6-month aerobic walking training increases T1w/T2w signal in the white matter of healthy older adults

Participants

Participants were 247 community-dwelling older adults (average age of 65 yrs., 68% women) enrolled in a 24-week randomized controlled exercise trial that examined the effects of aerobic exercise on cognitive performance and brain health. The trial is registered with United States National Institutes of Health ClinicalTrials.gov (ID: NCT01472744). Individuals were eligible to participate if they met the following inclusion criteria: (a) 60–80 years-old; (b) able to read and speak English; (c) scored <10 on the geriatric depression scale (GDS-15); (d) scored ≥ 75% right-handedness on the Edinburgh Handedness Questionnaire; (e) demonstrated normal or corrected-to-normal vision of at least 20/40 and no color blindness; (f) low-active (i.e., participated in 30 or min of moderate physical activity fewer than 2 days per week over the past 6 months); (g) local to the study location for the duration of the program; (h) willing to be randomized to one of four interventions; (i) not involved in another physical activity program; and (j) scored >21 on the Telephone Interview of Cognitive Status questionnaire and >23 on the Mini Mental State Exam (Fong et al., 2009). Eligibility also included meeting inclusion criteria for completing a magnetic resonance imaging (MRI) assessment, consisting of: (a) free from neurological disorders; (b) no history of stroke, transient ischemic attach, head trauma or surgeries including the removal of brain tissue; and (c) no implanted devices or metallic bodies above the waist. Thus, our sample consisted of healthy, community-dwelling, typically low active older adults. For more information on participant recruitment and screening, see (Baniqued et al., 2018; Burzynska et al., 2017; Ehlers et al., 2016, 2017; Fanning et al., 2016; Voss et al., 2018). Participants underwent a series of MRI imaging, cognitive, and cardiorespiratory testing, before and after the 6-month intervention program.

The study was approved by and carried out in accordance with the recommendations of the Institutional Review Board at the University of Illinois at Urbana-Champaign with written informed consent from all participants. All participants provided written informed consent in accordance with the Declaration of Helsinki.

Intervention

After all baseline data were collected, participants were assigned to one of four interventions implemented over four waves from October 2011 to November 2014. Participants were randomized using a computer data management system and baseline-adaptive randomization scheme, taking into account equal distributions of age and gender (Begg & Iglewicz, 1980). Participants in all conditions attended three 1-h exercise sessions per week for 24 weeks (∼6 months)(Burzynska et al., 2017; Ehlers et al., 2017). The four intervention groups included: Strength/stretching/stability (SSS) involved exercises designed to improve flexibility, strength, and balance with the aid of yoga mats and blocks, chairs, and resistance bands, specifically designed for individuals 60 years of age and older. This intervention served as the active control group to account for the social engagement and novelty in the other interventions, with the difference that SSS was not aimed to increase CRF. The Walking intervention was designed to increase CRF. Thus, it involved walking sessions at 50–60% of maximal heart rate, as ascertained from a maximal graded exercise test. Walking duration increased from 20 to 40 min during the first 6 weeks of the program. During the remaining 18 weeks, participants walked for 40 min at 60–75% of their maximal heart rate each session. Frequent assessment of heart rate, using either palpation or Polar Heart Rate Monitors, and rating of perceived exertion ensured that participants’ exercise was performed at the prescribed intensity. Exercise logs were completed after each exercise session. The Walking + Nutrition group, in addition to the above walking intervention, received a nutritional supplement containing antioxidants, anti-inflammatories, vitamins, minerals, and beta alanine (Abbott Nutrition, Abbott Park, Illinois). Beta-alanine is thought to promote an increase in lean muscle mass (Zoeller et al., 2007), thereby enhancing the effect of increased CRF. However, the analyses of the primary outcomes indicated no differences in gain in CRF between the walking interventions (Baniqued et al., 2018; Ehlers et al., 2017; Voss et al., 2019) therefore, Walking and Walking + Nutrition were collapsed for the present analyses. The Dance intervention was designed to provide simultaneous cognitive and social enrichment combined with aerobic physical activity (PA). The choreographed dance combinations became progressively more challenging over the course of the 6-months program. Group social dance styles were selected (i.e., Contra and English Country dancing) to minimize lead-follow roles. In each session, participants learned ∼4 dances and recorded their heart rate and perceived exertion after each dance. Each participant learned and alternated between two roles for each dance, increasing the cognitive challenge.

