Sex Differences in Dynamic and Static Measures of Brain Integration Derived from Resting-state Functional Magnetic Resonance Imaging

doi:10.21203/rs.3.rs-7906846/v1

Background: Understanding the impact of biological sex on the functional organization and dynamics of the brain is crucial for elucidating sex-specific differences in cognitive functions and neuropsychiatric disorders. Systems neuroscience often models the brain as a network of interconnected brain regions with functional connectivity (FC), the correlation between signal time courses, as a measure of connection strength. FC matrices, here derived from resting-state functional magnetic resonance imaging (rs-fMRI), define a network graph that can be characterized by its segregation or, inversely, level of module integration. Such parameters can be generated for the full length of the acquired data (static) or for short periods implying dynamically changing brain states. We recently made the interesting observation in a separate study (N = 63) that measures of brain integration and segregation based on dynamic functional connectivity (dFC) data differed between sexes, while graph-based measures based on static FC (sFC) did not, which we investigated in more detail in this study.

Methods: We preregistered to repeat our analysis from the small sample in N = 501 subjects of the Human Connectome Project dataset. We did cross-sectional comparisons between sexes of the static rs-fMRI graph parameters modularity and global efficiency, as well asthe dFC parameters state prevalence, mean dwell time, mean inter-state transition time (ITI), and variability derived from a two-state model. Additionally, we explore whether sex differences in 66 cognitive and behavioral parameters are mediated by the FC integration measure with the strongest sex effect.

Results: All static and dynamic measures of integration/segregation showed higher levels of functional integration in males with effect sizes up to 0.60 for the dFC parameter prevalence. For 3 of the 66 explored cognitive and behavioral parameters, we observed that the prevalence of the integrated state mediated the sex difference: dexterity, agreeableness, and self-reported aggression.

Conclusion: We found robust evidence in two data sets that rs-fMRI-based measures of brain integration are increased in males. An exploratory analysis, which needs replication, suggests that such differences mediate personality differences. This study highlights that biological sex differences in brain functional organization may explain sex-typical behaviors.

resting-state fMRI

dynamic functional connectivity

sliding window analysis

sex difference

functional integration

functional segregation

human brain connectome

Functional connectivity measures the similarity of activity between two brain regions over time. It is thought to be a measure of coupling strength between these regions and is used to define functional networks or modules of multiple brain regions. This aspect of brain architecture is similar to public transportation networks between the cities of a country. This architecture can be presumed to be stable over minutes (static) or change every couple of seconds (dynamic). Both approaches can be used to define how integrated the modules of the brain are, which is a property of the underlying architecture. While it is known that this architecture is different between women and men, the ways in which it is different and the consequences of such differences are unclear.

In this study, we show that men have higher levels of modular integration than women in both static and dynamic measures of this property. We also demonstrate that such differences may lead to known variations in behaviors, such as agreeableness and self-reported aggression. This is an important step forward in our endeavor to understand whether and how differences in brain structure and function between the sexes may lead to or contribute to known differences in behavior.

Males show higher brain integration than females in static and dynamic parameters.
Sex differences were stronger in dynamic than static parameters.
Brain integration mediated sex effects in dexterity, agreeableness and aggression.
Replication in N = 501 confirmed preregistered findings in N = 63.
The results help to understand the neural basis of behavioral sex dimorphism.

Studying differences between biological sexes with respect to intrinsic brain activity enhances our comprehension of individual differences in brain function and subsequent behavior and cognition (1, 2). Additionally, knowledge of sex differences in brain function in addition to biological, environmental and sociocultural variables is an important component of understanding sex-related differences not only in behavior but also in neuropsychiatric disease (3). This could inform tailored interventions and personalized treatment strategies in various psychiatric and neurological disorders with sex disparities (4, 5).

Functional connectivity (FC) (6), a measure of synchrony between brain activity time courses from different brain regions, is widely used to characterize brain function. It has also been employed to quantify fundamental differences between sexes in resting-state functional magnetic resonance imaging (fMRI) scans. Early functional studies reported that resting-state FC is more efficient in the right hemisphere of males and in the left hemisphere of females (7, 8), suggesting that sex differences in cognition may, in part, be related to divergent neural patterns in the brain (9). Moreover, stronger FC within the default mode network and reduced FC in sensorimotor cortices during rest have been reported in females (10, 11). Echoing these findings in healthy people, research on disorders with known sex differences in clinical manifestations has also revealed FC sex differences in multiple resting-state networks, e.g., the default mode, limbic, ventral attention and cerebellum in depression (12), autism (13) and Alzheimer's disease (14), as well as reduced network efficiency in male patients with multiple sclerosis (15).

Most relevant studies implicitly assume that FC remains temporally stationary over the course of the scan (usually 5–10 min.). More realistically, FC varies on such time scales, which led to the development of dynamic functional connectivity (dFC) metrics (16), which capture aspects of the dynamic nature of FC (17, 18). Reports on sex differences in dFC measures are still rare, possibly due to the novelty and greater complexity of the approach. Menon & Krishnamurthy reported that sex could be identified with approximately 80% accuracy based on dFC matrices using a 4-state model compared with 68% accuracy with static FC (sFC) bivariate Pearson correlation matrices and 90% accuracy with sFC partial correlation matrices (19). Sen and Parhi reported 94% accuracy using a tensor parallel factor decomposition technique for dFC matrices (20). Although these findings suggested distinct static and dynamic FC patterns between males and females, they did not report how FC differs between the sexes. Furthermore, females and males showed different dwell times in some whole-brain states with opposite effects in particular networks, i.e., the task-negative and specific task-positive networks such as the sensory-motor network and the executive network (3, 21), suggesting a possible link between differential neurocognitive performance in males and females and brain functional dynamics (3).

