BDNF Val66Met polymorphism moderates associations between physical activity and neurocognitive outcomes in older adults

Abstract

INTRODUCTION

Increased brain‐derived neurotrophic factor (BDNF) release through physical activity (PA) is thought to underlie protective effects of PA on brain aging. The BDNF Val66Met single‐nucleotide polymorphism (rs6265) reduces activity‐dependent BDNF release and has been linked to early Alzheimer's disease (AD) pathology and cognition. We examined whether BDNF genotype influences the association of PA with plasma markers of AD, axonal degeneration, and neuroinflammation, along with consequences for cognition, in older adults without dementia.

METHODS

One hundred eighty older adults (M_age = 73.1; SD_age = 9.1; 61% female; 42% BDNF Met allele carriers) from the University of California San Francisco (UCSF) Memory and Aging Center completed 30 days of actigraphy monitoring, plasma assays of phosphorylated tau (p‐tau181), neurofilament light chain (NfL), glial fibrillary acidic protein (GFAP), and BDNF Val66Met genotyping. One hundred twenty‐three of the sample completed comprehensive neuropsychological evaluation. Habitual PA levels were operationalized via average daily step count. Composite z‐scores were calculated for cognitive domains of memory and executive functioning.

RESULTS

BDNF genotype moderated the relationship between PA and plasma p‐tau181, whereby higher PA was associated with lower plasma p‐tau181 concentration in Val/Val participants only. In moderated mediation analyses examining cognitive outcomes, plasma p‐tau181 selectively mediated the relationship between PA and executive function in Val/Val participants. In analyses including sex as a biological factor, there was a three‐way interaction of PA, BDNF genotype, and sex on plasma GFAP concentration, whereby higher PA was associated with lower plasma GFAP only in Val/Val male participants.

DISCUSSION

The Val/Val BDNF genotype may facilitate the neuroprotective relationships of PA, including lower AD‐relevant biology and better executive function. We further show there may be a sex‐specific negative relationship of PA with neuroinflammation in Val/Val males. These results further elucidate sources of individual variation observed in relationships between PA and brain health and will contribute to guiding personalized neurotrophic treatments for older adults.

Highlights

• Higher physical activity (PA) is associated with lower phosphorylated tau (p‐tau181) in brain‐derived neurotrophic factor (BDNF) Val66Met Val/Val carriers.

• In Val/Val carriers, p‐tau181 mediated the association of PA and executive function.

• There was a negative association of PA and glial fibrillary acidic protein (GFAP) in Val/Val male, but not female participants.

• The neuroprotective benefits of PA may be more pronounced in BDNF Val/Val carriers.

• There may be a sex‐specific PA pathway for neuroinflammation in Val/Val males.

In older adults, physical activity (PA) is associated with a reduced risk of dementia, ¹ lower levels of biomarkers associated with neurodegeneration, ² , ³ and better cognition. ⁴ However, randomized controlled trials of PA interventions for brain outcomes highlight notable inter‐individual variability in therapeutic responses to PA. ⁵ Delineating person‐specific factors and mechanisms underlying this variability is an important next step toward evidence‐based prescription of PA to reduce dementia risk in older age.

A major mechanistic hypothesis underlying the benefits of PA in the brain is increased release of brain‐derived neurotrophic factor (BDNF). ⁶ BDNF is a neurotrophin that supports neuron proliferation ⁷ and survival, ⁸ as well as functional and structural neuroplasticity. ⁹ BDNF Val66Met is a common single‐nucleotide polymorphism (SNP rs6265) associated with reduced activity‐dependent release of BDNF, ¹⁰ , ¹¹ and is linked to increased markers of neurodegeneration and steeper cognitive decline in individuals at risk of Alzheimer's disease (AD). ¹² , ¹³ To date, the potential impact of the BDNF Val66Met polymorphism on the relationship between PA and neurodegenerative markers remains unknown. As the dementia prevention field advances toward a precision medicine framework, understanding the impact of this common genetic variant will inform the need for consideration of BDNF genotype when evaluating PA interventions, and will contribute toward optimization of PA recommendations based on individual genetic differences.

