Derived Neurons of Children with Neurodevelopmental Disorders

Alterations in Transcription and Connectivity in Stem Cell- Derived Neurons of Children with Neurodevelopmental Disorders

Kirill Zaslavsky

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Molecular Genetics University of Toronto

Alterations in Transcription and Connectivity in Stem Cell-Derived Neurons of Children with Neurodevelopmental Disorders

Kirill Zaslavsky

Doctor of Philosophy

Department of Molecular Genetics University of Toronto

2017 Abstract

Many neurodevelopmental disorders (NDD) have a genetic etiology. Some arise from mutations at a specific locus whereas others are polygenic. For example, Williams-Beuren

Syndrome (WBS) is caused by hemizygous deletions at 7q11.23 that result in intellectual disability and increased socialization. In contrast, autism spectrum disorder (ASD) is common, results in impaired socialization, and has a polygenic, sex-dependent pattern of inheritance.

While these NDDs involve genes in transcription and synapse function, it has been challenging to discover associated neuronal phenotypes owing to a lack of suitable human tissue. The recent development of human induced pluripotent stem cells (IPSC) allows to bridge this gap by providing a renewable source of neurons specific to individuals with NDDs. To examine transcription, I used IPSCs from a child with WBS and two children with ASD with mutations at the PTCHD1/PTCHD1AS locus, which encodes a protein and long non-coding RNA (lncRNA) of unknown function. In WBS neurons, I found an overall reduction in expression of synapse, voltage-gated potassium channel and extracellular matrix genes. In ASD neurons, I found a reduction in genes involved in RNA splicing, export, cellular respiration, and an increase in

ii extracellular matrix genes. The large range of transcriptional alterations suggests that mechanisms of disease in these NDDs are complex and varied.

To assess synaptic connectivity in ASD, I compared IPSC-derived neurons from two cases with mutations in SHANK2, a synaptic scaffolding gene, and engineered SHANK2 knockouts to parental and isogenic controls. Because heterogeneity among IPSC-derived neurons can compromise connectivity comparisons, I developed a Sparse coculture for Connectivity

(SparCon) assay that increases statistical power by sparsely plating differentially labeled control and mutant neurons onto a lawn of unlabeled control neurons. Using SparCon, I found that dendrite complexity and total synapse number are increased in SHANK2 mutant cells, which was paralleled by increased excitatory synaptic activity. These findings are the first example of excitatory hyperconnectivity in IPSC-derived neurons from subjects with ASD and provide evidence that SHANK2 functions in early neuronal development as a suppressor of dendrite branching.

In summary, IPSC-derived neurons are beginning to reveal transcriptional and synaptic neuronal phenotypes in vitro that are associated with NDDs. As the model system matures, methodological refinements like SparCon will further increase the ability of this system to discover disease phenotypes. Investigating alterations in transcription of synaptic genes in WBS,

RNA splicing in PTCHD1/PTCHD1AS ASD, and hyperconnectivity in SHANK2 ASD will yield further insight into the pathophysiology of these NDDs.

iii

Acknowledgments

The PhD is a physically and mentally draining process. I would not have been able to complete it without the support of my laboratory, my supervisor, my dear friends, and relatives.

In particular, I want to thank my supervisor James Ellis for giving me the opportunity to work in his lab and for allowing me to pursue my ideas. I am indebted to Joel Ross for inspiring me to work in this field. I am lucky I have found a great friend in labmate Deivid Rodrigues, who helped me keep Fresh Start in business by dutifully buying coffee every day and lending a patient ear. He has helped me shape my ideas and has been crucial source of support. I would also like to thank Eric Deneault, for I have not seen anybody that works harder. My work would also not have been possible without the great support of the lab technicians who help us feed and differentiate our cells – namely Alina Piekna, Wei Wei, and Asli Romm. Peter Pasceri, our lab manager, made the lab a welcoming and warm place to work. I also thank Rebecca Mok for helpful discussions on the project.

My friends have been an unwavering source of support. I’d like to thank Ilya

Mukovozov, who has been by my side for the past 7 years going through MD and PhD together.

Richard Wu has been the most amazing workout buddy and has kept me focused. I would be remiss if I did not acknowledge others that have helped me through it all: Rob Vanner for being a great mentor, Brian Ballios for always being an inspiration, Amanda Khan for her unwavering positivity and support, Priya Dhir for listening to me rant about things that almost always turned out better than I expected, and Richard Gao for being my one of my best friends since grade 8.

Lastly, this would not have been possible without the support of my parents. I thank my mother, Elena, who is the living embodiment of the American Dream, an immigrant who arrived with nothing and achieved so much. I am also thankful to my father, Ilya, for helpful discussion

iv and support. I am constantly inspired by my little brother, Maxim, who is the smartest person I know and at the age of 20 is set to begin grad school. I cannot wait to see what you will accomplish.

Table of Contents

Acknowledgments ...... iv

Table of Contents ...... vi

List of Tables ...... x

List of Figures ...... xi

Chapter 1 Introduction ...... 1 1.1 Overview of NDDs ...... 1 1.2 Overview of WBS ...... 3 1.2.1 Brief History of WBS ...... 3 1.2.2 Neurodevelopmental Alterations in WBS ...... 4 1.2.3 Genetic Architecture of WBS ...... 5 1.2.4 Structural brain changes in WBS ...... 7 1.2.5 Transcriptional and connectivity alterations in WBS ...... 8 1.3 1.3 Overview of ASD ...... 10 1.3.1 Brief history of ASD ...... 10 1.3.2 Diagnosis and Clinical Presentation ...... 12 1.3.3 Genetic Architecture of ASD ...... 14 1.3.4 Transcriptional and morphological alterations in post-mortem ASD brains ...... 20 1.3.5 Genetic evidence for PTCHD1/PTCHD1AS ...... 22 1.3.6 Genetic evidence for SHANK2 ...... 23 1.4 SHANK proteins are critical for development of synaptic connectivity...... 24 1.4.1 SHANKs coordinate synapse development and function ...... 25 1.4.2 SHANK1, SHANK2, and SHANK3 have overlapping and distinct functions...... 27 1.4.3 Non-synaptic roles of SHANK ...... 29 1.4.4 Mouse models of SHANK2-associated ASD exhibit inconsistent phenotypes ...... 30 1.5 IPSC in NDD modeling & Drug Discovery ...... 32 1.5.1 Gene editing in human iPSC ...... 33 1.5.2 Three Broad Types of Neuronal Differentiation Protocols ...... 36 1.5.3 Recent Progress in IPSC models of NDDs ...... 41 1.5.4 Challenges in IPSC models of NDDs...... 43 1.6 1.6 Rationale ...... 46 vi

Chapter 2 Transcriptional Alterations in WBS and ASD ...... 49

...... 51 2.1 Abstract ...... 51 2.2 Brief Introduction and Rationale ...... 52 2.3 Summary of electrophysiological phenotypes in neurons of NDD cases ...... 53 2.3.1 WBS ...... 53 2.3.2 PTCHD1/PTCHD1AS ASD ...... 54 2.4 Results ...... 57 2.4.1 Reduction in expression of potassium channel, ECM, synapse genes in WBS ...... 57 2.4.2 Elevated ECM and altered DNA-binding gene expression in purified neurons of PTCHD1/PTCHD1AS cases ...... 64 2.5 Summary and Discussion ...... 74 2.5.1 Transcriptional alterations in WBS neurons ...... 74 2.5.2 Transcriptional alterations in PTCHD1/PTCHD1AS mutant neurons ...... 76 2.6 Materials and Methods ...... 78 2.6.1 WBS microarrray analysis ...... 78 2.6.2 Microarray analysis of PTCHD1/PTCHD1AS mutant neurons ...... 79

Chapter 3 SparCon reveals hyperconnectivity of SHANK2 mutant human neurons ...... 81

...... 82 3.1 Abstract ...... 82 3.2 Brief Introduction and Rationale ...... 83 3.3 Results ...... 84 3.3.1 Generation of IPSCs from controls, ASD cases, and CRISPR/Cas9n knockout of SHANK2 84 3.3.2 Development of the SparCon Assay ...... 91 3.3.3 Generation and characterization of connectivity data using SparCon ...... 100 3.3.4 Within-well normalization ...... 101 3.3.5 Power Simulations ...... 109 3.3.6 Hyperconnectivity in SHANK2 mutant human neurons ...... 111 3.4 Brief Summary and Discussion ...... 117 3.5 Materials and Methods ...... 120 3.5.1 Generation of iPSCs ...... 120 3.5.2 Generation of SHANK2 knockout cells...... 121

vii

3.5.3 Characterization of off-target effects resulting from CRISPR/Cas9n knockout of SHANK2 122 3.5.4 Differentiation to Neural Precursor Cells (NPCs) ...... 123 3.5.5 Neuronal Differentiation ...... 124 3.5.6 Sparse-seeding co-culture ...... 125 3.5.7 Virus preparation ...... 126 3.5.8 Astrocyte Preparation ...... 127 3.5.9 Cloning ...... 127 3.5.10 Western blot ...... 127 3.5.11 Mycoplasma testing ...... 128 3.5.12 Immunocytochemistry ...... 128 3.5.13 Image acquisition ...... 129 3.5.14 Synapse and morphology analysis ...... 130 3.5.15 Statistical analysis ...... 131 3.5.16 Power Simulations ...... 132 3.5.17 Electrophysiological recordings ...... 133 3.5.18 Code ...... 134

Chapter 4 Discussion and Future Directions ...... 152

...... 153 4.1 Transcriptome analysis in IPSC-derived neurons of NDDs ...... 153 4.1.1 WBS neuronal transcriptome - differences with published datasets ...... 153 4.1.2 PTCHD1AS in ASD neuronal transcriptome ...... 159 4.1.3 Importance of ECM in NDDs ...... 163 4.1.4 Towards an ideal gene expression experiment using IPSC-derived neurons ...... 164 4.2 SparCon assays enable connectivity measurements and guide mechanistic investigation 168 4.2.1 SparCon allows for isolated investigation of synaptogenesis and guides mechanistic investigation ...... 168 4.2.2 Comparison of SparCon with other methods ...... 169 4.3 Role of SHANK2 as an early regulator of neuronal development ...... 172 4.3.1 Alterations in neuronal connectivity in ASD ...... 174 4.4 Future Directions ...... 175 4.4.1 Increasing the range of phenotypes assayable by SparCon ...... 175 4.4.2 Alternative splicing and chromosome conformation in PTCHD1AS neurons ...... 176

viii

4.4.3 Testing ECM function in NDD ...... 177 4.4.4 SHANK2 beyond the synapse ...... 178 4.4.5 Towards understanding molecular mechanisms of social ability ...... 179

References ...... 181

List of Tables

Table Page

Table 1. Summary of neuronal differentiation approaches. 40

Table 2. Summary of IPSC lines used for transcriptome analyses 56

Table 3. Analysis of SHANK2 KO (Vic tag) insertion at potential gRNA 91 binding sites with up to three mismatches.

Table 4. Comparisons between control and mutant cells on each coverslip used 107 for immunocytochemistry-based analysis of connectivity.

Table 5. Comparisons in each well for electrophysiological analysis of synapse 108 function.

Table 6. Summary of experiments performed. 135

Table 7. Batches of neurons used in sparse coculture experiments. 136

Table 8. Materials used. 137

List of Figures

Figure Page

Figure 1. Architecture of the 7q11.23 locus 6

Figure 2. Diagnosis of ASD 13

Figure 3. The distribution of ASD cases by gender and non-verbal IQ. 15

Figure 4. De novo variation influences a continuum of functional outcomes in 16

ASD

Figure 5. Summary of changes observed in postmortem ASD brains. 21

Figure 6. SHANK2 function at the synapse. 24

Figure 7. SHANK2 exists as multiple isoforms as a result of alternative 26 splicing and differential promoter usage.

Figure 8. Approach to modeling NDDs using IPSC-derived neurons. 33

Figure 9. Three general approaches for IPSC differentiation into neuronal 36 tissue.

Figure 10. Hierarchical clustering of data from our lab integrated with that 58 from the Gage lab.

Figure 11. qRT-PCR validation of 25 neuronally expressed genes. n=3 59 independent samples

Figure 12. Changes in expression of top 15 downregulated genes in the 60 voltage-gated potassium channel complex gene set from gene ontology database.

Figure 13. Top two negatively enriched gene sets with the top ten 61 downregulated genes by fold change in WBS neurons relative to controls.

Figure 14. The largest interconnected network (46 out of a total 136 nodes) is 63 comprised of downregulated gene sets governing synaptic function.

Figure 15. Interconnected network of gene sets governing extracellular matrix 64 function is downregulated in WBS relative to WT neurons.

Figure 16. Overview of the MACS procedure to purify neurons. 66

Figure 17. Quality control identifies systematic aberrations in replicate b of 68

Proband 2 line #15.

Figure 18. Differential transcript expression between neurons of control and 70 probands with mutations at the PTCHD1 locus.

Figure 19. Genes in the smoothened signaling pathway are not differentially 72 expressed in Proband 1 relative to controls except for PTCHD1.

xii

Figure 20. Gene set enrichment analysis in Proband 1 (a) and Proband 2 (b) 73 reveals that RNA splicing, RNA export, cellular respiration and extracellular matrix gene sets are differentially expressed in the same direction in both probands relative to controls.

Figure 21. Sparse co-culture for connectivity (SparCon) assays of IPSC- 85 derived SHANK2 ASD neurons compare marked mutant and control neurons seeded on the consistent synaptogenic environment of a lawn of unlabeled control neurons.

Figure 22. Recruited families, iPSC generation and characterization. 87

Figure 23. Characterization of CTRL1 excerpted from manuscript in 88 preparation by Ross et al.

Figure 24. Generation of SHANK2 homozygous knockout. 89

Figure 25. Summary of challenges with measuring connectivity. 93

Figure 26. Generation of NPCs and neurons. 95

Figure 27. Western blot and quantification of SHANK2 protein levels in iPSC, 97

NPC, and iPSC-derived neurons.

xiii

Figure 28. Generation of functional neurons within 9 weeks in the sparse co- 99 culture

Figure 29. Batch-to-batch and well-to-well noise in connectivity measures can 102 be smoothed by within-well normalization.

Figure 30. Distribution of sEPSC frequency and sEPSC amplitude among 104 controls.

Figure 31. No bias due to fluorophore expression on connectivity 105 measurements.

Figure 32. Anderson-Darling k-samples test is more sensitive than 110

Kolmogorov-Smirnov or T tests for most measures and within-well normalization significantly increases the statistical power at smaller sample sizes.

Figure 33. Within-well normalization increases sensitivity of the assay to 111 detect connectivity differences.

Figure 34. Synapse numbers and neuronal complexity are enhanced in 113

SHANK2 mutant neurons.

xiv

Figure 35. Excitatory synaptic function is enhanced in SHANK2 mutant 115 neurons.

Figure 36. Increased sEPSC Frequency in SHANK2 R841X neurons in the 116 presence of BDNF (10 ng/mL) throughout the course of differentiation.

Figure 37. Unnormalized measurements from the connectivity datasets 119 generated for the study yield fewer statistically significant discoveries.

Chapter 1 Introduction 1.1 Overview of NDDs

Neurodevelopmental disorders (NDDs) affect up to 15% of all children (Boyle et al. 2011).

NDDs are diagnosed early in life and result in impairments in daily functioning that may be accompanied by learning disabilities. The causes of these disorders are complex, but a significant subset is known to have a genetic component (J. A. Lee & Lupski 2006; Geschwind & Flint

2015). Some arise from disruption at a single locus and result in recognized syndromes (J. A.

Lee & Lupski 2006). For example, Williams-Beuren Syndrome (WBS) affects 1/7500 of all children, is characterized by hypersociability, intellectual disability, craniofacial and heart abnormalities, and results from in a hemizygous loss of 26-28 genes at 7q11.23. The deleted genes affect multiple cell functions, some of which govern transcription and neuronal connectivity (Schubert 2009).

In contrast, autism spectrum disorder (ASD) affects 1/68 children (Christensen et al. 2016), is defined by impairments in social communication and the presence of restricted/repetitive behaviours, and is highly genetically heterogenous (Lai et al. 2014). With the introduction of high-throughput sequencing at the onset of the 21st century, the repertoire of genetic loci associated with ASD with high confidence has expanded to exceed 70 (Sanders et al. 2015), suggesting that possible underlying mechanisms are complex and varied. Collectively, these findings strongly implicate alterations in transcription and neuronal connectivity in ASD, as a large number of associated loci are clustered in these functional domains (De Rubeis et al. 2014;

Sanders et al. 2015). As with WBS patients, individuals with ASD and mutations at the same locus can vary highly in their clinical presentation, necessitating an individual-centered approach

1 2 to elucidate mechanisms of disease (Berg & Geschwind 2012). Here, I focused on cases with mutations at the PTCHD1/PTCHD1AS locus, hypothesized to alter the transcriptome, and on cases with mutations affecting SHANK2, thought to alter neuronal connectivity.

1.2 Overview of WBS

1.2.1 Brief History of WBS

First observations of what is now recognized as Williams-Beuren Syndrome can be traced back to reports in 1950s among children with infantile hypercalcemia, a condition that reached epidemic proportions in Britain due to vitamin D oversupplementation. Following a readjustment of vitamin D levels in food and the subsequent resolution of the condition, a proportion of infants became evident with persistent systolic murmurs, developmental delay and failure to thrive

(Bongiovanni et al. 1957; Stapleton et al. 1957; Schlesinger et al. 1956). Subsequent reports by

Williams and Beuren identified a population of children with supravalvar aortic stenosis

(SVAS), dysmorphic facial features and intellectual disability (Williams et al. 1961; Beuren et al.

1962). SVAS was rapidly identified as having a genetic cause. It segregated in families in an autosomal dominant fashion (Eisenberg et al. 1964) and nonfamilial cases shared a “peculiar facies suggesting kinship” (Garcia et al. 1964).

Discovery of genetic linkage of the ELN locus to SVAS (Ewart et al. 1993) led to the rapid identification of ELN deletions among WBS patients (Lowery et al. 1995). The deletions were found to be likely caused by unequal chromosomal rearrangements during meiosis resulting in hemizygous loss of 1.5 to 1.8 Mb at 7q11.23, affecting up to 26 to 28 genes (Schubert 2009), implying that multiple mechanisms may underlie the syndrome. Proteins involved in transcription (GTF2I, BAZ1B), translation (EIF4H), extracellular matrix (ELN), synaptic function (STX1A), and diverse signaling pathways (FZD9, LIMK1) are all part of the deleted region. Consequently, a large amount of effort has gone into establishing genotype-phenotype correlations and delineating effects of individual genes.

1.2.2 Neurodevelopmental Alterations in WBS

WBS is diagnosed early in life and affects multiple systems. Children with WBS have craniofacial abnormalities, mild microcephaly and cardiovascular problems. For a review of non- neurological alterations in WBS that is outside of the scope of this work, see Pober et al. 2010.

One of the striking features of WBS is an uneven cognitive profile (Pober 2010). Most children with WBS have mild-to-moderate intellectual disability and present with an average intelligence quotient (IQ) of 50-60, with a range of 40-112, which remains stable over time

(Osborne & Mervis 2007). This is accompanied by delay in acquisition of language skills

(Järvinen-Pasley et al. 2008) and impairments in discerning semantic meaning (Karmiloff-Smith et al. 1998). Despite this, children with WBS usually have a normal vocabulary and some may present with superior verbal abilites than age- and IQ-matched controls (Gosch et al. 1994;

Jarrold et al. 1998). However, nonverbal abilities in WBS are generally poor. Children with

WBS have impaired visuospatial ability and motor skills (Udwin & Yule 1991). They are less coordinated, learn to walk later, and lack fine motor skills and dexterity relative to controls.

The disposition of children with WBS is described as friendly, highly sociable and empathic

(Pober 2010). However, they are predisposed to develop phobias and experience higher levels of anxiety. Despite this, they generally show less activation of the amygdala than controls, the opposite of that observed in individuals with social anxiety disorder, suggesting a distinct pathology (Binelli et al. 2016). Indeed, they display an absence of social anxiety that is nevertheless accompanied by prominent anticipatory anxiety (Pober 2010). Many have a strong engagement with music, and show a relative strength in that domain. However, they are less able to attach semantic meaning to emotion in music (Hopyan et al. 2001). WBS is also associated

5 with attention deficit hyperactivity disorder, with children displaying poor concentration, social disinhibition and hyperactivity (Leyfer et al. 2006).

1.2.3 Genetic Architecture of WBS

WBS is caused by hemizygous deletions at the 7q11.23 locus that may affect up to 28 coding genes. Most deletions range from 1.5 to 1.8 Mb, though some as small as 0.1 and 2.5 Mb have been found. In almost all cases, deletions occur de novo. The WBS deletion region consists of a

1.2 Mb single-copy gene region that is flanked on each side by large low copy repeats (LCR) blocks A, B and C. These blocks are ordered in 320kb complexes located on centromeric, medial and telomeric parts of 7q11.23. LCR blocks have high sequence similarity and are separated by

Alu elements, predisposing to chromosomal misalignment during non-allelic homologous recombination (NAHR) in meiosis, which can cause deletions leading to WBS. Centromeric and medial B blocks have highest sequence similarity at over 99.5%, and NAHR between these B blocks accounts for 90% of deletions in WBS, causing a ~1.5 Mb loss. Recombination between centromeric and medial LCR block A copies accounts for most of other deletions and causes a

~1.8 Mb loss (Figure 1). The remaining smaller deletions result because of breakpoints within the single-copy gene region. These can be restricted to just the ELN gene and cause SVAS, or cover a larger number of gene and cause either a partial or full spectrum of WBS phenotypes

(Schubert 2009).

Figure 1. Architecture of the 7q11.23 locus. A single-copy gene region is flanked by low- copy repeat blocks with high sequence similarity. The most common incidence of NAHR occurs between centromeric and medial B blocks, with the second most common between centromeric and medial A blocks. These result in a hemizygous deletions of 26-28 genes.

It is worth noting that in addition to deletions, duplications represent 1/3 of all genomic lesions at 7q11.23 and lead to a neurodevelopmental condition distinct from WBS. These individuals have severe delay in the development of expressive language, and are at high risk of

ASD and ADHD (Sanders et al. 2011). This may be accompanied by macrocephaly, hypotonia, epilepsy, cerebellar dysfunction. In addition, most of these individuals have concomitant

7 cerebellar vermis hypoplasia, decreased white matter volume and ventriculomegaly (Morris et al.

2015).

Lastly, the structure of the WBS locus can result in inversions after NAHR. These are usually asymptomatic and are present at 5% in the non-WBS population. However, these present a higher risk of chromosomal misalignments during NAHR, that can lead to deletion and duplications. The risk of WBS-inversion carrier for a child with WBS is 1/500 (Schubert 2009;

Pober 2010).

1.2.4 Structural brain changes in WBS

Individuals with WBS typically have a 10-15% reduction in overall brain size, which likely arises due to early reduction in cortical surface area (Green et al. 2016). Parietal, occipital lobes, thalamus, basal ganglia and midbrain are disproportionately decreased in overall volume, while the temporal and frontal lobes are not (Chiang et al. 2007; Campbell et al. 2009). Indeed, some studies reveal increases in grey matter volume in the frontal and temporal lobes (Campbell et al.

2009), while others point to decreases in surface area that are accompanied by increased cortical thickness (Meda et al. 2012). Decreases in white matter are significantly greater than those in grey matter and may underlie impaired long-range connectivity (Jackowski et al. 2009; Green et al. 2016).

Several attempts have been made to correlate changes in brain anatomy with performance in specific functional domains. For example, auditory cortex volume is increased in WBS patients with increased musical abilities (Martens et al. 2010). Volumetric differences in frontal lobes, caudate nucleus and cerebellum are associated with inattention and altered activity in the

8 prefrontal cortex may cause the lack of inhibition that could underlie hypersocialization

(Campbell et al. 2009; Barak & Feng 2016). Cerebellar dysfunction may underlie motor symptoms in WBS and while cerebellar volume is also decreased compared to controls, the decrease is proportionally smaller than the decrease in cerebral volume (Osório et al. 2014). In addition, volumetric alterations in cerebellum, temporal and parietal lobes are associated with hyperactivity. However, caution is necessary in interpreting these results as control individuals are rarely matched to WBS individuals by IQ (Jackowski et al. 2009).

1.2.5 Transcriptional and connectivity alterations in WBS

Significant effort has been expended to establish genotype-to-phenotype correlations in

WBS. However, heterogeneity in WBS phenotypes among patients, even for those with the same deletions, has confounded this analysis (Schubert 2009). Nevertheless, it is certain that ELN mutations cause SVAS and are responsible for the majority of cardiovascular symptoms (D. Y.

Li et al. 1997). Deletions that encompass the NCF1 gene decrease the risk of hypertension (Del

Campo et al. 2006). Some rare partial deletions suggest that GTF2I and GTF2IRD1 hemizygosity underlies the WBS neurocognitive profile (Antonell et al. 2010). Lastly, individuals with WBS bearing atypical 7q11.23 deletions sparing CLIP2 have mild facial anomalies, mostly spared intellectual ability, as well as normal visuospatial and motor performance compared to individuals with typical deletions that include CLIP2 (Gagliardi et al.

2003; Ferrero et al. 2010).

Analysis of deleted genes strongly implicates two processes in pathogenesis of neuronal symptoms, neuronal signaling and transcription. LIMK1 appears to be important for the visuospatial impairment in WBS, as a subset of patients with small deletions encompassing only

ELN and LIMK1 present with SVAS and impaired visuospatial constructive cognition, with decreased performance on tasks that require reproduction of viewed objects (Frangiskakis et al.

1996). Consistent with this, LIMK1 knockout mice show impaired re-learning of spatial information, altered fear responses and display abnormalities in synapse function and development of dendritic spines(Meng et al. 2002). Another gene, Syntaxin-1a (STX1A) is known to be important for exocytosis and plays a role in neurotransmitter release(Bennett et al.

1992; Xu et al. 2013). Interestingly, STX1A knockout mice show alterations in synaptic plasticity and impaired extinction of contextual fear memories, possibly paralleling the increased propensity to develop fears among children with WBS(Fujiwara et al. 2006).

WBS deletions affect at least four transcription factors (MLXIPL, GTF2I, GTF2IRD1,

BAZ1B), suggesting that transcriptional alterations may underlie some WBS phenotypes. Little is known about MLXIPL function. However, GTF2IRD1 knockout mice have craniofacial abnormalities resembling WBS, display increased social behaviour, reduced fear and aggression reminiscent of WBS, and have increased serotonin metabolism (Young et al. 2008). Furthermore, modeling using induced pluripotent stem cells (IPSCs) derived from WBS patients shows that

GTF2I may control a transcription program important for axonal guidance and development that can be observed in undifferentiated cell types, such as iPSCs and neural precursor cells (NPC)

(Adamo et al. 2015). Similar studies in iPSC-derived neurons revealed that BAZ1B controls multiple gene expression programs important for neural development, and may contribute to up to 42% of transcriptional dysregulation in WBS neurons (Lalli et al. 2016).

1.3 1.3 Overview of ASD

Autism spectrum disorder (ASD) is a neurodevelopmental condition that results in impairments in social communication and the presence of restrictive, stereotyped behaviours and interests. The reported prevalence of ASD has risen over the past several decades to 1/68 today

(Wingate et al. 2014). This is due to better ascertainment, broadened ASD diagnostic criteria and decreased stigma (Rutter 2005). The severity and phenotypic presentation is highly variable.

Approximately 40% of children with ASD also have intellectual disability, at least 30% ADHD, and other comorbid conditions (J. A. Chen et al. 2015). The causes of ASD are presumed to be predominantly genetic, with 38-90% concordance rates in monozygotic twins (Geschwind &

Flint 2015). Consequently, a large amount of effort has been directed towards uncovering genetic causes of ASD. In contrast to WBS, which is linked to deletions at 7q11.23, at least 71 loci are significantly associated with ASD, with most involved in synapse function and transcriptional control (De Rubeis et al. 2014; Sanders et al. 2015). The predominance of synaptic genes has given rise to the hypothesis that ASD is a disorder of altered neuronal connectivity. Mutations in transcriptional genes may contribute to this pathology by dysregulating neuronal maturation and impacting the development of connectivity. While post-mortem brains from ASD subjects and functional magnetic resonance imaging lends evidence in support of this idea (Rudie & Dapretto

2013), it has been difficult to test this hypothesis in live human neurons due to a lack of available samples.

1.3.1 Brief history of ASD

Autism (from Greek “autos” or “self”) was first used to describe patients with schizophrenia in early 1900s. Following Leo Kanner’s description of 11 children with extraordinary memory

11 yet with “extreme autistic aloneness” and “anxiously obsessive desire for sameness” in 1943

(Kanner 1943), this association persisted as autism was widely interpreted as childhood-onset schizophrenia. It was believed to have a psychogenic rather than an “inborn” cause as proposed by Kanner, and attributed to a lack of maternal attention (“refrigerator mothers”). The parents of these children were equally interesting: Kanner called them “successfully autistic”, as many were highly intelligent, yet more at ease relating to concepts rather than people. At the time, autism was thought to be secondary to mental retardation, as the vast majority of children with autistic features were also intellectually disabled. In 1966, the prevalence of autism was estimated at 2.3 children in 100000 (Rutter 2005).

As developmental psychologists started to recognize social communication as a distinct domain of normal development, autism began to be re-conceptualized as a failure to develop social capability separate from acquisition of intellectual ability. Far more important became the ascertainment of whether a child’s social functioning was similar to that of his or her age- matched peers. This paradigm shift led to the ability to diagnose autism in a variety of syndromes, such as tuberous sclerosis or congenital rubella. In 1980, autism finally appeared as a separate category in DSM-III, but its distinction from intellectual disability was not yet made clear (Rutter 2005).

Around this time, the research community began to appreciate that autism appeared on a spectrum with varying degrees of severity. The 1994 DSM-IV made diagnostic criteria more inclusive by creating an umbrella term of Pervasive Developmental Disorders (PDD, also known as Autism Spectrum Disorders (ASD)), which encompassed Autistic Disorder, PDD not otherwise specified, Childhood Disintegrative Disorder, Rett Syndome, and Asperger’s Disorder.

Autistic disorder retained its close association with intellectual disability, as Asperger’s Disorder

12 became the diagnosis for children with impaired social communication, yet normal language development and intellectual ability. At the same time as diagnostic criteria broadened, autism surveillance and detection programs improved, and led to markedly increased estimates of ASD prevalence.

The diagnostic criteria were further broadened in DSM-V in 2013, which combined PDD-

NOS, Asperger’s and Autistic Disorders under the umbrella diagnosis of ASD. The DSM-V criteria encouraged early detection of ASD in young children that present with impairments in social communication and the presence of repetitive/stereotyped behaviours. Additionally, ASD could now be diagnosed alongside other disorders, such as ADHD (Rutter 2012). With these changes, the prevalence of ASD is now estimated to be 1/68 children (Christensen et al. 2016).

1.3.2 Diagnosis and Clinical Presentation

ASD is diagnosed in children as young as two years old according that present with impairments in social communication (3 out of 3 criteria must be met) and evidence of restricted interests or repetitive behaviours (2 out of 4 criteria must be met, Figure 2). In addition, symptoms must manifest in early development, there must be a clinically significant impairment in areas of current functioning, and the impairment in social communication must be greater than that expected for general developmental level that is not better explained by intellectual disability (American Psychiatric Association n.d.).

Figure 2. Diagnosis of ASD. Diagnosis is made on the basis of impairments in social communication and the presence of repetitive behaviour and restricted interests.

1.3.3 Genetic Architecture of ASD

Most studies report ASD heritability in excess of 70% (Lai et al. 2014). ASD affects 4 times as many boys as girls (Wingate et al. 2014). This risk is modulated by intellectual function, with increased male:female bias among high-functioning cases and decreased bias in cases with intellectual disability (Yeargin-Allsopp et al. 2003; Banach et al. 2009; Werling & Geschwind

2013) (Figure 3). This profound gender bias is further reflected in a higher risk of ASD diagnosis in siblings of affected females rather than males (Jorde et al. 1991). These features are consistent with a sex-dependent multifactorial model, wherein ASD risk is associated with multiple loci, with females requiring a higher mutation burden (Rice et al. 1980). Given the phenotypic heterogeneity of ASD, even among related or IQ-matched cases, the possible repertoire of associated loci is likely to be very large (Ronemus et al. 2014).

The first step towards transitioning from a symptomatology of ASD to a mechanistic explanation is elucidation of the underlying genetic architecture. While the genetic etiology of

70% of cases is unknown, the severity of symptoms can be scored on a continuous scale and is likely due to polygenic risk arising from variants of small effect. For example, ASD cases in the

Simons Simplex Collection (SSC) can be separated from controls using the Vineland Adaptive

Behavior Score, with lower scores correlating with higher rates of de novo variation (Figure 4)

(Robinson et al. 2016).

Figure 3. The distribution of ASD cases by gender and non-verbal IQ. Cases in the

Simons Simplex Collection reveals that the male:female bias is modulated by non-verbal IQ.

Reproduced with permission from Ronemus et al. Nature Reviews Genetics 2014.

Figure 4. De novo variation influences a continuum of functional outcomes in ASD. (A)

ASD cases from the Simons Simplex Collections can be separated from controls along a continuous standard Vineland Adaptive Behaviour Score. (B) The rate of de novo variation in exomes of ASD cases is higher than that in controls. Reproduced with permission from Robinson et al. Nature Genetics 2016 . SSC: Simons Simplex Collection, LoF: loss-of-function, DCM: heterozygous missense mutations, ExAC: Exome Aggregation Consortium.

