Variation in the Oxytocin Receptor Gene Predicts Brain Region Specific Expression and Social Attachment

Abstract

Background

Oxytocin (OXT) modulates several aspects of social behavior. Intranasal OXT is a leading candidate for treating social deficits in autism spectrum disorder (ASD) and common genetic variants in the human oxytocin receptor (OXTR) are associated with emotion recognition, relationship quality and ASD. Animal models have revealed that individual differences in Oxtr expression in the brain drive social behavior variation. Our understanding of how genetic variation contributes to brain OXTR expression is very limited.

Methods

We investigated Oxtr expression in monogamous prairie voles, which have a well characterized OXT system. We quantified brain region-specific levels of Oxtr mRNA and OXTR protein with established neuroanatomical methods. We used pyrosequencing to investigate allelic imbalance of Oxtr mRNA, a molecular signature of polymorphic genetic regulatory elements. We performed next-generation sequencing to discover variants in and near the Oxtr gene. We investigated social attachment using the partner preference test.

Results

Our allelic imbalance data demonstrates that genetic variants contribute to individual differences in Oxtr expression, but only in particular brain regions, including the nucleus accumbens (NAcc), where OXTR signaling facilitates social attachment. Next-generation sequencing identified one polymorphism in the Oxtr intron, near a putative cis-regulatory element, explaining 74% of the variance in striatal Oxtr expression specifically. Males homozygous for the high expressing allele display enhanced social attachment.

Discussion

Taken together, these findings provide convincing evidence for robust genetic influence on Oxtr expression and provide novel insights into how non-coding polymorphisms in the OXTR might influence individual differences in human social cognition and behavior

INTRODUCTION

Oxytocin (OXT) is a neuromodulator that influences reproductive and social behavior through signaling via a single G-protein coupled oxytocin receptor (OXTR) in the brain. OXTR affects a range of social behaviors in animals including maternal nurturing and bonding (1, 2), social reward and gregariousness (3, 4), social recognition (5), and pair bonding in monogamous species (6–8). It has been proposed that OXT influences these complex social behaviors by increasing the salience and reinforcing value of social stimuli (9).

In humans, intranasal OXT reportedly enhances many aspects of social cognition, including trust, emotion recognition, and eye gaze (1, 10). However, individual reports of the effects of intranasal OXT should be interpreted cautiously (11). Nevertheless, the OXT system is a leading candidate target for improving social function in psychiatric disorders such as autism spectrum disorder (ASD) and schizophrenia (12–17), and one report suggests that OXTR expression may be reduced in ASD (18). Single nucleotide polymorphisms (SNPs) in noncoding regions of the human OXTR have been associated with pair bonding behaviors (19), parenting (1), face recognition skills (20), and autism (21). Some individuals have heterogeneous responses to OXT (22, 23), which may involve interaction with OXTR genetic polymorphisms (24–27). Early life stress in humans can interact with OXTR variation to influence adult social behavior and emotional regulation (28–31). Despite these many associations with human social behavior and disorders, the neural mechanisms by which non-coding SNPs in the OXTR could influence behaviors has yet to be explored.

One potential mechanism is that typical expression of OXTR is disrupted when such SNPs occur in regulatory elements that primarily lie within non-coding portions of the DNA (cis-REs). OXTR is distributed throughout the brain of many vertebrates and, importantly, the pattern of OXTR distribution is diverse between species (32, 33). OXTR is often enriched within regions of crucial behavioral circuits such as the mesolimbic reward (MLR) network and social decision making network and appears to modulate these networks to generate species-specific social strategies such as monogamous social attachments (32) and social gregarity (7). Thus, the manner in which the Oxtr gene is regulated between species appears to have profound consequences for the manner in which neural networks activate in response to the social environment.

