# MOLECULAR, CELLULAR AND MODEL ORGANISM APPROACHES FOR UNDERSTANDING THE BASIS OF NEUROLOGICAL DISEASE

EDITED BY: Robert J. Harvey and Kirsten Harvey PUBLISHED IN: Frontiers in Molecular Neuroscience

#### *Frontiers Copyright Statement*

*© Copyright 2007-2017 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

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

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

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

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

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

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

ISSN 1664-8714 ISBN 978-2-88945-173-9 DOI 10.3389/978-2-88945-173-9

#### 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

# **MOLECULAR, CELLULAR AND MODEL ORGANISM APPROACHES FOR UNDERSTANDING THE BASIS OF NEUROLOGICAL DISEASE**

Topic Editors: **Robert J. Harvey,** UCL School of Pharmacy, London, UK **Kirsten Harvey,** UCL School of Pharmacy, London, UK

The advent of next-generation sequencing technologies has resulted in a remarkable increase our understanding of human and animal neurological disorders through the identification of disease causing or protective sequence variants. However, in many cases, robust disease models are required to understand how changes at the DNA, RNA or protein level affect neuronal and synaptic function, or key signalling pathways. In turn, these models may enable understanding of key disease processes and the identification of new targets for the medicines of the future. This e-book contains original research papers and reviews that highlight either the impact of next-generation sequencing in the understanding of neurological disorders, or utilise molecular, cellular, and whole-organism models to validate disease-causing or protective sequence variants.

**Citation:** Harvey, R. J., Harvey, K., eds. (2017). Molecular, Cellular and Model Organism Approaches for Understanding the Basis of Neurological Disease. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-173-9

# Table of Contents

*05 Editorial: Molecular, Cellular and Model Organism Approaches for Understanding the Basis of Neurological Disease*

Robert J. Harvey and Kirsten Harvey

*07 Missense Mutation R338W in ARHGEF9 in a Family with X-linked Intellectual Disability with Variable Macrocephaly and Macro-Orchidism* Philip Long, Melanie M. May, Victoria M. James, Simone Grannò, John P. Johnson,

Patrick Tarpey, Roger E. Stevenson, Kirsten Harvey, Charles E. Schwartz and Robert J. Harvey

*15 Novel Missense Mutation A789V in IQSEC2 Underlies X-Linked Intellectual Disability in the MRX78 Family*

Vera M. Kalscheuer, Victoria M. James, Miranda L. Himelright, Philip Long, Renske Oegema, Corinna Jensen, Melanie Bienek, Hao Hu, Stefan A. Haas, Maya Topf, A. Jeannette M. Hoogeboom, Kirsten Harvey, Randall Walikonis and Robert J. Harvey

*25 Rare Variants in Neurodegeneration Associated Genes Revealed by Targeted Panel Sequencing in a German ALS Cohort*

Stefanie Krüger, Florian Battke, Andrea Sprecher, Marita Munz, Matthis Synofzik, Ludger Schöls, Thomas Gasser, Torsten Grehl, Johannes Prudlo and Saskia Biskup


Katharina Kuenzel, Oliver Friedrich and Daniel F. Gilbert


Pekka Poutiainen, Merja Jaronen, Francisco J. Quintana and Anna-Liisa Brownell

#### *128 Function Over Form: Modeling Groups of Inherited Neurological Conditions in Zebrafish*

Robert A. Kozol, Alexander J. Abrams, David M. James, Elena Buglo, Qing Yan and Julia E. Dallman


Yasuhito Watanabe, Michaela K. Müller, Jakob von Engelhardt, Rolf Sprengel, Peter H. Seeburg and Hannah Monyer

*171 Molecular Mechanisms Regulating LPS-Induced Inflammation in the Brain* Olena Lykhmus, Nibha Mishra, Lyudmyla Koval, Olena Kalashnyk, Galyna Gergalova, Kateryna Uspenska, Serghiy Komisarenko, Hermona Soreq and Maryna Skok

# Editorial: Molecular, Cellular and Model Organism Approaches for Understanding the Basis of Neurological Disease

Robert J. Harvey \* and Kirsten Harvey

*Department of Pharmacology, UCL School of Pharmacy, London, UK*

Keywords: next-gen sequencing, genetics, cell line, zebrafish, mouse

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

#### **Molecular, Cellular and Model Organism Approaches for Understanding the Basis of Neurological Disease**

Next-generation sequencing technologies have resulted in remarkable increases in our understanding of human neurological disorders through the identification of disease-causing or protective sequence variants. However, in many cases, new molecular, cellular, and whole-organism models are required to understand how changes at the DNA, RNA, or protein level affect neuronal and synaptic function. In turn, these models may enable understanding of key disease processes and the identification of new targets for the medicines of the future. What is also evident is the high quality and impact of the research conducted in this field. This is reflected in the reviews and research articles in this Special Issue entitled "Molecular, cellular and model organism approaches for understanding the basis of neurological disease."

Mutation and gene discovery in neurological diseases has recently been transformed by largescale DNA sequencing approaches coupled with stringent variant filtering. This is particularly true of intellectual disability, characterized by significantly impaired intellectual and adaptive function. Long et al. and Kalscheuer et al. highlight the utility of this methodology by resolving two families with X-linked intellectual disability caused by mutations in ARHGEF9 and IQSEC2, which both encode neuronal GDP–GTP exchange factors. One important aspect of these studies was functional validation of the pathogenic variants using molecular modeling, phosphoinositide binding, gephyrin clustering, or GDP–GTP exchange activity. By contrast, Krüger et al. reported on the use of panel sequencing to study a cohort of 80 German patients with Amyotrophic lateral sclerosis (ALS), a rapidly progressive, fatal neurological disease that causes degeneration of motor neurones. This approach allowed deep sequencing of 39 confirmed ALS genes and candidate genes, as well as 238 genes associated with other neurodegenerative diseases. They identified 79 rare potentially pathogenic variants in 27 ALS-associated genes, as well as pathogenic hexanucleotide repeats in C9orf72. Based on this study, the authors recommend two-staged genetic testing for ALS in patients with familial and sporadic ALS, comprising C9orf72 repeat analysis followed by comprehensive panel sequencing.

It is also important to remember that not all rare genetic variants are disease causing. Nixon-Abell et al. reported the detailed molecular and cellular characterization of a protective variant in LRRK2, encoding leucine-rich repeat kinase 2 (LRRK2) which is intimately associated with the pathogenesis of Parkinson's disease. Importantly, the p.R1398H variant affects GTPase function, axon outgrowth, and Wnt signaling in a manner opposite to pathogenic LRRK2

Edited and reviewed by: *Nicola Maggio, The Chaim Sheba Medical Center, Israel*

> \*Correspondence: *Robert J. Harvey r.j.harvey@ucl.ac.uk*

Received: *14 February 2017* Accepted: *03 March 2017* Published: *20 March 2017*

#### Citation:

*Harvey RJ and Harvey K (2017) Editorial: Molecular, Cellular and Model Organism Approaches for Understanding the Basis of Neurological Disease. Front. Mol. Neurosci. 10:74. doi: 10.3389/fnmol.2017.00074* mutations. The authors concluded that LRRK2-mediated Wnt signaling and GTPase function are fundamental in conferring disease susceptibility, and that this has clear implications for future therapeutic interventions in Parkinson's disease.

Sophisticated cellular models of disease are increasingly vital as high-throughput screening tools for recreating events at normal and disrupted synapses in vitro. Kuenzel et al. reported a new tool for screening of in vitro neurotoxicity (NT) and developmental neurotoxicity (DNT) mediated by inhibitory γ-aminobutyric acid type A (GABAA) and glycine receptors (GlyRs). They generated a human pluripotent stem cell line (NT2) that stably expresses YFPI152L, a halide-sensitive variant of YFP, which allows for fluorescence-based functional analysis of Cl<sup>−</sup> channels. Importantly, upon stimulation with retinoic acid, these NT2 cells undergo neuronal differentiation, allowing pharmacological and toxicological evaluation of native GABAARs and GlyRs at different developmental stages. By contrast, Dixon et al. reported an "artificial synapse" system a neuron-HEK293 cell co-culture technique for generating inhibitory synapses incorporating defined combinations of wildtype or mutant GABAAR or GlyR subunits. As well as allowing control over the subunit composition of the GlyRs under study, the electrotonically compact shape of HEK293 cells combined with rapid agonist application allows IPSC waveforms to be resolved with high fidelity using electrophysiology.

The role of the GlyR M3-M4 intracellular domain in health and disease was also reviewed by Langlhofer and Villmann who comprehensively documented molecular determinants of phosphorylation, intracellular sorting, protein–protein interactions, subunit topology, and modulation by G proteins, ethanol, and cannabinoids. Barral and Kurian also reviewed the current and future potential of patient-derived induced pluripotent stem cells (iPSCs) in the field of childhood neurological disorders. Importantly, iPSCs can now be routinely differentiated into specific neuronal subtypes which represent valuable in vitro models of disease. They are also vital tools for testing existing drugs with repurposing potential, or novel compounds and gene therapies, which then can be translated to clinical practice. Imaging also has a key role to play in diagnosis of disease. Poutiainen et al. reviewed advances in non-invasive imaging techniques such as positron emission tomography (PET) and prospective biomarkers for the diagnosis of multiple sclerosis.

Whole-organism in vivo models also continue to make an impact in this field. Kozol et al. and Ogino and Hirata reviewed the many advantages of zebrafish in the study of neurological disease. Validation of new disease genes and mutations is now possible through CRISPR/Cas9 mutagenesis of zebrafish gene orthologs, which rapidly creates new disease models which are amenable to phenotyping and high-throughput drug screening. Watanabe et al. also highlighted the use of sophisticated mouse models, showing how doxycycline-sensitive, Cre-mediated gene ablation of NMDA receptors in hippocampal excitatory neurons results in neurodegeneration in aging mice. This study highlighted the potentially damaging effects of long-term administration of NMDAR antagonists for therapeutic purposes. Lykhmus et al. used an acute model of LPS-induced neuroinflammation in mice and an in vitro model in cultured glioblastoma U373 cells to study neuroinflammation, which accompanies and often precedes the development of neurodegenerative pathologies such as Parkinson's and Alzheimer's diseases. They found that acute LPS-induced inflammation induced the cholinergic antiinflammatory pathway in the brain, with down-regulation of α7 subunit-containing nAChRs limiting these effects. Curiously, nAChR α7 subunit-specific antibodies aggravated neuroinflammation, by inducing the pro-inflammatory interleukin-6 and dampening anti-inflammatory miRNAs. However, the authors also highlighted that nAChR α7-specific antibodies may protect brain mitochondria and decrease the levels of pro-apoptotic miRNAs, preventing LPS-induced neurodegeneration.

We thank all contributors for their interesting and informative articles and the reviewers for their constructive and thoughtful suggestions.

#### AUTHOR CONTRIBUTIONS

KH and RJH wrote the manuscript and both authors approved the final version for publication.

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

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

# Missense Mutation R338W in ARHGEF9 in a Family with X-linked Intellectual Disability with Variable Macrocephaly and Macro-Orchidism

Philip Long<sup>1</sup> , Melanie M. May <sup>2</sup> , Victoria M. James 1† , Simone Grannò<sup>1</sup> , John P. Johnson<sup>3</sup> , Patrick Tarpey <sup>4</sup> , Roger E. Stevenson<sup>2</sup> , Kirsten Harvey <sup>1</sup> , Charles E. Schwartz <sup>2</sup> \* and Robert J. Harvey <sup>1</sup> \*

<sup>1</sup> Department of Pharmacology, UCL School of Pharmacy, London, UK, <sup>2</sup> JC Self Research Institute, Greenwood Genetic Center, Greenwood, SC, USA, <sup>3</sup> Department of Medical Genetics, Shodair Children's Hospital, Helena, MT, USA, <sup>4</sup> Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, UK

#### Edited by:

Jean-Marc Taymans, UMR1172, Jean-Pierre Aubert Research Center, France

#### Reviewed by:

Angel L. De Blas, University of Connecticut, USA Silvia Bassani, CNR Institute of Neuroscience, Italy

\*Correspondence:

Charles E. Schwartz ceschwartz@ggc.org; Robert J. Harvey r.j.harvey@ucl.ac.uk

†Present address: Victoria M. James, Wellcome Trust, London, UK

Received: 16 October 2015 Accepted: 14 December 2015 Published: 20 January 2016

#### Citation:

Long P, May MM, James VM, Grannò S, Johnson JP, Tarpey P, Stevenson RE, Harvey K, Schwartz CE and Harvey RJ (2016) Missense Mutation R338W in ARHGEF9 in a Family with X-linked Intellectual Disability with Variable Macrocephaly and Macro-Orchidism. Front. Mol. Neurosci. 8:83. doi: 10.3389/fnmol.2015.00083 Non-syndromal X-linked intellectual disability (NS-XLID) represents a broad group of clinical disorders in which ID is the only clinically consistent manifestation. Although in many cases either chromosomal linkage data or knowledge of the >100 existing XLID genes has assisted mutation discovery, the underlying cause of disease remains unresolved in many families. We report the resolution of a large family (K8010) with NS-XLID, with variable macrocephaly and macro-orchidism. Although a previous linkage study had mapped the locus to Xq12-q21, this region contained too many candidate genes to be analyzed using conventional approaches. However, X-chromosome exome sequencing, bioinformatics analysis and segregation analysis revealed a novel missense mutation (c.1012C>T; p.R338W) in ARHGEF9. This gene encodes collybistin (CB), a neuronal GDP-GTP exchange factor previously implicated in several cases of XLID, as well as clustering of gephyrin and GABA<sup>A</sup> receptors at inhibitory synapses. Molecular modeling of the CB R338W substitution revealed that this change results in the substitution of a long electropositive side-chain with a large non-charged hydrophobic side-chain. The R338W change is predicted to result in clashes with adjacent amino acids (K363 and N335) and disruption of electrostatic potential and local folding of the PH domain, which is known to bind phosphatidylinositol-3-phosphate (PI3P/PtdIns-3-P). Consistent with this finding, functional assays revealed that recombinant CB CB2 R338W SH3<sup>−</sup> was deficient in PI3P binding and was not able to translocate EGFPgephyrin to submembrane microaggregates in an in vitro clustering assay. Taken together, these results suggest that the R338W mutation in ARHGEF9 is the underlying cause of NS-XLID in this family.

Keywords: ARHGEF9, collybistin, gephyrin, PH domain, XLID

## INTRODUCTION

Intellectual disability (ID) is characterized by significantly impaired intellectual and adaptive function, and is often defined by an IQ score below 70 in addition to deficits in two or more adaptive behaviors (e.g., social skills, problem solving) that affect everyday life. ID is also subdivided into syndromal ID, where ID is associated with other clinical, morphological, or behavioral symptoms or non-syndromal ID, where intellectual deficits appear without other associated defects (Stevenson et al., 2012). X-linked intellectual disability (XLID) refers to forms of ID typically associated with Xlinked recessive inheritance. Mutations in monogenic XLID have been reported in >100 genes, many of which are now used in routine diagnostic screening panels (Basehore et al., 2015). Despite screening for mutations in selected known XLID genes by conventional linkage/candidate gene analysis or array CGH for examining copy number variants (CNVs), large number of families mapping to the X-chromosome remained unresolved (Lubs et al., 2012). These cases either represent undiscovered disease-relevant mutations in known genes, or causal mutations in novel XLID loci that remain to be identified.

Mutation and gene discovery in XLID has recently been transformed by large-scale DNA sequencing approaches coupled with stringent variant filtering (Tarpey et al., 2009; Rauch et al., 2012; Gilissen et al., 2014; Redin et al., 2014; Hu et al., 2015; Niranjan et al., 2015; Tzschach et al., 2015). For example, a recent study in a large cohort of unresolved families with XLID revealed that 20% of families carried pathogenic variants in established XLID genes (Hu et al., 2015), as well as revealing seven novel XLID genes (CLCN4, CNKSR2, FRMPD4, KLHL15, LAS1L, RLIM, and USP27X) and two candidates (CDK16 and TAF1). Another strategy, known as Affected Kindred/Cross-Cohort Analysis (Niranjan et al., 2015) has also identified variants in known and novel XLID genes including PLXNA3, GRIPAP1, EphrinB1 and OGT. However, analysis of next-generation data sets has also highlighted a number of XLID genes where truncating variants or previously published ''mutations'' are observed at a relatively high frequency in normal controls, calling into question whether certain single nucleotide variants (SNVs) are indeed causal (Tarpey et al., 2009; Piton et al., 2013). This highlights the need for integrating structure-function based approaches into the analysis pipeline for validating potentially disease-causing variants.

In this study, we have combined next-generation sequencing, variant filtering and structure-function assays to resolve the cause of XLID in a large family (K8010) with NS-XLID, with variable macrocephaly and macro-orchidism (Johnson et al., 1998). The previous clinical study in this family revealed ten affected males and two affected females in two generations, as well as four obligatory carriers. Most affected males exhibited macrocephaly and macro-orchidism, which are typical signs of the fragile X syndrome. However, cytogenetic testing and analysis of FMR1 indicated they did not have this syndrome. It was also notable that some normal males in the family also exhibited macro-orchidism and macrocephaly. Linkage analysis suggested that the causative gene was located on Xp11-q21 (Johnson et al., 1998). We present compelling evidence that the likely cause of XLID in this family is a missense mutation in ARHGEF9, encoding a neuronal RhoGEF known as collybistin (CB) involved in both inhibitory synaptic organization and mammalian target of rapamycin complex 1 (mTORC1) signaling pathways (Machado et al., 2015).

### MATERIALS AND METHODS

#### Subjects

Family K8010 was previously reported by Johnson et al. (1998). Briefly, the males resembled those with fragile X syndrome in that they had macrocephaly (abnormally large head, typically 2.5 standard deviations above normal for weight and gender), macro-orchidism (abnormally large testes), blue eyes, prominent jaw and long facies. Additionally, one carrier female was described as being ''slow''. However, using linkage analysis, the locus was mapped to Xq12-q21 rather than Xq27.3. Additionally, FMR1 gene analysis was negative (Johnson et al., 1998).

### Exon Capture and DNA Sequencing

Next generation sequencing was conducted as a partial followup of 15 probands from a large scale sequencing of 718 X-chromosome genes in 208 XLID probands (Tarpey et al., 2009). The deep sequencing was conducted using the Agilent SureSelect Human X chromosome kit (Takano et al., 2012). A novel and unique mutation in ARHGEF9, c.1012C>T, was noted in K8010. Segregation analysis was conducted using Sanger sequencing.

#### Polymorphism Analysis

Screening of 566 normal individuals (420 males, 146 females) for the ARHGEF9 c.1012C>T variant was carried out using allele-specific amplification. Primers used were: ARHGEF9- ASOF 5<sup>0</sup> -TACGGCCGCAACCAGCtGt 3<sup>0</sup> and ARHGEF9- ASOR 5<sup>0</sup> -CCCATCAGTATTTGCCCACT-3<sup>0</sup> . The ASOF primer recognized the mutation, which is indicated by the ''t'' at the 3 0 end of the primer. The third base from the 3<sup>0</sup> end was also changed from an ''a'' to a ''t'' to increase the specificity of the PCR. The ASOR primer was designed so that the T<sup>m</sup> of both primers were similar and to generate a PCR product of above 500 bp. Gradient duplex PCR analysis was conducted using mutation and normal samples to choose the optimum annealing temperature. High-throughput duplex PCR analysis with a mutation and normal sample as positive and negative controls.

#### Molecular Modeling of the Collybistin R338W Mutation

The non-synonymous R338W substitution was modeled into the structure of rat CB (PDB 2DFK; Xiang et al., 2006) using the swapaa command in Chimera (Pettersen et al., 2004) using the Dunbrack backbone-dependent rotamer library (Dunbrack, 2002). This took into account the lowest clash score, highest number of H-bonds and highest rotamer probability. Electrostatic potential of wild-type and R338W mutant CB was calculated using the Adaptive Poisson-Boltzmann Solver (APBS) web server (Baker et al., 2001; http://www.poissonboltzmann.org/).

### Site-Directed Mutagenesis and Expression Constructs

Full-length human CB cDNAs were cloned into the vector pRK5 as previously described (Kalscheuer et al., 2009). Mutations were introduced into pRK5myc-hCB3SH3<sup>−</sup> construct using the QuikChange site-directed mutagenesis kit (Agilent) and confirmed by Sanger DNA sequencing of the entire coding region.

## PI3P Pull-Down Assays

Human embryonic kidney (HEK293) cells were grown in DMEM supplemented with 10% (v/v) fetal bovine serum at 37◦C, 5% CO<sup>2</sup> and transfected with 4 µg pRK5myc-hCB3SH3<sup>−</sup> wild-type or R338W mutants using FuGENE (Roche). After 24 h, transfected cells were solubilized in a buffer containing Triton X-100 (Sigma-Aldrich), 1%; 150 mM NaCl; 50 mM Tris, pH 7.4, with protease inhibitor cocktail (Roche, Sussex, UK). Insoluble material was removed by centrifugation at 16, 100× g for 20 min. Phosphatidylinositol-3-phosphate (PI3P/PtdIns-3-P) agarose beads (40 µl; Eschelon Biosciences) were incubated with cell lysates for 2 h at 4◦C. Beads were washed four times in buffer. Proteins were eluted from beads by heating at 98◦C for 3 min in 2 × sample loading buffer and then subjected to SDS-PAGE. Proteins binding to beads were detected by Western blotting using mouse anti-c-myc antibody (Sigma, 1:1000) and HRP-conjugated goat anti-mouse (Santa Cruz, 1:2000). Immunoreactivity was visualized using West Pico Chemiluminescent Substrate (Pierce). Expression levels of hCB3SH3<sup>−</sup> and hCB3SH3−R338W, and PI3P pulldown assay results were assessed using an unpaired, two-tailed Student's t-test.

#### Gephyrin Clustering Assays

These were performed essentially as previously described (Harvey et al., 2004). HEK293 cells were co-transfected with pRK5myc-hCB3SH3<sup>−</sup> wild-type, pRK5myc-hCB3SH3<sup>+</sup> wild-type, pRK5myc-hCB3SH3<sup>−</sup> R338W, pRK5myc-hCB3SH3<sup>−</sup> R290H or pRK5myc-hCB3SH3<sup>−</sup> R356N/R357N constructs at a 1:1 ratio with pEGFP-gephyrin using electroporation (Gene Pulser II, Bio-Rad). Cells were fixed after 24 h for 2 min in 4% (w/v) PFA in PBS. Immunostaining to detect CB was performed using a mouse anti-c-myc antibody (1:200, Sigma) and detected using an AlexaFluor 546 goat anti-mouse secondary antibody (1:600; Invitrogen). Counterstaining for cell nuclei was performed with DAPI (1:500; Life Technologies). Confocal microscopy was performed using a Zeiss LSM 710 META. All images were taken with a × 63 objective.

### RESULTS

### Identification of a R338W Mutation in ARHGEF9 in Family K8010

X-chromosome exome sequencing of an individual male in family K8010 followed by bioinformatics analysis and filtering against publicly-available datasets revealed a novel missense change in ARHGEF9, chrX:62,885,810G>A, c.1012C>T; p.R338W, predicted as probably damaging (PolyPhen-2, score 1.000), damaging (SIFT) and a Combined Annotation Dependent Depletion (CADD) score of 19.37 (possibly pathogenic). This suggested that this missense mutation could be responsible for XLID, macrocephaly and macro-orchidism in this family. Subsequent segregation analysis using Sanger DNA sequencing indicated that the ARHGEF9 c.1012C>T variant cosegregated with the phenotype in all individuals tested (**Figures 1A,B**).

### Collybistin Mutation R338W is Predicted to Disrupt PH Domain Folding

CB belongs to the Dbl family of guanine nucleotide exchange factors, occurs in multiple splice variants (Kins et al., 2000; Harvey et al., 2004) and is specific for Cdc42, a small GTPase belonging to the Rho family (Xiang et al., 2006). CB has a multi-domain structure consisting of a regulatory SH3 domain, a catalytic RhoGEF domain and a pleckstrin homology (PH) domain (**Figure 1C**). Residue R338W is the first reported ARHGEF9 missense mutation affecting a highly conserved residue in the PH domain (**Figure 1D**). Previous nonsense and missense mutations in CB have affected residues in the N-terminus (p.Q2X; Shimojima et al., 2011), regulatory SH3 domain (p.G55A; Harvey et al., 2004) and catalytic RhoGEF domain (p.R290H; Lemke et al., 2012; Papadopoulos et al., 2015). To assess how the R338W substitution might disrupt CB function, molecular modeling was performed using the structure of rat CB (PDB 2DFK; Xiang et al., 2006), which has a sequence identity of 85.5% to human CB [aligned using HHalign algorithm (Söding, 2005) within Clustal-Omega (Sievers et al., 2011)]. The R338W change replaces a long, electropositive side-chain (arginine) with a large non-charged, hydrophobic side-chain (tryptophan). Positively-charged residues are thought to be critical for interaction of the PH domain with the membrane (Xiang et al., 2006) and R338W clearly changes the electrostatic potential of the PH domain, as visualized using APBS (**Figures 2A–C**). R338W is also predicted to introduce a number of clashes with surrounding residues (e.g., N335, K363; **Figures 2D,E**), which is also predicted to affect interactions with membrane and the fold of the PH domain.

### Mutation R338W in Collybistin Disrupts Phosphatidylinositol-3-Phosphate Binding

The CB PH domain has previously been shown to play a key role in binding PI3P/PtdIns-3-P, a phosphoinositide with an emerging role in membrane trafficking and signal

domain. Note that R338W is not one of the known PI3P binding residues (R356 and R357, green).

transduction. Deletion of the CB PH domain, or mutation of two key arginine residues (R356/R357) involved in PI3P binding has been demonstrated to abolish CB-mediated gephyrin clustering in functional assays (Harvey et al., 2004; Kalscheuer et al., 2009; Reddy-Alla et al., 2010). In order to determine whether the R338W mutation affected CB binding to PI3P, we performed pulldown assays using PI3P immobilized on agarose beads incubated with lysates of HEK293 cells transfected with either wild-type CB variant CB3SH3<sup>−</sup> or mutant CB3SH3<sup>−</sup> R338W. Total expression of CB3SH3<sup>−</sup> R338W was not significantly different to wild-type CB3SH3<sup>−</sup> when normalized to β-actin expression (**Figure 3A**, left and right panels, wild-type CB2SH3<sup>−</sup> 1.00 ± 0.23 vs. R338W 1.16 ± 0.33; normalized to wild-type ± SEM, n = 5). However, when the PI3P pull-down fraction was expressed as a percentage of raw input, a significant reduction of PI3P binding was observed for the R338W variant (**Figure 3A**, middle and right panels, wild-type 100 ± 13.5 vs. R338W 38 ± 8.6; pull-down fraction ± SEM, n = 5, p < 0.006).

FIGURE 2 | The R338W substitution is predicted to influence the electrostatic potential and folding of the CB PH domain. (A) Arginine 338 is located on the surface of the CB structure within the pleckstrin homology (PH) domain (circled by a black dotted line). It resides within a cleft between turns of β-sheet secondary structure (D), in close proximity to other electropositive side-chains. Therefore, an area of hydrophobicity is created (B) close to the polar heads of the cell membrane. It is predicted that substitution of arginine with tryptophan at position 338 changes the hydrophobicity of this area, making it neutral in charge (C). In addition, the bulky side-chain of tryptophan introduces clashes with surrounding side chains of lysine 363 and asparagine 335 (E), disrupting the overall structure of the protein in this region.

## Mutation R338W Disrupts Collybistin-Mediated Gephyrin Clustering

Given the key role of CB in mediating gephyrin clustering at inhibitory synapses, we also investigated whether the R338W substitution affected the ability of CB to translocate gephyrin to submembrane microaggregates in a cellular clustering assay (Harvey et al., 2004; Kalscheuer et al., 2009). This involved coexpression of myc-tagged human CB (myc-CB3SH3−; Kalscheuer et al., 2009) with EGFP-gephyrin in HEK293 cells. CB variants containing the regulatory SH3 domain (CB3SH3+) typically colocalize with EGFP-gephyrin in large intracellular aggregates (**Figures 3B,D**; Kins et al., 2000; Harvey et al., 2004; Kalscheuer et al., 2009) and require neuroligins, GABAAR α2 or the small Rho-like GTPase TC10 for activation (Poulopoulos et al., 2009; Saiepour et al., 2010; Mayer et al., 2013). However, variants lacking the regulatory SH3 domain (e.g., CB3SH3−) typically result in the formation of EGFP-gephyrin submembrane clusters (**Figure 3C**). However, the CB3SH3<sup>−</sup> R338W variant did not result in the formation of submembrane microaggregates with EGFP-gephyrin, but rather co-localized with EGFP-gephyrin in large intracellular aggregates (**Figure 3E**) similar to the distribution previously observed for CB3SH3<sup>+</sup> or CB mutants that disrupt PI3P binding, such as R290H and the double mutant R356N/R357N (**Figures 3F,G**; Reddy-Alla et al., 2010; Papadopoulos et al., 2015). This demonstrates that the R338W substitution disrupts CB-mediated accumulation of gephyrin in submembrane microclusters.

### DISCUSSION

This study reports the identification and functional characterization of a novel mutation (p.R338W) in ARHGEF9 that is likely to represent the cause of XLID in family K8010. Using next-generation X-exome sequencing, Affected Kindred/Cross-Cohort Analysis and inheritance testing, we found a novel SNV (c.1012C>T; p.R338W) in ARHGEF9 that segregated with the disease phenotype. Using molecular modeling and functional assays for CB PI3P binding and gephyrin clustering, we were able to establish the likely pathomechanism for p.R338W: a local disruption in the PH domain structure, leading to a reduction in PI3P binding and/or PH domain folding, and consequent loss in the ability of CB to mediate gephyrin clustering in an in vitro assay. The identification of ARHGEF9 as the causative gene for family K8010 is consistent with previous studies that have identified CB as a neuronally-expressed RhoGEF with a key role in inhibitory synaptic transmission (Kins et al., 2000; Harvey et al., 2004). At selected inhibitory synapses, CB interacts with gephyrin (Kins et al., 2000; Harvey et al., 2004), a scaffolding protein with dual roles in inhibitory receptor clustering and molybdenum co-factor synthesis (Feng et al., 1998). CB knockout mice show increased anxiety and impaired spatial learning associated with a selective loss of GABAARs in the basolateral amygdala and hippocampus (Papadopoulos et al., 2007). Unsurprisingly, loss of CB clearly leads to significant changes in GABAergic inhibition, network excitability and synaptic plasticity (Jedlicka et al., 2009). Recent studies have also implicated CB in mTOR signaling: CB physically interacts with mTOR and inhibits mTORC1 signaling pathway and protein synthesis (Machado et al., 2015). This suggests that disruption of mTORC1 signaling pathways could also contribute to ID in patients with ARHGEF9 loss-of-function mutations.

A number of mutations in ARHGEF9 have been identified in patients encompassing missense and nonsense mutations, deletions and complex rearrangements (**Table 1**). Curiously, the associated phenotypes vary quite substantially. For example, Harvey et al. (2004) reported a p.G55A missense mutation in ARHGEF9 associated with hyperekplexia, early infantile epileptic encephalopathy and severe psychomotor retardation (p.G55A, SH3 domain). By contrast, Shimojima et al. (2011) identified an ARHGEF9 nonsense mutation (p.Q2X) in an individual with refractory seizures, right frontal polymicrogyria and severe psychomotor retardation. Lemke et al. (2012) also reported a p.R290H missense mutation in the CB RhoGEF domain associated with epilepsy and ID. Furthermore, large de novo deletions affecting ARHGEF9 as well as neighboring genes SPIN4 and LOC92249 have been reported to be associated with complex phenotypes that include features such as partial seizures, delayed psychomotor development and generalized overgrowth (**Table 1**; Lesca et al., 2011; Shimojima et al., 2011). Lastly, a balanced chromosomal translocation (Kalscheuer et al., 2009) and a paracentric inversion (Marco et al., 2008) have been reported with yet more clinical features, including disturbed sleep-wake cycle, increased anxiety and aggressive behavior or hyperarousal, respectively.


The exact reasons behind this clinical variability remains unknown, but are likely to be linked to several factors. Firstly, certain CB mutations (e.g., p.G55A, p.R290H, C-terminal truncations) clearly cause dominant-negative effects on gephyrin and GABA<sup>A</sup> receptor clustering in neuronal systems (Harvey et al., 2004; Kalscheuer et al., 2009; Papadopoulos et al., 2015). Secondly, for female patients with CB mutations, the clinical features observed may depend on the degree of X-inactivation skewing. At least two studies involving a translocation or inversion in ARHGEF9 have indicated skewed X inactivation in favor of the abnormal X chromosome (Marco et al., 2008; Kalscheuer et al., 2009). However, one emerging theme in functional studies of CB missense mutations appears to be loss of PI3P binding. Our own functional analysis suggests that the p.R338W variant causes a local disruption in the PH domain structure, leading to a reduction in PI3P binding and/or PH domain folding, and consequent loss in the ability of CB to mediate gephyrin clustering. Similar findings have recently been reported for the p.R290H mutation linked to epilepsy and ID, which appears to alter the strength of intramolecular interactions between the RhoGEF and PH domains, also leading to a loss of PI3P binding affinity (Papadopoulos et al., 2015). These results highlight the key role of phosphoinositide binding and correct

#### REFERENCES


localization of CB for synaptic function. However, given the variability in clinical phenotypes associated with CB mutations, it is also evident that next-generation sequencing diagnostics have a pivotal role to play in the diagnosis of X-linked disorders.

#### AUTHOR CONTRIBUTIONS

RJH and CES designed the experiments; JPJ contributed DNA samples; RES provided clinical evaluation; PL, SG, MMM, VMJ, PT and KH performed the experiments; CES, SG, PT, KH and RJH analyzed the data; CES and RJH wrote the paper. All authors were involved in revising the paper for important intellectual content, and gave final approval of the version to be published.

### FUNDING

This work was supported by the Medical Research Council (J004049 to RJH and KH), a NINDS grant (R01NS073854 to CES) and in part by the South Carolina Department of Disabilities and Special Needs (SC DDSN). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Dedicated to the memory of Ethan Francis Schwartz (1996–1998).

phenotypic expression among males with a recurrent nonsense mutation - 109C>T (p.R37X). Clin. Genet. 87, 461–466. doi: 10.1111/cge.12420


exploratory research and analysis. J. Comput. Chem. 25, 1605–1612. doi: 10. 1002/jcc.20084


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

Copyright © 2016 Long, May, James, Grannò, Johnson, Tarpey, Stevenson, Harvey, Schwartz and Harvey. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Novel Missense Mutation A789V in IQSEC2 Underlies X-Linked Intellectual Disability in the MRX78 Family

Vera M. Kalscheuer 1,2\*, Victoria M. James <sup>3</sup> , Miranda L. Himelright <sup>4</sup> , Philip Long<sup>3</sup> , Renske Oegema<sup>5</sup> , Corinna Jensen<sup>1</sup> , Melanie Bienek <sup>1</sup> , Hao Hu<sup>1</sup> , Stefan A. Haas <sup>6</sup> , Maya Topf <sup>7</sup> , A. Jeannette M. Hoogeboom<sup>5</sup> , Kirsten Harvey <sup>3</sup> , Randall Walikonis <sup>4</sup> and Robert J. Harvey <sup>3</sup> \*

<sup>1</sup> Department of Human Molecular Genetics, Max Planck Institute for Molecular Genetics, Berlin, Germany, <sup>2</sup> Research Group Development and Disease, Max Planck Institute for Molecular Genetics, Berlin, Germany, <sup>3</sup> Department of Pharmacology, UCL School of Pharmacy, London, UK, <sup>4</sup> Department of Physiology and Neurobiology, University of Connecticut, Storrs, CT, USA, <sup>5</sup> Department of Clinical Genetics, Erasmus MC University Medical Center Rotterdam, Rotterdam, Netherlands, <sup>6</sup> Department of Computational Molecular Biology, Max Planck Institute for Molecular Genetics, Berlin, Germany, <sup>7</sup> Department of Biological Sciences, Institute for Structural and Molecular Biology, Birkbeck College, London, UK

#### Edited by:

Nicola Maggio, The Chaim Sheba Medical Center, Israel

#### Reviewed by:

Hiroyuki Sakagami, Kitasato University School of Medicine, Japan Cheryl Shoubridge, University of Adelaide, Australia

#### \*Correspondence:

Vera M. Kalscheuer kalscheu@molgen.mpg.de; Robert J. Harvey r.j.harvey@ucl.ac.uk

Received: 23 October 2015 Accepted: 14 December 2015 Published: 11 January 2016

#### Citation:

Kalscheuer VM, James VM, Himelright ML, Long P, Oegema R, Jensen C, Bienek M, Hu H, Haas SA, Topf M, Hoogeboom AJM, Harvey K, Walikonis R and Harvey RJ (2016) Novel Missense Mutation A789V in IQSEC2 Underlies X-Linked Intellectual Disability in the MRX78 Family. Front. Mol. Neurosci. 8:85. doi: 10.3389/fnmol.2015.00085 Disease gene discovery in neurodevelopmental disorders, including X-linked intellectual disability (XLID) has recently been accelerated by next-generation DNA sequencing approaches. To date, more than 100 human X chromosome genes involved in neuronal signaling pathways and networks implicated in cognitive function have been identified. Despite these advances, the mutations underlying disease in a large number of XLID families remained unresolved. We report the resolution of MRX78, a large family with six affected males and seven affected females, showing X-linked inheritance. Although a previous linkage study had mapped the locus to the short arm of chromosome X (Xp11.4-p11.23), this region contained too many candidate genes to be analyzed using conventional approaches. However, our X-chromosome exome resequencing, bioinformatics analysis and inheritance testing revealed a missense mutation (c.C2366T, p.A789V) in IQSEC2, encoding a neuronal GDP-GTP exchange factor for Arf family GTPases (ArfGEF) previously implicated in XLID. Molecular modeling of IQSEC2 revealed that the A789V substitution results in the insertion of a larger side-chain into a hydrophobic pocket in the catalytic Sec7 domain of IQSEC2. The A789V change is predicted to result in numerous clashes with adjacent amino acids and disruption of local folding of the Sec7 domain. Consistent with this finding, functional assays revealed that recombinant IQSEC2<sup>A</sup>789<sup>V</sup> was not able to catalyze GDP-GTP exchange on Arf6 as efficiently as wild-type IQSEC2. Taken together, these results strongly suggest that the A789V mutation in IQSEC2 is the underlying cause of XLID in the MRX78 family.

Keywords: ArfGEF, BRAG1, IQ-ArfGEF, IQSEC2, MRX78, XLID

### INTRODUCTION

Intellectual disability (ID) is a developmental brain disorder characterized by impaired intellectual and adaptive functions, and can be defined by an IQ below 70 and limitations in intellectual functioning and adaptive behaviors. As a result of the excess in males affected by ID, and the identification of families where ID shows clear X-linked segregation, significant attention has focused on the genetics of X-linked intellectual disability (XLID)—a common, clinically and genetically complex disorder often arising from mutations in one of >100 genes on the X chromosome. XLID may be associated with other clinical, morphological, or behavioral symptoms (syndromic XLID) or appear without other associated defects (nonsyndromic XLID). However, despite extensive genetic studies using conventional linkage/candidate gene analysis or analysis of copy number variants (CNVs), for a significant number of families, the underlying cause of XLID has remained unclear. Fortunately, the genetics of XLID has recently been accelerated by next-generation DNA sequencing and novel bioinformatics approaches for variant filtering (de Ligt et al., 2012; Rauch et al., 2012; Redin et al., 2014; Hu et al., 2015; Tzschach et al., 2015). However, analysis of large exome data sets from the general population has also revealed a number of ''XLID genes'' where truncating variants or previously published ''mutations'' are observed at a relatively high frequency in normal controls, calling into question whether they are indeed causal. For example, recent studies questioned the implication of AGTR2, MAGT1, ZNF674, SRPX2, ATP6AP2, ARHGEF6, NXF5, ZCCHC12, ZNF41, ZNF81 and RAB40AL in XLID (Piton et al., 2013; Ołdak et al., 2014). This highlights the vital importance of structure-function analyses for validating potentially diseasecausing variants.

In this study, we have combined next-generation sequencing, variant filtering and structure-function assays to resolve the cause of XLID in a large family known as MRX78 (de Vries et al., 2002). A previous investigation of the MRX78 family revealed linkage to the short arm of chromosome X (Xp11.4-p11.23; de Vries et al., 2002). We present compelling evidence that the likely cause of XLID in the extended MRX78 pedigree with six affected males and seven affected females in four generations is a missense mutation in the known XLID gene IQSEC2 (MIM<sup>∗</sup> 300522; Shoubridge et al., 2010) encoding a neuronal ArfGEF (known as IQSEC2, BRAG1 and IQ-ArfGEF) involved in cytoskeletal organization, dendritic spine morphology and excitatory synaptic organization, along with a review of previously published IQSEC2-related XLID patients.

#### MATERIALS AND METHODS

#### Genetic Analysis

Written informed consent was obtained from the legal guardians of patients III-5 and III-8 regarding next-generation sequencing. This study was declared exempt from approval by the medical ethics committee of the Erasmus MC University Medical Center (decision MEC-2012-387). Confirmation of the mutation in the additional family members was performed on DNA previously obtained and used for diagnostic purposes and was carried out according to Erasmus MC University Medical Center regulations for secondary use of tissue after diagnostic procedures. In addition, patients or legal guardians were informed and gave verbal consent to use remaining samples for scientific research. X-chromosome exome resequencing and bioinformatics analysis was performed as recently described (Hu et al., 2014, 2015). DNA from the affected male individual III-8 was used for constructing the sequencing library using the Illumina Genomic DNA Single End Sample Prep kit (Illumina, San Diego, CA, USA). Enrichment of the Xchromosomal exome was then performed using the Agilent SureSelect Human X Chromosome Kit (Agilent, Santa Clara, CA, USA), which contains 47,657 RNA baits for 7591 exons of 745 genes of the human X chromosome. Single-end deep sequencing was performed on the Illumina Genome Analyzer GAIIx (Illumina, San Diego, CA, USA). Reads were subsequently mapped to the human reference genome (hg18 without random fragments) with RazerS (Weese et al., 2009) tolerating a sequence difference up to 5 bp per read. We applied the split mapping tool SplazerS (version1.0; Emde et al., 2012) in order to detect short insertions (<30 bp) and larger deletions (<50 kb) from unmapped and indelcontaining reads. In addition, large insertions/deletions were predicted using ExomeCopy (Love et al., 2011) by analysing changes in depth of coverage along the targeted regions. Singlenucleotide polymorphisms (SNPs) and short indels (<5 bp) were called with snpStore. In parallel, we used the Medical Resequencing Analysis Pipeline (MERAP; Hu et al., 2014). Here, the mapping was performed using SOAP2 allowing at most two mismatches, and requiring at least four reads for single-nucleotide variant (SNV) and indel detection. For both approaches all sequence variants were screened against public databases [dbSNP138, 1000 Genomes project, Exome Variant Server (ESP6500)], and the in-house database of the Max Planck Institute, Berlin for annotating likely non-pathogenic and previously reported neutral variants. In addition, the OMIM catalog and the Human Gene Mutation Database (HGMD) were used as a filter to identify all previously described mutations. PCR primers for mutation confirmation and segregation analysis were IQSEC2-D221F 5<sup>0</sup> -cctcttgctgtccttttcca-3<sup>0</sup> and IQSEC2-D221R 5 0 -tgggcccaaaattagttcaa-3<sup>0</sup> .

#### Building a Homology Model of Human IQSEC2

Structural templates for homology modeling of IQSEC2 were determined using HHPred<sup>1</sup> (Söding et al., 2005). Homology modeling of the Sec7 and PH domains of IQSEC2 relied on three template crystal structures. The human BIG1 SEC7 domain (PDB ID: 3LTL, sequence identity 42%) was used to model the SEC7 domain, human GEP100 IQ motif (PDB ID: 3QWM, sequence identity 70%) was used to model the PH domain and finally, the mouse Grp1 Arf GTPase exchange factor (PDB ID: 2R0D; DiNitto et al., 2007) was used to model parts of both the Sec7 and PH domains and to guide the orientation of the domains

<sup>1</sup>http://toolkit.tuebingen.mpg.de/hhpred

relative to each other (overall sequence identity of 38%). A model of the 24 amino acid linker between the SEC7 and PH domains (residues 942–966) was calculated using I-TASSER<sup>2</sup> (Zhang, 2008; Roy et al., 2010, 2012), which predicts proteins structure from sequence using multiple threading alignments via LOMETS (Wu and Zhang, 2007) and iterative assembly simulations based on structural templates or by ab initio modeling. For this linker, I-TASSER built a model with a straight α-helix, based on structural coordinates from various homologous structures in PDB (PDB IDs: 3CC2G, 2KEGA, 1K73I, 1S72G, 1M90I, 1QVFG) with sequence similarities of 21–28% [confidence score (C-score) of −0.9]. This was consistent with secondary structure prediction methods; PSIPRED<sup>3</sup> (Jones, 1999) and RaptorX<sup>4</sup> (Källberg et al., 2012), which predicted residues 948–960 to form a helix with a high confidence (>80%). Single pairwise, sequence-structure alignments were calculated using HHPred, then the alignments were combined manually to generate one multiple alignment of the entire IQSEC2 sequence with the four template structure sequences (three PDB structures and one model of the linker). Based on this alignment, 30 models were generated using MODELLER-9.10 and assessed with the Discrete Optimized Protein Energy (DOPE) statistical potential score (Shen and Sali, 2006). The model with the lowest DOPE score was selected as representative. The p.A789V mutation was modeled with the swapaa command in Chimera (Pettersen et al., 2004) using the Dunbrack backbone-dependent rotamer library (Dunbrack, 2002) and taking into account the lowest clash score, highest number of H-bonds and highest rotamer probability.

#### Site-Directed Mutagenesis and Expression Constructs

The missense mutation c.C2366T, p.A789V was introduced into full-length human pCAGGS-IQSEC2 using the QuikChange sitedirected mutagenesis kit (Agilent) and the primers hIQSEC2- A789V1 5<sup>0</sup> -accggtgggagtggttcacttcatcctgg-3<sup>0</sup> and hIQSEC2-A789 V2 5<sup>0</sup> -ccaggatgaagtgaaccactcccaccggt-3<sup>0</sup> . Expression constructs were verified by Sanger DNA sequencing of the entire coding region.

### Arf Activation Assay

Golgi-localized, ear-containing ARF-binding protein 3 (GGA3) pull-down of activated ARF GTPases was performed as previously described in Shoubridge et al. (2010). Plasmids encoding HA-ARF6 in pXS and FLAG-tagged wild-type IQSEC2, IQSEC2A789V or IQSEC2E849K in pCAGGS were co-transfected into HEK293 cells with Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA). Cells were harvested in lysis buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 2 mM MgCl2, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton X-100, 10% glycerol, and HALT protease inhibitor cocktail (Pierce, Rockford, IL, USA), and lysates were incubated with GST:GGA3 coupled to glutathione beads. Beads were washed in lysis buffer without protease inhibitors and boiled in SDS-PAGE buffer. Samples were run on SDS-PAGE gels and transferred to PVDF membranes. The membranes were probed with primary antibodies against HA (Covance, Princeton, NJ, USA) or FLAG (Sigma, St. Louis, MO, USA) and IRDye secondary antibodies and visualized and quantified with the use of a LiCor Odyssey Infrared Imaging System. The fluorescence intensity of each of the bands was quantified with Image Studio Lite. ARF6-GTP bands were normalized to total ARF6 expression for each of the conditions, and then the ratio of normalized ARF6-GTP to normalized ARF6 was calculated for wild-type IQSEC2, IQSEC2A789V and IQSEC2E849K. The GGA pulldown assay was assessed by one-way ANOVA followed by Tukey's Multiple Comparison Test for post hoc analysis.

### RESULTS

### Clinical Re-assessment of Family MRX78

The revised pedigree of the MRX78 family is depicted in **Figure 1A** and a summary with clinical findings is presented in **Table 1**. Individual I-2 had mild ID and was illiterate. She had a stroke at age 71 and developed dementia thereafter. She passed away at age 80. She gave birth to 15 children; two were stillborn. Two sons and a daughter died in childhood, they were said to be ''handicapped'', one son had spina bifida. All children were placed in foster homes. She had two brothers with normal cognition and three sisters. II-2 had mild ID with difficulty reading and writing. II-6 has ID. II-8 has learning difficulties and is unable to read or write. Her chromosome studies were normal and urine analysis was negative for MAO-A deficiency. Her husband (II-7) also had learning difficulties/ID. II-9 has severe ID, intractable epilepsy and does not speak, or interact socially. He does like physical contact (e.g., holding hands) and responds to his name. He developed walking difficulties in his fifties, with unstable gait (''as a drunk person'') and is now wheelchair bound. There is no tremor and his vision and hearing are good. III-1 and III-2: These brothers both have severe ID. They do not seem to recognize a familiar person and have no speech. They can be physically aggressive towards caretakers, and both are on levomepromazine therapy. Vision and hearing are normal. Urine metabolic investigations were normal, and there was no MAO-A deficiency. Subject III-1 has had one seizure as a teenager. III-3 has moderate ID and no seizures. His height is 176.5 cm, occipital frontal circumference (OFC) 55 cm, and he has no dysmorphic features. Urine metabolic investigations were normal, and MAO-A deficiency was excluded. III-4 had learning difficulties. III-5 was born at term after an uncomplicated pregnancy with birth weight 4400 grams. Developmental delay was evident from early age. At age 12 extensive investigations were performed. At this time he had a moderate to severe ID and severe behavioral problems, refractory to treatment, including Haldol and valproic acid. An EEG showed a diffuse encephalopathy, without epileptic discharges, a brain CT-scan was normal (data not shown). IQ testing age 16 years showed verbal IQ (VIQ) compatible to age 4.7 years, and performance IQ (PIQ) compatible to 4.1 years. He showed minor regression when he was retested at 28 years with the Wechsler Intelligence

<sup>2</sup>http://zhanglab.ccmb.med.umich.edu/I-TASSER

<sup>3</sup>http://bioinf.cs.ucl.ac.uk/psipred/

<sup>4</sup>http://raptorx.uchicago.edu/

splice site. (D) Sequence alignments of IQSEC1, IQSEC2 and IQSEC3 showing the location of missense mutations in the Sec7 domain. Note that IQSEC2 Sec7 domain mutations do not affect predicted GTPase binding residues (bold green type below alignments) as defined in the structure of the related Sec7 domain in the ArfGEF ARNO. Bold type indicates amino acids that are identical in all four sequences.

Scale (WISC) when his cognitive abilities were compatible to a 3.5 year-old. His social-emotional functioning was lower, at the level of a 6–18-month-old. Currently, at age 43 years, he is physically healthy, height is 181 cm and weight 90 kg. Neurological examination showed no abnormalities. He has good hearing and vision, with only mild hypermetropia (1.25 dpt). He is independent in basic daily activities such as dressing and eating, but needs incentive. He is on risperidone treatment (0.5 mg twice daily) and laxantia. He communicates mostly through pictograms. He is quiet and introverted, with occasional verbal aggression, but not physical. He likes puzzles. He is easily distracted. He is suspected of autism spectrum disorder but has not been formally tested. Laboratory investigations in the past, including metabolic screen, Fragile X and chromosome studies



<sup>∗</sup>Cause unknown.

were normal. III-6: During pregnancy her mother had several hospital admissions due to vaginal bleeding and premature contractions. Chromosome studies in chorion villi biopsy were normal (46, XX). She was born at term, with cleft lip/palate, her birth weight was 2785 grams and length 50 cm. She has mild learning difficulties, her OFC is 54 cm. She has three daughters, in whom no molecular testing was performed. She has had two first trimester miscarriages. The eldest daughter (IV:1) was born at 38 weeks of gestation, BW 2900 gram. She has learning difficulties and receives special education, her OFC is 57 cm. She is overweight. Her second daughter (IV:2) was born with ventricular septal defect, her BW was 2700 grams after 38 weeks of gestation. She has normal development, and no learning difficulties. Chromosome studies in chorion villi biopsy in the middle and youngest daughter were normal (46, XX). The youngest is still too young to assess her development. III-8: the pregnancy was complicated with episodes of vaginal bleeding and premature contractions. Developmental delay became evident in the first year of life. He had psychiatric consultation at age 11, due to problematic and aggressive behavior. Two years later he was diagnosed with a pervasive developmental disorder. Currently, at age 36, he has severe ID and autism spectrum disorder. He has a fixation on mirrors and sunshades, and will go around and close them. He has shown severe aggressive outbursts with destructive behavior towards furniture and other objects. He receives risperidone 2.5 mg daily. WISC testing age 20 years showed VIQ compatible to age 4.7 years, and PIQ compatible to 4.1 years. He was last tested at age 34 (Vineland Adaptive Behavior Scale) and he had shown mild regression over a 3-year period. His adaptive skills are now at a level for communication of a 1 year 7 month old, for daily activities at a 3-year-old level, for motor skills at a 1 year 5 month old level, and socialization skills compatible to a 10 month old. He is physically in good health. He underwent adenotomy, strabismus surgery, and excision of a follicular jaw cyst. His current height is 177 cm, and weight 82 kg. Neurological examination showed only brisk reflexes of the lower extremities. He has normal hearing and myopia, with normal media and fundi. Fragile X testing was negative.

#### Identification of a p.A789V Mutation in IQSEC2 in Family MRX78

X-chromosome exome sequencing followed by bioinformatics analysis and filtering against publicly available datasets revealed three novel missense changes: IQSEC2, chrX:53277996G>A, p.A789V, consensus score 4.06, predicted as probably damaging (PolyPhen-2) and damaging (SIFT) with a CADD score of 19; TXLNG, chrX:16859571G>C, p.E423D with a low conservation score (0.02) and predicted as benign (Polyphen-2) and tolerated (SIFT) and CADD of 16; WAS, chrX:48542814T>G, donor, conservation score 3.69, and CADD of 12. This strongly suggested that the novel missense mutation identified in IQSEC2 could be responsible for XLID in this family. Subsequent segregation analysis using Sanger DNA sequencing indicated that the IQSEC2 chrX:53277996G>A variant co-segregated with the XLID phenotype in all individuals tested (**Figures 1A,B**).

#### IQSEC2 p.A789V Mutation is Predicted to Disrupt SEC7 Domain Folding

IQSEC2 has a multi-domain structure consisting of a regulatory IQ-like motif, a catalytic Sec7 domain, a pleckstrin homology (PH) domain and a PDZ binding motif (**Figure 1C**). Residue A789 is located in the catalytic Sec7 domain of IQSEC2 and is highly conserved within IQSEC1–3 (**Figure 1D**), whereas seven out of nine rare missense variants identified in this domain in normal controls (ExAC database<sup>5</sup> ) are mostly present in single individual, and are not conserved residues between IQSEC1−IQSEC3. It is also notable that 3 of 4 previously reported IQSEC2 missense mutations identified in XLID families are located in the Sec7 domain (**Figure 1C**; Shoubridge et al., 2010), including the previously investigated R863W mutation

<sup>5</sup>http://exac.broadinstitute.org/gene/ENSG00000124313

present in the large family MRX1. This R863 residue is the only highly conserved amino acid that is substituted to R863Q in a single normal female. To visualize the potential structural consequences of the p.A789V change, we built a comparative model of the IQSEC2 Sec7 and PH domains using the structures of three sequence-related ArfGEFs: BIG1, GEP100 and Grp1 (PDB ID: 3LTL, 3QWM). The resulting model revealed that A789 is located on the third short helix of nine in the Sec7 domain, although it is not one of the residues predicted to interact with Arf GTPases, GDP or Zn2<sup>+</sup> binding (**Figures 1D**, **2A**). The third and fourth helices come together closely and appear to form a hydrophobic pocket. The residues on these two helices are highly conserved and form a very similar structure in all Sec7-containing crystal structures in PDB, suggesting that the interaction between these helices may be of importance for proper folding of the Sec7 domain. Substitution of alanine at residue 789 with valine introduces a larger side-chain into this packed hydrophobic pocket between helices, which is predicted to cause numerous clashes with surrounding side-chains and/or backbones of residues V818, C821 and V822 (**Figure 2B**).

#### Mutation p.A789V in IQSEC2 Disrupts ArfGEF Activity

We also tested the GEF activity of full-length IQSEC2 and the IQSEC2A789V mutant in a cellular model using a pull-down assay using the adaptor protein Golgi-localized, ear-containing ARFbinding protein 3 (GGA3; **Figure 3A**). GGAs specifically interact with active, GTP-bound ARF but do not interact with inactive ARF GTPases. As a control, we used the IQSEC2E849K dominantnegative mutation, which reduces the exchange activity of the IQSEC2 Sec7 domain by several orders of magnitude (Shoubridge et al., 2010). Co-transfection of ARF6 with wild-type IQSEC2 resulted in a ∼5-fold increase (5.5 ± 0.91, mean ± SEM, n = 3) in GTP-bound ARF6, but only a 2.5- to 1-fold increase when expressed with IQSEC2A789V (2.5 ± 0.19) or IQSEC2E849K (1.2 ± 0.19) respectively (**Figures 3B,C**). This is consistent with a significant loss of ArfGEF activity in the IQSEC2A789V mutant. It is also noteworthy that the IQSEC2A789V mutation does not appear to affect protein stability (**Figure 3B**), at least in cellular models.

#### DISCUSSION

This study reports the identification and functional characterization of a novel mutation (c.C2366T, p.A789V) in IQSEC2 that based on variant filtering and segregation analysis likely presents the cause of XLID in the large family MRX78. A previous analysis in this family revealed X-linked inheritance with linkage to a 15.16 cM region on Xp11 (de Vries et al., 2002). Using molecular modeling of wild-type and mutant protein and assays for ArfGEF activity, we were

FIGURE 2 | IQSEC2 A789V mutation is predicted to disrupt SEC7 domain folding. Side views of the molecular models of IQSEC2 Sec7 and PH domains, with the Sec7 domain in red, the PH domain in blue and the linker between them in gold. Boxed areas show the areas expanded in the inserts to the right. (A) Normal IQSEC2 structure, and (B) mutant IQSEC2 harboring the p.A789V mutation. Note that the larger valine side-chain in the packed hydrophobic pocket between helices is predicted to cause numerous clashes with surrounding side-chains and/or backbones of residues V818, C821 and V822, indicating a potential disruption of the local fold in this region.

(SEM) from three independent experiments. ∗∗p ≤ 0.01; ∗∗∗p ≤ 0.001.

able to establish the likely pathomechanism for p.A789V: a disrupted hydrophobic pocket in the Sec7 domain structure, leading to loss of IQSEC2 ArfGEF activity. The identification of IQSEC2 as the causative gene for MRX78 is consistent with previous studies that have identified IQSEC2 as a neuronally-expressed ArfGEF with a key role in excitatory synaptic transmission. At excitatory synapses, IQSEC2 interacts with multivalent postsynaptic density (PSD) proteins such as IRSp53 (Sanda et al., 2009) and PSD-95 (Sakagami et al., 2008), forming a complex with N-methyl-D-aspartate (NMDA) receptors.

A recent reassessment of the MRX78 family by one of the co-authors (RO) revealed that except for one all affected males presented with moderate to severe ID, several affected males did not acquire speech skills and presented with behavioral disturbances. Females from this family who carry the mutation on one of her X-chromosomes presented with learning difficulties or mild ID. A summary with genetic findings and clinical presentations of other affected individuals who carry missense, nonsense, deletions, insertions and complex rearrangements is given in (**Table 2**). Shoubridge et al. (2010) initially reported four missense mutations in IQSEC2 associated with non-syndromic XLID (IQ domain p.R359C, Sec7 domain p.R758Q, p.Q801P and p.R863W). However, seizures, autistic traits, psychiatric problems and delayed early language skills were noted in different male individuals in this study. Furthermore, several de novo deleterious IQSEC2 mutations have been reported in males and females who presented with seizure disorders, including Rett-like syndrome and Lennox-Gastaut syndrome: a frameshift mutation (p.N91Kfs<sup>∗</sup> 112) and a truncating chromosome breakpoint in intron 1 [46, X(tX;20)(p11.2;q11.2)] in the N-terminal part of IQSEC2 (Morleo et al., 2008; Olson et al., 2015), a two base pair deletion resulting in p.C684X (region between IQ-like motif and Sec7 domain) (Gandomi et al., 2014), two additional mutations affecting the Sec7 domain (p.R855X and p.S861T that is predicted to abolish the native splice donor site; Rauch et al., 2012; Gandomi et al., 2014; Tran Mau-Them et al., 2014), and one de novo mutation (p.Q1108X) C-terminal to the PH domain (Allen et al., 2013). Additionally, Tran Mau-Them et al. (2014) reported two severely affected males with Rett-like features who carried de novo intragenic duplications, both of which were thought to disrupt IQSEC2. Furthermore, a p.I888Afs followed by a subsequent premature stop codon has been identified in a male with severe non-syndromic ID (Tzschach et al., 2015). IQSEC2 mutations in ID patients which affect the PH domain of the protein have also been reported, including a p.L992X change in a patient with schizophrenia (Purcell et al., 2014), a de novo p.Q1033X mutation in a male with severe ID, epilepsy, strabismus and autistic features (Redin et al., 2014), and a de novo mutation (p.R1055X) in a female with severe ID, epilepsy and borderline macrocephaly (Tzschach et al., 2015). A partial de novo duplication of TENM3 inserted into IQSEC2 and subsequent formation of an in-frame IQSEC2-TENM3 fusion gene was also recently reported. The resulting disruption of IQSEC2 was thought to contribute to ID, epilepsy, progressive spasticity, and microcephaly in that patient (Gilissen et al., 2014). Furthermore, three maternally inherited duplications with disrupted IQSEC2 in five males who presented with ID and behavioral disturbances were reported

#### TABLE 2 | IQSEC2 mutations and associated phenotypes.




#### Gross deletions and insertions without HUWE1 gene involvement


#### TABLE 2 | (Continued).



† seizures, autistic traits, psychiatric problems and delayed early language skills were noted in different individuals in this study, although none of these additional phenotypes were consistent in all affected male individuals in the families studied.

by Moey et al. (2015), and a de novo deletion of IQSEC2 and the XLID gene KDM5C (MIM<sup>∗</sup> 314690; Jensen et al., 2005) has been identified in a girl with severe ID and autistic behavior (Fieremans et al., 2015). Although originally described as nonsyndromic XLID, newly acquired clinical data from the MRX78 family suggests that additional features might be associated with the IQSEC2 p.A789V mutation, including variable seizures in males, which is consistent with other reports, and behavioral disturbances in five out of six affected males. By contrast, heterozygous female have learning disabilities. As IQSEC2 is one of the few genes that escape X-inactivation in females with an expression level similar in males and females (Moey et al., 2015) we hypothesize that, in these female carriers, the missense mutation is sufficient to produce symptoms but that the mutant protein has residual function and in addition that there is some compensation from the normal allele. This is in contrast to early truncating mutations in IQSEC2, which in females are associated with a more severe phenotype including infantile spasms, epilepsy, autistic features, Lennox Gastaut syndrome and Rett-like syndrome. Given that in several severely affected females IQSEC2 is either deleted on one of the Xchromosomes or the truncating change is located early in the N-terminus and therefore can be assumed to result in protein degradation due to nonsense-mediated mRNA decay, it is unlikely that in these cases the mutations produce a dominant negative effect. Thus, it is very likely that IQSEC2 is a dosage-sensitive gene that needs to be tightly regulated for

#### REFERENCES


normal cognitive function in males and females and that IQSEC2 mutations can cause a spectrum of clinical features in both sexes.