In our own work (osf.io/6gswx – in review), we investigated interindividual differences in FC patterns in terms of the concept of network (graph) integration and segregation, i.e., the idea that FC patterns may be described by stronger long-range, between-module connections and weaker within-module connections (integration) or vice versa (segregation) (22). Considering a static FC graph, these anticorrelated features can be captured by the graph-theoretical parameters global efficiency (increasing with functional integration, decreasing with functional segregation) and modularity (increasing with functional segregation, decreasing with functional integration), respectively. Moreover, by using the dFC approach with sliding window analysis (SWA) and a two-state model, a study revealed that the brain switches between a segregated state (State S) and an integrated state (State I) (23). From this model, the dynamic parameters mean dwell time (MDT), state prevalence (Prev - % scan time in this state), intertransition interval (ITI; average of MDTs for an even number of states) and state variability (Var) were derived. We consider the prevalence of the segregated state Prev_S as well as the MDT of the segregated state as measures of brain segregation, as these values prove to be correlated with sFC modularity and anti-correlated with sFC global efficiency and vice versa for Prev_I and MDT_I (see Supplement Table S1).

In our previous preregistered study (OWN), we found no correlation of these parameters with need for cognition, a questionnaire measure of a person’s tendency to like cognitively demanding tasks, in 63 analyzed participants (osf.io/286fb), but an effect of the sex covariate. A more detailed analysis revealed significant sex differences in dFC state prevalence and MDT of the segregated state with higher values of segregation in females, but no significant effects for the two sFC parameters above. This suggested not only a difference in functional brain organization between sexes, but also that dFC parameters may be more sensitive than sFC parameters for detecting such differences. However, our sample size of N = 63 provided only limited statistical power to confirm this. Consequently, we preregistered a secondary analysis in a much larger sample, i.e., 501 subjects of the publicly available human connectome project (HCP) (osf.io/p8usv).

Here, we present the results of this preregistered analysis. Our hypotheses for the dFC parameters were as follows: higher Prev_S (H1), higher MDT_S (H2), and lower Var_I (H3) in females than in males. We did not preregister a hypothesis on MDT_I, which did not show a significant sex effect in our OWN study. With respect to sFC parameters, we hypothesized lower global efficiency (H4) and greater modularity (H5) in females than in males. H1 to H5 are in line with the general notion that males show higher levels of brain integration. Finally, with respect to relative effect sizes, we hypothesized that effect sizes for the sFC parameters would be smaller than for the dFC parameter Prev_S (H6). Additionally, we conducted an exploratory analysis of 66 cognitive and behavioral measures to test whether sex disparities are mediated by brain integration (i.e., Prev).

2.1 Participants and data acquisition

We report data from two studies (see Table 1) here: the initial findings stem from our OWN data, whereas the confirmatory and preregistered analyses (osf.io/p8usv) were performed on HCP data.

2.1.1 OWN data

Within the Collaborative Research Center (CRC) 940 on Volition and Cognitive Control funded by the Deutsche Forschungsgemeinschaft (DFG), 80 subjects of subproject C1 on Volitional Dysfunctions in Self-Control Failures and Addictive Behaviors (24) agreed to participate in an additional session of MRI scans. Of these, 11 participants were excluded because more than 7.5% of frames had framewise displacements (FD) of 0.5 mm or more, five were missing behavioral data needed for the original investigation and one participant was excluded because of problems with normalization to MNI space (see also the preregistration osf.io/286fb; manuscript is under review). Among the resulting N = 63 participants, 49% had a mostly mild, addictive disorder and 51% were healthy controls. All the participants received financial compensation after MRI data collection. The study was approved by the Ethics Committee of the Technische Universität Dresden (EK 4012016) and all participants signed informed consent forms after receiving a detailed description of the experiment. The collection parameters of the MRI data were the same as those reported in a previous study (23). Rs-fMRI data (16 min 27s) based on a multi-band oblique-axial (T > C ~ -17°) 2D EPI sequence (25) were acquired with TR 987 ms, TE 32.6 ms, voxel size 2.0 mm × 2.0 mm × 2.0 mm, slice gap 0 mm, FOV 192 mm ×192 mm, multiband factor 6, flip angle 62°, matrix 96 × 96, BW 1860 Hz/Px, 72 interleaved slices, and 1000 volumes. All the participants received foam padding for head-movement reduction and earplugs for hearing protection and were instructed to close their eyes and to try not to fall asleep.

2.1.2 HCP data

The HCP data consisted of 501 subjects from the HCP S1200 release (26), whose collection parameters have been detailed in the previous study (23). For the rs-fMRI data (14 min 24sec), a simultaneous multi-slice pulse sequence with an acceleration factor of eight (27) was used with TR 720 ms, TE 33.1 ms, voxel size 2.0 mm × 2.0 mm × 2.0 mm, FOV 208 mm × 208 mm, multi-band factor 8, flip angle 52°, matrix 96 × 96, BW 2290 Hz/Px, 72 interleaved slices, and 1200 volumes. We used the data from the two sessions with left-to-right phase-encoding directions.

2.2 Preprocessing

2.2.1 OWN data

The preprocessing pipeline used fMRIPrep 1.2.5 (zenodo.org/record/4252786#.X7TzMGhKhPZ) based on Nipype 1.1.6 (zenodo.org/record/4035081#.X7Ty32hKhPY) (28) as in the previous study (23). FD was computed (29) to control for head motion (see section 2.1.1). We subsequently regressed out six head motion parameters (i.e., 3 translations and 3 rotations), signals of cerebrospinal fluid and white matter. For sFC, we used temporal bandpass filtering with a range of 0.01 ~ 0.1 Hz. For dFC computation, we used filtering with a range of 1/(w*TR) ~ 0.1 Hz, where w = 40, which resulted in a low frequency boundary of 0.025 Hz (30).

2.2.2 HCP data

We used the preprocessed rs-fMRI data with FIX cleaning (31, 32) from the HCP1200 dataset and regressed out the six head motion parameters as well as cerebrospinal fluid and white matter signals. We used the same bandpass filtering parameters as those used for the OWN data for the sFC in the HCP data. For the dFC, the band-pass filtering was employed with the consistent high-frequency boundary of 0.1 Hz and lower frequency boundary of around 0.025 Hz as used in OWN data, which corresponds to a window width of 55 TR. Since the HCP provided data from two sessions with left-to-right phase-encoding direction, we calculated the average dynamic and static parameters across the two sessions.

