Fatness but not Fitness Linked to BrainAge: Longitudinal Changes in Brain Aging during an Exercise Intervention

Abstract

Purpose:

Fitness, physical activity, body composition, and sleep have all been proposed to explain differences in brain health. We hypothesized that an exercise intervention would result in improved fitness and body composition and would be associated with improved structural brain health.

Methods:

In a randomized controlled trial we studied 485 older adults who engaged in an exercise intervention (n=225) or a non-exercise comparison condition (n=260). Using MRI, we estimated the physiological age of the brain (BrainAge) and derived a predicted age difference compared to chronological age (BrainPAD). Aerobic capacity, physical activity, sleep, and body composition were assessed and their impact on BrainPAD explored.

Results:

There were no significant difference between experimental groups for any variable at any timepoint. The intervention group gained fitness, improved body composition, and increased total sleep time but did not have significant changes in BrainPAD. Analyses of changes in BrainPAD independent of group assignment indicated significant associations with changes in body fat percentage (r(479)=0.154, p=0.001), and visceral adipose tissue (VAT) (r(478)=0.141, p=0.002), but not fitness (r(406)=−0.075, p=0.129), sleep (r(467) range −0.017 to 0.063, p range 0.171 to 0.710), or physical activity (r(471)=−0.035 p=0.444). With linear regression, changes in body fat percentage and VAT significantly predicted changes in BrainPAD (β=0.948, p= 0.003) with one kg change in VAT predicting 0.948 years change in BrainPAD.

Conclusions:

In cognitively normal older adults, exercise did not appear to impact BrainPAD, although it was effective in improving fitness and body composition. Changes in body composition, but not fitness, physical activity, or sleep impacted BrainPAD. These findings suggest that focus on weight control, particularly reduction of central obesity, could be an interventional target to promote healthier brains.

INTRODUCTION

Changes in brain structure are clearly associated with advancing age. These changes include reduced cortical thickness (1), volumetric decline of both grey and white matter (2, 3) and an increase in the absolute number and total volume of white matter hyperintensities (4). However, there is substantial individual variability in the prevalence of these declines (5), as well as the rate at which they progress when observed (3). Better understanding of the mechanisms underlying, and the behaviors that contribute to or inhibit, these changes could help inform interventions targeted at slowing brain structural and functional declines.

High levels of physical fitness, regular engagement in formal exercise, and high levels of overall physical activity have all been hypothesized to explain some of the observed individual differences in brain health. Indeed, several large observational studies have found associations between levels of physical activity and both brain volume and the risk of developing cognitive dysfunction (6). Further, systematic reviews of observational studies have observed that both higher levels of fitness and engagement in regular moderate to vigorous physical activity are often associated with higher volume of gray matter across key regions of the brain, including the hippocampus and prefrontal cortex as measured by magnetic resonance imaging (MRI) (7). Longitudinally, multiple studies have found associations between increased physical activity and changes in brain volume in key regions among both healthy individuals (6) and those with mild cognitive impairment (MCI) (8), Further, meta-analyses of interventional studies suggest that increasing the number of minutes of moderate to vigorous intensity physical activity (MVPA) improves cognitive function in older adults (9) particularly in individuals who are cognitively and physically healthy at baseline (10). In general, these observed improvements are further enhanced by the presence of multi-modal instruction that includes strength training (10). However, positive associations both cross-sectionally and longitudinally are not universally observed depending upon clinical status and the presence of comorbidities (2), and some studies have found no association between cardiorespiratory fitness and brain volume in healthy (11,12) and cognitively impaired (13) populations.

Independent of fitness and activity levels, body composition, particularly body fatness, has also been hypothesized to contribute to changes in brain volume and cognitive function. For example, higher BMI has been associated with decreased grey matter volume across multiple brain regions (14) and links between central adiposity, as measured by waist circumference, and executive function have been observed both in children (15) and older women (16). Recent systematic reviews of cross-sectional studies have indicated that obesity, particularly central obesity, is commonly correlated with reduced cortical thickness and gray matter volume (17) and with cognitive impairment in older adults (18). However, prospective longitudinal studies in both children (15) and mid-life adults (19) have observed a bidirectional predictive relationship between cognition and central obesity indicating that there may be a common causal pathway contributing to the development of both conditions. Kullmann et al (20) may have identified at least some portion of this shared pathway, noting that insulin sensitivity in the brain is strongly associated with volume of visceral fat, and insulin (in)sensitivity is associated with cognitive capacity.

