This study was approved by the Institutional Review Board at the University of Southern California and was carried out in accordance with the United States Code of Federal Regulations (45 C.F.R. 46) and with the Declaration of Helsinki. Participants were recruited through community outreach (e.g., using advertisements and flyers) and/or through healthcare professionals who had referred volunteers for neurocognitive assessments and neuroimaging. All subjects who satisfied the inclusion criteria and who could provide written informed consent were invited to participate. Participants included 136 individuals with mTBI [N = 136; 52 females; age range: 19–79 years (y), 72 participants below age 40 years; age mean μ = 42 years, standard deviation σ = 17 years, Table 1]. To reduce recruitment bias, all individuals satisfying the study’s inclusion criteria were invited to participate and all those who did also provided written informed consent. Inclusion criteria were (a) a TBI diagnosis due to a ground-level fall involving direct head trauma; (b) a Glasgow Coma Scale score above 12 upon initial clinical evaluation (μ ± σ = 14 ± 1); (c) loss of consciousness shorter than 30 min (μ ± σ ≃ 9 ± 4 min); (d) post-traumatic amnesia of less than 24 h (μ ± σ ≃ 3.6 ± 2.1 h); (e) availability of T₁-weighted MRI scans acquired both acutely and chronically after the injury, i.e., within ∼7 days and ∼6 months (μ ± σ = 5.6 ± 0.3 months), respectively, and (f) no gross TBI pathology findings on clinical MRIs. Exclusion criteria were (a) a pre-traumatic history of cognitive impairment, neurological and/or psychiatric disease, and (b) a history of psychotropic substance abuse. This retrospective study did not involve any therapeutic intervention common across all mTBI participants. A group of 35 HCs whose fMRIs had been acquired at two timepoints were selected from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), whose inclusion criteria are described elsewhere (Petersen et al., 2010).

T₁-weighted MRIs were acquired from mTBI participants using a magnetization-prepared rapid acquisition gradient echo sequence with the following parameters: repetition time = 1.95 s; echo time = 2.98 ms; inversion time = 900 ms; voxel size = 1 mm × 1 mm × 1 mm. BOLD RS fMRI time series with at least 140 volumes per session were acquired (repetition time = 3 s, echo time = 30 ms, flip angle = 80 degrees, voxel size = 3.3125 mm × 3.3125 mm × 3.3 mm, acquisition matrix = 64 × 64 × 49). All data were anonymized and de-linked. HC subjects’ T₁-weighted and functional MRIs were obtained from the ADNI database¹. ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early AD. Cognition was evaluated at the two timepoints using the Brief Test of Adult Cognition by Telephone (BTACT) (Tun and Lachman, 2006; Kliegel et al., 2007). BTACT measures episodic verbal memory [immediate recall (EVMI) and delayed recall (EVMD) of words on a 15-item list], working memory span (WMS, assessed using a backward digit span task), inductive reasoning (IR, assessed using a number series completion task), processing speed (PS, measured using a backward counting task) and verbal fluency (VF, assessed using a category fluency task). The BTACT is a phone-based cognitive assessment, so administering the BTACT is logistically convenient because participants do not have to be interviewed in person. The BTACT is also reliable, as it has been extensively validated against face-to-face assessments and against longer cognitive assessments, including the 90-min in-person Boston cognitive battery (Lachman et al., 2014). p-values for statistical tests involving cognitive assessments were corrected for multiple comparisons using the Benjamini–Hochberg procedure with a false discover rate (FDR) of 0.05 (Hochberg and Benjamini, 1990).

T₁ MRIs were segmented using FreeSurfer 6.0 (FS²) with default parameters (Dale et al., 1999; Fischl, 2012). FS (a) removes non-cerebral voxels, (b) transforms volumes into Talairach space, (c) normalizes signal intensities across voxels, (d) segments the gray matter, (e) tessellates the gray matter/white matter boundary, and (f) corrects surface topology. fMRI preprocessing was undertaken using the FreeSurfer Functional Analysis Stream (FS-FAST³) with default parameters. The stream includes frame-to-frame motion correction, frame censoring, frequency filtering for removal of scanner and physiological noise, brain masking, intensity normalization, co-registration between T₁ and fMRI volumes, FS atlas surface sampling, smoothing using a kernel with a full width at half maximum of 5 mm, and volume resampling to the Montreal Neurological Institute’s regional parcellation space generated using data from 305 subjects. Nuisance timeseries due to motion, WM, and cerebrospinal fluid were regressed out and the first four volumes in each time series were removed for signal equilibration. Within a general linear model, FS-FAST was used to calculate the main effects of sex and age, as well as their interaction, on changes in FC between each of the 17 functional networks (seeds) and the rest of the cortex. Importantly, by default, FS-FAST implements a correction for multiple comparisons during the procedure for identifying statistically significant clusters. This default option for multiple comparison correction was used throughout this study.

