Metabolomics of Dietary Acid Load and Incident Chronic Kidney Disease

Abstract

Objective:

Blood biomarkers of dietary intake are more objective than self-reported dietary intake. Metabolites associated with dietary acid load were previously identified in two chronic kidney disease (CKD) populations. We aimed to extend these findings to a general population, replicating their association with dietary acid load, and investigating whether the individual biomarkers were prospectively associated with incident CKD.

Methods:

Among 15,792 participants in the Atherosclerosis Risk in Communities (ARIC) cohort followed from 1987–1989 (baseline) to 2019, we evaluated 3,844 black and white men and women with dietary and metabolomic data in a cross-sectional and prospective analyses. We hypothesized that higher dietary acid load (using equations of potential renal acid load and net endogenous acid production) was associated with lower serum levels of 12 previously identified metabolites: indolepropionylglycine, indolepropionate, N-methylproline, N-δ-acetylornithine, threonate, oxalate, chiro-inositol, methyl glucopyranoside, stachydrine, catechol sulfate, hippurate, and tartronate. In addition, we hypothesized that lower serum levels of these 12 metabolites was associated with higher risk of incident CKD.

Results:

Eleven out of 12 metabolites were significantly inversely associated with dietary acid load, after adjusting for demographics, socioeconomic status, health behaviors, health status, and estimated glomerular filtration rate: indolepropionylglycine, indolepropionate, N-methylproline, threonate, oxalate, chiro-inositol, catechol sulfate, hippurate, methyl glucopyranoside (α + β), stachydrine, and tartronate. N-methylproline was inversely associated with incident CKD (HR: 0.95, 95% CI: 0.91, 0.99, p=0.01). The metabolomic biomarkers of dietary acid load significantly improved prediction of elevated dietary acid load estimated using dietary data, beyond covariates (difference in C statistics: 0.021-0.077, p≤1.08×10⁻³).

Conclusion:

Inverse associations between candidate biomarkers of dietary acid load were replicated in a general population. N-methylproline, representative of citrus fruit consumption, is a promising marker of dietary acid load and could represent an important pathway between dietary acid load and CKD.

INTRODUCTION

Acid-producing foods such as those from animal sources (e.g. cheese, meat, and eggs) result in higher dietary acid load.^1-3 In contrast, base-producing foods such as those from plant sources (e.g. fruits and vegetables) result in lower in dietary acid load.^4,5 In the Atherosclerosis Risk in Communities (ARIC) study, higher quartiles of dietary acid load were associated with higher risk for incident chronic kidney disease (CKD) independent of demographic characteristics, socioeconomic status, total energy intake, lifestyle factors, comorbid conditions, antihypertensive medication use, and baseline kidney function.⁶ The kidneys play an important role in the excretion of acid and other toxins to maintain acid-base homeostasis.⁷ Thus, the degree of acid load may impair the normal function of the kidney, acutely or chronically, and may promote decline of estimated glomerular filtration rate (eGFR).⁸

The measurement of dietary intake often relies upon the use of self-reported data which is subject to recall bias.⁹ To address this issue, candidate blood biomarkers of dietary acid load were discovered and replicated using untargeted metabolomics data in two studies of kidney disease patients: the African American Study of Kidney Disease and Hypertension (AASK) and the Modification of Diet in Renal Disease (MDRD).¹⁰ Of the hundreds of metabolites evaluated, 13 metabolites were inversely associated with dietary acid load in both study populations: S-methylmethionine, indolepropionylglycine, indolepropionate, N-methylproline, N-δ-acetylornithine, threonate, oxalate, chiro-inositol, methyl glucopyranoside, stachydrine, catechol sulfate, hippurate, and tartronate. Whether these metabolites relate to dietary acid load in a general population and their relationship with kidney outcomes is unknown.

In the present study, we aimed to test the hypothesis that higher dietary acid load is associated with lower serum levels of the previously identified metabolites in a population of generally healthy adults. In addition, we aimed to test the hypothesis that higher serum levels of these metabolites are associated with lower risk of incident CKD.

METHODS

Study Population and Design

The ARIC study is a multi-center, prospective study of 15,792 middle-aged (45-65 years) adults enrolled in 1987-1989 from four US communities and followed prospectively.¹¹ We conducted a cross-sectional analysis to investigate metabolomic markers of dietary acid load and a prospective analysis to determine whether dietary acid load-related metabolites were related to kidney disease risk.

