Metabolites Associated with Coffee Consumption and Incident Chronic Kidney Disease

Abstract

Visual Abstract

Background and objectives 

Moderate coffee consumption has been associated with lower risk of CKD; however, the exact biologic mechanisms underlying this association are unknown. Metabolomic profiling may identify metabolic pathways that explain the association between coffee and CKD. The goal of this study was to identify serum metabolites associated with coffee consumption and examine the association between these coffee-associated metabolites and incident CKD.

Design, setting, participants, & measurements 

Using multivariable linear regression, we identified coffee-associated metabolites among 372 serum metabolites available in two subsamples of the Atherosclerosis Risk in Communities study (ARIC; n=3811). Fixed effects meta-analysis was used to pool the results from the two ARIC study subsamples. Associations between coffee and metabolites were replicated in the Bogalusa Heart Study (n=1043). Metabolites with significant associations with coffee in both cohorts were then evaluated for their prospective associations with incident CKD in the ARIC study using Cox proportional hazards regression.

Results 

In the ARIC study, mean (SD) age was 54 (6) years, 56% were daily coffee drinkers, and 32% drank >2 cups per day. In the Bogalusa Heart Study, mean (SD) age was 48 (5) years, 57% were daily coffee drinkers, and 38% drank >2 cups per day. In a meta-analysis of two subsamples of the ARIC study, 41 metabolites were associated with coffee consumption, of which 20 metabolites replicated in the Bogalusa Heart Study. Three of these 20 coffee-associated metabolites were associated with incident CKD in the ARIC study.

Conclusions 

We detected 20 unique serum metabolites associated with coffee consumption in both the ARIC study and the Bogalusa Heart Study, and three of these 20 candidate biomarkers of coffee consumption were associated with incident CKD. One metabolite (glycochenodeoxycholate), a lipid involved in primary bile acid metabolism, may contribute to the favorable kidney health outcomes associated with coffee consumption. Two metabolites (O-methylcatechol sulfate and 3-methyl catechol sulfate), both of which are xenobiotics involved in benzoate metabolism, may represent potential harmful aspects of coffee on kidney health.

Introduction

Dietary changes are an important low-risk intervention that can improve health outcomes. Current guidelines recommend restricting protein, phosphorous, and sodium intake in patients with CKD (^1–3). In addition, there is increasing evidence on the effects of beverages on kidney health (⁴). Moderate coffee consumption (three to five 8-ounce cups per day) has been associated with lower risk of incident kidney disease (^5–7). However, given the complex chemical composition of coffee, the biologic mechanisms linking coffee to kidney disease are unclear (^5–7). Furthermore, there is a need to identify new biomarkers of food and beverages.

Metabolomic profiling is one approach to understanding biologic mediators of disease associations due to the close proximity of metabolites with disease phenotypes. Metabolomics allows for the detection of small molecules between 50 and 1500 D in size in an untargeted manner. Studies have recently demonstrated that metabolite levels in blood may reflect food consumption and its metabolism (⁸,⁹).

We aimed to use metabolomics to find metabolites that may reflect coffee consumption and to explore the prospective association between these coffee-associated metabolites and incident CKD. We hypothesized that some metabolites, particularly those within the xenobiotics category and xanthine metabolism pathway, would be associated with coffee consumption. In addition, we hypothesized that a subset of metabolites altered by coffee consumption would also be associated with incident CKD.

Materials and Methods

Study Design and Populations

We used two independent study populations to discover and validate coffee-associated metabolites. In the discovery analysis, we identified serum metabolites cross-sectionally associated with coffee consumption in the Atherosclerosis Risk in Communities (ARIC) study. These associations were validated in Bogalusa Heart Study (BHS) participants. We then evaluated associations of validated coffee-associated metabolites with incident CKD in ARIC.

The ARIC study is a prospective community-based cohort that enrolled 15,792 predominantly Black and White men and women between 45 and 64 years of age. In 1987–1989 (visit 1), participants were enrolled from four US communities: Forsyth County, North Carolina; Jackson, Mississippi; suburbs of Minneapolis, Minnesota; and Washington County, Maryland. Participants were followed at clinical visits in 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), 2011–2013 (visit 5), 2016–2017 (visit 6), and 2018–2019 (visit 7). Participants also underwent active surveillance for all deaths, hospitalizations, and development of kidney failure. An ethics committee at each site approved the study protocol, study participants provided informed consent at each study visit, and procedures were followed in accordance with the Declaration of Helsinki (¹⁰).

For this study, we included participants who had metabolomic profiling data from serum specimens obtained during visit 1 (n=4032). We then excluded participants who had implausible levels of total energy intake (n=20; defined as <500 or >3500 kcal/d for women and <700 or >4500 kcal/d for men), who had missing coffee consumption data (n=4), or who had prevalent CKD at visit 1 (n=72), defined as eGFR<60 ml/min per 1.73 m². We additionally excluded participants with missing covariate data: body mass index (BMI; n=3), physical activity (n=14), education (n=7), diabetes status (n=5), antihypertensive medication use (n=3), Dietary Approaches to Stop Hypertension (DASH) diet score (n=74), and alcohol consumption status (n=19). In total, 3811 ARIC study participants were included in the discovery analysis, which was analyzed according to the sample in which metabolites were measured: 1807 participants in sample 1 and 2004 participants in sample 2 (Figure 1).

