Sex differences in brain and behavior in adolescence: Findings from the Philadelphia Neurodevelopmental Cohort

Abstract

Sex differences in brain and behavior were investigated across the lifespan. Parameters include neurobehavioral measures linkable to neuroanatomic and neurophysiologic indicators of brain structure and function. Sexual differentiation of behavior has been related to organizational factors during sensitive periods of development, with adolescence and puberty gaining increased attention. Adolescence is a critical developmental period where transition to adulthood is impacted by multiple factors that can enhance vulnerability to brain dysfunction.

Here we highlight sex differences in neurobehavioral measures in adolescence that are linked to brain function. We summarize neuroimaging studies examining brain structure, connectivity and perfusion, underscoring the relationship to sex differences in behavioral measures and commenting on hormonal findings. We focus on relevant data from the Philadelphia Neurodevelopmental Cohort (PNC), a community-based sample of nearly 10,000 clinically and neurocognitively phenotyped youths age 8–21 of whom 1600 have received multimodal neuroimaging. These data indicate early and pervasive sexual differentiation in neurocognitive measures that is linkable to brain parameters. We conclude by describing possible clinical implications.

Introduction

An extensive literature on brain and behavior has documented sex differences in cognitive, affective and brain imaging parameters. Such measures have been informative in evaluating aberrations in neurodevelopmental disorders where sex differences are prominent, including attention deficit, learning disabilities and autism spectrum disorder. Sexual differentiation of behavior has been related to organizational factors during sensitive periods of development, with the prenatal period most investigated across species. There is growing evidence that puberty is another organizational period with long lasting effects on brain and behavior. Adolescence presents an especially informative and dynamic period as brain maturation is accelerated, hormonal changes associated with puberty emerge and social factors increase their impact. The transition to adulthood is influenced by complex interactions where the effects of this critical period may differ for males and females with implications for healthy functioning and psychopathology.

We will begin this review by highlighting sex differences in neurobehavioral measures in adolescence that are linked to brain function. We will then summarize neuroimaging studies examining brain structure, connectivity and perfusion. We will conclude by summarizing literature on the role of hormonal measures and discuss clinical implications.

Behavior linked to brain function

The developmental course of specific behavioral domains has been well documented. Executive-control (e.g., Conklin, et al., 2007; Goldberg, Maurer, & Lewis, 2001; Pickering, 2001), language and reasoning (e.g., Friederici & Wartenburger, 2010; Kuhl, 2010) and, more recently, social cognition (e.g., Burnett, et al., 2011; Shaw et al., 2012) show improved performance from childhood to young adulthood, especially pronounced during adolescence for executive domains of attention and working memory (Ang & Lee, 2010). Neural substrates for such age-related differences are being examined extensively with structural and functional neuroimaging, initially in cross-sectional studies and more recently expanding to longitudinal investigations. Results highlight childhood and adolescence as periods during which important age-related differences are observed in parameters of neural structure and function (Casey, Duhoux & Malter, 2010; Giedd, et al., 1999; Matsuzawa et al., 2001; Shaw et al., 2008). Integrating neuroimaging with behavioral findings, Jung and Haier (2007) identified a central role for frontal and parietal regions in the neurodevelopment of cognition, and this hypothesis has received support in large-scale studies (Deary, Penke & Johnson, 2010).

Sex differences have been extensively documented in behavioral measures (e.g., Halpern, et al., 2007; Hines, 2010). Males perform better than females on spatial (Linn & Petersen, 1985; Voyer, Voyer & Bryden, 1995) and motor tasks (e.g., Moreno-Briseño, et al., 2010; Thomas & French, 1985), while females perform better than males on some verbal and memory tasks (e.g., Hedges & Nowell, 1995; Hyde & Linn, 1988; Saykin et al., 1995) as well as measures of social cognition (Erwin et al., 1992, Gur et al., 2010, 2012; Moore et al., 2015; Williams et al., 2008). Some sex differences have been related to structural neuroimaging (e.g., De Bellis et al., 2001; Goldstein et al., 2001; Gur et al., 1999; Lenroot, et al., 2007) and functional imaging measures (e.g., Gur et al., 1982; 1995; 2000; Lenroot & Giedd, 2010), including volumetric differences in executive and memory related areas, supporting neural substrates for sex differences in cognition. However, the developmental course of sex differences in brain-behavior relationship, especially in adolescence and across neurobehavioral domains, remains to be elucidated, particularly with longitudinal studies.

Shortcomings of most cognitive measures currently used limit their applicability in establishing further links between brain function and behavioral domains. Most are broadly defined and load heavily on the “g factor” (Salthouse, 2004) without separating accuracy from speed. This feature precludes rigorous testing of hypotheses on the effects of brain connectivity on performance, which is expected to differentially influence processing speed. Additionally, the paper-and pencil administration format of many tests precludes their use in large-scale neuroimaging genomic studies. More narrowly defined behavioral tasks, used in functional neuroimaging, have been adapted as computerized tests to obtain rapid and efficient quantification of individual differences (Gur, Erwin & Gur, 1992, Gur et al., 2010). The literature is especially limited in the application of an identical neurocognitive test battery across a population ranging from childhood through puberty and young adulthood.

