Introduction
Most neurodevelopmental disorders (NDDs) present more often in males than females [
1], and autism spectrum disorder (ASD) in particular shows a striking sex bias with a 4:1 male-to-female ratio. However, despite the evidence that male sex—and masculinity more generally, as posited by the “extreme male brain” theory [
2]—is a strong risk factor for ASD, the cause of this sex bias is not yet well understood. The “extreme male brain” theory suggests that the male sex bias may reflect the role of male sex hormones (i.e., androgens) in the etiology of ASD; specifically, risk for ASD may stem from increased androgen (e.g., testosterone) exposure during critical periods of fetal development, as the brain undergoes substantial organization and sexual differentiation [
3,
4]. Indeed, numerous studies have found that greater prenatal androgen exposure is associated with ASD and ASD-related traits [
5‐
8], although other studies have observed no relationship [
9,
10]. Postnatal studies utilizing clinical and community samples of adults have also identified links between testosterone levels and ASD-related traits [
11,
12], although postnatal studies of children show mixed results [
13,
14]. Interestingly, despite evidence from many of these studies that greater testosterone is associated with phenotypes that characterize neurodevelopmental disorders more broadly—for instance, poor language abilities [
15,
16]—the majority of studies that examine androgen exposure focus on ASD. Given the mixed evidence and dearth of studies examining NDDs outside of ASD, a better understanding of whether and how testosterone exposure may contribute to the male bias observed across all NDDs is warranted [
8].
The links between androgen exposure and neurodevelopment have primarily been established via cross-sectional studies that measure testosterone levels at a single time point during perinatal (e.g., amniotic fluid, umbilical cord blood) or postnatal (e.g., blood serum, saliva) development. Beyond the practical challenges of measuring testosterone via biological samples, particularly during the perinatal period, these measures provide only a snapshot of androgen exposure at the time of collection, rather than capturing an individual’s cumulative exposure across development. To overcome the limitations of measuring testosterone levels directly, morphological and genetic measures are increasingly being leveraged to indirectly assess androgen-related risk. Digit ratio, or the ratio in length of the second and fourth digit (i.e., 2D:4D ratio), is a sexually dimorphic feature that is lower in males than in females [
17] and has been linked to fetal testosterone exposure [
18]. Numerous studies indicate that individuals with ASD exhibit lower (i.e., more masculine) digit ratios than their typically developing counterparts [
19‐
23], and evidence suggests this may be true for other NDDs as well [
24‐
26]. However, a number of recent studies have failed to replicate the association between fetal testosterone exposure and digit ratio [
27,
28], indicating that digit ratio may be a useful biomarker of NDDs, but its biological underpinnings require further clarification.
Facial masculinity has also been proposed as a potential indicator of fetal testosterone and NDD-related risk. A twin study found that females with a male twin, who were presumably exposed to greater fetal testosterone levels, exhibited greater facial masculinity than those with a female twin [
29] and another study showed that greater umbilical cord testosterone levels were associated with greater adult facial masculinity in both sexes [
30]. Moreover, recent studies have found that facial masculinity predicts ASD traits in a variety of populations including children with ASD [
31], siblings of individuals with ASD [
32], as well as non-clinical samples [
33,
34]. However, these studies provide inconsistent support for the “extreme male brain” theory [
35,
36]. While some studies found that greater masculinity was associated with greater ASD traits in both sexes, others suggested that androgynous features may confer greater risk. For example, one study found that female subjects with ASD demonstrated more masculine faces than their typically developing peers, but males with ASD exhibited
less masculine faces than their peers [
36]. Given that these early findings are mixed, and that a dearth of studies has examined facial masculinity in the context of other NDDs, it remains unclear whether facial masculinity can yield insight into NDDs and, if so, how it may reflect androgen-related risk.
Running parallel to morphological proxies of androgen exposure, genetic studies of NDDs can yield insight into the hypothesized role of androgens in the etiology of NDDs. While large-scale studies have so far not identified individual genomic loci that strongly implicate sex hormone pathways as a major risk factor, some smaller studies have yielded clues worthy of further investigation. For example, one study (
n = 174) found that three genes (the estrogen receptor
ESR2 and two steroid enzymes,
CYP11B1 and
CYP17A1) related to the synthesis, transport, or metabolism of sex hormones were associated with ASD-related traits in non-clinical samples [
37]. Another study found that
CYP19A1, a steroid enzyme that contributes to sexual differentiation of the brain, was associated with dyslexia, a form of language impairment [
38]. Yet another (
n = 1171) identified three genes—that encoded an estrogen receptor (
ESR1), steroid enzyme (
SRD5A2), and sex-hormone binding globulin (
SHBG)—as conferring risk for ASD-related traits in males [
39], although replication in a larger sample (
n = 10,654) found that only variation in
SHBG emerged as a significant predictor [
40]. Interestingly, levels of SHBG are inversely associated with levels of active (i.e., “free”) testosterone. Testosterone becomes inactive when it is bound to SHBG, and studies show that low levels of SHBG are found in individuals with excessive androgen activity [
41]. These findings suggest that SHBG and other hormone-binding proteins may play an important role in the etiology of ASD through regulation of testosterone exposure. Moreover, given evidence that SHBG and free testosterone levels have been linked to attention-deficit/hyperactivity disorder (ADHD) [
42] and that genome-wide studies have identified significant genetic overlap between ASD and other NDDs [
43,
44], it is possible that variation in SHBG and other testosterone regulators may contribute to NDDs more broadly. Further investigation of these potential mechanisms is needed.
