Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

Contents lists available at ScienceDirect

Neuroscience and Biobehavioral Reviews

journal homepage: www.elsevier.com/locate/neubiorev

Review article

Beyond the binary: Rethinking sex and the brain

Daphna Joel
School of Psychological Sciences and Sagol School of Neuroscience, Tel-Aviv University, Tel-Aviv, Israel

A R T I C L E I N F O A B S T R A C T

Keywords: The paper reviews the relations between sex and brain in light of the binary conceptualization of these relations
Brain structure and the challenges posed to it by the ‘mosaic’ hypothesis. Recent formulations of the binary framework range
Magnetic resonance imaging (MRI) from arguing that the typical male brain is different from the typical female brain to claiming that brains are
Hypothalamus
typically male or female because brain structure can be used to predict the sex category (female/male) of the
Mosaic
Male-female continuum
brain’s owner. These formulations are challenged by evidence that sex effects on the brain may be opposite under
Typical female brain different conditions, that human brains are comprised of mosaics of female-typical and male-typical features, and
Typical male brain that sex category explains only a small part of the variability in human brain structure. These findings led to a
Stress new, non-binary, framework, according to which mosaic brains reside in a multi-dimensional space that cannot
Human meaningfully be reduced to a male-female continuum or to a binary variable. This framework may also apply to
Rat sex-related variables and has implications for research.

1. Background: how the binary framework affects the brain that show sex/gender differences (i.e., these measures do not
conceptualization of the relations between sex and the brain appear in distinct female and male forms, reviewed in Joel, 2011). Yet,
in spite of the fact that most scientists nowadays acknowledge this
“The problem with the sex binary is that there has never been a hypothesis overlap and would not argue that brains of males and females belong to
or a theory to test— it is an epistemological framework that runs behind, two distinct types, the binary framework still dominates thinking about
above, and beyond particular theories and research projects” Sanz (2017, the relations between sex and the brain, and the ‘male brain - female
p. 20). brain’ or ‘typical male brain - typical female brain’ terminology still
prevails. These terms, however, may have different meaning for
When we talk about female and male genitalia, we have quite a clear different scientists.
and agreed-upon understanding of what this means – two distinct sets of Some scientists hold that there is a typical male brain which is
organs, one comprised of only genital organs with a form typical (i.e., distinct from the typical female brain. This is often evidenced in phrases
common) of females, and the other comprised of only genital organs of the sort - male brains are like this, female brains are like that – as in:
with a form typical of males. Genitalia that do not fall into one of these “During developmental periods, male brains tend to be structured to
distinct sets, because of having either one or more genital organs with a facilitate within-lobe and within-hemisphere connectivity… In contrast,
form intermediate between the female- and male-typical forms, or some female brains tend to have better interhemispheric connectivity and
genital organs with the female-typical form and others with the male- better cross-hemispheric participation…” (Tyan et al., 2017, p. 380).
typical form, are termed intersex, rather than ‘male’ or ‘female’. Esti­ Other scientists assume that human brains are aligned along a
mates of the prevalence of humans with intersex genitalia typically do male-female continuum, yet still hold that the typical female brain is
not exceed 0.2 % (on the basis of Table 8 in Blackless et al., 2000). different from the typical male brain. This hypothesis underlies, for
This is clearly not the case in the human brain, in which, if we were example, the extreme male brain theory of autism (e.g., Baron-Cohen,
to apply the terminology used to describe genitals, most brains would be 2002), as evident in this citation: “to examine the probability of autism
‘intersex’. This is because there is overlap between the distributions of spectrum disorder along a normative phenotypic axis ranging from the
females and of males on all currently known measures of the human characteristic female to male brain phenotype” (Ecker et al., 2017, p.

E-mail address: djoel@tauex.tau.ac.il.

https://doi.org/10.1016/j.neubiorev.2020.11.018
Received 6 December 2019; Received in revised form 12 November 2020; Accepted 14 November 2020
Available online 10 January 2021
0149-7634/© 2021 The Author. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

330). on the existence of internal consistency. According to this new formu­

Common to these descriptions of the typical female and male brain is lation, “brains are indeed typically male or typically female” (Rose­
the implicit assumption that different features within a single brain nblatt, 2016, p. E1966) because brain structure can be used to predict
would be similarly located along each feature’s male-female continuum whether the brain’s owner is female or male (Chekroud et al., 2016; Del
(i.e., all features would be located at the male-end1 of their distribution, Giudice et al., 2016; Rosenblatt, 2016). As I explain in Section 6.3, this
or all would be located at the female-end, or all would be located in- definition of a typical male and female brain is very different from the
between the two extremes, Fig. 1a). If this were the case, then indeed previous ones and from our understanding of male and female genitalia.
brains would be aligned along a female-male continuum, with the Below I shortly summarize the results of animal studies that led to
typical female brain different from the typical male brain (Fig. 1b). Yet, the mosaic hypothesis and human studies that support it. I then discuss
in 2015 we found that ‘mosaic’ brains – that is, brains consisting of a the validity and usefulness of the different formulations of the binary
mixture of features – some located at the male-end of their distribution view in light of the evidence. Next, I suggest a new, non-binary,
and others located at the female-end, are much more common than framework for thinking about human brains, according to which
internally consistent brains consisting of only one type of features2 mosaic brains reside in a multi-dimensional space that cannot mean­
(Fig. 1c–d; see Section 5 for more details on this analysis). On the basis of ingfully be reduced to a male-female continuum or to a binary variable
this finding we concluded that brains of women and of men do not (male and female brain). Last, I shortly discuss the mosaic nature of sex-
belong to two distinct categories nor aligned along a female-male con­ related variables, suggest that these too reside in a multi-dimensional
tinuum (Joel et al., 2015). space that cannot meaningfully be reduced to a binary variable, and
To illustrate the importance of the internal consistency assumption discuss the implications of these conclusions for research.
for the binary formulations of the relations between sex and the brain
described above, consider two studies of human brain connectivity. Both 2. Evidence leading to the mosaic hypothesis: sex effects on
studies reported the existence of sex/gender differences in some con­ brain structure may be opposite under different conditions
nections (Ingalhalikar et al., 2014; Joel et al., 2015), but only one
explicitly assessed the presence (or lack of) internal consistency (Joel According to the classical view of sex effects on the brain, the brain
et al., 2015). The first study (Ingalhalikar et al., 2014) concluded that undergoes sexual differentiation, with testosterone, secreted by the
“This analysis revealed conspicuous and significant sex differences that fetus’ testes, masculinizing the brain of the male away from the default
suggest fundamentally different connectivity patterns in males and females” female form. According to this view, all the features within a single brain
(p. 824, emphasis added) and pointed the reader to Fig. 2 (reproduced that are affected by testosterone are expected to be similarly located
here as Fig. 1e). This figure depicts in the upper pair of brain images, all along their female-male continuum. Moreover, brains are expected to be
the connections that were significantly stronger in men compared to located along a female-male continuum, depending on each brain’s
women, and in the lower pair of brain images, all the connections that testosterone levels during development.
were significantly stronger in women compared to men (blue here This scenario, however, is highly unlikely given what is currently
stands for intra-hemispheric connections, and orange for known about sex effects on the developing (and mature) brain (for a
inter-hemispheric connections). Although this was never explicitly detailed review and examples, see Joel et al., 2020). Thus, while animal
stated, this type of presentation and the accompanying conclusion sug­ studies provided plenty of evidence that testosterone affects multiple
gest that the authors’ underlying assumption is that the differences aspects of brain structure, brain structure is also influenced, in both
consistently add up within individual brains, so that the upper images males and females, by other sex-related hormones and by sex-related
represent the connectivity pattern typical of males - having male-typical genes (for reviews see Arnold, 2012; Arnold and Chen, 2009; Grgur­
connections, while the lower images represent the connectivity pattern evic and Majdic, 2016; McCarthy and Arnold, 2011; McEwen and Mil­
typical of females - having female-typical connections. While this may ner, 2017; Ngun et al., 2011; Sekido, 2014). This is expected to lead to
seem to be a reasonable assumption (and is true of genitalia), it was higher variability in the ‘femaleness-maleness’ of different features
refuted by an analysis of internal consistency (Joel et al., 2015). An within a single brain than the one expected in the case of a single factor.
analysis of the seven connections showing the largest sex/gender dif­ Moreover, sex-related hormones, including testosterone, act on different
ferences (out of over 4000 connections assessed), revealed that none of brain features via multiple independent mechanisms, so that even fea­
the brains was internally consistent (i.e., had all seven connections in the tures affected by the same hormone may vary considerably in their
female-end range, or all in the male-end range, both defined with the 33 location along their female-male continuum (for review see Joel and
% cutoff) but about half of the brains were mosaic (had at least one McCarthy, 2017; McCarthy and Arnold, 2011; McEwen and Milner,
connection, out of the seven analyzed, in the female-end range, and at 2017). Finally, at least some of the effects of sex-related genes and
least one connection in the male-end range; for the results with addi­ hormones may be opposite under different external conditions, suggest­
tional cutoffs see Table S2 in Joel et al., 2015). Thus, the brain images in ing that within a single brain, some features may not only be poorly
Ingalhalikar et al. (2014) do not represent the connectivity patterns correlated in their ‘femaleness-maleness’, but located at opposing ends
typical of males and females, but rather connectivity patterns that are of each feature’s female-male continuum (for review and references see
nonexistent or extremely rare. Instead, the connectivity pattern common Joel, 2011, 2012; Joel et al., 2020).
in both women and men is a mosaic of blue connections (which are more At the group level, the result of this multitude of mechanisms and
common in men than in women) and orange connections (which are interactions is that the form3 of a particular brain feature that is typical
more common in women than in men). of females under one set of conditions (e.g., individual housing) may be
In response to Joel and colleagues (2015), a new version of the bi­ typical of males under another set of conditions (e.g., group housing),
nary view of the human brain was formulated - one that does not depend and vice versa (i.e., the form typical of males under the first set of
conditions may be typical of females under the second set of conditions).

