key: cord-0306932-hq06ntqo
authors: Pak, ChangHui; Danko, Tamas; Mirabella, Vincent R.; Wang, Jinzhao; Zhang, Xianglong; Ward, Thomas; Grieder, Sarah; Vangipuram, Madhuri; Huang, Yu-Wen Alvin; Liu, Yingfei; Jin, Kang; Dexheimer, Philip; Bardes, Eric; Mittelpunkt, Alexis; Ma, Junyi; McLachlan, Michael; Moore, Jennifer C.; Urban, Alexander E.; Dage, Jeffrey L.; Swanson, Bradley J.; Aronow, Bruce J.; Pang, Zhiping P.; Levinson, Douglas F.; Wernig, Marius; Südhof, Thomas C.
title: Cross-Platform Validation of Neurotransmitter Release Impairments in Schizophrenia Patient-Derived NRXN1-Mutant Neurons
date: 2020-11-03
journal: bioRxiv
DOI: 10.1101/2020.11.03.366617
sha: 5106ad6f85282967caee2691228230d8d92786a3
doc_id: 306932
cord_uid: hq06ntqo

Heterozygous NRXN1 deletions constitute the most prevalent currently known single-gene mutation predisposing to schizophrenia. Previous studies showed that engineered heterozygous NRXN1 deletions impaired neurotransmitter release in human neurons, suggesting a synaptic pathophysiological mechanism. Utilizing this observation for drug discovery, however, requires confidence in its robustness and validity. Here, we describe a multi-center effort to test the generality of this pivotal observation, using independent analyses at two laboratories of patient-derived and newly engineered human neurons with heterozygous NRXN1 deletions. We show that in neurons that were trans-differentiated from induced pluripotent stem cells derived from three NRXN1-deletion patients, the same impairment in neurotransmitter release was observed as in engineered NRXN1-deficient neurons. This impairment manifested as a decrease in spontaneous synaptic events and in evoked synaptic responses, and an alteration in synaptic paired-pulse depression. Nrxn1-deficient mouse neurons generated from embryonic stem cells by the same method as human neurons did not exhibit impaired neurotransmitter release, suggesting a human-specific phenotype. NRXN1 deletions produced a reproducible increase in the levels of CASK, an intracellular NRXN1-binding protein, and were associated with characteristic gene expression changes. Thus, heterozygous NRXN1 deletions robustly impair synaptic function in human neurons regardless of genetic background, enabling future drug discovery efforts.

Schizophrenia is a devastating brain disorder that affects millions of people worldwide and exhibits a strong genetic component. In a key discovery, deletions or duplications of larger stretches of chromosomal DNA that lead to copy number variations (CNVs) were identified two decades ago (Sebat et al., 2004; Sebat et al., 2007) . CNVs occur unexpectedly frequently, are often de novo, and usually affect multiple genes depending on the size of the deleted or duplicated stretch of DNA. Strikingly, the biggest genetic risk for schizophrenia was identified in three unrelated CNVs, a duplication of region 16p11.2 and deletions of 22q11.2 and of 2p16.3 (Kirov, 2015; Coelewij and Curtis, 2018; Malhotra and Sebat, 2012; Marshall et al., 2017; Anh et al., 2014; Kirov et al., 2014; Rees et al., 2014) . Of these CNVs, 16p11.2 and 22q11.2 CNVs affect more than 20 genes, whereas 2p16.3 CNVs impact only one or more exons of a single gene, NRXN1, which encodes the presynaptic cell-adhesion molecule neurexin-1 (Hu et al., 2019; Coelewij and Curtis, 2018; Kasem et al., 2018; Sudhof, 2017; Kirov, 2015; Marshall et al., 2017) . NRXN1 CNVs confer an approximately ten-fold increase in risk of schizophrenia, and additionally strongly predispose to other neuropsychiatric disorders, especially autism and Tourette syndrome (Lowther et al., 2017 , Castronovo et al., 2020 . Moreover, genome-wide association studies using DNA microarrays identified common changes in many other genes that predispose to schizophrenia with smaller effect sizes (SCZ working group of PGC, 2014; Parnidas et al., 2018; Fromer et al., 2014; Fromer et al., 2016; Ripke et al., 2017; Sullivan and Geschwind 2019; Flaherty et al., 2019; Hall et al., 2020) . Viewed together, these studies indicate that variations in a large number of genes are linked to schizophrenia. Among these genetic variations, heterozygous exonic CNVs of NRXN1 are rare events, but nevertheless constitute the most prevalent high-risk single-gene association at present.

Neurexins are central regulators of neural circuits that control diverse synapse properties, such as the presynaptic release probability, the postsynaptic receptor composition, and synaptic plasticity (Missler et al., 2003; Aoto et al., 2013 and Anderson et al., 2015; Chen et al., 2017 , Dai et al., 2019 Luo et al., 2020) . To test whether heterozygous NRXN1 mutations might cause functional impairments in human neurons, we previously generated conditionally mutant human embryonic stem (ES) cells that enabled induction of heterozygous NRXN1 deletions using Cre-recombinase (Pak et al., 2015) . We then analyzed the effect of the deletion on the properties of neurons trans-differentiated from the conditionally mutant ES cells using forced expression of Ngn2, a method that generates a relatively homogeneous population of excitatory neurons that are also referred to as induced neuronal (iN) cells (Zhang et al., 2013) . These experiments thus examined isogenic neurons without or with a heterozygous NRXN1 loss-of-function mutation that mimicked the schizophrenia-associated 2p16.3 CNVs, enabling precise control of the genetic background.

In these experiments, the heterozygous NRXN1 deletion produced a robust but discrete impairment in neurotransmitter release without major changes in neuronal development or morphology (Pak et al., 2015) . These results were exciting because they suggested that a discrete impairment in neurotransmitter release could underlie the predisposition to schizophrenia conferred by the 2p16.3 CNVs, but these experiments did not reveal whether the actual NRXN1 mutation that is observed in schizophrenia patients induces the same synaptic impairment (Hyman, 2015) .

The present project was initiated to achieve multiple overlapping aims emerging from the initial study of Pak et al., 2015 . First, we aimed to validate or refute the results obtained with neurons generated from engineered conditionally mutant ES cells, but now using neurons generated from patient-derived iPS cells with NRXN1 mutations (Fig. 1A) . This goal was pursued in order to gain confidence in the disease-relevance of the observed phenotypes.

Second, we wanted to test whether the observed phenotype is independent of the laboratory of analysis, i.e. whether it is sufficiently robust to be replicated at multiple sites (Fig. 1A ). This goal was motivated by the observation of poor reproducibility of neuroscience research.

However, we hypothesized that the lack of reproducibility is often due to variations in experimental conditions rather than true failed replication testing of the original findings and designed our studies to demonstrate robustness of the findings through replication. Third, we aimed to generate reagents that could be broadly used by the scientific community for investigating the cellular basis of neuropsychiatric disorders (Panchision, 2016) . This goal was prompted by the challenges posed by the finding that many different genes appear to be linked to schizophrenia. Fourth, we aimed to definitively establish or exclude the possibility that human neurons are uniquely sensitive to a heterozygous loss of NRXN1 as compared to mouse neurons (Fig. 1B) . The goal here was to establish whether at least as regards to NRXN1, mouse and human neurons exhibit fundamental differences. Fifth and finally, we hoped to gain further insights into the mechanisms by which NRXN1 mutations may predispose to schizophrenia, an obviously needed objective given our lack of understanding of this severe disorder. As described in detail below, our data provide advances towards meeting these goals, establishing unequivocally that heterozygous NRXN1 deletions in human but not in mouse neurons cause a robust impairment in neurotransmitter release that is replicable in multiple laboratories.

