# IRON AND NEURODEGENERATION

EDITED BY : Isabella Zanella, Massimiliano Filosto and Giorgio Biasiotto PUBLISHED IN : Frontiers in Neuroscience and Frontiers in Neurology

#### 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 978-2-88963-451-4 DOI 10.3389/978-2-88963-451-4

#### 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

# IRON AND NEURODEGENERATION

Topic Editors: Isabella Zanella, University of Brescia, Italy Massimiliano Filosto, Civil Hospital of Brescia, Italy Giorgio Biasiotto, University of Brescia, Italy

Citation: Zanella, I., Filosto, M., Biasiotto, G., eds. (2020). Iron and Neurodegeneration. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-451-4

# Table of Contents


*174 Oxidative Stress Related to Iron Metabolism in Relapsing Remitting Multiple Sclerosis Patients With Low Disability*

Mariacristina Siotto, Maria Maddalena Filippi, Ilaria Simonelli, Doriana Landi, Anna Ghazaryan, Stefano Vollaro, Mariacarla Ventriglia, Patrizio Pasqualetti, Mauro Ciro Antonio Rongioletti, Rosanna Squitti and Fabrizio Vernieri


*183 Combined Visualization of Nigrosome-1 and Neuromelanin in the* 

*348 Iron, Ferritin, Hereditary Ferritinopathy, and Neurodegeneration* Barry B. Muhoberac and Ruben Vidal

# Editorial: Iron and Neurodegeneration

#### Giorgio Biasiotto1,2 \*, Massimiliano Filosto<sup>3</sup> \* and Isabella Zanella1,2 \*

*<sup>1</sup> Department of Molecular and Translational Medicine, University of Brescia, Brescia, Italy, <sup>2</sup> Section of Cytogenetics and Molecular Genetics, Clinical Chemistry Laboratory, ASST Spedali Civili di Brescia, Brescia, Italy, <sup>3</sup> Neurology Unit, Centre for Neuromuscular Diseases and Neuropathies, ASST Spedali Civili di Brescia, Brescia, Italy*

Keywords: iron, neurodegeneration, autophagy, neuroinflammation, exosomes

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

#### **Iron and Neurodegeneration**

As a co-factor of many proteins, iron is involved in essential biological processes, from DNA synthesis to mitochondrial respiration, and oxygen transport. In the central nervous system (CNS), it is required for further crucial functions like myelination and synthesis of neurotransmitters. Iron can cycle between two redox states, Fe2<sup>+</sup> and Fe3+. While this cycling is essential for crucial ox-redox reactions, it can be harmful, generating reactive oxygen species (ROS) through the Fenton reaction. Both brain iron deficiency and overload are noxious and in health conditions brain iron homeostasis is tightly regulated at the blood brain barrier (BBB) and in all cells. Brain iron increases in healthy aging, due to BBB dysfunctions or chronic inflammation that are typical of the elderly, but it raises also in several neurodegenerative diseases, in which iron dyshomeostasis and redistribution are frequently observed. Since distinct neurodegenerative disorders have their own etiology, pathogenesis and course, iron dysregulation often affects different cells, and pathways. However, common hallmarks may be identified as drivers or consequences of brain iron homeostasis disruption, like protein aggregation, impaired autophagy, ox-stress, ferroptosis, neuroinflammation, and altered immunity.

The aim of this Topic Issue was to provide an updated overview on the role of brain iron dysregulation under different neurodegenerative conditions. Several researchers contributed interesting point of views on this subject that continues to be a matter of debate among neuroscientists.

The central theme of the Topic is masterly illustrated by Ndayisaba et al., who provided an accurate picture of brain iron metabolism in both health conditions and aging. The authors reviewed shared dysfunctional pathways coexisting with iron accumulation or dyshomeostasis in different neurodegenerative disorders. They focused on dysfunctions of mitochondria, site of the synthesis of heme and Fe-S clusters (ISCs); protein misfolding and aggregation in which iron seems implicated; disruption of autophagy, possibly resulting in dysfunctional degradation of ferritin (FT), the main intracellular iron storage protein, and inadequate or excessive release of bioavailable iron; iron activation of microglia and astroglia and neuroinflammation; ferroptosis induction linked to iron dysregulation. The authors provided evidence for disease modification by modulators of iron homeostasis and for the usefulness of Magnetic Resonance Imaging (MRI) technology to study iron as early or disease progression biomarker.

#### Edited and reviewed by:

*Mark P. Burns, Georgetown University, United States*

#### \*Correspondence:

*Giorgio Biasiotto giorgio.biasiotto@unibs.it Massimiliano Filosto massimiliano.filosto@unibs.it Isabella Zanella isabella.zanella@unibs.it*

#### Specialty section:

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

Received: *22 November 2019* Accepted: *06 December 2019* Published: *20 December 2019*

#### Citation:

*Biasiotto G, Filosto M and Zanella I (2019) Editorial: Iron and Neurodegeneration. Front. Neurosci. 13:1382. doi: 10.3389/fnins.2019.01382*

**5**

The dual role of iron dyshomeostasis and protein aggregation neurotoxic effects in neurodegenerative disorders was thoroughly reviewed by Joppe et al., who suggested their mutual interplay in amplifying their detrimental effects. They provided evidence of an iron role in protein aggregation through its direct binding to amyloidogenic proteins like α-synuclein in Parkinson's disease (PD), Aβ and tau in Alzheimer's disease (AD), superoxide dismutase 1 (SOD1) and TAR DNA binding protein (TDP43) in Amyotrophic Lateral Sclerosis (ALS) and prion protein (PrP) in prion diseases. Authors highlighted the indirect iron role in the regulation of α-synuclein aggregation through ox-stress induction and autophagy inactivation and the modulation of secretase activity in amyloid precursor protein (APP) processing, Aβ release, and tau phosphorylation. They emphasized the impact of mutated SOD1 on iron dyshomeostasis and the influence of PrP on cellular iron deficiency and neuroinflammation. Finally, they reviewed updated clinical applications of iron chelators as symptomatic and neuroprotective agents in PD, AD, and ALS.

The connection between iron levels and the unfolded protein response (UPR) was explored in a research report (Healy et al.). The authors took advantage of their established neonatal hippocampal slice-based model for their study. They exposed slices to iron overload and found that iron loading induced UPRrelated transcripts and loss of oligodendrocytes. These effects were partially prevented by UPR activation. In silico analysis of putative transcription factor binding sites within promoter regions of iron-related genes identified binding sites for UPRassociated transcription factors, suggesting further studies in this direction.

The role of ferroptosis was illustrated by further contributions. De Gregorio-Rocasolano et al. described iron transport to the brain through the BBB and iron uptake by brain cells, wellsummarizing the actually known cellular receptors and the main proteins involved in intracellular iron handling in different brain cells. The authors then focused their attention on the role of iron in ischemic stroke (IS) and intracerebral hemorrhage (ICH), reviewing how iron can induce ROS production and ferroptosis and the beneficial effects of iron chelators and ferroptosis inhibitors in reducing neuronal death.

Wu et al. after depicting a detailed review of known molecular mechanisms involved in ferroptosis, showed intriguing evidence of its potential role in peripartum asphyxia and intraventricular/periventricular hemorrhage in preterm infants. The authors emphasized that the neonatal immature brain is more prone to ferroptosis, due to higher rate of oxygen consumption, higher concentration of polyunsaturated fatty acids and lower endogenous antioxidant defense mechanisms and proposed a new treatment approach, based on ferroptosis inhibition, encouraging future research in this unexplored direction.

Quiles del Rey and Mancias illustrated the ferroptotic role of nuclear receptor co-activator 4 (NCOA4), involved in ferritinophagy. Although its physiological role has been mainly studied in erythropoiesis, several studies showed a link between ferroptosis and ferritinophagy. Currently there is no direct evidence for ferritinophagy dysregulation in neurodegeneration, however both in health aging and neurodegenerative conditions brain iron increases and autophagy decreases and several neurodegenerative disorders due to mutations in proteins involved in the autophagy machinery are associated with brain iron accumulation. The dual role of NCOA4 in both iron homeostasis and autophagy in neurodegeneration offers new insights for future research.

The role of crucial proteins other than NCOA4 but strictly related to brain iron dyshomeostasis and neurodegeneration were reviewed in further contributions. FT is a 24-mer formed by the assembly of two different subunits, heavy (FTH) and light (FTL) chains, with distinct functions: while FTH has ferroxidase activity, FTL promotes iron mineralization. Hereditary neuroferritinopathy (HF), reviewed by Muhoberac and Vidal, is an autosomal dominant movement disorder with brain iron accumulation, caused by FTL gene mutations. The authors reviewed in vitro and in vivo models of HF, with an emphasis on functional anomalies of the mutant FTL and detailing the role of FTL aggregation, ROS generation, iron accumulation, ferritinophagy inhibition, and ferroptosis in its pathogenesis.

Ingrassia et al. focused on the role of divalent metal transporter 1 (DMT1), a transmembrane ferrous iron transporter that takes up non-transferrin bound iron (NTBI) at the plasma membrane and iron internalized through the transferrin (Tf)/transferrin receptor 1 (TfR1) cycle at the endosome membrane. Together with Tf, NTBI is a physiologic form of iron within interstitial fluid and cerebrospinal fluid (CSF) and the main source of iron uptake for some brain cells. The authors extensively reviewed its complex expression regulation, resulting from alternative splicing, transcriptional and posttranscriptional control and summarized proofs of upregulation during aging and neurodegenerations, providing evidence of its role in their pathogenesis.

All cells maintain a proper cytosolic iron level through the post-trascriptional regulation of TfR1, DMT1, FT, and the iron exporter ferroportin 1 (FPN1), through the binding/unbinding of iron regulatory proteins 1 and 2 (IRP1 and IRP2) to the iron regulatory elements (IREs) of TfR1, DMT1, FT, and FPN1 transcripts in iron depletion/repletion conditions. IRP2 is degraded in iron-replete conditions through proteasome via the F-box and leucine rich repeat protein 5 (FBXL5), in turn regulated by iron. IRP2 stability is also sensitive to ROS, although H2O<sup>2</sup> treatment seems to stabilize IRP2 in H1299 cells or it does not affect IRP2 stability in other cells (Cairo et al., 2002; Hausmann et al., 2011). Conversely, Jiao et al. showed that H2O<sup>2</sup> treatment induces IRP2 degradation through FBXL5 and proteasome, leading to reduced cellular iron uptake in the SH-SY5Y neuroblastoma cells. These intriguing findings needed further studies.

Llorens et al. reviewed the role of frataxin (FTX) in Friedreich's ataxia (FRDA), an early-onset neurodegenerative disease that affects CNS, heart and pancreas and caused by loss of function homozygous FTX mutations. The authors provided an updated overview on FTX functions in mitochondria and the mechanisms through which its loss of function can lead to iron dyshomeostasis. The authors provided an update on how FTX deficiency and iron overload can lead to neurotoxicity in FRDA through ROS-dependent and independent mechanisms.

Ceruloplasmin (CP) is a ferroxidase that oxidates exported cellular Fe2<sup>+</sup> to Fe3<sup>+</sup> for serum Tf binding and transport in blood stream. When this activity is compromised, cells retain iron. Patients with defective CP show progressive neurodegeneration due to iron accumulation. Ryan et al. studied CP and iron metabolism in wild-type and CP null animal models of cerebral ischemia. After ischemia, mice CP expression significantly increases in wild-type mice, while iron-related proteins were partly differently modulated in the two models. These data showed that the absence of CP generated iron dyshomeostasis and ox-stress and worsen the recovery of function when cerebral ischemia occurred.

Aceruloplasminemia is a rare autosomal recessive disorder with adult onset. The clinical features include also neurological symptoms directly proportional to brain iron accumulation. Marchi et al. reviewed recent literature, highlighting the importance of recognizing early evidence of this pathology to prevent the heavy burden of clinical manifestations. The therapeutic approach to aceruloplasminemia is deepened by the review of Piperno and Alessio, who analyzed the use of iron chelators to attenuate systemic iron overload and described the poor ability of these molecules to counteract neurological symptoms and the side-effect of worsening anemia. The authors presented recent studies in animal models in which parenterally administered CP showed a reduction of systemic iron overload and a promising neuroprotective effect.

The role of iron in neuroinflammation was depicted with intriguing points of view in some contributions.

Hepcidin is the master regulator of systemic iron metabolism. Hepcidin seems implicated on iron overload and neuroinflammation. It is expressed in the brain and where perhaps there is a contribution of the hepatic form when BBB leakage occurs like in inflammation. Hepcidin expression is low in brain cells, but increases in iron overload and inflammation. In this condition, hepcidin reduces brain microvascular endothelial cells ability to transport iron. Inflammation increases neuronal iron uptake and this condition is worsened by FPN1 degradation induced by hepcidin. The therapeutic potential of hepcidin needs additional studies since it could be of importance in the treatment of neurodegenerative diseases (Vela).

Healthy individuals maintain a balance between trained and tolerized immunity. Iron changes this balance during microbe infections due to the competition between host and pathogens. The strategy of innate immune cells to seize and accumulate iron in order to limit microbe iron availability could damage mitochondria, increase ROS, and generate autoimmune inflammation. Iron and aging could be responsible for intestinal and brain microvessel permeability, causing the migration of intestinal microbes to the CNS with disruption of immune tolerization, particularly deleterious in AD. Modification of trained and tolerized immunity using exosomes and nanoparticules charged with iron chelators and/or other therapeutic molecules could be useful to treat AD (Sfera et al.).

The increase of gut permeability permits microbiota to migrate in other tissues and stay as dormant microbes. Pretorius et al. suggested iron dysregulation and reactivation of dormant microbes as the trigger of AD. Iron dyshomeostasis observed in AD might increase ox-stress, cause the reactivation of dormant microbes and increase the amyloid formation. Bacteria could contribute to amyloid deposition and neuroinflammation. Glial cell activation and the increase of BBB permeability could fuel the inflammatory progression of AD.

Siotto et al. contributed with an interesting research article, showing a connection between peripheral inflammation, oxstress, and iron homeostasis in patients with relapsing-remitting multiple sclerosis (RR-MS) and low disability. Analyzing oxstress and iron metabolism through the measure of several biomarkers in peripheral blood, the authors found that in the early phases of the disease the ox-stress status was impaired and that disease-modifying therapy could affect iron homeostasis and ox-stress. Ashraf et al. studied CSF levels of melanotransferrin (MTf), an iron-binding protein belonging to the Tf superfamily that exists as both a plasmamembrane-anchored protein and a secreted form. Although its functions are currently unclear, it was found highly expressed in reactive microglia within senile plaques and significantly increased in the serum and CSF of AD patients. The authors hypothesized the association of baseline levels of CSF MTf to classic AD CSF biomarkers in a clinically well-defined cohort of AD patients, subjects with mild cognitive impairment (MCI) both converted and not converted to AD and cognitively normal subjects and found that MCI subjects that converted to AD had lower baseline MTf levels compared to non-converters and lower CSF MTf levels were associated with greater cognitive deficits and lower hippocampal volumes in AD converter, suggesting that iron homeostasis perturbations may be an early contribution to the disease process prior to acquiring AD.

The role of iron in AD pathogenesis was extensively reviewed by Liu et al. The authors described how iron can participate in the deposition of Aβ plaques through its effect on APP expression and cleavage, binding to Aβ, and accelerating its aggregation. Iron is also implicated in the deposition of tau tangles, another pathological feature of AD, since it binds tau and induces its phosphorylation and aggregation. Authors reviewed the potential of iron-related biomarkers for AD diagnosis and progression and classic and new therapeutic approaches based on the use of iron chelators.

The role of iron in PD pathogenesis was extensively analyzed in further contributions. MRI biomarkers are increasingly applied in the diagnosis of movement disorders. Neuromelaninsensitive magnetic resonance imaging (NM-MRI) combined with nigrosome-1 (N1) imaging by quantitative susceptibility mapping (QSM) in the substantia nigra (SN) may be used to distinguish de novo PD from patients affected by essential tremor (ET) (Jin et al.). This study indicated that PD patients have greater SN iron content when compared to ET patients. Implications of iron dyshomeostasis in PD and Parkinsonisms were highlighted by Lee and Lee. Atypical parkinsonian syndromes are characterized by brain iron accumulation. Multiple system atrophy (MSA), progressive supranuclear palsy (PSP) and PD patients can be differentiated using MRI, analyzing iron-related signal changes in tomographic patterns of widespread iron deposition in deep nuclei. The authors described the reasons that can influence region specificity and underlined the pivotal importance of considering age-related, sub-structural and structure-specific patterns when MRI is used to investigate iron related neurodegeneration.

Another contribution analyzed PD patients, differentiated on the basis of symptomatology. The interesting study by Lian et al. improved our knowledge on iron-related proteins, serum and CSF inflammatory factors in PD. The authors demonstrated that tremor dominant PD patients showed increased interleukine-6 (IL-6) and iron levels and decreased FTL in CSF and increased IL-6 and FT serum levels when compared with postural instability and gait difficulty-dominant PD patients and controls. The authors concluded that an interconnection exists between dysregulated iron metabolism and inflammation in tremor dominant patients.

Dopamine was studied, considering its possibility to generate aminochrome and ROS, specifically in neurons that are characterized by pathological iron accumulation (Sun et al.). Authors observed that post-mortem tissue of brain from PD patients showed acidosis. Studies in the pH range of 7.4–6.5 indicated that iron increased toxic quinones like aminochrome. This effect was amplified by Fe2<sup>+</sup> when compared to Fe3<sup>+</sup> and aminochrome accumulation was more elevated in brain with acidosis. In acidic environment, the elevated iron redox cycling enhanced the amount of ROS in the presence of dopamine.

Nitric oxide (NO) supports physiological metabolism in the brain, but it has been linked to neurodegeneration when produced at high levels. NO metabolism is strictly connected to iron homeostasis, it regulates IRPs influencing the expression of iron-related proteins and promotes the S-nitrosylation of these proteins modifying their functions. The interdependence between NO and iron homeostasis, the modification of iron metabolism resulted by change in NO concentration in PD and AD and the chelation therapy effects were reviewed by Liu et al. Some biological questions remain however open. The improvement of the techniques used in NO detection and evaluation of iron redox state could progress this field.

Recent findings evidenced an interconnection between iron and calcium homeostasis in neurons. Iron activates calcium signaling with downstream activation of kinases that influence synaptic plasticity. Ox-stress mediated by iron increase excessively activates calcium signaling that impairs mitochondrial function and dysregulates mitochondrial iron metabolism. This anomalous crosstalk between iron and calcium homeostasis leads to neuronal death. Núñez and Hidalgo reviewed recent literature in the field, showing that the dangerous deregulation involved excitotoxicity, lipid peroxidation and damage of important mechanisms of iron and calcium pathways, contributing to neurodegenerative disorders, like AD and PD.

The role of iron in ALS was explored and reviewed by Petillon et al. The authors summarized evidence of iron accumulation and dysregulation in brain and spinal cord both in patients and animal models. The authors reviewed also studies on hepcidin levels in ALS founding contrasting results. Next, the authors dealt with the questions regarding the diversity of models, methodologies, biomarkers and heterogeneity of findings, critically reviewing the literature in order to identify the limits of currently published studies and suggesting relevant parameters to include in future studies on iron metabolism in ALS patients.

Halon-Golabek et al. contributed an original work in which mechanisms involved in iron-derived ox-stress in skeletal muscle are discussed, mainly focusing on ALS. They summarized evidence showing iron dysregulation in the skeletal muscle of subjects with muscle atrophy, suggesting that it may activate the ubiquitin-proteasome and autophagy-lysosome systems, protein degradation, and muscle mass loss. They provided evidence of a link between impaired insulin signaling pathways and iron accumulation. Further, they described iron dysregulation in the skeletal muscle of ALS patients and animal models, showing that it may contribute to ROS formation and impair muscle endocrine function, with an impact on the brain. Finally, they showed evidence of the protective role of exercise training on insulin sensitivity, iron homeostasis and myokine secretion.

Chiabrando et al. reviewed the pathological role of heme in neurodegenerative disorders and discussed the potential mechanisms generating toxicity in the brain, highlighting the importance of understanding these mechanisms to design therapeutic approaches. Traumatic brain injury is a causal factor for neurodegenerative diseases. Iron overload can take place through hemorrhage or microhemorrages due to heme-bound iron or labile free iron. Daglas and Adlard reviewed the role of iron in this pathological condition highlighting the interconnections with neurodegenerative diseases. Therapeutic options aimed to attenuate neuropathology and neurodegeneration risks involve the use of iron chelators that can penetrate BBB and antioxidants agents.

Taken together, the papers collected in this Issue present the most recent knowledge and experimental evidence about the role of iron in neurodegenerative disorders and offer new perspective and interesting hypotheses on this topic.

# AUTHOR CONTRIBUTIONS

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

# ACKNOWLEDGMENTS

We are grateful to all authors that greatly contributed to the success of the Topic Issue, allowing us to provide the readers an updated overview on this matter. We wish also to thank all reviewers that highly contributed with their knowledge and useful advices. Finally, a heartfelt thank you to the Editorial Office that with expertise and in a professional manner helped us in the management of this collection.

# REFERENCES

Cairo, G., Recalcati, S., Pietrangelo, A., and Minotti, G. (2002). The iron regulatory proteins: targets and modulators of free radical reactions and oxidative damage. Free Radic. Biol. Med. 32, 1237–1243. doi: 10.1016/S0891-5849(02)00825-0

Hausmann, A., Lee, J., and Pantopoulos, K. (2011). Redox control of iron regulatory protein 2 stability. FEBS Lett. 585, 687–692. doi: 10.1016/j.febslet.2011.01.036

**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 Biasiotto, Filosto and Zanella. 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.

# Iron and Alzheimer's Disease: From Pathogenesis to Therapeutic Implications

Jun-Lin Liu<sup>1</sup>† , Yong-Gang Fan<sup>1</sup>† , Zheng-Sheng Yang<sup>2</sup> , Zhan-You Wang1,3 \* and Chuang Guo<sup>1</sup> \*

<sup>1</sup> College of Life and Health Sciences, Northeastern University, Shenyang, China, <sup>2</sup> Department of Dermatology, First Hospital of Qinhuangdao, Qinhuangdao, China, <sup>3</sup> Key Laboratory of Medical Cell Biology of Ministry of Education, Institute of Health Sciences, China Medical University, Shenyang, China

#### Edited by:

Massimiliano Filosto, Asst degli Spedali Civili di Brescia, Italy

#### Reviewed by:

Scott Ayton, Florey Institute of Neuroscience and Mental Health, Australia Peng Lei, West China Hospital of Sichuan University, China

#### \*Correspondence:

Zhan-You Wang wangzy@mail.neu.edu.cn Chuang Guo guoc@mail.neu.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 03 July 2018 Accepted: 22 August 2018 Published: 10 September 2018

#### Citation:

Liu J-L, Fan Y-G, Yang Z-S, Wang Z-Y and Guo C (2018) Iron and Alzheimer's Disease: From Pathogenesis to Therapeutic Implications. Front. Neurosci. 12:632. doi: 10.3389/fnins.2018.00632 As people age, iron deposits in different areas of the brain may impair normal cognitive function and behavior. Abnormal iron metabolism generates hydroxyl radicals through the Fenton reaction, triggers oxidative stress reactions, damages cell lipids, protein and DNA structure and function, and ultimately leads to cell death. There is an imbalance in iron homeostasis in Alzheimer's disease (AD). Excessive iron contributes to the deposition of β-amyloid and the formation of neurofibrillary tangles, which in turn, promotes the development of AD. Therefore, iron-targeted therapeutic strategies have become a new direction. Iron chelators, such as desferoxamine, deferiprone, deferasirox, and clioquinol, have received a great deal of attention and have obtained good results in scientific experiments and some clinical trials. Given the limitations and side effects of the long-term application of traditional iron chelators, alpha-lipoic acid and lactoferrin, as self-synthesized naturally small molecules, have shown very intriguing biological activities in blocking Aβ-aggregation, tauopathy and neuronal damage. Despite a lack of evidence for any clinical benefits, the conjecture that therapeutic chelation, with a special focus on iron ions, is a valuable approach for treating AD remains widespread.

Keywords: Alzheimer's disease, iron, chelation, alpha-lipoic acid, lactoferrin

# INTRODUCTION

Alzheimer's disease (AD) is a neurodegenerative disease that occurs in the elderly population. Most patients show early loss of memory, and as the condition worsens, language disorders, loss of directionality, and anxiety behaviors will also be present (Cheng et al., 2013). Regarding late-stage patients, their mental activities, such as cognition, emotion and behavior, are abnormal, and their bodily functions are gradually lost (Ikonomovic et al., 2011). With the development of society and changes in the human environment, the incidence of AD has increased year by year. In the epidemiological survey performed by the Alzheimer's Association in the United States in 2017, the number of AD patients in the United States exceeded 5.5 million, and people over the age of 65 were found to be twice as likely to suffer from AD. However, the pathogenesis of AD remains unclear, and drugs that can completely cure AD or relieve symptoms have not yet been developed (Blennow et al., 2006; Huang and Mucke, 2012; Mullard, 2012; Reese et al., 2012; Selkoe, 2012).

# THE PATHOGENIC HYPOTHESIS OF ALZHEIMER'S DISEASE

fnins-12-00632 September 6, 2018 Time: 19:35 # 2

The two major histopathological features of AD in the brain are senile plaques (SPs), formed by the deposition of extracellular β-amyloid protein (Aβ), and neurofibrillary tangles (NFTs), formed by hyperphosphorylation of tau proteins associated with microtubules in neurons (Colvez et al., 2002). Based on these obvious pathological features, scholars have proposed two hypotheses about the developmental mechanism of AD: the amyloid cascade hypothesis and the NFTs hypothesis. With the deepening of research, more recent scholars also proposed the hypothesis of inflammation and the metal ion hypothesis and continuously enriched the developmental mechanism of AD.

The amyloid cascade hypothesis stems from the amyloid degradation pathway of β-amyloid precursor protein (APP). This hypothesis states that Aβ1-42, produced by the amyloid degradation pathway of APP, has significant neurotoxicity, can induce aggregation and hyperphosphorylation of tau protein, form NFTs, cause neuronal damage, and eventually lead to dementia (Klein et al., 2001). The NFTs hypothesis originates from the presence of a large number of fiber tangles formed by the aggregation of hyperphosphorylated tau proteins in the neurons of AD patients. This hypothesis suggests that hyperphosphorylated tau protein competes with normal tau protein for binding to tubulin, disrupting the dynamic balance of microtubule assembly and disaggregation (Cardenas-Aguayo Mdel et al., 2014) and resulting in impaired axonal transport and accumulation of intracellular waste. Neurons gradually degenerate and cause dementia. In addition to amyloid plaques and NFTs, the researchers found a large amount of activated astrocytes and microglia in the brains of AD patients, accompanied by increased expression levels of TNF-α, IL-1β, IL-6 and other inflammatory factors (Cunningham et al., 2005). Therefore, the neuroinflammation hypothesis was proposed. This hypothesis suggests that neuroinflammation is not a passive system in AD that is activated by SPs and NFTs, but, like plaques and tangles, plays an important role in the development of diseases (Zhang et al., 2013).

Interestingly, substantial evidence has indicated that the steady-state dysregulation of metal ion metabolism in vivo can be involved in the pathology of AD (Kim et al., 2018). An imbalance in the metal levels in the brains of AD patients has been identified and is accompanied by metal-catalyzed oxidative damage (Nunomura et al., 2001; Perry et al., 2003). A large number of studies have shown that metal ions, such as copper, iron, zinc, magnesium and aluminum, are involved in the occurrence and development of AD (Wang and Wang, 2017). Clinical studies have also shown elevated levels of copper, iron and zinc in the brains of AD patients (Bush, 2013). Metal ions can affect neuronal metabolism, cause oxidative stress, and promote the deposition of Aβ and the formation of SPs (Lovell et al., 1998). At the same time, studies have also shown that the deposition of Aβ in the brain and its toxicity are directly related to metabolic disorders of zinc, copper, iron, and other metal ions in the cortex and the hippocampus (Liu et al., 2006). Studies have also shown that an imbalance of metal homeostasis can directly cause neuronal dysfunction (Myhre et al., 2013) and lead to neuronal cell death (Wright and Baccarelli, 2007). Moreover, the successful application of metal (zinc, copper, iron) chelators in several animal models of AD and patients with early AD provided strong evidence that AD is a transition metal-overloading disease (Guo et al., 2013b, 2015, 2017; Dusek et al., 2016; Giampietro et al., 2018; Kawahara et al., 2018; Zhang et al., 2018). Based on the research described above, the metal ion hypothesis was proposed to emphasize the role of metal ions in the pathogenesis of AD, which further complemented the pathogenesis of AD.

# BRAIN IRON DYSHOMEOSTASIS AND THE PATHOPHYSIOLOGY OF ALZHEIMER'S DISEASE

Iron is the second most abundant metal on earth, second only to aluminum, and it is also an essential element for the survival of all living things on earth (Crielaard et al., 2017). The biological activity of iron depends, to a large extent, on its effective electron transfer properties, allowing it to accept or provide electrons during the transition between divalent, ferric and tetravalent iron states of ferrous iron, thus serving as a catalytic cofactor in a variety of biochemical reactions (Hohenberger et al., 2012). Iron also promotes the activity of various biological enzymes in the process of DNA replication and repair in the form of iron–sulfur clusters (Fe-S). Simultaneously, iron is also a component of hemoglobin and myoglobin, which is involved in the transport of oxygen and carbon dioxide in organisms (Takeda, 2004).

Many important physiological activities in the brain involve iron. If iron is absent during the development of the brain, it will cause irreversible developmental delays; however, if the iron overload in the brain also has a neurotoxic effect, this will damage the normal physiological activities of the brain. The iron content in the brain gradually increases with age. Interestingly, using magnetic resonance imaging (MRI), it was found that the iron content in the brains of AD patients was significantly increased (Du et al., 2018). This finding was also confirmed in a comparative study of APP/PS1E9 double transgenic AD mice and wild-type mice of the same age (Dwyer et al., 2009). Besides, some scholars believe that an iron metabolism disorder is an important cause of late-onset AD formation (Corder et al., 1995). Based on the various discoveries related to iron, researchers realized that iron plays an extremely important role in the occurrence of AD. Therefore, research on iron as a target has gradually become a new direction for scientists to explore the pathogenesis of AD.

## Iron Absorption and Transport Into the Brain

Iron mainly exists in non-heme iron and heme iron in foods, and non-heme iron accounts for 90% of them. Non-heme iron is reduced to Fe2<sup>+</sup> in the upper part of the small intestine and enters mucosal epithelial cells via a divalent metal ion transporter (DMT1) on the membrane of the small intestine epithelium.

The remaining 10% of heme iron reacts with heme oxygenase (HO) in the proximal part of the small intestine, releasing Fe2<sup>+</sup> and is directly taken up by small intestinal mucosa epithelial cells via DMT1 (Krishnamurthy et al., 2007; Horl, 2008). Fe2<sup>+</sup> uptake by epithelial cells can be used directly by cells, while unutilized Fe2<sup>+</sup> is oxidized by ferrous oxidase (hephaestin) or ceruloplasmin to Fe3+. The resulting Fe3<sup>+</sup> is transported out of the cell via the ferroportin 1 (FPN1) of the basolateral membrane of the intestinal mucosal epithelium. The transferred out Fe3<sup>+</sup> is mainly combined with the transferrin (Tf) in the blood to form the iron-transferrin complex and can also be combined with lactoferrin to form a non-transferrin-bound iron into the peripheral blood circulation (Frazer and Anderson, 2005; Garrick and Garrick, 2009).

The iron-transferrin complex circulating in the peripheral blood to the brain enters the cells through endocytosis of brain capillary endothelial cells. This endocytosis is mediated mainly by transferrin receptor (TfR) on the surface of endothelial cells. Non-transferrin-bound iron in peripheral blood can enter the brain through the lactoferrin/lactoferrin receptor pathway (Ke and Qian, 2007). Iron-transferrin complexes form endosomes into cells via endocytosis. Due to the action of the proton pump on the endosome membrane, the PH in the endosome is reduced, which leads to the dissociation of iron and Tf/TfR complexes, at the same time Fe3<sup>+</sup> is reduced to Fe2+. Fe2<sup>+</sup> enters the endothelial cytoplasm through DMT1 on the endosome membrane (**Figure 1**). The detached iron Tf/TfR complex exudes from the vesicle to the lateral lumen of the endothelial cells. In the pH 7.4 environment, Tf dissociates from TfR and re-enters the blood (Dringen et al., 2007).

# Iron Regulation in the Brain

According to autopsy reports, total iron deposition in the human brain is positively related to age and contains high concentrations of iron in the basal ganglia of the putamen, globus pallidus, and substantia nigra (Connor et al., 1995), whereas the cerebral cortex, the brainstem and cerebellum contain low concentrations of iron (Zecca et al., 2004; Ramos et al., 2014). Iron homeostasis in nerve cells is mainly regulated by the transcriptional levels of mRNAs involved in iron metabolism. The proteins involved in brain iron metabolism mainly include iron regulatory proteins (IRPs), Tf, TfR1, ferritin, FPN1, DMT1,

Transferrin-bound iron binds TfR1 through endocytosis to form endosomes in the cells. Due to the action of the proton pump on the endosome, trivalent iron dissociates from the Tf/TfR1 complex and is reduced to divalent iron, which enters the cytoplasm via DMT1. Part of the ferrous iron that enters the cytoplasm is used by the cell itself (such as mitochondria), and part of it is oxidized to ferrous iron by ferritin and stored. Another part is oxidized to ferric iron by the ferroxidase on the cell membrane, and the cells are exported by FPN1 and recombined with extracellular Tf.

and so on (Crielaard et al., 2017) The mRNAs encoding TfR1, ferritin, FPN1 and DMT1 all contain a special amino acid sequence called the iron regulatory element (IRE). Iron regulates the transcription of iron-related proteins by controlling the binding of IRPs to IRE, thereby maintaining intracellular iron homeostasis (Zhou and Tan, 2017). After continuous in-depth research, people further realized that hepcidin, an antibacterial peptide, is an important factor in iron homeostasis, particularly brain iron homeostasis (Vela, 2018). FPN1 is the major receptor for hepcidin in vivo. A series of studies demonstrated that hepcidin regulates iron homeostasis via direct interaction between hepcidin and FPN1, inducing the internalization and degradation of FPN1 reducing the ability of cells to export iron (De Domenico et al., 2008), and thereby increasing the possibility of intracellular iron overload (Daher et al., 2017). Later, cell experiments showed that hepcidin not only downregulated the expression of FPN1 in astrocytes and neurons but also downregulated the expression of TfR and DMT1 (Du et al., 2011, 2015). The above results suggested that hepcidin regulates iron homeostasis not only by controlling the iron output of cells but also by regulating the iron input of cells.

In addition to iron-related proteins that regulate iron, APP and tau also act to regulate iron. Studies have shown that APP is a certain regulator for iron homeostasis, which can interact with FPN1 to regulate the efflux of ferrous ions (Kawahara et al., 2017). Indeed, APP knockout or haplo-insufficiency preferentially mediates brain iron accumulation in mice (Duce et al., 2010). Tau acts as an intracellular microtubule-associated protein, which can transport the produced APP to the cell surface to promote iron output (Li et al., 2015). Interestingly, tau knockout mice develop age-dependent iron accumulation and brain atrophy, and iron retention in the primary cultured neurons is caused by decreasing surface trafficking of APP, indicating that tau-mediated iron homeostasis might be APP-dependent(Lei et al., 2012; Tuo et al., 2017).

# Iron Participates in the Occurrence of Alzheimer's Disease

In the brain, iron is not only involved in the synthesis of myelin and neurotransmitter synthesis and metabolism but also plays an important role in maintaining the high metabolic capacity of neurons (Gerlach et al., 1994). Under normal physiological conditions, iron metabolism maintains homeostasis in the brain. Once iron metabolism is out of balance, it will have different effects on brain function. As early as Goodman (1953) found that iron was increased in the SPs in the brains of AD patients. Later, the use of quantitative susceptibility map (QSM) once again confirmed the co-localization of brain iron and Aβ plaques, and showed that the co-localization of brain iron deposition and Aβ plaques promoted the development of the disease (van Bergen et al., 2016). In fact, there are also progressive iron deposits that occur in the normal aging process of the brain, particularly in the substantia nigra, globus pallidus, caudate nucleus, and cortex, and these brain regions are closely related to neurodegenerative diseases (Rodrigue et al., 2011; Callaghan et al., 2014; Collingwood and Davidson, 2014; Ward et al., 2014). Compared with healthy people of the same age, the iron deposition in patients with AD is more serious in these areas. Further, the APP mRNA of peripheral blood mononuclear cells of AD patients was significantly lower than that of the control group by fluorescence quantitative PCR, which indirectly indicated the iron dyshomeostasis in AD (Guerreiro et al., 2015).

#### Iron Participates in the Deposition of Aβ Plaques and Tau Tangles

Studies have demonstrated that iron metabolic disorders can induce the production and accumulation of Aβ because iron can act on the IRE site of APP mRNA, thereby enhancing the translation and expression of endogenous APP (Cahill et al., 2009). It has also been found that long-term administration of high concentrations of iron in APP/PS1E9 double transgenic mice results in an increase in the number of SPs in the brain (Smith et al., 1997). When extended iron exposure to 12 months, the increased brain iron with 3,5,5-trimethylhexanoyl ferrocene diet accelerated the formation of SPs and microglial iron inclusions in APP mice (Peters et al., 2018).

Although there is a lot of evidence that iron and Aβ plaques co-localize, it is not known what form iron is present in plaques. Recently, Plascencia-Villa et al. (2016) used transmission electron microscopy (TEM) to confirm that iron is present in the core of SPs in the form of iron oxide (Fe3O4) magnetite nanoparticles. This provides evidence of metal biology associated with iron accumulation and Aβ aggregation. Later, using in situ X-ray magnetic circular dichroism again revealed the existence of magnetite in human SPs (Everett et al., 2018). Magnetite, as a polycrystalline iron oxide, is not a normal feature in the human brain, and its elevated content indicates that the anomalous iron redox chemistry affects AD (Ayton et al., 2017b). Furthermore, Telling et al. (2017) used advanced sub-microscopic resolution of X-ray microscopy to find evidence of the direct correlation between the morphology of SPs and iron and the formation of iron-amyloid complexes. Importantly, Aβ binds to iron through three histidine residues and one tyrosine residue in the hydrophilic N-terminal region of the peptide, which helps to stabilize these iron ions (Lane et al., 2018). In turn, studies also found that the binding of ferrous ions to Aβ reduced the peptide helix structure and increased the β-sheet content of the peptide, indicating that ferrous ions promote amyloid monomers to form oligomers and fibrils by enhancing the interaction between peptide-peptides (Boopathi and Kolandaivel, 2016; Tahirbegi et al., 2016). Except promoting Aβ aggregation, high iron levels can impact amyloidogenic processing of APP. Early studies found that iron had a modulatory effect on the α-secretase cleavage activity of APP (Bodovitz et al., 1995). Later, studies found that the process of converting α-secretase and β-secretase from the inactive state to the active state was regulated by furin, and iron could regulate the expression of furin at the transcriptional level (Silvestri and Camaschella, 2008). Excessive iron inhibits the expression of furin, which favors the activation of β-secretase, thereby promoting the production of Aβ from the amyloid pathway (Ward et al., 2014). Another study found that presenilin enhancer 2 (PEN-2), a γ-secretase components, could bind to iron through the ferritin light chain, enhancing

γ-secretase activity, thereby increasing Aβ formation (Li et al., 2013).

In vitro studies have also found that iron can promote aggregation of Aβ peptides and increase their cytotoxicity (Tahmasebinia and Emadi, 2017; Galante et al., 2018). However, there are different opinions on the role of iron and Aβ. Earlier studies have found that Fe2<sup>+</sup> and Fe3<sup>+</sup> interacted with APP and Aβ to promote aggregation of Aβ into fibrous forms (Ha et al., 2007). Fe2<sup>+</sup> can also interact with the amino acids of the Aβ protein, which may impart changes in the form of amyloid in a different manner than copper and zinc (Dahms et al., 2012). Fe3<sup>+</sup> bound to Aβ is easily reduced to Fe2<sup>+</sup> and increases reactive oxygen species (ROS) production, which causes β-secretase to cleave monomer Aβ42 into more toxic Aβ oligomers, accelerating neuronal death (Cohen et al., 2013; Balejcikova et al., 2018). Importantly, Aβ can damage mitochondrial function, convert Fe3<sup>+</sup> into Fe2<sup>+</sup> with redox activity, and induce oxidative stress, thereby aggravating iron overload and aggravating the AD condition (Everett et al., 2014; Mena et al., 2015). Further, it has been shown that iron exposure promotes the accumulation of APP in cultured SHSY5Y cells, along with the increase of β-secretase activity and Aβ42 in the medium (Banerjee et al., 2014). Inconsistently, a recent study showed that iron treatment of neurons promoted the APP non-amyloid pathway, altered the distribution of sAPPα and retained it in cell lysates rather than secreted outside the cell, while iron did not change β-secretion enzyme expression, but significantly inhibits its activity (Chen et al., 2018). Another study also found that Aβ can significantly reduce the redox ability of iron, which may indicate neuroprotection and metal chelation of Aβ during the pathogenesis of AD, but becomes toxic under certain conditions (Lane et al., 2018).

Neurofibrillary tangles are another major pathological feature of AD, and phosphorylated tau protein is the main component of NFTs. Studies have found iron deposition in neurons with NFTs (Smith et al., 1997). In addition to Aβ peptides, iron can bind to tau protein, induce tau protein phosphorylation, and aggregate phosphorylated tau protein, whereas this phenomenon can be reversed by iron chelators (Amit et al., 2008). Fe3<sup>+</sup> induces the aggregation of hyperphosphorylated tau protein, and when Fe3<sup>+</sup> is reduced to Fe2+, its induced aggregation can be reversed (Yamamoto et al., 2002). These results reveal that iron may play

an important role in the accumulation of hyperphosphorylated tau protein to form NFTs. Recent studies have shown that tau protein indirectly participates in the transmission of iron ions in brain neurons during the pathogenesis of AD (Lei et al., 2012). Moreover, in vitro and in vivo experiments have shown that iron is involved in the hyperphosphorylation of tau protein through the activation of the cyclin-dependent kinase (CDK5)/P25 complex and glycogen synthase kinase-3β (GSK-3β) (Xie et al., 2012; Guo et al., 2013a), but no relevant experiments have reported whether iron is also related to the inactivation of the protein phosphatase PP2A (**Figure 2**). Additionally, Fe3<sup>+</sup> can promote the reduction of superoxide radicals released by mitochondria (Kudin et al., 2004; Aliaga et al., 2011). The reduction of Fe3<sup>+</sup> can lead to the production of superoxide, and the reaction of superoxide with nitric oxide (NO) to produce per-nitrate can damage tyrosine residues that function normally (Nakamura and Lipton, 2011). In AD, nitration of tau prevents its stabilization of the microtubule lattice, and nitrating tau protein has been observed in tau entanglement and SPs (Reynolds et al., 2006; Kummer et al., 2011). Interestingly, the accumulation of tau in NFTs is also associated with increased induction of HO1 (Wang et al., 2015). HO1 is a potent antioxidant that can metabolize heme released from damaged mitochondria, and it also promotes the release of Fe2+, which may cause free radicals to initiate additional oxidative stress (Ward et al., 2014). Thus, iron-induced oxidative stress may also promote hyperphosphorylation and aggregation of tau (Lane et al., 2018).

To date, although the specific mechanism of iron involvement in the aggregation of Aβ plaques and hyperphosphorylated tau protein is not yet clear, it has been shown that iron can accelerate this process by affecting related signaling pathways and the three-dimensional conformation of proteins. Nevertheless, the changes in iron levels in AD need to be emphasized, and homeostatic imbalance may have dual effects through iron induction. On the one hand, iron overload areas in the brain contribute to oxidative stress and cell death around SPs and NFTs; on the other hand, other areas of the brain may suffer from impaired neuronal function due to iron deficiency (Belaidi and Bush, 2016).

#### Ferroptosis and AD

Cell death plays an important role in the growth and development of organism and tissue homeostasis. Studies have found that cell death is dysregulated in AD (Yang and Stockwell, 2016). Based on the unique pathological state caused by iron overload, scholars have also proposed a fourth cell death mode that differs from apoptosis, necrosis, and autophagy, namely, ferroptosis (Dixon et al., 2012).

Ferroptosis refers to iron-dependent lipid peroxidationinduced cell death that depends on ROS production and iron availability, with severe lipid peroxidation (Dixon, 2017). One of the hallmarks of ferroptosis is the iron-dependent accumulation of lipid ROS, a form of death that is dependent on intracellular iron rather than other metals (Abdalkader et al., 2018). The morphological features of ferroptosis are mainly reflected in intracellular mitochondria. Compared to the mitochondria of normal cells, the mitochondrial volume of ferroptosis cells is smaller, the density of mitochondrial membranes is reduced, mitochondrial hemorrhoids are reduced or have disappeared, and the mitochondrial outer membrane ruptures (Xie et al., 2016). Moreover, studies have found that the occurrence of ferroptosis consists of the accumulation of lipid ROS triggered by the inactivation of the intracellular antioxidant glutathione (GSH). Therefore, ferroptosis is caused by the imbalance in cellular redox homeostasis (Gao et al., 2016). Glutathione peroxidase 4 (GPx4), an antioxidant defense enzyme that repairs oxidative damage to lipids, is a central endogenous suppressor of ferroptosis (Chen et al., 2015). Studies have found that the GPx4 gene knockout mice involve in both three pronounced hallmarks of ferroptosis (iron dysregulation, lipid peroxidation, inflammation) and prodromal indices of AD (behavior dysfunction, hippocampal neurodegeneration), and these pathological changes can be ameliorated or prevented by a ferroptosis inhibitor (Seiler et al., 2008; Hambright et al., 2017). Intriguingly, erastin, a ferroptosis attractant, can induce neuronal death accompanied by ferroptosis (Hirata et al., 2018). Conversely, iron chelators and antioxidants specifically involved in protecting cells against ferroptosis (Hambright et al., 2017). Taken together, although the physiological function of ferroptosis is still unclear, its role in age-related neurodegenerative diseases (including AD) has been established. This suggests that considering ferroptosis the center, the development of ferroptosis inhibitors may be a new direction to alleviate the symptoms of AD.

# BIOMARKERS FOR CLINICAL DIAGNOSIS OF AD PROGRESSION

The pathogenesis of AD has not yet been fully explained. If special methods can be used to detect AD in the early stage, it is very important to increase our understanding of the disease and active prevention of the disease. Currently, the clinical diagnosis of AD is based primarily on family history and certain cognitive dysfunctions (Alzheimer's, 2016). Other methods mainly detect the levels of Aβ1-42 and phosphorylated tau and total tau in cerebrospinal fluid to predict the development and severity of AD (Bulk et al., 2018). A common method for detecting Aβ plaques is positron emission tomography (PET). With the deepening of research, more and more new technologies are used to identify changes in the Alzheimer brain, and gradient echo multiple contrast imaging (GEPCI) technology is one of them. The study confirmed a strong correlation between the GEPCI brain tissue index and the Aβ load defined by PET, providing a new method for assessing AD-related histopathology in the preclinical and early symptom stages of AD (Zhao et al., 2017).

Although Aβ and tau degeneration are considered to be key factors in AD, iron dyshomeostasis is increasingly reported as a potential cause of AD pathophysiology. Can iron be used as a biomarker for the progression of AD to detect the occurrence of AD and reflect the severity of AD? For this reason, iron as a biomarker of AD has become a research hotspot for scientists. The current effective method for detecting brain iron levels is QSM. QSM has better specificity and can be used to

non-invasively assess tissue magnetic susceptibility, which has been confirmed to have a good correlation with brain iron levels (Du et al., 2018). Through QSM, Du et al. (2018)found that the magnetic susceptibility values of the caudate nucleus on the left side of the brain may be used as biomarkers for the severity of mild and moderate AD disease. As early as Moon et al. (2016) have found that the magnetic susceptibility values of the core and caudate nucleus of AD patients are significantly different from those of the control group. Later, van Bergen et al. (2018) jointly used QSM and PET to show that the local correlation of Aβ plaque load and iron deposition can provide relevant information about the cognitive performance of healthy elderly people. Studies have shown that AD patients have higher iron accumulation in the frontal cortex, and the degree is related to the number of Aβ plaques and NFT, and there is a different iron distribution, and these changes appear to occur after the development of AD pathology markers (van Duijn et al., 2017). These findings suggest that changes in iron-based magnetic resonance (MRI) contrast can be used to indirectly determine the extent of AD pathology. Although iron concentration was identified as the main source of comparison in QSM in the brain, O'Callaghan et al. (2017) found a correlation between QSM and tau concentration, suggesting that QSM may be a useful biomarker for early detection of tau pathology in AD.

As a major iron storage protein in the body, ferritin is closely related to AD. Elevated ferritin in CSF has been shown to be associated with poor cognitive function and increases the risk of mild cognitive impairment to AD (Ayton et al., 2015). Moreover, research has shown ferritin's potential to contribute to a blood biomarker panel for preclinical AD (Goozee et al., 2017). CSF ferritin levels were negatively correlated with cognitive performance and strongly correlated with apolipoprotein E (ApoE) in CSF, suggesting that CSF ferritin as an indicator of brain iron load may be a biomarker of AD cognitive function (Ayton et al., 2015). However, another study combining QSM and PET to predict the value of longitudinal cognitive deterioration found that iron can worsen cognitive function in the presence of Aβ pathology. There is no Aβ pathology, and iron has no correlation with cognitive function (Ayton et al., 2017a). More advanced, Ayton et al. (2018) further found that high concentrations of CSF ferritin can accelerate the decrease of Aβ in CSF of 296 participants, which supports the possibility that iron may promote Aβ deposition and accelerate disease progression. This is the first clinical evidence that iron is associated with amyloid plaque formation.

Due to the invasiveness of lumbar puncture, the use of this CSF biomarker limits the widespread clinical application, while serum or plasma biomarkers are relatively simple to acquire and less invasive, and thus have great potential for application in AD. Studies have shown that serum and iron-related protein levels in AD patients are significantly elevated (Sternberg et al., 2017). Unlike Aβ peptides, iron and iron-related proteins are significantly associated with cognitive assessment tests, neuroimaging and clinical data (Sternberg et al., 2017). This at least partly indicates that iron can act as a biomarker for AD.

# IRON-TARGETING TREATMENT STRATEGIES

Metal-chelating agents can bind metal ions to the inside of a chelating agent through the strong binding action of a chelating agent molecule with a metal ion to become a stable compound with a larger molecular weight, thereby preventing metal ions from acting. Since iron overload plays an important role in the occurrence and development of AD, the use of metal chelators to reduce excessive iron in certain areas of the brain of AD patients, to achieve the strategy of relieving or even treating AD, has received an increased amount of attention. For a metal chelator to effectively exert chelation, it must have the following characteristics: (1) easy to penetrate the cell membrane and blood-brain barrier (BBB); (2) target iron-enriched areas without depleting transferrin-bound iron from plasma; (3) remove chelated iron from iron accumulation sites or transfer it to other biological proteins, such as circulating transferrin; and (4) no side effects or minor side effects on the body (Boddaert et al., 2007).

# Clioquinol

Clioquinol (CQ) has the chemical name 5-chloro-7-iodo-8 hydroxyquinoline and is an effective metal (iron, copper, and zinc) chelating agent. Studies have found that treatment of animal models of AD with CQ can reduce the deposition of amyloid in the brain and improve memory impairment (Cherny et al., 2001; Grossi et al., 2009). The probable cause of this phenomenon is that the high binding affinity of CQ to iron, zinc and copper ions allows it to competitively seize these metals from Aβ and prevent aggregation of Aβ (Opazo et al., 2006). Another study showed that 15-month-old APP Tg2576AD mice treated with CQ demonstrated a significantly reduced number and sizes of SPs in the brain compared with those of the sham-treated littermates (Cherny et al., 2001). Simultaneously, an in vivo experiment also found that, compared with the control group, CQ reduced the expression of APP by inhibiting the expression of β-secretase (BACE1) and γ-secretase (PS1) in the brains of APP/PS1 double transgenic mice (Wang et al., 2012). In coincidence with the results in animal models of AD, a small phase 2 clinical trial, it was also found that, compared with the control group, patients with moderately severe AD after oral CQ demonstrated slower decline in cognitive function and decreased Aβ42 levels in cerebrospinal fluid (Ritchie et al., 2003). However, it still lacks of the direct evidences supporting that CQ rescues the AD-like phenotypes via targeting iron in vivo. As is well known that tau deficiency induced age-dependent iron accumulation can be prevented by oral treatment with CQ (Lei et al., 2012, 2015). Moreover, CQ treatment can effectively prevent an iron-synuclein interaction in hA53T transgenic mice (Billings et al., 2016), as well as reverse the Fe3+-induced fibrin formation in vitro (Pretorius et al., 2013). Importantly, the formation of Aβ40 and Aβ42 aggregates in the presence of Fe3<sup>+</sup> and Cu2<sup>+</sup> were investigated, the study demonstrated that Fe3+, but not Cu2+, promotes the aggregation of Aβ40 and Aβ42, and CQ significantly reduces the Fe3+-induced Aβ42 aggregation (Tahmasebinia and Emadi, 2017). These researches provide the evidence that the anti-AD ability of CQ may, at least in part, via targeting iron and, surely, the underlying mechanism need to be further elucidated, Although it is now thought that CQ is toxic to the body, it at least opens up a promising direction for us.

# Desferoxamine, Deferasirox and Deferiprone

fnins-12-00632 September 6, 2018 Time: 19:35 # 8

Desferoxamine (DFO) is a well-proven iron chelator that inhibits the toxicity of iron or aluminum and the ROS that it induces on the body. It was first thought that DFO was a chelating agent for aluminum ions, and aluminum was thought to be an independent factor that increased the risk of AD (Campbell and Bondy, 2000). The results of Crapper McLachlan et al. (1991) showed that the degree of decline in the daily living abilities of AD patients given intramuscular injections of DFO was alleviated compared with AD patients given a placebo. In vitro experiments showed that DFO can inhibit the formation of β-sheets of Aβ1-42 and dissolve preformed plaque-like amyloid plaques (House et al., 2004). There are also studies that have shown that DFO can inhibit the translation of APP mRNA and the expression of APP whole protein and reduce the secretion of Aβ peptides (Rogers et al., 2002; Morse et al., 2004). Our research revealed that nasal feeding of DFO can reverse iron-induced memory deficits in AD mice and inhibit the formation of APP (Guo et al., 2013b). In addition to its effect on APP, DFO also influences the phosphorylation of tau protein. Fine et al. (2012) have found that DFO has the ability to phosphorylate GSK-3β, which in turn reduces the level of phosphorylated tau, but the mechanism of its inhibition of tau protein phosphorylation is not yet clear. Although DFO can inhibit the phosphorylation of tau protein in the brain of AD model mice, the effect of DFO on the phosphorylation of tau protein has not been determined in the presence of iron. Our previous experiments, which involved feeding APP/PS1 transgenic mice high concentrations of iron and then nasally administered DFO to the mice, showed that intranasal DFO treatment inhibited iron-induced tau phosphorylation through the CDK5 and GSK-3β pathways (Guo et al., 2013a). We also found that DFO can also attenuate synaptic loss in the brain of APP/PS1 transgenic mice through the P38/HIF-1α pathway (Guo et al., 2015).

Although DFO has achieved certain results in a variety of experimental mouse models and has been approved by the Federal Drug Administration (FDA) for the treatment of iron overload disease, there are still many problems in the clinical application of DFO. First, the bioavailability of DFO is poor, and the molecular size and hydrophilicity of DFO prevent it from freely crossing the BBB, reducing its availability in the central nervous system; Second, DFO cannot be taken orally and must be given by injection. The time of a single injection is long (up to 10 h) and the frequency of injections is high (5–7 times per week), resulting in low patient compliance (Crielaard et al., 2017); Again, there are many side effects, including neurotoxicity after long-term treatment and systemic metal ion depletion accompanied by anemia (Cuajungco et al., 2000), gastrointestinal malabsorption and rapid degradation (May and Bulman, 1983).

Deferasirox is the first FDA-approved oral iron repellent that can be routinely used. Its chemical name is 4-[3,5-bis(2 hydroxyphenyl)-1,2,4-tris Oxazol-1-yl]benzoic acid, and it is commonly used as a treatment for patients with thalassemia iron overload. Deferasirox's ability to bind iron is limited. This drug can only bind part of the iron and supply it to the extracellular and intracellular iron receptors. It is not easy to induce iron deficiency, but the efficacy of reducing iron accumulation is also relatively low. At present, studies have found that deferasirox plays a role in reducing brain iron accumulation. However, some studies have shown that deferasirox does not reduce brain iron accumulation or reduce iron toxicity in the brain (Sripetchwandee et al., 2016). This may be because deferasirox does not easily pass through the BBB, and its ability to bind iron is also weak. Three molecules of deferasirox are required for each binding molecule of iron, and thus higher doses are required.

The chemical name of deferiprone is 3-hydroxy-l,2-dimethyl-4-(lH)-pyridone. Like deferasirox, deferiprone is also approved for the treatment of patients with thalassemia iron overload. Deferiprone can bind almost all of the iron in the body to make it unable to further induce the production of ROS, and can also supply the bound iron to the iron receptors inside and outside the cell. Since deferiprone has a high iron-binding capacity, two molecules of deferiprone are required for each binding of one molecule of iron, and thus iron aggregation can be effectively reduced. Studies have found that deferiprone and DFO exert protective effects by reducing the rate of BBB disintegration, reducing brain iron accumulation and brain mitochondrial dysfunction in the presence of iron overload (Sripetchwandee et al., 2016). Although both deferiprone and DFO exert a protective effect, deferiprone can achieve greater advantages by oral administration.

Currently, DFO, deferasirox and deferiprone are recommended as iron overloaded first-line iron chelators, but they all have certain side effects, including allergic reactions, liver and renal dysfunction, and neuronal hearing loss (Borgna-Pignatti and Marsella, 2015). This requires further development of iron chelation therapy.

#### α-lipoic Acid

α-lipoic acid (LA) is a small molecule compound that can be naturally synthesized in mammals. It is a coenzyme that exists in mitochondria. Its structure contains hydroxyl and disulfide bonds, so it has both fat-soluble and water-soluble properties, and it easily crosses the BBB. The study found that, after 60 min of intravenous injection of LA in mice, the traces of LA were observed in the mouse cerebral cortex (Panigrahi et al., 1996). After a continuous injection over 7–14 days, the presence of LA was detected in multiple parts of the mouse brain (Arivazhagan et al., 2002). Surprisingly, however, when a certain amount of LA was administered to mice daily by gavage, LA accumulation was not detected in the mouse brain after a period of time (Chng et al., 2009). The most likely reason for this is that LA is quickly reduced to dihydro-lipoic acid (DHLA) in the stomach and is transported through the blood circulation to tissues throughout the body to participate in metabolism and is eventually excreted in urine.

In Europe, LA has been used as a therapeutic drug for more than 50 years, and it is mainly used for the treatment of diabetic polyneuropathy. Later, in a clinical study, it was found that the decline in cognitive abilities of 129 patients who may have been suffering from AD was effectively alleviated after LA treatment for some time (Fava et al., 2013). However, the sample size of this study was relatively small, and there was no randomized sample. The study subjects were likely AD patients and were not confirmed by neuropathology. Therefore, the results were not very convincing. Several studies have shown that supplementation with LA can increase the activity of acetylcholine transferase (ChAT) in the brains of rodents and relieve their cognitive impairment (de Freitas, 2010; Dwivedi et al., 2014). The activation of ChAT positively regulates the content of acetylcholine in the brain, and the decreased expression of acetylcholine is closely related to the impairment of cognitive function in AD. Therefore, LA may increase the expression of acetylcholine by activating ChAT and improve the cognitive function of AD patients. At the same time, studies have shown that the activation of ChAT is also beneficial to the neurogenesis of cholinergic neurons (Park et al., 2013), which further demonstrates that LA may improve the severity of central nervous system diseases. A large number of studies have shown that LA can reduce the release of IL-1β, IL-6, and TNFα by inhibiting the activity of the NF-κB and MAPK signaling pathways (Bierhaus et al., 1997; Wong et al., 2001; Zhang and Frei, 2001). Simultaneously, LA also inhibits the infiltration of inflammatory cells into the central nervous system and downregulates the expression of vascular cell adhesion molecule-1 (Kunt et al., 1999; Guo et al., 2016). In addition to its ability to chelate metal ions, LA can also play a role in anti-inflammation, antioxidant and activated glucose uptake and utilization. So far, no serious side effects of LA treatment have been observed, which also indicates that LA may become a trend for clinical drug development for AD treatment.

In vitro studies have shown that LA can bind with metal ions, such as Cu2+, Zn2+, and DHLA, and can form complexes with Cu2+, Zn2+, Pb2+, Hg2+, and Fe3+, thereby exerting metal chelation (Ou et al., 1995). Treatment of lens epithelial cells with LA causes a significant decrease in iron uptake rates and dynamic iron pools within the cells (Goralska et al., 2003). These results suggested that LA not only reduces the level of iron that enters the cell but also reduces the dynamic iron pool within the cell by increasing iron stores. In vivo studies have shown that feeding of LA for 2 weeks in aged rats can reduce iron accumulation in the age-related cortical regions (Suh et al., 2005). The results of Fonte et al. (2001) showed that LA enhances the solubility of Aβ in the frontal cortex of APP-overexpressing transgenic mice, confirming that LA, like other metal ion chelators, can successfully resolubilize Aβ and reduce amyloid deposition in the brains of AD patients. However, in another study, learning and memory retention of LA-treated AD mice was significantly improved, but there was no significant changes in soluble or insoluble Aβ levels in the brain (Quinn et al., 2007). Although the metal chelation of LA has been proven by many experimental results, there is controversy as to whether it can inhibit the deposition of Aβ and increase the solubility of Aβ by chelating iron. Most recently, our group also found that chronic treatment with LA effectively inhibited tau hyperphosphorylation and alleviated neuronal degeneration and abnormal behavior in P301S tau transgenic mice; the improvement is accompanied by alleviation of oxidative stress, inflammation and ferroptosis in the brains of transgenic mice (Zhang et al., 2018).

# Lactoferrin

Lactoferrin (LF) is a non-heme iron-binding glycoprotein with a molecular weight of 80 kDa that is widely present in various secretions, such as milk, saliva, and urine. Because its amino acid sequence is 60% identical to that of transferrin, it is classified as a transferrin family member (Metz-Boutigue et al., 1984). LF is composed of two spherical leaves folded from a polypeptide chain of 703 amino acids. Each molecule can bind reversibly with two iron, zinc, copper or other metal ions. The binding site is located on the two globular domains of the protein. The affinity of LF for iron is 300 times that of transferrin, and the affinity is further enhanced in weakly acidic environments, which may be related to the transfer of iron from transferrin to LF when inflammation occurs (Dhennin-Duthille et al., 2000). It has also been found that, in the central nervous system, activated microglia can also generate and release LF (Xu et al., 2017). The results of immunohistochemistry studies showed that LF accumulated in the brain tissue of patients with neurodegenerative diseases. Further studies have found abnormal LF content and distribution in the brains of AD patients (Brown et al., 1985), with a large amount of LF deposition in the SP- and NFT-enriched areas (Valverde et al., 1990). Wang et al. (2010) detected LF deposition in the brains of AD model mice, but no LF deposition was found in wild mice, and the formation of SPs preceded LF. Further studies showed that LF deposition was localized in SPs and amyloid lesions. LF deposition increased with the age of transgenic mice, but after 18 months of age, most SPs showed a decrease in LF positives (Wang et al., 2010). Using laser confocal technology in AD transgenic mice to colocalize LF and Aβ in the brain, it was found that LF is expressed on Aβ, Aβ formation precedes deposition of LF, and Aβ plaques develop as the age of the rat increases and as both the size and number of LF accumulations increase, indicating that there is a close relationship between LF and AD (van de Looij et al., 2014). In combination with other biological functions of LF, such as its participation in metal ion metabolism and the regulation of cell proliferation and apoptosis, we hypothesize that LF may delay the occurrence of active AD.

With the presence of LF receptors (LFR) on the blood membrane of BBB vascular endothelial cells, exogenous LF can easily cross the BBB; thus, in recent years, LF has been widely used as a carrier for drug targeting of the brain. Interestingly, Kamalinia et al. (2013) used the PC12 cell line to demonstrate that lactoferrin-DFO conjugates are able to interfere with apoptosis. The expression levels of autophagy markers, including Atg7, Atg12-Atg5 and LC3-II/LC3-I, increased, and the LF-conjugated peritoneal cavity mainly affected the expression levels of Capsase-3, PARP, Bax and bcl-2. Furthermore, intraperitoneal injection of LF conjugates can significantly improve the learning abilities of AD rats and reduce Aβ (Kamalinia et al., 2013),

which provides a theoretical basis for the use of LF in the treatment of neurodegenerative diseases.

Therefore, the use of exogenous LF as a therapeutic agent has been investigated to identify its roles in AD. Our group was the first to investigate whether exogenous LF administration could stimulate the non-amyloidogenic processing of APP and α-secretase catalytic activity and expression, which consequently reduce Aβ deposition and ameliorate cognitive decline in AD model mice (Guo et al., 2017). We also addressed the molecular mechanisms by which LF modulates APP processing. In fact, it has been reported in the literature that the function of human LF is similar to that of the iron chelator DFO, and it can also induce the neuroprotective effect of hypoxia-inducible factor (HIF-1α) expression under hypoxic conditions (Kawamata et al., 1993). Specifically, our results suggested that LF can enhance the α-secretase-dependent APP process through the ERK1/2-CREB and HIF-1α pathways in vitro and in vivo (Guo et al., 2017), once again proving that LF plays an active role in the treatment of AD. Consistent with previous observations, Aizawa et al. (2017) recently reported that the binding of LF to low-density lipoprotein receptor-associated protein-1 (LRP1) leads to the activation of the AMP-activated protein kinase signaling pathway, which in turn, promotes cell autophagy. This shows that LF may regulate the production of Aβ and NFTs through anti-inflammation, regulation of immunity, chelation of metal ions, regulation of cell autophagy, and may ultimately affect the occurrence and development of AD.

#### CONCLUSION

The pathogenesis of AD has been studied for decades, and although it is clear that SPs formed by Aβ deposition and NFTs formed by hyperphosphorylation of tau protein are two major pathological features of AD pathogenesis, there are many opinions about the inducing factors of SPs and NFTs that have not been clearly clarified yet. In light of the complex causes

#### REFERENCES


of AD, the study of its pathogenesis from multiple perspectives has been accepted by the public and has achieved good results. The study of the pathogenesis of AD using iron as a target has become a research hotspot in recent years. The use of iron chelators to chelate excess iron in the brains of AD patients has also become a new strategy for the treatment of AD. However, unlike other tissues and organs, there is a BBB in the brain that has strict specificity for the entry of drugs. Therefore, finding an iron chelator that easily crosses the BBB and that can function effectively must be sought from the body's self-synthesizing substances. There are many proteins or molecules that bind to iron in the body, and LA and LF are among these candidate drugs. In vitro and in vivo experiments have confirmed that direct administration of LA and LF can alleviate the symptoms of AD (Guo et al., 2017; Zhang et al., 2018), but the relevant mechanism of action on them has not been clarified, and further studies are needed. At the same time, LF-based nano-drug molecules are also under study. This drug molecule attaches to the LF molecule and can easily cross the BBB, thereby exerting a therapeutic effect on lesions and achieving better therapeutic effect. Future research will also use iron as an opportunity to study the mechanism of the occurrence and development of AD from the iron steady state to more fully explain the causes of AD.

#### AUTHOR CONTRIBUTIONS

J-LL and Y-GF wrote the manuscript. Z-SY, Z-YW, and CG approved and revised the final manuscript. All the authors have read and approved the final manuscript.

#### FUNDING

This work was supported by the Natural Science Foundation of China (U1608282).

various brain regions of aged rats. Exp. Gerontol. 37, 803–811. doi: 10.1016/ S0531-5565(02)00015-3


SHSY5Y cells: protective action of negletein. Neuromolecular Med. 16, 787–798. doi: 10.1007/s12017-014-8328-4


fnins-12-00632 September 6, 2018 Time: 19:35 # 11


memory impairment in a transgenic mouse model of Alzheimer's disease. J. Alzheimers Dis. 17, 423–440. doi: 10.3233/JAD-2009-1063


fnins-12-00632 September 6, 2018 Time: 19:35 # 12


of the Alzheimer's amyloid precursor protein transcript. J. Biol. Chem. 277, 45518–45528. doi: 10.1074/jbc.M207435200


**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 Liu, Fan, Yang, Wang and Guo. 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.

# Unraveling the Role of Heme in Neurodegeneration

#### Deborah Chiabrando\* † , Veronica Fiorito† , Sara Petrillo and Emanuela Tolosano

Molecular Biotechnology Center, Department of Molecular Biotechnology and Health Sciences, University of Torino, Turin, Italy

Heme (iron-protoporphyrin IX) is an essential co-factor involved in several biological processes, including neuronal survival and differentiation. Nevertheless, an excess of free-heme promotes oxidative stress and lipid peroxidation, thus leading to cell death. The toxic properties of heme in the brain have been extensively studied during intracerebral or subarachnoid hemorrhages. Recently, a growing number of neurodegenerative disorders have been associated to alterations of heme metabolism. Hence, the etiology of such diseases remains undefined. The aim of this review is to highlight the neuropathological role of heme and to discuss the major heme-regulated pathways that might be crucial for the survival of neuronal cells. The understanding of the molecular mechanisms linking heme to neurodegeneration will be important for therapeutic purposes.

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Michael Huang, The University of Sydney, Australia Raffaella Gozzelino, Centro de Estudos de Doenças Crónicas (CEDOC), Portugal

\*Correspondence:

Deborah Chiabrando deborah.chiabrando@unito.it

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 19 July 2018 Accepted: 19 September 2018 Published: 09 October 2018

#### Citation:

Chiabrando D, Fiorito V, Petrillo S and Tolosano E (2018) Unraveling the Role of Heme in Neurodegeneration. Front. Neurosci. 12:712. doi: 10.3389/fnins.2018.00712 Keywords: heme, neurodegeneration, FLVCR, FLVCR1, FLVCR1a, neuropathy

# INTRODUCTION

Neurodegeneration is a complex process leading to the progressive and selective loss of neurons. Mitochondrial dysfunction, oxidative stress, protein aggregation and endoplasmic reticulum stress are well established pathways driving the neurodegenerative process (Lansbury and Lashuel, 2006; Lin and Beal, 2006; Xiang et al., 2017; Morris et al., 2018). Recent data suggest that iron and heme metabolism dysfunction may also play a crucial role. The involvement of iron in neurodegeneration has been well documented (Levi and Finazzi, 2014; Mena et al., 2015; Stockwell et al., 2017) and iron chelation has been proposed as a therapeutic option (Ward et al., 2015; Ashraf et al., 2018). Moreover, several neurodegenerative disorders have been associated to dysfunction of heme metabolism, thus supporting a crucial role for heme in the pathogenesis of such diseases.

The neurotoxicity of heme is evident during intracerebral or subarachnoid hemorrhages. In these pathological conditions, large amount of hemoglobin and heme are released in the brain promoting oxidative stress, lipid peroxidation, inflammatory response and finally cell death (Righy et al., 2016). To counteract the toxic effects exerted by hemoglobin and heme, human cells have evolved several detoxification mechanisms. During brain injury, the plasma proteins Haptoglobin and Hemopexin have proved to be protective against free hemoglobin and heme as primary scavenger systems (Hahl et al., 2013; Ma et al., 2016). Moreover, the induction of the hemedegrading enzyme heme oygenase-1 (HMOX1) in the brain has been reported in intracerebral and subarachnoid hemorrhage, and other neurodegenerative conditions (Panahian et al., 1999; Wang and Doré, 2007; Guan et al., 2009; Gozzelino, 2016).

Hemoglobin-derived heme is not the only form of heme that the nervous system can handle. Endogenously synthesized heme is also deleterious if not properly managed. Heme synthesis

is an endogenous and essential process occurring in almost all tissues, including the nervous system (Smith et al., 2011; Gozzelino, 2016). Heme is endogenously synthesized to regulate a plethora of biological processes that may be particularly relevant in the nervous system. Among them, heme regulates energy production, ion channels, gene expression and microRNA processing (Smith et al., 2011). For these reasons, heme is essential for neuronal survival and differentiation. Recent data clearly indicate that the amount of intracellular heme available for regulatory functions (the so called "Heme Regulatory Pool" or "Labile Heme") is determined by the balance between heme synthesis, catabolism, import and export (Chiabrando et al., 2014b; Reddi and Hamza, 2016). The identification of a growing number of neurodegenerative disorders due to mutations in genes involved in heme metabolism strongly support the idea that the control of labile heme is crucial to avoid neurodegeneration. How heme drives the neurodegenerative process is still incompletely understood. However, it is becoming clear that heme neurotoxicity could be only partially explained by the release of toxic iron.

Here, we provide an overview of heme-related neurodegenerative disorders. Moreover, we discuss the potential mechanisms leading to heme neurotoxicity. We believe that the understanding of these mechanisms will be essential in the future to identify novel therapeutic targets.

# NEURODEGENERATIVE DISORDERS CAUSED BY HEME METABOLISM DYSFUNCTIONS

The physiological relevance to control labile heme in the nervous system is highlighted by the identification of neurodegenerative disease-causing mutations in genes involved in heme metabolism. Indeed, mutations in genes involved in heme biosynthesis are responsible for neuropathic Porphyrias and Friederich Ataxia; mutations in the heme importer FLVCR2 have been found in Fowler Syndrome; mutations in the heme exporter FLVCR1 cause Posterior Column Ataxia and Retinitis Pigmentosa, non-syndromic Retinitis Pigmentosa and Hereditary Sensory and Autonomic Neuropathies (**Figure 1**). However, the molecular mechanisms underlying heme neurotoxicity are still unclear. Below, we summarize the major clinical features and the molecular genetics of these rare disorders.

## Neurodegenerative Disorders Due to Defects in Heme Synthesis

Heme synthesis is a well characterized and ubiquitous process that involves eight enzymatic steps. Briefly, ALAS1 (δ-aminolevulinic acid synthase 1) catalyzes the condensation of succynil-CoA and glycine in the mitochondrial matrix, to form δ-aminolevulinic acid (ALA). ALA is transported into the cytosol where it is converted to coproporphyrinogen III through a series of enzymatic reactions. Then, coproporphyrinogen III is translocated back into mitochondria for the final steps of heme synthesis (Chiabrando et al., 2014a). Mutations in each gene involved in the heme biosynthetic pathway are responsible for a group of rare disorders collectively named porphyrias (Besur et al., 2014; Ramanujam and Anderson, 2015). The partial deficiency of these enzymes results in reduced heme synthesis and accumulation of toxic porphyrin precursors in multiple organ systems, including the skin, liver and nervous system. Here we focus on neuropathic porphyrias which are associated with neurologic manifestations. Furthermore, defects in the last steps of heme synthesis have been reported in Friederic Ataxia (Schoenfeld et al., 2005; Huang et al., 2009).

#### Neuropathic Porphyrias

The most common neurologic manifestations in neuropathic Porphyrias are a combination of autonomic and peripheral neuropathy. The autonomic neuropathy is characterized by abdominal pain, tachycardia, hypertension, constipation and nausea. The peripheral neuropathy is predominantly a motor axonal neuropathy leading to muscle pain and weakness. Sensory neuropathy is manifested as neuropathic pain and distal paraesthesiaes. The central nervous system (CNS) may also be affected, leading to psychosis, anxiety, depression and seizures. Porphyric neuropathies are characterized by relative quiescent phases followed by acute neurovisceral attacks involving severe abdominal pain, peripheral neuropathies and psychiatric disturbances. Such attacks are often exacerbated by external stimuli (drugs and hormones) that induce endogenous heme synthesis thus leading to an increase of toxic heme precursors (Albers and Fink, 2004; Simon and Herkes, 2011; Tracy and Dyck, 2014). Neuropathic Porphyrias include delta-aminolevulinate dehydratase deficiency, acute intermittent porphyria, hereditary coproporphyria and variegate porphyria.

Delta-aminolevulinate dehydratase deficiency (ALAD deficiency; OMIM: #612740) is the only type of neuropathic porphyria with an autosomal recessive mode of inheritance. ALAD deficiency is an extremely rare disorder with childhood onset and severe neurologic implications. ALAD deficiency is caused by mutations in the gene encoding the second enzyme in the heme biosynthetic pathway. These mutations cause almost complete lack of ALAD activity and patients excrete a large amount of ALA into urine (Ramanujam and Anderson, 2015).

Acute intermittent porphyria (AIP; OMIM: #176000) is an autosomal dominant disorder with incomplete penetrance. It is the most common type of neuropathic porphyria. AIP is caused by mutations in the hydroxymethylbilane synthase (HMBS) gene, resulting in the partial deficiency of porphobilinogen deaminase (PBGD). PBGD is the third enzyme in the heme biosynthetic pathway and its reduced activity leads to the accumulation and urine excretion of ALA and porphobilinogen (PBG) (Pischik and Kauppinen, 2015; Ramanujam and Anderson, 2015).

Hereditary coproporphyria (HCP; OMIM: #121300) is an autosomal dominant disorder with incomplete penetrance. HCP results from mutations in the gene encoding coproporphyrinogen oxidase (CPOX). CPOX represents the sixth enzyme in the heme biosynthetic pathway. Its partial deficiency

FIGURE 1 | Neurodegenerative disorders associated to heme metabolism dysfunction. Schematic representation of the main pathways involved in the regulation of intracellular labile heme: heme synthesis, heme incorporation into hemoproteins, heme degradation, heme export and import. The picture highlights rare neurodegenerative disorders directly associated to heme metabolism dysfunction: defective heme biosynthesis causes neuropathic porphyria and is observed in FRDA; defective heme export (FLVCR1 mutations) is responsible for PCARP, RP and HSAN. For simplicity, these disorders have been depicted near FLVCR1a; however, the specific contribution of FLVCR1a and FLVCR1b to these disorders still remains to be addressed. Finally, defective heme import (FLVCR2 mutations) leads to the Fowler syndrome. Alteration of heme incorporation into hemoproteins and heme degradation have not been directly associated to a specific neurodegenerative disorder. However, we cannot exclude a role for these pathways in neurodegeneration.

causes the accumulation of ALA, PBG and coproporphyrin III (Horner et al., 2013; Ramanujam and Anderson, 2015).

Variegate porphyria (VP; OMIM: #176200) is an autosomal dominant disorder with incomplete penetrance. VP is due to mutations in the gene encoding protoporphyrinogen oxidase (PPOX). PPOX catalyzes the oxidation of protoporphyrinogen IX to form protoporphyrin IX and its partial deficiency causes the elevation of plasma porphyrins (Horner et al., 2013; Ramanujam and Anderson, 2015).

The clinical presentation is generally milder in HCP and VP compared to AIP and ALAD deficiency.

The etiology of the neurological manifestations in neuropathic porphyrias remained undefined for long time. Nowadays, two major mechanisms have been proposed to contribute to the nerve failure: direct neurotoxicity of the porphyrin precursors and mitochondrial dysfunction (Lin et al., 2011).

#### Friederich Ataxia

Friederich Ataxia (FRDA; OMIM #229300) is a progressive neurodegenerative disease usually associated with cardiomyopathy and diabetes. Although a rare disorder, FRDA represents the most frequent type of inherited ataxia. FRDA is characterized by progressive gait and limb ataxia, dysarthria, areflexia, loss of vibratory and position sense, and a progressive motor weakness of central origin. Neurodegeneration first affects the dorsal root ganglia (DRG) but then involves spinal cord, peripheral nerves and cerebellum (Marmolino, 2011; Bürk, 2017).

FRDA is an autosomal recessive disorder caused by the abnormal expansion of the GAA triplet in the first intron of FRATAXIN (FXN) gene (Campuzano et al., 1996). FXN function remained elusive for long time. FXN is a mitochondrial iron chaperone involved in iron–sulfur (Fe–S) clusters and heme biosynthesis. Here we focus on the role of FXN on heme biosynthesis and readers are referred to more comprehensive reviews on the role of FXN in Fe–S clusters biosynthesis (Stemmler et al., 2010; Martelli and Puccio, 2014; Braymer and Lill, 2017; Rouault and Maio, 2017). FXN delivers iron to Ferrochelatase (FECH) (Yoon and Cowan, 2004), the enzyme responsible for the insertion of iron into protoporphyrin IX. Structural studies on the interaction between FXN and FECH shed light on the mechanism of iron delivery between them (Bencze et al., 2007; Söderberg et al., 2016). Evidences of defective heme biosynthesis have been reported in both cellular and mouse models of FXN deficiency as well as in FRDA patient-derived cells. Collectively these studies showed decreased expression of enzymes involved in the heme biosynthetic pathway (Schoenfeld et al., 2005; Huang et al., 2009), induction of HMOX1 (Huang et al., 2009), increased cellular protoporphyrin IX levels (Schoenfeld et al., 2005) and reduced heme content (Schoenfeld et al., 2005; Huang et al., 2009). Although the majority of these experiments have been performed in different cell and tissue types, it is conceivable that these findings can be translated to the nervous system. Fe-S clusters and heme biosynthesis are the two major iron consuming processes in mitochondria. Therefore, reduced iron utilization for Fe–S clusters and heme biosynthesis contributes to iron accumulation observed in FRDA (Huang et al., 2009). Moreover, mitochondrial iron accumulation in FRDA was reported to be due to heme-dependent upregulation of mitoferrin-2, a mitochondrial iron importer (Martelli et al., 2015). Both Fe–S clusters and heme are essential cofactors of the electron transport chain required for energy production. Indeed, FXN deficiency finally causes mitochondrial dysfunction, that actively contributes to the disease pathogenesis (Lodi et al., 1999; Martelli et al., 2015; Chiang et al., 2018).

Several yeast and mouse models of FXN deficiency have been generated (Perdomini et al., 2013). These models have been extremely useful to understand the molecular mechanisms of neurodegeneration associated to FXN loss and to test therapeutic approaches for the disorder.

#### Neurodegenerative Disorders Due to Defects in Heme Import

The import of heme inside the cell is achieved through multiple transporters: the heme responsive gene 1 (HRG1), the Heme carrier protein 1/Proton-coupled folate transporter (HCP1/PCFT) and the Feline leukemia virus subgroup C receptor 2 (FLVCR2).

The role of HRG1 in the import of heme is well established (Rajagopal et al., 2008; White et al., 2013). HRG1 is highly expressed in the brain and its transient downregulation in zebrafish leads to hydrocephalus and yolk tube malformations suggesting an important role for HRG1 in neurodevelopment (Rajagopal et al., 2008). Conversely, the substrate-specificity of HCP1 and FLVCR2 have been debated. HCP1 was initially identified as a heme importer (Shayeghi et al., 2005). Nevertheless, Qiu A. et al. reported that HCP1 is a folate transporter implicated in hereditary folate malabsorption (Qiu et al., 2006). However, evidences suggest that the heme import function of HCP1 could be relevant in some cell types or in particular physiologic or pathologic situations (Chiabrando et al., 2014b). Because of the low expression level of HCP1 in the brain, the role of HCP1 in the import of heme in the nervous system is likely marginal.

FLVCR2 is a member of the major facilitator superfamily (MFS) of transporters (Pao et al., 1998; Law et al., 2008) initially proposed as a calcium-chelate transporter (Meyer et al., 2010). Subsequent studies revealed that FLVCR2 is an importer of heme (Duffy et al., 2010). FLVCR2 is ubiquitously expressed. In the nervous system, FLVCR2 is widely expressed throughout the brain and spinal cord (Duffy et al., 2010; Meyer et al., 2010). However, FLVCR2 expression and subcellular localization still remain to be investigated in detail due to the lack of a specific antibody. Among all these heme importers, FLVCR2 is the only gene directly associated to a neurodegenerative disorder, the Fowler Syndrome. Therefore, the discussion below is focused exclusively on this disease.

#### Fowler Syndrome

Fowler syndrome, also known as Proliferative Vasculopathy, Hydrancephaly Hydrocephaly syndrome (PVHH; OMIM #225790), is a rare neurodegenerative disorder. The hallmark of the disease is the presence of proliferative glomerular vasculopathy in the CNS associated with severe hydrocephaly, ventriculomegaly, cortical thinning and hypoplastic cerebellum. Secondary features are hypokinesia and joint contractures (Meyer et al., 2010; Williams et al., 2010; Kvarnung et al., 2016). It is still unclear whether the proliferative vasculopathy is the primary or secondary event in the disease pathogenesis. Massive endothelial cells proliferation might be the first event leading to adjacent tissue damage, calcification and neuronal cells loss. Otherwise, proliferative vasculopathy might be a consequence of neurodegeneration (Meyer et al., 2010). The Fowler syndrome is characterized by early prenatal onset and is incompatible with life in most cases (Lalonde et al., 2010; Meyer et al., 2010; Thomas et al., 2010; Williams et al., 2010). Recently, two cases of survival beyond infancy have been reported. These patients were characterized by severe intellectual and neurologic disability (Kvarnung et al., 2016).

The Fowler syndrome is an autosomal recessive disorder caused by mutations in the FLVCR2 gene (Meyer et al., 2010). Different kinds of mutations in the FLVCR2 gene have been reported (Lalonde et al., 2010; Meyer et al., 2010; Thomas et al., 2010; Kvarnung et al., 2016), including missense and nonsense mutations, deletions and insertions. Homology modeling of FLVCR2 structure suggests that the missense mutations related with Fowler syndrome affect transmembrane domains that may modify the channel proper function or folding (Radio et al., 2018). It has been proposed that the mutations might lead to alteration of heme-iron metabolism. However, this is merely an

hypothesis since in vitro or in vivo evidences are still lacking. The generation of appropriate in vitro and in vivo model systems will be important to investigate the molecular mechanisms underlying Fowler syndrome.

# Neurodegenerative Disorders Due to Defects in Heme Export

The export of heme is mediated by two widely expressed proteins: the ATP binding cassette subfamily G member 2 (ABCG2) and the Feline Leukemia Virus Subgroup C Receptor 1 (FLVCR1).

Besides heme, ABCG2 is involved in the transport of a variety of substrates: urate, chemotherapeutics, antibiotics, xenobiotics and food metabolites (Krishnamurthy and Schuetz, 2006). In the nervous system, ABCG2 is mainly located on the luminal side of endothelial cells (ECs) (Krishnamurthy and Schuetz, 2006), suggesting an important role for ABCG2 in the bloodbrain-barrier to protect brain from drugs. Furthermore, ABCG2 has been implicated in brain protection following ischemic reperfusion injury (Shin et al., 2018) and in Alzheimer's disease (Xiong et al., 2009; Zeng et al., 2012; Shen et al., 2010). However, the relevance of ABCG2-mediated heme efflux to neurodegeneration still remains to be elucidated.

Feline Leukemia Virus Subgroup C Receptor 1 (FLVCR1) is a member of the MFS of transporters (Pao et al., 1998; Law et al., 2008), implicated in the transport of heme and other planar porphyrins (Yang et al., 2010). FLVCR1 gene encodes for two isoforms: FLVCR1a expressed at the plasma membrane and FLVCR1b in mitochondria (Chiabrando et al., 2012). FLVCR1 is ubiquitously expressed (Quigley et al., 2004; Keel et al., 2008; Chiabrando et al., 2012; Fiorito et al., 2014, 2015; Vinchi et al., 2014; Mercurio et al., 2015; Petrillo et al., 2017). In the nervous system, Flvcr1 mRNA has been detected in the mouse brain (neocortex, striatum, hippocampus, and cerebellum), posterior column of the spinal cord, retina and retinal pigment epithelium. The highest Flvcr1 mRNA levels have been found in the retina and spinal cord (Rajadhyaksha et al., 2010; Gnana-Prakasam et al., 2011). Unfortunately, no information is available concerning FLVCR1 protein expression levels and localization in neurons and glia, due to the lack of a reliable antibody against FLVCR1. Mutations in FLVCR1 gene have been reported in distinct disorders affecting the sensory nervous system: Posterior Column Ataxia and Retinitis Pigmentosa, Retinitis Pigmentosa and Hereditary Sensory and Autonomic Neuropathy, as reviewed below.

#### Posterior Column Ataxia and Retinitis Pigmentosa (PCARP)

PCARP (OMIM: #609033) is a rare neurodegenerative syndrome characterized by sensory ataxia and retinitis pigmentosa. Sensory ataxia is a consequence of the degeneration of the posterior columns of the spinal cord, resulting in loss of proprioception. Retinitis pigmentosa is due to the progressive degeneration of photoreceptors in the retina, leading to night blindness and progressive restriction of the visual field (Higgins et al., 1997; Rajadhyaksha et al., 2010). PCARP has been described for the first time in Higgins et al. (1997) as an autosomal recessive disorder associated with the AXPC1 locus (Higgins et al., 1999). In 2010, the analysis of three inbreed families of different ethnic origins revealed that mutations in the FLVCR1 gene were responsible for the disorder (Rajadhyaksha et al., 2010). In the majority of cases, PCARP is caused by homozygous mutations in the FLVCR1 gene (c.361A > G – c.721G > A – c.574T > C – c.1477G < C) (Rajadhyaksha et al., 2010; Ishiura et al., 2011). The identified mutations are missense mutations that affect highly conserved residues in potential transmembrane domains of the heme exporter. In vitro studies suggest that these mutations result in the mislocalization of FLVCR1 protein and in the loss of its heme export function. Compound heterozygous mutations in the FLVCR1 gene have been also reported in three siblings (c.1547G > A and c.1593 + 5\_ + 8delGTAA) (Shaibani et al., 2015).

#### Retinitis Pigmentosa (RP)

Non-syndromic retinitis pigmentosa (RP; OMIM: #268000) encompasses a group of genetically heterogeneous disorders characterized by the progressive neurodegeneration of photoreceptors. RP-associated genes are involved in different processes: phototransduction, retinal metabolism, tissue development and maintenance, cellular structure and splicing (Hartong et al., 2006). Recently, FLVCR1 mutations have been reported in some patients with retinitis pigmentosa (RP) without evidences of posterior column degeneration or cerebellar degeneration (Tiwari et al., 2016; Yusuf et al., 2018). Interestingly, all the patients carry a specific splice-site variant in FLVCR1 (c.1092 + 5G > A) in homozygosity or as compound heterozygosity. Functional studies showed that the c.1092 + 5G > A mutation interferes with the correct splicing of FLVCR1 resulting in the skipping of exon 4, the frame-shift deletion of 68bp and a premature stop codon (Tiwari et al., 2016; Yusuf et al., 2018).

#### Hereditary Sensory and Autonomic Neuropathy (HSAN)

HSANs are a group of neurodegenerative disorders of the peripheral nervous system mainly characterized by impaired nociception and autonomic dysfunction. The hallmark of the disease is the degeneration of sensory neurons, which transmit information about sensations such as pain, temperature and touch. The inability to experience pain causes unintentional self-injuries and chronic ulcerations. Soft tissue infections and osteomyelitis, often requiring amputations, are common. The disease is also associated to mild abnormalities of the autonomic nervous system: swallowing and feeding problems, apnea, gastroesophageal reflux and delayed development are variable features of the disorder (Verpoorten et al., 2006; Rotthier et al., 2012; Auer-Grumbach, 2013).

HSANs arise from mutations in genes that are crucial for distinct molecular pathways: sphingolipid-metabolism, membrane-shaping of organelles, neurotrophin action, DNA methylation, regulation of ion channels, endoplasmic reticulum turnover and axonal trafficking and heme metabolism (Rotthier et al., 2012; Khaminets et al., 2015; Chiabrando et al., 2016; Tanaka et al., 2016). Recently, mutations in the FLVCR1 gene have been reported in a subset of HSAN patients. Compound

heterozygous mutations in FLVCR1 (c.574T > C and c.610del – c.661C > T and c.1324dup) have been reported (Chiabrando et al., 2016). Functional studies in patient-derived cells showed that these mutations reduce heme export activity, enhance oxidative stress and increase the sensitivity to programmed cell death (Chiabrando et al., 2016). Finally, Castori et al. (2017) reported the homozygosity for a previously identified pc.661C > T variation in a sporadic case with a mixed phenotype of HSAN, PCARP and acute lymphocytic leukemia.

Due to the limited number of cases described to date, it is very difficult to understand whether FLVCR1-related PCARP, RP and HSAN are distinct disorders or rather a single clinical spectrum.

The finding of FLVCR1 mutations in PCARP, RP and HSAN indicates that heme is particularly required for the survival of specific subpopulation of neurons. However, the elucidation of the role of heme in these subpopulations is far to be clearly understood.

The available mouse models of FLVCR1 deficiency did not recapitulate the major disease features of HSAN, PCARP or RP. The targeted disruption of Flvcr1 gene causes embryonic lethality in two different mouse models. Mice lacking both Flvcr1a and Flvcr1b die in utero because of a block of erythropoiesis (Keel et al., 2008). Mice lacking only Flvcr1a die during embryonic development due to severe hemorrhages, edema and skeletal malformations (Chiabrando et al., 2012). The downregulation of FLVCR1 isoforms in zebrafish confirm the essential role of FLVCR1 isoforms in erythropoiesis and ECs maintenance (Mercurio et al., 2015).

The discrepancy between the phenotype of these animal models and the human diseases is likely due to different degree of Flvcr1a downregulation. Indeed, HSAN patients-derived cells still expressed FLVCR1a but are characterized by reduced heme export activity (Chiabrando et al., 2016). Moreover, treatment of patient-derived cells with Hemopexin, that facilitates heme export through FLVCR1a (Chiabrando et al., 2016), improved cell survival. These data clearly indicate that FLVCR1 mutations do not completely abrogate FLVCR1a function (Chiabrando et al., 2016). Reasonably, embryonic lethality is expected for patients carrying null mutations in the FLVCR1 gene.

The reason why the loss of an ubiquitously expressed heme exporter affects specific sensory modalities (vision, proprioception and/or nociception) is still unknown. The development of appropriate cellular and mouse models will help dissecting the role of FLVCR1 in these disorders.

#### HEME-REGULATED PROCESSES CRUCIAL FOR NEUROURODEGENERATION

In the paragraphs above, we summarized the most relevant neurodegenerative disorders associated to alterations in heme metabolism. However, the etiology of such diseases remains mostly undefined. Heme bears toxic properties due to its ability to promote iron mediated Fenton's reaction. This awareness contributed over the years to underestimate the importance of iron-independent mechanisms of heme toxicity in neuronal diseases. The misconception that heme neurotoxicity is due to iron release from protoporphyrin IX mainly contributed to the lack of interest in the topic. To stimulate research in the field, we will discuss the major heme-regulated pathways that might be relevant during neurodegeneration (**Figure 2**).

## Heme and Oxidative Stress: Implications for Neurodegenerative Disorders

Oxidative stress is a condition occurring when the production of reactive oxygen species (ROS) overcomes the endogenous antioxidant defenses. Accumulation of ROS causes the oxidative damage of cellular components, included lipids, proteins and nucleic acids. As a pro-oxidant molecule, heme can itself promote oxidative stress, eventually leading to cell death. In particular, due to its hydrophobic nature, free heme tends to aggregate and bind to cell membranes. Oxidation of membrane components leads to increased permeability and altered lipid organization, in turn promoting cell lysis and death (Schmitt et al., 1993). The redox activity of heme-derived iron also contributes to the cytotoxic effects of heme. Indeed, in aqueous phase heme becomes unstable and releases free iron, responsible for the formation of ROS via the Fenton's reaction (Kumar and Bandyopadhyay, 2005). The CNS is particularly susceptible to oxidative stress due to its high oxygen consumption rates and weak antioxidant systems. Consistently, oxidative stress is involved in the pathogenesis of several neurodegenerative disorders as Parkinson's disease, Alzheimer's disease, Huntington's disease and FRDA (Schulz et al., 2000; Calabrese et al., 2005; de Vries et al., 2008; Li et al., 2013; Puspita et al., 2017). Increased ROS levels have been observed also in HSAN patient-derived cells (Chiabrando et al., 2016; Castori et al., 2017). To counteract the toxic effects exerted by ROS accumulation, human cells have evolved several detoxification mechanisms. In particular, the primary defense systems against ROS consists of oxidant-inactivating enzymes, such as superoxide dismutase or HMOX1, and endogenous antioxidants, including glutathione and thioredoxin (Li et al., 2013). The transcriptional response to oxidative stress is mediated by a cis-regulatory element known as ARE (antioxidant response element) found in the promoters of genes codifying for antioxidant proteins. Nuclear factor E2-related factor 2 (Nrf2) is the major trans-acting factor responsible for the activation of gene transcription through its binding to the AREs (Nguyen et al., 2009). Notably, alterations in Nrf2 signaling pathway have been found in Parkinson's disease, Alzheimer's disease, and Huntington's disease (PMID: 18824091) as well as in a cardiac mouse model of FRDA (Anzovino et al., 2017). Several proteins and kinases modulate the Nrf2 pathway at various levels to ensure a tight control of its activity. Among these regulators, BTB domain and CNC homolog 1 (Bach1) acts as a repressor of Nrf2 activity by competing with Nrf2 for ARE binding sites. Notably, the activity and localization of Bach1 is regulated by heme (Sun et al., 2004; Suzuki et al., 2004; Marro et al., 2010), thus suggesting that alterations in heme homeostasis might affect the Nrf2-Bach1 axis and the cell ability to respond to the oxidative insult. This mechanism could be particularly relevant in the progression

of neurodegenerative diseases, which is indeed exacerbated by oxidative stress. Hence, oxidative stress might exert neurotoxicity also by impairing the ubiquitin-proteasome system, as found in Parkinson's disease (PD) where a decreased protein clearance rate due to elevated ROS results in the accumulation of alphasynuclein (Hashimoto et al., 1999; McNaught and Jenner, 2001).

# The Role of Heme in Cellular Proteostasis: Implications for Neurodegenerative Disorders

The maintenance of protein homeostasis, also called proteostasis, is essential for cell function and survival (Hartl et al., 2011; Morimoto and Cuervo, 2014; Kaushik and Cuervo, 2015). The constant control of cellular proteostasis requires the proper activity of two main cellular components (Kaushik and Cuervo, 2015). On one hand, the activity of chaperones is fundamental to correct protein misfolding eventually occurring during protein synthesis, assembly in three-dimensional and functional structures and trafficking. On the other hand, protein clearance mechanisms mediate the removal of irreversibly misfolded proteins. In particular, this second purpose is achieved most prominently through two molecular systems: the ubiquitinproteasome system (UPS) and the autophagy system. These systems are able to sense disturbances in the proteome and to be rapidly activated, thus exerting a constant protein quality control. In this way, the proteostasis network ensures the cell the ability to adapt to environmental changes and transitory stress. Reduced proteostasis capacity may lead to the accumulation of misfolded proteins, which form cytotoxic aggregates (Morimoto and Cuervo, 2014). Deficiencies in the machinery of protein homeostasis promote the onset and progression of several human pathologies, including neurodegenerative diseases (Hartl et al., 2011). For instance, deposits of aberrant proteins such as tau and α-synuclein, have been associated with dementia and Parkinson's disease. The use of proteasome inhibitors as Bortezomib in cancer therapy has been shown to induce peripheral neuropathy as a prominent side effect (Alé et al., 2014; Kaplan et al., 2017). Moreover, an impairment of mitochondrial turnover via selective autophagy has been found in Parkinson's disease (Andreux et al., 2013) and in sensory neuropathy (Khaminets et al., 2015). In addition to the above described

classical proteostasis network, which assures the quality control of de novo synthetized cytosolic proteins, organelle-specific systems can be activated in response to proteotoxic stress occurring in specific subcellular compartments (Kaushik and Cuervo, 2015). These signaling pathways include the unfolded protein response (UPR) of the endoplasmic reticulum (ER) and mitochondria. In particular, the presence of unfolded proteins in the ER promotes the activation of three ERresident transmembrane proteins: PERK (protein kinase RNA (PKR)-like ER kinase), ATF6 (activating transcription factor-6) and IRE1α (inositol-requiring enzyme-1α). The signaling cascade mediated by downstream effectors leads to an overall translational attenuation as well as enhanced transcription of ER chaperons. A similar mechanism has been described in mitochondria, which are particularly exposed to stressors that can trigger protein damage. Notably, alterations in the UPR machinery are involved in the pathogenesis of neurodegenerative diseases such as Alzheimer's disease, Parkinson's diseases, Huntington's disease and prion disease, which are indeed characterized by accumulation and aggregation of misfolded proteins (Scheper and Hoozemans, 2015; Xiang et al., 2017). Finally, mutations in genes associated with ER function have been described in neurodegenerative disorders as Charcot-Marie-Tooth disease (CMT), hereditary neuropathies with proximal dominant involvement (HMSN-P) and amyotrophic lateral sclerosis (Beetz et al., 2013; Tsai et al., 2014; Yagi et al., 2016; Murakami et al., 2017).

Growing evidences indicate that the alteration of intracellular heme levels affects proteostasis. For instance, heme directly binds and inhibits the proteasome machinery promoting the formation of perinuclear "aggresomes" of ubiquitinated proteins (Vallelian et al., 2015). In particular, Vallelian et al. (2015) analyzed changes in protein expression upon heme treatment in mouse embryo fibroblasts (MEF) isolated from wild type and mice lacking HMOX1 (Hmox1-null mice). The study of the proteome in Hmox1-null MEF by mass spectrometry revealed a specific signature characterized by a strong upregulation of networks related to the response to unfolded proteins. These data suggest that excessive intracellular heme can trigger proteostasis disruption. In particular, heme overload correlates with accumulation of ubiquitinated proteins, thus indicating a reduced protein degradation capacity in heme-stressed cells (Vallelian et al., 2015). Consistent with these previous observations, the formation of protein aggregates has been described in macrophages during heme stress (Vasconcellos et al., 2016). In particular, cell treatment with increasing concentration of heme leads to the formation of aggresome-like induced structures (ALIS), which are characterized by the presence of p62/SQSTM1 (sequestosome-1) and ubiquitinated proteins. Importantly, the pre-treatment with ROS scavengers completely abolishes heme-induced ALIS formation in bone marrowderived macrophages, thus indicating that ROS generation has a key role in this process (Vasconcellos et al., 2016). In addition, p62 is an inhibitor of the Nrf2 repressor Keap1 (Kelchlike ECH-associated protein 1). Therefore, the heme-induced sequestration of p62 in ALIS might affect the Nrf2-Keap1 axis (PMID: 24011591), interfering with the antioxidant response mediated by Nrf2-regulated genes. Enhanced ROS production and proteasome inhibition thus represent two different activities of heme, which turned out to be extremely cytotoxic in conditions of heme excess. These findings, taken together, point out a novel model of heme toxicity mediated by accumulation of damaged proteins and disruption of cellular proteostasis.

Another evidence of the proteotoxic effect mediated by heme comes from studies on ECs lacking FLVCR1a. The loss of the heme export way in ECs leads to the accumulation of endogenously synthetized heme, which in turn triggers programmed cell death via paraptosis and impairs the angiogenic process during embryonic development (Petrillo et al., 2017). Importantly, an altered swollen morphology of the ER as well as an increased expression of ER stress markers have been described in ECs isolated from embryos lacking Flvcr1a in endothelial cells (Petrillo et al., 2017). Interestingly, paraptosis has been described in neural development and neurodegeneration (e.g., amyotrophic lateral sclerosis and Huntington's disease), in the ischemic damage and in the context of retina pathophysiology (Lee et al., 2016).

Although specific studies in neuronal cells are still lacking, we proposed that similar mechanisms might be relevant in the nervous system. Therefore, the disruption of cell proteostasis might be an important mechanisms contributing to neurodegeneration in conditions of altered intracellular heme levels.

#### The Role of Heme in Mitochondria: Implications for Neurodegenerative Disorders

An additional mechanism through which heme could potentially participate to neurodegeneration is through the control of mitochondrial function. Mitochondrial dysfunction is a common pathogenetic mechanism of several neurodegenerative disorders. Particularly, the alteration of mitochondria by different mechanisms has been proposed as one of the underlying cause of Parkinson's disease, Alzheimer's disease, Huntington's disease and amyotrophic lateral sclerosis (Shi et al., 2010; Costa and Scorrano, 2012; Bennett et al., 2014; Cabezas-Opazo et al., 2015; Arun et al., 2016; Bose and Beal, 2016).

Heme is a crucial cofactor for cytochromes c and cytochromes in complexes II, III and IV of the mitochondrial electron transport chain (ETC) (Kim et al., 2012), thus modulation of heme metabolism could interfere with oxidative phosphorylation. Moreover, some of the steps of heme synthesis occur in mitochondria and initiate by the condensation of succynil-CoA with glycine. Succynil-CoA is an intermediate of the TCA cycle, so heme production belongs to the group of TCA cycle cataplerotic pathways (Frezza et al., 2011; Fukuda et al., 2017), a set of processes whose function is to avoid the accumulation of TCA cycle metabolites. Alterations in heme biosynthesis, consequently, could impair this system thus altering mitochondrial function. In addition, heme has been reported to directly or indirectly influence adenosine triphosphate (ATP) translocation among mitochondria and cytosol (Sabová et al., 1993; Giraud et al., 1998; Azuma et al.,

2008) and, therefore, modulation of heme homeostasis could profoundly affect the mitochondrial contribution to cell energy supply.

Supporting the role of heme in mitochondrial function, analyses on ECs demonstrated that alterations in heme metabolism, in addition to promote lipid peroxidation and activation of autophagy, induce mitophagy and apoptosis, indicating mitochondrial dysfunction (Higdon et al., 2012). Furthermore, we previously reported that loss of FLVCR1 impairs mitochondrial function in HeLa cells (Chiabrando et al., 2012) and mitochondrial morphology in human microvascular ECs (Petrillo et al., 2017).

Considering the role of heme in mitochondria, it is tempting to speculate that, by affecting mitochondrial function, the impairment of heme metabolism in neuronal cells could potentially initiate or sustain processes responsible for neuronal cells defects and death. Supporting this hypothesis, studies in the brain of HMBS−/− mice, a model of AIP, showed alterations in the activity of ETC complexes likely related to the compromised heme biosynthesis (Homedan et al., 2014). Moreover, it has been suggested that an energy failure may lead to a defective activity of the Na/K+ ATPase in the axon, leading to its degeneration (Lin et al., 2011). Since heme is required for ATP synthesis, we cannot exclude that heme metabolism could have a role in this context. Furthermore, defects in mitochondrial iron content, in the activity of the respiratory chain complexes and in ATP production have been observed in FRDA (Lodi et al., 1999). Although these defects in FRDA can be ascribed to alterations in Fe–S clusters trafficking, the contribution of heme to these processes cannot be excluded. Finally, in three cases of Fowler syndrome it was suggested the presence of a defect in complex III and IV of the ETC (Castro-Gago et al., 1999, 2001), directly indicating that mitochondrial defects due to alterations in heme trafficking could be the underlying cause of the neurological outcomes of this syndrome.

The contribution of mitochondrial dysfunction to PCARP, HSAN and RP has not been investigated. However, the implication of heme in these pathologies suggests the possibility that mitochondrial function could contribute to them. Notably, in sensory neurons the importance of heme in mitochondria could be related not only to the production of ATP, but also to mitochondrial dynamics (fusion, fission, motility). Mitochondrial dynamics are crucial for neurotransmission, synaptic maintenance and neuronal survival. Proper mitochondrial trafficking is particularly important in neurons as compared to other cell types due to their exceptional cellular morphology. Indeed, neurons extend their axons and dendrites for very long distances. Therefore, the neurons represent an extreme case of mitochondrial distribution: dysfunctions in mitochondrial distribution that are not dangerous for other cells, could be fatal for neuronal survival (Schwarz, 2013). Supporting this concept, compromised mitochondrial motility has been reported in Alzheimer's disease (Cabezas-Opazo et al., 2015), Parkinson's disease (Yang et al., 2008; Carelli et al., 2015), Huntington's disease (Shirendeb et al., 2011) and in amyotrophic lateral sclerosis (De Vos et al., 2007; Shi et al., 2010).

In the case of human peripheral nerves or corticospinal tracts, axons can extend up to a meter, so the population of sensory neurons could be highly sensitive to defects in mitochondrial dynamics. Modulation of the cellular labile iron pool and mitochondrial iron content have been observed in FXN deficiency in association to alterations in both heme biosynthesis and OPA1 (mitochondrial dynamin like GTPase) mediated mitochondrial fusion (Martelli et al., 2015). Although a causative link among these phenomena has not been proposed in FRDA, it is tempting to speculate that the modulation of iron and heme homeostasis could promote alterations in mitochondrial dynamics, with deleterious consequences in this pathology and, more likely, in neurological disorders affecting sensory neurons.

#### Heme Regulates Ion Channels: Implications for Neurodegenerative Disorders

Emerging evidences suggest that labile heme is a signaling molecule that impact on the function of various ion channels involved in neuron excitability. Among them, voltage-gated K+ channels are a large family of K+ channels that open on membrane depolarization and contribute to maintain the resting potential and to determine the shape and frequency of action potentials in excitable cells (Sahoo et al., 2014). Heme directly binds to the large-conductance calcium-dependent Slo1 BK channels and Kv1.4 A-type K+ channels. The high affinity binding of heme to Slo1 BK channels decreases the frequency of channel opening leading to an inhibition of transmembrane K+ currents (Tang et al., 2003). Its binding to the N-terminal domain of the Kv1.4 channels inhibits the fast inactivation of the channel, thus reducing cellular excitability (Sahoo et al., 2013).

The activity of several K+ channels is modulated by carbon monoxide (CO) (Tang et al., 2004) and ROS (Sahoo et al., 2014; Peers and Boyle, 2015). Heme may also indirectly regulate the activity of several K+ channels through the generation of CO and ROS. Indeed, the degradation of heme by HO1 produces iron, biliverdin and CO. Moreover, the released iron is responsible for the formation of ROS via the Fenton's reaction.

An alteration of the activity, function or expression of diverse K+ channels have been reported in Alzheimer's disease, Parkinson's disease, Huntington's disease and ataxias (Peers and Boyle, 2015). As mentioned before, increased ROS is common to all these neurodegenerative diseases. Oxidative modulation of diverse K+ channels has been proposed as a pathogenetic mechansim leading to neurodegeneration. Finally, several studies highlight a crucial role of K+ channels in nociception (Busserolles et al., 2016). The identification of FLVCR1 mutations in patients affected by HSAN (Chiabrando et al., 2016; Castori et al., 2017) suggests that the alteration of K+ channels activity may explain sensory defects in these patients. Considering all these data, it is possible to speculate that the alteration of labile heme may affect the activity of K+ channels, thus contributing to the neurodegenerative process.

#### CONCLUDING REMARKS

fnins-12-00712 October 5, 2018 Time: 19:9 # 10

The recent identification of several neurodegenerative disorders due to mutations in genes involved in heme metabolism highlights a previously unappreciated role for heme in neurodegeneration. Here, we discussed the potential mechanisms leading to heme neurotoxicity, thus suggesting future directions for investigation. The understanding of such mechanisms might be relevant for several neurodegenerative disorders. Indeed, alteration of heme metabolism have been observed also in Alzheimer's disease, Parkinson's disease and other aging-related neurodegenerative disorders (Atamna and Frey, 2004; Atamna, 2006; Scherzer et al., 2008; Dwyer et al., 2009; Smith et al., 2011; Hayden et al., 2015; Sampaio et al., 2017; Santiago and Potashkin, 2017; Fiorito et al., 2018). In the future, the comprehensive

#### REFERENCES


understanding of the molecular mechanisms linking heme to neurodegeneration will be potentially important for therapeutic purposes.

#### AUTHOR CONTRIBUTIONS

DC and VF conceptualized the study and wrote the manuscript. SP wrote the manuscript. ET wrote and reviewed the manuscript.

# FUNDING

This research was funded by University of Torino (Bando Ricerca Locale ex-60%).


erythroid differentiation. J. Clin. Invest. 122, 4569–4579. doi: 10.1172/JCI6 2422



pathophysiological role and translational implications. Mol. Neurobiol. doi: 10.1007/s12035-018-1028-6 [Epub ahead of print].




**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 Chiabrando, Fiorito, Petrillo and Tolosano. 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 Dual Role of Hepcidin in Brain Iron Load and Inflammation

Driton Vela\*

Department of Physiology, Faculty of Medicine, University of Pristina, Pristina, Kosovo

Hepcidin is the major regulator of systemic iron metabolism, while the role of this peptide in the brain has just recently been elucidated. Studies suggest a dual role of hepcidin in neuronal iron load and inflammation. This is important since neuronal iron load and inflammation are pathophysiological processes frequently associated with neurodegeneration. Furthermore, manipulation of hepcidin activity has recently been used to recover neuronal damage due to brain inflammation in animal models and cultured cells. Therefore, understanding the mechanistic insights of hepcidin action in the brain is important to uncover its role in treating neuronal damage in neurodegenerative diseases.

Keywords: hepcidin, brain, neurodegenerative diseases, glial cells, inflammation, iron load

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Torben Moos, Aalborg University, Denmark Andrzej Friedman, Medical University of Warsaw, Poland

> \*Correspondence: Driton Vela driton.vela@uni-pr.edu

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 25 July 2018 Accepted: 26 September 2018 Published: 15 October 2018

#### Citation:

Vela D (2018) The Dual Role of Hepcidin in Brain Iron Load and Inflammation. Front. Neurosci. 12:740. doi: 10.3389/fnins.2018.00740

#### INTRODUCTION

Iron is a crucial element for maintaining brain homeostasis. It is used by brain cells to synthesize myelin, produce neurotransmitters, and realize a wide range of important cellular biochemical reactions (Ward et al., 2014). But, iron excess is toxic for brain cells. This is observed during brain hemorrhage, where inflammatory signaling is responsible for the increase in brain iron load, which is associated with oxidative damage and cognitive decline (Wu et al., 2003; Ward et al., 2014; Garton et al., 2016). Furthermore, iron dysregulation has been proposed as a pathogenic factor in neurodegenerative diseases as well (Ward et al., 2014). Recent research suggests that brain iron dysmetabolism can be tackled via hepcidin manipulation, since local hepcidin production in the brain has been shown to affect cellular iron transport (Urrutia et al., 2013; Du et al., 2015; Gong et al., 2016; Xiong et al., 2016; Zhou et al., 2017). But, it seems that the presence or absence of inflammation dictates the deleterious vs. beneficial effects of hepcidin in the brain. Therefore, unraveling mechanistic aspects of the dual nature of hepcidin in the brain is important to understand its therapeutic potential.

#### REGULATION OF SYSTEMIC HEPCIDIN AND ITS ACTIONS

Most of systemic hepcidin is produced in hepatocytes via paracrine signaling from liver sinusoidal endothelial cells (LSECs) (Canali et al., 2017). LSECs respond to iron-mediated pathways by producing bone morphogenetic protein 6 (BMP6), which then activates BMP receptor (BMPR) in hepatocytes (Steinbicker et al., 2011). Activated BMPR induces s-mothers against decapentaplegic (SMAD) protein phosphorylation. Phosphorylated SMADs migrate to nucleus where they cause hepcidin upregulation (Casanovas et al., 2009). Although iron-mediated signals act mainly through BMP/SMAD pathway, other pathways are also involved, like mediation through transferrin receptor 2 (TFR2) and hemochromatosis protein (HFE) (Fleming, 2009). The second most studied

**38**

stimulator of hepcidin expression is inflammation. It induces hepcidin expression via interleukin-6/janus kinase 2/signal transducer and activator of transcription 3 (IL-6/JAK2/STAT3) pathway (Pietrangelo et al., 2007). This pathway does not dominate hepcidin expression in basal conditions, but during high levels of inflammation it can override hepcidin regulation by iron signaling. In addition to these pathways, hepcidin is controlled by negative regulation, which includes signaling via erythroferrone (ERFE), heparin, vitamin D (Poli et al., 2011; Bacchetta et al., 2014; Kautz et al., 2014). Tight regulation of hepcidin production ensures a direct control of iron metabolism. This is attributed to the ability of hepcidin to induce ferroportin (FPN) degradation in target cells (Ramey et al., 2010; Zhang et al., 2011). FPN is known as the major cellular iron exporter, which is the main target protein of hepcidin.

#### REGULATION OF BRAIN IRON METABOLISM

Iron homeostasis in the brain must be finely regulated to serve the purposes of brain cells without damaging cell structures. There are different mechanisms that allow for this to occur; first, blood brain barrier (BBB) is a natural barrier that does not allow passing of iron with ease, which means that iron has to be transported transcellularly via TFR (Bien-Ly et al., 2014; McCarthy and Kosman, 2015a). The iron-TFR complex undergoes endocytosis, while iron gets released intracellularly through divalent metal transporter 1 (DMT1) (Ward et al., 2014). Iron exits from brain cells via FPN (Ward et al., 2014). Higher density of FPN in neurons compared to other brain cells shows why FPN downregulation causes more marked cellular iron load in neurons (Moos and Rosengren Nielsen, 2006; Boserup et al., 2011; Urrutia et al., 2013). Studies suggest that astrocytes encircling brain microvascular endothelial cells (BMVECs) influence their FPN activity by releasing local hepcidin, but also by secreting ferroxidases [including amyloid-β precursor protein (APP)] to stabilize FPN (McCarthy and Kosman, 2014, 2015a,b). In any case, iron released into brain parenchyma enters neuronal cells via TFR1, transient receptor potential canonical (TRPC) channel, DMT1 and newly discovered players, such as Zip8 and Steap2, with DMT1 role being more prominent during pathophysiological processes (Ji and Kosman, 2015; Skjørringe et al., 2015).

#### ROLE OF SYSTEMIC HEPCIDIN IN BRAIN HOMEOSTASIS

It is believed that systemic hepcidin can cross blood barriers such as BBB (Xiong et al., 2016). Hepcidin is a member of the big family of antimicrobial peptides (AMPs), which have been shown to cross BBB via different pathways (Welling et al., 2015). This argument is further enforced by the ability of a specific group of AMPs named defensins (which includes hepcidin) to cross intact BBB rather easily (Schluesener and Meyermann, 1995). On the other hand, the idea that hepcidin passes intact BBB has been proposed by Raha-Chowdhury et al. (2015). Authors base their conclusion on their observation that hepcidin expression was seen in nearly all cellular layers of BBB, and also due to the fact that protein expression of hepcidin is higher and more robust than its mRNA expression, which has also been corroborated by other studies (Zechel et al., 2006). The idea that hepcidin could pass through intact BBB has been shared by other authors as well (Myhre et al., 2013; Hofer and Perry, 2016). It seems that the amphipathic and cationic chemical structure of defensins in general and hepcidin in particular, might make them potential candidates for BBB pass via transcytosis (Kumagai et al., 1987; Rousselle et al., 2000; Hunter et al., 2002), though this property of hepcidin has still not been conclusively shown in intact BBB. Still, hepcidin passage through BBB is significantly increased with its disruption, and it is observed even with larger peptides than hepcidin (Ostergaard et al., 2009).

It has to be mentioned that models with liver hepcidin knockout do not result in significant brain iron dysmetabolism (Xiong et al., 2016). This would be the reason why we do not observe significant brain iron overload in hemochromatosis (Rutgers et al., 2007). This means that autonomous mechanisms in the brain are responsible for keeping iron metabolism in check. These mechanisms are mostly realized through the activity of glial cells, which are capable of buffering excess iron (Pelizzoni et al., 2013; Codazzi et al., 2015; McCarthy et al., 2018). Astrocytes are especially active during iron transport through BBB (McCarthy and Kosman, 2015b), while microglia are responsible for the control of parenchymal brain iron transport (McCarthy et al., 2018). Microglia are important "managers" of the response to brain damage due to their first-time reaction and commanding role in this respect (Shinozaki et al., 2017). On the other hand, astrocytes are able to transport the much needed iron for neuronal needs as well, which is why conditional deletion of FPN in astrocytes damages the process of myelination (Schulz et al., 2012). FPN is also important for the iron transport through BMVEC, which is why specific BMVEC FPN mutations cause transient brain iron deficiency. This process is reversible due to upregulation of iron import proteins in BMVEC (Iacovelli et al., 2009). This means that significant injury to brain cells from iron overload will occur if both, cellular iron import and export proteins are affected. Cellular iron overload seems to be more detrimental to neurons due to their lower iron buffering capacity (Bishop et al., 2011). But, studies suggest that glial cell beneficial phenotype can also be dysregulated in the presence of high ironload (Song et al., 2018).

Animal models suggest that liver hepcidin has important consequences in brain pathologies, such as brain hemorrhage/ischemia. During brain hemorrhage/ischemia liver hepcidin is upregulated, which is accompanied with increased global brain iron content after peak levels of hepcidin (Xiong et al., 2016). Importantly, hepcidin knockout decreases brain iron content in models with brain hemorrhage, but not in physiological conditions (Tan et al., 2016; Xiong et al., 2016). Xiong et al. (2016) study found a significantly lower brain iron content in hepcidin knockout rodents compared to normal rodents during intracerebral hemorrhage (ICH). In addition, brain markers of oxidative stress and brain water content were

lower in hepcidin knockout rodents compared to the brain of normal rodents during ICH. Finally, hepcidin knockout rodents performed better than their normal counterparts during cognitive tests. Together these data suggest that the presence of hepcidin impacts brain iron content and function during ICH. Data from human studies seem to complement findings from animal studies because they suggest that systemic hepcidin could contribute to increased brain damage during ICH. This observation is based on the existence of the correlation between serum levels of hepcidin and clinical scores that predict poor outcome in these patients (Xiong et al., 2015). This is the reason why Xiong et al. (2016) conclude that hepcidin affects brain iron efflux during ICH, but how could this action occur is unknown. It is interesting to notice that hepcidin has opposite effects on brain iron content during brain iron-overload that occurs without inflammation (Du et al., 2015; Gong et al., 2016). Du et al. (2015) study showed that the use ad-hepcidin after injection of holotransferrin into left carotid artery results in reduced brain iron content compared to rodents not treated with ad-hepcidin (Du et al., 2015). In vitro results from this study showed that ad-hepcidin and hepcidin peptide reduced brain iron uptake by decreasing TFR1, DMT1, and FPN expression in brain endothelial cells. So, how could hepcidin presence during ICH cause opposite results? During ICH brain iron is increased even many weeks after the insult (Dietrich and Bradley, 1988; Wu et al., 2003; Hua et al., 2006; Liu et al., 2016). The increase in brain iron content in this setting might occur because during BBB disruption iron leakage to brain parenchyma cannot be accompanied with reactive iron efflux since hepcidin blocks excess iron from getting out of brain cells. In this respect, hepcidin further enhances the effect of inflammatory signals on cellular iron accumulation (Sansing et al., 2011; Urrutia et al., 2013; Zhou et al., 2014; Xiong et al., 2016; Goldstein et al., 2017). This occurs because during brain inflammation, cytokines, and hepcidin have agonistic effects in suppressing cellular iron efflux (Urrutia et al., 2013; Xiong et al., 2016; Zhao Y. et al., 2018). In addition, inflammatory signaling overrides the blocking effect of hepcidin on cellular iron uptake which creates an ideal environment that promotes brain iron overload by both, inflammatory signals and hepcidin (Urrutia et al., 2013; Du et al., 2015; Gong et al., 2016; Xiong et al., 2016; Zhou et al., 2017; Zhao Y. et al., 2018).

#### BRAIN HEPCIDIN EXPRESSION AND MECHANISMS OF REGULATION

Local hepcidin expression in the brain has been an object of investigation of different studies. In physiological conditions hepcidin expression in brain structures and cells (neurons, glial cells, endothelial cells, epithelial cells of choroid plexus) has been consistently observed by in vivo studies in humans and rodents, albeit in low levels (Krause et al., 2000; Pigeon et al., 2001; Zechel et al., 2006; Wang et al., 2008, 2010; Hänninen et al., 2009; Raha et al., 2013; Wei et al., 2014; Farajdokht et al., 2015; Raha-Chowdhury et al., 2015; Graf et al., 2016; Li Y. et al., 2016; Pan et al., 2016; Tan et al., 2016; Lu et al., 2017; You et al., 2017; Zhang et al., 2017). Data from human and animal studies suggest that local hepcidin is more robustly expressed in pathophysiological states (Sun et al., 2012; Urrutia et al., 2013; Tan et al., 2016; Xiong et al., 2016; You et al., 2017; Zhang et al., 2017). Similar to other cells, hepcidin main target protein in brain cells is FPN, but also iron import proteins (Sun et al., 2012; Urrutia et al., 2013; Tan et al., 2016; Xiong et al., 2016; You et al., 2017; Zhang et al., 2017).

Hepcidin upregulation can be elicited by acute iron load in astrocytes and microglia, while in neurons this response seems to become more evident with higher doses of iron supplementation (Sun et al., 2012; Urrutia et al., 2013; Simpson et al., 2015). On the other hand, according to Burkhart et al. (2016) hepcidin expression in the brain stem is scarce and probably has no effect during physiological conditions. It has to be mentioned that Burkhart et al. (2016) model of study has many differences compared to other authors. Burkhart et al. (2016) study did not include a bichamber model of BBB as compared to others, which imitates the cellular environment of brain endothelial cells and adjacent astrocytes (McCarthy and Kosman, 2014; Simpson et al., 2015). Still, even authors of this study acknowledge that hepcidin has an important role in brain iron homeostasis during iron-overload and conditions associated with BBB disruption (Burkhart et al., 2016). Unfortunately, compared to liver hepcidin, we still do not know the detailed mechanistic aspects of iron-induced regulation of hepcidin expression in brain cells. It is interesting to notice that results from studies with BMP6 pretreatment experiments are similar to those that have used pretreatment with hepcidin in terms of protection of brain cells from oxidative stress. But, whether BMP6 can control hepcidin expression in response to ironload in brain cells is still not known (Wang et al., 2001; Urrutia et al., 2017).

On the other hand, there is quite robust data concerning hepcidin regulation in the brain via an inflammatory cascade which involves lipopolysaccharide (LPS), toll-like receptor 4 (TLR4), IL-6, STAT3 molecules (**Figure 1**). The magnitude of cell-specific response to LPS is highest in glial cells and lowest in neurons (Urrutia et al., 2013; You et al., 2017). Furthermore, TLR4 as the main membrane receptor for LPS is more consistently present in microglia (Trotta et al., 2014). In addition, studies suggest that IL-6 (which is a downstream molecule induced by LPS activity) is the mediator of hepcidin upregulation in the brain during inflammation (Zhang et al., 2017). Knockdown of IL-6 during brain inflammation reduces significantly the extent of STAT3 phosphorylation and resultant hepcidin upregulation and FPN downregulation (Zhang et al., 2017). Furthermore, IL-6 activity is associated with overexpression of DMT1, though this effect is not as robust as the consequences in reduction of iron export through FPN (Zhang et al., 2017). Urrutia et al. (2013) and Zhang et al. (2017) have revealed that although the effect of IL-6 in brain hepcidin expression is robust, it is not the only cytokine through which LPS regulates hepcidin. Intriguingly, FPN downregulation during inflammation is not related only to STAT3 or hepcidin levels, but due to an unknown IL-6 dependent pathway (Zhang et al., 2017). This finding is in-line with the suggestion that inflammatory signaling independent of hepcidin expression is the main culprit

behind downregulation of FPN in brain cells. Cellular analysis shows that in an inflammatory setting the most evident changes in cellular iron load are observed in neurons and microglia, but less in astrocytes (Urrutia et al., 2013). This is probably due to the ability of astrocytes to facilitate iron transport in and out of their cellular environment (Xu H. et al., 2017). This would make sense if we take into account the general role of astrocytes as supporting cells that mediate transport and metabolism of different molecules.

Recently, two other pathways have been observed as inducers of brain hepcidin expression. This includes the canonical BMP/SMAD pathway, which is the classical pathway of hepcidin regulation in liver, and CCAAT/enhancer-binding protein (C/EBP) homologous protein (CHOP), which has also been involved in regulating liver hepcidin expression during endoplasmic reticulum (ER) stress (Oliveira et al., 2009). BMP/SMAD pathway has been shown to be involved in hepcidin expression in activated microglia in response to LPS, albeit in a reduced manner compared to IL-6/STAT3 pathway (Qian et al., 2014; Shin et al., 2018). On the other hand, CHOP pathway is involved in hepcidin expression in neurons, especially during brain hemorrhage. Similarly to liver, data suggest that CHOP regulates hepcidin expression in neurons via C/EBPα (Zhao J. et al., 2018). In this way, the picture of hepcidin regulation in the brain is unfolding, which will help in choosing the appropriate therapeutic strategies in different brain pathologies. Indeed, experiments with brain hepcidin knockout have resulted in decreased brain iron content in models with brain hemorrhage, while BMP/SMAD pathway suppression has recently been used in animal models to treat iron overload caused by hepcidin upregulation during brain ischemia (Ding et al., 2011; Tan et al., 2016; Shin et al., 2018). Similarly, CHOP silencing with siRNA reduces brain edema and improves neuronal function (Zhao J. et al., 2018). Also, TLR4 knockdown prevents brain iron accumulation during inflammation (Xiong et al., 2016).

#### DUAL ROLE OF HEPCIDIN IN BRAIN PATHOLOGIES; IMPLICATIONS FOR ITS THERAPEUTIC POTENTIAL

It is evident that local and/or systemic hepcidin is induced during brain inflammation and iron load (Urrutia et al., 2013; Xiong et al., 2016). In vivo experiments during brain

inflammation show that the use of iron chelation protects from brain iron overload, reduces microglial activation and improves cognitive functions in rodents by reducing levels of hepcidin and increasing levels of FPN in hippocampus (Li Y. et al., 2016; Pan et al., 2016). In addition, maternal diet and calorie restriction have been proposed as factors that offer neuroprotection by reducing brain iron load through suppression of brain hepcidin (Wei et al., 2014; Graf et al., 2016). But, it is interesting to notice that accumulating data suggest that hepcidin role in neuronal iron load and inflammation is ambiguous; hepcidin can protect from iron load, but also can cause iron load during inflammation. The reason for this duality seems to occur due to the timing of hepcidin action; studies show that pretreatment with hepcidin protects brain cells from iron load, while hepcidin induction during inflammation aggravates iron load (Xiong et al., 2016; Urrutia et al., 2017; Zhao Y. et al., 2018). Inflammation in the brain is a strong stimulus for the increase in iron import, which increases iron load. But, inflammation has a double effect on iron export; it blocks iron export via FPN inhibition through hepcidin-independent and hepcidin-dependent mechanisms in neurons and glial cells (Urrutia et al., 2013; Zhang et al., 2017). The double effect of inflammation in blocking iron export enhances the cellular iron load in the brain even further. Previous studies have indicated that LPS can indeed affect FPN expression independently of cytokines, though cellular specific differences do exist (Liu et al., 2005; Deschemin and Vaulont, 2013). Cellular specific reactions to inflammatory stimuli have revealed that astrocytes and microglia (compared to neurons) react more strongly to LPS by increasing their hepcidin expression and decreasing FPN expression (Urrutia et al., 2013; Ma et al., 2018). Isolated neurons do not show signs of robust increase in hepcidin levels and FPN downregulation in reaction to inflammatory stimuli (Ma et al., 2018; Puy et al., 2018). But, when neurons are stimulated with high dose of inflammatory cytokines their hepcidin expression is increased significantly (Urrutia et al., 2013). This environment is created when neurons are co-cultured with microglia and astrocytes (You et al., 2017). In this scenario, prime responders to LPS are microglia, which through IL-6 act on astrocytes to increase hepcidin expression. Then, hepcidin acts on neurons in a paracrine manner to reduce FPN expression (You et al., 2017). It has to be mentioned that microglia can also affect neuronal hepcidin expression via LPS (Qian et al., 2014; Li Y. et al., 2016). So, the present picture suggests that during inflammation astrocytes are responsible for directing iron influx from plasma into the brain, where this iron is deposited in neurons via microglial and astrocyte activity, which can increase neuronal oxidative damage and even cause neurodegeneration.

On the other hand, hepcidin pretreatment has the potential to relieve the damage caused by inflammatory signaling. This happens because hepcidin pretreatment downregulates IL-6 and tumor necrosis α (TNFα) expression in astrocytes and microglia (Urrutia et al., 2017) (**Figure 2**). Treatment with hepcidin in these conditions protects neurons from oxidative stress (Urrutia et al., 2017). In addition, treatment with ad-hepcidin in non-inflammatory conditions protects neurons from iron load (Du et al., 2015; Gong et al., 2016; Zhou et al., 2017). The protective role of hepcidin occurs due to its effect in inhibiting iron transport through BMVEC and through its effects in reducing iron import in neuronal cells (Du et al., 2015; Gong et al., 2016; Zhou et al., 2017). These data show that timing of hepcidin treatment and presence of inflammation dramatically influences hepcidin actions in brain cells. It seems that hepcidin pretreatment primes the brain cellular milieu against the stimulative effects of inflammatory signals in increasing brain iron load (Urrutia et al., 2017). Previous studies with cultured macrophages have revealed that hepcidin-FPN interaction induces suppressor of cytokine signaling 3 (SOCS3) activity, which as its name suggest, is a suppressor of cytokine signaling during inflammation (De Domenico et al., 2010). Also, human studies have shown that levels of hepcidin are related with the inflammatory response (Burté et al., 2013). It would be interesting for future studies to reveal if beneficial effects of pretreatment with hepcidin can be sustained with increasing levels of inflammation.

#### ROLE OF INFLAMMATION AND IRON LOAD IN NEURODEGENERATIVE DISEASES

Pathogenesis of neurodegenerative diseases is complex and still not entirely understood. But, in recent years accumulating evidence are linking neurodegenerative processes with iron metabolism and neuroinflammation. Iron dysmetabolism is an early feature of Alzheimer's disease (AD) and in isolated nuclei in Parkinson's disease (PD) (Nunomura et al., 2001; Smith et al., 2010; Guan et al., 2017). In AD, levels of ferritin in cerebrospinal fluid (CSF) are predictors of worsening cognitive performance and are also correlated with apolipoprotein E (ApoE) CSF levels and AD risk allele ε4 (Ayton et al., 2015). Similarly, higher iron deposition in substantia nigra (SN) of AD patients makes them more prone to develop PD (Brar et al., 2009). Use of compounds with moderate iron-binding potency reduces levels of α-synclein and protects neurons from oxidative damage in PD (Finkelstein et al., 2017). This effect is realized by restoring the function of FPN as a cellular iron exporter (Finkelstein et al., 2017). Iron load as a risk factor in PD has been further confirmed in a meta-analysis, which has identified the presence of iron overload in SN of PD patients (Wang et al., 2016). It seems that in PD iron overload is generally more present in advanced stages of the disease compared to AD (Wang et al., 2016; Guan et al., 2017). In patients with mild and moderate PD there is also evidence of systemic perturbations of iron metabolism, with ferritin and malondialdehyde (MDA) serum levels (marker of oxidative damage) serving as significant biomarkers of PD (De Farias et al., 2017). On the other hand, it has to be mentioned that the most affected brain structure in PD, that is SN, is characterized with iron overload and significant damage in early stages of the disease (Ziegler et al., 2013; Guan et al., 2017). This would imply that iron dysmetabolism in SN in PD could occur due to a similar pathogenic mechanism observed in AD.

It is interesting to notice that APP dysfunction (which is a hallmark of AD) induces dementia-like symptoms in PD (Lei et al., 2012). Importantly, APP dysfunction in AD results in its inability to stabilize the function of FPN as a cellular iron exporter, which would affect intracellular iron load (Duce et al., 2010).

Similar to iron load, inflammation is also being recognized as an important pathogenic marker of neurodegenerative diseases. It is observed early in AD and PD, often bearing evidence of peripheral (intestinal) aberrant response to different stimuli (Qin et al., 2016; Houser and Tansey, 2017; Raj et al., 2017; Gelders et al., 2018; Gurel et al., 2018; King et al., 2018; Steeland et al., 2018). This means that in AD and PD the pathogenic process is initiated and "fueled" by immunologic disturbances occurring in the gastrointestinal tract. In relation to this scenario are data coming from studies with the TNF receptor (TNFR), which have caught attention for their importance in neurodegenerative diseases. TNF and Aβ-induced inflammation is significantly reduced via TNFR1 loss (Steeland et al., 2018). It seems that TNF and Aβ oligomers act as agonists on TNFR1 to activate microglial cells. In addition, TNFR1 loss protects brain cells from propagation of peripheral inflammatory reaction by restoring blood-CSF barrier functionality (Steeland et al., 2018). In addition, direct evidence from human studies with inflammatory bowel disease (IBD) suggests that early TNF inhibition dramatically reduces the incidence of PD (Peter et al., 2018). TNF involvement in PD is further strengthened by the increased risk of early onset of PD in patients with higher expression of TNF (Lindenau et al., 2017). On the other hand, role of inflammation in AD and PD is characterized with subtle but important differences between these conditions; in vivo models with mice have shown that Aβ oligomer accumulation in AD is related to TLR4 activity, which is in contrast with α-synuclein accumulation in PD, where the role of TLR4 is low or inexistent (Noelker et al., 2013; Forloni et al., 2016; Calvo-Rodríguez et al., 2017). Also, microglial activation is somewhat more prominent with AD related oligomers than with PD related oligomers (Forloni et al., 2016). Furthermore, in PD, microglial activity is accompanied with increased number of lymphocytes in damaged brain nuclei (Wang et al., 2015). These data suggest distinct inflammatory routes of activation by pathogenic oligomers in AD and PD.

Alzheimer's disease and PD are diseases that increase in prevalence with age. It is believed that the ability of microglia and astrocytes to secure the homeostatic equilibrium for neurons is reduced with aging. The homeostatic disequilibrium might be a result of the genetic factors that are related with increased activity of inflammatory signals, decreased ability of microglia to clear oligomer accumulation and other factors (Heppner et al., 2015; Lindenau et al., 2017; Hansen et al., 2018). "Pathogenic" microglia release cytokines which are responsible for neuronal damage, partly by inducing iron load. In experiments with cultured neurons, accumulation of pathological Aβ oligomers produces significant neurotoxic effects when neurons are cocultured with glial cells (Urrutia et al., 2017). This shows that glial cell activity is an important factor in potentiating the toxic effect of Aβ oligomers. In order to protect themselves from cellular damage neurons increase ferritin depots, but the accumulating iron is very reactive and once it accumulates in sufficient amount it will cause oxidative damage. Iron load initiates a vicious cycle of neuronal damage, caused by the ability of excess iron to

accelerate oligomer toxicity (Rottkamp et al., 2001; Wan et al., 2017).

Although evidence for the role of inflammation and resultant iron load in neurodegenerative diseases is accumulating, it is of therapeutic importance to explain how do these pathogenic factors change the dynamic of cellular biochemistry. Data from animal models show that aging creates a favorable milieu for neuronal damage via iron load. In rats, levels of DMT1 and hepcidin are increased, while levels of FPN do not change (Wang et al., 2010; Lu et al., 2017). This would indicate that cellular iron load with aging occurs due to increased iron import. Increased cellular iron stimulates hepcidin expression, which is not able to control FPN expression. It is interesting to notice that FPN is downregulated in aged rats with APP knockout (Belaidi et al., 2018). Furthermore, APP knockout increases the rate of brain iron accumulation that occurs with aging. This means that the preservation of cellular iron efflux is important for slowing age-related brain iron overload. In brains of AD patients and mice hepcidin expression is located in damaged neurons and blood vessels of amyloid plaques (Raha et al., 2013). On the other hand, similar to changes occurring with age, we observe increased DMT1 expression, but also co-localization of DMT1 with amyloid plaques in AD patients and in animal models with AD (Zheng et al., 2009). Animal models suggest that DMT1 expression is directly linked with the metabolism of pathogenic peptides that accumulate during AD (Zheng et al., 2009). Furthermore, DMT1 silencing increases cell viability in AD (Zheng et al., 2009). Similarly, in human patients, animal models and cell cultures with PD, DMT1 is also upregulated and related to oxidative stress (Salazar et al., 2008; Chen et al., 2015). Consequently, blocking DMT1 and hepcidin improves cell viability, while FPN upregulation rescues neuronal function in cultured cells and mice brain with PD (Chen et al., 2015; Xu et al., 2016; Xu Q. et al., 2017). So, at least in animal models with neurodegenerative disease, suppression of iron import and stimulation of iron export (via hepcidin manipulation) ameliorates neuronal degeneration.

In other diseases accompanied with neurodegeneration (Huntington disease, amyotrophic lateral sclerosis) iron deposition is also increased in different brain structures (Chen et al., 2013; Muller and Leavitt, 2014; Szeliga et al., 2016; Moreau et al., 2018). Changes in expression of iron protein carriers and the therapeutic potential of iron chelation evoke striking similarities with AD and PD (Salazar et al., 2008; Zheng et al., 2009; Chen et al., 2013; Belaidi and Bush, 2016; Szeliga et al., 2016; Moreau et al., 2018). It is speculative to suggest that the role of hepcidin in neurodegenerative diseases might be of primary importance, due to the dominating role of innate immune system in the pathophysiology of these diseases. Hepcidin has been evidenced quite robustly as part of our innate immune response, which is helpful in acute situations, because it controls iron availability to pathogens, but also it can regulate actions of inflammatory cytokines (De Domenico et al., 2010; Armitage et al., 2011; Burté et al., 2013). But, during chronic innate immune activity the negative feedback control between hepcidin and inflammation observed in acute situations turns into an agonistic relationship which will eventually deteriorate brain cell damage.

## THE RATIONALE FOR USING HEPCIDIN THERAPEUTICS IN NEURODEGENERATIVE DISEASES

Stimulation and inhibition of hepcidin has been used in cultured cells and animal models to ameliorate brain damage. The rationale is evident; hepcidin affects FPN and iron import proteins, which means its manipulation can control cellular iron load. Furthermore, it has been shown that decreased activity or expression of FPN promotes neuronal damage (Song et al., 2010; Yang et al., 2011; Crespo et al., 2014). In animal models with PD neurodegeneration is associated with increased microglial activity and FPN downregulation (Zhang et al., 2014). In addition, human trials with iron chelators have shown promise in retarding the progress of neuronal damage (Crapper McLachlan et al., 1991; Martin-Bastida et al., 2017; Moreau et al., 2018). But, long-term results of this therapy are still unknown and pending future trials (Adlard and Bush, 2018).

In animal models with brain inflammation and increased oxidative stress, direct and indirect suppression of local and systemic hepcidin offers neuroprotection (Chen et al., 2015; Li W.-Y. et al., 2016; Pan et al., 2016; Xiong et al., 2016). But, in conditions with lack of inflammation, the use of ad-hepcidin has also beneficial effects in neuronal function, probably due to hepcidin ability to protect neurons during iron overload via suppression of cellular iron import proteins. Still, there exist practical issues concerning the use of delivery methods for hepcidin in these scenarios. In animal models this has been done via intracerebroventricular injections, but the feasibility of this method in humans is unknown (Gong et al., 2016; Urrutia et al., 2017). Recent advances suggest that nanotechnology could make an important breakthrough in treating brain diseases with targeted delivery of peptides (Goldsmith et al., 2014; Silva Adaya et al., 2017).

