Oxytocin Receptor Polymorphisms are Differentially Associated with Social Abilities across Neurodevelopmental Disorders

Oxytocin is a pituitary neuropeptide that affects social behaviour. Single nucleotide polymorphisms (SNPs) in the oxytocin receptor gene (OXTR) have been shown to explain some variability in social abilities in control populations. Whether these variants similarly contribute to the severity of social deficits experienced by children with neurodevelopmental disorders is unclear. Social abilities were assessed in a group of children with autism spectrum disorder (ASD, n = 341) or attention deficit hyperactivity disorder (ADHD, n = 276) using two established social measures. Scores were compared by OXTR genotype (rs53576, rs237887, rs13316193, rs2254298). Unexpectedly, the two most frequently studied OXTR SNPs in the general population (rs53576 and rs2254298) were associated with an increased severity of social deficits in ASD (p < 0.0001 and p = 0.0005), yet fewer social deficits in ADHD (p = 0.007 and p < 0.0001). We conclude that these genetic modifier alleles are not inherently risk-conferring with respect to their impact on social abilities; molecular investigations are greatly needed.

genome wide significance level 20 . Meta-analyses indicate that certain neuropeptide variants are more common in individuals with ASD than in controls 21 , and may influence social abilities in the general population 22 .
ASD and attention deficit hyperactivity disorder (ADHD) are common childhood onset neurodevelopmental disorders that are highly heterogeneous, present along a spectrum of severity, and often comorbid 23,24 . Both disorders are associated with social deficits, inattention, and hyperactivity to varying degrees [23][24][25] , leading some to propose that they may be different manifestations of an overarching disorder 23 . In addition to rare genetic mutations and environmental factors, the additive effects of common genetic variants are hypothesized to contribute to the development of these disorders 26 , and may explain some of the observed heterogeneity in behaviour, abilities, and psychopathology. Preliminary investigations suggest that single nucleotide polymorphisms (SNPs) in OXTR may act as modifier alleles with respect to the severity of social deficits experienced by individuals with ASD (For a review of the literature, see Supplementary Table S1). Two studies have also examined OXTR SNPs in small groups of children with ADHD 27,28 . It remains unclear whether OXTR SNPs act as: 1) general risk factors for social impairment across populations (i.e. effects span diagnostic boundaries), or 2) specific modifier alleles in certain behavioural or biological context (i.e. disorder specific effects). Two recent studies have shown that OXTR variants were associated with a similar magnitude and direction of social differences in children with ASD and their unaffected relatives 6,29 . This question, however, has not yet been examined across different diagnostic groups, where the environmental, behavioural, and genetic architecture differ. Given the increasing number of studies employing trans-diagnostic approaches in psychiatry 30 , it is important to examine whether behavioural risk variants function similarly across disorders.
In this study, we examined the association between four SNPs in OXTR and social abilities in a large sample of children and youth with ASD or ADHD and present the first cross disorder comparison of genotype effects. To extend on preliminary research to date, we used two validated continuous social metrics, and restricted initial hypotheses to Caucasian participants only, given evidence of possible ancestral differences in the effects of OXTR on social abilities 31,32 . First, in our group of children with ASD, we confirmed several associations between social deficits and OXTR variants (OXTR rs53576 G-allele, rs237887 A-allele, and rs2254298 A-allele) that had been previously shown in smaller samples. New associations were detected in the group of children with ADHD. Second, we compared the magnitude and direction of the impact of common variation in OXTR on social abilities between diagnostic groups (ASD vs. ADHD). Based on the family studies described above, and the high degree of symptom overlap and comorbidity between ASD and ADHD, we hypothesized that genotype effects would not differ by diagnosis. However, unexpectedly, highly significant diagnostic differences were detected, with effectively a reciprocal direction of association between genotype and social abilities between the two diagnostic groups. Our findings indicate that common polymorphisms in neuropeptide receptor genes likely represent a specific and not a general risk modifier, with effects that vary by diagnosis in the context of different biological, behavioural, or environmental conditions. These results have implications for all analyses involving common alleles and complex traits.

Results
Study sample. The sample consisted of 617 participants with either ASD or ADHD (n = 404 Caucasian). The ASD group had lower mean intelligence quotients (IQs), and higher anxiety levels (see Table 1). Genotype frequencies and linkage analyses are presented in Supplementary Tables S2 and S3, compared to/based on reference genome data from the 1000 Genomes Project 33 . OXTR allele frequencies did not deviate from Hardy Weinberg equilibrium (Supplementary Table S4), except for rs237887 in ADHD.