Functional correlation seeds were identical to those defined by Yeo et al. (2011), who had assigned each cortical region to one of seven canonical RSNs (DMN, DAN, VAN, FPN, LN, SMN, or VN, Figure 1A). Although these networks consist of non-contiguous cortical parcels, most such parcels within each RSN exhibit significantly intercorrelated RS fMRI signals, whence their grouping into RSNs. The seven canonical RSNs can be hierarchically subdivided into 17 more granular networks N_i where i = 1, …, 17 (Figure 1B). These 17 networks map onto the seven canonical RSNs approximately as follows: N₁ and N₂ map onto the VN, N₃ and N₄ onto the SMN, N₅ and N₆ onto the DAN, N₇ and N₈ onto the VAN, N₉ and N₁₀ onto the LN, N₈, N₁₂ and N₁₃ onto the FPN, N₁₆ and N₁₇ onto the DMN (Figure 2). Because the 17-network parcellation scheme is more anatomically granular, each of its RSNs was used as a seed in the FC analysis. In the remainder of this subsection, each RSN and its functions are described to facilitate conceptual interpretation.

The VN’s primary role is in processing visual information (DeAngelis et al., 1995; Culham et al., 2001), and TBI can lead to visual deficits underpinned by VN dysfunction (Palacios et al., 2017). The VN fractionates into two subnetworks comprising central (N₁) and peripheral visual (N₂) areas, respectively, divided across by the calcarine fissure. N₁ spans the lateral and the posterior part of the medial occipital lobe (i.e., most of the primary visual cortex, V1), whereas N₂ comprises other visual areas (i.e., V2–V5). Thus, the VN contains neuronal populations attuned to detecting visual features (across both N₁ and N₂), object motion (N₂), as well as the speed and direction of such motion (e.g., visual area V5, which is part of N₂) (Dubner and Zeki, 1971; Maunsell and Van Essen, 1983). The superior temporal gyrus/auditory cortex is a part of the visuospatial attentional system which, along with the VN, is engaged in exogenous orientation to verbal instructions to perform a visual task (Mayer et al., 2004; Halgren et al., 2011).

The SMN is subdivided by separating the superior and inferior portions of the precentral gyrus (PCG) and postcentral gyrus (PoCG) by a line proceeding, in anteroposterior fashion, along the crown of the middle frontal gyrus (MFG). This produces one dorsal and one ventral network (N₃ and N₄, respectively), reflecting a functional subdivision of somatomotor representation areas. Insights facilitated by the 17-network parcellation help to conceptualize post-traumatic somatomotor deficits as reflecting changes in the FCs of highly specialized somatomotor representation areas. Specifically, N₃ (which includes the superior dorsolateral aspects of the PCG and PoCG), is responsible for proprioceptive sensation and for the control of hands and fingers, whereas N₄ (which includes the inferior aspects of the PCG and PoCG, the posterior aspect of the insula and the temporal plane), processes sensory information from the face, eyes, nose, mouth, tongue, and jaws (Grodd et al., 2001). Wernicke’s area (N₄) is associated with both visual and auditory language comprehension (Binder et al., 1997). FC between the SMN (N₃) and the superior frontal gyrus (SFG) is implicated in inhibitory control during no-go tasks (Nakata et al., 2009). This may partly explain why TBI patients often experience behavioral impulsivity and reduced inhibitory control (Dimoska-Di Marco et al., 2011).