The other 30% are spread between syndromes associated with ASD arising from specific genomic lesions (10-20% of cases), copy number variants (CNVs), small indels, and single nucleotide variants (SNVs). (Devlin & Scherer 2012; J. A. Chen et al. 2015). This subset of cases is particularly informative, as it appears to be caused by rare genetic variation - infrequent, but highly penetrant mutations affecting at most 1% of cases each.

1.3.3.1 Insights into ASD pathophysiology from syndromic ASD cases

Syndromic forms of ASD arising from mutations in a single gene have provided the first indication that alterations in transcription and neuronal connectivity may underlie ASD pathophysiology. Mutations causing Rett Syndrome (MECP2), Timothy Syndrome (CACNA1C),

17 and Fragile X Syndrome (FMR1) all impair neuronal function. The absence of MECP2, a chromatin regulator, results in decreased transcription, and smaller, poorly branched neurons that make few synapses (Marchetto et al. 2010; Cheung et al. 2011; Y. Li et al. 2013). Mutations in

CACNA1C alter voltage-gated calcium channel function resulting in lowered neuronal activity and abnormal dendrite development, particularly when it depends on synaptic and neuronal activity (S. P. Paşca, Portmann, Voineagu, Yazawa, Shcheglovitov, Paşca, Cord, Palmer,

Chikahisa, Nishino, Bernstein, Hallmayer, Geschwind & Dolmetsch 2011a; Krey et al. 2013).

Mutations in FMR1, which encodes Fragile X Mental Retardation Protein (FMRP), among many other changes, impair neuronal development (Halevy et al. 2015) and synaptic plasticity (Huber et al. 2002). FMRP is a regulator of mRNA translation and functions primarily at neuronal synapses on the receiving (postsynaptic) side to control expression of synaptic proteins in response to neuronal activity. Strikingly, many of its mRNA targets have been identified as ASD candidate genes (Darnell et al. 2011).

Multigenic ASD syndromes arise due to CNVs causing either duplication or deletion of multiple genes. In these cases, genotype-to-phenotype relationships are less clear, but careful examination of patients with partial or smaller deletions can provide possible clues about underlying pathophysiology. For example, ~600 kb deletions at 16p11.2 are among the most prevalent and penetrant with respect to ASD, representing ~1% of all cases (Devlin & Scherer

2012). While the region encodes 29 transcripts, narrow deletions of KCTD13 have been observed in a subject with macrocephaly, a key feature of the syndrome. Manipulation of KCTD13 levels in mice and zebrafish allows for bidirectional control of brain size, indicating a casual relationship (Golzio et al. 2012). Deletions at Xp22.3 encompass genes NLGN3 and NLGN4, which have now been strongly associated with ASD (Thomas et al. 1999; Jamain et al. 2003) and are important for synapse formation (Bourgeron 2009). Deletions at 22q13 cause Phelan-

McDermid Syndrome (PMDS), a severe neurodevelopmental condition in which at least 50 % of children are diagnosed with ASD. In PMDS, one of the deleted genes is SHANK3 (SH3 and multiple ankyrin repeats protein 3), which is found mutated in non-syndromic ASD. SHANK3 is regulated by FMRP and involved in formation, maintenance, and function of excitatory synapses

(Phelan & McDermid 2012). Genotype-to-phenotype correlations for single genes have been difficult to find for other syndromes, but important associations exist between ASD and 15q11-

13 duplication, 7q11.23 duplication (WBS deletion region), and 22q11.2 deletion (J. A. Chen et al. 2015).

1.3.3.2 Insights in ASD pathophysiology from nonsyndromic ASD cases

With the advent of whole-genome screening in the mid-2000s, the repertoire of loci involved in non-syndromic ASD has rapidly expanded. In a recent meta-analysis, 71 loci containing 65 genes have passed rigorous statistical thresholds for association with ASD. Mutations at these loci mostly occur de novo, and no locus can account for more than ~1% of all cases. The majority of these encode proteins that interact with each other physically, suggesting a convergent biology (Sanders et al. 2015). Based on protein-protein interactions, these proteins can be separated into two clusters - synapse function and transcription/chromatin regulation (De

Rubeis et al. 2014; Sanders et al. 2015). For example, mutations in both SHANK2 and SHANK3 are associated with ASD, with mutations in SHANK2 being exclusively associated with nonsyndromic ASD (Leblond et al. 2014). SHANK2 has a hub-like position in the protein-protein interaction network composed of high-confidence ASD genes (Sanders et al. 2015). On the other hand, mutations at the PTCHD1/PTCHD1AS (Patched domain containing protein 1 and the

19 associated divergently transcribed long noncoding RNA) locus are highly prevalent among ASD cases (0.5%), but the function of its products is unknown (Devlin & Scherer 2012).

Most of the variation discussed above predominantly occurs de novo, with a small role for common and rare inherited variation. Children born with ASD have a higher de novo CNV burden (2-7% relative to 1% in controls), and the CNVs are larger, affecting more genes. The number of de novo single nucleotide variants (SNVs) is also increased and mostly comes from the paternal chromosome, reflecting increased paternal age as a risk factor for ASD (Sanders et al. 2015; Sandin et al. 2016). While such a mechanism is consistent with the “copy error” hypothesis, which states that de novo mutations could propagate and accumulate in successive generations of spermatogonia (Kong et al. 2012), it is unlikely to be sufficient as a sole explanation for paternal age as a risk factor. Some of the risk may be explained by possible failure of parental imprinting, as several parentally-imprinted regions have been associated with

ASD risk, most notably 15q11.13 (Chaste & Leboyer 2012). Lastly, the observation that parents of children with ASD may exhibit mild autistic behaviours (and may be categorized as having a broader autism phenotype) suggests that individuals with mild social impairment and genetic risk of ASD may take longer to father children due to a lower chance of establishing romantic relationships (Jobe & Williams White 2007; Byers et al. 2013).

The lack of known common variants for ASD is almost certainly due to insufficiently powered studies with samples sizes (all less < 1500 to date) that make it difficult to discover loci of small effect sizes. However, given the rate at which common risk variants are found for schizophrenia, a heritable neuropsychiatric disorder, in much larger cohorts, it can be estimated that up to 1000 loci can be associated with ASD if the study design was enlarged to include

60000 controls and cases each. Under the sex-dependent multifactorial model, 15-40% of ASD

20 risk should be attributable to common variation (J. A. Chen et al. 2015). Until this type of variation is characterized, the greatest insights can be gained by careful analysis of cases with specific mutations that have a relatively high prevalence (PTCHD1/PTCHD1AS) or have very strong genetic linkage to the disease and appear to be indicated as important functional hubs for processes affected in ASD, such as SHANK2.

1.3.4 Transcriptional and morphological alterations in post-mortem ASD brains

Post-mortem analyses represent an important step towards understanding the mechanisms underlying ASD pathophysiology. With recent increases in sample sizes used in post-mortem analysis, researchers have noted widespread disorganization in the prefronal cortex, limbic structures, and the cerebellum. Cortical minicolumns, considered the basic processing units of the brain, are smaller, but more dense (Casanova et al. 2006). Density of dendritic spines is increased in ASD subjects relative to controls in layer 2 of the frontal and parietal cortex and in both layers 2 and 5 of the temporal cortex (Hutsler & H. Zhang 2010; Tang et al. 2014), suggesting overproduction of synapses, impaired elimination of synapses (Tang et al. 2014), or both. The cerebellum exhibits a decreased number and size of Purkinje cells, but it is uncertain how this contributes to ASD (Fatemi et al. 2012) (Figure 5).

Figure 5. Summary of changes observed in postmortem ASD brains.

Transcriptional alterations find numerous differentially expressed genes that appear to be important at different stages of development. By analyzing samples from 2 to 56 year-old individuals with ASD, Chow et al. segregated genes into developmental and adult clusters. The developmental cluster involved genes that regulate cell cycle, cortical patterning, and neuronal differentiation. The adult set contained genes important for signaling and repair (Chow et al.

2012), concordant with a previous study (Voineagu et al. 2011). Lastly, two recent studies noted alterations in alternative splicing (AS). One found dysregulated AS in key neurodevelopmental genes, many of which are known ASD candidates. This is likely due to differential expression of

RBFOX1, which regulates developmental genes throughout development (Voineagu et al. 2011).

The second study discovered that ASD neurons have alterations in a conserved developmental

AS program for microexons of many genes, such as SHANK2. The functional significance of these microexons is not fully characterized, but may be important for specificity of protein- protein interactions (Irimia et al. 2014).

It is surprising that a large degree of convergent pathology has been found in postmortem studies of ASD, given the heterogeneity of ASD and the small sample sizes used in the studies.

The findings continue to strongly implicate transcriptional and connectivity alterations predicted by genetic screens. Thorough testing of the impact of mutations of ASD candidate genes on transcription and the development of neuronal connectivity is necessary to elucidate the underlying pathophysiology of ASD.

1.3.5 Genetic evidence for PTCHD1/PTCHD1AS

Deletions and mutations at the X-linked PTCHD1/PTCHD1AS locus are among the most commonly observed in ASD, affecting approximately 0.5% of all cases, all male. The locus features divergent transcription of a coding gene PTCHD1 (patched-domain containing protein

1) and the associated lncRNA PTCHD1AS. Despite the strong association with ASD, very little is known about the function of either product. It is also difficult to ascertain which is more important, as some mutations affect only PTCHD1, only PTCHD1AS or both. Further complicating the analysis is the fact that intellectual disability can also be present in many subjects with mutations at this locus (Noor et al. 2010; Devlin & Scherer 2012). It is possible that PTCHD1 is less important for ASD symptoms, as PTCHD1-/y knockout mice do not exhibit autistic behaviour, but instead have attention deficits due impairments in thalamoreticular circuits (Wells et al. 2016). However, such inferences must be checked against the vast differences in brain physiology and social biology between humans and mice. Careful study designs that delineate the individual roles of these products in neuronal function are necessary to understand the impact of mutation at this locus.

1.3.6 Genetic evidence for SHANK2

Over the last six years, SHANK2 has come to be recognized as a high-confidence risk gene for non-syndromic ASD. It occupies ~760kb on chromosome 11 and is relatively isolated from nearby genes. As such, all known deletions at the locus affect SHANK2 exclusively. Association between SHANK2 and non-syndromic ASD was first reported by Berkel et al. in 2010, identifying two male cases with de novo heterozygous putative loss-of-function mutations, one due to a nonsense mutation (R841X) and the other to a 66kb deletion affecting exons 15 and 16 common to all isoforms, causing a frameshift and premature stop (Berkel et al. 2010).

Subsequent studies and case reports identified seven additional male children with heterozygous large deletions or small indels, all de novo (Leblond et al. 2012; Chilian et al. 2013; Leblond et al. 2014). To date, no loss-of-function mutations affecting all isoforms have been observed in over 20000 controls. SNVs affecting evolutionarily conserved residues of SHANK2 are enriched among ASD cases, affecting up to 4.6% of cases (Leblond et al. 2012).

It is important to note that 3 cases with de novo SHANK2 deletions have an additional duplication of CHRNA7 gene, either parentally transmitted or due to a de novo translocation(Leblond et al. 2012; Chilian et al. 2013). While CHRNA7 is part of the 15q11-13 region, deletions of which cause ASD, there is insufficient evidence for an association of

CHRNA7 alone with ASD (Leblond et al. 2012). CHRNA7 has known associations with intellectual disability and schizophrenia (Bakanidze et al. 2013; Gillentine & Schaaf 2015).

Strikingly, SHANK2 has also been recently associated with both (Peykov et al. 2015; Leblond et al. 2014). All three children with SHANK2 deletions and CHRNA7 duplications have a diagnosis of intellectual disability (Leblond et al. 2014). Therefore, to elucidate the functional consequences of ASD-associated SHANK2 mutations in human neurons, a careful analysis of cases with SHANK2 mutations both with and without CHRNA7 duplications is necessary.

1.4 SHANK proteins are critical for development of synaptic connectivity.

At the turn of the millennium, four independent groups simultaneously reported the existence of large multi-domain neuronal scaffolding proteins (Du et al. 1998; Lim 1999; Boeckers et al.

1999; Zitzer et al. 1999), which appeared to be central to the function of the postsynaptic density

(PSD), a protein-dense meshwork of at least 2000 proteins responsible for processing excitatory synaptic input (Filiou et al. 2010). Eventually, the naming convention settled on SHANK (SH3 domain and multiple ankyrin repeats protein), with the three members of the protein family called SHANK1, SHANK2, and SHANK3. Through protein-protein interactions, SHANKs link together all major classes of glutamate receptors at the postsynaptic membrane with regulators of the actin cytoskeleton (Sheng & E. Kim 2000) (Figure 6).

Figure 6. SHANK2 function at the synapse. SHANK2 forms a scaffold that links glutamate receptors with actin regulators through protein-protein interactions. Through its multiple domains, SHANK2 interacts with other scaffolding proteins: with GKAP via its PDZ domain, which links it to PSD95 and NMDA receptors; with GRIP via its SH3 domain, which links it to AMPA receptors; with Homer1 via the Homer binding site (H in proline-rich region)), which links it to metabotropic glutamate receptors. SHANK2 interacts with several actin regulators: ⍺-fodrin via Ank Repeats; βPIX via the PDZ domain, which acts as a guanine exchange factor for RAC1 and CDC42; cortactin via the cortactin binding domain (C in proline- rich region). It also interacts with the molecular motor Dynamin2 with the dynamin-binding site

(D in the proline-rich region). SHANK2 is able to oligomerize with itself or possibly SHANK3 via the sterile alpha motif (SAM) in a zinc-dependent manner, which allows it to form a large scaffold or organize the post-synaptic density.

1.4.1 SHANKs coordinate synapse development and function

Given that synapses are highly dynamic structures that can be subject to extensive structural plasticity in response to synaptic input, SHANKs were immediately thought to be critical to synapse assembly and function. They were quickly shown to interact with multiple proteins of the PSD (Du et al. 1998; Boeckers et al. 1999; Tu et al. 1999; Sala et al. 2001; Okamoto et al.

2001; Romorini et al. 2004; Hwang et al. 2005; Hayashi et al. 2009; Park 2003), many of which were known or later discovered to be critical to processing of activity dependent signaling and associated with ASD (Ebert & Greenberg 2013). SHANKs were shown to be one of the most highly ubiquitinated proteins in the PSD, both during basal conditions and especially synaptic activity (Ehlers 2003). They are phosphorylated by CaMKII⍺, and together with another

26 scaffolding protein GKAP, support homeostatic synaptic scaling (Shin et al. 2012). It is likely that these posttranscriptional modifications alter the set of SHANK protein-interacting partners and could form the basis for one of the mechanisms for coordinating synaptic changes in response to activity. For example, following NMDA stimulation, SHANK-interacting protein cortactin, a regulator of actin cytoskeleton, translocates away from synapses to the dendrite

(Hering & Sheng 2003).

These functions are likely supported by the great variety of SHANK isoforms, which arises due to differential promoter usage and alternative splicing (Y.-H. Jiang & Ehlers 2013). These variations may alter the set of protein-interacting domains in the SHANKs and support cell-type specific functions. For example, SHANK2 exists as at least four isoforms, with the longest one containing the ankyrin repeats enriched in the cerebellum (Figure 7) (Leblond et al. 2012). Loss of SHANK2 from the cerebellum results in abnormal inhibitory signaling, with a higher frequency of inhibitory synaptic events that result in disorganized network activity (Peter et al.

2016).

Figure 7. SHANK2 exists as multiple isoforms as a result of alternative splicing and differential promoter usage. SHANK2 RNA can be detected in most tissues except skeletal

27 muscle and heart. The RNA of the longest SHANK2E isoform is known to be expressed in epithelial cells, the cerebellum, and variably in the neocortex. The RNA of the “brain-specific”

SHANK2A isoform is found in the neocortex. SHANK2B and SHANK2C RNA is widely expressed, with strongest levels in the brain. The repertoire of possible SHANK2 isoforms likely exceeds the four presented here, given the existence of additional RNA isoforms that can be found across different databases.

1.4.2 SHANK1, SHANK2, and SHANK3 have overlapping and distinct functions.

Though SHANKs are frequently studied together and assumed to overlap in function, mutations in different SHANKs are associated with different phenotypes in ASD cases, which merits the study of their individual roles. A critical aspect of the differences is the gradient in cognitive impairment: among individuals with heterozygous loss-of-function mutations, ASD subjects with SHANK1 mutations have normal intellectual ability (mean IQ 107), those with

SHANK2 generally have mild to moderate intellectual disability (mean IQ 62) , and those with

SHANK3 overwhelmingly have severe intellectual disability (mean IQ 31) (Leblond et al. 2014).

This gradient is paralleled in mouse knockout studies of SHANKs. Homozygous SHANK1 knockout mice have enhanced spatial learning (Hung et al. 2008), while SHANK3 knockout mice are mildly impaired (Wang et al. 2011; Kouser et al. 2013; J. Lee et al. 2015; Speed et al. 2015;

Jaramillo et al. 2016; Wang et al. 2016). Some cases with SHANK2 mutations are additionally reported as having craniofacial abnormalities (Chilian et al. 2013; Leblond et al. 2014) and those with SHANK3 mutations have manifestations of the Phelan-McDermid Syndrome (PMDS)

(Leblond et al. 2014). Heterozygous loss-of-function mutations in SHANK2 are specific for nonsyndromic ASD and are more frequent than mutations in SHANK1 (Leblond et al. 2014).

Unlike SHANK1, SHANK2 and SHANK3 have a zinc-sensitive sterile alpha motif domain, which allows these proteins to self-aggregate (Baron 2006; A. M. Grabrucker et al. 2011). Zinc is an important transsynaptic signal important for synapse formation and function, and there is some evidence of zinc deficiency in children with ASD (A. M. Grabrucker 2014).

In vitro, dosage of all SHANKs appears to bidirectionally control synapse density, and disease-associated variants frequently fail to induce excess synapse formation upon overexpression relative to the full-length WT proteins (A. M. Grabrucker et al. 2011; Arons et al.

2012; Leblond et al. 2012; Peykov et al. 2015). The effect size on the decrease of synapse density in the context of knockdown or deletion of individual SHANKs is variable, and tends to be approximately 20% (Sala et al. 2001; A. M. Grabrucker 2014). For example, SHANK2 knockdown results in a 20-50% decrease in synapse density (A. M. Grabrucker et al. 2011;

MacGillavry et al. 2015; Berkel et al. 2012), which could stem from variable knockdown efficiency and be mitigated by compensatory upregulation of other SHANKs, most likely

SHANK3 (Schmeisser et al. 2012).

SHANKs have different developmental kinetics. SHANK2 is expressed in early neuronal development, and can be found in the growth cones of developing rodent neurons. SHANK2 is the first to arrive at sites of nascent synapses, followed by SHANK3 and finally, by SHANK1 (A.

M. Grabrucker et al. 2011). The switch toward SHANK1 also appears to temporally coincide with functional maturation of synapses, as the NMDA receptors at those synapses switch from slow-conducting GLUN2B to fast-conducting GLUN2A subunits (Wyllie et al. 2013).

1.4.3 Non-synaptic roles of SHANK

While the vast majority of the focus has been on the role of SHANKs at the synapse,

SHANKs appear to have numerous roles in other neuronal functions and even in non-neuronal cell types. It has been recently shown that all SHANKs appear to be expressed not only in dendrites and synapses, but also in axons (Halbedl et al. 2016). SHANK3 undergoes activity- dependent synaptonuclear shuttling (S. Grabrucker et al. 2014). SHANK2 is more ubiquitously expressed than either SHANK1 and SHANK3, is enriched in epithelial cells, particularly those of the kidney and liver (Lim 1999; Han et al. 2006), and mutations at the SHANK2 locus have been associated with several squamous cancers (Freier et al. 2006; Carneiro et al. 2008; Sugahara et al. 2011; Ying et al. 2012; Qin et al. 2016). While the particulars of non-neuronal findings are outside of the scope of this work, collectively these observations suggest the existence of extensive pleitropy for SHANK2 function. SHANK2 knockdown results in excessive dendrite branching (Berkel et al. 2012). Conversely, overexpression of either SHANK2 or SHANK3 suppresses dendrite branching, suggesting that SHANK2 and SHANK3 have shared function and act as suppressors of dendrite branching (Quitsch et al. 2005). However, given that dendrite outgrowth precedes synaptogenesis and SHANK2 expression precedes SHANK3, SHANK2 is likely to play a more important role in dendrite growth, while SHANK3 could be more important for synapse formation.

1.4.4 Mouse models of SHANK2-associated ASD exhibit inconsistent phenotypes

Predicting how ASD-associated mutations in SHANK2 affect brain function is very challenging because SHANK2 governs multiple processes that impact development of neuronal connectivity, such as formation of dendrites and synapses. SHANK2 knockout models have yet to provide clarity on this question. Three different mice have been generated, one with a deletion of exons 15-16 (SHANK2e15-16) (Won et al. 2012), one with exon 16 (SHANK2e16) (Schmeisser et al. 2012) and another with a cerebellum-specific deletion of exon 16 (L7-SHANK2e16) (Peter et al. 2016). All targeting strategies are designed to terminate transcription and result in loss of protein. However, none cause ASD-like social deficit phenotypes, such as a decrease in time spent with a novel mice rather than objects, in a heterozygous background, suggesting that humans may be particularly sensitive to SHANK2 haploinsufficiency.

Cerebellum-specific homozygous knockout of SHANK2 causes autism-like behaviour in mice and provides the first experimental evidence of cerebellar involvement in ASD pathophysiology (Peter et al. 2016). However, non cell type-specific knockouts of SHANK2 that affect the whole brain result in somewhat discordant findings. These homozygous knockout mice exhibit autistic behaviours but have inconsistent, and in some cases, opposite alterations in brain functions. The SHANK2e16 exhibits an increase in NMDA signaling and synaptic plasticity where as the SHANK2e15-16 mouse exhibits a decrease. With respect to neuronal connectivity, the SHANK2e16 shows a mild 10% decrease in frequency of miniature excitatory postsynaptic currents (mEPSC), whereas the SHANKe15-16 mouse shows no difference (Schmeisser et al.

2012; Won et al. 2012). It is possible that genetic background of cases with ASD and SHANK2 mutations or differences in neuronal development between mice and humans can account for these discrepancies (Y.-H. Jiang & Ehlers 2013; Sala et al. 2015). Experiments conducted on

31 human SHANK2-null or SHANK2-heterozygous neurons may therefore elucidate the effects of

ASD-associated SHANK2 mutations on human neuronal connectivity.

1.5 IPSC in NDD modeling & Drug Discovery

The work in section 1.5 partly reflects contributions to Kim, D.-S., Ross, P. J., Zaslavsky, K.,

& Ellis, J. (2014). Optimizing neuronal differentiation from induced pluripotent stem cells to model ASD. Frontiers in Cellular Neuroscience, 8, 109. http://doi.org/10.3389/fncel.2014.00109

Recently, it has become possible to generate live human neurons from individual people by reprogramming their somatic cells to induced pluripotent stem cells (IPSC). This is accomplished by exposing an individual’s dermal, dental pulp, hair or blood cells grown in vitro to high levels of pluripotency-associated transcription factors OCT4, SOX2, KLF4, CMYC

(Takahashi et al. 2007). While initial methods employed integrating viruses (I.e., retrovirus or lentivirus) to overexpress these proteins, recently non-integrating vectors, such as episomes and

Sendai Virus, have become more widely used (Schlaeger et al. 2015). The resulting IPSCs can self-renew indefinitely in culture and theoretically generate any cell type in the body. Critically, this allows the generation of human neurons in cases where brain tissue is not accessible, such as living children with neurodevelopmental disorders. The development of these neurons can be followed over time and their transcriptome and neuronal connectivity phenotypes can be interrogated (Figure 8). In turn, genome editing approaches, such as CRISPR/Cas9, TALEN, or

ZFN (Gaj et al. 2013), can be used to test effects of suspected disease-causing variants on any observed phenotypes. However, discovery of reproducible phenotypes remains a challenge for most IPSC-based phenotyping approaches for NDDs owing to small sample sizes and line-to-line heterogeneity, and is a major barrier towards mechanistic investigation and drug-screening.

Figure 8. Approach to modeling NDDs using IPSC-derived neurons.

1.5.1 Gene editing in human iPSC

An ideal control for disease modeling using iPSCs is the generation of isogenic controls.

While random X-chromosome inactivation allows isolation of isogenic pairs of iPSC lines for X- linked disorders (Cheung et al., 2011), the development of targeted DNA endonucleases expands this possibility to autosomal mutations. Two types of technology exist that differ by means with which they target DNA sequences: endonucleases bound to programmable DNA binding domains, and RNA-guided endonucleases. The former includes zinc-finger nucleases (ZFNs)

(Cathomen & Keith Joung 2008) and transactivator-like effector nucleases (TALENs) (Cermak et al. 2011), while the latter rely on recently-discovered RNA-guided endonucleases (e.g., Cas9,

Cpf1, C2c1, C2c3) of bacterial clustered regular interspaced short palindromic repeat (CRISPR)

34 adaptive immunity systems (Wright et al. 2016). While ZFNs and TALENs have very high levels of on-target specificity (Pattanayak et al. 2011; Guilinger et al. 2014), they are complex to design and assemble, as they require ligating multiple DNA fragments into a specific sequence to build the DNA binding domain. Owing to this, the uptake of ZFNs and TALENs in academic research has been slow. On other hand, the use of CRISPR/Cas9 for genome editing is rapidly becoming a standard laboratory procedure, as researchers only need to design an oligonucleotide complementary to the target DNA sequence and clone it into an appropriate vector. Multiple genome-wide CRISPR/Cas9 screens have been published to date (Shalem et al. 2015), a feat that is virtually impossible with ZFNs or TALENs.

CRISPR-Cas systems are found in roughly 50% of bacteria and 90% of archaea. They are adaptive immune systems in that they retain records of past infections in order to mount an efficient defense upon reinfection. To accomplish this, they store sequences specific to past viral infections in a CRISPR array of spacer sequences separated by of 20-50 bp-long direct repeats.

The CRISPR locus is transcribed and processed to generate CRISPR RNA (crRNA), which acts in tandem with a trans-activating crRNA (tracrRNA) and guides an effector complex to digest any DNA sequences to which it is complementary. Class 1 CRISPR systems use a multi-protein effector complex, and Class 2 systems use a single endonuclease as an effector. Class 2 systems have been subsequently adapted for use in the laboratory for gene editing in eukaryotic cells, with Cas9 being the most popular, likely because it was the first to be applied for this purpose.

To ease application, tracrRNA and crRNA are fused together to make a single guide RNA

(gRNA) in most CRISPR/Cas9 cloning vectors. Using gRNAs, Cas9 will create double strand breaks (DSBs) at specific target sites, which are then corrected by endogenous DNA repair mechanisms, such as non-homologous end joining (NHEJ) and homology-directed repair (HDR) in the presence of a homologous template DNA. NHEJ is error-prone and can generate small

35 insertions or deletions, causing frameshifts that can cause the creation of premature termination codons, knocking out the target protein. On the other hand, HDR allows for precise genome modification at specific sites in the genome(Wright et al. 2016).

However, the application of the CRISPR/Cas9 system to human IPSCs is complicated for several reasons. Variable transfection efficiency of IPSCs, low frequency of HDR, and poor clonal survival of human IPSCs make it challenging to isolate pure gene-edited lines.

Transfection issues are usually solved by empiric determination of optimal transfection conditions on a line-by-line basis, and typically involve electroporation. To facilitate HDR,

NHEJ can be inhibited using SCR7 (Chu et al. 2015). To work around the issue of poor clonal survivability, Miyaoka et al. adapted a method from yeast genetics, sibling selection, to facilitate selection-free recovery of gene edited clones. The strategy relies on using digital droplet PCR

(ddPCR), a very sensitive method of absolute quantification of DNA, to detect genetically edited clones. Clones from a transfection are arrayed in a 96 well-plate, and ddPCR is used to find the well with the highest proportion of modified alleles. That well is then expanded into another 96 well-plate and the procedure is repeated. After several rounds of enrichment, a well with 100% of gene edited cells can be isolated. While this procedure is highly versatile and sensitive, it is possible that it may not be sensitive enough to detect genetically-modified cells if the gene editing efficiency is low. In these cases, standard and lengthy drug-selection based methods may yield clones with desired genetic modifications.

1.5.2 Three Broad Types of Neuronal Differentiation Protocols

Figure 9. Three general approaches for IPSC differentiation into neuronal tissue. IPSCs can be infected with a NGN2 or NEUROD1 expressing lentivirus, which will directly convert them to neurons. IPSCs can be subjected to directed differentiation. This can proceed from an

IPSC monolayer or through an embryoid body patterning step. From this point, neural precursor

37 cells are isolated and can be differentiated to a 2D neuronal culture. Alternatively, embryoid bodies are induced to form neuroectoderm tissue, which is then immersed in a matrigel droplet and suspended in a spinning bioreactor to generate 3D cerebral organoids over several months.

Multiple neuronal differentiation protocols have been developed to allow generation of neurons of different subtypes and increase efficiency. Three major categories exist: direct conversions, 2D methods that may include a transient 3D patterning step, and 3D organoids

(Figure 9).

Direct conversion methods rely on overexpression of transcription factors NGN2 or

NEUROD1. They can generate homogenous neuronal populations very quickly and are technically easy to execute. Furthermore, the resulting neuronal cultures can be used for synaptic assays as early as 3 weeks of age. The high degree of similarity between cultures grown in separate wells generally allows for well-controlled comparisons of single-cell connectivity.

However, while advantageous from an experimental design perspective, the possible compression of normal neuronal development may cause the experimenter to miss critical developmentally-important mechanisms of disease. In addition, these protocols are inflexible in generating different neuronal subtypes, which may hamper attempts to model phenotypes of mutations in genes that are not expressed in these neuronal populations (Yingsha Zhang et al.

2013). As such, direct conversions are at risk of failing to generate disease-relevant cell types, and their speed can compress normal neuronal development, limiting the search space for detecting disease-associated phenotypes.

There are several 3D-culture based approaches. The 3D organoids developed by Lancaster et al. are remarkable in their ability to capture important aspects of neocortical development. For example, these organoids harbor subregion-specific neuronal progenitors, as well as abundant oblique and vertically oriented radial glia more commonly found in human tissues. In addition, in these 3D organoids, neuronal development occurs in a layered inside-out pattern reminsceent of in vivo brain development. Gene expression analyses show that organoids are most similar to the developing fetal brain, with the oldest organoids resembling late 2nd / early 3rd trimester brain.

They have shown utility in modeling gross phenotypes like microcephaly (Lancaster et al. 2014) and have been applied to modeling of idiopathic ASD, wherein they revealed increased production of inhibitory GABAergic neurons relative to controls (Mariani et al. 2015). However, these types of organoids are technically challenging to maintain, require a long investment of time, and have yet to convincingly display synaptic activity. An alternative 3D approach was advanced by Pasca et al. in 2015 that generates 3D spheroids. Like the 3D organoids, the spheroids also develop in an inside-out layered cytoarchitecture that resembles the mammalian brain. They also produce non-reactive glial cells. However, they do not make GABAergic neurons, limiting the range of possible phenotypes that can be assayed. Like the organoids, the spheroids also require a long maturation time to generate synaptically-active neurons: Pasca et al. reported NMDA- and AMPA-dependent synaptic currents after 180 days culture (A. M. Paşca et al. 2015; Hartley & Brennand 2016). Given the long developmental time course and the technical expertise required for the generation, 3D organoids are perhaps best suited to investigating pathways important for early brain development. However, the long developmental time-frame and the accompanying technical requirements to maintain these 3D structures may complicate single-cell neuronal connectivity assays. It is unclear how network connectivity develops in these

39 structures, and the extent of variability between organoids generated from the same line and between different lines is unknown.

The oldest and most widely used category of methods for neuronal generation and phenotyping generate 2D cultures that may be generated through a 3D patterning step. These typically begin with dual SMAD inhibition to decrease the likelihood of generation non-neuronal cell fates, and generally allow good flexibility with respect to subregional specification. For example, early use of Wnt inhibition can be used to dorsalize the resulting culture, whereas sonic hedgehog agonists can serve to ventralize it (D.-S. Kim et al. 2014). These approaches also offer some flexibility in that they allow for the generation of a renewable pool of neural precursor cells

(NPCs) which can be maintained and expanded for use in multiple experiments. Differentiating these NPCs to neurons typically results in excitable neurons by 2 weeks, and some AMPA- and

NMDA-dependent synaptic activity by 6 weeks (Gupta et al. 2013). Differentiating these cultures to an older age, such as 9- or 12-weeks can result in more robust synaptic activity

(Livesey et al. 2014). The gene expression profile of these neurons typically matches that of the developing fetal brain (Stein et al. 2014). However, there is substantial variability between different differentiations of the same line and among different IPSC lines. The resulting cultures may differ in density and cell composition. For example, Kirwan et al. assayed the development of network connectivity in 2D cultures over time and found that different cultures do not develop network activity synchronously, with some developing several weeks earlier than others (Kirwan et al. 2015). This feature may introduce significant error in comparisons of synaptic connectivity between disparate neuronal cultures grown using this protocol. Like organoids, these 2D methods also generate a plethora of different neuronal subtypes and assaying disease-relevant cells also remains a challenge (Table 1).