Based on the accumulating evidence about the role of inflammation on cellular iron content, it is reasonable to assume that neuroprotection can be achieved by blocking the inflammatory pathways through already established drugs. One such drug is acetylsalicylic acid, which has already been used in cultured cells to protect neuronal cells and microglia from inflammation-induced damage (Li W.-Y. et al., 2016; Huang et al., 2018) (**Table 1**). The protective effect of acetylsalicylic acid is realized through increase in iron export in neuronal cells (Huang et al., 2018), while in microglia it reduces iron import and increases iron export (Xu et al., 2015).

As it was previously mentioned heparins can affect liver hepcidin expression. One such heparin, called dalteparin, has been administered via intraperitoneal injections in mice to suppress brain hepcidin expression during inflammatory conditions (Farajdokht et al., 2015). Dalteparin can also suppress serum IL-6 levels in these conditions, which means that its beneficial properties in reducing increased brain iron during inflammation could be due to suppression if TABLE 1 | Therapeutic potential of the manipulation of hepcidin and its target proteins in neurodegeneration.


6-OHDA, 6-hydroxydopamine; APP, amyloid precursor protein; BMVEC, brain microvascular endothelial cells; DMT1, divalent metal transporter 1; EGCG, epigallocatechin-3-gallate; FPN, ferroportin; IL-6, interleukin-6; L-LYC, liposomal lycopene; LPS, lipopolysaccharide; siRNA, small interfering RNA; TFR1, transferrin receptor 1; TNFα, tumor necrosis factor α.

systemic inflammation and brain hepcidin. Still the cause/effect relationship of dalteparin in the brain has not been studied thoroughly in terms of its impact on brain iron metabolism via hepcidin regulation. Future studies should also explain whether dalteparin-induced brain hepcidin suppression is dosedependent, because of the side effects of this drug on brain homeostasis (Perry et al., 2010).

Another therapeutic option that has been used is based on methods of delivery that increase the efficiency of neuroprotective substances. One example is the use of liposomes, which serve as particles that deliver drugs to target cells, whilst keeping intact the stability of the substance that is being used for therapeutic purposes. Liposomal lycopene has been delivered in rat brains to reduce oxidative stress and ameliorate apoptosis caused by ischemia. This effect is at least partly attributed to the ability of lycopene to reduce hepcidin expression in astrocytes via its suppressive effect on IL-6 signaling (Zhao Y. et al., 2018).

#### CONCLUSION

Hepcidin is emerging as an important peptide for brain (patho)physiology. The source of hepcidin in the brain is local and probably systemic. It is believed that systemic hepcidin can pass through BBB, especially during BBB leakage (for example during inflammation). Neurons, glial cells, endothelial cells, and other brain cells may express hepcidin, although basal levels of brain hepcidin are low. They increase significantly during inflammation and most likely during brain iron load. The local molecular regulation of brain hepcidin expression is unfolding, while studies suggest that it bears resemblance with liver hepcidin regulation. Still, the details of this regulation await further clarification from future studies. Hepcidin produced by the brain cells in pathological conditions may limit iron transport through BMVEC, and thereby also limit neuronal iron load. High iron load is detrimental for neuronal function. In addition, high iron load can also affect glial cell activity by turning them into cells that promote neuronal damage. The observation of robust production of hepcidin by glial cells compared to neurons during changes in brain homeostasis is

#### REFERENCES


in-line with the already established role of glia as supportive, regulatory and protective cells in the brain milieu. During inflammatory conditions hepcidin promotes neuronal iron load, while the opposite occurs in non-inflammatory conditions. This observation is based on studies that have revealed the neuroprotective nature of hepcidin suppression during inflammation. This occurs due to the increase in iron import caused by inflammation, which is further exacerbated by FPN downregulation caused by hepcidin-dependent and hepcidinindependent mechanisms. On the other hand, use of adhepcidin during neuronal iron load (not caused by inflammation) protects these cells from oxidative stress. Intriguingly, hepcidin pretreatment also protects neurons from the deleterious effects of inflammation. It seems that pretreatment with hepcidin blocks the initiation of inflammation-induced biochemical cascade inside brain cells. This effect of hepcidin is not without precedent, because it has been observed in mice and human studies. Unfortunately, therapeutic implications of this physiological effect of hepcidin have not been studied in details, despite its potential crucial importance in brain and other pathologies. Still, pharmacologic manipulation of hepcidin is emerging as e new therapeutic territory in neurodegenerative diseases. Animal models have shown that the use of ad-hepcidin and suppression of hepcidin protects brain cells in models of neurodegenerative diseases. This has been further confirmed with the use of standard anti-inflammatory drugs such as acetylsalicylic acid or lycopene.

Finally, considering the ever increasing importance of hepcidin for brain physiology, it is paramount for future studies to examine models with different cell specific hepcidin knockout in AD and PD that will reveal the temporal changes of hepcidin, but that will also define whether hepcidin is of primary importance for the pathophysiology of neurodegenerative diseases.

#### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and has approved it for publication.



ferroportin are reduced in the brain in Alzheimer's disease. Acta Neuropathol. Commun. 1:55. doi: 10.1186/2051-5960-1-55



models. Neuroscience 257, 20–30. doi: 10.1016/J.NEUROSCIENCE.2013. 09.037


**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 Vela. 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.

# Iron Dysregulation and Dormant Microbes as Causative Agents for Impaired Blood Rheology and Pathological Clotting in Alzheimer's Type Dementia

#### Lesha Pretorius<sup>1</sup> , Douglas B. Kell2,3 \* † and Etheresia Pretorius<sup>1</sup> \*

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Zhou Liu, Guangdong Medical University, China Andrea Ticinesi, Università degli Studi di Parma, Italy

#### \*Correspondence:

Douglas B. Kell dbk@manchester.ac.uk orcid.org/0000-0001-5838-7963 Etheresia Pretorius resiap@sun.ac.za orcid.org/0000-0002-9108-2384

#### †Present address:

Douglas B. Kell, Dept of Biochemistry, Institute of Integrative Biology, University of Liverpool, Liverpool, United Kingdom

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 04 September 2018 Accepted: 30 October 2018 Published: 16 November 2018

#### Citation:

Pretorius L, Kell DB and Pretorius E (2018) Iron Dysregulation and Dormant Microbes as Causative Agents for Impaired Blood Rheology and Pathological Clotting in Alzheimer's Type Dementia. Front. Neurosci. 12:851. doi: 10.3389/fnins.2018.00851 <sup>1</sup> Department of Physiological Sciences, Stellenbosch University, Stellenbosch, South Africa, <sup>2</sup> School of Chemistry, The University of Manchester, Manchester, United Kingdom, <sup>3</sup> The Manchester Institute of Biotechnology, The University of Manchester, Manchester, United Kingdom

Alzheimer's disease and other similar dementias are debilitating neurodegenerative disorders whose etiology and pathogenesis remain largely unknown, even after decades of research. With the anticipated increase in prevalence of Alzheimer's type dementias among the more susceptible aging population, the need for disease-modifying treatments is urgent. While various hypotheses have been put forward over the last few decades, we suggest that Alzheimer's type dementias are triggered by external environmental factors, co-expressing in individuals with specific genetic susceptibilities. These external stressors are defined in the Iron Dysregulation and Dormant Microbes (IDDM) hypothesis, previously put forward. This hypothesis is consistent with current literature in which serum ferritin levels of individuals diagnosed with Alzheimer's disease are significantly higher compared those of age- and gender-matched controls. While iron dysregulation contributes to oxidative stress, it also causes microbial reactivation and virulence of the so-called dormant blood (and tissue) microbiome. Dysbiosis (changes in the microbiome) or previous infections can contribute to the dormant blood microbiome (atopobiosis<sup>1</sup> ), and also directly promotes systemic inflammation via the amyloidogenic formation and shedding of potent inflammagens such as lipopolysaccharides. The simultaneous iron dysregulation and microbial aberrations affect the hematological system, promoting fibrin amylodiogenesis, and pathological clotting. Systemic inflammation and oxidative stress can contribute to blood brain barrier permeability and the ensuing neuro-inflammation, characteristic of Alzheimer's type dementias. While large inter-individual variability exists, especially concerning disease pathogenesis, the IDDM hypothesis acknowledges primary causative factors which can be targeted for early diagnosis and/or for prevention of disease progression.

Keywords: Alzheimer's type dementia, iron, gut dysbiosis, iron dysregulation and dormant microbes hypothesis, inflammation

<sup>1</sup>Presence of microbes in the wrong place, i.e., sterile environments such as blood.

# INTRODUCTION

fnins-12-00851 November 16, 2018 Time: 12:57 # 2

Alzheimer's type dementia (AD) is the most prevalent example of dementia, accounting for 75% of all cases (Qui et al., 2009), with an estimated global economic impact of \$604 billion per year (Langa, 2015). In 2013 this neurodegenerative disorder was diagnosed in 44.4 million individuals worldwide. This number is expected to increase to 75.6 million by 2030 (Vradenburg, 2015; Pretorius et al., 2016a). The increasing aging population and anticipated increase in prevalence of AD has severe socioeconomic implications (Prince et al., 2014). Familial AD can be promoted by mutations of the amyloid precursor protein (APP), presenilin (Bu et al., 2015) and apolipoprotein E (Ayton et al., 2015) genes, however, most AD patients suffer from the sporadic form, and even after decades of research, the etiology and pathogenesis of this disease remains largely unknown (Makin, 2018), and with essentially no existing disease-modifying treatments (Vogt et al., 2017). Despite the differences between familial and sporadic AD, the resultant neuropathology is shared by both (Pritchard et al., 2017).

Vascular dementia (VaD), also known as multi-infarct dementia, is caused by obstructions in the supply of blood to the brain, which over time results in stepwise neurodegeneration (Karantzoulis and Galvin, 2011). The pathology of VaD is characterized by atheromas of primarily cerebral arteries and arterioles. Vascular lesions characteristic of VaD often co-exist with AD, with 25–80% of individuals with dementia showing both AD symptoms and cerebrovascular lesions (Jellinger, 2008), a condition termed mixed dementia (MD). In early AD, critically located small vascular lesions in subcortical regions may promote cognitive decline. This suggests a synergistic relation between disorders (Jellinger and Attems, 2007). The distinction between isolated AD, VaD, and MD, where both pathologies coexist in the same individual, remains a controversial issue and one of the most difficult diagnostic challenges (Zekry et al., 2002; Jellinger and Attems, 2007). Additionally the population-based prevalence of MD in prospective and retrospective autopsy studies, ranges from 2 to 58%, emphasizing the coexistence of AD with multiple cerebrovascular lesions in patients with cognitive impairments. Moreover inflammation and alterations of inflammatory biomarkers are common to both AD and VaD (Schmidt et al., 2002). For these reasons, VaD and the pathogenesis thereof, fall under the AD umbrella of this review.

For the individual, AD shortens life expectancy significantly, with the cognitive deficit often causing institutionalization, physical disability and reduced quality of life (Qui et al., 2009; Pritchard et al., 2017). Hallmark features of AD pathology in the brain often include extracellular plaques composed of amyloid-β (Aβ), intracellular neurofibrillary tangles which are composed of hyperphosphorylated tau, basal forebrain cholinergic insufficiency and extensive neuronal and synaptic loss in the cortex and hippocampus (Law et al., 2001; Vogt et al., 2017). Aβ42 peptides either as monomers, dimers or intermediate amyloids stimulate arrays of inflammatory gene expression characteristic of the innate immune system (Zhao et al., 2015). These Aβ42 peptides are not only highly immunogenic, but they may self-aggregate into structures that interact with the plasma membrane and cause uncontrolled seepage of ions into and/or out of the neurons (Zhao and Lukiw, 2015), disrupting their normal physiological functioning.

Another hallmark of AD is activated neuroglial cells which produce significant amounts of inflammatory molecules. At low levels this may have a protective function but the increased expression in AD could induce neurodegeneration (Kamer et al., 2008). This said, cerebral inflammation caused by the activation of glial cells is generally asymptomatic as this occurs in the normal aging brain. According to Barrientos et al. (2016) prolonged production of pro-inflammatory cytokines as well as increased neuroinflammatory responses are common in the normal aging brain. The primary source of this exaggerated response is sensitized microglia. Other causes of this phenomenon may include neuroendocrine system dysregulation and potentiation of neuroinflammatory responses following an immune challenge, due to persistently elevated glucocorticoid levels in older individuals (Barrientos et al., 2016). Excessive age-related glial priming, present in AD, could also be due to microbial infiltration from disturbed microbiomes across a permeabilized blood brain barrier, providing slow inflammatory damage (Pritchard et al., 2017).

Recent research (Dorey et al., 2014; Filiou et al., 2014; Steardo et al., 2015) also suggests that neuro-inflammation may play a central role in the pathological progression of AD. Several neuro-inflammatory mediators including reactive oxygen species, chemokines, cytokines and activated complement are produced and secreted by microglia and astrocytes in the AD brain (Pretorius et al., 2016a). While neuro-inflammation is a part of the normal aging brain, and is a hallmark of acute injury (Lucas et al., 2006) and also of infection (McColl et al., 2009), in AD excessive systemic inflammation also occurs (Pretorius et al., 2016a), simultaneously exacerbating and perpetuating neuroinflammation. Previous studies have verified that peripheral inflammatory markers such as interferon-γ, tumor necrosis factor-alpha (TNF-α), interleukin-1β (IL-1β), and IL-6 are related to the advancement of AD (Rubio-Perez and Morillas-Ruiz, 2012; Bu et al., 2015).

Some individuals who develop AD seem to possess genetic susceptibilities that are influenced by co-occurring environmental factors (Pritchard et al., 2017). In this review, we rehearse the Iron Dysregulation and Dormant Microbes (IDDM) hypothesis, and its ability to relate events which trigger and promote the pathogenesis of inflammatory conditions such as AD. The IDDM hypothesis emphasizes two main external triggers, namely (1) stress-induced iron dysregulation, and (2) its ability to reactivate dormant or non-replicating microbes which the host had acquired via previous infections or dysbiotic communities. We believe the primary origin of such microbes comes from a dysbiotic gut (Kell and Pretorius, 2015a). These microbes are capable of sloughing off functionally significant inflammagens such as lipopolysaccharide(s) (LPS) from Gram-negative bacteria and lipoteichoic acid (LTA) from Grampositive bacteria. The consequences of this include significant coagulopathies, for example the amyloidogenic clotting of blood, which increases cell mortality (Kell and Pretorius, 2018). The contents of this manuscript are summarized in **Figure 1**.

#### IRON DYSREGULATION

Iron dysregulation is any form of deviation from normal, homeostatic iron metabolism; in particular it includes cerebrospinal fluid ferritin and serum ferritin levels as implicated in Alzheimer's type dementias (Ayton et al., 2015). According to the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort study, increased ferritin levels, measured in cerebrospinal fluid, were negatively correlated to cognitive performance (Mueller et al., 2005), and these levels predicted conversion of mild cognitive impairment to Alzheimer's disease in 144 individuals over 7 years (Ayton et al., 2015). Furthermore, a strong association between ferritin levels and cerebrospinal fluid apolipoprotein E levels was observed in this study, with the major risk allele, APOE-ε4, inducing 22% higher levels of cerebrospinal fluid ferritin levels when compared to non-ε4 carriers (Ayton et al., 2015). In the same study a modest association between cerebrospinal fluid ferritin and plasma ferritin levels was noted (p = 0.0002). Thus, there may be great clinical relevance for the use of systemically elevated serum ferritin (SF) levels as cognitive performance markers (Kell and Pretorius, 2014; Pretorius et al., 2016a).

#### Causes of Iron Dysregulation

Major sources of iron dysregulation stem from externally induced stressors (Kell and Pretorius, 2018). This form of iron dysregulation can be initiated by several factors that contribute to or cause cell death, such as mechanical damage (Zhang et al., 2013), nutritional stress (Schaffer, 2016), pharmacological stress (Primohamed et al., 2004), and of course oxidative stress (Kerley et al., 2018). Another source of free iron is via heme metabolism, due to the functioning of heme oxygenase-1 (HO-1), which catalyzes the degradation of heme (Pretorius and Kell, 2014). Since upregulation of HO-1 activity occurs in systemic inflammatory disorders in which erythrocytes are lysed, it may also be an important marker of inflammation and iron dysregulation.

Additionally, hepcidin, produced by the liver, is a key regulator of iron metabolism (Michels et al., 2015; Reichert et al., 2017). Decreases in hepcidin levels enhance surface exposure of ferroportin (Ganz and Nemeth, 2012) on enterocytes, macrophages and hepatocytes to increase serum ferritin levels (illustrated by **Figure 2**). Hepcidin expression is induced by inflammatory markers such as LPS, IL-1β, and IL-6, while increases in 1,25(OH)2D<sup>3</sup> (calcitriol) levels cause hepcidin levels to decrease (Kell and Pretorius, 2018). According to a report by Bacchetta et al. (2014), decreases in hepcidin levels by 1,25(OH)2D<sup>3</sup> are due to suppression of the HAMP gene by the vitamin D receptor (VDR). Chromatin immunoprecipitation assays confirmed the binding of VDR to the vitamin D response element within the proximal promotor region of the HAMP gene (Bacchetta et al., 2014). While this process is intricate, it appears that alterations in vitamin D metabolism could potentially instigate iron dysregulation.

Intestinal inflammation caused by gut dysbiosis can impact iron homeostasis within the GI tract (Cherayil et al., 2011), however, whether these findings have been extrapolated to serum iron homeostasis has not yet been elucidated. While iron dysregulation within the GI tract and gut dysbiosis potentially exacerbate one another, Constante et al. (2017) concluded that luminal heme originating from gastrointestinal bleeding or dietary components more likely contributes to dysbiosis of the gut microbiota in mice than vice versa. However, dietary nonheme iron intake from food has been associated with a 30% increased risk of Parkinson's disease (p = 0.02) (Logroscino et al., 2008). In the same study authors also observed that supplemental

iron intake was associated with a borderline increase in Parkinson's disease among men (Logroscino et al., 2008). Nonetheless, the most prominent cause of iron dysregulation in the form of elevated serum ferritin levels is cell death (Kell and Pretorius, 2018).

#### Unliganded Iron and Oxidative Damage

In AD, iron dysregulation and the benefits of its chelation have been known for decades (Crapper McLachlan et al., 1991), and many reviews point to the role of poorly liganded iron in AD development (Exley, 2006; Castellani et al., 2007, 2012; Kell, 2009, 2010; Sandro and Muckenthaler, 2009; Grunblatt et al., 2011; Weinreb et al., 2011; Peters et al., 2015; Wood, 2015; Belaidi and Bush, 2016). Unliganded iron and the accompanying oxidative damage is causatively related to neuro-inflammation (Bester et al., 2013; Hongkuan et al., 2015). According to Nunomura et al. (2001), neurons with neurofibrillary tangles show a 40– 56% decrease in relative 8-hydroxyguanosine (8-OHG) levels when compared to neurons without neurofibrillary tangles in AD patients. 8-OHG is a marker of hydroxyl radical formation (Kell, 2009, 2010), so this finding indicates that oxidative damage is an early event in AD progression, which dissipates with lesion formation (Nunomura et al., 2001). This is further supported by (Moreira et al., 2005) who found one of the earliest pathological events in AD to be an imbalance between free radical scavenging and generation. The relationship between poorly liganded iron and oxidative stress involves iron's catalytic contribution to the Fenton and Haber-Weiss reactions. These reactions are involved in the formation of the highly reactive hydroxyl radical (OH−), the most biologically detrimental free radical species, specifically with regards to macromolecules such as lipids, proteins and nucleic acids (Moreira et al., 2005; Kell, 2009, 2010; Lipinski and Pretorius, 2013; Pretorius et al., 2016a). Notably, iron chelation can reverse tau protein hyperphosphorylation in mice brains (Guo et al., 2013), thereby reducing the formation of intracellular neurofibrillary tangles.

# Microbial Reactivation

fnins-12-00851 November 16, 2018 Time: 12:57 # 5

Dormancy is extremely common in microbiology, even among non-sporulating bacteria (Kaprelyants et al., 1993; Lewis, 2007; Kell et al., 2015; Kell and Kenny, 2016), for sound evolutionary reasons (Mukamolova et al., 2003). Our major hypothesis involves shared symptoms across numerous chronic inflammatory diseases caused by dormant microbes (Pretorius et al., 2016a). As reviewed in 2016 (Pretorius et al., 2016a), microbial reactivation or resuscitation, which may be autocatalytic (Mukamolova et al., 1998), indicates that these so-called dormant microbes which are non-growing and appear operationally dead (Kell et al., 1998), can recover culturability. Metals, although explicitly involved in production of free radicals and potentially oxidative stress in AD, are also involved in the process of dormant microbe reactivation (Bester et al., 2015; Kell et al., 2015). All microbes (with the possible exception of Borrelia) require free available iron to grow. This follows from the in vivo limitation of microbial growth in the absence or decreased availability of free iron (Markel et al., 2007; Reid et al., 2009; Chu et al., 2010; Nairz et al., 2010, 2014; Skaar, 2010; Haley and Skaar, 2012; Armitage and Drakesmith, 2014; Subashchandrabose and Mobley, 2015; Damron et al., 2016; Carver, 2018). It is its absence in normal metabolism that causes dormant bacteria to remain in that state. With this in mind, iron dysregulation specifically in the form of high SF levels (Kell and Pretorius, 2014), accelerates AD pathology in two ways (Pretorius et al., 2016a) which suggests the opportunity of its use as a biomarker for early diagnosis and/or a target for slowing down disease progression (Bester et al., 2013).

# GUT MICROBIOTA

The human gut microbiome contains approximately 10–100 trillion microbes, outnumbering human genes by 100-fold (Zhu et al., 2010). The gut microbiome forms the largest diffuse organ system in the body, and is more metabolically active than is the liver (Zhao et al., 2015). Through various systemic effects, the well-being and health of the human host is largely impacted by these microbes (Zhao and Lukiw, 2015). However, the human microbiome displays a high degree of variation at inter-individual levels (Costello et al., 2009). For example, even as infants the composition of the gut microbiota is idiosyncratic with significant inter-individual variation being evident from the 1st day after birth (Chong et al., 2018). As a consequence it is has been impossible to define a healthy microbiome (Dave et al., 2012), but, it has become apparent that certain compositions of microbial communities may be able to promote both health and disease (Blum, 2017).

Intriguingly the human gastrointestinal (GI) tract has coevolved with two major phyla: (1) Firmicutes which constitutes 80% and, (2) Bacteriodetes which constitutes 20% of all GI tract bacteria (Pritchard et al., 2017; Zhao et al., 2017). There is increasing evidence implicating host–microbiome interactions at all stages of complexity including the central nervous system (Stilling et al., 2014). According to Stilling et al. (2014) gut-microbial products can distress chromatin plasticity, leading to changes in neuronal transcription and host conduct. This is due to the fact that the microbiota is an essential mediator of gene–environment interactions and may themselves be regarded as an epigenetic entity. Current research which supports this view includes reports which have characterized small non-coding RNA secreted from microbial cells in the GI tract (Ghosal et al., 2015; Zhao et al., 2017).

Alterations of the gut microbiota can promote proinflammatory cytokine release and increase intestinal permeability, which has also been associated with AD (Pistollato et al., 2016). In particular, changes of gut microbial diversity and density can result in systemic and neuro-inflammation as well as dysfunction of the cerebellum and hippocampus (Pistollato et al., 2016). Considering the multitudes of LPSs and amyloids in the GI tract, it is plausible that the microbiota is involved in the pathogenesis of neurological disorders hallmarked by amyloidogenic features (Tran and Greenwood-Van Meerveld, 2013; Shoemark and Allen, 2015; Zhao and Lukiw, 2015) such as AD.

#### Dysbiosis

Dysbiosis refers to alteration in the composition of the microbiota and has been implicated in the etiology of various disease states (de Oliveira et al., 2017; Kho and Lal, 2018). Perturbations of the gut microbiota can occur as consequences of antibiotic treatment (especially during infancy), dietary changes, sedentary behavior, food additives, non-steroidal anti-inflammatory and other drugs (Wallace, 2013; Ticinesi et al., 2017; Maier et al., 2018) and various other conditions (Pistollato et al., 2016). Dietary changes, for example, may explain up to 57% of the total structural variation the gut microbiome (Brown et al., 2012). Using a humanized mouse model in which adult human fecal microbiota were transplanted into germ-free mice, Brown et al. (2012) reported a shift in the composition of microbiota after switching the mice from a low-fat, plant polysaccharide-rich diet to a Western diet. While diets rich in refined sugars result in different changes to the microbiota from diets lacking fiber, due to the intricate balance that exists between microbial species, alterations in one species can disrupt the entire microbial community (Brown et al., 2012). Thus, alterations in diet, although modifiable, could potentially be a major hurdle in treatment dysbiosis-related conditions if not adequately addressed. Furthermore, physical activity and its relative intensity can induce or prevent gut dysbiosis by influencing splanchnic blood flow, intestinal permeability and inflammatory cytokine expression (Monda et al., 2017). For example, in mice, even in combination with a high-fat diet, exercise may reduce inflammatory infiltrate into epithelial cells, protecting the integrity of the GI wall (Campbell et al., 2016).

Broad-scale fluctuations in the gut microbiota play significant roles in disease advancement through immune and systemic activation (Vogt et al., 2017). There is considerable evidence for the existence of the gut-brain axis which allows bidirectional communication through various pathways such as

neural, endocrine and immune mechanisms (Gareau, 2014; Lyte, 2014; Bienenstock et al., 2015; Köhler et al., 2016; Fung et al., 2017; Sandhu et al., 2017). Within this context, changes in microbial communities could lead to pathophysiological changes in the brain of individuals with AD. Dysbiosis is often associated with increased anxiety and memory impairment due to decreased secretion of neurotrophic factors such as brain-derived neurotrophic factor (Stilling et al., 2014).

#### Dysbiosis in Alzheimer's Type Dementia

The recognition of dysbiosis and its possible link to neurodegenerative diseases are increasing as our understanding about the gut-brain axis improves (Miklossy and McGeer, 2016). A recent study in which transgenic AD mice were raised under germ-free conditions, indicated that these mice, compared to conventionally-raised AD mice, had less cerebral amyloid deposition (Harach et al., 2017). This indicates that a dysbiotic gut may influence progression of amyloid pathologies.

The gut microbiome diversity of AD patients is notably decreased and compositionally distinct from age- and gendermatched controls (Vogt et al., 2017). In this study Vogt et al. (2017), the authors reported decreases in Firmicutes and Bifidobacterium spp. and increased Bacteriodetes in AD patients. Bifidobacterium spp. are important in gut health and their beneficial effects are well-documented (Arboleya et al., 2016). For instance, particular Bifidobacterium species, such as B. bifidum, B. cantenulatum, B. breve, and B. adolescentis, have anti-inflammatory properties such as inhibiting LPS-induced IL-8 production and TNFα expression (Khokhlova et al., 2012) Bifidobacteria are also known to reduce intestinal permeability (Underwood et al., 2015). Moreover, Bifidobacterium supplementation has shown to decrease intestinal LPS levels and mend the GI-mucosal barrier in mice (Griffiths et al., 2004; Wang et al., 2006). The phylum Bacteriodetes encompasses an abundant and diverse group of Gram-negative bacteria, which have been detected as being increased in patients with Parkinson's disease (Keshavarzian et al., 2015). The shedding of LPS and subsequent induction of inflammation is also associated with the increase in numbers of fraction of these bacteria. Considering these findings, individuals with AD may present with a gut microbial phenotype that has an increased propensity for the translocation of inflammatory bacterial components (Vogt et al., 2017).

#### Amyloidogenesis

Amyloid by definition is protein that is deposited as an insoluble fibril as a result of successive alterations in the protein folding process, known as amyloidosis. Similarly to prions, there is no change in the primary amino acid sequence of the proteins when they adopt an insoluble amyloid form (Kell and Pretorius, 2018). Interestingly, self-associating amyloidogenic lipoproteins are secreted by most microbial species (Schwartz et al., 2016), and amyloid is one of the main secretory products of the gut microbiome (Zhao et al., 2017). The role of the gut in dissemination of amyloid proteins has been reviewed by Pistollato et al. (2016), in which the authors adopted the unifying prion concept to explain transmission of these prion-like proteins. The cumulative life-long contribution of microbial amyloid to neurodegenerative pathophysiology is little recognized. Progressive production and aggregation of amyloids contributes to the pathogenesis of diseases in which amyloids accumulate, which is often mediated by microglial cells (Zhao et al., 2017). Thus, bacterial amyloids originating from the gut microbiome may enhance inflammatory responses to cerebral accumulation of Aβ and play a role in the pathogenesis of AD (Pistollato et al., 2016).

For example, functional amyloid from Klebsiella pneumoniae can depolymerise from its fibrillary state to release oligomers which induce cytotoxicity equal to that caused by pathological Aβ in AD (Shahnawaz and Soto, 2012). The molecular mimicry theory (Ebringer and Rashid, 2009; Proal et al., 2013, 2017) includes microbial curli fibers<sup>2</sup> and Aβ as a protein–protein interaction which could result in cross-seeding, even if these proteins are essentially unalike (Friedland, 2015; Zhao et al., 2015). These extrinsic curli proteins act as pathogen-associated molecular patterns with increased β-pleated sheet structures, and can cross-react with antibodies to Aβ plaques (Miklossy, 2016). It is plausible that extracellular senile plaques, due to their morphological appearance and density, may in fact be constituted of curli-like Aβ compositions. In theory, this gives bacterial phylotypes a more blatant role in AD causality (Pritchard et al., 2017). The influence of gut microbiota on amyloid formation and propagation is more significant throughout aging as both the epithelium of the GI tract and the blood-brain barrier (BBB) increase in permeability to smaller molecules (Tran and Greenwood-Van Meerveld, 2013; Pistollato et al., 2016).

#### Shedding of Inflammagens

Some of the foremost bacterial species of the GI tract, such as the Gram-negative bacilli Bacteroides fragilis and Escherichia coli, secrete a multifarious selection of pro-inflammatory neurotoxins which are not only pathogenic but also detrimental to the homeostatic functioning of neurons (Zhao et al., 2017). The shedding of inflammagens such as LPS from gut-microbiota-related dysbiosis causes systemic and neuroinflammation and promotes gut permeability. Shedding can occur in response to varying physiological and environmental signals with the most extreme example of microbial shedding known as the Jarisch-Herxheimer reaction. This is basically an uninhibited cytokine storm, which is caused by prompt release of inflammagenic material, often following antibiotic treatment of syphilis (Kell and Pretorius, 2018), or other diseases caused by spirochetes. As mentioned earlier, this may promote neuronal dysfunction and apoptosis, impairment of synaptic plasticity and induced vulnerability to cognitive decline (Daulatzai, 2014). Bacterial LPS has also been found in AD hippocampal brain lysates where the mean LPS levels were three-fold higher in the hippocampus of AD patients than in age-matched controls. This was said to increase to 26-fold in advanced AD hippocampal cases (Zhao et al., 2017).

<sup>2</sup>Extracellular bacterial amyloids.

LPSs are sparingly soluble but over a period of time characteristically form large heterogenous aggregates which are exceptionally immunogenic (Zhao et al., 2015). Interestingly the glycosylphosphatidyl-inositol-anchored LPS and microbedetecting CD41 receptor, which is crucial in the neutralization of invading microbes, is similarly, stimulated by Aβ fibrils (Dowhan, 2014), relating innate-immune signaling with amyloidogenesis in AD. It has also been noted that previous bacterial infections which resulted in the formation of antibodies to amyloids or bacterial inflammagens may predispose central nervous system amyloids to ensuing attack by antibodies, which results in upregulation of neuro-inflammation (Via et al., 2015; Zhao et al., 2015). The inflammagenic potency of LPS is so pronounced that it is frequently used to stimulate in vivo AD models (Lee et al., 2008; Catorce and Gevorkian, 2016; Zakaria et al., 2017).

LTA is the cell wall equivalent of LPS in Gram-positive bacteria and is equally capable of inducing an inflammatory response through its interactions with toll-like receptor (TLR) 2 (Kell and Pretorius, 2018). LTA has not been as well studied as LPS with regards to inflammagenesis, however, recent work suggests that LTAs may in fact be more potent than LPSs (Pretorius et al., 2018). The stimulation of inflammatory cytokine production by both LPS and LTA are mediated through TLR binding (reviewed by Kell and Pretorius, 2018). In response to inflammagens, concentrations of acute phase biomarkers and inflammatory cytokines such as TNF-α, IL-1β, IL-6, IL-8, C-reactive protein (CRP), serum amyloid A (SAA), and fibrinogen increase significantly (deRosset and Strutz, 2015; De Buck et al., 2016; Pindjakova et al., 2017).

Additionally, increased BBB permeability typically observed in individuals with AD can be caused by shed bacterial inflammagens. The BBB plays a fundamental role in the initiation and continuance of chronic inflammation during AD (Zenaro et al., 2017). While Aβ accumulation in vasculature results in inflammatory events that increase BBB permeability in AD (Erickson and Banks, 2013), proteolytic enzymes such as carbonic anhydrases, peptidyl deiminases and gingipains, and appendages such as curli fibers, fimbriae and other amyloidlike proteins, which are carried with LPS (Pritchard et al., 2017) and contribute significantly to BBB permeabilization. BBB dysfunction during AD effects Aβ clearance, endothelial cell function, tight junction integrity and may activate glial cells which accelerates migration of leukocytes to the brain (Zenaro et al., 2017). This could stimulate the beginning of chronic neuro-inflammation in AD.

#### Atopobiosis

Atopobiosis refers to the appearance of some of the gut microbes in the wrong place (Potgieter et al., 2015). The sterility of blood is seemingly a controversial topic. The presence of a "physiological" blood microbiota has been reported by several studies using 16S ribosomal DNA quantitative polymerase chain reactions (Amar et al., 2013; Lelouvier et al., 2016; Martel et al., 2017; Proal and Marshall, 2018). Yet, in these studies the plausible source of these microbes from gut permeability was not investigated. Authors of the same studies have reported that changes in the composition of this blood microbiota can be associated with disease. However, based on immunological and clinical understanding (Ochei and Kolhatkar, 2000) the detection of bacteria in blood is always abnormal (Kinnby and Chavez de Paz, 2016; Andreadou et al., 2017) and thus referred to as atopobiosis (Potgieter et al., 2015). As seen in **Figure 3**, atopobiosis is caused by bacterial translocation in which bacteria move from the GI tract to normally sterile blood and tissues (Potgieter et al., 2015). Once translocation has taken place, these bacteria can cause direct infection and inflammation. The process of bacterial translocation has been reviewed by Potgieter et al. (2015) but can briefly be described by the simultaneous dysfunction of three role players: (1) dendritic cells (2) GI epithelium and, (3) M cells which overlay Peyer's patches. It is a combination of various compromised mucosal defenses that leads to bacterial translocation. When the influx of gut microbes and toxins is particularly abundant, it is often referred to as a "leaky gut" (Maes, 2009; Fasano, 2012; Quigley, 2016; Mu et al., 2017; Kell and Pretorius, 2018) and is due to severely compromised epithelial tight junctions (Fasano, 2012).

#### Dormant Blood Microbiome

Bacteria that have successfully translocated to the blood have the ability to become dormant and reactivate upon appropriate stimuli, such as iron dysregulation in the form of high SF levels as discussed in section Causes of Iron Dysregulation. In contrast to conventional infections, these bacteria do not multiply but enter dormant states (Kell and Kenny, 2016; Pretorius et al., 2016a; Kell and Pretorius, 2018). This is especially relevant in clinical settings when toxic concentrations of antibiotics promote bacterial persistence or adoption of a dormant state that permits their survival (Kell and Pretorius, 2018). Even though the major source of this dormant blood microbiome has been reviewed here as a dysbiotic and permeabilized GI tract, we also recognize periodontitis, gingivitis, urinary tract infections and even insemination as possible points of microbial infection and subsequent translocation (Kamer et al., 2008; Flores-Mireles et al., 2015; Kenny and Kell, 2017; Pretorius et al., 2017; Pritchard et al., 2017). Current research has drifted toward the correlation between an individual's infectious burden and risk of developing AD, with particular emphasis on herpes simplex virus 1 (HSV1) and human herpesvirus 6 and 7 infections (Honjo et al., 2009; Itzhaki, 2014, 2017; Bu et al., 2015; Itzhaki et al., 2016; Itzhaki and Tabet, 2017; Eimer et al., 2018; Itzhaki and Lathe, 2018; Readhead et al., 2018; Tzeng et al., 2018). Dormant bacteria can survive in leukocytes (Thwaites and Gant, 2011; Liehl et al., 2015) and erythrocytes (Potgieter et al., 2015) to establish the dormant blood microbiome. In Potgieter et al. (2015) transmission electron micrographs of cellular inclusions thought to be L-forms of bacteria inside erythrocytes of individuals diagnosed with AD can be seen (see also Mattman, 2001). **Figure 4** illustrates the association of bacteria with pathological coagulations within the blood of individuals diagnosed with AD. This dormant blood microbiome affects the integrity of the hematological system, promoting pathological coagulation.

FIGURE 4 | Scanning electron micrographs of bacteria associated with pathological coagulations found within the blood of individuals diagnosed with Alzheimer's disease. Unpublished data from Potgieter et al. (2015). (A) low magnification, (B) higher magnification.

#### PATHOLOGICAL COAGULATION

LPS is well known to cause inflammatory cytokine production which can cause hypercoagulation, often referred to as endotoxin-mediated hypercoagulation (Slotta et al., 2008; Kell and Pretorius, 2017). LPS may activate the coagulation pathway via upregulation of tissue factor (TF), which leads to activation of pro-thrombin (Landsem et al., 2015). Two main components of the hematological system namely fibrinogen and erythrocytes and the changes induced during inflammation and AD, will be discussed below.

#### Fibrinogen

Fibrinogen is the precursor of fibrin and thus plays a major role in thrombosis and hemostasis (Pretorius and Kell, 2014). During inflammation, circulating fibrinogen levels are increased,

while fibrinolysis is impaired. Amyloidogenesis of the fibrin fibers occurs, resulting in the formation of dense matted deposits which trap other blood cells and distort their shape. The combination of the increased propensity to form a stiffer fibrin network and reduced fibrinolysis is a feature of numerous inflammatory diseases such as AD (Ahn et al., 2010; Zamolodchikov and Strickland, 2012). **Figure 5** illustrates some of the morphological changes of fibrin fibers due to the addition of LPS.

The presence of excessive (unliganded) iron and the consequential production of hydroxyl radicals is one of the leading causes of altered fibrin cross-linking (Pretorius and Kell, 2014; Kell and Pretorius, 2015b). Dense matted deposits produced in the presence of ferric ions are abnormally resistant to chemical and proteolytic degradation, hypothesized to be due to the existence of intermolecular hydrophobic bonds (Lipinski and Pretorius, 2012). As reviewed, it seems that iron dysregulation may serve as a causative factor for multiple pathologies involved in AD progression. This altered fibrinogen structure, with excessive cross-linking and increased β-pleated sheet structure, has been implicated in the development of neuroinflammation and memory impairments. Increased fibrinogen levels are a strong indicator of cerebrovascular risk since fibrinogen binds to Aβ which further delays clot degradation (Bester et al., 2013; Pretorius et al., 2016a). In addition, pharmacological inhibition of the fibrinogen–Aβ interaction with Ru-505 altered thrombus structure and paused cognitive decline in mice (Ahn et al., 2014), providing a strong association between pathological clotting and neurodegenerative diseases.

#### Erythrocytes

The dynamics of erythrocytes represent an important but understudied feature of the cardiovascular system. Central to this is their rheology, viscosity, aggregation and deformability (Baskurt et al., 2011; Pretorius and Kell, 2014). It is thought that alterations in erythrocytes and their rheology can contribute to the pathogenesis of AD by hindering oxygen delivery (Tripathy et al., 2013). Dense matted deposits may also trap erythrocytes, impairing their effective delivery of oxygen to the brain. Proficient oxygen delivery is vital for normal functioning of the brain, underpinned by that fact that 20% of an individual's total oxygen intake is used by the brain. Neurons are particularly susceptible to hypoxic periods, which can cause irreversible neurological consequences (Lipinski and Pretorius, 2013) after only a few minutes. Altered functioning of erythrocytes is also strongly suggested by changes in their size distribution (Ozturk et al., 2013; Weuve et al., 2014; Pilling et al., 2017; Winchester et al., 2018).

As reviewed by Pretorius and Kell (2014), erythrocyte deformability is a complex process which is significantly altered by pathophysiological conditions. Briefly, reduced erythrocyte deformability is an important feature of inflammation in which reactive oxygen species cause degradation of spectrin and band 3, for example, which are important membrane proteins. Since the plasma membrane and the cytoskeleton are responsible for

of 0.2 ng/l−<sup>1</sup> 0111:B4 LPS on fibrin morphology. LPS, lipopolysaccharide. ∗, outlier measurement.

the preservation of erythrocyte shape and stability, modifications of the phospholipid bilayer can affect deformability. In 2010, Mohanty et al. (2010) reported that 15% of erythrocytes in AD were elongated and suggested that a potential link between changes in the erythrocyte proteome could contribute to AD pathology. Deformability and impaired reformability, particularly after erythrocytes have moved through capillaries, reduces their oxygen carrying capacity, potentiating transient states of hypoxia. **Figure 6** illustrates erythrocyte deformability as well as loss of membrane integrity in typical blood samples from individual's diagnosed with AD. It is clear to see how the loss of the typical bi-concave structure of these erythrocytes would impair their physiological functioning.

## NITRIC OXIDE REGULATION

The dysregulation of nitric oxide (NO) synthesis is prevalent in many diseased conditions (Cooke and Dzau, 1997; Kolios et al., 2004; Naseem, 2005; Karpuzoglu and Ahmed, 2006; Pacher et al., 2007). Recently, the physiological role of erythrocytes in regulating nitric oxide levels has also been investigated (Saldanha, 2016). Due to the consequences of chronic systemic inflammation and pathological coagulation, such as erythrocyte deformability and lysis, it is plausible that aberrant NO regulation could occur and causatively contribute to AD pathogenesis.

NO is synthesized by three isoforms of nitric oxide synthases. Although NO has vasoactive, immunological and neurophysiological functions, it can be neurotoxic at high concentrations and has been implicated in neurodegenerative diseases (Law et al., 2001). In this regard, erythrocytes have an important role in vascular function. Normally, erythrocytes regulate hemostasis by balancing oxygen delivery and NO scavenging and production. More specifically, hemoglobin (Hb) compartmentalization and encapsulation reduces an erythrocyte's ability to scavenge NO. However, in various diseases hemolysis introduces cell-free Hb into circulation, increasing NO scavenging and inducing hypertension (Helms et al., 2018). A decrease in NO may lead to a reduced ability to learn and

memorize due to impairment of long-term potentiation, since NO is responsible for the synaptic efficiency of pre-synaptic glutamatergic neurons (Panthi et al., 2018). Hypoperfusion of the brain and increased oxidative stress within the vasculature are common phenomena in AD (Edwards and Rickard, 2007) and some researchers have reported abnormalities in NO synthase expression as early symptoms of cognitive impairment and AD (Lüth et al., 2002; Malinski, 2007).

Furthermore, NO can exert its neurotoxic effects through three main mechanisms. Firstly, NO is a free radical which can react with superoxide anions to produce another damaging radical, peroxynitrite (Eliasson et al., 1999). This can produce significant oxidative stress which leads to oxidative damage and eventually neuronal death. Secondly, NO causes nitrosylation in a variety of proteins such as glyceraldehyde-3 phosphate dehydrogenase (GADPH) and protein kinase C (PKC), inhibiting their functioning. Lastly, NO can impair glycolysis and the latter energy manufacturing by potentiating ADP-ribosylation of GADPH (Law et al., 2001) as well as binding aconitase and thereby inhibiting electron transfer in the electron transport chain.

Together, pathological coagulation and altered nitric oxide regulation can enhance the propensity of spontaneous clot formation in cerebral arteries and arterioles. These clots are more resistant to degradation, due to amylodiogenesis of fibrin fibers in the presence of ferric irons, and may thus potentiate transient periods of hypoxia in the brain, promoting the occurrence of MD.

# CONCLUSION

While numerous hypotheses for the pathogenesis of AD exist (extensively reviewed by Law et al., 2001; Hardy and Selkoe, 2002; Swerdlow, 2007; Armstrong, 2011;

Dong et al., 2012; Barage and Sonawane, 2015; Selkoe and Hardy, 2016; Area-Gomez and Schon, 2017; Makin, 2018), the IDDM hypothesis (illustrated in **Figure 7**) recognizes that a combination of a dormant blood or tissue microbiome can be re-activated by unliganded iron. This in turn can induce shedding of highly inflammagenic LPS and LTA molecules, from which all known sequlae of AD can be seen to follow.

As reviewed here, dysbiotic bacterial communities from the GI tract or microbes from other sources of infection, translocate into normally sterile tissues and blood to establish a dormant blood microbiome. Characteristic of AD (and many other chronic, inflammatory diseases; Kell, 2009) is iron dysregulation, that amongst many other affects, causes microbial reactivation. Iron dysregulation, specifically in the form of elevated SF levels, which can causatively modulate oxidative damage, the formation of amyloidogenic fibrinogen and the aforementioned microbial reactivation as part of AD pathogenesis. Microbes contribute to amyloidogenic formations, inflammagen shedding, pathological clotting and systemic as well as neuro-inflammation. Various changes to the hematological system ensue, increasing the risk of pathological coagulation and transient hypoxic events in the AD brain. Increasing BBB permeability and glial cell activation initiate the slow inflammatory progression of AD, in combination with loss of synaptic plasticity and neuronal death. While AD is a disease too complex to fit to any particular model, the IDDM hypothesis highlights several causative role players which can be targeted for early diagnosis and/or for prevention of disease progression.

#### REFERENCES


# CONSENT FOR PUBLICATION

All authors have read the paper and agree that it can be published.

#### AUTHOR CONTRIBUTIONS

LP wrote the paper. DK co-wrote the paper and edited the paper. EP co-wrote the paper and is the study leader.

#### FUNDING

Funders include the Biotechnology and Biological Sciences Research Council (Grant No. BB/L025752/1) as well as the National Research Foundation (NRF) of South Africa (91548: Competitive Program) and the Medical Research Council of South Africa (MRC) (Self-Initiated Research Program) for supporting this collaboration.

#### ACKNOWLEDGMENTS

This is paper 18 in the series "A dormant blood microbiome in chronic, inflammatory disease". We thank the Biotechnology and Biological Sciences Research Council, the National Research Foundation (NRF) of South Africa and the Medical Research Council (MRC) of South Africa, for financially supporting this collaboration.







**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 Pretorius, Kell and Pretorius. 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.

# Kinetic Modeling of pH-Dependent Oxidation of Dopamine by Iron and Its Relevance to Parkinson's Disease

Yingying Sun<sup>1</sup> , A. Ninh Pham<sup>1</sup> , Dominic J. Hare2,3 and T. David Waite<sup>1</sup> \*

<sup>1</sup> Water Research Centre, School of Civil and Environmental Engineering, The University of New South Wales, Sydney, NSW, Australia, <sup>2</sup> Atomic Pathology Laboratory, Melbourne Dementia Research Centre at the Florey Institute of Neuroscience and Mental Health and The University of Melbourne, Parkville, VIC, Australia, <sup>3</sup> Department of Clinical Pathology, The University of Melbourne, Parkville, VIC, Australia

Parkinson's disease is the second most common neurodegenerative disease. While age is the most significant risk factor, the exact cause of this disease and the most effective approaches to mitigation remain unclear. It has long been proposed that dopamine may play a role in the pathology of Parkinson's disease in view of its ability to generate both protein-modifying quinones such as aminochrome and reactive oxygen species, especially in the presence of pathological iron accumulation in the primary site of neuron loss. Given the clinically measured acidosis of post-mortem Parkinson's disease brain tissue, the interaction between dopamine and iron was investigated over a pH range of 7.4 to 6.5 with emphasis on the accumulation of toxic quinones and generation of reactive oxygen species. Our results show that the presence of iron accelerates the formation of aminochrome with ferrous iron (Fe[II]) being more efficient in this regard than ferric iron (Fe[III]). Our results further suggest that a reduced aminochrome rearrangement rate coupled with an enhanced turnover rate of Fe[II] as a result of brain tissue acidosis could result in aminochrome accumulation within cells. Additionally, under these conditions, the enhanced redox cycling of iron in the presence of dopamine aggravates oxidative stress as a result of the production of damaging reactive species, including hydroxyl radicals.

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Ana Virel, Umeå University, Sweden Luigi Bubacco, Università degli Studi di Padova, Italy

> \*Correspondence: T. David Waite d.waite@unsw.edu.au

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 02 August 2018 Accepted: 02 November 2018 Published: 26 November 2018

#### Citation:

Sun Y, Pham AN, Hare DJ and Waite TD (2018) Kinetic Modeling of pH-Dependent Oxidation of Dopamine by Iron and Its Relevance to Parkinson's Disease. Front. Neurosci. 12:859. doi: 10.3389/fnins.2018.00859 Keywords: pH, iron, dopamine, aminochrome, oxidative stress, Parkinson's disease

# INTRODUCTION

Parkinson's disease is the second most common neurodegenerative disorder after Alzheimer's disease (Guttmacher et al., 2003). Even though the exact cause of Parkinson's disease is still unknown, two main pathological indicators are observed in the post-mortem brain; namely (i) loss of dopamine-producing neurons in the substantia nigra pars compacta (SNc; Kordower et al., 2013) and (ii) the widespread deposition of amyloid-like Lewy bodies rich in α-synuclein both in the SNc and other brain regions where overt neurodegeneration is not observed (Surmeier et al., 2017).

Accumulation of iron (Fe) in the SNc is an additional pathological feature common to all forms of Parkinson's disease (Ayton and Lei, 2014) that appears to precede the onset of clinical symptoms (Berg et al., 2015). Unlike synucleinopathy, brain Fe accumulation beyond that observed in normal aging is an early-stage event (He et al., 2015) restricted to the degenerating nigrostriatal pathway in Parkinson's disease (Wang et al., 2016) with this association supporting a potential causative role of Fe in neuron loss. In healthy neurons, Fe uptake is primarily via transferrin receptor

**68**

mediated endocytosis, which releases labile Fe into the cytosol for immediate distribution to various organelles and Fe-storage and regulatory proteins. In dopaminergic neurons a small amount of Fe is bound to neuromelanin, a biopolymer formed during the oxidation of dopamine (DA). The labile iron pool is thought to be transient and consists primarily of both ferrous (Fe[II]) and ferric (Fe[III]) species bound to low molecular weight ligands such as citrate and ATP (Double et al., 2002; Hare et al., 2013). While characterization of the labile Fe pool in a living system, particularly human tissue, remains an analytical challenge (New et al., 2018), it has been argued that an increase in reactive Fe in the labile iron pool may contribute to generation of reactive oxygen species (ROS) in Parkinson's disease at levels above those that can normally be attenuated by endogenous antioxidant mechanisms (Ward et al., 2014). Biochemically, Fe-mediated generation of ROS is primarily via traditional Fenton/Haber-Weiss chemistry (Equations 1, 2) where cytosolic iron cycles between Fe[II] and Fe[III] to produce a range of harmful oxidants, including superoxide (O•− 2 ), hydrogen peroxide (H2O2) and hydroxyl radicals (•OH; Graham et al., 1978; Halliwell and Gutteridge, 1984; Segura-Aguilar et al., 2014; Sun et al., 2016):

$$\text{Fe[III]} + \text{O}\_2\text{"}^\bullet \rightarrow \text{Fe[II]} + \text{O}\_2 \tag{1}$$

$$\text{Fe[II]} + \text{H}\_2\text{O}\_2 \rightarrow \text{Fe[III]} + \text{^\bullet OH} + \text{OH}^-\tag{2}$$

The Fenton/Haber-Weiss reaction is not, however, unique to dopaminergic neurons and normal Fe accumulation with age occurs in other neuroanatomical regions (Acosta-Cabronero et al., 2016) without causing generalized cell loss. Numerous antioxidant mechanisms, including superoxide dismutase 1 (SOD1), reduced glutathione (GSH) and glutathione peroxidase, and vitamin E-mediated scavenging of O•− 2 , H2O<sup>2</sup> and •OH, attenuate potential ROS-induced neurotoxicity from gradual increase in Fe levels, suggesting that dysfunction of attenuating mechanisms is involved in Fe-mediated dopaminergic cell death in Parkinson's disease (Zecca et al., 2004).

An emerging theory, supported by initial observations of Fe dysregulation dating back nearly 100 years (Lhermitte et al., 1924), suggests that abnormal interactions between Fe and DA represent a biochemical mechanism unique to the microchemical environment of vulnerable neurons (Hare et al., 2014). Healthy dopaminergic neurons have particularly high endogenous oxidative load, owing to their extensive and complex axonal network, large soma and high metabolic output (Blesa et al., 2015). Further, cytosolic DA present in the mid-µM range in these neurons (Mosharov et al., 2009) gives rise to a pathway of Fe-mediated ROS generation independent of Fenton/Haber-Weiss chemistry (Hare and Double, 2016). Dopamine-derived quinones, such the DA o-quinone (DAQ) and aminochrome (DAC) can also be produced via the Fe-driven generation of the transient precursor DA semiquinone (DA•−) intermediary (Hare and Double, 2016; Sun et al., 2018b). It is unclear if endogenous antioxidant enzymes are capable of attenuating damage from DAquinone production, though one proposed mechanism involves polymerization as the dark pigment, neuromelanin (Zhang et al., 2012).

As a major neurotoxic metabolite of DA, DAC has been used as a preclinical model compound to examine neurotoxicity in view of its apparent ability to cause indiscriminate neuronal damage, including mitochondrial dysfunction (Herrera et al., 2016; Segura-Aguilar and Huenchuguala, 2018), likely via lipid peroxidation and disruption of membrane integrity. It has been proposed that DAC can induce aggregation of α-synuclein and eventual Lewy body deposition through formation of a DAC-synuclein adduct (Berman and Hastings, 1999; Conway et al., 2001; Bianco et al., 2002; Norris et al., 2005), and αsynuclein mRNA has a predicted iron-response element in the 5 ′ -untranslated region similar to the Fe-storage protein ferritin (Friedlich et al., 2007), suggesting this hallmark protein of Parkinson's disease also plays a role in Fe-mediated toxicity within vulnerable neurons. Previous studies also reported that the presence of DA quinones may also aggravate oxidative stress as a result of the inactivation in the electron transport chain of mitochondrial complexes I and III and the subsequent leakage of electrons from the respiratory chain (Stokes et al., 1999; Adam-Vizi, 2005; Lin and Beal, 2006; Gautier et al., 2008).

Much of our understanding of Fe-mediated DA oxidation has come from ex vivo studies where the chemical environment remains relatively constant, or in vivo investigations using cell culture or simple animal models of parkinsonism (Jiang et al., 2013; Panicker et al., 2015; Sampson et al., 2016). While useful and of most relevance, in vivo experiments generally provide insight into the overall consequence of the whole process with limited insights regarding the exact pathway, or pathways, of DA transformation. As such, it is unlikely that location of the most toxic intermediates or their relative concentrations will be determined in such studies with resultant restrictions in identifying the most efficacious therapeutic strategies targeting toxic DA metabolites. While previous work has firmly established that aberrant reactions between Fe and DA give rise to increased levels of free radicals and oxidative stress markers (Hermida-Ameijeiras et al., 2004; Jiang et al., 2013), comparatively little attention has been paid to the quantitative study of the DA intermediates that are directly implicated in causing neuronal dysfunction. Mitochondrial complex I inhibition by DA quinones and subsequent mitochondrial dysfunction has been identified as a possible molecular basis of neurodegeneration in Parkinson's disease (Schapira, 1994) and resulting intraneuronal acidosis is likely to decrease cytoplasmic pH (Balut et al., 2008) and influence the kinetics of Fe-mediated DA oxidation in vivo. Further, while the frequently observed decrease in pH of postmortem brain tissue in the numerous tissue biobanks may result from ex vivo handling (Hare et al., 2012), physiological effects of

**Abbreviations:** 6-OHDA, 6-hydroxydopamine; AFO, amorphous iron oxide; DA, dopamine (1-amino-2-(3,4-dihydroxyphenyl)ethane); DA•−, semiquinone radical; DAC, dopaminochrome; DAL, leukoaminochrome; DAQ, dopamine-oquinone; DHI, 5,6-dihydroxyindole; ECF, extracellular fluid; Fe[II], inorganic ferrous ion; Fe[III], inorganic ferric ion; Fe[III]DA, mono-complex; Fe[III]DA2, bis-complex; Fe[III]DA3, tris-complex; Fe[III]<sup>I</sup> , total inorganic Fe[III]; H2O2, hydrogen peroxide; LMCT, ligand to metal charge transfer; MQ, Milli-Q water; O •− 2 , superoxide; •OH, hydroxyl radical; ROS, reactive oxygen species; SNc, substantia nigra pars compacta; TOR, turnover rate.

cardiorespiratory failure, cerebrovascular accident and end-stage neurodegenerative diseases generally result in lower brain tissue pH ranging from 7.0 to ∼ 6.0 (Hardy et al., 1985; Yates et al., 1990; Harrison et al., 1995; Monoranu et al., 2009; Genoud et al., 2017). Mitochondrial dysfunction, which is a cardinal feature of neurodegenerative disease (Lin and Beal, 2006), also decreases tissue pH via leakage of protons into the cytosol (Brand and Nicholls, 2011). Thus, a reduction in brain tissue pH as a result of disease progression may have a substantial role in accelerating the progression of Fe-mediated DA oxidation in Parkinson's disease.

To demonstrate the relevance of interplay between Fe and DA to Parkinson's disease under acidified conditions mimicking predicted effects of disease-specific pre-mortem and non-specific post-mortem changes, we examined interactions between DA and Fe in vitro at a range of pH values–from pH 7.4, the typical pH of extracellular fluid (ECF) to pH 6.5, which is the reported average pH value of human Parkinson's disease tissue from the Sydney Brain Bank (Genoud et al., 2017) (though the specific location within the brain from which these samples were drawn was not reported). In the majority of previous studies, biochemical assays were undertaken at pH 7.4 as opposed to intraneuronal values that are typically around 6.96 to 7.05. This is a major limitation given that the transformation and fate of a range of biochemical compounds such as DA is strongly dependent upon pH. As Femediated DA oxidation is an aerobic process, we also examined the effects of dissolved O<sup>2</sup> concentration within aqueous systems with respect to both the kinetics of Fe[II] oxidation and the production of reactive DA intermediates. In order to facilitate accurate and precise determination of reactive and unstable quinone intermediates, these investigations of the influence of pH and oxygen concentration on Fe-mediated DA oxidation have been performed ex vivo. These fundamental studies are critical for understanding Fe-DA reaction kinetics and set the stage for future studies where the techniques described are applied to the study of both the formation of neurotoxic DA metabolites in the Parkinson's disease brain and to the new therapies designed to prevent abnormal reactions of Fe within dopaminergic neurons that account for the changing pH of a degenerating dopaminergic neuron.

## MATERIALS AND METHODS

#### Chemical and Reagents

All analytical grade chemicals were purchased from Sigma-Aldrich (Castle Hill, Australia) or as otherwise stated. All solutions were prepared using 18 M·cm ultrapure Milli-Q water (MQ; Merck-Millipore, Bayswater, Australia). All glassware was acid washed in 5% HCl (v/v) for at least 1 week, then thoroughly rinsed in MQ before use. Stock solutions were stored in amber glass bottles at 4◦C prior to use. To mimic the intracellular environment and focus on the effect of pH on iron catalyzed oxidation of DA, all experiments were performed in a light-free environment at a controlled temperature of 22 ± 0.6◦C. This temperature, rather than a physiological temperature of 37◦C was selected to allow direct comparison with previous work at pH 7.4. Investigation of the effect of temperature on the processes of interest in this work would be worthwhile though it should be

of added iron at pH 6.5, pH 7.0, and pH 7.4 in 0.1 M NaCl solution. Error bars are standard errors from triplicate measurements and solid lines represent the model fit.

recognized that thermodynamic and kinetic data for many of the key reactions underpinning these processes has been obtained by other investigators in the temperature range 20–25◦C. Details of the preparation of stock and working solutions can be found in **Supplementary Materials**.

#### Preparation of Buffer Solutions

In general, despite often being considered "chemically inert" (Yu et al., 1997; Thiel et al., 1998), buffers such as 3-(N-morpholino)propanesulfonic acid (MOPS), 2-(N-morpholino)ethanesulfonic acid (MES) and 4-(2 hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES) may still exert an influence on the spectroscopic measurement and/or formation of different target substances. As such, to eliminate any buffer induced experimental artifacts, studies were undertaken in 0.1 M NaCl, 2 mM NaHCO<sup>3</sup> solutions containing 10 mM of MOPS, MES or HEPES by taking into consideration their pH control range and the influence they may exert on the measurement. Specifically, MOPS was used for the measurement of the generation of Fe[III]DA2, DAC and H2O<sup>2</sup> over pH 6.5–7.4 (**Figures 1**–**4** and **Supplementary Figure 5**) as it does not exhibit spectrum broadening effects. For the measurement of Fe[II], MES was used at pH 6.5, while HEPES was used at both pH 7.0 and 7.4 (**Figure 4** and **Supplementary Figure 7**) as a result of the pH control range of these buffers and the negligible buffer facilitated reduction of Fe[III] in the presence of ferrozine (FZ; 4-[3-pyridin-2-yl-5-(4 sulfophenyl)-1,2,4-triazin-6-yl]benzenesulfonate) at pHs below 7.0.

#### Control of Dissolved O<sup>2</sup> Content in Buffer Solutions

For experiments using variable O<sup>2</sup> concentrations (0, 2.5, and 5% dissolved O2), buffer solutions prepared above were sparged with special gas mixtures (297 ± 6 ppm CO<sup>2</sup> with or without 5% O2, in Ar; BOC Gases, Preston, Australia) for at least 1 h before the addition of DA and Fe (see Control of Dissolved O<sup>2</sup> Content in

solutions containing 20µM DA and 5µM Fe[II] (B) and 5µM Fe[III] (C) at pH 6.5, pH 7.0, and pH 7.4. Error bars are standard errors from triplicate measurements and solid lines represent the model fit.

Buffer Solutions). Constant O<sup>2</sup> concentration was maintained by continuous sparging with the gas mixture over the course of the entire experiment.

#### Analytical Methods

#### pH Measurements

All pH measurements were made using a Hanna Instruments HI9025 pH meter (Keysborough, Australia) with a glass electrode and Ag-AgCl reference electrode. The pH meter and electrode

solutions containing 20µM DA and 5µM Fe[II] (A) and 5µM Fe[III] (B) at pH 6.5, pH 7.0, and pH 7.4. Error bars are standard errors from triplicate measurements and solid lines represent the model fit.

were calibrated prior to each experiment using NIST-traceable buffer solutions (pH 4.01, 7.01, and 10.01).

#### Ferrous Iron Quantification

Quantitative measurements of Fe[II] were made using the modified FZ method as the reduction of DA bound iron occurs in the presence of FZ at low pH (Garg et al., 2013). The UV absorbance of Fe[II]FZ<sup>3</sup> (both from inorganically and organically bound Fe[II]) was monitored at 562 nm with baseline correction at 690 nm. The concentration of total Fe[II] was then calculated per the following (Garg et al., 2013):

$$\text{[Fe[III]]} = \frac{\left(A\_{562} - \epsilon\_{\text{Fe[III]}} \times \text{[Fe]}\_T \times l\right)}{\left(\varepsilon\_{\text{Fe[II]}} - \varepsilon\_{\text{Fe[III]}}\right) \times l} \tag{3}$$

where A<sup>562</sup> is the recorded absorbance at 562 nm, εFe[II] and εFe[III] are the molar absorption coefficients of the Fe[II]FZ complex at 562 nm arising from the presence of Fe[II] and Fe[III], respectively. [Fe]<sup>T</sup> is the total iron concentration, and l is the path length (10 cm).

#### Iron-Dopamine Complex Measurements

The concentration of Fe[III]DA<sup>2</sup> was determined spectrophotometrically by measuring the absorbance at

FIGURE 4 | Oxidation of 5µM Fe[II] at pH 6.5, pH 7.0, and pH 7.4 in air-saturated (21% O2) 0.1 M NaCl solutions containing 10µM DA (A); oxidation of 5µM Fe[II] at pH 7.0 in the presence of 2.5, 5, and 21% O<sup>2</sup> (B); and the formation of FeIIIDA<sup>2</sup> (C) in 0.1 M NaCl solutions in the same O<sup>2</sup> conditions. Formation of FeIIIDA<sup>2</sup> in 0.1 M deoxygenated NaCl solutions containing 5µM Fe[III] and 50µM DA at pH 6.5, pH 7.0, and pH 7.4 (D). Error bars are standard errors from triplicate measurements and solid lines represent the model fit. The experimental data at pH 7.4 was taken from Sun et al. (2016).

580 nm with baseline correction at 850 nm. Calibration curves for quantification of the concentration of Fe[III]DA<sup>2</sup> were constructed from measurements performed under anoxic conditions (Sun et al., 2016). Given the low concentrations and weak absorptivity of the mono-complex, the effect of this species on the measurement was considered negligible. It should be noted, however, that the Fe[III]DA complex is a precursor to the formation of the highly neurotoxic 6-OHDA quinone (Hare and Double, 2016). The molar absorptivity of Fe[III]DA<sup>2</sup> was determined to be 3,121 M−<sup>1</sup> cm−<sup>1</sup> which was within 7% of the previously published value (Sever and Wilker, 2004). As a result of the slightly acidic pH and DA concentration used in this study, the extent of formation of the tris-complex should be negligible (Kowalchyk et al., 1995; Sun et al., 2018a). Even though the spectrum of DAC can overlap with that of the Fe[III]DA<sup>2</sup> complex at 580 nm, the influence of DAC was not considered given the low concentration present (nominally <1µM) under all conditions investigated herein (Pezzella et al., 1997) and the small molar absorptivity of DAC at 580 nm (ε<sup>580</sup> = 439 M−<sup>1</sup> cm−<sup>1</sup> ).

#### Hydrogen Peroxide Measurements

The H2O<sup>2</sup> formed from Fe-mediated DA oxidation was quantified using the modified DPD method (Bader et al., 1988; Voelker and Sulzberger, 1996; Sun et al., 2016). Briefly, 1 mM DTPA was added to quench H2O<sup>2</sup> generation at each time point in each assay (Sun et al., 2016). To eliminate interference from subsequent H2O<sup>2</sup> generation and/or consumption as a result of the presence of high Fe[II] concentrations, 500µM BPY was added prior to DTPA addition (Voelker and Sulzberger, 1996).

Quantitative data were interpolated by linear regression analysis of absorbance at 551 nm by increasing concentrations of H2O<sup>2</sup> in 0.1 M NaCl with 60µM DPD and 500 U/L HRP added, including a standard blank. No apparent interferences arising from the presence of moderate Fe[III] (5µM) and DA (30µM) were identified (**Supplementary Figure 1**). In order to enhance the accuracy of the measurements and remove potential interference arising from DA residuals on H2O<sup>2</sup> absorbance measurements, the corresponding amount of DA, which was exactly the same as that used in a specific experiment set, was added to develop calibration curves for the experiments containing high concentrations of Fe[II].

#### Aminochrome Quantification

The concentration of DAC was determined by measuring absorbance at 475 nm with baseline correction at 850 nm (**Supplementary Figure 2**; Herlinger et al., 1995; Pham and Waite, 2014). The molar absorptivity derived from the calibration curves indicated that the precursor DAQ was unlikely to exert any influence on the measurement of DAC while most phenolic organics, including DA (**Supplementary Figure 3**) and 5,6-dyhydroxyindole (DHI), only absorb at around 280 nm (Il'ichev and Simon, 2003). As such, the influence arising from the rearrangement products of DAC should be minimal at 475 nm. Absorbance at a particular wavelength is the sum of the absorbances contributed from different species, thus spectral overlap may cause false-positive measurements. Given the coexistence of Fe[III]DA<sup>2</sup> and DAC in solutions containing iron and DA, the concentration of each species was determined by solving the linear equations as described previously (Sun et al., 2018b); i.e.,

$$A\_{475} = \left(\varepsilon\_{475}^{\text{Fe}^{\text{[III]}}\text{DA}\_2} \mathbf{C}\_{\text{Fe}^{\text{[III]}}\text{DA}\_2} \mathbf{l}\right) + \left(\varepsilon\_{475}^{\text{DAC}} \mathbf{C}\_{\text{DAC}} \mathbf{l}\right) \tag{4}$$

where A is the total absorbance at 475 nm, ε j i is the molar absorptivity of species j at wavelength i, C<sup>j</sup> is the concentration of species j and l is the pathlength (10 cm). The concentration of Fe[III]DA<sup>2</sup> was initially quantified based on the absorbance at 580 nm as DAC absorbed negligibly at this wavelength.

Calibration curves of DAC were developed by adding different concentrations of the freshly prepared DAC working solution into the air-saturated MOPS buffer solutions at pH 7.0 (Sun et al., 2018a). The molar absorptivity of DAC calculated in this study (ε475 nm = 3,245 M−<sup>1</sup> cm−<sup>1</sup> ) is similar to that reported by Segura-Aguilar and Lind (1989) (ε<sup>475</sup> = 3,085 M−<sup>1</sup> cm−<sup>1</sup> ) and Pham and Waite (2014) (ε<sup>475</sup> = 3,281 M−<sup>1</sup> cm−<sup>1</sup> ).

#### Speciation and Kinetic Modeling

The pH-dependent distributions of DA, Fe[III] and Fe[II] species were determined using the program Visual Minteq (Gustafsson, 2005). Details of the distributions are shown in **Supplementary Figure 4** with stability constants used in this study provided in **Supplementary Table 1**.

The kinetic model developed to describe the experimental data at pH 6.5, 7.0, and 7.4 was implemented using the software Kintek Explorer (Johnson et al., 2009). Specifically, the kinetic model is a set of reactions describing the key processes involved in the interplay between Fe and DA. To apply the model for the quantification of the time-dependent transformation of reactants, intermediates and products, the rate constant for each key process was either adopted from previous work or fitted in this study. The "goodness of fit" was judged by the ability of the reaction set used (and the associated set of coupled differential equations representing the rate expressions for each reaction) to describe the time-dependent transformation of a range of substances. If the model provided a poor description of the data, it indicated either a flaw in the reaction set or rate constant(s) used. In addition to the data collected in the current work, portions of the experimental data collected at pH 7.4 described by Sun et al. (2016) were used to complement this dataset and fully elucidate the effect of pH on DA oxidation. Given the critical role of dissociation of DA bound Fe at lower pH and the improved constraints provided by analysis of DAC, slight amendments were applied to the kinetic model previously developed for pH 7.4 (Sun et al., 2016).

#### RESULTS

#### Development and Rationale of Experimental Model

Given the relatively complicated model developed in this study, several intermediates were measured in order to better constrain the rate constants used herein. Briefly, (i) generation of DAC and H2O<sup>2</sup> (shown in **Figures 1**–**3**) are used to constrain the rate constants for transformation of DA both in the absence and presence of iron including those for the oxidation of DA and leukoaminochrome (DAL) and cyclization of DAQ; (ii) decay of Fe[II] and formation of Fe-DA complexes in the presence of O<sup>2</sup> (shown in **Figures 4A–C**) are used to constrain the rate constant for the DA-induced transformation of Fe; and (iii) formation of Fe[III]-DA complexes (shown in **Figure 4D** and **Supplementary Figure 5**) in the absence of O<sup>2</sup> is used to constrain the rate constants for mono-complex formation, DAinduced reductive dissolution of ferrihydrite (amorphous ferric oxyhydroxide; or AFO) as well as the dynamic equilibrium between the mono- and bis-complexes. Details of the various reactions hypothesized to play a role are presented in **Tables 1**–**3**. Sensitivity analysis was used to determine the relative importance of the proposed reactions. Specifically, the greater the variation of the relative residuals that occurred with the change in magnitude of the rate constant, the more influential the reaction is. The lowest point or the "shift point" shown in the sensitivity analysis represents the optimal rate constant for the overall model applied. To simplify the applied model, rate constants were maintained consistent with those deduced in our previous study (Sun et al., 2016) with the exception of those with significant sensitivity to change of pH.

The sensitivity of the model to changes in individual rate constant values, defined as the relative residual,r, was determined using the program Kintecus (Ianni, 2003) combined with a Visual Basic for Applications (VBA) program. Note that the relative residual is defined as:

$$r = \frac{1}{n} \sum\_{i=1}^{n} \frac{|MP\_i - ED\_i|}{ED\_i} \tag{5}$$

where MP<sup>i</sup> is the model prediction, ED<sup>i</sup> is the experimental data at the same condition and time interval and n represents the total number of measured data points.

As shown in **Supplementary Figures 6A,B**, significant influence of DAQ cyclization and DAC decay on the transformation of DA was observed at the two pH values investigated here (the results of previous studies at pH 7.4 are provided by Sun et al., 2016). The most sensitive point of the relative residual r increases in line with pH. This is in agreement with the proposed rate constants (**Tables 1**–**3**) in the main text as deprotonation is generally the prerequisite for DAQ cyclization and DAC decay. In contrast, a relatively insensitive relative residual r is observed below the deduced upper rate constant (k<sup>5</sup> = 5.3 × 10<sup>6</sup> ) for the redox exchange between DAQ and DAL (Land et al., 2003; Sun et al., 2016), especially at pH 7.0 (**Supplementary Figure 6C**). As such, the rate constant of the TABLE 1 | Modeled reactions and rate constants for the autoxidation of DA at pH 6.5, 7.0, and 7.4.


<sup>a</sup>Value modified from that used in Sun et al. (2016). <sup>b</sup>Rate constant taken from Sun et al. (2016). <sup>c</sup>The rate constant proposed for reaction 9 incorporates the influence of Fe. DA, dopamine; DA•−, semiquinone radical; DAQ, dopamine-o-quinone; DAC, aminochrome; DAL, leukoaminochrome; and O•− 2 , superoxide. Refs: (1) Pham and Waite (2014); (2) Borovansky et al. (2006); (3) Land et al. (2003); (4) Zafiriou (1990) and (5) Sun et al. (2016).

TABLE 2 | Modeled reactions and rate constants for Fe[III]-catalyzed oxidation of DA at pH 6.5,7.0 and 7.4.


<sup>a</sup>Modified value of the model developed at pH 7.4 in Sun et al. (2016). <sup>b</sup>Rate constant taken from Sun et al. (2016). DA, dopamine; DA•−, dopamine semiquinone radical; DAQ, dopamine-o-quinone; O•− 2 , superoxide; Fe[III], inorganic ferric ion; Fe[III]<sup>I</sup> , total inorganic Fe[III]; AFO, ferrihydrite; and Fe[II], inorganic ferrous ion. Refs: (6) Pham et al. (2006); (7) Blesa and Matijevi (1989); (8) Rose and Waite (2003); (9) El-Avaan et al. (1997) and (10) Rush and Bielski (1985).

redox exchange reaction at both pH 6.5 and 7.0 was consistent with these previously reported findings.

As shown in **Supplementary Figure 6D**, the DA-induced dissolution of AFO is an important process as the relative residual r is very sensitive to the change in the magnitude of the rate constant. A slight increase in the shift point with increase in pH is observed with this result in accord with the proposed rate constants applied. However, compared with that of DA-induced dissolution of AFO, the reductive dissolution after the adsorption of DA onto the surface of AFO (Reaction 12, **Table 2**) is relatively insensitive at both pH values used (**Supplementary Figure 6E**), indicating that reductive dissolution is not as important as DAinduced dissolution of AFO at each pH value investigated herein. As shown in **Supplementary Figures 6F,G**, the formation of both Fe[III]DA (Reaction 13, **Table 2**) and Fe[II]DA (Reaction 24, **Table 3**) are key reactions in this study as can be seen from the relative sensitivities of the relative residuals for these reactions. The rate constant for the formation of the bis-complex from Fe[III]DA with another DA molecule (Reaction 14, **Table 2**) is assumed to be similar to the rate


<sup>a</sup>Modified value of the model developed at pH 7.4 in Sun et al. (2016). b rate constant taken from Sun et al. (2016). DA, dopamine; DA•−, dopamine semiquinone radical; O•− 2 , superoxide; Fe[III], inorganic ferric ion; Fe[II], inorganic ferrous ion; H2O2, hydrogen peroxide; and •OH, hydroxyl radical. Refs: (11) González-Davila et al. (2005).

constant for water-loss from Fe(OH)(H2O)2<sup>+</sup> 5 of 4.50 × 10<sup>5</sup> M−<sup>1</sup> s −1 (Blesa and Matijevi, 1989) in view of the fact that the replacement of a coordinated H2O by the additional DA molecule is generally faster than the formation of the mono-complex (Ludwig et al., 1995). The results of sensitivity analysis shown in **Supplementary Figure 6H** indicate that this assigned value should be reasonable for the sensitivity of relative residual r. Similar to the formation of the Fe-DA complexes, dissociation of these complexes is also of great significance given the considerable sensitivity of the rate constants over several orders of magnitude (**Supplementary Figures 6I–K**). In contrast to increase in the optimal value on decrease in pH (**Supplementary Figure 6I**), the increase in the dissociation rate constant shown in **Supplementary Figure 6J** is much less significant. This supports the supposition that dissociation is generally important at low pH with the bis-complex much more stable in view of the iron sequestration. As shown in **Supplementary Figure 6L**, relative insensitivity of the relative residual r is evident for the deduced rate constant (3.7 × 10<sup>5</sup> M−<sup>1</sup> s −1 ) for the reaction between superoxide and AFO (Reaction 19, **Table 2**). As such, in order to simplify the model, the rate constant of this reaction at high pH is considered to be the same as that proposed previously (Sun et al., 2016). However, a value for this rate constant of one order of magnitude lower is deduced from model fitting in this study at pH 6.5 with this lower value possibly a result of the increased proportion of HO• 2 at this lower pH and the subsequent reduced electrostatic attraction between superoxide and the AFO surface.

Theoretically, change in pH would typically result in variation of the Fe[II] oxidation rate in the presence of oxidants such as DA•−. However, as shown in **Supplementary Figure 6M**, considerable insensitivity of the relative residual is observed around the value proposed at physiological pH. As such, a consistent rate constant was used. Reduction of Fe[III] by DA•− is important at pH 6.5 given the significant decrease in relative residuals on variation of the rate constant (from 1 to 10<sup>10</sup>

M−<sup>1</sup> s −1 ; shown in **Supplementary Figure 6N**). Therefore, the reaction between Fe[III] and DA•− is considered in the proposed reaction scheme. The reduction of DA bound Fe[III] by DA•− is not as significant as the reduction of Fe[III] in view of its lower reduction potential. As such, reduction of organically complexed Fe[III] by DA•− is not considered in the reaction scheme.

#### Dopamine-Derived Free Radical Production Is pH Dependent

To assess how pH influences the generation of toxic DA metabolites, the autoxidation of DA over a range of pH was initially investigated with attention given to the accumulation of H2O<sup>2</sup> and DAC. As shown in **Figure 1**, spontaneous oxidation of DA is highly pH-dependent with 20µM DA producing H2O<sup>2</sup> at rates of 0.08, 0.12, and 0.24µM h−<sup>1</sup> at pH 6.5, 7.0, and 7.4, respectively. On the other hand, <0.1µM of DAC was produced at pH 7.0 in the absence of Fe (**Figure 2A**).

The oxidation of DA by O2:

$$\text{DA} + \text{O}\_2 \rightarrow \text{DA}^{\bullet-} + \text{O}\_2^{\bullet-} \tag{6}$$

is often considered to be a critical step in the production of H2O<sup>2</sup> and DAC. Firstly, DA•− can disproportionate to form DAQ which subsequently cyclizes to DAL. DAC and H2O<sup>2</sup> can then be produced either through the redox exchange between DAQ and DAL (Equation 7) or as a result of the direct oxidation of DAL (Equation 8), O•− 2 disproportionation (Equation 9) and/or the O•− 2 -mediated transformation of radicals such as DA•− (Equation 10; Hawley et al., 1967; Graham, 1978; Sun et al., 2016).

$$\text{DAQ} + \text{DAL} \rightarrow \text{DA} + \text{DAC} \tag{7}$$

$$\text{DAL} + \text{O}\_2 \rightarrow \text{ DAC} + \text{H}\_2\text{O}\_2\tag{8}$$

$$\text{O}\_2^{\bullet-} + \text{O}\_2^{\bullet-} \rightarrow \text{ H}\_2\text{O}\_2 + \text{O}\_2 \tag{9}$$

$$\text{O}\_2\text{"}^- + \text{DA}^{\bullet-} \rightarrow \text{ DAQ} + \text{H}\_2\text{O}\_2 \tag{10}$$

Sun et al. pH, Iron, and Dopamine Oxidation

Among the various reactions shown above, the direct oxidation of DA by O<sup>2</sup> to DA•− and O•− 2 (Equation 6) is expected to be rate limiting in view of the spin restriction inherent between DA and O<sup>2</sup> and, accordingly, exhibits a small rate constant (**Table 1**). With this in mind, any factors that influence this rate-limiting step (such as the presence of Fe[II] and Fe[III]; see Section Iron Accelerates Dopamine Oxidation) will alter the rate of formation of products arising from the oxidation of DA.

In general, the apparent oxidation rate of DA is determined by the contribution of different DA species. The pH-dependent formation of H2O<sup>2</sup> shown in **Figure 1** is attributed, at least partially, to the increase in the proportion of deprotonated DA ions on increase in pH; i.e.,

$$\begin{split} k\_{\rm app} &= \alpha\_{\rm H\_2DA} k\_{\rm H\_2DA} + \alpha\_{\rm HDA} k\_{\rm HDA^-} + \text{O}\_2\\ &+ \alpha\_{\rm DA^{2-}} k\_{\rm DA^{2-}} + \text{O}\_2 \end{split} \tag{11}$$

where kapp is the apparent oxidation rate constant, α<sup>i</sup> is the fraction of total DA species present as species i and k<sup>i</sup> is the intrinsic oxidation rate constant of species i. Previous work has suggested that the abstraction of a hydrogen atom from the mono-deprotonated form of DA is the rate limiting step in the oxidation of DA (Herlinger et al., 1995). As such, it is not unexpected that an increase in the proportion of deprotonated DA on increase in pH would give rise to the enhanced generation of both DA•− and O•− 2 via Equation 6, eventually resulting in the subsequent increase in the concentration of H2O2. Compared with the significantly enhanced production of H2O<sup>2</sup> in the presence of iron (shown in section Iron Accelerates Dopamine Oxidation), while continuously formed, the toxicity induced by the autoxidation of DA is unlikely to be significant given the presence of in vivo oxidant removal enzymes, such as superoxide dismutase 1 (SOD1) and the glutathione peroxidases (GPx), though there is evidence that both enzymes are dysfunctional in Parkinson's disease (Cardoso et al., 2017; Trist et al., 2017, 2018). Additionally, the presence of DA-derived quinones is expected to be negligible (as shown in **Figure 2A**).

#### Iron Accelerates Dopamine Oxidation

In order to quantify the rate of generation of ROS and reactive quinones arising from the interaction between Fe and DA over pHs ranging from those of the physiological neuronal cytosol to those that have experienced disease-related and post-mortem acidosis, the accumulation of DAC and the concentrations of the key Fenton reagents–Fe[II] and H2O2-were subsequently investigated.

In contrast to the negligible production of DAC in the absence of Fe, a substantially higher concentration was generated in the presence of 5µM iron at pH 7.0. In general, our results show that Fe[II] was more effective in catalyzing DAC production than was Fe[III] (**Figure 2A**). The initial rate of formation of DAC in the presence of both Fe[II] and Fe[III] increased significantly at pH 7.0 and 7.4 (**Figures 2B,C**). By contrast, the concentration of DAC plateaued in the latter stages of the 120 min assessment period at pH 7.4, more so in the presence of Fe[II].

Accordingly, labile Fe also accelerated the production of H2O2, again with Fe[II] being generally more efficient than Fe[III], particularly during the initial ∼10 min following addition of reagents (**Figure 3A**). While H2O<sup>2</sup> was continuously produced at pH 7.0 and 7.4 in the presence of both Fe species, at pH 6.5 only Fe[III] showed an increase in H2O<sup>2</sup> production though the concentration present at t = 120 min was only ∼10% of that produced at pH 7.4 (**Figure 3B**). For Fe[II], H2O<sup>2</sup> concentration remained stable after a subtle increase from t = 0–10 min.

Accompanying the rapid generation of both DAC and H2O<sup>2</sup> was a significant decrease in the concentration of Fe[II]. The apparent oxidation rate of Fe[II] significantly increased in the presence of DA. In the absence of DA, the concentration of Fe[II] halved over 2 h at pH 7.0 (**Supplementary Figure 7A**) while Fe[II] levels were near-totally depleted within 30 min in the presence of 10µM DA at the same pH (**Figure 4A**).

According to the rate law:

$$-\frac{d\,[\text{Fe[II]}]}{dt} = k[\text{Fe[II]}][\text{O}\_2] \tag{12}$$

the oxidation of Fe[II] depends on the concentration of dissolved O<sup>2</sup> present. As the O<sup>2</sup> concentration in brain tissue is typically in the range of 10–60µM (Koppenol and Butler, 1985; Ndubuizu and Lamanna, 2007), which corresponds to <5% O<sup>2</sup> saturation in aqueous solutions, the effect of O<sup>2</sup> (at 2.5 and 5% saturation) on the transformation of Fe[II] in the presence of DA was examined at pH 7.0. As expected, both the rate and extent of Fe[II] oxidation decreased with lower O<sup>2</sup> concentration, with only a 10% decrease in Fe[II] concentration at pH 7.0 in 5% O<sup>2</sup> at t = 2 h compared with a 40% decrease at 21% O<sup>2</sup> saturation (**Supplementary Figure 7B**). Oxidation of Fe[II] followed an exponential decay on addition of 10µM DA at pH 7.0 with initial Fe[II] exhausted after ∼70 min in 21% O<sup>2</sup> and 80–90% loss at 2 h for physiologically-relevant O<sup>2</sup> concentrations (**Figure 4B**). Accordingly, the initial formation of FeIIIDA<sup>2</sup> was strongly dependent on O<sup>2</sup> concentration with markedly lower rates observed at 2.5 and 5% O<sup>2</sup> (**Figure 4C**).

#### Dopamine Induces Iron Mobilization

As predicted by thermodynamic and kinetic data, the majority of Fe[III] should be present in an insoluble phase, even in the presence of 20µM DA. As such, the continuous generation of both DAC and H2O<sup>2</sup> in the presence of Fe[III] shown in **Figures 2C**, **3B** suggests that DA is capable of inducing the slow release of reactive soluble Fe species. Indeed, dissolution of precipitated Fe by DA may pose a potential risk in view of its contribution to the labile iron pool (Dixon and Stockwell, 2014) and subsequent DA oxidation. Thus, to understand the extent of DA-induced mobilization of Fe[III], the temporal change in concentration of Fe[III]DA<sup>2</sup> was measured over the pH range of 6.5 to 7.4. To eliminate potential confounding effects of O2-mediated transformation of both Fe and reducing radicals, Fe[III]DA<sup>2</sup> formation was investigated under anoxic conditions. As shown in **Figure 4D**, Fe[III]DA<sup>2</sup> formation increased gradually at pH 7.0 and 7.4 in the presence of high initial DA concentration (50µM) though not at pH 6.5. At this pH 6.5, Fe[III]DA<sup>2</sup> formation was ∼60% of that produced at circumneutral pH; a trend conserved at a lower DA concentration (20µM; **Supplementary Figure 5**).

Compared with pH 6.5, the increase in the overall concentration of FeIIIDA<sup>2</sup> at pH 7.0 and 7.4 can most likely be attributed to: (i) the reduced rate of LMCT resulting from the increased proportion of FeIIIDA<sup>2</sup> present, and (ii) more efficient formation of transient >Fe[III]-DA surface complexes that precede thermal detachment of DA-bound Fe (Equation 13) and/or DA-induced reductive dissolution (Equation 14):

$$\text{ >Fe[III]}\_{\text{n}} + \text{DA} \rightarrow \text{ >Fe[III]}\_{\text{n}-1} + \text{Fe}^{[III]}\text{DA} \tag{13}$$

$$\text{\textbullet Fe[III]}\_{\text{n}} + \text{DA} \rightarrow \text{\textbullet Fe[III]}\_{\text{n}-1} + \text{Fe[II]} + \text{DA}^{\bullet-} \text{\textbullet \textbullet }$$

#### Ex vivo Modeling Predicts Potential Toxicity of Iron-Mediated Dopamine Oxidation

The kinetic model developed in this study was used to infer potential long-term effects of the combined mechanisms of DA oxidation in the presence of Fe. Particular emphasis was given to the effect of system conditions on Fe turnover rate (TOR; used as a measure of Fe-specific catalysis) and associated hydroxyl radical and DAC production, encompassing changing DA and H2O<sup>2</sup> concentrations and steady-state dissolved O2. In this study, given the overall goodness of the model fit to the experimental data (**Figures 1**–**5**, **Supplementary Figures 5**, **7**), the TOR was determined from the oxidation rate of Fe[II] as predicted by the model under particular conditions (e.g., pH, total concentration of Fe, the concentration ratio of Fe to DA and O<sup>2</sup> content).

#### Dopamine Concentration Mediates Iron Redox Cycling

In general, cycling rate of Fe was highest during the initial phase of each experiment and was amplified by increasing concentration of DA (**Figure 5A**) and higher pH (**Supplementary Figure 9**). After the initial dynamic cycling period, the TOR steadily decreased and appeared to reach equilibrium after ∼30 min. Both acidic pH (pH 6.5) and DA concentration dictated higher Fe TOR at the near-midpoint of the reaction time (t = 50 min); the presence of 30µM DA induced a 10-fold increase in turnover rate compared to DAabsent conditions at pH 6.5 whereas at pH 7.0 and 7.4 the increase in TOR represented only 2 and 1.3-fold change, respectively (**Figure 5B**). From this, we conclude that the effect of DA on Fe TOR is much more substantial in acidic conditions with this likely contributing to the abeyance of H2O<sup>2</sup> generation (shown in **Figure 3A**) and predicted increase in •OH production as a result of both Fenton chemistry (Equations 1, 2) and Fe[II]-mediated DA oxidation (Reactions 23 and 26, **Table 3**).

#### Kinetic Modeling of Sustained Parkinsonian Oxidative Stress

As a result of the attenuation by endogenous antioxidants, including SOD1, GPx, and endogenous ascorbate ions (Harrison and May, 2009), in vivo steady-state concentrations of H2O<sup>2</sup> stemming from normal mitochondrial respiration in the brain are estimated to be maintained at around 5 nM (Adam-Vizi, 2005). However, concomitant hypoxia thought to occur in the Parkinson's disease brain (Adams and Odunze, 1991) may result in both the accumulation of lactate as a result of the anaerobic conditions, which, in turn, may contribute to the decrease in

pH, as well as a release of DA from chemically-isolated vesicles (Phebus et al., 1986).

To predict the rate of •OH generation in a chemical environment more reflective of continued replenishment of prooxidants in a DA neuron, the kinetic model was applied where pseudo-equilibrium was reached in the presence of a range of fixed H2O<sup>2</sup> and DA concentrations. While it must be emphasized that this model is illustrative, as it is specific to only DAderived oxidative stress, it represents an important advance in understanding the biochemical mechanism by which gradual neuronal Fe accumulation with age can become pathological in cells with both high metabolic output and an abundance of a pro-oxidant catecholamine in DA.

As shown in **Figure 6A**, at a fixed concentration of 50µM DA, increased •OH production rate was observed with increasing H2O<sup>2</sup> concentration and decreasing O<sup>2</sup> concentrations over a 10–60µM range typical of that within the brain (Ndubuizu and Lamanna, 2007). The increased production of hydroxyl radicals as H2O<sup>2</sup> becomes more concentrated can be attributed to the relatively slow oxygenation rates at these conditions and the potential for active peroxidation of any Fe[II] present. In contrast, the effect of DA concentration on •OH production as predicted by the model was more complex (**Figure 6B**; **Supplementary Figure 10**) as DA can both sequester

Fe[II] and Fe[III] in a variety of coordination complexes and mobilize Fe[III] from ferric oxide precipitates via DA induced dissolution. As a whole, DA concentrations under 0.1µM produce minimal amounts of •OH regardless of the steady-state H2O<sup>2</sup> concentration with this effect principally a result of there being insufficient labile Fe present to promote oxidation. In contrast, as DA concentrations exceed 10µM, production of •OH is accelerated, provided the steady-state H2O<sup>2</sup> concentration is in excess of ∼100 nM, indicating that the leakage of DA at the later stage may aggravate the Fe and DA induced oxidative stress.

Again, the effect of increasing pH that promotes FeIIIDA<sup>2</sup> complex formation partially attenuates •OH production (**Supplementary Figure 10**), whereas FeIIIDA production favored at pH 6.5 results in an increase in •OH generation rate (**Figure 6B**; **Supplementary Table 2**) compared to the behavior observed at both higher pH values and equivalent DA and H2O<sup>2</sup> concentrations in an Fe-free environment.

Dopamine concentration had the largest effect on DAC production for all pH values investigated (**Figure 6C**; **Supplementary Figure 11**) with the highest rate of quinone formation observed at pH 6.5. On the other hand, while the model indicated H2O<sup>2</sup> concentration had minimal influence on the rate of DAC formation at pH 6.5, low concentrations of H2O<sup>2</sup> at pH 7.0 and pH 7.4 favor the accumulation of DAC, particularly at the lower range of DA concentrations used. Highlighting the versatility of Fe redox activity and why it is tightly regulated under normal physiological conditions, these results suggest that the oxidative load in a dopaminergic neuron arising from Fe-mediated •OH production at basal H2O<sup>2</sup> levels are somewhat mitigated by the concomitant oxidation of Fe[II], effectively sequestering Fe in the ferric state, which is less active with respect to neurotoxic DAC production.

#### DISCUSSION

The value of this study is in the detailed kinetic and thermodynamic profiling of DA oxidation, which provides direct chemical evidence to support the involvement of this unique source of oxidative stress in a potentially pathogenic mechanism of dopaminergic neurotoxicity.

#### Iron, Dopamine, and Selective Neuronal Loss in Parkinson's Disease

Iron accumulates in multiple regions of the aging human brain (Acosta-Cabronero et al., 2016), and is most marked in the Ferich deep gray matter within the basal ganglia (Li et al., 2014). The degree of accumulation in Parkinson's disease patients is markedly higher in the SNc and putamen (Wang et al., 2016), where dopaminergic neurons project, and is associated with disease severity (Wallis et al., 2008), suggesting elevated levels of Fe are intrinsically linked to the biochemical mechanism underlying neuron loss (Hare et al., 2015). The diseasespecific hyperaccumulation of Fe throughout the nigrostriatal pathway (He et al., 2015) indicates a particular vulnerability of dopaminergic neurons, and a lack of apparent pathology in non-Parkinson's-affected regions that also accumulate Fe during normal aging (Acosta-Cabronero et al., 2016) support the hypothesis that DA oxidation and free radical generation, occurring either independently of, or concurrent to, classical Fenton chemistry is a major source of oxidative stress in Parkinson's disease.

This characteristic feature may explain how two essential neurochemicals, which are normally segregated by vesicular confinement of DA to prevent oxidation by cytoplasmic Fe, are able to interact in the Parkinson's disease brain, and how they relate to α-synuclein dysfunction. Both mutations to the SNCA gene encoding α-synuclein and post-translational oxidative damage to the protein (Lotharius and Brundin, 2002b), the latter being an effect of elevated levels of Fe and Fentontype ROS production, can lead to permeabilized vesicles that effectively "leak" DA into the pro-oxidant environment of the cytoplasm (Lotharius and Brundin, 2002a). This initial •OHdriven oxidative damage may trigger a cascade of free radical production that accelerates the rate of α-synuclein modifications, impaired vesicular transport, and DA oxidation, in turn, aggravating cell loss by triggering DAC-induced neurotoxicity.

Central to the view that elevated Fe facilitates DA breakdown is the notion of a labile Fe pool. The propensity of unbound cytoplasmic Fe[II] to react with by-products (such as H2O2) of mitochondrial respiration necessitates tight metabolic regulation of neuronal Fe levels involving various regulatory proteins (Moos et al., 2007), a number of which have been recognized as being dysfunctional in Parkinson's disease (Hare et al., 2013). Whether a labile Fe pool exists in actuality is the subject of much debate and both rapid oxidation of Fe[II] in the O2-rich environment and obvious difficulties in directly speciating Fe within a living human neuron represent two analytical challenges that preclude obtaining conclusive evidence that Fe dyshomeostasis actively promotes increased oxidative stress in Parkinson's disease. Regardless, as nigral Fe accumulation is now recognized as an indisputable pathological feature of Parkinson's disease (Ayton and Lei, 2014; Wang et al., 2016), understanding the mechanism of how Fe promotes oxidative stress in dopaminergic neurons is critical as Fe chelation therapies enter clinical trials (Devos et al., 2014; Martin-Bastida et al., 2017; Moreau et al., 2018; Sun et al., 2018c).

## Iron-Induced Dopamine Radical Production Is Independent of Fenton Chemistry

Iron-mediated DA oxidation differs from Fenton/Haber-Weiss chemistry in that, in addition to redox cycling of Fe[II]/Fe[III], both species are capable of forming intermediary complexes immediately preceding free radical production. For instance, both Fe[II] and Fe[III] are capable of d orbital "bridging" between DA and O2, and inducing the subsequent accelerated transfer of electrons and generation of DA•− and O•− 2 free radicals (Miller et al., 1990). The reactions and associated rate constants involved in Fe[II]/Fe[III] redox cycling and DA oxidation are listed in **Tables 2**, **3**.

The particularly strong ability of Fe[II] to generate DAC and H2O<sup>2</sup> evident in **Figures 2**, **3** may result from the relative insolubility and precipitation of Fe[III] as AFO over the pH range investigated. This is an important factor to consider, as ferritin sequesters Fe within a protein nanocage in a chemical state resembling AFO (Jian et al., 2016). However, the ready oxidation of Fe[II] in the presence of DA (Reaction 24 and Reaction 25 in **Table 3**) favors the conversion of Fe[II] to the Fe[III]DA complex and subsequent O•− 2 generation (Sun et al., 2016). Fe[III]DA is not particularly stable and may either react with another DA molecule to form FeIIIDA<sup>2</sup> (shown in **Figure 4C** and **Supplementary Figure 7C**), reversibly react with O•− 2 to reform Fe[II]DA and O<sup>2</sup> or undergo ligand-to-metal charge transfer (LMCT) with release of Fe[II] and DA•− as follows:

$$\text{Fe}^{\text{[III]}}\text{DA} \rightarrow \text{Fe[II]} + \text{DA}^{\bullet-} \tag{15}$$

Compared to the relatively inactive iron oxyhydroxide precipitates, the more soluble DA-bound Fe[III] oxidation product formed from ferrous DA complexes should enhance the production of both DAC and H2O<sup>2</sup> by accelerating generation of both DA•− and O•− 2 radicals.

Distinguishing iron-mediated dopamine oxidation from Fenton-generated ROS is important, as the free radicals produced do not have the multiple redundant detoxification pathways that exist for H2O2. While quinones are a relatively minor component of dopamine metabolism, which ordinarily favors providing a precursor for adrenaline biosynthesis or excretion from the central nervous system as homovanillic acid, it is unclear precisely how these highly-reactive species are neutralized. Several endogenous enzymes have been shown to interact with dopamine o-quinones, including SOD1, glutathione transferase (by way of glutathione conjugation) and macrophage migration inhibitory factor (Solano et al., 1999; Haque et al., 2003), though quinone detoxification would be considered a secondary function. In the absence of metal catalysis, polymerization of quinones to neuromelanin is a remarkably slow process with a normal aged brain containing only around 5 mg g−<sup>1</sup> (Aime et al., 2000). Without an efficient mechanism of quinone removal, these products have greater neurotoxic potential. This is especially the case in human dopaminergic neurons, in light of recent data

indicating that dopaminergic neurons derived from fibroblasts of sporadic and familial Parkinson's disease patients contain higher concentrations of DA and oxidation products than similarlyprepared murine cell cultures (Burbulla et al., 2017).

## Relevance of Iron and pH to Disease Progression

The rapid formation of both DAC and H2O<sup>2</sup> may be attributed to the pH-dependent oxidation of DA-bound Fe[II] shown in **Figure 4A**. Variation in pH reflecting transition from a physiologically normal microchemical environment to Parkinassociated mitochondrial dysfunction (Pickrell and Youle, 2015) and post-mortem acidosis (Donaldson and Lamont, 2013) had a marked effect on DAC and H2O<sup>2</sup> production concomitant to pH-dependent oxidation of Fe[II]. The catechol moiety of DA is capable of forming a strong complex with Fe[III] (Avdeef et al., 1978; Sever and Wilker, 2004):

$$\text{Fe}[\text{III}] + \text{DA} \rightarrow \text{Fe}^{\text{[III]}}\text{DA} \tag{16}$$

with the mono-complex (Fe[III]DA) dominant at acidic pH and the bis Fe[III]DA<sup>2</sup> complex favored at circumneutral pH (**Supplemental Figure 4**). At pH 6.5, the ratio of [Fe[III]DA2]/[Fe[III]DA] is ∼5 in conditions reflective of otherwise physiologically normal concentrations of Fe and DA. Once formed, radical generation would also occur, increasing potential protein and lipid peroxidation events with the rate of LMCT for Fe[III]DA (0.23 s−<sup>1</sup> ) substantially higher than that for Fe[III]DA<sup>2</sup> (7.26 × 10−<sup>5</sup> s −1 ; El-Avaan et al., 1997; Sun et al., 2016). From the rate law

$$\frac{d\,\,[\text{Fe}[\text{II}]]}{dt} = \frac{d\,\,[\text{DA}^{\bullet-}]}{dt} = k[\text{Fe}^{[\text{III}]}\text{DA}] \tag{17}$$

continuous generation of DAC at pH 6.5, especially in the presence of Fe[II], was unsurprising. The slightly higher concentrations of DAC formed in the presence of Fe[III], particularly during the initial half of the total reaction time assessed (**Figure 2C**), may be attributed to the more efficient Fe mobilization induced by DA at higher pH.

In contrast to the slower yet continuous generation of H2O<sup>2</sup> at pH 7.0 and 7.4 (**Figure 3A**), the non-linear formation of DAC at these pHs (**Figure 2B**) suggests that DAC levels are maintained at a steady-state under physiological conditions through rearrangement of DAC and the subsequent formation of indoles or even possibly neuromelanin (**Supplementary Figure 8**). It is reasonable to deduce that the rearrangement of DAC coupled with the ensuing polymerization on increase in pH plays an important role in the removal of DAC.

It should be noted that the presence of metals, such as Fe and calcium (as Ca2+) can also promote this process (Sun et al., 2018a). The model developed adequately describes the generation of reactive oxygen species and accumulation of DAC but comprehensive description of the specific role of Fe in DAC decomposition will require an expanded model. Regardless, the accumulation of protein-modifying DAC is likely to be enhanced at low pHs as a result of the presence of high concentrations of the active Fe[II] catalyst seen in both parkinsonian animal models (Hare et al., 2014; Billings et al., 2016) and the human Parkinson's disease brain (Dexter et al., 1991).

The cessation of H2O<sup>2</sup> accumulation observed at pH 6.5 (**Figure 3A**) may be related to the high concentration of Fe[II] present in the solution. Under these conditions, the H2O<sup>2</sup> initially generated as a result of DA-enhanced oxidation of Fe[II] is subsequently consumed via peroxidation of the remaining Fe[II]. Apart from the obvious pH-dependent oxidation process, attenuation of Fe[II] oxidation at pH 6.5 (**Figure 4A**), especially during the later stages of the reaction, suggests that Fe[II] may be regenerated under these conditions with the regeneration of Fe[II] related to formation of DA intermediates such as DA•−. Given the continuous generation of DAC (**Figure 2B**), the high concentrations of Fe[II] observed may, in turn, result from DA•−-induced reduction of ferric iron. Thermodynamically, the low reduction potential of the DAQ/DA•− couple indicates that reduction of aqueous Fe[III] is feasible (Pham and Waite, 2014). Indeed, the rapid reduction of Fe[III] by 6-hydroxydopamine (6- OHDA) semiquinones has been proposed (Jameson and Linert, 2001) with 6-OHDA itself a neurotoxic breakdown product of DA that is facilitated by Fe (Hare and Double, 2016). However, as pH increases, it is expected that the majority of the Fe present would be efficiently converted into AFO and, with a low reduction potential at pH 7.0 (Watt et al., 1985), DA•− is unable to mobilize Fe from AFO. It is of note that neuronal ferritin iron is stored in an AFO-like ferrihydrite core (Hagen et al., 2017) which is prone to mobilization under acidic conditions (La et al., 2017).

## Considerations for Analysis of Dopamine Metabolites in Post-mortem Tissue

When viewed within the context of existing literature, the results described here present something of a conundrum for those wishing to quantify DAC and related DA oxidation products in human SNc tissue. While acidosis occurring in degenerating neurons may contribute to the progression of Parkinson's disease by increasing the rate of oxidant generation and accumulation of DAC, post-mortem decreases in tissue pH, stemming from sample handling, exposure to the environment after removal at autopsy, and even post-mortem interval may present inaccurate assessment of DA oxidation.

Considering mitochondrial dysfunction is common to most neurodegenerative diseases (Lin and Beal, 2006), it is not surprising that data supplied from tissue housed in the Sydney Brain Bank, where 72% of cases with a neurodegenerative disease, including pathologically-diagnosed Parkinson's disease, exhibited tissue pH as low as 5.86 (Genoud et al., 2017). Tissue pH that deviates from normal physiology, regardless of whether the cause is endogenous or artifact, will undoubtedly alter the speciation of DA and Fe, and rate of formation, reactivity and fate of toxic DA intermediates (El-Avaan et al., 1997; Pham et al., 2006; Sun et al., 2015, 2016, 2018a). Accordingly, to avoid any misinterpretation of DA metabolites, quantitative assessment of DA metabolism in post-mortem tissue should account for multiple confounding factors that influence tissue pH.

# CONCLUSIONS

The results of this study indicate that the formation of DA bound Fe[II] and Fe[III] complexes as well as the cyclization and rearrangement of DA-derived quinones are the most important pathways of DA metabolism, with each process heavily dependent on pH. A schematic showing the relative importance of these processes together with expected changes in key metabolites on decrease in pH is presented in **Figure 7**. The presence of Fe can accelerate the oxidation of DA and the accumulation of deleterious protein-modifying DAC with Fe[II] being more efficient in this regard than Fe[III]. Although DAC was slowly formed at pH 6.5, the decrease in the rearrangement rate of this species in acidic conditions resulted in long-term accumulation of DAC. As a result of both the rapid reduction of the unbound-Fe[III] by DA•− and the enhanced LMCT that occurs with the change in the speciation of DA-bound Fe, a substantially higher concentration of Fe[II] was generated at pH 6.5 compared to that at pH 7.0 and 7.4. Even though the presence of both Fe[III] and Fe[II] resulted in increased accumulation of H2O2, the enhanced Fe[II] regeneration coupled with the slow rate of oxygenation of Fe[II] at pH 6.5 gave rise to substantially greater potential for the peroxidation of Fe[II] with concomitant enhanced generation of •OH. Model predictions indicate that, in the presence of the same concentrations of DA and H2O2, acidosis in the Parkinson's disease brain results in an increase in DAC accumulation and DA-mediated production of hydroxyl radicals, with these potential toxicants likely to further aggravate the progression of Parkinson's disease and severity of symptoms arising from DA denervation.

The model developed in this study places emphasis on the purely chemical interactions between Fe and DA and is of particular value in facilitating prediction of the long-term consequences of these interactions by incorporating a range of conditions resulting from the complicated in vivo homeostasis of dopaminergic neurons such as continuous DA leakage into the cytosol and increased steady-state concentrations of H2O<sup>2</sup> as a result of the dysfunctional of antioxidant enzymes. It is also important to view these data within the broader biochemical context of Parkinson's disease neuropathology as a result of the limitations of current techniques. Regardless, the data presented here shows that pH, Fe and H2O<sup>2</sup> concentrations are intrinsically linked in the rate of formation of neurotoxic DA metabolites.

#### AUTHOR CONTRIBUTIONS

All the authors were involved in the experiments design. YS conducted the experiments and prepared the manuscript. AP, DH, and TW revised the manuscript. All authors reviewed the results and approved the final version of the manuscript.

#### ACKNOWLEDGMENTS

The authors would like to acknowledge the Sydney Brain Bank for advice regarding the pH measurements of post-mortem

#### REFERENCES


brain tissue from patients with neurodegenerative disease. The Sydney Brain Bank is supported by The University of New South Wales and Neuroscience Research Australia. We also gratefully acknowledge the China Scholarship Council and the University of New South Wales for scholarship support to YYS; and the National Health and Medical Research Council's Career Development Fellowship (Industry) support to DJH (GNT1122981). Supplementary data associated with this article can be found in the online version.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2018.00859/full#supplementary-material


natural organic matter solutions. Environ. Sci. Technol. 47, 1861–1869. doi: 10.1021/es3035889


in Parkinson's disease. Front. Neurosci. 12:106. doi: 10.3389/fnins.2018. 00106


using reversed-phase ion-pairing chromatography coupled in tandem with UV–visible and ESI-MS detections. J. Chromatogr. B 911, 55–58. doi: 10.1016/j.jchromb.2012.10.026

**Conflict of Interest Statement:** DH receives research and materials support from Agilent Technologies.

The remaining 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 Sun, Pham, Hare and Waite. 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.

# Rusty Microglia: Trainers of Innate Immunity in Alzheimer's Disease

#### Adonis Sfera1,2 \*, Roberto Gradini <sup>3</sup> , Michael Cummings <sup>2</sup> , Eddie Diaz <sup>2</sup> , Amy I. Price<sup>4</sup> and Carolina Osorio<sup>1</sup>

<sup>1</sup> Psychiatry, Loma Linda University, Loma Linda, CA, United States, <sup>2</sup> Patton State Hospital, San Bernardino, CA, United States, <sup>3</sup> Department of Pathology, La Sapienza University of Rome, Rome, Italy, <sup>4</sup> Evidence Based Medicine, University of Oxford, Oxford, United Kingdom

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Ilias Kounatidis, Diamond Light Source, United Kingdom Po-Wah So, King's College London, United Kingdom

> \*Correspondence: Adonis Sfera dr.sfera@gmail.com

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neurology

Received: 09 August 2018 Accepted: 21 November 2018 Published: 04 December 2018

#### Citation:

Sfera A, Gradini R, Cummings M, Diaz E, Price AI and Osorio C (2018) Rusty Microglia: Trainers of Innate Immunity in Alzheimer's Disease. Front. Neurol. 9:1062. doi: 10.3389/fneur.2018.01062 Alzheimer's disease, the most common form of dementia, is marked by progressive cognitive and functional impairment believed to reflect synaptic and neuronal loss. Recent preclinical data suggests that lipopolysaccharide (LPS)-activated microglia may contribute to the elimination of viable neurons and synapses by promoting a neurotoxic astrocytic phenotype, defined as A1. The innate immune cells, including microglia and astrocytes, can either facilitate or inhibit neuroinflammation in response to peripherally applied inflammatory stimuli, such as LPS. Depending on previous antigen encounters, these cells can assume activated (trained) or silenced (tolerized) phenotypes, augmenting or lowering inflammation. Iron, reactive oxygen species (ROS), and LPS, the cell wall component of gram-negative bacteria, are microglial activators, but only the latter can trigger immune tolerization. In Alzheimer's disease, tolerization may be impaired as elevated LPS levels, reported in this condition, fail to lower neuroinflammation. Iron is closely linked to immunity as it plays a key role in immune cells proliferation and maturation, but it is also indispensable to pathogens and malignancies which compete for its capture. Danger signals, including LPS, induce intracellular iron sequestration in innate immune cells to withhold it from pathogens. However, excess cytosolic iron increases the risk of inflammasomes' activation, microglial training and neuroinflammation. Moreover, it was suggested that free iron can awaken the dormant central nervous system (CNS) LPS-shedding microbes, engendering prolonged neuroinflammation that may override immune tolerization, triggering autoimmunity. In this review, we focus on iron-related innate immune pathology in Alzheimer's disease and discuss potential immunotherapeutic agents for microglial de-escalation along with possible delivery vehicles for these compounds.

Keywords: inflammasomes, astrocytes, microglia, exosomes, trained immunity, tolerized immunity

# HIGHLIGHTS


# INTRODUCTION

Alzheimer's disease (AD), the most common form of dementia, is marked by progressive memory impairment and functional decline believed to reflect synaptic and neuronal loss. Elimination of these structures may be carried out by an aggressive astrocytic phenotype, defined as A1. Preclinical studies show that LPS-activated microglia enable the conversion of trophic to A1 astrocytes by releasing several cytokines, including tumor necrosis factor alpha (TNF-alpha), interlukin-1 alpha (IL-1 alpha), and complement component C1q. These molecules alter astrocytic transcriptomes, promoting the neurotoxic A1 cells observed to engage in the phagocytosis of healthy neurons and oligodendrocytes, contributing to their loss (1). In this article we take the position that A1 astrocytes are autoimmune in nature, resulting from the prolonged microglial activation that overrides LPS tolerization.

LPS tolerization refers to the absence of an inflammatory response after repeated or prolonged exposure to this microbial endotoxin as re-challenged "tolerant" innate immune cells are incapable of immunological activation. This is relevant to AD because the constantly elevated brain LPS levels, detected in this disorder, fail to trigger microglial tolerization and lower neuroinflammation (2, 3).

Excess LPS in the brain microenvironment, a danger signal, is detected by microglial TLRs which activate the pathogenmediated iron sequestration via hepcidin-ferroportin axis (4) (**Figure 1**). Indeed, recent neuroimaging studies detected ironcontaining microglia in the hippocampi of AD patients, likely reflecting LPS-triggered iron sequestration (5).

The source of brain LPS is unclear at this time, but it has been hypothesized that it may derive from the gastrointestinal (GI) tract microbial community. These bacteria were found to lie dormant in various tissues, possibly including the brain, for prolonged periods of time restrained by the lack of free iron (3, 6). Later in life, impaired iron homeostasis caused by aging and/or defective ferritinophagy may provide these microbes with enough metal to resume their growth and LPS shedding, engendering inflammatory neuropathology (6).

Iron is closely linked to immunity as it plays a key role in immune cells proliferation and maturation, but it is also indispensable to pathogens and malignancies which compete for its capture (discussed in The Trained Arm of Innate Immunity section). To safeguard iron, innate immune cells, including microglia and astrocytes, sequestrate it intracellularly bound to cytosolic and mitochondrial ferritin, an arrangement which can generate excessive ROS and inflammation when disrupted (4). Indeed, iron-ferritin dissociation was demonstrated in AD animal models in which LPS-induced ferritinophagy increased intracellular free iron, engendering neuroinflammation (7, 8). Moreover, iron-damaged mitochondria were shown to act as danger signals, activating the inflammatory cascade by igniting inflammasomes, including NLRP3 (9).

LPS-activated microglia release cytokines, including TNFalpha, IL-1 alpha, and C1q, that have been associated with autoimmune inflammation, an association that led to the development of vaccines, including anti-TNF-alpha antibodies (for the treatment of autoimmune disorders) and anti-C1q antibody, currently in phase I trials for both AD and autoimmunity (10, 11).

In contrast to immune tolerization which may be an adaptive mechanism to prevent autoimmune inflammation, allowing the clearance of damaged self-components, immune training may serve the purpose of overriding immune anergy encountered in various pathologies, including sepsis and cancer (12, 13). Indeed, the peripheral and central innate immune cells are engaged in a balancing act as they must be sufficiently trained to effectively defend the host against pathogens and malignancies, while at the same time tolerant enough to accept self-components, food, the fetus and commensal microbes. Disruption of this fine equilibrium may either cause unchecked inflammation or excessive tolerance (12). Iron interferes with this balance by promoting activation and inflammation, a property exploited in oncology to train tumor-associated macrophages (TAMs) in fighting malignant cells (14). On the other hand, iron chelators oppose training, providing therapeutic benefit in AD. Indeed, deferiprone, currently in phase 2 clinical trials, may soon become the first FDA approved iron chelator indicated for AD (15). Along these lines, a recent preclinical study demonstrated that in response to peripherally applied inflammatory stimuli, microglia can undergo either activation (training) or tolerization, the former augmenting, the latter lowering neuroinflammation (13). This is important for AD because this study also found that training increases while tolerization decreases β-amyloid accumulation, suggesting that central tolerance can be promoted by peripheral stimulation, comprising a novel therapeutic strategy (13). Previous AD studies are in line with this model as it was demonstrated that immune trained subjects (positive for TNF-alpha) presented with memory loss, while the tolerized patients (TNF-alpha negative) lacked cognitive deterioration (16). Moreover, trained microglia and astrocytes were found able to undergo transcriptomic alterations, morphing into antigen presenting cells (APCs) that can establish immunologic synapses with infiltrating peripheral lymphocytes, further perpetuating neuroinflammation (17).

Traditionally, immunological memory has been associated with adaptive immunity, however novel studies indicate that

innate immune cells, including microglia and astrocytes can "recall" prior vaccinations or infections (13). For example, β-glucan, a cell wall component of Candida Albicans and bacillus Calmette-Guerin (BCG), was shown to train monocytes, improving the outcome of sepsis-induced immune paralysis (18). Indeed, the development of treatments based on innate immune memory manipulation has been encouraged by the recent FDA approval of immune training-inducers, the BCG vaccine for the treatment and prophylaxis of urinary bladder carcinoma in situ (CIS) and muramyl tripeptide for osteosarcoma (19). Along these lines, innate immune memory manipulation may improve AD outcome by de-escalating microglia via peripherallyapplied tolerizing stimuli (13). Since tolerized immunity, unlike immunosuppression, is antigen specific and natural, harnessing it may not only benefit AD but also cancer and autoimmune disorders (20, 21). The recent finding that altered endogenous molecules, including dysfunctional mitochondria, can trigger microglial training will likely result in novel AD strategies, including mitophagy-activating therapeutics (22).

In this article we look beyond iron chelators, focusing on lowering the consequences of iron pathology, especially microglial training and defective tolerization. The immunotherapeutic agents discussed here include inflammasome inhibitors, histone deacetylase inhibitors, activators of co-inhibitory receptors, acetylcholinesterase inhibitors and selective serotonin reuptake inhibitors.

# IRON HOMEOSTASIS AND THE INNATE IMMUNE SYSTEM

Iron is an indispensable but potentially toxic trace element and for this reason it has to be circulated throughout the body attached to its carrier protein, transferrin (Tf). Dietary iron is absorbed in the gut by crossing the enterocyte brush border membrane via divalent metal-ion transporter 1 (DMT-1). After conversion from ferrous to ferric iron by hephaestin and/or ceruloplasmin it is exported through the basolateral membrane via ferroportin (FPN) channels (23).

Novel studies have demonstrated that gram-negative commensal gut microbes are able to generate more than one type of LPS, thus activating or inhibiting innate immune responses (24). For example, Escherichia coli LPS was found to be an activator of innate immunity, while LPS derived from Bacteroides dorei was linked to immune tolerance (25). Interestingly, Bacteroides fragilis LPS (BF-LPS) was shown to contribute to autoimmune pathology by selectively inhibiting innate immune tolerization (26). Since the host and microbes share a common iron pool, its availability is directly linked to the survival and growth of the LPS-shedding bacteria.

Iron enters the CNS together with Tf (iron-Tf complex) by binding to transferrin receptor-1 (TFR-1) expressed by the endothelial cells of brain capillaries. After crossing into the cytosol of these cells and satisfying their metabolic needs, excess iron is stored in the cytosolic and mitochondrial ferritin. The remaining iron is exported via FPN into the extracellular space (27).

One of the key roles of the innate immune system is pathogen-mediated intracellular iron sequestration to deny it to bacteria and malignancies which are dependent on this biometal for growth (discussed in The Trained Arm of Innate Immunity section). Iron sequestration is initiated in response to exogenous and endogenous danger signals, consisting of pathogen-associated molecular patterns (PAMPs) or, damage-associated molecular patterns (DAMPs), respectively. For example, LPS, the cell wall component of gram-negative bacteria, is a PAMP that activates microglial TLR-4. In contrast, the endogenously produced ROS is a DAMP that activates the cytosolic inflammasome NOD-like receptor pyrin domaincontaining-3 (NLRP3) (28). Iron, itself and/or iron-induced ROS were shown to activate NLRP3 inflammasomes, engendering neuroinflammation (9) (**Figure 1**).

Iron enters neurons via DMT-1 and TFR-1 receptors but can only exit these cells through FPN, a receptor controlled by hepcidin, the master regulator of iron metabolism. Hepcidin is biosynthesized in the liver, but since only a small fraction of this this peptide hormone crosses the brain-blood barrier (BBB), the brain produces its own hepcidin. High hepcidin levels are associated with intracellular iron accumulation, as this biomolecule binds to FPN, internalizing this receptor (29). LPS and pro-inflammatory cytokines stimulate hepcidin production, inducing hypoferremia, a mechanism involved in the anemia of chronic inflammation pathogenesis (30). Interestingly, astrocytes were shown to induce neuronal death in response to LPS-activated microglia and to synthesize hepcidin via the IL-6/STAT3 pathway, likely linking hepcidin to A1 astrocytes (31). In addition, iron activates nuclear factor kappa-light-chainenhancer of activated B cells (NF-kB), a transcription factor involved in innate immune training and inflammation (27).

Microglia are yolk-sac-derived myeloid cells, comprising up to 20% of the total brain glial cell population (32). Their main function consists of brain parenchyma surveillance, searching for endogenous or exogenous danger signals to which they respond by activation (33). Another major function of microglia, assisted by astrocytes, consists of tolerization or acceptance of self-molecules, food, commensal microbes, and the fetus. This task is accomplished by phagocytosis via prompt clearance of molecular debris, including dead or dying cells that were shown to trigger inflammation and autoimmunity when allowed to accumulate (34). Taken together, disruption of phagocytosis can lead to unchecked inflammation, impaired immune tolerance and autoantibodies against self-components. Indeed, increased levels of autoantibodies were documented in AD, suggesting that autoimmunity may play a role in this disorder (35).

# THE TRAINED ARM OF INNATE IMMUNITY

Pathogen-mediated iron sequestration is an effective mechanism that protects the host against bacteria and malignancies at the risk of inflammation and ROS triggered by increased intracellular iron (36). To lower this risk, iron must be properly stored into the cytosolic and mitochondrial ferritin, a process that may be disrupted by inflammation and age-related pathology. In the CNS, pathogen-mediated iron sequestration is initiated upon detection of PAMPs and/or DAMPs by microglia and astrocyte, followed by the release acute phase proteins, including hepcidin and lipocalin-2 (LCN-2), to internalize iron (29, 37).

Aging augments the expression of hepcidin and LCN-2, which along with impaired ferritinophagy, increases the risk of free intracellular iron and prolonged inflammation via NLRP3 inflammasomes activation (16, 38). Both hepcidin and LCN-2 have been associated with neurotoxicity and AD, despite some studies describing them as neuroprotective (39–42). These findings are contradictory in appearance only because hepcidin and LCN-2 may have both beneficial and detrimental effects, depending on the phase of inflammation. For example, during the acute phase, iron sequestration may be adaptive as it is denied to pathogens, limiting infection and inflammation. On the other hand, with chronic inflammation and increasing age, intracellular iron may become detrimental as it activates NLRP3 likely overwhelming immune tolerization and triggering neuronal loss. In addition, iron also activates NF-kB implicated in tolerization disruption and autoimmunity when in excess or inappropriately activated (43) (**Figure 1**). Iron may impair immune tolerance directly by altering microglial secretome, enhancing the release of TNF-alpha, IL-1, and C1q, cytokines that promote A1 astrocytes (1). Indeed, the pathological phagocytosis that A1 cells engage in may reflect the loss of neuronal and oligodendrocytic tolerization. Interestingly, a different study reported A1 astrocytes' upregulation in healthy elderly individuals, linking these cells to the chronic inflammation of aging or inflammaging, considered by some an AD prodrome (44, 45). Moreover, a novel study found that glucagon-like peptide-1 receptor (GLP1R) agonists can block the conversion of trophic to A1 astrocytes, suggesting that Diabetes mellitus type 2 treatments, including liraglutide, may benefit inflammaging and AD (46). Interestingly, liraglutide was documented to lower inflammatory responses by decreasing the expression of NLRP3, suggesting that this drug may function as an inflammasome inhibitor (47).

Inflammasomes are cytosolic multiprotein complexes activated in innate immune cells in response to PAMPs or DAMPs (16, 48). Inflammation is ignited by the inflammasome assembly with caspase1-induced maturation of interleukin-1β and IL-18 which in turn activate multiple inflammatory pathways (49) (**Figure 2**). NLRP3 inflammasomes, abundantly expressed by microglia and astrocytes, may play a major role in AD as they can be activated by beta amyloid, iron and ROS, maintaining

chronic inflammation (48, 50, 51). Indeed, continuous immune training in inflammaging and AD was associated with perpetually elevated NLRP1 and NLRP3 mRNA levels (52–54). Similarly, chronic NLRP3 activation was associated with atherosclerosis and Western diets, linking metabolism and chronic inflammation (55, 56).

Recently the mitochondrion was identified as the interface connecting immunity and metabolism as this organelle contains the mammalian target of rapamycin (mTOR), a protein involved in shifting the metabolic gears from oxidative phosphorylation (OXPHOS), characterizing tolerized immunity, to aerobic glycolysis, marking immune training (11). Iron has been demonstrated to induce mitochondrial damage, probably lowering OXPHOS and prompting NLRP3 inflammasome assembly and microglial training (57, 58) (**Figure 2**). In fact, iron-disrupted OXPHOS may leave aerobic glycolysis as the only metabolic option for microglia and a possible source of neuroinflammation in AD. Along these lines, a recent study described a direct cross-talk between mTOR and iron metabolism, suggesting a connection between this biometal and aerobic glycolysis (59, 60). Mitophagy or elimination of damaged mitochondria (including that which is iron-induced) was reported to inhibit NLRP3 and lower immune training in AD models, pointing to mitophagy as a potential AD treatment (61–64). Interestingly, liraglutide, a GLP-1R agonist, was shown to promote mitophagy in addition to, or because of NLRP3 inhibition, indicating a potential "off label" use of this drug (65).

A recent preclinical study linked the NLRP3 component, apoptosis-associated speck-like protein containing a caspase recruitment domain (ASC), with the prion-like seeding of beta amyloid (66). Similarly, in a previous study, ASC was reported as capable of migrating through the extracellular space (67). Another novel study found a link between NLRP3 inflammasomes and prion disease (68). Furthermore, multiple studies over the past decade established a close connection between various forms of prion diseases and autoimmunity, suggesting a common denominator (69, 70). Interestingly, iron was associated with both prion disease and autoimmunity, and chelation was found therapeutic in CNS autoimmune disorders, including MS and experimental autoimmune encephalomyelitis (EAE) (71–74). For example, brain iron dyshomeostasis was demonstrated in prion disease patients, while another study found a higher incidence of autoimmune disorders in individuals with hemochromatosis (71, 74). A preliminary study found that deferiprone, a BBB-crossing iron chelator, was therapeutic in MS patients (73). Moreover, some iron chelators were found beneficial in AD and Parkinson's disease (PD), possibly pointing to the role of autoimmune inflammation in these disorders (75– 77). Indeed, over the past decade, several studies have associated autoimmunity with AD as serum and CSF autoantibodies against β-amyloid, tau, ceramide and adenosine triphosphate (ATP) synthase were detected in this disorder (35, 78–82). Interestingly, these molecules were also associated with iron and NLRP3, making it worthwhile to search for ASC autoantibodies in AD (83–87).

#### THE TOLERIZED ARM OF INNATE IMMUNITY

Effective protection from pathogens and malignant cells is an important function of the innate immune system, but an equally important task is protection against autoimmune inflammation by maintaining immunologic tolerance to self-structures, food, the fetus, and commensal bacteria. Tolerized immunity or absence of immune responses to repeated or prolonged LPS stimulation was originally described in rodents which survived the administration of a lethal endotoxin dose after previous exposure to sublethal concentrations (88). The adaptive role of this mechanism in the CNS may include "tolerating" elevated levels of LPS demonstrated in the brains of aging individuals and AD patients (2). Indeed, aging and AD have been associated with increased gram-negative bacteria titers in the GI tract and gums (89). For example, individuals with AD were found to have three times higher LPS blood levels compared to controls (90). Interestingly, novel studies have shown that B. fragilis generates a structurally different LPS (BF-LPS), that selectively inhibits immune tolerization, possibly contributing to the pathogenesis of both autoimmune disorders and AD (25, 26).

Aside from the LPS response, brain tolerized immunity may serve the purpose of inhibiting inflammation during phagocytosis of apoptotic cells, synapses, or damaged endogenous molecules, a process we define as tolerant phagocytosis (91, 92). Preclinical studies have shown that disruption of this pathway and accumulation of molecular debris activates autoimmune inflammation, inflicting neuronal damage (93, 94). A novel study has reported that LPS downregulates the expression of the triggering receptor expressed on myeloid cells 2 (TREM2), the microglial receptor associated with immune tolerance (95) (**Figure 1**). Apolipoprotein E (ApoE) is a TREM2 ligand and their interaction may be necessary for tolerant phagocytosis (96–98). Indeed, TREM-2 blockade was demonstrated to exacerbate EAE and MS, linking this receptor to the CNS autoimmune disorders (99, 100).

A recent pivotal study established the existence of an ApoE-TREM-2 axis which under normal circumstances engenders tolerized or homeostatic microglial phenotypes (101). In contrast, a dysfunctional ApoE-TREM-2 axis was associated with deficient tolerization and unopposed inflammation. For example, ApoE4 isoform, linked to late onset AD, presents with poor TREM-2 binding, leading to the activation of NLRP3 and immune training instead of tolerization (102, 103).

In AD, the tolerized arm of innate immunity appears to be impaired as persistently elevated LPS brain levels fail to induce tolerance, probably triggering the autoimmune elimination of neurons and oligodendrocytes via A1 astrocytes (1, 2, 98, 104). Moreover, LPS elevation in AD was linked to iron as dormant microbes, including B. fragilis, may survive in the CNS in an inactive state limited by the lack of free iron (6). During

tolerization via ApoE -TREM-2 axis. Both excessive immune training or insufficiently tolerized microglia may trigger neuronal loss. ApoE4 isoform binding C1q triggers inappropriate elimination of healthy synapses by astrocytes, likely by promoting the A1 cells.

aging, impaired iron homeostasis provides these microbes with enough metal to resume their growth and LPS shedding, possibly explaining the LPS upregulation observed in AD (105). Indeed, iron may facilitate B. fragilis resuscitation in the brain as this microbe was shown to require iron or heme for surviving outside of the GI tract (106). In addition, iron-induced endothelial cells' damage may open the BBB (a property exploited in the treatment of gliomas), allowing microbial passage (107, 108). On the other hand, iron chelators were demonstrated to inhibit B. fragilis growth, lowering LPS brain levels, perhaps explaining some of the beneficial effects of these compounds in AD (109). In addition, LPS, not only promotes ferritinophagy, but also the astrocytic expression of hepcidin via IL6/STAT3, further augmenting the cellular free iron pool along with the chances of dormant bacterial resuscitation (31).

At the molecular level, tolerized immunity was shown to be associated with microglial TREM-2 receptors, interacting with anti-inflammatory molecules, including ApoE, and complement components C1q/C3 (110–112) (**Figure 3**). In addition, tolerized immunity was shown to be modulated by iron-dependent hypoxia-inducible factor 1α (HIF-1α), suggesting another tolerization target (113).

Under physiological circumstances C1q and C3 facilitate phagocytosis by tagging dysfunctional synapses for engulfment by microglia and astrocytes (114, 115). In addition, C1q enhances immune tolerance and inhibits NLRP3-induced immune activation (116). Pathologically, deficient C1q was found in over 90% of systemic lupus erythematosus (SLE) patients, linking this protein to autoimmune inflammation (117). In contrast, ApoE4-induced C1q upregulation was associated with inappropriate elimination of healthy synapses by astrocytes, connecting this phenomenon to autoimmunity and the A1 cells (118). Indeed, the development of anti-C1q antibodies may comprise a novel therapeutic strategy for both AD and autoimmunity (10).

Iron disrupts tolerized immunity by lowering ApoE expression in microglia, impairing the ApoE-TREM-2 axis and amyloid plaque phagocytosis (119–121). Indeed, ApoEimmunoreactive microglia were found closely associated with senile plaques, suggesting a key role in AD pathology (122). In addition, iron disrupts tolerized immunity by upregulating C3 and altering the C1q expression, likely preventing their ApoE binding (123, 124). Indeed, C3 upregulation was linked to increased beta amyloid and phosphorylated tau, linking once more iron to AD pathogenesis (110, 125) (**Figure 3**). Other recent studies connected iron-upregulated C3 with the A1 astrocytes by demonstrating high complement levels in astrocyte-derived exosomes obtained from AD patients (114, 126).

Aside from ApoE and C1q/C3, iron inhibits α-secretase or ADAM-10, the enzyme involved in the generation of soluble TREM-2 (sTREM-2), linking again this biometal to autoimmunity (102, 123, 124, 127–129) (**Figure 4**).

Adenosine-triphosphate (ATP) synthase, the mitochondrial enzyme involved in ATP biosynthesis, is a newly identified AD target, especially because extracellular ATP (eATP) is a potent NLRP3 activator associated with autoimmune disorders (81– 85). This is relevant as preclinical studies show iron-associated

augmentation of F1 and F0 components of ATP synthase with loss of mitochondrial ATP, possibly suggesting ATP displacement to the extracellular compartment (130). In addition, ironbinding protein, lactoferrin clears eATP, presenting with a beneficial role in AD (see the section on exosomes and Inflammasomes). Furthermore, blockade of purinergic receptors P2Y1 was associated with improved cognition in AD models, while P2X7 inhibition was found beneficial in autoimmune disorders, linking eATP and iron to impaired immune tolerance (87, 130, 131).

# THE INNATE-ADAPTIVE CONTINUUM

Novel preclinical studies indicate a close cooperation between the innate and adaptive immune systems via NLRP3 inflammasomes and IL-1β (132, 133). Microglia and astrocytes communicate extensively with peripheral lymphocytes, inducing their activation and CNS migration (10, 134). IL-1β is known for increasing BBB permeability, facilitating the infiltration of peripheral cytotoxic T-cells (CTCs) into the CNS (135). In addition, iron is known for altering the BBB by inflicting endothelial cells damage, facilitating peripheral cells access to the CNS (107). In addition, peripheral lymphocytes may access the brain via the newly discovered CNS-draining lymphatics (17).

The newly arrived cells were shown to activate NLRP3, attracting more blood-borne immune cells into the brain (136). On the other hand, tolerized glia may orchestrate the CNS import of T-regs or the newly identified regulatory B1a cells which in turn lower microglial training (137–140). In addition, B-cells have been known to facilitate the recruitment of T-regs to the CNS, augmenting their immunosuppressive activities (141).

T-regs, the preservers of immune tolerance, were found to delay AD progression, linking this condition to autoimmune inflammation (142, 143). The anti-inflammatory IL-10 released by tolerized microglia and T-regs was demonstrated to negatively regulate iron metabolism, possibly explaining the beneficial role of Tregs in AD (144). In addition, IL-4 also regulates iron metabolism and promotes immune tolerance (145). Indeed, iron has been shown to facilitate the CNS infiltration of peripheral CTCs, while iron chelation produces the opposite effect (146). Furthermore, IL-10 was demonstrated to inhibit aerobic glycolysis, the metabolic driver of trained immunity, promoting the tolerant OXPHOS (144, 147, 148). Interestingly, a recent preclinical study discovered a CNS population of cerebral T-regs, documenting that aside from blood borne, the brain maintains its own tolerogenic lymphocyte population (139). Other novel studies demonstrated the existence of various T-regs subpopulations, expressing different co-inhibitory Tcell receptors, including the T-cell immunoglobulin and ITIM Domain (TIGIT) that regulates immunity exclusively via ironlowering IL-10, linking this receptor to iron downregulation (149, 150).

Co-inhibitory receptors, including cytotoxic T-lymphocyte associated protein 4 (CTLA-4), programmed cell death protein 1 (PD-1), lymphocyte-activation protein 3 (Lag-3), and TIGIT play an essential role in the prevention of autoimmune disorders by promoting tolerized responses (151, 152). A growing body of evidence reveals that dysfunctional co-inhibitory receptors may contribute to AD neuroinflammation as under normal circumstances they function to inhibit inflammation-generating co-stimulatory signals (153). For example, upregulation of PD-1 expressed on both microglia and T-regs, was found beneficial in AD animal models (154). Another study demonstrated that non-PD-1 expressing T-regs were upregulated in mild cognitive impairment (MCI), demonstrating the beneficial role of PD-1 receptors in preventing the development of full-blown AD (155). Indeed, tolerized microglia also express PD-1, activating T-regs and probably facilitating their CNS import (156, 157).

These findings suggest that other co-inhibitory receptors, including TIGIT, may activate T-regs. Interestingly, a recent study found elevated TIGIT in the peripheral blood of healthy elderly individuals (158). As upregulated A1 astrocytes were also found in normal aging, TIGIT and A1 may comprise novel AD risk markers (45). In addition, as TIGIT modulates immunity via the iron-lowering IL-10, this receptor may be elevated in normal aging in a compensatory manner in response to iron overload (149). TIGIT is yet to be demonstrated on glial cells, however like PD-1, it may be a coordinator of adaptive and immune responses.

Taken together, this data suggests that innate and adaptive immunity operate on a continuum as trained immunity may activate CTCs, while tolerized immunity the T-regs, therefore targeting trained and tolerized immunity may lower the CNS infiltration of peripheral lymphocytes.

# MICROGLIAL REPROGRAMING: TRAINED AND TOLERIZED THERAPIES

The recent FDA approval of immune training-inducers prompted research for similar therapies for cancer and neurodegeneration (18). Immune stimulatory vaccines, comprised of antigen and pro-inflammatory molecules, activate TLRs or NLRP3, inducing immune activation. These therapeutics have been utilized to train APCs and reverse cancer or sepsisinduced T cell anergy. Iron as an inducer of trained immunity has been utilized to activate TAMs and several nanoparticles are currently being designed for its delivery (14). On the other hand, the development of specific tolerized therapies, including "inverse vaccines," capable of improving pathology without causing generalized immunosuppression, has been vexing (20). The same is true regarding the first-generation AD vaccines, like AN1793, based on anti-Aβ1-42 antibodies which was abandoned due to adverse effects (159). However, despite these draw backs, a new generation of AD vaccines, including anti-C1q mentioned above, are currently being developed. Since neuroinflammation represents the predominant pathology in many neurodegenerative and even psychiatric disorders, de-escalating iron-trained microglia may comprise an effective treatment strategy. Indeed, the anti-inflammatory iron chelator deferiprone is currently in phase 2 trials for AD (15, 160). Deferiprone may have a dual mechanism of action, lowering trained immunity and augmenting tolerization, suggesting that simultaneously targeting both arms of innate immunity may yield better therapeutic outcomes than addressing each one in isolation. Another dual-mechanism-of-action compound, hydroxamic acid, is a promising histone deacetylase-6 inhibitor (HDAC6i) which presents with the added benefit of being an iron chelator (161, 162). Histone deacetylases are enzymes involved in the removal of acetyl groups from histone proteins, altering the chromatin landscape and gene expression. HDACs have been associated with AD pathogenesis, while some HDACis were found to improve cognition (163, 164). In addition, selective HDAC inhibitors were shown to suppress LPS-trained microglia via IL-6 and TNF-α, suggesting a therapeutic potential in AD neuroinflammation (165). Moreover, HDACs 1 and 2 inhibitors are augmenters of microglial β amyloid phagocytosis, linking these compounds to the immuno-metabolic reprograming via OXPHOS augmentation (165, 166).

# INFLAMMASOME INHIBITORS

NLRP3 inhibition to lower immune training and the subsequent neuroinflammation is an emerging therapeutic strategy in AD and traumatic brain injury (TBI) (167–169).

Purinergic receptor blockers, ASC inhibitors, diarylsulphonylurea inhibitors, β-sulfonyl nitrile compounds, and caspase-1 blockers are the broad categories of recently developed NLRP3 inhibitors (170, 171). The most studied compound, MCC950 is a diarylsulphonylurea inhibitor, a potent anti-inflammatory drug and a facilitator of β amyloid clearance in AD animal models (172). Another recently developed molecule, OLT1177, is a β-sulfonyl nitrile which inhibits NLRP3 by limiting ROS generation after LPS challenge. This compound is currently in phase I clinical trials.

Suppressing NLRP3 by ASC inhibition is an attractive AD strategy as ASC was recently found to disseminate β amyloid throughout the CNS (66, 172). Interestingly, saxagliptin, a dipeptidyl peptidase-4 (DPP-4) inhibitor, which increases plasma GLP-1 concentration, appears to selectively inhibit ASC, suggesting a role in preventing the spread of AD pathology (173). This is significant since GLP-1R agonists were found to inhibit the conversion of trophic into A1 astrocytes, probably linking ASC to this neurotoxic phenotype (1, 46). Moreover, Tranilast, an anti-allergic drug derived from tryptophan, was also reported to selectively inhibit ASC, suggesting a preventive potential in AD (174).

# OF EXOSOMES AND INFLAMMASOMES

Exosomes are (50–150 nm) extracellular nanovesicles released by most cell types, including neurons and glia, which can carry molecular cargos and trigger functional changes in target cells. Exosomes cross the BBB, facilitating the transfer of inflammation as they can carry a variety of peripheral molecules, including inflammasome-derived proteins (175, 176).

Exosomes, involved in innate immune cells' cross talk, participate in microglial training or tolerization by augmenting or lowering neuroinflammation, depending on their cells of origin (17, 177, 178). Novel studies have found that neuronal exosomes have beneficial effects, while astrocytesderived exosomes (ADE) are mostly detrimental, inducing neurodegeneration. Neuronal exosomes present with the adaptive function of clearing β amyloid by delivering it to microglia for phagocytosis. For this reason, pharmacological facilitation of neuronal exosomes' synthesis could be developed into an AD therapy (176). For example, designer neuronal exosomes were recently developed and utilized in animal models of Parkinson's disease (170). Interestingly, microglial cells were shown to release exosomes that were previously reported to originate from B cells and dendritic cells, suggesting pro-inflammatory roles (179). CNS autoimmune disorders have been associated with exosomes transporting histocompatibility complex (MHC) class II proteins released by interferon-γ activated microglia (180).