Measures. Severity of social deficits were quantified by examining the number of incorrect items selected by the participants on the Reading the Mind in the Eyes Test (RMET) 34 , and through their parent/caregiver's ratings of concerns on the Social Communication Questionnaire (SCQ) (see methods) 35 . Therefore, throughout the manuscript, higher scores on either metric suggest greater social deficits. Correlations between measures in ASD and ADHD are shown in Supplementary Tables S5 and S6. Twelve percent of the participants with ADHD had SCQ scores that fell above the cut-off for concern for ASD (Total SCQ > 15). For comparison, 23% of the participants with ASD fell above the cut-off for clinical concern regarding ADHD on the Child Behavior Checklist (T-score > 70). Aim 1. Confirmation of previous OXTR genotype effects on social abilities. Using logistic regression, we initially examined OXTR SNPs with prior evidence of genotype effects on social abilities within each diagnostic group (in ASD: rs53576, rs2254298, and rs237887; in ADHD: rs53576 and rs13316193), restricting the sample to those of Caucasian ancestry only, and adjusting for other possible confounding factors (see Methods). When considering multiple testing, five comparisons (3 in ASD and 2 in ADHD) across two social metrics yielded a corrected p-value of 0.005 as the threshold for significance for an overall alpha of 0.05.
The results from our primary hypotheses are displayed in Fig. 1 (white panels); values are indicated in Table 2 for genotype risk group comparisons, and mean scores across all genotypes are shown in Supplementary  Table S7. Covariates from these categorical logistic regression models in the Caucasian cohort are displayed in Supplementary Tables S8 and S9. In summary, several previous associations between OXTR SNPs and social abilities were confirmed in the group of children with ASD, but not ADHD. Aim 2. Examination across diagnostic groups. Next, we combined data across both diagnostic groups, and tested for genotype*diagnosis interactions in the Caucasian participants. After correcting for multiple comparisons (n = 8), a p-value of 0.006 was used for significance regarding the interaction term. In this context, we also tested for genotypic differences in social abilities across all SNPs in the remaining diagnostic groups (i.e. those not previously tested as part of our primary aims, including OXTR rs2254298 and rs237887 in ADHD, as well rs13316193 in ASD) (Fig. 1, grey panels, Table 2).
For rs53576, the diagnostic interaction terms were significant on both the SCQ and the RMET (SCQ: Wald χ 2 = 25.0, p < 0.0001; RMET: Wald χ 2 = 11.4, p = 0.0007), indicating that the 'risk variant' in ASD showed a significantly different effect in ADHD (as shown above, the OR of social deficits with the hypothesized risk variant was >1 on both metrics in ASD, but <1 on both metrics in ADHD, see Table 2). For OXTR rs2254298, the genotype*diagnosis interaction term was significant for the SCQ only (Wald χ 2 = 31.8, p < 0.0001). In ADHD,  Attention deficit hyperactivity disorder. Analyses using the RMET were adjusted for IQ, therefore only participants with an RMET score and IQ data were included for these analyses. *P-values from ANOVA's comparing demographic variable between ASD and ADHD, adjusted for ancestry.
the AA/GA-allele was associated with fewer social deficits on the SCQ [Z = −4.7; p < 0.0001, OR 0.7 (0.6-0.8)], opposite to the direction initially detected in the ASD group (OR = 1.2 in ASD). The diagnostic interaction term for OXTR rs2254298 was not significant on the RMET (Wald χ 2 = 0.3, p = 0.6), where the risk allele had no significant effect in ADHD or ASD (Table 2). Likewise, there were no significant differences in social abilities on either metric for rs13316193 in ASD nor in ADHD (Table 2), resulting in non-significant SNP*diagnosis interactions. Although no significant allele effects were found for rs237887 after correcting for multiple comparisons (Table 2), the trend towards effects in opposite directions across the two diagnoses resulted in similarly borderline interaction effects across both outcomes (SCQ: Wald χ 2 = 5.9, p = 0.02; RMET: Wald χ 2 = 3.0, p = 0.08). In summary, across multiple OXTR variants, there was evidence that the hypothesized risk allele was associated with more significant social deficits in ASD, but fewer social deficits in ADHD.

Secondary aims.