Of the 15,792 participants, there were 4,006 participants with metabolomic data at visit 1 (1987-1989). A total of 3,844 participants were available for analysis after excluding participants for the following reasons: (1) baseline eGFR ≤60 mL/min/1.73 m² or prevalent end-stage kidney disease (ESKD), as identified by linkage to the U.S. Renal Data System (USRDS) registry (n=73); (2) missing dietary intake data (n=54); or (3) missing covariates, i.e., body mass index (BMI) (n=4), physical activity (n=14), education (n=7), diabetes (n=5), smoking status (n=3), and total energy intake (n=2) (Supplemental Figure 1). The Institutional Review Board at each of the participating study sites reviewed and approved the study protocol. Informed consent was obtained from each participant.

Assessment of Dietary Intake

Dietary intake of participants was assessed using a validated, semi-quantitative food frequency questionnaire (FFQ) and administered by trained interviewers.¹² Participants reported their consumption frequency of a food item of a given portion size on average during their previous year.¹³ Total energy intake and nutrient intake were calculated by multiplying self-reported frequency of consumption and portion size by the nutritional content of each food item using U.S. Department of Agricultural data sources.

Quantification of Dietary Acid Load

Dietary acid load was estimated with dietary data using two established equations: potential renal acid load (PRAL) = (0.49×protein + 0.037×phosphorus − 0.021×potassium − 0.026×magnesium − 0.013×calcium) and net endogenous acid production (NEAP) = (54.5×(protein/potassium) − 10.2).^7,14

Metabolomic Profiling

Metabolomic profiling was completed using fasting serum samples that had been stored at −80°C since their collection at baseline in 1987–1989 (visit 1). Untargeted metabolomics was conducted using a Waters ACQUITY ultra-performance liquid chromatography system and a ThermoFisher Scientific Q-Exactive high resolution mass spectrometer with a heated electrospray ionization source and Orbitrap mass analyzer-based metabolomic quantification protocol by Metabolon, Inc. (Durham, North Carolina).^15,16 The metabolites were identified by matching features to a library of reference standards on the basis of retention time, mass-to-charge ratio, and chromatographic data. Profiling was conducted in two samples of serum specimens where the first sample (n=1,834) was a random sample of ARIC study participants from the Jackson, Mississippi study center, and the second sample (n=2,010) consisted of participants with sequencing data.^17-20

The present study focused on the 13 metabolites that were previously found to be significantly associated with dietary acid load in both AASK and the MDRD study, with the exception of S-methylmethionine, which was not detected in the ARIC study.¹⁰ In sample 1, 7 out of the 12 metabolites were available for analysis: indolepropionate, N-methylproline, threonate, chiro-inositol, catechol sulfate, hippurate, and stachydrine. All 12 metabolites were available in sample 2.

We scaled the metabolites to a median value of 1 and log-transformed the values. The metabolites were analyzed as continuous variables with the exception of chiro-inositol which was analyzed as a categorical variable (<lower limit of detection, <median, ≥median) given the large proportion of missing data. For all other metabolites, missing values (levels below the lower limit of detection) were imputed to the lowest detectable value for each metabolite. Missing data for each metabolite in the two samples before imputation is presented in Supplemental Table 1.

Ascertainment of Incident Chronic Kidney Disease

Kidney function was assessed using blood levels of creatinine and eGFR was estimated using the creatinine-based Chronic Kidney Disease Epidemiology Collaboration (CKD-Epi) equation.^21,22 For the prospective analysis, we used a composite definition of incident CKD: (1) eGFR <60 mL/min/1.73 m² accompanied by ≥25% eGFR decline from baseline, (2) International Classification of Diseases (ICD), Ninth/Tenth Revision code for a kidney disease-related hospitalization or death, or (3) ESKD (dialysis or transplantation) identified by linkage to the USRDS registry between baseline (study visit 1, 1987-1989) and December 31, 2017.^23,24 This definition was designed to decrease selection bias by supplementing visit-based measures with surveillance efforts, thus allowing for complete outcome ascertainment during follow-up.