Figure 1.: Flow chart of selection of study participants in the Atherosclerosis Risk in Communities (ARIC) study. BMI, body mass index; DASH, Dietary Approaches to Stop Hypertension.

BHS is a community-based long-term epidemiologic study in a biracial community in Bogalusa, Louisiana (¹¹). The main study consisted of a series of cross-sectional studies of children and adults performed from 1973 to 2016 (n=16,164). The study’s current core cohort includes 1298 participants born between 1959 and 1979 who were screened at least twice during childhood and twice during adulthood. In this study, we included BHS participants with serum metabolite and dietary data (n=1043). Informed consent was obtained from all BHS participants after detailed explanation of the study, the study was approved by the institutional review boards at all participating institutions, and procedures were followed in accordance with the Declaration of Helsinki (¹²).

Metabolomic Profiling

Metabolites were measured by Metabolon, Inc. (Durham, NC) using an untargeted, ultrahigh-performance liquid chromatography-tandem mass spectrometry approach using fasting serum samples from visit 1 in the ARIC study and using serum samples from the 2013–2016 visit in BHS (¹³,¹⁴). In the ARIC study, serum samples were analyzed separately in two samples, with sample 1 consisting of a random sample of ARIC study participants at the Jackson, Mississippi site and sample 2 consisting of ARIC participants with sequencing data; 374 known metabolites and 759 known metabolites were identified in sample 1 and sample 2, respectively. A total of 372 known metabolites that were identified in both samples were used in the primary analysis. In BHS, 1055 known metabolites were identified. Metabolite values were rescaled to a median of one and log transformed for analysis due to their non-normal distribution.

Coffee Consumption

In the ARIC population, coffee consumption was assessed via a 66-item semiquantitative food frequency questionnaire, which was administered in person by a trained interviewer at visit 1. Participants were asked to report how frequently they consumed an 8-ounce cup of regular (non-decaffeinated) coffee on average over the past year, with nine options ranging from “almost never” to “>6 cups per day.” In BHS, coffee consumption was assessed in a validated (¹⁵), self-administered, semiquantitative 131-item food frequency questionnaire (¹⁶). Participants were asked how frequently they typically consumed 1 cup of “not decaf” coffee over the past year. Additional information on the assessment of covariates is available in Supplemental Material.

Definition of Incident CKD

Incident CKD was defined by at least one of the following four criteria: (1) development of reduced kidney function (eGFR<60 ml/min per 1.73 m²) accompanied by ≥25% eGFR decline at any subsequent study visit relative to baseline, (2) International Classification of Diseases 9/10 (ICD 9/10) code for a hospitalization related to CKD stage 3+ identified through active surveillance, (3) ICD 9/10 code for a death related to CKD stage 3+ identified through linkage to the National Death Index, or (4) kidney failure identified by linkage to the United States Renal Data System registry (¹⁷).

Statistical Analyses

For the discovery analysis of the cross-sectional association between coffee and metabolites in ARIC, we used multivariable linear regression models conducted separately for the two samples to calculate β-coefficients and 95% confidence intervals adjusted for age, sex, race-center, education, BMI, physical activity, total energy intake, eGFR, smoking, alcohol consumption, and DASH diet score. Fixed effects meta-analysis was used to pool the results from the two ARIC samples. All metabolites available in both samples of ARIC (n=372) were included in the multivariable linear regression models in ARIC, and metabolites significantly associated with coffee consumption in the meta-analysis of ARIC were included in the replication analysis in BHS. For the replication analysis in BHS, we used the same linear regression model but replaced the race-center covariate with race. For the main analysis, coffee was analyzed as a continuous variable. In addition, we assessed whether there was a dose-response relationship between coffee, as a categorical variable, and metabolites.

The prospective analysis was limited to metabolites that were statistically significantly associated with coffee consumption in both the ARIC meta-analysis and BHS replication analysis. Cox proportional hazards models were used in both ARIC samples to calculate hazard ratios and and 95% confidence intervals for the association between metabolites and risk of incident CKD adjusted for age, sex, race-center, education, BMI, physical activity, total energy intake, eGFR, smoking, alcohol consumption, DASH diet score, diabetes, systolic BP, and antihypertensive medication use. Follow-up time was calculated from baseline until the date of an event (incident CKD), the date of a death (for deaths that were unrelated to CKD), or the end of the study. There was no loss to follow-up given that the CKD definition combined surveillance with visit-based measures, and no participants withdrew consent for follow-up.

For those metabolites that were statistically significantly associated with both coffee and incident CKD, we investigated whether the metabolites were associated with other lifestyle factors, and we calculated differences in absolute risk of incident CKD according to quartiles of metabolites with quartile 1 as the reference group.

