Association between cardiorespiratory fitness and cerebrovascular reactivity to a breath-hold stimulus in older adults: influence of aerobic exercise training

Abstract

Cerebrovascular reactivity (CVR) to a physiological stimulus is a commonly used surrogate of cerebrovascular health. Cross-sectional studies using blood oxygen level dependent (BOLD) neuroimaging demonstrated lower BOLD-CVR to hypercapnia among adults with high compared with lower cardiorespiratory fitness (CRF) in contrast to transcranial Doppler studies. However, whether BOLD-CVR changes following chronic aerobic exercise in older, cognitively intact adults is unclear. This study evaluated relations between BOLD-CVR with CRF (V̇o_2peak) using a cross-sectional and interventional study design. We hypothesized that 1) greater CRF would be associated with lower BOLD-CVR in older adults (n = 114; 65 ± 6.5 yr) with a wide range of CRF and 2) BOLD-CVR would be attenuated after exercise training in a subset (n = 33) randomized to 3-mo of moderate- or light-intensity cycling. CVR was quantified as the change in the BOLD signal in response to acute hypercapnia using a blocked breath-hold design from a region-of-interest analysis for cortical networks. In the cross-sectional analysis, there was a quadratic relation between V̇o_2peak (P = 0.03), but not linear (P = 0.87) and cortical BOLD-CVR. BOLD-CVR increased until a V̇o_2peak ∼28 mL/kg/min after which BOLD-CVR declined. The nonlinear trend was consistent across all networks (P = 0.04–0.07). In the intervention, both the active and light-intensity exercise groups improved CRF similarly (6% vs. 10.8%, P = 0.28). The percent change in CRF was positively associated with change in BOLD-CVR in the default mode network only. These data suggest that BOLD-CVR is nonlinearly associated with CRF and that in lower-fit adults default mode network may be most sensitive to CRF-related increases in BOLD-CVR.

NEW & NOTEWORTHY Earlier studies evaluating associations between cardiorespiratory fitness (CRF) and cerebrovascular reactivity (CVR) have demonstrated conflicting findings dependent on imaging modality or subject characteristics in individuals across a narrow range of CRF. This study demonstrates that CRF is nonlinearly associated with CVR measured by blood oxygen level dependent (BOLD) fMRI in a large sample of middle-aged and older adults across a wide range of CRF, suggesting that conflicting prior findings are related to the range of CRFs studied.

INTRODUCTION

Engagement in habitual physical activity and aerobic exercise is recommended as a prophylactic treatment to slow brain aging (1, 2) and reduce the risk of cognitive dysfunction (3, 4) in aging adults. Although the mechanisms underlying cognitive benefits with aerobic exercise remain unclear, attenuated age-related declines in cerebrovascular function are commonly implicated given the importance of the cerebrovascular endothelium in maintaining cerebral blood flow (CBF) regulation to the highly metabolic brain (5). In support of this idea, prior studies have demonstrated that lifelong masters athletes exhibit higher basal regional CBF and lower basal vascular impedance than healthy recreationally active older adults (6). Moreover, several studies have demonstrated exercise-induced improvements in basal CBF in regions sensitive to aging (7–11) in randomized control interventional studies in older adults, suggesting that aerobic exercise, even if initiated in middle-to-older age, may be an effective treatment to prevent declines in cerebrovascular function and CBF hemodynamics with aging.

Although prior studies have evaluated the effects of cardiorespiratory fitness (CRF) on basal CBF, the inability to augment CBF in response to physiological stimuli [i.e., cerebrovascular reactivity (CVR)] is associated with the development of Alzheimer’s disease (AD)-related pathology (12, 13), neurovascular uncoupling (14), and cognitive decline (15, 16). In parallel with the development of AD-related pathology, evidence also suggests brain regions and networks affected most by Alzheimer’s disease are the first to show CVR dysfunction, including the default mode network (17). Hypercapnia is commonly used as an experimental research technique to evaluate the functional vasomotor responsiveness of the cerebrovasculature as an index of cerebrovascular health. The cerebrovasculature is approximately twice as sensitive to hypercapnia (increased Pco₂) as compared with hypoxia or hypocapnia because CO₂ crosses the blood-brain barrier and dissociates to H⁺ ions, thereby reducing brain pH (5). Reductions in the local pH (e.g., more acidic pH) stimulate endothelium-dependent cell signaling mechanisms to reduce local cerebrovascular resistance (18) to augment CBF and restore local cerebral metabolic homeostasis. It has been hypothesized that if CVR represents an index of cerebrovascular health, individuals with higher CRF would exhibit greater CVR indicative of enhanced vascular endothelial responsiveness consistent with literature in the peripheral vasculature (19). However, prior studies investigating the role of CRF on CVR are mixed with some studies demonstrating positive (20–22), negative (23, 24), or no association between CRF and CVR (25) in humans. The reasons for the inconsistencies between these findings remain unclear but have been commonly explained by differences in subject characteristics, narrow CRF range, or neuroimaging methodologies [e.g., transcranial Doppler (TCD), functional magnetic resonance imaging (fMRI)] that may be further complicated by the use of cross-sectional study designs.

In a seminal study by Thomas et al. (23), lifelong masters athletes (average age: 75 yr; mean V̇o_2peak = 41 mL/kg/min) exhibited significantly lower CVR to hypercapnia compared with inactive older adults as measured by blood oxygen level dependent (BOLD-CVR) fMRI imaging. These fMRI data were replicated by Intzandt and colleagues (24) using both BOLD fMRI and pseudo-continuous arterial spin labeling (pCASL) in healthy, older, physically active (average age: 63 yr; V̇o_2peak = 29 mL/kg/min) adults, suggesting that the negative relation between CRF and BOLD-CVR is not explained by lifelong engagement in aerobic exercise alone. However, the cross-sectional study design and relatively high CRF of the participants (23, 24) precludes the ability to extrapolate these findings to other populations, including older adults across a wide range of low-to-high CRF or to evaluate changes in BOLD-CVR with regular and tailored aerobic exercise training.

