Age‐ and Sex‐Specific Patterns in Adult Brain Network Segregation

ABSTRACT

The human brain is organized into several segregated associative and sensory functional networks, each responsible for various aspects of cognitive and sensory processing. These functional networks become less segregated over the adult lifespan, possibly contributing to cognitive decline that is observed during advanced age. To date, a comprehensive understanding of decreasing network segregation with age has been hampered by (1) small sample sizes, (2) lack of investigation at different spatial scales, (3) the limited age range of participants, and more importantly (4) an inadequate consideration of sex (biological females and males) differences. This study aimed to address these shortcomings. Resting‐state functional magnetic resonance imaging data were collected from 357 cognitively intact participants (18.2–91.8 years; 49.9 ± 17.1 years; 27.70 ± 1.72 MoCA score, 203 [56.8%] females), and the segregation index (defined as one minus the ratio of between‐network connectivity to within‐network connectivity) was calculated at three spatial scales of brain networks: whole‐brain network, intermediate sensory and associative networks, as well as core visual (VIS), sensorimotor (SMN), frontoparietal (FPN), ventral attention (VAN), dorsal attention (DAN), and default mode networks (DMN). Where applicable, secondary within‐, between‐, and pairwise connectivity analyses were also conducted to investigate the origin of any observed age and sex effects on network segregation. For any given functional metric, linear and quadratic age effects, sex effects, and respective age by sex interaction effects were assessed using backwards iterative linear regression modeling. Replicating previous work, brain networks were found to become less segregated across adulthood. Specifically, negative quadratic decreases in whole‐brain network, intermediate associative network, VAN, and DMN segregation index were observed. Intermediate sensory networks, VIS, and SMN exhibited negative linear decreases in segregation index. Secondary analysis revealed that this process of age‐related functional reorganization was preferential as functional connectivity was observed to increase either between anatomically adjacent associative networks (DMN‐DAN, FPN‐DAN) or between anterior associative and posterior sensory networks (VIS‐DAN, VIS‐DMN, VIS‐FPN, SMN‐DMN, and SMN‐FPN). Inherent sex differences in network segregation index were also observed. Specifically, whole‐brain, associative, DMN, VAN, and FPN segregation index was greater in females compared to males, irrespective of age. Secondary analysis found that females have reduced functional connectivity between associative networks (DAN‐VAN, VAN‐FPN) compared to males and independent of age. A notable linear age‐related decrease in FPN SI was also only observed for females and not males. The observed findings support the notion that functional networks reorganize across the adult lifespan, becoming less segregated. This decline may reflect underlying neurocognitive aging mechanisms like neural dedifferentiation, inefficiency, and compensation. The aging trajectories and rates of decreasing network segregation, however, vary across associative and sensory networks. This study also provides preliminary evidence of inherent sex differences in network organization, where associative networks are more segregated in females than males. These inherent sex differences suggest that female functional networks may be more efficient and functionally specialized compared to males across adulthood. Given these findings, future studies should take a more focused approach to examining sex differences across the lifespan, incorporating multimodal methodologies.

Resting state networks become less segregated across the adult lifespan, likely reflective of neurocognitive aging mechanisms (i.e., dedifferentiation, neural compensation, and neural inefficiency). Functional network reorganizational patterns are preferential and vary between resting state networks. Females and males exhibit inherent differences in network segregation, especially among associative networks, independent of age.

Summary.

• Functional networks become less segregated across the adult lifespan, possibly reflective of neurocognitive aging mechanisms like neural dedifferentiation, inefficiency, and compensation.

• Functional reorganization patterns are preferential and vary between functional networks.

• Females and males exhibit inherent differences in network segregation, especially among associative networks, independent of age.

1. Introduction

Aging is generally accompanied by a progressive decline in cognitive and motor skills, even in the absence of a disease or disorder (see Harada et al. 2013; Murman 2015 for reviews). Older adults (≥ 65 years), for example, exhibit reduced working memory (Cansino et al. 2013), executive functioning (Ferguson et al. 2021; Zelazo et al. 2004), and motor performance (Leversen et al. 2012), compared to younger adults (≤ 35 years). These cognitive differences are believed to stem, at least in part, from age‐related differences in brain function (Deery et al. 2023; Edde et al. 2021).

Resting‐state functional magnetic resonance imaging (rs‐fMRI) is readily used to non‐invasively and indirectly examine brain function. By analyzing the synchronization of neural activity across different brain regions during rest, rs‐fMRI provides insights into functional connectivity patterns, which in turn reflect the functional organization of the brain (Biswal et al. 1995; Chang and Glover 2010; Friston 1994). Research has shown that the brain is organized into multiple associative and sensory functional networks, which support higher‐order cognitive and sensory processing, respectively (Achard and Bullmore 2007; Hrybouski et al. 2021; Thomas Yeo et al. 2011). In younger adults, these networks are highly segregated, exhibiting greater within‐network functional connectivity relative to between‐network functional connectivity (Ballard et al. 2023; Chan et al. 2014; Grady et al. 2016; Hou et al. 2019; King et al. 2018; Kong et al. 2020; Pedersen et al. 2021; Varangis et al. 2019). This high degree of segregation is believed to facilitate specialized functional processing, supporting more efficient neural communication among brain regions belonging to the same network, and has also been shown to be associated with better cognitive performance irrespective of age (e.g., long‐term episodic memory and fluid processing) (Fornito et al. 2015; Gonneaud et al. 2017; Grady 2012; Park et al. 2004).

The degree of associative and sensory network segregation, however, gradually decreases as adults age (Ballard et al. 2023; Grady et al. 2016; Hou et al. 2019; Kong et al. 2020; Pedersen et al. 2021). This reorganization of functional network architecture appears to be associated with age‐associated increases in between‐network functional connectivity and relatively smaller reductions in within‐network functional connectivity (Betzel et al. 2014; Chan et al. 2014; Geerligs et al. 2015; King et al. 2018; Malagurski et al. 2020; Siman‐Tov et al. 2017; Varangis et al. 2019). Associative networks are particularly vulnerable to the effects of aging, with studies reporting negative quadratic decreases in associative network segregation with age compared to sensory networks that exhibit linear decreases (Malagurski et al. 2020). Specifically, associative network segregation is reported to decline gradually until midlife (50–60 years of age), after which this decline accelerates through to the 8th decade (ages 70–80) (Chan et al. 2014; Pedersen et al. 2021). This dichotomy suggests that age‐related network reorganization is likely not a uniform process across the cerebral cortex; however, additional studies are needed to better elucidate possible heterogenous reorganization patterns that may exist.

Adding to the complexity of age‐related differences in network organization, limited evidence of sex (biological females and males) differences in functional connectivity and network organization also exists (Allen et al. 2011; Ballard et al. 2023; Filippi et al. 2013; Foo et al. 2021; Goldstone et al. 2016; Hicks et al. 2023). For example, greater within‐network functional connectivity (Allen et al. 2011; Filippi et al. 2013; Foo et al. 2021) and reduced between‐network functional connectivity (Allen et al. 2011; Filippi et al. 2013) has been observed in female adults relative to males, with differences being the most pronounced among associative networks (Allen et al. 2011; Filippi et al. 2013). This pattern suggests that functional networks, especially associative functional networks, may be more segregated in females compared to their male counterparts. However, conflicting and limited findings complicate this interpretation. For instance, although Foo et al. (2021) observed greater within‐network connectivity across several networks in females, they also noted that whole brain network segregation, as measured by modularity, was lower in females relative to males (Foo et al. 2021). Similarly, Zang et al. reported reduced network segregation in young adult females compared to males (Zhang et al. 2016). More recent studies, however, suggest that females may exhibit greater network segregation amongst associative networks compared to males, with possible interactions between age and sex (Ballard et al. 2023; Hicks et al. 2023). Given that sex differences have also been reported in studies of age‐related cognitive decline (Levine et al. 2021; Lobo et al. 2023) and neurodegenerative disease prevalence (Home | Alzheimer's Association n.d.; Subramaniapillai et al. 2021), additional studies that more thoroughly investigate possible interactions of age and sex on functional network segregation across adulthood are required.

Network segregation has been quantified in the literature using several metrics, including modularity, participation coefficient, clustering coefficient, and the segregation index (Deery et al. 2023). Most of these metrics require additional preprocessing steps like matrix thresholding, which can complicate direct comparisons between participants due to inter‐individual variability in the number and location of retained connections at different thresholds (Rubinov and Sporns 2010). Despite this, the observed effects of declining network segregation across adulthood in cognitively intact individuals have been largely consistent, regardless of the metric used (see Deery et al. 2023 for a recent review). Unlike other metrics, the segregation index (defined as one minus the ratio of between‐network functional connectivity to within‐network connectivity), provides an average measure of network segregation without requiring threshold optimization or additional community detection (Chan et al. 2014). However, existing studies on age‐related network reorganization using the segregation index have been limited by (1) small sample sizes (Chong et al. 2019; Farràs‐Permanyer et al. 2019; Grady et al. 2016; Hou et al. 2019; King et al. 2018; Kong et al. 2020; Manza et al. 2020), (2) a narrow range of examined spatial scales (e.g., whole‐brain or specific core network) (Chan et al. 2014; Kong et al. 2020; Pedersen et al. 2021), and (3) restricted participant age ranges or group comparisons (Farràs‐Permanyer et al. 2019; Huang et al. 2015; Malagurski et al. 2020; Zonneveld et al. 2019). More importantly, (4) there is a need for sex to be examined as a moderating variable of age‐related alterations in segregation as existing studies have largely excluded it from their analyses (Chan et al. 2014; Chong et al. 2019; Farràs‐Permanyer et al. 2019; Grady et al. 2016; King et al. 2018; Kong et al. 2020; Manza et al. 2020; Pedersen et al. 2021). Thus, in the current study, we investigated age and sex differences in network segregation at various network scales across the adult lifespan in large sample of cognitively intact individuals. We hypothesized that the segregation index would decrease with age for all examined networks. Further, we expected that the segregation index would be greater in females than in males for associative networks.