In a socially monogamous rodent, the prairie vole (Microtus ochrogaster), OXTR is enriched in important MLR regions such as the nucleus accumbens (NAcc) and prefrontal cortex (PFC) that constitute part of a neural network for pair bonding (6, 8, 34). OXTR density is much higher in the NAcc of prairie voles than promiscuous vole species and OXTR signaling in the NAcc is required for mating-induced partner preference formation, a laboratory proxy of pair bonding (6). Infusion of an OXTR antagonist into the NAcc or the PFC, but not the caudate putamen (CP) blocks mating-induced partner preferences in females (35) and males (A.C. Keebaugh and LJY, unpublished data). OXTR density also varies between individual prairie voles, especially in the NAcc along with the CP (36). Increasing OXTR density in the NAcc using viral-vector mediated gene transfer facilitates partner preference formation, while decreasing OXTR density in the same region using RNA interference inhibits such bonding (37–39). Variation in NAcc OXTR density is correlated with individual differences in monogamy-related behavior in males in naturalistic settings (40). Furthermore, variation in prairie vole NAcc OXTR confers susceptibility or resilience to the effects of daily neonatal isolations, a model of neglect, in relation to the ability to form social attachments as adults (41). Mechanisms responsible for OXTR diversity in the prairie vole NAcc may be important determinants of individual differences in social behavior as well.

One likely causal explanation for OXTR diversity is that genetic polymorphisms in cis-REs regulating Oxtr generate variation in expression in a brain region-specific manner. Variation in gene expression mediated by cis-REs plays an important role in evolutionary phenotypic change (32, 42–46). In prairie voles, a microsatellite in the 5’ flanking region of the vasopressin receptor gene (Avpr1a) containing cis-REs has been shown to have functional influence over species differences and individual variation in Avpr1a expression and is associated with variation in social behavior (47, 48). The influence of cis-REs can be detected by assaying for allelic imbalance, which is observed when two alleles of a gene in a heterozygous individual are expressed at different rates, creating an imbalance in the respective mRNAs (49–51). Any differences in mRNA levels between alleles occur in the same nuclear environment, where both alleles should be affected equally by environmental, hormonal or epigenetic factors, unless cis-REs proximal to the alleles are variable. Allelic imbalance is commonly observed in a tissue-specific manner (51, 52).

To determine whether prairie vole Oxtr gene expression is influenced by polymorphic cis-REs, we analyzed brain region-derived mRNA for allelic imbalance in animals heterozygous for a SNP in the Oxtr transcribed region. We found that robust allelic imbalance of Oxtr occurs within the striatum, but not in several other brain regions. Voles with alternative homozygous genotypes for this SNP had significant differences in NAcc OXTR density. Finally, to gain a more thorough understanding of the relationship between genetic polymorphisms in the prairie vole Oxtr and neural OXTR density, we sequenced 70 kilobases of DNA around the gene in 45 voles. We observed strong associations between several genetic markers and OXTR density that were particularly robust in the NAcc. A bioinformatics analysis using ENCODE data suggests that an intronic SNP is the most likely functional candidate for further investigation. This intronic SNP is strongly associated with OXTR density in the NAcc and was also found to associate with the propensity to form social attachments. Our results demonstrate for the first time that non-coding SNPs in the Oxtr can profoundly predict OXTR binding and Oxtr expression in a brain region-specific manner. These findings implicate that cis-regulation drives the remarkable variation in Oxtr transcription and has a more modest, but significant, influence on social behaviors. This is the first study to demonstrate that SNPs in the non-coding region of the oxytocin receptor gene robustly effect receptor density in the brain.

MATERIALS AND METHODS

Animals

Prairie voles (Microtus ochrogaster) were housed in same-sex groups with 2–3 voles/cage from post-natal day (PND) 21. Housing consisted of a ventilated 36×18×19 cm Plexiglass cage filled with Bed-o-Cob Laboratory Animal Bedding under a 14:10 h light/dark cycle (lights on 7:00 AM–9:00 PM) at 22 °C with access to food (rabbit LabDiet) and water ad libitum. Our laboratory breeding colony was originally derived from field captured voles in Illinois. All procedures were approved by the Emory University Institutional Animal Care and Use Committee.

Sanger sequencing and polymorphism discovery for an allelic imbalance marker

DNA was isolated using a Qiagen DNeasy kit (Germantown, MD). We designed primers to amplify five loci spanning the two coding exons plus the 5’ UTR and parts of the 3’ UTR. For details see Supplemental Information. Nucleotide 204321 (NT204321), located in the 3’UTR, was polymorphic (minor allele frequency: 0.33) and was used for the detailed allelic imbalance study.

Allelic imbalance

Subjects were euthanized with CO₂. Brains were frozen in crushed dry ice and stored at −80° C. Nucleic acids for the allelic imbalance assay were isolated from microdissected brain tissues using the Qiagen mRNA/DNA Micro Kit. For details see Supplemental Information.