Measurement of Covariates

At baseline, demographic characteristics (age, sex, and race), socioeconomic status (education level), health history (diagnosed co-morbidities and medication usage), and health behaviors (physical activity and cigarette smoking) were ascertained by trained interviewers using a structured questionnaire.¹¹ During visit 1, weight was measured, and BMI was calculated as kilograms divided by height in meters squared. The average systolic blood pressure (SBP) of the 2^nd and 3^rd measurements was used in this analysis. Diabetes was defined as fasting glucose ≥126 mg/dL, non-fasting glucose ≥200 mg/dL, self-reported diagnosis of diabetes by a physician, or diabetes medication use within the previous two weeks.

Statistical Analysis

We reported socio-demographic characteristics, clinical factors, and dietary factors for the overall study population and according to the sample of metabolomic data using descriptive statistics, i.e., mean and standard deviation for continuous variables and frequency and proportion for categorical variables.

For the cross-sectional analysis, multivariable linear regression models were used to examine the association between metabolites and dietary acid load (PRAL and NEAP), after adjustment for age, sex, race-center, education level, physical activity, smoking status, BMI, blood pressure, diabetes, total energy intake, and eGFR. Given the different racial distribution at each of the ARIC centers, a combined variable for race and center was created. For sample 1, race and center were omitted from the model since the sample consisted of participants from a single race group and a single study center, i.e., blacks from Jackson, Mississippi. We adjusted for education as a proxy for socioeconomic status because there was considerable missing data for income. For the prospective analysis, we used Cox proportional hazards regression to examine the association between dietary acid load-related metabolites and incident CKD after adjustment for the same covariates used in the cross-sectional analysis. Results were presented separately for the two samples and, for metabolites detected in both samples, results were meta-analyzed using fixed effects across samples to produce a single summary measure for the cross-sectional analysis and prospective analysis. For meta-analyzed results that were statistically significant, we examined race differences by stratifying and testing for interaction.

We calculated C-statistics for the prediction of the highest quartile of dietary acid load versus lower three quartiles of dietary acid load in models including metabolites and covariates compared to models with covariates only. We tested for differences in C-statistics for the two models.

All analyses were performed using Stata 15.1 (StataCorp, College Station, Texas).²⁵

RESULTS

Baseline Characteristics

Overall, 60% were female, 62% were African-American, mean age was 54 years, and mean eGFR was 108 mL/(min×1.73 m²) (Table 1). A third of the participants (32%) had education at a college level or above, mean BMI was 29 kg/m², 28% were current smokers, 13% had diabetes, and mean SBP was 125 mmHg. Sample 1 was exclusively African-American from the Jackson, Mississippi study center whereas 27% of participants were African-American in sample 2.

Table 1.

Baseline Characteristics of ARIC Participants According to Sample and for the Overall Study Population at Visit 1 (1987-1989)

Characteristics	Sample 1 (N=1,834)	Sample 2 (N=2,010)	Overall (N=3,844)
Age, years	53 (6)	54 (6)	54 (6)
Female	1,180 (64%)	1,142 (57%)	2,322 (60%)
African American	1,834 (100%)	536 (27%)	2,370 (62%)
Center
Forsyth County, North Carolina	-	574 (29%)	574 (15%)
Jackson, Mississippi	1,834 (100%)	410 (20%)	2,244 (58%)
Minneapolis, Minnesota	-	506 (25%)	506 (13%)
Washington County, Maryland	-	520 (26%)	520 (14%)
Diabetes	289 (16%)	218 (11%)	507 (13%)
Body mass index, kg/m2	30 (6)	28 (5)	29 (6)
Systolic blood pressure, mmHg	128 (21)	121 (20)	125 (21)
Education
Some high school or less	749 (41%)	498 (24%)	1,238 (32%)
High school graduate	512 (28%)	802 (40%)	1,314 (34%)
Some college or more	573 (31%)	719 (36%)	1,292 (34%)
Current smoker	518 (28%)	548 (27%)	1,066 (28%)
Physical activity index	2.1 (0.7)	2.4 (0.8)	2.3 (0.8)
Total energy intake, kcal/day	1,574 (613)	1,651 (610)	1,614 (613)
eGFR*, mL/min per 1.73 m2	114 (16)	102 (15)	108 (17)

Note: Values are mean (standard deviation) for continuous variables and n (%) for categorical variables.

