Sex mechanisms as nonbinary influences on cognitive diversity

Abstract

Essentially all neuropsychiatric diagnoses show some degree of sex and/or gender differences in their etiology, diagnosis, or prognosis. As a result, the roles of sex-related variables in behavior and cognition are of strong interest to many, with several lines of research showing effects on executive functions and value-based decision making in particular. These findings are often framed within a sex binary, with behavior of females described as less optimal than male “defaults”– a framing that pits males and females against each other and deemphasizes the enormous overlap in fundamental neural mechanisms across sexes. Here, we propose an alternative framework in which sex-related factors encompass just one subset of many sources of valuable diversity in cognition. First, we review literature establishing multidimensional, nonbinary impacts of factors related to sex chromosomes and endocrine mechanisms on cognition, focusing on value- based decision-making tasks. Next, we present two suggestions for nonbinary interpretations and analyses of sex-related data that can be implemented by behavioral neuroscientists without devoting laboratory resources to delving into mechanisms underlying sex differences. We recommend (1) shifting interpretations of behavior away from performance metrics and towards strategy assessments to avoid the fallacy that the performance of one sex is worse than another; and (2) asking how much variance sex explains in measures and whether any differences are mosaic rather than binary, to avoid assuming that sex differences in separate measures are inextricably correlated. Nonbinary frameworks in research on cognition will allow neuroscience to represent the full spectrum of brains and behaviors.

1. Introduction

Neuroscientists have long understood that sex-related mechanisms have the potential to influence both brain and behavior. However, whether the potential role for such mechanisms in brain and behavior was an opportunity or a problem depended on a researcher’s point of view. For many, this role was clearly perceived as a problem. Female animals and humans were systematically excluded from most research for many years, based on a largely unfounded fear of reduced power to observe significant effects due to a perceived risk of increased variability in females, ascribed in large part to hormonal cycling. A significant body of literature has now repeatedly demonstrated that females are in fact not more variable than males, which has begun to turn the tide (Becker et al., 2016; Beery and Zucker, 2011; Prendergast et al., 2014; R. M. Shansky, 2019). As is well known, in 2016 the NIH mandated the “consideration of sex as a biological variable (SABV)”, which followed the mandated inclusion of women as well as men in clinical research in 1993 (Beery and Zucker, 2011). The goal of these mandates was to avoid losing valuable data about natural variability in humans and model species, as lack of this testing was believed to have contributed to several incidents of unanticipated and dangerous side effects of approved drugs in some women (Nunamaker and Turner, 2023; Shansky and Murphy, 2021) although whether these related at all to sex mechanisms per se remains contentious (Greenblatt et al., 2019; Lee et al., 2023; Rushovich et al., 2023; Zhao et al., 2023). Although the SABV mandate reduced sex bias against using female animals in neuroscience publications, significantly increasing the number of papers using both males and females, it did not eliminate it, as single-sex publications in neuroscience-relevant fields continued to be male- biased (Li et al., 2021; Mamlouk et al., 2020; Nunamaker and Turner, 2023). Moreover, we would argue that the concern about sex influences did not disappear, but began to be expressed differently, with many more publications subsequently including females in their experiments consistent with NIH policy, but not analyzing data to test for potential sex effects—either analyzing it incorrectly (Garcia-Sifuentes and Maney, 2021) or not at all (Mamlouk et al., 2020; Woitowich et al., 2020). In short, while data collection in behavioral neuroscience has shifted towards better incorporating biological diversity related to sex, data analysis and interpretation necessary to benefitting from this effort has lagged significantly behind. This review and commentary aims to address one aspect of what may be limiting these efforts: the use of binary frameworks to represent variability in brains and behaviors.

Due to trailblazing research prior to these mandates, and a wealth of data that has emerged subsequently when sex factors have been analyzed, we now know more than ever before about numerous sex mechanisms and the ways they potentially impact neural structure and function, especially in model species (Becker et al., 2005; Becker and Chartoff, 2019; Hodes and Kropp, 2023; McCarthy et al., 2012; Shansky and Murphy, 2021). “Sex-related mechanisms” in this research can refer to one or more factors that are typically required for the successful production of one kind of gamete (egg or sperm) and/or successful reproduction with that form of gamete in a species. These are not all necessarily active in an individual or in a sample at a given time, reflecting the contextual and life history influences on what constitutes sex (Griffiths, 2021; Richardson, 2022). In the domains most likely to influence neural function, at minimum these mechanisms include 1) sex chromosome genes and their regulation, and 2) gonadal effects, including both early life and pubertal gonadal steroid hormones and their regulation. Much of this research has elucidated intricate genetic and endocrine mechanisms regulating reproductive and other innate behaviors in rodent models (Gegenhuber et al., 2022; McCarthy, 2023; McCarthy and Arnold, 2011). Although for simplicity many scientists are used to “rounding up” sex into two binary categories of “male” and “female”, this literature has clearly described not two inflexible, binary “male and female” states, but instead a nonbinary spectrum of sex influences. This can be surprising, as one might expect sex mechanisms’ role in regulating behavior to be the most dichotomous and/or least flexible in reproductive behaviors. However, even reproduction-relevant “innate” behaviors in rodents are in fact remarkably experience dependent (Karigo et al., 2021; Mei et al., 2023; Remedios et al., 2017), suggesting that even sex-biased behaviors relevant to reproductive roles are not inevitable or binary, but reflect interactions between an individual’s unique sex-related variables and their environmental milieu. This complex, nonbinary relationship between sex mechanisms and behavior is even more likely to be the case in the context of higher-order cognition, but this has been understudied compared to binary frameworks, described below.