Approach Yield Technical Age at which Cell-type Maturity

Variability synaptic heterogeneity relative to

phenotypes human brain

can be

assayed

Direct Minimal (99%

Conversion High Low 3 weeks excitatory Unknown

layer II/III)

3D organoids Similar to Fetal brain human brain, High / Unknown Low but may be unknown 3D spheroids spatially 180 days Fetal brain regulated

2D culture, Large, but 6-12 weeks dual SMAD Medium to likely High (the older, the Fetal brain inhibition high missingseveral better) cell types

Table 1. Summary of neuronal differentiation approaches.

1.5.3 Recent Progress in IPSC models of NDDs

The choice of protocol heavily depends on the research question and the expected effect size.

Initial studies focused on severe ASD-associated disorders, such as Rett Syndrome (RTT),

Timothy Syndrome (TS) and PMDS. Using a 2D differentiation protocol with a transient 3D embryoid body step, multiple groups were able to report decreases in cell size, dendritic arborization, many alterations in electrophysiological properties, as well as profound decreases in synaptic connectivity and network activity in RTT (Marchetto et al. 2010; Cheung et al. 2011;

K.-Y. Kim et al. 2011; Ananiev et al. 2011). Transcriptome analyses of RTT revealed global downregulation of transcription that was accompanied by decreases in translation that could be rescued by mTOR activation through IGF-1 or BDNF treatment (Y. Li et al. 2013). In these neurons, MECP2 protein level appears critical for appropriate neuronal development. Cheung et al. were able to rescue RTT phenotypes only with a lentiviral vector in which MECP2 expression was driven by its endogenous promoter MeP. In turn, Nageshappa et al. showed the MECP2 duplication neurons have increased synaptogenesis, dendrite growth and altered neuronal network development. Nageshappa et al. then performed a screen using multi-electrode arrays and found that a histone deactylase inhibitor restored network phenotypes in MECP2 duplication neurons (Nageshappa et al. 2015).

Pasca et al. and Krey et al. reported several early developmental abnormalities in TS neurons, demonstrating the potential of IPSC-derived neuronal cultures to capture diverse phenotypes that may be associated with NDDs. In particular, neuronal cultures differentiated from TS IPSCs had altered expression of cortical patterning markers, increased expression of tyrosine hydroxylase, indicating an increased abundance of dopaminergic neurons. Neurons also exhibited a widening of action potentials and a sustained rise of intracellular calcium following depolarization, consistent with a decrease in calcium- and voltage-dependent inactivation of the Cav1.2 calcium

42 channel encoded by the CACNA1C gene mutated in TS. Because intracellular calcium regulates activity-dependent signaling, TS neurons also displayed several differentially-expressed genes, some of which were involved in transduction of intracellular calcium signaling (S. P. Paşca,

Portmann, Voineagu, Yazawa, Shcheglovitov, Paşca, Cord, Palmer, Chikahisa, Nishino,

Bernstein, Hallmayer, Geschwind & Dolmetsch 2011b). Krey et al. tested whether these alterations in neuronal excitability and in intracellular calcium signaling could also alter neuronal development. While a reasonable prediction is that neurons with a greater intracellular [Ca2+] would show an exaggerated response to activity, TS neurons exhibited an activity-dependent retraction of dendrites, whereas control neurons increased in length by approximately 20%. This occurred even despite an absence of ion flow through the channel, suggesting that a depolarization-associated conformational change in Cav1.2 channel was responsible. Indeed, Krey et al. were able to show that this conformational change activated RhoC signaling, which caused the retraction in dendrites. These changes in excitability, intracellular [Ca2+] and dendrite development can significantly affect the establishment of proper neuronal connectivity (Krey et al. 2013).

Shcheglovitov et al. used a monolayer differentiation protocol and showed a 50% decrease in excitatory synaptic activity in PMDS (Shcheglovitov et al. 2013). Remarkably, IGF-1 treatment may be able to rescue many of the phenotypes in both RTT and PMDS neurons, and is currently being used in clinical trials on RTT patients (Khwaja et al. 2014). Bidinosti et al. later confirmed the importance of mTOR signaling for synaptic deficits in SHANK3-deficient mice and PMDS neurons, by showing the CLK2 inhibition was able to increase mTOR activity and synaptic activity (Bidinosti et al. 2016).

Mariani et al. used 3D organoids to attempt to detect alterations in brain development of patients with nonsyndromic idiopathic ASD. Using a combination of RNA sequencing and knockdown experiments, they found a dysregulation of FOXG1-dependent transcription which increased the number of inhibitory neurons and the density of inhibitory synapses (Mariani et al.

2015). While such findings are exciting given the hypothesized excitation-inhibition imbalance in ASD, the 3D organoid system did not permit to examine network phenotypes owing to relative immaturity of most neuronal cells. For their part, direct conversions have been predominantly used to test the effect of ASD-associated mutations caused by CRISPR mutagenesis in hESC, and have generally shown decreases in excitatory synaptic signaling. Yi et al. validated and extended Shcheglovitov et al.’s findings by demonstrating impaired synaptic function and reduced IH currents in SHANK3 mutant neurons (Yi, Danko, Botelho, Patzke, Pak, Wernig &

Südhof 2016a). Pak et al. showed that deletions in NRXN1, another high-confidence ASD risk gene, also impair excitatory synaptic transmission by impairing presynaptic release (Pak et al.

2015). Direct conversions also have potential to elucidate transcriptomic alterations by virtue of generating mostly homogenous neuronal cultures. Lalli et al. used this system on WBS patient cells to show that BAZ1B was responsible for approximately 35% of transcriptome alterations in

WBS neurons (Lalli et al. 2016).

1.5.4 Challenges in IPSC models of NDDs.

In the absence of knowledge about which neurons are responsible for NDD pathophysiology, and at what timepoint possible phenotypes manifest, 2D differentiation methods offer the most versatility. The widely used 2D differentiation process begins with IPSCs either cultured as a monolayer or as free-floating 3D cellular aggregates (embryoid bodies). In this state, cells are

44 directed to neuroectoderm fates by dual SMAD inhibition to prevent endoderm and mesoderm differentiation. Further modulation of signaling pathways can promote subregional specification.

Sonic hedgehog (SHH) agonists can induce ventralization and Wnt inhibitors can bias cells towards dorsal forebrain fates. The resulting cultures are rich in tripotent neural precursor cells

(NPCs) that appear as neural rosettes when seeded on a flat surface. They can differentiate along a typical embryonic timeline, as they first generate neurons, followed by astrocytes and oligodendrocytes in a temporally controlled manner (D.-S. Kim et al. 2014). Critically, neurons generated in this way appear to acquire network properties in a manner that parallels in vivo cortical development. First, synchronous oscillatory networks develop in an ordered progression over several weeks, likely supporting development of synaptic connectivity. This phase is followed by a decrease in oscillations that culminates in the rise of non-synchronous, yet ordered activity patterns (Kirwan et al. 2015).

However, there are several critical issues that complicate the use of this approach for transcriptome and connectivity phenotyping that must be addressed. Line-to-line variability in neuronal differentiation can complicate transcriptome analyses due to the presence of unwanted cells types, such as astrocytes, oligodendrocytes, residual NPCs, and cells from non-neuronal lineage whose presence, identity, and proportion is unknown and variable (Sandoe & Eggan

2013). This kind of variability could be potentially addressed by exploiting cell-surface markers, such as CD44, CD84, and CD184 for cell sorting (Yuan et al. 2011; Woodard et al. 2014). The technical difficulty in working with multiple lines also limits sample sizes of studies, and combined with line-to-line variability, raises the question of what can be considered a representative ‘control’ or ‘wild-type’ line. A study by the Geschwind and Gage labs shows that neurotypic IPSC-derived neurons follow a fetal developmental transcriptome program in vitro and can be used as a reference transcriptome to benchmark control lines generated in other

45 studies (Stein et al. 2014). One or a combination of these steps are essential for characterizing neuronal transcriptome changes in WBS and ASD, particularly for cases where the effect of mutations is difficult to predict, such as for PTCHD1/PTCHD1AS mutations.

Perhaps most challenging is detecting reproducible differences in neuronal connectivity, a feature critical for modeling ASD. Neoteny in human neurons requires long periods of maturation (typically, 9 weeks or more) to facilitate synapse formation (Geschwind & Rakic

2013; Livesey et al. 2014; Otani et al. 2016). Over this time, initial variability between different lines can cause dramatic changes in cellular environment between different cultures. The experimenter has limited control over how NPCs differentiate into neurons and exit the cell cycle, and this often occurs asynchronously when multiple lines are grown in parallel. Because of the presence of other cell types in these cultures that can accumulate to different proportions by the 9 week time point, each of which can exert unknown effects on neuronal development, neurons under comparison may develop in drastically different synaptogenic environments, which will affect measurements of their morphology and connectivity. This is further complicated by the need to phenotype disease-relevant neurons, as the gene of interest may not be ubiqituously expressed among differentiated neurons and present at different proportions among different neuronal cultures. Taken together, these factors can vastly increase noise in connectivity measurements and obscure biological differences that are not sufficiently large.

Because the effect size from ASD mutations on connectivity is unknown, it is imperative to design experiments that reduce variability and increase statistical power. This is particularly important for proper phenotyping of cases with SHANK2 mutations, given SHANK2’s possible dual role in regulating dendrite morphology and synapse density, both of which can be influenced by the surrounding cellular environment.

1.6 1.6 Rationale

Our understanding of pathophysiological mechanisms of NDDs severely lags behind our knowledge of the underlying genetic causes. Given the finite availability of research resources, focusing efforts on disorders of great severity, prevalence or combination of the two, such as

WBS and ASD, can maximize the potential to advance knowledge of disease mechanisms and benefit to patients. Functions of genes mutated in WBS and ASD converge on pathways involved in transcriptional control and synaptic connectivity. With the recent development of

IPSC technology, live human neurons can be generated from individuals with these disorders to test specific questions raised by genetic findings.

The first hypothesis that is critical to test is whether neurons from children with WBS and

ASD exhibit transcriptomic changes. Proper temporal coordination of transcription is essential to proper development of neurons and, ultimately, brain wiring. WBS deletions affect multiple transcription factors, and one of the most frequently mutated loci in ASD is the

PTCHD1/PTHCD1AS locus, which may encode products that regulate transcription. Generation of neurons specific to these cases can elucidate potential alterations that could be underlying brain dysfunction in these disorders. The first part of the second chapter of the thesis examines the transcriptome of WBS neurons compared to a WT control. To increase certainty in the transcriptome analysis, the WT control is compared to a published resource of wild-type IPSC- derived neurons. The second part of the chapter uses a larger sample size and employs a novel protocol to purify neurons to examine the transcriptome of ASD cases with

PTCHD1/PTCHD1AS mutations. Because WBS and ASD lie at the opposite ends of the social communication spectrum, I compared the transcriptomic changes found in these studies to search for any disparities in gene expression.

The second hypothesis to test is whether neurons from children with ASD and SHANK2 mutations exhibit altered synaptic connectivity, as identification of reproducible phenotypes can permit downstream mechanistic investigation and drug screening. To increase confidence in my results, the first aim is to use gene editing with CRISPR/Cas9 to derive isogenic knockouts of

SHANK2 in cells of unrelated controls. However, reproducible synaptic connectivity assays are extremely challenging in human iPSC-derived neurons owing to multiple factors. Therefore, the second aim of the third chapter is to develop a reproducible and high-powered assay to permit measuring connectivity of stem cell derived neurons. Ideally, the method must allow to directly characterize connectivity by quantifying all synapses of a given neuron, as well as simultaneously reconstruct the dendritic morphology of the same neuron. Using this system, the third aim is to characterize connectivity in SHANK2 mutant human neurons.

Taken together, the work I describe in the following chapters represents advances that will help bridge the knowledge gap between NDD genetics and underlying biology.

Chapter 2 Transcriptional Alterations in WBS and ASD

The work described in this chapter has contributed to one publication in print:

Khattak, S., Brimble, E., Zhang, W., Zaslavsky, K., Strong, E., Ross, P. J., et al. (2015). Human induced pluripotent stem cell derived neurons as a model for Williams-Beuren syndrome.

Molecular Brain, 8(1), 77. http://doi.org/10.1186/s13041-015-0168-0

Contributions: SK, EB, WZ, SM, MWS, LRO and JE conceived the study and designed the experiments. SK performed neural differentiations and immunoassays (not reported in the

Chapter), SK and JH performed single-cell qRT-PCR assays (not reported in the Chapter), WZ performed electrophysiological analyses (not reported in the Chapter), EB and ES performed qRT-PCR analyses, KZ performed microarray analyses, PJR generated WT IPSC.

The work in this chapter also contributed to an additional manuscript in preparation:

Ross, P.J., Zhang, W., Zaslavsky, K., Deneault, E., Mok, R.S.F., D’Abate, L., Rodrigues, D.C.,

Yuen, R.K.C., Wei, W., Piekna, A., Pasceri, P., Salter, M.W., Scherer, S.W., and Ellis, J.

Autism-associated deletions of the lncRNA PTCHD1-AS result in synaptic dysfunction in human neurons.

49 50

Contributions: PJR, SSW, MWS, JE conceived the study and designed the experiments. WZ performed electrophysiological analyses, KZ performed microarray analyses, ED performed gene editing (not shown in Chapter), RSFM performed morphological anlaysis (not shown in

Chapter), LA and RKCY perfomed genetics analyses on novel PTCHD1AS patients, DCR developed the MACS purification method, PJR, WW, AP, PP performed neuronal differentiations were involved in cell culture.

In summary, for both studies, I analyzed microarrays comparing gene expression between neurons from typical controls and affected cases and interpreted the findings.

2.1 Abstract

Genetics of WBS and ASD strongly implicate transcriptional dysregulation as a mechanism of disease. Differentiation of neurons from stem cells derived from individuals with these disorders allows for phenotypic investigation. To determine the possible transcriptional alterations underlying these changes, I analyzed microarray data comparing neurons from affected individuals to controls in two separate experiments. WBS neurons had lowered expression of several voltage-gated K+ ion channel genes, as well as downregulation of gene sets related to synaptic function and the extracellular matrix. These transcriptional changes may underlie the observed electrophysiological alterations in WBS neurons, such as impaired action potential firing and voltage-gated K+ ion channel function. Neurons from ASD individuals with mutations in the PTCHD1/PTCHD1AS locus had downregulation of RNA export, RNA splicing, cellular respiration, and an upregulation of extracellular matrix gene sets. While these changes make it difficult to predict specific functional abnormalities, electrophysiological analysis of affected neurons revealed impaired synapse and NMDA receptor function. Taken together, these findings highlight the diversity of neuronal alterations in these NDDs and the possible mechanisms underlying them.

2.2 Brief Introduction and Rationale

The complex genetic architecture and heterogeneity of NDDs implicate multiple underlying mechanisms. Both WBS and ASD feature mutations in genes regulating transcription(Pober 2010; De Rubeis et al. 2014). Children with WBS have increased sociability, intellectual disability and cardiovascular problems caused by a heterozygous loss of 26-28 genes at 7q11.23, several of which could regulate transcription (MLXIPL, GTF2I, GTF2IRD1, BAZ1B,

FZD9). On the other hand, children with ASD have impairments in socialization and increased restrictive and repetitive behavior. While approximately one third of the 100+ loci implicated in

ASD tend to harbor genes that may have roles in regulating transcription, the

PTCHD1/PTCHD1AS locus is one of the most highly prevalent and accounts for 0.5% of cases.

The discovery of several deletions affecting only PTCHD1AS raise the possibility that this lncRNA may be responsible for regulating transcription, either by cis or trans effects (Noor et al.

2010; Kung et al. 2013).

To investigate transcriptional alterations in WBS, I integrated microarray gene expression data comparing one control with 3 WBS lines with publicly available control iPSC-derived neuron data, as closely matched by differentiation protocol as possible and run on the same microarray platform. Using unsupervised hierarchical clustering, I found that our control was representative of others and discovered a downregulation of voltage-gated potassium channels, consistent with action potential deficits in WBS neurons observed in our lab. In addition, multiple gene sets affecting synaptic functions were downregulated.

To investigate transcriptional alterations in ASD, I focused on children with mutations at the locus containing patched domain containing protein 1 (PTCHD1) and the nearby divergently- transcribed lncRNA PTCHD1AS. The PTCHD1/PTCHD1AS locus is strongly associated with

ASD, with deletions or putative loss-of-function mutations affecting either PTCHD1,

PTCHD1AS, or both. For our study, one child had a deletion of both PTCHD1 and PTCHD1AS and the other had a deletion of PTCHD1AS exon 3 that also encompasses a nearby gene DDX53.

Electrophysiological analyses revealed decreases in mEPSC frequency in the probands, suggesting impairments in synapse function. To elucidate potential mechanisms underlying dysfunction, we performed microarray analyses. To address the issue of sample heterogeneity following neuronal differentiation, our lab developed a neuronal purification approach based on magnetic activated cell sorting. In a subsequent microarray analysis on purified 4-week old neurons, I found that extracellular matrix genes were upregulated, while RNA export, RNA splicing, and cellular respiration genes were downregulated in both cases, suggesting a specific function of PTCHD1AS, an uncharacterized lncRNA.

2.3 Summary of electrophysiological phenotypes in neurons of NDD cases

2.3.1 WBS

We generated 3 iPSC lines from one WBS individual and 1 line from a female WT individual (Table 2). Using directed differentiation, we generated a mixed population of cortical neurons within 6-8 weeks. Single-cell qRT-PCR analyses revealed that 80-90% of cells expressed neuronal markers (DCX, NCAM1, MAP2), a mix of upper and lower cortical neuron markers, as well as markers of glutamatergic and GABAergic neurons. No statistically significant differences in marker expression were found among the 4 lines. Electrophysiological analyses revealed that three WBS lines, compared to the WT line and an additional 2 lines from another unrelated individual had marked differences in action potential properties. Specifically,

54 neuronal repolarization was impaired, as the decay of action potential was prolonged and K+ voltage-gated channel currents were smaller. Additionally, input resistance was higher in WBS neurons. However, unitary neurotransmission as measured by miniature excitatory postsynaptic currents (mEPSCs) was not altered, suggesting that the deficit is confined to intrinsic membrane properties of WBS neurons.

2.3.2 PTCHD1/PTCHD1AS ASD

We identified multiple individuals with ASD harboring mutations at the

PTCHD1/PTCHD1AS locus. A majority of the individuals with the ASD diagnosis had mutations encompassing the PTCHD1AS locus, suggesting PTCHD1AS mutations could be causally linked to ASD. Therefore, we generated three iPSC lines from two affected children, one with a deletion of both PTCHD1 and PTCHD1AS (Proband 1) and the other with the PTCHD1AS exon 3 deletion that also affects a nearby gene DDX53 (Proband 2), as well as three lines from the maternal control of the child with the PTCHD1/PTCHD1AS deletion and one line from an unaffected male control (Table 2). We differentiated these iPSCs into neurons using a protocol that generates a mixture of cortical neurons (Brennand et al. 2011). Neurons from proband 1 had no expression of either PTCHD1 or PTCHD1AS. Neurons from the proband 2 had normal expression of PTCHD1 and altered splicing of PTCHD1AS. Although expression of DDX53 was absent in the cells of both affected individuals, it was present in the controls. We showed that 12 week-old PTCHD1/PTCHD1AS mutant neurons have lower frequency of miniature excitatory postsynaptic currents (mEPSC) and lower amplitude of NMDA currents relative to controls, suggesting that PTCHD1/PTCHD1AS leads to impaired synaptic function. The coincident impairment in NMDA signaling and frequency of mEPSCs suggests impairments in synaptic

55 development of neuronal maturation. In addition, we used CRISPR/Cas9n to knock in a polyadenylation signal into exon 3 of PTCHD1AS in the cells of the unrelated male control in order to terminate transcription through the locus and simulate PTCHD1AS exon 3 deletion.

These neurons recapitulated the mEPSC frequency decrease, but also exhibited a decrease in mEPSC amplitude.

Individual IPSC Lines

WBS experiments

Control WT

WBS A100

B100

I100

PTCHD1/PTCHD1AS experiments

CTRL-M (Unrelated male CTRL-M #2 control)

CTRL-F (Maternal control of CTRL-F #2.6 PTCHD1/PTCHD1AS case) CTRL-F # 2.14

CTRL-F #2.93

ASD case with Prb1 #2.27 PTCHD1/PTCHD1AS Prb1 #2.22 deletion (Proband 1, male)

ASD case with PTCHD1AS Prb2 #5 exon 3 deletion (Proband 2, Prb2 #15 male) Prb2 #36

Table 2. Summary of IPSC lines used for transcriptome analyses

2.4 Results

2.4.1 Reduction in expression of potassium channel, ECM, synapse genes in WBS

2.4.1.1 Transcriptome of control neurons generated in our laboratory is representative of those from another lab

To investigate potential mechanisms underlying the electrophysiological alterations in WBS,

I analyzed microarray data from 6 week-old neuronal cultures from the three WBS lines compared to the one WT control hybridized to the Illumina HumanHT12-v4 platform. Because we had only one control cell line, I needed to determine whether its transcriptome was representative of other WT lines relative to our WBS samples. To accomplish this, I integrated our microarray data with that produced on the same platform from iPSC-derived neurons differentiated in a similar manner by the Gage lab (Stein et al. 2014). I used the ComBat algorithm to reduce non-biological batch-to-batch variation (W. E. Johnson et al. 2007) arising from samples being hybridized and processed in different labs. Visual inspection of hierarchical clustering revealed that the WT sample clustered with the WT Gage samples (Figure 10), suggesting that it is representative of the gene expression pattern on WT iPSC-derived neurons.

Alternatively, the effect of WBS is large enough to be observed regardless of the variance among

WT datasets.

Figure 10. Hierarchical clustering of data from our lab integrated with that from the

Gage lab. WT neurons cluster within the Gage lab samples and not with the WBS samples

(A100, B100, I100), indicating WT sample from our lab is representative of other WT samples.

2.4.1.2 Downregulation of voltage-gated K+ channel genes in WBS neurons

Examination of genes in the deleted 7q11.23 region showed that most genes were expressed in iPSC-derived neurons and had lower expression values in WBS samples relative to the control. Those genes that did not meet the expression threshold on the array (Detection P-value

<0.01) were excluded from further analysis. Importantly, follow-up qRT-PCR validation assays on genes from the region confirmed that their abundance was reduced by 50%, indicating that

59 there was no transcriptional compensation from the loss of one allele. Next, I chose 25 neuronally expressed genes that had lower, similar or higher expression values in WBS relative to the control sample for validation by qRT-PCR in independently collected samples, which was carried out by our collaborators in the Osborne Lab (Figure 11). All selected genes exhibited fold changes in the same direction as that predicted by the microarray.

Figure 11. qRT-PCR validation of 25 neuronally expressed genes. n=3 independent samples.

To determine whether the electrophysiological alterations in voltage-gated K+ signaling could be correlated with changes in gene expression, I examined expression of genes in the voltage-gated potassium channel complex gene set from Gene Ontology (GO:0008076). I found that the vast majority of genes had lower expression values in WBS samples relative to the

60 control, including qRT-PCR validated reductions in KCNIP4, KCNMB1, KCNMB2, KCNMA1 in addition to WBS-deletion region genes STX1A and SNAP25 (Figure 12).

Figure 12. Changes in expression of top 15 downregulated genes in the voltage-gated potassium channel complex gene set from gene ontology database.

2.4.1.3 Gene set enrichment analysis reveals downregulation of gene sets associated with synapse function.

Because I only had one control sample, no genes were detected as differentially expressed at a false-discovery rate of 0.05 after comparing the samples using linear regression analysis for microarrays with the R package limma (Smyth 2004). Therefore, to determine which functions were perturbed in WBS neurons, I performed a continuous gene set enrichment analysis

(Subramanian et al. 2005) with gene sets from the gene ontology database. We ranked differentially expressed probes by the t-statistic and tested which gene sets were overrepresented

61 among top-ranked (i.e. upregulated in WBS) or bottom- ranked (i.e. downregulated in WBS) genes. Out of 5370 gene sets, 136 were enriched in WBS samples at a false discovery rate of 0.1.

Strikingly, all 136 gene sets had negative enrichment scores, suggesting that they are downregulated and their constituent genes are enriched with those expressed at lower levels in

WBS relative to controls. The top negatively enriched gene sets were ‘neurotransmitter receptor activity’ (GO:0030594) and ‘synapse assembly’ (GO:0007416) (Figure 13).

Figure 13. Top two negatively enriched gene sets with the top ten downregulated genes by fold change in WBS neurons relative to controls. Left: each vertical bar is the position of a gene in the given gene set in the ranked list of differentially expressed genes between WBS and

WT.

To discover functional themes among the 136 gene sets, I visualized their overlap using the Enrichment Map (Merico et al. 2010) plugin in Cytoscape. Here, gene sets are visualized as a network of interconnected nodes based on the number of genes they have in common. The largest connected network was found for gene sets governing synaptic function (46 gene sets total, Figure 14). This analysis suggested that ion channels, presynaptic function and GABA receptor function were particularly affected in WBS neurons. In addition, I found a network of 7 gene sets governing extracellular matrix function (Figure 15), with several gene sets showing particularly strong enrichment, which may be analogous to the known role of ELN insufficiency in the cardiovascular WBS phenotype.

Figure 14. The largest interconnected network (46 out of a total 136 nodes) is comprised of downregulated gene sets governing synaptic function. Enriched synaptic gene sets are highly interconnected. Enrichment Map in Cytoscape was used to visualize overlap between enriched gene sets (FDR <0.1) as a network of interconnected nodes. The size of each node corresponds to

64 the size of the gene set. The node color corresponds to the normalized enrichment score, with a darker green corresponding to stronger negative enrichment (more genes expressed lower in

WBS relative to WT). The size of the edges between nodes corresponds to the number of genes the gene sets share in common.

Figure 15. Interconnected network of gene sets governing extracellular matrix function is downregulated in WBS relative to WT neurons.

2.4.2 Elevated ECM and altered DNA-binding gene expression in purified neurons of PTCHD1/PTCHD1AS cases

2.4.2.1 Use of magnetic activated cell sorting to purify neurons

To determine how disruptions at the PTCHD1/PTCHD1AS locus could lead to disruption of synaptic function, we compared the transcriptome of the PTCHD1/PTCHD1AS deletion and

PTCHD1AS-ex3 deletion neurons to that of an unrelated male control and the maternal control of the PTCHD1/PTCHD1AS case. The experiment contained two technical replicates (a and b) each of: 3 lines of the case with the PTCHD1/PTCHD1AS disruption, 3 lines of the case with the

PTCHD1AS disruption, 3 lines from the maternal control of PTCHD1/PTCHD1AS case, as well

65 as one line from an unrelated male control. Because it was difficult to determine what fraction of differentially expressed gene sets in WBS neurons was due to non-neuronal cell types, for these experiments we developed a protocol to isolate RNA from pure populations of neurons. We depleted neural precursor cells and astrocytes using magnetic activated cell sorting by negative selection with the antibodies CD44 and CD184 (Yuan et al. 2011) at 3 weeks of differentiation, which can remove (Figure 16). Cells recovered for 1 week, after which RNA was harvested.

Figure 16. Overview of the MACS procedure to purify neurons.

2.4.2.2 Quality control and preprocessing of HTA 2.0 arrays

The samples were hybridized to the Affymetrix Human Transcriptome 2.0 (HTA 2.0)

Array at The Centre for Applied Genomics. Following quality control analysis, I excluded the second replicate of the third line (#15) PTCHD1/PTCHD1AS case based on low correlation with probe intensities of other samples, observation of systematic aberration in probe intensities by visual inspection of probe intensities following construction of a probe level model, and significantly increased normalized unscaled standard error in that sample relative to other samples (Figure 17).

Figure 17. Quality control identifies systematic aberrations in replicate b of Proband 2 line

#15. (a) correlation matrix of probe intensities among analyzed samples. (b) probe-level model residuals of Prb2 #15b show multiple systematic aberrations (red boxes). Each point in the matrix represents one probe. Intensity of each point represents deviation from the average expression value of the probe among all samples. (b) normalized unscaled standard error is elevated in Prb 2 #15 relative to all other samples.

Next, I normalized the samples using using RMA normalization and summarized probes at the transcript cluster level. Transcript clusters are collections of probes on the HTA 2.0 array that quantify expression of different transcript isoforms and non-coding RNAs.

2.4.2.3 Alterations in RNA export, RNA splicing, cellular respiration, and extracellular matrix in PTCHD1/PTCHD1AS neurons.

In order to compare expression at the transcript cluster level, I used linear analysis of microarrays package (limma) in R (Smyth 2004). Following this analysis, I found that the neurons from ASD-affected samples clustered separately from the controls by DEGs (Figure

18a).

Very few DEGs were shared in common between the two ASD subjects, indicating that the lncRNA deletions led to detectable changes in a restricted set of potential targets (Figure

18b). Only 18 transcripts from 14 genes exhibited a fold-change of >1.25 and achieved statistical significance (false discovery rate < 0.05) in both ASD subjects. Of these 5 have little or no annotation, 2 encode lncRNAs of unknown function (LOC648987, ZNF667-AS1) and 7 encode proteins. The protein coding genes include four C2H2-type zinc finger proteins with KRAB domains (ZNF300, ZNF471, ZNF667, ZNF677), as well as TPM1, IER3, and IQCB1. Joel Ross validated the microarray data by qRT-PCR. We selected several up- and down-regulated genes that were misregulated in the same direction in both probands and were either statistically significantly different or had potential roles in neuronal function. These qRT-PCR experiments largely validated the fold change values detected in the microarray (Figure 18c).

I checked the microarray data for potential mechanisms for impaired function of excitatory synapses. Expression of GRIN2B – which encodes GLUN2B, the dominant subunits of NMDARs in iPS cell-derived neurons (W.-B. Zhang et al. 2016) – and GRIN2D are unchanged in proband neurons (Figure 6F). Expression of GRIN2A is elevated in neurons from both probands, although qRT-PCR failed to detect a statistically significant difference for proband 2. An increase in GRIN2A expression may suggest that majority of NMDA channels feature fast deactivation, resulting in less total increase in intracellular [Ca2+] and possibly preventing efficient synapse development and maturation.

Figure 18. Differential transcript expression between neurons of control and probands with mutations at the PTCHD1 locus. (a) Hierarchical clustering of differentially expressed transcripts at a FDR <0.05 and fold change > 1.25. (b) Venn diagram of differentially expressed

71 transcripts shows very few were differentially expressed in common between Prb1 and Prb 2 relative to controls. (c) Validation of differentially expressed transcripts in independent samples with qRT-PCR. n=3. *P<0.05 **P<0.01

I observed no differences in neural precursor cell and astrocyte marker expression or neuronal specification markers, suggesting that samples did not differ significantly in cell composition. Because PTCHD1 encodes a transmembrane protein with a putative patched domain, its absence in Proband 1 could affect sonic hedgehog signaling pathway. To test this hypothesis, I analyzed expression of all of the genes in the smoothened signaling pathway gene set in the gene ontology database. The only gene that had a log2(fold change) > 1 in either direction (i.e. a fold change greater than 2) was PTCHD1, which had a log2(fold change) < -1.5

(Figure 19), suggesting that sonic hedgehog signaling is unaltered.

Figure 19. Genes in the smoothened signaling pathway are not differentially expressed in

Proband 1 relative to controls except for PTCHD1.

After performing continuous GSEA, I found that stem cell-derived neurons from both children share a large increase in expression of ECM genes, and decreases in RNA export, RNA splicing. The PTCHD1AS-ex3 case had an additional large cluster of downregulated gene sets important for chromatin regulation and transcription (Figure 20). The vast majority of genes within these gene sets had small (10-20%) but consistent statistically significant differences in gene expression between ASD cases and controls, with very few of those genes passing criteria to be defined as differentially expressed (fold change > 1.25 and false discovery rate < 0.05). A slight, but consistent perturbation across multiple related genes across multiple pathways

73 important for RNA metabolism implies that alternative splicing and developmental patterns of gene expression could be altered.

Figure 20. Gene set enrichment analysis in Proband 1 (a) and Proband 2 (b) reveals that

RNA splicing, RNA export, cellular respiration and extracellular matrix gene sets are differentially expressed in the same direction in both probands relative to controls.

Enrichment Map in Cytoscape was used to visualize overlap between enriched gene sets (FDR

<0.001) as a network of interconnected nodes. The size of each node corresponds to the size of the gene set. The node color corresponds to the normalized enrichment score, with a darker blue corresponding to stronger negative enrichment (more genes expressed lower in Prb relative to

Ctrl) and a darker red corresponding to a stronger positive enrichment in the Prb relative to Ctrl

(more genes expressed higher in Prb relative to Ctrl). The size of the edges between nodes corresponds to the number of genes the gene sets share in common.

2.5 Summary and Discussion

2.5.1 Transcriptional alterations in WBS neurons

My analysis of the microarray data shows that WBS neurons derived from patient iPSCs have dysregulated expression of genes including those encoding many ion channels that lie outside the WBS deletion. The expression profile is consistent with the observed electrophysiological defects with prolonged action potential repolarization resulting from a deficit in voltage- activated K+ currents. The observed alterations in K+ currents and action potential repolarization are predicted to dramatically alter integration, output and action potential- dependent transmission in networks of WBS neurons. Such altered cellular and network functioning may be a critical determinant underlying deficits in visuospatial construction and cognitive ability in patients with WBS.

The gene sets downregulated in WBS neurons extend our understanding of WBS-specific alterations in neuronal gene expression. There is significant overlap among these gene sets with those reported in the gene expression study on WBS iPSCs and NPCs (Adamo et al. 2015).

