Causal association of genetically determined caffeine intake from tea or coffee with bone health: a two-sample Mendelian randomization study

Abstract

Background

Relationship of caffeine intake and consumption of caffeinated beverages, such as tea and coffee, with bone health remains controversial. This study aimed to evaluate whether genetically determined caffeine intake from tea or coffee has causal effects on overall total body bone mineral density (TB-BMD) and fracture. We also assessed the association with TB-BMD in five age strata.

Methods

Using two-sample Mendelian randomization approach, summary statistics were retrieved from genome-wide association studies (GWAS)/GWAS meta-analyses of caffeine intake from tea (n = 395 866)/coffee (n = 373 522), TB-BMD (n = 66 628), and fracture (n = 426 795). Inverse variance weighted method was adopted as the main univariable analysis. Multivariable analysis was conducted to evaluate whether the causal effect is independent.

Results

In univariable analysis, genetically determined caffeine intake from tea had positive association with overall TB-BMD (per SD increase in genetically determined caffeine intake, beta of TB-BMD [in SD]: 0.166; 95% confidence interval (CI): 0.006–0.326) and inverse association with fracture (OR = 0.79; 95% CI: 0.654–0.954). Genetically determined caffeine intake from coffee was also positively associated with overall TB-BMD (beta = 0.231; 95% CI: 0.093–0.369). The association remained significant after adjustment for smoking in multivariable analysis. Genetically determined caffeine intake from tea or coffee was both positively associated with TB-BMD in the age strata of 45–60 years, but we lacked evidence of association in other strata.

Conclusions

Genetically, caffeine intake from tea or coffee may be beneficial to bone health. Due to the ascertainment method of caffeine intake from tea, our study also implied genetically higher tea consumption may improve TB-BMD and lower fracture risk.

What is already known on this topic 

• Published studies demonstrated controversial findings on the relationship of caffeine intake and consumption of caffeinated beverages, such as tea and coffee, with bone health.

What this study adds 

• This two-sample Mendelian randomization study demonstrated that genetically determined caffeine intake from tea or tea consumption was causally associated with improved overall TB-BMD and reduced fracture risk. Genetically determined caffeine intake from coffee was also positively associated with overall TB-BMD.

How this study might affect research, practice or policy 

• Caffeine and other active ingredients in tea may improve bone health. Tea may be included in the diet as a strategy to improve BMD and prevent osteoporosis-associated fracture.

Introduction

Osteoporosis is a chronic bone disease characterized by low bone mineral density (BMD) and deterioration of bone microarchitecture, leading to increased fracture risk. BMD measured by dual-energy X-ray absorptiometry (DXA) is the gold standard for diagnosis of osteoporosis. Osteoporosis-related fracture is estimated to incur significant economic burden, particularly amid the ageing population [1]. Moreover, the median 1-year all-cause mortality rate post hip fracture remained high at 22.8%, even a decreasing or stabilized trend was observed [2]. Therefore, identifying causal and modifiable risk factors (such as diet) is urgently required for timely prevention of osteoporosis and its associated fracture.

Tea and coffee are the most frequently consumed beverages [3], and also the major dietary sources of caffeine [4]. Yet, whether caffeine intake has any effect on bone health remains controversial [5–9]. Notably, individuals’ habitual caffeine intake [10], consumption of tea and coffee [11, 12] were linked to genes involved in the pharmacokinetic and pharmacodynamic response to caffeine, such as aryl hydrocarbon receptor and cytochrome P450 family 1 subfamily A member. Instead of solely habitual preference, the role of genetics enables the use of genetic variants as instrumental variables for caffeine intake or consumption of tea/coffee in Mendelian randomization (MR) studies. Since genetic variants are randomly allocated at conception, MR approach is less susceptible to residual confounding and reverse causality, providing stronger evidence of causal inference compared with conventional observational studies [13]. Aiming to clarify the controversial role of caffeine intake and explore the optimal dietary requirement in improving bone health, we assessed the causal effect of genetically determined caffeine intake from tea or coffee separately on DXA-derived total body bone mineral density (TB-BMD) and fracture using MR approach.

