or a peptide from the D<sup>2</sup> Dopamine receptor. A Protein Structure Network (PSN) accounting for persistent non-covalent interactions between amino acids was built for each protein state based on exhaustive Molecular Dynamics simulations. Structural network analysis helped unveiling the role of key amino acids in allosteric mechanisms and their evolutionary conservation among homologous proteins. Results for NCS1 highlighted allosteric inter-domain interactions between Ca2+-binding motifs and residues involved in target recognition. Robust long range, allosteric protein-target interactions were found also in Rec, in particular originating from the EF3 motif. Interestingly, Tyr 86, involved the hydrophobic packing of the N-terminal domain, was found to be a key residue for both intra- and inter-molecular communication with EF3, regardless of the presence of target or Ca2<sup>+</sup> ions. Finally, based on a comprehensive topological PSN analysis for Rec, NCS1, and GCAP1 and multiple sequence alignments with homolog proteins, we propose that an evolution-driven correlation may exist between the amino acids mediating the highest number of persistent interactions (high-degree hubs) and their conservation. Such conservation is apparently fundamental for the specific structural dynamics required in signaling events.

Keywords: NCS1, recoverin, GCAP1, protein structure network, molecular dynamics, neuronal calcium sensors

# INTRODUCTION

Calcium (Ca2+) is a universal second messenger whose changes in concentration contribute to the regulation of a variety of biological processes ranging from muscle contraction (Ebashi and Endo, 1968) to signal transduction (Berridge et al., 2003) and neuronal signaling (Augustine et al., 2003). The biological importance of Ca2<sup>+</sup> ion requires a family of proteins called Ca2+-sensor proteins,

#### Edited by:

Ildikó Rácz, Universitätsklinikum Bonn, Germany

#### Reviewed by:

John G. Partridge, Georgetown University, United States Pere Garriga, Universitat Politecnica de Catalunya, Spain

#### \*Correspondence:

Valerio Marino valerio.marino@univr.it

Received: 29 October 2018 Accepted: 11 February 2019 Published: 07 March 2019

#### Citation:

Marino V and Dell'Orco D (2019) Evolutionary-Conserved Allosteric Properties of Three Neuronal Calcium Sensor Proteins. Front. Mol. Neurosci. 12:50. doi: 10.3389/fnmol.2019.00050 which are able to change conformation in response to variations in intracellular Ca2<sup>+</sup> concentration. Calmodulin is a ubiquitous Ca2+-sensor protein (Vetter and Leclerc, 2003) whose structural plasticity allows the regulation of a plethora of biological targets (Ikura and Ames, 2006). Neuronal Calcium Sensor proteins (NCS) though, are a tissue-specific and highly specialized class of proteins able to regulate a limited number of targets (Burgoyne, 2007; Burgoyne and Haynes, 2015) involved in a large array of neuronal transmission processes. Some sensors like Calcium and Integrin Binding Protein 2 (CIB2), for instance, are involved in hearing and vision (Riazuddin et al., 2012; Patel et al., 2015; Michel et al., 2017; Vallone et al., 2018), Recoverin (Rec) and Guanylate Cyclase Activating Proteins (GCAPs) are fundamental players of the phototransduction cascade (Koch and Dell'orco, 2013, 2015) and Neuronal Calcium Sensor 1 (NCS1) regulates sight (Baksheeva et al., 2015), synaptic plasticity (Saab et al., 2009), and neuronal differentiation (Dason et al., 2012). Moreover, NCS proteins may have only one predominant biological target, as is as case for Rec and GCAP1, or even regulate the same effectors, as is the case for the regulation of Rhodopsin Kinase (GRK1) by both Rec and NCS1.

A number of high resolution three-dimensional structures are available for some NCS proteins in their isolated states or in complex with peptides from their native targets (Ames and Lim, 2012). While being fundamental for understanding the mechanisms regulated by NCS proteins at atomistic resolution, static structures offer just a frozen picture of the indeed highly dynamic nature of such proteins and of the conformational changes that constitute the essence of their function. If a rigorous and consistent sampling of the conformational space is achieved, Molecular Dynamics (MD) simulations can complement structural information with time evolution to synergistically investigate even small conformational changes (Marino et al., 2015b) in NCS proteins (Marino and Dell'orco, 2016). Non-covalent interactions are continuously formed and broken in protein dynamic processes, however the most persistent ones are the physical forces responsible for the assembly of native protein structures. The time evolution of such persistent interactions monitored by MD simulations defines Protein Structure Network (PSN) rearrangements, which are a direct consequence of conformational variations in the proteins. PSN analysis can help identify allosteric properties (Schueler-Furman and Wodak, 2016; Greener and Sternberg, 2018) as well as residues of crucial importance in NCS proteins such as GCAP1 (Marino and Dell'orco, 2016).

Here we present a thorough analysis of PSN derived by MD simulations of Rec and NCS1 to assess potential allosteric mechanisms of intra- and inter-molecular information transfer that are related to specific signaling states. Moreover, we investigate the communication routes from Ca2+-binding sites to target interface, which allows us to identify residues that play an important role in establishing state-specific topology and in target communication. Finally, by analyzing residue conservation in homologs and among NCS1, Rec, and GCAP1, we infer evolutionary conserved properties regarding target interfaces and topological features of the PSN.

# METHODS

# Structural Modeling of Rec and NCS1

The starting structures for MD simulations were either specific frames of deposited PDB files or partially modeled to avoid structural gaps. The structures with the highest resolution and sequence coverage available were chosen for both Rec and NCS1, resulting in the following models: Ca2+-free Rec ("tense Recoverin" or Rec-T, Tanaka et al., 1995), Rec with one Ca2<sup>+</sup> bound to EF3 ("intermediate state Recoverin" or Rec-I, Ames et al., 2002), Ca2+-loaded Rec ("relaxed Recoverin" or Rec-R, Ames et al., 1997), Ca2+-loaded Rec bound to GRK1 peptide (Rec-GRK1, Ames et al., 2006; Zernii et al., 2011; Ames and Lim, 2012), Ca2+-loaded uncomplexed NCS1 ("isolated NCS1" or NCS1-iso), Ca2+-loaded NCS1 bound to D<sup>2</sup> Dopamine receptor peptides (NCS1-D2R) and Ca2+-loaded NCS1 bound to Rhodopsin Kinase peptide [NCS1-GRK1, (Bourne et al., 2001; Pandalaneni et al., 2015)]. Details for molecular modeling are provided in **Supplementary Methods**.

#### Molecular Dynamics Simulations

MD simulations of Rec and NCS1 states were performed using GROMACS 2016.1 simulation package (Abraham et al., 2015) and CHARMM36m (Huang et al., 2017) all-atom force field, where parameters for N-myristoylated Gly were generated manually (available upon request). Simulations were performed in a dodecahedral box with Periodic Boundary Conditions applied, where proteins were located at 12 Å distance from box boundaries. Solvent was modeled as TIP3P water, system was neutralized first with 1 mM MgCl2, then after addition of 150 mM KCl. The size for each simulated system was the following: 33,837 atoms for Rec-T, 38,878 atoms for Rec-I, 56,010 atoms for Rec-R, 86,456 atoms for Rec-GRK1, 46,809 atoms for NCS1-iso, 42,882 atoms for NCS1- D2R and 50,584 atoms for NCS1-GRK1.

All structures underwent substantially the same preproduction steps as previous studies concerning GCAP1 (Marino et al., 2015b), briefly consisting of steepest descent (F = 1,000 kJ/mol∗nm) and conjugate gradients (F = 500 kJ/mol∗nm) energy minimization, then equilibration at 310 K for 2 ns backbone-constrained and 2 ns unrestricted MD simulations in NVT ensemble. After equilibration, 200 ns unrestrained MD simulations were performed in NPT ensemble at 310 K

**Abbreviations:** NCS, Neuronal Calcium Sensors; NCS1, Neuronal Calcium Sensor 1; Rec, Recoverin; GCAP1, Guanylate Cyclase Activating Protein 1; MD, Molecular Dynamics; GRK1, Rhodopsin Kinase; D2R, D<sup>2</sup> Dopamine receptor; PSN, Protein Structure Network; CIB2, Calcium and Integrin Protein 2; "tense Recoverin" or Rec-T, Ca2+-free Rec; "intermediate state Recoverin" or Rec-I, Rec with one Ca2<sup>+</sup> bound to EF3; "relaxed Recoverin" or Rec-R, Ca2+-loaded Rec; Rec-GRK1, Ca2+ loaded Rec bound to GRK1 peptide; "isolated NCS1"or NCS1-iso, Ca2+-loaded uncomplexed NCS1; NCS1-D2R, Ca2+-loaded NCS1 bound to D<sup>2</sup> Dopamine receptor peptides; NCS1-GRK1, Ca2+-loaded NCS1 bound to Rhodopsin Kinase peptide; PCn, n-th Principal Component; ES, Essential Subspace; RMSIP, Root Mean-Square Inner Product; LDn, n-th Linear Discriminant; CR, Communication Robustness; SB, Selective Betweenness; s.i., sequence identity; hEn, entering

helix of n-th EF-hand; hFn, exiting helix of n-th EF-hand; MSA, Multiple Sequence Alignment.

and 1 atm for each system. Equilibration and production MD simulations were independently replicated 5 times by changing the random seed for initial velocity generation as previously described (Marino and Dell'orco, 2016), to achieve exhaustive sampling of the conformational space.

### Principal Component Analysis and Linear Discriminant Analysis

Collective protein motions were identified using Principal Component Analysis (PCA) after diagonalization of the Cα covariance matrix (Amadei et al., 1999) calculated on single 200 ns replicas and concatenated 1 µs trajectories for each protein form, after superimposition to the final structure of the equilibration phase of replica one. Eigenvalues and eigenvectors extracted from the diagonalized matrix represent amplitude and direction of collective motions (Principal Components, or PC) and are ranked in decreasing order of eigenvalues, meaning that the first PC (PC1) accounts for the largest collective motion, PC2 for the second largest and so on.

The Essential Subspace (ES) of the first 20 PC, accounting for 78–93% of total motion of Rec and NCS1 states, was identified as in Marino and Dell'orco (2016) to compare the conformational space sampled by each replica and 1 µs trajectories using the Root Mean-Square Inner Product (RMSIP) index (Amadei et al., 1999). Details for cosine content (c1) (Hess, 2002) and RMSIP index calculations are provided in **Supplementary Methods**.

The sampling convergence of MD simulations was also assessed by performing Linear Discriminant Analysis (LDA, Martinez and Kak, 2001) on the projection of the frames of the 5 replicas on PC1 and PC2 calculated on the concatenated 1 µs trajectories. This supervised classification method decreases the number of features describing data and combines them to find a hyperplane that maximizes separation between means of projected classes and minimizes the variance within each projected class. In our case a two-dimensional feature space (PC1 and PC2) was classified using a single Linear Discriminant (LD1) and if a single conformation from a replica was classified as possibly belonging to a different replica, then the two replicas could be considered as consistent and therefore concatenable. Data presented in **Figures S3**, **S4** was smoothed using Kernel Density Estimation smoothing (Rosenblatt, 1956; Parzen, 1962).

#### Protein Structure Network Generation and Analysis

Dynamic information from concatenated 1 µs trajectories was converted into a PSN using PyInteraph software (Tiberti et al., 2014), with the same parameters as in Marino and Dell'orco, 2016, the only exception being the mass of atoms defined according to CHARMM36m force field (Huang et al., 2017). Briefly, non-bonded interaction between side chains (H-bonds, electrostatic and hydrophobic interactions) was computed as a percentage of frames where distance and angle constraints were satisfied. Interactions were filtered using the hydrophobic cluster size criterion (Vishveshwara et al., 2009), therefore the size of the biggest hydrophobic cluster was calculated at 0.1% persistence intervals, and the persistence threshold (pT) was computed as in Marino and Dell'orco (2016) and rounded at the lowest decimal value. For each case p<sup>T</sup> was calculated and all interactions above the threshold were joined in the unweighted PSN graph representing the specific protein state.

The degree of connectivity was calculated for all residues in each PSN, amino acids with ≥ 4 connections were considered hubs and reported in **Tables S1–S3**. The Communication Robustness index (CR, Marino and Dell'orco, 2016) of two residues permits the evaluation of the structural information transfer between amino acids within a protein threedimensional structure, and to highlight potential allosteric mechanisms. Details for CR calculation are provided in **Supplementary Methods**. CR index was therefore computed to measure the communication between Rec bidentate Ca2+ coordinating residues E85 and E121 and GRK1 interface residues (**Table S4**, Ames et al., 2006; Zernii et al., 2011). Analogously for NCS1, CR index was calculated between bidentate Ca2+ coordinating residues E84, E120, and E168 and with GRK1 or D2R interface residues (**Table S4**, Pandalaneni et al., 2015). For the sake of clarity, only interface residues with the 5 highest CR values for each presented state were reported in **Figures S5**, **S6**. Moreover, intramolecular communication was evaluated by analyzing CR calculated between the myristoyl group of Rec and GRK1-interface residues (**Figure S7A**) and among the previously mentioned Ca2+-coordinating residues of Rec and NCS1 (**Figure S7B**), to investigate intramolecular communications between EF-hand motifs.

Finally, also intermolecular communication was investigated by computing CR index between GRK1 peptide residues that were structurally solved (L6-I16) complexed with both Rec and NCS1 and their respective EF-hand representatives (**Figure S8**).

For visualization purposes, pathways connecting EF-hand representatives and GRK1 residues with the highest CR were chosen according to the highest cumulative Selective Betweenness (SB, Marino and Dell'orco, 2016). Details for SB calculation are provided in **Supplementary Methods**.

#### Sequence Alignment of Homologous NCS

Protein sequences for bovine Rec (Uniprot: P21457), human NCS1 (Uniprot: P62166), human GCAP1 (Uniprot: P43080) were subjected to Multiple Sequence Alignment (MSA) with Clustal Omega (Sievers et al., 2011) to evaluate residue conservation among homologous NCS. Moreover, each protein sequence was also subjected to MSA with Clustal Omega with up to 250 Uniprot sequences having at least 50% sequence identity (s.i.) and 80% sequence coverage with respect to the longest sequence in the cluster. In detail, Rec was aligned with 122 sequences from Uniref50\_P21457 and NCS1 was aligned with 250 sequences from Uniref50\_P62166. Human GCAP1 was instead aligned with 101 sequences from Uniref50\_P46065, referring to the bovine sequence, constituting the seed for the linked UniRef50 cluster (s.i. 94%).

Residue conservation in UniRef50 clusters shown in **Figure 6** and in **Tables S1-S3, S5** was calculated as the ratio between the number of sequences where the residue of the seed sequence was conserved and the total number of sequences in the cluster. When a residue was a hub in different signaling states, even with a different degree, only one occurrence of the residue was considered for the calculation of the average of a given hub degree. Average values for each protein were mediated and subjected to linear regression after passing Shapiro-Wilk (Shapiro and Wilk, 1965) (p = 0.05) and constant variance (p = 0.05) tests, with R <sup>2</sup> = 0.62.

Hub degree shown in **Figure 5** represents the highest hub degree exhibited by a residue in one of the simulated states for each protein, data about GCAP1 are taken from Marino and Dell'orco (2016).

Pairwise sequence alignments were performed with Needleman-Wunsch (Needleman and Wunsch, 1970) algorithm.

### RESULTS

#### Consistent and Exhaustive Conformational Sampling of Rec and NCS1 Signaling States Achieved by Molecular Dynamics Simulations

Rec-T, Rec-I, Rec-R, and Rec-GRK1 were simulated to evaluate the network of persistent interactions and any potential difference in the inter- and intra-molecular allosteric properties. Such differences are ascribable to the physiological conformational transition of the sensor as a consequence of the decrease of intracellular [Ca2+] upon activation of the phototransduction cascade (Ames et al., 1996). On the other hand, NCS1-iso, NCS1-GRK1, and NCS1-D2R were simulated to evaluate differential communication of the same protein in the absence and in the presence of different targets. Furthermore, the presence of the same target, namely GRK1, interacting with both Rec and NCS1 allowed for a comparative analysis of the communication routes between Ca2+-binding sites and a common target.

For each protein state 5 × 200 ns replicas of MD simulations were run; individual replicas and 1 µs concatenated trajectories were subjected to PCA of Cα motions to assess consistency and exhaustiveness, as previously described (Marino and Dell'orco, 2016). Cosine content c<sup>1</sup> of PC1 was the first necessary condition to achieve sampling convergence, as it was demonstrated that values close to 1 are indicative of insufficient sampling (Hess, 2000; Papaleo et al., 2009). The c<sup>1</sup> for the concatenated trajectories of Rec ranged between 0.0006 and 0.1244, those for NCS1 ranged between 0.0150 and 0.1597, while the average c<sup>1</sup> of the five replicas ranged between 0.4670 and 0.5792 for Rec and between 0.1421 and 0.3302 for NCS1. The decrease of c<sup>1</sup> values from the average of the single replicas to that of the concatenated trajectories suggests that the longer timescale allows for a more significant and exhaustive conformational sampling.

The RMSIP of the first 20 PC of the ES highlighted a substantial overlapping, and therefore reproducibility, of the conformational subspaces sampled by each individual replica and by the concatenated 1 µs trajectories, as shown by RMSIP values for NCS1 and Rec, ranging from 0.763 to 0.895 (**Figure S1**), and from 0.665 and 0.828, respectively (**Figure S2**).

LDA was performed on the projection of MD trajectories along PC1 and PC2 for each protein state. Results showed that for each couple of replicas in all NCS1 (**Figure S3**) and Rec (**Figure S4**) states there was at least one conformation that could belong to two different replicas, thus allowing these replicas to be treated as one concatenated trajectory.

Overall, all results confirm that individual 200 ns MD replicas are independent and consistent representations of the dynamic structural behavior of both Rec and NCS1 in their different signaling states. Moreover, concatenated 1 µs trajectories exhaustively sample each protein conformational space and can be subjected to further detailed PSN analyses.

#### Time-Persistent Non-covalent Interactions Involved in Intra- and Inter-molecular Communication Define Protein-Specific Structural Networks

Non-covalent interactions are crucial for defining protein threedimensional structure, yet hydrophobic, electrostatic, and Hbond interactions can have a transient or more persistent essence, depending on whether they are involved in structural information transfer or not. Therefore, we monitored the persistence of these interactions throughout the concatenated 1 µs trajectories of Rec and NCS1 signaling states and determined their significance in shaping the specific PSN (Vendruscolo et al., 2002; Atilgan et al., 2004; Papaleo, 2015) for each protein state. In the resulting graph residues are nodes and persistent interactions represent edges.

Dynamic information from 1 µs MD simulations was encoded in a static PSN (see Methods) and the analysis of graphs identified the presence of key residues (hubs) persistently interacting with many close-contact amino acids (**Tables S1–S3**). To discriminate whether the identified interactions were significant or transient, p<sup>T</sup> was calculated for each trajectory to filter out short-lived interactions (see Methods) and was found to range between 15.7 and 17.6 for Rec states and between 36.5 and 37.2 for NCS1 variants.

#### PSN Analysis Identifies Peculiar Residues With a Key Role in Intramolecular Information Transfer

To assess which ones among the interface residues are responsible for transmitting structural information regarding the Ca2+ loading state to their target, the CR index was calculated for each residue of NCS1 (**Figure S5**) and Rec (**Figure S6**) target interfaces with each respective Ca2+-coordinating Glu residues chosen as representative for each EF-hand motif.

Results for NCS1, summarized in **Figure 1**, highlight that in the absence of target peptides (orange arrows) there is a long-range communication from EF2 (**Figures 1A,D**) to both EF-hands of the C-domain. In detail, L101 (**Figure 1A**) and Q130 (**Figure 1D**) belong to the entering (hE) and exiting (hF) helix of EF3, respectively, and F169 (**Figures 1A,D**) is located at helix hF of EF4 (hF4). On the other hand, communication from EF3 (**Figures 1B,E**) and EF4 (**Figures 1C,F**) is overall shortranged in the absence of the target, reaching for both interfaces only residues located in adjacent EF2 or EF4. Specifically, E120 communicates with F64 and V68 (**Figure 1B**) or with V68 and

F72 (**Figure 1E**), which are all located on helix hE2. The EF4 representative residue E168 exhibits robust communication with M121 and I124 of helix hF3 (**Figure 1C**) or W103 and Q130, located, respectively, on helix hE3 and hF3.

