Growth of White Matter Signal Aberrations and Cortical Changes in Men and Women: A 9‐Year Multiple‐Time‐Point Longitudinal MRI Study in Older Adults

ABSTRACT

In aging, white matter hyperintensities (WMH) or white matter signal aberrations (WMSA), a proxy for cerebral small vessel disease, are the most common age‐related finding on brain MRI scans, with women reported to have a greater burden. Previous studies found that WMH/WMSA were linked to lower cortical volume and thickness. However, no study has investigated the connection between WMSA and both cortical thickness and area using multiple assessments over 9 years. We hypothesized that increasing WMSA was associated with a decrease in both cortical thickness and area over time, compounding the age‐typical decrease, and that the effect was more pronounced in women than men. Our cohort consisted of 104 older adults from the general population, born between 1936 and 1942, scanned at the same 3T scanner at baseline (mean age 72.4 years), and followed up at 1‐, 3‐, 5, and 9 years. We analyzed the 3D MPRAGE scans in FreeSurfer's longitudinal pipeline and extracted the volume of WMSA, cortical thickness and area, grey matter‐white matter intensity ratio, plus hippocampal and intracranial volume. Linear mixed‐effects models included the interaction between sex and WMSA on cortical thickness and area, and hippocampal volume. We also performed sex‐specific analyses. Finally, complete case and sensitivity analyses with known confounders were run. In both men and women, baseline WMSA volume was 3.6 cm³ (95% CI, 2.1–6.2), followed by an average annual growth rate of 7.2% (95% CI, 6.1–8.3). Increasing WMSA over time was associated with accelerated cortical thinning. In men, the association between WMSA and cortical thinning was more pronounced and present in more lobes than in women, despite similar WMSA volume and growth in both sexes. In the lobes with accelerated cortical thinning, increasing WMSA was also associated with an increase in cortical area present only in men. The sex‐specific analyses revealed even more pronounced differences in the relationships between WMSA growth and cortical features over time. The changes in cortex associated with WMSA volume were superimposed on the changes related to aging in both sexes. Change in WMSA over time was not directly associated with change in hippocampal volume. Complete case and sensitivity analyses supported the main findings. Thus, contrary to our hypothesis, WMSA change over 9 years was connected to greater cortical thinning alongside a surprising increase in cortical area in men, despite men and women having similar WMSA volume and growth. Our results underscore the presence of marked sex differences in cortical plasticity in the aging brain in the presence of WMSA. WMSA, the most common age‐related proxy for small vessel disease, contributed significantly and above that related to aging, to cortical changes, more so in men than women.

Greater growth of white matter signal aberration (WMSA) volume, the most common age‐related finding on brain MRI, is connected to increased cortical thinning, more so in men than women. In men only, the cortical regions experiencing increased thinning due to WMSA growth displayed a concomitant cortical area expansion.

1. Introduction

Brain maintenance, that is, preventing or delaying the development of age‐related brain pathology and structural changes (Stern et al. 2020), is essential for aligning health span with the increasing life span of today's aging population (Crimmins 2015). The most common lesion detected in the aging brain is white matter hyperintensities (WMH) on FLAIR or other T2‐weighted images (Fazekas et al. 1998; Håberg et al. 2016; Biesbroek et al. 2025; de Leeuw et al. 2001). WMH are also referred to as leukoaraiosis (Hachinski et al. 1987) or white matter (WM) lesions (Kim et al. 2008). When depicted on T1‐weighted images, these WM lesions are denoted WM hypointensities (Olsson et al. 2013). White matter signal abnormalities or aberrations (WMSA) is another term describing these lesions irrespective of scan weighting (Fazekas et al. 1993; Salat et al. 2009; Lindemer et al. 2015). In clinical practice, WMH are assessed on FLAIR scans, often with semi‐quantitative scales such as Fazekas (Fazekas et al. 1987) or Scheltens (Scheltens et al. 1993) scores. In research, WMH/WMSA volume is frequently used, with the gold standard being manually delineated WMH volume from FLAIR (Tuladhar et al. 2015; Vangberg et al. 2019). Nevertheless, most studies today use automated methods (in‐house (Rizvi et al. 2021) or open source (Mayer et al. 2021; Arild et al. 2022)) or semi‐automated (Jacobs et al. 2014; Landman et al. 2021) methods for segmenting WMH/WMSA. Clinical semi‐quantitative WMH scores and WMH/WMSA lesion mapping are used mainly in cross‐sectional scientific studies (Vangberg et al. 2019; Honningsvåg et al. 2018; Rostrup et al. 2012; Prins et al. 2004). In the following, we will refer to age‐related WMSA on FLAIR and/or T2‐weighted scans as WMH, while corresponding lesions on T1‐weighted scans or combinations of T1‐ and other scan weightings will be denoted WMSA.

WMH/WMSA are visible lesions that vary in size with a predilection for periventricular and deep WM regions, considered to reflect the presence of cerebral small vessel disease (SVD) (cf. Figure 1) (Debette and Markus 2010; Pantoni 2010; Wardlaw et al. 2013). Within WMH/WMSA, a range of pathologies are described, including axonal injury and loss, dys‐ and demyelination, microglial and glial activation as well as gliosis, enlarged perivascular spaces, disruption of the blood–brain‐barrier, hypoxia, and edema (Wardlaw et al. 2015; Iordanishvili et al. 2019; Simpson et al. 2007; Gouw, Seewann, et al. 2008; Fernando et al. 2006). WMH lesions are usually more spatially extensive than WMSA, with WMSA considered to reflect more severe WM pathology and loss of myelin (Olsson et al. 2013; Dadar, Maranzano, et al. 2018; Melazzini, Vitali, et al. 2021; Melazzini, Mackay, et al. 2021). Nevertheless, studies comparing the volume of WMH (manual or automated) and WMSA (automated) find that these measures are correlated (Olsson et al. 2013; Dadar, Maranzano, et al. 2018; Wei et al. 2019). Furthermore, WMH and WMSA volumes correlate with Fazekas and Scheltens scales (Vangberg et al. 2019; Andere et al. 2022; Gao et al. 2011). Importantly, studies report similar associations between manually and automated segmented WMH and WMSA volumes as well as semi‐quantitative scores in the same individuals and their neurological diagnosis, cognitive abilities and/or vascular risk factors (Honningsvåg et al. 2018; Dadar et al. 2019; Dadar, Zeighami, et al. 2018). Taken together, these results support that the different methods of visualizing and segmenting WMH/WMSA capture neurobiological pathology relevant to brain function and structure.

FIGURE 1.

Example of T1W and FLAIR scans, both overlaid with the WMSA segmentation for a woman and a man with approximately the same WMSA volume at baseline.