Statistical tests were two tailed. Statistical significance was assessed after accounting for multiple comparisons using the Bonferroni approach (i.e., P=0.05/372=1.34 × 10⁻⁴ for both samples of ARIC and the meta-analysis for the cross-sectional analysis of metabolites and coffee, P=0.05/40=1.25 × 10⁻³ for BHS, and P=0.05/20=0.0025 for the prospective analysis between coffee and incident CKD). All analyses were performed using Stata software (version 15.0; StataCorp).

Results

Baseline Characteristics

In the overall ARIC study population, mean (SD) age was 54 (6) years, 60% were women, and 61% were Black (Table 1). About a third of ARIC participants had at least some college education, mean BMI was 28 kg/m², and 28% were current smokers. Sample 1 was composed entirely of Black participants from Jackson, Mississippi, whereas in sample 2, 27% of participants were Black and all four ARIC study centers were represented. Participants who consumed coffee more frequently were less likely to be Black and to have diabetes, had lower BMI and systolic BP, and were more likely to be current smokers and current drinkers (Supplemental Table 1).

Table 1. - Baseline characteristics of the Atherosclerosis Risk in Communities study and Bogalusa Heart Study participants

Baseline Characteristics, n (%) or Mean ± SD	Atherosclerosis Risk in Communities			Bogalusa Heart Study
	Sample 1, n=1807	Sample 2, n=2004	Total Sample, n=3811	Total Sample, n=1043
Age, yr	53±6	54±6	54±6	48±5
Women	1166 (65)	1135 (57)	2301 (60)	627 (60)
Race
White	0 (0)	1470 (73)	1470 (39)	717 (69)
Black	1807 (100)	534 (27)	2341 (61)	326 (31)
Education level
Less than or equal to high school or equivalent	1240 (69)	1291 (64)	2531 (66)	495 (48)
Greater than or equal to college	567 (31)	713 (36)	1280 (34)	548 (53)
Body mass index, kg/m2	29.6±6	27.9±5	28.7±6	31.4±7.6
Center
Forsyth County, North Carolina	0 (0.0)	572 (29)	572 (15)	—
Jackson, Mississippi	1807 (100)	410 (20)	2217 (58)	—
Minneapolis, Minnesota	0 (0.0)	506 (25)	506 (13)	—
Washington County, Maryland	0 (0.0)	516 (26)	516 (14)	—
Smoking status
Never smoker	887 (49)	812 (41)	1699 (45)	559 (54)
Former smoker	406 (22)	643 (32)	1049 (28)	306 (29)
Current smoker	514 (28)	549 (27)	1063 (28)	178 (17)
Alcohol intake status
Never	891 (49)	491 (25)	1382 (36)	120 (12)
Former	375 (21)	388 (19)	763 (20)	345 (33)
Current	541 (30)	1125 (56)	1666 (44)	579 (56)
Total energy intake, kcal/d	1588±617	1662±608	1627±613	2240±1029
Baseline eGFR, ml/min per 1.73 m2	114±17	102±15	108±17	94±17
Physical activity a	2.1±0.7	2.4±0.8	2.3±0.8	4905±4669
Systolic BP, mm Hg	128±21	122±20	125±21	123±17
Diabetes	288 (16)	225 (11)	513 (13)	184 (18)
Antihypertensive medication	730 (40)	642 (32)	1372 (36)	354 (34)
Coffee consumption, 8-ounce cups per d
>3	78 (4)	347 (17)	425 (11)	207 (20)
2–3	331 (18)	473 (24)	804 (21)	190 (18)
1	519 (29)	375 (19)	894 (23)	50 (5)
<1	230 (13)	264 (13)	494 (13)	185 (18)
Almost never	649 (36)	545 (27)	1194 (31)	411 (39)

—, not applicable (the centers listed are involved in the Atherosclerosis Risk in Communities study).

^aPhysical activity was quantified as a score (with a range from one to five) in the Atherosclerosis Risk in Communities study and as metabolic equivalent task (MET) minutes/week in the Bogalusa Heart Study.

In BHS, mean (SD) age was 48 (5) years, 60% were women, and 31% were Black. A little over half of participants continued their education past high school, mean BMI was 31 kg/m², 17% were current smokers, and 56% were current drinkers. In BHS, a slightly larger proportion of participants drank 2 or more cups of coffee per day (38% in BHS versus 32% in ARIC), but a smaller proportion drank 1 cup of coffee per day (5% in BHS versus 24% in ARIC).

Association between Serum Metabolites and Coffee Consumption

In sample 1 of ARIC, 13 metabolites were statistically significantly associated with coffee consumption (Supplemental Table 2). In sample 2 of ARIC, 37 metabolites were statistically significantly associated with coffee consumption (Supplemental Table 3), including 12 of the 13 that were statistically significantly associated with coffee in sample 1.

In the meta-analysis of both ARIC samples, 41 metabolites were statistically significantly associated with coffee consumption (Table 2). The majority of significant metabolites were xenobiotics (n=23), followed by lipids (n=9), amino acids (n=6), carbohydrates (n=1), peptides (n=1), and energy (n=1) (Figure 2).