Common gene sets belong to synapse assembly, synaptic transmission, ECM organization and two networks of gene sets governing heart and renal function that are known to be perturbed in

WBS. The significance of these heart and renal networks to neuronal function is currently unclear and I cannot rule out the possibility that their appearance is due to the presence of contaminating disease-relevant cell types in our differentiations. Novel to WBS neuronal function, we have found a reduction of transcript levels of voltage-gated potassium channel genes (Fig 4b), as well as neurotransmitter receptor genes (Fig 4c). These changes may affect electrophysiological properties, as well as alter neurotransmission in WBS neurons. The prolongation of action potential repolarization in WBS neurons in our experiments provides

75 support for the former. Taken together, our results extend the findings of Adamo et al. to include abnormalities in neuronal gene expression.

Of those transcripts identified as belonging to significantly enriched gene sets affecting synapse function, many encoded ion channels. Using qRT-PCR, we validated the upregulation of glutamate receptor subunit GRIK1 and the downregulation of GABA receptor subunit GABRA3.

Dysregulation of either gene would suggest perturbations in neuronal signaling and function. In addition, I observed downregulation of transcripts for voltage-gated Na+ and K+ channels in

WBS neurons. Downregulation of SCN3B and SCN2A Na+ channel subunits may explain the observed smaller amplitude of action potentials (Figure 2d). The prolonged decay time observed in WBS neurons could be caused by altered function of voltage-gated K+ channels (Figure 2e).

This could be attributed to the observed downregulation of one, or a combination, of KCNIP4

(voltage-gated potassium channel gene), KCNMB1, KCNMB2 and KCNMA1 (calcium-activated potassium channel genes). However, I cannot exclude the possibility that the decrease of voltage- gated K+ current was due to the altered expression of other proteins directly and/or indirectly regulating K+ channels.

Voltage-gated K+ channels are broadly expressed in various types of neurons in the central nervous system and play critical roles in regulating neuronal excitability. Mutations of the proteins mediating the voltage-gated K+ currents can cause cellular dysfunction and diseases, and the channels have been used as therapeutic targets(Wulff et al. 2009). Decreased expression of gene KCTD7, a progressive myoclonus epilepsy gene encoding voltage-gated K+ channel tetramerisation domain containing 7, has been reported in WBS patients (Merla et al. 2006).

Additional reports have indicated that WBS patients suffer from progressive hearing loss of

76 unknown cause (Marler et al. 2010). Mutations in the proteins mediating voltage-gated KCNQ4 channel, a subtype of KV7/ KCNQ/M channels, can lead to progressive hearing loss (Jentsch

2000). Prolongation of action potential repolarization in our present study, due to the decrease in voltage-gated K+ currents in WBS neurons, may perturb temporal and spatial integration for input at the somata, and thus alter the neuronal firing output.

2.5.2 Transcriptional alterations in PTCHD1/PTCHD1AS mutant neurons

Our results suggest that ASD-associated disruptions at the PTCHD1 locus, including those of both the protein and the divergently-transcribed lncRNA PTCHD1AS or just

PTCHD1AS, can cause synaptic dysfunction in human neurons. Specifically, neurons exhibit

~50% reduction in mEPSC frequency and 25-50% decrease in NMDA receptor currents. The decrease in NMDAR currents despite a concurrent increase in GRIN2A steady-state mRNA levels may imply that either post-transcriptional regulation negatively regulates expression of

GRIN2A protein. While a decrease in NMDA-receptor currents may decrease the likelihood that newly formed synapses will be retained, we observed no difference in synapse density or dendrite length in proband neurons. These findings could be explained by alterations in presynaptic function that lower the probability of neurotransmitter release and consequently decrease mEPSC frequency. Critically, these findings, together with novel genetic data strongly implicating PTCHD1AS in ASD, imply that PTCHD1AS may be the causative agent underlying these changes.

To investigate possible underlying mechanisms, we therefore performed micorarrays on purified neurons to determine whether neurons from the two ASD probands shared similar

77 alterations in transcriptome profiles. However, we found very few transcripts differentially expressed (fold change > 1.25 at a false discovery rate <0.05) in common between the two probands, suggesting that mechanisms underlying synapse dysfunction were different between the probands or that transcriptome changes were more subtle. Analysis using a continuous gene set enrichment analysis argues for the latter interpretation as the vast majority of constituent genes in these sets had small but consistent alterations (10-20% in the same direction) in expression.

Such consistent alteration in gene expression across multiple related genes indicates that these changes may be caused by PTCHD1/PTCHD1AS mutations that in sum may perturb cell function to impair synapse function. Four clusters of gene sets were found to be misregulated in the same direction among both probands: RNA splicing (down), RNA export (down), cellular respiration (down), and extracellular matrix (up). Because the generally-matched maternal control of Prb1 comprised the majority of control samples in this study, the additional gene sets discovered in the Prb2 vs. Ctrl comparison may reflect differences in genetic background. Given this assumption, the data could support the notion that PTCHD1AS dysfunction could underlie dysregulated expression of those four gene sets clusters as PTCHD1AS deletion is common to both probands. Critically, it implicates PTCHD1AS exon 3 as necessary for normal synapse function and gene expression, as proband 2 has a deletion affecting PTCHD1AS exon 3 that alters PTCHD1AS alternative splicing, but does not abrogate PTCHD1AS expression.

2.6 Materials and Methods

2.6.1 WBS microarrray analysis

Microarray analysis was done using Illumina HumanHT-12 v4 Beadchip. Three wells of each WT, WBS A100, WBS B100, and WBS I100 differentiated neurons were pooled to extract

RNA. 1 µg of RNA from each iPSC derived neuronal sample; WT, WBS A, WBS B, and WBS I; was submitted to The Centre for Applied Genomics at the Hospital for Sick Children, Toronto, for analysis on the Illumina HT-12 v4 Beadchip array (Illumina, San Diego, CA), which targets over 47,000 probes. Raw intensity values were exported and analyzed using the R limma package. To determine whether gene expression data from our control sample (WT) is representative of other wild-type samples, we integrated our data with expression data produced on the same platform (Illumina HumanHT12-v4) from wild-type iPSC-derived neurons in Stein et al. [16] samples Gage-1, Gage-2, Gage- 3, Gage-4, Gage-5, Gage-6. Supplementary data containing background-corrected unnormalized probe intensity values was downloaded from the gene expression omnibus (GEO Accession: GSE57595). These data was merged with unnormalized, background-corrected (nec function in limma [36]) values from our dataset. The merged dataset was log2-transformed, quantile normalized and batchcorrected using the R package ComBat.

To compare gene expression between our WT sample and the three WBS samples, the probe IDs were converted to nuIDs to improve annotation consistency. The neqc function [36] was used for background correction and quantile normalization. The data were log2 transformed and probes with detection p -values >0.01 in all samples were removed for downstream analyses

(22134 probes remaining out of 47320). To perform gene set enrichment analysis, all probes were ranked by the t-statistic produced by an empirical Bayes statistics (eBayes) of the resulting

79 linear model (lmfit) of the microarray. We used gene sets from the Gene Ontology: Biological

Process database and filtered out datasets that were overly large (>900 genes) or overly small

(<15 genes). Gene sets enriched at a false discovery rate (FDR) of <0.1 were used to construct an

Enrichment Map [25] in Cytoscape [37] to visualize gene set interconnectedness and identify perturbed cellular functions.

2.6.2 Microarray analysis of PTCHD1/PTCHD1AS mutant neurons

RNA was extracted with TRIzol as described above and global gene expression was analyzed using the Human Transcriptome Array 2.0 (Affymetrix), with sample processing and array hybridization performed by the The Centre for Applied Genomics at the Hospital for Sick

Children. Data was processed using multiple BioConductor packages in R. Probe-level data was summarized at transcript level using the RMA algorithm (Gautier et al., 2004) in the oligo package.

Before differential expression analysis, 2995 control probe sets that were not annotated to a chromosome were removed. Differential expression of transcript clusters was determined using limma(Smyth 2004). Transcript clusters were defined as differentially expressed if the Benjamini-

Hochberg adjusted p-value was lower than 0.05 and log2(Fold Change) was greater than 1.25. For gene set enrichment analysis, transcript clusters were collapsed such that an individual gene was represented by the transcript cluster with the maximum average expression value (Subramanian et al. 2005) and all transcript clusters not mapping to HUGO gene names were removed. The GSEA analysis was pre-ranked by t-statistic generated by limma. The results of continuous gene set enrichment analysis were visualized as an enrichment map with an overlap coefficient of 0.5

(Merico et al. 2010), showing only those gene sets that met the cut-off false discovery rate of 0.1%.

Clusters containing 4 or fewer nodes were omitted from the enrichment map.

Chapter 3 SparCon reveals hyperconnectivity of SHANK2 mutant human neurons

The work described in this chapter has contributed to a publication in submission:

Zaslavsky K*, Zhang W*, Deneault E, Zhao M, Rodrigues DC, Ross PJ, Romm A, Piekna A,

Wei W, Zhuozhi W, Thompson T, Pasceri P, Scherer W, Salter MW, Ellis J. SparCon assays reveal hyperconnectivity in SHANK2 mutant human neurons.

*Co-first

Contirbutions: KZ, JE, MWS, SSW conceived the project, TT and AP isolated IPSC, KZ and TT characterized IPSC, KZ and ED designed SHANK2 knockout approach, KZ isolated SHANK2 knockout IPSC, ZW perofmred off-target analyses in SHANK2 knockout

IPSC, KZ, WW, AR, AP performed neuronal differentiations and cell culture, PP performed mycoplasma monitoring, KZ and DCR performed Western blots, KZ designed and optimized the SparCon assay and analytical approach, KZ and MZ acquired and quantified immunocytochemistry connectivity data, WZ acquired and quantified electrophysiology data, KZ analyzed ICC and sEPSC connectivity data, KZ performed power simulations.

81 82

3.1 Abstract

Connectivity assays on cortical neurons derived from induced pluripotent stem cells

(IPSC) are complicated by cell heterogeneity. To overcome this variability, we cocultured differentially labeled neurons from Autism Spectrum Disorder (ASD) donors and related controls that were sparsely seeded on a lawn of unlabeled neurons. These Sparse coculture for

Connectivity (SparCon) assays have enhanced statistical power and reveal a striking increase in total synapse number, dendrite complexity and excitatory synapse function in heterozygous

SHANK2 ASD and gene-edited null neurons. Our SparCon assay highlights unexpected hyperconnectivity of SHANK2 mutant human neurons.

3.2 Brief Introduction and Rationale

Genetic findings indicate many ASD-associated mutations occur in loci that participate in synaptic pathways, but few reports study their effect on connectivity of human neurons (Pinto et al. 2010; Sanders et al. 2015). To date, IPSC models of severe syndromic forms of ASD and engineered hESC knockouts of ASD candidate genes provide evidence for reduced synaptic function of cortical neurons in vitro (Shcheglovitov et al. 2013; Pak et al. 2015; Djuric et al.

2015; Griesi-Oliveira et al. 2015; Yi, Danko, Botelho, Patzke, Pak, Wernig & Südhof 2016b).

Despite these initial findings, extensive heterogeneity among neurons generated from different

IPSC lines has come to be a widely acknowledged issue in the field (Sandoe & Eggan 2013;

Brennand et al. 2015). This experimental variability hinders accurate modeling of altered connectivity of affected neurons relative to controls and may prevent discovery of all but the most extreme phenotypes. Here, I develop the SparCon assay to address confounding sample variability, evaluate its statistical power to detect connectivity phenotypes, and apply it to IPSC- derived neurons from nonsyndromic ASD cases with mutations in the scaffolding protein

SHANK2 (SH3- and multiple ankyrin repeats protein 2). SHANK2 heterozygous loss-of-function mutations are high confidence ASD risk factors (Berkel et al. 2010; Leblond et al. 2014).

However, recent findings in SHANK2 knockout mice are discordant as heterozygotes do not display ASD behaviours and homozygotes exhibit inconsistent changes in synaptic connectivity in the hippocampus (Schmeisser et al. 2012; Won et al. 2012). To resolve this, I used the

SparCon assay to compare single-cell connectivity phenotypes between control and SHANK2 haploinsufficient or CRISPR/Cas9-engineered knockout human stem cell-derived neurons.

3.3 Results

3.3.1 Generation of IPSCs from controls, ASD cases, and CRISPR/Cas9n

knockout of SHANK2

To directly address the effect of heterozygous SHANK2 mutations on human neuronal morphology and connectivity, we first generated IPSCs from two families with ASD-affected males with SHANK2 haploinsufficiency, as summarized in Figure 21a. One subject harbours a nonsense mutation (SHANK2 R841X) and the other a 66 kb deletion (SHANK2 DEL) that causes a frameshift and premature stop, with whole genome sequencing revealing no additional variants of clinical significance beyond those already reported in the literature for SHANK2

DEL (CHRNA7 duplication and ARHGAP11B deletion) (Leblond et al. 2012). Multiple IPSC lines were isolated from both subjects, and control lines were generated from 3 unaffected parents and one unrelated control subject (Figure 21-23). To account for the effects of genetic background, I also generated an isogenic CRISPR/Cas9n-edited homozygous knockout

(SHANK2 KO) from the unrelated control using a selection-free strategy based on sequential enrichment of targeted cells (Miyaoka et al. 2014). Specifically, I knocked in a single-stranded oligodinucleotide repair template bearing a V5 tag and termination codons in all reading frames in SHANK2 exons 16, which is common to all known isoforms of SHANK2. Western blotting revealed no presence of V5+ bands, suggesting that nonsense-mediated decay prevents expression of the truncated protein (Figure 24). Whole genome sequencing confirmed correct targeting (Table 3) with no off-target effects.

Figure 21. Sparse co-culture for connectivity (SparCon) assays of IPSC-derived

SHANK2 ASD neurons compare marked mutant and control neurons seeded on the consistent synaptogenic environment of a lawn of unlabeled control neurons. (A) Induced pluripotent stem cells (IPSC) generated from multiple control and affected individuals are

86 differentiated into neural precursor cells (NPCs). NPCs are differentiated in separate wells for 4 weeks and then differentially fluorescently labeled control and mutant cells are sparsely seeded onto a large unlabeled neuronal population (the lawn) and co-cultured with astrocytes. (B)

Timeline of the experiment starting with seeding of NPCs. Measurements of mutant cells are normalized to control cells in the same well. (C) Sparse seeding allows simultaneous analyses of cell morphology and connectivity (total number of SYN1 punctae) of single neurons. Scale bars

100 m. (D) To compare synaptic function, sEPSCs are recorded from neurons grown in the same well.

Figure 22. Recruited families, iPSC generation and characterization. (A) Pedigrees of recruited families with autistic children. IPSC were generated from individuals in coloured squares. The number of iPSC lines used per individual is indicated. Dermal fibroblasts from

CTRL1, CTRL3, CTRL4, and SHANK2 R841X were reprogrammed using retrovirus and

CTRL2 and SHANK2 DEL iPSC were reprogrammed using Sendai Virus (see Methods). (B)

Representative characterization of iPSC from each individual. CTRL1 characterization is in

Figure 13 excerpted from a submitted manuscript (Ross et al.). Expression of pluripotency markers (OCT4, SSEA4, NANOG, TRA1-60), differentiation into three germ layers in vitro and

88 in vivo (teratoma assay) and karyotype analysis are shown. Scale bar for immunofluorescence images is 50 µm (white bar); scale bar for H&E stains is 200 µm (black bar).

Figure 23. Characterization of CTRL1 excerpted from manuscript in preparation by Ross et al. (A) iPS cell lines synthesized the pluripotency-associated proteins OCT4 and SSEA4 (scale bar: 100 µm). (B) iPS cells differentiated into the three germ layers in vitro (embryoid body assays, scale bar: 25 µm). (C) iPS cells differentiated into the three germ layers in vivo (teratoma assays, scale bar: 100 µm). (D) iPS cells had a normal karyotype.

Figure 24. Generation of SHANK2 homozygous knockout. (A) iPSC generated from CTRL1, unrelated to autistic children in the study, were nucleofected with Cas9 D10A nickase, two paired guide RNAs and a homologous recombination template single stranded oligonucleotide

(SHANK2 KO ssODN) to introduce a 3xSTOP (stop codons in all three reading frames) in exon

16 of SHANK2. This exon is common to all known SHANK2 isoforms and is one of the two missing exons in SHANK2 DEL (longest SHANK2 protein isoform shown). Transfected cells were plated in a 96 well plate, and targeted cells were distinguished from WT cells using digital

90 droplet PCR (ddPCR) with Fam-tagged probes recognizing wild-type SHANK2 sequence and

Vic-tagged probes recognizing the V5 tag, Mre1 restriction site with the 3x STOP cassette (Vic tag) in the SHANK2 KO ssODN. From the first 96 well plate, the well with the highest fractional abundance of SHANK2 tag alleles with respect to SHANK2 WT alleles was expanded into a new 96 well plate and the enrichment procedure was repeated until a well with 100% SHANK2

KO alleles was isolated. (B) Dot plots showing raw counts of SHANK2 WT (blue) and

SHANK2 KO (Vic tag, green) alleles from the wells with the highest fractional abundance of

SHANK2 KO from each 96 well plate. The well shown for plate 4 had zero detected SHANK2

WT alleles and was used to establish the SHANK2 knockout (SHANK2 KO) line. (C) ddPCR verification of homozygous SHANK2 KO line. Bar graph shows the concentration of SHANK2

WT allele, an endogenous X-linked allele expected to be present as one copy (AFF2, X-linked), an autosomal allele expected to be present at two copies (CACNA1C), and the SHANK2 KO allele in gDNA from the SHANK2 KO cell line; n=3 reactions for each paired combination (e.g.

Fam-tagged SHANK2 WT vs. Vic-tagged SHANK2 KO). (D) G-band karyotyping reveals a normal 46X,Y karyotype. (E) Immunostaining of pluripotency markers SSEA4, OCT4, TRA1-60 and NANOG. Scale bar 100 µm. (F) Differentiation into three germ layers in vivo (teratoma assay). (G) Expression of SHANK2 and V5-tagged proteins in 4 week-old neurons (left to right,

CTRL1, SHANK2 KO, H9-derived neurons overexpressing V5-tagged FUBP3). Abbreviations:

Ank Repeats: Ankyrin repeats; SH3: Src homology 3 domain; PDZ: PDZ domain; Proline-rich: proline-rich domain; SAM: sterile alpha motif; H: Homer1 binding site; D: Dynamin2 binding site; C: Cortactin binding site.

Table 3. Analysis of SHANK2 KO (Vic tag) insertion at potential gRNA binding sites with up to three mismatches. The possible gRNA binding sites were predicted by the algorithm on the Zhang laboratory website (crispr.mit.edu). **Duplication of 67 bp

(TGCCCCCAAGAATGTAACATCACTATGCTAGCACTGGTTGACCTGGTTGGGACCAA

CTCAAGGTCAG) at position 67177791 (150 bp away from gRNA A binding). The duplication is intergenic and part of a short interspersed nuclear element (SINE).

3.3.2 Development of the SparCon Assay

I observed that directed differentiation of individual IPSC lines into cortical neurons(Rodrigues et al. 2016) produced neuronal cultures that were heterogeneous in their final cell density (Figure 25), and reasoned that the final cell type composition and maturity may also

92 differ over the prolonged experimental time course necessary for synaptic development in human neurons. The accompanying differences in neuronal synaptogenesis, which is modulated by extracellular cues, may increase experimental noise and obscure biological differences in connectivity between control and mutant neurons. In principle, maximal consistency would result if a controlled synaptogenic environment were used in every experiment. This is the foundation of the SparCon assay (Figure 21a), in which a small number of differentially labeled pairs of SHANK2 mutant and control neurons are sparsely seeded together onto a lawn of unlabeled control neurons. In order to maximize consistency across different comparisons, the lawn is derived from the same line (CTRL1 in all experiments described here) and held constant in all experiments.

Figure 25. Summary of challenges with measuring connectivity. (A) In a typical experiment,

NPCs generated from different iPSC lines will proliferate and differentiate at different rates, resulting in variable density and cell composition between cell cultures under comparison. Given the extended time frame human neurons require to develop synaptic connectivity, such technical variation may amplify to the extent that variability between the synaptogenic environments in different cultures will be so large that it will obscure the biological differences in connectivity between the cells. (B) Typical variance in density at the end of a typical experiment among control and ASD lines.

It is critical to characterize the appropriate population of neurons to assay by SparCon. We used an embryoid-body based differentiation protocol to generate SOX1+ neural precursor cells

(NPCs) that were expanded and frozen for use in multiple experiments (Rodrigues et al. 2016)

(Figure 26a,b). We found that upon terminal differentiation, the NPCs gave rise to MAP2+ cortical neurons, subsets of which expressed markers BRN2, CTIP2, SATB2, or TBR1 representative of mixed cortical layers (Figure 26c). To evaluate SHANK2 expression in the

IPSC derived neurons and validate the genotypes, Western blot analyses showed that SHANK2 protein levels were halved in neurons from heterozygous SHANK2 R841X and DEL ASD cases.

The SHANK2 KO neurons displayed a faint band of the wrong molecular weight, suggesting either nonspecific binding of antibody or existence of an isoform not including exon 16 of

SHANK2 (Figure 27a). Furthermore, SHANK2 protein levels significantly increased upon differentiation from NPCs to neurons and peaked at 4 weeks of differentiation, demonstrating that SHANK2 is expressed during the onset of synaptogenesis in vitro (M. A. Johnson et al.

2007; Gupta et al. 2013) and developmentally regulated in human IPSC-derived neurons (Figure

27b). Thus the differentiation protocol produces SHANK2-expressing control and mutant cortical neurons suitable for connectivity studies.

Figure 26. Generation of NPCs and neurons. (A) Differentiation protocol. (B) Left:

Representative image of expression of NPC (SOX1) and neural crest (AP2) markers in NPC cultures. Right: quantification of the percentage of cells expressing these markers across

96 differentiations of all lines used in the study. Means + SEM, n=4-10 wells totaling at least 1000 cells each. IMR90 cells were used as a negative control. (C) Representative images of 9-week old neurons expressing cortical markers BRN2, CTIP2, SATB2, and TBR1 grown on mouse astrocytes. A median filter (8 x 8 pixels) was applied to the red channel to remove noise and approximate staining to a median value. Scale bar 50 µm.

Figure 27. Western blot and quantification of SHANK2 protein levels in iPSC, NPC, and iPSC-derived neurons. (A) Left: Representative western blot of SHANK2 protein in 4 week old neurons among control (magenta), ASD (green), and knockout lines (blue). Right: Quantification of 3 western blots which collectively represent 3 control lines (CTRL1 #2, CTRL2 #4, CTRL3

#39) and 5 mutant lines (SHANK2 R841X #2,5,13, SHANK2 DEL #2, SHANK2 KO H6).

Protein levels in SHANK2 +/- samples are normalized to those of controls on each blot.

Student’s t-test * P < 0.05 (B) Left: Representative blot of SHANK2 protein levels throughout

98 differentiation in iPSC, NPC, 2-, 4-, and 6-week old neurons. CTRL1 sample shown. Right:

Quantification of 3 Western blots from different lines (CTRL1 #2, CTRL3 #39, SHANK2

R841X #2). Quantification is normalized to protein levels in IPSC for each line. One-way

ANOVA with Tukey’s post-hoc test. * P<0.05

To set up SparCon assays, I transduced 3 week-old differentiating control and SHANK2 mutant neurons with CaMKII⍺-mKO2 or CaMKII⍺-mKate2 (both red), or with CaMKII⍺-GFP

(green) lentivirus, which preferentially labels neurons likely to express SHANK2 (Shcheglovitov et al. 2013). Following a 1 week recovery, differentially labeled control and SHANK2 mutant neurons were sparsely seeded (at 2% of total neurons each) onto an unlabeled control lawn (96% of total neurons) on a layer of mouse astrocytes to promote synaptogenesis over 5 weeks (Figure

21a,b). After a total of 9 weeks of differentiation in this context, neurons marked by the reporters also expressed cortical (BRN2, CTIP2, TBR1) and excitatory (VGLUT1) markers as expected(Shcheglovitov et al. 2013) (Figure 28a). These neurons were electrophysiologically active and capable of generating action potentials (Figure 28b). SHANK2 mutant neurons did not differ from controls in intrinsic membrane properties (Figure 28c).

Figure 28. Generation of functional neurons within 9 weeks in the sparse co-culture. (A)

Phase contrast image of live neurons and fluorescent images of both mKO2 and GFP labeled

100 neurons co-staining with dendritic markers (MAP2), excitatory markers (VGLUT1), and cortical markers (CTIP2, BRN2, TBR1). Scale bar 50 µm. (B) Representative traces of evoked action potentials generated by stepwise injection of current (+5 pA) from CTRL, ASD, and engineered

SHANK2 KO cells. (C) Quantification of intrinsic membrane properties grouped by co-culture.

Each dot represents one measurement from one neuron. Co-cultures from left to right: CTRL

(CTRL1, 1 line, n=10) vs SHANK2 KO (1 line, n=10 neurons), CTRL (CTRL1, CTRL2,

CTRL3, 1 line each, n = 32) vs. SHANK2 R841X (3 lines, n=35), CTRL (CTRL1, 1line CTRL4,

2 lines, n = 28) vs. SHANK2 DEL (2 lines, n = 27); 146 neurons total. T-tests were used to compare samples in each co-culture and no statistically significant differences at ⍺ = 0.05 were found.

3.3.3 Generation and characterization of connectivity data using SparCon

We used SparCon to analyze connectivity differences between the SHANK2 mutant and control IPSC derived neurons. Because sparse seeding allows visualization of entire marked neurons by immunostaining, I directly measured connectivity by counting the total number of

Synapsin I punctae, common to 99% of mammalian central synapses (Micheva et al. 2010), for each neuron along fluorescently labeled neuronal dendrites (Figure 21c). This was accompanied by simultaneous acquisition of morphological data for dendrite length and complexity on the same cells, providing a heretofore unrealized level of detail on connectivity of human stem cell- derived neurons. To characterize synaptic activity, we performed electrophysiological recordings of AMPA receptor-dependent spontaneous excitatory postsynaptic currents (sEPSCs). The recordings were done in the absence of action potential blockers and hence the sEPSCs are driven by the network activity onto the neuron under study. (Figure 21d).

101

We evaluated 5 experimental batches of neurons by measuring at least 5 mutant and control cells in each of 87 wells, resulting in acquisition of synapse number and morphology data from

479 neurons and electrophysiological recordings of sEPSCs from 480 neurons. Neurons from 3 lines each of the R841X and DEL ASD subjects were cocultured with 2-5 control lines

(including at least one parental control) in pairs using an experimental design in which the reporter used to mark each genotype was blinded. The KO neurons were compared to their isogenic control and one unrelated control line. (Tables 2, 3).

3.3.4 Within-well normalization

To examine anatomic synaptic connectivity, we first quantified absolute numbers of synapses per neuron to explore the range of variation between batches and between wells within a batch

(Figure 29a). Variation was immediately evident with total synapse number per cell in batch 2 being lower and batch 3 being higher than batch 1 (Figure 29a). To minimize sources of error and facilitate comparisons across wells and batches, the SparCon setup allows relative calculations of total synapse numbers by within-well normalization. Because absolute synapse number and dendrite length in control neurons both fit lognormal distributions (Figure 29b), we normalized measurements of each neuron in a given well to the geometric mean of control neuron values in the same well. Following normalization, data across batches still fit the lognormal distribution for control neurons, but was largely devoid of systematic error (Figure

29c-e). The same held true for sEPSC frequency and amplitude of control neurons (Figure 30).

We also performed colour swapping experiments that compared the same line to itself in different colors (Tables 4 & 5). We did not detect any reporter gene bias as normalized synapse number and dendrite length were not dependent on which fluorescent protein was expressed

102

(Figure 31a), and the same held true for normalized sEPSC frequency and amplitude (Figure

31b). Taken together, I conclude that normalization of SparCon data addresses confounding sample variability, and that the reporter genes have no effect on neuronal connectivity.

Figure 29. Batch-to-batch and well-to-well noise in connectivity measures can be smoothed by within-well normalization. (A) Well-to-well variation in synapse number across experimental batches. Red dots: control neurons; Green dots: SHANK2 +/- ASD neurons; Blue dots: SHANK2 KO -/- neurons. Mean +/- s.e.m. shown. Inset: batch-to-batch variation in

103 synapse number in neurons derived from the same line (CTRL1) (B) Histogram of absolute measurements among control neurons in one experimental batch overlaid with a fitted lognormal distribution (red line). Synapse number and dendrite length in control neurons from batch 1 (n =

115). (C) Cumulative distribution of data in (B) is consistent with a lognormal distribution (red line); each circle is one neuron. (D) Cumulative distribution of measurements across all batches before (left) and after (right) within-well normalization. Synapse number (top) and dendrite length (bottom) in control neurons from batches 1, 2, and 3 (n = 210). (E) Normalized values for synapse number (top) and dendrite length (bottom) in neurons of the same line across different experimental batches. Each dot represents one neuron. Mean +/- s.e.m. shown.

104

Figure 30. Distribution of sEPSC frequency and sEPSC amplitude among controls. (A)

Histogram of absolute measurements among control neurons in one experimental batch overlaid with a fitted lognormal distribution (red line). sEPSC frequency (top) and sEPSC amplitude

(bottom) from control neurons from batch 5 (n=120). (B) Cumulative distribution of data in (A) is consistent with a lognormal distribution (red line); each circle is one neuron (C) Cumulative distribution of sEPSC frequency (top) and sEPSC amplitude (bottom) in controls across all batches (n=180) before (left) and after (right) within-well normalization.

105

106

Figure 31. No bias due to fluorophore expression on connectivity measurements. (A) Left: measurements of synapse and dendrite length among GFP and mKO2 cells from 6 wells where the same line was cultured against itself in different colors (left: CTRL3 #39, middle: R841X #2, right: R841X #5). mKO2 measurements are normalized to the geometric mean of GFP measurements in each well. Right: Cumulative probability and dot plots of pooled data from the wells on the left. Each dot represents one neuron, GFP n= 32 and mKO2 n=34. (B) Left: measurements of sEPSC Frequency and sEPSC Amplitude among GFP and mKate2 cells from 6 wells where the same line was cultured against itself in different colors (CTRL1 #2). mKate2 measurements normalized to geometric mean of GFP measurements in each well. Cumulative probability and dot plots of pooled data from the wells on the left. Each dot represents one neuron, n = 15 each. Anderson-Darling k-samples tests were used to compare samples in each co-culture and no statistically significant differences at = 0.05 were found.

107

Table 4. Comparisons between control and mutant cells on each coverslip used for immunocytochemistry-based analysis of connectivity. Rows highlighted in grey indicate wells used for color-swapping control experiments where the same line was cultured against itself in different colors.

108

109

Table 5. Comparisons in each well for electrophysiological analysis of synapse function.

Rows highlighted in blue indicate wells to which BDNF (10 ng / µL) was added. Rows highlighted in grey indicate wells used for color-swapping control experiments where the same line was cocultured against itself. Row 24 was used for color control experiments and had BDNF added.

3.3.5 Power Simulations

Since the data fit lognormal distributions, we hypothesized that nonparametric statistical tests applied to within-well normalized data may be more suitable for subsequent calculations. We performed Monte Carlo simulations to determine which statistical test was the most sensitive and whether relative within-well normalization increased statistical power at ⍺=0.05. These simulations revealed that the Anderson-Darling k-samples (AD) test is more sensitive than the

Kolmogorov-Smirnov and more sensitive or equal to Student’s t-test for all measures (Figure

32). Critically, while within-well normalization significantly improved sensitivity for all measures, the effect was most pronounced on connectivity measures, such as synapse number and sEPSC frequency. For unnormalized synapse number, 45 neurons per group are necessary to detect a 1.5-fold difference for a power > 0.8 at ⍺ = 0.05, but only 20 are required after within- well normalization (Figure 33a). At this sample size of 20, skipping the normalization step requires a 2 fold-difference in order to achieve a power > 0.8 (Figure 33b). Similarly, for sEPSC frequency, within-well normalization reduces the required sample size to detect a 2-fold difference from 50 neurons to 35. At this sample size of 35, a 2.5 fold-difference is necessary to achieve a power > 0.8 if the data are unnormalized. While similar trends hold true for dendrite length and sEPSC amplitude, the impact is most pronounced on synapse number and sEPSC

110 frequency, suggesting that these two measures of connectivity are particularly susceptible to experimental variation arising from heterogeneity of IPSC-derived neuronal cultures (Figure

33). We therefore used within-well normalization with the Anderson-Darling k-samples test to evaluate SHANK2 connectivity phenotypes.

Figure 32. Anderson-Darling k-samples test is more sensitive than Kolmogorov-Smirnov or

T tests for most measures and within-well normalization significantly increases the statistical power at smaller sample sizes. In silico simulations comparing the sensitivity of

Anderson-Darling k-samples, Kolmogorov-Smirnov, and Student’s t tests to detect the indicated differences (1.5-fold for synapse number, 2-fold for sEPSC frequency, 1.25-fold for dendrite length and sEPSC amplitude) at ⍺ = 0.05 for different sample sizes (from 5 to 180 in steps of 5) drawn from (A) unnormalized data or (B) within-well normalized data. Control neurons from the respective datasets were used as the population from which the samples were randomly drawn with replacement (N=210 for synapse number and dendrite length, N=180 for sEPSC frequency and sEPSC Amplitude). 1000 simulations were run at each sample size.

