Temporal and Spatial Scales of Human Resting-State Cortical Activity across the Lifespan
- John Bero1,
- Colin Humphries1,
- Yang Li1,
- Aviral Kumar1,
- Heungyeol Lee1,
- Maxwell Shinn2,
- John D. Murray3,
- Timothy J. Vickery4, and
- Daeyeol Lee1,5
- 1Neurogazer USA, Inc., Baltimore, Maryland 21231
- 2UCL Queen Square Institute of Neurology, University College London, London WC1E 6BT, United Kingdom
- 3Department of Psychological and Brain Sciences, Dartmouth College, Hanover, New Hampshire 03755
- 4Department of Psychological and Brain Sciences, University of Delaware, Newark, Delaware 19716
- 5The Zanvyl Krieger Mind/Brain Institute, Department of Neuroscience, Department of Psychological and Brain Sciences, Kavli Neuroscience Discovery Institute, Johns Hopkins University, Baltimore, Maryland 21218
-
Author contributions: J.B., H.L., T.J.V., and D.L. designed research; J.B., C.H., Y.L., A.K., M.S., J.D.M., T.J.V., and D.L. performed research; J.B., C.H., and T.J.V. analyzed data; J.B. and D.L. wrote the paper.
Abstract
Sensorimotor and cognitive abilities undergo substantial changes throughout the human lifespan, but the corresponding changes in the functional properties of cortical networks remain poorly understood. This can be studied using temporal and spatial scales of functional magnetic resonance imaging (fMRI) signals, which provide a robust description of the topological structure and temporal dynamics of neural activity. For example, timescales of resting-state fMRI signals parsimoniously predict a significant amount of the individual variability in functional connectivity networks identified in adult human brains. In the present study, we quantified and compared temporal and spatial scales in resting-state fMRI data collected from 2,352 subjects of either sex between the ages of 5 and 100 in Developmental, Young Adult, and Aging datasets from the Human Connectome Project. For most cortical regions, we found that both temporal and spatial scales decreased with age throughout the lifespan, with the visual cortex and the limbic network consistently showing the largest and smallest scales, respectively. For some prefrontal regions, however, these two scales displayed non-monotonic trajectories and peaked around the same time during adolescence and decreased throughout the rest of the lifespan. We also found that cortical myelination increased monotonically throughout the lifespan, and its rate of change was significantly correlated with the changes in both temporal and spatial scales across different cortical regions in adulthood. These findings suggest that temporal and spatial scales in fMRI signals, as well as cortical myelination, are closely coordinated during both development and aging.
Significance Statement
Temporal and spatial scales of resting-state cortical activity in humans measured by fMRI largely decreased throughout the lifespan, except that for some regions in the prefrontal cortex they peaked similarly during adolescence. In addition, whereas cortical myelination consistently increased throughout the lifespan, its variation across different cortical networks and the rate of age-related changes were correlated with the dynamics of temporal and spatial scales of rs-fMRI activity, suggesting that the spatiotemporal scales of cortical activity and cortical myelination might be coregulated during development and aging.
Introduction
The human brain goes through massive changes throughout the lifespan that manifest as development and decline in cognitive functions. Sensorimotor systems tend to mature first in early childhood followed by language and other high-order cognitive skills (Gogtay et al., 2004; Sowell et al., 2004; Casey et al., 2005). Fluid abilities, such as working memory, show a steady decline from early adulthood, whereas crystalized abilities, such as semantic memory, improve and then plateau around age 60 (Levy, 1994; Salthouse, 2010; Harada et al., 2013; Murman, 2015). These different phases of cognitive development and aging are a result of changes in the underlying anatomical and cytoarchitectural properties of the brain. Brain size, cortical thickness, and myelination increase during the early stages of life, followed by the loss of cortical tissues (Stiles and Jernigan, 2010; Blinkouskaya, et al., 2021; Bethlehem, et al., 2022). However, these anatomical changes are not uniform, exhibiting a regional heterogeneity in different development phases, which could be the main driver for the differences in cognitive abilities throughout the lifespan (Casey et al., 2005; Stiles and Jernigan, 2010). Similarly, functional connectivity (FC) computed using resting-state fMRI (rs-fMRI) also exhibits age-related changes. For example, FC during childhood is dominated by short-range local interactions, which are gradually replaced by long-distance interactions in adulthood (Fair et al., 2009). Aging studies have demonstrated the opposite effect, with networks exhibiting decreased connectivity as the brain matures (Geerlings et al., 2015). This trajectory implies a peak in development sometime between childhood and young adulthood (Betzel et al., 2014; Coupé et al., 2017; Edde et al., 2021; Bethlehem, et al., 2022).
Despite these previous findings, however, many aspects of cortical functions during development and aging remain poorly understood. For example, elementary statistical properties of the rs-fMRI signal, such as temporal and spatial autocorrelations, vary within and across individuals (Friston et al., 1995; Murphy et al., 2013; Burt et al., 2020; Shinn et al., 2023). Temporal and spatial scales in the fMRI signals might reflect important biological factors, including molecular (Huang et al., 2018; Burt et al., 2020), structural (Sethi et al., 2017; Fallon et al., 2020), activity state (Arbabshirani et al., 2019; Fascianelli et al., 2019), and network (Honey et al., 2012; Song et al., 2014) properties of the brain. Therefore, identifying the factors that influence the intrinsic temporal and spatial timescales of rs-fMRI signals is important to fully understanding the brain’s dynamics throughout lifespan. Nevertheless, due to the restricted age ranges used in previous studies on the topic, the complete age profile of spatial and temporal scales in the human brain has not been fully characterized (Gao et al., 2020; Ito et al., 2020; Raut et al., 2020; Han et al., 2022; Shinn et al., 2023; Truzzi and Cusack, 2023; Wu and Gollo, 2025).
