Participants were recruited through posters and online advertisements. We included 52 healthy, non-athletic volunteers (30 women and 22 men, aged 18–30 years old, Table 1) in a 3-month randomized, controlled trial (Netherlands Trial Register - NL5847; Figure 1). After stratification for age, sex, and baseline cardiorespiratory fitness, participants were randomized to a 12-week intervention of high- or low-intensity exercise training. Before and after the intervention, participants performed a maximal exercise test to measure cardiorespiratory fitness and underwent MRI measurements (Figure 2A).

Participants that were classified as athletic based on the definition by Haddad Herdy and Uhlendorf (38), i.e., VO₂max_males > 55 ml/kg/min, VO₂max_females > 45 ml/kg/min, were excluded. Furthermore, participants who engaged in intensive sports (>3 times/week) were also excluded. In addition, we excluded participants based on the following criteria: BMI >30 kg/m² (>class 1 obesity), MRI contra-indications, a history of chronic renal insufficiency, allergy to gadolinium-containing compounds, a history of psychiatric disorders, excessive smoking (>1 pack/day), excessive alcohol consumption (>21 units/week), or other regular drug use. Additionally, females were only included if they were on hormonal contraceptives to control for the effects of the hormonal cycle. We obtained written informed consent from all participants, and the study was approved by the local Medical-Ethical Committee of the Amsterdam University Medical Centre, University of Amsterdam (NL55943.018.15).

All participants were enrolled in an exercise program for 12 weeks, in which they were instructed to exercise three times a week for 45 min (1, 39) at the university sports center (USC). Their presence and active engagement were monitored by tracking their sports center visits using an automated fingerprint entrance system and by using weekly questionnaires on exercise duration and activities. Additionally, participants received a HR monitor (Polar, Finland) to measure HR during each training session (Table 1). Participants randomized to high-intensity exercise performed high-intensity interval training, targeting HR zones above 80% of their maximum HR. Participants randomized to low-intensity exercise performed stretching and toning exercises (active control condition), targeting HR zones under 60% of their maximum HR. In collaboration with the sports scientists of the USC, we provided a list of generic exercise group classes offered at the USC which were supervised by an experienced fitness instructor, that participants were allowed to choose from [Supplementary Material 1.5; Supplementary CERT Checklist (40)]. Participants that were engaged in physical activity before the study were instructed to do the recommended sports classes on top of their usual activities. For motivation purposes, participants were contacted regularly to check in on their progress and one experimenter joined them at least once during the intervention period to train with them (more detail in Supplementary Material 1.5). The Dutch version of the International Physical Activity Questionnaire [IPAQ; (41)] was used to measure physical activity during walking, intermediate and vigorous intensities before and after the intervention.

Participants underwent a cardiopulmonary exercise test on an ergometer before and after the exercise intervention to assess individual cardiorespiratory fitness. After 2 min of rest (baseline measurement), an incremental bicycle protocol (which was dependent on weight and sex) was started with a 3-min warm-up period, followed by an increase in resistance (watts) every minute until maximal effort (maximum resistance) or exhaustion, which was immediately followed by a 2-min recovery period at 50 watts resistance. Breath-by-breath gas exchange measurement data were obtained to determine maximum oxygen uptake [VO₂max (mL/kg/min)] (39). VO₂max data were time-averaged using 10s intervals (26). For exercise tests to be considered maximal, participants had to reach both a plateau in VO₂max and a respiratory exchange ratio of > 1.1 CO₂/O₂. VO₂max tests took place at least 24 h before the MRI scans.

Participants were scanned on a 7T whole-body MR system (Philips, Best, The Netherlands) using a dual-channel transmit coil and a 32-channel receive head-coil, and on a 3T whole-body MR system (Philips, Best, The Netherlands) using a body transmit coil and a 32-channel receive head-coil. A 24h gap between the last workout and MRI scanning was ensured to minimize the potential influences of dehydration on brain volume and acute exercise effects (42).