2.3 FC analyses

We used the first version of the Automated Anatomical Labeling atlas (AAL) (33), which consists of 116 regions of interest (ROIs), and grouped it into nine networks on the basis of Yeo’s seven functional networks on the cerebral cortex (34), and anatomical parcellations of the subcortical regions and cerebellum (i.e., visual network, sensory-motor network, dorsal attention network, ventral attention network, limbic network, frontoparietal network, default mode network, basal ganglia network, and cerebellar network). Mean time series were extracted by spatially averaging all voxels within each ROI.

SFC was computed as the Pearson correlation z-value on the basis of the ROI-wise time series. dFC was calculated as Pearson correlation z-values for each window frame (i.e., brain state instance) via SWA in DynamicBC toolbox (35) with a step size of 1 TR and a window size of 40 TRs = 39.48 s in the OWN data and 55 TRs = 39.60 s in the HCP data. Since one of our previous studies (23) showed that within-subject centering of the dFC matrices, i.e. subtracting the within-subject mean matrix, drastically reduces the reliability of the dFC parameters, we focus here on uncentered data and present results for centered data only for completeness and discussion purposes. As the resulting cluster centroids no longer reflect functional connectivity for centered data, we add across-sample mean connectivity matrices back to the centroid when appropriate. K-means clustering (k = 2) with cosine distance was used to classify centered and uncentered brain state instances as brain states I and S for all dFC matrices. For the OWN data, the extraction steps of the brain states are described in our preregistration (osf.io/286fb), which uses the same dataset. For the HCP data, the extraction was the same as the previous study based on 501 subjects (23).

2.3.1 Static FC parameters

To quantify static functional segregation and integration, graph theory-based parameters of modularity and global efficiency were computed in the brain connectivity toolbox (BCT) (36, 37). The former is computed as:

$$\:Q=\sum\:_{i=1}^{m}({e}_{ii}-{a}_{i}^{2})$$

1

,

where e_ii is the fraction of all edges that connect two nodes within module i, a_i is the fraction of edges that connect a node in module i to any other node, and m is the total number of modules computed via Newman’s modularity algorithm (38). Global efficiency is defined as:

$$\:{E}_{global}=\frac{1}{N*\left(N-1\right)}\sum\:_{i\ne\:j}\frac{1}{{L}_{ij}}$$

2

,

where N is the number of nodes, L_ij is the minimum path length between node i and j, and only positive connectivity was used for the calculation.

2.3.2 Dynamic FC parameters

We employed seven different parameters based on functional integration (state I) and segregation (state S) (6) to characterize features of dynamic brain states for each subject (23) (preregistrations in osf.io/286fb and osf.io/c3xvt):

1) Mean dwell time (MDT) – the average time one participant stayed in state I or S continuously during a run.

2) Prevalence (Prev) – a proportion of windows spent in state I or S with respect to the total number of windows within a recording. Note that Prev_I = 1- Prev_S; thus, we report primarily results for Prev_S only as effects for Prev_I will be merely opposite in sign.

3) Intertransition interval (ITI) – the length of time residing in any state before transitioning to a new state.

4) State variability (Var) – the mean Euclidean distance of the state instances belonging to one subject to the run-average of all windows within the same state and subject [see also (23) and preregistration (osf.io/c3xvt)]. Moreover, in the HCP dataset, where two sessions with left-to-right phase-encoding direction are available, we employed the averaged parameters of these sessions as the final brain-state values of the HCP for this study. Note that MDT, Prev, and Var are specific for the I and S states, whereas ITI is not, resulting in 7 parameters.

2.4 Statistical analyses

Since some of the dynamic parameters were right-skewed or otherwise not normally distributed, we conducted independent-samples t-tests and Mann-Whitney U tests on all of the variables to test the statistically significant (p < 0.05) differences between the male and female groups. We used two-sided tests in the OWN data and in the HCP data for MDT_I, ITI, and Var_S. To test the preregistered hypotheses listed in the introduction (i.e., Prev_S, MDT_S, Var_I and the two static parameters), one-sided tests were employed for uncentered data.

2.5 Mediation analyses

In an exploratory fashion, we investigated whether sex effects are mediated by brain integration for all 58 behavioral/cognitive parameters provided by the National Institutes of Health Toolbox (http://www.healthmeasures.net/explore-measurement-systems/nih-toolbox) (see Table S2), as used in previous studies (39, 40). For more details, please refer to the file wiki.humanconnectome.org/docs/assets/HCP_S1200_DataDictionary_Aug_22_2023.csv. Additionally, we examined the effects on eight alcohol consumption parameters (see Table S3), which we selected on the basis of our own research interest in alcohol use disorder. As a measure of brain integration, we employed the appropriate parameter that showed the strongest sex difference.

Before the mediation analyses, we calculated Pearson correlations of all the dependent NIH and alcohol measures with sex and our measures of brain integration. A two-sample t-test for sex differences was also performed, which produced the same p-value as the correlation. We then employed structural equation modeling for mediation analysis with 5000 random bootstrapping samples via the bootstrapLavaan function from the R software package lavaan (41) if the dependent variable was significantly correlated with both sex and the selected integration parameter (p < 0.05 after Bonferroni correction for 66 parameters). The dependent measures and the integration measure were standardized (variance = 1). Sex was coded as 1 for males and 2 for females. We computed the direct effect of sex on the dependent measure (i.e., path c’), the influence of sex on the integration measure (i.e., path a) and the influence of the integration measure on the dependent measure (i.e., path b). We report whether the indirect mediation effect (i.e., a*b) is significantly different from zero on the basis of the 95% confidence interval.