Cardiovascular variables

Cardiorespiratory fitness (CRF) was assessed before and after the intervention on a motor-driven treadmill by employing a modified Balke protocol (graded exercise test). The protocol involved walking at a self-selected pace with incremental grades of 2-3% every 2 minutes. We continuously collected measurements of oxygen uptake, heart rate and blood pressure. We measured oxygen uptake (VO₂) from expired air samples taken at 30-second intervals until a peak VO₂ (the highest VO₂) was attained; test termination was determined by symptom limitation, volitional exhaustion, and/or attainment of VO₂ peak as established by the American College of Sports Medicine guidelines (American College of Sports Medicine, 2013).

MRI Acquisition

We acquired images on a 3T Siemens Trio Tim system with 45 mT/m gradients and 200 T/m/sec slew rates (Siemens, Erlangen, Germany). T1-weighted images were acquired using a 3D MPRAGE (TR = 1900 ms; TE = 2.32 ms; TI: 900 ms; matrix = 256 × 256; FOV = 230 mm; 192 slices; 0.9 × 0.9 × 0.9 mm³ voxels size; GRAPPA acceleration factor 2). The non-diffusion weighted images from the diffusion-weighted acquisition were used as T2-weighted images (b-value = 0 s/mm², TR = 5500 ms; TE = 98 ms, matrix = 128 × 128; 1.7 × 1.7 x 3 mm³ voxels size; GRAPPA acceleration factor 2), because the study protocol did not include a T2-W image scan besides FLAIR, which is suboptimal for the T1w/T2w calculation since it has a decreased grey-WM contrast due to the inversion pulse (Ganzetti et al., 2014). Out of 213 participants who completed the intervention, 180 had good quality MRI data at pre and post-intervention (see Supplementary Material 1 for participant flow for the current analyses). AMC and AZB independently checked for image quality. Images were excluded from the analyses if they had motion or ghost artifacts that affected the grey-WM boundary or image co-registration; 8 subjects were excluded due to brain anatomical concerns that affected image co-registration and could lead to partial volume effects (e.g., ventriculomegaly or asymmetrical ventricles); 8 subjects were excluded due to insufficient brain coverage of their T2-W images for intensity calibration with the MRTool.

Skeletonization and region selection

We used Tract-Based Spatial Statistics (TBSS) in FSL (Smith et al., 2006) to restrict the analyses to the center of WM tracks. This was to minimize the effects of possible partial volume due to individual and age differences in anatomy, to focus the analyses on the normal-appearing WM, and to allow direct comparison with our earlier DTI findings from this sample (Burzynska et al., 2017). We used the non-FA TBSS pipeline for the T1w/T2w images to project them onto the group WM skeleton with a threshold of FA > 0.2, as we described earlier (Burzynska et al., 2017). To confirm that the T1w/T2w voxels were correctly projected onto the WM skeleton, we de-projected all skeletonized T1w/T2w images for visual inspection in subject’s native space; the deprojection was accurate for all participants and regions except for regions 3 and 4 of the corpus callosum in 5 participants, which were treated as missing values.