Sex differences in cognition have been of longstanding interest because of the potential for sex mechanisms to influence differential neuropsychiatric risk and resilience between individuals. Essentially all neuropsychiatric diagnoses show some impact of sex and/or gender in their etiology, diagnosis, or prognosis (Grissom and Reyes, 2019). A prominent subdomain of cognition that has been appealed to as potentially influenced by sex mechanisms are executive functions, as alterations in executive functions are found in high-profile neuropsychiatric challenges that show an apparent sex and/or gender bias, especially neurodevelopmental diagnoses and addictions (Becker et al., 2017; Becker and Chartoff, 2019; Santos et al., 2022). Available evidence suggests there is a role for sex-related mechanisms in these diagnosis biases (Kiraly et al., 2018; Mossa and Manzini, 2019; Shansky and Murphy, 2021; Werling and Geschwind, 2013), but we note that this does not exclude a role for gender roles and experiences in the development and expression of these diagnoses (Ellemers, 2018).

One domain of executive function, value- based decision making, has been shown to be sensitive to sex-related factors (Grissom and Reyes, 2019). During this type of task, individuals choose between two or more options that differ in expected value as well as in some risk or cost (Grissom and Reyes, 2019; Orsini and Setlow, 2017; Weafer and de Wit, 2014). In humans and rodent models, these findings can be roughly summarized as showing that males on average are less avoidant of rewarding choices that may also result in negative outcomes (punishment, nonreward, or reduced reward). This is a binary framing of the literature, however, that is inconsistent with the complex, nonbinary relationship we already know exists between sex mechanisms and “less cognitive” behaviors. This is a challenge for our field for two reasons. First, systemic and implicit sex biases have led to a literature which often describes the behavior of female animals and people as less optimal than male “defaults”, an interpretation which can only exist if male and female are pitted against each other as binary opposites. Descriptions of performance on tasks measuring cognition, executive functions, and learned behaviors are often imbued with value judgements reflected in terms like “correct performance”, “errors”, “superior”, and “deficit”, implying that neutral mechanisms of variation, such as those related to sex, are necessarily incorrect or nonoptimal (Grissom and Reyes, 2019). Second, framing females and males as opposites of a binary ignores the enormous overlap in fundamental neural mechanisms across all mammals, and obscures important variation that can occur not only across sexes but within them. Sex differences in brain and behavior reported in the literature reflect differences in group means, but significant overlap between groups tells us that important variability in an individual may not be captured by knowing their overall sex category (Joel et al., 2018; Maney, 2016). Given interest in sex mechanisms in executive functions and evidence for these mechanisms in affecting value based decision making, we propose to use this literature to offer evidence against a sex binary and an alternative to binary interpretations of sex in data, to avoid sexism and increase researcher comfort and interest in understanding sex mechanisms in their own work.

We offer this review and commentary to provide evidence for a key fact - sex influences on cognition are nonbinary. By “nonbinary”, we mean simply that sex mechanisms in mammals do not exist in one of two discrete and immutable states. A nonbinary view of sex influences is now supported by overwhelming evidence from multiple fields including psychology, neuroscience, and behavioral ecology and evolution (Hyde et al., 2019; Joel, 2021; McLaughlin et al., 2023), providing a key conceptual framework for understanding sex in the context of brain and behavior. In this framework, “sex” is a proxy term for numerous mechanisms which can sometimes correlate, but not always, because they are dissociable from each other. In other words, sex is multidimensional. This does not mean that sex mechanisms do not influence neural systems - far from it. However, it does mean that we cannot conclude that an observed average sex difference in behavior is the result of two separate ‘patterns’ of neural mechanisms acting in a single dimension to create two opposed sex-behavior states. Instead, available evidence suggests that multidimensional sex variables are “tuning factors” which interact with non-sex genetic and environmental variability to influence neural systems and behavior in an individual. Using value-based decision making as a key example, we can demonstrate how this may serve to create useful variation in strategies within and across individuals to navigate decisions in a restless, non-optimizable world.

There are several broad dimensions by which sex is widely understood to influence neural function. At a cell-autonomous level, sex chromosome copy number and gene variants influence development and function across the lifespan. At an organismal level, gonadal steroid hormones such as estrogen, progesterone and testosterone exert endocrine effects throughout the body, including the brain. This endocrine function is enriched in the brain by local neurosteroid synthesis and metabolism. Lastly, endocrine mechanisms and sex chromosome mechanisms can mutually influence each other, via the transcriptional role of steroid hormone receptors on both autosomal and sex-linked loci. We will first review current understanding for how any of these dissociable sex mechanisms can influence reward-guided decision making as a subdomain of executive function, including motivated behavior, outcome processing, exploratory, habitual, and perseverative behaviors, and relevant neuroanatomical findings. Following this, we will use this evidence to offer two suggestions to reduce bias and binary framings in research that are intended to be adoptable by any laboratory without needing to measure sex mechanisms directly (Fig. 1). We offer these approaches in the service of interpreting sex influences as contributors of species-wide variation in the neural mechanisms of cognition across individuals.

Fig. 1.

Two suggestions for experimenters on how to interpret potential sex influences in their data while reducing the risk of binary or sexist bias. The suggestions are illustrated with yellow and purple dots representing individuals to avoid any color associations with male or female sexes. Suggestion One is to consider using strategy assessments instead of performance metrics to describe differences in behavior. We suggest avoiding framing central findings as statements of better or worse performance between sexes, such as “yellow made more correct responses than purple”. Instead, consider interpreting behavioral differences across individuals as reflecting different prioritizations or tradeoffs. In the framework on the right, yellow dots display a strategy that prioritizes outcome X over other possible outcomes, while purple dots display a pattern that prioritizes outcome Y. This has the potential to avoid sexist interpretations of behavior based on a narrow focus on a specific outcome measure, and has the benefit of supporting potential neural circuits hypotheses that would support selection of different strategies across individuals and environments. Suggestion Two is to consider asking how much variance in behavior is contributed by sex, and whether sex impacts are mosaic. When considering sex differences in multiple outcome measures, such as Behavior A and Behavior B, we suggest avoiding conclusions that these outcome measures are correlated and regulated by the same sex mechanism without evidence. Instead, consider asking whether sex differences are a major dimension by which your measures differ, and whether differences in one measure are correlated with sex differences in another (or not). In the framework on the right, yellow dots are both on average different in Behavior A and Behavior B than purple dots are, but Behaviors A and B are not strongly correlated. It is likely that the reasons an individual shows some level of Behavior A have little to do with the reasons for showing Behavior B, even though there are average differences between purple and yellow. This has the potential to avoid sex binary interpretations of neural and behavioral measures, and has the benefit of supporting potential hypotheses disentangling different biological mechanisms contributing to Behaviors A versus B.

