The patterns of analysis for multivariate and univariate phenotypes are quite consistent. Specifically, for multivariate analyses, the lowest P-values obtained were those for PRLR (P = .000005, with the multivariate indicator of Communication Skills and rs7727306 and P=.000249, with the multivariate ADI phenotype, ADI Group and rs35614689) and PRL (P = .000085 and P = .006957, with the multivariate ADI phenotype and rs1341239 and rs1205961, respectively). The pattern of univariate results is supportive, highlighting, for associations with the PRLR gene, the phenotype ADOS-Stereotyped Behaviors (P = .039409 for rs35614689) and, for the associations with the PRL gene, the Vineland–Social (P = .044005 for rs1341239) and ADI-Based Diagnosis (P = .025527 for rs1205961) phenotypes.
Similarly, there is convergent evidence for the association with the OXTR and OXT genes, both for a variety of multivariate (P = .007972 for Clinical Diagnosis, P = .001843 for ADI Group, P = .018209 for Communication Skills and P = .006633 for Stereotyped Behaviors with rs2268493 of the OXTR gene, and P = .015761 for Stereotyped Behaviors with rs2740204 of the OXT genes, respectively) and univariate (for rs2268493 of OXTR, P = .000978, .011046, .012727, and .035048, for ADI-Based Diagnosis, ADI-Restricted/Repetitive Behaviors, ADOS-Stereotyped Behaviors, and Clinical Diagnosis, respectively; for rs2740204 of the OXT gene, P = .012738 for ADOS-Stereotyped Behaviors) phenotypes.
Finally, there is convergent although not strong evidence from both multivariate and univariate phenotypes regarding associations with autism indicators and the remaining two genes investigated in this study, DβH and FOSB. The presence of association with the two SNPs of the DβH gene is captured by a repetitive set of multivariate P-values (see ) and supported by univariate ones (for rs2519148 with the Vineland’s Communication, P = .035981, Daily Living Skills, P = .004197, Socialization, P = .006667, Motor Skills, P = .032184, and Composite Score, P = .007323; for rs2073837 with the ADOS’s Stereotyped Behaviors, P = .014762 and Imaginative Skills, P = .018242 and with the ADI-Based Diagnosis, P = .033303). Similarly, both sets of phenotypes consistently implicate FOSB as associated with ASD. Three of the SNPs in this gene show consistent associations with multiple phenotypes (rs1049739 and rs7256242, both multivariately, see , and univariately, with ADOS Stereotyped Behaviors, P = .038450 and ADI Restricted/Repetitive Behaviors, P = .008365 and P = .007427, respectively; and rs2276469, only univariately, with Clinical Diagnosis, P = .014081, and with ADI’s Communication, Restricted/Repetitive Behaviors, and ADI-Based Diagnosis, P =.029669, .012449, and .010176, respectively).
Haplotype analyses of the studied genes also provided support for the presence of association. Specifically, the PRLR T-T-G haplotype (rs37370, rs249522, rs35614689) appeared to be overtransmitted (P-values of .0319, .0232, and .0406, for ADOS-Based Diagnosis and ADI’s Restricted/Repetitive Behaviors and Communication, respectively). The ADOS’s Stereotyped Behaviors was associated with the G-G haplotype of the OXT gene (P = .0166), but no PRL or OXTR haplotypes were overtransmitted in this sample, as indicated by a lack of association with any of the analyzed phenotypes. The A-G haplotype of the DβH gene was associated with the ADI’s Onset (P-value = .0476) and ADI-Based Diagnosis (P-value = .0257) phenotypes. Finally, the SNPs in the FOSB gene revealed a consistent association with the phenotypes of Clinical Diagnosis and all ADI-based phenotypes. Specifically, all pairwise combinations and combinations of three and four SNPs were statistically significant, with P-values ranging from .0158 to .0051.
Given that the assessment techniques resulted in the creation of 23 phenotypes (7 multivariate and 16 univariate) described earlier and that we used 17 SNPs, we introduced a methodology to correct for multiple testing. To mitigate somewhat the conservative nature of the traditional Bonferroni correction, we used a modified Bonferroni correction. We implemented a two-stage strategy that uses PBAT to estimate powers of phenotype–allele combinations and select the most powerful such combinations, and then uses FBAT to test the selected combinations. The key to the validity of the strategy is that disjointed portions of the data are used in the two stages. Consider the combination of a given subset of phenotypes and a given marker allele. PBAT produces power estimates for that combination by making use of the parental genotypes and children’s phenotypes, and ignoring child genotypes that would drive any subsequent FBAT analysis. PBAT estimates genetic effect sizes in a model of the subset of phenotypes, and those effect sizes are then used to estimate power for a hypothetical FBAT analysis of that phenotype–allele combination. The second stage, which uses children’s genotypes in an FBAT analysis, is performed only if the power estimated in the first stage is sufficiently high. In our case, we selected the phenotype–allele combinations with estimated power higher than the median of the estimated powers of all such combinations, reducing the number of combinations to be tested by a factor of two. The reduced numbers of combinations were used in the two Bonferroni corrections (one for multivariate and one for univariate phenotypes); the obtained P-value thresholds were 0.00042 for multivariate and 0.00018 for univariate P-values, for overall levels of .05 taking multiple testing into account.
The pattern of results changes after the introduction of such a correction, so that only three multivariate P-values and no univariate ones survive. Specifically, the strongest P-values implicated PRLR and PRL genes as associated with the multivariate phenotypes ADI-Group (P = .000249 for PRLR’s rs35614689 and P = .000085 for PRL’s rs1341239) and Communication Skills (P = .000005 for PRLR’s rs7727306). Yet, as indicated earlier, the univariate and haplotype analyses supported the presence of association of ASD with these genes overall rather consistently. Because our analyses extend previously reported findings or hypotheses in the literature, we present here the general pattern of results, but this pattern of results should be interpreted with caution.