ADE were shown to promote the conversion of trophic to A1 astrocytes by releasing exosomes containing complement components C1q and C3 (126, 181–183). Preventing ADE release may comprise another potential AD strategy. Moreover, a novel study has demonstrated that excessive interstitial ATP can facilitate the generation of A1 astrocytes, suggesting clearance of ATP from the extracellular space (ECS) as a potential AD treatment modality (174). Interestingly, the natural iron chelator, lactoferrin, was found to facilitate the ECS removal of ATP, indicating a novel therapeutic application (184). Indeed, a recent study suggested that engineered lactoferrin-containing exosomes could accomplish this goal (185).

# HISTONE DEACETYLASES INHIBITORS (HDACI)

Innate immune cells are endowed with an epigenetically regulated memory of previous antigen encounters, suggesting that epigenome manipulation may comprise an efficient reprograming strategy in these cells (17). Acetylation and deacetylation of DNA and histone proteins are the most studied epigenetic mechanisms regulating innate immune cells. Indeed, HDACi were shown to lower the LPS-induced microglial training, promoting tolerization (165). For example, valproic acid, a HDACi utilized in the treatment of bipolar disorder and epilepsy, was shown to downregulate the mammalian target of rapamycin (mTOR), restoring OXPHOS, the driver of immune tolerance (11, 186–188). Furthermore, HDACi were recently shown to augment astrocytic ApoE secretion (189). Since ApoE inhibits NLRP3 activation, HDAC inhibition may have a potential therapeutic value in the fight against AD neuroinflammation (**Figure 3**).

HDACi are facilitators of ferritin storage, lowering intracellular free iron and the risk of NLRP3 activation (161, 190). In addition, HDACi reverse the iron-induced augmentation of glycogen synthase kinase-3β (GSK3β), a proinflammatory enzyme implicated in AD (191, 192). GSK-3β inhibitors, including lithium, were shown to promote immune tolerance, suggesting a therapeutic role in lowering iron levels (193–195). Interestingly, the tolerizing receptor TREM-2 is a negative regulator of GSK-3β, further associating this protein with decreasing iron-induced neuroinflammation (191, 196).

# CO-INHIBITORY T-CELL RECEPTORS IN AD

Dysfunctional co-inhibitory receptors may contribute to AD neuroinflammation as under normal circumstances, they block co-stimulatory signals, lowering inflammation (153). For example, stimulation of PD-1 receptors, expressed on both microglia and T-regs, was found beneficial in AD animal models (154). On the other hand, increased number of non-PD-1 expressing T-regs in MCI, shows the importance of this co-inhibitory receptor in AD prevention (155).

T cell activation by APCs is crucial for mounting adequate adaptive immune responses. In the CNS, microglia and astrocytes can become APCs, engendering immunologic synapses with peripheral T cells (10). The proliferation, differentiation, and migration of these cells is triggered by the interplay between co-stimulatory and co-inhibitory signals which fine tune T cell responses (197). For example, the co-stimulatory receptors promote effector T-cells function, while co-inhibitory receptors function to inhibit co-stimulatory signals, promoting T-regs (6).

Activation of co-inhibitory receptors, including CTLA-4, PD-1, LAG-3, TIM-3, and TIGIT, induce immune tolerance, therefore their stimulation would be beneficial in autoimmune disorders (198). On the other hand, their blockade was shown to train TAMs in a manner similar to iron and eliminate cancer cells (199). Interestingly, TIGIT competes with the co-stimulatory signal CD226 for a common ligand CD155, suggesting that promoting the TIGIT-CD155 axis may lower intracellular iron as TIGIT operates exclusively via the iron-lowering IL-10 (200). Indeed, IL-10 or CD-155 loaded exosomes may enhance the co-inhibitory receptor TIGIT, lowering microglial intracellular iron. In addition, the CD226 genetic variant, Gly307Ser, was associated with several autoimmune disorders, suggesting that CD 226 inhibition may facilitate tolerized immunity and benefit the treatment of these conditions as well (201, 202).

# TOLERIZED IMMUNITY AND ACETYLCHOLINESTERASE INHIBITORS

Acetylcholinesterase inhibitors (AChEIs) are cholinergic compounds widely utilized in AD. AChEIs, including donepezil, rivastigmine, and galantamine function by inhibiting cholinesterase, the acetylcholine-degrading enzyme, increasing its synaptic availability in the cholinergic tracts. Some of these drugs may be modulators of trained or tolerized immunity as they oppose iron-induced NLRP3 activation and the inhibition of ADAM-10, the protective α-secretase, in charge of cleaving both TREM-2 and APP (203–205) (**Figure 4**). In addition, α7 and α9 nicotinic acetylcholine receptor (nAChR) agonists were found to lower autoimmune inflammation, by augmenting tolerization (206, 207). Moreover, activation of α7nAChRs was found to reverse post-surgical cognitive impairment, suggesting that AChEIs-linked cognitive enhancement may be due to their anti-inflammatory actions (208). On the other hand, a recent study demonstrated that iron and ROS can inhibit α7nAChRs signaling, likely disrupting tolerized immunity (209, 210). Interestingly, conjugation of an AChEI, such as galantamine, with the natural iron chelator, lactoferrin, was recently proposed as an AD therapy (211).

# TOLERIZED IMMUNITY AND SELECTIVE SEROTONIN REUPTAKE INHIBITORS (SSRIs)

Serotonin reuptake inhibitors (SSRIs) are widely prescribed antidepressant drugs that block 5-hydroxytryptamine reuptake in the presynaptic neurons of brain serotonergic tracts, augmenting the synaptic availability of this neurotransmitter. However, these compounds were found to also lower trained immune responses. For example, a recent study indicated that the SSRI, fluoxetine reduces the central and peripheral levels of IL-1β by inhibiting NLRP3 (212, 213).

Another group demonstrated the existence of a cross talk between the brain innate immunity and serotonin signaling, possibly linking major depressive disorder with NLRP3 activation (214). Along the same lines, it was recently shown that indoleamine 2,3-dioxygenase-1 (IDO-1), an enzyme associated with tryptophan catabolism, is abundantly expressed by microglia and astrocytes, connecting serotonergic signaling with tolerized immunity (215). IDO-1 induces immune tolerance by processing tryptophan to L-kynurenine, an inhibitor of immune cells' proliferation and maturation (216). IDO-1 is known for orchestrating mother's immune acceptance of the fetus, however some bacteria and cancers were found to exploit this pathway (217, 218). Indeed, an IDO vaccine to block cancer-induced immune tolerance is currently under development (219). Moreover, IDO-1 contributes to the biosynthesis of quinolinic acid (QUIN), a neurotoxic tryptophan catabolite that was linked to AD pathogenesis (220). Interestingly, iron-QUIN complexes were demonstrated in this neurodegenerative disorder, indicating that QUIN toxicity may be iron-related (221, 222). QUIN was demonstrated to activate microglia and astrocytes, resulting in neuronal death, possibly by inducing A1 cells (223–225).

# CONCLUSIONS

The healthy immune system must keep trained and tolerized immunity in balance to ensure effective defenses against pathogens and malignancies, while at the same time accepting the self-components, food, the fetus, and commensal microbes. Iron interferes with this process as it triggers a competition between the host and pathogens. Both biological systems have developed mechanisms to capture iron and deny it to the opposing side. The innate immune cells sequestrate iron to restrict its availability to pathogens, but elevated intracellular iron increases the risk of mitochondrial damage, ROS generation, and autoimmune inflammation.

Aging and iron are associated with increased permeability of both GI and brain microvessels, allowing gut microbes, including B. fragilis, to migrate into the CNS and shed LPS, disrupting immune tolerization. Manipulation of trained and tolerized immunity, especially via exosomes and nanoparticles loaded with iron chelators could represent future AD strategies (76). Indeed, iron-oxide loaded exosomes for magnetic hyperthermia have been successfully used in oncology (226).

Tolerogenic immunization also known as "inverse vaccines," can for example, utilize CD-155 loaded exosomes to enhance the co-inhibitory receptor TIGIT and lower AD neuroinflammation (227). In addition, nanoparticles containing NLRP3-inhibitors, HDACi or ASC blockers may soon become parts of the AD armamentarium.

# AUTHOR CONTRIBUTIONS

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

# REFERENCES


inflammation and atherosclerosis in apoE-deficient mice. Lab Invest (2017) 97:922–34. doi: 10.1038/labinvest.2017.30


signaling in the retinal pigment epithelium. J Biol Chem. (2015) 290:11918– 34. doi: 10.1074/jbc.M115.645903


current situation and future expectations. Int J Mol Sci. (2017) 18:E802. doi: 10.3390/ijms18040802


**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 Sfera, Gradini, Cummings, Diaz, Price and Osorio. 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.

# Aceruloplasminemia: Waiting for an Efficient Therapy

#### Alberto Piperno1,2 \* and Massimo Alessio<sup>3</sup>

<sup>1</sup> Department of Medicine and Surgery, University of Milano-Bicocca, Monza, Italy, <sup>2</sup> Medical Genetic Unit, San Gerardo Hospital, ASST-Monza, Monza, Italy, <sup>3</sup> Division of Genetics and Cell Biology, IRCCS-Ospedale San Raffaele, Milan, Italy

Aceruloplasminemia is an ultra-rare hereditary disorder caused by defective production of ceruloplasmin. Its phenotype is characterized by iron-restricted erythropoiesis and tissue iron overload, diabetes, and progressive retinal and neurological degeneration. Ceruloplasmin is a ferroxidase that plays a critical role in iron homeostasis through the oxidation and mobilization of iron from stores and subsequent incorporation of ferric iron into transferrin (Tf), which becomes available for cellular uptake via the Tf receptor. In addition, ceruloplasmin has antioxidant properties preventing the production of deleterious reactive oxygen species via the Fenton reaction. Some recent findings suggest that aceruloplasminemia phenotypes can be more heterogeneous than previously believed, varying within a wide range. Within this large heterogeneity, microcytosis with or without anemia, low serum iron and high serum ferritin, and diabetes are the early hallmarks of the disease, while neurological manifestations appear 10–20 years later. The usual therapeutic approach is based on iron chelators that are efficacious in reducing systemic iron overload. However, they have demonstrated poor efficacy in counteracting the progression of neurologic manifestations, and also often aggravate anemia, thereby requiring drug discontinuation. Open questions remain regarding the mechanisms leading to neurological manifestation and development of diabetes, and iron chelation therapy (ICT) efficacy. Recent studies in animal models of aceruloplasminemia support the possibility of new therapeutic approaches by parenteral ceruloplasmin administration. In this review we describe the state of the art of aceruloplasminemia with particular attention on the pathogenic mechanisms of the disease and therapeutic approaches, both current and perspective.

Keywords: aceruloplasminemia, neurodegeneration, iron overload, ferroxidase, diabetes, ceruloplasmin, iron chelation, plasma

# INTRODUCTION

Aceruloplasminemia (ACP) was firstly described in 1987 as an autosomal recessive disease caused by an inactivating mutation of the ceruloplasmin gene (CP) (Miyajima et al., 1987). The rarity of the disease (estimate prevalence about 1:2,000,000) is a major limit for a comprehensive definition of phenotype, genotype-phenotype associations, therapeutic efficiency, and development of disease markers and drugs. Indeed, the typical manifestations make ACP a unique iron overload disease being: (i) The only one among the Neurodegeneration with Brain Iron Overload disorders, to whom ACP belongs, manifesting systemic iron overload; (ii) The unique systemic iron overload disease

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Scott Ayton, Florey Institute of Neuroscience and Mental Health, Australia Emanuela Tolosano, Università degli Studi di Torino, Italy

\*Correspondence:

Alberto Piperno alberto.piperno@unimib.it

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 26 September 2018 Accepted: 19 November 2018 Published: 04 December 2018

#### Citation:

Piperno A and Alessio M (2018) Aceruloplasminemia: Waiting for an Efficient Therapy. Front. Neurosci. 12:903. doi: 10.3389/fnins.2018.00903

**102**

characterized by neuropathy as major cause of morbidity and microcytic anemia with low serum iron and transferrin (Tf) saturation as a common manifestation. This phenotype is related to the function of ceruloplasmin (Cp) in iron homeostasis at both systemic and brain levels. The lack of Cp induces cellular iron retention and progressive overload on the one hand, and low cellular iron release leading to iron-restricted erythropoiesis and iron deficiency anemia, on the other hand. However, the physiopathology of organ damage, as well as the pathways and efficiency of chelators in brain iron removal, are highly controversial issues which remain to be fully elucidated.

# CLINICAL HETEROGENEITY OF ACERULOPLASMINEMIA

Recent studies showed that ACP phenotypes are more heterogeneous than previously believed and that some manifestations may precede neurological manifestation even by decades (Pelucchi et al., 2018). Thus, this could allow early diagnosis, hopefully preventing the development of neurological derangements, which are the most severe complication of this disease. Neurological symptoms usually appear in the fifth decade of life and vary within a wide spectrum that includes cerebellar ataxia, involuntary movements, parkinsonism, mood and behavior disturbances, and cognitive impairment (McNeill et al., 2008; Kono, 2012). As magnetic resonance imaging (MRI) is increasingly used in neurological diagnostics, brain iron accumulation can be found in a few patients with neuropsychiatric symptoms addressing diagnosis to ACP. However, ACP patients without brain iron overload or patients with brain iron overload without or with only very mild neurological manifestations even after 50 years of age have been reported (Skidmore et al., 2008; Kassubek et al., 2017; Pelucchi et al., 2018), suggesting that genetic and/or acquired factors may partially modify neurologic phenotype.

Retinal manifestations are reported in over 75% of Japanese patients (Miyajima, 1993; Kono, 2012), but these were less frequent in Italian patients (Pelucchi et al., 2018). However, accurate descriptions of retinopathy are available in only two patients (He et al., 2007) ranging from retinal discoloration to macular degeneration, indicating that further studies are needed to better characterize ACP-dependent retinopathy.

Diabetes mellitus is considered an early manifestation of ACP, being reported as the first symptom in 68.5% of patients at a median age of 38.5 years in a recent overview. Diabetes was insulin dependent in 60% of patients, undefined in 30%, and less than 10% orally and/or dietary treated (Vroegindeweij et al., 2015). Little information is available on family susceptibility, concomitant overweight, and beta-cells function, which would be useful in understanding diabetes pathogenesis in ACP.

Often not included in the classical triad of ACP (diabetes, retinopathy, and neuropathy), anemia and/or microcytosis are common manifestations being reported in 80% of ACP patients in Japan (Miyajima, 1993) and recorded as the earliest manifestations of disease in 86% of Italian patients, preceding diagnosis even by decades (Pelucchi et al., 2018). However, in a retrospective review of reported cases, anemia was recorded as the first symptom only in 24% of patients (Vroegindeweij et al., 2015). The high frequency of mild anemia and/or microcytosis in the general population, which are likely to be related to iron deficiency or thalassemia traits, can lead to the underestimation of these signs as early ACP-related manifestations. Thus, despite ACP rarity, Cp measurement should be included in the diagnostic work-up of anemia. The presence of low serum iron and Tf saturation with normal/high serum ferritin may serve to further enhance disease suspicion.

# PHYSIOPATHOLOGY OF ACERULOPLASMINEMIA

Ceruloplasmin is part of the multicopper ferroxidase family, a group of tissue specific proteins (Cp, hephaestin, and zyklopen) that facilitate cellular iron efflux in conjunction with the membrane ferrous iron exporter ferroportin (Fpn), by oxidizing ferrous iron to the ferric state (Vashchenko and MacGillivray, 2013). Cp is recognized as a serum protein secreted by the liver, but it has also been found as glycosylphosphatidylinositol (GPI)-linked protein in astrocytes and leptomeningeal cells in the central nervous system (CNS), but also in macrophages, hepatocytes and many other tissues (Musci et al., 2014). Cp-KO mice revealed an impairment in hepatocyte and reticuloendothelial iron efflux (Harris et al., 1999), and showed that Cp-GPI is relevant for brain iron metabolism in regulating iron efflux from astrocytes (Jeong and David, 2003). A molecular connection between Cp and Fpn has been established with the finding that ferroxidase activity is required to stabilize Fpn at the cell surface in cells expressing Cp-GPI (De Domenico et al., 2007). Thus, Cp can be considered as a second determinant of Fpn stability after hepcidin. These findings suggest that differently from most disorders of iron overload that reflect changes in the absolute amount of iron, ACP results, at least in early stages, from iron imbalance caused by impairment in the rate of iron efflux from storage sites. Serum hepcidin is decreased in ACP patients, possibly related to the low Tf saturation, suggesting that suppressed hepcidin synthesis might contribute to the development of iron overload (Kaneko et al., 2010). While it is clear that microcytosis, anemia and low serum iron depend on reduced iron availability due to cellular iron sequestration, the development of diabetes and neurodegeneration in ACP are not evidently explained at cellular and molecular levels. It has been suggested that both diabetes and neurodegeneration are secondary to iron accumulation and toxicity (Kono, 2012), however, no clear evidence of intracellular iron deposition in the endocrine pancreas and in neurons has been reported in neither ACP patients nor mice models of ACP (Kato et al., 1997; Jeong and David, 2006; Chen et al., 2018). These findings suggest that neurons and islet beta-cells distress might be related to an elevated sensitivity/vulnerability to iron toxicity or to the iron

deficient status induced by iron retention in the surrounding cells (**Figure 1**).

# Physiopathology of Brain Iron Overload and Damage

Iron enters the CNS via the brain barrier systems, namely the blood brain barrier (BBB) and the blood-cerebrospinal fluid barrier (BCSFB), which separate the blood from the interstitialand cerebrospinal-fluid (CSF), respectively. The mechanism of iron uptake/export in cells depends on the iron transport molecules they express. Transferrin-bound-iron, internalized by transferrin receptor 1 (TfR1) in brain capillary endothelial cells (BCECs), can cross the BBB by: (1) transcytosis and direct release as ferric iron in the brain parenchyma, or (2) iron release from endosomes to cytosol via divalent metal ion transporter 1 (DMT1), and exported via Fpn assisted by the GPI-Cp expressed on perivascular astrocytes (McCarthy and Kosman, 2015; Simpson et al., 2015). The absence of iron accumulation in BCECs in both ACP patients and murine models indicate transcytosis and direct iron release as the more plausible mechanism in ACP (Miyajima, 1993; Zanardi et al., 2018). The BCSFB comprises choroid plexus epithelial cells (CPEpiCs) that are in contact with the CSF on one side, and with fenestrated endothelial cells on the other. In CPEpiCs iron is internalized via TfR1 and then exported to the CSF by the action of Fpn and Cp, which are both expressed by CPEpiCs (Rouault et al., 2009; Zheng and Monnot, 2012). Large iron deposition in CPEpiCs precedes accumulation in other brain regions in Cp-KO mice (Zanardi et al., 2018) suggesting an early deficit in iron handling by these cells in ACP. However, the pathophysiology of choroid plexus damage in ACP patients is largely undefined.

Neurons acquire iron prevalently from Tf and TfR1 while oligodendrocytes and astrocytes do not express TfR1, which suggests they import iron using other mechanisms (Zhang et al., 2006; Belaidi and Bush, 2016). Astrocytes may acquire iron released by BCECs being in close contact with them via endfeets that highly express DMT1 (McCarthy and Kosman, 2015). However, DMT1 uptake requires ferric iron to be reduced, but the mechanism of iron reduction by astrocytes is still unclear. Oligodendrocytes and astrocytes are rich in ferritin, suggesting that they function as iron storage site for the whole brain (Finazzi and Arosio, 2014). Regardless of import mechanisms, iron efflux from all brain cells involve Fpn, and depends on the extracellular ferroxidase activity fostered primarily by both the membrane Cp-GPI and the soluble Cp circulating in the interstitial fluid and CSF (Patel and David, 1997; Olivieri et al., 2011). Therefore, in ACP, the absence of Cp-ferroxidase activity promotes Fpn internalization and degradation favoring intracellular iron accumulation (Kono et al., 2010; Persichini et al., 2012).

Brain iron accumulation, particularly in the basal ganglia and cerebellum, is reported by MRI in a large number of ACP patients (Grisoli et al., 2005; Kono, 2012). Nevertheless, the few autopsy studies performed indicated that iron deposition was more evident in glial and astrocytes than in neurons, in particular in the perivascular regions of brain areas where neuronal loss is prevalent (Kawanami et al., 1996; Kaneko et al., 2002, 2012; Oide et al., 2006). Thus, it has been hypothesized that

apoptosis/ferroptosis of iron-engulfed astrocytes/glial cells might release large amounts of toxic ferrous iron promoting oxidative neuronal damage, as inferred by large lipids peroxidation observed in the brain of ACP patients (Miyajima et al., 1998, 2001a; Kohno et al., 2000; Yoshida et al., 2000). Neuronal damage could be aggravated by the loss of glial-derived growth factors necessary for neuronal survival (Jeong and David, 2006). It is also conceivable that neuronal injury might depend on iron deficiency as the consequence of impaired mobilization of the accumulated iron in glial cells (Jeong and David, 2006; Kono, 2013).

#### Physiopathology of Systemic Iron Overload and Damage

In the liver of ACP patients, iron accumulation prevails in hepatocytes, a surprising finding considering the role of Cp in regulating cellular iron efflux via Fpn. (Morita et al., 1995; Kawanami et al., 1996; Hellman et al., 2000; Bosio et al., 2002; Loreal et al., 2002; Finkenstedt et al., 2010; Rusticeanu et al., 2014). Indeed, liver iron distribution differs in patients with classical Fpn disease and ACP, being prevalent in reticuloendothelial cells in the former and in hepatocytes in the latter, indicating the need to expand our knowledge on the relative function of Fpn and Cp and their interactions in different tissues. Despite hepatocellular iron overload, cirrhosis is uncommon in ACP, however, some patients developed moderate/severe fibrosis in the presence of heavy hepatic iron overload (Pelucchi et al., 2018). This suggests that in ACP patients the risk of liver fibrosis depends on the iron burden, a similar scenario to hereditary hemochromatosis (HH) (Wood et al., 2008), and that liver iron overload in ACP patients generally does not reach the threshold associated with liver damage risk.

Clinically, diabetes may precede neurological derangements even by decades, but the mechanism of endocrine pancreatic damage in ACP is unclear. Iron accumulation in the exocrine pancreas has been commonly reported in mouse models characterized by systemic iron loading (hypotransferrinemic mice, bone morphogenetic protein 6, hemojuvelin-, and hepcidin-deficient strains) (Chen et al., 2018). About 20% of HH patients show diabetes, all of them being heavy iron overloaded and the majority having liver fibrosis/cirrhosis (McClain et al., 2006). In HH-patients iron deposition is present in the exocrine pancreas and, varied from case to case, in beta-cells together with loss of endocrine granules (Rahier et al., 1987). In mouse models of HFE-HH, iron accumulates specifically in the endocrine pancreas resulting in increased beta-cell death (Cooksey et al., 2004). The total pancreaticand per islet-insulin content were low in Hfe-KO mice (Huang et al., 2011), suggesting an iron-related beta-cell oxidative stress and decreased insulin secretory capacity secondary to betacell apoptosis and desensitization of glucose-induced insulin secretion (Backe et al., 2016). However, in other mouse models of hemochromatosis, the endocrine pancreas was unaffected. Hepcidin-KO mice developed chronic pancreatitis owing to exocrine iron overload (Lunova et al., 2017), and, similarly, a murine model of Fpn mutant resistant to hepcidin binding showed exocrine pancreatic failure and iron overload (Altamura et al., 2014). Iron deposition and functional deficit restricted to the exocrine pancreas has also been reported in the double Cp/Hephaestin-KO mice model that did not display structural or functional (insulin secretion) deficiency of pancreatic islets (Chen et al., 2018).

Few studies describe pancreatic alteration in ACP patients mainly reporting generic iron accumulation without histological description of the cells and structures involved (exocrine vs. endocrine) (Daimon et al., 1995; Kawanami et al., 1996). Three reports described the absence or mild iron accumulation in the islets compared to the exocrine pancreas (Morita et al., 1995; Kato et al., 1997; Miyajima et al., 2001b). In a single ACP patient, an autopsy study showed marked reduction of insulin-containing cells without iron accumulation and degeneration/loss of islet cells, despite massive iron deposition in the exocrine pancreas (Kato et al., 1997). The patient's long type-2 diabetes history may suggest beta-cell exhaustion as a possible explanation for such findings. Nevertheless, the pathogenesis of diabetes in ACP and the role of iron remain elusive, also considering that

#### TABLE 1 | Therapeutic approaches for aceruloplasminemia.

Doyle et al., 2015

Pelucchi et al., 2018

#### Iron chelation

Deferoxamine Deferasirox Deferoxamine-Deferiprone Miyajima et al., 1997 Skidmore et al., 2008 Fasano et al., 2008 Yonekawa et al., 1999 Bethlehem et al., 2010 Mariani et al., 2004 Loreal et al., 2002 Roberti Mdo et al., 2011 Badat et al., 2015 Haemers et al., 2004 Suzuki et al., 2013 Bove and Fasano, 2015 Finkenstedt et al., 2010 Tai et al., 2014 Poli et al., 2017 Hida et al., 2010 Rusticeanu et al., 2014 Pelucchi et al., 2018 Pan et al., 2011 Lindner et al., 2015 Deferiprone Deferiprone-Deferasirox Deferoxamine-Deferasirox Pelucchi et al., 2018 Pelucchi et al., 2016 Calder et al., 2017

#### Zinc administration

Kuhn et al., 2007

Minocycline administration

Hayashida et al., 2016 Iron chelation + Vitamin E and C

Kuhn et al., 2007 Pelucchi et al., 2016 Pelucchi et al., 2018

#### Iron chelation + Fresh Frozen Plasma

Logan et al., 1994 Yonekawa et al., 1999

Poli et al., 2017

Iron chelation + Fresh Frozen Plasma with high Cp level "under development"

Enzyme Replacement Therapy

Harris et al., 1999\*

Zanardi et al., 2018\* Gene Therapy

"desirable"

<sup>∗</sup>Preclinical model.

diabetes can occur early when systemic iron overload is still minor (Hatanaka et al., 2003; Muroi et al., 2006; Pelucchi et al., 2018).

# TREATMENT OF ACERULOPLASMINEMIA

fnins-12-00903 November 30, 2018 Time: 15:11 # 5

Current treatment in ACP (**Table 1**) is primarily focused on reducing iron overload using iron chelators (deferasirox, deferoxamine, and deferiprone). In the majority of patients, iron chelation therapy (ICT) was effective in reducing systemic iron overload, but there is no information on whether it can improve glucose metabolism derangement and retinopathy. ICT was less or not effective on neurological symptoms [reviewed in (Kono, 2013; Dusek et al., 2016)]. Thus, there is the need to find treatments efficient in rescuing/preventing neurological manifestations, taking advantage of the window between the appearance of the first manifestations and neurological symptoms (Vroegindeweij et al., 2015; Pelucchi et al., 2018). The variable efficacy of ICT on neurological symptoms may depend on the ability of different chelators to cross the BBB, the timeliness of the treatment, the large heterogeneity of therapeutic protocols, and the unclear kinetics of iron chelators in the brain in vivo [reviewed in (Dusek et al., 2016)]. In the large majority, therapeutic efficacy was evaluated in a short time, and the few data available in long-term follow-up studies showed progression of neurological derangements (Pelucchi et al., 2018). ICT seems more promising if started before the onset of neurological symptoms, although the majority of the cases that remained asymptomatic have not yet reached the risk age [reviewed in (Dusek et al., 2016)]. Also, ICT must be often discontinued because of the aggravation of functional iron deficiency anemia, limiting the long-term therapy required to mobilize iron from the brain (Loreal et al., 2002; Mariani et al., 2004; Fasano et al., 2008; Finkenstedt et al., 2010). Due to the antioxidant properties of zinc sulfate and minocycline, the inhibitory effect on iron absorption of zinc sulfate, and chelating properties of minocycline iron, these have been proposed as alternatives to ICT when the latter must be discontinued, but the results, although promising, are limited to only two patients (Kuhn et al., 2007; Hayashida et al., 2016). To prevent tissue damage, antioxidants like vitamin E and C have been used along with ICT (Kuhn et al., 2007; Pelucchi et al., 2018).

In some cases, ICT was combined with fresh-frozen plasma (FFP) administration. Due to the half-life of Cp (5.5 days) (Hellman and Gitlin, 2002), FFP administration could partially/temporarily restore circulating Cp. In two cases, the combined therapy improved neurological symptoms, visibly reducing brain iron deposition (Yonekawa et al., 1999; Poli et al., 2017). This implies that circulating Cp can enter the CNS and be functionally effective. Since physiological serum Cp concentration ranges between 21 and 54 mg/dL (Gibbs and Walshe, 1979; Gaasch et al., 2007), the selection of high-Cp-content FFP transfusions might be more effective, thereby advancing this therapeutic approach. A post-transfusion Cp level of 8–10 mg/dL, comparable to that of ACP heterozygotes, could be enough to rescue iron homeostasis as inferred by the absence of clinical symptoms in the vast majority of heterozygous subjects (Miyajima, 1993; Kono, 2013). Nevertheless, the risks associated with long-term repeated transfusion could limit this approach.

A neuroprotective effect of ceruloplasmin administration was reported in various pathological models (Ayton et al., 2013, 2014; Tuo et al., 2017; Zanardi et al., 2018). Indeed, it has been demonstrated that intraperitoneally administered Cp was able to enter the Cp-KO brain crossing the barrier systems (Ayton et al., 2013; Zanardi et al., 2018). The potential of the Cp-enzyme replacement therapy (ERT) in reducing neurological manifestation was recently demonstrated in the preclinical model of ACP (Zanardi et al., 2018). The ferroxidase activity was restored in Cp-KO brain, inducing reduction in iron deposition, rescue of neuronal loss, and amelioration of motor incoordination (Zanardi et al., 2018). The mechanism by which administered Cp crosses the brain barrier systems in Cp-KO, but not in wild-type mice, is not known. However, the iron accumulation found in the choroid plexus of Cp-KO mice suggested that Cp enters the brain likely through BCSFB impairment (Zanardi et al., 2018). A similar impairment would also explain the effectiveness of the FFP therapy in ACP patients. Cp administration also mobilizes iron and temporarily restores iron homeostasis in Cp-KO mice (Harris et al., 1999; Zanardi et al., 2018). Thus, ERT seems to be efficacious on both systemic and neurological symptoms, and may be a potential therapeutic opportunity in humans.

As for many monogenic rare diseases, the ideal ultimate cure consists in gene therapy approaches to correct the defect, but further studies and several years will be necessary before its clinical application.

# CONCLUSION

In conclusion, current therapies in ACP are effective in reducing systemic iron overload but have little or no effect on neurological symptoms. Their capability in improving pancreatic and retinal damage is unclear, and might have deleterious side effects that limit long-term therapy. FFP administration can be a suitable source of functional Cp, and although cost-benefit effects need to be further evaluated, can represent a step toward the development of more efficient therapies in ACP.

# AUTHOR CONTRIBUTIONS

AP and MA contributed the same to conception and writing of the work.

#### FUNDING

Associazione per lo Studio dell'Emocromatosi e della Malattie da Sovraccarico di Ferro-ONLUS will pay for the review.

The Association was funded in Monza in 1996 and is part of the European Federation of Associations of Patients with Hemochromatosis (EFAPH). This no-profit Italian Association is involved in supporting information to patients, general practitioners, specialized physician, institutions, research in the field of iron overload disorders.

#### REFERENCES


#### ACKNOWLEDGMENTS

We thank the Associazione per lo Studio dell'Emocromatosi e delle Malattie da Sovraccarico di Ferro (+Fe – ONLUS) for supporting our work. We also thank Dr. Serena Melgrati, Imperial College London, for English language editing of the manuscript.




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

Copyright © 2018 Piperno and Alessio. 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.

# UPR Induction Prevents Iron Accumulation and Oligodendrocyte Loss in ex vivo Cultured Hippocampal Slices

#### Sinead Healy, Jill McMahon and Una FitzGerald\*

Galway Neuroscience Centre, School of Natural Sciences, National University of Ireland Galway, Galway, Ireland

The accumulation of iron within the brain occurs in many chronic disorders including Alzheimer's and Parkinson's disease and multiple sclerosis. Outside the CNS, a link between levels of iron and the unfolded protein response has already been established. To determine if such a relationship operates in within the brain, we used our ex vivo hippocampal slice-based model of iron accumulation. Ferrocene addition caused accumulation of iron within slices and loss of oligodendrocytes, an effect that was partially inhibited when ferrocene and ER stressor tunicamycin (Tm) were added together. An upward trend (not found to be statistically significant) in the expression of UPR transcripts in response to ferrocene was demonstrated using real-time PCR, while a significant upregulation of mRNA for B cell immunoglobulin-binding protein (BiP) and C/EBP homologous binding protein (CHOP) occurred following exposure to Tm. In silico analysis revealed consensus DNA-binding sequences for UPR-associated transcription factors within the promoter regions of eight iron-regulatory genes. In addition, dualstaining for CHOP and oligodendrocyte transcription factor 2 (OLIG2) or Ionized calcium binding adaptor molecule 1 (Iba1) showed nuclear expression of CHOP in some oligodendrocyte-lineage cells in response to Tm or Tm+ferrocene, but CHOP was rarely found in microglia. Co-expression of UPR-associated activated transcription factor 6 (ATF6) was detected in the nuclei of some oligodendrocyte-lineage cells exposed to Tm alone, or to Tm and ferrocene, but rarely in microglia. These data highlight the therapeutic potential of targeting UPR-associated proteins when developing novel treatments for chronic brain disorders that are affected by dysregulated iron.

Keywords: iron, unfolded protein response, UPR, tunicamycin, ferrocene

# INTRODUCTION

The unfolded protein response (UPR) is a 'check and balance' program activated upon disturbed homeostasis in the endoplasmic reticulum. Occasional but infrequent hints have documented an interplay between the UPR and iron metabolism (Ye and Connor, 2000; You et al., 2003; Oliveira et al., 2009; Vecchi et al., 2009). However, reports are few on the possible links between the UPR

Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Safikur Rahman, Yeungnam University, South Korea Raffaella Gozzelino, Centro de Estudos de Doenças Crónicas (CEDOC), Portugal

> \*Correspondence: Una FitzGerald una.fitzgerald@nuigalway.ie

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 13 September 2018 Accepted: 04 December 2018 Published: 18 December 2018

#### Citation:

Healy S, McMahon J and FitzGerald U (2018) UPR Induction Prevents Iron Accumulation and Oligodendrocyte Loss in ex vivo Cultured Hippocampal Slices. Front. Neurosci. 12:969. doi: 10.3389/fnins.2018.00969

and iron within the brain (Liu and Connor, 2012; Rahman et al., 2017, 2018; Lumsden et al., 2018). A better understanding of this cross-talk might provide further clues regarding disease pathogenesis, given that aberrant iron metabolism and increased expression of markers of the UPR have been independently reported in several neurodegenerative diseases (Hetz and Mollereau, 2014), including multiple sclerosis (Stephenson et al., 2014), Alzheimer's disease and Parkinson's disease (Mercado et al., 2013; Ward et al., 2014). Researchers have exploited glial and neuronal mono- and dual-cultures in vitro, as well as ex vivo systems to investigate the regulation of iron within the brain [see (Healy et al., 2017) for comprehensive review of this topic]. As a foundation for future studies, the overall goal of this work was therefore to confirm and characterize the reciprocal relationship between UPR and iron homeostasis within our hippocampal slice-based model (Healy et al., 2016) of the accumulation of iron within the CNS.

# METHODS

# Ethics Statement

The research described was approved by the Animal Care Research Ethics Committee Filing ID 12/FEB/04) and was carried out in accordance with European directive 2010/63/EU.

# Hippocampal Slice Culture and Reagent Exposure

Neonatal hippocampal slices were maintained as previously described (Healy et al., 2016).

Five µg/ml tunicamycin (Tm, T7765, Sigma-Aldrich, Dublin, Ireland) exposure was used to activate a UPR response, while iron loading was induced by a 12-h treatment with 1 µM ferrocene, a membrane-permeable compound that enables uptake of iron by cells (F408, Sigma-Aldrich, Dublin, Ireland). Slices were harvested immediately after iron loading.

#### Analysis of RNA Transcripts

RNA was isolated from control and treated slice cultures using TRIreagent (T9424, Sigma-Aldrich, Dublin, Ireland) as per the manufacturer's instructions. Following DNAse I treatment (18068-015, Thermofisher-Scientific, Dublin, Ireland) and cDNA synthesis using manufacturer protocol (1804014, Superscript II Reverse Transcriptase, Invitrogen, Dublin, Ireland), real-time PCR was accomplished using Applied Biosystems 4346906 MicroAmp <sup>R</sup> Fast Optical 96-Well Reaction Plate and Fast SYPR green master mix (applied Biosystems, 4385612) and the Step One real-time PCR System (Thermo-Fisher).

Unless otherwise indicated, oligonucleotide primer sequences (Sigma-Aldrich, Dublin, Ireland) used were designed using Primer3. Primer sequence details are summarized in the **Supplementary Table 1**. Results were quantified using the 11CT method following standardization relative to β-actin as an internal control.

# Tissue Viability

Tissue viability was measured as described in Healy et al. (2016), using a lactate dehydrogenase (LDH) and MTT [3-(4,5- Dimethylthiazol-2-yl)-2,5-Diphenyltetrazolium Bromide] assay.

#### Immunohistochemistry and Imaging

Tissue was fixed and stained as previously described (Healy et al., 2016). Briefly, slices were fixed in 4% paraformaldehyde (PFA) for 30 min followed by three washes in PBS and then blocked and permeabilised for 1 h with 10% normal goat serum and 0.4% Triton X-100 in PBS. Subsequently, the slices were incubated with primary antibody (for single and dual labeling) in 2.5% NGS (or 2.5% horse serum) and 0.1% Triton X-100 in PBS for 48 h at 4◦C and then washed three times for 15 min before incubation with the appropriate secondary antibody at 4◦C overnight. Source information and dilutions used for all antibodies is available in the **Supplementary Materials**. Slice cultures were mounted in Vectashield containing diamino-2-phenylindole (DAPI) (H-1200, Vector Laboratories, Peterborough, United Kingdom) to allow visualization of nuclei. Negative "no primary antibody" controls were included. Slides were stored in the dark at 4◦C until imaged. All samples were imaged on a laser scanning confocal microscope (Olympus Fluoview 1000) using a 40 × oil-immersion lens (numerical aperture of 1.3).

# In silico Analysis of UPR Transcription Factor Binding Sites

The ALGGEN PROMO algorithm<sup>1</sup> was used to determine if any UPR transcription factor consensus DNA binding sequences were present within the promoter regions of genes that encode iron homeostasis regulators. Only hits that had a dissimilarity index <15% and a random expectation <0.5 were included in this analysis.

#### Statistical Analysis

Data was analyzed using Prism 6 (GraphPad Software). All measurements were expressed as mean ± standard error of the mean (SEM) unless otherwise indicated. Normality was assessed using the Shapiro-Wilk test. Statistical analyses were carried out using Student's t-test, Mann-Whitney test or an ANOVA with Dunnet post-test, as appropriate, and as indicated in figure legends. Differences were considered statistically significant if P<0.05.

# RESULTS

# UPR Transcript Expression in Cultured Slices Exposed to Ferrocene Remains Unchanged

Real-time PCR analysis of changes in the transcripts for B-cell immunoglobulin binding protein (BiP), calreticulin (CRT), X-box binding protein 1 (XBP1), activated transcription factor

<sup>1</sup>http://alggen.lsi.upc.es/

FIGURE 1 | Cross-talk between regulators of the UPR and iron homeostasis. (A) Real-time PCR analysis of RNA samples derived from cultured slices treated for 12 h with 1 µM ferrocene. An upward trend in the expression of BiP, CRT, XBP1, ATF6, ATF4, and CHOP was detected (n = 3 experimental replicates). (B) Increased BiP and CHOP transcripts detected following 12 h treatment with 5 µg/ml Tm (n = 3). Data is expressed as mean + 95% confidence interval following completion of three experimental repeats. <sup>∗</sup>denotes p<0.05. (C) Upward trend in the expression of FTL and FTH transcripts detected following 12 h Tm treatment (5 µg/ml, n = 2 experimental repeats). (D) Effect of Tm exposure (5 µg/ml, 12 h) on iron and protein content and on tissue viability (n = 3 experimental replicates). Cultures were loaded with 1 µM ferrocene (F), 5 µg/ml Tm or both treatments concurrently (TmF). Data is expressed as percentage of DMSO control, which is indicated by the dashed line. Values are mean + SEM from three experiments. ♦ denotes a significant difference (p<0.05) when compared with DMSO and <sup>∗</sup>denotes significant difference between F-treated slices and those treated with Tm or TMF (p<0.05). (E) CHOP expression (green) was detected within the nuclei of olig2-positive cells (red) in hippocampal slices treated with Tm or TmF (lower panels), but was not seen within the Olig2-positive population of control or F-treated slices (upper panels). Scale bar = 100 µm (F) When cultured slices were dual-labeled using antibodies to Iba1 (red) and CHOP (green, RH panels), CHOP was not found in DMSO-exposed slices but was seen in non-Iba1-positive slices treated with F or TmF and occasionally in Iba1-positive nuclei in slices treated with Tm alone (white arrowheads). Scale bar = 100 µm. (G) The significant loss of oligodendrocyte-lineage cells caused by exposure to F is partially reversed when slices are co-treated with Tm or TmF (n = 2 experimental replicates). ♦denotes a significant difference (p<0.05) when compared with DMSO; <sup>∗</sup>denotes a significant difference between F-treated slices and those treated with Tm or TMF (p<0.05). (H) Cytoplasmic expression of ATF6 (red) detected in the non-olig2-expressing cells of DMSO- and F-treated hippocampal slices (top panels). Treatment with Tm or TmF led to nuclear expression of ATF6 within olig2-positive cells (green, arrows). Scale bar = 50 µm.

6 (ATF6), activated transcription factor and C/EBP homologous binding protein (CHOP) revealed an approximate 1.4–2.1-fold increase, which was not found to be statistically significant following three replicate experiments (**Figure 1A**).

# UPR Induction Does Not Increase Expression of Iron Storage Protein Transcripts

The ability of cultured hippocampal slices to mount a response to ER stress was confirmed when a significant increase in BiP and CHOP transcription (approximately 30-fold for both), was detected following a 12-h treatment with 5 µg/ml tunicamycin (Tm, **Figure 1B**). Tm inhibits N-glycosylation of proteins trafficked through the ER (Foufelle and Fromenty, 2016) hence causing a buildup of mis-folded proteins and ER stress. However, Tm was not found, after two experimental repeats, to cause a significant increase in the expression of ferritin light chain (FTL) or ferritin heavy chain (FHL) transcripts (**Figure 1C**).

# UPR Activation Prevents Ferrocene-Induced Iron Accumulation and Glial Perturbations

In line with our previous findings (Healy et al., 2016), 1 µM ferrocene produced a 1.3-fold increase in the levels of iron in slice cultures when compared to vehicle (**Figure 1D**). Similarly, we confirmed that ferrocene-induced a significant 1.5-fold increase in LDH release, and, once again, did not report any iron-induced loss in metabolic activity as assessed by MTT assay (**Figure 1D**). Interestingly, concurrent exposure to 5 µg/ml Tm (TmF) resulted in amelioration in iron accumulation and toxicity (**Figure 1D**). TmF treatment led to a significant reduction in iron content and a significant decrease in LDH release, when compared with ferrocene alone, but not when compared with the DMSO control. In keeping with our data relating to FTL and FTH transcriptional expression, Tm exposure alone did not lead to a significant difference in iron content. Tm treatment alone also had no effect on tissue viability when compared to DMSO control conditions. There were no significant differences in metabolic activity as measured using an MTT assay, or in protein content between groups (**Figure 1D**).

# UPR Transcription Factor Expression in Glia Following Tm and Iron Exposure

Classically, CHOP is upregulated in response to phosphorylation of PERK and is frequently exploited as a marker of activation of the UPR. Immunofluorescent staining demonstrated an absence of CHOP expression in control and ferrocenetreated slice cultures, but robust upregulation of CHOP in Tm-treated tissue. Dual immunofluorescent staining confirmed CHOP expression in Olig2-positive cells within Tm-treated slices, which appeared to increase when slices were co-exposed to Tm and Ferrocene (**Figure 1E**). Less frequently, CHOP was found to be present in microglia within cultured slices exposed to Ferrocene, Tm or both (**Figure 1F**). Similar to our prior finding of significantly reduced numbers of microglial endpoints, following slice treatment with F, we also found, following two experimental repeats, fewer end-points after F treatment, which was rescued when tissue was treated with TmF (data not shown). ATF6 is the second major regulator of the UPR whose protein expression we analyzed by immunofluorescence within cultured control and treated slices. While non-nuclear ATF6 was detected in control DMSO-treated cultures, an increase in nuclear ATF6 staining was observed following treatment with Tm, F, or a combination of both reagents (not shown).

In our previous study, we reported a significant loss of oligodendrocyte-lineage cells following iron loading (Healy et al., 2016). This result was confirmed here (**Figure 1G**). Intriguingly, slice exposure to Tm or TmF reversed this loss, to the extent that Olig2-lineage cells responded by increasing their number significantly (**Figure 1G**). Dual immunofluorescence was then used to determine if the localization of ATF6 expression within Olig2-positive cells was affected by the same conditions. Our images showed that while nuclear (active) ATF6 was not detectable within DMSO-treated Olig2-lineage cells, it was present in Tm-, F- or TmF-treated cultures (**Figure 1H**).

#### In silico Analysis of Putative Transcription Factor Binding Sites Within Promoter Regions of Iron-Regulatory Genes

Using the ALGGEN PROMO algorithm to predict consensus transcription factor (TF) DNA binding sequences, we identified a number of putative binding sites for UPRassociated transcription within the promoter regions of iron homeostasis genes (**Table 1**). None of the TFs were found to be commonly expressed on all the promoter regions examined. Activated transcription factor 3 (ATF3) and CCAAT-enhancer-binding protein alpha (C/EBP-α) consensus-binding sequences were found on six of the eight genes (**Table 1**). Consensus sequences for XBP1, ATF and ATF2 were found within two promoter regions; the CREB consensus sequence was found on three genes and C/EBP-β was not found on any. Therefore, there was not a consistent pattern of transcription factor hits across the different genes (**Table 1**).

# DISCUSSION

In an attempt to elucidate whether or not cross-talk between iron-regulatory proteins and ER stress, we exploited our rat hippocampal slice-based model of iron accumulation. Following slice exposure to Tm, F or both, real-time

PCR and fluorescent immunohistochemistry yielded the following primary findings: (1) Tm exposure does not affect the level of iron in hippocampal slices, but it triggers a UPR that appears to prevent iron overload and possibly ameliorates toxic effects of iron on oligodendrocytes and microglia; (2) iron overload causes an upward trend in the levels of transcripts encoding the major markers of the UPR, although this was not statistically significant; (3) the observation that Tm treatment led to no significant change in the expression of FTL or FTH transcripts was surprising, since the promoter regions of eight genes that encode iron-regulatory proteins, including FTL and FTH, contain putative binding sites for ER stress-associated transcription factors; (4) transcription factor ATF6 is activated following exposure of cultured hippocampal slices to Tm, F and TmF; (5) Activated ATF6 is detectable within the Tm- and TmF-treated oligodendrocycte population but rarely in similarly treated microglia.

While intriguing, it remains unclear how concurrent iron accumulation and UPR activation, demonstrated here, leads to


Summary of the putative binding site of the UPR-associated transcription factors in the gene sequence of an iron-homeostasis molecule with a dissimilarity index <15% and random expectation <0.5. Transcription factors tested: C/EBP-beta, C/EBP-α, CREB, XBP1, ATF, ATF3, ATF-2, and ATF1.

the ameliorated accumulation of iron. Outside the brain, the main description of the intersection between the UPR and iron metabolism is the action of CHOP on hepcidin (via modulation of C/EBPα binding and/or through CREBH) in the liver and spleen (Oliveira et al., 2009; Vecchi et al., 2009) although ER stress associated chaperones calreticulin and BiP have also been implicated in iron UPR cross-talk [reviewed in (Oliveira et al., 2011)].More recently, an increase in hepcidin expression leading to accumulated iron within the brain of rats subjected to subarachnoid hemorrhage, was found to be regulated by CHOP (Zhao et al., 2018).

Our observation that CHOP immunofluorescence appears to be increased in Olig-2 positive cells in slices that were co-exposed to Tm and Ferrocene compared with only Tm is consistent with these studies and it hints at the potential therapeutic benefit of UPR proteins induced by Tm to protect against iron-induced toxicity within oligodendrocyte lineage cells. However, further work is required to elucidate or reasonably speculate on such a mechanism.

### REFERENCES


# AUTHOR CONTRIBUTIONS

SH carried out the PCR. SH and JM completed the tissue staining. JM and UF conceived and oversaw research described. SH, JM, and UF wrote the manuscript.

# FUNDING

SH was supported by a College of Science Fellowship from The National University of Ireland, Galway. SH and JM were supported by the Foundation Office of the National University of Ireland, Galway.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2018.00969/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 Healy, McMahon and FitzGerald. 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 Involvement of Iron in Traumatic Brain Injury and Neurodegenerative Disease

#### Maria Daglas and Paul A. Adlard\*

The Florey Institute of Neuroscience and Mental Health, The University of Melbourne, Parkville, VIC, Australia

Traumatic brain injury (TBI) consists of acute and long-term pathophysiological sequelae that ultimately lead to cognitive and motor function deficits, with age being a critical risk factor for poorer prognosis. TBI has been recently linked to the development of neurodegenerative diseases later in life including Alzheimer's disease, Parkinson's disease, chronic traumatic encephalopathy, and multiple sclerosis. The accumulation of iron in the brain has been documented in a number of neurodegenerative diseases, and also in normal aging, and can contribute to neurotoxicity through a variety of mechanisms including the production of free radicals leading to oxidative stress, excitotoxicity and by promoting inflammatory reactions. A growing body of evidence similarly supports a deleterious role of iron in the pathogenesis of TBI. Iron deposition in the injured brain can occur via hemorrhage/microhemorrhages (heme-bound iron) or independently as labile iron (non-heme bound), which is considered to be more damaging to the brain. This review focusses on the role of iron in potentiating neurodegeneration in TBI, with insight into the intersection with neurodegenerative conditions. An important implication of this work is the potential for therapeutic approaches that target iron to attenuate the neuropathology/phenotype related to TBI and to also reduce the associated risk of developing neurodegenerative disease.

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Kimberly R. Byrnes, Uniformed Services University of the Health Sciences, United States Dong Sun, Virginia Commonwealth University, United States

#### \*Correspondence:

Paul A. Adlard paul.adlard@florey.edu.au; padlard@unimelb.edu.au

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 22 October 2018 Accepted: 07 December 2018 Published: 20 December 2018

#### Citation:

Daglas M and Adlard PA (2018) The Involvement of Iron in Traumatic Brain Injury and Neurodegenerative Disease. Front. Neurosci. 12:981. doi: 10.3389/fnins.2018.00981 Keywords: iron, metals, traumatic brain injury, neurodegeneration, oxidative stress, inflammation

# INTRODUCTION

Traumatic brain injury (TBI) is a leading cause of death and disability worldwide, particularly amongst young adults. Ten million individuals are affected by TBI annually, costing a staggering \$9–10 billion/year (Gardner et al., 2017). The aged population have a greater risk of sustaining a TBI, with frequent falls being the major cause of injury, and they also have worse outcomes post-injury compared to other age groups (Stocchetti et al., 2012). Aging is also accompanied by a number of co-morbidities which may contribute to poorer outcomes in these individuals following TBI (Stocchetti et al., 2012). Common secondary events that follow the primary impact are neuronal cell death, oxidative stress, brain oedema, blood-brain barrier (BBB) breakdown, and inflammation (Toklu and Tumer, 2015). TBI often results in debilitating long-term cognitive and motor impairments, and there are currently no approved treatments available for TBI patients (Bramlett and Dietrich, 2015). This highlights the need for therapeutic agents that can alleviate brain damage and the deficits caused by the primary injury and more specifically the reversible secondary pathologies that develop after TBI.

Iron homeostasis appears to be an important process in the pathobiology of TBI. Iron is essential for normal brain functioning where it acts as an essential cofactor for several enzymatic/cellular processes (Ke and Qian, 2007). However, impaired regulation of iron can result in the production of reactive oxygen species (ROS) and the consequent promotion of oxidative stress, which can wreak havoc on an already compromised brain in the context of TBI (Nunez et al., 2012). Interestingly, the accumulation of iron in various tissues and cells in the body and brain is an inevitable consequence of aging (Hagemeier et al., 2012; Del et al., 2015). A concomitant increase in the iron storage protein (i.e., ferritin), which can scavenge any excess iron and prevent undesired production of ROS, is also evident with aging (Andersen et al., 2014). However, failed or weakened antioxidant defenses and mitochondrial dysfunction that progresses with aging can disrupt the balance and allow for excessive iron to be released (Venkateshappa et al., 2012a,b; Andersen et al., 2014). This can cause pathological iron overload resulting in cellular damage that is considered to be a contributing factor in several degenerative diseases that are more prevalent with age, such as cancer, liver fibrosis, cardiovascular disease, diabetes (type II), and particularly neurological conditions such as Alzheimer's disease (AD) (Smith et al., 1997; Kalinowski and Richardson, 2005; Ward et al., 2014; Hare et al., 2016). Abnormal brain iron deposition has also been discovered in other neurodegenerative diseases, such as Parkinson's disease (PD) (Griffiths et al., 1999; Zhang et al., 2010; Barbosa et al., 2015), multiple sclerosis (MS) (Bergsland et al., 2017), amyotrophic lateral sclerosis (ALS) (Oshiro et al., 2011), Huntington's disease (Agrawal et al., 2018), and Friedreich's ataxia (Martelli and Puccio, 2014), and there is now increasing evidence of altered iron levels in TBI patients (Raz et al., 2011; Lu et al., 2015). This raises the interesting proposition of an intersection between aging, iron, TBI and neurodegenerative disease. Whilst the role of iron in TBI is not well known, the literature suggests that the levels of iron (and other metals such as zinc) are abnormally regulated following injury, and that the pharmacological targeting (e.g., using chelators or chaperones) of these metals may be beneficial in improving outcomes. Iron chelation therapy has been approved for decades for the treatment of iron overload conditions, and there is a recent heightened interest for their use in neurodegenerative diseases (Kalinowski and Richardson, 2005; Dusek et al., 2016). Here, we review the role of iron dyshomeostasis in TBI and gain a deeper understanding of its involvement in neurodegeneration as well as neuroinflammation (**Figure 1**). We further examine the potential benefit of utilizing iron chelation therapy with the hope of limiting iron-induced neurotoxicity in TBI.

#### IRON HOMEOSTASIS AND ITS ROLE IN NEUROLOGICAL FUNCTION

Iron is an important cofactor involved in a variety of bodily functions including cellular metabolism, energy production, cell growth and differentiation, and gene expression. It is primarily

recognized for its ability to bind oxygen, contained within the hemoglobin complex, to transport oxygen throughout the body (Pantopoulos et al., 2012). Iron is also, however, fundamental for neuronal development, synaptic plasticity, neurotransmitter processing, and myelination (Ke and Qian, 2007; Ward et al., 2014). In the brain, iron, as well as other biometals, are found in numerous cell types including neurons, astrocytes, microglia, and most abundantly in oligodendrocytes (Burdo et al., 1999; Angelova and Brown, 2015). Astrocytes that are an integral part of the BBB have the ability to take up circulating iron and redistribute it to other cells in the CNS (Dringen et al., 2007). Iron is crucial for the maintenance of myelin, and therefore oligodendrocytes require a constant supply of iron (Connor and Menzies, 1996). Iron is also essential for optimal mitochondrial function (Urrutia et al., 2014), which is critical for neurons and other brain residing cells that have significant energy and metabolic requirements.

Iron homeostasis is carefully regulated in mammals by a plethora of proteins and enzymes produced in the liver such as hepcidin and ceruloplasmin, and other iron transporters (divalent metal transporter 1: DMT1, ferroportin, and transferrin) and their receptors (Ke and Qian, 2007). The expression levels of these proteins/receptors are in turn regulated by iron responsive elements (IRE) and iron regulatory proteins (IRP) especially when iron levels are low (Torben et al., 2007). Iron generally exists in an insoluble ferric (Fe3+) or soluble ferrous (Fe2+) state and is an active participant in electron transfer/redox reactions, which can drive oxidative stressinduced cellular toxicity (Graf et al., 1984; Nunez et al., 2012). Iron is solely absorbed through the diet, usually in its ferric form, and needs to be reduced to Fe2<sup>+</sup> for its essential cellular uptake via transporter DMT1, with ferrous iron also released from cells via ferroportin (Gunshin et al., 1997; Donovan et al., 2000). In the circulation, iron is transported by binding to transferrin (ferric state; 2:1 ratio) whereas in tissues it is stored in a stable conformation by ferritin, a protein that consists of two subunits (i.e., ferritin heavy chain and light chain) (Ke and Qian, 2007; Nunez et al., 2012; Ward et al., 2014). The main regulator and inhibitor of iron release is hepcidin, which is secreted by the liver in response to stimulants (e.g., human hemochromatosis protein (HFE), inflammatory mediator IL-6, lipopolysaccharides (LPS) or high transferrin-Fe3<sup>+</sup> complexes) ultimately to reduce levels of iron in the body (Ganz, 2003; Ward et al., 2014).

## THE METABOLISM OF IRON IN THE BRAIN

The entry of iron into the brain parenchyma is restricted by the BBB due to its hydrophilic nature. Iron (ferric iron: Fe3+) can enter through endothelial cells that line the BBB typically via endocytosis of transferrin/transferrin receptors and to a lesser extent either independently (non-transferrin bound iron) or as a low molecular weight complex (Ke and Qian, 2007; Torben et al., 2007; Ward et al., 2014). Iron is released from transferrin due to the acidic pH environment of the endosome, is then reduced to ferrous iron and thus contributes to the labile ferrous iron pool (Ke and Qian, 2007). Ferroportin is the only known iron exporter for almost all cell types including neurons and astrocytes (Donovan et al., 2000). Ceruloplasmin, which is primarily a copper containing α2-globulin (binds six copper ions) and mainly present as a GPI-linked membrane protein in the brain, is also an important regulator of iron-mediated transport via binding ferroportin and driving the conversion of Fe2<sup>+</sup> to Fe3<sup>+</sup> by its ferroxidase activity (Attieh et al., 1999). Amyloid precursor protein (APP), a key protein in AD pathology, stabilizes ferroportin to facilitate iron export in neurons, and its activity has recently been found to be regulated by iron (Rogers et al., 2008; Wong et al., 2014). Ceruloplasmin and APP expression are both increased alongside iron accumulation in the brain of TBI patients and brain injured mice, and were found to elicit neuroprotective effects (Ayton et al., 2014). **Figure 2** summarizes the proteins/receptors involved in iron metabolism in the brain.

Since iron is a transition metal, Fe2<sup>+</sup> and Fe3<sup>+</sup> can react with hydrogen peroxide and oxygen, respectively, catalyzing hydroxyl and superoxide radicals and can also react with lipids leading to the formation of alkoxy and peroxy radicals (Gutteridge, 1982). This is known as the Haber-Weiss and Fenton reaction. Hence, iron can increase ROS by participating in redox reactions and in turn lead to cellular oxidative stress and eventually cell death (Ke and Qian, 2007). Sufficient iron storage in the brain either by ferritin, neuromelanin (mainly in substantia nigra neurons) or hemosiderin is essential to prevent oxidative damage (Angelova and Brown, 2015).

Deficiency of iron or excessive iron, or changes in ironregulating protein expression, can therefore be catastrophic to cells/tissues and are known to contribute to numerous disorders such as anemia, aceruloplasminemia, neuroferritinopathy, mitochondrial membrane protein-associated neurodegeneration (MPAN) and pantothenate kinase-associated neurodegeneration (PKAN) (Levi and Finazzi, 2014; Hogarth, 2015). Patients with inherited conditions that develop increased brain iron accumulation present with cognitive, motor and neuropsychiatric decline (Schröder et al., 2012; Hogarth, 2015). Excessive iron accumulation is harmful because it can promote the formation of free radicals resulting in oxidative stress, lipid peroxidation, protein aggregation and eventually cell/neuronal death (Nunez et al., 2012). On the other hand, iron deficiency can result in impaired myelin generation, neurotransmission and enzymatic processing. Iron deficiency anemia has been linked to the development of AD, and individuals with anemia are twice as likely to develop mild cognitive impairments over time (Faux et al., 2014). A recent study consisting of 939 patients with TBI found that anemia (i.e., presenting with low hemoglobin levels and low plasma iron), which affects approximately half of hospitalized TBI patients, is associated with poor outcomes following TBI (Litofsky et al., 2016). Taken together this shows that a delicate balance of iron metabolism needs to be achieved to protect the brain against its potentially harmful effects, in both healthy and diseased states.

#### IRON IN NEURODEGENERATIVE DISEASES

Iron dyshomeostasis appears to be a central factor in neurodegenerative conditions such as AD, PD, ALS, Huntington's disease, and Friedreich's ataxia (Oshiro et al., 2011; Agrawal et al., 2018). For decades, deposits of iron have been detected in brain lesions of patients with these neurodegenerative conditions (Hagemeier et al., 2012; Barbosa et al., 2015; Adlard and Bush, 2018). How iron accumulates in the diseased/injured brain is not known, however, heme-bound iron that is derived from the circulation and non-heme bound iron (i.e., all iron not bound to heme protein) are both present in affected tissue (Nisenbaum et al., 2014; Ward et al., 2014). It is likely that the deposition of iron into the brain parenchyma in these conditions is the result of either hemorrhage/microbleeds (i.e., heme-bound iron), damaged cells (i.e., neurons, oligodendrocytes and microglia),

FIGURE 2 | Iron metabolism in the healthy and traumatically injured brain. (A) Circulating iron (Fe3+) passes the blood-brain barrier (BBB) either bound to transferrin or independently, and is then taken up by endothelial cells expressing transferrin receptors via endocytosis. The acidic pH environment of the endosome detaches Fe3<sup>+</sup> from transferrin and Fe3<sup>+</sup> is then reduced to Fe2<sup>+</sup> by ferric reductase (Ward et al., 2014). In the brain, Fe2<sup>+</sup> is uptaken via DMT1 and distributed amongst astrocytes, neurons, oligodendrocytes, and microglia. Inside these cells, Fe2<sup>+</sup> is converted to Fe3<sup>+</sup> and stored by ferritin to prevent oxidative damage (Oshiro et al., 2011). Oligodendrocytes express Tim2 receptor that binds ferritin and iron can be imported through this mechanism, in addition to DMT1. Neurons and microglia can also uptake transferrin-bound iron via transferrin receptors (Ward et al., 2014). Ferrous iron is exported from cells via the iron exporter, ferroportin, which is present on all cell types. Fe2<sup>+</sup> is rapidly oxidized to Fe3<sup>+</sup> once outside the cell by ferroxidases such as hephaestin or ceruloplasmin (expressed on astrocytes mainly). Amyloid precursor protein (APP) is expressed on neurons, astrocytes and microglia, and has been found to facilitate iron export in neurons by stabilizing ferroportin (Wong et al., 2014). (B) Injury to the brain/neurovascular unit causes BBB damage, microglia activation, astrogliosis, and eventually damage to neurons and the myelin sheath surrounding axons. Dysregulation of iron metabolism in the brain following TBI can result in the accumulation of redox-active ferrous iron in various brain cells. This is possibly due to alterations in the expression/function of regulatory proteins such as ferroportin, ferritin and ceruloplasmin, which fail to export iron from cells and thereby increases the labile iron pool. Iron accumulation in microgia can cause increased pro-inflammatory cytokine production, and vice versa (Urrutia et al., 2014). Taken together, iron accumulation in the brain can promote oxidative stress that contributes to neurodegeneration.

and myelin, inflammatory processes, degradation of erythrocytes and heme proteins, and/or dysregulation of important ironrelated proteins (e.g., ceruloplasmin, ferroportin) (Hametner et al., 2013; Hare et al., 2013; Nisenbaum et al., 2014). Excessive iron that is localized in affected neurons and glia can be found in brain regions such as the substantia nigra (dopaminergic neurons) in PD (Barbosa et al., 2015), and can also accompany regions of amyloid plaques and neurofibrillary tangles in AD (Smith et al., 1997). Similarly, patients with MS present with increased iron deposition in deep gray matter regions correlating with poor integrity of the white matter tissue (Bergsland et al., 2017). Increased brain iron accumulation in these neurological conditions have correlated with poorer functional outcomes and disease severity in patients (Zhang et al., 2010; van Duijn et al., 2017).

Although it is not yet clear whether iron is a causative or a contributing factor, its presence drives redox reactions, inflammatory processes as well as mitochondrial dysfunction (Urrutia et al., 2014). Iron accumulation in dopaminergic neurons, which are typically affected in PD, can exacerbate α-synuclein pathology and induce autophagy dysfunction (Wan et al., 2017). Ferric iron (Fe3+) can cause the formation of neurofibrillary tangles by binding to aggregated hyperphosphorylated tau proteins and can also become reduced to its destructive redox-active form (free Fe2+) during the process (Yamamoto et al., 2002; Oshiro et al., 2011). Iron can promote

the toxicity of amyloid beta (Aß) by affecting aggregation, oligomerization, and amyloidosis of the peptide (Liu et al., 2011). It is also likely that the pathogenicity of Aß (e.g., in AD) is at least in part due to its interaction with iron and the potentiation of redox reactions (Rottkamp et al., 2001). In addition, expression of proteins that are involved in the regulation of iron are altered in these disease states; ceruloplasmin, ferroportin, and furin (regulates hepcidin) are decreased while DMT1 (iron importer) and transferrin are increased (Hare et al., 2013; Ward et al., 2014; Peters et al., 2015). The main iron storage protein in the brain, ferritin, is elevated in the cerebrospinal fluid (CSF) of patients with AD as well as areas containing amyloid plaques, and this protein also increases with age (Jellinger et al., 1990; Zecca et al., 2001; Ayton et al., 2018). These regulatory proteins in addition to iron are also associated with the production of Aß and hyperphosphorylated tau (Rogers et al., 2008). Together this reveals that the failure lies with various crucial proteins involved in the regulation of iron and is a possible explanation for the accumulation of iron in the brain.

Overall, increased iron accumulation in the brain, over and above the age-related changes, may be an indirect indicator of the progression and severity of AD and PD, and possibly other neurological conditions (van Duijn et al., 2017; Adlard and Bush, 2018). Elevated brain iron as measured by quantitative susceptibility mapping (via MRI) and increased CSF ferritin levels observed in patients with AD and mild cognitive impairment, were found to be predictive of cognitive decline over the long-term, and correlated with amyloid-β pathology (Ayton et al., 2017) and apolipoprotein E genotype (Ayton et al., 2015), respectively. Considering that brain iron content increases with age and that age is a major risk factor associated with neurodegenerative disease, the detection of iron using noninvasive imaging techniques may therefore prove to be a useful diagnostic tool as an early indicator of disease onset.

## INTERPLAY OF IRON, INFLAMMATION AND NEURODEGENERATION

The immune system similarly plays an important role in the pathogenesis of neurodegenerative diseases although its interaction with biometals is unappreciated. Immune cells are also equipped with transferrin and ferroportin receptors to enable the influx/efflux of iron to meet their cellular needs. In particular, phagocytic immune cells such as microglia (brain-specific macrophages) have an important role in the regulation and distribution of iron since they function to engulf and recycle damaged/dying cells and thereby release iron into the brain interstitial space (Moos, 2002; Andersen et al., 2014). Intracellular iron accumulation is associated with macrophage infiltration/polarization and increased TNF-α expression (Kroner et al., 2014). Indeed, inflammatory cytokines (mainly IL-6 and TNF-α) and mediators such as LPS induce the expression of iron transporter receptors (e.g., DMT-1) and promote an increase in the accumulation of iron in neurons and microglia (Urrutia et al., 2014).

Iron can accumulate in myeloid cells including microglia and monocytes, causing oxidative damage to the cell and promoting chronic neuroinflammation. This is a typical feature observed in several neurological conditions including MS and AD (Andersen et al., 2014; Stuber et al., 2016). In vitro studies reveal that iron can directly increase NFκB activity and stimulate pro-inflammatory cytokine production in microglia (Saleppico et al., 1996). Indeed, hemosiderin-laden macrophages were present in the frontal and temporal lobes of a patient with mild TBI (Bigler, 2004). Damaged oligodendrocytes and myelin in neuroinflammatory conditions including MS were determined as the main sources of extracellular iron in the brain parenchyma (Hametner et al., 2013).

This interplay of iron deposition with phagocytic cells (macrophages) can lead to neuronal damage not only due to oxidative stress and free radical formation but also through the promotion of pro-inflammatory mediators that further contributes to secondary damage in TBI. There is also the hypothesis that the inflammatory cells that migrate to the site of injury in TBI (or plaque regions in neurodegenerative diseases) are largely responsible for depositing iron (Sastry and Arendash, 1995; Andersen et al., 2014). Hence, an unrecognized link between iron accumulation and the involvement of the immune response in TBI warrants investigation considering both these processes are persistently active in the brain from weeks to months following injury. Whether iron dyshomeostasis in the brain is the process that promotes glial reactivity and drives neuroinflammatory responses, or vice versa, is unknown. Perhaps both these processes are simultaneously contributing to brain damage and consequential neurological impairment following TBI. Supporting this, a recent review proposed a positive feedback cycle in which mitochondrial dysfunction, inflammation, and iron accumulation are critical synergistic mechanisms that induce and enhance one another, and eventually become responsible for neuronal/cell death in the pathogenesis of neurodegenerative diseases (Nunez et al., 2012; Urrutia et al., 2014).

# THE ROLE OF IRON IN TBI

Iron deposits that accumulate in the brain can consist of both non-heme and heme iron sub-types. These features are more commonly observed in neurodegenerative conditions such as AD, although increasing evidence suggests that both these subtypes of iron are increased in the injured brain after TBI (Nisenbaum et al., 2014). Heme-bound iron is commonly found coinciding with intracranial hemorrhage, along with deposition of hemosiderin and ferritin due to the phagocytosis of erythrocytes by microglia/macrophages (Koeppen et al., 2008; Nisenbaum et al., 2014).

A feature seen in TBI as well as several neurodegenerative diseases is the formation of cerebral microhemorrhages. These have been observed at both the acute and chronic stages of TBI, in experimental models of single and repetitive TBI (Donovan et al., 2012; Glushakova et al., 2014) as well as in a subgroup of mild TBI patients (Park et al., 2009; Hasiloglu et al., 2011; Nisenbaum et al., 2014). The clinical relevance of cerebral microhemorrhages on TBI outcome is not yet clear (Scheid et al., 2006; Irimia et al., 2018; van der Horn et al., 2018) although

the presence of microhemorrhages following TBI is a suggested predictor of injury severity (Huang et al., 2015; Lawrence et al., 2017). Microbleeds have been negatively associated with cognitive outcomes in patients with mild cognitive impairment, stroke and MS (Werring et al., 2004; Zivadinov et al., 2016; Li X. et al., 2017). Microhemorrhages are interestingly associated with areas of white matter damage, BBB breakdown, demyelination, and inflammation in experimental TBI (Glushakova et al., 2014). Microbleeds, which can be detected by Prussian blue iron staining, reveal an abnormal accumulation of iron and ferritin/hemosiderin that have been found to be toxic to brainspecific cells (neurons, astrocytes, and microglia) and endothelial cells (Glushakova et al., 2014). More importantly, non-heme bound iron and free iron are also elevated in the brain following TBI and are considered more problematic (Nisenbaum et al., 2014), as shown in **Figure 2**.

The origin of brain iron deposits independent of heme-bound iron in TBI is not yet known although several hypotheses exist. Dysregulation of ceruloplasmin (e.g., decreased levels) can lead to an increase in free copper ions and a concomitant decrease in ferroxidase activity, thereby reducing clearance of iron from the CNS (i.e., increasing iron deposition). Ceruloplasmin deficient mice subjected to TBI showed elevated brain iron levels and exacerbated infarct volume (Ayton et al., 2014). TBI (and specifically intracerebral hemorrhage: ICH) often accompanies hemorrhage, increased vascularisation and BBB breakdown, which likely results in heme-bound iron deposition in the brain initially. In this instance, the enzyme heme-oxygenase 1 (HO-1) is involved in the degradation of heme-bound iron into free iron or non-heme bound iron (Beschorner et al., 2000). HO-1 is also suggested to be involved in induction of antioxidants bilirubin and biliverdin, and can therefore protect against oxidative stress. Indeed, HO-1 expression is heightened following TBI in both experimental models (Fukuda et al., 1995; Okubo et al., 2013) and in humans over a long-term period (Beschorner et al., 2000). HO-1 is likely protective at the acute stage through its antioxidant properties and clearance of heme-bound iron but it also could be responsible for non-heme bound iron accumulation in the brain, thereby increasing labile iron content over time. Free iron released from the degradation of heme proteins is a main source of oxidative stress, particularly in the injured brain (Chang et al., 2005). Excessive quantities of labile iron (**Figure 2**) can be harmful by inducing excitotoxic-related neuronal death, impaired neurotransmission, and mitochondrial dysfunction (Nunez et al., 2012; Urrutia et al., 2014; Ward et al., 2014).

Literature characterizing the metal profile and spatiotemporal distribution of iron is relatively sparse in TBI of varying injury severities and repetitive injuries, as well as ICH and stroke, in both clinical (see **Table 1**) and animal studies (see **Table 2**). Importantly, the consequence of iron accumulation in the brain following TBI is currently unknown, although it is associated with the development of post-traumatic epilepsy (Willmore et al., 1978; Willmore and Ueda, 2009; Ueda et al., 2013). Brain-region specific accumulation of iron has been documented in patients with mild TBI, via magnetic resonance imaging (MRI) approaches such as Magnetic field correlation (MFC) and susceptibility weighted imaging (SWI) (Raz et al., 2011; Nisenbaum et al., 2014; Lu et al., 2015). Various regions of the brain are affected including the hippocampus, globus pallidus, thalamus, substantia nigra, and corpus callosum, as summarized in **Table 1**. Non-heme iron is more sensitively detected by MR imaging techniques although limitations do exist (e.g., tissue fluid content can affect iron quantitation) (Raz et al., 2011). However, it can be concluded that non-heme bound iron accumulates in deep gray matter areas of the brain and may contribute to pathological secondary injury (e.g., oxidative stress) following TBI since abnormal iron deposition correlated with poor cognitive outcome (Raz et al., 2011; Lu et al., 2015).

Recent studies (Portbury et al., 2016, 2017) thoroughly analyzed the time-dependent alteration in metals including iron, zinc, and copper, in the brain by laser ablationinductively coupled plasma-mass spectrometry (LA-ICPMS) in a mouse controlled cortical impact (CCI) model. Elevated iron concentrations were detected at the impact site as early as 3 days and were persistently elevated up to 28 days in both young (Portbury et al., 2016) and aged mice (Portbury et al., 2017) following injury. Supporting this, histological staining revealed intracellular iron accumulated in the cortex and hippocampal regions of the injured hemisphere (Portbury et al., 2016). Another study subjected aged mice (21–24 month old) to the CCI model and found increased iron deposition using T2-weighted MRI in the ipsilateral thalamus that accompanied astrogliosis and microgliosis at 1 and 2 months post-TBI (Onyszchuk et al., 2009). The authors suggest that free iron may be elevated as a result of degenerating thalamic neurons (Onyszchuk et al., 2009). Therefore, the continued presence of both inflammation and iron accumulation in the lesion likely contributes to ongoing pathology following TBI, as summarized in **Figures 1**, **2**.

#### IRON IN AGING, INJURY AND NEURODEGENERATION

It is well recognized that iron as well as iron-related proteins (e.g., ferritin and HO-1) progressively increase in the aging brain (Hirose et al., 2003; Stankiewicz et al., 2007). Microglia and astrocytes, but not oligodendrocytes, appear to accumulate iron more with age (Zecca et al., 2001; Angelova and Brown, 2015). Regions of the brain typically affected by age-related iron accumulation are the hippocampus, cortex, cerebellum, and most prominently the basal ganglia. Interestingly, these regions are commonly linked to neurodegenerative diseases (Angelova and Brown, 2015). Indeed, age remains one of the greatest risk factors associated with poor outcomes following TBI, and the development of neurodegenerative/neuroinflammatory conditions (Stocchetti et al., 2012; Ward et al., 2014). In aged individuals, iron accumulation in the brain could be caused by events that coincide with the aging process including increased permeability of the BBB, inflammation, dysregulation of iron-related proteins, mitochondrial dysfunction, and altered iron distribution within the CNS (Farrall and Wardlaw, 2009; Ward et al., 2014). Therefore, it is not clear whether agerelated iron accumulation causes, or is rather a consequence of, neurodegeneration that commonly occurs with aging. Importantly, iron deposition in the aged brain has been linked


TABLE 1 | Brain iron accumulation in brain injury; evidence from clinical studies.

CTE, chronic traumatic encephalopathy; IHC, immunohistochemistry; ICH, intracerebral hemorrhage; MRI, magnetic resonance imaging; MFC, magnetic field correlation; MMSE, Mini-Mental State Examination; N/A, not available; PCR, polymerase chain reaction; SWI, susceptibility weighted imaging.

to cognitive and motor deficits in the elderly (Pujol et al., 1992), and motor dysfunction in rhesus monkeys (Cass et al., 2007). Since aged individuals generally exhibit co-morbidities and present with systemic inflammation and altered neurovascular unit integrity among other age-related pathological processes (Stocchetti et al., 2012), this renders the brain vulnerable to ironmediated oxidative damage (Stankiewicz et al., 2007). This can be particularly problematic for an already compromised brain following TBI. The extent of iron deposition between adults and elderly TBI patients has not yet been directly examined clinically. However, a recent study showed that aged (24 month old) mice have higher levels of iron (as well as other metals) in both the injured and contralateral (uninjured) hemisphere following TBI compared to young (3 month old) mice (Portbury et al., 2017). This heightened increase in brain iron levels can potentiate oxidative stress and neurotoxicity (as extensively covered in this review), and therefore is a possible explanation for the cognitive decline and overall worse outcomes observed in the aging TBI population. Further research is needed to better understand the role of iron in aging and its link to TBI and neurodegenerative disease.

## THERAPEUTIC RELEVANCE AND FUTURE DIRECTIONS

Iron-mediated brain damage occurs in patients with TBI over a chronic time-period, and it is speculated that increasing iron deposition (with age) can predispose these individuals to the development of neurodegenerative diseases later in life or to post-traumatic epilepsy (see **Figure 1**). Hence, a targeted therapeutic approach to regulate iron levels in TBI patients will be an important advancement in limiting progressive neurodegeneration and neurological deficits that accompanies

TABLE 2 | Brain iron accumulation in brain injury; evidence from experimental studies.


APP, amyloid precursor protein; HO-1, heme oxygenase-1; LA-ICPMS, laser ablation-inductively coupled plasma-mass spectrometry; MRI, magnetic resonance imaging; MCAO, middle cerebral artery occlusion.

TBI of varying severities, and thereby reduce the risk of neurological disorders. Iron chelation therapies work by binding free iron and removing excess iron from the blood. Iron chelation not only works by scavenging labile free iron but some can also act as an antioxidant that neutralizes the radical reaction by donating hydrogen (e.g., deferiprone, HBED iron chelator) (Khalaf et al., 2018). They can also bind/remove other metals yet have a higher affinity for iron, and the absorption and tissue penetration properties of the different iron chelators can vary (e.g., desferrioxamine is poorly absorbed in the gut while deferiprone and deferasirox are orally active) thereby different routes of administration are needed (Kalinowski and Richardson, 2005; Dusek et al., 2016). Iron chelation therapy has only been investigated in a few pilot clinical trials of common neurological conditions including AD (Crapper McLachlan et al., 1991), PD (Devos et al., 2014; Martin-Bastida et al., 2017), and MS (Lynch et al., 1996). As yet, no compelling evidence has come to light with the use of iron chelators in these conditions, although they have proved relatively effective in attenuating disease progression in PD. The brain permeable iron chelator, deferiprone, decreased iron deposits in the substantia nigra, and slowed disease-severity progression in a small-scale study involving PD patients (n = 40) (Devos et al., 2014). However, another pilot clinical trial assessing deferiprone in PD patients only found a trending improvement in motor function over 6 months (Martin-Bastida et al., 2017); hence more extensive clinical and experimental studies are warranted. Baicalin and deferoxamine, which are iron chelating drugs that can (baicalin) or cannot (deferoxamine) pass through


the BBB, were assessed in a rat model of PD. Both these treatments inhibited iron deposition in the substantia nigra as well as the striatum, dentate gyrus, dentate-interpositus, and cerebellum, protected dopaminergic neurons from death (Xiong et al., 2012).

This strategy of chelating iron has yet to be trialed in patients with TBI, or chronic traumatic encephalopathy (CTE) patients with repetitive head injury, nor has it been greatly explored in animal models of TBI. However, it remains a potential treatment option based on the few promising experimental TBI studies that have assessed the effect of iron chelators (i.e., deferoxamine or dextran-coupled deferoxamine) in experimental models that will be described hereafter.

#### IRON CHELATION IN TBI

Only a handful of studies to date have investigated the therapeutic potential of chelating iron in experimental TBI models, mainly via the ferric iron chelator deferoxamine, as summarized in **Table 3**. However, deferoxamine has several limitations including its inability to cross the BBB, rapid elimination rate and it can theoretically generate superoxide radicals through binding Fe3<sup>+</sup> leading to Fe2<sup>+</sup> oxidation (Klebanoff et al., 1989; Hall et al., 2010).

In a recent study, the brain permeable iron chelator HBED was assessed in its effectiveness to attenuate iron-mediated brain damage in a mouse model of TBI (i.e., CCI) (Khalaf et al., 2018). They found that HBED improved locomotor function along with a reduction in lesion volumes and hippocampal swelling. Interestingly, HBED also limited microgliosis and the expression of immune markers in the corpus callosum (Khalaf et al., 2018). This pointed to a potential mechanism through which the drug mediates its protective effects. However, a limitation of this study is that HBED treatment did not significantly reduce iron in the brain as measured by Perls' staining (Khalaf et al., 2018). On the other hand, deferoxamine treatment reduced the levels of iron and iron-storage proteins (ferritin and transferrin) in the brain, which were elevated at 28 days post-TBI in rats. This was accompanied by a reduction in brain atrophy in the injured rats treated with deferoxamine as well as an improvement in spatial memory and learning, at 28 days after TBI (Zhang et al., 2013). A similar protective effect with deferoxamine treatment was also observed in earlier studies at an acute period following TBI (Panter et al., 1992; Long et al., 1996).

Several key limitations of these studies exist, including the short period of treatment, limited outcome measures and only one recent study assessed brain pathology at one time-point. Hence, there is a great need for long-term sequential analysis for the use of iron chelators (specifically those that are brainpermeable) in experimental TBI models, in which the influence of iron chelation on functional outcomes, pathology (e.g., oxidative stress, cell death, BBB breakdown), other biochemical alterations and also inflammation are essential before moving into clinical trials.

However, iron chelation therapy has been more thoroughly explored as a strategy for combating brain iron overload in ICH and to some degree in stroke (See **Table 2**), although these are still in the preclinical phase with only a few currently active clinical trials (Yeatts et al., 2013; Yu et al., 2015; Zeng et al., 2018). In particular, a small-scale clinical trial using deferoxamine in ICH patients (n = 21) showed improvement in oedema volume and hematoma absorption over a period of 30 days (Yu et al., 2015). Deferoxamine reduces odema, cell death, hippocampal degeneration and also inflammation following experimental ICH (Garton et al., 2016). Neuronal death and behavioral deficits were attenuated by deferoxamine treatment in aged rats with induced ICH (Hatakeyama et al., 2013). Deferiprone treatment

has shown promise in reversing iron toxicity in the brain and improving outcome. Treatment was remarkably effective when deferiprone was combined with the antioxidant n-acetyl cysteine, in rats fed with a high iron-supplemented diet (Sripetchwandee et al., 2014, 2016). This combined treatment had a greater efficacy in improving learning and memory, long-term potentiation, mitochondrial function, BBB integrity and cell death, compared to individual treatment in iron-overloaded rats (Sripetchwandee et al., 2014, 2016). In addition, brain-related pathology including Aβ deposition, tau-hyperphosphorylation and dendritic spine loss were all significantly reduced to a greater extent with combined treatment (Sripetchwandee et al., 2016).

Most recently, the use of a ferroptosis (a form of irondependent cell death) inhibitor was shown to reduce iron deposition and result in a host of downstream benefits, including decreased neuronal degeneration and improved behavioral outcomes (motor and cognitive function) (Xie et al., 2018). Thus, ferroptosis may represent a new pathway of relevance for TBI, and such studies are currently underway in our lab.

#### CONCLUSION

Brain iron accumulation is a common feature of most neurodegenerative/neurological conditions including TBI. It is increasingly becoming clear that iron plays a detrimental role in the pathogenesis of various neurodegenerative diseases, however, its relationship to TBI pathobiology still requires consolidation. Although iron chelation with deferoxamine has shown limited potential in experimental TBI models, using a chelator that can penetrate the BBB (e.g., deferiprone, HBED, VK-28), with a known ability to lessen iron-induced neurotoxicity in ICH and neurodegenerative diseases, could prove to be superior than previously tested iron chelators in improving outcome after TBI. Whether a specified time-window for treatment is required (e.g., immediate vs delayed treatment) also needs to

#### REFERENCES


be addressed (as does the notion that a combination of iron and zinc-targeted treatments may have a synergistic benefit in TBI). Targeting both iron accumulation and subsequent oxidative damage and/or inflammation by proposed combined therapy (e.g., other antioxidants or anti-inflammatory agents) has not yet been assessed preclinically in TBI (Hall et al., 2010). This approach has the potential of limiting acute secondary injury and protecting against neurological deterioration, as well as attenuating long-term neurodegeneration thereby reducing the risk of neurodegenerative disease development. Lastly, there is a great need for experimental studies to focus on the long-term consequences of iron-mediated damage considering iron deposition progresses over time; as seen with age, neurodegenerative diseases as well as TBI.

#### AUTHOR CONTRIBUTIONS

MD and PA conceived and constructed the manuscript.

#### FUNDING

PA was supported by funding from the National Health and Medical Research Council of Australia, the Australian Research Council and the Department of Defense (United States). The funders played no role in the content of this review or in the decision to publish.

#### ACKNOWLEDGMENTS

The Florey Institute of Neuroscience and Mental Health acknowledge the strong support from the Victorian Government and in particular the funding from the Operational Infrastructure Support Grant.





T2<sup>∗</sup> -weighted gradient-echo MRI. Brain 127, 2265–2275. doi: 10.1093/brain/ awh253


**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 Daglas and Adlard. 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.

# Ceruloplasmin Plays a Neuroprotective Role in Cerebral Ischemia

#### Fari Ryan, Juan G. Zarruk, Lena Lößlein and Samuel David\*

Centre for Research in Neuroscience, The Research Institute of the McGill University Health Centre, Montreal, QC, Canada

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Yongting Wang, Shanghai Jiao Tong University, China Liliana Quintanar, Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Naciona (CINVESTAV), Mexico Zahoor A. Shah, The University of Toledo, United States

> \*Correspondence: Samuel David sam.david@mcgill.ca

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 11 October 2018 Accepted: 10 December 2018 Published: 08 January 2019

#### Citation:

Ryan F, Zarruk JG, Lößlein L and David S (2019) Ceruloplasmin Plays a Neuroprotective Role in Cerebral Ischemia. Front. Neurosci. 12:988. doi: 10.3389/fnins.2018.00988 Ceruloplasmin (Cp) is a ferroxidase that also plays a role in iron efflux from cells. It can thus help to regulate cellular iron homeostasis. In the CNS, Cp is expressed as a membrane-anchored form by astrocytes. Here, we assessed the role of Cp in permanent middle cerebral artery occlusion (pMCAO) comparing wildtype and Cp null mice. Our studies show that the lesion size is larger and functional recovery impaired in Cp null mice compared to wildtype mice. Expression of Cp increased ninefold at 72 h after pMCAO and remained elevated about twofold at day 14. We also assessed changes in mRNA and protein expression of molecules involved in iron homeostasis. As expected there was a reduction in ferroportin in Cp null mice at 72 h. There was also a remarkable increase in DMT1 protein in both genotypes at 72 h, being much higher in wildtype mice (19.5-fold), that then remained elevated about twofold at 14 days. No difference was seen in transferrin receptor 1 (TfR1) expression, except a small reduction in wildtype mice at 72 h, suggesting that the increase in DMT1 may underlie iron uptake independent of TfR1-endosomal uptake. There was also an increase of ferritin light chain in both genotypes. Iron histochemistry showed increased iron accumulation after pMCAO, initially along the lesion border and later throughout the lesion. Immunofluorescence labeling for ferritin (a surrogate marker for iron) and GFAP or CD11b showed increased ferritin in GFAP+ astrocytes along the lesion border in Cp null mice, while CD11b+ macrophages expressed ferritin equally in both genotypes. Increased lipid peroxidation assessed by 4HNE staining was increased threefold in Cp null mice at 72 h after pMCAO; and 3-nitrotyrosine labeling showed a similar trend. Three key pro-inflammatory cytokines (IL-1β, TNFα, and IL-6) were markedly increased at 24 h after pMCAO equally in both genotypes, and remained elevated at lower levels later, indicating that the lack of Cp does not alter key inflammatory cytokine expression after pMCAO. These data indicate that Cp expression is rapidly upregulated after pMCAO, and loss of Cp results in dysregulation of iron homeostasis, increased oxidative damage, greater lesion size and impaired recovery of function.

Keywords: stroke, ferroxidase, iron, oxidative damage, ferritin

# INTRODUCTION

fnins-12-00988 December 26, 2018 Time: 18:2 # 2

Dysregulation of cellular iron homeostasis is considered to contribute to neural damage after ischemic stroke (Millerot-Serrurot et al., 2008; Texel et al., 2011; Garcia-Yebenes et al., 2012). Iron plays an important role in many physiological processes by catalyzing a variety of enzymatic reactions (Rouault and Tong, 2005). The reason for these various functions is the redox active nature of iron, which can transition between the ferrous (Fe2+) and ferric (Fe3+) states. Divalent ferrous iron (Fe2+) can react with hydrogen peroxide in the Fenton reaction to catalyze the formation of free radicals, which cause oxidative damage (Crielaard et al., 2017). Therefore, the levels of cellular iron needs to be tightly regulated via a complex system of interactions involving several cell surface and intracellular proteins (Hentze et al., 2004; Crielaard et al., 2017).

Ceruloplasmin (Cp) is a ferroxidase that safely converts ferrous (Fe2+) iron to its ferric (Fe3+) form (Williams et al., 1974; Calabrese et al., 1989). The soluble form of Cp is expressed by the liver and found in serum (Gitlin, 1988). We identified a glycosylphosphatidylinositol-anchored form of Cp (GPI-Cp) in the central nervous system (CNS) that is expressed by astrocytes (Patel and David, 1997; Patel et al., 2000; Jeong and David, 2003). We have also reported that GPI-Cp is expressed abundantly by meningeal cells that surround the brain and spinal cord (Mittal et al., 2003). The choroid plexus located in the ventricular spaces in the CNS, and which produces cerebrospinal fluid, expresses the soluble form of Cp (Patel and David, 1997). We provided early evidence that GPI-Cp partners with the iron efflux transporter ferroportin to regulate iron efflux from astrocytes (Jeong and David, 2003). We and others have reported that Cp gene knockout mice show disruption of iron homeostasis in the CNS and other tissues such as the liver (Harris et al., 1999; Patel et al., 2002) and lead to iron accumulation in the CNS with age (>18 months of age), resulting in free-radical mediated neuronal loss (Patel et al., 2002; Jeong and David, 2006). Other work also indicates that the absence of GPI-Cp results in the internalization and degradation of the iron exporter, ferroportin (Fpn), from the cell surface, which results in iron accumulation (De Domenico et al., 2007). Iron accumulation in the CNS and other organs, which increases with age is a characteristic feature of humans with aceruloplasminemia, which results from mutations in the Cp gene (Harris et al., 1995; Morita et al., 1995; Yoshida et al., 1995; Xu et al., 2004). We have also reported a slow accumulation of iron in the brain in Cp null mice with age (>18 months of age) (Patel et al., 2002; Jeong and David, 2006).

Mechanisms that lead to iron accumulation in cells in the CNS in different neurological conditions are not fully understood but can give rise to oxidative damage and neurodegeneration. In CNS injury and disease, increased iron loading of cells occurs either when dying cells release their intracellular iron content into the extracellular space (as in cell necrosis or physical damage) and/or are taken up by microglia and macrophages (phagocytosis) as occurs in ischemic stroke. Increased iron load in the CNS also occurs acutely after hemorrhage and uptake of red blood cells by phagocytes, as occurs in hemorrhagic stroke. In addition, dysregulation of molecules involved in maintaining cellular iron homeostasis can also result in iron accumulation, i.e., increased iron uptake or reduce iron efflux. The extent of iron-mediated free-radical damage will depend on how well excess iron is taken up and safely sequestered in ferritin or exported/effluxed out of the cells (Finazzi and Arosio, 2014). Moreover, in CNS disorders such as multiple sclerosis (MS), iron accumulation is seen in microglia that eventually appear to become dystrophic and undergo cell death (Hametner et al., 2013), which could lead to further cycles of iron uptake and cell death. We have reported recently, that iron accumulation in macrophages in a mouse model of MS [experimental autoimmune encephalomyelitis (EAE)] may be due to hepcidin-mediated internalization of Fpn (Zarruk et al., 2015).

In the present study, we examined the role of Cp in ischemic stroke using wildtype and Cp null mice. We examined the role of Cp on lesion size and functional recovery, as well as changes in the expression of various molecules involved in iron homeostasis in the ischemic brain. We also studied the extent of iron deposition, free radical damage and changes in the expression of inflammatory cytokines. This work suggests that Cp plays a significant role in functional recovery and tissue neuroprotection after permanent cerebral ischemia.

## MATERIALS AND METHODS

#### Induction of Permanent Middle Cerebral Artery Occlusion (pMCAO)

Male Cp−/− and Cp+/+ mice on a C57BL/6 background, 8– 12 weeks of age, were used. Ceruloplasmin knockout mice were generated in our lab several years ago as previously described (Patel et al., 2002). All surgical procedures were approved by the Animal Care Committee of the Research Institute of the McGill University Health Centre and followed the guidelines of the Canadian Council on Animal Care, and the ARRIVE guidelines for reporting animal research (Kilkenny et al., 2010). The pMCAO surgery was done as described previously (Zarruk et al., 2012) and is a variant of an earlier model (Chen et al., 1986). Briefly, animals were deeply anesthetized with isoflurane in O<sup>2</sup> (0.5–1 L/min; inhalation) during the entire procedure. During surgery, body temperature was maintained at 37.0 ± 0.5◦C using a homeothermic system with a rectal probe (Harvard apparatus). A small craniotomy was made over the region of the trunk of the left middle cerebral artery (MCA) above the rhinal fissure. The pMCAO was done by ligating the trunk of the MCA just before its bifurcation between the frontal and parietal branches using a 9-0 suture. Complete interruption of blood flow was confirmed under visual inspection under the operating microscope. In addition, the left common carotid artery was then occluded. After surgery, animals were returned to their cages with free access to water and food. Spontaneous mortality was not observed after the surgical procedure.

#### Infarct Volume

Nissl stained tissue sections (30 µm) spanning the entire lesion were assessed to estimate the infarct volume using the ImageJ software. To calculate the infarct volume, two measurements



were taken: the whole ipsilateral hemisphere and the ipsilateral hemisphere excluding the lesion. The 2 values were used to calculate the % of infarcted hemisphere. This is done to accurately estimate the lesion in each animal and to exclude any variations in brain size or distortions due to histological preparation. These values were used for statistical analysis.

#### Functional Assay

The "adhesive-tape removal" assay was used to assess functional recovery. This test assesses the integrity of motor and sensory function. It was originally described for rats and then adapted to mice (Bouet et al., 2009; Zarruk et al., 2011). In brief, a small piece of adhesive tape is placed over the hairless parts of the forepaw (pads, thenar, and hypothenar). This is repeated on both sides. Habituation and training was done for 5 days before surgery, each time alternating the order in which the tape was placed (left or right paw). The test was done on 1, 2, 7, and 14-days post-surgery (n = 13–15 mice per group). Three trials were done per day per mouse and the mean of each day was calculated. On the evaluation day, after the adhesive tape was placed, the mice were placed back into the evaluation box/cage, and the time taken to remove the tape is measured. Sham surgery animals were used as controls. The asymmetry between the right and left sides was calculated by dividing the time taken to remove the tape on the affected side (right) versus the non-affected side (left). The higher the number meant greater disability which is explained because ischemic lesions will affect the sensorimotor abilities of the mice on the contralateral side, impairing their ability to feel and/or remove the tape.

# Collection of Tissue for mRNA and Western Blotting

Under deep anesthesia the mice were perfused via the heart with phosphate buffered saline and the brain rapidly exposed and removed; and the cortex dissected out and rapidly frozen in liquid nitrogen. The most rostral and caudal parts of the brain (1 mm thick) were excluded to avoid regions that lack the infarct. The ipsi and contralateral sides of the cerebral hemisphere were taken for analysis. Samples were taken for mRNA and protein at different time intervals as indicated in the results section.

#### Quantitative Real Time Polymerase Chain Reaction (qPCR)

Total RNA was extracted using RNeasy Mini Kit (Qiagen, Mississauga, ON, Canada) following manufacturer's instructions. cDNA was reverse transcribed with the Qiagen Quantinova Kit (Cat. No. 205411) and amplified using an ABI OneStep cycler (Applied Biosystems) using specific primer pairs as indicated in **Table 1**. GAPDH was used as an internal control gene. The results were quantified using the 11CT method following standardization relative to GAPDH (Livak and Schmittgen, 2001).

#### Western Blotting

Brain tissue was obtained after intracardiac perfusion with PBS as described above. Total protein was extracted with 1% Nonidet P-40 (Sigma), 1% sodium deoxycholate (BDH Chemicals), 2% SDS, 0.15 M sodium phosphate (pH 7.2), and 2 mM EDTA, containing a mixture of protease inhibitors (Roche Diagnostics) as described previously (Jeong and David, 2003). Protein concentrations were estimated using the Bio-Rad DC protein assay following manufacturer's instructions (Cat. No 500- 0121). Then 25 µg of proteins per sample were separated by sodium dodecyl sulfate polyacrylamide electrophoresis (SDS– PAGE) gels 4–12% (Novex, life Technologies), transferred to polyvinylidene fluoride membrane (PVDF) (Millipore, Billerica, MA). Membranes were blocked in 5% milk in 0.05% PBS Tween-20 and incubated overnight at 4◦C with the following primary antibodies: rabbit anti-ferritin (1:500; Sigma-Aldrich Inc.; F5012), rabbit anti-DMT1 (1:1000; Alpha Diagnostics), mouse anti-TfR1 (1:1000; Invitrogen; 13–6800), rabbit anti-Cp (1:1000 DakoCytomation; Q 0121), or rabbit anti-Fpn (1:300 Alpha Diagnostic; MTP11-A). Blots were washed and incubated with horseradish peroxidase-conjugated IgG (1:10,000 –100,000; Jackson ImmunoResearch, West Grove, PA, United States) and visualized with enhanced chemiluminescence (PerkinElmer). Equal loading of proteins was estimated by re-probing the membranes with rabbit anti-β-actin antibodies (1:1000; Sigma-Aldrich). The blots were quantified using ImageJ software (version 1.48r NIH-USA).

## Immunofluorescence and Confocal Microscopy

Mice were perfused with phosphate buffer saline followed by 4% paraformaldehyde in 0.1 M phosphate buffer. After post-fixation in the same fixative for 24 h, the tissue was cryoprotected in

30% sucrose and 14 µm thick coronal sections of the brain cut with a cryostat and used for the immunofluorescence labeling. Tissue sections were incubated with 0.3% Triton X-100 (Sigma-Aldrich), 5% normal goat serum (Jackson ImmunoResearch Inc.), 2% ovalbumin (Sigma-Aldrich) in phosphate buffer for 3 h room temperature (RT) to block non-specific binding of antibodies. Sections were then incubated overnight at 4◦C with primary antibodies against ferritin (1:200). These primary antibodies were combined with the following cell type specific antibodies: rat anti CD11b (for macrophages/microglia; 1:200; Serotec; MCA711), guinea pig anti-glial acidic protein (GFAP) (for astrocytes; 1:500; Synaptic Systems cat No. 173004). Sections were washed in 0.05% PBS-Tween-20 and incubated with appropriate fluorescentconjugated secondary antibodies, anti-rabbit Alexa Fluor 488, anti-rat Alexa Fluor 568 or anti-guinea pig Alexa Fluor 568 (all 1:500, Invitrogen). Tissue sections were viewed with a confocal laser scanning microscope (FluoView FV1000, Olympus) and micrographs taken with the FV10-ASW 3.0 software (Olympus). For comparing between groups, the same setting was applied in all images of each immunostaining. The quantification of the double labeling of GFAP and ferritin at 72 h after pMCAO was assessed using the Image J software to estimate Pearson's correlation coefficient as described previously (Dunn et al., 2011).

### Diaminobenzidine-Enhanced Turnbull's Blue Staining for Iron

Turnbull's blue iron histochemistry was done to detect ferric and ferrous iron as previously described (Meguro et al., 2007). Briefly, at different times after pMCAO mice were perfused with 0.1 M phosphate buffer followed by 4% paraformaldehyde. The brains were dissected out, and cryoprotected in 30% sucrose and 14 µm sections cut with a cryostat (Leica CM 3050 S). Sections were incubated in 10% ammonium sulfide for 90 min at RT, washed with distilled water (dH2O) and incubated for 30 min in a solution of 20% potassium ferricyanide (Sigma-Aldrich) and 1% hydrochloric acid. Sections were then washed in double distilled water and endogenous peroxidase blocked for 60 min at room temperature in 99 ml PBS + 0.01 M NaN3 (65 mg) + 1 ml 30% H2O2. After washes in 0.1 M phosphate buffer, iron staining in the tissue sections was visualized by incubation with 0.025% 3,3<sup>0</sup> diaminobenzidine with 0.005% H2O<sup>2</sup> in 0.1 M phosphate buffer (20 min at RT). The reaction was stopped with tap water and sections were counterstained with Mayer's hemalum solution (EMD Millipore, 1.09249.1000) and mounted with Entellan (Electron Microscopy Sciences). For quantification of iron labeled cell at 72 h post-pMCAO, the number of labeled cells were counted and estimated per µm<sup>2</sup> . Direct cell counts of iron labeled cells could not be done at 14 days because the cells were densely packed in the lesion. Therefore, for quantification of iron staining at 14 days, we used densitometry to estimate the integrated density as follows: microscope images were prepared and analyzed with Image J (version 1.48r NIH-USA). The background was first subtracted with a rolling ball of 200 pixels. Staining intensity was then quantified using the densitometry 1 channel plug-in with a tolerance of 100 and the mean of integrated density was calculated for each animal and each group and plotted on the y-axis.

# Immunohistochemistry for 4HNE and 3-NT

Brain cryostat sections (14 µm thick) from 4% paraformaldehyde perfused-fixed mice, 72 h post-pMCAO were used for DAB immunohistochemistry for 4HNE (4-Hydroxynonenal) and 3- NT (3-Nitrotyrosine) as described previously (Zarruk et al., 2015). In brief, endogenous peroxidase was blocked with 0.2% H2O<sup>2</sup> in PBS for 30 min at RT. Then non-specific binding was blocked for 3 h at RT in blocking solution (3% Triton, 5% normal goat serum, in PBS) with avidin D solution (Vector laboratories; SP2001) followed by overnight incubation with primary rabbit anti-4HNE antibody (1:300; Abcam; AB46545) or rabbit anti-3NT antibody (1:500; Invitrogen; A21285), with the biotin solution (Vector laboratories; SP2001). The Avidin/Biotin blocking kit (Vector) was used as an extra step to block nonspecific avidin/biotin binding. Sections were washed in 0.05% PBS-Tween and incubated for 1 h at RT with secondary goat antirabbit biotinylated (1:1000; Jackson ImmunoResearch), washed and incubated with avidin peroxidase 1 h (RT) (1:100; Sigma #A-2693151). Labeling was visualized with 3,3<sup>0</sup> diaminobenzidinetetrahydro chloride using the ImmPACT DAB peroxidase substrate kit according to manufacturer instructions (Vector; SK-4105) and counterstained with Mayer's hemalum. The number of cells per µm<sup>2</sup> was quantify in each group from 10x images taken with a Zeiss Axioskop II light microscope using Image J (version 1.48r NIH-USA).

#### Statistical Analyses

All types of analyses were carried out blind, so the evaluator was not aware of the groups. Data are shown as mean ± Standard Error of the Mean (SEM). Two-way ANOVA followed by Bonferroni's post hoc test was used to analyze the results from "adhesive tape removal" test at different time points. The colocalization of GFAP and ferritin was quantified using Pearson's colocalization coefficient (Dunn et al., 2011) with the ImageJ colocalization Threshold plug-in. A zero value is given for the pixels below the thresholds. That means that all the pixels which have intensities above the two thresholds have greater than zero correlation, and the pixels below the thresholds have none or negative correlated intensities. Therefore, the closest the value is to 1 the better the colocalization (1 = full colocalization, 0 = no colocalization). Mann–Whitney non-parametric test was used to compare two groups. A value of p ≤ 0.05 was considered significant.

# RESULTS

## Decreased Functional Recovery in Cp Null Mice After pMCAO

We first assessed functional recovery after pMCAO using the "adhesive-tape removal" test. In this test, a small piece of adhesive tape is stuck on the paw of each forelimb, and the time taken

by the mouse to remove the tape is estimated. This test showed that Cp KO mice took a longer time to remove the tape as compared to wildtype mice (**Figure 1A**). Interestingly, the Cp KO and wildtype mice showed similar levels of deficit at 1 and 2 days after pMCAO. However, at 7 and 14 days, the two groups diverged with the Cp null mice showing increasingly worse deficits, while the wildtype mice showed recovery to normal values at 14 days. The latter may be due to neural plasticity in wildtype mice, which appears to be impaired in Cp null mice, possibly due to free radical mediated secondary damage to neurons (e.g., dendritic spine formation) and glia (microglia and astrocytes). For example, pruning of synaptic arbors by microglia is required for establishment of proper function (Stephan et al., 2012). Astrocytes are also required for recycling of neurotransmitters and the proper functioning of synapses (Allen, 2014). We have reported previously that Cp null mice show changes in CNS glia with reduction of astrocytes with age (Jeong and David, 2006). Sham controls did not show any deficits.

#### Infarct Size Is Larger in Cp Null Mice

We next determined the infarct size using Nissl stained tissue sections obtained from mice 24 h and 14 days after pMCAO. The ImageJ software was used to measure the total infarct volume and presented as a percentage of the ipsilateral hemisphere (**Figure 1B**). This analysis showed a significant increase in the size of the lesion in Cp null mice 24 h after pMCAO (17.5 mm<sup>3</sup> ) as compared to wildtype mice (15.8 mm<sup>3</sup> ) (**Figures 1B,C**). At 14 days post-pMCAO there was a trend to an increase in the Cp null mice, but these differences were not statistically significant (data not shown). This may be because by 14 days there is marked shrinkage of the lesion leading eventually to a scar, making the differences less prominent.

# Changes in the Expression of Iron Homeostasis Proteins at the mRNA and Protein Levels

#### Ceruloplasmin

In wildtype mice, there is a modest but significant increase in Cp mRNA expression in the ipsilateral ischemic hemisphere as compared to the uninjured contralateral hemisphere at both 72 h and 14 days after pMCAO (**Figure 2A**). Interestingly, Western blot analysis showed an 8.8-fold increase in Cp expression at 72 h after pMCAO and remained significantly elevated at about twofold at day 14 (**Figures 2B,C**). These findings suggest upregulation of Cp expression at the protein level in ischemic stroke. This increase in Cp expression is in keeping with its protective

role as a ferroxidase to reduce the toxic effects of ferrous iron and to export iron out of cells, particularly astrocytes in the CNS, and thus prevent intracellular iron accumulation.

#### Ferroportin

Cp partners with ferroportin (Fpn), the ubiquitously expressed iron efflux transporter, to export iron out of cells. We therefore assessed the expression of Fpn at the mRNA and protein levels. At the mRNA level there was a modest increase in Fpn expression on the ipsilateral ischemic cortex in both genotypes at 72 h and 14 days (**Figure 2D**). However, at the protein level there was about a 50% reduction in Fpn expression at 72 h in the ipsilateral side in Cp null mice versus the contralateral uninjured cortex (**Figures 2E,F**), as well as a significant reduction in the ipsilateral side in Cp null mice compared to wildtype mice at 72 h (**Figures 2E,F**). Our previous studies have shown that GPI-Cp is required for the stability of Fpn on the cell surface (De Domenico et al., 2007). In the absence of GPI-Cp, Fpn gets internalized and eventually degraded. This may account for the reduction of Fpn protein acutely in Cp null mice at 72 h. In addition, the partial reduction in Fpn protein may be because Cp is only expressed by astrocytes but no other cell types in the CNS (Patel and David, 1997). No differences in Fpn protein were seen in wildtype mice at 72 h and between any of the groups at 14 days post-pMCAO despite increases in Fpn mRNA on the ipsilateral side. Why the increase in mRNA levels on the ipsilateral side are not reflected in increase in protein levels in the Western blots is not known at present but may involve regulated internalization and degradation of the protein.

#### Transferrin Receptor 1 (TfR1) and Divalent Metal Ion Transporter 1 (DMT1)

There was a small but significant reduction in TfR1 mRNA and protein in the ipsilateral cortex in wildtype mice at 14 days post-ischemia. No other differences were seen (**Figures 3A–C**).

No differences were observed in the mRNA level of DMT1 after pMCAO in either genotype (**Figure 3D**). In contrast, Western blot analysis of tissue taken at 72 h post-pMCAO showed a large almost 19.5-fold increase in DMT1 protein in the ipsilateral ischemic cortex in wildtype mice and a smaller (sixfold) increase in the ipsilateral cortex in Cp null mice (**Figures 3E,F**). A smaller (twofold) significant increase in DMT1 protein continued to be detected in the ipsilateral cortex of both genotypes at 14 days (**Figures 3E,F**). DMT1 is known to be involved in release of endosomal iron taken up via the Tf-TfR1 complex. DMTI is also expressed on the surface of astrocytes (Jeong and David, 2003) and so may be involved in transferrin-independent iron uptake into cells. Our data showing no increases in TfR1 expression would suggest that the increase in DMT1 might contribute to direct uptake of iron into astrocytes after cerebral ischemia. It is also important to note that DMT1 protein expression was 3 times

lower in Cp null mice compared to wildtype mice at 72 h (**Figure 3E**).

#### Ferritin

Ferritin is the iron storage protein in cells. Each molecule of ferritin consists of 24 heavy and light chain subunits that form a sphere in which iron is stored. Each molecule of ferritin can store up to 4,500 atoms of iron and so is a very efficient iron storage protein (Arosio and Levi, 2002, 2010). Excess iron is stored in ferritin. There is a significant increase at the mRNA level of ferritin light chain (the major iron storage form) in the ipsilateral hemispheres in both genotypes compared to the contralateral hemispheres. The mRNA increase at 72 h in wildtype mice is about threefold and in Cp null mice about fourfold (**Figure 3G**). In contrast, mRNA expression of ferritin heavy chain, which has ferroxidase activity, showed no difference after ischemia in either genotype, except for a small but significant increase in the ipsilateral side in Cp null mice at 72 h (data not shown). At the protein level in Western blots, the antibody does not distinguish between heavy and light chains, so the results are a summation of the levels of both chains. In the ipsilateral ischemic cortex there is a significant increase in ferritin protein at 72 h and 14 days in both genotypes (**Figures 3H,I**).

# Iron Histochemistry and Ferritin Immunofluorescence Labeling

To investigate further the changes in iron accumulation in ischemic stroke, we performed diaminobenzidine (DAB) enhanced Turnbull's blue iron histochemistry. These studies showed that iron labeled cells are seen in the lesion along the lesion border at 72 h after pMCAO in both wildtype and Cp knockout mice (**Figures 4A,B**). The cell counts of labeled cells did not differ between the two genotypes at 72 h (**Figure 4C**). However, at 14 days after pMCAO, there was a marked increase in iron staining throughout the lesion in both genotypes (**Figures 4D,E**). The labeled cells were densely packed in the lesion making it difficult to obtain cell counts. Densitometry analysis, however, showed a significant increase in Cp null mice compared to wildtype mice (**Figure 4F**).

Iron histochemistry is not compatible with immunostaining, therefore, to assess which cell types accumulate iron in the ischemic brain, we carried out double immunofluorescence

labeling for ferritin, a good surrogate marker for iron, and co-labeled the sections with antibodies to GFAP (astrocytes) (**Figure 5**) and CD11b (macrophages/microglia) (**Figure 6**) at 72 h after pMCAO. The double labeling showed that in the ipsilateral ischemic cortex of Cp null mice, ferritin expression is more intense (**Figures 5E,K**) and co-localizes to significantly more GFAP-positive astrocytes as compared to wildtype mice (**Figures 5F,L,M**). Colocalization of ferritin and GFAP was quantified using Pearson's correlation coefficient, as the two labels are in the cytoplasmic compartment (**Figure 5M**). These ferritin+ astrocytes are located in the intact CNS tissue along the border region of the infarct. Co-localization of ferritin in CD11b+ macrophages/microglia is seen in the ischemic lesioned tissue along the border region of the infarct (**Figures 6A–F**). The number of ferritin+/CD11b+ macrophages/microglia at 72 h post-pMCAO was similar in Cp null and wildtype mice (**Figure 6G**).

These results suggest that at 72 h post-pMCAO, the iron labeled cells seen by the Turnbull's stain along the lesion border are likely to include both astrocytes and macrophage/microglia.

#### Changes in Markers of Oxidative Stress

Two indicators of lipid and protein oxidative damage were assessed. Lipid peroxidation in tissue sections was evaluated by immunohistochemical detection of changes in 4-hydroxynonenal (4HNE). Protein oxidation was assessed by immunohistochemical labeling for 3-nitrotyrosine (3-NT) a product of tyrosine nitration that is mediated by peroxynitrite. Immunohistochemical staining for 4HNE and 3-NT was done on tissue sections of the brain taken at 72 h after pMCAO. The expression of 4HNE was visualized with the DAB technique and appeared as dark brown staining of cells along the border of the infarct region (**Figures 7A–D**). Adjacent intact regions of the cortex did not show such labeling. The number of 4HNE+ cells in the ipsilateral cortex was significantly higher in Cp null mice as compared to wildtype mice (**Figure 7E**).

Immunohistochemistry for 3-nitrotyrosine occurred in the infarct area and there appeared to be a trend to an increase in CpKO mice compared to wildtype mice but this did not reach statistical significance (**Figure 8**).

# Expression of Inflammatory Cytokines

It is known that the iron status of cells can affect their inflammatory profile (Kroner et al., 2014; Rathore et al., 2012). We therefore assessed whether changes in iron homeostasis after cerebral ischemia also influenced the inflammatory response in Cp null mice. We assessed the expression of four key cytokines that include IL-1β, TNFα, IL-6, and IL-10. Changes in these cytokines was assessed at 24, 72 h, and 14 days after pMCAO. There were marked increases in the mRNA expression of the three proinflammatory cytokines (IL-1β, TNFα, and IL-6) at 24 h, which remained elevated at lower levels at 72 h after pMCAO

(**Figure 9**). No differences were noted between genotypes. The highest increases were detected at 24 h when IL-1β increased by 15-fold, TNFα by 10-fold and IL-6 by about 55-fold. These values dropped markedly by 72 h but were still significantly higher in the lesioned side versus the unlesioned side. By 14 days the values had almost returned to unlesioned control levels (**Figure 9**). IL-10 levels were not significantly changed but there appeared to be a trend to an increase at 72 h and 14 days. The important point to

note here is that the absence of Cp does not alter the expression of key inflammatory cytokines after pMCAO as compared to wildtype mice.

# DISCUSSION

We provide evidence that the lack of Cp leads to loss of functional recovery and greater lesion size after pMCAO in mice. This was accompanied by alterations in expression of molecules involved in regulating iron homeostasis in the CNS, including reduction in expression of iron efflux transporter (ferroportin), increase in iron uptake transporter (DMT1) and increased iron accumulation (ferritin). Iron histochemistry showed deposition of iron in cells along the border of the lesion at early timepoints after pMCAO and throughout the infarct area by 14 days. Double immunofluorescence labeling for ferritin (a surrogate marker for iron) and cell type specific markers revealed that

macrophage/microglia and astrocytes along the lesion border accumulate iron. The labeling in astrocytes being greater in Cp null mice. These changes were also accompanied by increased lipid peroxidation (4HNE staining) and greater nitration of tyrosine residues in proteins as evidenced by increased 3 nitrotyrosine labeling. Increases in key inflammatory cytokines in both genotypes early after pMCAO indicate that inflammation contributes to the evolving pathology but is not influenced by the lack of Cp.

This study provides clear evidence that lack of Cp results in larger lesions and worsening of functional recovery as determined by the "sticky-tape" removal test. Wildtype mice take longer time to remove the adhesive tape between 1 and 7 days after pMCAO as compared to sham-control mice, and subsequently return to almost normal values at day 14. This may be due to neural plasticity-mediated reorganization of neural circuits that lead to recovery of function (Carmichael et al., 2001; Arvidsson et al., 2002; Winship and Murphy, 2009). Of interest to note is that function worsens in Cp null mice as compared to wildtype mice at 7 and 14 days, suggesting that the lack of Cp may induce neurogenerative processes during the second week after pMCAO in Cp null mice. This may be related to free radical mediated damage triggered by the absence of the two major functions of Cp, namely (i) its ability as a ferroxidase to safely convert ferrous iron (Fe2+) to its ferric (Fe3+) form; (ii) its ability to partner with Fpn to export iron out of cells (Calabrese et al., 1989; Patel et al., 2002).

Ferrous iron can react with hydrogen peroxide, which is produced in mitochondria through normal cellular metabolism and via the Fenton reaction produce highly toxic hydroxyl radicals that can damage lipids, proteins and DNA (Hentze et al., 2004; Rouault and Tong, 2005). Ceruloplasmin has six domains with three mononuclear type 1 coppers in domains 2, 4 and 6, as well as a trinuclear copper center that lies between domains 1 and 6 (Calabrese et al., 1989; Zaitsev et al., 1999). The mononuclear coppers in domains 4 and 6 participate in the oxidation of ferrous to ferric iron (Zaitsev et al., 1999; Quintanar et al., 2004, 2007). This ferroxidase activity is coupled via electron transfer to the reduction of dioxygen to 2H2O at the trinuclear copper center (Calabrese et al., 1989; Zaitsev et al., 1999; Quintanar et al., 2004; Bento et al., 2007). Ceruloplasmin, therefore, safely oxidizes ferrous iron without the generation of free radicals as would happen with the Fenton reaction. Mutations in the trinuclear copper center lead to loss of ferroxidase activity and the degradation of Cp as seen in patients with aceruloplasminemia (Kono, 2013). These patients show iron accumulation in the CNS which increases with age. We have seen increased oxidative stress and iron deposition in the CNS in the Cp null mice with age (Patel et al., 2002; Jeong and David, 2006). On the other hand, iron accumulation occurs because Cp partners with Fpn, the iron efflux transporter, to oxidize ferrous iron that is exported via Fpn (Jeong and David, 2003). In the absence of Cp, Fpn is internalized and degraded (De Domenico et al., 2007), leading to lack of iron efflux and thus iron accumulation. It therefore makes sense that we see significant reduction in the expression of Fpn in the ischemic brain of mice that lack Cp, as well as see increased oxidative damage (4HNE and 3NT) in these mice.

Using ferritin immunofluorescence labeling as a surrogate marker for iron we previously reported that iron accumulation in normally aging Cp null mice occurs in astrocytes, and that the neuronal loss seen may be due to the inability of astrocytes to deliver iron to neurons (Jeong and David, 2006). This may be because unlike neurons which do not directly contact CNS blood vessels, astrocytes are intimately associated with blood vessels via their astrocytic end-feet (Jeong and David, 2006; McCarthy and Kosman, 2015). In other words, astrocytes are equipped to acquire iron from blood vessels in the CNS and deliver it to neurons (McCarthy and Kosman, 2015). After CNS injury, such as spinal cord injury, Cp null mice show greater iron accumulation that is localized to macrophages/microglia at the site of injury (Rathore et al., 2008). These are mainly macrophages that have phagocytosed red blood cells (Rathore et al., 2008). However, in CNS lesions in which hemorrhage does not occur, as in multiple sclerosis or its animal model (EAE), macrophages and microglia accumulate iron likely from phagocytosis of dead and dying cells (Hametner et al., 2013; Zarruk et al., 2015). In the present study, we also see iron accumulation in CD11b+ macrophages/microglia initially around the border of the lesion and later throughout the infarcted area, suggesting that these iron laden cells may be infiltrating macrophages and resident microglia that have phagocytosed dying cells. Interestingly, in Cp null mice, increased ferritin staining was seen in GFAP+ astrocytes lining the lesion border as GPI-Cp is normally expressed selectively by astrocytes in the CNS (Patel and David, 1997). This may add to the iron burden in Cp null mice compared to wildtype mice with pMCAO.

It is not clear at present what might underlie the dysregulated massive 19.5-fold increase in DMT1 protein expression in wildtype mice 72 h after pMCAO. Interestingly, the increase in DMT1 at this time point in Cp null mice was 3 times less (sixfold). The latter might be a way to counteract reduced iron efflux and resultant increase in iron accumulation in Cp null mice due to the reduction in Fpn (50%). The retention of Fpn on the cell surface requires the presence of GPI-Cp (De Domenico et al., 2007). Given that Cp is expressed only in astrocytes in the CNS, the 50% reduction in Fpn seen by Western blot is likely to largely affect astrocytes. In fact, the immunofluorescence labeling experiments showed that there is increased ferritin expression in GFAP+ astrocytes in Cp null mice compared to wildtype mice. So, the lower level of increase in DMT1 in Cp null mice could be a compensatory mechanism that prevents even greater increases in intracellular iron in astrocytes. As mentioned earlier, since the increase in DMT1 in both genotypes is not accompanied by increases in TfR1 expression, it is unlikely that the increased DMT1 is involved in the release of endosomal iron from Tf-TfR1-dependent iron uptake but instead is likely to be involved in Tf-independent cellular iron uptake. We have previously reported that astrocytes express DMT1 on their cell surface but not TfR1, and so can take up iron independent of TfR1 (Jeong and David, 2003) as seen in other non-neural cell types (Wessling-Resnick, 2000).

Inflammation is a well-recognize feature of ischemic stroke (Offner et al., 2006; Wang et al., 2007; Denes et al., 2010; Zarruk et al., 2012; Amantea et al., 2015) in which there is rapid upregulation of pro-inflammatory cytokines. It is important to note that we see a rapid increase in expression in proinflammatory cytokines IL-1β, TNFα and IL-6 in both genotypes with no differences between them over the first 14 days after pMCAO. This suggests that the lack of Cp has no influence on the expression of these key inflammatory mediators after ischemic stroke. Recent studies have reported that parenteral treatment with soluble Cp improved lesion size and functional outcome after transient ischemic stroke in mice and rats (Tuo et al., 2017). What we show here is that in addition, the normal levels of Cp present in wildtype mice already has a significant beneficial effect, since lack of this Cp leads to worsening of functional recovery and to larger lesion size.

#### CONCLUDING REMARKS

This study further elucidates the role of Cp in ischemic stroke. Ceruloplasmin has two fundamental roles (i) in oxidizing ferrous iron and (ii) in the egress of ferrous iron out of cells via Fpn. Lack of Cp, therefore, leads to excess iron buildup in cells. Using wildtype and Cp null mice, we provide clear evidence that Cp has tissue protective effects in the acute phase (24 h) as the ischemic lesion size is larger in Cp null mice. Further evidence of the importance of this antioxidant in the acute phase after pMCAO is suggested by the ninefold increase in its expression at 72 h. In addition, our functional analysis revealed that the absence of Cp leads to worsening of functional recovery at 7 and 14 days after onset of ischemia. This suggests a late effect which could be due to detrimental effects on synaptic reorganization and circuitry via free-radical mediated effects on neurons, astrocytes and microglia. Future work will need to address if Cp influences microglial and astrocyte responses after ischemic stroke that can impact on

#### REFERENCES


functional recovery. Analysis of dendritic arborizations and dendritic spines in the penumbral regions of the cortex in wildtype and Cp null mice will also be worth assessing. Evidence of increased iron in the ischemic tissue was detected by iron histochemistry. Furthermore, immunofluorescence staining for ferritin (a surrogate marker for iron) showed increased ferritin in both astrocytes and macrophages/microglia after pMCAO in wildtype mice but this is significantly higher in astrocytes in Cp null mice. These perturbations in iron levels in the tissue was accompanied by increased oxidative damage to lipids and proteins. There is evidence that systemic treatment with Cp improves outcome after ischemic stroke in rodents. Our current study shows that the increased expression of Cp in the ischemic cortex of wildtype mice has important neuroprotective effects after cerebral ischemia.

#### AUTHOR CONTRIBUTIONS

JZ, FR, and SD planned the study and interpreted the results. JZ, FR, and LL carried out the experiments. JZ and FR analyzed the data and edited the manuscript. SD wrote the manuscript.

# FUNDING

This work was supported by a grant from the Canadian Institutes of Health Research (CIHR) (MOP-142231) to SD. JZ was supported by Postdoctoral Fellowships (PDF) from the Heart and Stroke Foundation of Canada, and the Fonds de Recherche du Québec – Santé.

#### ACKNOWLEDGMENTS

We thank Ourania Tsatas and Laura Curran for technical assistance.

sensorimotor deficits in mice. Nat. Protoc. 4, 1560–1564. doi: 10.1038/nprot. 2009.125


for iron accumulation. Neurobiol. Dis. 81, 93–107. doi: 10.1016/j.nbd.2015. 02.001


**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 Ryan, Zarruk, Lößlein and David. 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 Relevancy of Data Regarding the Metabolism of Iron to Our Understanding of Deregulated Mechanisms in ALS; Hypotheses and Pitfalls

Camille Petillon<sup>1</sup> , Rudolf Hergesheimer<sup>2</sup> , Hervé Puy<sup>3</sup> , Philippe Corcia2,4 , Patrick Vourc'h1,2, Christian Andres1,2, Zoubida Karim<sup>3</sup> and Hélène Blasco1,2 \*

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Raffaella Gozzelino, Centro de Estudos de Doenças Crónicas (CEDOC), Portugal Valentina Bonetto, Istituto di Ricerche Farmacologiche Mario Negri, Italy David Devos, Université de Lille, France

> \*Correspondence: Hélène Blasco helene.blasco@univ-tours.fr

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 14 October 2018 Accepted: 19 December 2018 Published: 15 January 2019

#### Citation:

Petillon C, Hergesheimer R, Puy H, Corcia P, Vourc'h P, Andres C, Karim Z and Blasco H (2019) The Relevancy of Data Regarding the Metabolism of Iron to Our Understanding of Deregulated Mechanisms in ALS; Hypotheses and Pitfalls. Front. Neurosci. 12:1031. doi: 10.3389/fnins.2018.01031 <sup>1</sup> Laboratoire de Biochimie, CHRU de Tours, Tours, France, <sup>2</sup> INSERM, U1253, Université de Tours, Tours, France, <sup>3</sup> Centre de Recherches sur l'Inflammation, Equipe "Hème, Fer et Maladies Inflammatoires", UMR 1149/ERL CNRS 8252, Université Paris Diderot Paris 7, UFR de Médecine Site Bichat, Paris, France, <sup>4</sup> Centre SLA, Service de Neurologie, CHRU de Tours, Tours, France

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease caused by the loss of motor neurons. Its etiology remains unknown, but several pathophysiological mechanisms are beginning to explain motor neuronal death, as well as oxidative stress. Iron accumulation has been observed in both sporadic and familial forms of ALS, including mouse models. Therefore, the dysregulation of iron metabolism could play a role in the pathological oxidative stress in ALS. Several studies have been undertaken to describe iron-related metabolic markers, in most cases focusing on metabolites in the bloodstream due to few available data in the central nervous system. Reports of accumulation of iron, high serum ferritin, and low serum transferrin levels in ALS patients have encouraged researchers to consider dysregulated iron metabolism as an integral part of ALS pathophysiology. However, it appears complicated to suggest a general mechanism due to the diversity of models and iron markers studied, including the lack of consensus among all of the studies. Regarding clinical study reports, most of them do not take into account confusion biases such as inflammation, renal dysfunction, and nutritional status. Furthermore, the iron regulatory pathways, particularly involving hepcidin, have not been thoroughly explored yet within the pathogenesis of iron overload in ALS. In this sense, it is also essential to explore the relation between iron overload and other ALS-related events, such as neuro-inflammation, protein aggregation, and irondriven cell death, termed ferroptosis. In this review, we point out limits of the designs of certain studies that may prevent the understanding of the role of iron in ALS and discuss the relevance of the published data regarding the pathogenic impact of iron metabolism deregulation in this disease and the therapeutics targeting this pathway.

Keywords: iron metabolism, amyotrophic lateral sclerosis, ferritin, biomarkers, ALS

# INTRODUCTION

fnins-12-01031 January 12, 2019 Time: 17:3 # 2

Amyotrophic lateral sclerosis (ALS) is a fatal, neurodegenerative disorder characterized by the selective loss of motor neurons (MN) in the spinal cord, brainstem, and cerebral cortex. This leads to progressive paralysis and death by respiratory failure (Hadzhieva et al., 2013). The median survival is three to five years from onset, and the only drug approved for treatment in Europe is riluzole that extends the average patient's lifespan for only a few more months. ALS is either familial (5–10% of cases) or sporadic (90–95% of cases). Genetic factors including mutations in the C9orf72, SOD1 (Superoxide Dismutase 1), TARDBP (Transactive Response DNA-Binding Protein 43), and FUS (Fused in Sarcoma) genes could explain nearly 50% of familial cases (van Blitterswijk et al., 2012). Although the specific etiology of sporadic ALS remains unknown, several pathophysiological mechanisms have been described, such as oxidative stress, glutamate-mediated excitotoxicity, impaired axonal transport, mitochondrial dysfunction, and protein aggregation (Shaw, 2005).

Oxidative stress, characterized by reactive oxygen species (ROS), has been suggested to play a major role in the disease, because it has been described in all cases (D'Amico et al., 2013), even in those independent of mutations in SOD1 (corresponding to 80% of familial ALS patients, and most of the cases of sporadic forms). One of the crucial factors associated with the production and presence of ROS is iron, especially via the Fenton reaction, producing highly reactive hydroxyl radicals (Hadzhieva et al., 2013). Furthermore, an iron-dependent mechanism of cell death has recently been described, termed ferroptosis (Dixon et al., 2012). Various reports have suggested a neurodegenerative role of ferroptosis in which the affected neurons exhibit an accumulation of iron and lowered glutathione levels in rodent models (Wu et al., 2018).

Interestingly, perturbed iron regulation has notably been observed in many pre-clinical and clinical studies of ALS (Kwan et al., 2012; Golko-Perez et al., 2017). Determining the etiology of this deregulation, the specific role of iron, whether or not iron is only linked to oxidative stress and, finally, its potential as a diagnostic tool, prognostic marker, or therapeutic target all remain challenges. Studies evaluating iron metabolism in ALS, yet rarely included confusion biases or related these mechanisms to other pathophysiological pathways. In addition, many studies have focused on metabolic markers in the bloodstream, but ALS is mainly a disease of the central nervous system (CNS) and iron metabolism is different in the brain.

Accordingly, we question whether the quantity and the quality of the published data enable us to propose one or several mechanisms linking iron to ALS and to justify their relevance in clinical practice.

We focus this review on the publications concerning the metabolism of iron in the brain, the methods applied to explore iron metabolism in ALS, and on the published data on ALS patients and mutant SOD1 mice. We discuss the role of each marker of iron metabolism evaluated in this pathological context. We also highlight limits of the designs of certain studies that may prevent the understanding of the role of iron in ALS. Lastly, we offer suggestions regarding pitfalls to avoid when determining the role of iron metabolism in this disease's pathogenesis and discuss the associated therapeutic perspectives.

## BODY IRON METABOLISM

Body iron metabolism relies on the coordination of several proteins having crucial roles in iron maintenance. Ferritin and transferrin (Tf) are the main carriers of iron in the blood, while proteins such as intracellular iron regulatory proteins (IRPs) and systemic hepcidin are key factors of iron metabolism regulation. Other proteins, including divalent metal transporter-1 (DMT1) and ferroportin (FPN1) in association with ferroxidases such as duodenal cytochrome B (DcytB), hephastin (Hp) and ceruloplasmin (Cp), are involved in the cellular membrane trafficking of iron. Finally, the production of hemoglobin (Hb), myoglobin, and various enzymes (cytochromes, aconitases, etc.) that require iron to function, are the final actors of iron metabolism.

Iron storage and recycling are directed by the liver and the spleen's macrophages that store excess iron with ferritin. They recycle iron following phagocytosis of senescent red blood cells and degradation of hemoglobin. Released iron is exported by FPN1 then oxidized by Cp to be re-bound to serum Tf for reuse. Iron absorption is mediated by the duodenal epithelial cells, which absorb both heme and non-heme iron from the diet according to the body's needs. Heme is taken up by specific carrier proteins and is catabolized by heme-oxygenase-1 to release ferrous iron (Fe2+). Non-heme ferric iron (Fe3+), is reduced to Fe2<sup>+</sup> by the duodenal Dcytb before uptake by the apically expressed DMT1. Ferrous iron is delivered to the circulation via the basolateral exporter FPN1 in association with the ferroxidase Hp. The complex (Fe3+-Tf) is delivered to cells by binding to its receptor, transferrin receptor 1 (TfR1), then the complex is taken up by the cell through the invagination of the membrane, initiating endocytosis. Finally, iron is released from transferrin and is transported out of the endosome by DMT1. In all cells of the body, excess Fe2<sup>+</sup> is converted to Fe3<sup>+</sup> by the ferroxidase activity of ferritin H-chain (Ft-H) before storage in the ferritin mineral core (Ft-H and Ft-L) (Hentze et al., 2010).

Hepcidin is a central player in iron homeostasis, acting as a 25 amino-acid mature peptide predominantly expressed in the liver (Nicolas et al., 2001). Hepcidin acts by triggering FPN1 activity and/or protein level and degradation of duodenal DMT1 protein (Nemeth et al., 2004; Brasse-Lagnel et al., 2011; Aschemeyer et al., 2018). Therefore, hepcidin is a hyposideremic factor that reduces both the intestinal absorption of iron and its release by macrophages, hepatocytes, and other cell types. Hepcidin is eliminated in the urine, and small fragments from degradation of 20–25 aa have been found (Park et al., 2001). Hepcidin is expressed by the HAMP gene located on chromosome 19 (19q13.1). This gene is upregulated by iron overload, infection, and inflammation. It is downregulated by iron deficiency, hypoxia, anemia, and stimulated erythropoiesis (Hentze et al., 2010).

#### BRAIN IRON METABOLISM

fnins-12-01031 January 12, 2019 Time: 17:3 # 3

Iron is a vital cofactor in many metabolic processes of the CNS, including oxidative phosphorylation, myelin synthesis, neurotransmitter production, and oxygen transport (Crichton et al., 2011). The presence of the blood–brain barrier (BBB) protects the brain from the daily fluctuations of systemic iron and also from iron disorders that affect all peripheral tissues. The precise mechanism by which the BBB mediates iron influx from the blood to the parenchyma of the brain still remains to be uncovered (**Figure 1**). One hypothesis currently accepted for the transfer of iron across the blood capillary endothelial cells (BCECs) states that Tf-TfR1 transcytosis creates a passive conduit without intracellular processing. However, a new model based on the classical process of iron and experimental data has been recently proposed (Simpson et al., 2015). In this model, iron that enters the CNS is delivered to BCECs by the Tf-TfR complex. Upon internalization into the endosome, iron is released from Tf, exported to the cytosol by DMT1, then exported out of the cell by FPN with the help of Cp ferroxidase expressed at the external cell surface of astrocytes (Jeong and David, 2003; Wu et al., 2004). The bioGPS database also shows Cp and FPN transcripts evenly distributed throughout the CNS (BioGPS – your Gene Portal System [Internet], 2018).

Released iron is most likely captured by brain Tf that is synthesized and secreted by oligodendrocytes and the choroid plexus into the brain interstitium. Intriguingly, in this proposed model, evidence of the presence of DMT1 in BCECs is yet to be demonstrated since, not until recently, there are conflicting studies indicating both the presence and absence of DMT1 transcripts and protein (Gunshin et al., 1997; Moos et al., 2006; Skjørringe et al., 2015). However, several arguments promote that in BCECs, the expression level of DMT1 is probably low but that its amount is sufficient to ensure the iron export from the endosome (Skjørringe et al., 2015). A third, alternative explanation of iron delivery involving the uptake of ferritin across the BBB has been proposed, but further support is still necessary to confirm this mechanism (Fisher et al., 2007; Todorich et al., 2011).

The acquisition of iron by brain cells is mediated by different pathways. Neurons express abundant TfR1 and acquire most of their iron via transferrin. Astrocytes do not appear to express TfR1 (Dringen et al., 2007), although this is controversial (Qian et al., 1999). However, they express DMT1 and FPN at the plasma membrane (Wu et al., 2004; Lane et al., 2010). Oligodendrocytes accumulate the most amount of brain iron and are the predominant producers of Tf in the brain (Burdo and Connor, 2003). Transferrin is widely distributed in the CNS but is preferentially expressed in the spinal cord, pre-frontal cortex, and hypothalamus (Mackenzie et al., 2007). Oligodendrocytes also express the ferritin receptor Tim-2, suggestive of an ability to acquire iron from ferritin (Hulet et al., 1999). Most brain cells express cytosolic ferritin for storing excess iron, and relatively high transcript levels are found in the spinal cord, hypothalamus, and pre-frontal cortex; neurons contain the least, while microglia contain the most (Mackenzie et al., 2007; Crichton et al., 2011; Singh et al., 2014).

Despite the high concentration of hepcidin in the liver, recent studies have also revealed the widespread expression of hepcidin in the brain, specifically in the spinal cord, hippocampus, and olfactory bulb (Zechel et al., 2006; Mackenzie et al., 2007; Wang et al., 2008). These findings imply that hepcidin might also

be a key regulator in brain iron homeostasis. Hepcidin, similar to its action in the intestine, may regulate the expression of iron transport proteins DMT1 and ferroportin (Puy et al., 2018). This could explain the maintenance of iron balance in neurons and glial cells through the control of their iron uptake and release (Du et al., 2015).

One regulatory protein of brain iron that is currently receiving increased attention is the HFE protein. Mutations in the HFE gene are commonly associated with hereditary hemochromatosis in which an iron overload is the consequence of hepcidin deficiency. HFE transcripts in the CNS are relatively concentrated in the spinal cord, temporal and occipital lobes, and the cortex (Mackenzie et al., 2007). The presence of the HFE protein at the interface between the brain and endothelial cells of the microvasculature, and in the cerebrospinal fluid (CSF), may also affect iron content and contribute to iron overload in neurodegenerative disorders. Indeed, HFE mutations have been investigated as risk factors for neurodegenerative disorders (Nandar and Connor, 2011).

## METHODS TO STUDY IRON METABOLISM

Imaging and biological methods have been utilized on cohorts of patients as tools to study the dysregulation of iron metabolism. In SOD1 transgenic mice and cell models, several approaches, such as concentration measurements, protein expression followed by Western blotting and/or immunofluorescence, and measuring mRNA expression have been employed. **Table 1** summarizes the different studies performed to explore iron metabolism in ALS. It is important to notice the significant diversity among models, tissues, and markers studied. For example, cell lines, especially neurons or MN (NSC-34, SH-SY5Y), have been transfected with mutant SOD1 gene constructs in order to attempt to mimic ALS in humans (Schlachetzki et al., 2013). Glial cells are also used in co-culture with neurons, given the hypothesis that ALS is a non-cell-autonomous disease, meaning that multiple cell types can play into the disease's pathogenesis (Smethurst et al., 2016; Zeineddine et al., 2017). Astrocytes expressing WT or mutant SOD1 have been co-cultured with primary cortical neurons (Kunze et al., 2013). Microglia expressing mutant SOD1 have been co-cultured, as well, with iPSC-derived MN (Frakes et al., 2014). Another example is the co-culture of human adult primary sALS astrocytes and human embryonic stem cell-derived MN (Re et al., 2014).

Nevertheless, transgenic mice and rats represent the gold standard of preclinical ALS modeling. Rodent models exist for the genetic mutations in SOD1, TARDBP, and FUS and mouse models for C9orf72 repeat expansions have also been developed (Chew et al., 2015; Jiang et al., 2016; Liu et al., 2016). In particular, the models in which human genomic mutant SOD1 is overexpressed have been used extensively with the following most studied mutations: SOD1 G93A, SOD1 G37R, and SOD1 G85R (Shaw, 2005). Clinical studies of cohorts of ALS patients also add to the knowledge of iron metabolism.

#### MARKERS OF DEREGULATED IRON METABOLISM IN ALS

We searched the Pubmed database to identify potentially relevant studies on the association between iron metabolism and ALS. We used free text and the medical subject heading (MeSH) terms "amyotrophic lateral sclerosis," "ALS," "iron," "iron metabolism," "iron accumulation," "biomarker," and "iron chelation." Then, we searched for each marker of iron metabolism (DMT1, TfR1, ferritin, ferroportin, hepcidin, HFE,. . .) AND ALS (OR "amyotrophic lateral sclerosis").


BCECs, blood capillary endothelial cells; BBB, blood–brain barrier; TfR1, receptor transferrin 1; DMT1, divalent metal ion transporter 1.

#### Iron Accumulation

fnins-12-01031 January 12, 2019 Time: 17:3 # 5

Increased iron levels were reported in spinal cord MN in two ALS SOD1-mouse models, SOD1 G93A (Lee et al., 2015) and SOD1 G37R (Jeong et al., 2009), by staining ventral MN with Perl's (Fe3+) and Turnbull's (Fe2+) reactants.

Iron accumulation was also observed in the CNS of ALS patients by magnetic resonance imaging (MRI) (Oba et al., 1993; Ignjatovic et al., 2013). T2-weighted MRI scans were used to detect areas of hypo-intensity (iron deposits) in the cerebral cortex of ALS patients. Likewise, they used Perl's-reactant for the analyses of postmortem brains and revealed a large amount of ferric iron in the pre-central cortices (Oba et al., 1993). More recently, accumulated iron was detected in multiple brain regions, including the motor cortex, substantia nigra, globus pallidus, red nucleus, and the putamen in ALS patients by quantitative susceptibility mapping (Acosta-Cabronero et al., 2018).