2. Materials and Methods

MR data were obtained from the Calgary Normative Study (McCreary et al. 2020); an ongoing, single‐center study that performs MR brain imaging in individuals ranging in age from 18 to 90+ years with no cognitive deficits (as determined by Montreal Cognitive Assessment score ≥ 24 at time of study enrollment; Ihle‐Hansen et al. 2017). For the current study, resting‐state fMRI and T1‐weighted images were obtained from 357 individuals (154 males, 203 females), ranging in age from 18.2 to 91.8 years (mean ± SD: 49.9 ± 17.1 years). The T1‐weighted images were used for anatomical registration and region of interest (ROI) segmentation. Other recorded demographic variables included handedness (n = 327 [91.6%] right‐handed), MoCA score (mean 27.70 ± 1.72), as well as self‐reported high blood pressure (n = 52 [14.6%]), elevated cholesterol (n = 47 [13.2%]), and diabetes (n = 17 [4.8%]). Participant demographics are outlined in Table 1.

TABLE 1.

Participant demographics.

	Age group (years)
	18–29	30–39	40–49	50–59	60–69	70+
N	55	68	36	54	62	50
Females (N)	30 (54.5%)	38 (55.9%)	16 (44.4%)	36 (66.7%)	36 (58.1%)	28 (56.0%)
Mean age (years) ± SD	25.31 ± 2.91	34.81 ± 2.63	44.07 ± 2.63	55.12 ± 2.63	64.06 ± 2.59	75.05 ± 4.12
Mean MoCA ± SD	28.67 ± 1.35	27.96 ± 1.50	27.86 ± 1.57	27.83 ± 1.81	27.03 ± 1.66	26.7 ± 1.68
Mean FWD (mm) ± SD	0.05 ± 0.02	0.05 ± 0.03	0.05 ± 0.03	0.06 ± 0.03	0.06 ± 0.03	0.09 ± 0.03
Mean education (years) ± SD	16.62 ± 2.23	16.78 ± 2.59	17.44 ± 2.90	16.28 ± 2.37	15.42 ± 3.19	15.12 ± 2.84
High cholesterol (N)	0 (0.0%)	6 (8.8%)	3 (8.3%)	6 (11.1%)	11 (17.7%)	17 (34.0%)
Diabetes (N)	1 (1.8%)	0 (0.0%)	0 (0.0%)	4 (7.4%)	4 (6.5%)	8 (16.0%)
High blood pressure (N)	0 (0.0%)	1 (1.5%)	1 (2.8%)	8 (14.8%)	18 (29.0%)	21 (42.0%)

Abbreviations: FWD = average framewise displacement, MoCA = Montreal Cognitive Assessment score, SD = standard deviation.

All imaging was performed with a 3 T MR scanner (Discovery MR750; General Electric Healthcare, Waukesha, Wisconsin, USA) using a 12‐channel head, neck, and spine imaging coil. Resting‐state fMRI images were collected with a T2*‐weighted gradient‐recalled echo, echo planar imaging sequence in an oblique‐axial orientation (TR/TE = 2000/30 ms, flip angle = 70, 64 × 64 matrix, 240‐mm field of view, 3.8‐mm slice thickness, 150 volumes, ~5 min in duration). Participants were not explicitly instructed to keep their eyes open or closed during rs‐fMRI acquisition. To minimize the impact of participant fatigue or sleep, rs‐fMRI images were collected early in the scanning session, as part of the Calgary Normative Study's comprehensive MRI protocol, which spans approximately 1 h in total acquisition time. T1‐weighted anatomical images were collected as a 3D volume in either a sagittal or coronal orientation (TR/TE = 7/2.5 ms, flip angle = 8°, 256 × 256 in‐plane matrix, 256 slices, 1‐mm isotropic resolution).

2.1. 3D T1‐Weighted Image Processing

T1‐weighted images were segmented into the 17‐network 200‐node parcellation scheme of the Schaefer‐Yeo functional atlas (Fischl 2012; Schaefer et al. 2018) using FreeSurfer v6.0.0, as per previous MR studies of brain aging (Betzel et al. 2014; Chan et al. 2014; Malagurski et al. 2020). The 17 networks were further reduced to seven core and more readily studied functional networks (Edde et al. 2021), including the visual (VIS), somatosensory (SMN), default mode (DMN), ventral attention (VAN), dorsal attention (DAN), frontoparietal (FPN), and limbic (LIM) networks; however, the LIM (14 nodes) was excluded from analysis due to rs‐fMRI signal dropout in fronto‐orbital and temporal brain regions. Network‐specific node assignments are listed in Table S1.

2.2. Resting‐State Functional Magnetic Resonance Imaging Processing

The rs‐fMRI data processed using the tools provided by the FMRIB's Software Library (FSL, v 6.0; https://fsl.fmrib.ox.ac.uk/fsl). The processing pipeline consisted of skull‐stripping using BET (Smith 2002), motion correction using MCFLIRT (Jenkinson et al. 2002), interleaved slice timing correction, and 6‐mm FWHM spatial smoothing (Smith et al. 2004). Spatial independent component analysis was then performed using MELODIC (Beckmann and Smith 2004) and neuronal and non‐neuronal components were automatically classified using FIX (Salimi‐Khorshidi et al. 2014). The processed rs‐fMRI data were registered to the T1‐weighted images using FLIRT (Greve and Fischl 2009) based on a linear boundary‐based registration (six degrees of freedom) and visually assessed for registration accuracy. Based on the Schaefer‐Yeo parcellation, the average timeseries for the 200 nodes were extracted and denoised using Python NiBabel libraries (10.5281/zendo.7633628) (Baum et al. 2020; Geerligs et al. 2015; Malagurski et al. 2020). This included six motion parameters (as estimated by MCFLIRT), their derivates, squares, and squared derivates, as well as the average signal estimates from white matter tissue and cerebrospinal fluid. High‐motion rs‐fMRI volumes (≥ 0.9 mm relative framewise displacement) were also censored (zeroed) from timeseries, along with one volume before and two volumes after the high motion volume (four volumes in total). Temporal bandpass filtering in the 0.008–0.15 Hz frequency range was also applied to the timeseries. If ≥ 20% of rs‐fMRI volumes were censored the subject was excluded from additional analysis due to the decrease in the amount of data relative to other participants (n = 0 subjects were censored) (Smyser et al. 2010).

2.3. Resting‐State Functional Connectivity Matrix Generation

The average time‐series for each of the 186 nodes was used to estimate the Pearson correlation coefficient (r) between each combination of node pairs. Node‐pairs were classified as a within‐ or between‐network connection. Fisher's r‐to‐z transformation was applied to the subject‐specific 186 × 186 functional connectivity matrices. The diagonal element of the matrix and negative z‐values were zeroed (Hyde and Jesmanowicz 2012; R: The R Project for Statistical Computing n.d.). Negative values were zeroed as the physiological basis of these connections is heavily debated and not well understood (Fornito et al. 2015; Power et al. 2018; Rubinov and Sporns 2011; Schwarz and McGonigle 2011). Moreover, Chan and colleagues recommend using positive values to calculate the segregation index (SI) (Chan et al. 2014).

2.4. Calculation of Segregation Index (SI), Mean Within‐ and Between‐Network Connectivity, and Pairwise Functional Connectivity

SI is a ratio that estimates the relative contributions of within‐and between‐network connectivity over a series of nodes. Specifically, SI is defined as one minus the ratio of between‐network functional connectivity to within‐network connectivity:

$$\mathrm{SI}=1-\frac{\overline{z_b}}{\overline{z_w}}$$

where, $\overline{z_w}$ is the mean z correlation value for node‐pairs that are connected within a network (e.g., VAN node 1 and VAN node 2) and $\overline{z_b}$ is the mean z correlation value for all node‐pairs that are connected between that network and all other networks (e.g., VAN node 1 and DMN node 1). SI values approaching 1 suggests greater within‐ than between‐network connectivity, and thus greater network segregation. Lower SI values reflect a greater contribution of between‐network connectivity and thus lower network segregation.

SI was computed across three spatial scales of brain networks: whole brain; intermediate sensory and associative networks; and core VIS, SMN, FPN, DMN, DAN, and VAN networks. For each of the nine networks, mean within‐ and between‐network connectivity was also computed to evaluate age‐ and sex‐effects on these measures of network organization. In a secondary analysis, functional connectivity between pairs of core networks (pairwise functional connectivity) was computed to evaluate whether age‐ and sex‐effects were omnipresent between all core network pairs. Mean functional connectivity across all node pairs was also computed for supplementary statistical analysis.