Accordingly, the primary objective of the present study was to assess the relation between CRF, measured by V̇o_2peak and BOLD-CVR to hypercapnia in older adults without cardiovascular disease (CVD) or cognitive impairment across a wide range of CRF (0.1–97th percentile of CRF). We hypothesized that higher CRF would be associated with lower BOLD-CVR to hypercapnia. We aimed to quantify alterations in BOLD-CVR in previously inactive, lower-fit older adults following 12 wk of an aerobic exercise intervention. We hypothesized that BOLD-CVR could be attenuated after 12 wk of habitual aerobic exercise training in previously inactive older adults based on prior cross-sectional findings in the literature suggesting that greater exposure to habitual physical activity and higher CRF are related to lower BOLD-CVR. Evaluation of these two hypotheses would help to clarify inconsistent findings in the prior literature regarding the relations between CRF and BOLD-CVR as well as potential insight into physiological adaptations in the cerebrovasculature to aerobic exercise training.

METHODS

Participants

Healthy older adults between 50 and 79 yr were recruited from the greater Iowa City area through email and flyer advertisements. One hundred and fourteen participants were pooled from four parent studies with matching assessments of CRF and BOLD-CVR in which BOLD-CVR was not a primary outcome to evaluate cross-sectional associations between BOLD-CVR and CRF. In these studies, participants were recruited across all ranges of self-reported physical activity, resulting in a relatively wide range of CRF. Participants were free of overt or subclinical CVD as determined by a detailed health history questionnaire and maximal exercise test with 12-lead ECG. Additional exclusion criteria in common across all four studies included: 1) self-reported psychiatric or neurological disorders that may affect cerebrovascular or cognitive function; 2) prior history of brain injury or loss of consciousness for longer than 30 min; 3) self-reported diagnoses of disease or conditions associated with increased CVD risk or alterations in cardiovascular function including heart disease, chronic obstructive pulmonary disease, renal or liver disease, unregulated thyroid disorder (i.e., not on stable dose of medication for at least 3 mo before enrollment); 4) smoking or living with someone who smokes in the past 3 mo; and 5) self-reported regular use of steroid-based medication, psychotropics and/or chemotherapy. Female participants were postmenopausal and had not taken hormone therapy for at least 6 mo before enrolling in the study. Participants were also excluded for a failed cognitive screening using either the Mini-Mental State Examination (MMSE) (score <24 out of 30) or the Montreal Cognitive Assessment (MoCA) (score <20 out of 30) to ensure participants were free from moderate-to-severe cognitive impairment.

To evaluate changes in CVR following an aerobic exercise intervention, a subset of individuals (n = 33) underwent 12 wk of exercise training. Exclusion criteria in the interventional study were consistent with the cross-sectional study, with cognitive screening using MMSE and with the exception of exercise frequency before recruitment. Participants were excluded from the aerobic exercise intervention if they regularly engaged in moderate-to-vigorous physical activity (MVPA) or exercised four or more days per week (>30 min per session) in the prior 6 mo. Finally, all participants in both the cross-sectional and intervention studies provided written informed consent to all study protocols that were approved by the Institutional Review Board at the University of Iowa. All parent studies from which baseline data were pooled for cross-sectional analyses (NCT01775865, NCT02453178, NCT03114150) and the intervention (NCT02453178) were preregistered on clinicaltrials.gov before participant enrollment.

Assessments of CRF

Peak oxygen uptake (V̇o_2peak) was quantified as an absolute (absV̇o_2peak, L/min) and relative value (relative to bodyweight, relV̇o_2peak, mL/kg/min), as measured using a graded maximal exercise test with respiratory gases analysis (True One, Parvomedics, Inc., Sandy, UT) on a cycle ergometer (Lode, Medical Technology, Goningen, The Netherlands). All participants completed a standardized protocol including a brief, 1-min warm-up at low resistance (10 W) followed by 2-min stages of increasing resistance until maximal exertion. Participants were instructed to maintain between 60 and 100 revolutions per minute for the duration of the test. Oxygen consumption was averaged from expired air samples at 15-s intervals to calculate the volume of oxygen consumption (V̇o₂) until maximal effort was achieved or the test was terminated. Criterion for a “successful” V̇o_2peak test included achievement of three of the following four criteria: 1) respiratory exchange rate (RER) ≥1.10; 2) achievement of >90% of age-predicted heart rate maximum; 3) heart rate and/or oxygen uptake plateau despite an increase in resistance level; and 4) rating of perceived exertion greater than or equal to 18 on the 6–20 Borg scale. Tests were terminated if the participant showed signs of distress including abnormal blood pressure or ECG responses (e.g., evidence of myocardial ischemia or arrhythmia) to exercise.

Breath-Hold Stimulus

Hypercapnia was administered using a blocked, breath-hold (BH) stimulus constructed using E-Prime and visually administered in the scanner. A BH challenge is an easy, well-tolerated method for inducing hypercapnia in aging and populations with comorbidities (26). Participants were familiarized with MRI and to the BH task by a trained research assistant and practiced the task in a mock scanner on a separate day before their neuroimaging visit to ensure compliance on the day of MRI testing. Participants were refamiliarized with the BH task immediately before the MRI testing session. The use of paced breathing before initiation of the BH challenge and the use of a BH at the expiration has been demonstrated to increase reproducibility in the BH response. At the end of expiration, the lungs reach equilibrium at their functional reserve capacity which has been demonstrated to be less variable compared with self-directed inspiration and does not produce a biphasic BOLD response (27, 28). Accordingly, the task consisted of seven blocks (∼5 min total in duration) with each block containing three separate phases including 14 s of normal breathing, followed by 8 s of paced breathing with the words “breathe in” and “breathe out” displayed on the screen lasting ∼2 s each, and a 16-s end-expiratory BH prompted as “breathe out and hold.” During the BH stimulus, circles of decreasing size were visually presented as an indication of time left during the BH stimulus (Fig. 1A). Participants were cautioned against taking a large inhalation after the BH to minimize motion and instead instructed to begin breathing normally immediately at the conclusion of the BH stimulus to minimize head motion in the MRI scanner. Participants wore an MRI-compatible respiratory belt to ensure compliance during the scan. We also verified empirical support for reliability of BH-induced CVR in our protocol, based on intraclass correlation coefficients >0.90 for global and network-level measurements (Supplemental Fig. S2; all Supplemental material is available at https://doi.org/10.6084/m9.figshare.19314251.v1) (29). Strong interblock reliability supports our observations that participants were following instructions with consistent breath-holds throughout the task.