^*Estimated glomerular filtration rate (eGFR) based on 2009 Chronic Kidney Disease-Epidemiology (CKD-EPI) creatinine equation.

ARIC, Atherosclerosis Risk in Communities study.

Cross-Sectional Analysis between Metabolites and Dietary Acid Load

Eleven out of 12 serum metabolites (indolepropionylglycine, indolepropionate, N-methylproline, threonate, oxalate, chiro-inositol, catechol sulfate, hippurate, methyl glucopyranoside, stachydrine, and tartronate) were statistically significantly inversely associated with PRAL and NEAP, after adjusting for demographic characteristics, socioeconomic status, health status, health behaviors, total energy intake, and eGFR (Table 2). The magnitudes of the associations were generally similar for metabolites that were detected in the samples. Likewise, the magnitudes of the associations were similar for PRAL (Figure 1) and NEAP (Figure 2). For example, higher levels of N-methylproline were associated with lower levels of PRAL in sample 1 (β = −1.21, p=6.44×10⁻⁸), sample 2 (β = −1.37, p=2.33×10⁻⁵), and the meta-analysis (β = −1.26, p=6.87×10⁻¹²) and with NEAP in sample 1 (β = −1.74, p=2.75×10⁻⁹), sample 2 (β = −2.00, p=1.85×10⁻⁹), and the meta-analysis (β = −1.85, p=2.35×10⁻¹⁷).

Table 2.

Cross-Sectional Association of Metabolites with Dietary Acid Load in the ARIC Study

Super-pathway	Metabolite	Sample 1			Sample 2			Meta-analysis
		β*	SE	P-value	β*	SE	P-value	β*	SE	P-value
Potential Renal Acid Load
Amino acid	Indolepropionylglycine	-	-	-	−0.41	0.19	3.23×10−2	-	-	-
Amino acid	Indolepropionate	−1.96	0.42	3.66×10−6	−0.83	0.33	1.17×10−2	−1.26	0.26	1.16×10−6
Amino acid	N-methylproline	−1.21	0.22	6.44×10−8	−1.37	0.32	2.33×10−5	−1.26	0.18	6.87×10−12
Amino acid	N-δ-acetylornithine	-	-	-	−0.36	0.65	5.82×10−1	-	-	-
Cofactors and vitamins	Threonate	−2.69	0.40	2.04×10−11	−3.44	0.61	1.78×10−8	−2.92	0.33	2.28×10−18
Cofactors and vitamins	Oxalate (ethanedioate)	-	-	-	−2.21	0.71	1.81×10−3	-	-	-
Lipid	Chiro-inositol	−1.73	0.47	2.14×10−4	−1.26	0.37	7.67×10−4	−1.44	0.29	7.48×10−7
Xenobiotic	Catechol sulfate	−2.49	0.53	2.95×10−6	−2.52	0.43	5.48×10−9	−2.51	0.33	6.24×10−14
Xenobiotic	Hippurate	−1.17	0.39	2.52×10−3	−1.97	0.34	1.05×10−8	−1.61	0.26	2.90×10−10
Xenobiotic	Methyl glucopyranoside (α + β)	-	-	-	−0.82	0.28	3.27×10−3	-	-	-
Xenobiotic	Stachydrine	−1.52	0.22	1.12×10−11	−1.09	0.22	1.01×10−6	−1.30	0.16	1.01×10−16
Xenobiotic	Tartronate (hydroxymalonate)	-	-	-	−2.29	0.42	9.84×10−8	-	-	-
Net Endogenous Acid Production
Amino acid	Indolepropionylglycine	-	-	-	−0.63	0.20	1.24×10−3	-	-	-
Amino acid	Indolepropionate	−2.68	0.55	1.26×10−6	−0.83	0.33	1.17×10−2	−1.63	0.29	1.74×10−8
Amino acid	N-methylproline	−1.74	0.29	2.75×10−9	−2.00	0.33	1.85×10−9	−1.85	0.22	2.35×10−17
Amino acid	N-δ-acetylornithine	-	-	-	−1.06	0.67	1.14×10−1	-	-	-
Cofactors and vitamins	Threonate	−4.22	0.52	6.92×10−16	−4.65	0.62	1.37×10−13	−4.40	0.40	2.83×10−28
Cofactors and vitamins	Oxalate (ethanedioate)	-	-	-	−3.53	0.73	1.26×10−6	-	-	-
Lipid	Chiro-inositol	−2.46	0.61	5.44×10−5	−1.35	0.39	4.85×10−4	−1.67	0.33	3.10×10−7
Xenobiotic	Catechol sulfate	−3.88	0.69	2.34×10−8	−2.34	0.45	1.70×10−7	−2.79	0.37	9.41×10−14
Xenobiotic	Hippurate	−1.80	0.50	3.39×10−4	−1.77	0.35	6.95×10−7	−1.78	0.29	8.35×10−10
Xenobiotic	Methyl glucopyranoside (α + β)	-	-	-	−1.21	0.29	2.57×10−5	-	-	-
Xenobiotic	Stachydrine	−2.38	0.29	2.35×10−16	−1.55	0.23	1.63×10−11	−1.87	0.18	1.38×10−25
Xenobiotic	Tartronate (hydroxymalonate)	-	-	-	−3.23	0.44	2.39×10−13	-	-	-

