Serum Metabolome of Coffee Consumption and its Association With Bone Mineral Density: The Hong Kong Osteoporosis Study

Abstract

Background

Inconsistent associations between coffee consumption and bone mineral density (BMD) have been observed in epidemiological studies. Moreover, the relationship of bioactive components in coffee with BMD has not been studied. The aim of the current study is to identify coffee-associated metabolites and evaluate their association with BMD.

Methods

Two independent cohorts totaling 564 healthy community-dwelling adults from the Hong Kong Osteoporosis Study (HKOS) who visited in 2001–2010 (N = 329) and 2015–2016 (N = 235) were included. Coffee consumption was self-reported in an food frequency questionnaire. Untargeted metabolomic profiling on fasting serum samples was performed using liquid chromatography–mass spectrometry platforms. BMD at lumbar spine and femoral neck was measured by dual-energy X-ray absorptiometry. Multivariable linear regression and robust regression were used for the association analyses.

Results

12 serum metabolites were positively correlated with coffee consumption after Bonferroni correction for multiple testing (P < 4.87 × 10^–5), with quinate, 3-hydroxypyridine sulfate, and trigonelline (N’-methylnicotinate) showing the strongest association. Among these metabolites, 11 known metabolites were previously identified to be associated with coffee intake and 6 of them were related to caffeine metabolism. Habitual coffee intake was positively and significantly associated with BMD at the lumbar spine and femoral neck. The metabolite 5-acetylamino-6-formylamino-3-methyluracil (AFMU) (β = 0.012, SE = 0.005; P = 0.013) was significantly associated with BMD at the lumbar spine, whereas 3-hydroxyhippurate (β = 0.007, SE = 0.003, P = 0.027) and trigonelline (β = 0.007, SE = 0.004; P = 0.043) were significantly associated with BMD at the femoral neck.

Conclusions

12 metabolites were significantly associated with coffee intake, including 6 caffeine metabolites. Three of them (AFMU, 3-hydroxyhippurate, and trigonelline) were further associated with BMD. These metabolites could be potential biomarkers of coffee consumption and affect bone health.

Coffee, bone mineral density, osteoporosis, metabolomics study, caffeine

Coffee is a widely consumed beverage. It contains bioactive compounds such as carbohydrates, lipids, nitrogenous compounds, vitamins, minerals, and phenolic compounds (1), which are related to various health outcomes. Large epidemiological studies suggest that habitual coffee consumption is associated with reduced risk of all-cause mortality, cardiovascular mortality, and total cancer (2). Coffee consumption has been shown to also be associated with a reduced risk of type 2 diabetes (3), liver diseases (4), Parkinson disease (5), and depression (5). Based on these findings, dietary guidelines for Americans in 2015 suggested that moderate daily consumption of coffee in healthy adults is associated with reduced risk of total mortality, cardiovascular disease, type 2 diabetes, and Parkinson disease (6). A UK Biobank study demonstrated that coffee intake was associated with reduced all-cause mortality (7), reaffirming that coffee drinking can form part of a healthy diet.

Although coffee intake has been shown to be associated with reduced risk of several diseases and mortality, its association with osteoporosis remains controversial. Caffeine, one of the most studied bioactive components of coffee, adversely affects calcium balance in humans by increasing calcium excretion or decreasing calcium absorption (8). Nonetheless, coffee intake has been associated with both increased (9, 10) and reduced (11, 12) bone mineral density (BMD) in epidemiological studies. Similarly, inconsistent findings have been shown in meta-analyses of observational studies (13). In general, a null association has been suggested between coffee intake and fracture (14), possibly due to a potential gender-specific effect, with daily consumption of coffee associated with increased risk of fracture in women and decreased risk in men (15). The inconsistent findings are likely multifactorial, in which the instrument used, a self-reported food frequency questionnaire (FFQ), could be a contributing factor.