111

Figure 33. Within-well normalization increases sensitivity of the assay to detect connectivity differences. (A) In silico simulations comparing the sensitivity of Anderson-

Darling k-samples test on normalized (green) and unnormalized (magenta) data to detect the indicated differences (1.5-fold for synapse number, 1.25-fold for dendrite length, 2-fold for sEPSC frequency, 1.25-fold for sEPSC amplitude) at ⍺ = 0.05 for different sample sizes (from 5 to 180 in steps of 5). (B) In silico simulations comparing the sensitivity of Anderson-Darling k- samples test on normalized (green) and unnormalized (magenta) data to detect a range of possible differences (fold changes ranging from 1 to 2.5 in steps of 0.1) at the indicated sample size (20 for synapse number, 30 for dendrite length, 35 for sEPSC frequency, 20 for sEPSC amplitude) at =0.05.

3.3.6 Hyperconnectivity in SHANK2 mutant human neurons

After immunocytochemistry staining for the fluorescent proteins, we traced dendrite morphology of the differentially marked control and SHANK2 mutant neurons (Figure 34a).

Synapsin1 puncta localized on the labeled neurons revealed a significant increase in synapse

112 number in SHANK2 mutant neurons (Figure 34b) (1.91-fold for R841X, 1.54-fold for DEL,

1.67-fold for SHANK2 KO), suggesting that lowered SHANK2 causes an increase in synaptic connections. This was accompanied by increases in dendrite length (Figure 34c) in all SHANK2 mutant cells (1.50-fold for R841X, 1.13-fold for DEL, 1.51-fold for SHANK KO). Consistent with a previous report on SHANK2 knockdown in rat neurons in vitro (Berkel et al. 2012), we found that dendrite branching complexity, measured by Sholl analysis, was increased in all

SHANK2 mutant cells (Figure 34d). Taken together, these findings indicate that lowered

SHANK2 dosage increased the total synapse number resulting in neuronal hyperconnectivity, which was partly driven by increases in dendrite length and complexity.

113

Figure 34. Synapse numbers and neuronal complexity are enhanced in SHANK2 mutant neurons. (A) Paired representative traces of control, SHANK2 ASD, and engineered SHANK2 knockout neurons. (B) Total synapse number, normalized within-well by dividing synapse number of individual cells by the geometric mean of the synapse number of control cells in a given well. (C) Total dendrite length, normalized within-well by dividing dendrite length of individual cells by the geometric mean of dendrite length of control cells in a given well. (D)

Sholl analysis, normalized within-well by subtracting the mean of control crossings from the number of crossings made by individual cells at a given radius. Radius step 10 µm. For (B,C,D)

114

SHANK2 mutant cells are compared to co-cultured control cells from the same wells: SHANK2

KO (n=40) vs. CTRL (CTRL1, isogenic, n=46), SHANK2 R841X (n=89) vs. CTRL (CTRL1,

CTRL2 (paternal), CTRL3 (maternal), n=96), SHANK2 DEL (n=74) vs. CTRL (CTRL1,

CTRL2, CTRL3, CTRL4 (paternal), n=68). Means + s.e.m. plotted. Total number of neurons =

407 from 5 WT lines from 4 CTRL individuals, 6 SHANK2+/- lines from 2 ASD individuals and

1 engineered SHANK2-/- KO line. Means + s.e.m. plotted,* P <0.05, ** P <0.01, *** P<0.005;

Anderson-Darling k-samples test for (B) and (C) and two-way ANOVA for (D).

To test whether increases in synapse number are associated with alterations in excitatory synaptic function, we performed patch-clamp recordings of AMPA receptor sEPSCs in

SHANK2 mutant and control neurons. There were no consistent differences in sEPSC amplitude across all SHANK2 genotypes (Figure 35b). A modest 1.18-fold increase was detected for

SHANK2 KO but not SHANK2 DEL, and the 1.14-fold difference in SHANK2 R841X was not observed after chronic addition of BDNF to examine more sEPSCs per neuron (Figure 36a).

However, when we examined connectivity using sEPSC frequency, we found a consistent and striking increase across all genotypes (5.4-fold for R841X, 2.8-fold for DEL, 6.4-fold for

SHANK2 KO) (Figure 35b) and with BDNF treatment (2.9-fold for R841X) (Fig 36b). Given that both control and mutant neurons synapse with the same population of neurons in SparCon assays, increased sEPSC frequency is likely to reflect an increase in synapse number in the neurons lacking SHANK2. The relatively bigger fold-change in sEPSC frequency compared to that in synapsin I puncta suggests that SHANK2 mutant neurons may preferentially make excitatory synapses, which can be expected with an elaboration of dendrite structure (Spruston

2008). The relatively smaller fold-change increase in R841X cells treated with BNDF likely

115 reflects a stimulatory effect of BDNF on control cells, as BDNF treatment increased sEPSC frequency n controls by 69%. Therefore, increased synapse numbers in SHANK2 mutant neurons are associated with markedly increased sEPSC frequency, providing strong evidence that SHANK2 regulates neuronal connectivity.

Figure 35. Excitatory synaptic function is enhanced in SHANK2 mutant neurons. (A)

Representative sEPSC traces with an averaged sEPSC trace (inset) from single control, ASD, and engineered SHANK2 KO neurons. (B) sEPSC frequency and sEPSC amplitude, normalized within-well by dividing measurements of individual cells by the geometric mean of control cells in the same well. SHANK2 mutants are compared to co-cultured control neurons from the same wells: SHANK2 KO (n=60) vs. CTRL (CTRL1 (isogenic), CTRL2, n=60), SHANK2 R841X

(n=60) vs. CTRL (CTRL1, CTRL3 (maternal), n=60), SHANK2 DEL (n=60) vs. CTRL

(CTRL4 (paternal), n=60). Total number of neurons = 360 from 5 WT lines from 4 CTRL

116 individuals, 6 SHANK2+/- lines from 2 ASD individuals and 1 engineered SHANK2-/- KO line.

Means + s.e.m. plotted, * P <0.05, ** P <0.01, *** P<0.005.; Anderson-Darling k-samples test.

Figure 36. Increased sEPSC Frequency in SHANK2 R841X neurons in the presence of

BDNF (10 ng/mL) throughout the course of differentiation. (A) Representative sEPSC traces with an averaged sEPSC trace (inset) from single control and SHANK2 R841X neurons. (C)

Cumulative probability and dot plots of SHANK2 R841X (3 lines, n=45) sEPSC frequency and amplitude normalized to geometric mean of CTRL (CTRL1, CTRL3, n=45) neurons in each well. *** P < 0.005; Anderson-Darling k-samples test.

117

3.4 Brief Summary and Discussion

Taken together, our findings show that human neurons with ASD-associated SHANK2 mutations make more functional excitatory connections in a cell-autonomous manner relative to controls, and this phenotype is distinct from the lowered connectivity associated with mutations in SHANK3 or other ASD-associated genes (Shcheglovitov et al. 2013; Pak et al. 2015; Yi,

Danko, Botelho, Patzke, Pak, Wernig & Südhof 2016a). The presence of the same hyperconnectivity phenotype in CRISPR/Cas9 engineered homozygous knockout neurons indicates that SHANK2 mutations cause this phenotype. Our results provide further evidence that SHANK2 functions in early neuronal development (Du et al. 1998) and serves an important role as a suppressor of excessive dendrite branching (Vessey & Karra 2007). Given this early role for SHANK2, it is possible that it plays a less important role in synapse formation, which may be supported by other SHANKs or scaffolding proteins. Thus, an increase in dendrite branching could lead to an overproduction of synapses by simply providing more dendrite length on which synapses can be formed. These findings strongly suggest that in human neurons,

SHANK2 has distinct function from SHANK3. Whereas SHANK3 protein levels appear critical for excitatory synapse function and formation (Shcheglovitov et al. 2013; Bidinosti et al. 2016),

SHANK2 influences dendrite development well before the onset of synaptogenesis.

The SparCon method is well-suited for comparing multiple IPSC lines from several different patients and controls where cell heterogeneity is pronounced due to genetic background or batch effects. The use of a consistent lawn mitigates heterogeneity arising from line-to-line variability and co-cultures permit within-well normalization that mitigates the impact of technical well-to- well and batch-to-batch variation. Within-well normalization is crucial, as our absolute number comparisons (Figure 37) are statistically less powered to detect differences. Critically, SparCon

118 measures the cell-autonomous synaptogenic propensity of human neurons in isolation from other functions. In addition, sparse seeding allows simultaneous characterization of total synapse number and dendrite morphology, which can guide interpretation of results.

The SparCon method has increased power to detect connectivity phenotypes in IPSC-derived human neurons that would have been otherwise obscured by experimental noise. It has additional advantages in its adaptability to study many neuronal subtypes with potential to model a variety of neurological disorders that are thought to arise due to altered synaptic connectivity (State &

Geschwind 2015). For example, lentiviral labeling with other cell-type specific promoters such as hvGAT to selectively label inhibitory neurons (DeRosa et al. 2015) can restrict connectivity analyses to disease-relevant populations. Alternatively, labeling with markers specific to neuronal maturational stages such as GDAP1L1 (Bardy et al. 2016) can restrict analyses to neurons of similar age and electrophysiological properties. In addition, experimenters have the option to engineer lawns of specific neuronal subtypes (e.g. purely excitatory lawns generated by

NGN2 conversion (Yingsha Zhang et al. 2013)), lawns composed of defined ratios of mixed populations, or lawns expressing optically- or chemically-gated modulators of activity, which can greatly expand the range of possible testable phenotypes. Overall, using SparCon, we find that in vitro derived human neurons are sensitive to SHANK2 dosage, as engineered knockout or neurons derived from ASD cases with heterozygous SHANK2 mutations become hyperconnected. This phenotype provides the first functional evidence for hyperconnectivity of neurons derived from ASD subjects.

119

Figure 37. Unnormalized measurements from the connectivity datasets generated for the study yield fewer statistically significant discoveries. For (A, B, C, D), SHANK2 mutant cells are compared to co-cultured control cells from the same wells. For (A, B) see caption of Figure

34 for sample information. For (C, D) see caption of Figure 35 for sample information. * P <

0.05, ** P < 0.01, *** P < 0.005; Anderson-Darling k-samples test.

120

3.5 Materials and Methods

3.5.1 Generation of iPSCs

All pluripotent stem cell work was approved by the Canadian Institutes of Health Research Stem

Cell Oversight Committee. With approval from the SickKids Research Ethics Board, we obtained dermal fibroblasts from skin-punch biopsies at the Hospital for Sick Children, after informed consent was obtained from all subjects. Induced pluripotent stem cells were generated by either retroviral reprogramming (CTRL1, CTRL3, CTRL4, SHANK2 R841X) as described(Hotta et al. 2009) or Sendai virus reprogramming (CTRL2, SHANK2 DEL) using the

Invitrogen CytoTune-iPS kit. During reprogramming and clonal expansion, iPSC colonies were grown on gelatin-coated plates on a layer of Mitomycin-C inactivated mouse embryonic fibroblasts extracted at embryonic day 12.5 (acquired from Hospital for Sick Children

Embryonic Stem Cell Facility) in hESC media (Knockout DMEM supplemented with 15%

Knockout serum replacement, 1x GlutaMAX, 1x penicillin/streptomycin, 1x non-essential amino acids, 0.5 mM 2-mercaptoethanol, and 10 ng/ml basic fibroblast growth factor (bFGF)). At this stage, the cells were passaged manually or with collagenase IV as described. Established iPSC lines were transitioned to plates coated with matrigel and grown in either mTeSR or StemMACS iPS Brew medium. From this point on, the cells were passaged either manually, using Dispase or

ReleSR every 5-7 days.

To verify genome integrity, G-band karyotyping was performed by the Centre for Applied

Genomics (Hospital for Sick Children) and only iPSC lines without karyotypic abnormalities were used for analyses. To test expression of pluripotency associated factors OCT4, NANOG,

TRA1-60, and SSEA4, immunocytochemistry was performed. To test whether cells were functionally pluripotent in vitro, iPSCs underwent spontaneous differentiation through an

121 embryoid-body based step as described (Cheung et al. 2011), fixed, and stained for expression of endoderm (-fetoprotein, AFP), mesoderm (smooth muscle actin, SMA), and ectoderm (3-tubulin,

TUBB3). Functional pluripotency in vivo was tested using teratoma assays. IPSC lines were grown in 10 cm matrigel-coated dishes, digested with collagenase IV and injected into the flank of NOD/SCID mice as cell clumps immersed as described (Cheung et al. 2011). Two months later or when obvious tumors appeared, the mice were sacrificed and the tumors were resected, stained with hematoxylin and eosin, and imaged.

3.5.2 Generation of SHANK2 knockout cells.

Two million iPSCs were nucleofected using the Amaxa Nucleofector II with program B-

016 in Solution I with 5 g CRISPR D10A nickase plasmid (Addgene, #44720), 2.5 g each of two gRNA cloning plasmids (Addgene, #41824) containing paired gRNAs A and B targeting exon 16 of SHANK2, and 1 l of 10 M ssODN containing two 60 nucleotide-long arms of homology and

SHANK2 KO sequence comprised of V5 tag, Mre1 restriction site and stop codons in all three reading frames. Following nucleofection, the cells were distributed in a 96 well plate. Following

2 weeks of growth, 1/3 of the cells were passed to a new 96 well plate and maintained, 1/3 were passed to a new 96-well plate and frozen in 90% KOSR - 10% DMSO mix with 10 µM ROCK inhibitor, and 1/3 had gDNA extracted as in Miyaoka et al. (Miyaoka et al. 2014) Digital droplet

PCR was used to determine the proportion of SHANK2 tag alleles in each well with a Fam- conjugated probe for the SHANK2 WT sequence and Vic-conjugated probe for the SHANK2

KO (Vic-tag) sequence. The well with the highest proportion was expanded into a new 96 well- plate and the procedure was repeated until wells with 100% SHANK2 tag allele and 0%

SHANK2 WT were found. The isolated wells were expanded and the resulting iPSC lines were

122 characterized to assay pluripotency (in vitro via ICC for pluripotency markers, in vivo via teratoma assays) and karyotyping. Off-target effects were assessed using whole-genome sequencing performed by The Centre for Applied Genomics and analyzed as described below.

3.5.3 Characterization of off-target effects resulting from CRISPR/Cas9n knockout of SHANK2

Genomic positions with designed gRNA sequences matching the human genome (hg38) were downloaded from the CRISPR Design Tool at the Zhang Lab website (http://crispr.mit.edu/).

Because our gRNA cloning protocol sets the first 5’ nucleotide as G which is dispensable for specificity, mismatches at that position were ignored. Likewise, mismatches at the PAM sequence were ignored. Sequences with up to 3 mismatches in the remaining fragments were then selected for analysis. Because hg38 genome may be missing parts of hg19 reference genome that may contain additional gRNA binding sites, the positions of those fragments were remapped to hg19, but did not yield any additional possible binding sites.

Whole genome sequencing was performed on both the control line and the derived

SHANK2 KO line at The Centre for Applied Genomics. The reads were mapped using the bwa algorithm (0.7.12) (H. Li & Durbin 2010) and the genome was analyzed with GATK(McKenna et al. 2010).

To verify that the SHANK2 KO (Vic tag) sequence was inserted at the right location, 100 bp of SHANK2 KO sequence was reconstructed in each direction at each locus predicted by the

CRISPR Design Tool. The only insertion was found at the on-target site, and reconstruction revealed it matched the SHANK2 KO (Vic tag) sequence (Table 3). The sequence was verified by PCR and Sanger sequencing.

123

To search for off-target SHANK2 KO (Vic Tag) insertions at sites not predicted by

CRISPR Design Tool, GATK (GenotypeConcordance) was used to identify de novo indel and

SNP calls in SHANK2 KO genome with respect to the control genome. Because of the length of the SHANK KO sequence, it could have appeared as a series of GATK calls in close proximity to one another. Therefore, all calls occurring within a 200bp window were grouped, and GATK

FastaAlternateReferenceMaker was used to reconstruct the SHANK2 KO sequence within that window to identify any potential insertion. We only found the SHANK2 KO (Vic Tag) insertion at the expected locus.

To determine if the SNPs/indels unique to the control line can create additional gRNA binding sites and facilitate off-target mutagenesis, we reconstructed 100 bp of genome sequence of the control line in both directions at each of the identified unique calls. The gRNA sequences

A and B were then mapped to those sequences using bwa to identify potential gRNA binding sites. Both A and B gRNAs mapped only to regions containing the exact match of the sequence near the on-target location of SHANK2 exon 16, suggesting an absence of off-target effects caused by CRISPR off-target effects.

3.5.4 Differentiation to Neural Precursor Cells (NPCs)

NPCs were generated using dual SMAD and Wnt inhibition with manual isolation of neuronal rosettes (Chambers et al. 2009; Brennand et al. 2011; Maroof et al. 2013). Briefly, iPSC colonies were digested for with collagenase IV (1 mg/mL) until they lifted off the plate.

Subsequently, the colonies were grown as free-floating cellular aggregates for 4 days in EB media (20% knockout serum replacement, DMEM/F12, 1x NEAA, 1x penicillin/streptomycin,

20ng/mL bFGF), after which they were adhered to a matrigel coated plate (1/71) and grown for

124 an additional 4 days in N2B27 media (DMEM/F12, N2 1x, B27 1x, 1x non-essential amino acids, 1x penicillin/streptomycin, 20ng/mL bFGF). For the first 8 days of differentiation, the cells were grown in the presence of SB431542 (10 M) and dorsomorphin (2.5 M). The cells were additionally treated with XAV939 (2 M) during the first 4 or 8 days of differentiation, depending on the cell line, to facilitate more efficient differentiation. After 8 days, neural rosettes were lifted by a brief incubation (3-5 min) with Dispase and tapping, washed with PBS, and adhered to another matrigel-coated plate and fed with N2B27 media. Following that, secondary or tertiary rosettes were manually dissected to purify neural precursor cells (NPCs). NPCs were then expanded and frozen in 10% DMSO-90% N2B27 media. NPCs were characterized by immunocytochemistry for expression of NPC marker SOX1 and neural crest marker AP2 using the ThermoFisher Cellomics automated imaging and analysis platform. NPCs were seeded at a density of 15000 / well of a 96-well Greiner microclear plate. IMR-90 fibroblasts were used as a negative control and seeded at a density of 10000 cells/well. Using a customized

TargetActivation algorithm, DAPI nuclei were detected and defined as SOX1 and AP2 positive if the average intensity of signal was greater than that observed in 99% of IMR-90 fibroblasts.

Only NPC cultures with 60% SOX1+ cells and less than 5% AP2+ cells were used for subsequent differentiation into neurons and phenotyping.

3.5.5 Neuronal Differentiation

NPCs were seeded at a density of roughly 1000 cells / mm2 (1000000 cells / well of a 6- well plate) on matrigel coated plastic plates (matrigel concentration 4x that used for iPSC, as determined by the dilution factor supplied by Corning with each matrigel batch). The cells were fed with Neuronal Differentiation (ND) media (Neurobasal, 1x N2, 1x B27, 1x glutamax, 1x penicillin/streptomicin, 1/500 laminin) supplemented with ascorbic acid (200 nM), BDNF (10

125 ng/L), GDNF (10 ng/L), cAMP (10 M) every other day by changing all of the media in a given well. For days 10-14, the cultures were treated with 5 M DAPT to attempt to curb excess NPC proliferation and promote differentiation (Figures 21, 26).

3.5.6 Sparse-seeding co-culture

To begin neuronal differentiation, NPCs were seeded at 1 million cells per well of a 6 well matrigel coated plate, as above, with one well per line per color. Nine to twelve wells of lawn

NPCs (CTRL1 line #2 was used as the lawn; see Tables 4-7) were seeded at the same time. For the first four weeks, the cells were fed as above. At day 21, neurons were infected with virus bearing different fluorescent colors (e.g. CaMKII-GFP for ASD and CaMKII-mKO2 or

CaMKII-mKate2 for control cells; 10 µL virus / well from a standard preparation, see Virus

Preparation). At this time, an appropriate number of vials of mouse astrocytes were thawed into a

T-75 flask such that at least 1.25 million astrocytes would be present at day 26. At day 26, astrocytes at a minimum density of 250 cells / mm2 (~50000 per well of 24 well plate) were seeded on poly-L-ornithine and laminin coated coverslips (12 mm in diameter) or coated plastic wells. At day 27, neurons of the lawn were digested with Accutase for 15 minutes, carefully triturated using a Gilson P1000 pipetman a maximum of 15 times, washed with PBS, counted, and seeded at a density of at least 1000 cells/mm2 (200000 cells / well of a 24 well plate).

Neuronal clumps were not removed at this stage to ensure there were enough cells of the lawn to seed. On day 28, infected neurons were digested with accutase for 15 minutes, carefully triturated using a Gilson P1000 pipetman a maximum of 15 times, strained with a 70 µm cell strainer to remove clumps, washed with PBS, and counted. After counting, each infected neuronal line was diluted to a concentration of 200000 cells / mL, and equal volumes of lines to

126 be compared with each other were combined into desired comparisons and seeded onto the lawn.

In order to account for error associated with efficiency of differentiation, infection, pipetting, and counting, three coverslips were seeded per comparison, with 2000, 4000, or 6000 cells (1%, 2%, or 3% of total neuron number) per line. While the viability of differentially transduced cells was not assessed, it is likely that it did not differ given the negative findings in color control experiments (Figure 21). Over the following week, the coverslips were monitored - those where there were at least 10 visible cells per line that were seeded sparsely enough so as to facilitate acquisition of images of full individual neurons were kept. Following re-seeding, the co-cultured neurons were fed with ND media supplemented with ascrobic acid (200 nM) or with ascorbic acid and BDNF (10 ng/mL) by replacing 1/2 the media every other day. See Tables 4 and 5 for information regarding which lines were co-cultured together for each experiment.

3.5.7 Virus preparation

Seven million HEK293T cells were plated in a T-75 flask, grown in 10% fetal bovine serum in DMEM. The next day, cells were transfected using Lipofectamine 2000 with plasmids for gag-pol (10 µg), rev (10 µg), VSV-G (5 µg), and the target construct (15 µg, CaMKII⍺-GFP or CaMKII⍺-mKate2 (Shcheglovitov et al. 2013) or CaMKII⍺-mKO2. Next day, the media was changed. The day after that, the media was spun down in a high-speed centrifuge at 30000g at

4°C for 2 hours. The supernatant was discarded and 80 µl PBS was added to the pellet and left overnight at 4°C. The next day, the solution was triturated and distributed into 10 µl aliquots and frozen at -80°C. As noted, above, one full aliquot was used to infect neurons in one well of a 6- well plate.

127

3.5.8 Astrocyte Preparation

Astrocytes were prepared from postnatal day 1 CD-1 mice as described. (H. J. Kim &

Magrané 2011) Before use, astrocytes are checked for mycoplasma contamination.

3.5.9 Cloning

Guide RNAs were cloned into the gRNA cloning plasmid (Addgene #41824) by following option B in the instructions supplied with the plasmid

(https://www.addgene.org/static/data/93/40/adf4a4fe-5e77-11e2-9c30-003048dd6500.pdf).

Guide RNAs were designed using the CRISPR design tool on the Zhang Lab website

(crispr.mit.edu). (Hsu et al. 2013) The CaMKII⍺-mKO2 plasmid was constructed in lab and described in Rodrigues et al. 2016.

3.5.10 Western blot

Protein was extracted from cells grown in 6-well plates by washing the cells 2 times with ice-cold PBS, followed by immersion in RIPA buffer (150 mM sodium chloride. 1.0% NP-40 or

Triton X-100 , 0.5% sodium deoxycholate, 0.1% SDS (sodium dodecyl sulfate), 50 mM Tris, pH

8.0) with a complete Mini protease inhibitor. The cells in RIPA buffer were kept on ice for 10 mins and vortexed every 2-3 min. After 10 min, the samples were sonicated with on a Fisher

Scientific sonic dismembrator F-60 and stored at -80°C. Proteins were quantified using the

BioRad DC Assay kit. At least 20 µg of protein was loaded per sample per lane of a BOLT 4-

12% gradient gel loaded into a BOLT mini gel tank. Proteins were transferred overnight onto nitrocellulose membrane at 60V. The membranes were blocked in either 5% BSA in TBS-T or

128

5% milk in PBS-T for 1 hour at room temperature. The membranes were then incubated with primary antibodies (1:1000 to 1:5000 dilution depending on antibody) overnight at 4°C. The following day, the membranes were washed 3 times for 5 minutes and incubated with HRP- conjugated secondary antibodies (1:5000) for 1 hour at room temperature. The membranes were then washed 3 times for 5 minutes. To visualize protein, either SuperSignal West Pico

Chemiluminescent Substrate (used for loading controls) or SuperSignal West Femto Maximum

Sensitivity Substrate (used for all other proteins) were applied to the membrane as evenly as possible. Shortly thereafter the membranes were exposed and imaged on a Biorad Gel Doc XR system. The resulting images were quantified using BioRad ImageLab software.

3.5.11 Mycoplasma testing

All IPSC, NPC, IMR90 and HEK293T cell lines were shown to be mycoplasma-free every

2-3 months using a standard method, (Otto et al. 1996) except Platinum Taq was used for the

PCR reactions.

3.5.12 Immunocytochemistry

IPSCs and NPCs were fixed in 4% PFA in PBS for 7 minutes at room temperature. The cells were then washed 3 times for 5 min with PBS and permeabilized with 0.1% Triton-X in

PBS for 10 minutes and blocked for 1 hour in 10% donkey or goat serum in PBS (blocking solution). After blocking, the cells were incubated with primary antibodies in blocking solution overnight at 4°C. The cells were then washed with PBS and incubated with secondary antibodies

129 in blocking solution for 1 hour. The cells were then washed three times in PBS or incubated with

DAPI (1:1000) in PBS and washed, if the experimental protocol required it.

Neurons were fixed by 1 min immersion in 37°C PKS (4% PFA in Krebs-Sucrose buffer, 50 mM

KCl, 1.2 mM CaCl2, 1.3 mM MgCl2, 20 mm HEPES pH 7.4, 12 mM NaH2PO4, 400 mM sucrose, 145 mM NaCl, 10 mM glucose in water) to preserve fine structure (Kwiatkowski et al.

2007) followed by 10 min in ice cold methanol. PKS was added immediately after removal of culture medium, and pipetting was done slowly to prevent dislodging cells from coverslips. The neurons were then washed 3 times for 5 min with PBS and blocked with blocking solution for 1 hr at room temperature. After blocking, the cells were incubated with primary antibodies diluted in blocking solution overnight at 4°C. The cells were washed 3 times for 5 min with PBS and incubated with secondary antibodies at room temperature for 1 hour. If the goal of staining was to visualize synaptic puncta, all antibody solutions were mixed thoroughly and spun in a centrifuge at max speed for 3 min to remove antibody clumps. Coverslips were rinsed in sterile water, briefly dried with a Kimwipe and mounted on a frosted glass slide with 7-8 µL of Prolong

Gold Anti-Fade mounting medium. The mounted coverslips were cured in the dark at room temperature for at least 24 hours before imaging.

3.5.13 Image acquisition

For simultaneous morphometric and synaptic connectivity analysis, we acquired images of neurons at 400x magnification using the Nikon A1R laser scanning confocal microscope. First, we identified neurons by tracking axons to their origin. To take images of entire neurons, we acquired 4x4 tiled z-stacks at 0.75 µm spacing spanning 4.5 µm total. The pinhole was widened to 2.6 Airy units, which increased speed of image acquisition by allowing to increase Z-spacing

130 and lower pixel dwell time without sacrificing Z-resolution to the extent that would compromise co-localization of synapsin I puncta to neurites. Each tile was 512x512 pixels, with 10% overlap, bringing the total field of view that was acquired to ~760 by 760 µm (0.4 µm / pixel). Image acquisition was done in a semi-automated manner, with manual picking of individual neurons and batch acquisition. Individual channels were acquired in sequential manner to prevent bleed- through of fluorophores. We aimed to acquire at least 7 neurons per color per coverslip, acquired in separate batches. Three channels would be acquired per batch for Synapsin I, GFP or mKO2.

Image acquisition was done in a blinded manner. While most neurons would be acquired successfully with the above setup, we omitted those neurons with neurites that either extended in x,y, or z direction beyond the 760 x 760 x 4.5 µm space, encountered regions of high cell density which resulted in blurriness, or overlapped significantly with other fluorescent nearby.

To image iPSCs, embryoid bodies, as well as acquire phase-contrast images of live neurons, a Leica DMI4000B microscope with a DFC 340 FX camera was used. To acquire images of fluorescent neurons co-stained with cortical and synaptic markers, Zeiss Axiovert

200M epifluorescence microscope with a Hamamatsu C9100-13 EM CCD camera was used. To acquire and quantify images of NPCs and unlabeled neurons (Figure 26), the high-throughput imaging platform ThermoFisher Cellomics VTi was used.

3.5.14 Synapse and morphology analysis

All analyses were completed in Fiji. ND2 files from the Nikon NIS Elements software were imported using the Bio-Formats plugin with channels pseudocolored such that the fluorescent protein and synapse channels were always colored in the same way from image to image to prevent the experimenter from inadvertently deciphering the genotype of the cells. The

131 analyses were completed on maximum intensity projections of z-stacks. For morphological analysis, the longest neurite was assumed to be an axon and not traced. Dendrites were traced and morphology was analyzed using the Simple Neurite Tracer plugin to obtain measurements for total dendrite length and Sholl analysis with 10 µm radius. Total synapse number was quantified by counting the number of Synapsin I puncta overlapping with signal from the fluorescent protein using the Cell Counter plugin. Images in which portions of the neurons could not be unambiguously distinguished, either due to blurriness or to excessive were not analyzed.

All image analysis was completed in a blinded manner.

3.5.15 Statistical analysis

3.5.15.1 Sparse-seeding co-culture.

For development of the method, we aimed to characterize the variability in connectivity across batches and wells. Because we could not derive an estimate for this variability in advance, we characterized 480 neurons for sEPSCs and 409 neurons for total synapse number across 48 and 39 wells, respectively. Probability distributions were fitted to the data using the fitdistrplus R package. Goodness-of-fit was evaluated using the Anderson-Darling test in the kSamples R package and the distribution with the lowest Akaike information criterion score was considered best fit. To control for well-to-well variability, measurements of all cells in a given well were normalized to the control cells in that well. For synapse number, dendrite length, sEPSC frequency and amplitude, normalization was done by dividing the values of all cells by the geometric mean of the controls. Anderson-Darling k-samples test was used to compare pooled samples that were co-cultured together. For Sholl analysis, normalization was done by subtracting the mean number of crossings of the controls from that of a given cell in the same

132 well at a particular radius. The change in crossings at a particular radius was then compared.

Two-way ANOVA was used to compare Sholl analysis curves between conditions. P-values of less than 0.05 were considered statistically significant. All analyses were performed with R. R

Code, unnormalized and normalized data are provided as supplementary files.

3.5.15.2 Other experiments

When two groups were compared, two-tailed Student’s t-test was used. P-values of less than 0.05 were considered statistically significant. Where suitable, analyses were completed in either R or GraphPad Prism.

3.5.16 Power Simulations

To determine which statistical tests were most sensitive and whether normalization improved sensitivity of those tests, we performed power simulations across a range of sample sizes. Conditions reaching a power of 0.8 with lower n were considered more sensitive. The following procedure was repeated for each of the four generated datasets (unnormalized synapse number and dendrite length, within-well normalized synapse number and dendrite length, unnormalized sEPSC frequency and amplitude, normalized sEPSC frequency and amplitude).

Briefly, control cells from a given dataset grown in ND + AA were sampled with replacement at sample sizes ranging from 5 to 180 in steps of 5. The values of sampled control cells were then multiplied by a fold-change factor to obtain hypothetical values for mutant cells (1.5-fold for synapse number and sEPSC frequency, 1.25-fold for dendrite length and sEPSC amplitude).

Statistical tests (Anderson-Darling, Kolmogorov-Smirnov, Student’s t-test) were then used to

133 compare the samples. The procedure was repeated 1000 times at each n, and the proportion of tests reaching significance (i.e. power) at ⍺ = 0.05 was recorded.

3.5.17 Electrophysiological recordings

Conventional patch-clamp recordings were made, as described previously(W.-B. Zhang et al. 2016), in human iPSC-derived neurons after 9-11 weeks in culture at room temperature

(22°C), and experimenter was blinded to genotype in electrophysiological recordings and data analysis. An Axopatch 1-D amplifier (Molecular Devices, USA) and a DigiData 1200 series interface (Molecular Devices, USA) were used for patch-clamp recordings. Electrical signals were sampled at 10 kHz and filtered at 2 kHz. Recording pipettes with resistance of about 5 to 8

MΩ were pulled on a P-87 pipette puller (Sutter Instrument Co., USA) using capillary glass

(World Precision Instruments, Inc., USA). The pipettes, for the recordings on intrinsic membrane properties, were filled with a solution that contained (in mM): 144 K+-gluconate, 10 KCl, 2

EGTA, 10 HEPES, and 2 Mg-ATP, and pH was adjusted to 7.2 with KOH. Extracellular recording solutions were composed of (in mM): 140 NaCl, 1 MgCl2, 5.4 KCl, 10 glucose, 15

HEPES, and 2 CaCl2, and pH was adjusted to 7.35 with NaOH. Under current-clamp condition, a liquid junction potential of 16 mV with K+-gluconate based internal solutions was subtracted from the measured membrane potential values. Action potentials from the membrane potentials of approximate -75 mV were evoked by the injection of a series of current steps from -5 pA to

+50 pA (in 5 pA increments) for 1 s. Under voltage-clamp condition, membrane potentials were

134 held at -70 mV and voltage-gated currents were elicited by depolarizing the membrane potentials to a series of potentials from -80 mV to + 60 mV for 400 ms. sEPSCs were recorded in voltage-clamp condition with the holding membrane potentials of -60 mV. Extracellular recording solutions were composed of (in mM): 140 NaCl, 15 HEPES, 5.4

KCl, 25 glucose, 0.003 glycine, 0.001 strychnine, 0.01 bicuculline, 1 MgCl2, and 1.3 CaCl2, and pH was adjusted to 7.35 with NaOH. Recording pipettes were filled with a solution that contained (in mM): 137 CsF, 1.5 CsCl, 10 BAPTA, 10 HEPES, 5 QX-314, and 4 Mg-ATP, and pH was adjusted to 7.2 with CsOH. Detection and analysis of sEPSCs were performed by Mini

Analysis Program in a blinded manner (Synaptosoft Inc, NJ, USA).

