Loss of [dgr]-catenin function in severe autism

Autism is a multifactorial neurodevelopmental disorder affecting more males than females; consequently, under a multifactorial genetic hypothesis, females are affected only when they cross a higher biological threshold. We hypothesize that deleterious variants at conserved residues are enriched in severely affected patients arising from female-enriched multiplex families with severe disease, enhancing the detection of key autism genes in modest numbers of cases. Here we show the use of this strategy by identifying missense and dosage sequence variants in the gene encoding the adhesive junction-associated δ-catenin protein (CTNND2) in female-enriched multiplex families and demonstrating their loss-of-function effect by functional analyses in zebrafish embryos and cultured hippocampal neurons from wild-type and Ctnnd2 null mouse embryos. Finally, through gene expression and network analyses, we highlight a critical role for CTNND2 in neuronal development and an intimate connection to chromatin biology. Our data contribute to the understanding of the genetic architecture of autism and suggest that genetic analyses of phenotypic extremes, such as female-enriched multiplex families, are of innate value in multifactorial disorders. Autism is a common neurodevelopmental disorder with a profound sex-bias: four times more males than females are affected1 whereas disease recurrence risk to siblings of autistic females is larger than to siblings of affected males2. Both features can be explained through autism’s multifactorial inheritance where females are affected at higher biological thresholds of an underlying liability than males. Under this model, females escape the effect of deleterious mutations unless the alleles are severe and at key developmental steps. To accelerate discovery, we examine families with highest recurrence risk and, consequently, probably enriched for severe mutations in such genes. We hypothesize that one group of families that have this property, and yet are underrepresented in autism sequencing efforts, are those with two or more severely affected females (female-enriched multiplex families (FEMFs)). The first genes discovered in autism were through syndromes (Supplementary Table 1), such as Rett and fragile X syndromes3. Today, genomic analyses have definitively identified 12 genes, from an estimated 500 (ref. 4), with an excess of de novo or segregating mutations in typical isolated cases that are overwhelmingly male (Supplementary Table 1). Given such heterogeneity, it may be crucial to identify those genes whose mutations impart the greatest autism risk. Increased recurrence risk is associated with lower incidence (the ‘Carter’ effect), since any rare class must arise from higher genetic liability (Fig. 1a)5. Consequently, gene discovery in epidemiologically rarer classes, namely female gender, high phenotypic severity and familial cases, may be fruitful; this is further enhanced if we increase the genetic load by considering individuals who have all three features. These genetically loaded cases have either a greater number or frequency of deleterious alleles that are probably severe coding variants. This prediction arises from our studies of Hirschsprung disease, a neurodevelopment disorder of enteric nervous system ganglionosis. Hirschsprung disease is a multifactorial disorder with a sex ratio of 4:1 in favour of males and whose risk factors are gender, phenotypic severity, and familiality6. Although more than 15 genes for Hirschsprung disease have been identified, the major gene encodes the receptor tyrosine kinase RET, which harbours numerous rare loss-of-function coding and one common enhancer variant7. We estimated the proportion of 174 patients with Hirschsprung disease with damaging RET coding variants conditional on their having 3, 2, 1 or 0 risk factors, where higher risk categories were female gender, long segment aganglionosis and familiality (Fig. 1b), to show that rare classes are significantly associated with a higher proportion of deleterious alleles, varying linearly between 46% and 2% from the highest to lowest risk class (P = 3.1 × 10−6); the non-coding variant had the reverse trend. Therefore, exome sequencing in autism can be similarly efficient in FEMFs. Since female incidence of autism is 0.0016, fewer than 10% of families are multiplex and fewer than 10% are severe, FEMFs have a crude incidence of less than 1.6 × 10−5 and represent a rare autism disorder enriched for deleterious coding variants7. Here we demonstrate the