#### AUTHOR CONTRIBUTIONS

VMK, RW and RJH designed the experiments; AJMH and RO contributed DNA samples and clinical evaluation; VMJ, MLH, PL, CJ, MB, MT and KH performed the experiments; VMK, HH, SAH, MT, KH, RW, and RJH analyzed the data; VMK, RW and RJH wrote the paper. All authors were involved in revising the paper for important intellectual content, and gave final approval of the version to be published.

#### FUNDING

This work was supported by the Medical Research Council (J004049 to RJH and KH) and the EU FP7 project GENCODYS, grant number 241995 (to HH and VMK) and the Whitehall Foundation (grant number 2010-08- 68 to RSW). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The authors would like to thank the Exome Aggregation Consortium and the groups that provided exome variant data for comparison. A full list of contributing groups can be found at http://exac.broadinstitute.org/about.


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

Copyright © 2016 Kalscheuer, James, Himelright, Long, Oegema, Jensen, Bienek, Hu, Haas, Topf, Hoogeboom, Harvey, Walikonis and Harvey. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Rare Variants in Neurodegeneration Associated Genes Revealed by Targeted Panel Sequencing in a German ALS Cohort

Stefanie Krüger<sup>1</sup> , Florian Battke<sup>1</sup> , Andrea Sprecher<sup>1</sup> , Marita Munz1,2, Matthis Synofzik2,3 , Ludger Schöls2,3, Thomas Gasser2,3, Torsten Grehl<sup>4</sup> , Johannes Prudlo5,6† and Saskia Biskup1,2 \* †

<sup>1</sup> CeGaT GmbH, Center for Genomics and Transcriptomics, Tübingen, Germany, <sup>2</sup> Department of Neurodegenerative Diseases, Hertie-Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany, <sup>3</sup> German Research Center for Neurodegenerative Diseases, Tübingen, Germany, <sup>4</sup> Department of Neurology, BG-Kliniken Bergmannsheil GmbH, Ruhr-University Bochum, Bochum, Germany, <sup>5</sup> Department of Neurology, University of Rostock, Rostock, Germany, <sup>6</sup> German Research Center for Neurodegenerative Diseases, Rostock, Germany

#### Edited by:

Robert J. Harvey, UCL School of Pharmacy, UK

#### Reviewed by:

Margaret M. DeAngelis, University of Utah, USA Stephane Pelletier, St. Jude Children's Research Hospital, USA

\*Correspondence: Saskia Biskup saskia.biskup@cegat.de †These authors have shared last author.

Received: 16 May 2016 Accepted: 20 September 2016 Published: 13 October 2016

#### Citation:

Krüger S, Battke F, Sprecher A, Munz M, Synofzik M, Schöls L, Gasser T, Grehl T, Prudlo J and Biskup S (2016) Rare Variants in Neurodegeneration Associated Genes Revealed by Targeted Panel Sequencing in a German ALS Cohort. Front. Mol. Neurosci. 9:92. doi: 10.3389/fnmol.2016.00092 Amyotrophic lateral sclerosis (ALS) is a progressive fatal multisystemic neurodegenerative disorder caused by preferential degeneration of upper and lower motor neurons. To further delineate the genetic architecture of the disease, we used comprehensive panel sequencing in a cohort of 80 German ALS patients. The panel covered 39 confirmed ALS genes and candidate genes, as well as 238 genes associated with other entities of the neurodegenerative disease spectrum. In addition, we performed repeat length analysis for C9orf72. Our aim was to (1) identify potentially disease-causing variants, to (2) assess a proposed model of polygenic inheritance in ALS and to (3) connect ALS with other neurodegenerative entities. We identified 79 rare potentially pathogenic variants in 27 ALS associated genes in familial and sporadic cases. Five patients had pathogenic C9orf72 repeat expansions, a further four patients harbored intermediate length repeat expansions. Our findings demonstrate that a genetic background of the disease can actually be found in a large proportion of seemingly sporadic cases and that it is not limited to putative most frequently affected genes such as C9orf72 or SOD1. Assessing the polygenic nature of ALS, we identified 15 patients carrying at least two rare potentially pathogenic variants in ALS associated genes including pathogenic or intermediate C9orf72 repeat expansions. Multiple variants might influence severity or duration of disease or could account for intrafamilial phenotypic variability or reduced penetrance. However, we could not observe a correlation with age of onset in this study. We further detected potentially pathogenic variants in other neurodegeneration associated genes in 12 patients, supporting the hypothesis of common pathways in neurodegenerative diseases and linking ALS to other entities of the neurodegenerative spectrum. Most interestingly we found variants in GBE1 and SPG7 which might represent differential diagnoses. Based

on our findings, we recommend two-staged genetic testing for ALS in Germany in patients with familial and sporadic ALS, comprising C9orf72 repeat analysis followed by comprehensive panel sequencing including differential diagnoses that impair motor neuron function to meet the complexity of ALS genetics.

Keywords: amyotrophic lateral sclerosis, neurodegeneration, next generation sequencing, genetic heterogeneity, polygenic inheritance

#### INTRODUCTION

fnmol-09-00092 October 7, 2016 Time: 15:23 # 2

Amyotrophic lateral sclerosis (ALS) is a devastating multisystemic neurodegenerative disorder characterized by degeneration of upper and lower motor neurons in the motor cortex, brain stem, and spinal cord (Peters et al., 2015). ALS can be inherited in an autosomal dominant, autosomal recessive or X-linked manner. About 10% of cases are considered as being familial (fALS), whereas the remaining 90% seem to occur sporadically (sALS) with no family history of ALS. Since the first discovery of SOD1 mutations being causative for ALS1 in 1993 (Rosen et al., 1993), researchers all over the world have made great effort to further delineate the genetic basis underlying ALS. Today, more than 30 confirmed major disease genes are listed by the Amyotrophic Lateral Sclerosis Online genetics Database (ALSoD<sup>1</sup> ), the most frequently affected being C9orf72 (40% fALS, 5–6% sALS; pathogenic repeat expansion in the non-coding region between exons 1a and 1b, detection by repeat analysis), SOD1 (20% fALS, 3% sALS), FUS (5% fALS, <1% sALS) and TARDBP (3% fALS, 2% sALS) (Abel et al., 2012; Su et al., 2014).

Screening of the known ALS genes identifies pathogenic mutations in more than 60% of fALS cases. However, the same genes that can be affected in fALS can also be found mutated in sporadic cases, e.g., due to incomplete penetrance, false paternity, recessive inheritance or de novo mutations (Su et al., 2014). Mutations in disease genes affect different molecular pathways which promote motor neuron degeneration and include protein misfolding and subsequent aggregation, mitochondrial dysfunction and oxidative stress, impaired RNA processing, glutamate excitotoxicity and impaired axonal transport (Redler and Dokholyan, 2012; Shaw, 2005). These findings provided fundamental insight into basic underlying pathomechanisms and additionally linked ALS to other disease entities like frontotemporal dementia (FTD) or hereditary spastic paraplegia (HSP).

With the application of genome-wide association studies (GWASs) and high throughput sequencing technologies (next generation sequencing, NGS), a large number of additional disease genes, disease modifiers, and risk factors have been identified especially in sALS. GWASs suggest that genetic factors might contribute to a minimum of 23% of disease risk, whereupon such factors do not necessarily have to be directly causative but instead may act as risk factors or disease modifiers (e.g., age of onset, disease progression) in the interplay with environmental and stochastic factors (Renton et al., 2014; Marangi and Traynor, 2015). Numerous GWASs have been published which showed associations of various loci with ALS containing potential risk genes such as FGGY, ITPR2 and UNC13A (Marangi and Traynor, 2015) but until now, causative variants in most of these genes have not been identified. As GWASs are based on the "common disease – common variant" hypothesis and odds ratios associated with risk alleles are usually low, they are solely suitable for the identification of common disease modifiers with low effect size in complex disorders rather than rare causative variants with large effect sizes. By contrast, NGS represents a powerful, groundbreaking approach to detect rare variants with moderate or high penetrance in Mendelian diseases without having access to large pedigrees (He et al., 2014). ALS and other neurodegenerative diseases which are characterized by great genetic heterogeneity and sometimes overlapping symptoms or even atypical phenotypes benefit to a great extent from NGS and the possibility to analyze all genes implicated in the disease in one approach. During the last years, the use of NGS encompassed and considerably increased the number of identified disease genes and risk factors for ALS, generating further insight into underlying pathomechanisms at the same time. One example is the recent discovery of the mitochondrial protein CHCHD10 as being implicated in ALS which for the first time proves a direct impact of mitochondria in the pathogenesis of the disease, a result obtained by exome sequencing in several families affected by ALS (e.g., Bannwarth et al., 2014; Müller et al., 2014; Kurzwelly et al., 2015). As sequencing costs and turnaround times substantially decreased during the last years, the broad application of NGS has triggered a fundamental shift not only in clinical genetics but also in research on rare heritable diseases. Additionally, by the analysis of large numbers of genes in parallel, it has become evident that some patients carry potentially pathogenic variants in genes that are associated with other entities of the neurodegenerative spectrum. Besides this, one emerging theme in ALS genetics is the presumption that ALS might be a complex disease. This view arises mainly from the observation of reduced penetrance in pedigrees affected by fALS and the partially missing heritability in sporadic cases (van Blitterswijk et al., 2012; He et al., 2014). In recent studies, the authors applied NGS to identify patients who carried pathogenic or potentially pathogenic variants in more than one disease gene with frequencies ranging from 1.6% to 31.7% in fALS and sALS cohorts (van Blitterswijk et al., 2012; Kenna et al., 2013; Cady et al., 2015). However, these studies additionally point out that the genetic basis underlying ALS in cohorts of different European countries and the US differs due to founder effects and thus should not be assumed to be homogeneous.

<sup>1</sup>http://alsod.iop.kcl.ac.uk/home.aspx

Here we hypothesize that ALS is caused by polygenic contributions from many disease-causing or disease-modifying gene variants which encompass not only known ALS genes but also other genes from the neurodegenerative disease spectrum. To investigate this hypothesis, we used a highcoverage targeted high-throughput sequencing approach to detect variants in 39 ALS associated genes as well as 238 additional genes that are linked to other neurodegenerative diseases in a German cohort of 80 clinically well characterized ALS patients. We aim at identifying known causative mutations and novel variants, to report on patients who carry multiple potentially disease causing variants or variants in genes which are implicated not only in ALS, but also in other neurodegenerative disorders. To our knowledge, this is the first report on extensive genetic screening in a German ALS cohort including not only confirmed ALS genes but also possible candidate genes, modifiers and risk factors to assess the great genetic heterogeneity of ALS in Germany.

#### MATERIALS AND METHODS

#### Study Participants

Our cohort includes 80 unrelated clinically diagnosed ALS patients (55% male, 45% female; 7.5% familial, 92.5% sporadic; 82.5% ALS, 6.25% ALS-FTD, 2.5% flail leg, 2.5% flail arm, 6.25% primary lateral sclerosis (PLS)). Mean age of disease onset was 60.1 years (range 29–88 years). Patients were recruited consecutively in ALS outpatient clinics at the university hospitals Rostock and Bochum (Germany). Relationship was excluded by evaluation of family history. Only one affected individual per family was included in this study and there was no evidence of relationship between any study participants. Informed written consent was obtained from all participants. The study was approved by the local medical ethics committee of Rostock University (A2009-10 and A2011- 56) and conducted in accordance with the Declaration of Helsinki.

#### DNA Extraction

Genomic DNA was extracted from EDTA blood using the QIAamp DNA Blood Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol.

#### C9orf72 Repeat Analysis

All subjects were screened for a pathological repeat expansion in the C9orf72 gene (GenBank NM\_018325.3, NM\_145005.5) using fragment length analysis of fluorescence labeled PCR products as repeat expansions cannot be detected by NGS (method according to DeJesus-Hernandez et al., 2011). Based on a repeat primed PCR we determined the size of GGGGCC repeats (method according to Renton et al., 2011). Repeat lengths of ≥ 30 units were considered as being pathogenic, whereas repeat lengths of 20 to 29 units are considered as intermediate.

#### Targeted Resequencing

Genomic DNA was enriched using a custom design Agilent SureSelect in solution kit. The design of our diagnostic panel for neurodegenerative diseases (277 genes in total) included 14 genes which were classified as disease genes when this study was initiated, 25 putative candidate genes, modifiers, and risk factors identified by literature research as being most presumably implicated in ALS (e.g., by GWAS, experimental evidence, or connected pathways; **Table 1**), as well as 238 genes associated with other neurodegenerative diseases (for example genes associated with FTD, HSP and others; see Supplementary Data, 763 kb in total). Sequencing was performed using barcoded libraries on the SOLiD 5500xl platform according to the manufacturer's instructions (Fragment Library Preparation 5500 Series SOLiDTM Systems, User Guide, Applied Biosystems by Life Technology). Approximately 2.3 million on target reads were generated per sample and the mean coverage on target was 184.2 sequencing reads with a mean mapping quality of 85.3. On average 89.4% of bases were covered by ≥10 reads/base per sample. The primary data analysis was performed using Lifescope (versions v2.5-r0 and v2.5 r2.5.1).

#### Variant Filtering

Only variants (SNVs/small indels) with a minor allele frequency (MAF) of ≤1% in coding and flanking intronic regions (±8 base pairs) and the UTR regions were evaluated. Known disease causing mutations which are listed in the HGMD database were evaluated in coding and flanking intronic regions up to ±30 base pairs and up to a MAF of ≤5%. Population frequencies are adapted from the following databases: 1000 Genomes, dbSNP, Exome Variant Server, ExAC and an internal database. Our quality criteria required coverage of ≥10 quality reads per base and a novel allele frequency (NAF) of ≥0.3. Detected variants were assessed based on their MAF, current literature and widely used Online databases [e.g., OMIM (McKusick-Nathans Institute of Genetic Medicine, Johns Hopkins University, Baltimore, MD, USA), HGMD (Stenson et al., 2014), Uniprot (UniProt Consortium, 2015), locus or disease specific databases] and prediction tools [MutationTaster (Schwarz et al., 2014), PolyPhen2 (Adzhubei et al., 2010), SIFT (Choi et al., 2012), NetGene2 Server (Brunak et al., 1991) and Splice Site Prediction by Neural Network (Reese et al., 1997)].

#### Comparison of Observed Frequencies

We compared the observed frequencies of affected genes in ALS cohorts from the US (Couthouis et al., 2014), Ireland (Kenna et al., 2013), Italy (Chiò et al., 2012) and Great Britain (Morgan et al., 2015) with detected frequencies in our cohort.

#### Generation of a Protein–Protein Interaction Network

To visually link candidate genes and possible modifiers to ALS, and to put them in relation to each another and to confirmed ALS genes, we created a protein–protein interaction network containing 21 disease genes and 13 candidate genes, possible

#### TABLE 1 | Genes analyzed in this study.

fnmol-09-00092 October 7, 2016 Time: 15:23 # 4


The top 14 genes were classified as disease genes when this study was initiated; a further 25 candidate genes, modifiers and risk factors were also included. Gene names are HGNC symbols, transcripts are identified by RefSeq accessions.

risk factors, and modifiers covered by our sequencing panel (**Figure 2**). The protein-protein interaction network was created using the STRING database v10<sup>2</sup> by searching for multiple proteins: ALS2, ANG, ATXN1, ATXN2, C9orf72, CHCHD10, CHMP2B, DPP6, ERBB4, FGGY, FIG4, FUS, GBE1, GLE1, GRN, HNRNPA1, ITPR2, KIFAP3, MATR3, NEFH, OPTN, PFN1, PON3, SETX, SIGMAR1, SLC2A1, SOD1, SPG11, SPG7, TARDBP, UBQLN2, UNC13A, VAPB, VCP. Standard settings were used, network edges set to show confidence, and structural previews inside network bubbles were disabled.

#### RESULTS

#### Identification of Variants in ALS Associated Genes

By analyzing 39 ALS associated genes (**Table 1**), we were able to detect 79 rare variants (European–American MAF ≤ 1% in dbSNP, EVS or ExAC) in 27 genes which passed defined filter criteria (see Variant Filtering) and manual assessment in the Integrated Genome Viewer (IGV, v2.1.28 rev release 175, Robinson et al., 2011; see **Table 2**). Of these, 34 variants have been published previously whereas 45 have not been described before and therefore are considered as being novel. Excluding synonymous substitutions, we identified 54 rare variants in 23 male and 25 female patients (48 patients representing 60% of our cohort). We found that 20 patients of whom 95% (19 out of 20 patients) are considered as sporadic cases carry variants in 14 known disease genes. Additionally we identified variants in candidate genes, modifiers or risk factors in 28 patients (see **Figure 1**).

Pathogenic repeat expansions in the C9orf72 gene were identified in five (6.25%) sporadic patients (mean age of onset: 67.6 years, range 49–76 years). Two of these patients carried additional variants in FIG4 and UNC13A (pat #10), and ITPR2 (pat #373), respectively (see **Table 3**). Furthermore, we identified four patients carrying intermediate length repeat expansions (mean age of onset: 57 years, range 40–68 years). Of these, two individuals carried additional missense and splice variants in ALS2 and UNC13A (pat #26), and SPG11 (pat #729) respectively (see **Table 2**). Given the size of this sample, the remarkable difference in mean age of onset between the patients with intermediate length expansions and carriers of pathogenic repeat expansions is not statistically significant (p = 0.11, Wilcoxon– Mann–Whitney test).

By focusing on candidate genes, modifiers, and risk factors, one interesting finding is the identification of four missense variants in the GRN gene (see **Table 2**). Of these variants, three have already been described as being probably benign in FTD cases (p.T182M), of unknown clinical relevance in FTD and progressive non-fluent aphasia (p.A324T), or as being potentially pathogenic in FTD spectrum disease (p.V77I), respectively (Guerreiro et al., 2008; Pickering-Brown et al., 2008; Yu et al., 2010). Besides this, we detected seven missense variants in the ITPR2 gene which was linked to ALS by several GWASs in the past (van Es et al., 2007), eight variants in FGGY, and three variants in UNC13A, as well as variants in ATXN1, DPP6, GLE1, KIFAP3, NEFH, PON3 and SLC1A2 (see **Table 2**).

#### Co-occurrence of Variants in ALS Associated Genes

Earlier studies supported a complex genetic basis for ALS, which is also supported by protein–protein interactions between known ALS-associated genes, candidate genes, risk factors, and possible

<sup>2</sup>http://string-db.org/


Frontiers in Molecular Neuroscience | www.frontiersin.org October 2016 | Volume 9 | Article 92 |

TABLE 2 |

Identified variants in ALS associated

 genes.




modifiers included in our gene panel (**Figure 2**). In our example, each of the proteins interacts in the context of key proteins for motor neuron degeneration (except CHCHD10 and PON3), pointing toward possible modifying effects of certain variants.

We searched our cohort for patients who carry multiple variants in ALS-associated genes and could identify 15 individuals carrying at least two variants (18.8%, synonymous variants excluded) in ALS-associated genes (**Table 4**). For example, missense variants in the ITPR2 gene were found in co-occurrence with clearly or potentially pathogenic variants in seven (8.75%) individuals. Four of these variants were also detected in an ALS cohort screening by Kenna et al. (2013). In our cohort the mean age of onset in patients who carried a variant in the ITPR2 gene in co-occurrence was 64.0 years compared to 66.6 years in patients carrying any other variants in co-occurrence (differences are not statistically significant, Wilcoxon–Mann–Whitney test). We detected two additional synonymous variants in ITPR2 but according to current knowledge we cannot assess their actual impact on the ITPR2 protein. Four of the 15 patients carried an expanded (Pat #10 and Pat #373) or intermediate (Pat #26 and Pat #729) C9orf72 repeat expansion in co-occurrence.

The mean age of onset in patients where no variant could be detected was 57.8 years, patients who carried one variant showed

TABLE 2 |

Continued

fnmol-09-00092 October 7, 2016 Time: 15:23 # 8

TABLE 3 | Carriers of pathogenic and intermediate C9orf72 repeat expansions.


a mean age of onset of 61.3 years and patients carrying two or more variants had a mean age of disease onset of 65.0 years. In comparison, the overall mean age of disease onset in our cohort was 60.1 years. However, these differences in age of onset are not statistically significant (Kruskal–Wallis Rank Sum Test).

#### Variants in Other NDD Genes

To match the hypothesis of common pathways in different neurodegenerative diseases (NDDs) and to link ALS to other entities of the NDD spectrum, we additionally searched for potentially pathogenic or disease causing variants in 238 genes which are associated with possible differential diagnoses or overlapping phenotypes that are included in our NDD gene panel. We identified 12 patients who carried potentially pathogenic variants in genes that are linked to other entities (**Table 5**).

In patient #38, we detected two heterozygous variants in the GBE1 gene (p.S378R and p.P40T, see **Table 5**). Mutations in GBE1 can cause autosomal recessively inherited adult Polyglucosan body disease (APBD) which is characterized by upper motor neuron signs similar to ALS, early neurogenic bladder, cognitive impairment and decreased or absent activity of the glycogen branching enzyme (Klein, 2013). APBD is one of the conditions that should be considered when establishing the diagnosis of ALS. Unfortunately, we could not investigate whether both variants occur in the compound-heterozygous state in our patient because samples for segregation analysis could not be obtained. Long-range PCR with mutation-specific primers was impossible due to the large distance of more then 170 kb between the variants.

Another interesting finding is the identification of heterozygous variants in the SPG7 gene in four sporadic patients (see **Table 5**). Mutations in SPG7 can cause autosomal recessively inherited spastic paraplegia type 7, but there are also some published cases of obviously autosomal dominant inheritance (e.g., Sánchez-Ferrero et al., 2013). The disease is mainly characterized by spasticity and weakness of the lower limbs. Additional neurologic symptoms might appear in more complex phenotypes. In our cohort, we identified the truncating mutation p.R213\* and the missense mutations p.I743T and p.G349S which are both described as acting disadvantageous on SPG7 protein function (Brugman et al., 2008; Bonn et al., 2010). None of the four patients had further relevant variants in ALS associated genes (only one patient carries an additional missense variant of unknown clinical relevance in the FGGY gene).

We also identified a high number of variants in the NOTCH3, SYNE1, and VPS13A genes as expected in genes of this size. For SYNE1, as mainly loss-of-function mutations are considered as being pathogenic in motor neuron disease (Gros-Louis et al., 2007; Izumi et al., 2013; Noreau et al., 2013). Similarly, only variants which result in a loss or gain of one cysteine residue within epidermal growth factor (EGF)-like repeat domains (Dichgans et al., 2001) are considered pathogenic in NOTCH3, and for VPS13A mostly loss-of-function variants are considered as pathogenic (Tomiyasu et al., 2011). Thus we assume that detected variants in our cohort represent rare polymorphisms. We identified variants in further genes that are included in our gene panel (see **Table 5**) but are unlikely to be implicated in our patients' phenotypes.

By comparing the number of patients identified to carry potentially pathogenic variants in ALS related genes in our cohort with previously published cohort studies, we show that the frequency of affected genes may vary in different populations (**Table 6**). For example, in the VAPB gene we detected variants in 5% of German patients (four cases) whereas in other populations no variants in VAPB were identified at all. Striking differences in frequencies across populations can also be observed for FIG4, FGGY, GRN, ITPR2, and UNC13A. The studies used vastly different strategies for sequencing and variant evaluation and analyzed different gene sets [from 6 genes, partially hotspots only sequenced by Sanger in Chiò et al. (2012) to 169 genes sequenced by NGS in Couthouis et al. (2014)]. Thus we consider this comparison solely to hint toward possible differences in gene frequencies among populations as a consequence of founder effects.

### DISCUSSION

By using next-generation sequencing we analyzed 39 ALSassociated genes in a German cohort of both familial and sporadic ALS patients. In total, we detected 54 rare variants in approved disease genes and possible candidate genes, risk factors, and modifiers (synonymous variants excluded) in 48 patients which represents 60% of our total cohort.

We identified pathogenic or potentially pathogenic variants in 14 analyzed disease genes in 20 patients of whom 19 patients (95%) are affected by sporadic ALS. This finding is unexpected, as it demonstrates that a genetic background can actually be found in a major proportion of seemingly sporadic cases (25%; 19 of 74 patients with sALS). We also would have expected to find more variants in familial cases. Although guidelines and recommendations on how to evaluate unknown variants are published (see for example Richards et al., 2015), the assessment of the actual pathogenicity of detected unknown variants with regard to the patients' phenotypes remains challenging and clear evidence on how a certain variant impairs the phenotype can only be achieved by extensive functional studies.

By focusing on possible candidate genes, risk factors, and modifiers, an interesting finding is the detection of heterozygous missense variants in the GRN gene in four patients affected by pure ALS (see **Table 2**). Three of the identified variants (p.V77I, p.V121M and p.A324T) are classified as being potentially pathogenic. Loss-of-function mutations in GRN are considered causative for frontotemporal lobar degeneration with ubiquitinpositive inclusions (Mackenzie et al., 2006). Recent evidence though suggests that missense mutations in GRN are also linked to the pathogenesis of ALS, especially as ALS and frontotemporal dysfunction are considered to represent a continuum of overlapping phenotypes, and a large proportion of ALS patients additionally experience frontotemporal dysfunction and vice versa (Sleegers et al., 2008; Cannon et al., 2013). Based on our findings, we recommend that GRN gene analysis should be included in routine molecular diagnostic settings and should also be considered in cases of pure ALS without frontotemporal involvement. Further, we detected seven missense variants in the ITPR2 gene. Although Fernández-Santiago et al. (2011) as well as Chen et al. (2012) could not confirm an association of variants in ITPR2 with ALS in a German and a Chinese cohort by SNP genotyping, we speculate that variation in the ITPR2 gene could act as a modulating factor in ALS. A modulating effect might also exist for variants in FGGY (eight variants), GRN (four variants) and UNC13A (three variants).

These findings reflect the overall challenges in assessing the relevance of rare variants with respect to the phenotype as functional studies investigating the actual effect of these variants are largely missing. However, by the implementation of NGS in clinical genetics, we are now faced with increasing numbers of genes published as being possibly implicated in the pathogenesis of ALS. Such candidate genes gain further support from protein-protein interaction data. As rare variants


#### TABLE 4 | Continued

fnmol-09-00092 October 7, 2016 Time: 15:23 # 12


in ALS associated genes according to current knowledge rather represent modifiers with effect on risk of developing the disease, age of onset, severity, or progression rate than disease causing mutations, further effort has to be made to understand how these modulating effects become evident in ALS. Investigating such modulating effects might lead to the identification of pathways that are not yet linked to ALS, enhancing our knowledge of ALS pathogenesis and higher-level neurodegenerative processes.

By performing repeat length analysis we identified five sporadic patients (6.25%) carrying pathogenic repeat expansions in the C9orf72 gene. This is in line with Majounie et al. (2012) who reported on 5.2% of C9orf72 repeat expansion carriers amongst German ALS patients. In two carriers of a pathogenic repeat expansion, we detected additional variants in ALS-associated genes. Although van Blitterswijk et al. (2012) suggested that additional genetic factors contribute to ALS pathogenesis in some carriers of a pathogenic C9orf72 repeat expansion, we cannot assess the impact of additional variants on the patients' phenotypes in our cohort study. We identified four further patients carrying intermediate length repeat expansions. According to recent literature, these might be pathogenic in ALS as patients carrying 20–29 repeats are phenotypically similar to those with more than 30 repeats (Byrne et al., 2014). However, as intermediate length repeats have been detected in both patients and healthy controls, their actual pathogenicity still remains unclear (Rohrer et al., 2015). Of the four individuals with intermediate length repeat expansions, two patients carried additional variants in disease related genes. In our cohort, patients with intermediate length repeats had an earlier age of onset than carriers of a pathogenic repeat expansion (averages of 57.0 and 67.6 years, respectively). This counter-intuitive result leads us to speculate that age of onset was not primarily influenced by the length of repeat expansions but possibly by other factors such as additional variants in other genes. However, we cannot draw a firm conclusion due to our limited cohort size. Surprisingly, we did not detect pathogenic repeat expansions in any of the familial cases, although this might also be because of the small sample size.

To evaluate the hypothesis that ALS might be of complex genetic origin, we searched our cohort for patients carrying more than one potentially disease-causing variant. We found that 15 patients (18.8% of our cohort, synonymous variants excluded) carry two or more variants in ALS-associated genes and that four of these 15 patients additionally carry an expanded or intermediate C9orf72 repeat expansion. According to current findings, a complex model of inheritance is used to explain phenomena like reduced penetrance or even intrafamilial phenotypic variability. A hypothesis by Cady et al. (2015) for example implies that disease onset is influenced by the burden of rare variants in ALS-associated genes. The authors reported that 3.8% of 391 study participants harbored two or more variants in 17 analyzed disease genes and that these individuals had disease onset 10 years earlier than patients carrying only one variant. The considerable difference in percentage of patients carrying two or more variants (3.8% in Cady et al., 2015 vs. 18.8% in this study) might be explained by the fact that we included not only variants in approved disease genes but also in candidate genes, modifiers, and risk factors. In contrast, Cirulli et al. (2015) did not report an effect of the number of variants on the age of onset in their cohort of 2869 ALS patients and 6405 controls, but they do not draw a strong conclusion as they did not test for pathogenic C9orf72 repeat expansions. In our data, we do see a later age of onset in patients carrying two or more variants. However, due to our smaller sample size, we cannot make statistically significant observations on a possible correlation and we cannot exclude that co-occurrence of multiple variants might have a disadvantageous effect on disease onset, severity, disease duration, or site of onset by affecting disease causing variants. As an example, the identification of ITPR2 variants in co-occurrence in seven patients might hint at a possible negative effect of additional variants in the ITPR2 gene. Further studies should include both next-generation sequencing and tests for pathogenic repeat expansion in a large cohort to resolve this open question.


To genetically and mechanistically link ALS to other pathologies of the NDD spectrum, we searched our cohort for potentially pathogenic variants in 238 genes that are associated with overlapping phenotypes and are covered by our diagnostic panel.

We identified potentially pathogenic variants in neurodegeneration-related genes in 12 patients. Although compound-heterozygosity for the detected variants in GBE1 in pat #38 is not proven, we speculate that both variants might be at least concurrently causative, especially as the patient revealed UMN-dominant ALS, cognitive impairment, and progressive non-fluent aphasia (PNFA) upon his last clinical examination in 2012. GBE1 is a glycogen branching enzyme which is involved in glycogen synthesis. According to Ngo and Steyn (2015), there is a link between the selective degeneration of neurons in ALS and metabolic alterations: Deficits caused by decreased glucose metabolism may trigger hyperexcitability and subsequent selective degeneration of upper and lower motor neurons. Although the underlying mechanisms are still unclear, Wang et al. (2015) could show that the FUS protein (juvenile ALS) interacts to a great extent with mitochondrial enzymes and proteins involved in glucose metabolism. With regard to these presumptions, we speculate that pathogenic variants in GBE1 might be causative

TABLE 6 | Percentage of patients carrying potentially pathogenic variants in ALS associated genes (missense, splicing, small Indels only) (American: Couthouis et al., 2014; Irish: Kenna et al., 2013; Italian: Chiò et al., 2012; British: Morgan et al., 2015).


for ALS or motor neuron degeneration, and that metabolic processes and involved genes must be taken into account in ALS genetics.

We detected known heterozygous variants in SPG7 (paraplegin) in four patients. Recent evidence suggests that mutations in SPG7 might be relevant in PLS as Mitsumoto et al. (2015) reported on the identification of a pathogenic heterozygous variant in SPG7 in a patient affected by PLS. Paraplegin is part of the metalloprotease AAA complex, an ATP-dependent proteolytic complex located on the inner mitochondrial membranes, and functions in controlling protein quality and ribosomal assembly. Ferreirinha et al. (2004) showed that paraplegin-deficient mice develop axonal swellings as a consequence of accumulation of mitochondria and neurofilaments in the spinal cord which precedes axonal degeneration by impaired anterograde axonal transport. Although further studies are needed to assess the functional role of SPG7 in human motor neurons, these findings hint at an important role of SPG7 in motor neuron survival and support our hypothesis, that paraplegin is implicated in the pathogenesis of ALS and those pathogenic mutations in SPG7 must be taken into account regarding genetic testing in ALS.

In summary, our results support recent observations whereby a genetic background is implicated in the sporadic form of ALS to a higher extent than assumed so far, and strengthen the upcoming hypothesis of ALS being a distinct manifestation of higher-level neurodegenerative processes rather than representing a discrete entity. Further, our results contribute to current discussions on a possible pathogenicity of intermediate repeat expansion in the C9orf72 gene, especially in the interplay with additional variants in other ALS associated genes. In contrast to previously published studies, we could not prove an earlier age of disease onset in patients carrying multiple variants but speculate that variants in the ITPR2 gene might act as a modulating factor in ALS. Additionally, our results lead us to assume that variants in GRN and SPG7 might be implicated in the pathogenesis of ALS which is in line with the aforementioned hypothesis of common neurodegenerative processes leading to distinct phenotypes. Surprisingly, we did not detect clearly pathogenic variants in SOD1 in our cohort, even though this gene is supposed to have a high impact on disease, encouraging us to launch a debate on the actual significance of SOD1 in Germany.

#### CONCLUSION

We investigated 39 ALS-associated genes in a German cohort of 80 familial and sporadic ALS patients utilizing nextgeneration sequencing. We identified 22 variants in diseasecausing genes in 20 patients and additionally 32 variants in candidate genes, risk factors, and modifiers in 28 patients. Thus we detected variants in ALS-associated genes in 60% of our study participants, of whom the vast majority are sporadic cases. Surprisingly, pathogenic repeat expansions in C9orf72 and potentially pathogenic variants in SOD1 were both detected at lower frequencies than expected. Instead we identified potentially pathogenic variants in the GRN gene in four patients, indicating that the impact of GRN mutations is not limited to ALS-FTD and might account for pure ALS, too.

Furthermore, our cohort enabled us to evaluate the hypotheses that ALS is of complex genetic origin. According to this hypothesis, numerous variants have some degree of influence on the clinical phenotype caused by the pathogenic mutation. We did in fact identify patients carrying variants in more than one ALS-associated gene. In contrast to other studies, however, our results do not show that patients with multiple variants have an earlier age of onset.

As ALS should be seen in the context of wider neurodegenerative disorders, we investigated our cohort for potentially pathogenic variants in 238 neurodegeneration related genes. The most interesting findings are the identification of two variants in the GBE1 gene that might be causative in a patient with UMN-dominant ALS and the detection of heterozygous variants in SPG7 in four ALS patients. These findings would benefit from extensive high-throughput sequencing in large patient and control cohorts of different ethnic background in order to more accurately assess the overall variability in ALSassociated genes and to better evaluate their impact on the disease.

Our results support the notion that next-generation sequencing could help uncover the genetic heterogeneous basis of ALS and thus argue for the broader application of NGS techniques in routine diagnostic settings. Therefore, our results are of immediate relevance for clinical genetics as we recommend that genetic testing in German patients should be offered not only to those with familial ALS but also to those with apparently sporadic ALS. We propose a two-stage strategy starting with a C9orf72 repeat analysis, followed by comprehensive gene panel sequencing if C9orf72 negative. To meet the high number of possible differential diagnoses that mimic ALS, genes causing FTD, HSP, spinal muscular atrophy (SMA) and other entities that impair motor neuron function should be included. Whereas Sanger sequencing focused on a few commonly affected genes such as SOD1, panel sequencing offers the opportunity to cover all disease-associated genes in only one approach and thus reveals the genetic heterogeneity of ALS and increases detection rates. Additionally, panel sequencing allows for the detection of multiple variants acting on the individual phenotype which might enable statements for example on disease progression or severity. We hope that our results will contribute to deeper knowledge which will allow the identification of new therapeutic targets for example by interfering with distinct pathways or personalized therapeutic approaches in the future.

It was our aim to broaden the genetic landscape of ALS. We detected previously identified ALS-causing mutations, novel variants within recognized disease-causing genes and candidate genes, in addition to modifiers and risk factors. Assessing the impact of newly detected variants and their potential contribution to the ALS phenotype requires further investigation in order to determine their functional relevance. For several patients who gave their informed consent, we collected fibroblasts to provide the basis for the necessary functional work up.

#### AUTHOR CONTRIBUTIONS

fnmol-09-00092 October 7, 2016 Time: 15:23 # 16

Study concept and design: SK, MS, JP, and SB. Acquisition of clinical data and blood sample collection: JP and TGr. Analysis and interpretation of genetic data: SK, FB, AS, and MM. Drafting of manuscript: SK. Critical revision of manuscript: SK, FB, AS, MM, MS, LS, TGa, TGr, JP, and SB.

#### REFERENCES


#### FUNDING

Supported by the German Center for Neurodegenerative Diseases (DZNE), Intersite project RO010 to J.P.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2016.00092



lateral sclerosis: a genome-wide association study. Lancet Neurol. 6, 869–877. doi: 10.1016/S1474-4422(07)70222-3


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

Copyright © 2016 Krüger, Battke, Sprecher, Munz, Synofzik, Schöls, Gasser, Grehl, Prudlo and Biskup. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Protective LRRK2 R1398H Variant Enhances GTPase and Wnt Signaling Activity

Jonathon Nixon-Abell1,2† , Daniel C. Berwick1,3† , Simone Grannó<sup>1</sup> , Victoria A. Spain<sup>1</sup> , Craig Blackstone<sup>2</sup> and Kirsten Harvey<sup>1</sup> \*

<sup>1</sup> Department of Pharmacology, UCL School of Pharmacy, University College London, London, UK, <sup>2</sup> Neurogenetics Branch, National Institute of Neurological Disorders and Stroke – National Institutes of Health, Bethesda, MD, USA, <sup>3</sup> Department of Life, Health and Chemical Sciences, The Open University, Milton Keynes, UK

Mutations in LRRK2 are a common cause of familial and idiopathic Parkinson's disease (PD). Recently, the LRRK2 GTPase domain R1398H variant was suggested in genetic studies to confer protection against PD but mechanistic data supporting this is lacking. Here, we present evidence that R1398H affects GTPase function, axon outgrowth, and Wnt signaling in a manner opposite to pathogenic LRRK2 mutations. LRRK2 R1398H GTPase domain dimerization and GTP hydrolysis were increased whereas GTP binding was reduced, leading to a decrease in active GTP-bound LRRK2. This protective variant also increased axon length of primary cortical neurones in comparison to wild-type LRRK2, whereas the R1441G LRRK2 pathogenic mutant decreased axon outgrowth. Importantly, R1398H enhanced the stimulatory effect of LRRK2 on canonical Wnt signaling whereas the G2385R risk variant, in accordance with all previously tested pathogenic LRRK2 mutants, had the opposite effect. Molecular modeling placed R1398H in close proximity to PD-causing mutations suggesting that this protective LRRK2 variant, like familial mutations, affects intramolecular RocCOR domain interactions. Thus, our data suggest that R1398H LRRK2 is a bona fide protective variant. The opposite effects of protective versus PD associated LRRK2 variants on GTPase function and canonical Wnt signaling activity also suggests that regulation of these two basic signaling mechanisms is important for neuronal function. We conclude that LRRK2 mediated Wnt signaling and GTPase function are fundamental in conferring disease susceptibility and have clear implications for therapeutic target identification.

Keywords: GTPase activity, LRRK2, Parkinson's disease, protective genetic variant, Wnt signaling

## INTRODUCTION

Autosomal-dominant mutations in LRRK2, encoding leucine-rich repeat kinase 2 (LRRK2), are the most common known cause of inherited Parkinson's disease (PD; Paisán-Ruíz et al., 2004; Zimprich et al., 2004). Patients with LRRK2 mutations display symptoms and brain pathologies that are largely indistinguishable from those of individuals with idiopathic PD (Paisán-Ruíz et al., 2004; Zimprich et al., 2004; Ross et al., 2006; Kumari and Tan, 2009). Thus, determining the biological role of LRRK2 is of paramount importance to understanding the etiology of PD, and likely to help uncover new therapeutic strategies.

#### Edited by:

Joe Lynch, University of Queensland, Australia

#### Reviewed by:

Giovanni Piccoli, Università degli Studi di Trento, Italy Sunghoe Chang, Seoul National University, South Korea

#### \*Correspondence:

Kirsten Harvey kirsten.harvey@ucl.ac.uk †These authors have contributed equally to this work.

Received: 09 December 2015 Accepted: 22 February 2016 Published: 08 March 2016

#### Citation:

Nixon-Abell J, Berwick DC, Grannó S, Spain VA, Blackstone C and Harvey K (2016) Protective LRRK2 R1398H Variant Enhances GTPase and Wnt Signaling Activity. Front. Mol. Neurosci. 9:18. doi: 10.3389/fnmol.2016.00018

LRRK2 is a multifunctional protein containing both kinase and GTPase activities and a number of protein–protein interaction domains (**Figure 1**). The 'catalytic core' is contained within the Roc (Ras of complex proteins), COR (C-terminal of Roc) and kinase domains (**Figure 1**) and appears essential for LRRK2 function (Berwick and Harvey, 2011). As the only hereditary mutations that are proven to cause PD fall within exons coding for the Roc, COR and kinase domains, the effects of pathogenic mutations on LRRK2 enzymatic activities require further investigation.

The precise extent of the LRRK2 GTPase domain remains controversial. Restricting the LRRK2 GTPase domain to the Roc domain can be justified by the similarity to Ras family small GTPases. Arguing against this are, the observations that isolated Roc protein appears to hydrolyse GTP at a greatly reduced rate compared to full-length LRRK2 (Gilsbach and Kortholt, 2014) and that throughout evolution, Roc and COR domains never occur without each other (Bosgraaf and van Haastert, 2003). This supports the definition of a LRRK2 GTPase "RocCOR tandem" domain.

Based on homology to C. tepidum Roco protein, the LRRK2 RocCOR tandem is predicted to fall into the GAD (G proteins activated by nucleotide-dependent dimerization) class of molecular switches (Gotthardt et al., 2008; Gasper et al., 2009; Gilsbach and Kortholt, 2014). Under the GAD model, two LRRK2 molecules are expected to dimerize via a constitutive interaction between the COR domains holding the two Roc domains in close proximity (**Figures 1** and **3A,B**). Binding of GTP to the Roc domains results in protein dimerization allowing binding of effector proteins. GTP hydrolysis disrupts Roc dimerization, leading to the dissociation of effector proteins (Gotthardt et al., 2008; Gasper et al., 2009; Gilsbach and Kortholt, 2014).

All known PD-causing mutations located within the LRRK2 RocCOR domain have been reported to increase GTP-binding, decrease GTPase activity, or both (**Table 1**) supporting the idea that shifting LRRK2 to the GTP-bound 'on'-state promotes neurodegeneration.

A consensus on the cellular role of LRRK2 is still lacking, with numerous competing – though not mutually exclusive – functions reported. Nonetheless, a comprehensive literature review identified cell biological processes involving LRRK2 that appear to be reproducible (Berwick and Harvey, 2013) including effects on membrane trafficking (Piccoli et al., 2011; Ramonet et al., 2011; reviewed by Gómez-Suaga et al., 2014), cytoskeletal function (Parisiadou et al., 2009; Law et al., 2014; reviewed by Gómez-Suaga et al., 2014) and signal transduction pathways including MAPK, Wnt, TLR, and NFAT pathways (Berwick and Harvey, 2011; Boon et al., 2014; reviewed by Gómez-Suaga et al., 2014).

A growing body of data supports the importance of deregulated canonical Wnt signaling in neurodegenerative disease pathogenesis, including PD (Berwick and Harvey, 2014; Inestrosa and Varela-Nallar, 2014). At least six proteins implicated in PD have been described to modulate this pathway, whilst development of the midbrain dopaminergic neurones that are typically lost in PD is acutely dependent on canonical Wnt signaling (Berwick and Harvey, 2014). Furthermore, decreased Wnt signaling has been reported in PD patients (Cantuti-Castelvetri et al., 2007), as well as in various animal models of parkinsonism (L'Episcopo et al., 2011; Gollamudi et al., 2012). Since Wnt ligands are well established as neuroprotective (Berwick and Harvey, 2014; Inestrosa and Varela-Nallar, 2014), the idea that decreased canonical Wnt signaling is involved in PD pathogenesis is an attractive hypothesis. Importantly, evidence for a crucial role of LRRK2 in canonical Wnt signaling is accumulating. Protein–protein interactions between LRRK2 and a number of Wnt signaling components have been reported, including interactions with disheveled (DVL) proteins, the serine/threonine kinase GSK3β, and LRP6, a Wnt signaling transmembrane receptor (Sancho et al., 2009; Lin et al., 2010; Berwick and Harvey, 2012). Furthermore, over-expressed wildtype LRRK2 enhances the signal strength of activated canonical Wnt signaling (Berwick and Harvey, 2012). Intriguingly, this effect of LRRK2 is weakened by PD-causing mutations in three distinct catalytic domains of LRRK2 – R1441C in the Roc domain, Y1699C in the COR domain, and G2019S in the kinase domain (Berwick and Harvey, 2012) – suggesting that impaired Wnt signaling is a common pathogenic mechanism of familial LRRK2 mutations.

Recently, a number of genetic screens have reported an inherited R1398H LRRK2 variant in the Roc domain (Chen et al., 2010; Tan et al., 2010; Ross et al., 2011; Heckman et al., 2013, 2014) that appears to confer decreased risk of PD. This could prove extremely informative, since a protective variant can be expected to display the opposite behavior to pathogenic variants in disease-relevant assays. In a single study, R1398H was reported to display decreased kinase activity compared to wildtype LRRK2 (Tan et al., 2010). However, this observation should be treated with caution, as this study also found the G2385R



Based on published data, the pathogenic N1437H, R1441C, R1441G, R1441H, and Y1699C mutations all elicit an increase in the fraction of GTP-bound LRRK2. Only one construct, an R1441C RocCOR-kinase, displayed no changes in GTP-binding or GTP hydrolysis; however, this mutation decreased GTPase activity in full length LRRK2 in the same publication. Note that mutations at residues equivalent to R1441 and Y1699 in C. tepidum Roco protein produce consistent results. Key: up and down arrows represent increases and decreases, respectively; dashes represent no change; 'ns' indicates 'not studied.'

risk variant to have increased kinase activity, in contrast to other reports (Jaleel et al., 2007; West et al., 2007; Nichols et al., 2010; Rudenko et al., 2012). In any case, the need to elucidate the functional relevance of the R1398H experimentally is clear.

Here, we report the behavior of the LRRK2 R1398H variant in five assays for which the effect of a bona fide protective mutation in the LRRK2 Roc domain can be predicted: LRRK2 RocCOR tandem domain dimerization, LRRK2 GTP-binding, LRRK2 GTPase assays, axon outgrowth, and canonical Wnt signaling assays. Remarkably, R1398H displays the opposite behavior to pathogenic mutants in all experiments. Furthermore, molecular modeling studies suggest that this amino acid substitution is likely to affect intramolecular RocCOR interactions, consistent with the predicted mode of action for PD-causing mutations in the RocCOR tandem. Thus our data (1) provide strong experimental support for the status of LRRK2 R1398H as a genuine protective variant; (2) increase the weight of evidence that GTP-bound LRRK2 is pathogenic; and (3) provide further data indicating that decreased canonical Wnt signaling is a key pathomechanism underlying PD.

#### MATERIALS AND METHODS

#### Molecular Cloning

pDS-BAIT (pDS; Dualsystems Biotech) plasmids containing the LRRK2 Roc and RocCOR domains (encoding amino acids 1330–1515 and 1335–1845, respectively), pACT2 (Clontech) containing the LRRK2 RocCOR domain and pYTH16 containing the intracellular domain of LRP6 (amino acids 1416–1613) have been described previously (Daniëls et al., 2011; Berwick and Harvey, 2012). pCHMWS vectors expressing 3× FLAGtagged wild-type LRRK2 and LRRK2-T1348N were a generous gift from Dr. Jean-Marc Taymans (Daniëls et al., 2011). pRK5 myc-LRRK2 has also been described previously (Sancho et al., 2009). R1398H, R1398H/R1441G and G2385R mutations were introduced using the QuikChange Lightening site-directed mutagenesis kit (Agilent) according to the manufacturer's instructions. All constructs were verified by DNA sequencing.

#### Culture of Immortalized Cell Lines

HEK293 cells and SH-SY5Y cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% (v/v) fetal bovine serum, 100 U/ml penicillin G and 100 µg/ml streptomycin at 37◦C and 5% CO2. Transient transfection was performed using FuGENE HD (Roche) according to the manufacturer's instructions, using a 2.5 µL transfection reagent to 1 µg DNA ratio. In all cases, cells were harvested 24 h after transfection.

#### Quantitative Yeast-Two Hybrid

The L40 yeast strain (Invitrogen) was co-transformed with pDS Roc or RocCOR bait and pACT2 wild-type or mutant RocCOR prey constructs, and the Y190 yeast strain (Clontech) was cotransformed with the pYTH16 LRP6 intracellular domain bait and pACT2 wild-type or mutant RocCOR prey constructs. Transformations were spread on selective dropout media (Clontech) lacking leucine and tryptophan for transformation controls, or leucine, tryptophan and histidine, supplemented with 0.5 mM 3-aminotriazole (L40 strain) or 10 mM 3-aminotriazole (Y190 strain; for suppression of 'leaky' histidine expression;

Sigma-Aldrich) for nutritional selection. After incubation at 30◦C for 3 days, prototrophic colonies were picked and used to inoculate minimal SD (Clontech) media lacking leucine and tryptophan. Samples were subsequently incubated shaking at 30◦C overnight. Cell pellets were then resuspended in Z-buffer (60 mM Na2HPO4, 40 mM NaH2PO4, 10 mM KCl, 1 mM MgSO4.7H2O) containing 40 mM β-mercaptoethanol, followed by lysis in 0.1% (w/v) SDS (Sigma-Aldrich) and 0.1% (v/v) chloroform (Sigma-Aldrich). After the addition of chlorophenored-β-D-galactopyranoside (Sigma-Aldrich), the color change was recorded at 540 nm and readings adjusted for turbidity of the yeast suspension at 620 nm. The background signal (bait plus empty pACT2 vector) was subtracted from each reading and values were normalized to the wild-type RocCOR response, which was set at 100%. All protein interactions were assayed in three to five independent experiments in triplicate.

#### Molecular Modeling

Molecular modeling was performed on the C. tepidum Roco structure (PDB: 3DPU; Gotthardt et al., 2008) using Chimera (Pettersen et al., 2004). Amino acid substitutions were performed with the swapaa command using the Dunbrack backbonedependent rotamer library (Dunbrack, 2002).

#### GTP-Binding Assay

HEK293 cells were transfected with 3× FLAG-tagged T1348N, R1398H or wild-type LRRK2. The GTP binding assay was performed similarly as described by others (Korr et al., 2006). Briefly, cells were lysed for 10 min on ice in lysis buffer G (100 mM Tris/HCl pH 7.5, 50 mM KCl, 1 mM EDTA, 0.1 mM DTT, 5 mM MgCl2, 1% Triton X-100, protease inhibitor cocktail, Roche), and lysates were centrifuged for 10 min at 20,000 g and 4 ◦C. Supernatants containing 100 µg protein each, as assessed by QuickStart Bradford assay (Bio-Rad), were incubated for 80 min at 4◦C with 30 µl of GTP-Sepharose bead suspension (Sigma) that was pre-treated with 100 µg/ml BSA (Pierce) at 4◦C for 1 h. Samples were washed three times with 500 µl lysis buffer G, before bound protein was eluted using 100 µM GTP in lysis buffer G. The resulting eluate was added to 1× NuPAGE sample buffer (Invitrogen) and heated for 6 min at 96◦C. The eluates were analyzed by SDS-PAGE and immunoblotting. Initially, protein was loaded into 4–12% (w/v) BisTris pre-cast gels (Invitrogen), prior to transfer to polyvinylidine fluoride membranes (Millipore). Non-specific bands were blocked for 1 h at 37◦C with 5% (w/v) skimmed milk in PBS plus 0.1% (v/v) Tween 20. Anti-Calnexin antibody (Abcam) was used at 1:4000, and anti-FLAG antibody (Sigma-Aldrich) was used at 1:3000 at 4◦C overnight. For detection, an HRP-conjugated anti-rabbit secondary antibody (Santa Cruz Biotechnology) was used at a final dilution of 1:2000, together with the SuperSignal West Pico Chemiluminescent Substrate (Pierce).

#### GTPase Assay

GTPase assays were carried out according to Daniëls et al. (2011). Initially, HEK293 cells were transfected with 3× FLAGtagged T1348N, R1398H or wild-type LRRK2 and lysed after 24 h in GTPase lysis buffer [20 mM Tris/HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 10% Glycerol, protease inhibitor cocktail (Roche), phosphatase inhibitor cocktail 2 (Sigma-Aldrich)]. Cell lysates were clarified as above, added to 40 µl of anti-FLAG M2 affinity gel (Sigma-Aldrich) and incubated overnight at 4◦C on a turning disk in order to purify the FLAG-tagged proteins. The affinity gel was subjected to centrifugation (4◦C, 100 g, 3 min), followed by two washes in GTPase lysis buffer and a brief rinse in GTPase buffer (20 mM Tris/HCl pH 7.5, 150 mM NaCl, 10 mM MgCl2, 0.02% Triton X-100). Proteins were eluted from beads with 3× FLAG peptide (Sigma-Aldrich) in GTPase buffer according to manufacturer's instructions. The protein concentration of eluates were calculated from a serial BSA dilution curve, with purity assessed by running a small volume of each sample on an SDS-PAGE gel and staining with GelCode Blue Stain Reagent (Pierce). GTPase assays were performed according to Margalit et al. (2004) using 80 nM LRRK2 protein in GTPase buffer containing 20 U/ml pyruvate kinase (EC 2.7.1.40)/lactate dehydrogenase (EC 1.1.1.27; Sigma-Aldrich), 600 µM NADH (Sigma-Aldrich), 1 mM PEP (Sigma-Aldrich) and 500 µM GTP (Sigma-Aldrich) in a final volume of 200 µl. Reaction mixes were equilibrated to 30◦C for 10 min before the reactions were initiated by the addition of GTP and thorough mixing of the contents. Depletion of NADH was measured by monitoring the decrease in absorbance at 340 nm every 5 min across a 50 min period using a VersaMax microplate reader (Molecular Devices).

#### Primary Cortical Neuronal Cultures

Primary cultures of rat cortical neurones were prepared from E18 embryos obtained from timed-pregnant Sprague-Dawley rats (Taconic Biosciences, Hudson, NY, USA) narcotized with CO<sup>2</sup> (in cylinders) then decapitated using a guillotine. All animal studies were approved by the National Institute of Neurological Disorders and Stroke/National Institute on Deafness and Other Communication Disorders Animal Care and Use Committee (Protocol 1151-12). Neurones were transfected with wild-type and mutant myc-tagged LRRK2 using the Amaxa Rat Neuron Nucleofector Kit, program 0-03, according to the manufacturer's protocol (Lonza Group, Basel, Switzerland). Neurones were then plated at a density of approximately 2.6 × 10<sup>4</sup> /cm<sup>2</sup> on cover slips and maintained as described previously (Zhu et al., 2006). At 7 days in vitro (DIV) neurones were fixed for 10 min with 4% paraformaldehyde, and permeabilized for 15 min with 0.05% Triton-X (Sigma-Aldrich) prior to a 1 h block in 5% NGS (GIBCO). Slides were then immunostained with primary and Alexa Fluor secondary antibodies, mounted using ProLong gold (Life Technologies), and imaged using a Zeiss LSM710 laser-scanning confocal microscope. Primary antibodies against the following proteins were used: myc-epitope (Santa Cruz Biotechnology), MAP2 (Abcam) and Tau-1 (Abcam). Alexa Fluor 488 (rabbit), 555 (mouse) and 633 (goat) secondary antibodies (Thermo Scientific) were used against myc, tau and MAP-2 epitopes respectively.

#### Axon Length Measurements

Axon outgrowth properties of neurones were quantified manually. Three to six coverslips from three independent

experiments for each genotype were analyzed. Axonal outgrowth and branching of dissociated neurones were quantified manually and verified using the NeuronJ plugin for ImageJ. Transfected neurones were identified using the myc-epitope antibody, whilst the length of the longest Tau-1 stained process from each neurone was measured. At least 40 neurones were quantified for each genotype, myc vector control, myc-LRRK2 wild-type, myc-LRRK2 R1398H, myc-LRRK2 R1441G, and myc-LRRK2 R1398H/R1441G from at least three coverslips.

#### Luciferase Assays

Canonical Wnt activity was measured using the TOPflash reporter plasmid (Veeman et al., 2003) in human dopaminergic SH-SY5Y cells as described previously (Berwick and Harvey, 2012). Cells were extracted 24 h post-transfection using Passive Lysis Buffer (Promega) and assays performed using a Dual Luciferase Reporter Assay kit (Promega) and Turner Instruments 20/20 luminometer. Luciferase values were normalized to co-transfected Renilla plasmid to adjust for transfection efficiency, and then corrected to values from parallel experiments performed using the FOPflash control plasmid (Veeman et al., 2003).

### Statistical Analysis

GTPase assays (**Figure 4D**) were tested by two-way ANOVA with repeated measures, with the independent variables genotype, time, and time × genotype, followed by post hoc analysis by two-sided Dunnett's testing. Axonal branching was assessed by Kruskal–Wallis (**Figure 6B**) or Mann–Witney (Supplementary Figure S2A) tests. Axon length was analyzed by one-way ANOVA followed by Bonferroni post hoc analysis (**Figure 6C**) or Student's t-test (Supplementary Figure S2B). For axon length analysis, outliers (values defined as differing from the mean by 2 or more standard deviations) were first excluded. All other experiments were analyzed by one-way ANOVA for the effect of genotype followed by a two-sided Dunnett's test with wild-type LRRK2 considered the control. Student's t-test was performed using Excel, all other statistics were performed with SPSS software. All error bars represent the standard error of the mean.

Frontiers in Molecular Neuroscience | www.frontiersin.org March 2016 | Volume 9 | Article 18 |

### RESULTS

### In Contrast to LRRK2 GTPase Mutants Causing PD, R1398H Increases RocCOR Dimerization

To investigate functional effects of R1398H, we studied this variant in assays of LRRK2 RocCOR domain dimerization. These sensitive quantitative yeast-two hybrid (Q-YTH) assays have been used previously in our laboratory to show that the PD-causing R1441C, R1441G, R1441H, and Y1699C mutations significantly weaken RocCOR domain dimerization (Daniëls et al., 2011). Intriguingly, R1398H elicited an effect opposite to these pathogenic mutations, with increased RocCOR dimerization observed (**Figure 2A**). The R1441G mutant was studied as a pathogenic control showing as expected a significant decrease in RocCOR dimerisation (**Figure 2A**). Consistently, R1398H also strengthened interaction between the LRRK2 RocCOR tandem domain and an isolated wild-type LRRK2 Roc domain, whilst the R1441G, R1441H, and Y1699C pathogenic mutations weakened this interaction (**Figure 2B**). R1441C was also studied in this experiment, displaying a non-significant trend toward decreased interaction (**Figure 2B**). These effects were not due to changes in protein expression (**Figure 2C**). Thus, in contrast to proven PD-causing RocCOR mutations, R1398H enhanced intermolecular dimerization within the LRRK2 GTPase domain.