In the presence of GRK1 (blue arrows) all three EFhands show allosteric intradomain properties: EF2 shows robust communication with F55 of helix hE1 and F169 of helix hF4. (**Figure 1A**); EF3 exhibits high CR values with M131 of helix hF3 and to a lower extent with F55 of helix hE1; finally, EF4 robustly communicates with L107, located on the adjacent helix hE3, and with Y52 of helix hF1.

The pattern of structural information transfer appears different in the presence of another biological target. Overall, communication of NCS1 EF-hands with D2R interface travels shorter distances (**Figures 1D–F**, blue arrows). In detail: EF2 communicates with hE1 residue F56 and hE3 residue Q130 (**Figure 1D**); EF4 communicates with hE3 residue L107 and hF3 residue I128 (**Figure 1F**); finally, EF3 communication reaches Y129, also belonging to hF3 helix, and the distant W30 of helix hE1, the only residue exhibiting long-range communication (**Figure 1E**). In summary, allosteric communication between EF-hands and target interface is a peculiar feature of NCS1- GRK1 complex and of the EF2 motif of NCS1-iso, while robust communication is on average shown on a mid-to-short range by NCS1-iso EF3 and EF4 and by NCS1- D2R complex.

Results for Rec, summarized in **Figure 2**, highlighted robust short-range communication from EF2 to GRK1 interface in the absence of target. In Rec-T (**Figure 2A**) the absence of Ca2<sup>+</sup> ions bound to the EF-hands is communicated from EF2 to F23 (hE1) and Y53 (hF1). In Rec-I (**Figure 2B**), where only EF3 is occupied by a Ca2<sup>+</sup> ion, the communication of unoccupied EF2 is robust with F57 (hF1) and L90 (hF2). Finally, in Rec-R (**Figure 2C**), with both EF-hands occupied by Ca2+, again EF2 communicates with hE1 residue F23 and hF2 residue L90.

The presence of the GRK1 target changed somewhat the scenario. Rec-GRK1 complex, indeed, revealed a strikingly high allosteric communication between EF2 (E85) and hF4 residues E189 and K192 (**Figure 2D**), a residue previously identified as crucial for the interaction with GRK residue F3 (Zernii et al., 2011). Long-range communication was observed also when originated in EF3, as shown by the high CR exhibited by Rec-T L90, located in hF2 (**Figure 2A**), the previously mentioned Y53 in Rec-I (**Figure 2B**), hE1 residue F35 in Rec-R (**Figure 2C**), and hF1 residue F49 in Rec-GRK1 complex (**Figure 2D**). Surprisingly, regardless of the presence of Ca2<sup>+</sup> or GRK1, EF3 showed a significant robust communication with the first residue of helix hF2, namely Y86. Moreover, Y86 counterpart in NCS1 (F85) was identified as one of the most robustly communicating interface residues with EF2 and EF4, particularly in the absence of targets (**Figure S5**).

The N-terminal myristoyl group of Rec was also found to robustly communicate with GRK1 interface residues (**Figure S7A**) in the absence of Ca2<sup>+</sup> (Rec-T), but this robustness decreased on average upon Ca2+-binding to EF3 (Rec-I). Moreover, no CR index could be calculated upon Ca2+-binding to EF2 (Rec-R and Rec-GRK1) due to the extrusion of the fatty acid from the hydrophobic crevice via myristoyl switch mechanism (Tanaka et al., 1995), which prevented any persistent interaction with the protein.

Finally, CR index was calculated among EF-hands of Rec and NCS1 (**Figure S7B**), highlighting substantial medium-range information transfer between EF2 and EF3 of Rec only in the absence of Ca2+. NCS1, on the other hand, showed no significant communication between adjacent EF-hands, but the NCS1- GRK1 complex exhibited a surprisingly high robustness in the allosteric communication between EF2 and EF4 (**Figure S7B**).

### Intermolecular Communication Pathways Between Ca2<sup>+</sup> Binding Sites and Common Target

Intermolecular communication between GRK1 and EF-hands of Rec and NCS1 was assessed via CR index to compare how the same target could be regulated by two proteins with overlapping interfaces. Results (**Figure S8**) highlighted that GRK1 residue L6 communicated specifically with NCS1-EF4, T8 with NCS1- EF3, A11 with Rec-EF3 and S13 with NCS1-EF2. Interestingly, residue N12 of GRK1 was found to communicate with both Rec-EF2 and NCS1-EF3, while F15 with EF2 in NCS1 and EF3 in Rec. Overall, the analysis highlighted that the robustness of NCS-target interaction is highly protein-dependent in terms of specificity of the signaling state.

In addition, the previously mentioned GRK1 residues with highest CR were investigated from a topological standpoint concerning the most probable PSN pathways that would allow information transfer to/from EF-hands. The most probable pathways were chosen according to the highest collective SB, nevertheless in some cases multiple pathways with the same score were reported (**Figure 3**). Pathway analysis (**Figure 3**) indicated that the pathway from NCS1 EF2 to S13 (**Figure 3A**; **Video V1**) included Ca2<sup>+</sup> ion and that the three GRK1 residues of the pathway, namely L6, V9 and S13, were also involved in the communication of the peptide with NCS1 EF3 and EF4. The pathway from NCS1 EF3 to N12 was the most branched out (**Figure 3B**; **Video V2**) with two forks and a double intermolecular communication required to correctly route Ca2+ binding information to V9. In addition, EF3-N12 pathway shared the GRK1 residues V9 and S13 with EF2-S13 pathway and both GRK1-I16 and NCS1-Y108 residues with I16 EF4-L6 pathway (**Figure 3C**; **Video V3)**. Pathway EF4-L6 shared GRK1 residue L6 with EF2-S13 pathway.

The same pathway analysis performed on Rec EF2 with GRK1-N12 (**Figure 3D**; **Video V4**) and EF3 with GRK1- A11 (**Figure 3E**; **Video V5**) suggested that each EF-hand independently communicates with the target, with no overlapping of the pathways.

Overall, PSN pathway analysis indicated that intermolecular information transfer concerning Ca2+-binding was routed by selective pathways connecting EF-hands with GRK1.

#### Evolutionary Conservation of Hub Residues Identified in PSN

The connectivity of a PSN is an interesting feature that can be used to compare different states of the same protein and to

identify hub residues (**Tables S1–S3**), that is amino acids involved in persistent interactions with many other residues, therefore playing a key role in maintaining network topology (Raimondi et al., 2011; Fanelli et al., 2013; Marino and Dell'orco, 2016).

An example of PSN is represented in **Figure 4**, where all persistent interactions in Ca2+-loaded NCS1 and Rec are shown. Although the number of connections (and consequently hubs) identified in NCS1 and Rec are different due to intrinsic protein properties, the highest degree hubs (degree = 8) in both proteins share their function, as NCS1 D73 and Rec D74 are the first Ca2+-coordinating residues of EF2 Ca2+-binding loop, while NCS1 D109 and Rec D110 are the equivalent for EF3.

Hub analysis was performed for all NCS1 (**Table S1**) and Rec (**Tables S2**, **S3**) states and was compared to our previously published data (Marino and Dell'orco, 2016) on EF3-Mg2<sup>+</sup> and Ca <sup>2</sup>+-loaded human GCAP1 (**Table S5**, and **Table S4** in the reference).

Multiple Sequence Alignment (**Figure 5**) showed that these three NCS proteins have 9 identical and 6 conserved hubs whose maximum degree was at least 5 in any of the simulated states. Interestingly, 4 identical hubs represent the first and the last residue of the Ca2+-binding loops of EF2 and EF3, implying that any mutation of these residues could have a detrimental effect on protein structure/function relationship, as previously demonstrated (Kitiratschky et al., 2009; Dell'orco et al., 2010; Vocke et al., 2017; Marino et al., 2018).

Finally, multiple sequence alignment was run for each protein with their respective UniRef50 cluster (proteins sharing at least

50% s.i. and 80% sequence coverage), and the conservation of each hub residue among homolog proteins was calculated (**Tables S1**–**S3**, **S5**) and compared to their degree (**Figure 6**). While no apparent mathematical relationship could describe this behavior in general, a clear, protein-specific trend could be identified. All three proteins reported a decrease of the average conservation concurrent with the decrease of the hub degree, with a 62% probability to be linearly correlated. Although the R 2 value of 0.62 is not sufficiently high to clearly infer a correlation between the average conservation and the hub degree, it has to be noticed that there is an intrinsic limitation in the number of points in the x-axis, as the hub degree is a discreet quantity. Therefore, R 2 value is greatly affected by the presence of some poorly conserved (<50%) high degree hub residues of Rec, namely I155, V87, E181, and K194.

# DISCUSSION

NCS1, Rec, and GCAP1 belong to a highly conserved family of proteins called NCS, which are able to specifically regulate a limited number of biological targets in a Ca2+-dependent manner. Such homologous proteins share a very similar fold despite a relatively low s.i., specifically: (human) NCS1 and (bovine) Rec have a 44% s.i. (**Figure S9A**), NCS1 and (human) GCAP1 have a 32% s.i. (**Figure S9B**) and Rec and GCAP1 have a 31% s.i. (**Figure S9C**).

The higher s.i. shown by Rec and NCS1 with respect to that exhibited with GCAP1 is probably related to their structural similarities, also reflected in the regulation mechanisms. As previously reported, Rec and NCS1 undergo a Ca2+-dependent conformational change known as myristoyl-switch (Ames and Lim, 2012), where the fatty acid chain is extruded from a highly hydrophobic crevice upon Ca2+-binding, allowing target interfaces to be solvent-exposed. On the other hand, the myristoyl group in GCAP1 is buried regardless of [Ca2+] and mediates communication between N- and C-domains (Marino and Dell'orco, 2016) via the mechanism known as myristoyltug (Peshenko et al., 2012). In addition, the three NCS proteins analyzed in the present study have intrinsically diverse structural dynamics, as assessed by the significantly different p<sup>T</sup> values exhibited by each protein state, which may be ascribable to the

specificity of their targets. The only common target is in fact GRK1, shared by NCS1 and Rec.

Analysis of the PSN deriving from thorough MD simulations has been proven to be a valuable tool for the investigation of allosteric properties of NCS proteins (Marino and Dell'orco, 2016). Robust long-range communication from functional EF-hands to target interface was exhibited by both NCS1- GRK1 (**Figure 1**, left), Rec-GRK1 (**Figure 2**), and to a lesser extent by NCS-D2R (**Figure 1**, right) complexes. The NCS1- GRK1 complex showed also a strikingly robust communication (**Figure S7B**) between the C-terminal low affinity Ca2+-binding site EF4 (Bourne et al., 2001; Aravind et al., 2008) and the


N-terminal high affinity EF2, similarly to what was previously reported for GCAP1 (Marino and Dell'orco, 2016).

white, conserved residues are highlighted in gray, identical residues are highlighted in red.

Interestingly, Rec-Y86 was found to be particularly important for mediating communication as to the cation loading state of EF3 in all signaling states (**Figure 2**), while its counterpart NCS1-F85 is involved in robust communication between target interface and EF2 and EF4 in the absence of the target (**Figures S5A,D,F**). The importance of the two evolutionary conserved residues resides in their peculiar spatial location as both residues belong to a network of tightly packed hydrophobic interactions (**Figure 7**) involving EF1 and EF2, which is eventually responsible for the stabilization of the entire C-terminal domain.

By comparing state-specific PSN (**Figure 4**) of NCS proteins, we identified evolutionary conserved hub residues (**Figure 5**), whose degree appears to have a certain degree of correlation with the conservation among homologs (**Figure 6**). Such residues evidently play an essential role in converting Ca2+-sensing information from EF-hands to a conformational switch required for the regulation of the targets (**Tables S1–S3**). On the other hand, even in the presence of the same biological target, as is the case for the regulation of GRK1 by Rec and NCS1, intermolecular interaction pathways with the targets are substantially different (**Figure 3**). Specific intra- and intermolecular networks of communication are therefore chosen for optimizing the unique function of the NCS protein, despite the high conservation of the hub residues with the family.

Hub residues Rec-E85, Rec-E121, NCS1-E84, NCS1-E120, GCAP1-E75, GCAP1-E111, are conserved in all three NCS proteins, which is not surprising since they are directly involved in Ca2+-coordination. Interestingly, the NCS1-E120Q mutation was proven to abolish Ca2<sup>+</sup> binding in the highest affinity site EF3 of NCS1 and subsequently the myristoylswitch mechanism (Weiss et al., 2000), to impact stability and unfolding profiles (Muralidhar et al., 2005) and ultimately to prevent D<sup>2</sup> receptor desensitization (Kabbani et al., 2002).

FIGURE 6 | Correlation between hub degree and average residue conservation in UniRef50 clusters for Rec, NCS1, and GCAP1. Average residue conservation in UniRef50 clusters is plotted against hub degree for Rec (green diamonds), NCS1 (blue circles), and GCAP1 (red squares). Residue conservation in UniRef50 clusters shown in Figure 6 and in Tables S1–S3,S5 was calculated as the ratio between the number of sequences where the residue of the seed sequence was conserved and the total number of sequences in the cluster. When a residue was a hub in different signaling states, even with a different degree, only one occurrence of the residue was considered for the calculation of the average of given hub degree. Linear regression of average residue conservation is shown as a black line.

Similar conclusions were drawn for Rec-E85Q mutation (Ames et al., 2002), which prevented Ca2+-binding to the high affinity site EF2 and impaired myristoyl switch. Other conserved hubs, though, were responsible for correctly routing intermolecular information in either NCS1 and GCAP1 (namely F69/F60, D73/D64, K100/K91, respectively) or GCAP1 and Rec (namely GCAP1-I119 and Rec I129).

Our topological analysis of the PSN resulting from proteinspecific dynamics also highlights amino acids that are crucial

for the physiological function of GCAP1 and suggests that any mutation in these key positions may lead to dysfunctional states. In fact, we previously identified GCAP1 hub residues that, when mutated, were associated with retinal dystrophies (Marino and Dell'orco, 2016), namely: D100, which is the target of the D100E/G substitutions (Kitiratschky et al., 2009; Dell'orco et al., 2010; Nong et al., 2014); L84, which is the target of the L84F substitution (Kamenarova et al., 2013; Marino et al., 2015a); Y99, associated with the Y99C mutation (Payne et al., 1998; Sokal et al., 1998); E155, associated with the E155A and E155G mutations (Wilkie et al., 2001); I143, found to be mutated in I143T/N (Nishiguchi et al., 2004). Other conserved residues highlighted in the present study, namely L176 and E111, were recently found to be associated with retinal dystrophies (Vocke et al., 2017; Marino et al., 2018). This group of genetic diseases is characterized by a dysregulation of the second messengers Ca2<sup>+</sup> and cGMP homeostasis caused by different molecular phenotypes (Koch and Dell'orco, 2013; Dell'orco et al., 2014). Most of the identified mutations indeed affect GCAP1 Ca2<sup>+</sup> affinity, making such mutants unable to correctly inhibit the target (retinal guanylate cyclase 1 mostly) upon light detection. The decreased Ca2+ affinity is either caused by mutations directly involved in Ca2+ coordination, as is the case for D100, E111, E155, or by structural effects of mutations, either neighboring ion-coordinating EFhand loops, such as Y99 and I143, or spatially distant like L84 and L176.

In conclusion, our unbiased PSN analysis highlighted the presence of disease-associated residues in the ensemble of the most robust communication pathways ensuring the correct conformational switch of GCAP1, suggesting a detrimental effect of the point mutation for the protein dynamics under physiological conditions. Thorough topological and dynamic analyses such as the ones performed here could be therefore extended to other cases and eventually they could help understanding at high level of resolution the molecular basis of diseases affecting key protein-protein interactions in a signaling pathway.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

# AUTHOR CONTRIBUTIONS

VM and DD conceived the study. VM performed MD simulations and data analysis, and wrote the manuscript with contributions from DD.

# FUNDING

The financial support of Fondazione Telethon-Italy (grant no. GGP16010 to DD) is gratefully acknowledged.

# ACKNOWLEDGMENTS

MD simulations were performed at the Computational Platform of the Centro Piattaforme Tecnologiche of University of Verona. Technical support with Python scripting by Alberto Borsatto is gratefully acknowledged.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2019.00050/full#supplementary-material

### REFERENCES


novel mutation (D100G) in the GUCA1A gene. Doc. Ophthalmol. 128, 59–67. doi: 10.1007/s10633-013-9420-z


**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 © 2019 Marino and Dell'Orco. 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) and the copyright owner(s) 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.

# Calcium, Dopamine and Neuronal Calcium Sensor 1: Their Contribution to Parkinson's Disease

Cristina Catoni<sup>1</sup> , Tito Calì<sup>2</sup> and Marisa Brini<sup>1</sup> \*

<sup>1</sup> Department of Biology, University of Padova, Padua, Italy, <sup>2</sup> Department of Biomedical Sciences, University of Padova, Padua, Italy

Parkinson's disease (PD) is a debilitating neurodegenerative disorder characterized by loss of dopaminergic neurons in the substantia nigra pars compacta. The causes of PD in humans are still unknown, although metabolic characteristics of the neurons affected by the disease have been implicated in their selective susceptibility. Mitochondrial dysfunction and proteostatic stress are recognized to be important in the pathogenesis of both familial and sporadic PD, and they both culminate in bioenergetic deficits. Exposure to calcium overload has recently emerged as a key determinant, and pharmacological treatment that inhibits Ca2<sup>+</sup> entry diminishes neuronal damage in chemical models of PD. In this review, we first introduce general concepts on neuronal Ca2<sup>+</sup> signaling and then summarize the current knowledge on fundamental properties of substantia nigra pars compacta dopaminergic neurons, on the role of the interplay between Ca2<sup>+</sup> and dopamine signaling in neuronal activity and susceptibility to cell death. We also discuss the possible involvement of a "neglected" player, the Neuronal Calcium Sensor-1 (NCS-1), which has been shown to participate to dopaminergic signaling by regulating dopamine dependent receptor desensitization in normal brain but, data supporting a direct role in PD pathogenesis are still missing. However, it is intriguing to speculate that the Ca2+-dependent modulation of NCS-1 activity could eventually counteract dopaminergic neurons degeneration.

#### Edited by: Daniele Dell'Orco, University of Verona, Italy

#### Reviewed by:

Rosanna Parlato, University of Ulm, Germany Teresa Duda, Salus University, United States

> \*Correspondence: Marisa Brini marisa.brini@unipd.it

Received: 25 October 2018 Accepted: 14 February 2019 Published: 22 March 2019

#### Citation:

Catoni C, Calì T and Brini M (2019) Calcium, Dopamine and Neuronal Calcium Sensor 1: Their Contribution to Parkinson's Disease. Front. Mol. Neurosci. 12:55. doi: 10.3389/fnmol.2019.00055

# NEURONAL CALCIUM SIGNALING

Calcium (Ca2+) homeostasis is essential for neuronal function and survival. Intracellular Ca2<sup>+</sup> signaling in neurons is extremely fine-tuned, because it controls gene transcription, membrane excitability, neurotransmitters secretion and many other cellular processes, including synaptic plasticity (Berridge, 1998; Brini et al., 2014). Like other cells, neurons use both extracellular and intracellular sources of Ca2<sup>+</sup> and, as a consequence of their excitability, they are exposed to large Ca2<sup>+</sup> fluctuations and thus to a major risk of Ca2<sup>+</sup> overload.

Keywords: calcium signaling, Cav1.3 calcium channel, ncs-1, dopamine, Parkinson's disease

The coordinated action of the different systems that handle Ca2<sup>+</sup> fluxes guarantees the generation of high Ca2<sup>+</sup> concentration microdomains with precise spatiotemporal features that are crucial to specifically activate different cellular processes (La Rovere et al., 2016; Filadi et al., 2017a; Samanta and Parekh, 2017). For instance, those generated upon the opening of the endoplasmic reticulum Ca2<sup>+</sup> channels are sensed by mitochondria that use them to drive

bioenergetic metabolism for the production of ATP and mitochondrial substrates for anabolic process (Tarasov et al., 2012).