Table 1
Demographic information
Dataset	N (female)	Age [years (SD)]	Difference in sex within datasets		Difference in sex between datasets
			Chi-Square	p value	Pearson Chi-Square	p value
OWN	63 (29)	25.98 (1.61)	0.397	0.529	1.056	0.304
HCP	501 (265)	28.85 (3.63)	1.679	0.195

Table 2
T-tests and Mann-Whitney U tests were used to compare individual head motions (median FD) between males and females in the two datasets.
Parameters	sex	N	Mean (SD)	T test		Cohen’s d CI	Median	U test
Parameters	sex	N	Mean (SD)	t value (df) CI	p value	Cohen’s d CI	Median	p value
OWN data	male	34	0.146 (0.041)	0.487(61) [-0.016, 0.026]	0.628	0.123 [-0.373, 0.619]	0.136	0.730
OWN data	female	29	0.141 (0.041)				0.140
HCP data	male	236	0.149 (0.054)	0.033(499) [-0.009, 0.009]	0.974	0.003 [-0.172, 0.178]	0.137	0.672
HCP data	female	265	0.149 (0.046)				0.139

As there was no significant difference in sex number within or between the two groups (Table 1), we considered the sex distribution in both datasets to be balanced. Additionally, we extracted individual median FD values and compared the differences between the two groups. The results revealed that there were no significant differences in head motion between sex groups for the two datasets (Table 2), which suggests that head motion was not a confounder of the results of this study. The centroids (averaged median matrices of individual brain states within groups) for each state and dataset for both centered and uncentered data are shown in Figures S1 and S2 for OWN data and can be found in the previous publication for the HCP data (23).

3.1 Static parameters

OWN data

There were no significant effects of sex on global efficiency (i.e., a measure of integration) or modularity (i.e., a measure of segregation) in the OWN data (Table 3; upper part in Fig. 1).

HCP data

The results revealed significant effects of sex on both global efficiency and modularity, which is in line with the general notion that males show higher levels of brain integration (Table 3; lower part in Fig. 1).

Table 3

Results of two-sample t-tests and Mann-Whitney U tests comparing global efficiency and modularity, two static FC measures of brain integration and segregation, respectively, between females and males for the two datasets.
Parameters	sex	N	Mean (SD)	t-test		Cohen’s d [CI]	Median	U test
Parameters	sex	N	Mean (SD)	t value (df) [CI]	p value	Cohen’s d [CI]	Median	p value
OWM data
Global efficiency	male	34	0.45 (0.10)	1.079 (61) [-0.022, 0.075]	0.285	0.273 [-0.226, 0.769]	0.430	0.282
Global efficiency	female	29	0.42 (0.09)				0.403
Modularity	male	34	0.13 (0.05)	-1.032 (61) [-0.038, 0.012]	0.306	-0.261 [-0.757, 0.238]	0.126	0.301
Modularity	female	29	0.14 (0.05)				0.144
HCP data
Global efficiency	male	236	0.39 (0.08)	5.632 (444.457) [0.023,0.047]	0.000**	0.511 [0.333, 0.689]	0.379	0.000**
Global efficiency	female	265	0.36 (0.06)				0.352
Modularity	male	236	0.14 (0.04)	-5.504 (499) [-0.028, -0.013]	0.000**	-0.493 [-0.670, -0.314]	0.145	0.000**
Modularity	female	265	0.16 (0.04)				0.159

^{*: p < 0.05; **: p < 0.005. SD: standard deviation; df: degrees of freedom; underlined parameters in the HCP data: one−sided t−test for hypothesis testing; two−sided t−tests otherwise}.

3.2 Dynamic parameters

Uncentered data

OWN data

MDT_S was significantly greater in females than in males across all tests and Prev_S was significantly greater in females than in males (Table 4; upper part in Fig. 2). Notably, as Prev_I = 1- Prev_S, this indicates that Prev_I was lower in females. MDT_I was greater in males (upper part in Fig. 2), but this effect was not significant (Table 4). The effect sizes for these parameters are given in Table 4. Var_I was significantly greater in males (p < 0.05) for both t- and U-tests (Table 4). Var_S was marginally significantly greater (p < 0.1; Table 4) in females according to t-tests. There were no significant differences in the other parameters. These results (of two-sided tests) have already been published in our preregistration (osf.io/p8usv/) and are the basis for our hypotheses to be tested in the HCP data.

HCP data

We observed significant differences between sexes in MDT_S, MDT_I, Prev_S, VAR_S and VAR_I. Notably, MDT_S, Prev_S, and VAR_I displayed significance across the two datasets, whereas MDT_I and VAR_S displayed significance solely in the HCP dataset. Importantly, all the significant findings maintained consistent directional effects across the two datasets. Moreover, similar to the OWN data, there were no significant differences observed for ITI. In addition, the t-tests with significant results revealed small to medium effect sizes (Table 4). Importantly, the effect size for Prev_S (and thus, by definition for Prev_I) was the largest of all observed effects with Cohen’s |d| = 0.6, which was ~ 20% above the effect size for the sFC parameters. Thus, all hypotheses H1-H6 could be confirmed with the limitation that the difference in effect size (H6) is not significant on the basis of the confidence intervals.

Centered data

OWN data

There were no significant sex differences in the centered data except for Var_S in the U-Test. Effect sizes varied between small to medium (t-tests) for all the parameters (Table S4).

HCP data

There were no significant differences between the groups for any dynamic parameters in the centered data with, at best, small effect sizes (Table S4).