In the present study, we estimated the timescale of the rs-fMRI signals using the correlation coefficient between the signals in two successive time steps. In addition, we estimated the spatial scale of the rs-fMRI signals and its regional variation by measuring how steeply the FC decays with distance over the cortical surface. We examined how these temporal and spatial scales of rs-fMRI signals as well as cortical myelination change from infancy into late adulthood by examining multiple datasets spanning the ages from 5 to 100 years from the Human Connectome Project. We found that the spatial and temporal scales of rs-fMRI have similar non-monotonic changes for some prefrontal regions, peaking during adolescence followed by a decrease for the rest of the lifespan. Furthermore, we found that age-related changes in temporal and spatial scales of rs-fMRI signals were correlated with the level of myelination across different cortical areas during childhood, but with the rate of changes in myelination during adulthood. These results suggest that the coordination between spatiotemporal scales of cortical activity and myelination might have important implications for brain development and aging.
Materials and Methods
Datasets and participants
The data analyzed in the present study were taken from three different datasets of the Human Connectome Project (HCP); developmental dataset (HCP-D) composed of youth and adolescents (age 5–22y), 1,200 subject dataset (HCP-YA) composed of young adults (age 22–37 years), and Aging dataset (HCP-A) composed of middle aged and elderly adults (age 37–100 years). In the following, we will refer to these three datasets as Child, Adult, and Old age groups, respectively. We analyzed the preprocessed resting-state fMRI time series in these datasets. Subjects were excluded based on their average framewise displacement (FD), computed as the sum of the absolute values of the differentiated realignment estimates (by backward differences) at every time point (Power et al., 2012). Based on the distribution of mean framewise displacement, a value corresponding to the distribution mean plus two times the standard deviation was used as a cutoff threshold. The final sample consisted of 627 Child subjects, 1,036 Adult subjects, and 689 Old subjects, for a total of 2,352 subjects (1,151 males, 1,201 females; Table 1). As described below, there were some variations in data collection methods, quality control procedures, and preprocessing steps across the datasets. In addition, the sampling density (number of subjects/age) was not uniform across ages and was maximal for the Adult dataset.
Demographic information for subjects
Data acquisition
Child and Old data were collected on a 3T SIEMENS Prisma scanner and included T1w anatomical (multiecho MPRAGE; resolution, 0.8 mm × 0.8 mm × 0.8 mm; TR/TI/TE, 2,500/1,000/1.8 ms; flip angle, 8°), T2w anatomical (variable-flip-angle TSE; resolution, 0.8 mm × 0.8 mm × 0.8 mm; TR/TE, 3,200/564 ms; turbo factor, 314), and rs-fMRI scans (GRE EPI; resolution, 2 mm × 2 mm × 2 mm; TR/TE, 800/37 ms; measurements, 488; MB factor, 8; flip angle, 52°; scan time, 6.30 min). Adult data were collected on a custom 3T SIEMENS Skyra scanner and included T1w anatomical (MPRAGE; resolution, 0.7 mm × 0.7 mm × 0.7 mm; TR/TI/TE, 2,400/1,000/2.4 ms; flip angle, 8°), T2w anatomical (variable-flip-angle TSE; resolution, 0.7 mm × 0.7 mm × 0.7 mm; TR/TE, 3,200/565 ms; GRAPPA, 2), and rs-fMRI scans (GRE EPI; resolution, 2 mm × 2 mm × 2 mm; TR/TE, 720/33 ms; measurements, 1,200; MB factor, 8; flip angle, 52°; scan time, 14.4 min).
Data analysis
Preprocessing
Structural and functional MRI data were processed using the HCP minimal preprocessing pipelines as described previously (Glasser et al., 2013; Bero et al., 2023). The T1w and T2w images were aligned with the anterior commissure–posterior commissure (AC–PC) line, and brain were extracted, corrected for intensity bias, and registered to the MNI template. The white matter and pial surfaces were traced, and the data were projected onto the cortical surface via FreeSurfer (Fischl, 2012). Myelin content was estimated by taking the log ratio of T1w/T2w signal intensities on a voxel-by-voxel basis (Glasser and Van Essen, 2011; Ganzetti et al., 2014) on the standard CIFTI grayordinate surface. The resting-state fMRI data underwent motion correction, susceptibility distortion correction, brain extraction, registration to MNI and anatomical space, time series normalization, component-based denoising using ICA-FIX, and high-pass filtering (0.0005 Hz). Processed volumetric data were projected onto the standard CIFTI grayordinate surface (59,412 voxels) and smoothed with a full-width half max of 2 mm. Other details of the processing were identical to those used in previous studies (Glasser et al., 2013; Glasser et al., 2016; Harms et al., 2018).
Postprocessing
To facilitate comparisons between datasets, some common processing steps were applied to all final outputs following their individual preprocessing pipelines. All functional surface maps were parcellated based on the Schaefer parcellation, which is a functional parcellation based on resting-state functional connectivity data with 400 total regions of interest (ROIs) separated into 17 resting-state networks (Fig. 1a; Schaefer et al., 2018). In addition, a stricter high-pass filter was applied to all time series at 0.008 Hz, and all ROI time series were standardized prior to analysis. Since the current study focuses on regional variation, global patterns and correlations were removed in all our data at the subject level for the main results presented in the present study. The global signal for each frame was computed as the average of signals across all ROI, and global signal regression was performed for each participant (Murphy and Fox, 2017; Li et al., 2019). Nevertheless, we also tested how the estimates of temporal and spatial scales of rs-fMRI signals are affected by comparing the results obtained with and without the global signal regression.