Blood samples were collected before the MRI measurements, pre and post-intervention, to obtain: 1) brain-derived neurotrophic factor (BDNF) levels as a proxy for exercise effects on hippocampal neuronal health, plasticity, and possibly neurogenesis (59–61); 2) insulin-like growth factor 1 (IGF) as a proxy for cell proliferation and the inhibition of cell death (62), 3) free vascular endothelial growth factor (VEGF) levels, as a prime regulator of angiogenesis (63); and 4) hematocrit levels. For BDNF and IGF, a total of 4 mL serum was collected (15 min centrifugation at 1,000 × g). For VEGF, 8 mL serum was collected in PECT tubes (64) through an open system, drop by drop, without using a tourniquet (60 min centrifugation at 4°C at 1,700 × g). All samples were aliquoted and stored at −80°C. Growth factors were quantified using enzyme-linked immunosorbent assays (ELISAs) according to the manufacturer's instructions (R&D Systems; DVE00 for VEGF; DBD00 for BDNF; DG100 for IGF), and optical densities were converted into concentrations using an LP4 logistic fit (Graphpad Prism 5).

Sample size calculations can be found in the Supplementary Methods (1.4). All data were checked for normality and, in the case of non-normality, transformed accordingly. To account for missing data and the longitudinal nature of the trial, linear mixed-effects models were used to investigate the condition (high- vs. low-intensity exercise) × time (pre- vs. post-intervention) interaction effects in Rv.3.5.3 (65) using the lme4 package (66). Sex (female vs. male) was tested as a possible predictor but did not contribute to any of the models. Model selection was based on an adjusted top-down procedure, in which the resulting models were compared using the Bayesian information criterion (BIC), and subsequently, the model best capturing the data was reported using χ² tests and BICs (67, 68). Bayes factors (BF) were calculated, indicating the strength of evidence, using BIC approximation (69). The evidence categories of Wetzels et al. were used (70) (Supplementary Table 1). We regarded changes in cardiovascular fitness and hippocampal and dACC volume as primary hypotheses, and changes in neuro-metabolites, CBV, myelination, CBF, and neurotrophic factors as secondary hypotheses.

Statistical tests regarding the interaction effects were corrected for multiple comparisons within modalities using Sidak's correction: α^* = 1 – (1 – α)^∧(1/m), with α = 0.05 and m being the number of interaction and main effects (m = 3), which resulted in an α^* = 0.02. Additionally, Tukey corrected t-tests were used as post-hoc tests.

Furthermore, exploratory analyses testing associations of all variables with changes in VO₂max were conducted using linear models in R, including the baseline measure of the explanatory variables and VO₂max as covariates (α = 0.05). We additionally tested exercise-condition as a possible covariate, which did not contribute to the model. Mean and standard deviation per time-point per variable are reported in Supplementary Table 4.

Hours spent exercising demonstrated high compliance with the exercise program in both exercise groups (Table 1). As expected, participants in the high-intensity condition spent significantly more time exercising in the intended higher HR regime than the low-intensity condition, which did not explain the change in VO₂max [t₍₄₀₎ = 1.34, p = 0.19; Figure 3C]. The low- and high-intensity exercise groups did not show a significant change from pre- to post-intervention on the total score or scores for walking and intermediate intensity activities as measured with the IPAQ questionnaire. However, a significant condition x time effect was found on vigorous-intensity activities [χ²₍₁₎ = 5.46, p = 0.02], indicating an increase in the high-intensity group but not in the low-intensity group (Supplementary Results 2.3). Nevertheless, contrary to our expectations, we found no condition x time effect on VO₂max; instead, we found decisive evidence (BF > 100) for an effect of time [χ²₍₁₎ = 15.43, p < 0.001; low-intensity: 4.7%, high-intensity: 12.65% change] (Figure 3A). Nevertheless, post-hoc tests revealed only a significant increase in the high-intensity condition [low: t₍₄₉₎ = 1.72, p = 0.09; high: t₍₄₉₎ = 4.20, p < 0.01]. In line with the results on VO₂max, we found no interaction effect, but decisive evidence (BF > 100) for a main effect of time [χ²₍₁₎ = 38.92, p < 0.001] on the maximal resistance attained. Post-hoc tests revealed a significant increase in both conditions [t₍₂₃₎ = 4.67, p < 0.01; t₍₂₄₎ = 7.02, p < 0.01; Figure 3B). No group effects on HR during the VO₂max test were found (Supplementary Table 2).