The DAN is a task-positive network partly responsible for visuospatial attention (Umarova et al., 2010). The DAN contains one anterior and one posterior subnetwork (N₅ and N₆, respectively, Figure 2). The anterior subnetwork of the DAN (N₅) comprises the anterior aspects of the inferior occipital gyrus and sulcus, the lateral occipito-temporal gyrus, the lingual part of the medial occipito-temporal gyrus, and the superior parietal lobule. The posterior subnetwork of the DAN (N₆) includes posterior portions of the MFG, as well as the supramarginal gyrus, the superior parietal lobe, the precentral sulci, and the precuneus (PCun). The DAN can act in concert with the VAN, which is involved in redirecting attention to novel stimuli of behavioral relevance (Corbetta and Shulman, 2002; La et al., 2014). The fusiform gyrus (N₅) is involved in high-level visual computations undertaken during face perception, object recognition, and reading (Weiner and Zilles, 2016). Worse performance on tasks related to these cognitive functions has been reported after TBI (Alnawmasi et al., 2019; Turkstra et al., 2020).

The VAN consists of two subnetworks (N₇ and N₈) extending across the anterior insula and parts of the frontal lobe. N₇ areas are adjacent and posterior to N₈ and include the SFG, portions of middle cingulate and supramarginal cortices, as well as the anterior opercular part of the inferior frontal gyrus. N₈ encompasses the orbital and triangular parts of the inferior frontal gyrus, as well as the MFG, SFG, and supramarginal gyri. Specifically, the VAN specializes in detecting (A) unexpected and unattended stimuli, and (B) selecting a stimulus upon which the DAN should focus its neural processing capabilities related to sustained attention (Vossel et al., 2014). In right-handed individuals, the opercular and triangular parts of the inferior frontal gyrus together form the right-hemisphere homolog of Broca’s area (N₇), which is activated by relatively complex tasks involving speech processing and word retrieval (Just et al., 1996; Drager et al., 2004; Papoutsi et al., 2009; Schremm et al., 2018). One year after TBI, patients often exhibit deficits in their voluntary control of attention, and these deficits are correlated with reductions in FC between the VAN and other regions, relative to HCs (Richard et al., 2018).

The LN consists of two subnetworks, one temporal and another frontal (N₉ and N₁₀, respectively). Deficits pertaining to the recognition of faces and facial expressions, both of which involve the LN, are more prevalent after TBI than after brain tumors or stroke by 20 and 77%, respectively (Prigatano and Pribram, 1982; Valentine et al., 2006; Knox and Douglas, 2009).

The FPN is partitioned into N₁₂ [comprising the inferior frontal sulcus, inferior frontal gyrus, MFG, and posterior aspect of the middle temporal gyrus (MTG)] and N₁₃ (comprising the inferior temporal sulcus, MTG, medial aspect of the SFG, superior frontal and lateral orbital sulci, as well as parts of the angular and supramarginal gyri). The cortical regions belonging to N₁₂ and N₁₃ are typically adjacent. Thus, N₁₂ includes the MFG, parts of the postcentral sulci, and parts of the inferior temporal gyri (posterior aspects). N₁₃ includes parts of the MFG, lateral orbital sulci, postcentral sulci, inferior temporal gyri, and small parts of the SFG (medial aspects). N₁₆ overlaps with the anterior cingulate cortex, PCun, the ventral aspects of supramarginal gyri, and with part of the anterior aspect of the MTG. FC degradation in the MTG is associated with cognitive decline after injury in males (Konstantinou et al., 2018).

The DMN includes N₁₆,which overlaps with the anterior cingulate cortex, PCun, the ventral aspects of supramarginal gyri, and part of the anterior aspect of the MTG; and N₁₇, which contains the SFG, MTG, superior frontal sulcus, and the angular gyrus.

Our methodological approach relies on a theoretical framework for age- and sex-related differences in mTBI-related changes in FC that emerges from the functional neuroanatomy literature. This framework has not been expounded adequately before, yet its specifications are important for understanding the rationale of our approach and for the interpretations of our results. For these reasons, this theoretical framework is synthesized in what follows. Thus, age-related decline in cognitive performance is related to cortical thinning and related functional plasticity to compensate for this thinning (Greenwood, 2007). Older age has been associated with reduction in FC; for example, in older adults, Wang et al. (2010) found both a decline in frontotemporal and temporoparietal FC, as well as an increase in DMN FC during memory encoding and recognition.

The relationship between cognitive functions and FC is also modulated by sex, although a comprehensive theoretical framework explaining these differences has not been established. Male mTBI patients have been found to exhibit decreased FC in the VN compared to females, as well as increased FC across multiple networks, including an executive function-related network associated with insomnia severity (Wang et al., 2018). Functions of the PCun that involve the VN pertain to the processing of visuospatial imagery and to episodic memory retrieval (Cavanna and Trimble, 2006), both of which are often affected by TBI (Kwon and Jang, 2011; Gillis and Hampstead, 2015; Gilmore et al., 2016). PCun (N₂) involvement in these functions is also modulated by sex (Butler et al., 2006; Zilles et al., 2016).