We extracted T1w/T2w regional values for statistical analyses. Total WM was defined as all voxels on the WM skeleton. We examined the five subsections of the corpus callosum (CC) (Hofer & Frahm, 2006) given the anterior-to-posterior gradient of CC’s vulnerability to aging (Head et al., 2004). Other WM regions included the association fibers connecting regions known to be affected by aging: the fornix (FX), the superior longitudinal fasciculus (SLF), the external capsule (EC), the cingulum (CING), and the uncinate fasciculus (UNC). In addition, we included two other major WM landmarks: the forceps minor (fMIN) and forceps major (fMAJ), containing callosal fibers and thalamic projections to the frontal lobes and the occipital lobes, respectively. The corticospinal tract (CST) represented long projection fibers, typically more resilient to aging but possibly involved in the physical activity interventions. To define fMIN, fMAJ, UNC, SLF and CST on the WM skeleton, we used the tract probability maps from the John Hopkins University WM tractography atlas (Hua et al., 2008); http://cmrm.med.jhmi.edu). We thresholded the tract probability maps at 10-15%, depending on a tract, with the aim to maximize the overlap with WM skeleton but avoid including voxels from neighboring tracts (Figure 1). For the FX and EC, we used the John Hopkins University WM labels in FSL. Finally, since the prefrontal cortex is vulnerable to aging (Head et al., 2004) and its volume and function has been shown to benefit from greater CRF or aerobic exercise (Colcombe & Kramer, 2003; Voss et al., 2013), we defined prefrontal WM using a cutoff of y > 12 in MNI space (Burzynska et al., 2013). Then, we winsorized T1w/T2w values for each region at one percent of their distributions, separately for pre and post-intervention.

• Download figure
• Open in new tab

Figure 1

6-month change in T1w/T2w signal in the SSS, Walking and Dance groups

Note. The points represent the group means at both preintervention (PRE) and postintervention (POST) for each intervention group, and error bars represent 95% confidence intervals.

Finally, we used Shapiro-Wilk tests with visual inspection of the histograms and QQ plots to assess the normality and distribution of the T1w/T2w data. Frequency histograms of T1w/T2w in the whole sample showed a bimodal distribution in the following regions: forceps major, uncinate fasciculus, external capsule and corticospinal tract. This could have diluted the between-groups mean differences in these regions, leading to underestimation of the intervention effects and overestimation of the effects of time. Therefore, we removed these four regions from the main analyses and included the analyses in the Supplementary Material 2.

Cognitive assessment

Cognitive assessment included the Virginia Cognitive Aging (VCAP) battery (Salthouse, 2009, 2017) and two additional experimental executive function tasks (task switching and spatial working memory (Baniqued et al., 2018; Erickson et al., 2011). We have previously shown that the task switching and spatial working memory load on the reasoning construct of the VCAP battery (Baniqued et al., 2018; Voss et al., 2018). Thus, we grouped VCAP reasoning tasks (matrix reasoning, Shipley abstraction, letter sets, spatial relations, paper folding, and form boards) with the task switching and the spatial working memory task to create an executive function composite (Baniqued et al., 2018; Voss et al., 2018). In addition to the executive function composite, the VCAP assessed episodic memory (word recall, paired associates, logical memory tasks), perceptual speed (digit symbol substitution, letter comparison, pattern comparison), and vocabulary (Wechsler Adult Intelligence Vocabulary, picture vocabulary, and synonym/antonym). We used the vocabulary construct only for sample description, because there is no evidence linking physical activity interventions with gains in crystallized abilities.

We winsorized values for each task at one percent of their distributions, separately for pre and post-intervention. Then, we expressed both pre and post-intervention individual values as standardized scores (z-scores) using the mean and standard deviation of the pre-intervention distribution. Finally, we calculated composite scores for both pre- and post-intervention as mean z-scores of tasks within each cognitive domain (using the mean and standard deviation from the pre-assessment).

Statistical analyses

We combined the two Walking intervention groups because earlier results indicated no difference in CRF gains between the Walking and Walking + Nutrition groups (Voss et al., 2018). We used linear mixed-effects (LME) models with parameter estimates fitted using the R lme4 package (Bates et al., 2015) to compare change in T1w/T2w between the three groups (Walking, Dance, and SSS). Models included fixed effects of time, group, and the time-by-group interaction as well as random intercepts. The group factor was coded using Helmert contrasts. This allowed us to compare the SSS control against the average of all the Walking and Dance groups. Then, to contrast the effects of Walking vs SSS and Dance vs. SSS we fitted additional LME models using a contrast matrix with dummy codes for the three groups, such that SSS was the reference. We standardized all quantitative variables, but not factors, to create partially standardized regression coefficients. The standardization of our variables rendered regression coefficients (β) that are loosely interpreted like correlation coefficients in terms of effect size (Ferguson, 2009). We tested the assumptions of the LME models by visually inspecting the normality of residuals, as well as the distribution of the residuals vs. fitted values.