Materials and methods

Study design

The univariable two-sample MR approach was adopted, primarily aimed to examine the total causal effect of genetically determined caffeine intake from tea or coffee (exposure) on overall TB-BMD in the total sample and fracture (outcome). TB-BMD and fracture were the primary outcomes as the genome-wide association study (GWAS) meta-analysis of TB-BMD [14] was the largest GWAS of DXA-derived BMD to-date while fracture was the clinical outcome of osteoporosis. The GWAS of combined caffeine intake from tea and coffee was also publicly available, but this phenotype was specifically related to caffeine intake from both tea and coffee, not either of them [10]. Notably, an individual with the habit of regular coffee intake is less likely having the concurrent habit of regular tea consumption, which was also implied by the weak phenotypic correlation between caffeine intake from tea and coffee, respectively (Spearman’s rank correlation coefficient r = −0.33) [10]. As the combined caffeine intake phenotype may not provide much clinical implication, its relationship with bone health was evaluated as an additional analysis. Due to the age-dependent features of both the genetic loci and their effect size for TB-BMD [14], if causal association was identified for overall TB-BMD, we also evaluated whether genetically determined caffeine intake has differential causal effects on TB-BMD in five age strata (below 15 years, 15–30 years, 30–45 years, 45–60 years, and 60 years or above) as secondary analysis. As a supplementary analysis for comparison, the causal association of genetically determined caffeine intake with DXA-derived BMD at lumbar spine (LS-BMD), femoral neck (FN-BMD), and forearm (FA-BMD) was assessed despite the relatively low statistical power.

If significant causal relationship was observed in the above univariable analyses, multivariable MR (MVMR) analysis was conducted to confirm whether caffeine intake was independently affecting bone health. Given the close correlation between smoking and caffeine intake demonstrated by both conventional observational [15] and genetic correlation [16] studies, and genetic instruments may be associated with several correlated exposures [17], MVMR analysis adjusted for beta estimates of cigarette smoking was performed. The study design and assumptions of MR analyses are illustrated in Fig. 1.

Data sources

Summary statistics of caffeine intake from tea, coffee, combined caffeine intake from both beverages, TB-BMD, fracture, LS-BMD, FN-BMD, FA-BMD, and cigarette smoking were retrieved from the largest publicly available GWAS. Summary statistics of caffeine intake from tea, coffee, and combined caffeine intake were obtained from a GWAS by Said et al. conducted in the UK Biobank participants, who were asked the average number of cups of tea and coffee they drank each day, and the type of coffee they usually drank [10]. The descriptions of data sources for all exposures and outcomes are detailed in Table 1. The selection of genetic instruments was described in Supplementary Methods 1.

Univariable MR analysis

The inverse variance weighted (IVW) method was adopted as the main analysis in the current MR study [18]. Several MR methods with different assumptions were applied as sensitivity analyses, including weighted median [19], MR-Egger regression [20], contamination mixture [21], and MR Pleiotropy Residual Sum and Outlier (MR-PRESSO) [22] methods, which are described in Supplementary Methods 2. Upon exclusion of pleiotropic outliers identified by MR-PRESSO [22], the main and sensitivity MR analyses were repeated. Cochran’s Q test was also conducted to evaluate the heterogeneity across the final genetic instruments. For analyses with TB-BMD as the outcome, results were presented as change in TB-BMD in standard deviation (SD) per SD increase in genetically determined caffeine intake. For analyses with fracture as outcome, the results were presented as the odds ratio (OR) of fracture per SD increase in genetically determined caffeine intake. Statistical power was calculated online (https://sb452.shinyapps.io/power/) [23]. If there was participant overlap between exposure and outcome datasets, the bias and Type I error were estimated by a web-based tool (https://sb452.shinyapps.io/overlap/) [24]. All analyses were conducted in R (version 4.1.3) using the ‘MendelianRandomization’, ‘TwoSampleMR’, and ‘MRPRESSO’ packages.

MVMR analysis

Several MVMR methods were developed as the extended versions of univariable MR methods, including MVMR-IVW [25], MVMR-Egger [26], and MVMR-PRESSO. ‘MVMR’ and ‘MRPRESSO’ packages in R were utilized to provide the causal estimates. The MVMR-Egger intercept test and MVMR-PRESSO global test were employed to detect horizontal pleiotropy in MVMR analyses.