In terms of the SMN, meta-analytic findings on sex-related differences pertaining to deficits of motor function and working memory are inconclusive (Gupte et al., 2019). HC females typically achieve better inhibitory control during no-go tasks (Li et al., 2006), and males’ exhibit reduced inhibitory control manifested as willful hand stillness during motor tasks (Nakata et al., 2009). The intrinsic FC of areas recruited by motor function is affected by TBI in a sex-dependent manner (Wang et al., 2018). The supplementary motor area (N₃) is likely involved in motor performance, particularly motor control (Goldberg, 1985), and TBI patients often suffer from motor deficits (Kuhtz-Buschbeck et al., 2003; Choi et al., 2012). Post-TBI, females tend to score worse than men on motor skill tests (Moen et al., 2014). Females typically demonstrate fewer somatosensory deficits following mTBI compared to males indicating sex differences in somatosensory deficit severity (Covassin et al., 2006; Bay et al., 2009). Some of these clinical observations (Irimia and Bradshaw, 2005a,b) have been attributed to females’ greater neuroprotective immune responses after TBI, which may be due to sex differences in hormone levels and endocrine function after injury (Stein, 2008). Sex-related differences in somatosensation (whose processing is localized to the PoCG) are modulated by the estrous cycle, which also predicates sex differences in neural plasticity (Alexander et al., 2018).

One of the clusters whose post-traumatic FC to the DAN (N₅) changes in a sex-dependent manner overlaps, to a large extent, with the fusiform gyrus. Poorer performance during face perception, object recognition, and reading has been reported in male TBI patients (Moore et al., 2010; Wang et al., 2018). Hypoconnectivity between the VAN and occipital cortex underlies attention deficits (Farrant and Uddin, 2016), and females’ ability to sustain attention is typically greater in the chronic stages of TBI (Ratcliff et al., 2007).

Functional correlation between the LN (N₉) and the lingual gyrus is directly involved in face recognition (Kanwisher et al., 1997; Rotshtein et al., 2001), which is involved in the Diagnostic Analysis of Nonverbal Accuracy task (on which males perform significantly better) (Nowicki and Carton, 1993; Babbage et al., 2018). The superior parietal lobule, which overlaps with N₃ and is involved in language processing, has altered FC with the FPN after TBI in a sex-dependent manner (Segal and Petrides, 2012; Banaszkiewicz et al., 2021). After injury, this structure undergoes BOLD signal reductions, which translate into altered FC between the superior parietal lobule and other regions (Sanchez-Carrion et al., 2008).

The inferior parietal lobule (IPL) deserves individual treatment here due to its well-documented and important sexual dimorphism (Frederikse et al., 1999), which may be responsible for sex differences in visuospatial ability (Culham and Kanwisher, 2001; Zago and Tzourio-Mazoyer, 2002). Emotion perception is among the primary functions of the IPL, which overlaps with N₁, N₅, N₁₅, and N₁₆ (Engelen et al., 2015). Following injury, the ability to perceive emotions evoked by facial and by other stimuli tends to degrade (Green et al., 2004; Bornhofen and McDonald, 2008a,b,c). Furthermore, females exhibit weaker deficits in emotion perception after injury (Rigon et al., 2016b). The connectome hub localized within the IPL is among the largest in the connectome; aside from emotion processing, the IPL is also involved in language production (Barbeau et al., 2017; Southwell et al., 2017; Jiao et al., 2020), which can be affected by TBI (DePompei and Hotz, 2001; Steel et al., 2017; Koebli et al., 2020) in a sex-dependent manner. Specifically, after injury, females typically preserve their overall language abilities better than males (Covassin and Bay, 2012; Eramudugolla et al., 2014).