1
As shown in Fig. 1a, the female-end and the male-end correspond to the two
3
extremes of the distribution, where there are large differences between the ‘Form’ here may relate to the size of a brain region, the morphology of
frequencies of females and males. Data reviewed below relates to an operational neurons, the density of receptors, or any other measure of brain structure. If, for
definition of the female- and male-end zones as the scores of the 33% most example, a region is larger, on average, in males compared to females, then the
extreme females and males, respectively. form typical of males would correspond to a volume-range in which more males
2
This conclusion was true over operational definitions of the female- and than females fall, and the form typical of females would correspond to another
male-end zones with cutoffs of 10%, 20%, 33%, and 50%. volume-range, in which more females than males fall.

166
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

Fig. 1. Internal consistency versus mosaic. a.

Grey matter volume of 116 regions in the brains
of 169 women and 112 men were assessed using
voxel-based morphometry (VBM). The figure
presents the frequency distribution of the grey
matter volume in women (red) and men (green)
of two of the regions showing the largest sex/
gender differences in this sample (left hippo­
campus [top, Cohen’s d = 0.74, p < 0.0001] and
left caudate [bottom, Cohen’s d = 0.84, p <
0.0001]). b. A schematic representation of in­
ternal consistency – if all regions in a brain are
similarly located at each feature’s female-male
continuum, then brains would be similarly
located along a female-male continuum, ranging
from a brain in which all features are at the
female-end (pink) to a brain in which all fea­
tures are at the male-end (blue). c. Ten regions
from the data set described in 1a, that showed
the largest sex/gender differences (0.70 < d ≤
0.84, all p’s < 0.0001) were chosen for the
mosaic analysis. d. A continuous color repre­
sentation of each region’s location along its
male-female continuum was created separately
for each of the 10 regions. Volumes falling in the
“male-end” and in the “female-end” zones
(defined as the scores of the 33 % most extreme
males and females, respectively) are colored
using continuous blue-white and pink-white
scales, respectively; Volumes falling in the “in­
termediate” zone are colored in white. In the
tables, each horizontal line represents the brain
of one individual and each column represents a
single brain region. The number above each
column corresponds to the region’s number in
1c. (a and d were created with permission on the
basis of Fig. 1 in Joel et al., 2015; For details of
the sample, imaging methods, and data analyses
see Joel et al., 2015). e. A summary diagram of
sex/gender differences in brain connectivity.
The top pair of brain images depicts connections
that were significantly stronger in men
compared to women, and the bottom pair of
brain images depicts connections that were
significantly stronger in women compared to
men. Intra-hemispheric connections are shown
in blue, and inter-hemispheric connections are
shown in orange. (Reproduced with permission
from Fig. 2 in Ingalhalikar et al., 2014). f. A
schematic summary of the results of Reich et al
(2009). The density of CB1 receptors that is
typical of females and males that are kept in
standard laboratory conditions is marked in pink
and blue, respectively. In male rats that were
exposed to chronic stress, the density of the re­
ceptors in both the ventral and dorsal hippo­
campus is the one typical of non-stressed
females. In female rats that were exposed to
chronic stress, the density of the receptors in the
ventral hippocampus is the one typical of
non-stressed females, whereas the density of
receptors in the dorsal hippocampus is the one
typical of non-stressed males.

At the individual level, the multiplicity of mechanisms and interactions The general principles described above may be illustrated by the
is expected to result in brains comprised of unique combinations of results of a single study, which assessed the effects of three weeks of mild
features that greatly vary in their location along each feature’s female- stress on the density of CB1 cannabinoid receptors in the rat hippo­
male continuum, leading to brains consisting of a mosaic of both campus (Reich et al., 2009): In rats kept under standard laboratory
female-end and male-end features. Moreover, these mosaic brains would conditions, the density of CB1 receptors was on average 3–4 times
not be meaningfully aligned along a male-female continuum (Joel, higher in males compared to females in both the ventral and dorsal
2011, see also Section 6.2). hippocampus. Following the stress exposure, the effects of sex in the

167
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

Fig. 2. Mosaic in the human brain response

to stress. The volume of 68 cortical regions and
26 subcortical structures was assessed in 34
participants (15 women) using surface-based
analysis. Each participant was scanned twice:
Once during their first week of a premilitary
paramedic preparation course, and again 36
months later. During their military service, all
participants experienced at least one highly
stressful event, which was similarly accompanied
in women and men by intense negative emotions
and an increase in stress-related symptoms. a.
Seven regions, in which the most commonly
observed (i.e., mode) change (increase, decrease,
or no change) in volume in women was different
from the most commonly observed change in
men, and the Sex x Time interaction for the re­
gion was significant. b. The tables present for
each women (left) and each men (right) whether
the change in each brain region was the one
typical of women (pink), of men (blue) or of
neither (white). Each horizontal line represents
the brain of one individual and each column
represents a single brain region. The number
above each column corresponds to the region’s
number in 2a. (Created with permission on the
basis of Fig. 1 and Table 2 in Shalev et al., 2020;
For more details of the sample, imaging methods,
and data analyses see Shalev et al., 2020).