### Molecular Modeling Suggests that LRRK2 R1398H Affects Intramolecular RocCOR Interaction

To examine the molecular mechanism underlying altered RocCOR dimerization, the predicted location of R1398 was examined by molecular modeling of the closest available protein structure: the RocCOR tandem domain of C. tepidum Roco protein (Gotthardt et al., 2008). Supporting the importance of R1398 in RocCOR dimerization, the equivalent residue in Roco (Q519) resides on the internal face of each Roc domain. To examine the role of human R1398, this amino acid was swapped to arginine in our model (**Figure 3C**). Using the most probable rotamer conformation, the long basic side chain of arginine projected toward the COR domain of the same molecule. Indeed, R1398 was predicted to bond with the hydroxyl

Roco protein is shown as a space-filled structure, viewed from two orientations. Roc and COR domains are labeled. For clarity, the Roc and COR domains of one molecule are depicted in shades of blue, the domains of the second molecule are yellow. (B) A magnified top view of the Roco dimer as a ribbon model. The internal location on both Roco molecules of the equivalent residue to human R1398 is colored red, and labeled. Arginine side chains are shown. (C) A magnified image of the modeled R1398, showing the arginine side chain projecting toward the equivalent residue S1671 in human LRRK2. A potential hydrogen bond between these residues is shown. (D) Swapping arginine to histidine at the R1398-equivalent site prevent hydrogen bonding to S1671. (E) Comparison of the sequence around R1398 and S1671 in human LRRK2 with those in human LRRK1 and C. tepidum Roco protein. Note that a serine at the 1671-position is conserved in mammalian LRRK2 proteins and amongst Roco proteins in lower organisms. By contrast, LRRK1 proteins contain a histidine residue at this position.

group of a conserved serine in the COR domain (S778 in Roco, S1671 in LRRK2, predicted hydrogen bond distance: 2.776 Å). This is an intriguing possibility, since the R1441C/G and Y1699C pathogenic mutations have also been predicted to modify intramolecular RocCOR interactions (Gotthardt et al., 2008; Daniëls et al., 2011). In agreement, conversion of R1398 to the much shorter histidine – representative of R1398H – prevented bonding with the conserved serine (**Figure 3D**). Taken together, our molecular modeling suggests that the observed increase in intermolecular RocCOR dimerization caused by R1398H (**Figure 2**) likely occurs via an indirect mechanism, involving changes to intramolecular interactions between Roc and COR domains.

#### R1398H Decreases the Fraction of GTP-Bound LRRK2

Numerous reports indicate that PD-causing mutations in the LRRK2 RocCOR tandem domain increase the ratio of GTPbound ('on') LRRK2 to GDP-bound ('off') LRRK2, by weakening GTPase activity and/or facilitating GTP binding (**Table 1**). In principle, a protective amino acid substitution located within this portion of LRRK2 would be expected to have the opposite effect. Thus, the effect of the R1398H variant on LRRK2 GTPase function was examined directly, using full-length human LRRK2 expressed in mammalian cells. Firstly, this variant was studied in GTP binding assays. As reported by others, wildtype LRRK2 bound strongly to immobilized GTP, whilst a

well-characterized mutant, LRRK2 T1348N, had negligible GTPbinding capacity (**Figures 4A,B**). Strikingly, the R1398H variant also displayed weakened GTP binding (∼47% of wild-type), although this value was more than threefold greater than for T1348N LRRK2, indicating that GTP binding is not abolished entirely.

We next examined the effect of the LRRK2 R1398H variant on GTP hydrolysis in vitro, using steady-state GTPase assays. FLAG-tagged wild-type LRRK2, and LRRK2 containing T1348N and R1398H amino acid substitutions were purified from HEK293 cells (**Figure 4C**), and equimolar amounts were used in subsequent experiments (**Figure 4D**). This protein produced a steady turnover of GTP, confirming that wild-type LRRK2 possesses intrinsic GTPase activity (**Figure 4D**, open circles). Unsurprisingly, since T1348N LRRK2 is almost unable to bind GTP, this mutant possessed very little GTPase activity, with GTP hydrolysis undetectable until the 20 min time point (**Figure 4D**, gray circles). However, R1398H LRRK2 showed a marked increase in GTP hydrolysis relative to wild-type LRRK2 (**Figure 4D**, black triangles). In summary, the protective R1398H LRRK2 variant weakens GTP binding but increases steady-state GTP hydrolysis, both of which are consistent with a decrease in the proportion of GTP-bound LRRK2.

### The Protective R1398H LRRK2 Variant Increases Canonical Wnt Signaling

We have previously reported that PD-causing mutations in the Roc, COR and kinase domains of LRRK2 weaken the activation of canonical Wnt signaling that is elicited by disheveled (DVL) proteins (Berwick and Harvey, 2012). Consistent with this finding, the LRRK2 R1441G pathogenic mutant and importantly the WD40 domain G2385R PD risk variant also reduce pathway activation relative to wild-type LRRK2 in SH-SY5Y cells (**Figure 5A**). By contrast, LRRK2 R1398H enhanced DVL1-driven Wnt activation almost twice as much as wildtype LRRK2 (**Figure 5A**). This opposing effect of the protective LRRK2 variant to that described for pathogenic variants in four distinct LRRK2 domains is notable, and cannot be attributed to altered expression levels (**Figure 5B**). These experiments suggest a strong correlation between PD risk conferred by LRRK2 variants and regulation of canonical Wnt signaling activity.

To further examine a possible causal mechanism that affects canonical Wnt signaling, we investigated the interaction between the LRRK2 R1398H variant and the canonical Wnt co-receptor LRP6 in Q-YTH experiments. Using the LRP6 intracellular domain as bait and the LRRK2 RocCOR tandem domain as prey, we observed a decrease in the interaction strength with the R1398H protective variant relative to wildtype LRRK2 (Supplementary Figures S1A,B). Since pathogenic LRRK2 GTPase mutants also show a decrease in protein– protein interaction (Supplementary Figure S1A, Berwick and Harvey, 2012) altered LRRK2-LRP6 interactions do not explain the increase in canonical Wnt signaling activity of the R1398H protective variant relative to the decrease observed for pathogenic LRRK2 variants.

### Protective R1398H LRRK2 Variant Increases Axon Length in Cultured Cortical Neurones

Temporary differences in neurite outgrowth between LRRK2 knockout, mutant and wild-type neurones have been reported in various experimental systems (MacLeod et al., 2006; Dächsel et al., 2010; reviewed by Gómez-Suaga et al., 2014). As LRRK2 GTPase activity and Wnt signaling activity (Salinas, 2012; Gómez-Suaga et al., 2014) affect neurite outgrowth, we decided to examine axon length and branching as a correlate for neurite outgrowth and complexity in rat cortical neurones in primary culture overexpressing LRRK2 wild-type and variants at 7 DIV. Over-expression of wild-type LRRK2 had no effect on mean axonal length (p = 0.082) or axon branching (p = 0.695) relative to neurones transfected with empty vector control (Supplementary Figure S2). Furthermore, no difference in axon branching was observed between the different LRRK2 genotypes (Kruskal–Wallis χ <sup>2</sup> = 0.943, p = 0.815; **Figure 6B**). However, the overexpression of LRRK2 mutants had a marked effect on axon length (F = 23.52, p < 0.001; **Figure 6C**). In agreement with previous work, neurones overexpressing the pathogenic LRRK2

R1441G mutant showed a reduction in axon length relative to wild-type LRRK2 (MacLeod et al., 2006; Cho et al., 2013). By contrast, the protective R1398H variant increased axon length in comparison to wild-type overexpressing cells. Remarkably, R1398H was able to rescue the effect of R1441G, as a double R1398H/R1441G mutant phenocopied the effect of the R1398H single mutant (**Figure 6**).

### DISCUSSION

Our data provide robust experimental evidence consistent with the idea that the familial LRRK2 R1398H variant is protective against PD. Firstly, in contrast to the pathogenic R1441C, R1441G, R1441H, and Y1699C variants (**Figure 2A**; Klein et al., 2009; Daniëls et al., 2011; Law et al., 2014), R1398H

increases LRRK2 RocCOR domain dimerization (**Figure 2**). Secondly, our molecular modeling suggests that this amino acid substitution will affect the interaction between the Roc and COR domains of the same molecule (**Figure 3**). Similarly, the R1441C/G/H and Y1699C mutations, and indirectly the N1437H mutation, have been suggested to affect intramolecular RocCOR interactions and consequently RocCOR dimerization as a proposed molecular pathomechanism (Gotthardt et al., 2008; Klein et al., 2009; Daniëls et al., 2011). Thirdly, our GTP-binding and GTP hydrolysis assays point toward R1398H decreasing the ratio of GTP-bound to GDP-bound LRRK2. This is in direct contrast to PD-causing mutations in the LRRK2 RocCOR tandem domain (**Table 1**). Fourthly, the R1398H protective variant increases axon length in cortical neurones, whereas LRRK2 mutants cause shortening of axons in equivalent assays (**Figure 6**, MacLeod et al., 2006; Dächsel et al., 2010; reviewed by Gómez-Suaga et al., 2014). And fifthly, R1398H has the opposite effect to pathogenic mutations and variants located throughout LRRK2 in cellular assays of canonical Wnt activity (**Figure 5**; Berwick and Harvey, 2012). Taken together, we believe our data make a persuasive case in support of the genetic evidence that the R1398H variant is a genuine protective variant.

Curiously, structure-based alterations at the R1398 site have already been studied in assays of LRRK2 GTPase function and produced results that are consistent with our R1398H data. This is encouraging, since mutation of R1398 to any other amino acid can be expected to prevent hydrogen bonding to S1671. The best-studied structure-based LRRK2 R1398 mutant, R1398L, was designed from Ras GTPase homology models, with the expectation that it would behave similarly to the Q61L amino acid substitution that renders Ras proteins 'GTP-locked.' However, R1398L had the opposite effect. This change increased GTPase activity, both in full-length LRRK2 and in a deletion construct containing the Roc, COR and kinase domains of LRRK2 (RocCOR-kinase; Xiong et al., 2010; Stafa et al., 2012; Biosa et al., 2013). R1398L also weakened GTP-binding in RocCOR-kinase constructs, although no effect was seen in full-length LRRK2 (Xiong et al., 2010; Biosa et al., 2013). Although in disagreement with expectations, these data are clearly in accordance with our results for LRRK2 R1398H. An R1398Q mutant was also studied, with the rationale that this mutation would render LRRK2 more like wild-type Ras, but no statistically significant changes to GTP-binding or GTP hydrolysis were detected using full-length LRRK2 (Biosa et al., 2013). However, when R1398Q was introduced into a RocCOR-kinase construct alongside a second Ras-like amino acid substitution, T1343G, GTP-binding was decreased and GTP hydrolysis increased (Xiong et al., 2010).

As mentioned, PD-causing mutations outside the RocCOR tandem domain do not appear to operate via the same mechanism as those affecting GTPase function, i.e., they do not shift LRRK2 GTPase toward the GTP-bound 'on'-state. This is curious, since logic would dictate that the effects of all PD-causing mutations in LRRK2 should converge on the same process or processes eventually. By extension, a protective variant would be expected to affect the same process, but in the opposite direction. With this in mind, it is striking that the protective LRRK2 R1398H variant appears to enhance canonical Wnt signaling. This observation is opposite to our data for pathogenic mutations throughout the catalytic core of LRRK2 (Berwick and Harvey, 2012), and a PD risk variant in the C-terminal WD40 domain (**Figure 7**).

Our data add to the growing body of evidence indicating that deregulation of canonical Wnt signaling is involved in LRRK2 PD (Berwick and Harvey, 2014). Previous studies have revealed that LRRK2 interacts with DVL proteins that play a central role in all branches of Wnt signaling: (i) canonical; (ii) planar cell polarity (PCP); and (iii) Wnt/Ca2+-signaling (Sancho et al., 2009; Berwick and Harvey, 2012). LRRK2 also interacts with multiple components of the β-catenin destruction complex in vivo and associates with the Wnt co-receptor LRP6 at membranes. Importantly, expression of familial LRRK2 mutants results in decreased activation of Wnt/β-catenin signaling (Berwick and Harvey, 2012). Strikingly, in this study we show that the R1398H mutant has the opposite effect.

Our data also have implications beyond LRRK2 PD. Wnt signaling pathways have emerged as essential regulators of neuronal development and maintenance (Inestrosa and Arenas, 2010). Wnt ligands are known to activate signaling pathways that lead to remodeling of the cytoskeleton and promote neurite outgrowth via small GTPases (Inestrosa and Arenas, 2010). Deficiencies in Wnt signaling pathways have been shown to affect synaptic stability in the striatum (Arenas, 2014), whilst antagonism of canonical Wnt signaling in the substantia nigra promotes dopaminergic neurone death (L'Episcopo et al., 2014). Wnt signaling is also important in the interplay between the immune system (astrocytes, microglia) and neurones, and deregulation affects adult neurogenesis (SVZ plasticity) with age (L'Episcopo et al., 2014). Furthermore, both aging and neurotoxin exposure are reported to down-regulate canonical Wnt signaling in the adult midbrain, thereby increasing the vulnerability

of dopaminergic neurones. Taken together these observation provide strong evidence that Wnt signaling regulates multiple cell biological functions in the midbrain dopaminergic neurones that degenerate in PD, and that the age-related decrease in canonical Wnt activity may have a central role in the pathogenesis of sporadic forms of PD.

#### CONCLUSION

Our data support a model where pathogenic RocCOR mutants display altered intramolecular RocCOR interactions, leading to weakened RocCOR dimerization, increased GTP binding and decreased GTP hydrolysis. Together, these effects lead to a greater proportion of LRRK2 molecules existing in the GTP-bound 'on'-state. The protective R1398H Roc domain mutation also affects RocCOR interactions but with the opposite result, leading to more LRRK2 in the 'off'-state. As such, developing small molecules to decrease LRRK2 GTP-binding and/or stimulate LRRK2 GTPase activity seems a promising strategy for the development of PD modifying treatments and has already shown some encouraging results in LRRK2 GTP and kinase domain mutant PD models (Li et al., 2014, 2015).

#### AUTHOR CONTRIBUTIONS

JN-A, DB, CB, and KH designed the experiments; JN-A, DB, SG, and VS performed the experiments; JN-A, DB, SG, and KH analyzed the data; DB and KH wrote the paper. All authors were involved in revising the paper for important intellectual content, and gave final approval of the version to be published.

#### REFERENCES


#### FUNDING

This work was supported by The Wellcome Trust [WT088145AIA, WT095010MA to KH], the Medical Research Council [MR/M00676X/1 to KH] and a Vera Down British Medical Association Research Grant [to KH]. CB and JN-A were also supported by the Intramural Research Program of the NINDS, National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

#### ACKNOWLEDGMENTS

We thank Dr. Jean-Marc Taymans (KU Leuven, Leuven, Belgium) for guidance in performing GTPase assays, Dr. Victoria James (University College London, London, UK) for assistance with molecular modeling, Dr. Peng-Peng Zhu (National Institutes of Health, Bethesda, MD, USA) for instruction on culturing primary neurones and Emma Schul (University College London, London, UK) for use of the microplate reader for GTPase assays.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2016.00018



altered in familial Parkinson's disease R1441C/G mutants. J. Neurochem. 103, 238–247.



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

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

# A Recombinant Human Pluripotent Stem Cell Line Stably Expressing Halide-Sensitive YFP-I152L for GABAAR and GlyR-Targeted High-Throughput Drug Screening and Toxicity Testing

#### Katharina Kuenzel 1, 2, Oliver Friedrich1, 2 and Daniel F. Gilbert 1, 2 \*

<sup>1</sup> Department of Chemical and Biological Engineering, Institute of Medical Biotechnology, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, <sup>2</sup> Erlangen Graduate School in Advanced Optical Technologies, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany

GABAARs and GlyRs are considered attractive drug targets for therapeutic intervention and are also increasingly recognized in the context of in vitro neurotoxicity (NT) and developmental neurotoxicity (DNT) testing. However, systematic human-specific GABAAR and GlyR-targeted drug screening and toxicity testing is hampered due to lack of appropriate in vitro models that express native GABAARs and GlyRs. We have established a human pluripotent stem cell line (NT2) stably expressing YFP-I152L, a halide-sensitive variant of yellow fluorescent protein (YFP), allowing for fluorescence-based functional analysis of chloride channels. Upon stimulation with retinoic acid, NT2 cells undergo neuronal differentiation and allow pharmacological and toxicological evaluation of native GABAARs and GlyRs at different stages of brain maturation. We applied the cell line in concentration-response experiments with the neurotransmitters GABA and glycine as well as with the drugs strychnine, picrotoxin, fipronil, lindane, bicuculline, and zinc and demonstrate that the established in vitro model is applicable to GABAAR and GlyR-targeted pharmacological and toxicological profiling. We quantified the proportion of GABAAR and GlyR-sensitive cells, respectively, and identified percentages of approximately 20% each within the overall populations, rendering the cells a suitable model for systematic in vitro GABAAR and GlyR-targeted screening in the context of drug development and NT/DNT testing.

Keywords: YFP-I152L, glycine receptor chloride channel (GlyR), gamma-aminobutyric acid receptor type-A chloride channel (GABAAR), human pluripotent embryonal teratocarcinoma stem cells, NT2 cells, NT2-N cells

### INTRODUCTION

GABA type-A receptors (GABAAR) and strychnine-sensitive glycine receptors (GlyR) are ligandgated chloride ion channels that mediate inhibitory neurotransmission in the central nervous system (CNS). In adult neurons, GABAAR and GlyR ion channels conduct an inhibitory anion current, mainly carried by chloride (Cl−) upon activation by <sup>γ</sup>-aminobutyric acid (GABA) and the amino acid glycine, respectively. In embryonic neurons however, due to a higher intracellular

#### Edited by:

Robert J. Harvey, UCL School of Pharmacy, UK

#### Reviewed by:

Henry J. Waldvogel, University of Auckland, New Zealand Joe Lynch, The University of Queensland, Australia

> \*Correspondence: Daniel F. Gilbert daniel.gilbert@fau.de

Received: 10 May 2016 Accepted: 13 June 2016 Published: 28 June 2016

#### Citation:

Kuenzel K, Friedrich O and Gilbert DF (2016) A Recombinant Human Pluripotent Stem Cell Line Stably Expressing Halide-Sensitive YFP-I152L for GABAAR and GlyR-Targeted High-Throughput Drug Screening and Toxicity Testing. Front. Mol. Neurosci. 9:51. doi: 10.3389/fnmol.2016.00051 chloride concentration compared to adult neurons, receptor activation causes an outward directed, depolarizing and excitatory Cl<sup>−</sup> flux (Webb and Lynch, 2007). GABAARs and GlyRs are both members of the pentameric ligand-gated ion channel (pLGIC) family and require five subunits to form a single functional oligomer. For GABAARs there are 19 genes known (α1–6, β1–3, γ1–3, δ, ε, θ, π, and ρ1–3) exhibiting a broad range of heterogeneity and many hundreds of theoretically possible subunit combinations (Olsen and Sieghart, 2009). For GlyRs there are four genes known (α1–3, β) in humans, exhibiting far less diversity compared to GABAARs (Lynch, 2009). Each GABAAR or GlyR isoform as well as each subunit combination has a unique physiological and pharmacological profile. The subunit combination can change during development, in a tissue-specific manner or as a consequence of pathophysiological events (Lynch, 2004; Webb and Lynch, 2007; Esmaeili and Zaker, 2011; Rudolph and Möhler, 2014; Deidda et al., 2015). Genetic or molecular perturbation of the channels' function has been associated with severe neurological disorders including neuropathic pain (Lian et al., 2012; Xiong et al., 2012; Chen et al., 2014), chronic pain sensitization (Harvey et al., 2004; Zeilhofer, 2005; Lynch and Callister, 2006), hyperekplexia (Chung et al., 2010; Bode and Lynch, 2014), epilepsy (Meier et al., 2005; Eichler et al., 2008, 2009; Macdonald et al., 2010), fragile X mental retardation syndrome (D'Hulst et al., 2006), learning and memory deficits (Deidda et al., 2015) as well as neurodegeneration (Kang et al., 2015). In addition, GABAARs and GlyRs are increasingly acknowledged in the context of immunomodulation (Stoffels et al., 2011; Gunn et al., 2015), amyotrophic lateral sclerosis (Martin and Chang, 2012) and cancer (Neumann et al., 2004; Cuddapah and Sontheimer, 2011). Therefore, GABAARs and GlyRs, including individual isoforms as well as the various subunit combinations in their native neuronal environment are increasingly considered highly attractive drug targets for therapeutic intervention (Alexander et al., 2015). Due to their fundamental role in inhibitory neurotransmission GABAARs and GlyRs are increasingly recognized in the context of neurotoxicity (Suñol et al., 1989; Hall and Hall, 1999; Narahashi, 2002; Vale et al., 2003; Mohamed et al., 2004; Islam and Lynch, 2012) and in vitro neurotoxicity testing (NT) (Talwar et al., 2013; Tukker et al., 2016).

Although GABAARs and GlyRs play fundamental roles during brain development (Avila et al., 2013, 2014), these receptors have only sparsely been associated with developmental neurotoxicity and developmental neurotoxicity (DNT) testing. This is even more surprising as the incidence of neurological diseases including learning and developmental disorders has increased in recent years (May, 2000; Colborn, 2004; Rauh et al., 2006; Herbert, 2010). At the same time, the number and volume of worldwide registered and traded chemical substances has also increased. There is no doubt that developing brain is particularly vulnerable to damage by chemicals (Rice and Barone, 2000) and evaluation of chemicals for developmental neurotoxicity is critical to human health (Grandjean and Landrigan, 2006, 2014). However, only a very small number of chemicals has been tested for developmental toxicity in recent years (Middaugh et al., 2003; Makris et al., 2009), presumably because the current guidelines for DNT testing exclusively involve animal experiments (OECD, 1997, 2007) that are of poor reproducibility and predictive quality, low in throughput, prohibitively expensive and limited with regard to mechanistic insights into the toxicant's mode of action (Smirnova et al., 2014).

DNT testing is conducted for identification of chemicalinduced adverse changes in the structure and function of the developing central nervous system. At present, NT and DNT testing is only officially acknowledged by regulatory authorities when done with standardized in vivo animal test methods and when conducted according to guidelines provided by the Organization for Economic Cooperation and Development (OECD). For example, NT testing involves daily oral dosing of rats for acute, sub chronic or chronic assessments for 28 days, 90 days, 1 year or longer (OECD, 1997). Primary observations include behavioral assessments and evaluation of nervous system histopathology. DNT testing evaluates in utero and early postnatal effects by daily dosing of at least 60 pregnant rats from implantation through lactation. Offspring are evaluated for neurologic and behavioral abnormalities and brain weights and neuropathology are assessed at different times through adulthood (OECD, 2007). The type of exposure (single or repeated dose) and the outcome (lethal or nonlethal; immediate or delayed effects) will result in different classifications for substances under the Globally Harmonized System (GHS).

Since there are various methods available for toxicological profiling of GABAARs and GlyRs (Gilbert et al., 2009a,b,d; Talwar et al., 2013) these receptors can serve as valuable molecular targets for in vitro developmental neurotoxicity testing (DNT) and provide mechanistic insights into the neurotoxicants or developmental neurotoxicants mode of action.

However, systematic screening for potentiating or inhibiting modulators of GABAARs and GlyRs in the context of drug development and NT/DNT testing is hampered due to lack of appropriate in vitro models. Recombinant expression systems using e.g., human embryonal kidney-derived (HEK293) cells allow systematic large scale screening for GABAAR and GlyR modulators in high throughput format (Kruger et al., 2005; Gilbert et al., 2009a,b,d; Talwar et al., 2013; Walzik et al., 2015). Despite recombinant models being successful in the identification of GlyR chloride channel modulators (Balansa et al., 2010, 2013a,b), these systems lack of fundamental neuronal genetic programs and cell intrinsic regulators influencing the functional properties of mature neurons in vivo and are restricted to physiological, pharmacological and toxicological analysis of individual GABAARs and GlyRs isoforms in isolation. Modulators identified or investigated using recombinant expression systems have been reported to yield contradictive results comparing recombinant systems and native neurons. For example, NV-31 an analog of bilobalide, a major bioactive component of Ginkgo biloba herbal extracts, has been reported to inhibit recombinant GlyRs but to potentiate native hippocampal neuron GlyRs (Lynch and Chen, 2008). Hence, screening data generated using recombinant expression system may be only partially relevant to GABAARs and GlyRs expressed in vivo and always require time and resource intensive retesting using secondary and individual approaches. Terminally differentiated neuronal cells of human origin, e.g., primary cells from biopsy samples are rarely available, enable only a limited number of experiments and are typically derived from pathogenic tissue, rendering these cells unsuitable to systematic large-scale screening for modulators of GABAARs and GlyRs. Neuronal cells of animal origin such as mouse or rat are widely used for studying mammalian inhibitory neurotransmission in general and the physiological properties of GABAARs and GlyRs in particular but are not optimal for identification of human-specific therapeutic leads or pharmacological probes as well as for GABAAR and GlyR mediated neurotoxicity or developmental neurotoxicity as the physiology of animals may strongly differ from human physiology. Stem cells including induced pluripotent stem cell (iPSC) and pluripotent embryonal carcinoma cells provide an enormous potential for both GABAAR- and GlyR-targeted drug development and NT/DNT testing as they allow standardized high-throughput in vitro screening of a large number of chemicals in maturing and adult human neurons, that is time, cost and resource-effective.

Human pluripotent NTERA-2 (NT2 or TERA2.cl.SP12) stem cells are increasingly considered as a suitable model for in vitro NT and DNT studies (Couillard-Despres et al., 2008; Hill et al., 2008; Laurenza et al., 2013; Pallocca et al., 2013; Stern et al., 2014). Upon exposure to retinoic acid, the cells undergo neuronal differentiation, i.e., mimic the process of differentiation in the developing brain, and are potentially suitable to NT/DNT testing at different developmental stages ranging from non-differentiated stem cells, committed neural progenitors to differentiated neuronal, so called NT2-N cells, and glial cells (Lee and Andrews, 1986; Pleasure et al., 1992; Sandhu et al., 2002; Stewart et al., 2003; Ozdener, 2007; Coyne et al., 2011). Electrophysiological studies and extracellular recordings with NT2-N cells have demonstrated voltage-activated calcium, TTX-sensitive sodium and potassium currents, spontaneous synaptic currents as well as glutamate, N-methyl-D-aspartate (NMDA), GABA and strychnine-sensitive glycine-induced currents (Pleasure et al., 1992; Munir et al., 1996; Neelands et al., 1998; Gao et al., 2004; Coyne et al., 2011; Laurenza et al., 2013) demonstrating that these cells exhibit properties similar to those described in native human neurons thus, making them an excellent experimental model for both GABAAR- and GlyR-targeted drug development and NT/DNT testing. Also, a variety of neuronal markers has been reported to be expressed in differentiated NT2-N cells, including β-tubulin type III, MAP-2 and synapsin I (Pleasure et al., 1992; Stewart et al., 2003; Hsu et al., 2014; Saporta et al., 2014; Stern et al., 2014).

To address the limitations of conventional in vitro models for GABAAR- and GlyR-targeted drug, NT/DNT screening as well as of animal-based in vivo NT/DNT testing approaches described above, we aimed to establish a cell line stably expressing YFP-I152L under the control of the human ubiquitin promoter C. The promoter has been reported to drive selective protein expression in principal neurons in the mammalian brain (Wilhelm et al., 2011). NT2 cells have previously been reported to provide a suitable system for expressing exogenous proteins in terminally differentiated neurons (Pleasure et al., 1992).

YFP-I152L, an engineered variant of yellow fluorescent protein (YFP) with greatly enhanced anion sensitivity, is quenched by small anions and is thus suited to reporting anionic influx into cells (Galietta et al., 2001). The fluorescent protein has been successfully applied for structure-function analysis and compound screening with many different chloride channel types (Kruger et al., 2005; Gilbert et al., 2009a,b,d; Balansa et al., 2010, 2013a,b; Chung et al., 2010; Gebhardt et al., 2010; Talwar et al., 2013; Walzik et al., 2015).

We further aimed to apply the cell line in concentrationresponse experiments with GABA and glycine as well as with a selection of chemicals with known toxicity profiles on GABAARs and GlyRs and to compare GABA and glycine EC<sup>50</sup> and drug IC<sup>50</sup> values with published electrophysiological data and data from fluorescence based functional imaging.

To evaluate the suitability of the in vitro model to systematic large-scale functional screening in the context of GABAARand GlyR-targeted drug development and NT/DNT testing, we intended to quantify the proportion of GABAAR- and GlyRpositive cells.

Our in vitro model and methodological approach will be applicable within a framework of various individual strategies assessing NT/DNT at different stages during neuronal differentiation as well as to systematic large-scale in vitro neurophysiological, -pharmacological and -toxicological screening with GABAARs and GlyRs that is time, cost and resource-effective. Furthermore, in the context of in vitro-based experimental and analytical approaches for NT/DNT prediction, our methodology can contribute to reduce or even replace animal experiments and to further promote the concept of the 3Rs in biomedicine (Russell and Burch, 1959).

### RESULTS

We have established a recombinant human pluripotent stem cell line, stably expressing halide-sensitive YFP-I152L under the control of the human ubiquitin C promoter. The cell line allows fluorescence-based GABAAR and GlyR-targeted drug screening and in vitro NT/DNT testing at different stages of neuronal maturation. The workflow of cell line generation, stem cell differentiation and functional imaging is shown in **Figure 1**. Images of recombinant non-differentiated NT2-YFP-I152L stem cells and differentiated NT2-N-YFP-I152L cells as well as the principle of functional imaging using the cells are shown in **Figure 2**.

#### Functional Profiling of GABAARs and GLyRs in Recombinant NT2-N-YFP-I152l Cells

To assess whether the generated recombinant stem cell line is suitable to chloride imaging and functional profiling of GABAARs and GlyRs, NT2-YFP-I152L cells were differentiated into neuronal NT2-N- YFP-I152L cells as described in the Methods section and were seeded at defined density of 2 × 10<sup>4</sup> cells in each well of a 96-well plate. Two days later and approximately 1 h prior to fluorescence imaging, the culture medium was completely removed and was replaced by 50 µl NaCl control solution. The standard NaCl control solution contained (in mM): NaCl 140, KCl 5, CaCl<sup>2</sup> 2, MgCl<sup>2</sup> 1, HEPES 10, glucose 10, pH 7.4 using NaOH. The 96-well plate was placed onto the motorized stage of a high-content imaging system and cells were imaged in control solution to record cellular YFP fluorescence in unquenched state. Because YFP-I152L is almost insensitive to chloride, its fluorescence intensity is highest in NaCl solution allowing for optimal focussing into the optical layer of cells. Subsequently, cells were perfused with 100 µl NaI solution containing increasing concentrations of

and functional imaging. (A) Generation of NT2-YFPI52L cell line. NT2 stem cells were seeded into 60 mm culture dishes and transfected with ubiquitin-YFP-I152L-encoding vector using the calcium phosphate precipitation method. YFP-I152L-positive cells were selected over 5 weeks using the antibiotic geneticin. Subsequently, NT2-YFP-I152L cells were enriched by flow cytometry (FACS). (B) Neuronal differentiation of human pluripotent NT2 stem cells and preparation for functional imaging. Cells were differentiated based on 5-week retinoic acid (RA) treatment and NT2-N cells were enriched by a double-replating strategy. As a preparatory step for imaging of GABAAR and GlyR-function, cells were plated into 96-well plates. (C) Preparation of imaging experiments. Approximately 30 min prior to imaging experiments, the culture media was removed and replaced by control solution supplemented with pharmacological drugs. Functional imaging was conducted as shown in Figure 2.

GABA (0.01–100 µM, **Figures 3A,B**) or glycine (0.1–1000 µM, **Figures 3C,D**). The NaI test solution was similar to NaCl control solution except that the NaCl was replaced by equimolar NaI. Cells were imaged throughout the complete procedure of agonist perfusion for a total of 30 seconds and with an acquisition rate of 2 Hz. Fluorescence quench was calculated at single cell level by quantitative analysis as described in the Methods section. The principle of functional profiling of ligand-gated chloride channels is depicted in **Figures 2D–G**. Average time-courses of quench (mean ± SD, n = 10) following the addition of NaI plus the indicated GABA or glycine concentrations are shown in **Figures 3A,C**, respectively and were constructed by pooling results from wells exposed to different solutions with 10 cells per well. Average agonist dose–response curves constructed from the experiments shown in **Figures 3A,C** are shown in **Figures 3B,D**, respectively. Calculated half-maximal activation concentrations (EC50) for GABA (1.1 ± 0.2 µM) and glycine (4.3 ± 0.5 µM) are listed in **Table 1** and are overall smaller compared with data from functional Cl<sup>−</sup> imaging and electrophysiology previously reported in the literature (see Discussion). However, Hill coefficients (nH) for GABA (1.8 ± 0.4) and glycine (1.5 ± 0.2) correspond well with data from the literature. Although EC<sup>50</sup> values measured in this study are overall smaller compared to values from the literature, these data demonstrate that the cell line is suitable to chloride imaging and functional profiling of GABAARs and GlyRs.

#### Toxicological Profiling of GABAARs and GlyRs in Recombinant NT2-N-YFP-I152L Cells

To evaluate the suitability of the established cell line for systematic screening for GABAAR and GlyR modulators, we conducted concentration-response experiments with the chemicals strychnine, picrotoxin, fipronil, lindane, bicuculline, and zinc in combination with the neurotransmitters GABA or glycine. To this end, human pluripotent NT2-YFP-I152L cells were differentiated into neuronal NT2-N-YFP-I152L cells as described in the Methods section and were seeded at a density of 2 × 10<sup>4</sup> cells in each well of a 96-well plate. 48 h later, the culture medium was completely removed and was replaced by 50 µl NaCl control solution. The cells were imaged in control solution and during perfusion with 100 µl NaI solution containing 1µM GABA or 3µM glycine—basically representing half-maximal activation concentrations—and increasing concentrations of the drugs strychnine (100 pM–10 µM), picrotoxin (1 nM–100 µM), fipronil (1 nM–100 µM), lindane (1 nM–100 µM), bicuculline (1 nM–100 µM) and zinc (10 nM–1 mM). Average (mean ± SD,

TABLE 1 | Calculated half-maximal activation concentration and hill coefficients (EC50, nH) for the used neurotransmitters.


FIGURE 2 | Images of NT2 and NT2-N cells stably expressing YFP-I152L and principle of functional imaging. (A) Transmission light (left) and fluorescence image (right) of non-differentiated NT2-YFP-I152L cells after 1 day in vitro (DIV) indicating YFP-I152L-expression in most cells of the depicted population. Scale bar 100 µm. (B,C). Transmission light (left) and fluorescence images (right) of differentiated NT2-N-YFP-I152L cells after 35 days in vitro. Cell clusters (B) and neurites (C), highlighted by white arrows in the respective fluorescence micrographs, indicate successful differentiation of NT2-YFP-I152L cells into NT2-N-YFP-I152L cells and demonstrate that stable integration of the YFP-I152L-gene does not interfere with pluripotency of NT2 stem cells and their ability to undergo neuronal differentiation. Scale bar: 100 µm (B) and 30 µm (C). (D,E) Principle of fluorescence-based functional imaging of ligand-gated chloride channels (LGCC) such as GABAARs and GlyRs. (F) NT2-N-YFP-I152L cells in control solution (NaCl), captured in unquenched state as reflected in the schematic drawing shown in (D). (G) NT2-N-YFP-I152L cells in quenched state, recorded upon addition of test solution (NaI) supplemented with the GlyR agonist glycine and as reflected by (E). Scale bar: 50 µm.

n = 10) drug dose–responses are shown in **Figures 4**, **5**. Calculated half-maximal inhibition concentration (IC50) for the tested chemicals are summarized in **Table 2** and indicate that the recombinant cell line is suitable to GABAAR and GlyR-targeted toxicological profiling and identification of ion channel-specific drugs.

#### Quantification of the Proportion of GABAAR and GlyR Positive NT2-N-YFP-I152L Cells

The applicability of the established recombinant in vitro model for high-throughput GABAAR and GlyR-targeted screening in the context of lead identification and NT/DNT testing strongly depends on the proportion of GABAAR and GlyR positive NT2- N cells in the whole cell population. A very small percentage of GABAAR and GlyR expressing NT2-N cells requires a large number of individual experiments including technical and biological replicates to be conducted in order to obtain statistically sound results, potentially compromising the usability of the established cell line for the envisioned application. To obtain a robust estimate of the proportions of GABAAR and GlyR TABLE 2 | Calculated half-maximal inhibition concentration (IC50) for the tested chemicals indicating that the recombinant cell line is suitable to GABAAR and GlyR-targeted toxicological profiling and identification of ion channel-specific drugs.


positive NT2-N-YFP-I152L cells in the overall cell population, we calculated the percentage of cells per image indicating a fluorescence quench of at least 20% upon application of saturating GABA or glycine concentration. The threshold of 20% fluorescence quench was chosen because, from our experience, any measured cellular fluorescence change within a range of −20

to 20% reflects biological noise typical to the employed assay. As a starting point toward analysing the proportion of GABAAR and GlyR-positive cells we segmented the images of a total of 24 individual experiments from two different differentiation batches (GABAAR, batch #1: n = 7, batch #2: n = 6; GlyR, batch #1: n = 6, batch #2: n = 5) using a modified version of DetecTIFF© software (Gilbert et al., 2009c) employing two individual sets of parameters for the identification of small neuronal NT2-N-YFP-I152L and larger non-neuronal NT2- YFP-I152L cells. **Figure 6A** shows a representative fluorescence micrograph of large NT2-YFP-I152L and small NT2-N-YFP-I152L cells, highlighted with white and gray arrows, respectively. Exemplary segmentation masks created from the image shown in **Figure 6A** for quantitative analysis of fluorescence intensity and calculation of "% fluorescence quench" in small neuronal NT2- N-YFP-I152L cells and large non-neuronal cells are depicted in **Figures 6B,C**. The average cell size (in pixel, mean ± SD) of GABA or glycine-exposed small and large cells calculated from two independent differentiation batches is shown in **Figure 6D** and is as follows: GABA-exposed small cells: 177 ± 94 (batch #1, n = 1869) and 152 ± 89 pixel (batch #2, n = 2560), GABAexposed large cells: 1369 ± 760 (batch #1, n = 1216) and 1228 ± 709 pixel (batch #2, n = 1371); glycine-exposed small cells: 163 ± 93 (batch #1, n = 1631) and 158 ± 92 pixel (batch #2, n = 1071); glycine-exposed large cells: 1356 ± 757 (batch #1, n = 1212) and 1316 ± 752 pixel (batch #2, n = 828). In a next step, we calculated the percentage of cells per image indicating a fluorescence quench of at least 20% upon application of saturating GABA and glycine concentration (see histogram in **Figure 6E**) that is as follows: GABA-exposed small cells: 28 ± 13 (batch #1, n = 7) and 32 ± 12 % (batch #2, n = 6), GABA-exposed large cells: 3 ± 1 (batch #1, n = 7) and 6 ± 2% (batch #2, n = 6); glycine-exposed small

cells: 36 ± 14 (batch #1, n = 6) and 35 ± 7% (batch #2, n = 5); glycine-exposed large cells: 2 ± 1 (batch #1, n = 6) and 4 ± 2% (batch #2, n = 5). Approximately one third of the small cell population is represented by GABAAR and GlyR-positive cells, equaling 17 ± 8 (GABA, batch #1), 21 ± 9 (GABA, batch #2), 20 ± 9 (glycine, batch #1) and 19 ± 4% (glycine, batch #2) and rendering ∼20% of the overall population GABAAR and GlyRpositive cells. These data indicate a population of GABAAR and GlyR-positive cells that is stable between individual experiments and differentiation batches, highlighting the applicability of the established in vitro model for systematic GABAAR and GlyRtargeted high-throughput pharmacological and toxicological screening in the context of drug development and NT/DNT testing.

#### DISCUSSION

For the sake of feasibility, throughput and cost reasons, GABAAR and GlyR-targeted screening is typically conducted in nonneuronal recombinant expression systems assessing individual GABAARs and GlyRs isoforms in isolation. Such expression

systems usually also lack fundamental neuronal genetic programs and cell intrinsic regulators of mature neurons in vivo. The initial benefit is often compromised by subsequent time-consuming and cost-intensive re-screens for validation and specificity-evaluation of potential GABAAR and GlyR-modulators. In the context of neurotoxicity (NT) or developmental neurotoxicity (DNT) testing, GABAARs and GlyRs can serve as molecular targets and can contribute to deciphering the molecular mechanisms of toxic effects. However, so far there are no suitable in vitro models available which comprehensively reflect the complex situation of the central nervous system in vivo and NT/DNT testing is still solely based on animal experimentation that is morally questionable, low in throughput, expensive and of rather poor predictive quality. To address these issues, we have established a recombinant human pluripotent stem cell line, stably expressing YFP-I152L that is advantageous for GABAAR and GlyR-targeted screening in the context of lead identification and drug development as well as for NT/DNT testing for several reasons. First, the recombinant cell line allows functional profiling of GABAARs and GlyRs in neuronal environment, providing fundamental neuronal genetic programs and cell intrinsic regulators critical to reflecting the functional properties

of mature neurons in vivo. Second, the in vitro model is of human origin, allowing for GABAAR and GlyR-targeted screening relevant to human physiology, including identification of leads for therapeutic intervention and toxicity evaluation. Third, as the cell line used in this study mimics brain maturation to a certain extent, it allows analysis of the physiological properties of GABAAR and GlyR during development of the central nervous system. Fourth, the cell model has been applied over a large range of passages—up to passage number 50 until now—without noticeable loss in YFP-I152L fluorescence intensity, further highlighting its applicability to standardized in vitro NT/DNT testing. Finally, differentiation of NT2 cells can be conducted in large scale format allowing for systematic high-throughput experimentation.

We exposed the cell line to GABA or glycine and demonstrated that the in vitro model is suitable to functional imaging of both GABAARs and GlyRs. Despite the fact that Hill coefficients compare well with data from the literature, the calculated half-maximal activation concentration for GABA and glycine is overall smaller compared to EC<sup>50</sup> concentrations previously measured in NT2-N cells using whole cell patch clamp electrophysiology (Neelands et al., 1998; Gao et al., 2004; Gao and Greenfield, 2005; Coyne et al., 2011). Although it is difficult to isolate the reason for this without directly comparing concentration-response data from the same cells generated with different methodologies, the apparent high GABA and glycine sensitivities may have been caused by the relatively long incubation time in agonist solution, i.e., the time between application of the agonist and quantification of the fluorescence intensity (30 s). Furthermore, we have previously observed differences in glycine EC50, i.e., increased sensitivity for glycine, in recombinantly expressed α1, α2, and α3 GlyR, assessed with the fluorescence-based assay vs. patch clamp electrophysiology (Talwar et al., 2013).

We used the recombinant in vitro model in a case study with the drugs strychnine, picrotoxin, fipronil, lindane, bicuculline and zinc. The imaging data indicate that YFP-I152L-expressing NT2-N cells are applicable to GABAAR and GlyR-targeted toxicological profiling and identification of ion channel-specific drugs. The drug strychnine specifically inhibited glycine but not GABA-induced anion-influx with average IC<sup>50</sup> values that correspond very well with results measured by either electrophysiological or equilibrium [3H]-strychnine displacement studies using recombinantly expressed homomeric wildtype α1 GlyRs and variants (Lynch et al., 1997; Vafa et al., 1999). Also, concentration-response experiments with pictrotoxin, a standard pharmacological tool for identifying the presence of β-subunits in recombinant and native GlyRs (Lynch, 2004) revealed half-maximal inhibition concentrations and slope values for glycine and GABA-induced anion influxes, that correspond well with results measured by patch clamp electrophysiology in recombinantly expressed homomeric a2 GlyR (Wang et al., 2007) as well as in human pluripotent stem cell-derived neurons (James et al., 2014), respectively. The alkaloid is known to strongly inhibit GABAARs at low micromolar concentrations, whereas GlyRs in vivo are considered much less sensitive (Lynch, 2004). However, we observed a reversed sensitivity sequence with a picrotoxinsensitivity of GlyR that is approximately 5-fold higher compared to GABAARs. As it has been reported that αβ-heteromeric GlyRs were less sensitive to picrotoxin-inhibition than were α-homomeric GlyRs (Pribilla et al., 1992) we speculate that NT2-N-YFP-I152L cells predominantly express α-homomeric GlyR rather than αβ-heteromeric GlyRs, although this is no explanation for the reversed picrotoxin-sensitivities of GlyRs and GABAARs. Bicuculline, a purported selective antagonist of the GABA<sup>A</sup> receptor, inhibited the GABA-induced fluorescence quench with a half-maximal inhibition concentration similar to values revealed from rat primary neurons by whole cell electrophysiological recordings (Kumamoto and Murata, 1995). Bicuculline also inhibited glycine induced I<sup>−</sup> influx in NT2-N-YFP-I152L cells with about 10-fold lower sensitivity compared to GABA-dependent anion influx, highlighting its specificity to GABAAR on the one hand, but also confirming previous reports on its inhibitory action assessed in both recombinantly expressed (Sun and Machu, 2000) and native glycine receptors (Bhattarai et al., 2016) on the other hand. The IC<sup>50</sup> value for glycine-induced fluorescence quench revealed in presence of increasing concentration of the insecticide lindane is about one order of magnitude smaller compared to electrophysiological recordings from recombinantly expressed GlyRs (Islam and Lynch, 2012). These data suggest that NT2-N-YFP-I152L cells predominantly express homomeric GlyRs as lindane has proven to selectively inhibit homomeric but not heteromeric GlyRs (Islam and Lynch, 2012; Talwar et al., 2013). This conclusion is further supported by the above mentioned high picrotoxinsensitivity of GlyRs. Surprisingly, lindane showed no dosedependent inhibitory activity on GABA-induced anion influx, although lindane-sensitivity of GABAARs has been reported by many studies conducted in various culture models and cell types (Narahashi, 1996; Ogata et al., 1988; Vale et al., 2003). A study conducted by Belelli et al. (1999) demonstrated no effect on GABA-evoked currents mediated by anesthetic-insensitive wildtype GABA receptors composed of the (rho 1) subunit (Belelli et al., 1999), suggesting expression of GABA<sup>A</sup> rho 1 receptor in the examined NT2-N-YFP-I152L cells. Indeed there are a number of studies indicating profound changes in subunit expression and pharmacology during neuronal differentiation—similar to developmental changes in GABAAR occurring in neurons of the developing central nervous system—and demonstrating a large variety of neurotransmission phenotypes of nondifferentiated NT2 and differentiated NT2-N cell in vitro (Neelands et al., 1998, 1999; Guillemain et al., 2000). This question however was not further evaluated in the present study. Fipronil, a broad-spectrum insecticide, inhibited both GABA and glycine induced anion influx in NT2-N cells with sensitivities that correspond very well to results measured by patch clamp electrophysiology in recombinantly expressed GlyRs and GABAAR (Ratra et al., 2001; Li and Akk, 2008; Islam and Lynch, 2012). Zinc is widely recognized as a modulator of both, GABAAR and GlyR (Smart et al., 1994). For GlyRs, potentiation as well as inhibition of glycine-activated currents by low (<10 µM) and high concentrations (>100 µM) of zinc, respectively has been reported (Harvey et al., 1999). For GABAARs, inhibition has been shown at concentrations >10 µM (Kumamoto and Murata, 1995). Although we observed a minor increase of a glycine-induced fluorescence quench (Approximately 10%) in presence of 1 and 10 µM zinc, this effect is not significant. In contrast, zinc-dependent inhibition was observed at concentrations >100 µM for both GABA and glycine-induced anion influx, confirming previous reports on zinc-sensitive GABA-induced currents recorded from NT2-N cells (Gao et al., 2004). The zinc IC<sup>50</sup> could not be calculated due to a too small range of applied concentrations but presumably exceeds 1 mM, suggesting the presence of γ subunits which have shown to confer zinc insensitivity to αβ GABAARs (Smart et al., 1991).

We quantified the proportion of GABAAR and GlyR positive NT2-N-YFP-I152L cells based on images of the cells and we identified a percentage of approximately 20% of cells expressing GABAARs and GlyRs, respectively. Despite the fact that the proportion of glycine and GABA sensitive cells are comparable, it cannot be concluded from our data whether the cells are simultaneously sensitive to both glycine and GABA or respond to either of the neurotransmitters. Although the calculated proportions have proven stable between different batches of differentiation cultures, the total percentage of GABAAR and GlyR positive NT2-N-YFP-I152L cells is not optimal for plate reader, i.e., photomultiplier or cell populationbased screening, that typically requires a high number of reacting, e.g., GlyR or GABAAR-positive cells and allows experimentation in high-throughput screening mode. If desired or required, the proportion of NT2-N-YFP-I152L cells may be increased by size-exclusion filtering using e.g., a simple nylon filter of defined mesh size or even more sophisticated cell sorting approaches. As highlighted by gray and white arrows in the fluorescence micrograph depicted in **Figure 6A** and the histogram shown in **Figure 6D**, NT2-N cells are much smaller compared to non-neuronal cells, presumably facilitating sizedependent population-increase of NT2-N cells. Despite the fact that plate reader-based screening allows experimentation in high or even ultra-high-screening mode it is disadvantageous with regard to cellular heterogeneity. Cultures of NT2 cells have been reported to express a large variety of non-neuronal and neuronal phenotypes (Guillemain et al., 2000), as we have shown in the case study using drugs with known and partly GABAAR and GlyR-specific pharmacological and toxicological profiles. Besides their neuronal properties and human origin, the heterogeneity of NT2-N cell cultures is probably one of the most important advantages over homogeneous, e.g., recombinant expression systems, as it reflects the in vivo situation to a much higher extent than simplifying in vitro models. Thus, microscopy-based highcontent imaging of NT2-N cells, evaluating the physiological pharmacological and toxicological properties at single cell level, and in the context of cellular heterogeneity, is likely to be the optimal approach for systematic GABAAR and GlyR-targeted drug screening as well as NT/DNT testing.

To confirm the validity of the morphology or sizebased selection criterion, we conducted functional imaging experiments with non-treated NT2-YFP-I152L stem cells and quantified the average cell size as well as the percentage of stem cells, responding with at least 20% fluorescence quench upon exposure to saturating GABA or glycine concentration, respectively (see Supplementary Figure S1). The average cell size (pixel), calculated from two individual experiments each (see Supplementary Figure S1C) is comparable between experiments (GABA, replicate #1: 397 ± 254 pixel, n = 650 cells, replicate #2: 385 ± 252 pixel, n = 547 cells; glycine, replicate #1: 403 ± 243 pixel, n = 589, replicate #2: 400 ± 256 pixel, n = 654) and differs from the size of small and large cells, respectively, as indicated in **Figure 6D**. The percentage of nontreated NT2-YFP-I152L stem cells, responding with at least 20% fluorescence quench upon exposure to 1 mM saturating GABA (replicate #1: 0.62%, replicate #2: 0.37%) or glycine (replicate #1: 0.68%, replicate #2: 0.92%) concentration, respectively (see Supplementary Figure S1D), also differs considerably from data obtained with retinoic acid-exposed NT2-N-YFP-I152L cells. These data clearly demonstrate that the morphology, e.g., the size, as well as the functional properties, i.e., the sensitivity to GABA or glycine, vary between non-treated stem cells and NT2- YFP-I152L differentiation cultures containing small neuronal and large non-neuronal cells and confirming the validity of the employed size-based selection criterion.

We have previously published a method allowing for assessment of ligand-gated ion channels in the context of cellular heterogeneity that is based on progressive receptor activation and iterative fluorescence imaging (Talwar et al., 2013). This method could easily be adapted for application with the established in vitro model and has the power to deliver imaging data of unsurpassed functional content presumably outranging any other existing technique for assessing physiological properties of ion channels with regard to throughput.

Due to the fact that our method is based on microscopic evaluation using a single fluorescence indicator, it allows additional fluorescence or luminescence-based markers to be implemented for multiplexing, such as ion or pH-sensitive probes e.g., for parallel analysis of further ion channels or transporters, to assess the activity of intracellular, e.g., toxicity-relevant signaling pathways, or for mapping cellular morphology to functional phenotypes.

It is important to mention that the applicability of the cell line as well as the experimental approach depends on the proteins to be evaluated and targeted, on the individual experimental setup, the available instrumental infrastructure and the biological question to be assessed. While this work focuses on functional profiling and GABAAR and GlyR-targeted drug screening as well as neurotoxicity and developmental neurotoxicity, our in vitro model and methodological approach could also be adapted for other proteins and strategies, such as RNAi or combined compound and RNAi screening, for singleendpoint or time-resolved functional expression analysis or for approaches using overexpression libraries. Altogether, this work contributes to furthering the applicability of cell-based highthroughput functional screening and provides a means for largescale characterization of neuronal proteins in the context of in vitro-based NT/DNT prediction thus, promoting a systems-level understanding of human physiology in homeostasis and disease.

### METHODS

### Pharmacological Reagents

Glycine, GABA, strychnine, picrotoxin, fipronil, lindane, zinc chloride and bicuculline were obtained from Sigma. Stock solutions of glycine (1 M), GABA (1 M) and zinc chloride (1 M) were prepared in water, strychnine (10 mM), picrotoxin (100 mM), fipronil (30 mM), lindane (30 mM) and bicuculline (100 mM) were dissolved in dimethylsulphoxide (DMSO). All stocks were frozen at −20◦C. From these stocks, solutions for experiments were prepared on the day of recording.

### Reagents for Differentiation of NT2 Pluripotent Stem Cells

Retinoic acid and poly-D-lysine (PDL) were obtained from Sigma. Retinoic acid was prepared as 100 mM stock in DMSO. Uridine was prepared as 100 mM stock in water. These stocks were kept at −80 to −20◦C. PDL was prepared as 10x stocks in water and stored at 4◦C. Solutions for differentiation cultures were prepared freshly at the days of experimentation.

### Molecular Constructs

The regulatory region of the human ubiquitin C promoter (1204 bp) was amplified by PCR, the vector pFUGW (Addgene) was used as the template as well as specific primers (Invitrogen). The fragment was then inserted into the promoterless vector pDsRed2-1 (Clontech). PCR of the YFP-I152L gene (720 bp) was performed using vector pcDNA3.1-YFPI152L (Invitrogen) as a template and specific primers (Invitrogen). The YFP-I152Lcoding construct was kindly provided from Prof. Joe Lynch (Queensland Brain Institute, The University of Queensland, Brisbane, Australia). After removing the sequence of the red fluorescent protein from the vector phuUbC-DsRed2, YFP-I152L was sub-cloned into the vector. The expression of YFP-I152L is under the control of the constitutive active human ubiquitin C promoter, the plasmid carries the neomycin resistance gene for the selection with the antibiotic geneticin.

#### Cell Line

Human pluripotent teratocarcinoma NT2 cells (NTERA-2 cl.D1, CRL-1973TM) were purchased from The American Type Culture Collection (ATCC).

#### Generation of Stable Cell Line

For generating a cell line stably expressing YFP-I152L, NT2 stem cells were seeded into 60 mm culture dishes (TPP) at a density of 5 × 10<sup>5</sup> cells and were incubated at standard conditions over night. The next day, cells were transfected with ubiquitin-YFP-I152L-encoding vector using the calcium phosphate precipitation method. YFP-I152L-positive cells were selected over 5 weeks with 0.5 mg/ml geneticin (Roth, Germany). The geneticin concentration had previously been examined in a concentration-response experiment with 0.1–1.0 mg/ml geneticin by measuring the fluorescence intensity of Hoechst 33342 stained (1µM) cells exposed to geneticin for 48 h in 96-well plate format using a VICTOR X4 plate reader (Perkin Elmer). The kill curve for evaluation of the geneticin concentration is shown in **Figure 1A**, subpanel #2. Geneticinselected NT2-YFP-I152L cells with strongest fluorescence signal were enriched by flow cytometry (FACS, approximately top 30%, see **Figure 1A**, subpanel #3) in a single step. Enriched cells were maintained in geneticin-supplemented medium at the above mentioned concentration until initiation of the differentiation procedure. Recombinant cells were used over a large range of passages between passage number 15 and 50, without noticeable decline in YFP-I152L fluorescence intensity.

#### Cell Culture

Cells were maintained in DMEM (Invitrogen) supplemented with 10% fetal bovine serum (FBS, Biochrom) and penicillin (100 U/ml)/streptomycin (100 mg/ml) (Invitrogen) and were cultured in T75 flasks (TPP) at 37◦C, 5% CO2 in a humidified incubator according to standard procedures. Cells were passaged every 2–3 days and used in differentiation cultures when approximately 80–90% confluent.

#### Differentiation Culture

We have previously developed an optimized method for neuronal differentiation of human pluripotent NT2 stem cells in monolayer cultures. In brief, NT2 cells were seeded at a defined density into PDL coated 60 mm dishes (TPP) in standard culture medium supplemented with 10% FBS and penicillin (100 U/ml)/streptomycin (100 mg/ml) and 10 µM RA and were cultured for 5 weeks at 37◦C, 5% CO<sup>2</sup> in a humidified incubator. In the second week, 10 µM uridine was added to the media.

#### Preparation of Cells for Experiments

Following retinoic acid treatment, cells were replated into T25 flasks and cultured overnight at standard conditions. The next day, NT2-N-YFP-I152L cells located on top of non-differentiated cells were removed by tapping and were seeded at a density of 2 × 10<sup>4</sup> into the wells of a BD MatrigelTM matrix (BD) coated 96-well plate. Cells were used in functional imaging experiments 48 h later.

#### Preparation of Imaging Experiments

Approximately 1 h prior to commencement of experiments culture media in 96-well plates was removed manually and was replaced by 50µl standard control solution, which contained (in mM) NaCl 140, KCl 5, CaCl<sup>2</sup> 2, MgCl<sup>2</sup> 1, HEPES 10, and glucose 10 (pH 7.4, NaOH) upon washing the cells once in 50 µl standard control solution. The NaI test solution was similar in composition to NaCl control solution except the NaCl was replaced by equimolar NaI. For imaging experiments the agonist NaI test solution was supplemented with 1 mM glycine or GABA and was serially diluted with NaI test solution to obtain agonist solutions containing 0.01, 0.03, 0.1, 0.3, 1, 3, 10, 30, 100 µM final GABA or 0.1, 0.3, 1, 3, 10, 30, 100, 300 and 1000 µM final glycine concentration. For antagonist concentrationresponse experiments, the control solution was supplemented with 1 µM GABA or 3 µM glycine and increasing concentrations of either of the drugs strychnine (100 pM–10 µM), picrotoxin (1 nM–100 µM), fipronil (1 nM–100 µM), lindane (1 nM–100 µM), bicuculline (1 nM–100 µM) and zinc (10 nM–1 mM). All drugs were diluted from stocks at the day of the experiment and experiments were conducted at room temperature.

#### Imaging Infrastructure

The 96-well plate was placed onto the motorized stage of a highend long-term imaging system (Nikon Eclipse Ti, Nikon, Japan) and was imaged with a 10x objective (CFI Plan Fluor DL 10X Phase, N.A. 0.30, Nikon, Japan). Illumination from a xenon lamp (Lambda LS, Sutter Instruments, USA), passing through a filter block (C-FL Epi-FL FITC, EX 465-495, DM 505, BA 515-555, Olympus, Japan) was used to excite and detect YFP fluorescence signal. Fluorescence was imaged by a sCMOS camera (NEO, Andor, Ireland) and digitized to disk onto a personal computer (Dell Precision T3500, Dell, USA) with Windows 7 operating System (Microsoft Corporation, USA). The primary resolution of the camera was 2560 × 2160 pixel, although images were binned (2 × 2), resulting in a resolution of 1280 × 1080 pixel. Each image typically contained 350–700 cells. The CCD image acquisition rate was 2 Hz.

#### Imaging Experiments

The experimental protocol involved imaging each well for 30 s at 2 Hz acquisition rate capturing the initial fluorescence intensity in the control situation as well as the test situation upon receptor activation, respectively. Liquid handling was performed manually.

### Single Cell-Based Quantitative Image Analysis

Cells depicted in fluorescence micrographs were selected manually based on strongest quench for low and intermediate concentrations and based on cell morphology in experiments where high drug concentrations were applied. Registered images of fluorescent cells were segmented and quantitatively analyzed using NIS-Elements software (Nikon, Japan). The fluorescence signal of identified cells was measured in images acquired before and after the addition of agonist solution as the mean of all pixel values within the area of a cell. "% fluorescence quench" in chloride imaging experiments is defined as

```
%fluorescence quench =

              Finit − Ffinal
                             ∗ 100/Finit
```
where Finit and Ffinal are the initial and final values of fluorescence, respectively.

#### Calculation of Concentration-Response Relationships

Individual concentration responses were constructed by pooling results from ten cells in one well exposed to agonist solution. Concentration-response relationships were fitted with the following equation:

$$F = \, Fmax + \, \frac{Fmin - \, Fmax}{1 + ([Agonist]/EC50)^{nH}}$$

where F is the fluorescence corresponding to a particular agonist concentration, [Agonist]; Fmin and Fmax are the minimal and maximal fluorescence values, respectively; EC<sup>50</sup> is the concentration that elicits half-maximal activation; and nH is the Hill coefficient. Curve fits were performed using a least squares fitting routine (Origin 7G, OriginLab Corporation). All averaged results are expressed as mean ± SD.