Table 2. - Metabolites associated with coffee intake in a meta-analysis of two subsamples of the Atherosclerosis Risk in Communities study and the Bogalusa Heart Study

Metabolite	Superpathway	Subpathway	Atherosclerosis Risk in Communities		Bogalusa Heart Study
			β (95% Confidence Interval)	P Value	β (95% Confidence Interval)	P Value
Quinate	Xenobiotics	Food component/plant	0.311 (0.277 to 0.345)	7.14×10−73	0.438 (0.389 to 0.487)	1.00×10−28
Paraxanthine	Xenobiotics	Xanthine metabolism	0.292 (0.258 to 0.327)	5.74×10−62	0.097 (0.067 to 0.126)	2.08×10−10
1-Methylurate	Xenobiotics	Xanthine metabolism	0.239 (0.211 to 0.267)	3.80×10−61	0.098 (0.075 to 0.121)	1.96×10−16
3-Hydroxypyridine sulfate	Xenobiotics	Chemical	0.223 (0.195 to 0.251)	1.72×10−55	0.185 (0.153 to 0.217)	1.30×10−28
1-Methylxanthine	Xenobiotics	Xanthine metabolism	0.189 (0.164 to 0.215)	9.62×10−48	0.130 (0.100 to 0.160)	1.61×10−16
5-Acetylamino-6-amino-3-methyluracil	Xenobiotics	Xanthine metabolism	0.226 (0.194 to 0.258)	2.68×10−44	0.145 (0.114 to 0.176)	1.45×10−19
Theophylline	Xenobiotics	Xanthine metabolism	0.208 (0.175 to 0.242)	8.27×10−35	0.105 (0.074 to 0.136)	3.67×10−11
Caffeine	Xenobiotics	Xanthine metabolism	0.215 (0.178 to 0.252)	1.13×10−29	0.060 (0.017 to 0.104)	0.006
Catechol sulfate	Xenobiotics	Benzoate metabolism	0.102 (0.084 to 0.120)	2.63×10−29	0.051 (0.032 to 0.070)	1.81×10−7
Hippurate	Xenobiotics	Benzoate metabolism	0.135 (0.111 to 0.158)	5.91×10−29	0.048 (0.025 to 0.070)	3.27×10−5
1,7-Dimethylurate	Xenobiotics	Xanthine metabolism	0.152 (0.125 to 0.179)	5.01×10−28	0.097 (0.065 to 0.128)	2.59×10−9
3-Methyl catechol sulfate (1)	Xenobiotics	Benzoate metabolism	0.134 (0.108 to 0.161)	2.87×10−23	0.093 (0.061 to 0.126)	1.68×10−8
O-methylcatechol sulfate	Xenobiotics	Benzoate metabolism	0.081 (0.062 to 0.100)	2.70×10−16	0.040 (0.018 to 0.060)	2.10×10−4
Theobromine	Xenobiotics	Xanthine metabolism	0.126 (0.095 to 0.156)	1.03×10−15	0.043 (0.013 to 0.073)	0.005
3-Phenylpropionate (hydrocinnamate)	Amino acid	Phenylalanine and tyrosine metabolism	0.098 (0.074 to 0.122)	2.03×10−15	0.032 (0.004 to 0.060)	0.03
3-Hydroxyhippurate	Xenobiotics	Benzoate metabolism	0.136 (0.101 to 0.170)	6.04×10−15	0.095 (0.056 to 0.134)	1.53×10−6
7-Methylxanthine	Xenobiotics	Xanthine metabolism	0.102 (0.075 to 0.128)	4.19×10−14	0.077 (0.051 to 0.104)	1.33×10−8
Cinnamoylglycine	Xenobiotics	Food component/plant	0.106 (0.077 to 0.136)	1.16×10−12	0.057 (0.015 to 0.099)	0.007
Homostachydrine	Xenobiotics	Food component/plant	0.054 (0.039 to 0.069)	1.84×10−12	0.035 (0.019 to 0.051)	1.72×10−5
1,3,7-Trimethylurate	Xenobiotics	Xanthine metabolism	0.109 (0.078 to 0.140)	4.91×10−12	0.060 (0.027 to 0.093)	3.37×10−4
3-Methylxanthine	Xenobiotics	Xanthine metabolism	0.078 (0.050 to 0.107)	8.92×10−08	0.073 (0.047 to 0.099)	3.06×10−8
Docosahexaenoate (DHA; 22:6n3)	Lipid	Polyunsaturated fatty acid (n3 and n6)	−0.022 (−0.031 to −0.013)	4.51×10−07	0.004 (−0.007 to 0.015)	0.50
1,3-Dimethylurate	Xenobiotics	Xanthine metabolism	0.090 (0.055 to 0.125)	4.94×10−07	0.121 (0.095 to 0.148)	2.52×10−18
Adrenate (22:4n6)	Lipid	Polyunsaturated fatty acid (n3 and n6)	−0.027 (−0.038 to −0.016)	1.64×10−06	0.001 (−0.012 to 0.014)	0.89
Dihomo-linoleate (20:2n6)	Lipid	Polyunsaturated fatty acid (n3 and n6)	−0.024 (−0.035 to −0.014)	1.83×10−06	−0.002 (−0.014 to 0.010)	0.71
Erythronate	Carbohydrate	Aminosugar metabolism	−0.017 (−0.024 to −0.010)	2.94×10−06	−0.003 (−0.009 to 0.003)	0.30
Malate	Energy	TCA cycle	−0.018 (−0.026 to −0.010)	6.19×10−06	−0.006 (−0.014 to 0.001)	0.09
Stearidonate (18:4n3)	Lipid	Polyunsaturated fatty acid (n3 and n6)	−0.033 (−0.048 to −0.019)	6.33×10−06	0.005 (−0.018 to 0.027)	0.69
Glycocholate	Lipid	Primary bile acid metabolism	−0.069 (−0.099 to −0.039)	6.53×10−06	−0.040 (−0.068 to −0.012)	0.006
Cyclo(leu-pro)	Peptide	Dipeptide	0.048 (0.027 to 0.069)	7.12×10−06	—	—
α-Hydroxyisovalerate	Amino acid	Leucine, isoleucine, and valine metabolism	−0.031 (−0.045 to −0.018)	7.31×10−06	−0.024 (−0.039 to −0.009)	0.003
Isovalerate	Amino acid	Leucine, isoleucine, and valine metabolism	−0.019 (−0.027 to −0.010)	1.53×10−05	0.002 (−0.007 to 0.012)	0.65
O-sulfo-L-tyrosine	Xenobiotics	Chemical	−0.015 (−0.022 to −0.008)	1.84×10−05	−0.003 (−0.009 to 0.003)	0.37
Erythritol	Xenobiotics	Food component/plant	−0.014 (−0.021 to −0.008)	1.94×10−05	−0.002 (−0.013 to 0.009)	0.79
Phenol sulfate	Amino acid	Phenylalanine and tyrosine metabolism	−0.045 (−0.066 to −0.024)	2.09×10−05	0.009 (−0.012 to 0.029)	0.41
Docosapentaenoate (n3 DPA; 22:5n3)	Lipid	Polyunsaturated fatty acid (n3 and n6)	−0.025 (−0.037 to −0.013)	3.90×10−05	0.006 (−0.006 to 0.018)	0.36
Myristoleate (14:1n5)	Lipid	Long-chain fatty acid	−0.026 (−0.039 to −0.014)	4.63×10−05	0.005 (−0.012 to 0.021)	0.58
2-Hydroxy-3-methylvalerate	Amino acid	Leucine, isoleucine, and valine metabolism	−0.036 (−0.054 to −0.019)	5.16×10−05	−0.023 (−0.037 to −0.009)	0.001
Palmitoleate (16:1n7)	Lipid	Long-chain fatty acid	−0.024 (−0.036 to −0.012)	6.39×10−05	0.001 (−0.017 to 0.018)	0.93
Glycine	Amino acid	Glycine, serine, and threonine metabolism	0.016 (0.008 to 0.025)	8.66×10−05	0.004 (−0.003 to 0.011)	0.28
Glycochenodeoxycholate	Lipid	Primary bile acid metabolism	−0.083 (−0.125 to −0.041)	1.14×10−04	−0.049 (−0.074 to −0.023)	2.31×10−4