The Philadelphia Neurodevelopmental Cohort (PNC) includes a large well-characterized community sample of youths, age 8–21 years. The PNC received a computerized neurocognitive battery (CNB; Gur et al., 2010, 2012, Roalf et al., 2014a) that is based on functional neuroimaging studies (Roalf et al., 2014b), has established validity (Moore et al., 2015) and heritability (Calkins et al., 2010; Greenwood et al., 2007; Gur et al., 2007). The age range from childhood to young adulthood enables to examine the pattern of performance, both accuracy and speed, during adolescence. The cross sectional sample was obtained between 2009–2011 and a subsample of about 500 is followed longitudinally with clinical, neurocognitive and neuroimaging measures.

Performance scores on each neurocognitive domain at baseline were standardized to the average of the entire sample (n= 9,122: 4405 males, 4,717 females). The z-scores were entered into a repeated-measures ANOVA in SAS (SAS Institute Inc., Cary, NC, operating on Linux LIN 64 platform), using PROC GLM separately on the 12 Accuracy measures and the 14 Speed measures with Age group (7 levels, 2-year spans from 8–21) and Sex as grouping factors and Test as a repeated measures (within–group) factor. The Age group, Sex, and Test effects and their interactions for each test are presented in Table 1.

Table 1.

Summary of overall ANOVA results for neurocognitive domains by test for main effects of age group (A), sex (S), Test (T) and their interactions (upper Table) and followup ANOVAs by neurocognitive domain and Test (Lower Table).

ANOVA		ACCURACY		SPEED
	DF	F	P	DF	F	P
A	6;8516	588.76	<.0001	6;8354	374.80	<.0001
S	1;8516	<1	NS	1;8354	<1	NS
T	11;93676	3.36	0.0001	13;108602	7.81	<.0001
A*S	6;8516	1.87	0.0826	6;8354	2.56	0.0177
A*T	66;93676	34.89	<.0001	78;108602	99.06	<.0001
S*T	11;93676	60.47	<.0001	13;108602	67.43	<.0001
A*S*T	66;93676	1.47	0.0074	78;108602	2.83	<.0001
DOMAIN	TEST	Effect	DF	Accuracy F	P	Speed F	P
EXECUTIVE	ABF	A	6;8354	82.85	<.0001	22.99	<.0001
		S	1;8354	23.16	<.0001	0.02	0.8916
		A*S	6;8354	0.46	0.8386	2.63	0.0152
	ATT	A	6;8354	297.91	<.0001	902.90	<.0001
		S	1;8354	35.50	<.0001	80.03	<.0001
		A*S	6;8354	1.60	0.1433	6.26	<.0001
	WM	A	6;8354	231.70	<.0001	247.23	<.0001
		S	1;8354	0.03	0.865	71.66	<.0001
		A*S	6;8354	0.79	0.5798	2.61	0.0156
MEMORY	VMEM	A	6;8354	31.95	<.0001	373.99	<.0001
		S	1;8354	29.63	<.0001	14.15	0.0002
		A*S	6;8354	0.25	0.9605	0.98	0.4342
	FMEM	A	6;8354	144.12	<.0001	93.06	<.0001
		S	1;8354	21.21	<.0001	0.61	0.4343
		A*S	6;8354	0.69	0.6576	0.51	0.7977
	SMEM	A	6;8354	11.87	<.0001	63.66	<.0001
		S	1;8354	13.41	0.0003	3.58	0.0584
		A*S	6;8354	2.64	0.0148	0.90	0.4957
COMPLEX	LAN	A	6;8354	486.77	<.0001	220.88	<.0001
COGNIITION		S	1;8354	7.29	0.007	41.30	<.0001
		A*S	6;8354	2.65	0.0144	0.23	0.9684
	NVR	A	6;8354	169.29	<.0001	36.15	<.0001
		S	1;8354	35.10	<.0001	64.46	<.0001
		A*S	6;8354	3.57	0.0015	3.25	0.0034
	SPA	A	6;8354	229.96	<.0001	17.93	<.0001
		S	1;8354	221.85	<.0001	28.68	<.0001
		A*S	6;8354	1.08	0.373	3.92	0.0006
SOCIAL	EID	A	6;8354	149.50	<.0001	201.12	<.0001
COGNITION		S	1;8354	36.75	<.0001	161.12	<.0001
		A*S	6;8354	1.40	0.2093	3.22	0.0037
	EDI	A	6;8354	272.77	<.0001	30.46	<.0001
		S	1;8354	36.48	<.0001	11.36	0.0008
		A*S	6;8354	1.66	0.1273	2.69	0.0132
	ADI	A	6;8354	306.22	<.0001	13.28	<.0001
		S	1;8354	58.32	<.0001	16.75	<.0001
		A*S	6;8354	2.40	0.0254	3.10	0.005
SENSORIMOTOR	MOT	A	6;8354			738.10	<.0001
SPEED		S	1;8354			213.44	<.0001
		A*S	6;8354			10.13	<.0001
	SM	A	6;8354			121.92	<.0001
		S	1;8354			26.35	<.0001
		A*S	6;8354			0.75	0.6057

Abbreviations: ABF-Abstraction and Mental flexibility; ATT-Attention; WM-Working Memory; VMEM-Verbal Memory; FMEM-Facial Memory; SMEM-Sptial Memory; LAN-Language; NVR-Nonverbal Reasoning; SPA-Spatial Processing; EID-Emotion Identification; EDI-Emotion Differentiation; ADI-Age Differentiation; MOT-Motor Speed; SM-Sensorimotor Speed.