Beginning in the early 2010’s, tools utilizing the capabilities of machine learning algorithms to provide a comprehensive assessment of structural changes within the brain have emerged. These algorithms are applied to MRI images to use volumetric measures of multiple brain regions drawn from large samples who range widely in age to provide an estimation of the physiological age of the brain (commonly termed BrainAge). The difference between this BrainAge and chronological age can be calculated to provide a Brain Predicted Age Difference (BrainPAD). Using this we can determine if an individual’s brain structure is younger (negative BrainPAD), or older (positive BrainPAD) than expected. These tools have successfully predicted age across the human lifespan, including in healthy adolescents (21) and older adults (22). Additionally, these algorithms show very good test/retest reliability (23) and have correctly identified larger BrainPAD (i.e., higher values/older brains) in populations with expected negative changes in brain structure and/or with evident cognitive decline including multiple sclerosis (24), stroke (25) and TBI (26). As such, BrainAge (and the associated BrainPAD) may offer meaningful public health research applications, although there has been little longitudinal research into predictors of BrainPAD or the likelihood of changes in BrainPAD in response to changes in modifiable behaviors.

Our research question(s) centered on how BrainPAD was affected by an exercise intervention and associated changes in fitness, fatness, activity, and sleep. We hypothesized that a six month moderately intense multi-modal exercise intervention focused on a combination of aerobic exercise, traditional resistance training, and functional movements would result in improved physical fitness and body composition (i.e., greater aerobic capacity, less body fat and visceral adipose tissue (VAT), and greater lean mass). We further hypothesized that this intervention would improve BrainPAD, and that those improvements would be associated with changes in the metrics of interest. As a secondary, but associated research question, we explored changes in fitness and/or body composition independent of their experimental grouping, with a hypothesis that beneficial changes over six months (i.e. greater fitness, less fat) would be associated with changes in BrainPAD indicating brains that are growing “younger” compared to chronological age. We had a final hypothesis that changes in sleep would be associated with changes in BrainPAD (with more sleep leading to a more negative/younger BrainPAD value), although likely minimally affected by the intervention.

METHODS

Participants

Data were drawn from a multicenter randomized interventional clinical trial approved by Institutional Review Boards at both the University of California, San Diego and Washington University in St. Louis, and informed consent to participate in the research study was obtained from all participants. This group has been described in depth elsewhere (27). In brief, participants were sedentary adults aged 65 to 84, not currently using glucocorticoid or diabetes medications, and without diagnosed cognitive impairment or neurodegenerative or cardiovascular disease.

Exercise Intervention

The exercise intervention was designed with the goal of integrating progressive aerobic and resistance training with functional movement and balance training and has been described in detail previously (28). In short, minimal heart rate targets during aerobic training were generated, personalized resistance training goals were established, and a comprehensive manual was developed to ensure consistency across sites and cohorts. All sessions were led by a trainer licensed by a nationally accredited organization (either The American College of Sports Medicine, The American Council on Exercise, or The National Academy of Sports Medicine) with extensive (>2 years) on-the-job experience working with older adults. In addition to demonstrated experience working with the target population, trainers went through a twelve hour training specific to the intervention during which the goals of the intervention, the exercise prescription and progression plans, and the specific exercises to be utilized were discussed and practiced in detail using team members and “friends and family” as example participants. Across the course of the intervention a total of five (three at one location and two at the other) trainers were engaged, with one “lead” trainer at each location leading ~60% of all classes at that intervention site. Classes were ninety (90) minutes and were held twice weekly for six months. Classes started with a thirty (30) minute warm-up period that included integrated movement designed to warm the body and raise the heart rate to a level at (or above) 55% of Heart Rate Reserve (HRR). Following the warm-up classes were split roughly in half with one group beginning aerobic exercise and the other beginning strength training. After thirty (30) minutes the groups switched training positions.