Brain maps for resting-state functional networks and sensorimotor-association axis. a, Functional networks identified in Schaefer et al. (2018) parcellation. b, Ranking of ROI in Schaefer parcellation along the sensorimotor-association (S-A) axis identified by Sydnor et al. (2021).
Analysis of temporal and spatial scales
To estimate the timescale of rs-fMRI signals, we computed the first-order (lag-1) autocorrelation of each parcellated time series, namely, Pearson’s correlation between the blood oxygen dependent-level (BOLD) signals in two successive time bins. We also computed the lag-1 autocorrelation in the global signal for each subject. For a first-order autoregressive process, this lag-1 autocorrelation (lag-1 TA, or ϕ1) is related to its time constant (τ) as τ = −TR/ln ϕ1 and can efficiently characterize the individual variability in the temporal autocorrelations in the rs-fMRI signals (Shinn et al., 2023). Lag-1 autocorrelation is also correlated with multiple temporal features of BOLD signals including the so-called long-memory dynamics (Shafiei et al., 2020; Shinn et al., 2023). To compare these long-memory or long-range processes to lag-1 autocorrelation, we also estimated the fractional integration constant d from an ARFIMA model by using a univariate wavelet-based Whittle estimator, which has been used in previous studies to capture long-memory dynamics (Achard et al., 2008; Sela and Hurvich, 2009; Achard and Gannaz, 2015).
To determine the spatial scale, a distance metric must first be established on the cortical surface. For this study, geodesic
distances were computed on the pial brain surfaces from the centroids of each Schaefer ROI, creating a 400 × 400 distance
matrix. Functional connectivity matrices were computed by taking Pearson’s correlation between each pair of parcellated time
series (also with dimensionality of 400 × 400). For a given ROI, all other brain regions were binned according to their geodesic
distance, using 1 mm bins, and the mean correlation with each bin was used to generate a spatial autocorrelation (SA) function
for that ROI. Then, the resulting SA was fit by an exponential function with two parameters, the rate at which the correlation
falls off with distance (SAλ) and the baseline (SA∞), using the method used in Shinn et al., (2023), as follows:<mml:math display="block"><mml:mrow><mml:mi mathvariant="normal">corr</mml:mi><mml:mo stretchy="false">(</mml:mo></mml:mrow><mml:msub><mml:mi>x</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>x</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mrow><mml:mo
stretchy="false">)</mml:mo></mml:mrow><mml:mo>=</mml:mo><mml:mi>S</mml:mi><mml:msub><mml:mi>A</mml:mi><mml:mrow><mml:mi mathvariant="normal">∞</mml:mi></mml:mrow></mml:msub><mml:mo>+</mml:mo><mml:mo
stretchy="false">(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>S</mml:mi><mml:msub><mml:mi>A</mml:mi><mml:mrow><mml:mi
mathvariant="normal">∞</mml:mi></mml:mrow></mml:msub></mml:mrow><mml:mo stretchy="false">)</mml:mo><mml:mrow><mml:mi mathvariant="normal">exp</mml:mi><mml:mo
stretchy="false">(</mml:mo></mml:mrow><mml:mo>−</mml:mo><mml:msub><mml:mi>d</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mo>,</mml:mo><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:mo>/</mml:mo><mml:mi>S</mml:mi><mml:msub><mml:mi>A</mml:mi><mml:mi>λ</mml:mi></mml:msub><mml:mrow><mml:mo
stretchy="false">)</mml:mo><mml:mo>,</mml:mo></mml:mrow><mml:mspace width="2em"></mml:mspace><mml:mo stretchy="false">(</mml:mo><mml:mn>1</mml:mn><mml:mo
stretchy="false">)</mml:mo></mml:math>corr(xi,xj)=SA∞+(1−SA∞)exp(−di,j/SAλ),(1)
where xi is the time series for region i and di,j is the distance between regions i and j. An ADAM optimizer (a variation of stochastic gradient descent) parallelized with 9 CPU cores was used to estimate the model
parameters. We used SAλ, akin to a length constant, as a measure of spatial scale in the rs-fMRI signals. Network averages were obtained by averaging
the parameters of all ROIs within a network. To test the reliability of the resulting brain map of the temporal and spatial
scales, we computed average brain maps in two randomly divided subgroups in each age group and computed the average correlations
between them over 100 repetitions.
Age dependence of spatial and temporal scales
To quantify the age-related change in spatial and temporal scales of rs-fMRI signals, we first used linear regression to generate brain maps of age coefficients for spatial and temporal scales in each age group. In addition, statistical significance of the age effect was tested using a linear mixed effects model with age and network coded as continuous and discrete variables, respectively. The statistical significance of overall age–network interaction was assessed by comparing a full model (AC = Age + Network + Age × Network) and a reduced model (AC = Age + Network) using a χ2 test (Chen et al., 2013). In the full model, individual age–network interactions were tested using the difference relative to the mean as contrast vectors. For visualization of age-related changes in the spatial and temporal scales over the entire lifespan, curves were fit separately for the network averages in each age group. This was done using a locally weighted scatterplot smoothing (LOWESS) regression model (with a fraction of 0.5), which is not influenced by the uneven sampling distributions as present in the data (Cleveland, 1981). Separate fits were performed for each age group to remove any artificial features that might arise due to differences in acquisition or preprocessing between the datasets. Significance of non-monotonic trends in each age group were tested using a two-tailed t test on the second-order age term from a regression model with a quadratic term for each network (p < 0.05) with the requirement that the maximum or minimum of the model must fall within the age range of the dataset in question. We used the same methods to analyze age-related changes in the global signal. Both spatial and temporal scales estimated in this study revealed some discontinuities at the age boundaries between different datasets, apparently due to the differences in the scanning protocol and preprocessing methods. All the statistical analyses reported herein were performed separately for each age group and therefore did not require any harmonization. For the purpose of visualization, however, we computed the offset in the timescale, spatial scale, and cortical myelination (see below) for the Adult age group at the two age boundaries and subtracted their mean from the corresponding values for individual networks.