A self-reported FFQ is commonly used to estimate habitual intake in nutritional epidemiological studies. As it relies on participant memory and understanding of portion size, it is subject to measurement and recall bias and not quantifiably precise (16), resulting in inconsistent diet-disease association. A clinical trial is considered the gold standard in inferring causality between coffee and health outcomes, but this is usually not practical unless the outcome can be observed in a short period of time. Moreover, food metabolism is affected by several factors including lifestyle, environment, gut microbiota, and genetics. As a result, there is a large variation between food intake and production of bioactive metabolites among individuals, and hence association with clinical outcomes. Metabolomics has recently been shown to be an objective methodology to reflect individual food intake (17). It measures metabolites in biofluids and may capture the metabolic products of food and better reflect the dietary exposure after metabolism. These metabolites may be related to disease pathogenesis. For example, serum metabolites have been shown to be associated with schizophrenia (18) and osteoporosis (19). In the current study, we aimed to identify biomarkers of coffee intake and evaluate their association with BMD.

Methods

The Hong Kong Osteoporosis Study

All participants were from the Hong Kong Osteoporosis Study (HKOS), a large prospective cohort initiated in 1995 to investigate the incidence of osteoporosis. The detailed study design has been described elsewhere (20). In brief, about 9000 Southern Chinese participants were recruited from the community at baseline (1995–2010). The in-person follow-up visits commenced in 2015, when data related to bone-associated comorbidities, such as metabolic syndrome and sarcopenia, in addition to bone phenotypes, were further collected. For both baseline and follow-up participants, serum samples (≥ 8 hours fasting), the self-reported FFQ, and the questionnaire about lifestyle, physical activity, medical condition, smoking, and drinking habits were collected on the same day. BMD at the lumbar spine and femoral neck were measured using dual energy x-ray absorptiometry (DXA) (Hologic Inc) at the study site by a trained technician. Medical records of study participants, such as demographics, admission, and diagnosis were retrieved from the Clinical Data Analysis Reporting System (CDARS), an electronic medical database managed by the Hong Kong Hospital Authority. Hip fracture diagnoses subsequent to the participant visits were identified by International Classification of Diseases, 9th Revision (ICD-9) code 820.XX in CDARS, which is a validated tool with 100% positive predictive value for identifying hip fractures (21). Ethics approval was obtained from the Institutional Review Board, HKU/HA HKW, HKSAR.

Only participants aged ≥ 20 years were included in this study. Participants with undetermined coffee consumption, missing demographic data, and BMD values were excluded. Among the remaining participants, 340 and 287 unrelated and nonoverlapping participants were selected from the baseline (serum stored for > 10 years) and follow-up (serum stored for < 2 years) study, respectively (22). Medical records from CDARS were screened and only participants without diagnoses of chronic diseases, such as cardiovascular diseases and diabetes, were selected. An extreme phenotype sampling design was adopted in cohort 1 (N = 340) that comprised 170 participants from the baseline with randomly selected “high BMD” and 170 participants from “low BMD,” who had a BMD z-score ≥ +1 and ≤ −1.28 measured at either lumbar spine or total hip, respectively. These participants represented the highest 15th percentile and the lowest 10th percentile of the baseline cohort. Such study designs using extreme phenotypes were reported to enhance statistical power (23). Although T-score was used to define osteoporosis (T < −2.5), it compares the BMD of an individual with healthy 30-year old adults of the same sex, irrespective of the age of the individual under measurement. Nevertheless, age plays an important role in the development of osteoporosis and the risk increases with age. Meanwhile, z-score compares the BMD of an individual among those with same sex and age group. Use of z-scores allows individuals to be defined as having low/high BMD with reference to their specific age. It lowers the chance of misclassification of high or low BMD due to omittance of age as an important risk factor. A random cohort design was adopted for cohort 2 (N = 287), in which the independent participants were randomly selected from the HKOS follow-up cohort. They did not overlap with cohort 1 and were considered a “replication cohort” for the metabolomics study (22). Notably, all fasting serum samples were collected on the same day that clinical assessment was performed and stored at −80°C. Since levels of some metabolomics markers may decrease after prolonged storage, the sample storage time was used as a covariate in the overall statistical analysis. Eleven participants from cohort 1 and 52 from cohort 2 were excluded due to incomplete FFQ and missing values for coffee consumption or covariates. A total of 564 participants (329 in cohort 1, 235 in cohort 2) were included in the final analysis.