2. What evidence is there that sex differences in cognition and behavior are nonbinary?

2.1. Nonbinary sex chromosome impacts

It is understandable that many people assume sex impacts on neural systems are binary, because of how we usually think of sex chromosomes. A karyotype of two X chromosomes for female mammals and XY for males seems to argue for a categorical difference between two fundamental groups, one of which develops female reproductive gonads and characteristics, while the other develops a male variety. However, it is not so simple. Sex in (most) mammals begins with a single gene, the testis-determing factor SRY on the Y chromosome, the presence of which initiates a developmental program that usually results in male gonadal development (Wallis et al., 2008). However, the presence of other genes on the Y chromosome, the number of copies of X chromosome an individual carries, and the gene variants on these chromosomes also exert influences. Thus, even on a purely chromosomal or genetic level sex is multidimensional, with correlated but dissociable factors including 1) presence or absence of SRY 2) the number of each type of sex chromosome, 3) genetic variation on sex chromosomes, and 4) interactions between autosomal gene variants and sex chromosomes. There is evidence that many of these mechanisms independently influence cognition and executive function.

Mouse models to manipulate sex chromosome complement (reviewed in (Arnold, 2020)) began with the Four Core Genotypes line, which arose from a spontaneous deletion of SRY on the Y chromosome which was later followed by deliberate translocation of SRY onto an autosome. Thus, XX animals that are also SRY+ can develop testes, and XY animals that are SRY- can develop ovaries, in addition to the more usual matchup of XX to ovaries and XY to testes. Likewise, the XY* line arose from a spontaneous mutation in mouse Y chromosome leading to aberrant recombination with X chromosome, producing offspring modeling copy number variation in the X chromosome that are functionally similar to XO and XX females, and XXY and XY males.

Research with these two mouse lines demonstrates that sex chromosomes have an impact on cognition and neural circuits relevant to value-based decision making that is dissociable from gonadal and endocrine sex mechanisms. The relevant domains include motivation, hedonic processing, and action repetition. Motivation to perform operant responses are higher in XY animals than XX regardless of gonads (Barker et al., 2010; Seu et al., 2014). However, actual consumption of reward was additively increased by both XY status and development of testes (Kopsida et al., 2013; Seu et al., 2014), corresponding with increased body weight in all animals with testes, which following gonadectomy revealed increased body weights in XY carriers (Chen et al., 2013). These data are consistent with a general dissociation between motivational drive and hedonic value (Berridge, 2018), with motivation apparently more completely influenced by XX/XY chromosomes, while hedonics shows influences of gonadal status that overshadow or synergize with chromosomal influences. Consistent with a role for gonadal sex regardless of sex chromosomes in the processing of emotional valence, mice with testes show increased exploration in uncertain and anxiogenic environments, such as elevated plus maze and zero mazes (Kopsida et al., 2013). Tasks assessing action selection and automaticity have repeatedly found evidence for chromosomal impacts. Perseverative errors in a probabilistic reversal learning task are higher in XY than XX animals, suggesting reduced aversion to mistakes or reduced value updating in XY (Aarde et al., 2021), and higher in XXY than XY (Aarde et al., 2019). Habit based responding for palatable foods are higher in XX carriers (Quinn et al., 2007), but habitual responding for alcohol is higher in XY carriers (Barker et al., 2010), an intriguing discrepancy that may be related to differences in the pharmacological impacts of alcohol. Lastly, the structure and gene expression of numerous brain regions implicated in reward, motivation, cognition, and executive function are influenced by chromosomal status independent of gonadal status (Chen et al., 2009; Corre et al., 2016; Vousden et al., 2018). Collectively, these findings suggest that cognitive domains such as motivation, perseveration, and habit may be sensitive to chromosomal status when gonadal status is held constant, while hedonics and valence may be sensitive to gonadal status, holding chromosomal status constant. However, a final caveat to this work is evidence that SRY itself is functional in the brain, regulating the synthesis of dopamine in the ventral tegmental area and substantia nigra and leading to increased risk of death of these neurons in XY + SRY mice (Czech et al., 2012, 2014; Dewing et al., 2006; Lee et al., 2019), raising the possibility that at least some effects attributed to male gonadal development in the Four Core Genotypes model may be related (instead of, or in addition to) SRY expression.

It is sometimes remarked that these mouse lines represent rare cases of intersex biology that do not have a bearing on typical sex differences as understood by most neuroscientists (or laypersons). One caveat to this argument is the base rate of sex chromosome copy number variants in any species is difficult to confidently ascertain without widespread karyotyping, leading to underrecognition of their prevalence. Both the Four Core Genotypes and XY* lines exist due to spontaneous mutations in sex chromosomes in laboratory mice that happened to have been investigated rather than disregarded (Arnold, 2020). Sex chromosomes are sites of continuous and heightened evolution relative to other chromosomes (Furman et al., 2020), to the point that there are rodent species that have eliminated Y chromosomes entirely (Gileva et al., 1982; Kuroiwa et al., 2010; Ortega et al., 2019). In humans, variation in the number of sex chromosomes are uncommon, but not as rare as sometimes assumed. Current estimates for alternative sex karyotypes are at least 1 out of 440 people (Breman and Stankiewicz, 2021), but some chromosomal combinations are more likely to have carriers that are asymptomatic, such as Kleinfelter’s syndrome (XXY), which is thought to be unidentified in up to 65 % of carriers (Bojesen et al., 2003; Groth et al., 2013). While variation in the absolute number of chromosomes may ultimately be rare but observable, there also are well-recognized impacts of functional variations in sex chromosome genes on brain and behavior.