As secondary aims, we also examined (1) whether social abilities varied by arginine vasopressin receptor 1a (AVPR1a) RS3 microsatellite length, and (2) the impact of ancestry on OXTR findings. Greater length of the AVPR1a RS3 microsatellite was previously associated with greater receptor expression levels and superior social abilities in typically developing populations [36][37][38] , although no effect was detected in a large sample of children with ASD 29 . Previous studies have also found evidence for a differential direction of effects of OXTR polymorphisms in Caucasian vs. non-Caucasian individuals, particularly for rs2254298 31,32 .
(1) AVPR1a RS3. AVPR1a is the gene coding for the corresponding vasopressin receptor. Upstream of the exon is a microsatellite promoter region, which includes an RS3 segment of CT, TT, and GT repeats. The RS3 segment varies in length; in our sample the range was 310-346 base pairs (bp), with a median of 328. We categorized each individual's promoter length as 'long' or 'short, with the cut-point placed at 328 bp, as this was both the median in our sample, and similar to cut-points used in previous studies (e.g. 326 and 328 36 , 326 39 , or median value 40 ,). We considered those carrying two 'short' alleles to be the high-risk group, and compared them to those with two 'long' alleles as the low-risk group. In the group of Caucasian participants with ASD, there were no differences in social abilities based on microsatellite length on the SCQ [short/short vs. long/long, Z = −0.9, p = 0.3, OR = 0.9 (95% CI: 0.8-1. (2) Examination of OXTR across ancestry groups. We subsequently combined data across Caucasian and non-Caucasian participants, and tested for ancestry*genotype interactions. The aim was to determine whether the direction of genotype effects was sufficiently similar between the Caucasian and non-Caucasian groups to justify performing the genotype group comparisons on the entire combined sample, irrespective of ancestry. For the ASD cohort, ancestry*genotype interaction terms were non-significant for all OXTR SNPS, on the SCQ and on the RMET (p > 0.05 for all). In the ADHD group, there were no significant ancestry*genotype Higher scores across both metrics suggest greater social deficits. All analyses were adjusted for age and sex differences, and those involving the RMET were adjusted for IQ as well. The high-risk/ low-risk groupings were based on the patterns observed in previous studies. P-values are uncorrected. Error bars represent 95% confidence intervals (CI). White panels indicate primary hypotheses from the literature (aim 1). Exact sample sizes detailed in Tables 1 and 2.
SCIENTIfIC REPORtS | 7: 11618 | DOI:10.1038/s41598-017-10821-0 interactions on the RMET (p > 0.05 for all), but significant effects were observed on the SCQ for rs2254298 (Wald χ 2 = 29.1, p < 0.0001), where the opposite direction of an association between the risk allele and social deficits was detected in the non-Caucasian cohort [OR = 1.4 (95% CI: 1.1-1.7), p = 0.002] compared to the Caucasian cohort [OR = 0.7 (95% CI: 0.6-0.8) p < 0.001]. Interaction terms were also significant in ADHD for rs53576 (Wald χ 2 = 7.6, p = 0.006) and rs13316193 (Wald χ 2 = 5.1, p = 0.02) on the SCQ. For SNPs where the ancestry*genotype interactions were conservatively non-significant (p > 0.05), we pooled data across ancestry groups, and calculated the combined odds ratios in the larger sample (Table 2, right side). Overall, the direction of effects was very similar in the combined and Caucasian only samples, although the larger combined sample size led to more significant associations. Exploratory analyses. The finding of differential effects of genotype on social deficits between diagnostic groups was unexpected. To further explore and confirm this pattern, we repeated the analyses using additive genotype risk models (GG vs. GA vs. AA), as opposed to categorical (GG/GA vs. AA). Results were similar with either model (see Supplementary Table S10).
Next, given an emerging hypothesis that oxytocin may affect social behaviour by decreasing anxiety or fear responses, potentially at the level of the amygdala 41,42 , and given the high rate of comorbid anxiety disorders in ASD (up to 40%) 43 , we post-hoc hypothesized that differential anxiety levels between the two clinical groups could potentially explain the reciprocal pattern. Anxiety was quantified through parent report on the Child Behavior Checklist anxiety subscale (CBCL-anxiety) and the Revised Children's Anxiety and Depression Scale (RCADS). We compared anxiety scores by diagnosis, then by genotype, and subsequently included anxiety as an additional covariate in the original SNP-SCQ models. Anxiety analyses were restricted to participants who also had SCQ scores available and were of Caucasian ancestry (CBCL n = 377, RCADS n = 235), to facilitate comparison to previous findings.