Exposure assessment

Coffee intake was measured in both cohorts by self-reported FFQ. Participants were asked how frequently they drank coffee over the last year and how many cups (250 mL per cup) they drank each time. In cohort 2, details were asked about both decaffeinated and regular coffee, and coffee intake was determined by adding both together (as in cohort 1). In the general questionnaire, participants were asked about their smoking status of tobacco products (including cigarettes, pipes, or cigars). They were classified as current or former smokers and nonsmokers. Alcohol consumption was categorized as drinker (current or former) or nondrinker.

Metabolite assessment

All fasting serum samples were collected in the morning and stored at −80°C for approximately 10 years and within 2 years, for cohort 1 and cohort 2, respectively. These samples were sent to Metabolon Inc, without being previously thawed, for untargeted metabolomic profiling analysis using liquid chromatography–mass spectrometry (LC-MS) platforms. Details of the platform have been mentioned previously (24). This measurement platform had been used by multiple large cohort studies, such as TwinsUK (25), Cancer Prevention Study II (CPS-II) (17), and the Atherosclerosis Risk in Communities (ARIC) Study (26). The metabolites within the batch were normalized by dividing an individual’s metabolite level by the batch median of all non-missing values.

Statistical analysis

For descriptive statistics, the demographic characteristics are expressed as mean ± standard deviation (SD) for continuous variables or frequencies for categorical variables, and the differences between 2 cohorts were compared by t-test (continuous variable) and chi-square test (categorical variable). To evaluate the association between serum metabolome and coffee consumption, a multivariable linear regression model was used with adjustment for age, gender, weight, height, sample storage duration, smoking, and drinking habits. The threshold of statistical significance was 4.87 × 10^–5 (0.05/1027 metabolites) after Bonferroni correction for multiple comparisons. For association between coffee-associated metabolites and BMD, multivariable robust regression was used to account for extreme phenotype study design, with adjustment for known risk factors of bone metabolism, including age, gender, weight, height, smoking, and drinking habits. Association between self-reported coffee consumption and BMD was evaluated by multivariable robust regression modeling. As an additional analysis, association between coffee consumption and hip fracture was assessed using the Cox Proportional Hazard model. In Cox regression analysis, follow-up period for each participant was defined as the duration from the date of visit to the date of hip fracture diagnosis, death, or December 31, 2018, whichever earlier. All analysis were performed in R 3.5.1.

Results

In the current study, most participants were female (81.2%) and did not smoke (91.8%) or consume alcohol (84.8%). Compared with cohort 2, participants in cohort 1 were younger and taller. More participants from cohort 1 did not consume alcohol and there were fewer nonsmokers compared with cohort 2. In both cohorts, nearly half did not consume coffee (43.2% in cohort 1 and 41.3% in cohort 2). Overall, 42.4% of participants were non-coffee drinkers (Table 1) and only 14.4% had more than 1 cup of coffee each day.

Table 1.

Demographics of the Hong Kong Osteoporosis Study Participants

	Overall	(N = 564)	Cohort 1	(N = 329)	Cohort 2	(N = 235)	P  4	HKOS	(N = 7004)
Age, years 1	51.7	± 14.3	48.4	± 15.9	56.2	± 10.1	<0.001	53.7	± 16.8
Female, n (%)	458	(81.2%)	233	(70.8%)	225	(95.7%)	<0.001	5216	(74.5%)
Weight, kg	57.1	± 11.3	57.3	± 12.4	56.9	± 9.5	0.638	56.6	± 10.6
Height, cm	157.7	± 7.8	158.6	± 8.4	156.6	± 6.5	0.002	1.6	± 0.1
BMI, kg/m2	22.9	± 3.8	22.7	± 4.0	23.2	± 3.5	0.135	22.7	± 3.6
Sample storage, years	7.6	± 6.1	12.5	± 2.5	0.7	± 0.4	<0.001	NA
Non-alcohol, n (%)	478	(84.8%)	288	(87.5%)	190	(80.9%)	0.029	6196	(88.5%)
Nonsmoker, n (%)2	518	(91.8%)	289	(87.8%)	229	(97.4%)	<0.001	6148	(87.8%)
Coffee (cups/week) 3	1.89	± 3.61	1.77	± 3.81	2.06	± 3.32	0.348	1.438	± 3.1
Non-coffee drinkers, n (%)	239	(42.4%)	142	(43.2%)	97	(41.3%)	0.338	3756	(53.6%)
<1 cup/day, n (%)	244	(43.3%)	135	(41.0%)	109	(46.4%)		2393	(34.2%)
≥1 cup/day, n (%)	81	(14.4%)	52	(15.8%)	29	(12.3%)		855	(12.2%)
BMD, g/cm2
Lumbar spine	0.921	± 0.202	0.929	± 0.224	0.911	± 0.166	0.309	0.890	± 0.170
Femoral neck	0.712	± 0.166	0.724	± 0.192	0.695	± 0.118	0.038	0.682	± 0.136