The reduced availability of BDNF associated with BDNF Val66Met may impact the efficacy of interventions hypothesized to protect against neurodegeneration in aging, such as PA. ¹⁴ Peripheral BDNF levels are increased acutely in humans following a single session of exercise, with an increase in both the acute exercise‐related response and resting BDNF levels following repeated engagement in PA. ¹⁵ Only a few studies have examined the impact of BDNF Val66Met on this PA‐related boost in circulating BDNF in humans. Findings are inconsistent, with one study showing a trend toward larger increases in Met allele carriers, albeit without significant group differences, ¹⁶ whereas another study showed no effect. ¹⁷ In contrast, BDNF Val66Met associates with attenuated PA‐related benefits for brain‐related outcomes, including hippocampal and temporal lobe volume, ¹⁸ executive function, ¹⁹ memory, ²⁰ and cognitive trajectories over time. ¹⁹ In addition, an interaction between BDNF Val66Met and sex has been demonstrated. In one study, self‐reported PA was more strongly associated with cognitive performance in cognitively unimpaired older Val/Val males relative to male Met allele carriers and females. ²¹ In contrast, in a sample of older adults with vascular cognitive impairment, a 6‐month randomized controlled trial of aerobic training benefited cognition most in Val/Val females. ²² These studies suggest that BDNF availability and/or downstream function, including genetic influence on its quantal release by neurons, may be important for older adults to obtain optimal brain benefits of PA, and biological sex may impact these relationships.

It is unclear whether BDNF Val66Met influences the effects of PA on neurodegenerative processes. Furthermore, although previous studies support the notion that BDNF Val66Met may moderate the effects of PA on brain aging, most studies are limited by the use of self‐reported PA measures, which are prone to inflated estimation ²³ and show poorer construct validity than objective measures. ²⁴ Understanding the sources of individual variation observed in the relationship between PA and neurodegenerative markers is critical for developing evidenced‐based PA recommendations to reduce dementia risk in older adults. In this study of older adults without dementia, we used wearable actigraphy monitoring to determine whether BDNF Val66Met influences the association of PA with plasma markers of AD, axonal degeneration, and neuroinflammation, along with potential consequences for cognition.

RESEARCH‐IN‐CONTEXT

1. Systematic review: The authors reviewed literature using traditional sources (e.g., PubMed) and conference abstracts. Although evidence suggests the brain‐derived neurotrophic factor (BDNF) Val66Met polymorphism may attenuate the benefits of physical activity (PA) for some brain health outcomes in older adults (e.g., hippocampal volume), no studies have extended this to plasma biomarkers of neurodegeneration.

2. Interpretation: Our findings show that BDNF genotype moderates the association of PA with Alzheimer's disease pathological burden in functionally intact older adults, with consequences for executive function. Furthermore, we show sex‐specific associations with neuroinflammation observed in older males.

3. Future directions: Studies with the opportunity to incorporate central measures of BDNF protein and sex hormones will allow for investigation of mechanisms driving the moderating role of BDNF genotype and interactions with sex. Furthermore, future work should seek to optimize PA parameters (e.g., type, intensity) to maximize neuroprotective effects in BDNF Val66Met carriers and to guide person‐specific PA recommendations.

1. METHODS

1.1. Participants

Participants included 180 older adults without dementia enrolled in the University of California San Francisco (UCSF) Memory and Aging Center's Brain Aging Network for Cognitive Health study (BrANCH). All participants had a Clinical Dementia Rating (CDR) global score of 0 (n = 168) or 0.5 (n = 12). Individuals in the study cohort were selected for inclusion based on availability of BDNF Val66Met (rs6265) genotyping, blood draws with plasma assayed via Quanterix Simoa, and 30‐days of Fitbit actigraphy monitoring. One hundred twenty‐three participants had available cognitive data obtained via comprehensive neuropsychological evaluation. The study was approved by the UCSF Human Research Protection Program and Institutional Review Board and was conducted in accordance with the latest Declaration of Helsinki, including written informed consent from all participants.

1.2. Measures

1.2.1. Genotyping

Genomic DNA was extracted from peripheral blood using standard protocols (Gentra PureGene Blood Kit, Qiagen). BDNF Val66Met rs6265 was genotyped using the Illumina Omni 2.5 Exome genotyping array (n = 77, Illumina Inc., San Diego, CA), Sequenom iPLEX Technology (n = 67, Sequenom, San Diego, CA), or TaqMan genotyping assay (n = 35, Thermo Fisher Scientific Inc., Waltham, MA). ²⁵ BDNF genotype distribution across participants was 105 (58.3%) Val/Val, 64 (35.6%) Val/Met, and 11 (6.1%) Met/Met, consistent with Hardy‐Weinberg equilibrium (χ² = 0.09, p = .765).