### Quantification of the Proportion of GABAAR and GlyR-Positive Cells

Images of fluorescent cells were segmented and quantitatively analyzed using a modified version of DetecTIFF <sup>R</sup> software (Gilbert et al., 2009c). In brief, images were segmented using an iterative size and intensity-based thresholding algorithm and the fluorescence signal of identified cells was calculated as the mean of all pixel values within the area of a cell.

### REFERENCES


#### Data Analysis and Visualization

Imaging data were annotated in Microsoft Excel and analyzed using Origin 7G (OriginLab Corporation).

#### AUTHOR CONTRIBUTIONS

KK, OF, and DG conceived the project. KK conducted experiments. KK and DG analyzed the data. KK and DG wrote the paper. All authors commented and agreed on the manuscript.

#### ACKNOWLEDGMENTS

The authors gratefully acknowledge funding of the Staedtler Stiftung, Bavarian Equal Opportunities Sponsorship—Förderung von Frauen in Forschung und Lehre (FFL)—Promoting Equal Opportunities for Women in Research and Teaching and the Erlangen Graduate School in Advanced Optical Technologies (SAOT) by the German Research Foundation (DFG) in the framework of the German Excellence Initiative. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2016.00051


Insecticide Fipronil—a GABAA-Gated Chloride Channel Blocker. J. Toxicol. Clin. Toxicol. 42, 955–963. doi: 10.1081/clt-200041784


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

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

# Generation of Functional Inhibitory Synapses Incorporating Defined Combinations of GABA(A) or Glycine Receptor Subunits

#### *Christine L. Dixon1†, Yan Zhang1 and Joseph W. Lynch1,2\**

*<sup>1</sup> Queensland Brain Institute, University of Queensland, Brisbane, QLD, Australia, <sup>2</sup> School of Biomedical Sciences, University of Queensland, Brisbane, QLD, Australia*

Fast inhibitory neurotransmission in the brain is mediated by wide range of GABAA receptor (GABAAR) and glycine receptor (GlyR) isoforms, each with different physiological and pharmacological properties. Because multiple isoforms are expressed simultaneously in most neurons, it is difficult to define the properties of individual isoforms under synaptic stimulation conditions *in vivo*. Although recombinant expression systems permit the expression of individual isoforms in isolation, they require exogenous agonist application which cannot mimic the dynamic neurotransmitter profile characteristic of native synapses. We describe a neuron-HEK293 cell coculture technique for generating inhibitory synapses incorporating defined combinations of GABAAR or GlyR subunits. Primary neuronal cultures, prepared from embryonic rat cerebral cortex or spinal cord, are used to provide presynaptic GABAergic and glycinergic terminals, respectively. When the cultures are mature, HEK293 cells expressing the subunits of interest plus neuroligin 2A are plated onto the neurons, which rapidly form synapses onto HEK293 cells. Patch clamp electrophysiology is then used to analyze the physiological and pharmacological properties of the inhibitory postsynaptic currents mediated by the recombinant receptors. The method is suitable for investigating the kinetic properties or the effects of drugs on inhibitory postsynaptic currents mediated by defined GABAAR or GlyR isoforms of interest, the effects of hereditary disease mutations on the formation and function of both types of synapses, and synaptogenesis and synaptic clustering mechanisms. The entire cell preparation procedure takes 2–5 weeks.

Keywords: inhibitory postsynaptic current, IPSC, GABAergic, glycinergic, neuropharmacology, synaptogenesis, electrophysiology

#### INTRODUCTION

The central nervous system is comprised of circuits of interconnected neurons that serve to process specific types of information. These circuits regulate their own output by feedback and feedforward connections. Knowledge of the physiological properties of the inhibitory and excitatory synapses that mediate these connections is crucial for understanding the electrical behavior of circuits and ultimately of brain function. Fast inhibitory neurotransmission in these circuits is mediated by GABA type-A receptor (GABAAR) and glycine receptor (GlyR) chloride channels.

*Edited by:*

*Kirsten Harvey, University College London, UK*

#### *Reviewed by:*

*Jochen C. Meier, Technical University Braunschweig, Germany Piotr Bregestovski, Aix-Marseille University, France*

> *\*Correspondence: Joseph W. Lynch j.lynch@uq.edu.au*

*†Present address:*

*Christine L. Dixon, Institute of Neurology, University College London, London, UK*

*Received: 12 November 2015 Accepted: 07 December 2015 Published: 23 December 2015*

#### *Citation:*

*Dixon CL, Zhang Y and Lynch JW (2015) Generation of Functional Inhibitory Synapses Incorporating Defined Combinations of GABA(A) or Glycine Receptor Subunits. Front. Mol. Neurosci. 8:80. doi: 10.3389/fnmol.2015.00080*

GABAARs exhibit a particularly broad range of heterogeneity. As members of the pentameric ligand-gated ion channel (pLGIC) family, five subunits are required to form a single functional oligomer. There are 19 GABAAR genes (α1–6, β1–3, γ1–3, δ, ε, θ, π, and ρ1–3) with the most common synaptic isoform comprising α1, β2, and γ2 subunits in a 2:2:1 stoichiometry. Although many hundreds of other subunit combinations are theoretically possible, it is thought that around one hundred exist naturally in the brain (Olsen and Sieghart, 2009). GlyRs exhibit far less diversity with only four genes (α1–3, β) in humans (Lynch, 2009). They also belong to the pLGIC family and synaptic GlyR isoforms comprise heteromeric assemblies of α and β subunits in a 2:3 or 3:2 stoichiometry (Durisic et al., 2012; Yang et al., 2012).

Each GABAAR or GlyR isoform has a unique physiological and pharmacological profile and it is the unique properties of a particular isoform that are important for the appropriate functioning of a particular network. Disruptions to these properties can result in neurological disorders. For example, hereditary mutations that affect the function of GABAARs or GlyRs can lead to epilepsy (Macdonald et al., 2010) or human hyperekplexia (Bode and Lynch, 2014), respectively. Other disruptive mechanisms are also possible. For example, a post-transcriptional RNA editing mechanism that is upregulated in temporal lobe epilepsy increases the prevalence of α3 GlyRs incorporating the P185L mutation (Meier et al., 2005; Eichler et al., 2008, 2009). Finally, a range of neurological disorders is known to result from aberrant changes to pLGIC phosphorylation status (Talwar and Lynch, 2014). For example, chronic pain sensitization is caused by prostaglandin-induced phosphorylation of α3 GlyRs (Harvey et al., 2004; Zeilhofer, 2005; Lynch and Callister, 2006) and ethanol-induced phosphorylation of the γ2 GABAAR subunit contributes to alcoholism (Qi et al., 2007).

Thus, characterizing the physiological and pharmacological properties of defined GABAAR and GlyR isoforms under synaptic activation conditions is essential for understanding how neuronal circuits function in health and disease. However, it is difficult to study individual isoforms in their native neuronal environment due to the multitude of other isoforms present, and the difficulty in pharmacologically or genetically isolating the receptor isoform of interest. The neuron-HEK293 cell co-culture protocols we describe here solve this problem by providing a simple, efficient means of generating functional recombinant inhibitory synapses that selectively incorporate the recombinant GABAAR or GlyR isoform of interest.

Co-culture approaches have previously been developed to understand the roles of synaptic adhesion molecules (including neurexin and neuroligin) in the formation of glutamatergic or GABAergic synapses (Scheiffele et al., 2000; Biederer et al., 2002; Dean et al., 2003; Graf et al., 2004; Sara et al., 2005; Kim et al., 2006; Dong et al., 2007; Fuchs et al., 2013) or to investigate the impact of disease-causing neuroligin mutations on GABAergic synaptogenesis (Chubykin et al., 2005; Sun et al., 2011). They have also been employed to characterize the functional properties of inhibitory post-synaptic currents (IPSCs), and have revealed kinetic differences among different GABAAR isoforms (Wu et al., 2012; Dixon et al., 2014) and GlyR isoforms (Zhang et al., 2014).

The original co-culture protocol, involving postnatal hippocampal neurons and transfected HEK293 cells, was optimized for the immunohistochemical analysis of glutamatergic and GABAergic synapse development (Biederer and Scheiffele, 2007). A more recent protocol outlined an improved procedure for generating recombinant GABAergic synapses between striatal medium spiny GABAergic neurons and transfected HEK293 cells (Brown et al., 2014). However, this was also optimized for monitoring synapse development rather than for recording IPSCs in mature synapses. We have extended the co-culture approach in three ways. First, we describe the first spinal neuron-HEK293 cell co-culture preparation suitable for the efficient generation of recombinant glycinergic synapses. Second, we have simplified the technique for creating GABAergic synapses by using embryonic cortical neurons grown in serumfree media that does not promote the growth of glia (Brewer et al., 1993; Brewer, 1995). We have also have optimized the technique to facilitate the electrophysiological analysis of GABAergic and glycinergic IPSCs.

### MATERIALS AND METHODS

#### Overview

Protocols for all procedures described in this study are detailed in the Supplementary Information. Lists of reagents and equipment are also provided. An overview of the coculture procedure is presented in **Figure 1**. The cerebral cortex contains large populations of GABAergic interneurons that were used to provide presynaptic GABAergic terminals onto HEK293 cells that recombinantly express the GABAAR subunits of interest. Similarly, spinal neurons contain large populations of glycinergic interneurons that were used to provide glycinergic presynaptic terminals onto HEK293 cells expressing GlyR subunits of interest. The main steps in preparing the neuronal cultures are summarized in **Figure 1** (blue box). The steps involved in HEK293 cell transfection are also described in **Figure 1** (green box). An image of a GABAergic neuron forming synaptic contacts with HEK293 cell is shown in the inset.

#### Preparation of Neuron Cultures

Euthanasia of timed-pregnant rats was performed via CO2 inhalation, in accordance with procedures approved by the University of Queensland Animal Ethics Committee. To produce GABAergic interneuron cultures, E18 rat embryos were surgically removed from timed-pregnant rats and placed into chilled Ca-Mg-free Hank's Balanced Salt Solution (CMF-HBSS) under sterile conditions. The cortical neuronal tissue was then pinched off using fine forceps, taking care to peel away the meninges to keep glial cell numbers down. The dissected neurons were then triturated, centrifuged and resuspended in Dulbecco's Modified Eagles Medium supplemented with 10% fetal bovine serum (DMEM-FBS). To produce glycinergic interneuron cultures, E15 rat embryos were surgically removed and placed into ice cold CMF-HBSS under sterile conditions. The spinal cords were then removed and pinned at the wider proximal end while

meninges were carefully detached. The dissected neurons were then triturated, centrifuged and resuspended in DMEM-FBS.

In both cases, the cells were then counted and between 40,000 and 80,000 neurons were plated onto each 12 mm poly-Dlysine-coated coverslip in four-well plates. As previously noted, neuronal density is a key consideration: if it is too low it impairs neuronal survival and if it is too high it encourages neuron clumping (Fuchs et al., 2013). Neuronal cultures were always maintained in a 5% CO2 incubator at 37◦C. After 24 h the entire DMEM-FBS medium was replaced with Neurobasal medium including 2% B27 and 1% GlutaMAX supplements. A second (and final) feed 1 week later replaced half of this medium. In contrast to a previous protocol (Fuchs et al., 2013), we found that antibiotics were unnecessary. Neurons were used in coculture experiments between 3 and 6 weeks later (for GABAergic co-cultures) or 1–4 weeks later (for glycinergic co-cultures).

#### Culture and Transfection of HEK293 Cells

HEK293 cells were cultured in T75 flasks in DMEM-FBS and maintained in a 5% CO2 incubator at 37◦C. The cells were passaged weekly. Prior to transfection, they were trypsinized and plated onto 35 mm culture dishes at a density of 5000 cells/dish. Following overnight incubation, the cells were transfected via a calcium phosphate co-precipitation protocol, using a total of 0.5– 2.5 μg DNA per dish. Then, following incubation for 5–20 h in a 3% CO2 incubator, the transfection was terminated by washing twice with divalent cation-free phosphate buffered saline. The cells were then trypsinized, centrifuged, and resuspended in Neurobasal medium (including 2% B27 and 1% GlutaMAX supplements) then seeded onto the neurons. One 35 mm dish of HEK293 cells was sufficient to seed four coverslips of neurons. Once seeded with HEK293 cells, the co-cultures were returned to the incubator overnight to allow synapses to form. Cultures were used for patch clamp recording over the following 2–3 days.

## Plasmid DNA

A total of 0.5–2.5 μg plasmid DNA should be added to each 35 mm dish of HEK293 cells. This amount may vary according to the individual plasmid expression efficiency, the number and ratios of plasmids to be transfected and the transfection method. When using a calcium phosphate co-precipitation protocol, our recommendations are as follows. When expressing synaptic GlyRs, the total plasmid DNA should comprise: 0.2 μg neuroligin 2A, 0.2 μg gephyrin, 0.1 μg EGFP, with the remainder comprising GlyR subunit DNA that varies according to the number of subunits and the ratio of subunit DNA required. For example, when transfecting GlyR α subunits as homomers, add 0.5 μg DNA. When transfecting α and β GlyR subunits in 1:10 or 1:50 ratios add a total of 2 μg DNA. We recommend transfecting α1:β, α2:β, and α3:β subunits in 1:50, 1:50, and 1:10 ratios, respectively (Zhang et al., 2014). When expressing α1β2γ2 GABAARs, the α1, β2, γ2, EGFP and neuroligin 2A plasmid DNAs should be transfected in a 1:1:4:1:1 ratio with a combined total of 0.5 μg DNA.

In experiments described in **Figures 2** and **3** we employed plasmid DNAs encoding the human α1 (pCIS), rat α3L (pcDNA3.1), and human β (pcDNA3.1) GlyR subunits, plus mouse neuroligin 2A (pNice) and rat gephyrin (pCIS). In experiments described in **Figure 4**, we employed human α2 (pcDNA3.1), human α4 (pCIS), β2 (pcDNA3.1), and γ2L

(pcDNA3.1) GABAAR subunits. Site-directed mutagenesis was performed using the QuikChange mutagenesis kit (Agilent Technologies) according to manufacturers' instructions and the successful incorporation of mutations was confirmed by DNA sequencing.

### Patch Clamp Electrophysiology and Data Analysis

Standard patch-clamp electrophysiology equipment can be used, with the only specific requirement being a fluorescence microscope for identifying GFP fluorescent cells. Coverslips containing the co-cultured cells were placed gently into the recording chamber on the microscope stage and perfused continuously with an extracellular solution comprising (in mM): 140 NaCl, 5 KCl, 2 CaCl2, 1 MgCl2, 10 HEPES, and 10 D-glucose, adjusted to pH 7.4 with NaOH. Patch pipettes were filled with an intracellular solution containing (in mM): 145 CsCl, 2 CaCl2, 2 MgCl2, 10 HEPES, 10 EGTA, and 2 MgATP, adjusted to pH 7.4 with NaOH. HEK293 cell selection is largely a matter of trial and error. A good starting point is to select large, strongly fluorescent green cells that are closely surrounded by many neurons, especially small clumps of neurons. Cells with a textured (rather than smooth) appearance often yield abundant IPSCs.

The electrophysiological techniques may vary according to the experimental requirements. For example, if precise quantitation of rise times is required, it is extremely important that the filtering and digitisation rates are high and that pipette series resistance is low to avoid artefactually slowing down the event. In contrast, testing the effect of a drug on IPSC decay rate is less sensitive to filtering, and it may be necessary to use higher resistance pipettes to obtain a membrane seal that is stable enough to permit recordings that are long enough to apply and wash out the drug.

In all experiments described below, series resistance was compensated to 60% of maximum and was monitored throughout the recording. Spontaneous and action potentialevoked IPSCs in HEK293 cells were recorded at a holding potential −60 mV and currents were filtered at 4 kHz and sampled at 10 kHz. Only cells with a stable series resistance of <25 M throughout the recording period were included in the analysis. Patch pipettes (4–8 M resistance) were made from borosilicate glass (GC150F-7.5, Harvard Apparatus). Analyses of IPSC amplitude, 10–90% rise time, and decay time constant (single-exponential) were performed using Axograph × (Axograph Scientific). Single peak IPSCs with amplitudes of at least three times above the background noise were detected using a semi-automated sliding template. Each detected event was visually inspected and only wellseparated IPSCs with no inflections in the rising or decay phases (suggestive of superimposed events) were included. The respective parameters from all selected events from a single

cell were averaged and are presented as a single data point in **Figures 2–4**. The averages from multiple cells were then pooled to obtain group data. Statistical analysis and plotting were performed on group data with Prism 5 (GraphPad Software). All data are presented as mean ± SEM. One-way and two-way ANOVA were employed for multiple comparisons. For all tests, the number of asterisks corresponds to level of significance: ∗*p* < 0.05, ∗∗*p* < 0.01, ∗∗∗*p* < 0.001 and ∗∗∗∗*p* < 0.0001.

#### RESULTS

#### Glycinergic IPSCs

While we found that some co-cultures exhibited spontaneous activity in almost all green fluorescent HEK293 cells, it was more typical to observe spontaneous glycinergic IPSCs in around 20% of cells. This success rate is adequate for most experiments.

It is important to establish that the spontaneous IPSCs produced by the co-culture synapses exhibit similar characteristics to those mediated by native synapses incorporating the same subunits. **Figure 2A** shows sample IPSC recordings from HEK293 cells expressing α1β GlyRs. An IPSC averaged from >200 events recorded from multiple cells is shown in **Figure 2B**. Mean amplitudes, 10–90% rise times and decay time constants are presented in **Figure 2C**. Frequency distributions of IPSC amplitudes, 10–90% rise times and decay time constants all exhibit monotonic distributions suggesting a single functional population of synapses (**Figure 2C**). In adult hypoglossal motor neurons, where the α1β GlyR isoform predominates (Lynch, 2009), the 10–90% rise times and decay time constants range between 0.6–1.8 and 4.9–7.7 ms, respectively (Singer et al., 1998; Graham et al., 2006; Hirzel et al., 2006; Muller et al., 2006). The mean decay time constant (9.3 ms) and the 10–90% rise time (1.9 ms) recorded in our co-culture synapses correspond well with these results.

We performed a similar analysis on α3β co-culture synapses and found the mean IPSC rise and decay times to be remarkably similar to those mediated by α1β GlyRs (**Figures 3A–C**). These parameters were also distributed monotonically, again suggesting a single population of synapses (**Figure 3C**). Although α3βmediated IPSCs have yet to be recorded in isolation in native neurons, evidence to date suggests their rise and decay times are indistinguishable from those mediated by α1β GlyRs (Harvey et al., 2004). This fits well with the results from our engineered synapses.

The α1 GlyR subunit D80A and A52S mutations result in startle disease phenotypes in mice (Graham et al., 2006; Hirzel et al., 2006). We previously demonstrated that engineered synapses incorporating α1D80Aβ and α1A52Sβ GlyRs exhibited accelerated IPSC decay rates closely resembling those recorded in native synapses from mutant mice homozygous for these mutations (Zhang et al., 2014). This provides an important validation of our technique. In this study we sought to determine whether GlyRs may be located both synaptically

averaged from >100 events from a single cell, before and after the application of 10 mM ethanol or 1 μM diazepam. (D) The decay time constants of IPSCs mediated by α4β2γ2 GABAARs were significantly prolonged by 10 mM ethanol but not by 1 μM diazepam (left). In contrast, IPSCs mediated by α2β2γ2 GABAARs were significantly prolonged by 1 μM diazepam but not by 30 mM ethanol (right). Diazepam data for α2β2γ2 GABAARs were reproduced from (Dixon et al., 2014). ∗*p* < 0.05 relative to drug-free control in same cell.

and peri-synaptically in HEK293 cells by introducing mutations that dramatically enhanced the glycine sensitivity. We reasoned that if GlyRs are located peri-synaptically then enhancing their glycine sensitivity may render them susceptible to activation by synaptically released glycine, and if so this should be detectable as an additional slow rise time component (Wu et al., 2012). As noted above, α3P185L results from a post-transcriptional RNA editing mechanism that is upregulated in (and is causative of) human temporal lobe epilepsy (Eichler et al., 2008, 2009). This mutation reduces the glycine EC50 from 70.9 to 7.4 μM (Legendre et al., 2009). We also investigated the α1A288G mutation (which is not associated with a disease) because it reduces the glycine EC50 from 30.9 to 6.0 μM (Lynagh and Lynch, 2010).

As shown in **Figures 2A–C**, IPSCs mediated by α1A288Gβ GlyRs exhibited significantly slower rise times and decay time constants, although additional slow components were never observed on the rising phase of IPSCs. As with unmutated α1β GlyRs, these properties were monotonically distributed (**Figure 2C**) suggesting a single postsynaptic receptor population. Similarly, **Figures 3A–C** shows that IPSCs mediated by α3P185Lβ GlyRs also exhibited significantly slower rise times and decay time constants that were distributed monotonically. Thus, we were not able to unequivocally distinguish a putative perisynaptic GlyR population in either case.

#### GABAergic IPSCs

As with glycinergic IPSCs, we typically observed GABAergic IPSCs in around 20% of HEK293 cells. The rise times and decay time constants of IPSCs recorded from the dominant (α1β2γ2) synaptic subtype (1.2 and 4.0 ms, respectively) are in close accordance with those recorded from neurons known to predominantly express this subtype (Dixon et al., 2014). The co-culture system has revealed that other GABAAR subunit combinations can yield IPSCs with dramatically different rise and decay times (Wu et al., 2012; Dixon et al., 2014), although it is as yet unclear how these properties relate to those of the same isoforms when expressed in native neuronal synapses.

We have previously demonstrated that the effects of some clinically important drugs on co-culture GABAergic synaptic IPSCs are similar to those recorded at corresponding neuronal synapses. For example, 1 μM diazepam or 0.1 μM flunitrazepam significantly increased the decay time constants of IPSCs mediated by α2-containing GABAARs (Dixon et al., 2014) and 1 μM zolpidem or 1 μM eszopiclone increased IPSC magnitudes and decay time constants of IPSCs mediated by α1-containing GABAARs (Dixon et al., 2015). Here we extended this characterisation by performing a 'reciprocal' pharmacological comparison of α2β2γ2 and α4β2γ2 GABAARs, based the knowledge that α4-containing GABAARs are highly sensitive to ethanol and insensitive to benzodiazepines, whereas α2-containing GABAARs have the opposite profile (Knoflach et al., 1996; Wallner et al., 2006). As shown in **Figures 4A,B**, IPSCs mediated by recombinant α2β2γ2 and α4β2γ2 GABAARs exhibit identical amplitudes, 10–90% rise times and decay time constants. A physiologically relevant (10 mM) ethanol concentration significantly increased the IPSC decay time constant in α4β2γ2 GABAARs whereas 1 μM diazepam had no effect (**Figures 4C,D**). On the other hand, 1 μM diazepam significantly prolonged the IPSC decay time constant in α2β2γ2 GABAARs (Dixon et al., 2014), whereas even a very high (30 mM) ethanol concentration had no effect (**Figure 4D**).

### DISCUSSION

#### Applications of the Protocol

We have described protocols for reliably generating recombinant inhibitory synapses that incorporate defined GlyR or GABAAR isoforms of interest. These are suitable for investigating (1) the kinetics of IPSCs mediated by defined GABAAR or GlyR isoforms, (2) the effects of drugs on IPSCs mediated by defined GABAAR or GlyR isoforms, (3) the effect of posttranslational modifications (e.g., phosphorylation) and hereditary disease mutations on the formation and function of both types of synapses, and (4) synaptogenesis and synaptic clustering mechanisms in both types of synapses. We now expand on each of these points.

#### IPSC Kinetics

Inhibitory post-synaptic currents mediated by different synaptic GABAAR or GlyR isoforms exhibit unique physiological and pharmacological profiles. It is useful to quantitate these properties because they may help in identifying the presence, or even the role, of a particular isoform in a particular neuron and also because accurate parameters provide key inputs into computational models of neuron or network function. Although studying recombinant receptors in standard heterologous expression systems such as HEK293 cells or *Xenopus* oocytes allows the electrophysiological properties of a single isoform to be studied in isolation, this approach is limited because the neurotransmitter must be applied artificially and it cannot mimic the fast (μs) dynamic neurotransmitter concentration profile that exists in the synaptic cleft.

#### Investigating Drug Efficacy and Selectivity

The GABAAR is an established therapeutic target for clinical indications including epilepsy, anxiolysis, muscle spasms, sedation and anesthesia. GABAAR-targeted drugs currently in clinical use are not strongly subtype-selective and this can lead to dose-limiting side effects. For example, diazepam produces effective anxiolysis by positively modulating α2-containing GABAARs, although it also elicits the side effect of sedation by modulating α5-containing GABAARs (Trincavelli et al., 2012). Drugs specific for other isoforms are also being sought. For example, selective modulators of α5-containing GABAARs are being developed for a range of indications including stroke, cognitive impairment, and schizophrenia (Soh and Lynch, 2015). Although GlyRs are not currently targeted by clinically useful drugs, molecules that selectively enhance α3-containing GlyRs are considered promising as new generation treatments for chronic pain (Zeilhofer, 2005; Lynch and Callister, 2006). When evaluating new molecules as potential therapeutic lead compounds for synaptically localized receptors, it is important to test their potency, efficacy and subtype-selectivity under realistic synaptic activation conditions. The system we describe provides the most definitive means available of evaluating drug efficacy and selectivity at IPSCs mediated by defined receptor isoforms.

### Investigating Disease Mutations

Engineered synapses have yet to be used to study diseasecausing GABAAR mutations or modifications, and hence, the method has not realized its full potential as a model system for understanding the molecular pathology of neurological disorders. Mutations in GABAAR α1 and γ2 subunits have long been associated with genetic epilepsy syndromes (Macdonald et al., 2010). Thus far, the function and pharmacology of epilepsycausing mutant GABAARs have only been investigated using whole-cell recordings of steady-state GABA-activated currents in heterologous expression systems. The differing approaches that have been used to analyze the effects of these mutations have lead to controversy, particularly in the case of the γ2*R*43*<sup>Q</sup>* mutation (Petrou and Reid, 2012). Moreover, of all the identified epilepsycausing mutant GABAARs that exhibit partial or full expression at the cell membrane, there is an animal knock-in model of only one (Petrou and Reid, 2012). Transgenic animal models are afflicted by compensatory mechanisms that can obfuscate data, especially those involving ion channel genetic manipulations that affect GABAergic transmission (Harris et al., 2011). Because coculture α1β2γ2 GABAAR synapses successfully recapitulate the kinetics of neuronal IPSCs (Dixon et al., 2014), they may provide a promising means of investigating the synaptic signaling defects induced by hereditary epilepsy mutations to α1 and γ2 subunits.

Glycinergic co-culture synapses incorporating α1D80A<sup>β</sup> and α1A52S<sup>β</sup> GlyRs have been shown to exhibit accelerated IPSC decay rates that strongly resemble those recorded in native synapses from mutant mice homozygous for the same mutations (Zhang et al., 2014). This suggests that the co-culture system should be useful for modeling the effects of hyperekplexia mutations to α1 and β subunits (Bode and Lynch, 2014) and autism mutations to α2 subunits (Pilorge et al., 2015). Here we investigated the effect of the α3P185L mutation that is associated with temporal lobe epilepsy. The mutation resulted in slowing of the IPSC decay rate (**Figure 3**). However, glycinergic synapses are absent in the temporal lobe and it is thought that the affected receptors are located presynaptically at glutamatergic synapses. In this case, our results suggest the mutant GlyR would remain open for longer during each synaptic event, thus potentiating an excitatory Cl flux leading to enhanced excitatory neurotransmission that could underlie the disorder (Eichler et al., 2008, 2009; Legendre et al., 2009).

#### Synaptogenesis and Synaptic Clustering Mechanisms

Co-culture synapses have been used extensively to probe the roles of synaptic adhesion molecules in the formation of glutamatergic or GABAergic synapses (Scheiffele et al., 2000; Biederer et al., 2002; Dean et al., 2003; Graf et al., 2004; Sara et al., 2005; Kim et al., 2006; Dong et al., 2007; Fuchs et al., 2013) or to investigate the impact of disease-causing neuroligin mutations on GABAergic synaptogenesis (Chubykin et al., 2005; Sun et al., 2011). The strengths and weaknesses of co-cultures in this respect have recently been reviewed (Fuchs et al., 2013).

#### CONCLUSION

As the presynaptic terminals of our engineered synapses are provided by real neurons, their function is likely to resemble those of synapses *in vivo*. Indeed, serial electron microscopic reconstructions of GABAergic terminals onto HEK293 cells have confirmed that their ultrastructures are similar to those of native neurons (Fuchs et al., 2013). However, the postsynaptic specializations are less likely to resemble those of

#### REFERENCES

Biederer, T., Sara, Y., Mozhayeva, M., Atasoy, D., Liu, X., Kavalali, E. T., et al. (2002). SynCAM, a synaptic adhesion molecule that drives synapse assembly. *Science* 297, 1525–1531. doi: 10.1126/science.1072356

neurons given that HEK293 cells do not endogenously express all necessary postsynaptic clustering proteins at appropriate levels for synaptogenesis. In addition, some proteins that they do express may lack neuron-specific post-translational modifications required for correct synaptic function. These factors could ultimately alter the geometry of the synaptic cleft and the postsynaptic receptor clustering density, leading to non-physiological changes in the neurotransmitter concentration profile that could affect IPSC kinetics. This uncertainty is the main limitation of the technique. We have addressed this as far as possible by comparing the properties of engineered synapses with those of real synapses in cases where we can be reasonably sure about the synaptic subunit composition.

However, due to their non-physiological status, engineered synapses also offer opportunities to investigate new clustering mechanisms. If, for example, substitution of a particular pLGIC subunit results in a drastic, unexpected slowing of the IPSC rise time, it is possible that synaptic receptors have been de-clustered in a manner that may not occur in a neuron. This could in turn lead to the identification of novel clustering molecules and mechanisms. HEK293 cells are ideal for investigating such questions: they do not express all proteins necessary for synaptogenesis, but they do provide a high efficiency of transfection, faithful protein translation and a small, electronically compact shape appropriate for accurate quantitation of IPSC rise and decay times (Thomas and Smart, 2005).

#### AUTHOR CONTRIBUTIONS

CD, YZ, and JL conceived the project and developed the protocols; CD and YZ performed experiments and analyzed the data; and CD, YZ, and JL wrote the manuscript.

#### ACKNOWLEDGMENTS

This research was supported by project grants from the Australian Research Council (DP120104373) and the National Health and Medical Research Council (APP1062183). JL is supported by a Principle Research Fellowship from the National Health and Medical Research Council (APP1058542). We thank Dr Nela Durisic for critical review of the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2015.00080

Biederer, T., and Scheiffele, P. (2007). Mixed-culture assays for analyzing neuronal synapse formation. *Nat. Protoc.* 2, 670–676. doi: 10.1038/nprot.2007.92

Bode, A., and Lynch, J. W. (2014). The impact of human hyperekplexia mutations on glycine receptor structure and function. *Mol. Brain* 7, 2. doi: 10.1186/1756- 6606-7-2


**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 © 2015 Dixon, Zhang and Lynch. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

## The Intracellular Loop of the Glycine Receptor: It's not all about the Size

#### Georg Langlhofer and Carmen Villmann\*

Institute of Clinical Neurobiology, University of Würzburg, Würzburg, Germany

The family of Cys-loop receptors (CLRs) shares a high degree of homology and sequence identity. The overall structural elements are highly conserved with a large extracellular domain (ECD) harboring an α-helix and 10 β-sheets. Following the ECD, four transmembrane domains (TMD) are connected by intracellular and extracellular loop structures. Except the TM3–4 loop, their length comprises 7–14 residues. The TM3–4 loop forms the largest part of the intracellular domain (ICD) and exhibits the most variable region between all CLRs. The ICD is defined by the TM3–4 loop together with the TM1–2 loop preceding the ion channel pore. During the last decade, crystallization approaches were successful for some members of the CLR family. To allow crystallization, the intracellular loop was in most structures replaced by a short linker present in prokaryotic CLRs. Therefore, no structural information about the large TM3–4 loop of CLRs including the glycine receptors (GlyRs) is available except for some basic stretches close to TM3 and TM4. The intracellular loop has been intensively studied with regard to functional aspects including desensitization, modulation of channel physiology by pharmacological substances, posttranslational modifications, and motifs important for trafficking. Furthermore, the ICD interacts with scaffold proteins enabling inhibitory synapse formation. This review focuses on attempts to define structural and functional elements within the ICD of GlyRs discussed with the background of protein-protein interactions and functional channel formation in the absence of the TM3–4 loop.

#### Edited by:

Robert J. Harvey, University College London (UCL), UK

#### Reviewed by:

Verena Tretter, Medical University Vienna, Austria Raphael Lamprecht, University of Haifa, Israel Sarah Lummis, University of Cambridge, UK

> \*Correspondence: Carmen Villmann villmann\_c@ukw.de

Received: 13 March 2016 Accepted: 17 May 2016 Published: 03 June 2016

#### Citation:

Langlhofer G and Villmann C (2016) The Intracellular Loop of the Glycine Receptor: It's not all about the Size. Front. Mol. Neurosci. 9:41. doi: 10.3389/fnmol.2016.00041 Keywords: GlyR receptors, synaptic inhibition, intracellular domain, interaction partners, posttranslational modifications

### INTRODUCTION

Glycine receptors (GlyRs) are the major inhibitory neurotransmitter receptors in adult spinal cord and brainstem. They are important for motor coordination and respiratory rhythm. Disturbances in glycinergic neurotransmission by: (i) mutated genes encoding various GlyR subunits or adjacent proteins of the glycinergic receptor complex; (ii) receptor editing or; (iii) receptor modulation by posttranslational mechanisms lead to neuromotor deficits (hyperekplexia), pain sensitization and autism spectrum disorders (Lynch, 2004; Schaefer et al., 2013; Bode and Lynch, 2014; Pilorge et al., 2015).

**Abbreviations:** CLRs, Cys-loop receptors; ECD, extracellular domain; ICD, intracellular domain; TM, transmembrane; GlyR, glycine receptor; wt, wild-type.

GlyRs are members of the superfamily of Cys-loop receptors (CLRs) such as nicotinic acetylcholine receptors (nAChR), 5HT<sup>3</sup> receptors, and GABAA/<sup>C</sup> receptors. They all share a common disulfide bridge in the extracellular N-terminal domain between conserved cysteine residues. GlyRs are pentameric receptors composed of 2α and 3β subunits (Grudzinska et al., 2005). Four different α subunits and one β subunit are known. Functional diversity is enhanced by alternative splicing processes, which has been described for all subunits (Kuhse et al., 1991; Malosio et al., 1991; Nikolic et al., 1998; Oertel et al., 2007; Hirata et al., 2013).

Most of the knowledge about GlyR signal processing comes from in vitro mutagenesis studies on structure-function relationships. Recently the x-ray structure of GlyRα3 and the cryo-electron microscopic structure of α1 were solved (Du et al., 2015; Huang et al., 2015). These structures provided deeper insights into the mechanisms of signal processing and gating. Interestingly, x-ray crystallography of CLR members was only possible when the large intracellular loop between TM3–4 was replaced by a short peptide. The TM3–4 loop harbors the highest variability among all CLRs in terms of length and sequence variations. These loop structures mediate subfamilyspecific interactions with intracellular binding partners (Goyal et al., 2011). In GlyRs, the TM3–4 loops interact with the scaffold protein gephyrin important for synaptic anchoring or signal transduction processes. In addition, the TM3–4 loop is modified by posttranslational modifications and binds allosteric modulators that in turn influence functional ion channel properties (**Figures 1A–D;** Ruiz-Gómez et al., 1991; Kirsch and Betz, 1995; Yevenes et al., 2008; Yevenes and Zeilhofer, 2011). Subdomains of the GlyR TM3–4 loop have been demonstrated to be important for receptor trafficking to the cellular membrane and the nucleus (Sadtler et al., 2003; Melzer et al., 2010).

#### IMPORTANCE OF GLYCINE RECEPTORS FOR INHIBITORY NEUROTRANSMISSION

In the nerve muscle circuit, GlyRs control excited motoneurons in spinal cord and brainstem. Motoneuron activation is enabled by released glutamate from dorsal root ganglia. In turn, activated motoneurons fire action potentials towards the neuromuscular endplate where the signal is transmitted via acetylcholine to propagate along muscle fibers resulting in muscle contraction. To balance motoneuron firing, inhibitory GlyRs localized within the motoneuronal membrane are activated by release of glycine from neighboring interneurons. These interneurons are excited by collateral axons of the motoneurons. As a consequence, motoneurons are hyperpolarized and excitation is dampened. This feedback control by GlyRs restores the balance between excitation and inhibition (Schaefer et al., 2012). Using similar mechanisms, GlyRs mediate respiratory rhythms in PreBöt (pre-Bötzinger complex) nuclei of the brainstem (Winter et al., 2009; Janczewski et al., 2013). An impaired glycinergic inhibition in the brainstem of the mouse mutant oscillator leads to decreased breathing frequency caused by prolongation of expiratory duration. This results in death of affected mice around postnatal day 21 due to respiratory acidosis (Markstahler et al., 2002). Minor GlyR expression has been determined in the retina, inner ear, and the hippocampus (Harvey et al., 2004; Heinze et al., 2007; Dlugaiczyk et al., 2008; Lynch, 2009; Aroeira et al., 2011).

In the hippocampus, GlyRs are mainly found at extrasynaptic sites pointing to a function in tonic activation processes (Aroeira et al., 2011). These extrasynaptic receptors are formed by homomeric α2 and α3 GlyR subunits. A gain of function GlyRα3 variant (α3 P185L) was previously identified in human hippocampectomies from patients with temporal lobe epilepsy (Meier et al., 2005; Eichler et al., 2008). Additionally, the hippocampus of patients with epilepsy expresses predominantly the long splice isoform of α3 (α3L; Eichler et al., 2009). Both findings were used to generate a mouse model with neuron-type specific expressions of the GlyRα3LP185L to study homeostatic effects that control synaptic neurotransmission. The estimated presynaptic expression of GlyRα3 P185L in glutamatergic terminals facilitated neurotransmitter release (Winkelmann et al., 2014). As a consequence, enhanced hyperexcitability leads to recurrent epileptoform discharge impairing cognitive function and discriminative associative memory (Winkelmann et al., 2014). Changes in cognitive function and discriminative associative memory have been analyzed with the reward-based 8-arm radial maze test that discriminates between working memory (number of entries into an arm that was never baited) and reference memory (re-entries into an arm visited in the ongoing trail).

In contrast, specific expression of GlyRα3LP185L in parvalbumin-positive interneurons generated hypoexcitability and triggered anxiety-like behavior (Winkelmann et al., 2014). Increased anxiety of GlyRα3LP185L mice was verified by a preference for the dark using the dark/light test, decreased entries into the center in an open field, and less time spent and decreased numbers of entries into the open arms using the elevated plus maze test (Winkelmann et al., 2014). In conclusion, increased presynaptic function represents a pathogenic mechanism able to alter neural network homeostasis and thereby control neuronal network excitability and trigger neuropsychiatric symptoms (Winkelmann et al., 2014).

Inhibition of postsynaptic GlyRα3 by PGE2- (prostagladin E2) induced phosphorylation underlies central inflammatory pain sensitization. This process depends on the activation of protein kinase A that phosphorylates α3 at residue S346 localized in the TM3–4 loop (Harvey et al., 2004). These findings initiated a series of pharmacological studies with GlyRα3 as a promising target in pain therapy (Lynch and Callister, 2006).

The involvement of GlyRs in autism spectrum disorders is based on genetic findings and knockout mice although the molecular mechanisms behind their involvement in the excitation/inhibition imbalances are not completely understood (Tabuchi et al., 2007; Pilorge et al., 2015). The analysis of a rare human X-linked GLRA2 microdeletion (deletion of exons 8 and 9 that refer to the TM3–4 loop) associated with autism exhibited lack of surface GlyR expression in vitro and severe axon-branching defects in zebrafish (Pilorge et al., 2015). A knockout of Glra2 in mice revealed deficits in object recognition memory and impaired long-term potentiation in the prefrontal

(+; left): loop A (green), loop B (yellow, with F159), loop C (red, with Y202 and F207); complementary subunit (−; right): loop D (blue), loop E (magenta), loop F (brown). (C) TM3–4 loop sequences of the human GlyRα1 (residues 309–400) and human α3 (last line) are shown. Constant (C) and variable (V) regions of the TM3–4 loop are marked. Bold black letters—all residues that have been functionally investigated in vitro (structurally and functionally important residues for ion permeation and desensitization, residues that bind intracellular proteins, residues involved in receptor trafficking and TM3 integration, residues that bind drugs and Gβγ proteins, posttranslational modifications, residues affected in human patients). (D) TM3–4 loop sequences of the human α1 (α1ins) and α3 splice variants (long-α3L and short-α3K) are shown. Splice inserts are marked with black bold letters. In the GlyRβ TM3–4 loop sequence binding sites for gephyrin (underlined) and syndapin are marked. Note, the β TM3–4 loop is longer (residues 327–453) compared to α1 (309–400) and α3 (α3L 309–400 and α3K 309–385).

cortex. In summary, these data provide evidence for a link of altered glycinergic inhibition to social and cognitive impairments (Pilorge et al., 2015).

The role of GlyRs detected in non-neuronal tissues, e.g., immune cells, endothelial cells, hepatocytes, renal cells is not completely understood but argues for other functions than a neuronal ion channel (Van den Eynden et al., 2009).

### HUMAN AND MURINE MUTATIONS FOUND IN GlyRα1 INTRACELLULAR DOMAIN (ICD)

GlyR mutations can result in the neuromotor disorder hyperekplexia. The most common cause for hyperekplexia are mutations in the GLRA1 gene which was mapped to the disease in 1993 (Shiang et al., 1993). The second most common cause for hyperekplexia results from mutations in the SLC6A5 gene encoding the presynaptic glycine transporter 2 (GlyT2; Rees et al., 2006). Mutant GlyT2 variants represent the presynaptic component of the disease. Rare forms of the disease are generated by mutations in genes encoding other postsynaptic proteins of the inhibitory synapse, e.g., gephyrin and collybistin (CB).

GlyRα1 mutations are distributed over the entire sequence. Among these, most of the dominant inherited mutations are localized in the ion channel domain (TM2) and adjacent loop structures. These mutants are accompanied by functional deficits such as lower maximal currents, reduced single channel conductance, enhanced desensitization or decreased ligandbinding efficacy (Saul et al., 1999; Becker et al., 2008; Chung et al., 2010). In contrast, recessive mutants influence receptor biogenesis, trafficking, and receptor stability (Villmann et al., 2009b; Schaefer et al., 2015).

So far, only five human mutations, R316X, G342S, E375X, D388A, and R392H have been identified in the GlyRα1 TM3–4 loop (**Figure 1C**). Three of them (R316X, D388A, R392H) are compound heterozygous. Compound heterozygosity refers to two recessive alleles (W68C/R316X, L291P/D388A, and R252H/R392H) that result in hyperekplexia in a heterozygous state (Vergouwe et al., 1999; Rees et al., 2001; Tsai et al., 2004; Chung et al., 2010; Bode and Lynch, 2013). In vitro studies on R392H revealed decreased inward currents, reduced expression and less stability as the underlying pathological mechanism. These effects were more pronounced when R392H was coexpressed with R252H. Receptors composed of R252H and R392H were non-functional, arguing for a dominant effect of R252H localized in close proximity to the ion channel pore (Villmann et al., 2009b).

GlyRα1 variants R316X and E375X lead to truncated α1 subunits. Truncations of receptor proteins result in significantly decreased surface expression due to protein misfolding and abnormal receptor trafficking (Villmann et al., 2009a; Kang et al., 2015; Schaefer et al., 2015). As a consequence, insufficient receptor densities lead to deficiency of functional ion channels.

A similar TM3–4 loop truncation of the closely related GABAAR γ2 subunit is associated with generalized epilepsy with febrile seizures plus (GEFS+; Kang et al., 2015).

An in vitro analysis of α1 E375X revealed no surface expression of the truncated α1 protein when expressed alone to form homomeric receptor complexes. Coexpression of α1E375X with wild-type (wt) α1 or α1β led to functional ion channel formation. The observed current amplitudes were smaller and EC<sup>50</sup> values were increased for GlyRs formed by α1wt/α1E375X/β in comparison to homomeric α1 and heteromeric α1β wt (**Figure 2A**). This simulation of the in vivo configuration constitutes the potential of E375X to integrate into pentamers, its transport to the cell surface and finally its impact on GlyR function (Bode and Lynch, 2013). Similar effects have been observed for the GlyRα1 ICD variant D388A. Mutant α1D388A receptors were only recruited to the cellular membrane in presence of either α1 or α1β wt (Bode et al., 2013).

R316X showed impaired trafficking with a small fraction of mutated GlyRs expressed at the cellular surface but insufficient to generate functional ion channels (Schaefer et al., 2015).

A TM3-4 loop truncation in the mouse mutant oscillator results in absence of truncated protein from the organism. Oscillator carries a 7 bp deletion and depending on the use of an alternative splice acceptor site generates two different transcripts although neither is translated into α1 protein in vivo (Kling et al., 1997). Lack of translation of both transcripts induces severe neuromotor deficits in homozygous oscillator mice starting at postnatal day 14. These deficits increase progressively until death at postnatal day 21. During this period GlyRs undergo a subunit switch from homomeric α2 (embryonic isoform) to heteromeric adult GlyRs (α1β, α3β). Obviously, there is no compensation by other GlyRs to the lack of functional α1β receptors in homozygous oscillator mice (Buckwalter et al., 1994; Kling et al., 1997). Thus, oscillator represents a GlyR NULL mutation.

An in vitro coexpression of the truncated oscillator GlyRα1 protein (spdot-trc) together with a complementary truncated wild-type α1 construct (harboring most of the TM3–4 loop sequence, TM4, and the C-terminus = myc-α1-iD-TM4; **Figure 2B**) restored surface expression of both GlyR domains arguing for lack of precise quality control in the overexpression system (Villmann et al., 2009a). The coexpression of the nonfunctional truncated GlyRα1 isoform (spdot-trc) together with the lacking protein portion (myc-α1-iD-TM4) on a separate plasmid in the same cell regenerated ion channel functionality (GlyRα1 rescue = functional complementation of an ion channel from for themselves non-functional ion channel domains). These findings suggest that GlyRs are composed of independent folding domains able to interact with each other to complement channel functionality (**Figure 2B**; Villmann et al., 2009a). Using similar GlyR N- and C-terminal domains, it was further shown that non-functionality of truncated GlyRs lacking the TM3–4 loop, TM4 and the C-terminus is due to the inability to form pentameric receptor complexes (**Figure 2C**; Haeger et al., 2010).

How do these independent folding domains interact? An interaction between differently charged residues was analyzed by stepwise truncation of the complementation construct from its N- to the C-terminus. A lack of more than 55 residues from the TM3–4 loop resulted in non-functionality. Interestingly, the coexpression of three GlyR domains regenerated functionality at least to some extent further supporting the finding for independent folding domains of the GlyR (Unterer et al., 2012).

An application of the domain complementation approach to truncated human variants yielded similar results. The human α1 variant R316X was coexpressed with a corresponding C-terminal complementation construct (iD-TM4-C). The functional restoration of the respective GlyRs achieved 20% of ion channel efficacy compared to the wild-type situation. R316X was identified in a patient concomitant to W68C. The mutant W68C significantly decreased receptor trafficking to the cellular surface. A coexpression of W68C, the complementation construct, and R316X generated functional ion channels indistinguishable from GlyRs lacking W68C (**Figure 2D**). Therefore, it was concluded that the mutant W68C in the extracellular domain (ECD) does not hinder R316X from forward trafficking and integration into the pentameric arrangement (Schaefer et al., 2015).

Hence, GlyRs are able to assemble from independent folding domains and generate functional ion channels. This process does not require the integrity of the GlyR ICD rather subdomain interactions may mediate the efficacy of GlyR ion channel functionality.

In addition to the TM3–4 loop, the ICD also comprises the short intracellular loop connecting TM1 and TM2. The role of the TM1–2 loop in hyperekplexia has been defined by functional studies of the mutant P250T (Saul et al., 1999). Residue P250 is localized in very close proximity to the inner vestibule of the ion channel. The introduction of a threonine at position 250 leads to fast-desensitizing receptors with decreased glycine sensitivity. A mutagenesis series of residue 250 determined side volume and

hydropathy as important mediators in the pathology underlying P250T (Breitinger et al., 2001).

#### GLYCINE RECEPTOR STRUCTURE

Since 2011, the x-ray structures of several CLR members have been solved. These structures together with electron cryo-microscopy structures revolutionized our current knowledge about conformational rearrangements of the ion channel in the presence of agonists and antagonists leading to open and closed channel conformations (Unwin, 2005; Hassaine et al., 2014; Miller and Aricescu, 2014; Du et al., 2015). A closer view onto the CLR structure revealed an architecture of two domains: the ECD able to bind the ligand and the transmembrane domain (TMD) encompassing four α-helical transmembrane segments, connected by intraor extracellular loop structures (**Figure 1A**). The crystal structures of the large intracellular loops of the GABA<sup>A</sup> receptors, the 5HT<sup>3</sup> receptors, and the GlyRs between transmembrane segments 3 and 4 have not been solved yet most probably due to hindrance of crystal formation when present.

The recently solved structures of GlyRα1 and GlyRα3 provided novel insights into GlyR functioning. Conformational rearrangements involve specific loop structures of the ECD as well as the ECD-TMD interface. These rearrangements enable ion channel gating as a consequence of an anti-clockwise outward rotation of TMD during opening of the ion channel pore. A prerequisite for glycinergic signal transduction is agonistbinding to the ligand-binding pocket formed by residues of loops A-F (**Figures 1A,B**). Ligand-binding is stabilized by aromatic residues e.g., F159, Y202, F207 within the pocket. Following binding, the signal is transmitted via extensive interactions near the ECD-TMD interface including the β1–2 loop, the Cys loop, and the M2–M3 loop at the principal side of the ligand-binding interface with loops β1–2, β8–9 and pre-M1/M1 of the complementary side (-) of the pocket (Du et al., 2015). Due to flexibility of loops C and β8–9, these loop structures initiate the rearrangement of the conformation from the open into the closed form by a backward movement involving the same loop structures and domains (Du et al., 2015). From crystallographic analysis there are so far no hints for an involvement of the intracellular loop between TM3–4 in signal transduction processes due to lack of its presence in constructs used for x-ray crystallography. Voltageclamp fluorometry experiments however provided evidence for the participation of the TM3–4 loop structure in the rearrangement of M3 and M4 during ion channel opening. In this context it was demonstrated that M3 and M4 undergo large transitions compared to M1 and M2 movements (Han et al., 2013a).

#### STRUCTURAL DETERMINANTS OF THE GlyR ICD

In contrast to eukaryotic CLRs (nAChRs, GABAA/CRs, GlyRs, and the 5HT<sup>3</sup> receptors), the prokaryotic CLR-homologs ELIC (Erwinia chrysanthemi ligand-gated ion channel) and GLIC (Gloeobacter violaceus ligand-gated ion channel) carry very short intracellular loop structures (Hilf and Dutzler, 2008; Nury et al., 2011).

Chimeric CLRs (5HT3A-GLIC, GlyR-GLIC) harboring mainly the short heptapeptide SQPARAA (TM3–4 loop of GLIC) instead of their receptor-specific TM3–4 loop were able to form functional ion channels, which differ in single channel conductances and desensitization compared to wildtype receptors. Their overall properties, such as ion selectivity, efficiency of ligand-binding and current amplitudes were unaffected (Jansen et al., 2008; Papke and Grosman, 2014; Moraga-Cid et al., 2015). Thus, the amino acid sequence of the TM3–4 loop determines subclass-specific ion channel properties. All studies concerning chimeric receptors have been performed in overexpression systems in vitro leaving the question for an in vivo effect of chimeric proteins unanswered.

Our structural knowledge of the TM3–4 loop is limited to small segments close to TM3 and TM4. The rest of the TM3–4 loop seems to be disordered (Unwin, 2005). The C-terminal end of the TM3–4 loop of cation-selective CLRs forms an α-helical domain, called the MA stretch (membrane-associated stretch; Unwin, 2005; Hassaine et al., 2014). A large content of charged residues within the MA stretch face a lateral tunnel or portal. These portals enable the permeation of the incoming ions and influence ion channel conductance of the appropriate channel (Kelley et al., 2003).

The structure of the serotonin receptor provided some hints that there is a second α-helical stretch at the beginning of the TM3–4 loop (**Figures 1C**, **3A**). The formation of intracellular portals is allocated by the C-terminal MA-stretch and obstructed by the N-terminal helix called MX-helix in a presumably closed channel conformation (Hassaine et al., 2014). The existence of such portals in GlyRs has been proposed due to sequence homology (Carland et al., 2009). Mutations of eight basic residues within the supposed glycinergic portals resulted in non-functional receptors. Moreover, quadruple mutations of positively charged residues (α1 R377A/K378A/K385A/K386A and α1 R377E/K378E/K385E/K386E) reduced ion channel conductance at negative membrane potentials (**Figure 3**). Therefore, these portals are indeed features of an extended glycine receptor permeation pathway (**Figures 1C**, **3A,B**). The positive charges surrounding the intracellular portals are assumed to electrostatically attract incoming anions to the intracellular compartment (Carland et al., 2009). CD spectroscopy further revealed the existence of α-helical elements close to TM3 and TM4 in GlyRα1 (Burgos et al., 2015).

The TM3–4 sequence of GlyRs can be subdivided into variable and conserved regions (Melzer et al., 2010; **Figure 1C**). Basic stretches are highly conserved among various GlyRs. Two other motifs have been determined to the variable region, a poly ''NNNN'' motif and a proline-rich stretch present in α and β subunits. The role of the asparagine-rich subdomain is completely unsolved.

The existence of a poly-proline helix type II (PPII) within the TM3–4 loop of the GlyR formed by the poly-proline stretch has been proposed by CD-spectroscopy (Cascio et al., 2001; Breitinger et al., 2004). PPII helices are helical secondary structures with a perfect 3-fold rotation symmetry forming SH3 consensus sequences (SRC homology 3 domain consensus sequences, Rath et al., 2005). The recognition motif for the PPII helix xxPxxP is highly conserved among all GlyR subunits and is involved in binding of intracellular partners to the GlyRβ loop (**Figure 1D**; Koch et al., 2011; Del Pino et al., 2014). Syndapin was identified as a binding partner of the <sup>384</sup>KxxPxxPxxP<sup>394</sup> motif in GlyRβ. The interaction between syndapin I and GlyRβ was greatly diminished when the second proline was exchanged by another residue (Del Pino et al., 2014). A miRNA knockdown of syndapin I in cultured primary spinal cord neurons assigned syndapin I as a mediator in GlyR trafficking or even anchoring (Del Pino et al., 2014). The latter needs further investigations to be proven.

Neuroligin 2 or the GABA<sup>A</sup> receptors α2 harbor prolinerich sequences similar to the <sup>365</sup>PPPAP<sup>369</sup> motif in GlyRα1 and <sup>385</sup>PPPAKP<sup>390</sup> GlyRβ subunits. The interactions of these proline-rich stretches of neuroligin 2 or GABAAR α2 with the SH3 domain of CB underlie a novel regulatory mechanism for

formation and function of inhibitory postsynapses (Soykan et al., 2014). CB has, however, never been shown to directly interact with GlyRs.

A further intracellular protein interaction has been attributed to the 15 residues splice cassette of GlyRα3L in the TM3–4 loop. GlyRα3L binding to the vesicular trafficking protein Sec8 targets GlyRα3L to presynaptic sites. Colocalization with the vesicular presynaptic marker VGLUT1 confirmed axonal trafficking of GlyRα3L towards presynaptic terminals (Winkelmann et al., 2014).

In conclusion, emerging evidences suggest a so far underestimated role of the GlyR TM3–4 loop in the interaction with other intracellular proteins beside gephyrin connecting the receptor to cytoskeletal elements, regulating receptor trafficking and synaptic localization.

### MOTIFS IMPORTANT FOR TRAFFICKING AND MODULATION OF CHANNEL PHYSIOLOGY BY PHARMACOLOGICAL SUBSTANCES

Basic residues <sup>316</sup>RFRRKRR<sup>322</sup> localized within the proposed MX-helix at the N-terminal portal of the TM3–4 loop determine ion channel properties (**Figure 1C**). The integrity of this positively charged domain is important for proper membrane integration of the apolar TM3 (Sadtler et al., 2003). Neutralization of one or two basic residues resulted in translocation to the endoplasmic reticulum (ER).

Furthermore, some residues of the basic motif (318RRKRR<sup>322</sup> in GlyRα1; <sup>324</sup>RRKRK<sup>328</sup> GlyRα3) are parts of a nuclear localization signal (NLS). Residues of the NLS interact with karyopherins α3/α4 and are actively involved in the nuclear import of GlyRs (**Figure 1C**; Melzer et al., 2010). Although, the function of GlyRs within the nucleus is unknown, an important function of nuclear import in non-neuronal tissue (Van den Eynden et al., 2009) and brain tumors has been demonstrated (Förstera et al., 2014). In glioma, a knockdown of the NLScontaining GlyRα1 reduced the self-renewal capacity of glioma formation in vivo and therefore impaired tumor progression.

Within the basic stretches, residues <sup>316</sup>RFRRK<sup>320</sup> and <sup>385</sup>KK<sup>386</sup> are critical for binding cytosolic G-protein subunits (Gβγ; Yevenes et al., 2006) which in turn enhance the glycineinduced chloride currents in vitro (Yevenes et al., 2003). It has been further estimated that the interaction of the sequences <sup>316</sup>RFRRK<sup>320</sup> and <sup>385</sup>KK<sup>387</sup> with the G-protein subunit Gβγ correlates with an allosteric interaction of the same motifs with ethanol (Yevenes et al., 2010). A peptide composed of the motif <sup>316</sup>RFRRKRR<sup>322</sup> was able to inhibit binding of Gβγ to the GlyRα1 intracellular loop and thus decreased the positive modulation by ethanol (**Figure 1C**; San Martin et al., 2012). Further determinants for ethanol binding are localized in TM2, the alternative splicing cassette within the TM3–4 loop of the α1 subunit and within the short extracellular C-terminus (Sánchez et al., 2015). Directly correlated to these data is knowledge from knock-in mice carrying K385A/K386A substitutions which show a reduced sensitivity for ethanol (Aguayo et al., 2014). K385 also plays an important role in the allosteric modulation by endocannabinoids (Yevenes and Zeilhofer, 2011). Although the GlyRα3 subunit shares sequence similarities with the GlyRα1 in terms of basic residues, GlyRα3 subunits have not been modulated by either ethanol or by Gβγ proteins. Using a chimeric approach between α1 and α3, it was demonstrated that the 15 residues alternative splice cassette of α3 and the C-terminus contains modulatory sites for Gβγ interaction in addition to the required, but not sufficient residue G254 (Sánchez et al., 2015).

### POSTTRANSLATIONAL MODIFICATIONS—UBIQUITINATION AND PHOSPHORYLATION

Residues within the ICD of GlyRs are modulated by posttranslational modifications. Ubiquitination of postsynaptic proteins marks proteins for proteolytic degradation (Christianson and Green, 2004). Many recessive hyperekplexia mutations cause an accumulation of GlyR protein in the ER and within Golgi compartments and influence ubiquitin-mediated receptor degradation (Villmann et al., 2009b; Schaefer et al., 2015). It is proposed that ubiquitination of the GlyRα1 subunit takes place at 3 out of 10 lysine residues within the TM3–4 loop triggering receptor internalization and proteolytic degradation (**Figure 1C**). Proteolytic cleavage of the full-length GlyRs generates two fragments of 13 kD and 35 kD (Buttner et al., 2001). These two fragments have never been observed at the cellular surface. Processing of GlyR receptors is therefore a downstream process of ubiquitination within the endocytic degradation pathway.

GlyR subtypes are phosphorylated by protein kinases A and C (PKA and PKC; **Figure 1C**). Both kinases influence the maximal chloride influx and desensitization (Vaello et al., 1994; Gentet and Clements, 2002). Residue S391 within the TM3–4 loop of GlyRα1 was identified as a PKC-binding site (Ruiz-Gómez et al., 1991). Phosphorylated α1 receptors regulate channel activity and modulate the interaction with other intracellular proteins (Changeux et al., 1984). A stimulation of PKC by phorbol 12 myristate (PMA) led to an enhanced GlyR internalization rate via endocytosis. Mutation of a di-leucine motif (L314/L315) within the TM3–4 loop prevented the PMA-stimulated receptor endocytosis (Huang et al., 2007). Phosphorylation of S403 of the GlyRβ subunit reduces the affinity between the GlyRβ TM3–4 loop and gephyrin resulting in enhanced lateral diffusion of GlyRs and less synaptic GlyR levels (Specht et al., 2011).

Phosphorylation of the GlyRα3 subunit plays an important role in pain sensitization processes. PGE2 inhibits glycinergic neurotransmission via a PKA-dependent pathway (Harvey et al., 2004). The sequence Arg-Glu-Ser-Arg in the TM3–4 loop of GlyRα3 represents a strong consensus sequence for PKA. PGE2 receptors activate PKA, which in turn enhances the fraction of phosphorylated GlyRα3 via residue S346 within the PKA consensus sequence. A decrease in glycinergic signal transduction is a consequence of increased internalization of phosphorylated GlyRα3. Residue S346 is not conserved in α1 and therefore α1 lacks modulation by PKA (Harvey et al., 2004). This study clearly showed the unique role of phosphorylated GlyRα3 in spinal nociceptive processes, whereas phosphorylation of GlyRα1 controls spinal motor circuits.

