Introduction

Oxytocin is an evolutionarily conserved neuropeptide implicated in an array of social and reproductive behaviors; its role in complex behavioral traits and in the pathophysiology of mental health conditions has attracted considerable attention1. Oxytocin is mostly synthesized in neurons in the supraoptic nucleus and the paraventricular nucleus for both systemic and central release. Research on humans has shown beneficial effects of intranasal oxytocin on performance on tests assessing social cognition2 and gaze to the eye region3. Moreover, single nucleotide polymorphisms in oxytocin pathway genes have been associated with social behavior and psychiatric disorders4. Emerging evidence also points to the oxytocin system’s role in energy metabolism, with relevance for metabolic functions5.

The distribution of oxytocin receptor (OXTR) mRNA across the brain provides a proxy for the distribution of oxytocin binding6, allowing for a detailed mapping of the anatomical geography of the oxytocin system in the brain. Seminal animal work using histochemistry and immunohistochemistry revealed high concentrations of OXTR mRNA in the hypothalamus, amygdala, olfactory bulb, ventral pallidum, and the dorsal vagal nucleus in rodents7,8. Further, experimentally increasing9 or decreasing10 OXTR expression in the prairie vole nucleus accumbens modulated partner preference behavior, suggesting a correspondence between the spatial distribution of OXTR mRNA, its functional neuroanatomy, and behavioral relevance. While OXTR mRNA localization in the rodent brain is well-described11, its anatomical distribution across the human brain is poorly understood, as investigations have historically tended to sample very few brain regions12. While more recent work has exclusively examined whole-brain distribution of OXTR genes against a limited set of genes13 and overexpression in broad functional brain networks (e.g., limbic network)14, researchers have yet to explore OXTR’s associations with all protein-coding genes or with gene expression patterns associated with specific cognitive states.

Characterizing interactions of OXTR with other key elements of the oxytocin signaling pathway and biological systems beyond this pathway is critical for determining oxytocin’s behavioral and functional relevance. Along with OXTR, CD38 and OXT genes in the oxytocin signaling pathway have been implicated in human social behavior4. Specifically, CD38 is involved in oxytocin secretion via Ca2+ release15, and oxytocin-neurophysin I (OXT) encodes the oxytocin prepropeptide containing the nonapeptide oxytocin and the carrier protein neurophysin-I16. OXT mRNA has been shown to be highly expressed in human paraventricular nucleus of the hypothalamus, the lateral hypothalamic area, and the supraoptic nucleus, and there is evidence of co-expression with OXTR mRNA and the μ and κ types opioid receptor mRNA13, providing a putative avenue for interactions between the oxytocin and opioid pathways. The interactions with the oxytocin system extend beyond the opioid pathway, including the dopaminergic17,18 and muscarinic acetylcholine12,19 circuits, with possible implications for social behavior and psychiatric disorders. For instance, the dopamine D2-receptor subtype (DRD2), has been implicated in various putative intermediate phenotypes in psychiatric illness, including motivational processing20 and pair bonding in animal models21. Moreover, the muscarinic acetylcholine M4 receptor (CHRM4) has been associated with schizophrenia22 and implicated in cognitive flexibility23 and dopamine release24. Finally, oxytocin can also bind to AVPR1A receptors25, which have also been linked to social functioning26. However, the mRNA co-expression of these systems, and potentially others, with the oxytocin system is not well characterized.

While brain regions with high oxytocin pathway gene expression in humans have been identified12,13, inference of specific cognitive states from the activity of single brain regions is elusive. For instance, commonly observed increases in medial and lateral frontal region activity during emotion and pain processing seem to be better explained by more general sustained attention processes27. By collecting data from more than 14,000 fMRI studies, NeuroSynth27 allows for reverse inference of cognitive states based on a given brain gene expression map with high specificity. Establishing the specific cognitive state correlates of oxytocin pathway genes will provide a deeper understanding of the human oxytocin system and its relevance for brain functions and mental health.

Here we characterize the anatomical distribution of oxytocin pathway mRNA expression in the human brain, identify putative gene interactions, and explore the functional relevance of these patterns. Expression of critical oxytocin pathway genes are enriched in subcortical and olfactory regions and these genes are highly co-expressed with several dopaminergic and muscarinic acetylcholine genes, along with genes associated with metabolic regulation. Moreover, OXTR and CD38 contribute to a co-expression gene network that is significantly enriched in a waist-hip ratio GWAS along with genesets associated with cognition and feeding behavior. The results from a large-scale fMRI meta-analysis suggest that oxytocin pathway gene expression maps correspond with the processing of anticipatory, appetitive, and aversive cognitive states. These results indicate that the oxytocin signaling system may operate synergistically with the dopaminergic and muscarinic acetylcholine signaling systems to exert its complex effects on cognition.