Long-range PCRs for target enrichment of 70kb surrounding Oxtr

DNA was isolated from previously sectioned brains stored at −80° C with the Qiagen DNeasy Kit. All PCRs were performed using the Qiagen LongRange PCR Kit. Ten loci between 6.6 – 10 kb were amplified. For details see Supplemental Information.

Amplicon Library Preparation

Sequencing library preparation and sequencing analyses were performed by the Yerkes Nonhuman Primate Genomics Core (Atlanta, GA). PCR amplicons from each animal were pooled and cleaned using Solid Phase Reversible Immobilization (SPRI) beads (Beckman Coulter). Libraries were generated using the Illumina Nextera XT DNA Library Prep Kit (San Diego, CA), dual barcoding and sequencing primers were added according to the manufacturer protocol. Libraries were validated by microelectrophoresis, quantified, pooled and clustered on Illumina TruSeq v3 flowcell. Clustered flowcell was sequenced on an Illumina HiSeq 1000 in 100-base single-read reactions.

Amplicon sequencing analysis

Sequencing reads were mapped to Microtus ochrogaster target 220 haplotype 2 genomic scaffold (DP001215.2) using the Burrows-Wheeler Aligner (bwa version 0.7.10) (53). The aligned reads were processed with the DNAseq Variant Analysis workflow of the Genome Analysis toolkit (GATK v3.2.2) (54), including marking duplicate reads and local realignment around insertion/deletions. Variants were called on a per-sample basis and then combined to produce a joint variant call file.

Oxytocin receptor autoradiography

Oxytocin receptor autoradiography was performed as previously reported (37). For details see Supplementary Information.

In situ hybridization

Sense and antisense ³⁵S-UTP-labeled RNA probes for prairie vole Oxtr mRNA were generated as described previously (55). For details see Supplementary Information.

Partner preference test

The partner preference was performed as previously described (56). For details see Supplemental Information

Statistical analysis

All statistical analyses were performed in the R statistical software package version 3.1.1, unless stated otherwise. Associations between genetic information and brain data were examined using linear regression with Bonferroni corrections for multiple comparisons. Regarding the factor analyses, to determine how many factors to extract we used the “nFactors” package in R and the factor extraction decisions were based on the eigenvalues-greater-than-one rule, parallel analysis, the optimal coordinates method and acceleration factor. Exploratory factor analyses were performed with the factanal() function, using “varimax” as the rotation method. Processing of next-generation sequencing data was performed in VCFtools, a program package designed for working with VCF files from sequencing projects (57). For associations between individual markers in the Oxtr sequence and autoradiography expression data, linear regressions were performed using PLINK/SEQ, an open-source library for working with genetic variation data (URL: https://atgu.mgh.harvard.edu/plinkseq/). For further details see Supplemental Information.

RESULTS

As expected based on previous experiments, the NAcc and caudate putamen (CP) exhibited more individual variation in OXTR density than other brain regions (Figure 1a–c). Furthermore, OXTR binding density appeared to be correlated with Oxtr mRNA levels based on in situ hybridization signal (Figure 1a). To test the hypothesis that the high variability in OXTR density within the NAcc was due to the influence of putative cis-REs, we first assayed for allelic imbalance. We sequenced the transcribed region of the Oxtr in a small sample of voles to identify any SNP in our prairie vole colony with a relatively high minor allele frequency. One SNP in the 3’ untranslated region (minor allele frequency 33%), heretofore referred to as NT204321 based on the position of this nucleotide on a sequenced prairie vole bacterial artificial clone (58) (DP001215.2), was identified using this method. In heterozygous animals for this SNP, we found significant allelic imbalance in NAcc (cDNA: 3.16, gDNA threshold: 1.11), caudate putamen (CP) (cDNA: 5.24, gDNA threshold: 1.13) and to a lesser degree, amygdala (Amyg) (cDNA: 1.19, gDNA threshold: 1.14). The allelic imbalance was pronounced in the two striatal sub-regions, NAcc and CP, with the T-allele transcript being 3–5 times more prevalent than the C-allele in the same animals in these regions (Figure 1d). These data strongly suggest that cis-REs linked to NT204321 generate individual variation in expression of Oxtr in select brain tissues.

Figure 1.