Figure 1 shows performance scores on each domain. As can be seen, there is overwhelming age associated improvement in performance across multiple neurobehavioral domains. Against that background, there is some variability among domains and between accuracy and speed measures and, most importantly, sex differences modulate these effects in a manner related to adolescence.

Figure 1.

Performance effect sizes in standard deviation units (+/-SEM) of males (blue) and females (red) in the Philadelphia Neurodevelopmental Cohort on the Computerized Neurocognitive Battery domains for Accuracy (left) and Speed (right).

Several effects are notable in Figure 1. Among the Executive domain measures (A), abstraction and mental flexibility shows the least age related improvement in accuracy and speed shows a trend toward decline post pubescence. Attention shows the greatest improvement in both accuracy and speed while working memory has intermediate age-related effect sizes. Sex differences are not prominent in executive functions except for higher accuracy in females for attention and greater working memory speed for males. Both effects emerge after age 11. For Episodic Memory tests (B), effect sizes are considerably smaller than for attention; memory is apparently a major strength of the developing brain already in childhood. Age-related improvement is most pronounced for verbal memory speed and for face memory, two domains in which females outperform males across the age range. As with sex differences in the Executive domain, the magnitude of the sex difference increases in post pubescence age bands. For the Complex Cognition domain (C), age-related improvement is seen primarily in accuracy where the effect sizes are large as well as in verbal reasoning speed. Sex differences appear again after age 11, where males begin to show improved accuracy while females begin to show better speed. For Social Cognition (D), females outperform males from childhood onward in both accuracy and speed across all three measures. Nonetheless, this difference seems accentuated in post pubescent years, especially for speed. The opposite sex difference is observed for motor speed, where males outperform females across the age range. Here too, however, these differences become greater in the post pubescent age groups. Thus, while most age-related trajectories flatten after age 18, both the rate of age-related differences and the magnitude of the sex differences increase after age 11.

In addition to the uneven rate at which cognitive systems mature, abilities develop unevenly within individuals (e.g., Luna, et al., 2004). The study of within-individual variability (WIV) in performance can shed light on the developmental course of variability in performance on different domains, across tasks, within a single individual. While most studies of development examine age effects on average performance, a parameter that reflects the degree of variability in performance within an individual would be sensitive to deviations of specific abilities from the global level of performance of that individual. There are likely individual differences in WIV, with low values indicating “cognitive generalists” and high values identifying “cognitive specialists” and changes in WIV could reflect differential improvement or deterioration in specific performance domains. Thus, assessing variability could provide important information about typical development (e.g., Van Geert & Van Dijk, 2002) and help identify individuals at risk for brain disorders affecting neurocognition. Notably, WIV is higher in developmental disorders such as attention deficit–hyperactivity disorder (Leth-Steensen, Elbaz, & Douglas, 2000) and schizophrenia (Roalf, et al., 2013). The PNC enables to study WIV across domains through development and examine sex differences in a sample that received the same measures across the age range.

WIV showed a non-linear, U-shaped, relationship with age (Roalf et al., 2014a) both for accuracy (Figure 2A) and for speed (Figure 2B). WIV decreased with age from childhood to adolescence, indicating the expected leveling of performance with maturation. Unexpectedly, however, WIV increased after age 17 for accuracy and after age 13 for speed into early adulthood, especially in males. Notably, WIV is consistently higher in males but this sex difference becomes accentuated after age 13 and into adulthood for both accuracy and speed. These results suggest that after maturation reaches a level of evenness among cognitive abilities, further maturation of behavioral performance is characterized by increased variability, most likely related to specialization. That specialization is more strongly reflected in speed variability than in accuracy.

Figure 2.

Within-individual variability (WIV) in performance of males (blue) and females (red) in the Philadelphia Neurodevelopmental Cohort on the Computerized Neurocognitive Battery domains for Accuracy (a) and Speed (b). (From Roalf DR, Gur RE, Ruparel K, Calkins ME, Satterthwaite TD, Bilker WB, Hakonarson H, Harris LJ, Gur RC. Within-individual variability in neurocognitive performance: age- and sex-related differences in children and youths from ages 8 to 21. Neuropsychology. 2014;28:506–518).

