1 INTRODUCTION

For many decades, the majority of biomedical knowledge was based on studies of males, leading to major disparities in our understanding of disease etiology, symptom presentation, treatment strategy, and clinical response in females. In 1993, the National Institutes of Health (NIH) in the United States implemented the Revitalization Act (https://orwh.od.nih.gov/resources/pdf/NIH-Revitalization-Act-1993.pdf), which mandated that females must be included in NIH-funded clinical trials. Unfortunately, sex biases in findings from basic and preclinical research persisted in both human and animal studies. A 2009 study of sex biases in animal research revealed that 80% of all animal studies examined male rodents only, across eight different scientific fields (Beery & Zucker, 2011), with the strongest male biases in neuroscience and pharmacology. Male animals still dominate the biomedical animal literature, particularly in cardiovascular research—a field with known sex differences in health risks, symptom presentation, and treatment response (Ramirez et al., 2017).

Sex disparities in human biomedical research have begun to be addressed in recent years, but they are still understudied. A recent bibliometric analysis of over 11 million papers (Sugimoto, Ahn, Smith, Macaluso, & Larivière, 2019) outlined the severity of sex biases across various fields of medicine, with the vast majority of studies neglecting to report sex characteristics of the sample. Of the disciplines studied, psychiatric and neurological studies reported sex in approximately 80% and 65% of research papers, respectively. The poorest sex reporting came from pharmacology studies, where only 24% of papers disclosed sex characteristics. More disturbingly, greater sex reporting was found in publications in lower impact journals over time, meaning lower visibility for papers that may adequately address sex effects (Sugimoto et al., 2019). To address these persisting biases, the NIH mandated that all grant proposals must address sex as a biological variable (NOT-OD-15-102, http://grants.nih.gov/grants/guide/notice-files/NOT-OD-15-102.html), even for basic and preclinical research. NIH investigators are now required to provide a detailed plan to analyze equal numbers of males and females, or provide sufficient justification if sex distributions would be unequal (Clayton, 2018). Early reports suggest this mandate has improved sex reporting and awareness of its importance in clinical and preclinical studies (Lee, 2018; Zucker & Beery, 2019), but decades of biased work still dominate the literature and continue to be published. For example, a systematic review of 1,827 neuroscience papers published in 2017 (25% human studies) showed that 44% of studies included both males and females but did not consider sex as an experimental variable,1 26% were male-only, 16% did not report sex, 8% included males and females and did consider sex as an experimental variable, 5% were female-only, and 1% were hermaphrodites (Mamlouk, Dorris, Barrett, & Meitzen, 2020). These numbers indicate that only 8% of studies adhered to the NIH mandate implemented 1 year prior (~57% were NIH-funded) (Mamlouk et al., 2020).

Previous preclinical studies justified sex imbalances by claiming that females introduced too much variability into research designs due to hormonal fluctuations along the oestrous cycle (Sugimoto et al., 2019; Wang, ; Zucker & Beery, 2010). However, there is considerable evidence from animal studies showing that variability in behavioral, biological, and molecular end points is consistent between females and males (Sugimoto et al., 2019). Further, a meta-analysis of 293 rodent studies revealed greater variability in males than females on indices of hormones, metabolism, and morphological traits (Prendergast, Onishi, & Zucker, 2014). The female reproductive cycle also has been used to justify explicit recruitment biases against women in clinical research. For example, pregnant women were considered a vulnerable population that was “protected” (i.e., excluded) from clinical research until the U.S. Department of Health and Human Services (HHS) amended the Federal Policy for the Protection of Human Subjects (“Common Rule”) in 2018 https://www.hhs.gov/ohrp/regulations-and-policy/regulations/45-cfr-46/index.html (Biggio Jr, 2020). The amended policy (implemented in 2019) was a response to the U.S. Task Force on Research Specific to Pregnant Women and Lactating Women, which stated that pregnant women are fully capable of making medical decisions for themselves and their fetus, and that the term “vulnerable” restrained the right to autonomy (Costantine, Landon, & Saade, 2020; Heyrana, Byers, & Stratton, 2018). Research exclusion of pregnant women also led to gross knowledge gaps in treatments and interventions that can be safely administered during pregnancy. While it is too soon to tell whether the new amendment will reduce explicit research biases against pregnant women, the overwhelming exclusion of pregnant women in studies of COVID-19 suggest that these biases are still in play (Costantine et al., 2020). Finally, implicit sex biases are also well-documented in healthcare and clinical research and contribute to lower participation of women in clinical trials (Chadwick & Baruah, 2020; Chapman, Kaatz, & Carnes, 2013; Daugherty et al., 2017; Hansen et al., 2019; Kannan et al., 2019; Salles et al., 2019).