Results

Oxytocin gene expression patterns in the brain

The full dataset of protein coding genes (n = 20,737) from six donor brains were collected from the Allen Human Brain Atlas (http://human.brain-map.org/). Each brain was sampled in 363–946 distinct locations, either in the left hemisphere only (n = 6), or over both hemispheres (n = 2) using a custom Agilent 8 × 60 K cDNA array chip. Analyses were performed on left hemisphere samples due to a larger sample size. Individual brain maps were non-linearly registered to the MNI152 (Montreal Neurological Institute) template using Advanced Normalization Tools28. Next, we extracted region specific statistics for 54 brain regions based on the Automated Anatomical Label (AAL) atlas.

Of primary interest were the three oxytocin pathway genes that have been associated with social behavior in both animal and human research: OXTR, CD38, and OXT4. Four other selected sets of mRNA, which are thought to co-express with oxytocin pathway mRNA and modulate social behavior, were also of specific interest: A dopaminergic set (DRD1, DRD2, DRD3, DRD4, DRD5, COMT, and DAT1)17,18, a muscarinic acetylcholine set (CRHM1, CRHM2, CRHM3, CRHM4, and CRHM5)12,19, a vasopressin set (AVPR1A, AVPR1B)26,29 and an opioid set (OPRM1, OPRD1, and OPRK1)13. Of additional interest were complete sets of dopaminergic (n = 63), cholinergic (n = 79), and oxytocinergic (n = 94) genes, which were identified using the Kyoto Encyclopaedia of Genes and Genomes (KEGG) database (http://www.kegg.jp). Full gene set lists are provided in Supplementary Data 1.

The Genotype-Tissue Expression (GTEx) project database30 was used for independent sample validation. Although this dataset provides gene expression data from fewer brain regions (i.e., 10) compared to the Allen dataset, the data is derived from a larger dataset of donors (mean sample size for mRNA expression across brain regions = 131.7, range = 88–173). Median gene expression profiles from 10 distinct brain regions were extracted for the above-specified 20 genes of interest from the GTEx database and median values were calculated for these same 10 regions from the Allen dataset (see Methods, Supplementary Fig. 1, and Supplementary Data 1). For independent sample validation, the rank-order correlation of gene expression between the Allen and GTEx datasets using the 10 brain regions reported in the GTEx database was calculated.

Region-of-interest gene expression

Compared to average expression across the brain, statistically significant higher expression of OXTR mRNA levels was observed in the olfactory bulbs (p = 0.048, d = 2.1; Supplementary Data 1). Subcortical structures had higher than average expression of OXTR, associated with Cohen’s d values >1, but these did not reach FDR corrected (54 tests) thresholds of statistical significance (Pallidum: p = 0.1, d = 1.7; Caudate: p = 0.12, d = 1.5; Putamen: p = 0.13, d = 1.3; Thalamus, p = 0.18, d = 1.1). There was significantly lower OXTR expression in the cerebral crus II (p = 0.01, d = 4), cerebellum 7b (p = 0.01, d = 3.6), and cerebellum 8 (p = 0.01, d = 4.3; Fig. 1a). OXTR expression patterns in the brain were comparable to patterns in the GTEx database (rs = 0.81, p = 0.01). CD38 gene expression was statistically significantly higher than average in the caudate (p = 0.004, d = 4.1), pallidum (p = 0.004, d = 4.6), olfactory bulbs (p = 0.01, d = 3.4), putamen (p = 0.02, d = 2.7), thalamus (p = 0.03, d = 2.1), and cingulate anterior (p = 0.03, d = 2; Fig. 1b). There was lower CD38 expression in the cerebral crus II (p = 0.045, d = 1.8), cerebellum 7b (p = 0.03, d = 2.1), and cerebellum 8 (p = 0.03, d = 1.9). CD38 expression patterns in the Allen database were comparable to CD38 expression patterns in the GTEx database (rs = 0.67, p = 0.04). There were no brain regions with significantly higher than average OXT expression after FDR correction (54 tests; Supplementary Fig. 2). OXT expression patterns in the Allen database were not significantly associated to the GTEx expression patterns (rs = 0.22, p = 0.54). Expression of genes included in the selected dopaminergic, muscarinic acetylcholine, vasopressin, and opioid sets are presented in Supplementary Figures 3–6, respectively. Of note, DRD1, DRD2, CHRM4, and OPRK1 genes were highly expressed in subcortical regions, including the caudate, pallidum, putamen, and thalamus (Supplementary Data 1).

Fig. 1
figure 1

Oxytocin pathway gene expression in the human brain. Each point represents mean expression from six donors with standard errors for a given brain region for a OXTR and b CD38. The bolded dashed lines represent mean expression across the all regions with 1 standard deviation (+/−) also shown. Compared to average brain expression, there is increased expression of OXTR and CD38 in central and temporal brain structures, along with the olfactory region. Lower than average expression is observed in the cerebellum. *p < 0.05 (FDR corrected for 54 tests)

Full size image

Given the role of the hypothalamus in oxytocin signaling, expression patterns within hypothalamic substructures were also summarized using anatomic labels from the Allen dataset (Fig. 2; Supplementary Data 1). Oxytocin pathway gene expression in nearby thalamic substructures using anatomic labels from the Allen dataset were also summarized for comparison (see Supplementary Fig. 7 for anatomical map of substructures). There was significantly higher expression in the hypothalamic structures compared to the thalamic structures for OXTR (p < 0.001), CD38 (p = 0.002) and OXT (p < 0.001). Notably, there was significantly higher expression of OXT mRNA in the PVN, SON, preoptic region, and dorsomedial hypothalamic nucleus substructures compared to mean expression across the hypothalamus and thalamus (FDR corrected p < 0.001).