utility of this strategy by exome sequence analyses of 13 unrelated females and identifying 18 candidate genes of which at least four, CYFIP1, DLG1, PLXNA3 and CTNND2, are of interest to autism aetiology. We have evaluated one of them, CTNND2 (the δ-2-catenin gene encoding the δ-catenin protein), in depth using a combination of genetic, genomic and functional studies to show that (1) CTNND2 harbours a significant excess of deleterious missense and copy number variants (CNVs) in autism; (2) these variants, by functional testing, are loss-of-function and affect Wnt signalling; (3) expression of CTNND2 is highest in the fetal brain and is highly correlated with other autism genes; and (4) CTNND2 correlated genes are enriched for chromatin and histone modification, as well as dendritic morphogenesis, functions. These results are consistent with the roles of CTNND2 in the formation of dendritic spines8, and the regulation of beta (β)-catenin in neurons9. Given the recent finding of de novo autism mutations in pathways regulating β-catenin (Supplementary Table 1), loss-of-function of CTNND2 is probably rate-limiting for dendritic morphogenesis and maintenance. We sampled 13 unrelated females, negative for deleterious variants in MECP2, from multiplex families who had severe autism (measured with the Autism Diagnostic Interview-Revised and the Autism Diagnostic Observation Schedule). Proband exomes were sequenced and analysed with sequence data from 71 European females (1000 Genomes Project; Extended Data Fig. 1). To identify pathogenic alleles, we focused on missense variants absent in public databases (dbSNP129, 1000 Genomes Project) and conserved to zebrafish, nonsense and canonical splice site variants. This led to 3,090 variants of interest in the combined 84 exomes within 2,516 genes, with 447 of these having two or more variants of interest; among them, the 13 autism cases harboured two or more variants of interest in 24 genes of which 18 reached significance (P < 1 × 10−4) (Supplementary Table 2). By searching their expression profiles (Supplementary Table 3), we identified four genes, with an excess of deleterious alleles, as candidates: CYFIP1, DLG1, PLXNA3 and CTNND2. On the basis of our previous genome-wide association study implicating chromosome 5p10, we followed up CTNND2 at this locus. CTNND2 harboured two deleterious variants, G34S and R713C, both of which were absent in 3,889 European controls (1000 Genomes Project and Exome Variant Server); G34S was present at a frequency of 5.3 × 10−4 in 1,869 African ancestry samples (Exome Variant Server) and in one Luhyan sample (NA19020) (Extended Data Fig. 2). To estimate their frequency, we genotyped 10,782 samples from the HapMap and autism collections: the only additional individuals with G34S were an affected female and her mother (SSC02696, SSC03276) from the Simons Simplex Collection (SSC). Principal component analysis on polymorphism data from individuals with G34S found that our autism cases were not of African ancestry, identifying a new ancestral origin for G34S (Extended Data Fig. 3). For R713C, only our FEMF samples were heterozygous. Next-generation CTNND2 sequencing in 362 additional females with autism (Extended Data Fig. 4) identified a total of seven variants (G34S, R713C and five new variants: P189L, P224L, G275C, R454H, T862M), of which four (G34S, G275C, R713C, T862M) were conserved to zebrafish (Fig. 2a and Supplementary Table 4). We also identified Q507P in an autistic male from 170 SSC probands. An identical analysis of 379 European ancestry control samples (1000 Genomes Project) yielded three variants after validation (R330H, D465N, A482T), one conserved to zebrafish. On aggregate, variants at these conserved CTNND2 residues are significantly more frequent in autism than in controls (P = 0.04 versus 1000 Genomes Project; P = 7.8 × 10−4 versus Exome Variant Server). We next assessed whether CNVs within CTNND2 were enriched in autism. First, from the literature, we identified six deletions and one duplication. Second, we identified two deletions and one duplication from the Emory University and Baylor College of Medicine clinical cytogenetics laboratories. Third, from the Autism Genetic Resource Exchange (AGRE), we identified two previously unreported valid deletions (Extended Data Fig. 5). Therefore, we detected 12 CNVs (ten deletions, two duplications), seven overlapping one or more exons (Fig. 3 and Supplementary Table 5). As a control, we searched the Database of Genomic Variants to identify 33 variants, with only two overlapping exons (58.3% in