Table 4

Results of two-sample t-tests (t) and Mann-Whitney U tests to compare differences between females and males in dynamic FC parameters for uncentered data from the two datasets.
Parameters	sex	N	Mean (SD)	t-test		Cohen’s d [CI]	Median	U test
Parameters	sex	N	Mean (SD)	t value (df) [CI]	p value	Cohen’s d [CI]	Median	p value
OWN data
MDT_S	male	32	41.09 (34.06)	-2.149 (28.235)	0.040*	-0.658	32.853	0.027*
MDT_S	female	23	77.92 (76.94)	[-71.919, -1.742]		[-1.205, -0.105]	49.914
MDT_I	male	30	80.06 (100.40)	1.135 (54)	0.261	0.304	50.516	0.286
MDT_I	female	23	54.37 (61.02)	[-19.690, 71.078]		[-0.226, 0.831]	31.584
Prev_S	male	34	45.761 (35.457)	-2.654 (61)	0.010*	-0.671	42.04	0.007*
Prev_S	female	29	68.273 (31.160)	[-39.471, -5.553]		[-1.178, -0.159]	75.234
ITI	male	30	61.83 (45.33)	-0.546 (51)	0.588	-0.151	53.621	0.43
ITI	female	23	68.38 (40.52)	[-30.654, 17.553]		[-0.694, 0.394]	55.713
Var_S	male	33	45.44 (11.96)	-1.929 (34.912)	0.062†	-0.462	49.62	0.386
Var_S	female	29	49.55 (2.40)	[-8.433, 0.215]		[-0.966, 0.045]	49.823
Var_I	male	33	48.14 (5.79)	2.749 (32.994)	0.010*	0.768	48.822	0.019*
Var_I	female	27	40.10 (14.26)	[2.090,13.989]		[0.237, 1.292]	47.381
HCP data
MDT_S	male	236	28.639 (21.946)	-4.323 (497.995)	0.000**	-0.385	23.869	0.000**
MDT_S	female	264	37.614 (24.478)	[-13.054, -4.896]		[-0.562, -0.207]	30.521
MDT_I	male	232	49.504 (35.943)	2.656 (495)	0.008*	0.239	41.357	0.000**
MDT_I	female	265	40.897 (36.131)	[2.239, 14.974]		[0.062, 0.416]	31.582
Prev_S	male	236	35.325 (20.914)	-6.721 (499)	0.000**	-0.602	32.853	0.000**
Prev_S	female	265	48.022 (21.278)	[-16.408, -8.985]		[-0.781, -0.422]	46.466
ITI	male	232	39.035 (17.769)	0.136 (494)	0.892	0.012	34.271	0.717
ITI	female	264	38.818 (17.759)	[-2.924, 3.358]		[-0.164, 0.189]	35.198
Var_S	male	236	45.470 (6.559)	-4.886 (307.272)	0.000**	-0.456	47.578	0.000**
Var_S	female	265	47.713 (2.743)	[-3.146, -1.340]		[-0.633, -0.278]	48.315
Var_I	male	236	49.327 (2.214)	6.345 (428.436)	0.000**	0.551	49.390	0.000**
Var_I	female	265	47.561 (3.874)	[1.218, 2.312]		[0.372, 0.730]	48.102

Effect size Cohen’s d with 95% confidence intervals. The unit of MDTs and ITI is seconds; prevalence of segregated state (percentage) Prev_I = 1- Prev_S; *: p < 0.05; **: p < 0.005; †: p < 0.1 (trend). SD: standard deviation; df: degrees of freedom; CI: 95% confidence interval; underlined parameters in the HCP data: one-sided t-test for hypothesis testing; two-sided t-tests otherwise.

3.3 Mediation analyses

Thirty of the 58 behavioral and cognitive parameters and seven of the eight alcohol measurements showed sex differences (Tables S2 and S3). Of these 37, six measures, all from the former group, showed a bivariate correlation with Prev_I (Table S2): Penn matrix test—number of correct responses (PMAT24_A_CR), short Penn continuous performance test—specificity (SCPT_SPEC), nine-hole pegboard test (Dexterity_Unadj), five-factor model factor summary scores—agreeableness (NEOFAC_A), five factor model factor summary scores—conscientiousness (NEOFAC_C), and negative affect— sadness, fear, and anger (AngAggr_Unadj). All of them were also correlated with modularity and global efficiency. These six measures were subjected to mediation analyses to test whether the sex effect on Prev_I could explain the sex differences (Table S5).

In the mediation analyses, we observed significant [i.e., bootstrapping confidence interval (Boot CI) excluded 0] partial indirect/mediation effects of Prev_I in three cases for the effect of sex on dexterity (Dexterity_Unadj), on the agreeableness subscale of the five factor model of human personality NEO-FFI (NEOFAC_A) (42) and on self-reported aggression (AngAggr) (Fig. 3 and Table S5).

This study investigated biological sex differences in global FC during rest with a focus on markers of functional brain integration and segregation (22), inspired by a finding in a separate analysis currently in review in the OWN data (osf.io/286fb/). Our results indicate that males show higher measures of brain integration, whereas females show higher measures of segregation. This finding is consistent across multiple static and dynamic FC parameters associated with the concepts of brain integration and segregation. Specifically, FC networks in males show higher global efficiency and lower modularity than those in females. With respect to dFC, females spend more time in segregated brain states (Prev_{S [females]} = 48% versus Prev_{S [males]} = 35% for HCP data) and stay longer in the segregated brain state before transitioning to the integrated state (MDT_{S [females]} = 38 s versus MDT_{S [males]} = 29 s for HCP data) and shorter in an integrated state (MDT_{I [females]} = 41 s versus MDT_{I [males]} = 50 s for HCP data). This reciprocal pattern is consistent with the lack of a sex difference in the time between state switches (ITI), which approximates the mean of the MDT times. Thus, sexes do not differ in the rate of state changes but rather in their preferred brain states. Notably, while the dFC parameters Prev, MDT, and ITI are not independent (see Table S1), they reveal different interpretable features of brain state dynamics that cannot be obtained from sFC markers. Consequently, the analysis of dFC measures leads to more insights into the underlying mechanisms of FC group differences. Overall, our findings provide specific and reproducible evidence for sex differences in resting-state FC networks, thereby advancing our understanding of sex differences in human brain function.

Additionally, we observed sex differences in state variability, i.e. the mean distance of all state instances of a particular state to its within-subject centroid, which is consistent with the observed correlations between these variables and Prev and MDT of the respective state [Tables S1 and S7; see also (23)]. This association may be explained by presuming that the state distributions are, in first order, merely shifted between subjects. Higher values of Prev_S arise when this distribution is shifted toward the centroid of state S. Consequently, the extent of the segregated state cloud is increased (i.e. VAR_S) because of the fixed boundary between the states.