dorsal hippocampus were reversed – the average receptor density of the and dorsal hippocampus would be in the range typical of females in this
stressed female rats was as high as that observed in non-stressed males, sample. In females, receptor density in the dorsal hippocampus would be
whereas the average receptor density of the stressed males was as low as in the male-typical range whereas receptor density in the ventral hip­
that found in non-stressed females. In the ventral hippocampus the ef­ pocampus would be in the female-typical range. Thus, in terms of the
fects of sex on CB1 receptor density were again reversed in males, but density of CB1 receptors in the hippocampus, these females would
were unaffected in females (leading to the disappearance of the exhibit a mosaic of female-typical and male-typical features (Fig. 1f).
group-level sex difference that was observed under the no-stress con­ Taking into account that interactions between sex and other factors
dition) (Reich et al., 2009). have been reported for additional brain measures (e.g., spine density,
Thus, at the group level, for both the ventral and dorsal hippocam­ number of neurons), brain regions (e.g., amygdala, cortex, cerebellum),
pus, the density of CB1 receptors that is typical of females and of males and types of manipulation (e.g., housing conditions, drug exposure, for
depends on an external factor (exposure to stress). Moreover, consid­ review see Joel, 2011, 2012, 2020) and moving from considering two
ering the density of CB1 cannabinoid receptors in the dorsal and ventral brain features and two environmental conditions to considering the
hippocampus together, the hippocampus could be found in one of three entire brain and the huge complexity of the environment from the
forms: low receptor density in both the ventral and dorsal hippocampus moment of conception throughout life, it is difficult to imagine that
(in non-stressed females and in stressed males); high receptor density in brains would be internally consistent in the ‘sex-typicality’ of their
both the ventral and dorsal hippocampus (in non-stressed males); and different features. Instead, the mosaic hypothesis holds that most brains
high receptor density in the dorsal hippocampus and low receptor would consist of unique mosaics of features - some in the form typical of
density in the ventral hippocampus (in stressed females). These three the females in that sample and others in the form typical of the males in
forms of the hippocampus cannot be meaningfully sorted into a male- that sample - and that these mosaics would not fall into two distinct
typical and a female-typical form, nor be meaningfully aligned along a types nor be meaningfully aligned along a male-female continuum (Joel,
female-male continuum. 2011).
This example also demonstrates how the interactions between sex I would like to stress that the mosaic hypothesis does not hold that
and other factors may lead to the formation of a mosaic brain at the sex does not affect the brain or that there are no group-level sex dif­
individual level. Consider for example a sample of rats, all of which are ferences in specific brain features. Rather, the mosaic hypothesis holds
kept under standard laboratory conditions. Most males would exhibit that the multiplicity of mechanisms by which sex affects the brain
high CB1 receptor density in the dorsal and ventral hippocampus, combined with the repeated observation that sex-related effects depend
whereas most females would exhibit low receptor density in the two on other factors, result in brains with features that greatly vary in their
hippocampal regions (Fig. 1f). A few rats in the sample may be exposed location along each feature’s male-female continuum.
to stress (because, for instance, they were unintentionally housed with a One prediction of the mosaic hypothesis would therefore be that
dominant and aggressive rat). These rats would exhibit sex-atypical mosaicism would be greater under conditions of greater genetic and
features - in males, the density of CB1 receptors in both the ventral environmental variability. Thus, little mosaicism is expected in an

168
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

inbred strain of laboratory animals kept under the same external con­ men, we found that 25 out of the 34 participants exhibited a mosaic of
ditions from utero, more mosaicism is expected in an out-bred strain of female-typical and male-typical structural changes, whereas in only one
laboratory animals kept under different sets of external conditions, and a participant all changes were of the same type (Fig. 2, created with
lot of mosaicism is expected in wild-type animals living in the wild. permission on the basis of Fig. 1 and Table 2 in Shalev et al., 2020).
A note on mosaic and variance. Variance in biological systems is
always expected. There is large variability in the form of genital organs 4. A note on sex effects and sex differences
within females and within males. But variance in sex effects on the brain
differs from variance in sex effects on the genitalia, in that the former Studies in which sex-related genes or hormones are directly manip­
may be so large as to lead to the existence of female-typical and male- ulated demonstrate that sex affects brain structure and function (for
typical features in the same brain - a situation that is very rarely reviews see Arnold, 2012; Arnold and Chen, 2009; Grgurevic and Maj­
observed in human genitalia. It is evidence for the existence of this type dic, 2016; McCarthy and Arnold, 2011; McEwen and Milner, 2017; Ngun
of variance that led to the formulation of the mosaic hypothesis and to et al., 2011; Sekido, 2014). Yet, most of the evidence for sex effects on
the construction of methods to test it. (See more on mosaic versus noise the brain derives from studies reporting a difference between a group of
in Section 5.2.1). females and a group of males on some endpoint(s) (e.g., regional vol­
ume, receptor density). While such studies show that sex-related vari­
3. Mosaic in the brain’s response to external events ables affect the endpoint, they do not suffice to identify the variable(s)
responsible for this effect, nor even to provide information on whether
The observation that various manipulations alter how sex affects the these variables are part of “sex itself” (i.e., sex-related genes and hor­
brain means that the effects of these manipulations on the brain are mones, Richardson, 2013), are affected by sex-related genes or hor­
different in females and males (e.g., stress decreased the density of CB1 mones (e.g., body size), or are correlated with sex category (e.g., single
receptors in the dorsal hippocampus in males, but increased it in fe­ versus group housing) (for a detailed discussion of the direct and indi­
males, Reich et al., 2009). Indeed, most studies that reported in­ rect effects of sex, see Joel and McCarthy, 2017). This problem is
teractions between sex and other factors framed their results as sex intensified in studies of the human brain, as many more variables
differences in the effects of the other factor (e.g., the title of the study by (environmental, psychological and social) correlate with sex category
Reich et al. (2009) described above is: “Differential effects of chronic (e.g., Fausto-Sterling, 2000; Fine, 2010; Joel and Fausto-Sterling, 2016;
unpredictable stress on hippocampal CB1 receptors in male and female Joel and McCarthy, 2017; Jordan-Young and Rumiati, 2012; Kaiser,
rats”). Could it thus be that the brains of females and males are distinct 2012; Maney, 2015; Rippon et al., 2014). I therefore refrain from using
not in their structure but rather in their response to environmental the term ‘sex effects’ when discussing the human brain, and use instead
conditions (e.g., there’s a female-typical and a male-typical neural ‘sex/gender differences’.
response to stress)? This would be the case if all the features in a single
brain responded to an environmental event (such as stress) in the way 5. Mosaic in human brain structure
typical of females or all responded in the way typical of males. This
would not be the case, however, if in an individual brain, some features To assess whether sex differences add up consistently or ‘mix’ to
would change in the way typical of males while others would change in create mosaics, one has to consider at least two measures showing sex
the way typical of females – that is, if the response of each brain con­ differences for each brain. Below I describe two studies that tested the
sisted of a mosaic of female-typical and male-typical changes. Such mosaic hypothesis in the human brain – one used postmortem data of the
‘mixing’ of responses would occur if the way in which a brain feature type often assessed in laboratory animals (namely, the number of neu­
responds to an environmental event depends not only on sex, but on an rons in two hypothalamic nuclei, Joel et al., 2020); the second used
interaction between sex and other factors. different types of measures obtained from MRI studies of the entire brain
Animal studies are seldom designed in a way suitable for answering (Joel et al., 2015).
this question - that is, they rarely test the effects of a specific manipu­
lation (e.g., stress) under different conditions (e.g., individual versus 5.1. Mosaic in the human hypothalamus: analysing post-mortem data
group housing) in females and males. The results of one study, which
was designed this way, suggest that the effects of a manipulation on We (Joel et al., 2020) have recently co-analyzed three hypothalamic
females and males may depend on other factors. Horovitz et al. (2014) measures that show large sex/gender differences - differences that are
assessed the behavioral effects of stress experienced in adulthood in amongst the largest known to date in the human brain. Specifically, we
male and female rats that were either exposed to stress early in life or assessed mosaicism in the total number of neurons in the interstitial
not. Thus, it was possible to appreciate whether sex differences in the nucleus of the anterior hypothalamus, subdivision 3 (INAH3, a
response to stress experienced in adulthood depend on other factors – in sub-nucleus of the uncinate nucleus), and in the number of
this case, early exposure to stress. Horovitz et al. (2014) found that at the galanin-stained and non-galanin stained neurons in the INAH1 (also
group level, the early exposure to stress interacted with sex to determine called sexually dimorphic nucleus or intermediate nucleus) (Garcia-­
the average response to stress experienced in adulthood. At the indi­ Falgueras et al., 2011; Garcia-Falgueras and Swaab, 2008). There was
vidual level, while the behavioral response to adulthood stress that was relatively little overlap between the distribution of scores for women
typical (i.e., common) of females exposed to early stress was different and for men in each of the three measures - the probability that a man
from the one typical of males exposed to early stress, there were some picked at random will have a higher score than a woman picked at
females and males that exhibited the response typical of the other sex. random (Del Giudice, 2019) was 0.88, 0.74 and 0.73, respectively. This
These observations suggest that additional factors, that were not allowed the delineation of a male-typical and a female-typical range of
measured or manipulated in Horovitz et al’s (2014) study, interacted scores (a range of scores which are very common in men but rare in
with sex to determine an individual’s response to stress, and that women, or are very common in women but rare in men, respectively4),
mosaicism may also occur in sex-related responses to stress. and subsequently the assessment of mosaicism within each brain. We
The possibility that mosaicism may also be seen in sex-related re­
sponses to stress was recently supported by a small-scale magnetic
resonance imaging (MRI) study in humans exposed to real-life extreme 4
For example, in the INAH3, the female-typical and the male-typical range of
stress (Shalev et al., 2020). Considering seven regions (listed in Fig. 2) in scores were defined as scores below and above 2,000 neurons, as 82% of the
which the change in volume that was most common in women (increase, women in the sample had fewer than 2,000 neurons and 93% of the men had
decrease or no change) was different from the change most common in more than 2,000 neurons (Joel et al., 2020).