Results

Genetic instruments adopted in the MR analyses

The final lists of genetic instruments applied in each univariable and MVMR analysis were included in Supplementary Tables S1–S25, and Supplementary Tables S26–S30, respectively. Caffeine intake from tea or coffee shared two common genetic instruments, while the combined caffeine intake shared up to five and six genetic instruments with caffeine intake from tea or coffee, respectively (Supplementary Tables S1–S6). Information of the genetic instruments adopted in each analysis is summarized in Table 2. The F-statistic was relatively high in all analyses (≥90.87). The Cochran’s Q tests were all statistically insignificant, indicating the absence of heterogeneity among the instruments. The Type I error and bias incurred due to sample overlap between the exposure and outcome datasets were estimated to be 0.002–0.004 and 0.05, respectively (Table 2), which is minimal. Power calculation for MR analysis of genetically determined caffeine intake from tea (Supplementary Figs S1–S3), coffee (Supplementary Figs S4–S6), and combined caffeine intake (Supplementary Figs S7–S9) with TB-BMD, fracture, LS-BMD, FN-BMD, and FA-BMD are presented in Supplementary Results 1.

Primary analysis—causal association of genetically determined caffeine intake from tea or coffee with overall TB-BMD and fracture

Univariable IVW analysis suggested genetically determined caffeine intake from tea had a positive causal association with overall TB-BMD (per SD increase in genetically determined caffeine intake from tea, beta of overall TB-BMD in SD: 0.166; 95% confidence interval [CI]: 0.006–0.326; P = 0.042; Fig. 2a and Supplementary Fig. S10) and an inverse association with reduced risk of fracture (OR = 0.79; 95% CI: 0.654–0.954; P = 0.014; Fig. 2a and Supplementary Fig. S11). The significant associations were also observed in the sensitivity analyses of contamination mixture and/or MR-PRESSO method, with insignificant MR-Egger intercept and MR-PRESSO global tests (Fig. 2a). Leave-one-out analysis showed that the significant associations with TB-BMD and fracture were not driven by any single instrument (Supplementary Figs S12–S13). Upon adjustment for the beta estimates of smoking in MVMR analysis, the association with overall TB-BMD (beta = 0.172; 95% CI: 0.007–0.336; P = 0.041) and fracture (OR = 0.811; 95% CI: 0.671–0.978; P = 0.029) remained significant in MVMR-IVW method. Similar estimates were observed from MVMR-PRESSO method. Both MVMR-Egger intercept and MVMR-PRESSO global tests were insignificant (Fig. 2b), suggesting horizontal pleiotropy was unlikely.

Similarly, genetically determined caffeine intake from coffee had a positive association with overall TB-BMD in univariable IVW (beta = 0.231; 95% CI: 0.093–0.369; P = 0.001; Fig. 2c and Supplementary Fig. S14), contamination mixture, and MR-PRESSO analyses. As MR-PRESSO global test was insignificant but not the MR-Egger intercept test (Fig. 2c), we cannot rule out the possibility of horizontal pleiotropy. No sufficient evidence could support the association of genetically determined caffeine intake from coffee with fracture (Fig. 2c and Supplementary Fig. S15). Based on the leave-one-out analysis, no single instrument drove the association (Supplementary Figs S16–S17). Upon adjustment for smoking, the causal association with overall TB-BMD remained significant in MVMR-IVW (beta = 0.222; 95% CI: 0.075–0.37; P = 0.003) and MVMR-PRESSO (beta = 0.222; 95% CI: 0.062–0.383; P = 0.01) methods. Insignificant MVMR-Egger intercept and MVMR-PRESSO global tests suggested that horizontal pleiotropy was less likely (Fig. 2d).

In the additional analysis, we had insufficient evidence to support any causal association of genetically determined combined caffeine intake from both tea and coffee with overall TB-BMD and fracture (Supplementary Results 2 and Supplementary Figs S18–S22).