It is worth noting that the larger study was designed with a 2x2 factorial design in which approximately half of the individuals within the exercise intervention also received a mindfulness-based stress reduction (MBSR) intervention. Similarly, approximately half of the individuals within the control condition also received MBSR training, while the other half received a series of lectures aimed at health promotion. These lectures specifically avoided topics related to exercise and/or mindfulness. All individuals were included in these analyses without differentiation between those receiving or not receiving MBSR.

Neuroimaging Acquisition

Neuroimaging was gathered at baseline and following the six-month intervention period. All baseline scans were acquired >1 day, but <30 days prior to intervention initiation and +/− 2 weeks from intervention completion. For individuals involved in the exercise intervention, neuroimage scanning was completed on a non-exercise day (i.e., not on a day where formal training occurred) and participants were asked to report to the scanning location well rested. While all scans were acquired during standard operating hours (8A-5P) time of day and day of the week were not controlled.

3T MRI scanners were used to acquire high-resolution (1x1x1mm) T1-weighted sagittal, magnetization-prepared rapid gradient echo with 1 scanner used at one site (GE, Signa--MP-RAGE; repetition time (TR)=2300 ms; inversion time (TI)=900 ms, echo time (TE)=2.95 ms, flip angle=9°; Acquisition time=5 minutes)) and 2 used at another (Siemens Prisma and Tim Trio—MP-RAGE TR=2400 ms, TI= 1000 ms TE=3.16 ms, flip angle = 8°). Both used real-time motion correction (PROMO). Scans were processed using FreeSurfer (version 6.0) to provide quantitative measures of image quality. All images were reviewed for incidental findings or excessive head movement by the study neurologist(s).

BrainAge Processing of T1 Weighted MRI images

The BrainAge model developed by James Cole, commonly called BrainAgeR, was used for these analyses (22). This model was deemed to be the most appropriate available model as the algorithm training was done on a group which contained a comparatively large number of individuals over the age of 65. To derive the BrainAge score, T1-weighted MRI scans were segmented and normalized using SPM12. Vectors with mutually exclusive compartments for GM, WM and CSF were established using the Rnifti package in R. The Kernlab package was then applied to provide a BrainAge score using the 435 established input variables. To provide visual quality control beyond the point-of-acquisition review described below, multiple slices of the brain were provided as visual images in .html format using a FSL program. These images were reviewed for obvious object- or movement-artifact by a specially trained researcher. BrainPAD scores were calculated by subtracting chronological age from the BrainAge score provided by the algorithmic scoring. Positive values reflect brains that are older than chronologically expected, while negative scores indicate brains that are younger than the chronological age of the individual.

Assessment of BrainAge values and images for inclusion

Study neurologist’s recommendations regarding image usability were applied so that individuals who had uninterpretable findings were not analyzed. The Euler number, which is derived from the FreeSurfer algorithm and provides a quantified description of the number of holes in an image, was applied to further exclude individuals whose scans were poorly visualized and likely to be subject to error. Additionally, individuals who had a change in BrainPAD from baseline to the end of the intervention greater than three standard deviations from the group change calculated in absolute values were excluded on the presumption that one (or both) of their scans had features that led to inaccurate scoring.

Physical Measures (GXT, DXA, Accelerometery)

These physical measures have been presented in greater detail in Wing et al, 2022 (29) and Wetherell et al, 2020 (28). However, we have provided key elements of the physical measures below.

A Graded Exercise Test (GXT) to 85% of age predicted maximal heart rate (220-age--APMHR) was conducted on either a treadmill (Quinton QStress, CardiacScience, Chelmsford, Mass) or cycle ergometer (LODE Excalibur Sport, Netherlands) using two-minute stages that increased by 2.5% elevation (treadmill) or 0.33 W/kg (cycle) per stage and continued until the participant reached the pre-determined 85% value or the study physiologist ended the test based on physiological changes. Exercise capacity was calculated in metabolic equivalents of task (METs) using formulas published by the American College of Sports Medicine based upon speed and grade. METs were chosen as the metric of interest based on its common usage in clinical contexts, but changes in estimated oxygen uptake (VO₂) normalized for body weight at 85% of APMHR could be calculated by multiplying the METs value by 3.5 ml/kg/min.