Comparison with other brain maps
We constructed brain maps corresponding to the mean values and age coefficients for temporal and spatial scales of rs-fMRI signals for each ROI in the Schafer parcellation. To examine age-related changes in cortical microstructures, similar maps were also created for cortical myelination using the log ratio of T1w/T2w as a proxy (Glasser and Van Essen, 2011; Ganzetti et al., 2014) and were analyzed using the same mixed effects model and the LOWESS regression fit used to analyze temporal and spatial scales. We compared these brain maps as well as the ranking of each ROI along the sensorimotor-association (S-A) axis that accounted for many anatomical and functional features of the human cortex (Fig. 1b; Sydnor et al., 2021, https://pennlinc.github.io/S-A_ArchetypalAxis/). The statistical significance of correlations between different brain maps was determined using a t test.
Results
Age-related changes in timescales of fMRI signals
We first computed the lag-1 autocorrelation to examine how the timescales of rs-fMRI signals in different cortical areas changed across different age groups (Fig. 2a). The resulting brain maps of timescales were robust in all age groups, as indicated by the correlation coefficient between the maps generated for randomly divided two halves in each group (r = 0.82, 0.78, and 0.73, for Child, Adult, and Old, respectively). The timescales were relatively short in the insular cortex and cingulate gyrus and long in the occipital and parietal regions (Fig. 2a). These brain maps were highly similar across different age groups at the network level (r > 0.95, p < 0.001; Fig. 2b,c). We also found that the whole-brain mean timescales significantly decreased from Child to Adult to Old age groups (mean ± SD = 0.59 ± 0.20, 0.41 ± 0.19, and 0.39 ± 0.20, for Child, Adult, and Old, respectively; one-way ANOVA, p < 0.05; Fig. 2b,c), which presumably resulted from the differences in imaging protocols across datasets. These results were nearly identical when we used the long-memory parameter d estimated with the ARFIMA model instead of the lag-1 autocorrelation. Indeed, the long-memory parameter and lag-1 autocorrelation were highly correlated across networks for all age groups (r > 0.99, p < 0.001).
Timescales of rs-fMRI activity. a, Brain maps of mean timescale (lag-1 TA) maps for Child (left), Adult (center), and Old (right) age groups. b, c, Scatter plots of the timescales for Child versus Adult (b) and Adult versus Old (c) age groups. Small and large symbols show the values for individual ROI and network averages. Dotted lines show group means.
Overall, we found that the timescales tended to decrease for most FC networks in all age groups, as indicated by the significantly negative age coefficient in a linear mixed effects model (Table 2). We also computed the age coefficient of a simple linear regression model for each ROI and found that they are negative in most ROI, except that some ROIs around Brodmann area 10 in the Child age group showed significantly positive age coefficients (p < 0.05; Fig. 3a). In addition, we found that for all age groups, the full model with age × network interaction terms fit the data better than a reduced model without them, suggesting that the slopes of individual network trajectories were not uniform within each age group. We found that the trajectory of the timescale was more complex and heterogeneous in the Child age group than that in the Adult and Old age groups. Most FC networks showed significant age × network interaction in the Child age group (n = 11 out of 17), whereas this was the case for fewer networks in Adult (n = 4) and Old (n = 6) age groups (Table 2). The Limbic-A, Limbic-B, and Default-C networks showed significantly less negative (i.e., more positive) slopes than the average in both Adult and Old age groups.
Age-related changes in the timescale of rs-fMRI signals. a, Brain maps of the age coefficient for timescale in Child (left), Adult (middle), and Old (right) age groups. Gray regions indicate the ROI in which the age effect was not statistically significant (p > 0.05). b, Scatter plots of the average timescale (lag-1 TA) for the entire cortex across individual subjects. The red line represents the LOWESS fit and vertical dotted lines show the border between age groups. c, LOWESS fit for timescale in each FC network. Dotted/dashed lines show networks that are significantly more positive/negative than the average (see also Table 2). The values for the Adult age group were increased by 0.087 to compensate for the discontinuations at the age boundaries shown in b. d, Peak ages for timescales in networks with significant non-monotonic time course in the Child age group. The black vertical line and gray bar correspond to the mean ± SD.
Age-related changes in timescale, spatial scale, and cortical myelination across cortical networks estimated by linear mixed effects models for different age groups
To visualize the age-related changes in timescale within each age group, we fit the LOWESS curves to the timescales of individual participants for the entire cortex (Fig. 3b) and also separately for different FC networks (Fig. 3c). The average timescale for the entire cortex as well as the timescales of individual networks tended to decrease with age. However, these plots also revealed non-monotonic age-related changes in the timescales, which was further tested using a regression model that included a quadratic age term. The average timescale for the entire cortex showed a significant non-monotonic trend in the Child age group (p < 0.005), reaching its peak at 11.1 years of age (Fig. 3b). Significant non-monotonic age-related changes in timescales were also found for 13 of 17 FC networks, but only in the Child age group (Fig. 3c). The peak age varied from 6 to 16 across these 13 networks, with the mean of 11.9 years of age (Fig. 3d). This implies that the timescales of rs-fMRI signals still increase during early childhood and do not reach their full maturation until adolescence. Therefore, while age has a global effect on timescales of rs-fMRI signals, these age-related changes vary across functional networks.