We found no interaction effect of condition and time, and no main effects of time for BDNF, VEGF, or IGF levels (BF < 100⁻¹) (Supplementary Table 2). No significant interaction effect of condition and time was found for hematocrit [χ²₍₂₎ = 0.03, p = 0.98], but a main effect of time was found, indicating an increase in both exercise groups [χ²₍₁₎ = 12.90, p < 0.01; BF = 8.4].

We found decreased right hippocampal volume, particularly in the high-intensity exercise condition. This effect was not found to be driven by changes in specific hippocampal subfields as previously suggested (35). Although most studies, particularly in older adults, have reported increases in hippocampal volume following exercise (9, 12, 73), Wagner et al. (26) also demonstrated that young, healthy participants (age 21–28) who failed to benefit from an exercise program showed decreases in hippocampal volume. These findings suggest that changes in hippocampal volume are highly variable between individuals, especially in young adults (74). Indeed, a meta-analysis concluded that exercise does not stimulate the hippocampal growth in young participants but instead prevents its volume decline as it occurs with increasing age (11). Additionally, many studies that found hippocampal volume increases through exercise interventions in younger adults were conducted in patient populations instead of healthy individuals (75–79). Hence, exercise effects on hippocampal volume may be dependent on age and disease (80), and therefore exercise could be regarded as a restorative, rather than stimulatory intervention.

To further understand the potential underpinnings of the volume reductions of the right hippocampus, we investigated neuro-metabolite concentrations and peripheral BDNF and IGF concentrations as markers of neuronal remodeling. Utilizing the potential of ultra-high field MRI, we resolved numerous (low-concentration) neuro-metabolites, such as Glu, Gln, and GSH additionally to NAA. This was important as previous studies in young adults reported both increases (81) and decreases (26) in Glx (Glu + Gln) and NAA after aerobic exercise.

We found increases in GSH levels in the left hippocampus over time, independent of exercise intensity, which were mainly driven by an increase in the low-intensity condition. GSH is known to be responsible for the survival and function of neural cells and for sustaining dendrite integrity and cognitive function (82). Reducing GSH levels in hippocampal neurons of mice resulted, e.g., in dendritic disruption, glial activation in CA1, and cognitive impairment (83). GSH levels have previously been shown to increase in rats after long-term exercise (84), but not in the hippocampus; therefore, this finding was somewhat surprising, particularly given the absence of findings in other metabolites involved in neuronal integrity in the hippocampus. We can speculate that increases in GSH levels detected here could be indicative of cell proliferation, but replication of these findings is needed, and associations with fitness changes will need to be confirmed. Furthermore, neuro-metabolite levels in this study were only measured in the left hippocampus, whereas volume decreases were found particularly in the right hippocampus. Previous studies have found metabolite changes due to exercise interventions in both the left (81) and right hippocampus (26), and therefore future studies should consider acquiring bilateral hippocampal MRS data.

Although no associations between fitness and hippocampal NAA were found, increased fitness was associated with increased NAA levels in the dACC (the control region). NAA is a well-accepted marker of neuronal viability, and exercise-induced increases in NAA could potentially reflect improvements in neuronal health. In the rodent hippocampus, NAA changes have been linked to neurogenesis (85–88), a form of structural hippocampal plasticity that was recently re-confirmed to take place also in the human brain (89–92) and is thought to underlie associations between exercise and hippocampal volume (9, 13, 24, 35, 61, 93). In our exploratory analyses, we found a novel positive association between NAA in the dACC and fitness. A previous study (81) has found changes in NAA in the hippocampus that we could not replicate. This exploratory finding thus requires further replication, in order to investigate whether exercise-induced alterations in neurometabolites are region-specific or global.

