Physical activity as a modifiable risk factor in preclinical Alzheimer's disease

It is estimated that nearly half of AD cases worldwide are attributable to modifiable risk factors^1,2,3. Physical inactivity, in particular, plays a prominent role in the USA and Europe^1,2. Although animal studies support a benefit of physical activity on AD progression^4,5, important knowledge gaps remain in human literature, limiting the effective translation into AD prevention trials. Most existing observational studies rely on self-reported physical activity^{6,7,8,9,10,11,12}, which is prone to recall bias and misreporting, especially in populations with or at risk for cognitive impairment. In addition, most studies focused on the clinical syndrome of AD^7,9,12,13,14, with few directly examining core AD biomarkers^{15,16,17,18,19,20,21}, and even fewer have done so longitudinally^22,23. A clearer understanding of how objectively measured physical activity is associated with the progression of AD neuropathology is essential, particularly during the preclinical period, when there is likely the greatest potential to modify disease trajectory.

With the increasing popularity of digital wearables, daily step count has become an easily accessible and understood measure of physical activity. Emerging literature suggests that higher step counts are linked to lower all-cause mortality^24,25. Our previous study demonstrated that higher step counts in cognitively unimpaired (CU) older adults with elevated baseline beta-amyloid (Aβ) burden were associated with slower prospective cognitive decline²⁶, supporting a potential protective role of physical activity in the preclinical stage of AD. However, it remains unknown whether this protective association with cognition is mediated by differences in AD neuropathology burden, or what levels of physical activity are associated with slower cognitive decline.

The current study addressed these questions by leveraging an expanded Harvard Aging Brain Study (HABS) cohort of CU older individuals with pedometer-measured physical activity, longitudinal Aβ and tau positron emission tomography (PET) data, and annual cognitive assessments up to 14 years. We examined whether physical activity is associated with slower cognitive and functional decline through different rates of Aβ and tau accumulation. We further examined the dose–response associations with physical activity levels to help inform future AD prevention trials and public health policies.

Participant characteristics and longitudinal trajectories

The current study examined 296 participants from HABS who were CU at baseline and followed longitudinally²⁷. Cognition and function were assessed annually using the Preclinical Alzheimer’s Cognitive Composite-5 (PACC5) and Clinical Dementia Rating²⁸ Sum of Boxes (CDR-SOB) scores, respectively. Global Aβ and inferior temporal cortex (ITC) tau burdens were measured longitudinally in a subset of participants (Aβ n = 241; tau n = 172), with participants undergoing their first tau scan at 2.2 ± 1.5 years after baseline because the technique was introduced mid-study. Table 1 summarizes the participant characteristics and longitudinal follow-up. The unadjusted longitudinal trajectories for global Aβ, ITC tau, PACC5 and CDR-SOB are shown in Extended Data Fig. 1. For illustration purposes only, to more clearly visualize the individual trajectories, the data were plotted according to high versus low baseline Aβ burden in columns and physical activity (pedometer-measured mean steps per day) in rows, defined by above and below the median values.

Baseline associations with physical activity

We first examined the associations between physical activity and baseline demographics, imaging and cognitive variables. There was a negative association between physical activity and age (Spearman’s r = −0.27, P < 0.001). Males had greater physical activity than females (Sex [males compared to females]: β = 927 [95% confidence intervals 247 to 1,607], P = 0.008). There were no significant associations between physical activity and education (r_partial = 0.08, P = 0.16), or baseline Aβ burden (r_partial = −0.004, P = 0.95), adjusting for age and sex. There was a significant interaction between lower baseline physical activity and elevated Aβ on greater initial ITC tau burden, adjusting for age, sex, years of education and time interval between study baseline and first tau scan (Physical activity × Aβ: β = −0.19 [−0.30 to −0.08], P = 0.001) (Extended Data Fig. 2). There was no additional independent association between baseline physical activity and initial ITC tau burden (Physical activity: β = −0.06 [−0.18 to 0.05], P = 0.28). Importantly, there were no interactive or independent associations of initial Aβ and tau burden with baseline physical activity, adjusting for the same covariates (Aβ × ITC tau: β = −0.04 [−0.15 to 0.07], P = 0.44; Aβ: β = 0.10 [−0.07 to 0.28], P = 0.25; ITC tau: β = 0.01 [−0.21 to 0.22], P = 0.96) (Extended Data Fig. 3). There were also no significant interactive (Physical activity × Aβ) or independent effects of physical activity on baseline PACC5 (Physical activity × Aβ: β = 0.003 [−0.07 to 0.08], P = 0.93; Physical activity: β = 0.003 [−0.07 to 0.08], P = 0.92). Baseline associations with CDR-SOB were not examined because participants were clinically unimpaired at baseline.