We found that throughout the lifespan, the timescales of the global signal were longer than the average timescales of rs-fMRI signals estimated with the global signal regression (paired t test, t > 80.0, p < 10−100 for all age groups; Figs. 3b, 4a). In addition, the timescale of the global signal significantly decreased with age in the Adult and Old age groups (p < 0.01), although this change was significantly smaller than for the average timescale of FC networks estimated with the global signal regression (z test, p < 10−6). In contrast, for the Child age group, the timescale of the global signal showed a significant non-monotonicity, peaking at the age of 16.2. Thus, the timescales of the global signal and regional rs-fMRI signals showed qualitatively similar age-related changes. Nevertheless, we found that the timescales of rs-fMRI for individual FC networks were largely unaffected by the global signal regression. The timescales of rs-fMRI signals from different FC networks with and without the global signal regression were highly correlated in all age groups (r > 0.97, p < 0.001; Fig. 4b).
Age-related changes in the global signal and its effect on the temporal and spatial scales of rs-fMRI activity. a, Timescale of the global signal across individual subjects. Red line indicates LOWESS fit. b, c Scatter plots for temporal (b) and spatial (c) scales estimated with (ordinate) and without (abscissa) the global signal regression (GSR).
Age-related changes in spatial scales of fMRI signals
Next, we examined age-related changes in spatial scale of rs-fMRI signals. The brain map of the average spatial scale in each age group showed that the spatial scales tended to be larger in the somatosensory-motor, visual, and frontopolar regions (Fig. 5a). These maps were robust, although the correlation coefficient between the randomly divided two halves was somewhat lower (r = 0.63, 0.60, and 0.58, for Child, Adult, and Old groups, respectively) than that for the timescales. We also found strong and significant correlations between the maps of spatial scales across age groups (Corr > 0.95, p < 0.001; Fig. 4b,c). Although the anatomical distribution of spatial scales remains similar across different age groups, the whole-brain average spatial scale decreased significantly and consistently across three different age groups (mean ± SD = 15.5 ± 6.0, 13.1 ± 5.1, and 12.5 ± 5.0 mm for Child, Adult, and Old groups, respectively; one-way ANOVA, p < 0.05).
Spatial scales of rs-fMRI activity. a, Brain maps of mean spatial scale of rs-fMRI signals for Child (left), Adult (center), and Old (right) age groups. b, c, Scatter plots showing spatial scales for Child versus Adult (b) and Adult versus Old (c). Dotted lines show group means.
Using linear mixed effects models, we also found that spatial scales decreased with age in all age groups (Table 2, Fig. 5a). This analysis also revealed significant age × network interactions for some networks in each age group (n = 9, 3, and 7 for Child, Adult, and Old groups, respectively; Table 2). Similar to the pattern with the timescale, some FC networks show significant age × network interactions in both Adult and Old age groups. For example, Limbic-B network showed significantly less negative slopes than the average in both age groups, whereas Visual-Cent and Somatosensory-Motor-B networks showed significantly more negative slopes (Table 2).
The age coefficients for spatial scales in individual ROI showed that spatial scales tended to decrease with age for most brain areas throughout the lifespan. However, similar to the age coefficients for timescales, they were significantly positive for some areas of the prefrontal cortex in the Child age group (Fig. 6a). Furthermore, although LOWESS fits to the whole brain and network averages for subjects in each group similarly showed overall decrease in spatial scales with age (Fig. 6b), some networks displayed non-monotonic trajectories in the Child age group (Fig. 6c). Although the number of FC networks for which the quadratic regression model fit age-related changes in spatial scale better was smaller (n = 5) than that for the timescales (n = 13), the average peak ages were similar (mean = 11.9 and 11.3 for temporal and spatial scales, respectively; Figs. 3d, 6d). Similar to the timescales of rs-fMRI signals, we also found that the spatial scales estimated with and without the global signal regression were highly correlated (r > 0.9, p < 0.001; Fig. 4c).
Age-related changes in the spatial scale of rs-fMRI signals. a, Brain maps of the age coefficients for the spatial scale (SAλ) in Child (left), Adult (center), and Old (right) age groups. Gray regions indicate the ROI in which the age effect was not statistically significant (p > 0.05). b, Scatter plots of the average spatial scale for the entire cortex across individual subjects. c, LOWESS fit for the spatial scale in each FC network. Dotted/dashed lines show networks that are significantly more positive/negative than the average. The values for the Adult age group were increased by 0.86 to compensate for the discontinuations at the age boundaries shown in b. d, Peak ages for spatial scale in networks with significant non-monotonic time course in the Child age group. Same format as in Figure 3.
Comparison of temporal and spatial scales
The above results suggest that age-related changes in temporal and spatial scales of fMRI signals might show similar patterns. In particular, the age-related changes in temporal and spatial scales were non-monotonic for some prefrontal regions in the Child age group, which would not be fully captured by the age coefficient in a linear regression model. Therefore, for each network and age group, we tested whether these two scales were correlated across different subjects. We found that these two scales were strongly and significantly correlated for all networks in each age group (mean correlation coefficient ± SD = 0.64 ± 0.09, 0.74 ± 0.10, 0.74 ± 0.09, for Child, Adult, and Old age groups, respectively). The correlations between the two scales for the networks with non-monotonic trajectories in the Child age group ranged from 0.45 to 0.78, and they were not significantly different from the monotonic network correlations (t test, p > 0.05). We found that the correlation between the temporal and spatial scales at the network level was not statistically significant (Fig. 7, top). On the other hand, we found a significant correlation between the age coefficients for temporal and spatial scales in each age group (Fig. 7, bottom), suggesting that the age-related changes in temporal and spatial scales of rs-fMRI signals might be coordinated throughout the lifespan.
Relationship between the temporal and spatial scales of rs-fMRI activity. Scatter plots comparing the temporal and spatial scales (top) and the corresponding age coefficients (bottom). Small transparent dots indicate the values for individual ROI, while the large opaque circles indicate network averages.