¹Mean ± SD (all such values).

²Self-reported habitual alcohol intake, smoking of any tobacco (cigarettes, pipes, or cigars).

³Coffee intake was assessed by self-reported food-frequency questionnaire and measured in cups (250 mL) of total coffee per week (included caffeinated and decaffeinated).

⁴Differences between cohort 1 and cohort 2 were assessed by t-test, and chi-square tests for continuous and categorical variables, respectively.

Coffee-associated metabolites

A total of 1194 serum metabolites were profiled in the untargeted metabolomic profiling. Among these, 835 were of known identity and 359 were unknown. Metabolites that were below the detection limit in ≥ 50% of the samples (N = 167) were excluded to reduce noise and increase statistical power. In cohort 1, 88 metabolites were nominally associated with habitual coffee consumption. All supplementary material and figures are located in a digital research materials repository (27). These metabolites were mostly xenobiotics, peptides, and amino acid–related metabolites. Among these, 20 were replicated (as defined by P < 0.05) in cohort 2, with a total of 120 metabolites showing nominal association with habitual coffee consumption (27). Among those replicated metabolites, 12 were positively and significantly associated with coffee consumption after Bonferroni correction in the meta-analysis (cohorts 1 + 2; Table 2). The metabolite with the strongest association was quinate, with β of 0.257 (SE = 0.027; P = 2.33 × 10^–20) cups of coffee consumed per week with every 1 SD increase in this metabolite. Other metabolites with a significant association were 3-hydroxypyridine sulfate, trigonelline (N’-methylnicotinate), 5-acetylamino-6-formylamino-3-methyluracil, 5-acetylamino-6-amino-3-methyluracil, 1-methylxanthine, paraxanthine, 3-methyl catechol sulfate (1), 1-methylurate, X–23639, 1,7-dimethylurate, and 3-hydroxyhippurate (Table 2). There were 8 metabolites showing nominally significant association with coffee consumption in both cohorts 1 and 2 but did not pass the Bonferroni’s significance threshold (27). Eleven of the 20 replicated metabolites were previously shown to be related to coffee consumption (27).

Table 2.

Metabolites Significantly Associated With Habitual Coffee Consumption Among Both Cohort 1 and Cohort 2, and Significant in Overall Analysis