Cohort of cases and controls. Peripheral blood mononuclear cell specimens and genomic DNA were obtained from the NIMH Repository and Genomics Resource (NRGR), donated by schizophrenia patients carrying heterozygous NRXN1 exonic deletions and control individuals who were participants in the Molecular Genetics of Schizophrenia (MGS2)

European-ancestry cohort (Shi et al., 2009) . The controls met criteria that predicted low genetic risk for schizophrenia. Cases and controls were aged 35-51 years at collection (for details, see Methods; for information on the availability of non-identified clinical information and of biomaterials, see www.nimhgenetics.org; for patient and control properties of the cell lines reported here, see Table 1 ). iPS cell lines were generated from peripheral blood mononuclear cells by the Rutgers University Cell and DNA Repository (RUCDR) via integration-free Sendai virus reprogramming, and three subclones were generated from each line for analysis ( Fig. 1) . Reprogramming included extensive quality control measures to verify normal karyotypes, morphology, and pluripotency of iPS cells, which were also tested pre-and post-freezing for mycoplasma ( which is predicted to be benign.

Of the five lines with NRXN1 deletions (referred to as NRXN1 del ) and the six control lines available, we selected three pairs of cases and controls based on similarity in gender and age at donation (Table 1) . Two of the lines selected had deletions affecting only NRXN1a, while the third line had a deletion ablating expression of both NRXN1a and NRXN1b (Fig.   S1 ). All subsequent differentiations, functional assays, and data analyses were performed using this pairing system, such that each patient-derived human iPS cell line was consistently paired with its own control human iPS cell line to minimize experimental and line-to-line variability. This approach does not imply genetic similarity between the cases and controls, and the pairing is by necessity intrinsically random. Frozen vials of iPS cells were shipped for cell line expansion, banking, and iN cell trans-differentiation to generate human neurons at two sites, Stanford University and FUJIFILM Cellular Dynamics Inc.

(FCDI) (Fig. 1) . The neurons generated at Stanford University were then also analyzed at Stanford, whereas the neurons generated at FCDI were shipped for analysis to Rutgers

University.

into excitatory neurons (Zhang et al., 2013; Pak et al., 2015; Yi et al., 2016) . These human neurons, also referred to as iN cells, are composed of a population of excitatory, relatively homogeneous neurons resembling cortical layer 2/3 pyramidal neurons (Zhang et al., 2013) .

Since production of Ngn2-induced human neurons is robust and scalable, we decided to further optimize it with two specific goals in mind: 1) to generate human neurons by transdifferentiation of iPS cells that are grown on feeder cell layers, which was necessary to minimize karyotypic abnormalities, and 2) to manufacture human neurons at an industry scale, thus enabling distribution of human neurons to multiple sites for downstream functional studies.

Guided by our published protocols (Pak et al., 2018;  see Methods), we modified the iPS cell dissociation step to efficiently remove mouse feeder cells from human iPS cell colonies. We also improved the initial induction step to provide the optimal medium combination for cell survival and TetO-inducible Ngn2 expression via doxycycline. However, we observed unexpected differences between iPS cell lines. Every iPS cell line pair had to be optimized separately to achieve workable conditions for the induction, differentiation and survival of Ngn2-induced neurons (Fig. S3 ). Adjustments of lentivirus titers, starting cell numbers, and timing of the window of doxycycline application before puromycin selection were made for each iPS cell line, and no standard treatments with reliable survival and trans-differentiation of iPS cell lines into neurons was possible. This variability likely reflects clonal differences in iPS cell lines that are not immediately apparent in standard assays.

reproducible and reliable in an individual lab setting, it is not well suited for up-scaling.

Therefore, FCDI optimized a large-scale manufacturing strategy to produce Ngn2-induced human neurons via the PiggyBac transposon system (see Methods). This system utilizes stable transgene integration by PiggyBac, which contains an inducible Ngn2 and a constitutive puromycin expression cassette. Puromycin-resistant cells were selected and expanded prior to doxycycline induction. Next, cryopreservation of Ngn2-induced human neurons was optimized, such that post-mitotic neurons can be cryopreserved in a fashion where they retain their pre-cryopreservation phenotype after thawing, and batches of cryopreserved human neurons were benchmarked against fresh cultures (see Methods).

Lastly, post-thaw culture conditions were optimized so that Ngn2-induced human neurons can be cultured long-term in any laboratory CO2 incubator under defined media conditions and yield electrically and synaptically functional human neurons for imaging and whole-cell patch-clamp electrophysiology experiments (see Methods).

Heterozygous NRXN1 deletions (NRXN1 del ) predisposing to schizophrenia do not affect neuronal morphology or synapse numbers. We previously characterized the effect of conditional heterozygous NRXN1 mutations on the molecular, cellular and electrophysiological properties of human neurons (Pak et al., 2015) . In these studies, NRXN1mutant neurons were derived from engineered ES cells carrying conditional heterozygous NRXN1 deletions or truncations that could be produced by recombinases. We found that heterozygous NRXN1 mutations did not impair dendritic arborization or synapse formation of human neurons, nor did they alter their passive membrane properties or action potential generation properties (Pak et al., 2015) . Therefore, we tested the same parameters in Ngn2-induced neurons generated from pairs of schizophrenia patient-derived NRXN1 del and control iPS cells. Neurons were either generated with the small-scale method and analyzed at Stanford, or were generated by the PiggyBac-based large scale method at FCDI and analyzed at Rutgers (Fig. 1A, S3 ).

In examining the first two pairs of NRXN1 del and control neurons, we did not detect any significant changes in neurite outgrowth, number of primary dendritic processes or dendritic branch points, or soma size ( Fig. 2A neurons exhibited a large (~2-fold) decrease in mEPSC frequency, but no change in the mEPSC amplitude (Fig. 4) . This decrease was observed in two independent pairs of NRXN1 del vs. control neurons, and was similarly detected in small-scale neurons produced and analyzed at Stanford, and in large-scale neurons produced at FCDI and analyzed at Rutgers. These results again phenocopy those previously obtained for engineered NRXN1mutant human neurons (Pak et al., 2015) , suggesting that the heterozygous NRXN1 mutation impairs synaptic transmission.

Schizophrenia-associated NRXN1 del CNVs decrease the neurotransmitter release probability. To determine whether the decrease in mEPSC frequency reflects a decrease in synaptic strength and to determine whether this decrease may be caused by a change in neurotransmitter release probability, we measured action potential-evoked excitatory postsynaptic currents (EPSCs) mediated by α-amino-3-hydroxy-5-methyl-4isoxazolepropionic acid receptors (AMPARs). Consistent with the decrease in mEPSC frequency, we observed a large decrease (~2-fold) in EPSC amplitude in NRXN1 del neurons compared to controls both for the two pairs studied in the analyses described above, and also for a third pair of NRXN1 del vs. control neurons (Fig. 5A ).