our 12 CNVs versus 6.1% in the Database of Genomic Variants; P = 5 × 10−4). This significant excess of exon-disruptive deletions suggests CTNND2 haploinsufficiency in autism. Most of our patients had an autism diagnosis; however, some probands were referred with a diagnosis of neurodevelopment disorder. To test whether CTNND2 CNVs may be enriched in neurodevelopment disorders generally, we assessed CNVs from 19,556 independent cases referred for clinical diagnostic studies and 13,898 controls from population-based studies11. Considering all dosage imbalances, we observed 25 instances in cases and three in controls, corresponding to an odds ratio of 5.9 (P = 4.10 × 10−4) (Extended Data Fig. 6 and Supplementary Table 6). The impact of loss-of-function (deletions, unbalanced translocations) mutations at this locus is significant, with an odds ratio of 14.7 (P = 8.28 × 10−5), with specificity for CTNND2 since the effect size is comparable for intragenic deletions (eight cases, one control; P = 0.059, odds ratio = 5.68) as for all CNVs. To assess the in vivo functional consequences of autism CTNND2 variants, we used a complementation assay in zebrafish embryos. Zebrafish have two genes for δ-catenin that are as divergent from each other (18.3%) as they are from humans (19.9%, 20.7%), at the protein level. We examined expression of both genes by PCR with reverse transcription (RT–PCR) at six developmental time points (Fig. 2b) and focused on ctnnd2b because it was expressed at all stages. Using a splice-blocking morpholino oligonucleotide targeting ctnnd2b, we injected one- to eight-cell embryos and analysed them at the eight- to ten-somite stage. Morphant embryos had gastrulation phenotypes consistent with abnormal Wnt signalling (shortened body axes, longer somites, and broad and kinked notochords) (Fig. 2c). RT–PCR of axin2 messenger RNA (mRNA), a direct target of canonical Wnt signalling12, from ten-somite ctnnd2b morphants, showed significant decrease (P < 0.01), reinforcing the hypothesis of defective Wnt signalling (Extended Data Fig. 7a). Specificity of the morpholino was tested by co-injection of wild-type mRNA to observe significant (P < 0.001) rescue (Fig. 2d). To investigate the effect of each variant on protein function, injection cocktails containing morpholino and mutant variants were injected and compared with rescue with wild-type mRNA: five variants (G34S, P189L, P224L, R454H, Q507P) were better than morpholino alone (P < 0.001) but worse than wild-type rescue (P < 0.001), implicating these as hypomorphic (Fig. 2d). One variant (R713C) was functionally null while G275C and T862M were benign, and all four controls were benign, demonstrating specificity. To preclude the possibility of mRNA toxicity, we injected mutant mRNA corresponding to all alleles and observed no significant differences in the gastrulation phenotypes (Fig. 2e). To replicate these findings with an in vivo assay querying Wnt signalling earlier in development, we assessed the consequences of ctnnd2b suppression on chordin expression during epiboly, whose ectopic expression is known in Wnt mutants13. Consistent with the role of chordin in Wnt-dependent dorsalization14, we observed shortening and widening of the chordin expression domain as well as loss of anterior-specific expression fields in ctnnd2b morphants (Extended Data Fig. 7b); this phenotype could be rescued by wild-type mRNA. Further, testing of two control alleles that scored benign in our mid-somitic assays (A482T, G810R) showed significant rescue (P < 0.001); the hypomorphic allele G34S rescued chordin expression to a level significantly worse than wild-type mRNA rescue (P < 0.001), while the null allele R713C did not rescue chordin expression (Extended Data Fig. 7c). Since CTNND2 can bind CTNNB1 (ref. 15), we tested this interaction with mutant CTNND2. Expression of green fluorescent protein (GFP)-tagged CTNND2 and Flag-tagged CTNNB1 revealed that wild-type CTNND2 could immunoprecipitate CTNNB1; however, its interaction with CTNNB1 was diminished upon expression of G34S or R713C (Extended Data Fig. 7d), suggesting in vivo Wnt phenotypes may result from attenuated CTNNB1–CTNND2 interaction. Finally, we asked if these major CTNND2 sequence variants could affect neuronal circuitry by using a well-established in vitro model system. Dendritic spines are the primary sites for excitatory synapse formation, and their dysregulation underlies many neuropsychiatric disorders16. To test if CTNND2 variants interfere with development and maintenance of spines, we prepared primary hippocampal