Figure 1.

A: schematic of the three phases of the breath-hold task paradigm. The sequence of events was shown slide by slide for times shown and repeated seven times. B: whole brain statistical maps showing the evoked blood oxygen level dependent (BOLD) response to the breath-hold relative to baseline comprising normal and paced breathing. Statistical maps are thresholded at z = 4.26, P < 0.05 cluster extent; the cross-hairs indicate the voxel of peak response in the posterior cingulate cortex; images are in radiological orientation so R = L. The colormaps illustrate the network masks from the Yeo/Schaefer 7 network parcellation, including the visual (red), somatomotor (blue), default (yellow), and salience/ventral attention (purple) networks, and their corresponding average percent signal change (C) reflecting cortical cerebrovascular reactivity (CVR) with each dot representing an individual in study 1 (n = 103).

Neuroimaging Protocol

MRI testing was conducted at the Magnetic Resonance Research Facility at the University of Iowa. Initial data collection (n = 47) occurred on a Siemens 3 T TIM Trio with a 12-channel head coil before the acquisition of a new MRI scanner in June 2016. Therefore, the remaining subjects (n = 67) underwent neuroimaging testing on 3 T General Electric (GE) Discovery MR750W MRI Scanner using a 32-channel head coil and scanner type was adjusted for as a covariate in statistical analyses. A reference T1-weighted anatomical brain image was collected [Siemens: echo time (TE) =3.09 ms, repetition time (TR) =2,530 ms, inversion time (TI) = 900 ms, flip angle = 10°, acquisition matrix = 256 × 256 × 240 mm, bandwidth = 219 Hz/pixel, voxel dimensions = 1.00 × 1.00 × 1.00, number of slices = 240; GE: TE = 3.376, TR = 8.588 ms, TI = 450 ms, flip angle = 12°, acquisition matrix = 256 × 256 × 240, FOV = 256 × 256 × 240, voxel dimensions = 1.00 × 1.00 × 1.00, number of slices = 240) for coregistration of all functional images to the participant’s individual anatomical T1 image.

BOLD imaging was used to measure CVR to the BH stimulus as previously described (30). Briefly, all functional images were acquired with a voxel size of 3.45 × 3.45 × 4 mm with ascending axial slice acquisition and no gap between slices (Siemens: TE = 30 ms, TR = 2,000 ms, flip angle = 80°, FOV = 220 × 220 mm, acquisition matrix = 64 × 64 mm, number of slices = 31, volumes = 163; GE: TE = 30 ms, TR = 2,000 ms, flip angle = 80°, FOV = 220 × 220, acquisition matrix = 64 × 64 mm, number of slices = 37).

Neuroimaging Analysis

Preprocessing was performed using fMRIPrep (31), a Nipype (32)-based tool. Each T1w (T1-weighted) volume was corrected for intensity nonuniformity using N4BiasFieldCorrection v2.1.0 (33) and skull-stripped using antsBrainExtraction.sh (v 2.1.0) using the OASIS template. Brain surfaces were reconstructed using recon-all from FreeSurfer v6.0.1 (34) and refined with a custom variation of the method to reconcile ANTs-derived and FreeSurfer-derived segmentation of the cortical gray matter (GM) from Mindboggle (35). Spatial normalization to the ICBM 152 Nonlinear Asymmetrical template version 2009c (36) was performed through nonlinear registration with the antsRegistration tool of ANTs v2.1.0 (37) using brain-extracted versions of both T1w volume and template. Brain tissue segmentation of cerebrospinal fluid (CSF), white matter (WM), and gray matter (GM) was performed on the brain-extracted T1w using the FAST tool available with the FSL software (38).

Functional data were motion-corrected using FSL’s MCFLIRT (39) and coregistered to the corresponding T1w using boundary-based registration (40) with nine degrees of freedom, using bbregister (FreeSurfer v6.0.1). Motion correcting transformations, BOLD-to-T1w transformation, and T1w-to-template (MNI) warp were concatenated and applied in a single step using (ANTs v2.1.0) antsApplyTransforms using Lanczos interpolation. ICA-based automatic removal of motion artifacts (AROMA) was used to nonaggressively remove motion-related artifacts from the data (41), which was subsequently used for estimating the BOLD response to the BH stimulus with a general linear model (GLM). The BOLD hemodynamic responses to the blocked BH task were visually inspected in each participant and time point to ensure adherence to the BH task.

Following preprocessing, in a subset of n = 35 representative participants, the global mean timeseries was extracted for every subject to derive a group-averaged timeseries. Following Murphy et al. (42), a 13-s delay was added to the 14-s block regressor for the BH period, which maximized the correlation between the block regressor and the global BOLD signal across all subjects (42). The GLM was then applied to all participants, estimated using FSL’s FEAT (43), with the high-pass filter set to 100 s. This resulted in β-coefficient brain maps representing BH-induced BOLD activation for each participant. FSL’s Featquery was used to convert β-coefficients to estimated percent BOLD signal change induced by the BH relative to breathing normally and paced breathing (Fig. 1). Because evidence suggests that aging affects CVR in some brain networks more than others, CVR was also examined within two associative networks sensitive to age-related cognitive decline (default mode, salience/ventral attention), and two sensory networks (somatomotor, visual), as defined by a data-driven cortical brain network parcellation (Fig. 1, B and C) (44–46). CVR from each network map was also averaged to create a composite CVR variable, expressing the relationship between CRF and average cortical CVR.

Aerobic Exercise Intervention

A subset of participants (n = 33) from the cross-sectional study participated in an aerobic exercise intervention that has been described in depth previously (see Supplemental Fig. S1 for the flow diagram of participant enrollment and completion) (30). Briefly, participants were randomized to 12 wk of either a moderate-to-vigorous (n = 22) or light (control group, n = 11) intensity exercise program using a 2:1 ratio of moderate-to-light exercise training. All participants completed three supervised training sessions per week performed on a cycle ergometer (Theracycle 200; Franklin, MA) with individuals exercising separately. Heart rate (HR; Polar HR Monitor, Model RCX5) and rating of perceived exercise (standard Borg scale from 6–20) was monitored every 5–10 min starting at minute 5 of all training sessions. Target HR for each training condition was calculated based on the maximal HR achieved during the baseline maximal aerobic exercise test.