Furthermore, evidence of conformational GlyR modulation by phosphorylation have been obtained in a combined approach of voltage clamp fluorometry and pharmacological measurements. The GlyRα3 S346 mutant was unable to induce conformational changes in the extracellular ligand-binding site compared with wild-type α3. These data showed for the first time that phosphorylation encompasses structural changes in the TM3–4 loop that propagate towards the ECD of the receptor (Han et al., 2013b).

SUMOylation is another type of posttranslational modification influencing receptor endocytosis and ion channel function. Although direct SUMOylation of GlyRs has never been shown, SUMOylation of kainate receptors indirectly influences GlyR endocytosis (Konopacki et al., 2011; Chamberlain et al., 2012). Recently, another kainate-induced mechanism for GlyR endocytosis has been resolved. This process involves a calcium-dependent de-SUMOylation of PKC. Activation of PKC by de-SUMOylation reduced GlyR-mediated synaptic activity concomitant to GlyR endocytosis (Sun et al., 2014). This crosstalk between excitatory and inhibitory receptors may serve to maintain the excitatory–inhibitory balance in the CNS.

#### ICD INTERACTION WITH SCAFFOLD PROTEINS ENABLES INHIBITORY SYNAPSE FORMATION

The best analyzed interaction between the GlyR and an intracellular binding partner is the interaction of the GlyRβ subunit with the scaffold protein gephyrin. This direct interaction involves GlyRβ residues 398–410 (Kim et al., 2006).

Gephyrin itself is a cytoplasmic protein, which consists of N-terminal G domains and C-terminal E domains (homologous to E. coli proteins MogA and MoeA—molybdenum cofactor biosynthetic proteins, Schwarz et al., 2001) connected by a central linker region. These domains form a hexagonal structure built up by G domain trimers and E domain dimers (Saiyed et al., 2007) anchoring GlyRs at the postsynaptic membrane (Kneussel and Betz, 2000). The binding motifs of the gephyrin E domain to GABA<sup>A</sup> receptors (Maric et al., 2014) and the GlyRβ TM3–4 loop sequence <sup>398</sup>FSIVGSLPRDFELS<sup>411</sup> (**Figure 1D**) have been identified (Meyer et al., 1995). Besides its role as an anchoring protein, gephyrin undergoes interactions with polymerizing tubulin (Kirsch et al., 1991) as well as the microtubuli-associated motor proteins KIF5 and dlc1/2. These interactions are involved in anterograde and retrograde transport mechanisms of GlyRs at inhibitory synapses (Fuhrmann et al., 2002; Maas et al., 2009). Among numerous intracellular proteins bound to gephyrin, the GDP/GTP-exchange factor CB is especially interesting (Kins et al., 2000; Fritschy et al., 2008). Knockout of CB results in a region-specific loss of gephyrin in the hippocampus and gephyrin-binding GABA<sup>A</sup> receptor subtypes in the forebrain of knockout mice (Papadopoulos et al., 2007, 2008). Although several attempts have been started to identify novel interaction partners of the GlyR TM-3–4 loop using yeast two hybrid screens, mostly gephyrin has been detected due to its high affinity for the GlyRβ loop. One might conclude that the affinity between other intracellular binding partners and GlyRs may be too low with respect to the sensitivity of a yeast two hybrid approach.

Using mass spectrometry, transport proteins Vps35 and neurobeachin (Nbea) and the F-bar protein syndapin I were detected as binding partners of the GlyRβ TM3–4 loop (Del Pino et al., 2011, 2014). Syndapines are important for vesicle formation at the cellular membrane, within the trans-Golgi network and the proteasome (Qualmann and Kelly, 2000; Kessels and Qualmann, 2004). Thus, the GlyRβ TM3–4 loop acts as an adapter for other intracellular binding partners involved in transport processes of receptor complexes towards the cellular membrane.

#### DESENSITIZATION

Desensitization is defined as the transition of the agonistbound open channel into a closed ion channel configuration in the presence of agonist. Wild-type α1 and α3 GlyRs show very small portions of desensitizing currents. In vitro mutagenesis studies on the TM3–4 loop of various GlyRα subunits revealed single amino acids and grouped residues involved in the desensitization process of GlyR channels (Nikolic et al., 1998; Breitinger et al., 2009; Meiselbach et al., 2014). The human GlyRα3 carries an alternativesplicing cassette of 15 residues within the TM3–4 loop. The resulting variants α3L (including the 15 residues) and α3K (short, lacking the alternative-splicing cassette) differ significantly in their desensitization behavior (Nikolic et al., 1998). These data provided first evidences for the importance of the intracellular TM3–4 loop for ion channel desensitization (**Figure 1C**). The lack of this alternative-splicing cassette generated fast desensitizing currents in contrast to almost no desensitization observed for the long GlyRα3 variant (Nikolic et al., 1998). The alternative-splicing cassette of GlyRα1 subunit does not influence receptor desensitization most probably due to differences in amino acid composition compared to α3. The α3 cassette harbors three possible phosphorylation consensus sites. A substitution of residues carrying hydroxyl side chains (α3L1OH = α3LT358A/Y367F/S370A) within the 15 amino acid insert generated an intermediate state of desensitization between α3L and α3K suggesting that hydroxyl groups mediate desensitization processes (**Figure 4A**; Breitinger et al., 2002). In a follow-up study, the secondary structure analysis of α3K and α3L suggested a stabilization of the overall spatial structure of the TM3–4 loop by the α3 splice cassette (Breitinger et al., 2009). The importance of the alternativesplicing cassette was further supported in an in vitro study of α1α3 chimeric proteins. The analysis of α1α3 chimera allocated that desensitization properties are transferable between GlyR subunits (**Figures 4B–D**; Meiselbach et al., 2014). Chimeras containing the α3 insert desensitized significantly slower than chimeras lacking the splice cassette.

The TM3–4 loop length differences between prokaryotic and eukaryotic CLRs (Tasneem et al., 2005) posed the following question: Is the TM3–4 loop essential for CLR function? Crystal structures of the prokaryotic channels ELIC and GLIC revealed both the open conformation (GLIC) and the closed channel conformation (Hilf and Dutzler, 2008, 2009;

FIGURE 4 | Desensitization determined by alternative splicing cassette in GlyRα3. (A) Desensitization of recombinant α3 glycine receptors, left—α3L almost non-desensitizing (black curve; contains alternative splicing cassette), middle—α3K fast desensitizing (black curve; without alternative splicing cassette), right α3L1OH intermediate desensitization (shown in blue; with alternative splicing cassette but mutated hydroxylated residues α3L1OH = α3LT358A/Y367F/S370A). The curve of the intermediate state is also shown in panels of α3L and α3K for comparison (dotted blue line). Note, differences in desensitizing current fractions: α3L 18%, α3K 83%, and α3L∆OH 45%, modified from Breitinger et al. (2002). (B) GlyRα1-α3 chimera with either the TM3–4 loop, TM4 and the C-terminus of α3L or α3K, or the TM3–4 loop only of α3L or α3K with the remaining sequence of α1. The 15 residues of the alternatively spliced segment are depicted above the scheme (positions 358–372). (C) Maximal glycine-evoked currents (Imax) recorded using whole-cell configurations from HEK293 cells expressing chimeric GlyRs. All chimeras responded to saturating glycine concentrations but differed in their desensitization kinetics. Variants harboring the α3K TM3–4 loop are fast desensitizing shown by blue overlays of the appropriate current traces. (D) Fractions of desensitizing currents of α1α3 chimera compared to α3L (non-desensitizing) and α3K (desensitizing), blue boxes refer to chimeras containing the TM3–4 loop of α3K (modified from Meiselbach et al., 2014). n.s., not significant. Level of significance, <sup>∗</sup>p < 0.05, ∗∗p < 0.01.

Bocquet et al., 2009). Although first studies indicated a nondesensitized GLIC in an acidic environment (Bocquet et al., 2007), GLIC desensitization became obvious at a pH lower than 5 (Gonzalez-Gutierrez and Grosman, 2010; Parikh et al., 2011). These data again argue for subtype-specific regulatory elements of desensitization within the CLR superfamily. An exchange of the whole TM3–4 loop of various CLRs (5HT<sup>3</sup> and GABA<sup>C</sup> receptor) with the ICD of GLIC (SQPARAA) did not lead to changes in the macroscopic electrophysiological properties of the chimeric ion channels (Jansen et al., 2008; Papke and Grosman, 2014). In a recent study, the full-length loop of GlyRα1 was either replaced completely by the prokaryotic heptapeptide (i), or (ii) basic stretches <sup>318</sup>RRKRR and <sup>393</sup>KKIDK close to TM3 and 4 have been left intact carrying the heptapeptide in between (GlyRα1-GLIC(+)bm). (iii) A third construct contained a short TM3–4 loop only composed of both basic stretches (GlyRα1-∆TM3–4(+)bm; **Figure 3C**). The pure heptapeptide between TM3 and TM4 resulted in intracellular aggregation, lack of surface receptors and non-functionality. Constructs GlyRα1- GLIC(+)bm (ii) and GlyRα1-∆TM3–4(+)bm (iii) were able to form functional ion channels that differed significantly in their desensitization behavior. The presence of both basic stretches resulted in a fast transition of GlyRα1 channels into a closed conformation. The insertion of SQPARAA between both basic motifs (GlyRα1-GLIC(+)bm) decreased the desensitizing current significantly in comparison to wild-type GlyRα1 (**Figure 3D**). Thus, the sequence between both basic stretches determines the desensitization behavior of GlyRα1 (Langlhofer et al., 2015). The introduction of the prokaryotic heptapeptide at another position within the GlyRα1 TM3–4 loop between residues Q310 and K385 depicted also differences on the fraction of desensitizing currents (Papke and Grosman, 2014). The common conclusion from studies concerning the length of TM3–4 loop and the determination of desensitization rates revealed that separation of both basic stretches at the N- and C-terminal end of the TM3–4 loop represent a critical determinant of ion channel functionality.

To complete the knowledge on desensitization determined by the GlyR ICD, the human mutation P250T needs to be mentioned. This mutant localized in the M1-M2 loop is associated with very fast desensitization. The original proline introduces conformational rigidity to the short M1-M2 linker. The given higher flexibility by the introduced threonine allows TM2 rearrangements resulting in fast ion channel closure. Thus, fast desensitization underlies the pathology of patients carrying P250T and in turn contributes to enhanced muscle tone delineating a major clinical feature in startle disease patients (Saul et al., 1999; Breitinger et al., 2001). Further support for a key role of the M1-M2 loop in desensitization derives from a recent study on the identification of the desensitization gate in CLRs. The TM1–2 loop interacts with the internal end of TM3 determining the desensitization gate. An exchange of GlyR residues with residues from the GABA<sup>C</sup> ρ1 subunit elicited the intracellular end of TM3 as the key component for desensitization (Gielen et al., 2015). Further hints for an association of enhanced desensitization and disease were given by studies of the nAChR. The enhanced desensitization of presynaptic nAChRs at GABAergic terminals generates lower inhibitory input at dopaminergic neurons and concomitantly enhanced activity of the dopaminergic rewards system (Mansvelder et al., 2002). An enhanced desensitization rate of nAChRs has also been described to underlie a special form of frontal lobe epilepsy (Bertrand et al., 2002).

#### CONCLUSIONS AND OUTLOOK

The ICD of the glycine receptor harbors subdomains important for trafficking and functionality of the inhibitory GlyR. Basic residues are crucial determinants in both processes. Since trafficking is a prerequisite for functional modulation, the basic domains represent key regulators of this receptor family. This is further supported by their involvement in binding of Gβγ proteins and ethanol.

Studies on chimeric proteins have helped us to understand the functional role of the TM3–4 loop. Lack of this large intracellular loop does not lead to non-function, rather to a disruption of ion channel modulation. Except for the cytoplasmic portals that

#### REFERENCES


are proposed to resemble an α-helical structure, the TM3–4 loop is suggested to be unfolded. Unfolding might represent an advantage for the interaction with intracellular proteins important for regulation of receptor recruitment to synaptic sites, ion channel function, and finally degradation initiation. Further research is required to enhance our knowledge on other so far non-identified interactions partners modulating synaptic strength and fine-tuning of GlyR function depending on the surrounding neuronal network.

#### AUTHOR CONTRIBUTIONS

GL and CV wrote the manuscript.

#### ACKNOWLEDGMENTS

This work was supported by the Deutsche Forschungsgemeinschaft (DFG VI586 to CV) and the Bayerische Forschungsstiftung. GL was further supported by the Graduate School of Life Science Würzburg.

govern glycine receptor activation and desensitization. J. Biol. Chem. 276, 29657–29663. doi: 10.1074/jbc.M100446200


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

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

# Utility of Induced Pluripotent Stem Cells for the Study and Treatment of Genetic Diseases: Focus on Childhood Neurological Disorders

Serena Barral <sup>1</sup> and Manju A. Kurian1,2 \*

<sup>1</sup> Neurogenetics Group, Molecular Neurosciences, UCL Institute of Child Health, University College London, London, UK, <sup>2</sup> Department of Neurology, Great Ormond Street Hospital, London, UK

The study of neurological disorders often presents with significant challenges due to the inaccessibility of human neuronal cells for further investigation. Advances in cellular reprogramming techniques, have however provided a new source of human cells for laboratory-based research. Patient-derived induced pluripotent stem cells (iPSCs) can now be robustly differentiated into specific neural subtypes, including dopaminergic, inhibitory GABAergic, motorneurons and cortical neurons. These neurons can then be utilized for in vitro studies to elucidate molecular causes underpinning neurological disease. Although human iPSC-derived neuronal models are increasingly regarded as a useful tool in cell biology, there are a number of limitations, including the relatively early, fetal stage of differentiated cells and the mainly two dimensional, simple nature of the in vitro system. Furthermore, clonal variation is a well-described phenomenon in iPSC lines. In order to account for this, robust baseline data from multiple control lines is necessary to determine whether a particular gene defect leads to a specific cellular phenotype. Over the last few years patient-derived neural cells have proven very useful in addressing several mechanistic questions related to central nervous system diseases, including early-onset neurological disorders of childhood. Many studies report the clinical utility of human-derived neural cells for testing known drugs with repurposing potential, novel compounds and gene therapies, which then can be translated to clinical reality. iPSCs derived neural cells, therefore provide great promise and potential to gain insight into, and treat early-onset neurological disorders.

Keywords: iPSCs, childhood neurological disorders, in vitro disease modeling, gene therapies, drug screening, isogenic control

### INTRODUCTION

Over the last decade, significant advances such as whole exome and genome sequencing have facilitated genetic screening of patients, resulting in an ever-increasing number of inherited human diseases. Despite this genetic revolution, the molecular mechanisms downstream of a specific gene mutation or genetic variant remain yet to be fully elucidated for the majority of diseases. Future research priorities must therefore lie in studying such disorders in more depth, to not only understand the disease, but also to develop novel therapies for clinical translation.

#### Edited by:

Kirsten Harvey, University College London, UK

#### Reviewed by:

Erik Maronde, Goethe-Universität, Germany Hansen Wang, University of Toronto, Canada

> \*Correspondence: Manju A. Kurian manju.kurian@ucl.ac.uk

Received: 02 June 2016 Accepted: 15 August 2016 Published: 06 September 2016

#### Citation:

Barral S and Kurian MA (2016) Utility of Induced Pluripotent Stem Cells for the Study and Treatment of Genetic Diseases: Focus on Childhood Neurological Disorders. Front. Mol. Neurosci. 9:78. doi: 10.3389/fnmol.2016.00078

To date, transgenic animal models and transformed cell lines, have allowed clarification of pathophysiological pathways affected by genetic mutations. Despite their benefits, both methods have a number of limitations, in that they often do not fully mimic human physiology, only partially recapitulate progression of disease, and do not accurately recapitulate human metabolism and homeostasis. It has long been recognized that patient derived cells are a potentially better in vitro tool for studying human disease. However, human tissue is often either unavailable or simply not accessible. This is clearly exemplified by neurological disorders, where accessing the brain and neuronal tissue for cell culture and future study is near impossible. Since the first human embryonic stem cells (ESCs) were isolated in 1998 (Thomson et al., 1998), the use of pluripotent stem cells (PSCs) has become a new reality in the study of human diseases, offering a challenging but incredibly useful model to move from clinic to bench and potentially vice versa.

The discovery of cellular reprogramming techniques has been a major step forward in the in vitro modeling of human disease, theoretically allowing the study of all genetic disease with specific patient cells as the starting point. Yamanaka and colleagues (Takahashi et al., 2007) elegantly reprogrammed adult dermal fibroblasts to a pluripotent state, by inducing ectopic expression of four factors: Oct4, Sox2, Klf4, and cMyc (Takahashi et al., 2007; Takahashi and Yamanaka, 2016). The induced PSCs generated are highly similar to human ESCs, with the ability to indefinitely proliferate and differentiate in cells derived from the three germ layers. Since publication of Yamanaka's landmark article, reprogramming techniques have been further refined, and many new strategies have been developed to effectively reprogram somatic cells into pluripotency. Integrating retroviruses and lentiviruses have been superseded by the use of non-integrating systems, including adenovirus, Sendai virus, mRNA, episomal vectors, proteins and small molecules (Fusaki et al., 2009; Kim et al., 2009; Zhou and Freed, 2009; Warren et al., 2010; Okita et al., 2011; Bar-Nur et al., 2014).

Neural stem cells (NSCs) have been successfully derived from PSCs and several protocols for PSCs differentiation into a broad variety of mature neurons and glial cell subtypes have been published (Srikanth and Young-Pearse, 2014). Patientderived neural cells have the specific advantage of retaining the genetic background of the donor and thus offer a unique in vitro neuronal disease model. They are an unlimited source of cells that allows the analysis of the cellular mechanisms involved in disease. Furthermore, they provide a novel platform to test new drugs and genetic therapies as well as a source of cells that could potentially be used for cell replacement therapy (**Figure 1**).

When approaching the study of neurological diseases using human induced pluripotent stem cell (iPSC)-derived neurons as an in vitro model system, several considerations need to be taken into account, including the accurate generation of truly pluripotent cells, the relative efficiency of neuronal differentiation, and the strengths, utility and limitations of generated neurons. In this review, we provide a brief overview of what we consider to be the most important

FIGURE 1 | Use of patient-derived induced pluripotent stem cells (iPSCs) for modeling genetic neurological diseases. Genetic screening of patients affected by a neurological disorder may lead to the identification of specific mutations causing disease. Patient-derived somatic cells (fibroblast and other cell types) can be then reprogrammed to a pluripotent state. iPSCs carrying the disease-related mutation (indicated in red) can be then differentiated into the neural cells type (neurons and glial cells) which are affected in the disease. This allows for the in vitro study of the molecular mechanisms downstream the genetic mutation. In order to overcome genetic background variability and to validate the effect of the genetic mutation on phenotype observed in vitro, isogenic control iPSCs can be generated via genomic correction of the mutation. Moreover, in vitro differentiated cells can be used for high-throughput screening of drugs or the validation of specific genetic therapies that can then be translated into clinical practice.

advantages and disadvantages of using human iPSCs to model neurological diseases, and their translational utility at a clinical level. Our main focus will be to evaluate this model system for early-onset genetic neurological disorders (**Table 1**), although, where relevant and appropriate, we will use examples from other later-onset neurological diseases.

#### NEURAL DIFFERENTIATION OF HUMAN iPSC: A WIDE VARIETY OF CELL TYPES

Differentiation of iPSCs into neural cells is based on recapitulating embryonic development and relies on the use of specific factors that can promote or inhibit specific signaling pathways. All methods published so far can guarantee high purity of NSCs, but it is more challenging to obtain decent percentages of specific subtypes of desired mature neural cells. To date, it is possible to derive a wide variety of neuronal cell types from PSCs, including forebrain neuronal neurons (Espuny-Camacho et al., 2013; Lancaster et al., 2013); motor neurons (Wada et al., 2009; Nizzardo et al., 2010); dopaminergic neurons (Kriks et al., 2011; Kirkeby et al., 2012); GABAergic neurons (Maroof et al., 2013; Nicholas et al., 2013); medium spiny neurons (Delli Carri et al., 2013); forebrain cholinergic neurons (Wicklund et al., 2010; Hu et al., 2016); serotonergic neurons (Erceg et al., 2008); caudal neurons (Kirkeby et al., 2012); cerebellar neurons (Erceg et al., 2010); astrocytes (Emdad et al., 2012; Juopperi

#### TABLE 1 | Utility of induced pluripotent stem cells (iPSC) in childhood-onset neurodevelopmental and neurological disorders.


#### TABLE 1 | (Continued).


et al., 2012); oligodendrocytes (Nistor et al., 2005; Ogawa et al., 2011). Neuronal populations generated are typically heterogeneous, presenting both mature and immature cells, and thus need further technologies to achieve a high level of purity. Sorting techniques are often useful, but there are few neuronal subtype-specific surface markers available to select desired neural subpopulations (Pruszak et al., 2009; Yuan et al., 2011; Doi et al., 2014). To overcome this issue, sorting can sometimes be achieved by the expression of a selectable marker included as a reporter under expression of specific transcription factors or proteins (DeRosa et al., 2015; Toli et al., 2015).

#### IN VITRO DERIVED NEURAL CELLS: WHAT IS THEIR TRUE DEVELOPMENTAL STAGE?

During reprogramming into iPSCs, somatic cells return to a developmental stage similar to that of ESCs, independent of their original age. Indeed, age-related characteristics of the original cells, (such as nuclear abnormalities, telomere length, and mitochondrial activity) are lost during this re-set to an embryonic stage (Marion et al., 2009; Suhr et al., 2010). Differentiation protocols for the generation of neuronal subtypes from PSCs require a significant amount of time spanning from weeks to months (Srikanth and Young-Pearse, 2014) to produce neurons that show a relatively mature morphological, molecular and electrophysiological phenotype. Despite this maturation process, generated neurons are still reminiscent of human fetal neurons (Mariani et al., 2012; Lancaster et al., 2013; Miller et al., 2013; Vera and Studer, 2015). It is therefore conceivable that such in vitro model systems could fail to recapitulate the disease phenotype especially for late-onset disorders.

Studer and colleagues (Miller et al., 2013) have addressed this issue, by developing a genetic strategy for introducing agingrelated features in iPSC-derived neurons, specifically studying Parkinson disease (PD). Specifically, PD patient iPSC-derived midbrain dopaminergic neurons (mDA) recapitulate some PD disease features, including α-synuclein (αSYN) accumulation, oxidative stress, defects in neural outgrowth and mitochondrial dysfunction (Byers et al., 2011; Jiang et al., 2012; Reinhardt et al., 2013). However key late disease features of PD, such as neural degeneration, were only evident in model systems exposed to external stressors (Byers et al., 2011; Nguyen et al., 2011). Importantly, the accumulation of Lewy bodies and appearance of neuromelanin, a distinct feature of adult mDA neurons, have not been observed in iPSC-derived neurons. In this case, Studer and colleagues transiently overexpressed progerin, and showed restoration of aging features in both fibroblasts and mDA neurons derived from iPSCs. In particular, in iPSC-derived mDA neurons, they observed features of normal neural aging, with degeneration of dendrites in vitro and neuromelanin accumulation after grafting in vivo (Miller et al., 2013).

The relatively immature characteristics of iPSC-derived neurons should therefore always be a major consideration when modeling postnatal-, childhood- and adult-onset neurological diseases, where pathological features only manifest during postnatal development or the aging process.

### VARIABLE GENETIC BACKGROUND: DEFINING GOOD CONTROLS

Using patient derived iPSCs guarantees a unique opportunity to study the phenotype associated with a specific mutation in the context of the genetic background in which the mutation leads to disease. However, genetic background and potential genetic modulators of disease could conceivably affect the phenotype of both healthy control and patient lines. This is a definite limiting factor in the study of both Mendelian and more complex multigenic/multifactorial disorders. One solution is to use iPSCs derived from un-affected relatives for comparison, or to compare several control and several different patient lines in the same study. However, both these methods can be costly and time consuming. In more recent times, gene-editing technologies have become a more robust method by which the effect of a specific genotype on the iPSC model system can be unequivocally validated. Indeed, for monogenic disorders, correction of the single mutation in an iPSC cell line allows development of a unique ''isogenic'' control which harbors the same genetic background of the patient, thereby decreasing any ''background noise'' that could mask or affect cell phenotype. Furthermore, the insertion of a specific mutation into a control line can be utilized to show that such mutagenesis can induce disease phenotypes into a control line, akin to those seen in patient lines. Such concepts have been elegantly illustrated by Reinhardt et al. (2013) in their study of LRRK2-related PD, where they generated and compared several control lines, from healthy age- and gendermatched individuals, vs. isogenic controls using gene editing tools. No significant difference in αSYN levels was observed when comparing mDA neurons from wild type to LRRK2-mutated patient derived iPSC lines. However, αSYN levels were markedly reduced in patient lines when compared to the corresponding corrected isogenic lines. Gene expression profiles after 30 days of differentiation revealed significant differences in the age- and gender-matched iPSC lines when compared to both patient and isogenic lines. It is therefore clear that genetic background can have effects on gene expression, and the comparison of patient lines with isogenic controls can help overcome such genetic variability.

A number of different techniques can be used to generate isogenic controls. Reinhardt et al. (2013) used Zinc Finger Nuclease (ZFNs) technology. Both ZFN and the similar Transcription Activator-Like Effector Nuclease (TALENs) have been used successfully for gene editing in iPSCs (Hockemeyer et al., 2009, 2011). Both methods rely on the generation of costumed DNA binding domains conjugated to Fok1 nuclease, which can induce double strand breaks in a non-specific manner. The landmark discovery of the Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)-Cas9 system has added a highly efficient tool to manipulate the iPSCs genome (Ding et al., 2013) using endonuclease Cas9 and a guide RNA, targeting a specific region. Due to the efficient, relatively fast and targeted approach of the CRISPR-Cas9 technique compared to ZFNs and TALENs, this newer technique appears to be the preferred genome editing methodology for many researchers. Furthermore, the different commercial companies now offer generation of custom isogenic control lines using CRISPR-Cas9 technology (Baker, 2014).

### NEURONAL CELLS IN A DISH: THE NEED FOR A 3D SYSTEM

Although iPSCs represent an advantageous tool to study molecular phenotypes in neurological disease, their two dimensional nature means that some of the environmental factors, regional identities and complex neural circuits are absent when compared to either an animal disease model or human patient. Different studies have shown the intrinsic ability of NSCs to spontaneously self-organize in 3-dimensional (3D) structures resembling whole organs (Lancaster and Knoblich, 2014). Lancaster et al. (2013) showed how simple cultivation of control iPSCs in suspension can give rise to organoids which display several brain regions along the rostro-caudal and dorsoventral pathways of mainly the forebrain and mid-hindbrain areas. The degree of cellular organization was maximal in the dorsal cortical region of the generated organoid where they observed a layered organization typical of the developing human forebrain. The same group then generated organoids from iPSCs derived from patients with microcephaly. The patient-derived cortical organoids mimicked a number of the features seen in disease, including smaller neural tissues with few progenitors regions, and radial glial maturation and orientation abnormalities, which were not previously recapitulated in a murine model. Organoids therefore represent a powerful tool to study both human fetal neocortical development (Camp et al., 2015) and developmental disorders. The generation of organoids representing more caudal fetal regions, (such as the midbrain and hindbrain) is also no doubt useful for studying neurological disorders affecting these areas of the central nervous system. Indeed, midbrain-like organoids (MLOs) have recently been derived from human ESCs (Jo et al., 2016). Neural cells of such MLOs not only expressed midbrain markers and displayed characteristic dopaminergic electrical activity, but also produced neuromelanin, a characteristic not observed in bi-dimensional systems.

### MOVING iPSCs INTO CLINICAL UTILITY: DRUGS, CELL AND GENE THERAPY FOR PHARMACORESISTANT CHILDHOOD NEUROLOGICAL DISORDERS

The development of new drugs to treat human disorders is a challenging field. Indeed, many drugs tested in animal models have failed in human clinical trials due to lack of efficacy or intolerability (Scannell et al., 2012). Overall, new drugs for pharmacoresistant disorders are an unmet need, and constitute a priority area for research. The development of new drugs is hindered by the lack of appropriate models. Human iPSCs offer a unique opportunity for high-throughput drug screening in patient derived cells to assess drug efficacy and toxicity. Despite being in the early stages, several studies have demonstrated the feasibility and usefulness of this approach for childhood neurological disorders. For example, motor neurons derived from patients affected by Spinal Muscular Atrophy (SMA) have been used to test specific drugs (Xu et al., 2016). SMA is characterized by mutations in SMN1, which leads to degeneration of spinal motor neurons associated with mitochondrial dysfunction. Treatment of SMA human iPSCs derived motor neurons with N-acetylcysteine (NAC) improved mitochondrial functionality, thereby rescuing motor neuron degeneration in vitro.

In Rett syndrome (RTT) patient-derived iPSCs have been utilized to test the effect of IGF1 and gentamicin in vitro (Marchetto et al., 2010). RTT iPSCs derived glutamatergic neurons showed a decreased level of glutamatergic synapses when compared to controls, which was increased by IGF1 treatment. Gentamicin, administrated at high dose acted as a suppressor of nonsense mutations which cause impaired function of MeCP2 in RTT. Furthermore, IGF1 has been used to rescue the phenotype observed in neurons differentiated from Phelan–McDermid syndrome (PMDS)-derived iPSCs (Shcheglovitov et al., 2013). PMDS neurons showed impairment in excitatory synaptic transmission and reduced number of excitatory synapses, which were restored after treatment with IGF1.

These and other studies (**Table 1**), have mainly tested small number of compounds on iPSC differentiated cells, thereby demonstrating the feasibility of using iPSCs for drug testing. In the future, high-throughput technologies allowing the screening of an extensive library of compounds will be useful. Highcontent imaging and analysis are likely to be helpful for such high-throughput approaches (Sirenko et al., 2014). High-content high-throughput assays have already been undertaken on iPSCderived neural cells in 348-well assays, with analysis focusing on neurite outgrowth, cell number and viability, mitochondrial integrity and membrane potential.

Two studies have used high-throughput assays to screen large numbers of candidate drugs for Fragile X syndrome, a neurodevelopmental disorder characterized by learning problems, autism, and anxiety. This syndrome is associated with CGG repeat expansion in the 5<sup>0</sup> -untranslated region of FMR1, which leads to the absence of FMR protein. Both studies used patient derived iPSCs and differentiated them either in NSCs or neural precursors. In the first study 5000 compounds (both novel and approved drugs) were tested (Kumari et al., 2015), while the second study expanded drug screening to over 50,000 compounds (Kaufmann et al., 2015). In both studies FMR1 neural cells, treated with compounds as decanehydroxamate, deserpidine or tibrofan, showed increased mRNA levels, though not to clinically significant levels. Nevertheless, both studies show how promising high-throughput techniques can be in expanding the potential of drug testing using iPSCs derived cells.

In addition to the screening of new drugs, human iPSCs also offer other therapeutic possibilities that can be translated into clinical practice, as evident in the Phase I SMA study (Chiriboga et al., 2016). The antisense oligonucleotide nusinersen was designed to alter splicing of SMN2 mRNA. SMN1 motor neurons compensate with the paralogous gene SMN2. SMN2 shows a high sequence homology to SMN1, the only difference resides in the C-to-T base change inside exon 7. This mutation leads to abnormal SMN2 splicing and to the generation of truncated highly unstable proteins that trigger neural degeneration of motor neurons. SMN2 splicing correction with the use of oligonucleotides resulted in the production of a greater amount of full-length SMN2. This strategy has been tested on patientderived iPSCs differentiated motor neurons (Corti et al., 2012) and showed the ability to convert SMA-differentiated motor neurons to a normal phenotype, both in vitro and after grafting into a mouse model of disease.

iPSC-derived neural cells can also be a patient-derived platform to test genetic therapies. Juvenile neuronal ceroid lipofuscinosis (NCL) disorder is caused by loss-of-function mutations in CLN3. Patient-derived iPSCs neurons showed abnormal lysosomal storage with abnormalities observed in mitochondria, Golgi apparatus and endoplasmic reticulum. After restoring the function of CLN3 via AAVrh.10 virus bearing wild-type human CLN3, in vitro differentiated neurons showed a rescued phenotype, without excess accumulation of storage material (Lojewski et al., 2014). Similarly, a lentiviral approach was used for a genetic early onset form of PD. iPSCderived mDA harboring mutations in PINK1, a gene encoding a mitochondrial kinase, showed dysfunctional mitochondrial function. Expression of the non-mutated PINK1 via lentivirus in patient-derived mDA neurons restored normal recruitment of PINK1 upon mitochondrial depolarization, and normalized mitochondrial number and biogenesis (Seibler et al., 2011).

The combination of genetic engineering techniques and the promise of human PSCs differentiated cells as a donor source for cell replacement therapies, could in the future lead to the generation of patient-derived ''corrected'' cells that could potentially be used in autologous transplantation to replace affected disease cells. Such use of patient derived, genetically corrected neural cells may also potentially overcome immunemediated responses that might be triggered by using allogenic neural cells. Overall, even though such approaches are extremely promising and although PSCs are already used in clinical trials (Kimbrel and Lanza, 2015), there remain many issues regarding the use of PSC as cell replacement therapy, particularly concerning cell identity, purity, safety and long term risks. For this reason, even though it remains an extremely promising approach, clinical translation of such a therapeutic approach is likely to take some time.

#### CONCLUSION

Patient derived iPSCs represent a unique and increasingly utilized tool for the study of human genetic neurological diseases of childhood. iPSCs are an extraordinary model that can facilitate new insight into the molecular basis of disease and aid the development of new therapies, especially for pharmacoresistant diseases where human tissue is inaccessible for research purposes. Like all other laboratory models, human iPSCs have some limitations, namely that the model can be time consuming and costly to establish, shows clonal variability and genetic background can influence phenotype. Despite this, the ever-growing number of studies using human iPSCs to both model genetic disease and discover new therapies, render them an extremely promising tool, capturing the attention of researchers worldwide.

#### AUTHOR CONTRIBUTIONS

SB: conceptional design and writing of the manuscript. MAK: conceptional design and writing of the manuscript.

#### REFERENCES


#### ACKNOWLEDGMENTS

SB is funded by the Wellcome Trust (158065). MAK is funded by a Wellcome Trust Intermediate Clinical Fellowship and receives funding from GOSHCC, the Gracious Heart Charity Foundation, Rosetrees Trust, the AADC Research Trust, and supported by the National Institute for Health Research Biomedical Research Centre at Great Ormond Street Hospital for Children NHS Foundation Trust and University College London.

a human Huntington's disease patient. Biochem. J. 446, 359–371. doi: 10. 1042/bj20111495


asymptomatic SMN1-deleted individuals. Cell. Mol. Life Sci. 73, 2089–2104. doi: 10.1007/s00018-015-2084-y


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

The handling Editor declared a shared affiliation, though no other collaboration, with the authors SB, MAK and states that the process nevertheless met the standards of a fair and objective review.

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

# Precision Medicine in Multiple Sclerosis: Future of PET Imaging of Inflammation and Reactive Astrocytes

#### Pekka Poutiainen<sup>1</sup> , Merja Jaronen<sup>2</sup> , Francisco J. Quintana<sup>2</sup> and Anna-Liisa Brownell <sup>1</sup> \*

<sup>1</sup> Athinoula A Martinos Biomedical Imaging Center, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, MA, USA, <sup>2</sup> Ann Romney Center for Neurologic Diseases, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA

Non-invasive molecular imaging techniques can enhance diagnosis to achieve successful treatment, as well as reveal underlying pathogenic mechanisms in disorders such as multiple sclerosis (MS). The cooperation of advanced multimodal imaging techniques and increased knowledge of the MS disease mechanism allows both monitoring of neuronal network and therapeutic outcome as well as the tools to discover novel therapeutic targets. Diverse imaging modalities provide reliable diagnostic and prognostic platforms to better achieve precision medicine. Traditionally, magnetic resonance imaging (MRI) has been considered the golden standard in MS research and diagnosis. However, positron emission tomography (PET) imaging can provide functional information of molecular biology in detail even prior to anatomic changes, allowing close follow up of disease progression and treatment response. The recent findings support three major neuroinflammation components in MS: astrogliosis, cytokine elevation, and significant changes in specific proteins, which offer a great variety of specific targets for imaging purposes. Regardless of the fact that imaging of astrocyte function is still a young field and in need for development of suitable imaging ligands, recent studies have shown that inflammation and astrocyte activation are related to progression of MS. MS is a complex disease, which requires understanding of disease mechanisms for successful treatment. PET is a precise non-invasive imaging method for biochemical functions and has potential to enhance early and accurate diagnosis for precision therapy of MS. In this review we focus on modulation of different receptor systems and inflammatory aspect of MS, especially on activation of glial cells, and summarize the recent findings of PET imaging in MS and present the most potent targets for new biomarkers with the main focus on experimental MS research.

Keywords: multiple sclerosis, inflammation, neuroreceptors, positron emission tomography, precision medicine, microglia, astrocyte

#### Edited by:

Kirsten Harvey, University College London, UK

#### Reviewed by:

Hartmut Lüddens, University of Mainz, Germany Samaneh Maysami, University of Manchester, UK

\*Correspondence: Anna-Liisa Brownell abrownell@mgh.harvard.edu

Received: 08 May 2016 Accepted: 30 August 2016 Published: 15 September 2016

#### Citation:

Poutiainen P, Jaronen M, Quintana FJ and Brownell A-L (2016) Precision Medicine in Multiple Sclerosis: Future of PET Imaging of Inflammation and Reactive Astrocytes. Front. Mol. Neurosci. 9:85. doi: 10.3389/fnmol.2016.00085

## INTRODUCTION

Multiple sclerosis (MS) is the most common disabling neurologic disease of young people, afflicting approximately a quarter of million Americans (Anderson et al., 1992; Islam et al., 2006; Brody, 2012; Ransohoff et al., 2015). It occurs more in women than in men by a ratio of nearly 2 to 1, and it strikes most often between the ages of 20 and 40 (Compston and Coles, 2008). MS results from the immune-driven demyelination of the central nervous system (CNS), which leads to axonal damage and progressive loss of neurological functions (Sofroniew and Vinters, 2010; Malpass, 2012; Sofroniew, 2015). Based on clinical characteristics, MS pathology can be divided into three different disease courses: relapsing-remitting (RR), secondary progressive (SP), and primary progressive (PP) (Goodin, 2014). Initially, most MS patients exhibit a RR-MS disease course (Morales et al., 2006), experiencing heterogeneous symptoms such as ataxia, visual disturbances, paresthesia, and muscle weakness (Ellwardt and Zipp, 2014). However, eventually the majority of these patients develop SP-MS characterized by the progressive and irreversible accumulation of neurological disability (Lublin and Reingold, 1996). PP-MS patients have continuous disease progression from onset, without relapses or remissions (Morales et al., 2006; Lopez-Diego and Weiner, 2008).

Recent findings of the innate and the adaptive immune system of CNS have shaken up the classical view of MS as being strictly an autoimmune disease of the white matter (Weiner, 2008; Gandhi et al., 2010; Hemmer et al., 2015). The studies have revealed the important role of infiltrating immune cells from the periphery as well as the role of resident activated glial cells leading ultimately to the T cells and macrophages reaction against myelin (see **Figure 1**) (Frohman et al., 2006; Compston and Coles, 2008). These advances have switched the focus of MS research toward neurodegenerative aspects of the disease, occurring early in the pathological process (Kiferle et al., 2011). Despite the recent progresses in the field of MS therapeutic strategies there is no curative treatment for progressive MS (Lopez-Diego and Weiner, 2008; Derwenskus and Lublin, 2014). Therefore, identifying new specific biomarkers for MS could reveal new potential drug targets and diagnostic markers. Moreover, there is an unmet clinical need for methods to monitor different mechanisms of disease pathogenesis in MS patients, therefore advanced non-invasive molecular imaging technologies are needed to expand our understanding of the controversial aspects of the MS pathology (Kiferle et al., 2011; Jacobs and Tavitian, 2012).

Historically, MRI has overruled other imaging technologies in the diagnosis of MS (Traboulsee and Li, 2006; Barkhof and Filippi, 2009). The classical McDonald criterion for MS diagnosis requires objective dissemination of lesions in time and space (Filippi and Rocca, 2011). The literature analysis has shown that the sensitivity of MRI has been between 35 and 100%, and specificity has been between 36 and 92% depending on the research protocol (Schäffler et al., 2011; Tillema and Pirko, 2013). Overall, T2-weighted MRI is effective way to detect MS lesions, but because the signal reflects the water content, it does not provide reliable information about the myelin content (Ge, 2006; Poloni et al., 2011). T1-weighted imaging together with contrast agents such as gadolinium-DTPA has increased the lesion detection sensitivity, however, signal frequency is associated with the opening of the blood brain barrier (BBB; Lund et al., 2013). This leads to the problem that MRI can vary greatly in terms of sensitivity and specificity, especially in MS-related pathological pathways (Barkhof et al., 2009; Lövblad et al., 2010). Early diagnosis and treatment is effective for the therapy and decreases the financial burden of the disease (Noyes et al., 2011; Guo et al., 2014). The annual mean cost is around \$47,000 per MS patient, which arises to a national cost of about \$13 billion in US per year (Olek, 2012). The disease modifying treatments (DMTs, typically used first-line interferon betas and glatiramer acetate) have been available for the last 25 years and are estimated to account for one third of the total cost. Unfortunately, these treatments suppress the disease only for a few years (Hartung et al., 2015) and the spectrum of treatment options is narrow (Oh and O'Connor, 2015). Conventional MRI gives anatomical information from the progressed lesions in the brain of MS patients but lacks the power to provide target for drug discovery and more specific molecular markers when compared to imaging modalities like positron emission tomography (PET; Filippi et al., 2012; Matthews et al., 2012).

PET research field is emerging and the researchers have been successful in developing novel tracers for multiple different aspects of MS to enhance understanding the pathophysiology of the disease. In this review, we summarize PET imaging in MS research and introduce some of the most potent imaging targets and applications that have been successfully investigated in inflammation and which can be implemented especially to astrocyte activation related pathways, which are presently of high interest in MS research (Maragakis and Rothstein, 2006; Nair et al., 2008; Miljkovic et al., 2011; Nash et al., 2011; Mayo et al., ´ 2014).

#### PET IMAGING TECHNIQUES

PET imaging is based on detection of isotope labeled tracers, which emit beta radiation (see **Table 1**). These tracers are administered into the subjects to monitor underlying biological processes (Kiferle et al., 2011). The radioisotope undergoes positron emission decay and emits positron, which travels into surrounding tissue until it interacts with an electron and the annihilation process takes part (see **Figure 2**). The formed two photons travel in approximately opposite directions and can be detected with the imaging device as a coincidence pair. Each detected coincident forms a line of response (LOR) where the point of origin is the location of annihilation event. The combination of LORs can be used for reconstruction of images to provide 3-dimensional (3D) distribution of the radiolabeled tracer (Gambhir, 2002).

The clinical history of positron emission techniques started in 1952 when Gordon Brownell was able to localize brain tumors from patients (Brownell and Sweet, 1953). Further technical progression led to 3D tomographical positron imaging by 1971 (Pizer et al., 1971). However, even today PET suffers

from Criste et al. (2014) and Friese et al. (2014).

TABLE 1 | Properties of discussed positron emitting radio-isotopes.


Saha et al., 1994; Partridge et al., 2006; Miller et al., 2008; Jødal et al., 2012.

from high cost because the production of radiopharmaceutical agents increase the imaging cost compared to CT or MRI, both which became available later. After the development of [ <sup>18</sup>F]fluorodeoxy glucose ([18F]FDG) PET imaging received more significant clinical role especially in oncological diagnosis (Portnow et al., 2013). In MS however, MRI has been regarded as the golden standard in assessing patients (Filippi et al., 2012; Miller et al., 2012). Combined PET/MRI imaging with high specificity to MS lesions, would have a potential to become a practical tool in clinics to follow up the treatment of MS patients and increase cost-effectiveness. This approach could reveal an optimized treatment regimen; increase the treatment effectiveness and safety of patients, especially in early stage and patients with aggressive disease (Catana et al., 2013). When comparing these two imaging techniques, PET imaging has at least four major advantages over conventional MR imaging: (Massoud, 2003; Kiferle et al., 2011; Poloni et al., 2011; Jacobs and Tavitian, 2012; Miller et al., 2012; Torigian et al., 2013; Faria Dde et al., 2014; Jadvar and Colletti, 2014; Bodini et al., 2015) (1) Specific information of disease mechanism and molecular contributors, (2) Enhance development of new medicines and therapeutic targets, (3) Efficient allocation of new costly therapeutics and personalized medicine, and (4) Improved prognostic method for the MS patients.

Although the first clinical positron emission imaging studies were done over 60 years ago, the spectrum of applications of PET imaging is still limited due to the high cost and lack of validated traces and state-of-the-art facilities including availability of cyclotrons and automated radiopharmaceutical production laboratories (Jones et al., 2012). Complete knowledge about pharmacokinetic and pharmacodynamic properties of injected tracers can assure the correct interpretation of the images from preclinical and clinical studies. Overall, PET is an extremely powerful technology and the in vivo receptor occupancy can help answer many vital questions in the MS research (Matthews et al., 2012; Bodini et al., 2015). Furthermore, PET offers an opportunity for the detection of enzyme reactions, ligandreceptor interactions, cellular metabolism, cell proliferation, protein-protein interactions, as well as gene and cell therapy (Herschman, 2003; Ono, 2009; Thorek et al., 2013). The development of new PET tracers is challenging because the binding affinity and selectivity of the tracer have to be high and the dissociation must be fast enough to obtain the binding equilibrium in time frame of scan (1–2 h) (Hicks, 2006; Sharma et al., 2010). The tracer should penetrate the BBB, but too lipophilic compound might have strong non-specific binding (Liu et al., 2008). The optimal radiotracer should have minimum amount of unwanted metabolism and fast synthetic method (usually in single half-life of the radioisotope).

PET imaging systems have been developed also for small animals enhancing significantly basic research. Modern micro-PET instrumentation (resolution < 1 mm) is rapidly expanding the use of non-invasive PET imaging techniques in

basic research. These advances have been progressively translated to human studies (Herschman, 2003; Liang et al., 2007; Lancelot and Zimmer, 2010). PET imaging offers tools to evaluate a great variety of molecular aspects of MS and neurodegeneration in animal models as well as in clinics (see **Figures 3**, **4**).

### PRECLINICAL MODELS FOR MS

Several animal models are used to study mechanisms of disease pathogenesis relevant for MS (Furlan et al., 2009; Denic et al., 2011; Ransohoff, 2012). Experimental autoimmune encephalomyelitis (EAE) is the most vividly used animal model especially to study the inflammation aspects of MS. In this model, rodents are immunized with myelin antigens to activate peripheral antigen specific T-cells, which travel to CNS and induce formation of demyelinating lesion (Baxter, 2007; Constantinescu et al., 2011). Based on the hypothesis that viral infections may cause MS, virus-induced demyelination animal models are also used to study the disease (Gilden, 2005; Owens et al., 2011; Tselis, 2012). A disadvantage of this model is that the experimental disease manifests months after the initial infection (Olson et al., 2001; Fatima et al., 2010). Demyelination and spontaneous remyelination processes relevant to MS are predominantly studied using toxin-induced models (Blakemore and Franklin, 2008). The induction with copper chelating agent, Cuprizone [oxalic acid bis(cyclohexylidene hydrazide)], is one of the frequently used methods, since it is highly reproducible, relatively simple, induces fast demyelination, and the model has spontaneous remyelination after halting the toxin exposure (Torkildsen et al., 2008; Kipp et al., 2009). The focal demyelination lesions are commonly induced with ethidium bromide and lysolecithin (Woodruff and Franklin, 1999). The small size of the disease model increases the technical aspects for imaging technology. In the following chapters we will discuss the current tracers designed to detect the main pathological features of MS.

FIGURE 3 | 3-Nitropropionic acid (3-NP, a naturally occurring plant toxin and mycotoxin) could be involved to the development of MS. This study demonstrates the advantages of PET imaging where specific tracers can be used to reveal different time dependent neurochemical processes. In this case significant decrease of glucose metabolism imaged by 18F-FDG, decrease of dopamine D2 receptor function imaged by 11C-raclopride and decrease of dopamine transporter function imaged with 11C-CFT follow after 3-NP administration. Modified from Brownell et al. (2004).

## PET IMAGING OF AXONAL DEGENERATION

The complex network of conditions leading to neuroaxonal degeneration and neuronal loss contribute the permanent disability related to MS pathology (Friese et al., 2014). Even though during the earlier days of MS research, axonal loss was considered to be late-occurring, it is now discovered to happen also in the early stages of MS (Trapp and Nave, 2008; Trapp

and Stys, 2009). In the earlier phases of the disease axonal damage can occur acutely in the new inflammatory lesions. Whereas later in the disease progression axonal damage is usually related to chronic and demyelinated regions and there is only little if any active inflammation present (Criste et al., 2014). In addition, there is growing amount of evidence from both MRI and histological studies proposing that the axonal degeneration contributes to the development of clinical disability (Edgar and Nave, 2009; Nave, 2010). These interesting facts further highlight the need of specific markers for imaging of disease stage.

Currently, the most promising marker for neuronal integrity is benzodiazepine site on the GABA<sup>A</sup> receptor (Sigel and Buhr, 1997). Flumazenil, antagonist for benzodiazepine site, has been already labeled with <sup>18</sup>F and <sup>11</sup>C (Suzuki et al., 1985; The MICAD Research Team, 2004). Interestingly, in MS patients the axonal reduction has been demonstrated with the use of [11C]flumazenil (Barkhof et al., 2009). Furthermore, it has been shown that focal brain inflammation causes reduced GABA<sup>A</sup> mediated inhibition in neurons (Rossi et al., 2012). In addition, the inhibition is also induced in gray matter in the acute relapsing phases of MS (Rossi et al., 2012). Rossi et al. suggest that neurodegeneration in white and gray matter lesions are accompanied by a loss of GABA<sup>A</sup> receptors. PET could visualize this with radiolabeled flumazenil. However, this strategy remains yet to be tested.

Another cell type, which suffers from axonal loss during MS, is cholinergic neurons (D'Intino et al., 2005). Degradation of these neurons can, at least party, contribute to the cognitive impairment of the MS patients (Kooi et al., 2011). Interestingly, when assessing the acetylcolinesterase (AChE) activity by [ <sup>11</sup>C]MP4A (11C-methyl-4-piper-idinylpropionate), an inverse correlation with the activity of AChE and cognitive impairment was observed in MS patients (Virta et al., 2011). This result is contradicting the demonstrated positive response seen in MS patients with AChE inhibitors (Krupp et al., 2004; Tsao and Heilman, 2005). However, it has been hypothesized that the controversial results with increased AChE expression might be due to induction by inflammatory response in glial cells (Virta et al., 2011).

In addition, the reduction of glucose metabolism in the degenerated regions has shown correlation between disease activity, hypo-metabolism and specific cognitive functions during the MS progression (Bakshi et al., 1998). [18F]FDG has some valuable characteristics for monitoring cognitive and mental dysfunctions associated with MS (Paulesu et al., 1996; Zarei, 2003; Buck et al., 2012; Colasanti et al., 2014).

#### PET IMAGING OF DEMYELINATION

Demyelination, the pathological removal of myelin sheaths surrounding the axons, has been thought to be an integral part of axonal degeneration, as chronic CNS demyelination has been demonstrated to lead to axonal pathology and degeneration (Wilkins et al., 2010). However, these two events can happen independently from one another as axonal degeneration has been demonstrated to occur without myelin loss (Nave, 2010) and recently it has been demonstrated that the loss of myelin does not necessarily lead to axonal degeneration (Smith et al., 2013).

Classically in MS, demyelination is thought to cause the axonal dysfunction and disease-related pathogen conditions (Lucchinetti et al., 2000). On the other hand, spontaneous remyelination, executed by oligodendrocytes that mature from oligodendrocyte precursor cells, may occur following demyelination, presumably allowing a partial, if not complete, recovery from disability (Brück, 2005; Compston and Coles, 2008). Both adaptive and innate immune systems control the fine balance between demyelination and remyelination during MS and determine the outcome of the disease (Zhang et al., 2013). However, recently researchers have demonstrated early loss of both neurons and oligodendrocytes, leading to the question whether inflammatory demyelination is primary or secondary in the disease process of MS (Trapp and Nave, 2008). Remyelination is usually seen to occur in the early phases of the disease, whereas in the later phases it fails to recover the demyelinated areas leading to chronic demyelinated lesions (Chang et al., 2002; Kuhlmann et al., 2008).

Several tracers have been developed to target the β-sheet structures of intact myelin (Wu et al., 2006; Mallik et al., 2014). The first tracer, [11C]BMB (1,4-bis(p-aminostyryl)- [ <sup>11</sup>C]2-methoxy benzene), had significant off-target affinity toward white and gray matter (Stankoff et al., 2006). Some of these downsides have been overcome with Congo red derivatives (e.q. [11C]CIC, Case Imaging Compound) and thioflavine-T derivatives (e.q. [11C]PIB, N-methyl-[11C]2-(4′ methylaminophenyl)-6-hydroxybenzothyazole). Moreover, these tracers have more reliable production and BBB penetration (Wang et al., 2009; Stankoff et al., 2011). Recent comparisons between these series and new [11C]MeDAS, (N-[11C]methyl-4,4′ diaminostilbene) tracers prefer the latter compound as the most promising ligand so far to detect MS-like lesions and spinal cord imaging (de Paula Faria et al., 2014b). [11C]MeDAS has been successfully used to image acute focal neuroinflammation in the brain, lyso-phosphatidyl choline induced focal demyelination in the spinal cord and EAE rodent models of MS (Wu et al., 2013). Furthermore, [11C]MeDAS was also able to highlight both demyelination and remyelination processes in cuprizone mouse model (de Paula Faria et al., 2014b). Interestingly, the uptake of [ <sup>11</sup>C]MeDAS was not interfered by inflammation (Wu et al., 2013). The current literature suggest that [11C]MeDAS is the most preferred PET agent so far to highlight the lesions as well as the myelin content in the spinal cord in motor disability related MS (de Paula Faria et al., 2014a). To this point the only PET tracer used to image myelin in MS patients is [11C]PIB, a tracer widely utilized to visualize β-amyloid plaques in Alzheimer's disease (Stankoff et al., 2011; Zhang et al., 2014).

Altogether, PET imaging of myelin integrity shows great potential in animal models of MS. It is interesting to validate these methods in patients, especially now that new remyelination therapies are introduced in clinical trials (Brugarolas and Popko, 2014).

#### PET IMAGING OF MICROGLIAL ACTIVATION

Neuroinflammation is a common characteristic of numerous neurodegenerative disorders, including MS (Glass et al., 2010). Reactive states of astrocytes (astrogliosis), and microglia (microgliosis), as well as the infiltration of the lymphocytes are the hallmarks of neuroinflammation (Carson et al., 2006). Although factors inducing inflammation vary between CNS related diseases, there is evidence that convergence mechanisms are accountable for the sensing, transduction, and amplification of inflammatory processes that eventually lead to the production of neurotoxic mediators (Glass et al., 2010). In fact, neuroinflammation is a highly dynamic and complex process combining local and systemic reactions of multiple cell types, chemical signals, and signaling pathways to adaptive response for restoring tissue homeostasis (Medzhitov, 2008; Aguzzi et al., 2013; Naegele and Martin, 2014). In the following, we will discuss the PET tracers used to visualize microglial and astrocytic activation.

Microglia are of mesenchymal origin and constantly monitor the extracellular environment as well as interact closely with astrocytes and neurons (Yamasaki et al., 2014; Michell-Robinson et al., 2015). As macrophages in the periphery, microglia are the first line of defense against infections or insults in the CNS (Olson et al., 2001; Hanisch and Kettenmann, 2007; Nau et al., 2014). Upon activation, microglia acquire an amoeboid appearance and secrete pro-inflammatory molecules such as interleukin 1β, interferon γ, and tumor necrosis factor-α (TNFα) (Boche et al., 2013): a classically activated M1 state (Mills et al., 2000; Martinez and Gordon, 2014). The aim of the proinflammatory reaction is to clear the hazardous material and correct the inflicted damage (Gordon, 2003; Martinez et al., 2009). Usually, the pro-inflammatory reaction is down-regulated by the anti-inflammatory molecules (Tambuyzer et al., 2009; Scheller et al., 2011). In addition to pro-inflammatory molecules, microglia can release trophic and anti-inflammatory factors such as interleukins 4 and 10 as well as insulin-like growth factor 1 (Cherry et al., 2014). These factors are aimed to contribute to the repair and limitation of the inflammation (Mantovani et al., 2004; Hanisch and Kettenmann, 2007; Michelucci et al., 2009). Astrocytes and inflammatory T-cell subsets surrounding microglia influence the state of microglia, and determine whether they are releasing pro- or anti-inflammatory factors (Shih, 2006; Goverman, 2009; Mayo et al., 2012, 2014; Quintana et al., 2014).

The role of microglial activation in MS progression has remained enigmatic (Correale, 2014). However, several theories have been offered. The first theory suggests that inflammatory processes similar to those observed in RR-MS cause the brain damage (Kutzelnigg and Lassmann, 2014). However, during the progressive disease stages, a microenvironment is created within the brain favoring the homing and retention of inflammatory cells, finally resulting in the failure of disease-modifying therapies (Frischer et al., 2009). According to the second theory, MS starts out as an inflammatory disease and after several years, neurodegeneration, a process autonomous of inflammatory response, becomes the mechanism responsible for progression of the disease (Meuth et al., 2008; Kutzelnigg and Lassmann, 2014). Finally, MS could be seen primarily as a neurodegenerative disease, where inflammation occurs as a secondary response, augmenting and modifying progressive stages (Kassmann et al., 2007; Fitzner and Simons, 2010). Needless to say, these theories are not mutually exclusive. Furthermore, it has been postulated that the lack of understanding the exact microglial function during course of MS, has led to the absence of therapies for SP-MS (Correale, 2014). Altogether, this clearly demonstrates the need for a consensus and better understanding of microglial activation, which can only be achieved by using appropriate methodology.

Inflammation related PET studies in MS are traditionally focused on monitoring changes in glucose metabolism and the presence of activated microglia/macrophages in sclerotic lesions (Schiepers et al., 1997; Kiferle et al., 2011). [18F]FDG was recently used to evaluate the inflammation in the spinal cord in the EAE rat model (Buck et al., 2012). However, the basal uptake of glucose is elevated in the brain reducing the usability of [18F]FDG as a marker for brain lesions. Results of different stage patients indicate that [18F]FDG could be used to classify white matter lesions as either acute (hyper metabolism) or chronic (hypo metabolism) based on the glucose consumption (Paulesu et al., 1996; Dimitrakopoulou-Strauss et al., 2004; Buck et al., 2012). It is obvious that more specific markers are required to image the inflammation related metabolism in MS.

The majority of current PET tracers used to detect microglial activation utilize the expression of the peripheral benzodiazepine receptor (PBR), also known as the translocator protein TSPO (18 kDa) (Ryu et al., 2005; Ching et al., 2012). Translocator protein is expressed in the outer mitochondrial membrane. It was assumed to contribute through the cholesterol transportation into mitochondria regulating the rate of the synthesis of neurosteroids. However, these views have recently been challenged (Rupprecht et al., 2010; Selvaraj and Stocco, 2015). Gene-expression studies in the brain of rodents, primates, and humans have shown that TSPO expression is nearly absent in microglia patrolling the intact CNS parenchyma but rapidly increases in inflammation (Venneti et al., 2008; Ching et al., 2012). TSPO is highly expressed in activated microglia, in the choroid plexus and in reactive astrocytes, but its expression is globally low in the normal brain (Chauveau et al., 2008; Banati et al., 2014; Liu et al., 2014). These findings indicate that TSPO is a biomarker and an attractive target for the imaging microglial activation and reactive gliosis in cerebral inflammation (Rupprecht et al., 2010; Ching et al., 2012).