neurons from embryonic day (E)18 rat embryos and introduced either GFP or GFP fusion to wild-type CTNND2 or to its mutant variants at day in vitro (DIV)8. At DIV15, neurons were fixed and analysed to assess spine density. We found that wild-type CTNND2 had a significantly higher spine density than GFP controls17. However, neurons expressing G34S had a significantly lower spine density than those expressing GFP or wild-type CTNND2. Neurons expressing R713C, on the other hand, had the same spine density as those expressing GFP but significantly less than the one that expressed wild-type CTNND2, suggesting a loss-of-function effect. In contrast, the A482T polymorphism had an effect similar to wild-type CTNND2 (Extended Data Fig. 8). To test if observed changes in spine density reflected changes in excitatory synapse number in the networks, we analysed excitatory synapses: that is, overlapping region between postsynaptic marker PSD95 and presynaptic marker vGluT1 in mouse hippocampal neurons at DIV14 (Fig. 4A). As with spine density, we found an increase in excitatory synapse number in neurons that overexpressed wild-type but not mutant CTNND2. Further, loss-of-function of CTNND2 led to a decrease in overall excitatory synapse density, as well as active synapses that expressed the GluA subunit of the AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid)-type glutamate receptors (Fig. 4B, C). Taken together, these results suggest that CTNND2 is critical to the formation and/or maintenance of synapses, in accord with other studies18, 19. Moreover, unlike wild-type CTNND2, the tested mutants failed to rescue the reduction in synapse density in CTNND2 null background, demonstrating loss-of-function. Therefore, G34S and R713C impair development and/or the maintenance of mammalian neural circuitry. To understand CTNND2 expression, we tested mRNA levels in 16 adult and eight fetal human tissues: CTNND2 expression was highest in the fetal brain (20× the adult brain) (Extended Data Fig. 9). Therefore, we used the Allen Brain Atlas of the Developing Human Brain microarray data to identify other CTNND2 co-expressed genes. We used the data normalized to 17,630 genes and linear regression on age and brain regions for estimating Pearsonian correlations between CTNND2 and all other genes (absolute correlation > 0.3, P = 2.84 × 10−6 given multiple comparisons). First, we performed pathway analysis on the 826 positively and 662 negatively correlated genes (Supplementary Table 7). The positive set was significantly enriched for genes encoding proteins localized to the cytoskeleton, cell junction, neuronal projection, with GTPase regulatory activity, and functioning in cell morphogenesis, chromatin modification, neuronal development and neuron projection formation. Of these, the role of CTNND2 in dendritic development and spine morphogenesis is known17 as well as its involvement in actin dynamics and GTPase regulatory activity20, 21. However, its role in chromatin modification is novel. The closest known function of CTNND2 to chromatin is based on CTNND2 binding to ZBTB33 (ref. 22), a protein regulating transcription and Wnt pathway genes23, and its possible nuclear localization and function24. Second, we searched for transcription factors that may regulate CTNND2: among the correlated genes we identified 75 of which PAX6 is the most biologically significant25. A Pax6 mutant rat show autism-related features26 and genetic variation disrupting PAX6 has been identified in individuals with autism27. Also, Pax6 can regulate Ctnnd2 expression in cells, including the binding of Pax6 to its promoter25, 28. We searched the correlated genes for autism29 (https://gene.sfari.org/autdb/) and intellectual disability candidates (Supplementary Table 7). Of 529 autism genes, 71 (61 positively, 10 negatively) were significantly correlated with CTNND2, representing significant enrichment (P = 2.83 × 10−6). Next, we examined the correlations between these 71 genes and CTNND2 (Fig. 5a) to find an intimate relationship between CTNND2 and autism genes. To interrogate the function of the 61 positively correlated genes, we again performed pathway analyses (Fig. 5b) to find significant enrichment of genes involved in dendrite morphogenesis (P = 2.96 × 10−3; PDLIM5, MAP2, SHANK1, CDKL5, DLG4) as well in chromatin modification (P = 2.96 × 10−3; HDAC3, HUWE1, CREBBP, EP300, YEATS2, EP400, ATXN7, HCFC1, ARID1B, NSD1). Our studies strongly implicate δ-2-catenin (CTNND2) as a critical gene in autism and an important neurodevelopmental protein given its role in