Moderate-Intensity Training Group

Training sessions for the moderate-intensity group were standardized across the group and gradually increased in duration of time spent in the moderate-intensity HR zone (64%–75% of HRmax) because participants were inactive upon enrollment. Initially, training sessions consisted of a 5-min warm-up, followed by 20-min of moderate-intensity exercise and 20 min of passive cycle in which a motor in the bike moved the bike pedals for the participant, and 5-min of cool-down per session to achieve a total of 50 min per session. Five minutes of moderate-intensity exercise per session was added and 5 min of passive cycling was subtracted each successive week until the total time of moderate-intensity exercise reached 40 min per session (week 5). Participants unable to attend an exercise training session in the laboratory were provided instructions for completing their training session at home. Overall adherence to the 36 total exercise sessions was excellent (99.6% of sessions completed) with an average of 4.5% of sessions completed outside the laboratory.

Light Exercise Control Group

Older adults randomized to the “light” exercise control group reported to the laboratory for the same duration and frequency each week as the moderate-intensity exercise group. A passive cycling intervention was used as “light” exercise control group compared with the more commonly used “stretching” control group to separate the effects of exercise intensity from those of movement alone and to use a similar movement between the interventional and control arms. A motor in the stationary bicycle moved the bicycle pedals for the participant (30). However, to maintain interest in the passive exercise intervention, short bouts (approximately 1-min in duration) of moderate-intensity activity were permitted once every 10 min of passive exercise to accrue ∼12 min of moderate-intensity exercise per week. The short bouts of moderate-intensity cycling were designed to maintain interest in the study but to be less effective for substantially increasing CRF over the course of the 12-wk intervention.

Statistical Analysis

In the cross-sectional study, separate mixed models were used to quantify if baseline CRF (e.g., absV̇o_2peak, relV̇o_2peak,) was associated with average cortical and regional (e.g., visual cortex, salience, default mode, somatomotor cortex) BOLD percent signal change, and extent to which the relation was linear or curvilinear. Weight was included as a covariate for models including absV̇o_2peak. Covariates for all cross-sectional models included chronological age, scanner model (SE, GE), biological sex, and scanner head motion during the BH task quantified as average frame-wise displacement (47) calculated by MRIQC (48). In the interventional study, group differences in subject characteristics and regional BOLD-CVR were quantified using a Student’s t test at baseline. Repeated-measures analysis of variance models were used to test the effects of time (pre vs. post intervention) and intervention group (light control vs. moderate-intensity exercise) on variables of weight and CRF (49). Next, linear mixed models were used to test whether intervention-related changes in CRF were associated with changes in BOLD-CVR in the entire cohort. As reported before (30), both intervention groups improved their CRF and therefore, a percent change in CRF and BOLD-CVR was calculated and correlated as an exploratory analysis in the entire cohort to test the change in BOLD-CVR with CRF, regardless of intervention group. A two-tailed α level of 0.05 was used to indicate significance for all models. All data presented are the mean ± standard deviation.

RESULTS

Study 1: Cross-Sectional Study

CRF and BOLD-CVR data were available in 114 participants across the four parent studies. Eleven participants were excluded from analyses because they did not complete fMRI testing or had poor BOLD-CVR data (n = 8), BOLD-CVR values that met criterion for outliers (n = 2) or had incomplete CRF data (n = 1) resulting in a sample of 103 used in statistical analyses. Participant characteristics for study 1 (n = 114) are presented in Table 1. On average, participants were older (65 ± 6.5 yr), overweight [body mass index (BMI): 28.2 ± 5.3 kg/m²], and had “average” CRF (relV̇o_2peak: 24.3 ± 7.9 mL/kg/min) as defined by normative data published by the American College of Sports Medicine (50). Approximately, 57% were female and ∼20% reported chronic use of vascular-altering medications [e.g., antihypertensives (20%), statins (18%), nonsteroidal anti-inflammatory drugs (23%)].

Table 1.

Participant characteristics in cross-sectional study

		Study 2: Intervention sample
	Study 1: Cross-sectional	Pre	Post	Change (SE) [d]
n	114	33
Sex n (%), female	65 (57)	21 (64)
Age, yr	65 ± 6.5	67.3 ± 4.3
Education, yr	17.4 ± 2.9	17.5 ± 2.7
Weight, kg	81.9 ± 16.7	82.2 ± 18.6	77.8 ± 22.4	−4.72 (0.28) [−0.28]
BMI, kg/m2	28.2 ± 5.3	29.2 ± 5.4	27.7 ± 7.35	−1.48 (0.92) [−0.28]
Relative V̇o2peak, mL/kg/min	24.3 ± 7.9	19.8 ± 4.9	21.6 ± 5.3	1.38 (0.31)* [0.77]
Absolute V̇o2peak, L/min	1.97 ± 0.69	1.63 ± 0.58	1.74 ± 0.63	0.11 (0.03)* [0.69]
V̇o2peak percentile, ACSM	21.3 ± 27.3	6.24 ± 6.37	9.96 ± 11.20	3.7 (1.33)* [0.49]
Max Watts, W	159 ± 54.5	130 ± 40.2	138 ± 48.1	6.91 (4.93) [0.25]
CVR
Average cortical, %	0.65 ± 0.24	0.59 ± 0.29	0.61 ± 0.26	0.03 (0.05) [0.09]
Default, %	0.66 ± 0.24	0.59 ± 0.23	0.67 ± 0.30	0.08 (0.05) [0.27]
Salience, %	0.58 ± 0.24	0.53 ± 0.28	0.56 ± 0.26	0.03 (0.05) [0.12]
Somatomotor, %	0.61 ± 0.23	0.54 ± 0.24	0.57 ± 0.23	0.02 (0.04) [0.10]
Visual, %	0.75 ± 0.35	0.68 ± 0.51	0.66 ± 0.33	−0.02 (0.10) [-0.02]

All data are means ± SD. Education was collected as highest degree of obtainment in one study therefore years of education was only available in n = 100 participants. *P < 0.05 (two-tailed paired t test); change = post-pre [d = Cohen’s d for post-pre mean difference]. BMI, body mass index; CVR, cerebrovascular reactivity; V̇o_2peak, maximal volume of oxygen uptake.