We then examined whether the decreased synaptic strength in NRXN1 del neurons is due to a decrease in release probability, since the lack of a change in mEPSC amplitudes suggested that postsynaptic AMPARs were normal. We assessed the release probability by measuring the coefficient of variation (C.V.) of evoked EPSCs, which is inversely proportional to the release probability (Hefft et al., 2002) . The coefficient of variation of EPSCs was increased significantly (~1.4-fold) in NRXN1 del vs. control neurons, suggesting a decrease in release probability (Fig. 5A ). To independently test this conclusion, we measured the paired-pulse ratio (PPR) of EPSCs, which is the ratio of the EPSC amplitudes evoked by two closely spaced action potentials. The PPR depends on the release probability because the extent of release induced by the first action potential determines, among others, how much additional release can be induced by the second action potential (Abbott and Regehr, 2004; Neher and Brose, 2018) . In control neurons, the PPR exhibited a decrease in the second response (referred to as paired pulse depression) because the release probability under our recording conditions is high (Fig. 5B ). However, in the NRXN1 del neurons generated from patient-derived iPS cells, the PPR was greatly increased (i.e., paired-pulse depression was decreased), confirming a deficit in the initial release probability. This increase in PPR was replicated in all three pairs of NRXN1 del vs.

control neurons (Fig. 5B ).

We performed similar experiments with the two pairs of NRXN1 del and control neurons that we primarily analyzed at Rutgers U., with comparable results (Fig. 5C, 5D ). Again, we observed a decrease in the EPSC amplitude and an increase in the coefficient of variation of the EPSC, although the difference was not significant (Fig. 5C ). Moreover, we detected an increase in the PPR, which was highly significant (Fig. 5D ). Taken together, these results strongly indicate that patient-derived NRXN1 del neurons exhibit a robust and reproducible decrease in neurotransmitter release probability.

A newly engineered conditional NRXN1 del iPS cell line reproduces the synaptic impairments observed in patient-derived NRXN1 del neurons. In analyzing patientderived neurons, a pressing question is whether the observed phenotypes in neurons with disease-associated CNVs or mutations are truly caused by these genetic changes, or are produced by polygenic effects resulting from a combination of a specific mutations with a particular genetic background (Hyman, 2015) . Mutations associated with neuropsychiatric disorders often predispose to multiple disease conditions. For example, NRXN1 CNVs are among the most frequent mutations associated with not only schizophrenia, but also intellectual disability, autism-spectrum disorders, epilepsy and others (Castronovo et al., 2020; Lowther et al., 2017 , Kasem et al., 2018 Szatmari et al., 2007; Guilmatre et al., 2009; Liu et al., 2012; Ching et al., 2010; Nag et al., 2013; Huang et al., 2017) . We thus decided to test whether the similarity of the synaptic impairments we observed in NRXN1 del neurons generated from schizophrenia patient-derived iPS cells to those we previously described for NRXN1-deficient neurons generated from ES cells can be validated with neurons generated from iPS cells carrying a newly engineered conditional NRXN1 mutation. For this purpose, we generated a new iPS cell line (from our control line C3141a, Table 1 ) carrying a heterozygous conditional KO (cKO) of NRXN1 using a strategy that is identical to the approach we previously employed in mice and in human ES cells (Fig. S4 , Table 2 ). We produced this additional validation tool not only to confirm our conclusions, but also to generate a conditionally mutant NRXN1 iPS cell line that can be freely distributed for further studies, and is not subject to the commercial constraints imposed on the previously generated conditionally mutant NRXN1 ES cell lines (Pak et al., 2015) .

We limited our analysis of the newly engineered conditional NRXN1 del neurons to key measurements, namely assessments of mEPSCs and evoked EPSCs. The newly engineered NRXN1 del neurons exhibited the same decrease in mEPSC frequency and EPSC amplitude and the same increase in PPR as the patient-derived NRXN1 del neurons ( Fig. 6 neurons. For this purpose, we generated ES cells from mice with exactly the same mutation as the human conditional NRXN1 del iPS and ES cells Trotter et al., 2019) .

We converted these mouse ES cells into Ngn2-induced neurons, and analyzed the electrophysiological phenotype of these neurons. Strikingly, the mouse neurons did not exhibit a significant synaptic impairment (Fig. 7, S5B ). They displayed a normal mEPSC frequency, and a trend towards a decrease in EPSC amplitude that was not statistically significant ( Fig. 7A-7D) . Moreover, the mouse Nrxn1 del neurons showed no change in paired-pulse ratio, possibly the most sensitive electrophysiological phenotype (Fig. 7E ).

These results indicate that the synaptic phenotypes conferred by the heterozygous human NRXN1 del mutations are unique to the human neuronal context.

neurons. One striking finding previously obtained in ES cell-engineered NRXN1 del neurons was the increase in CASK protein stability, which was detected with two different engineered NRXN1 mutations (Pak et al., 2015) . CASK is a cytoplasmic scaffolding protein that interacts with neurexins (Hata et al., 1996) and forms a tight complex with other presynaptic proteins (Wei et al., 2011; Cohen et al., 1998; Hsueh et al., 1998; Butz et al., 1998; Tabuchi et al., 2002) . In human patients, mutations in CASK -which is X-linked-are associated with autism spectrum disorders and X-linked mental retardation in addition to brain malformations (Najm et al., 2008; Saitsu et al., 2012; Piluso et al., 2009; Hackett et al., 2010; Sanders et al., 2012; Neale et al., 2012) . We thus tested whether CASK protein is also increased in schizophrenia patient-derived NRXN1 del neurons. Consistent with previous studies, we observed a large increase (>50%) in CASK protein levels in NRXN1 del neurons, while all other synaptic proteins measured were not altered, with the exception of NRXN1 protein levels, which were downregulated by ~50% (Fig. 8A, 8B ).

Moreover, we detected a similar increase in CASK levels in neurons derived from the newly engineered heterozygous NRXN1 del iPS cell line (Fig. 8C, 8D ), further validating this change. Since no change in CASK mRNA levels was detected in the RNAseq experiments (see below), these results indicate that CASK protein is stabilized and protected from degradation upon heterozygous deletion of NRXN1 in human neurons.

four-weeks old, relatively mature human neurons generated from the three pairs of patientderived NRXN1 del and control iPS cells analyzed in Fig. 5 , we performed bulk RNAsequencing (RNAseq) on total RNA in triplicate (Fig. 9A ). In addition, we performed bulk RNAseq analyses (in triplicate) on isogenic pairs of human neurons without or with the heterozygous NRXN1 deletion that were trans-differentiated from the newly engineered iPS cell line carrying a conditional NRXN1 del allele (Fig. 6, S4 ), and on 3 unrelated wild-type iPS cell lines (in duplicate). In total, we performed differential gene expression analyses on 30 samples (9 controls vs. 9 schizophrenia-NRXN1 mutants, 3 controls vs. 3 engineered NRXN1 mutants, and 6 wild-type iPS cell samples) in triplicates or duplicates (replicates refer to independent cultures performed at different time points).

