Genome-Wide Gene-Environment Analysis Identifies Genetic Variation within A2BP1 As a Potential Modifier of the Risk Effect of Maternal Smoking on the Expression of Autistic Traits in Middle Childhood

Background: Maternal smoking during pregnancy has been discussed as a potential risk factor for the increase of autistic symptoms in children, though findings have been inconsistent. It is possible however that maternal smoking might only elevate the expression of autistic symptoms in genetically susceptible individuals, and thus GxE may conceal the true underlying relationship between risk factor and behavioural outcome.

Objectives: We aimed to identify genetic moderators of the risk effect of maternal smoking on the expression of autism-related social communication difficulties during childhood and adolescence using a genome-wide GxE analysis.

Methods: We performed a 2-stage genome-wide GxE analysis with respect to social communication problems as captured by the Social and Communication Disorders Checklist (SCDC). Measurements were assessed in a large UK population-based birth cohort at 8, 11, 14 and 17 years of age (Avon Longitudinal Study of Parents and Children, N ≤ 5539). During the first stage, we performed a genome-wide screen for differences in phenotypic variance per genotype (Levene’s test) using 2.5 million imputed or genotyped SNPs and investigating each time-point individually. This identified a set of variants, which is likely to be involved in interactions. During the second-stage, we investigated all independent signals with respect to maternal smoking during the first trimester (adjusted for child’s age and sex), performing a joint likelihood ratio test for both G+GxE and E+GxE respectively using negative binomial regression.

Results: Maternal smoking during the first trimester was identified as a strong predictor of increased autistic traits at 8 years of age (β=0.23(0.05), measured in log-counts of social-communication problems, P=4.4x10^-7), but the risk effect declined during the course of development. During stage one of our GxE screen, we identified 132 independent offspring SNPs (Levene-P<10^-5, LD based clumping: r²=0.2, ±500kb, in addition to all SNPs in known ASD candidate regions with P<10^-4) across all 4 time-points, which are likely to be involved in interaction effects. The strongest E+GxE signal (Levene-P=5.2x10^-5_, P_GxE=0.00016 (P_{GxE_adj}=0.021), P_E+GxE=1.7x10^-9) was observed for variation in the ASD-candidate gene A2BP1 (alias RBFOX1) and was associated with social-communication problems at 8 years of age. There was no evidence for a genetic main effect (P_G=0.96). Stratification by genotype, showed that the risk effect of maternal smoking was only present in homozygote carriers of the major (risk) allele (0 risk allele: β(smoking) = -0.23(0.20), P =0.27, 1 risk allele: β(smoking) = 0.057(0.084), P=0.50, 2 risk alleles: β(smoking)=0.35(0.057), P=5.8x10^-10). The interaction was attenuated with progressive age due to the decline in risk effect of smoking. We found no joint G+GxE effect (P<10^-5), which provided more evidence for association than G alone.  Replication of the interaction signal is currently being sought within independent cohorts.

Conclusions: Our variance-based genome-wide screen identified a small set of SNPs that are probably involved in interactions either with environmental or other SNP variation. We identified one GxE interaction effect, which, if replicated, might indicate that offspring of some smoking mothers are at particularly high risk for social-communication problems in childhood and adolescence.