Adjusted for age, sex, race, education, cigarette smoking, alcohol drinking, physical activity, total energy intake, Dietary Approaches to Stop Hypertension diet score, tea consumption, eGFR, and body mass index. Bold P values are statistically significant (Bonferroni-corrected thresholds for statistical significance of P=0.05/372=1.34 × 10⁻⁴ for the Atherosclerosis Risk in Communities study and P=0.05/40=1.25 × 10⁻³ for Bogalusa Heart Study). DHA, docosahexaenic acid; TCA, tricarboxylic acid cycle; —, not applicable [cyclo(leu-pro) was not available in the Bogalusa Heart Study]; DPA, docosapentaenoic acid.

Figure 2.: Volcano plot depicting β -coefficients and −log ₁₀ P values from meta-analysis of serum metabolites and coffee consumption in the ARIC study. Each dot represents a single metabolite. Metabolites above the red horizontal line were statistically significant (P=0.05/372=1.34 × 10⁻⁴). The color of the dots represents the superpathway classification of the metabolites.

Of 41 metabolites statistically significantly associated with coffee consumption in the meta-analysis of ARIC samples, 40 metabolites were available to be analyzed in BHS (Table 2). Among these 40 metabolites, 20 metabolites replicated in terms of the direction and statistical significance of their associations with coffee consumption (P=0.05/40=1.25 × 10⁻³), including 18 metabolites positively associated and two metabolites inversely associated with coffee consumption. The replicated metabolites included 18 xenobiotics, one amino acid, and one lipid.

For all of the 20 replicated metabolites, there was a dose-response relationship across categories of coffee consumption (Supplemental Table 4).

Association between Serum Metabolites and Incident CKD

Among ARIC participants, there were 1180 (31%) cases of incident CKD that occurred over a median follow-up of 24 years. Of the 20 metabolites consistently associated with coffee consumption (i.e., in the ARIC meta-analysis and the replication analysis in BHS), three metabolites were statistically significantly associated with incident CKD (Figure 3). Glycochenodeoxycholate, a lipid involved in primary bile acid metabolism, was inversely associated with coffee consumption and positively associated with incident CKD (Table 3). O-methylcatechol sulfate and 3-methyl catechol sulfate (¹), two xenobiotics representative of benzoate metabolism, were positively associated with both coffee consumption and incident CKD.