Statistical analysis was performed in R. All results were given as mean ± SEM, and “n” stands for the number of neurons recorded. Statistical comparisons of data were performed by Student t- test, or one-way ANOVA, as appropriate, and difference was considered to be significant when

P < 0.05.

3.5.18 Code

R package used include: data.table, tidyr, purrr, dplyr, stringr, ggplot2, ggthemes,

RColorBrewer, scales, psych, kSamples, fitdistrplus.

135

Table 6. Summary of experiments performed. *Excerpted from manuscript by Ross et al. Unr

= Unrelated, Pat = paternal, Mat = Maternal. iPSC: induced pluripotent stem cell; NPC: neural precursor cell; ICC: immunocytochemistry; sEPSC: spontaneous excitatory postsynaptic current.

136

Table 7. Batches of neurons used in sparse coculture experiments. ICC: immunocytochemistry; sEPSC: spontaneous excitatory postsynaptic current.

137

Primary Antibodies

Item Supplier Cat # ICC WB

Concentration Concentration

Chicken anti-GFP ThermoFisher A10262 1:1000

Rabbit anti-GFP ThermoFisher A6455 1:1000

Mouse anti-mKO2 Amalgaam M168-3 1:400

Guinea Pig anti-MAP2 Synaptic Systems 188 004 1:1000

Mouse anti-TUBB3 Chemicon MAB1637 1:200 1:2000

Rabbit anti-Synapsin I Millipore AB1543P 1:250 1:1000

Sheep anti-SHANK2 R&D Systems AF7035 1:1000

Mouse anti-V5 ThermoFisher 37-7500 1:1000

Mouse anti-SATB2 Abcam ab51502 1:250

Goat anti-BRN2 SantaCruz Sc-6029 1:500

Rabbit anti-TBR1 Abcam ab31940 1:250

138

Rat anti-CTIP2 Abcam ab18465 1:100

Rabbit anti-LMNB1 Abcam ab16048 1:1000

Mouse anti-vGlut1 Synaptic Systems 135 511 1:1000

Goat anti-SOX1 R&D Systems AF3369 1:200

Mouse anti-AP2 DSHB 3B5-c 1:200

Goat anti-OTX2 R&D Systems AF1979 1:200

Rabbit anti-PAX6 BioLegend PRB-278P 1:2000

Mouse anti-Nestin Millipore AB5326 1:200

Mouse anti-actin Sigma A5316 1:2000

Mouse anti-SSEA4 Invitrogen 41-4400 1:100

Mouse anti-TRA1-60 Invitrogen 41-4000 1:100

Rabbit anti-NANOG Cell Signaling 4903P 1:400

Rabbit anti-OCT4 Abcam Ab19857 1:200

Mouse anti-SMA Invitrogen 18-0106 1:200

139

Mouse anti-AFP R&D Systems MAB1368 1:200

Secondary Antibodies

Item Conjugated Supplier Cat # ICC WB

Donkey anti-rabbit Alexa 488 ThermoFisher A21206 1:500

Donkey anti-goat Alexa 488 ThermoFisher A11055 1:500

Donkey anti-mouse Alexa 488 ThermoFisher A21202 1:500

Donkey anti-rat Alexa 488 ThermoFisher A21208 1:500

Donkey anti-goat Alexa 555 ThermoFisher A21432 1:500

Donkey anti-sheep Alexa 555 ThermoFisher A21436 1:500

Donkey anti-rabbit Alexa 555 ThermoFisher A31572 1:500

Donkey anti-mouse Alexa 555 ThermoFisher A31570 1:500

Donkey anti-rat DyLight 550 Pierce SA5-10027 1:500

140

Donkey anti-mouse Alexa 647 ThermoFisher A31571 1:500

Donkey anti-sheep Alexa 647 ThermoFisher A21448 1:500

Donkey anti-rabbit Alexa 647 ThermoFisher A31573 1:500

Donkey anti-goat Alexa 647 ThermoFisher A21447 1:500

Donkey anti-guinea pig Alexa 647 JacksonImmuno 706-645- 1:100

148

Donkey anti-rabbit DyLight 405 JacksonImmuno 711-475- 1:100

152

Donkey anti-chicken Alexa 488 JacksonImmuno 703-545- 1:100

155

Donkey anti-sheep HRP ThermoFisher A16047 1:5000

Goat anti-rabbit HRP ThermoFisher 65-6120 1:5000

Goat anti-mouse HRP ThermoFisher 626520 1:5000

141

General immunocytochemistry

Item Supplier Cat #

Normal Donkey Serum Millipore 530

Triton X-100 Sigma P9284

Tween 20 Sigma P1379

Paraformaldehyde, 16% Alfa Aesar 43368

Prolong Gold Antifade Mountant ThermoFisher P36934

Frosted Glass Slides VWR 16004-368

Methanol Sigma 179337-4L

Neuronal Differentiation

Item Supplier Cat #

Neurobasal Invitrogen 21103-049

N2 Invitrogen 17502-048

B27 Invitrogen 12587-010

BDNF Peprotech 450-02

142

GDNF Peprotech 450-10

Cyclic AMP Sigma D0260

Ascorbic acid Sigma A4403

SB-431542 StemGent 04-0010-10

Dorsomorphin StemCell Technologies 72102

XAV939 Sigma X3004

Poly-L-ornithine Sigma P8638

Laminin Roche 11243217001

DAPT Sigma D5942

143 iPSC Reprogramming and Culture

Item Supplier Cat #

Cytotune iPS Kit ThermoFisher A16518

Mouse Embryonic Fibroblasts Hospital for Sick

E12.5 Children (Rossant Lab

Human ES Cell

Facility)

Knockout Serum Replacement Invitrogen 10828028

Matrigel Corning 354277

ReleSR Stemcell Technologies 05873

-mercaptoethanol Invitrogen 21985023

mTeSR StemCell Tech 05851

StemMACS iPS Brew XF Miltenyi 130-107-086

Stem Cell Nucleofector Kit I Amaxa VPH-5012

144

General Cell Culture

Item Supplier Cat #

HEK293T cells ATCC CRL-3216

IMR-90 fibroblasts ATCC CCL-186

DMEM Invitrogen 11965-092

DMEM/F12 Invitrogen 11330-032

Knockout DMEM/F12 Invitrogen 12660-012

Accutase Innovative Cell Tech AT-104

Inc.

Dispase StemCell Technologies 07923

Trypsin 0.25% Invitrogen 25200-056

Basic fibroblast growth factor Peprotech 100-18C

Opti-MEM ThermoFisher 31985088

Lipofectamine 2000 ThermoFisher 11668019

Non-essential amino acids Invitrogen 11140050

145

GlutaMAX Invitrogen 35050061

Penicillin/streptomycin Invitrogen 15140-122

Fetal Bovine Serum Invitrogen 10439-024

G418 ThermoFisher 10131035

Puromycin Sigma P8833

Phosphate Buffered Saline (- WISENT 311-010-CL

Mg2+, -Ca2+)

Sterile Water WISENT 809-115-CL

DMSO Sigma D2650

40 m cell strainer Falcon 352340

70 m cell strainer Falcon 352350

0.2 m filter Sarstedt 83-1826-001

Coverslip 12mm Fisher Scientific 1943-10012A

146

General Molecular Biology

Item Supplier Cat #

Phusion Polymerase NEB M0530S

Q5 polymerase NEB E0555S

Gibson cloning kit NEB E5510S

Plasmid Prep Kits ThermoFisher K210007

BOLT mini gel tank ThermoFisher A25977

4-12% bis-Tris gels ThermoFisher NW04122BO

Nitrocellulose membrane GE Healthcare RPN203D

HyBond GCL

MES Buffer 20x ThermoFisher B0002

SuperScript III RT ThermoFisher 18080044

DNase I, Amplification grade ThermoFisher 18068015

SYBR Select PCR Master Mix ThermoFisher 4472908

147

dNTP ThermoFisher 18427013

Random hexamer ThermoFisher 48190011

Platinum HiFI Taq Invitrogen 10966-018

Complete Mini Protease Roche 11 836 170 001

Inhibitor, EDTA-free

Plasmids

Item Source Cat #

pCas9D10A_GFP Addgene 44720

gRNA Cloning Plasmid Addgene 41824

pLenti-CaMKII-GFP Gift from Drs.

Shcheglovitov &

Dolmetsch

pLenti-CaMKII-mKate2 Gift from Drs.

Shcheglovitov &

Dolmetsch

148

pLenti-CaMKII-mKO2 Constructed in lab

(Rodrigues et al. 2016)

pMXs-hOCT4 Addgene 17217

pMXs-hKLF4 Addgene 17219

pMXs-hSOX2 Addgene 17218

pMXs-hcMyc Addgene 17220

PmaI-SHANK2 Constructed in lab

Oligos / guide RNAs

Target/Purpose Sequence of FWD primer Sequence of REV primer

gRNA SHANK2 CACGGACTCCAGGTACTGTAGGG

ex 16 – 1

gRNA SHANK2 TGGCGTGGCAAGCCGGACTAAGG

ex16 – 2

149

SHANK2 ex 16 ACACACCCATTGAAGAATTCACACCAACACCGGCTTTCCCAGC

Knockout tag (V5 CCTACAGTACCTGGGGAAACCCATTCCCAATCCCCTTCTTGGGC

– Mre1 – 3x TTGATTCAACGCCGGCGTAACTAGCTGAGGGGTGGCGTGGCAA

STOP) ssODN; GCCGGACTAAGGACCGGGGACTTCTTGATTGAGGTAGGGACAC black – homology AGGTGTCTGTAT

arms

SHANK2 ex16 - CACCCATTGAAGAATTCAC GCTCTCCCCAAGCAGAAAGA

ddPCR ACCA

Fam-SHANK2- TGGAGTCCGTGGATG

Vic-tag (for CAACGCCGGCGTAACTA detecting SHANK2

KO alleles)

CACNA1C-ddPCR CTGAGCACATCCCCACCC GGTTTCCCATATTGCTGCCG gRNA CACNA1C CGATGGCCGCCTGCCACGA GCCCGGCAGGCTAAGCTGATGG

CAGG G

150

Fam-CACNA1C- AGGCGGCCATCGACG

AFF2-ddPCR AGCAACAGAAAATCAAAA CCTGTGTCAAAGTCTGCATCTTG

CCTGAGT

gRNA AFF2 TGACTATCACGTGACCACT TAGCACTGTACTGGCAAGCCAG

CAGG G

Fam-AFF2-WT TCACGTGATAGTCATAAC

OCT4 TGTCTCCGTCACCACTCTGG GTTCCCAATTCCTTCCTTAGTG

KLF4 CGGACATCAACGACGTGAG GACGCCTTCAGCACGAACT

(primerBank)

CMYC GGCTCCTGGCAAAAGGTCA CTGCGTAGTTGTGCTGATGT

(primerBank)

SOX2 TACAGCATGTCCTACTCGC GAGGAAGAGGTAACCACAGGG

(primerBank) AG

pMXs-OCT4 TGTCTCCGTCACCACTCTGG TCCCCCCTTTTTCTGGAGAC

151

pMXs-KLF4 GACCACCTCGCCTTACACA GCGCTCAGCTGGAATATCACC

pMXs-CMYC ACACAAACTTGAACAGCTA GCGCTCAGCTGGAATATCACC

CGG

pMXs-SOX2 CCCAGCAGACTTCACATGT CCCACCCTTTCACATGTGTG

SHANK2_commo AAGCCGGTGTTGGTGTGAA AAGCCGGTGTTGGTGTGAATTC

n (PDZ) TTC

MYCO-A GGCGAATGGGTGAGTAACA CGGATAACGCTTGCGACCTATG

RTb-actin CCACTGCCGCATCCTCTTCC CTCGTTGCCAATAGTGATGACCT

Table 8. Materials used.

Chapter 4 Discussion and Future Directions

In my studies, I hypothesized that altered transcription and synapse function contribute to the

Neurodevelopmental disorders WBS and ASD. Here, I discuss how they inform our understanding of NDD biology, how the findings compare with those from post-mortem studies and animal models, what experimental challenges were encountered and how they were overcome. In particular, I explore what are ideal controls for IPSC-based modeling studies, the impact of line-to-line heterogeneity on experimental results, and how differences in experimental methods such as 3D organoids and direct conversion methods may affect interpretations of synaptic connectivity in different studies. I conclude by discussing the potential and key challenges which iPSC model systems must overcome towards becoming a platform for investigation of NDD biology and drug screening.

152 153

4.1 Transcriptome analysis in IPSC-derived neurons of NDDs

4.1.1 WBS neuronal transcriptome - differences with published datasets

WBS is severe, affects multiple organ systems, and is caused by a hemizygous deletion of 28 genes, at least five of which regulate transcription. Therefore, it is reasonable to hypothesize that the effect of the deletion on transcription would be similar in magnitude to changes caused by other multigenic deletions associated with other neuropsychiatric disorders, such as 22q11.2 (Lin et al. 2016), 16p11.2 (Arbogast et al. 2016), and therefore robustly observed across different experimental settings. Recent studies provide some evidence for this idea, as they converge on a small set of biological pathways that could be underlying WBS, but also significantly diverge in their findings overall.

4.1.1.1 Studies on WBS human IPSC-derived neurons

To investigate transcriptional alterations in WBS, I compared microarray data generated from

6 week-old neurons from one WBS patient (3 lines) and one control (1 line). Though the sample size was small, I was able to show that the control line was representative of controls neurons generated in another lab and make several conclusions that were validated by qRT-PCR. In particular, I found that most genes were downregulated in WBS neurons. I found a decrease in expression of multiple voltage-gated potassium genes which could explain the observed deficit in action potential firing in WBS neurons. Overall, gene sets governing synaptic, axonal and extracellular matrix function were enriched. These data were the first to be published to describe alterations in neuronal function in WBS. One possible explanation is that the decrease in expression of these gene sets is due to delayed maturation of WBS neurons. However, a recent single cell RNA-seq study of developing IPSC-derived neurons (X. Chen et al. 2016) had little

154 overlap with our set of differentially expressed gene sets save for alterations in extracellular matrix genes. Chen et al. describe differences in mitochondrial electron transport chain as the most predictive of neuronal maturation, which were not altered in our dataset. In addition, gene sets normally upregulated in immature neurons, such as those involved in axon guidance, were downregulated in our samples. Lastly, our single cell qRT-PCR analyses aimed at quantifying expression of markers associated with different stages of development, different neuronal layers, and subtypes, revealed no difference.

In addition to the microarray dataset I analyzed for Chapter 2 (Khattak et al. 2015), three additional studies used iPSCs to investigate transcriptional alterations in WBS. Chailangkarn et al. used a 2D differentiation method similar to the one used in Chapter 2 to compare WBS NPCs and neurons with typically-developing (TD) controls (Chailangkarn et al. 2016). Lalli et al. used a direct conversion protocol which relied on overexpression of NEUROD1 and analyzed IPSCs and neurons (Lalli et al. 2016). Lastly, Adamo et al. compared transcriptomes of IPSCs and

NPCs from WBS and 7q11.23 duplication patients with that of controls, and found several gene sets altered at these early developmental stages (Adamo et al. 2015). Taken together, two studies compared transcriptome in IPSCs (Adamo et al. and Lalli et al.), two studies in NPCs (Adamo et al. and Chailangkarn et al.), and three studies in neurons (Khattak et al., Chailangkarn et al., and

Lalli et al.)

4.1.1.2 Dysregulation of extracellular matrix genes

While comparing neuronal gene expression findings between different studies (Khattak et al.

(Chapter 2), Chailangkarn et al., and Lalli et al.) yields few differentially expressed gene sets in common, several patterns emerge. In order to facilitate this comparison, I analyzed the lists of

155

DEGs provided by Lalli et al. using g-profiler (Reimand et al. 2016) to determine the full spectrum of down- and up-regulated gene sets, as Lalli et al. report only the top four. Neuronal samples generated using similar protocols in Khattak et al. and Chailangkarn et al. exhibit a modest downregulation in genes regulating the extracellular environment. However, no such differences are observed in Lalli et al.’s samples. Given that the method used by Lalli et al. generates pure populations of neurons by direct conversion of IPSCs, these findings suggest that mature WBS neurons do not have differences in expression of extracellular matrix genes.

Rather, the observed differences in Khattak et al. and Chailangkarn et al. could stem from residual NPCs present in the heterogeneous neuronal cultures generated by directed differentiation protocols. Regulation of cell adhesion and extracellular environment can provide important cues to survival and differentiation and may represent an additional mechanism underlying reduced proliferation of NPCs in WBS, which is partially explained by lowered

FZD9 expression and Wnt signaling in Chailangkarn et al. Alternatively, this difference could represent a failure of direct conversion methods to faithfully recapitulate neuronal development and thus omit stages at which differences in extracellular matrix genes would become apparent between WBS and TD neurons. There are several lines of evidence that such developmental compression is occurring. First, directly-converted neurons become synaptically active at three weeks of differentiation, roughly twice as fast as neurons generated using directed differentiation. Second, synaptic activity is exclusively AMPA-receptor mediated, with little expression of NMDA receptors (Yingsha Zhang et al. 2013), representing an inversion of the typically-accepted chronology of synaptic development (Hall et al. 2007).

It is therefore necessary to determine which cell population is responsible for alterations in extracellular matrix genes in WBS. While the above studies examine alterations in NPCs and

156 neurons, an important contribution may also come from astrocytes, which would have been present in neuronal cultures from Chapter 2 and Chailangkarn et al. Astrocytes can promote or inhibit dendrite extension and synaptogenesis by extracellular interactions (Eroglu & Barres

2010). It is likely that complex interactions between multiple cell types are ultimately responsible for the WBS phenotype, and employing new differentiation or cell sorting methods to separately study the contribution of each is necessary for future studies (Yuan et al. 2011; Ye

Zhang et al. 2016).

4.1.1.3 Dysregulation of synaptic genes

One of the largest collections of downregulated gene sets in WBS neurons in Chapter 2 was related to synaptic function. In addition to the downregulation of WBS region genes (STX1A,

LIMK1), I also found reductions in expression among genes related to synapse assembly, synaptic transmission, synaptic vesicles, and ion channel complexes.

Similar to what I observed, directly converted neurons by Lalli et al. had downregulated synaptic function gene sets. It is difficult to ascertain whether neurons from Chailangkarn et al. had the same deficit. While no synapse-related gene sets are enriched in Chailangkarn et al., single-cell qRT-PCR of 6 week-old neurons showed decreases in Synapsin I, which is common to 99% of mammalian synapses (Micheva et al. 2010), GRIN1, an essential subunit of NMDA receptors, and the GRIA2 subunit of AMPA receptors, suggesting that excitatory neurotransmission could be diminished. However, multi-electrode arrays (MEAs) and calcium imaging in the same study revealed increased activity in WBS neuronal cultures, and imaging analyses showed longer dendrites and higher dendritic spine density in WBS neurons compared to controls, all suggestive of a functional enhancement of excitatory synaptic function in WBS. It

157 can therefore be difficult to reconcile the apparent increase in excitatory synaptic activity with a concurrent decrease in expression of synaptic genes.

Several factors can potentially account for this discrepancy. First, the differences in protocols used (Lalli et al.) or lowered sample size (Chailangkarn et al. compared 4 controls vs. 4 WBS patients, more than either in Lalli et al. or in Chapter 2) could be partly responsible. Second, it is possible that increased activity detected via MEAs and calcium imaging is related to altered intrinsic membrane properties increasing the likelihood of spontaneous neuronal firing. This is unlikely as Chailangkarn et al. did not detect differences in intrinsic membrane properties, and in

Chapter 2, Khattak et al. discovered action potential deficits in WBS neurons, which would dampen spontaneous excitatory activity. An MEA experiment blocking AMPA-receptor dependent neurotransmission (e.g. with CNQX) could distinguish whether the increase in neuronal firing is due to spontaneous action potential generation or network activity. Third, the increased neuronal activity could indicate a relative absence of GABAergic interneurons in the

WBS neuronal cultures of Chailangkarn et al. This is has a higher likelihood of being true, as lowered Wnt signaling in WBS neuronal cultures (Chailangkarn et al.) could promote increased generation of excitatory neurons by increasing the proportion of PAX6+ neural precursors (Mi et al. 2013), as shown by single-cell qRT-PCR analysis of NPCs in Chailangkarn et al.

Furthermore, while persistently low Wnt can later bias NPCs towards ventralization and generation of GABAergic inhibitory neurons (D.-S. Kim et al. 2014), increased apoptosis in

WBS cultures may prevent this from occurring in appreciably large numbers.

Taken together, the downregulation of synaptic genes I discovered in Chapter 2 is paralleled in the only two other studies on WBS neurons. However, it is at present difficult to reconcile

158 reductions in synaptic genes with an apparent increase in excitatory synaptic activity observed by

Chailangkarn et al.

4.1.1.4 Alterations in axon-associated genes

Genes governing axonal function were downregulated in WBS neurons described in Chapter

2, a finding later replicated by all follow-up studies and detected at multiple stages of differentiation. The WBS deletion region contains LIMK1, which regulates neuronal maturation, neurite outgrowth, and synaptic plasticity (Cuberos et al. 2015). The transcriptomic alterations in axonal gene expression suggest that additional pathways may be involved, and thus suggest that axonal growth could be additionally altered. However, axonal biology is a relatively unexplored area of WBS pathophysiology, and studies of axons in human iPSC-derived neurons, as well as in the new mouse model bearing a heterozygous WBS deletion (Segura-Puimedon et al. 2014), could be informative. These mice recapitulate the decrease in brain size observed in WBS post- mortem brains and have impairments in excitatory synaptic plasticity (Borralleras et al. 2016).

4.1.1.5 Summary

Taken together, the current human neuronal studies converge on very early processes in neuronal development in WBS. Dysregulation of extracellular matrix genes in NPCs and lowered

Wnt signaling may alter the developmental trajectory of WBS neurons. Concurrent alteration of genes regulating axonal development can then predispose developing neurons to wire inappropriately and alter neuronal network function. The striking concordance of morphological and dendritic spine abnormalities in WBS IPSC-derived neurons and post-mortem WBS brains

159 observed by Chailangkarn et al. suggests that these processes may underlie altered brain development and function in WBS. However, it is likely that this is only the tip of the iceberg, as transcriptomic heterogeneity arising from human genetic variation and heterogeneity among

IPSC-derived neuronal cultures only allows discovery of transcriptomic alterations of high effect size (Adamo et al. 2015). Increasing sample sizes, as well as taking steps to reduce heterogeneity among human stem cell-derived neuronal cultures subject for transcriptomic analysis should improve discovery efforts.

4.1.2 PTCHD1AS in ASD neuronal transcriptome

Deletions at the X-linked PTCHD1/PTCHD1AS locus affect 0.5% of all ASD cases, and disruption of either the protein-coding gene PTCHD1 or the divergently transcribed lncRNA

PTCHD1AS increase the risk of ASD. While the function of these products is unknown, IPSC- derived neurons from these subjects have impaired excitatory neurotransmission, and I discovered a set of transcriptional alterations in cases that can be attributed to PTCHD1AS. To improve discovery of DEGs, we made several improvements to the experimental design relative to the WBS study described above. First, we increased the sample size and used 4 lines from 2 controls and 6 lines from 2 ASD subjects. To investigate which transcriptomic changes could be attributed to compromised PTCHD1AS function, we used neurons from an ASD subject with a mutation in both PTCHD1 and PTCHD1AS, and a subject with a deletion restricted to

PTCHD1AS-exon3. Second, we used IPSC-derived neurons derived from the mother of the

PTCHD1/PTCHD1AS subject as one of the controls, which could reduce the impact of human genetic variation. Third, we used a novel magnetic activated cell sorting (MACS) protocol to

160 purify neurons to enable comparisons uncompromised by the presence of variable numbers of

NPCs or glial cells among neuronal cultures.

4.1.2.1 Discordance in DEGs among probands

To identify genes that are misregulated in common between the two probands, I compared expression in each case and then compared the lists of DEGs. Those DEGs that are found in common between the two cases are potential candidates for regulation by PTCHD1AS. While I detected multiple DEGs for each proband, the overlap was restricted to a very small number. In particular, I identified downregulation of zinc finger-containing genes. While the functions of most of these genes are relatively unknown, ZNF300 is expressed during early brain development, and ZNF667 suppresses apoptosis (L. Jiang et al. 2014). In addition, most of the downregulated zing finger proteins were situated on chromosome 19. This suggests a possible function for PTCHD1AS in trans regulation of these genes, which could occur due to 3D conformation of the genome that would bring the two loci together (Kung et al. 2013). Another possibility that cannot be ruled out is that DDX53 is responsible for the observed phenotype.

DDX53 located in close to PTCHD1/PTCHD1AS and is deleted in the case with the PTCHD1AS- exon3 deletion. It is expressed in control neurons, but diminished or absent in neurons from both

ASD cases, suggesting that transcription through the PTCHD1/PTCHD1AS locus may be necessary for its expression. While the function of DDX53 is unknown, it is a member of the

DEAD box protein family, which is involved in all aspects of RNA metabolism (Cordin et al.

2006). As such, the alterations observed in common between the two cases could stem from

DDX53 dysregulation. Generation of isogenic knockouts specific to DDX53 or PTCHD1AS may help to test this hypothesis.

161

4.1.2.2 PTCHD1AS dysfunction could lead to the disruption of RNA export, RNA splicing, Cellular Respiration, and Extracellular Matrix genes.

Gene set enrichment analysis revealed four clusters of gene sets in common between the two subjects: downregulation of RNA export, RNA splicing, Cellular Respiration, and upregulation of Extracellular Matrix genes. Most genes in the downregulated gene sets had relatively small but statistically significant and consistent downward deviations in transcript abundance.

Alternatively, most upregulated ECM genes had large effect sizes, but were variable across samples. The overlap in these genes among ASD cases suggests that PTCHD1AS could be responsible for subtly tuning expression of multiple genes, leading to altered function. These effects are likely to be secondary as PTCHD1AS is present at very low abundance in neurons and is localized to the nucleus. It is possible that its primary effect is on expression of zinc finger proteins, which could cause widespread alterations in transcription and alternative splicing. Yet another possibility is that transcription through of PTHCD1AS is necessary for proper expression of DDX53. Given DDX53’s possible roles in RNA metabolism, its diminished expression of

DDX53 in PTCHD1/PTCHD1AS-mutant cases may underlie a part of the observed transcriptional alterations.

The lack of additional gene sets unique to the subject with a deletion of both PTCHD1 and

PTCHD1AS argues against a significant role for PTCHD1 in transcription. Consistent with this idea, despite the fact that PTCHD1 contains a putative sonic-hedgehog binding domain and causes weak suppression of hedgehog signaling in a medulloblastoma cell line (Chung et al.

2014), I found no alterations in the hedgehog signaling in IPSC-derived neurons of the

PTCHD1/PTCHDAS case.

162

At first glance, it can be somewhat perplexing that these analyses did not reveal changes in synaptic gene expression, given the decreases in excitatory synapse function. While using MACS allows for comparison of pure neurons seeded at the comparable densities, it is currently restricted to neurons that are relatively young. MACS is carried out on 3 week-old neuronal cultures in order to maximize survival during the procedure. Following reseeding, neurons are grown for an additional week before samples are collected, as current culture protocols cannot prevent purified neurons from clumping and lifting off the cell culture plate beyond approximately 2 weeks. Therefore, the lack of differences in synaptic gene expression could be due to low expression of synaptic genes in 4 week-old neurons. In contrast, 12-16 week old neurons co-cultured with astrocytes for electrophysiological assays are expected to have robust expression of synaptic genes, such that differences in excitatory synapse function could be detected. In addition, sensitivity of the experiments can be improved by switching to RNA- sequencing from microarrays (Zhao et al. 2014). In addition, RNA-seq can yield information about alternative splicing (Voineagu et al. 2011), which has been recently found to be altered in autistic postmortem brains and may be important for elucidating PTCHD1AS-dependent effects on the transcriptome.

4.1.2.3 Alterations in RNA splicing in PTCHD1/PTCHD1AS ASD

Downregulation of genes involved in RNA splicing suggests that alternative splicing (AS) may be dysregulated in the neurons of these ASD cases. This is consistent with a neurodevelopmental pathology of ASD and observed in multiple other studies. NLGNs and

NRXNs, high confidence ASD risk genes involved in synapse development, have been found to be aberrantly spliced in ASD (Smith & Sadee 2011). Blood drawn from boys with ASD shows

163 significant evidence of genome-wide AS alterations (Stamova et al. 2013). In addition, a conserved microexon AS splicing program regulated by NSR100 is also dysregulated in ASD

(Irimia et al. 2014). Voineagu et al. showed that genes underlying synapse development and function were particularly likely to be aberrantly spliced in post-mortem ASD brains. This may offer an additional explanation for the discordance between decreased excitatory neurotransmission and absent synapse function gene expression changes in microarray analyses.

While the human transcriptome array technically offers an opportunity to screen for potential alternative splicing events, it has an unacceptably high false positive rate (40% under the best conditions) (Seok et al. 2015). Recently, alterations in lncRNA AS have been observed in ASD, consistent with AS we observed in PTCHD1AS upon exon 3 deletion (Parikshak et al. 2016).

Alterations in alternative splicing may also be correlated with the observed deficits in mEPSC frequency, which implies hypoconnectivity of PTCHD1/PTCHD1AS neurons. A relative scarcity of synaptic connections may cause PTCHD1/PTHCD1AS neurons to be activated less than controls, and thus alter activity-dependent alternative splicing programs. Consistent with this idea, absence of NSR100, which regulates alternative splicing during neuronal development, is correlate with a decreased frequency of mEPSCs and alterations in activity-dependent alternative splicing (Quesnel-Vallières et al. 2016).

4.1.3 Importance of ECM in NDDs

One of the rather interesting findings, especially in contrast with WBS, is the increase in expression of extracellular matrix genes. WBS and ASD are thought to be opposites of each other in the domain of social communication, with WBS resulting in increased sociability and friendliness and ASD causing impairments in socialization. This is bolstered by the finding that

164

ASD can be caused by the duplication of the WBS deletion region, suggesting that dosage of

WBS region genes could be important for development of social communication. The opposing alteration in expression of ECM genes is an additional correlate of this divergence. While the functions of individual ECM genes in social communication specifically are relatively unexplored, some could be important for neuronal development. For example, tenascin C, downregulated in WBS and upregulated in ASD subjects with PTCHD1 locus mutations, is expressed by neurons and regulates proliferation of NPCs and glial differentiation. These findings suggest that interaction between neurons and glial cells could be an important aspect of pathophysiology of WBS and ASD and warrants further investigation. However, caution in interpretation of these findings from IPSC-derived neurons is warranted due to the potential presence of unknown cells types that may have an impact on ECM gene expression.

4.1.4 Towards an ideal gene expression experiment using IPSC-derived neurons

The studies described in Chapter 2, as well as others published in recent years, raise several points critical to carrying out gene expression experiments in IPSC-derived neurons. While the standard recommendations of increasing sample size and technical replicates apply, the choice of control lines and ensuring that neuronal cultures compared are as similar as possible in cell composition are particularly important for these experiments.

165

4.1.4.1 What is the ideal control line?

Given the extent of human genetic variation and its impact on transcription in IPSCs, the question of what constitutes an appropriate control line for gene expression experiment is critical. To accurately ascribe observed changes in transcriptome to a particular genotype, the impact of human genetic variation must therefore be addressed. First, gene editing technology such as CRISPR/Cas9 could be used to induce suspected disease-causing lesions in control lines or correct them in lines of affected cases to generate isogenic controls. This approach can be of high import in cases of highly-penetrant SNVs or small indels and has been successfully applied to detect alterations in presynaptic gene expression in IPSC-derived neurons with mutations in

DISC1, which can predispose to schizophrenia (Wen et al. 2014). This was accomplished in

Chapter 3 by generating isogenic SHANK2 knockout cells.