However, exaggerated mitochondrial Ca2<sup>+</sup> accumulation may be dangerous, since can lead to mitochondrial permeability transition pore (mPTP) opening, cytochrome c release and can activate apoptotic cell death (Bernardi et al., 2015). Thus, once Ca2+-regulated processes have been engaged, Ca2<sup>+</sup> ions must be rapidly extruded (and/or buffered) to avoid that their excessive accumulation could trigger mitochondrial dysfunction (Calì et al., 2012a; Muller et al., 2018). The "Ca2<sup>+</sup> machinery" that is in place to tune Ca2<sup>+</sup> concentration includes transport proteins such as channels, exchangers and pumps that move the ion across the membranes (i.e., the plasma membrane and the membranes of organelles), and Ca2<sup>+</sup> binding proteins that act as Ca2<sup>+</sup> buffer and/or transducer (**Figure 1**).

Increasing evidence suggests that defective Ca2<sup>+</sup> handling plays an important role in aging and neurodegeneration (Berridge, 1998; Calì et al., 2014; Pchitskaya et al., 2018). Despite of neurodegenerative diseases are a large group of heterogeneous disorders characterized by relative selectivity in the death of neuronal subtypes, they share some common tracts that include disturbance in cellular quality mechanisms (i.e., ER stress, autophagy, accumulation of aggregated proteins), oxidative stress, neuroinflammation and defective Ca2<sup>+</sup> signaling (Brini et al., 2014; Hetz and Saxena, 2017; Kurtishi et al., 2018; Muller et al., 2018). Furthermore, recent studies have indicated that defective ER-mitochondria communication, by impinging on energetic metabolism, lipid synthesis, autophagy, could have detrimental consequences for cell function and survival (Filadi et al., 2017b). Many regulators of ER-mitochondria interface are proteins whose mutations are linked to familial forms of Alzheimer's disease (AD) and PD, suggesting that defects at the ER-mitochondria contact sites could have a role in the onset and/or the progression of these neurodegenerative diseases (Calì et al., 2013a; Filadi et al., 2016; Area-Gomez and Schon, 2017).

As mentioned above, in addition to the Ca2<sup>+</sup> transport across the membranes, another important mechanism that contributes to the regulation of Ca2<sup>+</sup> homeostasis is the processes of Ca2<sup>+</sup> buffering that is managed by mitochondrial Ca2<sup>+</sup> uptake but largely relies on the existence of several cytosolic Ca2<sup>+</sup> binding proteins. Among them, the ubiquitous EF-hand Ca2<sup>+</sup> protein calmodulin (CaM) is mainly responsible for translating the increases of the cytosolic Ca2<sup>+</sup> concentration into a biochemical signal through conformational changes of its targets (Sharma and Parameswaran, 2018). It is present at high concentration in the brain, where it plays a key role in the regulation of ions channels activity and synaptic plasticity (Xia and Storm, 2005). Other Ca2+-binding proteins such as Calbindin D-28K (CB-28K), calretinin (CR), and parvalbumin (PV) are also present in the nervous system. By buffering Ca2<sup>+</sup> levels with different capacity, affinity and kinetics (Schwaller, 2012; Paillusson et al., 2017) and thanks to their cell-specific abundance, they guarantee the selective activation of different biological processes. Celltype-specific distribution of Ca2<sup>+</sup> binding proteins could also account for the selective susceptibility to cell death of the specific neuronal populations affected in different neurodegenerative diseases. Indeed, it has been observed that CB-28K containing cells are spared from cell death in pharmacological-induced parkinsonism in mice and that CB-28K immunoreactivity in cholinergic neurons of the basal forebrain (the same neurons that are preferentially loss in AD) was reduced in an age-related manner in humans, suggesting a role for CB-28K also in the selective neuronal vulnerability in AD (Yamada et al., 1990; German et al., 1992; Mouatt-Prigent et al., 1994; Damier et al., 1999; Geula et al., 2003; Zallo et al., 2018).

#### PARKINSON'S DISEASE, CALCIUM AND SELECTIVE VULNERABILITY OF SUBSTANTIA NIGRA PAR COMPACTA

PD is the second most common neurodegenerative disorder in humans after AD. PD patients present motor symptoms such as resting tremor, bradykinesia and postural rigidity. However, the appearance of other disturbances such as constipation, sleep disorders, olfactory deficit, apathy, pain, sexual difficulties, and in some case cognitive decline is currently observed to anticipate motor deficits in many patients (de Lau and Breteler, 2006) and indicates that regions of the brain that are not involved in motor symptoms are also compromised. At histological levels, the hallmarks for PD are the selective loss of the dopamine (DA)-containing neurons of the substantia nigra pars compacta (SNc) and the presence of proteinaceous cytosolic inclusions known as Lewy bodies, mainly constituted by alpha-synuclein (Goedert et al., 2013). The progressive SNc DA cells death leads to decreased DA levels and the worsening of the symptoms. SNc DA cells release DA from their axonal terminals and from their cell bodies and dendrites within both the dorsal striatum and the midbrain, respectively. DA release is crucial for voluntary movement and it is strictly Ca2+- and electrical activity-dependent. Indeed, the continuous supply of DA to the connected brain areas is guaranteed by autonomous pacemaking, which occurs in the absence of conventional synaptic input and thank to the orchestrated action of different ion channels. In particular, the presence of voltagedependent L-type Ca2<sup>+</sup> channels containing a distinctive Cav1.3 pore forming subunit, which opens at relatively hyperpolarized potentials, allows Ca2<sup>+</sup> entry with an oscillatory pathway that contributes to the membrane potential threshold, underlying autonomous pacemaking (Chan et al., 2007; Puopolo et al., 2007; Guzman et al., 2010). Continuous Ca2<sup>+</sup> influx is necessary to modulate physiological DA release by SNc DA neurons, but, its long-lasting presence may synergize with the exposure to risk factors (i.e., aging, mitochondrial toxins, mutations) and generate metabolic stress and mitochondrial damage (Surmeier et al., 2011; Guzman et al., 2018).

It is widely recognized that in PD, the major risk of Ca2+ induced toxicity is represented by Ca2<sup>+</sup> entry through the voltage gated Ca2<sup>+</sup> channels during the normal activity of the dopaminergic neurons (Ilijic et al., 2011; Liss and Striessnig, 2019), that, in this way, become more vulnerable to death than other neuronal populations. Cell damage could be further exacerbated by environmental factors such as exposure to

mitochondrial toxins [i.e., MPTP (1-methyl-4-phenyl-1,2,3,6 tetrahydropyridine), rotenone, 6-hydroxy dopamine (6-OHDA), paraquat (1,1<sup>0</sup> -dimethyl-4,4<sup>0</sup> -bipyridine)] or upon loss of function of specific proteins such as alpha-synuclein, Parkin, PINK1 and DJ-1, whose mutations are linked to genetic forms of PD. Interestingly, all these proteins, despite their different intracellular localization and function, are able to interfere with Ca2<sup>+</sup> signaling (Calì et al., 2014). Indeed extracellular and intracellular deposition of alpha-synuclein aggregates has been proposed to enhance Ca2<sup>+</sup> influx through the plasma membrane by forming pore-like structures (Danzer et al., 2007; Surguchev and Surguchov, 2015; Angelova et al., 2016) or by interfering with Ca2<sup>+</sup> channels (Liu et al., 2013; Ronzitti et al., 2014), as well as PINK1 has been proposed to participate to the regulation of both influx or efflux of Ca2<sup>+</sup> ions from mitochondria (Gandhi et al., 2009; Marongiu et al., 2009). We have found that the overexpression of PD-linked alpha-synuclein, parkin and DJ-1 proteins enhanced ER-mitochondria Ca2<sup>+</sup> transfer by favoring ER-mitochondria juxtaposition, and provided evidence that through this action, physiological amounts of these proteins are able to tune ATP production (Calì et al., 2012b, 2013b; Ottolini et al., 2013). The loss of this function is likely to be particularly damaging to neurons that are heavily dependent on proper Ca2+signaling and ATP production. Accordingly, Paillusson et al. (2017) have documented loss of ER-mitochondria

association in induced pluripotent stem cells derived neurons from PD patients harboring alpha-synuclein gene triplication.

In summary, if by one side Ca2<sup>+</sup> entry through Cav1.3 pore subunit is essential to sustain pacemaking activity of SNc DA neurons, by the other it exposes these neurons to metabolic burden and mitochondrial stress. Differently, DA neurons from the ventral tegumental area (VTA), which are also autonomous pacemakers, are significantly less vulnerable than SNc DA neurons from which they differ in respect with two main features: they have smaller Ca2<sup>+</sup> currents (Khaliq and Bean, 2010) and strong intrinsic Ca2<sup>+</sup> buffering capacity due to higher calbindin levels (Dopeso-Reyes et al., 2014).

The most convincing argument in favor of the "Ca2<sup>+</sup> hypothesis" in PD onset is that epidemiologic studies on patients under clinical trial with L-type channel antagonists for the treatment of hypertension have shown a reduced risk of developing PD (Becker et al., 2008; Ritz et al., 2010; Pasternak et al., 2012). The voltage gated L-type Ca2<sup>+</sup> plasma membrane channels inhibitor isradipine has been demonstrated to be neuroprotective in a mouse model of PD (Ilijic et al., 2011) and phase III of clinical trial is currently under evaluation to establish whether treatment with isradipine is able to slow the progression of PD in humans (Liss and Striessnig, 2019).

Despite general consensus agrees with the fact that the anatomical, physiological, and biochemical phenotype of the SNc

DA neurons predisposes them to mitochondrial dysfunction, the molecular bases of the subtype-selective neuronal vulnerability are still obscure and of big interest.

Interestingly, computer imaging and immunohistochemical staining techniques have revealed a strict correlation between the distribution of the Ca2+-binding proteins CB-28K and CR and cell survival in midbrain dopaminergic regions: cells that are spared from death in animals treated with the DA neurotoxin MPTP (German et al., 1992; Mouatt-Prigent et al., 1994) are those that display higher expression levels of CB-28K and CR in control untreated animals. Interestingly, this observation has been reinforced by a comparative study performed on postmortem brain from neurologically normal individuals and PD patients in which the distribution of calbindin, calmodulin and calretinin did not associate with the regions prone to neurodegeneration. It has also been observed that the expression of Cav1.3 subtypes increased in the brain of patients at early stage of the disease, even before the appearance of recognized pathological signs (Hurley et al., 2013), suggesting that Ca2<sup>+</sup> dysregulation could be an early event in PD pathogenesis.

Low expression levels of Ca2+-binding proteins in the brain area more susceptible to cell death in PD suggest that those neuronal populations are characterized by low Ca2<sup>+</sup> buffering capacity. This parameter has been directly evaluated in neurons from the ventral and medial SNc by applying a protocol originally developed by E. Neher (Neher and Augustine, 1992; Zhou and Neher, 1993; Neher, 1998). Foehring and colleagues (Foehring et al., 2009) have loaded the cells with an exogenous Ca2+ indicator/buffer and calculated the Ca2<sup>+</sup> binding ratio (KS) by measuring changes in Ca2+-bound buffer and dividing by the free Ca2<sup>+</sup> increase. Interestingly, despite the intrinsic Ca2<sup>+</sup> buffering in DA cells increases with postnatal age (K<sup>S</sup> ' 110 at postnatal day 13–18 and '179 at postnatal day 25–32), it remains low at both age ranges. Other neuronal populations (e.g., neocortical pyramidal cells or cortical GABAergic interneurons), that are not endowed with pacemaking activity, display similar or higher values and Purkinje cells have the highest K<sup>S</sup> values (∼2,000) (Fierro and Llano, 1996).

Considering that, in addition to Ca2<sup>+</sup> binding proteins, also mitochondria play a role in buffering cytosolic Ca2+, a reduction of mitochondria amount or/and the presence of dysfunctional mitochondria could account for differences in Ca2<sup>+</sup> buffering capacity among different neuronal midbrain populations. In line with these considerations, a study has found that the mitochondrial mass in SNc DA neurons is reduced in respect with that of other neurons from the midbrain (Liang et al., 2007). Thus, also this peculiarity may account for selective vulnerability of DA SNc neurons.

At the end of this discussion, it is worth to mention that other observations suggest that additional sources of Ca2<sup>+</sup> (other than Ca2<sup>+</sup> entry from the extracellular ambient) could contribute to SNc DA neurons vulnerability. In this respect, defects in intracellular Ca2<sup>+</sup> stores handling and ER stress have been frequently documented in PD cellular models (Wang and Takahashi, 2007; Mercado et al., 2013).

All together it is clear that the equilibrium between Ca2<sup>+</sup> signaling and SNc DA neurons activity is extremely critical: upon conditions of increased metabolic demand, i.e., when continuous dopamine release into the dorsal striatum is required for movement, elevated metabolic burden could originate a vicious cycle that further impairs mitochondrial function, resulting in increased metabolic stress. Interestingly, it has been proposed that Ca2<sup>+</sup> load may further contribute to exacerbate neurodegeneration by promoting an increase of the neurotoxic catecholamine intracellular levels (Mosharov et al., 2009).

#### DOPAMINE RELEASE AND NEURONAL CALCIUM SENSOR 1: POSSIBLE IMPLICATIONS IN PARKINSON DISEASE?

Among the Ca2+-binding proteins, the components of the subfamily of Neuronal Ca2<sup>+</sup> Sensors (NCS) are particularly abundant in neurons and photoreceptors and deserve special attention since their properties distinguish them from CaM or CB-28K, CR and PV and allow them to play non-reduntant roles. Differences in Ca2<sup>+</sup> affinities, in cellular expression and distribution and in target proteins are at the basis of the specialization of NCS function (McCue et al., 2010). Neuronal Ca2<sup>+</sup> Sensor-1 (NCS-1) is the most ancient member of the family (Pongs et al., 1993), and it is implicated in the regulation of cell-surface receptors and ion channels, and in neurotransmitter release, gene transcription, cell growth and survival (Burgoyne, 2007).

NCS-1 has been linked to a large spectrum of diseases possibly because its differential interaction with partners. Changes in the abundance of NCS-1 result in altered relationship with target proteins and determine cell dysfunction. An up-regulation of NCS-1 mRNA was found in a variety of non-neurological and neurological diseases. NCS-1 has been proposed to be a biomarker in aggressive breast cancer (Moore et al., 2017). In the heart, altered Ca2<sup>+</sup> signaling mediated by NCS-1 and inositol 1,4,5 trisphosphate receptor interaction was linked to cardiac arrhythmias (Zhang et al., 2010). Schizophrenia, bipolar disorder (BD) (Koh et al., 2003) and autism (Piton et al., 2008; Handley et al., 2010) have been associated with upregulation or mutations in NCS-1 protein.

Increased levels of NCS-1 mRNA were measured in neurons from SNc of PD patients (Dragicevic et al., 2014) and NCS-1 was shown to co-localize with the D2 type-dopamine receptors in dendrites, spines, and occasionally in axonal buttons of rat and monkey striatal neurons (Kabbani et al., 2002), thus supporting the involvement of NCS-1 in the process of dopaminergic signaling, but also suggesting its possible link with PD.

As mentioned above, numerous convincing biophysical and pharmacological findings support the hypothesis that Cav1.3 channels by sustaining pacemaker-activity exposes SNc DA neurons to continuous Ca2<sup>+</sup> load and mitochondrial stress (Surmeier et al., 2011). However, other studies investigating dopamine receptor mediated autoinhibition of neuronal activity have shown that Ca2<sup>+</sup> entry through Cav1.3 channels, in addition to sustain pacemaker activity, regulates dopamine autoreceptors (Dragicevic et al., 2014). Considering that current

therapies to alleviate PD symptoms and progression are based on the administration of dopamine precursor L-Dopa and/or dopamine D2 receptor agonists (Oertel and Schulz, 2016), the understanding of Cav1.3 physiology becomes crucial to better define the pathways involved in PD onset and develop therapeutic strategies.

Dopaminergic transmission is dependent on two main families of DA receptors, namely D1- and D2-type (Beaulieu and Gainetdinov, 2011) that are both members of the G proteincoupled receptor (GPCR) superfamily. The D1-like receptors activate Gαs/olf and stimulate cAMP production, whereas the D2-like receptors activate Gαi/<sup>o</sup> and inhibit adenylate cyclase activity and cAMP production. The two DA receptor types differ in their localization: the D1-like receptors are predominately localized post-synaptically (Levey et al., 1993), whereas the D2 like receptors are present post-synaptically on dopaminergic target neurons (Levey et al., 1993; Sesack et al., 1994), but pre-synaptically and as autoreceptors (D2-AR) on DA neurons (Mercuri et al., 1997; L'hirondel et al., 1998). The response of SNc neurons to DA is highly regulated and chronic loss of dopamine leads to receptor sensitization (Schultz and Ungerstedt, 1978). In particular, DA binding to the D2-AR leads to activation of G-protein-coupled, inwardly rectifying potassium channels (GIRK2) (Luscher and Slesinger, 2010; Beaulieu and Gainetdinov, 2011) that promotes K<sup>+</sup> efflux and hyperpolarization, and consequently reduces SNc DA activity (Beckstead et al., 2004). At the same time, however, D2-AR internalization occurring in response to DA stimulation reduces the DA-induced inhibitory effect on SNc DA neurons firing and tonic Ca2<sup>+</sup> entry through L-type voltage channels promotes desensitization of D2 receptor-dependent activation of GIRK channels (Gantz et al., 2015).

In other words, DA itself, upon release, acts in a negative feedback loop: by binding to D2-subtype receptors, it inhibits SNc DA neurons electrical activity and further DA release, but both Ca2<sup>+</sup> influx and receptor desensitization limit this action.

Dragicevic et al. (2014) have observed that, in contrast to juvenile SNc neurons, mature neurons have lost D2 autoreceptors desensitization, and, accordingly, upon in vivo exposure to high DA level also juvenile neurons present the same D2-autoreceptors desensitizing response. According to their results, Cav 1.3 mediated Ca2<sup>+</sup> influx is essential for age-dependent modulation of somatodendritic D2-autoreceptors responses and D2 autoreceptor sensitization requires both Cav1.3 and NCS-1 activation.

NCS-1 and D2 receptors co-localize both in primate and rodent brain (Kabbani et al., 2002) and NCS-1 attenuates agonistinduced receptor internalization via a mechanism that involves a reduction in D2 receptor phosphorylation. Interestingly, amino acid substitutions that affect NCS-1 Ca2<sup>+</sup> binding ability abolished its modulation on D2 receptor signaling (Kabbani et al., 2012) and NCS-1 deletion in mouse has been reported to decrease DA secretion (Ng et al., 2016), thus implying important contribution of NCS-1 impairment in defective dopaminergic signaling.

The finding that, in juvenile mice, Cav1.3 can adapt SNc DA neurons activity in response to high extracellular DAlevels by providing the Ca2<sup>+</sup> source for neuronal Ca2<sup>+</sup> sensor NCS-1 (Dragicevic et al., 2014) strongly indicates the existence of an adaptive signaling network (Cav1.3/NCS-1/D2/GIRK2)

motor function. Dopamine binding to D2-autoreceptors (D2-AR) on SNc DA neurons controls their firing rate by promoting the inhibitory effect of GIRK2 K<sup>+</sup> channels and, at the same time, the D2-AR internalization thus contributing to desensitization process. NCS-1 participates in the regulation of dopaminergic signaling since upon Cav 1.3 channels-mediated Ca2<sup>+</sup> influx it becomes active and, by blocking D2-AR internalization, prevents receptors desensitization. Upon pharmacological treatment with L-DOPA and/or Cav.1.3 antagonist isradipine a vicious loop may be activated: D2-AR desensitization could be facilitated, since reduced Ca2<sup>+</sup> influx may affect NCS-1 inhibition of receptor internalization and the ability of dopamine to inhibit neuronal activity may be compromised, altogether leading to excitotoxicity. This possibility deserves further investigation, even if no evidence in this direction has been provided so far.

that may have protective role by preventing D2 autoreceptors desensitization. A simplified model that summarizes this concept is shown in **Figure 2**. According to it, increases in the intracellular Ca2<sup>+</sup> concentration activate NCS-1 that opposes somatodendritic D2-autoreceptors internalization and blocks their desensitization counteracting in this way the inhibitory effect mediated by GIRK2 channels on Cav1.3 and finally promotes dopamine release also through this mechanism. Apparently, this could result in a sort of vicious circle that exacerbate Ca2<sup>+</sup> entry. However, no desensitization was found during development in KO mice for Cav1.3 and no evidence for exacerbated excitotoxicity upon treatment with the dihydropyridine L-type Ca2<sup>+</sup> channel blocker isradipine has been reported so far, thus suggesting that other compensatory mechanisms intervene.