Physical activity associations with longitudinal Aβ

For primary analyses, we examined the effects of baseline physical activity on longitudinal imaging and cognitive measures. Using a linear mixed effects model, we found no association between baseline physical activity and longitudinal Aβ burden (Physical activity × Time: β = −0.0006 [−0.01 to 0.01], P = 0.92) (Fig. 1a).

Fig. 1: Associations of baseline physical activity with longitudinal Aβ, tau and cognition.

a, Linear mixed effects model revealed no association between baseline physical activity and longitudinal Aβ burden (β = −0.0006 [−0.01 to 0.01], P = 0.92; n = 241). b–d, By contrast, there were significant interactions between baseline physical activity and Aβ burden on longitudinal ITC tau burden (b), longitudinal cognition measured with PACC5 (c) and longitudinal functional decline measured with CDR-SOB scores (d). Individuals with high baseline physical activity and elevated Aβ (solid red line) showed slower ITC tau accumulation (β = −0.13 [−0.19 to −0.06], P < 0.001; n = 172) (b), slower PACC5 decline (β = 0.10 [0.05 to 0.16], P < 0.001; n = 296) (c) and slower CDR-SOB progression (β = −0.14 [−0.22 to −0.05], P = 0.001; n = 296) (d). Statistical significance was assessed using two-tailed t-tests, with P < 0.05 considered statistically significant without adjustment for multiple comparisons. Baseline physical activity (mean steps per day) and Aβ burden were modeled as continuous variables. To visualize the model results, the estimated trajectories based on representative levels of low versus high baseline physical activity and (for tau, PACC5 and CDR-SOB models) low versus high baseline Aβ burden are presented, with error bands representing 95% confidence intervals for the estimated trajectories. Low and high physical activity are represented by −1 and +1 s.d. relative to the mean (low, 2,800 steps per day; high, 8,700 steps per day). Low and high Aβ are represented, for illustration purposes, by the mean Aβ burden of Aβ-negative (PiB PVC-DVR = 1.17) and Aβ-positive (PiB PVC-DVR = 1.85) participants, respectively. The numbers of participants contributing longitudinal data to each 2.5-year segment for the respective statistical models are summarized in Extended Data Table 5.

Physical activity–Aβ interactions on longitudinal tau, cognition

We then examined the interactive associations between baseline physical activity and Aβ burden on longitudinal cognition and function. Baseline Aβ was modeled as a continuous variable to leverage the full dynamic range of global Aβ burden²⁹. Consistent with our earlier work²⁶, this larger HABS sample with extended follow-up showed a significant physical activity by Aβ interaction on longitudinal PACC5, whereby higher physical activity was associated with slower Aβ-related PACC5 decline (Physical activity × Aβ × Time: β = 0.10 [0.05 to 0.16], P < 0.001; Fig. 1c). In addition, there was an independent association between higher physical activity and slower PACC5 decline (Physical activity × Time: β = 0.06 [0.004 to 0.12], P = 0.04). We further extended our previous findings and identified a significant physical activity by Aβ interaction on longitudinal CDR-SOB, whereby higher physical activity was associated with slower Aβ-related functional decline (Physical activity × Aβ × Time: β = −0.14 [−0.22 to -0.05], P = 0.001) (Fig. 1d). There was no additional main effect of physical activity on CDR-SOB progression (Physical activity × Time: β = −0.05 [−0.13 to 0.04], P = 0.29).

We then addressed the question of whether the associations between physical activity and cognitive and functional decline were mediated by differences in tau pathology. We found a concordant interaction, whereby higher baseline physical activity was associated with attenuated Aβ-related ITC tau accumulation (Physical activity × Aβ × Time: β = −0.13 [−0.19 to −0.06], P < 0.001) (Fig. 1b). There was additionally an independent association between higher physical activity and slower tau accumulation (Physical activity × Time: β = −0.07 [−0.13 to −0.003], P = 0.04). Moderated mediation analysis demonstrated that slower ITC tau accumulation fully mediated the association between higher physical activity and slower PACC5 decline in individuals with elevated baseline Aβ (mediated effect: β = 0.59 [0.32 to 0.91], P < 0.001; direct effect: β = 0.11 [−0.30 to 0.52], P = 0.61; 84% mediated) (Fig. 2a). For CDR-SOB, slower ITC tau accumulation partially mediated the association between higher physical activity and slower CDR-SOB progression in individuals with elevated baseline Aβ (β = −0.43 [−0.71 to −0.20], P < 0.001; 40% mediated) (Fig. 2b). There remained a significant direct effect of physical activity on functional decline (β = −0.64 [−1.16 to −0.13], P = 0.01). In individuals with low baseline Aβ, there were no significant total or mediated effects of physical activity on PACC5 decline or CDR-SOB progression (P > 0.75).