The current state of knowledge on sex differences in human brain structure is sparse given the established sex differences in disease prevalence rates, age of onset, and symptom patterns for many psychiatric and neurodegenerative conditions. In this review, we will illuminate these gaps in the human neuroimaging literature and discuss commonly used methods and design strategies that miss opportunities to sufficiently address the role of sex in brain health and disease. To provide mechanistic context for the purported sex differences in various brain disorders and conditions, we first provide a brief overview of the function and trajectory of the primary sex hormones and their associations with neuroimaging indices. We then discuss normative sex differences in brain structure and function from population-based samples, as these differences are preserved in many clinical conditions and should not be interpreted as an outcome of disease. We chose to review both psychiatric and neurodegenerative diseases, as many psychiatric conditions are risk factors for neurodegenerative diseases and dementia later in life (Ahearn et al., 2020; Almeida, Hankey, Yeap, Golledge, & Flicker, 2017; Diniz et al., 2017; Gimson, Schlosser, Huntley, & Marchant, 2018; Kørner, Lopez, Lauritzen, Andersen, & Kessing, 2009; Kuring, Mathias, & Ward, 2020; Mrabet Khiari et al., 2011; Ribe et al., 2015; Singh-Manoux et al., 2017; Truelsen et al., 2002; Yaffe et al., 2010; Zilkens, Bruce, Duke, Spilsbury, & Semmens, 2014), and are associated with abnormalities in similar neuroimaging measures. Importantly, psychiatric disorders typically have an earlier age of onset than neurodegenerative conditions, with peak prevalences and symptom severities during the critical decades (ages 20–40) when neurodegenerative pathologies are seeded (Jones, 2013; Kessler et al., 2007; Zilkens et al., 2014). Thus, understanding the role of sex on psychiatric neuroimaging phenotypes may inform etiologic mechanisms of neurodegenerative disease and dementia, and help in the development of novel interventions.

Our literature review was conducted using a combination of search terms in PubMed, Google Scholar, and bioRxiv. Search terms to query information about sex included “sex,” “gender,” “males,” “females,” “men,” and “women” in conjunction with comparison terms such as “differences,” “biases,” “disparities,” and “confounds.” Additional phrases included “sex-specific,” “sex-by-age interaction,” “sex-by-diagnosis interaction,” “sex covariate,” “adjusting for sex,” “nuisance covariate,” “sexually dimorphic,” “biological sex,” and “genetic sex.” We prioritized the most recent publications first and gradually expanded our search in 2-year increments; all years were searched if terms yielded no hits in the last decade. As this is not a systematic review and our paper covers 14 distinct neurological and psychiatric conditions (in addition to normative studies), it was not possible to account for every published MRI study on sex effects or every imaging modality within these published studies. Instead, we prioritized large-scale studies from biobanks (e.g., UK Biobank), consortia (e.g., ENIGMA), and systematic reviews that focused on traditional structural MRI (e.g., volumetrics, thickness, surface area) and diffusion MRI (e.g., diffusion tensor imaging [DTI] scalar metrics) outcomes when available. Small-scale studies were reviewed when large studies of sex effects had not been conducted, were inconclusive, or contradicted other work. For conditions with a considerable literature on sex effects on neuroimaging (e.g., multiple sclerosis [MS], Alzheimer's disease [AD]), we prioritized studies that contributed to key themes or findings related to sex and neuroimaging. A summary of the studies reviewed is provided in Table S1 in the Supplementary Material. Although many of the conditions in this review may have developmental origins in childhood and adolescence, this literature is highly complex and has been reviewed in detail in other work (Deak et al., 2015; Earls, 1987; Schwarz & Bilbo, 2012). As such, we elected to focus on sex differences in the adult brain to adhere to page constraints and avoid an overly exhaustive and redundant review. Finally, we specifically use the term “sex” to refer to biological differences between males and females, rather than socially constructed roles that vary across time and cultures (i.e., “gender”). Although sex and gender continuously interact to influence the human healthspan, there is limited empirical data that demonstrates these interactive effects on brain structure in the context of major brain diseases. Addressing these research gaps is a central goal of the newly formed ENIGMA Gender Studies and Transgender Working Groups.