In line with this suggestion, loss of Cav1.3 (or its pharmacological inhibition) does not severely compromise pacemaking activity both in juvenile and adult SNc DA neurons, but rather altered its precision and regular occurrence (Poetschke et al., 2015). The appearance of compensatory response due both to NCS-1 upregulation and to the existence of alternative Ca2<sup>+</sup> source in SNc DA cells that are able to mediate NCS-1/D2-AR interactions could explain the findings. Indeed, an upregulation of both T-type Ca2<sup>+</sup> Cav 1.2 channels and NCS-1 protein has been found in Cav1.3 KO mice (Poetschke et al., 2015).

All together these observations strongly support the idea that Ca2<sup>+</sup> and DA are critical components in the disease and underline the complexity of their interplay in the modulation of dopaminergic response.

#### CONCLUSION

The distinctive physiology of the DA midbrain neurons within the SNc has attracted attention as possible reason for their selective vulnerability. Slow rhythmic activity (accompanied by oscillations in intracellular Ca2<sup>+</sup> concentration) and high dendritic arborization distinguishes these neurons from the other neurons in the brain. Cav1.3 mediated Ca2<sup>+</sup> influx is essential to sustain DA release, to guarantee high energy demands that are required for this function and to provide necessary amount of ATP at axonal and dendrites sites. But if continuous Ca2<sup>+</sup>

#### REFERENCES


entry sustains DA secretion and mitochondrial metabolism, at the same time it exposes cells to "Ca2<sup>+</sup> stress," that may synergize with intrinsic low Ca2<sup>+</sup> buffering capacity, aging, mutations or mitochondria damage and culminate in cell degeneration. In vitro and in vivo studies strongly implicated Cav1.3 activity in the high vulnerability of SNc DA neurons, however the complexity of DA metabolism that includes an autoregulatory nature of DA secretion underlines that selective vulnerability of SNc neurons is still an obscure issue. The characterization of the Cav1.3 Ca2<sup>+</sup> channels physiology and of the alternative pathways that are engaged to compensate pharmacological inhibition of Cav.1.3 channels upon isradipine treatment certainly deserves more investigations. The outcome of isradipine phase III clinical trial will shed light on these aspects.

At this point we can conclude that the deciphering of the molecular mechanisms involved in dopaminergic signaling is the best we can do to develop therapeutic strategy, but we have to be aware that the complexity of the system is increased by interactive pathways that are engaged in compensatory mechanisms and this makes the investigations very challenging.

### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the supplementary files.

#### AUTHOR CONTRIBUTIONS

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

# FUNDING

This work was supported by grants from the Università degli Studi di Padova (Progetto di Ateneo 2015 n. CPDA 153402 to MB, Progetto Giovani 2012 n.GRIC128SP0 to TC and Progetto di Ateneo 2016 n. CALI\_SID16\_01 to TC ) and from the Ministry of University and Research (Bando SIR 2014 n. RBSI14C65Z to TC).



dopaminergic neurons is attenuated by DJ-1. Nature 468, 696–700. doi: 10.1038/ nature09536



**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 © 2019 Catoni, Calì and Brini. 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) and the copyright owner(s) 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.

# NCS-1 Deficiency Is Associated With Obesity and Diabetes Type 2 in Mice

Olga Ratai , Joanna Hermainski † , Keerthana Ravichandran and Olaf Pongs\*

Center for Integrative Physiology and Molecular Medicine (CIPPM), Institute for Cellular Neurophysiology, University of the Saarland, Homburg, Germany

Neuronal calcium sensor-1 (NCS-1) knockout (KO) in mice (NCS-1<sup>−</sup>/<sup>−</sup> mice) evokes behavioral phenotypes ranging from learning deficits to avolition and depressive-like behaviors. Here, we showed that with the onset of adulthood NCS-1<sup>−</sup>/<sup>−</sup> mice gain considerable weight. Adult NCS-1<sup>−</sup>/<sup>−</sup> mice are obese, especially when fed a high-fat diet (HFD), are hyperglycemic and hyperinsulinemic and thus develop a diabetes type 2 phenotype. In comparison to wild type (WT) NCS-1<sup>−</sup>/<sup>−</sup> mice display a significant increase in adipose tissue mass. NCS-1<sup>−</sup>/<sup>−</sup> adipocytes produce insufficient serum concentrations of resistin and adiponectin. In contrast to WT littermates, adipocytes of NCS-1<sup>−</sup>/<sup>−</sup> mice are incapable of up-regulating insulin receptor (IR) concentration in response to HFD. Thus, HFD-fed NCS-1<sup>−</sup>/<sup>−</sup> mice exhibit in comparison to WT littermates a significantly reduced IR expression, which may explain the pronounced insulin resistance observed especially with HFD-fed NCS-1<sup>−</sup>/<sup>−</sup> mice. We observed a direct correlation between NCS-1 and IR concentrations in the adipocyte membrane and that NCS-1 can be co-immunoprecipitated with IR indicating a direct interplay between NCS-1 and IR. We propose that NCS-1 plays an important role in adipocyte function and that NCS-1 deficiency gives rise to obesity and diabetes type 2 in adult mice. Given the association of altered NCS-1 expression with behaviorial abnormalities, NCS-1<sup>−</sup>/<sup>−</sup> mice may offer an interesting perspective for studying in a mouse model a potential genetic link between some psychiatric disorders and the risk of being obese.

#### Edited by:

Daniele Dell'Orco, University of Verona, Italy

#### Reviewed by:

Tomoe Y. Nakamura, National Cerebral and Cardiovascular Center, Japan Lee Haynes, University of Liverpool, United Kingdom

> \*Correspondence: Olaf Pongs oupon@t-online.de

#### †Present address:

Joanna Hermainski, Sysmex Europe GmbH, Norderstedt, Germany

> Received: 19 December 2018 Accepted: 11 March 2019 Published: 03 April 2019

#### Citation:

Ratai O, Hermainski J, Ravichandran K and Pongs O (2019) NCS-1 Deficiency Is Associated With Obesity and Diabetes Type 2 in Mice. Front. Mol. Neurosci. 12:78. doi: 10.3389/fnmol.2019.00078 Keywords: neuronal calcium sensor-1, obesity, adipocyte, insulin receptor, insulin resistance, depression

# INTRODUCTION

Sensing and regulating intracellular levels of calcium are essential for proper function of neuronal and non-neuronal cells. This includes important cellular processes such as the release of neurotransmitters and hormones, vesicular traffic, exo- and endocytosis. Since its discovery in Drosophila melanogaster (Pongs et al., 1993), neuronal calcium sensor-1 (NCS-1), which is highly conserved from yeast to man (Stockebrand and Pongs, 2006; Burgoyne, 2007; Dason et al., 2012), has been implicated in some of these processes. Importantly, NCS-1 has been associated with psychiatric conditions including autism (Piton et al., 2008), bipolar disorder, schizophrenia (Koh et al., 2003; Bai et al., 2004), and X-linked mental retardation (Tessier and Broadie, 2011). The role, however, which NCS-1 plays in these disorders, remains unresolved in part because NCS-1 seems to have not one conserved cellular target, but multiple ones. Thus, many different target proteins have been reported as potential interaction partners of NCS-1, e.g., phosphatidylinositol 4-kinase (PI4K; Mora et al., 2002; Strahl et al., 2003), Ric8a (Mansilla et al., 2017), inositol 1,4,5-triphosphate receptor (IP3R; Boehmerle et al., 2006), A-type potassium channels (Nakamura et al., 2001; Guo et al., 2002), protein interacting with C kinase-1 (PICK1; Jo et al., 2008), P/Q-type Ca2<sup>+</sup> channel Cav2.1 (Tsujimoto et al., 2002), D2 dopamine receptor (D2R; Kabbani et al., 2002), adenosine A2A receptors (Navarro et al., 2012), and G-protein-coupled receptor kinase 1 (GRK1; Pandalaneni et al., 2015). Most of these interactions were studied mainly in cellular in vitro systems overexpressing NCS-1 and the physiological significance remains unclear.

In order to investigate the potential physiological roles of NCS-1, we used an NCS-1 knock-out (KO) mouse line (NCS-1−/−; Hermainski, 2012; Ng et al., 2016). Previous studies showed that young NCS-1−/<sup>−</sup> mice are viable and generally healthy. Though NCS-1−/<sup>−</sup> mice have a mild cardiac problem at a neonatal stage, i.e., a diminished systolic function, this contraction problem disappears in adulthood (Nakamura et al., 2011). NCS-1−/<sup>−</sup> mice show little change in their physical activities, as determined via treadmillanalysis (Nakamura et al., 2017) and open field locomotion (Hermainski, 2012; Ng et al., 2016), but some behavioral deficits are notable. NCS-1−/<sup>−</sup> mice showed impaired spatial learning and memory function in the Morris Water Maze Test (Nakamura et al., 2017) and a decreased willingness to work for food (Ng et al., 2016). These behavioral phenotypes were associated with a reduced release of dopamine and brain-derived neurotrophic factor (BDNF) in CA1 presynaptic neurons (Nakamura et al., 2017) and decreased presynaptic dopamine release in striatal neurons, respectively (Ng et al., 2016). Interestingly, NCS-1-deficiency also resulted in anxietyand depressive-like behaviors as demonstrated by elevated plus maze, large open field, forced swim and tail suspension tasks (De Rezende et al., 2014).

Though behavioral phenotypes of NCS-1−/<sup>−</sup> mice have been investigated extensively, their obesity, which represents the most apparent adult NCS-1−/<sup>−</sup> phenotype, remained uncharacterized so far. Here we show that adult NCS-1−/<sup>−</sup> mice, especially when fed a high-fat diet (HFD), are hyperglycemic and hyperinsulinemic, typical symptoms associated with obesity (Modan et al., 1985; Mehran et al., 2012). HFD-fed NCS-1−/<sup>−</sup> mice display a significant increase in fat body mass. NCS-1−/<sup>−</sup> adipocytes are dysfunctional giving rise to lower serum concentrations of resistin and adipokinine than adipocytes of wild-type (WT) littermates. Importantly, insulin receptor (IR) concentration in NCS-1−/<sup>−</sup> adipocyte membrane is severely reduced. Apparently, NCS-1 is required directly to up-regulate IR density in adipocyte membrane ensuring an adequate insulin response to changes in diet. Our data indicate that NCS-1 plays an important role in adipocyte function and that NCS-1 deficiency yields obesity and diabetes type 2 in adult NCS-1−/<sup>−</sup> mice. Combining these data with the previously reported anxiety- and depressive-like behaviors of NCS-1−/<sup>−</sup> mice (De Rezende et al., 2014; Ng et al., 2016) implicates NCS-1 in a relationship of diabetes type 2 and depression, frequently observed with human patients (Nouwen et al., 2010; Renn et al., 2011; Haljas et al., 2018).

# MATERIALS AND METHODS

# NCS-1 Knock-In and Knock-Out Mouse Lines

Generation of NCS-1-EGFP knockin mice and NCS-1−/<sup>−</sup> mice (C57BL6/J background) has been described earlier (Hermainski, 2012; Ng et al., 2016). Briefly, NCS-1-EGFP knockin mice were generated that had the last four translated exons of Ncs-1 flanked by lox-P sites as well as an Egfp modification in exon seven to generate mice expressing NCS-1-EGFP fusion protein. Knockin mice were then crossed with CMV-Cre transgenic mice (Schwenk et al., 1995), which had been backcrossed to C57BL/6 over 10 generations, producing mice with no detectable NCS-1. NCS-1−/<sup>−</sup> mice were backcrossed to C57BL/6J over 10 generations and maintained on the C57BL/6J background. Genotyping was done by PCR as described (Hermainski, 2012; Ng et al., 2016). Animal protocols were in accordance with guidelines for humane treatment of animals and were reviewed and approved by the Animals Ethics Committee of the Saarland, Germany. Mice were kept at a regular 12 h day/night cycle. They had ad libitum access to a normal chow diet (ssniffTM R/M-H Pellets, ssniff-Spezialdiäten GmbH) and water. HFD (ssniffTM EF R/M D12492, ssniff-Spezialdiäten GmbH) was fed from 6 weeks of age until the termination of the experiment as indicated in figures. Throughout we used male mice for experiments. Mice were killed by cervical dislocation at the end of the study.

# Measurement of Food and Water Consumption

Food and water consumption of mice was monitored using a nutrition monitoring system (Infra-e-Motion, Hamburg). The system also records every second cage activity of the mouse. Normal chow was provided as a powder (ssniffTM EF R/M-H, ssniff-Spezialdiäten GmbH). In case of HFD, the food (ssniffTM EF R/M D12492, ssniff-Spezialdiäten GmbH) was pulverized with a grinder and mixed with 25% ssniffTM EF R/M-H powder to make a pourable chow. After a 5-day trial, phase food and water supply were replenished, and data collection was started for a period of 4 days. Data were transmitted online to a central computer outside the animal facility.

# Blood and Serum Analysis

Blood glucose levels were determined from whole venous tail blood using a glucose monitoring system (Roche Accu Chek). Serum was obtained by collecting blood from the submandibular vein. Blood samples were then incubated at room temperature for 60 min and centrifuged at 3,000 g for 10 min. Resulting serum was stored at −80◦C for further use.

# Glucose and Insulin Measurements in Glucose Tolerance Tests

Glucose tolerance tests were carried out in 16-week-old animals following a 16 h fast. After measuring fasted blood glucose levels, mice were injected intraperitoneally with 2 mg glucose/g body weight (20% glucose solution, Sigma-Aldrich, St. Louis, MO, USA). Blood glucose levels were determined at 30, 60, 90, and 120 min postglucose injection. Insulin concentrations were measured after a 16 h fast and 30 min post glucose injection in blood samples taken from the vena facialis.

# Elisa Tests for Insulin, Adipokinines and Cytokinines

Serum concentrations of hormones were determined using the following commercially available ELISA kits: Ultra Sensitive Mouse Insulin ELISA kit (Crystal Chem), Mouse Leptin ELISA Kit (Crystal Chem), Mouse Adiponectin/Acrp30 Immunoassay (R&D Systems, Minneapolis, MN, USA), Quantikine Mouse Resistin Immunoassay (R&D Systems, Minneapolis, MN, USA), Legendplex Mouse Th2 Panel for tumor necrosis factor-alpha (TNF-α) and interleukine-6 (IL-6; BioLegend<sup>R</sup> , San Diego, CA, USA).

# Preparation of Fat Body Lysate

Fat body (1 g per 1 mL lysis buffer) was transferred to a Potter homogenizer and homogenized on ice in 20 mM HEPES (pH 7.4), 125 mM KCl. 0.05% Tween 80, 100 nM CaCl<sup>2</sup> containing protease inhibitor cocktail according to directions of the supplier (Roche). Then 1 M sucrose was added to a final concentration of 300 mM. Lysis was continued for 45 min at 4◦C on a rotarod. It was centrifuged twice for 5 min at 2,000 g at 4◦C. The supernatant was centrifuged at 50,000 g for 1 h at 4◦C. The supernatant was saved as cytosol fraction and kept in aliquots at −80◦C until further use. The membranous pellet was resuspended in 300 mM sucrose, 10 mM HEPES (pH 7.4), 10 mM Tris-HCl (pH 8.0), 0.1 mM MgCl2, 100 nM CaCl<sup>2</sup> buffer containing 1 tablet/10 mL protease inhibitor cocktail (Roche). Final protein concentration was ≥2 mg/mL.

#### Immunoprecipitation

Protein G dynabeads were washed with 50 mM Tris-HCl (pH 7.4), 150 mM KCl, 1% (v/v) Triton X-100, 100 nM CaCl<sup>2</sup> buffer containing 1 tablet/10 mL protease inhibitor cocktail (Roche). Fifty microliter Dynabead suspension was added to 500 µL lysate and incubated for 3 h at 4◦C for preclearing. The beads were discarded. Primary antibodies [rabbit anti-IR antibody (5 µg/mg lysate) Abcam ab137747; rabbit anti-NCS-1 antibody (1:100) Cell Signaling D12D2] were added and for control mouse immunoglobulin (5 µg/mg lysate—Invitrogen 02-6502). Then dynabeads were collected and washed repeatedly in 50 mM Tris-HCl (pH 7.4), 150 mM KCl, 0.5% (v/v) Triton X-100, 5 mM MgCl2, 100 nM CaCl<sup>2</sup> buffer containing 1 tablet/10 mL protease inhibitor cocktail (Roche). Finally, Dynabeads were incubated with 40 µL NuPage sample buffer and heated 10 min at 70◦C before loading for PAGE.

# Page Analysis of Fat Body Lysate

PAGE was carried out using the NuPAGE 4%–12% Bis-Tris Mini Gel System according to the manufacturer's specifications (ThermoFisher, Waltham, MA, USA). Proteins were transferred onto a PVDF membrane. Immunodetection was performed using the following antibodies: rabbit anti-IR (1:1,000 Abcam ab 137747), rabbit anti-NCS-1 (1:3,000 Cell Signaling D12D2), mouse anti—β-actin (1:5,000 Sigma-Aldrich, St. Louis, MO, USA 32430), rabbit anti-IRS-1 (1:1,000 Novus biologicals, Centennial, CO, USA NB100-82001), rabbit anti-IRS-2 (1:1,000 Cell Signaling 4502S), goat anti-rabbit IgG (HRP modified; 1:5,000 Millipore AQ132P).

# Densitometry Analysis

Scanned films of Western blots were analyzed and quantified using ImageJ software available in the public domain.

#### Preparation of Paraffin-Embedded Adipose Tissue Sections

Mice were perfused with Zinc-Formal-FixxTM (ThermoFisher Scientific, Waltham, MA, USA). Gonadal fat tissue was embedded in paraffin and sectioned at 8 µm. Deparaffinized and rehydrated tissue sections were stained with anti-NCS-1 or anti-GFP antibodies followed by incubation with biotinylated secondary anti-rabbit- and, respectively, anti-chicken IgG antibodies and staining with ABC/DAB solutions (Vector Laboratories).

#### Immunostaining of Pancreatic Islets

Sixteen micrometer cryosections of pancreatic islets of adult NCS-1−/−, NCS-1-EGFP, and WT littermates were immunostained with rabbit anti-GFP (1:2,000, Invitrogen, Carlsbad, CA, USA A11122), rabbit anti-NCS-1 (1:3,000, Cell Signaling D12D2) or rabbit anti-insulin antibody (1:1,000, Abcam 63820) followed by incubation with secondary goat anti-rabbit IgG labeled with Alexa FluorTM 488 (1:1,000, Invitrogen, Carlsbad, CA, USA A-11034) or Alexa FluorTM 546 dye. Confocal images were generated with a Leica TCS SP2 confocal microscope.

#### Statistical Analysis

All numerical data are given as mean ± SEM. Repeated measurements MANOVA with Bonferroni post hoc test or ANOVA and Bonferroni post hoc test were used for statistical analysis with the help of the programme Prism5 (GraphPad). Statistical significance is indicated with <sup>∗</sup> for P < 0.05, ∗∗ for P < 0.01, ∗∗∗ for P < 0.001. Replicates and number of animals are indicated in Legends.