Figure 3.: Volcano plot depicting hazard ratios and −log ₁₀ P values from prospective analysis of coffee-associated serum metabolites and incident CKD in the ARIC study. Each dot represents a single metabolite. Metabolites above the red horizontal line were statistically significant (P=0.05/20=0.0025). The color of the dots represents the superpathway classification of the metabolites.

Table 3. - Serum metabolites statistically significantly associated with coffee consumption and incident CKD in the Atherosclerosis Risk in Communities study

Metabolite	Superpathway	Subpathway	Cross-Sectional Analysis (Metabolites and Coffee) a		Prospective Analysis (Metabolites and CKD) b
			β (95% Confidence Interval)	P Value	Hazard Ratio (95% Confidence Interval)	P Value
Glycochenodeoxycholate	Lipid	Primary bile acid metabolism	−0.083 (−0.125 to −0.041)	1.14×10−4	1.05 (1.02 to 1.07)	6.44×10−4
O-methylcatechol sulfate	Xenobiotics	Benzoate metabolism	0.081 (0.062 to 0.100)	2.70×10−16	1.14 (1.07 to 1.21)	5.98×10−5
3-Methyl catechol sulfate (1)	Xenobiotics	Benzoate metabolism	0.134 (0.108 to 0.161)	2.87×10−23	1.09 (1.04 to 1.14)	1.39×10−4

Statistical significance was assessed after accounting for multiple comparisons using the Bonferroni approach (P=0.05/20=0.0025).

^aAdjusted for age, sex, race-center, education, body mass index, physical activity, total energy intake, eGFR, smoking, alcohol consumption, and Dietary Approaches to Stop Hypertension diet score.

^bAdjusted for age, sex, race-center, education, body mass index, physical activity, total energy intake, eGFR, smoking, alcohol consumption, Dietary Approaches to Stop Hypertension diet score, diabetes, systolic BP, and antihypertensive medication use. Hazard ratios are expressed per one unit higher in the metabolite.

There were small but statistically significantly higher risk of incident CKD for the highest versus lowest quartile for glycohenodeoxycholate, O-methylcatechol sulfate, and 3-methyl catechol sulfate (Table 4).

Table 4. - Absolute risk differences by quartile for serum metabolites statistically significantly associated with coffee consumption and incident CKD in the Atherosclerosis Risk in Communities study

Metabolite and Quartile	Absolute Risk Difference (95% Confidence Interval)	P Value
Glycochenodeoxycholate
Quartile 1	Reference
Quartile 2	0.00047 (−0.0019 to 0.0029)	0.70
Quartile 3	0.0010 (−0.0014 to 0.0035)	0.41
Quartile 4	0.0035 (0.0009 to 0.0061)	0.008
O-methylcatechol sulfate
Quartile 1	Reference
Quartile 2	0.00053 (−0.0019 to 0.0029)	0.67
Quartile 3	0.0040 (0.0013 to 0.0067)	0.003
Quartile 4	0.0088 (0.0059 to 0.0117)	<0.001
3-Methyl catechol sulfate (1)
Quartile 1	Reference
Quartile 2	0.0026 (0.0002 to 0.0050)	0.04
Quartile 3	0.0047 (0.0021 to 0.0074)	<0.001
Quartile 4	0.0094 (0.0065 to 0.0123)	<0.001

The three metabolites that were associated with coffee and incident CKD (glycohenodeoxycholate, O-methylcatechol sulfate, and 3-methyl catechol sulfate) were also associated with current smoking status, with higher levels of the metabolites among current smokers (Supplemental Table 5).

Discussion

Leveraging untargeted metabolomic data of known serum metabolites, we discovered 41 metabolites significantly associated with coffee consumption in 3811 middle-aged Black and White men and women in the ARIC study. Twenty of the 41 metabolites replicated in an independent sample of 1043 Black and White men and women in BHS. There was a dose-response relationship between metabolites and categories of coffee consumption. Higher levels of three of these coffee-related metabolites were significantly associated with higher relative and absolute risks of developing kidney disease during follow-up: glycochenodeoxycholate, O-methylcatechol sulfate, and 3-methyl catechol sulfate.

Of the 20 metabolites that were associated with coffee consumption in the ARIC meta-analysis and in BHS, ten metabolites (1-methylurate; 1,3,7-trimethylurate; 1,7-dimethylurate; 1,3-dimethylurate; 1-methylxanthine; 3-methylxanthine; 7-methylxanthine; theophylline; 5-acetylamino-6-amino-3-methyluracil; and paraxanthine) were involved in xanthine metabolism. Caffeine (1,3,7-trimethylxanthine) is a well-known component of coffee beverages, and theophylline, paraxanthine, and 1,3,7-trimethylurate are all direct products of caffeine metabolism by the cytochrome P450 oxidase enzyme system in the liver (¹⁸,¹⁹).