1.3. Plasma protein quantification

Venous blood was collected in tubes containing ethylenediaminetetraacetic acid (EDTA), and plasma samples were stored in 0.5 mL polypropylene tubes at −80°C. For analyses, plasma samples (one thawing only) were gradually brought to room temperature. The ultrasensitive HD‐X analyzer by Quanterix (Lexington, MA) was used for quantification of plasma proteins including phosphorylated tau at amino acid 181 (p‐tau181), neurofilament light chain (NfL), and glial fibrillary acidic protein (GFAP). GFAP and NfL were measured via multiplex single molecule arrays (Simoa, Quanterix Neurology 4‐Plex A), whereas p‐tau181 was measured using single analyte assays (Simoa, Quanterix). Lower limit of detection (LLOD) was 0.09 pg/mL for p‐tau181, 0.09 pg/mL for NfL, and 0.441 pg/mL for GFAP. All samples were >LLOD. All analyses were performed in duplicate, according to the manufacturer's published protocols and by researchers blinded to clinical data. Samples with coefficients of variance >20% were excluded from analyses (p‐tau181 = 18; NfL = 3; GFAP = 1). Data normality assessment confirmed that plasma biomarker distributions showed positive skew. Variables were transformed using a square root transformation, and all statistical models were run with the transformed variables. When compared to models using the raw data, results remained unchanged. Models with the raw data are presented to preserve the scale of the data and improve the interpretability of the findings. Assessment for outliers was conducted using a threshold of 12 median absolute deviations (MADs) from the median, used in previous studies to limit plasma biomarker data removal to extreme values. ²⁶ No outliers were identified.

1.4. Actigraphy monitoring

The Fitbit Flex2 (n = 165) or Inspire2 (n = 15) (Fitbit Inc., San Francisco, CA, USA; https://www.fitbit.com) monitored daily step count. Participants were blinded to all notifications and indication of their step count for the 30‐day monitoring period. They were instructed to wear the Fitbit during all hours (Inspire2) or all waking hours (Flex2). Individual days with fewer than 100 steps were removed from the analyses to control for nonadherence. All participants had at least 7 days of valid monitoring data. The primary outcome was average daily steps, calculated by taking the daily step count average across all days with sufficient data (maximum of 30 days). To increase the interpretability of analyses, average daily step count was divided by 1000. Using a criterion of 3 SD from the mean, one statistical outlier was identified (22,765 steps/day). Inclusion of these data point did not alter the results of the statistical models, and therefore, the data were retained in the sample. Fitbit model (Flex2, mean step count = 8005 [SD = 3605] vs Inspire2, mean step count = 7953 [SD = 3268]) was not a significant predictor of any plasma biomarker when included in the regression models (all p > .380), and step count data between the two models showed strong concordance (Figure S1).

1.5. Neuropsychological evaluation

One hundred twenty‐three participants in the sample completed a standardized battery of neuropsychological tests that were validated as neuroanatomically sensitive to age‐related neurodegeneration. ²⁷ Sample‐based z‐scores were calculated using normative data from the larger UCSF BrANCH cohort (all CDR scores = 0; n > 650 per test). To quantify memory and executive functioning in our sample, z‐scores for performance on several tests within each cognitive domain were averaged together to create composite z‐scores previously published by our group. ²⁸

Episodic memory. The episodic memory composite score included sub‐tests from the California Verbal Learning Test–Second Edition (CVLT‐II) (total immediate recall of a 16‐item list over five learning trials, long delay [20‐min] free recall total, and recognition discriminability [d’]) and total recall of the Benson figure following a 10‐min delay. ²⁷ Composite scores for nine participants were unavailable due to incomplete memory testing.

Executive functioning. Executive functioning measures included total number of items recalled (“60”) from the Stroop interference task, ²⁹ number of digits recalled in backwards order (Wechsler Adult Intelligence Scale, Forth Edition, Digit Span Backwards ³⁰ ), total completion time (seconds) for a modified version of the Trail Making Test (Part B), ²⁷ total number of words that start with “D” in 60 s from a phonemic fluency task, ²⁷ and total designs generated in 60 s from the Delis‐Kaplan Executive Function System (D‐KEFS) Design Fluency, Condition 1.

The collection of key study variables (blood biomarkers, cognitive assessment) was conducted an average of 3.35 days (SD = 9.96) before the beginning of the actigraphy monitoring period.

1.6. Statistical analyses

All analyses were conducted using JASP version 0.18.2, and an alpha level of 0.05. Given the small sample of BDNF Met/Met carriers, participants with at least one BDNF Met allele were combined into a Met allele carrier group. BDNF genotype (Val/Val vs Met allele carrier) and sex differences on demographic and clinical characteristics were examined using two‐tailed independent samples t‐tests. All regression models controlled for age, sex, apolipoprotein E (APOE) ε4 allele carrier status (yes/no) and self‐reported race/ethnicity. Analyses involving cognitive outcomes additionally controlled for education. All mediation analyses described below were conducted with bootstrapped bias‐corrected 95% confidence intervals (CIs; 5000 bootstrapped samples) for assessment of the indirect effect. ³¹

To determine the interactive effects of PA (average daily steps) and BDNF genotype on plasma markers, multiple linear regression analyses separately modeled p‐tau181, NfL, and GFAP as a function of PA, BDNF genotype, and their interaction. Models with significant PA by BDNF genotype interactions were followed up with a priori planned comparisons examining the relationship between PA and the plasma marker, stratified by BDNF genotype. Given previous reports of sex differences in associations of PA with neurobehavioral outcomes and by BDNF genotype, multiple regression analyses for p‐tau181, NfL, and GFAP were repeated with inclusion of sex and its three‐way interaction with BDNF genotype and PA. Significant three‐way interactions were followed up with a priori planned multiple regression analyses that separately modeled the plasma marker as a function of PA, BDNF genotype, and their interaction, stratified by sex. As above, models with significant PA by BDNF genotype interactions were then further probed by examining the relationship between PA and the plasma marker, stratified by BDNF genotype.