Region-specific variation in neural expression of the prairie vole Oxtr gene is caused by cis-regulatory elements. (a) An illustration that OXTR protein density (visualized by receptor autoradiography) and Oxtr mRNA levels (visualized by in situ hybridization) vary in the striatum, including the nucleus accumbens (NAcc). (b) In contrast to the individual differences in NAcc and CP OXTR density, other brain regions show less variation. Black scale bar represents 100 µm. (c) Quantification of individual variation in OXTR density. Each dot represents OXTR density (dpm/mg) for an individual prairie vole (N=12, males). NAcc, nucleus accumbens; CP, caudate putamen; LS, dorsal lateral septum; Cl, claustrum; Ins, insular cortex; PFC, prefrontal cortex; Amyg, amygdala. (d) Allelic imbalance was calculated as the average allelic ratio (%T/%C) for cDNA and genomic DNA (gDNA) from animals heterozygous at NT204321 (N=8). A significant allelic imbalance was detected in the cDNA derived from the nucleus accumbens (NAcc), caudate putamen (CP) and amygdala (Amyg), but not in cDNA derived from the prefrontal cortex (PFC) or lateral septum (LS). The striatal regions had very high allelic imbalance, with 3–5 fold differences between alleles. gDNA T and C-alleles are amplified at equal levels in all tissues. * cDNA allelic ratio is significantly greater than a threshold calculated by the mean of the gDNA allelic ratio + 3 gDNA standard deviations. Data are expressed as mean ± standard error of the mean (SEM). Except for CP gDNA (n=6), n=7.

To test if NT204321 might serve as a marker to predict overall OXTR density in the NAcc, we collected expression data for 12 brain regions (N=31) using autoradiography. Visual inspection of OXTR density across brain regions suggested that density in some regions co-varied with NAcc density, while density in other regions did not. Therefore, we used factor analysis to determine the correlation structure between OXTR density data from the 12 brain regions investigated. We hypothesize that correlations between OXTR densities from different brain regions can be explained by unobserved variables, possibly reflecting transcriptional processes giving rise to the patterns of correlation. Exploratory factor analysis is a method to identify such unobserved, latent variables. Our analysis revealed two factors together explaining the majority of variance (58%). Factor 1 strongly reflects co-variability in a set of regions involved in reward processing (NAcc, CP, and olfactory tubercle) while the second factor reflects covariation between cortical and subcortical regions that have relatively uniform levels of OXTR density (Table 1a). We identified similar patterns when we investigated the associations between NT204321 and the OXTR expression in the 12 brain regions, with regions loading into Factor 1 being more related to genotype than those loading into Factor 2 (Figure 2). We performed a second factor analysis in an additional sample (N=85) (Table 1b) and this analysis, like the first one, revealed two factors explaining most of the variance. OXTR binding in the NAcc was almost perfectly correlated with the first factor and binding in the insular cortex (Ins) was close to perfectly correlated with Factor 2. Thus further analyses including brain data focused on these two regions as representatives of Factor 1 and Factor 2.

Table 1.

Two factors encompass the covariation in OXTR density amongst brain regions

a	Factor 1	Factor 2
NAcc	0.99	0.14
Tu	0.82	0.09
CP	0.82	0.18
CeA	0.69	0.3
OB	0.56	0.5
LS	0.56	0.15
Cl	0.17	0.88
Ins	0.12	0.85
BLA	0.04	0.71
PFC	0.5	0.51
VMH	0.3	0.47
AON	0.28	0.41

b	Factor 1	Factor 2
NAcc	0.99	0.04
CP	0.92	0.1
Tu	0.91	−0.01
LS	0.56	0.14
Ins	−0.06	0.99
Cl	0.07	0.75
PFC	0.51	0.6

Values in the table represent factor correlations. NAcc, nucleus accumbens; Tu, olfactory tubercle; CP, caudate putamen; CeA, central amygdala; OB, olfactory bulb; LS, dorsal lateral septum. Cl, claustrum; Ins, insular cortex; BLA, basal lateral amygdala; PFC, prefrontal cortex; VMH, ventral medial hypothalamus; AON, anterior olfactory nucleus.

Figure 2.

OXTR binding density in striatal and olfactory regions is associated with NT204321. Samples sizes for each genotype are C/C=11, C/T=8 and T/T=12. (a) Brain regions are sorted based on the how strongly the regions loaded on Factor 1. Within Factor 1 regions, OXTR binding is significantly related to genotype in NAcc, CP and OB. (b) Brain regions are sorted based on the how strongly the regions loaded on Factor 2. In the Factor 2 grouping, only the AON was significantly associated with genotype. Associations were investigated using simple linear regression. * P < 0.004 (α corrected for 12 comparisons). NAcc, nucleus accumbens; Tu, olfactory tubercle; CP, caudate putamen; CeA, central amygdala; OB, olfactory bulb; LS, dorsal lateral septum; Cl, claustrum; Ins, insular cortex; BLA, basal lateral amygdala; PFC, prefrontal cortex; VMH, ventral medial hypothalamus; AON, anterior olfactory nucleus. Data is shown as individual OXTR density (dpm/mg) with trend line for the linear regression.