As a result of the co-culturing scheme of human neurons with mouse glia, the RNAseq data on neurons were composed of a mixture of human neuronal and mouse glia transcriptomes.

A key step in processing of the RNAseq data was to deconvolve the two transcriptomes and to normalize the relative abundance of each mRNA from each species to the total number of mRNAs from that species only. As described in the Methods, the Kallisto program was able to unambiguously assign each paired 150 basepair sequence to mouse and human reference transcriptomes (Bray et al., 2016) . 9A ). This approach provided a reproducible abundance measure for each gene in each sample. The resulting values were used to form a log2(TPM+1) gene-by-sample matrix that was used for differential gene expression analyses. suggested that the different clusters were associated with different biological categories, but could not be assigned to defined functional biological pathways. For example, clusters C1 and C3 are both associated with 'synaptic signaling' and 'presynapse', whereas clusters C6 and C7 are both associated with 'nucleic acid binding'.

We quantitatively compared the transcriptomes among the various groups of samples.

Principal to stringent quality control such as karyotyping, suggesting that the transcriptome differences are not due to large chromosomal arrangements (Fig. S2 ). The process of subcloning and engineering iPS cells may thus be sufficient to produce large shifts in iPS cell properties, resulting in fundamental gene expression changes between neurons derived from these iPS cells. On the background of these gene expression differences, the strong similarity in the synaptic phenotypes of patient-derived and engineered NRXN1 del neurons appears even more compelling.

A primary goal of the bulk RNAseq experiments was to assess the specific effect of heterozygous NRXN1 deletions on gene transcription. No differences between NRXN1 del and control neurons could be detected in the principal component analysis (Fig. S6 ), and none of the clusters of DEGs was driven by gene expression differences between NRXN1 del vs. wild-type genotypes (Fig. 9B, S7 ). These observations indicate that the heterozygous NRXN1 deletion does not produce major perturbations of gene expression, but do not rule out more subtle effects. To achieve a more granular analyses, we selected (after initial QC, 10A , S8). The increase in KYAT3 expression was highly reproducible in various experiments, suggesting a true expression change (Fig. S9 ). preferentially produced in neurons, it is also expressed in other cell types. When we measured NNAT mRNA levels in patient-derived NRXN1 del vs. control iPS cells, we found also a consistent increase in NNAT expression in all patient-derived iPS cells (Fig. 10F) .

Thus, the increased NNAT expression in patient-derived NRXN1 del cells was not specific to neurons.

We next asked whether the increased NNAT expression could be a direct consequence of the heterozygous NRXN1 deletion. We analyzed NRXN1 del neurons obtained from two different engineered, conditionally mutant stem cell lines (see Table 2 ), our originally described ES cell line (Pak et al., 2015; Fig. 10G ) and the newly engineered iPS cell line ( Fig. 10H ). Both populations of NRXN1 del neurons exhibited no increase in NNAT expression, even though they displayed the modest increase in RIMS1 expression observed in patient-derived NRXN1 del neurons (Fig. 10C, 10D) . Thus, NRXN1 del neurons derived from schizophrenia patients exhibited a consistent NNAT expression change that was absent from controls and from 'normal' neurons in which the same NRXN1 deletion was engineered, despite the fact that these neurons display the same functional phenotype. A possible reason for the difference in NNAT expression between the engineered NRXN1 del and the patient-associated NRXN1 del mutation could be that NNAT is an imprinted gene (Kagitani et al., 1997; Evans et al., 2005) . Silencing of NNAT by imprinting may be maintained in the engineered iPS cell line but not in the patient-derived iPS cell line. An alternative explanation could be that the relative immaturity of the engineered neurons is responsible, but this seems unlikely given that increased NNAT expression is already observed in patient-derived NRXN1 del iPS cells, and thus not dependent on neuronal transdifferentiation and maturation.

Schizophrenia is a devastating and prevalent mental disorder with a huge impact on millions of patients and their families. Recent advances in human genetics have identified variations in multiple chromosomal regions and genes that contribute to the genetic risk for schizophrenia and may provide clues to schizophrenia pathogenesis (Coelewij and Curtis, 2018; Kirov, 2015) . Enormous progress was made in describing the genetic landscape underlying schizophrenia, resulting in the realization that changes in a large number of genes can contribute to schizophrenia (SCZ working group of PGC, 2014; Pardinas et al., 2018; Marshall et al., 2017; , Purcell et al., 2014 . This genetic heterogeneity supports the notion that schizophrenia is a multifaceted and multidimensional disorder that may not be understandable in terms of a single general pathophysiological mechanism, possibly because no such mechanism exists owing to a great degree of disease heterogeneity, or because many reported findings on mechanisms in schizophrenia were based on underpowered and preliminary experiments.

A large number of 2p16.3 CNVs were described that cause different deletions of chromosomal DNA, but affect expression of only a single gene, NRXN1 (Marshall et al., 2017 , Castronovo et al., 2020 . These CNVs strongly predispose to schizophrenia. Although rare, NRXN1 CNVs affect thousands of patients. As a result, NRXN1 CNVs are at present the most prevalent known single-gene mutation associated with schizophrenia, although other CNVs involving multiple genes (such as 22q11.2) are more prevalent among schizophrenia patients (Marshall et al., 2017; Hu et al., 2019; Kasem et al., 2018) . In addition, NRXN1 del CNVs predispose to other neuropsychiatric and neurodevelopmental disorders (Lowther et al., 2017 , Castronovo et al., 2020 . Strikingly, our previous study suggested that heterozygous deletions of NRXN1 in human but not in mouse neurons cause a distinct and robust impairment in neurotransmitter release without changing the development or the morphology of neurons (Pak et al., 2015) . This was an exciting finding because it identified a robust phenotype caused by a schizophrenia-associated mutation that could potentially be used to gain further insight into schizophrenia pathophysiology and may even be amenable for drug development. Given the current reproducibility crisis, however, it was necessary to independently assess these results. Thus, the present study aimed for a multidimensional validation of these findings. Specifically, the present study had 6; objective 3) . Furthermore, when we compared human and mouse neurons obtained from stem cells with the same trans-differentiation protocol and carrying essentially the same NRXN1/Nrxn1 deletion, we observed a robust phenotype only in the human neurons but not the mouse neurons (Fig. 7) . Thus a human-specific functional phenotype is produced by the same gene in comparable experimental conditions, suggesting that differences between human and mouse genes may not be solely inherent in the genes themselves, but mediated by the genetic context, which is of potential importance in studying the pathophysiological role of human gene mutations (objective 4). Overall, our results therefore establish that in human neurons, heterozygous NRXN1 deletions, independent of whether they result from spontaneous unfortunate CNVs or from genetic engineering in stem cells, produce a robust but discrete functional phenotype that consists of a decrease in release probability without a change in synapse numbers of neuronal development. The definition of this synaptic impairment enables a mechanistic approach to a better understanding of schizophrenia pathophysiology.