For correlational analyses, 6-month change scores in the variables of interest were calculated as the post-intervention z-score minus pre-intervention z-score (note that we used the pre mean and standard deviation to transform both pre and post data). We used partial Pearson’s correlations in R ppcor to study the associations between change in T1w/T2w and cognition (controlling for age, sex and education), and between change in T1w/T2w and CRF (controlling for age and sex) within each intervention group. Because these correlational analyses were exploratory, we corrected for multiple comparisons using the false discovery rate method as implemented by p.adjust (p.value, method=“fdr”) in R. Statistical significance was accepted at p<0.05 for two-tailed tests. For the size of our correlations and effects, we assumed 0.50 as a large-sized effect, 0.30 a medium-sized effect and 0.10 a small-sized effect (Cohen, 1992).

We created figures using the ggplot function in the ggplot2 package (Wickham, 2009) and the multiplot function within the coefplot package (Lander, 2016). All statistical analyses were completed using R version 4.0.1.

Intervention effects

Comparing the combined Walking and Dance groups vs. SSS, we found a significant time-by-intervention interaction in 6 out of the 11 WM regions, which included: total WM, the anterior and posterior subsections of the corpus callosum, forceps minor, cingulum and superior longitudinal fasciculus (Table 1). A comparison between Walking and SSS showed significant time-by-intervention interactions in the total WM, genu and splenium of the corpus callosum, forceps minor and the cingulum. Similarly, the models comparing Dance and SSS showed significant time-by-intervention interactions in the total WM and the genu of the corpus callosum, and at trend level in the cingulum. Using Helmert contrasts we compared the effect sizes quantifying the intervention effects between the Dance vs. Walking groups and we did not find significant differences between groups (Supplementary Material 4). Together, our results showed positive intervention-related changes in the T1w/T2w when compared to the SSS control (Figure 1), with larger effect sizes in the Walking group.

6-month longitudinal change in T1/T2 in the control group

As predicted, we observed a consistent pattern of decline or stability of the T1w/T2w signal over a period of 6 months in the SSS control group in the majority of WM regions. The largest effect sizes were observed in WM regions where we found significant time-by-group interactions (forceps minor and cingulum). Figure 1 shows the means for the T1w/T2w at preintervention and postintervention for each intervention group, while Figure 2 shows the standardized β-coefficients for all WM regions for the effects of time in the SSS control. Together, Figures 1 and 2 show a significant negative effect of time on the T1w/T2w in all WM regions except the genu of the corpus callosum and prefrontal WM, which showed no 6-month change.

• Download figure
• Open in new tab

Figure 2 Standardized β coefficients for the fixed effects of time in the SSS control

Note. Asterisks indicate †<0.10, *p<0.05, **p<0.01. Error bars represent 95% confidence intervals.

Change in T1w/T2w and cognition

Because the Walking group produced small to medium sized effects in five WM regions in the LME models, we correlated 6-month change in T1w/T2w in these regions with change in memory, perceptual speed and executive function, while controlling for age, sex, and education. We found that a positive change in T1w/T2w correlated with a positive change in episodic memory, in the genu of the corpus callosum and cingulum, only in the Walking group (Table 2). None of these effects were significant in the SSS group and Dance group after controlling for multiple tests. Lastly, we did not find significant associations between baseline T1w/T2w and baseline cognitive scores (Supplementary Material 5).

T1w/T2w and CRF

Because CRF was the primary physiological outcome of the intervention, we tested whether the intervention-related changes in T1w/T2w were associated with increases in CRF. Pearson partial correlations, controlling for age and sex, revealed no significant associations between change in T1w/T2w and CRF after controlling for multiple tests (Table 3).