Within FS-FAST, a weighted least-squares GLM implemented by the FS mri_glmfit function was used to identify cortical areas exhibiting significant FC to each seed RSN. Such areas consisted of one or more contiguous cortical surface clusters whose overall BOLD signals had statistically significant partial correlations ρ with the BOLD signal of the seed RSN, both at the acute (ρ_A) and chronic (ρ_C) timepoints (null hypothesis: ρ_A = ρ_C). The correlation values were compared across timepoints over the entire cortical surface, rather than only across clusters of significant correlations identified by the BOLD signal analysis. The FS-FAST mri_glmfit-sim function was used to evaluate cluster-wise statistical significance using the –cwp flag. Multiple comparison correction was implemented using –3 spaces flag. We chose the partial correlation coefficient ρ rather than the standard contrast effect size as a measure of FC because ρ ranges from −1 to 1, thus making interpretations and comparisons more straightforward for our purposes. This choice does not affect the results of the study because FS-FAST takes into account both when calculating statistical significance. Here and throughout, the effect sizes discussed are those pertaining to partial correlations ρ, i.e., to FCs. Let ρ_F and ρ_M denote the partial correlation coefficients of females and males, respectively. Similarly, let ρ_O and ρ_Y denote the partial correlation coefficients of older adults (OAs) and younger adults (YAs), i.e., between adults younger vs. older than 40 years, respectively. There are three effect size types considered in this study: (A) the difference Δρ_T = ρ_C−ρ_A, (quantifying FC changes with time), (B) the difference Δρ_S = ρ_F−ρ_M (quantifying differences between sexes), and (C) the difference Δρ_A = ρ_O−ρ_Y (quantifying FC differences between age groups). Effect sizes were calculated for each subject, and all subjects’ timeseries were then concatenated for group-level statistical analysis. Significant group differences between YAs and OAs and group differences between sexes were identified within the same GLM. The age cutoff of 40 years was selected partly for convenience, as this cutoff splits our sample roughly into half. Testing null hypotheses of group contrasts allowed us to identify ρ values that differed significantly across groups. Δp-values were computed clusterwise (i.e., at the cluster rather than vertex level). To facilitate reporting and tabulation, group-wise averages of partial correlations were calculated across clusters and effect sizes were reported using Cohen’s d multiplied by either −1 or +1, depending on whether the effect was associated with a decrease or increase in FC, respectively. Thus, for example, d = −0.2 indicates an FC decrease with an effect size of 0.2; analogously, d = +0.5 indicates an FC increase with an effect size of 0.5. To relate FCs to cognitive measures, the maximum absolute value of Δρ was first identified for each subject within each of that subject’s spatial cluster of statistical significance. In other words, for each cluster of significance that had been identified across all subjects, the peak effect size Δρ_max was first identified within each cluster and for each subject. Pearson’s product moment correlation coefficient between each maximum effect size Δρ_max and each cognitive score (EVMI, EVMD, WMS, IR, and PS) was then calculated, and its statistical significance was tested using Student’s t-test with N-2 degrees of freedom, N being the sample size. Figures visualizing statistical results were generated in Mathematica (Wolfram Research, Urbana-Champaign, IL, United States); cortical maps of significant statistical findings were generated in MATLAB (Mathworks, Natick, MA, United States).

To improve anatomo-functional localization, our FC analysis was implemented using the 17-network parcellation of Yeo et al. (2011) to define FC seeds, rather than the coarser seven-network (canonical) RSN parcellation. Because the 17-network parcellation is more granular, this allows one to localize age- and sex-specific effects on the cortex at relatively higher spatial resolution. Additionally, this reduces both (A) spatial filtering effects due to time series averaging and (B) cancellation effects whereby the anatomical patterns and distributions of correlation coefficients with opposite signs are lost (canceled out) when the coefficients are averaged over across relatively small cortical patches. Nevertheless, because the seven RSNs are more frequently studied, better understood, and easier to interpret, we choose to report, tabulate, and discuss results using the nomenclature of the seven-network parcellation scheme. For completeness, however, we also highlight notable anatomic and functional insights facilitated by our use of the 17-network RSN parcellation scheme. To summarize results obtained from our 17-RSN FC analysis, we leveraged the information in Figure 2, which maps the 17-RSN scheme onto the 7-RSN scheme, to map results obtained within the former onto the latter.