Besides establishing NRXN1 del neurons as a robust model system and resource for schizophrenia studies, our results provide new information on the pathogenetic mechanisms involved ( Fig. 8-10 ; objective 5). The electrophysiological studies revealed that NRXN1 del neurons exhibit a selective impairment in neurotransmitter release without changes in other neuronal properties, suggesting a circumscribed dysfunction of synaptic transmission that is related to the release machinery. Consistent with this conclusion, we uniformly detected in NRXN1 del neurons an increase in CASK protein levels, in agreement with the notion that a decrease in NRXN1 signaling via CASK may lead to a compensatory increase in CASK levels (Fig. 8) . In the RNAseq studies, few major changes in gene expression were detected, possibly because NRXN1 is not in itself directly involved in regulating gene expression (Fig. 9, 10) . However, several interesting gene expression changes were noted. In all patient-derived NRXN1 del neurons, we detected a significant upregulation in NNAT expression. NNAT's exact function is currently unknown, but it appears to perform a regulatory role in secretory tissues, especially the brain (Joseph, 2014; Pitale et al., 2016) . We did not detect an upregulation of NNAT in NRXN1 del neurons derived from genetically engineered iPS cells, possibly because NNAT is an imprinted gene and the imprinting state may differ among iPS cells. The fact that NNAT is not increased in the engineered NRXN1 del neurons indicates that the increase in NNAT expression is not responsible for the synaptic phenotype that is uniformly observed in the patient-derived and engineered NRXN1 del neurons. Finally, the RNAseq experiments identified a prominent change in the expression of KYAT3, a gene linked to schizophrenia pathogenesis that mediates the biosynthesis of kynurenic acid. Kynurenic acid can act as an anticonvulsant, is an antagonist of ionotropic glutamate receptors, and has been proposed to be critically involved in schizophrenia pathogenesis (Erhardt et al., 2007) . In particular, kynurenic acid is thought to suppress NMDA-receptor function, which in turn is also regulated by NRXN1 (Dai et al., 2019) and may be decreased in schizophrenia (Nakazawa and Sapkota, 2020) .

Thus, an overall picture emerges whereby NRXN1 dysfunction might operate in the same pathway as kynurenic acid in schizophrenia pathogenesis.

In summary, we here present evidence that the robust synaptic phenotype observed upon NRXN1 deletions in human neurons is disease-relevant since it is fully reproducible in patient-derived neurons. Moreover, in generating iPS cells with a conditional NRXN1 deletion we have produced a generally available resource for studying schizophrenia pathophysiology. Finally, we have shown that the heterozygous NRXN1 deletion phenotype is not observed in mouse neurons under the same experimental conditions, strengthening the rationale for studying disease phenotypes in human neurons. The stage is now set to use these findings and reagents for insight into the molecular mechanisms involved.

Patient and control cohort. Case and control donors for this study are described in Table 1 . They were drawn from the Molecular Genetics of Schizophrenia (MGS) European-ancestry cohort (Shi et al., 2009 ; see also information on availability of non-identified clinical information, and biomaterials from the National Institute of Mental Health, nimhgenetics.org, studies 6, 29 and SZ0 for cases, and study 29 for controls; and of genome-wide association study data from dbGAP, accession numbers phs000021.v3.p2 for the "GAIN" and phs000167.v1.p1 for the "NONGAIN" subsets of the genotypes). Cases and controls were age 35-51 at collection. Selection criteria for controls included:

polygenic risk score in the bottom 20th percentile based on the Psychiatric Genomics Consortium schizophrenia GWAS data in 2013; employed, but not in a professional/managerial position; married;

high school education but without >2 years of college; denied psychiatric or substance dependence disorder based on an online screening online questionnaire (nimhgenetics.org). These individuals are predicted to be at low genetic risk of schizophrenia, and not outliers in cognitive function. For the overall project of which this work was a part, iPS cell lines were derived for 5 schizophrenia cases each with NRXN1 deletions, 22q11.2 deletions or 16p11.2 duplications, and for 6 controls. Note that aliquots of all subclones of these 21 iPS cell lines are available from NRGR for all 21 individuals, including the six clones used in the present work and the 4 subclones of the engineered conditional NRXN1 mutation line described below (Table 1) .

GWS and CNV-calling methods. For the six individuals (three NRXN1 deletion schizophrenia cases and three controls) for whom data are reported in this paper, 33-41x whole-genome DNA sequencing (WGS) was carried out from genomic DNA extracted from whole blood by Macrogen USA (Rockville, MD) on the Illumina X Ten platform, using Trueseq Nano library prep kits for paired-end 150bp reads.

Sequencing reads were aligned to the human reference genome (hg19) using BWA (Li et al., 2009 Human NRXN1 exon 19 PCR genotyping. Genotyping on engineered iPSC line was per-formed using the following primers described previously (Pak et al., 2015) : The cells were fed with E8 for three or more days, then selected with E8 + 200 µg/ml Geneticin (Gibco) during expansion of an engineered pool up to T-150 scale, followed by cryopreservation.

The engineered lines were confirmed to have a normal karyotype before induced neuron generation. Lentivirus. Lentiviruses were produced as described (Pang et al., 2010 , Pak et al., 2015 . Differential Gene Expression Analysis. Prior to DEG analyses, we removed gene types that are prone to mapping errors (fusion genes, antisense RNA genes, overlapping and intronic transcript genes) or incomplete capture by standard library prep (non poly-adenylated short RNA genes, e.g., micro-RNA and small nucleolar genes). For each cell type separately, we determined two values for each gene: max (the highest logTPM value for any culture in the set); and min (the minimum value in the set). Each DEG analysis was restricted to genes with max ≥ 2 and min > 0.1 in the relevant set as described below. The rational for the min criterion was that genes with "near-zero" values ≤ 0.1 yielded inflated DEG statistics, e.g., in genes with no near-zero values, for 9 patients vs. 9

controls, |log2(fold-change)| > 0.2 or > 0.8 was observed in 15.20% and 0.4%, 60.1% and 6.1% of genes with one near-zero value, and in most cases, that value was inconsistent with the other replicates for the same line, suggesting technical artifacts.

DEG analyses used "moderated" t-tests as implemented in limma v3.42.2 (Law et al., 2014) with the limma-trend procedure (fit = eBayes(fit, trend = TRUE)) that uses an Empirical Bayes Model to effect "shrinkage of the estimated sample variances towards a pooled estimate" (Smyth 2004) The following primers were used for genotyping to discriminate between different alleles: Genotyping on Nrxn1 cKO mouse ES cells was performed using following primers: F: to support glia viability, and iN cells were assayed on day14 in most experiments.

Immunofluorescence labeling experiments. Cultured iN cells were fixed in 4% paraformaldehyde + 4% sucrose in PBS for 20 min at room temperature, washed three times with 0.2% Triton X-100 in PBS (PBST) for 10 min each wash at room temperature. Cells were incubated with blocking buffer (PBS containing 2% normal goat serum (Sigma-Aldrich) and 0.02% Sodium-azide) for 1 hr at room temperature. Primary antibodies, diluted in blocking buffer, were applied for 1 hour at room temperature then washed 3x with PBST. Secondary antibodies were diluted in blocking buffer and applied for 1 hr at room temperature. Immunolabeled neurons were then mounted on glass slides with Fluoromount-G mounting medium. The following antibodies were used for our analysis: chicken anti-MAP2 (1:1000; ab5392; Abcam) and rabbit anti-synapsin (1:200; E028; TCS). Alexa-546-and Alexa-633-conjugated secondary antibodies were obtained from Invitrogen.