Aerobic exercise training increased T1w/T2w in the adult WM

Our results confirmed our hypothesis that the aerobic exercise intervention (Walking and Dance training) resulted in an increase in the T1w/T2w signal, relative to an active control condition. Our findings are in alignment with the evidence supporting a positive relationship between aerobic exercise and brain structure and function (Colcombe et al., 2006; Erickson et al., 2011; Kleemeyer et al., 2016; Voss et al., 2010). However, previous randomized controlled trials using DTI reported no group effects of aerobic exercise (Burzynska et al., 2017; Clark et al., 2019; Voss et al., 2013). To the best of our knowledge, this is the first study to report exercise-induced changes in WM microstructure in healthy older adults.

Interestingly, the effects of the interventions on the T1w/T2w were region-specific. In the Walking group, the T1w/T2w showed positive changes in total WM, the genu and splenium of the corpus callosum, as well as the forceps minor and the cingulum. The majority of these WM regions contain late-myelinating association fibers that interconnect with the prefrontal cortex, with myelination extending into young adulthood (Knyazeva, 2013; Lebel et al., 2019; Pfefferbaum et al., 2000). Because WM regions that myelinate later in development deteriorate earlier with age (Bartzokis et al., 2010; Brickman et al., 2012), our findings suggest that regions vulnerable to aging retain some level of plasticity that can be induced by exercise interventions. Our results are comparable to correlational DTI studies in healthy older adults that showed that aerobic fitness was positively correlated with FA in the body and genu of the corpus callosum (Johnson et al., 2012), and the cingulum bundle (Marks et al., 2011). This study shows promising evidence of neural plasticity of structural connections within the vulnerable frontal and prefrontal regions.

Dance vs. Walking interventions

Although we observed no difference between change in T1w/T2w between the Walking and Dance group, we observed larger effect sizes in the Walking group for the time-by-intervention interactions. It is tempting to conclude that aerobic walking may be more effective in increasing WM’s T1w/T2w than dancing. However, because the Walking group (n=86) had a larger sample size than the Dance group (n =51), the lack of significant effects in some of the regions could be due to lower statistical power in the models comparing Dance vs. SSS. In addition, the Dance intervention included a large amount of low-intensity instructional time and less intensive aerobic activity. This is consistent with previously observed lower gains in CRF in Dance vs. Walking groups (Voss et al., 2018).

It is important to note that our previous DTI analyses showed that 6-month Dance intervention (but not Walking) resulted in an increase in FA in the fornix, which declined in the corntrol group (Burzynska et al., 2017). It is possible that the Dance intervention elicits changes in WM that are better measured with DTI in the fornix, which is a thin tract containing small-diameter fibers with many unmyelinated axons, and it is surrounded by the lateral ventricles. It is possible that the T1w/T2w cannot capture changes in the fornix because of the partial volume effects induced by the cerebrospinal fluid or because of the microstructural characteristics in this region.

Within-person changes in WM measured with T1w/T2w

T1w/T2w decreased over a 6-month period in the majority of WM regions in the SSS control, with the largest effect sizes observed in the forceps minor, cingulum and superior longitudinal fasciculus and the smallest effect sizes observed in the anterior corpus callosum. Our results using T1w/T2w are consistent with our earlier findings using DTI. Namely, in a similar set of participants we observed 6-month decline in FA in the in association fibers (e.g., cingulum), limbic system structures (e.g., fornix), and in commisural fibers (corpus callosum) (Burzynska et al., 2017). However, for other regions, our current results suggest a different pattern of age-related decline in the T1w/T2w signal. For instance, we did not find 6-month change in the genu of the corpus callosum, but reported a change in the body and splenium of the corpus callosum. Perhaps, FA is more sensitive to changes in regions with smaller axons that are coherently oriented and densely packed (e.g., genu) (Burzynska et al., 2010, 2017). On the other hand, the T1w/T2w may be better suited to detect longitudinal changes in regions characterized by larger axonal diameter, such as the splenium of the corpus callosum (Lamantia & Rakic, 1990) or by high crossing fibers such as the cingulum bundle or the superior longitudinal fasciculus (Glenn et al., 2016). In sum, our results suggest that the degree of decline in the T1w/T2w varies by region, so it is possible that these regional differences in susceptibility to aging reflect distinct mechanisms of WM aging. However, a systematic examination of this issue using more advanced WM techniques is required to confirm such a speculation.

