Selection of Candidate Genes

The list of genes screened in the S2D project was generated from a repertoire of approximately 5000 potential synaptic genes compiled from several synaptic lists from different synapse databases, including the Genes-to-Cognition (G2C) database (which includes an extensive list of postsynaptic genes)²⁰ and the Synapse database (SynDB, which uses Synapse Ontology algorithms to mine potential synaptic genes),²¹ and from an extensive list of synaptic vesicle genes.²² It also comprises genes identified through manual searches of PubMed that are either localized at the synapse or affect synapse-related functions (i.e., plasticity, axon or dendrite outgrowth, dendritic spine morphology, learning, and memory).

The sequencing data reported here were generated, using Sanger technology, from the screening of the coding regions and splice site junctions of 122 X-linked and 279 autosomal potentially synapse-related genes (see Table S1 available online) in 142 ASD and 143 SCZ subjects. The excess of X-linked genes is due to the S2D selection procedure, which sought to include all potentially synaptic X-linked genes because of their importance in neurodevelopmental diseases²³ and because ASD is more common in males than females.²⁴ The genes on the autosomes were largely ones that encode glutamate receptors (including NMDA receptors, AMPA receptors, kainate receptors, and metabotropic glutamate receptors), as well as genes that encode proteins that interact with them, in particular those complexed with the NMDAR, a majority of which were identified by large proteomic studies²⁵ and reported in the G2C database.²⁰ The autosomal gene list is composed of 203 genes from the glutamate receptor complex (23 known glutamate receptors and 180 of their synaptic interactors) and 73 genes implicated in synapse function and/or cognition, interaction with ASD genes, or because of their disruption in the context of small CNVs or balanced translocations in patients with ASD, SCZ, or mental retardation. An additional three genes were included for which mutations were reported to cause ASD, SCZ, or the related neurodevelopmental disease, nonsyndromic mental retardation, which is known to coexist with ASD. Mutations in which multiple reports have previously found associations with diseases that are not related to ASD or SCZ were eliminated. In addition, data generated from the resequencing of 39 of these 401 genes in 285 population control samples were used. The 39 genes were resequenced in controls after discovering a de novo or deleterious mutation in ASD or SCZ samples.

DNA Preparation, Sequencing, and Variant Identification

Genomic DNA was extracted from peripheral blood lymphocytes and/or lymphoblastoid cell lines with Puregene extraction kits (Gentra System). A panel of seven microsatellite markers was used to confirm parentage for all samples.²⁶ To overcome the issue of limited DNA material, we performed the gene screening on DNA isolated from an Epstein-Barr virus transformed lymphoblastoid cell line for most cases (n = 224). The rest were done on blood DNA (n = 61). The ASD cell line samples had been frozen or regrown a maximum of two times. For the SCZ cell lines, 59 DNA samples were acquired from Coriell and are available through a request to J.L.E.D. All unique variants (heterozygous in a single individual) detected during the screen were tested in the parents' DNA. All potential de novo variants (not seen in the parents) were reconfirmed by reamplifying the fragment and resequencing the proband blood DNA and both his/her parents with reverse and forward primers in order to eliminate PCR or sequencing artifact. All de novo variants identified originally from cell line DNA were retested in the subject DNA extracted from blood to rule out variations that could have occurred during production or growth of the lymphoblastoid cell line. An identity DNA test was performed to eliminate any cell lines and/or blood inconsistencies caused by sample identification errors or nonpaternity. Primers were designed with the Exon Primer program from the UCSC Genome Browser. PCR products were sequenced on one strand with Sanger technology, done at the Genome Quebec Innovation Centre on a 3730XL DNA Analyzer System. PolyPhred (v. 5.04), PolyScan (v. 3.0), and Mutation Surveyor (v. 3.10, SoftGenetics) were used for variant detection.

Estimation of Base Pairs Screened

To estimate the number of base pairs (bp) screened, we determined the amount of coding and noncoding (intronic, UTR) sequence screened for each gene based on our PCR amplicon designs (Table S2). Briefly, genomic intervals were calculated based on forward and reverse PCR primer sequences for each amplicon. Because the DNA sequence overlapping each PCR primer is not surveyed, these genomic intervals did not include those corresponding to each PCR primer. Furthermore, because the first ~30 bp of sequence from each sequence read are of low quality, we trimmed 30 bp from each genomic interval (usually at the forward primer end). Overlapping amplicons were merged to form a single genomic interval. Then each genomic interval was annotated as to coding and noncoding sequence for each gene with tools available on the refGene table from the UCSC Genome Browser.

We defined functional and nonfunctional sites sequenced as those in which a nucleotide change would or would not lead to an altered protein sequence, respectively. We estimated that 71.2% of coding sites were functional, whereas nonfunctional sites were estimated at 28.8% of the coding and 100% of the intronic sites.²⁷ The number of CpG sites was estimated as 2.8% for functional sites and 1% for nonfunctional sites.²⁸ The corrections for increased mutation rates in CpG regions was 10× the number of CpG sites, yielding an effective bases sequenced of 10 × CpG sites + non-CpG sites for both functional and nonfunctional sites.

Evaluation of False-Negative Mutation Calls

A subset of 71 of our ASD samples was also genotyped by Affymetrix 500K arrays. We tested our ability to detect heterozygous variants by comparing our heterozygous calls from the resequencing screen of these 71 samples with calls made for overlapping genotyped SNPs on the array (Table S3). Of the 1649 heterozygous calls made in 90 autosomal SNPs on the Affymetrix 500K chip overlapping our screened amplicons, we failed to detect 41 of those heterozygous calls in our resequencing survey, suggesting that our false-negative rate is ~2%.