The isoquinoline carboxamide derivate PK11195 (N-butan-2 yl-1-(2-chlorophenyl)-N-methylisoquinoline-3-carboxamide), a nonbenzodiazepine ligand specifically binding to TSPO, has been widely used for its functional characterization and for the identification of its cellular origin in brain tissue (Banati et al., 1997, 2000; Chauveau et al., 2008). The issues regarding sensitivity and specificity of traditional PK11195 has been discussed (Venneti et al., 2008; Dickens et al., 2014; Boutin et al., 2015). Fortunately, recently developed radioligands such as DPA-714 (James et al., 2008; Chauveau et al., 2009), PBR28 (Imaizumi et al., 2007), PBR111 (Van Camp et al., 2010), SSR18075 (Chauveau et al., 2011), CLINME (Arlicot et al., 2008; Van Camp et al., 2010), and GE-180 (Dickens et al., 2014) have demonstrated better binding potency and bioavailability compared to the classical PK11195 and could overcome the problems of the classical tracers in MS and its models. A number of other TSPO targeting tracers have been developed to study the inflammation including but not limiting to DAA1106, FE-DAA1106, DPA-713, and vinpocetine, and reviewed by (Chauveau et al., 2008; James et al., 2008; Winkeler et al., 2010; Ciarmiello, 2011; Kiferle et al., 2011). Microglial activation was demonstrated in clinical MS studies with [11C]PK11195, unfortunately only in a limited number of patients (Banati et al., 2000; Debruyne et al., 2003; Versijpt et al., 2005; Vas et al., 2008). Radiotracer binding was increased in areas of acute and relapse-associated inflammation detected by classical Gd-DTPA enhanced T1-weighted MRI imaging (Rissanen et al., 2014). Interestingly, a significant increase in [11C]PK11195 binding was observed in activated microglia outside the histopathologically or MRI defined borders of MS plaques in both cerebral central gray-matter areas, which are not normally reported as sites of pathology in MS, as well as in normal appearing white matter (Banati et al., 2000; Debruyne et al., 2003). Unfortunately, 2nd generation TSPO targeting agents suffer from unexpected low binding status in over 30% of the population, which limits their use in clinics and demands genetic testing of the TSPO polymorphism. However, clinical studies have shown increased uptake with <sup>18</sup>F-PBR111 and <sup>11</sup>C-PBR28 in white matter lesions but not with all 2nd generation compounds like <sup>18</sup>F-FEDAA1106. Additional studies are required to further investigate the specificity of these radiotracers for activated microglia over other activated glial cells. Overall, imaging of microglial activation in MS patients may serve as a complementary biomarker for disease progress (Abourbeh et al., 2012; Airas et al., 2015).

The type 2 cannabinoid receptor (CB2R) is part of the human endocannabinoid system and is involved in both central and peripheral inflammatory processes (Ehrhart et al., 2005; Pacher, 2006; Chiurchiù et al., 2014). CB2R can be found in immune cells, such as macrophages, perivascular T lymphocytes, astrocytes and reactive microglia, and it is thought to mediate antiinflammatory as well as immunomodulatory effects (Docagne et al., 2008; Rodgers et al., 2013). 2-oxoquinoline and oxadiazolyle derivatives have been synthesized and radiolabeled with <sup>11</sup>C and <sup>18</sup>F, representing promising candidates for brain imaging in mice (Evens et al., 2009; Teodoro et al., 2013; Slavik et al., 2015). CB2R is almost undetectable in a healthy brain, whereas it is expressed in the activated glial cells (Stella, 2004; Cabral et al., 2008; Atwood and Mackie, 2010). This demonstrates that the effective PET ligands for CB receptors have the potential to act as biomarkers in the studies of pathophysiology of MS (Sanchez-Pernaute et al., 2008; Evens et al., 2009; Horti et al., 2010; Turkman et al., 2011; Vandeputte et al., 2011). In addition, new microglial targets, like P2X purinoceptor 7 (P2X7; Yiangou et al., 2006; Monif et al., 2009) and matrix metalloproteinases (Wagner et al., 2007; Iwama et al., 2011), have been explored for imaging of MS.

Overall, the benefits of PET contribute to the understanding of personalized status of MS patients, disease profiling, prognosis, and response, which are all combined in precision medicine. Specific biomarkers are the backbone for capturing the different aspects of MS heterogeneity, which could be useful for diagnosis, treatment stratification, and personalization of the therapeutic approach. Simplified, the precision medicine aims to provide the right drug with the right dose for the right indication in the right patient at the right time. Such as the case with the current 2nd generation TSPO markers, precision medicine relies on variability of genes, environment and life style of each person rather than on the data from large clinical trials. The customization of the treatment is based on the characterization of the genotype and phenotype induced effects on imaging in the individual patient. Biomedical imaging offers a great tool for mapping data from biomarkers, genomics, and physiology. There is a great interest for the monitoring of microglial activation in MS. However, the recent results with TSPO ligands suggest that the reactive astrocytes might increase the signal levels in MS (Lavisse et al., 2012). Since the role of reactive astrocytes in MS is recently of great interest, more specific markers are needed for reliable imaging of neuroinflammation (Rostami and Ciric, 2014; Zeis et al., 2015).

#### IMAGING ASTROCYTE ACTIVATION

Astrocytes are one of the most abundant cell types in the CNS. They have complex function ranging from supporting the surrounding neurons to the regulation of synaptic activity and BBB integrity (Sofroniew and Vinters, 2010). Although astrocytes are not immune cells per-se they can in specific conditions, such as in CNS inflammation, exert both pro- and anti-inflammatory effects on microglia (Min et al., 2006; Farina et al., 2007; Sofroniew, 2015).

Astrocytes were regarded to be non-participating bystanders in MS, responding secondarily to insults by undergoing astrogliosis and producing a glial scar (Brosnan and Raine, 2013). However, since the T-cell mediated immunity has been strongly associated with MS there are several plausible means by which astrocytes could contribute to autoimmunity. Astrocytes may facilitate immune cell extravasation into the CNS by releasing chemokines. They can modulate the activity of innate immune cells, such as microglia and inflammatory monocytes recruited to the CNS, by boosting their ability to promote neurodegeneration. Finally, astrocytes also have direct neurodegenerative functions mediated by the production of TNFα and nitric oxide (NO). However, these actions can also represent potential mechanisms by which astrocytes could reduce inflammation to promote remyelination (Claycomb et al., 2013).

Activation of glial cells is a common feature of MS as discussed earlier. Acetate is reported to accumulate into astrocytes and the [11C]acetate accumulation is increased in MS lesions (Takata et al., 2014). However, brain uptake of [11C]acetate is insufficient for obtaining a quantitative image of astrocytes' oxidative metabolism (Okada et al., 2013). To overcome this drawback benzyl [11C]acetate has been synthesized (Okada et al., 2013). Although the quantitative measurement remains under development, acetate is specific for astrocyte lipid metabolism (Brekke et al., 2015) and could serve as a marker for activated astrocyte metabolism in MS (Takata et al., 2014). In addition, <sup>18</sup>F labeled derivative of acetate could increase the signal to noise ratio compared to <sup>11</sup>C analog. It is expected that this tracer will be used in MS (Ponde et al., 2007).

One critical function of astrocytes is acting as sentinels and monitoring the BBB, a complex barrier composed of endothelial cells, astrocytes, pericytes, and myeloid cells such as perivascular macrophages and mast cells (Abbott et al., 2006). BBB functions as an anatomical mechanism for the highly selective passage of water, ions, nutrients, and cells from peripheral circulation into and out of the brain parenchyma (Abbott et al., 2006; Daneman and Rescigno, 2009; Larochelle et al., 2011). Under inflammatory conditions the BBB opens and it enables higher leukocyte passage into the CNS (Claycomb et al., 2013). Astrocytes play a critical role in shielding and protecting the CNS under inflammatory conditions (Voskuhl et al., 2009). Furthermore, astrocyte ablation has been shown to cause enhanced monocyte, but not T-cell, migration into the CNS (Toft-Hansen et al., 2011). To date, there is no clinically relevant PET tracer for BBB integrity, although several candidates have been proposed [13N]glutamate, [82Rb]Cl, or <sup>68</sup>Gallium-ethylene-diamine-tetra-acetic acid (EDTA) (Saha et al., 1994; Wunder et al., 2012).

Imaging of astrocyte function is still a young field and it needs development of suitable imaging ligands. Astrocytes are involved in several neurological diseases and the main obstacle using imaging techniques has been the lack of proper tracers.

#### TARGETS FOR PET IMAGING IN MS

The recent increased availability of PET tracers to assess activated glial cells, disease pathology, and signaling pathways give PET a promising role in MS research. Since the underlying mechanisms of neurodegeneration and regeneration are still poorly understood the non-invasive techniques will enhance understanding these processes to develop better drug candidates, early diagnosis, and reliable monitoring of the treatment response. Several possible targets for PET imaging in MS are discussed in this section. These candidates may serve as more specific targets and may reveal some of the missing links in MS treatment and pathology, especially in the glial cell mediated actions.

Neuroinflammation is a dynamic and complex adaptive response process, which involves multiple cell types and various signaling routes, pathways, and receptors (Singhal et al., 2014). As discussed earlier, neuroinflammation can be imaged in MS. However, new tracers are needed to gain practical importance in clinics. The greatest potential may lay in the imaging of the dynamic interplay between neuroinflammation and the molecular mechanisms that contributes to the disease progression. The recent findings support three major neuroinflammation components in MS: astrogliosis, cytokine elevation, and significant changes in specific proteins, which offer a great variety of specific targets for imaging purposes.

TNF-α is associated with self-propagation of neuroinflammation and the expression of TNF-α is elevated in MS patients (Rossi et al., 2014). Microglia, inflammatory monocytes recruited to the CNS and astrocytes are major sources of TNF-α in CNS, interestingly proposing TNF-α expression as a marker in MS (Welser-Alves and Milner, 2013). PET tracers, like [64Cu]DOTA-etanercept and [64Cu]pegylated dimeric c(RGDyK), have been developed to target TNF-α in both acute and chronic inflammation in mice (Cao et al., 2007). TNF-α may be a target for MS imaging in the future. Overall, cytokines are highly related to oxidative stress in the brain (Di Penta et al., 2013). The expression of inducible nitric oxide synthase (iNOS) is increased in MS lesions, increasing generation of NO as well as reactive nitrogen species like peroxynitrite (Kröncke et al., 1998; Ortiz et al., 2013). The accumulation of these molecules induces lipid peroxidation, resulting in damage to DNA and neuronal degeneration (Haider et al., 2011).

In the healthy CNS tissues, the expression levels of iNOS are low but become highly expressed in astrocytes and neurons during inflammation (Saha and Pahan, 2006). In chronic pathology the reactive nitrogen species produced by iNOS are not efficiently eliminated, which leads to cellular dysfunctions (Fulda et al., 2010). The number of tracers for iNOS is minimal and the current [18F]NOS (6-(1/2)(2-[18F]fluoropropyl)-4-methylpyridin-2-amine) needs further modification and improvement. Importantly, the feasibility of iNOS PET imaging has been demonstrated in human inflammation (Herrero et al., 2012; Huang et al., 2015). In addition, active iNOS enzyme has been demonstrated in astrocytes in both acute and chronic active MS lesions and might therefore be an interesting target for imaging purposes (Liu et al., 2001).

The expression of another proinflammatory cytokine mediator, cyclooxygenase-2 (COX-2), is extensively increased in MS lesions and it has been tightly linked to increased iNOS expression (Rose et al., 2004). Furthermore, COX-2 expression was found in the cells expressing microglial marker, highlighting the importance of immune-derived cells. COX-2 has also been suggested to act as a link between neuroinflammation and glutamate mediated neuronal excitotoxicity (Kelley et al., 1999). These facts clearly indicate a need for methods to detect COX-2 expression. PET tracers for COX-2 have been developed, but the in vivo imaging properties have not been very effective (de Vries et al., 2003, 2008; Takashima-Hirano et al., 2011; Ji et al., 2013). The most promising COX-2 tracer so far is [ <sup>11</sup>C]Rofecoxib (4-(4-methylsulfonylphenyl)-3-phenyl-5Hfuran-2-one), demonstrating in vitro usability, but lacking necessary affinity for in vivo studies (Ji et al., 2013). Nevertheless, cyclooxygenases is presently an important target for PET tracer development.

Besides stimulating production of reactive oxygen species, cytokines are known to modulate the lipid metabolism and increase the production of neurodegeneration promoters such as eicosanoids and ceramides (Adibhatla and Hatcher, 2007). As previously mentioned, acetate is converted into fatty acid by acetyl-CoA synthase and [11C]acetate PET has proven useful for imaging in several diseases (Grassi et al., 2012). In addition, acetate is preferentially absorbed into astrocytes by the monocarboxylate transporter, which is overexpressed in MS (Nijland et al., 2014). Moreover, bioactive lipids exert significant effects on inflammation during autoimmunity targets or regulators of the immune response (Rinaldi et al., 2009). In addition, the appearance of cytosolic lipid synthesis is one the corner stones of macrophage foam cell formation (Matthäus et al., 2012). The intracellular concentrations of different individual lipids or the receptors involved the synthesis of particular bioactive lipids could reveal novel aspects of the disease progression (Mayo et al., 2014). Recently β-1,4 galactosyltransferase 6 (B4GALT6) was found to promote astrocyte activation and neuroinflammation during chronic EAE. The lactosylceramide (LacCer) synthesized by B4GALT6 in astrocytes controls the production of chemokines and cytokines, such as CCL2 and GM-CSF, which regulates the recruitment and activation of inflammatory monocytes and microglia and clearly highlights the importance of a specific lipid profile for disease progression (Mayo et al., 2014).

In summary, comprehensive profiling of lipid metabolism and the BBB function are likely to reveal new targets for therapeutic intervention in MS as well as for other neurological disorders where astrocyte activation contributes to disease pathology (Neu and Woelk, 1982; Pannu et al., 2005; Adibhatla and Hatcher, 2007; Wheeler et al., 2008; Kooij et al., 2012; Prüss et al., 2013). The imaging of specific bioactive lipids, receptors or enzymes that are involved in their synthesis may be novel targets for PET imaging.

### BIOMARKERS FOR THE EARLY PHASES OF MS

In the search of better treatments for MS, cerebrospinal fluid (CSF) biomarkers have been used to identify high risk MS patients as well as patients with other neuronal disorders. Recently, high levels of astrocyte derived chitinase 3-like protein 1 (CHI3L1) were associated with the strong prediction of MS. This finding further demonstrates the increased importance of astrocyte activation and the specific role of astrocyte as a source for biomarkers in MS, already at the early disease phase.

The activated lipid metabolism in astrocytes demands increased acetate and lipid transportation (Lev, 2012). ATP and glutamate stimulation can significantly enhance the dynamin-independent endocytosis and their receptors control the microglial physiology and pathology (Jiang and Chen, 2009). For example ATP related purinergic receptors control microglial cytokine release among several other functions (Sperlágh and Illes, 2007). Moreover, purinergic pathways regulate neuroinflammation (Burnstock, 2008). The increasing evidence suggests that the P2X7 receptor is an interesting neuroinflammation associated molecular target (Lister et al., 2007; Monif et al., 2009; Gandelman et al., 2010). PET tracers have been developed to image P2X7 receptor, [11C]A-740003 (N-[1-[[(Cyanoamino)(5-quinolinylamino)methylene]amino]- 2,2-dimethylpropyl]-3,4-dimethoxybenzeneacetamide) and [ <sup>11</sup>C]GSK1482160 ((S)-N-(2-chloro-3-(trifluoromethyl)benzyl)- 1-[11C]methyl-5-oxopyrro-lidine-2-carboxamide; Janssen et al., 2014; Gao et al., 2015). Purinergic system might serve as a sensitive target for MS imaging.

Furthermore, the adenosine receptors, whose expression is modulated by microglial activation, moderate immune function (Haskó et al., 2008; Orr et al., 2009; Domercq et al., 2013; Luongo et al., 2014). Especially A2A receptors are up-regulated during inflammation (Rissanen et al., 2013). It is clear that adenosine signaling play a significant role in MS as a neuromodulator and the clinical studies with [11C]TMSX (7-methyl-[11C]-(E)- 8-(3,4,5-Trimethoxystyryl)-1,3,7-trimethylxanthine) PET will likely open new perspective to develop new tracers to this target in the future (Rissanen et al., 2015).

In addition, the cholinergic system shows decreased function in MS patients (Kooi et al., 2011). PET imaging studies of cholinergic activity may define which patient will respond to the treatment which will further increase the knowledge of MS. A similar approach has been already used in Alzheimer's disease using radiolabeled choline derivatives and these techniques could be easily transferred to MS clinical research (Volkow et al., 2001; Rinne, 2003; Kooi et al., 2011).

The cannabinoid receptors CB2 are expressed in very low levels in a healthy brain, but the expression increases during microglial activation (Benito et al., 2007). CB2 is an interesting target for PET imaging in MS models especially with [11C]A836339 (2,2,3,3-Tetramethylcyclopropanecarboxylic acid [3-(2-[11C]methoxyethyl)-4,5-dimethyl-3H-thiazol-(2Z) ylidene]amide) and [11C]NE40. Recently there has been great progress in developing new tracers for this target (Docagne et al., 2008; Horti et al., 2010; Evens et al., 2011, 2012; Slavik et al., 2015; Yrjölä et al., 2015).

During inflammation microglia will release glutamate in response to the production of reactive oxygen species (ROS; Bal-Price and Brown, 2001; Brown and Neher, 2010; Takaki et al., 2012). The cysteine-glutamate exchange modulates the release of ROS and cytokines which impairs the function of glutamate transporters and leads to increased extracellular glutamate levels as well as excitotoxicity (Rao et al., 2003; Matute et al., 2006). Metabotropic glutamate receptors (mGluRs) are transmembrane proteins that are expressed in glial cells and play a pivotal role in cell function and glial-neuronal co-operation (Kritis et al., 2015). Immunohistochemical analyses have revealed that mGluR5 is expressed in reactive astrocytes surrounding the MS lesion site and the expression is higher than in non-activated astrocytes (Geurts et al., 2003). In addition, the activation of mGluR5 reduced the microglial activation in an inflammation model (Byrnes et al., 2009; Loane et al., 2009). During the last 15 years the subtype selective allosteric modulators have been identified for different mGluRs (see **Figure 5**; Zhang and Brownell, 2012). Many PET tracers have been synthesized by radiolabeling the derivatives of MPEP and MTEP and to date over 15 mGluR5 targeting PET ligands have been reported (Mu et al., 2010; Zhang and Brownell, 2012). Tracers like [18F]FPEB ((3-[18F]Fluoro-5- (2-pyridinylethynyl)benzonitrile) have been already developed for automated synthesis and evaluated in humans (Lim et al., 2014).

Monoamine oxidase type B located in the outer membrane of mitochondria and is expressed in astrocytes, where its activity is increased in neurodegenerative diseases (Mallajosyula et al., 2008; Veitinger et al., 2014). It catalyzes the deamination reaction thus modulating neurotransmitter concentrations and has been

FIGURE 5 | Coronal and sagittal sections of fused PET and CT images in 10 days old pups of mice. PET studies using [18F]FPEB show enhanced mGluR5 expression in the brain of the pups, whose mothers were injected with LPS compared to saline injection (control). Coronal slices show highest accumulation in the hippocampal area of the mouse, whose mother had LPS administration. Sagittal images show spine based on CT images and high accumulation of [18F]FPEB in the brain and gut. Modified from Arsenault et al. (2014).

a major target for drug development, especially in movement related diseases (Talati et al., 2009; Deftereos et al., 2012). The <sup>11</sup>C-L-deprenyl indicates increased monoamine oxidase type B content in reactive and proliferating astrocytes in AD (Gulyás et al., 2011). Results from the MS patients studies are likely to be published soon (Hurley, 2015).

The above exploration shows that the combined PET imaging of activated microglia and astrocytes is presently of special interest in MS research.

## PET AS A TOOL FOR PRECISION MEDICINE IN MS

More specific features of MS lesions have been described in parallel with the identification of body fluid markers such as CHI3L1 and B4GALT6. Recent progress with biomarkers and imaging tracers suggests that precision medicine is becoming a reality in MS. The prevalence of MS is increasing and there is relatively little data available to personalize the treatments and increase the cost effectiveness. Sophisticated tools are needed to handle the complex data to obtain more detailed insight of the clinical status of the patient's condition. The combined information from various biomarkers and imaging studies can be used to predict the disease evolution individually. PET imaging can provide precise data for the cross-roads of multiple fields, like biomedical imaging, pharmacology, neurology, genomics etc. Achieving precision medicine in MS requires high quality data, large samples, and consistent interdisciplinary approach.

## CONCLUSION

Inflammation and glial activation play an important role in numerous neurodegenerative diseases, such as Alzheimer's disease, Parkinson disease, amyotrophic lateral sclerosis and MS. Although factors inducing inflammation vary between diseases, there is evidence of greatly converging mechanisms for the sensing, transduction, and amplification of inflammatory processes that eventually lead to the production of neurotoxic mediators. PET imaging provides a powerful method for dynamic imaging of these events. The full potential of PET is not yet recognized, mainly due to the lack of validated tracers; the complicated and costly process of validating new tracers needs partnerships, human resources, expertise, funding, and access to patients, but it is something that needs to be focused on to obtain the essential information of the biological processes in disease pathology. This will ultimately produce more reliable diagnosis, better treatments and effective prevention methods for MS. The role of PET imaging will increase in clinics, when onsite cyclotrons, the development of new tracers, and imaging equipment become available.

[ <sup>18</sup>F]-FDG is still the most extensively used PET imaging tracer for inflammation even though it tends to produce controversial results. New imaging tracers for TSPO ([11C]PK11195, [11C]PBR28, etc.) have gained a great interest for detection of inflammation and evaluation of therapy. Using these new tracers, PET imaging has greatly improved our

#### TABLE 2 | Examples of PET tracers in MS research.


#### TABLE 2 | Continued


#### TABLE 2 | Continued


#### TABLE 2 | Continued


understanding of the mechanism of inflammation and increased the diagnostic specificity and accuracy of inflammation. As summarized in **Table 2**, various radiopharmaceutical approaches have been developed for PET imaging to detect inflammation, including biomarkers targeting to specific receptors and lipid metabolism.

So far, the clinical PET studies in MS are limited to evaluation of two biological processes: glucose metabolism and inflammation. It is clear that the use of combined PET/MR imaging is increasing also in MS research. One of the main interests is to develop combined imaging markers and methods for MS pathology to stage, cell type and record activity related changes in lesions. However, presently PET imaging is relatively expensive and it also requires sophisticated quantification, which demands special software and skilled operators. MS is a complex disease, which remains difficult to treat before more specific disease mechanisms are revealed. PET research community is looking for the first ligands to be recommended for routine clinical practice in MS diagnosis and follow up of therapy. PET has already shown to be one of the most sophisticated, sensitive, reliable, effective, and safest tools for the monitoring of several cancers in clinics and there is no reason why it could not be same in the neurodegenerative disorders as well. Clinical imaging and research modalities should be combined to expand the knowledge of clinical findings, genetics, phenotyping, pharmacology, and drug targeting. Advanced imaging technologies, including PET, could be used to reveal the causes of MS rather than concentrating on correlations. MS is a complex and heterogeneous disease, which could benefit from precision medicine in the future. The genomic approach can be used to individualize the imaging data as presently done with 2nd generation TSPO markers ([11C]PBR28 etc.). Astrocyte activation and their ability to modulate the complex neuronal network and inflammation related pathways have a great potential to reveal disease stage specific markers for personalized medicine. Despite the astrocyte related research in MS is still in early stages, and the recent promising results suggest new techniques to diagnose, monitor and treat this cruel disease. The combined pathogenic characteristics of MS are still unknown and the key to prevent and cure this devastating disease is still waiting for discovery.

#### AUTHOR CONTRIBUTIONS

PP did literature search, prepared tables and participated writing. MJ participated literature search, writing and composing the manuscript. FQ did critical evaluation of the manuscript. AB participated writing, preparation of figures, and final evaluation of the manuscript.

#### REFERENCES


#### ACKNOWLEDGMENTS

AB was supported by the grant NIBIB-R01EB012864. PP was supported by the Orion Farmos Research Foundation and Kuopio University Foundation, Finland. MJ was supported by the Sigrid Juselius Foundation, Finnish MS-Society, Orion Farmos Research Foundation and Saastamoinen Foundation, Finland.


model of neuroinflammation. PLoS ONE 8:e54722. doi: 10.1371/journal.pone. 0054722


lesions. Am. J. Pathol. 158, 2057–2066. doi: 10.1016/S0002-9440(10) 64677-9


Eur. J. Nucl. Med. Mol. Imaging 42, 1081–1092. doi: 10.1007/s00259-015- 3043-4


at: http://doi.wiley.com/10.1002/14651858.CD010386.pub2 (Accessed May 20, 2015).


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

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

# Function Over Form: Modeling Groups of Inherited Neurological Conditions in Zebrafish

Robert A. Kozol <sup>1</sup> \*, Alexander J. Abrams <sup>2</sup> , David M. James <sup>1</sup> , Elena Buglo<sup>2</sup> , Qing Yan<sup>1</sup> and Julia E. Dallman<sup>1</sup> \*

<sup>1</sup> Department of Biology, University of Miami, Coral Gables, FL, USA, <sup>2</sup> Department of Human Genetics, John P. Hussman Institute for Human Genomics, Dr. John T. Macdonald Foundation, University of Miami, Miami, FL, USA

Zebrafish are a unique cell to behavior model for studying the basic biology of human inherited neurological conditions. Conserved vertebrate genetics and optical transparency provide in vivo access to the developing nervous system as well as high-throughput approaches for drug screens. Here we review zebrafish modeling for two broad groups of inherited conditions that each share genetic and molecular pathways and overlap phenotypically: neurodevelopmental disorders such as Autism Spectrum Disorders (ASD), Intellectual Disability (ID) and Schizophrenia (SCZ), and neurodegenerative diseases, such as Cerebellar Ataxia (CATX), Hereditary Spastic Paraplegia (HSP) and Charcot-Marie Tooth Disease (CMT). We also conduct a small meta-analysis of zebrafish orthologs of high confidence neurodevelopmental disorder and neurodegenerative disease genes by looking at duplication rates and relative protein sizes. In the past zebrafish genetic models of these neurodevelopmental disorders and neurodegenerative diseases have provided insight into cellular, circuit and behavioral level mechanisms contributing to these conditions. Moving forward, advances in genetic manipulation, live imaging of neuronal activity and automated high-throughput molecular screening promise to help delineate the mechanistic relationships between different types of neurological conditions and accelerate discovery of therapeutic strategies.

Edited by: Robert J. Harvey, UCL School of Pharmacy, UK

#### Reviewed by:

Ellen J. Hoffman, Yale University, USA Hiromi Hirata, Aoyama Gakuin University, Japan

#### \*Correspondence:

Robert A. Kozol robkozol@bio.miami.edu Julia E. Dallman Jdallman@bio.miami.edu

Received: 05 April 2016 Accepted: 23 June 2016 Published: 07 July 2016

#### Citation:

Kozol RA, Abrams AJ, James DM, Buglo E, Yan Q and Dallman JE (2016) Function Over Form: Modeling Groups of Inherited Neurological Conditions in Zebrafish. Front. Mol. Neurosci. 9:55. doi: 10.3389/fnmol.2016.00055 Keywords: zebrafish, disease modeling, autism spectrum disorder, intellectual disability, schizophrenia, ataxia, Charcot-Marie tooth, hereditary spastic paraplegia

#### INTRODUCTION

As genetically tractable vertebrates (Streisinger et al., 1981) that share with humans many pathways targeted by FDA approved pharmaceuticals (Renier et al., 2007; Rihel et al., 2010) zebrafish are a powerful model for inherited neurological conditions, both in terms of delineating underlying mechanisms and developing therapeutic strategies. Their small size and optical transparency enable in vivo visualization of cell- and systems-level processes throughout early developmental stages (McLean and Fetcho, 2011; Rasmussen and Sagasti, 2016) while, precocious development of quantifiable behaviors (Brustein et al., 2003) and reduced complexity of the zebrafish nervous system (Goulding, 2009) simplify functional studies of neural circuits. These advantages combined with conserved vertebrate genetics lend themselves to keeping pace with the extraordinary discovery rate of genetic mutations that cause inherited neurological conditions in humans. With the sequencing of the human genome, the formation of worldwide consortia of human geneticists and clinicians and ever-cheaper sequencing technologies, these discoveries have revealed that many inherited disorders with related clinical diagnoses consist of large sets of rare molecular genetic variation (Buxbaum et al., 2012; Gonzalez et al., 2015). Here we focus on these parallel and synergistic frontiers of disease gene discovery and systems-level analyses in zebrafish that promise to yield insight into disease mechanisms and therapies.

To assess zebrafish as a model, we compare studies of two broad classes of inherited neurological conditions: developmental disorders and degenerative diseases, with each class presenting a spectrum of overlapping genotypes and phenotypes. For developmental disorders, we include Autism Spectrum Disorders (ASD), Intellectual Disability (ID) and Schizophrenia (SCZ) that can all affect executive functions, social and overall intellectual abilities (American Psychiatric Association, and DSM-5 Task Force, 2013). Such disorders can co-occur in the same individual (Amaral et al., 2011) supporting overlapping disease etiologies. The second general class we consider are a subset of degenerative diseases including Cerebellar Ataxia (CATX), Hereditary Spastic Paraplegia (HSP), spinal motor atrophy (SMA), amyotrophic lateral sclerosis (ALS) and Charcot-Marie Tooth Disease (CMT) that impair movement due to degeneration of long axon tracts (Züchner and Vance, 2005). In both developmental and degenerative cases there are examples of: (1) distinct clinical features within a given class that result from mutations in the same genes; and (2) shared clinical phenotypes produced by many different types of genetic mutation (Espinós and Palau, 2009; Kaufman et al., 2010; Timmerman et al., 2013; Vissers et al., 2016). For example, ASD affects roughly 1% of the population but even the most commonly mutated genes only account for 1–3% of ASD with hundreds of suspected causal loci (Miles, 2011; De Rubeis and Buxbaum, 2015; Geschwind and State, 2015). It is likely that these two groupings also have phenotypic overlap since age of onset and progression of symptoms varies within each grouping. For example, SCZ that we group with ''developmental'' disorders also shares symptoms with neurodegenerative conditions. None-the-less, by reviewing the literature on zebrafish models of these two groups of disorders, we hope to highlight the role we feel the zebrafish model has to play in revealing grouped mechanisms of shared clinical features that result from diverse genetic mutations.

### COMPARING HUMAN AND ZEBRAFISH BRAINS AND GENETICS

In addition to significant advantages of using zebrafish to model human disease, there are also challenges to modeling disease conditions in zebrafish. For example, in HSP symptoms are associated with degeneration of corticospinal tracts that have no clear homologous cell type in zebrafish. Moreover, for many human brain regions, the baseline studies to determine whether some of these brain regions function similarly to humans still need to be done. Finally, in terms of genetics, zebrafish have retained gene duplicates from a ray-finned fish whole genome duplication (Glasauer and Neuhauss, 2014) that likely provides both advantages (sub-functionalization of pleiotropic phenotypes) and disadvantages (genetic redundancy) for generating disease models.

## Conserved Brain Regions?

Many brain regions relevant to human disease show molecular and structural homology in zebrafish (**Figure 1**). Gene expression patterns that molecularly define large-scale regions of fore-, mid-, hindbrain and spinal cord of the central nervous system (CNS) are generally conserved in vertebrates (Myers et al., 1986; Lumsden and Krumlauf, 1996; Wurst and Bally-Cuif, 2001; McLean et al., 2007; Wullimann et al., 2011). These major regions exhibit both neurochemical identities (e.g., neurotransmitters and receptors; Higashijima et al., 2004; Renier et al., 2007; Jones et al., 2015) and regional connectivity of sub-structures, including the thalamus (Mueller, 2012), optic tectum (Wullimann, 1994), hypothalamic regulatory nuclei (Herget et al., 2014), cerebellum (Hashimoto and Hibi, 2012), medulla oblongata (Kinkhabwala et al., 2011; Koyama et al., 2011), and spinal cord (Higashijima et al., 2004; Wen and Brehm, 2005). Determining homology between human and zebrafish forebrain structures is more challenging because of developmental differences affecting telencephalon topology (eversion vs. invagination) and the elaboration of the mammalian cerebrum (Striedter, 2005). Recent studies have made headway supporting the existence of zebrafish basal ganglia- (Filippi et al., 2014; Wullimann, 2014), cortex- (Ganz et al., 2014), amygdala- (Maximino et al., 2013) and hippocampus-like circuits (O'Connell and Hofmann, 2011; Maximino et al., 2013; Ganz et al., 2014). While these studies support the existence of structurally homologous brain regions, work is still needed to resolve connectivity and functional homology among fore- and midbrain structures. Furthermore some structures in the human brain appear absent in zebrafish including the pons (Wullimann et al., 2011), cortico-thalamic (Mueller, 2012) and cortico-spinal tracts (Babin et al., 2014). With these detailed analyses, we have the necessary anatomical map needed for neurological disease research that is quickly being enriched by functional studies to test the relevance to these brain regions for both zebrafish behaviors and human disorders.

To address the involvement of brain regions in a particular behavior, advances in functional imaging have been critical. Importantly, these studies focus on larval stages, 5–7 post fertilization, when it is possible to image the entire brain and the larva has already acquired an impressive repertoire of behaviors (Haesemeyer and Schier, 2015). For example, recent studies support functional homology of the zebrafish cerebellum during visual-motor behaviors (Hsieh et al., 2014; Matsui et al., 2014). These findings were made possible by combining behavioral experiments with both genetically encoded calcium imaging to visualize neuronal activity (Matsui et al., 2014) and loose patch recordings from cerebellar Purkinje neurons (Hsieh et al., 2014). These pioneering studies now open the door to sophisticated functional modeling of cerebellar-based neuropathies that can manifest in both neurodevelopmental disorders and neurodegenerative diseases, particularly CATX. Extending these functional technologies

to forebrain circuits will be essential to support past studies suggesting similar cognitive and emotional functions exist in subdivisions of the zebrafish forebrain (Northcutt, 2006; O'Connell and Hofmann, 2011; Maximino et al., 2013).

telencephalon; Hip, hippocampus; Hy, hypothalamus; MN, motor neuron; PT, posterior tuberculum; Th, thalamus; ON, optic nerve.

#### Conserved Genomes?

Human and zebrafish genomes are highly conserved, with 76–82% of human disease genes present in zebrafish and an average of 20–24% of zebrafish genes duplicated (Howe et al., 2013). In some cases duplicates become sub-functionalized providing an advantage for studying pleiotropic phenotypes (Fleisch et al., 2008; Good et al., 2012; Lagman et al., 2015); in other cases, duplicates have redundant functions providing phenotypic buffering and complicating the generation of disease models (Hinits et al., 2012; Manoli and Driever, 2014). To determine if genes linked to a particular human neurological condition are enriched in gene duplicates we compared gene duplication rates in disease gene orthologs. Because past studies have shown duplicate enrichment in genes associated with neuronal development, signaling pathways and neuronal activity (Howe et al., 2013; Glasauer and Neuhauss, 2014), we hypothesized that neurodevelopmental duplicates would have a higher retention rate compared to neurodegenerative duplicates. Several gene sets showed a high duplicate retention frequency: over 60% (ASD-ID) and 45% (ASD and CMT;

FIGURE 2 | Zebrafish orthologs of human neurological disease genes vary with respect to duplicate retention and average protein size. (A) Gene duplicate retention rates in zebrafish are graphed for neurodevelopmental and neurodegenerative disease groups. Yellow numbers at the base of bars represent sample size. (B) Protein sizes of zebrafish orthologs of human disease genes with floating box plots (median with upper and lower quartile-box and range-whiskers). Note some larger proteins are outliers that fall outside of the calculated range. These gene sets for each human disease only incorporate a small percentage of associated genes and selection criteria varied because of the heterogeneity in genes linked to each disease and an emphasis on producing conservative lists. Each gene set was selected using data from research groups and review articles with the goal of including only the highest confidence disease genes based on either statistical thresholds and/or reoccurrence. Autism spectrum disorders (ASD) genes were chosen from the Simons Foundation Autism Initiative (SFARI.org) "high confidence" and "strong candidate" gene lists, which uses a multi-variable scoring analysis that includes sample size, statistical significance, replication, and functional analysis (Basu et al., 2009). ASD-intellectual disability (ID) genes were found in four separate reviews that provide evidence for reoccurrence in (Continued)

#### FIGURE 2 | Continued

both ASD and ID (Kaufman et al., 2010; Krumm et al., 2014; Srivastava et al., 2014; Vissers et al., 2016). Schizophrenia (SCZ) genes were chosen from a SCZ genetics review, however this list is small and does not provide a confidence level for the disease contribution of each gene (Escudero and Johnstone, 2014). X-linked ID genes (Piton et al., 2013) and Charcot-Marie Tooth Disease (CMT; Timmerman et al., 2014) genes were chosen from recent meta-analyses to which we included a threshold of >5 cases per gene. Autosomal dominant and recessive and X-linked ataxia (ATX; Bird, 2016) and Hereditary Spastic Paraplegia (HSP; Fink, 2014) genes were chosen with well-known inheritance pedigrees. Human proteins for these gene lists were generated using BioMart (Kinsella et al., 2011) and the longest isoforms were used to identify zebrafish orthologs (Supplementary Material, Tables S1–7). Human proteins were then Blasted (Flicek et al., 2014) against the zebrafish proteome and ortholog information was recorded. Zebrafish proteins with low percent coverage, protein identity, e-value and ambiguous gene annotation (e.g., gene-like, LOC1084, etc.) were reciprocally blasted to confirm orthology.

**Figure 2A**). In comparison, ID, ATX and HSP zebrafish orthologs had duplicate retention frequencies of 22–26% similar to the estimated genome-wide (20–24%) duplicate retention rate (Postlethwait et al., 2004; Howe et al., 2013). To determine if neurodevelopmental processes were enriched in sets with high duplicate retention rates we compared gene ortholog Gene Ontology (GO) term enrichments within orthologs sets (Mi et al., 2013). This program compares the frequency of a biological term (e.g., neuron development) with the expected frequency calculated for the genome. ASD (n = 35) and ASD-ID (n = 22) orthologs showed enrichments for GO terms associated with nervous system and neuronal development that confirm large studies analyzing extensive data sets of ASD and ID genes (Parikshak et al., 2013; Pinto et al., 2014). By contrast, CMT orthologs (n = 19) did not show any GO term enrichments overall, but those retained zebrafish CMT duplicates were enriched in genes associated with neuronal development (7/9 genes; Supplementary Material, Table S12). Therefore these genes may be associated with earlier onset cases of CMT, which can occur anywhere between birth and adulthood (De Jonghe et al., 1997). Although these gene sets are small and not statistically powerful, these rates show a general trend for retaining duplicates associated with neural development.

To continue comparing these ortholog classes we looked at the length of the longest protein-coding isoform for each disease gene ortholog. This parameter impacts modeling due to numerous exons and complex splicing. For example SH3 and multiple ankyrin repeat domain 3 (SHANK3) is a large gene and mouse Shank3 knockout models exhibit variable phenotypes depending on whether a mutation is expressed in a predominant isoform (Peça et al., 2011; Zhou et al., 2016). Therefore we compared protein sizes between disease gene orthologs (**Figure 2B**). Between groups developmental disorders all had larger median protein sizes when compared to degenerative disease orthologs (**Figure 2B**). However, most notably proteins encoded by ASD genes were more than twice as long when compared to all degenerative disease proteins (**Figure 2B**). Because ASD genes have been suggested to encompass relatively large genes this result was perhaps not surprising (King et al., 2013; Uddin et al., 2014). Together these results on isoform length and duplicate enrichment suggests that ASD gene modeling in zebrafish will need to grapple with both extensive gene duplication and large complex genes that present challenges for knocking out all protein-coding isoforms and for generating rescue constructs.

### NEURODEVELOPMENTAL AND NEURODEGENERATIVE ZEBRAFISH MODELS

#### Developmental Disorders

Zebrafish models of developmental disorders have benefitted from the accessible embryonic stages and simplified nervous system that reveal an important role for signaling that patterns in the early embryo. Developmental disorders including ASD, ID and SCZ start to manifest phenotypically early in development and include deficits in social, learning and occupational functions (DSM-V). Because brain regions mediating human cognitive symptoms may lack parallels in zebrafish, modeling has focused on embryonic development and disorder comorbidities (such as sensory hypo-/hyper-sensitivity, sleep disruptions, and epilepsy) that still allow testing of etiological theories for these developmental disorders. These studies have paved the way for comprehensive functional assessments that link cellular- and circuit-level phenotypes to changes in behavior.

#### Molecular to Cellular Mechanisms: Signaling Pathways and Head Size

Mounting evidence links the etiology of neurodevelopmental disorders to embryonic stages. For example, teratogen exposure during gestation can cause developmental disorders (Arndt et al., 2005; Levy, 2011) and embryonic phenotypes in knockout mouse models provide support for an embryonic component underlying neurodevelopmental disorders (Knuesel et al., 2005; Lee et al., 2011; Durak et al., 2015). Zebrafish knockdown models of ASD and ID genes suggest that disrupted patterning of presumptive neural tissue in developmental disorders can occur as early as blastula stages (Yimlamai et al., 2005) and during gastrulation (De Rienzo et al., 2011; Turner et al., 2015). At the molecular level, these disruptions in patterning are likely due to changes in conserved signaling pathways. Several ASD and SCZ zebrafish models have investigated disease genes associated with the Wnt pathway (De Rienzo et al., 2011; Bernier et al., 2014; Brooks et al., 2014). For example, the Wnt interacting protein Chromodomain Helicase DNA binding domain 8 (CHD8) directly affects brain development during gastrulation and increases the size of the optic tectum, mirroring macroencephaly seen in ASD patients carrying CHD8 mutations (Bernier et al., 2014; Sugathan et al., 2014). These genotype-specific features (e.g., macrocephaly) provide a phenotypic screen that can be used to investigate genetic classes within disorders. For example zebrafish potassium channel tetramerization domain containing 13 (kctd13) was shown to have a dose-dependent affect in producing macrocephaly (knockdown) and microcephaly (overexpression) that supports a role for KCTD13 copy number variants causing head size phenotypes (Golzio et al., 2012). Moreover, Disrupted In Schizophrenia 1 (DISC1) interacts with canonical and non-canonical Wnt signaling and zdisc1 morphants and mutants exhibit disorganized axon tracts at larval stages that can be rescued by activating Wnt signaling (De Rienzo et al., 2011). These studies provide clear examples of utilizing zebrafish as an embryonic model to determine molecular and cellular mechanisms that define morphological phenotypes seen in individuals with developmental disorders.

#### Systems-Level Mechanisms: Disrupting the Balance of Neuronal Activity

Mechanisms affecting neuronal activity can contribute to neurodevelopmental disorders and zebrafish have been used to relate circuit-level changes in activity to behavior (Rubenstein and Merzenich, 2003; Eichler and Meier, 2008; Nelson and Valakh, 2015; Scharf et al., 2015). Circuit-level changes include disrupting the excitatory and inhibitory (E/I) balance, an operational set-point of excitation and inhibition within neural circuits that maintains functional behaviors (Borodinsky et al., 2004; Gatto and Broadie, 2010; Turrigiano, 2012; Vitureira et al., 2012; Davis, 2013). Zebrafish ASD and ID models have looked at E/I balance using transgenic fish lines expressing fluorescent glutamatergic (excitatory) and GABAergic (inhibitory) neurons (Kozol et al., 2015; Hoffman et al., 2016). Recently, Hoffman et al. found that populations of GABAnergic neurons were significantly decreased in contactin associated protein-like 2 (cntnap2ab) mutants, recapitulating a mutant mouse Cntnap2 model and suggesting that in the absence of cntnap2ab larvae fail to maintain inhibitory neuronal populations. This inhibitory decrease was shown to increase seizure susceptibility in cntnap2ab−/<sup>−</sup> mutants by applying a GABA receptor antagonist (Hoffman et al., 2016). In addition to seizure susceptibility, cntnap2ab−/<sup>−</sup> mutants had increased nighttime activity providing a circadian disruption for high-throughput drug screening. To identify potential therapies, they screened for drugs that reduced nighttime activity and identified a phytestrogen that restored wild type-like activity states. Like decreased inhibition, increased excitation is also known to contribute to developmental disorders. One well-studied example of this is augmented metabotropic glutamate receptor (mGluR) signaling in Fragile X Syndrome (Scharf et al., 2015). Similar to Fragile x mental retardation 1 (Fmr1) knockout models in mice, a zebrafish fmr1 knockdown model showed behavioral deficits that were ameliorated when treated with an mGluR inhibitor (Tucker et al., 2006). These studies demonstrate how zebrafish genetic models can be used to explore disorder etiology at multiple levels and efficiently test molecular theories for drug discovery.

#### Systems-Level Mechanisms: Comorbidities and Connecting Cells to Behavior

Individuals with developmental disorders are more likely to have accompanying medical conditions, or comorbidities, than typically developing individuals (Gurney et al., 2006; American Psychiatric Association, and DSM-5 Task Force, 2013; Chen et al., 2013a). Non-cognitive comorbidities such as sensory hypo- or hyper-sensitivity, epilepsy and gastrointestinal (GI) discomfort have revealed cellular-level mechanisms that may underlie behavioral phenotypes in developmental disorders. Several zebrafish knockdowns models of ASD, ID and epilepsy genes have looked at impaired touch sensitivity. Knockdown models of the ASD genes autism susceptibility candidate 2 (auts2) and shank3a exhibit hyposensitivity with concomitant neuronal cell death and morphological changes in skin innervating sensory neurons (Oksenberg et al., 2013; Kozol et al., 2015). Also exploring sensitivity, chromodomain helicase DNA binding protein 2 (chd2) knockdowns and sodium channel, voltage gated, type II, alpha (scn1lab) knockouts display hyper-excitable phenotypes that are characterized by extended or disorganized swimming with epileptiform-like activity in the brain (Baraban et al., 2013; Suls et al., 2013; Galizia et al., 2015). These epileptic swimming bouts provide a stereotyped behavior for highthroughput drug screening. Such a screen in scn1lab mutants identified anti-histamine clemizole as a novel anti-epileptic drug (Baraban et al., 2013). Although more focus has been paid to conditions such as epilepsy, other comorbidities like GI distress in ASD have yet to be investigated comprehensively (Hsiao, 2014; Bresnahan et al., 2015). For instance, chd8 morphants have a decrease in HuC/D positive enteric neurons innervating the gut and have impaired gut motility (Bernier et al., 2014). Again this example provided a cellular to systems level mechanism for GI distress seen in a majority of ASD patients carrying a CHD8 mutation. These examples all show the utility of zebrafish for studying comorbidities that impact the quality of life of large cohorts of patients; therefore a better understanding of the basis for these comorbidities would likely improve patient care.

#### Hereditary Neurodegenerative Disorders

Some common cellular mechanisms underlying degenerative diseases have been elucidated through gene discovery and zebrafish modeling of rare hereditary diseases. The degeneration of axon tracts in the central and peripheral nervous system are a clinical feature in neurodegenerative disorders such as Charcot-Marie-Tooth disease type 2 (CMT2), HSP, SMA or spinal muscle atrophy (SMA), ALS, as well as some forms of CATX which have phenotypic and mechanistic overlap (Züchner and Vance, 2005; Timmerman et al., 2013; Bargiela et al., 2015; Burté et al., 2015). The early development of zebrafish peripheral, motor and sensory neurons provide a foundation that has been used to dissect molecular mechanisms at both the cellular and systems level especially in models of SMA and ALS (McGown et al., 2013; Wiley et al., 2014). Using an innovative strategy to develop SMA therapies, one study used a high-throughput synthetic genetic array (SGA) screen in fission yeast to identify gene networks that when targeted with drugs reversed motor axon outgrowth deficits in a zebrafish SMA model (Wiley et al., 2014). Zebrafish models of these neurodegenerative diseases have also focused on molecular mechanisms such as axonal transport, mitochondrial dynamics, and autophagy, while also measuring morphological changes at the systems level such as alterations at the neuromuscular junction, degeneration of motor

#### Cellular Level Mechanisms: Axonal Transport

A subset of the causative genes in these disorders are directly involved in axonal transport processes (Timmerman et al., 2013). The optical transparency of zebrafish and transgenic lines available make zebrafish an ideal model to study the relationship between axonal transport and axon degeneration in vivo (Pluci ´nska et al., 2012; O'Donnell et al., 2014). Mutations in Kinesin Family member 5A (KIF5A), a molecular motor for transporting microtubule-mediated cargo, have been reported in both CMT2 (Crimella et al., 2012), and HSP patients (Reid et al., 2002; Fichera et al., 2004). A kif5a mutant zebrafish shows decreased touch response, and defective sensory neuronal maintenance all within the larval stages of development (Campbell et al., 2014). Furthermore the authors found that kif5a specifically affects the transport and distribution of mitochondria in neurons, but not lysosomes or presynaptic vesicles. Dominant mutations in Atlastin GTPase 1 (ATL1) encoding atlastin-1 cause an early onset form of HSP (Dürr et al., 2004). Morpholino knockdown of atl1 in zebrafish causes decreased mobility in larval fish and specifically disrupts axon tracts of spinal motor neurons (Fassier et al., 2010). Fassier et al. (2010) further demonstrated that the phenotype is the result of altered BMP signaling and that atlastin may play a role in BMP receptor trafficking. This link to BMP receptor trafficking suggested blocking BMP receptors as a therapeutic strategy that indeed ameliorated both cellular and behavioral phenotypes in the atl1 morphant zebrafish model. These zebrafish models support axonal transport as a cellular mechanism that could explain why long axons in CMT2 and HSP are primarily affected by genetic mutations in genes associated with transport processes.

#### Cellular Level Mechanisms: Mitochondrial Neuropathies

Mitochondrial dysfunction is another common mechanism in neurodegeneration. Dominant mutations in Mitofusin 2 (MFN2) are the primary cause of axonal degeneration in Charcot-Marie-Tooth Neuropathy (CMT2), and MFN2 has been implicated in the fusion and transport of mitochondria in neurons (Chen et al., 2003; Züchner et al., 2004; Baloh et al., 2007). Murine Mfn2 knockout and knock-in models are embryonic or postnatal lethal and do not develop a peripheral neuropathy, however a conditional knockout model did produce cerebellar degeneration and neonatal lethality (Chen et al., 2003, 2007; Strickland et al., 2014). In contrast, a stable mfn2L285X loss-of-function zebrafish model does recapitulate the motor neuron degenerative phenotype showing progressive loss of swimming ability, loss of neuromuscular junctions (NMJs), and early lethality by 1 year of age (Chapman et al., 2013). The authors found that the transport of mitochondria is disrupted in cultured motor neurons from the homozygous mfn2L285X at 24 hpf suggesting that a primary transport defect occurs before the onset of symptoms. In addition to axon degeneration of motor neurons a portion of patients with MFN2 mutations also develop optic atrophy (Züchner et al., 2006). The optic nerve seems to be particularly susceptible to mitochondrial dysfunction and is often affected in clinical spectrum phenotypes classified as mitochondrial optic neuropathies (Yu-Wai-Man et al., 2009). Recessive mutations in the nuclear encoded mitochondrial gene Optic Atrophy 3 (OPA3) cause a spectrum disorder classified as Costeff syndrome and includes optic atrophy, ataxia, extra pyramidal dysfunction, and increased urinary excretion of 3-methylglutaconic acid (MGC; Costeff et al., 1989). Zebrafish opa3 null mutants show increased MGC at both 5 dpf and at 2–5 months (Pei et al., 2010). At 1 year they show decreased optic nerve thickness and retinal ganglion cell density. Mutants have detectable changes in movement behaviors at larval stages and adults show loss of horizontal swimming. The authors speculate that the swimming phenotype can be attributed to ataxia, however TUNEL and histological staining of the cerebellum did not reveal any abnormalities. A third disease gene linked to mitochondrial dynamics and a spectrum of degenerative neurological conditions that include optic atrophy, CMT and cerebellar degeneration is Solute Carrier family 25, member 46 SLC25A46; Abrams et al., 2015). Studies in zebrafish and patient stem cells linked disruption of slc25a46 to reduced mitochondrial fission, altered distribution of mitochondria in motor neurons, and defective maintenance of neuronal processes. Even though cellular phenotypes were dramatic, swimming deficits in slc25a46 morphants were mild.

#### Cellular Level Mechanisms: Cerebellar Purkinje Neurons

Ataxia and associated sensory and motor phenotypes result from genetic mutations that affect various cell types within the spinocerebellar circuit. Cerebellar PCs are commonly affected and appear especially sensitive to peroxisomal dysfunction associated with PC loss or cerebellar atrophy (Akbar and Ashizawa, 2015). Consistent with this, zebrafish ataxia models, such as sorting nexin 14 (snx14) morphants show decreases in PC progenitors while cwf19-like 1 (cwf1911) morphants show disruptions in overall hindbrain morphology (Burns et al., 2014; Akizu et al., 2015). Alternatively, one group has focused on primary motor neurons for functional studies of Potassium Channel, voltage gated shaw related subfamily c, member 3 (KCN3) mutations that cause Spinocerebellar ataxia type 13 (Waters et al., 2006). They found that zebrafish kcn3a is expressed in the primary motor neurons and overexpression of a dominant negative version of this potassium channel decreases in neuronal excitability during fictive swimming (Issa et al., 2012). To follow up this study, Issa et al. also investigated the affect of KCN3 mutations associated with infant onset ataxia. Overexpression of two human KCN3 mutations demonstrated axonal defects that were only found in a mutation associated with infantile onset of SCA13. Given that neurodegenerative phenotypes become more severe with age, there has been doubt as to whether modeling these diseases in zebrafish larvae would be informative. The studies reviewed above indicate that indeed it is possible to gain functional insight into the basic biology of these disease genes from modeling in the larva. As a newer model, however, approaches in zebrafish are not quite as standardized as those in longer established models like Drosophila and mouse resulting in the diversity of modeling strategies employed by different researchers to model these inherited neurological conditions.

### LESSONS LEARNED FROM MODELING INHERITABLE DISEASE GENES

### Homeostatic Plasticity: Unlinking Cellular and Systems Level Phenotypes

In some less common examples, genetic mutations cause cellular- and molecular-level phenotypes without leading to behavioral phenotypes. In zebrafish where's waldo and strumpy mutants, neuromuscular synaptogenesis is defected but the mutants exhibit normal motility (Hutson and Chien, 2002; Panzer et al., 2005). Similar phenomena have been observed in mouse hypoxanthine-guanine phosporibosyltransferase (HPRT) and HPRT-adenine phosporibosyltransferase (APRT) mutant models of Lesch-Nyhan syndrome genes that produced an expected drop in dopamine but lacked self mutilation behaviors (Kuehn et al., 1987; Engle et al., 1996; Jinnah et al., 1999). These observations indicate that molecularand cellular-level phenotype does not always correspond to behavioral phenotype, suggesting the existence of compensatory mechanisms at the systems-level. Besides genetic compensation from other genes, the nervous system also has remarkable capacity to stabilize its functions via homeostatic plasticity. Compensatory homeostatic plasticity operates on multiple levels, regulating synaptogenesis, synaptic strength and intrinsic excitability to stabilize neural circuit output in the context of genetic and/or environmental perturbations (Turrigiano, 2012; Vitureira et al., 2012; Davis, 2013). In zebrafish strumpyp37er mutants have enlarged NMJ acetylcholine receptor clusters are compared to wild-type (Panzer et al., 2005), but lack motor phenotypes, indicating homeostatic plasticity. In addition, defective homeostatic plasticity has been associated to a variety of human neurological diseases, including ASD, ID and Fragile X Syndrome (Rubenstein and Merzenich, 2003; Eichler and Meier, 2008; Gatto and Broadie, 2010; Yizhar et al., 2011; Wondolowski and Dickman, 2013; Nelson and Valakh, 2015). Mutant animal models with cellular but not behavioral phenotype have the potential to shed light on mechanisms of homeostatic plasticity at the systems-level. Given the diversity of molecular genetic pathways that contribute to developmental and degenerative disorders, a potential therapeutic target is to boost compensatory mechanisms that act to re-establish systems-level function.

### Genetic Buffering: Molecular Compensation for Genetic Lesions

It is also common, though not necessarily well-represented in the literature, for knockout models to lack phenotypes. In a zebrafish study that generated mutant lines with 32 distinct lesions in 24 genes, most of the mutants exhibit a wild-type phenotype (Kok et al., 2015). Rather than a unique characteristic of zebrafish, such phenotypic buffering is found across singlecelled to multi-cellular organisms. For example, >70% in fission yeast and >80% in bakers yeast, of genes can be individually mutated with little effect on haploid viability in the laboratory setting (Kim et al., 2010) 1 . Also in mice, though it is hard to accurately estimate the proportion of knockout mice without detectable phenotypes due to a lack of publications, the number is considerable (Barbaric et al., 2007). This phenomenon can be explained by the recent finding that other genes in a regulatory network can provide genetic compensation in mutants (Rossi et al., 2015). Such compensation can vary depending on genetic background (Gerlai, 1996; Pearson, 2002). Another recent study in zebrafish shows that the oncogenic B-RAF proto-oncogene (BRAFV600<sup>E</sup> ) mutations rarely covert carrier cells into cancer cells unless in a p53 mutant background (Kaufman et al., 2016). The importance of genetic background for gene regulatory compensation likely contributes to disease penetrance and expressivity in humans as well (Zlotogora, 2003; Andersen and Al-Chalabi, 2011; Cooper et al., 2013; Persico and Napolioni, 2013), and suggests the importance to move from analyzing single gene to systems-level analysis of gene regulatory networks in disease models (Döhr et al., 2005; Barabási et al., 2011; Chen et al., 2014; Wiley et al., 2014; Marbach et al., 2016).

#### Broad Gene Expression; Local Pathology

Both developmental and degenerative diseases are associated with dysfunction in specific regions of the CNS (**Figure 1**). None-the-less most genes have heterogeneous spatial and temporal expression patterns that extend well beyond the windows of time and specific neural circuits associated with the disorder. Therefore we did a limited meta-analysis of mRNA expression of zebrafish disease orthologs. Unfortunately a zebrafish gene expression atlas does not exist and the most complete data sets only span embryonic development. Therefore we chose to compile previously published in situ hybridization data and determine the relative enrichment of gene expression in brain regions and spinal cord (**Figure 3**; Suplementary Material, Tables S13–18). During the first day of development, both gene sets show enriched expression in the hindbrain with the developmental set having enriched expression in the forebrain. By the second and third day, expression patterns become broader showing similar enrichment throughout the brain at these stages of morphogenesis and circuit formation. Broad CNS expression patterns suggest that genes play functional roles throughout development and across the nervous system despite being associated with symptoms that disrupt specific circuits during particular times of life.

#### LOOKING FORWARD: NEURAL CIRCUITS, BEHAVIOR, AND THERAPY

Because of the large number of rare mutations linked to inherited nervous system diseases, an important frontier for disease modeling is strategies that leverage to make stable F<sup>0</sup> mutant models of inherited neurological disorders (Jao et al.,

1997; Kudoh et al., 2001; Thisse et al., 2001; Wurst and Bally-Cuif, 2001; Groth et al., 2002; Rauch et al., 2003; Thisse and Thisse, 2004, 2005; Croushore et al., 2005; Imamura and Kishi, 2005; Meyer et al., 2005; Thompson et al., 2005; Yimlamai et al., 2005; Liu et al., 2006; Mendelsohn et al., 2006; Meyer and Smith, 2006; Cheng et al., 2007; George et al., 2007; Goruppi et al., 2007; Katsuyama et al., 2007; Patten et al., 2007; Anichtchik et al., 2008; Stuebe et al., 2008; Sun et al., 2008; Yoshida and Mishina, 2008; Zhou et al., 2008; Emond et al., 2009; Ferrante et al., 2009; Monnich et al., 2009; Patten and Ali, 2009; Titus et al., 2009; Wood et al., 2009; Appelbaum et al., 2010; Davey et al., 2010; Fassier et al., 2010; Rissone et al., 2010; Takada and Appel, 2010; Mapp et al., 2011; Yeh et al., 2011; Artuso et al., 2012; Dresner et al., 2012; Gomez et al., 2012; Imai et al., 2012; Mueller, 2012; Pujol-Martí et al., 2012; Xing et al., 2012; Yanicostas et al., 2012; Baraban et al., 2013; Campbell and Marlow, 2013; Haug et al., 2013; Ng et al., 2013; Recher et al., 2013; Suls et al., 2013; Vatine et al., 2013; Bernier et al., 2014; Garbarino et al., 2014; Housley et al., 2014; Hsieh et al., 2014; Galizia et al., 2015; Kozol et al., 2015; Wakayama et al., 2015).

2013; Shah et al., 2015). Such models enable the rapid screening of candidate disease genes for whether they produce disease relevant phenotypes in the zebrafish model. To this end, many

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

Frontiers in Molecular Neuroscience | www.frontiersin.org July 2016 | Volume 9 | Article 55 |

groups currently augment their MO studies by demonstrating similar phenotypes in F<sup>0</sup> mutants (Bernier et al., 2014; Aspatwar et al., 2015; Bögershausen et al., 2015; O'Rawe et al., 2015; Wheeler et al., 2015; Xing et al., 2015). Still others have found mismatches between morphant and stable mutant phenotypes (Kok et al., 2015; Rossi et al., 2015; Stainier et al., 2015). Clearly in some cases these mismatches ascribed to non-specific effects of the morpholino (Kok et al., 2015) can also be explained by compensatory mechanisms masking the phenotype in the stable mutant (Rossi et al., 2015). To address the challenges of gene duplicates and multiple mutation causes of disease (Shah et al., 2015), several labs have further pioneered a strategy to pool guides targeting multiple genes and inject them together to efficiently screen multiple mutations in the F<sup>0</sup> generation. Combined with a large repertoire of behaviors that develop within 5 days of fertilization and diverse transgenic lines for rapid screening of cellular phenotypes, F<sup>0</sup> CRISPR zebrafish mutagenesis promises to contribute significantly to our understanding of genetic variation linked to nervous system disorders.