Episodic Memory and T1w/T2w

In two out of the five regions with significant intervention effects, we found a positive association between change in T1w/T2w and change in episodic memory in the Walking group. One of these regions was the genu of the corpus callosum, which is known to mediate interhemispheric integration in episodic memory processes. Specifically, the genu of the corpus callosum has been shown to be implicated in the recruitment of the ventrolateral prefrontal cortex and to mediate the relationship between perceptual speed and episodic retrieval in healthy older adults (Bucur et al., 2008). We also observed significant associations in the total WM, possibly driven by the effects observed in all WM regions combined, and in the cingulum (this latter at a trend level). This is consistent with evidence showing that decreased FA in the dorsal cingulum has been associated with episodic memory impairment (Lockhart et al., 2012).

Given the effects of exercise on executive functions and processing speed (Colcombe & Kramer, 2003; Kramer & Colcombe, 2018) and the reliance of processing speed on WM integrity (Chopra et al., 2018), we were surprised to find no associations between change in T1w/T2w and change in these two cognitive domains. Although we observed performance decline trends in the control group for the executive function and processing speed components, these were not statistically significant. Similarly, in our previous work, we did not find significant time-by-group interactions showing training-related changes on the cognitive level (Voss et al., 2018). A systemic meta-analysis confirmed no beneficial effects of short-term aerobic training interventions on these cognitive domains (Young et al., 2015). Overall, our results suggest that the T1w/T2w is capable of detecting positive changes of WM in regions known to be involved in episodic memory processes. However, different modifiers can influence the relationship between aerobic exercise and cognition, such as physical activity intensity, intervention duration, diet, social engagement and genetic factors. Thus, future studies should test the associations between change in T1w/T2w and cognition over longer periods of time while testing for important individual differences that can moderate the effects of aerobic exercise on brain health.

Cardiorespiratory fitness and WM plasticity

We did not find an association between change in T1w/T2w and change in CRF. Even though numerous studies have reported positive associations between increased CRF and brain function and structure (Kramer & Colcombe, 2018), our results do not support our prediction that increased T1w/T2w signal would be associated with increased CRF, suggesting that the increased T1w/T2w in the aerobic exercise groups is mediated by other mechanisms other than CRF. However, it is important to note that increased CRF is associated with moderate-to-high intensity aerobic exercise, as corroborated by results from this randomized controlled trial showing that the Walking groups increased CRF (Voss et al., 2018). Aerobic exercise training is associated with physiological adaptations that result in increased oxygen-carrying capacity and extraction by tissues, which rapidly improves CRF in the first 5-11 weeks of training (Lundby et al., 2017). Subsequently, smaller improvements in CRF would be expected after months of training (3-12 months) (Erickson et al., 2011; Vidoni et al., 2015; Voss et al., 2018). Because our study sample was composed of healthy individuals, with relatively normal blood pressure values (Supplementary Material 3), it is possible that the associations between CRF and arterial health are stronger among individuals with hypertension or other chronic diseases (Haapala et al., 2020) and thus not observable in our healthy sample. Since vascular health is likely a key mechanism underlying the protective effects of aerobic exercise training on brain health, future studies are needed to further characterize individual differences that mediate or moderate the protective effects of aerobic exercise on the aging WM.

However, we found that increased T1w/T2w correlated with episodic memory in frontal and prefrontal WM regions only in the Walking group. This resembles findings by Ruscheweyh and colleagues, where they found that a 6-month aerobic exercise intervention correlated with increased gray matter volumes (mostly in the prefrontal and cingulate cortex) and episodic memory performance (word list learning), independently of CRF (Ruscheweyh et al., 2011). Our results suggest that any form of aerobic exercise is associated with positive changes in T1w/T2w, but only Walking is associated with better episodic memory functions.