Fig. 2: Tau accumulation mediated the associations between physical activity and cognitive/functional decline in preclinical AD.

a,b, Individual slopes for ITC tau, PACC5 (a) and CDR-SOB (b) were extracted from linear mixed effects models for moderated mediation analyses (n = 172; in participants who have both longitudinal tau and cognitive data). We modeled physical activity (mean steps per day) as predictor, ITC tau slope as mediator and PACC5 or CDR-SOB slopes as outcome. Both physical activity and Aβ burden were modeled as continuous variables in the mediation models. For the moderation analyses, low and high levels of Aβ burden were represented by the mean Aβ burden of Aβ-negative (PiB PVC-DVR = 1.17) and Aβ-positive (PiB PVC-DVR = 1.85) participants, respectively. Statistical testing was performed using a quasi-Bayesian Monte Carlo method based on 10,000 simulations to generate the estimates and 95% confidence intervals, with two-tailed P < 0.05 considered statistically significant without adjustment for multiple comparisons. Results demonstrated that at an elevated level of baseline Aβ burden, slower ITC tau accumulation fully mediated the association between higher physical activity and slower PACC5 decline (β = 0.59 [0.32 to 0.91], P < 0.001, 84% mediated) (a) and partially mediated the association between higher physical activity and slower CDR-SOB progression (β = −0.43 [−0.71 to −0.20], P < 0.001, 40% mediated) (b). In individuals with low baseline Aβ burden, there were no significant total or mediated effects of physical activity on PACC5 decline (total effect: β = 0.03 [−0.27 to 0.31], P = 0.85; mediated effect: β = −0.002 [−0.16 to 0.15], P = 0.98) or CDR-SOB progression (total effect: β = 0.05 [−0.28 to 0.37], P = 0.75; mediated effect: β = 0.001 [−0.11 to 0.12], P = 0.98). The error bars represent the 95% confidence intervals for the estimated mediated, direct and total effects.

Sensitivity analyses

The interaction between baseline physical activity and Aβ on longitudinal tau, PACC5 and CDR-SOB remained significant in sensitivity analyses assessing for possible confounds (Table 2): (1) using a more stringent physical activity cutoff excluding days that registered fewer than 1,000 steps; (2) using Aβ and tau PET data without partial volume correction (PVC); (3) including only participants age <70 years to minimize potential confounding effects of advanced age (the negative association between physical activity and age was absent in this age <70 subgroup; Spearman’s r = −0.01, P = 0.93); (4) adjusting for the season of physical activity measurement, self-reported physical activity, systemic vascular risk and depressive symptoms; (5) excluding participants who developed mild cognitive impairment or dropped out of the study within the first 2 years of follow-up; and (6) adjusting for baseline tau, PACC or CDR-SOB in their respective models to account for physical activity effects on baseline outcome measures.

Physical activity levels on Aβ-related changes in tau, cognition

Lastly, we examined the dose–response associations between levels of physical activity and longitudinal tau accumulation, cognitive and functional decline. Ordinal levels of physical activity were defined using cutoffs adapted from ref. ³⁰, which were modified to better suit the lower activity levels in older adults and to ensure an adequate number of participants in each subgroup (Extended Data Table 1): ‘inactive’ (≤3,000 per day), ‘low activity’ (3,001–5,000 steps per day), ‘moderate activity’ (5,001–7,500 steps per day) and ‘active’ (≥7,501 steps per day). Participant characteristics across physical activity levels are summarized in Extended Data Table 2. Higher physical activity levels were associated with younger age, but baseline demographics, imaging and cognitive variables were not significantly different across physical activity groups adjusting for age and sex. In linear mixed effects models using the inactive subgroup as reference, all higher levels of physical activity were associated with slower Aβ-related ITC tau accumulation, PACC5 decline and CDR-SOB progression, except that the slower CDR-SOB progression in the low activity group did not reach statistical significance (Table 3). To better visualize the dose–response relationships, we modeled the physical activity level by Aβ interaction on extracted ITC tau, PACC5 and CDR-SOB slopes (Fig. 3). In individuals with elevated baseline Aβ, even low levels of physical activity (3,001–5,000 steps per day) were associated with substantially slower rates of tau accumulation, cognitive and functional decline. There were further attenuations of tau accumulation and cognitive and functional decline at moderate activity (5,001–7,500 steps per day), with similar rates in the active group (≥7,501 steps per day), suggesting a plateauing of effects.

