Autism Genome Project

Background: Autism is a neurodevelopmental disorder that affects approximately 1 in 150 individuals and is characterized by deficits in reciprocal social interaction, communication and patterns of repetitive behaviors and restricted interests. Twin and family studies indicate high heritability, but evidence supports a highly complex architecture for the underlying genetic etiology. The Autism Genome Project (AGP) was formed to facilitate gene identification by uniting investigators and family data. AGP Phase I involved genome-wide (GW) analysis of linkage and copy number variation (CNV) in >1100 multiplex families. Linkage analysis revealed promising loci on 11p and 15q, with gender and ancestry influencing signals at these and other loci. Copy number analysis of 10k data showed a striking degree of CNV, with instances of both inherited and de novo CNV. Objectives: AGP Phase II involves GW analysis for association and copy number variation using the Illumina 1M SNP array in a dataset to ultimately reach ~3,000 families. We have completed analysis for an sample of ~1,500 parent-child trios for association and CNV, and combined analyses include data from >700 AGRE families genotyped for the 550k subset of markers (~2,200 families total). Thus, the AGP represents by far the largest genetic study undertaken for autism. Methods: Raw 1M SNP data was distributed for CNV analysis, while genotypes were assessed for quality control (QC), ancestry and Hardy-Weinberg equilibrium (HWE) prior to analysis for association. Results: In copy number analysis, multiple algorithms were utilized to infer CNV from intensity and genotype data, and ~59,889 CNVs were called by at least two algorithms. Mean and median CNV sizes were 95kb and 42kb, respectively, and 20 CNVs were detected per proband on average. Numerous de novo and inherited variants were identified in novel loci highlighting previously implicated pathways (e.g. neuronal cell adhesion molecules) and conditions with overlapping genetic etiologies (e.g. mental retardation, schizophrenia). Family-based association analysis of ~1,500 AGP families using 1M data alone, or combined with AGRE families, does not reach GW-significance, and this highlights the underlying genetic heterogeneity of autism and what are certain to be small effect sizes. Of SNPs with 10-4 < P < 10-7, numerous identify genes related to neuronal development and guidance, similar to those identified by CNV. Integration of CNV and association results is underway to prioritize follow-up studies. Replication of association findings in the next stage of the AGP will be important for their ultimate interpretation. Conclusions: We conclude from these data that collectively rare variation, particularly from CNV, contributes substantially to autism risk or causation. If, as seen in studies of type 2 diabetes, common allele effect sizes correspond to odds ratios of (e.g.) 1.2-1.3, much larger samples will be required in order to provide sufficient power for risk allele identification.