The YAs and OAs differ significantly in how the SMN, VAN, and DMN are functionally coupled to various cortical clusters (Figure 3 and Table 3). Here and throughout, d_Y and d_O stand for Cohen’s d (the effect sizes for YAs and OAs, respectively) multiplied by the sign of the change in FC associated with d. We find a significant age group difference in FC changes (d_Y = 0.05, d_O = −0.19, Δd_A = −0.24, p = 0.0027) between the SMN and a cortical cluster spanning a patch of lateral temporal cortex overlapping with the LN and DMN (Figure 3A). This cluster’s FC changes are significantly and negatively associated with changes in EVMI (r = −0.18, t₁₃₄ = −2.09, p = 0.02, q = 0.03, see cluster 36 in Table 2). Another significant age difference in FC change (d_Y = −0.04; d_O = −0.21, Δd_A = −0.25, p = 0.0305) is found between the VAN and a cluster spanning lateral occipital and inferior parietal regions (Figure 3B). Thirdly, a significant difference in FC change (d_Y = 0.10, d_O = −0.11, Δd_A = −0.21, p = 0.0078) is found between the DMN and a cluster spanning parts of the superior temporal, middle temporal, and anterior cingulate cortices (Figure 3C). Notably, OAs exhibit relatively larger FC decreases than YAs (Figure 4). In OAs, FC decreases across all three clusters whereas, in YAs, FC is relatively unchanged for the YAs, such that |d_Y| < |d_O|.

Table 4 lists (A) Cohen’s d_M and d_F, and (B) Cohen’s d_S for differences in FC change between sexes (Δρ_S). These data are tabulated for each seed-target pair; the seed is an RSN, and the target is a significant cortical cluster to which the RSN is coupled. Also listed is the overlap of each target cortical cluster with various RSNs. Here and throughout, d_F and d_M stand for Cohen’s d (the effect size for females and males, respectively) multiplied by the sign of the respective change in FC associated with d. RSN-cluster pairs are sorted in ascending order of males’ effect sizes, which range from d_M = −2.14 (FC decrease) to d_M = 0.77 (FC increase); females’ effect sizes range from d_F = −1.89 to d_F = 1.03. The clusters listed in Table 4 are also displayed on the cortical surface in Figure 5. The reason for which males—rather than females—were selected to determine the order of displays in Table 4 and Figure 5 is that males had been found to exhibit a wider range of FC changes. This made the comparison of their FCs relative to those of females easier because females’ relatively smaller average FC changes provided a pseudo-baseline, relative to which males’ FCs could be compared. To assist the reader in comparing females’ effect sizes relative to men’s, Supplementary Table 1 lists the data in Table 4 in ascending order of females’ effect sizes. A statistical comparison of the sexes across HCs identified no statistically significant results. Thus, across all seed networks, no significant difference in FC change across timepoints was found between sexes.

Each inset of Figure 5 displays a single seed RSN and the cortical clusters whose FCs to that RSN change in a way that is predicated significantly upon sex. Thus, as already stated, Figure 5 provides cortical displays of the results in Table 4. As Figure 5 and Table 4 indicate, males exhibit greater variability, across both source RSNs and target clusters, in the magnitudes of their FC changes. All seven RSNs exhibit sex-related differences in how their RS FCs to other areas change after injury (Figure 5 and Table 4). The magnitudes of most such changes are relatively smaller in females; for example, the effect size of the change in FC between the VN and the portion of the LN overlapping the insula is d_F = −0.06. By contrast, males’ FC changes are relatively larger in magnitude than females’ and most are negative (i.e., they involve FC decreases). Thus, FC between the VN and portion of the LN overlapping the insula changes significantly post-injury (d_M = −0.78, p < 0.0001). Males’ largest FC increases are between (A) the LN and the middle temporal portion of the DMN (d_M = 0.34, p < 0.0001), and (B) the FPN and the inferior parietal portion of the VN (d_M = 0.77, p < 0.0001, Figure 5).

Let Δd_S=d_F−d_M, where d_F and d_M are the effect sizes of females and males, respectively. Thus, Δd_s is the difference in effect sizes between sexes. Because |d_F| < |d_M| in most cases, most sex differences in FC change occur in males. Thus, Figure 5 indicates that the largest sex differences observed pertain to FC changes between portions of the VN and LN that overlap with the insula, entorhinal cortex, and other temporal regions. For example, in the insula and entorhinal cortex, males’ FCs decrease more than in females by |Δd_s| = 1.01 and by |Δd_s| = 0.52, respectively (p < 0.0001). The second-largest, sex-related difference pertains to changes in FC between the DMN and SMN, for which |Δd_s| = 0.72 (DMN to the SMN portion in the postcentral gyrus) and |Δd_s| = 0.64 (SMN to the DMN portion in the SFG), respectively. Larger effects are found in males (p < 0.0001 for both clusters, Figure 5 and Table 4).