NT204321 was not chosen based on any assumption of functional importance and we suspected that other SNPs across and outside the Oxtr transcribed region might be more closely associated with the cis-RE and better predict Oxtr expression. We first characterized the suite of polymorphisms across the Oxtr gene by sequencing 70 kb of DNA including and surrounding the gene. Amongst the 45 voles we sequenced, we identified 967 SNPs with a read depth no lower than 100 reads, a quality score of at least 1000, and for which all were variable in our sample. Figure 3a shows a quantile-quantile plot for the association between the SNPs in the Oxtr sequence and NAcc OXTR binding density. Clearly, as can be seen in Figure 3a many of the SNPs in our set were strongly associated with NAcc OXTR density. We were primarily interested in identifying variants that could potentially be functional and therefore focused on the SNPs with the lowest P-values (and largest effect sizes). Due to linkage disequilibrium, 15 SNPs spanning a 30 kb region were associated with NAcc OXTR density with the same, minimal P-value (P=1.06×10⁻¹⁵, adjusted R²=0.78), which is a remarkable effect size for a genotype-phenotype relationship.

Figure 3.

Identification of NT213739 as a marker of NAcc OXTR density. (a) Quantile-quantile plot of the associations between the 967 Oxtr markers and NAcc OXTR density. Each dot represents the –log P of the association between a particular SNP with OXTR density placed in ascending order. (b) A schematic of the mouse Oxtr gene with accompanying functional data from the ENCODE project. The prairie vole sequences containing the SNPs showing the strongest association with NAcc OXTR density that also map to mouse Oxtr sequence. Approximately 500 bp per SNP of prairie vole sequence surrounding each SNP was aligned to the mouse genome as indicated by red rectangles in the figure. ChIP-seq signal or DNaseHS signal is shown. OB, olfactory bulb; Brain, whole brain; CTCF, CCCTC-binding factor (red) ; H3K4m1, a single methyl modification of lysine 4 of the H3 histone (orange); H3K27a, acetylation modification of lysine 27 on the H3 histone (green); DNaseHS, deoxyribonuclease I hypersensitive sites (dark blue).

In order to assess whether any of the 15 most associated SNPs are more likely than others to lie in a putative cis-RE, we investigated their homology with regions of the mouse Oxtr gene that overlap with signatures of functional activity occurring within neural tissues where OXTR is expressed in the mouse (59). We mapped vole sequence containing these SNPs and surrounding sequence of approximately 500 bp per SNP to the mouse Oxtr gene. We compared our vole sequences to ENCODE tracks for markers of general transcriptional activity such as DNase hypersensitivity, H3K4me1, and H3K27a as well as binding of the transcription factor CTCF, which can act as a canonical transcription factor or as an organizer of genomic architecture (60). Only one SNP overlapped with strong signatures of transcriptional function, a SNP occurring at nucleotide 213739 (NT213739, minor allele frequency: 0.32) (Figure 3b). The sequence containing NT213739 overlapped peaks of DNase hypersensitivity and CTCF binding within the large intron, a region proposed to contain cis-REs in humans (20, 61, 62). Based on this evidence we chose to further investigate the predictive power of NT213739 in two additional samples.

After genotyping additional voles (Sample 2 n=35, Sample 3 n=31) for NT213739, we confirmed the SNP was robustly associated with OXTR density in the NAcc but not in the Ins (Figure 4a). Indeed, in both the second and third independent samples investigated, our findings from the sequenced sample were very closely replicated. In all of our relatively small samples, NT213739 was strongly associated with NAcc expression (all P-values<4×10⁻¹⁰). Similarly, the effect size of this association was very large in all samples (adjusted R²≥74%), strongly suggesting that NT213739 explains at least 74% of the variance in NAcc expression of OXTR in prairie voles. The effect was marked, such that OXTR density values in the NAcc between homozygous animals of the two genotypes did not overlap at all while heterozygous animals displayed an intermediate phenotype.