Fig. 2
figure 2

Expression of OXTR, CD38, and OXT in hypothalamic and thalamic substructures. Each point represents mean expression of a OXTR, b CD38, and c OXT with standard errors for a given brain region. The bolded dashed lines represent mean expression across the all regions with 1 standard deviation (+/−) also shown. Z-scores for d OXTR, e CD38, and f OXT expression are also presented in anatomic maps for hypothalamic and thalamic substructures (see Supplementary Fig. 7 for map legend). Substructures with expression more than 1 standard deviation greater than the mean are labeled. DMH: dorsomedial hypothalamic nucleus, LMH: lateral hypothalamic area (mammillary region), LT: lateral tuberal nucleus, Pa: paraventricular nucleus of the hypothalamus, PR: Preoptic region, SON: supraoptic nucleus, TM: tuberomammillary nucleus, VMH: ventromedial hypothalamic nucleus, AN: anterior group of nuclei, DLG: dorsal lateral geniculate nucleus, DL: lateral group of nuclei (dorsal division), VL: lateral group of nuclei (ventral division), MG: medial geniculate complex, MN: medial group of nuclei, PG: posterior group of nuclei, RT: reticular nucleus of thalamus. * p < 0.05 (FDR corrected for 16 tests)

Full size image

Voxel-by-voxel gene expression associations

Using the AAL atlas is a straightforward means to categorize and identify brain regions associated with increased gene expression. However, investigating specific brain regions of interest reduces spatial resolution. This is particularly an issue when correlating the expression patterns of two genes over a set of brain regions, as smaller anatomical regions may bias overall associations. Moreover, results may differ according to the selected atlas. Therefore, to increase precision, reduce the risk of bias, and examine associations independent of atlas, we used voxel-by-voxel mRNA maps for analysis (see Methods).

A Spearman’s correlation matrix demonstrating the relationships between genes of interest using the raw Allen data is presented in Fig. 3a. Hierarchical clustering revealed that oxytocin pathway genes (OXTR, CD38) cluster with some elements of the dopaminergic (DRD2, DRD5, and COMT) and cholinergic (CHRM4, CHRM5) systems, but not all, indicating the relationship between oxytocin pathway genes and dopaminergic/cholinergic gene sets were not globally strong. There were statistically significant relationships between the expression of the genes in this cluster (all FDR adjusted p-values < 0.05). The correlation of OXTR, CD38, and OXT within complete sets of dopaminergic, cholinergic, and oxytocinergic genes are visualized in Fig. 3b. For an independent sample validation of the clustering pattern of these 20 specified genes of interest, gene co-expression in the Allen dataset was compared with the independent GTEx dataset (see Methods). In both datasets, OXTR, CD38, DRD2, and CHRM4 clustered together (Supplementary Fig. 8). To explore whether these co-expression patterns are not driven by gene expression differences between cortical and non-cortical areas, we visualized the similarities in expression patterns based on cortical areas only for comparison against whole brain patterns (Fig. 3c). A comparison of expression patterns across all the brain regions with only cortical samples suggest that OXTR and CD38 co-expression patterns appear to be global, whereas OXT co-expression seems to be driven by correlations within cortical areas. The differential expression of the geneset cluster including OXTR and CD38 (Fig. 3a) in 30 tissue types across the body was assessed using FUMA, which extracted data from the GTEx database (version 7)30,31. Hypergeometric tests were used to assess statistically significant enrichment, revealing that the OXTR/CD38 geneset cluster was significantly enriched in brain tissue (Bonferroni corrected p-value < 0.05; Fig. 3d).

Fig. 3
figure 3

Co-expression of selected oxytocin, dopaminergic, muscarinic acetylcholine, vasopressin, and opioid gene sets in the brain. a Co-expression patterning for the expression of selected oxytocin, dopaminergic, muscarinic acetylcholine, vasopressin, and opioid gene sets. The complete linkage method was used to identify 3 clustering groups (black squares). OXTR and CD38 are clustered together, along with DRD2, DRD5, COMT, CHRM4, and CHRM5. b Whole-brain co-expression of OXTR, CD38, and OXT with selected vasopressin, muscarinic acetylcholine, and dopaminergic gene sets. Spearman’s correlations are visualized on a density distribution demonstrating all co-expression correlations between the of OXTR, CD38, and OXT and all genes in the complete oxytocinergic (n = 94), cholinergic (n = 79), and dopaminergic (n = 63) gene sets (see Supplementary Data 1 for full lists). c Cortex-only co-expression of OXTR, CD38, and OXT with selected vasopressin, muscarinic acetylcholine, and dopaminergic gene sets. d Differential expression of the OXTR/CD38 gene sets in 30 tissue types from the GTEx dataset, with -log 10 p-values representing the probability of the hypergeometric test. ***p < 0.001, **p < 0.01, *p < 0.05

