Cross‐sex shifts in two brain imaging phenotypes and their relation to polygenic scores for same‐sex sexual behavior: A study of 18,645 individuals from the UK Biobank. Christoph Abé Alexander Lebedev Ruyue Zhang Lina Jonsson Sarah E. Bergen Martin Ingvar Mikael Landén Qazi Rahman. Human Brain Mapping, February 26 2021. https://doi.org/10.1002/hbm.25370
Abstract: Genetic and hormonal factors have been suggested to influence human sexual orientation. Previous studied proposed brain differences related to sexual orientation and that these follow cross‐sex shifted patterns. However, the neurobiological correlates of sexual orientation and how genetic factors relate to brain structural variation remains largely unexplored. Using the largest neuroimaging‐genetics dataset available on same‐sex sexual behavior (SSB) (n = 18,645), we employed a data‐driven multivariate classification algorithm (PLS) on magnetic resonance imaging data from two imaging modalities to extract brain covariance patterns related to sex. Through analyses of latent variables, we tested for SSB‐related cross‐sex shifts in such patterns. Using genotype data, polygenic scores reflecting the genetic predisposition for SSB were computed and tested for associations with neuroimaging outcomes. Patterns important for classifying between males and females were less pronounced in non‐heterosexuals. Predominantly in non‐heterosexual females, multivariate brain patterns as represented by latent variables were shifted toward the opposite sex. Complementary univariate analyses revealed region specific SSB‐related differences in both males and females. Polygenic scores for SSB were associated with volume of lateral occipital and temporo‐occipital cortices. The present large‐scale study demonstrates multivariate neuroanatomical correlates of SSB, and tentatively suggests that genetic factors related to SSB may contribute to structural variation in certain brain structures. These findings support a neurobiological basis to the differences in human sexuality.
4 DISCUSSION
In this large‐scale study on SSB, we used brain imaging phenotypes from two imaging modalities and a multivariate classification algorithm to extract independent brain covariance patterns related to sex. We then tested for SSB related cross‐sex shifts in such patterns. For the first time, we also examined whether polygenic scores for SSB relate to brain imaging phenotypes.
Our results showed that the PLS classifier was effective in classifying males and females, and that patterns important for classification were less pronounced in non‐heterosexual individuals, indicative of a cross‐sex shift. The analysis of LVs demonstrated that one (LV1) displayed a sex‐by‐SSB interaction. This interaction remained following adjustment for potential confounding variables, including psychiatric diagnoses and victimization experiences, and was driven by the fact that nHeF showed larger LV1 scores than HeF. Since males showed the largest LV scores, this indicates an SSB‐related cross‐sex shift in multivariate brain patterns predominantly in females. This shift in LV1 was not observed in males, which could potentially arise because SSB‐related differences in males might have less of a covarying nature, regionally differ, be more focal, or less pronounced (smaller effect size) compared to females, as indicated by secondary univariate analyses (Figure 6). However, these differences could also be explained by the fact that the SSB measure does not capture all aspects of sexual orientation. While SSB correlates highly with other components of sexual orientation, nHeF and nHeM in our sample may differ in other components such as sexual attractions, sexual identity labels, or romantic attractions (J. M. Bailey et al., 2016). Hence, we cannot exclude the presence of sub‐groups among non‐heterosexual individuals. In line with that notion, in an explorative analysis excluding individuals with one or two reported lifetime same‐sex partners (see Supporting Information), the peak of the LV1 distribution in nHeM was shifted toward smaller values (the mean of females), indicating that a sub‐group of nHeM (e.g., those with more same‐sex partners) may show a more female‐like multivariate brain pattern. However, this effect requires further investigation. Nevertheless, our findings suggest sexuality‐related variation in multivariate brain data, supporting the utility of data‐driven classification and that multivariate pattern analyses are effective at identifying such associations on group level, at least in females.