Predictors of BH response.

Cortical BOLD-CVR was first averaged across all cortical regions sensitive to aging (Fig. 1B) to investigate cross-sectional associations with CRF, defined as both relative and absolute terms, in older adults with a wide range of V̇o_2peak values. Average cortical BOLD-CVR was associated with age when CRF was evaluated as a relative (relV̇o_2peak, β = −0.06 ± 0.02, t score = −2.57, P = 0.01) but not absolute (absV̇o_2peak, β = −0.04 ± 0.03, t score = −1.56, P = 0.12) value. However, average BOLD-CVR did not differ by sex in either model (P > 0.05) and did not differ between scanners (P > 0.05).

Relations between CRF with the average cortical BH response.

In the model with CRF defined as relV̇o_2peak, only age (β = −0.06 ± 0.2, t score = −2.57, P = 0.01) and relV̇o_2peak were significantly associated with the BOLD-CVR response. There was a significant quadratic (β = −0.04 ± 0.02, t score = −2.14, P = 0.03) but not linear (β = 0.01 ± 0.04, t score = 0.16, P = 0.87) relation between relV̇o_2peak and the average cortical BOLD-CVR (Fig. 2A, Table 2). Interestingly, the strength of the quadratic relation increased when CRF was defined as absV̇o_2peak term (absV̇o_2peak²: β = −0.05 ± 0.02, t score = −2.87, P = 0.01) and the association between BOLD-CVR and age was abolished (β = −0.04 ± 0.03, t score = −1.56, P = 0.12; Fig. 2A, Table 3). Adjustment for vascular altering medication use did not alter data interpretation (data not shown).

Figure 2.

Scatterplots displaying the relationship between cardiorespiratory fitness (CRF) and cerebrovascular reactivity (CVR) in cross-sectional (study 1 n = 103; assessed using mixed models adjusting for covariates) and intervention (study 2 n = 28; assessed using repeated measures ANOVA) data. A: cross-sectional data with a wide spectrum of CRF showed a quadratic relation between relative and absolute V̇o_2peak with cortical CVR. B: within the default network, the relationship was stronger for absolute CRF, with weight adjusted in the regression rather than as a ratio (see Tables 2 and 3). C: in lower-fit older adults, training-related change in CRF was observed for all participants (gray lines reflect individuals and error bars reflect SE); increased CRF was related to increased CVR specifically in default mode network (see Table 4). In all plots, shaded area reflects 95% confidence interval.

Table 2.

Model summary of multiple linear regressions assessing adjusted relations between global and regional CVR and relative V̇o_2peak in the cross-sectional study

	Average Cortical CVR	Visual Network CVR	Somatomotor Network CVR	Default Network CVR	Salience Network CVR
Intercept	0.67 (0.04)***	0.72 (0.04)***	0.60 (0.04)***	0.71 (0.06)***	0.64 (0.04)***
Scanner (GE = 0, SE = 1)	0.03 (0.06)	−0.04 (0.06)	0.03 (0.06)	0.10 (0.09)	0.04 (0.06)
Sex (M = 0, F = 1)	0.01 (0.06)	−0.03 (0.06)	0.02 (0.06)	0.09 (0.09)	−0.02 (0.06)
Head motion (FD)	0.05 (0.03)	0.04 (0.03)	0.06 (0.03)*	0.04 (0.04)	0.06 (0.03)*
Age, yr	−0.06 (0.02)*	−0.09 (0.02)***	−0.07 (0.02)**	−0.04 (0.04)	−0.06 (0.02)*
relV̇o2peak, mL/kg/min	0.01 (0.04)	−0.01 (0.04)	0.02 (0.04)	0.02 (0.06)	0.00 (0.04)
relV̇o2peak2, mL/kg/min	−0.04 (0.02)*	−0.04 (0.02)	−0.04 (0.02)*	−0.05 (0.03)	−0.03 (0.02)

Data are presented as β (SE) for continuous variables with predictors standardized to a common scale and outcome variable in raw units of percent signal change. Significance is indicated as *P < 0.05; **P < 0.01; ***P < 0.001. CVR, cerebrovascular reactivity; FD, frame-wise displacement; relV̇o_2peak, V̇o_2peak adjusted for bodyweight expressed as a linear function; relV̇o_2peak², V̇o_2peak adjusted for body weight expressed as a quadratic function. 

Table 3.

Model summary of multiple linear regressions assessing adjusted relations between global and regional CVR and absolute V̇o_2peak in the cross-sectional study

	Average Cortical CVR	Visual Network CVR	SomatoMotor Network CVR	Default Network CVR	Salience Network CVR
Intercept	0.64 (0.05)***	0.70 (0.05)***	0.57 (0.05)***	0.68 (0.07)***	0.63 (0.05)***
Scanner (GE = 0, SE = 1)	−0.03 (0.06)	−0.09 (0.06)	−0.03 (0.06)	0.03 (0.09)	−0.01 (0.06)
Sex (M = 0, F = 1)	0.12 (0.08)	0.09 (0.08)	0.13 (0.08)	0.21 (0.12)	0.04 (0.08)
Weight, kg	0.02 (0.03)	0.04 (0.03)	0.01 (0.03)	0.01 (0.05)	0.00 (0.03)
Head motion (FD)	0.05 (0.03)	0.04 (0.03)	0.06 (0.03)*	0.04 (0.04)	0.06 (0.03)*
Age, yr	−0.04 (0.03)	−0.06 (0.02)*	−0.04 (0.03)	−0.01 (0.04)	−0.05 (0.03)
absV̇o2peak, L/min	0.08 (0.05)	0.07 (0.05)	0.10 (0.05)	0.10 (0.08)	0.06 (0.05)
absV̇o2peak2, L/min	−0.05 (0.02)**	−0.06 (0.02)**	−0.05 (0.02)**	−0.06 (0.03)*	−0.04 (0.02)*

Data are presented as β (SE) for continuous variables with predictors standardized to a common scale and outcome variable in raw units of percent signal change. Significance is indicated as *P < 0.05; **P < 0.01; ***P < 0.001. CVR, cerebrovascular reactivity; FD, frame-wise displacement; absV̇o_2peak, absolute V̇o_2peak expressed as a linear function; absV̇o_2peak², absolute V̇o_2peak expressed as a quadratic function.

