Mendelian randomization study of coffee consumption and age at onset of Huntington’s disease

Abstract

Objective

The association between habitual coffee or caffeine consumption and age at onset (AAO) of Huntington’s disease (HD) is unclear. We employed Mendelian randomization to investigate the causal relationship between coffee consumption and AAO of HD.

Methods

We selected 14 independent genetic variants associated with coffee consumption from a genome-wide association study (GWAS) meta-analysis of 375,833 individuals of European ancestry. Genetic association estimates for AAO of HD were obtained from the Genetic Modifiers of Huntington’s Disease Consortium GWAS meta-analysis including 9,064 HD patients of European ancestry. The inverse variance weighted method was used to evaluate the causal estimate and a comprehensive set of analyses tested the robustness of our results.

Results

Genetically predicted higher coffee consumption was associated with an earlier AAO of HD (β = −1.84 years, 95% confidence interval (CI) = −3.47 to −0.22, P = 0.026). Results were robust to potential pleiotropy and weak instrument bias.

Conclusions

This genetic study suggests high coffee consumption is associated with an earlier AAO of HD. Coffee is widely consumed and thus our findings, if confirmed, offers a potential way to delay the onset of this debilitating autosomal dominant disease.

Introduction

Huntington’s disease (HD) is an autosomal dominant neurodegenerative disease caused by a CAG repeat expansion in huntingtin gene (HTT) and is characterized by movement, cognitive and psychiatric disorders.^{1, 2} The annual incidence of HD is about 0.38 per 100,000,³ with death occurring within 15–20 years after onset.⁴ CAG repeat length inversely correlates with and accounts for 50–70% of variance in age at onset (AAO), and the remaining variation is likely explained by other genetic and environmental factors.^{4, 5} Given that HD is a fatal disease and no disease-modifying therapies are available, it is important to identify interventions that can delay the AAO of HD. A single retrospective observational study reported that higher caffeine intake was associated with an earlier AAO of HD.⁶ Additional epidemiological support is warranted but challenged by lack of large prospective studies of HD, a rare disease.

Mendelian randomization (MR) is an alternative approach for inferring causality that uses genetic variants as instruments for traits of interest (e.g., coffee consumption) and is not susceptible to confounding and reverse causation in observational studies.^7–10 With available genome-wide association study (GWAS) results for consumption of coffee, the major source of caffeine for many populations, and AAO of HD,^{11, 12} MR offers an opportunity to efficiently investigate the casual relationship between coffee consumption and AAO of HD. Therefore, we performed this 2-sample MR study to investigate the causal effects of coffee consumption on AAO of HD.

Materials and methods

We used the summary statistics from published GWAS that previously obtained informed consent and ethical review board approvals.^{11, 12}

Genetic instruments for coffee consumption

We selected genetic instruments for coffee consumption (per 1% increment) based on a GWAS meta-analysis of the UK Biobank and three independent US cohorts including up to 375,833 individuals of European ancestry.¹¹ In UK Biobank (discovery stage), coffee consumption was collected from all participants at baseline using a touchscreen questionnaire.¹¹ Total coffee consumption was based on the question ‘How many cups of coffee do you drink each day (include decaffeinated coffee)? The association analyses were adjusted for age, sex, body mass index (BMI), and top 20 principal components.¹¹ Replication stage was carried out in three independent US corhorts: Women’s Genome Health Study (WGHS), Nurses’Health Study (NHS), and Health Professionals Follow-up Study (HPFS). For these three replication cohorts, coffee consumption data were collected by the same food frequency questionnaire (FFQ). For WGHS, the GWAS analysis was adjusted for age, total caloric intake, top 10 eigenvectors of population substructure, and BMI. For NIHS and HPFS, the GWAS analysis was adjusted for age, total caloric intake, case-control study, top 4 eigenvectors of population substructure, and BMI. Finally, this GWAS meta-analysis identified 15 single nucleotide polymorphisms (SNPs) associated with coffee consumption at genome-wide significance (P < 5 × 10⁻⁸).¹¹ We clumped these SNPs for independence (r² < 0.1, distance window, 10,000kb), and used the remaining 14 SNPs as the primary genetic instrument. The effect sizes for the SNP–coffee consumption associations were expressed per 1% increment in coffee consumption in the GWAS meta-analysis,¹¹ and were rescaled to 50% increment in coffee consumption in this study.

As a secondary instrument, we used two of these 14 SNPs which present with the largest effect estimates, have been confirmed in multiple studies^{11, 13–15} and are biologically relevant; SNPs near CYP1A1/A2 (rs2472297) and AHR (rs4410790). CYP1A2 encodes cytochrome P450(CYP)1A2 which accounts for ~ 95% of caffeine metabolism in humans, and AHR encodes the aryl hydrocarbon receptor which can regulate the expression of CYP1A1 and CYP1A2.^{13, 16, 17} The two variants of these loci associated with higher coffee consumption are also associated with lower blood caffeine levels in GWAS.¹⁸