169
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

found that even when considering only three brain measures each at the other extreme, namely, the female-end zone (where, on average,
showing a large sex/gender difference, about half the brains contained a only 17 % of males fall). Smaller variance that results in some features
mixture of female-typical and male-typical measures – a proportion falling at the male-end (or female-end) and all others falling at the in­
significantly higher than expected if brains were internally consistent termediate range (where, on average, ~50 % of males and females fall,
(Joel et al., 2020). Fig. 1a), would not be classified as mosaic (nor as internally consistent).
I want to reiterate that the mosaic hypothesis was built on the basis
5.2. Mosaic in the human brain: analyzing MRI data of the observation that sex effects on brain features may be opposite
under different conditions, and that the interactions of sex with other
We analyzed MRI data of over 1400 brains from four datasets (the variables may be different for different brain features. The mosaic
analysis of one dataset is described in Fig. 1). Because most MRI-derived analysis was specifically constructed to detect this type of variability –
brain measures show no or only small sex differences, we analyzed only created by features located at opposite ends of their male-female con­
a few (7–12) brain measures in each dataset - those showing the largest tinuum – and ignore variability due to random noise. Using simulations,
sex/gender differences (for example, in the analysis of the dataset we have shown that the pattern of results (i.e., the number of internally
described in Fig. 1, the Cohen’s d of the sex difference in the regions consistent brains and of mosaic brains) obtained by the mosaic analysis
included in the analysis ranged between 0.70 to 0.84). In addition, of human brain measures is different from the one expected were these
because the overlap between women and men in brain measures ob­ measures internally consistent but noisy (see Fig. S1 in Joel et al., 2015).
tained from MRI data is much greater than that observed in the human Finally, I would like to point out that the mosaic analysis is more
hypothalamic measures described above, a female-typical and a male- sensitive than correlation coefficients in detecting internal consistency
typical range of scores could not be defined even for the measures and in differentiating between mosaicism and noise. Clearly, if the
showing the largest sex/gender differences (because the scores common correlation coefficient between two variables is very low (as was the
in women are also common in men, and vice versa, e.g., Fig. 1a). We case for most correlations between the hypothalamic measures
therefore defined for each of these measures a female-end and a male- described above, Joel et al., 2020), then the two variables are not
end range of scores, each at one extreme of the distribution, where internally consistent (and a mosaic analysis would reveal many mosaic
there are large differences between the frequencies of females and males brains and very few internally consistent brains, see, Fig. S1E in Joel
(Fig. 1a). For example, with the female-end and the male-end ranges et al., 2015). Similarly, if the correlation coefficient between two vari­
defined as the scores of the 33 % most extreme females and males, ables is near 1, then the two variables are internally consistent (and a
respectively, the average percent of males with a female-end score and mosaic analysis would reveal some internally consistent brains and no
of females with a male-end score was 17 % (thus, the chances of falling mosaic brains, see, Fig. S1A in Joel et al., 2015). However, high corre­
at the end zone of the other sex were half the chances of falling at the end lation coefficients between variables (i.e., in the range of 0.7− 0.8) may
zone of one’s sex; Joel et al., 2015). We then assessed whether brains reflect either an internally consistent system with some degree of
were internally consistent (i.e., all the measures fell in the male-end random noise (Fig. S1B and S1C in Joel et al., 2015) or a system with no
zone, or all fell in the female-end zone) or mosaic (at least one mea­ underlying internal consistency (the simulated data in Del Giudice et al.,
sure fell in the male-end zone and at least one measure fell in the 2016). The mosaic analysis can differentiate between the two possibil­
female-end zone, Fig. 1d). ities – in the former case there would be more internally consistent
We found that regardless of the sample, the MRI-derived measure brains than mosaic brains, whereas in the second case, the opposite
analyzed (volume, cortical thickness or connectivity), or the male-end - would be true (for further discussion see Joel et al., 2016 and the Sup­
female-end cutoff (50 %, 33 %, 20 % or 10 %), mosaic brains were more plementary Material of Joel et al., 2015).
common than internally consistent brains (depending on the sample,
with a cutoff of 33 %, the percent of mosaic brains ranged between 23 6. The validity and usefulness of the different formulations of
and 53, and the percent of internally consistent brains ranged between the binary view of the human brain
0 and 8.2; the remaining brains were comprised either of male-end and
intermediate features, or of female-end and intermediate features, Joel That human brains do not belong to two distinct types, the way
et al., 2015). (The results with cutoffs of 10 %, 20 % and 50 % can be human genitalia do, stems from the observation that mosaic brains are
found in Table S2 in Joel et al., 2015). common whereas internally consistent brains are rare (in contrast to
Clearly, the number of internally consistent and mosaic brains de­ human genitalia, where the opposite is true).
pends on the choice of cutoff. The more lenient the cutoff (i.e., more
participants are included in the male- and female-end ranges), the 6.1. Is the typical female brain different from the typical male brain?
higher the number of both internally consistent and mosaic brains.
Therefore the number of internally consistent brains or of mosaic brains But does the prevalence of mosaic brains also contradict the view
by itself is meaningless; it is the comparison between the two that is that the typical female brain is different from the typical male brain? We
important. The mosaic hypothesis is supported when mosaic brains are have recently tested this question using MRI data of over 2100 brains
more prevalent than internally consistent brains, whereas the reverse from two datasets, and concluded that brain architectures typical of
scenario suggests the existence of two distinct types (Joel et al., 2015, women are also typical of men, and vice versa (Joel et al., 2018; Note
2016). Indeed, a higher number of internally consistent faces than of that in these analyses we used all brain measures, not only the ones
mosaic faces was found in an analysis of the facial morphology of three showing the largest sex/gender differences, as we had done in the
primate species (i.e., distinct types, Del Guidice et al., 2016). (For a mosaic analysis). Specifically, if the typical female brain were different
summary and discussion of the criticism of the mosaic analysis, see Joel, from the typical male brain, we should expect an anomaly detection
2020; Joel et al., 2020). algorithm that was trained on women’s brains to mark many more
brains of men as anomalous compared to brains of women. Instead, the
5.2.1. Mosaic versus noise anomaly detection algorithm marked very similar numbers of men’s and
I want to stress that with the above definition of a mosaic, a brain is women’s brains as anomalous, suggesting that the brain architectures
considered a mosaic only if it shows large variability in the location of its typical of women are also common in men (Joel et al., 2018). Training
features on each feature’s male-female continuum. For example, with a the algorithm on men’s brains and then testing it on men’s and women’s
cutoff of 33 %, a brain of a male would be considered a mosaic only if at brains yielded the same result - the brain architectures typical of men are
least one brain measure (of the 7–12 analyzed in that dataset) fell at the also common in women (Joel et al., 2018). An unsupervised cluster
male-end zone (where 33 % of males fall), and at least one measure fell analysis supported this conclusion by showing that large clusters –