# RESULTS

# NCS-1−/<sup>−</sup> Mice Are Obese

In agreement with previous reports (Nakamura et al., 2011; Hermainski, 2012; Ng et al., 2016), we observed that NCS-1−/<sup>−</sup> mice gain with onset of adulthood considerably more weight than their WT littermates (**Supplementary Figure S1A**). Body weights of 16-week-old NCS-1−/<sup>−</sup> mice were on average ∼10% higher than those of WT littermates [NCS-1−/−: 33.6 ± 0.7 g (n = 26); WT: 30.6 ± 0.6 g (n = 26); P < 0.01]. When mice were fed HFD, differences in body weight between NCS-1−/<sup>−</sup> mice and WT littermates readily enlarged (**Figure 1A**). For example, NCS-1−/<sup>−</sup> mice fed for 6 weeks HFD, had gained more weight than WT-littermates such that the NCS-1−/<sup>−</sup> mice had become obese displaying a body mass index (BMI) of 36.1 ± 0.5 g/dm<sup>2</sup> (n = 6), whereas WT-littermates were overweight, displaying a BMI of 28.5 ± 0.6 g/dm<sup>2</sup> (n = 6; P = 0.002). Sixteen week-old NCS-1−/<sup>−</sup> mice, having been fed HFD for 12 weeks, weighed on

(C) Food consumption of NCS-1−/<sup>−</sup> mice and WT littermates fed HFD. Food consumption was averaged over 24 h for a period of 4 days. Measurements were started at an age of 26 weeks. (D) Water consumption of NCS-1−/<sup>−</sup> mice and WT littermates fed HFD. Water consumption was averaged over 24 h for a period of 4 days. Measurements were started at an age of 26 weeks. Bars represent mean values. n.s.—not significant; <sup>∗</sup>P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001 (MANOVA and Bonferroni post hoc test).

average ∼16% more than HFD fed WT littermates [NCS-1−/−: 48.3 ± 0.8 g (n = 26); WT: 41.7 ± 0.8 g (n = 26); P < 0.001]. BMIs were 46.3 ± 0.8 g/dm<sup>2</sup> (n = 6) and 36.4 ± 0.6 g/dm<sup>2</sup> (n = 6) respectively. Obesity is characterized by an excess of adipose tissue (World Health Organization, 2015). Also, weight differences of HFD-fed NCS-1−/<sup>−</sup> and WT mice were mainly due to a significant increase in NCS-1−/<sup>−</sup> fat body mass [NCS-1−/−: 6.62 ± 2.2 g (n = 6); WT: 2.2 ± 0.47 g (n = 6); P < 0.001] (**Figure 1B**). Note cardiac tissue mass shows no significant difference between WT and NCS-1−/<sup>−</sup> mice (Nakamura et al., 2011). Taken together, the data indicates that NCS-1−/<sup>−</sup> mice, especially HFD-fed ones, are obese.

Excessive food intake and lack of physical activity combined with genetic susceptibility are the most common causes of obesity (Haslam and James, 2005). We wanted to know whether this also applies to obesity observed in NCS-1−/<sup>−</sup> mice. Therefore, we investigated their feeding behavior and their general physical activity (cage activity). The respective nutrition monitoring results showed no significant difference in eating and drinking behavior between chow-fed NCS-1−/<sup>−</sup> and WT mice (**Supplementary Figures S1B,C**). Twenty to twenty-fiveweek-old NCS-1−/<sup>−</sup> mice daily consumed on average (n = 12) 3.63 ± 0.13 g chow and drank 3.69 ± 0.15 mL water. WT littermates ate and drank similar daily quantities [3.83 ± 0.16 g chow (n = 13); 3.37 ± 0.16 mL water (n = 13)]. Also, daily food consumption of HFD-fed NCS-1−/<sup>−</sup> and WT mice showed no

significant difference [WT: 4.05 ± 0.10 g (n = 9); NCS-1−/−: 4.29 ± 0.12 g (n = 9); P = 0.074] (**Figure 1C**). On the other hand, HFD-fed NCS-1−/<sup>−</sup> mice apparently drank on average more than their WT littermates [NCS-1−/−: 3.93 ± 0.19 mL/24 h (n = 9); WT: 3.09 ± 0.19 (n = 9); P = 0.003] (**Figure 1D**). As a measure for their physical activity, we registered over a period of 96 h how many seconds per minute the mice moved in their cage. Twenty-four hours data averaged over four consecutive days (n = 12) showed that there was no difference in cage activity of NCS-1−/<sup>−</sup> and WT mice (**Supplementary Figure S2A**). It corroborates our previous data that NCS-1−/<sup>−</sup> mice and WT littermates show similar levels of open field locomotion (Ng et al., 2016). When fed HFD, both types of mice displayed reduced cage activity (**Supplementary Figure S2B**). Again, there was no significant difference and NCS-1−/<sup>−</sup> and WT cage activities were alike. The results suggested that NCS-1−/<sup>−</sup> obesity most likely had other reasons than excessive feeding behavior and lack of physical activity.

mean ± SEM. <sup>∗</sup>P < 0.05; ∗∗∗P < 0.001 (MANOVA and Bonferroni post hoc

# NCS-1−/<sup>−</sup> Mice Are Hyperglycemic and Hyperinsulinemic

Obesity, altered glucose homeostasis and hyperinsulinemia are closely linked (Mehran et al., 2012). Therefore, we hypothesized that obesity observed in NCS-1−/<sup>−</sup> mice also was associated with hyperglycemia and hyperinsulinemia. Indeed, blood glucose levels were significantly higher in NCS-1−/<sup>−</sup> mice than in WT littermates [NCS-1−/−: 131 ± 7 mg/dL (n = 8); WT:

test).

115 ± 3 mg/dL (n = 9); P < 0.05]. Upon fasting, blood glucose levels in NCS-1−/<sup>−</sup> mice and WT littermates decreased with comparable rates, reaching similar end points [WT: 80 ± 2 mg glucose/dL (n = 8); NCS-1−/−: 78 ± 2 mg glucose/dL (n = 9); P > 0.05] (**Figure 2A**). Evidently, NCS-1−/<sup>−</sup> mice needed considerably more time for reaching the same endpoint in glucose concentration than WT littermates (NCS-1−/−: ∼13 h; WT: 8 h), because at the onset of fasting they had a higher blood glucose concentration (**Figure 2A**). Furthermore, HFD produced in NCS-1−/<sup>−</sup> mice a significantly higher increase in blood glucose concentration than in WT littermates resulting in concentrations of 152 ± 5 mg glucose/dL blood (n = 8) in NCS-1−/<sup>−</sup> mice vs. 128 ± 4 mg glucose/dL blood (n = 10) in WT littermates (P < 0.001). Upon fasting, blood glucose concentrations also decreased at similar rates in HFD-fed NCS-1−/<sup>−</sup> and WT mice (**Figure 2B**). In comparison to normal-chow fed mice, rates in HFD-fed mice were, however, nearly 10-fold slower (NCS-1−/−: 0.56 ± 0.02 mg glucose/dL per hour; WT: 0.50 ± 0.02 mg glucose/dL per hour). Subsequently, we measured glucose clearance after intraperitoneal injection of 2 mg glucose/g body mass into normal chow- and HFD-fed mice, which had been fasted for 16 h. Then blood glucose concentrations were analyzed at 30 min time intervals (**Figures 2C,D**). The results showed no significant difference in glucose clearance rates of NCS-1−/<sup>−</sup> and WT mice [NCS-1−/−: 60.7 ± 6 mg glucose/dL per hour (n = 7); WT littermates: 58.6 ± 7.3 mg glucose/dL per hour (n = 10); P > 0.05] as well as in those measured between 60 and 150 min for HFD-fed NCS-1−/<sup>−</sup> and WT [NCS-1−/−: 51 ± 15 mg glucose/dL; WT: 45 ± 12 mg glucose/dL per hour; P > 0.05] (**Figure 2D**). In conclusion, the data indicate that NCS-1 deficiency has no influence on glucose clearance rates. However, it gives rise to a significant increase in the steady-state level of blood glucose rendering NCS-1−/<sup>−</sup> mice hyperglycemic.

Next, we tested NCS-1−/<sup>−</sup> mice for hyperinsulinemia. We determined plasma insulin concentrations in NCS-1−/<sup>−</sup> mice and their WT littermates, both after fasting and in response to intraperitoneal glucose injection. A 16-h fast gave rise in normal chow-fed NCS-1−/<sup>−</sup> mice to a six-fold higher plasma insulin concentration than in normal chow-fed WT littermates [NCS-1−/−: 1.18 ± 0.19 ng insulin/mL (n = 8); WT: 0.20 ± 0.03 ng insulin/mL (n = 9); P < 0.001] (**Figure 3A**). Fasted HFD-fed mice exhibited a dramatic increase in plasma insulin concentration (**Figure 3B**). Again, plasma insulin concentrations were significantly higher in NCS-1−/<sup>−</sup> mice [4.43 ± 0.41 ng insulin/mL (n = 8), P < 0.001] than in controls [3.23 ± 0.13 ng insulin/mL (n = 10), P < 0.01] (**Figure 3B**). Intraperitoneal injection of glucose (2 mg/g body mass) into fasted normal chow-fed mice significantly increased plasma insulin concentration in WT littermates from 0.20 ± 0.03 ng insulin/mL to 1.5 ± 0.16 ng insulin/mL (n = 15); P < 0.001. In contrast, glucose injection into normal chow-fed NCS-1−/<sup>−</sup> mice produced an insignificant increase of the already elevated plasma insulin concentration, which was 1.18 ± 0.19 ng insulin/mL (n = 8) before and 1.30 ± 0.16 ng insulin/mL (n = 13) after intraperitoneal glucose injection (P = 0.31; **Figures 3A,C**). On the other hand, plasma insulin

concentration in fasted HFD-fed WT littermates was raised to 3.81 ± 0.62 ng insulin/mL (n = 9; **Figure 3D**), whereas the one in fasted HFD-fed NCS-1−/<sup>−</sup> mice had increased to an even much higher level (6.17 ± 0.61 ng insulin/mL, n = 8, P < 0.01; **Figures 3B,D**). Elevated levels of plasma insulin concentration in NCS-1−/<sup>−</sup> mice indicates that NCS-1−/<sup>−</sup> mice are hyperinsulinemic. Taken together we observed in the data that HFD-fed NCS-1−/<sup>−</sup> mice are obese, hyperglycemic and hyperinsulinemic indicating a default of NCS-1−/<sup>−</sup> tissue in their response to insulin.

#### NCS-1 Deficiency Affects Adipokine Serum Concentration

Prime targets of insulin are muscle, liver and fat cells. NCS-1 expression is undetectable in liver and muscle (Nef et al., 1995), but prominant in adipose tissue. Western blots of fat body lysate and immunostaining of paraffin-embedded slices of gonadal adipose tissue demonstrated that both small and large adipocytes express NCS-1 (**Supplementary Figures S3A,B**). Since obesity-linked insulin resistance may be associated with dysfunctional adipocytes (Friedman and Haalas, 1998; Steppan et al., 2001; Galic et al., 2010; Ouchi et al., 2011), we concentrated on adipocytes for further analysis of the NCS-1−/<sup>−</sup> phenotype. Adipocytes are an endocrine organ and source of adipokines and cytokines, for example, of leptin, TNF-α, IL- 6, adiponectin, and resistin, which have been implicated in modulating insulin sensitivity and energy balance (Friedman and Haalas, 1998; Berg et al., 2001; Steppan et al., 2001; Mojiminiyi et al., 2007; Galic et al., 2010; Ouchi et al., 2011). Normally, plasma concentration of leptin, an important hormone involved in the regulation of energy expenditure, positively correlates with body mass (Friedman and Haalas, 1998). Also, our data revealed, a positive correlation between body mass and leptin for normal chow- and HFD-fed NCS-1−/<sup>−</sup> mice and WT littermates. Furthermore, TNF-α and IL-6 plasma concentrations were just as positively correlated with NCS-1−/<sup>−</sup> and WT body weight (**Supplementary Figures S4A–C**). It follows that plasma concentrations of leptin were positively correlated with TNF-α and IL-6 (**Supplementary Figure S4D**). This is in good agreement with previous data that leptin increases the production of TNF-α and IL-6 (Galic et al., 2010; Ouchi et al., 2011). Thus, deficiency of NCS-1 has no effect on the secretion of leptin, TNF-α and IL-6, respectively.

In rodents, adipocytes are the primary source of resistin. It may also influence the ability of the body to respond to insulin and metabolize glucose, and it may decrease secretion of adiponectin, an insulin-sensitizing adipocytokine that seems to play an opposite role to the one of resistin (Berg et al., 2001; Steppan et al., 2001; Fasshauer et al., 2002; Möhlig et al., 2002; Mojiminiyi et al., 2007; Ouchi et al., 2011; Chen et al., 2014). Resistin levels in serum of normal chow-fed NCS-1−/<sup>−</sup> and WT littermates showed no significant difference [NCS-1−/−: 20.2 ± 1.0 ng resistin/mL (n = 10); WT: 21.6 ± 0.9 ng resistin/mL (n = 10); P > 0.05] In contrast, HFD-fed NCS-1−/<sup>−</sup> and WT mice displayed significantly different resistin concentrations (NCS-1−/−: 19.4 ± 0.7 ng resistin/mL, n = 8; WT: 23.9 ± 0.8 ng resistin/mL, n = 10, P = 0.008; **Figure 4A**). Whereas resistin serum concentrations were positively correlated with WT body mass, those in NCS-1−/<sup>−</sup> mice were uncorrelated or even negatively correlated with body mass (**Figure 4A**). Likewise, normal chow-fed NCS-1−/<sup>−</sup> mice and WT littermates showed no significant difference in adiponectin serum concentration

FIGURE 4 | Reduced concentrations of resistin and adiponectin in NCS-1−/<sup>−</sup> serum. (A) Relation between resistin serum concentration and body weight for 25-week-old NCS-1−/<sup>−</sup> mice (n = 8–10) and WT littermates (n = 8–10). (B) Relation between adiponectin concentration and body weight for 25-week-old NCS-1−/<sup>−</sup> mice (n = 8) or WT littermates (n = 10). Mice were kept either on normal chow (n = 10) or fed HFD after week 6 (n = 8). Bars represent mean values ± SEM. n.s.—not significant; <sup>∗</sup>P < 0.05; ∗∗P < 0.01 (MANOVA and Bonferroni post hoc test).

(NCS-1−/−: 8.5 ± 0.4 µg adiponectin/mL, n = 10; WT: 8.2 ± 0.7 µg/mL, n = 10; P > 0.05; **Figure 4B**). In agreement with data in the literature (Berg et al., 2001; Steppan et al., 2001; Fasshauer et al., 2002; Möhlig et al., 2002; Mojiminiyi et al., 2007; Chen et al., 2014), adiponectin concentrations were negatively correlated with body mass in both types of mice (**Figure 4B**). But the negative correlation between body mass and adiponectin serum concentration was in HFD-fed NCS-1−/<sup>−</sup> mice (n = 10) more pronounced than in HFD-fed WT littermates (n = 8). The data indicate that NCS-1 deficiency affects resistin and adiponectin plasma levels. The fact that both levels are influenced by insulin (Berg et al., 2001; Fasshauer et al., 2002; Möhlig et al., 2002; Mojiminiyi et al., 2007; Ouchi et al., 2011; Chen et al., 2014), suggests that NCS-1 deficiency affects insulin sensitivity of NCS-1−/<sup>−</sup> adipocytes.

#### NCS-1 Deficiency Affects Insulin-Receptor Concentration in Adipocyte Membranes

Next, we asked if NCS-1 deficiency affects the insulin response of NCS-1−/<sup>−</sup> adipocytes. Unexpectedly, we found that the adipocyte membrane of NCS-1−/<sup>−</sup> mice contained significantly less IR than WT littermates. Quantitative evaluation of Western blots of membrane fractions from fat body lysate of normal chow-fed NCS-1−/<sup>−</sup> and WT mice showed that IR signal intensities normalized to those of actin were about twice as high in WT than in NCS-1−/<sup>−</sup> mice (IR/β-actin—WT: 0.61 ± 0.04, n = 4; NCS-1−/−: 0.30 ± 0.06, n = 4; P = 0.006; **Figures 5A,B**). Analysis of IR concentration in adipocyte membrane of HFD-fed WT mice revealed a significant rise that nearly tripled IR concentration (IR/β-actin: 1.60 ± 0.14, n = 5, P = 0.007). By contrast, there was no significant increase in IR concentration in adipocyte membrane of HFD-fed NCS-1−/<sup>−</sup> mice (IR/β-actin: 0.43 ± 0.05, n = 6, P = 0.12; **Figure 5B**). The data indicates that IR concentration in the membrane of WT adipocytes is exquisitely sensitive to diet. In contrast, IR expression in NCS-1−/<sup>−</sup> adipocytes is insensitive to HFD. In conclusion, NCS-1 deficiency leads to a default in diet-sensitive regulation of IR concentration in adipocyte membrane.

#### Increase of IR and NCS-1 Concentrations in Adipocyte Membrane Are Correlated

Obviously, NCS-1 is involved in HFD-associated IR upregulation in adipocyte membrane. When we analyzed Western blots of fat body lysate of normal chow- and HFD-fed mice, we observed that NCS-1 expression was also HFD sensitive (**Figures 5C,D**). The data showed that in normal chow-fed mice two-thirds of NCS-1 (63.3 ± 6.5%, n = 4) and in HFD-fed mice almost all NCS-1 protein (91.2 ± 4.5%, n = 6) resided in the membrane fraction of fat body lysate. Importantly, fat-body membrane isolated from HFD-fed mice contained more than two-fold higher amounts of NCS-1 than those of mice fed with normal chow (HFD: NCS-1/β-actin—1.12 ± 0.07; normal chow: NCS-1/β-actin—0.48 ± 0.09; P < 0.001; **Figures 5C,D**). Thus, both the concentration of NCS-1 and IR are sensitive to HFD. Next, we constructed a plot of NCS-1 vs. IR—concentrations that were observed in fat body lysate of normal chow and resepectively,

FIGURE 5 | NCS-1 and insulin receptor (IR) concentration in adipocyte membrane are correlated. (A) Western blot analysis of cytosol (C) and membrane (M) fraction of fat body lysate prepared from NCS-1−/<sup>−</sup> knockout (KO) and WT mice fed either normal chow or HFD. Blot was immunostained with anti-IR and with anti—β-actin antibodies as indicated at left. (B) Normalized IR signal intensities (IR/β-actin) obtained from Western blots of WT (n = 2–3) and NCS-1−/<sup>−</sup> (KO; n = 2–3) fat body lysate as exemplified in (A). IR concentrations (IR/β-actin) were determined in duplicate. (C) Western blot analysis of cytosol (C) and membrane (M) fraction of fat body lysate prepared from NCS-1−/<sup>−</sup> (KO) and WT mice fed either normal chow or HFD. Blot was immunostained with anti-NCS-1 and with anti—β-actin antibodies as indicated at left. (D) Normalized NCS-1 signal intensities (NCS-1/β-actin) obtained from Western blots of WT fat body lysate (n = 2–3) as exemplified in (C). NCS-1 concentrations (NCS-1/β-actin) were determined in duplicate. (E) Plot of normalized IR concentration vs. NCS-1 concentration. Values (± SEM) were calculated from the data shown in (B,D). Red box represents IR concentration measured in lysate of NCS-1−/<sup>−</sup> fat body as shown in (B). (F) Western blot analysis of immunoprecipitates obtained after incubation of fat body lysate with anti-NCS-1 antibodies (NCS-1 IP), anti-IR antibodies (IR IP) or with non-specific immunoglubulin (IgG IP). "SN" corresponds to supernatant of the immunoprecipitation reaction and "input" to starting material. Western blots were stained either with anti-IR or with anti-NCS-1 antibody as indicated at left. Molecular weights are given in kDa at left. Bars represent mean values. ∗∗P < 0.01; ∗∗∗P < 0.001 (unpaired two-tailed Student's t-test).

HFD-fed mice and normalized to β-actin. A linear fit to the data well described the correlation between NCS-1 and IR concentrations (**Figure 5E**). The fit provided two important informations. First, if the NCS-1 concentration equals zero, the fit indicated a remaining normalized IR-concentration of 0.28. A very similar IR concentration was detected in the lysate of NCS-1−/<sup>−</sup> adipocytes. These data indicate that there is a basal concentration of IR in the adipocyte membrane, which is independent of NCS-1 expression. Second, there is a diet-sensitive component of IR concentration, which is correlated with HFD-sensitive up-regulation of NCS-1 concentration. It suggests that NCS-1 is an important player in the regulation of IR membrane concentration in adipocytes in response to diet changes.

It has been reported that NCS-1 associates with D2 and adenosine A2A receptors (Kabbani et al., 2002; Navarro et al., 2012). We investigated the possibility that NCS-1 likewise associates with IR like. NCS-1 in membrane fractions of HFD-fed mice was precipitated with anti-NCS-1 antibodies. NCS-1 immunoprecipitates were then analyzed by immunostaining respective Western-blots with anti-IR and anti-NCS-1 antibodies. For control, we used non-specific IgG. The results showed that anti-NCS-1 antibodies immunoprecipitated NCS-1 completely and co-immunoprecipitated a significant amount of IR with it (**Figure 5F**). In the reverse experiment we immunoprecipitated IR and co-immunoprecipitated NCS-1 with it. In this case, anti-NCS-1 antibodies stained a band, which consistently migrated slightly more slowly in PAGE (n = 3) than NCS-1 of NCS-1 immuno-precipitates (**Figure 5F**). The altered mobility of NCS-1, which may correspond to an additional band seen in overexpression experiments (e.g., Mora et al., 2002), is difficult to understand. Our anti-NCS-1 antibodies do not stain any other protein than NCS-1 (**Figure 5C**, **Supplementary Figure S5**). Also a spillover of stained light-chain IgG is very unlikely. It may reflect a NCS-1 conformation altered by binding other ions than Ca2<sup>+</sup> (Choudhary et al., 2018; Tsvetkov et al., 2018) or altered by an as yet unknown post-translational modification. In summary, the co-immunoprecipitation experiments suggest that IR and NCS-1 form a complex. Whether this reflects a direct or indirect interaction between IR and NCS-1, however, requires further experiments.