Although WMH/WMSA appear as localized lesions in WM, they are associated with changes encompassing most of the hemispheric WM as shown with diffusion tensor imaging and quantitative MRI (Vangberg et al. 2019; Iordanishvili et al. 2019; James et al. 2023; van Leijsen et al. 2018). Correlations between WMH/WMSA and grey matter (GM) structures are also reported, with larger volumes of WMH/WMSA being associated with lower cortical volume or cortical thickness in cross‐sectional (Tuladhar et al. 2015; Vangberg et al. 2019; Mayer et al. 2021; Jacobs et al. 2014; Seo et al. 2012; Raji et al. 2012) and longitudinal studies (Rizvi et al. 2021; Bernal et al. 2024). Surprisingly, the relationship between WMH/WMSA and cortical area has not been established. Cortex is characterized by two dimensions, thickness and area, which are ontologically (Rakic 1988) and genetically (Grasby et al. 2020) distinct. Cortical thickness is considered more related to environmental and health factors, while area is regarded as more associated with genetic factors (Grasby et al. 2020). In aging, both cortical thickness and area decrease (Sele et al. 2021; Nyberg et al. 2023), although reduction in area appears less consistent (Yu 2024). Whether WMH/WMSA contribute additionally or synergistically with age to reducing both cortical thickness and area remains unknown. Some studies have also connected greater WMH/WMSA volume to smaller hippocampal volumes cross‐sectionally (Fiford et al. 2017; Rizvi et al. 2023) and longitudinally (Luo et al. 2023), or greater hippocampal atrophy with greater baseline WMSA volume (Fiford et al. 2017; Legdeur et al. 2019), but other studies report no relationship between WMSA and hippocampal volume cross‐sectionally or longitudinally (Legdeur et al. 2019; Gattringer et al. 2012). Furthermore, presence of WMH/WMSA, WMH volume as well as Fazekas and Scheltens scores are reported to be more common and extensive in women than in men at the same age (Håberg et al. 2016; Vangberg et al. 2019; Fatemi et al. 2018; de Kort et al. 2025; Sachdev et al. 2009). Still, whether WMH/WMSA influence GM structures differently in men and women remains unknown. Moreover, most studies on the relationship between WMH/WMSA and cortical thickness and volume, as well as hippocampal volume, are cross‐sectional or based on two scans within a 3–4‐year period. To assess brain maintenance, better evaluate age effects, and separate these from WMH/WMSA growth effects on GM, longitudinal data with more than two observations over time are required (Stern et al. 2020; Pfefferbaum et al. 2013).

In this study, we therefore investigated the relationship between WMSA volume, cortical thickness and area, and hippocampal volume across 9 years and five scan time points, in older adults from the general population. We hypothesized that increasing WMSA volume with age would add to cortical thinning as well as area loss previously reported in aging. Since women are found to have more WMH/WMSA than men, we predicted greater decreases in thickness and area over time in women than in men related to WMSA. For hippocampal volume, we assumed no relationship with an increasing WMSA volume over time.

2. Material and Methods

2.1. Ethics

The study was approved by the Regional Committee for Medical Research Ethics (REK 2012/849) and adhered to the Helsinki Declaration. Participants gave written informed consent.

2.2. Participants

The participants in this study are from the Generation 100 Brain Study, a study that assesses both the long‐term effect of a randomized controlled trial (RCT) exercise intervention (see below) and aging on brain structure until 2035. All adults born between 1936 and 1942 registered in the Norwegian National Population Registry as independently living citizens of Trondheim municipality were personally invited by letter in the regular mail in 2012 to a 5‐year RCT on the effect of exercising on mortality, the Generation 100 Study (NCT01666340, ClinicalTrials.gov registry) (Stensvold et al. 2015). Before randomization, participants were asked if they also wanted to join the Generation 100 Brain Study with brain MRI scanning during and following the completion of the RCT, to assess brain changes related to both the RCT and aging. Participants were randomized (2:1:1) into control, that is, following national physical activity guidelines of 30 min or more of moderate to vigorous physical activity almost every day of the week, or supervised exercise twice a week with either High Intensity Interval Training (HIIT) consisting of 10 min warm‐up followed by four 4 min exercise intervals at 85%–95% of peak heart rate, or Moderate Intensity Continuous Training (MICT) for 50 min at 70% of peak heart rate (Stensvold et al. 2015). Exclusion criteria were any disease (including dementia) or other issues limiting life expectancy and/or precluding exercising, plus participation in other studies. No significant effect of the RCT on mortality or morbidity was detected (Stensvold et al. 2020), and secondary analyses did not find any effect of group allocation on scores on the Montreal Cognitive Assessment (MoCA) (Zotcheva et al. 2022) or other cognitive tests (Sokołowski et al. 2021).

A total of 111 (55 men, 56 women) consented to participate in the Generation 100 Brain Study. Of these, six were excluded due to MRI contraindications or previous neurological/neurosurgical disorders that influenced brain structure or image quality, leaving 105 participants (53 men, 52 women). The RCT intervention group allocation did not significantly influence WMH or WMSA volumes (Arild et al. 2022) or WM microstructure as assessed with diffusion tensor imaging (Pani et al. 2022), or cortical volume from FreeSurfer analysis (Pani et al. 2021). But the supervised exercise group experienced greater hippocampal volume loss during and following the intervention than the controls (Pani et al. 2021; Reitlo et al. 2026), and had lower hippocampal NAA level (Reitlo et al. 2023), a MR spectroscopy proxy for synaptic loss and reduced neuronal well‐being.

2.3. Demographic and Health Data

Date of birth, sex, and level of education (primary school, high school, and university) were obtained at baseline (Stensvold et al. 2015). These sociodemographic variables are confounders linked to both the volume and/or rate of change in WMH/WMSA (Håberg et al. 2016; de Leeuw et al. 2001; Arild et al. 2022; Dadar, Maranzano, et al. 2018; Sachdev et al. 2009), as well as cortical (Wierenga et al. 2022; Pintzka et al. 2015) and subcortical (Pintzka et al. 2015; Homayouni et al. 2023; Noble et al. 2012) structural features.

The health variables at baseline were obtained from self‐report and clinical assessment, including blood tests, and are also known confounders, connected to volume of and/or change in WMH/WMSA, and/or cortical, and/or subcortical features. The variables were hypertension defined by self‐reported hypertension and/or use of antihypertensives, and/or systolic blood pressure of ≥ 140 mmHg and/or diastolic blood pressure of ≥ 90 mmHg (Rostrup et al. 2012; Sachdev et al. 2009; Ye et al. 2024), diabetes based on self‐report and/or hemoglobin A1c (HbA1c) ≥ 6.5% (Sims et al. 2014; Gouw, van der Flier, et al. 2008; Brundel et al. 2010), body mass index (BMI) in kg/m² (King et al. 2014; Lin et al. 2024), and smoking status, defined as current smoker or non‐smoker (Liao et al. 1997; Zhao et al. 2019). Finally, ApoE 4 allele (APOE4) status has been associated with larger WMH volume (Brickman et al. 2014; Kumar et al. 2022), thinner cortex (Fennema‐Notestine et al. 2011; Leclaire et al. 2024), and smaller hippocampus volume (O'Dwyer et al. 2012; Schuff et al. 2009).

2.4. MRI Scanning

The participants underwent the same standardized structural MRI scan protocol on the same 3T Magnetom Skyra scanner (Siemens AG, Erlangen, Germany) equipped with a 32‐channel head coil at baseline, 1‐, 3‐, 5‐, and 9 years. In this study, we used the 3D T1‐weighted, FLAIR, T2‐weighted, and susceptibility‐weighted scans (Table 1). Standardized neuroradiological reading was performed by two blinded experts in consensus, and included Fazekas score (Sims et al. 2014), presence and number of infarctions (lacunar, cortical, and subcortical), stratification of microbleeds into > 5 or ≤ 5 lesions, and perivascular spaces in the basal ganglia and/or centrum semiovale into > 5 or ≤ 5 visible spaces, to map the presence of cerebral SVD.

TABLE 1.

Parameters for MRI scans used at all time points.