T1w/T2w as an index of WM health

Because this is the first application using T1w/T2w to assess the effects of a clinical trial, the meaning of our findings need to be interpreted with caution. Although 3–4 weeks of aerobic exercise training was shown to stimulate myelination and increase axonal diameter in mice (Bobinski et al., 2011; Chen et al., 2019), relating changes in T1w/T2w to any particular microstructural mechanism, such as myelin content, is premature. Overall, evidence suggests that the T1w/T2w offers a promising measure of the functional aspects of WM microstructure (Uddin et al., 2019) that shows high concurrent validity after calibration (Arshad et al., 2017). Even though the T1w/T2w was first designed to map myelin content in the human neocortex (Glasser & van Essen, 2011), several biophysical processes in WM contribute to changes in T1w/T2w intensity (Deoni, 2010). The contrast of different tissues in T1-WI and T2-WI arise largely from distinct T1 and T2 relaxation properties of brain tissues (Sharma & Lagopoulos, 2010). However, T1 and T2 relaxations are not independent, because returning to equilibrium magnetization also results in a loss of magnetization from the transverse plane (Deoni, 2010). Indeed, previous studies using the T1w/T2w showed a strong correlation with myelin volume fraction in human WM (Saccenti et al., 2020), oligodendrocyte-specific gene expression in humans (Patel et al., 2020), and with cortical myelin as compared to histological myeloarchitectural studies in both humans and non-human primates (Glasser & van Essen, 2011). However, others have found that the T1w/T2w correlated with MRI estimates of axonal diameter (Arshad et al., 2017), axonal density (Fukutomi et al., 2018), and iron content in healthy adults (Shams et al., 2019). Overall, evidence suggests that the T1w/T2w offers a promising measure of the functional aspects of WM microstructure (Uddin et al., 2019).

Because sensitivity or specificity of the T1w/T2w to myelin or axons is not known, our results need to be replicated along with more specific myelin imaging methods, such as myelin water imaging or quantitative magnetization transfer (Lee et al., 2020; MacKay & Laule, 2016).

Limitations and future directions

Some methodological limitations warrant consideration. Because the simple division of the T1-WI by the T2-WI does not automatically cancel the different signal variations related to the different sensitivity profiles of the receiver coils of each image (Ganzetti et al., 2016), the variations in intensity of the MRI need to be corrected to obtain a standardized T1w/T2w contrast. Although the external calibration using non-brain tissues (i.e. eye and neck muscle) is considered the most robust (Ganzetti et al., 2016), we used the internal calibration proposed by Glasser and colleagues (Glasser & van Essen, 2011) due to insufficient brain coverage of our T2-W image to use the external hallmarks. The internal calibration rescales the image using the whole brain intensity distribution (Ganzetti et al., 2014; Glasser & van Essen, 2011). For this, we verified the resulting T1w/T2w pre and post calibration and our analysis yielded T1w/T2w histograms that had comparable intensity scales across subjects, in both the intervention and control groups. Still, our results need to be replicated using an external calibration approach.

Because we had insufficient brain coverage, we were not able to include other WM regions that have been associated with episodic memory, such as the WM tracts connecting the hippocampal formation (Burgess et al., 2002). Nevertheless, our study included subcortical connections of the hippocampus (e.g., fornix), and other direct projections (e.g., cingulate gyrus). Future research should include these relevant WM tracts to further characterize the relevant WM regions that can benefit from aerobic exercise interventions and can lead to increased episodic memory function.

Because the TBSS analysis used inherently focuses on normal appearing WM and the centers of the tracts, it may introduce bias by excluding some voxels affected by WM hyperintensities. We confirmed that the voxels sampled onto the WM skeleton from TBSS were derived from the same WM structures using the “deproject” function on FSL. Still, future studies should test different registration techniques to allow robust analyses in the entire WM.