FEMFs, functional association with other autism genes, cri-du-chat syndrome30 and other diseases31. Clearly, CTNND2 haploinsufficiency is common in autism and strongly associated with neurodevelopment disorder generally. Nevertheless, in the general population, the frequency of disease alleles we discovered is low (3.9 × 10−4 and 8.0 × 10−4 in individuals of European and African ancestry, respectively, in Exome Variant Server), consistent with their deleterious functional effects. CTNND2 is a plakoglobin/armadillo family member with identity to PKP4, CTNND1 and ARVCF. The armadillo domain is a key part of the protein that binds cadherins15, β-catenin15, presenilins 1 and 2 (ref. 32) and sphingosine kinase33. It also harbours a coiled-coil domain, a polyproline tract at amino acids 219–224 where src receptor kinases bind34, and a PDZ domain at the carboxy (C) terminus, which can bind Discs large homologue 4 (ref. 35) and erbin36. These features suggest that CTNND2 is important in neuronal actin dynamics and the cytoskeleton15, 34, as also supported by observations of induced branching of dendrite-like processes and enhanced dendrite morphogenesis by CTNND2 overexpression37. Importantly, CTNND2 can directly bind to actin37 and cortactin34, and act on the Rho family to induce filopodia within neurons20 and increase the number of dendritic spines17. Finally, we demonstrate a role of CTNND2 in canonical Wnt signalling through zebrafish analyses: although the precise mechanism is not understood, it can bind to proteins (GSK-3β, ZBTB33) that regulate Wnt signalling23 and transcription22, in concert with CTNNB1. The other novel CTNND2 function we implicate is its possible role in the nucleus, through its interaction with HDAC3 (Fig. 5). This is not unexpected, since CTNND2 can affect gene expression after nuclear translocation24. Furthermore, the armadillo family member p120ctn interacts with ZBTB33 (Kaiso) and the NCoR co-repressor complex containing HDAC3 (refs 22, 24, 38, 39). Thus, we hypothesize that CTNND2 may be a nucleo-cytoplasmic protein whose autism effect may arise from its cytoplasmic or nuclear loss-of-function or both. Published Ctnnd2 knockout mice8, 18, 19, and our analyses of their dendritic spines, give clues to the role of CTNND2 in autism and in cognition. Homozygote mice exhibit structural and functional abnormalities at the synapse, as well as impaired spatial learning and fear conditioning18, 19, with reduced levels of PSD-95, β-catenin associated with cadherin, and N-cadherin. The interaction of PSD-95 with CTNND2 was discovered as an important linkage to AMPA receptor binding protein and glutamate receptor interacting protein (GRIP)35. Our results confirm that CTNND2 is required for the maintenance of spine structures in vivo19, and stability of some key components of the synaptogenic machinery such as N-cadherins and PSD95 (refs 8, 18). We show that loss of spines and reduction in total levels of synaptic proteins in null mice reflect reduction in the number of functional excitatory synapses at the subcellular level. Interestingly, acute loss of δ-catenin in vitro impairs activity-dependent formation of spines40, reinforcing its importance in formation and maintenance of synaptic structures and cognitive functions. Studying FEMFs is unconventional for a complex disease where most mutations have small effects. Nevertheless, our data suggest that modest numbers of samples of rare extreme phenotypes, in contrast to large numbers of typical cases, can be important. Note, we identified 18 candidate genes among which at least three others are worthy of follow-up: CYFIP1 is in a 15q11-13 autism duplication, has altered expression in autism patients, interacts with FMRP and is involved in regulating dendritic spines through translational inhibition and actin dynamics41; DLG1 is a multi-scaffolding postsynaptic density protein lying within a 3q29 autism and/or intellectual disability deletion; PLXNA3 is known to alter dendritic spines and is a receptor for SEMA5A42, another autism gene10. The broader FEMFs hypothesis can thus be tested by sequencing larger numbers of cases for identifying genes critical to early brain development. We studied 13 unrelated females, 12 from FEMFs and one from a family with an affected girl and boy, from the AGRE49 and the NIMH collections (https://www.nimhgenetics.org/). Of these, 11 were of European, and one each of Hispanic and Native Hawaiian or Pacific Island ancestry. Studies of human subjects were approved by the Johns Hopkins Medicine Institutional Review Board (IRB NA_00015748). All