Parameter	3D T1W	3D T2W	3D FLAIR	2D SWI
Orientation	Sagittal	Sagittal	Sagittal	Axial
Fold‐over direction	AP	AP	AP	RL
TR	1900 ms	3200 ms	5000 ms	27 ms
TE	3.16 ms	412 ms	388 ms	20 ms
TI	900 ms	n/a	1650 ms	n/a
No. of slices	192	176	176	55
Slice thickness	1.0 mm	1.0 mm	1.0 mm	1.5 mm
Slice gap	—	—	—	0 mm
FOV	256 × 256 mm	250 × mm	256 × 256 mm	220 × 220 mm
Matrix (resolution)	256 × 256 mm	256 × 256 mm	256 × 256 mm	256 × 243 mm
Voxel size	1.0 × 1.0 × 1.0 mm	1.0 × 1.0 × 1.0 mm	1.0 × 1.0 × 1.0 mm	0.9 × 0.9 × 1.5 mm
Flip angle	9°	T2 variable	T2 variable	15°
Scan technique	Turboflash	Space IR	Space IR	GRE SWI

Abbreviations: AP, anterior–posterior; FOV, field of view; GRE, gradient echo; IR, inversion recovery; mm, millimeter; ms, milliseconds; RL, right–left; SWI, susceptibility‐weighted imaging; TE, echo time; TI, inversion time; TR, repetition time; W, weighting.

2.5. Image Analysis

The FreeSurfer (v.7.3.0, http://surfer.nmr.mgh.harvard.edu/) package and its longitudinal streams (Reuter et al. 2012) were used to calculate cortical area and thickness (Dale et al. 1999), WMSA and hippocampal volumes (Fischl et al. 2002), as well as estimated intracranial volume (ICV) from the T1‐weighted scans (Fischl 2012). We also extracted the grey matter (GM)‐white matter (WM) intensity ratio (GM/WM) for each lobe. The GM/WM intensity ratio was calculated as follows: GWR = (100(WM − GM))/(0.5(WM + GM)) with WM signal intensity measured 1 mm below the WM surface, and GM signal intensity measured 30% into the cortex from the gray/white border (https://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/dev_binaries/centos6_x86_64/pctsurfcon) (Putcha et al. 2023; Salat et al. 2011). Cortical thickness changes in aging are found to be associated with changes in the GM/WM intensity ratio (Vidal‐Piñeiro et al. 2016), which might be particularly relevant in the presence of WMSA, which impact WM properties far beyond the visible lesions (Vangberg et al. 2019). Also, iron depositions as part of brain aging could influence T1 relaxation in the junction between WM and GM (Ogg and Steen 1998), and thus influence cortical thickness measures.

The vertex‐wise estimates of cortical thickness and area, as well as the GM/WM intensity ratio results from the surface‐based stream, were combined with an automatic parcellation procedure (Fischl et al. 2004) that estimated the mean cortical thickness, total area, and mean GM/WM intensity ratio for the frontal, parietal, temporal, occipital, cingulate cortex, and insula lobes, as defined in FreeSurfer (https://surfer.nmr.mgh.harvard.edu/fswiki/FreeSurferWiki/LCN‐CortLobes). The values for the left and right hemispheres were aggregated by summing the cortical area and averaging the volumetric measures.

Regional differences in WMSA volume and growth between men and women were assessed with a “bullseye parcellation” of cerebral white matter divided into the four classical lobes: frontal, temporal, parietal, and occipital (Sudre et al. 2018). In short, we used the FreeSurfer segmentations to divide cerebral white matter into the four lobar zones in each hemisphere. Subsequently, each lobar zone was divided into four equally spaced concentric layers according to the distance to the ventricles, giving a total of 32 regions (So 2025). WMSA load was calculated as the percentage of WMSA volume in each region.

2.6. Statistics

Demographic and health data were summarized using means and standard deviations for numerical data and counts and percentages for categorical data. WMSA (in cm³) had a right‐skewed distribution and was log‐transformed before statistical analysis.

A Wilcoxon rank sum test for continuous variables, Pearson's chi‐squared test for categorical variables with cell counts ≥ 5 and Fisher's exact test for categorical variables with cell counts < 5 were used to investigate whether there were differences in the sociodemographic, clinical, and radiological variables at baseline between those who completed all five examinations and those that were lost to follow‐up to establish missing at random or not. The same methods were used to investigate if there were significant sex differences in Fazekas scores.

WMSA growth rate was estimated using a linear mixed model with random intercept and slope, and log(WMSA) as the outcome, age, and the interaction between age and sex as predictors, adjusting for ICV. Using log(WMSA) as the outcome allowed us to compute percentage change in WMSA volume with age. Regional differences between men and women in WMSA load and growth were assessed in a linear mixed model using random intercept and slope with WMSA load as the outcome, age, and the interaction between age and sex as predictors. The WMSA load was not log‐transformed because the measurements were deemed normally distributed based on visual inspection of the histograms.

Associations between WMSA and cortical thickness and area over 9 years were assessed using a linear mixed‐effects model. The outcome variables were mean cortical thickness and area of the combined left and right frontal, parietal, cingulate, insula, temporal, and occipital lobes. Natural logarithm‐transformed WMSA was the main predictor, with age at each time point and sex as covariates. The interaction between sex and WMSA was included to test whether WMSA affected men and women differently. ICV and GM/WM intensity were included as confounders to adjust for differences due to brain size and GM/WM intensity on the outcome variable. GM/WM intensity differences may bias cortical thickness estimates, with lower GM/WM intensity observed in aging (Vidal‐Piñeiro et al. 2016; Pieperhoff et al. 2026) being associated with an overestimation of cortical thickness in FreeSurfer (Westlye et al. 2009). Slope and intercept were included as random effects. The models were rerun separately for men and women to assess the validity of the full model and examine heterogeneity in associations between men and women. Estimates of the percentage change in cortical thickness and area were calculated as marginal effects (i.e., holding all predictors fixed except one). We calculated the effect of a one‐year change in age and WMSA, where the effect of age was the estimate for age in the regression models while the effect of WMSA was the average yearly change in log(WMSA) multiplied by the estimated effect of log(WMSA) (i.e., the beta estimates for each lobe). This gives the yearly change in thickness and area in (mm or cm²). However, since the average thickness and area of the lobes vary and for easier interpretation, we calculated the percentage change relative to the average thickness or area at baseline for men and women, separately.

Associations between WMSA and hippocampal volume over 9 years were assessed with a linear mixed‐effects model, in the same manner as for cortical thickness and area, excluding GM/WM intensity, but adding intervention groups since the intervention influenced hippocampal volume during and at the end of the intervention (Pani et al. 2021), as well as at the 9‐year follow‐up (Reitlo et al. 2026). Slope and intercept were included as random effects. As for cortex, analyses were also run separately for men and women. Additionally, we included a three‐way interaction between the intervention group, sex, and WMSA due to our earlier findings of greater hippocampal atrophy in those allocated to HIIT exercise (Pani et al. 2021; Reitlo et al. 2026, 2023).

2.7. Power Calculations

Interaction terms typically require larger sample sizes to achieve similar statistical power as main effects (Leon and Heo 2009). To assess the power needed to uncover significant interactions between sex and WMSA volume on GM structures, we performed power calculations using a Monte‐Carlo approach implemented in the “simr” package (Green and MacLeod 2016). An artificial dataset was constructed from parameters from the literature (Shaw et al. 2016; Crivello et al. 2014) and from an independent in‐house dataset in a general older population (average age at baseline 70.6 years, with approximately 2 years between baseline and follow‐up), scanned on a different 3T scanner but processed similarly in FreeSurfer. We ran simulations to estimate power across a range of effect sizes for cortical thickness and hippocampal volume. The analyses indicated that, with a sample size and attrition rate similar to those observed in our study, we had 80% power to detect a statistically significant positive or negative sex × WMSA interaction corresponding to approximately a 0.025 mm change in cortical thickness and a 0.035 cm³ change in hippocampal volume per unit change in WMSA in men. The full details of the power analysis, including the code, are provided in Supporting Information.