Full size image

To uncover co-expression patterns of the full oxytocin signaling pathway (94 genes), we applied weighted gene co-expression network analysis using the WGCNA R package32. This analysis identified a co-expression module containing both OXTR and CD38, among 28 genes in total (Fig. 4a, Supplementary Data 1; Supplementary Fig. 9). Submission of the OXTR/CD38 module to FUMA revealed that this geneset module is enriched in GWAS catalog reported genes33 for waist-hip ratio adjusted for BMI and smoking (p = 1.13 × 10−2). This OXTR/CD38 module was also enriched in genesets reported in the Molecular Signatures Database34 that are associated with several behavioral and cognitive state processes including feeding behavior (p = 5.83 × 10−3), cognition (p = 7.44 × 10−3), and behavioral response to stimuli (p = 1.05 × 10−2). This OXTR/CD38 module was also enriched in brain and breast tissue (Bonferroni corrected p-value < 0.05; Fig. 4b). Altogether, these results suggest that the identified OXTR/CD38 module is biological meaningful. The association between OXTR and CD38 was statistically significant (p < 0.0001) and among the top 5% of correlations compared to all 20,737 protein coding genes. Density plots of the correlations between the three oxytocin pathway mRNA probes and all other available protein coding genes are presented in Supplementary Figure 10.

Fig. 4
figure 4

Co-expression patterns of the full oxytocin signaling pathway. a Weighted gene co-expression network analysis was used to construct a hierarchical clustering tree (also see Supplementary Fig. 9) and network heatmap plot of gene–gene connectivity of full oxytocin signaling pathway in the brain. Darker red colors represent stronger topological overlap. This analysis identified a module (dark blue) containing both OXTR and CD38 (see Supplementary Fig. 9 and Supplementary Data 1 for a full list of genes and modules). b Differential expression of the OXTR/CD38 module in 30 tissue types from the GTEx dataset, with -log 10 p-values representing the probability of the hypergeometric test. *p < 0.05

Full size image

Correlation analyses using Spearman’s coefficient with all available 20,737 protein coding gene probes in the Allen Human Brain Atlas and each of the three oxytocin pathway gene probes are summarized in Supplementary Table 1, with the top 10 strongest positive and top 10 strongest negative correlations (the full set of correlations with the remaining 20,736 genes are presented in Supplementary Data 2). The p-values for these correlations were all < 0.001. Among the top ten most positive mRNA map correlations for OXTR were Neurotensin Receptor 2 (NTSR2; rs = 0.78), Glutamate dehydrogenase 2 (GLUD2; rs = 0.78), and Glutamate dehydrogenase 1 (GLUD1; rs = 0.77), and the top ten most positive correlations with CD38 included NTSR2 (rs = 0.82), C12orf39 (Spexin; rs = 0.81), and GLUD1 (rs = 0.78). Associations with these specific genes are notable given their reported role in metabolic regulation35,36,37,38. Additionally, the gene with the 6th strongest association with CD38 was PSAT1 (rs = 0.79), whose disruption has been linked to schizophrenia39.

Donor-to-donor reproducibility of gene expression patterns

Given differences in sex and ethnicity among donors, we examined the similarity of gene expression patterns across the six brains in the sample by using the concept of differential stability, which is the average Spearman's correlation between any possible combination of the 15 pairs between the six donors. This approach has previously been applied to the same dataset40, with data suggesting that genes with strong differential stability are highly biologically relevant. We found that both OXTR and CD38 had differential stability values in the top decile of all 20,737 protein coding genes in the dataset (Fig. 5a), indicating that gene expression patterning is reproducible, regardless of individual differences, such as ethnicity and sex. We also calculated donor-to-donor associations (Spearman’s rank correlation coefficient) of gene expression patterns, revealing statistically significant associations (p < 0.0001) for OXTR, CD38, and OXT expression patterns between all donors (Fig. 5b–d). Inspection of gene expression data from the GTEx database also corroborates our differential stability data as this did not reveal sex differences in oxytocin pathway gene expression in the brain (Supplementary Fig. 11).

Fig. 5
figure 5

Differential stability of protein-coding genes. a Differential stability for all protein coding genes (n = 20,737) was calculated to assess the similarity of gene expression patterns from donor-to-donor. OXTR and CD38 were in the top decile of all genes (ranking in gray boxes), suggesting high reproducibility. Donor-to-donor differential stability was also assessed for b OXTR, c CD38, and d OXT. Each individual association for all three genes were statistically significant, suggesting that gene expression patterns were stable between each donor (ethnicity and sex are presented for each donor). *p < 0.001

Full size image

Cognitive state term correlates

The NeuroSynth framework27 has extracted text and neuroimaging data from 14,371 fMRI studies (release 0.7). While this tool can be used to create meta-analytic brain activation maps for specific cognitive states (e.g., stress) using forward inference, it can also be used to “decode” cognitive states, given an activation map, via reverse inference. Here, the term “cognitive state” refers to the neural processing that occurs response to a specific stimulus (e.g., pain) or that underlies a psychological process (e.g., learning).