Relations between CRF with the regional BH response.

Secondary analyses were conducted to determine if the baseline relation between CRF and BOLD-CVR varied by individual regional networks. The quadratic relV̇o_2peak² term was associated with BOLD-CVR in all networks with a similar effect strength including the visual network (β = −0.04 ± 0.02, t score = −1.94, P = 0.05), default mode network (β = −0.05 ± 0.03, t score = −1.91, P = 0.06; Fig. 2B), somatomotor network (β = −0.04 ± 0.02, t score = −2.06, P value = 0.04), and salience network (β = −0.03 ± 0.02, t score = −1.86, P value = 0.07; Table 2). Consistent with the above findings, the relations were strengthened in most regions when CRF was calculated as an absV̇o_2peak²: visual (β = −0.06 ± 0.02, t score = −3.16, P < 0.001), default mode network (β = −0.06 ± 0.03, t score = −2.21, P = 0.03), somatomotor network (β = −0.05 ± 0.02, t score = −3.02, P < 0.001), and salience/ventral attention network (β = −0.04 ± 0.02, t score = −2.22, P = 0.03; Table 3, Supplemental Fig. S3). Neither relV̇o_2peak or absV̇o_2peak expressed as a linear term was associated with CVR in any region (P > 0.05 all models), suggesting the nonlinear association generalized to cortical networks as well. Adjustment for use of vascular altering medications did not change data interpretation (data not shown).

Study 2: Interventional Study

A subset (n = 33) of participants from the cross-sectional analyses participated in a 12-wk randomized aerobic exercise intervention (Table 1). Supplemental figures and tables are provided here: https://doi.org/10.6084/m9.figshare.19314251.v1). Five individuals were excluded from analyses due to poor preintervention data (n = 4) or an outlier score preintervention (n = 1) resulting in a sample of n = 28 included in analyses. Individuals randomized to light and moderate-intensity exercise treatments did not differ by any subject characteristics or by regional CVR at baseline (P < 0.05; Supplemental Table S1). Weight did not change with the intervention (time, P = 0.11) in either group (time × intervention, P = 0.55). Groups did not differ in CRF at baseline (P = 0.96), and both groups improved CRF (moderate vs. light: 6% vs. 10.8%, between groups t test on percent change, P = 0.28), defined by change in relV̇o_2peak, with the exercise intervention (time, P < 0.001). However, consistent with the t test, the change in CRF did not differ between groups (time × group, P = 0.18; see Supplemental materials for the full model results). Consistent with these results, absV̇o_2peak also increased after the intervention (moderate vs. light: 5.6% vs. 8.54%, between groups t test on percent change P = 0.51; time, P < 0.001) and did not differ between groups (time × group, P = 0.25). The change in BOLD-CVR over time did not differ by intervention group (P = 0.23). Notably, as shown in Fig. 2C, and Supplemental Figs. S4 and S5, because the participants in the intervention study were lower-fit before training, their training-induced gains in CRF were in the direction toward the peak of the inverted U-shaped relationships in the cross-sectional study. Based on these data, contrary to our initial prediction, we may expect their CVR to increase in relation to training-induced increases in CRF for these lower-fit individuals.

Relations between CRF with the average cortical BH response in the exercise intervention.

Because both groups improved in CRF and contained relatively modest sample sizes in each intervention group, a percent change value was calculated for all BOLD-CVR and CRF variables and entered to a linear regression that included scanner, sex, age, intervention group, change in weight, and change in absV̇o₂ (Table 4). AbsV̇o₂ was the focus of longitudinal analysis given its tighter relationship with BOLD-CVR in cross-sectional analyses. The mean percent change in absV̇o₂ for the subset of older adults who completed the intervention was 6.5%. Consistent with what would be expected from the cross-sectional study, results showed that, for this lower-fit population, increased CRF from training was related to increased BOLD-CVR in the default mode network and this was independent of changes in weight (Table 4, Fig. 2C).

Table 4.

Model summary of multiple linear regressions assessing adjusted relations between global and regional % change in CVR in the aerobic exercise intervention study

	Average Cortical CVR %Change	Visual Network CVR %change	SomatoMotor Network CVR %change	Default Network CVR %change	Salience Network CVR %change
Intercept	−6.23 (18.17)	−63.69 (32.43)	6.32 (22.83)	−3.06 (17.97)	−15.81 (17.80)
Scanner (GE = 0, SE = 1)	−9.32 (19.28)	−2.64 (34.40)	−4.14 (24.22)	−27.23 (19.07)	−1.11 (18.59)
Sex (M = 0, F = 1)	22.09 (19.27)	67.48 (34.39)	6.55 (24.21)	17.84 (19.06)	24.17 (18.77)
Age, yr	2.48 (9.22)	10.24 (16.46)	−1.44 (11.59)	5.19 (9.12)	−0.96 (9.17)
Intervention group (0 = light, 1 = moderate)	11.63 (17.76)	45.89 (31.69)	8.26 (22.31)	27.03 (17.56)	12.18 (17.83)
%Change weight, kg	−19.26 (9.90)	−35.97 (17.66)	−11.39 (12.43)	−17.28 (9.79)	−18.09 (9.72)
%Change absV̇o2peak	11.40 (8.37)	11.80 (14.93)	15.26 (10.51)	20.76 (8.28)*	9.74 (8.12)

Data are presented as β (SE) for continuous variables with predictors standardized to a common scale and outcome variable in raw units of percent change. Significance is indicated as *P < 0.05. CVR, cerebrovascular reactivity; %change = [(post-pre)/pre] × 100.