Outcome data sources

The primary outcome in this study was AAO of HD corrected for inherited CAG repeat length; herein ‘residual AAO’ as defined by the Genetic Modifiers of Huntington’s Disease (GeM-HD) Consortium.¹² Specifically, residual AAO was calculated as the difference between the observed (AAO of diagnostic motor signs) and expected (based on CAG repeat size) AAO, representing years of deviation from the expectation.¹² For example, a HD patient with a residual AAO of +4 indicates that he/she developed motor symptoms 4 years later than expected. Genetic association estimates for residual AAO in HD were obtained from GeM-HD Consortium GWAS meta-analysis including 9,064 HD patients of European ancestry (4,417 males; 4,647 females), which sought to discover disease-modifying loci that act prior to clinical diagnosis to delay onset.¹² This GWAS revealed 8 common (minor allele frequency > 1%) genome-wide significant (P < 5×10⁻⁸) loci; thus demonstrating sufficient power for the current MR.¹²

Statistical analysis

The primary MR estimates were performed using fixed-effects inverse variance weighted (IVW) meta-analyses.¹⁹ IVW estimates might be biased in the presence of directional pleiotropy.¹⁹ As a measure of pleiotropy, we evaluated heterogeneity across individual SNP estimate in the IVW analyses with the Cochran Q test (P < 0.05 was deemed statistically significant). We also used alternative MR methods that are relatively robust to pleiotropic SNPs, including MR-Robust Adjusted Profile Score (MR-RAPS),²⁰ weighted median²¹ and MR-Egger regression.²² The MR-RAPS method provides unbiased estimates in the presence of many weak instruments.²⁰ The weighted median method allows an instrumental variable based on up to 50% potentially invalid SNPs.²¹ The MR-Egger regression can control for bias from unbalanced pleiotropy, at the cost of reduced power.²² The intercept from MR-Egger was used as an indicator of unbalanced pleiotropy (P < 0.05 was considered significance).²² We further examined the presence of pleiotropic outlier SNPs using the MR-Pleiotropy Residual Sum and Outlier (MR-PRESSO) method.²³ The F-statistic was used to evaluate the strength of the instruments.²⁴ All statistical analyses were undertaken in R (v4.0.2) using the MendelianRandomization,²⁵ MR-PRESSO,²³ and mr.raps packages.²⁰

Results

The F statistic for each SNP was > 10, suggesting low risk of weak instrument bias (Supplementary Table 1). Genetically predicted 50% increase in coffee consumption was associated with an earlier AAO of HD (β = −1.84 years, 95% confidence interval (CI) = −3.47 to −0.22, P = 0.026; Figure 1). The MR-RAPS, weighted median, and MR-Egger analyses yield similar pattern of effect estimates, although with broader CIs due to lower statistical power (Figure 1). There was no evidence of heterogeneity (P for Cochran’s Q = 0.42) and unbalanced pleiotropy (P for MR-Egger intercept = 0.73). For the 14 SNPs, MR-PRESSO did not identify any significant potential outlier.

Figure 1.

Mendelian randomization estimates of coffee consumption and age at onset of Huntington’s disease.

An MR analyses using a genetic instrument restricted to two biologically relevant variants (rs2472297 and rs4410790) yielded similar results; genetically predicted 50% increase in coffee consumption was associated with an earlier AAO of HD (β = −2.09 years, 95% CI = −4.17 to −0.02, P = 0.048; Figure 1).

Discussion

By leveraging large-scale GWAS data, we employed the first MR to investigate the causal effects of coffee consumption on the AAO of HD. We found that genetically predicted 50% increase in coffee consumption was associated with about 2 years earlier AAO of HD.

Our genetic findings are consistent with the only observational study that has investigated the association between caffeine and AAO of HD.⁶ This study retrospectively evaluated caffeine consumption in 80 HD patients using a dietary survey, and suggested that pre-diagnosis caffeine consumption > 190 mg/day (~2 small cups of regular coffee) was significantly associated with 3.6 years earlier AAO of HD after adjusting for gender, smoking status and CAG repeat length.⁶ Our MR study protects against reverse causation and recall bias and thus strengthens these initial retrospective findings. Nevertheless, careful interpretations of our findings are warranted. Most of the coffee-SNPs, particularly those with large effect estimates are associated with increased coffee drinking behavior but lower biological exposure to caffeine according to GWAS for each trait. ^{18, 26} That is, higher coffee (caffeine) drinking is balanced by higher caffeine metabolism. Thus, our significant findings point to at least two causal hypotheses. First, higher biological exposure to caffeine may delay the onset of HD. Neurological research suggests caffeine exerts antioxidant effects and reduces oxidative stress and inflammation;^{27, 28} properties which may impact the onset and progression of HD in a beneficial way.^29–31 Second, exposure to one or more non-caffeine constituents of coffee may accelerate the onset of HD. We are unaware of any compound in coffee that is potentially neurotoxic. Both hypotheses warrant further investigation.

Our MR-designed investigation presents with a number of strengths to complement and extent traditional epidemiological studies but is not without limitations. First, while we employed rigorous methods to address it, we cannot completely discount potential bias due to pleiotropy. Second, our study was based on data from populations of European ancestry and thus findings may not generalize to other populations. Our focus on European ancestry was due to the limited availability of summary-level data but also to protect against bias resulting from population stratification. Third, our genetic instrument for coffee was dominated by caffeine-metabolism related SNPs which partly complicates interpretation of the results. Coffee is a major source of caffeine but is also an important source of other bioactive constituents and an instrument reflecting other aspects of coffee drinking behavior would have been ideal.

In conclusion, our genetic findings add to preliminary research suggesting caffeine-containing beverages may impact the AAO of HD. However, interpretation of the specific results warrants caution and further investigation. Given widespread consumption of caffeine-containing beverages the potential for caffeine or coffee to impact HD merits further investigation.