On the CBCL-anxiety, the ASD group had higher mean T-scores (Table 1) and higher total anxiety raw scores compared to the ADHD group [mean raw score For genotype group comparisons, we tested for effects of rs53576 and rs2254298 on anxiety measures in ASD and ADHD separately. No genotype effects were detected on the RCADS for either SNP (p > 0.05). For rs53576, a modest genotype effect was detected on the CBCL-anxiety raw score in the ASD group in line with the direction of effects seen with the SCQ [AA = 2.9 (95% CI: 1.8-4.0); GG/GA = 4.6 (4.2-5.1), t = −2.9, p = 0.004]. There were no significant effects of rs53576 on CBCL-anxiety in the ADHD group (difference: −0.2, p = 0.7). The SNP*diagnosis interaction term was non-signification (p = 0.09), suggesting a lack of differential effects of genotype on anxiety between the two diagnostic groups. For rs2254298, CBCL-anxiety scores were modestly affected by genotype in the ADHD group only (AA/GA vs. GG: −1.0 points, p = 0.03). The diagnostic interaction term was again non-significant (p = 0.3).
We subsequently included CBCL-anxiety score as an additional covariate in the SCQ models for rs53576 and rs2254298. Although CBCL-anxiety was significantly associated with SCQ in the overall model, results regarding the effects of OXTR on SCQ scores were otherwise unchanged [rs53576 in ASD: OR = 1.4 (1.1-1.7), rs53576 in ADHD: OR = 0.8 (0.6-0.9), rs53576*dx: p < 0.0001; rs2254298 in ASD: OR = 1.3 (1.1-1.5), rs2254298 in ADHD: OR = 0.7 (0.6-0.8), rs2254298*dx: p < 0.001]. Overall, data suggest that OXTR genotype may affect anxiety to some extent, but a reciprocal pattern between diagnostic groups as seen on social measures was not observed, and anxiety did not explain the differential effects of genotype on social outcomes by diagnosis.

Discussion
In this study, we examined how common polymorphisms in neuropeptide receptor genes affected social abilities in a large group of children and youth with ASD or ADHD. This is the first study comparing genotype effects across two neurodevelopmental disorders, with the largest ADHD cohort, and one of the largest ASD cohorts for this type of analysis to date. Other strengths of this study include: participants from two clinical groups who were recruited and characterized using the same research protocol; detailed phenotyping on both participant and caregiver completed metrics; and examination for potential ancestral differences in genotype effects. Our main findings were: 1) confirmation of several previous associations regarding neuropeptide receptor genotype and social abilities in our large group of children with ASD, 2) a lack of replication of previous associations between OXTR SNPs and social abilities seen in smaller ADHD cohorts, although several new associations were detected in this larger sample, and 3) significant diagnostic differences in the association between OXTR genotype and social abilities (i.e. a genetic modifier allele associated with more social deficits in ASD had either no effect, or conferred a benefit with respect to social abilities in ADHD).
Our findings are in line with direction of effects described in previous ASD cohorts and family studies. For example, Parker et al. examined genotypic differences in social abilities in a group of 79 children with ASD, 53 unaffected siblings, and 62 controls, of varied ethnicities 6 . They found that carriers of the A-allele of OXTR rs2254298 had higher levels of social impairment, and carriers of the G-allele of OXTR rs53576 performed worse on an affect recognition task 6 . Skuse et al. tested for associations between 60 OXTR SNPs and 31 AVPR1a SNPs and social abilities in an ASD cohort of 198 families from the United Kingdom and Finland, of mostly Caucasian ancestry 29 . Only OXTR rs237887 genotype affected face recognition memory, where AA carriers performed significantly worse. Length of AVPR1a RS3 microsatellite promoter region did not impact social abilities in their sample 29 , nor in ours. A recent meta-analysis found both the OXTR rs237887 A-allele and the rs2254298 A-allele were also risk-conferring with respect to an ASD diagnosis 21 . It would follow that in certain vulnerable families, specific OXTR common variants may push a small subset of those at risk over the diagnostic threshold.