Image Acquisition and Quantification of Neurite outgrowth and Excitatory Synapses. Image acquisition and quantification were performed as described (Pak et al., 2015) . Cells were chosen at random from three or more independent cultures. Images were taken from at least three coverslips per experiment. Fluorescent images were acquired at room temperature with an inverted Nikon A1

Eclipse Ti confocal microscope (Nikon) equipped with a 60 x objective (Apo, NA 1.4) and operated by NIS-Elements AR acquisition software. Images were taken at 1024x1024 pixel resolution with a z-stack distance of 0.5 μm. Laser power and photomutiplier settings were set so bleed-through was negligible between channels. Within the same experiment, the same settings for laser power, PMT gain, and offset were used. These settings provided images where the brightest pixels were just under saturation. General Analysis was performed with NIS-Elements Advanced Research software (Nikon). For quantifications, images were thresholded by intensity to exclude background signals and puncta were quantified by counting the number of puncta whose areas ranged from 0.1-4.0 mm 2 .

For each experiment, at least 15 cells per condition were analyzed and the mean and SEM were calculated. Data shown represent the average of the mean values from at least 3 independent experiments. For analysis of dendritic arborizations, neurons were sparsely transfected with AAV-Syn-EGFP to obtain fluorescent images of individual neurons. Images were acquired using Nikon A1 Eclipse Ti confocal microscope (Nikon) equipped with a 20× objective and operated by NIS-Elements AR acquisition software. Images from 20 to 30 neurons per condition (per n=1) were reconstructed using the MetaMorph neurite application, scoring for total neurite length, neurite branch points, and soma area.

For image acquisition and morphometrics performed at Rutgers site, images were taken on a Zeiss LSM700 confocal microscope. The number of Synapsin positive puncta density corresponding to MAP2 dendrites, soma sizes, puncta sizes and dendritic tree area (MAP2) were identified using Intellicount (Fantuzzo et al., 2017) , a high-throughput, automated synapse quantification program. The system uses adaptive segmentation for selection of synaptic puncta (Synapsin) as well as dendritic boundaries of MAPs signals. The primary processes of dendrites were counted manually after MAP immunostaining. All image quantifications were conducted blindly.

Electrophysiological recordings Electrophysiological recordings in cultured iN cells were performed in the whole cell configuration as described previously (Maximov and Südhof, 2005; Zhang et al., 2013; Pak et al., 2015) . Patch pipettes were pulled from borosilicate glass capillary tubes (Warner to around 80%. In these experiments, first minimal currents were introduced to hold membrane potentials around −70 mV (typically 1-10pA). To elicit action potentials, increasing amount of currents were injected (typically from -10 pA to +60 pA, with 5 pA increments) for 1s in a stepwise manner.

Vrest was determined after break-in but before the injection of any currents, as the mean steadystate voltage between any spontaneous sodium spikes. Action potential parameters were determined by analyzing the first AP in any triggered trains of APs. Input resistance (Rin) was calculated as the slope of linear fits of current-voltage plots generated from a series of small current injection steps. To determine whole-cell membrane capacitance (Cm), square wave voltage stimulation was used to produce a pair of decaying exponential current transients that were each analyzed using a least-squares fit technique (Clampfit 9.02). Data were digitized at 10 kHz with a 2 kHz low-pass filter using a Multiclamp 700A amplifier (Molecular Devices). Data were analyzed using Clampfit 9.02 software. For all electrophysiological experiments, the experimenter was blind to the condition/genotype of the cultures analyzed. All experiments were performed at room temperature.

Data and reagent availability. (Note to reviewers: the following arrangements are either completed, or are in process. If this paper is reviewed and we are invited to submit revisions, final accession details will be included.) All biomaterials and data are available to the scientific community as follows:

(i) The 3 schizophrenia/NRXN1 del and 3 control iPS cell lines ( AEU oversaw RNA-seq processing and data analysis CP, DFL, MW, and TCS planned the experiments, analyzed the data, and wrote the paper in consultation with all co-authors.

JLD is an employee and shareholder of Eli Lily and Company.

MM reports personal fees from FUJIFILM Cellular Dynamics, Inc. during the conduct of the study. The frequency but not amplitude of spontaneous mEPSCs is decreased ~2-fold in NRXN1 del neurons derived from schizophrenia patients compared to controls (A1 and A2) (top, representative traces of mEPSC recordings; bottom left, cumulative probability plots of interevent intervals and summary graphs of the mEPSC frequency; bottom right, cumulative probability plots and summary graphs of the mEPSC amplitude). mEPSCs were recorded in the presence of tetrodotoxin (TTX, 1 µM) at Stanford University.

B. Similar as A, but spontaneous EPSCs (sEPSCs) were analyzed in the absence of TTX with an independently generated set of neurons at Rutgers University (cross-lab and cross-platform validation).

Data are means ± SEM (numbers in bars represent number of cells/experiments analyzed). Statistical analyses were performed by Student's t-test for the bar graphs, and by Kolmogorov- Data are means ± SEM (numbers in bars represent number of cells/experiments analyzed). Statistical analyses were performed by Student's t-test for the bar graphs (A, C) and by two-way RM ANOVA with Tukey's post-hoc test for summary plots (B, D) , comparing NRXN1 del neurons to controls (* = p<0.05; ** = p<0.01; *** = p<0.001; non-significant comparisons are not indicated).

Neurons were analyzed at 6 weeks in culture. Table 2 and Fig. S4 for a description of the newly engineered conditional NRXN1 mutation.

Conditional heterozygous NRXN1 deletion in neurons generated from a newly engineered iPS cell line causes a nearly 2-fold decrease in mEPSC frequency but not amplitude (left, cumulative probability plot of the mEPSC interevent intervals and summary graph of the mEPSC frequency; right, cumulative probability plot and summary graph of the mEPSC amplitude).

C. Sample traces of evoked EPSCs elicited by closely spaced pairs of action potentials (intervals: 20 ms, 50 ms, 100 ms, 500 ms, 1000 ms) to measure both the EPSC amplitude and the EPSC paired-pulse ratio from neurons derived from the NRXN1 cKO iPS cell line.

Conditional heterozygous NRXN1 deletion in engineered neurons induces a >2-fold decrease in the amplitude of evoked EPSCs (summary graphs of the amplitude of the first evoked EPSC).

Conditional heterozygous NRXN1 deletion in engineered neurons causes a >2-fold reduction in the paired-pulse depression of evoked EPSCs (summary plot of the paired-pulse ratio (PPR), measured as the amplitude ratio of the second over the first evoked EPSC, and plotted as a function of the inter-stimulus interval).

Data are means ± SEM; numbers of cells/cultures analyzed are shown in the bar diagrams. Statistical significance was evaluated with the Kolmogorov-Smirnov-test (cumulative probability plots), onetailed t-test (summary graphs), or two-way ANOVA (PPR plots) (*, p<0.05; **, p<0.01). All recordings were performed from 6 week-old neurons. Heterozygous Nrxn1 deletions in mouse neurons cause no significant changes in mEPSC frequency or amplitude. ES cells isolated from heterozygous Nrxn1 cKO mice were trans-differentiated into neurons using Ngn2 expression, infected with lentiviruses expressing Cre (to induce the heterozygous Nrxn1 deletion) or ΔCre (control), and analyzed at DIV14-16 (A, representative mEPSC traces; B, cumulative probability plots of the mEPSC interevent intervals [inset = summary graph of the mEPSC frequency] and of the mEPSC amplitude [inset = summary graph of the mEPSC amplitude]).