2.5. Statistical Analysis

Statistical analysis was performed using R (V.4.2.1) (R: The R Project for Statistical Computing n.d.). Welch's t‐test's (α = 0.05) was used to determine if (1) the average age of participants, (2) average functional connectivity, and (3) average MoCA score was statistically different between females and males (Ross and Willson 2017). The Kolmogorov–Smirnov (KS) test (α = 0.05) was used to determine if the age distribution differed between females and males (Massey 1951). All examined functional connectivity metrics (SI, within‐network, between‐network, and pairwise functional connectivity) were visually and quantitively assessed for normality by examining histograms of the data and the KS test, respectively.

2.6. Backward Iterative Linear Regression Modeling

A backwards iterative linear regression modeling approach was used to identify the most appropriate linear model for each examined metric (SI, within‐network, between‐network, and pairwise connectivity) across each network or network pair (Breiman 1995; Derksen and Keselman 1992; Harrell 2015; Keith 2019; Weisberg 1985). Based on existing literature (Ballard et al. 2023; Betzel et al. 2014; Chan et al. 2014; Foo et al. 2021; Goldstone et al. 2016; Hicks et al. 2023; Pedersen et al. 2021), the initial comprehensive linear model accounted for linear and quadratic age effects, sex effects, and interactions between age and sex.

$$\text{Functional Metric}=\mathrm{Age}+{\mathrm{Age}}^2+\mathrm{Sex}+\mathrm{Age}\times \mathrm{Sex}+{\mathrm{Age}}^2\times \mathrm{Sex}$$

Next, least significant terms (i.e., those with the largest p‐values) were iteratively removed from the more complex models and reduced complexity models re‐estimated. Successive models were compared using both the Akaike Information Criterion (AIC) and Analysis of Variance (ANOVA) F‐test to statistically assess differences in model fits. Specifically, a reduced model was accepted only if both AIC and ANOVA results indicated that the reduced model provided a comparable or better fit than the more complex model, without significantly losing explanatory power (Akaike 1974; Faraway 2002). This iterative process continued for each metric until either AIC no longer improved (i.e., became more negative) or ANOVA (i.e., p < 0.05) indicated that removing a term significantly worsened the degree of variation explained. Once these conditions were no longer satisfied, the last accepted iteration was selected as the final best‐fitting model, irrespective of the statistical significance of individual predictor terms.

To account for multiple comparisons and mitigate the risk of inflated Type I errors, Holm‐Bonferroni multiple comparisons correction was applied based on the spatial scale of identical metrics (Holm 1979). Specifically, intermediate network, individual network, and pairwise connectivity results were adjusted for 2, 6, and 15 comparisons, respectively.

2.7. Supplementary Analysis

Supplementary linear model analyses of network SI were also examined to account for covariates pertaining to individual health status, years of education, handedness, and MoCA score. Specifically, a self‐reported diagnosis of or use of medication to treat diabetes, high cholesterol, and high blood pressure, were included as nuisance variables to control for differences in health states amongst individuals. Intra‐individual differences in mean functional connectivity and mean relative framewise displacement (i.e., motion) were also accounted for in supplementary linear models. Supplementary analyses were done by individually adding the nuisance variables to the final model for each network. A Pearson R correlation analysis between the various segregation indices and the MoCA score was also done. Holm‐Bonferroni multiple comparisons correction was applied to these analyses based on the spatial scale of the metric.

3. Results

No statistical significance between males and females in terms of average age (t _328.13 = −0.53, p = 0.601), average functional connectivity (t _299.32 = 1.92, p = 0.06), average motion (t _328.94 = 1.08, p = 0.279) or age distribution (D = 7.80 × 10⁻², p = 0.654) was observed. Statistical significance was observed in the average MoCA score between males and females (t _330.0.31 = −2.40, p = 0.017), with the average MoCA score being higher in females (mean ± SD: 27.9 ± 1.72) than males (mean ± SD: 27.5 ± 1.71). Average head motion (framewise displacement) was observed to be significantly, positively correlated with age (r = 0.33, t ₃₅₀ = 6.60, p < 0.001).

3.1. Backward Iterative Linear Regression Modeling

The final linear model for SI varied between spatial scales and respective networks (Table 2). Based on AIC and ANOVA tests, all excluded terms were iteratively deemed to statistically not improve model fit (AIC_complex model > AIC_reduced model) or better model the variance in the data (ANOVA, p > 0.05). Similar results were observed for within‐, between‐network, and pairwise connectivity. Final models for all metrics and networks are provide in Tables 3, 4, 5.

TABLE 2.

Summary of backwards iterative linear regression modeling for functional network segregation indices (SI).

Network	Linear model	Adjusted R 2	AIC	ANOVA
Whole‐brain	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.213	−1177.80	F = 0.01, p = 0.936
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}$	0.215	−1179.79
				F = 0.51, p = 0.478
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathbf{sex}$	0.216	−1181.28
				F = 0.32, p = 0.570
	$\mathrm{SI}~\mathbf{ag}{\mathbf{e}}^{\mathbf{2}}+\mathbf{sex}$	0.218	−1182.95
Sensory	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.074	−490.25	F = 0.25, p = 0.614
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.076	−491.99
				F = 0.37, p = 0.544
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}$	0.077	−493.62
				F = 0.04, p = 0.854
	$\mathrm{SI}~\mathbf{age}+\mathrm{sex}$	0.080	−495.59
Associative	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.168	−1215.42	F = 1.53, p = 0.217
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathbf{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.168	−1215.87
				F < 0.001, p = 0.969
	$\mathrm{SI}~\mathbf{ag}{\mathbf{e}}^{\mathbf{2}}+\mathbf{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.170	−1217.87
Visual network (VIS)	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.079	−710.19	F = 0.06, p = 0.813
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.081	−712.13
				F = 0.18, p = 0.674
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}$	0.084	−713.95
				F = 0.21, p = 0.645
	$\mathrm{SI}~\mathbf{age}+\mathrm{sex}$	0.086	−715.74
Somatosensory network (SMN)	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.102	−734.59	F = 0.01, p = 0.914
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.105	−736.58
				F = 0.22, p = 0.636
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}$	0.107	−738.35
				F = 0.19, p = 0.666
	$\mathrm{SI}~\mathbf{age}+\mathrm{sex}$	0.109	−740.17
				F = 0.29, p = 0.590
	$\mathrm{SI}~\mathbf{age}$	0.110	−741.87
Frontoparietal network (FPN)	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathbf{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.081	−919.12	F = 1.72, p = 0.191
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathbf{sex}+\mathbf{age}\times \mathbf{sex}$	0.081	−919.38
				F = 0.63, p = 0.429
	$\mathrm{SI}~\mathrm{age}+\mathbf{sex}+\mathbf{age}\times \mathbf{sex}$	0.081	−920.74
Dorsal attention network (DAN)	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.019	−882.22	N/A
Ventral attention network (VAN)	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.198	−798.59	F = 0.03, p = 0.868
	$\mathrm{SI}~\mathrm{age}+\mathbf{ag}{\mathbf{e}}^{\mathbf{2}}+\mathbf{sex}+\mathrm{age}\times \mathrm{sex}$	0.200	−800.56
				F = 0.28, p = 0.594
	$\mathrm{SI}~\mathrm{age}+\mathbf{ag}{\mathbf{e}}^{\mathbf{2}}+\mathbf{sex}$	0.208	−802.27
				F = 0.44, p = 0.507
	$\mathrm{SI}~\mathbf{ag}{\mathbf{e}}^{\mathbf{2}}+\mathbf{sex}$	0.208	−803.82
Default mode network (DMN)	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}+{\mathrm{age}}^2\times \mathrm{sex}$	0.102	−779.31	F = 0.41, p = 0.522
	$\mathrm{SI}~\mathrm{age}+\mathrm{ag}{\mathrm{e}}^2+\mathrm{sex}+\mathrm{age}\times \mathrm{sex}$	0.103	−780.89
				F = 0.26, p = 0.611
	$\mathrm{SI}~\mathrm{age}+\mathbf{ag}{\mathbf{e}}^{\mathbf{2}}+\mathbf{sex}$	0.105	−782.63
				F = 1.28, p = 0.260
	$\mathrm{SI}~\mathbf{ag}{\mathbf{e}}^{\mathbf{2}}+\mathbf{sex}$	0.105	−783.34

Note: The initial full model included age, age², sex, age × sex, and age² × sex as predictor variables. During each iteration, the least significant predictor was removed, and the regression model was re‐estimated. Model comparisons were conducted using the Akaike Information Criterion (AIC) and the Analysis of Variance (ANOVA) F‐test. Adjusted R ² values are also provided. A reduced model was accepted only if AIC and ANOVA results demonstrated that excluding a specific predictor either improved the model fit or maintained a statistically comparable fit, indicating that the excluded term did not significantly enhance the explanation of data variation. This iterative process continued for each metric until either AIC no longer improved (i.e., became more negative) or ANOVA indicated that removing a term significantly worsened the fit. This approach ensured selection of the most statistically appropriate model for each network segregation index. All intermediate models demonstrating improved fit during the iterative process are included in the table. When no further improvements were observed (not shown), the final model was selected and is presented in the table as the final and most appropriate regression model for each respective metric. Importantly, final models could include non‐significant terms if they improved the overall fit or provided a better explanation of data variation. Statistically significant age, age², sex, age × sex, and age² × sex (p < 0.05) appear in bold.