#### The Alteration of Iron Transport

A study reported the upregulation of DMT1 and TfR1 mRNAs in G93A-SOD1 cells (SH-SY5Y) (Hadzhieva et al., 2013). In another study, the expression of various proteins involved in iron transport, such as DMT1, TfR1, ferroportin, and ceruloplasmin was evaluated in the spinal cord of SOD1 G37R mice (Jeong et al., 2009). Data at 12 months of age (end stage) showed an increase in the mRNA and protein expression of DMT1, ferroportin, and ceruloplasmin in the cervical spinal cord. Furthermore, double immunofluorescence labeling revealed that in 12-monthold SOD1 G37R transgenic mice the expression of DMT1 was augmented in MN in the ventral horn of the spinal cord. It was also shown that these iron transport markers were more abundant in SOD1 G37R mice at 4 months of age, and mRNA levels were higher in the lumbar region of the spinal cord in the early stages of the disease.

Two independent clinical studies have shown increased values of transferrin saturation (TS) in a cohort of ALS patients and controls (Nadjar et al., 2012; Veyrat-Durebex et al., 2014). However, only the study by Nadjar et al. (2012), performed on a large ALS population (694 patients), thus presenting higher statistical power (Veyrat-Durebex et al., 2014), showed a significant decrease in serum transferrin levels in ALS subjects. Interestingly, it was also demonstrated that low serum transferrin concentrations were associated with a higher loss in body weight at diagnosis in ALS patients (Veyrat-Durebex et al., 2014), which could link iron metabolism to nutritional status – largely associated with energetic metabolism in ALS.

#### Increased Serum Ferritin

Serum ferritin is also increased in ALS patients (Goodall et al., 2008; Qureshi et al., 2008; Su et al., 2015). Six case-control studies published between 2008 and 2015 investigating the possible relationship between ALS patients and their susceptibility to elevated serum ferritin levels were included in a meta-analysis (Hu et al., 2016), including 1053 cases of ALS and 760 healthy control cases from the United Kingdom, United States, France, and Japan. The serum ferritin concentrations in ALS were statistically higher compared to healthy controls (p < 0.00001). The limitations revealed by this meta-analysis were a high heterogeneity (p = 0.03; I<sup>2</sup> = 50%) and a publisher bias.

#### Alteration of HFE Gene and Hepcidin Alteration

A meta-analysis assessing the roles of the H63D and C282Y variants of the HFE gene in ALS from 14 observational studies illustrated a significant association of the C282Y variant to the disease, meaning a decreased risk for ALS, while confirming that the H63D variant showed no relevant association (Li et al., 2014). In fact, hemochromatosis linked to mutations in the HFE gene has been attributed solely to the homozygous C282Y variant. The results of this meta-analysis imply that the C282Y HFE variant could be a protective factor against ALS, which is a priori surprising given the physiopathology of ALS, in which iron is expected to play a toxic role.

In one report, hepcidin plasma concentrations were measured in SOD1 transgenic rats (G93A) at 12 weeks (asymptomatic, ALS stage I), 21 weeks (disease onset, ALS stage II) and 24 weeks (end-stage disease, ALS stage III) of age, and were significantly increased in the blood of ALS II and III animals (Halon et al., 2014). On the other hand, the plasma concentrations of prohepcidin, the precursor to hepcidin, were not significantly altered in a cohort of 29 ALS patients (from nine to 28 months postdiagnosis) compared to control group. Nevertheless, there was a positive correlation between pro-hepcidin and IL-6 levels in ALS patients compared to controls, suggesting exacerbated iron release by macrophages and inflammation in ALS patients (Mitchell et al., 2010).

#### CAN WE INTEGRATE PUBLISHED DATA TO SUGGEST A ROLE OF IRON DEREGULATION IN ALS?

Regarding the diversity of models and the methodology used, as well as the studied markers, the findings with respect to iron metabolism seem heterogeneous. It appears complicated to suggest a global mechanism of dysregulation. Accumulation of iron, higher serum ferritin, and lower serum transferrin levels in ALS patients have clearly encouraged researchers to consider iron deregulation as an essential mechanism in ALS. The critical review of the literature led us to identify the limits preventing the understanding of iron metabolism disturbance in ALS, and more specific findings in the context of ALS must be provided.

#### PITFALLS TO AVOID WHEN STUDYING IRON METABOLISM IN ALS

#### Types of Models

The lack of the complex interplay between MN and their surrounding environment is the precise problem that we face with the published data on cellular models, encouraging us to turn to other models. Although SOD1 mutations are

rare in sALS, there are phenotypic similarities with fALS, and it has been assumed that studying the familial form could provide insight into the sporadic form. However, the validity of these assumptions has been questioned (Clerc et al., 2016) and, even if clinical manifestations are not so different, the pathophysiological mechanisms are obviously distinct. For instance, TDP-43 aggregates were not observed in patients suffering from ALS forms associated with SOD1 mutations (Mackenzie et al., 2007), whereas they were in a SOD1G93A mouse model (Jeon et al., 2018). Due to this discrepancy between postmortem patient and mouse tissues, one should consider the TDP-43 aggregation to be unrelated to the SOD1 ALS pathology if this artifactual aggregation is unavoidable in SOD1 mouse models. Moreover, exploring iron metabolism closely linked with the consequence of SOD1 activity, i.e., production of ROS, seems biased. So, this model that is subjected to oxidative stress may not fit with the investigation of iron metabolism. In addition, the relatively rapid neurodegeneration that occurs in these mice can reduce the quantity of iron in the affected neurons, potentially confounding the interpretation of the corresponding results. Altogether, we are aware that SOD1 mouse models remain valid for the study of certain aspects of ALS pathogenesis, but the scope of the conclusions must be limited in relation to the restraints of this model, especially in the exploration of iron metabolism.

#### Localization of Iron Metabolism Exploration

Data reporting the main genetic loci involved in iron accumulation would be the first step to the understanding of the role of iron in ALS. Another study, in fact, showed that inducing inflammation with intravenous injections of lipopolysaccharide (LPS) in mice led to hepcidin expression in the cortex, hippocampus, and striatum (Wang et al., 2008). In fact, hepcidin along with the other iron regulators are shown to be expressed in the spinal cord and cortex, the two main areas affected by ALS. Therefore, investigating their behavior in these tissues in ALS models could yield insight regarding iron accumulation hot spots. Published data on mouse models have revealed that the spinal cord is especially affected with increased expression of iron transporters (DMT1, ferroportin, and TfR1). Much attention has been turned toward ALS patients' brains, but incomplete data prevent us from drawing conclusions in humans. Although there is limited data for factors of iron metabolism in the CSF of ALS patients, data from MRI and postmortem brain exploration have presented interesting information to locate the main regions of accumulation.

In addition, the identification of cells types is also essential to integrate all of these findings. Some researchers have highlighted the role of glia (Jeong et al., 2009) by observing an increased cytosolic ferritin expression not identified in MN, suggesting that neurons and glia accumulate iron via different mechanisms.

Finally, it is necessary to be aware of the differences in the expression levels between mice and humans, since murine models are the "gold standard" in studying ALS. Indeed, the differences in distribution and expression levels pose limitations on the conclusions inferred from these models (**Supplementary Table S1**). Even though certain regulators exhibit differential distributions and expression levels in the mouse CNS compared to that of humans, mouse microglia consistently show expression for most, if not all, the regulators mentioned here. Because cells other than neurons can also become affected in ALS and that microglia have quite an unequivocal relationship with iron-driven inflammation, paying attention to the levels of these iron regulators in the microglia of murine ALS models could increase the relevancy and decrease the limitations of murine models.

#### Extrapolation From Liver to the Brain

The interaction between the liver and the brain regarding iron metabolism must be taken into account, but it is difficult to extrapolate from the published data. Indeed, most of the factors of iron metabolism have not yet been studied in the brain, and we cannot deduce a mechanism from their activity in other organs. For example, it was suggested that the HFE C282Y variant could be a potentially protective factor for ALS. But, it is difficult to extrapolate the role of HFE from the liver to the brain, given that ALS is not a disease primarily affecting the liverapart from the manifestation of a fatty liver (Dupuis et al., 2008). Similarly, it could be misleading to compare the levels of metabolic regulators of iron in the serum and the brain, since the main carriers and regulatory processes are not fully understood in healthy persons and even less so in ALS patients.

#### RELEVANT PARAMETERS TO INCLUDE IN IRON METABOLISM EXPLORATION IN ALS PATIENTS

#### Dynamic Variation of Iron Metabolism

Importantly, a multi-factorial disease, such as ALS, is characterized by regulation and counter-regulation, which is dependent on the progress of the disease and on other pathophysiological mechanisms. Information regarding the relation between iron metabolism and disease progression would help the integration of iron's role in the pathogenesis. One does, therefore, wonder if iron accumulation in the CNS of ALS patients is a factor for poor prognosis and if increasing concentrations of iron might cause the disease to progress more quickly over time. However, very few data on iron concentrations in the brain of ALS patients are available, and even fewer about variations in these concentrations throughout the progression of the disease. Nadjar et al. (2012) showed that high levels of serum ferritin (>156 µg/L), were significantly associated with a reduced survival time. ALS patients with low levels of ferritin lived around 300 days longer than patients with high levels (p < 0,04). But, serum ferritin does not necessarily reflect the iron accumulation in the CNS.

The pathophysiological mechanisms that could explain why ALS patients have increased iron body storage and how that is responsible for a decreased survival is, at present, unknown.

With respect to the largely unknown mechanism of regulation and/or rebound effect, very little can be concluded for a general

mechanism of iron metabolism in ALS, such as with data concerning hepcidin that has been clearly understudied and whose studies yield contradictory outcomes.

# Confounding Factors for the Interpretation of Iron Metabolism Status

Regarding clinical studies, the first issue is the heterogeneity among ALS patients, since the site of onset, the form of ALS, the evolution, and genetic mutations add and multiply the diversity among patients. To add to this expected and known limit of all cohort studies of ALS, some biological concerns are often overlooked. Several parameters that can disrupt iron metabolism are not taken into account. For example, inflammation, renal dysfunction, and nutritional status are commonly overlooked. Since serum ferritin is an acute-phase reactive protein whose concentration can rise in the presence of acute or chronic inflammation, it might fluctuate depending on the inflammatory status of ALS patients (Hu et al., 2016). Also, CRP concentrations were not measured in all the clinical studies (Qureshi et al., 2008; Ikeda et al., 2012; Nadjar et al., 2012). This was taken into account in the study by Veyrat-Durebex et al. (2014), and CRP values were not significantly different between patients and controls. In Goodall et al. (2008), CRP and ceruloplasmin levels were both negatively correlated with iron and iron saturation in ALS patients showing that iron metabolism disturbance was not due to inflammation.

Kidney dysfunction may alter some parameters of iron metabolism (Maras et al., 2015). The age at onset of ALS [between 50 and 60 years (Shaw, 2005)] could suggest that kidney function starts to be altered, but this data has not been directly reported; no ideal marker has been identified or used in these cases, and clinicians are hesitant to insist on kidney function exploration. Unfortunately, creatinine is not a very reliable marker of kidney function in the context of ALS due to muscular atrophy (Corbett et al., 1982), leading to the mis-evaluation of the glomerular filtration rate (eGF). Veyrat-Durebex et al. (2014) reported a creatinine value, used in the MDRD (modified diet in renal disease) equation, to estimate the glomerular filtration rate. The estimated eGFR was higher (p < 0.0001) in ALS patients than in controls, showing not that renal function was better but that creatinine levels were low probably due to muscular atrophy and that the role of potential defective renal elimination on differences in iron metabolism cannot be explored by this approach (Veyrat-Durebex et al., 2014). The evaluation of creatinine values after taking into account body composition (impedancemetry) or using other markers of renal function (cystatin C, iohexol, etc.) may help with the interpretation of iron metabolism, independent of muscle atrophy and renal elimination. Similarly, nutritional factors essential to ALS have to be included in iron metabolism interpretation, as it may completely influence the results, either due to disease or management. For example, it has been illustrated that low serum transferrin concentrations are associated with a higher body weight loss at the time of diagnosis in ALS patients (Veyrat-Durebex et al., 2014). Further investigations on nutritional care in ALS patients will be necessary.

We expect promising and complementary findings from CSF studies, but few data has been published to date (Goodall et al., 2008).

## THERAPEUTIC PROSPECTS

The pathogenic impact of iron metabolism deregulation in ALS has been further supported by the partially protective effects of iron chelators in ALS mouse models (Hadzhieva et al., 2014). In recent years, various iron chelators have demonstrated a prolonged survival of SOD1 transgenic mice. A classic iron chelator, desferrioxamine, previously studied in models of neurodegenerative diseases, such as Huntington's disease or Parkinson's disease (Hilditch-Maguire et al., 2000; Zhang et al., 2005), as well as the new iron chelators 5-[4-(2-hydroxyethyl) piperazine-1-ylmethyl]-8- hydroxyquinoline (VK-28) and 5-(Nmethyl-N-propargylaminomethyl)-8-hydroxyquinoline (M30), have also been studied in models of ALS disease (Wang et al., 2011).

Iron chelators significantly delay disease onset (6–12 days on average, p < 0.05), extend the lifespan (8–13 days on average, p < 0.05), and reduce spinal cord motor neuron loss in ALS SOD1 transgenic mice. These chelators are able to prevent ferroptosis, as well (Masaldan et al., 2018). In support of this, a mouse model of aggravated ferroptosis displayed ALS features, such as intense muscular atrophy, rapid paralysis, and ferroptotic neurodegeneration of the motor neurons of the spinal cord (Chen et al., 2015; Masaldan et al., 2018). Thus, the therapeutic effects of iron chelation in SOD1 mice seem to partly stem from the suppression of ferroptosis, which could result from iron accumulation. Besides these findings, VK-28 and M30 reduced ROS generation and inhibited activation of glial cells in the spinal cord (Wang et al., 2011).

In both a SOD1 murine model of fALS and sALS, a low dose of deferiprone, another classic iron chelator, was associated with a decrease in pathological iron accumulation in the central motor pathways (Moreau et al., 2018).

In ALS patients, iron chelation could reduce iron accumulation and the related excess of oxidative stress in the motor pathways. Classical iron chelation could induce systemic iron depletion. For this reason, the safety and efficacy of conservative iron chelation (chelation with a low risk of iron depletion) was assessed in a single-center, single-arm, 12 month-long pilot clinical trial (NCT02164253). After 12 months of treatment with deferiprone, none of the 23 ALS patients enrolled in the trial displayed signs of anemia. The decreases in the ALS Functional Rating Scale (ALSFRS-R) and in the body mass index were significantly smaller in the first 3 months of deferiprone treatment than in the parallel 3-month treatmentfree period (Moreau et al., 2018). A significant decrease in iron concentration, shown by MRI, was observed in the cervical spinal cord and the motor cortex while not in areas outside the motor system (the cerebellum and the occipital cortex) (Moreau et al., 2018).

Currently, the efficacy of deferiprone in ALS patients is being evaluated in a randomized, double-blind, placebo-controlled,

multi-center French study (FAIR-ALS II) with an estimated enrollment of 210 participants and a 12-month period (NCT03293069).

Evidence has revealed that the neuroprotection of iron chelation may result from the attenuation of iron-related oxidative stress, iron accumulation, and ferroptosis. However, we have little perspective on the long term regarding the effects of these iron chelators, and we are not yet certain about their precise mechanism of action. It could very well be that the chelation treatment brings about a long-term effect; it could also simply be temporarily effective, requiring life-long treatment. Moreover, to our knowledge no study has determined the exact specificity of these chelators for iron. We question to which extent one can attribute the therapeutic effect of iron chelators in animal models to the interaction with iron. Therefore, future studies should pay attention to the concentrations of other heavy metals besides iron in order to better assess the overall effect of the chelator on the cell.

#### CONCLUSION

Since iron heavily participates in ROS production and has been seen to accumulate in the context of ALS, we have focused this review on the published findings concerning iron metabolism in this disease. Future studies will need to address the experimental limitations that we have acknowledged in this review. The limitations in the murine models originate from the experimental bias projected from the use of pathogenic SOD1. Because this condition primarily generates ROS, the conclusions relating iron metabolic regulators drawn from these studies are confined to the ALS cases in which SOD1 is mutated, which is only a minute proportion of both familial and sporadic cases. Therefore, it would be of particular interest to compare the effects on iron regulation between the SOD1 models and another type, such as TDP-43 models.

Other limitations in clinical studies analyzing the metabolic alterations of iron stem from the fact that many overlook

#### REFERENCES


the roles of the inflammatory status, renal function, and nutritional status. These factors become very important when considering therapeutic strategies relying on iron chelation. The clinical trials attempt to employ chelators as an early form of treatment but, even though patients show a considerable retention in body mass and a decrease in iron levels in the spinal cord (Wang et al., 2011), this seems only temporary. Furthermore, studies on patients do not necessarily contain a homogeneous cohort of people affected by SOD1-driven ALS, unlike SOD1 murine models. This confounds the interpretations of iron metabolism, since the sub-type of ALS is not controlled. However, these clinical limitations might be nearly impossible to avoid. Therefore, while complementing treatment with nutritional maintenance and keeping track of kidney and inflammatory status, the chelation-guided therapy could be prolonged and possibly more applicable to the general population of ALS patients. Indeed, there appears to be a global dysregulation of iron in ALS cases, and we must gain a more complete understanding of the causes and consequences.

#### AUTHOR CONTRIBUTIONS

CP and RH wrote most of the sections of the manuscript. CP did the bibliography and the literature review on iron in ALS. ZK wrote Section "Body Iron Metabolism" and completed Section "Brain Iron Metabolism". RH did the article proofreading (English). All authors contributed to manuscript revision, read and approved the submitted version.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2018.01031/full#supplementary-material



at MR imaging. Radiology 189, 843–846. doi: 10.1148/radiology.189.3. 8234713


**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 Petillon, Hergesheimer, Puy, Corcia, Vourc'h, Andres, Karim and Blasco. 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 Contribution of Iron to Protein Aggregation Disorders in the Central Nervous System

Karina Joppe<sup>1</sup> \*, Anna-Elisa Roser<sup>1</sup> , Fabian Maass<sup>1</sup> and Paul Lingor1,2,3,4 \*

<sup>1</sup> Department of Neurology, University Medical Center Göttingen, Göttingen, Germany, <sup>2</sup> Center for Biostructural Imaging of Neurodegeneration, Göttingen, Germany, <sup>3</sup> German Center for Neurodegenerative Diseases, Göttingen, Germany, <sup>4</sup> Rechts der Isar Hospital, Technical University of Munich, Munich, Germany

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Neena Singh, Case Western Reserve University, United States Luigi Zecca, Italian National Research Council, Italy Heather R. Lucas, Virginia Commonwealth University, United States

#### \*Correspondence:

Karina Joppe Karina.Joppe@med.uni-goettingen.de Paul Lingor plingor@gwdg.de

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 29 October 2018 Accepted: 08 January 2019 Published: 22 January 2019

#### Citation:

Joppe K, Roser A-E, Maass F and Lingor P (2019) The Contribution of Iron to Protein Aggregation Disorders in the Central Nervous System. Front. Neurosci. 13:15. doi: 10.3389/fnins.2019.00015 The homeostasis of iron is of fundamental importance in the central nervous system (CNS) to ensure biological processes such as oxygen transport, mitochondrial respiration or myelin synthesis. Dyshomeostasis and accumulation of iron can be observed during aging and both are shared characteristics of several neurodegenerative diseases. Iron-mediated generation of reactive oxygen species (ROS) may lead to protein aggregation and cellular toxicity. The process of misfolding and aggregation of neuronal proteins such as α-synuclein, Tau, amyloid beta (Aβ), TDP-43 or SOD1 is a common hallmark of many neurodegenerative disorders and iron has been shown to facilitate protein aggregation. Thus, both, iron and aggregating proteins are proposed to amplify their detrimental effects in the disease state. In this review, we give an overview on effects of iron on aggregation of different proteins involved in neurodegeneration. Furthermore, we discuss the proposed mechanisms of iron-mediated toxicity and protein aggregation emphasizing the red-ox chemistry and protein-binding properties of iron. Finally, we address current therapeutic approaches harnessing iron chelation as a disease-modifying intervention in neurodegenerative disorders, such as Parkinson's disease, Alzheimer's disease, and amyotrophic lateral sclerosis.

Keywords: iron, protein aggregation, neurodegeneration, disease mechanism, iron chelator

# INTRODUCTION

Neurodegenerative disorders (NDDs) rapidly gain importance due to their age-related prevalence and the resulting socio-economic burden (Hindle, 2010; Abbott, 2011). Aging is one of the main risk factors for NDDs (Ashraf et al., 2018) and the constantly growing life expectancy will result in their increased prevalence (Oeppen and Vaupel, 2002). Although different NDDs present a variety of symptoms ranging from cognitive, motor, sensory and/or autonomic failure, neuronal loss is the shared characteristic. Therefore, it is of great relevance to identify common pathophysiological features that are present in multiple NDDs to elucidate general mechanisms of neurodegeneration and potential pathways for intervention.

Iron does not only play a main role in cellular senescence but also in NDDs such as Parkinson's disease (PD), Alzheimer's disease (AD), amyotrophic lateral sclerosis (ALS) or prion diseases (PrD) (Ashraf et al., 2018). As protein aggregation is another shared hallmark among many NDDs

(Iadanza et al., 2018), it is suggestive to assume a mutual interplay of iron and protein aggregation amplifying their detrimental effects. In addition to iron, other transition metals such as copper and manganese are also considered in the pathogenesis of NDDs. This review, however, is focusing on the role of iron in protein aggregation disorders and the reader is referred to alternative publications on the role of other metals (e.g., Carboni and Lingor, 2015; Kim et al., 2018).

In this review, we discuss the mode of action of iron in NDDs and proposed mechanisms of iron-mediated protein aggregation. Finally, we outline recent therapeutic approaches targeting iron as promising treatment option for NDDs.

#### IRON DYSHOMEOSTASIS AND ACCUMULATION

Iron is the most abundant trace metal in the human brain, present in neuronal and glial cells. It acts as a catalytic center for multiple enzymes and supports the synthesis of DNA, neurotransmitters and myelin in the brain. Furthermore, it participates in oxygen transport, neurotransmitter metabolism and mitochondrial respiration (Ward R.J. et al., 2014; Ashraf et al., 2018).

Iron metabolism is tightly regulated by iron responsive elements (IREs). In the presence of iron, binding of the iron regulatory protein (IRP) to IREs controls the translation of mRNAs related to proteins of iron metabolism, e.g., iron import, export and storage proteins. Under physiological conditions, extracellular ferric iron (Fe3+) is mainly bound to the glycoprotein transferrin (Tf), being delivered to cells by transferrin receptors (TfR). Non-transferrin-bound ferrous iron (Fe2+) is mainly transported into the cell via the divalent metalion transporter 1 (DMT1) (Hider and Kong, 2013; Jiang et al., 2017). DMT1 is also an important player in mitochondrial uptake of Fe2<sup>+</sup> (Wolff et al., 2018). Intracellularly, ferritin is the major iron storage protein complex. Before storage in both H- or L-ferritins, Fe2<sup>+</sup> is oxidized by H-ferritin to Fe3+. Under normal conditions, the labile Fe2<sup>+</sup> pool and ferritin molecules are at an equilibrium. The export of Fe2<sup>+</sup> is regulated by ferroportin-1 (FPN1) being controlled by hepcidin (Hider and Kong, 2013; Ashraf et al., 2018).

Impaired iron metabolism coupled with its accumulation in various brain regions are hallmarks of physiological aging. H-and L-ferritins also are more abundant with age (Zecca et al., 2001). During life, both ferritin subunits increase in their concentration within the SN but stay constant within the locus coeruleus. Both regions are important target areas for PD whereas the locus coeruleus is also affected in AD (Zecca et al., 2004a). Brains of patients suffering from NDDs, e.g., PD and AD, are lacking the age-associated rise of both ferritins. In PD, reduced ferritin levels in SN and pathological iron accumulation were found (Dexter et al., 1991; Connor et al., 1995).

Excessive ROS production resulting in oxidative stress is a common feature of NDDs and accumulated redox-active iron triggers ROS formation by the Fenton and Haber-Weiss reactions, providing the basis for catalyzed oxidation processes. Accordingly, iron reacts with hydrogen peroxide, which is a by-product of the mitochondrial respiration and intracellularly abundant, resulting in hydroxyl free radicals (HO•). Therefore, iron fosters the formation of ROS that lead to oxidative stress, inducing mitochondrial dysfunction and cell death (Zecca et al., 2004b).

This said, the reasons for iron accumulation and its precise effects on pathomechanisms in neurodegeneration remain still incompletely understood. Its contribution to the aggregation of disease-relevant proteins may be a major effector of its toxicity in NDDs.

# PROTEIN AGGREGATION

A shared hallmark of numerous NDDs is protein aggregation. For example, α-synuclein aggregates are the main components of Lewy bodies in PD (Spillantini et al., 1998), whereas neurofibrillary tangles and plaques in AD are composed of Tau and Amyloid beta (Aβ), respectively (Glenner and Wong, 1984; Brion, 1998). Aggregation of TDP-43 or SOD1 are observed in ALS (Brown, 1998; Neumann et al., 2006). Recent data demonstrate, however, that aggregation of one particular protein is not specific for one disease (e.g., Cisbani et al., 2017; Trist et al., 2018). Under physiological conditions, the ubiquitin proteasome system (UPS), autophagosomes and chaperone activity ensure the clearance of protein aggregates (Stroo et al., 2017). However, genetic or environmental factors can disturb the balance of aggregate formation and clearance, so that native soluble proteins or peptides start misfolding and assemble into insoluble betasheet oligomers and protofibrils. This filamentous aggregation results in amyloid fibrils and protein inclusion formation. For different disease-dependent proteins this aggregation process is likely to follow similar pathways (Soto and Pritzkow, 2018).

Whereas for protein inclusions a possible neuroprotective role is still discussed, oligomers and protofibrils of the abovementioned species are very likely neurotoxic. Amyloid structures are believed to impair axonal transport, DNA transcription and the UPS, and trigger mitochondrial dysfunction, synaptic dysfunction and oxidative stress (Dhouafli et al., 2018; Iadanza et al., 2018). Furthermore, oligomers increase the lipid bilayer conductance and, therefore, induce calcium dyshomeostasis (Verma et al., 2015). Altogether, these mechanisms contribute to cellular dysfunction and cytotoxicity.

#### IRON AND PROTEIN AGGREGATION

Via interaction with redox-active metal ions, amyloidogenic forms of, e.g., Aβ or α-synuclein triggered ROS production and oxidative cytotoxicity (Liu et al., 2011; Deas et al., 2016). Especially iron was shown to enhance aggregation processes of α-synuclein (Ostrerova-Golts et al., 2000), Aβ (Rottkamp et al., 2001) or Tau (Sayre et al., 2000). How iron enhances protein aggregation is not fully understood, but two distinct mechanisms are considered as relevant. First, the direct binding of iron to amyloidogenic proteins, and second, an indirect iron-mediated

Indirectly, iron also influences α-synuclein on its transcriptional and translational level. α-synuclein acts as a ferrireductase and can induce iron accumulation by overexpression. (B) Iron fosters aggregation of both Aβ und Tau by binding. Whereas Aβ reduces levels of ferritin-bound iron, an overexpression of mitochondrial ferritin reduces Aβ toxicity. APP controls iron efflux and together with iron it affects the Aβ release. Furthermore, there is evidence for both, Aβ-induced iron

(Continued)

#### FIGURE 1 | Continued

fnins-13-00015 January 18, 2019 Time: 17:28 # 4

accumulation and Aβ-induced iron depletion. Whereas iron increases Tau-phosphorylation via CDK5 and GSK3ß pathways, iron-induced oxidative stress reduces Tau-phosphorylation. (C) Iron binds SOD1, inducing oxidative stress and toxicity. Mutations of SOD1 lead to an upregulation of iron metabolism proteins followed by iron influx. Iron is suggested to affect TDP-43 aggregation indirectly via oxidative stress-mediated ROS accumulation. An interaction of iron and TDP-43 has not been objectified so far. (D) PrP operates as a ferrireductase partner of ZIP14 and DMT1 increasing Fe3<sup>+</sup> uptake. Furthermore, PrP-ferritin aggregates induce iron deficiency and an upregulation of total iron, Fe2<sup>+</sup> and iron uptake proteins. Inflammation processes may contribute to iron deficiency. Vice versa, Fe3<sup>+</sup> triggers PrP accumulation within the cell.

process, where the above-mentioned Fenton and Haber-Weiss reaction of Fe2<sup>+</sup> triggers aggregation by ROS production and resulting oxidative stress. An overview on relevant interactions of iron and below-mentioned proteins is shown in **Figure 1**.

#### α-Synuclein

α-Synuclein is a 140 amino acid protein expressed in neuronal cytosol and presynaptic terminals that is thought to participate in vesicle packaging, release and trafficking as well as in membrane remodeling. Furthermore, interactions of α-synuclein with histones as well as with nuclear DNA are suspected, but its concrete function in the nucleus and presynaptic terminals needs to be further investigated (Bendor et al., 2013; Rocha et al., 2018). α-Synuclein is intensively studied in regard to the pathophysiology of PD, since some inherited forms of PD can result from point mutations and from overproduction of α-synuclein through multiplications of the SNCA gene encoding the human α-synuclein protein. In total, six point mutations (A30P, E46K, H50Q, G51D, A53T, and A53E) of the SNCA gene were identified so far leading to a biophysical change of amino acid substitutions (Uchihara and Giasson, 2016). Therefore, models overexpressing α-synuclein or a mutated form are used to understand the role of α-synuclein and its interaction with iron (e.g., Zhu et al., 2016; Carboni et al., 2017).

Analyzing the interaction of iron and α-synuclein, different studies could show that α-synuclein fibrillation can be induced by iron (Uversky et al., 2001; Golts et al., 2002; Kostka et al., 2008). In vitro, using BE-M17 neuroblastoma cells iron had stronger effects on aggregation of A53T and A30P mutant compared to wild-type α-synuclein (Ostrerova-Golts et al., 2000). In vivo, iron treatment showed reduced survival of α-synuclein mutant (A53T, A30P) and α-synuclein wild-type flies compared to w1118 wildtype controls, but only the mutant flies showed a strong motor decline (Zhu et al., 2016). Transmission electron microscopy resolved that ex vivo Fe3<sup>+</sup> addition to wild-type and mutant (A53T, A30P, E46K) α-synuclein generates fibrils, resembling fibril conformations formed without iron incubation, whereas copper addition to mutant α-synuclein led to the formation of amorphous aggregates. These results indicate that the fibril morphology is metal-specific (Bharathi et al., 2007).

Even micromolar concentrations of Fe3<sup>+</sup> increased α-synuclein aggregation and produced larger SDS-resistant oligomers. Since H2O<sup>2</sup> treatment did not lead to the same effects as iron treatment, oxidation of α-synuclein per se cannot be the reason for the oligomerization, showing that trivalent ions play a relevant role (Kostka et al., 2008). However, further in vitro studies showed that Fe2<sup>+</sup> also promoted α-synuclein aggregation, transmission and affected viability of SK-N-SH and SN4741 cells (Li et al., 2011; Xiao et al., 2018). Specifically, in SK-N-SH cells α-synuclein aggregation was increased around the nucleus (Li et al., 2011). Under aerobic conditions in vitro, Fe2<sup>+</sup> treatment caused a polymerization into antiparallel soluble α-synuclein oligomers, whereas under anaerobic conditions both Fe2<sup>+</sup> and Fe3<sup>+</sup> induced parallel ß-sheet aggregates (Abeyawardhane et al., 2018).

How iron influences α-synuclein aggregation is not completely understood. However, α-synuclein has a high metal binding affinity and it is known that both Fe2<sup>+</sup> and Fe3<sup>+</sup> can bind α-synuclein, revealing a binding constant of 1.2 x 10<sup>13</sup> M−<sup>1</sup> for Fe3<sup>+</sup> and 5.8 x 10<sup>3</sup> M−<sup>1</sup> for Fe2<sup>+</sup> (Peng et al., 2010). Fe2<sup>+</sup> binds at the C-terminus, specifically at Asp-121, Asn-122, and Glu-123 (Binolfi et al., 2006) and the binding affinity is increased in phosphorylated (pY125 or pS129) α-synuclein (Lu et al., 2011). Additionally, Fe3<sup>+</sup> can bind α-synuclein, having two binding sites, which are likely at the C-terminus (Davies et al., 2011).

Interestingly, there is a close homology in the 5<sup>0</sup> -UTR of human α-synuclein mRNA to the IRE of the ferritin mRNA, which could explain the regulation of α-synuclein levels by intracellular iron (Friedlich et al., 2007). Fe3<sup>+</sup> was suggested to control the translation of α-synuclein mRNA since the iron chelator deferoxamine (DFO) highly decreased the polysomeassociated endogenous α-synuclein mRNA in HEK293 cells (Febbraro et al., 2012). Furthermore, a knockdown of the IRP in SK-N-SH cells enhanced α-synuclein aggregation by upregulation of α-synuclein transcription, indicating that iron partially controls the aggregation process through the IRE/IRP system (Li et al., 2011). In addition to its direct binding there is thus evidence that iron also influences α-synuclein on a transcriptional and translational level.

Iron also promotes α-synuclein aggregation indirectly by regulating the nuclear transcription factor EB (TFEB), which is a transcriptional regulator of the autophagosome-lysosome pathway. Iron enrichment decreased the TFEB expression and inhibited its nuclear translocation through the activation of the Akt/mTORC1 pathway resulting in increased α-synuclein aggregation in cell lysates by the inhibition of TFEB-mediated autophagosome-lysosome fusion. Furthermore, iron increased α-synuclein cell-to-cell transmission, which was attenuated by TFEB overexpression (Xiao et al., 2018). As an indirect pathway, oxidative stress contributes to the iron-induced aggregation. This was further supported by the finding that supplementation of the antioxidative vitamin E attenuated aggregation in SK-N-SH cells (Li et al., 2011). Additionally, an ex vivo study indicated that oxidative stress enhances α-synuclein aggregation indirectly via oxidation of iron from Fe2<sup>+</sup> to Fe3<sup>+</sup> (Levin et al., 2011).

Vice versa, iron levels were increased by the overexpression of α-synuclein itself in primary midbrain neurons and PC12 cells

which were analyzed with particle induced X-ray emission. Iron accumulation quantified by X-ray fluorescence was specifically observed within the perinuclear regions of PC12 cells (Ortega et al., 2016).

Furthermore, α-synuclein was also found to act as a ferrireductase, reducing Fe3<sup>+</sup> to Fe2<sup>+</sup> (Davies et al., 2011). The ferrireductase active form of α-synuclein is suggested to be a membrane-associated helical-rich tetramer (McDowall et al., 2017; Angelova and Brown, 2018). Previous studies suggest a resistance of the tetramer to fibril and aggregate formation, making the tetramer to an interesting subspecies in vivo (Bartels et al., 2011; Wang W. et al., 2011; Dettmer et al., 2015).

Regarding treatment strategies to reduce or prevent α-synuclein aggregation, iron seems to be a promising target. Iron recycling by Nramp1 (Soe-Lin et al., 2009) was shown to degrade microglial α-synuclein oligomers in vitro and in vivo, highlighting natural defense mechanisms in iron overload conditions (Wu et al., 2017). Furthermore, different iron chelators showed various beneficial effects in preclinical cell or animal PD models (e.g., Sangchot et al., 2002; Mandel et al., 2004; Billings et al., 2016; Finkelstein et al., 2016, 2017; Carboni et al., 2017; Das et al., 2017).

Iron and α-synuclein thus influence each other mutually: whereas iron contributes to α-synuclein aggregation by direct binding and indirectly via oxidative stress and transcription factors, α-synuclein shows ferrireductase activity influencing iron homeostasis.

#### Aβ and Tau

Aβ is a metalloprotein consisting of 39–43 residues that is derived from the transmembrane amyloid precursor protein (APP) by proteolytic cleavage. Aβ aggregates are found as amyloid plaques in AD and Aβ plays a role in metal sequestration and homeostasis, synaptic activity and neuronal plasticity (Smith et al., 2007; Rajasekhar et al., 2015). Aβ was found to be strongly colocalized with brain iron in risk patients for AD measured by magnetic resonance imaging and positron emission tomography (Van Bergen et al., 2016). X-ray studies analyzing Aβ plaques in cortex tissue of transgenic mice and Aβ plaque cores of AD patients confirmed a direct correlation of iron and Aβ localization, suggesting the formation of an iron-Aβ complex (Telling et al., 2017; Everett et al., 2018). Ex vivo, Fe3<sup>+</sup> was found to promote aggregation of Aβ1-40 and Aβ1- 42 visualized by fluorescence spectroscopy and atomic force microscopy (Tahmasebinia and Emadi, 2017). Fe3<sup>+</sup> was also shown to bind Aβ using the phenolic oxygen of tyrosine 10 as binding site (Miura et al., 2001). Fe3+-mediated generation of Aβ aggregates was also reported in vitro. However, these aggregates were shorter and less ordered than in iron-free Aβ incubation (Liu et al., 2011). In drosophila, the affinity of Aβ for iron is mediated by three N-terminal histidines enhancing Aβ dimerization and leading to histidine-dependent oxidative damage (Ott et al., 2015).

Furthermore, treatment with iron chelators (clioquinol, YM-F24) increased survival and locomotor function of flies expressing Aβ1-42 compared to control wild-type flies, highlighting the relevance of oxidative stress for neurotoxicity of Aβ (Rival et al., 2009; Liu et al., 2011). Iron chelation also showed beneficial effects in mammalian models (Fine et al., 2015; Zhao et al., 2017) and reduced Aβ1-42 aggregation in an ex vivo study (Tahmasebinia and Emadi, 2017). Other studies showed that the presence of the important iron storage protein ferritin has beneficial effects on Aβ pathology. Accordingly, in vitro and in vivo, mitochondrial ferritin reduced neurotoxic effects exerted by Aβ (Wu et al., 2013; Wang P. et al., 2017). The overexpression of mitochondrial ferritin in SHSY5Y cells prevented the activation of the MAPK signaling pathway, which is related to oxidative stress-induced cell death (Wu et al., 2013).

Iron also strongly influences APP. Treating SHSY5Y cells with Fe3<sup>+</sup> caused APP accumulation in membrane-enriched cellular fraction and increased activity of β-secretase that both triggered increased release of Aβ1-42 (Banerjee et al., 2014). Another study using SHSY5Y cells confirmed increased APP steady state levels and APP production after iron treatment (Rogers et al., 2016).

Furthermore, APP influences iron export by controlling the persistence of FPN1 on the neuronal surface, even if it does not function as a ferroxidase (Wong et al., 2014). There is evidence that Aβ induces iron accumulation in a cell-free ex vivo study (Everett et al., 2014b). Other studies suggest that Aβ controls the redox-activity of iron and reduces iron chemically in vivo (Everett et al., 2018). Accordingly, Aβ was able to inhibit ascorbatedependent hydroxyl radical generation by free Fe3<sup>+</sup> (Nakamura et al., 2007). Furthermore, interaction with Aβ led to a reduction of ferrihydrite (ferritin-bound iron) to pure redox-active Fe2<sup>+</sup> (Everett et al., 2014a).

Iron was not only shown to modulate the aggregation of Aβ but also of Tau (Kim et al., 2018). Tau is a microtubuleassociated protein that is the main component of neurofibrillary tangles in AD. Mostly, it is located in axons and sometimes in dendrites (Nisbet et al., 2015). So far, little is known about the interplay of iron and Tau and present studies are not consistent. Iron treatment enhanced Tau aggregation in ironenriched hippocampal regions and iron was shown to bind Tau (Sayre et al., 2000). Some studies showed that only trivalent metal ions, as Fe3+, trigger Tau aggregation but not divalent ions (Yamamoto et al., 2002; Bader et al., 2011). Fe3+-generated Tau oligomers were even more stable than DMSO-generated ones (Nübling et al., 2012). However, a recent electrochemical study showed that both, Fe3<sup>+</sup> and Fe2+, bind Tau at different binding sites, inducing a structural change, which was more pronounced with Fe2<sup>+</sup> (Ahmadi et al., 2017).

Iron-induced oxidative stress reduced Tau phosphorylation by interfering with the function of the CDK5/p25 system of hippocampal neurons (Egaña et al., 2003). Another study using primary cultures of rat cortical neurons showed ironinduced Tau phosphorylation by activation of GSK3 (Lovell et al., 2004). Ebselen, an organo-selenium compound with antioxidant activity, inhibits the CDK5 and GSK3β pathway leading to less Tau phosphorylation. These effects were not only caused by the antioxidant effects of Ebselen, but rather by the inhibition of DMT1 in SHSY5Y cells (Xie et al., 2012). The same effects on CDK5 and GSK3β as well as on Tau phosphorylation were shown with the iron chelator DFO analyzing APP/PS1 transgenic mouse brains (Guo et al., 2013).

In summary, iron binds both, Aβ and Tau, partially modifying their structure and fostering their aggregation process. Furthermore, the Aβ and Tau phenotypes can be partially rescued by iron chelation.

#### SOD1 and TDP-43

fnins-13-00015 January 18, 2019 Time: 17:28 # 6

SOD1 is an abundant antioxidant protein predominantly located within the cytosol. Its aggregation leads to, e.g., neuronal degeneration and changes in DNA/RNA metabolism, neurofilament and axonal transport (Pasinelli and Brown, 2006). An NMR study showed binding of iron to SOD1 with binding sites closely located to Cu/Zn binding pockets leading to Fe2+ bound SOD1 complexes, which are likely toxic (Lim and Song, 2015). Even if there is no direct evidence for iron-induced SOD1 aggregation, several studies emphasize a coherence of iron load and SOD1 pathology.

Therapeutic approaches showed that in SOD1 mouse models (G93A, G37R, G86R) iron chelators [VK-28, M30, HLA20, VARced, deferiprone (DFP)] extended lifespan, increased locomotor function and motoneuron survival and decreased oxygen free radicals, iron levels and TfR expression (Jeong et al., 2009; Kupershmidt et al., 2009; Wang Q. et al., 2011; Golko-Perez et al., 2016; Moreau et al., 2018).

Furthermore, several studies observed an impact of SOD1 on iron metabolism. Accordingly, in vivo studies with SOD1.G93A mutant mice showed increased mRNA expression of ferritin, TfR1 and DMT1. These results indicate enhanced iron load, since iron is known to regulate ferritin expression (Jeong et al., 2009; Wang Q. et al., 2011). In vitro studies analyzing cell lysates showed an increased iron content and altered iron metabolism mediated by an impaired Akt signaling pathway. Accordingly, SOD1.G93A overexpression triggered an inactivation of Akt, activation of the transcription factor FOXO3a and subsequently increased ferritin synthesis and iron accumulation (Hadzhieva et al., 2013; Halon-Golabek et al., 2018). SOD1.G37R mutants showed in vivo and in vitro increased mRNA levels of TfR1, ferritin, DMT1 but also of mitochondrial ferritin (Jeong et al., 2009). The above-mentioned results of an increased iron content and increased TfR levels suggest an iron dyshomeostasis, which is not primarily controlled by IRE/IRP1 mechanisms (Lovejoy and Guillemin, 2014). Other studies showed diverse effects of SOD1 on the cytosolic iron sensor IRP1. SOD1 activation in G93A mutant mice showed no changes of IRP1 expression but more activated IRP1 (Jeong et al., 2009; Gajowiak et al., 2016). In vivo, SOD1 deficiency and the resulting oxidative stress caused IRP1 downregulation (Milczarek et al., 2017). Furthermore, in SOD1.G93A mutant mice ROS induction led to an upregulated iron import, triggering oxidative stress (Hadzhieva et al., 2013).

SOD1 is also thought to interact with the ALS-relevant protein TDP-43 (Higashi et al., 2010). Aggregates of the RNA-binding protein TDP-43 are incorporated in ubiquitinated inclusions within the neuronal cytoplasm found in ALS and in syndromes jointly named 'neurodegeneration with brain iron accumulation' (NBIA) (Neumann et al., 2006; Haraguchi et al., 2011). SOD1 was shown to initiate modification and accumulation of TDP-43 (Zeineddine et al., 2017; Jeon et al., 2018). In SOD1 mutant mice, iron chelators reduced TDP-43 aggregation, whereas vehicle-treated animals showed TDP-43 aggregates located in the cytoplasm of motor neurons (Wang Q. et al., 2011). Since oxidative stress-mediated accumulation of ROS fosters the TDP-43 aggregation in vitro (Cohen et al., 2012), iron chelator effects on TDP-43 aggregation suggest an indirect effect of iron by oxidative stress induction.

In conclusion, proteins of the iron metabolism are altered in SOD1 mutants, which partially could explain enhanced iron loading. Iron chelator effects in SOD1 models indicate an impact of iron on SOD1 pathology. Even if iron accumulation is a common feature of ALS (e.g., Moreau et al., 2018), a contribution of iron to aggregation of SOD1 or TDP-43 is yet unproven.

# Prion Protein

The prion protein (PrP) is located intracellularly and is also an important membrane-bound protein at the cell surface. Therefore, PrP is involved in exocytotic and endocytic synaptic vesicle processing as well as in signaling pathways, but also in myelination and neurogenesis (Liebert et al., 2014; Legname, 2017; Watts et al., 2018).

By using PrP knockout or PrP-overexpressing mouse models, studies showed that upregulation or downregulation of PrP levels, respectively, affect the iron homeostasis (Singh et al., 2009a,b; Pushie et al., 2011; Ashok et al., 2018). Accordingly, iron deficiency is a crucial characteristic in brains of humans, hamster and mice affected with prion pathology. Analyzing pathological brain tissue as well as scrapie-infected ScN2a and SMB cells, iron dyshomeostasis was supposedly caused by the iron sequestration in detergent-insoluble PrP-scrapie-ferritin aggregates, resulting in a decreased bio-available iron pool and a state of cellular iron deficiency. These results also explain increased amounts of total iron and Fe2<sup>+</sup> as well as iron uptake proteins (Singh et al., 2009a). Another study investigated PrP-mediated iron deficiency in retinas of scrapie-injected hamsters and demonstrated an accumulation of detergent-insoluble ferritin. Furthermore, they showed a correlation of the ferritin accumulation with microglia activity. These results suggested a contribution of chronic inflammation as a side effect of PrP accumulation to functional iron deficiency (Asthana et al., 2017).

Another explanation for the influence of PrP on the iron homeostasis is that PrP operates as a plasma membrane ferrireductase. Accordingly, PrP-expressing neuroblastoma cells showed a significant increase in ferrireductase activity compared to non-transfected cells. Furthermore, mutant PrP forms showed that for an optimal ferrireductase function of PrP the presence of NADH, the copper binding octa-peptide repeat region and the linkage to the plasma membrane is needed (Singh et al., 2013). Also in HepG2-cells, expressing PrP, ferrireductase activity was indicated by an increased uptake of Fe3<sup>+</sup> but not of Fe2+. Since Fe3<sup>+</sup> uptake was significantly increased after a co-expression of PrP with metal transporter ZIP14 and DMT1, PrP is suggested to be the ferrireductase partner of both metal transporter (Tripathi et al., 2015).

So far, there is not much evidence for the influence of iron on the PrP pathology. One study using PrP-deficient HpL3-4 cells showed that Fe3<sup>+</sup> but not Fe2<sup>+</sup> induced accumulation of internalized PrP after iron exposure and PrP treatment of cells (Choi et al., 2013). However, it is not clarified if a direct binding to the PrP protein or indirect mechanisms triggered by iron foster the aggregation of PrP.

Vice versa, PrP was shown to influence the iron homeostasis via its ferrireductase activity and PrP-ferritin aggregates but it needs further investigation to explain if these two modes of action or indirect mechanisms such as the activation of inflammation processes lead to iron deficiency.

#### CLINICAL APPLICATION

Iron accumulation in NDDs argues for a clinical evaluation of iron chelators as symptomatic or neuroprotective agents. Iron chelation by DFO, DFP or deferasirox (DFX) is clinically approved by the FDA for systemic conditions like acute iron intoxication and chronic iron overload<sup>1</sup> . DFO application is problematic due to the required continuous injection (short plasma half-life period) and dose-dependent neurotoxicity. DFX and DFP can be administered orally. DFX must be especially monitored for renal failure and DFP for agranulocytosis and neutropenia (Mobarra et al., 2016). With its ability to relocate iron and to cross the blood brain barrier, DFP presents the most promising candidate for targeting iron load in the CNS, being tested in NDDs with regional iron overload, e.g., Friedreich ataxia (e.g., Velasco-Sánchez et al., 2011; Pandolfo et al., 2014) and NBIA (e.g., Lim et al., 2018; Rohani et al., 2018).

In PD patients (FAIR-PARK-I, NCT00943748), DFP treatment showed benefits in motor performance and reduced MRquantified substantia nigra iron content (Devos et al., 2014). In another trial (DeferipronPD, NCT01539837), administration of a lower dosage resulted in a reduced T2<sup>∗</sup> MRI iron content in the dentate and caudate nucleus (Martin-Bastida et al., 2017). Based on these results, a European multicenter, parallel-group, placebocontrolled, randomized phase III trial is ongoing to assess disease-modifying effects of DFP in PD patients (FAIRPARKII, NCT02655315).

In a pilot trial (SAFEFAIRALS, NCT02164253), DFP treatment of ALS patients reduced R2<sup>∗</sup> iron content in the cervical spinal cord, medulla oblongata and motor cortex. Additionally, patients showed a smaller decrease in the ALS function rating scale and in the body mass index in the first 3 months of treatment compared to the first treatment-free period (Moreau et al., 2018). Based on these results, a phase II/phase III study with a larger sample size has been initiated testing DFP (FAIR-ALSII, NCT03293069).

In AD, intramuscular application of DFO significantly reduced the decline of daily living skills (Crapper McLachlan et al., 1991). Besides, cognition of AD patients benefited from treatment with the metal-attenuating compound clioquinol or its derivate (Ritchie et al., 2003; Lannfelt et al., 2008). Now, AD

<sup>1</sup> www.fda.gov

patients are recruited for an ongoing phase II study investigating effects of DFP (The 3D Study, NCT03234686).

Details of the mentioned studies can be found in **Supplementary Table 1**. Summing up, iron chelators already showed promising results in smaller trials and this advocates for further assessment in larger studies.

# CONCLUSION

Current investigations focus on iron accumulation and protein aggregation as two pathological hallmarks of multiple NDDs. The interaction of both features appears to be an essential part of common neurodegeneration mechanisms. So far, iron was found to bind amyloidogenic proteins like α-synuclein, Aβ and Tau fostering their aggregation. Indirectly, iron influences diseaserelated proteins via oxidative stress or manipulation of their transcription and translation. On the other hand, α-synuclein, Aβ and SOD1 were found to affect iron metabolism and iron accumulation, proposing a mutual interplay and an amplification of their detrimental effects in the disease state. Even if precise effects of iron on neurodegenerative pathomechanisms remain incompletely understood, translational trials in human patients already showed beneficial effects of iron chelation as a treatment strategy for NDDs. Thus, iron may be a substantial contributor to neurodegeneration and merits further investigation as molecular and therapeutic target.

# AUTHOR CONTRIBUTIONS

KJ developed the idea under the lead of PL, performed literature research, wrote and finalized the manuscript, and prepared the figure. A-ER and FM were involved in literature research and writing of the manuscript. PL developed the idea for this review and revised the manuscript. All authors have seen and approved the final version.

# FUNDING

We acknowledge support by the German Research Foundation and the Open Access Publication Funds of the Göttingen University.

# ACKNOWLEDGMENTS

The support of the SFB 1286 "Quantitative Synaptology" was greatly acknowledged.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.00015/full#supplementary-material

#### REFERENCES

fnins-13-00015 January 18, 2019 Time: 17:28 # 8



model of amyotrophic lateral sclerosis. FASEB J. 23, 3766–3779. doi: 10.1096/fj. 09-130047


activity of copper and iron. Biochemistry 46, 12737–12743. doi: 10.1021/ bi701079z


mitochondrial aggregation and α-synuclein translocation in SK-N-SH cells in culture. Dev. Neurosci. 24, 143–153. doi: 10.1159/000065700



**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 Joppe, Roser, Maass and Lingor. 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.

# Oxidative Stress Regulated Iron Regulatory Protein IRP2 Through FBXL5-Mediated Ubiquitination-Proteasome Way in SH-SY5Y Cells

#### Qian Jiao, Xixun Du, Jie Wei, Yong Li and Hong Jiang\*

Department of Physiology, Shandong Provincial Key Laboratory of Pathogenesis and Prevention of Neurological Disorders and State Key Disciplines: Physiology, School of Basic Medicine, Qingdao University, Qingdao, China

Iron regulatory protein 2 (IRP2) plays a key role in the cellular iron homeostasis and could be regulated by a variety of factors, such as oxidative stress, hypoxia and iron, etc. IRP2 depletion results in neurodegenerative movement disorder with the loss of neurons and accumulations of iron. Since oxidative stress extensively exists in several neurodegenerative diseases where iron accumulation also exists, it is important to clarify the mechanisms underlying the effects of oxidative stress on IRP2 expression and its consequence. 200 and 300 µM H2O<sup>2</sup> could result in the reduced cell viability in SH-SY5Y cells. The intracellular levels of reactive oxygen species (ROS) were increased by 52.2 and 87.3% with 200 and 300 µM H2O<sup>2</sup> treatments, respectively. The decreased levels of mitochondrial transmembrane potential (19m) were only observed in 300 µM H2O2-treated group. The protein levels of IRP2, but not for its mRNA levels, were observed decreased in both groups, which resulted in the lower TfR1 expression and decreased iron uptake in these cells. Pretreatment with MG132, the decreased IRP2 levels caused by H2O<sup>2</sup> treatment could be antagonized. The protein levels of F box and leucine-rich repeat protein 5 (FBXL5), the only E3 ligase of IRP2, were observed decreased accordingly. When knockdown the intracellular FBXL5 levels by si-FBXL5, the protein levels of IRP2 were found increased with H2O<sup>2</sup> treatment. Our results suggest that FBXL5 is involved in the degradation of IRP2 under oxidative stress in dopaminergic-like neuroblastoma cells, which implies that its role in the neuronal regulation of IRP2 in neurodegenerative diseases.

Keywords: oxidative stress, IRP2, FBXL5, iron metabolism, ubiquitination

# INTRODUCTION

Oxidative stress occurs when the accumulated production of reactive oxygen species (ROS) and the decrease of antioxidant activity. It is thought to be one of common underlying mechanisms that lead to cellular dysfunction in neurodegenerative disease (Pong, 2003; Dias et al., 2013). As such, the substantia nigra (SN) exhibits increased levels of oxidized lipids, damaged proteins, and DNA in Parkinson's disease (PD) (Hwang, 2013). Moreover, iron levels in the SN have been reported to be elevated in patients with PD (Sofic et al., 1988; Dexter et al., 1989) and PD animal models (Wang et al., 2004; Jiang et al., 2010; You et al., 2015). Markers of oxidative stress also increase in both Alzheimer's disease (AD) patients models brain

#### Edited by:

Massimiliano Filosto, Asst degli Spedali Civili di Brescia, Italy

#### Reviewed by:

Philipp Janker Kahle, Hertie-Institut für klinische Hirnforschung (HIH), Germany Xuping Li, Houston Methodist Research Institute, United States

> \*Correspondence: Hong Jiang hongjiang@qdu.edu.cn

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 21 May 2018 Accepted: 10 January 2019 Published: 29 January 2019

#### Citation:

Jiao Q, Du X, Wei J, Li Y and Jiang H (2019) Oxidative Stress Regulated Iron Regulatory Protein IRP2 Through FBXL5-Mediated Ubiquitination-Proteasome Way in SH-SY5Y Cells. Front. Neurosci. 13:20. doi: 10.3389/fnins.2019.00020

**166**

areas in which amyloid β (Aβ) is abundant (Butterfield et al., 2001). Aβ plaques can efficiently generate ROS and accelerate iron accumulation (Smith et al., 2007; van Duijn et al., 2017). Thus, the oxidative stress may play an important role in inducing iron metabolic disorder in neurodegeneration associated with several brain pathologies.

Iron homeostasis is regulated by coordination proteins that are responsible for iron uptake, storage, exports, and utilization in the cellular (Hentze et al., 2010). Iron regulatory protein(IRPs), including IRP1 and IRP2, are RNA-binding proteins that interact with RNA stem loops known as iron responsive element (IRE) to regulate the translation and stability of mRNAs that encode proteins required for iron homeostasis, including ferritin, ferroportin (Fpn), transferrin receptor (TfR) and divalent metal transporter 1 (DMT1) (Rouault, 2006; Muckenthaler et al., 2008). It has been reported that a syndrome of progressive neurodegenerative disease and anemia develops with iron deposits in white matter and a loss of Purkinje cells in IRP2−/<sup>−</sup> mice, whereas IRP1−/<sup>−</sup> mice develop polycythemia and pulmonary hypertension, indicating a prominent role of IRP2 in controlling neuronal iron metabolism (Meyron-Holtz et al., 2004; Wilkinson and Pantopoulos, 2014). But some studies considered that IRP2 deficiency without symptomatic neurodegeneration in the mouse (Galy et al., 2006). Our previous study showed that IRP2 expression is decreased in 1-Methyl-4-phenylpyridinium [MPP(+)]-induced cellular model of PD (Zhang et al., 2009), whereas IRP2 expression is increased in 6-hydroxydopamine (6- OHDA)-treated PD model (Jiang et al., 2010). In AD patients, IRP2 immunoreactivity is present in the intracellular neurofibrillary, including neurofibrillary tangles and neuropil threads, which is striking differences from the control brain (Smith et al., 1998). These suggest that IRP2 expression disruption exists in neurodegenerative diseases where oxidative stress extensively occurs. In peripheral system, such as liver and kidney, IRP2 is rapidly degraded through ubiquitin-proteasomal system in high iron and high oxygen pressure conditions (Kuhn, 2015). Additionally, hypoxia and low iron level can prevent IRP2 degradation. However, in nerve system, the underlying mechanisms of oxidative stress regulate IRP2 expression and the homeostasis of iron is not fully understood.

Iron regulatory protein 2 has unique 73–amino acid that is different from IRP1 (Iwai et al., 1995). Deletion of this sequence eliminates the rapid turnover of IRP2, and does not show a sensitivity to F box and leucine-rich repeat protein 5 (FBXL5) (Iwai et al., 1995; Salahudeen et al., 2009). FBXL5, an iron- and oxygen-regulated SCF-type ubiquitin ligase (E3), has been shown to contribute to iron-dependent degradation of IRP2 in liver or human embryonic kidney cells (Salahudeen et al., 2009; Vashisht et al., 2009; Moroishi et al., 2011). Here, we used SH-SY5Y cells, a dopaminergic-like neuroblastoma cell, to investigate the regulation mechanism of IRP2 under oxidative stress condition in nerve system. We found that oxidative stress induced by H2O<sup>2</sup> reduced IRP2 protein levels through ubiquitination-proteasome way. Our results also revealed that E3 FBXL5 plays a pivotal role in the regulation of IRP2 protein levels under oxidative stress condition in nerve system.

# MATERIALS AND METHODS

# Cell Culture and Treatment

The SH-SY5Y cell line was purchased from the Cell Bank of the Shanghai Institute of Cell Biology and Biochemistry, Chinese Academy of Sciences (Shanghai, China). SH-SY5Y cells were cultured in DMEM/F12 (Dulbecco's modified Eagle medium and Ham's F12, 1:1, pH 7.4) with 10% fetal bovine serum (FBS), 100 U/mL of penicillin and 100 U/mL of streptomycin (all from Invitrogen, United States) at 37◦C in a humidified atmosphere containing 5% CO2. For experiments, cells were seeded in 6 well plates with 1 × 10<sup>5</sup> cells/mL and grown to 80–90% confluency before they were treated with H2O<sup>2</sup> (20, 30, 50, 200, 300, 500, and 1000 µmol/L) for 24 h. After 24 h, the medium was replaced with medium containing H2O2, and cells were treated for another 24 h and then harvested for experiments.

# Transfection of shRNA

The FBXL5 short hairpin RNA (shRNA) and the scramble shRNA control were pre-designed and purchased from Genechem (Shanghai, CN). SH-SY5Y were transfected using 1-µg plasmid and 2.5-µL Lipofectamine 2000 reagent (Invitrogen, United States) in 800-µL transfection medium per well in 12 well plate for 24 h optimal transfection and then used for the following experiments.

# Cell Viability

The cell viability was detected by the conventional 3-(4,5 dimethylthiazol-2-yl) -2, 5-diphenyltetrazolium bromide (MTT) (Beyotime, CN) assay. The MTT assay is a colorimetric assay for measuring the activity of cellular enzymes that reduce the tetrazolium dye. Cells were incubated in MTT (5 mg/mL) for 3–4 h. 100 µL DMSO was added to each well after medium was removed. The formazan dye crystals were solubilized for 10 min, and absorbance was measured at 494 and 630 nm with a spectrophotometer (Molecular Device, M5, United States).

## Measurement of the Mitochondrial Transmembrane Potential (19m) and the Production of Intracellular Reactive Oxygen Species (ROS)

Changes in the levels of intracellular ROS and 19m with H2O<sup>2</sup> (200 and 300 µmol/L) treatment of SH-SY5Y were measured using flow cytometry with Rhodamine 123 (Sigma, United States) or carboxy-H2DCFDA dye (Invitrogen, United Kingdom) as previously described (Jiang et al., 2010). After cells were washed with HEPES buffered saline (HBS, 10 mM of HEPES, 150 mM of NaCl, pH 7.4) three times, carboxy-H2DCFDA (5 µM) was added and incubated for 30 min at 37◦C. Cells were washed with HBS two times, followed by centrifugation at 800 rpm for 5 min, and re-suspended in 1 mL HBS. Intracellular ROS generation was also assayed by Rhodamine 123 (5 µg/mL). Fluorescent intensity was recorded at 488 nm excitation and 525 nm emission wavelengths (Fluorescence 1, FL1). Results were demonstrated as FL1-H (Fluorescence 1-Histogram), setting the gated region M1

and M2 as a marker to observe the changing levels of fluorescence intensity using Cellquest Software.

# Calcein Loading of Cells and Ferrous Iron Influx Assay

Ferrous iron influx were detected as previously described (Zhang et al., 2009; Du et al., 2016). Calcein-AM (Molecular Probes, United States) is a membrane-permeative, and forms fluorescent calcein by cytoplasmic esterases upon intracellular cleavage. This reaction is pH independent and can be quenched rapidly by divalent metals and reversed easily by chelators. The cells were planted in glass coverslips and incubated with calcein-AM (0.5 µM) in HBS at 37◦C for 30 min. The excess calcein on the cell surface was washed for three times with HBS. Calcein fluorescence was recorded at an excitation wavelength of 488 nm and an emission wavelength of 525 nm and the fluorescence intensity was measured every 3 min for 30 min with continuous perfusion of 100 µM of ferrous iron (ferrous sulfate in an ascorbic acid solution, 1:44 molar ratio; prepared immediately before the experiments). Ascorbic acid maintained the reduced status of ferrous iron, in addition, ascorbate acted as a chelator to maintain the iron in solution. The mean fluorescence signal of 25–30 single cells in four separate fields was monitored at 200× magnification and processed with Fluoview 5.0 software.

#### Quantitative Real-Time PCR

SH-SY5Y cells were collected for the investigation of FBXL5 and IRP2 mRNA expression by quantitative real-time PCR (qRT-PCR). Total RNA was extracted by using the Trizol reagent (Invitrogen) in accordance with manufacturer's protocol and was quantified by spectrophotometry (Bio-Rad, United States). RNAs were reversely transcribed to cDNA by a reverse transcriptase kit (RevertAid First Strand Cdna Synthesis Kit, Thermo, United States). Relative abundance of each mRNA was quantified by qRT-PCR using specific primers and the SYBR Premix Ex TaqII (TaKaRa, CN). Primers for rat FBXL5 (forward 5<sup>0</sup> - TTAACTAACAAGGGCATTGGAGAAG -3 0 ; reverse 5 0 - TCAGCCAAATCTTCAGCATCTAAC -3<sup>0</sup> ), IRP2 (forward 5<sup>0</sup> -CGCCTTTGAGTACCTTATTGAAACA-3 0 ; reverse 5 0 -CGTACAGCAGCTTCCA ACAAGA-3<sup>0</sup> ) and GAPDH (forward 5<sup>0</sup> -GCACCGTCAAGGCTGAGAAC-3<sup>0</sup> ; reverse 5 0 -TGGTGAAGACGCCAGTGGA-3<sup>0</sup> ) were synthesized by TaKaRa. QRT-PCR reactions were carried out by using Real-Time PCR Detection System (Eppendorf,GER). Data were analyzed by the 211CT method, and GAPDH was taken as an internal control.

## Western Blot Analysis

The SH-SY5Y cells were washed with ice-cold PBS and lysed in lysis buffer (Cwbio, CN) containing protease inhibitors cocktail (Rhoche, GER) for 30 min. The lysates were centrifuged at 12,000 × g for 10 min, and the protein concentration of the supernatants was determined with a PierceTM BCA Protein Assay Kit (Thermo, United States). A total of 20 µg of protein was electrophoresed on 8% SDS polyacrylamide gels and transferred onto PVDF membranes (300 mA, 90 min). After blocking with 10% non-fat milk for 1 h at room temperature, the membranes were incubated overnight at 4◦C with rabbit anti-FBXL5 (1:1000, Abcam, United Kingdom), rabbit anti-IRP2 (1:1000, Abcam, United Kingdom), rabbit anti-TfR1 (1:1000, Abcam, United Kingdom ) and rabbit anti-β -actin monoclonal antibody (1:10000, BIOS, CN). The cross-reactivity was visualized using ECL Kit (Millopore, United States) and analyzed through scanning densitometry with a UVP image system.

FIGURE 1 | The oxidative effects of different concentrations of H2O<sup>2</sup> on SH-SY5Y cells. (A) Cell viability was changed with the increased concentration of H2O<sup>2</sup> treatment. Cell viability of the 0 µM H2O<sup>2</sup> group was set to 100%. Data were presented as mean ± SEM of 6 independent experiments. <sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 compared with 0 µM H2O<sup>2</sup> group. (B) Fluorometric assay on 19m in different groups (a) and statistical analysis (b). 19m was decreased with the increased concentration of H2O2. Fluorescence values of the 0 µM H2O<sup>2</sup> was set to 0. Data were presented as mean ± SEM of 6 independent experiments. ∗∗∗P < 0.001 compared with H2O<sup>2</sup> 0 µM group. (C) Fluorometric assay on ROS levels in different groups (a) and statistical analysis (b). Fluorescence values of the 0 µM H2O<sup>2</sup> group was set to 100%. Data were presented as mean ± SEM of 4 independent experiments. ∗∗∗P < 0.001 compared with 0 µM H2O<sup>2</sup> group.

ANOVA, data were presented as mean ± SEM of 3 independent experiments.∗P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 compared with H2O<sup>2</sup> 0 µM group. (C) Calcein-indicated ferrous iron (FeSO4) influx in SH-SY5Y cells. The fluorescence intensity of SH-SY5Y cells in 200 µM group and 300 µM group were significantly higher than that in 0 µM H2O<sup>2</sup> group. Two-way ANOVA, <sup>∗</sup>P < 0.05. The mean fluorescence intensity of 35 separate cells from 4 separate fields at each time point was presented as mean ± SEM of 6 independent experiments. (D) TfR1 protein levels were changed with H2O<sup>2</sup> treatment. One-way ANOVA, data were presented as mean ± SEM of 3 independent experiments.∗P < 0.05 and ∗∗P < 0.01compared with H2O<sup>2</sup> 0 µM group.

## Statistical Analysis

Results are presented as means ± SEM. One-way analysis of variance (ANOVA) followed by the Tukey's Multiple Comparison Test was used to compare difference between means in more than two groups. The iron uptake experiment was carried out using two-way ANOVA followed by the Student-Newman-Keuls test. A probability of P < 0.05 was taken to indicate statistical significance.

#### RESULTS

## The Oxidative Effects of H2O<sup>2</sup> on SH-SY5Y Cells

To elucidate the oxidative effects of H2O2, we assessed cell viability, 19m and ROS in the present study. SH-SY5Y cells were treated with different concentrations of H2O<sup>2</sup> for 24 h and then MTT method was used to detect the cell viability. The results showed that the cell viability was decreased with the increased concentration of H2O<sup>2</sup> (**Figure 1A**). In comparing with the 0 µM H2O<sup>2</sup> group, the cell viability in 20, 30, 50, 200, 300, 500 and 1000 µM H2O<sup>2</sup> group was decreased by 10.2, 12.9, 13.7, 15.3, 21.2, 33.6, and 38.6%, respectively, which displayed significant differences (**Figure 1A**, P < 0.05). The 200 and 300 µM effective concentration of H2O<sup>2</sup> were applied for the following studies. Oxidative stress induced 19m reduction and excessive generation of ROS which contribute to DNA or RNA damage. We found that the 19m in cells was decreased significantly in 300 µM H2O<sup>2</sup> treatment group (**Figure 1B**, P < 0.001). The intracellular ROS levels were increased by 52.2 and 87.3%, respectively, and the difference was statistically significant (P < 0.001) when compared with the 0 µM H2O<sup>2</sup> group

(**Figure 1C**). Thus, these observations indicated that appropriate concentrations of H2O<sup>2</sup> could induce oxidative stress in SH-SY5Y cells.

# H2O<sup>2</sup> Induced a Reduction in IRP2 Expression and Ferrous Iron Uptake

To test the relationship of IRP2 expression and oxidative stress induced by H2O2, we examined the expression of IRP2 in mRNA and protein levels. After SH-SY5Y cells were treated with 200 µM or 300 µM of H2O<sup>2</sup> for 24 h, the mRNA levels of IRP2 showed that no significant difference (**Figure 2A**, P > 0.05), whereas the protein levels of IRP2 were decreased in comparison with the 0 µM H2O<sup>2</sup> group (**Figure 2B**, P < 0.01). Thus, H2O<sup>2</sup> regulated IRP2 expression in protein levels. The SH-SY5Y cells were incubated with 100 µM Fe2<sup>+</sup> to detect the ability of cell iron uptake. Cell iron uptake ability was observed using laser confocal microscopy, and the results showed that the intracellular fluorescence intensity was increased with time, indicating a decreased Fe2<sup>+</sup> uptake in both 200 µM and 300 µM H2O<sup>2</sup> treatment groups (**Figure 2C**, P < 0.05). Moreover, the expression of TfR1in 200 and 300 µM H2O<sup>2</sup> treatment groups were also decreased (**Figure 2D**, P < 0.05). These results suggested that H2O<sup>2</sup> treatment resulted in a decrease of IRP2 protein levels rather than mRNA levels, leading to downregulation of IRP2 targets and iron uptake ability.

# The Down-Regulative Effects of H2O<sup>2</sup> on IRP2 Protein Levels Through Ubiquitination Pathway

We next examined how such H2O<sup>2</sup> affected IRP2 protein levels. To determine the expression of IRP2 that induced by H2O<sup>2</sup> through ubiquitination pathway, we incubated SH-SY5Y cells with the ubiquitin proteasome inhibitor MG132. **Figure 3** showed that the expression levels of IRP2 levels were decreased in H2O<sup>2</sup> treatment groups and were restored by MG132 treatment, the difference was statistically significant (P < 0.05).

# FBXL5 Was Involved in the Degradation of IRP2 Induced by H2O<sup>2</sup>

F box and leucine-rich repeat protein 5, an E3 ubiquitin ligase subunit, is a key event to mediate IRP2 degradation for controlling iron homeostasis (Moroishi et al., 2011). To test our hypothesis that IRP2 reduction was raised by FBXL5 in the oxidative environment, we measured the FBXL5 expression both in mRNA and in protein levels. Results showed that the mRNA levels of FBXL5 showed no significant difference (**Figure 4A**, P > 0.05), whereas the protein levels of FBXL5 were increased in both H2O<sup>2</sup> treatment groups (**Figure 4B**, P < 0.05). These data indicated that oxidative factors H2O<sup>2</sup> could cause the decrease of IRP2 and the increase of FBXL5.

Sh-FBXL5 were transfected into SH-SY5Y cells, and we found that FBXL5 levels were decreased in all H2O<sup>2</sup> treatment groups (**Figures 4Ca,b**, P < 0.05), and IRP2 levels were increased in all groups (**Figures 4Ca,c**, P < 0.05), the difference was statistically significant in comparing with negative control. The

results suggested that H2O<sup>2</sup> could down regulate IRP2 through FBXL5 mediated ubiquitination pathway.

## DISCUSSION

H2O<sup>2</sup> group. T-test,#P < 0.05 and ##P < 0.01.

We have shown that the oxidative stress induced by H2O<sup>2</sup> in SH-SY5Y cells resulted in a down-regulation of IRP2 protein levels, which in turn leads to a decrease of iron uptake ability. The IRP2 protein stability was regulated by ubiquitination pathway. FBXL5, the E3 of IRP2, was up-regulated after treatment of H2O2. The knockdown of FBXL5 could attenuate the IRP2 down-regulation that induced by H2O2. Our findings indicate that H2O<sup>2</sup> regulated IRP2 expression through FBXL5 mediated ubiquitination pathway in SH-SY5Y cells.

Molecules or molecular fragments containing one or more unpaired electrons are called free radicals. These free radicals in living systems are mostly derived from oxygen (Jomova et al., 2010). The ROS, such as (O<sup>2</sup> <sup>−</sup>), derived from metabolic processes, and then further interacts with other molecules via enzyme- or metal-catalyzed to generate "secondary" ROS (Jomova et al., 2010; Farina et al., 2013). Excessive exposure to ROS may induce oxidative stress particularly in mitochondria, and lead to neurodegenerative diseases. H2O<sup>2</sup> generates oxygen by catalase, which lead to hyperoxia. And the free intracellular Fe2+and H2O<sup>2</sup> participate in the Fenton reaction to generate high reactive hydroxyl radical. Thus, in the present study, we used H2O<sup>2</sup> mimic oxidative stress (**Figure 1**). The cell viability was reduced with the increased H2O<sup>2</sup> concentrations (**Figure 1A**). The 19m reduction and excessive generation of ROS increased both in H2O<sup>2</sup> (**Figures 1B,C**) treatment groups. These results indicated that cells were damaged by the artificial oxidative stress in vitro.

In the present study, we found that H2O<sup>2</sup> reduced the expression of IRP2 just in protein levels rather than mRNA levels in the dopaminergic-like neuroblastoma cells (**Figures 2A,B**).

T-test,#P < 0.05, ##P < 0.01, and ###P < 0.001.

Correspondingly, TfR1 expression and cell iron uptake ability were reduced. As a key regulator of brain iron metabolism, IRP2 expression is accurately modulated by the microenvironment factors, including oxygen, iron, free radicals, and cytokines. Generally, neurodegenerative diseases have high oxidative stress level (Emerit et al., 2004; Kim et al., 2015). For example, oxidative stress is especially substantiated by the complex I mitochondrial dysfunction with increased ROS production that might induced by accumulated α-synuclein and iron in the SN of PD patients (Jomova et al., 2010; Farina et al., 2013; Hwang, 2013). In our pervious study, we found that the expression of IRP2 mRNA levels was not affected by FAC at concentration of 100 µM (Zhang et al., 2009). Moreover, sodium nitroprusside, a NO donor and an oxidative generator, could inhibit the IRP2 protein expression (Wei et al., 2017). The mechanism underlying oxidative stress-mediated down-regulation of IRP2 is complex. IRP2 mRNA transcription is regulated by HIFα of transcription factors that affected by oxygen and iron (Kaelin and Ratcliffe, 2008). In post-transcriptional regulation, IRP2 was recognized by ubiquitin ligase and then degraded by the proteasome in kidney cell line (Salahudeen et al., 2009; Takahashi-Makise et al., 2009; Vashisht et al., 2009). However, the mechanisms of oxidative stress affect IRP2 expression in neurodegenerative diseases are largely unknown. We found that MG132 treatment could prevent the reduction of IRP2 that induced by H2O<sup>2</sup> (**Figure 3**), suggesting that oxidative stress regulated IRP2 expression in dopaminergic-like neuroblastoma cells might be achieved through ubiquitination pathway.

There are two E3 ubiquitin ligases for IRP2, including FBXL5 and haem-oxidized IRP2 ubiquitin ligase 1 (HOIL-1). HOIL-1 belongs to RING finger protein and recognizes IRP2 through a signal created by heme-mediated oxidative modification of the protein (Iwai et al., 2005). However, HOIL-1 is independent and is not required for iron-dependent degradation of IRP2 (Zumbrennen et al., 2008). The selective iron- and oxygen-dependent degradation of IRP2 is mediated

by the Skp1/Cul1/Fbox (SCF) E3 ubiquitin ligase complex containing FBXL5 (Salahudeen et al., 2009; Vashisht et al., 2009). Disruption of the FBXL5 expression are failed to sense the increased cellular iron availability, which results in constitutional accumulation of IRP2 and the disordered expression of its target genes (Salahudeen et al., 2009; Vashisht et al., 2009). In FBXL5 conditional deleted mice, hematopoietic stem cells are cellular iron overload and reduced in cell number (Muto et al., 2017). It is also found that FBXL5−/<sup>−</sup> mice die for overwhelming accumulation of oxidative stress during embryogenesis (Moroishi et al., 2011). Here, we found that FBXL5 increased and its targets IRP2 reduced when treated by H2O<sup>2</sup> in SH-SY5Y cells. Thus, FBXL5 might be also referred to regulating IRP2 expression under oxidative stress condition in nerve system.

F box and leucine-rich repeat protein 5 contains a hemerythrin domain in the N-terminus that belongs to a family of iron- and oxygen-binding proteins, and contains the leucine-rich repeats in C-terminal region that binds to IRP2 (Salahudeen et al., 2009; Chollangi et al., 2012; Ruiz and Bruick, 2014). The N-terminal hemerythrin domain has iron and oxygen sensing properties for binding with Fe-O-Fe center. The Fe-O-Fe center is formed in iron replete and oxygenated cells. FBXL5 is stable when directly binds to iron in the hemerythrin domain, whereas unstable under iron-deficient conditions. With this iron-sensing ability, FBXL5 controls the abundance of IRP2 in an iron-dependent manner, and degrades IRP2 by the stabilized FBXL5 under iron-replete conditions. In addition to iron bioavailability, IRP stability and activity are regulated by oxygen. Acting as ROS sensors, the hemerythrin domain could also facilitate regulation by ROS that stabilizes FBXL5 and antagonizes the stabilization

#### REFERENCES


of IRP2 (Ruiz and Bruick, 2014). In our present study, oxidative stress generators H2O<sup>2</sup> could up-regulate FBXL5 expression (**Figures 4A,B**) and using si-FBXL5 could alleviate the reduction of IRP2 protein levels (**Figure 4C**), suggesting that the close relationship of FBXL5 and oxidative stress in regulation nerve system cellular iron metabolism.

In conclusion, we present novel evidence that the oxidative generators H2O<sup>2</sup> induced an increase of FBXL5-mediated ubiquitination degradation in dopaminergic-like neuroblastoma cells. The results of this study implicate an important role of oxidative stress in regulating iron metabolism in nerve system.

#### AUTHOR CONTRIBUTIONS

HJ and QJ conceived the project and designed the study. QJ, JW, and XD performed the experiments, analyzed the data, and interpreted the results. QJ wrote the manuscript. HJ and YL reviewed and edited the manuscript. All authors have read and approved the final version of the manuscript.

#### FUNDING

This work was supported by the NSFC (31771110, 31500837, 31471114, and 81430024), National Key Research and Development Program of China (2016YFC1306500), the Key Research and Development Program of Shandong Province (2016GSF201053), Qingdao Municipal Science and Technology Project (16-2-2-nsh) and Taishan Scholars Construction Project.



**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 Jiao, Du, Wei, Li and Jiang. 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.

# Oxidative Stress Related to Iron Metabolism in Relapsing Remitting Multiple Sclerosis Patients With Low Disability

Mariacristina Siotto<sup>1</sup> \*, Maria Maddalena Filippi<sup>2</sup> , Ilaria Simonelli<sup>3</sup> , Doriana Landi<sup>4</sup> , Anna Ghazaryan2,5, Stefano Vollaro<sup>5</sup> , Mariacarla Ventriglia<sup>2</sup> , Patrizio Pasqualetti<sup>3</sup> , Mauro Ciro Antonio Rongioletti<sup>6</sup> , Rosanna Squitti<sup>7</sup> and Fabrizio Vernieri<sup>5</sup>

1 IRCCS Fondazione Don Carlo Gnocchi, Milan, Italy, <sup>2</sup> Fatebenefratelli Foundation for Health Research and Education, AFaR Division, Rome, Italy, <sup>3</sup> Service of Medical Statistics and Information Technology, Fatebenefratelli Foundation for Health Research and Education, AFaR Division, Rome, Italy, <sup>4</sup> Multiple Sclerosis Clinical and Research Unit, Department of Systems Medicine, Tor Vergata University, Rome, Italy, <sup>5</sup> Neurology Unit, Campus Bio-Medico University of Rome, Rome, Italy, <sup>6</sup> Department of Laboratory Medicine, Research and Development Division, "San Giovanni Calibita", Fatebenefratelli Hospital, Rome, Italy, <sup>7</sup> IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Ajay Vikram Singh, Max Planck Institute Stuttgart, Germany Jorge Cervantes, Texas Tech University Health Sciences Center, United States

\*Correspondence:

Mariacristina Siotto mary.siotto@gmail.com orcid.org/0000-0003-0837-0594

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 31 October 2018 Accepted: 25 January 2019 Published: 11 February 2019

#### Citation:

Siotto M, Filippi MM, Simonelli I, Landi D, Ghazaryan A, Vollaro S, Ventriglia M, Pasqualetti P, Rongioletti MCA, Squitti R and Vernieri F (2019) Oxidative Stress Related to Iron Metabolism in Relapsing Remitting Multiple Sclerosis Patients With Low Disability. Front. Neurosci. 13:86. doi: 10.3389/fnins.2019.00086 Oxidative status may play a role in chronic inflammation and neurodegeneration which are considered critical etiopathogenetic factors in Multiple Sclerosis (MS), both in the early phase of the disease and in the progressive one. The aim of this study is to explore oxidative status related to iron metabolism in peripheral blood of stable Relapsing-Remitting MS with low disability. We studied 60 Relapsing-Remitting MS patients (age 37.2 ± 9.06, EDSS median 1.0), and 40 healthy controls (age 40.3 ± 10.86). We measured total hydroperoxides (dROMs test) and Total Antioxidant Status (TAS), along with the iron metabolism biomarkers: Iron (Fe), ferritin (Ferr), transferrin (Tf), transferrin saturation (Tfsat), and ceruloplasmin (Cp) panel biomarkers [concentration (iCp) and enzymatic activity (eCp), copper (Cu), ceruloplasmin specific activity (eCp:iCp), copper to ceruloplasmin ratio (Cu:Cp), non-ceruloplasmin copper (nCp-Cu)]. We computed also the Cp:Tf ratio as an index of oxidative stress related to iron metabolism. We found lower TAS levels in MS patients than in healthy controls (CTRL) and normal reference level and higher dROMs and Cp:Tf ratio in MS than in healthy controls. Cp and Cu were higher in MS while biomarkers of iron metabolism were not different between patients and controls. Both in controls and MS, dROMs correlated with iCp (CTRL r = 0.821, p < 0.001; MS r = 0.775 p < 0.001) and eCp (CTRL r = 0.734, p < 0.001; MS r = 0.820 p < 0.001). Moreover, only in MS group iCp correlated negatively with Tfsat (r = -0.257, p = 0.047). Dividing MS patients in "untreated" group and "treated" group, we found a significant difference in Fe values [F(2, 97) = 10.136, p < 0.001]; in particular "MS untreated" showed higher mean values (mean = 114.5, SD = 39.37 µg/dL) than CTRL (mean 78.6, SD = 27.55 µg/dL p = 0.001) and "MS treated" (mean = 72.4, SD = 38.08 µg/dL; p < 0.001). Moreover, "MS untreated" showed significantly higher values of Cp:Tf (mean = 10.19, SD = 1.77∗10−<sup>2</sup> ; p = 0.015), than CTRL (mean = 9.03, SD = 1.46 <sup>∗</sup>10−<sup>2</sup> ). These results suggest that chronic oxidative stress is relevant also in the remitting phase of the disease in patients with low disability and short disease duration. Therefore, treatment with antioxidants may be beneficial also in the early stage of the disease to preserve neuronal reserve.

Keywords: multiple sclerosis, oxidative stress, iron metabolism, total antioxidant status, hydroperoxides, ceruloplasmin:transferrin ratio, ceruloplasmin

#### INTRODUCTION

fnins-13-00086 February 7, 2019 Time: 18:22 # 2

Multiple sclerosis (MS) is a chronic immune-mediated condition that can affect the brain and the spinal cord and is characterized by a relapsing remitting (RRMS) course eventually followed by secondary progression (SPMS) or gradual progression of disability since the beginning (primary progressive MS – PPMS). MS has been traditionally considered a focal inflammatory demyelinating disease of the white matter, while today the role of chronic and diffuse gray matter neurodegeneration in the disability accrual is well established (Stys et al., 2012). Neurodegeneration accompanies demyelination since the early phases of the disease and becomes the main pathological feature in the secondary and primary progressive forms.

Inflammation and neurodegeneration are mutually dependent phenomena. Inflammation, in fact, induces degeneration, likely through excitotoxicity mechanisms (Centonze et al., 2009); on the other hand neurodegeneration can induce inflammatory response, both in the central nervous system (CNS) and in peripheral blood (Träger and Tabrizi, 2013), as demonstrated also in other neurodegenerative conditions (i.e., amyotrophic lateral sclerosis, Alzheimer's disease, Parkinson's disease, and Huntington's disease) (Träger and Tabrizi, 2013). Nevertheless, consolidated biomarkers of chronic inflammation in MS are lacking, although many studies have identified oxidative stress markers as particularly promising to estimate peripheral inflammation in MS since inflammation leads to oxidative stress and vice-versa (Naidoo and Knapp, 1992; Karg et al., 1999; Besler and Comoglu, 2003 ˇ ; Ferretti et al., 2005; Koch et al., 2006; Alimonti et al., 2007; Ortiz et al., 2009, 2013; Ghabaee et al., 2010; Ristori et al., 2011; Miller et al., 2012; Oliveira et al., 2012; Tasset et al., 2012). The majority of these studies have demonstrated an increase of peripheral inflammation in the progressive forms of MS or during a relapse; few of them have suggested that oxidative stress and antioxidant capacity are elevated also in RRMS during the remitting phase (Ferretti et al., 2005; Koch et al., 2006; Miller et al., 2012; Oliveira et al., 2012).

Oxidative status is considered to be related to iron (Fe) metabolism. This metal might play a role in the pathogenesis of inflammation and neurodegeneration in MS, causing microglia activation, induction of mitochondria dysfunction and free radicals in the body and in the CNS (Drayer et al., 1987; Bizzi et al., 1990; Zivadinov et al., 2010, 2018; Sheykhansari et al., 2018). Ceruloplasmin (Cp) is an acute phase protein (Hellman and Gitlin, 2002), playing a fundamental role in copper (Cu) and Fe metabolism and holding a strong antioxidant function due to its ferroxidase activity (Gutteridge, 1995): indeed, the Cp:Transferrin (Tf) system (measured by the Cp:Tf ratio) is reported to be the main antioxidant system in peripheral blood (Kozlov et al., 1984). Cp has also been recognized as marker of inflammation in systemic pathologies (Göçmen et al., 2008; Tang et al., 2012) and we found that the Cp:Tf ratio is elevated in Alzheimer's disease (Squitti et al., 2010; Siotto et al., 2016), in stroke (Altamura et al., 2009; Squitti et al., 2018b) and in subacute post-stroke patients affected by neuropathic pain (Siotto et al., 2017), where it correlates with the clinical status.

Our study aimed to identify possible and easy to test markers of peripheral inflammation and to explore Fe metabolism in relation to the clinical status in MS patients with low disability in the remitting phase. For this purpose, we used two commercially available biomarker of oxidative stress, along with a panel of biomarkers related to Fe metabolism, strictly associated with oxidative stress.

#### MATERIALS AND METHODS

#### Subjects

The study was performed at Neuroscience Department of the Fatebenefratelli Hospital, Isola Tiberina, Rome and at Neurology Unit of the Campus Bio-Medico University of Rome.

Sixty Relapsing-Remitting MS patients (45 females, age 37.2 ± 9.06) fulfilling the 2010 revision of diagnostic criteria of MS (Polman et al., 2011) were recruited. Forty healthy unrelated volunteers (22 females, age 40.3 ± 10.86) of comparable age were also selected as control group (CTRL). All the included patients were free from relapses for at least 6 months before the blood sampling.

Exclusion criteria were the following: therapy with corticosteroids or ACTH in the month before the blood sampling, pregnancy, anemia, alcohol, and drug abuse, use of dietary supplements, chronic diseases potentially inducing systemic inflammation (heart or pulmonary diseases, diabetes, autoimmune diseases, etc.), primary or secondary hepatic diseases, hemochromatosis, aceruloplasminemia and any other diseases with known or presumable effect on Cu/Fe metabolism.

Local institutional ethics committees approved the study and all participating subjects gave written informed consent to be included in the study, in line with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and the standards established by the Authors' Institutional Review Board.

#### Blood Sampling and Biochemical Assay

Fasting blood samples were collected in the morning and sera fractions were separated by centrifugation (3000 rpm, 10 min, and 4◦C). They were then aliquoted and rapidly stored at −80◦C.

The subject's aliquots and the reference samples were thawed just before the assay.

All the serum analyses were performed in duplicate on the biochemical analyzer Horiba Pentra 400 (ABX Diagnostic, Montpellier, France).

Total antioxidant blood capacity (TAS) was measured using the TAS kit (Randox Laboratories, Crumlin, United Kingdom) (Rice-Evans and Miller, 1994). Hydroperoxide content was assessed by d-ROMs test (Diacron, Italy) and expressed in arbitrary units (U.CARR) where 1 U.CARR corresponding to 0.08 mg/100 ml of hydrogen peroxide (Alberti et al., 2000).

Iron (Fe) was measured by the photometric test using Ferene (Higgins, 1981) (Horiba ABX, Montpellier, France). Transferrin (Tf) levels were measured by immunoturbidimetric assay (Skikne et al., 1990) and Ferritin was measured by latex-enhanced turbidimetric immunoassay (Simó et al., 1994) (Horiba ABX, Montpellier, France). Concentration of immunological Cp (iCp) was measured by immunoturbidimetric assay (Wolf, 1982) (Futura System SRL, Rome, Italy). The enzymatic activity of Cp (eCp) was tested with an automated version of the manual Schosinsky o-dianisidine eCp assay (Lehmann et al., 1974; Schosinsky et al., 1974), adapted by our laboratory (Siotto et al., 2014). The serum copper (Cu) concentration was measured using the colorimetric assay of Abe et al. (1989) (Randox Laboratories, Crumlin, United Kingdom).

For each sample, we also computed the specific activity of Cp (enzymatic activity per mg of Cp concentration in IU/mg<sup>∗</sup> 10−<sup>1</sup> ), as the ratio between the Cp enzymatic activity and immunoturbidimetric Cp concentration (eCp/iCp) (Siotto et al., 2014) and the ratio between Cp and Tf serum concentrations (Cp:Tf <sup>∗</sup> 10−<sup>2</sup> ) (Nobili et al., 2013). nCp-Cu was calculated by means of the equation provided by Walshe [appendix of Walshe (2003)] on the basis of total copper and iCp concentrations in serum. For each serum copper and ceruloplasmin pair, we computed the amount of copper bound to ceruloplasmin (CB) and the amount of nCp-Cu, following standard procedures (Appendix 1: "Calculation of 'free copper' concentration"). The Cu:Cp ratio (Twomey et al., 2006; Squitti et al., 2014) was calculated as reported in Twomey et al. (2006). Moreover, Tf saturation (Tf-sat) was calculated by dividing serum Fe (µg/dL) by the total iron binding capacity (TBC = TF in mg/dL<sup>∗</sup> 1.25) and multiplying by 100.

#### Statistical Analysis

Smirnov test was applied to test the normality distribution of continuous variables. To test difference in age and clinical characteristics between two groups was applied the Student's t-test or, when necessary, non-parametric Mann-Whitney test. A logarithmic transformation was applied to minimize the variability and to better approximate the data distribution to normality.

Correlation between all biochemical variables were calculated by the Pearson's correlation coefficient separately in the MS and CTRL groups to look for any differences in the reciprocal behavior of the variables between the two groups. Non-parametric Spearman's correlation was calculated to test the correlation between the biochemical variables and the clinical variables (EDSS, disease duration). Analysis of variance (ANOVA) model was applied to evaluate the difference in the biochemical variables between CTRL and MS patients and between CTRL and MS divided in "untreated MS" and "treated MS," adjusting for sex and age. The results were presented as mean and standard deviation (SD). Benjamini-Hochberg procedure was applied to adjust the p-value in multiple comparisons. A p-value < 0.05 has been considered statistically significant. The statistical analysis was performed using IBM SPSS Statistics for Windows version 19.0.0.

# RESULTS

All patients had a low disability [Expanded Disability Status Scale, EDSS 1.0 (0.0–4.5)], except for a single male patient with EDSS 4.5. Patients on disease modifying therapy (DMT) were 78%, of which 85% treated with interferon-beta (Significant demographic and clinical data are listed in **Table 1**). Patients and controls did not diverge for age but were different for sex distribution (p = 0.018, see **Table 1**).

Differences in biochemical variables, adjusted for sex and age in ANOVA analyses, between MS patients and CTRL were reported in **Table 2**. TAS was significantly lower in patients than in controls (MS mean = 1.24, SD = 0.14 vs. CTRL mean = 1.39, SD = 0.13 mmol/L; p = 0.001) and lower than normal reference range indicated by the manufacturer (1.30–1.77 mmol/L). The total hydroperoxides in circulation were higher in MS patients than in healthy volunteers (MS mean = 329.6, SD = 75.15 vs. CTRL mean = 295.2, SD = 61.38 UCarr; p = 0.032). Moreover, the Cp:Tf ratio was higher in MS patients than in controls (MS mean = 9.89, SD = 1.48<sup>∗</sup> 10−<sup>2</sup> vs. CTRL mean = 9.03, SD = 1.46 ∗ 10−<sup>2</sup> ; p = 0.005) (**Table 2** and **Figure 1**).

The iCp values were higher in patients with respect to healthy controls (MS mean = 26.9, SD = 3.86 vs. CTRL mean = 24.8, SD = 3.28 mg/dL; p = 0.006) and coherently the Cu values were a slightly higher in patients (MS mean = 13.7, SD = 0.29 vs. CTRL mean = 12.8, SD = 0.33 µmol/L; p = 0.043; **Figure 1**).

The following biological variables correlated both in healthy controls and MS patients: dROMs with iCp (CTRL: r = 0.821, p < 0.001; MS: r = 0.775 p < 0.001) and eCp (CTRL: r = 0.734, p < 0.001; MS: r = 0.820 p < 0.001). Moreover, in MS patients iCp correlated negatively with Tfsat (r = -0.257, p = 0.047) but it not survived after the adjustment for multiple comparisons (B-H adjusted p = 0.094) (**Figure 2**). No additional correlations were found among any of the other biochemical parameters and EDSS, disease duration (all p > 0.2).

Considering untreated MS patients and treated MS as two separate groups, we did not observe differences in disease duration (Mann Whitney test p = 0.10) and EDSS (Mann Whitney test p = 0.448).

We performed the ANOVA analyses comparing the two groups of MS patients with CTRL. All analyses were adjusted for sex and age. TAS was significantly lower in both patients' groups than in controls ("untreated MS" mean = 1.25, SD = 0.14 mmol/L and "treated MS" mean = 1.23, SD = 0.15 mmol/L vs. CTRL mean = 1.39 SD = 0.15 mmol/L; p = 0.015 and


TABLE 1 | Demographic of multiple sclerosis patients (MS) and healthy volunteers (CTRL).

In non-bold are reported data of Female MS and Female CTRL.

TABLE 2 | Biological variable differences in multiple sclerosis patients (MS) and in healthy volunteers (CTRL).


<sup>a</sup>ANOVA model was applied to evaluated the difference in the biochemical variables between healthy controls and SM patients adjusting for sex and age. <sup>b</sup>ANOVA model was applied considered the logarithmic transformation of the variable.

Bold indicates p < 0.05.

p = 0.005, respectively) which is consistent with our previous observations. The total hydroperoxides in circulation were no longer significantly different between the three groups (p = 0.081). Dividing MS patients in "untreated" group and "treated" group, we found a significant difference in Fe values [F(2, 97) = 10.136, p < 0.001]; in particular "untreated MS" showed higher mean values (mean = 114.5, SD = 39.37 µg/dL) than CTRL (mean 78.6, SD = 27.55 µg/dL p = 0.001) and "treated MS" (mean = 72.4, SD = 38.08 µg/dL; p < 0.001). Moreover, "untreated MS" presented significantly higher values of Cp:Tf (mean = 10.19, SD = 1.77<sup>∗</sup> 10−<sup>2</sup> ; p = 0.015), than CTRL (mean = 9.03, SD = 1.46 <sup>∗</sup> 10−<sup>2</sup> ) (**Figure 3**).

#### DISCUSSION

The main result of this study is that in RRMS patients with low disability in the early phase of disease, the oxidative stress status is altered, as it is revealed by low levels of TAS, and high levels of total hydroperoxides and of Cp:Tf (**Table 2** and **Figure 1**). In fact, even though EDSS scores were low, suggesting a low rate of neurodegeneration, all the serum biomarkers of oxidative status analyzed revealed a marked imbalance in systemic oxidative stress.

Another interesting result is that dividing MS patients in "treated" and "untreated," we found increased level of Fe and Cp:Tf ratio in "MS untreated," suggesting a role of therapy in oxidative stress related to Fe levels (**Figure 3**).

We employed TAS and dROMs as oxidative stress markers, two easy to use and common laboratory assays to test general oxidative stress status of the body, and the Cp: Tf ratio which is an index optimized from our laboratory to explore oxidative stress mediated by Fe.

The TAS is a measure of the total antioxidant capacity of general circulation. The biochemical assay is based on the principle of inducing the antioxidant defense of serum; this is achieved by introducing a radical reagent opportunely and previously formed in the sample; the serum sample is then added. Serum can exert a suppression of the radical in a way that is directly proportional to the ability of all antioxidant components in serum to counteract the radical oxidation. We found that serum samples of MS patients were not able to counteract the induced oxidative stress, as revealed by TAS values lower than healthy control and reference range (**Table 2** and **Figure 1**). No differences in TAS levels were found between "treated MS" and "untreated MS"; TAS values were significantly lower in both patients' group and under the reference range.

The dROMs is a test that measures directly the hydroperoxides circulating in serum (Cesarone et al., 1999; Alberti et al., 2000). High dROMs have been found in patients with different chronic conditions, correlating with the clinical outcome (Daniil et al., 2008; Capone et al., 2012; Vassalle et al., 2012; Taguchi et al., 2013) and with C reactive protein (Kotani and Taniguchi, 2012; Taguchi et al., 2013). Moreover, antioxidant administration has been demonstrated to reduce dROMs (Cornelli et al., 2001). In this study we noticed a very strong positive correlation

between iCp and eCp and dROMs (both in healthy CTRL and in MS with a Spearman's rho close to 0.8, **Figure 2**). On this basis, dROMs test appears as an alternative measurement of Cp enzymatic activity, or related to the antioxidant activity of Cp. Erel et al. (Harma, 2006) sustained that the chromogen employed in dROMs test was a substrate of the ferroxidase Cp while another study asserted that there is no correlation between ceruloplasmin and dROMs (Colombini et al., 2016). Nevertheless, the hypothesis that enzymatic o-dianisidine assays such as eCp might in turn be influenced by the presence of blood hydroperoxides cannot be excluded. Whether measuring oxidative stress or the ferroxidase activity of Cp (as sustained by Erel), dROMs is a reliable method to detect a peripheral response to a systemic stress.

In this study we found dROMs levels higher in MS compared to healthy controls (**Table 2** and **Figure 1**) but no differences were found dividing patients in treated or not.

In order to further explore the oxidative stress status in MS, we also measured the Cp:Tf ratio and we found higher values in MS than in CTRL (**Table 1** and **Figure 1**) and in the group of "untreated MS" (**Figure 3**). This ratio is reported to be the main antioxidant system in serum, and it was measured with Electron Paramagnetic Resonance Spectroscopy by Kozlov et al. (1984). The Cp:Tf variable analyzed in this study, is a calculated index that provides a good quantification of this antioxidant system (Altamura et al., 2009; Nobili et al., 2013). In fact, it quantifies the existing ratio between Cp and Tf. A correct "handling" and distribution of Fe is evinced by a correct stoichiometry between these two proteins. When this occurs, there is consequently a lower probability to develop oxidative stress because the two proteins work in concert reducing levels of ferrous ions, Fe2+, the Fe ion more toxic (Singh et al., 2012) that reacts with hydrogen peroxides initiating oxidative stress chain reactions via Fenton's and Heber Weiss reactions and generating reactive oxygen species.

According to the hypothesis of cerebrospinal venous insufficiency (Zamboni et al., 2009; Zivadinov et al., 2010) abnormality in blood drainage from the brain and spinal cord may contribute to nervous system damage in MS through Fe overload and deposition (Singh and Zamboni, 2009; Singh, 2010). In 2018 a clinical trial study showed no efficacy of venous percutaneous transluminal angioplasty (Zamboni et al., 2018); in fact, the treatment did not increase the proportion of patients who improved functionally nor did it reduce the mean

number of new combined brain lesions on magnetic resonance imaging (MRI) at 12 months. However, the CCSVI hypothesis has been linked with the potential effects of Fe deposition in the brain parenchyma. Together with radiological data which suggest excessive Fe accumulation in the brain of MS patients (Drayer et al., 1987; Bizzi et al., 1990; Zivadinov et al., 2010), this hypothesis drove researchers attention upon the Fe as a possible cause of neurodegeneration, although the reasons of Fe overload are not clear yet and presumably they do not depend on CCSVI. The brain Fe accumulation could lead to chronic cell stress, with consequent axonal and neuronal death. Moreover, there are some heme-involved processes that caused lipid oxidation. During de-myelinization, Fe could be released from myelin sheath (for an extensive review see Adamczyk and Adamczyk-Sowa, 2016).

However, we did not find any significant difference between MS group and controls in the markers strictly related to Fe as Fe itself, Ferr, and Tf. On the other hand, we found significant higher level of Fe in "untreated MS" respect to both control and "treated MS" (**Table 2** and **Figure 3**). Despite preliminary and limited by the small sample size, these results are quite interesting as they highlight the importance of better investigating the role of treatment in Fe metabolism.

Looking at the other variables, we also found iCp and Cu higher in MS patients compared to healthy controls. Cp is acute-phase reactant in inflammation processes (such as infection, chronic diseases, arthritis and several neoplasia) and Cu follows the same trend being structurally bound to Cp. Holo-Cp biosynthesis occurs in liver where the Cu is charged in the

apo-form of the protein. The Cu in the Cp is not exchangeable, being part of the protein structure. So, in MS, the rise of Cp can explain also the rise of Cu due to an inflammation status. Recently, a study from De Riccardis et al. (2018) on 38 MS patients with EDSS scores under 3.5 reported that Cp and Cu values were higher in MS than in controls of an age and sex matched group of 39 subjects, according to our results. Moreover, they found a significant increase in dROMs values in MS patients.

In our study we found no significative difference in other important markers of Cu status like nCp-Cu that measures the portion of Cu not structurally bound to Cp, and Cu:Cp ratio which is an index of the correct stoichiometry between Cu and Cp (**Table 2**). Conversely, in various study on different neurological diseases, such as Alzheimer's or Wilson's disease (Siotto et al., 2016; Squitti et al., 2018a) or stroke (Siotto et al., 2017; Squitti et al., 2018b) we found these specific markers of Cu status altered, with a clear indication of Cu dishomeostasis.

In our opinion, in our MS group iCp and Cu were higher due to systemic inflammation and not as a result of an imbalance systemic Cu management.

Cp is involved in the metabolism of Fe being the main ferroxidase of the serum, as it oxidizes Fe from Fe2<sup>+</sup> to Fe3+, which could be charged from Tf for being transported into the circulation. The increase of iCp in MS and its negative correlation with TfSat stands for an impairment in the Fe handling, being that in MS the more circulating Cp correlates with less TfSat (Letendre and Holbein, 1984; Hellman and Gitlin, 2002). Transferrin itself showed no differences between MS and CTRL as reported also in a study on CSF samples (LeVine et al., 1999).

Summarizing our results, the oxidative stress is clearly an evidence in MS patients, but while the values of TAS and dROMs appeared not to be influenced by therapy, underlining the general systemic oxidative stress in MS, Cp:Tf ratio is a good index to detect oxidative stress strictly related to Fe and potentially related with therapy.

Cp:Tf ratio values is an easy and good marker to investigate the mismanagement of Fe. A higher Cp:Tf ratio has been observed in two studies on stroke patients during acute phase (Altamura et al., 2009; Squitti et al., 2018b). Moreover, we recently found an increase of Cp:Tf ratio in subacute stroke patients affected by neuropathic pain (Siotto et al., 2017). In this group of patients, TAS levels were lower than normal, whereas the Cp:Tf ratio was higher. Decreased TAS levels indicated depletion of the antioxidant system compounds and a reduced ability to counteract the increased oxidative stress generated by stroke injury. The parallel increase of Cp:Tf ratio revealed the activation of systemic processes leading to the cellular internalization of Fe in order to contain oxidative stress in subacute post-stroke patients with neuropathic pain (Carbonell and Rama, 2007).

Similarly, TAS levels, being lower in our group of MS patients, than both in healthy controls and in the reference range, reveal an inability to counteract oxidative stress in patients with early MS. In the meantime, the Cp:Tf system responds to a general status of oxidation (as depicted also from higher levels of hydroperoxides) and is activated to exert a cellular internalization of Fe, especially in patients with higher levels of Fe.

In a study by Oliveira et al. (2012) oxidative stress in MS was evaluated by measuring various oxidative biomarkers. Compared to controls, MS patients with EDSS > 3.5 exhibited higher plasma levels of tert-butyl hydroperoxide-initiated chemiluminescence and carbonyl protein, as well as lower plasma levels of nitric oxide, total radical-trapping antioxidant parameter (measured by the same test we used in current study). In agreement with our results, Oliveira et al. (2012) demonstrated that oxidative stress is important in the physiopathology of MS progression. Is it well known that reactive oxygen species are generated in excess by macrophages which have been implicated as mediators of demyelination and axonal damage in both MS and experimental autoimmune animal models (Gilgun-Sherki et al., 2004).

Another study exploring the value of oxidative stress biomarkers in MS patients reported that plasmatic advanced oxidation protein products were significantly higher while ferric reducing ability and thiol group levels were lower in MS patients in comparison with healthy controls (Pasquali et al., 2015).

Our results underline the importance to monitor oxidative stress and Fe in patients, both at the beginning of the disease and during the DMT. The blood tests employed in this study are easier to perform and cheaper than other serum and radiological biomarkers. MRI is a very well-established method for qualitative and quantitative assessment of MS related damage, but it needs clinical expertise and cumbersome instrumentation set up.

## CONCLUSION

An altered status of oxidative stress and/or antioxidant response is detectable in MS patients with low disability analyzed during a relapse-free period, suggesting the occurrence of a systemic subclinical inflammation that accompanies MS even in the early stages and in the remitting phase. Assessment of the oxidative stress biomarkers and particularly of Cp:Tf ratio, which is strictly related to Fe management, is an easy way to monitor oxidative stress in MS. Moreover, adding antioxidants to conventional immunotherapy in MS may be reasonable and highly beneficial for MS patients due to their ability to reduce oxidative stress. Further research should be performed to test new antioxidants, and to develop new methods to monitor oxidative stress.

# AUTHOR CONTRIBUTIONS

MMF, DL, AG, SV, and FV contributed to patient enrolment and clinical evaluation. MS, RS, and MV contributed to blood sample collection and biochemical evaluation. MS, IS, MMF, PP, MCAR, RS, and FV contributed to study design, statistical analysis, data interpretation, and manuscript preparation.

# FUNDING

The research leading to these results has received funding from: Grant Merck Serono SPC0188 "Impact of systemic metal homeostasis and oxidative status on clinical and radiological parameters of disease severity in Multiple Sclerosis."

# REFERENCES

fnins-13-00086 February 7, 2019 Time: 18:22 # 8



**Conflict of Interest Statement:** SR is the inventor of a Cu-related kit for AD diagnosis and minor shareholder in IGEA Research Co (3% shares) and in Canox4Drug SPA (3%) without monetary compensation.

The remaining 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 Siotto, Filippi, Simonelli, Landi, Ghazaryan, Vollaro, Ventriglia, Pasqualetti, Rongioletti, Squitti and Vernieri. 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.

# Combined Visualization of Nigrosome-1 and Neuromelanin in the Substantia Nigra Using 3T MRI for the Differential Diagnosis of Essential Tremor and de novo Parkinson's Disease

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Fabiana Novellino, Italian National Research Council (CNR), Italy Luigi Bubacco, University of Padova, Italy

#### \*Correspondence:

Mengsu Zeng zengmengsu@outlook.com Chunjiu Zhong zhongcj@163.com

†These authors share co-first authorship

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neurology

Received: 30 September 2018 Accepted: 25 January 2019 Published: 12 February 2019

#### Citation:

Jin L, Wang J, Wang C, Lian D, Zhou Y, Zhang Y, Lv M, Li Y, Huang Z, Cheng X, Fei G, Liu K, Zeng M and Zhong C (2019) Combined Visualization of Nigrosome-1 and Neuromelanin in the Substantia Nigra Using 3T MRI for the Differential Diagnosis of Essential Tremor and de novo Parkinson's Disease. Front. Neurol. 10:100. doi: 10.3389/fneur.2019.00100 Lirong Jin1†, Jian Wang2,3†, Changpeng Wang1†, Danlan Lian<sup>4</sup> , Ying Zhou<sup>5</sup> , Yong Zhang<sup>6</sup> , Minzhi Lv <sup>7</sup> , Yuanfang Li <sup>5</sup> , Zhen Huang<sup>1</sup> , Xiaoqin Cheng<sup>1</sup> , Guoqiang Fei <sup>1</sup> , Kai Liu2,3 , Mengsu Zeng2,3 \* and Chunjiu Zhong<sup>1</sup> \*

<sup>1</sup> Department of Neurology, Zhongshan Hospital, Fudan University, Shanghai, China, <sup>2</sup> Department of Radiology, Zhongshan Hospital, Fudan University, Shanghai, China, <sup>3</sup> Shanghai Medical Imaging Institute, Shanghai, China, <sup>4</sup> Department of Radiology, Xiamen Branch, Zhongshan Hospital, Fudan University, Xiamen, China, <sup>5</sup> Department of Neurology, Xiamen Branch, Zhongshan Hospital, Fudan University, Xiamen, China, <sup>6</sup> MR Research, GE Healthcare, Shanghai, China, <sup>7</sup> Department of Biostatistics, Zhongshan Hospital, Fudan University, Shanghai, China

Differentiating early-stage Parkinson's disease (PD) from essential tremor (ET) remains challenging. In the current study, we aimed to evaluate whether visual analyses of neuromelanin-sensitive magnetic resonance imaging (NM-MRI) combined with nigrosome-1 (N1) imaging using quantitative susceptibility mapping (QSM) in the substantia nigra (SN) are of diagnostic value in the differentiation of de novo PD from untreated ET. Sixty-eight patients with de novo PD, 25 patients with untreated ET, and 34 control participants underwent NM-MRI and QSM. NM and N1 signals in the SN on MR images were visually evaluated using a 3-point ordinal scale. Receiver operating characteristic (ROC) analyses were performed to determine the diagnostic values of the visual ratings of NM and N1. The diagnostic values of the predicted probabilities were calculated via logistic regression analysis using the combination of NM and N1 visual ratings, as well as their quadratic items. The proportions of invisible NM and invisible N1 were significantly higher in the PD group than those in the ET and control groups (p < 0.001). The sensitivity/specificity for differentiating PD from ET was 0.882/0.800 for NM and 0.794/0.920 for N1, respectively. Combining the two biomarkers, the area under the curve (AUC) of the predicted probabilities was 0.935, and the sensitivity/specificity was 0.853/0.920 when the cutoff value was set to 0.704. Our findings demonstrate that visual analyses combing NM and N1 imaging in the SN may aid in differential diagnosis of PD and ET. Furthermore, our results suggest that patients with PD exhibit larger iron deposits in the SN than those with ET.

Keywords: Parkinson's disease, essential tremor, neuromelanin, nigrosome-1, QSM, substantia nigra

# INTRODUCTION

Parkinson's disease (PD) and essential tremor (ET) are common movement disorders, especially among older adults (1). PD is characterized by motor symptoms including bradykinesia, resting tremor, and rigidity, while ET often manifests as isolated tremor in the bilateral upper limbs. Although they are distinct entities, these two movement disorders may share some clinical characteristics, such as non-motor features and resting/postural tremor, as well as genetic and pathological mechanisms (2, 3). Hence, the differentiation of PD and ET remains challenging, especially early in the disease course.

Neuroimaging may aid in the differentiation of the two movement disorders. Dopamine transporter (DAT) imaging of the striatum is recommended in the differential diagnosis of PD (4). However, DAT imaging is expensive, subjects the patient to low doses of radiation, and is only available in specialized centers, limiting its clinical application. In contrast, MRI is widely available and does not subject the patient to radiation. Recent studies have provided evidence that MRI biomarkers may aid in the diagnosis of movement disorders (5, 6) and help to reveal the pathological changes correlated with motor and non-motor symptoms (7).

The pathological hallmarks of PD include progressive neurodegeneration of dopaminergic neurons and iron overload in substantia nigra (SN) (8). In contrast, whether ET is a degenerative disease is still debated, although some studies suggest that it is associated with cerebellar degeneration (9). Dopaminergic neurons in the SN contain a black pigment called neuromelanin (NM). Based on the paramagnetic properties of NM, high-resolution T1-weighted fast spin echo (FSE) imaging at high field strength (e.g., 3T) can visualize NM-generated contrast. This technique is referred to as neuromelanin-sensitive MRI (NM-MRI) (10). Previous studies have indicated that the signal intensity of NM is decreased in patients with PD (10, 11), even in the early stage of the disease (12), while it remains unchanged in patients with ET (13, 14), when compared with that in healthy controls. In addition, nigrosome-1 (N1), the largest of the five described nigrosomes, is most affected in patients with PD (∼98% neuronal loss in N1) (15). N1 represents the pockets of high signal intensity in the dorsal part of the healthy SN, at intermediate and caudal levels on high resolution T2<sup>∗</sup> /SWI, and can be visualized as a "swallow-tail sign" (16, 17). However, hyperintensity of N1 is absent in most patients with PD (17), possibly due to increases in iron deposition that occur in parallel to the loss of dopaminergic cells (16). Quantitative susceptibility mapping (QSM) may overcome several nonlocal restrictions of SWI and phase imaging, allowing for the quantification of iron content (18, 19). Indeed, this method may be more sensitive for detecting iron-related changes in patients with PD (20). To our knowledge, however, no studies have investigated N1 appearance in patients with ET.

Unlike voxel-based morphometry, diffusion tensor imaging, or blood oxygenation level-dependent imaging, which require complicated post-processing or quantitative measurements, NM on NM-MRI and N1 on QSM can be assessed visually, making these methods feasible for clinical application. To the best of our knowledge, no previous studies have investigated the combination of these two MR sequences for the differential diagnosis of PD and ET. In the current study, we aimed to evaluate whether visual analyses of NM imaging using NM-MRI combined with N1 imaging using QSM in the SN are of diagnostic value in the differentiation of de novo PD from untreated ET.

#### PARTICIPANTS AND METHODS

#### Participants

Sixty-eight patients with de novo PD, 25 patients with untreated ET, and 34 healthy controls were voluntarily recruited between September 2016 and April 2018. Patients with PD were diagnosed in accordance with MDS clinical diagnostic criteria for PD (4), while patients with ET were diagnosed in accordance with the criteria outlined in the Consensus Statement of the Movement Disorders Society on Tremor (21), by two movement specialists (J.L.R. and H.Z.). All patients were drug naïve. All control participants were recruited as volunteers from the community and had no history of neurological/psychiatric disorders. Exclusion criteria were as follows: history of other neurological/psychiatric disorders, severe infection, liver dysfunction, renal insufficiency, past/current substance abuse, tremor-related dysmetabolism including thyroid dysfunction and drug toxicity, and abnormal signals that affected further analyses on structural MRI. Unified Parkinson's Disease Rating Scale (UPDRS) motor scores were obtained for all patients with PD and ET. This study was approved by the Committee on Medical Ethics of Zhongshan Hospital, Fudan University. Written informed consent was obtained from all participants.

#### MRI Protocol and Imaging Analysis Imaging Protocol

All MR images were acquired using a 3T MR unit (DiscoveryTM MR750, GE Healthcare, Milwaukee, WI). A T1-weighted fast spin-echo sequence was obtained for NM-MRI images, as previously described (22), and the imaging parameters were as follows: repetition time/echo time (TR/TE), 600/13 ms; echotrain length, 2; section thickness, 2.5 mm, with no intersection gap; number of slices, 16; matrix size, 512 × 320; field-of-view (FOV), 220 mm; NEX, 5. A three-dimensional multi-echo GRE sequence was used to acquire T2<sup>∗</sup> -weighted images, and the scanning parameters were as follows: TR: 51.5 ms; number of echoes, 16; first TE, 2.9 ms; TE spacing, 3 ms; bandwidth, 62.50 kHz; flip angle, 12◦ ; FOV, 22 cm; matrix, 220 × 220; slice thickness, 2 mm; acceleration factor, 2; slices, 66. Afterwards, the QSM images were reconstructed from T2<sup>∗</sup> -weighted images as described in previous studies (20). In addition, conventional MRI sequences including T1-weighted images, T2-weighted fluid-attenuated inversion recovery (FLAIR) images, and diffusion-weighted images (DWI) were obtained to exclude other pathological imaging findings that may have interfered with further assessment. The axial sections were scanned parallel to the anterior commissure-posterior commissure line with whole-brain coverage for QSM and routine MRI scans, and with coverage from the posterior commissure to the pons for NM-MRI.

#### Visual Analysis of Imaging Data

The NM-MR images were transferred to a workstation (ADW4.6, GE Healthcare) and displayed using certain settings (window width: 400, window level: 800–900) for analysis. The method for visual analysis was based on that described in a previous study (23), with modifications. We classified NM-MR images according to an 3-point ordinal scale, as follows: 0, normal view of the SN with high signal intensity bilaterally and no volume loss, indicating a healthy SN; 1, possible abnormality with reduced signal or volume of the SN unilaterally or bilaterally, indicating possible SN pathology; 2, definite abnormality with reduced signal or volume of the SN, indicating SN pathology (**Figure 1**).

QSM data were transferred to a local computer, and images were viewed using MRIcro software (Version: 1.40 build 1). N1 was visualized as an oval-shaped area of low signal intensity surrounded by hyperintensity on QSM in the dorsal part of the healthy SN, at intermediate and caudal levels. Visual analysis of N1 was performed using a 3-point ordinal scale, as follows: 0, normal, N1 present bilaterally; 1, non-diagnostic, indecisive presence of N1 unilaterally or bilaterally; 2, pathological, N1 absent bilaterally (**Figure 1**). Two radiologists (rater 1: W.J. and rater 2: L.D.L.), who were blinded to participants information, independently performed visual analyses twice, with an interval of at least 7 days. The intra- and inter-rater agreement of visual scores was determined using weighted kappa values. For the conflicting cases between the two raters, visual analyses were conducted by a third radiologist (with 30 years of experience) to acquire a final rating for statistical analyses.

#### Statistical Analyses

The results were expressed as the mean ± SD. One-way analyses of variance (ANOVA), Mann-Whitney U-test, and chi-square test were used to compare demographic data. Logistic regression analyses were employed to estimate the combined predicted probabilities of the visual ratings for NM and N1. Diagnostic performances in visual assessments of NM and N1, separately, and in the predicted probabilities were analyzed using receiver operating characteristic (ROC) curves. The chi-square test was used to compare the area under the curve (AUC) values of the different diagnostic models. The level of significance was set to 0.05, and all tests were two-sided. Statistical analyses were performed using SPSS version 22.0 (SPSS Inc., Chicago IL) and Stata/SE version 14.0 (StataCorp LP).

#### RESULTS

#### Demographic and Clinical Data

The demographic and clinical characteristics of all participants are summarized in **Table 1**. No significant differences in gender, age, Mini Mental State Examination (MMSE) scores, or Montreal Cognitive Assessment (MoCA) scores were observed among the three groups. Disease duration was significantly longer in patients with ET than in patients with PD (p < 0.001), although there was no significant difference in UPDRS tremor scores between the ET and PD groups. All patients with PD had mild disease severity (Hoehn and Yahr stages 1 to 2).

### Visual Analyses of the SN on NM-MRI and N1 on QSM

The proportion of conflicting cases was 22.8% (29/127) for NM-MRI analyses and 11.8% (15/127) for N1-QSM analyses for rater 1. For rater 2, the proportion of conflicting cases was 28.3% (36/127) for NM-MRI analyses and 22.0% (28/127) for N1-QSM analyses. The proportion of conflicting cases between the two observers was 25.2% (32/127) for NM-MRI analyses and 24.4% (31/127) for N1-QSM analyses. Thus, the weighted kappa coefficient was calculated to evaluate intra- and inter-rater agreement for visual analyses. For rater 1, intra-rater agreement values for NM and N1 were 0.837 and 0.903, respectively. For rater 2, intra-rater agreement values for NM and N1 were 0.828 and 0.815, respectively. Furthermore, the inter-rater agreement was 0.827 for NM and 0.777 for N1. Thus, visual analyses were highly consistent.

For NM visual analysis, scores were 0 in 8 patients with PD (11.8%), 1 in 27 patients with PD (39.7%), and 2 in 33 patients with PD (48.5%). In the ET group, NM scores were 0 in 20 patients (80.0%) and 1 in 5 patients (20.0%). In the control group, NM scores were 0 in 30 participants (88.2%) and 1 in 4 participants (11.8%) (**Figure 2A**). NM scores of 2 were not observed in the ET and control groups, and the proportion of NM ratings did not significantly differ between these two groups (p = 0.385). However, the proportion of NM ratings in the PD group differed significantly from that in the ET group (p < 0.001).

N1 scores were 0 in 3 patients with PD (4.4%), 1 in 11 patients with PD (16.2%), and 2 in 54 patients with PD (79.4%). In the ET group, N1 scores were 0 in 11 patients (44.0%), 1 in 12 patients (48.0%), and 2 in 2 patients (8.0%). In the control group, N1 scores were 0 in 10 participants (29.4%), 1 in 19 participants (55.9%), and 2 in 5 participants (14.7%) (**Figure 2B**). The proportion of N1 ratings in the ET group did not significantly differ from that in controls (p = 0.454). However, the proportion of patients with N1 scores of 2 was significantly lower in the ET group than in the PD group (p < 0.001). Representative images obtained from patients with PD and ET and controls are presented in **Figure 1**.

#### Diagnostic Performances of Visual Analyses of NM and N1, and of the Predicted Probabilities Combining the Two Biomarkers

Based on our findings, we then employed ROC analysis to assess the diagnostic values of several models for differentiating PD from ET. The AUC of NM-MRI (model 1) for differentiating PD from ET was 0.890 (95% CI 0.822, 0.958), and the sensitivity and specificity were 0.882 and 0.800, respectively, when the cutoff value for NM scores was set to ≥1. The AUC of N1 on QSM (model 2) for differentiating PD from ET was 0.882 (95% CI 0.802, 0.962), and the sensitivity and specificity were 0.794 and

be visualized in the dorsal part of the healthy SN on QSM images (D, arrow). (D–F) represent that N1 was present, indecisively present and absent, respectively. (G,H), a control subject, female, 65 years, neuromelanin was normal (G) and N1 was present (H, arrow) in bilateral SN. (I,J), an ET patient, 59 years, female, neuromelanin was normal (I) and N1 was present (J, arrow) in bilateral SN. (K,L), a de novo PD patient, 75 years, female, neuromelanin was definitely abnormal in unilateral SN (K, arrowhead) and N1 was absent in bilateral SN (L).

#### TABLE 1 | Clinical data for patients with PD, patients with ET, and controls.


\*Significantly different; PD, Parkinson's Disease; ET, Essential Tremor; MMSE, Mini Mental State Examination; MoCA, Montreal Cognitive Assessment; UPDRS, United Parkinson's Disease Rating Scale.

0.920, respectively, when the cutoff value for N1 scores was set to 2. No significant differences in AUC values were observed between these two models (p > 0.05, **Table 2**).

We calculated the predicted probabilities using logistic regression to further explore the diagnostic performance of these two biomarkers. Model 3 was established using the predicted probabilities obtained by simply combining these two biomarkers (Log it = −2.176 + 1.923 × NM + 1.429 × N1). The AUC of model 3 was 0.933 (95% CI 0.883, 0.983). The sensitivity and specificity of model 3 were 0.809 and 0.960, respectively, when the cutoff value was set to 0.848. The AUC of model 3 was significantly higher than that of model 1 (p = 0.009), whereas it did not significantly differ from that of model 2 (p = 0.051, **Table 2**). Furthermore, the specificity of model 3 was higher than those of model 1 and 2 while the sensitivity of model 3 was lower than that of model 1.

Model 4 was established using the predicted probabilities obtained by combining NM, N1, and N1<sup>2</sup> (Log it = −1.499 + 2.014 × NM−1.194 × N1 + 1.267 × N1<sup>2</sup> ). The AUC was 0.935 (95% CI 0.884, 0.986). The sensitivity and specificity of model 4 were 0.853 and 0.920, respectively, when the cutoff value was set to 0.704. The AUC of model 4 was significantly higher than those of model 1 (p = 0.041) and model 2 (p = 0.014). The sensitivity and specificity of model 4 were best among the four diagnostic models (**Table 2**). We further established model 5 using the predicted probabilities obtained by combining NM, N1, NM<sup>2</sup> , and N1<sup>2</sup> . However, the results of model 5 were not better than those of model 4 (data not shown). The ROCs of the four diagnostic models are shown in **Figure 3**.

nigrosome-1. Note that no definitely abnormal neuromelanin signals (score: 2) were observed in the ET and control groups, and the frequency of nigrosome-1 absence (score: 2) was significantly lower in the ET group than in the PD group (p < 0.001).

#### DISCUSSION

In the current study, we aimed to evaluate whether visual analyses of NM and N1 imaging in the SN are of diagnostic value in the differentiation of de novo PD from untreated ET. Patients with PD exhibited reduced signal intensity on NM imaging and an absence of N1 in the SN, relative to patients with ET and healthy controls. Moreover, when visual analyses of NM and N1 imaging were combined, the model exhibited high diagnostic accuracy for differentiating de novo PD from untreated ET. To date, although ET is a common neurological disease, the pathogenic mechanisms remain poorly understood (24), and there are no adequate diagnostic biomarkers. Our findings suggest that non-invasive neuroimaging studies may aid in the differential diagnosis of tremor disorders, particularly PD and ET.

NM plays an important role in the pathogenesis of PD (25). Previous MRI studies have indicated that NM signal intensity in the SN are decreased even in the early stages of PD (22, 26). NM has a high binding affinity for iron. However, Reimao et al. reported that there is no significant correlation between NM and iron content in the SN (26). A recent review demonstrated that the NM is not only directly involved in reactive oxygen species (ROS) reduction but also in Ca2<sup>+</sup> homeostasis, with NM loss leading to the death of dopaminergic neurons (27). Consistent with our quantitative results, a previous analysis study (14) demonstrated that NM levels in the SN are not significantly decreased in patients with ET. Indeed, only 20% of patients with ET obtained scores of 1. All others obtained scores of 0. In contrast, 39.7 and 48.5% of patients with PD obtained scores of 1 and 2, respectively. These results may provide evidence against a possible pathogenic link between PD and ET.

Consistent with the findings of previous reports (17), our results indicated that N1 was absent on QSM images in 79.4% of patients with PD. While NM is known to participate in intracellular iron metabolism, loss of nigrosome signals may be related to iron deposition in the brain (28). At present, QSM is the optimal imaging method for quantifying iron content in the brain in vivo (18, 20). Thus, our findings indicated that iron overload occurs in N1 in patients with PD. Previous studies have indicated that changes in N1 can aid in the differential diagnosis of PD (29), as they can be observed in both the early and late stages of the disease (17). Our data also demonstrated the presence of N1 in most patients with ET, in contrast to our findings for patients with PD.

To date, several studies have indicated that ET is likely to be a neurodegenerative disease, especially affecting cerebellar system proven by clinical, neuroimaging, and postmortem studies (30, 31). In addition, patients with ET exhibit a 4-fold higher risk of developing PD than those without ET (32). One MRI T2<sup>∗</sup> -relaxometry study revealed that ET is associated with iron deposition in the SN (33). Another functional MRI study provided further evidence of neurodegeneration in patients with ET, reporting over-activation in the parietal cortex and dorsolateral prefrontal cortex (34). Mild abnormalities in striatal DATs have also been observed in patients with ET, along with a typical PD-like pattern of uptake loss (35). However, there is no loss of DAT binding over time in patients with ET, providing evidence against the neurodegeneration hypothesis (36). To our knowledge, our study is the first to report that the rate of N1 absence is significantly lower in patients with ET than in patients with PD, suggesting that ET is associated with a lower iron deposition than PD.

By combining our NM and N1 findings for patients with PD and ET, we developed diagnostic models for the differentiation of the two disorders. Molecular imaging techniques such as DAT imaging may further improve diagnostic accuracy. Novellino et al. reported that DAT-SPECT and MIBG scintigraphy findings are abnormal in patients with probable PD, while they are normal in patients with ET (37). Another study also reported that DAT imaging can be used to differentiate early-stage PD and ET with high sensitivity (84.4%) and specificity (96.2%) (38). However, this technology is not feasible for clinical application. Several transcranial sonography (TCS) studies have also aimed to distinguish patients with PD from those with ET (38–41), achieving moderate sensitivity and high specificity. Despite these results, recording failures due to an insufficient acoustic bone window may limit the use of

TABLE 2 | Diagnostic models for differentiating PD from ET.


<sup>a</sup>Model 1: NM.

<sup>b</sup>Model 2: N1.

#No significant difference was observed between the AUCs of Model 1 and Model 2.

<sup>c</sup>Model 3: predicted probabilities. Log it = −2.176 + 1.923 × NM + 1.429 × N1.

§The AUC of model 3 was significantly higher than that of Model 1.

<sup>d</sup>Model 4: predicted probabilities. Log it = −1.499 + 2.014 × NM – 1.194 × N1 + 1.267 × N1<sup>2</sup> .

\$The AUC of model 4 was significantly higher than those of Model 1 and Model 2.

PD, Parkinson's Disease; ET, Essential Tremor; NM, Neuromelanin; N1, Nigrosome-1; AUC, Area Under the Curve.

TCS (42). A previous NM-MRI study reported sensitivity and specificity values of 67.7 and 93.3%, respectively, when high-signal areas in the SN were used to distinguish ET from PD (13). Because visual analysis is fast and more convenient for clinical applications, we performed visual analyses of NM and N1 to aid in the differential diagnosis of de novo PD and untreated ET. Sensitivity values were >79% for both biomarkers, while specificity values were equivalent to 80 and >90% for NM and N1, respectively. Despite the relatively good diagnostic ability of both visual assessments for NM and N1, further combining the two biomarkers may provide higher diagnostic accuracy in clinical practice for individual diagnosis. Indeed, the combination of both biomarkers achieved a sensitivity of 85.3% and specificity of 92%—values higher than those for each biomarker.

The present study possesses some limitations of note. First, the number of participants was relatively small, especially the number of patients with ET, and all participants were recruited from clinical settings. Therefore, replication in larger population-dwelling samples is warranted to confirm the effectiveness of the diagnostic models used in our study. Second, NM-MRI acquisition times were relatively long in our study, which many patients may be unable to tolerate. However, a recent study introduced a three-dimensional NM-MRI sequence that took only slightly more than 4 min, which may facilitate clinical application of this method (43). Third, the analysis in our study was qualitative, which may be more suitable for clinical application, rather than quantitative. However, QSM can be used to quantify iron content (20), while NM content can be quantified on NM-MRI based on volume/width (22, 44). Further studies should therefore focus on the quantitative differences between PD and ET. Notably, the patients recruited in this study were all drug naïve, which eliminated the undefined confounding factors associated with medication use (45).

In conclusion, our findings indicated that visual analyses combing NM and N1 may represent a diagnostic biomarker for the differentiation of tremor disorders. Furthermore, our results suggest that iron deposition is greater in patients with PD than in those with ET.

#### AUTHOR CONTRIBUTIONS

All authors have carefully reviewed this manuscript. In this study, each author has contributed significantly. LJ, YL, CW, YiZ, XC, ZH, ML, CZ, and GF were responsible for collecting and confirming clinical data. JW, DL, KL, MZ, and YoZ were responsible for the MRI study. CZ designed the study and wrote the manuscript.

#### FUNDING

This study was supported by grants from Science and Technology Commission of Shanghai Municipality (18ZR1406500), the National Key Research and Development Program Foundation of China (2016YFC1306403), and Natural Science Foundation and Major Basic Research Program of Shanghai (16JC1420502).

# REFERENCES


by quantitative susceptibility mapping. Hum Brain Mapp. (2015) 36:4407–20. doi: 10.1002/hbm.22928


45. Fahn S, Oakes D, Shoulson I, Kieburtz K, Rudolph A, Lang A, et al. Levodopa and the progression of Parkinson's disease. N Engl J Med. (2004) 351:2498– 508. doi: 10.1056/NEJMoa033447

**Conflict of Interest Statement:** CZ holds shares of Shanghai Rixin Biotech Co., Ltd., which focuses on the development of new drugs against Alzheimer's disease.

The remaining 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 Jin, Wang, Wang, Lian, Zhou, Zhang, Lv, Li, Huang, Cheng, Fei, Liu, Zeng and Zhong. 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.

# Brain Iron Accumulation in Atypical Parkinsonian Syndromes: in vivo MRI Evidences for Distinctive Patterns

Jae-Hyeok Lee<sup>1</sup> \* and Myung-Sik Lee<sup>2</sup> \*

<sup>1</sup> Department of Neurology, Research Institute for Convergence of Biomedical Science and Technology, Pusan National University Yangsan Hospital, Pusan National University School of Medicine, Yangsan, South Korea, <sup>2</sup> Department of Neurology, Gangnam Severance Hospital, Yonsei University College of Medicine, Seoul, South Korea

Recent data suggest mechanistic links among perturbed iron homeostasis, oxidative stress, and misfolded protein aggregation in neurodegenerative diseases. Iron overload and toxicity toward dopaminergic neurons have been established as playing a role in the pathogenesis of Parkinson's disease (PD). Brain iron accumulation has also been documented in atypical parkinsonian syndromes (APS), mainly comprising multiple system atrophy (MSA), and progressive supranuclear palsy (PSP). Iron-sensitive magnetic resonance imaging (MRI) has been applied to identify iron-related signal changes for the diagnosis and differentiation of these disorders. Topographic patterns of widespread iron deposition in deep brain nuclei have been described as differing between patients with MSA and PSP and those with PD. A disease-specific increase of iron occurs in the brain regions mainly affected by underlying disease pathologies. However, whether iron changes are a primary pathogenic factor or an epiphenomenon of neuronal degeneration has not been fully elucidated. Moreover, the clinical implications of iron-related pathology in APS remain unclear. In this review study, we collected data from qualitative and quantitative MRI studies on brain iron accumulation in APS to identify disease-related patterns and the potential role of iron-sensitive MRI.

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Patrice Péran, Institut National de la Santé et de la Recherche Médicale (INSERM), France Sara Pietracupa, Sapienza University of Rome, Italy

#### \*Correspondence:

Jae-Hyeok Lee jhlee.neuro@pusan.ac.kr Myung-Sik Lee mslee@yuhs.ac

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neurology

Received: 18 October 2018 Accepted: 21 January 2019 Published: 12 February 2019

#### Citation:

Lee J-H and Lee M-S (2019) Brain Iron Accumulation in Atypical Parkinsonian Syndromes: in vivo MRI Evidences for Distinctive Patterns. Front. Neurol. 10:74. doi: 10.3389/fneur.2019.00074 Keywords: iron, neurodegeneration, magnetic resonance imaging, atypical parkinsonian syndromes, multiple system atrophy, progressive supranuclear palsy

# INTRODUCTION

Iron overload in the substantia nigra and its toxicity toward dopaminergic neurons has been suggested to play a key role in the pathogenesis of Parkinson's disease (PD) (1). Brain iron accumulation has also been repeatedly documented in atypical parkinsonian syndromes (APS), mainly comprising multiple system atrophy (MSA), and progressive supranuclear palsy (PSP). Pathological studies have shown that iron levels in the putamen, globus pallidus, and substantia nigra are higher in MSA patients than in PD and control patients, and resembles the levels found in PSP patients (2–4).

Iron-sensitive magnetic resonance imaging (MRI) has been applied to identify iron-related signal changes for the diagnosis and differentiation of these disorders. However, the clinical and pathogenic implications of iron-related pathology detected by MRI in APS remain unclear. In this review study, we collected data from qualitative and quantitative MRI studies on brain iron accumulation in APS to identify disease-related patterns and the potential role of ironsensitive MRI.

**191**

# Potential Pathogenic Mechanisms of Iron Accumulation in APS

Brain iron accumulation can be caused by several factors, such as impaired iron homoeostasis, neuroinflammation, and increased blood–brain barrier permeability (1). The key proteins associated with degenerative parkinsonism, α-synuclein and tau, are believed to play roles in iron homeostasis (5). α-synuclein is a cellular ferrireductase responsible for reducing ferric iron (Fe3+) to ferrous iron (Fe2+) (6). In the presence of functional tau, the export of ferrous iron in neurons is controlled by the binding of the amyloid precursor protein (APP) to tau, which results in the trafficking of the protein complex to the neuronal surface, where APP interacts with ferroportin (7). Changes in these proteins might result in altered cellular iron homeostasis (5).

Dysregulation of the proteins involved in iron homeostasis has been reported in APS patients. In PSP patients, the amount of ferritin, an iron-storing protein, has been reported to be considerably lower than that in control patients, which is the opposite trend of increasing iron burden (8). High levels of iron might exceed the iron-buffering capacity of complexes, such as neuromelanin and ferritin in the substantia nigra (1, 9). In MSA, an increase in ferritin iron coupled with a reduction in ferroportin expression has been detected in the pons and putamen (10). These findings suggest a potential deficit in bioavailable iron despite an excess of tissue iron.

Impaired iron homeostasis can induce toxic protein oligomers and abnormal intracellular aggregates under pathological conditions. The aggregation of α-synuclein and tau has been shown to be triggered by iron in vitro (1). Iron promotes α-synuclein aggregation by inducing β-sheet conformational changes and increasing oxidative stress and microglial activation (4, 11). Abundant evidence has demonstrated the role of iron in tau hyperphosphorylation and aggregation into fibrils (12, 13).

In postmortem brain tissues, high concentrations of iron have been colocalized with α-synuclein and hyperphosphorylated tau aggregates. Notably, the highest iron content in the putamen, as assessed by histochemistry, is correlated with the highest density of glial cytoplasmic inclusions and α-synuclein aggregation in MSA brains (14). Hyperphosphorylated tau aggregates from the brains of PSP patients have been shown to colocalize with ferritin (15).

Iron overload in brain regions that undergo protein misfolding and aggregation can promote oxidative stress and neuroinflammation (4). In turn, iron-induced oxidative stress leads to apoptosis and ferroptosis, a novel mode of regulated cell death dependent on iron and lipid peroxidation (1, 16). Scavenging of accumulated labile iron by chelation can protect against iron-mediated neuronal damage (17).

# CURRENT MRI METHODS FOR DETECTION AND QUANTIFICATION OF BRAIN IRON

#### Iron-Sensitive MRI Techniques

Iron-sensitive MRI techniques have been used to differentiate parkinsonian disorders and monitor disease progression (18–20). Various methods have been used to study iron content, including R2 (1/T2), R<sup>∗</sup> 2 (1/T<sup>∗</sup> 2 ), R′ 2 (1/T′ <sup>2</sup>), susceptibility weighted imaging (SWI), and quantitative susceptibility mapping (QSM) (20, 21). SWI enhances image contrast by using the susceptibility differences between tissues; images are obtained by combining T2<sup>∗</sup> -weighted magnitude images with filtered phase images in the gradient echo data. SWI further enhances the contrast between tissues of different susceptibilities (22). Because of its high degree of sensitivity in detecting and visualizing iron deposition, SWI has been applied in the diagnosis and differentiation of parkinsonian disorders (23). However, SWI does not provide quantitative measurement of magnetic susceptibility, which is a limitation that has been addressed by the recent development of QSM (22, 23). Both QSM and the transverse relaxation rate (R2<sup>∗</sup> ) are highly sensitive and represent the most accurate measurement of iron content in the brain (20); results have been shown to correlate closely with iron concentrations measured in postmortem brain tissues (24, 25). QSM can also overcome the confounding effect of increased water content influencing relaxation values, thus counteracting the effect of increased iron levels (20). Recent studies have shown that QSM achieves stronger diagnostic performance than R2<sup>∗</sup> for detecting iron deposition in the subcortical gray matter (26, 27). However, the opposite effect of myelin on QSM may impede its positive correlation with iron. A histological validation study found lower correlation coefficients for QSM and iron in white matter and cortical gray matter regions than in the basal ganglia when correlating R2<sup>∗</sup> and iron (28).

#### Quantitative MRI Measurements of Brain Iron

The region-of-interest (ROI) method is commonly used to calculate regional iron content. The method is based on the manual or automated segmentation of target structures (29). Automated ROI analysis eliminates bias in the process of selecting images and drawing ROIs (30). Computer-based MRI analyses using machine learning techniques and support vector machine (SVM) classification have proven potential in the discrimination of parkinsonian syndromes and the tracking of disease progression (31, 32). SVM pattern recognition of SWI data can accurately discriminate PD from APS (32).

Focal or unevenly increased iron deposition should be considered in determining the iron concentration within a structure (23, 33). Single or two consecutive ROIs in twodimensional images without delineation of the exact boundaries of a structure may not sufficiently reflect unevenly or focally increased iron content of the whole three-dimensional structure. Voxel-based analysis can provide additional information about the topographic distribution of sub-regional iron deposition (34).