However, this may not be suitable for those NDD cases that are associated with large deletions that span multiple genes. In this scenario, the gene expression profile of the set of controls used will influence the results. This is a possible explanation for the divergence in gene expression findings among WBS studies using IPSC-derived neurons. Therefore, either the number of control individuals needs to be increased or the cell lines must be compared to an existing ‘reference transcriptome’, preferably both. Such a reference transcriptome should be generated from a large dataset of controls, preferably from pure neurons, and benchmarked to temporal progression brain development and subregional specification. Two efforts towards this goal have been recently published. Joyce van de Leemput et al. created the Cortecon database describing the temporal progression of gene expression in neurons generated from human embryonic stem cells. While it can serve as a very useful tool to explore developmentally- regulated patterns of gene expression, it is based on profiling a single human emrbyonic stem cell line that may not be representative of IPSC lines generated from the wider population. Stein

166 et al. arrive much closer to the mark by profiling expression of human cortical NPCs from multiple donors differentiated in vitro over the course of 12 weeks. By comparing these findings with transcriptome data from developing human brain samples, they are able to define pathways that are upregulated both in vitro and in vivo during human development and develop a framework for classifying neuronal samples by age and subregional specification based on their gene expression profile. Stein et al. go on to demonstrate that neuronal cultures derived by standard directed differentiation from IPSCs in different labs have either a cortical or cerebellar identity, reflecting substantial lab-to-lab variation in the kind of neuronal cultures generated.

Such variation could significantly bias interpretation of transcriptome results. Given the recent availability of these resources, it may be beneficial to compare the transcriptional profile of control neuronal cultures to determine which subregional specialization they are most consistent with before proceeding with analyses.

4.1.4.2 Purity of cells

A significant concern with human IPSC-derived neuronal cultures is heterogeneity in cell composition and density. NPCs can give rise to neurons, glial cells, and other NPCs. For example, constant generation of immature neurons by continuously dividing NPCs can introduce significant noise and bias gene expression profiles to reflect that of immature cells.

Unpredictable contribution of glial cells to different cultures can additionally complicate results by making it difficult to determine what fraction of transcriptomic changes are due to changes in proportions of different cell populations. Additionally, glial cells affect transcription in nearby neurons by secretion of trophic factors and thus create false positive findings. Lastly, cell density

167 can influence neuronal development and survival, introducing further noise in transcriptome measurements.

It is thus necessary to develop methods to compare purified cell populations that are similar in density. The MACS protocol used in Chapter 2 for studies of PTCHD1/PTCHD1AS-mutant cells solves this issue by allowing removal of NPCs and glial cells from 3-week old cultures by negative selection using cell surface markers CD44 and CD184. The subsequent reseeding also allows control of cell density. However, pure neurons tend to clump together following reseeding and lift off the plate after approximately 2 weeks, limiting analyses to early developmental stages. As such, further optimization of culture conditions is necessary in order to promote long- term survival.

As an alternative to MACS, it may be possible to infer which subpopulations of cells are present in a given sample. For example, the recently developed bioQC algorithm by Zhang et al. relies on more than 150 signatures of tissue-enriched genes to detect tissue heterogeneity (J. D.

Zhang et al. 2017). However, applying these methods to human stem cell-derived neurons has not yet been tested. It will be important to determine which possible populations of cells can be present in a neuronal culture of a certain age generated by a specific protocol, and have a way to account for human genetic variation and how it may affect tissue-specific expression. The former consideration will likely require a reliance on a greater repertoire of reference samples stratified by maturity level and differentiation protocol. The second may require the construction of a reference transcriptome.

168

4.2 SparCon assays enable connectivity measurements and guide mechanistic investigation

In Chapter 3, I investigated neuronal connectivity in human neurons with mutations in

SHANK2, a synaptic scaffolding gene strongly associated with ASD. Given genetic evidence pointing to altered synapse function as a key to understanding NDD pathophysiology, perhaps the most important use for IPSC-derived neurons in NDD modeling is their potential for use in functional tests of neuronal connectivity. Linking synaptic phenotypes to specific genetic mutations is necessary before mechanistic studies can be undertaken. In addition, identification of robust synaptic phenotypes can facilitate drug screening approaches that are targeted to the most directly relevant phenotype.

4.2.1 SparCon allows for isolated investigation of synaptogenesis and guides mechanistic investigation

There is currently a lack of tools for IPSC-derived neurons to isolate synaptic neuronal function for investigation. Sparse co-cultures address this need by allowing for investigation of synaptogenesis in a cell-autonomous manner. In sparse co-cultures, the lawn supplies an excess of axons, keeps cell culture conditions consistent and within-well normalization eliminates systematic error. The only remaining variable that affects connectivity of neurons is their synaptogenic capacity.

It allows for visualization of entire neurons and simultaneous acquisition of morphological and connectivity data from the same neuron. This can provide clues about mechanisms underlying changes in connectivity. For example, the increase in total synapse number in

SHANK2 mutant neurons is accompanied by substantial increases in dendrite growth and

169 complexity. This strongly points to a role of SHANK2 as a suppressor of dendrite branching.

Therefore, this prompts one to consider actin-dependent mechanisms of dendritic growth. It is possible that wild-type levels of SHANK2 facilitate Rho kinase signaling which can prevent excessive dendrite growth. Because synaptotropic dendrite growth is coupled to excitatory synaptogenesis and NMDA signaling, and SHANK2 has roles in both processes, dysregulation of appropriate dendrite growth may result in development of aberrant connectivity between neurons.

4.2.2 Comparison of SparCon with other methods

Two other methods have been recently advanced that have potential for investigating synaptic connectivity, 3D organoids and directly converted neurons. Lancaster et al. generated human cortical organoids that recapitulate many aspects of human cortical development and demonstrated its applicability in modeling microcephaly. Using this system, Mariani et al. discovered a FOXG1-dependent dysregulation of inhibitory neuron generation in idiopathic

ASD, leading to increased numbers of inhibitory neurons and inhibitory synaptic puncta, potentially altering excitation-inhibition balance. While organoids appear very promising for studying human neuronal development, Mariani et al. was not able to appreciably detect synaptic function. This is a challenge that is likely to be overcome with time, but several important questions persist. First, how variable are different organoids? As observed in chapter 3, perturbations in the synaptogenic environment can drastically affect measurements of neuronal connectivity. Given the complexity and time required for organoid development, it is likely that variability will be very high. Secondly, it is presently unclear how neuronal connectivity analyses can be standardized in organoids. It is conceivable that patches of high and low neuronal

170 connectivity can coexist in a given organoid, introducing substantial noise. 3D organoids require further technical refinement before they can be relied on to yield reproducible connectivity data.

Zhang et al. directly converted IPSCs to a pure population of excitatory neurons by forced expression of either NGN2 or NEUROD1. This approach has several important advantages for studying neuronal connectivity. First, it is able to generate a pure population of layer II/III excitatory neurons. Second, it is very rapid as neurons achieve synaptic activity by 3 weeks of age. In comparison, directed differentiation requires a minimum of 12 weeks to proceed from

IPSCs to neurons. These factors alone serve to drastically reduce variability among neuronal cultures, as well as restrict analyses to a specific neuronal subtype. Owing to this rapidity, it can be expected that in the short term direct conversion will be a mainstay of basic research using stem cell-derived neurons. However, it has several important limitations. First, while it generates a uniform and subtype-restricted population of neurons, it may not be suitable for modeling those disorders in which the affected neuronal subtype differs from those generated by the protocol.

Second, the rapid conversion to synaptically active neurons may compress important developmental processes and thus omit potential disease-relevant phenotypes. Third, development of synapses appears atypical with an absence of NMDA response despite a robust

AMPA response that can be inhibited by CNQX. While NMDA responsiveness does develop later, this appears to be an inversion of the typically accepted chronology of synaptic development. This may be an artifact of the system, or it could represent a genuine peculiarity of development of this particular neuronal subtype.

On the other hand, sparse co-cultures retain the flexibility of directed differentiation in generating a variety of neuronal subtypes and largely recapitulating embryonic neuronal development, but unlike 3D organoids, provide a way to control extensive cellular heterogeneity

171 that can arise as a result of this process. While the NGN2 system restricts analysis to a specific neuronal subtype, in sparse cocutlures, lentiviral labeling potentially allows the same degree of control, as well flexibility to label different neuronal subtypes. Additionally, it can permit analyses of inhibitory synapse function, a potentially important aspect of ASD, while direct conversion is restricted to analyses of excitatory synapses.

172

4.3 Role of SHANK2 as an early regulator of neuronal development

The results I obtained in Chapter 3 support the idea that SHANK2 functions during early neuronal development. Protein expression data show that SHANK2 protein increases sharply upon differentiation into neurons but begins to decline after approximately 4 weeks of age. This is consistent with findings from rodent literature that shows that SHANK2 is expressed before either SHANK3 or SHANK1 (A. M. Grabrucker et al. 2011), is localized to growth cones of immature dendrites (Du et al. 1998) and suppresses dendrite growth and branching complexity

(Berkel et al., 2012). However, my data challenge the idea that SHANK2 reduction in ASD could lead to deficit in excitatory synapse function, as frequently hypothesized based on in vitro knockdown in rodent neurons (Leblond et al. 2014). On the contrary, I find that total synapse number is increased by 54-91% and frequency of synaptic EPSCs by 180-590% in SHANK2 mutant neurons. This divergence between rodent and human findings suggests that temporal expression of SHANK2 in human neurons has less impact on synapse function than that of its paralog SHANK3. According to the Cortecon database, SHANK2 is expressed at its highest during early differentiation and then decreases, while SHANK3 mirrors it by rising later in the differentiation and staying upregulated. This may also partially explain why PMDS IPSC- derived neurons, which harbor deletions encompassing SHANK3, have impairments in excitatory neurotransmission (Shcheglovitov et al. 2013; Bidinosti et al. 2016). Additionally, compensatory upregulation of SHANK3 in SHANK2 mutant background has precedent (Schmeisser et al. 2012) and may partially underlie the large increase in connectivity in SHANK2 mutant neurons. In addition, a lowered amount of SHANK2 at the synapse may dysregulate actin-dependent mechanisms of synaptic plasticity and cause a greater number of AMPA and NMDA receptors to be inserted into the postsynaptic membrane, increasing the likelihood that a given synapse will

173 become stabilized following neuronal activity (Hanley 2014; Bu et al. 2015). The slight increases in sEPSC amplitude in some samples partially support this idea.

174

4.3.1 Alterations in neuronal connectivity in ASD

Genetic, post-mortem and functional imaging all suggest that alterations in neuronal connectivity may underlie ASD pathophysiology. Post-mortem studies show increases in synaptic density, suggesting overconnectivity in neurons. Functional magnetic resonance imaging point towards local overconnectivity alongside long-range hypoconnectivity. The observation of a propensity of SHANK2 mutant IPSC-derived neurons to make an overabundance of synapses suggests a possible mechanism about how this may arise. To my knowledge, this marks the first contribution of IPSC-derived neurons toward explaining overconnectivity of excitatory neurons in ASD. This is currently a second possible mechanism the development of overconnectivity in ASD, in addition to a deficit in mTOR-dependent autophagy (Tang et al.

2014). Importantly, the discovery of this phenotype only became possible by using sparse cocultures, as several of my previous attempts using standard culturing methods failed to reveal any difference.

175

4.4 Future Directions

4.4.1 Increasing the range of phenotypes assayable by SparCon

The sparse co-culture is a modular system, whose parts can be modified to develop new assays. The ability to engineer specific lawns can allow experimenters to create synaptogenic environments designed to emulate specific conditions. For example, for disorders resulting from mutations on the X-chromosome that cause neuronal mosaicism, such as RTT, creating a lawn that is an equal mixture of mutant and wild-type cells can better emulate brain environment. To restrict analyses to a specific synaptic subtype, lawns consisting of primarily inhibitory (T.-G.

Kim et al. 2014), excitatory (forced NGN2 expression - (Yingsha Zhang et al. 2013)), or serotonergic neurons (Lu et al. 2016) are now possible to generate. Network phenotypes can also be assayed by generating lawns whose activity can be optically or chemically modulated by expression of light- or chemically-gated ion channels. Synapse function may also be assayed in live neurons by adapting the mammalian GFP reconstitution across synaptic partners (mGRASP) system to sparse cocultures (J. Kim et al. 2011). By engineering the lawn to express the presynaptic mGRASP fragment and the assayed neurons the postsynaptic fragment, synapse formation and dendrite extension can be simultaneously observed in vitro. This can facilitate investigation of synaptotropic dendrite growth (S. X. Chen et al. 2010), as well as mechanisms of synapse formation, maintenance and pruning.

Elimination of systematic error by within-well normalization can enable drug screening programs if the speed of phenotyping can be increased. This can be achieved by adapting the system for calcium imaging. Instead of using GFP and mKO2 fluorophore pair to label neurons, one can switch from GFP to a blue protein such as mTagBFP2, leaving the green channel open for calcium indicators. A simple calcium indicator dye like Fluo4-AM can then be used to

176 monitor activity in both assayed cells and the lawn. This can allow to estimate the degree to which fluorescent cells have integrated into the neuronal networks in the dish. In addition, use of pharmacological agents such as CNQX to block AMPA receptors, APV to block NMDA receptor, bicucculline to block GABA receptors, as well as others can distinguish whether any observed differences are due to function of specific synapses or due to intrinsic changes. The system is theoretically adaptable to all-optical electrophysiology analyses using the Optopatch system which uses archaeorhodopsin-based voltage indicators and channelrhodopsin to stimulate neurons (Hochbaum et al. 2014). However, this will require a highly specialized imaging setup.

4.4.2 Alternative splicing and chromosome conformation in PTCHD1AS neurons

The possibility that alterations in RNA splicing gene sets are due to PTCHD1AS disruption support the idea that alternative splicing is altered in ASD, which has been recently proposed by several post-mortem studies. A long noncoding RNA as the potential cause of such alteration is particularly interesting and novel. However, this needs to be directly tested by isolating the effect of PTCHD1AS deletion and using RNA sequencing experiments, rather than microarrays(Irimia et al. 2014; Seok et al. 2015). The former can be completed by generating IPSCs from ASD cases with deletions restricted to exon 3 of PTCHD1AS or by targeted deletion of the same using

CRISPR/Cas9. This can be further bolstered by testing whether knockout of DDX53, a neighbouring gene downregulated in both ASD cases with PTCHD1 locus mutations, leads to alterations in AS.

LncRNAs are traditionally thought to affect AS by directly masking splice sites on nascent transcripts (Kung et al. 2013). However, the low abundance of PTCHD1AS, as well as the

177 widespread nature of AS changes in ASD brains, argues against this function. An alternative way by which PTCHD1AS may exert its effects is by regulating transcription of other genes in trans.

Because a cluster of ZF proteins on chromosome 19 has altered transcription in PTCHD1AS mutant neurons suggests that PTCHD1AS may function by regulating that cluster. This could occur by 3D conformation of the genome that brings the PTCHD1AS locus in close apposition to the locus on chromosome 19. This can be tested with chromosome conformation capture experiments, with subsequent mutagenesis of important sequences to directly test the role for any discovered association.

The absence of PTCHD1AS in mutant cells may prevent normal activation of transcription of genes in that cluster and lowered expression of those ZF proteins can cause downstream alterations in gene expression of splicing factors, as well as directly affect AS. RNA sequencing experiments to an appropriate depth of sequencing may elucidate potential AS changes in these neurons and suggest possible mechanisms.

4.4.3 Testing ECM function in NDD

WBS studies using IPSC-derived neurons converge on the intriguing possibility that ECM genes may be downregulated. Strikingly, the opposite is observed in PTCHD1/PTCHD1AS mutant neurons, which exhibit an upregulation of ECM function. The observation of opposite sociability phenotypes between WBS and ASD, opposite alterations in ECM gene expression, and the presence of altered neuronal connectivity (Chailangkarn et al. 2016) suggests the possibility that ECM may be critical to development of neuronal circuits underlying social communication. Extracellular cues appear to be critical for development of neuronal connectivity by regulating neuron number, dendrite morphogenesis and synaptogenesis (Kurshan et al. 2014;

178

Valnegri et al. 2015). Consistent with this idea, dysregulation of matrix metalloproteinaises is thought to underlie neuronal hyperactivity and hyperplasticity in ASD (Abdallah & Michel

2013).

This correlation can be bolstered if neurons from ASD cases caused by duplication of WBS genes have increases in ECM expression. In addition, cells can be compared in their ability to grow on different substrates. For example, WBS neurons with downregulation of ECM genes may be less able to grow on substrates that do not provide much trophic and structural support, such as plastic.

4.4.4 SHANK2 beyond the synapse

Results from chapter 3 strongly indicate that it is important to consider SHANK2 functions beyond the synapse. In addition to its abundance in neurons, SHANK2 is expressed in epithelial tissues throughout the body and CNVs in SHANK2 have been found in squamous cancers.

SHANK2 interacts with ARHGEF7, which has been recently shown to act as a scaffold for hippo pathway kinase LATS, promoting its activity in cancer cells. Lowered dosage of SHANK2 may cause a reduction in ARHGEF7, and thus a reduction in LATS activity. In turn, lack of LATS activity may aberrantly activate the hippo pathway by translocation of YAP/TAZ to the nucleus and increased transcription of target genes (Heidary Arash et al. 2014). Because the hippo pathway has been shown to regulate dendrite growth and complexity in Drosophila(Y.-N. Jan &

L. Y. Jan 2010), alteration in hippo signaling during early stages of neuronal development may promote excessive neuronal outgrowth. Therefore, lowered SHANK2 could cause neuronal growth through dysregulation of hippo signaling. A possible way to test whether this is occurring is to test for hippo pathway activation in SHANK2 mutant neurons by measuring transcript levels

179 of hippo pathway target genes. If they are found to be altered, it will be imperative to characterize the extent of transcriptomic alterations in SHANK2 mutant neurons. Hippo pathway dysregulation has been observed in IPSC-derived human neurons with mutations in CHD8, one of the highest confidence ASD genes (Cotney et al. 2015). Regulation of neuronal connectivity by hippo and its connection to ASD make this a novel and intriguing area of study.

4.4.5 Towards understanding molecular mechanisms of social ability

Social ability is a fundamental function of the human brain (Dunbar 2009). It exists on a spectrum - for example, ASD cases can be separated from typically developing controls on a continuous scale using the standardized Vineland Adaptive Behaviour Score (Figure 4)

(Robinson et al. 2016), which measures measures adaptive behaviour across domains of

Communication, Daily Living Skills, Gross Motor Skills and Socialization. While WBS results in increased sociability and is caused by 7q11.23 deletions, ASD encapsulates a broad range of impaired social ability and has a vast array of genetic risk factors. The plethora of loci increasing risk for ASD, incomplete penetrance, phenotypic variability and sex-dependent, polygenic inheritance of ASD all suggest a model where common genetic variation may coexist with rare genetic variants to increase the probability of developing ASD (Geschwind & Flint 2015). These findings strongly imply the existence of multiple mechanisms by which the function of brain circuits supporting social ability can be disrupted.

Genetic studies have uncovered a large number of ASD genetic loci that comprise two broad functional clusters of transcriptional control and synaptic function. While this is an important first step, a molecular understanding of what pathways are perturbed and how they impact synaptic connectivity requires transition to suitable biological models. Given the complex

180 genetic architecture of ASD, using IPSCs that largely preserve the genetic makeup of donors and allow functional studies of human neurons are critical. In Chapters 2 and 3, I described progress towards improving reproducibility and sensitivity of this new model system in detecting transcriptional and synaptic changes in WBS and ASD. For example, applying the sparse coculture method to a broader array of ASD samples can elucidate how connectivity can be altered and transcriptomics can point toward mechanism. With time, this approach may ultimately permit a stratification of ASD that is based on functional and molecular signatures rather than relying on a broad diagnosis. In turn, identification of specific altered pathways underlying subsets of ASD can promote the development of targeted therapies.

References

Abdallah, M.W. & Michel, T.M., 2013. Matrix metalloproteinases in autism spectrum disorders. Journal of molecular psychiatry, 1(1), p.16.

Adamo, A. et al., 2015. 7q11.23 dosage-dependent dysregulation in human pluripotent stem cells affects transcriptional programs in disease-relevant lineages. Nature Genetics, 47(2), pp.132–141.

American Psychiatric Association, Diagnostic and statistical manual of mental disorders 5 ed., Arlington, VA: American Psychiatric Publishing.

Ananiev, G. et al., 2011. Isogenic pairs of wild type and mutant induced pluripotent stem cell (iPSC) lines from Rett syndrome patients as in vitro disease model. A. Wutz, ed. PLoS ONE, 6(9), p.e25255.

Antonell, A. et al., 2010. Partial 7q11.23 deletions further implicate GTF2I and GTF2IRD1 as the main genes responsible for the Williams-Beuren syndrome neurocognitive profile. Journal of Medical Genetics, 47(5), pp.312–320.

Arbogast, T. et al., 2016. Reciprocal Effects on Neurocognitive and Metabolic Phenotypes in Mouse Models of 16p11.2 Deletion and Duplication Syndromes. G. S. Barsh, ed. PLoS Genetics, 12(2), p.e1005709.

Arons, M.H. et al., 2012. Autism-associated mutations in ProSAP2/Shank3 impair synaptic transmission and neurexin-neuroligin-mediated transsynaptic signaling. The Journal of neuroscience : the official journal of the Society for Neuroscience, 32(43), pp.14966–14978.

Bakanidze, G., Roinishvili, M. & Chkonia, E., 2013. Association of the nicotinic receptor α7 subunit gene (CHRNA7) with schizophrenia and visual backward masking. Frontiers in ….

Banach, R. et al., 2009. Brief Report: Relationship between non-verbal IQ and gender in autism. Journal of autism and developmental disorders, 39(1), pp.188–193.

Barak, B. & Feng, G., 2016. Neurobiology of social behavior abnormalities in autism and Williams syndrome. Nature Publishing Group, 19(6), pp.647–655.

Bardy, C. et al., 2016. Predicting the functional states of human iPSC-derived neurons with single-cell RNA-seq and electrophysiology. Molecular Psychiatry, 21(11), pp.1573–1588.

Baron, M.K., 2006. An Architectural Framework That May Lie at the Core of the Postsynaptic Density. Science, 311(5760), pp.531–535.

Bennett, M.K., Calakos, N. & Scheller, R.H., 1992. Syntaxin: a synaptic protein implicated in

181 182

docking of synaptic vesicles at presynaptic active zones. Science, 257(5067), pp.255–259.

Berg, J.M. & Geschwind, D.H., 2012. Autism genetics: searching for specificity and convergence. Genome Biol.

Berkel, S. et al., 2012. Inherited and de novo SHANK2 variants associated with autism spectrum disorder impair neuronal morphogenesis and physiology. Human Molecular Genetics, 21(2), pp.344–357.

Berkel, S. et al., 2010. Mutations in the SHANK2 synaptic scaffolding gene in autism spectrum disorder and mental retardation. Nature Genetics, 42(6), pp.489–491.

Beuren, A.J., Apitz, J. & Harmjanz, D., 1962. Supravalvular aortic stenosis in association with mental retardation and a certain facial appearance. Circulation, 26, pp.1235–1240.

Bidinosti, M. et al., 2016. CLK2 inhibition ameliorates autistic features associated with SHANK3 deficiency. Science.

Binelli, C. et al., 2016. Facial emotion processing in patients with social anxiety disorder and Williams-Beuren syndrome: an fMRI study. Journal of psychiatry & neuroscience : JPN, 41(3), pp.182–191.

Boeckers, T.M. et al., 1999. Proline-Rich Synapse-Associated Protein-1/Cortactin Binding Protein 1 (ProSAP1/CortBP1) Is a PDZ-Domain Protein Highly Enriched in the Postsynaptic Density. Journal of Neuroscience, 19(15), pp.6506–6518.

Bongiovanni, A.M., Eberlein, W.R. & Jones, I.T., 1957. Idiopathic hypercalcemia of infancy, with failure to thrive; report of three cases, with a consideration of the possible etiology. The New England Journal of Medicine, 257(20), pp.951–958.

Borralleras, C. et al., 2016. Synaptic plasticity and spatial working memory are impaired in the CD mouse model of Williams-Beuren syndrome. Molecular brain, 9(1), p.76.

Bourgeron, T., 2009. A synaptic trek to autism. Current Opinion in Neurobiology, 19(2), pp.231–234.

Boyle, C.A. et al., 2011. Trends in the prevalence of developmental disabilities in US children, 1997-2008. Pediatrics, 127(6), pp.1034–1042.

Brennand, K.J. et al., 2015. Creating Patient-Specific Neural Cells for the In Vitro Study of Brain Disorders. Stem cell reports, 5(6), pp.933–945.

Brennand, K.J. et al., 2011. Modelling schizophrenia using human induced pluripotent stem cells. Nature, 473(7346), pp.221–225.

Bu, Y. et al., 2015. Myosin IIb-dependent Regulation of Actin Dynamics Is Required for N- Methyl-D-aspartate Receptor Trafficking during Synaptic Plasticity. The Journal of biological chemistry, 290(42), pp.25395–25410.

183

Byers, E.S. et al., 2013. Sexual well-being of a community sample of high-functioning adults on the autism spectrum who have been in a romantic relationship. Autism : the international journal of research and practice, 17(4), pp.418–433.

Campbell, L.E. et al., 2009. Brain structural differences associated with the behavioural phenotype in children with Williams syndrome. Brain Research, 1258, pp.96–107.

Carneiro, A. et al., 2008. Prognostic Impact of Array-based Genomic Profiles in Esophageal Squamous Cell Cancer. BMC Cancer, 8(1), p.98.

Casanova, M.F. et al., 2006. Minicolumnar abnormalities in autism. Acta Neuropathologica, 112(3), pp.287–303.

Cathomen, T. & Keith Joung, J., 2008. Zinc-finger Nucleases: The Next Generation Emerges. Molecular Therapy, 16(7), pp.1200–1207.

Cermak, T. et al., 2011. Efficient design and assembly of custom TALEN and other TAL effector-based constructs for DNA targeting. Nucleic Acids Research, 39(12), pp.e82–e82.

Chailangkarn, T. et al., 2016. A human neurodevelopmental model for Williams syndrome. Nature, 536(7616), pp.338–343.

Chambers, S.M. et al., 2009. Highly efficient neural conversion of human ES and iPS cells by dual inhibition of SMAD signaling. Nature Biotechnology, 27(3), pp.275–280.

Chaste, P. & Leboyer, M., 2012. Autism risk factors: genes, environment, and gene-environment interactions. Dialogues in clinical neuroscience, 14(3), pp.281–292.

Chen, J.A. et al., 2015. The emerging picture of autism spectrum disorder: genetics and pathology. Annual review of pathology, 10(1), pp.111–144.

Chen, S.X. et al., 2010. Neurexin-neuroligin cell adhesion complexes contribute to synaptotropic dendritogenesis via growth stabilization mechanisms in vivo. Neuron, 67(6), pp.967–983.

Chen, X. et al., 2016. Coupled electrophysiological recording and single cell transcriptome analyses revealed molecular mechanisms underlying neuronal maturation. Protein & cell, 7(3), pp.175–186.

Cheung, A.Y.L. et al., 2011. Isolation of MECP2-null Rett Syndrome patient hiPS cells and isogenic controls through X-chromosome inactivation. Human Molecular Genetics, 20(11), pp.2103–2115.

Chiang, M.-C. et al., 2007. 3D pattern of brain abnormalities in Williams syndrome visualized using tensor-based morphometry. NeuroImage, 36(4), pp.1096–1109.

Chilian, B. et al., 2013. Dysfunction of SHANK2 and CHRNA7 in a patient with intellectual disability and language impairment supports genetic epistasis of the two loci. Clinical Genetics, 84(6), pp.560–565.

184

Chow, M.L. et al., 2012. Age-Dependent Brain Gene Expression and Copy Number Anomalies in Autism Suggest Distinct Pathological Processes at Young Versus Mature Ages G. Gibson, ed. PLoS Genetics, 8(3), p.e1002592.

Christensen, D.L., Bilder, D.A. & Zahorodny, W., 2016. Prevalence and characteristics of autism spectrum disorder among 4-year-old children in the Autism and Developmental Disabilities Monitoring Network. … of Developmental & ….

Chu, V.T. et al., 2015. Increasing the efficiency of homology-directed repair for CRISPR-Cas9- induced precise gene editing in mammalian cells. Nature Biotechnology, 33(5), pp.543–548.

Chung, J.H. et al., 2014. A PTCH1 homolog transcriptionally activated by p53 suppresses Hedgehog signaling. The Journal of biological chemistry, 289(47), pp.33020–33031.

Cordin, O. et al., 2006. The DEAD-box protein family of RNA helicases. Gene, 367, pp.17–37.

Cotney, J. et al., 2015. The autism-associated chromatin modifier CHD8 regulates other autism risk genes during human neurodevelopment. Nature Communications, 6, p.6404.

Cuberos, H. et al., 2015. Roles of LIM kinases in central nervous system function and dysfunction. FEBS Letters, 589(24 Pt B), pp.3795–3806.

Darnell, J.C. et al., 2011. FMRP stalls ribosomal translocation on mRNAs linked to synaptic function and autism. Cell, 146(2), pp.247–261.

De Rubeis, S. et al., 2014. Synaptic, transcriptional and chromatin genes disrupted in autism. Nature, 515(7526), pp.209–215.

Del Campo, M. et al., 2006. Hemizygosity at the NCF1 Gene in Patients with Williams-Beuren Syndrome Decreases Their Risk of Hypertension. The American Journal of Human Genetics, 78(4), pp.533–542.

DeRosa, B.A. et al., 2015. hVGAT-mCherry: A novel molecular tool for analysis of GABAergic neurons derived from human pluripotent stem cells. Molecular and cellular neurosciences, 68, pp.244–257.

Devlin, B. & Scherer, S.W., 2012. Genetic architecture in autism spectrum disorder. Current opinion in genetics & development, 22(3), pp.229–237.

Djuric, U. et al., 2015. MECP2e1 isoform mutation affects the form and function of neurons derived from Rett syndrome patient iPS cells. Neurobiology of disease, 76, pp.37–45.

Du, Y. et al., 1998. Identification of a novel cortactin SH3 domain-binding protein and its localization to growth cones of cultured neurons. Molecular and Cellular Biology, 18(10), pp.5838–5851.

Dunbar, R.I.M., 2009. The social brain hypothesis and its implications for social evolution., 36(5), pp.562–572.

185

Ebert, D.H. & Greenberg, M.E., 2013. Activity-dependent neuronal signalling and autism spectrum disorder. Nature, 493(7432), pp.327–337.

Ehlers, M.D., 2003. Activity level controls postsynaptic composition and signaling via the ubiquitin-proteasome system. Nature Neuroscience, 6(3), pp.231–242.

Eisenberg, R. et al., 1964. FAMILIAL SUPRAVALVULAR AORTIC STENOSIS. American journal of diseases of children (1960), 108, pp.341–347.

Eroglu, C. & Barres, B.A., 2010. Regulation of synaptic connectivity by glia. Nature, 468(7321), pp.223–231.

Ewart, A.K. et al., 1993. A human vascular disorder, supravalvular aortic stenosis, maps to chromosome 7. Proceedings of the National Academy of Sciences of the United States of America, 90(8), pp.3226–3230.

Fatemi, S.H. et al., 2012. Consensus Paper: Pathological Role of the Cerebellum in Autism. The Cerebellum, 11(3), pp.777–807.

Ferrero, G.B. et al., 2010. An atypical 7q11.23 deletion in a normal IQ Williams-Beuren syndrome patient. European journal of human genetics : EJHG, 18(1), pp.33–38.

Filiou, M.D. et al., 2010. Profiling of mouse synaptosome proteome and phosphoproteome by IEF. Electrophoresis, 31(8), pp.1294–1301.

Frangiskakis, J.M. et al., 1996. LIM-kinase1 hemizygosity implicated in impaired visuospatial constructive cognition. Cell, 86(1), pp.59–69.

Freier, K. et al., 2006. Recurrent coamplification of cytoskeleton-associated genes EMS1 and SHANK2 with CCND1 in oral squamous cell carcinoma. Genes, chromosomes & cancer, 45(2), pp.118–125.

Fujiwara, T. et al., 2006. Analysis of knock-out mice to determine the role of HPC-1/syntaxin 1A in expressing synaptic plasticity. The Journal of neuroscience : the official journal of the Society for Neuroscience, 26(21), pp.5767–5776.

Gagliardi, C. et al., 2003. Unusual cognitive and behavioural profile in a Williams syndrome patient with atypical 7q11.23 deletion. Journal of Medical Genetics, 40(7), pp.526–530.

Gaj, T., Gersbach, C.A. & Barbas, C.F., 2013. ZFN, TALEN, and CRISPR/Cas-based methods for genome engineering. Trends in biotechnology, 31(7), pp.397–405.

Garcia, R.E. et al., 1964. Idiopathic Hypercalcemia and Supravalvular Aortic Stenosis. The New England Journal of Medicine, 271(3), pp.117–120.

Geschwind, D.H. & Flint, J., 2015. Genetics and genomics of psychiatric disease. Science, 349(6255), pp.1489–1494.

Geschwind, D.H. & Rakic, P., 2013. Cortical evolution: judge the brain by its cover. Neuron,

186

80(3), pp.633–647.

Gillentine, M.A. & Schaaf, C.P., 2015. The human clinical phenotypes of altered CHRNA7 copy number. Biochemical …, 97(4), pp.352–362.

Golzio, C. et al., 2012. KCTD13 is a major driver of mirrored neuroanatomical phenotypes of the 16p11.2 copy number variant. Nature, 485(7398), pp.363–367.

Gosch, A., Städing, G. & Pankau, R., 1994. Linguistic abilities in children with Williams-Beuren syndrome. American Journal of Medical Genetics Part A, 52(3), pp.291–296.

Grabrucker, A.M., 2014. A role for synaptic zinc in ProSAP/Shank PSD scaffold malformation in autism spectrum disorders S. Carbonetto & T. Bourgeron, eds. Developmental Neurobiology, 74(2), pp.136–146.