To identify cognitive state correlates using brain maps of gene expression41 we performed quantitative reverse inference via large-scale meta-analysis of functional neuroimaging data using mRNA brain expression maps on voxel-by-voxel left hemisphere brain maps, representing the average of 6 individuals (see Methods). Using the NeuroSynth framework, we correlated our voxel-by-voxel mRNA maps (Fig. 6) with association Z maps, reflecting the posterior probability maps, which display the likelihood of a given cognitive state term being used in a study if activation is observed at a particular voxel. We assessed the 15 strongest relationships (Spearman’s r) between OXTR, CD38, and OXT expression maps (5 unique terms for each gene) and all available cognitive state maps available in the NeuroSynth database. To assess the comparative strength of these relationships compared to other genes, we also assessed the relationship between these 15 cognitive state maps and the 20,737 mRNA gene maps. We plotted these distributions and identified the rank of OXTR, CD38, and OXT Spearman’s r values compared to all other associations.

Fig. 6
figure 6

Voxel-by-voxel brain gene expression maps for OXTR, CD38, and OXT. mRNA expression maps were created using left hemisphere data from the Allen Human Brain Atlas dataset. These maps were submitted to the NeuroSynth framework to assess associations with cognitive state activation maps. SD: standard deviation

Full size image

Decoding cognitive states meta-analytically from voxel-by-voxel mRNA maps (Fig. 6) via quantitative reverse inference revealed that OXTR, CD38, and OXT mRNA expression maps were most highly correlated with functional imaging maps that can be broadly categorized as anticipatory, appetitive, and aversive (Fig. 7a, Supplementary Table 2). The “Emotional” and “Facial expression” cognitive states were considered both aversive and appetitive given that emotions and facial expressions can either be positive or negative and that intranasal oxytocin has been shown to modulate the processing of both positive and negative emotions42. Figure 7b shows the full distribution of Spearman’s correlation coefficients for each cognitive state term across all 20,737 gene maps, with labeled oxytocin pathway genes and their absolute rank for each term (Supplementary Table 2). Notably, the OXTR map’s relationship with the cognitive state maps of “sexual”, “motivation”, “incentive”, and “anxiety” were all ranked among the top 0.5% strongest associations out of 20,737 genes and were statistically significant (p < 0.001; Fig. 7b; Supplementary Table 2), The “taste”, “stress”, “reward”, “monetary”, “fear”, and “emotional” cognitive state maps were within the top 2.5% of all associations with OXTR expression and also statistically significant (p < 0.001; Fig. 7b; Supplementary Table 2). In other words, not only were these cognitive state maps the most highly correlated with the OXTR expression map, but these correlations with specific cognitive states were amongst the strongest correlations across all 20,737 protein coding genes. All correlation coefficient p-values for the relationship between each cognitive state term and oxytocin pathway genes (OXTR, CD38, and OXT) are presented in Supplementary Table 2. Additionally, OXTR (p = 0.0002; Fig. 7c) and CD38 (p = 0.006; Fig. 7d), had significantly greater mRNA expression in brain regions associated with social behavior compared to brain regions associated with non-social cognitive states (Supplementary Data 1). There was no significant difference in OXT mRNA expression (p = 0.14; Fig. 7e) between these social and non-social brain regions.

Fig. 7
figure 7

Cognitive state correlates of central oxytocin pathway gene expression patterns. a Cognitive states were meta-analytically decoded from central oxytocin pathway mRNA maps (Fig. 6) using the NeuroSynth framework. The top five strongest relationships for OXTR, CD38, and OXT (Spearman’s r) are shown, with duplicates removed. b The distribution of Spearman’s correlations between each protein coding gene map (n = 20,737) and cognitive state maps. The absolute ranking for each oxytocin pathway gene out of 20,737 correlations are shown (also see Supplementary Table 2). mRNA expression of c OXTR, d CD38, and e OXT in brain regions associated with social and non-social mental states (see Supplementary Data 1 for lists of social and non-social terms)

Full size image

Discussion

The anatomical distribution of gene expression in the brain is heterogeneous and highly coordinated40. Dynamic alterations in gene expression are essential in response to environmental demands, and are critically involved in a range of cognitive states, learning processes, and diseases43,44, while the basic organization likely partly reflects the evolutionary conserved modular layout of the brain45. Gene-gene co-expression patterns form specific genetic signatures in the brain, representing distinct tissues and biological systems40, and likely reflect the potential of complex and differential gene-gene interactions with implications for brain disorders and cognitive health. Here, we leveraged the unique human brain mRNA expression library from Allen Brain Atlas to show that mRNA reflecting specific genes in the oxytocin pathway are highly expressed in subcortical and temporal brain structures, along with the olfactory region. We also show reduced expression of OXTR and CD38 in the cerebellum, consistent with prior animal research46. Importantly, an independent sample comparison revealed that the observed OXTR and CD38 expression patterns from the Allen database were consistent with OXTR and CD38 expression patterns observed in the GTEx database. Expression patterns of OXT in the Allen database were less stable from donor-to-donor and were not significantly related to the expression patterns in the GTEx database, which limits the generalizability of the results relating to this specific gene. OXT expression did not demonstrate strong whole-brain correspondence with OXTR and CD38 expression, which was likely due to OXT transcripts only showing a strong presence in oxytocin producing cells within hypothalamic regions.