Please note that we were unable to estimate power for cortical area since we could not find longitudinal cortical area measurements in the cohort's age range.

2.8. Sensitivity Analyses

In the sensitivity analyses, we added the confounders education, smoking, diabetes, hypertension, BMI, and APOE4 status as covariates to the regression models with cortical thickness, area, and hippocampal volume as outcomes. In the cortical thickness and area models, intervention group was added even though the exercise intervention was shown not to influence cortical volume (Pani et al. 2021).

2.9. Complete Brain MRI Sample Analysis

The main linear mixed‐effects analyses assessing associations between WMSA and cortical thickness and area, and hippocampal volume over 9 years, including the interaction between sex and WMSA, were rerun in participants with complete data at all five time points.

Statistical analyses were performed using R version 4.5.3, with linear mixed models computed using the lme4 package (Bates et al. 2015). Model fit was assessed with R ² and conditional R ². Multicollinearity was evaluated with the variance inflation factor (VIF) computed with the “performance” package (Lüdecke et al. 2021). We considered VIF values < 5 as acceptable. Bonferroni's method was used to correct for multiple comparisons across the six lobes for the thickness and area models and across the 32 regions when testing for regional differences in WMSA load. The statistics on hippocampus volumes were not corrected for multiple comparisons. A corrected p value < 0.05 was considered significant. The ggplot2 and ggseg packages were used for plotting (Mowinckel 2020).

3. Results

The flow chart in Figure 2 shows the inclusion and attrition in the study. After FreeSurfer processing, 104 participants had brain morphological measures at one or more time points. A total of 64 participants had morphological measurements at all time points.

FIGURE 2.

Flowchart of participant inclusion, loss to follow‐up, number of successful brain MRI scans, and successful analysis in FreeSurfer during the 9‐year study period.

3.1. Sociodemographic and Health Variables

Sociodemographic and health variables at baseline relevant to WMSA volume and brain morphometry are presented in Table 2. In general, most participants held a university degree and had health values within normal ranges.

TABLE 2.

Participants' characteristics at baseline.

Characteristic	Overall, N = 104	Men, N = 52	Women, N = 52
Age (years)	72.4 (1.9)	72.5 (1.9)	72.3 (1.9)
Intervention group
Control	48 (46%)	23 (44%)	25 (48%)
MICT	24 (23%)	11 (21%)	13 (25%)
HIIT	32 (31%)	18 (35%)	14 (27%)
Cohabitation status
No one	31 (30%)	7 (14%)	24 (46%)
Spouse/partner	72 (70%)	44 (86%)	28 (54%)
Education
Primary school	9 (8.7%)	3 (5.8%)	6 (12%)
High school	28 (27%)	12 (23%)	16 (31%)
University	67 (64%)	37 (71%)	30 (58%)
APOE e4	19 (22%)	11 (23%)	8 (20%)
Body mass index (kg/m2)	25.9 (3.3)	26.3 (2.9)	25.6 (3.6)
Hypertension	53 (51%)	25 (48%)	28 (54%)
Diabetes	3 (2.9%)	2 (3.9%)	1 (1.9%)
Current smoker	9 (8.7%)	3 (5.8%)	6 (12%)

Note: Values are reported as mean (standard deviation) for continuous data, and the number of observations (%) for categorical data. Intervention group: Control = following national physical activity guidelines; HIIT = high intensity interval training; MICT = moderate intensity continuous training. APOE4 is the presence of one e4 allele, as none of the participants were homozygous for APOE4. APOE e4 status was missing in 15 participants as analysis for APOE4 status included a separate consent and blood draw. Cohabitation status is included since it was used in the randomization to the intervention groups.

3.2. Missing Data

Table S1 provides an overview of the distribution of sociodemographic, clinical, and radiological characteristics at baseline in the lost‐to‐follow‐up group compared to the group that remained in the study for 9 years. The group lost to follow‐up was significantly older at baseline at 73.0 years, compared to 72.0 years in the group that remained in the study, and included more individuals with hypertension at baseline. The lost‐to‐follow‐up group had higher WMSA volume and Fazekas scores indicating more cerebral SVD. The brain morphometric measurements showed that the lost‐to‐follow‐up group had thinner frontal cortex, smaller occipital lobe surface area, and lower hippocampus volume after adjusting for ICV and sex. Thus, those lost to follow‐up were not completely missing at random.

3.3. WMSA Volume and Fazekas Score Across 9 Years

MRI proxies of cerebral SVD; Fazekas scores, number of microbleeds, visible perivascular spaces, and infarctions increased from baseline to 9 years in the cohort (Table 3) and similarly in men and women (Table S2). The proportion of participants with Fazekas score 3, considered pathological at any age (Inzitari et al. 2007), rose with age from 9.2% at baseline to 28.0% at the 9‐year follow‐up (Table 3). There were no significant differences in Fazekas score of 3 between men and women (Table S3), while there was a strong agreement between Fazekas score and WMSA volume per mille (⁰/₀₀) ICV in all participants at each time point (Figure S3).

TABLE 3.

Age, number, and percentage of men and women, Fazekas scores, WMSA volume, prevalence of infarctions, microbleeds, and visible perivascular spaces, cortical thickness and area, hippocampal volume, and GM/WM signal intensity from baseline and at the 1, 3, 5, and 9‐year follow‐ups.

Characteristic	Baseline, N = 98	Year 1, N = 94	Year 3, N = 87	Year 5, N = 85	Year 9, N = 72
Age (years)	72.4 (1.7)	73.8 (1.9)	75.4 (1.8)	77.5 (1.8)	81.2 (1.8)
Sex
Women	51 (52%)	47 (50%)	44 (51%)	41 (48%)	34 (47%)
Men	47 (48%)	47 (50%)	43 (49%)	44 (52%)	38 (53%)
Fazekas score
0	21 (21%)	20 (21%)	15 (17%)	12 (14%)	10 (14%)
1	36 (37%)	33 (35%)	27 (31%)	26 (31%)	17 (24%)
2	32 (33%)	31 (33%)	30 (34%)	31 (36%)	25 (35%)
3	9 (9.2%)	10 (11%)	15 (17%)	16 (19%)	20 (28%)
Infarcts	12 (12%)	11 (12%)	13 (15%)	16 (19%)	19 (26%)
Microbleeds a	4 (4.1%)	2 (2.1%)	4 (4.6%)	5 (5.9%)	7 (9.7%)
Perivascular spaces b	63 (64%)	63 (67%)	62 (71%)	68 (80%)	60 (83%)
WMSA (cm3)	3.6 (2.1, 6.2)	3.9 (2.2, 7.0)	4.8 (2.4, 10.6)	5.5 (3.0, 9.5)	6.7 (3.2, 11.9)
Average cortical thickness (mm)	2.3 (0.1)	2.3 (0.1)	2.3 (0.1)	2.2 (0.1)	2.2 (0.1)
Cortical area (103 cm2)	177 (20)	178 (20)	179 (22)	178 (19)	177 (20)
Hippocampus volume (cm3)	3.9 (0.4)	3.9 (0.4)	3.8 (0.4)	3.8 (0.4)	3.6 (0.3)
Average GM/WM	18.1 (2.8)	18.0 (2.9)	18.0 (3.5)	17.7 (3.5)	17.2 (3.0)

Note: Continuous data are reported as mean (standard deviation), except for WMSA volume, which is presented as median (interquartile range). Categorical data are reported as the number of observations (%). Infarcts: number of participants with one or more cortical or subcortical infarction. Hippocampus volume is the mean of the left and right hemisphere volumes. Average cortical thickness and average GM/WM are the mean values over the cerebral cortex.