Body composition was assessed by Dual X-Ray Absorptiometry (DXA) images gathered using a GE Lunar Prodigy densitometer at one site and an iDXA (GE/Lunar, Madison, WI) at the other. Both scanners utilized EnCore software (versions 14.1 and 16.1 respectively) for estimation of body composition. Values of body fat, lean tissue, bone, and visceral adipose tissue (VAT) were generated. Body fat percentage was derived by dividing the total body fat by the sum of fat, lean and bone components. Appendicular Lean Muscle Index (ALMI) was derived to control for differences in lean tissue attributable to differences in height. This variable was derived using the sum of the lean tissue (in kilograms) in the arms and legs divided by the participant’s height in meters squared.

A tri-axial accelerometer, the Actigraph GT9X+ Link (Actigraph Inc., Pensacola, FL) was used to objectively measure physical activity and sleep. Participants were asked to wear the device on their non-dominant wrist continuously for ten days except while bathing or swimming. This location and duration of wear are consistent with best practice as they result in a high degree of wear compliance and have been shown to capture sufficient wear time to be indicative of normal activity (30). After participant wear, devices were downloaded and screened for sufficient wear and potential device malfunction using commonly accepted methods (30) and algorithms (31) with acceleration data process into vector magnitude counts per minute (VM CPM). This metric incorporates intensity, frequency, and duration of movement and has been recognized as a reliable method to assess total volume of physical activity across 24-hour (or longer) periods of observation (39) as well as being able to distinguish between sleep and wakefulness (32).

Participants were asked to maintain sleep journals recording the time they tried to fall asleep and the time that they first woke during the period(s) of accelerometer wear. These time windows were analyzed on a minute-by-minute basis to determine sleep time using an algorithm designed for use in healthy adults (32). In addition to total sleep time, sleep efficiency and wake after sleep onset (both in terms of number of events and total time of events) were calculated.

Statistical analysis

Power calculations were conducted a priori to answer the primary research questions of the larger study that these data are drawn from. Specifically, based upon prior investigations completed by the primary investigators of the larger study, power calculations were completed to detect changes in performance based assessments of cognition, and hippocampal volume. Further analysis of power was not conducted for the specific outcomes analyzed here as these data were drawn from the available participant pool. Participants were excluded from any analysis for which they had missing values. SPSS version 27 was used to complete all statistical analysis. Descriptive statistics (percentages, means and standard deviations) were used to characterize demographic variables and identify potential outliers. Change scores were derived by subtracting baseline values from follow up values on an individual level.

Independent t-tests were conducted to assess differences across groups at baseline, and 2x2 (time x group) mixed measures ANOVAs were conducted to evaluate the effects of the intervention on BrainPAD, aerobic fitness, body composition, activity levels, and sleep. When there were significant effects for both the interaction and time, groups were split with the effect of time evaluated independently using paired sample t-tests.

Independent of intervention group, associations between changes in BrainPAD and changes in fitness, body composition, activity, and sleep were examined using Pearson’s correlation without controlling for any covariates. When associations between change scores were observed, univariate linear modelling was conducted with sex, site, and chronological age included as covariates. These were included based on the known systematic bias in BrainAge estimation toward younger appearing brains in older individuals, the possibility of systematic differences across sites, and the substantial differences in absolute values for body composition and fitness associated with sex. Years of education were also initially included as a covariate, but excluded when it did not contribute at all to model fit.

RESULTS

After excluding individuals without sufficient imaging (n=1 at baseline, and n=47 at follow up), those with suboptimal scans (n=8 at baseline and n=1 at follow up), individuals with BrainPAD changes >3 SD of absolute change (n=10; 5 positive and 5 negative), and those missing all comparator values for aerobic capacity, accelerometry, and body composition (n=1), a total of 485 participants were included. Due to partially missing data, an additional 77 participants were excluded from analysis of fitness, 4 from body composition, 12 from accelerometry based physical activity and 16 from accelerometer-based sleep. Overall, the sample was 72.6% female and showed some racial diversity (365 (75.3%) Non-Hispanic White, 54 (11.1%) Black, 33 (6.8%) Hispanic, and 23 (4.7%) Asian, with the remaining 10 (2.1%) claiming either more than one category or declining to answer). Descriptive data and results of 2x2 (intervention group x time) mixed measures ANOVA are detailed in Table 1.