Fig. 3: Physical activity levels and changes in tau and cognition in preclinical AD.

a–c, Using extracted slopes for ITC tau (n = 172) (a), PACC5 (b) and CDR-SOB (c) (PACC5 and CDR-SOB, n = 296), we examined the interactive effects of baseline physical activity level (ordinal) and Aβ burden (continuous) using linear regression models. Levels of physical activity (ordinal) were defined as inactive (≤3,000 steps), low activity (3,001–5,000 steps), moderate activity (5,001–7,500 steps) and active (≥7,501 steps). The number of individuals in each physical activity subgroup included in the tau, PACC5 and CDR-SOB analyses are summarized in Extended Data Table 1. Aβ burden was modeled as a continuous variable. For illustration purposes, low and high Aβ are represented by the mean Aβ burden of Aβ-negative (PiB PVC-DVR = 1.17) and Aβ-positive (PiB PVC-DVR = 1.85) participants, respectively. The error bars represent the 95% confidence intervals for the estimated effects of physical activity levels on tau and cognitive slopes at representative levels of low and high Aβ burden. Results demonstrate that in individuals with elevated baseline Aβ, even low levels of physical activity (3,001–5,000 steps) were associated with substantially slower rates of tau accumulation and cognitive decline compared to inactive individuals. There were further attenuations of tau accumulation and cognitive and functional decline at moderate activity (5,001–7,500 steps per day), with similar rates in the active group (≥7,501 steps per day).

To help contextualize the associations between physical activity levels and cognitive and functional decline, we used the linear mixed effects models from Table 3 to estimate the longitudinal PACC5 and CDR-SOB trajectories at representative levels of low versus high baseline Aβ burden (represented by the mean global Aβ burden of Aβ-negative and Aβ-positive participants, respectively) across physical activity levels (Extended Data Fig. 4). For individuals with elevated Aβ, from baseline to the median cognitive follow-up of 9 years, estimated PACC5 z-scores declined by 2.5 (inactive), 1.5 (low activity), 1.1 (moderate activity) and 1.2 (active) points, respectively. Compared to inactive individuals, cognitive decline was lower by 40%, 54% and 51% across increasing physical activity levels. Over the same period, estimated CDR-SOB scores increased by 2.2 (inactive), 1.4 (low activity), 1.2 (moderate activity) and 1.1 (active) points in individuals with elevated Aβ. Compared to inactivity, there were 34%, 45% and 51% slower functional decline across increasing physical activity levels. Using −1.5 PACC5 z-score as a threshold for impaired cognition³¹, CU individuals with elevated Aβ were estimated to reach this threshold for cognitive worsening at 6.5 (inactive), 9.6 (low activity), 13.6 (moderate activity) and 12.7 (active) years from baseline, respectively. Similarly, using a CDR-SOB increase of 1.5 as a threshold for clinically meaningful functional decline in early AD³², individuals with elevated Aβ were estimated to reach this threshold at 7.1 (inactive), 10.2 (low activity), 11.9 (moderate activity) and 13.6 (active) years from baseline, respectively.

In an expanded cohort of 296 CU older adults with pedometer-measured physical activity, baseline Aβ PET and up to 14 years of annual cognitive follow-up, we replicated our previous finding of a protective association between higher physical activity and slower Aβ-related cognitive decline, and further demonstrated a similar relationship with functional decline. Importantly, we demonstrated that these associations with cognition and function were not related to differences in cross-sectional or longitudinal Aβ pathology. Instead, in a subset of 172 individuals with longitudinal tau PET, we found a new and concordant association between higher physical activity and slower Aβ-related early neocortical tau accumulation, which significantly mediated the relationships with slower cognitive and functional decline. Our dose–response analyses further suggest that the largest incremental differences between increasing physical activity levels were observed in the most sedentary individuals, with plateauing associations with both tau and cognition at a moderate level of activity (5,001–7,500 steps per day), potentially offering a more approachable physical activity goal for older sedentary individuals. Collectively, our results support targeting physical inactivity, either alone or in combination with anti-Aβ therapy, as an intervention in future prevention trials to slow the progression of tau pathology and delay the onset of cognitive and functional decline in preclinical AD.

Many observational studies have demonstrated associations between physical activity and reduced risks of cognitive impairment and AD. However, most existing literature relied on self-reported physical activity, which is susceptible to recall bias or misreporting, especially among populations with or at risk for cognitive impairment. A few longitudinal studies have examined the effects of objectively measured physical activity (by pedometer or accelerometer) in older adults without dementia and found associations with reduced risks of cognitive decline, AD or all-cause dementia^13,33,34,35. Our previous study²⁶ extended these findings to CU individuals and demonstrated an association between higher step counts and slower prospective cognitive decline in preclinical AD. The current study replicated this finding using a considerably larger cohort, including younger individuals in late midlife, and over a substantially longer duration of follow-up (median of 9.3 years instead of 6 years). The latter is notable given that some observational studies found diminished benefits of physical activity with longer follow-up time^36,37,38. We further expanded our cognitive findings and demonstrated a similar association with slower Aβ-related functional decline measured by CDR-SOB, which has been frequently used as an outcome measure in AD clinical trials^39,40.