The multimodal MRI approach combines MRI techniques for imaging iron and other MRI sequences that are sensitive to complementary tissue characteristics, including volume atrophy and microstructural damage (35). This approach may aid early diagnosis, monitoring of disease progression, and highlighting of pathogenic mechanisms (36, 37).

# IRON-DEPOSITION PATTERNS IN APS: THE DETERMINANTS OF REGION SPECIFICITY

## Age-Related and Structure-Specific Patterns

The regional distribution of iron is heterogeneous in normal adult brains. The basal ganglia have the highest iron concentrations, whereas low concentrations are detected in the cortical gray matter, white matter, brainstem, and cerebellum (1). Typically, total iron concentrations increase with age in the substantia nigra, putamen, globus pallidus, caudate nucleus, and cortices. However, the rate of iron accumulation varies among strictures throughout the adult life span (38, 39). For example, a previous study found high iron concentration in the globus pallidus regardless of age, but a significantly greater presence of iron in the putamen with advancing age (38). Regional heterogeneity of brain iron and its changes with age have been confirmed in vivo by MRI (1).

The region- or structure-specific pattern of iron deposition in normal aging can be enhanced in pathologic conditions. In the parkinsonian variant of MSA (MSA-P), iron deposition predominantly in the posterolateral part of the putamen is in line with the aging pattern (40, 41). The motor-related subcortical structures may present with different patterns in both volume and iron level. In previous MRI studies, the putamen demonstrated a decrease in volume and an increase in iron level, whereas the thalamus did not show iron deposition despite volume shrinkage (42, 43).

#### Pathoanatomy-Related Distribution

Total iron levels have been found to be elevated in most areas of the basal ganglia in the postmortem brain tissue of patients with APS (2). However, in vivo iron-sensitive MRI has shown topographical differences in excessive iron accumulation among the distinct subtypes of parkinsonism (23, 33, 34, 43– 46). Significant increases in iron-related signals (SWI phase and QSM susceptibility) have been found in the midbrain and globus pallidus of patients with PSP, and in the putamen of patients with MSA (**Figure 1**) (34, 46).

Regions of higher iron content are correlated with the main sites of underlying pathology. Increased iron in the putamen is the most consistent finding in MSA (23, 33, 34, 43–47), and specifically in MSA-P (48, 49), as well as being correlated with hypometabolism on <sup>18</sup>F-flurodeoxyglucose positron-emission tomography (PET) (50). In a previous study, in vivo <sup>18</sup>F-flortaucipir tau PET revealed that PSP patients had the highest overall uptake in the globus pallidus with the greatest discriminant power (51).

The multimodal MRI approach combining volumetry and iron estimation has shown that excessive iron accumulation is associated with advanced atrophy in a region-specific manner (4, 43, 49). R2<sup>∗</sup> values are negatively correlated with volumes in the putamen and globus pallidus (43). In a longitudinal study, iron accumulation in the putamen increased in parallel with the extent of atrophy (49).

#### Uneven or Sub-structural Distribution

Iron distribution can be uneven across the entire structure of the brain. Iron concentrations in MSA-P are highest in the posterolateral subregion of the putamen (33, 41, 44, 48). Substructural differences have also been demonstrated by voxelbased analysis (34, 43). A previous study reported significant increases in SWI hypointensity in the posterolateral putamen and adjacent lateral aspect of the globus pallidus in patients with MSA-P, and in the anterior and medial aspects of the globus pallidus in patients with PSP (34). Similarly, higher R2<sup>∗</sup> values have been reported in more atrophic sub-regions (43). In MSA-P, increases in R2<sup>∗</sup> values in the putamen and globus pallidus have been noted more posteriorly and dorsally than those in PSP.

Conventional MRI techniques are insufficient to delineate the inner structure of the substantia nigra and measure the loss of melanized dopaminergic neurons and the increase of iron deposition (52). Recently, iron-sensitive MRI sequences at high field strengths have been reported to consistently display a hyperintense ovoid area within the dorsolateral border of the otherwise hypointense substantia nigra pars compacta, which is referred to as dorsolateral nigral hyperintensity (53–55). A postmortem 9.4 T MRI study with histopathological correlation indicated that this MRI feature corresponds to nigrosome-1 (56). Loss of dorsolateral nigral hyperintensity, a typical finding for patients with PD, is also observed in the majority of patients with MSA or PSP (53–55).

## CLINICAL IMPLICATIONS OF BRAIN IRON DEPOSITION IN APS

#### Discriminant Markers

Distinctive patterns of iron deposition detected on iron-sensitive MRI can be used in clinical practice to differentiate PD, MSA, and PSP (see **Table 1**). Patients with PSP have more widespread iron deposition in the motor-related subcortical nuclei than patients with other parkinsonian disorders (**Figure 1**) (34, 61). In receiver operating characteristic (ROC) analyses, the putamen has been consistently demonstrated to have the highest area under the curve (AUC) values (∼0.8–0.9) in differentiating MSA-p from PSP and PD (34, 45, 46, 58, 59, 63). Furthermore, the red nucleus and globus pallidus are the two most valuable nuclei in the diagnosis of PSP (34, 46). However, direct comparisons among studies are difficult because of differences in patient characteristics (e.g., clinical severity and clinical subtypes) and methodology (e.g., MRI parameters, selected brain structures, and analysis methods) (31). The role of iron-sensitive MRI in other APS, such as diffuse Lewy body disease and corticobasal degeneration, has not yet been determined.

SWI can provide improved diagnostic accuracy through assessment of the degree and distribution of hypointensity in the putamen using a visual rating scale (**Supplementary Figure 1**) (41). A pattern of posterolateral putaminal hypointensity with a striking lateral-to-medial gradient on SWI is a highly specific sign of MSA-P (**Supplementary Figures 1D–F**). However, grading of putaminal hypointensity focused only on the signal intensity without consideration of the distributional

pattern fails to differentiate MSA-P from PD (57). Slit-like hypointensity along the lateral margin of the putamen or evenly distributed hypointensity throughout the putamen can be a nonspecific and age-related sign of physiological mineralization (**Supplementary Figures 1A–C**) (40, 41). Compared with SWI, T2- or T2<sup>∗</sup> -weighted images have limited diagnostic value with lower sensitivity (66, 67).

Because of uneven and focal accumulation of iron within the structure of the brain, subregional analyses could have higher diagnostic value than total average iron calculations (33). The posterior and inner subregion of the putamen is the most valuable among four subregional ROIs in differentiating MSA-P from PD. Moreover, high iron deposition percentages can be used to detect smaller increases in iron with higher sensitivity than average SWI phase shifts.

Multimodal MRI can identify specific markers to discriminate patients with PD from patients with MSA with high accuracy (37). The combination of T2<sup>∗</sup> relaxation rates and mean diffusivity (MD) in the putamen achieves discrimination of >95% between patients with MSA-P and PD (65).

#### Clinical Correlates

Increased iron content in motor-related subcortical nuclei may be associated with the severity of parkinsonian motor deficits. Nigral iron accumulation in PD correlated with symptoms linked to dopaminergic neurodegeneration (68). A recent study identified a significant correlation between Unified MSA Rating Scale II scores and putaminal R2<sup>∗</sup> values in patients with MSA (65). The phase-shift value of the posterior aspect of the putamen has demonstrated significantly higher asymmetry in MSA-P patients than in PD patients (60). These findings are frequently detected in the contralateral symptomatic side of patients.

In a recent study of correlations between R2<sup>∗</sup> values and the degree of Unified Parkinson's Disease Rating Scale scores in PSP, the burden of iron-related PSP pathologies in the lenticular nucleus was associated with the severity of rigidity, and nigrostriato-pallidal and dentate iron content were associated with the severity of tremors (62). Other studies have failed to show a correlation of iron-sensitive MRI markers with clinical severity scale scores in MSA and PSP (34, 43, 49, 64), but have instead reported that diffusion tensor metrics, such as MD values, are



(Continued)

#### TABLE 1 | Continued


AUC, area under the curve; CN, caudate nucleus; D, dimensional; DN, dentate nucleus; DTI, diffusion tensor imaging; FDG, Fluorodeoxyglucose; GP, globus pallidus; MD, mean diffusivity; MSA, multiple system atrophy; MSA-P, parkinsonian variant of MSA; MSA-C. cerebellar variant of MSA; MT, magnetization transfer; NC, normal control; PD, Parkinson's disease; PRESTO, principles of echo shifting using a train of observations; PSP, progressive supranuclear palsy; PT, pulvinar thalamus; PUT, putamen; QSM, quantitative susceptibility mapping; RN, red nucleus; ROI, region-of-interest; SI, signal intensity; SN, substantia nigra; SUVR, standardized uptake value; SWI, susceptibility-weighted imaging; T, tesla; TH, thalamus; T1-Vol., T1 volumetry; UMSARS, unified multiple system atrophy rating scale; UPDRS, unified Parkinson's disease rating scale; WM, white matter.

closely correlated with clinical severity (63, 64). Iron-sensitive MRI markers may have less sensitivity to reflect clinical status than diffusion tensor metrics.

Structural MRI abnormalities in MSA are segregated into basal ganglia and brainstem-cerebellar factors (64). Alterations in the R2<sup>∗</sup> of the basal ganglia region are not significantly correlated with the MRI variables associated with brainstemcerebellar degeneration. In a longitudinal study, a higher rate of iron accumulation was observed in the putamen of MSA-P patients compared to patients with the cerebellar variant of MSA and PD (49). These findings may underlie the stratification of the motor subtype of MSA. Thus, far, no attempt has been made to correlate regional iron deposition with the severity of motor deficits according to the subtypes of PSP.

A recent post-hoc analysis demonstrated that MSA-specific MRI abnormalities are associated with more rapid progression and worse overall prognosis (69). In a longitudinal study, greater baseline iron content in the putamen was correlated with smaller volumes at follow-up in MSA patients (49). Similarly, accumulation of iron in the putamen predicts volume loss in healthy older adults (70). Excessive iron accumulation in the putamen may serve as a marker for impending progression of neurodegeneration in MSA.

#### Current Limitations and Future Directions

Although recent data suggest a mechanistic link between perturbed iron homeostasis, oxidative stress, and misfolded protein aggregation, whether iron accumulation is a primary pathogenic factor or an epiphenomenon of neuronal degeneration remains to be elucidated. In a single case report, putaminal iron deposition preceded the occurrence of the initial symptoms of MSA-P (71). However, when iron deposition appears during the course of disease is unknown. A longitudinal study on the relationship between multimodal MRI abnormalities at earlier stages of the disease is necessary to identify the spatiotemporal patterns of iron-related neurodegeneration.

High-field MRI developments have advanced the ability to map the distribution of iron in vivo (1). However, studies that have investigated the pathological correlates of iron-sensitive

#### REFERENCES


MRI measurements are rare. Signal hypointensities observed using T2 or T2<sup>∗</sup> methods are correlated with iron deposits (Fe3+) and ferritin in histologic analysis (8, 48, 72). A recent study showed that R2<sup>∗</sup> is significantly associated with nigral α-synuclein. Iron-sensitive MRI may capture the pathological aspects of disorders other than iron (73). Further research is needed to verify the exact underlying pathology of MRI signal alterations.

## CONCLUSION

A distinctive pattern of iron deposition in MSA and PSP has been described. We highlighted the main factors that can determine region specificity. Disease-specific increases of iron occur in the brain regions mainly affected by underlying disease pathologies. Therefore, in vivo MRI mapping of brain iron deposition may serve as an indirect marker of neurodegenerative changes in pathoanatomically relevant sites. Age-related, structure-specific, and sub-structural patterns of iron accumulation should be considered when investigating iron-related neurodegeneration using MRI.

## AUTHOR CONTRIBUTIONS

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

### FUNDING

This research was supported by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: HI18C0713).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fneur. 2019.00074/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 © 2019 Lee and Lee. 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.

# Noxious Iron–Calcium Connections in Neurodegeneration

#### Marco Tulio Núñez<sup>1</sup> \* and Cecilia Hidalgo2,3

1 Iron and Neuroregeneration Laboratory, Department of Biology, Faculty of Sciences, Universidad de Chile, Santiago, Chile, <sup>2</sup> Calcium Signaling Laboratory, Biomedical Research Institute, CEMC, Physiology and Biophysics Program, Institute of Biomedical Sciences, Faculty of Medicine, Universidad de Chile, Santiago, Chile, <sup>3</sup> Department of Neuroscience, Faculty of Medicine, Universidad de Chile, Santiago, Chile

Iron and calcium share the common feature of being essential for normal neuronal function. Iron is required for mitochondrial function, synaptic plasticity, and the development of cognitive functions whereas cellular calcium signals mediate neurotransmitter exocytosis, axonal growth and synaptic plasticity, and control the expression of genes involved in learning and memory processes. Recent studies have revealed that cellular iron stimulates calcium signaling, leading to downstream activation of kinase cascades engaged in synaptic plasticity. The relationship between calcium and iron is Janus-faced, however. While under physiological conditions iron-mediated reactive oxygen species generation boosts normal calcium-dependent signaling pathways, excessive iron levels promote oxidative stress leading to the upsurge of unrestrained calcium signals that damage mitochondrial function, among other downstream targets. Similarly, increases in mitochondrial calcium to nonphysiological levels result in mitochondrial dysfunction and a predicted loss of iron homeostasis. Hence, if uncontrolled, the iron/calcium self-feeding cycle becomes deleterious to neuronal function, leading eventually to neuronal death. Here, we review the multiple cell-damaging responses generated by the unregulated iron/calcium selffeeding cycle, such as excitotoxicity, free radical-mediated lipid peroxidation, and the oxidative modification of crucial components of iron and calcium homeostasis/signaling: the iron transporter DMT1, plasma membrane, and intracellular calcium channels and pumps. We discuss also how iron-induced dysregulation of mitochondrial calcium contributes to the generation of neurodegenerative conditions, including Alzheimer's disease (AD) and Parkinson's disease (PD).

Keywords: neurodegenerative diseases, reactive oxygen species, mitochondria, HIF-1, Nrf-2, inflammation, ferroptosis

## INTRODUCTION

Iron and calcium ions are both essential for maintaining normal brain function. Iron is required for oxidative phosphorylation, Krebs cycle, iron–sulfur cluster and heme synthesis, synaptic plasticity, and the development of cognitive functions (Oexle et al., 1999; Bolton et al., 2000; Hidalgo and Núñez, 2007;Tong and Rouault, 2007; Zorov et al., 2007; Huang et al., 2011; Muñoz et al., 2011; Murray-Kolb, 2013; Mena et al., 2015; Requejo-Aguilar and Bolanos, 2016), while increases in cytoplasmic calcium concentration known as calcium signals mediate the secretion

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Veronica Perez de la Cruz, Instituto Nacional de Neurología y Neurocirugía (INNN), Mexico Nicola B. Mercuri, University of Rome Tor Vergata, Italy

> \*Correspondence: Marco Tulio Núñez mnunez@uchile.cl

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 16 November 2018 Accepted: 18 January 2019 Published: 12 February 2019

#### Citation:

Núñez MT and Hidalgo C (2019) Noxious Iron–Calcium Connections in Neurodegeneration. Front. Neurosci. 13:48. doi: 10.3389/fnins.2019.00048

**200**

of neurotransmitters, synaptic plasticity, and axonal growth (Berridge, 1998). Moreover, nuclear calcium signals command the expression of neuronal genes involved in learning and memory processes (Bading, 2013).

Recent studies have revealed that cellular iron stimulates neuronal calcium signaling, leading to downstream activation of kinase cascades engaged in synaptic plasticity, a neuronal response associated to memory and learning (Hidalgo and Núñez, 2007; Hidalgo et al., 2007; Muñoz et al., 2011). The relationship between calcium and iron is Janus-faced, however. Dysregulation of iron levels induces calcium dyshomeostasis and abnormal calcium signaling, whereas increased calcium levels enhance redox-active iron levels; neurodegenerative conditions exhibit both types of dysregulation. Thus, current evidence indicates that increased intracellular calcium and aberrant calcium signaling occur in neurodegenerative disorders that include Alzheimer's disease (AD), Parkinson's disease (PD), and amyotrophic lateral sclerosis (ALS) (reviewed in Pchitskaya et al., 2018). Similarly, disruption of neuronal iron homeostasis linked to iron accumulation is a common feature in a plethora of neurodegenerative diseases, which included and other parkinsonisms (Lee et al., 2013), AD (Bulk et al., 2018), Friedrich's ataxia, pantothenate kinaseassociated neurodegeneration (Swaiman, 1991), and other neuronal pathologies that entail brain iron accumulation (Wiethoff and Houlden, 2017).

Here, we will review the evidence that points to a rarely acknowledged relationship between iron homeostasis and calcium homeostasis and signaling, in which dysregulation of one of these processes results in dysregulation of the other, generating a vicious cycle that ends up in neuronal death. The multiple celldamaging responses generated by the unregulated iron/calcium self-feeding cycle, such as excitotoxicity, free radical-mediated lipid peroxidation, and the oxidative modification of crucial components of iron and calcium homeostasis and signaling, including the iron transporter DMT1, plasma membrane and intracellular calcium channels and pumps will be presented. This review article concludes with a discussion of how dysregulation of calcium homeostasis and signaling contributes to the generation of neurodegenerative diseases with an iron accumulation component, focusing in particular on AD and PD.

## IRON HOMEOSTASIS IN NEURONAL CELLS

Through the Fenton reaction, redox-active iron is a net producer of the hydroxyl radical, the most reactive radical species in nature (Davies, 2005). In a reductive environment, such as the intracellular milieu, through redox cycling redox-active iron promotes the production of hydroxyl radicals at the expense of O<sup>2</sup> and GSH consumption (Núñez et al., 2012). Consequently, under physiological conditions, redox-active iron levels that conform the labile iron pool are very tightly regulated to a cytosolic concentration of 0.5–1.5 µM, which comprises <5% of total intracellular iron (Cabantchik, 2014). In mammals, cellular iron levels are maintained by transcriptional, translational, and vesicular flux mechanisms (Núñez, 2010). The iron regulatory protein (IRPs: IRP1 and IRP2)/iron-responsive element (IRE) is the best-characterized system that regulates at the translation level the expression of iron homeostasis proteins. These proteins include the transferrin receptor 1 (TfR1, involved in iron uptake), the divalent metal transporter 1 (DMT1, involved in iron uptake), ferroportin 1 (FPN1), involved in iron efflux from cells), and ferritin (involved in iron storage) (Wilkinson and Pantopoulos, 2014). The IRP/IRE is a plentiful-oriented system, since it is activated under low iron conditions to ensure an adequate supply of iron for cellular processes. Of note, the IRE–IRP regulatory system is not only regulated by cellular iron status but it is also regulated by reactive oxygen species (ROS), whereby cells elicit a defense mechanism against iron toxicity and iron-catalyzed oxidative stress (Ray et al., 2012).

The iron uptake protein DMT1 (SLC11A2) is a crucial component of cell iron homeostasis. SLC11A2 transcription generates four alternatively spliced mRNAs that differ at their 5 0 -untranslated region (coding for the DMT1 isoforms 1A and 1B) and at its 3<sup>0</sup> -untranslated region (coding for isoforms +IRE and –IRE) (Garrick et al., 2006). Thus, the expression of the 1A and 1B isoforms of DMT1 is subjected to differential transcriptional regulation.

## THE CRUCIAL RELATIONSHIP BETWEEN IRON AND THE HYPOXIA-INDUCIBLE TRANSCRIPTION FACTOR (HIF)

At the systemic level, the hypoxia-inducible transcription factor (HIF) transcription factor family coordinates the cellular response to low oxygen levels by regulating the expression of a large array of target genes during hypoxia, which results in adaptive changes in the hematopoietic, cardiovascular, and respiratory systems (Smith et al., 2008; Mole et al., 2009). The HIF-1 is kept at basal levels by HIF prolyl hydroxylase domain (PHD) enzymes; prolyl-hydroxylation of HIF-1 via PHD signals for its degradation via the ubiquitin-proteasome system (Bruick and McKnight, 2001; Myllyharju, 2013; Yeh et al., 2017). The PHD enzymes are both oxygen- and iron-dependent; thus, hypoxia and iron chelation results in decreased PHD activity and increased HIF-1α activity (Hewitson et al., 2003; Nandal et al., 2011; Flagg et al., 2012).

In recent years, a series of iron chelating agents that exert neuroprotective effects have been developed (Núñez and Chana-Cuevas, 2018). In particular, M30, which is an 8-hydroxyquinoline-based iron chelator developed by the group of Moussa Youdim at Technion-Israel Institute of Technology (Weinreb et al., 2016), stabilizes HIF-1α, most probably by inactivating HIF-1α PHD. In the brain, HIF-1α stabilization by M30 leads to the expression of a broad number of neuroprotective-adaptive mechanisms and prosurvival signaling pathways (Kupershmidt et al., 2011). Realtime RT-PCR revealed that M30 differentially induces the expression of a variety of cellular components, including vascular

endothelial growth factor, erythropoietin, enolase-1, TfR1, heme oxygenase-1, inducible nitric oxide synthase (iNOS), glucose transporter 1, brain-derived neurotrophic factor (BDNF), glial cell-derived neurotrophic factor, and the antioxidant enzymes catalase, superoxide dismutase-1, and glutathione peroxidase (Kupershmidt et al., 2009). Further reports have supported the role of iron chelators in inducing neuronal survival pathways (Mechlovich et al., 2014; Guo et al., 2015, 2016; Xiao et al., 2015). It follows that the capacity of iron chelators to induce HIF-1α-mediated neuroprotection adds to the recognized neuroprotective effects of iron chelators, through their ability to prevent hydroxyl radical production via the Fenton reaction. A tempering note comes from the report that treatment of human skin cells with the iron chelator N-(2-hydroxybenzyl)-Lserine (HBSer) does not induce HIF-1α activation, as opposed to desferrioxamine (DFO) and salicylaldehyde isonicotinoyl hydrazone (SIH) used as positive controls (Creighton-Gutteridge and Tyrrell, 2002). The authors conjectured that the lack of HIF-1α activation by HBSer might be related to its lower affinity for iron as compared to DFO and SIH. Of relevance to the theme of this review, however, is the fact that the transcription factor HIF-1α activates the expression of several genes associated with iron homeostasis (Lee and Andersen, 2006), which in non-excitable cells results in an increase in cellular iron content (Qian et al., 2011).

### INCREASED REACTIVE OXYGEN/NITROGEN SPECIES GENERATION INDUCES IRON DYSHOMEOSTASIS

A significant number of studies have shown that physiological levels of ROS and reactive nitrogen species (RNS) act as signaling molecules in a variety of biological responses (Sen, 2001; Ray et al., 2012; Asiimwe et al., 2016; Lourenco et al., 2017; Moldogazieva et al., 2018; Nemes et al., 2018). The brain is an organ highly susceptible to oxidative stress (Cobley et al., 2018). Hence, neuronal cells have to maintain physiological levels of ROS and RNS to avoid oxidative or nitrosative stress, which arises when excessive ROS/RNS production overcomes the cellular antioxidant systems, which by affecting the redox environment favors excitotoxicity. Henceforth, we will use the broad term oxidative stress to refer to both oxidative and nitrosative stress. Oxidative stress perturbs the function of many cellular processes; it causes glutathione deficiency (Wu et al., 2004) and promotes cell death by inducing mitochondrial depolarization, membrane lipid peroxidation, and DNA damage, among other harmful effects. To counteract the damaging effects of excess ROS/RNS (Duncan and Heales, 2005; Halliwell, 2006; Moncada and Bolaños, 2006), cells make use of enzymatic and nonenzymatic defense mechanisms, some of which engage calcium signals (Gordeeva et al., 2003). Depending on the extent and persistence of the cellular damage, oxidative stress may play a key role as a causative agent of neurodegenerative disease emergence and or progression.

Iron uptake into primary hippocampal neurons stimulates ROS generation (Muñoz et al., 2011), and decreases the redox potential established by the intracellular levels of oxidized and reduced glutathione in neuroblastoma cells (Núñez et al., 2004). The relationship between increased redox-active iron and increased oxidative stress is well accepted (Núñez et al., 2012). Similarly, in cellular models convincing evidence points to a link between increased oxidative tone and iron accumulation. Initial observations indicated that hydrogen peroxide activates IRP1 (Martins et al., 1995; Pantopoulos and Hentze, 1995; Hanson and Leibold, 1999; Caltagirone et al., 2001; Mueller et al., 2001), and that this activation results in increased cellular iron uptake mediated by increased translation of TfR1 and decreased ferritin synthesis (Caltagirone et al., 2001; Sureda et al., 2005). A subsequent study reported that, besides activation of IRP1, exposure of SH-SY5Y cells to H2O<sup>2</sup> results in the degradation of the exo-transporter FPN1, which coupled to a decrease in ferritin translation leads to a time-dependent increase in the cellular labile iron pool (Dev et al., 2015), with the ensuing generation of oxidative damage through the production of hydroxyl radicals by the Fenton reaction. In animal models, evidence linking increased ROS as a cause, and iron homeostasis loss as an effect, is still lacking. An early report indicated that vitamin A deficiency results in increased hepcidin mRNA expression and increased iron spleen concentrations, accompanied by increased protein oxidation. These results suggest that vitamin A protects the liver against protein oxidation by maintaining iron homeostasis (Arruda et al., 2009). In a mouse model of experimentally induced hemolysis, excess heme increases intracellular ROS production, which signals for the induction of the unique iron exporter ferroportin, resulting in iron export from macrophages (Jeney et al., 2002; Marro et al., 2010). Of note, decreasing ROS levels by treatment with the antioxidant N-acetylcysteine prevents ferroportin induction and normalizes intracellular iron levels in a mice model of experimentally induced hemolysis (Tangudu et al., 2018). In a separate study using the APPswe/PS11E9 AD mice model, administration for 3 months of antioxidants of plant origin decreased iron levels and malondialdehyde content and increased antioxidant defenses (Zhang et al., 2018). Overall, these results are consistent with the notion that oxidative stress dysregulates iron homeostasis both in cellular and animal models.

Nitric oxide also modifies IRP1 activity; it increases IRP1 binding to IRE and the increase IRE binding activity of IRP1 is coincident with inhibited ferritin synthesis and increased TfR1 mRNA levels (Phillips et al., 1996). Nitric oxide activates IRP1 directly by destabilizing the 4Fe-4S cluster of IRP1 (Drapier et al., 1993; Soum and Drapier, 2003), and may activate IRP1 indirectly by mobilizing intracellular Fe, thus decreasing the intracellular labile iron pool and activating IRP1 (Pantopoulos et al., 1996; Wardrop et al., 2000).

Altogether, these studies point to a robust association between increased ROS, nitric oxide, and increased levels of redox-active iron, which in turn mediate oxidative damage to lipids, protein, and nucleic acids. Yet, a recent study showed that in cultured macrophages, the labile iron pool apparently attenuates peroxynitrite-dependent damage (Damasceno et al., 2018). Whether this protective effect of the labile iron pool is

translatable to other cell types and to neurons in particular, remains to be established.

#### IRON DYSHOMEOSTASIS IN NEURODEGENERATION

Increased iron levels and the associated ROS generation are important initiators and mediators of cell death (Dixon and Stockwell, 2014), whereas iron dyshomeostasis has a key role in neurodegeneration. Mutations of genes encoding proteins involved in iron homeostasis are associated with degeneration of central nervous system (CNS) cells (Ponka, 2004; Zecca et al., 2004), while oxidative stress induces cell injury by disrupting cellular iron balance (Carocci et al., 2018; Kerins and Ooi, 2018), thus leading to a vicious circle. Both AD and PD entail increased brain iron content and oxidative stress (Crichton et al., 2011; Ward et al., 2014; Tonnies and Trushina, 2017; Guo et al., 2018), with the implied threat of hydroxyl radical-induced neuronal damage. Of note, binding of Cu or Fe to Aβ peptides generates H2O<sup>2</sup> (Huang et al., 1999), which may induce neuronal oxidative stress during AD progression (Cristovao et al., 2016). Moreover, iron overload increases neuronal amyloid-β production and enhances cognitive impairment in a transgenic mice model of AD (Becerril-Ortega et al., 2014) while accumulating evidence suggests that impaired iron homeostasis is an early event in AD progression (Peters et al., 2015).

Parkinson's disease is a progressive neurodegenerative disease characterized by rigidity, tremor, and slowness of movement as well as by a wide range of debilitating non-motor symptoms. A key feature of PD is the selective loss of the dopaminergic neurons of the substantia nigra; ample evidence shows iron accumulation in these neurons (Jellinger, 1999; Zecca et al., 2004; Berg and Youdim, 2006; Muñoz et al., 2016; An et al., 2018). The observation that pharmacological agents with iron chelation capacity prevent neuronal death in AD or PD experimental models (recently reviewed in Núñez and Chana-Cuevas, 2018) highlights the pivotal role of iron as a mediator of neuronal death in AD and PD.

In addition, spinal neurons from ALS patients display increased iron and calcium levels (Kasarskis et al., 1995). The success of iron chelation therapy in ALS mouse models sustains the central role of iron in ALS pathogenesis (Jeong et al., 2009; Kupershmidt et al., 2009). Alterations of proteins involved in iron metabolism, inhibition of anterograde axonal transport leading to iron accumulation in ventral motor neurons, and increased mitochondrial iron load in neurons and glia have been proposed as pathogenic mechanisms to explain the abnormal iron accumulation displayed by the neurons and glia of ALS mice (Jeong et al., 2009).

## IRON DYSHOMEOSTASIS IN INFLAMMATION

Inflammation, a condition often found in neurodegenerative environments, is yet another factor that induces iron accumulation (reviewed in Urrutia et al., 2014). Both the p50 and the p65 subunits of the transcription factor NF-κB bind to a NFκB-responsive element present in the 5<sup>0</sup> untranslated region of the DMT1-1B promoter, inducing the expression of the 1B isoforms of DMT1 (Paradkar and Roth, 2006a,b). The transcription factor NF-κB is often activated by inflammatory signals (Rothwell and Luheshi, 2000; Hanke and Kielian, 2011; Shih et al., 2015). Accordingly, the possible association between inflammation and iron accumulation has been recently explored (Pelizzoni et al., 2013; Urrutia et al., 2013, 2014; McCarthy et al., 2018). To this purpose, the effects of lipopolysaccharide (LPS) and the pro-inflammatory cytokines TNF-α and IL-6 on mRNA and protein levels of DMT1, FPN1, and hepcidin were investigated in astrocytes, microglia, and neurons isolated from rat brain. A causal association between inflammation and iron accumulation was reported, since in addition to increasing neuronal iron content, treatment with LPS, TNF-α, or IL-6 increases DMT1 expression and protein levels in neurons (Urrutia et al., 2013). Similar results were reported in studies of ventral mesencephalic neurons in primary culture, in which treatment with TNF-α or IL-1β induces an increment in DMT1 and TfR1 protein levels, together with a reduction of FPN1 levels (Wang et al., 2013). Considering that NFκB activation is downstream of the TNF-α, IL-1, and LPS signaling pathways, these results are consistent with a circuit in which inflammatory stimuli induce DMT1 expression and iron accumulation via NFκB activation and hepcidin expression.

An additional link between inflammation and increased oxidative stress resides in the observation that inflammatory stimuli activate the NADPH oxidase (NOX) enzyme, the main regulated source of cellular ROS generation (Green et al., 2001; Urrutia et al., 2014; Sorce et al., 2017). In particular, proinflammatory cytokines activate the phagocytic NOX2 enzyme, which is highly expressed in microglia (Sorce and Krause, 2009; Lim et al., 2013; Barrett et al., 2017). In turn, NOX2 activation generates an oxidative extracellular environment that increases ROS levels in neighboring neurons (Schiavone et al., 2009; Gao et al., 2012; Surace and Block, 2012; Zhang et al., 2014).

Of note, NOX activation has also been reported in experimental models of PD and AD. In particular treatment with MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine), a prodrug to the Parkinsonian neurotoxin 1-methyl-4 phenylpyridinium (MPP+), results in increased synthesis of the pro-inflammatory cytokine IL-1β, and increased membrane translocation of the cytosolic NOX sub-unit p67phox, which is prevented by minocycline, a tetracycline derivative that exerts multiple anti-inflammatory effects (Wu et al., 2002). In addition, iron supplementation to midbrain neuron-glia cell cultures causes the selective death of dopaminergic neurons. This death is mediated by microglial NOX2 since cells derived from NOX2−/− mice do not present ironinduced neuronal death (Zhang et al., 2014). Similarly, brain tissue from AD patients reveals an elevation in NOX activity coupled to membrane translocation of cytosolic NOX2 subunits and increased NOX1 and NOX3 mRNA transcripts (Shimohama et al., 2000; Ansari and Scheff, 2011). Significantly, the authors found an inverse correlation between

postmortem NOX activity and ante-mortem cognitive status (Ansari and Scheff, 2011). Altogether, these results indicate that inflammatory stimuli generate a neuronal phenotype of increased iron influx and increased NOX-produced ROS, with the consequent risk of oxidative damage observed in neurodegenerative diseases.

# ROLE OF IRON IN EXCITOTOXICITY

Earlier reports indicate that iron plays a role in excitotoxicity, a death process mediated by high levels of intracellular calcium caused by excessive activity of excitatory neurotransmitters. Seminal observations by Cheah and colaborators described a signaling cascade in which stimulation of N-methyl-D-aspartate (NMDA) receptors activates the small GTPase Dexras1 via S-nitrosylation, mediated by activation of neuronal nitric oxide synthase (nNOS). Activated Dexras binds to the peripheral benzodiazepine receptor-associated protein PAP7, which in turn binds to the iron transporter DMT1 (Cheah et al., 2006). Most importantly, treatment with neurotoxic concentrations of NMDA elicits a major increase in iron uptake, resulting in threeto fivefold increase of hydroxyl radical levels and the death of more than 90% of cells, whereas co-treatment with NMDA and a cell-permeant iron chelator largely blocked NMDA-induced cell death. Overall, these results indicate that glutamate-NMDA excitotoxicity elicits DMT1 activation and iron-mediated cell death (Cheah et al., 2006). Although the role of calcium was not explored, the authors hypothesized initial calcium entrance through the NMDA receptor may activate nNOS, thus initiating this neurodegenerative process. As discussed below, crosstalk between NMDA receptor-mediated calcium influx and iron is likely to have a relevant role in neurodegenerative diseases such as AD and PD.

## NEURONAL CALCIUM HOMEOSTASIS AND SIGNALING

Calcium is a universal second messenger that regulates numerous cellular processes over a wide temporal range (Berridge et al., 2003). In neurons, the free calcium concentration at rest lies in the 70–100 nM range (Clapham, 2007), whereas extracellular calcium concentration is ∼1.5 mM. This large concentration gradient, combined with resting potential values of 70–90 mV (negative inside), generate a significant electrochemical gradient that favors calcium entry. Calcium homeostasis and signaling are vital for cellular function and survival; thus, neuronal cells possess powerful systems to maintain calcium homeostatic and signaling systems (Berridge, 1998; Brini et al., 2014), which regulate the neuronal processes underlying synaptic plasticity and complex brain cognitive functions, such as information processing, learning, and memory. Calcium is stored also in intracellular organelles such as the endoplasmic reticulum (ER), where its free concentration lies in the 0.5 mM range. The main systems in charge of maintaining calcium homeostasis are two plasma membrane calcium (PMCA) transporters, the PMCA pump and the sodium/calcium exchanger (NCX), plus the sarco/ER Ca2+-ATPase (SERCA) pump that maintains the steep concentration gradient between the ER and the cytoplasm (Brini et al., 2014). These transporters utilize different mechanisms and display a range of calcium affinities, transport rates, and capacities to ensure the proper managing of the variety of intracellular increases in free calcium concentration, known as calcium signals, generated by the diverse stimuli neuronal cells receive.

Neuronal activity generates calcium signals that via sequential activation of calcium-dependent signaling cascades and transcription factors promote the expression of genes underlying dendritic development, synaptic plasticity, and neuronal survival (Brini et al., 2014). In response to neuronal stimulation, calcium influx through plasma membrane pathways, including agonistoperated, store-operated, or voltage-gated calcium channels (VGCCs), generates neuronal calcium signals. In addition, the release of calcium from intracellular stores such as the ER also contributes to neuronal calcium signal generation (Brini et al., 2014). All neuronal compartments crucial for neurotransmission have the two types of calcium release channels: the inositol 1,4,5-trisphosphate (IP3) receptor (IP3R) and the ryanodine receptor (RyR) channels (Berridge, 1998). These large channels have complex gating and regulation and mediate a cellular process known as calcium-induced calcium release (CICR), whereby calcium promotes its own release from intracellular calcium stores (Berridge, 1998). This CICR mechanism likely contributes to propagate calcium signals that reach distant targets such as the nucleus (Berridge, 1998), where calcium promotes short or long-lasting changes in neuronal function and structure (Bading, 2013).

#### ROS LEVELS INFLUENCE CALCIUM HOMEOSTASIS AND SIGNALING

Significant evidence gathered in the last years indicates that increased cellular ROS levels modify the function of key proteins engaged in calcium homeostasis and signaling (Gordeeva et al., 2003; Hidalgo and Donoso, 2008; Gorlach et al., 2015). Redoxdependent regulation of components that maintain neuronal Ca2<sup>+</sup> homeostasis may influence the direction and/or the efficiency of Ca2+-signaling pathways. Interactions among ROS and calcium-dependent signaling pathways are bidirectional; while ROS regulate cellular calcium signaling, calcium signals are essential for ROS production because a number of ROSgenerating and antioxidant systems of living cells are calciumdependent (Gordeeva et al., 2003).

Reversible modifications of proteins via ROS-promoted oxidation of their amino acids modify the properties of neuronal proteins engaged in signal transduction, such as protein kinases, protein phosphatases, and transcription factors (Wang et al., 2012). An increase in cellular ROS levels results in the inhibition of NMDA receptor function and in decreased activity of both the PMCA and the SERCA calcium pumps, while it enhances the activity of IP3R and RyR calcium channels (Hidalgo and Donoso, 2008; Joseph et al., 2018). Consequently, ROS generation in neuronal cells increases cellular calcium levels by promoting

calcium release mediated by IP3R and RyR channels and by inhibiting both the PMCA and the SERCA pumps: the ROS-mediated increase in calcium levels promotes changes in several calcium-dependent neuronal pathways (Yermolaieva et al., 2000; Riquelme et al., 2011). Depending on the magnitude of the calcium increase, activation of kinases or phosphatases implicated in synaptic plasticity process will occur (reviewed in Hidalgo and Arias-Cavieres, 2016). In addition, activation of the large-conductance Ca2+-activated K<sup>+</sup> (BKCa) by oxidation of SH cysteine residues, by decreasing neuronal excitability reduces the vulnerability of hippocampal neurons to hypoxia (Hepp et al., 2005).

#### DEFECTIVE CALCIUM HOMEOSTASIS AND SIGNALING IN NEURODEGENERATION

A link between excessive intracellular calcium accumulation and neuronal death is by now firmly established (Demuro et al., 2010; Stefani et al., 2012; Fukui, 2016). Neuronal cells are particularly sensitive to calcium-mediated cytotoxic damage and the failure of neurons to maintain Ca2<sup>+</sup> homeostasis is a common feature of aged-linked neurodegenerative pathologies (Berrocal et al., 2015; Brini et al., 2017; Pchitskaya et al., 2018; Strehler and Thayer, 2018). Calcium dyshomeostasis (Pchitskaya et al., 2018) and altered neuronal calcium signaling (Chan et al., 2009; Overk et al., 2015) have been reported in a number of disease conditions, including AD, PD, and ALS (for recent reviews, see Franco-Iborra et al., 2018; Popugaeva et al., 2018; Post et al., 2018; Sirabella et al., 2018). Likewise, defective calcium signaling in glia and autophagy-related pathways has been implicated in neurodegenerative disease (Mustaly-Kalimi et al., 2018). Accruing evidence implicates decreased levels and functional decline of PMCA isoforms in neurodegeneration (Hajieva et al., 2018; Strehler and Thayer, 2018). In addition, in many pathological conditions, disruptions of ER Ca2<sup>+</sup> signaling disrupt neuronal function and promote neuronal cell death (Okubo et al., 2018). To date, however, the detailed mechanisms of how neuronal Ca2<sup>+</sup> homeostasis and signaling are perturbed in neurodegenerative diseases are not well understood.

The most common neurodegenerative disorder and the leading cause of dementia in the elderly is AD, whereas PD is the most prevalent age-associated movement disorder that entails early and selective degeneration of dopaminergic neurons. The pathological hallmarks of AD are neuronal loss and the presence of amyloid plaques and neurofibrillary tangles in the brain of affected individuals. Yet, before the emergence of plaques and/or neurofibrillary tangles other defects, including Ca2<sup>+</sup> and iron dyshomeostasis, mitochondrial dysfunction, increased contacts between the ER and the mitochondria, neuroinflammation, and alterations in lipid metabolism have been reported (Mattson, 2010; Area-Gomez and Schon, 2017; Lane et al., 2018). How these early defects relate to the emergence and progression of AD remains to be established. Of note, a recent report by the Alzheimer's Association Calcium Hypothesis Workgroup [AACHW] (2017) analyzed how defective calcium signaling may be an early upstream event in AD progression, and how alterations of the many cellular components that control calcium hemostasis and signaling leads to the decline in the function of more vulnerable neuronal cells. In particular, these more vulnerable neuronal cells present early synaptic dysfunction, upregulation of some calcium signaling related genes and significant decreases in the expression of genes related to synaptic neurotransmission, neurotrophic factor activity, mitochondrial metabolism, and energy production, among others (Alzheimer's Association Calcium Hypothesis Workgroup [AACHW], 2017).

In PD, dopaminergic neurons of the substantia nigra are especially vulnerable to Ca2<sup>+</sup> dyshomeostasis, because their autonomous pace-making activity, through which these neurons generate action potentials in the absence of synaptic input (Guzman et al., 2009). This particular property of dopaminergic neurons imposes a high pressure on adequate handling of activity-generated Ca2<sup>+</sup> signals, presumably mediated by the significant calcium influx through plasma membrane L-type calcium channels that markedly surpasses the calcium influx displayed by other neurons (Surmeier et al., 2010).

Dopaminergic neurons are highly vulnerable to oxidants and mitochondrial inhibition compared to other neuronal types (Sherer et al., 2003; Halliwell, 2006; Burbulla et al., 2017). Thus, brain permeant toxins such as rotenone, an inhibitor of NADH dehydrogenase, exclusively affect dopaminergic neurons (Sherer et al., 2003; Cannon et al., 2009). This special vulnerability is likely to produce the mitochondrial Ca2<sup>+</sup> imbalance that occurs in PD (Ludtmann and Abramov, 2018). As discussed below, iron-induced ROS increases result in calcium signal generation, which may contribute to the basal calcium burden of dopaminergic neurons.

Several studies have reported defective Ca2<sup>+</sup> signaling in other neurodegenerative conditions such as ALS (Siklos et al., 1998; Grosskreutz et al., 2010; Jaiswal, 2013; Muhling et al., 2014) and multiple sclerosis (MS; Gonsette, 2008). The death of motor neuron is a characteristic feature of ALS. Excitotoxicity, which involves glutamate and calcium overload, or cell death caused by high levels of intracellular calcium caused by excessive activity of excitatory neurotransmitters, may contribute to ALS pathology. Calcium dysregulation is also present in MS, where the associated disability results from neuronal and axonal loss initiated by microglia activation and mediated by oxidative stress, excitotoxicity, calcium dysregulation, and mitochondrial dysfunction, leading to proteolytic enzyme production and apoptosis (Gonsette, 2008).

Additionally, calcium dysregulation has been associated with neuroinflammation, a CNS response that comprises biochemical and cellular responses to injury, infection, or neurodegenerative diseases (DiSabato et al., 2016). Neuroinflammation markers increase in some neurodegenerative conditions; microglia, the innate CNS immune cells, play key roles as mediators of these neuroinflammatory responses. Of note, Ca2<sup>+</sup> and neuroinflammatory signaling mechanisms exhibit extensive crosstalk and bidirectional interactions. As an example, chronic brain inflammation leads to a feedback reduction in NMDA

receptors caused by excess synaptic glutamate activity during microglial activation (Rosi et al., 2004).

Despite the abundant literature that points to defective calcium homeostasis and signaling in neurodegeneration, scarce information is available in the literature regarding the protective effects of calcium modulators in preventing these defects. Yet, all recent reports on this subject highlight the fact that this is an urgent task. The non-competitive NMDA receptor antagonist Memantine, which reduces excitoxicity by inhibiting neuronal calcium entry (Bezprozvanny and Mattson, 2008), has been the most common drug for treating moderate to severe AD; nevertheless, Memantine provides only limited benefits for AD patients. A recent report showed that immunotherapy with a murine analog of the anti-Aβ antibody Aducanumab restores calcium homeostasis in a transgenic AD mice model (Kastanenka et al., 2016), whereas a recent review addresses the evidence linking neurodegeneration to RyR dysfunction and describes novel therapeutic approaches to target abnormal RyR function (Kushnir et al., 2018). There are no reports, however, on the efficacy of these strategies for treating AD patients.

#### THE HIF-1–CALCIUM RELATIONSHIP

In SH-SY5Y neuroblastoma cells, chelation of calcium by 1,2 bis(2-aminophenoxy)ethane-N,N,N',N'-tetraacetic acid (BAPTA) induces HIF-1α protein accumulation and nuclear localization, whereas the cytosolic calcium elevation produced by inhibition of the SERCA pump attenuates these effects (Berchner-Pfannschmidt et al., 2004). Further experiments indicated that calcium chelation attenuates the interaction of HIF-1α with the von-Hippel-Lindau protein, thus decreasing the proteasomal degradation of HIF-1α (Berchner-Pfannschmidt et al., 2004). Control experiments indicated that BAPTA did not complex intracellular iron ions, thus the above effects were caused by calcium chelation and not by stimulation of ironactivated PHD enzymes (Berchner-Pfannschmidt et al., 2004). Subsequent studies showed that intermittent hypoxia causes HIF-1α accumulation through a mechanism that involves increased cell calcium and increased generation of ROS by NOX, which induces a complex signaling pathway that results in increased HIF-1α synthesis and decreased hydroxylase-dependent HIF-1α degradation (Yuan et al., 2008).

Although brain levels of HIF-1α are considered neuroprotective (Kietzmann et al., 2001; Siddiq et al., 2005; Guo et al., 2016; Lee et al., 2017; Merelli et al., 2018), under certain conditions, HIF-1α acts as a pro-apoptotic factor (Piret et al., 2002; Zhang et al., 2007; Merelli et al., 2018). For example, the anesthetic isoflurane induces apoptotic neurodegeneration in a process mediated by HIF-1α, since knockdown of HIF-1α expression attenuates isoflurane-induced neurotoxicity (Jiang et al., 2012). In addition, isoflurane induces a significant elevation of cytoplasmic calcium levels in cultured neurons. Based on previous observations, showing that calcium mediates HIF-1α expression (Yuan et al., 2005; Riganti et al., 2009), the authors hypothesized that the high calcium levels induced by isofluorane promote HIF-1α-mediated apoptotic neurodegeneration. Studies discerning the calcium concentrations and other cell physiological condition, which elicit each of the above responses, should shed light on these apparent contradictory responses.

Overall, the emerging picture from the above results is consistent with the idea that low calcium concentrations stabilize HIF-1α protein levels by decreasing its proteasomal degradation. The effect of calcium concentrations above homeostatic levels is not clear and may depend on cell type: high calcium levels may induce a protective response consisting in increased HIF-1α synthesis coupled to decreased degradation or may mediate HIF-1α degradation.

# IRON MODIFIES NEURONAL CALCIUM HOMEOSTASIS AND CALCIUM SIGNALING PATHWAYS

Iron overload increases intracellular calcium levels, which in turn activate calcineurin (Lee et al., 2016), a calcium-dependent phosphatase which has a key role in synaptic plasticity and memory processes (Baumgartel and Mansuy, 2012). Increases in cytoplasmic calcium levels via the NMDA receptor stimulate the neuronal NOS enzyme (Ghosh and Salerno, 2003) and generate superoxide anion in hippocampal neurons (Brennan et al., 2009), while in cortical neurons, this calcium increase induces a signaling cascade that enhances superoxide generation through NOX activation (Girouard et al., 2009).

Both synaptic plasticity and memory processes require activity-dependent hippocampal ROS generation (Bindokas et al., 1996; Kahlert et al., 2005; Kishida and Klann, 2007). In primary hippocampal neurons, the ROS increase induced by iron promotes calcium release from the ER mediated by redoxsensitive RyR channels; the resulting increase in intracellular calcium levels link NMDA receptor stimulation to ERK1/2 activation and nuclear translocation (Muñoz, 2012). Since several reports indicate that CREB-dependent transcription of synaptic plasticity related genes entails ERK1/2-mediated longterm CREB phosphorylation (Wu et al., 2001; Lynch, 2004; Thomas and Huganir, 2004). Accordingly, these findings indicate that iron-induced, RyR-mediated, calcium release contributes to this activity-dependent neuronal response.

In summary, the current evidence suggests that crosstalk between ROS and calcium plays an essential role in many pathophysiological conditions including neurodegenerative diseases such as PD and AD. Here we complement this idea by adding the component of iron overload, and the ensuing oxidative stress, as a significant factor in promoting calcium dysregulation in neurodegeneration.

#### THE SYNERGISM BETWEEN IRON AND CALCIUM IN LIPID PEROXIDATION

A link between iron-mediated lipid peroxidation and calcium dyshomeostasis was reported more than 30 years ago. Peroxidation of rat brain synaptosomes by Fe2<sup>+</sup> and

H2O<sup>2</sup> generates rapid and large uptake of Ca2<sup>+</sup> by isolated synaptosomes; Fe2<sup>+</sup> also enhances Ca2<sup>+</sup> uptake by spinal cord neurons in culture, an effect that is coupled to lipid peroxidation and the release of arachidonic acid from cells, in a process prevented by the iron chelator DFO (Braughler et al., 1985). The authors concluded that Fe2<sup>+</sup> and Ca2<sup>+</sup> act synergistically to produce lipid peroxidation and damage to neuronal membranes (Braughler et al., 1985). A further report proposed that this is not a direct effect of calcium but the result of independently initiated processes of lipid peroxidation and Ca2<sup>+</sup> translocation, which interact subsequently in a synergistic manner (Bors et al., 1990).

The relationship between iron-induced lipid peroxidation and dysregulation of Ca2+-ATPase is long-dated (Pelizzoni et al., 2008; Mata and Sepulveda, 2010; Zaidi, 2010). Plasma membrane lipid peroxidation initiated by the Fenton reaction results in increased calcium content, derived from the inhibition of the plasma membrane and ER Ca2+-ATPases (Pereira et al., 1996). This inhibitory effect is shared by other radical species, in particular by peroxynitrite (Zaidi and Michaelis, 1999). Oxidants also inhibit SERCA activity while promoting calcium release from the ER (Gordeeva et al., 2003; Hidalgo and Donoso, 2008). Arguably, the observed upsurge in intracellular Ca+<sup>2</sup> induced by iron-mediated lipid peroxidation could be compounded by an oxidantinduced loss of PMCA and SERCA activity, which combined with ROS-mediated activation of IP3R/RyR channels would result in further and deleterious increases in the intracellular Ca2<sup>+</sup> concentration.

# IRON, CALCIUM, AND HIF-1 ACTIVITY

Considering the overlapping effects of iron and calcium on HIF-1 activity detailed above, a Fe/Ca bipartisan regulation of HIF-1 can be hypothesized (**Figure 1**). In neurodegenerative diseases with an iron accumulation component, decreased HIF-1 activity is predicted by a mechanism that involves the maintenance of maximal HIF-1 hydrolase activities and the increase of intracellular calcium mediated by dysregulation of redox-sensitive calcium homeostatic and signaling pathways. In neurons, decreased HIF-1 activity results in decreased neuroprotective, neuroregenerative, and antioxidant responses associated to HIF-1 activity (see above).

## CALCIUM-INDUCED IRON DYSHOMEOSTASIS: POSSIBLE ROLE IN NEURODEGENERATION

Excessive intracellular calcium levels, such as those associated with neurodegenerative conditions, are likely to produce high H2O<sup>2</sup> levels via NOX2 and to enhance mitochondrial calcium uptake and ROS production. These two ROS sources, NOX2, and mitochondria should affect jointly iron homeostasis by increasing the cellular labile iron pool, as occurs in SH-SY5Y cells exposed to H2O<sup>2</sup> (Dev et al., 2015). These findings raise the possibility that conditions that increase postsynaptic calcium levels following NMDA receptor activation, which promote H2O<sup>2</sup> generation in neocortical neurons via a cascade engaging the NOS and NOX2

activity of HIF-1 hydrolases and, through enhanced ROS production increase the cellular levels of calcium by activation of RyR, IP3R, and L-type calcium channels.

enzymes (Girouard et al., 2009), may lead to an upsurge of the labile iron pool.

The role of voltage-gated Ca2<sup>+</sup> channels as mediators of iron transport into neuronal cells is well established (Gaasch et al., 2007; Lockman et al., 2012). Another link between iron and calcium during neurodegeneration was provided by the observation that the hippocampal and nigral neuron loss induced by intra-cerebroventricular FeCl<sup>3</sup> injection in rats is largely prevented by pre-treatment with nicardipine, a blocker of L-type VGCCs (Bostanci and Bagirici, 2013). The authors concluded that these channels mediate the neurotoxic effects of extracellular iron by inducing either calcium and/or iron uptake. The question remains as to whether the protective effect of nicardipine is due to decreased iron influx, decreased calcium influx, or to a calciummediated process leading to neurodegeneration downstream of iron-induced oxidative stress. Other studies have also provided evidence of iron influx through NMDA receptors or voltageoperated calcium channels in primary hippocampal neurons (Pelizzoni et al., 2011) and the dopaminergic MES23.5 cell line (Wang et al., 2017).

## THE MITOCHONDRION MELTING POT: CALCIUM, IRON, AND APOPTOTIC CELL DEATH

Mitochondria, by taking up and releasing calcium, act as important regulators of cellular calcium levels. The large mitochondrial membrane potential difference (negative inside) provides the driving force for calcium entry through the mitochondrial calcium uniporter (MCU), a component of the inner mitochondrial membrane; in physiological conditions, calcium antiporters rapidly extrude calcium from the mitochondria and restore the basal calcium levels (Giorgi et al., 2018). Mitochondrial calcium concentration modulates ATP production and mitochondrial metabolism, whereas mitochondrial calcium overload induced by deregulated intracellular calcium levels promotes apoptotic cell death, necrosis, and autophagy (Giorgi et al., 2012; Rimessi et al., 2013; Marchi et al., 2018). In pathological conditions, the MCU has been implicated in excitotoxicity, iron overload, inflammation, and oxidative stress-induced mitochondrial dysfunction and cell death (Liao et al., 2017). The ROS generated by mitochondrial enzymes or the electron transport chain presumably regulate diverse cellular physiological pathways, whereas dysregulated ROS generation may contribute to the development of human diseases (Angelova and Abramov, 2017).

Neurodegenerative disorders such as AD, PD, and ALS display dysregulation of calcium homeostasis and mitochondrial dysfunction; in addition, excitotoxicity leads to mitochondrial calcium overload and promotes necrotic or apoptotic-like excitotoxic cell death (Hansson Petersen et al., 2008; Rosales-Corral et al., 2012; Bose and Beal, 2016; Gibson and Thakkar, 2017; Giorgi et al., 2018). An increase in mitochondrial calcium is of particular importance in dopaminergic neurons, which are constantly exposed to calcium influx (Surmeier et al., 2011). The interaction between calcium overload and excessive ROS production (Chinopoulos and Adam-Vizi, 2006) causes further damage to the electron transport chain and promotes oxidative stress, which induces the oxidation of mitochondrial lipids, proteins, and DNA leading to cytotoxicity (Muravchick and Levy, 2006). The vicious cycle between calcium overload and oxidative stress favors the sustained opening of the mitochondrial permeability transition pore (mPTP), leading to the collapse of the mitochondrial membrane potential and mitochondrial swelling, which in turn promote the apoptotic response (Di Lisa and Bernardi, 2009). Opening of the mPTP has a central role in the pathogenesis of neurodegenerative disorders (Vila and Przedborski, 2003; Perier et al., 2005).

Excessive iron levels promote oxidative stress leading to the upsurge of unrestrained calcium signals that damage mitochondrial function. Iron accumulation in the cytoplasm induces excessive iron influx into mitochondria through mitoferrin1/2 (Paradkar et al., 2009), which causes mitochondrial dysfunction by increased ROS production leading to oxidative damage. Hence, if uncontrolled, this calcium/iron self-feeding cycle becomes deleterious to mitochondrial and neuronal function, leading eventually to neuronal death. In this regard, it is of note that, by preserving mitochondrial calcium homeostasis, the iron-binding protein lactoferrin protects vulnerable dopamine neurons from degeneration (Rousseau et al., 2013).

In primary hippocampal neurons, iron-induced stimulation of redox-sensitive RyR-mediated calcium release causes significant mitochondrial fragmentation (Sanmartin et al., 2014). Iron overload induces oxidative stress in the ER earlier than in the mitochondria, thereby increasing ER stress and calcium levels, and consequently causing mitochondrial fragmentation and neuronal cell death (Lee et al., 2016). Mitochondrial fragmentation presumably contributes to the impairment of neuronal function produced by iron overload, since inhibition of mitochondrial division protects neuronal cells from excitotoxicity (Ruiz et al., 2018). Moreover, iron overload of HT-22 hippocampal neuron cells increases intracellular calcium levels, causes mitochondrial fragmentation via dephosphorylation of Drp1, and increases apoptotic neuronal death; of note, calcium chelation and inhibition of calcineurin prevented mitochondrial fragmentation and cell death (Lee et al., 2016). Based on these findings, the authors suggested that calcium-mediated calcineurin signaling, by regulating mitochondrial dynamics, has a key role in iron-induced neurotoxicity.

In summary, mitochondrial calcium overload combined with excessive iron-mediated ROS generation and mPTP opening compose a cell death-inducing sequence implicated in neuronal dysfunction and neurodegeneration.

#### OXYTOSIS/FERROPTOSIS A POSSIBLE MEETING POINT FOR CALCIUM AND IRON-INDUCED CELL DEATH

Oxytosis was first described as a glutamate-induced, calciumdependent form of cell death (Murphy et al., 1988). Initial studies on the molecular mechanisms of oxytosis pointed to a decrease

in intracellular glutathione levels, enhanced mitochondrial ROS production, lipid peroxidation, and massive calcium influx as causes of death (Miyamoto et al., 1989). The mechanistic link between glutamate exposure and glutathione depletion was the inhibition of the cysteine/glutamate antiporter xc- by glutamate, which deprived cells of cystine, a precursor in glutamate synthesis (Murphy et al., 1989). As glutathione levels decrease, ROS levels increase exponentially (Tan et al., 1998), which gives rise to the activation of signaling pathways that culminate in large influx of calcium and cell death (Maher et al., 2018). Thus, the increased calcium influx is an essential mediator of the cell death program, as cells do not die in calcium-free medium.

Ferroptosis is a form of regulated cell death that occurs because of lethal lipid peroxidation. Ferroptotic cell death is blocked by iron chelators, lipophilic antioxidants, and by agents that block propagation of lipid peroxidation (Stockwell et al., 2017; Fricker et al., 2018). Ferroptosis is induced by inhibition of the cystine-glutamate antiporter xc- or by inhibition of glutathione peroxidase 4 (Gpx4). Induction of ferroptosis by erastin (an inhibitor of the antiporter xc-) results in cell death, which is morphologically characterized by mitochondrial shrinkage, with no overt signs of apoptosis or necrosis (Dixon et al., 2012). Ferroptotic death derives from depletion of cell glutathione, increased oxidative stress, and iron-mediated lipid peroxidation to lethal levels (Stockwell et al., 2017). Inhibition of Gpx4 also causes ferroptosis (Yang et al., 2014). Gpx4 is a lipid peroxidase that directly reduces phospholipid hydroperoxides at the expense of reduced glutathione. In mice, conditional

Increases in ROS due to mitochondrial dysfunction, inflammation, and NOX activation promote excessive ROS generation that promotes calcium release via RyR channels and increases redox-active iron by IRP1 activation. Through the Fenton reaction, iron induces lipid peroxidation that inhibits the PMCA calcium pump. In addition, excitotoxic conditions over-activate NMDA receptors (NMDAR), which results in increased cytoplasmic calcium levels and increases iron levels via DMT1 activation.

deletion in forebrain neurons of Gpx4 resulted in hippocampal neurodegeneration and behavior dysfunction, associated with markers of ferroptosis such as elevated lipid peroxidation, ERK1/2 activation, and neuroinflammation (Hambright et al., 2017). Given the large similarity in their cell death mechanisms, the current view is that oxytosis and ferroptosis are similar forms of cell death (Fricker et al., 2018; Lewerenz et al., 2018; Maher et al., 2018).

Although a lethal influx of calcium is an intrinsic feature of oxytosis, the participation of calcium in ferroptosis is still poorly defined. Initial work by Dixon et al. (2012) revealed that calcium chelators do not inhibit erastin-induced HT-1080 cell death. Nevertheless, as mentioned above, an increase in cellular ROS levels enhances the activity of IP3R and RyR calcium release channels and decreases the activity of both the PMCA and the SERCA calcium pumps. These effects increase cellular calcium levels by promoting calcium release from the ER while inhibiting its removal by the calcium pumps.

Lipid peroxidation-derived modification of some calcium channels could also be intermediates in ferroptotic death (Maher et al., 2018). In particular, 4-hydroxynonenal, a product of hydroxyl radical-mediated lipid peroxidation, induces the opening of NMDA receptor and VGCCs (Lu et al., 2002). Indeed,

the store-operated calcium entry (SOCE) system, which activates calcium influx channels in the plasma membrane upon the release of intracellular calcium from the ER participates in the neuronal death induced by MPP+, a Parkinsonian toxin that inhibits mitochondria electron transport chain complex I. Inhibition of the SOCE response decreased apoptotic cell death, reduced ROS production, and decreased lipid peroxidation induced by MPP+ (Li et al., 2013). In the same vein, a recent report shows that the pharmacological or transcriptional inhibition of SOCE proteins protects against oxidative glutamate toxicity and against MPP+ injury (Maher et al., 2018).

Overall, these results point to a possible participation of calcium in ferroptosis, a participation driven by raises in cytoplasmic calcium because of oxidative- and lipid peroxidation-derived modifications of the calcium homeostasis machinery (**Figure 3**). This participation may be specific to cells with weak antioxidant defenses, which could allow for the oxidative activation of calcium channels.

# PROTECTION OF IRON OVERLOAD BY NRF2; THE CALCIUM CONNECTION

Under oxidative stress conditions, the transcription factor known as nuclear factor (erythroid-derived 2)-like 2 (Nrf2) directs the expression of a large array of cytoprotective genes (Tonelli et al., 2018). In particular, Nrf2 protects cells from the injurious effects of iron overload by regulating the expression of genes involved in iron storage and iron export (Chorley et al., 2012; Kerins and Ooi, 2018). These findings raise the possibility that Nrf2 may counteract the oxidative stress induced in neuronal cells by iron overload. Thus, Nrf2 may restore the intracellular labile iron pool by inducing the expression of the iron exporter FPN1 and ferritin, which oxidizes Fe2<sup>+</sup> to Fe3<sup>+</sup> and stores it within its structure making it unavailable for the Fenton reaction (Orino et al., 2001). In addition, intracellular iron levels act as a link between HIF-1α and Nrf2: Nrf2-induced ferritin upregulation activates HIF-1α by making iron unavailable to HIF prolyl-hydroxylase (Siegert et al., 2015), whereas Nrf2 inhibition results in decreased activation of HIF-1α by hypoxia (Kim et al., 2011; Ji et al., 2014).

Recent studies point to a connection between ROS, cellular calcium signals, and Nrf2. The neurotrophin BDNF, besides inducing complex neuronal signaling cascades underlying synaptic plasticity (Bolton et al., 2000), also induces neuronal antioxidant responses through BDNF-induced Nrf2 nuclear translocation (Bouvier et al., 2017; Bruna et al., 2018). In this regard, a recent study showed that BDNF-mediated activation of Nrf2 in astrocytes, which is under circadian control, protects dopaminergic neurons from ferroptosis (Ishii et al., 2018). Of note, in primary hippocampal neurons, ROS and RyR-mediated calcium signals are required for BDNF-induced nuclear translocation of Nrf2 (Bruna et al., 2018). It remains to be established if iron overload in hippocampal neurons engages BDNF and RyR-mediated calcium signals to promote Nrf2-induced expression of protective genes.

accumulation. Excess iron promotes lipid peroxidation by hydroxyl radicals derived from the Fenton reaction. Lipid peroxidation modifies the activity of a variety of proteins involved in calcium homeostasis that results in massive calcium influx. Increased cytoplasmic calcium levels result in increased mitochondrial calcium and accentuated mitochondrial dysfunction, oxidative stress, and damage. If uncontrolled, this ROS–iron–calcium self-feeding cycle becomes deleterious to mitochondrial and neuronal function, leading eventually to apoptotic or ferroptotic cell death. In yet another aspect of this cycle, high intracellular calcium levels increase the rate of HIF-1 proteasome degradation, which results in increased ROS levels, product of decreased expression of antioxidant proteins.

# CONCLUSION

fnins-13-00048 February 8, 2019 Time: 19:37 # 12

This article reviews the little acknowledged relationship between iron and calcium in neurodegeneration. The main premise of this article is that increased ROS production generated by iron accumulation profoundly influences calcium homeostasis and signaling (**Figure 4**).

The influence of increased ROS is clearly shown in the process of oxytosis, in which lipid peroxidation initiated by Fenton-derived hydroxyl radicals results in the modification of a variety of proteins involved in calcium homeostasis that brings to calcium upsurges and cell death. Increased cytoplasmic calcium levels caused by iron dysregulation result in increased mitochondrial calcium and accentuated oxidative stress and damage. If uncontrolled, this calcium/iron self-feeding cycle becomes deleterious to mitochondrial and neuronal function, leading eventually to neuronal death. In yet another aspect of neurodegeneration, high intracellular calcium levels increase the rate of HIF-1 proteasomal degradation while high levels of iron both maintain the activity of HIF-1 hydrolases and, through oxidative stress

#### REFERENCES


increase the cellular levels of calcium by activation of ER and PMCA channels. The main conclusion we draw from this review is that dysregulated iron or calcium levels promote deleterious crosstalk between iron and calcium that can result in neuronal dysfunction and death. To our knowledge, this is the first attempt in literature to systemize this important relationship.

#### AUTHOR CONTRIBUTIONS

MN and CH wrote this article.

# FUNDING

This work was supported by FONDEF project 17I10095 from the National Commission of Science and Technology (CONICYT) of Chile; Fondo Nacional de Desarrollo Científico y Tecnológico (FONDECYT; Grants Nos. 1170053 and 11140580), and by Biomedical Neuroscience Institute (Grant No. BNI P09-015-F).




hydrogen peroxide through metal ion reduction. Biochemistry 38, 7609–7616. doi: 10.1021/bi990438f




synaptosomal [Ca2+]i increase during oxidative stress. Brain Res. 713, 269–277. doi: 10.1016/0006-8993(95)01554-X


uptake in primary hippocampal neurons. Front. Mol. Neurosci. 7:13. doi: 10. 3389/fnmol.2014.00013



**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 Núñez and Hidalgo. 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 Potential Role of Ferroptosis in Neonatal Brain Injury

Yanan Wu<sup>1</sup> , Juan Song<sup>1</sup> , Yafeng Wang<sup>1</sup> , Xiaoyang Wang1,2, Carsten Culmsee1,3,4 and Changlian Zhu1,5 \*

<sup>1</sup> Henan Key Laboratory of Child Brain Injury, Third Affiliated Hospital of Zhengzhou University, Zhengzhou, China, <sup>2</sup> Institute of Neuroscience and Physiology, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden, <sup>3</sup> Institute of Pharmacology and Clinical Pharmacy, University of Marburg, Marburg, Germany, <sup>4</sup> Center for Mind, Brain and Behavior, University of Marburg, Marburg, Germany, <sup>5</sup> Center for Brain Repair and Rehabilitation, Institute of Neuroscience and Physiology, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden

Ferroptosis is an iron-dependent form of cell death that is characterized by early lipid peroxidation and different from other forms of regulated cell death in terms of its genetic components, specific morphological features, and biochemical mechanisms. Different initiation pathways of ferroptosis have been reported, including inhibition of system X<sup>c</sup> −, inactivation of glutathione-dependent peroxidase 4, and reduced glutathione levels, all of which ultimately promote the production of reactive oxygen species, particularly through enhanced lipid peroxidation. Although ferroptosis was first described in cancer cells, emerging evidence now links mechanisms of ferroptosis to many different diseases, including cerebral ischemia and brain hemorrhage. For example, neonatal brain injury is an important cause of developmental impairment and of permanent neurological deficits, and several types of cell death, including iron-dependent pathways, have been detected in the process of neonatal brain damage. Iron chelators and erythropoietin have both shown neuroprotective effects against neonatal brain injury. Here, we have summarized the potential relation between ferroptosis and neonatal brain injury, and according therapeutic intervention strategies.

Keywords: hypoxic ischemic brain injury, intraventricular hemorrhage, cell death, iron toxicity, lipid peroxidation, neonate, reactive oxygen species, lipid peroxidation

#### INTRODUCTION

Neonatal brain injury is an important cause of developmental impairment and of permanent neurological deficits such as cerebral palsy in children. Among many etiological factors, hypoxic– ischemic encephalopathy in term infants and intraventricular/periventricular hemorrhage in preterm infants are the most common causes of neonatal brain damage (Bennet et al., 2012; Shankaran et al., 2014). The progression of brain injury depends on the balance between persistent injury and the repair response, which can be modulated by therapeutic intervention (Kaandorp et al., 2012; Azzopardi et al., 2016; Song et al., 2016). Emerging evidence indicates that there is great potential for improving the treatment of acute brain injury in these children as well as opportunities for more effective regenerative treatment of these patients (Juul and Ferriero, 2014; O'Gorman et al., 2015; Tagin et al., 2015). Today, cooling the body is the only established method of treating newborns after asphyxia, but the protective effects are limited and only moderately injured children benefit from this treatment (Azzopardi et al., 2014). It is imperative, therefore, that we continue our

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Qitao Ran, The University of Texas Health Science Center at San Antonio, United States Teresa Gasull, Health Sciences Research Institute of the Germans Trias i Pujol Foundation (IGTP), Spain

#### \*Correspondence:

Changlian Zhu zhuc@zzu.edu.cn; changlian.zhu@neuro.gu.se

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 02 November 2018 Accepted: 30 January 2019 Published: 14 February 2019

#### Citation:

Wu Y, Song J, Wang Y, Wang X, Culmsee C and Zhu C (2019) The Potential Role of Ferroptosis in Neonatal Brain Injury. Front. Neurosci. 13:115. doi: 10.3389/fnins.2019.00115

**218**

efforts to identify the mechanisms of injury and repair in the developing brain and to identify new therapeutic strategies. An improved understanding of the mechanisms of brain injury is needed in order to develop strategies for the next generation of treatments for brain injuries in both term and preterm infants.

Neuronal cell death after an insult takes several different forms, but the underlying mechanisms of neuronal death share common features and signaling pathways. However, the cell death mechanisms in the developing brain have been shown to be quite different compared to those in the adult brain (Zhu et al., 2005; Wang et al., 2009). Neuronal cell death can be classified into accidental and regulated forms. Accidental cell death is induced by severe insults that cause immediate cellular demise, usually necrosis, and these cells cannot be rescued (Galluzzi et al., 2018). In contrast, regulated cell death has been differentiated into many types defined by morphological and/or biochemical features, including apoptosis, necroptosis, autophagy, pyroptosis, eryptosis, and more recently also ferroptosis (Galluzzi et al., 2018). Ferroptosis was first described in RAS mutant cancer cells in 2012 and has been defined as an iron and lipid peroxidationdependent form of cell death that is genetically, biochemically, morphologically, and mechanistically distinct from other types of cell death (Dixon et al., 2012; Stockwell et al., 2017).

Studies elucidating the mechanisms of different ferroptosis inducers have found that system X<sup>c</sup> <sup>−</sup> and glutathione-dependent peroxidase 4 (GPX4) inhibition or depletion can trigger ferroptosis through reduced glutathione (GSH) levels and subsequent accumulation of reactive oxygen species (ROS), especially through enhanced lipid peroxidation (Seiler et al., 2008; Stockwell et al., 2017; Seibt et al., 2018). Ferroptosis has been found not only in cancer cells, but also in dying neurons in model systems of neurological disorders (Weiland et al., 2018; Wu et al., 2018). Prior to the discovery of ferroptosis, the iron chelator deferoxamine (DFO) had been shown to have great potential in protecting brain cells from death (Lee et al., 2011; Masaldan et al., 2018); however, the mechanism behind this activity remained unknown, but recent studies on ferroptosis might provide an explanation (Dixon et al., 2012; Xie Y. et al., 2016; Morris et al., 2018). Programmed cell death is important for normal development, and such activity is more pronounced in neonates than in adults (Wang et al., 2009). Furthermore, metabolic iron imbalance is common and antioxidant capacity is low in neonates; and this suggests that ferroptosis might be important in neonates under pathological conditions. However, there are as yet no reports or reviews linking ferroptosis and neonatal brain injury. Here, we summarize recent advances in our understanding of ferroptosis, and we discuss the potential relationship between ferroptosis and neonatal brain injury.

# DEFINITION AND DISCOVERY OF FERROPTOSIS

Ferroptosis is a non-apoptotic form of cell death that depends on cellular iron and ROS (Dixon et al., 2012), and ferroptosis inducers had been applied before the mechanisms of ferroptosis were first proposed. Erastin, a synthetic compound, was discovered in Dolma et al. (2003). They found that erastin could cause non-apoptotic cell death in cells that expressed an engineered mutant Ras protein, but not in their wildtype counterparts. Later in 2008, they used high- throughput screening of small-molecule libraries to identify two Ras-selective lethal small molecules (RSL3 and RSL5) that induce nonapoptotic and iron-dependent oxidative cell death, and they found that the cell death could be inhibited by the iron chelator desferrioxamine mesylate and by the antioxidant vitamin E (Yang and Stockwell, 2008). The mode of cell death induced by RSL3 was found to be non-apoptotic because these cells showed no apoptotic hallmarks and because such cell death still occurred in cells where the factors of the core apoptosis machinery – i.e., caspases, BAX, and BAK – were suppressed (Yang and Stockwell, 2008; Wolpaw et al., 2011; Dixon et al., 2012). In 2012, Dr. Stockwell coined the term "ferroptosis" to describe this newly discovered form of cell death (Dixon et al., 2012). The morphological features of ferroptosis include an intact cell membrane without ruptures or blebs, normal nucleus size, and a lack of chromatin condensation, but with a shrinking mitochondrial membrane that shows increased membrane density and reduced or complete absence of mitochondrial cristae as well as outer mitochondrial membrane rupture (Dixon et al., 2012; Xie Y. et al., 2016; Doll and Conrad, 2017). Ferroptosis inducers can be roughly divided in two classes, one is represented by system X<sup>c</sup> <sup>−</sup> inhibitors such as erastin and sulfasalazine and the other acts through GPX4 inhibition such as RSL3 (Stockwell et al., 2017). Accompanied with the discovery of inhibitors of ferroptosis, the veil of ferroptosis has gradually been lifted.

## METABOLIC PATHWAYS AND MOLECULAR MECHANISMS IN FERROPTOSIS

#### Inhibition of System X<sup>c</sup> <sup>−</sup> Triggers Ferroptosis

System X<sup>c</sup> <sup>−</sup> is a heterodimeric cystine/glutamate antiporter composed of SLC7A11 (xCT) and SLC3A2, and it is responsible for maintaining redox homeostasis by importing cystine into the cell where it is further reduced to cysteine for synthesizing the major antioxidant GSH. Cysteine is the rate-limiting factor in cellular GSH biosynthesis because this amino acid is relatively rare in food. It has been shown that the cysteine-glutathione pathway is a pivotal upstream signaling regulator of ferroptosis (Dixon et al., 2014). Impairment of the system X<sup>c</sup> <sup>−</sup>-dependent antioxidant defense system results in oxidative injury and cell death. The small molecule erastin has been shown to inhibit system X<sup>c</sup> <sup>−</sup> activity and lead to the accumulation of lipid ROS and thus to trigger ferroptosis (Dixon et al., 2012, 2014). Several other agents also induce ferroptosis through inhibiting system X<sup>c</sup> <sup>−</sup>, such as glutamate, sulfasalazine and sorafenib (Dixon et al., 2014) (**Figure 1**).

Glutamate is exchanged for cystine in a 1:1 ratio by system Xc <sup>−</sup>, thus excess glutamate acts as an equivalent of erastin and can also induce ferroptosis, which perhaps provides new insight into

FIGURE 1 | Hypothetical mechanisms of hypoxia ischemia and intracranial hemorrhage-induced ferroptosis in the immature brain. Following intracranial hemorrhage, the lysed erythrocytes release Hb, which produces a degradation product, heme, that is then degraded by HO-1 into carbon monoxide CO, biliverdin, and free iron. Excess Fe2+, the reactive form of iron, will cause membrane lipid damage and ferroptosis. The selective ferroptosis inhibitors Fer-1 and Lip-1 inhibit lipid peroxidase activity and inhibit ferroptosis. Iron chelators such as DFO can reduce the level of unbound iron and inhibit the production of ROS and the occurrence of ferroptosis. EPO reduces the level of unbound iron by promoting erythropoiesis. The inactive form of iron (Fe3+) is recognized by TF and delivered into the cell by TfR1 and stored in endosomes, where Fe3<sup>+</sup> is then converted into Fe2<sup>+</sup> by STEAP3. Free iron (Fe2+) can be transported by DMT1 out of the endosome. Some of the free iron is stored in ferritin, some is transported out of the cell by ferroportin, and some might cause lipid ROS. Free iron can be released from ferritin degradation via the ferritinophagy pathway, which is mediated by NCOA4. These proteins work together to maintain the balance of iron as well as to control cell fate to some content. Erastin impedes cystine transport by inhibiting system Xc−, while Nrf acts in an opposing role by upregulating system Xc<sup>−</sup> transcription. Once inside the cell, cystine is reduced to cysteine, and the level of cysteine can be supplemented by the trans-sulfuration pathway. Subsequently, cysteine is used for the biosynthesis of GSH. GPX4 uses two GSH molecules as electron donors to reduce phospholipid hydroperoxides (PL-OOH) to the corresponding alcohols leaving GSSG (oxidized GSH) as a byproduct. ACSL4 is required for activation of PUFAs, especially arachidonic acid (AA) and adrenic acid (AdA) to AA-CoA and AdA acyl co-A derivatives. These derivatives are esterified by LPCAT3 into AA-PE and AdA-PE, which then be catalyzed by the iron-containing enzyme lipoxygenase (LOX) to generate fatty acid hydroperoxides. After hypoxia-ischemia insult, the acidulated environment will cause the accumulation of excess iron and glutamate. Accumulated glutamate will inhibit system Xc−, which in turn will lead to insufficient levels of cellular cysteine and GSH production. As a result, GPX4 will be inactivated and lead to lipid peroxidation and ferroptosis. RSL3 is a ferroptosis activator that binds to and inactivates GPX4. During hypoxia insult, HIF is stabilized by PHDs and is translocated into the nucleus where it turns on transcription of EPO. EPO stimulates the activity of GPX4 and inhibits lipid peroxides and thus acts as an inhibitor of ferroptosis. PHD inhibitors like DFO and AQ not only inhibit PHDs, but also inhibit ATF4, which is a ferroptosis activator. AA/AdA, arachidonic acid/adrenic acid; ACSL4, acyl-CoA synthetase long chain family member 4; AIF, apoptosis inducing factor; AQ, adaptaquin; ATF4, activating transcription factor 4; CO, carbon monoxide; DFO, deferoxamine; DMT1, Divalent metal transporter 1; EPO, Erythropoietin; Fer-1, ferrostatin-1; GPX4, glutathione peroxidase 4; GR, glutathione reductase; GSSG, oxidized GSH; GSH, reduced glutathione; Hb, hemoglobin; HO-1, heme oxygenase 1; HIF, hypoxia-inducible factor; Lip-1, liproxstatin-1; LPCAT3, lysophosphatidylcholine acyltransferase 3; PE: phosphatidylethanolamine; PHDs, prolyl-hydroxylases; PL-PUFA (PE), polyunsaturated-fatty-acid-containing phospholipids; PL-PUFA(PE)-OOH, polyunsaturated-fatty-acid-containing- phospholipid hydroperoxides; PL-PUFA(PE)-OH; PUFA, polyunsaturated fatty acid; ROS, reactive oxygen species; RSL3, RAS-selective lethal 3; STEAP3, 6-transmembrane epithelial antigen of the prostate 3; TF, transferrin; TfR1, transferrin receptor 1; VDAC, Voltage-dependent anion-selective channel protein.

explaining the mechanism of glutamate toxicity in the nervous system. Previous studies have shown that glutamate, at millimolar concentrations, can also block X<sup>c</sup> <sup>−</sup> activity in fibroblasts and in a neuronal cell line, thereby exerting pathways of oxidative cell death that involve GSH depletion and lipid peroxidation (Tan et al., 1999; Tobaben et al., 2011; Neitemeier et al., 2017). This form of cell death in immortalized hippocampal neurons (HT22 cell line) was previously defined as oxytosis (Tan et al., 2001), and more recent data demonstrated similarities between ferroptosis and oxytosis at the level of X<sup>c</sup> <sup>−</sup> inhibition, GSH depletion, and Gpx4 inhibition and in mitochondrial pathways of oxidative cell death in neurons that involve activation of the pro-apoptotic BID

and the mitochondrial release of apoptosis inducing factor (AIF) (Neitemeier et al., 2017; Jelinek et al., 2018). The mitochondrial mechanisms merged the previously separated pathways of oxytosis and ferroptosis in neuronal cells, and this is supported by data demonstrating the involvement of AIF in cell death induced by genetic Gpx4 depletion (Seiler et al., 2008), and the particular role of mitochondrial AIF translocation to the nucleus of damaged neurons in models of neonatal hypoxia/ischemia (Zhu et al., 2003) and in transient focal cerebral ischemia (Plesnila et al., 2004; Culmsee et al., 2005). It is well established that glutamate toxicity plays an important role in neuronal cell death in neonatal brain injury (Zhu et al., 2010; Jantzie et al., 2015). The neurotoxicity of glutamate in differentiated neurons and brain tissue is dependent on excitotoxic (Tobaben et al., 2011) disruption of the Ca2<sup>+</sup> homeostasis and oxidative stress mediated by oxidative iron species, and this can be inhibited by iron chelators or ferroptosis-specific inhibitors, including 12/15 LOX inhibition (Neitemeier et al., 2017). These findings demonstrate the close relationship between glutamate toxicity, system X<sup>c</sup> − inhibition, lipid peroxidation, and ferroptosis.

In addition to glutamate, recent studies have found other biological molecules that regulate ferroptosis through system Xc <sup>−</sup>. A recent study reported that beclin1 promotes ferroptosis by directly blocking system Xc<sup>−</sup> activity. Mechanistically, beclin1 is phosphorylated by AMP activated protein kinase at Ser 90/93/96 and then interacts with SLC7A11, a core component of system Xc−, to prevent cysteine transport and glutathione formation (Song et al., 2018). SLC7A11 is also a target of p533KR , and p533KR transcriptionally downregulates SLC7A11 expression and induces ferroptosis upon cysteine uptake limitation and ROS induction (Jiang et al., 2015). Moreover, transcription factor Nrf2 upregulates system Xc<sup>−</sup> in various cancers, and this increases the redox-sensitivity of the cell and induces ferroptosis (Sun X. et al., 2016).

#### Inactivation of GPX4 Induces Ferroptosis

Elucidating the mechanism of RSL3-mediated cell death provided further major insight into the regulation of ferroptosis. Cell death induced by erastin and RSL3 share common features of ferroptosis, such as a dependence on iron and ROS; however, RSL3 toxicity is dependent on either the voltage-dependent anion-selective channel protein 2/3 (VDAC2/VDAC3) (Yagoda et al., 2007) or system X<sup>c</sup> <sup>−</sup> indicating that it functions through a different initiating mechanism (Yang et al., 2014). Analysis of affinity-based chemoproteomics with LC-MS and western blotting confirmed the interaction between GPX4 and RSL3 (Yang et al., 2014), and further study found that RSL3 covalently interacts with the selenocysteine in the active site of GPX4 to inhibit its enzymatic activity (Yang et al., 2016). Knockdown of GPX4 by shRNA induces ferroptosis, while overexpression of GPX4 renders cells resistant to RSL3 toxicity (Yang et al., 2014), and this suggests that different initiating mechanisms converge to a similar form of ferroptotic cell death. GPX4, which uses GSH as an essential cofactor, catalyzes the reduction of hydrogen peroxide and organic hydroperoxides, especially lipid-hydroperoxides, to water and the corresponding alcohols, respectively (**Figure 1**).

# Role of ROS and Lipid Peroxidation in Ferroptosis

Erastin induces ferroptosis by inhibiting system X<sup>c</sup> <sup>−</sup> and thus inducing the accumulation of ROS, which have been shown to be lipid ROS according to flow cytometry assays using the fluorescent probes dichlorofluorescein (DCF), H2DCFDA, and C11-BODIPY (Yagoda et al., 2007; Dixon et al., 2012). Further support for the involvement of lipid ROS in ferroptosis comes from a study showing that RSL3 induces ferroptosis by binding to and inhibiting GPX4, thereby increasing lipid peroxidation as measured by C11-BODIPY fluorescence (Yang et al., 2014).

Several lipophilic antioxidants have been identified as strong suppressors of erastin-induced cell death, including α-tocopherol, butylated hydroxytoluene, β-carotene, ferrostatin-1 (Fer-1), and liproxstatin-1. Fer-1 and liproxstatin-1 are regarded as specific ferroptosis suppressors because they suppress the cell death induced by ferroptosis inducers (erastin, RSL3) and fail to save cells from apoptosis or necroptosis induced by staurosporine and H2O2, respectively (Dixon et al., 2012; Yang et al., 2014).

Lipidomics analysis revealed that polyunsaturated fatty acids (PUFAs) are the most susceptible lipids to peroxidation in the course of ferroptosis compared to other classes of lipids (Yang et al., 2016). A recent study identified acyl-CoA synthetase longchain family member 4 (ACSL4) as an essential component of ferroptosis through a genome-wide CRISPR-based genetic screen and a microarray analysis of ferroptosis-resistant cell lines. ACSL4-enriched cellular membranes with long polyunsaturated omega-6 fatty acids and GPX4−/ACSL4<sup>−</sup> double-knockout cells show marked resistance to ferroptosis (Doll et al., 2017). Mechanistically, ACSL4 is required for activation of PUFAs, especially arachidonic acid (AA) and adrenic acid (AdA) to AA-CoA and AdA acyl co-A derivatives (Hirschhorn and Stockwell, 2018). Next, lysophosphatidylcholine acyltransferase 3 (LPCAT3) can esterify these derivatives into PE (phosphatidylethanolamine) to form AA-PE and AdA-PE, which then be catalyzed by an ironcontaining enzyme lipoxygenase (LOX) to generate fatty acid hydroperoxides in a stereospecific manner (Stockwell et al., 2017; Hirschhorn and Stockwell, 2018) (**Figure 1**). Lipid peroxidation seems play a role in the final stage of ferroptosis, and this is evidenced by pronounced upregulation of aldo-keto reductase family 1 member C family genes in cell lines that are resistant to erastin-induced ferroptosis (Dixon et al., 2014; Stockwell et al., 2017). The products of these genes can detoxify the end products of oxidized PUFAs, such as 4-hydroxynonenal which are likely produced by the oxidative lipid fragmentation processes that occur during the execution of ferroptosis (Stockwell et al., 2017).

# The Role of Iron in Ferroptosis

Although the exact role of iron in ferroptosis remains enigmatic, there is considerable evidence that iron is a necessary component in this form of oxidative cell death (Dixon et al., 2012; Li et al., 2017). It has been shown that erastin induces ferroptosis through the accumulation of ROS in the cytosol as measured by flow cytometry using the fluorescent probe DCF when incubation with exogenous sources of iron rather than other divalent

transition metal ions (Cu2+, Mn2+, Ni2+, and Co2+), and this process is suppressed by co-treatment with the iron chelator DFO (Dixon et al., 2012). Therefore, the state of cellular free iron seems to be important to ferroptosis induction. Consistently, studies have demonstrated that proteins that control cellular iron balance, such as transferrin, ferritin, ferroportin, and other iron-containing proteins for iron uptake, storage, utilization, and degradation participate in promoting ferroptosis. In fact, treatment with recombinant iron-loaded rather than iron-free transferrin promoted ferroptotic cell death (Gao et al., 2015). Moreover, iron chelation, depletion of transferrin from serum, and knockdown of TfR1 prevents erastin-induced ferroptosis. Erastin-induced cell death can also be inhibited by Fer-1, a specific ferroptosis inhibitor (Dixon et al., 2012), which also blocks cell death triggered by either full amino acid starvation or cystine starvation (Gao et al., 2015). Inhibition of ferritin autophagy, referred to as ferritinophagy, through blockage of autophagy or knockdown of cargo receptor NCOA4 (nuclear receptor coactivator 4) which recruits ferritin to autophagosomes for lysosomal degradation, abrogates the accumulation of labile iron and ROS, and prevents ferroptosis (Santana-Codina and Mancias, 2018) (**Figure 1**). In addition, iron response element binding protein 2 was identified as an essential gene for the induction of ferroptosis due to its role in the regulation of iron metabolism and iron accumulation (Dixon et al., 2012). Collectively, these results strengthen the view that iron is necessary in initiating ferroptosis.

Although iron is necessary for ferroptosis, it remains unclear how iron acts in ferroptosis. One hypothesis is that iron acts as free iron and generates hydroxyl radicals or hydroperoxyl radicals through the Fenton reaction, which is an important source of ROS (Dixon and Stockwell, 2014). Another hypothesis is that iron functions in ferroptosis as a cofactor of iron-containing enzymes. As mentioned before, lipoxygenases are a family of iron-containing enzymes that mediate PUFA oxidation and exert a lethal effect under conditions of GSH depletion (Yang et al., 2016). Other studies have found that other iron-containing enzymes – the hypoxia-inducible factor prolyl hydroxylases (HIF-PHDs) – are related to the process of ferroptosis (Speer et al., 2013; Karuppagounder et al., 2016) (**Figure 1**). Also, adaptaquin, a specific inhibitor of HIF-PHD enzymes, reduced neuronal death and behavioral deficits after intracranial hemorrhage (ICH) in a rodent model without affecting total iron or zinc distribution in the brain (Karuppagounder et al., 2016). The protection from oxidative death in vitro or from ICH in vivo by adaptaquin was associated with suppression of the activity of activating transcription factor 4 (ATF4) rather than activation of a HIF-dependent pro-survival pathway.

#### Other Molecules Regulate Ferroptosis

Several other metabolic pathways and molecules regulate ferroptosis sensitivity. As a limited building block of GSH, the level of cysteine acts as the upstream signal of ferroptosis. Cysteine starvation and inhibition of system X<sup>c</sup> <sup>−</sup>-induced ferroptosis can be rescued by the trans-sulfuration pathway (biosynthesis of cysteine from methionine) in some cells. Cysteinyl-tRNA synthetase (CARS) was recently discovered to be involved in this pathway, and knockdown of CARS increases intracellular free cysteine and inhibits erastin-induced ferroptosis (Hayano et al., 2016; Stockwell et al., 2017). However, cysteine deficiency does not induce the generation of lipid peroxidation and ferroptosis when there is a lack of glutamine or when there is inhibition of glutaminolysis (Gao et al., 2015; Stockwell et al., 2017). Glutamine is a major cellular energy source and can provide elements for biosynthesizing amino acids, nucleic acids, and lipids by generating intermediates through glutaminolysis. Glutaminase 1 (GLS1) and glutaminase 2 (GLS2) both catalyze glutamine into glutamate as the first reaction of glutaminolysis, but only suppression of GLS2 prevents ferroptosis, which is transcriptionally controlled by the P53 P47S variant (Jennis et al., 2016). Mevalonate-derived antioxidant coenzyme Q10 (CoQ10), which is derived from the mevalonate pathway, is a negative regulator of ferroptosis by reducing the accumulation of lethal lipid peroxidation induced by FIN56 (Shimada et al., 2016) (**Figure 1**). Many other molecules and metabolic pathways need to be explored.

#### NEONATAL BRAIN INJURY

Neonatal brain injury is a major public health issue and is a leading cause of neonatal mortality and morbidity, especially in preterm infants. Neonatal brain injury is not a single welldefined entity, and many factors contribute to such injury, but the most common etiologies are hypoxic–ischemic encephalopathy in term infants and intraventricular/periventricular hemorrhage in preterm infants (Gale et al., 2018). Brain injury evolves over time and goes through different stages, and multiple mechanisms contribute to this process, including energy depletion, excitatory amino acids, mitochondrial impairment, generation of ROS, and inflammation, all of which lead to different types of cell death (Hagberg et al., 2014; Sun et al., 2017; Albertsson et al., 2018; Davidson et al., 2018; Nazmi et al., 2018). Apoptosis and necrosis have been identified as the two main mechanisms of cell death in many different variants of brain injury (Li et al., 2010; Zhu et al., 2010; Northington et al., 2011; Thornton et al., 2017), but more and more studies have demonstrated that different forms of cell death occur simultaneously or successively (Sun Y. et al., 2016; Xie C. et al., 2016; Sun et al., 2017). After the discovery of ferroptosis, recent studies have also demonstrated connections between ferroptosis and neurological diseases (Tonnus and Linkermann, 2016; Hambright et al., 2017; Zille et al., 2017).

Compared to the adult brain, the neonatal brain has a high rate of oxygen consumption, high concentrations of unsaturated fatty acids, and low concentrations of antioxidants, which make it particularly sensitive to oxidative damage (Blomgren et al., 2003). The PUFA content of the brain increases during gestation and indicates that the preterm brain is even more sensitive to lipid peroxidation than the term brain and that lipid peroxidation might be a major factor in the white-matter damage seen in preterm infants who suffer from brain injury (Millar et al., 2017). Furthermore, the brain's endogenous antioxidant defense mechanisms show less activity in the immature brain compared

to the mature brain (Lafemina et al., 2006). Altogether, this suggests that the immature brain is more sensitive to oxidative stress-induced cell death and brain injury. Because perinatal hypoxia and ICH are two dominant causes of neonatal brain injury, we focus on the potential contribution of ferroptosis on asphyxia and ICH-induced neonatal brain injury.

# Ferroptosis and Peripartum Asphyxia

Despite important progress in obstetric and neonatal care in recent years, perinatal asphyxia is still one of the leading causes of death and adverse developmental outcomes (Zhu et al., 2009; Azzopardi et al., 2016; Liu et al., 2016). Perinatal hypoxicischemic insult-induced cell death peaks at 24–48 h, but this pathological process continues for weeks after injury (Geddes et al., 2001; Fleiss and Gressens, 2012), and this long-lasting cell death causes significant loss of brain volume as indicated by cerebral MRI (Wideroe et al., 2011). Further studies that delayed the chronic phase of cell death for several days after acute injury reduced brain injury and tissue loss volume (Han et al., 2014; Xie et al., 2014). Several types of cell death have been shown to be involved in neonatal brain injury. Apoptosis, necrosis, and autophagy are the three types of cell death that have been commonly identified (Zhu et al., 2005; Li et al., 2010; Xie C. et al., 2016), but with the identification of ferroptosis, new potential mechanisms of cell death should be taken into consideration.

Excitotoxicity, oxidative stress, and inflammation play important roles in the mechanism of neonatal hypoxic-ischemic (HI) brain injury (Li et al., 2011; Bi et al., 2015). HI results in depolarization of neurons and glial cells and the subsequent release of excitatory amino acids such as glutamate into the extracellular space. Elevated glutamate has been documented in the cerebrospinal fluid of infants who have suffered severe HI injury (Pu et al., 2008). Accumulated glutamate activates the NMDA receptors that mediate normal brain development and function by promoting the proliferation and migration of neuronal precursors, synaptic development, and plasticity (Rocha-Ferreira and Hristova, 2015), but excessive glutamate is also a risk factor for triggering ferroptosis (**Figure 1**). Studies have shown that GSH is reduced after hypoxia in the neonatal periventricular white matter and in primary cultured oligodendrocytes when exposed to hypoxia or to conditioned medium from hypoxic microglial cells (Kaur et al., 2010). Such glutamate toxicity-mediated X<sup>c</sup> <sup>−</sup> inhibition and GSH depletion is also established in a model system of oxytosis in immortalized hippocampal neurons, the HT22 cell line, which is widely used to study mechanisms of oxidative cell death involving enhanced lipid peroxidation. As mentioned before, recent data demonstrated that glutamate-mediated X<sup>c</sup> − inhibition in oxytosis resembles the major features of ferroptosis. Thus, glutamate toxicity in vivo might also involve ferroptosis mechanisms similar to those described for erastin and RSL3 in cancer cells.

Iron is another dangerous factor during HI injuries. Iron homeostasis in the brain involves the regulation of iron movement between the blood and the brain, between the intracellular and extracellular spaces in the brain, and between different iron pools within such spaces (Ke and Qian, 2007; Garton et al., 2016). This homeostasis is maintained by a series of iron transport proteins (e.g., transferrin) and iron storage proteins (e.g., ferritin). Under normal physiological conditions, most of the brain iron is sequestered within storage proteins because protein-bound iron is safe while free iron can generate highly damaging reactive molecules, which can cause damage to proteins and nucleotides and lipid peroxidation via Fenton chemistry. During the actual HI event, protein-bound iron is liberated from its binding proteins due to the low intracellular pH. After HI, the rapid accumulation of iron is seen in damaged neurons and in the periventricular white matter of neonatal rats (Rathnasamy et al., 2011), and increased levels of iron-bound proteins IRP1, IRP2, and TfR, and accumulated free iron are seen in the serum and cerebrospinal fluid of human infants (Shouman et al., 2008). Intracerebral injection of the lipid-soluble form of transferrin (apotransferrin) attenuates white matter damage in a neonatal rat model of cerebral HI, which suggests that overloaded free iron can protect cells from oxidative stress (Guardia Clausi et al., 2012). Indeed, after HI, production of free radicals is activated, which then attack unsaturated fatty acids and lead to the production of neurotoxic lipid peroxides (LPOs). Rodent models have shown abundant formation of LPOs in the brain after HI, and the serum levels of malondialdehyde, which is an end-product of lipid peroxidation, increase in neonates after HI (El Bana et al., 2016). Serum LPO concentrations increase significantly in asphyxiated infants, and this correlates with the clinical grading of hypoxic–ischemic encephalopathy and mortality (Ramy et al., 2016). The level of LPO, as well as the severity of cell damage, can be reduced by the iron chelator DFO (Peeters-Scholte et al., 2003; Rathnasamy et al., 2011).

Ferroptosis has been reported in ischemic injuries in other tissues in vivo and in vitro, and different iron chelators and their analogs have shown great potential in preventing these injuries. Treatment with DFO reduces the infarct size in hearts suffering from ischemia/reperfusion stress (Gao et al., 2015) and protects against acute renal failure and organ damage in a model of severe kidney ischemia/reperfusion injury (Linkermann et al., 2014). A study in a newborn mouse model found that iron pretreatment aggravates periventricular cystic whitematter lesions (Dommergues et al., 1998), and this has been proposed to be related to iron overload-induced oxidative stress (Gong et al., 2016). Erythropoietin (EPO) is thought to have neuroprotective effects through multiple mechanisms, including neurotrophic, anti-oxidant, anti-apoptotic, and antiinflammation activities and the promotion of neural stem cell proliferation and differentiation (Wang et al., 2004; Juul and Pet, 2015). Furthermore, EPO promotes erythropoiesis, which increases iron utilization and potentially reduces free iron (Bany-Mohammed et al., 1996). Preterm infants who received EPO treatment had significantly reduced serum ferritin accompanied by reduced serum lipid peroxidation (Akisu et al., 2001), and our clinical studies have shown that EPO treatment reduces neurological sequelae in both term and preterm infants (Zhu et al., 2009; Song et al., 2016). During a hypoxic insult, HIF is stabilized by PHDs and translocates into the nucleus where it turns on transcription of EPO. Previous studies have shown that EPO stimulates the activity of GPX4 and inhibits lipid peroxides

and thus might act as an inhibitor of ferroptosis (Genc et al., 2002) (**Figure 1**). The neuroprotective effect of EPO might be related to reduced levels of unbound iron and oxidative stress and thus to reduced levels of ferroptosis (Bailey et al., 2014).

# Ferroptosis and Intracranial Hemorrhage in Preterm Infants

Intracerebral hemorrhage is one of the most common complications in preterm infants, and its rate of occurrence increases with decreasing birth weight and decreasing gestational age (Bolisetty et al., 2014). ICH occurs in 15% of very preterm infants, and more than half of all infants with severe ICH develop post-hemorrhagic ventricular dilation and around 30% develop cerebral palsy (Bolisetty et al., 2014). Currently there is no efficient therapy to protect infants from neurological disability as a result of ICH. ICH-induced brain injury includes both the primary injury, caused by the physical presence of a hematoma, and secondary injury caused by the effect of neurotoxic compounds released from the hematoma, including free iron and cell-free hemoglobin, and iron accumulation contributes to ventricular dilation (Chen et al., 2015; Xiong et al., 2016; Garton et al., 2017). It has also been shown that iron chelator treatment alleviates ventricular dilation after ICH, and this confirms the important role of iron in the development and progression of brain injury after ICH (Meng et al., 2015).

There is growing evidence that clot-derived factors such as hemoglobin, iron, and fibrinogen have an important role in ICHinduced secondary injury (Ley et al., 2016; Garton et al., 2017; Petersen et al., 2018). Hemoglobin and iron-induced neuronal toxicity have attracted much attention by neonatologists (Li et al., 2017), and hemoglobin and its metabolites are cytotoxic and are capable of inducing oxidative stress and inflammation (Agyemang et al., 2017). Hemoglobin is degraded in the brain by heme oxygenase into iron, carbon monoxide, and biliverdin, and a large amount of iron is released from hemoglobin into the extracellular space following hemorrhage (Savman et al., 2001), and this can result in free radical production through the Fenton reaction and ultimately cause oxidative damage to DNA, proteins, and lipids (**Figure 1**). Non-protein-bound iron is elevated in cerebrospinal fluid from preterm infants with posthemorrhagic ventricular dilatation (Savman et al., 2001). The infusion of hemoglobin and its degradation products to rats causes ventricular dilation and brain damage (Hua et al., 2003) and leads to upregulation of heme oxygenase-1 and ferritin and to increased iron deposition (Strahle et al., 2014). Treatment with DFO, an iron chelator, has shown promise as a treatment for ICH due to its beneficial effects in reducing iron-induced neuronal death and inflammation (Hatakeyama et al., 2013; Strahle et al., 2014; Karuppagounder et al., 2018).

Although studies have shown that necrosis, apoptosis, and autophagy contribute to brain injury after ICH (Chen et al., 2012; Lule et al., 2017; Li et al., 2018), more and more evidence indicates that cell death occurs through other mechanisms as well (Zille et al., 2017; Li et al., 2018). Iron toxicity and glutamate accumulation, and thus ROS generation, also occur after ICH, which supports the hypothesis that ferroptosis is involved in these brain injuries (Dixon et al., 2014), and it has been shown that experimental ICH exhibits many features of ferroptotic and necroptotic cell death (Zille et al., 2017). In addition, ultrastructural analysis of neuronal death after ICH has shown the co-occurrence of ferroptosis, autophagy, and necrosis (Li et al., 2018). Fer-1, a specific inhibitor of ferroptosis, prevents neuronal death and reduces iron deposition induced by hemoglobin in organotypic hippocampal slice cultures, and it also improves neurological function in mice after ICH. Fer-1 also reduces lipid ROS production and attenuates the increased expression level of PTGS2 and its gene product cyclooxygenase-2 both ex vivo and in vivo (Li et al., 2017). All of these adult animal model studies confirm that ferroptosis occurs in both the collagenase-induced ICH model and in autologous blood injection-induced cell death (Li et al., 2017; Zhang et al., 2018). However, there is no report yet regarding ferroptosis in neonatal intraventricular hemorrhage. An intervention study using DFO in a neonatal rat germinal matrix hemorrhage model showed neuroprotection in terms of both brain morphology and behavior (Klebe et al., 2014), and this indicates that ferroptosis might play an even more important role in the secondary injury after neonatal ICH. This requires further investigation to provide evidence for potentially clinically translatable therapeutic strategies.

# CONCLUDING REMARKS

As a novel form of regulated cell death, ferroptosis occurs in cells when iron accumulation and lipid peroxidation are activated. However, our understanding of ferroptosis is still at an early stage, and no specific biomarkers have been found for the identification of ferroptosis in vivo. Overall, ferroptosis is largely defined through pharmacological activators and inhibitors. According to pharmacological effects of pro- and anti-ferroptotic substances, the mechanism of ferroptosis mainly includes iron toxicity, the inactivation of GPX4 and system X<sup>c</sup> <sup>−</sup>, and ultimately lipid peroxidation. Ferroptosis is involved in adult ischemic and intraventricular hemorrhage-induced neuronal cell death, and ferroptosis inhibition reduces neuronal death and behavioral deficits. Even though there are so far no direct reports of ferroptosis in neonatal brain injury, there is evidence to suggest that ferroptosis should be more prone to occur in the neonatal brain. To mimic PVL in vitro, cultured oligodendrocytes under cystine-free conditions showed depleted GSH and cell death, and this type of cell death could be blocked by ferroptosis-inhibiting ferrostatins (Skouta et al., 2014). Considering the emerging evidence, ferroptosis should be investigated further in models of neonatal brain injury and should be considered as a potential therapeutic target for the treatment of neonatal brain injury.

# AUTHOR CONTRIBUTIONS

CZ conceptualized and designed the review, and participated in literature searching and analysis, prepared the initial draft, and finalizing the manuscript. YNW participated in literature searching and collection and analysis, and

participated in drafting and finalizing the manuscript. JS, YFW, XW, and CC participated in drafting and finalizing the manuscript.

#### FUNDING

This work was supported by the National Natural Science Foundation of China (31761133015, U1704281), the Department

#### REFERENCES


of Science and Technology of Henan Province, China (171100310200), the National Key Research and Development Program of China (2018YFC1004604), the Swedish Research Council (2015-02845, 2018-02667), the Swedish Childhood Cancer Foundation (PR2016-072, NCp2016-0019), the Swedish Cancer Foundation (CAN2017/509), Swedish Governmental grants to scientists working in health care (ALFGBG-717791), Brain Foundation (FO2018-0034), and Deutsche Forschungsgemeinschaft (DFG CU43/12-1).


murine astroglial glutathione peroxidase production in vitro. Neurosci. Lett. 321, 73–76. doi: 10.1016/S0304-3940(02)00041-1


can synergize with PGE2 to inhibit ferroptosis and improve outcomes following hemorrhagic stroke in mice. Ann. Neurol. 84, 854–872. doi: 10.1002/ana.25356


fnins-13-00115 February 12, 2019 Time: 18:47 # 9



**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 Wu, Song, Wang, Wang, Culmsee and Zhu. 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.

fnins-13-00115 February 12, 2019 Time: 18:47 # 11

# Nitric Oxide, Iron and Neurodegeneration

Chao Liu1,2,3,4 \*, Mui Cheng Liang1,4 and Tuck Wah Soong1,4,5 \*

<sup>1</sup> Department of Physiology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore, <sup>2</sup> Jiangsu Province Key Laboratory of Anesthesiology, Xuzhou Medical University, Xuzhou, China, <sup>3</sup> Jiangsu Province Key Laboratory of Anesthesia and Analgesia Application Technology, Xuzhou Medical University, Xuzhou, China, <sup>4</sup> Neurobiology/Ageing Program, Centre for Life Sciences, National University of Singapore, Singapore, Singapore, <sup>5</sup> National Neuroscience Institute, Singapore, Singapore

Iron is a crucial cofactor for several physiological functions in the brain including

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Veronica Perez De La Cruz, Instituto Nacional de Neurología y Neurocirugía (INNN), Mexico Hans Lassmann, Medical University of Vienna, Austria Sho Kakizawa, Kyoto University, Japan

\*Correspondence:

Chao Liu flowingsands@hotmail.com Tuck Wah Soong phsstw@nus.edu.sg

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 22 October 2018 Accepted: 30 January 2019 Published: 18 February 2019

#### Citation:

Liu C, Liang MC and Soong TW (2019) Nitric Oxide, Iron and Neurodegeneration. Front. Neurosci. 13:114. doi: 10.3389/fnins.2019.00114 transport of oxygen, DNA synthesis, mitochondrial respiration, synthesis of myelin, and neurotransmitter metabolism. If iron concentration exceeds the capacity of cellular sequestration, excessive labile iron will be harmful by generating oxidative stress that leads to cell death. In patients suffering from Parkinson disease, the total amount of iron in the substantia nigra was reported to increase with disease severity. High concentrations of iron were also found in the amyloid plaques and neurofibrillary tangles of human Alzheimer disease brains. Besides iron, nitric oxide (NO) produced in high concentration has been associated with neurodegeneration. NO is produced as a co-product when the enzyme NO synthase converts L-arginine to citrulline, and NO has a role to support normal physiological functions. When NO is produced in a high concentration under pathological conditions such as inflammation, aberrantly S-nitrosylated proteins can initiate neurodegeneration. Interestingly, NO is closely related with iron homeostasis. Firstly, it regulates iron-related gene expression through a system involving iron regulatory protein and its cognate iron responsive element (IRP-IRE). Secondly, it modified the function of iron-related protein directly via S-nitrosylation. In this review, we examine the recent advances about the potential role of dysregulated iron homeostasis in neurodegeneration, with an emphasis on AD and PD, and we discuss iron chelation as a potential therapy. This review also highlights the changes in iron homeostasis caused by NO. An understanding of these mechanisms will help us formulate strategies to reverse or ameliorate iron-related neurodegeneration in diseases such as AD and PD.

Keywords: oxidative stress, Parkinson's disease, nitric oxide, iron homeostasis, S-nitrosylated proteins

# INTRODUCTION

Iron, one of the most abundant metals in the earth crust (Weber et al., 2006), is a transition metal. Iron has the ability to dynamically form compounds with organic ligands, while switching between Fe2<sup>+</sup> (ferrous) and Fe3<sup>+</sup> (ferric) state. Because of this property, iron has a crucial role in catalyzing electron transfer (redox) in certain enzymatic reactions. It is critical in cellular respiration, oxygen transport, and many other biological reactions (Aisen et al., 2001). For example, deficiency of iron

has led to hypomyelination in some children (Oski et al., 1983; Lozoff et al., 2006). Iron-mediated production of free radicals via the conversion of hydrogen peroxide can damage many cellular structures. Most cellular iron are bound as iron–sulfur (Fe–S) clusters and by heme proteins. Cytoplasmic ferrous iron can be found in a labile and chelatable state that is known as the labile iron pool (also termed "free iron"). It is thought to be the main contributor of oxidative stress during iron overload (Piloni et al., 2016). Iron in the labile pool can be loosely bound to peptides, carboxylates and phosphates as compounds with low-mass, while some might exist as hydrated free iron. In mammalian cells, the labile iron concentration is less than 1 µM, and less than 5% of total iron (Kakhlon and Cabantchik, 2002). The labile iron exerts its toxicity by generating reactive oxygen species. Fe2<sup>+</sup> catalyzed reactions produce free radicals via the Fenton's reaction:

$$\mathrm{Fe^{2+} + H\_2O\_2 \to Fe^{3+} + HO \bullet + OH^-}$$

$$\mathrm{Fe^{3+} + H\_2O\_2 \to Fe^{2+} + HOO \bullet + H^+}$$

On the other-hand, labile iron taken up by the mitochondrial mitoferrin can be incorporated into Fe–S clusters and heme groups (Hentze et al., 2010). Cellular oxidative stress induced by iron overload is characterized by increased lipid peroxidation and protein and nucleic acid modifications (Ward et al., 2014). These free radicals interact with the surroundings indiscriminately. In the brain, iron plays a role in other enzymatic reactions that produce neurotoxins. The cells which have active iron metabolism are more prone to iron toxicity, such as dopaminergic neurons, which need iron for dopamine metabolism (Hare and Double, 2016). MRI imaging showed that iron deposition was observed in the putamen, globus pallidus, red nucleus and substantia nigra, but yet not all of them have dramatic neuronal loss in PD patients. The major neuronal loss in PD happens in dopaminergic neurons of the substantia nigra pars compacta, which is unique because dopamine metabolism produces several neurotoxins. These toxins are more harmful than widespread oxidative stress caused by iron deposition (Zhou et al., 2010).

Inflammation is commonly observed in the brains of animal models or patients with neurodegenerative diseases like AD and PD (Block and Hong, 2005). Neuroinflammation can produce a variety of proinflammatory factors including TNF-α, IL-6 and nitric oxide (NO). At low levels, NO supports normal neuronal functions via S-nitrosylation of target proteins. However, when NO is produced in a high concentration under pathological conditions such as inflammation, aberrantly S-nitrosylated proteins can initiate neurodegeneration (Nakamura et al., 2013).

We discuss in this review how iron homeostasis is affected in neurodegenerative disease, especially by NO, and how dysregulation of iron can lead to neurodegeneration.

#### IRON HOMEOSTASIS

About half of the total iron is contained in hemoglobin which is needed to transport oxygen along the blood-stream. The other half is, however, stored in complex with ferritin in various cells, mostly in bone marrow, liver and spleen. Only a small portion is circulated in the plasma, bound to transferrin (Conrad and Umbreit, 2000). The absorption of dietary iron takes place in the duodenum. The iron is absorbed by enterocytes of the duodenal lining bound by a protein, such as heme protein, or through divalent metal transporter 1 (DMT1) in ferrous form. The lining cells of the intestine can either bind iron by ferritin for storage, or release it to the body via ferroprotein, the only known mammalian iron exporter (Gropper et al., 2013). Ferroportin (Fpn) can be post-translationally repressed by a peptide hormone of a 25-amino acid length, named hepcidin, Hepcidin binds to Fpn and internalizes it within the cell (Nemeth et al., 2004). Fpn expression can be modulated by the IRP-IRE regulatory mechanism. If the iron concentration is too low, the IRP-IRE binding increases, thus inhibiting Fpn translation. On the other hand, Fpn translation can also be regulated by the microRNA, miR-485-3p (Sangokoya et al., 2013). The ferroxidase, hephaestin, found in the small intestine, oxidizes Fe2<sup>+</sup> to Fe3<sup>+</sup> and therefore helps Fpn transfer iron across the small intestine basolateral membrane (Ward et al., 2014). Iron metabolism in the body is thus regulated by regulating each of these steps. For instance, under iron-deficient anemic conditions, enterocytes will produce higher levels of Dcytb, DMT1 and transferrin receptor (TfR) proteins (Fleming and Bacon, 2005). Absorption of dietary iron is also enhanced by vitamin C and, in contrast, reduced by excess calcium, zinc, or manganese.

In the brain, iron uptake is through the blood-brain barrier (**Figure 1**). Iron is taken up by the capillary endothelial cell TfR, in the form of transferrin-Fe3<sup>+</sup> (Tf-Fe3+). The iron is transported to cerebral compartment from the basolateral membrane of endothelial cells, and is then made available to neurons and glia. Of note, oligodendrocytes stained for most of the detectable iron in the brain. Transferrin is also found predominantly in these cells (Bloch et al., 1985), which is important for myelination formation. Oligodendrocytes support myelin membranes that are many times of the weight of their somata, and need to generate a lot of ATP (Cammer, 1984), hence they need a lot of iron to maintain their metabolism. The TfR is expressed in blood vessels, and in cortical, striatal and hippocampal neurons (Connor and Menzies, 1995). Most types of cells uptake iron primarily through transferrin-mediated endocytosis, via TfR1, TfR2, and GAPDH (West et al., 2000; Kumar et al., 2011; **Figure 1**). Fe3<sup>+</sup> is dissociated from transferrin in the endosome, reduced to Fe2<sup>+</sup> by a STEAP family reductase, and exported from endosome into cytosol by DMT1 (Hentze et al., 2010). Alternatively, Fe2<sup>+</sup> can be absorbed directly via the plasma membrane divalent cation transporters such as DMT1 and ZIP14 (Lane et al., 2015). Fpn is the only protein know for iron efflux so far (Zecca et al., 2004). Neurons and astrocytes express TfR, DMT1, and Fpn (Moos et al., 2007; Zarruk et al., 2015). The astrocyte end feet form intimate contact with brain capillary epithelial cells and play an important role to mediate iron export from blood to brain (Moos et al., 2006). Astrocytes express ceruloplasmins which oxidize Fe2<sup>+</sup> to Fe3<sup>+</sup> and facilitate iron export from astrocytes (**Figure 1**). During brain inflammation, iron is accumulated in microglia stead of astrocyte. The efflux mechanism of iron in microglia was disrupted in brain

complex. The Hb:Hp complex is bound with the receptor CD163 and undergo endocytosis. The complex are degraded after endocytosis and the iron is absorbed into intracellular iron pool and stored in ferritin. The iron flow direction is indicated by black arrows.

inflammation because Fpn is endocytosed, induced by hepcidin, and microglia cells do not express ferroxidase (Zarruk et al., 2015). In contrast, astrocytes do not accumulate iron, showing robust expression of influx and efflux proteins including Fpn and ferroxidase, with normal iron recycling capability (**Figure 1**). Ferritin, a major iron storage protein in the brain, compromises the H- or L-ferritin monomers. Neurons and oligodendrocyte express H-ferritin, which have a high iron metabolite rate. Microglia express L-ferritin, which is associated with iron storage (Connor et al., 1994). Although oligodendrocytes have the highest iron accumulation in normal aging brain, there is little effect on myelination and oligodendrocyte in neurodegenerative diseases. Oligodendrocytes have high levels of ferritin expression (Connor et al., 1994), which provide more buffering capacity for free iron and therefore affording more protective effect. Fpn needs to couple with ferroxidase to export iron. Astrocytes express ceruloplasmin, while oligodendrocytes express another ferroxidase, hephaestin, to facilitate iron export from oligodendrocyte (Schulz et al., 2011). Oligodendrocytes down regulate TfR1 to reduce iron uptake after maturation (Han et al., 2003). These iron level-regulating mechanisms protect oligodendrocytes in neurodegenerative diseases. The iron absorbed from blood during different age is more than what is accumulated in the brain, indicating that there is a iron efflux mechanism in the brain. The iron export from brain to blood maybe through the cerebrospinal fluid (Moos et al., 2007).

In addition, the highest iron load in the brain is seen in relation to chronic microbleeds. Cerebral microbleeds are likely to be caused by cerebral atherosclerosis in aged people or people with neurodegenerative diseases (Martinez-Ramirez et al., 2014). The iron from hemolysis of erythroid cells may be the major source of iron load in aged or neurogenerative brain (**Figure 1**).

Erythroid cells contain most of the body iron in hemoglobin (Hb). Hb contains 70% of the total body iron in its heme moiety. Hb is not restricted in erythroid cells, but expressed in glia, macrophage and neurons. Hb is an oxygen reservoir inside the cell and regulate mitochondria function. While Hb mainly exists intracellularly, it will be released extracellularly during hemolysis. Haptoglobin (Hp) is mostly produced by hepatic cell and binds free Hb in plasma with very high affinity (Wicher and Fries, 2010). In the brain, Hp is produced locally by oligodendrocyte (Zhao et al., 2009). The Hb–Hp complex has a membrane receptor CD163. The interaction between them leads to internalization of CD163 (**Figure 1**), Hb dissociating from Hp, Hb heme degradation, and iron being absorbed into the intracellular iron pool. Following intravascular hemorrhage, hippocampal and cortical neurons express CD163 in the brain (Garton et al., 2016), which facilitates iron absorption, post-hemolysis.

Although rodent models are widely used for iron study, it is noteworthy that iron accumulation and cellular storage are different between humans and experimental rodent models. Rodent generally have a lower basal level of iron deposition in the brain, especially in young animals (less than one year old), as shown by several groups (Jeong and David, 2006; Liu et al., 2018).

#### IRON DEPOSITION CHANGES IN AD AND PD

In AD, iron hemostasis is disrupted. Transferrin has been shown to be decreased consistently, and in particular in the white matter, White matter is thought to play a major role in neurodegeneration, and increased peroxidative damage to white matter is known to take place in AD (Connor et al., 1992). High concentration of iron is accumulated in Aβ plaques and tau tangles which is characteristic of AD. The Aβ plaques contain a fairly large amount of labile iron, while the neighboring cells express significant levels of ferritin and transferrin receptors (Connor et al., 1995). Moreover, iron influence APP translation via IRP-IRE system. There is an IRE within the 5<sup>0</sup> -UTR of APP mRNA (Rogers et al., 2002), and in response lowered intracellular iron as a result of chelation by desferrioxamine and clioquinol, translation of APP was selectively down-regulated. APP interacts with Fpn, and it has ferroxidase activity as it possessed the conserved H-ferritin-like active site, and Zn2<sup>+</sup> can specifically inhibit this active site (Duce et al., 2010; Bush, 2013). Lei et al. showed that tau deficiency caused iron accumulation in brain and dopaminergic neuron degeneration, which led to parkinsonism in mice with dementia. Tau promotes the export of neuronal iron by facilitating the trafficking of APP to the plasma membrane. The study suggested that Alzheimer disease, Parkinson disease and tauopathies that are associated with the iron toxicity due to the loss of soluble tau could in principle be rescued by a pharmacological agent such as clioquinol, an iron chelator (Lei et al., 2012).

Several studies reported that iron deposition was increased in the substantia nigra according to the severity of the disease in PD patients (SN) (Dexter et al., 1987, 1989; Hirsch et al., 1991). The researchers used plasma spectroscopy to detect iron concentration quantitatively in various brain regions (Dexter et al., 1989). In PD brain, histology studies showed that iron accumulate in neurons and glia in SN (Jellinger et al., 1990). In vitro, the formation of α-synuclein fibrils can be accelerated by the presence of Fe3<sup>+</sup> (Uversky et al., 2001) and it was also reported that redox-active iron was sequestrated by Lewis bodies in SN (Castellani et al., 2000). Furthermore, there are reports that a dysfunction in the IRP-IRE system that results in iron accumulation gave rise to α-Syn-induced toxicity (Li et al., 2010; Febbraro et al., 2012), that led to PD pathogenesis (Rocha et al., 2017). Similarly, in almost all PD patient brains the Lewy bodies contained aggregated α-Syn (Wakabayashi et al., 2007).

The reason for iron accumulation in SN is unclear. Several hypotheses proposed include increased brain-blood-barrier (BBB) permeability (Faucheux et al., 1999; Kortekaas et al., 2005), increased neuroinflammation (Block and Hong, 2005; Conde and Streit, 2006; Machado et al., 2011), increased DMT1 isoform expression or function (Block and Hong, 2005; Salazar et al., 2008; Machado et al., 2011; Liu et al., 2018), and altered transferrin or lactoferrin transport (Faucheux et al., 1995; Mastroberardino et al., 2009). Cellular iron accumulation in PD brain may be caused by elevated influx or decreased efflux. Inflammation could contribute to iron accumulation by either increasing DMT1 uptake activity or TfR transport activity. In a mouse model, DMT1 activity was increased to mediate the iron uptake (Salazar et al., 2008), and this increase may be due to direct S-nitrosylation of DMT1 (Liu et al., 2018).

## NO PRODUCTION IN IMMUNE RESPONSE AND NEURODEGENERATION

NO is a gaseous signaling molecule that initially was thought as a dilator in blood vessels, with guanylyl cyclase as the major effector. NO binds to the heme group of guanylyl cyclase and activates it in the presence of iron. High cGMP level are associated with release of neurotransmitters including glutamate, acetylcholine and glycine. NO participates in tumor and bacteria immunity and in the central nervous system, it acts as a retrograde neurotransmitter. In the nervous system, NO has both physiological and pathological functions. For example, NO contributes to long-term potentiation (LTP) and long-term depression (LTD), and thus it plays a role in learning and memory (Schuman and Madison, 1991; Shibuki and Okada, 1991; Lev-Ram et al., 1997; Kakegawa and Yuzaki, 2005; Kakizawa et al., 2012). NO enhances CREB expression to mediate the response to brain-derived neurotrophic factor (Riccio et al., 2006). NO also mediates glutamate-NMDA receptor (NMDAR) signaling as glutamate activates NMDAR in neurons, leading to Ca2<sup>+</sup> influx, and activation of nNOS (Xu et al., 2018). The synaptic NMDARs mediate neuroprotection, while the extrasynaptic NMDARs mediate neurodegeneration (Talantova et al., 2013; Molokanova et al., 2014), implicating the diverse signaling pathways triggered by NO. NO also binds to other iron-containing proteins, such as mitochondrial aconitase. Mitochondrial aconitase has a [4Fe-4S]2<sup>+</sup> cluster, which is a possible target for NO-induced toxicity

(Drapier and Hibbs, 1986, 1988). The interaction between mitochondrial aconitase and superoxidase are the major cause to mitochondrial damage (Vasquez-Vivar et al., 1999). Most of the cytotoxicity of NO is attributed to the production of peroxynitrite (Pacher et al., 2007), which is more powerful to produce radicals. Peroxynitrite is produced from the diffusion-controlled reaction between NO and superoxide in vivo (Squadrito and Pryor, 1995). Peroxynitrite is a strong oxidant and it interacts with electron-rich groups, including Fe–S cluster. It is reported that peroxynitrite is far more effective to produce hydroxyl radicals than Fenton's reaction (Beckman et al., 1990; Darley-Usmar et al., 1992; Hogg et al., 1992). Peroxynitrite is an important intermediator for protein nitration and oxidation, lipid peroxidation, mitochondria dysfunction, and finally causes apoptosis and necrosis (Radi, 2018).

NO is produced by NOS through the conversion of L-arginine to citrulline. Three distinct NOS isoforms have been identified in the brain (Forstermann et al., 1991). Neuronal NOS (nNOS) is expressed in neurons, while endothelial (eNOS) is expressed in brain endothelial cells. They are Ca2+/calmodulin-dependent and synthesize NO in a short period in response to receptor activation or extracellular stimuli (Moncada et al., 1991). Inducible NOS (iNOS) is expressed in glia cells upon brain injury or inflammation. Inducible NOS produces a large amount of NO upon stimulation by proinflammatory cytokines over a long period of time (Green et al., 1994).

In human immune response, NO is produced by phagocytes such as monocytes, macrophages, and neutrophils. In phagocytes, interferon-gamma (IFN-γ) or tumor necrosis factor (TNF) activates iNOS (Green et al., 1993). On the other hand, transforming growth factor-beta (TGF-β), interleukin-4 (IL-4) or IL-10 weakly inhibits iNOS. As such, phagocytes contribute to inflammatory and immune responses via NO (Green et al., 1994). In an immune response, NO is secreted as free radicals that is toxic to intracellular pathogens. The modes of action are via DNA damage (Wink et al., 1991; Nguyen et al., 1992) and degradation of Fe–S centers into iron ions and ironnitrosyl compounds (Hibbs et al., 1988). The molecular effects of NO depend on two kinds of reactions: S-nitrosylation of thiols and the nitrosylation of some metalloenzymes. Guanylate cyclase, a NO activated heme-containing enzyme, is an essential component of the relaxing function of NO on smooth muscles (Derbyshire and Marletta, 2009). cGMP activates protein kinase G that lead to the re-uptake of Ca2<sup>+</sup> and the rise in cytoplasmic Ca2<sup>+</sup> activates calcium-activated potassium channels triggering the relaxation of smooth muscle (Rhoades and Tanner, 2003).

In addition to neuro-inflammatory stimuli, induction of iNOS expression in astrocytes, macrophages, and microglia by Aβ oligomers or by toxins such as 1-methyl-4-phenyl-1,2,3,6 tetrahydropyridine (MPTP) have been reported to increase NO levels in the degenerating brain (Liberatore et al., 1999; Medeiros et al., 2007; Nakamura et al., 2013). Knockdown of iNOS in APP/PS1 AD mouse model ameliorate AD-related symptom including Aβ plaque formation, premature death, astroglioses and microgliosis (Nathan et al., 2005). However, in the Tg2576 APP AD mouse model, ablation of iNOS exacerbated spatial learning and memory and tau pathology, providing evidence that NO may have a neuroprotective role (Wilcock et al., 2008).

#### NO REGULATION ON IRON HOMEOSTASIS

NO targeted proteins have been partially characterized. NO can interact with Fe–S cluster containing protein and influence their enzyme activity. One of the Fe–S cluster-containing proteins is IRE-binding protein (also termed "iron regulatory protein," IRP). Cytosolic iron concentrations sensed by IRPs could posttranscriptionally adjust the expression of iron metabolizing genes to optimize the availability of labile iron. IRPs bind to iron-responsive elements (IRE), which are specific non-coding mRNA sequences, to control iron metabolism. IREs are of 30 nucleotide in length found along RNA motifs, and they contain the CAGUGN sequence (the classic IRE motif) that form a stemloop structure (Molokanova et al., 2014). IREs are found either within the 3<sup>0</sup> -UTR (untranslated region) or 5<sup>0</sup> -UTR regions of a specific mRNA. IRP1 and IRP2 are examples of two RNA-binding proteins that interact with IRE to modulate the translation of either the ferritin or Fpn mRNA, and they also control the stability of TfR and DMT1 mRNAs. The binding of IRPs and IREs is regulated by free iron concentration. Therefore, IRPs can act as either a translation enhancer or inhibitor (Pantopoulos, 2004; Piccinelli and Samuelsson, 2007). As ferritin and Fpn transcripts contain IRE in their 5<sup>0</sup> -UTRs, their translation will be inhibited by IRPs when there is iron deficiency (Hentze et al., 2010). The decreased expression of ferritin and Fpn reduces free iron binding and export, leading to an increased in availability of labile iron for use by the cell. In contrast, the TfR and DMT1 transcripts contain 3<sup>0</sup> -UTR IREs that bind IRPs, and when cytoplasmic iron is deficient, the stabilization of transcripts will increase synthesis of TfR and DMT1 proteins to enhance iron uptake (Hentze et al., 2010). Examples of transcripts that contain IREs include those that encode the ferritin subunits L and H, TfR, Fpn, DMT1, mitochondrial aconitase, succinate dehydrogenase, erythroid aminolevulinic acid synthetase, amyloid precursor protein and a-synuclein. These downstream genes suggest that iron has close regulation of iron metabolism, redox and neurodegeneration via the IRP-IRE system. NO could regulate IRP-IRE binding, which in turn regulates many iron metabolism-related proteins (**Figure 2**). NO acts on DMT1 by increasing IRP1 binding to the IRE sequence located at the 3<sup>0</sup> -UTR, to stabilize and increase DMT1 transcript level, akin to cells exposed to low iron condition (Weiss et al., 1993; Jaffrey et al., 1994; Wallander et al., 2006; Skjorringe et al., 2015). Similarly, NO can regulate ferritin, Fpn and TfR via regulating the interaction of IRP-IRE binding, and hence regulate iron metabolism (**Figure 3**). Another report also showed that NO can enhance iron deposition in the brain via decreasing APP expression (Ayton et al., 2015). The authors reported obvious decrease in expression of APP in substantia nigra of PD brain. APP KO mice have iron-dependent dopaminergic neuron loss, while APP overexpressing mice have protection effect in MPTP mouse model, as APP facilitates iron efflux. NO decreases APP expression via the IRP-IRE system

regulates the transcription of iron-metabolism-related proteins, and elevates intracellular iron level. NO also directly S-nitrosylates DMT1, which enhances DMT1 transporter function. In addition, NO S-nitrosylates Dexras 1 and enhances the binding of Das1-PAP7-DMT1 complex and finally enhances iron uptake.

and this may explain how NO leads to dopaminergic neuron loss in PD.

NO also regulates iron metabolism-related proteins in other ways, such as S-nitrosylation (**Figure 2**). Another pathway to increase iron uptake via DMT1 is by S-nitrosylation of Dexras 1 to indirectly regulate DMT1 function and enhance Fe2<sup>+</sup> uptake as reported in PC12 and cortical neurons (Cheah et al., 2006). Recently, we have shown that NO directly modulated DMT1 and enhanced its function via S-nitrosylation. This is unexpected as S-nitrosylation of proteins important in PD such as Parkin and XIAP resulted in compromised functions. Besides, many S-nitrosylated proteins have been identified in the past decade, and of note, those that have been functionally characterized have a loss-of-function (Nakamura et al., 2013). In this regard, enhanced DMT1 activity arising from S-nitrosylation would therefore present a mechanism by which Fe2<sup>+</sup> could be accumulated over aging and contribute to age-dependent neurodegeneration (**Figure 3**).

# POTENTIAL THERAPY FOR NEURODEGENERATIVE DISEASES TARGETED TO IRON DEPOSITION

The potential therapeutic use of iron chelators gained much attention in recent years (Ward et al., 2014; Zhou and Tan, 2017). The strategy to target iron deposition is either to chelate iron directly or to regulate iron homeostasis, including NOregulated iron absorption. The candidate compounds should be

BBB-permeant and easily penetrate cell membrane, chelate free iron and minimize the side effect to normal iron metabolism.

Several iron chelators were used to deplete excessive iron and yielded promising clinical outcome. The clinical trial using deferiprone at 30 and 20 mg/kg per day was carried out in PD patient, and the compound was well-tolerated. The syndrome was improved and the iron content in SN significantly decreased as monitored by MRI (Devos et al., 2014). Deferiprone seems quite promising so far, and is waiting to be tested in further clinical trial in larger population. Furthermore, in several in vivo PD models, iron chelators, including deferasirox, deferrioxamine, VAR10303 and D-607, have been used to significantly attenuate DA neuronal loss (Ghosh et al., 2010; Dexter et al., 2011; Bar-Am et al., 2015; Das et al., 2017).

Desferrioxamine has been shown to decelerate AD progression (Rogers and Lahiri, 2004). However, as DFO is unstable with poor BBB permeability (Bandyopadhyay et al., 2010), clioquinol, a BBB-permeant iron chelator, was used instead in clinic to mitigate cognitive loss by reducing plasma Aβ levels in AD patients (Ritchie et al., 2003). As clioquinol has side effects associated with myelopathies (Zhang et al., 2013), PBT2, a second-generation 8-hydroxyquinoline analog metal chaperone which can bind transition metal, was used in clinical trial for AD patients. The patients who received 250 mg/kg PBT2 have substantially reduced Aβ level in cerebral spinal fluid and showed cognitive improvement in a Phase II trial (Lannfelt et al., 2008; Faux et al., 2010). However, in another Phase II trial announced by the Australian company Prana Biotechnoloy in 2014, PTB2 failed to improve brain amyloid deposition, neuronal function, brain atrophy and cognition in a one-year course treatment. Several other iron chelators, such as VK28 (Fe3<sup>+</sup> chelator), HLA20 and M30 (Fe3<sup>+</sup> chelator with N-propargylamine-like properties), can not only suppress APP expression but also reduce Aβ level in the brain (Bandyopadhyay et al., 2010).

Besides direct chelation of iron, potent nontoxic IRE inhibitors with excellent BBB penetrating capacity were also thought to have high therapeutic significance in neurodegenerative diseases. The IRE inhibitors that downregulate translation of APP and α-Syn and prevent protein aggregation can support survival of neurons. To date, a few promising drug candidates of IRE inhibitors have been characterized and are being tested in various clinical trials for AD and PD patients (Zhou and Tan, 2017). Posiphen is a natural product that has been shown to inhibit translation of both APP and a-synuclein proteins. Furthermore, it is nontoxic and potent (Rogers et al., 2011). The inhibitory effect has been validated by various experiments done in vitro and in vivo (Lahiri

#### REFERENCES


et al., 2007; Yu et al., 2013). A phase I human clinical trial in subjects with mile cognitive impairment have shown that posiphen lowered APP production by 50 % in CSF and was well-tolerated (Maccecchini et al., 2012). The compound is still waiting for further clinical trials to test the efficacy for AD and PD (Bandyopadhyay et al., 2010). Another compound JTR-009, screened from a 110,000-compound library, was identified to have a more potent effect on APP translation than posiphen. JTR-009 is thought to bind to 5<sup>0</sup> -UTR IRE directly and selectively inhibit APP expression, which was validated in SH-SY5Y cells (Bandyopadhyay et al., 2013). Another case for targeting iron homeostasis is inhibiting DMT1 function. To inhibit NOmediated DMT1 functional increase, we used the NOS inhibitor L-name to reduce NO-mediated iron deposition in LPS-evoked mouse inflammatory model. L-NAME significantly ameliorated SN dopaminergic neuron loss and LPS-induced behavior deficit (Liu et al., 2018).

#### CONCLUSION AND FUTURE PERSPECTIVES

Iron homeostasis is elaborately regulated in the human brain, and iron accumulation is closely associated with neurodegenerative diseases. NO regulates iron deposition at several levels. So far, therapeutic targeting of iron deposition has yielded some promising results, and further clinical trials in larger populations are still needed. There are still some questions that remained unresolved. For example, how iron is transported through brain capillary epithelial cells and more specifically, how S-nitrosylation of DMT1 enhanced its transporter activity. The technique for iron and NO concentration detection is a limitation, especially for accurate detection of Fe2<sup>+</sup> and Fe3<sup>+</sup> concentrations. A real-time, quantitative and in vivo detection technique will be extremely valuable for the field.

#### AUTHOR CONTRIBUTIONS

CL and MCL drafted the manuscript. TWS critically edited the manuscript. All authors approved the final version of the manuscript.

#### FUNDING

This work was supported by the Singapore National Medical Research Council, NMRC, Grant no: NMRC/CBRG/0042/2013.

of beta-amyloid precursor protein. J. Neurosci. 35, 3591–3597. doi: 10.1523/ JNEUROSCI.3439-14.2015

Bandyopadhyay, S., Cahill, C., Balleidier, A., Huang, C., Lahiri, D. K., Huang, X., et al. (2013). Novel 5<sup>0</sup> untranslated region directed blockers of iron-regulatory protein-1 dependent amyloid precursor protein translation: implications for down syndrome and Alzheimer's disease. PLoS One 8:e65978. doi: 10.1371/ journal.pone.0065978



Rhoades, R. A., and Tanner, G. A. (2003). Medical Physiology, 2nd Edn. Philadelphia, PA: Lippincott Williams & Wilkins.


**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 Liu, Liang and Soong. 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.

fnins-13-00114 February 14, 2019 Time: 19:2 # 10

# The Role of Iron in Friedreich's Ataxia: Insights From Studies in Human Tissues and Cellular and Animal Models

José Vicente Llorens1,2 \*, Sirena Soriano3,4, Pablo Calap-Quintana1,2 , Pilar Gonzalez-Cabo5,6,7 and María Dolores Moltó1,2,8 \*

<sup>1</sup> Department of Genetics, Faculty of Biological Sciences, University of Valencia, Valencia, Spain, <sup>2</sup> Unit for Psychiatry and Neurodegenerative Diseases, Biomedical Research Institute INCLIVA, Valencia, Spain, <sup>3</sup> Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, TX, United States, <sup>4</sup> Jan and Dan Duncan Neurological Research Institute, Texas Children's Hospital, Houston, TX, United States, <sup>5</sup> Department of Physiology, Faculty of Medicine and Dentistry, University of Valencia, Valencia, Spain, <sup>6</sup> Center of Biomedical Network Research on Rare Diseases CIBERER, Valencia, Spain, <sup>7</sup> Associated Unit for Rare Diseases INCLIVA-CIPF, Biomedical Research Institute INCLIVA, Valencia, Spain, <sup>8</sup> Center of Biomedical Network Research on Mental Health CIBERSAM, Valencia, Spain

Edited by: Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Jordi Tamarit, Universitat de Lleida, Spain Ian Alexander Blair, University of Pennsylvania, United States

#### \*Correspondence:

José Vicente Llorens j.vicente.llorens@uv.es María Dolores Moltó dmolto@uv.es

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 15 November 2018 Accepted: 23 January 2019 Published: 18 February 2019

#### Citation:

Llorens JV, Soriano S, Calap-Quintana P, Gonzalez-Cabo P and Moltó MD (2019) The Role of Iron in Friedreich's Ataxia: Insights From Studies in Human Tissues and Cellular and Animal Models. Front. Neurosci. 13:75. doi: 10.3389/fnins.2019.00075 Friedreich's ataxia (FRDA) is a rare early-onset degenerative disease that affects both the central and peripheral nervous systems, and other extraneural tissues, mainly the heart and endocrine pancreas. This disorder progresses as a mixed sensory and cerebellar ataxia, primarily disturbing the proprioceptive pathways in the spinal cord, peripheral nerves and nuclei of the cerebellum. FRDA is an inherited disease with an autosomal recessive pattern caused by an insufficient amount of the nuclear-encoded mitochondrial protein frataxin, which is an essential and highly evolutionary conserved protein whose deficit results in iron metabolism dysregulation and mitochondrial dysfunction. The first experimental evidence connecting frataxin with iron homeostasis came from Saccharomyces cerevisiae; iron accumulates in the mitochondria of yeast with deletion of the frataxin ortholog gene. This finding was soon linked to previous observations of iron deposits in the hearts of FRDA patients and was later reported in animal models of the disease. Despite advances made in the understanding of FRDA pathophysiology, the role of iron in this disease has not yet been completely clarified. Some of the questions still unresolved include the molecular mechanisms responsible for the iron accumulation and iron-mediated toxicity. Here, we review the contribution of the cellular and animal models of FRDA and relevance of the studies using FRDA patient samples to gain knowledge about these issues. Mechanisms of mitochondrial iron overload are discussed considering the potential roles of frataxin in the major mitochondrial metabolic pathways that use iron. We also analyzed the effect of iron toxicity on neuronal degeneration in FRDA by reactive oxygen species (ROS)-dependent and ROS-independent mechanisms. Finally, therapeutic strategies based on the control of iron toxicity are considered.

Keywords: Friedreich's ataxia, frataxin, iron, animal models, oxidative stress, lipid deregulation, iron chelators

# INTRODUCTION

fnins-13-00075 February 14, 2019 Time: 19:4 # 2

Nicholaus Friedreich, a German neurologist among a family of professors of medicine, had described a new disease entity in successive papers published between 1863 and 1877 (Koeppen, 2013). The new disease, named later Friedreich's ataxia (FRDA), was characterized by degenerative atrophy of the posterior columns of the spinal cord, leading to the development of progressive ataxia, sensory loss and muscle weakness in patients and was associated with cardiomyopathy in many cases. Clear diagnostic criteria for FRDA were established by Anita Harding in the 1980s (Harding, 1981), who specified the hereditary nature of the disease with an autosomal recessive pattern, an onset during puberty, progressive ataxia of the limbs and gait with the absence of tendon reflexes in the legs, extensor plantar responses, and dysarthria. In 1996, the identification of the causative gene of FRDA and expansion of a trinucleotide repeat as the most frequent mutation (Campuzano et al., 1996) led to the development of a molecular tool for a definitive diagnosis of the disease. It was crucial to address the differential diagnosis with respect to other peripheral neuropathies and ataxias with similar clinical features, as well as the clinical heterogeneity in FRDA, including atypical phenotypes (Bidichandani and Delatycki, 2017). Moreover, the availability of this molecular approach allowed for appropriate genetic counseling.