protocols for animal care, use and euthanasia were reviewed and approved by the Institutional Animal Care and Use Committees of Johns Hopkins University (protocol MO12M412) and Duke University (protocol A229-12-08), and were in accordance with the Association for Assessment and Accreditation of Laboratory Animal Care guidelines. MECP2 sequencing. Each of the 13 individuals with autism was assessed for the four coding exons of MECP2 by PCR amplification of each exon and Sanger sequencing, performed at Beckman Coulter Genomics. The sequence traces were analysed in Sequencher version 4.7. Exome sequencing and read mapping. We analysed ten female autism cases and one HapMap sample (NA18507) using the Agilent SureSelect Whole Exome capture (38 Mb) and SOLiD 3+ and 4 technologies. For three additional female individuals with autism, sequencing used the Illumina TruSeq Whole Exome capture (62 Mb) and Illumina technology; all Illumina exome experimental steps were performed at the Illumina sequencing centre. The SOLiD and Illumina data were mapped to the human genome build 37 using the BFAST50 and BWA51 programs, respectively. Subsequently, the SAM output was converted to BAM output, duplicates were marked using Picard, and indel realignment and quality score recalibration were performed in GATK52. Variants were called across all exomes as well as 71 of the 1000 Genomes European female exomes. Each variant was annotated for genetic features using ANNOVAR53. Additional annotations included presence in 1000 Genomes, conservation to zebrafish, and presence in autism, autism candidate, or intellectual disability genes based on the published literature (ref. 29 and https://gene.sfari.org/autdb/). In total, we identified 37,424 non-synonymous, 486 stop gain, 32 stop loss, 35,549 synonymous and 273 splice variants. CTNND2 sequencing and read mapping. Three hundred and sixty-two females with autism (300 unrelated, independent), ten HapMap samples and a pooled individual sample replicated eight times were sequenced for all of CTNND2 coding exons. To amplify the 22 RefSeq exons and seven Ensembl exons in CTNND2, 87 amplicons were designed on the Fluidigm Access Array Targeted Resequencing platform. Amplification and addition of barcodes were accomplished as described in the manual using the bidirectional sequencing primer strategy. Next, each sample was purified using Agencourt AMPure beads following the manufacturer’s protocol. All samples were run on Agilent High Sensitivity DNA Chips on an Agilent 2100 Bioanalyzer to confirm size range and purity of the PCR product, followed by quantitative PCR for quantification before pooling all 384 samples for sequencing. The library was sequenced on a single lane of an Illumina HiSeq (100 base, single pass reads) instrument following the Illumina Sequencing Strategy as described in the Fluidigm Access Array manual. Each sample fastq read was assigned and partitioned to an amplicon based on its primer sequence using sabre (https://github.com/najoshi/sabre), then aligned only to that amplicon using BWA. All the resulting sam files for each individual were combined using Picard into one sample bam. Variants were called per individual using GATK and hard filters to get high-quality variants. To assess genotype quality of the HapMap samples, comparisons were made to HapMap genotype, OMNI genotype and 1000 Genomes data. The pooled individual replicate was also used for data quality control. Variant validation. All CTNND2 missense changes identified were sequenced by Sanger chemistry for validation. Primers were designed to cover the exons in which the variants were found. A portion of each PCR product was run on a 1.8% agarose gel for 1.5 h to check for the expected product size. Upon confirmation, the rest of the product underwent PCR purification. The purified samples were quantified by nanodrop, diluted to 25 ng μl−1 and sent to Beckman Coulter Genomics for Sanger sequencing. Subsequently, the reads were analysed in Sequencher version 4.7. CTNND2 TaqMan assay for the G34S and R713C variants. To test the frequency of the two CTNND2 variants we found in the autism exomes, we used TaqMan genotyping and created synthetic homozygous reference/mutant genotypes within a plasmid containing DNA from our patients, and used the patient DNA on each plate as a heterozygous control, to ensure that we would get three cluster plots in the SDS software. We ran a total of 11,788 reactions including 1,006 duplicates for which there was 100% genotype concordance. To genotype the G34S and R713C variants, custom