Table 1:

Descriptives of Key Variables at Baseline, and 6 months

	Total Group		Exercise Group		Non Exercise Group
Variable	Baseline Mean (SD)	6 month Mean (SD)	Mean (SD)	6 month Mean (SD)	Mean (SD)	6 month Mean (SD)	p for time by group interaction
BrainAge (yrs)	69.2 (7.6)	70.0 (7.4)	69.1 (8.1)	69.8 (8.0)	69.4 (7.0)	70.1 (6.9)	0.959
n=	485	485	225	225	260	260
BrainPAD (yrs)	−2.0 (6.3)	−1.9 (6.2)	−2.5 (6.6)	−2.4 (6.6)	−1.6 (5.9)	−1.5 (5.9)	0.996
n=	485	485	225	225	260	260
Fitness (METS)	4.7 (1.5)	5.2 (1.5)	4.6 (1.4)	5.3 (1.4)	4.8 (1.6)	5.1 (1.5)	0.001
n=	471	414	216	192	255	222
Body Fat (%)	39.8 (7.6)	39.2 (7.8)	40.5 (7.0)	39.0 (7.5)	39.3 (8.0)	39.4 (8.1)	0.001
n=	485	481	225	224	260	257
Lean Tissue (kg)	43.7 (9.2)	43.8 (9.0)	43.0 (8.8)	43.6 (8.8)	44.2 (9.6)	43.9 (9.2)	0.001
n=	485	481	225	224	260	257
ALMI (kg/m2)	6.97 (1.24)	6.93 (1.35)	6.94 (1.18)	7.01 (1.26)	6.99 (1.30)	6.86 (1.43)	0.004
n=	484	481	224	224	260	257
VAT (kg)	1.32 (0.93)	1.28 (0.88)	1.32 (0.91)	1.24 (0.83)	1.33 (0.95)	1.32 (0.92)	0.001
n=	480	476	224	223	256	253
Sleep Efficiency (%)	84.2 (6.8)	84.2 (6.5)	84.1 (6.6)	84.1 (6.7)	84.3 (6.9)	84.3 (6.4)	0.806
n=	483	471	225	221	258	250
Time Asleep (min)	384.6 (56.3)	384.9 (58.8)	382.9 (59.9)	388.9 (60.6)	386.1 (53.1)	381.4 (57.0)	0.011
n=	483	471	225	221	258	250
WASO (min)	72.5 (33.2)	71.8 (30.4)	72.4 (31.4)	72.7 (31.0)	72.6 (34.8)	71.1 (29.8)	0.608
n=	483	471	225	221	258	250
VM (CPM)	1937 (506)	1952 (539)	1935 (484)	1964 (533)	1938 (525)	1941 (545)	0.385
n=	483	475	225	222	258	253

BL=Baseline; 6m= 6 months; yrs=Years; METS=Metabolic Equivilant of Task; %=percentage; kg=kilogram; kg/m2=kilogram per meter squared; min=minutes; WASO=Wake After Sleep Onset; VM=Vector Magnitude; CPM=Counts per minute;

There were no significant differences at baseline between those randomized to exercise versus non-exercise conditions for any variables (p range = 0.075 to 0.947), nor were there any cross-sectional group differences at 6 months (p= between 0.118 and 0.944). Variables that were the closest to significant at baseline were total body percent fat (p=0.075) with those in the intervention group having an average body fat percentage of 40.5% vs. 39.3%, and BrainPAD (p=0.119) with those in the exercise group having a BrainPAD of −2.5 years vs. −1.6 years in the non-exercise group.