^a Microbleeds: number of participants with > 5 microbleeds.

^b Perivascular spaces: number of participants with > 5 visible perivascular spaces in the basal ganglia and/or centrum semiovale.

The volume of WMSA increased with age (Figures 3, 4, 5, Table 3, and Table S2), with an average annual growth of 7.2% (95% CI, 6.1–8.3) assessed by the linear mixed model adjusting for sex and ICV (Figure 3 and Table S3). Neither WMSA volume nor growth rate differed significantly between men and women (Table S3). The quantitative assessment of regional WMSA load with the bullseye segmentation revealed that neither WMSA load, growth, nor spatial distribution differed between men and women (Figure 5), but there was a significant effect of age, with the greatest increase in WMSA load in frontal and temporal WM regions (Figure S1).

FIGURE 3.

Maps of the spatial distribution and frequency of WMSA from baseline to 9 years for women and men. The numbers below the images are the median (95% confidence interval) of the WMSA volume in cm³.

FIGURE 4.

Change in log(WMSA volume in cm³) over the 9‐year study period. Red lines represent women, and blue lines represent men. The thicker red and blue lines are the mixed‐effects trends for women and men, respectively.

FIGURE 5.

Bullseye plot of WMSA load for men and women at each time point. The red hue indicates the percentage WMSA in each region: F = frontal lobe, P = parietal lobe, T = temporal lobe, O = occipital lobe in each hemisphere, presented in neurological convention.

3.4. Relationships Between WMSA Volume, Cortical Thickness, and Area Over Time

There was a slight decrease in cortical thickness across all lobes with higher age (Table 3 and Figure 6). Men had, on average, a thinner cortex than women, and average cortical thickness decreased more in men than in women (Figure 6). For men and women, a small decrease in cortical area was observed with age (Figure 7).

FIGURE 6.

Cortical thickness (mm) in the six lobes from baseline to the 9‐year follow‐up. Women are plotted in red, and men are plotted in blue. The thicker red and blue lines are the linear mixed‐effects trends for women and men based on the estimates in Table 4. See Supporting Information for details.

FIGURE 7.

Cortical area (cm²) in the six lobes from baseline to 9 years of follow‐up. Women are plotted in red, and men are plotted in blue. The thicker red and blue lines are the linear mixed‐effects trends for women and men based on the estimates in Table 5. See Supporting Information for details.

Increasing WMSA with higher age was associated with significantly greater cortical thinning in the frontal, temporal, insula, and cingulate lobes (Table 4 and Table S5 with confidence intervals). The interaction between sex and WMSA revealed accelerated cortical thinning with WMSA growth in men compared with women in the temporal, insula, and cingulate lobes. The rate of cortical thinning associated with WMSA in men was approximately twice that of women (Table 4 and Figure 6), even though men and women had similar WMSA volumes, load, and growth rates across 9 years. Estimates from the separate models for men and women agreed with the model including both sexes, but also revealed that in men, the effect of WMSA on cortical thinning was significant in all lobes except the occipital lobe. In contrast, for women, WMSA was only associated with significant cortical thinning in the temporal, insula, and cingulate lobes (Tables S7 and S8). The parameter estimates from the sex‐specific models (Figure 8 and numerical values in Table S15) showed that at the average annual WMSA growth rate of 7.2% present in both sexes, men experienced an additional 33%–324% greater cortical thinning across all lobes on top of the age‐related effect. In women, the same WMSA growth was associated with more modest cortical thinning, adding 25%–183% to the effect of age, with significant findings only in the temporal, insula, and cingulate cortex.

TABLE 4.

Estimates of the linear mixed‐effects models with log(WMSA) volume at baseline, 1, 3, 5, and 9 years as predictors and cortical thickness per lobe at the same time points as outcomes.

	Frontal	Temporal	Parietal	Occipital	Insula	Cingulate
Age	−0.01***	−0.01***	−0.01***	0.00***	−0.01**	−0.01***
Sex (men)	−0.04	−0.02	−0.08	−0.02	0.09	0.02
WMSA	−0.04*	−0.07***	−0.03	−0.01	−0.10***	−0.08***
ICV	0.00	0.00	0.00	0.00*	0.00	0.00
GM/WM	−0.01***	−0.02***	−0.01**	−0.02***	−0.02***	−0.04***
Sex (men) × WMSA	−0.02	−0.07**	−0.01	0.00	−0.10***	−0.06**
Conditional R 2	0.907	0.957	0.891	0.904	0.962	0.965
Marginal R 2	0.498	0.439	0.323	0.415	0.493	0.471

Note: N = 104 (436 observations), p values corrected with Bonferroni's method. Cortex thickness in mm, age in years, WMSA is the natural logarithm of the WMSA volume in, ICV in cm³, and GM/WM is the lobar gray matter to white matter intensity ratio. Variable inflation factors ≤ 2.04 for all models.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

FIGURE 8.

Visualization of the estimated change in thickness and area in the different lobes resulting from a 1‐year change in age and WMSA volume. The change represents the percentage difference in average lobe thickness or area relative to the baseline values for men and women, separately.

There was no effect of intervention group on cortical thickness or area when group was added to the main models, but the main effect of WMSA remained similarly associated with thickness and area as described above (results not shown).

Across all participants, increasing WMSA volume with higher age was associated with a significant positive effect on cortical area in the temporal, parietal, and cingulate lobes, and positive but non‐significant associations in frontal, occipital, and insula lobes (Table 5 and Table S6 with confidence intervals). A significant interaction between sex and WMSA on cortical area was present, with men having a greater positive association of WMSA on area in the frontal, temporal, and cingulate lobes (Table 5 and Figure 5). The sex‐specific models demonstrated that WMSA growth was significantly associated with increasing cortical area only in men (Tables S9 and S10). The parameter estimates from the sex‐specific models, using the annual WMSA growth rate of 7.2% revealed that for men, increasing WMSA volume opposed the area reduction due to increasing age (Figure 8 and Table S16). Thus, in men, the aging‐related area reduction was nearly offset by the area increase associated with WMSA volume growth. This led to only a slight net reduction in cortical area, except in the insula, where a net annual increase in area of 0.01% was present in men with average WMSA growth (Table S15). The sex‐specific models further showed a small, 0.17%–0.46% additive effect of WMSA volume growth on area increase in men. In women, WMSA growth was not significantly associated with cortical area, with estimates between −0.09% and 0.04%. It is worth noting that the effect of age on cortical area was much smaller for women and only significant in the occipital and cingulate lobes.

TABLE 5.

Estimates (beta values) of linear mixed‐effects models with WMSA volume at baseline, 1, 3, 5, and 9 years as predictors and cortical area per lobe at the same time points as outcomes.

	Frontal	Temporal	Parietal	Occipital	Insula	Cingulate
Age	−1.1***	−1.0***	−1.1***	−0.71***	−0.10*	−0.35***
Sex (men)	23	11	23	11	1.0	3.3
WMSA	3.5	8.7**	7.6*	1.5	0.71	1.7*
ICV	0.25***	0.10***	0.19***	0.08***	0.02***	0.03***
GM/WM	4.3***	4.3***	2.4***	2.0***	−0.06	1.3***
Sex (men) × WMSA	11*	10**	0.30	−0.41	1.0	1.9*
Conditional R 2	0.985	0.986	0.988	0.991	0.982	0.986
Marginal R 2	0.731	0.664	0.716	0.475	0.549	0.612

Note: N = 104 (436 observations), p values corrected with Bonferroni's method. Cortical area in cm², age in years, WMSA is the natural logarithm of the WMSA volume in cm³, ICV in cm³, and GM/WM is the lobar gray matter to white matter intensity ratio. Variable inflation factors ≤ 2.12 for all models.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

3.5. The Relationship Between WMSA Volume and Hippocampal Volume

Hippocampus volume decreased similarly in men and women as they got older (Table 6 and Figure S2), with significant and similar estimates in men and women (Table S11). There was no significant main effect of WMSA or interaction between sex and WMSA on hippocampal volume over time (Table 6). An effect of intervention group on hippocampal volume, with a greater atrophy rate in the HIIT group, was present as expected (Pani et al. 2021; Reitlo et al. 2026) (Table 6). The three‐way interaction between intervention (HIIT) group, sex, and WMSA was significant in both men and women (Table S11). Thus, allocation to HIIT was associated with greater hippocampal atrophy in those with greater WMSA volumes over time.