170
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

which represent common human brain architectures, include a similar there are no sex/gender differences in the brain or that these differences
number of brains from women and from men. Large sex/gender differ­ cannot be used to cluster brains according to sex category (Joel, 2011;
ences were found only in some of the small clusters, which represent rare Joel et al., 2016, 2018, see also Section 6.3).
brain architectures (Joel et al., 2018). (Although this has not been
tested, these small clusters could potentially account for sex/gender
differences in the prevalence of some neuro/psychiatric conditions, such 6.2. From a male-female continuum to considering mosaic brains in a
as autism, which are rare in the population but show large sex/gender multi-dimensional space
differences.)
The conclusion that the brain architectures typical of women are also The view that emerges from the two studies (Joel et al., 2015, 2018)
common in men, and vice versa, is consistent with the observation that is that, when human brains are described by the vector of their feature
when total brain size it taken into account, there are only few and mostly values (e.g., the volume of 116 regions of grey matter, Fig. 3a), human
small sex/gender differences in MRI-derived brain measures (e.g., brains constitute a cloud of points in a multi-dimensional space, with
Jancke et al., 2015; Sanchis-Segura et al., 2019) and sex/gender cate­ women and men sharing quite equally the dense central part, and
gory accounts for less than 2% of the variance in human brain structure differing in some of the sparser periphery. Even in the bivariate scat­
(Eliot, 2020). terplot of the two principal components that differentiate most between
Note that whereas the lack of large differences in the proportion of women and men (Fig. 3b), the overlap is all encompassing.
women and men in the large clusters indicates that sex category is less I suggest that this new multi-dimensional description should replace
important than other variables (such as age, Jancke et al., 2015) in the image of brains aligned along a male-female continuum (e.g., Bar­
explaining human variability in brain structure, it does not indicate that on-Cohen, 2002; Ecker et al., 2017; Phillips et al., 2019). The
male-female continuum may be useful for describing the distributions of

Fig. 3. From a male-female continuum to a multi-

dimension space. a. The grey matter volume of all
116 regions of grey matter in females (left) and in males
(right) from the sample described in Fig. 1a,c,d is rep­
resented using a continuous green (large) – white –
yellow (small) scale. Each horizontal line represents the
brain of one individual and each column represents a
single brain region. The continuous large-small scale
represents the volume of a brain region in a given brain
relative to the volume of this brain region in all other
brains (of both females and males). (Created with
permission on the basis of Fig. 3 in Joel et al., 2015). b.
A principal component analysis of the dataset presented
in (a) was conducted and the two principal components
that differentiate most between women and men were
selected. The graph presents the bivariate scatter gram
of the scores of brains of females (red) and males (blue)
on these two components. c. A bivariate scatter gram of
the number of regions (out of the ten regions listed in
Fig. 1c) at the “female-end” (x axis) and at the “mal­
e-end” (y axis) in females (red) and males (green) in
that sample (the actual mosaics can be found in
Fig. 1d). d. The difference between the number of
female-end characteristics and the number of male-end
characteristics was computed for each brain, and is
presented in a histogram. e. A histogram of the number
of all possible combinations of female-end, intermedi­
ate and male-end characteristics for each “female-end
minus male-end characteristics score”. f. Hypothetical
brains with different combinations of “male-end”
(blue), “intermediate” (white), and “female-end” (pink)
features.

171
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

women and men on a single brain feature (e.g., Fig. 1a), but fails to et al., 2019; Chekroud et al., 2016; Del Giudice et al., 2016; Joel et al.,
account for the observations obtained when several brain features (or 2016, 2018; Rosenblatt, 2016; van Putten et al., 2018v; Zhang et al.,
the brain as a whole) are considered together (Joel, 2020; Joel et al., 2018; Note, however, that Sanchis-Segura et al., 2020 showed that the
2015, 2018, 2020). Moreover, I claim that although it is mathematically accuracy of sex prediction on the basis of brain structure drops to around
possible to align brains on a male-female continuum (there are many ~60 % when total brain size is properly controlled for). Here, instead of
mathematical ways to achieve this, e.g., Phillips et al., 2019), such an reducing the information in the multi-dimensional space into a single
alignment would carry little information. dimension (a female-male continuum), it is reduced into a binary vari­
To illustrate this, take for example the set of brains shown in Fig. 3a. able – female or male. Clearly, this binary variable carries very little
A mosaic analysis of the ten brain regions showing the largest sex/ information about a person’s specific brain mosaic. It merely assures us
gender differences in this dataset (listed in Fig. 1c) reveals that most that had we known the structure of this person’s brain, we could have
women have more female-end features than male-end features, whereas guessed her/his sex category with high accuracy. Yet, it is a person’s sex
the opposite is true for men (Figs. 1d, 3 c). We can thus align the brains category that led us to predict that s/he has a female or a male brain in
on a female-male continuum by assigning each brain a score calculated the first place, so what kind of information have we gained from the
as the difference between the number of its female-end and male-end “prediction” definition of a male and a female brain? Knowing that
characteristics (Fig. 3d; a similar method has been used by Baron- someone is, say, male, gives you much more information about the form
Cohen and colleagues to align humans along a systemizer-empathizer of his genital organs than that you’re very likely to conclude he is male
continuum, e.g., Greenberg et al., 2018). On this continuum, brains had you seen his genitalia. The latter is of course true, but knowing that
with only female-end characteristics would be at one pole (+10), and someone is male allows you in addition to very safely predict that he has
brains with only male-end characteristics would be at the other pole a penis, scrotum, prostate and vas deference and surely does not have a
(-10). The two poles are thus well defined, having either all regions in clitoris, minor and major labia, vagina, fallopian tubes and uterus. The
the female-end form or all regions in the male-end form. Note that these studies cited in the present review clearly demonstrate that not only
poles do not represent typical male and female brains, but rather brain such a prediction is not possible for brain structure (because most brains
types that are extremely rare (Fig. 3c,d). In contrast, scores along the rest are unique mosaics of female-typical and male-typical features), sex
of the continuum, where most brains reside, are ill defined as they may category provides little information about the structure of an in­
include very different brain mosaics. Continuing with the ten regions dividual’s brain.
example, a brain with a score of +3 may be comprised of any of the Is it then worth maintaining the “prediction” formulation of the male
following four combinations of female-end, intermediate and male-end and female brain only for the sake of preserving a binary view of the
characteristics – [3,7,0], [4,5,1], [5,3,2], [6,1,3] – and each of these human brain?
combinations potentially includes many different mosaics, depending I do not think so.
on which of the ten regions is in which form. For example, the number of
potential mosaic brains with five female-end characteristics, three in­ 6.4. “Costs” of the binary view of the human brain
termediate characteristics and two male-end characteristics is 2520
(Fig. 3e; the actual mosaics observed among the 169 women and 112 Maintaining the binary framework interferes with our efforts to un­
men of this dataset are depicted in Fig. 1c). Thus, two mosaics with the derstand the human brain because it diverts us from studying other
same female-end - male-end score (e.g., Mosaics 1 and 2 in Fig. 3f) or variables, which may be more important in understanding the human
even the same number of female-end, intermediate and male-end brain in health and disease (e.g., Mitricheva et al., 2019). In addition,
characteristics (e.g., Mosaics 1 and 3 in Fig. 3f) may be very different the focus on sex differences often leads researchers and readers to
from one another, and more similar to other mosaics with a different overestimate their importance. The title of too many studies declares
female-end - male-end score (e.g., Mosaics 1 and 4 in Fig. 3f). More that females and males differ in brain structure, function or connectivity,
generally, it is the specific composition of a brain, not the difference whereas careful reading of the Methods and Results sections reveals that
between the number of its female-end and male-end features, that de­ of the hundreds or even thousands of variables assessed, a significant sex
termines whether it is similar to or different from other brains. Indeed, difference was found in only a few. For example, a recent study of
the unsupervised cluster analysis described above revealed that the functional connectivity in utero was titled “Sex differences in functional
chances of a woman and a man to be in the same cluster are very similar connectivity during fetal brain development”, even though there were
to the chances of two women or two men to be in the same cluster (Joel no sex differences in connectivity patterns, and of the 128 correlations
et al., 2018). between sex and age that were assessed, there were significant differ­
The above example demonstrates the type of information that is ences in only three (Wheelock et al., 2019).
being lost when information residing in a ten-dimension space (1c) is The binary view of the human brain and the accompanying practice
reduced to a single dimension (Fig. 3d). Moreover, brains have many of looking for sex differences may also send researchers chasing false
features in addition to those showing sex/gender differences (e.g., positive results. As I explain elsewhere (e.g., Joel, 2011, 2020; Joel and
Fig. 3a), and information about these features is also being lost when Fausto-Sterling, 2016), when the population is highly heterogeneous
brains are aligned along a male-female continuum. Given that sex/ and the samples are relatively small (as, for example, in functional MRI
gender accounts for a very small part of the variability in human brain studies), comparing two samples from this population (one of females
structure (Eliot, 2020) and probably also function (e.g., Kersey et al., and the other of males) is likely to yield some significant differences. But
2019; Mitricheva et al., 2019) it is clear why even though sex/gender these would not reflect genuine sex differences worth pursuing, but
differences may be used to align brains on a male-female continuum, rather false-positive errors. The results of a recent study support this
such alignment carries little information about an individual’s brain claim. David et al. (2018) assessed the relations between sample size and
structure. the number of significant sex/gender differences in human functional
MRI studies. If the “prediction”, or any other definition, of the typical
6.3. The “prediction” version of the binary view of human brains male and female brain were meaningful, then larger samples, which
have greater power, should have discovered more sex/gender differ­
This discussion brings us to the question of prediction, and specif­ ences. Yet no correlation was found between sample size and the number
ically to how does this new multi-dimensional view of human brains of significant sex/gender differences (David et al., 2018). The authors
reconcile with the repeated observation that the structure and function concluded: “The extremely high prevalence of “positive” results and the
of the brain can be used to predict with high accuracy (often 80 % or lack of the expected relationship between sample size and the number of
higher) whether the brain’s owner is female or male (e.g., Anderson discovered foci reflect probable reporting bias and excess significance

