Targeted Massively Parallel Sequencing of GWAS Association Peaks in a Case and Control Cohort Identifies Rare Autism Spectrum Disorder Risk Variants

Background: Genome-wide association studies (GWAS), copy number variation screens, and candidate gene studies have found no single factor accounting for a large percentage of genetic risk for autism spectrum disorders (ASD). This has led to the hypothesis that many genetic variants in potentially hundreds of genes contribute to ASD.  To identify these rare risk variants sequencing of  large datasets for rare or low frequency variants with potential functional significance to ASDs is essential.

Objectives: To identify new variants contributing to ASD risk by sequencing ASD candidate genes and  GWAS associated regions in a large dataset.

Methods: Candidate regions were chosen from GWAS Noise Reduction analyses of two autism datasets with prioritization of haplotype blocks based on the Truncated Product Method (TPM) (Hussman et. al., 2011). We designed an Agilent SureSelect probe set covering 17Mb corresponding to: 1) exons of 681 genes overlapping blocks with TPM p-values<0.05; 2) evolutionarily conserved regions in those genes plus 5kb from their transcriptional starts and ends; 3) evolutionarily conserved regions within non-genic blocks with TPM p-values<0.05; and 4) entire blocks with TPM p-values<0.01. Illumina HiSeq2000 reads were processed with the Burrows-Wheeler Aligner, genotypes called with the GATK Universal Genotype Caller, and annotated with SeattleSeq134, PolyPhen2, and SIFT.  We have completed analysis of 951 unrelated ASD cases and 872 controls. Among these, Eigenstrat stratification identified 598 white cases and 433 controls for further analyses. Gene-based association testing between targeted genes and ASD was performed with the Sequence Kernel Association Test (SKAT) utilizing a regression approach adjusting for covariates. A follow-up validation study of an additional 2000 individuals is underway.

Results: In the dataset, 87.9±6.3% of targeted bases are covered at least 10X with an average on target depth coverage of 78.1±25.9X. A total of 231,945 single nucleotide variants (SNVs) pass quality controls with a call rate of at least 99% across all samples. Of these, 21,263 SNVs are exonic, 6,512 are non-synonymous changes, and 4,028 are predicted to adversely affect protein function . We examined 69 genes that had been previously implicated in ASD.  These contain 47,111 SNVs: 3,579 exonic, 1,118 non-synonymous, and 656 damaging.  Of the damaging variants, 307 are unique to cases and several are in established ASD genes including 10 SNVs in CNTNAP2, 7 in MACROD2, 4 in NRXN1, and 5 in SEMA5A.  Moreover, we found 47 genes in which only cases have more than one rare (MAF<0.01) damaging alteration in a single individual, including the autism candidates, CNTN3, CDH8, IL1RAPL2, and PSD3. Rare variant association testing with SKAT identified nominally significant association of sets of rare exonic variants (p<0.05) with ASD in 16 genes including the voltage dependent calcium channel CACNA2D1 (p=0.00079) and the intracellular transport regulators GNPTAB (p=0.0049) and BICD1 (p=0.01).

Conclusions: These studies yield important findings regarding rare, potentially functional, SNVs found uniquely in ASD cases in previously identified and new candidate genes establishing targeted sequencing of ASD candidate genes/regions as a powerful method for discovery of new genetic variants contributing to ASD risk.