DISCUSSION

The primary objectives of the present study were to investigate relations between CRF and BOLD-CVR in response to a BH stimulus 1) in a cross-sectional study in older adults without CVD across a wide range of CRF (0.1 - 97th percentile using normative data adjusted for age and sex) and 2) after 12 wk of aerobic exercise training in healthy, previously inactive older adults. In the cross-sectional analysis, BOLD-CVR was nonlinearly associated with CRF at baseline such that BOLD-CVR increased with CRF until ∼28 mL/kg/min (relV̇o_2peak) or ∼ 3 L/min (absV̇o_2peak) before declining with increasing CRF in older adults. Associations between BOLD-CVR and absV̇o_2peak were stronger than with relV̇o_2peak. Regional analyses revealed that the nonlinear effect of CRF was relatively consistent across all cortical regions of the brain in the cross-sectional analysis. To evaluate the modulatory role of CRF on BOLD-CVR to the BH stimulus, previously inactive, healthy older adults were enrolled for 12 wk of moderate or primarily light intensity cycling in study 2. CRF was modeled as a percent change because both groups improved in CRF with the intervention. In contrast to our initial hypothesis, a greater percent change in absV̇o_2peak with the intervention was associated with increased BOLD-CVR in the default mode network, a cortical network of regions sensitive to aging, in previously inactive middle-aged and older adults. Collectively, these data demonstrate for the first time that BOLD-CVR to a hypercapnic BH stimulus is nonlinearly associated with CRF in middle-aged and older adults with a wide range of CRF and underscore the importance for consideration of CRF in studies evaluating cerebrovascular health.

Benefits associated with aerobic exercise on cognition are widely thought to be mediated in part through improvements in cerebrovascular function and neuroplastic mechanisms even in aged adults (51–53). Neurovascular coupling is the functional hyperemic response that couples neuronal metabolic demand with CBF supply. The cerebrovascular endothelium plays a critical role in neurovascular coupling by propagating local vasodilatory signaling intramurally from the neuron upstream to the pial arteries to reduce cerebrovascular resistance and augment CBF to local metabolically active neurovascular units (5), underscoring the importance for maintaining functional responsiveness of the cerebrovascular endothelium to vasodilatory stimuli in aging. Prior cross-sectional studies in adults with narrower ranges of CRF demonstrate inverse and positive associations in older adults with higher and lower CRF, respectively, contributing to the relative heterogeneity in prior literature (20, 23, 24). However, the cross-sectional findings in the present study demonstrate an inverse nonlinear association between CRF and BOLD-CVR with a positive association until a relative CRF of 28 mL/kg/min before an inverse association was observed with increasing CRF. The inverse nonlinear association suggests that the relative heterogeneity in prior study findings may be dependent on differences in participant CRF between studies. In support of this, prior studies have demonstrated that older adults with relV̇o_2peak approximately <26 mL/kg/min (20) exhibit a positive association between CRF and CVR, whereas older adults with a relV̇o_2max ∼>29 mL/kg/min exhibited a negative association (23, 24), consistent with the cross-sectional findings of the present study. Moreover, CRF measured by absV̇o_2peak and adjusted for bodyweight was a stronger predictor of CVR compared with relV̇o_2peak in the present study_, consistent with a recent study evaluating relations between CRF and brain volumes that demonstrates absV̇o_2peak may be a more sensitive measure of CRF compared with relV̇o_2peak for brain health (54).

In a previous study, CRF was negatively associated with slope of the hemodynamic BOLD response to hypercapnia in healthy fit older adults (>50th percentile of CRF) (24). Moreover, there was a qualitative dose-dependent reduction in the BOLD hemodynamic response with increasing quintile of CRF, measured by V̇o_2peak (24), collectively supporting the hypothesis that the attenuated response in higher fit older adults is related to aerobic exercise-related adaptation. However, in the present study, older adults with very low CRF enrolled in the intervention demonstrated similar magnitudes of improvements in CRF regardless of randomization group. The percent change in CRF was positively associated with increased BOLD-CVR in the default network only, consistent with other studies demonstrating that these regions are sensitive to the beneficial effects of aerobic exercise (55).

Although mechanisms underlying the varying effects of CRF on CVR remain unclear, interventional data from the present investigation and others (23, 24) suggest that physiological adaptations to habitual aerobic exercise on the cerebrovasculature or inherent differences in neuroimaging modalities and the BOLD signal may be involved (56, 57) Thomas et al. (23) hypothesized that attenuations in CVR to hypercapnia were related to reductions in central chemoreceptor sensitivity associated with habitual sub-lactate threshold exercise training in highly fit master’s athletes. Exercise below lactate threshold augments cerebral spinal fluid levels of CO₂ that are not buffered by respiratory compensation and may over time reduce central chemoreceptor sensitivity. However, central chemoreceptor sensitivity was not altered following 12 wk of aerobic exercise training (20) or in deep-sea breath-hold divers who experience significant repetitive increases in Pco₂ (58).

Alternatively, recent data demonstrate that age-related changes in CVR to hypercapnia may be maintained in older adults by an augmented blood pressure response compared with young adults (59). Chronic aerobic exercise training reduces blood pressure reactivity to physiological stimuli (60) and may reduce or delay the onset of cerebral autoregulation (61). Therefore, aerobic exercise training may also reduce the important compensatory blood pressure response driving cerebral perfusion pressure and CVR in older adults that may not be immediately buffered by cerebral autoregulation, contributing to a negative association observed in individuals with CRF >28 mL/kg/min in the cross-sectional study. In support of this, recent data demonstrate that 12 mo of aerobic exercise training reduces CVR (via TCD) and blood pressure reactivity to hypercapnia induced by bag rebreathing in older adults with mild cognitive impairment (62). However, we were unable to measure if blood pressure was significantly altered in response to the acute BH stimulus or reduced following our shorter 12-wk intervention because of the lack of specialized MRI-compatible blood pressure equipment.

Finally, differences in the cerebral vascular bed interrogated by BOLD fMRI versus TCD cannot be excluded as a possible explanation for differences between studies and neuroimaging methodologies. Cerebrovascular signaling mechanisms mediating hypercapnia-induced vasodilation differ between cerebral vascular beds with local vasoactive factors (e.g., nitric oxide, adenosine, prostaglandins) (5) and shear-mediated (63) vasodilation mediating alterations in the microvasculature and large cerebral arteries (e.g., middle cerebral artery), respectively. Moreover, TCD measures cerebral blood velocity in a large cerebral artery (commonly the middle cerebral artery) under the assumption that vessel diameter is not altered with the CO₂ stimulus. Increasing data support that hypercapnia induces large cerebral artery vasodilation, therefore, limiting the ability to interpret changes in cerebral blood velocity (64–66). In contrast, the fMRI BOLD signal provides an indirect measure of CBF by measuring the change in the ratio of oxygenated-to-deoxygenated hemoglobin in the cerebral venous/venules circulation under the assumption that hypercapnia does not alter neuronal activity or cerebral metabolic rate of oxygen consumption (56). BOLD fMRI is sensitive to changes in the concentration of deoxyhemoglobin (i.e., BOLD hemodynamic response) that are relatively greater on the venous side compared with the arterial side of the cerebral circulation and fluctuates in response to a physiological or cognitive stimulus. Therefore, although BOLD fMRI is favored for its high signal-to-noise ratio compared with other CBF MRI techniques (e.g., arterial spin labeling), it is important to recognize that BOLD fMRI is an indirect measure of CBF measured in the venules, in contrast to TCD.