^*Adjusted for age, sex, race, center, education level, physical activity, smoking status, body mass index, blood pressure, diabetes, total energy intake, and estimated glomerular filtration rate. For sample 1, race and center variables were omitted from the model since the sample consisted of participants from a single race group and a single study center, i.e., blacks from Jackson, Mississippi. ARIC, Atherosclerosis Risk in Communities study; SE, standard error.

Figure 1.

Scatterplot of β Coefficients and −Log₁₀(P-values) for the Association between Metabolites and Potential Renal Acid Load

Figure 2.

Scatterplot of β Coefficients and −Log₁₀(P-values) for the Association between Metabolites and Net Endogenous Acid Production

Prospective Analysis of Dietary Acid Load-Related Metabolites and Incident CKD

During a median follow-up of 21 years, there were 1,452 (37.8%) incident CKD cases. After meta-analyzing across samples, N-methylproline was significantly inversely associated with incident CKD (HR: 0.95, 95% CI: 0.91, 0.99, p=0.01) (Table 3). The direction of the association between N-methylproline was the same in both samples, but the association was statistically significant only in sample 1 (sample 1 HR: 0.93, 95% CI: 0.87, 0.98, p=9.87×10⁻³; sample 2 HR: 0.98, 95% CI: 0.91, 1.06, p=0.64). This association was slightly stronger among black participants (HR: 0.94, 95% CI: 0.89, 0.99, p=0.02) compared to white participants (HR: 0.98, 95% CI: 0.90, 1.08, p=0.72), but there was no statistical evidence of interaction (p=0.23).

Table 3.

Prospective Association between Metabolites and Incident Chronic Kidney Disease in the ARIC Study

Super- pathway	Metabolite	Sample 1		Sample 2		Meta-analysis
		HR (95% CI)*	P-value	HR (95% CI)*	P-value	HR (95% CI)*	P-value
Amino acid	Indolepropionylglycine	-	-	1.04 (0.99, 1.09)	8.42×10−2	-	-
Amino acid	Indolepropionate	0.96 (0.86, 1.08)	0.52	1.01 (0.93, 1.10)	0.77	0.99 (0.93, 1.06)	0.87
Amino acid	N-methylproline	0.93 (0.87, 0.98)	9.87×10−3	0.98 (0.91, 1.06)	0.64	0.95 (0.91, 0.99)	0.01
Amino acid	N-δ-acetylornithine	-	-	1.22 (1.05, 1.41)	0.01	-	-
Cofactors and vitamins	Threonate	0.99 (0.89, 1.10)	0.84	0.99 (0.85, 1.14)	0.88	0.99 (0.91, 1.07)	0.80
Cofactors and vitamins	Oxalate (ethanedioate)	-	-	1.01 (0.85, 1.20)	0.90	-	-
Lipid	Chiro-inositol	0.98 (0.87, 1.11)	0.77	1.05 (0.97, 1.15)	0.22	1.03 (0.96, 1.10)	0.45
Xenobiotic	Catechol sulfate	0.96 (0.83, 1.11)	0.61	1.14 (1.03, 1.27)	0.01	1.06 (0.97, 1.16)	0.17
Xenobiotic	Hippurate	0.91 (0.82, 1.00)	0.05	1.06 (0.98, 1.15)	0.13	0.99 (0.93, 1.05)	0.66
Xenobiotic	Methyl glucopyranoside (α + β)	-	-	0.98 (0.92, 1.05)	0.60	-	-
Xenobiotic	Stachydrine	1.00 (0.95, 1.07)	0.91	1.00 (0.95, 1.05)	0.97	1.00 (0.96, 1.04)	0.97
Xenobiotic	Tartronate (hydroxymalonate)	-	-	0.92 (0.84, 1.02)	0.11	-	-