## Epidemiological, Pathophysiological and Molecular Aspects of FRDA

Friedreich's ataxia (OMIM #229300; ORPHA:95) is a rare neurodegenerative disease affecting 1:20,000–1:50,000 people of Indo-European and North African descendent (Vankan, 2013), and with a carrier frequency of 1:60–1:100, FRDA is the most frequently inherited ataxia in these populations. The neurological signs and symptoms in FRDA are the results of pathological changes that disturb both the central and peripheral nervous systems (NSs) (reviewed in Marmolino, 2011; Koeppen and Mazurkiewicz, 2013). The dorsal root ganglia (DRG) are particularly vulnerable and have been identified as the first site of neuropathology with the degeneration of the large sensory neurons and posterior columns, leading to the loss of position and vibration senses. Next, affectation of the corticospinal and spinocerebellar tracts of the spinal cord occurs, with loss of myelinated fibers. In the cerebellum, dentate nuclei (DN) also show progressive atrophy. Cerebral abnormalities are being progressively reported with the application of neuroimaging techniques (Selvadurai et al., 2018). Non-neurological phenotypes of FRDA mainly include hypertrophic cardiomyopathy, which is the leading cause of mortality (Raman R. et al., 2011; Payne and Wagner, 2012; Weidemann et al., 2013; Koeppen et al., 2015). The heart typically maintains adequate systolic function (Kipps et al., 2009) in FRDA patients who develop a severe hypertrophic cardiomyopathy until shortly before death (Rajagopalan et al., 2010). Additionally, diabetes mellitus and skeletal abnormalities are also present in this disease (Parkinson et al., 2013; Bidichandani and Delatycki, 2017).

Friedreich's ataxia is caused by loss-of-function mutations in the frataxin gene (FXN) (Campuzano et al., 1996). The most common mutation appears in homozygosis in most patients and consists of an abnormally expanded GAA trinucleotide repeat (usually between 600 and 900 repeats) in the first intron of the gene FXN (Campuzano et al., 1996; Monrós et al., 1997). The remaining patients (approximately 4%) are compound heterozygous with the expanded GAA repeat on one allele and another pathogenic variant (point mutation, deletion or insertion) on the other allele (Galea et al., 2016). The pathological GAA expansion affects the expression of FXN by blocking the transcription of mRNA through the formation of sticky DNA triplexes and R-loop structures (reviewed in Kumari and Usdin, 2012) and/or inducing abnormal heterochromatinization (Saveliev et al., 2003). Thus, the level of frataxin is strongly reduced from 5 to 30% of the physiological level (Campuzano et al., 1997). This reduction is related to the disease severity and length of the GAA repeat. In general, longer expansions are present in patients with a more severe phenotype such as earlier onset, faster progression and/or presence of non-neurological features (Parkinson et al., 2013). Although most of the pathogenic point mutations also decrease the level of functional protein, some lead to the production of a less active protein (De Castro et al., 2000; Bridwell-Rabb et al., 2011).

#### Protein Frataxin

Frataxin is a small acidic protein, synthetized in the cytoplasm as a 210-residue polypeptide that is then imported into the mitochondria, where it is proteolytically processed to the mature and most abundant form of 130 residues [frataxin (81–210)] (Condò et al., 2007). Although frataxin has historically been considered a protein exclusively confined to the mitochondrial matrix, several studies have reported the existence of an extramitochondrial pool of frataxin in different human cell types (Acquaviva et al., 2005; Condò et al., 2006; Lu and Cortopassi, 2007). Extramitochondrial frataxin corresponds to the (81–210) mature form of the protein (Condò et al., 2010) and earlier studies suggested that after the initial transport and processing in the mitochondria, mature frataxin was transported back to the cytosol (Acquaviva et al., 2005; Condò et al., 2006). However, a recent study describes an N-terminal acetylated extramitochondrial form of frataxin in erythrocytes. This (76– 210) isoform does not contain the mitochondrial targeting sequence, and remains in the cytosol where it is cleaved to produce an (81–210) protein identical to the mitochondrial mature form (Guo et al., 2018).

The three-dimensional structure of frataxin shows seven antiparallel β-sheets flanked by two α-helices, producing a characteristic globular αβ fold (Dhe-Paganon et al., 2000). The protein is ubiquitously expressed (Campuzano et al., 1996, 1997), but different cells types show distinct susceptibility to its deficiency. This could be explained by frataxin reduction in different tissues being associated with distinct transcriptomic profiles, as described recently (Chandran et al., 2017). Frataxin is a highly conserved protein through evolution, and its cellular function is critical for life in multicellular organisms. A strong reduction of frataxin in Drosophila melanogaster seriously affects

Llorens et al. Iron in Friedreich's Ataxia

viability (Anderson et al., 2005; Llorens et al., 2007). Knockout of the frataxin gene causes embryonic lethality in mice (Cossée et al., 2000) and in the plant Arabidopsis thaliana (Vazzola et al., 2007). In line with this, FRDA patients with non-GAA-repeat mutations in both frataxin alleles, resulting in totally deficient frataxin function, have not been reported.

At the time of FXN identification, there was no evidence about the function of frataxin. Experiments in yeast Saccharomyces cerevisiae promptly suggested a potential role for frataxin in iron homeostasis regulation. Deletion of the ortholog of frataxin in yeast (Yfh1) causes iron accumulation in the mitochondria (Babcock et al., 1997; Foury and Cazzalini, 1997) that was initially suggested at the expense of the cytosolic iron pool (Knight et al., 1998). However, measurements of cytosolic iron in frataxin-deficient yeast cells show that the cytosolic iron levels are not affected by the reduction of the frataxin synthesis (Li et al., 2012). Also associated with frataxin deficiency are reduced activities of mitochondrial iron-sulfur cluster (ISC) enzymes and aconitase, mitochondrial dysfunction and oxidative stress (reviewed in Calabrese et al., 2005; Lupoli et al., 2018). Since then, efforts have been made to answer several questions about iron in FRDA, including whether its accumulation is a primary or secondary event in the disease, what is the mechanism causing the deregulation and buildup of iron, and the role of this metal in the neuropathology of FRDA. Here, we review the contribution of the cellular and animal models of this disease and relevance of the studies using patient samples to gain knowledge about these issues.

#### IRON ACCUMULATION IN FRDA

Iron accumulation is a hallmark in FRDA and initially was suggested to be a primary pathogenic event triggered by frataxin deficiency. Tissues in which iron overload has been reported in FRDA patients are showed in **Supplementary Table S1**. The first observations linking iron and FRDA were made in postmortem heart samples from a few patients, where dense intracytoplasmic deposits of iron particles were detected in the cardiac fibers (Sanchez-Casis et al., 1976; Lamarche et al., 1980). These findings were later corroborated in subsequent studies that use different techniques for iron quantitation, such as Perls Prussian blue staining for histological detection or atomic absorption spectroscopy to measure total iron levels in tissue. The excess iron in the heart of FRDA patients appears as multifocal aggregates highly localized rather than as a diffuse pattern (Ramirez et al., 2012; Koeppen et al., 2015). This regional distribution of the excess of iron could be responsible for the findings of iron levels equivalent to healthy controls in bulk extracts or from random biopsied cardiac samples (Michael et al., 2006; Koeppen et al., 2015; Kruger et al., 2016). In addition to the heart, positive iron staining was observed in the liver and spleen of patients, showing a distribution consistent with a mitochondrial location (Bradley et al., 2000) as has been described in the frataxin-deficient yeast model (Babcock et al., 1997; Foury and Cazzalini, 1997).

In regions of the NS, such as DN and DRG, several scenarios regarding the iron content in FRDA have been reported: normal levels (Bradley et al., 2000; Koeppen et al., 2007, 2013; Solbach et al., 2014), accumulation (Bonilha da Silva et al., 2014; Harding et al., 2016), and limited and regional increases of this metal (Koeppen et al., 2009). These diverse results could be partly due to the technique used for iron measurements and their capability to discriminate between the accumulation and redistribution of iron. In most of these studies, the iron content has been quantified from a small number of patient samples, a situation that could also contribute to the different results reported. A more recent study with a higher number of FRDA subjects analyzed used magnetic resonance imaging, which allows in vivo iron quantitation (Harding et al., 2016). The authors found a significant increase in the iron concentration in the FRDA cohort compared with that in controls, in the DN and Red nucleus in the midbrain. However, the technical approach used in this study did not allow the authors to discriminate between deposition and redistribution or whether the location of the iron was extracellular or intracellular. Changes in the distribution of iron and other metals, such as copper and zinc, in the DN of FRDA patients have also been reported (Koeppen et al., 2012), although the total amount of these metals was similar to that of controls. It is worth highlighting that copper can also generate oxygen radicals and, in the context of FRDA, may exacerbate the oxidative injury caused by the increased amount of iron. Mitochondrial iron accumulation has also been documented in fibroblasts and lymphoblasts of FRDA patients. Although some authors have found an increase in the total iron content (Delatycki et al., 1999) or labile iron fraction (Tan et al., 2001), normal levels of iron have been reported in other studies (Wong et al., 1999; Sturm et al., 2005).

Several mouse models of FRDA have been obtained using different experimental strategies, and most of them show iron overload because of frataxin reduction (**Supplementary Table S2**). However, no iron accumulation was detected in embryonic samples of the frataxin knockout (KO) mice (Cossée et al., 2000), and the conditional model of frataxin deficiency in striated muscle (MCK-KO mice) showed heart intramitochondrial iron buildup only in late stages of the disease (Puccio et al., 2001). This led to the hypothesis that the increase in iron is not the causative pathological mechanism in FRDA and is a secondary effect.

Iron accumulation has been studied, particularly in muscle creatine kinase (MCK) conditional Fxn knockout mice (MCK-KO). This mutant develops a severe cardiac phenotype associated with almost ablation of frataxin in heart and skeletal muscle (Puccio et al., 2001). Different studies on this mouse line have reported distinct ages of onset for the iron overload despite using the same iron detection methodology (**Supplementary Table S2**). Some authors (Puccio et al., 2001) found iron deposits in 10-week-old mice, whereas animals 7 weeks of age did not present iron accumulation but showed deficits in ISC enzyme activities. By contrast, other authors (Huang et al., 2013) observed progressive iron accumulation from 5 to 10 weeks of age, with the presence of the iron deposits being the first observable histological change in the MCK-KO hearts. Therefore, there is no consensus about when iron starts to accumulate and whether the presence of iron deposits is relevant for the development of heart pathology in FRDA.

An interesting result found in the MCK-KO model is that iron levels increase in other tissues besides the heart (**Supplementary Table S2**), such as the liver, kidney and spleen, as well as serum (Huang et al., 2009; Whitnall et al., 2012). Because the amount of frataxin in these non-muscular tissues is within the normal range in the MCK-KO model (Huang et al., 2009; Whitnall et al., 2012), the iron increase was explained as an alteration of systemic iron metabolism orchestrated by the frataxin reduction in heart and skeletal muscle. It could be operated through a signaling mechanism involving hepcidin (Huang et al., 2009; Whitnall et al., 2012), a key regulator of systemic iron metabolism in mammals. These explanations indicate that excess iron plays a significant role in the disease. The mitochondrial iron in the heart of MCK-KO mutants is in the form of biomineral iron, which aggregates with phosphorus and sulfur (Whitnall et al., 2012). This is important because excess of iron that is not bound to ferritin, the main cellular protein that stores iron, can be very toxic by generating reactive oxygen species (ROS).

In the liver of mice with hepatic Fxn deficiency (FxnAlb mice), a fivefold increase in the mitochondrial iron level has been observed at 28 days of age compared with that in control mice. Meanwhile, 14-day-old mutant showed deficits in ISC enzymes in the liver but normal levels of iron (Martelli et al., 2015), suggesting that iron accumulation is a secondary event also in the liver. As in FRDA patients, iron deposition in the NS of FRDA mouse models has not been clarified yet. In conditional knockout models with frataxin reduction in neurons and other tissues (Puccio et al., 2001) or in parvalbumin-positive cells in the DRG, cerebellum and brain (Piguet et al., 2018), no iron accumulation has been observed. Other studies have reported few iron deposits in NS late in disease progression (Simon et al., 2004) or that iron accumulates both intracellularly and extracellularly in the brain following the loss of Fxn (Chen et al., 2016a). To understand these diverse findings, the authors have indicated that the different sensitivities of the techniques used to measure iron could be an explanation, or that, in some mouse models, cells loss may occur before sufficient iron is accumulated to be detected.

Other animal models of FRDA have also shown an increase in iron levels in the mitochondria (**Supplementary Table S2**), as is the case of D. melanogaster (Soriano et al., 2013). In this species, severe loss of frataxin function results in iron accumulation in multiple tissues, including the brain (Chen et al., 2016b). This was the first FRDA model clearly demonstrating that iron can accumulate in the NS. Additionally, frataxin-deficient flies also showed increased levels of other transition metals, such as zinc, copper and manganese, suggesting that frataxin reduction affects the homeostasis of other metals besides iron (Soriano et al., 2016). In the plant A. thaliana, frataxin is localized in both the mitochondria and chloroplasts, where it is suggested to have similar functions; its deficiency alters the iron content, and this metal accumulates in the two organelles (Martin et al., 2009; Turowski et al., 2015). These results indicate that frataxindefective organisms show a common propensity to accumulate Fe into organelles where this protein is located. Although many studies have suggested that iron deposits are not instrumental and represent a late event in the pathogenesis of FRDA, the role of Fe in the disease pathophysiology is not fully understood. Recently, frataxin function has been nicely discussed (Alsina et al., 2018), and it can offer some insight into the mechanisms underlying iron accumulation in FRDA tissues and in model organisms. We review this subject in the following section.

# FRATAXIN FUNCTION AND POTENTIAL MECHANISMS OF MITOCHONDRIAL IRON ACCUMULATION

Iron is vital for the cell because it participates in essential biological processes such as DNA synthesis, gene regulation, cellular respiration, and energy production. The mitochondria play a major role in the cellular metabolism of this element and synthesis of heme and ISCs, which are needed for the function of many proteins involved in such cellular processes (Napier et al., 2005; Richardson et al., 2010). Iron homeostasis must be highly controlled because the depletion of its level can produce cellular death, and an increased level of iron promotes the production of ROS. In its Fe+<sup>2</sup> state, iron can interact with oxygen via the Fenton and Haber-Weiss reactions to catalyze the production of ROS, which oxidize and damage proteins, lipids and nucleic acids (Eaton and Qian, 2002). Therefore, precise regulatory systems are needed to ensure the proper amount and oxidative state of iron in the cell and its different compartments.

Although frataxin has been proven to be a mitochondrial ironbinding protein (Foury et al., 2007) involved in the homeostasis of iron, its role remains unclear. Many studies have indicated that this protein could play roles related to iron storage (Park et al., 2003; Huang et al., 2011), heme (Yoon and Cowan, 2004; Napier et al., 2005) and ISC synthesis (Foury et al., 2007; Richardson et al., 2010; Lane et al., 2015) and could function as an iron sensor (Adinolfi et al., 2009) and a metabolic switch (Becker et al., 2002). Some of these possible roles have been more thoroughly tested and proven that others, with the involvement in ISC synthesis being the most accepted one thus far. We discuss these functions next and its possible relationship with mitochondrial iron accumulation in FRDA.

#### Iron Storage

The observation that yeast frataxin (Yfh1), under aerobic conditions and an excess of iron, can form spheroidal oligomers that have ferroxidase activity (Adamec et al., 2000; Gakh et al., 2002; Park et al., 2002) promptly suggested that frataxin could function as an iron storage protein, similar to ferritin (Pandolfo and Pastore, 2009). These Yfh1 multimers have the capacity to bind up to 50 atoms of iron per monomer and maintain it in a redox-inactive state through the ferroxidase activity, which converts redox-active Fe+<sup>2</sup> to Fe+<sup>3</sup> (Gakh et al., 2006). On this line, frataxin could store iron in a bioavailable form in the mitochondria, thus its deficit would lead to iron loading in this organelle.

Frataxins are proteins capable of binding iron. Nuclear magnetic resonance (NMR) studies predominantly mapped the iron-binding sites in the conserved acidic α1 and β1 regions, which contain a clustering of negatively charged aspartate and glutamate residues (reviewed in Bencze et al., 2006;

Pandolfo and Pastore, 2009). However, the hypothesis of frataxin as an iron storage protein presents several problems. Perhaps, one of most serious questions is that, in the case of human frataxin (FXN), the mature form of the protein is not prone to oligomerization. Only the precursor and intermediate forms of FXN can form oligomers and in an iron-independent manner (Adinolfi et al., 2002; O'Neill et al., 2005). Nevertheless, it has lately reported that the mature form of FXN, under specific experimental conditions can form oligomers, but they are unstable (Ahlgren et al., 2017). Other important caveat to this hypothesis is the redundant role that FXN would play with mitochondrial ferritin. In addition, it has been reported that oligomerization-deficient mutant Yfh1 is able to substitute functionally wild-type Yfh1 (Aloria et al., 2004), indicating that oligomerization is not required for frataxin function.

## Heme Biosynthesis

Frataxin can bind iron directly without forming oligomers (Bencze et al., 2006; Cook et al., 2006). Depending on the organism (bacteria, yeast or human) and state of iron (Fe+<sup>2</sup> or Fe+<sup>3</sup> ), different possible iron-binding sites in the protein have been described (Yoon and Cowan, 2003; Bou-Abdallah et al., 2004; Cook et al., 2006; Yoon et al., 2007; Huang et al., 2008). The primary site that binds Fe+<sup>2</sup> involves the residues 18, 19 and 22 of CyaY (and corresponding ones in Yfh1 and FXN) at the acidic ridge in the α1 helix of frataxin. This raised the possibility of new functions for frataxin in irondependent processes, in which some of the involved machinery components seem to interact with FXN. Frataxin interactions have been reported with mitochondrial aconitase, ferrochelatase (Fech), succinate dehydrogenase and the components of the ISC synthesis machinery (Gerber et al., 2003; Yoon and Cowan, 2003; Bulteau et al., 2004; Bencze et al., 2006). Among these interactions, the binding of frataxin with components of the ISC complex has been reproduced and characterized in independent studies (reviewed in Martelli and Puccio, 2014).

The heme biosynthetic pathway is carried out in mitochondria. This pathway includes several enzymes (Sassa, 2006; Furuyama et al., 2007; Chung et al., 2012; Dailey and Meissner, 2013; Chiabrando et al., 2014), with the 5 aminolevulinic acid synthase (ALAS) being the first one in that pathway and the rate-limiting enzyme. ALAS has two isoenzymes, ALAS1 and ALAS2. Whereas ALAS1 expression is negatively regulated by heme directly bound to a cysteineproline dipeptide motif (Furuyama et al., 2007), the ALAS2 gene contains an iron-responsive element (IRE) that interacts with iron regulatory proteins (IRPs) regulating its expression at the posttranscriptional level. Under iron deficiency conditions, the binding of IRP to ALAS2 mRNA inhibits its translation; by contrast, under iron sufficiency conditions, IRPs detach from the IRE, resulting in increased ALAS2 translation. In this manner, the cytosolic iron availability regulates the rate of heme synthesis (Ponka, 1997).

The final enzyme in the heme biosynthetic pathway is Fech, which inserts ferrous iron into protoporphyrin IX. This enzyme contains a [2Fe-2S] cluster that is necessary for enzymatic activity (Wu et al., 2001). The disruption in ISC synthesis could adversely affect the rate of heme synthesis (Richardson et al., 2010; Huang et al., 2011). It has been shown that mitoferrin (Mfrn1) is responsible for iron transport into mitochondria, allowing the direct transfer of ferrous iron for heme and/or ISC synthesis (Fleming, 2011). The role of frataxin in heme synthesis is not clear, although a reduction of heme synthesis was observed in yeast and mouse frataxin mutants (Lesuisse et al., 2003; Huang et al., 2009). Different studies have suggested a direct role of frataxin in the heme synthesis pathway by binding directly to Fech and acting as an iron donor (Yoon and Cowan, 2004; Napier et al., 2005). Thus, this metal would accumulate because of frataxin deficiency.

#### ISC Biogenesis Chaperone

Iron-sulfur clusters are protein cofactors present in almost all living organisms. They play a critical role in several cellular functions, from electron transport in mitochondrial respiratory complexes to DNA repair or metabolism. The de novo synthesis of ISC occurs in the mitochondria (Lill, 2009; Beilschmidt and Puccio, 2014). The components of the ISC synthesis machinery are the cysteine desulfurase (NFS1), which converts L-cysteine to L-alanine with the generation of sulfur, an accessory protein (ISD11), a reductant (ferredoxin), a scaffold protein (ISCU) and an iron delivery protein that is at the core of the debate (Lill and Mühlenhoff, 2006; Py and Barras, 2010). Besides ISD11, recent findings showed that human NFS1 needs another accessory protein, the mitochondrial acyl carrier protein (ACP), making the cysteine desulfurase complex (NFS1-ISD11-ACP) (Van Vranken et al., 2016). In addition, NFS1 requires FXN for its full function (Tsai and Barondeau, 2010; Fox et al., 2015). Regarding the role of FXN in the ISC synthesis, there are two main hypothesis. It was first suggested that frataxin could be the iron donor (Gerber et al., 2003; Yoon and Cowan, 2003; Layer et al., 2006). Using a combination of different techniques based on isothermal titration calorimetry and multinuclear NMR spectroscopy, Cai et al. (2018) have recently showed that FXN can function as an iron donor in vitro. They found that FXN-Fe2<sup>+</sup> interacts weakly to the NFS1-ISD11-ACP complex, but the presence of ISCU largely improves this interaction and that dissociation of Fe2<sup>+</sup> from FXN occurs when the <sup>L</sup>-cysteine and ferredoxin enters into the equation. These results suggest that FXN is the proximal source of iron for ISC assembly, but in vivo characterization is required to confirm this role for frataxin.

Other studies have proposed that frataxin could actually act as a regulator in the synthesis of ISC (Adinolfi et al., 2009; Tsai and Barondeau, 2010; Bridwell-Rabb et al., 2012). The binding of frataxin to the ISCU/NFS1/ISD11 complex can stabilize it, activates the cysteine desulfurase activity and control the entry of iron to the complex (Tsai and Barondeau, 2010; Colin et al., 2013; Pandey et al., 2013).

Mutation of specific aspartate and glutamate residues in the acidic ridge of frataxin, where iron-binding sites were previously mapped, alters the interaction of frataxin with ISCU. Unexpectedly, these mutations does not reduce the total number of iron-binding sites (Schmucker et al., 2011). These results suggested that some of these residues might be critical for frataxin interaction with the core components of the ISC complex, but

not for binding iron. Consequently, some authors (Gentry et al., 2013) proposed that frataxin might have unknown binding sites for this metal. Using different spectroscopic and structural methodologies, a new iron-binding site was identified at His86 in the N-terminus of the mature FXN form, therefore not positioned in the acidic ridge. Interestingly, this residue was essential for iron binding and for in vitro productive synthesis of ISC by ISCU, without showing a direct interaction with the scaffold protein (Gentry et al., 2013). These results support the proposal that frataxin is the iron donor for ISC assembly.

Extramitochondrial frataxin has been shown to physically interact with IscU1, the cytosolic isoform of the human ISC, indicating a role for frataxin in the Fe/S cluster assembly in the cytosol (Acquaviva et al., 2005). The cytosolic form of frataxin has also been reported to interact and possibly regulate the function of the cytosolic aconitase/iron regulatory protein-1 (IRP1) (Condò et al., 2010). Other functions suggested for frataxin in the cytosol include preventing oxidative stress and apoptosis (Condò et al., 2006); and in the nucleus, frataxin could promote DNA repair and telomere stability (Lill et al., 2015).

#### Iron Sensor and Metabolic Switch

As indicate above, frataxin could act as an iron sensor to tightly regulate ISC synthesis (Adinolfi et al., 2009). In E. coli, CyaY, far from facilitating iron delivery, can inhibit the ISC synthesis process. With normal iron levels, CyaY has a low affinity for the IscS-IscU system (the bacterial ISC synthesis machinery), and ISC is loaded to the final acceptor. However, under conditions of iron excess, the affinity of CyaY for IscS would increase, slowing the rate of ISC synthesis and avoiding an overproduction of ISC. This would avoid the degradation of ISC not inserted into an acceptor protein and the release of free iron that could catalyze the production of ROS. However, the consequence of frataxin deficiency in FRDA patients is a decrease in ISC protein function, suggesting that FXN might operate as an activator of the ISC synthesis. In vitro studies also support the role of FXN as an allosteric activator of the ISC assembly (Tsai and Barondeau, 2010). In contrast, both FXN and CyaY can complement a yeast strain with deletion of Yfh1, indicating that eukaryotic and bacterial frataxins can at least partially conserve their functions (Cavadini et al., 2000; Bedekovics et al., 2007). To clarify this controversial results, Bridwell-Rabb et al. (2012) conducted enzyme kinetic assays in which analogous components of the human and E. coli ISC complex were interchanged. Interestingly, they found that CyaY could substitute FXN as an activator of the cysteine desulfurase in the human ISC complex. Conversely, FXN could also replace CyaY as an inhibitor in the assembly of the bacterial ISC. Therefore, both frataxin proteins have the property to modulate the ISC formation, but the mechanism of action of any of them, as an activator or inhibitor of ISC biogenesis, depends on another component of the ISC assembly machinery. This component was found to be the cysteine desulfurase, which controls how the frataxin protein operates in the ISC complex (Bridwell-Rabb et al., 2012). Differences between the human and bacteria cysteine desulfurase might explain the different role of frataxin in humans with respect to E. coli and by extension in eukaryotes vs prokaryotes. Further studies are needed to define the nature of these differences that dictate how frataxin acts in the process of the ISC synthesis.

Examination of the role of frataxin in ISC and heme metabolism in differentiating erythroid cells has suggested that frataxin acts as a metabolic switch in the diversion of iron from one metabolic pathway to another. This was supported by the observation that an increased level of protoporphyrin IX (PPIX) leads to the downregulation of frataxin expression and potential diversion of iron from ISC synthesis toward heme synthesis (Becker et al., 2002). The molar ratio between frataxin and Fech affects the rate of heme synthesis (Yoon and Cowan, 2004), and these findings have suggested that frataxin could function as a metabolic switch in regulating mitochondrial iron metabolism, independently of the expression levels of PPIX or Fech. This metabolic switch allows the mitochondria to favor heme or ISC syntheses, depending on the frataxin levels. The possible mechanism behind this switching may be the relative affinities of frataxin for Fech vs ISCU. This may result in the preferential binding of frataxin to Fech over ISCU at relatively lower frataxin levels (Richardson et al., 2010).

When the frataxin level is reduced, there would be no control in the rate of ISC formation, causing iron accumulation (Adinolfi et al., 2009). Regarding the metabolic switch role of frataxin, its depletion would provoke the enhancement in the reduction of ISC formation at the ISCU machinery level and at the end, subjecting the cell to increased iron accumulation and ROS production (Yoon and Cowan, 2004).

# Dysregulation of Iron Homeostasis in FRDA

Although the mechanisms responsible for mitochondrial iron loading in FRDA are not fully understood, there are several reasonable hypotheses that could explain this alteration. In the MCK-KO mouse, Huang et al. reported altered heme and ISC synthesis and mitochondrial iron storage. Significantly diminished activity and expression of ISC enzymes, such as NFS1 and ISCU1/2, were observed that agreed with the reduction in the expression of succinate dehydrogenase complex, subunit A (Sdha) (Huang et al., 2009). At the same time, other studies have shown a downregulation in the number of key heme synthesis enzymes and decreased heme levels in the heart of the MCK KO model (Schoenfeld et al., 2005; Huang et al., 2009). A decrease in the synthesis of different products that require iron in the mitochondria could explain the accumulation of this metal because it cannot be exported from the organelle being part of those products. However, frataxin deficiency can also affect the expression of several other proteins, some of them critical for the proper control of iron metabolism (**Figure 1**). In DN, Koeppen et al. (2007) observed modification of the expression of proteins such as transferrin receptor 1 (TFR1), ferritins (FRTs) and ferroportin (FPN), alterations shared among animal models of FRDA. In the MCK-KO mouse (Martelli and Puccio, 2014), has been described marked TFR1 upregulation, leading to increased iron uptake, and the downregulation of FPN, decreasing cellular iron release from the cytosol. There are also decreased levels of ferritin light chain 1 (FRTL1)

and heavy chain 1 (FRTH), implying reduced cytosolic iron incorporation and storage. Finally, an increased expression of Sec15l1 (an exocyst complex protein) and the mitochondrial iron importer mitoferrin-2 (MFRN2) in the mutant could facilitate mitochondrial iron influx, which was seen to be enhanced in this model (Whitnall et al., 2008).

The link between frataxin deficiency and alteration in the expression of some of these proteins could be IRP1 and IRP2. Both proteins bind to IRE elements in the 5<sup>0</sup> and 3<sup>0</sup> UTR of the mRNAs of iron metabolism-related proteins (Chiang et al., 2016). When bound to a 5<sup>0</sup> IRE, IRP1 and IRP2 repress the translation of the mRNAs while binding of the IRPs to a 3<sup>0</sup> IRE increases the stability and translation of the mRNAs. The activity of both IRPs is increased by the low levels of cellular iron by different mechanisms. Interestingly, IRP1 activity is controlled by the insertion of an ISC, which converts the protein into a cytosolic aconitase and makes it unable to bind to IRE elements. Since frataxin depletion reduces ISC synthesis, it is reasonable to suggest a role for IRP1 in the deregulation of iron metabolism. Nevertheless, the activity of IRP2, which is controlled by irondependent controlled ubiquitination and degradation, is also increased in MCK-KO mouse, possibly due to cytosolic iron depletion (Whitnall et al., 2008). More recently, Martelli et al. (2015) studied a double KO mouse with complete ablation of Irp1 and liver-specific deletion of Fxn. In these mice, the authors observed that alterations in the expression of proteins such as TFR1 and divalent metal transporter 1 (DMT1) were indeed due, a least partially, to IRP1, while the regulatory protein seemed to exert little effect on the expression of other proteins such as FRTL and FPN. Other models of the disease also manifest similar alterations in the expression of proteins involved in iron metabolism. Frataxin depletion in D. melanogaster produces alterations in the expression of mitoferrin, IRP-1A and ferritin, while genetic modifiers for Malvolio, Tsf1 and Tsf3 (Drosophila homologs of DMT1 and transferrins) and IRP-1A and IRP-1B could correct the motor deficiency phenotype of frataxindeficient flies (Soriano et al., 2016; Calap-Quintana et al., 2018; Monnier et al., 2018). Although some hints point to the possible causes of iron dysregulation in FRDA, more research is needed to propose a more precise and final hypothesis.

# MECHANISMS OF IRON NEUROTOXICITY IN FRDA

Initially, it was postulated that iron overload in FRDA activates the formation of ROS via the Fenton reaction, which would arbitrate the pathogenesis of the disease. The highly redox active environment in the mitochondria could promote ROS generation when the free iron content increases. Nevertheless, an ROS independent mechanism mediated by iron has recently been proposed for FRDA neurodegeneration. These two mechanisms are described in the following paragraphs (**Figure 2**).

# ROS-Dependent Mechanism of Iron Neurotoxicity

Iron accumulation in the NS is a common feature in neurodegenerative diseases, including Alzheimer's disease (Connor et al., 1992), Parkinson's disease (Gorell et al., 1995), Huntington's disease (Dumas et al., 2012) or amyotrophic lateral sclerosis (Kwan et al., 2012). In these disorders, iron deposits are observed in specific regions of the NS. As indicated previously, in Friedreich ataxia patients, deposits of iron have not been clearly reported in NS, and the increment of iron levels in the DN remains controversial (Waldvogel et al., 1999; Koeppen et al., 2007; Harding et al., 2016). This region is interesting because the patients develop neuronal atrophy and abnormal proliferation of the synaptic terminal, called grumose degeneration. The Koeppen's group confirmed the presence of iron metabolism alteration in the NS of FRDA patients by changes in the expression of proteins related to iron, such as FPN, FRT and TFR1, in the DN (Koeppen et al., 2012) and DRG (Koeppen et al., 2009). They observed a relocation of iron from the affected neurons to the microglia, in the case of DN, or from neurons to the satellite cells in the DRGs. The increment of iron levels correlates with the production of ROS in these areas, suggesting that alterations of iron homeostasis could contribute to the pathogenesis of the disease through ROS production. Additionally, frataxin deficiency is related to low antioxidant defenses (Chantrel-Groussard et al., 2001; Paupe et al., 2009), making the nervous cells have an oxidative stress-prone phenotype.

The toxicity of iron may be due to its localization, biochemical form or cell ability to prevent the generation and propagation of iron. In living organisms, iron can be found in its reduced form, ferrous iron (Fe2+), and its oxidized form, ferric iron (Fe3+). Most iron binds, as a prosthetic group, to proteins such as hemoglobin, myoglobin, cytochromes or storage proteins; only less than 5% of iron in the cell is in the labile iron pool

(LIP) (Li and Reichmann, 2016). The cellular LIP is defined as a pool of chelatable and redox-active iron complexes (Kakhlon and Cabantchik, 2002). LIP has the capacity to participate in the formation of harmful free radicals including the hydroxyl radical. Fe2<sup>+</sup> is catalyzed to Fe3<sup>+</sup> in a chemical reaction mediated by hydrogen peroxide (H2O2), which leads to the formation of hydroxide radical (OH•). Cytotoxic events are started in the cell via the Fenton reaction. To avoid the negative effects of excess free iron, it is coupled to proteins such as FRT, which is responsible for iron storage inside the cell. In the satellite cells of DRG from FRDA patients, an increment of cytosolic FRT (Koeppen et al., 2009) was observed; in DN, ferritin is expressed mostly in oligodendrocytes, satellite cells around the neurons. As the disease progresses, the neurons begin to undergo atrophy, and these cells disappear, producing microglia FRT-positive cells (Koeppen et al., 2007). Noteworthy, mitochondrial FRT has not been detected in DRG neurons, a protein responsible to store free mitochondrial iron and with a very important role in antioxidant activity (Koeppen et al., 2007; Campanella et al., 2009). FPN is a transmembrane protein that exports iron and is expressed in the central NS and DRG. Interestingly, in the DRG from FRDA patients, the expression of FPN is exclusive to satellite cells (Koeppen et al., 2009), probably as a response to the iron release from DRG neurons to these cells (Koeppen et al., 2016). In the synaptic terminals of neurons from the DN of FRDA patients, is observed an increment of the ferroportin levels (Koeppen et al., 2007). The nerve endings present an elevated number of mitochondria, suggesting a response in the biosynthesis of this protein due to the dyshomeostasis of iron. Thus, the toxicity of this metal could be an important factor in grumose degeneration.

Because mitochondria are an important generator of ROS, the continuous presence of iron in the mitochondrial matrix makes it susceptible to new production of ROS. Proteins with ISC clusters, such as complex I, II and III of the electron transport chain, can be damaged by ROS. These complexes show reduced activity in FRDA patients (Rötig et al., 1997), probably because of iron toxicity. Moreover, these proteins will generate new free iron that will participate in the Fenton reaction and carry out new oxidative processes. In plasma and urine from FRDA patients (Emond et al., 2000; Schulz et al., 2000), similar to in all the disease models generated (González-Cabo and Palau, 2013), oxidative stress markers have been detected.

In the last years, neuronal models of the disease have been developed to better understand the neurodegenerative process. Oxidative stress, mitochondrial dysfunction, and iron dyshomeostasis are features shared among the neuronal models. For example, the granulate cells of the cerebellum from YG8R mice show an increment of ROS, lipid peroxidation and reduced glutathione levels (Bradshaw et al., 2016); the neurons of DRG from YG8R mice present an increment of ROS, dysfunctional mitochondria, and calcium dyshomeostasis (Mollá et al., 2017); the motor neuron cell line (NSC34) with reduced expression of the frataxin gene shows an increment in the GSSG/GSH ratio and more production of ROS (Piermarini et al., 2016) and alteration in complex I activity of the electron transport chain (Carletti et al., 2014); reduction of Drosophila frataxin (fh) expression in glial cells produces massive accumulation of fatty acids and increased sensitivity to oxidative stress (Navarro et al., 2010).

Under oxidative stress, the transcription factor Nrf2 binds to antioxidant response element (ARE) and modulates the expression of many genes involved in the antioxidant defense. It has been reported that Nrf2 signaling is defective in fibroblasts from FRDA patients (Paupe et al., 2009), the DRG and cerebellum of YG8R mice (Shan et al., 2013) and motor neuron and

Llorens et al. Iron in Friedreich's Ataxia

Schwann cell models (D'Oria et al., 2013; Shan et al., 2013), making these cells and tissues more vulnerable to oxidative stress damage and the neurodegeneration process. Concretely, frataxin deficiency in neurons promotes axonal dystrophy with multiple axonal spheroid formation mainly due to Ca2<sup>+</sup> imbalance and oxidative stress (Mollá et al., 2017). The axonal swellings are associated with specific cytoskeletal changes (Mincheva-Tasheva et al., 2014; Piermarini et al., 2016; Mollá et al., 2017), improper axonal transport and autophagic flux (Shidara and Hollenbeck, 2010; Mollá et al., 2017). Moreover, the oxidative stress damage effects to the neurites growth leads to shorter dendrites due to alterations in the dynamics of the microtubules (Shan et al., 2013; Carletti et al., 2014; Piermarini et al., 2016). When oxidative stress is reduced in these neurons, either by direct addition of antioxidants (Piermarini et al., 2016) or through drugs that increase the expression of Nrf2 (Petrillo et al., 2017), axonal growth is improved. In any case, these alterations affect the cellular viability of both neurons and glia, triggering cell death processes either by apoptosis (Mincheva-Tasheva et al., 2014; Igoillo-Esteve et al., 2015; Katsu-Jiménez et al., 2016) or autophagy (Bolinches-Amorós et al., 2014; Edenharter et al., 2018). It has been shown that iron is a key element in the induction of cell death (Dixon et al., 2012). This new process, called ferroptosis, is dependent on iron and is accompanied by an increase in ROS and lipid peroxidation and a decrease in GSH. As we have reviewed previously, these characteristics are observed in frataxin-deficient cells; therefore, new studies are needed to determine the possible cell death due to a ferroptosis process.

## ROS-Independent Mechanisms of Iron Neurotoxicity

The extent to which oxidative stress is necessary for the pathogenesis of FRDA is, however, controversial. The deficit in ISC enzymes and increase in iron levels were not paired with an increase in ROS in the MCK-KO mouse (Seznec et al., 2005), in FxnAlb mice with specific depletion of Fxn in liver (Martelli et al., 2015), and in the inducible and reversible FRDAkd mouse model (Chandran et al., 2017). Similarly, increased oxidative stress was not detected in several Drosophila knockdown models, and overexpression of ROS scavengers did not rescue the viability, heart dysfunction or neurodegeneration phenotypes (Anderson et al., 2005; Shidara and Hollenbeck, 2010; Tricoire et al., 2014; Chen et al., 2016b). Moreover, FRDA neural tissue did not show differences in the susceptibility to oxidative stress in a study using patient samples (Macevilly and Muller, 1997). The absence of ROS markers in these studies might be attributed to differences in methodology or the distinctive features of each of the animal models or tissues employed. Alternatively, it was suggested that the discrepancies in the presence of ROS would be explained by differences in the timepoints studied, in a scenario where oxidative stress would be a secondary effect rather than a primary cause, occurring later in the progression of FRDA (discussed in Lupoli et al., 2018). Recent findings in the fly and mouse models of FRDA actually support a mechanism that is independent of ROS that links neurodegeneration and iron accumulation via the sphingolipid pathway (Chen et al., 2016a,b).

Alterations in the homeostasis of lipids are not exclusive of FRDA (Huang et al., 1978; Filla et al., 1980; Thurman and Draper, 1989; Raman S.V. et al., 2011; Worth et al., 2014) and have been reported in neurodegenerative disorders, including Parkinson's, Alzheimer's, Huntington's, multiple sclerosis and Niemann–Pick diseases (reviewed in Shamim et al., 2018). In the case of FRDA, lipid droplet accumulation was first described in the cardiac muscle of the NSE-KO mouse model (Puccio et al., 2001) and was further characterized as pathological in a glial frataxin knockdown model in Drosophila (Navarro et al., 2010). The accumulation and dyshomeostasis of lipids have been reported as well in several other cellular and animal models of the disease (Martelli et al., 2012; Schiavi et al., 2013; Obis et al., 2014). More recently, the expression of a mutant allele of the Drosophila fh in mosaic clones produced an age-dependent degeneration of photoreceptor neurons with impaired mitochondrial function, increased levels of Fe2<sup>+</sup> and Fe3<sup>+</sup> in multiple tissues and dramatic lipid droplet accumulation in glial cells (Chen et al., 2016b), consistent with the work from Navarro et al. (2010). Because the Fenton reaction can induce lipid peroxidation, oxidative stress has been classically proposed to be the link between abnormal lipid homeostasis and neurodegeneration in FRDA. However, Chen et al. did not observe an increase in free radical production and, neither providing the antioxidant AD4 nor overexpressing the ROS scavengers catalase, SOD1 and SOD2, suppressed neurodegeneration (Chen et al., 2016b). Moreover, reducing the number of lipid droplets by overexpressing brummer lipase or lipase 4 was insufficient to improve the neurodegeneration defect in the knockout photoreceptors. Because high iron conditions increase sphingolipid synthesis in wild-type yeast (Lee et al., 2012), Chen et al. (2016b) pursued the characterization of this pathway in their fly model. The synthesis of sphingolipids, including dihydrosphingosine, dihydroceramide and sphingosine, was actually increased in the fh mutants. Both the genetic knockdown and chemical inhibition of the fly serine palmitoyltransferase, an enzyme essential for the de novo synthesis of sphingolipids, rescued neurodegeneration. Additionally, they found that downstream targets of the transcription factor Mef2 were upregulated in response to the increase in sphingolipids in fh mutants and knocking down the Pdk1/Mef2 pathway led to the suppression of photoreceptor degeneration. To ask whether this novel pathway described in Drosophila was conserved in vertebrates, Chen et al. (2016a) developed a mouse model with reduced Fxn levels in the NS. Both ferrous and ferric iron were found to be increased in the cerebral cortex and, consistent with their observation in the fly model, they could not find increased ROS production measured as levels of lipid peroxidation. Loss of Fxn in the NS led to the activation of the PDK1/Mef2 pathway in the mouse before the appearance of the neurological manifestations. Sphingolipids and PDK1 activation were also found to be increased in cardiac tissue from FRDA patients (Chen et al., 2016a). Nevertheless, alterations in the PDK1/Mef2 pathway were not replicated in the FRDAkd-inducible mouse model of FRDA (Chandran et al., 2017). In this study, the authors did not find changes in PDK phosphorylation in the brain, muscle, heart or liver in the chronic

adult knockdown mice. The levels in the cerebellum and heart of five target genes of Mef2 were analyzed as well and displayed changes that were inconsistent across tissues. Alternatively, other biochemical pathways could be altered in relation to iron under frataxin-deficient conditions. In line, several genes involved in hemochromatosis, an iron-overload disorder, are upregulated in the FRDAkd-inducible mouse model, in which no increase in ROS and no strong evidence of Pdk1/Mef2 activation were observed (Chandran et al., 2017).

## POTENTIAL TREATMENT OF FRDA BY REDUCING IRON TOXICITY

An optimal therapy in FRDA could consist of providing patients with sufficient frataxin amount to restore the functions of this protein. Several strategies are being developed to obtain this objective, such as gene therapy (Perdomini et al., 2014; Piguet et al., 2018), transplanting hematopoietic stem and progenitor cells (Rocca et al., 2017) or improving FXN transcription (Soragni and Gottesfeld, 2016). While advancing the development of such strategies for its application to patients, rational treatment could be considered to remove excess iron.

The utilization of iron chelators is one of the first strategies proposed to treat FRDA patients to eliminate excess iron accumulation in the mitochondria. To that end, an iron chelator should be able to cross the cell membranes, access mitochondria and redistribute the excess labile iron. Deferiprone (DFP, 1, 2 dimethyl-3-hydroxy-pyrid-4-one), an orally administered lipid soluble iron chelator can easily cross the blood–brain barrier and cellular membranes. DFP shows relatively low affinity for iron and can transfer this metal to circulating transferrin, as well as to cellular acceptors—e.g., for heme synthesis (Sohn et al., 2008). Therefore, its use is less probable to produce iron depletion than other known chelators. For all these reasons, DFP has been considered as the prototypic compound for iron redeployment (Cabantchik et al., 2013). It is also used to treat iron-overloaded patients with β-thalassemia who need continuous blood transfusions (Totadri et al., 2015), and their properties make it an interesting therapeutic compound for FRDA.

The use of DFP in cellular and animal models of FRDA has beneficial results, e.g., frataxin-deficient HEK-293 cells could restore mitochondrial dysfunction after DFP treatment. It was attributed to the ability of this compound to make available the iron accumulated into mitochondria (Kakhlon et al., 2008). These conclusions were also supported by our group using a Drosophila FRDA model, in which DFP improved the survival and climbing ability phenotypes because of the removal of excess free mitochondrial iron, thereby preventing its toxic effects (Soriano et al., 2013). Recently, it has been reported that DFP is effective to reduce ROS production and improve calcium handling kinetics in an FRDA humaninduced pluripotent stem cell-derived cardiomyocyte model (Lee et al., 2016).

The efficacy of DFP has already been assessed in clinical trials in FRDA patients, but the results are inconclusive. However, considering these studies, DFP has shown beneficial effects on cardiac parameters achieved at low doses of the compound, while treatment with higher doses of DFP could aggravate the disorder by affecting the ataxia phenotype (Pandolfo et al., 2014). A high dose of DFP could remove too much iron, provoking a deficiency of this metal, which could affect the cellular functions that require iron. On the other hand, the positive action of this drug on the heart compared with NS has also been related to the initiation of treatment before cardiac dysfunction is present (Elincx-Benizri et al., 2016). Overall, DFP seems to have therapeutic value in FRDA and is quite safe at lower doses, but systematic, large scale, placebo controlled longer studies are needed to evaluate its usefulness (Strawser et al., 2017). In line with these issues, our results using a Drosophila model of the disease indicate that DFP treatments at the larval stage recover FRDA-like phenotypes in flies, but no improvement was observed when the drug is administered to adults (Soriano et al., 2013). Additionally, we found that reducing the excess of iron accumulating into mitochondria through a genetic strategy has benefit in this FRDA model (Navarro et al., 2015; Soriano et al., 2016). These results show a causal relationship between iron and frataxin deficiency. Thus, iron is a critical factor in the pathophysiology of FRDA and appears not be a simple epiphenomenal event in the disease. Consequently, research on how to deal with the iron excess and eradicate the iron toxicity is still an area of great interest.

In the last decades, some small molecules with pharmacophore characteristics for iron chelation are being explored for treatment in neurodegenerative diseases with regional increase of iron levels (reviewed in Singh et al., 2018). These compounds are derivatives of the 8-hydroxyquinoline, hydroxypyridone, and arylhydrazones, and some of them are able to reduce selectively the iron excess in the mitochondria (Mena et al., 2015). The use of pro-chelating compounds, which are activated in the target region, have the benefit to remove the excess of iron in the affected regions. Thus, this strategy would protect cells in non-affected areas from the risk of iron subtraction. Novel pro-chelators triggered by H2O<sup>2</sup> were obtained as promising candidates for management of Parkinson's disease (Perez et al., 2008). Similarly, this approach might be useful in FRDA since H2O<sup>2</sup> is a critical pathogenic agent associated to frataxin deficiency (Anderson et al., 2008). Finally, new compounds with multifunctional actions, as previously indicated for DFP, are also being examined for effective treatment of neurodegenerative diseases (Singh et al., 2018). Several of these diseases, including FRDA, share some pathological features such as regional iron increase, oxidative stress and dysregulation of calcium signaling. Recently, a novel compound, named CT51 (N-(1,3-dihydroxy-2-(hydroxymethyl)propan-2 yl)-2-(7-hydroxy-2-oxo-2H-chromen-4 yl) acetamide), has been synthesized that seems to suppress these cellular impairments (García-Beltrán et al., 2017). CT51 can easily permeate the cell membrane and distribute to both mitochondria and the cytoplasm. It works as a highly selective iron chelator and possesses significant capability for free radical quenching, protecting cells against oxidative damage. The use of CT51 in primary hippocampal neurons decreases the continuous release

of calcium provoked by an agonist of ryanodine receptorcalcium channels, supporting a protective role of this compound in neuronal function. Although developments of this kind of drugs are not a curative strategy, because the frataxin deficiency will persist, they might improve the current pharmacological treatments and might produce increases in quality of life and slowing of disease progression.

#### CONCLUSION

The relationship between neurodegeneration and iron accumulation is well established. However, in Friedreich's ataxia, it remains unclear whether iron loading is the cause or consequence of the pathophysiology. The studies in model organisms suggested both hypotheses in relation to the role of iron in this disease: a key role in the pathophysiologic mechanisms or a secondary event that appeared late in the disease progression. It was also debated the extent to which oxidative stress is crucial to explain the pathogenesis of FRDA. Indeed, additional pathways might be involved in the toxicity of iron independent of ROS production. Differences in the distinctive features of each of the FRDA models studied, vulnerability of the different tissues and cells to frataxin deficit and timepoints analyzed could partially explain the contradictory results. Nevertheless, from the therapeutic point of view, it is interesting to explore whether new iron chelators showing

#### REFERENCES


protective properties on cellular function could improve the mitochondrial dysfunction in FRDA.

#### AUTHOR CONTRIBUTIONS

JL and MM planned the manuscript. JL, SS, PC-Q, PG-C, and MM wrote the manuscript and agreed on the finally submitted version of the manuscript.

## FUNDING

Financial support of this work came from Generalitat Valenciana, Spain (PROMETEOII/2014/067) to MM, the Spanish Ministry of Economy and Competitiveness [SAF2015-66625-R] within the framework of the National R+D+I Plan to PG-C, Fundación Ramón Areces, Spain (CIVP18A3899) to PG-C and MM. PC-Q was a recipient of a fellowship from Generalitat Valenciana, Spain, and JL is supported by a research contract from FARA and FARA Ireland.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.00075/full#supplementary-material


marker in Friedreich's ataxia. Eur. J. Neurol. 21, 1131–1136. doi: 10.1111/ene. 12448


activation, leading to neurodegeneration. eLife 5:e16043. doi: 10.7554/eLife. 16043




cardiomyopathy by gene therapy in a mouse model of Friedreich's ataxia. Nat. Med. 20, 542–547. doi: 10.1038/nm.3510



chelation. Proc. Natl. Acad. Sci. U.S.A. 105, 9757–9762. doi: 10.1073/pnas.08042 61105


**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 Llorens, Soriano, Calap-Quintana, Gonzalez-Cabo and Moltó. 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.

# Deciphering the Iron Side of Stroke: Neurodegeneration at the Crossroads Between Iron Dyshomeostasis, Excitotoxicity, and Ferroptosis

Núria DeGregorio-Rocasolano<sup>1</sup> , Octavi Martí-Sistac1,2 \* and Teresa Gasull<sup>1</sup> \*

<sup>1</sup> Cellular and Molecular Neurobiology Research Group, Department of Neurosciences, Germans Trias i Pujol Research Institute (IGTP), Badalona, Spain, <sup>2</sup> Department of Cellular Biology, Physiology and Immunology, Universitat Autònoma de Barcelona, Bellaterra, Spain

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by:

Alessandro Fanzani, Università degli Studi di Brescia, Italy Ashley Ian Bush, The Florey Institute of Neuroscience and Mental Health and The University of Melbourne, Australia Gladys Oluyemisi Latunde-Dada, King's College London, United Kingdom

#### \*Correspondence:

Octavi Martí-Sistac octavi.marti@uab.cat Teresa Gasull tgasull@igtp.cat; teresagasull@yahoo.com

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 28 October 2018 Accepted: 25 January 2019 Published: 19 February 2019

#### Citation:

DeGregorio-Rocasolano N, Martí-Sistac O and Gasull T (2019) Deciphering the Iron Side of Stroke: Neurodegeneration at the Crossroads Between Iron Dyshomeostasis, Excitotoxicity, and Ferroptosis. Front. Neurosci. 13:85. doi: 10.3389/fnins.2019.00085 In general, iron represents a double-edged sword in metabolism in most tissues, especially in the brain. Although the high metabolic demands of brain cells require iron as a redox-active metal for ATP-producing enzymes, the brain is highly vulnerable to the devastating consequences of excessive iron-induced oxidative stress and, as recently found, to ferroptosis as well. The blood–brain barrier (BBB) protects the brain from fluctuations in systemic iron. Under pathological conditions, especially in acute brain pathologies such as stroke, the BBB is disrupted, and iron pools from the blood gain sudden access to the brain parenchyma, which is crucial in mediating strokeinduced neurodegeneration. Each brain cell type reacts with changes in their expression of proteins involved in iron uptake, efflux, storage, and mobilization to preserve its internal iron homeostasis, with specific organelles such as mitochondria showing specialized responses. However, during ischemia, neurons are challenged with excess extracellular glutamate in the presence of high levels of extracellular iron; this causes glutamate receptor overactivation that boosts neuronal iron uptake and a subsequent overproduction of membrane peroxides. This glutamate-driven neuronal death can be attenuated by iron-chelating compounds or free radical scavenger molecules. Moreover, vascular wall rupture in hemorrhagic stroke results in the accumulation and lysis of iron-rich red blood cells at the brain parenchyma and the subsequent presence of hemoglobin and heme iron at the extracellular milieu, thereby contributing to ironinduced lipid peroxidation and cell death. This review summarizes recent progresses made in understanding the ferroptosis component underlying both ischemic and hemorrhagic stroke subtypes.

Keywords: iron, reactive oxygen species, ferroptosis, stroke, excitotoxicity, iron dyshomeostasis, neurodegeneration, transferrin saturation

**Abbreviations:** AIS, acute ischemic stroke; ALOX5: arachidonic acid lipoxygenase 5; APP, amyloid precursor protein; BBB, brain–blood barrier; CSF, cerebrospinal fluid; Dexras1, dexamethasone-induced Ras protein 1; DFO, deferoxamine; DMT1, divalent metal transporter 1; EAATs, excitatory amino acid transporters; FPN, ferroportin; FT, ferritin; FTH, ferritin heavy chain; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; GPX4, glutathione peroxidase 4; GSH, glutathione; Hb, hemoglobin; HO, heme oxygenase; Hp, haptoglobin; Hpx, hemopexin; HTf, holotransferrin; ICH, intracerebral hemorrhage; I/R, ischemia/reperfusion; IREs, iron regulatory elements; IRPs, iron regulatory proteins; NAC, N-acetylcysteine; NMDA, N-methyl-D-aspartate; NTBI, non-transferrin-bound iron; OGD, oxygen and glucose deprivation; PUFA, polyunsaturated fatty acids; ROS, reactive oxygen species; TIM, T-cell immunoglobulin and mucin domain-containing receptor; Tf, transferrin; TfR1, transferrin receptor 1; TfR2, transferrin receptor 2; TSAT, saturation of transferrin with iron; UTRs, untranslated regions; X<sup>−</sup> c , cystine/glutamate exchanger or antiporter, i.e., SLC7A11/xCT.

#### IRON TRANSPORT TO THE BRAIN: THE ROLE OF THE BLOOD–BRAIN BARRIER

Iron is essential for life; numerous proteins require iron as a cofactor for their activity. Iron acts as an electron donor or acceptor, and thus is pivotal for oxygen transport (coordinated with hemoglobin), cellular respiration (as a part of hemecontaining cytochromes and proteins containing Fe–S groups in the electron transport chain), and DNA synthesis (in ribonuclease reductase). Moreover, in the nervous system, iron is essential for myelinization and neurotransmitter biosynthesis.

Iron exists both as Fe2<sup>+</sup> and Fe3+; redox-active and cycling between these two states lead to the generation of reactive oxygen species (ROS) through the Fenton reaction. The ROS promote oxidative stress and cause extensive lipid peroxidation (Gutteridge et al., 1979). The brain is particularly vulnerable to lipid peroxidation damage because it is rich in polyunsaturated fatty acids (PUFAs) and iron, but relatively poor in antioxidant defenses (Cobley et al., 2018). The iron content remarkably differs across different brain regions: it is high in the striatum, medium in the hippocampus and cortex, and low in the pons and medulla (Ramos et al., 2014). Although it has recently attracted considerable attention, iron metabolism in the brain is yet not completely understood as that in other murine tissues (Ashraf et al., 2018). Brain imports iron from systemic circulation, and the levels of iron are determined by diet, intestinal iron absorption, release of iron initially stored in the hepatocytes, and export of iron recycled from macrophages. In addition, the levels of iron in the plasma are regulated by the hormone hepcidin through its effect on the transmembrane iron export protein ferroportin (FPN), which is mainly present at the cell membrane of macrophages, hepatocytes, and enterocytes.

In the blood and most other extracellular fluids, iron is essentially carried bound with high affinity to the iron transport protein transferrin (Tf), which shields Fe3<sup>+</sup> from redox activity. In addition to Tf, iron also associates with other carrier proteins in the blood such as ferritin (FT), haptoglobin (Hp), Hpx, or albumin. Recently, half a dozen different non-proteinaceous lowmolecular-weight iron species (<10 KDa) have been shown to exist in the blood (Dziuba and Lindahl, 2018); further studies are required to determine the precise identity and physiological role of these species. Once iron is bound to Tf, Fe3<sup>+</sup> is rendered redox-inactive and virtually non-exchangeable or displaceable by other physiological metals or molecules. Tf−iron complexes circulate in the blood until they bind transferrin receptors (TfRs) in cellular membranes of target cells. In general, in non-ironoverload conditions, little, if any, non-transferrin-bound iron (NTBI) is found in the blood of healthy individuals (Gaasch et al., 2007). The high affinity of Tf for iron and the large amount of Tf in the blood exceed the total requirement for iron binding; in normal human blood, the saturation of transferrin with iron (TSAT) is only of around 30% (De Valk et al., 2000; Van Dijk et al., 2008). Therefore, iron transport to the brain is mainly from iron-loaded transferrin (holotransferrin, HTf) through the endothelial cells of the BBB and the epithelium of the choroid plexuses and arachnoid membrane, the blood–cerebrospinal fluid (CSF) barrier.

Although barrier cells of the brain form tight junctions laterally, epithelial cells of the choroid plexuses are fenestrated and more permeable than the endothelial cells of the BBB (Strazielle and Ghersi-Egea, 2016). The abluminal side of brain capillary endothelial cells is surrounded by pericytes and astrocytes of the neurovascular unit that provide extra structural and functional support. TfRs are abundantly expressed in the luminal side of membranes of polarized endothelial cells; once they bind Tf, the TfR–Tf complexes undergo endocytosis. The following steps are still controversial, and different pathways are thought to be involved in brain iron acquisition (Bradbury, 1997; Khan et al., 2018). Thus far, the main contribution to this process is thought to include the following pathways: (1) endocytic vesicles containing TfR–HTf traveling from the luminal membrane to fuse to the abluminal membrane and exocytose either HTf or free iron into the brain extracellular space; (2) under low pH conditions within the endosome of endothelial cells, as in most cell types, iron is extracted from Tf, converted to Fe2<sup>+</sup> by endosomal ferrireductases such as the six-transmembrane epithelial antigen of prostate protein 3 (Steap3), and released to the cytosol through divalent metal transporter 1 (DMT1) or the ion channels TRPML, also known as mucolipin 1 or MCOLN1 (detailed information can be found in references Luck and Mason, 2012; Skjorringe et al., 2015). Subsequently, Fe2<sup>+</sup> from the cytosol is exported through FPN present in the abluminal membrane of endothelial cells, which are in contact with astrocytes expressing ceruloplasmin, that rapidly oxidize Fe2<sup>+</sup> to Fe3<sup>+</sup> (this Tf/TfR endocytosis mechanism is common for endothelial and neuronal cells and is depicted in **Figures 1**, **2**). A recent study performed using comprehensive mathematical models concluded that optimal iron homeostasis is dependent on the degree of saturation and the transport of HTf across the BBB from the blood (Khan et al., 2018).

In addition, in mice exposed to experimental stroke under normal iron conditions, iron deposits in the microvasculature were observed early in the border zone of the damaged areas (Desestret et al., 2009), suggesting that, when signaled, the BBB might function as an iron reservoir from which iron is released into the brain. Regarding signaling to release, in bovine retinal endothelial cells, the FPN-mediated iron export is known to be blocked by the FPN inhibitor hepcidin (Hadziahmetovic et al., 2011). In addition, new insights into the communication between systemic iron status and brain iron uptake reveal the regulation of brain iron uptake at the level of brain microvascular endothelial cells rather than the BBB acting merely as a retaining wall (Chiou et al., 2018b). In support of this regulatory hypothesis, endothelial microvascular cell distribution of TfR is altered in luminal, intracellular, and abluminal membranes depending on brain iron status (Simpson et al., 2015), and microvascular cell iron availability and efflux have been found to be tightly modulated by diffusible ceruloplasmin and hepcidin released from astrocytes surrounding the BBB (McCarthy and Kosman, 2014). Moreover, a recent study showed that inhibition of DMT1 alters the transport of iron and Tf across the endothelial cells and proposed that the levels of irondevoid Tf (apotransferrin) and iron delivered to the brain are strongly regulated by the ratio apotransferrin:HTf as well as by

FIGURE 1 | Schematic drawing of some of the main players in iron transport to the brain in control (left hand upper side), iron overload (right-hand upper side), ischemic stroke (left-hand lower side), or intracerebral hemorrhage (right-hand lower side). The thicker the arrows the more increased the transport. Abbreviations: ATf, apotransferrin; FPN, ferroportin; HTf, holotransferrin; FT, ferritin, GPX4, glutathione peroxidase 4; GSH, glutathione; Hb, hemoglobin; NTBI, non-transferrin-bound iron; ROS, reactive oxygen species; TJ, tight junction; ZA, zonula occludens.

hepcidin levels in the extracellular fluid of each brain region (Duck et al., 2017).

Further, very recently, a new function for FT heavy chain (FTH) as a transporter of NTBI across the BBB by binding to the T-cell immunoglobulin and mucin domain-containing 1 receptor 1 (TIM-1) has been reported in human endothelial cells (Chiou et al., 2018b).

In the context of physiological iron-overload conditions found, for instance, in individuals on iron-rich diets, a significant pool of NTBI is found in the blood and might be readily absorbed by microvascular endothelial cells. This effect might be relevant considering that regional differences in the brain uptake of <sup>59</sup>Fe–NTBI versus <sup>59</sup>Fe–TfR have been reported (Deane et al., 2004), and that <sup>59</sup>Fe–NTBI seems to be transported faster than <sup>59</sup>Fe–TfR into the brain parenchyma of iron-overloaded diseasefree mice. Furthermore, the iron storage protein FT has been found to be upregulated in epithelial cells of the choroid plexus and endothelial cells of the brain during systemic iron overload (Tripathi et al., 2017).

## IRON UPTAKE AND HANDLING BY PARENCHYMAL BRAIN CELLS

Once in the brain interstitial fluid, both TBI and NTBI are potential iron sources for brain cells (Gaasch et al., 2007); unlike

glutathione; GSSG, glutathione disulfide; Hb, hemoglobin; HFE, hereditary hemochromatosis protein; HIF, hypoxia-inducible factor; HO1, heme oxygenase 1; Hp, haptoglobin; Hpx, hemopexin; HTf, holotransferrin; IRE, iron-responsive element; IRPs, iron regulatory proteins; LIP, labile iron pool; MCOLN1, mucolipin 1; MtFT, mitochondrial ferritin; NCOA, nuclear receptor coactivator; NMDAR, NMDA receptor; NTBI, non-transferrin-bound iron; OH, hydroxyl radical; PCBP, poly(rC)-binding protein; PUFA, poly-unsaturated fatty acid; PUFA-OOH, PUFA peroxide; ROS, reactive oxygen species; Se, selenium; Steap, six-transmembrane epithelial antigen of prostate; system X<sup>−</sup> c , cystine–glutamate antiporter; TfR1, transferrin receptor 1; TfR2, transferrin receptor 2; ZIP8/14, zinc transporters 8/14. The picture depicts the effect of one of the major players in the pathophysiology of brain I, the disruption of glutamatergic neurotransmission homeostasis that produces elevated extracellular glutamate levels, overactivation of the NMDA subtype of glutamate receptors (NMDAR), and excitotoxic cell death. In the recent past, different authors have shown that this NMDAR overactivation boosts neuronal iron uptake and produces an iron-dependent form of neuronal death associated with massive lipid peroxidation that fits with the definition of ferroptosis. Further, in the hemorrhage type of stroke, iron derived from hemoglobin and/or heme enters neurons and produces massive lipid peroxidation. Damage by ischemic and hemorrhagic stroke is alleviated by the inhibitors of ferroptosis.

blood Tf, CSF Tf is considered to be completely saturated with iron (Bradbury, 1997). Within the extracellular milieu of the brain parenchyma, iron is carried loaded into Tf or bound to low-molecular-weight molecules such as citrate, ascorbate, or ATP. Each brain cell type uptakes extracellular TBI or NTBI depending on the iron-transport receptors and channels present in their membranes and, physiologically, safely handles iron by using the network of proteins they specifically express to import, store, traffic, and export iron. Players of cellular iron homeostasis are cell type-specific, and the list of proteins involved in iron homeostasis is still growing. As an example of iron-related new proteins, only 3 years ago, the members of the ZIP family of metal ion transporters Zip8 and Zip14, which, unlike the DMT1 transporter are only active at normal physiological pH, were reported to be expressed by hippocampal neurons, and Zip8 was found to play a major role in the accumulation of NTBI by hippocampal neurons (Ji and Kosman, 2015), whereas both Zip8 and Zip14 were considered to play a role in iron homeostasis in rat retina (Sterling et al., 2017).

# Receptors Involved in Iron Uptake by Brain Cells

In addition to the well-known TfR1 as a main player of the TfR/Tf endocytic pathway in most brain cells, other proteins, e.g., transferrin receptor 2 (TfR2) (Kawabata, 2018) or glyceraldehyde-

3-phosphate dehydrogenase (GAPDH) (Raje et al., 2007), have been described as receptors for Tf and are currently being investigated. TfR1 binds HTf with a Kd of 1 nM; it is expressed in response to changes in iron levels through modulation by iron regulatory elements (IREs) present in its mRNA and binds hereditary hemochromatosis HFE protein, a molecule that competes with HTf for binding to TfR1. In contrast, TfR2 has lower affinity for HTf; its levels are not modulated by IREs (West et al., 2000) and, although still controversial, seems to bind HFE protein in a sequence non-overlapping to that of the Tf-binding site (for a review see Worthen and Enns, 2014).

The TfR2 is present in two main isoforms, TfR2α and TfR2β. The TfR2α form has been reported to be more abundant in the brain, especially in the hippocampus; brain TfR2α levels were reported to significantly differ across animals fed iron-deficient or iron-enriched diets (Pellegrino et al., 2016). A specific role for TfR2 in the control of the iron regulatory network in the brain tissue has been recently suggested, as altered brain IRPs, iron accumulation, and reactive microglia signaling were found in TfR2-KO mice (Pellegrino et al., 2016). In contrast, a previous study on TfR2 mutant mice showed changes in hepatic, but not brain iron levels, although that study reported a 30% increase in brain FT levels (Acikyol et al., 2013).

In addition to TfR1 and TfR2, the multifunctional molecule GAPDH has been reported to work as a TfR in different cell types (Raje et al., 2007; Kumar et al., 2012) with a role in iron uptake and/or iron export (Sheokand et al., 2016) and as a chaperone for intracellular heme traffic (Sweeny et al., 2018). Relevant to brain and excitotoxicity, GAPDH interacts with the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid subtype of glutamate receptors in the neurons without glutamate stimulation and has also been reported to be pivotal for axon development (Lee et al., 2016).

Another receptor-mediated means to import iron into brain cells, at least in oligodendrocytes, is the uptake of FTH through TIM-1/2 receptors (Todorich et al., 2008, 2011; Chiou et al., 2018a). In other non-neuronal cells, FTH internalize through TfR1 (Li et al., 2010), whereas scavenger receptor class A member 5 (SCARA5) has been found to play also a role as a receptor for FT light chain uptake (Li et al., 2009). To the best of our knowledge, the role of SCARA5 has not yet been investigated in any brain cell type.

# Mechanisms and Proteins Involved in Cytosolic Iron Handling in Different Brain Cell Types

Inside the cell cytosolic and nuclear compartments, iron chaperone poly(rC)-binding proteins serve the binding and delivery of iron to iron-requiring enzymes, ubiquitous protein of iron storage FT, or the iron export membrane protein FPN (Philpott et al., 2017). In the cytosol, unused iron is stored into FT, a molecule composed of 24 subunits of a variable proportion of heavy and light chains, that shields up to 4500 iron atoms in a redox inert state. Neurons express both TfR1 and DMT1 (Pelizzoni et al., 2012; DeGregorio-Rocasolano et al., 2018); DMT1 is mainly found in the cytoplasm colocalizing with endosomes/lysosomes. Further, neurons express the iron export molecule FPN (Burdo et al., 2001; Zhou et al., 2017). Astrocytes express TfR1, DMT1, and FPN (Yang et al., 2012; Huang et al., 2014). Hence, although TfR1 is considered to be the main gate for cellular iron import, control astrocytes show low TfR1 levels, and most mature oligodendrocytes or microglial cells do not express TfR1 (Kaur and Ling, 1995; Rathnasamy et al., 2016). However, under hypoxia, both astrocytes and oligodendrocytes are found to upregulate TfR (Yang et al., 2012; Rathnasamy et al., 2016). Moreover, oligodendrocytes and cells of choroid plexuses, but not other brain cell types, express Tf under normal condition (Leitner and Connor, 2012). All the above are examples of the important heterogeneity in the expression and induction of iron homeostatic or IRPs observed in different brain cell types and situations. Most of the proteins involved in iron metabolism, including cytosolic FT levels, are mainly regulated by iron, oxidative stress, and inflammation. The irondependent regulation occurs mainly at the post-transcriptional level via the IREs and iron regulatory protein (IRP) machinery. When iron levels are high, iron binds to the IRP, which then loses affinity for the IRE sequence in FT mRNA, resulting in increased FT expression (the cytosolic iron handling is depicted in **Figure 2**) (Arosio et al., 2017). Several studies have shown that the mobilization of iron from FT requires FT degradation either by lysosome–autophagy or ubiquitin proteasome pathways (Asano et al., 2011; Mancias et al., 2015). In addition to its role in storing iron within the cell, FT can be secreted (Truman-Rosentsvit et al., 2018).

In addition to post-transcriptional regulation, transcriptional regulation has been reported for some of the proteins involved in iron homeostasis through the hypoxia-inducible factor family of transcription factors or the heme-dependent transcription factors (e.g., for TfR1 or FPN) (Marro et al., 2010; Man-man et al., 2017). Moreover, the master regulator of bodily iron homeostasis hepcidin is subjected to a sophisticated transcriptional regulation through the bone morphogenetic protein/SMAD pathway, interleukin 6 pathway, or erythropoietin/erythroferrone pathways when regulated according to the iron, inflammatory, or erythropoietic demands (Sangkhae and Nemeth, 2017). The regulation of cellular iron balance has been recently reviewed (Recalcati et al., 2017).

Iron also plays an important role in specific organelles; this is especially true for mitochondria. Mitochondria house the assembly of heme, the majority of the Fe–S cluster biosynthesis apparatus, and contain the respiratory complexes that need ironcontaining cofactors. In addition, mitochondria are a major source of endogenous ROS. Sustained iron exposure increased mitochondrial ROS levels in dopaminergic neuroblastoma SH-SY5Y cells (Huang et al., 2017), and scavenging of mitochondrial ROS was protective against iron overload damage in hippocampal neurons, preserving mitochondrial morphological integrity and membrane potential (Pelizzoni et al., 2011). In addition, a recent study showed that sequestration of iron by mitochondrial FT was neuroprotective against oxidative stress and played an important role in preventing neuronal damage in some other conditions (You et al., 2016; Guan et al., 2017). Moreover, mitochondrial FT overexpression significantly reduced cellular labile iron pool,

prevented H2O2-induced elevation of labile iron, and rescued cells from H2O2-induced damage (Gao et al., 2017).

In several cell types, although not yet studied in neurons, mitochondria−endosome interactions have been observed, suggesting a possible direct transfer of iron from endosomes to mitochondria (Das et al., 2016; Hamdi et al., 2016) through the "kiss-and-run" hypothesis.

#### IRON DYSHOMEOSTASIS IN THE BRAIN

Brain iron abnormalities are associated to rare, but severe, neurodegenerative conditions, and there is growing evidence that the more common systemic iron overload disorders such as hereditary hemochromatosis exert important effects on the brain iron content and pathological brain iron deposition (Nandar and Connor, 2011; Kumar et al., 2016). Moreover, the hemochromatosis H63D polymorphism of the HFE gene could be a disease-modifying gene in frontotemporal lobar degeneration, fostering iron deposition in the basal ganglia (Gazzina et al., 2016), or even macro- and microanatomically altering some brain structures associated with changes in Tf levels in the blood (Jahanshad et al., 2012). Recently, abnormal recycling of TfR1 due to reduced post-translational palmitoylation has been reported to be crucial to affect the iron import in the socalled neurodegeneration with brain iron accumulation disease (Drecourt et al., 2018). Further, neurodegeneration with brain iron accumulation disease-linked genes that showed altered expression in response to iron loading has been recently reported to be directly or indirectly related to myelin metabolism (Heidari et al., 2016).

A list of mild conditions, including aging (Ward et al., 2014), continuous uptake of some bioavailable iron sources (Peters et al., 2018), or obesity (Han et al., 2017), have been reported to alter brain iron content and/or distribution. In addition, a recent report showed that iron administration increases iron levels in the brains of healthy rats, which induces brain changes and triggers a hormetic response that reduces oxidative damage (Piloni et al., 2018). Moreover, other studies reveal a complex interplay between inflammation and brain iron homeostasis (Righy et al., 2018; Sankar et al., 2018), with acute inflammation increasing non-Tf iron uptake by brain microglia (McCarthy et al., 2018). This dyshomeostasis is especially important in acute pathologies such as stroke, in which the impairment of the BBB regulatory role is rapidly and massively affected following either rupture of an artery (ICH) or aberrant increase of brain microvascular endothelial permeability following ischemic stroke (AIS). Dysregulation of iron resulting in the accumulation of free iron and brain iron overload, which can increase the production of ROS, is evident in aging and stroke-related pathologies and is also a hallmark of chronic and long-term neurodegenerative pathologies. In this regard, iron overload in the brain has been reported in Huntington, Parkinson, or Alzheimer diseases [readers interested in the topic are referred to recent detailed reviews by Morris et al. (2018) and Uranga and Salvador (2018)]. Interestingly, marked age-related changes in brain iron homeostasis have also been observed in amyloid precursor protein (APP)-knockout mice (Belaidi et al., 2018).

#### IRON OVERLOAD WORSENS NEUROLOGICAL DAMAGE BOTH IN ISCHEMIC STROKE AND INTRACEREBRAL HEMORRHAGE

Stroke is a life-threatening disease that causes high rates of permanent disability subsequent to neuronal loss. There are two major types of strokes: ischemic stroke or AIS, which is caused by a blood clot blocking an artery and accounts for 85% of all strokes, and hemorrhagic stroke or ICH, caused by leakage or rupture of an artery and accounts for 15% of stroke cases.

#### Iron Overload Condition in Stroke Damage and Outcome

In ischemic stroke, neurons die because of interrelated processes that include glutamate excitotoxicity, excess of free radical production, and inflammation (Nagy and Nardai, 2017). Each of these mechanisms has already been targeted in clinical trials, although unsuccessfully (for a review see Chamorro et al., 2016). Moreover, free radical-induced oxidative stress has been for long recognized as a pathogenic factor in the ischemia/reperfusion (I/R) injury that follows reoxygenation. The hydroxyl radical, the most harmful of the free radicals, is generated through reactions catalyzed by iron (Kell, 2009; Chen H. et al., 2011). Of relevance to stroke damage and final outcome, several clinical studies indicate that a previous systemic iron overload condition, measured as high levels of serum FT at admission, is associated with increased brain damage and worse outcome induced by ischemic stroke (Davalos et al., 1994; Millan et al., 2007). A similar detrimental effect (increased ischemic damage) was observed in iron-overloaded animals exposed to experimental ischemic stroke (Castellanos et al., 2002; Gamez et al., 2003; Mehta et al., 2004; García-Yébenes et al., 2012; DeGregorio-Rocasolano et al., 2018). Iron overload is a major source of oxidative stress in ischemic brains (Carbonell and Rama, 2007). In addition, iron overload exacerbated the risk of hemorrhagic transformation in animal models of transient ischemia plus early reperfusion (García-Yébenes et al., 2012; García-Yébenes et al., 2018); therefore, in iron-overloaded animals, an initial ischemic event has higher probability to become an ICH. The contribution of iron overload to widespread the damage in the ICH is supported by the fact that systemic iron overload (estimated using high serum FT levels) was associated with higher perihematoma edema and poor outcomes in patients with ICH (Mehdiratta et al., 2008; Perez de la Ossa et al., 2010; Bakhshayesh et al., 2014; Garton et al., 2017). Further, a combination of high serum FT and low serum iron and Tf was associated with poor outcome in ICH patients (Yang et al., 2016). Therefore, systemic iron overload impacts the outcome of both ischemic and hemorrhagic stroke patients.

The brain is physiologically sheltered from fluctuations in systemic iron because of the BBB, even under experimentally

induced diet iron overload conditions or intraperitoneal iron administration (Castellanos et al., 2002; Millerot et al., 2005). However, in the ischemic brain, the detrimental effect of iron overload might result from systemic iron pools reaching the brain parenchyma. Following artery occlusion, a physiological increase in transport through a still somehow intact BBB is noted, and thereafter pathological halt of the BBB function that results in leaky or broken capillaries that allow blood spill out (**Figure 1**). Some BBB impairment with aberrant exchange of brain and blood molecules occurs relatively early following ischemic stroke (within the first hours of post-symptom onset) showing a biphasic behavior in some studies (Klohs et al., 2009). Moreover, leakage through the BBB varies in different experimental stroke models (Klohs et al., 2009; Abulrob et al., 2011). Of note, post-stroke BBB leakage also varies in different areas of the brain, whereas the administration of the broadly used thrombolytic agent tissue plasminogen activator increases leakage (Jaffer et al., 2013), resulting in hemorrhagic transformation. Moreover, a recent evidence of the pivotal contribution of BBB leakage to ischemic stroke damage has been revealed by a study in which a vascular leakage blocker reduced stroke damage (Zhang et al., 2017).

#### Effect of Stroke on Brain Iron

Regarding iron content of the brain following ischemic stroke, an increase in total iron was observed in the ischemic areas after permanent or transient ischemic stroke (Helal, 2008; Millerot-Serrurot et al., 2008; Tuo et al., 2017). In addition, increased TfR and iron levels were observed in susceptible hippocampal areas following transient global cerebral ischemia (Park et al., 2011), or increased extravasation of iron-loaded HTf was observed following acute focal ischemic stroke (DeGregorio-Rocasolano et al., 2018). In fact, HTf from blood accumulates in the brain parenchyma of the rat ischemic brain hemisphere at 1 h after reperfusion (DeGregorio-Rocasolano et al., 2018). Subsequently, iron accumulates into the endothelial cells of brain capillaries near ischemic areas (Desestret et al., 2009), whereas increased levels of Tf were observed 24 h following a transient experimental stroke at the ischemic hemisphere (Lo et al., 2007). In fact, this Tf seems to accumulate in neurons since Tf immunoreactivity was found located at the neuronal cytoplasm 24 h after stroke onset (DeGregorio-Rocasolano et al., 2018).

In the ICH stroke type, the systemic iron pools play a direct role in brain damage since the rupture of the microvascular wall and concomitant blood release results in neurons, astrocytes, and other cells around the hemorrhagic area to become immediately exposed to blood-derived iron, either free or bound to Tf (**Figures 1**, **2**). Moreover, the brain is exposed to a bulk amount of erythrocytes that undergo lysis and subsequent spill of hemoglobin (Hb). Erythrocyte lysis occurs within minutes and continues for days after the formation of brain hematoma, leading to the release of red blood cell-derived products. Cells of the brain are then sequentially exposed to aberrantly high extracellular levels of Hb, the oxidized Hb form methemoglobin, and free heme or the iron ferric heme form hemin. The main mechanisms of neurodegeneration in hemorrhagic stroke are iron- and heme-induced ROS production, amplification of an inflammatory response, direct toxic effects of iron and heme, and glutamate-induced excitotoxicity (Righy et al., 2016).

# BRAIN REGULATION OF IRON METABOLISM DURING STROKE

# Endogenous Mechanisms Involved in the Protection of Brain Cells From Excess Iron Following AIS

Increased brain Tf or TfR1 levels (Ding et al., 2011; DeGregorio-Rocasolano et al., 2018) were observed in lethal ischemia. Consensus exists about the protective role of increased FT expression during experimental stroke, and the same applies to hepcidin in AIS; however, the role of blunting the ischemia-induced upregulation of Tf and TfR remains controversial. Reductions of Tf and TfR were considered to be associated with neuroprotection in one specific stroke model (Lo et al., 2007). However, other neuroprotective conditions were not associated with Tf and/or TfR reductions (He et al., 2012).

Hepcidin seems to protect from neurodegeneration in AIS. It is an iron-inducible peptide hormone that is considered to be the main regulator of iron homeostasis. Hepcidin is produced as a propeptide mainly in the liver and then secreted to the blood where it interacts with the only known iron exporter FPN located at the cell membrane. Hepcidin acts as a negative regulator of cellular iron release by binding FPN and causing its internalization and degradation; this inhibits FPN-mediated iron egression from enterocytes, macrophages, hepatocytes and, as recently found, some other cell types (e.g., endothelial cells or neurons). The inhibition of iron egression, mainly from macrophages and hepatocytes, prevents the presence of NTBI in the blood as well as reduces the TSAT in the blood. In line with its main function, hepcidin deficiency leads to iron overload (Nicolas et al., 2001), and hepcidin overexpression causes severe iron deficiency (Nicolas et al., 2002a). Hepcidin has a rapid and direct impact on circulating iron pools. In fact, even a single intraperitoneal administration of hepcidin reduces the TSAT in blood to less than half the normal levels for 6−12 h (Sillerova et al., 2012). Both blood hepcidin and HTf/apotransferrin ratio might be considered as the master regulatory factors of systemic iron metabolism, since hepcidin senses iron by sensing HTf levels in the circulation and iron-FT levels inside hepatocytes. Relevant to this review on iron and stroke, hepcidin expression is also responsive to hypoxia (Nicolas et al., 2002b; Sonnweber et al., 2014) mainly by the oxygen-regulated hypoxia-inducible factor. In general, hypoxia inhibits hepcidin expression to allow the mobilization of iron to sustain erythropoietic expansion. However, a recent study showed that the exposure of mice to an hypoxic environment induces biphasic modulation of hepcidin expression with a transient increase in liver expression within the first 6 h and a reduction after 15 h of hypoxia (Ravasi et al., 2018).

Among ischemic stroke patients, who have areas of the brain exposed to oxygen and glucose deprivation (OGD), blood hepcidin levels increased within the first 6 h from symptom onset compared to those in controls; serum hepcidin levels in

ICH patients were similar to those of controls (unpublished results from our laboratory). This early increase in blood hepcidin during ischemic stroke is in agreement with the early increase reported in the hypoxic mice model (Ravasi et al., 2018) and with the two other reports on stroke patients (Petrova et al., 2015; Slomka et al., 2015) and might be a reflection of an increased expression and secretion of hepcidin from the ischemic brain cells to the extracellular medium, and a subsequent spillover out of the brain during stroke. Although hepcidin expression in the brain is thought to be low in basal conditions, hepcidin is now emerging as an important player in brain iron homeostasis (Vela, 2018). Hepcidin mRNA levels increased early in the brain in response to ischemia in a model of endothelin-induced vasoconstriction (Bickford et al., 2014), and hepcidin levels were significantly elevated in the ischemic side of the brain 24 h after occlusion in a stroke model due to middle cerebral artery occlusion; this ischemic side also showed increased TfR1 and FT levels and reduced FPN levels (Ding et al., 2011). In addition, knock-down of hepcidin slows the ischemicmediated changes in brain FT and FPN, revealing an important contribution of this molecule to parenchyma iron regulation in cerebral ischemia. Further, normoxic iron-overloaded rats, hepcidin, either overexpressed in the brain or administrated to the brain parenchyma, reduced brain uptake of iron in vivo as well as reduced iron uptake in cultures of neurons or brain microvascular endothelial cells, owing to the dowregulation of iron-transport proteins (Du et al., 2015). In microglial cells, the bone morphogenetic protein antagonist noggin abrogated the OGD-induced changes in hepcidin, FPN, and FT levels, and this inhibition of bone morphogenetic protein/hepcidin axis in vitro shifted reactive microglia from an iron-storing to an iron-releasing phenotype after OGD plus reperfusion (Shin et al., 2018). Factors released from noggin- and OGD-exposed microglial culture medium were found to increase myelin production in oligodendrocyte cell cultures; since this effect is blunted by the iron chelator deferoxamine (DFO), iron released from the microglia is proposed to serve in the remyelination in the ischemic brain (Shin et al., 2018).

#### Endogenous Mechanisms Involved in the Protection of Brain Cells From Excess Iron Following ICH

Similar to that observed in AIS, subarachnoid hemorrhage has been shown to increase hepcidin expression in neurons (Zhao et al., 2018) associated with a reduction of its downstream target FPN (Tan et al., 2016). However, in this hemorrhage model, hepcidin seems to be harmful since its administration potentiated and its siRNA decreased apoptosis and early brain damage (Tan et al., 2016). In agreement with a harmful effect of hepcidin on brain hemorrhage, higher blood hepcidin levels were observed 1−7 days following ICH in patients with poorer outcomes (Xiong et al., 2015). The rationale for this harmful effect of hepcidin following ICH might be an impaired clearance of iron from brain parenchyma, since increased hepcidin expression caused by inflammation was found to inhibit intracellular iron efflux from brain microvascular endothelial cells and further draining of iron into circulation, leading to the aggravation of oxidative brain injury (Xiong et al., 2016). Conversely, addition of hepcidin or adenovirusdriven overexpression of hepcidin has been shown to protect neurons from hemin-induced injury (Zhou et al., 2017); this protection was thought to be related to the effect of hepcidin reducing hemin-induced iron uptake and reduced expression of TfR1 and DMT1.

Moreover, importantly, stroke has an inflammatory component, and some inflammatory inducers or mediators (e.g., lipopolysaccharide or interleukin 6) have been found to increase the expression of hepcidin in astrocytes, microglia, or neurons (for a recent review see Vela, 2018), although the effect on neurons is usually weaker (Ma et al., 2018).

The brain has protective mechanisms that can be activated after ICH and erythrocyte lysis, the most important one being the Hb-scavenging molecule haptoglobin (Hp) and the hemescavenging system hemopexin (Hpx) (**Figure 2**). Hb itself is avidly bound by Hp to form a complex that prevents the release of the heme group from Hb or methemoglobin. However, Hp cannot prevent the release of the heme group from the highest oxidative Hb form ferrylHb. In the brain, Hp is almost exclusively synthesized by oligodendrocytes, and mice overexpressing Hp were more resistant, whereas Hp-knockout mice were more susceptible, to ICH injury (Zhao et al., 2011). The complex Hp:Hb binds to the Hp-specific receptor CD163 that is present in macrophages, allowing them to remove Hb from the extracellular space. In addition, CD163 is upregulated in neurons after ICH (Liu et al., 2017). The role of Hp in neuroprotection was not devoid of controversy since Hp was found to exacerbate the vulnerability of CD163-expressing neurons to Hb, an effect that was reduced by iron chelators (Chen-Roetling and Regan, 2016). However, most Hb was not bound to Hp, indicating that the CD163–Hb–Hp system is saturated, and that the primary route for Hb clearance from the CNS occurs through the BBB (Galea et al., 2012; Righy et al., 2018). A good clearance is associated with neuroprotection since lower CSF Hp levels were found in ICH patients without clinical and/or radiological evidence of delayed cerebral ischemia (Galea et al., 2012). Hb accumulated in the brain parenchyma is converted to ferrylHb, which causes the upregulation of proinflammatory adhesion molecules and increases the recruitment of inflammatory neutrophils or macrophages (Jeney et al., 2013).

Iron–hemin uptake has been reported in neurons and astrocytes and is especially high in microglial cell cultures; neurons benefit from a high export of iron following the initial iron–hemin loading, and this export was exacerbated by treatment with DFO, a compound that attenuates hemin neurotoxicity (Chen-Roetling et al., 2014). As already mentioned, the brain produces the heme scavenger protein Hpx that protects from deleterious effects of heme. In this regard, deletion/knockout of the Hpx aggravates brain injury by ICH (Chen L. et al., 2011; Ma et al., 2016) and, although endogenous levels of Hpx are insufficient to counteract the massive heme overload following ICH, increasing brain Hpx levels has been found to improve the outcome after ICH (Leclerc et al., 2018). Conversely, Hpx has recently been reported to increase the

neurotoxicity of Hb in the absence of Hp (Chen-Roetling et al., 2018). Therefore, although both Hp and Hpx might provide a robust line of defense during ICH, additional knowledge on the fine tuning of this system is still required to design new possible neuroprotective treatments for ICH. The low-density lipoprotein receptor-related protein CD91 is expressed in several cell types, including macrophages and neurons, and was identified as the Hpx–heme receptor, mediating Hpx–heme uptake by endocytosis (Hvidberg et al., 2005). Once inside the cell, the heme oxygenase (HO) system is responsible for cellular heme degradation. In neurons, HO generates biliverdin, which has been reported to scavenge free radicals. Astrocytes rapidly upregulate HO1 and FT and are resistant to heme-mediated injury, and astrocyte HO1 is known to play a robust neuroprotective role after hemorrhage (Chen-Roetling et al., 2017). A recent study indicated that Hpx activates endothelial progenitor cells and reduces synaptic damage in a rodent ischemic stroke model; these effects require HO1 activity (Yang et al., 2018).

Moreover, the presence of heme groups in the extracellular milieu activates the endothelial expression of adhesion molecules and promotes the activation of neutrophil extracellular traps through a mechanism dependent on ROS generation (Chen et al., 2014). Further, inflammatory response is a well-documented mechanism of brain damage after ICH (Wang and Doré, 2007), and the mechanisms underlying heme-induced inflammation have been reviewed elsewhere (Dutra and Bozza, 2014). Iron and heme associated with pro-inflammatory mediators in the CNS following hemorrhagic stroke and inflammatory profiles are associated with poorer prognosis. Similarly, iron-induced damage following ischemic stroke in iron-overloaded animals has also been related to increased inflammatory responses associated with increased serum levels of interleukin 6 and TNF-alpha (Castellanos et al., 2002).

## THE FERROPTOTIC COMPONENT OF STROKE-INDUCED NEURODEGENERATION

#### The Concept of Ferroptosis: Molecular Players Involved in Ferroptosis in the Neurons

The concept of ferroptosis was introduced as a form of cell death, which is morphologically, biochemically, and genetically distinct from apoptosis, necrosis, and autophagy; it is associated with excessive iron-mediated accumulation of lipid ROS in cancer cells as well as in hippocampal slice cultures exposed to glutamate excitotoxicity (Dixon et al., 2012). As excitotoxicity is the main mechanism involved in neuronal death in ischemic stroke, the involvement of ferroptosis in ischemic stroke could be suspected. In addition, the role of high systemic iron exacerbating stroke-induced damage, concomitant with increased brain iron after ischemic stroke onset (reviewed in the previous section), further suggested a possible ferroptosis component in stroke; in the recent past, mounting evidence indicated that ferroptosis plays such a pivotal role in stroke-induced neurodegeneration in both ischemic and hemorrhagic stroke subtypes. Although the concept is still evolving, ferroptosis is now considered a unique form of regulated cell death characterized by cytosolic accumulation of iron and executed by the massive peroxidation of PUFA in the plasma membrane and associated with reduced glutathione (GSH). Compounds that induce only either cytosolic or mitochondrial ROS production are not considered as ferroptosis inducers. A recent study proposed that both Fenton reactions, in which Fe2<sup>+</sup> converts to Fe3<sup>+</sup> producing ROS, and enzymatic lipoxygenase-mediated reactions in cellular membranes contribute to ferroptosis (Feng and Stockwell, 2018). Some authors considered that the ferroptosis concept and the previously described oxidative oxytosis concept have many common features to be regarded as independent processes (Lewerenz et al., 2018). In fact, the widely used oxidative stress stimulus tert-butylhydroperoxide was found to induce neuronal cell death that was blocked by ferroptosis inhibitors, thereby implying a crosstalk between the initial oxidative damage and the occurrence of ferroptosis (Wu et al., 2018). However, ferroptosis is considered a major driver of cell death in many neurodegenerative and neurological diseases (Lewerenz et al., 2018; Morris et al., 2018).

Ferroptosis can be triggered by the inhibitors of the lipid repair enzyme glutathione peroxidase 4 (GPX4), such as the RAS-selective lethal 3 compound, or depletion of the antioxidant GSH by starving the cells of GSH precursors [inhibiting the cystine/glutamate antiporter, i.e., system X<sup>−</sup> c (X<sup>−</sup> c ) with erastin, sulfasalazine, sorafenib, or, alternatively, challenging the exchange capacity of X<sup>−</sup> <sup>c</sup> with high extracellular glutamate]. Extracellular Tf in its iron-loaded HTf form is considered as an inducer of ferroptosis as well, and inhibiting TfR expression prevents ferroptosis (Gao et al., 2015).

At present, the existence of X<sup>−</sup> c in neurons is still controversial, with one report showing expression lack of X<sup>−</sup> c in neurons (Ottestad-Hansen et al., 2018), whereas another showing neurons expressing significant X<sup>−</sup> c levels (Soria et al., 2016). However, neurons in culture overexpress X<sup>−</sup> c shortly after being exposed to OGD (Soria et al., 2014). The physiological role of X<sup>−</sup> c as a cystine importer is linked to its antiporter glutamate export and requires that extracellular levels of glutamate and other excitatory amino acids are maintained low by the action of excitatory amino acid transporters (EAATs). Thus, one of these glutamate transporters, the sodium-dependent EAAT3, might be selectively expressed in neurons and could also be an unidirectional cysteine import channel, with extracellular glutamate inhibiting its cysteine import capacity (Watts et al., 2014). EAAT3 might be as important as X<sup>−</sup> c in providing GSH precursors to neurons, considering that mice lacking EAAT3 have lower GSH levels and age prematurely (Aoyama et al., 2006). Notably, low GSH levels might also inhibit GPX4 activity since GPX4 uses GSH as a cofactor to catalyze the reduction of lipid peroxides and protect cells membranes against peroxidation (see Latunde-Dada, 2017).

Conversely, ferroptosis can be prevented by (1) ironbinding compounds such as DFO and ciclopirox; (2) small molecules, free radical scavengers, or lipophilic antioxidants such as ferrostatin-1, α-tocopherol, butylated hydroxytoluene, or liproxstatin-1; (3) treatments with molecules such as deuterated

PUFA, inhibitors of acyl-CoA synthetase long-chain family member 4, lipoxygenase inhibitors, glutaminolysis inhibitors, or cycloheximide; or (4) boosting cystine and/or cysteine import by reducing extracellular glutamate.

Glutamate is the most abundant excitatory neurotransmitters in the vertebrate nervous system, with 90% of the synaptic neurotransmission considered glutamatergic (Hofmeijer and Van Putten, 2012; Mayor and Tymianski, 2018). Under physiological conditions, extracellular glutamate levels are low, and glutamate exerts its physiological role mainly at the neuronal synapse. The ability of astrocytes near glutamatergic neurons to rapidly clear synaptic glutamate after its physiological action is critical in the functioning of synapses and neuronal circuits, and the initial increase of extracellular glutamate is inhibited mainly by glial EAAT1 and 2. Neuronal glutamate transporters, mainly EAAT3, also play a role in clearing extracellular glutamate excess. However, ischemia leads to energy failure and subsequent electrochemical gradient impairment, causing EAATs to act in a reversed manner, thereby extruding nonvesicular intracellular glutamate and resulting in large excess of extracellular glutamate and impairment of EAAT3 as a unidirectional cysteine import channel.

The impact of excess extracellular glutamate on neuronal survival is more than that on the cystine/glutamate exchanger X<sup>−</sup> c . In fact, most of the effect is produced by glutamate binding to specific glutamate receptors present in cell membranes of several cell types and, especially in neurons, N-methyl-D-aspartate (NMDA) subtype of glutamate receptors (NMDARs) (Guirao et al., 2017) that are coupled to an intracellular downstream complex of signaling effectors, which are known to initiate excitotoxic neuronal death following ischemia.

Although both glutamate and iron are essential for neuronal survival, excess glutamate can be toxic, and excess free iron through the Fenton reaction is a potentially toxic substance that can catalyze the production of extremely damaging ROS. During stroke, the impairment of BBB function allows increased access to brain of iron-carrying substances and, preceding neurodegeneration, iron accumulates in the ischemic areas (Helal, 2008); therefore, not surprisingly, iron overload conditions are associated with larger areas of brain damage in animals exposed to experimental ischemic stroke (Castellanos et al., 2002; Gamez et al., 2003; Mehta et al., 2004; García-Yébenes et al., 2012; DeGregorio-Rocasolano et al., 2018). Following experimental stroke or NMDAR overactivation, prevention of NMDA-induced neuronal iron uptake was found to be neuroprotective (DeGregorio-Rocasolano et al., 2018). Therefore, the crosstalk between neuronal homeostasis of iron and glutamatergic signaling is increasingly being supported by evidence and should not be disregarded as a therapeutic intervention target.

#### Glutamate-Induced Iron Uptake in Neurons and Ferroptosis

Cheah et al. (2006) from Dr. Snyder's laboratory published a study before the concept of ferroptosis was introduced and showed that the overactivation of NMDAR in neurons induces, in addition to the well-known NMDAR-mediated calcium influx, an NMDAR-mediated iron uptake. They also identified a signaling cascade whereby NMDAR regulated neuronal iron homeostasis and excitotoxicity through an NMDAR/nitric oxide/dexamethasone-induced Ras protein 1 (Dexras1) pathway that increased both non-Tf and Tf-mediated iron uptake from the extracellular medium through the neuronal plasma membrane (Cheah et al., 2006). A similar increase of iron uptake in hippocampal neurons was observed in a subsequent study in which either synaptic activity or NMDAR activity was boosted (Pelizzoni et al., 2011). Moreover, either ischemia or NMDA overactivation increased the uptake of HTf, resulting in an increase of cytosolic redox-active iron in neurons, whereas blockade of TfR1 reduced iron dyshomeostasis and NMDAinduced excitotoxicity (DeGregorio-Rocasolano et al., 2018). The original work from Snyder's laboratory hypothesized that iron in its free form gains access to the neuronal cytosol through DMT1 present in the neuronal membrane. However, this would not explain the uptake of Tf-bound iron since the DMT1 isoform endogenously overexpressed in primary cultures of hippocampal neurons exposed to NMDA is DMT1-1B/IRE(+) (Haeger et al., 2010) and is found in endosomes/lysosomes in the cytoplasm (Pelizzoni et al., 2012), thereby not supporting a direct role of DMT1 in iron entry through the neuronal membrane. Following the above rationale of crossroads between excitotoxicity and ferroptosis, several in vitro and in vivo studies showed that neurons survive otherwise lethal NMDA or glutamate concentrations when treated with iron chelators such as DFO or deferasirox (Yu et al., 2009; Tian et al., 2017; Sakamoto et al., 2018). Moreover, a study showed that the deletion of Dexras1 reduced NMDA-induced iron uptake and NMDA toxicity in cortical neurons and retinal ganglion cells (Chen et al., 2013). Moreover, Dexras1-KO mice showed increased expression of the NR2A subunit of the NMDAR (Carlson et al., 2016), a subunit that is preferentially located at synapses and related to NMDAR pro-survival signaling; this strongly suggests the existence of a functional crosstalk between the NMDA-induced iron intracellular signaling and the physiological NMDAR function. Interestingly, intracellular iron released from lysosomes has been shown to modulate NMDAR synaptic excitability via Dexras1 (White et al., 2016), and iron is needed post-synaptically to activate Ca2+-dependent pathways downstream NMDAR (Muñoz et al., 2011). Further, glutamateinduced neuronal excitotoxicity was described as ferroptosis (Dixon et al., 2012).

In the widely reported scenario of elevated extracellular glutamate during ischemic stroke, NMDAR-induced iron import was upregulated in neurons associated with increased ROS production (Cheah et al., 2006), and system X<sup>−</sup> c and EAAT3 cystine/cysteine import are impaired by high extracellular glutamate (Conrad and Sato, 2012), thus compromising neuronal iron homeostasis and GSH-mediated free radical detoxification, eventually leading to an excess lipid oxidative stress that causes ferroptotic damage. High GSH stores are pivotal for cell survival since GSH is required for GPX4 activity and is also thought to be the only molecular species having a significant role as a cytosolic ligand for Fe2<sup>+</sup> (Philpott and

**264**

Ryu, 2014). Even when ischemia is resolved in the brain owing to a successful reperfusion, other mechanisms involved in ferroptotic damage might become dangerous. In this regard, a recent study on Sertoli cells revealed that GPX4 is inactivated via GSH depletion following I/R injury (Li et al., 2018), and several studies showed that GPX4 gene inactivation causes hippocampal neurodegeneration (Seiler et al., 2008; Hambright et al., 2017) and a remarkable degeneration of motor neurons (Chen et al., 2015).

# Ferroptosis in AIS and ICH Stroke Subtypes: Effect of Ferroptosis Inhibitors and Iron Chelators

In addition to a pivotal role of ferroptosis in I/R-induced damage in several peripheral organs (Friedmann Angeli et al., 2014; Linkermann et al., 2014; Gao et al., 2015), a role of ferroptosis has been specifically identified in I/R-induced brain damage in transient ischemia, as brain damage was substantially reduced by the ferroptosis inhibitor liproxstatin-1 in an AIS model (Tuo et al., 2017). These authors also showed that a reduction of tau levels, a protein that facilitates iron export by trafficking APP to membranes to stabilize FPN, was found to precede a pro-ferroptotic iron accumulation in the brain following I/R (Tuo et al., 2017). APP mRNA possesses a functional IRE with sequence homology to that of FT, and its translation is responsive to free iron levels in the cytosol (Duce et al., 2010). In addition, APP has been reported to interact with FPN to promote iron release in several brain cell types (Duce et al., 2010; McCarthy et al., 2014). APP overexpression reduces free iron content in neuroblastoma cells (Wan et al., 2012), and either preventing APP transport to the cell surface or knocking down of APP results in iron accumulation in primary cortical neurons in culture (Lei et al., 2012). The complete role of APP in ferroptosis and neurodegeneration in AIS requires additional studies. However, further evidence supporting the involvement of ferroptosis mechanisms in AIS brain damage is provided by the overexpression of lipoxygenases following brain ischemia and the protection afforded by treatment with lipoxygenase inhibitors (Cui et al., 2010; Karatas et al., 2018). Further, a recent study showed that N-acetylcysteine (NAC), an FDA-approved cysteine prodrug, prevents hemin-induced ferroptosis by neutralizing toxic lipids generated by arachidonate-dependent activity of 5 lipoxygenases (ALOX5) (Karuppagounder et al., 2018). They reported that ALOX5 inhibitors such as zileuton and molecules such as NAC that inhibit toxic metabolic products of ALOX5 or prevent ALOX5 recruitment to its site of action, protect against hemin or ICH-induced ferroptosis (Karuppagounder et al., 2018). Notably, NAC has been previously reported to prevent neuronal damage induced by AIS (Khan et al., 2004; Turkmen et al., 2016).

Further, the role of free iron in ferroptosis and neurodegeneration is further supported by the protective effect of treatments with exogenous iron chelators, e.g., DFO or 2,2<sup>0</sup> -dipyridyl, in animal models of stroke. After permanent ischemic stroke (Millerot-Serrurot et al., 2008), the increase in free iron, the iron form prone to induce lipid peroxidation and susceptible to be chelated, precedes the increase in total iron. In experimental models of transient ischemic stroke, DFO administered intramuscularly (Xing et al., 2009) or intranasally (Hanson et al., 2009) after the induction of ischemia significantly decreased infarct volume. Another iron chelator, the aforementioned 2,2'-dipyridyl was found to be cytoprotective after permanent arterial occlusion in rats; this was associated with reduced infarct size and preservation of GSH levels (Demougeot et al., 2004) or reduction of the apoptotic component and prevention of the enlargement of the lesion (Van Hoecke et al., 2005).

The role of iron in ICH neurodegeneration has been known for long. Recently, neuronal death following ICH was found to have features similar to those of ferroptosis, since chemical inhibitors of ferroptosis protected against Hb- and hemininduced neurotoxicity (Zille et al., 2017). The nonheme iron present in brain parenchyma after ICH is a source of redoxactive iron. The marked increase in brain nonheme iron was not cleared for weeks following the ICH event, and brain Tf, TfR, and FT levels were also increased 72 h after ICH, thereby making FT upregulation long-lasting (Wu et al., 2003). Iron chelators such as DFO reduce hemin- and iron-mediated neurotoxicity, perihematoma edema, and neuronal damage, leading to good neurologic outcomes after ICH (Nakamura et al., 2004; Hatakeyama et al., 2013). Other recent studies indicated that the specific inhibitor of ferroptosis ferrostatin-1 reduces neuron degeneration and improves neurological deficit in experimental ICH models (Li Q. et al., 2017; Zhang et al., 2018), and GPX4 levels were found to be reduced after ICH, whereas pharmacological inhibition of GPX or genetic knockdown exacerbated brain injury after ICH (Zhang et al., 2018). Moreover, inhibitors of ALOX5 or inhibition of the ALOX5 metabolic products protected against hemin- or ICH-induced ferroptosis (Karuppagounder et al., 2018).

Moreover, DFO significantly reduced hemorrhagic transformation after focal I/R (Xing et al., 2009; Li Y. et al., 2017). Some studies have shown that the effect of DFO in preventing ischemic neuronal death is related to the DFOinduced expression of hypoxia-inducible factor 1 and its downstream targets genes rather than with its iron-chelating properties or, recently, with microglial/macrophage HO1 expression in ICH (LeBlanc et al., 2016). A systematic review and meta-analysis in experimental models of ICH showed that DFO was neuroprotective, e.g., reduction of brain edema and improvement of neurobehavioral outcomes, especially when it was administered 2−4 h after ICH induction (Cui et al., 2015). In clinical studies, conclusive evidence of the benefit of DFO treatment in ICH (Zeng et al., 2018) or AIS patients is still lacking; some clinical trials of DFO in ICH and AIS are currently ongoing.

The rationale for iron chelation as a neuroprotectant relies primarily on the idea that chelation removes free iron present in the blood and enhances iron elimination in the urine. After gaining access to other body compartments, DFO could also act by chelating free extracellular and cytosolic iron, whereas its effect on blood TSAT or blood total iron levels is modest (Tauchenová et al., 2016). In fact, DFO was shown to have

an extremely slow removal rate of iron from Tf (Abergel and Raymond, 2008). This, together with the fact that DFO half-life in the plasma is extremely short (20 min), poses serious doubts about the effectiveness of the treatment with intravenous DFO unless administered with continuous infusion. Recently, another iron-regulator, apotransferrin, administered after stroke AIS onset, was found to be neuroprotective because of the systemic reduction of blood TSAT and a reduction of NMDA ischemia-induced neuronal iron import (DeGregorio-Rocasolano et al., 2018).

#### CONCLUSION

Increasing number of studies show that ferroptosis is involved in several neurodegenerative diseases. This review summarizes the knowledge available thus far regarding the role of iron and ferroptosis in the execution of neuronal cell death during AIS and ICH. In AIS-induced cell death, the role of systemic iron, Tf, TSAT, and hepcidin on excitotoxic damage, infarct size, and neurological outcome in rodent models of stroke might have important clinical implications. Following the AIS event, high levels of extracellular glutamate and impairment of the BBB allow the NMDAR-mediated influx of iron, either bound or not to Tf, into neurons. In addition, high extracellular glutamate prevents the neuronal production of the antioxidant GSH or the function of GPX4, thereby promoting a form of cell death in which glutamate receptor overactivation boosts neuronal iron uptake, which in turn promotes an overwhelming production of membrane peroxides. The role of iron in ICH neurodegeneration

#### REFERENCES


has been known for long, since blood spillover results in iron-Hb and iron–heme complexes at the extracellular milieu of brain parenchymal cells, thereby contributing to iron-induced lipid peroxidation and neuronal death. The review documents the beneficial effects of iron chelators such as DFO, physiological molecules with neuroprotective potential, and specific inhibitors of ferroptosis liproxstatin-1 or ferrostatin-1 in reducing neuron degeneration and improving neurologic deficits induced by ischemic and hemorrhagic stroke subtypes.

#### AUTHOR CONTRIBUTIONS

The three authors have contributed in a similar way to the writing of the article.

# FUNDING

This study was supported by the following grants: Instituto de Salud Carlos III (ISCIII) INVICTUS PLUS RD16/0019/0020 that was susceptible to be co-financed by FEDER funds and Agència de Gestió d'Ajuts Universitaris i de Recerca 2017 SGR 1520. We acknowledge the funding received from "la Caixa" Foundation CI15-00009 and from the European Institute of Innovation and Technology (EIT) PoC-2016-SPAIN-04. EIT receives support from the European Union's Horizon 2020 research and innovation program.



of hemorrhagic transformation after tPA (tissue-type plasminogen activator) administration in thromboembolic stroke mice. Stroke 49, 2163–2172. doi: 10.1161/STROKEAHA.118.021540





**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 DeGregorio-Rocasolano, Martí-Sistac and Gasull. 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.

# Iron in Neurodegeneration – Cause or Consequence?

#### Alain Ndayisaba, Christine Kaindlstorfer and Gregor K. Wenning\*

Department of Neurology, Medical University of Innsbruck, Innsbruck, Austria

Iron dyshomeostasis can cause neuronal damage to iron-sensitive brain regions. Neurodegeneration with brain iron accumulation reflects a group of disorders caused by iron overload in the basal ganglia. High iron levels and iron related pathogenic triggers have also been implicated in sporadic neurodegenerative diseases including Alzheimer's disease (AD), Parkinson's disease (PD), and multiple system atrophy (MSA). Iron-induced dyshomeostasis within vulnerable brain regions is still insufficiently understood. Here, we summarize the modes of action by which iron might act as primary or secondary disease trigger in neurodegenerative disorders. In addition, available treatment options targeting brain iron dysregulation and the use of iron as biomarker in prodromal stages are critically discussed to address the question of cause or consequence.

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Paul Anthony Adlard, Florey Institute of Neuroscience and Mental Health, Australia Bing Zhou, Tsinghua University, China Po-Wah So, King's College London, United Kingdom

\*Correspondence:

Gregor K. Wenning gregor.wenning@i-med.ac.at

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 05 November 2018 Accepted: 14 February 2019 Published: 01 March 2019

#### Citation:

Ndayisaba A, Kaindlstorfer C and Wenning GK (2019) Iron in Neurodegeneration – Cause or Consequence? Front. Neurosci. 13:180. doi: 10.3389/fnins.2019.00180 Keywords: iron, neurodegeneration, neurodegenerative mechanisms, mitochondrial dysfunction, autophagiclysosomal dysfunction, protein aggregation, neuroinflammation

# INTRODUCTION

Iron is involved in an abundant number of cellular processes in the brain including mitochondrial respiration, myelin synthesis, DNA synthesis, oxygen transportation, neurotransmitter synthesis and cellular metabolism (Stankiewicz and Brass, 2009; Ward et al., 2014). In the CNS, iron is present in neurons, oligodendrocytes, astroglia and microglia cells.

Within the cell, iron mediates essential functions due to its capability to participate in reaction of electron transfer thereby switching between two states: ferrous (II) and ferric (III) iron. This mechanism represents a double-edge sword because distinct levels of reactive oxygen species (ROS) are produced during these reactions enabling calcium-mediated basal synaptic transmission and long-term potentiation (Muñoz et al., 2011), but on the other hand may catalyze the so-called Fenton reaction, which generates ROS in high amounts detrimental for cellular function and eventually survival (Halliwell, 2006). Thus, tight regulation of intracellular iron homeostasis is required which is facilitated by sequestration of iron into iron-binding proteins like heme, ferritin, neuromelanin, and iron-sulfur clusters among others (Kruszewski, 2003).

Iron levels in the brain and body increase sharply up to 30 years of age due to a metabolic need during the growth process and remain stable during adulthood (Bolognin et al., 2009; Stankiewicz and Brass, 2009; Apostolakis and Kypraiou, 2017). In the aging brain however, region-specific increase of total iron is observed, probably triggered by inflammation, increased blood-brain barrier (BBB) permeability, redistribution of iron within the brain, and changes in iron homeostasis (Conde and Streit, 2006; Farrall and Wardlaw, 2009; Ward et al., 2014) and shows highest iron levels in the basal ganglia. In addition, iron accumulation varies among brain cell types, as neurons, micro- and astroglia accumulate iron over their lifespan, whereas oligodendroglial iron levels remain stable (Connor et al., 1990). Changes in regional iron distribution have been demonstrated consistently in neurodegenerative diseases via magnetic resonance imaging (MRI) (Duyn, 2010)

and at post-mortem examinations (Hare et al., 2012) and since aging represents the number one risk factor for neurodegeneration, there may be a link between age-related iron accumulation and neurodegeneration (Zecca et al., 2001, 2004; Ward et al., 2014).

Alzheimer's disease (AD), Parkinson's disease (PD), multiple system atrophy (MSA), dementia with Lewy bodies, amyotrophic lateral sclerosis, Huntington's disease (HD), frontotemporal dementia, corticobasal degeneration, and progressive supranuclear palsy (PSP) are primarily characterized by the deposition of insoluble protein aggregates which colocalize with iron (Coffey et al., 1989; Berg and Hochstrasser, 2006; Muller and Leavitt, 2014; Wang et al., 2016; Fernández et al., 2017; Lee et al., 2017; Sheelakumari et al., 2017; Kaindlstorfer et al., 2018; Lane et al., 2018; Moreau et al., 2018), suggesting a link between those clinically and pathologically distinct disease entities. This raises the question whether iron dyshomeostasis represents a critical factor in initiating neurodegeneration, whether it contributes to acceleration of widespread pathology as a result of nerve cell death and the consecutive release of intracellular components or whether neurodegeneration and iron accumulation constitute two completely unrelated events appearing in parallel.

In this review we aim to outline the potential links between pathophysiological mechanisms and the role of iron in neurodegeneration.

#### BRAIN IRON METABOLISM IN HEALTH AND AGING

Both excess and deficiency of iron lead to impaired brain function, thus tight regulation of iron metabolism is critical.

Iron enters the brain either bound to transferrin, thereby crossing the BBB or the blood-cerebrospinal fluid (CSF) barrier (Moos and Morgan, 2000), or possibly unbound, especially in conditions that result in iron overload as transferrin becomes saturated with iron. The exact uptake mechanisms in the latter case are unknown, but it is hypothesized that free serum iron might be reduced by cellular reductants such as ascorbate (Lane and Lawen, 2008; Lane et al., 2010; Lane and Richardson, 2014) to then be imported by divalent metal transporter 1 (DMT1) or ZRT/IRT-like proteins (ZIPs) like ZIP14 or ZIP8 (Jenkitkasemwong et al., 2012). Iron uptake into astroglia is thought to be mediated mainly by non-Tf bound iron (Ashraf et al., 2018), as TfR1 has so far only been reported in vitro (Pelizzoni et al., 2013). DMT1, a transporter located in cellular and endosomal membranes, is found in astrocytes and is believed to facilitate transport of non-transferrin bound iron into the cytoplasm (Xu and Ling, 1994). How iron is exported from the endothelial cells is still elusive (Ward et al., 2014). Within the CSF, iron occurs mainly as holo-transferrin (two ferric iron atoms bound to apo-transferrin) that interacts with TfR1. Neurons internalize the Tf-TfR1 complex into endosomes, where iron is separated from transferrin after acidification, converted into its ferrous form via reductase STEAP3 (Ohgami et al., 2005) and transported into the cytoplasm via DMT1 (Moos and Morgan, 2004).

Iron, prone to contribute to oxidative stress, can be (i) stored within ferritin (Zecca et al., 2004), (ii) imported into mitochondria, probably via so-called mitoferrins and TfR2 (Mastroberardino et al., 2009; Horowitz and Greenamyre, 2010), to enable biosynthesis of heme and iron-sulfur clusters and contribute in the respiratory chain reaction, or (iii) be released from the cell via ferroportin 1 (Ward et al., 2014). Intracellular iron homeostasis is tightly modulated by the iron regulatory protein (IRP) and iron-responsive element (IRE) signaling pathways (Pantopoulos, 2004; Zhang D.L. et al., 2014). IRP1 and IRP2 are two RNA-binding proteins that interact with IREs, non-coding sequences of messenger RNA (mRNA) transcripts to alter translation of ferritin, ferroportin and TfR mRNA. Ferritin H and L subunits or ferroportin mRNA transcripts carry IREs within the 5<sup>0</sup> -untranslated region (UTR), whereas mRNA transcripts for TfR and DMT-1 carry IRE motifs at the 3<sup>0</sup> - UTR. Cytosolic iron binds to IRPs and induces a conformational change within the molecule that does not allow attachment to IREs. Decreased iron levels on the other hand facilitate IRP–IRE interaction: IRP binding at the 5<sup>0</sup> -UTR inhibits further mRNA translation of ferritin subunits and ferroportin; at the 3<sup>0</sup> -UTR, binding protects against endonuclease cleavage (Pantopoulos, 2004; Zhou and Tan, 2017). Ferritin represents the dominant iron storage protein in the CNS, mostly found in glia and also within neurons, whereas neuromelanin (NM) captures large amounts of iron in certain neuronal populations for longer-term storage (Zucca et al., 2017). Recent studies have demonstrated that human poly-(rC)-binding proteins 1–4 (PCBPs 1–4) are implicated in iron transfer to ferritin (Philpott, 2012; Leidgens et al., 2013; Frey et al., 2014; Yanatori et al., 2014), which is a 24 subunit heteropolymer with heavy chains (H-type ferritin) with ferroxidase activity and light chains (L-type ferritin) crucial for subsequent iron storage. H-type ferritin occurs more abundantly in neurons for rapid mobilization and use, whereas in astroand microglia L-type ferritin predominates for iron storage. In oligodendrocytes, both forms of ferritin are expressed (Ashraf et al., 2018). Neuromelanin (NM), a dark brown pigment, is present in dopaminergic neurons of the substantia nigra, the noradrenergic neurons of locus coeruleus, the ventral tegmental area, the ventral reticular formation and the nucleus of the solitary tract in the medulla oblongata (Zecca et al., 2004; Fedorow et al., 2005), but it has also been detected in the putamen, premotor cortex and cerebellum in lower amounts (Zecca et al., 2008; Engelen et al., 2012). Ferritin degradation by the autophagy-lysosome system (Asano et al., 2011) initiates iron release which can then be reutilized or exported, mainly through ferroportin 1 (Biasiotto et al., 2016). This requires ferroxidases ceruloplasmin and hephaestin to oxidize iron for export (Hentze et al., 2004). In addition, heme-oxygenase 1 represents a stress protein which degrades heme to ferrous iron in order to maintain iron homeostasis (Nitti et al., 2018). Systemic ferroportin levels are regulated by circulating hepcidin, the main iron regulatory hormone in the body – during iron overload and inflammation, hepcidin induces ferroportin internalization and degradation (Wang and Pantopoulos, 2011). The origin of hepcidin within the brain is unknown, It may be locally produced or systemically derived by passing the BBB (Vela, 2018). Conditional ferroportin

knock-out mice for example do not show any significant intracellular iron accumulation in the brain, nor do they show behavioral or histological deficits compared to wildtype mice (Matak et al., 2016), suggesting that other cellular iron export mechanisms exist.

Iron accumulates as a function of the aging brain and thereby the levels of labile, potentially harmful iron increase (Ward et al., 2014). Iron accumulating at toxic levels within neurons, as seen in neurodegeneration, may lead to cell death via apoptosis, autophagy, necrosis or ferroptosis, a recently discovered mechanism of iron-mediated cell death distinct from apoptosis (Dixon et al., 2012). In glial cells however, iron accumulation triggers the release of pro-inflammatory cytokines, thereby creating a pro-inflammatory environment (Williams et al., 2012) which promotes neurodegeneration (Xu et al., 2012).

The underlying mechanisms of age and disease related region and cell specific differences in iron levels have not been fully elucidated yet, but the distinct iron richness of the basal ganglia may explain its increased vulnerability to neurodegeneration (Hallgren and Sourander, 1958; Zecca et al., 2004; Xu et al., 2012). In neurons of substantia nigra and locus coeruleus, premotor cortex, putamen, and cerebellum, neuromelanin levels are increased for sequestration of iron, however neuromelanin may itself exhibit toxic functions which are further explained in later sections (Zecca et al., 2004, 2008).

Astro- and microglia exhibit increased iron deposition and ferritin concentrations throughout their lifespan. In contrast, oligodendrocytes that constitute the main iron reservoir in the CNS (Connor et al., 1990; Kaindlstorfer et al., 2018) do not accumulate iron by aging. Iron is essential in the development of oligodendrocytes to allow proper axon myelination, as iron deficiency induces hypomyelination that persists even after the iron imbalance has been corrected (Todorich et al., 2009). The role of iron in oligodendrocytes is further exemplified by Tim-2, a receptor selective for H-ferritin uptake (Todorich et al., 2008). Neuroinflammation in the elderly brain is induced via activated microglia, a lot of which are ferritin-positive (Kaneko et al., 1989). Of those, most exhibit a dystrophic-type morphology, and it has been suggested that iron uptake into the activated microglia cells may represent another source of toxic iron, as seen in neurodegenerative diseases (Rathnasamy et al., 2013; McCarthy et al., 2018; Refolo et al., 2018).

## IRON IN NEURODEGENERATIVE DISEASES – A ONE-SIZE-FITS-ALL?

Iron accumulation as a hallmark feature has been confirmed in many neurodegenerative diseases. **Figure 1** depicts dysfunctional pathways colocalizing with iron accumulation in neurodegeneration. Here, we review the contribution of iron to key mechanisms in neurodegeneration and evaluate its impact on pathogenesis.

#### Iron and Mitochondrial Dysfunction

Mitochondria provide neurons and glia with energy in the form of ATP via oxidative phosphorylation of glucose. Functional disruption of this pathway is likely to contribute to neurodegeneration. In PD, environmental toxins that disrupt mitochondrial complex I function like rotenone induce parkinsonism (Talpade et al., 2000; Dauer and Przedborski, 2003). Also, genetic causes of early-onset PD including PARK2, PINK1, and DJ-1 may interfere with mitochondrial autophagy, thus leading to mitochondrial dysfunction (Batlevi and La Spada, 2011). Moreover, transcriptional dysregulation of coactivator PGC-1α and PPAR transcription factors, including PPAR-δ (Dickey et al., 2016) may play an important role in sporadic AD (Katsouri et al., 2016), PD, and HD. Vulnerability of neural tissue to mitochondrial dysfunction is further demonstrated in singlegene mutations causative of systemic defects in mitochondrial quality control, that can trigger neurodegeneration without having an effect on peripheral tissue (Batlevi and La Spada, 2011). In an iPSC-model of familial PD, accumulation of oxidized dopamine as an effect of mitochondrial oxidative stress has downstream effects on lysosomal function and α-synuclein (α-syn) aggregation (Burbulla et al., 2017). Inversely, lysosomal dysfunction resulting in insufficient mitophagy shows that dysfunction of one system inevitably impairs the other (Plotegher and Duchen, 2017). A large part of intracellular iron is utilized for synthesis of heme and iron-sulfur clusters in mitochondria (Singh et al., 2014). Thus, iron homeostasis is heavily dependent upon proper mitochondrial function.

Neurodegeneration with brain iron accumulation comprises a number of pediatric and adult neurological diseases characterized by parkinsonism and dementia, and excessive iron accumulation in the basal ganglia among other regions (Hogarth, 2015). Distinct genetic causes have been described within this disease spectrum, yet no direct link between neurodegeneration and iron accumulation can be spanned in most forms. Mutations in PANK2, a gene implicated in coenzyme A biosynthesis, lipids synthesis and citric acid cycle (Zhou et al., 2001), results in panthoten kinase-associated neurodegeneration (PKAN). Though this mutation may not affect iron homeostasis directly, subsequent iron accumulation suggests that mitochondrial dysfunction may trigger iron dyshomeostasis. In iPSC-derived neurons from a PKAN patient, early disease stage phenotype is reflected by mitochondrial deficiency but absent iron dysregulation and iron chelation treatment causes further exacerbation of cellular dysfunction through iron deficiency (Arber et al., 2017).

In Friedreich's ataxia, expansions of unstable nucleotide repeats in the FXN gene result in reduced expression of frataxin (Campuzano et al., 1996), a protein thought to act as molecular chaperone in iron-sulfur clusters and heme biosynthesis as well as iron storage site (Alsina et al., 2018). Thus, lack of frataxin leads to mitochondrial iron overload and generation of ROS, which in turn impairs mitochondrial function creating a vicious cycle (Horowitz and Greenamyre, 2010). Decreased levels of frataxin were also observed in HD human and mouse brains, together with an increase of iron uptake protein mitoferrin 2 and mitochondrial iron accumulation (Agrawal et al., 2018). Interestingly, neonatal iron supplementation increases mitochondrial iron accumulation and mitochondrial dysfunction exacerbating neurodegeneration in HD mice. From

this dataset it is difficult to conclude whether early-life iron exposure may cause mitochondrial dysfunction as an initiating step of neurodegeneration. It has been discussed that iron fortified infant formula within a critical window of susceptibility to iron overload may contribute to neurodegeneration in later life (Hare et al., 2015). Still, this remains highly speculative without evidence from prospective studies, but might indicate a causal role of iron in mitochondrial dysfunction. In a Drosophila model of Friedreich's ataxia, upregulation of mitoferrin associated with frataxin deficiency is observed. Interestingly, targeted downregulation of mitoferrin ameliorates the phenotype while worsening overall life span, indicating that due to its physiological function dysregulation of mitoferrin has detrimental consequences (Navarro et al., 2015). Knockdown of mitoferrin-1 in an Alzheimer model of Caenorhabditis elegans reduces paralysis rate and slows progression of AD by decreasing mitochondrial iron accumulation and ROS generation (Huang et al., 2018), showing that mitochondrial iron homeostasis may substantially impact neurodegeneration.

Another aspect favoring a causal role of iron in mitochondrial dysfunction is that damage of mitochondrial DNA due to oxidative stress is detected consistently at an early stage of neurodegeneration, and its vulnerability can be explained by a lack of protective histones (Shokolenko et al., 2009). Dysfunctional mitochondria produce less heme and ISCs, which causes activation of IRPs and further iron accumulation leading to positive feedback loop (Mena et al., 2015). In nigral dopaminergic neurons of human PD and the rotenone mouse model, transferrin accumulates intracellularly and large parts locate in mitochondria, which may be explained by the expression of TfR2 in the mitochondrial membrane of substantia nigra dopaminergic neurons (Mastroberardino et al., 2009). Elevated ROS levels in mitochondria oxidize holo-transferrin, which then tends to release ferrous, redox-reactive iron (Mastroberardino et al., 2009). Treating rats and monkeys with rotenone leads to downregulation of TfR1 in the presence of iron accumulation, whereas TfR2 levels remain elevated (Mastroberardino et al., 2009), supporting previous findings on TfR2 stability during iron overload in hepatic tissue (Fleming et al., 2000; Mifuji et al., 2006). TfR2 stability in conditions of neurodegeneration might thus exacerbate mitochondrial dysfunction, which in turn further affects iron handling.

#### Iron, Protein Misfolding, and Aggregation

Neurodegenerative diseases are characterized by extra- and/or intracellular accumulation of abnormally folded proteins and selective regional cell death in a progressive and predictable fashion (Fu et al., 2018). This includes AD, PD, MSA, dementia with Lewy bodies, PSP, frontotemporal dementia, corticobasal degeneration, HD, amyotrophic lateral sclerosis, spinocerebellar ataxias, chronic traumatic encephalopathy, and prion diseases (Fu et al., 2018). These disease entities are distinguished by clinical symptoms reflecting areas of the brain and cell types affected, by prevalence rates, risk factors and genes involved. The underlying mechanisms of protein aggregation are believed to include β-sheet-rich structures that aggregate into small oligomers and large fibrillar inclusions. There is increasing evidence that the misfolding and aggregation of proteins follows the seeding-nucleation model (Jarrett and Lansbury, 1993), where preformed seeds recruit endogenous native proteins into growing aggregates, which may then be fragmented into smaller oligomers again that are capable to spread between cells. This concept is being more and more accepted within the neuroscience community. Yet, it does not explain initial mechanisms underlying misfolding of distinct disease-associated proteins in sporadic neurodegeneration. Here, we summarize the current knowledge of interactions between iron and proteins implicated in neurodegenerative diseases.

#### Iron and α-Synuclein

Central to the pathogenesis of PD is the 17 kDa protein α-synuclein, which accumulates mainly into neuronal cytoplasmic inclusions termed Lewy bodies (Spillantini et al., 1997). Interestingly, α-synuclein can bind both ferrous and ferric iron (Peng et al., 2010), however only latter has been shown to promote aggregation (Levin et al., 2011). Phosphorylated α-synuclein at residue S129 is associated with Lewy bodies, and even though its role in PD pathogenesis has not been fully elucidated yet, it has been shown that phosphorylated synuclein exhibits increased binding affinity toward ferrous iron at the c-terminus (Lu et al., 2011). Iron and free radical generators like dopamine are able to induce aggregation of the wildtype protein in an in vitro model, and this effect is exacerbated by the cooccurrence of α-synuclein mutations A53T and A30P, suggesting a detrimental interplay in genetic forms of PD (Ostrerova-Golts et al., 2000). Importantly, α-synuclein possesses and IRE and IRP complexes during iron overload increase its translation resulting in elevated total α-synuclein burden (Friedlich et al., 2007; Zhou and Tan, 2017). On the other hand, it has been reported that α-synuclein exhibits ferrireductase activity that is increased in point mutations (A53T, A30P, or E46K), which may result in altered iron uptake and metabolism in the presence of divalent metal transporter 1 (Davies et al., 2011). In Saccharomyces cerevisiae and Caenorhabditis elegans models of Parkinson's disease, expression of α-synuclein may phenocopy a high iron condition by inhibiting recycling of iron transporter ortholog of ferroportin or membrane-bound ortholog of ceruloplasmin from endocytic vesicles back to the cell membrane. Iron chelation with desferoxamine partially rescued dopaminergic neurotoxicity in Caenorhabditis elegans (Patel et al., 2018).

Taken together, there seems to be an intricate interplay between α-synuclein and iron, both of which promote disease progression in synucleinopathies synergistically.

#### Iron and Aβ

More than half a decade ago Goodman provided first evidence for iron accumulation in senile plaques of post-mortem AD brains (Goodman, 1953), and more recently, brain imaging of patients with early stage AD demonstrated increased iron concentrations co-localizing with Aβ plaques which may both promote disease development and progression.

Long-term treatment of APP mice with 3,5,5 trimethyhexanoyl ferrocene reveal accelerated plaque formation and more senile morphology, microglial iron inclusions and increased amounts of Aβ (Peters et al., 2018). Although the redox state of iron located in plaques is still elusive, recent studies detected iron oxide magnetite nanoparticles in the core of senile plaques (Plascencia-Villa et al., 2016; Everett et al., 2018). Furthermore, Telling et al. (2017) provide evidence for a direct iron–amyloid interaction within senile plaques. Increased iron levels are believed to enhance Aβ production via downregulation of furin, a proprotein convertase participating in α-secretasedependent APP processing, which in turn activates β-secretase implicated in Aβ generation (Silvestri and Camaschella, 2008). SH5SY cells exposed to ferric ammonium citrate confirm an increase in β-secretase levels coupled with enhanced Aβ<sup>42</sup> production, in addition iron treatment promotes accumulation of APP (Banerjee et al., 2014). Interestingly however, in a recent study ferric iron treatment in rat cortical neurons facilitates nonamyloidogenic processing of APP via α-secretase, intracellular localization of sAPPα impaired enzymatic activity of β-secretase, and thus decreases levels of sAPPβ and Aβ (Chen et al., 2018). Boopathi and Kolandaivel (2016) show that ferrous iron induces a structural conformation in Aβ toward a β-sheet structure which tends to form oligomers and fibrils, whereas other groups demonstrate that ferric iron promotes aggregation of both Aβ<sup>40</sup> and Aβ<sup>42</sup> as well as cytotoxicity in vitro (Tahmasebinia and Emadi, 2017; Galante et al., 2018). Aβ<sup>42</sup> promotes magnetite nanoparticle formation (Tahirbegi et al., 2016), thus coming full circle and providing evidence for a causative role of iron in AD pathogenesis. The IRP-IRE signaling pathway is also involved in proteostasis of APP (Rogers et al., 2008). Thus, sustained iron overload may result in elevated Aβ-levels, the core constituent of extracellular plaques in AD. Intriguingly, genetic manipulation of iron metabolism substantially influences Aβ toxicity in model organisms. Overexpression of ferritin heavy chain, but also iron chelation, rescue Aβ-mediated toxicity in a Drosophila model expressing Aβ (Rival et al., 2009; Liu et al., 2011).

#### Iron and Tau

Tauopathies including AD, PSP, corticobasal degeneration, frontotemporal dementia, but also PD and HD, are characterized by neurofibrillary tangles, the core of which consists of paired helical filaments composed of hyperphosphorylated tau. Tau facilitates microtubule stabilization and regulation, regulation of

axonal transport (Magnani et al., 2007; Dixit et al., 2008) and likely plays a role in neurotransmission and iron metabolism by the trafficking of APP to the cell surface (Lei et al., 2012).

These functions may be disrupted following hyperphosphorylation of tau, and iron has been shown to mediate this modification depending on its redox state. Whereas Fe (II) forms a reversible interaction through threonine residues, Fe (III) induces an irreversible conformational change, both of which promote aggregation of tau (Ahmadi et al., 2017). Hyperphosphorylation of tau, one of the key steps toward aggregation, may be induced by iron via (CDK25)/P25 complex and GSK-3β (Xie et al., 2012; Guo et al., 2013). In addition, Fe (III) may mediate nitration of tau, which prevents microtubule stabilization and is commonly found in neurofibrillary tangles and senile plaques (Nakamura and Lipton, 2011). Tau accumulation in tangles on the other hand leads to induction of heme-oxygenase 1, an antioxidant that promotes release of the redox-active Fe (II), which releases free radicals to generate oxidative stress (Ward et al., 2014). This in turn promotes tau hyperphosphorylation and aggregation (Lane et al., 2018). Intracellular iron accumulation, a common feature in tauopathies, leads to loss of tau function, which prevents iron export through impaired transport of APP to the cell membrane, where it stabilizes ferroportin (Wong et al., 2014) resulting in a vicious cycle of iron accumulation and tau pathology.

# Iron and Autophagic-Lysosomal Dysfunction

Autophagic-lysosomal dysfunction is a hallmark feature of neurodegenerative proteinopathies, either due to mutations directly affecting the autophagy-lysosome machinery, or secondary to aggregation of misfolded proteins. Iron dyshomeostasis may occur as a consequence of inadequate ferritinophagy, a term coined to describe the lysosomal degradation of ferritin molecules mediated via nuclear receptor co-activator 4 (NCOA4) (Mancias et al., 2014; Biasiotto et al., 2016). As both autophagy activation and inhibition have been reported in neurodegeneration, aberrant ferritinophagy may lead to iron dyshomeostasis manifesting as iron deficiency or overload, both of which exert deleterious effects on cell survival. In line with that, quantifications of ferritin levels in substantia nigra of PD brain reveal contradictory findings in the literature (Zecca et al., 2001; Koziorowski et al., 2007; Sian-Hülsmann et al., 2011; Friedman and Galazka-Friedman, 2012; Galazka-Friedman et al., 2012). NCOA4 levels, targeting ferritin heavy chain 1 for ferritin degradation, may be critical for ferritinophagy as factors such as oxidative stress regulate NCOA4 gene transcription (Sharifi et al., 2008). Pharmacological inhibition of autophagy leads to NCOA4 accumulation (Dowdle et al., 2014), probably as a compensatory mechanism. In addition, NCOA4-depletion seems to have a protective effect during oxidative stress (Mancias et al., 2014), possibly contrabalancing transcriptional and translational regulation of ferritin in situations of stress (Torti and Torti, 2002). Another part of the autophagic machinery slowly accumulating iron high amounts of iron, copper and zinc as well as various toxins over the course of several decades constitutes neuromelanin (Zecca et al., 2008). In addition, intracellular dopamine in excess can be metabolized into the more stable NM molecule. However, externalization of neuromelanin organelles and subsequent release of content as seen in PD activates microglia, which may accelerate further cell death and release of neuromelanin-containing vesicles, thereby creating a self-perpetuating cycle (Zhang W. et al., 2013; Zucca et al., 2017). Recent data suggests that neuromelanin-containing vesicles in PD may represent a specialized autolysosomal compartment, where proteins and lipids not otherwise degraded are accumulated in a very slow turnover rhythm (Zucca et al., 2018). Neuromelanin accumulates within vesicles over time, and in PD, and a strong correlation between nigral cell loss and amount of pigmented neuromelanin-immunoreactive neurons could be shown, providing first evidence that neuromelaninpigmented cells may exhibit increased vulnerability during PD pathogenesis (Zucca et al., 2018). One possible explanation is that at a certain threshold of iron level in PD brains buffering capacities in the form of NM and ferritin are exhausted (Good et al., 1992; Jellinger et al., 1992), and then redox-active iron is freed and may have deleterious consequences and significantly contributes to subsequent pathology (Faucheux et al., 2003).

#### Iron and Neuroinflammation

Microglia are in a constant cross-talk with neurons and are capable of sensing neuronal dyshomeostasis. Microglial activation is a healthy reaction against disease and injury, accompanied by the release of pro- and anti-inflammatory factors, ROS and recruitment molecules (Colton and Gilbert, 1987; Graeber et al., 1988; Hurley et al., 1999). Chronic neuroinflammation however, a hallmark feature of neurodegenerative diseases, creates a pro-apoptotic environment resulting in neuronal death and thus becomes a driver of neurodegeneration itself (Lull and Block, 2010). AD and PD brains show extensive proliferation of reactive macrophages and activated microglia. In addition to the cocktail of cytokines and ROS released by microglia cells, they also produce high levels of NAPDH oxidase and nitric oxide synthase, two main players of oxidative stress and ROS generation during inflammation (Urrutia et al., 2014).

Interestingly, microglia turn out to be far more efficient in sequestering iron compared to other brain cell types, as levels of ferritin induced by iron exposure in an organotypic slice culture model are highest in microglia and oligodendrocytes, whereas ferritin is rarely detected in astrocytes and, surprisingly, never in neurons (Healy et al., 2016).

Exposing cultured neurons to ferric iron reveals accumulation of iron, however microglia accumulates eightfold more iron than neurons and 4.7-fold more than cultured astrocytes after 24 h (Bishop et al., 2011). Activated microglia in substantia nigra of post-mortem PD brains exhibit enhanced ferritin immunoreactivity (Mirza et al., 2000), yet iron burden does not correlate with ferritin in a linear fashion, as high iron saturation in microglia may be achieved despite low ferritin baseline, as seen in conditions of acute oxidative stress (Mehlhase et al., 2005; Thomsen et al., 2015). In neurodegeneration, iron-containing microglia co-localize with affected areas (Andersen et al., 2014) and there is some intriguing

evidence pointing toward a substantial contribution of microgliaassociated iron in disease. Ferrous iron in primary midbrain cultures causes progressive dopaminergic neurodegeneration mediated by microglial β-nicotinamide adenine dinucleotide phosphate oxidase 2, a superoxide generator, and inhibition of the enzyme prevents neurotoxicity reflected in decreased levels of superoxide (Zhang W. et al., 2014). Intriguingly however, Wang et al. report that iron accumulation in microglia alone is not sufficient to trigger release of proinflammatory cytokines TNF-α, IL-1β, and IL-6, instead MPP+ is required to trigger enzymatic release, which is then aggravated if iron accumulation is present (Wang et al., 2013). Inflammatory stimuli including TNF-α, IL-β and toll-like receptor-4 agonist lipopolysaccharide (Yang et al., 2002) on the other hand are able to trigger neuronal iron accumulation via alteration of DMT1 and FPN mRNA levels (Urrutia et al., 2013). This is supported by another study showing increased DMT1 and TfR1 and decreased FPN1 protein levels following TNF-α or IL-1β treatment (Wang et al., 2013).

Increased iron uptake in substantia nigra in PD patients has been linked with altered expression of lactoferrin receptor on vulnerable dopaminergic neurons (Faucheux et al., 1995). In addition, lactoferrin has been implicated in AD pathogenesis due to its presence in senile plaques and neurofibrillary tangles in the limbic system (Wang et al., 2015). Lactoferrin is produced only by activated microglia (Fillebeen et al., 2001; An et al., 2009; Wang et al., 2015) and it is suggested that it may have a protective function in neurodegeneration by modulating mitochondrial calcium homeostasis (Fillebeen et al., 1999; Rousseau et al., 2013) rather than sole iron chelation following MPP+ injury (Wang et al., 2015).