Prediction of Missense Severity

The potential consequence of each missense variant was evaluated with the MAPP,²⁹ PolyPhen,³⁰ SIFT,³¹ and PANTHER³² programs. Orthologous protein sequence alignments were obtained with tools available on the Galaxy Browser website for the generation of MAPP scores.

Estimates of the Neutral Human Mutation Rate

To calculate human mutation rates, we estimated the total initial count of bases we resequenced in all cohorts as 458.8 Mb (Table 4; Subjects and Methods). This includes 230.6 Mb of protein-coding and 228.3 Mb of intronic sequence. In the ASD and SCZ cohorts, exonic material sequenced is 215,186,702 bases and intronic is 218,145,769 bases. The total number of replacement sites in the cases is 153,704,787 and the number of silent (synonymous and intronic) sites is 279,627,684. In the QNTS cohort, 15,422,960 bases were exonic and 10,122,419 bases were intronic. The number of replacement sites is 11,016,400 and the number of silent sites is 14,528,979.

We distinguished functional from nonfunctional sites based on the effect of a mutation on transcription or translation of the protein at a given position (see Subjects and Methods). When addressing whether there was an excess of functional DNMs³⁴ relative to nonfunctional or silent base pairs, we calculated the “effective bp count.”³⁵ Because mutations are 10 times more likely to occur at CpG sites than at non-CpG sites, we calculated the total number of CpG and non-CpG sites for both functional and nonfunctional sites²⁸ and calculated the effective bp count (non-CpG sites + 10 × CpG sites) to account for increased mutation rates in CpG sites (Table 4).

We calculated the neutral mutation rate by examining the number of DNMs found in nonfunctional sites in our ASD, SCZ, and QNTS samples. In total, we sequenced ~294 Mb of nonfunctional DNA, in which we observed four DNMs (Table 2; in genes GSN [MIM 137350], MAP2K1 [MIM 176872], BSN [MIM 604020], and ATP2B4 [MIM 604020]). Because these mutations are unlikely to be pathogenic (referred to here as “neutral”), they allowed us to directly estimate the rate of neutral point mutations. We estimated this to be 1.36 × 10⁻⁸ mutations per site per generation (95% Poisson confidence interval: 0.34 × 10⁻⁸, 2.7 × 10⁻⁸). These estimates of neutral mutation rates are similar to, and not significantly different from, the estimate of 2.5 × 10⁻⁸ derived from phylogenetic analyses^1,2 and the intergeneration estimate of 1.1 × 10⁻⁸ derived from next-generation sequencing data.⁴

Excess of Functional DNMs in ASD and SCZ Cohorts

If DNMs cause sporadic cases of ASD and SCZ, then DNMs will be more common in functional than in nonfunctional sites in our disease cohorts. Based on an observation of four DNMs in 294 Mb of neutral (silent and intronic) DNA, we expect 1.3 DNMs in the 96 Mb of nonsynonymous DNA sites to be resequenced in the cases with no family history of disease (65% of ASD cases and 60% of SCZ cases). However, among trios without family histories (Table 1), we observed six nonsynonymous DNMs surveyed in individuals, representing a significant enrichment of nonsynonymous DNMs (p = 0.003 in one-tail binomial test; p = 0.022 FET; Table 4). This excess remains significant even when we take into account that CpG dinucleotides mutate faster than other sites and are more common in exons than introns (p = 0.008 in one-tail binomial test; p = 0.032 FET; see Table 4 and Subjects and Methods).

If DNMs cause disease, we also expect point mutations with larger effects to be more frequent than expected in the disease group. Among our five nonsynonymous DNMs in trios with no family history, two are nonsense mutations (ratio 1:2.5), similar to previous estimates³ for DNMs causing Mendelian diseases (1:3.9) that are cataloged in the Human Gene Mutation Database (HGMD). Also, the ratio of synonymous to missense DNMs in the ASD and SCZ cohort is similar to that observed for HGMD.³ Under a neutral model,³ we would expect a ratio of 1 nonsense to 19.7 missense DNMs³ when only point mutations are considered. In HGMD, the number of missense to nonsense DNMs was significantly higher than the neutral expectation. Using a binomial test, our observed number of missense to nonsense DNMs was also significantly higher than the neutral expectation (p = 0.04), suggesting that some of the mutations are predisposed to be pathogenic. All of these observations suggest an excess of potentially disease-predisposing DNMs in the SCZ and ASD cohort. Taken together, these lines of evidence suggest that mutations with functional effects are overrepresented within the synapse genes sequenced in individuals showing sporadic ASD and SCZ.

We would like to thank all the families and individuals who participated in this study. We are thankful for the efforts of the members of the Genome Québec Innovation Centre Sequencing and Bioinformatic groups. This work was supported by Genome Canada and Génome Québec and received cofunding from Université de Montréal for the Synapse-to-Disease (S2D) Project, funding from the Canadian Foundation for Innovation to both G.A.R and P.A., and cofunding from the Ministré de Exploration, Innovation et Economique of Québec. G.A.R. holds the Canada Research Chair in Genetics of the Nervous System; P.A. holds career awards from the Fonds de Recherche Santé Québec and Génome Québec.