2 ROLE OF PRIMARY SEX HORMONES IN THE BRAIN 2.1 Developmental influences and lifespan trajectories

In humans, the impact of sex differences on health and disease begins as early as 50 days postconception when sex is determined through a cascade of genetic interactions beginning with the sex-determining region of the Y chromosome (SRY gene, chromosome 9) (Mamsen et al., 2017). When SRY transcripts are present, they initiate sexual differentiation of bipotential gonads by activating the SRY-box 9 gene (SOX9) approximately 50 days postconception (Mamsen et al., 2017). SOX9 regulates the transcription of anti-mullerian hormone along with other male-specific genes that promote androgen biosynthesis and the development of male sex organs. In the absence of SRY, female reproductive genes (WNT4, RSPO-1, FOXL2) promote the development of ovaries and inhibit differentiation of the testis, resulting in estrogen and progesterone synthesis. The divergent mechanisms of primary sex hormones are enacted, in part, by the location of hormone synthesis and the function and regional distribution of target receptors. Biosynthesis of sex steroids primarily occurs in male and female reproductive organs, but they are also synthesized in the brain and other tissues de novo from cholesterol (Hu, Zhang, Shen, & Azhar, 2010). Many additional genetic and epigenetic factors have been identified in sex differentiating pathways, as detailed elsewhere (Mamsen et al., 2017; Rotgers, Jørgensen, & Yao, 2018).

Sex hormones are believed to have organizational (permanent) and activational (dynamic) effects that impact disease manifestation, timing, and neuropathological progression, with associated changes in brain structure and function (Herting & Sowell, 2017; Schulz & Sisk, 2016). Organizational effects of sex hormones are “permanent” effects related to sexual differentiation and development that occur during the perinatal period (Arnold & Breedlove, 1985) and puberty/adolescence (Schulz, Molenda-Figueira, & Sisk, 2009; Schulz & Sisk, 2016). During these periods, sex hormones are believed to impart lasting effects on brain structure through complex gene-biology interactions that influence dendritic spine growth, synaptogenesis, synaptic patterning, and pruning (Arnold & Breedlove, 1985; Herting & Sowell, 2017; McCarthy, 2008; Schulz et al., 2009; Schulz & Sisk, 2016). These organizational effects are implicated as underlying mechanisms of the “developmental origins” hypothesis—a widely accepted theory linking early life experiences to adult disease (McCarthy, Arnold, Ball, Blaustein, & de Vries, 2012). By contrast, activational effects of sex hormones refer to transient and dynamic effects of sex hormones that occur throughout life after neural circuits have been organized, typically during adulthood (Herting & Sowell, 2017; Schulz & Sisk, 2016).

The developmental trajectory of primary sex hormones is dynamic in both males and females. In males, testosterone levels rise during the perinatal period, reaching peak levels 1–3 months after birth. Afterward, testosterone levels decline sharply until they plateau around 7–12 postnatal months; levels sharply rise again at puberty and plateau again around age 17 (Forest, Sizonenko, Cathiard, & Bertrand, 1974; Johannsen et al., 2018; Mason, Schoelwer, & Rogol, 2020; Senefeld et al., 2020; Tomlinson, Macintyre, Dorrian, Ahmed, & Wallace, 2004). Estrogen levels in males remain low in early childhood and then modestly rise around age 8 until they peak between ages 16 and 18 years (Frederiksen et al., 2020). In females, estrogen levels rise shortly after birth until approximately age 1, when estrogen plateaus until puberty (~age 10). During puberty, estrogen levels increase until ages 15–16 (Bidlingmaier, Wagner-Barnack, Butenandt, & Knorr, 1973; Frederiksen et al., 2020). Testosterone levels in females remain low throughout infancy and childhood, and modestly increase around age 6 until around age 14 (Søeborg et al., 2014). At the onset of puberty, females experience monthly fluctuations in estrogen and progesterone according to the menstrual cycle, as detailed in the sections below. Deviations from these normal sex-specific hormone trajectories can permanently alter structural brain development and increase vulnerability to disease acutely and many years later (McCarthy, 2008; Pike, 2017).

Testosterone levels remain fairly stable throughout adulthood in both males and females (Handelsman, Sikaris, & Ly, 2016), with modest decline in males that begins around the fifth decade (Feldman et al., 2002; Harman et al., 2001). Testosterone levels are also higher in males than females throughout the lifespan (Rothman et al., 2011). Females experience more dynamic changes in primary sex hormones (estrogen, progesterone) during adulthood due to menstruation, pregnancy, and menopause (Del Río et al., 2018). Estrogen levels are higher in premenopausal females than males, but estrogen levels between males and females become similar after females experience menopause (Handelsman et al., 2016; Nugent et al., 2012; Rothman et al., 2011).