Variants of sex chromosome genes are also poised to directly impact cognition and learned behavior in a nonbinary fashion, although we do not always immediately think of these as “sex mechanisms”. Variants in almost 20 % of X chromosome genes are linked to autism and neurodevelopmental diagnoses, including prominent examples such as Fragile X syndrome (impairment of FMR1) and Rett’s syndrome (impairment of MeCP2) (Brand et al., 2021; Hunter et al., 2019; Ribeiro and MacDonald, 2020). Each of these syndromes show substantial sex biases in their clinical impacts. Fragile X syndrome is much more likely to be diagnosed in people with only one X chromosome, while Rett’s syndrome is typically not survivable in people with only one X chromosome, each case highlighting the importance of X mosaicism in multi-X carriers (see below). Impairment of the function of these X chromosome genes likewise produce profound, sex-biased impacts on the mechanisms of cognition and learned behavior in animal models (Armstrong et al., 2020; Nolan et al., 2017; Ribeiro and MacDonald, 2020; Schmitt et al., 2022). While extremely large effects emerge from substantial alterations in the function of these genes, variability in cognition and learned behavior can also result from subclinical variants in sex chromosome linked genes in humans. As an example, impairments in FMR1 resulting in a Fragile X Syndrome diagnosis involve a repeat sequence of CGGs outside of the coding region of the gene. Fragile X syndrome is typically associated with over 200 repeats of this sequence. People with any number of subthreshold variants (50–200 repeats range) are at higher risk of multiple neuropsychiatric diagnoses impacting executive function, including mood and neurodevelopmental disorders, and later in life are at higher risk of motor impairments (Bourgeois et al., 2009; Gossett et al., 2016). While the above considered all subthreshold carriers categorically, recent work has suggested a more granular or dose-dependent impact of subthreshold CGG repeat numbers on executive functioning and cognitive ability in humans (Maltman et al., 2023). Subclinical FMR1 variants demonstrate that variation in the function of sex chromosome genes may lead to nonbinary effects across individuals with the same chromosomal complement.

Chromosomal impacts are not often considered as explanations for sex variation in standard laboratory animal models, due to high genetic similarity between individuals in these strains. A counterpoint to this is that chromosomal sex effects are at play even when they are not dissociated via a genetic model. An additional source of sex chromosome variation that can play a role is X mosaicism. Within each cell of individuals with two or more X chromosomes, one X chromosome is (largely) inactivated, with the effect of equalizing the dosage of most (but not all) X-linked genes between XX and XY carriers. This process occurs early in embryonic development in a stochastic manner that generally results in “patching” of maternal or paternal X chromosome expression throughout the body, most easily visualized as the mechanism for calico or tortiseshell fur patterning in XX cats. This process is not easily visible in other species or tissues without a visual indicator for different parent Xs, or single cell genetic analysis, so our understanding of parent-of-origin X effects is currently limited to rodent models with these techniques. The degree to which one parent’s X chromosome over another is favored can vary across individuals, even across littermates (Wu et al., 2014), and can exert different impacts across brain regions and cell types. Skew towards maternal versus paternal X expression in cortical pyramidal neurons is substantially greater than the more even mixture of maternal and paternal X in cortical interneurons, reflecting stochastic progenitor effects in these two neuronal populations that is poised to influence cognition (Wu et al., 2014). Recent work has demonstrated a substantial lopsidedness to X inactivation in the mouse brain, favoring maternal X expression overall (Szelenyi et al., 2021; Wang et al., 2010). This can protect neural development and cognitive function from the developmental impacts of a deleterious paternal X allele more effectively than the reverse (Szelenyi et al., 2021), but driving expression of only the maternal X accelerates cognitive aging and impairs hippocampal learning and memory in female mice (Abdulai-Saiku et al., 2022). The stochastic variation of parental X expression between cells, brain regions, and individuals with two X chromosomes is therefore poised to be a cryptic, underrecognized, and nonbinary influence on cognition and behavior.

An underlying assumption of many researchers is that the impacts of sex mechanisms are the same within a species, if not within all mammals. However, it is increasingly recognized that the impact of autosomal gene variants within a species depends on the genetic background on which they are expressed (Cirnigliaro et al., 2023; Sittig et al., 2016), opening the question of whether this is true in sex gene variants as well. Indeed, in mouse models, autosomal gene variants linked with neurodevelopmental disorders have been repeatedly demonstrated to interact with sex to produce diverging phenotypes in cognition and learned behavior (Giovanniello et al., 2021; Grissom et al., 2018; Kim et al., 2023; Rojas et al., 2023). However, sex differences are also influenced by neutral variation in genetic background. An examination of 84 inbred mouse strains for operant administration of cocaine or saline revealed massive strain differences in overall lever pressing behavior, as well as cocaine preference over saline. However, only 9 of 84 strains showed a significant sex difference in this behavior, with half (5/9) displaying increased preference for cocaine in males versus females, with half showing the opposite trend (Bagley et al., 2022b). This is corroborated by sex differences in the psychomotor impacts of cocaine depending in large part on the background genetics of mice (Chapp et al., 2023). However, a large sample of over 400 diversity outcross mice did find overall increased errors in probabilistic reversal learning in males compared to females (Bagley et al., 2022a) (compare to increased errors in this task in XY compared to XX mice (Aarde et al., 2021)). Because model organism work in neuroscience usually employs only one strain, these findings urge caution in assuming a sex difference in one strain or species will be apparent in another, for example, sex differences in a particular strain of rats in discounting tasks (Hernandez et al., 2020; Orsini et al., 2016) versus no sex differences in a particular strain of mice in similar tasks (Rojas et al., 2022). It has even been suggested that background strain differences in Four Core Genotypes experiments may contribute to variation across labs (Aarde et al., 2021). At the same time, they bolster the need to investigate sex differences in rodent models to gain a complete picture. Collectively, these findings indicate that sex mechanisms may bias a cognitive phenotype across a population (eg, biasing the strategies employed in probabilistic learning), but that variation at a genetic and environmental level can mitigate or even change the nature of sex influences on the expression of a phenotype. This would be expected to produce nonbinary outcomes of sex.