#### DISCUSSION

We have made two major observations in our characterization of the obese phenotype of NCS-1−/<sup>−</sup> mice, which has a late onset and only emerges in adulthood. First, the NCS-1−/<sup>−</sup> mice appear to have diabetes type 2. Second, NCS-1 deficiency impairs the up-regulation of IR in adipocyte membrane in response to HFD.

The obese phenotype of HFD-fed NCS-1−/<sup>−</sup> mice, which exhibit abnormally increased fat body mass, is associated with hyperglycemia, hyperinsulinemia and dysfunctional adipocytes, in good agreement with a previous report that hyperinsulinemia and obesity are associated with dysfunctional adipocytes (Ouchi et al., 2011). We showed that NCS-1 deficiency specifically affects adipokine secretion of adipocytes. HFD-fed NCS-1−/<sup>−</sup> mice have significantly lower serum levels of adiponectin and resistin than HFD-fed WT littermates, whereas serum levels of leptin, IL-6, and TNF-α were unaffected. Previous data suggest that NCS-1 has a regulatory role in dopamine and hormone secretion. Thus, NCS-1−/<sup>−</sup> mice exhibit decreased presynaptic dopamine release in striatal neurons (Ng et al., 2016) and reduced release of dopamine and BDNF in CA1 presynaptic neurons, respectively (Nakamura et al., 2017). Conversely, overexpression of NCS-1 in PC12 cells increased evoked growth hormone release (McFerran et al., 1998). In this context, it is of note that NCS-1 interacts with IP3R in cardiac tissue and that NCS-1 overexpression enhances IP3R-stimulated phosphorylation of CaMKII-δ (Nakamura et al., 2011; Nakamura and Wakabayashi, 2012). Thus, NCS-1 may affect global Ca2<sup>+</sup> signals. It is likely that this effect is pertinent to the proposal that NCS-1 exerts regulatory functions in exocytosis of dense core vesicles (McFerran et al., 1998). A regulatory effect of NCS-1 on regulatory Ca2+-signaling may provide a common molecular basis for NCS-1 effects on hormone secretion in neurons as well as in adipocytes. Note that this does not concern insulin secretion of beta cells, which do not express NCS-1 (**Supplementary Figure S5**).

Adipocyte membrane of HFD-fed WT mice contained significantly more NCS-1 than WT mice fed with normal chow. In the light of reports that NCS-1 is up-regulated in the prefrontal cortex of schizophrenic and bipolar patients (Koh et al., 2003; Bai et al., 2004), it is interesting to see that a metabolic insult like HFD up-regulates NCS-1. The HFD-induced up-regulation of NCS-1 in adipocyte membrane was directly correlated with that of IR. By contrast, HFD-induced up-regulation of IR is impaired in NCS-1−/<sup>−</sup> adipocytes. We conclude that NCS-1 deficiency impedes HFD-sensitive up-regulation of IR density in adipocyte membrane. Our data also indicate that adipocytes require NCS-1 for HFD-sensitive IR up-regulation. The relatively low IR concentration in NCS-1 deficient adipocyte membrane may be insufficient for a normal adipocyte response to insulin and, thus, lead to the obese phenotype in NCS-1−/<sup>−</sup> mice (Soli et al., 1975). The consequences of insulin signaling in adipocytes will be discussed below.

IR activation by insulin induces cytoplasmic binding of IR substrates IRS-1 and IRS-2. They play key roles in transmitting signals from the IR to intracellular pathways. Phosphorylation on multiple sites upon activation of IR regulates IRS-1 and IRS-2 distribution between membrane and cytosol and their interaction with downstream partners. Down-regulation of IRS-1 and IRS-2 is associated with an obese phenotype (Kaburagi et al., 1997; Boucher et al., 2016; Kubota et al., 2017). Our Western-blot data indicated that IRS-1 concentrations in adipocytes of HFD-fed NCS-1−/<sup>−</sup> mice are significantly reduced in comparison to WT littermates, both in cytosol and membrane fractions of fat body lysate (**Supplementary Figures S6A,B**). This result is in good agreement with the observation that IRS-1 is in adipocytes a major target in insulin signaling. IRS-2 concentration, on the other hand, was similar in cytosol fractions of NCS-1−/<sup>−</sup> and WT fat body lysate. Only membrane fractions of NCS-1−/<sup>−</sup> contained a significantly reduced IRS-2 concentration in comparison to WT (**Supplementary Figures S6C,D**). The data shows that NCS-1 deficiency affects down-stream of IR IRS activity and signaling, most likely as a consequence of reduced IR expression in NCS-1−/<sup>−</sup> adipocyte membrane. In this context, it will be important to see whether the NCS-1 effect on IR concentration and IR signaling is specific for IR or whether NCS-1 affects also other receptors, for example, insulin growth factor-1. The activation of insulin signaling induces translocation of the glucose transporter GLUT4 to the plasma membrane. We observed that GLUT4 is expressed in NCS-1−/<sup>−</sup> adipocytes at a lower level in comparison to WT (**Supplementary Figures S6E,F**). The data suggest that adipocytes of HFD-fed NCS-1−/<sup>−</sup> mice have a lower rate of glucose-uptake than adipocytes of HFD-fed WT littermates. Previously it has been reported that overexpression of NCS-1 in 3T3L1 tissue culture cells inhibits the translocation of GLUT4 to the plasma membrane involving a phosphatidyl-4 kinase-dependent pathway (Mora et al., 2002). How NCS-1 in combination with phosphatidyl-4 kinase influences the sorting and/or translocation process of GLUT4 remained however unclear. At this stage, it is difficult to compare data obtained by overexpression of NCS-1 in tissue culture cells with ours, which were obtained with NCS-1−/<sup>−</sup> adipocytes in HFD-fed adult animals.

Our data indicate that there are at least two pathways for controlling IR concentration in the adipocyte membrane; one is NCS-1 independent, the other is NCS-1 dependent. The former provides a basic IR level observed in WT and NCS-1−/<sup>−</sup> mice alike. This situation provides a likely explanation for phenotypic differences observed with NCS-1−/<sup>−</sup> and FIRKO mice, where IR expression in adipose tissue is completely absent (Blüher et al., 2002; Entingh-Pearsall and Kahn, 2004; Boucher et al., 2014, 2016; Kubota et al., 2017). The latter involves a direct correlation between NCS-1 and IR concentrations in adipocytes. Whether this reflects a direct association between NCS-1 and IR, as suggested by our co-immunoprecipitation results, requires further study along the lines previously shown for D2- and adenosine A2A receptors (Kabbani et al., 2002; Woll et al., 2011; Navarro et al., 2012). This also includes the possibility that NCS-1 attenuates IR internalization, as was shown for D2 receptor internalization when NCS-1 is overexpressed in HEK 293 cells (Kabbani et al., 2002).

Insulin resistance, decreased plasma levels of adiponectin (hypoadiponectinemia) and decreased plasma levels of resistin are reportedly related to obesity and type 2 diabetes mellitus in patients (Way et al., 2001; Fasshauer et al., 2002; Möhlig et al., 2002; Heilbronn et al., 2004; Mojiminiyi et al., 2007; Chen et al., 2014). The observed effects of NCS-1 deficiency on IR concentration and on adipokinine secretion in combination with the observed hyperinsulinemia, hyperglycemia and obesity strongly suggest that the phenotype of NCS-1−/<sup>−</sup> mice resembles type 2 diabetes. Therefore, we propose NCS-1 as a novel player in the development of type 2 diabetes. In light of the potential clinical implications, it will be important to further explore insulin responses in NCS-1−/<sup>−</sup> mice.

Our study extends the phenotypic characterization of NCS-1−/<sup>−</sup> mice in an important aspect. It demonstrates that NCS-1 deficiency leads to diabetes type 2 and to behavioral phenotypes reminiscent of psychiatric disorders. Thus, NCS-1−/<sup>−</sup> mice display depressive-like behavior in forced swim and tail suspension tests (Friedman and Haalas, 1998), and as we previously showed, NCS-1−/<sup>−</sup> mice display decreased motivation to work for food and choose the less effortful option for obtaining food (Ng et al., 2016). NCS-1−/<sup>−</sup> mice, therefore, exhibit a genetic link for diabetes type 2 and mood-related behaviors. In human patients, depressive symptoms and diabetes type 2 are frequently associated (Nouwen et al., 2010; Renn et al., 2011; Haljas et al., 2018). In future studies, the generation of mouse lines with specific tissue-specific NCS-1-deficiencies may provide further insights into a potential genetic link between some psychiatric disorders and the risk of being obese.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

# AUTHOR CONTRIBUTIONS

OP conceived the project and wrote together with OR the manuscript. OP, OR, JH and KR designed the experiments, analyzed and interpreted the data. OR carried out the Western blot analyses and the immunoprecipitations. JH constructed the knock-out mouse line and measured glucose tolerance and insulin resistance. KR measured hormone levels in mouse serum.

# FUNDING

This work was supported by a grant (DFG—Po137/45-1) of the German Research Council (Deutsche Forschungsgemeinschaft).

#### ACKNOWLEDGMENTS

OP is grateful to Jens Rettig for hosting him in his CIPPM institute and his generous support and interest in this work. We especially thank Anja Ludes for her help in breeding mice.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2019.00078/full#supplementary-material

FIGURE S1 | Normal food and water consumption in overweight NCS-1−/<sup>−</sup> mice. (A) Body weight of male NCS-1−/<sup>−</sup> mice (n = 26) and wild-type (WT) littermates (WT; n = 26) fed normal chow. (B) Food consumption of NCS-1−/<sup>−</sup> mice and WT littermates fed normal chow. Food consumption was averaged over 24 h for a period of 4 days. Measurements were started at an age of 26–31 weeks. (C) Water consumption of NCS-1−/<sup>−</sup> mice and WT littermates fed normal chow. Water consumption was averaged over 24 h for a period of 4 days. Measurements were started at an age of 26–31 weeks. (A) Data points represent mean values ± SEM. <sup>∗</sup>P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001 (MANOVA and Bonferroni post hoc test). (B,C) Bars represent mean values. n.s.—not significant.

FIGURE S2 | NCS-1−/<sup>−</sup> and WT mice have similar cage activity. (A) Four days average of 24 h profile of cage activity of NCS-1−/<sup>−</sup> mice (red trace; n = 10)

and WT littermates (black trace; n = 10) fed normal chow. Measurements were started at an age of 26–31 weeks. Activity was measured every second and is plotted as number of events per minute. (B) Four days average of 24 h profile of cage activity of NCS-1−/<sup>−</sup> mice (red trace; n = 9) and WT littermates (black trace; n = 9). Measurements were started at an age of 26–31 weeks. Mice were fed high-fat diet from an age of week 6 onwards. Activity was measured every second and is plotted as the number of events per minute. Black bar indicates light/dark cycle.

FIGURE S3 | Expression of NCS-1 in adipose tissue. (A) Western blot analysis of fat body lysate of WT and NCS-1−/<sup>−</sup> mice. Blot was immunostained with anti-NCS-1 antibodies and with anti—β-actin antibodies for input control. (B) Eight micrometer paraffin embedded cryosections of gonadal fat tissue of NCS-1-EGFP (KI), NCS-1−/<sup>−</sup> (KO) and WT mice were immunostained either with anti—GFP antibodies (EGFP – panels in upper row) or with anti-NCS-1 antibodies (NCS-1—panels in lower row). Antigen-antibody complexes were visualized with secondary biotinylated antibodies followed by staining with 3,3<sup>0</sup> -diaminobenzidine (DAB). Black scale bar—100 µm.

FIGURE S4 | Relation between leptin, TNF-α, IL-6, and body weight of WT and NCS-1−/<sup>−</sup> mice. (A) Relation between leptin concentration and body weight for 25-week-old NCS-1−/<sup>−</sup> mice or WT littermates. Mice were kept either on normal chow (n = 7) or fed high-fat diet after week 6 (n = 13). Measurements were done in duplicate. (B) Relation between tumor necrosis factor-alpha (TNF-α) concentration and body weight for 25-week-old NCS-1−/<sup>−</sup> mice or WT littermates (WT). Mice were kept either on normal chow (n = 5) or fed high-fat diet after week 6 (n = 6). Measurements were done in duplicate. (C) Relation between interleukin 6 (IL-6) concentration and body weight for 26-week-old NCS-1−/<sup>−</sup> mice fed normal chow (n = 6) and WT littermates fed normal chow (n = 6) or high-fat diet (n = 6). Measurements were done in duplicate. (D) Correlation of leptin concentration with TNF-α and, respectively, IL-6 plasma concentrations was based on the linear relations shown in (A,B; Leptin vs. TNF-α) and in (A,C; Leptin vs. IL-6), which were used to read out respective plasma concentrations at a given body weight. (A–C) Data points represent mean values ± SEM.

FIGURE S5 | NCS-1 is not expressed in β-cells of pancreatic islets. Sixteen micrometer cryosections of pancreatic islets of NCS-1-EGFP (KI), NCS-1−/<sup>−</sup> (KO) and WT mice. Panels in upper row: KI cryosection immunostained with anti—GFP rabbit antibodies (EGFP), KO cryosection with anti-NCS-1 rabbit antibodies (NCS-1), WT cryosection with anti-NCS-1 rabbit antibodies. KO and WT cryosections were co-stained with 4<sup>0</sup> ,6-diamidin-2-phenylindol (DAPI). Panels in lower row: WT cryosection immunostained with anti-NCS-1 rabbit antibodies and, respectively, with anti—insulin mouse antibodies; ovl—overlay. Secondary antibodies were Alexa FluorTM 488—coupled goat anti-rabbit IgG and, respectively, Alexa FluorTM 546—coupled goat anti-mouse IgG. White scale bar—50 µm.

FIGURE S6 | Reduced insulin receptor substrate 1 and 2 concentration in NCS-1−/<sup>−</sup> membrane. (A) Western blot analysis of cytosol (C) and membrane (M) fraction of fat body lysate prepared from NCS-1−/<sup>−</sup> (KO) and WT mice fed high-fat diet (HFD). Blot was immunostained with anti—insulin receptor substrate-1 (IRS-1) and with anti—β-actin antibodies as indicated. (B) Normalized insulin receptor substrate-1 signal intensities (IRS-1/β-actin) obtained from Western blots of WT (n = 5) and NCS-1−/<sup>−</sup> (KO; n = 5) fat body lysate as exemplified in (A). Insulin receptor concentrations in lysate were determined in duplicate. (C) Western blot analysis of cytosol (C) and membrane (M) fraction of fat body lysate prepared from NCS-1−/<sup>−</sup> (KO) and WT mice fed high-fat diet (HFD). Blot was immunostained with anti—insulin receptor substrate-2 (IRS-2) and with anti—β-actin antibodies. (D) Normalized insulin receptor substrate-2 signal intensities (IRS-2/β-actin) obtained from Western blots of wild type (WT; n = 4) and NCS-1−/<sup>−</sup> (KO; n = 4) fat body lysate as exemplified in (C). IRS-2 lysate concentrations were determined in duplicate. (E) Western blot analysis of cytosol (C) and membrane (M) fraction of fat body lysate prepared from NCS-1−/<sup>−</sup> (KO) and WT mice fed high-fat diet (HFD). Blot was immunostained with anti – glucose transporter (GLUT4) antibodies and with anti—β-actin antibodies as indicated. (F) Normalized GLUT4 signal intensities (GLUT4/β-actin) obtained from Western blots of wild type (WT; n = 4) and NCS-1−/<sup>−</sup> (KO; n = 4) fat body lysate as exemplified in (E). Bars represent mean values. <sup>∗</sup>P < 0.05; ∗∗P < 0.01 (unpaired two-tailed Student's t-test).

#### REFERENCES


Linnaeus 1758). [Dissertation thesis]. Hamburg, Germany: University of Hamburg.


K. W. Koch and P. P Phillippov (New York, NY: Nova Science Publishers), 3–17.


World Health Organization. (2015). Obesity and overweight. Fact sheet N◦ 311.

Woll, M. P., De Cotiis, D. A., Bewley, M. C., Tacelovsky, D. M., Levenson, R., and Flanagan, J. M. (2011). Interaction between the D2 Dopamine Receptor and Neuronal Calcium Sensor-1 Analyzed by Fluorescence Anisotropy. Biochemistry 50, 8780–8791. doi: 10.1021/bi200637e

**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 © 2019 Ratai, Hermainski, Ravichandran and Pongs. 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) and the copyright owner(s) 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.

# NCS-1 Deficiency Affects mRNA Levels of Genes Involved in Regulation of ATP Synthesis and Mitochondrial Stress in Highly Vulnerable Substantia nigra Dopaminergic Neurons

Carsten Simons<sup>1</sup>† , Julia Benkert<sup>1</sup>† , Nora Deuter<sup>1</sup> , Christina Poetschke<sup>1</sup> , Olaf Pongs<sup>2</sup> , Toni Schneider<sup>3</sup> , Johanna Duda<sup>1</sup> and Birgit Liss1,4 \*

1 Institute of Applied Physiology, University of Ulm, Ulm, Germany, <sup>2</sup> Institute of Physiology, Center for Integrative Physiology and Molecular Medicine, University of the Saarland, Homburg, Germany, <sup>3</sup> Institute of Neurophysiology, University of Cologne, Cologne, Germany, <sup>4</sup> New College, University of Oxford, Oxford, United Kingdom

Edited by:

Daniele Dell'Orco, University of Verona, Italy

#### Reviewed by:

Tomoe Y. Nakamura, National Cerebral and Cardiovascular Center, Japan Marisa Brini, University of Padova, Italy

> \*Correspondence: Birgit Liss birgit.liss@uni-ulm.de †These authors have contributed equally to this work

Received: 20 December 2018 Accepted: 27 September 2019 Published: 27 November 2019

#### Citation:

Simons C, Benkert J, Deuter N, Poetschke C, Pongs O, Schneider T, Duda J and Liss B (2019) NCS-1 Deficiency Affects mRNA Levels of Genes Involved in Regulation of ATP Synthesis and Mitochondrial Stress in Highly Vulnerable Substantia nigra Dopaminergic Neurons. Front. Mol. Neurosci. 12:252. doi: 10.3389/fnmol.2019.00252 Neuronal Ca2<sup>+</sup> sensor proteins (NCS) transduce changes in Ca2<sup>+</sup> homeostasis into altered signaling and neuronal function. NCS-1 activity has emerged as important for neuronal viability and pathophysiology. The progressive degeneration of dopaminergic (DA) neurons, particularly within the Substantia nigra (SN), is the hallmark of Parkinson's disease (PD), causing its motor symptoms. The activity-related Ca2<sup>+</sup> homeostasis of SN DA neurons, mitochondrial dysfunction, and metabolic stress promote neurodegeneration and PD. In contrast, NCS-1 in general has neuroprotective effects. The underlying mechanisms are unclear. We analyzed transcriptional changes in SN DA neurons upon NCS-1 loss by combining UV-laser microdissection and RT-qPCRapproaches to compare expression levels of a panel of PD and/or Ca2+-stress related genes from wildtype and NCS-1 KO mice. In NCS-1 KO, we detected significantly lower mRNA levels of mitochondrially coded ND1, a subunit of the respiratory chain, and of the neuron-specific enolase ENO2, a glycolytic enzyme. We also detected lower levels of the mitochondrial uncoupling proteins UCP4 and UCP5, the PARK7 gene product DJ-1, and the voltage-gated Ca2<sup>+</sup> channel Cav2.3 in SN DA neurons from NCS-1 KO. Transcripts of other analyzed uncoupling proteins (UCPs), mitochondrial Ca2<sup>+</sup> transporters, PARK genes, and ion channels were not altered. As Cav channels are linked to regulation of gene expression, metabolic stress and degeneration of SN DA neurons in PD, we analyzed Cav2.3 KO mice, to address if the transcriptional changes in NCS-1 KO were also present in Cav.2.3 KO, and thus probably correlated with lower Cav2.3 transcripts. However, in SN DA neurons from Cav2.3 KO mice, ND1 mRNA as well as genomic DNA levels were elevated, while ENO2, UCP4, UCP5, and DJ-1 transcript levels were not altered. In conclusion, our data indicate a possible novel function of NCS-1 in regulating gene transcription or stabilization of mRNAs in SN DA neurons. Although we do not

provide functional data, our findings at the transcript level could point to impaired ATP production (lower ND1 and ENO2) and elevated metabolic stress (lower UCP4, UCP5, and DJ-1 levels) in SN DA neurons from NCS-1 KO mice. We speculate that NCS-1 is involved in stimulating ATP synthesis, while at the same time controlling mitochondrial metabolic stress, and in this way could protect SN DA neurons from degeneration.