Astroglia as part of the BBB are involved in strict separation and regulation of brain iron metabolism from the periphery (Moos et al., 2007; Burkhart et al., 2016). They express coeruloplasmin to facilitate iron uptake and distribution (McCarthy and Kosman, 2015) and hephaestin, indicating ferroxidase activity (Wang et al., 2007). However, astroglia contain relatively small amounts of ferritin and iron (Mirza et al., 2000; Zecca et al., 2004; Bartzokis et al., 2007), yet in vitro studies in cultured astrocytes show that cultured astrocytes are surprisingly resistant to ferrous iron compared to neurons and oligodendrocytes, even though intracellular iron burden is comparable (Kress et al., 2002; Oshiro et al., 2008). Possibly this is due to an extensive antioxidative system including metallothioneins (Waller et al., 2018). Astroglia possess heme oxygenase-1, which has been shown to be neuroprotective in different models of PD by impeding oxidative stress (Xu et al., 2016; Yu et al., 2016), it has to be noted however that during the degradation of heme catalyzed by the enzyme, bio-reactive iron is released to participate in Fenton's reaction (Cuadrado and Rojo, 2008). Astroglial increase of iron levels is observed in substantia nigra of PD (Jellinger et al., 1990), which is accompanied by astroglial activation, reflected in part by the up-regulation of lipocalin-2, which, if overexpressed in experimental conditions, initiates nigrostriatal dopaminergic neurodegeneration that is aggravated following iron treatment (Kim et al., 2016). Neurotrophic factors BDNF and GDNF, produced by astroglia besides oligodendrocytes, prevent neuronal iron accumulation by affecting IRPs linked with DMT1 expression (Zhang H.Y. et al., 2014), highlighting the crucial role of astroglial health on iron regulation. In case of iron accumulation, exemplified by a study from Rathore et al. including treatment of rat astrocytes with ferrous iron, increased expression of DMT1, ferroportin and both ferritin subunits are observed. However, co-treatment with TNF-α diminished the increase in H-ferritin, whereas the elevated L-ferritin levels are not altered (Rathore et al., 2012). This raises the question whether microglial activation in form of TNF-α secretion might impair astroglial iron handling and storage. In the 6-OHDA neurotoxin rat model of PD, altered BBB permeability is associated with microglial iron accumulation and increased levels of L-ferritin besides neurodegeneration in substantia nigra (Olmedo-Díaz et al., 2017). Using Mössbauer spectrometry, Friedman et al. (2011) determine decreased levels of L-ferritin in substantia nigra in PD brains in contrast to experimental results.

Iron as cause of neuroinflammation has also been shown in ALS by introducing the G93A mutation of ALS-associated human superoxide dismutase 1 (SOD1) in mice, which results in iron accumulation of ventral motor neurons. Intriguingly, an increase in the enzymatic activity of TNF-α – converting enzyme and subsequent enhanced secretion of soluble TNF-α is observed triggered by iron whereas chelation with deferoxamine mesylate delays disease onset and prolongs overall lifespan (Milanese et al., 2014).

#### Ferroptosis in Neurodegeneration

Ferroptosis represents a recently discovered form of cell death independent of the caspase pathway and involves iron dysregulation, lipid peroxidation and inflammation as major hallmarks (Dixon, 2017). Central to this phenomenon is the depletion of glutathione, an antioxidant that buffers ROS and binds to labile iron (Hider and Kong, 2011) to prevent Fenton reaction with subsequent ROS production. This imbalance in cellular redox homeostasis causes the accumulation of lipid ROS and thus lipid peroxidation (Gao et al., 2016; Abdalkader et al., 2018). Morphological changes reflecting the consequences include reduced mitochondrial volume (Dixon et al., 2012; Xie et al., 2016), reduced density of mitochondrial membranes (Xie et al., 2016), chromatin condensation (Dixon et al., 2012), cytoplasmic swelling, and eventually rupture of mitochondrial outer membrane (Friedmann Angeli et al., 2014; Xie et al., 2016) and plasma membrane (Dixon et al., 2012). A number of potent inducers of ferroptosis have been identified (Stockwell et al., 2017), among which the small molecule erastin potently induces ferroptosis via inhibition of the xCT cystine/glutamate antiporter, as cystine is essential for glutathione synthesis. Reduced gluthatione is utilized by glutathione peroxidase GPX4 (Yang et al., 2014) to detoxify phospholipid hydroperoxides and hydrogen peroxides (Ursini and Bindoli, 1987). Interestingly, GPX4 knock-out in mice revealed not only iron dysregulation, lipid peroxidation and inflammation, but in addition early signs of an AD phenotype including behavior dysfunction and hippocampal neurodegeneration (Seiler et al., 2008; Hambright et al., 2017). The deleterious effects on neuronal health and survival upon

ablation of GPX4 in motor neurons might confer a role of ferroptosis to degenerative motor neuron diseases like amyotrophic lateral sclerosis (Chen et al., 2015).

Key features of ferroptosis including lipid peroxidation and reduced glutathione levels have been detected previously in the substantia nigra in PD (Dexter et al., 1989; Jenner et al., 1992) and colocalize with accumulated iron. According or Jenner et al. these findings may represent early events in the disease (Jenner et al., 1992). In addition, mitochondrial damage, commonly observed in PD as well, may be mediated by the interaction of α-synuclein with mitochondrial membranes, thus aggravating mitochondrial vulnerability and oxidative stress further (Parihar et al., 2008).

Therefore, targeting ferroptosis in neurodegeneration may represent an attractive target for disease modification. Potential candidates include vitamin E analogs, liproxstatins and ferrostatins, deuterated lipids and ACSL4 (Acyl-CoA Synthetase Long Chain Family Member 4) inhibitors (Angeli et al., 2017).

## CAN IRON CAUSE NEURODEGENERATION?

Based on the literature reviewed to this point it is certain that iron contributes substantially to neurodegeneration, however so far, no compelling evidence was provided to attribute a causal role to iron, which would imply (1) detection at an early stage if not as first sign, and (2) evidence for disease modification if targeted appropriately. We aim to address both issues based on current knowledge and progress made to this point.

## Targeting Iron Dyshomeostasis for Disease Modification in Neurodegeneration

In the past few years numerous targets for treatment of iron dysregulation have been identified including reduction of iron burden but also inhibition of downstream effects. However, the numerous physiological functions of iron and defects resulting from aggressive iron depletion need to be considered. Therefore, the following features of treatment of brain iron dyshomeostasis could be taken into account: ability to pass BBB and consecutive cell membranes; chelation of excess iron with moderate affinity to avoid depletion of transferrin and iron-associated proteins; prevention of side effects due to iron removal in areas lacking iron overload (Boddaert et al., 2007; Hamilton et al., 2017; Loza-Rosas et al., 2017). Here, we will discuss potential candidates based on those selection criteria.

#### Deferiprone

Deferiprone is an iron chelator currently in use for systemic iron overload conditions like thalassemia major, but its benefit in neurodegenerative conditions lacking systemic iron dysregulation has been shown in a small trial in 9 patients with Friedreich's ataxia, who received deferiprone for 6 months. Treatment results in decreased iron load in dentate nuclei as visualized by MRI, an area implicated in neurodegeneration. Moreover, motor skills and signs of neuropathy are improved especially in the younger patients, and no severe side effects are reported (Boddaert et al., 2007). Deferiprone acts as "conservative iron chelator" by redistributing iron from overloaded areas to areas deprived of iron, thereby keeping overall iron levels stable (Cabantchik et al., 2013). In a 12 months double-blind pilot study (FAIRPARK-I), 40 PD patients were recruited, and treatment with deferiprone reduces substantia nigra hyperechogenicity in MRI and shows clinical improvement determined by the Unified PD Rating Scale (UPDRS) (Devos et al., 2014).

Interestingly however, in another phase II trial on deferiprone 22 early-disease patients (Martin-Bastida et al., 2017) reveal no change in iron content of putamen, striatum nor substantia nigra except for 3 cases. In dentate and caudate nuclei decreased iron load is determined by MRI. In this study, no improvement in motor-UPDRS is observed, even though a trend is achieved in the high dose-tier (30 mg/kg). An ongoing phase III trial, FAIRPARK-II, will reveal further insights into the potential benefits of deferiprone in PD.

#### Lactoferrin

Lactoferrin (LF) is an iron-binding glycoprotein with 60% amino acid sequence homology to transferrin (Metz-Boutigue et al., 1984). However, LF affinity for iron is 300 times higher than transferrin (Dhennin-Duthille et al., 2000) and it can act by binding two iron, zinc, copper particles or other metals. In AD brain, abnormal LF content was described already a few decades ago (Brown et al., 1985) and in 2010 Wang et al. could show that LF is present in AD model mice, but not healthy wildtype mice (Wang et al., 2010). There is a strong correlation between LF and Aβ, as LF levels increase following generation of Aβ, and with Aβ plaque formation lactoferrin continues to accumulate (van de Looij et al., 2014).

There is evidence of a protective function of lactoferrin as exogenous application results in reduced Aβ deposition and improvement of cognitive decline in AD mice via enhanced nonamyloidogenic processing of APP and α -secretase expression and activity through ERK/1/2-CREB and HIF-1α pathways (Guo et al., 2017). In MPTP mice modeling PD, administration of LF results in reduction of MPTP-induced iron accumulation via suppressed upregulation of DMT1 and TfR. Furthermore, neuroprotection is mediated by an increase in BDNF and HIF-1α with downstream activation of ERK/1/2-CREB pathway (Xu et al., 2018). In addition, LF receptors are found at the BBB, therefore LF may also be utilized as carrier for iron chelators that would otherwise be excluded from CNS. Kamalinia et al. conjugated LF with deferasirox to inject into AD mice, which ameliorates memory function, reduces Aβ levels, and leads to attenuation of apoptosis markers and increase in autophagy (Kamalinia et al., 2013). In a recent pilot study 50 AD patients were treated with lactoferrin for 3 months to find significant improvement in antioxidant and anti-inflammatory markers in serum, and decrease in Aβ42, phosphorylated tau, interleukin-6 and caspase-3 as well as improved cognitive function (Mohamed et al., 2019), possibly opening the doors toward new treatments in AD, however further follow-up data are required.

#### PBT2

Metal chaperone PBT2 is a derivate of clioquinol, an iron, copper and zinc chelator. In a phase II trial for AD clioquinol improved

cognition and decreased Aβ plasma levels (Ritchie et al., 2003), however in a transgenic AD mouse model toxicity manifested by myelinopathy was observed (Zhang Y.H. et al., 2013). PBT2 on the other hand improved cognition and decreased intracerebral Aβ levels in preclinical AD models (Adlard et al., 2011, 2014). A phase II trial in 78 AD patients revealed significant reduction in CSF Aβ<sup>42</sup> and improved cognition by two measures of executive function (Lannfelt et al., 2008; Faux et al., 2010).

#### IRE (50UTR) Inhibitors

Another promising target in the self-perpetuating cycle of iron and disease-associated protein accumulation represents the inhibition of the IRE in the 5<sup>0</sup> -UTR of α-synuclein and APP (Zhou and Tan, 2017). Two compounds have now been tested in clinical trials, these include Posiphen, a phenyl carbamoyl analog of (+)-physostigmine, second-generation 8-hydroxyquinoline analog metal chaperone PBT2, and paroxetine, an antidepressant of the selective serotonin-reuptake inhibitor class.

Posiphen potently inhibits translation of APP and α-synuclein in preclinical models (Shaw et al., 2001; Yu et al., 2013; Teich et al., 2018). In a phase I human trial and a small non-randomized study of five subjects with mild cognitive impairment, administration of posiphen is well tolerated and decreases levels of sAPPα, sAPPβ, total and phosphorylated tau, and inflammatory marker in CSF. In addition, a trend toward lower Aβ<sup>42</sup> is observed (Maccecchini et al., 2012).

Besides, paroxetine is widely used in the treatment of depression acting through selective inhibition of serotonine reuptake and may represent a suitable candidate for drug repurposing in AD. Preclinical in vitro and in vivo evidence supports the effects of paroxetine as chemical IRE modulator on APP expression (Payton et al., 2003; Tucker et al., 2006; Bandyopadhyay and Rogers, 2014), and its clinical safety has obviously been proven due to its indication in depression.

#### Iron as MRI Biomarker for Disease Progression

One of the major drawbacks in therapeutic research of neurodegenerative diseases is a lack of biomarkers for earlier diagnosis, providing a chance for successful disease-modifying treatments. Advances in MRI technology to facilitate more accurate and specific detection of iron may help to determine the biomarker status of iron accumulation at preclinical stages of neurodegeneration. For an extensive review on the use of MRI for brain iron detection we refer to excellent reviews (Schweser et al., 2011; Schipper, 2012).

In premotor PD patients, MRI scans revealed increased echogenicity in substantia nigra, probably indicative of labile iron (Berg et al., 2011). De Marzi et al. (2016) investigated a cohort of patients with idiopathic rapid eye movement sleep behavior disorder (iRBD), a majority of which went on to develop PD. Interestingly, loss of dorsolateral nigral hyperintensity in a T2<sup>∗</sup> map, a sign for low neuromelanin reflecting the loss of dopaminergic neurons (Blazejewska et al., 2013), was determined both in iRBD as well as in PD patients, indicating that loss of dorsolateral nigral hyperintensity may represent a promising biomarker for prodromal PD

(De Marzi et al., 2016). In a cohort with carriers of the apolipoprotein ε4 allele at risk for AD and patients with mild cognitive impairment, representing early stage AD, quantitative susceptibility mapping MRI revealed higher Ab plaque load as well as increased iron concentrations colocalizing with Ab plaques both in cortical as well as sub-cortical brain areas (van Bergen et al., 2016). In Huntington's disease however, examination of premanifest gene carriers and early Huntington's disease patients exhibit no differences in iron content in asymptomatic carriers, whereas in early patients a lower R2<sup>∗</sup> parameter value, indicating alterations in iron burden, was observed. In this case, iron measurement by MRI may rather be a marker of disease progression (Di Paola et al., 2014). In genetically and phenotypically distinct disorders like ALS with variable upper and lower motor neuron affection, a reliable biomarker is urgently needed for both early and accurate diagnosis. MRI in ALS patient cohorts shows hypointensity in deep layers of motor cortex, and increased hypointensity correlates with more severe affection of upper motor neurons (Kwan et al., 2012). Direct comparison of MRI signal changes in ALS with histological examination of post-mortem ALS brains revealed that motor cortex hypointensity corresponds to microglial iron accumulation, which may have utility as marker for disease progression (Pallebage-Gamarallage et al., 2018). Taken together, brain iron detection via MRI will be a powerful tool in the future both for early diagnosis and for evaluation of disease progression in neurodegenerative disorders. Detection of biochemical or structural components implicated in the disease process in addition to brain iron have immense potential to elucidate early pathophysiological changes in the otherwise inaccessible brain.

# CONCLUSION AND OUTLOOK

Experimental data show that many neurodegenerative diseases represent conditions of multifactorial cellular dysfunction associated with iron dyshomeostasis present already at early disease stages, yet preclinical models may not be able to answer what comes first in sporadic neurodegenerative diseases. Intriguingly however, iron substantially contributes to and drives many aspects of neurodegeneration, plus, genetic modification of iron metabolism resulted in rescue of neurodegeneration in different animal models, indicating a causal role of iron in neurodegenerative disorders. Advances in MRI technology to facilitate more accurate and specific detection of iron hold great promise to implement iron as a biomarker for preclinical stages of neurodegeneration, probably a crucial treatment window for upcoming therapies. Moreover, CSF levels of iron related proteins in sporadic neurodegenerative diseases may aid in early diagnosis in the future, like CSF ferritin in AD (Ayton et al., 2015, 2017). Currently, a number of various promising compounds targeting different aspects of iron dysregulation besides conventional iron chelation are being tested at different clinical and preclinical trial stages. Recently discovered mechanisms like ferroptosis may pose promising targets for disease modification via ferrostatins in the future.

#### AUTHOR CONTRIBUTIONS

fnins-13-00180 February 27, 2019 Time: 16:35 # 10

AN wrote the manuscript. CK and GW edited and revised the manuscript.

#### REFERENCES


#### FUNDING

This study was supported by grant F4414 of the Austrian Science Fund (FWF).


neuropathological findings. J. Neuropsychiatry Clin. Neurosci. 1, 400–404. doi: 10.1176/jnp.1.4.400


[125I]ferrotransferrin binding sites in the basal ganglia of control subjects, patients with Parkinson's disease and MPTP-lesioned monkeys. Brain Res. 691, 115–124. doi: 10.1016/0006-8993(95)00629-5



ordered aggregation. J. Biol. Chem. 286, 4248–4256. doi: 10.1074/jbc.M110. 158980




protein, α-Synuclein synthesis, interleukin-1β release, and cholinergic action. Antiinflamm. Antiallergy Agents Med. Chem. 12, 117–128. doi: 10.2174/1871523011312020003


**Conflict of Interest Statement:** CK and GW are employed by the Medical University of Innsbruck, Austria.

The remaining 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 © 2019 Ndayisaba, Kaindlstorfer and Wenning. 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.

# Low Cerebrospinal Fluid Levels of Melanotransferrin Are Associated With Conversion of Mild Cognitively Impaired Subjects to Alzheimer's Disease

Azhaar Ashraf, Jose Andres Alepuz Guillen, Manal Aljuhani, Chantal Hubens and Po-Wah So\* for the Alzheimer's Disease Neuroimaging Initiative

#### Edited by:

Giorgio Biasiotto, Università degli Studi di Brescia, Italy

#### Reviewed by:

Wilf Arthur Jefferies, The University of British Columbia, Canada Goran Simic, University of Zagreb, Croatia Eline A. J. Willemse, VU University Medical Center, Netherlands

> \*Correspondence: Po-Wah So po-wah.so@kcl.ac.uk

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 27 October 2018 Accepted: 14 February 2019 Published: 08 March 2019

#### Citation:

Ashraf A, Alepuz Guillen JA, Aljuhani M, Hubens C and So P-W (2019) Low Cerebrospinal Fluid Levels of Melanotransferrin Are Associated With Conversion of Mild Cognitively Impaired Subjects to Alzheimer's Disease. Front. Neurosci. 13:181. doi: 10.3389/fnins.2019.00181 Department of Neuroimaging, Institute of Psychiatry, Psychology and Neuroscience, King's College London, London, United Kingdom

The disruption of iron metabolism and iron transport proteins have been implicated in the pathogenesis of Alzheimer's disease (AD). Serum melanotransferrin (MTf), a transferrin homolog capable of reversibly binding iron, has been proposed as a biochemical marker of AD. MTf has also been shown to be elevated in iron-rich reactive microglia near amyloid plaques in AD. We examined the association of CSF MTf to hippocampal volumes and cognitive tests in 86 cognitively normal, 135 mild cognitive impairment (MCI) and 66 AD subjects. CSF was collected at baseline for MTf, Aβ, total-tau and phosphorylated-tau measurements. Serial cognitive testing with ADAS-Cog13, Rey's auditory visual learning test (RAVLT), mini-mental state examination (MMSE) were performed alongside hippocampal MRI volumetric analysis for up to 10 years after baseline measurements. High levels of baseline CSF MTf were positively associated with baseline hippocampal volume (R <sup>2</sup> = 22%, β = 0.202, and p = 0.017) and RAVLT scores (R <sup>2</sup> = 7.30%, β = −0.178, and p = 0.043) and negatively correlated to ADAS-Cog13 (R <sup>2</sup> = 17.3%, β = 0.247, and p = 0.003) scores in MCI subjects. Interestingly, MCI subjects that converted to AD demonstrated significantly lower levels of CSF MTf (p = 0.020) compared to MCI non-converters at baseline. We suggest the diminished CSF MTf observed in MCI-converters to AD may arise from impaired transport of MTf from blood into the brain tissue/CSF and/or increased MTf export from the CSF into the blood arising from attenuated competition with reduced levels of CSF Aβ. Further investigations are required to determine the source of CSF MTf and how brain MTf is regulated by cellular barriers, Aβ and activated microglia that surround plaques in AD pathophysiology. In conclusion, low CSF MTf may identify those MCI individuals at risk of converting to AD.

Keywords: Alzheimer's disease, CSF, iron, MCI, melanotransferrin, microglia

# INTRODUCTION

fnins-13-00181 March 8, 2019 Time: 13:15 # 2

A growing body of evidence implicates iron metabolism as a contributing factor to oxidative stress and neurodegeneration in Alzheimer's disease (AD) (Meadowcroft et al., 2015; Ashraf et al., 2018). The transition metal, iron, is crucial in essential processes including DNA synthesis, myelin synthesis, neurotransmitter synthesis and metabolism in the central nervous system (Ward et al., 2014). Iron has been shown to associate with insoluble amyloid plaques (Telling et al., 2017) and neurofibrillary tangles (Yamamoto et al., 2002), characteristic hallmarks of AD. Iron produces reactive oxygen species (ROS) via the Fenton reaction, damaging macromolecules such as lipids, proteins and nucleic acids (Smith et al., 1997; Ward et al., 2014; Ashraf et al., 2018). Deferoxamine, an iron chelator, demonstrated substantial improvement in cognitive performance in AD subjects (Crapper McLachlan et al., 1991).

Melanotransferrin (MTf) or p97 belongs to the transferrin superfamily and binds to a single ferric iron with high affinity (Baker et al., 1992). It has been demonstrated to exist as a plasma membrane glycosyl phosphatidylinositol (GPI) anchored protein (Alemany et al., 1993; Kennard et al., 1995) or a soluble and actively secreted protein (Food et al., 1994; Desrosiers et al., 2003). The physiological function of both forms of MTf remain to be established. MTf is expressed by brain capillary endothelium in cognitively normal (CN) individuals, but also shown to be specifically localized in the reactive microglia associated with senile plaques in AD brains (Jefferies et al., 1996; Rothenberger et al., 1996; Yamada et al., 1999). MTf levels have been demonstrated to be increased in the serum of AD subjects compared to healthy controls, and significantly increased in cerebrospinal fluid (CSF) of AD subjects compared to individuals suffering from other CNS diseases (Kennard et al., 1996; Feldman et al., 2001; Kim et al., 2001), highlighting the potential of MTf as a possible AD biomarker. Desrosiers et al. (2003) demonstrated no differences in levels of serum MTf between AD and control subjects. However, the study was not only statistically underpowered with a small sample size, the ages and sex of subjects were not reported, and also suffered from methodological concerns. The latter include non-optimal preparation and storage of samples; lack of a calibration curve for absolute quantification and a positive control for western blot analysis (as described in Kennard et al., 1996). In contrast, other groups independently validated the use of serum MTf as a potential biomarker of AD in statistically well-powered doubleblind studies (Feldman et al., 2001; Kim et al., 2001). The aim of the present study was to determine the association of baseline CSF MTf with AD biomarkers, cognitive and neuroimaging measures using the AD Neuroimaging Initiative (ADNI) cohort. We hypothesized that increased CSF MTf levels will be associated with cognitive impairment in the ADNI cohort.

#### MATERIALS AND METHODS

A total of 287 Alzheimer's Disease Neuroimaging Initiative (ADNI) subjects comprising of 86 CN, 135 mild cognitive impaired (MCI) and 66 AD subjects were included in the present study<sup>1</sup> . Of the MCI subjects, 85 converted to AD (MCIc), while the remaining 50 MCI-nc did not in a period of 10 years, with most continuing to satisfy the criteria for MCI with the exception of four, who became CN. CN subjects had MMSE scores of ≥25 and no history of significant cognitive or physical impairments. MCI subjects had a MMSE score of ≥24, a memory complaint but preservation of cognitive and functional performance. AD cases included had an MMSE score of ≥20 and met the NINCDS/ADRDA criteria for probable AD. Detailed inclusion/exclusion criteria are available on the ADNI website<sup>2</sup> . Subjects included in the study underwent lumbar puncture and blood collection at baseline; and serial cognitive testing – AD Assessment Scale-Cognitive Subscale (ADAS-Cog13) and Rey's auditory visual learning test (RAVLT) alongside magnetic resonance imaging (MRI)-assessment of hippocampal volume. Although not all subjects had [18F]FDG-PET (18Ffluorodeoxy-glucose positron emission tomography), the data was still included to provide information about synaptic glucose metabolism. ADNI uses serial clinical and neuropsychological assessments (MRI), PET, and baseline CSF biomarkers, in combination to monitor progression of MCI subjects to AD. ADNI was approved by the institutional review board and ethics committees of participating institutions, and written informed consent was obtained from all participants or their next of kin.

## CSF Analysis

Participants underwent CSF sampling as described fully on the ADNI website<sup>2</sup> . Briefly, a small sample of CSF was collected from the lower spine by lumbar puncture in the morning following an overnight fasting. Samples were frozen on dry ice within an hour of collection, and consequently shipped on dry ice to the ADNI Biomarker Core laboratory (University of Pennsylvania Medical Center). Aliquots of 500 µl were prepared after an hour of thawing at ambient temperature and gently mixed. The aliquots were kept at −80◦C prior to analysis.

Cerebrospinal fluid Aβ1−42, total-tau (ttau) and phosphorylated-tau (181p, ptau) were measured using the multiplex xMAP Luminex platform (Shaw et al., 2011). A multiplexed mass spectrometry (MS)-based assay using multiple reaction monoring (MRM) was used to detect CSF levels of MTf and developed by Caprion Proteomics in collaboration with the ADNI Biomarker Consortium Project team. The technology, quality control and validation of the MRM platform is fully described in the "Use of Targeted Mass Spectrometry Proteomic Strategies to Identify CSF-Based Biomarkers in Alzheimer's Disease Data Primer"<sup>3</sup> . Briefly, but fully described in the primer and elsewhere (Spellman et al., 2015), CSF (100 µl) was depleted of plasma proteins using a Multiple Affinity Removal System (MARS-14) column and digested with trypsin (1:25 protease:protein ratio). Following lyophilization, samples were desalted and reconstituted with five internal standard peptides and analyzed by LC/MRM-MS on a

<sup>1</sup>http://adni.loni.usc.edu/

<sup>2</sup>http://www.adni-info.org/

<sup>3</sup> ida.loni.usc.edu

5500 QTRAP LC-MS/MS system: Q1 isolates the characteristic MTf (trypsin-digested) peptide ion (TRFM\_ADTDGGLIFR) which then undergoes collision-induced-dissociation in Q2 to produce a characteristic fragment ion measured in Q3. The signal of the fragment ion was monitored over the chromatographic elution time and used for quantification. The peptide, TRFM\_ADTDGGLIFR, had been previously synthesized and used for method development prior to analysis. Also, note that absolute quantification by an external standard in a different matrix (and fully described in the data primer mentioned above) is only an approximation. While absolute quantification is possible with spiking of known amounts of stable isotope-labeled peptides into samples, this was not done, however, data comparisons between subjects remain valid.

## Structural MRI Volumes

Subjects underwent structural T1-weighted MRI at 1.5T using a sagittal 3D-volumetric magnetization prepared rapid gradient echo (MP-RAGE) sequence (Jack et al., 2008). Briefly, the acquisition parameters were: repetition time (for the inversion pulses), 2400–3000 ms; echo time, 4 ms; inversion time, 1000 ms and 8◦ flip angle. The field of view was 240 mm × 240 mm, matrix size 192 mm × 192, and 1.2 mm thick 160–208 slices collected covering the whole brain to give a nominal resolution of 0.94 mm × 0.94 mm × 1.2 mm. Hippocampal volumes were obtained using FreeSurfer (version 4.1.0) and fully detailed elsewhere (Fischl et al., 2002, 2004). In brief, motion correction, affline transformation to Talairach image space, intensity inhomogeneity, and removal of non-brain tissues were performed. Following intensity normalization and non-linear warping of the atlas brain image to the subject image, the resultant warped atlas brain image underwent atlas-based tissue segmentation to label various brain regions including the hippocampus. Hippocampal volume was calculated by multiplying the number of voxels by the voxel volume. MRI was performed at baseline, 6 months, 1 year, then yearly for 10 years.

#### [ <sup>18</sup>F] Fluorodeoxyglucose ([18F]FDG-PET)

[ <sup>18</sup>F]FDG-PET scans were acquired on multiple scanners with various resolutions, e.g., voxel dimensions of 2.0 mm × 2.0 mm × 2.0 mm with image size, 128 × 128 × 63, at 6 months, 1, 1.5, and 2 years (Jagust et al., 2010). The scans were acquired as 6 × 5-min images, from 30 min after injection of 185 MBq (5 mCi) of [18F]-FDG (for full details<sup>4</sup> ). Each image was registered to the first image to produce a dynamic image set which was then averaged to yield a single 30-min PET image. For comparison between subjects, each baseline average PET image was reoriented along the anterior-posterior commissure line and resliced to a 1.5 mm isotropic voxel space and smoothed using a standard 14 mm full-width half-maximum kernel to produce images of a uniform resolution. Each PET image was spatially normalized to Montreal Neurological Institute brain space and the mean hippocampal FDG uptake (normalized to pons uptake) measured<sup>5</sup> .

# Neuropsychological Assessments

All subjects underwent detailed neuropsychological testing including ADAS-Cog13 and RAVLT. ADAS-Cog13 is a 13 item scale used for assessing learning, memory, language production and comprehension, constructional and ideational praxis, orientation, has number cancelation and delayed free recall tasks. The word recall test was administered first, and the word recognition task given at the end with other cognitive tasks given in between. The twoword memory tasks were separated so that the risk of individuals confusing words from the two tasks was minimized. Objective testing was followed by subjective clinical ratings of language ability and aptitude of the participant to remember test instructions. The test is scored in terms of errors and range from 0 to 70, with higher scores indicative of poor performance<sup>6</sup> .

The RAVLT tests episodic verbal memory by assessing an individual's ability to acquire a list of 15 unrelated words (all nouns) over five trials. The words are presented orally to the subject at a rate of one word per second and immediate free recall of words is elicited. The number of correctly recalled words are recorded on each trial. Following a 30-min delay filled with unrelated testing (distractor list), the subject is required to repeat the original list of 15 words. Finally, a yes/no recognition trial is administered which consists of the original 15-words and 15 randomly interspersed distractor words. The number of target "hits" as well as false positive responses are recorded. The sum of scores from the first five trials was used to compute the RAVLT score. Cognitively intact individuals attain a higher score than individuals exhibiting cognitive impairment<sup>6</sup> .

## Statistical Analysis

ANCOVA models assessed differences in CSF levels of MTf, Aβ, tau, neuropsychological tests and neuroimaging measures across diagnostic groups, with age, sex and APOEε4 status as covariates. Since age, sex and genetic status have been known to affect the dependent variables under study, ANCOVA was chosen to adjust for the variance attributed to these factors (covariates), to understand the effect of disease on the dependent variables in question. The CSF Aβ and MTf were normally distributed, while ttau and ptau were natural log-transformed to ensure normality. For regression models, we tested the conditions necessary to satisfy assumptions by checking for collinearity, normal distribution of residuals, maintenance of homoscedasticity and normality of error terms. All models satisfied these conditions. Associations between baseline cognitive scores and neuroimaging measures as well as percentage longitudinal change, i.e., (follow time measure–baseline measure)/baseline measure × 100%, with baseline MTf were performed using linear regression.

<sup>4</sup>http://www.adni-info.org/Scientists/doc/PET-Tech\_Procedures\_Manual\_v9.5. pdf

<sup>5</sup>http://adni.loni.usc.edu/methods/pet-analysis/pre-processing/ <sup>6</sup>http://www.adni-info.org



Data is presented as mean ± standard deviation. P-values are presented for ANCOVA models of CSF proteins, hippocampal volume and FDG, cognitive measures and AD biomarkers, with adjustment for age, sex and APOEε4 status, with significant p-values denoted by bold italics (a.u., arbitrary units).

Since follow-up times were different between subjects, follow-up time was included as a covariate. We then used two-tailed T-test to determine differences in baseline MTf levels between MCI converters (MCI-c, n = 85) and nonconverters (MCI-nc, n = 50) to AD. A p-value of ≤0.05 was considered significant. All analysis was performed using SPSS IBM version 22.0 and GraphPad Prism 7.0 (GraphPad Inc., San Diego, CA, United States) was used to produce figures.

#### RESULTS

The demographics of individuals based on diagnosis are shown in **Table 1**, and for MCI-nc and MCI-c in **Table 2**, while **Supplementary Table 1** document the follow-up time for CN, MCI, and AD subjects. Levels of CSF MTf were not significantly different between CN, MCI, and AD subjects (**Table 1**). However, multiple regression modeling of established AD biomarkers and MTf in the total cohort showed higher levels of CSF MTf were positively associated with hippocampal volume (R <sup>2</sup> = 38.0, β = 0.169, p = 0.001; **Table 3** and **Supplementary Figure 1**) and percentage longitudinal change in RAVLT scores (R <sup>2</sup> = 20.6, β = 0.127, p = 0.025; **Table 3** and **Supplementary Figure 2**).

Multiple regression modeling was repeated to determine the associations between established AD biomarkers, cognitive scores and MTf based on diagnosis (**Tables 4**, **5** and **Supplementary Figures 3**, **4**). The regression model demonstrated higher levels of CSF MTf were positively associated with hippocampal volume in CN (R <sup>2</sup> = 24.8, β = 0.311, p = 0.005; **Table 4** and **Supplementary Figure 3A**) TABLE 2 | Demographics of MCI subjects based on their conversion status: MCI non-converters (MCI-nc) and converters (MCI-c) to AD.


Data is presented as mean ± standard deviation. P-values presented for two-tailed T-test of CSF proteins, hippocampal volume and FDG, cognitive measures and AD biomarkers, with significant p-values denoted in bold italics (a.u., arbitrary units).

and MCI individuals (R <sup>2</sup> = 21.6, β = 0.206, p = 0.016; **Table 4** and **Supplementary Figure 3A**). Although CSF MTf was not associated with hippocampal volume in AD, MTf were associated with longitudinal hippocampal volume change (R <sup>2</sup> = 43.8, β = 0.288, p = 0.036; **Table 4** and **Supplementary Figure 3A**). Also, CSF MTf was positively correlated to baseline glucose metabolism (R <sup>2</sup> = 17.7, β = 0.426, p = 0.019; **Table 4** and **Supplementary Figure 3B**) in

TABLE 3 | Modeling the association of CSF Aβ, total tau (ttau), phosphorylated tau (ptau), and melanotransferrin (MTf) with neuroimaging measures of the hippocampus and cognitive scores in the total cohort (Supplementary Figures 1, 2).


Age, sex and diagnosis were included as covariates. The % change represents the longitudinal change in neuroimaging/cognitive measures from baseline to the follow-up period for each respective patient. Since the follow-up time points were different between subjects, we included the follow-up time as a covariate. Data is presented as mean ± standard deviation. The standardized coefficient (β) with adjusted R<sup>2</sup> and p-values are stated, with significant p-values in bold italics.

TABLE 4 | Modeling the association of CSF Aβ, total tau (ttau), phosphorylated tau (ptau), and melanotransferrin (MTf) with neuroimaging measures of the hippocampus in cognitively normal (CN), mild cognitive impairment (MCI) or Alzheimer's disease (AD) (Supplementary Figure 3).


Age and sex were included as covariates. The % change represents the longitudinal change in neuroimaging scores from baseline to the follow-up period for each respective patient. Since the follow-up time points were different between subjects, the follow-up time was also included as a covariate. Data is presented as mean ± standard deviation. The standardized coefficient (β) with adjusted R<sup>2</sup> and p-values are stated, with significant p-values in bold italics.

AD subjects. The only associations between CSF MTf and cognitive scores were observed in MCI, with MTf negatively associated with ADAS-Cog13 scores (R <sup>2</sup> = 7.00, β = −0.172, p = 0.050; **Table 5** and **Supplementary Figure 4C**) and positively associated with longitudinal change in RAVLT scores (R <sup>2</sup> = 17.6, β = 0.248, p = 0.003; **Table 5** and **Supplementary Figure 4C**).

Interestingly, MCI subjects that converted to AD (MCIc) demonstrated significantly lower levels of baseline CSF MTf compared to those that did not (MCI-nc; p = 0.020; **Table 2** and **Figure 1**). MCI-c had significantly decreased Aβ (p = 5 × 10−<sup>4</sup> ; **Table 2** and **Figure 1**) and increased ptau compared to MCI-nc subjects (p = 0.028; **Table 2** and **Figure 1**). Levels of ttau were similar between MCI-c and MCI-nc (p = 0.0984; **Table 2** and **Figure 1**). Multiple regression modeling was also performed, including established AD biomarkers and CSF MTf according to the conversion status of MCI subjects (**Table 6** and **Supplementary Figures 5, 6**), where CSF MTf was found to be positively correlated to percentage longitudinal change in RAVLT score in MCI-nc but not MCI-c (**Table 6** and **Supplementary Figure 6C**).



Age and sex were included as covariates. The % change represents the longitudinal change in cognitive scores from baseline to the follow-up period for each respective patient. Since the follow-up time points were different between subjects, the follow-up time was also included as a covariate. Data is presented as mean ± standard deviation. The standardized coefficient (β) with adjusted R<sup>2</sup> and p-values are stated, with significant p-values in bold italics.

#### DISCUSSION

We demonstrate diminished levels of baseline CSF MTf are associated with lower hippocampal volumes in CN and MCI and worse cognitive scores in MCI subjects. Moreover, significantly lower levels of baseline CSF MTf were observed in MCI subjects converting to AD compared to non-converters, underscoring the possibility of CSF MTf to identify those individuals with increased susceptibility of converting to AD. In AD, lower CSF MTf levels was associated with a reduction in hippocampal volume over time and appear to reflect disease progression.

We found similar levels of CSF MTf in CN, MCI, and AD subjects. In contrast, a previous study demonstrated increased levels of CSF MTf in AD compared to individuals suffering from various neurodegenerative diseases (Kennard et al., 1996). However, our study cohort was a greater size, comprised mixed ethnicity and age-matched CN subjects, while the Kennard et al. study had only a Japanese cohort and the AD group older than the "control" group of individuals with non-AD neurodegenerative disease. With less genetic variability compared to our mixed ethnicity cohort, their results may not be representative of the general population. To the best of our knowledge, our study is the first report addressing the levels of CSF MTf levels in AD compared to CN subjects.

While elevated serum MTf have been found in AD (Kennard et al., 1996; Kim et al., 2001), others have observed this in early milder AD cases but no further increases in latter stages of AD (Kim et al., 2001). It is noted that serum MTf of varying glycosylated conformations were unchanged in another, but rather poor (see "Introduction"), study on AD subjects (Desrosiers et al., 2003). Desrosiers and co-workers used two-dimensional gel electrophoresis, whereas other, but validated studies, utilized sandwich fluorescent assay (Kennard et al., 1996), radioimmunoassay (Feldman et al., 2001), and dot-immunoblot assay (Kim et al., 2001) to measure serum MTf levels. The present study suffers from the limitation that plasma/serum levels of MTf were not available in the ADNI repository.

It is well documented that MCI subjects are at an increased risk of acquiring AD, and around 10–15% of these subjects convert to AD on a yearly basis (Risacher et al., 2009). By stratifying MCI individuals to MCI-c and MCI-nc, we found significantly diminished baseline CSF MTf levels in the former group. Indeed, lower CSF MTf were associated with greater cognitive deficits (ADAS-Cog13) in MCI and lower hippocampal volumes in both CN and MCI. While not associated with hippocampal volume, CSF MTf was associated with longitudinal hippocampal volume change in AD. MTf appears to have a role in mediating cellular iron uptake (Kennard et al., 1995), and hence likely to be involved in iron metabolism.

non-converters (MCI-nc) and converters (MCI-c) with p < 0.05 being considered significant.

Perturbed MTf expression may contribute to iron dysregulation and cellular iron accumulation, precipitating oxidative stress as iron is a potent source of free radicals, hastening AD pathogenesis. Greater iron content in the subcortical areas has been associated with poorer memory performance, lower general cognitive aptitude, mental retardation, and poorer cognitive and motor control in a healthy population (Sullivan et al., 2009; Penke et al., 2012; Rodrigue et al., 2013; Adamo et al., 2014; Daugherty and Raz, 2015). Furthermore, higher hippocampal iron has been correlated to smaller hippocampal volume, which in turn predicted poorer episodic memory (Rodrigue et al., 2013). Additionally, iron overload has been shown to accelerate cognitive impairment in human and transgenic mouse models of AD (Rodrigue et al., 2013; Becerril-Ortega et al., 2014). Interestingly, MCI subjects exhibited increased iron levels in the cortex and cerebellum (Smith et al., 2010). Another study reported an increase in the redox-active iron in the CSF of MCI but not AD cases, with levels correlating with the extent of cognitive impairment (Lavados et al., 2008). Consistent with these reports, we demonstrated a decrease in the baseline CSF MTf levels of MCI-c compared to MCI-nc, but no differences between CN, MCI, and AD groups, which suggests that iron dysregulation is an early event in AD pathogenesis (van Bergen et al., 2016).

There are two forms of MTf, one form is located on the cell surface via a GPI anchor on the plasma membrane, and the other is a soluble form that is secreted and found in the serum (Food et al., 1994; Desrosiers et al., 2003). Soluble MTf was originally thought to be derived from improper processing resulting in the protein evading the GPI-addition or endogenous (phosphatidylinositol-specific phospholipase D) cleavage of GPIanchored MTf (McNagny et al., 1996). Moreover, soluble MTf was proposed to be generated from an alternatively spliced mRNA transcript lacking a GPI signal coding sequence (McNagny et al., 1996). However, more recently, Yang et al. (2004) conducted a detailed study in which deletion of the GPI pre-anchor sequence in human p97 led to a soluble form of MTf, and proposed more convincing mechanisms that could account for the soluble forms of GPI proteins. Apparently, three critical recognition sites are needed for the processing of GPI proteins in the endoplasmic


TABLE 6 | Association of CSF Aβ, total tau (ttau), phosphorylated tau (ptau), and melanotransferrin (MTf) with hippocampal neuroimaging measures and cognitive scores in MCI non-converters (MCI-nc) and converters (MCI-c) to Alzheimer's Disease (Supplementary Figures 5, 6).

Age and sex were included as covariates. The % change represents the longitudinal change in neuroimaging/cognitive measures from baseline to the follow-up period for each respective patient. Since the follow-up time points were different between subjects, the follow-up time was also included as a covariate. Data is presented as mean ± standard deviation. The standardized coefficient (β) with adjusted R<sup>2</sup> and p-values are stated, with significant p-values in bold italics.

reticulum: a transamidase, the residues to which the GPI anchor is linked and a carboxyl terminal signal peptide. Disruption in any of these could potentially alter MTf processing to result in soluble MTf, without a GPI anchor (Alemany et al., 1993; Maxwell et al., 1995; Yang et al., 2004). MTf in the blood has been known to undergo a high rate of transcytosis across the blood-brainbarrier (BBB) from the bloodstream to the brain (Demeule et al., 2002). We postulate that impaired transcytosis of MTf from the bloodstream into the brain may in part account for the decrease in CSF MTf levels in MCI-c and consistent with reports of increased serum MTf in AD (Kennard et al., 1996; Feldman et al., 2001; Kim et al., 2001). It is important to mention that MTf is not exclusively located in the brain but is also found in the liver and intestinal epithelial cells (Sciot et al., 1989; Alemany et al., 1993).

The low-density lipoprotein receptor (LRP) has been identified as a receptor for MTf and appears to actively transport MTf from the blood across the BBB into the brain (Demeule et al., 2002). Genetic studies strongly implicate the LRP gene locus in enhanced susceptibility to AD with APOE and Aβ being key LRP ligands (Kounnas et al., 1995; Kang et al., 1997). Furthermore, LRP levels are lower in AD, and of the two isoforms, LRP1 and LRP2, higher levels of the former have been associated with later onset of disease in AD patients, suggesting LRP1 may be protective against AD (Kang et al., 2000). At the BBB, LRP1 has been shown to be essential for the elimination of Aβ from the brain into the blood (Storck et al., 2016), with escalating Aβ levels in the brain associated with reduced LRP1 expression (Shibata et al., 2000). This reduction in LRP at the BBB may explain the lack of transcytosis of MTf from the blood into the brain and so, the low baseline CSF levels in MCI-c. However, it has been hypothesized that the brain capillary endothelial cells themselves produce MTf and that there is transcytosis in both directions at the BBB (Rothenberger et al., 1996). Thus, further investigations are needed to understand MTf import and export through the BBB and possible interactions with Aβ transport in aging and AD.

Neuroimaging and post-mortem studies have implicated BBB dysfunction as an early and common occurrence in AD, characterized by microbleeds, impaired glucose transport, disrupted functioning of P-glycoprotein 1, perivascular deposits of blood-derived proteins, cellular infiltration and degeneration of endothelial cells (Sweeney et al., 2018). Since the function of

the BBB is to strictly regulate the blood-to-brain and brain-toblood transport of solutes, MTf may simply leak through the disrupted BBB into the general circulation, leading to diminished levels of CSF MTf in the MCI-c at baseline. As the BBB is also known to be disrupted in AD, CSF MTf will also be expected to be reduced in AD, but this was not observed.

Additionally, the lower MTf levels in the CSF of MCI-c may, in part, arise from increased export of MTf from the CSF into the blood. LRP1 has also been detected at the choroid plexus (Pascale et al., 2011; Spuch et al., 2012) and suggests that MTf may be exported from the CSF to the blood via this route. Aβ (1–40) has been shown to be actively eliminated from the CSF, thought to be partly via LRP1 at the choroid plexus (Fujiyoshi et al., 2011). We propose that the lower CSF MTf in MCI-c in our study may result from its increased export from the CSF to the blood due to concomitant attenuated competition with the diminishing CSF Aβ levels arising from impaired Aβ clearance from the brain parenchyma. Future experiments are required to perform detailed investigations of whether other LRP substrates can influence the transcytosis of MTf, and the mechanisms governing MTf transport from the CSF to the blood and vice versa.

If CSF MTf is determined by Aβ clearance from the parenchyma into the CSF, this would imply that CSF MTf would be reduced in AD as levels of CSF Aβ are significantly reduced. A reduction in CSF MTf in AD was not observed in our study, but this may be explained by the increased production of MTf in the brain parenchyma from a subset of reactive microglia associated with amyloid plaques (Jefferies et al., 1996; Yamada et al., 1999). The subset of reactive (dysregulated) microglia surrounding plaques appear to be an exclusive hallmark of AD pathology and not observed in other neurodegenerative diseases (Jefferies et al., 1996; Rothenberger et al., 1996; Yamada et al., 1999). Whatever the mechanism(s) for reduced CSF MTf levels in MCI-c (described above), they are likely to be operational in AD as well and ought to lead to reduced CSF MTf in AD, we propose this is not observed as levels are maintained by MTf production from these reactive microglia surrounding plaques. The microglia appear to be laden with iron, as evident by high expression of the iron storage protein, ferritin (McGeer et al., 1987; Grundke-Iqbal et al., 1990; Food et al., 1994). We speculate that increased iron uptake by the reactive microglia via expression of GPI-anchored MTf leads to cellular iron accumulation, further exacerbating microglial dysfunction. Chinese hamster ovary (CHO) cells defective in the transferrin receptor but transfected to express MTf, showed doubling of iron intake. Iron-associated microglial-driven neuroinflammation may be a significant driver behind neuronal and synaptic destruction, working synergistically with Aβ (Gallagher et al., 2012; McGeer and McGeer, 2013). Determining the source of CSF MTf may aid determination of the defect that is contributing to AD pathogenesis.

Based on our results, CSF MTf levels appears to be significantly decreased in MCI-c compared to MCI-nc at baseline. However, no significant changes were found in MTf levels at the AD stage. MTf may be involved in iron metabolism, iron dyshomeostasis may be an early event in disease pathogenesis. Indeed, redoxactive CSF iron levels were shown to be increased from normal to MCI subjects, while in AD, there was an abrupt decrease in iron levels close to zero (Lavados et al., 2008). Through histochemical iron analysis, increased brain iron content was reported in MCI and preclinical cases of AD (Smith et al., 2010). It is likely that perturbations in baseline MTf levels and resultant dysfunction at the stage of MCI, may be a significant and early contribution to the disease process prior to acquiring AD.

# CONCLUSION

In conclusion, we demonstrate that baseline CSF MTf levels are significantly decreased in MCI-c compared to MCI-nc. However, our results remain to be validated in an independent cohort. Future directions would be required to elucidate the role of MTf in the context of AD, especially determining the source of MTf in the CSF and how brain MTf is regulated by cellular barriers, Aβ and activated microglial cells in human and transgenic AD models, requiring measurements in paired blood-CSF samples. Nevertheless, our study implies that baseline CSF MTf levels may be a useful marker to identify individuals with increased risk of conversion to AD, although much development still needs to be undertaken to ensure robust assay reproducibility across multiple clinical laboratories.

# DATA AVAILABILITY

Publicly available datasets were analyzed in this study. This data can be found here: http://adni.loni.usc.edu/.

# AUTHOR CONTRIBUTIONS

AA and P-WS contributed the concept and design of the study, wrote the first draft of the manuscript, and revised and approved the submitted version. AA performed the statistical analysis, aided by JA, MA, and CH, with interpretation performed by AA and P-WS.

# FUNDING

Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI was funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer's Association; Alzheimer's Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd. and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research and Development LLC.;

Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California. This paper represents independent research part funded by the National Institute for Health Research (NIHR)

#### REFERENCES


Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care. The authors would like to thank Biotechnology and Biological Sciences Research Council (BBSRC), King's College London and Perspectum Diagnostics Ltd., for funding AA's industrial Ph.D. studentship.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.00181/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 © 2019 Ashraf, Alepuz Guillen, Aljuhani, Hubens and So. 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.

# NCOA4-Mediated Ferritinophagy: A Potential Link to Neurodegeneration

Maria Quiles del Rey and Joseph D. Mancias\*

Division of Genomic Stability and DNA Repair, Department of Radiation Oncology, Dana-Farber Cancer Institute, Boston, MA, United States

NCOA4 (Nuclear receptor coactivator 4) mediates the selective autophagic degradation of ferritin, the cellular cytosolic iron storage complex, thereby playing a critical role in intracellular and systemic iron homeostasis. Disruptions in iron homeostasis and autophagy are observed in several neurodegenerative disorders raising the possibility that NCOA4-mediated ferritinophagy links these two observations and may underlie, in part, the pathophysiology of neurodegeneration. Here, we review the available evidence detailing the molecular mechanisms of NCOA4-mediated ferritinophagy and recent studies examining its role in systemic iron homeostasis and erythropoiesis. We propose additional studies to examine the potential role of NCOA4 in the brain in the context of neurodegenerative diseases.

#### Edited by:

Federico Benetti, Scuola Internazionale Superiore di Studi Avanzati (SISSA), Italy

#### Reviewed by:

Liliana Quintanar, Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Nacional (CINVESTAV), Mexico Raffaella Gozzelino, Centro de Estudos de Doenças Crónicas (CEDOC), Portugal

#### \*Correspondence:

Joseph D. Mancias joseph\_mancias@dfci.harvard.edu

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 22 November 2018 Accepted: 28 February 2019 Published: 14 March 2019

#### Citation:

Quiles del Rey M and Mancias JD (2019) NCOA4-Mediated Ferritinophagy: A Potential Link to Neurodegeneration. Front. Neurosci. 13:238. doi: 10.3389/fnins.2019.00238 Keywords: NCOA4, ferritinophagy, iron homeostasis, ferroptosis, neurodegenerative disorders, neurodegenerative disease, neurodegeneration

# INTRODUCTION

Iron is an indispensable element for nearly all living organisms as it takes part in a wide range of biological reactions such as oxygen transport, oxidative phosphorylation, and enzymatic reactions required for cellular proliferation (Pantopoulos et al., 2012; Bogdan et al., 2016; Muckenthaler et al., 2017). While iron is critical for these reactions, iron availability must be tightly regulated as high levels of free iron can generate deleterious reactive oxygen species via the Haber-Weiss/Fenton reactions (Dixon and Stockwell, 2014). Many neurodegenerative disorders (ND), including Alzheimer's Disease (AD), Parkinson's Disease (PD), Huntington's Disease (HD), and Amyotrophic Lateral Sclerosis (ALS) are associated with elevated free iron, dysfunctional iron homeostasis in general, and resultant oxidative stress that can precipitate ferroptosis, an irondependent form of cell death (Ward et al., 2014; Kim et al., 2015). Likewise, a group of inherited neurological disorders known as Neurodegeneration with Brain Iron Accumulation (NBIA) are characterized by inappropriate iron deposits associated with oxidative stress and neuronal cell death (Kruer, 2013).

To balance cellular requirements for iron with the potential deleterious effects of free iron, cells possess an intricate system of proteins involved in intracellular iron homeostasis (Pantopoulos et al., 2012). Critically, cellular iron is stored in a non-toxic form in ferritin, which is a cytosolic heteropolymer made of 24 subunits of ferritin heavy and light chains (FTH1/FTL) that can store up to 4500 iron atoms (Arosio et al., 2009). Iron is stored in ferritin during times of iron excess and must be released under periods of iron demand. The predominant pathway for iron release from ferritin is via Nuclear receptor coactivator 4 (NCOA4)-mediated selective

autophagy whereby NCOA4 binds ferritin to traffic it to the lysosome where ferritin is degraded and iron is released for use by the cell (Mancias et al., 2014; **Figure 1**).

Maintaining iron balance in the cell is a complex and highly regulated process, which, if impaired may lead to cell death. Recent studies show that NCOA4 plays a central role in maintaining this delicate balance on a cellular and systemic level (Mancias et al., 2014, 2015; Bellelli et al., 2016). Dysregulation of NCOA4-mediated ferritinophagy disrupts systemic iron homeostasis with deleterious effects on erythropoiesis and regulation of oxidative stress. Furthermore, recent studies show that NCOA4-mediated ferritinophagy modulates susceptibility to ferroptosis (Gao et al., 2016; Hou et al., 2016). As many NDs are associated with defective iron metabolism and autophagy, NCOA4 may play an important role in the onset and development of these disorders. Here, we review the literature to delineate new hypotheses to elucidate the potential significance of NCOA4 in ND.

#### NCOA4 MEDIATES FERRITINOPHAGY AND REGULATES INTRACELLULAR BIOAVAILABLE IRON

We recently identified NCOA4 as the selective autophagy receptor mediating delivery of iron-laden ferritin to the lysosome for degradation and release of iron, a process termed ferritinophagy (Mancias et al., 2014). Using a combination of autophagosome enrichment and quantitative mass spectrometrybased proteomics, NCOA4 was identified as one of the most robustly enriched proteins in autophagosomes. Affinity purification-mass spectrometry identified ferritin subunits (FTL, FTH1) and the HERC2 E3 ubiquitin ligase as NCOA4 interacting partners. In vitro assays with recombinant FTH1 and NCOA4 demonstrated that NCOA4 binds to ferritin via a direct interaction of a conserved surface arginine (R23) on FTH1 and a C-terminal domain in NCOA4 (only present in NCOA4α) (Mancias et al., 2015). Gryzik et al. (2017) further revealed that FTH1 can bind up to 24 NCOA4 fragments in in vitro binding assays. NCOA4 colocalized with ferritin in autophagosomes and lysosomes. Importantly, NCOA4 depletion inhibited delivery of ferritin to the lysosome and thereby ferritin degradation (Mancias et al., 2014) implicating NCOA4 in the selective autophagic degradation of ferritin. The decrease in ferritin turnover due to NCOA4 loss leads to a decrease in bioavailable cellular iron highlighting the importance of NCOA4 in regulating intracellular iron homeostasis.

Flux through the ferritinophagy pathway is determined by NCOA4 levels which are in turn regulated by iron levels in the cell (Mancias et al., 2015). When cellular iron levels are high, NCOA4 interacts with the HERC2 ubiquitin E3 ligase in an iron-dependent manner and is targeted for degradation via the ubiquitin proteasome system. This ensures NCOA4 levels are low under high iron conditions thereby decreasing ferritinophagy and favoring ferritin iron storage. Conversely, under low cellular iron conditions, the NCOA4 and HERC2 interaction is decreased leading to elevated NCOA4 levels and increased ferritinophagy flux to replenish cellular iron levels (Mancias et al., 2015).

The mechanism of NCOA4-ferritin trafficking to the lysosome is unclear with evidence for both classical ATG8-dependent autophagic delivery to the lysosome and a non-classical ESCRTmediated delivery system involving TAX1BP1, VPS34, ATG9A, and ULK1/2-FIP200 (Mancias et al., 2014; Goodwin et al., 2017). Further research is necessary to understand the mechanism(s) that direct NCOA4 to the lysosome. Alternatively, proteasomal degradation of ferritin has been reported in the context of Ferroportin-mediated iron export (De Domenico et al., 2006). While the multiple reported mechanisms of ferritin degradation may suggest context-dependent modes of ferritin turnover; NCOA4-mediated mechanisms appear to be the predominant means of ferritin degradation.

While NCOA4-mediated ferritinophagy appears to be a conserved function of NCOA4, initial studies of NCOA4 suggested it acts as a nuclear receptor coactivator, albeit a weak one (Alen et al., 1999; Gao et al., 1999; Monaco et al., 2001; Hu et al., 2004; van de Wijngaart et al., 2006). More recently, Lodish and colleagues reported that NCOA4 is stimulated by thyroid hormone promoting its recruitment to chromatin regions associated with transcripts abundant during erythropoiesis (Gao et al., 2017). NCOA4 has also been reported to act as a regulator of DNA replication origin activation via an interaction with the minichromosome maintenance 7 protein (MCM7) (Bellelli et al., 2014). Further work is required to understand the different reported functions of NCOA4 and whether there are context dependent variations in NCOA4 function.

#### PHYSIOLOGICAL ROLE OF NCOA4

Given NCOA4 is a critical component of the intracellular iron homeostasis machinery, understanding its role in modulating systemic iron homeostasis has been of paramount importance. The physiological role of NCOA4-mediated ferritinophagy has been studied in vivo initially in the context of erythropoiesis, a highly iron-dependent process. Early studies to understand the relationship between NCOA4 and erythropoiesis showed that NCOA4 mRNA levels were upregulated at erythropoietic sites in zebrafish (Weber et al., 2005). NCOA4 ablation in a zebrafish model leads to defects in globin synthesis and hemoglobinization (Mancias et al., 2015). Dowdle et al. (2014) revealed that NCOA4 is necessary for ferritin turnover and iron homeostasis in vivo as NCOA4-deficient mice showed accumulation of ferritin and increased iron deposits associated with splenic macrophages. A systemic NCOA4 knockout mouse model show that loss of NCOA4 leads to accumulation of ferritin in all organs examined. Mice developed iron accumulation in the liver and spleen and signs of iron overload in the serum. While all signs pointed to an iron overload phenotype, these mice developed a mild microcytic hypochromic anemia suggesting an inability to properly utilize the available iron. This anemia was exacerbated by a low iron diet with mice unable to mobilize iron stores from ferritin. An additional study in a systemic NCOA4

NCOA4 is unable to target ferritin for lysosome degradation leading to accumulation of ferritin and an iron overload phenotype (based on in vivo studies of NCOA4 knockout). Excess free iron is associated with reactive oxygen species generation and oxidative stress and may contribute to neurodegeneration. Further experimentation is necessary to link specific defects in the NCOA4-mediated ferritinophagy pathway to neurodegeneration.

knockout mouse model similarly showed anemia in experimental mice that was more pronounced in the immediately post-natal period suggesting a differential temporal dependence of NCOA4 function in erythropoiesis or more generally in iron mobilization from ferritin (Gao et al., 2017). Given the systemic nature of the NCOA4 knockout with associated global dysregulation in iron homeostasis in both of these mouse models, the relative erythroid cell autonomous and non-autonomous contributions toward the anemia phenotype remained unclear until recently. Multiple ex vivo cellular models of erythropoiesis suggested a cell autonomous role for NCOA4 in erythropoiesis (Mancias et al., 2015; Ryu et al., 2017). A recent study from our group using a conditional knockout mouse model of NCOA4 has clarified the cell autonomous role for NCOA4 in erythropoiesis (Santana-Codina et al., 2019). NCOA4fl/fl ; EpoR-Cre mice with erythroid specific deletion of NCOA4 developed a pronounced anemia in the postnatal period and a mild hypochromic microcytic anemia at adult stages. Notably, the anemia at both ages in NCOA4fl/fl ; EpoR-Cre mice was less severe than that of mice with systemic NCOA4 knockout suggesting both cell autonomous and nonautonomous roles for NCOA4 in supporting basal erythropoiesis. While these in vivo studies did not perform in-depth analysis of the consequences of NCOA4 depletion in the brain, acute systemic depletion of NCOA4 using an inducible NCOA4fl/fl ; UBC-cre/ERT2 murine model did reveal accumulation of FTH1 over the time span of a week in the brain suggesting a basal level of dependence on NCOA4 in the brain for FTH1 turnover (Santana-Codina et al., 2019).

An additional finding of interest in NCOA4 null mice was an increase in serum ferritin levels (Bellelli et al., 2016). Ferritin has been reported to be secreted via a non-canonical lysosomal secretory pathway (Cohen et al., 2010). Despite the role of NCOA4 in mediating lysosomal delivery, it appears in NCOA4-deficient mice, ferritin is still secreted via a lysosomal pathway or an alternative one such as the multivesicular bodyexosome pathway (Truman-Rosentsvit et al., 2018). This finding of elevated ferritin secretion in the setting of NCOA4 loss was corroborated by a cell culture study showing that loss of NCOA4 accelerated ferritin exocytosis (Kimura et al., 2017). This may have implications in the central nervous system (CNS) where ferritin has been identified in the cerebrospinal fluid and may act as an iron carrier (Todorich et al., 2008).

# NCOA4-MEDIATED FERRITINOPHAGY AND FERROPTOSIS

In determining the relevance of NCOA4-mediated ferritinophagy beyond erythropoiesis, recent studies have identified a link between NCOA4 and a newly discovered form of iron-dependent cell death, ferroptosis. Ferroptosis is an iron-dependent form of cell death characterized by lipid peroxidation accumulation (Dixon et al., 2012; Stockwell et al., 2017). Ferroptosis can be triggered by multiple mechanisms including iron accumulation leading to reactive oxygen species resulting in lipid peroxidation or via inactivation of the lipid repairing peroxidase, GPx4. Given NCOA4-mediated ferritinophagy modulates intracellular iron levels, ferroptosis sensitivity has recently been shown to be affected by NCOA4 levels (Gao et al., 2016; Hou et al., 2016; Tang et al., 2018). In cell culture models, NCOA4 depletion decreases intracellular bioavailable iron and thereby decreases sensitivity to ferroptosis-inducing insults. On the other hand, NCOA4 over-expression in cellular models increases ferritinophagy flux and thereby increases sensitivity to ferroptosis. Interestingly, in NCOA4 knockout mice, consistent with their iron overload phenotype and elevated basal levels of tissue iron, knockout mice fed an iron-enriched diet developed early liver toxicity and premature death (Bellelli et al., 2016). This was associated with elevated levels of oxidative stress as indicated by increased Gpx and superoxide dismutase (SOD) levels. This suggests that in the context of long-term NCOA4 depletion in vivo, sensitivity to ferroptosis may actually be enhanced. This observation may have implications for NDs that have been shown to be associated with defects in autophagy, excess iron accumulation, oxidative stress, and ferroptosis.

## NEURODEGENERATIVE DISORDERS AND IRON

While there is no direct evidence for dysregulation of NCOA4 mediated ferritinophagy as a cause for ND, there are multiple lines of correlative evidence related to defective iron metabolism, autophagy, and ferroptosis that suggest a link. Iron is critical for normal physiologic function in the brain with roles in such processes as DNA synthesis, oxygen transport, mitochondrial respiration, and neurotransmitter synthesis (Ward et al., 2014). Iron accumulation is a normal process in the aging brain; however, the mechanism by which this occurs is unknown (for in-depth review of normal iron homeostasis mechanisms in the brain, we refer the reader to: (Ward et al., 2014; Garza-Lombó et al., 2018). Healthy brains accumulate iron in cytosolic ferritin and in iron bound to neuromelanin found in lysosomes (Zecca et al., 2001). Iron normally accumulates in multiple regions of the brain including the globus pallidus, putamen, substantia nigra, caudate nucleus, and red nucleus (Zecca et al., 2004; Angelova and Brown, 2015; Biasiotto et al., 2016). Interestingly, these areas of the brain are the most susceptible to ND. The cells where elevated levels of iron are found are the oligodendrocytes with varying levels found in microglia and smaller amounts of iron in neurons and astrocytes (Rouault, 2013). While iron accumulation is normal with aging, inappropriate iron accumulation and dysregulation of iron homeostasis in combination with alterations in ferritin levels are associated with ND and worse outcomes amongst ND patients. However, it remains unclear whether this is a primary or secondary phenomenon in ND. One potential primary cause of ND related to inappropriate iron accumulation is ferroptotic cellular death as recently demonstrated in models of ND. The first evidence for a link between ND and ferroptosis came from in vitro rat organotypic hippocampal slice cultures treated with glutamate to mimic the pathophysiology of stroke and ND (Dixon et al., 2012). Cells were found to undergo ferroptosis that could be prevented by iron chelators and ferrostatin (Dixon et al., 2012). Furthermore, studies in multiple models of ND with genetic (Chen et al., 2015) or pharmacological inhibitors of ferroptosis (Skouta et al., 2014; Do Van et al., 2016) showed that inhibition of ferroptosis decreased neuronal cell death pointing to a potential future therapeutic strategy.

In addition to dysfunction in iron homeostasis as a potential cause of ND, alterations in protein and organelle quality control are associated with ND. While neuronal autophagy activity levels decrease in the normal aging brain, abnormal decreases in autophagy and genetic perturbations in the autophagy-lysosome pathway are associated with ND. Indeed, genetic mouse models of neuronal autophagy deficiency develop ND (Hara et al., 2006; Komatsu et al., 2006). One hypothesis links the changes in iron homeostasis seen in both normal aging and ND with decreased levels of autophagy. In support of this, there are a number of ND diseases with genetic alterations in autophagy machinery that are associated with inappropriate iron accumulation (for an exhaustive review, see Biasiotto et al., 2016). Direct links between defective autophagy, iron accumulation, and ND are illustrated by a newly identified form of Neurodegeneration with Brain Iron Accumulation (NBIA: a group of genetic neurological disorders with abnormal accumulation of iron in the basal ganglia), static encephalopathy of childhood with neurodegeneration in adulthood (SENDA), due to a mutation in WDR45 (Saitsu et al., 2013). WDR45 encodes WD-repeat domain 45 protein, also known as WIPI4, which has a role in autophagy initiation. Pathogenic mutations in WDR45 seen in SENDA patients lead to cellular defects in autophagy, diminished lysosomal function, iron accumulation, and oxidative stress (Seibler et al., 2018). Accordingly, patients presenting with these deleterious mutations have been found to have large iron depositions in the globus pallidus and substantia nigra (Saitsu et al., 2013). Interestingly, in ex vivo models of WDR45 loss of function, pharmacologic stimulation of the autophagy pathway with Torin1 leads to normalization of iron levels (Seibler et al., 2018). The precise molecular mechanistic details linking alterations in autophagy function to changes in iron levels have not been determined in this disease or in general in ND. However, given NCOA4 mediated ferritinophagy links iron homeostasis and autophagy, it is tempting to hypothesize that defects in ferritinophagy may underlie inappropriate accumulation of iron in ND. We next review the role of altered iron metabolism and autophagy in the two main neurodegenerative disorders, Alzheimer's Disease and Parkinson's Disease.

Alzheimer's disease is the most common cause of dementia or cognitive impairment in patients older than 65 years old and is the sixth leading cause of death in the United States (Atri, 2019). AD is characterized by accumulation of intracellular and extracellular aggregates of β-amyloid (Aβ) as well as tau protein leading to neuronal dysfunction and neurodegeneration. Interestingly, high concentrations of redox-active metals, including iron, are present in these aggregates. Excess iron in these aggregates may lead to deleterious iron loss in neurons affecting neuronal function (Roberts et al., 2012). Furthermore, altered iron homeostasis has been linked to the initial misfolding of tau and β-amyloid. Specifically, amyloid precursor protein (APP) has a putative iron response element (IRE) in its 5<sup>0</sup> untranslated region under the control of the IRP1/2 post-transcriptional regulatory system (Rogers et al., 2002). Iron excess may decrease IRP1/2 binding on this IRE thereby increasing APP translation making it available for β-amyloid production. A hallmark of AD is oxidative stressinduced damage stemming from a number of factors but likely accelerated by inappropriate accumulation of redoxactive metals, including iron. Indeed, reducing levels of iron/redox-active metals has recently become a clinical target in neurodegenerative diseases in general and in AD specifically (Nuñez and Chana-Cuevas, 2018). Recent phase II clinical trials of chelation therapy for AD patients revealed promising stabilization or improvement in AD symptoms and decreases in Aβ in plasma or cerebrospinal fluid (Ritchie et al., 2003; Lannfelt et al., 2008). While inappropriate accumulation of iron is clearly associated with AD in animal models of the disease and in human patients as detected by MRI (Schenck, 2003), the relationship between altered iron homeostasis and the pathogenesis of AD is unclear. Whether increased iron is a cause or a secondary effect of the pathogenic amyloid aggregates remains unclear.

Parkinson's disease is the second most common ND and is characterized by selective loss of dopaminergic neurons in the substantia nigra initially leading to motor dysfunction. Similar to AD, PD is also characterized by accumulation of aggregated alpha-synuclein in Lewy bodies, mitochondrial dysfunction, and oxidative stress (Ward et al., 2014). PD patients also demonstrate increases in total iron concentration in the substantia nigra. For instance, increased DMT1 expression (Salazar et al., 2008) and activating mutations in Transferrin both lead to an increase in cytosolic iron and are correlated with worse outcomes for PD patients (Rhodes et al., 2014). Interestingly, an increase in redox-active iron may also be related to a decrease in ferritin synthesis in PD patients (Faucheux et al., 2002). In general PD patients with the most severe neuronal loss have the highest levels of redox-active iron (Faucheux et al., 2003). As redoxactive iron is clearly associated with generation of reactive oxygen species, inappropriate iron accumulation may be central to PD pathogenesis. However, as for AD, it is unclear whether this is a cause of a symptom of PD pathogenesis.

NCOA4 expression and function in the brain and its role in ND is an unexplored field of study. There is evidence of NCOA4 expression in the murine and rat brain at both mRNA and protein levels (Kollara and Brown, 2010). However, there are no studies of NCOA4 expression levels or function during aging or in pathological specimens from ND patients. Furthermore, there are no known pathogenic mutations in NCOA4 associated with ND. Presumably, with a decrease in global autophagy as is observed in some ND, ferritinophagy flux would also be decreased (**Figure 1**). As described above, the outcome of decreased NCOA4 levels and ferritinophagy flux as it relates to iron status in the cell varies depending on the system examined. In vitro cellular models of NCOA4 depletion show a decrease in bioavailable iron due to a decrease in ferritin degradation resulting in trapped iron (Mancias et al., 2014). This leads to decreases in bioavailable iron, lower levels of oxidative stress, and decreased sensitivity to ferroptosis. This would suggest that NCOA4 depletion in the brain in the context of ND could be protective. However, in the setting of constitutive systemic NCOA4 depletion in animal models, animals show signs of iron overload and increased sensitivity to oxidative stress (Bellelli et al., 2016). These discordant results may be a consequence of a number of differences between in vitro versus in vivo systems, including the time scale of NCOA4 loss, relative dependence on NCOA4-mediated ferritinophagy for iron balance in cell culture models (most of which are tumor-derived cell lines) versus normal tissues, and differences in the iron availability between in vivo growth conditions and cell culture models which are predominantly grown in transferrin-rich serum-containing media. The results from in vivo models of NCOA4 depletion showing iron overload and increased sensitivity to oxidative stress could align with prior findings of decreased autophagy, increased iron accumulation, and oxidative stress demonstrated in ND (**Figure 1**). However, additional studies are required to assess the expression, regulation, and function of NCOA4 in the brain under physiologic and ND conditions. We speculate that long-term NCOA4 depletion in the brain in the context of animal models of ND would worsen the ND phenotype due to a further inappropriate accumulation of free iron and resulting oxidative stress as has been previously demonstrated in systemic longterm depletion of NCOA4 (Bellelli et al., 2016). Newly available conditional NCOA4 knockout models with targeted deletion of NCOA4 in the brain will be instrumental in understanding the potential role of NCOA4 function in the brain and in ND specifically.

#### AUTHOR CONTRIBUTIONS

MQ and JDM performed the literature searches and wrote and edited the manuscript.

# FUNDING

This review was supported by a Burroughs Wellcome Fund Career Award for Medical Scientists, Brigham and Women's Hospital MFCD Award, Sidney Kimmel Foundation Kimmel Scholar Program, and Dana-Farber Cancer Institute Claudia Adams Barr Program for Innovative Cancer Research Award to JDM.

# REFERENCES



**Conflict of Interest Statement:** JDM is an inventor on a patent pertaining to the autophagic control of iron metabolism.

The remaining 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 © 2019 Quiles del Rey and Mancias. 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.

# Iron Metabolism of the Skeletal Muscle and Neurodegeneration

Malgorzata Halon-Golabek<sup>1</sup>† , Andzelika Borkowska<sup>2</sup>† , Anna Herman-Antosiewicz<sup>3</sup> and Jedrzej Antosiewicz<sup>4</sup> \*

 Department of Physiotherapy, Faculty of Health Sciences, Medical University of Gdansk, Gda ´ nsk, Poland, ´ Department of Bioenergetics and Physiology of Exercise, Faculty of Health Sciences, Medical University of Gdansk, Gda ´ nsk, Poland, ´ Department of Medical Biology and Genetics, Faculty of Biology, University of Gdansk, Gda ´ nsk, Poland, ´ Department of Biochemistry, Gdansk University of Physical Education and Sport, Gda ´ nsk, Poland ´

Recent studies clearly indicate that the endocrine function of the skeletal muscle is essential for a long and healthy life. Regular exercise, which has been shown to stimulate the release of myokines, lowers the risk of many diseases, including Alzheimer's and Parkinson's disease, emphasizing the role of skeletal muscle in proper functioning of other tissues. In addition, exercise increases insulin sensitivity, which may also impact iron metabolism. Even though the role of iron in neurodegeneration is well established, the exact mechanisms of iron toxicity are not known. Interestingly, exercise has been shown to modulate iron metabolism, mainly by reducing body iron stores. Insulin signaling and iron metabolism are interconnected, as high tissue iron stores are associated with insulin resistance, and conversely, impaired insulin signaling may lead to iron accumulation in an affected tissue. Excess iron accumulation in tissue triggers iron-dependent oxidative stress. Further, iron overload in the skeletal muscle not only negatively affects muscle contractility but also might impact its endocrine function, thus possibly affecting the clinical outcome of diseases, including neurodegenerative diseases. In this review, we discuss possible mechanisms of iron dependent oxidative stress in skeletal muscle, its impact on muscle mass and endocrine function, as well as on neurodegeneration processes.

Keywords: myokine, iron, insulin signaling, ALS, skeletal muscle, neurodegeneration, Akt, JNK

# INTRODUCTION

Excess of iron in any tissue may induce oxidative stress and impair tissue function. In the skeletal muscle, oxidative stress not only causes muscle damage but also negatively impacts its endocrine function. The skeletal muscle is a source of myokines, which are cytokines produced and released by skeletal muscle capable of exerting protective effects on other tissues, including the neuronal tissue (Besse-Patin et al., 2014; Dai et al., 2018; Liu et al., 2018). This is supported by the observation that regular exercise, which pronouncedly increases myokine biosynthesis, reduces the risk of various diseases, including Parkinson's disease (PD) and Alzheimer's disease (AD) (Chen et al., 2005; Santos-Lozano et al., 2016). Conversely, disruption of balance between muscle protein synthesis and degradation, resulting from a wide variety of conditions, including cancer, immobilization (or disuse), denervation, or iron overload, can lead to oxidative stress dependent skeletal muscle atrophy and impairment of myokine synthesis (Tisdale, 2004; Argiles et al., 2014). Although muscle

#### Edited by:

Isabella Zanella, Università degli Studi di Brescia, Italy

#### Reviewed by: Lee J. Martin,

Johns Hopkins University, United States Andrew Paul Tosolini, University College London, United Kingdom

\*Correspondence:

Jedrzej Antosiewicz jant@gumed.edu.pl

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 31 October 2018 Accepted: 12 February 2019 Published: 15 March 2019

#### Citation:

Halon-Golabek M, Borkowska A, Herman-Antosiewicz A and Antosiewicz J (2019) Iron Metabolism of the Skeletal Muscle and Neurodegeneration. Front. Neurosci. 13:165. doi: 10.3389/fnins.2019.00165

function is widely studied in terms of adaptive changes induced by exercise, or atrophy induced by some morbidities, much less is known about its possible illness-related role as an endocrine tissue, and about the interconnections between oxidative stress and myokine production. This topic will probably represent one of the hot new areas in the study of pathomechanisms of neurodegeneration and other diseases.

Iron overload is a known contributor to multiple degenerative diseases, including liver fibrosis, heart attack, and cancer (Stevens et al., 1994; Klipstein-Grobusch et al., 1999; Ong and Farooqui, 2005). Importantly, excess iron accumulation in the brain is linked to neurodegenerative disorders (Bartzokis et al., 1999, 2000). Some neurodegenerative diseases are associated with the failure of muscle function (Busse et al., 2008). However, little is known about the link between iron accumulation in the muscle and neurodegeneration. Some interesting results come from studies on amyotrophic lateral sclerosis (ALS), a neurodegenerative disease characterized by a selective loss of motor neurons (Gajowiak et al., 2015). These findings, as well as current knowledge about iron metabolism in the skeletal muscle and its possible influence on neurodegenerative diseases, will be discussed in the current review.

#### ENDOCRINE FUNCTION OF THE SKELETAL MUSCLE AND NEURODEGENERATION

In recent years, the skeletal muscle has been recognized as a secretory organ that releases appreciable amounts of circulating proteins, called myokines. Currently, we know that the skeletal muscle produces several hundreds of peptides classified as myokines, and muscle contraction stimulates their release (Henningsen et al., 2010; Huh, 2018). Considering that the skeletal muscle represents the largest organ of the human body, (the muscles constitute approximately 40% of total body mass), its role in the regulation of metabolic processes via myokines appears to be very important. Myokines can act as autocrine, paracrine, or endocrine stimuli. Thus, they may affect different organs and tissues, e.g., the brain, bone, adipocyte tissue, heart artery, and many others (Giudice and Taylor, 2017). For instance, the myokines interleukin (IL) IL-6 and IL-10, released from the muscle during exercise or under ischemia, exert powerful local and systemic anti-inflammatory effects. Furthermore, IL-10 has been shown to provide cardio- and neuroprotection, mediated by the activation of anti-apoptotic protein kinase B (PKB or Akt) (Sharma et al., 2011; Cai et al., 2013). Physical activity induces central and peripheral production of neurotrophins, such as brain-derived neurotrophic factor (BDNF), glial cell line-derived neurotrophic factor (GDNF), neurotrophin-3 (NT-3), and neurotrophin-4 (NT-4). They support neural survival, growth, synaptic plasticity, and neuromuscular junctions (Zoladz and Pilc, 2010). In addition, myokines, such as myostatin, irisin, IL-15, IL-6, leukemia inhibitory factor (LIF), or apelin, play a major role in processes associated with regulation of hypertrophic muscle growth and myogenesis (Munoz-Canoves et al., 2013).

Limited data are available on the effect of oxidative stress on the biosynthesis of myokines, where myostatin is one of the examples. It is a member of transforming growth factor beta superfamily and negatively regulates muscle growth. The myostatin/follistatin ratio is significantly higher in ALS in comparison to control patients, and is positively correlated with muscle degeneration (Tasca et al., 2016). Oxidative stress has been shown to increase myostatin synthesis (Enoki et al., 2016) and, conversely, myostatin increases the production of reactive oxygen species (ROS) by NADPH oxidase in C2C12 cells (Sriram et al., 2011). Expression of myostatin is downregulated by regular exercise (Jones et al., 2004; Kim et al., 2005; Louis et al., 2007). Interestingly, compared with a sedentary ALS animal, swim training of ALS mouse significantly lowers oxidative stress and delays body weight reduction (Flis et al., 2018). In addition, it has been shown that swim training sustains the motor function and increases the ALS mouse life span by about 25 days. This beneficial effect is one of the most important therapeutic achievements in the strategy of ALS treatment. What is more, the analysis of muscle phenotype revealed maintenance of the fast phenotype in fast-twitch muscles, delayed spinal motoneuron death, and preserved astrocyte and oligodendrocyte populations in ALS spinal cord (Deforges et al., 2009). Recent data have shown that swimming exercise not only extends life span in mouse model of ALS, but also maintains the grip strength in ALS mice, lowers cholesterol content, and raises the caveolin-1 protein level in the skeletal muscle crude mitochondrial fraction. Moreover, higher activity of COX enzyme in swimming animals seems to be a marker of respiratory chain function improvement (Flis et al., 2018). However, the role of myokines in protective effects of swimming training on ALS development has not been studied.

The role of myostatin inhibitors as potential therapeutics for muscle-wasting diseases and muscle weakness in human and animals has been widely explored. Several myostatin inhibitors, including myostatin antibodies, anti−myostatin peptibody, activin A antibody, soluble (decoy) forms of soluble activin receptor type IIB (ActRIIB−Fc), anti−myostatin adnectin, and ActRIIB antibody have been tested in pre-clinical and clinical trials in the last decade. These inhibitors have currently progressed into clinical development in several indications, mainly sarcopenia, early recovery after surgery, and cachexia. Myostatin inhibitors for the treatment of muscular dystrophy are also being tested in early clinical trials (Saitoh et al., 2017). There are many papers showing positive effects of myostatin inhibitors on animal models with different types of muscle disorders (Holzbaur et al., 2006; Ohsawa et al., 2006; Morine et al., 2010). It has been demonstrated that treatment of an ALS mouse with myostatin inhibitor, ActRIIB−Fc, results in a delay in the onset of weakness, increases body weight and grip strength, and enlarges muscle size when applied either in a pre-symptomatic animal or after symptom onset (Morrison et al., 2009). Surprisingly, in an animal denervation model, myostatin inhibition is not effective against atrophy. By contrast, ActRIIB-Fc treatment protects immobilized mice against the loss of muscle mass (MacDonald et al., 2014). Myostatin is thought to disrupt the balance between protein synthesis and protein degradation in healthy skeletal muscle by inhibiting Akt

kinase, which regulates the muscle mass by inhibiting protein degradation and promoting protein synthesis (Morissette et al., 2009; Trendelenburg et al., 2009; Glass, 2010; **Figure 1**). The most common, myostatin-dependent, signaling pathways involved in muscle myogenesis and protein synthesis/degradation include: Akt/mTOR/mTORC/p70s6K; Akt/FOXO/Atrogin-1, MURF-1; Akt/GSK-3β/cyclinD1 (Elkina et al., 2011; Rodriguez et al., 2014).

Contrary to myostatin, other myokines, e.g., apelin and IL-15, exert a protective effect against oxidative stress and skeletal muscle atrophy.

Apelin is a peptide which is an endogenous ligand for the apelin receptor (APJ) (Tatemoto et al., 1998). The apelin/APJ system has several important functions in the body, such as blood pressure regulation, cardiac contractility, immunity, glucose metabolism, water homeostasis, cell proliferation, angiogenesis, and neuroprotection (Wu et al., 2017). According to research carried out in recent years, apelin is considered to be a myokine (Besse-Patin et al., 2014; Son et al., 2018; Vinel et al., 2018). Vinel et al. (2018) demonstrated in their in vitro and in vivo studies that apelin reverses age-related sarcopenia. During aging, apelin synthesis in skeletal muscle is reduced and plasma apelin levels decrease. Conversely, aged mice, supplemented with a daily injection of apelin or overexpressing apelin, exhibited improved muscle capacities and myofiber hypertrophy (Vinel et al., 2018). A number of studies demonstrated that apelin mediates neuroprotection in in vivo and in vitro models (Kasai et al., 2011; Cheng et al., 2012; Zou et al., 2016; Ishimaru et al., 2017). Moreover, epidemiological and clinical studies reported that physical activity could reduce the risk of developing of PD and AD (Bhalsing et al., 2018; Zhang et al., 2018). Interestingly, apelin stimulates endothelial nitric oxide release that plays a role in inhibition of amyloid beta (Aβ) production, synaptic plasticity, and Aβ clearance in brain (Masoumi et al., 2018). In cells treated with 6 hydroxydopamine, which imitates dopaminergic neurotoxic conditions in PD, pretreatment with apelin-13 significantly decreased the level of intracellular ROS. In heart, apelin protects from ROS-dependent damage or cardiac hypertrophy (Zeng et al., 2009; Foussal et al., 2010). Apelin pretreatment decreased the generation of ROS and malonaldehyde content as well as lactate dehydrogenase leakage in myocardial cells from neonatal rats under hypoxia/re-oxygenation. Furthermore, apelin enhanced superoxide dismutase activity and phosphorylation of extracellular signal-regulated kinase 1/2 and Akt after hypoxia/reoxygenation (Zeng et al., 2009). In a mouse model of ALS, mRNA levels of apelin and its receptor were significantly lower in spinal cord of the G93A hmSOD1 mouse than those in the wildtype mouse. Immunohistochemical analysis revealed a reduced number of motor neurons and activation of microglial cells in transgenic G93A hmSOD1 apelin-deficient mouse, indicating that apelin deficiency pathologically accelerates the progression of disease. Besides, apelin enhanced the protective effect of VEGF on hydrogen peroxide (H2O2)-induced neuronal death in primary neurons (Kasai et al., 2011).

Skeletal muscle tissue releases high amounts of IL-15, which is reported to increase transiently, immediately following resistance (Riechman et al., 2004) and aerobic exercise (Tamura et al., 2011). In the mice muscle and serum, IL-15 protein levels decline progressively with advanced age (Quinn et al., 2010). Further, Yalcin et al. (2018) demonstrated that the level of IL-15 is significantly lower in patients with sarcopenia compared to non-sarcopenic old people. A study conducted on a C2C12 muscle cell line presented the protective effects of IL-15 against H2O2-iduced oxidative stress. Pre-incubation with IL-15 reduced the intracellular creatine kinase and lactate dehydrogenase activities and decreased the ROS overproduction in H2O<sup>2</sup> exposured myoblasts (Li et al., 2014).

Irisin a novel myokine, which is secreted following proteolytic cleavage of its precursor fibronectin type III domain containing 5 (FNDC5). Irisin plays a role in metabolic diseases, aging, inflammation, and neurogenesis (Mahgoub et al., 2018). Physical activity increases irisin level in plasma (Jedrychowski et al., 2015). During AD, the level of FNDC5 /irisin was decreased in hippocamp and cerebrospinal fluid. Knockdown of FNDC5/irisin in the brain impairs long-term potentiation and novel object recognition memory in mice. Conversely, boosting brain levels of FNDC5/irisin rescues synaptic plasticity and memory in AD mouse models. Peripheral overexpression of FNDC5/irisin rescues memory impairment, whereas blockade of either peripheral or brain FNDC5/irisin attenuates the neuroprotective actions of physical exercise on synaptic plasticity and memory in AD mice. Irisin reduced ischemia-induced neuronal injury, and significantly suppressed the levels of nitrotyrosine, superoxide anion, and 4-hydroxynonenal in peri-infarct brain tissues. Mice administration with irisin increased Akt and ERK1/2 phosphorylation, while blockade of Akt and ERK1/2 by specific inhibitors reduced the neuroprotective effects of this compound. Finally, the exercised mice injected with irisin neutralizing antibody displayed more severe neuronal injury than the exercised mice injected with control IgG (Li et al., 2017). In obese patients and in chronic diseases, such as type I and type II diabetes or chronic kidney disease, the level of irisin is lower than in healthy normal subjects (Wen et al., 2013; Ebert et al., 2014; Belviranli et al., 2016; Lu et al., 2016; Shelbaya et al., 2017). The above-mentioned cases, as well as other chronic diseases, are accompanied by chronic inflammation.

#### DYSREGULATION OF IRON METABOLISM AND SKELETAL MUSCLE ATROPHY

Loss of muscle mass is caused by an imbalance between protein synthesis and muscle fiber degradation. Two main degradation pathways can be hyperactivated during muscle dystrophy: the ubiquitin-proteasome and autophagy-lysosome systems. These two pathways require ATP and are believed to serve separate functions. Proteasomes degrade myofibrillar and short-lived proteins (Solomon and Goldberg, 1996; Clarke et al., 2007; Fielitz et al., 2007; Cohen et al., 2009), whereas autophagy-lysosomes remove long-lived proteins and organelles (Levine and Kroemer, 2008; Mizushima et al., 2008).

The ubiquitin-proteasome system relies on a cascade of enzymatic reactions that culminate in the labeling of substrate

Akt in skeletal muscle and neuronal tissue thus protect against muscle and neurons atrophy. Loss of muscle mass thus, reduction in their endocrine functions may accelerate neurodegeneration and degeneration of motor neurons promotes muscle atrophy.

proteins with ubiquitin chains, for degradation by the 26S proteasome. E3 ubiquitin ligases confer substrate specificity and play a crucial role in this system. Two E3 ubiquitin ligases are essential for the development of skeletal muscle atrophy: muscle atrophy F-box (MAFbx)/atrogin-1 and muscle RING finger-1 (MuRF1). They are responsible for the selection and ubiquitination of myofibrillar proteins for subsequent proteosomal degradation (Bodine et al., 2001; Gomes et al., 2001).

The autophagy-lysosomal pathway involves sequestration of substrates within vacuoles called autophagosomes. These vacuoles subsequently fuse with lysosomes, and the cargo is hydrolyzed by lysosomal hydrolases. This process is controlled by autophagy-specific gene products, including Beclin 1. Crucial stages of the pathway rely on the transfer of small ubiquitin-like molecules (LC3 and others) from the conjugation system to the membranes, to allow their growth into double-membrane autophagosomes that engulf portions of the cytoplasm (Kabeya et al., 2000; Mizushima et al., 2004).

Interestingly, both pathways are upregulated during atrophy by Forkhead box (FOX) O3a, which regulates the transcription of genes coding for atrogin-1, MuRF1, LC3B, and its homolog Gabarap 1, as well as Beclin 1 (Sandri et al., 2004; Mammucari et al., 2007; Zhao et al., 2007). Transcriptional activity of FOXO3a is regulated by posttranslational modifications. Regulation by Akt has been most extensively investigated; the protein phosphorylates FOXO3a on Thr32 and Ser253, leading to its cytosolic retention by 14-3-3 (Brunet et al., 1999). Consequently, factors that activate Akt, such as insulin or the

growth factor/phosphatidylinositide 3-kinase (PI3K) pathway, cause FOXO3a inactivation and prevent the synthesis of proteins involved in muscle atrophy. By contrast, phosphorylation of Ser413/588 of FOXO3a by AMP-activated protein kinase, a protein that becomes activated during energy deficit (exercise, hypoxia, or nutritional stress), leads to its activation and, subsequently, the induction of protein degradation pathways (Sandri, 2010; Sanchez et al., 2012).

The relationship between muscle iron metabolism and muscle atrophy with age or disease is unclear; however, recent reports have shed some light on these processes. Ikeda et al. (2016) showed that iron administration results in a decrease of skeletal muscle mass in mouse. The molecular mechanism of this phenomenon involved the induction of oxidative stress and inhibition of the Akt-FOXO3a pathway, hence, upregulation of atrogin-1 and MuRF1. Silencing of FOXO3a expression in C2C12 myotube cells or application of ROS scavenger, TEMPOL, suppress iron-induced expression of atrogin-1 and MuRF1, and prevent cell atrophy (Ikeda et al., 2016). Furthermore, Huang et al. (2013) demonstrated that mouse fed a high-iron diet exhibits elevated AMP-activated protein kinase activity and impaired insulin signaling in the skeletal muscle and liver. These effects are abrogated by co-treatment with N-acetyl cysteine (Huang et al., 2013).

Another regulator of the degradation pathways is tumor necrosis factor receptor-associated factor (TRAF6). It induces the expression of muscle-specific E3 ubiquitin ligases and autophagyrelated molecules in the skeletal muscle on denervation and in Lewis lung carcinoma tumor-bearing mouse (Paul et al., 2010). It is worth noting that iron might stimulate this signaling pathway which was shown in hepatic macrophages (Zhong et al., 2012). Thus, iron accumulation in the skeletal muscle may play an underlying role in skeletal muscle atrophy (**Figure 1**).

#### IRON ACCUMULATION IN THE SKELETAL MUSCLE

Under physiological conditions, most iron is stored in the liver, spleen, and bone marrow; however, a high amount of iron has also been detected in the skeletal muscle. It has been determined that the total amount of stored iron in the skeletal muscle is comparable with that in the liver in healthy individuals (Torrance et al., 1968). Furthermore, muscle storage of iron can increase, e.g., in individuals with iron overload.

Iron metabolism seems to be tightly controlled. It is not entirely clear why under some conditions iron accumulates in the skeletal muscle and/or other tissues as well. It has been demonstrated that diet rich in highly bioavailable forms of iron promotes high iron stores, whereas foods containing phytate and other natural iron chelators reduce these stores. Conversely, under some pathological conditions, excessive iron accumulation is observed regardless of the diet. For example, in an animal model of ALS, the amount of iron and iron storage proteins, ferritin L and ferritin H, is elevated in the skeletal muscles and neurons (Jeong et al., 2009; Halon-Golabek et al., 2018). Skeletal muscle iron accumulation has also been observed after immobilization (Kondo et al., 1992). Further, hepatic iron content significantly increases after 2 weeks of a high-fructose diet (Ackerman et al., 2005). These data clearly indicate that tissue iron accumulation is not always associated with the consumption of food with high-iron content (Tsuchiya et al., 2013) but, rather, with impaired tissue iron metabolism. The mechanism of iron transport into a cell is well understood; however, the changes in iron metabolism that are responsible for excess iron accumulation are not fully known.

Iron overload can negatively affect skeletal muscle function, as it can induce oxidative stress (Schafer et al., 1981). Intracellular ROS formation is strongly associated with the amount of free iron. Lowering the levels of catalytic free iron in a cell by using chelators always results in reduced ROS formation and changes the composition of free radical species. For example, formation of the hydroxyl radical is iron-dependent.

It is not clear why increased iron stores correlate with enhanced iron-dependent oxidative stress since iron, stored mostly in ferritin, does not stimulate ROS formation. Despite this, a positive correlation between oxidative DNA damage and body iron stores has been observed (Barollo et al., 2004; Sullivan, 2004). Iron may affect the clinical course of diseases associated with the pathological disorders of the muscle. We have recently shown that in a transgenic rat bearing the G93A hmSOD1 gene (an animal model of familial ALS), iron levels in the muscle increased with the development of disease, and that was accompanied by increased oxidative stress (Halon et al., 2014; Halon-Golabek et al., 2018). Taken together, similarly to the brain, liver, and some other tissues, under certain conditions, the skeletal muscle may accumulate too much iron, which contributes to ROS formation.

#### STRESS AND IRON SIGNALING

Under stress conditions, numerous signaling pathways are activated within a cell, which may lead to an adaptive response to such conditions. One of such pathways is mediated by stressactivated protein kinases, and results in ferritin degradation and release of free iron [the so-called labile iron pool (LIP)]. In cell culture models, c-jun terminal kinase (JNK-1), a stress-activated protein kinase, together with p66Shc adaptor protein, mediates ferritin degradation by the proteasome (Antosiewicz et al., 2006, 2007; Borkowska et al., 2011). Ferritin is a protein that binds iron atoms and stores them in a "safe" (non-reactive) form. One molecule of ferritin can bind up to 4500 iron atoms in the form of ferric iron (Fe3+), creating a mineral core, in which the iron is stored in complex with phosphate. However, ferritin may be a source of free iron if it undergoes proteasomal or lysosomal degradation. Recent studies indicate that ferritin can be also degraded by autophagy mediated by nuclear receptor co-activator 4 (Philpott et al., 2017). These and some other studies clearly show that ferritin iron is not a "safe" form of iron, as it can be liberated and subsequently stimulate iron-dependent cell damage (Sullivan, 2004; **Figure 2**).

Despite the ability to induce cell damage, iron is also a physiological signaling molecule. Its signaling properties are associated with ROS formation. ROS can oxidize specific amino

acids in proteins and thus modulate their activity; e.g., ironresponsive protein 2 in the presence of high iron levels undergoes site-specific oxidation, which targets this protein for proteolytic degradation (Iwai et al., 1998). Cysteine is an amino acid with a very high affinity for chelatable iron. Interestingly, cysteine oxidation by H2O<sup>2</sup> is often not possible in the absence of iron. That is because of the high pKa of sulfydryl groups of most cysteine residues in proteins (approximately 8.5). Only cysteine thiolate anion (Cys-S-) is vulnerable to oxidation by H2O2. Thus, the amount of iron released during ferritin degradation may greatly impact cellular response to stress.

One of the most important signaling activities of iron is associated with its interaction with iron-responsive elementbinding proteins, which can upregulate ferritin gene expression. Increases in ferritin L and H levels lead to the sequestration of free iron, and reduction of iron-dependent ROS formation, lowering tissue damage. Such a scenario has been observed in an ischemic heart, in which ischemic preconditioning induces irondependent upregulation of ferritin, protecting the heart during full ischemia (Chevion et al., 2008). Certainly, the amount of iron liberated during ferritin degradation will determine tissue responses to stress. For example, in a mouse model of iron

overload, elevated levels of iron in the tibialis anterior muscle and a fourfold increase in ferritin light chains were observed. These changes were accompanied by elevated markers of oxidative stress; significant reduction in the fast-twitch (extensor digitorum longus) and slow-twitch (soleus) muscles mass; and decreased exercise capacity (Reardon and Allen, 2009). Moreover, cytotoxicity of the tumor necrosis factor was augmented by iron and significantly reduced by iron chelators on cell line model (Warren et al., 1993).

# INSULIN SENSITIVITY AND IRON ACCUMULATION IN THE SKELETAL MUSCLE

Insulin or growth factors/PI3K/Akt/FOXO3a is another important signaling pathway that is affected by stress. Insulin resistance is observed in numerous pathological conditions and is associated with the impaired Akt/FOXO3a signaling pathway. Insulin and growth factors-mediated activation of Akt is facilitated by its membrane recruitment upon interaction with phosphatidylinositol 3,4,5-trisphosphate synthesized by PI3K. After membrane anchoring, Akt is phosphorylated at Ser308 and Ser473 by phosphoinositide-dependent kinase 1 and mammalian target of rapamycin complex 2, respectively. Upon phosphorylation, Akt translocates from the plasma membrane to intracellular compartments, including the nucleus, where it phosphorylates a range of substrates (Song et al., 2005). One of these substrates is transcriptional factor FOXO3a, phosphorylated by Akt at Thr32, Ser253, and Ser315. FOXO3a phosphorylation leads to its exclusion from the nucleus and reduction of its DNA-binding activity. Stress conditions may activate JNK, which phosphorylates Ser574 of FOXO3a, antagonizes the Akt signaling pathway, and promotes nuclear translocation and transcriptional activity of FOXO3a (Sunayama et al., 2005; Wang et al., 2012). FOXO3a plays an important role in the upregulation of genes associated with oxidative stress resistance, including catalase and MnSOD genes.

Recently, it has been demonstrated that Caenorhabditis elegans DAF-16, an ortholog of the FOXO family of transcription factors, regulates iron metabolism by increasing the expression of ferritin H (Ackerman and Gems, 2012). We confirmed this observation using a mammalian cell line, demonstrating that an increase in FOXO3a activity leads to upregulation of ferritin protein levels in cells (Halon-Golabek et al., 2018). This observation was confirmed in the skeletal muscle of transgenic animals expressing SOD1 G93A in which Akt activity reduction, FOXO3a activity increase, and upregulation of ferritin protein and catalase activity were observed (Halon-Golabek et al., 2018).

Thus, impairment in insulin signaling increases FOXO3amediated induction of antioxidant proteins, such as catalase, MnSOD, ferritin, and increases cellular resistance to oxidative stress. Under some conditions, increasing ferritin H protein levels is sufficient to augment cell antioxidant potential. For example, in a cell culture model, overexpression of ferritin H leads to 50% reduction of LIP and increased resistance to oxidative challenge (Cozzi et al., 2000). In the short term, such changes are beneficial for the cell; however, in the longer term, such conditions can disturb iron metabolism. Accordingly, overexpression of ferritin H leads to reduced iron-dependent signaling and induction of an iron-deficiency phenotype. These changes are manifested by a fivefold increase in the activity of iron-responsive elementbinding proteins; 2.5-fold increase of transferrin receptor levels; and 1.8-fold increase in iron-transferrin uptake (Cozzi et al., 2000). Hence, it can be expected that during chronic stress, upregulation of ferritin may lead to reduced LIP levels.

Transfer of some labile iron to ferritin may generate irondeficiency phenotype, even if the total amount of iron in a cell is unchanged. Under such conditions, the cell will continue to import iron until LIP returns to the usual level. Therefore, as mentioned above, both Akt and JNK kinases can be involved in iron metabolism by modulating FOXO3a/ferritin activity. Further, it has been shown on cell lines that JNKmediated ferritin degradation is accompanied by increase in LIP and iron-dependent ROS formation (Antosiewicz et al., 2007; Borkowska et al., 2011). JNK activation and a decrease in Akt activity have been observed in ALS humans and animals (Kim and Choi, 2015). Interestingly, decreases in Akt levels, upregulation of ferritin L, and decreases in ferritin H levels are observed in the muscle of ALS pre-symptomatic animals. These changes are accompanied by increased levels of oxidative stress markers, possibly because of impaired iron metabolism. Hence, it can be speculated that LIP transiently increases before the first symptoms of disease in the muscle of ALS animals, unblocking the translation of ferritin and activating FOXO3a, to augment the transcription of ferritin genes. In later phases of disease, upregulation of ferritin may cause iron accumulation and iron-dependent induction of oxidative stress (**Figure 3**). Interestingly, impairment insulin signaling and iron accumulation have also been observed in the brain of PD and AD (Aviles-Olmos et al., 2013; Calsolaro and Edison, 2016; Rani et al., 2016; Apostolakis and Kypraiou, 2017). Insulin receptors are found in the basal ganglia and substantia nigra and growing evidences are suggesting that insulin plays an essential regulating role in neuronal survival and growth, dopaminergic transmission, and maintenance of synapses (Bassil et al., 2014). Thus, there are some evidences that patients with type 2 diabetes (DMT2) have an increased risk of developing PD and share similar dysregulated pathways suggesting common underlying pathological mechanisms (Dunn et al., 2014). In the early stage of the DMT2, patients develop insulin resistance, leading to a variety of detrimental effects on metabolism and inflammation. Accumulating evidence suggests that similar dysregulation of glucose and energy metabolism seems to be an early event in the pathogenesis of sporadic PD, indicating that the insulin signaling pathway may potentially be a novel target for disease modification (Athauda and Foltynie, 2016). There is a fundamental question, whether insulin resistance occurs as a cause or a consequence of neurodegeneration. Substantial evidence implicates that loss of Akt control (an important downstream target of insulin signaling pathway) is involved in DMT2 and AD (Griffin et al., 2005). There are strong evidences that an altered Akt signaling pathway could be a component of PD neurodegeneration (Greene et al., 2011). Reduction in

phosphorylated Akt kinase was observed in post mortem studies of PD patients (Malagelada et al., 2008; Timmons et al., 2009). In contrary to skeletal muscle, the interdependence between insulin signaling and iron metabolism in neuronal tissue has not been studied. We can only speculate that changes in insulin signaling manifested by the inhibition of PI3K/Akt/FOXO3a signaling pathway may trigger changes in iron metabolism which will lead to brain iron accumulation and iron-dependent oxidative stress.

# THE ROLE OF IRON IN PATHOMECHANISM OF ALS

Amyotrophic lateral sclerosis, also known as Lou Gehrig's disease, is a progressive, usually fatal, neurodegenerative disease in human adulthood. It is caused by the degeneration of motor neurons in the spinal cord, i.e., the nerve cells in the central nervous system that control voluntary muscle movement. ALS is characterized by escalating muscle weakness, spasticity, and atrophy as both the upper and lower motor neurons degenerate (Rowland, 2001; Lino et al., 2002; Gajowiak et al., 2016). Recent studies indicate that the nerve cells and non-neuronal cells may both play a decisive role in the onset of disease. Transgenic mice harboring SOD1 G93A gene recapitulates the symptoms of human ALS and are considered as an animal model of the disease. Interestingly, expression of the SOD1 variant exclusively in the motor neurons results in no apparent pathology or motor deficit suggesting that accumulation of the hm SOD1 gene in neurons is not a critical factor for the onset of the disease (Lino et al., 2002). Furthermore, studies involving the animal model indicate that restricted expression of the mutated SOD1 gene in astrocytes, neurofilaments, or oligodendrocytes is not sufficient to cause motor neuron degeneration in vivo. Surprisingly, these animals fail to develop the disease, which confirms the notion that the expression of SOD1 G93A in other cell types is critical for disease initiation, or that other factors beside the presence of mutated SOD1 gene are essential for the onset of disease (Gong et al., 2000; Pramatarova et al., 2001; Yamanaka et al., 2008). Many scientific reports indicate oxidative stress as an initiating factor in the onset of the disease. In a transgenic mouse model, selective expression of mutated SOD1 G93A (Dobrowolny et al., 2008) or G37R (Wong and Martin, 2010) exclusively in the skeletal muscle demonstrated progressive muscle atrophy, associated with a significant reduction in muscle strength, alterations in the contractile apparatus-sarcomere and sarcotubular system disorganization, and mitochondrial dysfunction, mimicking a pathologic phenotype consistent with ALS. These data demonstrate that skeletal muscle is a primary target of SOD1(G93A)-mediated toxicity and disclose the molecular mechanism whereby oxidative stress triggers muscle atrophy, where human SOD1 has a causal role in ALS and motor neuron degeneration explaining their selective vulnerability (Dobrowolny et al., 2008; Wong and Martin, 2010).

Thanks to recent advances in genetics, we now know that ALS is associated with mutations in at least 20 genes, coding for proteins related to cell functions as diverse as RNA metabolism (TARDBP, FUS/TLS, Senataxin, Ataxin2, HNRNPA2/B1, ELP3, HNRNPA1), vesicle trafficking (Alsin, FIG4, OPTN, VABP, CHMP2B), and proteasomal function (UBQLN2, VCP) (White and Sreedharan, 2016). The GGGGCC hexanucleotide repeat expansion in gene C9orf72 is the most numerous genetic variant of ALS, which together with genetic modifications in SOD1, TARDBP, FUS, Antaxin, or OPTN proteins, is extensively explored, in the ALS context.

The analyses of disorders in iron metabolism in ALS patients, in most cases, concern patients in whom the etiology of the disease is unknown. By contrast, in animal models, rodents of mutated hm SOD1 gene (most commonly SOD1 G93A and SOD1 G37R) predominate. In ALS patients, iron accumulation was observed in the spinal cord and cerebrospinal fluid (Kasarskis et al., 1995). MRI technic is a useful method to detect iron deposits in the brain in ALS patients. There are some evidence showing iron aggregates in the precentral gyrus's gray matter, which shift toward higher MRI scores after 6 months (Ignjatovic et al., 2013). In postmortem, histopathological analysis of a brains, Perls' DAB staining, revealed abundant cells containing intracellular iron and occasional extracellular iron deposits in the motor cortex of ALS patients. The intracellular iron was observed in microglia, which was accompanied by a higher level of ferritin (Kwan et al., 2012). In addition, patients with ALS have marked changes in the iron biochemical parameters measured in the blood. Studies have shown increased serum ferritin levels, transferrin saturation coefficient, and decreased transferrin in ALS patients, which could reflect a general increase in iron stores or be a consequence of on-going muscle degeneration (Goodall et al., 2008; Qureshi et al., 2008; Nadjar et al., 2012; Hu et al., 2016). In addition, elevated serum ferritin levels have been associated with reduced survival. Patients with high level serum ferritin had a shorter survival time compared to those with low level serum ferritin (618 days versus 921 days). However, increased serum ferritin levels are not specifically indicative of increased body iron storage. Systemic inflammation can also be associated with increased serum ferritin, especially during the end-stage of disease, and positively correlates with bronchial congestion in patients with respiratory muscle weakness (Nadjar et al., 2012).

Amyotrophic lateral sclerosis animal models allow for more accurate investigation of molecular mechanisms of iron accumulation in the nervous system or muscles. Jeong et al. (2009) proved iron accumulation in the spinal cord of SOD1G37R transgenic mice at 12 months of age. In the cell bodies of the large ventral horn motor neurons in the spinal cord, the level of cytoplasmic iron inclusions (measured by ferrozine) was increased by 56% in SOD1G37R mice, compared with age-matched wild-type controls (Jeong et al., 2009). On the other hand, changes in iron metabolism were observed in skeletal muscles of SOD1 G93A transgenic rats. Changes in expression of the iron metabolism proteins (H-ferritin and ferroportin) in skeletal muscles were already observed at the pre-symptomatic stage in SOD1G93A rats. During

disease progression, the level of ferritin H significantly increased and was accompanied by iron accumulation (Halon et al., 2014).

Information on iron metabolism disorders in ALS models other than those expressing mutations in the SOD1 gene is poor. In 2006, the gene coding for transactive response DNA binding protein 43 kDa (TDP-43), an RNA/DNA binding protein, was implicated in ALS as the major component of ubiquitinated inclusions (Neumann et al., 2006). TDP-43 is cleaved, hyperphosphorylated, ubiquitinated, or mis-localized in the cytoplasm in the form of insoluble inclusions. Guam-ALS patients presented higher iron levels and lower zinc levels in brain (Yasui et al., 1993). ALS and ALS/PD-dementia patients in Guam also present TDP-43 inclusions as a secondary pathology. In this paper, it has also been shown that the TDP-43 is a consistent component of the ubiquitinated inclusions in sporadic ALS and Guam ALS, but TDP-43 inclusions are absent or scarce in SOD1 familial ALS (Maekawa et al., 2009). On the contrary, evidence shows that TDP-43 was expressed by astrocytes and microglial cells in the spinal cord of hmSOD1(G93A) transgenic mice. In addition, the expression of phosphorylated and truncated TDP-43 increased in the spinal cord of ALS mice compared with age-matched non-transgenic (Tg) (Cai et al., 2015; Jeon et al., 2018). Furthermore, the serum iron concentration and expression of transferrin (a homeostasis-related iron protein) in the SP were increased relative to non-Tg. The protein expression level of heme oxygenase 1 related to oxidative stress was increased in the spinal cord of hmSOD1(G93A) Tg relative to non-Tg (Cai et al., 2015). In TDP-43 A315T transgenic mice, dysregulation of ion metal was noticed. However, only the levels of zinc, copper, and manganese were increased in the spinal cords significantly. The level of iron has not changed significantly (Dang et al., 2014). Many ALS patients (∼36–51%) also exhibit cognitive impairment, with frontotemporal lobar degeneration (FTLD) in about 20%. There is some probability that patients with FTLD might develop ALS. Pathologically, FTLD includes multiple subtypes, including FTLD-TDP-43, FTLD-FUS (fused in sarcoma/translocated in liposarcoma), or FTLD-Tau (Guerrero et al., 2016). Postmortem research conducted on human brains showed that a significant increase of iron deposition was observed in the claustrum, caudate nucleus, globus pallidus, thalamus, and subthalamic nucleus of the FTLD-FUS and FTLD-TDP groups, while in the ALS one, the iron increase was only observed in the caudate and the subthalamic nuclei. FTLD is often linked to mutations in FUS and TDP43or to an expanded C9orf72 repeat.

There are no data on skeletal muscle iron metabolism in ALS models other than ALS-SOD. What is interesting, in patients with FTLD, the mutation in hemochromatosis gene (H63D HFE) genotype was found. Further, increased iron uptake, and gradual iron accumulation in tissues and organs have been reported in patients with a mutation of the HFE gene, which encodes a protein responsible for the control of iron absorption by the intestinal epithelial cells. Other studies indicate that the H63D mutation within the gene is a contributing factor in the development of ALS (Wang et al., 2004; Sutedja et al., 2007). Interestingly, studies show increased prevalence of HFE gene mutations in individuals with AD (Percy et al., 2008). Conversely, studies on animal models demonstrated that double transgenic generated mouse line (SOD1/H67D) carrying the H67D HFE (homolog of human H63D) and SOD1(G93A) mutations had a shorter survival and accelerated disease progression comparing to SOD1 G93A mouse. This correlated with decreased transferrin receptor and a significant increase in ferritin L expression-indicators of iron status (L-ferritin expression in double transgenic mice was 86 and 79% higher than SOD1 mice at 90 and 110 days in the double transgenic and SOD1 mice starting at 90 days pre-symptomatic stage), indicating for dysregulation of iron homeostasis in these mice. What is more, compared to SOD1 mice, double transgenic mice performed significantly worse on both forelimb and hindlimb grip strength beginning at the age of disease onset (106 days) until the end of the test (127 days), suggesting an accelerated disease progression in these mice (Nandar et al., 2014).

In support of the role of iron in the development of ALS, it has been reported that the use of iron chelator salicylaldehyde isonicotinoyl hydrazone significantly delays the onset of disease and increases the life span (by approximately 5 weeks) depending on the dosage, which resulted in a five- to sixfold decrease in the number of accumulating iron cells in the spinal cord. These studies clearly indicate that the disturbance of iron homeostasis, or rather its excessive accumulation in central nervous system, leads to the progression of these animals' disease (Jeong et al., 2009). In addition, treatment with another chelator, deferoxamine, does not alter the permeability of the blood–spinal cord barrier to IgG, hemoglobin, or hemosiderin. Instead, it significantly reduced early accumulation of free iron, which was accompanied by a delayed onset of disease in comparison with warfarin- or saline-treated patients. These data suggest that deferoxamine prevents early iron accumulation in the spinal cord but does not exert a direct anti-inflammatory effect (Zhong et al., 2009; Winkler et al., 2014). Similarly to deferoxamine, VK-28 and monoamine oxidase inhibitor (M30) are brain-permeable iron chelators that exhibit neuroprotective and neurorestorative activities in NCS-34 motor neuron cells (Kupershmidt et al., 2009; Wang et al., 2011). Treatment with VK-28 and M30 delays the disease onset, prolongs the lifespan, and reduces spinal cord motor neuron loss in a mouse model of ALS. Furthermore, iron chelators attenuate the elevated level of iron and the expression of transferrin receptor; decrease the production of free oxygen radicals; and suppress microglial and astrocytic activation in the spinal cord of the SOD1 G93A mouse. Furthermore, iron chelators decrease toxic aggregation of the transcriptional regulator TDP-43, decrease the levels of proapoptotic Bax, and increased the levels of antiapoptotic protein Bcl-2 (Wang et al., 2011).

The notion of complex etiology of neurodegenerative disorders led to the discovery of a brain-permeable, nontoxic compound with antiapoptotic and iron-chelating properties—VAR10303. When co-administered with a highcalorie/energy diet to SOD1 G93A mouse, VAR10303 prolonged life span for about 10%, improved motor performance, and attenuated iron accumulation and motor neuron loss in the animal spinal cord (Golko-Perez et al., 2016). This slight extension of the life span in the context of chronic disease seems to have a remarkable meaning and what is far more important, VAR treatment significantly increased stride length distance—a useful gait footprint parameter in mouse model of ALS.

Recently published results indicate that conservative iron chelation—deferiprone (i.e., chelation with low risk of iron depletion), in a murine preclinical model and pilot clinical trial in SOD1 G86R mice, increased the mean life span compared with placebo. This corresponded to a 56% extension in survival (13 days) from disease onset in a female mouse compared to the vehicle group (defined as the peak in the body weight). What is more interesting, an interaction between dose and sex was observed, namely, the required dose was higher in males than in females. Treatment with deferiprone led to a decrease in iron accumulation not only in the murine model SOD1 G86R, but also in ALS patients where a significant decrease in iron concentration was observed in the cervical spinal cord, medula oblongata and motor cortex after 12 months of treatment, but the first visible signs were already seen after 3 months. All patients had a slight elevation in urine iron levels and presented a decreased level of 8-OHdG in the cerebrospinal fluid after 9 months of treatment. Summing up, iron chelators might be promising agents for clinical treatment, as there is a strong evidence indicating the role of iron dysregulation in neuronal cell death the pathophysiology of neuromuscular diseases. The presented and study demonstrates the safety of conservative iron chelation in ALS, as even at a low-dose levels, deferiprone can cross membranes, decrease iron accumulation, may reenter the captured iron into extracellular transferrin, and then spread the iron throughout the body, thus avoiding anemia (Moreau et al., 2018).

#### CONCLUSION

Neurodegenerative diseases are associated with insulin resistance and iron accumulation not only in the central nervous system but also in the muscle. The exact mechanism of iron accumulation in the skeletal muscle and neuronal tissue is yet unknown; however, there is evidence for the involvement of impaired insulin and growth factor/PI3K/Akt/FOXO3a signaling pathways. This accumulation may lead to increased production of ROS and

oxidative stress, which may impair endocrine function of the skeletal muscle and, indirectly, the function of other organs. Hence, under such conditions, the neuronal tissue suffers not only from iron-induced oxidative stress but also from a reduced exposure to trophic factors derived from the skeletal muscle. Thus, it is possible that changes in the iron metabolism might constitute a trigger of a disease, as opposed to being caused by the disease. This notion is supported by observation that changes in iron metabolism are observed before the first symptoms of disease in ALS rats. The protective effects of iron chelators in many models of neurodegeneration confirm the important role of this metal in the pathomechanism of these diseases and indicate that the chelators could have some therapeutic value. Exercise training which is known to increase insulin sensitivity and modulate iron metabolism could be recommended as a preventive approach.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

MH-G, AB, AH-A, and JA conceptualized the study and wrote the manuscript. AH-A and JA reviewed the manuscript.

## FUNDING

This study was supported by grants from the National Science Centre in Poland (2014/13/B/NZ7/02346).

#### ACKNOWLEDGMENTS

The authors would like to express thanks to Joanna Mackie Freelance Science Editor for language assistance.

modification? Prog. Neurobiol. 118, 1–18. doi: 10.1016/j.pneurobio.2014. 02.005



muscle in SOD1(G93A) Transgenic ALS Mice. J. Mol. Neurosci. 59, 504–510. doi: 10.1007/s12031-016-0763-2



in human skeletal muscle. J. Appl. Physiol. 103, 1744–1751. doi: 10.1152/ japplphysiol.00679.2007




**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 Halon-Golabek, Borkowska, Herman-Antosiewicz and Antosiewicz. 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.

# Tremor-Dominant in Parkinson Disease: The Relevance to Iron Metabolism and Inflammation

Teng-Hong Lian<sup>1</sup> , Peng Guo<sup>1</sup> , Li-Jun Zuo<sup>1</sup> , Yang Hu<sup>1</sup> , Shu-Yang Yu<sup>1</sup> , Qiu-Jin Yu<sup>1</sup> , Zhao Jin<sup>1</sup> , Rui-Dan Wang<sup>1</sup> , Li-Xia Li<sup>2</sup> and Wei Zhang3,4,5,6,7 \*

<sup>1</sup> Department of Neurology, Beijing Tiantan Hospital, Capital Medical University, Beijing, China, <sup>2</sup> Department of Internal Medicine, Beijing Tiantan Hospital, Capital Medical University, Beijing, China, <sup>3</sup> Center for Cognitive Neurology, Department of Neurology, Beijing Tiantan Hospital, Capital Medical University, Beijing, China, <sup>4</sup> China National Clinical Research Center for Neurological Diseases, Beijing, China, <sup>5</sup> Key Laboratory for Neurodegenerative Disorders of the Ministry of Education, Capital Medical University, Beijing, China, <sup>6</sup> Center of Parkinson's Disease, Beijing Institute for Brain Disorders, Beijing, China, <sup>7</sup> Beijing Key Laboratory on Parkinson's Disease, Beijing, China

#### Edited by:

Isabella Zanella, Brescia University, Italy

#### Reviewed by:

Antonio Martin-Bastida, Imperial College London, United Kingdom Andrea Pilotto, Brescia University, Italy

> \*Correspondence: Wei Zhang ttyyzw@163.com

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 31 October 2018 Accepted: 04 March 2019 Published: 27 March 2019

#### Citation:

Lian T-H, Guo P, Zuo L-J, Hu Y, Yu S-Y, Yu Q-J, Jin Z, Wang R-D, Li L-X and Zhang W (2019) Tremor-Dominant in Parkinson Disease: The Relevance to Iron Metabolism and Inflammation. Front. Neurosci. 13:255. doi: 10.3389/fnins.2019.00255 Background: Tremor is one of the most predominant symptoms of patients with Parkinson disease (PD), but the underlying mechanisms for tremor relating to iron and its metabolism-related proteins and the inflammatory factors in cerebrospinal fluid (CSF) and serum have not been fully elucidated.

Methods: A total of 135 PD patients were divided into a tremor-dominant (PD-TD) group (N = 74) and a postural instability and gait difficulty-dominant (PD-PIGD) group (N = 39) based on the ratio of mean tremor score to the mean bradykinesia/rigid score of the Unified Parkinson's Disease Rating Scale (UPDRS) III. Age and sex-matched healthy controls were recruited (N = 35). Demographic variables were evaluated; iron and its metabolism-related proteins and the inflammatory mediators in both CSF and serum were measured in these groups. The relevance of iron metabolism, inflammation and PD-TD were analyzed.

Results: (1) The PD-TD group had significantly decreased L-ferritin, increased iron levels in CSF and increased ferritin levels in the serum compared with the PD-PIGD and control groups (P < 0.05). (2) The PD-TD group had significantly enhanced IL-6 levels in both CSF and serum compared with the PD-PIGD and control groups (P < 0.05). (3) In CSF, the IL-6 level was increased as the iron level was elevated in the PD-TD group (r = 0.308, P = 0.022). In serum, the IL-6 level was increased as the ferritin level was elevated in the PD-TD group (r = 0.410, P = 0.004).

Conclusion: The interplay between disturbed iron metabolism and relevant inflammation might modulate clinical phenotypes of PD.

Keywords: Parkinson disease, tremor-dominant, iron metabolism, inflammation, IL-6

# INTRODUCTION

fnins-13-00255 March 25, 2019 Time: 18:13 # 2

Parkinson disease (PD) is a progressive neurodegenerative disease characterized by classical motor manifestations, including resting tremor, bradykinesia, rigidity, and gait and postural abnormalities. Based on these symptoms, PD can be divided into three motor phenotypes: tremor-dominant (PD-TD), postural instability and gait difficulty-dominant (PD-PIGD), and a mixed group (Isaias et al., 2011). It was reported that patients with PD-TD tend to have slower disease progression (Moretti et al., 2017), fewer non-motor symptoms (Ba et al., 2016), more chances of improvement when using levodopa and higher survival rates (Rajput et al., 2017) compared to patients with PD-PIGD.

The typical motor symptoms of PD are predominantly attributed to the neurodegeneration of dopamine-producing neurons in the substantia nigra (SN) and the dysfunction of the basal ganglia (Shulman et al., 2011; De Virgilio et al., 2016). Tremor and non-tremor phenotypes have distinct underlying pathophysiologic mechanisms in PD (Thenganatt and Jankovic, 2014). It is believed that the increased interactions between the basal ganglia and the cerebello-thalamo-cortical circuit may contribute to parkinsonian tremor (Helmich et al., 2011; Helmich, 2018). Moreover, the dopamine depletion in the globus pallidus is associated with the severity of a clinical tremor (Helmich et al., 2011). The tremor-associated drive from the subthalamic nucleus (STN) to the muscles has been confirmed, so STN may also be directly involved in tremor generation (Guan et al., 2017). However, as one of the supporting characteristics of PD, the underlying mechanisms for tremor have not been fully elucidated.

Abnormal iron accumulation is associated with a host of neurodegenerative diseases, such as PD and Alzheimer disease, via an abnormal deposition in specific brain regions or a malfunction of iron homeostasis (Sian-Hulsmann et al., 2011). Most of the research in PD patients focuses on the brain iron overload in regions such as the SN, putamen and globus pallidus (Sian-Hulsmann et al., 2011; Liu et al., 2017; Sjostrom et al., 2017; Xuan et al., 2017; Yu et al., 2018). A study of magnetic susceptibility weighted imaging show that the iron depositions in the dentate nucleus of the cerebellum and the red nucleus, are correlated with tremor in PD (Guan et al., 2017). However, few experiments have been conducted on iron and its metabolismrelated proteins in both the central and peripheral systems, within the different motor phenotypes of PD.

A growing set of observations from post-mortem and in vivo studies suggest that inflammation in both the central and peripheral systems may be one of the primary mechanisms involved in neuronal degeneration in PD (Hirsch and Hunot, 2009). Inflammation in SN may serve as a driving force of neuronal death (Block et al., 2007) and the active peripheral inflammation can promote dopaminergic cell loss by aggravating and synergizing central inflammatory responses (Tiwari and Pal, 2017). Moreover, inflammation is likely to cause iron deposition by inducing changes in iron metabolism proteins (Urrutia et al., 2013) and inhibiting the mitochondrial complex I (Urrutia et al., 2014). In return, excessive free iron induces cellular destruction and neurodegeneration via oxidative and nitrative stress, inflammation, and excitotoxicity (Sian-Hulsmann et al., 2011). The relevance of abnormal iron metabolism and related inflammation has been found in some parkinsonian non-motor symptoms, like pure apathy (Wang et al., 2016), sleep disorders (Yu et al., 2013), and especially rapid eye movement sleep behavior disorder (Hu et al., 2015). The ratio of neutrophillymphocyte, a marker of inflammation, has been observed to have a positive association with motor severity in the PD-TD group (Sanjari Moghaddam et al., 2018). Yet, the role among parkinsonian tremor, abnormal iron metabolism, and inflammation is seldom investigated.

In this study, the levels of iron and the proteins related to iron metabolism, including ferritin, H-ferritin and L-ferritin, transferrin, and lactoferrin, and inflammatory mediators, including interleukin (IL)-6, IL-1β, prostaglandin (PG) E2, nitric oxide (NO), and hydrogen peroxide (H2O2), were detected in both the CSF and serum in the PD-TD, PD-PIGD, and the healthy control groups. We then analyzed the relationship between iron and the proteins related to iron metabolism and inflammatory mediators in the CSF and serum of PD-TD patients.

# MATERIALS AND METHODS

# Subjects

#### PD Patients

A total of 135 PD patients were consecutively recruited from the Department of Neurology, Beijing Tiantan Hospital, Capital Medical University. All PD patients satisfied the Movement Disorder Society Clinical Diagnostic Criteria for Parkinson disease (Hughes et al., 1992).

Demographic variables, including sex, age, education level, disease duration, age at disease onset, vascular risk factors, like hypertension, diabetes, hyperlipidemia and smoking, the number and levodopa equivalent daily dose (LEDD) of patients under dopaminergic treatment for PD were recorded. The severity was evaluated using the Hohen-Yahr stage.

PD patients with systemic diseases, especially infectious and autoimmune diseases and hematological diseases, and patients who have experienced blood loss were excluded.

#### Control Subjects

Normal controls were required to satisfy the following criteria: no intracranial diseases, no neurological symptoms or signs and normal magnetic resonance imaging results of the head. The other exclusion criteria for the control subjects were the same as those for the PD patients. Finally, 35 age and sex-matched healthy controls were recruited for this study.

## Classification of Motor Phenotypes for PD Patients

Motor symptoms were assessed by the Unified Parkinson's Disease Rating Scale (UPDRS) III. Motor phenotypes were identified based on the ratio of mean tremor score (sum of items 20 and 21 in UPDRS III divided by four) to the mean bradykinesia/rigid score (sum of items 22–27 and 31 in UPDRS III divided by 15). Patients with a ratio greater than 1.0, less than 0.80, and between 0.80 and 1.0 were classified into PD-TD, PD-PIGD, and mixed-types (Schiess et al., 2000), respectively. The patients of mixed-type were excluded due to the mixed motor phenotype.

# Collections of CSF and Serum

fnins-13-00255 March 25, 2019 Time: 18:13 # 3

In total, 3 ml CSF and 2 ml venous whole blood was collected in polypropylene tubes at 07:00 to 10:00 in the fasting condition, with anti-parkinsonian medication withdrawal for 12–14 h if patients' condition permitted (Zuo et al., 2016). CSF and serum samples were centrifuged immediately at 3000 rpm at 4◦C. Approximately 0.5 ml volume of the supernatant of CSF and serum were put into separate Nunc cryotubes and frozen at −80◦C. Freezing and thawing and protein degradation were carefully avoided.

## Detections of the Levels of Iron and Its Metabolism-Related Proteins in CSF and Serum

The levels of iron and ferritin, H-ferritin and L-ferritin, transferrin, and lactoferrin in the CSF and serum were measured by using the Enzyme-Linked Immunosorbent Assay (ELISA) (Piao et al., 2017). We performed two measurements and took the average of the two results for each patient. Ab83366 kit, Ab108837 kit, and Ab108911 kits (Abcam Company, Cambridge, United Kingdom) were used for iron, ferritin, and transferrin, respectively. The E01L0224 kit (Shanghai Lanji Biological Limited Company, Shanghai, China) was used for lactoferrin.

## Detections of the Levels of Inflammatory Factors in CSF and Serum

The levels of inflammatory factors, including IL-6, IL-1β, and PGE<sup>2</sup> in the CSF and serum were measured using ELISA (Hu et al., 2015). A 1R040 kit (RB Company, Shanghai, China) was used for IL-1β. A human IL-6 ELISA kit (Beijing DOP Biotechnology Co., Ltd, Beijing, China) was used for IL-6. A CSB-E07965h kit (CUSABIO Company, Wuhan, China) was used for PGE2. The levels of free radicals, including NO and H2O2, were measured using the chemical colorimetric method. An A012 kit and A064 kit (Nanjing Jiancheng Biological Engineering Research Institute, Nanjing, China) were used for NO and H2O2, respectively.

### Data Analyses

Statistical analyses were conducted using SPSS Statistics 20.0 software (IBM Corporation, Armonk, New York, United States).

The normally distributed continuous variables were compared using a two-tailed t-test. The abnormally distributed continuous variables were compared using a non-parametric test. Discrete variables were compared using a Chi-square test. A P-value of less than an alpha level of 0.05 was defined as statistically significant.

Demographic variables were compared between the PD-TD and PD-PIGD groups. Multiple comparisons of the levels of iron and its metabolism-related proteins and inflammatory factors in the CSF and serum were compared among the control, PD-PIGD, and the PD-TD groups by using the Kruskal–Wallis test. A Bonferroni correction was made in the comparison of multiple inflammatory factors or iron/ferritin among the control, PD-PIGD and the PD-TD groups to control the family wise error rate. In order to adjust for the multiple tests, we set the significance to 0.05/6 in the comparison of the levels of iron and its metabolismrelated proteins in the CSF or serum, and we set the significance to 0.05/5 in the comparison of the levels of inflammatory factors in the CSF or serum. Post hoc analyses were made by using the Bonferroni correction. We set the significance to 0.05/3 in the pair-wise of each group.

The Spearman correlation was used to test the association between iron and its metabolism-related proteins and inflammatory factors in the PD-TD group.

# RESULTS

## Demographic Variables and Clinical Characteristics of PD-TD and PD-PIGD Groups

Among the 135 PD patients, 74 cases (54.81%) and 39 cases (28.89%) were classified into the PD-TD group and the PD-PIGD group, respectively.

Demographic variables, including sex, age, education level, age at disease onset, disease duration, vascular risk factors like hypertension, diabetes, hyperlipidemia and smoking, LEDD, and the number of patients under dopaminergic treatment in the PD-PIGD and PD-TD groups were compared (**Table 1**). The results demonstrated no significant differences in the above demographic variables between the two groups. More analyses of the correlation of the demographic variables with the following iron metabolism/inflammation markers can be seen in **Supplementary Tables 1**, **2**.

The PD-TD and PD-PIGD groups showed no significant difference in Hoehn-Yahr stage (**Table 1**), suggesting that the two groups did not differ in PD severity.

The UPDRS III score was significantly lower in the PD-TD group than in the PD-PIGD group (**Table 1**), indicating that the PD-TD group had significantly less severe motor symptoms than the PD-PIGD group.

## The Comparation of Iron and Iron Metabolism-Related Protein Between PD-TD and PD-PIGD Groups in Both Central and Peripheral Systems

The levels of iron, ferritin, H-ferritin, L-ferritin, transferrin, and lactoferrin in the CSF and serum were compared between the control, PD-PIGD and PD-TD groups (**Table 2**). Further comparation of the iron and its metabolism-related protein between the PD-PIGD and PD-TD groups of drug-naive patients can be seen in **Supplementary Table 3**.

In the CSF, iron levels in the PD-TD group were significantly higher than those in both the control and PD-PIGD groups (P < 0.001, P = 0.045, respectively). L-ferritin levels in the PD-TD

TABLE 1 | Clinical variables of PD-PIGD and PD-TD groups.


<sup>∗</sup>P < 0.05.

group were prominently lower than those in the control and PD-PIGD groups (P < 0.001, P = 0.019, respectively). These results were consistent with the comparison of the above factors between the PD-PIGD and PD-TD groups of the drug-naive patients (**Supplementary Table 3**).

In the serum, ferritin level in the PD-TD group was strikingly lower than those in the PD-PIGD and control groups (P < 0.001, P = 0.019, respectively).

#### The Comparation of Inflammatory Factors Between PD-TD and PD-PIGD Groups in Both Central and Peripheral Systems

The levels of inflammatory factors, including IL-1β, IL-6, PGE2, NO, and H2O<sup>2</sup> in the CSF and serum were compared among the control, PD-PIGD and PD-TD groups (**Table 3**). Further comparison of the inflammatory factors between the PD-PIGD and PD-TD groups of drug-naive patients could be seen in **Supplementary Table 4**.

In the CSF, the PD-TD group had significantly enhanced IL-6 levels compared with the PD-PIGD and control groups (P < 0.001, P = 0.020, respectively). In the serum, IL-6 levels in the PD-TD group were strikingly increased compared with the PD-PIGD and control groups (P < 0.001, P = 0.012, respectively). These results were consistent with the comparison of the inflammatory factors between PD-PIGD and PD-TD groups of the drug-naive patients (**Supplementary Table 4**).

Further analysis suggested that IL-6 level in the CSF had a positive and significant correlation with IL-6 levels in the serum (r = 0.257, P = 0.022).

#### The Relationship Between Iron Metabolism and Inflammation in PD-TD Patients

In the CSF, we observed that the IL-6 levels increased alongside elevated iron levels in the PD-TD group (r = 0.308, P = 0.022). In the serum, we found that the IL-6 levels increased alongside enhanced ferritin levels in the PD-TD group (r = 0.410, P = 0.004).

TABLE 2 | The levels of iron and its metabolism-related proteins in CSF and serum from the control, PD-PIGD and PD-TD groups.


P, Kruskal–Wallis test among control, PD-PIGD and PD-TD groups, α = 0.008; P<sup>1</sup> , PD-PIGD group vs. Control group, α = 0.017; P<sup>2</sup> , PD-TD group vs. Control group, α = 0.017; P<sup>3</sup> , PD-TD group vs. PD-PIGD group, α = 0.017. The bold values are the p-values less than α.


TABLE 3 | The levels of inflammatory factors in CSF and serum from the control, PD-PIGD and PD-TD groups.

P, Kruskal–Wallis test among control, PD-PIGD and PD-TD groups, α = 0.008; P<sup>1</sup> , PD-PIGD group vs. Control group, α = 0.017; P<sup>2</sup> , PD-TD group vs. Control group, α = 0.017; P<sup>3</sup> , PD-TD group vs. PD-PIGD group, α = 0.017. The bold values are the p-values less than α.

#### DISCUSSION

In this study, 62.22% of all cases were classified as PD-TD, indicating that tremor is one of the most predominant symptoms of PD. However, the PD-TD group displayed less severe motor symptoms (**Table 1**) indicated by a lower UPDRS III score than the PD-PIGD group, consistent with a previous longitudinal study (Rajput et al., 2017). More widespread biochemical abnormalities in the brain may explain the more severe motor symptoms of PD-PIGD patients (Rajput et al., 2008).

Iron plays an essential role on many biological activities, including DNA synthesis, oxygen transport and mitochondrial respiration (Sian-Hulsmann et al., 2011). However, elevated iron levels can elicit a cascade of cellular deleterious events (Munoz and Humeres, 2012). The consequence of the dysregulation of iron and its metabolism may be the primary cause of PD (Weinreb et al., 2013).

In the PD-TD group, iron levels in the CSF were strikingly increased compared with the PD-PIGD and control group (**Table 2**), indicating that excessive iron in the brain might be associated with PD-TD. Furthermore, the role of the proteins involved in iron metabolism was investigated. As it is known, ferritin, the most common iron storage protein in the brain, can reduce the toxic effects of excessive iron (Henle et al., 1996; Zhang et al., 2014). Ferritin comprises of heavy-ferritin (H-ferritin) and light-ferritin (L-ferritin). H-ferritin converts toxic Fe (II) into less toxic Fe (III) and participates in the absorption and utilization of iron (Harrison and Arosio, 1996). In contrast, L-ferritin contributes to the long-term safe storage of iron (Harrison and Arosio, 1996) and is predominantly expressed in microglia (Connor et al., 2001), which are mostly scavenger cells. In this study, compared with the PD-PIGD and control groups, L-ferritin levels in the CSF in PD-TD group were especially decreased (**Table 2**). In summary, abnormal iron metabolism, especially the decrease of L-ferritin levels, and subsequent excessive iron deposition in the brain, may be related to tremor in PD patients.

The plasma ferritin level was strongly increased in the PD-TD group (**Table 2**). Results of plasma ferritin changes in PD patients are inconsistent (Madenci et al., 2012; Costa-Mallen et al., 2015). Ferritin levels in the serum can be increased with systemic iron overloaded (Kuhn, 2015), but we failed to identify excessive iron in the serum of the PD-TD group. We speculate that the plasma ferritin level may be increased in two ways. On the one hand, PD-TD patients may have a damaged blood-brain barrier, allowing ferritin transport from the CSF to the plasma and inducing an elevation of ferritin in the plasma. Though the PD-TD group had a lower L-ferritin level in the brain, there was no significant difference in the ferritin level between the PD-TD and PD-PIGD groups (**Table 2**). On the other hand, it was found that plasma ferritin can also be increased with body iron deficiency, in the inflammatory condition, due to iron retention in macrophages (Cohen et al., 2010). Additionally, it was demonstrated that the ferritin level was positively correlated with the IL-6 level in the serum of PD-TD patients. Therefore, we speculate that abnormal peripheral iron metabolism may also play a role on PD-TD due to more severe inflammation in the serum.

Peripheral iron can be transferred across the blood–brain barrier via transferrin, lactoferrin and their receptor, as well as divalent cation metal transporters (Ponka, 2004). In this study, compared with the control group, iron and transferrin levels were increased in the CSF, but decreased in the serum, in PD patients (**Table 2**). We hypothesize that these changes may be the result of the abnormal transfer of transferrin from the peripheral to the brain, further leading to the abnormal deposition of iron in PD brains. Though an imbalance of transferrin in the peripheral and central systems was seen in PD patients, we failed to see differences in transferrin and lactoferrin levels in the CSF and serum between the PD-TD and the PD-PIGD groups (**Table 2**). More studies are needed to explore whether iron depositions in the brains of PD-TD patients are resulted from the transfer of iron from the periphery to the central nervous system.

Accumulating evidence reveals a pivotal role of chronic neuroinflammation on the pathologic features of PD

(Collins et al., 2012). Inflammatory mediators, such as NO, TNF-α, IL-1β, and IL-6 may contribute to the progression of PD (Niranjan et al., 2012; Williams-Gray et al., 2016; Kouchaki et al., 2018). This study showed that the PD-TD group had a significantly elevated IL-6 level in the CSF when compared with the PD-PIGD and control groups (**Table 2**). As it is known, IL-6 is a pleiotropic cytokine involved in the acute phase response. It can trigger degeneration and even the death of neurons after injury and infection in both the peripheral and central nervous system (Gadient and Otten, 1997; Jones and Jenkins, 2018). A de novo PD patient has significantly elevated IL-6 levels in the CSF when compared to the control group (Blum-Degen et al., 1995). Moreover, it has been demonstrated that men with elevated IL-6 levels in the serum had a higher risk of PD, but the same influence was not seen in other inflammatory biomarkers, like C-reactive protein, fibrinogen, and TNF-α (Chen et al., 2008). Moreover, a positive correlation has been found between plasma IL-6 levels and the UPDRS III score (Delgado-Alvarado et al., 2017). Some studies fail to certify the same difference of the plasma IL-6 level between PD and control groups and the inconsistency may be related to the recruitment of patients with a shorter disease duration than in the current study (Blum-Degen et al., 1995). Collectively, many existing studies have shown that increased IL-6 levels in both the brain and the serum may contribute to the progression of PD. However, to our knowledge, no studies have examined the role of IL-6 on parkinsonian tremor. This study is the first to reveal that inflammation, especially IL-6 in both the peripheral and central nervous system, may play a critical role on TD of PD patients. Moreover, the spearman correlation analysis further showed that IL-6 levels in the serum were significantly correlated with IL-6 levels in the CSF, indicating that peripheral inflammation exacerbates central inflammation in PD-TD patients.

Similar changes of iron accumulation and inflammation can also be seen in PD-TD and PD-PIGD groups of drugnaive patients, who do not have the following influence of anti-parkinsonian treatment. Different clinical symptoms of PD have a different response to anti-parkinsonian drugs. For example, bradykinesia generally responds very well to levodopa (Nutt et al., 2011), while tremor does not (Yahr et al., 1969). Anti-parkinsonian drugs may produce effects on inflammatory factors, like IL-6, IL-8 and NO (Parrado et al., 2012; Wang et al., 2019). For example, it has been found that DA alone does not regulate the inflammatory factors directly, like IL-6, TNF-α or NO, but it inhibits lipopolysaccharide-induced NO production in microglial cells. In our study, anti-parkinsonian drugs were withdrawn at least 12–14 h before obtaining the CSF and serum samples. Thus, the anti-parkinsonian treatment might have little influence on the results. Therefore, the changes of iron/ inflammation makers of the two groups remain the same in both total PD patients and the drugnaive PD patients.

The interplay between iron accumulation and inflammation in the neurodegenerative process may lead to cell death in PD (Urrutia et al., 2014). Here, positive correlations were found between the IL-6 level and iron level in the CSF and ferritin level in the serum in PD-TD patients, indicating that increased iron depositions in the brain and systemic abnormal iron metabolism worsened as neuroinflammation intensified. As a cross-sectional study in humans, this investigation could not reveal whether abnormal iron metabolism was the cause or the consequence. However, the vicious cycle of inflammation and iron-induced oxidative damage in both the peripheral and central nervous systems might be given considerable weight in contributing to PD-TD. As known, the basal ganglia nuclei are transiently activated at the onset of tremor episodes, whereas activity in the cerebello-thalamocortical circuit fluctuates with tremor amplitude (Helmich et al., 2011) in PD patients. Iron depositions in the dentate nucleus of the cerebellum, red nucleus, and the subcortical nuclei in the cerebello-thalamo-cortical circuit, have been reported to be correlated with tremor in PD (Guan et al., 2017). Moreover, patients with the mutation causing abnormal ferritin metabolism and iron deposition in the globus pallidus presented with symptoms of extrapyramidal dysfunction including tremor (Ponka, 2002). In conclusion, overloaded iron might be involved in the presence of tremor of PD. We put forward the hypothesis that insufficient L-ferritin is unable to handle the overload iron, leading to iron deposition in the brain. Iron deposition and related inflammation are part of a synergistic self-propelling cycle, and result in neuronal death in TD-associated regions, like the basal ganglia and the cerebello-thalamo-cortical circuit, contributing to tremor in PD. However, the role of each inflammatory cytokine on different forms of PD has not been well acknowledged. It has been found that different cytokines may be involved in the pathophysiology of specific motor and non-motor symptoms of PD (Menza et al., 2010; Hu et al., 2015; Pereira et al., 2016). Some researchers postulate the hypothesis that inflammation may accompany PD progress and through the Braak stage, inflammation may move from the peripheral to the central nervous system to destroy different neurons, resulting in different forms of PD (Barnum and Tansey, 2012). Therefore, longitudinal studies are necessary in order to evaluate the impact of inflammatory factors on the motor and non-motor progression in PD.