We further found individual BDNF and IGF changes to be associated with individual fitness changes. While BDNF has been suggested as a primary candidate, IGF also plays a vital role in stimulating neurogenesis in the hippocampus (63, 94). Most rodent studies determined hippocampal levels of BDNF and IGF, whereas we determined neurotrophic factors in peripheral blood, which may not directly reflect changes in the hippocampus or its subregions. Nevertheless, BDNF and IGF associations with fitness indicate that individual changes in cardiorespiratory fitness were indeed associated with markers of neuronal health.

In summary, we found exercise-independent increases of GSH, possibly indicating a gain in cell proliferation over time. Additionally, dACC NAA levels, as well as BDNF and IGF levels, were positively associated with fitness, hinting toward a relationship of cardiovascular fitness and neuronal remodeling in young, healthy adults.

Additionally, vascularization was investigated as a potential underlying mechanism of exercise-induced changes in hippocampal volume. Changes in vascularization can be estimated using CBF, CBV, as well as the peripheral neurotrophic factor VEGF.

One of the first studies on exercise-induced changes in vascularization suggested strong, positive effects of exercise on hippocampal angiogenesis, estimated with CBV, in both rodents and older adults (16). They argued that oxygen and neurotrophic factors can reach the brain more efficiently through formations of new blood vessels and could, therefore, positively affect cardio-pulmonary and cognitive functions. Subsequent studies have not yet replicated beneficial exercise-induced changes in blood volume but found a positive association of fitness changes and changes in hippocampal vasculature (CBF and CBV) in a population of older adults (95). In contrast, Thomas et al. found no effects of exercise on CBV (12) and we found a decrease in left hippocampal CBV over time in young adults which is suggestive of no or negative effects of exercise on hippocampal vasculature.

Furthermore, even though non-significantly, increases in individual GM CBF were associated with improved individual fitness, indicating global, whole-brain changes rather than specific fitness effects on hippocampal CBF, but these results need replication. CBF is a potential marker for neuronal activity, as blood supply needs to be guaranteed during higher energy demands (96). These findings are in line with several other studies that found widespread beneficial effects of exercise on CBF (97, 98), and no effects on CBF in the hippocampus (99).

In line with the absence of exercise effects on hippocampal vasculature, we also did not find any exercise-induced changes in peripheral VEGF levels. VEGF is thought to play a pivotal role in the formation of new blood vessels. Its peripheral levels were not changed in our study, even though we used collection in PECT tubes, which should provide an accurate estimation of in-vivo circulating VEGF levels (100).

In sum, independent of fitness, we found CBV in the left hippocampus to decrease over time, as a potential marker for angiogenesis. Interestingly, even though non-significant, we found indications that increased fitness is positively associated with global increases in CBF.

Contrary to our hypothesis and previous studies, mainly performed in middle-aged or older adults (9, 101), our high-intensity aerobic exercise intervention did not improve cardiorespiratory fitness significantly more than the low-intensity stretching and toning intervention. Although related exceptions have been reported (102), this was an unexpected finding, as low-intensity training was previously found not to influence VO₂max (101), and a meta-analysis stressed the importance of high-intensity training (HR > 80% of the maximum HR) for improving fitness in younger adults (18–45 years old) (33). Moreover, the two groups strictly adhered to the instructed intensities of the respective interventions, as is further evidenced by the significantly higher percentage of maximum HR attained during the training sessions in the high-intensity condition (Figure 3C). These findings, therefore, may suggest that engaging in stretching and toning activities might have caused sufficiently large changes in activity and/or lifestyle to increase fitness in young, otherwise non-athletic volunteers. Indeed, previous studies have argued that also low-intensity exercise such as yoga and pilates might be beneficial for cardiovascular fitness in young individuals (103, 104), even though sample sizes were small and effect sizes were medium. This would partially explain why we did not observe an interaction effect between the low- and high-intensity conditions (despite the fact that only the high-intensity group showed a significant increase in VO₂max). Nevertheless, our findings suggest that perhaps in some individuals, even minimal time of elevated HR might be enough to increase cardiorespiratory fitness, which is an important finding for those starting to exercise again, or in rehabilitation medicine. However, this suggestion needs to be interpreted with caution and requires further investigation, as time spent in these high HR regimes did not explain changes in VO₂max. As an alternative explanation, it has been suggested that some forms of exercise (regardless of their intensity) are more “mindful” than others, presuming possible synergistic effects of physical and cognitive activity on brain structure and function; however, this remains to be confirmed in future studies (105, 106).