Changes over time were significant and the degree of change also differed significantly between the exercise and non-exercise intervention groups for cardiovascular fitness (METS p=<0.001 for both time and group x time interaction), body fat percentage (p=<0.001 for both time and group x time interaction), total lean tissue (p=0.003 for time and <0.001 for group x time interaction), and VAT (p=0.002 for time and <0.001 for group x time interaction). Follow up tests indicated that the intervention group gained fitness and improved body composition by lowering body fat percentage and visceral adiposity and increasing lean tissue while the non-exercise group had a significant decrease in lean tissue, and non-significant changes in body fat and VAT. Somewhat unexpectedly, the non-exercise group also evidenced increased fitness, although not by as large a margin as the exercise group. There was also a significant effect of the intervention on total sleep time (group x time interaction, p=0.011), with small increases in the exercise group and non-significant decrease in the non-exercise group; the main effect of time was not significant (p=0.485). Specific results and confidence intervals of follow up tests are shown in Table 2.

Table 2:

Significant Interaction Effects From the Exercise Intervention.

Variable	Units	Exercise		No Exercise
		Estimate (CI)	p value	Estimate (CI)	p value
CRF	(METS)	0.695 (0.571 to 0.820)	<0.001	0.313 (0.189 to 0.437)	<0.001
Body Fat	(%)	−1.433 (−1.158 to −1.709)	0.001	−0.095 (−0.331 to 0.142)	0.432
Lean Tissue	(kg)	0.584 (0.418 to 0.751)	<0.001	−0.237 (−0.397 to −0.077)	0.004
ALMI	(kg/m2)	0.72 (0.004 to 0.140)	0.039	0.133 (−0.019 to −0.247)	0.022
VAT	(g)	−84 (−46 to −123)	0.001	4 (38 to −30)	0.803
TST	(min)	6.4 (0.7 to 12.1)	0.028	−3.7 (−8.9 to 1.6)	0.171

CRF=Cardiorespiratory Fitness; METS=Metabolic Equivalent of Task; %=percentage; kg=kilogram; kg/m2=kilogram per meter squared; min=minutes

Significant Results in bold

As would be expected with over an approximately six-month period, there was a significant effect of time for BrainAge (p=0.001) with BrainAge increasing 0.709 years on average (CI: 0.502 to 0.916). However, there was no significant group by time interaction (p=0.959). Additionally, there was no significant effect for time or group by time interaction for BrainPAD (p=0.345 for time and p=0.996 for interaction), nor for sleep efficiency (p=0.870 for time and 0.806 for interaction), number of minutes awake during sleep periods (WASO, p=0.096 for time and 0.608 for interaction), or overall daily physical activity (VM, p=0.503 for time and 0.385 for interaction).

As with many interventions, changes in the metrics of interest were not universal, and some individuals within the non-exercise group also experienced meaningful changes, particularly in fitness. With this in mind, we explored the correlations between changes in BrainPAD and changes in fitness, fatness, activity, and sleep without consideration of group. These analyses revealed that changes in BrainPAD were significantly associated with changes in body fat percentage (r(479)=0.154, p=0.001), and visceral adipose tissue (r(478)=0.141, p=0.002), but not fitness (r(406)=−0.075, p=0.129), metrics of sleep (r(467) range −0.017 to 0.063, p range 0.171 to 0.710) or physical activity (r(471)=−0.035 p=0.444).

When significant associations were explored independently (while controlling for chronological age at baseline, gender, and location) via linear regression, changes in both body fat percentage and VAT significantly predicted changes in BrainPAD (p=0.002 and 0.003 respectively), although when both are included in the model neither remains significant (p=0.054 and 0.089 respectively), likely due to a moderate amount of collinearity. The regression model including changes in visceral adipose tissue is included below as Table 3, which indicates that for each one kg change in VAT, there is a corresponding change of 0.948 years in BrainPAD when chronological age at baseline, sex, and location are controlled for.

Table 3:

Linear regression analysis of the association between changes in Visceral Adipose Tissue (VAT) and changes in Brain Predicted Age Difference (BPAD)

Model Summary	R	R2	Adjusted R2	SEE	P-value
	0.203	0.041	0.033	2.258	< 0.001
Predictors	Unstandardized β	SE	Standardized β coefficient	T	p-value
Constant	4.388	1.639		2.677	0.008
VAT mass a	0.948	0.316	0.137	2.996	0.003
Covariates
Chronological age at baseline	−0.054	0.022	−0.110	−2.424	0.016
Sex b	0.258	0.237	0.050	1.087	0.278
Site c	−0.468	0.208	−0.102	−2.246	0.025

Note: Values for VAT and BPAD are change scores based on the difference between values gathered at the 6 month visit minus values gathered at baseline.