Together, these previous studies point to several important yet unanswered questions. First, which morphological proxies of androgen exposure are most predictive of NDD diagnosis? Second, what specific behavioral and cognitive phenotypes are most related to these proxies of androgen exposure? Third, do morphological proxies such as digit ratio and facial morphology capture the same or different molecular aspects of masculinization, and how do these mechanisms relate to neurodevelopmental risk? Finally, how do quantitative effects related to androgen exposure in both sexes compare to the binary sex effect of being a Y-chromosomal male? The answers to these questions will deliver vital insight into the nature of male bias in NDDs. Toward this end, we carried out an investigation in two steps (Fig.
1). First, we assembled a neurodevelopmental cohort (
N = 763) that includes (1) genome-wide genotypes, (2) measures of digit ratio masculinity (DRM), (3) measures of facial landmark masculinity (FLM) derived by machine learning, and (4) extensive clinical and parent-reported phenotypic information. This is the first study to investigate genetic factors associated with both digital and facial masculinity and their connection to neurodevelopmental outcomes. Second, to ensure the robustness of findings generated by investigation of our discovery cohort, we employed SPARK, a large genetic study of autism (
N = 9419 in this analysis), as a means for testing the generalization of our observations.
Discussion
The current study employed a two-step approach to investigate genetic factors associated with both digital and facial masculinity and their connection to neurodevelopmental outcomes. In the first study, which was the first of its kind to examine both digit ratio and facial masculinity in the same genetically characterized neurodevelopmental cohort, we found convergent evidence that androgen exposure is associated with deficits in social functioning, even while correcting for binary sex. We found that facial landmark masculinity (FLM) was predictive of NDD diagnoses (ADHD, ASD, ID), in the direction consistent with the extreme male brain theory, while digit ratio masculinity (DRM) was not. The predictive power of FLM relative to DRM suggests that further pursuit of biomarkers and endophenotypes based on facial morphology may prove more fruitful than digit ratio. Particularly noteworthy is that social functioning showed convergent association with both FLM and DRM. Through analysis in connection with polygenic risk scores (PRS), our findings implicate both testosterone and sex hormone binding globulin (SHBG) levels as complementary and potentially additive factors that may impact social functioning. In the second study, our hypothesis found additional statistical support in the large SPARK cohort, and we found evidence suggesting that the autosomal genetic factors (i.e., the PRS) that predict testosterone and SHBG levels may exert effects on social functioning that are comparable in magnitude to the effect of binary (i.e., chromosomal) sex itself. This provides evidence that, at least in the context of its effect on social functioning, sex may be more accurately described as bimodal and continuous, rather than binary and discrete.
Notably, the direction of associations between FLM and NDD diagnosis was consistent with the extreme male brain theory (i.e., greater masculinity conferred greater risk for NDD traits), which was not the case for DRM. Although DRM was associated with social deficits in the expected direction, when considering clinical diagnoses, young- to middle-adulthood males with NDDs demonstrated more
feminine digit ratios than their typically developing peers. This is not the first study to find that androgynous features are associated with NDDs [
33] and that males with ASD demonstrate more feminine digit ratios [
36]. However, quantitative reviews have supported the association between masculine digit ratio and ASD [
71,
72], so the literature on DRM remains mixed. It is also unclear why the association between DRM and affected status did not emerge until young-adulthood, given that NDD-related outcomes emerge in early childhood. Longitudinal studies of typically developing children show that digit ratio tends to increase (i.e., becomes more feminine) across puberty [
73,
74], however the rank order of inter-individual differences remain stable. No study to date has examined the developmental trajectory of DRM in individuals with NDDs, or longitudinally compared groups of individuals with and without NDDs, so further research in this area may be useful. Nevertheless, our results suggest that FLM may provide greater utility than DRM as a biomarker with clinical relevance.