For each significant two‐way interaction (PA x BDNF genotype), moderated mediation analyses determined whether the association of PA (predictor) with plasma biomarkers (mediator) had downstream consequences for cognitive outcomes, and whether this pathway was dependent on BNDF genotype (moderator). For each significant three‐way interaction (PA x BDNF genotype x Sex), moderated mediation analyses with the same structure described above were conducted stratified by sex. Models with a significant index of moderated mediation were further examined with a priori planned mediation analyses examining the plasma marker as a statistical mediator of the relationship between PA and cognition, stratified by BDNF genotype.

1.7. Sensitivity analysis

Given that the Met allele prevalence is higher in East Asian ancestry populations compared to European populations, sensitivity analyses were conducted by excluding all participants who self‐reported Asian identity and re‐running the primary analyses. Our findings remain largely unchanged, showing patterns in the same direction as in the full cohort (Table S1, Table S2).

2. RESULTS

Demographic and clinical characteristics of the study sample are presented in Table 1. There were no statistically significant group differences between Val/Val and Met allele carriers on age, education, average daily steps, cognition, or plasma biomarkers (all p’s > .062; see Table 1). Although there was no significant between‐group difference on memory score, the Val/Val group trended toward better memory performance compared to Met allele carriers, consistent with prior studies ³² (p = .062).

TABLE 1.

Study sample demographic and clinical characteristics.

	Full sample	Val/Val	Met carriers	Independent‐sample t‐test
Total n	100% (180)	58.3% (105)	41.7% (75)	–
Sex, % female	60.5% (109)	61.9% (65)	58.7% (44)	–
Race
White	85.6% (154)	88.6% (93)	81.3% (61)	–
Black	1.7% (3)	2.9% (3)	0.0% (0)	–
Asian	11.1% (20)	6.7% (7)	17.3% (13)	–
Other	1.7% (3)	1.9% (2)	1.3% (1)	–
Global CDR, % score of 0.5	6.7% (12)	7.6% (8)	5.3% (4)	–
Fitbit model, % Inspire2	8.3% (15)	6.7% (7)	10.6% (8)	–
Age (years)	73.13 (9.10)	72.54 (8.74)	73.95 (9.58)	t(178) = −1.02, p = 0.309
Education (years)	17.53 (1.87)	17.51 (1.88)	17.57 (1.86)	t(178) = −0.24, p = 0.809
Valid days of step count data	29.89 (5.56)	30.44 (5.80)	29.12 (5.14)	t(178) = 1.57, p = 0.119
Daily step average (thousands)	7.93 (3.51)	7.83 (3.14)	8.07 (3.98)	t(178) = −0.45, p = 0.651
Memory (z‐score; n = 114) *	0.01 (0.79)	0.11 (0.75)	−0.16 (0.83)	t(119) = 1.88, p = 0.062
Executive function (z‐score; n = 123) *	0.18 (0.72)	0.20 (0.72)	0.07 (0.67)	t(129) = 1.04, p = 0.300
Glial fibrillary acidic protein (GFAP; n = 179; pg/mL)	175.64 (99.10)	170.84 (97.31)	182.30 (101.81)	t(177) = −0.76, p = 0.447
Neurofilament light chain (NfL; n = 177; pg/mL)	29.95 (18.13)	28.94 (17.29)	31.33 (19.26)	t(175) = −0.86, p = 0.390
Phosphorylated tau‐181 (p‐tau181; n = 162; pg/mL)	3.72 (2.16)	3.79 (2.27)	3.63 (1.20)	t(160) = 0.46, p = 0.645

Note: Mean (SD) or % (n) reported. CDR: Clinical Dementia Rating

Abbreviation: CDR, clinical dementia rating.

^*z‐scores were calculated based on the larger BrANCH cohort of healthy older adults.