Next, we examined further whether some FC networks tend to show non-monotonic age-related changes consistently for temporal and spatial scales. Given that the results in the present study were based on three separate datasets, non-monotonic patterns manifest in two different ways. First, the results from quadratic regression models have shown that for some FC networks in the Child age group, the temporal and spatial scales tend to peak approximately at the age of 12 (Fig. 8, left, green). Second, we also found that for some prefrontal areas in the Child age group, the age coefficients for both scales were significantly positive (Fig. 8, left, red). Given that the age coefficients for both scales were significantly negative for all areas in the Adult age group (22 years and older), this suggests that the temporal or spatial scale might reach their peak values around the age of 22 for areas with positive age coefficients in the Child age group. Therefore, we looked for the ROI that showed either a significant non-monotonic trend or a significant positive age coefficient in both temporal and spatial scales. For example, some areas showed positive age coefficient for the spatial scale but a quadratic change for timescale in the Child age group (Fig. 8, right, yellow). The results from this analysis revealed that for some regions in the prefrontal and parietal cortical areas as well as the insular cortex, both temporal and spatial scales might increase and reach their peak values before the age of 22 (Fig. 8, right).
Brain maps showing non-monotonic age-related changes in temporal and spatial scales. The ROI with significant positive age coefficient in the Child age group are shown in red, while the regions with a significant quadratic trend in the same group are shown in green for temporal (top left) and spatial (bottom left) scales. The ROI showing either significant quadratic trend or significantly positive age coefficient for both scales are also shown (right) with different colors indicating different combinations of these two patterns.
Age-related changes in cortical myelination
Next, we examined the regional variability in cortical myelination and its age-related changes. Consistent with the results in previous studies (Grydeland et al., 2019), brain maps of average myelin show an overall increase from childhood to adulthood and became stable in the Old age group, with the highest values seen around the central sulcus and posterior medial regions near the calcarine fissure (Fig. 9a). Similar to the brain maps for temporal and spatial scales, these myelin maps were highly correlated across different age groups at the network level (r > 0.85, p < 0.001; Fig. 9b,c).
Cortical myelination across the lifespan. a, Brain maps of the mean myelination maps for Child (left), Adult (center), and Old (right) age groups. b, c, Scatter plots for comparing the mean myelinations for Child versus Adult (b) and Adult versus Old (c) age groups. Dotted lines show group means.
Cortical myelination increased steeply throughout the brain in the Child age group and more slowly in the Adult and Old age groups (Fig. 10a). In particular, the age coefficient for cortical myelination was positive in all FC networks for the Child age group, and a significant non-monotonic trend was not observed either for the brain average (Fig. 10b) or for any individual networks (Fig. 10c). In addition, the full linear mixed effects model did not fit the age-related changes in cortical myelination better than the model without the interaction terms in the Adult age group (Table 2), indicating that cortical myelination changes uniformly and consistently throughout the brain during adulthood. In contrast, the full model fit the results better in the Child and Old age group, and the rate of age-related changes for cortical myelination was significantly different from the mean for some FC networks. For example, Somatosensory-Motor-B network showed a significantly more positive slope than the average in both Child and Old age groups, whereas Limbic-B and Default-C networks showed significantly more negative slopes in both age groups (Table 2).
Age-related changes in cortical myelination. a, Brain maps of the age coefficient for myelination in Child (left), Adult (center), and Old (right) age groups. Gray regions indicate the ROI in which the age effect was not statistically significant (p > 0.05). b, Scatter plots of the average myelination for the entire cortex across individual subjects. c, LOWESS fit for myelination in each FC network. Dotted/dashed lines show networks that are significantly more positive/negative than the average. The values for the Adult age group were decreased by 0.055 to compensate for the discontinuations at the age boundaries shown in b.
We then examined whether and how cortical myelination and its age-related changes might be related to temporal and spatial scales of rs-fMRI activity across different age groups. First, we found that cortical myelination was not significantly correlated with the timescale of rs-fMRI signals across FC networks in any age group (r = 0.02, 0.32, 0.21 for Child, Adult, and Old age groups, respectively, p > 0.2; Fig. 11, top). Second, the age coefficients for timescale and myelination were significantly correlated for the Adult (r = −0.51, p < 0.05) and Old age groups (r = −0.66, p < 0.005), but not for the Child age group (r = −0.182, p > 0.4; Fig. 11, bottom). This suggests that the age-related changes in temporal scale and myelination are more closely coordinated during adulthood than during childhood. Third, we found that spatial scale of rs-fMRI activity was significantly correlated with cortical myelination across FC networks in all age groups (r = 0.70, 0.53, 0.50 for Child, Adult, and Old age groups, respectively, p < 0.05; Fig. 12, top). Finally, the age coefficients for spatial scale and cortical myelination were significantly correlated only in the Adult age group (r = 0.15, −0.64, and −0.41 for the Child, Adult, and Old age groups, respectively; Fig. 12, bottom).
Relationship between timescale of rs-fMRI activity and myelination. Scatter plots for cortical timescale of rs-fMRI signals and cortical myelination (top) and their corresponding age coefficients (bottom). Small transparent dots indicate the values for individual ROI, while the large opaque circles indicate network averages.
Relationship between spatial scale of rs-fMRI activity and myelination. Scatter plots for spatial scale of rs-fMRI signals and cortical myelination (top) and their corresponding age coefficients (bottom). Same format as in Figure 11.