Structural Neuroimaging

Volumetric MRI

An extensive literature shows that adolescence is associated with changes in brain structure, including reduced gray matter (GM) volume and increased white matter (WM) volume, which have been related to sex differences (Blakemore, Burnett & Dahl, 2010; Giedd et al., 1999; Lenroot and Giedd, 2006; Paus, 2005; Sowell et al., 2003). Such maturational processes have been linked to cognitive and affective development during adolescence. Sex differences have been reported in overall brain volume as well as in regional volumes. For example, a steeper age-related increase in WM was reported in males compared with females (Giedd et al. 1999; De Bellis et al., 2001). Lenroot et al. (2007) compared the trajectories of WM development in a large sample and reported the same pattern of sex differences across the brain, and after covarying for total brain volume differences in WM trajectories between males and females in the frontal lobe persisted.

In the PNC sample, as previously detailed (Satterthwaite et al., 2014b), imaging data was acquired at a single site, on a single 3T Siemens TIM Trio whole-body scanner. Signal excitation and reception were obtained using a quadrature body coil for transmit and a 32-channel head coil for receive. Gradient performance was 45mT/m, with a maximum slew rate of 200 T/m/s. Volumetric analysis was performed using deformable registration via attribute matching and mutual-saliency weighting (DRAMMS; Ou et al., 2011, 2014). We examined age group and sex differences applying ANOVA to the volume measures obtained from the T1 weighted images in a sample of 1571 participants (745 males, 826 females). The ANOVA tested main effects for Age group, which divided the sample into seven groups (2-year spans from 8–21), and Sex as grouping factors, and Region (Frontal, Temporal, Parietal, Occipital) as repeated measures (within-group) factor, separately for GM and WM. Age group and Sex effects on deep GM were examined in a separate ANOVA. The overall ANOVAs for GM and WM and for individual regions are presented in Table 2 and the regional volumes are shown in Figure 3.

Table 2.

Summary of ANOVA results for volumetric analysis for main effects of age (A), sex (S), region (R) and their interactions overall and univariate by region.

OVERALL			GM		WM
	Effect	DF	F	P	F	P
	A	6;1557	28.33	<.0001	18.32	<.0001
	S	1;1557	395.27	<.0001	328.15	<.0001
	R	3;4671	51659.50	<.0001	68268.20	<.0001
	A*S	6;1557	2.19	0.0416	3.44	0.0022
	A*R	18;4671	23.96	<.0001	8.06	<.0001
	S*R	3;4671	144.18	<.0001	182.30	<.0001
	A*S*R	18;4671	1.85	0.0153	1.39	0.1248
REGION
FRONTAL	A	6;1557	28.53	<.0001	11.66	<.0001
	S	1;1557	334.77	<.0001	264.91	<.0001
	A*S	6;1557	2.31	0.0321	2.19	0.0417
TEMPORAL	A	6;1557	13.42	<.0001	10.21	<.0001
	S	1;1557	376.61	<.0001	344.91	<.0001
	A*S	6;1557	2.17	0.0437	4.06	0.0005
PARIETAL	A	6;1557	46.90	<.0001	40.07	<.0001
	S	1;1557	359.56	<.0001	342.52	<.0001
	A*S	6;1557	2.17	0.0437	4.51	0.0002
OCCIPITAL	A	6;1557	17.89	<.0001	21.77	<.0001
	S	1;1557	364.78	<.0001	264.48	<.0001
	A*S	6;1557	1.42	0.2051	3.75	0.001
DEEP GRAY	A	6;1557	2.05	0.0561
	S	1;1557	304.22	<.0001
	A*S	6;1557	2.36	0.0283

A=Age group; S=Sex; R=Region

Figure 3.

Magnetic Resonance Imaging volumetric measures (Means+/-SEM), for gray matter (GM) and white matter (WM), males (blue) and females (red), in the Philadelphia Neurodevelopmental Cohort. Age groups are in 2-year intervals, and results are shown for the four lobes (3a) and deep gray matter (3b).

Thus, we observed reduced cortical GM volume and increased WM volume in all lobes associated with age, which appeared more pronounced in post-pubescent groups compared to pre-pubescent children. In contrast, deep GM volume showed minimal age related changes. Sex differences in WM replicate earlier findings that indicate steeper increase in males from pre - to post–pubescence, especially in the frontal lobe (Lenroot et al., 2007).

Sex differences were also observed in the deep GM of the medial temporal lobe. Giedd et al. (2006) reported post-pubescence increase in hippocampal volume in females, but not in males. This effect may relate to better memory performance in females and the post-pubescence enhancement of this sex differences in memory described above (Gur et al., 2012). The PNC sample afforded the opportunity to test this link directly by examining pre-pubescent and post-pubescent hippocampal and amygdala volume in relation to memory performance (Satterthwaite et al., 2014a). We found that whereas pre-pubertal males and females had similar hippocampal volumes, post-pubertal females had larger hippocampi bilaterally. This effect was absent in the amygdala. Notably, post-pubertal sex differences were most prominent in the lateral aspect of the hippocampi corresponding to the CA1 subfield. The sex differences in hippocampal volume correlated with performance on memory tests.