2.2. Nonbinary impacts of endocrine mechanisms

Relative to chromosomal status, endocrine sex mechanisms are recognized as much more continuous variables, and there are numerous avenues through which their influence can vary in a nonbinary capacity. Ebbs and flows of gonadal steroid hormones, particularly estrogen, are often appealed to as a primary suspect in driving variations in learned behaviors and cognition in female animals compared to males, despite there being no increased variance in behavior between females compared to between males (Becker et al., 2016; Rebecca M. Shansky, 2019; Shansky and Murphy, 2021) and despite the fact that males also experience circadian rhythms in steroid hormone expression (Iwahana et al., 2008). More recently, androgen variation in males, for example due to social roles or hierarchies as well as during diurnal cycles, has been proposed as an additional potential driver of variation. However, despite the binary roles assigned to these hormones, estrogens, progestins, and androgens, and their receptors, are all expressed in mammals of all sexes (Bianchi et al., 2021; Dart et al., 2013; Hess and Cooke, 2018; Oettel and Mukhopadhyay, 2004). These levels are biased across individuals in sex categories in important and obvious ways, but can also vary widely between individuals of the same sex, as well as within individuals over cycles and across the lifespan, but our ability to recognize this is limited by inaccuracy and low sensitivity of inexpensive testing methods, with expensive and laborious technology necessary for reproducible hormone measurements (Kanakis et al., 2019; Stanczyk et al., 2007). In addition to the hormones themselves, individual variation in the expression and function of steroid hormone receptors and binding proteins which prevent receptor activation can vary across individuals (Deswal et al., 2018; Pasinetti and Falsetti, 1993; Zimmerman et al., 2014). Thus, individual variation in circulating levels of gonadal hormone ligands and their targets at these different timescales provide a first level of nonbinary influence of endocrine mechanisms. Second, the influence of these hormones are contributed not only by expression and synthesis in the gonads post pubertally, but also by prepubertal influences of gestational endocrine exposure and lifelong influences local steroid synthesis in the brains of males and females (Giatti et al., 2020) to varying degrees across sexes and brain regions (Cisternas et al., 2015, 2017). Local metabolism of estrogen in the brains of males regulates reproductive behaviors and systemic testosterone levels (Brooks et al., 2020), highlighting a persistent and important role for estrogens in males and the complex interactions between endocrine systems. These papers largely discuss the activational role of hormones post-puberty, but there are also substantial roles for these hormones prepubertally, including an organizational requirement for gestational estrogen exposure to yield a brain capable of producing male-typical reproductive behaviors in rodents (McCarthy and Arnold, 2011). Thus, estrogen, progesterone, and testosterone influences from beyond the gonads of an individual animal (maternal and neuronal) provides a second avenue for individual differences, and thus nonbinary influences, of hormones.

There is ample evidence that estrogen, progesterone, and/or testosterone influence the function of neuronal circuits essential to value-based decision making. Ovarian steroid hormone cycling has been repeatedly demonstrated to influence neuromodulator systems important to decision making and executive functions in adult animal models, including by regulating midbrain dopamine synthesis in adults (Becker and Chartoff, 2019; Yoest et al., 2014), and changing the excitability of dopamine neurons (Calipari et al., 2017). (Compare to the role for SRY gene expression in regulating dopamine synthesis (Czech et al., 2012, 2014)). Likewise, estrogen enhances norepinephrine synthesis and reduces clearance in adult female rats (Bangasser et al., 2016). Estrogens are widely recognized for having impacts on neuronal connectivity throughout the brain by modulating dendrite structure and spine density via estrogen receptor mediated potentiation of metabotropic glutamate signaling (Meitzen et al., 2018). These include modulating the number and connectivity of rodent medium spiny neurons in the striatum across postgestational development, important for outcome signaling and action selection (Meitzen et al., 2018). Estrogens modulate spine density of prefrontal cortical regions critical for executive control and choice (Premachandran et al., 2020). Prefrontal cortex dendritic spines are on average greater in number in males than females, an effect partially driven by male puberty (Delevich et al., 2020; Kolb and Stewart, 1991). Moving from pubertal development to gestation and early life, there is also evidence for nonbinary impacts of hormone effects. Testosterone exposure during gestation can increase the number of dopamine synthesizing cells in both male and female rodents and sheep (Brown et al., 2015; Gillies et al., 2014), although elevated expression of enzymes for dopamine synthesis can be elevated in midbrain dopamine neurons of male mice prior to significant prenatal androgen surges (Sibug et al., 1996), suggesting a synergistic or combinatorial influence of sex chromosomes and developmental endocrine mechanisms on dopamine function. Collectively, variability in endocrine mechanisms across the lifespan are poised to exert nonbinary effects on neural circuits of decision making.