Recent data comparing CVR to 5% CO₂ in young and older adults using BOLD fMRI and TCD demonstrate no association in CVR between methods and showed differential effects of age (56). Therefore, it is possible that differences between TCD and BOLD fMRI studies represent physiological differences in the two vascular beds and their response to CVR and/or aerobic exercise indicating extreme caution should be applied to comparing findings with CVR across different neuroimaging modalities. Taken together, our findings add to the increasing data assessing relations of CVR and CRF using different neuroimaging modalities and the effects of physiological adaptations to moderate-intensity exercise on CVR.

The results of this study should be interpreted within the context of the following limitations. A BH challenge is safe, low-cost, and well-tolerated for inducing hypercapnia that can be easily applied in an MRI setting with less participant discomfort and does not require specialized equipment or set-up. However, we acknowledge that the use of a BH stimulus is not without limitations when evaluating physiological mechanisms by which CRF may modify CVR. Previous studies have validated the use of a BH stimulus to be consistent with fixed concentration of CO₂ (67, 68). Prior studies have demonstrated that reproducibility of CVR is increased when adjusted for $\text{P}\text{E}{\text{T}}_{\text{C}{\text{O}}_2}$ values or expressed as percent change, consistent with the present study (26, 67). Nevertheless, we did not measure the change in end-tidal CO₂ ($\text{P}\text{E}{\text{T}}_{\text{C}{\text{O}}_2}$) to the BH stimulus and therefore, are not able to calculate the change in BOLD signal relative to change in $\text{P}\text{E}{\text{T}}_{\text{C}{\text{O}}_2}$, consistent with other studies (20, 23). This precludes our ability to evaluate if $\text{P}\text{E}{\text{T}}_{\text{C}{\text{O}}_2}$ achieved with the BH stimulus differed by CRF or following the intervention. Additionally, the use of a BH stimulus may reduce $\text{P}\text{E}{\text{T}}_{{\text{O}}_2}$ depending on the duration of the BH stimulus, potentially suppressing the BOLD signal (67, 69). However, the present study used a 16-s BH stimulus that is shorter compared with these studies, potentially reducing the risk of signal suppression associated with reductions in $\text{P}\text{E}{\text{T}}_{{\text{O}}_2}$.

Second, light intensity exercise in which a motor in the cycle ergometer moved participant’s legs except for the brief bouts of moderate-intensity exercise (totaling 12 min/wk) to maintain participant interest was used to control for the effects of movement alone, making it a theoretically more suitable control group compared with the commonly used stretch “control” group (70). We acknowledge that participants enrolled in this light intensity exercise arm of the intervention improved their CRF to a similar degree as participants randomized to moderate-intensity aerobic exercise, despite only engaging in 12-min of moderate (i.e., 1 min bouts of moderate-intensity every 10-min of passive cycling) exercise per week and not changing their physical activity levels outside the laboratory measured by accelerometry (30). These data suggest that small bouts of moderate intensity embedded into larger volumes of light-intensity exercise to maintain participant interest may not be a suitable control group in previously inactive older adults with very low CRF for physiological research studies, but provides support that changes in CRF associated with moderate-intensity aerobic exercise has effects on the cerebrovasculature. In addition, participants in the present study were predominately a non-Hispanic white population thereby potentially limiting the generalizability of the study results to other ethnic/racial groups. Further, our sample size for the intervention study was modest, not powered for testing how the training intervention affected CVR, and only lower-fit participants trained to improve CRF. Future studies should extend these results with well-powered evaluations of the dose-response effect of aerobic exercise intensity, evaluation of the duration on changes in CVR, and changes in CVR in individuals with very low, moderate and higher CRF to help provide insight into the observed inflection point at ∼28 mL/kg/min (relV̇o_2peak) or ∼ 3 L/min (absV̇o_2peak) in the present study. However, within the realm of exploratory analyses, our results provide the foundation for larger trials with more diverse participants to replicate and extend results presented here.

In summary, these data suggest that relations between CRF and CVR to a BH stimulus vary by CRF in older adults across a wide range of CRF. Future investigations are needed to elucidate the mechanism by which greater CRF, achieved through aerobic exercise training, modulates BOLD-CVR to a hypercapnic BH stimulus in older adults.

SUPPLEMENTAL DATA

Supplemental Table S1 and Supplemental Figs. S1–S5: https://doi.org/10.6084/m9.figshare.19314251.v1

GRANTS

This study was supported by the grants provided by National Institutes of Health: 5R01 AG055500-04 (to M.W.V.), 1R21 AG043722 (to M.W.V.), 1R21-AG-043722 (to G.L.P.), F32 AG071273 (to L.E.D.), T32AG000279 (to L.E.D.), KL2 RR024980-05 (to G.L.P.), and U54 TR001013. Neuroimaging was conducted using equipment funded by the National Institutes of Health 1S10OD025025-01.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

G.L.P. and M.W.V. conceived and designed research; L.E.D.,T.B.W., C.W., L.R., C.H., K.D., and A.L.-C. performed experiments; L.E.D., T.B.W., A.O., and M.W.V. analyzed data; L.E.D. and M.W.V. interpreted results of experiments; L.E.D. and M.W.V. prepared figures; L.E.D. drafted manuscript; L.E.D., T.B.W., G.L.P., C.W., L.R., C.H., A.O., K.D., A.L.-C., and M.W.V. edited and revised manuscript; L.E.D., T.B.W., G.L.P., C.W., L.R., C.H., A.O., K.D., A.L.-C., and M.W.V. approved final version of manuscript.