2.2 Testosterone

Testosterone is known for promoting muscle and bone growth, healthy libido, mood, and social behaviors such as aggression, competitiveness, and risk taking (Campbell et al., 2010; Casto & Edwards, 2016; Eisenegger, Haushofer, & Fehr, 2011; Tyagi, Scordo, Yoon, Liporace, & Greene, 2017; Walther, Wasielewska, & Leiter, 2019). The effects of testosterone on brain and behavior occur by binding to androgen receptors in the forebrain, midbrain, and brainstem, with the highest concentrations in the ventromedial hypothalamus, medial preoptic area, nucleus accumbens, basal nucleus of the stria terminalis, and septum (Davey & Grossmann, 2016). Neuroimaging investigations relating testosterone levels to brain structure are limited in healthy adults, but suggest a link between circulating testosterone and frontal-temporal brain integrity. Specifically, a structural MRI study of healthy young adults (N = 34, 50% female; ages 21–47, Mage = 26.6 ± 5.0) reported a negative association between testosterone levels and gray matter volume in the left inferior frontal gyrus (IFG) after adjusting for sex and total gray matter volume (Witte, Savli, Holik, Kasper, & Lanzenberger, 2010). However, testosterone only explained 2.2% of total model variance compared to 32% and 47.2% explained by sex and gray matter volume, respectively. Sex-stratified analyses did not show significant associations between testosterone and gray matter volumes, likely due to low statistical power from the small sample size and limited explanatory effect of testosterone on gray matter volume in the whole sample. More recently, a study of hippocampal volume in the Vietnam Era Twin Study of Aging cohort (N = 445 males, ages 51–60) showed that effects of salivary free testosterone (unbound to a receptor) on hippocampal volume differed based on a person's cortisol levels (Panizzon et al., 2018). Specifically, Panizzon et al. (2018) found that the effect of free testosterone on hippocampal volume was only significant when cortisol levels were >1 SD above or below the mean, such that hippocampal volumes were larger in individuals with high testosterone and high cortisol, but smaller in individuals with low testosterone and low cortisol. These associations were observed after covarying for age, ethnicity, twin pair, current alcohol use, depression, smoking status, and a history of cardiovascular disease, hypertension, and diabetes, and after correcting for lack of independence in the sample (i.e., twin clustering) (Panizzon et al., 2018).

2.3 Estrogen and progesterone

Estrogen and progesterone are important opposing sex hormones that fluctuate significantly across the female menstrual cycle. Estrogen impacts a wide range of positive biological functions beyond the reproductive system, including maintenance of bone mineral density, regulation of antioxidant defense systems and mitochondrial oxidation, maintenance of blood vessel structure and vascular tone, and enhanced neuron survival (Prabhushankar, Krueger, & Manrique, 2014; Ventura-Clapier, Piquereau, Veksler, & Garnier, 2019). At the same time, estrogen is associated with increased production of the stress hormone, cortisol, upregulation of excitatory neurotransmitters (acetylcholine, dopamine) and downregulation of the inhibitory neurotransmitter, GABA (Barth, Villringer, & Sacher, 2015). These latter biological effects of estrogen increase vulnerability to adverse psychosomatic symptoms under certain conditions.

Estrogen exerts its effects by binding to G-coupled receptors that activate second messenger systems and to intracellular estrogen receptor alpha (ER-α) and beta (ER-β) to modulate transcription (Fuentes & Silveyra, 2019). ER-α and ER-β, respectively, modulate the excitatory and inhibitory effects of estrogen, and both receptors are located throughout the limbic system, midbrain, and brainstem to regulate the stress response in the preoptic area, arcuate and lateral habenula, periaqueductal gray, locus coeruleus, and in nuclei of the amygdala, hypothalamus, pons, and medulla oblongata (Weiser, Foradori, & Handa, 2008). However, only ER-α is found in the ventromedial nucleus of the hypothalamus and subfornical organ, whereas only ER-β is found in the olfactory bulb, zona incerta of the subthalamus, ventral tegmental area, cerebellum, pineal gland, and hypothalamic nuclei of the supraoptic (SON), paraventricular, suprachiasmatic, and tuberal areas (Weiser et al., 2008). The distinct distribution of ER-α and ER-β in these brain regions allows for separate interactions with neurotransmitter systems to facilitate target functions.

Progesterone is a derivative of the hormone, pregnenolone, in both males and females. Although it is most commonly known for its role in female physiology, progesterone also facilitates sperm capacitation, fertilization and immunosuppression in both sexes (Maybin & Critchley, 2011; Oettel & Mukhopadhyay, 2004). In the brain, progesterone helps to maintain the structural integrity of myelin and regulates synaptogenesis, neuron survival and dendritic growth (Schumacher et al., 2012). Neural functions of progesterone occur primarily through membrane-associated progesterone receptors in various brain regions including the hippocampus, amygdala, olfactory bulb, cortex, cerebellum, locus coeruleus, hypothalamus, thalamus, basal ganglia, and brainstem (Schumacher et al., 2012).