172
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

bias in this literature” (David et al., 2018). The most obvious implication of the conclusion of the previous
section, that sex-related variables reside in a multi-dimensional space
7. Mosaics of sex-related variables? that cannot meaningfully be reduced into a binary variable (female,
male), is replacing the common practice of comparing a group of females
The above discussion reveals that, contrary to the privileged position to a group of males with an attempt to associate sex-related variables
and coordinated action of sex-related variables in determining the form with the phenomenon under study. As has been previously noted (e.g.,
of the genitalia, these variables are a fragment of a large array of vari­ Ritz et al., 2014), finding a difference between females and males may
ables that interact to determine brain structure. These interactions result provide the first clue that sex-related variables are relevant, but it is only
in poor correlations between sex effects on different brain measures, and a first step in understanding the relations between sex and this phe­
as a result, sex category provides little information about a brain’s nomenon. Subsequent steps should attempt to detect the sex-related
specific structure. variable(s) that affect the studied phenomenon – an endeavour that
As mentioned above (Section 4), there are many variables that would require the assessment of many sex-related variables at the in­
correlate with sex category (i.e., variables on which females and males dividual level and the use of statistical tools that allow the detection of
differ at the group level) and many of these may affect brain structure complex interactions between variables (Joel and Fausto-Sterling, 2016;
and function. These sex-related variables include aspects of “sex itself”, Joel et al., 2020; Ritz et al., 2014).
that is, sex-related genes and hormones, as well as many physiological Practically, I suggest always including females and males in a sam­
(e.g., body size), psychological (e.g., empathy), and social/environ­ ple, to capture the entire variability of the studied species, human or
mental (e.g., status) variables (e.g., Fausto-Sterling, 2000; Fine, 2010; non-human (Joel, 2015; Joel and Fausto-Sterling, 2016; Joel and
Joel and Fausto-Sterling, 2016; Joel and McCarthy, 2017; Jordan-Young McCarthy, 2017). This is important in both basic and clinical research
and Rumiati, 2012; Kaiser, 2012; Maney, 2015; Rippon et al., 2014; Ritz and should be done regardless of whether sex differences in the studied
et al., 2014). These variables may be affected by sex itself, by gender endpoint(s) have previously been reported. Prior knowledge of the ex­
(that is, the social construction of the sex categories), by both, and by istence or lack of sex differences should, however, direct the researcher
many other variables (e.g., one’s height also depends on genetic and in designing the study and in deciding whether to use sex category as a
environmental (e.g., nutrition) variation not related to sex) (e.g., Joel variable in the analysis of the results. There are three possible scenarios:
and McCarthy, 2017; Joel et al., 2020). there are no or only a few prior studies on sex differences in the studied
As noted by Maney (2016), one’s sex/gender category provides very phenomenon; there is strong evidence for the lack of sex differences;
little information about a person’s specific values on all of these there is strong evidence for the existence of sex differences.
sex-related variables. This is particularly problematic because only the Y In the first scenario - no or only few relevant studies - one should
chromosome, the gonads and the genitalia appear in a binary form (e.g., assess sex differences as a crude way to evaluate the possible involve­
present or absent for the Y chromosome; ovaries or testes for the go­ ment of sex-related variables, which could be assessed in subsequent
nads). For all other sex-related variables, including sex-related hor­ studies. In studies in which sex category is used as a variable, one should
mones (for a recent review see, Hyde et al., 2019), there is overlap be careful with generalizations of the results across environmental
between the values observed in females and in males – overlap that is conditions, strains and species, because sex effects may be different
often considerable (e.g., Hyde, 2005, 2014; Hyde et al., 2019; Zell et al., under different genetic, developmental, or environmental conditions
2015). Moreover, there is little reason to believe that the many (Joel and Fausto-Sterling, 2016; Joel and McCarthy, 2017). Special
sex-related variables are highly correlated. For example, there is no a caution is required in generalizing the results of animal studies, in which
priori reason to believe that muscle to fat ratio is strongly correlated the variability in these other variables is often very limited.
with empathy or socioeconomic status, or even with height. Therefore, In the second scenario - strong evidence for the lack of sex differences
specifically because men and women differ on average on many vari­ - it may be best not to include sex category as a variable when analysing
ables, it is highly likely that most humans possess a mosaic of values on the results. This is because adding a variable that does not account for
these variables, with some values falling on their male-end of the dis­ variability in the endpoint(s) detracts from the study’s power to detect
tribution and others on their female-end. differences on other variables (because of the reduction in the degrees of
If sex-related variables reside in a multi-dimensional space that freedom).
cannot meaningfully be reduced into a binary variable (female, male), In the third scenario - strong evidence for the existence of sex dif­
then the current practice of studying sex mostly in the context of sex ferences - it would be wise to collect data on sex-related variables that
differences should be replaced with the measurement of sex-related may be relevant for the phenomenon under study, because simply
variables and the assessment of their associations with the phenome­ finding (again) a sex difference would not advance much our under­
non under study. standing of the phenomenon nor its relations to sex. A famous example
for the importance of going beyond sex differences to consider sex-
7.1. Implications for research and diagnosis: considering sex as a related variables is the case of zolpidem. The sex difference in the
biological variable drug’s clearance, which probably contributes to the higher rate of
adverse side effects in women, became non-significant when partici­
The exclusion of females from clinical trials as well as from many pants’ weight was taken into account (Greenblatt et al., 2014). Another
areas of basic research harmed not only the health of women, but also example is the cardiovascular system, which is affected by variables that
the advance of science and medicine (for several examples, see correlate with sex category, such as smoking, height and physical ac­
https://genderedinnovations.stanford.edu). The requests of the Na­ tivity. It is clearly better to ask a patient whether they smoke and how
tional Institute of Health (NIH) and other funding agencies to include much, then rely on their sex category and the average difference be­
women in clinical trials (NIH revitalization Act, 1993) and later to tween women and men in smoking to predict outcome or assign
consider sex as a biological variable in basic research (e.g., Clayton and treatment.
Collins, 2014) were necessary steps to correct this situation. The prob­ Recent years have seen a welcome increase in the number of studies
lem is that the justified call to include both sexes in research is often that assess sex/gender-related variables in addition to sex category.
followed by a binary conceptualization of the physiology of females and Unfortunately, instead of using the powerful tools of deep learning to
males. This binary conceptualization is most evident in the common uncover the probably complex relations between these variables and
understanding of the request to consider sex as a biological variable as a specific endpoint(s) (e.g., disease outcome), many of these studies
request to study sex differences (see Ritz et al. (2014) for a similar reduce the data to a single continuum – the probability that the partic­
warning against a simplistic binary approach to sex and sex differences). ipant is a woman (or a man) – and then assess the correlations of this

