We have confirmed the involvement of GABRA4 in autism through identification of significantly associated SNPs within an independent African-American population. Furthermore, we have strengthened our original findings, including identification of additional associated SNPs and a significant interaction between GABRA4 and GABRB1 and in an extended dataset (N=557) of Caucasian autism families. The AA dataset contains only 54 families, and in general power can be a problem in small sample sets if an effect is not seen. However, for the GABRA4 gene, we did find significant results in the AA dataset. Furthermore, studies in small datasets such as the AA dataset can also be less robust than in larger datasets. However, the fact that we find the same gene, GABRA4, significant in two different ethnic groups is supportive of the role of GABRA4 in autism risk.
The identification of two different two-way interactions between GABRA4 and GABRB1 provides additional evidence of the complex interaction of these two genes in autism. The rs1912960 (GABRA4) with rs2351299 (GABRB1) interaction is between the same two SNPs previously reported in the study of Ma et al., and is still significant in our larger dataset. A second significant two-way interaction was found, also including rs2351299 in GABRB1 as well as rs17599416 in GABRA4, further supporting that interactions between these two genes are involved in autism risk. Although rs2351299 does not have a significant PDT result, it is found in both significant interaction pairs, and in each of the significant three-way interactions. The MDR approach used in these analyses is specifically designed to detect interaction effects both in the presence and absence of main effects.
It is possible that both of the two-way interactions are being identified due to LD between the two GABRA4 SNPs (rs1912960 and rs17599416). There is significant correlation between the two SNPs, although the r2 value (Table ) between them is not large (0.32). Given that these do not appear to be causative variants, it is likely that the true variant, yet to be identified, is in LD with these GABRA4 SNPs. Examination of interaction in the independent dataset of African-American families was not possible due to the limited sample size.
We also identified variants within GABRB1 as associated within the autistic population with seizures. One of these variants, rs2351299, also shows a significant interaction with SNPs in GABRA4 (Table ). While no effect was seen in GABRA4 for the seizure subset, the sample size may be too small to conclusively determine its role in seizure status in autism. However, the enhanced findings in GABRB1 in the seizure subset implicate GABRB1 as a contributor to genetic risk in these patients.
Despite the identification of GABRA4 in both ethnic groups, different SNPs were found to be associated. The identification of distinct SNPs within these populations may indicate differences in allele frequency and linkage disequilibrium within the two racial groups, differences in the haplotypic background in which identical causative variations originated, or differences in the causative variation. SNP rs16859788 for example, which is significant in the African-American group, shows little variation in the Caucasian dataset, therefore, providing no power for detection of an effect in this group. Other SNPs, however, show similar allele frequencies between the two populations.
Some differences in LD do exist between the two ethnic groups as well; however, the majority of the differences are small. The largest differences in LD are in pairwise values with rs16859788, which appear to mostly be due to the fact that the SNP is practically monoallelic in the Caucasian population. The Caucasian dataset suggests that there is a significant association of SNP haplotypes with autism risk, while the African-American set does not. However, this difference may be due to the small size of the African-American dataset. Therefore, while it appears that minor allele frequency differences can account for the lack of association of rs16859788 in the Caucasian dataset, additional studies are needed to determine whether the other differences in results between the two ethnic groups are due to sample size differences, differential LD with the causative variation in the two populations, or are population-specific risk alleles.
While we have identified several associated SNPs, we do not predict any of the ones in GABRA4 to have functional consequences; therefore, it is unlikely that these are primary variants leading to the autism susceptibility. One of the SNPs identified in GABRB1 in the seizure subset, however, is in the 3′ untranslated region (UTR). Given that multiple GABA receptor subunits combine in varying combinations to form a functional GABA receptor, even minor changes in levels of a particular subunit may alter the makeup of receptors within a particular cell type, and alter the GABAergic signaling. Therefore, variations within potential regulatory regions, such as untranslated regions and promoters, could play an important role. It will be important to look at potential changes that may result from this and other potential GABRB1 UTR variations, as well as sequence coding and potential regulatory regions to identify the primary variation, or variations leading to altered autism susceptibility.
In summary, these data show that the GABA receptors are implicated in the etiology of autism in two different ethnic populations and suggest seizures as a stratifying phenotype. Furthermore, these results support our earlier findings, indicating GABRA4 and GABRB1 as genes contributing to autism susceptibility, independently, in the case of GABRA4, and through complex interactions with each other.