On the other hand, however, other studies have found that moderately intense exercise benefits neuronal health most (107–109). Therefore, speculatively, our high-intensity intervention may have been too intense, which might have evoked an accompanying stress reaction, which in turn could have deteriorated, rather than improved, some of the measures studied here (110–112). Although we did not measure stress hormone levels, we found no association between cardiorespiratory fitness or volume measures and subjective stress scores (measured with the Depression Anxiety Stress Scale - DASS-S), indicating that this explanation is less likely.

Taken together, our findings highlight the importance of a passive control condition that refrains from exercise entirely in future studies; in addition to an active control condition (as implemented in this study), and different levels of exercise intensities (e.g., to control for possible environmental “enrichment” effects, including changes in social interactions, and individual motivation), to investigate if lifestyle changes or exercise training change brain health in young, healthy adults. In our study, we did not directly supervise the exercise sessions, but chose a more naturalistic approach, in which participants could choose from a predefined list of classes at the sports center. These classes were supervised by a certified fitness instructor. This approach has the advantage that it resembles better how exercise would be implemented in everyday life and introduces less stress and pressure. Nevertheless, it also meant that we had less control over the classes that participants chose to follow, which might explain some of the variance in the cardiovascular fitness measures in this study.

Notably, previous studies have been inconsistent in their operationalization of (maximal) cardiorespiratory fitness (e.g., VO₂plateau, VO₂peak, VO₂vat), thereby making it challenging to compare results (12, 26, 29, 30, 39, 81, 93, 101, 113). Therefore, we warrant it essential to harmonize the analyses and detailed reportings of such outcome measures.

It has further been suggested that individuals differ in the extent to which they are susceptible to fitness-based interventions, which may subsequently also influence the relationship between exercise and brain-related changes (26, 93, 114, 115). Upon confirmation in more extensive studies [e.g., the IGNITE study (116)], it would become essential to develop individualized exercise programs to confer neurobiological benefits (11). So far, the most prominent advantages of exercise for the brain seem to apply mainly to middle-aged or older adults and diseased populations, and less so to younger, healthy adults, due to possible ceiling effects and an already optimal neuronal health. Nevertheless, it is important to mention that we studied a relatively homogenous group (comparable BMI, age, fitness), which might have made ceiling effects even more likely. Overall, our results point toward the hypothesis that exercise benefits on the human brain are restorative rather than stimulatory.

Interestingly, evidence is accumulating that exercise effects might be transient and change relatively rapidly in young adults. For instance, Van Der Borght et al. (117) found that the vasculature in rodents changed rapidly 3 days after exercise training but also declined again after 24 h of inactive behavior. Thomas et al. (12) even observed temporary changes in young healthy adults, at least for the anterior hippocampus. Specifically, they found a temporary effect of exercise intensity on the volume and myelination of the anterior hippocampus. As these changes were temporary, consistency and regularity of training seem to be essential factors, which could have influenced our measures. In our study, we intentionally controlled for rapid exercise effects on perfusion by guaranteeing a 24 h gap between the last exercise session and MRI measurements, as we were interested in the prolonged effects of exercise intensity.