173
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

variable with the endpoints (e.g., Ballering et al., 2020; Norris et al., David, S.P., Naudet, F., Laude, J., Radua, J., Fusar-Poli, P., Chu, I., Stefanick, M.L.,
Ioannidis, J.P.A., 2018. Potential reporting Bias in neuroimaging studies of sex
2017; Pelletier et al., 2015; Smith and Koehoorn, 2016).
differences. Sci. Rep.-Uk 8.
The binary framework and the focus on sex differences have impli­ Del Giudice, M., 2019. Measuring sex differences and similarities. In: VanderLaan, D.P.,
cations also for clinical and diagnostic settings. Clearly, one’s sex cate­ Wong, W.I. (Eds.), Gender and Sexuality Development: Contemporary Theory and
gory provides crucial information for diagnosis in some situations – for Research. Springer, New York.
Del Giudice, M., Lippa, R.A., Puts, D.A., Bailey, D.H., Bailey, J.M., Schmitt, D.P., 2016.
example, a patient presenting with acute pelvic pain – and should surely Joel et al.’s method systematically fails to detect large, consistent sex differences.
be recorded, together with other data such as age, blood pressure, Proc. Natl. Acad. Sci. U.S.A. 113, E1965.
chronic disease, etc., in any medical encounter. But, as the examples Eberly, L.A., Richterman, A., Beckett, A.G., 2019. Identification of Racial Inequities in
Access to Specialized Inpatient Heart Failure Care at an Academic Medical Center.
above demonstrate, providing good medicine, and surely providing Circ. Heart Fail. 12 (11), e006214.
personalized medicine, requires much more than a person’s sex category Ecker, C., Andrews, D.S., Gudbrandsen, C.M., Marquand, A.F., Ginestet, C.E., Daly, E.M.,
– it requires gathering information on a variety of variables that are Murphy, C.M., Lai, M.C., Lombardo, M.V., Ruigrok, A.N., et al., 2017. Association
between the probability of autism Spectrum disorder and normative sex-related
related to sex and gender (in addition to many other variables). phenotypic diversity in brain structure. JAMA Psychiatry 74, 329–338.
Sex category is not only dull in the information it provides, the Eliot, L., 2020. Sex/gender differences in the brain and their relationship to behavior. In:
current dominance of the binary framework may lead to misdiagnosis Cheung, F., Halpern, D. (Eds.), Cambridge International Handbook on Psychology of
Women. Cambridge University Press, Cambridge, UK.
when a patient is suffering from a condition or is presenting with Fausto-Sterling, A., 2000. Sexing the Body: Gender Politics and the Construction of
symptoms that are associated with the other gender. One example is Sexuality. Basic Books, New York.
depression, and especially postpartum depression, which is under- Fine, C., 2010. Delusions of Gender: How Our Minds, Society, and Neurosexism Create
Difference. W. W. Norton, New York.
diagnosed in men, even though it is not as rare as commonly believed.
Garcia-Falgueras, A., Swaab, D.F., 2008. A sex difference in the hypothalamic uncinate
For example, a recent study revealed a 10:9 ratio of mothers to fathers nucleus: relationship to gender identity. Brain 131, 3132–3146.
suffering from postpartum depression (Cheng et al., 2018). Another Garcia-Falgueras, A., Ligtenberg, L., Kruijver, F.P.M., Swaab, D.F., 2011. Galanin
example is heart disease, which is considered a men’s disease, even neurons in the intermediate nucleus (InM) of the human hypothalamus in relation to
sex, age, and gender identity. J. Comp. Neurol. 519, 3061–3084.
though it is the number one killer of women in the United States (Canto Greenberg, D.M., Warrier, V., Allison, C., Baron-Cohen, S., 2018. Testing the
and Kiefe, 2014). The association between heart disease and men leads empathizing-systemizing theory of sex differences and the extreme male Brain
to a gender bias in referrals for both diagnostic and therapeutic pro­ theory of autism in half a million people. Proc. Natl. Acad. Sci. U. S. A. 115,
12152–12157.
cedures, which may result in poorer outcomes (e.g., Eberly et al., 2019; Greenblatt, D.J., Harmatz, J.S., Singh, N.N., Steinberg, F., Roth, T., Moline, M.L.,
Humphries et al., 2017). Harris, S.C., Kapil, R.P., 2014. Gender differences in pharmacokinetics and
pharmacodynamics of zolpidem following sublingual administration. J. Clin.
Pharmacol. 54, 282–290.
8. Summary Grgurevic, N., Majdic, G., 2016. Sex differences in the brain-an interplay of sex steroid
hormones and sex chromosomes. Clin. Sci. 130, 1481–1497.
The sex binary is a powerful framework through which we under­ Horovitz, O., Tsoory, M.M., Yovell, Y., Richter-Levin, G., 2014. A rat model of pre-
puberty (juvenile) stress-induced predisposition to stress-related disorders: sex
stand not only the social world but also physiology. However, current similarities and sex differences in effects and symptoms. World J. Biol. Psychiatry 15,
data suggest that this framework is not appropriate for understanding 36–48.
sex effects on the brain or even sex itself. Moreover, the dominance of Humphries, K.H., Izadnegahdar, M., Sedlak, T., Saw, J., Johnston, N., Schenck-
Gustafsson, K., Shah, R.U., Regitz-Zagrosek, V., Grewal, J., Vaccarino, V., et al.,
the binary framework interferes with the scientific endeavor to under­
2017. Sex differences in cardiovascular disease - Impact on care and outcomes.
stand sex, the brain, and the relations between them. The challenge for Front. Neuroendocrinol. 46, 46–70.
the future is to develop new analytical methods that take human vari­ Hyde, J.S., 2005. The gender similarities hypothesis. Am. Psychol. 60, 581–592.
ability on all measures, including sex-related ones, into account (Joel, Hyde, J.S., 2014. Gender similarities and differences. Annu. Rev. Psychol. 65, 373–398.
Hyde, J.S., Bigler, R.S., Joel, D., Tate, C.C., van Anders, S.M., 2019. The future of sex and
2014; Joel and Fausto-Sterling, 2016). gender in psychology: five challenges to the gender binary. Am. Psychol. 74,
171–193.
Acknowledgement Ingalhalikar, M., Smith, A., Parker, D., Satterthwaite, T.D., Elliott, M.A., Ruparel, K.,
Hakonarson, H., Gur, R.E., Gur, R.C., Verma, R., 2014. Sex differences in the
structural connectome of the human brain. Proc. Natl. Acad. Sci. U. S. A. 111,
This work was partly supported by the Israel Science Foundation 823–828.
(grant No. 217/16). Jancke, L., Merillat, S., Liem, F., Hanggi, J., 2015. Brain size, sex, and the aging brain.
Hum. Brain Mapp. 36, 150–169.
Joel, D., 2011. Male or female? Brains are intersex. Front. Integr. Neurosci. 5, 57.
References Joel, D., 2012. Genetic-gonadal-genitals sex (3G-sex) and the misconception of brain and
gender, or, why 3G-males and 3G-females have intersex brain and intersex gender.
Anderson, N.E., Harenski, K.A., Harenski, C.L., Koenigs, M.R., Decety, J., Calhoun, V.D., Biol. Sex Differ. 3, 27.
Kiehl, K.A., 2019. Machine learning of brain gray matter differentiates sex in a large Joel, D., 2014. Sex, gender, and brain – a problem of conceptualization. In: Schmitz, S.,
forensic sample. Hum. Brain Mapp. 40, 1496–1506. Höppner, G. (Eds.), Gendered Neurocultures Feminist and Queer Perspectives on
Arnold, A.P., 2012. The end of gonad-centric sex determination in mammals. Trends Current Brain Discourses. Zaglossus. e.U., Vienna.
Genet. 28, 55–61. Joel, D., 2015. The NIH call to consider sex as a biological variable is conceptually
Arnold, A.P., Chen, X., 2009. What does the “four core genotypes” mouse model tell us “captured” in the “sex differences” paradigm. Catal. Fem. Theory Technoscience 1,
about sex differences in the brain and other tissues? Front. Neuroendocrinol. 30, 4–5.
1–9. Joel, D., 2020. Beyond sex differences and a male-female continuum: Mosaic brains in a
Ballering, A.V., Bonvanie, I.J., Olde Hartman, T.C., Monden, R., Rosmalen, J.G.M., 2020. multidimensional space. In: Lanzenberger, R., Kranz, G.S., Savic, I. (Eds.), Sex
Gender and sex independently associate with common somatic symptoms and Differences in Neurology and Psychiatry. Elsevier, Amsterdam.
lifetime prevalence of chronic disease. Soc. Sci. Med. 253, 1–11. Joel, D., Fausto-Sterling, A., 2016. Beyond sex differences: new approaches for thinking
Baron-Cohen, S., 2002. The extreme male brain theory of autism. Trends Cogn. Sci. 6, about variation in brain structure and function. Philos. Trans. Biol. Sci. 371.
248–254. Joel, D., McCarthy, M.M., 2017. Incorporating sex as a biological variable in
Blackless, M., Charuvastra, A., Derryck, A., Fausto-Sterling, A., Lauzanne, K., Lee, E., neuropsychiatric research: where are we now and where should we be?
2000. How sexually dimorphic are we? Review and synthesis. Am. J. Hum. Biol. 12, Neuropsychopharmacology 42, 379–385.
151–166. Joel, D., Berman, Z., Tavor, I., Wexler, N., Gaber, O., Stein, Y., Shefi, N., Pool, J.,
Canto, J.G., Kiefe, C.I., 2014. Age-specific analyses of breast cancer versus heart disease Urchs, S., Margulies, D.S., et al., 2015. Sex beyond the genitalia: the human brain
mortality in women. Am. J. Cardiol. 113, 410–411. mosaic. Proc. Natl. Acad. Sci. U.S.A. 112, 15468–15473.
Chekroud, A.M., Ward, E.J., Rosenberg, M.D., Holmes, A.J., 2016. Patterns in the human Joel, D., Persico, A., Hanggi, J., Pool, J., Berman, Z., 2016. Reply to Del Giudice Et Al.,
brain mosaic discriminate males from females. Proc. Natl. Acad. Sci. U.S.A. 113, Chekroud Et Al., and Rosenblatt: do brains of females and males belong to two
E1968. distinct populations? Proc. Natl. Acad. Sci. U.S.A. 113, E1969–E1970.
Cheng, E.R., Downs, S.M., Carroll, A.E., 2018. Prevalence of depression among fathers at Joel, D., Persico, A., Salhov, M., Berman, Z., Oligschlager, S., Meilijson, I., Averbuch, A.,
the pediatric well-child care visit. JAMA Pediatr. 172, 882–883. 2018. Analysis of human brain structure reveals that the brain “Types” typical of
Clayton, J.A., Collins, F.S., 2014. Policy: NIH to balance sex in cell and animal studies. males are also typical of females, and vice versa. Front. Hum. Neurosci. 12, 399.
Nature 509, 282–283. Joel, D., Garcia-Falgueras, A., Swaab, D., 2020. The complex relationships between sex
and the brain. Neuroscientist 26, 156–169.