TABLE 3.

Summary of predictor estimates, standard errors (SE), t‐values, and uncorrected (p _raw ), corrected (p _cor ) p‐values, where appropriate, for the final whole‐brain and intermedia associative and sensory network regression models corresponding to within‐network, between‐network, and segregation index metrics.

Functional network	Predictor estimate (SE)	t	p raw	p cor
Whole‐brain
Within
age	—	—	—	—
age2	−2.28 × 10−6 (1.66 × 10−6)	−1.37	0.170	N/A
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	4.61 × 10−4 (1.34 × 10−4)	3.44	≤ 0.001 ***	N/A
age2	—	—	—	—
sex	−1.07 × 10−2 (4.63 × 10−3)	−2.32	0.021 *	N/A
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Segregation index
age	—	—	—	—
age2	−1.37 × 10−5 (1.41 × 10−6)	−9.70	≤ 0.001 ***	N/A
sex	1.42 × 10−2 (4.90 × 10−3)	2.90	0.004 **	N/A
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Sensory
Within
age	—	—	—	—
age2	−4.08 × 10−6 (2.60 × 10−6)	−1.57	0.112	0.174
sex	−1.58 × 10−2 (9.04 × 10−3)	−1.75	0.082	0.163
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	8.07 × 10−4 (2.23 × 10−4)	3.62	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	−1.72 × 10−2 (7.70 × 10−3)	−2.24	0.026 *	0.0520
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Segregation index
age	−2.06 × 10−3 (3.72 × 10−4)	−5.54	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	2.08 × 10−2 (1.28 × 10−2)	1.62	0.107	0.107
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Associative
Within
age	1.72 × 10−3 (1.07 × 10−3)	1.60	0.110	0.219
age2	−1.83 × 10−5 (1.07 × 10−5)	−1.72	0.087	0.174
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	2.60 × 10−3 (1.27 × 10−3)	2.06	0.041 *	0.041 *
age2	−2.36 × 10−5 (1.25 × 10−5)	−1.88	0.061	0.121
sex	4.13 × 10−2 (3.94 × 10−2)	1.05	0.295	0.295
age × sex	−2.56 × 10−3 (1.71 × 10−3)	−1.50	0.135	0.269
age2 × sex	2.65 × 10−5 (1.69 × 10−5)	1.57	0.118	0.236
Segregation index
age	—	—	—	—
age2	−7.39 × 10−6 (2.02 × 10−6)	−3.65	≤ 0.001 ***	≤ 0.001 ***
sex	3.34 × 10−2 (8.60 × 10−3)	3.88	≤ 0.001 ***	≤ 0.001 ***
age × sex	—	—	—	—
age2 × sex	−3.91 × 10−6 (2.70 × 10−6)	−1.45	0.148	0.296

Note: Predictor variables not included in the final model are represented with a dash (—). p‐values were corrected across the appropriate scales, regardless of whether the predictor variable was part of all network final models. Statistically significant effects (p < 0.05) appear in bold.

Abbreviation: SE = standard error.

^* p < 0.05.

^** p < 0.01.

^*** p ≤ 0.001.

TABLE 4.

Summary of predictor estimates, standard errors (SE), t‐values, and uncorrected (p _raw ), corrected (p _cor ) p‐values for the final visual, sensorimotor, frontoparietal, dorsal attention, ventral attention, and default mode network regression models corresponding to within‐network, between‐network, and segregation index metrics.

Functional network	Predictor estimate (SE)	t	p raw	p cor
Visual
Within
age	—	—	—	—
age2	−1.02 × 10−6 (2.70 × 10−6)	−0.37	0.701	1.00
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	7.18 × 10−4 (1.51 × 10−4)	4.74	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	−8.93 × 10−3 (5.22 × 10−3)	−1.71	0.088	0.240
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Segregation index
age	−1.57 × 10−3 (2.73 × 10−4)	−5.75	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	1.56 × 10−2 (9.43 × 10−3)	1.66	0.100	0.238
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Sensorimotor
Within
age	—	—	—	—
age2	−5.73 × 10−6 (3.36 × 10−6)	−1.71	0.089	0.365
sex	−2.33 × 10−2 (1.17 × 10−2)	−1.99	0.047 *	0.188
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	5.34 × 10−4 (1.67 × 10−4)	3.20	0.002 **	0.006 **
age2	—	—	—	—
sex	−1.01 × 10−2 (5.77 × 10−3)	−1.75	0.080	0.240
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Segregation index
age	−1.77 × 10−3 (2.64 × 10−4)	−6.72	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Frontoparietal
Within
age	3.63 × 10−3 (1.80 × 10−3)	2.02	0.044 *	0.176
age2	−3.20 × 10−5 (1.78 × 10−5)	−1.80	0.073	0.365
sex	1.33 × 10−1 (5.60 × 10−2)	2.37	0.018 *	0.090
age × sex	−6.00 × 10−3 (2.42 × 10−3)	−2.48	0.014 *	0.084
age2 × sex	5.42 × 10−5 (2.40 × 10−5)	2.26	0.025 *	0.148
Between
age	2.29 × 10−3 (1.28 × 10−3)	1.79	0.075	0.150
age2	−1.89 × 10−5 (1.27 × 10−5)	−1.49	0.138	0.554
sex	4.29 × 10−2 (3.99 × 10−2)	1.07	0.283	0.283
age × sex	−2.47 × 10−3 (1.73 × 10−3)	−1.43	0.154	0.925
age2 × sex	2.50 × 10−5 (1.71 × 10−5)	1.46	0.145	0.872
Segregation index
age	−3.30 × 10−4 (3.10 × 10−4)	−1.06	0.288	1.00
age2	—	—	—	—
sex	6.00 × 10−2 (2.14 × 10−2)	2.81	0.005 **	0.021 *
age × sex	−1.24 × 10−3 (4.13 × 10−4)	−3.00	0.003 **	0.017 *
age2 × sex	—	—	—	—
Dorsal attention
Within
age	8.04 × 10−4 (2.70 × 10−4)	2.97	0.003 **	0.018 *
age2	—	—	—	—
sex	−3.78 × 10−2 (9.32 × 10−3)	−4.05	≤ 0.001 ***	≤ 0.001 ***
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	6.14 × 10−4 (1.55 × 10−4)	3.96	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	−1.73 × 10−2 (5.36 × 10−3)	−3.22	≤ 0.001 ***	0.008 **
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Segregation index
age	1.69 × 10−3 (2.10 × 10−3)	0.80	0.422	1.00
age2	−1.55 × 10−5 (2.08 × 10−5)	−0.74	0.458	1.00
sex	1.15 × 10−1 (6.54 × 10−2)	1.76	0.079	0.238
age × sex	−4.90 × 10−3 (2.83 × 10−3)	−1.73	0.085	0.423
age2 × sex	−4.09 × 10−5 (2.81 × 10−5)	1.46	0.146	0.873
Ventral attention
Within
age	4.23 × 10−3 (1.71 × 10−3)	2.48	0.014 *	0.070
age2	−5.05 × 10−5 (1.69 × 10−5)	−2.97	0.003 **	0.018 *
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	2.33 × 10−3 (9.84 × 10−4)	2.37	0.018 **	0.055
age2	−2.07 × 10−5 (9.77 × 10−6)	−2.11	0.035 *	0.176
sex	−1.09 × 10−2 (5.20 × 10−3)	−2.10	0.036 *	0.144
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Segregation index
age	—	—	—	—
age2	−2.05 × 10−5 (2.40 × 10−6)	−8.56	≤ 0.001 ***	≤ 0.001 ***
sex	3.85 × 10−2 (8.33 × 10−3)	4.62	≤ 0.001 ***	≤ 0.001 ***
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Default mode
Within
age	1.89 × 10−3 (1.25 × 10−3)	1.51	0.131	0.393
age2	−2.15 × 10−5 (1.24 × 10−5)	−1.74	0.082	0.365
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Between
age	—	—	—	—
age2	2.96 × 10−6 (1.26 × 10−6)	2.35	0.019 *	0.117
sex	−7.23 × 10−3 (4.38 × 10−3)	−1.65	0.010 *	0.053
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Segregation index
age	—	—	—	—
age2	−1.34 × 10−5 (2.47 × 10−6)	−5.42	≤ 0.001 ***	≤ 0.001 ***
sex	3.36 × 10−2 (8.57 × 10−3)	3.92	≤ 0.001 ***	≤ 0.001 ***
age × sex	—	—	—	—
age2 × sex	—	—	—	—

Note: Predictor variables not included in the final model are represented with a dash (—). p‐values were corrected across the appropriate scales, regardless of whether the predictor variable was part of all networks final models. Statistically significant effects (p < 0.05) appear in bold.

Abbreviation: SE = standard error.

^* p < 0.05.

^** p < 0.01.

^*** p ≤ 0.001.

TABLE 5.

Summary of predictor estimates, standard errors (SE), t‐values, and uncorrected (p _raw ), corrected (p _cor ) p‐values for the pairwise functional connectivity regression models.