OXTR and CD38 showed considerable co-expression with DRD2, DRD5, COMT, CHRM4, and CHRM5 genes, providing evidence for putative interactions between dopaminergic and muscarinic acetylcholine systems with oxytocin pathway signaling. Exploratory analysis between oxytocin pathway mRNA and 20,737 mRNA probes revealed several relationships worth noting in the context of metabolic and feeding regulation, as well as psychiatric disorders. We discovered that OXTR is highly co-expressed with NTSR2, which has been found to regulate ethanol consumption in mice35 and GLUD1 and GLUD2, which are involved in energy homeostasis36,37. CD38 is highly co-expressed with NTSR2 and C12orf39 (Spexin), with the latter associated with weight regulation38. Moreover, weighted gene co-expression network analysis identified a co-expression module containing both OXTR and CD38, which was enriched in a waist-hip ratio GWAS33 and a gene set associated with feeding behavior34. OXTR mRNA expression was also strongly associated with the neural processing of taste stimuli. Altogether, these results are consistent with emerging evidence that the oxytocin system plays a role in the metabolic and feeding dysregulation, which are overrepresented in severe mental illnesses5. Notably, research indicates that intranasal oxytocin administration improves appetite regulation in humans47,48.

In regards to psychiatric disorders, there were strong negative correlations between oxytocin pathway genes and PAK7 and GABRD which have been associated with schizophrenia49,50,51,52. CD38 was highly correlated with PSAT1, a gene that has been associated with schizophrenia39. While these studies suggest intriguing links between specific genes and psychiatric illnesses, it is important to note that these are largely single report associations that require replication in independent samples. Quantitative reverse inference via meta-analysis of 14,371 fMRI studies revealed that the distribution of the oxytocin signaling pathway genes is most strongly associated with anticipatory, appetitive, and aversive cognitive states (Supplementary Table 2). Notably, out of the 20,737 gene expression brain maps that we compiled, the OXTR expression map was among the top 0.5% of strongest relationships with the “sexual”, “motivation”, “incentive”, and “anxiety”, cognitive state maps. Put another way, these cognitive state maps had a stronger correlation with OXTR expression patterns than at least 99.5% of all protein coding genes in the brain. These results are also consistent with a growing body of research indicating that intranasal oxytocin administration modulates emotion processing2,3,53,54, suggesting that intranasally administered oxytocin may exert its effects by binding to oxytocin receptors located in regions associated with emotion processing. Indeed, compelling rodent evidence suggests that direct oxytocin central administration influences social behavior and cognition via action on oxytocin receptors55. In these animals, oxytocin receptors are located in regions that are crucial for social behavior and the processing of social cues11.

Data from both our co-expression and meta-analytic analyses provides converging evidence that oxytocin may exert its effects synergistically with the dopaminergic network17. The oxytocin and dopaminergic systems have been shown to work in concert to promote rodent maternal behaviors56, and central D2 receptors modulate social functioning in prairie voles57. There is also evidence of direct dopaminergic fibers to the supraoptic nucleus and the paraventricular nucleus58, consistent with demonstrations that dopamine may stimulate oxytocin release in the hypothalamus59. This is also congruous with reported deficits in dopaminergic signaling in schizophrenia60 and autism61. The muscarinic acetylcholine system may also work with the oxytocin system to boost attention62 and to assist with the release of oxytocin63. Relatedly, the muscarinic acetylcholine system is thought to contribute to the cognitive dysfunction observed in schizophrenia64. Altogether, our findings corroborate emerging evidence for interactions between the oxytocin, acetylcholine12, and dopaminergic signaling systems17.

Our data revealed high expression of oxytocin pathway mRNA in the olfactory region, which is in line with animal research65. Indeed, social olfactory deficits in mice without the oxytocin gene66 are rescued with injection of oxytocin into rat olfactory bulbs67. Increased mRNA in the human olfactory region may be vestigial, as olfaction is not as important for human conspecific identification compared to most other mammals due to species specialization, however, intranasal oxytocin has been shown to improve olfactory detection performance in schizophrenia68, and intranasal oxytocin is thought to enter the brain via the olfactory bulbs69.

