In this study, we tested multiple common variants in GABRB3 for association with AS and related endophenotypes in a general population sample. To our knowledge, this is the first study to incorporate such a comprehensive group of phenotypes (including clinical diagnosis, self-report questionnaires, and observational measures on both social and non-social aspects of autistic cognition). Nine SNPs were nominally associated with AS, of which three (rs7180158, rs7165604, rs12593579) retained their significance after Bonferroni correction. All three SNPs are located in a 19 kb genomic region that forms an LD block in our sample. The entire block was found to be associated with AS when tested for global haplotype association. For the EQ, thirteen SNPs were nominally significant, of which two (rs9806546, rs11636966) remained significant after Bonferroni correction. Two SNP-SNP interactions were significant for the EFT (rs12438141-rs1035751 and rs12438141-rs7179514) and one for the MRT (rs7174437-rs1863455) after Bonferroni correction. None of the SNPs was located in regions of common structural variations, CpG islands, or affected micro RNA binding sites, as calculated by SNPnexus and the UCSC Genome Browser.
The three SNPs that were significant in AS case–control studies are located in intron 4, within a region of 16,218 base pairs. In our sample, they were found to be a part of a shared 19 kb LD block, which is also associated with AS. This genomic region extends from rs7180158 (Chr15: 26978238, dbSNP) to rs7174437 (Chr15: 26997923, dbSNP) in intron 4, and is 19,905 bp downstream of exon 3 and 16,326 bp upstream of exon 4. When queried in FastSNP, rs12593579 was suggested to disrupt an enhancer region. Both rs7165604 and rs12593579 are found in putative transcription regulatory sites according to F-SNP. Further, according to Haploreg, these SNPs disrupt transcription factor binding sites either directly or through common variants in high LD. The two SNPs significantly associated with the EQ are located in intron 5, and are a mere 358 bp apart. However, they are not in significant LD either in the CEU population (R2: 0.182; see Additional file 1: Table S1) or in our sample (R2 = 0.157) (see Additional file 1: Figure S2). It is hence likely that the effect of the SNPs have been independently measured, although this does not exclude the possibility that a single non-tested variant situated close by is contributing to the effect seen at the two SNPs. Genotyping all the variants in this short region and testing them for association with the EQ should be undertaken in a future study. The most significantly associated SNP rs11636966 markedly alters an En-1_2 binding site (Haploreg). There is considerable evidence to suggest a role for proteins of the Engrailed family in early neurodevelopment and ASC [49–52]. Both these SNPs are also in high LD with several SNPs that putatively alter other transcription factor binding sites. A previous study from our laboratory  tested for association between two SNPs in GABRB3 (rs2873027 and rs11161335) with both AS and the EQ. Neither of the two SNPs was significantly associated with AS, but rs2873027 was found to be associated with the EQ. In the current study, only rs2873027 was included in the 45 SNPs genotyped. rs2873027 was not significantly associated with the EQ in the current study.
Two SNP-SNP interactions were found to contribute to variation in the EFT scores. Three SNPs participate in these interactions and none of them is in significant LD with each other either in the CEU population sample or in our sample (see Additional file 1: Table S1; see Additional file 1: Figure S3). Two of these SNPs, rs7179514 and rs1035751, are located in intron 5, while rs12438141 is located in intron 6. According to both FastSNP and F-SNP, rs1035751 and rs12438141 alter transcription regulatory sites. All SNPs except rs1035751 disrupt regulatory motifs (Haploreg). A single SNP-SNP interaction, rs1863455-rs7174437, was significantly associated with the MRT. rs1863455 is located in intron 5 and rs7174437 in intron 4, and they are 79,754 bp apart. They are not in significant LD (R2 = 0.003) (see Additional file 1: Table S1) in the CEU population, and are not a part of an LD block in our sample (see Additional file 1: Figure S3). Both the SNPs disrupt regulatory motifs (FastSNP, F-SNP, and Haploreg). rs7174437 disrupts regulatory motifs of the Sox family of transcription factors (Haploreg), which play a pivotal role during development .
The putative functional annotation of the SNPs indicates that a number of the significantly associated SNPs alter GABRB3 transcription during and after embryonic development either directly or via other variants that are in high LD. Of the ten SNPs found to be significant, three are in intron 4, six in intron 5, and one in intron 6. GABRB3 has twelve exons (see Figure 1) and four transcript variants in NCBI. Transcript variants 1 and 2 are more common at a cellular level, but neither of them have exons 5 and 6. Transcript variant 3, which has been described only at the transcript level, starts from exon 5, whereas transcript variant 4, which has been described at the protein level, starts from exon 6. There are currently no studies describing the function of these transcripts. It is possible that the variants in introns 4 and 5 act to regulate levels of these alternate transcripts. Alternatively, the variants could also act as intronic regulatory elements to alter the levels of transcript variants 1 and 2. Cellular studies need to be performed to validate either of these hypotheses.
We did not replicate the exact SNP association reported by Chakrabarti et al. . However, the inference in our earlier paper was at the gene level (that is, the statistical inference from the permutation analysis was that one or more SNPs in the gene were significantly associated with the phenotype, and not specifically for any SNP). The current study replicates the association of GABRB3 with empathy at the gene level, and provides evidence for the association of GABRB3 with AS, the EFT, and the MRT.
A limitation of this study is the moderate sample size. Replication of the results in a larger sample size is required to validate the findings. Although no SNP was significant across two or more traits, the experiments were not completely independent as several individuals participated in two or more experiments. Additionally, the low variance in the phenotypic scores of EFT must be taken into account when interpreting the results. To assess if the limited variance was restricted to the genotyped sample, we checked for EFT scores in a larger database. Both the mean EFT score and standard deviation were similar between the two datasets, indicating that the low variance in the EFT scores in our sample was not an artifact. Although none of the cases studied were formally tested for IQ, an essential feature for the clinical diagnosis of AS is that individuals diagnosed as having AS or high-functioning autism (as was the case for the ASC sample in our study) is that individuals have an IQ in the average range . Average range is defined as two standard deviations either side of the mean, so this would mean an IQ of over 70. Further, it has been shown that IQ and autistic traits show limited genetic covariance . It is therefore reasonable to infer that IQ differences are unlikely to confound the conclusions drawn from the current study. Finally, it is important to remember that GABRB3 is not the sole gene that contributes to either AS or the traits tested, and that other genes could act either independently of or epistatically with GABRB3 to contribute to these complex phenotypes.