			Cohort 1 (N = 329)			Cohort 2 (N = 235)			Overall (N = 564)
	Metabolites	Sub Pathway	β	SE	P  1	β	SE	P  1	β	SE	P  2
		Xenobiotics
†	Quinate	Food Component/Plant	0.219	0.038	3.02 × 10–08	0.311	0.035	2.45 × 10–16	0.257	0.027	2.33 × 10–20
	3–Hydroxypyridine Sulfate	Chemical	0.363	0.055	2.22 × 10–10	0.275	0.054	8.74 × 10–07	0.329	0.039	1.50 × 10–16
†	5–Acetylamino–6–Formylamino–3–Methyluracil	Xanthine Metabolism	0.919	0.148	1.48 × 10–09	0.411	0.099	4.42 × 10–05	0.584	0.086	2.59 × 10–11
†	5–Acetylamino–6–Amino–3–Methyluracil	Xanthine Metabolism	0.706	0.147	2.51 × 10–06	0.890	0.187	3.45 × 10–06	0.755	0.114	9.53 × 10–11
†	1–Methylxanthine	Xanthine Metabolism	0.830	0.159	3.11 × 10–07	0.580	0.197	3.52 × 10–03	0.755	0.123	1.45 × 10–09
†	Paraxanthine	Xanthine Metabolism	1.130	0.204	6.70 × 10–08	0.556	0.217	1.10 × 10–02	0.908	0.149	2.30 × 10–09
	3–Methyl Catechol Sulfate (1)	Benzoate Metabolism	0.319	0.068	3.93 × 10–06	0.257	0.081	1.75 × 10–03	0.306	0.050	2.57 × 10–09
†	1–Methylurate	Xanthine Metabolism	1.192	0.220	1.16 × 10–07	0.545	0.225	1.63 × 10–02	0.885	0.159	3.86 × 10–08
†	1,7–Dimethylurate	Xanthine Metabolism	0.626	0.155	6.43 × 10–05	0.354	0.156	2.43 × 10–02	0.505	0.111	6.44 × 10–06
	3–Hydroxyhippurate	Benzoate Metabolism	0.357	0.094	1.80 × 10–04	0.334	0.144	2.12 × 10–02	0.350	0.077	6.71 × 10–06
	Cofactors and Vitamins
†	Trigonelline (N’–Methyl nicotinate)	Nicotinate and Nicotinamide Metabolism	0.670	0.118	3.09 × 10–08	0.837	0.154	1.46 × 10–07	0.704	0.093	1.51 × 10–13
	X – 23639	Unknown	2.333	0.742	1.81 × 10–03	2.420	0.586	5.08 × 10–05	2.260	0.474	2.34 × 10–06

P  ¹: Multivariable linear regression model of individual metabolites on coffee consumption, adjusting for age, sex, weight, height, drinking and smoking habit

P  ²: Multivariable linear regression model of individual metabolites on coffee consumption for overall cohort, adjusting for age, sex, weight, height, drinking and smoking habit, and sample storage duration

Only metabolites with association passes the significance threshold for multiple testing (P = 4.87 × 10^-05 for 1027 metabolites) were shown

† Known coffee and caffeine metabolites.

Suffixes (eg, (1)) indicate different sulfation locations on the molecular structure.

Relationship between habitual coffee consumption and BMD

We first tested the association of coffee consumption with BMD in the original HKOS cohort (N = 7004; Table 1) (27). There was a significant association of coffee intake with BMD at the lumbar spine (β = 0.002 per cup of coffee per week; SE = 5.35 × 10^–4; P = 1.81 × 10^–4) and femoral neck (β = 0.001 per cup of coffee per week; SE = 3.77 × 10^–4; P = 8.06 × 10^–4) (Table 3). In the 564 HKOS participants with serum metabolome data, a significant association was observed only for BMD at the lumbar spine (β = 0.005 per cup of coffee per week; SE 2.44 × 10^–3; P = 0.036) (Table 3).

Table 3.

Robust regression model of self-reported coffee consumption and BMD at lumbar spine and femoral neck

	Baseline HKOS Cohort (N = 7004)			Metabolomic Cohort (N = 564)
BMD	β	SE	P	β	SE	P
Lumbar Spine	0.002	5.35 × 10-04	1.81 × 10-04	0.005	2.44 × 10-03	0.036
Femoral Neck	0.001	3.77 × 10-04	8.06 × 10-04	0.002	1.30 × 10-03	0.165

Adjusting for age, sex, weight, height, smoking, and drinking habit

Relationship between habitual coffee consumption and hip fracture

Among the 7004 participants in the HKOS cohort, 951 did not have relevant medical records in CDARS. Among the remaining 6053 participants with a median follow-up of 14.88 years, 5.22% or 316 participants (73 male and 243 female) had hip fracture after the first assessment. Their demographics are shown in Supplementary Table 5 (27). Of the 564 participants with metabolomics data, medical records of 111 participants were unavailable in CDARS and they were excluded from the Cox regression analysis. The median follow-up time for the remaining 453 participants was 10.30 years, with their demographic characteristics summarized in Supplementary Table 6 (27). Eleven of them (3 male and 8 female) had hip fracture after assessment.