2.1. BDNF genotype and physical activity

Multiple linear regression analyses (Table 2) revealed a significant PA by BDNF genotype interaction for plasma p‐tau181 (β  = 0.19, p = 0.033), but not for plasma NfL (β = 0.62, p = 0.377) or plasma GFAP (β  = 1.06, p = 0.773). In analyses stratified by BDNF genotype, higher PA was associated with lower plasma p‐tau181 concentrations in Val/Val participants (β  = 0.19, p = 0.017), but not in Met allele carriers (β = 0.02, p = 0.762) (Figure 1). Moderated mediation analyses examining cognitive outcomes were significant for executive function (nVal/Val = 71, nMet = 52; index of mod‐med = –0.04, 95% CI: –0.09 to –0.003), but not for memory (nVal/Val = 68, nMet = 46; index of mod‐med = −0.02, 95% CI: –0.09 to 0.01). Mediation analyses stratified by BDNF genotype found that for Val/Val participants only, there was a significant indirect effect of PA on executive function, mediated by plasma p‐tau181 concentration (41.5% of variance explained; indirect effect, β = 0.02, 95% CI: 0.003 to 0.045; Figure 2). The total effect of PA on executive function was statistically significant for Met allele carriers (β = 0.05, p = 0.047), and approached statistical significance for Val/Val participants (β = 0.05, p = 0.058). When participants with a CDR of 0.5 were excluded from the sample, the pattern of results remained unchanged.

TABLE 2.

Interactive effects of PA and BDNF genotype on plasma markers.

	p‐tau181		NfL		GFAP
Predictor	β [95% CI]	p	β [95% CI]	p	β [95% CI]	p
PA	−0.18 [−0.31, ‐0.04]	0.012	−1.01 [−2.10, 0.09]	0.071	−4.06 [−9.68, 1.57]	0.156
BDNF (ref: Val/Val)	−1.71 [−3.19, ‐0.22]	0.024	−3.37 [−15.50, 8.76]	0.584	−1.50 [−64.63, 61.62]	0.962
Age	0.06 [0.02, 0.10]	0.002	0.78 [0.48,1.08]	<0.001	4.87 [3.30, 6.45]	<0.001
Sex (ref: male)	−1.50 [−2.11, ‐0.89]	<0.001	1.66 [−3.34, 6.66]	0.513	28.04 [1.97, 54.11]	0.035
APOE (ref: ε4 non‐carrier)	0.37 [−0.28, 1.03]	0.263	1.48 [−3.97, 6.93]	0.592	1.76 [−26.87, 30.39]	0.904
Ethnicity (ref: White)	−0.06 [−0.17, 0.05]	0.289	−0.63 [−1.57, 0.31]	0.186	−2.24 [−7.18, 2.70]	0.372
PA*BDNF	0.19 [0.02, 0.36]	0.033	0.62 [−0.77, 2.02]	0.377	1.06 [−6.19, 8.31]	0.773

Note: Unstandardized coefficients are reported. Analyses controlled for age, sex, APOE ε4 carrier status, and self‐reported ethnicity. p values in bold represent statistical significance of p < .05.

Abbreviations: APOE, apolipoprotein E gene; BDNF, brain‐dervived neurotrophic factor gene; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PA, physical activity; p‐tau181, phosphorylated tau at amino acid 181.

FIGURE 1.

(A) Two‐way interaction of BDNF genotype (Val/Val vs Met allele carrier) and physical activity (PA; average daily steps) on plasma p‐tau181 concentration (n = 162). Higher PA was associated with lower plasma p‐tau181 concentration in Val/Val participants only. The interaction of BDNF genotype and PA was not statistically significant for (B) plasma NfL concentration (n = 177) or (C) plasma GFAP concentration (n = 179). Analysis adjusted for age, sex, APOE ε4 carrier status, and self‐reported race/ethnicity. Shaded error bands represent 95% confidence intervals. *p < .05. **significant BDNF genotype by PA interaction. GFAP, glial fibrillary acidic protein, NfL, neurofilament light chain; p‐tau181, phosphorylated tau at amino acid 181.

FIGURE 2.

A moderated mediation analysis (n = 123) demonstrated an indirect positive effect of PA on executive function through plasma p‐tau181 for Val/Val, but not for Met allele carriers. Analyses controlled for age, sex, education, APOE ε4 carrier status, and self‐reported race/ethnicity. Bootstrapped bias‐corrected 95% confidence intervals were used for interpretation of indirect effects. BDNF, brain‐derived neurotrophic factor; p‐tau181, phosphorylated tau at amino acid 181.

