Identifying Biological Pathways Implicated in Defined Subgroups of Phenotypic Expression for Autism Spectrum Disorders

Background: Previous studies established a strong influence of genetic variation in the etiology of Autism Spectrum Disorders (ASD). However, estimated effect sizes for associated regions are small and combined evidence from many analyses does not explain the heritability. The difficulty in identifying genetic variation with strong effects may arise from the wide variability in clinical manifestation being explained by underlying genetic heterogeneity. More effective methods of ASD phenotype definition and applying pathway-based analysis to GWAS data are ways to address these obstacles. Pathway-based analysis organizes SNPs into biologically meaningful groups and evaluates multiple contributing factors together. This approach allows insights into biological mechanisms undetectable in the conventional approach to GWAS.

Objectives: Our hypothesis is that minimizing phenotypic heterogeneity will help account for genetic heterogeneity and will increase power to detect associations with genetic variation.

Methods: We used the AGRE family dataset for our initial modeling.  We used agglomerative hierarchical clustering to group individuals with ASD relative to behavioral and clinical exam information. We calculated the odds of cases being assigned to the same cluster given a familial relationship. We also estimated genetic relationships and compared relatedness within versus across clusters. P-values for association were calculated using the Family-Based Association Test. We used Pathway Analysis by Randomization Incorporating Structure (PARIS) to assign SNPs to genes based on chromosomal location and identify biological pathways of interest based on overrepresentation of nominally significant SNPs. We validated our data using a dataset derived from the AGP.

Results: The most valid clustering grouped the data into two clusters based on severity. We see increased odds for relatives being assigned to the same phenotype cluster (OR≈1.5, p<0.00001). We also see that cases in a given cluster are more genetically related when compared to the un-clustered dataset (Fst≈0.15±0.26). We identified 53,703 SNPs significantly associated (p<0.05) with ASD that are being evaluated in the pathway-based approach. Of these, only 1,607 are associated with ASD independent of phenotypic clustering. The levels of significance for all associated SNPs vary based on cluster assignment. We will test for replication of findings from the genetic studies by doing similar analyses using the two comparable clusters identified in the independent ASD dataset.

Conclusions: The results indicating that related individuals are more likely to be assigned to the same cluster suggest that the phenotypic clusters recapitulate genetic etiology. We anticipate by utilizing these effective methods for phenotype definition prior to pathway-based genetic analyses our power to detect factors influencing risk for ASD will be increased.