TABLE 6.

Estimates (beta values) of linear mixed‐effects models with WMSA volume at baseline, 1, 3, 5, and 9 years as predictors and hippocampus volume at the same time points as outcomes.

	Beta
Age	−0.04***
Sex (male)	0.13
Intervention group
Control	—
MICT	0.05
HIIT	−0.17*
WMSA	−0.06
ICV	0.00*
Sex (male) × WMSA	0.00
Conditional R 2	0.980
Marginal R 2	0.302

Note: N = 104 (436 observations), Hippocampus volumes in cm³, age in years, WMSA is the natural logarithm of the WM hypointensity volume in cm³, ICV in cm³. Intervention group: Control = at least 30 min physical activity per day; HIIT = high intensity interval training; MICT = moderate intensity continuous training. Variable inflation factors ≤ 2.7 for all models.

^* p < 0.05.

^*** p < 0.001.

3.6. Sensitivity Analysis

The results of the sensitivity analyses, including the confounders and intervention groups in the statistical models with cortical thickness, area, and hippocampus volume as outcomes, yielded similar results as in the main models. Notably, none of the added confounders were significantly associated with the outcome variables (Figure 9 and Tables S12–S14).

FIGURE 9.

Beta estimates with 95% confidence intervals for the sensitivity and complete data analyses with cortical thickness, area, and hippocampal volume as outcomes. The results of the main models are shown in orange, the sensitivity analysis adjusted for confounders in blue, and the complete data analyses limited to participants (n = 65) who had complete data at all 5 time points and run in the main model in red. The predictor variables were normalized for easier visualization and comparison by subtracting the total mean from the values and dividing by the standard deviation.

In the sensitivity analysis with hippocampal volume as outcome and including the three‐way interaction between intervention group, sex, and WMSA in addition to the confounders, the three‐way interaction between intervention group, sex, and WMSA remained significant for women but not for men. However, the estimates were similar, only less precise for men in the main analysis and the sensitivity analysis with the triple interaction term (Tables S11 and S14).

There were no significant three‐way interactions between intervention group, sex, and WMSA on cortical thickness and area with and without the confounders.

3.7. Complete Data Analyses

The analyses limited to the 65 participants with complete data (325 observations) revealed quite similar results as the main models (Figure 7). For cortical thickness, the estimates for the sex and WMSA interaction tended toward being lower, that is, more significant in the complete data set, while for cortical area and hippocampal volume, the differences were very small, but when present trending toward more significant.

4. Discussion

This study is the first to use longitudinal MRI data spanning 9 years with multiple measurements to assess relationships between WMSA volume, cortical thickness, and area, as well as hippocampal volume. As predicted, we found that increasing WMSA volume was associated with greater cortical thinning over time in older adults from the general population. Unexpectedly, a stronger relationship between WMSA growth and more widespread and greater cortical thinning was found in men than women, despite no sex differences in WMSA volume at baseline or progression over 9 years. For cortical area, a complex relationship with WMSA growth was revealed in response to the same WMSA volume and growth over time in men and women. Contrary to our prediction, increasing WMSA volume was not related to cortical area in women, but it was associated with greater cortical area over time in men. Indeed, the age‐related area reduction in the temporal, parietal, and cingulate lobes was almost completely offset by the increase in area associated with WMSA growth in men. Only cortical thickness and area in the occipital lobe were not related to WMSA growth in men. As expected, we did not find a direct relationship between changes in WMSA volume and hippocampal volume across 9 years. In the model with the three‐way interaction between WMSA, sex, and intervention group, though, both men and women allocated to HIIT exercise were shown to have greater hippocampal atrophy in the presence of greater WMSA increase over time. Thus, physiological effects of exercise mode in older adults with increasing cerebral SVD, as proxied by more WMSA, appeared to increase hippocampal volume loss over time. The sex‐specific, sensitivity and complete case analyses revealed similar results, suggesting robust relationships. Taken together, our study showed a strong relationship between changes in WMSA volume and cortical features in lobes with predominantly heteromodal and limbic association cortices, which differed by sex, with steeper relationships in men than women. In men, increasing WMSA was associated with both cortical thinning and cortical area increase in the same regions; a similar pattern was not observed in women. Thus, the cortex appeared to be differentially affected by WMSA in men and women.

4.1. WMSA Volume, Spatial Distribution, and Fazekas Score

In our study, both men and women had similar levels of cerebral SVD based on clinical MRI reading, and the Fazekas score correlated with WMSA volumes. Quantitatively, WMSA measures were also similar in both the sexes. Median baseline WMSA volumes were 3.5 cm³ for women and 3.9 cm³ for men (2.2‰ and 2.3‰ relative to ICV for women and men, respectively). These WMSA volumes were comparable to findings in other longitudinal cohort studies in middle‐aged to older adults using manual and semi‐automated segmentation of WHM (Landman et al. 2021; Prins et al. 2004; Melazzini, Vitali, et al. 2021; Dao et al. 2019). However, many studies also report larger WMH volumes based on semi‐automated and automated measurements in this age range (van den Heuvel et al. 2006; Stephen et al. 2021; Moon et al. 2018). In our cohort, men and women experienced a similar annual WMSA growth of 7.2% adjusted for ICV. This corresponded to an annual WMSA growth rate of 0.26 cm³, which was lower than the WMH growth of 0.55–1.00 cm³ (Prins et al. 2004; Dao et al. 2019; van den Heuvel et al. 2006; Stephen et al. 2021; Maillard et al. 2012), or 16%–45% (Venkatraman et al. 2020; Kellar et al. 2021) reported in previous intervention‐study controls and observational cohorts using manual, semi‐automated or automated volumetric WMH assessment. Larger WMH than WMSA volumes, and hence also greater WMH than WMSA growth over time, are as expected (Olsson et al. 2013; Dadar, Maranzano, et al. 2018; Melazzini, Vitali, et al. 2021; Melazzini, Mackay, et al. 2021). In summary, our results showed that WMSA volume, that is, the region with more severe WM pathology, expanded similarly in men and women with age.