Few studies have examined the impact of OXTR on social functioning in ADHD 27,28 . Park et al. found that rs53576 AA allele carriers had fewer social deficits in their sample of 119 participants. However, in our ADHD sample, OXTR rs53576 GG allele carriers and rs2254298 A allele carriers had fewer social deficits, opposite to the direction of effects seen in our ASD group. For OXTR rs53576, this finding may be more in keeping with patterns observed in control populations. In a recent meta-analysis of mostly typically developing adults, rs53576 G-allele homozygotes had better social abilities on a variety of metrics 22 . Similarly, a previous study using the RMET found that healthy adolescents who had the OXTR rs53576 GG genotype had higher accuracy overall 18 . Our data suggest that the direction of impact of common variants in these receptors vary by disorder-a protective variant in one condition may increase risk in another. Overall, this would indicate that common polymorphisms in OXTR represent a specific vulnerability factor, and not a universal risk factor with respect to social functioning.
Mechanistically, it may be that these non-coding intronic SNPs co-localize with specific receptor haplotypes, other important exons, or certain rare variants 44 , that affect the structure or function of OXTR. Alternatively, they may be markers of functional differences in the non-coding regulatory elements of neuropeptide receptor genes, indirectly affecting receptor expression levels. SNPs in non-coding regulatory elements were recently shown to exert a robust effect on oxytocin receptor expression levels within specific brain regions in prairie voles 45 . Similar investigations in humans are lacking and are greatly needed. At the same time, large and rare genetic variants may be more common in ASD, and could be potentially associated with more significant disruptions on oxytocin signaling 46,47 . For example, contactin-associated protein 2 (CNTNAP2) mutations have been associated with ASD and developmental delay 46 . A recent mouse model of ASD using CNTNAP2 knockout showed disrupted oxytocin expression and social deficits, which were corrected with exogenous oxytocin administration 47 . While usually we conceptualize rare variants as acting on a background of common variants, the implication of our study is genetic background may also play a critical role in modifying the impact of common 'risk' variants on behavioural phenotypes.
How exactly similar changes in receptor expression or function could result in disparate behavioural consequences in different diagnostic groups is unknown. Possible explanations include: 1) specific neurocircuits contributing to social deficits may vary between the disorders, 2) other diagnostic differences in environmental, hormonal, or epigenetic processes may be differentially accentuated through oxytocin signaling or 3) measurement factors or confounding.
With respect to neurocircuitry, emerging literature indicates that social and fear perception involve complex networks centred on the amygdala and reward centres 42,48 , and that oxytocin may potentially enhance transmission in these circuits through its effects on interneurons 41,49 . It may be that the brain circuits, regions, or underlying behavioural constructs contributing to differences in SCQ scores may vary between diagnostic groups. While our exploratory analyses involving anxiety measures suggested that OXTR genotype may also impact on anxiety levels to some extent, a reduction in anxiety alone was insufficient to explain the differential OXTR effects across diagnostic groups.
Alternatively, it is possible that other diagnostic differences in environmental, hormonal, or epigenetic processes may be differentially accentuated through oxytocin signaling. For example, several studies in typically developing cohorts have shown that the OXTR rs53576 G-allele was associated with increased environmental sensitivity, which led to more adverse outcomes under conditions of early childhood adversity or exclusion [13][14][15][16][17] . G-allele carriers were also shown to experience a more beneficial response to receiving positive social support as indicated by salivary cortisol levels 50 , and were more likely to seek social support under stress 51 . In this context, perhaps children with ASD are less likely to experience the stress buffering effects of social contact given their underlying social deficits, putting the more 'sensitive' allele group at greater risk. At the same time, there is evidence to suggest that oxytocin may play a role in suppressing the cortisol stress response 52 . Given evidence of divergent cortisol patterns in ASD vs. ADHD (i.e. enhanced cortisol responses to stress in ASD 53 , and attenuated cortisol profiles in ADHD 54, 55 ), it would follow that an OXTR variant leading to increased cortisol activity would be disadvantageous in ASD, but potentially advantageous in ADHD. Differential epigenetic regulation of OXTR gene expression between the two disorders could also potentially explain diagnostic differences in SNP effects (for a review see ref. 56). While there is evidence to suggest that increased methylation of certain OXTR sites may be associated with increased social deficits and autistic traits in ASD 57,58 , and callous/ unemotional traits in ADHD 59, 60 , to our knowledge, epigenetic profiles of OXTR have not been compared between the two disorders.