Figure 4.

NT213739 genotype robustly predicts Oxtr expression. (a) NAcc OXTR density was associated with NT213739 genotype across three independent samples (Sample 1, adjusted R²=0.81; Sample 2, adjusted R²=0.90; Sample 3, adjusted R²=0.74). OXTR density in the insula (Ins) was not associated with NT213739. (b) (c) NAcc Oxtr mRNA density was significantly associated with NT213739 genotype (N=31, adjusted R²=0.69). (c) NAcc Oxtr mRNA density is significantly correlated with OXTR protein binding density (N=31). (d) Representative images highlighting the differences between NT213739 genotypes in Oxtr expression within the striatum, particulary the nucleus accumbens (NAcc). Black scale bar represents 100 µm.* P < 1×10⁻⁸. Data is shown as individual OXTR density (dpm/mg) or individual mRNA density (ROD) with trend line for the linear regressions.

To confirm that the association between NT213739 and receptor density is mediated through mRNA levels, we performed in situ hybridization on adjacent sections of brains from 31 individuals. We found that NT213739 is also significantly associated with Oxtr mRNA levels in the NAcc (Figure 4b) and that mRNA levels and OXTR density were significantly correlated (Figure 4c, 4d). Together these data suggest that NT213739 is strongly associated with transcriptional variation of the Oxtr gene and tightly linked to a cis-RE.

Since NT213739 robustly predicted NAcc OXTR density, and NAcc OXTR signaling is important for regulating partner preference formation in prairie voles, we sought to determine whether NT213739 would influence mating-induced partner preference formation, a measure of social attachment that involves social information processing and social reinforcement. Males of varying genotype were housed with a female for 6 h and then tested in the partner preference test. There was a significant effect of genotype on partner preference formation, with C/C voles spending significantly more time huddling with the partner than the stranger, while C/T and T/T males failed to display a partner preference (Figure 5).

Figure 5.

NT213739 genotype influences partner preference formation in male prairie voles. The effect of genotype on behavior was investigated using two-way ANOVA. The interaction of genotype×stimulus on huddling duration was significant (F_(1,136)=4.45, P=0.037). Males with a C/C genotype (N=39) spent significantly more time huddling with the partner than the stranger while animals with a C/T (N=13) or T/T (N=18) genotype did not. * indicates a partner preference, mean partner huddling time is significantly greater than mean stranger huddling time (t-test, P < 0.01). Data are expressed as mean ± SEM.

DISCUSSION

We here demonstrate that genetic variation in the Oxtr exerts robust control over individual diversity in Oxtr expression and OXTR density in the prairie vole brain and that this influence on expression occurs in a region-specific manner. Allelic imbalance is strongest in the striatum and genotype-OXTR associations are most robust in this region. Exploratory factor analysis identified a cluster of striatal-olfactory regions with correlated OXTR density. One intriguing interpretation of this finding is that the unobserved latent variable represented by Factor 1 may reflect a set of transcriptional regulators with maximal effect on Oxtr expression activity in prairie vole olfactory-reward processing regions. Whatever transcription factor or factors leads to the covariation appear to interact with the cis-REs associated with NT213739 to generate the high variation in expression observed in these regions, but not other regions. In this manner cis-REs appear to contribute exquisite control over OXTR and through this process, influence behavioral diversity in prairie voles.

Little is known about the molecular mechanisms regulating brain region-specific Oxtr expression in any species. Prairie voles and montane voles display species-specific patterns of OXTR expression. Comparisons of ~1500 bp of 5’ flanking regions from the Oxtr gene reveal only a few SNPs and 99% homology between the species (63). Transgenic mice carrying a reporter gene driven by 5 kb of the prairie vole Oxtr 5’ flanking sequence express the reporter in the brain in a pattern resembling prairie voles (64), suggesting the sequence may suffice for some aspects of brain region-specific expression. DNA methylation of Oxtr differs between brain regions in rodents (65, 66). Interestingly, DNA methylation of OXTR in humans may be mediated by an intronic SNP (67). In some rodent brain regions, OXTR expression is regulated by gonadal steroids in a species specific manner. For example, testosterone increases OXTR in the hypothalamus of rats but decreases OXTR in mice (68, 69). Clearly a better understanding of the molecular mechanisms underlying species differences and individual variation in Oxtr transcription in the brain is needed and may inform our understanding of how genetic variation in human OXTR relates to psychiatric phenotypes or responses to OXT based therapies.