There is need to establish, rather than assume, a link between sex differences and puberty or age. Disentangling the effects of age from pubescence is difficult in cross sectional studies although some inroads can be made by examining same age groups at different stages of pubescence (Satterthwaite et al., 2014a). Elucidating effects of pubescence and separating age effects from hormonal effects requires longitudinal studies in which hormonal measures are obtained and pubescence stage is rigorously established. While the literature on longitudinal effects is growing thus far efforts to separate age from hormonal effects are limited. For example, Dennison et al. (2013) examined volumetric changes in subcortical structures of 60 adolescents (28 females, 32 males) at ages 12.5 and 16.5 years. Brain regions showed a heterogeneous pattern of maturation indicating that hemispheric specialization and volumetric sex differences play a role in maturation. Vijayakumar et al. (2016), in a mixed longitudinal design, examined maturation of cortical thickness, surface area and volume in 90 participants (ages 11–20 years). Surface area, across most of the cortex, showed non-linear increases, whereas thickness and volume were characterized by non-linear decreasing and increasing trajectories. Sex differences in volume and surface area were observed across time, but there were no differences in thickness. The authors consider their findings to suggest that thickness and surface area may be driven by different underlying mechanisms of brain development. However, neither study examined hormonal status and therefore age and pubescence effects could not be disentangled.

Diffusion Tensor Imaging Connectivity

Diffusion tensor imaging (DTI) has been increasingly applied to examine sex differences in WM microstructure and fiber tracts that connect among regions. For example, Schmithorst, Holland and Dardzinski (2008) reported that females had higher fractional anisotropy (FA) in the splenium of the corpus callosum than males. Males, however, had a higher FA in several regions including the frontal lobes. Furthermore, in females age and FA had a positive correlation across regions, while males showed no such correlation between FA and age. For mean diffusivity (MD), males had higher values, compared with females, in the corticospinal tract and in the frontal lobe, while females had higher MD values in several other regions. Thus, structural properties of WM are not uniform throughout the brain or across males and females.

Perrin et al. (2009) examined sex differences in the maturation of WM during adolescence, measuring lobular volumes throughout the brain to estimate a myelination index using magnetization-transfer ratio. They reported in male adolescents age-related increases in WM lobular volumes with decreases in the lobular values of WM magnetization-transfer ratio. Furthermore, WM density in the cortico-spinal tract decreased with age. This pattern was not evident in female adolescents. The investigators suggest that sex-specific mechanisms may underlie WM growth during adolescence, involving age-related increases in axonal caliber in males and increased myelination in females. The literature on sex differences in DTI-based measures such as FA, as highlighted above, indicates increased FA values in major WM regions and tracts in males (Clayden et al., 2012; Hertig et al., 2011; Hsu et al., 2008;), and in the corpus callosum in females (Kannan et al. 2012). Indeed, our volumetric finding of greater cortical WM volume in males and greater callosal prominence in females led us to hypothesize that male brains are optimized to communicate within a hemisphere, whereas female brains are optimized for inter hemispheric communication (Gur et al., 1999).

More recently, DTI has been used to study the communication architecture of brain networks (Bava et al., 2011; Clayden et al., 2012; Herting et al., 2011; Hsu et al., 2008; Ingalhalikar et al., 2014). The structural connectome examines brain connectivity locally and globally. The identification of network properties such as communities or the communication backbone can advance the understanding of how complex behavior emerges from the integration of segregated neuronal clusters (Schwarz, Gozzi &Bifone, 2008). We have evaluated the structural connectome to elucidate sex differences in the PNC and found stronger intra-hemispheric connectivity bilaterally in males and stronger inter-hemispheric connectivity in females (Ingalhalikar et al., 2014).

To examine the developmental course of sex differences requires large datasets spanning age ranges that include adolescence. The PNC dataset (Satterthwaite et al., 2014b), which includes structural, functional and behavioral parameters, provides a unique opportunity to identify age-related differences in the subnetworks of the structural connectome and elucidate how these differences relate to sex differences in behavior. We evaluated subnetworks in order to establish a reliable link between brain structure and behavior (Tunc et al., 2016). Our results suggest that sex differences in functional and behavioral dimensions are associated with related differences in the network properties of the structural connectome. We observed increased structural connectivity related to the motor, sensory and executive function subnetworks in males. In females, subnetworks associated with social cognition, attention and memory tasks had higher connectivity. Another measure of network structure is modularity, which indicates the prominence of division of networks into modules (sometimes referred to as “communities”). Highly modular networks are characterized by dense connections among nodes within a module relative to sparse connections among nodes that belong to different modules. We found that males showed higher modularity compared to females, with females having higher inter-modular connectivity. Thus, an increased separation between males and females emerges in the course of development, in behavioral patterns and in associated brain parameters. However, it is still unclear how specifically a greater within hemispheric modularity in males and higher inter-modular and inter-hemispheric connectivity in females contribute to sex differences in particular behavioral domains. We can speculate that tasks that require depth of processing within a single domain, verbal or spatial, would be more easily processed by males, whereas tasks that require integration of domains, such as fusion of verbal and spatial aspects of stimuli, would be better facilitated in a brain with female connectome features.