Cai et al. (21) reported sex differences in dwell time using a four-state model in late adolescents, identifying the differences in two of four states in one of two datasets. Specifically, females presented a shorter dwell time than did males in a state with overall low connectivity (i.e., a segregated state) and longer dwell times in a state with high connectivity (i.e., a more integrated state) in the visual and cognitive control networks, which is not in agreement with our observation of longer MDT_S in females. On the basis of a similar four-state model, de Lacy et al. (3) reported sex differences in dwell times between states, with females spending more time in brain states with anticorrelation between networks, which may be in line with our finding of longer MDT_S in females. They also reported greater functional dynamism, i.e. faster state switching, in males, whereas we observed no difference in ITI. Notably, in agreement with our finding of greater FC integration in males, they reported stronger sFC internetwork connections in males outside of default mode (sub-) networks. However, a further comparison of our work with these two studies is difficult because of the different methodologies used. This also applies to a previous study (43) that reported that males occupy more combinations of connectivity patterns on the basis of on a five-state model. These studies employed ICA-based brain parcellation to compute FCs and presumed four or more brain states. Conversely, we intentionally opted for atlas-based parcellation and only two states to maximize reliability and interpretability of the parameters (23). Notably, Fig. 4 in the first paper (21) illustrates that ICA-based parcellation results in substantial differences in parcels and derived states across groups, making the transfer of related findings to other datasets difficult and making a consistent description of sex differences impossible. In contrast, our findings are robust across groups. Thus, we have strong confidence that our findings are generalizable to other groups of healthy volunteers of a similar age range. Moreover, our parameters provide an interpretation for sex differences in whole-brain dynamics by quantifying global network integration and segregation.

We found no significant differences in the centered data across the two datasets except for VAR_S, which was uncorrected for multiple comparisons. The issue of data centering for dFC was already raised in our previous study (23). Here, we included this processing option in an applied context for completeness. Centering removes the between-subject differences in sFC, thus producing yielding states and parameters that are more clearly related to within-subject dynamics rather than a mixture of static and dynamic differences. This approach avoids issues with subjects who do not switch between states or are outliers (see Table S6), however, substantially reduces the reliability of the derived parameters (23). Conceptually, it is questionable whether it even results in a well-defined brain state because between-subject variations in sFC are eliminated. For these reasons, centering for dFC analyses is not recommended (23).

In structural brain graphs based on diffusion MRI, females displayed stronger features associated with functional integration than males did, for example, higher global efficiency (44). This is apparently opposite to our findings in the functional connectome. Sex differences in brain structure and function, however, are complex. For example, males display greater diffusion anisotropy and FC in unimodal sensorimotor cortices, whereas females have greater tract complexity and greater cortical thickness and greater FC in the default mode network (11). It has also been reported that the female structural brain graph has more edges, more spanning trees, and a larger minimal bisection width and is a better expander graph (45). Considering the foundational role of the structural connectome in shaping functional connectivity (46), further investigations that concurrently assess the properties of both functional and structural networks are warranted.

When screening 66 cognitive and behavioral measures, we identified six that showed both a sex difference and a correlation with Prev_I (Tables S2 and S3). All of them were also correlated with modularity and global efficiency. For three of these variables, we identified significant mediation effects of the brain integration measure. Although this finding is preliminary and explorative, it indicates that brain integration may explain part of the variance between sexes for particular behavioral traits such as dexterity, agreeableness or physical aggression. This needs to be further investigated in future studies. In our work, we found no correlation with need for cognition (Schwemmer et al. under review) in the OWN dataset.

Furthermore, we have currently investigated only global measures of functional integration/segregation. Considering such metrics for subsystems of the brain could increase the sensitivity and relevance of such measures for the above processes, which is beyond the scope of this article. Moreover, we have not yet investigated clinical populations. Changes in dFC are intricately linked to the pathophysiology of various brain diseases, including but not limited to depression (12), autism (13) Alzheimer's disease (14) and multiple sclerosis (15), all of which have demonstrated notable sex dimorphism. Thus, future work should investigate whether our findings hold translational relevance for understanding sex-specific neural mechanisms in clinical populations.

The results presented in this study need to be considered in the context of several methodological limitations. We currently do not know the underlying reasons for the observed sex differences, such as, potentially, fluctuations in autonomic system activity (47). Additionally, although we did not observe sex differences in head motion, a prevalent source of time-varying noise, methodological artefacts cannot be completely ruled out. Therefore, future studies should consider investigating the relationships between physiological variabilities and brain dynamics at rest. Second, more nuanced manipulations of subjects' internal states should be considered to facilitate the mapping and decoding of dynamic brain states on the basis of connectivity (17), which could involve employing multi-modal approaches, e.g., concurrent EEG-fMRI, to elucidate electrophysiological disparities between functional connectivity states. Third, we used k-means clustering, renowned for its efficiency and robustness, to identify the two brain states. Nonetheless, it is pertinent to recognize its susceptibility to outliers (48). Explorations into alternative methods for delineating dynamic brain states, e.g., coactivation analyses (49), are warranted to refine techniques for identifying functional connectivity states and state transitions. Methods that consider more than two brain states (3, 21, 43) may be vulnerable to reduced parameter reliability but could also be more sensitive to specific effects.

In conclusion, we found reliable sex differences in both sFC and dFC measures of brain integration and segregation with consistently higher values for integration and lower values for segregation in males than in females in two datasets. dFC parameters offered not only more specificity in this regard (e.g. sex differences in MDT but not in ITI) but also potentially larger effect sizes than the sFC parameters (Cohen’s d for sex difference is ~ 20% larger for Prev than for modularity or global efficiency). Taken together, our results underscore the utility and potential of dynamic FC brain analyses as a valuable tool for probing sex differences in brain function. Future endeavors should extend these inquiries to explore whether these dynamic properties undergo alterations throughout normal developmental trajectories, and cognitive processes, and in the context of neuropsychiatric disorders.