Functional system	Predictor estimate (SE)	t	p raw	p cor
Sensory—sensory connections
VIS‐SMN
age	8.07 × 10−4 (2.23 × 10−4)	3.62	≤ 0.001 ***	0.004 **
age2	—	—	—	—
sex	−1.72 × 10−2 (7.70 × 10−3)	−2.24	0.026 *	0.260
age × sex	—	—	—	—
age2 × sex	—	—	—	—
Associative—associative connections
DAN‐DMN
age	4.34 × 10−4 (1.42 × 10−4)	3.05	0.002 **	0.024 *
age2	—	—	—	—
sex	−1.13 × 10−2 (4.90 × 10−3)	−2.31	0.022 *	0.239
age × sex	—	—	—	—
age2 × sex	—	—	—	—
DAN‐FPN
age	7.40 × 10−4 (1.80 × 10−4)	4.11	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	−1.86 × 10−2 (6.21 × 10−3)	−2.99	0.003 **	0.039 *
age × sex	—	—	—	—
age2 × sex	—	—	—	—
DAN‐VAN
age	2.76 × 10−3 (1.29 × 10−3)	2.15	0.032 *	0.251
age2	−2.28 × 10−5 (1.28 × 10−5)	−1.79	0.075	0.902
sex	−2.44 × 10−2 (6.80 × 10−3)	−3.59	≤ 0.001 ***	0.005 **
age × sex	—	—	—	—
age2 × sex	—	—	—	—
VAN‐DMN
age	1.35 × 10−3 (9.15 × 10−4)	1.48	0.140	0.280
age2	−1.29 × 10−5 (9.08 × 10−6)	−1.42	0.155	1.00
sex	−7.38 × 10−3 (4.83 × 10−3)	−1.53	0.128	0.789
age × sex	—	—	—	—
age2 × sex	—	—	—	—
VAN‐FPN
age	1.96 × 10−3 (9.61 × 10−4)	2.04	0.042 *	0.251
age2	−1.74 × 10−5 (9.54 × 10−6)	−1.82	0.069	0.902
sex	−1.86 × 10−2 (5.08 × 10−3)	−3.67	≤ 0.001 ***	0.004 **
age × sex	—	—	—	—
age2 × sex	—	—	—	—
FPN‐DMN
age	3.15 × 10−3 (1.46 × 10−3)	2.16	0.031 *	0.251
age2	−3.19 × 10−5 (1.44 × 10−5)	−2.21	0.028 *	0.390
sex	9.25 × 10−2 (4.53 × 10−2)	2.04	0.042 *	0.377
age × sex	−4.70 × 10−3 (1.96 × 10−3)	−2.40	0.017 *	0.255
age2 × sex	4.93 × 10−5 (1.94 × 10−5)	2.54	0.012 *	0.174
Sensory—associative connections
VIS‐DAN
age	8.13 × 10−4 (2.22 × 10−4)	3.67	≤ 0.001 ***	0.003 **
age2	—	—	—	—
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
VIS‐DMN
age	—	—	—	—
age2	6.34 × 10−6 (1.46 × 10−6)	4.43	≤ 0.001 ***	≤ 0.001 ***
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
VIS‐FPN
age	7.30 × 10−4 (1.59 × 10−4)	4.60	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	−8.72 × 10−3 (5.48 × 10−3)	−1.59	0.113	0.789
age × sex	—	—	—	—
age2 × sex	—	—	—	—
VIS‐VAN
age	2.39 × 10−3 (1.23 × 10−3)	1.94	0.053	0.251
age2	−1.91 × 10−5 (1.22 × 10−5)	−1.57	0.118	1.00
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
SMN‐DAN
age	4.62 × 10−4 (2.22 × 10−4)	2.08	0.038 *	0.251
age2	—	—	—	—
sex	−1.94 × 10−2 (7.65 × 10−3)	−2.53	0.012 *	0.141
age × sex	—	—	—	—
age2 × sex	—	—	—	—
SMN‐DMN
age	5.08 × 10−4 (1.73 × 10−4)	2.93	0.004 **	0.032 *
age2	—	—	—	—
sex	−1.11 × 10−2 (5.98 × 10−3)	−1.85	0.065	0.521
age × sex	—	—	—	—
age2 × sex	—	—	—	—
SMN‐FPN
age	7.49 × 10−4 (1.68 × 10−4)	4.46	≤ 0.001 ***	≤ 0.001 ***
age2	—	—	—	—
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—
SMN‐VAN
age	2.71 × 10−3 (1.55 × 10−3)	1.75	0.080	0.251
age2	−2.45 × 10−5 (1.53 × 10−5)	−1.59	0.112	1.00
sex	—	—	—	—
age × sex	—	—	—	—
age2 × sex	—	—	—	—

Note: Predictor variables not included in the final model are represented with a dash (—). p‐values were corrected across the appropriate scales, regardless of whether the predictor variable was part of all network final models. Statistically significant effects (p < 0.05) appear in bold.

Abbreviation: SE = standard error.

^* p < 0.05.

^** p < 0.01.

^*** p ≤ 0.001.

3.2. Whole‐Brain Network Segregation

Whole‐brain network SI decreased in a negative quadratic manner with increasing age (b = −2.28 × 10⁻⁶, p < 0.001) and was also significantly higher in females than males, independent of age (b = 1.42 × 10⁻², p = 0.004), Figure 1a. Specifically, whole‐brain network SI gradually decreases until the 6th decade of life (ages 50–60), after which the decline accelerates into the 8th decade (ages 70–80). Whole‐brain between‐network connectivity increased in a linear manner with increasing age (b = 4.61 × 10⁻⁴, p < 0.001) and was significantly lower in females than males, independent of age (b = −1.07 × 10⁻², p = 0.021). No other statistically significant age, age², sex, age × sex, age² × sex interaction effects were observed for any whole‐brain metrics (p _cor  > 0.05, Table 3).

FIGURE 1.

(a) Whole‐brain; (b) sensory; (c) associative; (d) visual (VIS); (e) sensorimotor (SMN); (f) frontoparietal (FPN); (g) dorsal attention (DAN); (h) ventral attention (VAN); and (i) default mode network (DMN) segregation index (SI) with age. Sensory, VIS, and SMN SI exhibited linear decreases with increasing age, while quadratic decreases were observed for whole brain, associative networks, VAN, and DMN SI. Age‐related decreases in FPN SI were only observed in females (pink). Whole‐brain, associative, DMN, VAN, and FPN SI were greater in females compared to males (green), independent of age. Shaded areas represent 95% confidence intervals of the linear/quadratic fit.

3.3. Sensory and Associative Network Segregation

Sensory network SI decreased in a linear manner with increasing age (b = −2.06 × 10⁻³, p _cor  < 0.001), Figure 1b. Sensory network between‐network connectivity increased in a linear manner with increasing age (b = 8.07 × 10⁻⁴, p _cor  < 0.001). Associative network SI decreased in a negative quadratic manner with increasing age (b = −7.39 × 10⁻⁶, p _cor  < 0.001) and was also significantly higher in females than males, independent of age (b = 3.34 × 10⁻², p _cor  < 0.001), Figure 1c. Specifically, associative network SI gradually decreases until the 6th decade of life (ages 50–60), after which the decline accelerates into the 8th decade (ages 70–80). Associative network between‐network connectivity increased in a linear manner with increasing age (b = 2.60 × 10⁻³, p _cor  = 0.041). No other statistically significant age, age², sex, age × sex, age² × sex interaction effects were observed for any sensory or associative metrics (p _cor  > 0.05, Table 3).

3.4. Core Network Segregation

Both VIS and SMN SI decreased in a linear manner with increasing age (b = −1.57 × 10⁻³, p _cor  < 0.001; b = −1.77 × 10⁻³, p _cor  < 0.001, respectively), Figure 1d,e. VIS and SMN between‐network connectivity increased in a linear manner with increasing age (b = 7.18 × 10⁻⁴, p _cor  < 0.001; b = 5.34 × 10⁻⁴, p _cor  = 0.006, respectively). No other statistically significant age, age², sex, age × sex, age² × sex interaction effects were observed for any VIS and SMN metrics (p > 0.05, Table 4).

No statistically significant linear or quadratic age effects were observed for either FPN or DAN SI (Table 4), Figure 1f,g. Both VAN and DMN SI decreased in a negative quadratic manner with increasing age (b = −2.05 × 10⁻⁵, p _cor  < 0.001; b = −1.34 × 10⁻⁵, p _cor  < 0.001, respectively), Figure 1h,i. Specifically, VAN SI decreases until the 5th decade of life (ages 40–50), while DMN SI decreases until the 6th decade (ages 50–60), after which the decline for both networks accelerates into the 8th decade (ages 70–80). VAN within‐network connectivity decreased in a quadratic manner with increasing age (b = −5.05 × 10⁻⁵, p _cor  = 0.019). On the other hand, DAN within‐network connectivity increased in a linear manner with increasing age (b = 8.04 × 10⁻⁴, p _cor  = 0.019). DAN between‐network connectivity also increased in a linear manner with increasing age (b = 6.14 × 10⁻⁴, p _cor  < 0.001). A significant age × sex interaction was also observed for FPN SI, with only females displaying age‐associated linear decreases in FPN SI (b = −1.24 × 10⁻³, p _cor  = 0.017).