2.2. Sex differences

Given that sex differences are reported increasingly for both AD biomarkers and PA, ³³ and may influence the effects of BDNF Val66Met on PA‐related outcomes, ²¹ we examined how biological sex influenced the primary models. Demographic and clinical characteristics broken down by both BDNF genotype and sex are available in Table S3. There were no statistically significant group differences between male and female participants on proportion of Met allele carriers, age, Fitbit parameters, plasma GFAP, plasma NfL, or executive function z‐scores (all p’s > .087). Male participants had more years of education (p = 0.003), higher plasma p‐tau181 level (p ≤ .001), and lower memory z‐scores (p = 0.037) compared to female participants. Multiple linear regression analyses (Table 3) revealed a three‐way interaction of PA, BDNF genotype, and sex on plasma GFAP concentration (β = −24.58, p ≤ .001), but not on plasma p‐tau181 (β = −0.12, p = 0.516) or NfL (β = −1.88, p = 0.213). In analyses stratified by sex (Figure 3), there was a significant PA by BDNF genotype interaction for male (β = 16.82, p = 0.002) but not for female (β = −8.70, p = 0.091) participants. Specifically, higher PA was associated with lower plasma GFAP concentration in Val/Val males (β = −15.76, p = 0.007), whereas the association between PA and GFAP was attenuated in male Met allele carriers (β = 0.62, p = 0.816). Moderated mediation analyses examining GFAP on cognitive outcomes did not reach statistical significance for females (memory z‐scores: nVal/Val = 44, nMet = 27, index of mod‐med = –0.14, 95% CI: –2.94 to 0.75; executive function z‐scores: nVal/Val = 45, nMet = 31, index of mod‐med = 0.06, 95% CI: –0.92 to 2.22) or males (memory z‐scores: nVal/Val = 28, nMet = 22, index of mod‐med = 0.07, 95% CI: –2.38 to 5.35; executive function z‐scores: nVal/Val = 30, nMet = 24, index of mod‐med = −1.10, 95% CI: –3.09 to 0.45).

TABLE 3.

Interactive effects of PA, BDNF genotype, and sex on plasma markers.

	p‐tau181		NfL		GFAP
Predictor	β [95% CI]	p	β [95% CI]	p	β [95% CI]	p
PA	−0.34 [−0.59, ‐0.08]	0.011	−1.22 [−3.22, 0.78]	0.231	−13.31 [−23.09, ‐3.53]	0.008
BDNF (ref: Val/Val)	−2.66 [−5.24, ‐0.09]	0.043	−14.75 [−35.67, 6.17]	0.166	−151.48 [−254.22, ‐48.73]	0.004
Sex (ref: male)	−3.23 [−5.68, ‐0.78]	0.010	−2.13 [−22.30, 18.05]	0.835	−90.95 [−189.51, 7.61]	0.070
Age	0.06 [0.02, 0.09]	0.003	0.74 [0.44, 1.05]	<.001	4.48 [2.93, 6.01]	<.001
APOE (ref: ε4 non‐carrier)	0.33 [−0.33, 0.99]	0.325	1.66 [−3.79, 7.11]	0.549	1.31 [−26.50, 29.13]	0.926
Ethnicity (ref: White)	−0.07 [−0.19, 0.04]	0.200	−0.55 [−1.49, 0.40]	0.257	−2.16 [−6.99, 2.68]	0.380
PA*BDNF	0.30 [−0.01, 0.60]	0.046	1.59 [−0.75, 3.92]	0.182	15.68 [4.13, 27.23]	0.008
PA*Sex	0.21 [−0.08, 0.51]	0.149	0.13 [−2.19, 2.45]	0.911	11.90 [0.50, 23.30]	0.041
BDNF*Sex	1.01 [−2.22, 4.25]	0.537	21.29 [−4.98, 47.57]	0.112	249.93 [118.89, 380.96]	<.001
PA*BDNF*Sex	−0.12 [−0.50, 0.25]	0.516	−1.88 [−4.85, 1.09]	0.213	−24.58 [−39.46, ‐9.70]	0.001

Note: Unstandardized coefficients are reported. Analyses controlled for age, APOE ε4 carrier status and self‐reported ethnicity. p values in bold represent statistical significance of p < .05.

Abbreviations, APOE, apolipoprotein E gene; BDNF, brain‐derived neurotrophic factor gene; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PA, physical activity; p‐tau181, phosphorylated tau at amino acid 181.

FIGURE 3.

Three‐way interaction between Sex (male, female), BDNF genotype (Val/Val, Met allele carrier), and physical activity (PA; average daily steps) on plasma GFAP concentration (n = 179; p < .001). (A) In male participants, there was a significant BDNF genotype by PA interaction. PA was negatively associated with plasma GFAP concentration in Val/Val participants but not in Met allele carriers. (B) There was no significant association between PA and plasma GFAP concentration, or interaction between PA and BDNF genotype, in female participants. All analyses adjusted for age, APOE ε4 carrier status, and self‐reported race/ethnicity. Shaded error bands represent 95% confidence intervals. *p < .05; **significant BDNF genotype by PA interaction. GFAP, glial fibrillary acidic protein.