Using the Cox regression model with adjustment for age, gender, weight, height, smoking, drinking habit, and history of previous fracture (at hip, spine, wrist, or humerus), higher coffee consumption was associated with reduced fracture risk in both the HKOS and metabolomics cohorts. However, the result was not statistically significant (HKOS cohort: HR = 0.962; 95% CI, 0.906–1.022; P = 0.212; metabolomics cohort: HR = 0.726; 95% CI, 0.332–1.588; P = 0.423) (27).

Relationship between coffee-associated metabolites and BMD

All coffee-associated metabolites, including 10 xenobiotics, 1 cofactor and 1 unknown metabolite, were further evaluated for their association with BMD. Results of robust regression analysis are shown in Table 4. 5-acetylamino-6-formylamino-3-methyluracil (AFMU, β = 0.012, SE 0.005; P = 0.013) was significantly associated with BMD at the lumbar spine, while 3-hydroxyhippurate (β = 0.007, SE = 0.003, P = 0.027) and trigonelline (β = 0.007, SE = 0.004; P = 0.043) were significantly associated with BMD at the femoral neck. Other coffee-associated metabolites did not show significant association with BMD at either site (data not shown).

Table 4.

Robust Regression Model of Coffee-Associated Metabolites and BMD at Lumbar Spine and Femoral Neck

	Overall (N = 564)
Lumbar Spine BMD	β	SE	P  1
5-Acetylamino-6-Formylamino-3- Methyluracil†	0.012	0.005	0.013
X - 23639	0.079	0.027	0.004
Femoral Neck BMD	β	SE	P  1
Trigonelline (N’-Methyl nicotinate)†	0.007	0.004	0.043
3-hydroxyhippurate	0.007	0.003	0.027

P  ¹: Robust regression model adjusting for age, sex, weight, height, smoking, and drinking habit.

Only metabolites with association passing the significance threshold (P = 0.05) are shown.

† Known coffee and caffeine metabolites.

Discussion

In this study, we first identified 12 serum metabolites significantly associated with self-reported habitual coffee intake in community-dwelling Chinese adults, and a significant and positive association between coffee consumptions and BMD. Among these 12 coffee-associated metabolites, 4 of them were significantly associated with BMD. This study identified potential bioactive metabolites that underlie the beneficial effect of coffee on bone metabolism in humans.

Few studies have investigated the potential biomarkers of usual coffee consumption (25, 26, 28–32), and most have used urine samples and been conducted in Caucasian populations with a much higher coffee intake compared with Chinese (33) populations. In our cohort, all known coffee-associated metabolites (N = 11) were also shown to be significantly associated with dose-response coffee consumption in a 3-stage clinical trial (34); while 7 of the coffee-associated metabolites were also shown to be associated with coffee consumption in 498 US adults (27, 28). These findings suggest that measuring serum metabolites can objectively and sensitively reflect habitual coffee consumption, even in a population with relatively low coffee consumption. This statement is further strengthened by looking at the properties of these metabolites. Among the 11 known metabolites, six are related to caffeine metabolism in humans. These 6 metabolites are paraxanthine, one of the major metabolites of caffeine modulated by cytochrome P450 1A2 (CYP1A2) (35) and its secondary metabolites that include 5-acetylamino-6-formylamino-3-methyluracil, 1-methylurate, 1-methylxanthine, 5-acetylamino-6-amino-3-methyluracil (AAMU), and 1,7-dimethylurate (35). Other than caffeine metabolites, quinate (HMDB0003072) and trigonelline (HMDB0000875) are known to be present in coffee (36, 37) and are significant biomarkers of coffee consumption (38). In addition, other coffee-associated metabolites are also directly or indirectly related to coffee. 3-hydroxyhippurate and 3-methyl catechol sulfate are metabolites of chlorogenic acid (CGAs), a nutritional polyphenol mainly found in green coffee beans. These benzoate metabolites are generated by the colonic microbiota or during the roasting process (36). 3-hydroxypyridine sulfate is a trace metabolite of pyridine (vitamin B6), and can be produced in the thermal decomposition of amino acids (39). Among the 8 replicated metabolites that did not pass Bonferroni threshold, N-trimethyl 5-aminovalerate was a significant metabolite associated with habitual milk intake in 2 individual cohorts. It is related to lysine degradation by the gut microbiota and carnitine metabolism involved in energy generation (40). Cytosine is derived from pyrimidine metabolism and is one of the 4 main bases found in DNA and RNA. Androstenediol (3alpha, 17alpha) monosulfate and sphingomyelin (d18:2/24:2) are an androgenic steroid and sphingolipid respectively, both involved in lipid metabolism. These metabolites might reflect the phytosterol content of coffee or a presence in food components served with coffee, such as dairy products or dried milk powder.