DMN, VAN, and FPN SI was significantly higher in females than males, independent of age (b = 3.36 × 10⁻², p _cor  < 0.001; b = 3.85 × 10⁻², p _cor  < 0.001; b = 6.00 × 10⁻², p _cor  = 0.021, respectively). No statistically significant sex effects were observed for DAN SI (Table 4). DAN within‐network and between‐network connectivity, however, were both significantly lower in females than males, independent of age (b = −3.78 × 10⁻², p _cor  < 0.001; b = −1.73 × 10⁻², p _cor  = 0.008, respectively). No other statistically significant age, age², sex, age × sex, age² × sex interaction effects were observed for any VAN, DMN, FPN, and DAN metrics (p _cor  > 0.05, Table 4).

A unified comparison of SI aging trajectories (independent of sex effects) across all examined networks is illustrated in Figure 2.

FIGURE 2.

Summary of modeled fit aging trajectories for whole‐brain, sensory, associative, visual (VIS), sensorimotor (SMN), frontoparietal (FPN), default mode (DMN), ventral attention (VAN), and dorsal attention (DAN) network segregation index (SI). Fitted models revealed quadratic decreases in whole‐brain, associative, VAN, and DMN SI, and linear decreases in sensory, VIS, and SMN SI. No significant age‐related effects were observed for DAN or FPN SI. Fitted models only account for age effects.

3.5. Pairwise Resting‐State Functional Connectivity Analysis

Pairwise between‐network connectivity increased in a linear manner with increasing age between several sensory‐sensory, sensory‐associative, and associative‐associative networks pairs (Table 5). This included VIS‐SMN (b = 8.07 × 10⁻⁴, p _cor  = 0.004), DAN‐DMN (b = 4.34 × 10⁻⁴, p _cor  = 0.024), DAN‐FPN (b = 7.40 × 10⁻⁴, p _cor  < 0.001), VIS‐DAN (b = 8.13 × 10⁻⁴, p _cor  = 0.003), VIS‐FPN (b = 7.30 × 10⁻⁴, p _cor  < 0.001), SMN‐DMN (b = 5.08 × 10⁻⁴, p _cor  = 0.032), and SMN‐FPN (b = 7.49 × 10⁻⁴, p _cor  < 0.001) network pairs. Positive quadratic increases in connectivity between VIS‐DMN were also observed with increasing age (b = 6.34 × 10⁻⁶, p _cor  < 0.001). Specifically, connectivity between VIS and DMN gradually increase until the 6th decade of life (ages 50–60), after which the decline accelerates into the 8th decade (ages 70–80).

Pairwise between‐network connectivity was lower in females relative to males, independent of age in several associative‐associative network pairs (Table 5). This includes DAN‐FPN (b = −1.86 × 10⁻², p _cor  = 0.039), DAN‐VAN (b = −2.44 × 10⁻², p _cor  = 0.005), and VAN‐FPN (b = −1.86 × 10⁻², p _cor  = 0.004) network pairs. No other statistically significant age, age², sex, age × sex, age² × sex interaction effects were observed for any other pairwise connectivity metrics (p _cor  > 0.05, Table 5).

3.6. Supplementary Analysis

All statistically significant linear and quadratic age, sex, and respective age × sex effects that survived Holm‐Bonferroni multiple comparisons correction persisted even after accounting for intra‐individual differences in health status (high blood pressure, high cholesterol, and diagnosis of diabetes), educational attainment, handedness, mean functional connectivity, mean relative framewise displacement, and MoCA score. The outputs of these models are provided in Supporting Information (Tables S2–S9).

A statistically significant positive correlation between whole‐brain SI and MoCA scores was observed across participants (r = 0.165, p = 0.002), Figure 3a. A statistically significant positive correlation between intermediate network SI and MoCA scores was also observed across participants (r = 0.172, p _cor  = 0.001), Figure 3b. No other significant relationships between intermediate sensory network or core network SI indices and MoCA scores were observed (p _cor  > 0.05).

FIGURE 3.

(a) Whole brain segregation index (SI) with MoCA score (r = 0.165, p = 0.002). (b) Associative network SI with MoCA score (r = 0.172, p _cor  = 0.001).

4. Discussion

This study examined age‐ and sex‐related differences in functional network segregation (defined by SI) across a large sample of cognitively intact individuals that spanned the adult lifespan. Specifically, we examined several brain networks across different spatial scales and conducted secondary within‐, between‐, and pairwise connectivity analyses to investigate the origin of any observed age and sex effects on network SI. Replicating previous work, we found brain networks to become less segregated as individuals age. Specifically, negative quadratic decreases in whole‐brain network, associative network, VAN, and DMN SI were observed; while sensory network, VIS, and SMN SI displayed linear age‐related decreases. This process of age‐related functional network reorganization was found to be preferential as functional connectivity was observed to increase either between anatomically adjacent associative networks or between anterior associative and posterior sensory networks.

Inherent sex differences in network segregation were also observed. Specifically, whole‐brain, associative, DMN, VAN, and FPN SI was greater in females compared to males, regardless of age. These differences are likely reflective of the reduced functional connectivity that was observed between associative networks in females compared to males. A notable linear age‐related decrease in FPN SI was also only observed for females, highlighting an age × sex interaction.

4.1. Segregation Index as a Biomarker of Age‐Related Functional Network Reorganization

Functional brain networks were observed to become less segregated with increasing age across adulthood. Specifically, SI for sensory networks (intermediate sensory network, SMN, and VIS) exhibited a gradual linear decline throughout adulthood. In contrast, SI for the whole‐brain network and most associative networks (intermediate associative network, VAN, and DMN) followed a negative quadratic trajectory: decreasing gradually until midlife (around 50–60 years of age), after which the decline accelerated into the 8th decade of life (70–80 years). These findings align with existing cross‐sectional and longitudinal studies, which have reported diverging linear and quadratic SI trajectories for sensory and associative networks, respectively (Chan et al. 2014; Chong et al. 2019; Grady et al. 2016; Hou et al. 2019; Malagurski et al. 2020; Manza et al. 2020; Pedersen et al. 2021; Varangis et al. 2019).

Age‐related decreases in cortical atrophy (Blinkouskaya et al. 2021; Pedersen et al. 2021), glucose metabolism (Deery et al. 2022), cerebrovascular reactivity (Kong et al. 2020), and cerebral blood flow (Beason‐Held et al. 2008) also disproportionately affect brain regions that underly transmodal associative networks compared to unimodal sensory networks. Moreover, anterior and long‐range white matter tracts, which are crucial for associative network communication, are more susceptible to age‐related deterioration than posterior and short‐range tracts (Gunning‐Dixon et al. 2009; Salami et al. 2012). These age‐related structural, functional, and cerebrovascular changes likely contribute to the observed differences in SI trajectories between associative and sensory networks, highlighting the need for future multimodal aging studies.

Similar to existing studies, we found that decreases in network segregation were linked to both decreasing within‐network connectivity and increasing between‐network connectivity with age (Chan et al. 2014; Pedersen et al. 2021; Varangis et al. 2019). Our findings, however, were not as ubiquitous as those previously reported and, in most cases, did not survive multiple comparisons correction. A multitude of factors could have contributed to the lack of observed or strong age effects on within‐network and between‐network connectivity in our study. This includes shorter rs‐fMRI acquisition time, more complete statistical models that account for sex and quadratic age effects, as well as potential variations in rs‐fMRI processing pipelines (e.g., lack of global signal regression). Previous studies have reported that absolute values of within‐ and between‐network connectivity are sensitive to rs‐fMRI acquisition parameters, such as acquisition duration and slice timing, as well as motion correction and network definition (Parkes et al. 2018; Power et al. 2012; Rai et al. 2024). On the other hand, as SI reflects proportional differences between within‐ and between‐network connectivity rather than absolute values it is likely less affected by such variabilities (Chan et al. 2014). For example, both within‐ and between‐network connectivity values tend to be elevated in subjects with greater head motion due to the introduction of spurious global functional connections (Power et al. 2012; Van Dijk et al. 2012). As SI normalizes these proportional differences relative to within‐network connectivity it may reduce the influence of such confounds, making it a potentially more robust and reliable biomarker.

Reduced segregation across several associative and sensory functional networks suggests that the functional architecture of the brain undergoes dedifferentiation across the adult lifespan. Functional dedifferentiation, which is characterized by a loss of neural distinctiveness or reduced selectivity in brain activation patterns, is typically observed in task‐based fMRI studies of aging (Fornito et al. 2015; Grady 2012). For instance, Park and colleagues observed more diffuse (i.e., less specialized) recruitment of visual cortices in older adults relative to younger adults when performing a visual recognition task (Park et al. 2004). Similarly, Gonneaud and colleagues observed that younger adults recruit task specific and fewer cortical brain regions when performing prospective memory tasks, whereas older adults display more diffuse and less selective cortical recruitment. Older adults were also reported to take longer or not disengaging task irrelevant cortical regions (i.e., increased between‐network connectivity) compared to younger adults (Gonneaud et al. 2017). This pattern of neural dedifferentiation appears to be mirrored in resting brain states as most associative and sensory functional networks were observed to become less segregated with age. That is, the increased coherence of blood oxygenation level dependent signals, as measured by rs‐fMRI, between different functional networks may be analogous to the dedifferentiation patterns that are observed during task‐states.