Relationship among multiple cortical gradients
We compared the brain maps for temporal and spatial scales of rs-fMRI activity and cortical myelination together with their corresponding age coefficients to the ranking along the sensorimotor-association (S-A) axis (Fig. 1b, Fig. 13). Consistent with the findings from previous studies (Sydnor et al., 2023), we found that the cortical myelination had a strong and significant negative correlation with the ranking along the S-A axis (Fig. 13, right). We also found a significant negative correlation between the spatial scale and the rank along the S-A axis at both ROI and network levels in each age group, indicating that sensory cortical areas tended to have longer spatial scales (Fig. 13, center). We did not find any significant correlation between the timescale of resting-state activity and the rank along the sensorimotor-association axis in any age group (Fig. 13, left). On the other hand, we found that the age coefficients for the timescale were significantly correlated only for the Child age group at both ROI and network levels.
Relationship between the ranking along sensorimotor-association (S-A) axis and three measures examined in the present study (timescale, spatial scale, and cortical myelination; top) and their corresponding age coefficients (bottom) at ROI and network levels. Significance thresholds are indicated by dashed lines, with an asterisk indicating significant correlations for a given age group (p < 0.05).
To examine more systematically how the similarities among these different maps change across different age groups, we visualized the correlations among various brain maps in each age group as a graph in which an edge or link between any two nodes represents the correlation between the two corresponding brain maps (Fig. 14). These graphs show that some brain maps consistently show significant correlations across all age groups. For example, spatial scale and myelination were positively correlated, whereas both spatial scale and cortical myelination were negatively correlated with the S-A axis ranking, in all age groups (Fig. 14, light colors). In addition, correlations between some of these measurements were significant only in some age groups (Fig. 14, dark colors). For example, the negative correlation between the age coefficient for the spatial scale and the cortical myelination was present only in the Child and Old age groups, suggesting that they might be coordinated only during development and aging. On the other hand, correlations between some brain maps were significant only in the Child age group, suggesting that their age-related changes might be coregulated during development. For example, the age coefficient for the timescale was positively and negatively correlated with S-A rank and cortical myelination, respectively, only for the Child age group. Finally, coupling between some measures appeared only in the Adult and Old age groups, such as the negative correlation between the age coefficients for timescale and cortical myelination, and the positive correlation between the age coefficient for spatial scale and S-A rank.
Relationship between multiple cortical maps for different age groups. Correlations among different brain maps for temporal and spatial scales of rs-fMRI activity, cortical myelination, and the ranking along the S-A axis are visualized as an edge between the two nodes corresponding to different brain maps. The brain maps for the age coefficients for timescale, spatial scale, and myelination are indicated by the circles with symbol Δ. Only connections between the maps with significant correlations are shown, with the color and thickness of the line indicating the sign and magnitude of correlation coefficient. Significant correlations consistent across all age groups are shown in light colors.
Discussion
In the present study, we found that temporal and spatial scales of resting-state fMRI signals in the human cortex tended to decrease with age, while they increased in some prefrontal regions during childhood. Although the regional variations in these two scales were not correlated consistently across different age groups, the rates of their age-related changes were correlated across cortical networks in all age groups. We also found that the age-related changes in spatial and temporal scales were closely coordinated with the level of cortical myelination or the rate of its age-related changes across the lifespan. Although we did not find any correlation between the timescale and the ranking along the sensorimotor-association (S-A) axis, the rate of age-related changes in the timescale increased along the S-A axis in the Child age group. Finally, both the spatial scale and myelination decreased along the S-A axis consistently in all age groups, whereas the rate of age-related changes in the spatial scale increased along the S-A axis in the Adult and Old age groups.
Temporal and spatial scales of resting-state activity
In previous studies, the so-called intrinsic timescale of cortical activity was often measured by the decay rate of the temporal autocorrelation in neural activity (Murray et al., 2014; Ito et al., 2020; Raut et al., 2020). Here, we used the lag-1 autocorrelation to measure the timescale in rs-fMRI activity for several reasons. First, the lag-1 autocorrelation is strongly correlated with the first principal component extracted from diverse temporal features of the resting-state activity (Shafiei et al., 2020). Second, the regional variation in the lag-1 autocorrelation reliably captures the individual variability in the resting-state activity and its graphic topology (Shinn et al., 2023). Previous studies (Yeo et al., 2011, Demirtaş et al., 2019), especially those based on electrophysiological recordings (Murray et al., 2014; Gao et al., 2020; Siegle et al., 2021; Manea et al., 2022; Wolff et al., 2022; Manea et al., 2024; Song et al., 2024), have found a hierarchical ordering of timescales across the cortex, with unimodal and transmodal areas exhibiting shorter and longer timescales, respectively (Murray et al., 2014; Ito et al., 2020; Raut et al., 2020). In contrast, we found that the timescale estimate based on the lag-1 autocorrelation was not significantly related to the hierarchical gradient along the sensorimotor-association axis in any of the age groups (see also Shinn et al., 2023). In fact, a recent study also found that the first principal component estimated from various temporal features of rs-fMRI activity does not follow such a gradient (Shafiei et al., 2020). These results suggest that the resting-state BOLD activity might be characterized by multiple timescales (Spitmaan et al., 2020; Song et al., 2024). In contrast, using a novel method to estimate the anatomical gradient of the spatial scale in the rs-fMRI activity, we found that spatial scale was significantly related to the ranking along the S-A axis.