Keywords: ND1, ENO2, Parkinson's disease, mitochondrial uncoupling proteins, voltage-gated calcium channels, Cav2.3, KChip3/DREAM/Calsenilin, DJ-1/PARK7

#### INTRODUCTION

Ca2<sup>+</sup> signaling is important for a variety of neuronal functions, like membrane excitability, neurotransmitter release, gene transcription, and many other processes crucial for neuronal functions and viability (Berridge, 1998; Brini et al., 2014). The nature, magnitude, and location of the Ca2<sup>+</sup> signal is crucial for its distinct effects (La Rovere et al., 2016). Hence, intracellular Ca2<sup>+</sup> levels are tightly controlled (Gleichmann and Mattson, 2011; Heine et al., 2019). Neuronal Ca2<sup>+</sup> sensor proteins (NCS) respond to changes in intracellular Ca2<sup>+</sup> concentrations with conformational changes that allow them to bind diverse interaction partners, and to activate a variety of different signaling pathways (Burgoyne and Haynes, 2012; Choudhary et al., 2018; Burgoyne et al., 2019). The widely expressed neuronal Ca2<sup>+</sup> sensor NCS-1 (Pongs et al., 1993) modulates e.g., voltage-gated Ca2<sup>+</sup> channels (Cav) (Tsujimoto et al., 2002; Weiss et al., 2010), A-type K<sup>+</sup> channels (Nakamura et al., 2001; Guo et al., 2002), that are composed of Kv4 α- and KChip3 β-subunits in SN DA neurons (Liss et al., 2001), G-protein coupled receptor kinases (GRK), and G-protein coupled dopamine D2-receptors (Kabbani et al., 2002; Pandalaneni et al., 2015), to just name a few. Changes in NCS-1 expression will alter the relation with its target proteins and were described in a variety of diseases, including schizophrenia and Parkinson's disease, both characterized by dysfunctional dopaminergic signaling (Koh et al., 2003; Kabbani et al., 2012; Dragicevic et al., 2014; Boeckel and Ehrlich, 2018; Bandura and Feng, 2019; Catoni et al., 2019).

Parkinson's disease (PD) is the second most common neurodegenerative disease (Schulz et al., 2016; Obeso et al., 2017). Its motor-related symptoms are caused by a progressive loss of dopaminergic (DA) neurons, particularly within the Substantia nigra (SN) (Damier et al., 1999; Surmeier et al., 2017b; Giguere et al., 2018). While the cause for most PD cases is still unclear, ion channel activity, activity-related Ca2<sup>+</sup> homeostasis, mitochondrial dysfunction, and elevated metabolic stress constitute key interacting events in PD pathology (Duda et al., 2016; Michel et al., 2016; Surmeier et al., 2017b; Cherubini and Wade-Martins, 2018). In some familial inherited PD cases, disease-causing mutations in so-called PARK genes have been identified, most of them linked to elevated metabolic stress (van der Brug et al., 2015; Deng et al., 2018; Blauwendraat et al., 2019). SN DA neurons are particularly vulnerable to metabolic stress and other PD-stressors, due to their energetically demanding large axonal arborizations, as well as a stressful Ca2<sup>+</sup> entry during action potential firing, mediated by Cav channels. This distinguishes them from the more resistant DA neurons in the ventral tegmental area (VTA) that are rarely affected in PD (Pissadaki and Bolam, 2013; Pacelli et al., 2015; Duda et al., 2016; Surmeier et al., 2017a). These activity-related, voltage-gated Ca2<sup>+</sup> dynamics sustain electrical activity, ATP synthesis, and dopamine release of SN DA neurons and thus movement, but they constitute an intrinsic metabolic burden (Guzman et al., 2010, 2018; Duda et al., 2016; Zaichick et al., 2017). Moreover, the Ca2<sup>+</sup> buffer capacity of calbindin negative SN DA neurons is low compared to other neurons, e.g., the resistant calbindin positive VTA DA neurons (Foehring et al., 2009; Dopeso-Reyes et al., 2014; Blesa and Vila, 2019). They rely mainly on mitochondrial-ER Ca2<sup>+</sup> buffering (Cali et al., 2013; Lee et al., 2018) and on inhibitory regulatory feedback mechanisms that reduce activity related Ca2<sup>+</sup> entry and associated neurodegenerative events, like Ca2<sup>+</sup> dependent activation of K<sup>+</sup> channels (Wolfart et al., 2001; Anderson et al., 2013; Dragicevic et al., 2014; Subramaniam et al., 2014; Iyer et al., 2017). One of these Ca2<sup>+</sup> and K<sup>+</sup> channel dependent feedback mechanisms operant in SN DA neurons involves NCS-1 function (Dragicevic et al., 2014; Poetschke et al., 2015; Catoni et al., 2019).

NCS-1 has emerged as particularly important in this context of activity-related Ca2<sup>+</sup> stress and vulnerability of SN DA neurons in PD (Bandura and Feng, 2019; Catoni et al., 2019). In general, NCS-1 has been shown to stimulate mitochondrial function and neuronal survival promotion (Nakamura et al., 2006, 2019; Angebault et al., 2018; Boeckel and Ehrlich, 2018). Furthermore, especially in SN DA neurons, NCS-1 inhibits their stressful activity in a Ca2<sup>+</sup> and Cav dependent fashion, by stimulation of inhibitory, K<sup>+</sup> channel coupled dopamine D2 autoreceptors (Kabbani et al., 2002; Dragicevic et al., 2014; Robinson et al., 2017), and NCS-1 mRNA levels are elevated in remaining human SN DA neurons from post-mortem PD brains (Dragicevic et al., 2014).

Here, we aimed to gain insights into effects of general NCS-1 loss on gene expression in SN DA neurons. By combining UV-lasermicrodissection (UV-LMD) with retrograde tracing and quantitative PCR approaches, we examined a panel of candidate genes in SN DA neurons from NCS-1 KO mice and wildtype controls. We quantified mRNA levels of the voltagegated Ca2<sup>+</sup> channel α-subunits Cav1.3, Cav2.3, Cav3.1, the voltage- and Ca2+-gated A-type K<sup>+</sup> channel α- and β-subunits Kv4.3 and KChip3, the mitochondrial Ca2<sup>+</sup> transporters MCU, LETM1, mNCX, the mitochondrial uncoupling proteins UCP2 – UCP5, and of PARK genes that affect mitochondrial function and/or Ca2<sup>+</sup> homeostasis (SNCA, DJ-1, PGC-1α, and GBA1). Furthermore, we analyzed the expression of the glycolytic enzyme neuron-specific enolase 2 (ENO2) and of the NADH-ubiquinone oxidoreductase chain 1 (ND1), a subunit of the complex I of the mitochondrial respiratory chain. We gained evidence for selective transcriptional downregulation of proteins involved in ATP synthesis (ND1, ENO2) and metabolic stress defense (UCP4, UCP5, DJ-1) in SN DA neurons from NCS-1 KO mice, that were associated with but likely not causal to lower Cav2.3 mRNA levels.

We conclude that NCS-1 (directly or indirectly) is involved in stimulating the transcription or the mRNA stability of these genes in SN DA neurons.

# MATERIALS AND METHODS

fnmol-12-00252 November 25, 2019 Time: 15:43 # 3

#### Ethical Approval

All animal procedures were approved by the German Regierungspräsidium Tübingen (AZ 35/9185.81-3TV No. 1291, Reg. No. o.147) and conducted to the guidelines of the German Tierschutzgesetz.

#### Mice

All mice were bred in in-house breeding facilities at Ulm University. The NCS-1 KO is back-crossed at least 10 times into C57BL/6J, thus reaching a 99% analogy with C57BL/6J and losing the 129/SvJ original background (Ng et al., 2016). Cav2.3 KO mice were only back-crossed 4 times, leading to a 75% C57BL/6J and 25% 129/SvJ mixed background, as they do not breed well, likely due to the physiological function of Cav2.3 in mouse sperms (Wennemuth et al., 2000). SN DA neurons from NCS-1 and Cav2.3 KO mice were compared to their respective distinct +/+ wildtype background littermates. For NCS-1 KO and their WT, we analyzed juvenile mice (PN13), for Cav2.3 KO and their WT, we show results for adults (∼ PN90), as in contrast to NCS-1, Cav2.3 expression increases with post-natal maturation (Benkert et al., 2019). Data were derived from 18 NCS-1 KO and 18 NCS-1 WT mice, and from 5 Cav2.3 KO and 5 Cav2.3 WT mice. Only male mice were analyzed.

#### In vivo Retrograde Tracing

SN DA neurons from adult mice were retrogradely labeled as described (Liss et al., 2005; Lammel et al., 2008; Krabbe et al., 2015). Red fluorescent latex retrobeads (Lumafluor) were injected unilaterally into the dorsal striatum (bregma: 0.98 mm, lateral 1.9 and 2.7 mm, ventral: -3.2 mm, 2 × 60 nl) under stereotactic control (Kopf Instruments) with a NanoFil syringe attached to a micropump (UMP3 with SYS-Micro4 Controller; World Precision Instruments) at a rate of 50 nl/min and general isoflurane anesthesia. After 7 days, mice were sacrificed and striatal injection sites were verified using TH-immunofluorescence stain (primary antibody: rabbit anti-TH, 1:1,000, catalog no. 657012, Merck group; secondary antibody: Alexa Fluor 488 goat anti-rabbit IgG, 1:1,000, catalog no. A-11034, Thermo Fisher Scientific) and Nissl stain (Neuro Trace 640/660 deep-red fluorescent Nissl stain solution, 1:1,000 in 1x PBS, catalog no. N21483, Thermo Fisher Scientific), according to the mouse brain atlas (Paxinos and Franklin, 2013) as described (Liss et al., 2005; Krabbe et al., 2015).

### Tissue Preparation, UV-Laser Microdissection (UV-LMD), and Reverse Transcription

Carried out similarly as previously described in detail (Grundemann et al., 2011; Dragicevic et al., 2014; Duda et al., 2018). Briefly, 12 µm coronal mouse brain sections were cut with a cryomicrotome CM3050 S (Leica), mounted on PEN-membrane slides (Microdissect), and fixed with an ascending ethanol series. Sections were stored in sterile Falcon tubes with silica gel at -80◦C until used for UV-LMD. UV-LMD was carried out with a LMD7000 system (Leica Microsystems); laser-microdissected cells (pools of 10 SN DA neurons, each) were collected into the lid of a certified RNase free reaction tube (PCR Tubes Thinwall Clear 0.5 ml, Axygen) containing 4.7 µl lysis-buffer, and reverse transcription was performed without an RNA-isolation step by using random hexamer primers and superscript II RT enzyme (Thermo Fisher Scientific). For juvenile mice, mounted sections were stained (1 min) with a cresylviolet-ethanol solution, and SN DA neurons were identified by anatomical location, morphology and size in brightfield mode. In adult mice, fluorescence-traced SN DA neurons were identified under Y3 filter (565–610 nm, exposure time: 250 ms). The size/area of lasered cells was determined automatically after calibration by the LMD7000 software (Version 8.2.0.6739, Leica Microsystems). We ensured that lasered areas were similar for all analyzed animals/compared groups (**Supplementary Figure S1** and **Supplementary Table S2**).

### Qualitative Multiplex Nested PCR and Quantitative Realtime PCR

All cDNA reactions were precipitated as described (Liss, 2002) and resolved in 17 µl RNase free water (5Prime, Molecular biology grade, certified RNase free). Qualitative and quantitative PCRs were carried out, essentially as described (Liss et al., 2001; Grundemann et al., 2011; Poetschke et al., 2015; Duda et al., 2018). All primers except for ND1 were spanning at least one intron to quantify only cDNA-derived signals. We performed a qualitative multiplex PCR with the GeneAmp PCR System 9700, Thermo Fisher Scientific, with an aliquot of 5 µl (30%) of each individual cDNA-pool for respective marker genes: Tyrosine hydroxylase (TH) as a marker for dopaminergic midbrain neurons, the glutamic acid decarboxylase isoforms GAD<sup>65</sup> and GAD<sup>67</sup> as markers for GABAergic neurons, glial fibrillary acidic protein (GFAP) as a marker for astroglia cells and calbindind28<sup>k</sup> (CBd28k), that is strongly expressed only in less vulnerable DA midbrain neurons. Qualitative PCR products were analyzed in a QIAxcel Advanced System (Quiagen). Only cDNA pools expressing the correct marker gene profile (i.e., TH positive, GAD, GFAP, CBd28k negative) were further analyzed via qPCR. All primer pairs and amplicon information are given in **Supplementary Table S1**.

#### TABLE 1 | TaqMan <sup>R</sup> assay and standard curve information.

fnmol-12-00252 November 25, 2019 Time: 15:43 # 4


Threshold, Y-Intercept (± SD), slope (± SD), R<sup>2</sup> (± SD) were determined using serial dilutions over four magnitudes of cDNA as templates, derived from mouse midbrain tissue of juvenile NCS-1 WT mice (∗Cav2.3 WT-derived).

Quantitative realtime PCR was performed as described (Grundemann et al., 2011; Duda et al., 2018), by using the 7900 HT Fast realtime PCR System and QuantStudio 3 System, both Thermo Fisher Scientific Thermocyclers. TaqManTM primer/probe assays were marked with a 3<sup>0</sup> BHQ (black hole quencher) and 5<sup>0</sup> FAM (Carboxyfluorescein). TaqMan assays were carefully established and performance was evaluated by generating standard curves, using defined amounts of cDNA (derived from midbrain tissue mRNA), over four magnitudes of 10-fold dilutions as templates, in at least three independent experiments, as described (Liss, 2002; Duda et al., 2018). All TaqMan assays and standard curve details are summarized in **Table 1**.

#### Isolation of Genomic DNA From Laser-Microdissected Neurons and qPCR Quantification of Genomic ND1

Laser-microdissected pools of 10 SN DA neurons as well as single neurons were harvested into the lid of a reaction tube (PCR Tubes Thinwall Clear 0.5 ml, Axygen), containing 15 µl ATL-buffer (tissue lysis buffer, Qiagen). The QiaAmp DNA Micro-Kit (Qiagen) was used and the manufacturer protocol was adapted as described for genomic DNA isolation (Muhling et al., 2014). Genomic DNA was eluted in 30 µl water and 5 µl were used for ND1 qPCR. Mitochondrial genome copy number was estimated by quantifying the mitochondrially coded NADHubiquinone oxidoreductase chain I (ND1) from genomic DNA via qPCR, as described (Bender et al., 2006; Muhling et al., 2014). We quantified genomic ND1 levels in single SN DA neurons as well as in pools of 10 neurons (**Supplementary Table S2** and **Supplementary Figure S2**). As we detected no significant differences in mean copy-number per individual SN DA neuron between both approaches, respective data sets were pooled.

#### qPCR Data Analysis and Statistics

For qPCR data analysis as well as for graphical representation, the SDS 2.4 software (Thermo Fisher Scientific), the QuantStudioTM Design and Analysis Software (Thermo Fisher Scientific) and GraphPad Prism 6 (GraphPad Software Inc.) were used. The cDNA-amount per cell in relation to the utilized standard was calculated as described (Grundemann et al., 2011; Schlaudraff et al., 2014; Duda et al., 2018), according to the following formula:

$$\text{cDNA amount per cell} = \frac{S^{\{(Ct - Y\_{Intercept})/slops\}}}{No\_{cells} \times cDNA fraction}$$

S stands for the serial dilution factor of the standard curve (in our case 10 for serial dilution steps of 10), Nocells corresponds to the number of harvested neurons per cDNA sample (here 10), and

FIGURE 1 | General workflow illustrating procedure for combined UV-LMD and RT-qPCR based mRNA and genomic DNA analysis for individual mouse SN DA neurons. (A) Coronal cryo-sections from juvenile (PN13) mice were stained with cresylviolet (CV) and ethanol-fixed. SN of adult mice (PN90) was in vivo retrogradely traced and coronal cryo-sections were not CV-stained but only ethanol-fixed. SN DA neurons were isolated via UV-LMD into a sterile reaction tube. Subsequently, either genomic DNA was isolated from each sample for qPCR-based quantification of genomic ND1, or a combined lysis and cDNA synthesis protocol was applied, followed by cDNA precipitation for qPCR-based mRNA quantification. A fraction of each cDNA-pool was used for qualitative multiplex-nested marker gene PCR. Note that only samples expressing the correct marker gene profile (TH positive, GAD, CB, GFAP negative) were further analyzed via TaqManTM qPCR, as indicated. GAD, L-glutamate decarboxylase; CB, calbindind28<sup>k</sup> ; GFAP, glial fibrillary acidic protein; TH, tyrosine hydroxylase. Left photographs: Upper: Overview of a juvenile CV-stained wildtype mouse coronal section after UV-LMD of 10 SN DA neurons. The Substantia nigra (SN) is highlighted. Scale bar: 500 µm. Lower: An exemplary juvenile SN DA neuron before (left) and after (right) UV-LMD. Scale bars: 30 µm. Right photographs: Upper: View of an adult in vivo traced injection site (i.e., dorsal striatum), next to an illustration of the respective brain section according to the mouse brain atlas (Paxinos and Franklin, 2013). MC, motor cortex; SSC, somatosensory cortex; CPu, Caudate putamen; NAc, Nucleus Accumbens. Lower: Traced SN DA neuron in fluorescence (left) and brightfield mode (right). Scale bars: 10 µm. (B) Upper: Gel image after capillary electrophoresis of RT-PCR products indicates that all presumed mRNAs are expressed in standard cDNA, derived from PN13 C57BL/6J mouse midbrain tissue in a 1:10<sup>4</sup> dilution (positive control). Lower: Gel image after capillary electrophoresis of RT-PCR products of an individual SN DA neuron from a WT mouse indicates that all genes analyzed in this study, except for UCP3 and mNCX, are regularly expressed at the mRNA level in individual TH positive SN neurons from WT mice. Note, that we detected positive signals for mNCX only in ∼22% (2 of 9) and signals for UCP2 only in ∼50% (8 of 15) of analyzed WT SN DA neurons.

cDNA fraction refers to the fraction of the cDNA (or respectively genomic ND1 DNA) reaction product, used as a template in qPCR reactions (cDNA: 1.5/17 for ND1 and ENO2; 5/17 for all other analyzed genes. Genomic DNA: 5/30 for ND1). The YIntercept unit magnitude corresponds to the respective standard utilized (e.g., pg equivalents of standard cDNA, derived from midbrain tissue mRNA). To facilitate comparison, cDNA amounts were calculated with a YIntercept of 45.00 for all genes. Expression data are given as mean ± SD with and without normalization to the neuronal size per microdissected area (by dividing respective expression values to the corresponding area of the individual microdissected neurons). Note that the mitochondrially encoded ND1 gene contains no intron, thus the RT-qPCR results reflect number of cDNA and genomic ND1 molecules per sample.

Outlier tests were performed in GraphPad Prism 6 and outliers were removed according to ROUT-outlier test. For statistical analysis, the non-parametric Mann-Whitney U (MWU) test was used. Significant differences are marked by asterisks (∗p ≤ 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001).