Grabrucker, A.M. et al., 2011. Concerted action of zinc and ProSAP/Shank in synaptogenesis and synapse maturation. The EMBO Journal, 30(3), pp.569–581.

Grabrucker, S. et al., 2014. The PSD protein ProSAP2/Shank3 displays synapto-nuclear shuttling which is deregulated in a schizophrenia-associated mutation. Experimental Neurology, 253, pp.126–137.

Green, T. et al., 2016. Surface-based morphometry reveals distinct cortical thickness and surface area profiles in Williams syndrome. American journal of medical genetics. Part B, Neuropsychiatric genetics : the official publication of the International Society of Psychiatric Genetics, 171B(3), pp.402–413.

Griesi-Oliveira, K. et al., 2015. Modeling non-syndromic autism and the impact of TRPC6 disruption in human neurons. Molecular Psychiatry, 20(11), pp.1350–1365.

Guilinger, J.P. et al., 2014. Broad specificity profiling of TALENs results in engineered nucleases with improved DNA-cleavage specificity. Nature Methods, 11(4), pp.429–435.

Gupta, K., Hardingham, G.E. & Chandran, S., 2013. NMDA receptor-dependent glutamate excitotoxicity in human embryonic stem cell-derived neurons. Neuroscience Letters, 543, pp.95–100.

Halbedl, S. et al., 2016. Shank3 is localized in axons and presynaptic specializations of developing hippocampal neurons and involved in the modulation of NMDA receptor levels at axon terminals. Journal of Neurochemistry, pp.n/a–n/a.

Halevy, T., Czech, C. & Benvenisty, N., 2015. Molecular mechanisms regulating the defects in fragile X syndrome neurons derived from human pluripotent stem cells. Stem cell reports, 4(1), pp.37–46.

Hall, B.J., Ripley, B. & Ghosh, A., 2007. NR2B signaling regulates the development of synaptic AMPA receptor current. The Journal of neuroscience : the official journal of the Society for Neuroscience, 27(49), pp.13446–13456.

187

Han, W. et al., 2006. Shank2 associates with and regulates Na+/H+ exchanger 3. The Journal of biological chemistry, 281(3), pp.1461–1469.

Hanley, J.G., 2014. Actin-dependent mechanisms in AMPA receptor trafficking. Frontiers in Cellular Neuroscience, 8, p.381.

Hartley, B.J. & Brennand, K.J., 2016. Neural organoids for disease phenotyping, drug screening and developmental biology studies. Neurochemistry International.

Hayashi, M.K. et al., 2009. The Postsynaptic Density Proteins Homer and Shank Form a Polymeric Network Structure. Cell, 137(1), pp.159–171.

Heidary Arash, E. et al., 2014. Arhgef7 promotes activation of the Hippo pathway core kinase Lats. The EMBO Journal, 33(24), pp.2997–3011.

Hering, H. & Sheng, M., 2003. Activity-dependent redistribution and essential role of cortactin in dendritic spine morphogenesis. The Journal of neuroscience : the official journal of the Society for Neuroscience, 23(37), pp.11759–11769.

Hochbaum, D.R. et al., 2014. All-optical electrophysiology in mammalian neurons using engineered microbial rhodopsins. Nature Methods, 11(8), pp.825–833.

Hopyan, T. et al., 2001. Music skills and the expressive interpretation of music in children with Williams-Beuren syndrome: pitch, rhythm, melodic imagery, phrasing, and musical affect. Child neuropsychology : a journal on normal and abnormal development in childhood and adolescence, 7(1), pp.42–53.

Hotta, A. et al., 2009. Isolation of human iPS cells using EOS lentiviral vectors to select for pluripotency. Nature Methods, 6(5), pp.370–376.

Hsu, P.D. et al., 2013. DNA targeting specificity of RNA-guided Cas9 nucleases. Nature.

Huber, K.M. et al., 2002. Altered synaptic plasticity in a mouse model of fragile X mental retardation. Proceedings of the National Academy of Sciences of the United States of America, 99(11), pp.7746–7750.

Hung, A.Y. et al., 2008. Smaller Dendritic Spines, Weaker Synaptic Transmission, but Enhanced Spatial Learning in Mice Lacking Shank1. The Journal of neuroscience : the official journal of the Society for Neuroscience, 28(7), pp.1697–1708.

Hutsler, J.J. & Zhang, H., 2010. Increased dendritic spine densities on cortical projection neurons in autism spectrum disorders. Brain Research, 1309, pp.83–94.

Hwang, J.-I. et al., 2005. The interaction of phospholipase C-beta3 with Shank2 regulates mGluR-mediated calcium signal. The Journal of biological chemistry, 280(13), pp.12467– 12473.

Irimia, M. et al., 2014. A highly conserved program of neuronal microexons is misregulated in autistic brains. Cell, 159(7), pp.1511–1523.

188

Jackowski, A.P. et al., 2009. Brain abnormalities in Williams syndrome: a review of structural and functional magnetic resonance imaging findings. European journal of paediatric neurology : EJPN : official journal of the European Paediatric Neurology Society, 13(4), pp.305–316.

Jamain, S. et al., 2003. Mutations of the X-linked genes encoding neuroligins NLGN3 and NLGN4 are associated with autism. Nature Genetics, 34(1), pp.27–29.

Jan, Y.-N. & Jan, L.Y., 2010. Branching out: mechanisms of dendritic arborization. Nature Reviews Neuroscience, 11(5), pp.316–328.

Jaramillo, T.C. et al., 2016. Altered Striatal Synaptic Function and Abnormal Behaviour in Shank3 Exon4-9 Deletion Mouse Model of Autism. Autism research : official journal of the International Society for Autism Research, 9(3), pp.350–375.

Jarrold, C., Baddeley, A.D. & Hewes, A.K., 1998. Verbal and nonverbal abilities in the Williams syndrome phenotype: evidence for diverging developmental trajectories. Journal of child psychology and psychiatry, and allied disciplines, 39(4), pp.511–523.

Järvinen-Pasley, A. et al., 2008. Defining the social phenotype in Williams syndrome: A model for linking gene, the brain, and behavior. Development and Psychopathology, 20(1), pp.1– 35.

Jentsch, T.J., 2000. Neuronal KCNQ potassium channels: physiology and role in disease. Nature Reviews Neuroscience, 1(1), pp.21–30.

Jiang, L. et al., 2014. ZNF667/Mipu1 is a novel anti-apoptotic factor that directly regulates the expression of the rat Bax gene in H9c2 cells. G.-C. Fan, ed. PLoS ONE, 9(11), p.e111653.

Jiang, Y.-H. & Ehlers, M.D., 2013. Modeling Autism by SHANK Gene Mutations in Mice. Neuron, 78(1), pp.8–27.

Jobe, L.E. & Williams White, S., 2007. Loneliness, social relationships, and a broader autism phenotype in college students. Personality and Individual Differences, 42(8), pp.1479–1489.

Johnson, M.A. et al., 2007. Functional neural development from human embryonic stem cells: accelerated synaptic activity via astrocyte coculture. The Journal of neuroscience : the official journal of the Society for Neuroscience, 27(12), pp.3069–3077.

Johnson, W.E., Li, C. & Rabinovic, A., 2007. Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics (Oxford, England), 8(1), pp.118–127.

Jorde, L.B. et al., 1991. Complex segregation analysis of autism. American Journal of Human Genetics, 49(5), pp.932–938.

Kanner, L., 1943. Autistic Disturbances of Affective Contact. Nervous Child, 2, pp.217–250.

Karmiloff-Smith, A. et al., 1998. Linguistic dissociations in Williams syndrome: evaluating receptive syntax in on-line and off-line tasks. Neuropsychologia, 36(4), pp.343–351.

189

Khattak, S. et al., 2015. Human induced pluripotent stem cell derived neurons as a model for Williams-Beuren syndrome. Molecular brain, 8(1), p.77.

Khwaja, O.S. et al., 2014. Safety, pharmacokinetics, and preliminary assessment of efficacy of mecasermin (recombinant human IGF-1) for the treatment of Rett syndrome. Proceedings of the National Academy of Sciences, 111(12), pp.4596–4601.

Kim, D.-S. et al., 2014. Optimizing neuronal differentiation from induced pluripotent stem cells to model ASD. Frontiers in Cellular Neuroscience, 8, p.109.

Kim, H.J. & Magrané, J., 2011. Isolation and culture of neurons and astrocytes from the mouse brain cortex. Methods in molecular biology (Clifton, N.J.), 793(Chapter 4), pp.63–75.

Kim, J. et al., 2011. mGRASP enables mapping mammalian synaptic connectivity with light microscopy. Nature Methods, 9(1), pp.96–102.

Kim, K.-Y., Hysolli, E. & Park, I.-H., 2011. Neuronal maturation defect in induced pluripotent stem cells from patients with Rett syndrome. Proceedings of the National Academy of Sciences, 108(34), pp.14169–14174.

Kim, T.-G. et al., 2014. Efficient specification of interneurons from human pluripotent stem cells by dorsoventral and rostrocaudal modulation. Stem cells (Dayton, Ohio), 32(7), pp.1789– 1804.

Kirwan, P. et al., 2015. Development and function of human cerebral cortex neural networks from pluripotent stem cells in vitro. Development, 142(18), pp.3178–3187.

Kong, A. et al., 2012. Rate of de novo mutations and the importance of father/'s age to disease risk. Nature, 488(7412), pp.471–475.

Kouser, M. et al., 2013. Loss of predominant Shank3 isoforms results in hippocampus-dependent impairments in behavior and synaptic transmission. The Journal of neuroscience : the official journal of the Society for Neuroscience, 33(47), pp.18448–18468.

Krey, J.F. et al., 2013. Timothy syndrome is associated with activity-dependent dendritic retraction in rodent and human neurons. Nature Neuroscience, 16(2), pp.201–209.

Kung, J.T.Y., Colognori, D. & Lee, J.T., 2013. Long noncoding RNAs: past, present, and future. Genetics, 193(3), pp.651–669.

Kurshan, P.T. et al., 2014. Regulation of synaptic extracellular matrix composition is critical for proper synapse morphology. The Journal of neuroscience : the official journal of the Society for Neuroscience, 34(38), pp.12678–12689.

Kwiatkowski, A.V. et al., 2007. Ena/VASP Is Required for neuritogenesis in the developing cortex. Neuron, 56(3), pp.441–455.

Lai, M.-C., Lombardo, M.V. & Baron-Cohen, S., 2014. Autism. Lancet (London, England), 383(9920), pp.896–910.

190

Lalli, M.A. et al., 2016. Haploinsufficiency of BAZ1B contributes to Williams syndrome through transcriptional dysregulation of neurodevelopmental pathways. Human Molecular Genetics, 25(7), pp.1294–1306.

Lancaster, M.A. et al., 2014. Cerebral organoids model human brain development and microcephaly. Nature, 501(7467), pp.373–379.

Leblond, C.S. et al., 2012. Genetic and Functional Analyses of SHANK2 Mutations Suggest a Multiple Hit Model of Autism Spectrum Disorders M. State, ed. PLoS Genetics, 8(2), p.e1002521.

Leblond, C.S. et al., 2014. Meta-analysis of SHANK Mutations in Autism Spectrum Disorders: a gradient of severity in cognitive impairments. PLoS Genetics, 10(9), p.e1004580.

Lee, J. et al., 2015. Shank3-mutant mice lacking exon 9 show altered excitation/inhibition balance, enhanced rearing, and spatial memory deficit. Frontiers in Cellular Neuroscience, 9(2), p.14966.

Lee, J.A. & Lupski, J.R., 2006. Genomic rearrangements and gene copy-number alterations as a cause of nervous system disorders. Neuron, 52(1), pp.103–121.

Leyfer, O.T. et al., 2006. Prevalence of psychiatric disorders in 4 to 16-year-olds with Williams syndrome. American journal of medical genetics. Part B, Neuropsychiatric genetics : the official publication of the International Society of Psychiatric Genetics, 141B(6), pp.615– 622.

Li, D.Y. et al., 1997. Elastin point mutations cause an obstructive vascular disease, supravalvular aortic stenosis. Human Molecular Genetics, 6(7), pp.1021–1028.

Li, H. & Durbin, R., 2010. Fast and accurate long-read alignment with Burrows-Wheeler transform. Bioinformatics, 26(5), pp.589–595.

Li, Y. et al., 2013. Global Transcriptional and TranslationalRepression in Human-Embryonic- Stem-Cell-Derived Rett Syndrome Neurons. Stem Cell, 13(4), pp.446–458.

Lim, S., 1999. Characterization of the Shank Family of Synaptic Proteins. MULTIPLE GENES, ALTERNATIVE SPLICING, AND DIFFERENTIAL EXPRESSION IN BRAIN AND DEVELOPMENT. Journal of Biological Chemistry, 274(41), pp.29510–29518.

Lin, M. et al., 2016. Integrative transcriptome network analysis of iPSC-derived neurons from schizophrenia and schizoaffective disorder patients with 22q11.2 deletion. BMC Systems Biology, 10(1), p.105.

Livesey, M.R. et al., 2014. Maturation of AMPAR composition and the GABAAR reversal potential in hPSC-derived cortical neurons. The Journal of neuroscience : the official journal of the Society for Neuroscience, 34(11), pp.4070–4075.

Lowery, M.C. et al., 1995. Strong correlation of elastin deletions, detected by FISH, with Williams syndrome: evaluation of 235 patients. American Journal of Human Genetics,

191

57(1), pp.49–53.

Lu, J. et al., 2016. Generation of serotonin neurons from human pluripotent stem cells. Nature Biotechnology, 34(1), pp.89–94.

MacGillavry, H.D. et al., 2015. Shank-cortactin interactions control actin dynamics to maintain flexibility of neuronal spines and synapses. The European journal of neuroscience, pp.n/a– n/a.

Marchetto, M.C.N. et al., 2010. A Model for Neural Development and Treatment of Rett Syndrome Using Human Induced Pluripotent Stem Cells. Cell, 143(4), pp.527–539.

Mariani, J. et al., 2015. FOXG1-Dependent Dysregulation of GABA/Glutamate Neuron Differentiation in Autism Spectrum Disorders. Cell, 162(2), pp.375–390.

Marler, J.A. et al., 2010. Auditory function and hearing loss in children and adults with Williams syndrome: cochlear impairment in individuals with otherwise normal hearing. C. A. Morris, ed. American journal of medical genetics. Part C, Seminars in medical genetics, 154C(2), pp.249–265.

Maroof, A.M. et al., 2013. Directed differentiation and functional maturation of cortical interneurons from human embryonic stem cells. Cell stem cell, 12(5), pp.559–572.

Martens, M.A., Reutens, D.C. & Wilson, S.J., 2010. Auditory cortical volumes and musical ability in Williams syndrome. Neuropsychologia, 48(9), pp.2602–2609.

McKenna, A. et al., 2010. The Genome Analysis Toolkit: A MapReduce framework for analyzing next-generation DNA sequencing data. Genome Research, 20(9), pp.1297–1303.

Meda, S.A., Pryweller, J.R. & Thornton-Wells, T.A., 2012. Regional brain differences in cortical thickness, surface area and subcortical volume in individuals with Williams syndrome. Y. He, ed. PLoS ONE, 7(2), p.e31913.

Meng, Y. et al., 2002. Abnormal spine morphology and enhanced LTP in LIMK-1 knockout mice. Neuron, 35(1), pp.121–133.

Merico, D. et al., 2010. Enrichment Map: A Network-Based Method for Gene-Set Enrichment Visualization and Interpretation. PLoS ONE, 5(11), p.e13984.

Merla, G. et al., 2006. Submicroscopic deletion in patients with Williams-Beuren syndrome influences expression levels of the nonhemizygous flanking genes. American Journal of Human Genetics, 79(2), pp.332–341.

Mi, D. et al., 2013. Pax6 exerts regional control of cortical progenitor proliferation via direct repression of Cdk6 and hypophosphorylation of pRb. Neuron, 78(2), pp.269–284.

Micheva, K.D. et al., 2010. Single-synapse analysis of a diverse synapse population: proteomic imaging methods and markers. Neuron, 68(4), pp.639–653.

192

Miyaoka, Y. et al., 2014. Isolation of single-base genome-edited human iPS cells without antibiotic selection. Nature Methods, 11(3), pp.291–293.

Morris, C.A. et al., 2015. 7q11.23 Duplication syndrome: Physical characteristics and natural history. American Journal of Medical Genetics Part A, 167A(12), pp.2916–2935.

Nageshappa, S. et al., 2015. Altered neuronal network and rescue in a human MECP2 duplication model. Molecular Psychiatry.

Noor, A. et al., 2010. Disruption at the PTCHD1 Locus on Xp22.11 in Autism spectrum disorder and intellectual disability. Science Translational Medicine, 2(49), pp.49ra68–49ra68.

Okamoto, P.M. et al., 2001. Dynamin isoform-specific interaction with the shank/ProSAP scaffolding proteins of the postsynaptic density and actin cytoskeleton. The Journal of biological chemistry, 276(51), pp.48458–48465.

Osborne, L.R. & Mervis, C.B., 2007. Rearrangements of the Williams-Beuren syndrome locus: molecular basis and implications for speech and language development. Expert reviews in molecular medicine, 9(15), pp.1–16.

Osório, A. et al., 2014. Cerebral and cerebellar MRI volumes in Williams syndrome. Research in developmental disabilities, 35(4), pp.922–928.

Otani, T. et al., 2016. 2D and 3D Stem Cell Models of Primate Cortical Development Identify Species-Specific Differences in Progenitor Behavior Contributing to Brain Size. Cell stem cell, 18(4), pp.467–480.

Otto, E. et al., 1996. Quantitative detection of cell culture Mycoplasmas by a one step polymerase chain reaction method. Methods in cell ….

Pak, C. et al., 2015. Human Neuropsychiatric Disease Modeling using Conditional Deletion Reveals Synaptic Transmission Defects Caused by Heterozygous Mutations in NRXN1. Cell stem cell, 17(3), pp.316–328.

Parikshak, N.N. et al., 2016. Genome-wide changes in lncRNA, alternative splicing, and cortical patterning in autism. bioRxiv.

Park, E., 2003. The Shank Family of Postsynaptic Density Proteins Interacts with and Promotes Synaptic Accumulation of the betaPIX Guanine Nucleotide Exchange Factor for Rac1 and Cdc42. Journal of Biological Chemistry, 278(21), pp.19220–19229.

Paşca, A.M. et al., 2015. Functional cortical neurons and astrocytes from human pluripotent stem cells in 3D culture. Nature Methods, 12(7), pp.671–678.

Paşca, S.P., Portmann, T., Voineagu, I., Yazawa, M., Shcheglovitov, A., Paşca, A.M., Cord, B., Palmer, T.D., Chikahisa, S., Nishino, S., Bernstein, J.A., Hallmayer, J., Geschwind, D.H. & Dolmetsch, R.E., 2011a. Using iPSC-derived neurons to uncover cellular phenotypes associated with Timothy syndrome. Nature Medicine, 17(12), pp.1657–1662.

193

Paşca, S.P., Portmann, T., Voineagu, I., Yazawa, M., Shcheglovitov, A., Paşca, A.M., Cord, B., Palmer, T.D., Chikahisa, S., Nishino, S., Bernstein, J.A., Hallmayer, J., Geschwind, D.H. & Dolmetsch, R.E., 2011b. Using iPSC-derived neurons to uncover cellular phenotypes associated with Timothy syndrome. Nature Medicine, 17(12), pp.1657–1662.

Pattanayak, V. et al., 2011. Revealing off-target cleavage specificities of zinc-finger nucleases by in vitro selection. Nature Methods, 8(9), pp.765–770.

Peter, S. et al., 2016. Dysfunctional cerebellar Purkinje cells contribute to autism-like behaviour in Shank2-deficient mice. Nature Communications, 7, p.12627.

Peykov, S. et al., 2015. Identification and functional characterization of rare SHANK2 variants in schizophrenia. Molecular Psychiatry, 20(12), pp.1489–1498.

Phelan, K. & McDermid, H.E., 2012. The 22q13.3 Deletion Syndrome (Phelan-McDermid Syndrome). Molecular syndromology, 2(3-5), pp.186–201.

Pinto, D. et al., 2010. Functional impact of global rare copy number variation in autism spectrum disorders. Nature, 466(7304), pp.368–372.

Pober, B.R., 2010. Williams–Beuren Syndrome. dx.doi.org.myaccess.library.utoronto.ca, 362(3), pp.239–252.

Qin, H.-D. et al., 2016. Genomic Characterization of Esophageal Squamous Cell Carcinoma Reveals Critical Genes Underlying Tumorigenesis and Poor Prognosis. American Journal of Human Genetics, 98(4), pp.709–727.

Quesnel-Vallières, M. et al., 2016. Misregulation of an Activity-Dependent Splicing Network as a Common Mechanism Underlying Autism Spectrum Disorders. Molecular cell, 64(6), pp.1023–1034.

Quitsch, A. et al., 2005. Postsynaptic shank antagonizes dendrite branching induced by the leucine-rich repeat protein Densin-180. The Journal of neuroscience : the official journal of the Society for Neuroscience, 25(2), pp.479–487.

Reimand, J. et al., 2016. g:Profiler-a web server for functional interpretation of gene lists (2016 update). Nucleic Acids Research, 44(W1), pp.W83–9.

Rice, J., Cloninger, C.R. & Reich, T., 1980. General causal models for sex differences in the familial transmission of multifactorial traits: an application to human spatial visualizing ability. Social biology, 27(1), pp.36–47.

Robinson, E.B. et al., 2016. Genetic risk for autism spectrum disorders and neuropsychiatric variation in the general population. Nature Genetics, 48(5), pp.552–555.

Rodrigues, D.C. et al., 2016. MECP2 Is Post-transcriptionally Regulated during Human Neurodevelopment by Combinatorial Action of RNA-Binding Proteins and miRNAs. Cell reports, 17(3), pp.720–734.

194

Romorini, S. et al., 2004. A functional role of postsynaptic density-95-guanylate kinase- associated protein complex in regulating Shank assembly and stability to synapses. The Journal of neuroscience : the official journal of the Society for Neuroscience, 24(42), pp.9391–9404.

Ronemus, M. et al., 2014. The role of de novo mutations in the genetics of autism spectrum disorders. Nature Reviews Genetics, 15(2), pp.133–141.

Rudie, J.D. & Dapretto, M., 2013. Convergent evidence of brain overconnectivity in children with autism? Cell reports, 5(3), pp.565–566.

Rutter, M., 2012. Changing Concepts and Findings on Autism. Journal of autism and developmental disorders, 43(8), pp.1749–1757.

Rutter, M., 2005. Incidence of autism spectrum disorders: changes over time and their meaning. Acta paediatrica (Oslo, Norway : 1992), 94(1), pp.2–15.

Sala, C. et al., 2001. Regulation of Dendritic Spine Morphology and Synaptic Function by Shank and Homer. Neuron, 31, pp.115–130.

Sala, C. et al., 2015. Shank synaptic scaffold proteins: keys to understanding the pathogenesis of autism and other synaptic disorders. Journal of Neurochemistry, pp.n/a–n/a.

Sanders, S.J. et al., 2015. Insights into Autism Spectrum Disorder Genomic Architecture and Biology from 71 Risk Loci. Neuron, 87(6), pp.1215–1233.

Sanders, S.J. et al., 2011. Multiple recurrent de novo CNVs, including duplications of the 7q11.23 Williams syndrome region, are strongly associated with autism. Neuron, 70(5), pp.863–885.

Sandin, S. et al., 2016. Autism risk associated with parental age and with increasing difference in age between the parents. Molecular Psychiatry, 21(5), pp.693–700.

Sandoe, J. & Eggan, K., 2013. Opportunities and challenges of pluripotent stem cell neurodegenerative disease models. Nature Neuroscience, 16(7), pp.780–789.

Schlaeger, T.M. et al., 2015. A comparison of non-integrating reprogramming methods. Nature Biotechnology, 33(1), pp.58–63.

Schlesinger, B.E., Butler, N.R. & Black, J.A., 1956. Severe type of infantile hypercalcaemia. British medical journal, 1(4959), pp.127–134.

Schmeisser, M.J. et al., 2012. Autistic-like behaviours and hyperactivity in mice lacking ProSAP1/Shank2. Nature, 486(7402), pp.256–260.

Schubert, C., 2009. The genomic basis of the Williams – Beuren syndrome. Cellular and Molecular Life Sciences, 66(7), pp.1178–1197.

Segura-Puimedon, M. et al., 2014. Heterozygous deletion of the Williams-Beuren syndrome

195

critical interval in mice recapitulates most features of the human disorder. Human Molecular Genetics, 23(24), pp.6481–6494.

Seok, J. et al., 2015. RASA: Robust Alternative Splicing Analysis for Human Transcriptome Arrays. Scientific reports, 5, p.11917.

Shalem, O., Sanjana, N.E. & Zhang, F., 2015. High-throughput functional genomics using CRISPR-Cas9. Nature Reviews Genetics, 16(5), pp.299–311.

Shcheglovitov, A. et al., 2013. SHANK3 and IGF1 restore synaptic deficits in neurons from 22q13 deletion syndrome patients. Nature, 503(7475), pp.267–271.

Sheng, M. & Kim, E., 2000. The Shank family of scaffold proteins. Journal of Cell Science, 113 ( Pt 11), pp.1851–1856.

Shin, S.M. et al., 2012. GKAP orchestrates activity-dependent postsynaptic protein remodeling and homeostatic scaling. Nature Publishing Group, 15(12), pp.1655–1666.

Smith, R.M. & Sadee, W., 2011. Synaptic signaling and aberrant RNA splicing in autism spectrum disorders. Frontiers in synaptic neuroscience, 3, p.1.

Smyth, G.K., 2004. Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Mol Biol.

Speed, H.E. et al., 2015. Autism-Associated Insertion Mutation (InsG) of Shank3 Exon 21 Causes Impaired Synaptic Transmission and Behavioral Deficits. Journal of Neuroscience, 35(26), pp.9648–9665.

Spruston, N., 2008. Pyramidal neurons: dendritic structure and synaptic integration. Nature Reviews Neuroscience, 9(3), pp.206–221.

Stamova, B.S. et al., 2013. Evidence for differential alternative splicing in blood of young boys with autism spectrum disorders. Molecular Autism, 4(1), p.30.

Stapleton, T., MacDonald, W.B. & Lightwood, R., 1957. The pathogenesis of idiopathic hypercalcemia in infancy. The American journal of clinical nutrition, 5(5), pp.533–542.

State, M.W. & Geschwind, D.H., 2015. Leveraging genetics and genomics to define the causes of mental illness. Biological Psychiatry, 77(1), pp.3–5.

Stein, J.L. et al., 2014. A quantitative framework to evaluate modeling of cortical development by neural stem cells. Neuron, 83(1), pp.69–86.

Subramanian, A. et al., 2005. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proceedings of the National Academy of Sciences of the United States of America, 102(43), pp.15545–15550.

Sugahara, K. et al., 2011. Combination effects of distinct cores in 11q13 amplification region on cervical lymph node metastasis of oral squamous cell carcinoma. International journal of

196

oncology, 39(4), pp.761–769.

Takahashi, K. et al., 2007. Induction of pluripotent stem cells from adult human fibroblasts by defined factors. Cell, 131(5), pp.861–872.

Tang, G. et al., 2014. Loss of mTOR-dependent macroautophagy causes autistic-like synaptic pruning deficits. Neuron, 83(5), pp.1131–1143.

Thomas, N.S. et al., 1999. Xp deletions associated with autism in three females. Human Genetics, 104(1), pp.43–48.

Tu, J.C. et al., 1999. Coupling of mGluR/Homer and PSD-95 complexes by the Shank family of postsynaptic density proteins. Neuron, 23(3), pp.583–592.

Udwin, O. & Yule, W., 1991. A cognitive and behavioural phenotype in williams syndrome. Journal of Clinical and Experimental Neuropsychology, 13(2), pp.232–244.

Valnegri, P., Puram, S.V. & Bonni, A., 2015. Regulation of dendrite morphogenesis by extrinsic cues. Trends in Neurosciences, 38(7), pp.439–447.

Vessey, J.P. & Karra, D., 2007. More than just synaptic building blocks: scaffolding proteins of the post-synaptic density regulate dendritic patterning. Journal of Neurochemistry, 102(2), pp.324–332.

Voineagu, I. et al., 2011. Transcriptomic analysis of autistic brain reveals convergent molecular pathology. Nature, 474(7351), pp.380–384.

Wang, X. et al., 2016. Altered mGluR5-Homer scaffolds and corticostriatal connectivity in a Shank3 complete knockout model of autism. Nature Communications, 7, p.11459.

Wang, X. et al., 2011. Synaptic dysfunction and abnormal behaviors in mice lacking major isoforms of Shank3. Human Molecular Genetics, 20(15), pp.ddr212–3108.

Wells, M.F. et al., 2016. Thalamic reticular impairment underlies attention deficit in Ptchd1Y/− mice. Nature, 532(7597), pp.58–63.

Wen, Z. et al., 2014. Synaptic dysregulation in a human iPS cell model of mental disorders. Nature, 515(7527), pp.414–418.

Werling, D.M. & Geschwind, D.H., 2013. Sex differences in autism spectrum disorders. Current opinion in neurology, 26(2), pp.146–153.

Williams, J.C., Barratt-Boyes, B.G. & Lowe, J.B., 1961. Supravalvular aortic stenosis. Circulation, 24, pp.1311–1318.

Wingate, M. et al., 2014. Prevalence of Autism Spectrum Disorder Among Children Aged 8 Years — Autism and Developmental Disabilities Monitoring Network, 11 Sites, United States, 2010. MMWR, 63, pp.1–24.

Won, H. et al., 2012. Autistic-like social behaviour in Shank2-mutant mice improved by

197

restoring NMDA receptor function. Nature, 486(7402), pp.261–265.

Woodard, C.M. et al., 2014. iPSC-Derived Dopamine Neurons Reveal Differences between Monozygotic Twins Discordant for Parkinson’s Disease. Cell reports, 9(4), pp.1173–1182.

Wright, A.V., Nuñez, J.K. & Doudna, J.A., 2016. Biology and Applications of CRISPR Systems: Harnessing Nature’s Toolbox for Genome Engineering. Cell, 164(1-2), pp.29–44.

Wulff, H., Castle, N.A. & Pardo, L.A., 2009. Voltage-gated potassium channels as therapeutic targets. Nature reviews. Drug discovery, 8(12), pp.982–1001.

Wyllie, D.J.A., Livesey, M.R. & Hardingham, G.E., 2013. Influence of GluN2 subunit identity on NMDA receptor function. Neuropharmacology, 74, pp.4–17.

Xu, J. et al., 2013. SNARE proteins synaptobrevin, SNAP-25, and syntaxin are involved in rapid and slow endocytosis at synapses. Cell reports, 3(5), pp.1414–1421.

Yeargin-Allsopp, M. et al., 2003. Prevalence of autism in a US metropolitan area. Jama, 289(1), pp.49–55.

Yi, F., Danko, T., Botelho, S.C., Patzke, C., Pak, C., Wernig, M. & Südhof, T.C., 2016a. Autism-associated SHANK3 haploinsufficiency causes Ih channelopathy in human neurons. Science.

Yi, F., Danko, T., Botelho, S.C., Patzke, C., Pak, C., Wernig, M. & Südhof, T.C., 2016b. Autism-associated SHANK3 haploinsufficiency causes Ih channelopathy in human neurons. Science, 352(6286), pp.aaf2669–aaf2669.

Ying, J. et al., 2012. Genome-wide screening for genetic alterations in esophageal cancer by aCGH identifies 11q13 amplification oncogenes associated with nodal metastasis. M. Katoh, ed. PLoS ONE, 7(6), p.e39797.

Young, E.J. et al., 2008. Reduced fear and aggression and altered serotonin metabolism in Gtf2ird1-targeted mice. Genes, brain, and behavior, 7(2), pp.224–234.

Yuan, S.H. et al., 2011. Cell-surface marker signatures for the isolation of neural stem cells, glia and neurons derived from human pluripotent stem cells. M. Pera, ed. PLoS ONE, 6(3), p.e17540.

Zhang, J.D. et al., 2017. Detect tissue heterogeneity in gene expression data with BioQC. BMC Genomics, 18(1), p.277.

Zhang, W.-B. et al., 2016. Fyn Kinase regulates GluN2B subunit-dominant NMDA receptors in human induced pluripotent stem cell-derived neurons. Scientific reports, 6, p.23837.

Zhang, Ye et al., 2016. Purification and Characterization of Progenitor and Mature Human Astrocytes Reveals Transcriptional and Functional Differences with Mouse. Neuron, 89(1), pp.37–53.

198

Zhang, Yingsha et al., 2013. Rapid single-step induction of functional neurons from human pluripotent stem cells. Neuron, 78(5), pp.785–798.

Zhao, S. et al., 2014. Comparison of RNA-Seq and microarray in transcriptome profiling of activated T cells. S.-D. Zhang, ed. PLoS ONE, 9(1), p.e78644.

Zitzer, H. et al., 1999. Somatostatin receptor interacting protein defines a novel family of multidomain proteins present in human and rodent brain. The Journal of biological chemistry, 274(46), pp.32997–33001.

Figure 3 reprinted by permission from Macmillan Publishers Ltd: Nature Reviews Genetics Feb;15(2):133-41 Ronemus et al. 2014.

Figure 4 reprinted by permission from Macmillan Publishers Ltd: Nature Genetics May;48(5):552-5 Robinson et al. 2016.

199