The research, led by Jieping Wang, Huanyu Wang, and colleagues at Shanghai University of Sport, drew on the UK Biobank, one of the largest prospective health cohorts ever assembled. The team followed 435,551 participants who were free of Parkinson’s disease at baseline for a median of 15.7 years, during which 3,319 people developed the disorder. Everyone in the analysis carried a known genotype at CYP1A2 rs762551, a functional variant in the gene encoding the primary enzyme that metabolizes caffeine. The variant comes in three flavors: AA carriers, who are fast metabolizers; AC heterozygotes, who are intermediate; and CC homozygotes, who are slow. In the cohort, roughly 52 percent were AA, 40 percent AC, and 8 percent CC.

On the surface, the results looked like a null finding. When the researchers pooled everyone together, coffee consumption showed no significant association with Parkinson’s risk, and Kaplan-Meier survival curves for different intake levels and genotypes overlapped almost perfectly. Neither coffee nor CYP1A2 genotype alone predicted who would develop the disease. But that population-averaged picture concealed something important. A formal likelihood ratio test revealed a significant interaction between coffee intake and CYP1A2 genotype, with a p-value of 0.01, meaning the effect of coffee genuinely differed depending on metabolic phenotype. No significant interaction emerged between coffee and sex alone, and the three-way interaction among coffee, genotype, and sex was not statistically significant either, though sex-stratified patterns still proved revealing.

When the analysis was stratified by genotype, the heterogeneity became striking. Among AA fast metabolizers, drinking less than five cups per day was associated with a 17 percent lower risk of Parkinson’s disease, with a hazard ratio of 0.83 and a 95 percent confidence interval of 0.74 to 0.93. Restricted cubic spline analysis, a statistical technique that models the dose-response curve flexibly rather than forcing it into a straight line, showed the association was nonlinear, with the lowest risk occurring at around three cups per day. High intake, at five or more cups daily, showed no significant benefit in this group. In sharp contrast, among AC carriers, heavy consumption of five or more cups per day was associated with a 43 percent higher risk, with a hazard ratio of 1.43, and among the small CC group the estimate climbed to 1.84, nearly doubling the risk.

The sex-stratified results added another layer of complexity. The protective association among AA carriers was driven almost entirely by women: women with the AA genotype who drank less than five cups per day had a 27 percent lower risk, with a hazard ratio of 0.73, while men with the same genotype showed no significant benefit at any intake level. Conversely, the elevated risk among slow metabolizers at high consumption was concentrated in men. AC men drinking five or more cups daily had a 53 percent higher risk, with a hazard ratio of 1.53, and CC men showed a 2.37-fold increase. These patterns suggest that hormonal context, particularly estrogen’s known modulation of CYP1A2 activity and adenosine signaling, shapes how caffeine’s effects play out in different bodies.

The pharmacokinetic logic behind these findings is coherent. Fast metabolizers maintain moderate systemic caffeine levels that are high enough to antagonize adenosine A2A receptors, enhancing dopaminergic transmission and reducing neuroinflammation, but brief enough to avoid the downsides of prolonged exposure. Slow metabolizers, by contrast, experience extended caffeine clearance times, and the authors propose that sustained high exposure may lead to receptor desensitization, oxidative stress, and mitochondrial dysfunction, particularly at heavy doses. However, the researchers are careful to note that direct mechanistic evidence for caffeine-induced neurotoxicity at high habitual intake in humans remains limited, and the observational data cannot fully separate caffeine-specific effects from correlated lifestyle factors.

To probe whether the pattern was robust, the team ran six sensitivity analyses. Excluding the first two, three, or five years of follow-up to guard against reverse causation from prodromal disease left the results materially unchanged. Finer exposure categories, additional adjustment for cumulative smoking in pack-years, and alternative genotype groupings all produced stable patterns. When AC and CC carriers were combined, intake below five cups showed no association, while five or more cups daily was associated with a 48 percent higher risk. The researchers also constructed polygenic scores for caffeine metabolism, CMSG2 and CMSG4, incorporating variants near CYP1A2, AHR, POR, and CYP2A6, and found that the primary findings were broadly consistent across all score strata, suggesting that CYP1A2 function, rather than broader genomic influences, is the main driver of the heterogeneity.

The study has important caveats. Coffee consumption was self-reported only at baseline and assumed to remain stable over 15 years, which likely introduced exposure misclassification that would bias results toward the null. The UK Biobank coffee variable also included decaffeinated coffee, so the observed associations cannot be interpreted as direct evidence of caffeine-specific effects, and this may partly explain why the overall unstratified analysis showed no association, in contrast to a recent population-based study that found caffeinated coffee protective and decaffeinated coffee neutral. The cohort is also predominantly of White European ancestry, limiting generalizability, and Parkinson’s cases were ascertained from hospital and registry records, meaning early or undiagnosed cases may have been missed. The number of cases in some genotype-dose strata, especially among CC homozygotes, was small, reducing statistical precision.

Despite these limitations, the findings offer a compelling resolution to one of the most persistent puzzles in nutritional epidemiology. The association between coffee and Parkinson’s disease, the authors conclude, is not a single relationship but a family of them, shaped by metabolic phenotype and sex. Population-averaged analyses mask meaningful subgroup effects, and the intake range most consistent with lower risk is context-dependent. The work provides testable hypotheses for precision prevention and a framework for designing genotype-informed studies, suggesting that the next generation of coffee and Parkinson’s research should stop asking whether coffee is good or bad, and start asking for whom, at what dose, and through which metabolic machinery.

Subject of Research: Neurology

Article Title: Coffee and Parkinson’s disease: associations by CYP1A2 genotype and sex in UK Biobank

Article References: Wang, J., Wang, H., wan, K., Wang, R., Guo, S., Ma, F., Xiang, Y., Ma, H., Zong, G., Zhi, X., Tian, X., Zhou, G., & Wang, R. (2026). Coffee and Parkinson’s disease: associations by CYP1A2 genotype and sex in UK Biobank. npj Parkinson's Disease, 12(1), Article 226. https://doi.org/10.1038/s41531-026-01519-0

Image Credits: AI Generated

DOI: 10.1038/s41531-026-01519-0

Keywords: Coffee, Parkinson, disease, associations, CYP1A2, genotype, Biobank, scientific research