Other neurocognitive aging theories, including neural compensation and inefficiency, which can only be studied under task conditions could also potentially explain the observed results. The Scaffolding Theory of Aging and Cognition suggests that decreasing network segregation, or dedifferentiation, is likely reflective of a compensatory brain mechanism that is employed to offset age‐related physiological brain changes and preserve cognitive function. That is, as neural resources gradually deteriorate across adulthood, older adults may recruit additional or alternative functional connections within and across functional networks to meet cognitive demand (Goh and Park 2009). Network reorganization may also be reflective of underlying neural inefficiencies, where the ability to suppress non‐task specific functional networks weakens with age, resulting in non‐beneficial cortical recruitment (McDonough et al. 2022).

Taken together, these neurocognitive aging theories—dedifferentiation, compensation, and inefficiency—suggest that functional brain networks become less segregated as individuals age. Although these patterns are well‐documented in task‐based studies, the present findings suggest that similar mechanisms may underlie the reorganization of functional networks in resting states (Baltes and Lindenberger 1997; Cabeza et al. 2018; McDonough et al. 2022; Reuter‐Lorenz and Park 2014; Stern 2021). The rate of network reorganization, however, appears to vary across core sensory and associative functional networks (Figure 2), suggesting that the aging process may not affect the brain uniformly (Chan et al. 2014; Goh 2011; Malagurski et al. 2020; Pedersen et al. 2021).

4.2. Functional Network Reorganization Is Preferential

The pairwise between‐network connectivity analysis revealed two notable patterns of functional network reorganization with age (Figure 4). First, functional connectivity increased with age between spatially adjacent associative networks (DAN‐DMN and DAN‐FPN). These findings are consistent with both resting‐state and task‐based fMRI studies (Chan et al. 2014; Grady et al. 2016; Malagurski et al. 2020; Ng et al. 2016; Spreng and Schacter 2012; Varangis et al. 2019). For example, Malagurski and colleagues observed significant age‐associated increases in connectivity between adjacent network pairs (DAN‐DMN and VAN‐LIM, the latter of which was not examined in this study) in a sample of older adults (64–87 years) over a 4‐year period (Malagurski et al. 2020). Similarly, Spreng and colleagues reported increased connectivity between DAN‐DMN in older adults relative to younger adults during both an autobiographical planning task and resting state (Spreng et al. 2016). Our findings may reflect neurocognitive aging mechanisms like dedifferentiation and neural inefficiency (Baltes and Lindenberger 1997; Cabeza et al. 2018; McDonough et al. 2022; Reuter‐Lorenz and Park 2014; Stern 2021). That is, increased connectivity between adjacent associative networks may result from reduced neural selectivity and an inability to efficiently suppress irrelevant network activity. This pattern could also be indicative of dedifferentiation, as associative networks lose their functional distinctiveness with age (Chan et al. 2014; Goh and Park 2009; Kong et al. 2020; McDonough et al. 2022; Reuter‐Lorenz and Park 2014).

FIGURE 4.

Summary of pairwise between‐network connectivity with age. Examined networks include default mode (DMN), frontoparietal (FPN), ventral attention (VAN), dorsal attention (DAN), visual (VIS), and sensorimotor (SMN) networks. Functional connectivity increased with age in a linear manner between sensory networks (VIS‐SMN), anatomically adjacent associative networks (DMN‐DAN, FPN‐DAN), and anterior associative and posterior sensory networks (VIS‐DAN, VIS‐FPN, SMN‐DMN, and SMN‐FPN). However, connectivity between VIS and DMN increased in a quadratic manner with age. All effects were independent of sex.

Second, functional connectivity was observed to increase between sensory and associative networks (VIS‐DAN, VIS‐DMN, VIS‐FPN, SMN‐DAN, and SMN‐DMN), as well as between sensory networks (VIS‐SMN). These findings are consistent with existing task‐based and rs‐fMRI studies (Chan et al. 2014; Hou et al. 2019; King et al. 2018; Malagurski et al. 2020; Roski et al. 2013, 2014; Siman‐Tov et al. 2017; Stumme et al. 2020). For example, Roski et al. (2013) reported reduced activation in visual‐attention regions and increased activation in motor regions among older adults (relative to younger adults) during a visual‐attention task. Conversely, during an explicit motor task, older adults exhibited reduced activation in motor regions and increased activation in visual‐attention regions (Roski et al. 2013). This pattern of more coherent VIS and SMN activation was also observed in a follow‐up rs‐fMRI study, where older adults displayed greater connectivity between cortical regions corresponding to these two networks compared to younger adults (Roski et al. 2014). Similarly, King et al. (2018) reported age‐related increases in functional connectivity between sensory and associative network components (VIS‐DMN, VIS‐FPN, SMN‐DMN, and SMN‐DAN), though not all findings survived multiple comparisons correction. Our findings may reflect neurocognitive aging mechanisms like dedifferentiation and neural compensation (Baltes and Lindenberger 1997; Cabeza et al. 2018; Chan et al. 2017; McDonough et al. 2022; Reuter‐Lorenz and Park 2014; Stern 2021). That is, increased connectivity between sensory and associative networks may help mitigate age‐related cognitive decline by enabling broader network recruitment (Goh and Park 2009; Kong et al. 2020; McDonough et al. 2022). This is further supported by the observation that sensory network brain regions experience significantly fewer and less pronounced age‐related structural and metabolic changes compared to associative network regions (Beason‐Held et al. 2008; Blinkouskaya et al. 2021; Deery et al. 2022; Erramuzpe et al. 2020; Kong et al. 2020; Pedersen et al. 2021). Together, these findings suggest that increased connectivity between both spatially adjacent associative networks and sensory‐associative networks may be a hallmark of the aging process. These patterns may reflect the combined effects of reduced network specialization and neural inefficiency, as well as adaptive reorganization of the brain (dedifferentiation) to maintain cognitive function in the face of physiological decline (Baltes and Lindenberger 1997; Cabeza et al. 2018; Chan et al. 2017; McDonough et al. 2022; Reuter‐Lorenz and Park 2014; Stern 2021).

4.3. Dorsal Attention Network Segregation Is Less Prone to Aging Effects

We did not observe significant age‐associated decreases in DAN SI, because unlike other networks, both DAN within‐ and between‐network connectivity increased across adulthood. These findings align with existing studies that also report (1) increases in DAN within‐ and between‐network connectivity with age (Chan et al. 2014; Grady et al. 2016; Song et al. 2014), and (2) no statistically significant age effects on DAN SI over adulthood (when accounting for sex and/or age × sex) (Ballard et al. 2023; Grady et al. 2016; Malagurski et al. 2020). In studies that found DAN SI decreased with age, the magnitude of the reported age‐effect(s) was substantially smaller relative to the age effects of other associative networks. Moreover, these studies also reported that DAN within‐network functional connectivity did not significantly decrease with age (Chan et al. 2014; King et al. 2018). The DAN supports top‐down voluntary allocation of attention, which becomes increasingly difficult with age (ElShafei et al. 2022; Kok 2000; Vossel et al. 2014). As our study only includes cognitively intact adults, the observed increased DAN within‐ and between‐network connectivity may resemble complementary neural compensatory and dedifferentiation aging processes (McDonough et al. 2022; Stern 2021). Specifically, greater incorporation of the DAN (i.e., overactivation) alongside broader network recruitment may help minimize age‐associated declines in network specific cognitive performance (Geerligs et al. 2015; Grady et al. 2016; Song et al. 2014).

4.4. Associative Networks Are More Segregated in Females

Whole‐brain and several associative functional brain networks (intermediate associative network, FPN, DMN, and VAN) were observed to be more segregated in females compared to males, irrespective of age. Pairwise between‐network connectivity analysis further releveled that, relative to males, females had reduced connectivity between associative networks (DAN‐FPN, DAN‐VAN, and VAN‐FPN), independent of age (Figure 5). Although sex differences in network organization are less frequently reported than age effects, limited evidence of greater associative network segregation in females compared to males does exist (Allen et al. 2011; Ballard et al. 2023; Filippi et al. 2013; Foo et al. 2021; Goldstone et al. 2016; Hicks et al. 2023). For example, both Allen et al. (2011) and Filippi et al. (2013) observed greater within‐network connectivity and reduced between‐network connectivity among associative networks in female adults compared to males (Allen et al. 2011; Filippi et al. 2013). Similarly, Hicks et al. (2023) reported that salience network (or VAN) SI was higher in females than males across adulthood (Hicks et al. 2023). Task‐based fMRI studies have also reported comparable patterns of associative network organization. For instance, Bell et al. (2006) found that females exhibited less diffuse, or more specialized, cortical activation in frontal, parietal, and occipital brain regions across several task conditions compared to males (Bell et al. 2006). These findings suggest that greater segregation in female networks likely reflects enhanced efficiency of associative brain networks, which may contribute to more specialized functional organization in adult females relative to males.

FIGURE 5.

Pairwise between‐network connectivity with sex. Examined networks include default mode (DMN), frontoparietal (FPN), ventral attention (VAN), dorsal attention (DAN), visual (VIS), and sensorimotor (SMN) networks. Females exhibited significantly lower functional connectivity between associative networks (DAN‐VAN, VAN‐FPN) compared to males, independent of age.