Another five coffee-associated metabolites (hippurate, 3-hydroxyhippurate, catechol sulfate, 3-methyl catechol sulfate, and O-methylcatechol sulfate) were involved in benzoate metabolism. Benzoate is both present in coffee and a downstream metabolite of chlorogenic acids that are naturally present in coffee, and benzoic acid can be formed by gut microbial degradation of dietary phenols, polyphenols, and tannins present in coffee (^20–22). Hippurate has been identified as a potential serum marker of dietary acid load in a metabolomics study of 689 patients with CKD in the African American Study of Kidney Disease and Hypertension and 356 patients with CKD in the Modification of Diet in Renal Disease study, and coffee is known to affect dietary acid load (²³).

Previous metabolomic studies of coffee consumption provide support for our findings by identifying some of the same metabolites as potential biomarkers for coffee consumption. One study analyzed the serum metabolome of 47 Finnish habitual coffee consumers who underwent an investigator-blinded, three-stage clinical trial where participants refrained from drinking coffee for 1 month, consumed 4 cups of coffee per day in the second month, and consumed 8 cups per day in the third month (²⁴). This study replicated the positive association of all 18 xenobiotics that were associated with coffee in both ARIC and BHS. Another nested patient-control study of 251 patients with colorectal cancer and 247 matched controls from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial in the United States found associations between coffee consumption and nine (quinate; paraxanthine; catechol sulfate; 1-methylxanthine; theophylline; 1-methylurate; 1,7-dimethylurate; 3-hydroxyhippurate; and 1,3,7-trimethylurate) of the 20 replicated coffee-associated metabolites from this study (¹⁹). A cross-sectional analysis of 1595 women in two separate cohorts in the United States also found several of our replicated coffee-associated metabolites to be associated with coffee consumption, including xanthine metabolism metabolites (1,7-dimethylurate; 7-methylxanthine; and 5-acetylamino-6-amino-3-methyluracil) (²⁵). Lastly, a previous study focused on metabolites associated with food groups in only 1977 Black ARIC study participants (sample 1) found nine metabolites to be associated with coffee consumption (quinate; paraxanthine; 5-acetylamino-6-amino-3-methyluracil; 1,7-dimethylurate; 1-methylurate; 1-methylxanthine; caffeine; 1,3,7-trimethylurate; and 7-methylxanthine). All nine of these metabolites were significantly associated with coffee consumption in this ARIC meta-analysis, and eight of the nine metabolites (all metabolites except for caffeine) also replicated in BHS (²⁶). Further research is needed to confirm these metabolites as potential objective biomarkers of coffee consumption.

This is one of the first population-based studies to assess the role of coffee-associated metabolites in kidney disease among a diverse sample of middle-aged adults. We were able to identify three coffee-associated metabolites that were associated with incident CKD. Glycochenodeoxycholate, a lipid representative of primary bile acid metabolism, was negatively associated with coffee consumption but positively associated with risk for incident CKD in this analysis. Primary bile acids are synthesized from cholesterol in the liver and serve an important role in regulating lipid, glucose, energy, and cholesterol homeostasis (²⁷,²⁸). The metabolism of primary bile acids contributes toward the bile acid pool, the composition of which has been hypothesized to be associated with several disease states, including metabolic syndrome (²⁸,²⁹). In a nested case-control study of 221 patients with liver cancer and 242 patients with fatal liver disease in the α-Tocopherol, β-Carotene Cancer Prevention cohort, glycochenodeoxycholic acid was found to be inversely associated with coffee and positively associated with both liver cancer and fatal liver disease (³⁰). However, this study is the first to find an association between glycochenodeoxycholate and kidney disease, and the exact mechanisms through which this metabolite could potentially confer a harmful effect on the kidneys are unknown.

Two other metabolites, O-methylcatechol sulfate and 3-methyl catechol sulfate, were positively associated with both coffee consumption and higher risk for incident CKD. These two metabolites are both xenobiotics belonging to the benzoate metabolism pathway. Benzoate is a downstream metabolite of chlorogenic acids present in coffee. Chlorogenic acids have been hypothesized to be a potential contributor to the observed association between coffee consumption and lower kidney disease risk, as they may inhibit glucose absorption, reduce oxidative stress, and reduce liver glucose output, thereby lowering the risk of diabetic nephropathy (⁵,^31–35). Furthermore, chlorogenic acids, along with other chemical components within coffee, may protect the glomerular endothelium from oxidative stress and reduce both insulin resistance and systemic inflammation (³⁶,³⁷). However, the direction of association of both metabolites with coffee and incident CKD indicates that these metabolites may reflect potentially harmful aspects of the coffee on kidney health. Coffee preparation methods and the temperature and duration of coffee bean roasting, in particular, influence the presence of chlorogenic acids in coffee and may affect the health implications of coffee consumption.

All three metabolites—glycochenodeoxycholate, O-methylcatechol sulfate, and 3-methyl catechol sulfate—that were associated with coffee and incident CKD were also significantly associated with cigarette smoking in this study. It is known that coffee consumption is associated with cigarette smoking (³⁸). Although we controlled for smoking status in our study, because of residual confounding, levels of these metabolites may in part be influenced by smoking status.