Secondary analysis—causal association of genetically determined caffeine intake from tea or coffee with TB-BMD in five age strata

We further assessed whether the association of genetically determined caffeine intake from tea or coffee with TB-BMD was age-specific. Insufficient evidence could support any association of genetically determined caffeine intake from tea with TB-BMD in the age strata below 15 years, 15–30 years, 30–45 years, and 60 years or above (Table 3 and Supplementary Fig. S23). Positive causal association was observed for genetically determined caffeine intake from tea with TB-BMD in the age strata of 45–60 years in univariable IVW (beta = 0.313; 95% CI: 0.06–0.566; P = 0.015), weighted median, contamination mixture, and MR-PRESSO methods (Table 3 and Supplementary Fig. S23d). Leave-one-out analysis did not show any single instrument was driving the association (Supplementary Fig. S24). After adjustment for smoking, the association with TB-BMD in the age strata of 45–60 years remained significant in MVMR-IVW (beta = 0.326; 95% CI: 0.067–0.585; P = 0.014) and MVMR-PRESSO (beta = 0.326; 95% CI: 0.064–0.588; P = 0.019) methods (Supplementary Fig. S25).

Positive association was observed for genetically determined caffeine intake from coffee with TB-BMD in the age strata of 45–60 years in univariable IVW (beta = 0.24; 95% CI: 0.008–0.472; P = 0.043), contamination mixture, and MR-PRESSO methods, but not in the other four age strata (Table 3 and Supplementary Fig. S26). Leave-one-out analysis showed that no single instrument drove the association (Supplementary Fig. S27). After adjustment for smoking, the positive association in the age strata of 45–60 years remained significant in both MVMR-IVW (beta = 0.251; 95% CI: 0.015–0.488; P = 0.037) and MVMR-PRESSO (beta = 0.251; 95% CI: 0.029–0.474; P = 0.03) analyses (Supplementary Table S28).

In the secondary analysis, univariable and multivariable MR-Egger intercept and MR-PRESSO global tests were all statistically insignificant (Table 3, Supplementary Figs S25 and S28), suggesting the absence of horizontal pleiotropy.

In the supplementary analysis (Supplementary Results 3), no sufficient evidence could support the causal association of genetically determined caffeine intake from tea (Supplementary Table S31 and Supplementary Fig. S29), coffee (Supplementary Table S32 and Supplementary Fig. S30), and combined caffeine intake (Supplementary Table S33 and Supplementary Fig. S31) with LS-BMD, FN-BMD, and FA-BMD, respectively. Leave-one-out analysis suggested that no single instrument was driving the association (Supplementary Figures S32-S34).

Discussion

To the best of our knowledge, this is the first study demonstrating the causal association of genetically determined caffeine intake from tea or coffee with bone health. We showed that genetically higher caffeine intake from tea or coffee had an independent causal effect on improving overall TB-BMD. The positive association was also observed for TB-BMD in the age strata of 45–60 years. Notably, genetically higher caffeine intake from tea was additionally associated with a reduced risk of fracture independently, but such association was not observed for genetically higher caffeine intake from coffee.

The positive causal association of genetically determined caffeine intake from tea or coffee with overall TB-BMD and TB-BMD in the age strata of 45–60 years were robustly supported by the univariable IVW analysis, and the sensitivity analyses of contamination mixture and/or MR-PRESSO methods. Although similar causal estimates were yielded by weighted median method, statistical significance was not achieved in all analyses. This may be attributed to the relative low power of weighted median method when all instruments are valid, or there exists invalid instruments that have balanced pleiotropic effects on the outcome [21]. Notably, the positive causal association of caffeine intake from tea or coffee with TB-BMD remained significant in MVMR analysis after adjustment for smoking, indicating the beneficial effect of caffeine intake from either beverage on TB-BMD is independent. A hypothesized mechanism linking caffeine with bone metabolism is that caffeine is an antagonist of adenosine receptors. Caffeine inhibits the binding of adenosine to its receptors and thus limiting the functions modulated by the adenosine receptor signalling, including osteoclast formation, bone resorption [27], osteoblast proliferation, osteoblast differentiation, and bone formation [28]. While several observational studies were conducted to examine the relationship of caffeine intake with BMD, the results were contradictory. A few cross-sectional studies did not identify any significant association between caffeine intake and BMD [5–8], but one study suggested that caffeine consumption was positively associated with LS BMD in female aged 30–39 while inversely associated with BMD in male aged 40–49 [6]. A meta-analysis revealed an elevated risk of fracture among individuals with the highest caffeine consumption with reference to those with the lowest caffeine consumption [9], while a population-based cohort study detected a J-shaped relationship between caffeine consumption and hip fracture risk [29]. These conflicting findings can be explained by their different study designs, limitation of conventional observational studies such as reverse causality, the heterogenous assessment of caffeine intake, and definition of high versus low caffeine consumption in the constituting studies of meta-analysis. Another possible explanation of the discrepancy is the residual confounding in conventional observational studies. For instance, cigarette smoking was highly correlated with caffeine intake [15, 16], while cigarette smoking was also associated with higher rate of bone loss and increased risk of fracture [30, 31]. Observational studies without adjustment for cigarette smoking as a confounder likely underestimated the genuine association. In the present MR study, genetic instruments associated with known bone-related factors like cigarette smoking and body mass index (BMI) were excluded. The independence and exclusion restriction assumptions were likely valid. Our MVMR analysis further suggested caffeine intake from tea or coffee could improve TB-BMD independent of smoking.