Only a handful of seed-target pairs exhibit FC changes that are smaller in females (i.e., where females’ FC decreases more than males’ after mTBI), and most of these involve FCs between the DAN and other cortical regions, and between the VAN and other areas (Figures 5, 6C,D and Table 4). Examples include sex differences in FC between (A) the DAN and the superior frontal portion of the SMN (|Δd_s| = 0.25, p = 0.0019), (B) the DAN and the superior temporal portion of the LN (|Δd_s| = 0.60, p < 0.0001), and (C) the LN and the inferior temporal lobe portion of the DAN (|Δd_s| = 0.28, p = 0.0006). Other RSN-cluster pairs whose FC decreases are significantly larger in males include (A) the DAN and the entorhinal portion of the LN (|Δd_s| = 0.57, p < 0.0001), and (B) the VAN and the lateral occipital portion of the VN (|Δd_s| = 0.54, p < 0.0001) as shown in Figures 5, 6C,D and Table 4.

The clusters whose FCs to various RSNs change in ways that differ significantly by sex are mapped on the cortical surface in Figure 6 (for the VN, SMN, DAN, and VAN) and in Figure 7 (for the LN, FPN, and DMN). For each of the RSNs, there is at least one cluster whose FC to the seed RSN changes significantly in a sex-dependent manner. The pairs of regions exhibiting sex-dependent FC changes include:

Most clusters occur bilaterally, but do not typically overlap with a single RSN (Figures 6, 7).

The interaction between our main variables of sex and age were limited to small cortical clusters where significant differences were found in how FC changes depended on the sex × age interaction (Figure 8). The largest and most significant cluster is in the left supramarginal gyrus (p < 0.0010) and involves FC with the VN. The next most prominent clusters of interaction involve the VN seed and overlap with left rostral anterior cingulate cortex (p < 0.0010) and with the IPL (p < 0.0010).

Three FC clusters were found to be significantly different between OAs and YAs with mTBI (one in the left lateral part of the occipital lobe, and one in each hemisphere’s superior temporal lobe, Figures 3, 4). These clusters feature FC decreases that are significantly greater in OAs (Table 3), potentially reflecting their poorer outcomes, whether neurological, cognitive, or affective (Mosenthal et al., 2002; Mushkudiani et al., 2007; Senathi-Raja et al., 2010). In particular, the lateral occipital gyrus is recruited by visual memory, a cognitive process that is more severely affected in OAs than in YAs (Larsson and Heeger, 2006; Senathi-Raja et al., 2010; Nagy et al., 2012; Byom et al., 2019). The lateral occipital gyrus also exhibits reduced FC to the VAN after mTBI, a phenomenon we found to be associated with poorer immediate recall (EVMI) in OAs (see Figures 3, 4, and row 1 in Table 3). This abnormal connectivity between the lateral occipital gyrus and the VAN may underlie impairments of visual memory, since visual mnemonic aids supported by the lateral occipital gyrus can be employed as encoding strategies on verbal memory tasks to improve recall (Unsworth et al., 2019). Two other clusters were identified here whose FC changes differ significantly by age group, and both overlap substantially with the superior temporal gyri. These bilateral clusters exhibit FCs to the SMN and DMN, respectively, whose post-traumatic FC decreases are significantly larger in older adults (Table 3). The superior temporal lobe is involved in language processing and in emotion perception, both of which frequently degrade after TBI (Green et al., 2004; Bornhofen and McDonald, 2008a,c). Thus, our findings may provide insight on how age modulates the extent of cognitive impairments after mTBI.

The sex effects identified here involve all seven canonical RSNs. Typically, males demonstrated FC decreases compared to females, who either showed smaller decreases, no difference, or increases (Figures 5, 9, and Table 4). We also observed that the typical magnitude of FC changes in males is larger than in females (Figure 9 and Supplementary Figure 1). For these reasons, we highlight male sex as an independent risk factor for functional network degradation after mTBI, in agreement with previous work (Gupte et al., 2019; Robles et al., 2021). Tables 5–11 provide systematic interpretations for each result, cluster by cluster. Typically, across all seed networks, we found evidence for greater FC decreases in males; in many cases, decreases were correlated with poorer cognitive performance.