To model specific patient missense mutations and to better understand the basic biology of disease genes by tagging them in vivo or create conditional mutant alleles, several groups have also recently pioneered the use CRISPR/Cas9 for more sophisticated genome engineering. One novel strategy effectively ''enhancer traps'' a gene of interest by replacing the last exon with an engineered last exon encoding the C-terminal end of the coding sequence in frame with a cleavable p2A sequence followed by a fluorescent reporter (Li et al., 2015). In this way, the spatial and temporal expression dynamics of the protein can be captured. Also recently, Hoshijima et al. (2016) have developed streamlined strategies to precisely edit the genome and generate conditional mutant alleles flanked by LoxP sites. Such conditional mutant alleles have been used to great effect in mouse models to test when the mutation acts to produce different disease phenotypes. For example, in a mouse Shank3 autism model, rescuing the mutant Shank3 protein in the adult was sufficient to rescue social interactions and excessive grooming but not anxiety and repetitive motor behaviors (Mei et al., 2016).

The development of approaches that enable monitoring of behavior-relevant neural circuits in the intact larvae will be a boon for modeling inherited neurological disease (Ahrens et al., 2013; Fosque et al., 2015; Randlett et al., 2015; Dunn et al., 2016). Such approaches have the significant advantage of being unbiased. While many of these systems-level analyses use light-sheet or high-end microscopy to capture data (Ahrens et al., 2013; Fosque et al., 2015; Dunn et al., 2016), others use standard confocal microscopy to identify relevant brain circuits (Randlett et al., 2015) that are often spatially distributed across the nervous system. Several groups have also made concerted efforts towards establishing brain atlases to structurally and functionally annotate the brain (Ronneberger et al., 2012; Turner et al., 2014; Randlett et al., 2015) which is crucial since the ability to interpret imaging data is only as good as our understanding of brain regions (Arrenberg and Driever, 2013). Once these circuits are identified, they can then be studied in greater depth using in vivo calcium imaging with genetically encoded calcium indicators- (GECIs; Chen et al., 2013b), laser or enzymatic ablation of parts of the circuit (Liu and Fetcho, 1999; Tabor et al., 2014; Chen et al., 2016), electrophysiological recordings (Koyama et al., 2011; Baraban, 2013; Johnston et al., 2013; Wen et al., 2013), and optogenetics (Wyart et al., 2009; Kimura et al., 2013) to dissect circuit properties. Most of these approaches have yet to be broadly applied to zebrafish models of disease but once applied more broadly they promise to significantly contribute to our understanding of systemslevel neural circuit mechanisms that contribute to symptoms of inherited neurological disorders.

In addition to genetic screens, due to their small size and their tendency to absorb drugs added directly to the water, zebrafish larvae are uniquely amenable for high-throughput drug screens (Rihel et al., 2010; Rihel and Schier, 2013; Bruni et al., 2014). High-throughput behavioral screens in zebrafish have enabled the classification of neuro-active drugs with respect to their impact on whole organism behavior. The ability to screen compounds in this manner is crucial since neuroactive drug discovery is still more empirical—a matter of what works—rather than rational—a matter of what makes sense based on chemistry and known molecular targets (Bruni et al., 2014). As highlighted in neurodevelopmental and neurodegenerative sections above, several disease models have made great use of the ability to use drugs to enhance or suppress mutant phenotypes as a means to identify therapeutic strategies (Fassier et al., 2010; Baraban et al., 2013; Hoffman et al., 2016).

## CONCLUSION

The continued expertise and innovations of zebrafish genetic and developmental tools will continue to make zebrafish an attractive neurological disease model. Going forward, combining standard assays that allow comparisons across models with newer approaches would be ideal to enable a better understanding of the molecular, cellular, and systems-level groupings of these neurological conditions. Finally, zebrafish will certainly contribute to consortia of research groups that use multiple animal models for discovering essential molecular to circuit level mechanisms underlying neurological disease.

## AUTHOR CONTRIBUTIONS

RAK conducted all meta-analyses, made all figures and wrote introductory and developmental disorder sections and conducted extensive edits to coordinate sections. AJA wrote the bulk of the neurodegenerative section with the exception of the ataxia section that was written by EB. DMJ wrote comorbidity section that centered on GI distress in developmental disorders. QY wrote the lessons learned section. JED conceived the scope of the review, wrote frontiers section and helped RAK to conduct extensive edits to coordinate sections.

### FUNDING

This work was supported by support from the National Institutes of Health Institute of Mental Health R03MH103857 to JED and from the Institute of General Medicine, an IMSD graduate fellowship from parent grant R25GM076419 to DMJ.

#### ACKNOWLEDGMENTS

We would like to acknowledge our human genetics colleagues Stephan Züchner, Rebecca Schüle, Margaret Pericak-Vance, and

#### REFERENCES


Joseph Buxbaum without whom we would never have started modeling human genetic disorders and diseases in zebrafish.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol.2016.000 55/abstract


a mutant sod1 zebrafish model. Ann. Neurol. 73, 246–258. doi: 10.1002/ana. 23780


vertebrates: a comparative developmental analysis. Front. Neuroanat. 5:27. doi: 10.3389/fnana.2011.00027


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

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

# Defects of the Glycinergic Synapse in Zebrafish

#### Kazutoyo Ogino\* and Hiromi Hirata\*

Department of Chemistry and Biological Science, College of Science and Engineering, Aoyama Gakuin University, Sagamihara, Japan

Glycine mediates fast inhibitory synaptic transmission. Physiological importance of the glycinergic synapse is well established in the brainstem and the spinal cord. In humans, the loss of glycinergic function in the spinal cord and brainstem leads to hyperekplexia, which is characterized by an excess startle reflex to sudden acoustic or tactile stimulation. In addition, glycinergic synapses in this region are also involved in the regulation of respiration and locomotion, and in the nociceptive processing. The importance of the glycinergic synapse is conserved across vertebrate species. A teleost fish, the zebrafish, offers several advantages as a vertebrate model for research of glycinergic synapse. Mutagenesis screens in zebrafish have isolated two motor defective mutants that have pathogenic mutations in glycinergic synaptic transmission: bandoneon (beo) and shocked (sho). Beo mutants have a loss-of-function mutation of glycine receptor (GlyR) β-subunit b, alternatively, sho mutant is a glycinergic transporter 1 (GlyT1) defective mutant. These mutants are useful animal models for understanding of glycinergic synaptic transmission and for identification of novel therapeutic agents for human diseases arising from defect in glycinergic transmission, such as hyperekplexia or glycine encephalopathy. Recent advances in techniques for genome editing and for imaging and manipulating of a molecule or a physiological process make zebrafish more attractive model. In this review, we describe the glycinergic defective zebrafish mutants and the technical advances in both forward and reverse genetic approaches as well as in vivo visualization and manipulation approaches for the study of the glycinergic synapse in zebrafish.

#### Edited by:

Robert J. Harvey, University College London, UK

#### Reviewed by:

Jochen C. Meier, Technical University Braunschweig, Germany Julia Dallman, University of Miami, USA

#### \*Correspondence:

Kazutoyo Ogino kogino@chem.aoyama.ac.jp; Hiromi Hirata hihirata@chem.aoyama.ac.jp

Received: 31 March 2016 Accepted: 13 June 2016 Published: 29 June 2016

#### Citation:

Ogino K and Hirata H (2016) Defects of the Glycinergic Synapse in Zebrafish. Front. Mol. Neurosci. 9:50. doi: 10.3389/fnmol.2016.00050 Keywords: startle disease, glycine, receptor, synapse, transporter, zebrafish

## OVERVIEW OF GLYCINERGIC NEUROTRANSMISSION IN THE MAMMALIAN NERVOUS SYSTEM

Glycine is a major inhibitory neurotransmitter in the brainstem and spinal cord. The glycinergic transmission plays important roles in the reflex and the control of rhythmic motor behaviors such as locomotion and breathing (Schmid et al., 1991, 1996; Gomeza et al., 2003a; Grillner, 2003; Callister and Graham, 2010), as well as nociceptive processing (Baba et al., 2001; Ahmadi et al., 2002; Harvey et al., 2004; Zeilhofer, 2005). Glycine receptors (GlyRs) are also expressed in the retina (Sassoè-Pognetto et al., 1994; Haverkamp et al., 2003, 2004; Jusuf et al., 2005; Heinze et al., 2007) and various regions of the brain, such as amygdala (McCool and Botting, 2000; McCool and Farroni, 2001), hippocampus (Malosio et al., 1991; Chattipakorn and McMahon, 2002, 2003; Brackmann et al., 2004; Danglot et al., 2004; Eichler et al., 2009; Jonsson et al., 2012;

Chen et al., 2014; Winkelmann et al., 2014; Çali¸skan et al., 2016), midbrain (Malosio et al., 1991; Mangin et al., 2002; Jonsson et al., 2012), neocortex (Malosio et al., 1991; Jonsson et al., 2012; Salling and Harrison, 2014), to regulate neuronal network excitability. GlyRs are pentameric ligand-gated chloride channels taking the form of α-subunit homopentamers or αβ-subunit heteropentamars. The α-subunit is ligand-binding subunit (Becker et al., 1988; Grenningloh et al., 1990a,b; Kuhse et al., 1990), whereas the β-subunit binds to synaptic gephyrin scaffold consisting of tens to hundreds gephyrin molecules, thereby the αβ heteromeric GlyRs can be accumulated at postsynaptic sites (Kirsch et al., 1993; Meyer et al., 1995; Feng et al., 1998; Hanus et al., 2004; Sola et al., 2004; Grudzinska et al., 2005; Kim et al., 2006; Calamai et al., 2009; Herweg and Schwarz, 2012; Specht et al., 2013). However, the postsynaptic localization of GlyR is not permanent. GlyRs are constantly exchanged between synaptic sites and extrasynaptic plasmamembrane by means of lateral diffusion, and alterations of the diffusion properties have been considered to be mechanisms involved in the synaptic plasticity (Meier et al., 2001; Dahan et al., 2003; Ehrensperger et al., 2007; Lévi et al., 2008; Calamai et al., 2009; Specht et al., 2011). Furthermore, the postsynaptic gephyrin scaffold is involved in regulation of the properties for synaptic plasticity in both glycinergic synapses and GABAergic synapses through interactions with various proteins (Fritschy et al., 2008; Tyagarajan and Fritschy, 2014).

Two possible stoichiometric models of the heteromeric GlyR have been proposed from research using purified GlyR subunits: 3α2β and 2α3β (Langosch et al., 1988; Burzomato et al., 2003; Grudzinska et al., 2005). Durisic et al. (2012) published data supporting 3α2β stoichiometry by single-molecule stepwise photo-bleaching: whereas a homomeric channel of fluorescent protein-tagged α subunits exhibited bleaching by a five-step reduction of fluorescence, a heteromeric channel containing the fluorescent protein-tagged α-subunits exhibited bleaching by a three-step reduction (Durisic et al., 2012). However, another group proposed the likelihood of the 2α3β stoichiometry based on the observation of antibody bounded receptor consisting of FLAG-tagged α-subunit and His-tagged β-subunit using singlemolecule resolution atomic force microscopy (Yang et al., 2012). The single-molecular imaging showed that the maximum binding number of antibodies to each of the tag, anti-FLAG antibodies and anti-His antibodies could bind to the receptor up to two and three, respectively. The exact stoichiometry of the αβ GlyR heteromer thus remains a point of contention, and requires further investigation. Although the formation of a homomeric β GlyR has also been hypothesized, biochemical evidence has indicated that β subunits do not form pentamers (Griffon et al., 1999), and an electrophysiological analysis showed that recombinant expression of only the β subunit in HEK-293 cells does not result in glycine-gated currents (Bormann et al., 1993).

In contrast to heteromeric GlyR, homomeric GlyRs are mainly distributed in the extrasynaptic area or presynaptic terminal in the brain and spinal cord (Turecek and Trussell, 2001; Jeong et al., 2003; Ye et al., 2004; Wang et al., 2005; Deleuze et al., 2005; Eichler et al., 2008; Kubota et al., 2010; Hruskova et al., 2012; Kunz et al., 2012; Avila et al., 2013; Chen et al., 2014; Salling and Harrison, 2014; Winkelmann et al., 2014). The extrasynaptic homomeric GlyRs are activated by ambient glycine or taurine, thereby tonic glycinergic current that contributed to regulation of neuronal excitability both in mature and immature neural tissue is evoked (Chattipakorn and McMahon, 2003; Wang et al., 2005; Eichler et al., 2008; Chen et al., 2014; Maguire et al., 2014; Salling and Harrison, 2014). In contrast to the GlyRs on mature neurons, GlyRs on immature neurons typically induce depolarizing chloride efflux in accordance with the chloride gradient that is opposite to that of mature neurons (Reichling et al., 1994; Flint et al., 1998; Ben-Ari, 2002). The inverted chloride gradient is formed and maintained by Na+- K <sup>+</sup>-2Cl<sup>−</sup> cotransporter (NKCC1), which import chloride ions into the cell. On immature neurons, presynaptic GlyRs facilitate neurotransmitter release through activation of voltage gated calcium channel (VGCC) (Turecek and Trussell, 2001; Jeong et al., 2003; Ye et al., 2004; Eichler et al., 2009; Kunz et al., 2012; Winkelmann et al., 2014). Similarly, tonic activation of the extrasynaptic GlyR on the developing immature neurons induced Ca2<sup>+</sup> transients through VGCCs. The activation of VGCCs have been considered to promote the synaptic accumulation of GlyRs (Kirsch and Betz, 1998; Kneussel and Betz, 2000) or to regulate the development of cortical neurons by promoting migration (Avila et al., 2013, 2014). The Ca2<sup>+</sup> transients may also be involved in the K+-Cl<sup>−</sup> cotransporter type 2 (KCC2) expression (Brustein et al., 2013; Allain et al., 2015). It is generally accepted, the activity of NKCC1 declines with progress of development, whereas activity of KCC2, which establishes the chloride gradient of mature neurons by extrusion of chloride ions, becomes predominant (Kanaka et al., 2001; Ben-Ari, 2002; Mikawa et al., 2002; Wang et al., 2002; Delpy et al., 2008). The KCC2 activity is necessary for glycinergic synapse maturation. Knockdown of KCC2 impaired cluster formation of GlyR α1 subunit (adulttype subunit) and gephyrin in cultured spinal neurons, whereas α2 subunit (neonatal type subunit) cluster formation was not affected by the KCC2 knockdown (Schwale et al., 2016). In addition to the developmental role, the tonic depolarizing current contributes to regulate the neuronal excitability in immature hippocampus (Chattipakorn and McMahon, 2003; Eichler et al., 2008; Chen et al., 2014).

The α homomeric GlyR current can be pharmacologically distinguished from that of the αβ heteromeric GlyR by picrotoxin, a GlyR inhibitor: picrotoxin has a 100-fold greater affinity for the α homomeric GlyR than the αβ heteromeric GlyR, such that the α homomeric GlyR is specifically blocked by picrotoxin at low concentrations (Pribilla et al., 1992; Mangin et al., 2002). These currents can also be distinguished by electrophysiological analysis, because the conductance mediated by α1 or α3 homomeric GlyR is larger than that of α1β or α3β heteromeric GlyR (Takahashi et al., 1992; Bormann et al., 1993; Rajendra et al., 1995; Beato et al., 2002; Burzomato et al., 2004; Zhang et al., 2015).

In addition to the GlyR agonist, glycine acts as co-agonist of N-methyl-D-aspartate receptor (NMDAR) (Johnson and Ascher, 1987). The simultaneous binding of glutamate and the co-agonist is required to activate the NMDAR (Johnson and Ascher, 1987; Furukawa et al., 2005; Paoletti and Neyton, 2007). Although the

physiological significance of co-agonist glycine is unclear, the recent works have reported the involvement in induction of LTD (Papouin et al., 2012), morphogenesis of dopaminergic neurons (Schmitz et al., 2009, 2013).

### Mammalian Glycine Receptor Subunits

There exist four GlyR α subunit genes (GLRA1, GLRA2, GLRA3, GLRA4) and one β subunit gene (GLRB) in humans and rodents (Grenningloh et al., 1987, 1990b; Kuhse et al., 1990, 1991; Harvey et al., 2000), although the human GLRA4 is not expressed due to a premature stop codon in this gene, thus human GLRA4 is pseudogene (Simon et al., 2004). Electrophysiological studies on cultured cells expressing mammalian GlyR subunits have shown that the difference in conductance and kinetics between α1β/α3β and α2β, the conductance of α2β is larger than that of the former, activation kinetics of α2 homomeric and α2β heteromeric is slower than α1/α3 containing receptor (Takahashi et al., 1992; Bormann et al., 1993; Rajendra et al., 1995; Beato et al., 2002; Mangin et al., 2003; Burzomato et al., 2004; Zhang et al., 2015). The adult mammalian hindbrain and spinal cord predominantly expresses the α1 (GLRA1) and β (GLRB) GlyR subunits (Malosio et al., 1991; Singer et al., 1998; Jonsson et al., 2012; Weltzien et al., 2012). Hence, it is assumed that heteromeric α1β GlyRs mediate a majority of glycinergic neurotransmission in these matured neural tissues. In this physiological context, it has been shown that glycinergic synaptic transmission is important for the generation of rhythmic motor behaviors. Functional loss of human GLRA1 due to missense, nonsense, or frame-shift mutation leads to the development of a hyperekplexia syndrome that is characterized by various exaggerated startle responses to unexpected acoustic or tactile stimuli, as well as neonatal apnea (Harvey et al., 2008a; Davies et al., 2010; Bode and Lynch, 2014). In addition, mutations that are associated with hyperekplexia syndrome have been identified in the GlyR β subunit gene (Rees et al., 2002; Al-Owain et al., 2012; Chung et al., 2013; James et al., 2013; Mine et al., 2013; Rizk and Mahmoud, 2014), the gephyrin gene (Rees et al., 2003), and the collybistin gene (Harvey et al., 2004). Furthermore, mutation of the glycine transporter 2 (GlyT2) gene (slc5a6) leads to a reduction in presynaptic glycine release and also produces hyperekplexia (Harvey et al., 2008a).

Also α3 subunits are expressed in the spinal cord as heteromeric GlyRs. The α3 subunit distribution is restricted to the superficial layer of the adult dorsal horn of the spinal cord (Harvey et al., 2004), and the α3 subunit is involved in nociceptive processing by modulating the excitability of projection neurons that relay nociceptive input from peripheral afferents (Harvey et al., 2004, 2009; Zeilhofer, 2005). Prostaglandin E2, an important inflammatory mediator, induces protein kinase A-dependent phosphorylation at Ser346 of the α3 subunit to depress glycinergic synaptic transmission and facilitate inflammatory pain (Ahmadi et al., 2002; Harvey et al., 2004; Zeilhofer, 2005). Although the α3 knockout mice exhibited neither morphological abnormality nor neuro-motor phenotype, the prostaglandin E2 induced pain sensitization was abolished in the α3 knockout mice (Harvey et al., 2004). Hence the α3 subunit may therefore be a promising target for the regulation of inflammatory pain. In fact, cannabinoids that potentiate α3 subunit exhibited analgesic effect to chronic inflammatory pain without psychoactive side effect (Xiong et al., 2012). The α3 phosphorylation induced conformational change specifically on glycine binding site of the α3 subunit, because the serine residues is not conserved at the equivalent position of the α1 subunit (Harvey et al., 2004; Han et al., 2013). The specific conformational change may be a help to develop other type specific drugs potentiating the phosphorylated α3 subunit. Thus, the glycinergic synaptic transmission in spinal cord is mediated by the α1β and α3β heteromeric receptor, and synaptic transmission is an attractive for the development effective remedies. In addition to the pain regulation in spinal cord, changes in the activity of hippocampal α3 subunit are also associated with temporal lobe epilepsy (Eichler et al., 2008; Legendre et al., 2009; Winkelmann et al., 2014).

The α2 subunit mRNA encoded by GLRA2 gene is predominantly expressed in the developing spinal cord, and then the α2 subunit is largely replaced by the α1 subunit in this regions within 2 weeks after birth in mice (Kuhse et al., 1990; Malosio et al., 1991; Watanabe and Akagi, 1995; Singer et al., 1998; Liu and Wong-Riley, 2013). Functional α2 homomeric GlyRs also found in embryonic immature cortex neurons (Flint et al., 1998; Young-Pearse et al., 2006). Although a previous study using α2 knockout mice showed no morphological or molecular alterations in nervous system development (Young-Pearse et al., 2006), recent analysis in newly established α2 knockout mice indicated that the α2 subunit contributes to several neural development process, such as tangential migration in developing cortex (Avila et al., 2013), cerebral cortical neurogenesis (Avila et al., 2014), morphogenesis and synaptogenesis of somatosensory cortical neuron (Morelli et al., 2016). The importance of α2 subunit in development and maturation of brain was also underscored by the recent identification of a micro-deletion and two mutations in GLRA2 gene from patients with autism spectrum disorder (Pinto et al., 2010; Pilorge et al., 2015).

After the developmental switching in the spinal cord, α2 and α3 subunits are still expressed as the predominant subunits in some regions of adult brain such as hippocampus and frontal cortex; in these regions the GlyRs contribute to regulation of neural excitability and synaptic plasticity (Chattipakorn and McMahon, 2002; Song et al., 2006; Zhang et al., 2006, 2008; Eichler et al., 2009; Kubota et al., 2010; Aroeira et al., 2011; Jonsson et al., 2012). GlyR β subunit mRNA was abundantly detected throughout the embryonic and adult brain, from olfactory bulb to spinal cord (Fujita et al., 1991; Malosio et al., 1991). However, a recent immunohistochemical study using a novel monoclonal antibody to the GlyR β subunit exhibited distinctive punctate staining of the β subunit at synaptic sites only in spinal cord, brainstem, midbrain, olfactory bulb, and retina of adult mice (Weltzien et al., 2012). In contrast to these regions, only weak diffuse immunostaining signals were detected in the hypothalamus, the cerebellum, the hippocampus and the neocortex of adult mice (Weltzien et al., 2012). These observations suggest that most of GlyRs in adult brain are extrasynaptic homopentamer, as presented in previous studies about hippocampal GlyR (Chattipakorn and McMahon, 2002; Zhang et al., 2008; Aroeira et al., 2011). It has been suggested

that the tonic inhibition by the neocortex GlyRs have antiepileptic effect (Chattipakorn and McMahon, 2003; Kirchner et al., 2003; Zhang et al., 2008; Shen et al., 2015). Epilepsy upregulated the expression of high-affinity GlyR through RNA editing in GlyR mRNA; this high-affinity GlyR may facilitate tonic inhibition to suppress the epileptic activity (Eichler et al., 2008).

In other region of adult brain, GlyRs are considered to participate in drug addiction. Studies using rodent models have revealed that some of the cortex GlyRs are involved in the behavioral effect of ethanol, such as a motor incoordination or a hypnosis (Quinlan et al., 2002; Ye et al., 2009; Blednov et al., 2012; McCracken et al., 2013a,b; Aguayo et al., 2014), and in the ethanol preference (Blednov et al., 2015). Ethanol increases extracellular dopamine levels in nucleus accumbens of the brain reward system through enhancement of the accumbal GlyR activity (Molander and Söderpalm, 2005; Molander et al., 2005; Maguire et al., 2014). The accumbal GlyRs are also involved in the dopamine-elevating effects of tetrahydrocannabinol and nicotine in the nucleus accumbens (Jonsson et al., 2014). Since dopamine elevation is a basis of drug addiction, the accumbal GlyR could be therapeutic targets of the addiction of ethanol or nicotine.

In addition to the expression within the brain, immunolabeling studies have demonstrated that all α subunits are also expressed in retina, especially in inner plexiform layer (IPL), with subunit -specific expression patterns (Sassoè-Pognetto et al., 1994; Haverkamp et al., 2003, 2004; Jusuf et al., 2005; Heinze et al., 2007; Weiss et al., 2008; Zhang et al., 2014), suggesting that different GlyR α subunits may be involved in different levels of visual processing (Wässle et al., 2009). Monoclonal antibody against the GlyR β subunit revealed that the β subunits were almost completely colocalized with GlyR α1-3 subunits and gephyrin in the IPL of mouse retina, however, GlyR α4 subunit did not show the highly colocalization with the β subunits (Weltzien et al., 2012). Thus, most of the retinal GlyR are αβ heteromeric receptor. The physiological role of these glycinergic synapses have been investigated by pharmacological and genetic approaches. For example, strychnine, a specific GlyR blocker, mediated blockade of glycinergic feedback pathway that from neuron in IPL to photoreceptor cell reduced the amplitudes of light-evoked response in both ON and OFF bipolar cells, indicating that the feedback pathway regulates signal propagation in the distal retina (Jiang et al., 2014). Genetic manipulations of GlyR α subunits in vivo have informed the significance of GlyR α subunit expression in the retina. Inhibitory glycinergic transmission between amacrine and bipolar cells of the retina is absent in GLRA1-deficient mice (Ivanova et al., 2006), and the frequency of spontaneous glycinergic input to A-type ganglion cells is also significantly reduced in the GLRA1-deficient retina (Majumdar et al., 2007). A study of GLRA2 knockout mice revealed that α2-containing GlyRs are involved in glycinergic input to group II wide-field amacrine cells (Majumdar et al., 2009). In the GLRA2-deficient retina, glycine-evoked currents are also reduced in MA-S5 and ON-starburst amacrine cells, suggesting that the α2 subunit is also involved in information processing by the amacrine cells. In GLRA3 deficient mice, spontaneous glycinergic currents are absent in AII amacrine cells (Weiss et al., 2008), additionally suggesting a role for the α3 subunit in these cells. In addition, α2 and α3 subunits enhance the excitatory center response of retinal receptive fields through modulation of local receptive field interactions (Nobles et al., 2012). Since the kinetics of glycinergic transmission differ according to α subunit type, signaling via different GlyRs may facilitate the complex regulation of retinal neuronal networks in visual processing.

### Mammalian Glycine Transporters

Two glycine transporters have been identified in the mammalian CNS: glycine transporter 1 (GlyT1) and 2 (GlyT2). Both of these transporters belong to the Na+/Cl<sup>−</sup> dependent transporter family and mediate the uptake of glycine from the extracellular space into the cytosol (Eulenburg et al., 2005; Betz et al., 2006). GlyT1 is primarily expressed in glial cells throughout the CNS (Eulenburg et al., 2005; Betz et al., 2006). Glycinergic transporter 1 regulates glycine concentrations in the synaptic cleft of glycinergic synapses and glutamatergic synapses, where glycine acts as an essential co-agonist of NMDAR (Eulenburg et al., 2005; Betz et al., 2006). GlyT1 knockout mice display respiratory insufficiency as a result of elevated synaptic glycine and over-activation of GlyRs in the respiratory pathway (Gomeza et al., 2003a). Alternatively, GlyT2 is expressed in glycinergic neurons and mainly localizes to the presynaptic terminal. GlyT2 facilitates the uptake of glycine released into the presynaptic terminal in order to limit the extent of neurotransmission and allow the recycling of glycine for future synaptic transmission. Accordingly, GlyT2 knockout mice display attenuated glycinergic synaptic transmission (Gomeza et al., 2003b).

#### GLYCINERGIC NEUROTRANSMISSION IN NON-MAMMALIAN SYSTEMS

Many investigations to reveal the physiological role of glycinergic synapse have been conducted in the mammalian model animals or human; however, non-mammalian vertebrates have been used for investigation the physiological roles of glycinergic synapse in locomotive behaviors (Drapeau et al., 2002; Grillner, 2003; Korn and Faber, 2005; Roberts et al., 2008). Most teleosts have Mauthner cells (M-cells), which are paired large neurons found in their hindbrain (Eaton et al., 1977; Zottoli, 1977). The M-cell is activated by auditory input, and the M-cell firing triggers fast escape behaviors (Kohashi and Oda, 2008; Kohashi et al., 2012). Both excitatory (e.g., glutamatergic) and inhibitory (e.g., glycinergic and GABAergic) synapses are formed on dendrites and soma of the M-cell (Korn and Faber, 2005). In goldfish, glycinergic input to M-cells was enhanced by tetanization of the auditory pathway (Korn et al., 1992) or repeated sound stimulation (Oda et al., 1998). Glycinergic synaptic transmission reduces M-cell excitability by countering concurrent excitatory synaptic input (Korn and Faber, 2005; Curtin and Preuss, 2015). Thus, the plasticity of the glycinergic synapse has been shown to regulate auditory conditioning of the escape response (Oda et al., 1998). Recently, zebrafish (Danio rerio) have also been used as a model for investigating the physiological roles of glycinergic

neurotransmission (Cui et al., 2005; Hirata et al., 2005; Downes and Granato, 2006; Rigo and Legendre, 2006; Mongeon et al., 2008). In this review, we will describe advances in both forward and reverse genetic approaches as well as visualization techniques that have proven helpful for the study of the glycinergic synapse in zebrafish.

#### Zebrafish Glycine Receptors

In total, zebrafish express five α subunits (glra1, glra2, glra3, glra4a and glra4b) and two β subunits (glrba and glrbb) (Hirata et al., 2010). Four GlyR α subunits (αZ1, αZ2, αZ3, and αZ4) and two β subunits (βa/βZ and and βb) were initially reported in zebrafish (David-Watine et al., 1999; Imboden et al., 2001a,b,c; Hirata et al., 2005). Although αZ2 was originally thought to encode a GlyR α2 subunit (Imboden et al., 2001a), the gene was subsequently reclassified as a second α4 subunit in a more detailed phylogenetic analysis (Imboden et al., 2001b). Thus, αZ2 was designated glra4a. Phylogenetic analyses have suggested that αZ1, αZ3, and αZ4 subunits are orthologs of the mammalian α1, α3, and α4 GlyR subunits, respectively (Imboden et al., 2001b); thus, these subunits have been referred to as the zebrafish GlyRα1, GlyRα3, and GlyRα4b. Of note, the existence of distinct orthologs of a mammalian gene is common in the zebrafish genome due to suspected whole genome duplication during fish evolution (Amores et al., 1998). The spatial expression of GlyR genes has been examined using in situ hybridization (**Table 1**). Although the spatial expression patterns of glra2 and glra3 remain to be determined, quantitative PCR analysis of developing zebrafish embryos (24–72 hpf) has revealed the temporal expression patterns for these genes: glra1, glra3, and glra4a were observed at 24 hpf, whereas glra2 expression was not induced until 32 hpf (Ganser et al., 2013). Moreover, the expression levels of glra1, glra2, and glra4a were noted to steadily increase during development, while the expression of glra3 subsided after 48 hpf (Ganser et al., 2013). Knock down experiments using antisense morpholino oligo (MO) have reported the specific involvement of glra4a in the differentiation of spinal interneurons (McDearmid et al., 2006).

#### The Advantages of Zebrafish as a Vertebrate Model

Zebrafish offer several advantages as a model for vertebrate development. First, zebrafish quickly reach sexual maturity (within 3 months), and the adult zebrafish lay 100–200 eggs twice a week, such that fertilized eggs are readily available throughout the year with a relatively low cost, because a large number of zebrafish can be maintained in smaller space compared to model mammals. Second, zebrafish embryos are optically transparent, and their development progresses rapidly at 28.5◦C (Kimmel et al., 1995). These characteristics are helpful for developmental studies in vivo. Indeed, in vivo imaging of events such as synapse formation and subcellular CaMKII translocation has been achieved through the use of green fluorescent protein tagging in zebrafish embryos (Gleason et al., 2003; Niell et al., 2004). Moreover, the transparency of early zebrafish embryos has enabled the in vivo study and manipulation of neural activity (Douglass et al., 2008; Arrenberg et al., 2009; Zhu et al., 2009; Schoonheim et al., 2010; Muto et al., 2011). While the transparency of the zebrafish embryo is progressively diminished by the formation of melanophores after 24 hpf, 1-phenyl 2 thiourea (PTU) is a widely used agent to prevent the pigment formation (Karlsson et al., 2001). Alternatively, because it is not feasible to grow zebrafish into adulthood in the presence of PTU, a pigment-deficient mutant known as casper is now available, and facilitates the in vivo visualization and manipulation of neural activity in adult zebrafish (White et al., 2008). Third, zebrafish are suitable for electrophysiological analyses. The activity of neurons and muscles can be recorded using standard patch-clamp and extracellular recording techniques during most stages of zebrafish development (Grunwald et al., 1988; Legendre and Korn, 1994, 1995; Ribera and Nüsslein-Volhard, 1998; Drapeau et al., 1999;


Saint-Amant and Drapeau, 2001; Sidi et al., 2003; Higashijima et al., 2004; Kimura et al., 2006; Fetcho, 2007; McLean et al., 2007; Tanimoto et al., 2009). Finally, the high-efficient mutagenesis and transgenesis are available in zebrafish, as discussed in a later section of this review.

In addition to the usefulness in developmental study, zebrafish is also valuable model for high-throughput screening of novel drugs (Peterson et al., 2000; Goldsmith, 2004; Love et al., 2004). The high-throughput screening using mutant zebrafish model of human disease have identified novel compound that potentially ameliorate the phenotype associated with the mutation (Peterson et al., 2004; Cao et al., 2009; Paik et al., 2010; Kawahara et al., 2011; Peal et al., 2011; Baraban et al., 2013).

#### The Development of Locomotive Behavior

Zebrafish exhibit three distinct behaviors during embryogenesis: spontaneous coiling, touch-evoked escape contractions, and swimming. Spontaneous coiling appears after 17 hpf, and consists of side-to-side alternating contractions of the axial muscles (Saint-Amant and Drapeau, 1998; Downes and Granato, 2006; Pietri et al., 2009). The frequency of spontaneous coiling reaches a peak of 0.3–1 Hz at 19 hpf and gradually declines to less than 0.1 Hz by 26 hpf. After 21 hpf, zebrafish embryos respond to tactile stimuli with escape contractions that typically consists of two-to-three rapid, alternating contractions of the axial muscles (Saint-Amant and Drapeau, 1998; Hirata et al., 2005). By 28 hpf, tactile stimuli initiate swimming (Saint-Amant and Drapeau, 1998). The frequency of swimming contractions reaches 30 Hz at 36 hpf, which is comparable to the frequency of tail contraction in adult zebrafish (Buss and Drapeau, 2001). Thus, the neural network regulating swimming is functionally matured within 1.5 days of development. Interestingly, head-removed embryos transected at somites 5–7 are capable of responding to touch with an initial tail flip, although swimming fails to follow in most cases (Downes and Granato, 2006). Indeed, detailed observations suggest that the spinal cord can initiate touch responses, while supraspinal input is necessary for swimming. The spinal cord region located between somites 5–10 somite appears to be responsible for spontaneous coiling (Pietri et al., 2009). Given this knowledge, a variety of in vivo and ex vivo manipulations are possible in zebrafish.

#### Mutagenesis and Transgenesis in Zebrafish

Mutagenesis for the forward genetic screening of zebrafish mutants was first established in the early 1990's (Mullins et al., 1994), and two large-scale N-ethyl-N-nitrourea (ENU)-based mutagenesis projects were completed in Tübingen, Germany and Boston, USA by 1996. These screens identified more than 4,000 mutants, including motility defective mutants, caused by the dysfunction of glycinergic neurotransmission.

Additionally, more comprehensive ENU-based mutagenesis project named "Zebrafish Mutation Project (ZMP)" was launched in 2011. ZMP identified potentially disruptive mutations in more than 38% of all known zebrafish protein-coding genes (Kettleborough et al., 2013). The forward genetic approach provides an opportunity to identify novel genes participating in the investigating subject. The growing list of disruptive mutated alleles will be a valuable resource both for fundamental and clinical research to reveal the interested gene function.

On the other hand, reverse genetic approaches has also been applied to zebrafish biology. Targeting-induced local lesions in genomes (TILLING) has been introduced to zebrafish as the first available reverse genetics method (Wienholds et al., 2002). TILLING is a combinational method of chemical-induced mutagenesis and high-throughput screening; accordingly, TILLING can generate not only loss of function mutations, but also unexpected mutations due to non-directional missense mutagenesis (Wienholds et al., 2002, 2003; Moens et al., 2008). However, TILLING is time-consuming and less effective for intron-rich genes due to a decreased chance of obtaining null or hypomorphic alleles (Stemple, 2004). Subsequently, Zinc finger nucleases (ZFNs) (Doyon et al., 2008; Meng et al., 2008; Foley et al., 2009; Sander et al., 2011a) and transcription activator-like effector nucleases (TALEN) (Sander et al., 2011b; Bedell et al., 2012; Cade et al., 2012; Dahlem et al., 2012) were introduced to zebrafish genetics as targeted mutagenesis methods. ZFNs utilize an array of zinc finger DNA-binding motifs that bind to specific DNA triplet sequences (Urnov et al., 2005; Carroll, 2011). However, the exact DNA binding specificity of ZFNs has not been completely resolved (Carroll, 2011) and thus each ZFN construct requires careful design and screening of its target DNA-binding zinc finger motifs. Alternatively, TALEN DNA binding specificity is more predictable (Miller et al., 2011). The DNA-binding motif of TALEN is composed of repeated modules, where each module independently binds to a specific single nucleotide through repeat variable di-residues in a 1:1 manner. Thus, each module is a functional unit for the recognition of DNA sequences (Boch et al., 2009; Moscou and Bogdanove, 2009). This simple rule of nucleotide recognition has made TALEN more popular than ZFN for zebrafish genetics.

More recently, the clustered regularly interspaced palindromic repeats/CRISPR-associated 9 (CRISPER/Cas9) method was applied to zebrafish as a surprisingly simple and effective method. CRISPR and Cas are involved in bacterial adaptive immunity against the invasion of foreign nucleic acids derived from exogenous plasmids or bacteriophages (Barrangou et al., 2007; Garneau et al., 2010; Horvath and Barrangou, 2010). In type II CRISPR/Cas systems, Cas9 nuclease is guided to the target site by two RNA molecules, tracrRNA and crRNA. Alternatively, the use of a single chimeric guide RNA, generated by fusing the 3<sup>0</sup> end of a crRNA to the 5<sup>0</sup> end of a tracrRNA, has been used to guide Cas9 to its target site (Jinek et al., 2012). Thus, instead of the need to design a DNA binding domain consisting of several 10–100s amino acids as in the case of ZFN and TALEN, a single small guide RNA can be used in CRISPR/Cas9. Codon-optimized Cas9 with guide RNA efficiently induces target sequence-specific genome modification in humans and mammals (Cong et al., 2013; Li D. et al., 2013; Li W. et al., 2013; Mali et al., 2013). In zebrafish, CRISPR/Cas9 produces targeted gene

lesion with a similar efficiency to that of TALENs (Chang et al., 2013; Hwang et al., 2013). Furthermore, the mutagenesis efficiency was improved with zebrafish codon optimized Cas9 (Jao et al., 2013; Liu et al., 2014). The off-target effects of the CRISPR/Cas9 method are reported to be quite limited (Chang et al., 2013; Hruscha et al., 2013; Hwang et al., 2013; Jao et al., 2013), although some studies in cultured cells have reported high off-target mutagenesis (Fu et al., 2013; Hsu et al., 2013). This complication can be resolved through back-crossing with wildtype zebrafish. CRISPE/Cas9 system is able to introduce not only an in-del mutation leading to gene knockout, but also exogenouse coding DNA in site specific manner (Auer et al., 2014; Kimura et al., 2014; Hisano et al., 2015). The site specific insertion allows researchers to substitute any nucleotides or amino acids, or to tagging any protein with fluorescent protein or epitope tag.

Other than the knockout technologies, MO injection into fertilized eggs has been widely used as gene knockdown method (Nasevicius and Ekker, 2000). However, in some cases MO-induced phenotypes (morphant) were not observed to correspond with equivalent mutants generated by ZFN and TALEN (Kok et al., 2015). Phenotypic discrepancies between morphants and mutants may be due to the off-target effects of MO (Kok et al., 2015). Alternatively, genetic compensation induced by deleterious mutation has also been reported as a cause of phenotype variation between morphants and mutants, as compensatory gene upregulation is not typically observed in MO-injected embryos (Rossi et al., 2015). Regardless of the cause of phenotypic discrepancy, antisense MOs should be used with meticulous care, and consider dose dependency as well as the implementation of proper positive and negative controls.

Transposon or retroviral methods have also been used to generate transgenic zebrafish (Davidson et al., 2003; Kawakami et al., 2004; Ellingsen et al., 2005; Villefranc et al., 2007; Asakawa et al., 2008) with improved efficiency (Davidson et al., 2003; Kawakami et al., 2004; Ellingsen et al., 2005; Sivasubbu et al., 2006; Villefranc et al., 2007). This increased efficiency has facilitated the use of the yeast GAL4/UAS system (Scheer and Campos-Ortega, 1999; Inbal et al., 2006; Scott et al., 2007; Asakawa et al., 2008; Halpern et al., 2008). GAL4, a yeast transcriptional activator, is capable of binding UAS sequences and driving the expression of downstream genes (Brand and Perrimon, 1993; Asakawa and Kawakami, 2008). Thus, a gene of interest can be expressed in any tissue by crossing UAS transgenic zebrafish with tissue-specific GAL4 driver transgenic zebrafish. Recently, various GAL4 driver transgenic zebrafish have been established using the Tol2 transposon-mediated enhancer trap method (Asakawa and Kawakami, 2008; Kawakami et al., 2010). In fact, the GAL4/UAS system has proven to have great utility for in vivo imaging and the manipulation of neural activity in zebrafish (Douglass et al., 2008; Arrenberg et al., 2009; Zhu et al., 2009; Schoonheim et al., 2010; Muto et al., 2011, 2013). However, GAL4/UAS systems also have a disadvantage. It has been reported that the methylation at CpG nucleotides in the UAS sequences could silence the transgene expression in zebrafish that is offspring of the founder (Goll et al., 2009; Akitake et al., 2011; Pang et al., 2015). Although tryptophan repressor (TrpR) and its upstream activation sequence (tUAS) was proposed as an alternative gene expression system to overcome the silencing effect of Gal4/UAS, TrpR/tUAS system gave toxic effect to developing zebrafish may be due to strong transcriptional activity or toxic effect of TrpR protein (Suli et al., 2014). Four distinct Gal4 binding sites that placed in tandem (4x nr UAS) drove high levels of reporter expression with significantly less susceptible to the methylation compare to fourteen tandem copies of UAS (Akitake et al., 2011). Therefore, the fewere non-repetitive UAS sequence could be useful to circumvent the silencing by the methylation.

#### The Non-invasive Imaging of Neuronal Excitability

Inhibitory glycinergic and GABAergic synaptic inputs critically regulate neuronal excitability. Electrophysiological techniques are available for the monitoring of single neuron excitability through membrane potential and synaptic current recordings; however, it is difficult to investigate network excitability using this approach. An alternative approach is the monitoring of Ca2<sup>+</sup> transients as a surrogate of excitability. To this end, genetically encoded calcium indicators have been used for Ca2<sup>+</sup> imaging in zebrafish (Muto et al., 2011). GCaMP, the most widely used genetically encoded calcium indicator, is a fusion protein of circularly permuted EGFP, calmodulin, and calmodulininteracting M13 peptide (Baird et al., 1999; Nakai et al., 2001). Although the fluorescence intensity of circularly permuted EGFP is very low, conformational changes induced by Ca2<sup>+</sup> binding to calmodulin lead to the enhancement of fluorescence intensity and improved detection. The sensitivity of this method has been improved by the generation of GCaMP variants, which are now available for research use (Ohkura et al., 2005; Tallini et al., 2006; Tian et al., 2009; Akerboom et al., 2012; Chen et al., 2013; Muto et al., 2013). Muto et al. (2011) have successfully used this technique to visualize the alternating activation of spinal motor neurons during spontaneous coiling, and neural activity in the optic tectum during prey capture behavior (Muto et al., 2013).

### ZEBRAFISH MOTILITY DEFECT MUTANTS

In the Tübingen screen, 166 mutants showing motility defects between 48 and 60 hpf were identified (Granato et al., 1996). Among them, 63 mutants were assumed to be muscle-defective based on the simple observation of actin-myosin organization (birefringence intensity) under polarized light (Felsenfeld et al., 1990). Mutations in dystrophin, laminin, titin, Hsp90 and the cognate co-chaperone Unc45b were identified in this manner (Bassett et al., 2003; Etard et al., 2007; Hall et al., 2007; Steffen et al., 2007; Hawkins et al., 2008; Guyon et al., 2009). The other 103 mutants with normal muscle structure were divided into two groups: locomotion abnormal and reduced motility (Granato et al., 1996), and assumed to have impairments

in the nervous system, neuromuscular junction (NMJ), or muscular functional components. Electrophysiological analyses have been useful in delineating the exact cause of each abnormality. For example, electrophysiological recordings from the accordion (acc) mutant, a locomotion abnormal mutant, revealed normal neuronal outputs and thus implicated deficits in the functional components of the muscle (Hirata et al., 2004). In the atp2a1 mutant, it was determined that the simultaneous contraction of the bilateral trunk muscles results from the impaired clearance of cytosolic Ca2<sup>+</sup> by the sarcoplasmic reticulum Ca2+-ATPase SERCA1 (Gleason et al., 2004; Hirata et al., 2004). Several other muscle mutants (relaxed, relatively relaxed) have exhibited defects in excitation-contraction coupling (Schredelseker et al., 2005, 2009; Zhou et al., 2006; Hirata et al., 2007).

Six abnormal locomotive mutants (bandoneon, ziehharmonika, bajan, diwanka, quetschkommode, and expander) were previously compared with the acc mutant (Granato et al., 1996). In contrast to acc, electrophysiological recordings from the bandoneon (beo) mutant showed aberrant, arrhythmic fluctuations in response to tactile stimulation, indicating a defect in CNS output (Hirata et al., 2005). As will be described in detail, a molecular genetic study revealed that alteration of the glrbb gene is responsible for the beo mutant, and thus beo exhibits defects in glycinergic synaptic transmission (Hirata et al., 2005). Of note, beo is the only known acc subgroup mutant with dysfunctional glycinergic transmission (**Table 2**).

In addition to beo, the shocked (sho) mutant also shows reduced mobility due to a defect in glycinergic synaptic transmission. The sho gene encodes the glycine transporter GlyT1 (Cui et al., 2005; Mongeon et al., 2008). Based on its function, the loss-of-function mutation of GlyT1 should increase the availability of extracellular glycine in the CNS. Overactivation of the glycinergic synapse by elevated extracellular glycine can suppress the neural network responsible for motility, and therefore result in a phenotype of reduced mobility. Of note, strychnine, a specific inhibitor of GlyR, partially restores normal neuronal activity in sho mutants (Cui et al., 2005). Other


mutants with reduced mobility have gene mutations unrelated to glycinergic transmission (**Table 3**).

### Defective GlyR Clustering in the bandoneon Mutant

The tubingen mutant screen identified seven mutant alleles (tp221 = Y79X, tw38f = L255R, ta86d = Y79X, ta92 = K343X, tm115 = Q87X, tf242 = Y79D, and tu230 = lost) of glrbb (Granato et al., 1996; Ganser et al., 2013). In addition, we have isolated an eighth allele (mi106a = R275H) in a previous mutagenesis screen (Hirata et al., 2005). These

#### TABLE 3 | Mutants with reduced locomotion.

zebrafish glrbb mutants were named bandoneon after the South American accordion-like instrument. Phenotypically, the bandoneon mutations show simultaneous contraction of both axial muscles instead of alternating in response to touch stimuli; therefore, the mutant body length is shortened by tactile stimuli, similar to the movement of an accordion (Hirata et al., 2005). Since axial muscle contractions are controlled by the reciprocal inhibition of the left and right sides of the spinal cord, inhibitory synaptic transmission is required to produce alternating contractions for swimming (Grillner, 2003; Roberts et al., 2008). As abovementioned, GlyR clustering at synaptic


sites is necessary for effective glycinergic transmission and motor pattern generation. In bandoneon mutants, GlyR cluster immunostaining in the spinal cord has a diffuse appearance rather than a clustered appearance, suggesting that GlyRb is necessary for the synaptic aggregation of αβ GlyRs in zebrafish (Kirsch et al., 1993; Meyer et al., 1995; Feng et al., 1998; Kim et al., 2006).

#### Defective GlyT1 in the shocked Mutant

Shocked (sho) carries a mutation in slc6a9, which encodes GlyT1. Three mutant alleles of slc6a9 were identified in the Tübingen screen (ta229g = G81D, te301 = C305Y and ta51e = unknown), and these sho mutants exhibit trunk twitching instead of swimming in response to tactile stimuli at 2 dpf (Granato et al., 1996; Luna et al., 2004; Cui et al., 2005; Mongeon et al., 2008). In addition, the frequency of spontaneous coiling is reduced and the escape contraction at 1 dpf is abolished in sho mutants. An electrophysiological analysis revealed that tactile stimuli induces arrhythmic muscle activation in sho mutants rather than the rhythmic depolarization observed in wild-type muscle (Cui et al., 2004). The ta229g allele, which produces the strongest phenotype, results from a G81D missense mutation that disrupts GlyT1 function (Cui et al., 2005). Increased extracellular glycine is thought to be the cause of the sho mutant phenotype. This presumption is supported by the following observations. First, the perfusion of glycine-free cerebrospinal fluid following the removal of the dorsal roof of the fourth ventricle recovers touchevoked swimming in sho mutants, and this effect is lost when glycine is added to the perfusate (Cui et al., 2005; Mongeon et al., 2008). Second, the application of low concentrations of strychnine to sho mutants leads to partial motor recovery, similar to observations in GlyT1 knockout mice (Gomeza et al., 2003a; Cui et al., 2005). Thus, a lack of functional GlyT1 and elevated extracellular glycine are likely to underlie the potentiation of glycinergic transmission in the sho phenotype.

The GlyT1 knockout mice exhibited severe motor and respiratory defects and die at birth due to the respiratory defect (Gomeza et al., 2003a), whereas the GlyT1 defective zebrafishes exhibit a motility recovery by 4–5 dpf and survive thereafter (Mongeon et al., 2008). The motility recovery was accompanied by a reduction in GlyR expression that leads to a decrease in the amplitude of inhibitory potentials enhanced by the GlyT1 mutant (Mongeon et al., 2008). Although the alteration of GlyR expression may represent compensatory mechanisms, the molecular basis for this compensation is not revealed.

Glycinergic transporter 1 dysfunction has been posited in the pathogenesis of glycine encephalopathy, which is characterized by respiratory impairment (Gomeza et al., 2003a; Applegarth and Toone, 2004, 2006; Harvey et al., 2008b), and schizophrenia which may be arose from hypofunction of NMDAR (Krystal et al., 1994; Lahti et al., 2001; Tsai et al., 2004; Dalmau et al., 2007). Animal models are clearly required to investigate the biological roles of GlyT1 in the context of human disorders, and zebrafish may provide particular advantages in future efforts. For example, while GlyT1 knockout mice die on the first postnatal day due to severe respiratory deficits (Gomeza et al., 2003a; Tsai et al., 2004), zebrafish sho mutants can survive well into adulthood through careful feeding and are thus useful as a physiologically relevant animal model for GlyT1 dysfunction.

### CONCLUSION

Functional GlyRs are α homomeric or αβ heteromeric pentamers. Electrophysiological studies of α subunits in cultured cells have shown the difference in conductance and kinetics of each α subunit (Takahashi et al., 1992; Bormann et al., 1993; Rajendra et al., 1995; Beato et al., 2002; Mangin et al., 2003; Burzomato et al., 2004; Zhang et al., 2015). Moreover, both in zebrafish and in mammals, each α subunit exhibits a distinct pattern of temporal and spatial expression in the CNS. This diversity of the α subunits suggests that GlyRs play various roles in physiological functions. However, the details of these physiological roles are not well understood. Several studies have identified GlyR subunit selective modulator in Cannabinoids and endocannabinoid (Hejazi et al., 2006; Yang et al., 2008; Xiong et al., 2012), in compounds isolated from ginkgo and Australian marine sponges (Balansa et al., 2010, 2013a,b; Maleeva et al., 2015). Furthermore, recent publications have reported that heptapeptide exhibit the subunit selective modulating effects on α1β and α3β heteromeric receptor with zinc ion dependence (Cornelison et al., 2016). These subunitselective modulators would be useful tool to investigate the physiological roles of GlyR subunits in CNS. Together with the selective modulator, losses of function mutants of each GlyR subunit are powerful tools for uncovering the specific functions of each GlyR subunit. Now, zebrafish provide a cheaper, more convenient and highly useful alternative animal model for the study of GlyRs and glycinergic transmission. So far, the glrbb and slc6a9 mutants have been used to investigate the glycinergic synapse. Future work can further utilize CRISPR/Cas9-mediated gene disruption in zebrafish. In addition, in vivo observations of the glycinergic synapse in living zebrafish are a powerful method for the study of glycinergic synapse formation and plasticity. Taken together, a combination of readily available genetic mutants and innovative imaging techniques in zebrafish is sure to accelerate our understanding of glycinergic neurotransmission in the future.

### AUTHOR CONTRIBUTIONS

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

#### ACKNOWLEDGMENTS

We apologize to investigators whose work could not be cited in this manuscript owing to space limitations. This work was supported by a Grant-in-Aid for Young Scientists (B) from the Ministry of Education, Culture, Sports, Science, and Technology of Japan (MEXT) to KO and Grant-in-Aid for Scientific Research (B) from the MEXT, the Takeda Science Foundation, the Mochida Memorial Foundation for Medical and Pharmaceutical Research, the Naito Foundation, and the Suzuken Memorial Foundation and the Japan Epilepsy Research Foundation to HH.

#### REFERENCES








dihydropyridine receptor β1-null zebrafi sh relaxed is an exclusive function of the β1a subunit. J. Biol. Chem. 284, 1242–1251. doi: 10.1074/jbc.M807767200



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

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

# Age-Dependent Degeneration of Mature Dentate Gyrus Granule Cells Following NMDA Receptor Ablation

*Yasuhito Watanabe1, Michaela K. Müller2,3, Jakob von Engelhardt2,3, Rolf Sprengel4, Peter H. Seeburg4 and Hannah Monyer1\**

 *Department of Clinical Neurobiology, University Hospital and German Cancer Research Center Heidelberg, Heidelberg, Germany, <sup>2</sup> Synaptic Signalling and Neurodegeneration, German Center for Neurodegenerative Diseases, Bonn, Germany, Synaptic Signalling and Neurodegeneration, German Cancer Research Center Heidelberg, Heidelberg, Germany, Department of Molecular Neurobiology, Max Planck Institute for Medical Research, Heidelberg, Germany*

*N*-methyl-*D*-aspartate receptors (NMDARs) in all hippocampal areas play an essential role in distinct processes of memory formation as well as in sustaining cell survival of postnatally generated neurons in the dentate gyrus (DG). In contrast to the beneficial effects, over-activation of NMDARs has been implicated in many acute and chronic neurological diseases, reason why therapeutic approaches and clinical trials involving receptor blockade have been envisaged for decades. Here we employed genetically engineered mice to study the long-term effect of NMDAR ablation on selective hippocampal neuronal populations. Ablation of either GluN1 or GluN2B causes degeneration of the DG. The neuronal demise affects mature neurons specifically in the dorsal DG and is NMDAR subunit-dependent. Most importantly, the degenerative process exacerbates with increasing age of the animals. These results lead us to conclude that mature granule cells in the dorsal DG undergo neurodegeneration following NMDAR ablation in aged mouse. Thus, caution needs to be exerted when considering long-term administration of NMDAR antagonists for therapeutic purposes.

#### *Edited by:*

*Robert J. Harvey, UCL School of Pharmacy, UK*

#### *Reviewed by:*

*Hansjürgen Volkmer, Universität Tübingen, Germany Hansen Wang, University of Toronto, Canada*

*\*Correspondence: Hannah Monyer h.monyer@dkfz-heidelberg.de*

*Received: 06 November 2015 Accepted: 18 December 2015 Published: 12 January 2016*

#### *Citation:*

*Watanabe Y, Müller MK, von Engelhardt J, Sprengel R, Seeburg PH and Monyer H (2016) Age-Dependent Degeneration of Mature Dentate Gyrus Granule Cells Following NMDA Receptor Ablation. Front. Mol. Neurosci. 8:87. doi: 10.3389/fnmol.2015.00087*

Keywords: NMDA receptors, aging, neurodegeneration, hippocampus, dentate gyrus, mouse

#### INTRODUCTION

The dentate gyrus (DG) in the hippocampus has been the focus of numerous studies for two main reasons. Firstly, the DG is a major station in the trisynaptic pathway that conveys input from the cortex to the hippocampus. DG granule cells excite CA3 neurons, which in turn project to CA1 neurons. All hippocampal subregions have been much investigated in the context of distinct forms of plasticity, learning, and memory (e.g., McHugh et al., 2007; Niewoehner et al., 2007; Gu et al., 2012; Nakashiba et al., 2012). Secondly, the DG is one of the two principal neurogenic niches in the postnatal brain, and has been therefore frequently studied in the context of neurogenesis (reviewed in Jessberger and Gage, 2014). Some studies addressed questions that link these two research areas. Thus, young DG granule cells support pattern separation, whereas mature granule cells facilitate pattern completion (Nakashiba et al., 2012). The DG comprises a dorsal and a ventral part that can be distinguished on anatomical grounds, and also based on distinct functions that the two areas support. Thus, the dorsal DG is involved in cognitive functions such as spatial memory, whilst the ventral DG supports brain functions related to stress, emotion, and affect (reviewed in Fanselow and Dong, 2010).

Several decades of pharmacological and genetic research have taught us that *N*-methyl-*D*-aspartate receptors (NMDARs) are intricately linked to hippocampus-dependent memory (reviewed in Bannerman et al., 2014). Mouse genetics was fundamental to correlate subregions of the hippocampus, including the DG, with distinct functions supporting spatial memory (McHugh et al., 2007; Niewoehner et al., 2007). Thus, NMDARs in the DG are required for pattern separation (McHugh et al., 2007) and spatial working memory (Niewoehner et al., 2007). Furthermore, studies based on mouse mutants were instrumental in helping delineate the differential functions of NMDAR subunits. DG granule cells, like most neurons in the hippocampus, harbor NMDARs built from the obligatory GluN1 subunit (encoded by *Grin1*) and the developmentally regulated GluN2A and GluN2B subunits (encoded by *Grin2a* and *Grin2b*, respectively; Monyer et al., 1994; Rodenas-Ruano et al., 2012; Paoletti et al., 2013). NMDARs have been much discussed with respect to their function in postnatal neurogenesis. Thus, NMDARs in postnatally generated DG granule cells are a prerequisite for the survival of newborn neurons (Tashiro et al., 2006).

In mature neurons, over-activation or suppression of NMDARs are detrimental for cell function and survival. Thus, enhanced receptor activity as it occurs in many pathophysiological neurological diseases, e.g., stroke, epilepsy, induces excitotoxic cell death (reviewed in Choi, 1992; Paoletti et al., 2013). Conversely, pharmacological blockade of NMDARs for hours and days also has an adverse effect, as it induces cellular vacuolization in pyramidal neurons in many cortical brain areas (e.g., Ikonomidou et al., 1999). The effect of prolonged or even permanent NMDAR blockade on the survival of mature neurons has not yet been tested systematically. Moreover, a potential role of NMDARs for cell survival in aged animals has not been investigated. This is surprising, given that the NMDAR has been a key candidate when considering therapeutic strategies for stroke.

Here we took advantage of genetically modified mice and addressed the question whether the survival of mature and aged neurons depends on NMDARs. We demonstrate that in the hippocampus, mature neurons in the DG, but not in the CA1 region, require intact NMDAR function for survival.

#### MATERIALS AND METHODS

#### Animals

Animals were housed and handled according to the respected animal welfare guidelines and rules of the Max Planck Society and the Germany government animal welfare office in Karlsruhe, Germany. The following genetically modified male and female C57Bl/6N mice were used in this study: DG and CA1 selective *Grin1* knockout (*Grin1*-*DGCA*1) mice (Bannerman et al., 2012), DG selective *Grin1* knockout (*Grin1*-*DG*, previously called *NR1*-*DG*) mice (Niewoehner et al., 2007), *Grin2a* knockout (*Grin2a*−/−, previously called homozygous *GluR*ε*1* mutant) mice (Sakimura et al., 1995), and DG and CA1 selective *Grin2b* knockout (*Grin2b*-*DGCA*1, previously called *NR2B* -*HPC*) mice (von Engelhardt et al., 2008).

*Grin1*-*DGCA*<sup>1</sup> mice are homozygous for the floxed *Grin1* and carry two transgenes, *TgLC*<sup>1</sup> and *TgCN*12, which enable doxycycline-sensitive, Cre-mediated gene ablation in CA1 and DG excitatory neurons and piriform cortex in the adult brain by use of a *CamKIIa/Grin2c* hybrid promoter (Bannerman et al., 2012). *Grin1*-*DG* mice are homozygous for *Grin1* and carry the transgenes *TgLC*<sup>1</sup> and *TgCN*<sup>10</sup>−*itTA*, which are the same transgenes as in *TgCN*12, but yield differential expression in the two transgenic mouse lines. In *Grin1*-*DG* mice, Cre is specifically expressed in the DG granule cells and some CA1 pyramidal neurons (Niewoehner et al., 2007) which express calbindin (CB; data not shown). *Grin2b*-*DGCA*<sup>1</sup> mice are homozygous for the floxed *Grin2b* and carry *TgLC*<sup>1</sup> and *TgCN*12. For the selective expression of Cre in *Grin1*-*DGCA*1, *Grin1*-*DG* and *Grin2b*-*DGCA*<sup>1</sup> mice, doxycycline was given via the drinking water to pregnant mice to suppress Cre expression of the offspring during embryonic development, and was withdrawn after birth (Bannerman et al., 2012). Doxycycline treatment was performed under the license 35-9185.81/G71/10 of the governmental council in the Karlsruhe, Germany. In *Grin1*-*DGCA*<sup>1</sup> mice, Cre expression becomes detectable as early as postnatal day P28 (Bannerman et al., 2012). Control mice (i.e., mice with no Cre expression) to be compared with *Grin1*-*DGCA*1 and *Grin1*-*DG* mice were homozygous for floxed *Grin1* and carried one or none of the transgenes. Control mice for the *Grin2a* knockout mice are wild-type mice. Control mice for *Grin2b*-*DGCA*<sup>1</sup> mice were homozygous for floxed *Grin2b* and carried one or none of the transgene. As control mice for Cre expressing mice (carrying only *TgLC*<sup>1</sup> and *TgCN*<sup>12</sup> but no floxed gene) we used either wild-type mice or carriers of one of the transgenes. Following mice are available at EMMA (https://www.infrafrontier.eu) with respective EMMA ID: Floxed *Grin1* mice (EM: 09220), mice carrying *TgLC*<sup>1</sup> and *TgCN*<sup>12</sup> transgenes (EM: 09256), and mice carrying *TgCN*<sup>10</sup>−*itTA* transgene (EM: 09255). All animal experiments were performed according to the regulations of Heidelberg University/German Cancer Research Center/Max Planck Institute.

