Heterogeneity in Autism Spectrum Disorders (ASD): Linkage Analysis of Four ASD Subsets Defined by ADI-R Clustering

Background: Numerous susceptibility genes and chromosomal abnormalities have been associated with autism spectrum disorders (ASD), but most discoveries either fail to be replicated or account for a small effect. Inconclusive results could be in part a reflection of the heterogeneous phenotype of autism, suggesting the need to employ strategies to identify more homogeneous groups of subjects. Recently a new ASD phenotypic sub-classification was reported by employing clustering analyses on ADI-R data, the gold standard autism diagnostic assessment tool (Hu et al., 2009). According to this classification, four distinct phenotypic subgroups were identified: (G1) severe language impairment, (G2) milder symptoms across all domains, (G3) notable savant skills, and (G4) intermediate phenotype. Objectives: The objective of our study was to address clinical heterogeneity in ASD at the genomic level and to determine if the subject stratification method based on ADI-R clustering will reduce phenotypic heterogeneity and improve linkage analysis. Methods: Data for approximately 10,000 Single Nucleotide Polymorphism (SNP) markers derived from the Affymetrix 10K SNP array for 426 families was downloaded from the AGRE website. The SNP data were sorted into separate files for each of the four sub-categories. Four group-related main lists were prepared by selecting families having at least one autistic member belonging to one of the four groups (G1, G2, G3, or G4). In multiplex families, affected siblings that were not in the same phenotypic subgroup were then removed to reduce intra-family heterogeneity (G1s, G2s, G3s, and G4s). Further stratification was done based on the affected individual's gender. This intense subject stratification method resulted in 16 distinct lists for linkage analysis. Non parametric linkage was calculated using MERLIN package. The linkage disequilibrium command in MERLIN was used to remove unlikely genotypes and to rule out the possibility of false positive results. Results: When the combined (non-subtyped) samples were analyzed, the highest LOD score obtained for 426 families was 2.8, P=0.0002 at 10q23. It is expected that lowering sample size results in reducing the detection rate. Interestingly, when applying our subject stratification method, despite the sample size reduction, in several instances the LOD score either improved compared to no grouping or new group-specific suggestive linked regions were detected. An example of such group-specific results is CNTNAP2. Suggestive linkage to CNTNAP2, a candidate gene for autism and language impairment, was detected for group 1 autistic females, a severely language impaired ADI-R subtype. A positive suggestive linkage was also identified with the SEMA5A gene for group 4 only (LOD=1.9, p=0.002, 138 families). This group-specific linkage result is intriguing because a recent genome-wide association study identified SEMA5A as a new autism susceptibility gene. Furthermore, expression of SEMA5A is reduced in autistic subjects compared with controls using both blood and brain samples. Conclusions: Our results indicate that applying such a phenotypic sub-classification method will improve the detection power in genome-wide linkage studies by reducing heterogeneity in ASD study subjects. It also provides further evidence for both inter- and intra-family heterogeneity. Our study demonstrates a novel and effective method to address the heterogeneity in ASD.