The largest sex differences observed pertain to FC changes between portions of the VN and LN that overlap with the insula, entorhinal cortex, and with other temporal regions. Interestingly, these regions have been highlighted extensively by FC studies of AD and are associated with memory deficits (e.g., Killiany et al., 2000; Jack et al., 2002; Chen et al., 2016). TBI patients are at higher risk of AD, especially after repeated brain injuries (Faden and Loane, 2015). Irimia et al. (2020) found overlap in the DMNs of TBI and AD patients that include medial and lateral areas of the temporal lobe. These authors found extensive commonalities in DMN FC between AD and TBI patients, and between both patient groups and HCs. The extent of AD-like FC degradation within the DMNs of geriatric TBI patients were predicted based on their acute cognitive assessment scores, with specificity and sensitivity comparable to AD blood and imaging biomarkers. These findings may be related to the fact that the thickness of entorhinal cortex can predict cognitive decline with age and cognitive impairment severity in AD (Kordower et al., 2001; Velayudhan et al., 2013; Zhou et al., 2016). This is only one example of the complex relationship between mTBI-related changes in structural and functional connectomics, whose exploration is outside the scope of this study. Our results do support previous research that highlights similarities between mTBI and AD in terms of their shared FC degradation trajectories, and the reader is referred to our previous study on this topic for further details (Irimia et al., 2020).

In HCs, we found no significant sex differences in FC changes between baseline and follow-up. Prior HC studies found relatively small changes in FC—and even no significant changes—across follow-up periods of 2–3 years (Odish et al., 2015; Ousdal et al., 2020). The absence of significant findings in our analysis is consistent with these prior studies and suggests strongly that our findings are related to mTBI effects rather than to typical aging across the timespan between imaging sessions. Presumably, any changes in HCs across 6 months, if any, are likely to be very small relative to those identified here in the mTBI group.

Although present, the interactions between the main effects of age and sex were limited to small clusters located primarily in the left hemisphere (IPL, supramarginal gyrus, and anterior cingulate cortex). Interpretation of the age × sex interaction within these regions is not straightforward due to their small spatial extent and effect size. Furthermore, the available functional neuroanatomy literature does not provide the spatial specificity required to formulate cautious interpretations of our findings at a level of spatial localization similar to that of the interaction clusters identified here.

In this study, FC analysis results were mapped from the 17 RSN-scheme onto the seven canonical RSN-scheme. The advantages of this technique included (A) improved anatomo-functional localization, and (B) greater ability to summarize results across RSNs, since the 7-RSN scheme has fewer networks. Nevertheless, we did not use the same scheme for both the analysis and the summary, which may have caused relatively small inaccuracies when mapping results from one scheme onto the other. One additional limitation of this study is that, for logistical and methodological reasons, we did not study changes in FC between cortical regions and the subcortex. Future studies, however, should investigate these changes. Because fMRI can only resolve low-frequency BOLD signals, FCs involving high-frequency brain activity cannot be captured well using this technique. Alternative approaches like electro- and magnetoencephalography should be used to confirm and extend our studies further (Irimia et al., 2013a,b).

To find age-related effects, we split our cohort into two groups, each consisting of adults older or younger than 40, respectively. This threshold was convenient partly because it split our sample into approximately equal groups, thus resulting in a relatively balanced statistical design. However, because age is a continuous (rather than discrete) variable, the threshold in question is relatively arbitrary and other cutoffs may lead to different results (Irimia et al., 2015). One alternative to our approach could involve studying FC changes as a function of age at injury; nevertheless, due to the complexity of our findings, we deemed our current strategy to be preferable. Furthermore, although we did not interpret the statistical interaction between age and sex, this interaction was included in the GLM and its effect was regressed from our analysis before calculating effect sizes and FCs. Future studies should explore both the age-by-sex interaction and how FCs change as a function of age at injury.

Our sample focused on mTBI, which may limit the generalizability of our findings to moderate or severe TBI patients. Whereas explicit analysis of these severities is required, we speculate that the effects of older age and male sex on FC changes and cognitive outcomes would be exacerbated in a moderate-to-severe TBI group (Konstantinou et al., 2018). Bonnelle et al. (2011) found increased FC of the DMN in moderate-to-severe TBI patients, which correlated with deficits in sustained attention, indicating that higher TBI severity may lead to greater FC changes and poorer cognitive outcome. Finally, our study is observational rather than interventional; development and testing of interventions for mTBI-related impairments is beyond the scope of our work.