#### Immunohistochemistry

Mice were perfused with phosphate buffered saline (PBS, pH 7.4) followed by 4% paraformaldehyde (PFA). Brains were removed and postfixed with 2% PFA at 4◦C overnight. Coronal sections were prepared at 50 μm using a vibratome. Sections were blocked and permeabilized with 5% bovine serum albumin (BSA) in PBS containing 0.5% Triton X-100, incubated with primary antibodies overnight at 4◦C, then incubated with appropriate secondary antibodies. Primary and secondary antibodies were diluted in PBS containing 0.2% Triton X-100 before incubation. Sections were washed three times with PBS after every antibody incubation step. Sections were counterstained with DAPI (Invitrogen, Carlsbad, CA, USA) after the secondary antibody incubation step. For GluN1 staining, before blocking the sections, antigen retrieval was performed as follows. Sections were immersed in 10 mM sodium citrate (pH 6.0) and heated in a microwave oven for 5 min at 650 W twice with a 5 min break in between, kept for 20 min at room temperature, and washed twice with PBS. The following primary antibodies were used: rabbit active caspase-3 (catalog # AF835; 1:2000; polyclonal; R&D systems, Minneapolis, MN, USA), mouse CB D-28k (300; 1:2000; monoclonal; Swant, Switzerland), goat DCX (sc-8066; 1:500; polyclonal; Santa Cruz Biotechnology, Santa Cruz, CA, USA), goat Sox-2 (sc-17320; 1:500; polyclonal; Santa Cruz Biotechnology), mouse GFAP (G3893; 1:10000; monoclonal; Sigma–Aldrich, Saint Louis, MO, USA), rabbit S100β (721; 1:2000; polyclonal; Swant), rabbit NMDAR1 (for GluN1; AB9864R; 1:250, monoclonal; Millipore, Billerica, MA, USA), rabbit Cre [1:2000; polyclonal; a generous gift from Dr. Günther Schütz (Kellendonk et al., 1999)], mouse Mineralocorticoid receptor [6G1; 1:100; monoclonal; a generous gift from Dr. Elise P. Gomez-Sanchez (Gomez-Sanchez et al., 2006)].

#### Nissl Staining

Sections were mounted on microscope slides, air dried, immersed in 0.1% thionin, washed with water, dehydrated in 70, 95, and 100% ethanol for several minutes, cleared in xylene for 5 min twice. The slides were subsequently covered with Eukitt mounting medium, and mounted with coverslips.

#### Image Processing and Quantification

Fluorescent images were acquired using a LSM 700 confocal microscope (Zeiss, Oberkochen, Germany). A single focal plane picture of the whole DG area from one hemisphere was taken from each brain section using a tile scan function of the confocal microscope. Bright field images were acquired using a BX51W microscope (Olympus, Tokyo, Japan). All quantification analyses were done using ImageJ software. The DG thickness was measured from both dorsal/suprapyramidal blade and ventral/infrapyramidal blade. Care was taken to minimize variations in choosing sections in the rostrocaudal axis. Several, typically three, DG images from one animal were used for the analysis whenever possible. Measured values were averaged to create one representative value for each animal. To calculate DG thickness, a part of the DG with uniform thickness was selected as rectangular area of interest, and the area was then divided by the base length of the rectangular area. Dorsal DG thickness was calculated from the proximal areas to the crest where the two blades connect. Ventral DG thickness was calculated from the most ventral area close to the end of the blade. Degenerating cells were counted using the green channel after staining with antibody against active caspase-3. A cell was considered DCXpositive when the signal outlined the shape of a cell body with processes. To confirm that active capsase-3 (stained in green fluorescence) is expressed in CB-positive (stained in red fluorescence) cells, images of active caspase-3 positive cells were taken when there was no auto-fluorescence signal from degenerating cells in an unstained (far-red fluorescence) channel. Degenerating cells were counted as follows. The DG area to be examined was selected with selection tools. Putative degenerating cells were selected by MaxEntropy method of the autothreshold function, and signals bigger than 3 μm<sup>2</sup> were counted using "Analyze particles" function. We confirmed that this counting method matches manual counting. This method enabled us to quantify degenerating cells more objectively than manual

counting. Except for degenerating cells, all other fluorescently labeled cells were counted manually using cell counter function of ImageJ. Three to six images of the DG were used for counting fluorescent positive cells, and the obtained values were pooled to obtain one value for each mouse.

### Statistical Analysis

Quantifications were carried out without referring to the genotype information. Differences or correlations between groups were examined using the statistical tests indicated in the figure legends. Statistical analyses were performed using R 3.1.3. Values were expressed as mean ± standard deviation unless otherwise mentioned in the figure legends. Values of *p* < 0.05 were considered statically significant (∗*p* < 0.05, ∗∗*p* < 0.005, ∗∗∗*p* < 0.0005).

## RESULTS

### GluN1 Ablation Causes Progressive Degeneration of Granule Cells in the Dorsal DG

In 6 months old *Grin1*-*DGCA*<sup>1</sup> mice, in which *Grin1* is deleted in both the DG and CA1, we found a reduced thickness of the DG granule cell layer (**Figures 1A,B,H**). This phenotype could be readily detected by eye after DAPI staining, and was more pronounced in 18 months old mice (**Figures 1C,D,H**). In contrast, there was no apparent abnormality in the CA1 region in 18 months old *Grin1*-*DGCA*<sup>1</sup> mice (**Figures 1E,F,I**). Expression of active caspase-3 in CB-positive granule cells was indicative of mature granule cell demise (**Figure 1G**). The alteration in the granule cell layer was restricted to the dorsal DG. In 18 months old mice, the ventral DG in *Grin1*-*DGCA*1 mice was comparable to that of controls (Supplementary Figures S1A–C). We confirmed GluN1 ablation in the CA1 region and in both the dorsal and ventral DG of 6 months old *Grin1*-*DGCA*<sup>1</sup> mice by immunohistochemistry (Supplementary Figures S1D–G). Neuronal degeneration did not result from Cre expression itself, but was a consequence of *Grin1* deletion via Cremediated recombination, as the DG was normal in Cre expressing mice with wild-type *Grin1* alleles (Supplementary Figure S2). Thus, ablation of NMDARs in CA1 and DG causes selective degeneration of granule cells in the dorsal DG.

#### Neuronal Degeneration in *Grin1-DGCA*1 Mice Affects Mature DG Granule Cells

To corroborate that NMDAR deletion causes cell death of mature DG granule cells, we compared the distribution of immature doublecortin (DCX) expressing cells (**Figure 2A**) and of degenerating cells visualized by active caspase-3 immunostaining (**Figure 2B**) within the DG granule cell layer (**Figure 2C**). If degeneration affected primarily immature DG granule cells, degenerating cells should be confined to the innermost rim of the DG granule cell layer that harbors DCX expressing neurons. This, however, was not the case. Degenerating cells were found throughout the entire DG granule cell layer (**Figure 2D**). We next

FIGURE 1 | GluN1 deletion causes age-dependent neurodegeneration in the dorsal dentate gyrus (DG). (A) DAPI staining illustrating a section of a dorsal DG from a 6 months old control (Ctrl) mouse. (B) DAPI staining of a dorsal DG from a 6 months old *Grin1*-*DGCA*<sup>1</sup> mouse. Note the thinning of the dentate granule cell layer. (C) DAPI staining illustrating a section of a dorsal DG from an 18 months old control mouse. (D) DAPI staining of a dorsal DG from an 18 months old *Grin1*-*DGCA*<sup>1</sup> mouse. Note the thinning of the dentate granule cell layer following profound neurodegeneration. (E) DAPI staining illustrating a section of a dorsal CA1 from an 18 months old control mouse. (F) DAPI staining of a dorsal DG from an 18 months old *Grin1*-*DGCA*<sup>1</sup> mouse. There is no difference between the two genotypes. (G) Representative image from the DG showing an apoptotic, i.e., caspase-3-positive, CB-positive neuron in a 6 months old *Grin1*-*DGCA*<sup>1</sup> mouse. The section is counterstained with DAPI. (H) Quantitative evaluation of dorsal DG thickness in 6 and 18 months old control and *Grin1*-*DGCA*<sup>1</sup> mice. The thickness of the DG was analyzed by two-factor ANOVA, followed by pairwise comparison test with *p*-value modification by Holm's method, ∗∗∗*p* < 0.0005. The DG is thinner in *Grin1*-*DGCA*<sup>1</sup> mice [*F*(1,16) <sup>=</sup> 662.39, *<sup>p</sup>* <sup>&</sup>lt; 0.0005], and the difference increases with aging [*F*(1,16) <sup>=</sup> 59.92, *<sup>p</sup>* <sup>&</sup>lt; 0.0005]. There is a significant interaction between genotype and age [*F*(1,16) = 53.99, *p* < 0.005]. (I) Quantitative evaluation of CA1 thickness in the dorsal hippocampus in 6 and 18 months old control and *Grin1*-*DGCA*<sup>1</sup> mice. There is no difference in the thickness of the CA1 layer (two-factor ANOVA, no significant effect of genotype [*F*(1,16) <sup>=</sup> 0.594], no significant effect of age [*F*(1,16) = 2.793], and no significant interaction between age and genotype [*F*(1,16) = 0.168]. Scale bars, in (F), which applies to (A–E), and (G), 200 and 10 μm, respectively. Five and four mice for 6 months old control and *Grin1*-*DGCA*1, and six and five mice for 18 months old control and *Grin1*-*DGCA*<sup>1</sup> were used, respectively.

examined whether there was a correlation between the number of immature and degenerating DG granule cells. Since the number of immature DG granule cells decreases with age, both populations would have to decrease in parallel, if degenerating cells were primarily immature DG granule cells. However, while the number of immature DG granule cells decreased with age

indicate the maximum and minimum value within the 1.5-fold of the interquartile range from the higher (75th percentile) and the lower quartile (25th percentile). Dots in the boxplot are outliers. This analysis comprised 405 DCX-positive cells and 173 degenerating cells. (E) The number of DCX-positive cells in the DG decreases with age both in control and *Grin1*-*DGCA*<sup>1</sup> mice [two-factor ANOVA, *<sup>F</sup>*(1,16) <sup>=</sup> 30.090, *<sup>p</sup>* <sup>&</sup>lt; 0.00005]. There is neither a significant difference between genotypes [*F*(1,16) = 0.653], nor a significant interaction between age and genotype [*F*(1,16) = 0.423]. (F) The number of degenerating neurons in the DG is higher in *Grin1*-*DGCA*<sup>1</sup> mice [two-factor ANOVA, *<sup>F</sup>*(1,16) <sup>=</sup> 85.593, *<sup>p</sup>* <sup>&</sup>lt; 0.00005, followed by *post hoc* analysis with Holm's *<sup>p</sup>*-value modification, ∗∗∗*<sup>p</sup>* <sup>&</sup>lt; 0.0005]. There is neither an effect of age [*F*(1,16) = 0.247, ns], nor a significant interaction between age and genotype [*F*(1,16) = 0.888]. (G) There is no correlation between the number of degenerating and newborn neurons in the DG of *Grin1*-*DGCA*<sup>1</sup> mice (Spearman's rank correlation coefficient test). The same animals were used as in Figure 1. Scale bar in (B), 25 μm.

both in control and *Grin1*-*DGCA*<sup>1</sup> mice (**Figure 2E**), the number of degenerating cells was higher in *Grin1*-*DGCA*<sup>1</sup> mice than in controls both at 6 and 18 months of age (**Figure 2F**). Furthermore, in *Grin1*-*DGCA*<sup>1</sup> mice, there was no correlation between the number of immature DG granule cells and the number of degenerating cells, suggesting that degenerating cells were not immature cells (**Figure 2G**). Together these results indicate that degenerating DG granule cells are primarily mature DG granule cells.

#### NMDAR Dependent Survival is a Cell Type-Specific Property of Mature DG Granule Cells

In *Grin1*-*DGCA*<sup>1</sup> mice, DG granule cell degeneration might be a complex phenotype that results from changed interactions between CA1 and DG network activities, and might reflect, at least in part, altered output activity from CA1 neurons, in which NMDARs were also deleted. To directly test whether the observed degeneration of mature DG granule cells following GluN1 depletion is indicative of a specific DG granule cell vulnerability, we investigated the effect of NMDAR depletion in genetically modified mice (*Grin1*-*DG*), in which the manipulation was restricted to the DG. Indeed, a reduction in the thickness of the dentate granule cell layer with degenerating cells was seen also in 6 months old *Grin1*-*DG* mice (**Figures 3A–D,G,H**), indicating that adult granule cells require NMDAR-mediated activity for their survival.

Similarly to *Grin1*-*DGCA*<sup>1</sup> mice, also in *Grin1*-*DG* mice the number of immature DG granule cells was not reduced (**Figures 3E,F,I**), suggesting that GluN1 depletion was restricted to mature DG granule cells. Hence we determined Cre expression in mature and immature DG granule cells, and found that indeed, Cre expression was detected in CB-positive, but not DCX-positive neurons both in *Grin1*-*DG* (**Figures 3J–O**) and in *Grin1*-*DGCA*<sup>1</sup> mice (data not shown). Accordingly, GluN1 expression was present in immature DG granule cells (Supplementary Figures S3A–D). Together, these data provide evidence that GluN1 ablation was restricted to mature DG granule cells.

We wondered whether DG granule cell degeneration following NMDAR ablation was noticeable already at earlier developmental stages, and thus investigated 2 months old *Grin1*-*DG* mice. Despite of the GluN1 ablation, the number of immature and degenerating DG granule cells, and the overall thickness of the granule cell layer were comparable between control and mutant mice (Supplementary Figures S3E–I). It thus appears that the phenotype of DG granule cell degeneration following NMDAR ablation worsens in an age-dependent fashion.

Next we examined whether the genetic manipulation altered the number of stem cells in the DG of 6 months old *Grin1*-*DG* mice. The number of stem cells, i.e., GFAP/Sox2 positive cells, was not different between the two genotypes. The increased number of GFAP/Sox2/S100β positive cells in *Grin1*-*DG* mice (Supplementary Figure S4), reflects enhanced proliferation of astrocytes as often reported in brain areas with neurodegeneration. Hence NMDAR depletion in mature DG granule cells does not affect stem cell proliferation.

### Survival of Mature DG Granule Cells Requires GluN2B Expression

Finally, to examine the composition of the NMDARs involved in the degeneration of DG granule cells, we investigated the survival of mature DG granule cells in mice with either GluN2A or GluN2B receptor ablation. To this end we took recourse to 18 months old *Grin2a*−/<sup>−</sup> and *Grin2b*-*DGCA*<sup>1</sup> mice. In *Grin2a*−/<sup>−</sup> mice, the overall appearance and thickness of the DG granule cell layer was not different from that in control mice (**Figures 4A,B,E**). In contrast, the DG of *Grin2b*-*DGCA*<sup>1</sup> mice was reminiscent of the above-described phenotype in *Grin1*-*DGCA*1 and *Grin1*-*DG* mice (**Figures 4C,D,F**). These data demonstrate that GluN2B-containing NMDARs support the survival of aged dorsal DG granule cells.

### DISCUSSION

Here we demonstrate that NMDAR ablation causes neurodegeneration of the DG. This phenotype derives from the demise of mature DG granule cells and not from altered neurogenesis for the following reasons: First, enhanced caspase-3 expression and degenerating neurons were detected in the outer DG granule cell layer in the mutant mice. Second, the number of stem cells in the DG did not differ between control and mutant mice. Third, neurogenesis in the DG is known to decrease with age. Thus, if NMDAR ablation affected newborn cells, one would expect a progressive age-dependent decline of degenerating neurons. This, however, was not the case. Fourth, GluN1 was depleted primarily in mature DG granule cells as evidenced both by Cre expression and GluN1 immunohistochemistry in *Grin1*-*DG* mice. In addition, previous studies indicated that in several mouse mutants with reduced adult or absent adult neurogenesis the size or thickness of the DG granule cell layer was not visibly altered (Ansorg et al., 2012; Ouchi et al., 2013).

As evidenced in *Grin1*-*DGCA*<sup>1</sup> mice, neurodegeneration following NMDAR ablation was region- and cell type-specific. Thus, dorsal but not ventral DG granule cell or CA1 pyramidal neurons were affected. A similar phenotype as the one detected in *Grin1*-*DGCA*<sup>1</sup> and *Grin1*-*DG* mice was also reported in two other animal models of neurodegeneration. Thus, in rats, adrenalectomy causes degeneration of granule cells in the dorsal DG. The ventral DG is much less affected, and CA1 neurons or other hippocampal regions not at all (e.g., Sloviter et al., 1993). The pro-survival effect of corticosterone appears to be mediated by mineralocorticoid receptors (Gass et al., 2000). We do not think, however, that altered downstream signaling in mice with NMDAR ablation in DG granule cell involves signaling via mineralocorticoid receptors as their expression was not affected in our mutants (data not shown).

Notably, transgenic mice overexpressing Gsk3β also exhibit degeneration of granule cells preferentially in the dorsal DG

DCX-positive immature cells is comparable between genotypes (Welch's *t*-test). (J–O) In the DG of *Grin1*-*DG* mice, Cre (red) is expressed in CB-positive [green in (J–L)] mature neurons, but not in DCX-positive [green in (M–O)] immature neurons. Right panels (K,L,N,O) show a magnified view of the indicated area [white boxes in (J,M)]. Scale bars in (B,F,M,O), 100, 20, 100, and 20 μm, respectively. Five and seven mice were used for control and *Grin1*-*DG*, respectively.

(Fuster-Matanzo et al., 2011), and the alteration is accompanied by activated astrocytes (e.g., Fuster-Matanzo et al., 2011). Interestingly, Gsk3β is more active in the dorsal, whilst Akt, an inhibitor of Gsk3β, is more active in the ventral hippocampus (Fuster-Matanzo et al., 2011). Conversely, chronic application of lithium, an inhibitor of Gsk3β, ameliorates DG neurodegeneration induced by injection of amyloid-β fibrils (De Ferrari et al., 2003). Notably, there is evidence that β-catenin, a substrate of Gsk3β, directly interacts with the NMDAR (Al-Hallaq et al., 2007). Accordingly, GluN2Bcontaining NMDARs suppress apoptosis induced by Gsk3β overexpression in cultured neurons (Habas et al., 2006).

It was surprising that GluN2B, but not GluN2A, was required for the survival of DG granule cells, as several studies provided ample evidence that GluN2B-containing receptors mediate excitotoxic neuronal cell death (reviewed in Hardingham and Bading, 2010). There are nevertheless reports proposing that under certain conditions GluN2B

activation provides also pro-survival signals (Habas et al., 2006).

Our findings have implications for therapeutic approaches in which chronic NMDAR antagonist treatment is envisaged. Thus, NMDAR antagonists have been considered for treatment of many neurological diseases, including cerebral ischemia, traumatic brain injury, pain, Alzheimer's disease, Huntington's disease, Parkinson's disease, autism spectrum disorders, and depression (reviewed in Paoletti et al., 2013). In some cases, treatment appeared promising, at least in animal models (Paoletti et al., 2013; Patrizi et al., 2015). In humans, the NMDAR antagonist ketamine shows promising effects in patients with neuropathic pain and depression (reviewed in Niesters et al., 2014; Iadarola et al., 2015). Surveillance of the literature, however, indicates that in most studies the treatment was short, ranging from 1 day to few weeks. Importantly, ketamine, is also used for recreational purposes, and the longest period of drug intake exceeds 10 years (Liao et al., 2010). Chronic use of ketamine has long-lasting devastating effects on brain function, including working memory and episodic memory deficits (reviewed in Morgan and Curran, 2012). Based on our results, we propose that NMDAR antagonists, whilst potentially beneficial for acute treatment, bear the risk of causing DG degeneration following chronic administration, which would contribute to the worsening of cognitive functions, considering the crucial role of the DG in memory processes. It is not clear, however, to which extent the results that we obtained in mice following genetic receptor ablation can be transferred to humans, and in particular whether chronic NMDAR antagonist administration mimics the results reported here. Finally, the results warrant further investigations as to the downstream cellular signal cascade that eventually leads to the region-specific cell death. It is tempting to propose a

putative involvement of Gsk3β, but so far this hypothesis remains speculative.

In summary, mature granule cells in the dorsal DG undergo neurodegeneration following NMDAR ablation in aged mice. The study provides evidence that caution must be exerted, especially in aged patients, when long-term administration of NMDAR antagonists is considered for therapeutic purposes. Identification of downstream pathways leading to neurodegeneration following NMDAR ablation is warranted to eventually establish safe clinical administration of NMDAR antagonists.

#### AUTHOR CONTRIBUTIONS

YW and HM designed the experiments and wrote the manuscript. YW performed the experiments. YW and MM evaluated the data and performed the statistical analysis. All authors provided critical input for the generation of the final version of the manuscript. RS and PS provided the *Grin1*-*DGCA*1,

#### REFERENCES


*Grin1*-*DG*, and *Grin2a*−/<sup>−</sup> mice. All authors read and approved the final manuscript.

#### ACKNOWLEDGMENTS

We thank R. Hinz-Herkommer, I. Preugschat-Gumprecht, and U. Amtmann for technical assistance, M. Higuchi and A. Herb for animal management. This study was supported in part by a Postdoctoral Fellowship from the Uehara Memorial Foundation to YW.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2015.00087


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

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

# Molecular Mechanisms Regulating LPS-Induced Inflammation in the Brain

Olena Lykhmus <sup>1</sup> , Nibha Mishra<sup>2</sup> , Lyudmyla Koval <sup>1</sup> , Olena Kalashnyk <sup>1</sup> , Galyna Gergalova<sup>1</sup> , Kateryna Uspenska<sup>1</sup> , Serghiy Komisarenko<sup>1</sup> , Hermona Soreq<sup>2</sup> and Maryna Skok <sup>1</sup> \*

<sup>1</sup> Laboratory of Cell Receptors Immunology, O. V. Palladin Institute of Biochemistry, Kyiv, Ukraine, <sup>2</sup> The Edmond and Lily Safra Center of Brain Science and The Alexander Silberman Institute of Life Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel

Neuro-inflammation, one of the pathogenic causes of neurodegenerative diseases, is regulated through the cholinergic anti-inflammatory pathway via the α7 nicotinic acetylcholine receptor (α7 nAChR). We previously showed that either bacterial lipopolysaccharide (LPS) or immunization with the α7(1–208) nAChR fragment decrease α7 nAChRs density in the mouse brain, exacerbating chronic inflammation, beta-amyloid accumulation and episodic memory decline, which mimic the early stages of Alzheimer's disease (AD). To study the molecular mechanisms underlying the LPS and antibody effects in the brain, we employed an in vivo model of acute LPS-induced inflammation and an in vitro model of cultured glioblastoma U373 cells. Here, we report that LPS challenge decreased the levels of α7 nAChR RNA and protein and of acetylcholinesterase (AChE) RNA and activity in distinct mouse brain regions, sensitized brain mitochondria to the apoptogenic effect of Ca<sup>2</sup><sup>+</sup> and modified brain microRNA profiles, including the cholinergic-regulatory CholinomiRs-132/212, in favor of anti-inflammatory and pro-apoptotic ones. Adding α7(1–208)-specific antibodies to the LPS challenge prevented elevation of both the anti-inflammatory and pro-apoptotic miRNAs while supporting the resistance of brain mitochondria to Ca<sup>2</sup><sup>+</sup> and maintaining α7 nAChR/AChE decreases. In U373 cells, α7-specific antibodies and LPS both stimulated interleukin-6 production through the p38/Src-dependent pathway. Our findings demonstrate that acute LPS-induced inflammation induces the cholinergic anti-inflammatory pathway in the brain, that α7 nAChR down-regulation limits this pathway, and that α7-specific antibodies aggravate neuroinflammation by inducing the pro-inflammatory interleukin-6 and dampening anti-inflammatory miRNAs; however, these antibodies may protect brain mitochondria and decrease the levels of pro-apoptotic miRNAs, preventing LPS-induced neurodegeneration.

#### Edited by:

Kirsten Harvey, University College London, UK

#### Reviewed by:

Changiz Geula, Northwestern University, USA Julie A. Saugstad, Oregon Health & Science University, USA

> \*Correspondence: Maryna Skok skok@biochem.kiev.ua

Received: 11 December 2015 Accepted: 23 February 2016 Published: 08 March 2016

#### Citation:

Lykhmus O, Mishra N, Koval L, Kalashnyk O, Gergalova G, Uspenska K, Komisarenko S, Soreq H and Skok M (2016) Molecular Mechanisms Regulating LPS-Induced Inflammation in the Brain. Front. Mol. Neurosci. 9:19. doi: 10.3389/fnmol.2016.00019

Keywords: inflammation, brain, α7 nicotinic acetylcholine receptor, acetylcholine esterase, microRNA, antibody

## INTRODUCTION

Neuro-inflammation accompanies and often precedes the development of neurodegenerative pathologies such as Parkinson's and Alzheimer's diseases (AD; Wee Yong, 2010) and might be one of the pathogenic factors for neurodegeneration (Chung et al., 2010; Heppner et al., 2015). However, the molecular mechanisms linking inflammatory reaction to degeneration in the brain are poorly understood. One of the cross-points may include nicotinic acetylcholine receptors of the α7 subtype (α7 nAChRs), which are involved in both cholinergic anti-inflammatory and pro-survival intracellular pathways (Ji et al., 2014; Báez-Pagán et al., 2015; Terrando et al., 2015; Truong et al., 2015).

A number of studies point out the involvement of α7 nAChRs in pro-survival cell signaling, engaging the PI3K/Akt signaling pathway (Parada et al., 2010; Yu et al., 2011; Cucina et al., 2012; Huang et al., 2012; Cui et al., 2013). Such signaling was shown to protect cultured brain cells from apoptosis induced by various apoptogenic agents (Parada et al., 2010). Moreover, α7 nAChRs were found to regulate mitochondrial permeability transition pore formation and release of apoptogenic factors like cytochrome (cyt c) and therefore, to control the mitochondrialmediated pathway of apoptosis (Gergalova et al., 2012, 2014).

In addition, α7 nAChR directly interacts with the amyloidbeta (Aβ) precursor protein (Ikonomovic et al., 2009), which favors its normal processing by α-secretase (Kim et al., 1997; Qi et al., 2007) and promotes internalization of processed Aβ forms (Wang et al., 2000; Parri and Dineley, 2010). Consequently, the decrease of α7 nAChR density on the plasma membrane impairs Aβ metabolism and favors accumulation of extracellular Aβ (1–42) (Gouras et al., 2005; Dziewczapolski et al., 2009).

Previously, we found that regular injections of bacterial lipopolysaccharide (LPS) decreased the density of α7 nAChRs in the brain and brain mitochondria of mice and reduced nucleated cell numbers in the hippocampus and striatum, while stimulating astrocytosis, accumulation of Aβ (1–42) peptides and episodic memory decline—symptoms characteristic of the early stages of AD (Lykhmus et al., 2015). Antibodies, generated in vivo by immunization of mice with recombinant extracellular domain of α7 nAChR subunit, α7(1–208), facilitated symptoms similar to those induced by LPS but did not cause degeneration in the brain of mice (Lykhmus et al., 2015), indicating the involvement of specific regulatory processes.

Another important regulator of cholinergic signaling is acetylcholinesterase (AChE), the levels of which decrease during inflammation, increasing acetylcholine levels and stimulating the anti-inflammatory pathway (Soreq, 2015). Acetylcholine was shown to attenuate the release of pro-inflammatory cytokines, like IL-1 or TNFα, by peritoneal monocytes and macrophages in response to bacterial endotoxin—LPS through α7 nAChRs (Borovikova et al., 2000). This phenomenon, first described in 2000 and called ''Cholinergic Anti-Inflammatory Pathway'', was further observed in many organs and tissues including the brain (de Jonge and Ulloa, 2007; Tyagi et al., 2010; Thomsen and Mikkelsen, 2012; Ji et al., 2014; Báez-Pagán et al., 2015; Egea et al., 2015; Truong et al., 2015). AChE expression has been shown to be regulated by microRNAs (miRNAs), small non coding RNAs suppressors of entire pathways of gene expression (Chen et al., 2004; Soreq and Wolf, 2011). MiRNA-132 is reported to increase during inflammation in many tissues (Maharshak et al., 2013; Shaltiel et al., 2013; Nadorp and Soreq, 2015) and is validated to target AChE further to potentiate cholinergic antiinflammatory pathway (Shaked et al., 2009; Soreq and Wolf, 2011).

The present study was aimed to reveal the molecular mechanisms underlying the LPS and antibody effects in the brain, using a model of acute LPS-induced inflammation with or without α7-specific antibody injections. Specifically, we studied the involvement of α7 nAChRs in brain inflammation and mitochondrial apoptosis, measured changes in AChE levels with inflammation and profiled brain miRNAs under exposure to LPS, LPS and α7-specific antibody (Ab α7) or nicotine. Our findings indicate that LPS down-regulates α7 nAChR and AChE in the brain; exacerbates the mitochondrial pathway of apoptosis and changes brain miRNAs in favor of pro-apoptotic and anti-inflammatory ones. Inversely, the antibody supports the integrity of brain mitochondria and attenuates the LPSinduced pro-apoptotic miRNAs up-regulation while stimulating pro-inflammatory signaling and preventing the LPS-induced elevation of the anti-inflammatory miRNA-132/212 (Shaked et al., 2009; Shaltiel et al., 2013; Soreq, 2015).

#### MATERIALS AND METHODS

#### Animals and Reagents

Female 3 months old C57BL/6J mice were housed in a quiet, temperature-controlled room (22–23◦C) in the animal facility of the O.V. Palladin Institute of Biochemistry and were provided with water and dry food pellets ad libitum. Mice were sacrificed by cervical dislocation to remove the brain. All procedures of this study conformed to the guidelines of the Animal Care and Use Committee of Palladin Institute and were approved by the IACUC Protocol 1/7–421.

All reagents were of chemical grade and were purchased from Sigma-Aldrich unless specially indicated. Antibodies against α7(1–208), α7(179–190), α3(181–192), α4(181–192), β2(190–200) and β4(190–200) nAChR fragments were obtained and characterized by us previously (Skok et al., 1999; Koval et al., 2004; Lykhmus et al., 2011).

Glioblastoma U373 cells (ATCC HTB17–1055) were a kind gift of Prof. A.V.Ryndich from the Institute of Molecular Biology and Genetics in Kyiv.

### Animal Treatment and Samples Preparation

Mice were intra-peritoneally injected with LPS (30 µg/mouse in PBS) on days 0 and 2 (groups 1 and 2). Group 2 mice were additionally intravenously injected with α7(1–208)-specific antibodies (200 µg/per mouse in saline) on days 0, 1 and 2. Group 3 mice received nicotine in the drinking water (200 µl/l) for either 3 days or 1 month and the 4th group was intact (Ctrl).

On the 3rd day of each treatment, mice were sacrificed; their brains were removed, homogenized and fractionated into primary pellets (nuclei, cell debris) and mitochondria (additionally pelleted from the supernatant) by a standard procedure of differential centrifugation (Gergalova et al., 2012). To quantify nAChR subunits, both mitochondria and primary pellets were lysed in detergent-containing buffer. Live mitochondria were further examined for functional activities.

In another set of experiments, the brains of similarly treated mice were divided into two halves (hemispheres) and each half was dissected into Hippocampus, Striatum/Thalamus, Cerebellum and Frontal cortex. The one half sections were homogenized under liquid nitrogen and used for RNA extraction. Sections of the 2nd half were homogenized; lysed in detergent-containing buffer (0.01 M Tris-HCl, pH 7.4, 1 M NaCl, 1 mM EGTA, 1% Triton X-100) for 45 min on ice and centrifuged at 13,000 rpm. The resulting supernatant was used for quantifying AChE activity and α7 nAChR protein. Protein content was measured with the BCA kit (Thermo Scientific, France).

### Cytochrome c (cyt c) Release Assay with Live Mitochondria

Cyt c release from isolated mitochondria was measured as described previously (Gergalova et al., 2012). Briefly, purified mitochondria (120 µg of protein per ml) were incubated with different doses of CaCl<sup>2</sup> with or without the α7 nAChR agonist PNU282987 (30 nM) for 2 min at room temperature and were immediately pelleted by centrifugation. The supernatants were tested for the presence of cyt c by a sandwich ELISA assay.

#### Quantifying nAChR Subunits in the Brain or Mitochondria Preparations

The assay was performed as described Lykhmus et al. (2015). Briefly, immunoplates (NUNC, MaxiSorp) were coated with rabbit α7(1–208)-specific antibody (20 µg/ml), blocked with 1% BSA, and the tested preparations were applied into the wells (1 µg of protein per 0.05 ml per well) for 2 h at 37◦C. The plates were washed with water and the second biotinylated α3(181–192), α4(181–192), α7(179–190), β2(190–200) or β4(190–200)-specific antibody was applied for additional 2 h being revealed with Streptavidin-peroxidase conjugate and o-phenylendiaminecontaining substrate solution.

### RT-PCR for α7, AChE and microRNA Transcripts

RNA extraction from tissue samples was carried out using Trizol (Sigma, NY, USA) as described Shaked et al. (2009). RNA concentration and integrity were determined spectrophotometrically and by electrophoresis, respectively. RNA samples (500 ng) were reverse transcribed using the Quanta cDNA synthesis kit for mRNA and miRNA as per the manufacturer's (Quanta Biosciences) protocol. Real-time RT-PCR was performed using the ABI prism 7900 HT and SYBR green master mix (Quanta Biosciences). For miRNA transcripts, PerfeCTa microRNA assay primers (Quanta Biosciences) were used. Results were normalized to the expression of snoRD47 and actin for miRNA and mRNA respectively. Further all the results were normalized to respective regional control. The following primers were used for: AChE-S Forward (F): CTGAACCTGAAGCCCTTAGAG, Reverse (R): CCGCCTCG TCCAGAGTAT; nAChR7, F: CACATTCCACACAACGTCTT, R: AAAAGGGAACCAGCGTACATC; actin F: CCACACCCG CCACCAGTT, R: TACAGCCCGGGGAGCAT. Fold change values for both miRNAs and mRNAs were calculated using the ∆∆Ct method.

### RNA-seq Library Preparation and Sequencing

Libraries for next generation sequencing (NGS) were prepared from whole brain RNA using TruSeq Small RNA Library Prep Kit as per the manufacturer's protocol. A total of four libraries (pooled from four animals of each group) were prepared from RNA of four groups. Briefly, the total 600 ng of RNA samples were hybridized with Trueseq Adaptor Mix which is a set of oligonucleotides with a single-stranded degenerate sequence at one end and a defined sequence required for Miseq sequencing at the other end. The Adaptor Mix constrains the orientation of the RNA in the ligation reaction such that hybridization with it yields template for sequencing from the 5 0 end of the sense strand. After hybridization, the adaptors are ligated to the small RNA molecules using the Ligation Enzyme Mix, which is a mixture of an RNA Ligase and other components. Ligation requires an RNA molecule with a 5<sup>0</sup> -monophosphate and a 3-hydroxyl end; therefore, most small RNAs can participate in this reaction, and intact mRNA molecules with a 5<sup>0</sup> -cap structure are excluded. Next, the small RNA population with ligated adaptors of each sample was reverse transcribed, to generate cDNA libraries. Treatment with RNaseH followed, to digest the RNA from RNA/cDNA duplexes and to reduce the concentration of un-ligated adaptors and adaptor by-products. The cDNA libraries were amplified using bar coded primer sets and 15–18 cycles of PCR. The amplified cDNA libraries were cleaned up and size selected from gel—PCR products of 105–150 bp were isolated, corresponding to inserts derived from the small RNA population. The amplified cDNA libraries generated were there after used for Miseq sequencing.

miRNA pathway analysis was performed using the microT-CDS tool available through Diana Tools <sup>1</sup> .

### Determination of AChE Activity

The acetylcholine hydrolyzing activity of AChE was measured by the Ellman's assay (Ellman et al., 1961) as described

<sup>1</sup>http://diana.imis.athena-innovation.gr/DianaTools/index.php?r=mirpath/ index

Arbel et al. (2014). Briefly, samples were diluted 1:5 with 0.2 M phosphate buffer pH 7.4. Ellman's reagent was added to each sample, the mixture was incubated at RT for 20 min under darkness, acetylthiocholine iodide was added and absorbance of the corresponding plate wells at 405 nm was monitored with Stat-Fax 2000 ELISA Reader (Advanced Technologies, IL, USA) at 15 time points with 2 min intervals. AChE activity was calculated based on the concentration of the resultant 5-thio-2-nitrobenzoate anion (ε<sup>405</sup> = 13.6 M−<sup>1</sup> cm−<sup>1</sup> ); taking into account the average OD increment per minute and protein concentration in the sample.

#### U373 Culturing, Staining and Imaging

Glioblastoma U373 cells were cultured in RPMI 1640 medium supplemented with 20 mM L-glutamine, 20 mM HEPES, penicillin-streptomycin mixture and 10% fetal calf serum. For microscopy studies, the cells were attached onto glass slides, 3 × 10<sup>4</sup> per slide, in complete medium for 3 h and were incubated with biotinylated α7(179–190)-specific antibody (0.06 mg/ml) and MitoTracker Green (Invitrogen, USA) for 100 min at 37◦C. Cells were fixed with 4% paraformaldehyde and treated with Streptavidin-Cychrome3 conjugate to visualize the biotinylated antibody, followed by washes with PBS and water, embedding in MOWIOL-DABCO and examination in a Zeiss LSM 510 Meta confocal laser scanning microscope.

#### Measuring IL-6 Produced by U373 Cells

U373 cells (3 × 10<sup>5</sup> per ml) seeded into the wells of 96-well plates were cultured with LPS (clone 055 B5; 1 µg/ml) or α7(179–190)-specific antibody (10 µg/ml) in the presence or absence of the following kinase inhibitors: SB202190 (10 µM, p38 inhibitor), KN62 (1 µM, CaKMII inhibitor), PP1 (10 µM, Src kinase inhibitor), bisindolylmaleimide (50 nM, protein kinase C inhibitor) and Wortmaninn (1 µM, phosphatidylinositol-3-kinase inhibitor) for 24 h. The IL-6 concentration in the supernatants was detected using Diaclone test system as per the manufacturer's instructions.

### Statistical Analysis

Each experiment has been performed in minimum 7 mice and ELISA assays for each mouse have been performed in triplicates. The mean values for individual mice were used for statistical analysis using Student's t-test. The data are presented as M ± SE; <sup>∗</sup>p < 0.05; ∗∗p < 0.005.

#### RESULTS

#### Both LPS and α7(1–208)-Specific Antibodies Modulate nAChR Composition in the Brain

Sandwich ELISA performed with whole brain preparations demonstrated that LPS treatment decreased the level of α7 nAChR subunits while increasing the α3 and β4 subunits. Injections of α7(1–208)-specific antibodies additionally decreased the α4 and β2 subunits (**Figure 1A**). Nicotine (3 days) up-regulated both α7 and α4β2 nAChRs and did not influence α3β4 ones. LPS and LPS plus α7(1–208) specific antibody, but not nicotine, decreased α7 and α4β2 nAChRs and non-significantly increased α3β4 ones in mitochondria preparations (**Figure 1B**). LPS treatment decreased both α7 RNA and protein in the frontal cortex, striatum, hippocampus and cerebellum. The α7(1–208)-specific

FIGURE 2 | Modified levels of α7 nAChR protein (A) or RNA (B) in various brain regions of mice treated with LPS, α7(1–208)-specific antibody + LPS (Abα7 + LPS) or nicotine (Nic) compared to non-treated mice (Control). FC—frontal cortex, Str—striatum, Hip—hippocampus, Cer—cerebellum <sup>∗</sup>p < 0.05; ∗∗p < 0.005 compared to Control (n = 8).

antibody did not modify the effect of LPS on α7 nAChR RNA or protein expression. Nicotine slightly increased the α7 protein in the striatum but did not affect its RNA level (**Figures 2A,B**).

### LPS and α7(1–208)-Specific Antibodies Modulate Brain AChE Expression and Activity

LPS exposure significantly decreased the levels of the ''synaptic'' AChE-S variant (Soreq and Seidman, 2001) in the frontal cortex and non-significantly in the striatum, hippocampus and cerebellum that was interpreted as a tendency to decrease. The antibody additionally decreased AChE-S RNA in the hippocampus. Nicotine (3 days) caused non-significant up-regulation of AChE-S in the striatum (**Figure 3A**). Enzyme activity measurements demonstrated decreased AChE activity in the frontal cortex and cerebellum and a tendency to decrease in the striatum and hippocampus under the effect of LPS. The antibody accentuated the LPS effect in the frontal cortex, while nicotine tended to increase AChE activity in striatum (**Figure 3B**). A correlation between AChE-S expression and activity was observed in the frontal cortex, cerebellum and striatum of individual mice (Pearson coefficient being from 0.80–0.99; data not shown).

### LPS and α7(1–208)-Specific Antibodies Modify the Reaction of Brain Mitochondria to Apoptotic Stimuli

Live mitochondria isolated from the brains of LPS-treated mice released more cyt c in response to 0.9 and 9.0 µM Ca2<sup>+</sup> and became less sensitive to the normalizing effect of the α7-specific agonist PNU282987 than mitochondria of non-treated animals (**Figure 4**). Moreover, mitochondria of LPS-treated mice released some cyt c without any Ca2+, reflecting their unstable (preapoptotic) state. The α7(1–208)-specific antibodies decreased the LPS-induced cyt c release from mitochondria at low Ca2<sup>+</sup> doses and facilitated the normalizing effect of PNU282987.

### The Effect of LPS and α7(1–208)-Specific Antibody on the microRNA Spectrum in the Brain

To delineate the differentially expressed miRs in brain during inflammation we performed whole brain miR-deep sequencing analysis. A total of 24, 249, 523 sequencing reads were obtained. All the reads (50 bp long) were subjected to trimming of the tag end terminal base pairs and P1 start adapter (Miseq miRNA reverse primer sequences) using CLC genomics workbench V7.0 (CLC Inc, Aarhus, Denmark). The remaining reads (1, 88, 986) obtained were aligned against the mouse miRNA genome (miRBase release 20) and Ensemble mouse database (GRCm38) for non coding RNA. Through the annotation and merge counts, only reads longer than 15 bases were analyzed. Match parameters included for mature length variants (IsomiRs)-additional two upstream and downstream bases, and two missing upstream and downstream bases, and maximum allowed mismatches two using standard specific alignment protocol. On average, 70% (range 76.6–50.6%) of the sequences of all annotated reads were mapped and 85% (range 90.4–76.7%) of miRBase genes were detected. Of the mapped reads, on average 11% (range 12.2–10.8%) had perfect match to the aligned genes, 54% (range 54.5–55.2%) one mismatch, 23% and (range 21.7–24.7%) two mismatches. Mapping to Ensemble mouse database (GRCm38) yielded mapping to 10.5% (10–11%) of all annotated database sequences on average across libraries. Preference was given to miRBase, so the database which was not mapped to miRBase is mapped to this. Of them, 55.6% (range 55.6–37.3%) exhibited perfect match to the reference sequences, 39.6% (range 39.6–51.5%) with one mismatch, and 8.3 (8.3–11.2%) two mismatches. These alignments yielded an average of 2.06% aligned reads of the total number of reads (with a minimum of 0.2% and a maximum of 3.16%). Annotated samples were grouped by both precursor and mature sequence identity. Overall, a top 44 mature miRs that exhibited count of at least 30 per million reads in at least one sample were analyzed for differential expression between the different experimental conditions. For fold change analysis the above identified miRs of different groups were normalized to control group and that were 1.5 up-regulated or 0.5 fold down-regulated were identified as uniquely regulated miRs.

The wide screening of the brain RNA for the expression of 44 miRs demonstrated that the cluster of the highest expression contained miRs-99a, let7g and 9, that the cluster of medium expression included miRs-26a and let-7f, and that other miRs were of quite low or very low expression. The miR-21 and miR-434 were observed twofold up-regulated and let-7a-1 0.5 fold down-regulated. As shown in **Figure 5A**, all types of treatment (LPS, LPS + α7(1–208)-specific antibody or nicotine) influenced the level of many of them. MiR-99a was obviously upregulated by both nicotine and LPS and the effect of LPS was withdrawn by the antibody. In contrast, miR-let7g was downregulated by LPS and, less, nicotine but the effect of LPS was again withdrawn by the antibody. MiR-9 was up-regulated by nicotine and much less affected by the LPS; however, again, the antibody effect was opposite to that of LPS. MiRs-26a and let-7f were down-regulated in all groups of treated mice; the antibody obviously aggravated the LPS effect for miR-26a. Among other miRs, of considerable interest are miR-21a and miR-434: both of them were up-regulated in all treated mice, mostly by LPS, and the antibody prevented this effect more or less efficiently. In whole, the antibody obviously attenuated the LPS effect in 16 miRs tested and aggravated its effect in six miRs. The effects of nicotine and LPS were of similar direction for six miRs (miR-26, let-7f, let-7c, miR-30, miR-21a and miR-434) and showed an opposite impact for other four miRs (miR-9, let-7j, miR-218 and miR-125).

After miRNA differential expression analysis, we performed KEGG pathways analysis of differentially regulated miRNA of each group exposed to either LPS or LPS and Ab α7 or nicotine (**Figure 5B**). After removing redundant terms, our findings pointed that all pathways involved in LPS exposure were included in LPS and Ab α7 exposure group. In addition, this group was also associated with Notch signaling, apoptosis, mRNA surveillance, bacterial invasion and calcium signaling pathways. Similarly, nicotine group included all pathways involved in LPS inflammation; in addition it was associated

with addiction pathways like Cocaine, Amphetamine, Retrograde endocannabinoid signaling, apoptosis and Notch signaling. Common KEGG pathways found in all treatment groups are either neuronal or inflammatory pathways which included axon guidance, MAPK signaling pathway, PI3K-Akt signaling pathway, T cell receptor signaling pathway, Neurotrophin

signaling pathway, mTOR signaling pathway, TGF-beta signaling pathway, Wnt signaling pathway, Fc gamma R-mediated phagocytosis, Adherence junction, Glutamatergic synapse, Longterm potentiation, Dopaminergic synapse, Amyotrophic lateral sclerosis (ALS), Fc epsilon RI signaling pathway, Dorsoventral axis formation, Cholinergic synapse, B cell receptor signaling pathway, Jak-STAT signaling pathway, Toll-like receptor signaling pathway, chemokine signaling pathway.

#### qRT PCR Analysis of miRNA-132/212 in Different Regions of the Brain

To further understand the down-regulation in expression of the cholinergic α7 nAChR (global) and AChE-S (brain region specific) genes, we performed qRT PCR analysis for quantifying the expression of AChE-S targeting miRs: miR-132 and its co-clustered miR-212 in the above mentioned four brain regions. We observed significant LPS-induced increases in the expression levels of miR-212 in all of the tested brain regions, whereas miR-132 showed region-specific (frontal cortex and cerebellum) increases in its expression, correlating to AChE-S expression (**Figures 3A,B** vs. **Figure 6**). The antibody tended to cause decreases in miR-132 expression that was up-regulated by LPS in the frontal cortex, hippocampus and striatum. In comparison, the antibody treatment clearly prevented LPS-induced miR-212 up-regulation in all brain regions except cerebellum (**Figure 6B**). Nicotine (3 days) failed to significantly affect either miR-132 or miR-212 expression.

#### LPS and α7(1–208)-Specific Antibodies Stimulate IL-6 Production by U373 Cells through a Similar Signaling Pathway

Previously we demonstrated that α7(1–208)-specific antibodies and even more, α7(179–190)-specific antibodies stimulated IL-6 production in U373 cells via a p38-dependent pathway (Kalashnyk et al., 2014). To test if similar or different mechanisms are involved in LPS- or antibody-stimulated IL-6 production, we tested the effect of various kinase inhibitors on these consequences. As shown in **Figure 7A**, the antibody stimulated much weaker IL-6 production compared to LPS in U373 cells; however, both LPS-stimulated and antibodystimulated IL-6 levels were significantly decreased in the presence of p38 and Src kinase inhibitors, suggesting the involvement of a similar Src/p38-dependent signaling pathway.

To study if internalized α7-specific antibodies can bind mitochondria, we allowed their internalization by U373 cells for 100 min, followed by MitoTracker Green staining. Confocal microscopy showed no overlapping staining for the antibody and MitoTracker (**Figure 7B**); therefore, the internalized antibody seemed not to bind mitochondria.

### DISCUSSION

The influence of α7 nAChR signaling on pro-inflammatory cytokines production is well documented (de Jonge and Ulloa, 2007; Kalashnyk et al., 2014). Our data support, deepen and extend these findings, showing that inflammation regulates α7 nAChR and AChE-S expression and changes miRNA profiles in the brain. Intraperitoneal LPS injections resulted in down-regulation of brain α7 nAChR and AChE-S RNAs and protein levels. AChE down-regulation was accompanied by upregulation of its targeting miRNA-132 (Shaked et al., 2009; Shaltiel et al., 2013) in the frontal cortex and cerebellum of LPS-treated mice. This region-specific inter-related regulation of miRNA-132/AChE is compatible with the involvement of the cluster harboring miRNA-132/212 in the resolution of inflammation (Nahid et al., 2011; Rao et al., 2015). The

general up-regulation of miRNA-212 in all studied brain areas and its significant inhibition by the α7-specific antibody in the frontal cortex, striatum and hippocampus suggests its specific involvement in inflammation-related mechanisms, different from those regulated by miRNA-132. In general, this data indicates that the LPS-induced inflammatory reaction also stimulated the anti-inflammatory cholinergic pathway by leading to consequent increases in ACh levels due to downregulated AChE expression and activity. However, inhibiting the expression of α7 nAChRs (which presumably mediate the anti-inflammatory effect of ACh) makes inefficient this LPSstimulated anti-inflammatory response.

Down-regulation of α7 nAChRs was accompanied by the increase of α3β4 nAChRs, similarly to what we observed in α7 <sup>−</sup>/<sup>−</sup> mice or in mice chronically treated with LPS (Lykhmus et al., 2011, 2015). This means that α7 to α3β4 nAChRs substitution is an established mechanism, which could possibly be due to the chromosomal arrangement of nAChR subunit genes; and that a rather short LPS influence (3 days) is sufficient to stimulate this gene expression exchange. The antibody additionally decreased α4 and β2 protein in the brain, possibly due to the cross-reactivity of α7(1–208)-specific antibodies with the homologous α4 subunit, resulting in α4β2 receptors internalization and degradation. Nicotine treatment did not cause significant changes in the nAChR or AChE expression but up-regulated α4, α7 and β2 proteins in the whole brain that is in accordance with the suggested chaperon-like activity of nicotine (Sallette et al., 2005). We conclude that LPS and α7(1–208) specific antibodies manipulate the molecular components of the brain's cholinergic anti-inflammatory pathway.

Similarly to the chronic LPS treatment (Lykhmus et al., 2015), short-term LPS exposure decreased the level of mitochondrial α7 nAChRs and made the brain mitochondria more sensitive to Ca2+. The α7-specific antibody prevented the additional cyt c release from mitochondria and, therefore, supported their resistance to apoptogenic influence. Since α7(1–208)-specific antibodies attenuated Ca2+-stimulated cyt c release from isolated mitochondria (Gergalova et al., 2014), we hypothesized that the antibody could penetrate the brain cells to directly affect mitochondria, as it was recently suggested for anti-mitochondrial antibodies in patients with Pemfigus vulgaris (Chernyavsky et al., 2015). However, in our experiments, no binding of internalized antibody with mitochondria was observed in U373 cells (**Figure 7B**) assuming the involvement of indirect molecular mechanism, possibly, including miRNAs.

Over a decade many miRNAs showed functional involvement in neuro-inflammatory mechanisms (Soreq and Wolf, 2011; Maharshak et al., 2013; Nadorp and Soreq, 2015). Analysis of the present data suggests the involvement of miRNAs in regulating the brain cell survival under the effect of LPS and α7-specific antibodies. The brain-abundant miRNA-9, which was down-regulated by LPS exposure, inhibits the expression of a proapoptotic Bcl-2L11 found in the outer mitochondrial membrane (Li et al., 2014); its potential pro-apoptotic effect was largely avoided by the antibody. Another brain-abundant miRNA-99 regulates pro-survival Akt/mTOR signaling (Jin et al., 2013); its up-regulation with LPS was expected to decrease the brain cell viability; whereas the α7(1–208)-specific antibody limited this effect. Likewise, let-7g suppresses the expression of the anti-apoptotic protein Bcl-xL (Wu et al., 2015); therefore, its down-regulation by LPS might be pro-apoptotic, whereas the antibody treatment inhibited this effect. In whole, LPS exerted anti-survival changes in multiple brain miRNAs, and the α7-specific antibody could antagonize part of them, providing a tentative explanation for the protective antibody effect in mitochondria found here. According to our previously published data, α7 nAChR is located in the outer mitochondria membrane and its activation stimulates intramitochondrial Pi3K/Akt signaling pathway (Gergalova et al., 2012, 2014), that is in good agreement with the present results. The pro-survival antibody effect also explains the data of Kamynina et al. (2013) and Lykhmus et al. (2015), where α7-specific antibodies exerted neuroprotection in certain models of AD.

In addition to their involvement in inflammation-related processes, both miRNA-132 and miRNA-212 protect neurons against H2O2-mediated cell death, and their loss causes neuronal apoptosis via elevated levels of the cell deathassociated proteins PTEN, FOXO3 and P300 that antagonize Akt pro-survival signaling (Wong et al., 2013). Knockdown of miRNA-132 in the hippocampus impairs memory acquisition (Wang et al., 2013), which can be regarded as a marker for cognitive impairment (Xie et al., 2015). The brains of Alzheimer patients demonstrated decreased miR-132/212 level already at stages Braak III and IV of the disease and in a manner related to Tau pathology (Lau et al., 2014). Regarding our data, it may be suggested that neuro-inflammation upregulates miRNA-132/212 as an anti-inflammatory reaction, which also protects the brain cells from excessive reactive oxygen species toxicity. The α7(1–208)-specific antibodies weaken this reaction but do not decrease miRNA-132/212 below the control level. Obviously, the final pro- or anti-survival antibody effect is an integral interplay of multiple miRNAs and their targets.

The KEGG pathways analysis allowed deeper understanding of miRNA regulating activity in the brain under the effects of LPS, LPS and antibody or nicotine. According to the probability of engagement, the predicted signaling pathways could be classified into three groups (**Figure 5A**): extremely probable (−log p > 20: Axon guidance, MAPK and PI3K/Akt pathways); highly probable (20 > −log p > 10: TcR, mTOR, Wnt, TGFβ and Neurotrophin pathways) and probable (10 > −log p > 2: all other predicted pathways).

The first group includes pathways found under many different receptors; therefore, it is impossible to identify any definite one. However, comparison of −log p values found for LPS and LPS + Ab treatments allows suggesting that the antibody contributes additional signaling through these pathways. This is in accord with our experimental data on the ability of α7-specific antibody to stimulate IL-6 production in U373 cells through p38-dependent pathway (**Figure 7A**). The involvement of both MAPK and PI3K/Akt pathways in α7 nAChR signaling is well documented (Parada et al., 2010; Yu et al., 2011; Cucina et al., 2012; Huang et al., 2012; Cui et al., 2013). Previously we reported that signaling of α7 nAChRs expressed in mitochondria can be triggered by α7-specific antibody (Gergalova et al., 2014), therefore, the antibody could engage PI3K/Akt pathway in the brain.

The second group contains signaling pathways of TcR, TGFβ, Wnt and Trk (neurotrophin) receptors, as well as mTOR pathway. Again, the antibody contributes to these pathways compared to LPS alone. The maximal increase of −log p was observed for TcR (22 vs. 15), mTOR (17 vs. 12) and neurotrophin (10.5 vs. 7.7) signaling pathways. Neurotrophins acting through Trk receptors activate PI3K/Akt and MAPK pathways, and the mTOR functions downstream of PI3K/Akt signaling pathway in response to cytokines and growth factors (LoPiccolo et al., 2008; Longo and Massa, 2013); therefore, their engagement by the antibody may be explained by the antibody effect found in group 1. Neurotrophin signaling is strongly stimulated by nicotine (14 (Nic) vs. 7.5 (LPS) vs. 10.5 (LPS + Ab)) that also assumes the involvement of nicotinic receptors and, possibly, their cross-talk with Trk receptors. T lymphocytes are normally not found in the brain parenchyma but can penetrate there under neuroinflammation when the blood-brain barrier is disrupted (Engelhardt, 2006). In addition, they could be found in the brain capillaries (since the brains had not been perfused before RNA extraction). The antibodies significantly contributed to this pathway; therefore, they could facilitate T lymphocyte migration to or penetration into the brain and subsequent activation resulting in TcR signaling pathway involvement. The Wnt and TGFβ signaling pathways are engaged by all three treatments; however, there is no significant difference between nicotine, LPS or LPS + Ab-treated mice.

Among other signaling pathways of interest are the Notch and apoptosis signaling, which are found in nicotine- and LPS + Ab- but not LPS-treated mice and, therefore, relate to nicotinic receptors. The Notch signaling pathway components expressed in the brain were shown to be involved in the pathogenesis of AD (Woo et al., 2009) and, therefore, this pathway may contribute to the memory impairment found in α7(1–208)-immunized mice (Lykhmus et al., 2015). The involvement of nAChRs in regulating cell survival and apoptosis has already been discussed. The data of KEGG analysis predict the miRNA-mediated regulation of apoptosis pathway by α7(1–208)-specific antibodies that is in accord with the analysis of molecular targets of miRs 9, 99 and let-7g described above.

LPS notably stimulates pro-inflammatory cytokines production via toll-like receptor type 4 (TLR-4)/CD14 receptor complex, resulting in MAP kinases activation and a nuclear localization of NF-κB (Kawai and Akira, 2010; Nadorp and Soreq, 2015) also engaging TLR-4 to Src activation (Liu et al., 2012). Here, we show that similar enzymes (Src kinases and p38) are involved in the signal transduction pathway from either LPS or α7-specific antibodies, suggesting that the α7 nAChR can directly influence TLR-4 signaling. Together with the parallel TLR9/ACh interaction (Nadorp and Soreq, 2015), these findings explain how ACh may attenuate the LPS-induced pro-inflammatory cytokine production through α7 nAChR. Inversely, α7-specific antibodies stimulate TLR-4 pro-inflammatory signaling, predicting close TLR-4 and α7 nAChR proximity in the plasma membrane. Although LPS failed to prevent binding of the α7-specific antibodies to U373 cells in flow cytometry (data not shown), one might predict an

intersection of α7 nAChR signaling with TLR-4 at the level of adaptor proteins or Src-kinases, i.e., at the very upstream, plasma membrane-proximal stage.

Taken together, our current findings, summarized in **Figure 8**, demonstrate that:


This data support the idea that excessive inflammation is an important pathogenic factor stimulating neurodegenerative processes. In this context, α7 nAChR-specific antibodies may play a dual role: potentiating a certain level of inflammation but preventing its neurodegenerative consequences.

#### AUTHOR CONTRIBUTIONS

MS, SK, and HS: substantial contributions to the conception and design of the work; OL, NM, LK, OK, GG, and KU: the acquisition, analysis, and interpretation of data for the work; MS, NM, and HS: drafting the work; OL, LK, OK, GG, KU, and SK: revising it critically for important intellectual content; OL, NM, LK, OK, GG, KU, SK, HS, and MS: final approval of the version to be published. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved (OL, NM, LK, OK, GG, KU, SK, HS, and MS).

#### FUNDING

Support of this study by the European Research Council under the European Union's Seventh Framework Programme (FP7/2007–2013)/ERC Advanced Award 321501 and the Israel Science Foundation grant no.817/13 (to HS) is acknowledged. NM was a recipient of post-doctoral

#### REFERENCES


fellowships from the ELSC Brain Center and The Israeli Government.

#### ACKNOWLEDGMENTS

We are grateful to Dr S. Karakhim for the help in confocal microscopy studies.

mitochondria through kinase-mediated pathways. Int. J. Biochem. Cell Biol. 49, 26–31. doi: 10.1016/j.biocel.2014.01.001


**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 © 2016 Lykhmus, Mishra, Koval, Kalashnyk, Gergalova, Uspenska, Komisarenko, Soreq and Skok. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY)The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.