Decision making itself is also influenced by circulating endocrine function. In broad summary, ovarian steroid hormones are linked with increased sensitivity to negative outcomes in decision making, while testosterone either has no effect or serves to inhibit impulsive choices acutely, but promote seeking of risk when chronically elevated, in both rodents (Hernandez et al., 2020; Orsini et al., 2016, 2022) and in humans (Ambrase et al., 2021; Coenjaerts et al., 2021; Derntl et al., 2014). The impacts of specific manipulations of these hormones on decision making in animal models was recently thoroughly reviewed in (Orsini et al., 2022), so we will focus on potential nonbinary impacts of these endocrine mechanisms. The literature has sometimes tested the overtly nonbinary question of whether individual gonadal hormones have consistent impacts on behavior across adult individuals, or whether individuals in a male or female sex category tend to show different impacts of differing hormone levels. Because estrogen, progesterone, and testosterone are functional across sexes, the capacity to respond to these hormones might be expected to have similar impacts in the brain. On the other hand, the chromosomal and gestational influences on neural circuits relevant to decision making could be one of several avenues creating variability in the capacity to respond to these hormones across sex categories. Indeed, this literature is mixed. In rats, combined gonadectomy with exogenous hormone replacement revealed both disparate impacts of endogenous gonadal function, but shared impacts of exogenous estrogen. Female rats became more willing to risk a negative outcome during decision making following removal of the ovaries, while male rats became less willing to accept this negative outcome following removal of the testes. Following gonadectomy, sex steroid hormones were directly administered to the animals. Acute testosterone delivery had no effect on males, but estrogen reduced risk taking in both males and females (Orsini et al., 2021). However, a study asking a similar question in humans highlights not only the complexity of testing endocrine mechanisms, but the way in which a sex binary framing can impede our ability to understand these mechanisms. Acute administration of estrogen to men and women had opposing effects on their willingness to accept offers in an ultimatum game, with estrogen associated with a decreased acceptance of offers in women but an increased acceptance of offers in men. However, unlike the animal experiment, endogenous gonadal function was still having an influence, with treatment increasing estrogen to a much greater degree in women than in men, which could have contributed to opposing effects. In addition, participants’ (unfounded) assumptions about whether they had received estrogen or placebo also had a significant influence on offer acceptance of the placebo group, leading to acceptance of worse offers in participants who believed they had been given estrogen, especially men (Coenjaerts et al., 2021). This data provides a stark example of how binary and sexist assumptions about how sex mechanisms influence cognition can impede our ability to understand how these mechanisms actually work.

3. If sex is nonbinary, how can we interpret sex differences in cognition and behavior?

It would be understandable, but unfortunate, to review the above evidence and conclude that sex mechanisms are too convoluted to understand in the context of cognition and so we should not try, or to revert to a sex binary interpretation of cognitive data out of ease or familiarity. We hope that we have demonstrated that binary sex categories are ill-equipped to shed light on how biological variability linked with sex can influence cognitive diversity. It is important not to conflate the stereotypes attached to gender roles to sex mechanisms (Perfors et al., 2023). Every individual regardless of gender identity experiences variation in multiple sex mechanisms throughout their lifespan that have the potential to influence cognition, including factors such as sex chromosome gene variants, epigenetic alterations, gestational hormone exposure, adrenarche, puberty, changes in hormones due to pregnancy, child rearing, menopause, andropause, and reproductive, sexual, and gender-affirming healthcare. Sex is nonbinary and therefore sex influences on cognition, where they exist, must also be nonbinary.

However, this does not mean that all research programs on cognition, or even executive functions in particular, must devote themselves to elucidating the exact contributions of each sex-linked mechanism. Certainly, changes to experimental design to incorporate sex variable manipulations can be beneficial, but this is not always possible or easy. Not everyone is able or willing to transform their laboratory’s research into a pursuit of sex mechanisms per se, nor should this be an expectation as it would limit resources to pursue numerous other mechanisms. However, there is still an expectation from both funding agencies and the field at large that males and females will be assessed across studies to aid “consideration of the influence of sex” (“NOT-OD-15–102: Consideration of Sex as a Biological Variable in NIH-funded Research,”, n.d.), which mandates disaggregation of sex data in the name of uncovering biological variability. Although researchers have not always proceeded to test for sex differences in their datasets, this mandate means that research moving forward will inevitably identify average sex differences, and these will require interpretation even though investigators may not be comfortable with the process (Maney, 2016; Rebecca M. Shansky, 2019).

Because sex binary interpretations are not upheld by data, how are investigators to proceed? Perhaps the most straightforward approaches require only adjustments to how we think about and interpret data that has already been collected in male and females. Here we suggest two practices for interpreting sex influences on cognition that can be implemented by any laboratory, without additional cost or experimentation (Fig. 1). We wish to emphasize that these are practical recommendations for researchers to reveal nonbinary influences of sex factors even in datasets in which individuals are already assigned to binary sex categories as a minimum “consideration of the influence of sex”. First, shift interpretation of behavior away from performance metrics and towards strategy assessments, to avoid the fallacy that one sex is worse than another. Second, consider asking how much variance sex explains in measures, and whether sex impacts are mosaic rather than binary, to avoid assuming that sex differences in separate measures are inextricably correlated.

3.1. Suggestion one: consider strategy assessments instead of performance metrics

Our first suggestion is to be skeptical of frameworks, including in past literature, that labels the behavior of typical individuals of one sex category as inferior or superior to another in a given dependent measure (Grissom and Reyes, 2019). Instead, when examining data that presents a spread of possible actions in a task across sexes, a useful approach can be to consider the importance of context - in other words, to assess and describe strategies across individuals while asking what context would make each of these strategies optimal (Chen et al., 2021a, 2021b). This can lead to the discovery of neural tuning factors that prioritize and reassess decision variables based on different situations and individual biology, producing diversity in decision making for reasons beyond sex biology (Cazettes et al., 2023; Engelhard et al., 2019; Kane et al., 2022).

In animal models, and even more so in human tasks, we as experimenters often assume that the optimal strategy or goal for the task is obvious to the individual being tested. Sometimes this assumed strategy or goal is established from prior literature which made the normative definition of correct from male samples, such as the example of the prototypical response to fear conditioning being freezing (emitted pre-dominately by male rats) rather than escape behaviors (emitted pre-dominately by female rats) (Shansky and Murphy, 2021). In many decision making tasks, the assumed strategy can be calculated by the experimenter as the approach which extracts as many rewards as possible, while committing zero behaviors (impulsive, checking, punishment avoiding) that do not immediately lead to reward (Grissom and Reyes, 2019), This framing is based on the implicit assumption that collecting rewards as fast as possible is also the goal of the subject. However, these reflect only a single or limited set of goals an individual can have in a task compared to the wider distribution of goals a typical individual could consider - a distribution of goals that can continuously change as a result of external environment as well as internal state. As noted in the introduction, the results of these assumptions, combined with a binary sex framework, can lead to labeling the behavior of an entire sex as sub-optimal.