Longitudinal DTI studies provide information on developmental trajectories indicating WM growth during adolescence. Most studies include small samples and two time points (Bava et al., 2010; Giorgio et al., 2009; Lebel and Beaulieu, 2011; Wang et al., 2012). Simmonds et al., (2014) conducted repeated annual examinations over a course of five years in a large sample of 128 youths (ages 8–28). The age range enabled evaluation of patterns of growth from childhood to young adulthood. During adolescence, WM microstructure in fronto-cortical, fronto-subcortical and cerebellar connections reached adult levels, whereas cortico-limbic connectivity matured in adulthood. Sex differences were observed with females showing growth especially in mid-adolescence whereas males showing continuous WM growth across the age range. Notably, maturation was related to cognitive performance.

Functional Neuroimaging

Perfusion

Cerebral blood flow (CBF), critical for healthy brain function, is coupled to regional metabolism, responds to activation with cognitive tasks and shows a marked decline throughout childhood and adolescence (Chiron et al., 1992; Takahashi et al., 1999). Compared to the extensive literature on cognitive and structural brain parameters in relation to development, the literature on CBF has been limited. Early technologies to measure CBF, including the Kety–Schmidt nitrous oxide method, ¹³³Xe clearance and ¹⁵O labeled water with PET, were limited in their application to developmental samples due to invasiveness and use of ionizing radiation. Yet, consistently across studies CBF was found to be elevated during childhood then declining throughout adolescence. Small sample sizes did not allow examining sex differences in youth. Studies of adults have demonstrated increased perfusion in females (e.g., Gur et al., 1982, Ragland et al., 2000).

The application of arterial spin labeling (ASL; Aguirre & Detre, 2012; Detre et al., 1992) with MRI provides a quantitative noninvasive measure of CBF that has been validated with PET (Ye et al., 2000) and applied to pediatric samples. Taki et al. (2011) replicated prior findings of declining perfusion in adolescence and also reported that females had higher perfusion in the posterior cingulate cortex, with a steeper decline rate of CBF in males. The effects of puberty on such sex differences have not been evaluated. Using ASL data from the PNC, we examined developmental patterns of cerebral perfusion in males and females in relation to puberty (Satterthwaite et al., 2014c). We demonstrated differential patterns of developmental CBF in males and females with divergent nonlinear trajectories in multiple brain regions, including hubs of the executive and default mode networks. The decline in CBF was similar between males and females in early puberty but diverged in mid-puberty, with CBF increasing in females and continuing to decrease in males. Thus, higher CBF previously noted in adult females emerges in mid-puberty already and the contribution of hormonal levels needs to be established.

Resting State Functional Connectivity

Resting-state functional connectivity MRI (rsfc-MRI; Biswal, VanKylen & Hyde, 1997; Fox & Raichle 2007) offers a potentially effective tool for examining functional brain networks (Power et al., 2011; Yeo et al., 2011). Several studies reported sex differences in functional connectivity (Biswal et al., 2010; Tian et al., 2011; Wang et al., 2012; Wu et al., 2013; Zuo et al., 2010), but most of this work has been done in adults. Developmental studies confronted a methodological obstacle since motion affects rsfc-MRI and confounds interpretation because of its association with age (Power et al., 2012; Satterthwaite et al., 2012; van Dijk, Sabuncu & Buckner, 2011).

We examined sex differences in functional connectivity in relation to cognition in the PNC (Satterthwaite et al., 2015). We found that sex differences in cognitive performance were related to multivariate patterns of rsfc-MRI. Males outperformed females on motor and spatial cognitive tasks and displayed more between-module connectivity, whereas females were faster in tasks of emotion identification and nonverbal reasoning, showing more within-module connectivity. Multivariate pattern analysis with support vector machines classified subject’s sex on the basis of their cognitive profile with 63% accuracy, but was more accurate using functional connectivity data (71% accuracy). Notably, “masculinity” of a participant’s cognitive profile was related to that of their pattern of brain connectivity. These findings demonstrate that sex differences in cognition are associated with divergent patterns of brain functional connectivity. Whereas analysis of the structural connectome indicates greater modularity in males, functional connectome shows greater modularity in females. The relationship between sex differences in structural connectivity described in the section on DTI and functional connectivity based on resting state measures of regional co-activation is a topic of current interest that is yet to be elucidated. Such investigations require multimodal convergent analyses with datasets geared to test hypotheses in developmental samples. For example, it is not clear whether structural connectivity constrains the development of functional connectivity or, conversely the establishment of a functional connectome triggers the layering of myelinated fibers that create the structural connectome. Longitudinal studies in which the relative timing of the establishment of regional connectivity can be determined are prerequisite for answering such questions.