Measurement factors and confounding could also potentially contribute to observed diagnostics differences. While testing protocols were identical across diagnostic groups, it is possible that our social measures may differentially capture certain aspects of sociality in ASD vs. ADHD (e.g. attachment security, social reward, or other more complex social tasks). While common confounding factors were adjusted for in our analyses, the role of clinical severity, adaptive functioning, and parental stress or psychopathology were not explored, and could potentially affect response patterns on either social metric. The impact of different traits and measurement factors on SCQ scores in ASD and ADHD in particular merits further exploration.
In terms of other limitations, despite our sample size, we may have been underpowered to detect more subtle differences, including interactions with ancestry or analyses on the smaller group with RMET and IQ scores. Lack of statistical significance would therefore not necessarily rule out a possible biological effect of genotype on social behaviour for these measures. Analyses including a comorbid ASD/ADHD group, a control group, or a group of unaffected family members, would have proven interesting comparisons but were not feasible with the existing dataset. While results were statistically significant, differences between genotype groups were generally small and not meant to imply clinical utility. Human molecular/ expression data is greatly needed to clarify the mechanisms through which these variants may influence receptor function and downstream behaviours. In summary, our findings indicate that common polymorphisms in neuropeptide receptor genes act as specific (as opposed to general) modifier alleles regarding social phenotypes in children with neurodevelopmental disorders, with effects that vary by diagnosis. Our findings therefore contribute to the often-conflicting literature regarding these SNPs. We are the first to implicate potential diagnostic differences as one explanation for heterogeneity in findings. Future studies comparing OXTR genotype effects across these disorders ought to examine the role of the hypothalamic pituitary axis social neurocircuitry, environmental social stress, and rare genetic, or epigenetic differences as possible mechanisms. Molecular studies investigating the functional impacts of these SNPs in humans are also greatly needed.

Methods
Participants and study protocol. Participants were recruited via the Province of Ontario Neurodevelopmental Disorders (POND) Network, a coordinated multi-centre research initiative examining the neurobiology of neurodevelopmental disorders. Hospital Research Ethics Boards approved the study protocol at each participating institution (Holland Bloorview Kids Rehabilitation Hospital, Toronto; The Hospital for Sick Children, Toronto; McMaster Children's Hospital, Hamilton; and Lawson Health Research Institute, London); all experiments were performed in accordance with relevant guidelines and regulations. Informed consent was obtained from parents of all participants; further details of the protocol are described elsewhere 25 . All participants (n = 614) were between 4 and 21 years of age, and had a clinical diagnosis of ASD or ADHD (see Table 1). Clinical diagnoses were confirmed via in-depth diagnostic assessments using established measures. Ancestry was determined by parent/caregiver self-report on the ethnic origins of the participants' biological four grandparents. Where all four grandparents were identified as Caucasian (including Scandinavian, Ashkenazi Jewish, French Canadian, Mediterranean, Eastern European, Russian, Western European), the participant was classified as 'Caucasian.' Participants with one or more grandparents of non-Caucasian ancestry were classified as 'non-Caucasian. ' Quantitative behavioural phenotype. The Social Communication Questionnaire (SCQ) is a validated measure completed by a parent/caregiver to screen for ASD 35,61 , developed based on the items and factors in the Autism Diagnostic Interview-Revised 62 . Higher SCQ scores (out of 39) indicate a greater risk, with a cut-off of > 15 proposed to distinguish between those with and without ASD 35 . For context, typically developing children usually score under 10 on the SCQ, while those with ASD have mean scores closer to 20 and above 63,64 . Previous research suggests that mean scores on the SCQ are comparable in children with ASD across IQ strata, and it is valid for use in children with a mental age above 2.0 61 . For this analysis, we examined the total prorated scores on the 28 social interaction and communication items only, excluding the items pertaining to restricted/ repetitive behaviours. We did not adjust for IQ for this measure (see statistical analyses). The Reading the Mind in the Eyes Test (RMET) is a standardized test of social perception abilities 65 . Higher scores on the RMET are indicative of more accurate social perception. The child version of the RMET was used in this study, containing a total of 28 items 34 . Typically developing children tend to score between 15-20 out of 28 on this measure, while children with ASD tend to score 2-4 points lower than control subjects 25,34 . RMET scores are known to vary with IQ 25 , therefore IQ was included as a covariate for these analyses. To more easily facilitate comparison of genotype effects across both measures, we report RMET scores with respect to the number of incorrect items. Therefore, throughout the manuscript, higher scores on either metric suggest greater social deficits. Other behavioural traits were quantified using the Child Behavior Checklist (CBCL) subscales [66][67][68] , and the Revised Children Anxiety and Depression Scale (RCADS) 69 .