In this Chinese cohort (N = 7004), self-reported coffee consumption showed a significant association with BMD at the lumbar spine and femoral neck. This is consistent with a previous Korean study (N = 4066) showing that coffee consumption was positively associated with BMD at both skeletal sites (10). In contrast, a previous Japanese study showed that coffee consumption was positively associated with BMD at the lumbar spine only, not the femoral neck (41). This could be because coffee consumption seems to have a larger effect size on BMD at the lumbar spine than at the femoral neck (42). In addition, a small sample size might have insufficient power to detect an association between coffee consumption and BMD at the femoral neck, as observed in the Japanese study (N = 498) and in our subgroup analysis (N = 564; Table 3). Nevertheless, our study provides evidence that coffee consumption is associated with better bone health. Although inverse yet insignificant association was observed between coffee consumption and risk of hip fracture in both cohorts using Cox regression analysis, this may be explained by the small effect size of coffee consumption on BMD improvement and multifactorial nature of hip fracture.

We further analyzed the association between coffee-associated metabolites and BMD. Three metabolites with known identity were positively associated with BMD: AFMU for lumbar spine BMD, and 3-hydroxyippurate and trigonelline for femoral neck BMD. Identification of functional metabolites in improving bone mass may help in the development of novel therapeutic agents. AFMU is a metabolite of methylxanthines and bioactive compounds abundant in coffee, cocoa, and tea. Methylxanthines mainly act as adenosine receptor antagonist, but also inhibit phosphodiesterase, modulate GABA receptors, and regulate intracellular calcium level (43). Methylxanthines affect many tissues and are involved in regulation of cAMP that influences the activity of nonsteroidal hormones, estrogen, and androgens that might in turn affect bone homeostasis (44). Moreover, most methylxanthines are metabolized in the liver by CYP1A2. This reaction and effect on bones was suggested to be influenced by age, gender, diet, liver function, and external factors such as smoking, alcohol consumption, and calcium intake (13). Thus, individuals may response differently to the same dose of methylxanthines. The high toxicity level probably arises due to pharmacological treatment, not dietary consumption (8). Therefore, it should be considered safe for consumption as part of a typical diet.

The compound 3-hydroxyhippurate (3-hydroxyhippuric acid) is one of the microbial aromatic acid metabolites of the CGAs found in coffee (45). Although there is no evidence of a direct relationship between 3-hydroxyhippuric acid and bone health, it has been shown to be associated with regulation of blood glucose and liver and kidney function (45), which may be related to the antioxidant and anti-inflammatory properties of CGA. An in vivo study demonstrated that supplementing CGAs in ovariectomized rats could promote bone formation biomarkers, improve femoral bone microarchitecture, and increase osteoblast differentiation (46). As CGA is also widely available in fruits and vegetables, the potential role of its metabolites in prevention of bone loss warrants further study.

Trigonelline was significantly associated with BMD at the femoral neck. In a previous in vivo study, trigonelline was beneficial to bone health in nonhyperglycemic mice, but detrimental in hyperglycemic/diabetic mice (47). In the current study, most participants were free of diabetes at the assessment date, thus supporting the beneficial role of trigonelline on BMD in nonhyperglycemic population. Trigonelline is an alkaloid and a metabolite of niacin (vitamin B3). It is found in lentils, oats, and chickpeas, but also commonly as a biomarker of coffee consumption (31). It is reported to exert phytoestrogenic activity and affect bone metabolism. Daily trigonelline administration to rats showed negative bone turnover in the ovariectomized rats but insignificant changes in normal rats (48). The inconsistency with our results is possibly because our study did not perform subgroup analyses by gender or menopausal status due to the limited of sample size.