We found that prairie vole Oxtr expression is strongly influenced by polymorphic cis-REs that include or are associated with NT213739, a SNP in the intron of the gene. We focused our attention on NT213739 because out of 15 SNPs in perfect LD with one another, only NT213739 mapped to a site in the mouse Oxtr intron with robust evidence of transcriptional activity. We note that such comparisons should be made with the understanding that transcription factor binding sites may differ between species (70). We were particularly interested in the proximity of the mapped prairie vole sequence to a putative CTCF binding site, as CTCF binding peaks are found in the intron as well as near the promoter of both the mouse Oxtr (Figure 3b) and human OXTR (20). One role of CTCF is to act as a spatial organizer, aiding DNA looping to permit cis-RE-promoter interactions required for gene transcription (60). Polymorphic cis-REs have been shown to influence CTCF binding (71, 72) and DNA looping (73). In the human OXTR third intron, the CTCF peak is found near a SNP, rs237887, that was predicted to be near a cis-RE, and associated with face recognition abilities (20) and ASD diagnosis (21). Additionally, the human intron may contain other cis-REs (61) and accumulates rare SNPs in cases of ASD (62). If the large intron of Oxtr contains cis-REs in multiple species, spatial organization of DNA by factors such as CTCF could offer a potentially general regulatory mechanism required for proper function of species or region-specific cis-REs. While this evidence provides a potential molecular mechanism by which NT213739 could lead to region-specific differential transcription, it is important to note that our current genotype-phenotype relationships do not implicate NT213739 above any of the other SNPs in perfect LD with it spanning the 30 kb. Future studies using large samples of more genetically diverse animals, including wild caught specimens, may be needed to break this haplotype structure in order to identify which SNP is most likely functionally contributing to OXTR density variation. Furthermore, biochemical analyses, including chromatin immunoprecipitation followed by sequencing (ChIP-seq), chromosome conformation capture (3C) (74), and in vitro transcription assays could be used to investigate interactions between CTCF binding, DNA looping and Oxtr regulation.

One of the most remarkable findings of our study is the amount of OXTR density variance explained by genetic polymorphism. Our data suggests that NT213739, or any of the SNPs in perfect LD with it, explain 74% of the variation in NAcc OXTR. Behavioral genetic studies typically report that 1–10% of the variance in behavioral phenotype is explained by candidate gene polymorphisms. We presume that OXTR density is biologically more proximate to genotype than behavior, and in typical behavioral genetic studies many more variables are contributing to behavioral variation. Thus direct measures of brain phenotype are likely to yield stronger effect sizes than reported in behavioral studies.

Prairie voles strongly express OXTR in regions of the MLR, a conserved neural network (75) involved in reward processing and implicated in numerous human psychiatric disorders including depression (76), addiction (77), and schizophrenia (78). In a key region of the MLR, the NAcc, OXTR activation is necessary for prairie vole partner preference formation (35) and OXTR variation mediates individual differences in pair bonding behaviors (37–39). Here, we confirm a genetic role for naturally occurring OXTR density differences that contribute to individual variation in social behavior (36, 40). In addition to region-specific effects, OXTR signaling enhances functional connectivity within a network of regions during prairie vole pair bond formation (79). Recent research in humans found that intronic OXTR variants predicted individual differences in functional connectivity between brain regions during the processing of social information (80, 81). These results highlight a potentially conserved role for OXTR in social cognition between species despite likely differences in sites of expression. The prairie vole may prove a useful model to understand how individual differences in OXTR expression in key regions lead to variation in network connectivity.

A recent meta-analysis supports the conclusion that genetic variation in OXTR is associated with ASD diagnosis (21), and others have reported associations with endophenotypes of ASD (20, 82, 83), including structural variability in brain regions relevant to social cognition (80, 84). Such findings in humans call for a better understanding of molecular mechanisms regulating OXTR variation (21, 85). The present study suggests that these genotype-phenotype relationships may be mediated by polymorphisms in cis-REs in the OXTR gene that influence OXTR density in a brain region-specific manner. Here we use the prairie vole model to demonstrate for the first time that a single SNP can predict the majority of variance in OXTR expression in specific brain regions. Further studies to identify functional mechanisms leading to this difference in Oxtr transcriptional activity may provide exciting insights into the precise genetic mechanism generating OXTR mediated diversity in social behavior, including human psychopathology.