Estrogen and progesterone levels fluctuate throughout the female menstrual cycle, and these fluctuations have been associated with changes in mood, concentration, somatic sensations, and brain structure (Catenaccio, Mu, & Lipton, 2016). The follicular phase and luteal phase are the two primary phases of the menstrual cycle. Menstruation generally occurs between days 1 and 8 of the follicular phase, during which estrogen rises until reaching peak levels at the start of the peri-ovulatory period of the follicular phase (~days 11–12). Ovulation, which typically occurs around Day 14 of the menstrual cycle, marks the transition from the follicular phase to the luteal phase, and a significant drop in estrogen levels. Simultaneously, progesterone levels begin to rise during the luteal phase (progesterone levels are low during follicular phase), peaking at the start of the pre-menstrual period (~days 21–22) and declining rapidly thereafter (Catenaccio et al., 2016; Maybin & Critchley, 2011). A recent review of 25 neuroimaging studies (N = 1,321) (Catenaccio et al., 2016) described the brain signatures that corresponded to these phases. Four studies reported larger volumes during the follicular phase than the luteal phase, with the most consistent effects in the hippocampus, parahippocampal and middle frontal gyri. Five studies reported larger volumes during the luteal phase than the follicular phase using a mix of voxel-based morphometry (VBM) and region of interest (ROI) methods, but none reported effects in consistent regions. Three studies reported no volume differences between follicular and luteal phases. A few studies compared neuroimaging metrics (VBM, volumetry, ROI-based) between the menstrual and peri-ovulatory periods, but none reported consistent effects in any regions (Catenaccio et al., 2016). It is worth noting, however, that the sample sizes of each of the 25 reviewed studies was small (largest N = 128, mean N ~ 32) and effect sizes were not reported in any systematic way.

2.4 Exogenous sources of estrogen and progesterone

Studies of hormone contraceptive use provide naturalistic designs to understand relationships between hormonal fluctuations and brain health. These studies are important as approximately 25% of premenopausal females use oral hormone contraceptives or long-acting reversible contraceptives (Daniels & Abma, 2020). There is increasing recognition that hormone contraceptives can influence short and long-term changes in brain structure. Lisofsky, Riediger, Gallinat, Lindenberger, and Kühn (2016) showed that even brief use of oral contraceptives (the pill) can result in structural brain changes in regular cycling females (ages 16–35) using VBM. Specifically, female contraceptive users (n = 28) showed volume reductions in the left amygdala and anterior parahippocampal gyrus (PHG) compared to age-matched controls (n = 28) after 3 months of daily contraceptive use and after adjusting for age, intracranial volume (ICV), and total estrogen and progesterone levels. These regions are key hubs for emotion processing and regulation and may explain affective changes associated with contraceptive use (Montoya & Bos, 2017). However, individuals on oral contraceptives were not on the same pill type regimen, and it is unclear how different pill types (estrogen + progestin vs. progestin-only, etc.) may have influenced these results.

Our group recently expanded on this work in the UK Biobank cohort by examining the impact of oral contraceptive use on whole-brain white matter in premenopausal and postmenopausal females using DTI (Nabulsi & Lawrence, 2020)—a noninvasive imaging technique that measures patterns of water diffusion throughout the brain (Basser & Pierpaoli, 1996). The most common diffusion MRI metric is fractional anisotropy (FA), which measures the degree of diffusion restriction within an image voxel. In a trajectory analysis using fractional polynomial modeling, Nabulsi and Lawrence (2020) revealed higher whole-brain FA and tensor distribution function (TDF, a more rigorous multitensor diffusion model that accounts for intravoxel fiber crossing) FA in contraceptive users compared to never-users (n = 7,136 users, n = 1,177 nonusers; ages 45–80 years) after adjusting for age, years of education, socioeconomic status (SES), waist-to-hip ratio, and genetic ancestry. Conversely, longer duration and younger age at first contraceptive use were associated with lower FA and TDF-FA compared to never users. As higher FA is typically a good indicator of “healthy” white matter (Basser & Pierpaoli, 1996; Bennett, Madden, Vaidya, Howard, & Howard Jr, 2010), these results suggest that brief oral contraceptive use may serve a protective role for white matter microstructure, but chronic use—particularly at older ages—may be associated with a faster rate of white matter decline in older adulthood.