There is growing interest in understanding whether physical activity associations with reduced cognitive decline are mediated by differences in AD neuropathology, with most studies focusing on Aβ. Although animal models suggest that physical exercise may reduce Aβ burden^4,41, evidence in human studies has been mixed. A few cross-sectional studies found associations between higher self-reported physical activity and lower Aβ pathology^{15,16,17,18,19,20}. Other large CU cohort studies have found no associations between self-reported physical activity and cross-sectional or longitudinal Aβ PET burden^21,22,23. Consistent with the latter, our study found no associations between pedometer-measured physical activity and baseline or longitudinal Aβ burden, suggesting that our findings were unlikely to be driven by different rates of Aβ accumulation.

By contrast, we demonstrated a new relationship between physical activity and longitudinal tau burden, where higher step counts were associated with slower early neocortical tau accumulation in individuals with elevated baseline Aβ. We further demonstrated that reduced tau accumulation significantly mediated the associations between higher physical activity and slower cognitive decline in preclinical AD. These observational findings support a potential role for physical activity in modifying the AD pathological cascade, which remains to be confirmed in future randomized clinical trials. Interestingly, although tau change fully mediated the effects of PACC5 decline (84% mediated), it only partially mediated (40%) the effects on CDR-SOB progression, with a significant remaining direct effect. This suggests that a substantial portion of the physical activity association with slower functional decline was mediated by mechanisms besides tau-related injury, which should be further examined in future studies (for example, physical frailty and functional reserve).

A small number of prior studies have investigated the cross-sectional associations between physical activity and tau. A few have found associations between higher physical activity and lower cerebrospinal fluid phosphorylated tau or phosphorylated tau to Aβ42 ratios^19,20,42,43. Others have examined the cross-sectional relationships between self-reported physical activity and tau PET with mixed results. Two studies found that higher self-reported physical activity was associated with lower cross-sectional tau burden^44,45. However, a larger study of CU individuals found no cross-sectional associations between self-reported physical activity and regional tau burden and no significant mediation of physical activity effects on cross-sectional cognition by tau²¹. It is likely that our use of objectively measured physical activity and, more importantly, our longitudinal design increased our sensitivity to detect a relationship between physical activity and early aggregated tau pathology in preclinical AD.

Although our rich longitudinal PET and cognitive data are a substantial strength, this study remains observational, and we are unable to rule out potential reverse causality (that is, individuals with prodromal AD may have reduced physical activity from their underlying disease). However, several factors reduce this likelihood. First, there were no associations between physical activity (independently or interactively with Aβ) and baseline cognition. Although there was a physical activity and Aβ interaction on baseline tau burden, a reversed relationship was not observed—higher cross-sectional Aβ or tau burden (individually or interactively) was not associated with lower baseline physical activity. In addition, the association between higher physical activity and slower longitudinal ITC tau accumulation remained significant after adjusting for the effects of lower initial tau burden. Our findings were further unchanged in sensitivity analyses that excluded participants who developed mild cognitive impairment (or dropped out) within the first 2 years of follow-up, reducing the likelihood that our findings were driven by individuals with impending cognitive and/or clinical changes. Although these observations are reassuring, it is important to emphasize that randomized clinical trials are ultimately needed to establish causal relationships between physical activity and preclinical AD progression.

There is substantial public interest in understanding how much physical activity is needed to improve health. Although this question cannot be directly answered by observational studies, we examined the dose–response associations between physical activity levels and progression of tau and cognitive decline, which may help inform the design of future AD prevention trials. Our results suggest that even a modest increase in physical activity may be associated with attenuated tau accumulation and cognitive decline in sedentary individuals on a preclinical AD trajectory: compared to inactive individuals, low levels of physical activity were associated with 34% to 40% slower cognitive and functional decline over 9 years (the median duration of cognitive follow-up). Notably, the associations with more favorable tau and cognitive trajectories reached a plateau by moderate levels of physical activity (5,001–7,500 steps per day), which may be a less daunting goal for sedentary older adults than the popular goal of 10,000 steps per day commonly referenced in lay media. Several studies have demonstrated a similar curvilinear relationship between step count and all-cause mortality or dementia^24,25,34, with benefits plateauing around 6,000–8,000 steps per day for older adults and around 8,000–10,000 steps per day for younger populations, suggesting a ceiling effect. Specifically for the current study, our findings suggest that there may be a limit to how much increasing physical activity may dampen the detrimental effects of Aβ on tau and cognitive and functional decline, and support designing clinical trials that combine lifestyle interventions with anti-Aβ therapy to more effectively modify the trajectory of preclinical AD.