The differential performance between FLM and DRM in predicting NDD traits may reflect a number of physiological factors that differentiate the two measures of masculinity. First, combining distances between 12 facial landmarks naturally provides a richer basis for an estimator of masculinity than the average length of two digits. As such, the face inherently provides a more sensitive indicator for subtle differences in biological risk. In contrast to digits, faces are more proximal to the brain not just physically, but in development and biology. Prenatal brain development is intimately tied to craniofacial development through a combination of biochemical mechanisms and genetic signaling [
29]. Brain and craniofacial tissues originate from the same population of cells, such as neural crest cells, which differentiate early in fetal development to eventually form the brain and face [
75]. As such, developmental perturbations in the brain are more likely to manifest in craniofacial features than digits [
76‐
78]. These developmental mechanisms support our findings that FLM is associated with NDDs broadly, and not only to ASD. Given the ties between neurodevelopment and craniofacial development, it follows that facial morphology is a useful basis for predicting neurodevelopmental risk generally, rather than ASD in particular. This was underscored by the convergence of DRM and FLM on a factor that loads on social functioning. That this convergence on social impairment was found in a mixed NDD cohort is noteworthy: although ASD is most prominently characterized by deficits in social functioning, other NDDs are also marked by elevated rates of social dysfunction [
79,
80], underscoring the fact that key features of NDDs, and psychopathology more broadly, do not adhere to diagnostic boundaries [
43,
44].
To provide potential mechanistic insights, we investigated the association of specific polygenic risk scores to DRM, FLM, and factor 9, a latent variable capturing social impairment which was found to be a point of convergence between FLM and DRM. Both DRM and FLM were significantly predicted by androgen-related PRS: DRM by testosterone and FLM by SHBG, an attenuator of free testosterone. These and other results of our investigations suggest that DRM and FLM may access a latent masculinity through complementary androgen-related mechanisms. Social impairment (factor 9) showed a combined nominal association where increasing testosterone PRS and decreasing SHBG PRS were independently and additively associated with increased social deficits. Taken together, these results pointed toward a hypothesis of increased net androgen exposure resulting in reduced social functioning.
Our investigation in the much larger SPARK cohort (N = 9419) added further evidence in favor of this hypothesis, as male sex, testosterone PRS, and SHBG PRS were all significantly associated, in the anticipated direction, with cooperative/imaginative play, and having a best friend or friends. It is noteworthy that these indicators emerged from a background of other possible social phenotypes, including joint attention, facial expressions, and binary communication (yes/no by head nodding or shaking). Our working hypothesis was based on the initial observation of a testosterone/SHBG association with a factor that, as in the SPARK study, loaded on friendships and social interactions (and not on other social phenotypes such as social anxiety). Further investigation is needed, but the fact that both the discovery sample and the replication sample converged on a friendship-oriented social phenotype is striking.
When considering the genomic reservoir of androgen-related risk, it seems intuitive that the sex chromosomes (and therefore our binary concept of male and female sex) would play a dominant role. However, our comparison of sex chromosome and autosomal effects (binary sex and polygenic estimates of testosterone and SHBG, respectively, see Fig.
5d, e) led to the unexpected finding that sex chromosome and autosomal contributions are comparable to each other in magnitude, with respect to risk of social deficits. Although perhaps not intuitive, this should not be surprising, given that a sizable majority of genes involved in androgen metabolism and sex differentiation pathways are encoded on autosomes (see Additional file
1: Table 5), and that genetic variation in these genes will shape and tune an individual’s response to the presence of sex hormones throughout their development. While further research is needed, these findings suggest the possibility of a latent and continuous “autosomal sex” that, like binary sex, plays a major role in neurodevelopmental outcomes, especially those related to social functioning.
How might the proposed genetic risk mechanisms impact social functioning? Sex hormones affect brain structure and function through multiple mechanisms and on different time scales. Human and animal research suggests that both short-term (e.g., neurotransmitter and neuropeptide) and long-term (i.e., developmental) mechanisms downstream of androgen exposure can lead to reduced capacity for social functioning [
81‐
84]. Studies in humans that have linked fetal testosterone exposure (FTE) to later reduced quality of social relationships [
85] and reduced empathic descriptions of social interactions [
86] underscore the long-term processes, potentially related to neural circuit formation, that may explain androgen-related social deficits. Taken together, the early predictive value of FTE and our observation that affected individuals show elevated masculinity relative to typically developing peers, even at an early age (Fig
2. panels g, h), suggest that indices of sex hormone exposure may hold promise as early markers of developmental risk. The more immediate behavioral effects of androgens might be explained by their interactions with signaling pathways involving the neuropeptides oxytocin and vasopressin, which have been well-established as key mechanisms in social behavior [
87,
88], and have recently shown promise in clinical trials for their influence on social functioning [
89]. Our results build on this previous work by providing genetic evidence that increasing testosterone and decreasing SHBG (overall corresponding to more potent net androgen exposure) may underlie the observed social deficits that were indexed by our morphological estimates of masculinity, DRM and FLM. Furthermore, our results may offer a partial explanation for why males are diagnosed with NDDs at high rates than females [
1]: males, on average, will have higher androgen exposure and production throughout their lives, and are consequently at greater risk for disorders where social impairment is a factor in diagnosis, such as ASD and other NDDs. Conversely, females have lower androgen production and exposure throughout their lives, making this risk mechanism generally less applicable and resulting in a lower rate of diagnosis with disorders where social impairment is a decisive factor. A key contribution of this work is that we show that
autosomal androgen-related risk mechanisms (shown here as polygenic risk scores) can and do operate in males and females alike, contributing risk above and beyond that linked to sex chromosomes.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.