TaqMan genotyping assays were designed for each variant. The five individuals (03C16092, 03C16094, SSC02696, SSC03276, NA19020) containing the G34S variant were assessed for ancestry by principal component analysis. A set of approximately 6,000 autosomal single nucleotide polymorphisms genotyped in common in all five samples (Affymetrix 5.0, Affymetrix 5.0, Illumina 1MDuo, Illumina 1MDuo, Illumina OMNI 2.5) were analysed using the Eigenstrat program. Genotypes from reference populations came from the CEU, YRI and CHB/JPT populations. TaqMan Copy Number Assays were used for validation of CNVs in the AGRE samples (AU066818, AU075604, AU1178301, AU051503) and their family members. Three or four assays were run for each CNV region in each sample. NA10836 was used as a calibration sample in CopyCaller software version 2.0 for a copy number of 2 in each region. After alignment with ClustalW, Molecular Evolutionary Genetics Analysis (MEGA) software was used to generate a phylogenetic tree by the neighbour-joining method; a total of 1,163 and 678 positions were used to assess percentage identity for orthologues and paralogues, respectively. Exome sequence. This study focused on variants of interest, defined as those that were absent in both dbsnp129 and 1000 Genomes low-pass sequencing data and were probably functionally deleterious (missense at residues conserved to zebrafish (human, chimp, dog, cow, mouse, rat and zebrafish from the University of California, Santa Cruz (UCSC) 46-way alignment), nonsense and canonical splice site changes). We compared each gene and its number of variants of interest with that expected on the basis of 10,000 replications of random sampling of 13 exomes from 71 female European controls. Genes having two or more variants of interest only in autism exomes were considered to be relevant candidates. Allen Brain Atlas Data. The Allen Brain Atlas Microarray (Affymetrix Human Exon 1.0 ST data microarray summarized to genes (n = 17,630 genes)) data set for the Developing Human Brain (8 weeks after conception to 40 years) was downloaded from the Allen Brain Atlas website on 24 February 2012. Linear regression was performed on the data set for age and brain region (in R software). Pearsonian correlations were calculated for each gene (X) and CTNND2 (Y); genes with absolute values greater than 0.3 were retained, corresponding to an experiment-wise P = 0.05 (17,630 comparisons) significance level. Pathway analyses were performed using GeneMania (http://www.genemania.org). DAVID analysis was performed on all correlated genes using the following categories: GOTERMs (biological process, cellular compartment and molecular function) for function and UCSC TFBS. Generation of human CTNND2 and mouse Ctnnd2 constructs. Human CTNND2 was initially cloned into the pDONR221 Gateway vector. Subsequently, the human DNA was cloned into the pCS2 vector for zebrafish assays and pcDNA 6.2 N-EmGFP–DEST vector (Gateway) for the neuronal assays. Zebrafish gastrulation assays. Using a splice-blocking morpholino targeting zebrafish ctnnd2, one- to eight-cell stage embryos were injected (N = 50–180) and live embryos at the eight- to ten-somite stage were analysed for gastrulation phenotypes including shortened body axes, longer somites, and broad and kinked notochords in morphant embryos. Embryos with phenotypes were then classified as class I or II depending on their severity (features of the convergence/extension phenotype include a shortened body axis, wider somites and a kinked notochord, with class I having one or two and class II having all three of these components, respectively). Specificity of the morpholino reagent was tested by co-injection of wild-type human CTNND2 mRNA. To test CTNND2 variants, injection cocktails containing morpholino and mutant human CTNND2 variants were injected and compared with the rescue condition of wild-type human CTNND2. Zebrafish chordin expression assay. Zebrafish embryos were harvested at 90% epiboly stage and fixed in 4% paraformaldehyde at 4 °C. Whole-mount RNA in situ hybridization was performed with a digoxigenin-labelled anti-chordin RNA probe synthesized by in vitro transcription (Roche). The chordin expression domain was measured in lateral view (‘L’ in Extended Data Fig. 7b). The middle point of the expression domain length and the centre of the embryo was linked with a dashed line (Extended Data Fig. 7b), along which the width of chordin expression domain (‘W’ in Extended Data Fig. 7b) was measured. Length:width ratio was calculated to quantify ectopic expression. Immunoblotting. Cells