Although no other studies have found these three specific metabolites to also be associated with CKD, there have been a number of other studies that have identified serum metabolites associated with CKD (³⁹,⁴⁰). Of note, xanthine and 1-methylxanthine concentrations were associated with prevalent CKD in a case-control study in China, both of which belong to the xanthine metabolism pathway that has been consistently associated with coffee consumption (¹⁹,^24–26,⁴¹). However, both metabolites were not associated with incident CKD in this analysis. Another community-based cohort of 1104 older adults in Germany found spermidine to be associated with incident CKD (⁴²). However, the study only analyzed 140 serum metabolites and may not have included the three metabolites found to be associated with coffee and incident CKD in this study. Spermidine was not available in this analysis.

This study had several strengths. We analyzed data from a large study population with continuous surveillance for incident CKD over 30+ years in our discovery analysis. We replicated the cross-sectional analysis of coffee and metabolites in an independent sample (i.e., BHS). Additionally, our discovery study population was recruited from a variety of locations, so it may be more representative of the general US population than a study with participants from a single location. We adjusted for a large number of potential covariates to account for confounding. We used a stringent procedure for determining statistical significance to account for multiple comparisons and minimize the likelihood of false-positive findings.

Our study also had a number of limitations. We used baseline data on coffee consumption, metabolites, and covariates, and as such, our analysis did not capture the time-varying nature of these variables in the prospective analysis of incident CKD. Many factors affect the metabolome, and although we adjusted for a wide range of covariates, there is still potential for unmeasured confounders. There are also potential reporting biases in our study, as coffee consumption and many of the covariates were self-reported through questionnaires. Despite the two samples being selected from the same underlying study population (the ARIC study), there were differences between the two samples (i.e., sample 1 consisted entirely of Black participants from Jackson, Mississippi). Because of the lack of incident CKD data in BHS, we were unable to replicate the prospective association between metabolites and incident CKD. Using untargeted metabolomics for biomarker discovery optimizes the opportunity for novel findings but limits the ability to replicate across independent study populations.

In conclusion, we found 41 unique metabolites associated with coffee consumption in a meta-analysis of two subsamples of ARIC and replicated the findings in terms of direction and statistical significance for 20 of these metabolites among BHS participants. Some of these serum metabolites have been previously found to be associated with coffee consumption in other populations and are promising objective, quantifiable biomarkers of coffee intake. We also found three metabolites that were associated with coffee consumption and incident CKD: glycochenodeoxycholate, O-methylcatechol sulfate, and 3-methyl catechol sulfate. With replication of our findings and more research on the metabolic underpinnings of the coffee-kidney relationship, these metabolites may represent pathophysiologic processes that are relevant for preventing kidney disease through diet modification.

Disclosures

E. Boerwinkle reports ownership interest in Codified Genomics. J. Coresh reports consultancy agreements with Healthy.io, Kaleido, and Ultragenyx; ownership interest in Healthy.io; receiving research funding from the National Institutes of Health and the National Kidney Foundation (NKF; which receives industry support); and serving as a scientific advisor or member of Healthy.io and NKF. M.E. Grams reports receiving honoraria from academic institutions for giving grand rounds and American Diabetes Association for reviewing abstracts; serving as a scientific advisor or member of American Journal of Kidney Diseases, CJASN, the JASN Editorial Fellowship Committee, the Kidney Disease Improving Global Outcomes Executive Committee, the NKF Scientific Advisory Board, and the United States Renal Data System Scientific Advisory Board; and other interests/relationships with NKF, which receives funding from Abbvie, Relypsa, and Thrasos. E.A. Hu reports employment with and ownership interest in Foodsmart. C.R. Parikh reports consultancy agreements with Genfit Biopharmaceutical Company and Novartis; is a member of the advisory board of and owns equity in RenalytixAI; reports receiving research funding from the National Heart, Lung and Blood Institute and the National Institute of Diabetes and Digestive and Kidney Diseases; and reports serving as a scientific advisor or member of Genfit Biopharmaceutical Company and Renalytix. C.M. Rebholz reports serving as an editorial board member for Diabetes Care. All remaining authors have nothing to disclose.

Funding

The ARIC study has been funded in whole or in part with federal funds from National Heart, Lung, and Blood Institute, National Institutes of Health, US Department of Health and Human Services contracts HHSN268201700001l, HHSN26821700002l, HHSN268201700003l, HHSN268201700004l, and HHSN268201700005l. Metabolomics measurements were sponsored by National Human Genome Research Institute grant 3U01HG004402-02S1. BHS was supported by National Institutes of Health, National Institute on Aging awards R01AG041200 and R21AG051914. Research reported in this publication was partially supported by National Institutes of Health, National Institute of General Medical Sciences award P20GM109036. This work was also supported by National Institutes of Health, National Heart, Lung, and Blood Institute grant F30HL147486. C.R. Parikh is supported by National Institutes of Health grants R01HL085757, UH3DK114866, U01DK106962, and R01DK093770. C.M. Rebholz is supported by National Heart, Lung, and Blood Institute grants R56 HL153178 and R21 HL143089 and National Institute of Diabetes and Digestive and Kidney Diseases grants K01 DK107782 and R03 DK128386.