When confronted with an average difference in behavior across sexes in typical individuals, an alternative approach can be to examine the entire range of behaviors produced by all animals and consider whether the patterns elicited represent a tradeoff between options. One example readers may already be familiar with is a sex difference in the primary response to fear conditioning, revealed when female rats showing a fast-moving “darting” behavior, while males predominantly showed freezing behavior to the shock (Gruene et al., 2015). This literature was used to define non-freezing behavior in females in prior studies as reflecting poorer learning and memory. However, follow up work has shown that darting behavior increases in males with extended training (Mitchell et al., 2022), suggesting, if anything, that darting could reflect faster learning. However, this interpretation itself could lend itself to potential bias about which group is “better”. Rather than interpreting these data as reflecting a sex binary in the ability to learn, we would advocate that these findings suggest a tradeoff between passive and active coping strategies elicited by an aversive stimulus (Koolhaas et al., 2007; Radley and Herman, 2023). The argument that darting versus freezing may reflect changes in active versus passive coping are underscored by the observation that the proportion of animals of any sex that dart varies based on the severity of the shock (Mitchell et al., 2022). These findings suggest that neural circuit tradeoffs between active and passive coping are possible within individuals as well as between them, with sex mechanisms likely to access some portion of the mechanisms that could mediate this tradeoff.

When considering sex influences in value-based decision making in the context of tradeoffs, our own work suggests that across sexes, animals vary in their solution to the tradeoff between exploration and exploitation. The explore-exploit tradeoff is a well-established problem in decision science and reinforcement learning and, critically, no solution for this tradeoff is optimal in all situations (Addicott et al., 2017). While optimal strategies can be calculated for the many decision tasks used in laboratories, it is inevitable that a solution that would lead to more reward collection in one task design could be deleterious to performance in a different task design. The explore-exploit framework is thus an ideal testbed for our proposal that average sex differences be interpreted as reflecting a latent tradeoff between strategies across individuals. To test this, we can examine behavior in male and female mice across probabilistic decision making tasks, sometimes called “bandit” tasks, which engage the explore-exploit tradeoff with different optimal solutions attached with differences in reward probability, probability of change, kinds of choices, etc. We observe a range of balances in explore-exploit behaviors across individuals in bandit tasks, including an average sex difference. On average, we have found female mice are more likely to commit to a repeated action strategy over exploratory sampling of options on any given trial (Chen et al., 2021a, 2021b). However, in order to be successful in these tasks, individuals have to engage in both patterns of behavior, and male and female mice both show periods of exploratory sampling and periods of repetitive choice for a (usually) more advantageous option. Consistent with the nonbinary impacts of sex, there are overlaps in strategy usage (Chen et al., 2021a) and probability of being in an exploratory state (Chen et al., 2021b) in males and females. Consistent with the idea that there are no universally optimal solutions to explore-exploit tradeoffs, in one task female mice earned more rewards than male mice (Chen et al., 2021a) while in the other reward collection rates were equivalent (Chen et al., 2021b), despite the greater average exploitation and action repetition in females in both tasks. These findings provide an exciting framework to investigate how both sex and non-sex mechanisms can contribute necessary individual variability in explore-exploit behavior to enable success over a wide range of decision situations. Indeed, it has recently been demonstrated that attention deficit symptoms are associated with more optimal behavior in a naturalistic foraging-inspired decision making task (Barack et al., 2024), demonstrating that variability in decision making strategies driven by factors other than sex also do not have universal impacts, but depend on the current structure of the environment.

When we consider sex differences as influencing the neural mechanisms of strategy tradeoffs, it becomes clear that interpreting the behavior of one sex as suboptimal not only depends on a biologically implausible binary, but does not make evolutionarily sense. The evolutionary pressures of foraging under which reward-guided decision making behavior evolved explicitly require balancing multiple variables that cannot always be mutually optimized, such as probability of identifying food, costs of travel, and costs of predation (Addicott et al., 2017). Sex influences are not likely to produce inflexible impacts on behavior which must be sensitive to these varied environmental factors, and which is so important to survival. A fundamental principle of neural systems appears to be that they have evolved to be incredibly robust to neuromodulatory variations when achieving results important to survival, constraining the observed outputs of that neural system (Marder and Goaillard, 2006). Indeed, neural sex impacts in mammals are sometimes found to increase similarity in behaviors across sexes when similarity of those behaviors benefits the survival of the species (De Vries, 2004). This does not mean that sex mechanisms never have an impact on behavior, but evidence shows that these impacts are variable not only as a function of current sex mechanisms (eg age, reproductive status) but also environment. The ecology and evolutionary biology literature has demonstrated evidence for slightly sex-biased foraging strategies in a wide range of species (including many with non-mammalian sex-determining mechanisms which vary over the lifespan) such as seabirds, seals, freshwater fish, flies, crickets, and mice (Bennison et al., 2022; Brand et al., 2023; Lidgard et al., 2020; Maklakov et al., 2008; Perrigo and Bronson, 1985; Powolny et al., 2014; Reddiex et al., 2013). However, even when dichotomous or sex-specific nutrient foraging behaviors would be demonstrably more optimal, animals are constrained towards behavior that would benefit any individual regardless of sex (Maklakov et al., 2008; Reddiex et al., 2013). This, combined with the fact that sex influences vary across individuals and over a lifespan, indicates evolutionary pressure would be to maintain neural circuits for foraging that are fundamentally similar across individuals, with sex influences in behavior emerging only under specific conditions. Indeed, in wild mus musculus (the wild type for most laboratory mice), sex biases in foraging are less apparent with greater resource abundance. More resources and less competition over breeding leads male mice to show a greater overlap of smaller ranges (Chambers et al., 2000; Singleton and Krebs, 2007), more similar to the behavior of female mice, who can benefit reproductively from traveling shorter distances and overlapping ranges with other breeding females (Auclair et al., 2014; Schubert et al., 2008; Weidt et al., 2014). As we noted in the previous section, the sex mechanisms likely to influence these kinds of average sex differences can vary quite a bit between individuals of the same sex category. Therefore, a nonbinary interpretation of these data is that evolution largely constrains a species to a general foraging strategy, with sex mechanisms flexibly promoting cognitive diversity to allow for changes in strategy as environmental conditions change, or internal conditions change. Recognizing the evolutionary importance of maintaining flexibility in neural circuits regardless of sex influences brings us to our second practice for interpreting sex differences without relying on a sex binary.