174
D. Joel Neuroscience and Biobehavioral Reviews 122 (2021) 165–175

Jordan-Young, R., Rumiati, R.I., 2012. Hardwired for sexism? Approaches to sex/gender Ritz, S.A., Antle, D.M., Cote, J., Deroy, K., Fraleigh, N., Messing, K., Parent, L., St-
in neuroscience. Neuroethics 5, 305–315. Pierre, J., Vaillancourt, C., Mergler, D., 2014. First steps for integrating sex and
Kaiser, A., 2012. Re-Conceptualizing “Sex’’ and “Gender’’ in the Human Brain. gender considerations into basic experimental biomedical research. FASEB J. 28,
Zeitschrift für Psychologie 220, 130–136. 4–13.
Kersey, A.J., Csumitta, K.D., Cantlon, J.F., 2019. Gender similarities in the brain during Rosenblatt, J.D., 2016. Multivariate revisit to “sex beyond the genitalia”. Proc. Natl.
mathematics development. NPJ Sci. Learn. 4, 19. Acad. Sci. U.S.A. 113, E1966–E1967.
Maney, D.L., 2015. Just like a circus: the public consumption of sex differences. Curr. Sanchis-Segura, C., Ibanez-Gual, M.V., Adrian-Ventura, J., Aguirre, N., Gomez-Cruz, A.J.,
Top. Behav. Neurosci. 19, 279–296. Avila, C., Forn, C., 2019. Sex differences in gray matter volume: how many and how
Maney, D.L., 2016. Perils and pitfalls of reporting sex differences. Philos. Trans. R. Soc. large are they really? Biol. Sex Differ. 10, 32.
Lond., B, Biol. Sci. 371, 20150119. Sanchis-Segura, C., Ibanez-Gual, M.V., Aguirre, N., Gomez-Cruz, A.J., Forn, C., 2020.
McCarthy, M.M., Arnold, A.P., 2011. Reframing sexual differentiation of the brain. Nat. Effects of different intracranial volume correction methods on univariate sex
Neurosci. 14, 677–683. differences in grey matter volume and multivariate sex prediction. Sci. Rep. 10,
McEwen, B.S., Milner, T.A., 2017. Understanding the broad influence of sex hormones 12953.
and sex differences in the brain. J. Neurosci. Res. 95, 24–39. Sanz, V., 2017. No way out of the binary: a critical history of the scientific production of
Mitricheva, E., Kimura, R., Logothetis, N.K., Noori, H.R., 2019. Neural substrates of sex. Signs J. Women Cult. Soc. 43.
sexual arousal are not sex dependent. Proc. Natl. Acad. Sci. U.S.A. 116, Sekido, R., 2014. The potential role of SRY in epigenetic gene regulation during brain
15671–15676. sexual differentiation in mammals. Adv. Genet. 86, 135–165.
Ngun, T.C., Ghahramani, N., Sanchez, F.J., Bocklandt, S., Vilain, E., 2011. The genetics of Shalev, G., Admon, R., Berman, Z., Joel, D., 2020. A mosaic of sex-related structural
sex differences in brain and behavior. Front Neuroendocrin. 32, 227–246. changes in the human brain following exposure to real-life stress. Brain Struct. Funct.
NIH revitalization Act, 1993, https://orwh.od.nih.gov/sites/orwh/files/docs/NIH- 225, 461–466.
Revitalization-Act-1993.pdf. Smith, P.M., Koehoorn, M., 2016. Measuring gender when you don’t have a gender
Norris, C.M., Johnson, N.L., Hardwicke-Brown, E., McEwan, M., Pelletier, R., Pilote, L., measure: constructing a gender index using survey data. Int. J. Equity Health 15,
2017. The contribution of gender to apparent sex differences in health status among 82–90.
patients with coronary artery disease. J. Womens Health 26, 50–57. Tyan, Y.S., Liao, J.R., Shen, C.Y., Lin, Y.C., Weng, J.C., 2017. Gender differences in the
Pelletier, R., Ditto, B., Pilote, L., 2015. A composite measure of gender and its association structural connectome of the teenage brain revealed by generalized q-sampling MRI.
with risk factors in patients with premature acute coronary syndrome. Psychosom. Neuroimage Clin. 15, 376–382.
Med. 77, 517–526. van Putten, M., Olbrich, S., Arns, M., 2018v. Predicting sex from brain rhythms with
Phillips, O.R., Onopa, A.K., Hsu, V., 2019. Beyond a Binary Classification of Sex: An deep learning. Sci. Rep. 8, 3069.
Examination of Brain Sex Differentiation, Psychopathology, and Genotype. J. Am. Wheelock, M.D., Hect, J.L., Hernandez-Andrade, E., Hassan, S.S., Romero, R.,
Acad. Child Adolesc. Psychiatry 58 (8), 787–798. Eggebrecht, A.T., Thomason, M.E., 2019. Sex differences in functional connectivity
Reich, C.G., Taylor, M.E., McCarthy, M.M., 2009. Differential effects of chronic during fetal brain development. Dev. Cogn. Neurosci. 36, 100632.
unpredictable stress on hippocampal CB1 receptors in male and female rats. Behav. Zell, E., Krizan, Z., Teeter, S.R., 2015. Evaluating gender similarities and differences
Brain Res. 203, 264–269. using metasynthesis. Am. Psychol. 70, 10–20.
Richardson, S.S., 2013. Sex Itself: the Search for Male and Female in the Human Genome. Zhang, C., Dougherty, C.C., Baum, S.A., White, T., Michael, A.M., 2018. Functional
University of Chicago Press, Chicago. connectivity predicts gender: evidence for gender differences in resting brain
Rippon, G., Jordan-Young, R., Kaiser, A., Fine, C., 2014. Recommendations for sex/ connectivity. Hum. Brain Mapp. 39, 1765–1776.
gender neuroimaging research: key principles and implications for research design,
analysis, and interpretation. Front. Hum. Neurosci. 8, 650.

175