Heterozygous Nrxn1 deletions in mouse neurons also cause no significant impairments in evoked synaptic transmission. EPSCs were recorded in response to two sequential stimuli separated by defined inter-stimulus intervals (20 ms, 50 ms, 100 ms, 500 ms, 1000 ms) in neurons prepared as decribed in A (C, representative traces; D, summary graphs of the amplitudes; E, summary plot of the paired-pulse ratio of the EPSCs displayed as a function of the interstimulus intervals).

Numerical data are means ± SEM (number in bars or brackets show number of cells/cultures analyzed). No significant differences were detected between test and control conditions using an unpaired, one-tailed t-test (B & D) or a two-way ANOVA (E). 

Quantifications of protein levels in the two pairs of NRXN1 del and control neurons reveal that apart from a large increase in CASK protein levels and a decrease in neurexin levels, no other major changes were detected in the levels of the analyzed proteins. All analyses were carried out using quantifications with fluorescently labeled secondary antibodies and TUJ1 as a loading control.

Representative blots (C) and quantifications (D) reveal that the engineered conditional NRXN1 deletion in an iPS cell background (1215sub) also causes an increase in CASK protein levels. Experiments were performed as in A & B.

Data are means ± SEM (n = 3-5 independent cultures). Statistical analyses were performed by Student's t test comparing test samples to the control (*p<0.05). Table 1 ), we selected 12,910 genes meeting all QC criteria (see Methods) including log2(TPM+1) expression values ≥ 2 in at least one of the 18 cultures and >0.1 in all cultures, and excluding small non-poly-adenylated RNA genes; alternative spliced, readthrough and fusion genes to minimize mapping errors. Moderated paired ttests were performed using limma-trend, pairing the patient and control culture that were generated at the same time on the same plate. We examined these results in two complementary ways.

A. The volcano plot shows p-value in negative logarithmic scale and log2(fold-change) results for all genes, highlighting those with the largest observed effects: |log2(fold-change)| ≥ 0.8 and/or -log10(Pvalue) ≥ 3.

Shown are the two most significant Gene Ontology terms in each GO category, from functional enrichment analysis of the patient vs. control DEGs. A "broad" set of DEGs (403 up-regulated and 296 down-regulated) was selected with |log2(fold-change)| > 0.2 and p-value < 0.05, for this analysis (see Supplementary Data for a list of all DEGs) performed with ToppFun (ToppGene.org). There were 12,880 genes with a valid Gene Symbol, of which 11,781 (the Test Set) were annotated to at least one GO term, including 338 up-regulated and 257 down-regulated DEGs. Separately for upand down-regulated DEGs, and for each term containing 10-5,000 Test Set genes, we performed a hypergeometric test (to determine the probability, given n DEGs and m Test Set genes in the term, of observing n or more of 257 or 338 DEGs with the expectation of m of 11,781 Test Set genes). For each main category, the 5% significance threshold was Bonferroni-corrected for the N of terms in that category (1,105 for Molecular Function; 5,962 for Biological Process; 807 for Cellular Component). There were no significant terms for upregulated DEGs. For down-regulated DEGs, there were 3, 32 and 10 significant terms respectively for the three categories; for a 5% Benjamini-Hochberg FDR threshold, there were 12,278 and 34 terms (note there is substantial overlap among terms, making the test conservative).

NNAT expression is increased 10-100 fold in NRXN1 del neurons derived from three schizophrenia patients. Relative expression levels of the indicated mRNAs were measured by quantitative RT-PCR in neurons derived from three pairs of schizophrenia patients' NRXN1 del and control iPS cells. Average RQ values (normalized to MAP2 mRNA) were converted to a ratio of control/mutant and plotted on a logarithmic scale. Note that only a limited number of mRNAs were analyzed for pair #2 and only NNAT for pair #3 because of its enormous increase in expression in the NRXN1 del neurons.

F. NNAT expression is also increased several-fold in the iPS cells derived from the three NRXN1 del schizophrenia patients analyzed in this study (see B-D for measurement methods).

G & H. NNAT expression is not increased in neurons trans-differentiated from genetically engineered stem cells with conditional NRXN1 del mutations. Both neurons derived from previously described ES cells with a conditional NRXN1 truncation (cTr) that mimics a schizophrenia-associated NRXN1 del mutation (G; Pak et al., 2015) and neurons derived from a newly engineered iPS cell line (1215) that causes a conditional deletion of NRXN1 (H; Fig. S4 ) were analyzed (see B-D for measurement methods). Note that the modest but highly significant increase in RIMS1 (an active zone protein) expression observed in the patient-derived NRXN1 del neurons is fully reproduced in the engineered lines.

Data in C-H are means ± SEM (n ≥ 3 independent cultures for all experiments). Statistical significance was evaluated by Student's t-tests (* = p<0.05; ** = p<0.01; *** = p<0.001).

Synaptic computation

High rate of disease-related copy number variations in childhood onset schizophrenia

Β-neurexins control neural circuits by regulating synaptic endocannabinoid signaling

Presynaptic neurexin-3 alternative splicing trans-synaptically controls postsynaptic AMPA receptor trafficking

Distinct circuit-dependent functions of presynaptic neurexin-3 at GABAergic and glutamatergic synapses

Near-optimal probabilistic RNA-seq quantification

A tripartite protein complex with the potential to couple synaptic vesicle exocytosis to cell adhesion in brain

Phenotypic spectrum of NRXN1 mono-and bi-allelic deficiency: A systematic review

High-efficiency transformation of mammalian cells by plasmid DNA

ToppGene Suite for gene list enrichment analysis and candidate gene prioritization

Conditional Deletion of All Neurexins Defines Diversity of Essential Synaptic Organizer Functions for Neurexins

Deletions of NRXN1 (neurexin-1) predispose to a wide spectrum of developmental disorders

Mini-review: Update on the genetics of schizophrenia

Human CASK/LIN-2 binds syndecan-2 and protein 4.1 and localizes to the basolateral membrane of epithelial cells

Alternative splicing of presynaptic neurexins differentially controls postsynaptic nmda and ampa receptor responses

The kynurenic acid hypothesis of schizophrenia

Comparative phylogenetic analysis of blcap/nnat reveals eutherian-specific imprinted gene

Intellicount: High-Throughput Quantification of Fluorescent Synaptic Protein Puncta by Machine Learning

Neuronal impact of patient-specific aberrant NRXN1α splicing

De novo mutations in schizophrenia implicate synaptic networks

Gene expression elucidates functional impact of polygenic risk for schizophrenia

Recurrent rearrangements in synaptic and neurodevelopmental genes and shared biologic pathways in schizophrenia, autism, and mental retardation

CASK mutations are frequent in males and cause X-linked nystagmus and variable XLMR phenotypes

A transcriptome-wide association study implicates specific pre-and post-synaptic abnormalities in schizophrenia