#### RESULTS

To study possible transcriptional changes in vulnerable SN DA neurons upon NCS-1 loss, we quantified mRNA levels of PD and/or Ca2<sup>+</sup> stress-related genes. To analyze relative mRNA expression levels with best possible resolution, we used our well-established single cell UV-LMD RT-qPCR protocol, and optimized TaqMan primer/probe assays. **Figure 1A** summarizes the general work-flow, **Table 1** and **Supplementary Table S1** summarize details of all utilized PCR primer/probe assays. As illustrated in **Figure 1B**, all analyzed mRNAs are readily detected in wildtype mouse midbrain tissue, including NCS-1. This was also the case in individual laser-microdissected SN DA neurons from wildtype mice for most genes, with the following exceptions: the mitochondrial uncoupling protein UCP2 was detected only in about 50% (n = 8 of 15), the mitochondrial Ca2<sup>+</sup> exchanger mNCX only in about 22% (n = 2 of 9) of SN DA neurons, while UCP3 was not detected at all, in accordance with previous work (Andrews et al., 2005; Liss et al., 2005; Hoang et al., 2012). Hence, we focused on comparing mRNA levels of ND1 and ENO2 proteins, the mitochondrial Ca2<sup>+</sup> transporters MCU and LETM1, the mitochondrial uncoupling proteins UCP2, 4, and 5, the PARK genes DJ-1, SNCA, PGC-1α, and GBA1, and the ion channels Cav1.3, Cav2.3, Cav3.1, and Kv4.3/KChip3, in SN neurons that were positive for tyrosine hydroxylase, ND1 and ENO2, while negative for calbindind28<sup>k</sup> , GAD65/67, and GFAP. RT-qPCR data were normalized to cell sizes, derived from laser-microdissected areas. Mean cell sizes/lasered areas were similar for all compared groups (**Supplementary Figure S1** and **Supplementary Table S2**).

We first analyzed ND1 and ENO2 mRNA, as markers for mitochondrial and glycolytic ATP synthesis, respectively in SN DA neurons from NCS-1 KO and WT mice. Both mRNAs were readily detected in all samples, but levels for both genes were about 20% lower in NCS-1 KO (**Figure 2**, **Table 2** and **Supplementary Table S2**; ND1: n = 87 for WT, n = 93 for KO, KO/WT = 0.82, p = 0.0152; ENO2: n = 87 for WT, n = 84 for KO, KO/WT = 0.72, p = 0.0034). As the mitochondrially coded ND1 gene is intronless, the RT-qPCR results reflect the amounts of cDNA and genomic DNA. To address if the lower detected mRNA levels are rather caused by lower numbers of mitochondrial genomes than by lower number of ND1 transcripts, we compared genomic ND1 levels in SN DA neurons from WT and NCS-1 KO mice (**Figure 2**, **Table 2**, and **Supplementary Table S2**). As genomic ND1 levels were similar, this argues for lower ND1 transcription or lower ND1 mRNA stability, but similar numbers of mitochondrial genomes in SN DA neurons from NCS-1 KO mice.

Furthermore, we detected about 60% lower levels of the mitochondrial uncoupling proteins UCP4 and UCP5, while those of UCP2, MCU and LETM1 were not changed (**Figure 3A**, **Table 2**, and **Supplementary Table S3**; UCP4: n = 9 for WT, n = 10 for KO, KO/WT = 0.44, p = 0.0003; UCP5: n = 9 for WT, n = 9 for KO, KO/WT = 0.42, p = 0.0142). For mNCX, we detected signals only in 2 out of 9 WT samples and in none of the 10 NCS-1 KO samples (**Supplementary Table S3**), thus a statistical comparison is not meaningful. Similar as for UCP4 and UCP5, we detected about 55% lower mRNA levels of the PARK7 gene product, the deglycase DJ-1 (**Figure 3B**, **Table 2**,

FIGURE 2 | Lower relative cDNA levels of ND1 and ENO2 in SN DA neurons from NCS-1 KO mice compared to WT. Left: qPCR-derived relative cDNA and genomic DNA levels of mitochondrially coded NADH-ubiquinone oxidoreductase chain 1 (ND1) in individual TH positive SN DA neurons from juvenile NCS-1 KO and WT mice, relative to respective tissue cDNA derived standard curves (in pg/cell, upper), and in addition normalized to specific laser-microdissected neuron sizes (in µm<sup>2</sup> × 100, lower). Note that ND1 cDNA levels reflect cDNA and also genomic DNA-derived signals, as the mitochondrially encoded ND1 gene contains no intron, but genomic ND1 levels alone are similar between WT and KO. Right: Similar qPCR-derived relative cDNA levels for the same samples as in (A) for the neuron specific enolase (ENO2). Data are given as scatter plots with mean ± SD. Significant differences are indicated according to Mann–Whitney U-tests and marked with asterisks ( <sup>∗</sup>p ≤ 0.05; ∗∗p < 0.01). Numbers of analyzed individual SN DA neuron-derived cDNA samples (n) are given on the x-axis. All data and statistics detailed in Supplementary Figure S2 and Supplementary Table S2.

and **Supplementary Table S3**; n = 13 for WT, n = 11 for KO, KO/WT = 0.45, p = 0.0129). In contrast, mRNA levels of all other analyzed PD-related genes (SNCA, PGC-1α, and GBA1) were similar in SN DA neurons from NCS-1 KO and WT mice (**Figure 3B**, **Table 2**, and **Supplementary Table S3**). mRNA levels of Cav2.3 R-type channel α-subunits were also about 50% lower (n = 20 for WT, n = 20 for KO, KO/WT = 0.53, p < 0.0001) in SN DA neurons from NCS-1 KO mice. In contrast, those of Cav1.3 and of Cav3.1 were not changed. Likewise, transcript levels of Ca2<sup>+</sup> sensing KChip3 and of Kv4.3 A-type K<sup>+</sup> channel subunits were also similar in NCS-1 KO and WT mice (**Figure 3C**, **Table 2**, and **Supplementary Table S3**).

In summary, we detected at the transcriptional level an orchestrated downregulation, of ND1 and ENO2, both involved in ATP synthesis, of the mitochondrial uncoupling proteins UCP4 and UCP5, the PARK7 gene DJ-1, and of Cav2.3 in SN DA neurons from NCS-1 KO mice, compared to WT.

Cav channels can regulate gene expression (Gomez-Ospina et al., 2006; Barbado et al., 2009; Pinato et al., 2009), and in SN DA neurons Cav activity contributes to elevated metabolic stress and to their degeneration in PD (Surmeier et al., 2017a; Guzman et al., 2018; Tabata et al., 2018). To address if the lower mRNA levels of ND1, ENO2, UCP4, and 5, and DJ-1 in SN DA neurons from NCS-1 KO mice are possibly secondary to the lower Cav2.3 levels in NCS-1 KO, we analyzed the expression of these genes in SN DA neurons from Cav2.3 KO mice and respective wildtype controls. In SN DA neurons from Cav2.3 KO, ND1 mRNA as well as genomic DNA levels were about 15 and 25% higher compared to those of WT mice, respectively, pointing to more mitochondrial genomes

in SN DA neurons from Cav2.3 KO (**Figure 4A**, **Table 2**, and **Supplementary Table S2**; ND1 mRNA + genomic DNA: n = 42 for WT, n = 44 for KO, KO/WT = 1.14, p = 0.0062; genomic DNA: n = 15 for WT, n = 19 for WT, KO/WT = 1.24, p = 0.0153). For all other tested mRNAs, levels were similar between WT and Cav2.3 KO (ENO2: **Figure 4A**, **Table 2**, and **Supplementary Table S2**; UCP4 and5, DJ-1: **Figure 4B**, **Table 2**, and **Supplementary Table S4**).

Finally, we plotted the detected transcript levels for UCP4, UCP5, DJ-1, and Cav2.3 for NCS-1 KO and WT to that of ND1 and ENO2, determined in the same samples, to assess if the detected changes in SN DA neurons from NCS-1 KO mice were correlated with each other. Lower UCP4, UCP5 and Cav2.3 mRNA levels in NCS-1 KO were not correlated to that of ND1, as they were still significantly lower in KO when compared to WT in relation to ND1 mRNA levels, while in relation to ENO2, only UCP4 mRNA was still significantly lower in NCS-1 KO (**Figure 5**, **Table 3**, and **Supplementary Table S5**).

#### DISCUSSION

Here, we addressed transcriptional changes in SN DA neurons, caused by the loss of NCS-1. We combined UV-LMD with RT-qPCR approaches to analyze mRNA levels in individual SN DA neurons of NCS-1 KO mice, compared to their respective wildtype. **Figure 6** summarizes the main findings and conclusion of this study and illustrates complex Ca2<sup>+</sup> signaling in SN DA neurons.

#### NCS-1 – A Regulator of Gene Transcription in SN DA Neurons?

The finding that ND1, ENO2, UCP4, UCP5, DJ-1, and Cav2.3 transcripts are significantly lower in SN DA neurons from NCS-1 KO mice, indicates that NCS-1 is involved in regulating the transcription of these genes or the stability of their mRNAs in these neurons.

Ca2<sup>+</sup> dependent regulation of gene expression in general, and in particular of genes involved in Ca2<sup>+</sup> homeostasis and metabolic stress, is well described (Berridge et al., 2000; West et al., 2001; Greer and Greenberg, 2008; Naranjo and Mellstrom, 2012). For instance, a Ca2<sup>+</sup> dependent regulation by calcineurin is described for NCS-1 gene expression itself (Hamasaki-Katagiri and Ames, 2010) that is stimulated by the neurotrophic factor GDNF (Nakamura et al., 2019). How exactly NCS-1 in turn could regulate gene expression remains less clear. A role of NCS-1 for controlling activity-related nuclear Ca2<sup>+</sup> levels is reported for cardiomyocytes (Nakao et al., 2015). NCS-1 is present in the nucleus or in perinuclear regions

FIGURE 4 | Higher expression levels selectively of ND1 cDNA and genomic DNA in SN DA neurons from Cav2.3 KO mice compared to WT. (A) Left: qPCR-derived relative cDNA and genomic DNA levels of mitochondrially coded NADH-ubiquinone oxidoreductase chain 1 (ND1) in individual TH positive SN DA neurons from adult Cav2.3 WT and KO mice, relative to respective tissue cDNA derived standard curves (in pg/cell, upper), or normalized to specific laser-microdissected neuron sizes (in µm<sup>2</sup> × 100, lower). Note that ND1 cDNA levels reflect cDNA + genomic DNA-derived signals, as the ND1 gene contains no intron, and cDNA as well as genomic ND1 levels are elevated in the KO. Right: Similar qPCR-derived relative cDNA levels for the same samples as in (A, left) for the neuron-specific enolase (ENO2). (B) Relative RT-qPCR-derived data from individual TH positive SN neurons from Cav2.3 KO and WT mice for the genes as indicated (displaying lower mRNA levels in SN DA neurons from NCS-1 KO mice). All data are given as scatter plots with mean ± SD. Significant differences are indicated according to Mann–Whitney U-tests and marked with asterisks (∗p ≤ 0.05; ∗∗p < 0.01). Numbers of analyzed individual SN DA neuron-derived cDNA samples (n) are given on the x-axis. All data and statistics detailed in Supplementary Figure S2 and Supplementary Tables S2, S4.

( statistics detailed in Supplementary Table S5.

(Burgoyne, 2007; Nakao et al., 2015), its nuclear targeting is Ca2<sup>+</sup> dependent (O'Callaghan et al., 2002), and recent evidence suggests that NCS-1 might regulate gene transcription by sensing nuclear Ca2<sup>+</sup> (Naranjo and Mellstrom, 2012; Nakamura et al., 2019). NCS-1 could control gene expression by indirect mechanisms, such as stimulation of Calmodulin kinase II, PI3/AKT-signaling (Petko et al., 2009; Nakamura et al., 2011), cAMP responsive elements (Souza et al., 2011), or DJ-1 (Xu et al., 2005; Takahashi-Niki et al., 2017). DJ-1 itself can translocate into the nucleus, particularly in response to metabolic stress, and regulates expression, e.g., of UCPs (Xu et al., 2018). Hence, the detected lower mRNA levels of UCP4 and UCP5 in NCS-1 KO mice might be secondary to the transcriptional downregulation of DJ-1. Lower levels of UCP4 and UCP5 were also described in SN tissue from DJ-1 KO mice (Guzman et al., 2010). Though the complex functions of DJ-1 are still not entirely clear, a support of mitochondrial function and reduction of metabolic stress is established (Biosa et al., 2017), and loss-of-function mutations in DJ-1 (PARK7) cause familial inherited PD (Bonifati et al., 2003).

Cav channel activity has also been linked to Ca2<sup>+</sup> dependent regulation of gene expression (Gomez-Ospina et al., 2006; Barbado et al., 2009; Pinato et al., 2009). However, as mRNA levels of ENO2, UCP4, UCP5, and DJ-1 were all not altered in SN DA neurons of Cav2.3 KO mice, we conclude that their transcriptional downregulation in NCS-1 KO mice is rather not secondary to lower Cav2.3 levels. The higher ND1 mRNA and genomic levels in SN DA neurons from Cav2.3 KO might compensate for a reduced Ca2<sup>+</sup> mediated stimulation of enzymes for ATP production. However, these are only theoretical speculations.

We found no evidence for transcriptional compensation of NCS-1 loss by KChip3. KChip3 (also named Calsenilin or DREAM) has overlapping functions with NCS-1 (Naranjo and Mellstrom, 2012; Burgoyne et al., 2019), and not only constitutes a beta subunit for Kv4.3 channels (An et al., 2000), that are involved in PD pathology (Subramaniam et al., 2014; Dragicevic et al., 2015), but it can also shuttle from the plasma-membrane to the nucleus, and act as a Ca2<sup>+</sup> dependent transcription repressor by direct DNA binding (Carrion et al., 1999; Mellstrom and Naranjo, 2001; Gomez-Villafuertes et al., 2005; Mellstrom et al., 2014).

TABLE 2 | Relative mRNA levels in SN DA neurons of either NCS-1 KO or Cav2.3 KO, normalized to individual cell sizes, and to respective WT data.


Data and statistics for graphs shown in Figures 2, 3A–C, 4A,B. Data are normalized to lasered cell sizes and to WT, thus mean KO-values directly display fold-differences in relation to respective WT. (n) indicates the number of detected signals, from all tested TH positive SN neuron-derived cDNA samples (given in brackets), derived from (N) mice. (# ) data modified from Benkert et al. (2019). P-values according to Mann–Whitney U-tests, significant differences are marked with (<sup>∗</sup> ). Note that ND1 cDNA levels reflect cDNA + genomic DNA-derived signals, as the ND1 gene contains no intron.

TABLE 3 | Relative mRNA levels in SN DA neurons of NCS-1 KO and WT, normalized to cell sizes, in respect to that of ND1 or ENO2.


Data and statistics for graphs shown in Figure 5. Data are normalized to lasered cell sizes, to ND1 or ENO2 and to WT, thus mean KO-values directly display folddifferences in relation to respective WTs. (n) indicates the number of detected signals, from all tested TH positive SN DA neuron-derived cDNA-samples (given in brackets), derived from (N) mice. P-values according to Mann–Whitney U-tests, significant differences are marked with (<sup>∗</sup> ).

proliferator-activated receptor gamma coactivator 1-alpha; Pyr, pyruvate; ROS, reactive oxygen species; SNCA, alpha-synuclein gene; TFs, transcription factors;

Possible compensation is generally important to consider, particularly for global knockouts. Both analyzed KO mouse strains have been already studied intensively and displayed clear phenotypes, in line with the described functions of NCS-1 and Cav2.3. The here studied NCS-1 KO mice showed decreased motivation, associated with lower dopamine release in the nucleus accumbens (Ng et al., 2016), and they are prone to gain weight and develop type 2 diabetes (Ratai et al., 2019). A different NCS-1 KO mouse strain, lacking exon 1 and resulting in disrupted NCS-1 protein, displayed an anxiety- and depression-like phenotype, reduced novelty-induced exploratory behavior (de Rezende et al., 2014), as well as reduced stress tolerance in cardiomyocytes due to dysfunctional mitochondrial detoxification and Ca2<sup>+</sup> dependent pathways (Nakamura et al., 2016, 2019). Cav2.3-deficient mice display a mild cardiac, endocrine and neuronal phenotype (Pereverzev et al., 2005), assessed in four different KO mouse strains (Weiergraber et al., 2006). Mainly, they display slightly impaired insulin and somatostatin secretion (Jing et al., 2005; Zhang et al., 2007), mild cardiac arrhythmia (Weiergraber et al., 2005), and they are less prone to epilepsy (Weiergraber et al., 2007, 2010; Dibue-Adjei et al., 2017).

#### NCS-1 – A Regulator of ATP Synthesis and Metabolic Stress in SN DA Neurons?

Our data suggest that NCS-1 activity in SN DA neurons is correlated with the expression of genes important for glycolytic and mitochondrial ATP production (ND1, ENO2), as well as of genes that control mitochondrial function and reduce metabolic stress (UCP4, UCP5, DJ-1). This might offer an explanation for a possible but not yet demonstrated neuroprotective function of

UCP, uncoupling protein; 2-PG, 2-phosphoglycerate.

NCS-1 for SN DA neurons: stimulation of ATP synthesis while at the same time controlling metabolic stress levels.

This conclusion would be in line with a reported NCS-1 stimulation of mitochondrial function and of Ca2<sup>+</sup> dependent survival promotion in injured neurons in general (Angebault et al., 2018; Nakamura et al., 2019). However, it is important to note that we do not provide any functional data here to support this conclusion. We are currently addressing this issue by comparing respiration, ATP production capacity, and mitochondrial uncoupling in freshly-dissected vital SN slices from NCS-1 KO and wildtype mice via Seahorse XFe analysis.

Our findings are however well-complemented by a similar study in cardiomyocytes, at protein and functional levels (Nakamura et al., 2016). In cardiomyocytes of NCS-1 KO mice, the overall respiration and mitochondrial biogenesis was reduced, accompanied by a decreased functional expression of mitochondrial proteins. This phenotype could be rescued by NCS-1 overexpression, similar as described for respective mitochondrial dysfunction in fibroblasts from Wolfram Syndrome patients (Angebault et al., 2018). Furthermore, a reduced UCP-mediated proton leak in response to oxidative stress, accompanied with elevated mitochondrial oxidant stress, was described for cardiomyocytes of NCS-1 KO mice (Nakamura et al., 2016) – in line with the lower UCP4 and UCP5 mRNA levels that we report here for SN DA neurons.

Transcriptional downregulation of ND1, ENO2, and Cav2.3 might reflect a compensatory response to reduce stressful activity, related Ca2<sup>+</sup> load, and ATP synthesis in SN DA neurons in the absence of protective NCS-1. However, these theoretical considerations would need to be experimentally addressed.

On a wider note, cell-specific stimulation of NCS-1 function (Mansilla et al., 2017) might offer a novel therapeutic strategy for combating metabolic stress and neurodegeneration. However, given the ubiquitous expression of NCS-1 and its multiple, complex, and still not fully understood functions, manipulation of this intricate network should be considered with caution.

#### REFERENCES


#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the German Tierschutzgesetz, and the Regierungspräsidium Tübingen. The protocols were approved by the German Regierungspräsidium Tübingen (AZ 35/9185.81- 3TV No. 1291, Reg. Nr. 0.147).

#### AUTHOR CONTRIBUTIONS

CS, ND, and JB carried out the molecular biology experiments. JD, JB, and CP performed the in vivo retrograde tracing. JD and JB contributed the UV-LMD of adult mice. OP provided NCS-1 KO mice. TS provided Cav2.3 KO mice. BL designed the study. BL, JD, JB, and CS wrote the manuscript. All authors revised the manuscript.

#### FUNDING

This work was supported by the German DFG (Graduate Schools CEMMA and Molecular Medicine, and Li1754-1(1-3) to BL), the Austrian Science Fund (SFB F44-12 to BL), and the Alfried Krupp Foundation (to BL). CS was supported by the Medical Faculty of Ulm University, and the Excellence Initiative of the Federal and State Governments.

#### ACKNOWLEDGMENTS

We thank Desirée Spaich for training and excellent technical support.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2019.00252/full#supplementary-material




miR-133b in dopamine neurons in Parkinson's disease. Neurobiol. Aging 35, 2302–2315. doi: 10.1016/j.neurobiolaging.2014.03.016


new light on an enigmatic channel. Neurosci. Biobehav. Rev. 30, 1122–1144. doi: 10.1016/j.neubiorev.2006.07.004


**Conflict of Interest:** 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 © 2019 Simons, Benkert, Deuter, Poetschke, Pongs, Schneider, Duda and Liss. 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) and the copyright owner(s) 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.

digital media

of impactful research

article's readership