We also found that temporal and spatial scales of rs-fMRI activity undergo systematic changes in coordination with cortical myelination. Sensory and motor cortical areas are more myelinated than association cortical areas (Flechsig, 1901; Glasser and Van Essen, 2011; Huntenburg et al., 2018). This might reflect different levels of plasticity across cortical areas, since myelination inhibits neural plasticity (Takesian and Hensch, 2013). Previous studies showed that the time course of myelination as well as the level of myelination differs along the sensory-association cortical axis, proceeding earlier in the unimodal sensory and motor areas than the transmodal association cortical areas (Grydeland et al., 2019; Baum et al., 2022). Consistent with these previous findings, we found that the rate of myelination in the Child age group was negatively associated with the ranking along the S-A axis. In addition, the rate of age-related changes in the timescale during childhood was positively correlated with the ranking along the S-A axis during childhood and negatively with the changes in cortical myelination during adulthood and aging. This suggests that the factors regulating the timescale of rs-fMRI activity might differ for development and aging. In contrast, we found that the spatial scale was related to myelination and S-A axis consistently throughout the lifespan, although it decreased faster in sensory and motor cortical areas than in association cortical areas during adulthood and aging. In addition, during adulthood, age-related changes in spatial scale and cortical myelination were correlated. These results suggest that throughout the lifespan, age-related changes in temporal and spatial scales might be influenced differently by various anatomical and functional properties of cortical networks. Timescales of resting-state fMRI may be altered in some clinical conditions (Zhang et al., 2024; Uscătescu et al., 2023) and related to memory performance (Wu and Gollo, 2025). However, more studies are needed to understand the specific aspects of cognitive functions that can be predicted by the temporal and spatial scales of resting-state activity.
Implications for brain development and aging
Several studies have found an age-related decrease in intrinsic timescales of BOLD (Geerligs et al., 2017; Truzzi and Cusack, 2023; Wu and Gollo, 2025) and EEG (Gao et al., 2020) activity in the human cortex, although they did not cover the entire lifespan. Consistent with these previous findings, we found that the timescale of rs-fMRI activity mostly decrease throughout the lifespan. We also found that the spatial scale of rs-fMRI activity decreased with age. Given that the resting-state functional connectivity is dominated by the relatively low-frequency power (Cordes et al., 2001), an age-dependent decrease in the spatial scale of BOLD activity is consistent with the previous findings that the strength of functional connectivity tends to decrease with age (Ferreira and Busatto, 2013). Age-related changes in the temporal and spatial scales of rs-fMRI activity also suggest that functions of spontaneous activity related to plasticity might change throughout the lifespan. Longer spatial and temporal scales in spontaneous activity early during development might allow developing brains to integrate incoming information from their surroundings over longer spatial and temporal windows, helping them to learn more general features (Kiebel et al., 2008; Vogelstang et al., 2018). Subsequent decorrelation and reduced synchrony in neural activity during later development might contribute to finer encoding of temporal and spatial information (Golshani et al., 2009; Gribizis et al., 2019; Martini et al., 2021).
We found that age-related changes in temporal and spatial scales were positively correlated in every age group, indicating that changes in both scales might be coordinated throughout the lifespan. On the other hand, a recent study showed the spatial scale estimated globally for the entire cortex increased with age (Shinn et al., 2023). This discrepancy is likely due to the differential effects of age on local versus global functional connectivity, which has been attributed to delayed maturation of higher-order cognitive systems in the brain (Supekar et al., 2010; Betzel et al., 2014; Deery et al., 2023). Consistent with an age-related decline in the low-frequency fluctuation in the global signal reported previously (Ao et al., 2022), we found that the timescale of the global signal tended to decrease with age. Age-related changes in both temporal and spatial scales of rs-fMRI activity were, however, unchanged by whether the global signal was factored out or not, suggesting that they might arise from different underlying mechanisms.
For some regions in the prefrontal cortex, we found that both temporal and spatial scales increased with age during childhood and peaked between the ages of 11 and 12. Previous studies showed that the age-related changes in the average power of low-frequency (<0.1 Hz) fluctuation in the resting-state BOLD signals decreased monotonically in sensory and motor areas but increased in the prefrontal cortex during childhood and started to decrease late in adolescence (Sydnor et al., 2023). The finding that functional connectivity in the default mode network changed non-monotonically during childhood (Faghiri et al., 2019) is also consistent with the changes in the timescale observed in the present study. These results provide additional support to the hypothesis that maturation of the prefrontal cortex might be more delayed compared with the rest of the brain (Dumontheil et al., 2008; Fransson et al., 2011; Sydnor et al., 2021).
Some of the age-related changes in temporal and spatial scales reported in this study might be due to changes in hemodynamic response. In fact, previous studies have found age-related changes in hemodynamic response functions during development (Arichi et al., 2012; Schmithorst et al., 2015) and aging (Abdelkarim et al., 2019; West et al., 2019; Tsvetanov et al., 2021; Turner et al., 2022). Nevertheless, the exact contribution of age-related changes in neurovascular coupling to timescales of resting-state activity is still not fully understood. This is difficult to study because both spontaneous activity (Khazipov and Luhmann, 2006) and neural response to sensory stimuli change during development and aging (Lippé et al., 2007; Fabiani et al., 2014; Gao et al., 2020; Chettouf et al., 2022; Springer et al., 2023). More research is therefore needed to elucidate the precise manner in which the temporal and spatial changes in BOLD activity reflect the neural and neurovascular factors.
Data availability
The final numerical values used to make all figures in the current study, along with the code to generate them, are available from the authors upon request.
Footnotes
- Received March 23, 2025.
- Revision received November 6, 2025.
- Accepted November 11, 2025.
-
Data used in the preparation of this manuscript were obtained from the National Institute of Mental Health (NIMH) Data Archive (NDA). NDA is a collaborative informatics system created by the National Institutes of Health to provide a national resource to support and accelerate research in mental health. Dataset identifier: 10.15154/5f0d-z111. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or of the Submitters submitting original data to NDA. This work was supported by Neurogazer Inc.
-
The authors declare that they have known competing financial interests or personal relationships which could have appeared to influence the work reported in this paper, as some are employees of Neurogazer Inc. which funded this study. In addition, J.D.M. is a part-time contractor for Johnson & Johnson Innovative Medicine.
- Correspondence should be addressed to Daeyeol Lee at daeyeol{at}jhu.edu or Timothy J. Vickery at tvickery{at}udel.edu.
- Copyright © 2025 the authors