# CONCLUSION

TD-PD patients exhibit decreased L-ferritin levels and subsequent excessive iron deposition in the brain and increased ferritin levels in the serum, as well as inflammation in both the peripheral and central nervous systems. The interplay between disturbed iron metabolism and inflammation may cast a new light on the mechanism underlying tremor in PD.

This investigation had the following limitations. First, CSF samples were relatively insufficient in healthy controls since most of the subjects were old and some of them had spinal deformities and bone hyperplasia. A further analysis should include more healthy controls. Second, since PD-TD has a higher frequency than PD-PIGD, patients with PD-TD were more than those with PD-PIGD, as expected. The imbalanced sample

size of the two groups may cause a potential bias. In the future, we will recruit PD patients to amplify the sample size, aiming to reduce the bias. Third, the altered levels of iron and its metabolism-related proteins in the serum and CSF from PD-TD patients did not indicate that the iron accumulated in a certain brain region, such as the substantia nigra, therefore, neuroimaging studies are much needed in a further study. Fourth, though most of the PD patients had a significant response to levodopa, we still cannot guarantee that none of them will be diagnosed with parkinsonism, like multiple system atrophy (MSA), 2–3 years later since MSA patients may present symptoms very similar to PD and also have very good response to levodopa within 2–3 years of symptom onset. Hence, 2– 3 years of follow-ups is very important for the confirmation of diagnosis. We will perform a follow-up in a future study. Lastly, as a cross-sectional study, this study had limited grounds for drawing definite conclusions. Moreover, some iron metabolismrelated proteins, including divalent metal transporter 1 and ceruloplasmin were not included in this investigation. Therefore, we will continue to collect more CSF samples from healthy controls and a longitudinal study is being considered to explore the correlation of iron and iron metabolism-related proteins with inflammatory factors in PD-TD patients.

## DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of International Ethical Guidelines for Biomedical Research Involving Human Subjects, Council for International Organizations of Medical Sciences. The protocol was approved by the Ethics Committee of the Beijing Tiantan Hospital. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

# AUTHOR CONTRIBUTIONS

T-HL contributed to drafting the manuscript, study design, analysis of data, accepts responsibility for conduct of research and will give final approval, acquisition of data, and statistical analysis. PG contributed the study design, accepts responsibility for the conduct of research and will give final approval, and acquisition of data. L-JZ, YH, S-YY, Q-JY, ZJ, R-DW, and L-XL

# REFERENCES

Ba, F., Obaid, M., Wieler, M., Camicioli, R., and Martin, W. R. (2016). Parkinson disease: the relationship between non-motor symptoms and motor phenotype. Can. J. Neurol. Sci. 43, 261–267. doi: 10.1017/cjn. 2015.328

contributed to accepts responsibility for the conduct of research and will give final approval, acquisition of data. WZ contributed to the study design, analysis of the data, accepts responsibility for the conduct of research and will give final approval, acquisition of data, and statistical analysis.

# FUNDING

This work was supported by The National Key Research and Development Program of China (2016YFC1306000, 2016YFC1306300), the National Key R&D Program of China— European Commission Horizon 2020 (2017YFE0118800— 779238), The National Natural Science Foundation of China (81571229, 81071015, and 30770745), The Key Project of National Natural Science Foundation of China (81030062), The Key Project of Natural Science Foundation of Beijing, China (B) (kz201610025030), The Key Project of Natural Science Foundation of Beijing, China (4161004, kz200910025001), The Natural Science Foundation of Beijing, China (7082032), National Key Basic Research Program of China (2011CB504100), Important National Science & Technology Specific Projects (2011ZX09102-003-01), National Key Technology Research and Development Program of the Ministry of Science and Technology of China (2013BAI09B03), Project of Scientific and Technological Development of Traditional Chinese Medicine in Beijing (JJ2018-48), Project of Beijing Institute for Brain Disorders(BIBD-PXM2013\_014226\_07\_000084), High Level Technical Personnel Training Project of Beijing Health System, China (2009-3-26), Project of Construction of Innovative Teams and Teacher Career Development for Universities and Colleges Under Beijing Municipality (IDHT20140514), Capital Clinical Characteristic Application Research (Z12110700100000, and Z121107001012161), Beijing Healthcare Research Project, China (JING-15-2), Excellent Personnel Training Project of Beijing, China (20071D0300400076), Natural Science Foundation of Capital Medical University (PYZ2018077), Basic-Clinical Research Cooperation Funding of Capital Medical University, China (2015-JL-PT-X04, 10JL49, 14JL15), Youth Research Funding, Beijing Tiantan Hospital, and Capital Medical University, China (2015-YQN-15, 2015-YQN-14, 2015-YQN-17).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.00255/full#supplementary-material

Barnum, C. J., and Tansey, M. G. (2012). Neuroinflammation and non-motor symptoms: the dark passenger of Parkinson's disease? Curr. Neurol. Neurosci. Rep. 12, 350–358. doi: 10.1007/s11910-012-0283-6

Block, M. L., Zecca, L., and Hong, J. S. (2007). Microglia-mediated neurotoxicity: uncovering the molecular mechanisms. Nat. Rev. Neurosci. 8, 57–69. doi: 10. 1038/nrn2038


Thenganatt, M. A., and Jankovic, J. (2014). Parkinson disease subtypes. JAMA Neurol. 71, 499–504. doi: 10.1001/jamaneurol.2013.6233


**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 Lian, Guo, Zuo, Hu, Yu, Yu, Jin, Wang, Li 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.

fnins-13-00255 March 25, 2019 Time: 18:13 # 9

# Aceruloplasminemia: A Severe Neurodegenerative Disorder Deserving an Early Diagnosis

Giacomo Marchi\*, Fabiana Busti, Acaynne Lira Zidanes, Annalisa Castagna and Domenico Girelli

Department of Medicine, University of Verona, Verona, Italy

Aceruloplasminemia (ACP) is a rare, adult-onset, autosomal recessive disorder, characterized by systemic iron overload due to mutations in the Ceruloplasmin gene (CP), which in turn lead to absence or strong reduction of CP activity. CP is a ferroxidase that plays a key role in iron export from various cells, especially in the brain, where it maintains the appropriate iron homeostasis with neuroprotective effects. Brain iron accumulation makes ACP unique among systemic iron overload syndromes, e.g., various types of genetic hemochromatosis. The main clinical features of fully expressed ACP include diabetes, retinopathy, liver disease, and progressive neurological symptoms reflecting iron deposition in target organs. However, biochemical signs of the disease, namely a mild anemia mimicking iron deficiency anemia because of microcytosis and low transferrin saturation, but with "paradoxical" hyperferritinemia, usually precedes the onset of clinical symptoms of many years and sometimes decades. Prompt diagnosis and therapy are crucial to prevent neurological complications of the disease, as they are usually irreversible once established. In this mini-review we discuss some major issues about this rare disorder, pointing out the early clues to the right diagnosis, instrumental to reduce significant disability burden of affected patients.

#### Edited by:

Giorgio Biasiotto, University of Brescia, Italy

#### Reviewed by:

Xiao-Ping Wang, Shanghai Jiao Tong University, China Giulietta Maria Riboldi, New York University, United States

> \*Correspondence: Giacomo Marchi marchigiacomo@pec.it

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 31 October 2018 Accepted: 21 March 2019 Published: 05 April 2019

#### Citation:

Marchi G, Busti F, Lira Zidanes A, Castagna A and Girelli D (2019) Aceruloplasminemia: A Severe Neurodegenerative Disorder Deserving an Early Diagnosis. Front. Neurosci. 13:325. doi: 10.3389/fnins.2019.00325 Keywords: iron metabolism, iron overload disease, neurodegeneration with brain iron accumulation, rare anemias, Aceruloplasminemia

#### INTRODUCTION

With the increasing awareness of an abnormal local iron accumulation in highly impacting neurodegenerative diseases such as Alzheimer's and Parkinson's disease (for review see references Ke and Ming, 2003; Ward et al., 2014), attention in understanding the peculiarities of iron metabolism in the brain has steadily raised (Raha et al., 2013; Rouault, 2013). CP is a coppercontaining ferroxidase enzyme that plays a key role in cellular iron export, and has been postulated to have a neuroprotective function (Wang and Wang, 2018). In this context Aceruloplasminemia (ACP) represents a paradigmatic disorder highlighting how the loss of CP function causes iron accumulation and neurodegeneration.

Aceruloplasminemia is a rare, adult-onset, autosomal recessive disease caused by mutations in the CP gene, encoding CP. The impairment of CP ferroxidase activity results in pathological cellular iron retention and iron-mediated oxidative damage. The spectrum of clinical manifestations includes mild microcytic anemia, diabetes mellitus, liver disease, retinopathy, and progressive

neurodegeneration due to iron accumulation in the brain and other parenchymal organs (Miyajima and Hosoi, 1993– 2018; Kono, 2013). ACP is variably classified into different subgroups of rare diseases, such as neurodegeneration with brain iron accumulation (NBIA) (Hogarth, 2015), atypical microcytic anemias (Camaschella, 2013; Donker et al., 2014; Brissot et al., 2018), and non-HFE iron overload syndromes (Pietrangelo et al., 2011), depending on the specialists who see the patients for the first time. Apart from few neurologists and hematologists, the awareness of this disease is poor even among other specialists that could intercept the patients at various stages of their clinical history, such as diabetologists and ophthalmologists. Indeed, the commonly reported consistent diagnostic delay illustrates how ACP is under-recognized in most cases (Miyajima and Hosoi, 1993–2018; Vroegindeweij et al., 2015). This represents a major problem, since prompt diagnosis, and treatment are crucial to prevent neurological complications of the disease, which are usually irreversible once established. In this mini-review we outline some major issues about this rare disease, including the limited knowledge about epidemiology and genotypephenotype correlation, as well as the challenges in diagnosis and therapy.

# EPIDEMIOLOGY

Our knowledge on the epidemiology of ACP is largely based on studies in the Japanese population, where the disease was first reported in 1987 by Miyajima and collaborators, through the description of a 52-year-old woman presenting with blepharospasm, retinal degeneration and diabetes mellitus (Miyajima et al., 1987). The same Authors later estimated the prevalence of the disease as being roughly 1:2,000,000 in Japanese individuals born from non-consanguineous marriages. Such estimation was inferred by a study in an adult population where serum CP was systematically measured with subsequent sequencing of the CP gene in individuals having low serum values (Miyajima et al., 1999). This estimation has obvious regional limitations, and cannot be applied in non-Japanese population, where ACP prevalence is actually unknown. Few more than one hundred of patients have been described in the literature so far, with a worldwide geographical distribution (Miyajima and Hosoi, 1993–2018; Kono, 2013; Doyle et al., 2015; Vroegindeweij et al., 2015; Pelucchi et al., 2018; Riboldi et al., 2018; Yamamura et al., 2018). While Japanese patients with neurological involvement still represent the majority (Miyajima and Hosoi, 1993–2018) the second group is represented by Italian patients reported by some referral centers for disorders of iron metabolism (Pelucchi et al., 2018). Nonetheless, ACP has been described also in other European countries (Spain, France, Netherlands, Belgium, Poland), as well as in China and in African American patients (Shang et al., 2006; Parks et al., 2013; Vroegindeweij et al., 2017; Riboldi et al., 2018). The generally low disease awareness, as well as a negative publication bias of single cases, represents possible explanations for the limited number of reports. An effort is needed to increase disease awareness especially among specialists in neurology, hematology, and endocrinology, in order to reduce the unacceptable diagnostic delay that is usually of decades after the appearance of the first biochemical and/or clinical signs of the disease (Miyajima and Hosoi, 1993–2018; Vroegindeweij et al., 2015; Pelucchi et al., 2018).

# PATHOPHYSIOLOGY: RECENT ADVANCES AND UNRESOLVED ISSUES

Aceruloplasminemia is caused by biallelic mutations in the CP gene, a 20 exons gene encompassing approximately 65 kb of DNA, located in chromosome locus 3q24-q25 and encoding CP (Yoshida et al., 1995; Hellman and Gitlin, 2002; Kono, 2013). CP is a single polypeptide chain of 1,046 amino acids that can bind up to six atoms of copper. The crystallographic 3D structure of the protein shows its six structural domains with a three-copper catalytic center that is crucial for its oxidative function (Bento et al., 2007). Copper is incorporated into CP before its secretion and this phase is crucial for its function and stability (Sato and Gitlin, 1991). Indeed, secretion of a copper-deficient CP leads to its rapid degradation in the plasma (Hellman and Gitlin, 2002). Two distinct isoforms of CP are produced by an alternative splicing in exons 19 and 20: a soluble form that is present in the plasma, and a glycosylphosphatidylinositol (GPI)-anchored membrane form (Patel et al., 2000). The soluble isoform is almost exclusively synthetized by hepatocytes, and accounts for near 95% of copper in the plasma (Hellman and Gitlin, 2002). On the other hand, the GPI-anchored membrane isoform is produced and expressed by a number of cells, including brain astrocytes glial cells, hepatocytes, macrophages, pancreatic, and retinal epithelial cells (Kono et al., 2010). The soluble form is involved in NO oxidation and homeostasis (Shiva et al., 2006), while the membrane isoform plays a key role in cellular iron egress (Muckenthaler et al., 2017). Indeed, CP cooperates with ferroportin, the ubiquitous and unique transmembrane protein able to export ferrous iron (Fe2+) from the cells (Drakesmith et al., 2015). This process needs to be completed by copper-mediated oxidation to ferric iron (Fe3+), to ensure the appropriate binding of extracellular iron to transferrin (Miyajima and Hosoi, 1993– 2018). GPI-anchored CP has been also reported to interfere with the modulation of ferroportin activity by hepcidin (Nemeth et al., 2004; Kono, 2013; Aschemeyer et al., 2018), the master regulator of systemic iron homeostasis (Ganz, 2013). Membrane bound CP is therefore essential for ensuring a regular iron handling by various cells, including neurons that need iron for synthesis of neurotransmitters, energetic metabolism, and myelin formation (Miyajima and Hosoi, 1993–2018). According to the current model mainly based on animal experiments (Jeong and David, 2006), early neuronal injury in ACP may be related to the inability to uptake iron from the CP-deficient astrocytes. In advanced ACP, neurodegeneration may be due to iron accumulation and ensuing oxidative damage (Miyajima and Hosoi, 1993–2018; Kono and Miyajima, 2006; Kono, 2013), astrocytes loss and neuronal uptake of alternative and toxic iron sources, i.e., non-transferrin bound iron (NTBI) (Breuer et al., 2000). A basic model of astrocyte/neuronal iron handling

is depicted in **Figure 1**. Lack of CP impairs this model, and, in general, ferroportin-mediated iron export from several cells. While this model explains iron maldistribution in critical cell types of ACP patients (e.g., astrocytes, neurons, hepatocytes, pancreatic, and retinal cells), it does not explain well why total iron stores are increased in such disease. This point needs the

postulation of a ferroportin-mediated increase of dietary iron absorption by intestinal cells. Intestinal ferroportin, expressed on the basolateral membrane, works in synergy with the CP homolog hephaestin, another multicopper oxidase that is responsible for the transition Fe2+→Fe3<sup>+</sup> needed for iron incorporation into plasma transferrin (Vulpe et al., 1999; Muckenthaler et al., 2017). Thus, this mechanism should be unaffected in ACP. Nevertheless, increased iron absorption should imply inappropriately low levels of hepcidin, the main negative regulator of ferroportin activity (Drakesmith et al., 2015; Muckenthaler et al., 2017). Indeed, low level of hepatic hepcidin expression has been reported in an ACP mouse model (Guo et al., 2009), and few ACP patients tested for serum hepcidin levels had levels in the lower range (Kaneko et al., 2010). While the latter finding needs confirmation in larger series, it remains to be explained why lack of CP leads to hepcidin suppression, particularly in patients with typically increased liver iron deposits that should stimulate rather than inhibit hepcidin production (Girelli et al., 2016). The iron-restricted erythropoiesis typically found in ACP (see below), due to impaired iron recycling, could stimulate the production of erythroferrone (ERFE) by bone marrow erythroid precursors, which in turn is a potent inhibitor of hepcidin (Kautz et al., 2014). However, up to now ERFE levels have never been measured in ACP patients. Anyway, things are likely more complicated than expected. For example, even if soluble CP is not able to pass the blood-brain barrier, it is also present in the brain through secretion by epithelial cells of the choroid plexus into the cerebrospinal fluid (Hellman and Gitlin, 2002). Moreover, the role of hephaestin may not be limited to intestinal iron absorption, since HEPH-KO mice have shown a dysregulation of brain iron homeostasis (Jiang et al., 2015).

Overall, it emerges that CP is part of a complex regulatory system that protects brain from either systemic iron deficiency or oxidative damage due to iron overload. In ACP the loss of this regulation results in neurodegeneration, as well as increased iron stores and toxic damage localized in other organs, especially liver, pancreas, and retina. A better understanding of the ACP pathophysiology could result in more efficient treatments, which for the moment are essentially limited to poorly effective iron chelation (see below).

## GENOTYPE-PHENOTYPE CORRELATIONS

Disease-related mutations in the CP gene are usually private, leading to absent or severely reduced ferroxidase activity of CP. To the best of our knowledge 28 missense, 17 frameshift, 13 splicing and 8 nonsense mutations have been described (Miyajima and Hosoi, 1993–2018; Pelucchi et al., 2018). Missense mutations lead usually to impairment of protein stability or alteration in copper binding domains. The majority of cases are due to homozygous mutations (which should prompt to ask for history of consanguinity), but compound heterozygosity has been also described. A milder disease course has been reported in some individuals with a single CP mutation (simple heterozygosity) (McNeill et al., 2008b). Possible explanations for such cases could be a second mutation in CP non-studied regions, in other relevant genes involved in genetic iron overload (compound heterozygosity) (Camaschella, 2005), or a dominant-negative effect of mutant CP through silencing the wild-type CP, i.e., causing retention also of the wild type CP in the endoplasmic reticulum (Kono, 2013). While no clear genotype-phenotype correlation is usually reported, recent observations suggest that some CP genotypes (i.e., homozygosity for Cys338Ser or for Ile991Thr) could be associated to residual ferroxidase activity, resulting in no or less frequent neurological impairment (Pelucchi et al., 2018). On the other hand, homozygosity for the Gly631Arg mutation has been associated in almost all cases to extra-pyramidal symptoms in Caucasian patients (Vroegindeweij et al., 2017).

#### CLINICAL FEATURES

Aceruloplasminemia is the only known iron overload disorder in which brain and systemic iron overload are combined. Neurodegeneration typically involves the dentate nuclei of cerebellum, basal ganglia, and thalamus, and is detectable by MRI showing in these regions diffuse hypointensity in T2<sup>∗</sup> and T2 fast spin echo (FSE) sequences (McNeill et al., 2008a). The classical clinical triad of retinal degeneration, dementia and diabetes is frequently cited, mainly based on observations in Japanese patients (Miyajima et al., 2003). The largest review of clinical manifestations was based on 71 cases, most them being Japanese (Miyajima and Hosoi, 1993–2018). However, recent non-Japanese case series have highlighted a broader clinical and molecular heterogeneity (Vroegindeweij et al., 2017; Pelucchi et al., 2018). According to these descriptions, the clinical presentation of ACP usually leading to the diagnosis by neurologists includes cerebellar signs (dysarthria, ataxia of trunk, and limbs) and involuntary movements (dystonia, chorea, tremors), with an onset between 40 and 60 years of age (Miyajima and Hosoi, 1993–2018). At variance with this "typical" picture in Japanese ACP patients, cognitive-psychiatric changes and extrapyramidal signs are apparently more frequent in Caucasians, with a trend toward an earlier onset (Vroegindeweij et al., 2017). However, cognitive alterations (apathy, memory loss) and behavioral changes have a low specificity and can be often underestimated. When ACP is suspected on the basis of neurological symptoms, a brain MRI with specific sequences must be performed (McNeill et al., 2008a). In such cases, the differential diagnosis mainly includes other NBIAs and Wilson's disease.

On the other hand, increasing reports point out the presence of a mild microcytic anemia as the earliest biochemical sign of ACP, in both Japanese and non-Japanese cases, which can be often traceable in patient's history since childhood (Miyajima and Hosoi, 1993–2018; Pelucchi et al., 2018). Unfortunately, it rarely leads to the diagnosis of ACP in the early presymptomatic stage. Diabetes mellitus (DM) is another classical

manifestation of ACP. It generally presents in the fourth to sixth decade, often in subjects without classical risk factors for diabetes, and requires insulin treatment (Vroegindeweij et al., 2017). As for mild anemia, in most cases of ACP DM is not appropriately recognized as part of a systemic disorder. The mechanism leading to DM in ACP is poorly understood. Of note, iron accumulation appears prevalent in exocrine rather than endocrine pancreas cells (Kato et al., 1997). Liver iron accumulation, reflecting systemic iron overload, is often present in ACP but rarely leads to clinically overt manifestations such as cirrhosis and liver failure (Miyajima and Hosoi, 1993–2018; Pelucchi et al., 2018). When liver biopsy is performed, iron accumulation is prevalent in hepatocytes like in classical HFE-related hemochromatosis, which can also be misleadingly considered in the differential diagnosis (Hellman et al., 2000). Retinopathy is frequently reported in Japanese ACP patients, but a uniform morphological description is lacking and it infrequently causes clinically relevant visual impairment (Miyajima and Hosoi, 1993–2018; He et al., 2007). In non-Japanese case series, retinopathy is far less frequent, and its attribution to ACP per se is often uncertain (Pelucchi et al., 2018). Finally, iron overload in other organs, such as heart and endocrine glands other than pancreas, has been sporadically reported in both Caucasian and Japanese patients, but the true prevalence may be actually under-investigated (Miyajima and Hosoi, 1993–2018; Badat et al., 2015).

The lack of uniform description of ACP cases makes it difficult to draw firm conclusions about the prognosis in these patients. Nonetheless, it is clear that neurological manifestations, when present, have the main impact on the quality of life of the patients and their families.

A summary of ACP main clinical features is depicted in **Figure 2**.

# ACP DIAGNOSIS: INCREASING DIAGNOSTIC SENSITIVITY BY USING THE "BIOCHEMICAL" TRIAD RATHER THAN THE "CLINICAL" TRIAD

As mentioned above, mild microcytic anemia, low transferrin saturation (TSAT), and hyperferritinemia are the first detectable signs of ACP according to all the major case series (Miyajima and Hosoi, 1993–2018; Pelucchi et al., 2018). This biochemical triad is pivotal in the definition of an "atypical microcytic anemia," as opposed to typical cases due to iron deficiency in which serum ferritin levels are consistently low. Such triad may still be due to relatively frequent conditions, i.e., thalassemia syndromes and the so called anemia of inflammation, but once these hypotheses are easily discarded (i.e., by other obvious clinical clues and/or biochemical markers of inflammation), the differential diagnosis should consider a definite number of rare disorders. Besides ACP, they include ferroportin disease, congenital sideroblastic anemias, DMT1 deficiency and atransferrinemia (for extensive reviews see Camaschella, 2013; Donker et al., 2014; Brissot et al., 2018).

In ACP, microcytic anemia and low TSAT are likely due to an impairment of iron recycling leading to an iron-restricted erythropoiesis. The resulting anemia is generally mild likely because other oxidation systems independent from CP can counteract the lack of this protein, allowing a certain degree of availability of iron for erythroid precursors. Anyway, in this condition the diagnosis of ACP long before other clinical manifestations can be easily done by assessing serum CP levels, which are usually undetectable or clearly below the normal range (Kono, 2013). Low CP levels can be detected also in asymptomatic CP heterozygotes, Wilson disease, Menkes disease or hypoproteinemias, which can be distinguished from ACP because of different clinical features before eventually molecular analysis. The majority of CP pathogenic mutations produce a dysfunctional and unstable apo- CP that is rapidly degraded in the plasma resulting in undetectable levels by conventional assays. However, in a minority of cases, serum CP can be only mildly decreased or normal. In such ultra-rare cases, a functional assay that explores the CP ferroxidase activity can be useful (Erel, 1998). This assay has also been reported of prognostic value, since residual ferroxidase activity has been associated to better neurological outcome, although in few observations (Pelucchi et al., 2018). Finally, low hepcidin levels are usually found in ACP patients, helping in the differential diagnosis of iron overload disorders (Kaneko et al., 2010; Girelli et al., 2016).

## TREATMENT OF ACP: STILL UNSATISFACTORY

Up to now, our knowledge on ACP treatment is mainly based on case reports. The most common approach is based on iron-chelating agents. At variance with deferoxamine and deferasirox, deferiprone is the only available iron chelator that is able to cross the blood-brain barrier, due to its low molecular weight and lipophilic properties (Kono, 2013). In general, all these drugs have shown encouraging results in reducing serum ferritin and iron accumulation in the liver (Finkenstedt et al., 2010; Kono, 2013). However, their efficacy on brain iron accumulation and neurological manifestations is controversial. Some studies indicate that they may prevent or slow down neurodegeneration, therefore early initiation of treatment is crucial (Pelucchi et al., 2018). However, the overall efficacy of iron chelation therapy in neurodegenerative disorders with brain iron accumulation is difficult to assess because of the lack of standardized description of clinical and MRI findings in ACP patients. According to a recent review, "there is no compelling evidence of the clinical effect of iron removal therapy on any neurological disorder," including ACP (Dusek et al., 2016). Moreover poor tolerance to iron chelators, even including worsening of anemia due to further iron subtraction for erythropoiesis, has been often reported, limiting the long-term use of such agents that would be required to mobilize iron from the brain (Miyajima and Hosoi, 1993–2018; Mariani et al., 2004; Pelucchi et al., 2018). The efficacy of phlebotomies appears even lower than iron chelation, since CP deficiency likely impairs mobilization of tissue iron stores notwithstanding the induction of marked erythropoietic demand by blood removal. Indeed, failure to reduce liver iron overload, as well as development of neurological symptoms, have been occasionally reported in ACP patients treated with phlebotomies (Hellman et al., 2000; Watanabe et al., 2018). In few cases, iron chelation has been combined with fresh frozen plasma (FFP) administration (aimed to restore CP levels), with transient beneficial effects in a couple of studies (Yonekawa et al., 1999; Poli et al., 2017). Other strategies are based on preventing oxidative tissue damage by administration of vitamin E or zinc sulfate (Pelucchi et al., 2018). The latter has been associated to a dramatic neurological improvement in a patient with extrapyramidal signs and cerebellar movement disorder (Kuhn et al., 2007). Its effects on iron absorption, as well as the good tolerability make this drug a feasible option, also associated to iron chelation therapy (Donangelo et al., 2002). Tetracyclines have in vitro iron-chelating properties and are known to be able to cross the blood-brain barrier (Grenier et al., 2000). Amelioration of neurological dysfunction has been reported in a patient treated with minocycline (Hayashida et al., 2016). Finally, the administration of human CP to CP-knock-out mice was able to pass the blood-brain barrier, rescued brain ferroxidase activity, reduced neuronal death and brain iron deposits, and also ameliorated motor incoordination (Zanardi et al., 2018). Whether or not the use of a recombinant CP treatment could be successful in human ACP remains to be demonstrated.

#### CONCLUDING REMARKS

Aceruloplasminemia is a rare proteiform disorder that can be faced by different specialists at different times. A high degree of suspicion is needed, and a proper early diagnosis is critical. To this end, the biochemical triad of mild anemia with low TSAT and high ferritin levels not due to any obvious alternative explanation is likely the best clue. The disease should be always suspected in any case of unexplained liver iron overload, diabetes mellitus in young adults with no classical risk factors, as well as in adult onset neurological dysfunctions (behavioral changes, psychiatric disorders, extrapyramidal or cerebellar signs) with MRI showing hypointensity in T2 FSE and T2<sup>∗</sup> sequences in dentate nucleus of cerebellum, basal ganglia and thalamus. A better understanding of ACP molecular pathophysiology is needed, possibly leading to novel treatments in alternative to poorly effective current options. Considering the rarity of ACP and the lack of uniformity in cases described, a multicenter international registry should be instrumental to improve knowledge about this highly invalidating iron metabolism disorder.

#### AUTHOR CONTRIBUTIONS

GM wrote the manuscript. FB, ALZ, and AC co-wrote the manuscript. DG critically revised and edited the manuscript.

#### s090744490604947x Breuer, W., Hershko, C., and Cabantchik, Z. I. (2000). The importance of nontransferrin bound iron in disorders of iron metabolism. Transfus. Sci. 23,

185–192. doi: 10.1016/s0955-3886(00)00087-4 Brissot, P., Bernard, D. G., Brissot, E., Loréal, O., and Troadec, M. B. (2018). Rare anemias due to genetic iron metabolism defects. Mutat. Res. 777, 52–63. doi: 10.1016/j.mrrev.2018.06.003

Aschemeyer, S., Qiao, B., Stefanova, D., Valore, E. V., Sek, A. C., Ruwe, T. A., et al. (2018). Structure-function analysis of ferroportin defines the binding site and an alternative mechanism of action of hepcidin. Blood 131, 899–910.

Badat, M., Kaya, B., and Telfer, P. (2015). Combination-therapy with concurrent deferoxamine and deferiprone is effective in treating resistant cardiac ironloading in aceruloplasminaemia. Br. J. Haematol. 171, 430–432. doi: 10.1111/

Bento, I., Peixoto, C., Zaitsev, V. N., and Lindley, P. F. (2007). Ceruloplasmin revisited: structural and functional roles of various metal cation-binding sites. Acta Crystallogr. D Biol. Crystallogr. 63, 240–248. doi: 10.1107/


bjh.13401

REFERENCES

doi: 10.1182/blood-2017-05-786590

fnins-13-00325 April 4, 2019 Time: 18:3 # 7


iron transport, is defective in the sla mouse. Nat. Genet. 21, 195–199. doi: 10.1038/5979


**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 Marchi, Busti, Lira Zidanes, Castagna and Girelli. 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.

# DMT1 Expression and Iron Levels at the Crossroads Between Aging and Neurodegeneration

Rosaria Ingrassia<sup>1</sup> \*, Barbara Garavaglia<sup>2</sup> and Maurizio Memo<sup>3</sup>

<sup>1</sup> Section of Biotechnologies, Department of Molecular and Translational Medicine, University of Brescia, Brescia, Italy, <sup>2</sup> Medical Genetics and Neurogenetics Unit, Fondazione IRCCS Istituto Neurologico Carlo Besta, Milan, Italy, <sup>3</sup> Section of Pharmacology, Department of Molecular and Translational Medicine, University of Brescia, Brescia, Italy

Iron homeostasis is an essential prerequisite for metabolic and neurological functions throughout the healthy human life, with a dynamic interplay between intracellular and systemic iron metabolism. The development of different neurodegenerative diseases is associated with alterations of the intracellular transport of iron and heavy metals, principally mediated by Divalent Metal Transporter 1 (DMT1), responsible for Non-Transferrin Bound Iron transport (NTBI). In addition, DMT1 regulation and its compartmentalization in specific brain regions play important roles during aging. This review highlights the contribution of DMT1 to the physiological exchange and distribution of body iron and heavy metals during aging and neurodegenerative diseases. DMT1 also mediates the crosstalk between central nervous system and peripheral tissues, by systemic diffusion through the Blood Brain Barrier (BBB), with the involvement of peripheral iron homeostasis in association with inflammation. In conclusion, a survey about the role of DMT1 and iron will illustrate the complex panel of interrelationship with aging, neurodegeneration and neuroinflammation.

Keywords: iron homoeostasis, neurodegenerative diseases, aging, transport of iron and heavy metals, divalent metal transporter 1 up-regulation, non-transferrin bound iron transport, neurodegeneration with brain iron accumulation

# INTRODUCTION

Interest in the contribution of iron and DMT1 to both aging and neurodegeneration comes from relevant studies performed during the course of the last 20 years. As Knutson (2017) importantly highlighted, last year was the 20th anniversary of the discovery of DMT1, the first reported mammalian divalent metal transporter. DMT1 principally mediates the transport of ferrous iron and heavy metals in systemic iron homeostasis, from the plasma membrane or endosomes to the intracellular labile pool, mostly sustaining Non-Transferrin Bound Iron transport (NTBI). Although according to recent highlights, the uptake of NTBI into the liver seems to be mediated by ZIP14/SLC39A14 (Jenkitkasemwong et al., 2012) and hepatocyte DMT1 was shown to be dispensable for hepatic iron accumulation (Wang and Knutson, 2013). Interestingly, <sup>59</sup>Fe-NTBI uptake was impaired in ZIP14/SLC39A14 null mice when fed with iron-loaded diet or crossed with mouse models of hereditary hemochromatosis, while DMT1 siRNA reduced iron loading in a mouse model of hereditary hemochromatosis (Wang et al., 2019), thus highlighting the role of both NTBI transporters in the complex homeostasis of iron overload, with cerebral

#### Edited by:

Massimiliano Filosto, Azienda Socio Sanitaria Territoriale degli Spedali Civili di Brescia, Italy

#### Reviewed by:

Torben Moos, Aalborg University, Denmark Gladys Oluyemisi Latunde-Dada, King's College London, United Kingdom Gabriela Alejandra Salvador, Universidad Nacional del Sur, Argentina

#### \*Correspondence:

Rosaria Ingrassia rosaria.ingrassia@unibs.it

#### Specialty section:

This article was submitted to Neurodegeneration, a section of the journal Frontiers in Neuroscience

Received: 25 October 2018 Accepted: 20 May 2019 Published: 05 June 2019

#### Citation:

Ingrassia R, Garavaglia B and Memo M (2019) DMT1 Expression and Iron Levels at the Crossroads Between Aging and Neurodegeneration. Front. Neurosci. 13:575. doi: 10.3389/fnins.2019.00575

implications. Noteworthy, the comparison of these two NTBItransporters activity, in RNA injected Xenopus oocytes, showed a maximum at pH 7.5 for mouse ZIP14, compared to human DMT1 (Pinilla-Tenas et al., 2011), a proton co-transporter, shows an optimum ferrous iron uptake at pH 5.5 (Garrick et al., 2006; Mackenzie et al., 2007), which could exacerbate iron uptake during the pathological-associated extracellular acidosis. Other non-transferrin-bound iron transporters were later identified, such as ZIP14 (SLC39A14), ZIP8 (SLC39A8), L-type calcium channels (LTCCs) and T-type calcium channels (TTCCs), and TRPC6 (Martinez et al., 1999; Mwanjewe and Grover, 2004; He et al., 2006; Liuzzi et al., 2006; Jenkitkasemwong et al., 2012; Striessnig et al., 2014). However, here attention will be given into the complex structure and function of DMT1, underpinning the equilibrium of iron and heavy metals homeostasis, as an essential prerequisite for normal metabolic and neurological functions in relationship to heterogeneous pathophysiological regulation. In this regard, iron and DMT1 significantly contribute to the development of different neurodegenerative diseases, assuming the role of a common participant, as later described. In this review, we discuss the growing body of findings and the emerging concepts about DMT1 and NTBI involvement in both physiological aging and neurodegeneration/neuroinflammation, as found in Parkinson's disease (PD), ischemia, Neurodegeneration with Brain Iron Accumulation (NBIA), and Alzheimer's disease (AD). To this purpose, focus will be given to the potential role of DMT1 as an interesting target for the development of future neuroprotective strategies in the field.

#### FERROUS IRON TRANSPORTER DMT1 AND REDOX EQUILIBRIUM IN THE LIFE-SUSTAINING CELLULAR HOMEOSTASIS

Divalent Metal Transporter 1 (or Nramp2 gene) was first isolated by low stringency Nramp1 homology screening (Gruenheid et al., 1995) and expression cloning of a rat duodenal cDNA library (Gunshin et al., 1997). Subsequently, Lee et al. (1998) firstly defined the structure of the 30UTR splice variant without-IRE and Kishi and Tabuchi (1997) performed functional characterization in yeast of human DMT1 cDNA (Tabuchi et al., 1999). DMT1 expression and function are strictly dependent on this complex structure, with four different isoforms, generated by two alternative splicing (Garrick et al., 2006; Mackenzie et al., 2007). The 50UTR splicing produces two different promoter regions, 1A and 1B. The 1A promoter is responsive to hypoxia in PC12 rat cells (Lis et al., 2005) and HIF-2 alpha in Caco-2/TC7 cells (Mastrogiannaki et al., 2009). Conversely, the 1B isoform is responsive to NF-κB in P19 mouse embryonic carcinoma cells, mouse primary cortical neurons (Paradkar and Roth, 2006a,b; Ingrassia et al., 2012) and to HIF-1 alpha in HepG2 cells (Wang et al., 2010; Qian et al., 2011). The second splicing, at 3 <sup>0</sup>UTR, implies that either 1A and 1B isoforms possess or do not possess an Iron Responsive Element (IRE), that is sensitive to feedback regulation by intracellular iron levels, through IRE/IRP system (Hubert and Hentze, 2002; Pantopoulos, 2004). Moreover, in Caco-2 cell line, both 1A isoforms and (+) IRE 1B isoform show the predicted mRNAs up- and down-regulation in iron deficiency or overload, respectively, while the (–) IRE 1B isoform is regulated by iron-independent mechanisms (Hubert and Hentze, 2002). Particularly, Hubert and Hentze showed that, in Caco-2 cell line, the Transferrin Receptor (TfR) and (+) IRE DMT1 isoform are down-regulated by intracellular iron overload, due to the canonical IRE/IRP post-transcriptional control. In this respect, while (+) IRE DMT1 isoforms do not contribute to the increased uptake of ferrous iron during iron overload, (–) IRE DMT1 isoforms are not influenced by intracellular iron perturbation. Conversely, (–) IRE DMT1 can be regulated at the transcriptional level (Paradkar and Roth, 2006a,b; Ingrassia et al., 2012; **Figure 1**), post-translationally, via impairment of proteasome degradation (Garrick et al., 2012; Jia et al., 2014) and by autophagy (Ingrassia et al., 2017). In this regard, IRP1, IRP2, (+) IRE and (–) IRE DMT1 proteins are up-regulated in lesioned rat hippocampus after kainate treatment and expressed in GFAP positive astrocytes (Huang et al., 2006); the (–) IRE DMT1 increased expression leads to argue that mechanisms other than IRP regulation modulate DMT1 expression.

Importantly, (–) IRE DMT1, which co-localizes with TfR on early endosomes, is also involved in the life-sustaining Transferrin Receptor cycle (Tabuchi et al., 2002, 2010), where the release of ferric iron from transferrin is triggered by endosomal acidic pH, followed by its reduction to ferrous iron with loading by DMT1, whose activity is maximally stimulated at low pH (Garrick et al., 2006; Mackenzie et al., 2007).

Divalent Metal Transporter 1 isoforms play multiple roles and are characterized by complex subcellular distribution; in fact, they localize at the cell membrane, in the cytoplasm and at the nuclear and outer mitochondrial membranes (Roth et al., 2000; Lis et al., 2004; Wolff et al., 2014a,b), where they were recently found to play a role in mitochondrial iron and manganese uptake (Wolff et al., 2018). Through this heterogenous distribution, a tight control of ferrous iron homeostasis is guaranteed at intracellular level, avoiding, for example, excess of ferrous iron that may catalyze the formation of free radicals through the Haber–Weiss and Fenton reactions, and lead to cell damage (Connor and Ghio, 2009). In this context, pharmacological manipulation of the Fenton reaction, by preventing cellular oxidative damage and the formation of hydrogen peroxide and hydroxyl radicals, could represent a useful targeting system leading to potential therapeutic strategies in the treatment of neurodegenerative disorders (Youdim, 2018). Moreover, a physiological pathway for NTBI uptake and regulation of systemic iron homeostasis in different tissues has been extensively described (Knutson, 2018). In this scenario, DMT1 supports a crosstalk between peripheral districts and the central nervous system, with iron transport into the blood stream at the Blood Brain Barrier. According to previous results, DMT1 gene expression in the mouse brain presumably is mostly accounted for by 1B isoform, with both (+)/(–) IRE splicing, which has also been reported to be expressed in rat brain. Whereas no evidence for expression of

1A isoform was found (Hubert and Hentze, 2002; Ke et al., 2005). DMT1 mRNA was also found up-regulated in the cerebellum of ceruloplasmin-knockout mice, particularly in Purkinje and deep nuclei neurons (Jeong and David, 2006), in the frontal cortex of both wild-type and APPSWE/PS11E9 Alzheimer mouse model (Xian-hui et al., 2015), and (+)/(–) IRE DMT1 isoforms significantly increased in rat cortex, striatum, hippocampus and substantia nigra (Lu et al., 2017). DMT1 is also controversially reported to be expressed not only in neurons, but also in nonneuronal cells, such as astrocytes, microglia, oligodendrocytes, and over the brain capillary endothelial cells and choroid plexus cells, that form the brain barrier (Moos and Morgan, 2004; Mills et al., 2010; Skjørringe et al., 2015). Indeed, rat brain capillaries in vivo and isolated endothelial cells, with Blood Brain Barrier (BBB) competence, express DMT1, transferrin receptor, ferroportin, ceruloplasmin, ferrireductases STEAP2-3, and hephestin. This complex machinery sustains the distribution of iron in the central nervous system with epithelial cells of the choroid plexus expressing large amounts of DMT1 mRNA, in order to regulate metal transport across the BBB (Yang W.M. et al., 2011; Burkhart et al., 2016). Divalent metal transport was also studied in cerebral hemorrhage models subjected to iron chelation by deferoxamine (Li et al., 2017), in induced pluripotent stem cell (iPSC)-derived brain endothelial cells (huECs) and a human BBB cellular model where transferrin, hepcidin, and DMT1 sustain iron transport and release (Chiou et al., 2018).

## DMT1 AND IRON UP-REGULATION DURING AGING

Several studies demonstrated how (–)/(+)IRE DMT1 mRNAs expression levels are significantly increased during aging in rat brain cortex, hippocampus, striatum and substantia nigra (Ke et al., 2005; Lu et al., 2017). The authors of these studies showed how (–)/(+)IRE DMT1 mRNAs have a peculiar compartmentalization in both early development and aging, with significant up-regulation at 3 post-natal weeks in the cortex and at 28 post-natal weeks in the substantia nigra, that is surprisingly not influenced by dietary intake in rats, in spite of the expected iron-dependent response. DMT1 mRNA was then found up-regulated in the cerebellum of 24-month-old ceruloplasmin-knockout mice, and DMT1 protein was found increased in Purkinje and deep nuclei neurons (Jeong and David, 2006). DMT1 mRNA was also higher in the frontal cortex of 12 month-old wild-type mice and, with significant increase, in the APPSWE/PS11E9 Alzheimer mouse model (Xian-hui et al., 2015). However, (+)/(–) IRE DMT1 isoforms were found significantly up-regulated at the protein level in the cortex, striatum, hippocampus and substantia nigra of 3-, 12-, and 24-month-old rats, where a significant increase of hepcidin was also described by immunofluorescence (Lu et al., 2017), potentially sustaining a block of iron export. Interestingly, DMT1 up-regulation in the substantia nigra during aging links the transporter to the pathogenesis of other neurodegenerative diseases, such as PD, Parkinsonisms and NBIA, in which enhanced iron accumulation into the basal ganglia occurs. In this regard, in an NBIA mouse model with a mutation in phospholipase A2 beta (PLA2G6), aging significantly up-regulates IRPs and (+) IRE DMT1 in the cortex, striatum, substantia nigra and cerebellum of 100 weeks-old mice, with respect to age-matched wild-type controls (Beck et al., 2015). Furthermore, beside DMT1 increase during aging in the central nervous system, which is highly subjected to damage by iron-dependent oxidative stress, hepatic DMT1 protein was also found up-regulated by Deferoxamine treatment in 24-month-old rats, when compared to 3-month-old rats (Bloomer et al., 2014).

## DMT1 AND IRON UP-REGULATION IN THE NEURODEGENERATION

Oxidative stress leads to DNA damage and to the polymerization and denaturation of proteins that, together, can form insoluble structures typically known as plaques, a hallmark of neurodegenerative diseases (Kell, 2009), with altered cellular proteostasis. Arising from the previous reasons, the intracellular iron level and traffic need a tight control to avoid perturbations in

the expression of DMT1, which may induce downstream cellular damage and dysfunctions in iron metabolism, contributing to the pathogenesis of several neurodegenerative disorders (Biasiotto et al., 2016). In this respect, DMT1 mRNA localization in rat central nervous system by in situ hybridization showed a restricted pattern throughout different areas, as reported by Gunshin et al. (1997), with prominent DMT1 labeling in cerebellar granule cells, hippocampal pyramidal and granule cells in the preoptic nucleus and pyramidal cells of the piriform cortex, as well as in the substantia nigra. The authors also detected DMT1 labeling in the ventral area of the anterior olfactory bulb and in the olfactory epithelium, an important route for the delivery of environmental metals to the brain. The intranasal drug delivery may exert highly efficient effects in the central nervous system (Henkin, 2011) and at the circulatory level, inducing an effective systemic immunity (McGhee, 2011). In fact, the olfactory epithelium can be considered as a gateway for vaccines or peptide hormones administration, also in light of the high neuroprotective potential showed by intranasal administration of iron chelators (Guo et al., 2015). Accordingly, intranasal administration of neurotoxins in rodents, such as 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), has been used to develop an animal model suitable for testing neuroprotective drugs against PD (Prediger et al., 2011).

An important aspect concerns the role of DMT1 in the pathogenesis of several neurodegenerative diseases induced by known neurotoxic stimuli such as kainate, N-methyl-Daspartate (NMDA), agonist of the NMDA Receptors of the family of ionotropic glutamate receptors, levodopa or L-3,4 dihydroxyphenylalanine (L-DOPA), 1-methyl-4-phenyl-1,2,3,6 tetrahydropyridine (MPTP), 6-hydroxydopamine (6-OHDA) and hypoxia, as discussed later. DMT1 also plays a role in neuronal cell death in the hippocampus by excitotoxic glutamatergic stimuli, such as kainate and NMDA. DMT1, in fact, was shown to be up-regulated by kainate in rat hippocampus, both by an IRP-dependent and -independent mechanism, being the (+)IRE and the (–)IRE isoforms upregulated and mostly expressed in GFAP positive astrocytes (Huang et al., 2006). The kainate-dependent DMT1 upregulation and hippocampal injury were then associated with higher cerebral levels of lead and cadmium in rats, when treated with these metals dissolved in the drinking water (Ong et al., 2006). Interestingly, other reports suggested a possible role of DMT1 in the hippocampal spatial memory formation, due to the evidence of NMDA-dependent plasticity induction by 1B/(+)IRE DMT1 in rat primary hippocampal neurons, and evidence of its block by the transcription inhibitor Actinomycin D and the NMDA receptor antagonist MK-801 (Haeger et al., 2010). Thus, NMDA receptor can influence neurodegenerative diseases through DMT1 enhancement, with consequent iron overload-dependent NMDA-Receptor mediated synaptic plasticity. The NMDARs-dependent iron overload also challenges increased lysosomal iron release. In this regard, it was well reported that (–)/(+)IRE DMT1 isoforms have different subcellular localization (Tabuchi et al., 2002, 2010), characterized by peculiar sorting and recycling kinetics into endolysosomal compartment (White et al., 2016; Xu et al., 2017).

In fact, (+)IRE DMT1 has prevalent surface expression, it is internalized from the plasma membrane without efficient recycling and targeted to lysosomes, while (–) IRE DMT1 is efficiently sorted to recycling endosomes upon internalization (Lam-Yuk-Tseung and Gros, 2006). Moreover, it was shown that expression of (–) IRE DMT1, increased by L-DOPA, could have a role in the development of L-DOPA toxicity, significantly counteracted by (–) IRE DMT1 silencing in primary cortical neurons (Du et al., 2009). Again, iron and DMT1 are involved in the pathogenesis of neurodegenerations with multifactorial origin, such as idiopathic PD, as shown by Salazar et al. (2008). They reported an increase of iron and DMT1 levels in the substantia nigra of post-mortem brain of PD patients and in the ventral mesencephalon of the PD mouse model of 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) intoxication. Furthermore, the authors importantly showed that the mk/mk microcytic mice and the Belgrade rat model, both harboring a DMT1 point mutation G185R with consequent impairment of iron uptake (Fleming et al., 1997, 1998), were protected from neurotoxicity induced by MPTP or by 6-hydroxidopamine (6-OHDA), respectively, thus highlighting an iron- and DMT1-dependence of the toxins mechanism of action. Again, in MES23.5 dopaminergic neurons and in the substantia nigra of the 6-OHDA lesioned rats, as a PD model, iron accumulation was associated to NMDA receptors activation, through the up-regulation of (+)IRE DMT1 and down-regulation of the iron exporter ferroportin (FPN1) (Xu et al., 2018). In this model, iron increase was counteracted by the non-competitive NMDA antagonist MK-801 and by the selective NMDA antagonist (2R)-amino-5-phosphonovaleric acid (AP5). Accordingly, iron and DMT1 protein accumulation were also found in the substantia nigra of the NF-κB/c-Rel knockout mice, a model of neurodegeneration with Parkinsonism (Baiguera et al., 2012). Ferrous iron and DMT1 up-regulation are also linked to neuroinflammatory signaling pathways, downstream to the NF-κB/RelA Lys310 acetylation-mediated cell death, during the early phase of brain ischemia, as demonstrated in the well-studied models of in vivo mouse brain ischemic neurodegeneration, subjected to transient middle cerebral artery occlusion tMCAO, and in mouse primary cortical neurons subjected to oxygen-glucose-deprivation (OGD) (Ingrassia et al., 2012). Moreover, (–) IRE DMT1 was shown to contribute to increased ferrous iron uptake subsequent to treatment with 1 methyl-4-phenylpyridinium, MPP (+), in the MES23.5 model of dopaminergic neurons, fully antagonized by iron chelation with Desferal (Zhang et al., 2009), known to prevent dopaminergic neuronal death in MPTP-treated mice (Guo et al., 2016). Incidentally, the conditional overexpression of H-ferritin in TH positive neurons of the substantia nigra counteracted the increased level of iron and DMT1 in the transgenic mice, upon treatment with the proteasome inhibitor lactacystin, thus providing a genetic model of iron chelation (Zhu et al., 2010). In this regard, not only the epigenetic modulation but also the post-translational proteasomal degradation of DMT1 plays a role in DMT1-associated neurodegeneration. In fact, the E3-ubiquitin ligase Parkin, when overexpressed in SH-SY5Y neuroblastoma cells, increased proteasomal degradation of 1B

DMT1 for both (+) IRE and (–) IRE isoforms (Roth et al., 2010). The authors also showed an accumulation of both (+) IRE and (–) IRE DMT1 isoforms in human lymphocytes of early onset patients with familial PD, bearing homozygous deletion of Parkin exon 4, and in the mouse brain of the E3 ubiquitin ligase Parkin knockout. Incidentally, another form of familial PD was identified with a mutation in the gene coding for the Vacuolar Protein Sorting 35, VPS35 (Zavodszky et al., 2014), which is responsible for the correct endosomal recycling and trafficking of DMT1 (Tabuchi et al., 2010), underpinning the derangement of DMT1 in autophagosomal localization as a basal mechanism involved in the disease.

The recent development of a transgenic mouse model of DMT1 overexpression has highlighted the relevant point that DMT1 overexpression alone is not enough to cause neurodegeneration (Zhang et al., 2017). In fact, DMT1 transgenic mice showed selective iron accumulation in the substantia nigra at 18 months of age, only when fed with an iron-supplemented diet. DMT1 expressing mice also showed increased levels of Parkin, as a compensative neuroprotection mechanism. However, DMT1 overexpression in the double mutant mice with Parkin null background, generated to avoid Parkin neuroprotective effect, showed no vulnerability against dietary iron intake, while exposure to 6-hydroxydopamine led to increased neurotoxicity.

Interestingly, with respect to brain iron compartmentalization in PD, post-mortem brain sections of PD patients showed a peculiar redistribution, with decreased levels of iron and iron transporters in the temporal cortex, if compared with the substantia nigra (Yu et al., 2013). Furthermore, several reports have recently highlighted that, during aging, iron and DMT1 accumulate in the frontal cortex of the APPSWE/PS11E9 transgenic mouse model (Xian-hui et al., 2015), where both (–)/(+) IRE DMT1 were also increased in the cortex and hippocampus compared with wild type-control (Zheng et al., 2009), potentially contributing to increasing to the risk of developing AD. Interestingly, (–)/(+) IRE DMT1 up-regulation in the cortex and hippocampus of the APPSWE/PS11E9 transgenic mouse model was recently shown as a consequence of downregulation of Ndfip1, the NEDD4 family interacting protein 1 (Tian et al., 2018), which modulates DMT1 degradation through ubiquitination pathway (Foot et al., 2011). Again, (–)IRE DMT1 up-regulation was associated with a quantitative, age-dependent increase of heavy metals in the frontal cortex of a "natural" rodent model of AD, named Octodon degus (Braidy et al., 2017), with impaired lysosomal function. Interestingly, in support for the presence of common mechanisms involving iron metabolism in different neurodegenerative diseases, the knockout mice lacking the ferroxidase Ceruloplasmin at 6 months of age showed Parkinsonian neurodegeneration with nigral iron accumulation and neuronal loss, which was prevented by iron chelation (Ayton et al., 2014). However, more recently, it was reported that Alzheimer models were induced in the knockout mouse of Ceruloplasmin either by injection of Aβ25−<sup>35</sup> into the lateral ventricle of the brain or by transgenic APP expression, with a consequential increase in memory impairment and hippocampal iron accumulation mediated by (–) IRE DMT1 and changes in ROS levels (Zhao et al., 2018).

In conclusion, these findings highlight the role of DMT1 in the pathogenesis of several, multifactorial and genetic neurodegenerative diseases induced by known neurotoxic stimuli.

# NEURODEGENERATION WITH BRAIN IRON ACCUMULATION (NBIA)

Neurodegeneration with Brain Iron Accumulation (NBIA) encompasses a heterogeneous group of rare disorders characterized by abnormal progressive iron accumulation in specific brain areas. Their prevalence is estimated between 1 and three individuals in a million (Gregory and Hayflick, 2013). Clinical signs include dystonia, tremor, bradykinesia, rigidity, postural instability, optic atrophy or retinal degeneration, neuropsychiatric abnormalities ranging from rapid neurodevelopmental regression in infancy to minor cognitive impairment in adulthood (Gregory and Hayflick, 2013). To date, 12 disease-associated genes leading to NBIA have been identified; however, around 20% of cases are still genetically undefined (Arber et al., 2016; Levi and Tiranti, 2019). Only two genes, namely FTL (ferritin light chain) and CP (ceruloplasmin), are directly associated with iron homeostasis (Levi and Rovida, 2015; Kassubek et al., 2017). While the other ten genes encode proteins either with various functions in lipid metabolism, lysosomal activity and autophagic processes or with still unknown mechanisms: PANK2 (Pantothenate kinase 2), COASY (Coenzyme A synthase), PLA2G6 (Phospholipase A2), C19orf12 (Chromosome 19 open reading frame, with transcript of unknown function), FA2H (fatty acid 2-hydroxylase), ATP13A2 (ATPase cation transporting 13A2), WDR45 (WD repeat containing protein 45, beta-propeller protein), DCAF17 (DDB1 and CUL4-associated factor 17), SCP2 (sterol carrier protein 2) and GTPBP2 (GTP binding protein 2). Although demonstration of iron accumulation in the brain is essential for the diagnosis of NBIA, the impairment of iron metabolism in the different NBIA subclasses has not been clarified. Ingrassia et al. (2017) have recently shown an altered pattern of iron transporters with iron overload in patients' fibroblasts with WDR45 mutations, a NBIA gene with a predicted role in autophagy (Hayflick et al., 2013). PLA2G6 knockout mice (Beck et al., 2015) and fibroblasts of a patient with Neuroferritinopathy (Barbeito et al., 2010) also showed altered expression of iron transporters with DMT1 up-regulation.

# DMT1 AND SIGNALING CONTRIBUTIONS TO NEURODEGENERATION

Of abiding interest is the influence of inflammatory signaling pathways consequential upon heavy metal intoxication, which induces cytokine production and pro-inflammatory stimuli activation. Inflammation, concomitant to oxidative stress, is present in several neurodegenerative diseases. Noteworthy, the NF-κB signaling induces a transient rise of intracellular iron

immediately after a short stimulation with Lipopolysaccharide (LPS), with consequential NF-κB nuclear activation already present at 15 min, as revealed in hepatic macrophages by electromobility shift assay, with a kB specific probe (Xiong et al., 2003). Intriguingly, later findings showed how LPS and NF-κB can regulate specific DMT1 isoforms, as later discussed, with apparently divergent responses at the post-translational level, due the activity of both NEDD4 family interacting protein 1 (Ndfip1) and E3 ubiquitin ligase Parkin. This critical regulatory system may represent a useful target in the prevention of metal accumulation, by DMT1 post-translational degradation, and neuronal cell death. In fact, overexpression of Ndfip1 was demonstrated to have a neuroprotective effect from metal toxicity with ubiquitination and degradation of DMT1 in human primary cortical neurons (Howitt et al., 2009) and after acute cortical brain injury (Sang et al., 2006). Accordingly, Ndfip1 knockout mice revealed a significant increase of intestinal DMT1 with higher serum iron levels, transferrin saturation and inflammation (Foot et al., 2011). Moreover, ferrous iron and LPS induced Ndfip1 protein expression in MES23.5 dopaminergic cell lines, were counteracted by iron chelation with deferoxamine (Xu et al., 2013). Xu and colleagues described Ndfip1 regulatory mechanism mediated by NF-κB activation, in agreement with previous results obtained with LPS stimulation (Xiong et al., 2003), demonstrating how BAY11-7802, an inhibitor of NF-κB activation, significantly decreased mRNA expression levels of Ndfip1 in response to both ferrous iron and LPS.

However, while Ndfip1-mediated post-translational regulation was mostly found associated to pan-DMT1 response, a parallel DMT1 isoform-specific, translational and posttranslational, regulation has arisen. In fact, epigenetic studies contributed to elucidate the acetylome-dependent NF-κB regulation of 1B DMT1 promoter. Particularly, the acetylation of NF-κB/RelA at Lys310 activates the NF-κB-dependent response, a well-studied epigenetic regulation in post-ischemic injury (Lanzillotta et al., 2010), with consequent 1B/(–)IRE DMT1 upregulation in models of differentiated human neuroblastoma and in mouse primary cortical neurons exposed to OGD, or transient middle cerebral artery occlusion (tMCAO) (Ingrassia et al., 2012), and also in undifferentiated P19 embryonic carcinoma cells (Paradkar and Roth, 2006a,Paradkar and Roth, 2006b). In this regard, the isoform-specific regulation of DMT1 could possibly explain the apparent discrepancy emerged between translational and post-translational regulation of Ndfip1. In fact, the Parkin-dependent post-translational degradation of 1B DMT1 isoform was shown to influence metal transport in SH-SY5Y, human neuroblastoma cells, overexpressing wild-type and mutant forms (Park-T240R) of human Parkin, while 1A DMT1 isoform resulted unaffected (Roth et al., 2010). Roth et al. (2010) importantly showed similar findings in human B lymphocyte cell lines derived from a PD patient carrying the homozygous deletion of Parkin exon 4, and in Parkin knockout mice where a clear (–) IRE DMT1 isoform specific degradation occurs, while the (+) IRE isoform expression is not altered. To this purpose, clear-cut data obtained from studies on HEK293 cells highlighted the isoform-dependence in DMT1 post-translational regulation (Garrick et al., 2012).

Taking advantage of the peculiar expression level of 1A DMT1 isoform, which is high in HEK293 cell lines compared to neuronal cells, Garrick and colleagues clearly showed that in HEK293, 1A DMT1 is not a target for Parkin ubiqitination. In fact, transient transfection of Parkin in two HEK293F cell lines, with tetracycline-dependent overexpression of 1A/(+)IRE DMT1 and 1B/(–)IRE DMT1 isoforms respectively, allowed to detect a 1B isoform isoform-specificity of Parkin ubiquitination. In agreement with these findings, the Ndfip1 knockout mice showed a rise of intestinal pan-DMT1 (Foot et al., 2011), a district with predominant expression of 1A DMT1, further supporting the Parkin specificity for 1B/(–)IRE DMT1 isoform. Altogether, these results highlighted the heterogeneous post-translational regulation of DMT1 isoforms. On the other hand, data collected so far clearly claims the need for further studies in this regard. In fact, while Parkin induces DMT1 degradation in several models of PD, the loss of Ndfip1 was associated to DMT1 increase in ventral mesencephalic neuronal cultures of Ndfip1 knockout mice (Howitt et al., 2014). However, Howitt and colleagues concomitantly found increased Ndfip1 levels in the substantia nigra of PD patients, associated with the increase of (+) IRE DMT1. The authors suggested that this DMT1 up-regulation, parallel to Ndfip1 increase, could be accounted to the prevalence of translational regulation, respect to the post-translational one, with an increase of DMT1 isoforms due to iron exposure (Salazar et al., 2008), hypoxia (Lis et al., 2005), and NF-κB activation (Paradkar and Roth, 2006a,Paradkar and Roth, 2006b; Ingrassia et al., 2012).

In addition, signaling derangement with DMT1 misregulation was also presents in Alzheimer models. In fact, the decrease of Ndfip1 in the cortex and hippocampus of APPSWE/PS11E9 transgenic mice was associated with the up-regulation of both (–) and (+) IRE DMT1 isoforms (Tian et al., 2018). Tian et al. (2018) also showed how the down-regulation of DMT1, with reduced iron uptake and decreased Aβ(1– 42) peptide secretion, occurs after transient overexpression of Ndfip1 in SH-SY5Y human neuroblastoma cells stably overexpressing APPsw. Accordingly, in immortalized microglia cells, the Aβ (1–42) treatment induced pan-DMT1 up-regulation (McCarthy et al., 2018). Comprehensively, all these evidences lead to hypothesize a signaling mechanism accounting for the activation of specific ubiquitination pathways in the regulation of DMT1 isoforms.

In conclusion, the complex transcriptional and posttranslational coordination of DMT1 isoforms may be underpinning to divalent metals exposure and neuroinflammation in PD, AD or any other trauma to the CNS. To this purpose, evident signs of gliosis and microglia activation (Toklu et al., 2018), or a mild inflammatory profile at pre-motor stage in the NF-κB/c-Rel−/<sup>−</sup> mouse model of late-onset Parkinsonism, without gliosis at 18 months of age, were described (Porrini et al., 2017). Interestingly, the damage to the substantia nigra in PD induces TNF alpha and other cytokines that bind to their neuronal receptors with consequent NF-κB translocation to the nucleus, 1B DMT1 up-regulation and iron accumulation with cellular damage (Roth, 2009). In PD, iron and (+) IRE DMT1 are up-regulated in the dopaminergic,

neuromelanin positive neurons of the substantia nigra in PD patients (Salazar et al., 2008). In accordance with the previous evidence, the early signs of neuroinflammation present in the NF-κB/c-Rel knockout mice, considered as a model of late-onset Parkinsonism (Baiguera et al., 2012), induced an increase of nigral iron and DMT1 gene expression in the striatum and mesencephalon. An analogous mechanism is involved during the early phase of brain ischemia, both in vitro and in vivo (Ingrassia et al., 2012). Accordingly, pretreatment with Tanshinone IIA, a phenanthrenequinone derivative, with anti-inflammatory potential against the NF-κB pathway, exerted neuroprotection against ischemic and hypoxic damage with down-regulation of DMT1 and TfR in cerebral ischemic rats subjected to in vivo tMCAO (Yang L. et al., 2011).

Furthermore, a close relationship between inflammation and ferrous iron increase, mediated by enhanced DMT1 gene expression, emerges in primary hippocampal astrocytes upon interferon gamma treatment (Pelizzoni et al., 2013). In addition, the inflammation-dependent increase of NTBI via up-regulated expression of pan-DMT1 is evident in immortalized mouse brain microglia cells, where LPS treatment up-regulates both DMT1 mRNA and protein levels, with concomitant extracellular acidification, due to metabolic changes. This leads to a maximum DMT1 uptake, and treatment with Amiloyd β (1-42), increases DMT1 mRNA and NTBI uptake (McCarthy et al., 2018). Moreover, inflammatory stimuli led to DMT1 up-regulation and ferroportin down-regulation, at both mRNA and protein levels in primary rat hippocampal neurons (Urrutia et al., 2013), in which a significant increase in cell death and oxidative stress was observed after treatment with Aβ-treated astrocyte- or microglia-conditioned media (Urrutia et al., 2017). These evidences show a complex panel about the contribution of the altered homeostasis of DMT1 regulation and iron accumulation, closely related to neuroinflammation, in several neurodegenerative disorders.

Noteworthy, iron overload induces cellular oxidative stress with increased lipid peroxidation, protein and nucleic acid modifications associated to neurodegenerative diseases (Liu et al., 2019). Importantly, under pathological conditions with inflammation, Nitric Oxide (NO) production is increased and aberrantly S-nitrosylated proteins can induce neurodegeneration, like the endogenous S-nitrosylation of DMT1 found in the substantia nigra of post-mortem PD brains (Liu et al., 2018). Interestingly, Liu et al. (2018) pointed the attention on the important aspect that, like extracellular acidification, the increased S-nitrosylation may enhance DMT1 uptake and cell death in the substantia nigra. Moreover, intra-nigral injection of LPS induced NO production and increased neuronal cell death, specifically blocked by the NO synthase inhibitor L-NAME or by Ebselen, a selective inhibitor of ferrous iron uptake by DMT1. Interestingly, lipids peroxidation also increased in the striatum of 100 week aged knockout mice for calcium-independent phospholipase A2 beta (iPLA2β), an enzyme that catalyzes the hydrolysis of membrane glycerophospholipids into free fatty acids and lysophospholipids, as a model of PLA2G6/NBIA (Beck et al., 2015).

Again, inflammation, through the altered expressions of DMT1, ferroportin and hepcidin, induces iron accumulation in the central nervous system (Urrutia et al., 2013; You et al., 2017; Ganz, 2018).

In this respect, hepcidin, the liver-derived peptide hormone, with iron-sensing competence through posttranslational degradation of ferroportin, the only known iron-exporter so far identified, plays a remarkable role in the control of systemic iron homeostasis as its expression is up-regulated by inflammation (Ganz, 2013). Hepcidin significantly influences transferrin-bound-iron transport (Tf-Fe) and NTBI homeostasis with the involvement of the respective transporters (Du et al., 2015; Zhou et al., 2017). In addition, hepcidin deficiency has also been reported to be associated with iron overload in hereditary hemochromatosis (Papanikolaou et al., 2005). Recent evidences highlighted that hepcidin expression in the brain is low, according to Burkhart et al. (2016), and is upregulated

in the central nervous system during pathophysiological inflammation, exerting a paracrine action on neurons through down-regulation of ferroportin expression (Urrutia et al., 2013; You et al., 2017; Vela, 2018). Expression of ferroportin was assessed in endothelial cells of the blood–brain barrier, in neurons, oligodendrocytes, astrocytes, in the choroid plexus and ependymal cells, as well as being identified in the synaptic vesicles fraction of purified rat brain synaptosomes (Wu et al., 2004). However, the iron exporter ferroportin is a hepcidin target in neurons, astrocytes and microglial cells (Urrutia et al., 2013; You et al., 2017). During inflammation, astrocytes drive iron influx from blood flow to the brain, with neuronal iron accumulation through microglia and astrocyte activity, resulting in neurocytotoxicity and neurodegeneration. Systemic hepcidin can cross the blood brain barrier, particularly when BBB is damaged by pathological conditions such as intracerebral hemorrhage and acute inflammation (Ostergaard et al., 2009; Xiong et al., 2016). It is noteworthy that hepcidin serum levels were significantly increased in PD patients after deep brain stimulation (Kwiatek-Majusuak et al., 2018) and

#### REFERENCES


in AD patients (Sternbreg et al., 2017). Therefore, the role of heavy metals absorption and accumulation in the central nervous system may represent a central event for several neurodegenerative diseases of both idiopathic and genetic origin, which arise from nutrition, lifestyle, environmental exposure, genetic polymorphisms, traumatic and neuroinflammatory events affecting the central nervous system (**Figure 2**).

#### AUTHOR CONTRIBUTIONS

RI, BG, and MM contributed to the writing of the manuscript. RI reviewed and edited the manuscript.

#### FUNDING

The authors acknowledge financial support by research funds from the University of Brescia and from private donation to the corresponding author.


neuronal death via the activation of MAPK family proteins in MPTP-treated mice. Exp. Neurol. 280, 13–23. doi: 10.1016/j.expneurol.2016.03.016




**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 Ingrassia, Garavaglia and Memo. 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.

# Iron, Ferritin, Hereditary Ferritinopathy, and Neurodegeneration

Barry B. Muhoberac<sup>1</sup> \* and Ruben Vidal <sup>2</sup>

*<sup>1</sup> Department of Chemistry and Chemical Biology, Indiana University-Purdue University Indianapolis, Indianapolis, IN, United States, <sup>2</sup> Department of Pathology and Laboratory Medicine, Indiana Alzheimer Disease Center, Stark Neurosciences Research Institute, Indiana University School of Medicine, Indianapolis, IN, United States*

Cellular growth, function, and protection require proper iron management, and ferritin plays a crucial role as the major iron sequestration and storage protein. Ferritin is a 24 subunit spherical shell protein composed of both light (FTL) and heavy chain (FTH1) subunits, possessing complimentary iron-handling functions and forming three-fold and four-fold pores. Iron uptake through the three-fold pores is well-defined, but the unloading process somewhat less and generally focuses on lysosomal ferritin degradation although it may have an additional, energetically efficient pore mechanism. Hereditary Ferritinopathy (HF) or neuroferritinopathy is an autosomal dominant neurodegenerative disease caused by mutations in the FTL C-terminal sequence, which in turn cause disorder and unraveling at the four-fold pores allowing iron leakage and enhanced formation of toxic, improperly coordinated iron (ICI). Histopathologically, HF is characterized by iron deposition and formation of ferritin inclusion bodies (IBs) as the cells overexpress ferritin in an attempt to address iron accumulation while lacking the ability to clear ferritin and its aggregates. Overexpression and IB formation tax cells materially and energetically, i.e., their synthesis and disposal systems, and may hinder cellular transport and other spatially dependent functions. ICI causes cellular damage to proteins and lipids through reactive oxygen species (ROS) formation because of high levels of brain oxygen, reductants and metabolism, taxing cellular repair. Iron can cause protein aggregation both indirectly by ROS-induced protein modification and destabilization, and directly as with mutant ferritin through C-terminal bridging. Iron release and ferritin degradation are also linked to cellular misfunction through ferritinophagy, which can release sufficient iron to initiate the unique programmed cell death process ferroptosis causing ROS formation and lipid peroxidation. But IB buildup suggests suppressed ferritinophagy, with elevated iron from four-fold pore leakage together with ROS damage and stress leading to a long-term ferroptotic-like state in HF. Several of these processes have parallels in cell line and mouse models. This review addresses the roles of ferritin structure and function within the above-mentioned framework, as they relate to HF and associated disorders characterized by abnormal iron accumulation, protein aggregation, oxidative damage, and the resulting contributions to cumulative cellular stress and death.

#### Edited by:

*Giorgio Biasiotto, University of Brescia, Italy*

#### Reviewed by:

*Ikuo Tooyama, Shiga University of Medical Science, Japan Andrzej Friedman, Medical University of Warsaw, Poland Shashank Masaldan, University of Melbourne, Australia*

> \*Correspondence: *Barry B. Muhoberac bmuhober@iupui.edu*

#### Specialty section:

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

Received: *25 June 2019* Accepted: *21 October 2019* Published: *11 December 2019*

#### Citation:

*Muhoberac BB and Vidal R (2019) Iron, Ferritin, Hereditary Ferritinopathy, and Neurodegeneration. Front. Neurosci. 13:1195. doi: 10.3389/fnins.2019.01195*

Keywords: hereditary ferritinopathy, neurodegeneration, mutant ferritin, ROS, ferritinophagy, ferroptosis

# INTRODUCTION

Life has evolved incorporating iron as a crucial element in a number of metabolic processes by utilizing it in biochemical catalysis, electron transfer, and oxygen transport and storage. Familiar are the hemeproteins hemoglobin and myoglobin which perform the latter two processes, cytochrome-c for electron transport, and the cytochromes P-450 which perform drug metabolism. In addition, there are major metabolic pathways that require non-heme iron enzymes as centers of catalytic transformation. Neurotransmitter synthesis is intimately connected with such enzymes, e.g., tryptophan hydroxylase and tyrosine hydroxylase, which perform the initial reactions transforming tryptophan and tyrosine into neurotransmitters. These examples imply a clear requirement that cellular iron be adequately available and properly compartmentalized for normal physiology. However, the processes of acquisition, distribution, and catalytic utilization of iron in the cellular environment, which can be rich in oxygen and reducing equivalents, is not without risk (Muhoberac and Vidal, 2013). Iron is a redox active transition metal, which makes it substantially different from sodium, potassium, calcium, and zinc in that it is often involved in single electron transfers. Improperly coordinated iron in the labile iron pool (LIP) of cells can produce reactive oxygen species (ROS) that modify the structure of proteins, lipids, and nucleic acids, causing an immediate local problem of loss of normal function, but also protein aggregation through protein destabilization and unfolding. Additionally, proteins with mutations alone are often destabilized enhancing susceptibility to unfolding and aggregation. Finally, iron also has the ability to enhance aggregation of proteins that have regions of intrinsic sequence disorder, which are generally considered as normal functional sequences that are independent of mutation or ROS modification, and some of these aggregation-sensitive proteins are key players in several neurodegenerative diseases. For example with Parkinson disease (PD), alpha-synuclein, which has as its normal sequence a partially disordered stretch of amino acids that take up variable structural conformations, is especially sensitive to iron levels in that iron enhances its aggregation (Wolozin and Golts, 2002; Theillet et al., 2016). Similarly, the partially disordered amyloid β (Aβ) peptide in Alzheimer disease (AD) undergoes aggregation with iron (Liu et al., 2018). Furthermore, these aggregates have the ability to induce iron redox changes as well as generate ROS (Hands et al., 2011; Liu et al., 2018). Both dysregulation of iron metabolism and iron accumulation in specific regions of the brain are directly associated with several less common neurodegenerative diseases, i.e., the Neurodegeneration with brain iron accumulation (NBIA) diseases (Rouault, 2013; Meyer et al., 2015), but also apparently with several other more commonly encountered ones (Oshiro et al., 2011). To address the issues of availability of iron as metabolically crucial and the potential toxicity of improperly coordinated iron, ferritin has evolved as the end point protein to remove and store excess iron safely and provide it to the cell as required, and thus to reduce cellular damage and stress.

The importance of native ferritin in normal physiology, as well as its purified availability for in vitro research, has made it one of the more well-studied proteins over several decades (Crichton, 2009). While the mechanism of iron uptake and storage as an iron mineral in its interior is complex but relatively wellunderstood, the mechanism of iron release, although generally considered to involve lysosomal degradation through the process of ferritinophagy, has research suggesting alternative pathways. These alternatives are release (1) induced by small cytosolic molecules usually found close to ferritin or (2) by the proteasome (Liu et al., 2003; DeDomenico et al., 2009; Tang et al., 2018). Such pathways may be involved in general or perhaps more nuanced iron management. More recently, mutant forms of ferritin in which the C-terminal alpha helix is disordered and unraveled at the four-fold pores providing an iron exit and entry pathway that is normally considered closed, have been characterized (Muhoberac and Vidal, 2013). These mutant forms were discovered through clinical investigation and molecularlevel characterization of the neurological disorder hereditary ferritinopathy (HF) or neuroferritinopathy, which has some clinical characteristics similar to PD. Inclusion bodies (IBs) containing ferritin, increased iron levels, and oxidative damage (carbonylation) are found in brain samples of patients with HF upon autopsy (Vidal et al., 2004). These characteristics are to a great extent reproducible for investigation with cellular and animal models expressing mutant ferritin. Such ferritin expressed and purified from cell cultures undergoes both (1) precipitation with increasing iron and (2) oxidative damage, i.e., carbonylation, proteolysis, and crosslinking, in the presence of physiological concentrations of iron and ascorbate found in the brain (Baraibar et al., 2012). Here ascorbate functions as a reductant so that iron can produce ROS. Dealing with elevated iron, ROS formation, aggregation, oxidative damage, and IB formation in HF, requires the use of cellular synthesis, transport, repair, and disposal mechanisms and may provide a source of cumulative cellular stress and damage as cells age.

The finding of elevated iron in HF, PD, AD, and other neurodegenerative diseases suggests the importance of tight physiological control of iron compartmentalization and concentrations, and leads to consideration of ferroptosis, which is a newly described form of rapid programmed cell death clearly different in cellular structural changes and biochemical pathways from apoptosis and necrosis (Dixon et al., 2012; Stockwell et al., 2017). Characteristics of ferroptosis are lipid peroxidation and abundant or elevated iron. Ferroptosis may be enhanced through the process of ferritinophagy, by which ferritin is delivered to the lysosome for degradation and a large quantity of iron can be rapidly released enhancing ROS production (Latunde-Dada, 2017). However, the overall ferroptotic process is complicated in that damage can be repaired by a particular glutathione peroxidase if sufficient glutathione is available. Importantly, ferroptosis appears to be strongly connected with cell death in neurodegenerative diseases (Tang et al., 2018). In HF there are elevated iron and lipid peroxidation similar to ferroptosis, but at least three mechanisms of ferritinophagy inhibition may be operative, suggesting that it is the leaky four-fold pores and modestly elevated iron that triggers a long term cellular ferroptotic-like state gradually and cumulatively leading to cellular misfunction and decline.

TABLE 1 | *FTL* genetic variants associated with Hereditary Ferritinopathy (neuroferritinopathy).


A contribution to decline is the accumulation of ferritin aggregates from ferritin overproduction and iron binding, which causes a variety of stressful secondary effects on synthesis, transport, and disposal, as well as providing additional ROS generating sites. The complicated interrelationships between the molecular-level processes described above and HF are reviewed herein, with the further objective that this presentation will provide a better understanding of the common threads among protein aggregation, elevated iron, and ROS damage found in neurodegenerative diseases.

# GENETICS, CLINICAL PRESENTATION, AND PATHOLOGY OF HF

HF or neuroferritinopathy is inherited in an autosomal dominant pattern. Linkage analysis established a relationship between the disease and a locus on chromosome 19q13.3, which contains the ferritin light chain (FTL) gene, consisting of four exons and three introns (Curtis et al., 2001). Mutations in the FTL gene causing HF have been reported in individuals with a Caucasian ancestry and in East Asian populations from Japan, Korea, and China, presenting with abnormal involuntary movements (Curtis et al., 2001; Vidal et al., 2004; Mancuso et al., 2005; Ohta et al., 2008; Devos et al., 2009; Kubota et al., 2009; Storti et al., 2013; Nishida et al., 2014; Ni et al., 2016; Yoon et al., 2019). Mutations in FTL consist of nucleotide duplications in exon 4 that affect the C-terminal residues of the FTL polypeptide (**Table 1**). There are no known polymorphisms in the FTL gene that may affect the clinical and pathological phenotype. In addition to the cases indicated in **Table 1**, two more cases of HF have been described. One case was diagnosed pathologically and no genetic data is available (Schröder, 2005). The second case consists of a missense mutation (A96T) in the FTL gene in an individual without significant involvement of the putamen, thalamus, and substantia nigra that did not show autosomal dominant transmission since the mother of the proband, also a carrier of the A96T mutation, had similar MRI findings and was asymptomatic (Maciel et al., 2005). The A96T variant has been recently shown to be stable under physiological conditions and incorporate iron comparable to that of wild-type FTL ferritin (Kuwata et al., 2019).

The onset of clinical signs and symptoms may occur between the second and seventh decade of life and depending on the specific mutation, the disease may become evident from the third to the fifth decade of life (Vidal and Ghetti, 2015). HF occurs equally in males and females, with duration ranging from a few years to several decades (Curtis et al., 2001; Vidal et al., 2004, 2011; Mancuso et al., 2005; Ohta et al., 2008; Devos et al., 2009; Kubota et al., 2009; Ory-Magne et al., 2009; Storti et al., 2013; Nishida et al., 2014; Ni et al., 2016; Yoon et al., 2019). The clinical presentation of the disease may be highly variable not only between families but also within a family, suggesting that a heterogeneous presentation may be a characteristic of HF (Ory-Magne et al., 2009). The disease may present as a movement disorder with tremor, cerebellar signs, Parkinsonism, dystonic, and choreic movement; in addition, pyramidal and pseudo-bulbar symptoms may be present. Although dystonia and dysarthria are the main manifestations of HF, the disease cannot be diagnosed only on the basis of clinical symptoms alone. In more advance stages, psychiatric and cognitive symptoms may become evident (Curtis et al., 2001; Vidal et al., 2004, 2011; Mancuso et al., 2005; Chinnery et al., 2007; Ohta et al., 2008; Devos et al., 2009; Kubota et al., 2009; Ory-Magne et al., 2009; Storti et al., 2013; Nishida et al., 2014; Ni et al., 2016)

Neuroimaging studies allow regional characterization of normal vs. abnormal iron deposition through T2<sup>∗</sup> -weighted images (WI) and susceptibility weighted imaging (SWI). These studies show in the early clinical stage of HF the presence of an abnormal decrease in T2<sup>∗</sup> signal intensity reflected as hypointense lesions in the basal ganglia on particularly in the globus pallidus and putamen, which signifies abnormal accumulation of iron (Curtis et al., 2001; Vidal et al., 2004, 2011; Mancuso et al., 2005; Chinnery et al., 2007; McNeill et al., 2008; Ohta et al., 2008; Devos et al., 2009; Kubota et al., 2009; Ory-Magne et al., 2009; Ohta and Takiyama, 2012; Storti et al., 2013; Nishida et al., 2014; Ni et al., 2016; Yoon et al., 2019). With the progression of the disease the signal loss extends to the dentate nucleus, substantia nigra, and cerebral cortex.

Macroscopic examination of the brain of patients with HF reveals mild cerebral and cerebellar atrophy as well as cavitation of the putamen. Neuropathological data is available for individuals with the c.442dupC, c.460dupA, and c.497\_498dupTC mutations (Curtis et al., 2001; Vidal et al., 2004; Mancuso et al., 2005). The caudate nucleus and putamen may show a grayish discoloration. At neuropathologic examination, the main findings are the presence of ferritin IBs in the cytoplasm and nuclei of glial cells and some subsets of neurons, and abnormal iron deposition in IBs, in particular in the caudate nucleus, putamen, and globus pallidus. IBs can be seen as eosinophilic, homogenous bodies on haemotoxylin and eosin staining, and can be stained using the Perls' and Turnbull blue methods for iron (**Figure 1**; Vidal et al., 2004). Different from other IBs such as those containing αsynuclein in Lewy bodies in PD and dementia with Lewy bodies (DLB) (Spillantini and Goedert, 2018), ferritin IBs do not show electron dense fibrillar structures on electron microscopy (**Figure 2**) and cannot be stained by thioflavin S, but are strongly immunopositive using antibodies against ubiquitin. In

FIGURE 1 | Numerous inclusion bodies containing iron are present in the brain with a case of hereditary ferritinopathy. In HF, the most striking pathologic alteration is the presence of intranuclear and intracytoplasmic bodies in glial cells and in some subsets of neurons. Sections of the putamen show numerous ferritin bodies of various sizes stained by H&E (a). Iron-containing ferritin inclusion bodies stained with the Perls' method for iron (b) and immunostained using antibodies specific for the mutant ferritin light chain polypeptide (c). FTL immunoreactivity is seen in the nuclei and also in the cytoplasm. Perls' staining and antibodies against wild-type and mutant FTL subunits were used as previously described (Vidal et al., 2004). Scale bars: (a–c), 50µm.

FIGURE 2 | Electron microscopy of IBs in patients with HF. High-power electron micrograph showing ferritin accumulation in the putamen (a), in the nucleus of granule cells of the cerebellum (b), and skin (c). Note the chromatin accumulation toward the nuclear membrane.

the cytosol, ferritin may be dispersed or forming IBs showing a spherical appearance. Intranuclear IBs measure between 2 and 35µm in diameter, and may occupy almost completely the nucleus and as a result displace the chromatin up against the nuclear membrane. Antibodies against the N-terminus of FTL as well as antibodies against ferritin heavy chain (FTH1) and mutant-specific antibodies strongly label ferritin IBs of all sizes in glial cells and in nerve cells (Vidal et al., 2004). The use of antibodies also show the presence of diffuse deposits, perhaps early aggregates, in the cytoplasm (**Figure 1**; Vidal et al., 2004. Markers for lipid oxidation are also evident by immunostaining (Mancuso et al., 2005).

In the neocortex, intranuclear and intracytoplasmic ferritin immunopositivity is present throughout the cortical layers, with the exception of layer I and II (Vidal and Ghetti, 2015). Intranuclear IBs are seen in perineuronal satellite cells and in perivascular glia. The caudate nucleus, putamen, and globus pallidus contain the largest numbers of IBs. In these areas, extracellular ferritin aggregates may also be found, probably as the result from the coalescence of multiple bodies. Intranuclear IBs may be seen in neurons of the putamen, globus pallidus, and thalamus, and occasionally, ferritin immunopositivity may be seen in the leptomeningeal and parenchymal vessel walls as well as in leptomeningeal cells. In the cerebellum, IBs are seen in the

are present in all areas of the hippocampus, in particular in the granular layer of the dentate gyrus and CA4. (b) Intranuclear and intracytoplasmic ferritin deposits in the cerebral cortex immunolabeled by antibodies against the FTL polypeptide. (c–e) Ultrastructural studies showing a ferritin inclusion body in the nucleus of a neuron of the striatum (c). The section marked in (c) was magnified in (d,e). IBs (indicated with arrows) occupy most of the nucleoplasm. The chromatin appears to be centrifugally displaced to varying degrees, often forming a thin layer adjacent to the nuclear membrane (nm) (e). (a,b) Immunohistochemistry using antibodies against wild-type FTL subunits as previously described (Vidal et al., 2008). Scale bars: (a), 10µm; (b), 40µm; (c), 2µm.

nuclei of glial cells, including the Golgi epithelial cells, in granule cells and Purkinje cells. In the latter, IBs could also be found in the cytoplasm of the perikaryon and dendrites. Glial cells of the white matter are affected as well. IBs are also found in the nuclei of endothelial cells, in cells of the vascular adventitia and in the epithelium of the choroid plexuses; ependymal cells appear to be free of IBs. IBs have been reported in the skin, kidney, liver, and muscle in affected individuals from French and American families (Vidal et al., 2004; Mancuso et al., 2005). The detection of ferritin IBs in biopsies of the skin or muscle may be helpful for the pathologic diagnosis of symptomatic individuals (**Figure 2**; Vidal et al., 2004, 2011).

Low serum ferritin levels were initially reported in individuals with the c.460InsA mutation (Curtis et al., 2001); however, serum ferritin levels are decreased in some but not all patients with HF (Muhoberac and Vidal, 2013; Yoon et al., 2019). Interestingly, CSF ferritin levels were significantly lower than normal in a patient with the c.469\_484dup16 mutation who had normal serum ferritin values (Nishida et al., 2014). Additional studies are needed to establish whether CSF ferritin levels could be used as a novel biomarker for HF.

# MODELING HF IN CELLS AND MICE

A mouse model of HF (FTL-Tg) that expresses the human mutant ferritin FTL subunit p.Phe167SerfsX26 (mtFTL) has been extensively characterized and also used to generate astrocytes and primary mouse embryonic fibroblasts for in vitro studies (Barbeito et al., 2009; Li et al., 2015) that complemented studies using fibroblasts from patients with HF (Barbeito et al., 2010). Expression of the transgene in the mouse yields a progressive neurological phenotype, with a significant decrease in motor performance, shorter life span, misregulation of iron metabolism, and evidence of oxidative damage (Vidal et al., 2008; Barbeito et al., 2009; Deng et al., 2010). Mutant HF mice form nuclear and cytoplasmic ferritin IBs in glia and neurons throughout the CNS and in organ systems outside the CNS, as found in patients with HF (**Figure 3**; Vidal et al., 2004, 2008). Both cytosolic and nuclear inclusions grow larger in size and number as the animals age to the point of crowding other cellular structures potentially causing mechanical disruption of cellular processes like transport. DNA oxidative damage is detected associated with nuclear mitochondria (Deng et al., 2010) at 12 months but not 6

months of age. Ferric iron accumulation is detected throughout the CNS through Perls' Prussian blue staining in histological sections and T2-weighted imaging in vivo. There is evidence of proteosomal involvement at the ferritin inclusions, including ubiquitination and the presence of the 11S and 20S subunits. Ferritin inclusions from the transgenic mice are SDS insoluble as found in HF patients. The SDS insoluble IBs suggests a rather strong association of the ferritin 24-mers (Vidal et al., 2008), which could occur through iron bridging as might be expected with a disordered and extended C-terminus and perhaps including contributions from partial unfolding and proteolysis. Additional studies show a substantial increase in cytoplasmic ferritin (∼200–400%) depending on the subunit identity from antibody labeling and cortex vs. cerebellum sampling (Barbeito et al., 2009). There is a statistically significant increase in nonheme iron levels between controls and transgenic HF mice of ∼15% (cortex) to 20% (cerebellum), but these levels are more modest than the protein increases above. Evidence is found for elevated oxidative protein and phospholipid damage through detection of HNE-protein modifications, protein carbonyls, and lipid peroxidation products like MDA in the transgenic HF mice. Radical formation is found through immunohistochemistry of nitrone-protein adducts. Thus, this mouse model produces only modest increases in iron levels vs. controls, but substantial, cumulative overproduction and lack of clearance of ferritin, as well as oxidative damage.

Similar dysfunction is observed in primary culture skin fibroblasts with the FTL subunit p.Phe167SerfsX26 mutation obtained from a HF patient (Barbeito et al., 2010). The HF fibroblastsshow accumulation of mutant ferritin in the cytoplasm and nucleus. There is a substantial increase in ferritin content in the HF cells of ∼320 and 470% in the FTH1 and FTL subunit concentrations, respectively, but the increase in total iron content is again modest at 25%. Also, a modest increase in ROS concentration is found in the HF cells over normal, which was further enhanced by FAC treatment.

In primary mouse embryonic fibroblasts (MEFs) from FTL-Tg mice, there is initially a substantial increase in the levels of soluble mtFTL, but not in the levels of insoluble mtFTL. After incubation with ferric ammonium citrate (FAC), MEF cells became less viable (3–5 days) and contained more iron than MEF cells from wild-type mice by ∼40%. Thus, the combination of mutant-containing ferritin and iron leads to aggregation into SDS resistant aggregates (Garringer et al., 2016). Incubation with the iron chelator deferiprone (DFP) after iron loading removes substantially more iron from the cells than the controls, which suggests the iron is more loosely bound in the cells expressing mtFTL as would be expected if the iron is initiating aggregation and relatively loosely bound to the C-termini and surface region of ferritin spheres rather than internally as a mineral. In earlier studies, transgenic astrocytes were loaded with iron by FAC incubation causing substantial conversion of ferritin to a SDS insoluble form, but incubation with the iron chelator phenanthroline led to a substantial reversal back to a soluble form (Baraibar et al., 2008). Along these lines, in vitro studies using purified mtFTL homopolymeric recombinant ferritin that was precipitated by iron can be partially resolubilized by addition of the chelator deferoxamine (Baraibar et al., 2008).

An interesting but perhaps counterintuitive result is obtained in studies of FTL-Tg mice under iron loading and chelation treatment (Garringer et al., 2016). Here, iron overload and DFP treatment of the mouse model have remarkable effects on systemic iron homeostasis and ferritin deposition, without significantly affecting CNS pathology (**Figure 4**). The data obtained using the FTL-Tg mouse model is in agreement with previous observations indicating lack of effectivity of a chelation therapy in individuals affected by HF (Chinnery et al., 2007; Kubota et al., 2009; Storti et al., 2013). Chinnery et al. (2007) treated three patients with monthly venesection for 6 months. Two of the patients also were treated with intravenous deferoxamine (4,000 mg weekly subcutaneously for up to 14 months), and one had oral DFP (2 g, three times a day for 2 months). These treatments cause profound and refractory iron depletion without significant benefits for the patients. Kubota et al. (2009) treated a patient with monthly venesections (400 mL/mo) for 2 months without any change in the clinical condition of the patient. Storti et al. (2013) did a 6-month trial with DFP (15 mg/kg/day) in one patient without any signs of improvement. Work on the mouse model also suggests that systemic ferritin deposits in HF (Vidal et al., 2004) may not be useful to monitor therapeutic approaches since they may be modified independently from brain ferritin deposits.

To determine whether loss of function of the FTL subunit in the brain could lead to some of the pathologic features observed in HF by loss of the iron storage function of ferritin, a Ftl knock-out (Ftl−/−) mouse model was generated (Li et al., 2015). Interestingly, Ftl−/<sup>−</sup> mice are viable (knock-out of the Fth subunit is embryonic lethal) and do not show signs of neurodegeneration, presence of an inflammatory process, noticeable protein aggregates, or iron accumulation as in patients with HF, suggesting that the deleterious effect(s) caused by mutant FTL subunits in HF are driven by disruption of the ferritin pore structure and unraveling of the C-terminus in the heteropolymer rather than by a loss of normal function of the FTL subunit itself. Importantly, Ftl−/<sup>−</sup> mice show that Fth ferritin homopolymers are capable of maintaining brain iron homeostasis in vivo, paving the way for the development of a potential therapeutic approach for HF using RNA interference to induce sequence-specific post-transcriptional gene silencing of mutant FTL (Li et al., 2015).

#### FERRITIN STRUCTURE, FUNCTION, AND NORMAL IRON INCORPORATION

The functions of ferritin are to (1) sequester and store excess ferrous iron, when necessary in large amounts, removing it from the LIP and thus preventing deleterious cellular processes such as ROS generation, protein aggregation, and iron deposition, and (2) to allow release of this iron as needed for cellular processes. Storage is accomplished by converting ferrous iron into a less reactive and compact form of ferric iron oxide "mineral" that is stored in the interior of ferritin, which has the form of a spherical shell to contain it. Spatial storage efficiency (i.e., density of the iron mineral) is high in that storage does not require separate iron ions be chelated by a single protein or a small molecule chelator in a 1:1 or 1:2 ratio.

Ferritin is a molecule assembled from 24 protein subunits into a highly symmetric protein shell (**Figures 5a,b**) of exterior diameter ∼12 nm and a hollow core of ∼8 nm that can store up to 4,500 iron atoms within its shell (Harrison and Arosio, 1996; Crichton and Declercq, 2010). Native human ferritin is a heteropolymer composed of FTH1 (MW∼21 kDa) and FTL (MW∼20 kDa) subunits in varying ratios depending on the cell/organ of origin such that its storage rate and ability is optimized for its location. Both subunits are composed of four parallel and closely associated alpha-helix bundles (A–D) and a fifth alpha-helix (E) that is shorter and not aligned with the others but rotated approximately perpendicular (∼60 degrees; **Figure 5c**). The E helix points inwards from the ferritin spherical surface and forms the sides of a pore (see below). The two different subunits diverge in amino acid identity (∼55%), but fold into an almost identical 3-dimensional structure, which allows them to assemble into a highly symmetric shell with tight fitting inter-subunit surfaces and junctions exhibiting three-fold and four-fold symmetry (**Figures 5a,b**). This assembly forms eight more polar three-fold pores of ∼6 Å length and ∼3.4 Å diameter and six more apolar four-fold pores of <3 Å diameter and ∼12 Å length, with the latter generally considered closed to iron passage (Harrison and Arosio, 1996; Crichton and Declercq, 2010). The FTH1 subunit contains the ferroxidase center, which oxidizes the ferrous ions in preparation for mineral formation and the FTL subunit contains clusters of negatively charged residues which foster the nucleation and mineral growth. The three-fold pores are the ferrous iron entry pathways and there is a group of negatively charged residues that help transport the iron through the pore, along the interior of the shell to the ferroxidase site, and then to the nucleation and mineralization points. The specifics of the iron acquisition, transport, catalytic oxidation, and deposition processes in ferritin, i.e., identification of the specific metal binding amino acid residues involved in the ferroxidation and the overall pathway from ferrous iron to ferric mineral, have been elucidated in detail over a decade or more of study (Masuda et al., 2010; Tosha et al., 2010; Behera and Theil, 2014; Pozzi et al., 2015). Interestingly, homopolymeric ferritin 24-mers composed of only FTL subunits can load iron into the shell, but at a much reduced rate vs. heteropolymeric ferritins, which can coordinate the iron deposition process between subunits.

#### MUTANT FERRITIN STRUCTURAL AND FUNCTIONAL ANOMALIES

Over the last decade there has been a substantial amount of research describing the molecular level structural and functional differences between normal ferritins and those containing mutant FTL subunits made possible by cloning, bacterial expression, and purification. Our studies focused on the p.Phe167SerfsX26 mutation that causes a 9 amino acid substitution and a 16 amino acid extension of the C-terminus out of a total 174 amino acids (Baraibar et al., 2010). This substantial change into a completely non-native sequence does not prevent mtFTL subunits from assembling into homopolymeric 24-mer spheres, with ultrastructure quite similar to the native, that are crystallizable. This mutation encompasses part of the E helix, but as demonstrated by X-ray crystallography in different subunits of a 24-mer the helix unraveling can be found to range from just before the mutation to 10 or more amino acids toward the D helix. The unraveling distribution (i.e., the number of amino acids disordered as to not be visible by Xray crystallography) is statistical over the six four-fold pores and the four C-termini composing each of them (**Figures 5d–f**). The result is a substantial disordering of what was a tight four-fold pore into a more flexible, iron permeable structural defect and an unraveling of the C-terminal sequence that, if extended from the exterior of the shell, could reach as far as the entire diameter of ferritin. This statistical disordered structure is important not just for iron leakage, but because the disordered C-termini will take up various orientations and conformations as potential iron binding sites. Several amino acid residues of the disordered Cterminus are known iron binding ligands and the disorder and conformational sampling adds variability to their potential iron chelation geometry in a search for free energy minimization. As mentioned previously, redox active improperly coordinated iron is problematic as a ROS generator causing protein (including ferritin) modification, and the multiple C-termini conformations may enhance the generation probability. Additionally, sampling multiple conformations can enhance the probability of iron bridging two C-termini either on a single mutant-containing ferritin or between two mutant-containing ferritins enhancing aggregation (Muhoberac and Vidal, 2013). Indeed, because of the extremely close packing of the cytoplasm, this Cterminal extension has the potential to interact with multiple cellular constituents, e.g., small molecules, proteins, bilayers, cytoskeleton component, etc., with unpredictable results.

In vitro studies show that the four-fold pore-centered FTL mutation has several consequences. Firstly, homopolymeric ferritin composed of mtFTL subunits is more susceptible to iron-loading induced aggregation and precipitation than ferritin composed of wild type FTL (wtFTL) subunits. In standard ironloading assays, precipitation occurs starting at approximately an iron to mtFTL ferritin (24-mer) ratio of 1,500:1, with homopolymers of wtFTL remaining soluble to >4,000:1 ratio (Baraibar et al., 2008). Although the iron concentration was high, these experiments point out fundamental differences in the interaction with and handling of iron between mutant and wild-type ferritins. Secondly, with heteropolymeric ferritin, this aggregation tendency extends to the cases in which mtFTL subunits are combined with those of either wtFTL or FTH1 to form ferritin 24-mers (Muhoberac et al., 2011). Specifically, reduction in the percent of mtFTL subunits by half and having the remainder provided by either wtFTL or by FTH1 causes no aggregation at an iron to ferritin ratio of 1:1,000 (as above) but substantial aggregation at a 3,000:1 ratio. Clearly, protection is not provided by a large reduction in the number of mutant subunits. Thirdly, the overall stability of homopolymeric ferritin composed of mtFTL subunits is compromised with respect to that of the wtFTL in temperature denaturation and proteolysis studies (Baraibar et al., 2008). However, it is known that wtFTL homopolymers are more stable than FTH1 homopolymers such that the inclusion of wtFTL subunits in native heteropolymeric ferritin is considered stabilizing. Although the loss of stability of the homopolymeric mtFTL 24-mers is found to not be substantial, any loss suggests the possibility of more general enhanced local unfolding causing repositioning and exposure, if only transiently, of normally buried or hydrogen bonded residues. Supporting this unfolding process is the enhanced thermolysin proteolysis of mtFTL over wtFTL homopolymers in which a 17 kDa N-terminal fragment is produced from the mutant but not the wild type. The position of this fragment in the sequence strongly suggests substantial transient fluctuations in overall mutant ferritin structure even though ultrastructurally the mutant is spherical and crystallizable. Fourthly, susceptibility to ROS damage of mtFTL homopolymers is enhanced substantially over that of ferritin composed exclusively of wtFTL subunits. At physiological concentrations of iron and ascorbate, mtFTL homopolymers undergo substantial proteolysis (6 and 14 kDa fragments), crosslinking (27 kDa fragment), and carbonylation (Baraibar et al., 2012). Ferritin composed of wtFTL subunits alone does not undergo these changes and a hydroxyl radical scavenger prevents mtFTL homopolymer from undergoing such disruptive modifications implying a mutant ferritin-centered, ROS-generating reaction initiated by iron and ascorbate. These results strongly suggest that the mtFTL homopolymer has one or likely more iron binding sites available that generate ROS which attack ferritin and may attack other proteins and lipids, as discussed in the next section.

With respect to the function of iron incorporation, defects are found in both the mtFTL homopolymer and the 24 mer heteropolymers. A comparison of iron uptake between heteropolymers composed of half FTH1 and half of either mtFTL or wtFTL subunits demonstrates a 50% reduction in iron incorporation with a 1,000:1 ratio of iron to heteropolymeric ferritin (Muhoberac et al., 2011). This ratio is below that found to cause iron precipitation suggesting there exists an iron incorporation defect caused by the four-fold structural perturbation in HF that exists at the level of single ferritin heteropolymers containing the mutant subunit. In addition, small soluble aggregates not yet large enough to form precipitates could also hinder the sequestration of iron. Taken together, the above-mentioned in vitro studies suggest that the overall structural and functional effects of the mutation are quite substantial and damaging, and may require a variety of cellular processes for correction and compensation, which are energetically costly to the cells. It is interesting to note that in PD there is a decrease in the concentration of FTL polypeptides in the substantia nigra, which in a manner parallels the mtFTL dysfunction in HF in that both lead to similar changes in LIP iron (Koziorowski et al., 2007; Friedman et al., 2011).

#### IRON CHEMISTRY, INTERACTIONS, AND REACTIVE OXYGEN SPECIES

The importance of iron to cellular function must be understood within the context of its versatile, but complex chemical behavior and its difficulty in cellular processing, transport, and storage (Shriver and Atkins, 2006; Crichton and Declercq, 2010). Iron can not only undergo redox cycling between ferrous and ferric states, but its ability to be cycled at a particular redox potential strongly depends on its cellular ligation, i.e., the ligands from proteins or small molecules that are bound to it. Ferrous iron prefers coordination to nitrogen or sulfur-containing ligands whereas ferric iron prefers those that contain oxygen such as aspartate and citrate. This preference is part of a chemical classification of binding that is dependent on atomic radius and polarizability of the atoms in contact (Shriver and Atkins, 2006). Iron coordination not only varies its redox potential, but can determine its behavior as an electron relay or a catalytic center, and such processes generally require some level of conformational control of as well as outright exchange of iron coordination by the chelating protein and substrate. Reductant concentration and cellular packing also factor into the iron redox state. Iron can interact with dioxygen producing potentially toxic ROS that modify proteins and phospholipids, and this reaction is substantially enhanced by iron with <6 coordinate ligand shell (Graf et al., 1984). Such improperly coordinated iron is found, even transiently, in the LIP with small cytosolic molecules and with mutant, denatured, and aggregated proteins. Even protein aggregation and precipitation of unmodified proteins, especially those with regions of intrinsic sequence disorder, can be enhanced substantially by iron presence, as is found with αsynuclein and Aβ (Mantyh et al., 1993; Golts et al., 2002; Everett et al., 2014; Boopathi and Kolandaivel, 2016). Indeed, there are iron deposits and interactions with specific proteins that are routinely found in association with neurodegenerative diseases (Zecca et al., 2004; Carboni and Lingor, 2015; Ward et al., 2015; van Bergen et al., 2016; Liu et al., 2018). Aqueous iron solubility itself varies substantially between its ferrous and ferric forms, with the ferric form basically insoluble near neutral pH and requiring cellular carriers such as small molecule chelators or proteins, but with ferrous iron soluble near 100 mM (Crichton and Declercq, 2010). Because of this level of molecular behavioral complexity and the number of cellular constituents with which iron can interact, the cell has developed a variety of mechanisms to employ iron in biochemical catalytic processes as well as to deal with excess iron and iron-produced ROS, which potentially negatively effects proteins, lipids, and DNA.

The most common cellular ROS—hydrogen peroxide, superoxide, and the hydroxyl radical—vary substantially in their origin, interactions, lifetime, and toxicity (Halliwell and Gutteridge, 2007; Auten and Davis, 2009; El-Beltagi and Mohamed, 2013; Davies, 2016). The hydroxyl radical is considered the most toxic with a high reactivity and short diffusion distance, causing protein backbone cleavage, oxidative residue modification, crosslinking, nucleic acid modification, and lipid peroxidation close to its sight of generation, although once initiated the peroxidation damage can spread. Hydrogen peroxide is not particularly reactive with a large ability to diffuse and is part of a redox signaling mechanism in cells, and superoxide is considered of intermediate toxicity. There are enzymes that normally produce these ROS for biochemical and defensive purposes, and ROS may also originate with damaged enzymes or those that on occasion miscycle during normal catalysis (Turrens, 2003). The characterization of ROS availability is further complicated by the potential interconvertability of these species both by detoxifying enzymes and local concentrations of poorly coordinated iron in the LIP.

Hydroxyl radical formation is generally discussed with respect to two widely accepted and employed chemical generating systems of different composition (Cohen et al., 1981; Uchida et al., 1989; Ito et al., 1993). The first consisting of hydrogen peroxide and a ferrous salt is the Fenton/Haber-Weiss-type reaction which produces hydroxyl radicals (·OH) as follows:

$$\text{Fe}^{2+} + \text{H}\_2\text{O}\_2 \rightarrow \text{Fe}^{3+} + \cdot\text{OH} + \text{OH}^-$$

The ferrous ion is regenerated by a reaction between a ferric iron and hydrogen peroxide or perhaps superoxide (·O − 2 ) which is formed from ferrous iron and dioxygen, and then the cycle repeats. The second generating system consists of ascorbic acid, EDTA, and ferric iron and is called the Udenfriendtype reaction. This system appears to be more similar to what might occur in vivo in that it has a supply of physiological reductant and the iron is chelated by ligands other than water, hydroxide, and/or oxygen. Ascorbic acid reduces the iron, which reacts with dioxygen to produce superoxide. Then superoxide produces hydrogen peroxide which feeds a Fenton/Haber-Weisstype reaction producing the hydroxyl radical. Ascorbic acid cycles the iron back to the reduced state producing additional hydroxyl radicals. As might be expected, the ability of the Udenfriendtype reaction to generate ROS is strongly dependent on the type of iron chelation in that EDTA-iron must transiently become 5 coordinate opening a ROS generating site. Radical generation and the ability to modify target molecules is strongly modulated by the choice of iron chelators that are substituted for EDTA (Burkitt and Gilbert, 1990).

There are a number of cellular enzymatic processes to (1) eliminate the specific ROS, converting them to water or other ROS that are, under normal conditions in which iron levels are controlled, less damaging, or (2) repair cellular damage caused by them (Halliwell and Gutteridge, 2007; Kim et al., 2015). Superoxide is catalyzed to hydrogen peroxide by superoxide dismutase, thus exchanging one ROS for another, which can generate a potentially more toxic species through the non-enzymatic Fenton/Haber Weiss-type reactions. Hydrogen peroxide can be catalyzed to water by catalase, removing it from the toxicity cycle. Glutathione peroxidases can also convert hydrogen peroxide to water, but doing so requires reduced glutathione, which brings into focus the need for maintaining the reducing potential in the brain where normal levels of ascorbate and reduced glutathione (GSH) are relatively high (1– 2 mM) (Baraibar et al., 2012). However, high levels of reducing agents in the brain may be problematic with respect to ROS generation. Taken together, ROS management and damage repair are highly complex and interrelated processes that would likely become problematic as iron levels exceeded the norm. In addition, protein conformational changes and aggregation that result in iron binding sites have the potential to further contribute to ROS generation, as will be discussed later.

## PROTEIN AND LIPID MODIFICATION AND DEGRADATION BY ROS

The general characteristics of biological exposure to ROS are deleterious oxidative modifications of proteins, lipids, and DNA (Halliwell and Gutteridge, 2007; Suzuki et al., 2010; Yin et al., 2011; El-Beltagi and Mohamed, 2013; Davies, 2016). As with iron, the levels of oxidized cellular proteins and lipids increase with age and are found in HF and other neurodegenerative diseases. Additionally, modified proteins and their aggregates are associated with compromised proteasomal function inhibiting damaged protein clearance (Grune et al., 2001; Stadtman, 2004; Hipkiss, 2006; Radak et al., 2011). Thus, this lack of clearance and increased concentration not only leads to enhanced protein aggregation in a cyclical manner, but strains cellular repair and elimination processes to the point of accumulation, diverting cellular resources from other tasks (e.g., biomembrane maintenance). Although ROS-centered cellular damage is often characterized by protein carbonyl derivative formation (carbonylation), the kinds of modifications that are caused by ROS are much more varied (Halliwell and Gutteridge, 2007). Protein side chains can undergo substantial modification by the hydroxyl radical depending on their identity. For example, lysine and arginine can acquire a carbonyl, methionine can be oxidized, phenylalanine, and tyrosine can each acquire an added hydroxyl group, and histidine becomes 2-oxo-histidine. Such modifications change residue physical attributes, e.g., charge and size, and also chemical reactivity, which in turn can alter their interactions with substrates or other proteins, decrease protein stability, and enhance aggregation without direct iron binding or bridging. Protein modification and unfolding has the potential to enhance iron binding and iron binding-induced aggregation. Loss of protein stability (partial unfolding) and the aggregation process itself can produce new iron binding sites that can form centers of ROS generation. Potentially more extensive protein damage can occur because the hydroxyl radical can also cause polypeptide backbone cleavage and crosslinking, with the latter particularly problematic for protein repair and disposal pathways.

Not only proteins, but lipids are subject to modification by ROS (Yin et al., 2011; El-Beltagi and Mohamed, 2013), with those containing more than one double bond being particularly susceptible to radical attack and the resulting substantial structural damage (Wagner et al., 1994). Lipid peroxidation has been linked to a large number of diseased states including AD and PD as well as being found with HF (Yin et al., 2011). Such damage can clearly alter biomembrane structure, e.g., fluidity and phospholipid raft formation, and function, including that of passive transport and membrane bound enzymes and receptors. Even phospholipid-based signaling can be affected. However, the extent of potential cellular disruption and damage is magnified by the more complicated cyclical propagation behavior of ROS-modified lipids that attack unaltered lipids and the several kinds of reactive phospholipid byproducts that are formed, e.g., malondialdehyde (MDA) and 4-hydroxynonenal (HNE) that cause additional damage (Halliwell and Gutteridge, 2007). There are three steps involved in radical-centered damage to membranes: initiation, propagation, and termination, with various subreactions including direct iron involvement (Wagner et al., 1994; El-Beltagi and Mohamed, 2013). Lipid peroxidation most likely begins with an initiation step in which a hydrogen atom is extracted by a hydroxyl radical from a carbon adjacent to double bonded carbons in a polyunsaturated hydrocarbon chain (LH to L·). The ability energetically to produce this carbon-centered radical depends on its location relative to vinyl groups. The hydrogen of a carbon sandwiched between two vinyl groups, i.e., a bis allyic carbon, is most easily abstracted followed by a carbon adjacent to a single vinyl group, i.e., an allylic carbon. The most difficulty hydrogen to abstract is an acyl chain carbon away from double bonds explaining the susceptibility of polyunsaturated lipids to radical attack. In the propagation steps, dioxygen reacts with the carboncentered radical on the acyl chain (L·) forming a peroxyl radical (LOO·), which has the ability to extract another hydrogen from a second unmodified polyunsaturated hydrocarbon chain producing additional damage through a second carbon-centered radical (L·) and a lipid hydroperoxide (LOOH). This reaction is cyclical in propagating damage. The lipid hydroperoxide is relatively stable but can react with iron forming hydroxyl and alkoxyl radicals (LO·), of which both can abstract hydrogen. Iron involvement highlights increased iron levels and improperly coordinated iron in neurodegenerative diseases. Termination involves the formation of more stable or non-radical products (e.g., crosslinked LL), trapping by small molecule radical scavengers, or enzymatic repair. However, it should be noted that lipid hydroperoxide breakdown into reactive byproducts such as MDA and HNE, which are well-known markers for ROS damage and cause additional lipid and protein damage. For example, both MDA and HNE can cause numerous types of protein modifications, e.g., crosslinking between two proteins and modification of the structure of residues, thus hindering protein function and fostering aggregation. Taken together, iron can both initiate lipid and protein damage by hydroxyl radical formation, as well as enhance the rate of damage in a cyclical manner, especially with lipids. These two reactive byproducts are proximal markers for cell death involving uncontrollable lipid peroxidation through ferroptosis, as will be discussed shortly.

# FERRITIN IRON RELEASE BY FERRITINOPHAGY AND OTHER MECHANISMS

Under circumstances in which cellular iron is lacking, iron stored in native ferritin can be released for cytosolic distribution, and there are different potential mechanisms of various complexity and divergent pathways that may do so. Such mechanisms are lysosomal and proteasomal ferritin degradation, and removal of iron through the ferritin three-fold pores, which have clearly been established as the entrance channels for iron but not as exit pathways. Current focus as the most likely and dominant iron release mechanism in normal cells is lysosomal degradation of ferritin through an autophagic process called ferritinophagy (Biasiotto et al., 2015; Ndayisaba et al., 2019), which is complicated and requires substantial cellular machinery, i.e., transport, synthesis and degradation. However, there is an extensive literature on completely different mechanisms (Crichton, 2009; DeDomenico et al., 2009; Bou-Abdallah et al., 2018), and it is not unreasonable to suggest that a fine tuned cell would have more than one way to remove iron from ferritin under different conditions, especially one that minimizes use of cellular material and energy and is not spatially dispersed. Degradation of ferritin by ferritinophagy is a complex and spatially non-local process requiring significant cellular resources of synthesis and transport, and even proteosomal degradation is energetically costly and requires ATP both for labeling the target proteins with ubiquitin and the degradation process itself. Perhaps cells have also developed a more measured, finetuned, iron release pathway utilizing locally available cellular components such as cytosolic reductants and small molecule chelators, and such pathways have undergone substantial in vitro investigation (Melman et al., 2013; Koochana et al., 2018). Thus, there may be two (or more) mechanistic levels of stored iron release from native ferritin operating concurrently in the cell, one for nuanced, local iron balance and the other used under different circumstances, e.g., under iron-depleted cellular distress. In fact, given the importance of iron management, evolutionary sampling, and the complexity of cellular response to environmental change, it is not unexpected that there would be multiple mechanisms to inhibit and augment iron release from ferritin, which is an ancient protein. Below is discussed ferritinophagy in the context of an autophagic process as well as the alternate more local mechanisms of iron release.

Of the two major protein degradation pathways, ubiquitinproteasomal and lysosomal degradation, ferritinophagy depends upon the latter through a multistep pathway involving autophagosomes and the cargo protein Nuclear Receptor Coactivator 4 (NCOA4) (Mancias et al., 2014, 2015; Bellelli et al., 2016; Gatica et al., 2018). In this pathway, NCOA4 binds ferritin and is necessary for ferritin to be transported to lysosomes. First, ferritin binds to NCOA4 forming a complex through an FTH1 subunit. Next, the double walled proto-autophagosome membrane encompasses the ferritin NCOA4 complex forming a completely enclosed structure. Finally, the autophagosome fuses with the lysosome in which ferritin is degraded releasing iron. This pathway requires microtubules for transport and assembly (Hasan et al., 2006). Further details of the NCOA4 ferritin interaction are from (1) binding studies of a cloned and expressed portion of the NCOA4 protein (residues 383–522) and (2) its different interactions with FTH1, wtFTL, and single amino acid mutations of the FTH1 subunit (Gryzik et al., 2017). The FTH1 ferritin 24-mer can bind up to 24 NCOA4 proteins. Furthermore, FTH1 binds to the NCOA4 fragment tightly (in the nM range), but NCOA4 does not bind to wtFTL or to FTH1 after certain mutations. This binding specificity may be crucial to the explanation of the accumulation of ferritin into IBs, as will be discussed below.

HF is characterized by formation of IBs that are composed of mtFTL, wtFTL, and FTH1 subunits, and there is clear biochemical and histologic evidence of the accumulation of all three types of ferritin subunits in patients with HF and in the mouse model (Vidal et al., 2004, 2008). However, the ferritinophagy process is based on recognition of FTH1 in the ferritin 24-mer for lysosomal degradation and clearance, and structurally complete 24-mers can be formed from any combination of these subunits. Brain ferritin has FTH1 as the predominant subunit of its normal ferritin 24-mer assembly, and it is possible that imbalance in the ratio of mtFTL and/or wtFTL to FTH1 subunits caused by HF, i.e., a decrease in normal ferritin FTH1 subunit content in the 24-mer, could substantially and negatively affect NCOA4 binding to ferritin and its clearance by ferritinophagy. This could explain, at least in part, the buildup of IBs in HF. More likely, inhibition of NCOA4 binding to ferritin could be caused by just the presence of the mtFTL in the 24-mer because of interference from the extended Cterminus or by formation of ferritin aggregates by iron bridging. Thus, the buildup of cellular ferritin in the form of IBs is consistent with direct mutant-based inhibition of ferritinophagy through compromised NCOA4-FTH1 subunit binding. Inhibited ferritinophagy is particularly problematic in HF in that the elevated iron from four-fold pore misfunction is a driver to overproduction of both normal and mutant ferritin without appropriate clearance. This feedback process would likely tax cellular resources and eventually cause ferritin accumulation with multiple unwanted cellular interactions, e.g., association with susceptible proteins, inhibition of microtubule-based cellular transport, and altered spatial packing in a cumulative manner. In addition, mutant-containing ferritin, together with cellular reductants and iron would generate ROS. With ferritinophagy compromised and ferritin accumulation the cell would likely resort to proteasomal degradation of ferritin and the aggregates, and there is evidence of proteasomal constituents in IBs (Vidal et al., 2008). However, the proteasome is ill-equipped to handle ferritin because of its size vs. the dimensions of the proteasome interior. Ferritin 24-mers might dissociate into monomers, which are substantially smaller, however both normal and mutantcontaining ferritin are particularly stable. Further complicating disposal is that protein aggregates inhibit the proteasome. Partitioning of the proteasome into the vicinity of the ferritin aggregates may be a failed attempt to remove ferritin or it may be in response to ROS damage of other proteins caused by ferritin-iron aggregates. Taken together ferritin overexpression and compromised ferritinophagy are major components of HF, and mutant-induced inhibition of binding of ferritin to NCOA4 may contribute directly to small aggregate and IB formation.

Other proposed mechanisms of inhibition of ferritinophagy are dependent on elevated iron. More specifically, in the absence of HF, high iron levels (1) enhance proteosomal degradation of NCOA4 and (2) inhibit the binding between FTH1 subunit and NCOA4 (Mancias et al., 2015; Gryzik et al., 2017). However, the modest elevation of iron in HF suggests that neither process may dominantly inhibit ferritinophagy, and the mutation-based direct inhibition of may contribute in parallel with, or perhaps even dominate, the other two mechanisms. In addition, the mechanism by which ferritinophagy is decreased by iron-induced enhancement of proteosomal degradation of NCOA4 may not be operative in HF with the proteasome compromised by ferritin aggregates and unable to remove NCOA4. In any case, there may be as many as three mechanisms working to inhibit ferritinophagy at some stage of the progression of HF consistent with the gradual buildup of IBs.

Several alternative native ferritin iron release mechanisms that do not require complex cellular responses, but are instead more direct pathways centered on ferritin interaction with locally available cytosolic effector molecules, have been proposed from in vitro studies (Boyer et al., 1988; Crichton, 2009; Melman et al., 2013; Badu-Boateng et al., 2017; Johnson et al., 2017; Koochana et al., 2018). Most of the latter do however require (1) reduction of the iron mineral stored in the interior of the ferritin, (2) an iron chelator, and (3) a pathway through which the chelator and iron-chelator complex can pass through the shell. Cytosolic molecular reductants, e.g., NADPH and FADH, and chelators are relatively large bringing into question their ability to pass through the three-fold pores and gain access to the ferric iron mineral, as well as the rates in which these processes could occur. By Xray crystallography the pores are smaller than the dimensions of these cytosolic effector molecules. Even if the reductant does not need to enter the ferritin interior because electrons could transit the shell by a redox relay mechanism, the standard threefold pore configuration is not designed for chelated ferrous iron to have egress. Still, there have been reports of slow release of iron from the ferritin interior by treatment with combinations of reductants and chelators, and there is evidence that the crystallographic, nominal three-fold pore diameter of ∼3.4 Å relaxes to accommodate somewhat larger diameter molecules (Yang and Nagayama, 1995). In any event, such iron release would likely be slow compared with lysosomal iron release. In HF, limitations in reductant and chelator entry and exit are substantially reduced with the disordered four-fold pore.

Another iron release mechanism involves perturbing ferritin three-fold pore structure directly. Such studies were performed through amino acid mutations at the pore, binding of peptides to the surface of ferritin, and exposure to low levels of structural perturbants with human FTH1 24-mer ferritin and a frog FTH1 analog (Takagi et al., 1998; Jin et al., 2001; Liu et al., 2003, 2007; Hasan et al., 2008). Site-directed mutagenesis of specific amino acids e.g., FTH1 L138P, allows substantial modulation of the iron egress rate by increasing pore size and/or disorder. Furthermore, small peptides that bind to the surface of ferritin, which were found through screening assays, and even low concentrations of physiological perturbants modulate the iron exit rate of native frog ferritin. Thus, it is plausible that one or more pathways alternate to ferritinophagy may be operative in the overall process of cellular iron balance, although available reducing ability is necessary to dissolve the mineral core in all cases. These processes have the potential for more nuanced control of iron levels, perhaps with more rapid response and energetic efficiency than ferritinophagy. However, deconvoluting these competitive processes in a cellular environment without one masking another and the often use of excesses of perturbants, reductants, and chelators further complicates the already complicated scheme of iron balance.

## CELLULAR STRESS AND DESTRUCTION THROUGH FERROPTOSIS

Ferroptosis is becoming increasingly characterized as a major player in several neurodegenerative diseases (including AD and PD) as well as being understood to be a new form of Programmed Cell Death differing biochemically and morphologically from apoptosis, classical autophagy, and necrosis (Biasiotto et al., 2015; Gao et al., 2016; Angeli et al., 2017; Stockwell et al., 2017). Furthermore, close sequential biochemical linkage between the end point of ferritinophagy (iron release) and the ROS damage in ferroptosis suggests that ferritinophagy iron release is a component of, or perhaps an outright driver of, ferroptosis under certain conditions (Santana-Codina and Mancias, 2018; Sui et al., 2018, 2019). Interestingly, components of the ferroptotic process have been reported in the literature as isolated effects for some time now, but their connections into an overall picture were lacking. This was in part because cellular studies from diverse cell types found at different times that iron chelators, antioxidants (radical scavengers), and lipoxygenase inhibitors can cause inhibition of ferroptotic cell death (Cao and Dixon, 2016; Stockwell et al., 2017). Conversly, a different set of drugs and small molecules were found to induce ferroptosis. Thus,

it has been argued that ferroptosis is not really new and there is a suggested timeline to the evolution of its etiology (Hirschhorn and Stockwell, 2019). The two major cellular components characterizing ferroptosis are (1) iron presence or elevation and (2) accumulated lipid damage with inhibition of the lipid repair process. The pathways of several contributors to the overall ferroptotic process can be deconvoluted experimentally and recognized individually using small molecule ferroptosis inducers and inhibitors specific to each contributor (Cao and Dixon, 2016; Stockwell et al., 2017), although not all have been applied to cells involved in neurodegeneration. These studies are complicated in that ferroptosis is often induced by one small molecule and inhibited by another.

Paralleling HF, a picture of the crucial importance of iron availability or excess in ferroptosis is clear, and under some circumstances its excess may induce ferroptosis (Fang et al., 2018; Dixon and Stockwell, 2019). Rapid ferroptosis-induced cell death can be inhibited by preventing normal activity of the NCOA4 cargo protein and its ability to bring ferritin to the lysosomes for degradation through ferritinophagy, which releases toxic levels of iron into the LIP (Mancias et al., 2014; Gao et al., 2016). Iron chelators are able to circumvent ferroptosis initiated by several different small molecule inducers that work through different cellular pathways, but contribute to the overall ferroptotic process (Cao and Dixon, 2016). Furthermore, changing expression of ferroportin and transferrin receptor levels, i.e., exporters and importers of cellular iron, modulate ferroptosis as expected (Gao et al., 2016; Geng et al., 2018; Dixon and Stockwell, 2019). Recently, it was demonstrated that intracellular iron overload caused by treatment with a membrane permeable iron chelation can cause ferroptosis and cell death (Fang et al., 2018). Such studies implicate, in addition to the need for iron availability, that an increase in iron level itself may be a mechanism of ferroptosis initiation, especially in stressed cells where other compensating processes are depressed.

Lipid peroxidation is a defining process of ferroptosis leading to cell death, and there are two major pathways which can cause this: (1) ROS generated at improperly coordinated iron non-specifically attacking surrounding lipids and proteins, as well as iron enhancing the rate of lipid autooxidation, and (2) an iron-containing lipoxygenase (LOX) which detrimentally modifies phospholipids in a more specific (enzymatic) manner and requires dioxygen but not reducing equivalents. Both pathways likely occur to some extent in the ferroptotic process and the dominant one would depend on the amount of LIP iron accumulation without safe storage vs. LOX availability and expression. Recently, it was found that LOX inhibitors may more effectively function as radical scavengers than inhibitors (Shah et al., 2018), thus adding weight to improperly coordinated iron and ROS as a drivers of ferroptosis. Importantly, the process of ferroptosis-induced cell death is inhibited by phospholipid repair by a specific glutathione peroxidase (GPX4), but requires GSH to do so, and it is known that direct covalent inhibition of GPX4 prevents repair of damaged phospholipids enhancing ferroptosis (Yang and Stockwell, 2016). Restriction of GSH availability by inhibition of GSH synthesis or prevention of import of its precursor cystine also enhances ferroptosis.

In HF, the situation is more complicated in that (1) sequestration of iron into ferritin is already compromised to some extent with the leaky four-fold pores causing chronically elevated iron, which enhances ROS damage without the need for ferritinophagy releasing iron, and (2) the true extent of ferritinophagy inhibition in HF is unclear but may be substantial because there is significant but gradual ferritin IB accumulation. Moderately elevated iron level and oxidative damage in HF are suggestive of a gradual form of ferroptotic-like cell death, which is consistent with the relatively long life span of patients and animal models. Ongoing stress from unproductive and compensating cellular processes would likely be an important component. Protein aggregate formation from ferritin or other proteins may cause other forms of cumulative damage indirectly contributing to cell death on a time scale differing from ferroptosis, which is rapid. Further complicating the issue is that normal brain iron concentration increases with age (Weinreb et al., 2012) paralleling the likelihood of onset of several different neurodegenerative diseases, as is found with HF, and most of these diseases are suggested to have a mechanism-based iron component to their etiology.

# THE ROLE OF FERROPTOSIS IN HF—SIMILARITIES AND DIFFERENCES

Drawing a complete analogy between cell death from a long-term process like HF and a short term, small molecule-induced death process like ferroptosis is difficult in that the short term process would not necessarily exist long enough to allow production of long term, cumulative cellular damage. Both processes critically involve iron, and they have ROS and phospholipid damage as components. In HF ferritinophagy is likely eliminated or greatly reduced, with iron levels gradually increasing from four-fold pore leakage, which does not occur in ferroptosis. Although the time scales differ, both HF and ferroptosis appear to respond to increasing levels of iron in an increasingly negative manner. Because HF has an overproduction of ferritin and a C-terminus that allows for iron bridging, aggregates and IBs form over the long term. This process apparently does not occur in ferroptosis, but it would be interesting to examine what would occur if ferroptosis could be slowed into a more long-term process and the results compared to other forms of cellular dysfunction in neurodegenerative diseases. Perhaps protein aggregates from ROS damaged proteins and iron would occur in an extended form of ferroptosis more closely paralleling neurodegenerative diseases. In addition, GSH levels fall with age, in ferroptosis, and in neurodegenerative diseases (Emir et al., 2010), again pointing toward a parallel with a more long term ferroptotic-like state. Thus, the designation of long term ferroptotic-like state in HF is meant to capture the dependence on iron, ROS, lipid and protein damage, and stress, and perhaps declining GSH levels in both processes, but not to draw a complete molecular level parallel. Interestingly, this discussion brings up the question of time scales in contributions to neurodegenerative diseases.

# CONCLUSION

Not long ago, one could attend a seminar on major neurodegenerative diseases and hear no or only passing mention of transition metal ions. The same absence is still true of some current textbooks. These lectures and texts generally focused on aggregation and fibrillization alone. However, the current literature is becoming increasingly clear concerning the very important connections between iron and AD, PD, Prion disease, motor neurone disease, and other chronic slowly-progressing neurodegenerative diseases (Biasiotto et al., 2015; Ndayisaba et al., 2019; Weiland et al., 2019). This review emphasizes the complexity of the cellular problem at hand with respect to interactions of iron with certain proteins and small molecules as caused by HF. The inability of ferritin containing the mutant subunit to properly store iron and its modified and extended C-terminus have several consequences. These are seen as measured observables in elevated iron concentration, overproduction of ferritin, enhanced protein and lipid oxidation, ROS generation, and aggregation and deposition of ferritin into IBs. The elevated iron level in HF is causal to both production of ROS and ferritin overproduction to attempt to compensate for iron elevation. The mutant C-terminus and iron together foster ferritin aggregation leading to IBs. However, these cellular changes are not static endpoints and they themselves have far reaching cellular involvements and consequences.

ROS are generated (1) by reductants and oxygen at iron improperly coordinated by small molecules, mutant ferritin, and ferritin or other proteins that are unfolded or aggregated, (2) secondarily as protein and phospholipid reaction products from primary ROS attack, and (3) from enzymes that miscycle or for defensive purposes. Additionally, improperly coordinated iron may cause interconversions between various ROS, and these may increase toxicity. Once damaged by ROS, proteins tend to unfold exposing their hydrophobic interiors that associate and form aggregates independent of, but also potentially driven further by, iron binding. Additionally, this process may be cyclical because ROS can be generated from iron on aggregates and ROS attack and damage helps aggregate proteins. Cells must expand resource outlays and shift resources away from other tasks to attempt to correct for ROS overproduction, protein and lipid damage, and disposal of aggregates, which are common stressors in neurodegenerative diseases. These cellular outlays can take the form of additional synthesis of proteins to attempt to restore iron homeostasis, destroy ROS, repair proteins and phospholipids, and eliminate and transport protein aggregates, and also to maintain redox balance and energy availability for these as well as other normal cellular processes. These many processes are materially and energetically costly and stress the cell. Elevated iron and lipid peroxidation products as are found in HF are markers for ferroptosis, which may be occurring at a moderate level as the cells try to cope with the above-mentioned multiple stressors.

In HF, excess ferritin synthesis, in an attempt to control iron levels, diverts ribosomal proteins and energy away from other necessary cellular synthesis. Ferritin disposal is also a complex, interdependent and a somewhat redundant system perhaps involving both the proteasome and the lysosome, which require synthesis and energy. However, the extent and distribution between these becomes unclear in HF especially with respect to soluble ferritin vs. ferritin aggregates. Elements of the proteasomal machinery are evident in IBs in HF, however such is not likely to effectively clear excess ferritin because of the large size of the 24-mers and aggregates making the lysosomal clearance primary. NCOA4 binding to ferritin is required for ferritin to be transported to the lysosome and undergo ferritinophagy, but inclusion of the mtFTL subunit itself, or the resulting aggregation it induces, could hinder the binding of NCOA4 to ferritin for disposal thus enhancing accumulation into IBs. Under the circumstance that ferritin aggregates are not disposed of by ferritiophagy, the proteasome likely attempts compensation. However, the formation of IBs in HF suggests that the process of IB formation outpaces the ability of both proteasomal and lysosomal degradation, and proteasomal proteins may be synthesized but not effectively used for ferritin clearance. These several cellular processes meant to compensate or correct for the presence of mtFTL are all energetically and materially costly, and stress the cell.

This review makes clear the complex interrelationships in HF as the cells are stressed and muster their efforts at maintaining overall homeostasis. The two key pathogenic mechanisms for the phenotypic expression of the disease are best explained (1) as a loss of normal ferritin function with decreased iron incorporation that triggers intracellular iron accumulation and overproduction of ferritin polypeptides, and (2) a gain of toxic function through ROS production, ferritin aggregation, and oxidized protein and lipid formation (Muhoberac and Vidal, 2013; Vidal and Ghetti, 2015). However, this phenotype does not fully emphasize the layered cellular damage signified by oxidation and aggregate formation, which may have parallels in a number neurodegenerative diseases and lead to a major loss of normal cellular function. The responses to (1) and (2) above tax cells energetically, and with respect to material resource availability and management, causing gradual, but cumulative damage. Even proteasomal action is inhibited by aggregated proteins. Interestingly, it is found that there is only a modest increase in iron levels (10–20%) in cells and animal models vs. the rather large associated increase in ferritin (200–400%) found with them. Moderate iron level increases make sense in that a substantial increase would likely be more rapidly lethal to the cells through ROS production and aggregate formation, and could shut down ferritinophagy completely if any portion was still functional. Both humans with HF and animal models do exhibit moderate life spans and gradually, not rapidly form IBs. Still, because iron is redox active even low concentrations can cause ROS damage, especially if the cellular stress is long term and to some extent cumulative.

The presence of IBs is a diagnostic marker for HF, and they are increased much more than the iron level. The chronicprogressive increase of IB (number and size) through the

FIGURE 6 | circles), two caused by increased iron levels and the third by the C-terminal mutation, in addition to ferritin aggregates directly inhibiting microtubule transport and proteosomal degradation, leading to buildup of ferritin aggregates and IBs. The combination of accumulation of ferritin aggregates and ROS damage causes cellar dishomeostasis leading to several forms of cellular stress (rightmost box). (b) ROS can originate with cellular iron that is improperly coordinated (see box), as well as by enzymes that misfire or form them normally. ROS can be converted to other ROS forms, sometimes more toxic, by improperly coordinated iron or native enzymes. Certain enzymes have the function to eliminate ROS as cellular protectors. Both lipid and protein damage occur from ROS leading to local cellular loss of function, and protein damage can lead to aggregation further enhancing and spreading ROS production. (c) Mutant-containing ferritin clearly leads to increased iron levels, which in turn causes ROS-induced protein and lipid damage in HF. However, compensating for the long-term dishomeostasis (see box in a) caused by mutant ferritin stresses the cell, diverting resources and energy away from bulk metabolism and taxing cellular transport and repair mechanisms, which contributes to a long-term cellular weakening process. Increased iron and lipid peroxidation, which are both found in HF, are hallmarks of ferroptosis, which leads to cell death and is associated with several neurodegenerative diseases. The causative factors in cell death in HF appear to be not only iron accumulation and ferritin aggregate formation, but additionally, a combination of several gradual but cumulative forms of overall stress (a, rightmost box) causing cellular compensatory breakdown that taken together cause a long-term ferroptotic-like cellular state.

duration of the disease points toward a serious failure of the systems attempting to manage cellular ferritin disposition, i.e., lysosomal degradation through ferritinophagy. NCOA4-FTH1 binding is inhibited by high iron concentration in purified protein studies and NCOA4 degradation is enhanced (Mancias et al., 2015; Gryzik et al., 2017), however these may not be enough to substantially suppress ferritinophagy. Thus, the modest iron increases found in in vivo studies of HF do not rule out an additional, perhaps dominant contribution to hindered NCOA4-ferritin binding and decreased ferritinophagy that is mechanistically separate from the two above-mentioned iron level effects. Importantly, it is likely that the presence of the partially unfolded mtFTL subunit in ferritin would decrease NCOA4-ferritin binding directly through hindering the NCOA4- FTH1 interaction and also through enhanced ferritin aggregation making NCOA4-FTH1 interaction spatially difficult to achieve. Additionally, aggregates of ferritin would be difficult to transport to the lysosome or aggresomes. This may provide a mechanism for the gradual buildup of aggregates into dispersed IBs that is not dependent on high iron levels.

Even recently, the contribution of iron-induced aggregation has been suggested to be unimportant in the etiology of HF (Luscieti et al., 2010; Levi and Rovida, 2015). Such aggregation can occur through C-terminal bridging of ferritin containing the mtFTL subunit or by ROS oxidation and proteolysis of ferritin destabilizing its structure. ROS formation originates from a variety of cellular sources with improperly coordinated iron, such as small LIP carrier molecules, the mutant FTL C-terminus, protein proteolysis products, or unfolded or aggregated proteins. Reductants such as ascorbate are clearly available in the brain and the Haber-Weiss and Undenfriend-type reactions use ascorbate and iron to generate various damaging ROS depending on iron coordination. The process of protein aggregation, especially with ferritin containing the mtFTL subunit unraveled C-termini, will sample a variety of conformations as the initial aggregates are formed. It follows that iron, especially but not necessarily in great excess, could sample a variety of coordinations with some being facile ROS producers, and this sampling would likely occur not only with ferritin and HF but with other proteins and neurodegenerative diseases. Interestingly, the finding that HF and HF models produce a substantial fraction of SDS-insoluble ferritin brings up the possibility that the IBs, once formed, could hinder penetration of a reductant as large as ascorbate from their interior and keep iron safely six coordinate and oxidized. ROS generation on the surface of the IBs may occur, but the surface area to volume ratio would suggest a net reduction in damage by IB mutant ferritin over what would occur with small dispersed mutant ferritin aggregates and iron. The proteasome may target more than mutant ferritin alone, including repair or removal of other ROS-damaged proteins. However, IB buildup, transport and elimination is another issue.

On the negative side, the size and number of IBs increase with age, and formation of a large number of IBs would likely hinder crucial cytoplasmic transport or nuclear processes. Non-mutant ferritin aggregates are known to associate with tubulin (Hasan et al., 2006) and ferritin associates with kinesin (Jang et al., 2016), suggesting the mutant ferritin aggregates could be problematic in cytoskeletal transport disruption in that transport requires these proteins. Disruption of transport would scale with the age of the cells, and not only hinder ferritinophagy, but a number of other crucial cellular transport processes. Thus, damage could occur in a cyclical manner through inhibition of required ongoing cellular processes involving transport and clearance.

The overall pathway in HF from mutant ferritin expression to cellular dysfunction and death is a long-term stressful process. The gradual nature of HF development suggest that cells do cope with ferritin overexpression and elevated iron for some time before becoming critically dysfunctional. Increased iron in the LIP leads to ROS production and overproduction of ferritin with elimination of ferritin compromised (**Figure 6a**). These two misfunctions lead to a more general cellular dishomeostasis by causing several stressors taxing the cells and diverting energy and material away from bulk metabolism, transport, and repair. ROS production is strongly associated with improperly coordinated iron (**Figure 6b**) which can become more numerous as the ferritin mutant, is aggregates, and other misfolded proteins and their aggregates accumulate and provide iron-binding sites. Cells expend synthesis energy to eliminate ROS and repair damage. The increases in iron and lipid damage, which are known markers for ferroptosis, are clearly observed in HF, but the moderate increase in iron levels and gradual, cumulative nature of the overall damage suggests a more general, long term, cellular dishomeostasis. Still, the moderately elevated iron and ROS lipid damage, modified by the gradual, cumulative cellular dishomeostasis, may produce a long term ferroptotic-like state (**Figure 6c**). It is worth mentioning a cautionary note that many in vivo and in vitro studies of iron homeostasis and cellular stress are made (1) under conditions of chelator, reductant, and ROS concentrations and (2) over time scales that are designed to elicit a change that mimics some aspect of a disease state. However, these studies may be unrepresentative of the low physiological concentrations and long time course of the disease under investigation, especially if it takes years to develop such as with most neurodegenerative diseases.

This review on HF details biochemical aspects of metal ion interactions, protein misbehavior, and cellular responses that may have parallels or commonality to other neurodegenerative diseases, especially with respect to ROS formation and damage, as well as protein aggregation, inclusion formation, and disposal pathway inhibition. With respect to HF treatment, we can add two additional approach to our previous suggestion of simultaneous administration of a radical scavenger and appropriate iron ion chelator (Muhoberac and Vidal, 2013). Importantly, Ftl−/<sup>−</sup> mice show that because FTH1 ferritin homopolymers can maintain brain iron homeostasis in a mouse model, there is the potential for development of a therapeutic approach for HF treatment using RNA interference to induce sequence-specific post-transcriptional gene silencing of mutant

# REFERENCES


FTL or alternatively both, wild-type and mutant FTL alleles (Li et al., 2015). In addition, drug-based manipulation or enhancement at some level of the cellular disposal mechanisms may reduce ongoing ferritin and aggregate accumulation, and thus reduce cellular stress enhancing survival. It is hoped that researchers in the field of neurodegeneration will be able to find parallels and apply them to their individual systems of study and lead to more robust treatments.

#### AUTHOR CONTRIBUTIONS

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

## FUNDING

This study was supported by grants from the National Institute on Neurological Disorders and Stroke NS050227 and NS063056 and the National Institute on Aging AG10133.

Boopathi, S., and Kolandaivel, P. (2016). Fe2<sup>+</sup> binding on amyloid β-peptide promotes aggregation. Proteins 84, 1257–1274. doi: 10.1002/prot.25075


caused by the 458dupA FTL mutation. Brain 132:e109. doi: 10.1093/brain/ awn274


**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 Muhoberac and Vidal. 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