There are some important limitations worth noting. First, given the nature of human brain tissue collection for mRNA studies, the sample size was small and the donor group was variable in regards to age and ethnicity. These issues, along with individual differences in brain sizes and morphological features, may have introduced variability in sampling. However, there are several reasons to believe that donor-to-donor variability did not influence the outcomes of the study. The relatively small amount of measurement error for the mRNA probes is indicative of high precision and sufficient generalizability. We also independently validated general gene expression patterning in the larger GTEx dataset, however, it is worth noting that this comparison is somewhat limited due to fewer brain tissue sampling sites in the GTEx dataset compared to the Allen dataset. While there was only 1 female in the sample, which prohibited a comprehensive examination of sex differences in mRNA distribution from the Allen dataset, whole-brain expression patterns of OXTR, CD38, and OXT from the female donor was significantly related to the expression patterns of all the other male donors. Moreover, data from the GTEx dataset demonstrated little evidence of sex differences in central oxytocin pathway gene expression, consistent with preliminary human research12. However, it could be possible that sex differences in oxytocin pathway gene expression exist in brain regions not sampled in the GTEx dataset. Finally, we demonstrated that OXTR and CD38 demonstrate high differential stability, which is indicative of highly consistent patterning across donor brains40. Indeed, high differential stability appears to be a hallmark of genes with considerable biological relevance40. Second, despite animal models demonstrating that OXTR expression levels directly modulate social behavior9,10 and that increased OXTR gene expression is associated with increased oxytocin binding, research is yet to demonstrate whether the increased expression of OXTR (as measured by the presence of its transcript) in humans is associated with the activation of the receptor during specific cognitive states.

To translate oxytocin to the clinic, research needs to demonstrate engagement of drug targets, such as OXTR occupancy reflected by regional brain activity changes70. Without precise targets, it is unclear whether non-significant effects of intranasal oxytocin, beyond insufficient statistical power71, are due to an inefficacious drug or misidentified drug targets. By identifying precise oxytocin pathway targets in the human brain and the cognitive state correlates of these oxytocin pathway gene distribution, our study may help facilitate efforts to better understand oxytocin treatment efficacy as it identifies targets for oxytocin receptor engagement, which can facilitate dose-ranging studies69,70. Analysis of data from the Allen Brain Atlas database provides a unique opportunity to map the central oxytocin system and explore co-expression with other systems. Altogether, these results provide a proof-of-principle demonstration of corresponding cognitive and gene expression patterns of a neuropeptide pathway involved in complex human behaviors.

Methods

Post-mortem brain samples

mRNA distribution data was collected from the Allen Human Brain Atlas (http://human.brain-map.org/). Three donors were Caucasian males, one donor was a Hispanic female, and two donors were African-American males. Mean donor age was 42.5 (S.D. = 11.2) years. Data was collected 22.3 (S.D. = 4.5) hours after demise, on average (See Supplementary Table 3 for detailed donor profiles). If more than one probe for each mRNA was available, we selected the probe with the highest differential stability40 which represented the probe with least amount of spatial variability among donors. The collection of human data complied with relevant ethical regulations, with institutional review board approval obtained at each tissue bank and repository that provided tissue samples. Moreover, informed consent was provided by each donor’s next-of-kin. For more details regarding procedures and data collection associated with the Allen Human Brain atlas, see http://help.brain-map.org/display/humanbrain/Documentation.

Independent sample validation

Oxytocin pathway median expression values (Log10 transcripts per million) for 10 distinct brain regions (amygdala, anterior cingulate cortex, caudate, cerebellar hemisphere, cerebellum, cortex, frontal cortex, hippocampus, nucleus accumbens, putamen, substantia nigra) were extracted from the Genotype-Tissue Expression (GTEx) project database30 for independent sample validation. Although this dataset offers less spatial resolution compared to the Allen dataset, the data is derived from a larger dataset of donors (mean sample size for mRNA expression across brain regions = 131.7, range = 88 to 173). For the comparison between the Allen and GTEx datasets, median gene expression for these 10 distinct GTEx regions were calculated from the Allen dataset (Supplementary Fig. 1), as the GTEx platform reports median values. Spearman’s rank correlation coefficients (rs) were calculated to assess the relationship between oxytocin pathway expression profiles from the two databases and overall co-expression patterns. Complete linkage clustering72 was used to identify three clustering groups in both datasets.

Voxel-by-voxel gene expression maps

To create novel voxel-by-voxel volumetric expression maps, we first marked all the sample locations and expression values in native image space. To interpolate missing voxels, we labeled brain borders with the sample expression value that had the closest distance to a given border point (Supplementary Fig. 12). Next, we divided the space between scattered points into simplices based on Delaunay triangulation, then linearly interpolated each simplex with values to yield a completed map. All maps were computed in Matlab 2014a (The Mathworks Inc., Natick, MA, USA). We then created a composite brain map representing an average of 6 individuals on the left hemisphere, registered brains to MNI space using ANT’s non-linear registration and averaged so that each gene’s mRNA is represented by a single voxel-by-voxel brain map.