Although genetically determined caffeine intake from tea or coffee had an independent protective effect on bone health, our additional MR analysis did not support any relationship of combined caffeine intake from both tea and coffee with TB-BMD and fracture. One likely explanation is that GWAS of combined caffeine intake identified a substantial proportion of genetic variants that are specifically related to caffeine intake from both beverages, but not either of them [10]. This aligns with our MR study that only up to one-third of genetic instruments for combined caffeine intake was common with those for caffeine intake from tea or coffee. Another possibility may be the weak phenotypic correlation among the three caffeine intake phenotypes. Caffeine intake from tea was inversely and weakly correlated with caffeine intake from coffee (r = −0.33) [10]. While combined caffeine intake had a weak phenotypic correlation with caffeine intake from tea (r = 0.307), its correlation with caffeine intake from coffee was stronger (r = 0.715) [10].

Findings related to caffeine intake from tea in this study is also indicative of daily tea consumption, as the caffeine intake from tea was estimated by multiplying the number of cups of tea daily by 30 mg, the standardized caffeine content per cup of tea [10]. Our current study implies that genetically determined caffeine intake from tea, or genetically predicted tea consumption, is an independent protective factor of bone health in respect of TB-BMD and fracture. Comparison of our study finding with the conflicting results from published studies on the relationship of tea consumption with bone health was presented in Supplementary Discussion 1. Several studies also investigated the association between coffee consumption and bone health. A comparison of the current study with these published studies was made in Supplementary Discussion 2.

This study has potential clinical implication. Our finding suggests the lifelong effect of genetically higher caffeine intake from tea or tea consumption on improvement of the overall TB-BMD and reduction of fracture risk, leading to the hypothesis that appropriate tea consumption may benefit bone health. While BMD increases steadily at childhood and rises sharply during adolescence until peak bone mass is reached at approximately the age of 30s, BMD remains stable till the age of 50s. The age-specific effect of genetically determined caffeine intake on TB-BMD in the age strata of 45–60 years is particularly essential, as bone loss accelerates after the age of 50, while the prevalence of osteoporosis was estimated to be 6.6 and 22.1%, respectively, in male and female aged ≥50 in Europe in 2019 [32]. Tea may be included in the diet as a strategy to improve BMD and prevent osteoporosis-associated fracture, especially due to its benefit on TB-BMD during the age of 45–60 years at which individuals would shortly experience rapid bone loss. Although we observed similar causal effect of genetically higher caffeine intake from coffee on overall TB-BMD in the total sample and the age strata of 45–60 years, our study did not support its effect on fracture reduction. Similar null association was observed for genetically predicted coffee consumption and fracture [33]. It is likely that other active ingredients in tea like fluoride [34] and polyphenols [35], particularly flavonoids [36–39], act together with caffeine to improve bone health. As a caffeinated beverage, tea may serve as a better protector of bone than coffee. Despite the long speculation that caffeine was linked to the development of osteoporosis via altering of calcium metabolism and suppression of vitamin D receptor [40], our study has added new evidence to the current literature of contradictory findings. This would likely facilitate future review of the optimal dietary requirement in improving bone health, as the current MR study is unable to evaluate the previously reported J-shaped relationship between caffeine intake and fracture risk.