^aVAT derived from Dual X-Ray Absorptiometry measured in kg

^bFemale=1, Male=2

^cUCSD=1, WUSTL=2

DISCUSSION

As expected, it appears that the multi-modal exercise intervention was successful in increasing the cardiovascular fitness of the participants, as well as improving body composition both by decreasing fat and increasing lean tissue. Further, the intervention appeared to have a small, but potentially meaningful effect on visceral fat, which is strongly negatively implicated in several chronic disease states common among older adults (33). However, this 6-month exercise intervention did not appear to have a meaningful impact on BrainPAD. Additionally, the increased fitness experienced in the non-exercise group suggests that there may have been larger factors at play in this population that encouraged a focus on fitness regardless of the intervention.

We have previously observed cross-sectional associations between visceral adiposity, but not fitness and/or physical activity, and BrainPAD (29) in this sample of older adults, and now we show that, when changes in metrics of fitness, fatness, and sleep and their relationship(s) to changes in BrainPAD are explored independent of group assignment, there is a clear association between increased/faster aging brains and increased fatness, particularly increased visceral adipose tissue. However, there was no association between changes in BrainPAD and changes in either fitness or overall physical activity. Given the recent evidence presented by Vidal-Pineiro et al. (34) suggesting that early life behaviors have a strong(er) influence on brain structure, and consequently, BrainPAD, with only minimal contributions from behaviors during middle and older adulthood, it is notable that we found links between BrainPAD change and fatness, but not fitness, change later in life in the context of an intervention study.

These data contrast with published evidence linking fitness and brain health, both in terms of the volume of various brain structures (6) and cognitive performance (10). A possible explanation for this may be that the relatively modest changes in fitness observed in this study (increase of ~0.5 METS at 85% of APHRM) were too small to elicit meaningful changes in brain structure (and thus, BrainPAD), and that an intervention that was either longer or more intense might elicit significant changes. Similarly, the combination of strength and aerobic training within the same intervention may have reduced the effectiveness of structural changes that have been observed with interventions more focused on aerobic training exclusively (6,8,9). However, it is worth noting that many of the studies that have found positive associations between fitness and brain structure have looked at individual segments of the brain (i.e., the hippocampus or frontal lobe exclusively); when the whole brain is examined, data have indicated variable levels of association and have generally had small effect sizes (8, 35 ). Further, the data here extend cross-sectional data showing no association between fitness and/or physical activity and BrainPAD in a nearly identical population (29). Given the large number of brain regions/features contributing to the BrainAge score it is possible that subtle changes to small regions within the brain are not sufficient to meaningfully impact the score. Thus, while BrainAge has proven itself potentially useful in a number of clinical populations (i.e. TBI, MS, etc.) to provide a relatively easily understood metric of brain health, it may not be sufficiently sensitive to be useful in evaluating changes that are expected to affect localized regions within the brain, particularly if those regions are ones that are not particularly age-related and thus contribute less to the prediction of BrainAge.

Interestingly, the observed relationship between body fatness and BrainPAD does offer some evidence to suggest that BrainAge (and the associated BrainPAD) may have utility in evaluating interventional changes in brain health, provided that those changes are occurring across a number of age-related brain regions. Indeed, the results from these analyses match recent research which has identified links between high levels of body fatness and reduced brain health (17, 18). Combined with recent scholarship indicating associations between central obesity and declines in whole brain structure (29) and cognitive function (36), these data offer additional evidence to suggest that VAT is particularly deleterious to health and has downstream effects across multiple systems.