3.2. Suggestion two: consider examining how much variance sex contributes, and whether sex impacts are mosaic

Because sex is multidimensional, sex influences on multidimensional data are likely to be variable and mosaic within and across individuals. The preceding sections of this paper highlight the necessity not assuming that average differences across sexes reflect binary or dichotomous phenotypes between females and males. However, for Suggestion One, we limited ourselves to discussing a primary outcome measure and the extent to which sexes cover a tradeoff, or spectrum, of behavior on the single dimension of that measure. However, when researchers identify sex differences in cognitive tasks, behaviors, or in the brain, they usually report more than one dependent measure, often many more, reflecting the multidimensional aspects of cognitive and behavioral testing and neural measures. There is an unexamined assumption that the sex differences in one measure correlates with a sex difference in another measure, reflecting a concept of sex as a binary. A related assumption is if a sex difference is identified, sex contributes the largest amount of variance between individuals. The implication of these assumptions is that knowing an individual’s sex can allow you to predict which data points belong to them (Maney, 2016). However, this assumption is not well founded. In multidimensional datasets, a sex difference in one measure has been found to be uninformative about a sex difference in any other measure, and variance contributed by sex differences are overwhelmed by large individual differences from other sources. This can lead to the discovery of the dimensions linked with the greatest variability between individuals neural and behavioral measures regardless of sex category.

We can identify broad sex categories in mammals based on genitalia. Because of this simple classification, we can ask more easily about the contributions of sex categories on cognition and executive functions than we can about other unknown variables driving individual differences. However, sex is not the only contributor of individual variability, which can be large, even in highly controlled animal models. In fact, the influence of sex can be far outweighed by individual subject variability on both motor sequence behavior (Levy et al., 2023) and on decision strategy sequences (Chen et al., 2021a). In both of these manuscripts, behaviors of female mice were highly individualistic compared to one another, but females were less likely to differ in their behavior from themselves in the past, compared to males compared to their past selves. When multidimensional datasets are considered, it becomes possible to ask a number of questions about the relationships between measures in the dataset. First, one can ask how much of the variation between individuals is contributed by sex. Using principal components analysis, we have found that sex differences in decision strategies were a significant component, but outweighed by a first component reflecting overall preference for one kind of choice (Chen et al., 2021a). Likewise, effects of estrous cycle on motor behavior were far outweighed by an individual mouse’s preferred motor sequences (Levy et al., 2023). Second, one can ask whether variation within an individual is a sex difference; as noted above, both manuscripts found evidence of higher internal variability in actions generated by a given male compared to a given female. Third, one can ask whether individuals are “sex typical” in their behaviors, or whether the behavior of individuals reflects a mosaic influence of sex.

The “sex as mosaic” concept has been forwarded (Joel, 2021) as an alternative to a sex binary framework. It emerges from a series of observations primarily from human neural structure analyses, but which could easily be extended to behavioral research. Sex differences in neural architectures, including the sizes of brain areas, have been identified across many neuroimaging studies in humans (Joel, 2021; Maney, 2016), but how “typical” is any individual compared to the summed average for their sex? Although average sex differences can be identified, no individual in these datasets can be found that represents all, or even most, average phenotypes for their sex. This is a special case of a longstanding observation that although we can create statistical averages for any group, no “average” individual for that group, with mostly average measurements, ever exists (Daniels, 1952; Molenbroek, 2022; Perez, 2019). This implies that any individual categorized into a sex category is in fact very unlikely to display “extreme phenotypes” in multiple neural dimensions, which was forwarded for some time as a theory for sex biases in neurodevelopmental diagnoses (Baron-Cohen et al., 2005). The implications of the above work for studying cognition and executive function across species is that different dependent measures such as reward obtained, correct responses, reaction times, and even strategies may be partially dissociable from each other, and the individuals who drive a sex influence on one measure may not be the individuals who drive a sex influence on a different measure. In addition to providing evidence about the suitability of a sex mosaic framework in decision making research, this approach can serve to highlight those mechanisms for regulating executive functions that are dissociable from each other as potential therapeutic targets. For example, in our own work we have assumed that sex differences in reaction time are driven by sex differences in exploratory action selection (Chen et al., 2021b), consistent with an assumed tradeoff between speed and accuracy of responding. However, recent work suggests that speed and accuracy in decision tasks can be co-regulated, (Steverson et al., 2019; Wolf and Lappe, 2023) opening the possibility that the mechanisms governing these decision variables may be influenced by different biological factors. We are excited to evaluate our past and future work to determine the extent to which a mosaic pattern of sex is evident in our decision making data.

4. Conclusion

The potential for sex differences in behavioral and cognition, including executive function, are of strong interest to many researchers and the general public alike. There are numerous scientific justifications for pursuing this work, including the potential role for sex mechanisms in influencing neuropsychiatric risk and resilience. The excitement over new discoveries of sources of variation in brain and behavior linked with sex factors can lead to emphasizing binary framings of sex differences, such as by calling them “dichotomies” despite substantial overlap in phenotypes, as pointed out by (Becker et al., 2005). When we conceptualize sex differences as a binary, we not only do a disservice to the multidimensional complexity of sex but we also risk setting arbitrary limits on what individuals are capable of thinking and doing. A sex label alone cannot capture this complexity (Miyagi et al., 2021; Perfors et al., 2023) just as an average does not fully describe it (Joel, 2021). We advocate for adopting nonbinary frameworks into sex differences research on cognition to allow science to represent the full spectrum of brains and behaviors.

Data availability

No data was used for the research described in the article.