In addition to differences in network organization, females are reported to have greater gray matter density across anterior brain regions (Gennatas et al. 2017; Luders et al. 2009), better cognitive performance (Nooyens et al. 2022), and benefit from the neuroprotective effects of estrogen and progesterone (Brann et al. 2007). These sex differences in brain structure, cognition, and hormonal factors likely contribute to the observed differences in associative network segregation, particularly the enhanced efficiency of associative brain networks in females. In support of this rationale, we also found a statistically significant positive correlation between associative network SI and MoCA scores, but not sensory network SI, independent of sex. MoCA scores were also found to be significantly higher in females than males. However, when accounting for MoCA scores and sex in our linear model statistical analysis, the relationship did not exhibit a significant effect, suggesting that observed sex differences in associative network segregation cannot be fully explained by cognitive status alone. To better understand these complex relationships, future studies should employ multimodal approaches that integrate structural, functional, and cognitive measures.

4.5. Only Females Exhibit Age‐Related Decrease in Frontoparietal Network Segregation

A notable finding from our analysis was that FPN SI decreased with age in females, but not males. This contrasts with findings from Ballard et al. (2023), who reported that FPN SI decreases with age in both sexes. However, Ballard also noted that other associative networks, such as the DMN and DAN, exhibited age‐related decreases in SI exclusively in females when analyzed separately (Ballard et al. 2023). Similarly, Hicks et al. (2023) observed that age‐related decreases in salience network (or VAN) SI were more accelerated in females than males, suggesting a possible age × sex interaction. Hicks also reported that higher endogenous levels of progesterone in females were significantly associated with greater SI in FPN subnetworks, including the cingulo‐opercular task control and fronto‐parietal task control networks (Hicks et al. 2023). However, these associations did not survive multiple comparisons correction. These findings suggest that the FPN, and possibly other associative networks, may be more vulnerable to age‐related decreases in SI in females due to age‐related hormonal changes compared to males.

Hormonal changes with age, particularly those associated with menopause, may increase susceptibility to structural and functional deterioration in females. For instance, females are reported to experience greater cortical thinning and gray matter loss in the frontal, parietal, and occipital cortices than males with advancing age (Guo et al. 2016), despite having greater cortical thickness and gray matter density in these regions earlier in life (Sangha et al. 2021). These structural changes likely contribute to sex‐specific neurocognitive aging trajectories, including the observed age‐related declines in FPN SI in females. For instance, decreasing SI in associative networks like the FPN in females may reflect mechanisms such as dedifferentiation and neural inefficiency. Dedifferentiation, characterized by reduced functional specialization, and inefficiency, marked by diminished network suppression, could likely be exacerbated by structural changes linked to age‐related hormonal declines (Goh and Park 2009; McDonough et al. 2022; Reuter‐Lorenz and Park 2014; Stern 2021). These sex‐specific vulnerabilities highlight the need for further research into how hormonal, structural, and functional factors interact to shape neurocognitive aging trajectories across the adult lifespan.

It is important to note that our study excluded the frontotemporal limbic network and subcortical regions from the analysis due to well‐documented fMRI signal drop‐out in these areas (Chavhan et al. 2009; Liu et al. 2018). Additionally, we excluded anticorrelations from our analysis, as their physiological underpinnings are not well understood; unlike Ballard, who included these brain regions and anticorrelations in their SI calculations (Ballard et al. 2023). These methodological differences may partially explain the discrepancies between our findings and those of other studies. Moreover, the potential for age × sex interactions on network SI has often been overlooked in previous research, making it challenging to directly compare our results with the existing literature (Chan et al. 2014; Chong et al. 2019; Hou et al. 2019; King et al. 2018; Kong et al. 2020; Pedersen et al. 2021; Varangis et al. 2019). Thus, future research should consider the potential for age × sex interactions on network SI in their statistical analysis.

5. Study Limitations

Though this study advances our understanding of age‐ and sex‐effects on resting‐state network organization, it is not without limitations. We employed a cross‐sectional design that is inherently limited in its ability to account for intra‐individual differences and capture rates of change over time. Future studies using longitudinal data should be used to elucidate actual aging trajectories.

The parcellation scheme used in this study was derived from the functional data of younger, cognitively intact adults, which means that possible age‐related differences in network topology could not be accounted for (Doucet et al. 2021; Schaefer et al. 2018). Future research should consider implementing age‐group‐specific parcellation templates and evaluating the observed results across different rs‐fMRI processing techniques.

Several alternative parcellation schemes (Power et al. 2011) are also commonly used to investigate network segregation and age effects, which may partially explain some of the differences between our findings and those of other studies. Notably, although this study did not explicitly investigate this aspect, both Chan and Malagruski, and their respective colleagues, reported that aging trends were generally preserved across different parcellation schemes (Chan et al. 2014; Malagurski et al. 2020). Future research should continue to investigate the impact of methodological choices, such as parcellation schemes, on study outcomes and work towards harmonizing processing techniques.

The duration of the rs‐fMRI data acquisition (~5 min) used in this study is also suboptimal as longer durations (9–12 min) are suggested to increase intra‐ and inter‐session stability of functional connectivity measurements (Birn et al. 2013). As SI normalizes and quantifies the proportional differences in within‐ and between‐network connectivity relative to within‐network connectivity, however, it may be less prone to such temporal differences (Chan et al. 2014; Han et al. 2024). This may also explain why age‐related effects on network SI have been largely consistent across the literature relative to the reported age effects on within‐ and between‐network connectivity (Ballard et al. 2023; Chan et al. 2014; Chong et al. 2019; Grady et al. 2016; Hou et al. 2019; Malagurski et al. 2020; Manza et al. 2020; Pedersen et al. 2021; Varangis et al. 2019).

One limitation of backward iterative regression is that the final models vary across spatial scales, which limits comparability across and between scales. Although this lack of uniformity introduces some challenges in interpreting effects, it ensures that each model is tailored to best explain the variation observed in the data (Weisberg 1985). This approach was guided by objective statistical criteria, including AIC and ANOVA, to ensure the robustness of our findings. We also corrected observed effects using Holm‐Bonferroni Multiple comparisons correction, which is a relatively conservative correction technique that minimizes Type I error (Holm 1979). However, we acknowledge that certain effects hypothesized in the literature may not have been detected in our study, possibly due to sampling bias or limited statistical power. Future work with larger sample sizes or longitudinal data would strengthen the observed findings.

While the SI provides a straightforward and intuitive measure of network organization, it has several limitations. Specifically, the SI requires predefining within‐ and between‐network connections and evaluates between‐network connectivity relative to within‐network connectivity, which may oversimplify complex functional interactions. Future studies should aim to compare age‐ and sex‐related differences in network segregation using multiple metrics to gain a more comprehensive understanding of functional network organization across adult aging.

6. Conclusion

Resting‐state functional networks reorganize across the adult lifespan in cognitively intact individuals, with network segregation generally decreasing with age. This decline may reflect underlying neurocognitive aging mechanisms, such as neural dedifferentiation, inefficiency, and compensation. The aging trajectories and rates of decreasing network segregation were also found to vary across associative and sensory networks, suggesting that the aging process may not uniformly affect the cerebral cortex. Instead, two general patterns appear to underlie the aging process: increased functional connectivity (1) between spatially adjacent associative networks and (2) between sensory and associative networks. Interestingly, certain associative networks, like the DAN, were found to be less vulnerable to age‐related changes in segregation.

This study also provides preliminary evidence of inherent differences in network organization between females and males, independent of age. Specifically, females exhibited greater segregation across several associative networks, suggesting that these networks may be more efficient and functionally specialized in females compared to males. The notable age‐related decline in frontoparietal network segregation observed only in females highlights the possibility of sex differences in neurocognitive aging mechanisms. These findings underscore the need to better understand how sex influences network reorganization with aging.

Given these findings, future studies should take a more focused approach to examining sex differences across the lifespan, incorporating multimodal methodologies to better integrate structural, functional, hormonal, and cognitive data. Such approaches will be essential for disentangling the complex relationships between network organization, aging, and cognitive outcomes.

Author Contributions

A.S.S. performed conceptualization, data curation, formal analyses, visualization, and wrote the original draft of the manuscript. K.T.N.D. assisted in developing methodology, data processing, and reviewed and edited the manuscript. T.H.S. assisted in data processing and reviewed and edited the manuscript. R.J.S. was involved in conceptualization and reviewed and edited the manuscript. M.L.L. assisted with formal analyses and reviewed and edited the manuscript. M.S. performed data processing and reviewed and edited the manuscript. C.R.M. performed data curation, developed methodology, and reviewed and edited the manuscript. A.B.P. assisted with conceptualization and reviewed and edited the manuscript. B.G.G. was involved in conceptualization, provided methodological guidance and reviewed and edited the manuscript. R.F. supervised and acquired funding for the study. He was involved in study conceptualization and reviewed and revised the manuscript.

Ethics Statement

This study was carried out in accordance with the recommendations of the University of Calgary Conjoint Health Research Ethics Board (CHREB) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. This protocol was approved by the CHREB.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1.

Data Availability Statement

Processed and anonymized resting‐state functional connectivity matrices (200 × 200, includes limbic regions) and simple demographic information (age and sex) are available for academic purposes upon completion and approval of a short data sharing application submitted to the principal investigator (R.F.) via email (calgarynormstudy@ucalgary.ca). Details of the iterative linear models (i.e., summary tables) are also available via request to the principal investigator.