Presynaptic short-term depression is maintained during regulation of transmitter release at a GABAergic synapse in rat hippocampus

Sheng M Direct interaction of CASK/LIN-2 and syndecan heparan sulfate proteoglycan and their overlapping distribution in neuronal synapses

Genetic insights and neurobiological implications from NRXN1 in neuropsychiatric disorders

Rare Copy Number Variants in NRXN1 and CNTN6 Increase Risk for Tourette Syndrome

Enlisting hESCs to Interrogate Genetic Variants Associated with Neuropsychiatric Disorders

Neuronatin gene: Imprinted and misfolded: Studies in Lafora disease, diabetes and cancer may implicate NNAT-aggregates as a common downstream participant in neuronal loss

Peg5/Neuronatin is an imprinted gene located on subdistal chromosome 2 in the mouse

Neurexins and neuropsychiatric disorders

The penetrance of copy number variations for schizophrenia and developmental delay

CNVs in neuropsychiatric disorders

voom: Precision weights unlock linear model analysis tools for RNA-seq read counts

Fast and accurate short read alignment with Burrows-Wheeler transform

Mutation analysis of the NRXN1 gene in a Chinese autism cohort

Neurexins cluster Ca2+ channels within the presynaptic active zone

Molecular characterization of NRXN1 deletions from 19,263 clinical microarray cases identifies exons important for neurodevelopmental disease expression

Contribution of copy number variants to schizophrenia from a genome-wide study of 41,321 subjects

Cnvs: harbingers of a rare variant revolution in psychiatric genetics

Monitoring synaptic transmission in primary neuronal cultures using local extracellular stimulation

Alpha-neurexins couple Ca2+ channels to synaptic vesicle exocytosis

The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data

MetaSV: an accurate and integrative structural-variant caller for next generation sequencing

CNV analysis in Tourette syndrome implicates large genomic rearrangements in COL8A1 and NRXN1

Mutations of CASK cause an X-linked brain malformation phenotype with microcephaly and hypoplasia of the brainstem and cerebellum

The origin of NMDA receptor hypofunction in schizophrenia

Patterns and rates of exonic de novo mutations in autism spectrum disorders

Dynamically Primed Synaptic Vesicle States: Key to Understand Synaptic Short-Term Plasticity

Common schizophrenia alleles are enriched in mutation-intolerant genes and in regions under strong background selection

Human neuropsychiatric disease modeling using conditional deletion reveals synaptic transmission defects caused by heterozygous mutations in nrxn1

Rapid generation of functional and homogeneous excitatory human forebrain neurons using Neurogenin-2 (Ngn2)

Concise Review: Progress and Challenges in Using Human Stem Cells for Biological and Therapeutics Discovery: Neuropsychiatric Disorders

Induction of human neuronal cells by defined transcription factors

Genetic heterogeneity of FG syndrome: a fourth locus (Fgs4) maps to Xp11.4-p11.3 in an Italian family

Neuronatin Protein in Health and Disease

Wayne Howse 1, Marina Gorbatyuk 1 Neuronatin Protein in Health and Disease

A polygenic burden of rare disruptive mutations in schizophrenia

BEDTools: a flexible suite of utilities for comparing genomic features

Analysis of copy number variations at 15 schizophrenia-associated loci

Current Status of Schizophrenia Gwas

Integrative genomics viewer

CASK aberrations in male patients with Ohtahara syndrome and cerebellar hypoplasia

De novo mutations revealed by whole-exome sequencing are strongly associated with autism

Biological insights from 108 schizophrenia-associated genetic loci

Large-scale copy number polymorphism in the human genome

Strong association of de novo copy number mutations with autism

Linear models and empirical bayes methods for assessing differential expression in microarray experiments

Synaptic neurexin complexes: a molecular code for the logic of neural circuits

Defining the Genetic, Genomic, Cellular, and Diagnostic Architectures of Psychiatric Disorders

Mapping autism risk loci using genetic linkage and chromosomal rearrangements

CASK participates in alternative tripartite complexes in which Mint 1 competes for binding with Caskin 1, a novel CASK binding protein

The putative Neuronatin imprint control region is an enhancer that also regulates the Blcap gene

Synaptic neurexin-1 assembles into dynamically regulated active zone nanoclusters

Direct conversion of fibroblasts to functional neurons by defined factors

Liprin-mediated large signaling complex organization revealed by the liprin-α/CASK and liprin-α/liprin-β complex structures

Hierarchical mechanisms for direct reprogramming of fibroblasts to neurons

Autismassociated SHANK3 haploinsufficiency causes Ih channelopathy in human neurons

Rapid single-step induction of functional neurons from human pluripotent stem cells

BioMed Research International 2015, 621690. potential [Vrest] and action potential height [AP height]; bottom panels, summary graphs of the action potential half-width

Statistical analyses by Student's t-test comparing NRXN1 del neurons to controls revealed no significant differences. All recordings were performed from 6 weeks after the start of differentiation

Smirnov tests for cumulative probability plots, comparing NRXN1 del to control neurons (* = p<0

All recordings were performed from 6 weeks after the start of differentiation

Flow-diagram of RNAseq experiments. RNA-seq libraries from total RNA were prepared induced human neurons (~DIV28) co-cultured with mouse glia (triplicate cultures for all iN cell cultures

duplicate for iPS cell lines), using NEBNext Ultra Directional RNA Library Prep Kit for Illumina (NEB)

0 (Bray et al., 2016) and a concatenated mouse-human index. From each Kallisto-processed mouse-human RNAseq sample, an expression matrix was constructed to generate human-specific TPMs (transcripts per million) by (1) counting the per million human reads per mouse-human sample; (2) summing up transcript-level TPM values per gene; and (3) converting gene TPM values to log2(TPM+1) values. 19,701 genes were selected with log2(TPM+1) ≥ 1 in at least one of the thirty samples, and the resulting gene-level values were per-sample quantilepolished to reduce sample-sample variability

For 15,000 post-QC genes meeting expression criteria (log[tpm+1] ≥ 2 in at least one 1 culture; and > 0.1 in all cultures) in one or more types of cells (non-engineered neurons [18n

ToppFun (ToppGene.cchmc.org) then determined how many of the 15,000 genes and of the DEGs in each cluster were annotated to each Gene Ontology term (in all categories: MF [Molecular Function], BP [Biological Process] and CC [Cellular Component]). Terms containing at least 10 of 15,000 genes were retained, and the 13,392 genes mapped to at least one of those terms were considered the Test Set. For each term, a hypergeometric p-value was computed to determine the probability of the observed N (or more) DEGs being annotated to the term, given the proportion of all 13,392 Test Set genes assigned to that term (see Supplementary Data for detailed results). The heat map shows the relative expression level of each gene for each culture. The annotation lists results for selected terms with the most significant p-values for each Cluster: the p-value; the number and percentage of cluster DEGs and of all Test Set genes assigned to the term, and the odds ratio (OR). Because GO categories annotate the same genes multiple times, the 5% alpha level for Bonferroni correction considered the number of

Bio-samples for this publication were obtained from NIMH Repository & Genomics Resource, a centralized national biorepository for genetic studies of psychiatric disorders (nimhgenetics.org). These were peripheral blood mononuclear cell samples from study