Calculation of differential stability

The average correlation (Spearman’s r) across 15 pairs of 6 donors’ voxel-by-voxel brain maps were used to calculate differential stability using an approach described by Hawrylytz and colleagues40. Since each voxel-by-voxel map was generated based on a limited and variable number of samples, to calculate the statistical significance of Spearman’s coefficients we calculated p-values between two donors based on the smallest number of samples out of the two. That is, if one donor had samples from 353 locations and another one from 456, we would use the 353 samples to calculate a p-value for the pair.

Statistical analysis of gene expression data

The R statistical package (version 3.3.2) was used for statistical analysis. One-sample t-tests (two-tailed) were conducted to assess which of the 54 left hemisphere regions from six donor samples expressed mRNA to a significantly greater or lesser degree compared to average mRNA expression across the brain (for raw data, see Supplementary Data 1). To correct for multiple tests (54 in total), reported p-values were adjusted using a false discovery rate (FDR) threshold. Cohen’s d values for one-sample t-tests were calculated to yield a measure of effect size (Supplementary Data 1).

Based on raw expression data, we generated a 20 × 20 correlation matrix reflecting the spatial Spearman’s correlation for each donor for each mRNA map pair including the following selected genes (Oxytocin pathway set: OXTR, CD38, OXT; Dopaminergic set: DRD1, DRD2, DRD3, DRD4, DRD5, COMT, and DAT1; muscarinic acetylcholine set: CRHM1, CRHM2, CRHM3, CRHM4, and CRHM5; vasopressin set: AVPR1A, AVPR1B; opioid set: OPRM1, OPRD1, OPRK1). Next, we computed the average correlation in each cell of the correlation matrix across the 6 donors and clustered the mRNA maps using the complete linkage method72 to assess co-expression patterns. To assess putative gene–gene interactions beyond selected gene sets, we computed the spatial Spearman’s correlations between all available protein coding mRNA probes (n = 20,737) and three oxytocin pathway mRNA probes. As more than one mRNA probe may have been available for a single gene, we selected the probe with the highest differential stability value. Unlike Pearson’s r, which is the correlate statistic presented on the Allen Human Brain Atlas web interface, Spearman’s correlation coefficient is less sensitive to outliers and non-normally distributed data.

To assess gene expression across various body tissues, normalized gene expression values (reads per kilo base per million; RPKM) were extracted from the GTEx database30, via the FUMA platform31. As described by Watanabe and colleagues31, normalized expression [zero mean of log2(RPKM + 1)] was used to assess differentially expressed gene sets. Bonferroni adjusted p-values are then calculated using two-sided t-tests per gene per tissue against all other tissues. Genes with a Bonferroni adjusted p-value < 0.05 and absolute log fold change ≥0.58 were categorized as a differentially expressed gene set in a given tissue type. The presented -log10 p-values represent results from hypergeometric tests, which were used to assess if genes of interest were overrepresented in differentially expressed gene sets in specific tissues. Similarly, hypergeometric tests were used to assess if genes of interest were overrepresented in gene sets reported in the GWAS catalog33 and gene sets associated with behavioral and cognitive state processes reported within GO biological processes gene sets within Molecular Signatures Database34, using 20,119 protein coding genes as the background set. P-values were Benjamini-Hochberg adjusted for all genesets reported in the GO biological processes dataset and GWAS catalog, respectively.

Weighted gene co-expression network analysis

The WGCNA R package32 was used to identify gene modules for expression of the oxytocin signaling pathway in the brain. A measure of topological overlap, which reflects biological meaningful models73, was used to identify gene modules with tight interconnections. The WGCNA package uses average linked hierarchical clustering to group genes into modules based on their topological overlap. Soft thresholding power was selected based on a scale-free topology criterion. We selected power 10 (scale free topology R2 = 0.828) as this was the lowest point for which the scale-free topology fit index curve flattens out (Supplementary Table 4).

Cognitive state correlates using the NeuroSynth platform

We correlated a voxel-by-voxel mRNA expression maps with NeuroSynth (version 0.3.7), a Python package for large-scale synthesis of functional neuroimaging data27 to extract the top five cognitive correlates for each of three genes (OXT, OXTR, and CD38). We modified the NeuroSynth package to calculate Spearman’s correlation coefficient instead of the default Pearson’s correlation coefficient.

Next, to test specificity of these cognitive states (e.g. “Sexual”, “Motivation”), we extracted association Z maps, which reflect Z-scores of the association between the presence of activation and the presence of a feature. After extracting these maps, we calculated Spearman’s correlation between all 20,737 genes and association Z score maps and ranked them from largest to smallest.

A list of social and non-social cognitive states (Supplementary Data 1) was selected by two of the co-authors by consensus. Ambiguous cognitive states that could be easily classified as social or non-social (e.g., “Family”) were excluded. Next, we extracted association Z scores and FDR-corrected p-value maps for each cognitive state. We then thresholded each association map at FDR p < 0.05 and calculated the median association Z score for each statistically significant region, which resulted in an expression score for each cognitive state term. Finally, we calculated p-values by performing 5000 permutations for the difference in mean expression between social and non-social regions (Supplementary Data 1) via t-tests using the Permutation Analysis of Linear Models (PALM) tool74.