4.2. WMSA Growth and Cortical Morphometry Over 9 Years

Cortical thickness declined with age in our cohort, in line with previous longitudinal (Pfefferbaum and Sullivan 2015; Thambisetty et al. 2010; Jiang et al. 2014) and cross‐sectional (Tuladhar et al. 2015; Seo et al. 2012; Tan et al. 2022; Dickie et al. 2016) reports of age‐related changes in the cortex. Increasing WMSA volume over time was associated with a steeper decline in cortical thickness in frontal, temporal, insula, and cingulate cortices in our main analysis. Although men and women had equal WMSA volumes at baseline and WMSA volume increased at the same rate over 9 years, a significantly steeper relationship was present between WMSA growth and cortical thinning in men in the temporal, cingulate, and insular lobes compared to women. The sex‐specific analysis revealed significant relationships between WMSA and cortical thinning in all lobes except the occipital cortex in men, with the greatest effects of WMSA observed in the insula and cingulate regions. In women, significant associations between WMSA and cortical thinning were only present in the insula and cingulate lobes, where the estimated effect of WMSA was approximately half of that in men. Previous cross‐sectional and longitudinal studies in hospital, patient, and healthy population cohorts with a maximum follow‐up time of 4 years using various techniques for WMH/WMSA segmentation, cortical thickness assessment, and statistical approaches report lower cortical thickness or increased thinning in the frontal (Tuladhar et al. 2015; Rizvi et al. 2021; Seo et al. 2012; Bernal et al. 2024; Tan et al. 2022; Jiménez‐Balado et al. 2024; Kim et al. 2020), temporal (Tuladhar et al. 2015; Rizvi et al. 2021, 2018; Seo et al. 2012; Bernal et al. 2024; Tan et al. 2022; Jiménez‐Balado et al. 2024; Kim et al. 2020), parietal (Tan et al. 2022; Jiménez‐Balado et al. 2024; Kim et al. 2020), insula (Bernal et al. 2024; Tan et al. 2022), and cingulate (Bernal et al. 2024; Yang et al. 2022) cortices. Our longitudinal data showed that over a longer period, the inverse relationship between WMSA growth and cortical thinning was most prominent in the insula and cingulate cortices in both sexes, regions not previously identified as particularly linked to WMSA. Furthermore, the relationship between WMSA growth and cortical thinning was widespread across all lobes and steeper in men than in women. No study has examined the interaction between sex and WMSA on cortical thinning or performed sex‐specific analysis of this relationship (sex has been controlled for in the analysis or analyzed as an independent predictor of cortical thickness). Studies of sex differences in cortical thinning with age predominantly report greater cortical thinning in men than women in cross‐sectional studies (Wierenga et al. 2022; Thambisetty et al. 2010), while a longitudinal study found similar cortical thinning in men and women after age 43 years (Díaz‐Caneja et al. 2021; Forde et al. 2020). Both types of studies show greater variability in brain measures in (older) men than women. Based on our results, the previously described greater variability of cortical thickness across age in men may result from sex‐specific differences in response to WMSA and cohort differences in WMSA presence and volumes.

There was no association between WMSA and occipital cortical thinning, even though WMSA became more widespread in the occipital lobe during the 9 years (see Figures 1, 3, and 5). Previous studies have shown that the optical radiation, as assessed with DTI, appears relatively spared compared to frontal WM in the presence of WMH (Vangberg et al. 2019; Chao et al. 2013), which might lead to differences in cortical afferent and efferent connectivity, in turn influencing neuronal morphology, and thereby cortical thickness. Concurring with our results, the above‐mentioned studies on WMH/WMSA volume and cortical thickness also did not report group differences or change in the occipital cortex volume or thickness in relation to WHM/WMSA. For the other lobes where associations between WMSA volume and cortical thickness were present, the estimates were smaller in the lobes that included association, sensory and/or motor cortices (i.e., parietal and frontal cortex), and largest in (para)limbic lobes, which encompass regions with allocortex (i.e., temporal, insula, and cingulate lobes). Given that cortical thinning in typical aging emerges from changes in neuronal morphology and not neuronal loss (Peters 2002; Freire‐Cobo et al. 2021), the most parsimonious explanation is that, on average, greater morphological changes due to growth of WMSA take place in heavily interconnected cortical neurons of association cortices, perhaps mainly in regions with allocortex. Association cortex contains neurons that are extensively connected locally, intra‐ and inter‐hemispherically, compared to, for instance, the occipital cortex, where primary and secondary sensory regions make up most of the cortex, and processing is localized and hierarchical. These cytoarchitectural differences might make some neuronal populations more vulnerable to WMSA. As far as we know, no study has assessed cortical histology by sex and age in humans or non‐human primates (Peters 2002; Freire‐Cobo et al. 2021). Nevertheless, differences in the synapses (Kniffin and Briand 2024; Liu et al. 2020), WM organization (Eikenes et al. 2023; Gur et al. 2002), and GM metabolism (Goyal et al. 2019) between the sexes are reported and suggest greater neuroplasticity in women, which might explain why fewer lobes underwent accelerated cortical thinning in the presence of WMSA in women compared to men.

Unexpectedly, there was a positive association between increasing WMSA and cortical area across all participants. As far as we know, only decreasing cortical area has been described in aging in earlier studies based on cross‐sectional and/or longitudinal data from participants spanning larger age ranges than in our study, and modeling area change across the lifespan using different statistical approaches (Sele et al. 2021; Nyberg et al. 2023; Yu 2024; Storsve et al. 2014; Walhovd et al. 2016). The increase in cortical area associated with WMSA over time was specific to men, in whom the areas of all lobes, except the occipital lobe, were positively associated with WMSA growth, echoing the findings for cortical thickness. In men, the positive relationship between WMSA and cortical area, and the negative relationship between WMSA and age almost counterbalanced each other, resulting in a small net increase in cortical area over 9 years. We cannot rule out the possibility that the area expansion associated with WMSA was an artifact, but it seems unlikely. First, an artifact should be equally likely in men and women, especially since men and women had similar WMSA volume and growth over the 9 years. Second, FreeSurfer's cortical area measure is considered more precise than thickness (Storsve et al. 2014; Liem et al. 2015), and the longitudinal FreeSurfer pipeline provides highly reliable area estimates (Iscan et al. 2015). Third, changes in the GM/WM contrast that potentially could bias cortical area estimates were similarly associated with the area of the different lobes in men and women. Fourth, it appears unlikely that edema and/or inflammation would affect only the horizontal and not the vertical dimension of the cortex. Hence, we speculate that the area expansion represents a compensatory mechanism reflecting changes in local processing and/or horizontal connectivity, in response to the accelerated cortical thinning associated with WMSA in the same lobes in men. Previous studies have reported greater variability of cortical area in men compared to women (Wierenga et al. 2022; Díaz‐Caneja et al. 2021; Forde et al. 2020). Our results provide a potential explanation for this finding, suggesting that the increased male variability in cortical area is associated with sex‐specific differences in the cortex related to WMH/WMSA in addition to aging. Based on our findings, this relationship may be more prominent when the area with more severe WM pathology, that is, WMSA, is used in the statistical models. A previous study examining variability in area across both men and women reported the largest deviations between the sexes in the cingulum, entorhinal, insula, anterior, and inferior regions of the frontal lobe, and anterior temporal lobe, while the lowest variability was present in pericalcarine, occipital, superior temporal, and post‐central gyri (Sele et al. 2021). That is, the greatest variability in area was found in heteromodal‐limbic association cortices, and the lowest in sensory regions, which correspond with the regions where cortical thickness and area were differently associated with WMSA volume over time. Surprisingly, in women, cortical area was not associated with WMSA, and age was only linked to lower occipital and cingulate lobe areas. Thus, notable differences between men and women with regard to relationships between age and WMSA on cortical morphometry over time appeared to be present. Our results demonstrated the value of longitudinal imaging in participants with a limited age range and multiple scans for the study of brain aging (Pfefferbaum et al. 2013) and clearly illustrated the importance of including age‐related brain changes such as WMSA to uncover differences in brain aging trajectories. Furthermore, our results imply that the inconsistent changes in cortical area across age and cohorts, and greater variability in cortical measures in certain regions and in men than in women in the published literature (Wierenga et al. 2022; Díaz‐Caneja et al. 2021; Forde et al. 2020), could be due to differences in sex distribution, presence, and volume of WMH/WMSA across studies.