Activation with functional MRI

Cognitive and affective processes have been probed in fMRI paradigms in adolescents, commonly in separate studies (e.g. Forbes, et al., 2011; Luna et al., 2001, Moore et al., 2012). The elucidation of the interaction between cognitive and emotion processing measures has been more limited in adolescence (e.g., Somerville et al., 2011). Sex differences have been noted in fMRI studies of adolescents probing inhibitory control (e.g., Rubia et al., 2013), attention (e.g., Rubia et al., 2010), working memory (e.g., Alarcon et al., 2014), and emotion processing (e.g., Schneider et al., 2011). It is difficult to generalize from a growing number of studies that use divergent tasks and approaches to data analysis and often with relatively small samples. Nonetheless, these studies generally indicate that sex difference are evident already in adolescence and parameters of regional brain activation show changes related to improved performance. Of note, hormonal effects have been occasionally evaluated with fMRI (e.g. Alcaron et al., 2014; Goddings et al., 2012; Cservenka et al., 2015), as described below.

Hormonal Modulation

The extent to which human behavioral sex differences are influenced by sex hormones that change during sensitive periods of development has been an important area of investigation. Hormonal effects on behavior in early development have been well documented in animal studies establishing the prenatal and early neonatal stage as a sensitive period. More recently, based on rodent research, puberty has been considered as another sensitive period along the continuum where brain organization is influenced by sex hormones (Schultz, Molenda-Figueira & Sisk, 2009; Sisk & Zehr, 2005). This line of research is germane to advancing the understanding of how puberty during adolescence impacts brain organization.

In a comprehensive review, Berenbaum and Beltz (2011) suggest that sex differences in human behavior relate to hormonal exposure at multiple developmental periods, including puberty. While most literature evaluated the effects of androgens on male-typical behavior, there is a growing body of research supporting the role of ovarian hormones in female-typical behavior. Viewing sensitive periods as a continuum implies that variations in timing of puberty impact brain organization and behavior and may contribute to understanding sex differences during adolescence. Lacking is data from large-scale longitudinal studies in adolescence where cognitive performance, combined with parameters of brain structure and function, are obtained during specific time points. As noted in our summary of the neurocognitive findings, some sex differences in behavior remain constant throughout development, such as better performance of females on social cognition measures, while others emerge with pubescence.

Similar effects are reported for several parameters of brain structure and function (Raznahan et al., 2010). MRI studies of brain structure have linked increased testosterone to WM development in males (Perrin et al., 2009; Paus, 2010) and a negative correlation was reported between estradiol level and age-corrected GM volume in adolescent females (Pepper et al., 2008). Similarly, Satterthwaite et al., (2014c, 2015) using the PNC data, linked both structural and functional parameters to puberty and age-related differences in neurocognitive performance. Again, longitudinal studies are needed with concurrent measurements of performance, brain parameters and hormonal status. Such studies could focus on the period of transition from pre-pubescence to pubescence.

There is variability between sexes and within each sex in onset of puberty. Physical changes are evident earlier in females, about age 10, than males, about age 11.5 (Marshall and Tanner, 1969). There is also a notable concomitant increase in sex hormones - estradiol in females and testosterone in males (Nottelmann et al., 1987). Longitudinal studies enabling repeated measures of maturation can provide important information on onset and progression of puberty in relation to brain parameters. Several studies have undertaken such paradigms with structural MRI measures. For example, Raznahan and colleagues (2010) examined the relation between variation in signaling efficacy of the androgen receptor and neuroanatomic brain maturation (ages 9 to 22 years). Findings suggest sex specific and brain region specific effects: greater androgen receptor signaling attenuated age-related decreases in superior parietal and parts of temporal lobe in males, while accelerating age-related decreases in the left inferior frontal gyrus in females. Nguyen and colleagues (2013a; 2013b) evaluated sex differences in the association between androgen levels and cortical thickness in pre-pubertal and post-pubertal males and females (ages 4 to 22 years). They reported that higher dehydroepiandrosterone (DHEA) in males and females and testosterone in females predicted increases in cortical thickness in the pre-pubertal participants, while higher testosterone predicted decreases in cortical thickness in post-pubertal males and females. Tanner stages in a sample of males and females (ages 7 to 22 years) were noted to predict changes in subcortical volume of several structures including the hippocampus, amygdala, and caudate (Goddings et al. 2013). In a sample of 126 adolescents (ages 10 to 14 years) Hertig et al. (2014) employed growth curve modeling to examine how testosterone and estradiol relate to changes in subcortical brain volumes obtained in a longitudinal design 2-years apart. Hormonal levels and Tanner Stage predicted WM and right amygdala growth for males and females across adolescence independent of age. Such studies illustrate the power of integrating measures of puberty and developmental epoch in the study of brain and behavior. Efforts to examine gene environment contributions to structural brain measures were enhanced in the twin paradigm (Brouwer et al., 2015). Non-shared environmental factors contributed in females, ages 9–12 years, to the association between follicle stimulating hormone and regional GM density. Shared environmental factors contributed to the association of higher estradiol levels with lower regional GM density.

In addition to hormonal relations to structural brain parameters, highlighted above, some studies with fMRI likewise made efforts to evaluate pubertal status in relation to brain function (e. g., Forbes, et al., 2011; Klapwijk et al., 2013; Moore et al., 2012), including hormonal levels (e.g., Alcaron et al., 2014; Goddings et al., 2012; Cservenka et al., 2015). The convergence of brain measures with evaluation of hormonal studies will contribute to elucidating the mechanisms underlying sex differences. However, as indicated above, disentangling hormonal effects from age effects requires longitudinal studies with rigorous measurements of all related parameters.

Clinical implications

Here we focused on normative sex differences in brain and behavior. These differences should be considered when interpreting effects of diverse brain disorders with manifested psychopathology. Developmental disorders may emerge early and are more prevalent in males. Anxiety and mood disorder commonly emerge later in development and are more frequent in females. Some of the normative sex differences may explain or modulate effect of disorders such as schizophrenia (e.g., Ragland et al., 1999; Gur et al., 2004; Calkins et al., 2013). Therefore, comparative samples of males and females are needed when examining disorders. Such samples should represent the lifespan because hormonal factors mediate these sex differences and likely interact with the disorder in generating the symptoms. Future treatments can be informed by these relationships in tailoring interventions to males and females at different stages of development.

How could such future clinical application look like? For example, there is increased awareness that sport-related mild traumatic brain injuries can have cumulative adverse effects. With the increased participation of females in sports, normative growth charts for neurocognitive development and, eventually, neuroimaging based growth charts for brain development, it will be possible to identify and monitor such effects. Importantly, such a developmental database would also enable disentangling these effects from those of other insults to the brain that may relate to other adverse conditions characteristic of adolescence such as substance use, car crashes and brain disorders. Of note, many of these events affect frontal executive as well as limbic and striatal brain systems that would increase risk-taking behavior and potentiate sex characteristic psychiatric manifestations with depression in females and externalizing behaviors in males. Knowledge about normative sex differences is necessary to interpret and intervene.

Summary and future directions

Sex differences in brain organization that are evident in adults become accentuated during adolescence, implicating hormonal effects of pubescence. Improved executive function and complex and social cognition is associated with increase in the magnitude of sex differences in these domains and such changes in cognition are paralleled by age-related differences in brain parameters.

Results of the PNC, where the sample size permits detection of relatively small effects, show a striking ubiquity of significant sex differences on nearly all behavioral and brain parameters. We replicated effects from the literature, but also found sex differences that have not been observed before both in individual parameters and in age-related differences across the developmental epoch we examined. The prevalence of such differences indicates that the human species demonstrates complementarity between the sexes in behavior and underlying brain structure and function. While some of these differences are small their effects across humanity can be substantial and with clinical and societal implications. On the other hand, none of the results are of the kind that would justify considering the human brain “sexually dimorphic”, certainly not in the same sense as other sexually dimorphic organs in the human body. Furthermore, environmental effects certainly modulate sex differences and their interaction with the developing brain needs further study. Finally, biological sex can interact with gender identity in ways that can be illuminating and merit investigation. Thus, sex differences in brain and behavior should be taken in perspective. They are interesting and informative, but male brains and correspondingly behavior is more alike that of females than it is different.

What we have learned from the PNC data is that the period between childhood through adolescence and into young adulthood is characterized by pronounced improvement in accuracy and speed of performance especially in executive and reasoning tasks, combined with reduced within individual variability from childhood through adolescence followed by increased variability especially in speed. Sex differences were evident on most tasks already at childhood but their magnitude increased with development. These age-related effects were paralleled by differences in brain parameters of anatomy and its connectivity and of physiology and its connectivity. Novel approaches for data analysis are needed and have been applied to examine development of these parameters and they generally indicate pruning of gray matter accompanied by myelination and complementary age-related effects on inter-regional anatomic connectivity. Sex differences in anatomic connectivity indicate greater modularity and within hemispheric connectivity in males and greater inter-module and interhemispheric connectivity in females. This pattern of sex differences emerges during adolescence and becomes more pronounced during young adulthood. It is still unclear and a topic of further investigation how the anatomic connectivity relates to physiologic connectivity and how both relate to performance.

These general relationships, however, are yet to be examined extensively in a longitudinal context and only such studies can be sensitive to detect effects of changes and document trajectories. Longitudinal studies are also needed to elucidate effects of sex hormones related to puberty. Multimodal longitudinal studies are needed for advancing the understanding of sex differences in both healthy development and the effect of aberrant conditions that lead to illness. The multitude of variables involved in such deep phenotyping and the need to consider multiple social and environmental factors mandates large-scale studies. In this context, community studies have an advantage over investigation of help-seeking patients and convenience samples of control individuals because the former provide information on the distribution of continuous dimensions and may afford a better appraisal of the need for intervention regardless of specific conditions affecting the likelihood of help seeking.

The prospect of large-scale multimodal studies in which deeply phenotyped populations are evaluated longitudinally is daunting, such an effort is however necessary to have the information needed for mechanistic accounts of sex differences in behavior and neuropsychiatric disorders. With such information we will be in a position to detect early signs of impending psychopathology that may differ in boys and girls, and we will be equipped with evidence-based models for gender optimized prevention and intervention.

Highlights.

• Notable sex differences

• Adolescence is a critical period

• Multimodal data merging needed