were transfected with CTNND2 and CTNNB1 expression constructs and harvested 48 h later. Protein lysates were immunoprecipitated using an anti-GFP antibody (Roche 11814460001) and immunoblotted with an anti-Flag antibody (Sigma F7425). Neuronal cultures and synapse analysis. Hippocampi from day E18 rats or E17 mouse embryos were prepared and maintained as described elsewhere54. At DIV8, the cells were transfected with GFP constructs (pcDNA6.2/N-EmGFP–DEST: alone, fused to wild-type or variant allele containing CTNND2) and 500 ng of pCAG-DsRed2 using Lipofectamine 2000 (Life Technologies). On DIV16, cells were fixed with a 4% paraformaldehyde/4% sucrose solution, followed by immunolabelling with primary antibodies against the appropriate target as described and their respective secondary antibodies. Neurons were imaged either on a Zeiss 510 confocal for spine analysis or on Zeiss epifluorescence microscope for synapse analysis, and analysed using ImageJ. Synapses were defined as puncta with overlapping signal between vGluT1 (Millipore catalogue number AB5905) and PSD95 (K28/43 clone from Neuromab) or vGluT1 and GluA (polyclonal antibody raised in rabbit against the C terminus of GluA subunit). To assess the expression of transfected CTNND2 and its mutant alleles in the δ-catenin null background, we selected five pre-defined areas of interest with constant area in each dendrite (Extended Data Fig. 10). Expression of ctnnd2a and ctnnd2b in zebrafish. To assess mRNA expression of the two zebrafish orthologues (ctnnd2a, ctnnd2b) of human δ-catenin, we performed PCR on normalized cDNA libraries (a gift of S. Maragh) from zebrafish at various developmental stages (50% epiboly, 75% epiboly, bud, 13 somite, 24 h after fertilization, and 3 days). CTNND2 expression analysis in human and mouse tissues. To examine expression of CTNND2 in different human tissues, the Human MTC cDNA Panel 1 (Clontech catalogue number 636742, lot number 7080213), Human MTC cDNA Panel II (Clontech catalogue number 636743, lot number 6040176) and the Human Fetal MTC cDNA Panel (catalogue number 636747, lot number 5090557) were analysed by a TaqMan gene expression assay (catalogue number Hs00181643_m1) for CTNND2 and for a pipetting control (GAPDH, catalogue number 4333764T). Each tissue was tested in triplicate. Subsequently, the threshold cycle (Ct) values were averaged and the ΔCt values calculated between all of the tissues and the adult brain. The fold difference from brain was calculated as (1/(2ΔCt)) for each tissue. Download references We acknowledge the participation of all of the families in the AGRE, NIMH and SSC studies that have been a model of public participatory research. The AGRE is a program of Autism Speaks and is supported, in part, by grant 1U24MH081810 from the National Institute of Mental Health. The SSC used here was developed by the following principal investigators: A. Beaudet, R. Bernier, J. Constantino, E. Cook, E. Fombonne, D. Geschwind, D. Grice, A. Klin, D. Ledbetter, C. Lord, C. Martin, D. Martin, R. Maxim, J. Miles, O. Ousley, B. Peterson, J. Piggot, C. Saulnier, M. State, W. Stone, J. Sutcliffe, C. Walsh, E. Wijsman. We thank the Allen Brain Atlas for use of their publicly available developing human brain expression data. Finally, we thank V. Kustanovich (AGRE) for helping with access to Autism Diagnostic Observation Schedule severity score data, D. Arking for sharing DNA from the SSC for Taqman genotyping, S. Maragh for zebrafish complementary DNA (cDNA) libraries and eef1a1l1 primers, A. Kapoor for discussions, Q. Jiang for the translation of ref. 43, and J. A. Rosenfeld, L. G. Shaffer, Y. Shen and B.-L. Wu for sharing CNV data sets. Sequencing services were provided by the Johns Hopkins University Next Generation Sequencing Center, Sidney Kimmel Comprehensive Cancer Center, Illumina Sequencing Services and the Johns Hopkins University Genetic Resources Core Facility. E.C.O. is a National Alliance for Research on Schizophrenia and Depression young investigator. N.K. is a Distinguished George W. Brumley Professor. This work was funded by grants from the Simons Foundation to A.C. and to N.K., NIMH grant MH095867 to M.E.T., NIMH grants 5R25MH071584-07 and MH19961-14 to D.M.D.L. (Malison), National Institutes of Health grant RO1MH074090 to C.L.M., NIMH grant R01MH081754 to A.C. and an Autism Speaks Dennis Weatherstone pre-doctoral fellowship (number 7863) to T.T. All sequence data have been deposited in the National Database for Autism Research in NDAR Study 367 and are available at http://dx.doi.org/10.15154/1171641.