3. DISCUSSION

We investigated whether the BDNF Val66Met single‐nucleotide polymorphism influences the association between PA and plasma markers of neurodegeneration in older adults without dementia. We further examined the potential influence of these relationships for cognitive outcomes. We found that higher PA was associated with lower plasma p‐tau181 concentration for Val/Val participants, but not for Met allele carriers. Furthermore, plasma p‐tau181 statistically mediated the relationship between PA and executive function for Val/Val participants only. In analyses including sex, there was a negative association between PA and plasma GFAP for male Val/Val participants only. This did not associate with cognitive outcomes; however, statistical power for cognition models was limited. These findings suggest that the BDNF Val/Val genotype may facilitate the neuroprotective relationships of PA, including associations with lower AD pathological burden and better executive function, along with a sex‐specific association with lower neuroinflammation for males.

The presence of an association between PA and p‐tau181 in BDNF Val/Val participants and not in Met allele carriers is consistent with previous studies that suggest that BDNF Val66Met may attenuate the benefits of PA for other brain‐related outcomes. ¹⁸ Reasons for this are likely two‐fold. First, increases in BDNF protein associated with PA may be larger in Val/Val compared to Met allele carriers. ³⁴ Second, the BDNF Met allele has been associated with reduced efficacy of BDNF activity (e.g., intracellular packaging and vesicle release ¹¹ ). Given the role of BDNF in supporting neuronal survival and plasticity, ⁸ , ⁹ greater PA‐related BDNF abundance and efficacy may translate to greater resilience against pathological burden in Val/Val relative to Met allele carriers. For example, a recent study showed that higher mid‐life circulating BDNF was associated with lower late‐life tau burden in healthy older adults, consistent with our p‐tau181 finding. ³⁵ It is also possible that adequate BDNF activity is necessary to counteract the acute stress effects of PA, including increased cortisol levels and oxidative stress. ³⁶ , ³⁷ , ³⁸ Our data suggest that these associations may be specific to AD pathophysiology, as we did not observe associations between PA and plasma NfL concentration, a non‐specific marker of axonal degeneration, ³⁹ in either genotype. In contrast, p‐tau181 is a specific marker of AD pathology and is particularly sensitive to pre‐clinical stages of disease, similar to the composition of our aging study cohort. ⁴⁰

Analyses examining the interplay between PA, BDNF Val66Met, and plasma biomarkers on cognitive outcomes suggested that in the Val/Val group, PA was positively associated with executive function indirectly through lower p‐tau181 concentrations. Therefore, the influence of BDNF Val66Met on the relationship between PA and AD pathological burden may have a meaningful association with clinical outcomes in Val/Val individuals. Although the mediating effect of p‐tau181 was absent in Met allele carriers, there was still a positive association between PA and executive function in this group. This suggests that although lower AD pathology may be a pathway through which PA associates with cognition in Val/Val individuals, Met allele carriers still show positive associations of PA and cognition that are likely mediated through other pathways. Studies that replicate this finding are necessary to better understand the clinical relevance of BDNF genotype for biomarker changes.

Our results further suggest there may be a sex‐specific benefit of PA, such that Val/Val males who engage in higher levels of PA showed lower markers of neuroinflammation measured via GFAP. Produced by reactive astrocytes, GFAP is upregulated during an immune response to injury or neuropathological change. ⁴¹ Previous work from our group has shown that PA may have selective relationships with lower peripheral cytokine markers in older adult males relative to females. ⁴² Reasons for this may involve sex differences in immune function in both the central and peripheral nervous systems. ⁴³ Astrocyte function is also influenced by its interaction with sex chromosome expression and gonadal hormones, ⁴⁴ and astrocytes of males and females show divergent reactive responses to injury and other pathological changes (e.g., ⁴⁵ ). Given the key role of PA for immune homeostasis, ⁴⁶ a possible explanation for our finding is that males have greater need and/or capacity for immune‐related plasticity compared to the relatively more robust immune responses shown in females. ⁴³ Alternately, estrogen has been shown to directly influence BDNF signaling. ⁴⁷ Females show higher basal circulating levels of BDNF compared to males. ⁴⁸ However, in the context of exercise interventions, there may be larger acute increases in peripheral BDNF in males, ⁴⁹ but larger longer‐term basal increases in BDNF in females following PA engagement. ³³ It is unclear whether females gain less benefit from PA, or whether higher basal BDNF levels in females provide a greater level of protection against neurodegeneration that males may need to obtain via other routes, such as engagement in PA. Indeed, there is mixed evidence, ³³ , ⁴² and further research is needed to better understand sex‐specific benefits and mechanisms linking PA to brain health.