Future studies are needed to understand the mechanisms underlying the protective effects of physical activity on Aβ-related tau pathology and cognitive decline. Although adjusting for composite vascular risk did not impact our findings, better vascular fitness remains a likely potential mechanism. Greater physical activity has been linked to better cardiopulmonary fitness^46,47, which has, in turn, been shown to dampen the negative association between Aβ and cognition⁴⁸. In addition, greater accelerometer physical activity in CU older adults has been associated with greater cerebral blood flow⁴⁹, which has been shown to be impaired early in the AD cascade⁵⁰. Lastly, greater physical activity and/or exercise have also been shown in animal models and humans to decrease inflammation and upregulate VEGF-A, BDNF and Irisin^{51,52,53,54,55}, which are candidate pathways with the potential to modulate Aβ effects on tau and cognition^{56,57,58,59,60}.

The current study has several limitations. As discussed above in greater detail, this is an observational study and we are unable to fully rule out the potential influence of reverse causality. Future randomized clinical trials are required to establish causal relationships. Although our pedometer-measured physical activity is a strength, it was only assessed at baseline and did not classify the duration, intensity or history of physical activity, nor capture nonstep-related physical activity (for example, swimming, resistance training). Future studies with longitudinal physical activity measurements (for example, with actigraphy) and life course physical activity data are needed to better delineate temporal trends. The HABS cohort consists of highly educated, predominantly non-Hispanic white individuals and excluded those with symptomatic cerebrovascular disease and poorly controlled vascular risk factors at baseline, which may limit the generalizability of our findings to other populations.

In summary, leveraging a deeply phenotyped CU cohort with objectively measured physical activity and longitudinal neuroimaging and cognitive assessments, we demonstrated an association between higher physical activity and attenuated accumulation of early neocortical tau pathology, which significantly mediated the relationship with slower cognitive and functional decline in preclinical AD. Our dose–response analyses further suggest that the incremental differences in associations between higher physical activity levels and slower tau and cognitive changes were greatest for the most sedentary individuals, and appeared to reach a plateau by moderate levels of physical activity (5,001–7,500 steps per day). Taken together, our findings support targeting physical inactivity as a strategy in future randomized clinical trials to modify the trajectory of tau and cognition in preclinical AD, and potentially provide an easily understood and more attainable physical activity goal for older sedentary individuals at high risk of cognitive decline. Our results further suggest that AD prevention trials using activity-based interventions may benefit from preferentially enrolling sedentary individuals with elevated Aβ levels, longer trial durations, and consideration of examining biomarker endpoint with tau PET, because these approaches may maximize the likelihood of demonstrating a protective effect of physical activity on early AD progression.

Participants

The current study included 296 participants from HABS²⁷, a community cohort of individuals aged 50 to 90 who were CU at baseline (global CDR²⁸ of 0, education-adjusted Mini-Mental State Examination⁶¹ score of 27 or greater, and Logical Memory IIa Delayed Recall performance in the normal range⁶²). Inclusion and exclusion criteria and study protocol have been detailed previously²⁷ and published online (https://www.synapse.org/Synapse:syn53910452/wiki/631007). We examined all eligible participants with baseline physical activity measurement, Aβ PET imaging and at least two longitudinal cognitive assessments. Extended Data Fig. 5 illustrates that of 348 eligible HABS participants at baseline, 50 were excluded from the current study because of missing data. Importantly, the included group did not differ in demographics or baseline cognition (Extended Data Table 3). Baseline characteristics by self-reported sex are summarized in Extended Data Table 4. Additional subsets of participants were used in the longitudinal Aβ PET (n = 241) and tau PET (n = 172) analyses (Table 1). Data were collected from April 2010 to February 2025. The Mass General Brigham Institutional Review Board approved the HABS protocol and procedures, and all participants signed a written informed consent before the completion of any study procedures.

Physical activity

Baseline physical activity was measured using a waistband-mounted pedometer (HJ-720ITC; Omron Healthcare), which has been shown to accurately measure step counts⁶³. Participants were asked to wear the pedometer for seven consecutive days during waking hours. Using previously published cutoffs for pedometer data quality, days that registered <100 or >30,000 steps were excluded^26,64. We included participants with at least 4 days of recorded activity within these cutoffs⁶⁴. Mean steps per day were used as the primary measure of physical activity, which was square-root transformed to account for skewness.