In addition to contraceptive use, hormone replacement therapy (HRT) may have neuroprotective effects when administered during perimenopause (Eberling, Wu, Haan, & Mungas, 2003; Shao et al., 2012). Specifically, earlier studies suggested that HRT implemented early in menopause (Eberling et al., 2003; Shao et al., 2012) or over long periods may attenuate risk of AD diagnosis (Imtiaz et al., 2017) and death (Mikkola et al., 2017) in females. Indeed, neuroimaging work by Pintzka and Håberg (2015) showed that females who initiated HRT prior to menopause and remained on HRT for at least 3 years (n = 80) had greater whole hippocampal volume compared to HRT-naive females (n = 80) who were matched for ICV and age (ages 51–66). A recent voxelwise study by Boyle et al. (2020) also revealed larger gray and white matter volume in various regions in females with a history of HRT (N = 562, ages 71–94) after adjusting for data acquisition site, age, ethnicity, years of education, clinical diagnosis, history of heart disease, Type 2 diabetes, and hypertension, presence of white matter lesions, BMI, physical activity, and past-year estrogen use, but the duration of HRT was not associated with imaging variables.

Other work suggests that HRT does not prevent cognitive decline (Henderson et al., 2016) and that long-term HRT may slightly increase risk for dementia (Savolainen-Peltonen et al., 2019; Wu et al., 2020). Indeed, a randomized, double-blinded placebo controlled trial of HRT (specifically, conjugated equine estrogens) in 95 recently postmenopausal women (ages 42–56) showed decreases in whole brain volume, increased ventricle expansion, and increased white matter hyperintensity volume over 48 months of HRT compared to the placebo group. However, cognitive performance did not differ between groups (Kantarci, Tosakulwong, et al., 2016). Finally, Nabulsi & Lawrence, 2020] earlier mentioned work on contraceptive use also examined the impact of HRT on brain white matter in the UK Biobank using both DTI and neurite orientation dispersion density imaging (NODDI). Results revealed lower whole-brain white matter fiber coherence (orientation dispersion index) in females on HRT (N = 3,106) compared to never-users (N = 5,195), but less pronounced white matter changes with age (Nabulsi & Lawrence, 2020). Further, females on an estrogen-only regimen showed greater white matter disruption with age (lower fiber dispersion, and increased free water).

The inconsistent neuroimaging markers of HRT emphasize that there is much to learn regarding the therapeutic potential of exogenous estrogen use. A major limitation of previous work is the variability in HRT chemical composition used across individuals, even within the same study. Prior work on HRT has compared studies using various hormonal combinations (e.g., estrogen, androgen, progesterone treatment alone or together), chemical formulations, doses, and delivery routes that exert different effects on the body (Comasco, Frokjaer, & Sundström-Poromaa, 2014; Maki & Dumas, 2009; Moraga-Amaro, van Waarde, Doorduin, & de Vries, 2018; Yare & Woodward, 2020). In fact, estrogen and progesterone have various molecular and binding mechanisms that modulate neurotransmitter activity, sometimes in opposing directions (Del Río et al., 2018). This variability and lack of consensus of HRT effects on brain structure is a major gap in the literature that requires immediate attention given the growing population of older adults and the high percentage of females taking HRT worldwide (Boyle et al., 2020).

3 NORMATIVE SEX DIFFERENCES IN HUMAN BRAIN STRUCTURE 3.1 Brain size metrics and gray matter

Neuroimaging has been particularly informative for understanding how sex differences in brain size can be explained by differences in underlying gray and white matter microstructure. At the outermost level, ICV remains an important index of maximal brain size across the lifespan, as the intracranial vault is set at approximately age 7 and is not susceptible to developmental or degenerative processes (O'Brien et al., 2006). Sex differences in ICV have been established in both large and small-scale studies. On average, males have significantly larger ICV than females (Ruigrok et al., 2014), and these differences have been shown to account for some, but not all, regional sex differences in brain volumes (Jahanshad & Thompson, 2017). The literature also shows that total brain volume (TBV)—measured on T1-weighted brain scans—is approximately 9–12% larger in males than females in childhood, adolescence, and adulthood in both large and small scale studies (for review, see Kaczkurkin, Raznahan, & Satterthwaite, 2019; Lenroot & Giedd, 2010). This aligns with very early evidence that brain weight is approximately 10% larger in males than females at autopsy (Pakkenberg & Voigt, 1964; Voigt & Pakkenberg, 1983).