This study has several strengths. The MR approach utilizes genetic variants as instrumental variables representing lifelong effect of genetically determined caffeine intake from tea, coffee, or the combined caffeine intake, enabling the investigation of long-term effect on bone health when compared with cohort studies and RCT. By MR-Steiger filtering, we only included genetic instruments which explained the variance of exposure more than the outcome. The instruments are unlikely affected by reverse causality, strengthening the evidence of causal inference. The high F-statistic of the instruments (Table 2) indicated that weak instrument bias is unlikely. Moreover, the finding is robustly supported by various univariable and MVMR analyses with different assumptions. There are also limitations. First, two-sample MR approach assumes a linear relationship between the exposure and outcome. If the genuine relationship between caffeine intake and fracture is J-shaped as reported in a cohort study [29], we may have underestimated the causal effect. Second, the study has limited power in the secondary analysis evaluating the association of genetically determined caffeine intake with TB-BMD in the five age strata, and supplementary analyses assessing the causal effect on LS-BMD, FN-BMD, and FA-BMD. The absence of association may be due to the low power. Future MR studies are warranted when larger GWAS of DXA-derived BMD, preferably age- and sex-specific GWAS, become available. Nevertheless, we had sufficient power in the primary analysis (Supplementary Results 1, Supplementary Figs S1–S6). Third, the GWAS of caffeine intake [10], TB-BMD [14], and fracture [41] comprised UK Biobank participants. Participant overlap between exposure and outcome datasets may cause bias towards the confounded association [24]. Yet, we estimated the bias and Type I error rate to be 0.002–0.004 and 0.05, respectively (Table 2), which is minimal. This aligns with the findings of a simulation study that two-sample MR methods could be applied safely in large single sample like UK Biobank, except for MR-Egger [42]. In addition to the low statistical power of MR-Egger method, this might explain why insignificant associations were observed from our MR-Egger regression analysis. Nonetheless, due to the robust causal estimates obtained from other sensitivity analyses, this does not affect the conclusion of our study. Fourth, as we retrieved summary statistics of caffeine intake from the GWAS conducted by Said et al., this study also shared the same limitations of the GWAS [10]. For instance, tea and coffee drinking habit was collected from self-reported questionnaire at one time-point that did not capture change in the drinking behaviour. The different preparation methods of coffee were not considered. The standardization of caffeine content as 30 mg per cup of tea is considered inaccurate since different types of tea have variable caffeine content. The estimated caffeine intake may deviate from the actual intake [10]. Lastly, the summary statistics were retrieved from GWAS/GWAS meta-analysis of predominantly Europeans, limiting the generalizability of findings to other ethnicities.

To conclude, the current study provides robust evidence that genetically higher caffeine intake from tea and tea consumption may causally improve TB-BMD and reduce fracture risk. Meanwhile, genetically higher caffeine intake from coffee is associated with TB-BMD but not fracture. In addition to caffeine, other active ingredients in tea may play a role in protecting the bone. This may provide insights on the optimal dietary requirements amid the contradictory findings reported by conventional observational studies on the relationship between caffeinated beverages and bone health.

Acknowledgements

The authors would like to thank the study participants of the respective GWAS/GWAS meta-analysis from which the summary statistics were retrieved for this MR study.

Conflict of interest statement

None declared.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Author contributions

G.H.L. and C.L.C. contributed to the conceptualization and study design. G.H.L. performed the data analysis, wrote the draft, and revised the manuscript. C.M.T. contributed to validation and wrote the first draft of the manuscript. S.M.W. contributed to validation. All authors contributed to the data interpretation, and critical review of the manuscript.

Data availability

The summary statistics used in this study can be downloaded from the respective GWAS/GWAS meta-analyses cited in Table 1.

Ethical approval

Ethical approval and informed consent from study participants were obtained from the original GWAS/GWAS meta-analysis from which the genetic data were extracted for the current analysis. No separate ethical approval is required for analysis of these publicly available summary data.

References

Author notes