In our study, the unknown coffee-associated metabolite X-23639 was positively correlated with BMD at the lumbar spine and femoral neck. Another unknown metabolite, X-21442, has been shown to be a dominant predictor of coffee consumption among postmenopausal women (17) although this metabolite was excluded in our study due to very low detection level in majority of the cohort participants. Nevertheless, this may demonstrate a representative association between some unknown metabolites and bone health, and further investigation is needed.

There are several strengths in the study. First, all coffee-associated metabolites with known identity were consistently identified in a previous trial, demonstrating that the data (serum, metabolomics, and FFQ) used in the current study are of high quality. Second, a large sample size was used to elucidate the relationship between coffee consumption and BMD, thus providing a strong statistical power in demonstrating the association with femoral neck BMD; this could not be shown in studies with small sample size. Third, the robust design of 2 cohorts (extreme BMD design for cohort 1 versus random cohort design for cohort 2) in the metabolomics study accounted for potential limitations in storage time (> 10 years in cohort 1; < 2 years in cohort 2). Since degradation of metabolites is possible after prolonged storage, a consistent association in the replication cohort (with a serum storage time < 2 years) indicates a robust association that is less likely to have been affected by storage time. Consistent associations in both cohorts largely reduce false positive results.

Our study had several limitations. Self-reported FFQ was used as the reference to identify coffee-associated metabolites and may have led to a high false-negative rate due to various biases, although the current findings are consistent with similar studies conducted in Caucasians. Metabolites may have deteriorated due to the prolonged storage, even though serum samples were stored properly at −80°C as in the other cohort studies. We tried to address this by using 2 batches of samples obtained at distinct time periods. The sample size of the metabolomics study was not large, thus reducing the power to identify metabolites associated with coffee consumption with smaller effect size, such as theobromine (27). Nevertheless, the current study should have identified metabolites that are associated with coffee consumption with large effect size, and such data should be considered stronger biomarkers of coffee consumption. Like many epidemiological studies, there is residual confounding in the current study.

In conclusion, we identified 12 serum metabolites associated with habitual coffee consumption in a Chinese population. Among these, AFMU, trigonelline, and 3-hydroxyhippurate were also associated with BMD. They may contribute to the beneficial effect of coffee consumption on bone health that was observed in the current study. Identification of functional metabolites in improving BMD may help in the development of novel therapeutic agents.

Abbreviations

Abbreviations
 
• AFMU

  5-acetylamino-6-formylamino-3-methyluracil

 
• BMD

  bone mineral density

 
• CDARS

  Clinical Data Analysis Reporting System

 
• CGA

  chlorogenic acid

 
• DXA

  dual-energy X-ray absorptiometry

 
• FFQ

  food frequency questionnaire

 
• HKOS

  Hong Kong Osteoporosis Study

Acknowledgments

The authors gratefully acknowledge the support from staffs in Hong Kong Osteoporosis Study.

Financial Support: This work is supported by ECS grant funded by the Research Grants Council, HKSAR, China. (27100416).

Author Contributions: Yin-Pan Chau: Data curation, Formal analysis, Investigation, Writing - original draft. Philip C.M. Au: Formal analysis, Investigation. Gloria H.Y. Li: Formal analysis, Investigation, Writing - original draft. Chor-Wing Sing: Investigation, Writing - original draft. Vincent K.F. Cheng: Investigation. Kathryn C.B. Tan: Investigation, Writing - original draft. Annie W.C. Kung: Data curation, Investigation, Writing - original draft. Ching-Lung Cheung: Conceptualization, Data curation, Funding acquisition, Supervision, Investigation, Writing - original draft.

Additional Information

Disclosure Summary: The authors have nothing to disclose. No conflict of interest is declared. Ethical approval was obtained from the Institutional Review Board, HKU/HA HKW, HKSAR No.:UW 15–236.