As predicted, we did not find a direct association between WMSA growth and greater hippocampal atrophy in our longitudinal data set. This is in line with data from a longitudinal study using automated WMH and hippocampal segmentation in participants > 90 years of age with and without cognitive impairment, scanned up to 4 times across two years (Legdeur et al. 2019). Also, a study in ADNI participants with and without cognitive impairment and dementia reported no association between WMSA volume increase and hippocampal volume loss (Fiford et al. 2017). However, the latter study did report that a higher baseline WMSA volume was linked to greater hippocampal volume loss between two time points, mainly in controls. Some cross‐sectional studies suggest that the link between greater WMH/WMSA volume and smaller hippocampal volume is mediated by or directly linked to co‐occurring neurodegeneration, as reflected, for instance, in CSF Aβ42 levels (Freeze et al. 2017) and CNS amyloid beta pathology (Xhima et al. 2024). In our general population cohort, none of the participants were homozygous for APOE4, the major Alzheimer's disease risk allele. This might explain why WMSA was unrelated to hippocampal atrophy with increasing age in our study. However, when adding a triple interaction between WMSA, sex, and intervention group, greater hippocampal atrophy was present in the HIIT group. This is in line with our suggestion that HIIT exercise might be too strenuous for some older adults, potentially compromising their cerebral blood flow (Pani et al. 2021; Reitlo et al. 2026, 2023), in line with experimental findings in animals and humans (Calverley et al. 2020; Lucas et al. 2015; Vestergaard et al. 2020; Inoue et al. 2015; Shih et al. 2013). The pattern of vascularization and regulation of blood flow combined with the high metabolism of the hippocampus are suggested to make it particularly vulnerable to even modest reductions in perfusion (Johnson 2023). The finding that WMSA, a proxy for cerebral SVD, interacts with exercise mode on hippocampal volume over time has significant implications for guidelines related to physical activity and exercising in older adults.

The relationships between WMSA and cortical thickness, area, and hippocampal volume remained the same in the sensitivity models including the baseline confounders. It might be that any effect of these confounders on the processes relating to WMSA growth and morphological changes in the cortex started decades earlier than the participants' inclusion into our study between ages 70–77, as is known for dementia (Liu et al. 2020).

4.3. Strengths and Limitations

Our study had some notable strengths, including longitudinal, multiple scanning sessions on the same scanner with the same protocol over 9 years. A total of 74% of the cohort remained from baseline to 9 years, suggesting a representative longitudinal sample. Although the number of participants was modest, the power calculation indicated that we had sufficient numbers, which was also demonstrated in Results. Moreover, the longitudinal stream of FreeSurfer is shown to be highly reliable (Reuter et al. 2012; Hedges et al. 2022). The increase in cortical area was surprising, but the consistency of the sex difference across analyses points to a genuine effect and not an artifact. The robust longitudinal associations across main (with 436 observations), sex‐specific, complete case (in 65 participants scanned 5 times), and sensitivity analyses underscored the robustness of our results. Unfortunately, the number of participants did not allow us to perform, for instance, lag models to determine directionality between WMSA volume and cortical morphology over time. We used WMSA volume from T1‐weighted 3D instead of WMH from FLAIR. Thus, we studied relationships between volumes of more severely injured WM than in studies using WMH volumes. We may also have underestimated the true effect of WMSA on cortical thickness and area since participants who dropped out of the study had more signs of SVD at baseline, which may have biased the estimates toward healthier participants. Nevertheless, our consistent results suggested that WMSA captures biological processes relevant to GM.

Our results need to be replicated in other cohorts, and biological mechanisms and relationships with functional outcomes need to be established. If sex‐specific differences exist in cortical neuroplasticity in relation to cerebrovascular SVD, this has important implications for our understanding of brain aging.

5. Conclusion

Similar baseline volume and growth of WMSA across 9 years in overall healthy men and women from the general population were associated with greater cortical thinning in men compared to women, as well as increased area in the cortical regions with thinning in men, but not in women. Our results support sex differences in cortical plasticity in association, but not in primary sensory cortices, associated with WMSA growth, the most common age‐related finding and proxy for cerebral SVD in the aging brain.

Funding

The Generation 100 Brain study was supported by the Central Norway Regional Health Authority and MiDT National Research Center, supported by the Norwegian Department of Health and Social Services.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: Supplementary methods.

Table S1: Baseline characteristics of those lost‐to‐follow‐up compared to those that remained in the study over the 9‐year period.

Table S2: Change in proxies of cerebrovascular disease, WMSA volume and brain morphometry over the five time points stratified by sex.

Table S3: Test on significant differences in Fazekas score between men and women at each time point.

Table S4: Linear mixed‐effects estimates with sex and age at 1, 3, 5, and 9 years as predictors and log(WMSA) at the same time points as outcome.

Table S5: Linear mixed‐effects estimates of log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and cortical thickness per lobe at the same time points as outcome.

Table S6: Linear mixed‐effects estimates with log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and cortical area per lobe at the same time points as outcome.

Table S7: Linear mixed‐effects estimates for women, with log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and cortical thickness per lobe at the same time points as outcome.

Table S8: Linear mixed‐effects estimates for men, with log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and cortical thickness per lobe at the same time points as outcome.

Table S9: Linear mixed‐effects estimates for women, with log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and cortical area per lobe at the same time points as outcome.

Table S10: Linear mixed‐effects estimates for men, with log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and cortical area per lobe at the same time points as outcome.

Table S11: Linear mixed‐effects estimates for the main model, men and women separately and the model with a triple interaction between sex, intervention group and WMSA, with log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and hippocampus volume at the same time points as outcome.

Table S12: Linear mixed‐effects estimates of log(WMSA) volume at baseline, 1, 3, 5, and 9 years as main predictor and cortical thickness per lobe at the same time points as outcome, adjusted for confounders.

Table S13: Linear mixed‐effects estimates of log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and cortical area per lobe at the same time points as outcome, adjusted for confounders.

Table S14: Linear mixed‐effects estimates with log(WMSA) at baseline, 1, 3, 5, and 9 years as main predictor and hippocampus volume at the same time points as outcome, adjusted for confounders.

Table S15: Estimated %‐change in cortical thickness due to a one‐year change in age and WMSA for men and women.

Table S16: Estimated %‐change in cortical area due to a one‐year change in age and WMSA for men and women.

Figure S1: Bullseye illustration of the estimated regional growth of WMSA from linear mixed‐effects models. Red hue indicates regions with a significant association with age. The units are in percentage points WMSA load. Legend for bullseye sectors: F = frontal lobe, P = parietal lobe, T = temporal lobe, O = occipital lobe.

Figure S2: Change in total hippocampus volume over time. The thin lines are the change for each participant, while the thick lines are the linear trend for women in red and men in blue.

Figure S3: Association between Fazekas score and WMSA volume over the nine‐year period.

Figure S4: Change in GM/WM contrast ratio over time for the six lobes. The thin lines represent individual participants, and the thick lines are the linear trend lines. Women are plotted in red and males in blue.

Data Availability Statement

The study is still ongoing, and additional data collections will be performed. Because of privacy concerns and state regulations, the ethical and governance approvals for this study do not allow the MRI data, clinical, and cognitive data to be made available in a public repository. Data in this manuscript can be accessed by qualified investigators after ethical and scientific review (to ensure the data is being requested for valid scientific research) and must comply with the European Union General Data Protection Regulations (GDPR), Norwegian laws and regulations, and NTNU regulations. The completion of a material transfer agreement (MTA) signed by an institutional official will be required.