Similarly, sex differences in gray matter measures are well-established across the lifespan. In young adults (ages 20–34 years), VBM has revealed larger total normalized gray matter volume (total gray matter volume/total ICV) but also a faster rate of age-related decline in normalized gray matter volume in females (n = 71) than males (n = 71) (Farokhian, Yang, Beheshti, Matsuda, & Wu, 2017). Yang, Bozek, Han, and Gao (2020) revealed sex differences in several different cortical gray matter features, including sulcal depth and cortical thickness, in young adults (ages 19–37) from the Chinese and U.S. Human Connectome Project (HCP) cohorts using sample-specific surface templates based on FreeSurfer segmentations from 35 Desikan–Killiany (DK) and 75 Destrieux atlas structures in each hemisphere. In the Chinese HCP, males (n = 100) exhibited greater cortical thickness than females (n = 100) in the frontal, temporal, and parietal lobes after correcting for age and ICV, whereas in the U.S. HCP, males (n = 100) exhibited lower thickness than females in the caudal anterior cingulate cortex (ACC), but greater thickness in the insula, lateral orbitofrontal cortex (OFC), and the isthmus of the cingulate than females (n = 100).

The Enhancing Neuroimaging Genetics through Meta-Analysis (ENIGMA) consortium has begun to chart neuroimaging sex differences across the lifespan to provide a benchmark for evaluating individual brain health and improve disease detection and monitoring (Dima, Papachristou, Modabbernia, & Doucet, 2020; Frangou, Modabbernia, Doucet, Moser, et al., 2020; Wierenga et al., 2020). Recent mega-analyses of subcortical volumes and cortical thickness by the ENIGMA Lifespan Working Group has shown that while unadjusted overall cortical volume and thickness are larger in males than females, sex differences do not persist after covarying for ICV (Dima et al., 2020; Frangou, Modabbernia, & Doucet, 2020). Trajectory analyses, however, do show significant sex differences in the rate of change in cortical thickness across the lifespan. Specifically, Frangou, Modabbernia, and Doucet (2020) showed that males (n = 8,212), on average, had faster whole-brain cortical thinning than females (n = 8,863) during midlife (ages 30–59), but cortical thinning rates were comparable between males and females in early life (ages 3–29) and in older adulthood (ages 60–90). Males also had faster regional cortical thinning than females in motor, somatosensory, and visual association cortices during early life (ages 3–29), and in frontal-temporal cortical areas during midlife (Frangou, Modabbernia, & Doucet, 2020). Of note, sex was the only covariate used for trajectory modeling, as the imaging data were harmonized using the ComBat package in R, which adjusts for site and scanner-related variance.

In more recent work from the ENIGMA-Lifespan group, Wierenga et al. () examined cortical and subcortical brain metrics in 16,683 individuals between ages 1 and 90 years (47% females) using the ENIGMA FreeSurfer (Fischl, 2012) cortical and subcortical pipelines http://enigma.ini.usc.edu/protocols/imaging-protocols/ (ENIGMA, 2017). Results revealed sex differences both in brain structure metrics and in between-subject variability in brain metrics after adjusting for cohort, magnetic field strength, FreeSurfer version, and age. Specifically, males had larger volumes than females in all subcortical ROIs (Cohen's d range = 0.41 [left accumbens] to 0.92 [right thalamus], and these differences persisted with slightly smaller effect sizes after covarying for TBV (d range = 0.21 [left accumbens] to 0.58 [right thalamus]). Males also had greater regional surface area than females across the entire cortex with and without adjusting for total surface area (without: d range = 0.42 [left ACC] to 0.97 [left superior temporal gyrus, STG]; with: d = 0.21 [left ACC] to 0.59 [left STG]). Females had greater thickness than males in 38 of the 68 DK atlas-defined cortical regions, but effect sizes were comparatively small (largest effect, d = 0.12 in the right caudal ACC), and effects in several regions changed direction (males > females) or became nonsignificant when total thickness was included as a covariate. However, males showed significantly greater between-subject variability than females for all subcortical volumes and cortical surface area metrics, and for 60% of cortical thickness metrics, and these differences persisted throughout the lifespan (Wierenga et al., 2020).