Ethics approval and consent to participate The OWN study was approved by the Ethics Review Board of the Technische Universität Dresden (EK45022016). All the subjects in the OWN study signed informed consent forms after receiving a detailed description of the experiment. HCP participants are anonymous to the authors of this paper and signed informed consent forms within the HCP project (humanconnectome.org). Consent for publication Not applicable Availability of data and materials This study was preregistered on https://osf.io/c3xvt after an analysis of the OWN data. All extracted parameters, including time series for connectivity matrices and the required software are openly available at https://osf.io/mu3st/. The raw MRI data from our acquisition (i.e., OWN data) are not available since complex brain images may contain fingerprint-like information, which could lead to reidentification of the subjects. Discussions on how to treat such data at the university level are ongoing. Individual requests for data access can be sent to the corresponding author. OWN data time-courses and matrices are available at https://osf.io/6gswx/files/osfstorage including static and dynamic parameters. The code to produce dFC parameters is available on https://osf.io/mu3st. The HCP data are publicly available at HCP Young Adult - Connectome (humanconnectome.org). All the imaging files of the Human Connectome Project (HCP) are publicly available and can be downloaded from http://www.humanconnectome.org/data. Competing interests The authors declare that they have no competing interests. Funding This research was supported by the Deutsche Forschungsgemeinschaft (DFG) grants 178833530 (SFB 940), 402170461 (TRR 265), and 454245598 (IRTG 2773). Authors' contributions All the authors have reviewed and approved the final version of the manuscript being submitted and warrant that the article is the authors' original work, has not been published before and isn't under consideration for publication elsewhere. Conceptualization: MM, OS; Methodology: MM, XF, OS; Software: XF; Formal Analysis: XF, OS; Investigation: MM, XF; Data Curation: MM, XF, OS, AH; Writing – Original Draft: XF, MM; Writing – Review & Editing: XF, MM, OS, AH; Visualization: XF, OS; Supervision: MM; Project administration: MM; Funding acquisition: MM. Acknowledgments We thank all participating subjects and the staff involved in the data acquisition. We gratefully acknowledge the computing time provided to us on the high-performance computers Taurus and Barnard at the NHR Center NHR@TUD (~10,000 CPUh). This is funded by the Federal Ministry of Education and Research and the state governments participating on the basis of the resolutions of the GWK for the national high-performance computing at universities (www.nhr-verein.de/unsere-partner). We are also thankful to the DFG-supported Open Access Publication Fund by the Technische Universität Dresden managed by the Sächsische Landesbibliothek – Staats- und Universitätsbibliothek Dresden (SLUB).