Several mechanisms have been proposed to explain the detrimental impact of VAT. For instance, VAT has been linked to reduced immunity secondary to increased levels of inflammation (37)and to increased oxidative stress resulting from upregulated cytokine activation (38). However, potentially most compelling given the observed links between brain insulin resistance and decreased cognitive function and brain structure (39) is the fact that increased VAT contributes to decreased insulin sensitivity systemwide (40). While these data do not confirm a causal relationship between VAT and insulin (in)sensitivity in the brain, and there is a possibility of shared etiology that affects both independently, they do suggest the possibility of a causal pathway in which increased insulin resistance and visceral adiposity are linked both to each other and to structural brain health.

Although total sleep time was modestly affected by the intervention, neither changes in sleep time, nor sleep efficiency, were associated with changes in BrainPAD. Because this population was nearly exclusively comprised of normal sleepers, both in terms of time and efficiency, and the changes over the observation window were quite small, it is possible that there the modest changes in sleep time did not elicit structural changes, or that there was not sufficient variation within the group to detect meaningful differences in brain structure. Further, the lack of association is supported by evidence that suggests that it is only with an amount of sleep substantially above or below the recommended amount (for instance, <4 or >10 hours per night) that structural and functional decline is observed (41).

Given the many important positive health outcomes associated with increased fitness and larger volumes of physical activity (particularly moderate to vigorous physical activity) it is tempting to see it as a panacea that can promote good health across all organ systems, including the brain. Interestingly, the role of exercise in weight control is also often over-stated (42) further suggesting a desire for one mechanism of intervention to work to promote health in all areas. Unfortunately, likely even more than weight control, brain health is made up of many complementary and interconnected factors that are affected both independently and in coordination with each other. While these data do not preclude the possibility of positive adaptation in the brain with increased exercise, they do suggest that the physiological aging of the brain as a whole cannot be slowed/changed simply by increasing exercise levels by a moderate amount over a short period of time. However, these findings do contribute to a substantial literature that suggest that a focus on weight control, particularly reduction/prevention of central obesity, even in the short term, may be a useful target to promote healthier/younger brains, as well as benefiting other physiological systems.

Strengths of this study include the use of high-quality measurement tools in a large population of (presumptively) healthy older adults. Specifically, the use of whole brain MRI imaging for BrainAge calculation, graded exercise testing to estimate aerobic capacity, accelerometry to provide objective measures of physical activity and sleep, and DXA to estimate body composition and visceral adiposity mean that there are likely fewer sources of error compared to proxy measures, self-report, or epidemiologically derived estimation algorithms. However, there are some limitations which should be considered when applying these findings to intervention development or as a guide for future research. In particular is the possibility of a (n unmeasured) shared etiology that accounts for the observed changes in visceral adiposity, and BrainPAD. Further, it is possible that BrainAge, and consequently BrainPAD, is affected not just by the amount of VAT and body fatness, but also by the length of time those metrics are above “safe” levels. Additionally, because the BrainAgeR algorithm uses a large number of structural features drawn from multiple brain regions, it is possible that it is insensitive to isolated changes in areas that are less age-related and yet make meaningful contributions to cognition and/or function. Further, while six months is a reasonably long intervention period, it may be that it was not long enough with an intervention of the intensity used in this investigation to elicit changes in fitness sufficient to manifest as changes in brain structure. Similarly, a six-month window for observation, may not be long enough to see changes in BrainPAD that might be associated with small changes in fitness, fatness, or sleep that are maintained over time. Finally, the degree to which changes in BrainPAD are explained by changes in VAT is small, accounting for less than 4% of the total variance. However, given the potentially modifiable nature of VAT, and the possibility of benefiting multiple other systems through systematic reduction in VAT levels, further research is warranted. In particular, research that better elucidates the causal pathways linking VAT and brain health, or that identifies novel or particularly effective ways to reduce VAT, have the potential to lead to substantial public health benefit, likely including improved structural brain health.

CONCLUSIONS

Brain health as described by the difference between the biological age of the brain vs. the chronological age of the individual, appears to be modifiable with changes in body composition. Specifically, reducing body fatness in general, and visceral adiposity in particular, is associated with positive changes in BrainPAD consistent with brains growing younger compared to chronological age over time. However, changes in fitness levels, volume of physical activity, and sleep are not associated with changes in BrainPAD. This contributes to the body of evidence that suggests that body composition should be a primary target for behavioral interventions aimed at promoting brain health.