Our results inform mechanisms underlying inter‐individual variability in therapeutic responses to PA in older adults. Although previous studies support an association between PA and reduced dementia risk, ¹ the literature examining the relationship between PA and AD biomarkers continues to be mixed (for systematic review see ⁵⁰ ). Reasons for this are likely to include sample and design heterogeneity, whereby individual differences influence the extent of brain health benefits gained from PA interventions. We demonstrate that BDNF Val66Met genotype and sex are important moderating factors to consider when assessing the relationship between PA and neurodegenerative disease biomarkers, which may include the efficacy of PA‐based brain health interventions. Although the evidence reported in this study suggests that individuals with the Val/Val genotype may gain the most benefit from PA in protection against early AD pathology, PA is still likely to benefit healthy brain aging in Met allele carriers via other pathways, such as cardiovascular risk ⁵¹ and mood. ⁵² Future research should seek to understand how PA might be optimized for Met allele carriers (e.g., type, intensity, duration of PA) to maximize its benefits.

The study conclusions are limited by several factors. First, although the physiological consequences of BDNF Val66Met are well established for neuronal BDNF release and signaling, we did not obtain a direct measure of BDNF levels. The influence of PA on BDNF levels in the brain is not yet well established in humans, and BDNF shows poor penetrance of the blood–brain barrier, ⁵³ thereby limiting the use of plasma as a direct proxy for BDNF abundance in the brain. Given that increased PA‐related BDNF is hypothesized as the key driver behind the moderating effect of BDNF Val66Met on the relationship between PA and brain health outcomes, the inclusion of a measure of central nervous system BDNF (e.g., via cerebrospinal fluid or positron emission tomography [PET] ligand) would allow this hypothesis to be more thoroughly investigated. Similarly, despite the known relevance of sex hormones for BDNF activity, we did not obtain sex hormone levels in the sample, and we did not collect information on menopausal status or use of hormone‐replacement therapy for female participants. In particular, estrogen appears to play a key role in moderating the benefits of PA in the female brain. ⁴⁷ Future work using longitudinal capture of sex hormone variability and how this may be implicated in our current findings is needed. Furthermore, the small proportion of BDNF Met/Met carriers within the sample did not allow for the examination of potential differences between carriers of one versus two Met alleles. Future studies with larger sample sizes should seek to determine the potential dose‐dependent effects of the Met allele. Similarly, moderated mediation models that stratified by both BDNF genotype and sex were likely underpowered for cognitive models. Future studies with larger samples are needed to further examine this question. The analyses did not account for medical comorbidities that can influence blood biomarker concentrations (e.g., chronic kidney disease) and step count measurement (e.g., mobility impairments). In addition, wrist actigraphy monitoring may be less accurate for certain types of PA (e.g., swimming). Although this study was restricted to investigation of average daily PA, future studies that have access to other PA metrics (e.g., time in moderate‐to‐vigorous PA) will be critical to guide clinical recommendations of the specific PA characteristics most relevant for brain health. In addition, other phosphorylated tau isoforms (e.g., p‐tau217) may be more sensitive to AD pathology and could be assessed in future studies. The study sample was largely White with a high level of education, potentially limiting the generalizability of the study findings. Finally, the cross‐sectional, observational design of this study limits claims regarding causality. Although the use of genetic data mitigates some limitations regarding interpretation of directionality effects, confirming these mechanistic relationships between PA and healthy brain aging will require future longitudinal and randomized controlled trials.

Together our findings suggest that the BDNF Val/Val genotype may facilitate the protective effects of PA against markers of neurodegeneration in aging. These effects include associations with lower AD pathological burden and better executive function. In addition, we show that there may be a sex‐specific relationship between PA and lower neuroinflammation in older Val/Val males. These results assist in further elucidating sources of individual variation in the extent of brain‐related benefits obtained from engagement in PA, and further emphasize the importance of examining these factors when assessing the efficacy of PA interventions. Furthermore, as the field of precision medicine for dementia prevention continues to make progress, our results will contribute to guiding precise neurotrophic interventions to support healthy brain aging.

CONFLICT OF INTEREST STATENENT

None. Author disclosures are available in the supporting information.

CONSENT STATEMENT

This study was approved by the UCSF Human Research Protection Program and Institutional Review Board. Written informed consent was obtained from all participants.

Supporting information

Supporting Information

Supporting Information

Cadwallader CJ, VandeBunte AM, Fischer DL, et al. BDNF Val66Met polymorphism moderates associations between physical activity and neurocognitive outcomes in older adults. Alzheimer's Dement. 2025;11:e70106. 10.1002/trc2.70106

DATA AVAILABILITY STATEMENT

De‐identified data from this report will be made available on request from any qualified investigator (https://memory.ucsf.edu/research‐trials/professional/open‐science#Data‐Sharing). UCSF Memory and Aging Center data requests can be sent to the corresponding author.