PET imaging

Detailed methods have been published previously⁶⁵. Briefly, global Aβ burden was measured using PiB PET as a DVR across a composite of frontal, lateral temporal and parietal, and retrosplenial regional uptake defined using FreeSurfer (v.6.0). ¹⁸F-Flortaucipir PET was introduced mid-study to measure tau, with participants undergoing their first ¹⁸F-flortaucipir PET at 2.2 ± 1.5 years after baseline. Tau was measured as standardized uptake value ratio in the ITC, a prominent site of early neocortical tau accumulation in preclinical AD associated with emerging cognitive impairment⁶⁵. Both Aβ and tau PET used cerebellar gray matter as reference and the geometric transfer matrix method for PVC⁶⁶. Sensitivity analyses were performed using PET data without PVC.

Cognitive assessment

Cognition was assessed using the PACC5 (ref. ⁶⁷), a composite z-score that includes the Mini-Mental State Examination⁶¹, Wechsler Adult Intelligence Scale-Revised Digit Symbol Coding⁶⁸, Wechsler Memory Scale-Revised Logical Memory delayed recall⁶², Free and Cued Selective Reminding Test (free recall plus total recall)⁶⁹ and Category Fluency Test⁷⁰. Functional decline was assessed using the CDR-SOB score.

Statistical analyses

We used R v.4.3.1. Continuous variables were z-transformed before analysis to obtain standardized effect sizes, with the exception of PACC5 (already a z-score). For illustrative purposes, nonstandardized variables were used when plotting model results to enhance interpretability (‘ggplot2’ and ‘plot_model’ packages). A two-tailed P < 0.05 was considered statistically significant.

We first examined the associations between baseline physical activity with age and sex. We also examined the associations between physical activity and baseline imaging and cognitive variables, adjusting for age and sex.

For primary analyses, we used linear mixed effects model (‘nlme’ package) to examine: (1) the effects of physical activity on longitudinal Aβ; and (2) the interactive effects of physical activity and baseline Aβ on longitudinal ITC tau, PACC5 and CDR-SOB, respectively. Baseline Aβ was modeled as a continuous variable to leverage the full dynamic range of Aβ burden, which has been shown to capture additional information on preclinical AD progression over a dichotomous approach and avoids variability introduced by positivity threshold selection, which remains an evolving area of research²⁹. Time was operationalized as years from baseline assessment. We adjusted for age, sex, APOE ε4 carrier status, education, their interactions with time, and included random intercepts and slopes. There were no significant APOE ε4 effects on longitudinal tau, PACC5 or CDR-SOB; therefore, APOE was removed from these models to include three participants with missing genotype data.

We conducted the following sensitivity analyses to examine the robustness of our primary findings: (1) using a more stringent physical activity cutoff excluding days that registered <1,000 steps; (2) using PET data without PVC; (3) including only participants age <70 years to minimize potential confounding effects of advanced age; (4) adjusting for additional potential confounders and their interactions with time—season of physical activity measurement, total self-reported physical activity (CHAMPS questionnaire⁷¹), systemic vascular risk (office-based Framingham Heart Study cardiovascular disease risk score⁷²) and depressive symptoms (geriatric depression scale); (5) excluding participants who developed mild cognitive impairment (n = 5) or dropped out (n = 11) within the first 2 years of follow-up to reduce the potential for reverse causality; and (6) adjusting for baseline tau, PACC or CDR-SOB in their respective models to account for physical activity effects on baseline outcome measures.

We further used moderated mediation analysis (‘mediation’ package⁷³) to examine whether tau accumulation mediated the interactive effects of physical activity and Aβ on PACC5 decline or CDR-SOB progression. Individual slopes of ITC tau, PACC5 and CDR-SOB change were extracted from unadjusted linear mixed models with time as the only predictor, including both random slopes and intercepts. We conducted moderated mediation analysis with PACC5 or CDR-SOB slope as outcome, ITC tau slope as mediator, baseline Aβ (continuous) as moderator, and adjusted for age, sex, education and time interval between baseline Aβ and first tau scan. To examine for moderation effects by baseline Aβ, mediation models were run at both low and high levels of baseline Aβ, which were represented by the mean Aβ burden of Aβ-negative (PiB PVC-DVR = 1.17) and Aβ-positive (PiB PVC-DVR = 1.85) participants respectively, defined using the conventional Aβ-positivity threshold (PiB PVC-DVR = 1.324). Statistical testing was performed using a quasi-Bayesian Monte Carlo method based on 10,000 simulations to generate the estimates and 95% confidence intervals.