Population-based studies of middle-aged and older adults (ages 45–80 years) from the UK Biobank cohort generally align with the findings from the ENIGMA-Lifespan group. Specifically, Ritchie et al. (2018) reported greater cortical surface area in males (n = 2,466) than females (n = 2,750) in most regions after adjusting for TBV, age, and ethnicity. Conversely, females had greater thickness than males across most of the cortex, except the medial OFC and rostral ACC, which was thicker in males. Females also had larger volumes in the nucleus accumbens compared to males after adjusting for TBV, whereas males had larger volumes in the putamen, amygdala, and pallidum after adjusting for the same covariates (Ritchie et al., 2018). Other work in the UK Biobank revealed significant age-by-sex interaction effects in subcortical volume trajectories, such that males (n = 12,665) exhibited faster decline than females (n = 13,775) in all volumes across the full age range (44–81 years) after adjusting for ICV, education, and BMI (Ching et al., 2020). Interestingly, sex differences in volume loss were significantly attenuated after age 60, suggesting that the greatest sex effects on subcortical volumetry may occur during middle age.

3.1.1 Hippocampal volume

The hippocampus is a complex subcortical brain structure that serves as an essential integratory hub for memory (formation, storage, retrieval), spatial navigation, and emotional processing. As sex differences are reported in these functions, numerous imaging studies have investigated hippocampal volume in clinical and non-clinical populations (Yagi & Galea, 2019). Consistent with population-based studies from ENIGMA and the UK Biobank, a meta-analysis of hippocampal volumes in healthy participants from 76 studies (N = 4,418, ages 0–79 years, 45.3% females) showed that raw hippocampal volumes were larger in males than females by ~6–7%, but statistical adjustments for ICV or TBV nullify these sex differences (Tan, Ma, Vira, Marwha, & Eliot, 2016). Similarly, nomograms (percentile charts) of whole hippocampal volume (computed with FSL-FIRST) in 19,793 individuals from the UK Biobank (ages 45–80 years, 52.9% females) showed similar hippocampal volume measurements by sex after adjusting for age, scan date, and an automated metric of head size derived from the nomogram pipeline (https://lnobis.github.io/HippoFit_Tool/index.html) (Nobis et al., 2019). However, trajectory analyses revealed significant sex differences in the rate and temporal change of hippocampal volume loss with age, with accelerated loss in males around age 50 and accelerated loss in females between ages 60 and 65. The rate of hippocampal volume loss relative to total gray matter volume also differed in males and females, with peak inflection points around ages 63 and 67, respectively (Nobis et al., 2019).

Finally, while sex differences in total hippocampal volume do not persist after correcting for head size, prior work suggests that subregions of the hippocampus are sexually dimorphic. In a recent lifespan study of hippocampal subfield volumes manually segmented from 4.7 T scans (N = 129, ages 18–85, Mage = 47.6 ± 18.9), Malykhin, Huang, Hrybouski, and Olsen (2017) found larger subfield volumes in the hippocampal head (dentate gyrus [DG]), body (CA1-3, subiculum, DG), tail (all subfields), and DG in females (n = 70) than males (n = 59) after normalizing hippocampal volumes by ICV (raw hippocampal volume/ICV of same subject × sample averaged ICV) and removing the effects of age. Sex-by-age interactions were significant in the subiculum of the hippocampal tail, with a marginally significant nonlinear effect of age on subiculum volumes in females, but no significant effect in males (Malykhin et al., 2017).

Interestingly, a larger study of young adults (ages 21–36) from the Queensland Twin Imaging study (QTIM, 4 T scanner) and HCP (3 T scanner) revealed larger subfield volumes (segmented with the FreeSurfer-v.6.0 hippocampal subfield pipeline) in males (n = 692) than females (n = 995) in the fimbria, parasubiculum, fissure, and presubiculum after statistically adjusting for whole hippocampal volume (van Eijk et al., 2020). Importantly, sex differences persisted across four different statistical methods to control group differences in whole hippocampal volume: (a) allometric scaling—regresses out the effect of whole hippocampal volume (or comparable metric) after identifying the scaled relationship between whole hippocampal volume and each subfield via log–log regression, (b) covariate—models the effect of whole hippocampal volume as a covariate predictor in regression or analysis of covariance (ANCOVA), (c) residuals—regresses out the effect of whole hippocampal volume on the subfield ROI and uses the residuals as the dependent variable, and (d) matched—where groups are matched by whole hippocampal volume. When subfields were adjusted for brain segmentation volume (a FreeSurfer defined metric of total brain size that includes gray matter, white matter, and CSF) rather than whole hippocampal volume, males again had larger volumes than females in the hippocampal fissure, presubiculum, and parasubiculum using all four correction methods, albeit at smaller effect sizes for subicular subregions. Males also had larger volumes than females in the fimbria and subiculum using covariate, residual, and matching methods, but not when using allometric scaling. Sex differences were not detected in the CA2/3, CA4, hippocampal–amygdala transition area (HATA), or DG using any normalization technique or covarying for whole hippocampal volume or brain