Selection of genetic markers. Four SNPs in OXTR from the third intronic region (rs53576, rs2254298, rs237887, rs13316193) were selected for genotyping based on existing literature at the time of study design suggesting an association between these markers and social abilities in participants with ASD or ADHD (Table S1). These SNPs, either alone or contributing to a haplotype, have also been associated with ASD risk 21,31,[70][71][72] . Analysis for linkage disequilibrium (LD) was completed using Ldlink 33 , and the 1000 Genomes Project database of Caucasian samples (see Supplementary Table S3). Although rs13316193 and rs237887 showed moderate linkage, all four SNPs were sequenced to facilitate replication of previous findings within specific diagnostic groups. AVPR1a RS3 microsatellite length was sequenced as well, given initial findings suggesting an association between social abilities and genotype as well as receptor expression levels in typically developing subjects [36][37][38] , despite no association with social abilities in ASD in one study 29 . Genotyping. SNPs were analyzed on the MassARRAY Analyzer 4 system using iPLEX Gold chemistry (both Agena Biosciences, San Diego, CA, USA) using the primers in Supplementary Table S11 and the recommended manufacturer's protocol. Genotypes were called using Typer 4.0 (Agena Biosciences). The AVPR1A RS3 polymorphism primers and methods are also listed in Table S11. Primers were adapted from that used by Tansey et al. to match the Human Feb. 2009 (GRCh37/hg19) assembly and to elevate the Tm but otherwise maintain the amplified position 73 . Results were analyzed using the software GeneMapper v. 3.7 (Life Technologies).

Statistical analyses.
All analyses were two-tailed, with the alpha set at 0.05, performed in SAS 9.3 (2002-2010, SAS Institute Cary, NC, USA). We used a logistic regression analysis to examine the proportion of items selected as 'yes' (for the 28 SCQ social communication items) or incorrect (for the 28 RMET items), using the PROC LOGISTIC function in SAS, which allows us to specify both the total score and the number of trials that this score is based on in the model outcome. This approach permitted adjustments for total number of items (e.g. decreasing the denominator for participants who answered >75% of items but did not complete the entire metric, n = 3 total). We adjusted for the contribution of the following predictor variables: genotype group, age, sex, ancestry (and IQ for analyses involving the RMET) 25 . Logistic regression was chosen for the main analyses given that scores on both metrics are sums of binary variables and therefore, follow a binomial distribution. Initial hypotheses were examined in the ASD group (for OXTR rs53576, rs2254298, and rs237887) and ADHD group (for rs53576, and rs1331619), including Caucasian participants only. We selected one specific categorical genotype risk pattern per SNP (i.e. dominant, recessive, or comparison of homozygous groups) for statistical testing based on the positive findings described in previous papers (Table S1). We subsequently combined data for both the ASD and ADHD groups (Caucasian only) and assessed the SNP*diagnosis interaction term to determine whether the magnitude of the genotype effect on social abilities varied by diagnosis. We present the odds ratios (ORs) of increasing SCQ/RMET scores for the hypothesized 'high-risk' vs. 'low-risk' genotype groups, based on the literature. That is, the odds of obtaining a positive/incorrect score on a single SCQ/RMET item for a participant in the 'high-risk' group divided by the odds of obtaining a positive/incorrect score for a participant in the 'low-risk' group. Thus, an odds ratio of 1 represents no risk group differences, while odds greater than 1 suggest increased risk. For presentation purposes, we also provide the predicted SCQ/ RMET scores by genotype group (by multiplying the predicted proportion of items selected or incorrect by the total number of items) adjusted to age 11.0 years and mean IQ scores. Details on covariates are provided in Supplementary Tables S8 and S9. Unadjusted p-values are reported; a Bonferroni correction for multiple comparisons is described alongside results. Finally, we combined data across ancestry groups and tested for genotype* ancestry interactions. Where non-significant (suggesting the effect of neuropeptide receptor genotype on social abilities does not vary by ethnicity), we pooled data across groups, added ancestry as a covariate (along with genotype, age, sex +/− IQ) and calculated the effect size (OR) in the larger varied ancestry group. For exploratory analyses involving anxiety, we compared genotype groups using ANOVAs; we adjusted for age and sex using ANCOVAs.
Data availability. The datasets generated and analyzed during the current study are publicly available through Brain-CODE. http://www.braininstitute.ca/braincode/.