Our neuroanatomical findings support a number of previous small‐scale reports of sexual orientation‐related differences (Abé et al., 2014; Abé et al., 2018; Manzouri & Savic, 2018a, 2018b; Ponseti et al., 2007; Savic & Lindström, 2008) in that they indicate SSB‐related cross‐sex shifts in brain imaging phenotypes. Intriguingly, the calcarine sulcus (part of the visual cortex) appears to be the most consistently reported structure showing sexual orientation‐related differences (Abé et al., 2014; Abé et al., 2018; Manzouri & Savic, 2018b), which is consistent with results from our secondary univariate analyses (ROI approach: Figure 6, and whole brain analysis: Data S2). We did not replicate sexual orientation differences in the anterior cingulate cortex (Manzouri & Savic, 2018a, 2018b) and hippocampus (Abé et al., 2014) in males. Cross‐sex shifts in brain data are also consistent with a large body of empirical findings demonstrating cross‐sex shifted patterns of gender‐related behavior, cognitive ability (in tasks that typically differ between the sexes), and certain personality traits (Allen & Robson, 2020; Bailey et al., 2016; Li et al., 2017; Rieger et al., 2008; Xu et al., 2017). However, there is considerable overlap in the distribution of LV‐scores between the groups, and the magnitude of the effects for SSB‐related brain differences seem smaller than those reported for the aforementioned behavioral traits. Notably, effect sizes for SSB‐related differences in cortical volumes were also smaller than those of sex differences (Table S3, Data S2).
The imaging variables that loaded most strongly on LV1 (displaying the sex‐by‐SSB interaction) were measures of regional volumes in prefrontal, parietal, and occipital (including visual) cortices. In the context of SSB, the visual cortex is involved in visual perception and processing of sexual stimuli (Georgiadis & Kringelbach, 2012). Prefrontal areas are involved in the integration of sensory information and reward‐value representation of sexual stimuli (Georgiadis & Kringelbach, 2012). Together with the precuneus, involved in self‐referential processes (Cavanna & Trimble, 2006), these areas are also recruited during visuo‐spatial processing and selective visual attention (Cavanna & Trimble, 2006; Georgiadis & Kringelbach, 2012; Paneri & Gregoriou, 2017; Posner & Gilbert, 1999). However, this study does not allow conclusions about causality or the brain regions' functional involvement. It requires further testing how differences in brain structure relate to SSB. Note that although volumes of those brain regions that tended to successfully predict group membership largely overlap with those previously reported in other studies, in contrast to direct group comparisons in univariate analyses, PLS results should not necessarily be interpreted as evidence of structural differences between the groups, but rather as generalized covariance patterns in the brain data that discriminate between them. Another important finding is that while LV1 appeared to capture the hypothesized cross‐sex shift, LV2 appeared to capture a main effect of SSB. This may indicate that SSB‐related multivariate brain patterns may exist that do not follow a cross‐sex shift and are similar in both nHeM and nHeF (regardless of sex). It is also noteworthy that cortical volumetric measures showed the highest loadings, whereas those of subcortical structures and DTI‐based FA values were close to zero, indicating that sex‐related brain phenotype variation may be more pronounced in gray matter than white matter or subcortical measures.
The causes of sexual orientation‐related differences in brain structure are as yet unknown. Both genetic and non‐genetic factors have been proposed to play a role, with the most prominent hypothesis involving prenatal androgen influences (Bailey et al., 2016; Kevin, Khytam, & David, 2018). Genetic influences are modest based on existing twin models and molecular genetic studies (Bailey et al., 2000; Bailey et al., 2016; Ganna & Verweij, 2019; Langstrom et al., 2010) and are almost certainly polygenic in nature (Ganna & Verweij, 2019). Here, we investigated genetic influences on brain phenotypes by testing the associations between polygenic scores for SSB (PS‐SSB) and brain imaging phenotypes. Whereas PS‐SSB did not seem to predict multivariate brain patterns (LVs), we found that PS‐SSB was associated with cortical volumes in individual brain regions. These associations were observed mainly in lateral occipital and temporo‐occipital cortex. In lateral occipital cortex, higher PS‐SSB was associated with lower volumes in both males and females. In temporo‐occipital cortex, higher PS‐SSB was associated with lower cortical volumes in nHeM and larger volumes nHeF. These findings tentatively indicate that genetic factors related to SSB are associated with variation in some cortical structures and that a higher genetic predisposition to SSB has the opposite effect on cortical volume in males and females who reported SSB. These associations were small and PS‐SSB explained little of the variance in brain structure. Notably, we did not find significant genetic correlations in complementary analyses linking previously published SSB and brain phenotype GWASs (Elliott et al., 2018; Ganna & Verweij, 2019). Therefore, these genetic associations should be treated with caution, and additional factors are likely to explain brain variation associated with human sexuality. Mechanisms responsible for how genetic factors influence brain structure, function, and in turn behavior are complex and multi‐factorial. Given the general limitations of the applied methodology (see below), these cannot be derived from this study. We also want to note that, given the wide and overlapping range of LVs and PS‐SSB, as well as the weak classification performance when solely predicting SSB (AUC = 0.57), the present results cannot be used to predict an individual's sexual orientation based on genetic or neuroimaging data.