Lastly, we examined the dose–response relationships between the levels of physical activity and associations with Aβ-related tau accumulation and cognitive decline. Ordinal levels of physical activity were defined using cutoffs adapted from ref. ³⁰: inactive (≤3,000 steps per day), low activity (3,001 to 5,000 steps per day), moderate activity (5,001 to 7,500 steps per day) and active (≥7,501 steps per day). Using linear mixed effects models, we examined the interactive effects of baseline physical activity levels (ordinal) and Aβ (continuous) on longitudinal ITC tau, PACC5 and CDR-SOB, adjusting for the same covariates. To contextualize the effects of physical activity on cognitive decline, we used the model results to estimate predicted PACC5 and CDR-SOB trajectories across physical activity levels at representative levels of low versus high baseline Aβ burden, which were represented by the mean global Aβ burden of Aβ-negative (PiB PVC-DVR = 1.17) and Aβ-positive participants (PiB PVC-DVR = 1.85) respectively as described above. We calculated the predicted changes in PACC5 and CDR-SOB scores from baseline to year 9 (median duration of cognitive follow-up) and calculated the percent difference in decline across higher activity levels compared to inactive individuals. Lastly, we used literature-derived thresholds for cognitive impairment (lower than −1.5 PACC5 z-score³¹) and clinically meaningful functional decline in early AD (increase of 1.5 points on CDR-SOB³²) to estimate the time from baseline to these thresholds of cognitive and functional worsening for individuals with elevated Aβ across physical activity levels.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

References

Authors and Affiliations

1. Department of Neurology, Mass General Brigham, Boston, MA, USA

   Wai-Ying Wendy Yau, Dylan R. Kirn, Michael J. Properzi, Aaron P. Schultz, Zahra Shirzadi, Kailee Palmgren, Paola Matos, Courtney Maa, Stephanie A. Schultz, Rachel F. Buckley, Dorene M. Rentz, Keith A. Johnson, Reisa A. Sperling & Jasmeer P. Chhatwal

2. Harvard Medical School, Boston, MA, USA

   Wai-Ying Wendy Yau, Aaron P. Schultz, Zahra Shirzadi, Stephanie A. Schultz, Rachel F. Buckley, Dorene M. Rentz, Keith A. Johnson, Reisa A. Sperling & Jasmeer P. Chhatwal

3. Harquail Centre for Neuromodulation and Hurvitz Brain Sciences Program, Sunnybrook Research Institute, Toronto, Ontario, Canada

   Jennifer S. Rabin

4. Division of Neurology, Department of Medicine, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Ontario, Canada

   Jennifer S. Rabin

5. Rehabilitation Sciences Institute, University of Toronto, Toronto, Ontario, Canada

   Jennifer S. Rabin

6. Department of Neurology, Banner Alzheimer’s Institute, Phoenix, AZ, USA

   Jeremy J. Pruzin

7. Melbourne School of Psychological Sciences, University of Melbourne, Melbourne, Victoria, Australia

   Rachel F. Buckley

8. Department of Radiology, Mass General Brigham, Boston, MA, USA

   Keith A. Johnson

Contributions

W.-Y.W.Y., J.S.R., D.M.R., K.A.J., R.A.S. and J.P.C. conceived and/or designed the project. D.R.K., K.P., P.M., C.M., D.M.R., K.A.J., R.A.S. and J.P.C. contributed to the acquisition of data. W.-Y.W.Y., D.R.K., J.S.R., M.J.P., A.P.S., J.J.P., R.A.S. and J.P.C. contributed to data analysis. W.-Y.W.Y., J.S.R., Z.S., J.J.P., S.A.S., R.A.S., D.M.R., K.A.J., R.A.S. and J.P.C. contributed to interpretation of data. W.-Y.W.Y. drafted the initial manuscript. All authors provided substantive contributions to the revisions and approve the final version of the submitted manuscript.

Corresponding authors

Correspondence to Wai-Ying Wendy Yau or Jasmeer P. Chhatwal.

Competing interests

The authors declare no competing interests relevant to the current study. Potential conflicts of interest outside the submitted work are included below. J.J.P. has served as a consultant for Eisai. K.A.J. has served as a consultant for Novartis, Merck and Janssen. R.A.S. has served as a consultant for AbbVie, AC Immune, Acumen, Alector, Apellis, Biohaven, Bristol Myers Squibb, Genentech, Ionis, Janssen, Oligomerix, Prothena, Roche and Vaxxinity over the past 3 years. K.A.J. and R.A.S. have received research funding from Eisai and Eli Lilly for public–private partnership clinical trials but do not have any personal financial relationship with the companies. J.P.C. has served as a consultant for ExpertConnect.

Peer review information

Nature Medicine thanks Henry Brodaty, Susan Landau and Terry Therneau for their contribution to the peer review of this work. Primary Handling Editor: Jerome Staal in collaboration with the Nature Medicine team.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions