Physical activity and risk of Alzheimer disease

Abstract

Objective

Evidence from observational studies for the effect of physical activity on the risk of Alzheimer disease (AD) is inconclusive. We performed a 2-sample mendelian randomization analysis to examine whether physical activity is protective for AD.

Methods

Summary data of genome-wide association studies on physical activity and AD were used. The primary study population included 21,982 patients with AD and 41,944 cognitively normal controls. Eight single nucleotide polymorphisms (SNPs) known at p < 5 × 10⁻⁸ to be associated with average accelerations and 8 SNPs associated at p < 5 × 10⁻⁷ with vigorous physical activity (fraction of accelerations >425 milligravities) served as instrumental variables.

Results

There was no association between genetically predicted average accelerations with the risk of AD (inverse variance weighted odds ratio [OR] per SD increment: 1.03, 95% confidence interval 0.97–1.10, p = 0.332). Genetic liability for fraction of accelerations >425 milligravities was unrelated to AD risk.

Conclusion

The present study does not support a causal association between physical activity and risk of AD.

Alzheimer disease (AD) is the main cause of dementia and one of the great health care challenges of the 21st century.¹ Research since the discoveries of β-amyloid and tau, the main components of plaques and tangles, has provided considerable knowledge about molecular pathways of AD development; however, this knowledge has not yet been translated into the implementation of effective prevention measures for modifiable risk factors of AD.² Considerable research has focused on the potentially protective role of physical activity for AD. Several meta-analyses of observational studies suggested a protective effect of physical activity for cognitive decline and risk of dementia and AD.^3–10 Also, intervention studies have shown that exercise may improve cognitive performance.^4,11

Randomized trials have not revealed changes in the risk of dementia or AD through exercise interventions.^4,11 More recently, long-term observational studies have suggested that the inverse association between physical activity and dementia might be subject to reverse causation due to a decline in physical activity during the preclinical phase of dementia.^12–14 Mendelian randomization is a method that uses genetic variants as instrumental variables to uncover causal relationships in the presence of reverse causation and unobserved confounding.¹⁵ In the current study, we performed 2-sample summary data mendelian randomization analyses to provide evidence for the association between accelerometer-assessed physical activity and AD.

Methods

The mendelian randomization study design had 3 components: (1) identification of genetic variants to serve as instrumental variables for accelerometer-assessed physical activity; (2) the acquisition of summary data for the genetic instruments from genome-wide association studies (GWAS) on accelerometer-assessed physical activity; (3) acquisition and harmonization of instrumenting single nucleotide polymorphism (SNP)-outcome summary data for the effect of genetic instruments from GWAS on the risk of AD.

Physical activity measurement in the UK Biobank

A GWAS of the UK Biobank identified SNPs associated with objectively measured physical activity from accelerometers.¹⁶ The UK Biobank study is a community-based prospective cohort study that recruited over 500,000 men and women aged 40–69 years from different socioeconomic backgrounds from 22 centers across the United Kingdom between 2006 and 2010.¹⁷ A subset of 103,712 participants wore an Axivity AX3 wrist-worn triaxial accelerometer on their dominant hand at all times over a 7-day period between 2013 and 2015.¹⁸ It was set to capture triaxial acceleration data at 100 Hz with a dynamic range of ±8 g. This device demonstrated equivalent signal vector magnitude output on multi-axis shaking tests to the GENEActiv accelerometer, which has been validated using both standard laboratory and free-living energy expenditure.^18,19

We used genetic variants proxying 2 accelerometer-based physical activity measures: average accelerations (mean acceleration in milligravities) and the fraction of accelerations >425 milligravities,^16,18 the latter corresponding to an equivalent of vigorous physical activity (≥6 metabolic equivalent tasks [METs]).¹⁶ Participants who were more likely to consent to wearing the accelerometer were women, aged 55–74 years, and those with higher socioeconomic status, better physical health status, and less time since baseline assessment (table 1). The Spearman coefficient for the correlation of METs estimated from the International Physical Activity Questionnaire and average accelerations was 0.24 in men and 0.22 in women.²⁰

Table 1.

Demographic characteristics in genome-wide association studies

Instrumental variables for accelerometer-assessed physical activity

We selected 8 SNPs associated with average accelerations and 8 SNPs associated with fraction of accelerations >425 milligravities (vigorous physical activity) at a genome-wide significance level (p < 5 × 10⁻⁸) and suggestive significance level (p < 5 × 10⁻⁷), respectively, using a PLINK-clumping algorithm (r² threshold = 0.001 and window size = 10 Mb) from a genome-wide study of 91,084 UK Biobank participants of self-reported European descent.¹⁶ UK Biobank participants were genotyped using the UK BiLEVE array and the UK Biobank axiom array. Genotype imputation to a reference set combining the UK10K haplotype and Haplotype Reference Consortium reference panels was performed using IMPUTE2 algorithms.¹⁹ Table 2 presents the harmonized instruments.

Table 2.

Genome-wide significant single nucleotide polymorphisms (SNPs) for accelerometer-based physical activity

GWAS summary statistics for AD

Genetic variants associated with late-onset AD were obtained from meta-analysis of 4 previously published GWAS datasets.²¹ In stage 1, a meta-analysis of 46 studies including 21,982 clinically or autopsy-confirmed AD cases and 41,944 cognitively normal controls from European descent from 4 consortia including the Alzheimer Disease Genetics Consortium (ADGC), Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE), the European Alzheimer's Disease Initiative (EADI), and the Genetic and Environmental Risk in Alzheimer's Disease/Defining Genetic, Polygenic and Environmental Risk for Alzheimer's Disease Consortium (GERAD/PERADES) was performed. Datasets were imputed to the 1,000 Genomes Project using SHAEPEIT/IMPUTE2 or MACH/Minimac.²¹ All diagnoses were autopsy-confirmed or satisfied the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria or DSM-IV guidelines.²¹ The mean age at AD onset ranged from 74.4 to 81.9 years, and the average age at examination for 83% of controls was ≥76 years (table 1). Table 3 provides associations of genome-wide significant harmonized SNPs for accelerometer-based physical activity with AD. We conducted a replication analysis using summary statistics from another GWAS for AD including AD-by-proxy based on parental diagnoses.²² The summary statistics were based on a meta-analysis of a total of 71,880 cases and 383,378 controls (table 1).

Table 3.

Association of genome-wide significant single nucleotide polymorphisms (SNPs) for accelerometer-based physical activity with Alzheimer disease (AD) in the primary genome-wide association studies for AD²

Standard protocol approvals, registrations, and patient consents

Written informed consent was obtained from UK Biobank study participants and ethics approval of UK Biobank was given by the North West Multicentre Research Ethics Committee, the National Information Governance Board for Health & Social Care, and the Community Health Index Advisory Group. This study (UK Biobank project 15678) was covered by the general ethical approval of the UK Biobank studies from the NHS National Research Ethics Service on 17 June 2011 (Ref. 11/NW/0382).¹⁶ For the studies participating in the AD GWAS datasets,^21,22 written informed consent was obtained from study participants or, for those with substantial cognitive impairment, from a caregiver, legal guardian, or another proxy. Study protocols for all cohorts were reviewed and approved by the appropriate institutional review boards.^21,22

Statistical power

The a priori statistical power was calculated using an online tool at cnsgenomics.com/shiny/mRnd/.²³ We assumed that the 8 SNPs for average accelerations and fraction accelerations >425 milligravities explained 0.4% of the phenotypic variables.²⁴ Given a type 1 error of 5%, we had sufficient statistical power (>85%) for an expected odds ratios (ORs) per 1 SD of ≤0.88 AD in the primary dataset.²¹

Statistical analyses

A multiplicative random effects inverse-variance weighted (IVW) model was used as principal mendelian randomization analysis.^15,25 For sensitivity analyses, we applied weighted median, mendelian randomization–Egger (MR-Egger), and mendelian randomization–Pleiotropy Residual Sum and Outlier (MR-PRESSO).^26,27 Results are presented as OR per 1 -SD increment in average accelerations, for comparing engagement in vigorous physical activity (fraction accelerations >425 milligravities) and no engagement in vigorous physical activity (fraction accelerations ≤425 milligravities). One SD of average accelerations in the UK Biobank Study is approximately 8 milligravities (or 0.08 m/s²) of acceleration in a mean 5-second window.¹⁶

For the 2-sample mendelian randomization approach to be valid, several assumptions must be satisfied.^15,28 First, the genetic variants used as instruments for physical activity should be associated with the risk factor (F statistic >10). The F statistics ranged from 26 to 48 (table 2). The second assumption requires that the genetic variants are not associated with any confounders of the association between physical activity and AD (no horizontal pleiotropy). To assess this assumption, we checked each instrument and its proxies (r² > 0.8) in PhenoScanner²⁹ and the GWAS catalog³⁰ for previously reported associations (p < 5 × 10⁻⁶) with confounders. We regarded smoking, education, pulmonary function, and homocysteine as relevant confounders.^1,31–34 We additionally performed leave-one-SNP-out analyses to rule out possible pleiotropic effects, and we used the MR Egger regression approach where a close to zero intercept indicated no presence of directional pleiotropy.¹⁵ The third assumption requires the genetic variants to be associated with AD only through physical activity (exclusion restriction). This assumption is not verifiable in summary data mendelian randomization analysis.²⁸ Finally, mortality prior to study onset precludes individuals from entering the AD study.^35,36 We assessed the exposure effect on mortality as a means of evaluating sensitivity to survival bias.³⁷ Using data from a GWAS on longevity,³⁸ we associated genetically predicted accelerometer-based physical activity with surviving at or beyond the age corresponding to the 90th survival percentile. The current study was not preregistered. Analyses were performed using the TwoSampleMR (version 0.5.0)²⁷ and MR-PRESSO (version 1.0) packages in R (version 3.6.2). Reporting follows the STROBE-MR statement.³⁹

Data availability

This study is based on publicly available data. The summary statistics for the physical activity GWAS¹⁶ are available at klimentidis.lab.arizona.edu/content/data (access date: January 27, 2020); the AD GWAS summary datasets^21,22 are available at niagads.org/datasets/ng00075 and ctg.cncr.nl/software/summary_statistics (access date: January 27, 2020); and the longevity GWAS dataset can be accessed at longevitygenomics.org/downloads (access date: January 27, 2020).

Results

In our primary analysis, we used 21,982 AD cases and 41,944 cognitively normal controls. The associations between genetic variants and exposure and between genetic variants and outcome are provided in tables 2 and 3. We found that genetically predicted average accelerations were not associated with AD (IVW OR per 1-SD increment, 1.03; 95% confidence interval [CI], 0.97–1.10; p = 0.332; table 4). Likewise, fraction accelerations >425 milligravities were unrelated to AD (IVW OR, 0.91; 95% CI, 0.46–1.81; p = 0.794; table 4). These findings were confirmed using alternative mendelian randomization methods and leave-one-SNP-out analysis (tables 4 and 5). p Values of the Cochran Q statistic <0.05 indicated heterogeneity in effects among SNPs instrumenting average accelerations and fraction accelerations >425 milligravities (table 6). Our replication analysis using summary statistics from a different GWAS for AD²² supported our primary finding (table 4).

Table 4.

Mendelian randomization estimates for the relationship between accelerometer-based physical activity and Alzheimer disease (AD)

Table 5.

Inverse-variance weighted estimates for accelerometer-based physical activity and Alzheimer disease (AD) in the primary analysis with single nucleotide polymorphisms (SNPs) individually removed in leave-one-out analyses

Table 6.

Heterogeneity and mendelian randomization–Egger (MR-Egger) pleiotropy test for the primary analysis

None of our selected instruments or its proxies were associated with potential confounders (smoking, education, pulmonary function, homocysteine) of the association between physical activity and AD in PhenoScanner or the GWAS catalog. The intercept from the MR-Egger regression was not statistically significant in the analyses of average accelerations and fraction accelerations >425 milligravities, denoting lack of potential directional pleiotropy (table 6). Average accelerations were unrelated to the odds of surviving the 90th percentile of age (IVW OR per 1 SD, 1.01; 95% CI, 0.92–1.12; p = 0.794) in a sensitivity analysis to assess robustness to survival bias.

Discussion

We found no evidence that physical activity (assessed as average accelerations and fraction of accelerations >425 milligravities [vigorous physical activity]) has an effect on the risk of developing AD. We performed mendelian randomization analysis using a recently published GWAS on AD²¹ and replicated the null finding in another independent AD GWAS.²²

Previous observational studies concluded that higher levels of self-reported physical activity are associated with reduced risk of dementia and AD.^6,7,9 The most comprehensive meta-analysis comprising 15 cohort studies found a 35% (relative risk, 0.65; 95% CI, 0.56–0.74) relative reduction in risk of AD when comparing the highest and lowest levels of physical activity.⁷ Two previous mendelian randomization studies^40,41 found that a genetic liability to higher self-reported physical activity was associated with an increased risk of AD with ORs >2. Yet both studies used genetic variants as instruments that are additionally associated with cognition and AD (including variants in the APOE region),^16,42 which might have violated the no horizontal pleiotropy and exclusion restriction assumptions. These conclusions of observational studies and the 2 previous mendelian randomization studies are in contrast to meta-analyses of intervention studies, which do not show a protective effect of exercise interventions on the risk of AD.^4,11 Similarly, recent observational studies have found that when physical activity assessment and diagnosis of AD are ≥10 years apart there was no association between physical activity and risk of dementia and AD.^12–14 In particular, an individual-level meta-analysis¹² of 19 studies including 319,953 participants and 987 incident AD cases found a hazard ratio of 1.04 (95% CI, 0.91–1.19) when comparing physically active and inactive individuals when restricting follow-up time to ≥10 years. Similarly, an analysis of 1.1 million women with 5,873 AD cases found a hazard ratio of 0.95 (95% CI, 0.93–0.98) for the comparison of active and inactive women and AD risk during follow-up years 15+.¹⁴

Our mendelian randomization analysis and the findings from the latter 2 long-term observational studies support the notation that decline in physical activity levels occurs during the subclinical phase of dementia and that previous observational studies might have overestimated dementia risk associated with insufficient levels of physical activity as many studies were based on short follow-up times and thus may have been subject to reverse causation.^12,14

Our study has several notable strengths. The use of 2-sample mendelian randomization enabled us to use the largest GWAS on AD to date. We replicated our analysis using a different AD GWAS dataset. We used genetically predicted objectively measured physical activity, which is less prone to recall and response bias than measurement of self-reported physical activity.⁴³ Furthermore, because some genetic loci for self-reported physical activity are also related to cognitive function, genetically proxied self-reported physical activity measures may be prone to violations of the no horizontal pleiotropy assumption.^16,42 In contrast, SNP associations based on accelerometer-assessed physical activity are unrelated to cognitive performance or other potential pathways and shared common causes with AD, which reduces the likelihood of pleiotropic effects.^16,42

Our study has certain limitations. First, the IVW estimate for fraction of accelerations >425 milligravities and AD had low statistical power and a CI ranging from 0.46 to 1.81, which is compatible with halving AD risk. However, the lower bound of the CI of the corresponding replication analysis was 0.95, which provided reassuring evidence of a null association of vigorous physical activity and AD. Second, the discovery GWAS of physical activity consisted of UK Biobank participants aged 40–70 years.¹⁶ Previous twin studies suggest that the genetic contribution to physical activity decreases with age.⁴⁴ By estimating SNP–physical activity associations in a sample of middle-aged and older adults, we might have underestimated the denominator effect of the ratio estimator.⁴⁵ Thus, when the effect of the instrument on exposure changes over time, the ratio estimator will represent a biased estimate of the lifetime effect of physical activity on AD.⁴⁵ Third, as the AD study samples consist of a nonrandom subset of the population that has survived to be included, survival bias may distort findings.^35,36 We examined sensitivity to survival bias³⁷ and found that physical activity was unrelated to longevity.

Given the increase in life expectancy, AD has increasingly become a public health challenge and measures to prevent or delay the onset of dementia are urgently needed. The present study provides little evidence that recommending physical activity would help to prevent AD.

Glossary

AD

Alzheimer disease

CI

confidence interval

DSM-IV

Diagnostic and Statistical Manual of Mental Disorders, 4th edition

GWAS

genome-wide association studies

IVW

inverse-variance weighted

MET

metabolic equivalent task

MR-Egger

mendelian randomization–Egger

MR-PRESSO

mendelian randomization–Pleiotropy Residual Sum and Outlier

OR

odds ratio

SNP

single nucleotide polymorphism

Appendix. Authors

Study funding

The i–Select chips were funded by the French National Foundation on Alzheimer's Disease and Related Disorders. EADI was supported by the LABEX (Laboratory of Excellence Program Investment for the Future) DISTALZ grant, Inserm, Institut Pasteur de Lille, Université de Lille 2, and the Lille University Hospital. GERAD/PERADES was supported by the Medical Research Council (grant 503480), Alzheimer's Research UK (grant 503176), the Wellcome Trust (grant 082604/2/07/Z), and German Federal Ministry of Education and Research (BMBF): Competence Network Dementia (CND) grants 01GI0102, 01GI0711, 01GI0420. CHARGE was partly supported by the NIH/NIA grant R01 AG033193 and the NIA AG081220 and AGES contract N01–AG–12100, the NHLBI grant R01 HL105756, the Icelandic Heart Association, and the Erasmus Medical Center and Erasmus University. ADGC was supported by NIH/NIA grants U01 AG032984, U24 AG021886, and U01 AG016976, and Alzheimer's Association grant ADGC–10–196728.

Disclosure

The authors report no disclosures relevant to the manuscript. Go to Neurology.org/N for full disclosures.