Jaušovec N, Jaušovec K. Resting brain activity: differences between genders. Neuropsychologia. 2010;48(13):3918–25.
Gaillard A, Fehring DJ, Rossell SL. Sex differences in executive control: A systematic review of functional neuroimaging studies. Eur J Neurosci. 2021;53(8):2592–611.
de Lacy N, McCauley E, Kutz JN, Calhoun VD. Sex-related differences in intrinsic brain dynamism and their neurocognitive correlates. NeuroImage. 2019;202:116116.
Rezzani R, Franco C, Rodella LF. Sex differences of brain and their implications for personalized therapy. Pharmacol Res. 2019;141:429–42.
Filippi M, van den Heuvel MP, Fornito A, He Y, Hulshoff Pol HE, Agosta F, et al. Assessment of system dysfunction in the brain through MRI-based connectomics. Lancet Neurol. 2013;12(12):1189–99.
Friston KJ. Functional and effective connectivity in neuroimaging: a synthesis. Hum Brain Mapp. 1994;2(1–2):56–78.
Tian L, Wang J, Yan C, He Y. Hemisphere- and gender-related differences in small-world brain networks: a resting-state functional MRI study. NeuroImage. 2011;54(1):191–202.
Tomasi D, Volkow ND. Gender differences in brain functional connectivity density. Hum Brain Mapp. 2012;33(4):849–60.
Satterthwaite TD, Wolf DH, Roalf DR, Ruparel K, Erus G, Vandekar S, et al. Linked Sex Differences in Cognition and Functional Connectivity in Youth. Cereb Cortex. 2015;25(9):2383–94.
Bluhm RL, Osuch EA, Lanius RA, Boksman K, Neufeld RW, Theberge J, et al. Default mode network connectivity: effects of age, sex, and analytic approach. NeuroReport. 2008;19(8):887–91.
Ritchie SJ, Cox SR, Shen X, Lombardo MV, Reus LM, Alloza C, et al. Sex Differences in the Adult Human Brain: Evidence from 5216 UK Biobank Participants. Cereb Cortex. 2018;28(8):2959–75.
Talishinsky A, Downar J, Vertes PE, Seidlitz J, Dunlop K, Lynch CJ, et al. Regional gene expression signatures are associated with sex-specific functional connectivity changes in depression. Nat Commun. 2022;13(1):5692.
Walsh MJM, Wallace GL, Gallegos SM, Braden BB. Brain-based sex differences in autism spectrum disorder across the lifespan: A systematic review of structural MRI, fMRI, and DTI findings. Neuroimage Clin. 2021;31:102719.
Cieri F, Yang Z, Cordes D, Caldwell JZK. Sex Differences of Brain Functional Topography Revealed in Normal Aging and Alzheimer's Disease Cohort. J Alzheimers Dis. 2021;80(3):979–84.
Schoonheim MM, Hulst HE, Landi D, Ciccarelli O, Roosendaal SD, Sanz-Arigita EJ, et al. Gender-related differences in functional connectivity in multiple sclerosis. Mult Scler. 2012;18(2):164–73.
Chang C, Glover GH. Time-frequency dynamics of resting-state brain connectivity measured with fMRI. NeuroImage. 2010;50(1):81–98.
Allen EA, Damaraju E, Plis SM, Erhardt EB, Eichele T, Calhoun VD. Tracking whole-brain connectivity dynamics in the resting state. Cereb Cortex. 2014;24(3):663–76.
Mucha PJ, Richardson T, Macon K, Porter MA, Onnela JP. Community structure in time-dependent, multiscale, and multiplex networks. Science. 2010;328(5980):876–8.
Menon SS, Krishnamurthy K. A Comparison of Static and Dynamic Functional Connectivities for Identifying Subjects and Biological Sex Using Intrinsic Individual Brain Connectivity. Sci Rep. 2019;9(1):5729.
Sen B, Parhi KK. Predicting Biological Gender and Intelligence From fMRI via Dynamic Functional Connectivity. IEEE Trans Biomed Eng. 2021;68(3):815–25.
Cai B, Zhang G, Zhang A, Hu W, Stephen JM, Wilson TW, et al. A GICA-TVGL framework to study sex differences in resting state fMRI dynamic connectivity. J Neurosci Methods. 2020;332:108531.
Shine JM, Bissett PG, Bell PT, Koyejo O, Balsters JH, Gorgolewski KJ, et al. The Dynamics of Functional Brain Networks: Integrated Network States during Cognitive Task Performance. Neuron. 2016;92(2):544–54.
Fang X, Marxen M. Test-retest reliability of dynamic functional connectivity parameters for a two-state model. Netw Neurosci. 2025:1–41.
Kraeplin A, Hofler M, Pooseh S, Wolff M, Kronke KM, Goschke T, et al. Impulsive decision-making predicts the course of substance-related and addictive disorders. Psychopharmacology. 2020;237(9):2709–24.
Moeller S, Yacoub E, Olman CA, Auerbach E, Strupp J, Harel N, et al. Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration using partial parallel imaging with application to high spatial and temporal whole-brain fMRI. Magn Reson Med. 2010;63(5):1144–53.
Van Essen DC, Smith SM, Barch DM, Behrens TE, Yacoub E, Ugurbil K, et al. The WU-Minn Human Connectome Project: an overview. NeuroImage. 2013;80:62–79.
Ugurbil K, Xu J, Auerbach EJ, Moeller S, Vu AT, Duarte-Carvajalino JM, et al. Pushing spatial and temporal resolution for functional and diffusion MRI in the Human Connectome Project. NeuroImage. 2013;80:80–104.
Esteban O, Markiewicz CJ, Blair RW, Moodie CA, Isik AI, Erramuzpe A, et al. fMRIPrep: a robust preprocessing pipeline for functional MRI. Nat Methods. 2019;16(1):111–6.
Power JD, Barnes KA, Snyder AZ, Schlaggar BL, Petersen SE. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage. 2012;59(3):2142–54.
Leonardi N, Van De Ville D. On spurious and real fluctuations of dynamic functional connectivity during rest. NeuroImage. 2015;104:430–6.
Glasser MF, Sotiropoulos SN, Wilson JA, Coalson TS, Fischl B, Andersson JL, et al. The minimal preprocessing pipelines for the Human Connectome Project. NeuroImage. 2013;80:105–24.
Barch DM, Burgess GC, Harms MP, Petersen SE, Schlaggar BL, Corbetta M, et al. Function in the human connectome: task-fMRI and individual differences in behavior. NeuroImage. 2013;80:169–89.
Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage. 2002;15(1):273–89.
Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol. 2011;106(3):1125–65.
Liao W, Wu GR, Xu Q, Ji GJ, Zhang Z, Zang YF, et al. DynamicBC: a MATLAB toolbox for dynamic brain connectome analysis. Brain Connect. 2014;4(10):780–90.
Cohen JR, D'Esposito M. The segregation and integration of distinct brain networks and their relationship to cognition. J Neurosci. 2016;36(48):12083–94.
Rubinov M, Sporns O. Complex network measures of brain connectivity: uses and interpretations. NeuroImage. 2010;52(3):1059–69.
Newman ME, Girvan M. Finding and evaluating community structure in networks. Phys Rev E Stat Nonlin Soft Matter Phys. 2004;69(2 Pt 2):026113.
Kong R, Li J, Orban C, Sabuncu MR, Liu H, Schaefer A, et al. Spatial Topography of Individual-Specific Cortical Networks Predicts Human Cognition, Personality, and Emotion. Cereb Cortex. 2019;29(6):2533–51.
He T, Kong R, Holmes AJ, Nguyen M, Sabuncu MR, Eickhoff SB, et al. Deep neural networks and kernel regression achieve comparable accuracies for functional connectivity prediction of behavior and demographics. NeuroImage. 2020;206:116276.
Rosseel Y. lavaan: An R Package for Structural Equation Modeling. J Stat Softw. 2012;48(2):1–36.
McCrae RR, Costa PT Jr. A contemplated revision of the NEO Five-Factor Inventory. Pers Indiv Differ. 2004;36(3):587–96.
Yaesoubi M, Miller RL, Calhoun VD. Mutually temporally independent connectivity patterns: a new framework to study the dynamics of brain connectivity at rest with application to explain group difference based on gender. NeuroImage. 2015;107:85–94.
Gong G, Rosa-Neto P, Carbonell F, Chen ZJ, He Y, Evans AC. Age- and Gender-Related Differences in the Cortical Anatomical Network. J Neurosci. 2009;29(50):15684–93.
Szalkai B, Varga B, Grolmusz V. Graph Theoretical Analysis Reveals: Women's Brains Are Better Connected than Men's. PLoS ONE. 2015;10(7):e0130045.
Honey CJ, Sporns O, Cammoun L, Gigandet X, Thiran JP, Meuli R, et al. Predicting human resting-state functional connectivity from structural connectivity. Proc Natl Acad Sci U S A. 2009;106(6):2035–40.
Shokri-Kojori E, Tomasi D, Volkow ND. An Autonomic Network: Synchrony Between Slow Rhythms of Pulse and Brain Resting State Is Associated with Personality and Emotions. Cereb Cortex. 2018;28(9):3356–71.
Golalipour K, Akbari E, Hamidi SS, Lee M, Enayatifar R. From clustering to clustering ensemble selection: A review. Eng Appl Artif Intell. 2021;104:104388.
Karahanoglu FI, Van De Ville D. Transient brain activity disentangles fMRI resting-state dynamics in terms of spatially and temporally overlapping networks. Nat Commun. 2015;6:7751.

No competing interests reported.

SupplementaryInformation.docx

Sex Differences in Dynamic and Static Measures of Brain Integration Derived from Resting-state Functional Magnetic Resonance Imaging

Status:

Version 1

Abstract

Figures

Plain English Summary

Highlights

1 Introduction

2 Methods

2.1 Participants and data acquisition

2.1.1 OWN data

2.1.2 HCP data

2.2 Preprocessing

2.2.1 OWN data

2.2.2 HCP data

2.3 FC analyses

2.3.1 Static FC parameters

2.3.2 Dynamic FC parameters

2.4 Statistical analyses

2.5 Mediation analyses

3 Results

3.1 Static parameters

3.2 Dynamic parameters

3.3 Mediation analyses

4 Discussion and Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1