Function Based GWAS Identifies Novel Candidate Genes in Autism Spectrum Disorders

Background: Significant progress has been made in decoding the genetic architecture of autism spectrum disorder (ASD). ASD is highly heritable and polygenic, with risk spread across thousands of DNA variants throughout the genome (Klei et al. 2012). Genome-wide association studies (GWAS) utilizing sample sizes in the low thousands provide only weak and inconsistent evidence for risk alleles of main effect, suggesting that GWAS of ASD remains underpowered to identify individual risk alleles that survive multiple testing correction. Despite this challenge, studies from our group show that nominally associated results from a large family-based test of ASD association (Anney et al. 2010) are enriched for risk alleles that are also associated with gene expression in relevant brain tissues (i.e., expression quantitative trait loci; eQTL), but are not enriched for eQTLs in lymphoblastoid cell lines (Davis et al., 2012).

Objectives: Here we demonstrate a functional-unit based GWAS, which, instead of utilizing all SNPs on the array (the vast majority of which fall under the null expectation of no association), requires only a subset of SNPs that are coding (i.e., missense, nonsense or frameshift) as well those that have been show by previous studies to be cis- or trans-eQTLs involved in regulating gene expression in brain. Coding variants and eQTLs are then assigned to the genes that they affect. The main objective of this study is to use biologically informative genomic annotations to effectively reduce the number of multiple tests conducted, thereby maximizing our power to detect meaningful associations.

Methods: We applied our method to both a discovery and replication cohort, and conducted a meta-analysis including both sets of results. The discovery cohort included 654 probands from the Autism Genetic Resource Exchange (AGRE) and 1,593 unselected iControls. A second, independent replication cohort consisted of 1,679 probands from the Simons Simplex Collection (SSC) and 1,297 controls from the Study of Addiction: Genetics and Environment (SAGE). We constructed gene-level annotations for each known gene in the human genome by annotating each gene with regulatory variation (i.e., eQTLs) previously identified in the parietal cortex and the cerebellum as well as coding variants (nonsense, frameshift, and missense) within the gene.   A gene-level p-value was obtained using a combined statistic that integrates SNP-level evidence, derived from the traditional single-variant test, for association with the trait.  A total of 13,487 genes, represented by at least two variants, were annotated and included in each analysis.  A Bonferroni corrected p-value of 3.7x10^-6 was imposed to correct for the number of gene-based tests conducted.

Results: The most significant associations discovered were ZNF204P (2x10^-5) within the AGRE/iControl analysis and ITFG2 (2x10^-5) within the SSC/SAGE analysis, although neither result exceeded the threshold for genome-wide significance. However, upon meta-analysis, three genes were found to exceed genome-wide significance including ITFG2 (p=1.01x10^-6), MTHFD2 (p=1.05x10^-6), and ZNF2 (p=1.83x10^-6).

Conclusions: We present a novel gene-based approach to GWAS, which incorporates functional annotations including brain eQTLs and coding variants in a noise-reduction GWAS framework. We present meta-analysis data implicating three genes (ITGF2, MTHFD2, and ZNF2) and evaluate their potential as ASD candidate genes.