^*Adjusted for age, sex, race, center, education level, physical activity, smoking status, body mass index, blood pressure, diabetes, total energy intake, and estimated glomerular filtration rate. For sample 1, race and center variables were omitted from the model since the sample consisted of participants from a single race group and a single study center, i.e., blacks from Jackson, Mississippi.

Bold font indicates statistically significant associations.

ARIC, Atherosclerosis Risk in Communities study; CI, confidence interval; HR, hazard ratio.

In sample 2, higher levels of N-δ-acetylornithine (HR: 1.22, 95% CI: 1.05, 1.41, p=0.01) and catechol sulfate (HR: 1.14, 95% CI: 1.03, 1.27, p=0.01) were associated with higher risk of incident CKD (Table 3). However, for catechol sulfate, the direction of the association was not consistent across samples and was not statistically significant in sample 1 (HR: 0.96, 95% CI: 0.83, 1.11, p=0.61) or in the meta-analysis (HR: 1.06, 95% CI: 0.97, 1.16, p=0.17).

Prediction of Elevated Dietary Acid Load with Metabolites

Dietary acid load-related metabolites significantly improved the prediction of elevated dietary acid load beyond demographic characteristics, socioeconomic status, health status, health behaviors, total energy intake, and eGFR in both samples (Table 4). Results were similar for the prediction of elevated dietary acid load estimated using PRAL (sample 1 PRAL difference in C statistics: 0.028, 95% CI: 0.011, 0.045, p=9.2×10⁻⁴; sample 2 difference in C statistics: 0.021, 95% CI: 0.009, 0.034, p=1.08×10⁻³) and NEAP (sample 1 NEAP difference in C statistics: 0.077, 95% CI: 0.050, 0.103, p=1.10×10⁻⁸; sample 2 NEAP difference in C statistics: 0.033, 95% CI: 0.015, 0.051, p=2.85⁻⁴).

Table 4.

Prediction of Elevated Dietary Acid Load (Highest Quartile vs. Lower 3 Quartiles) with the Inclusion of Metabolites in Addition to Covariates

Dietary Acid Load	Sample	Model 1*	Model 2†	Difference in C Statistics (95% CI)	P-value‡
		C Statistic (95% CI)	C Statistic (95% CI)
PRAL	1	0.684 (0.656, 0.711)	0.712 (0.685, 0.738)	0.028 (0.011, 0.045)	9.20×10−4
	2	0.695 (0.669, 0.721)	0.716 (0.690, 0.741)	0.021 (0.009, 0.034)	1.08×10−3
NEAP	1	0.605 (0.575, 0.634)	0.681 (0.654, 0.709)	0.077 (0.050, 0.103)	1.10×10−8
	2	0.646 (0.618, 0.674)	0.679 (0.652, 0.707)	0.033 (0.015, 0.051)	2.85×10−4

^*Model 1 included covariates only (age, sex, race, center, education level, physical activity, smoking status, body mass index, blood pressure, diabetes, total energy intake, and estimated glomerular filtration rate). For sample 1, race and center variables were omitted from the model since the sample consisted of participants from a single race group and a single study center, i.e., blacks from Jackson, Mississippi.

^†Model 2 included covariates in Model 1 and metabolites. For sample 1, the 7 metabolites were: indolepropionate, N-methylproline, threonate, chiro-inositol, catechol sulfate, hippurate, and stachydrine. For sample 2, the metabolites 12 were: indolepropionylglycine, indolepropionate, N-methylproline, N-δ-acetylornithine, threonate, oxalate, chiro-inositol, catechol sulfate, hippurate, methyl glucopyranoside, stachydrine, and tartronate.

^‡P-value for the difference in C statistics for Model 1 vs. Model 2.

CI, confidence interval; PRAL, potential renal acid load; NEAP, net endogenous acid production

DISCUSSION

In this large, multi-site study of individuals without kidney disease, higher dietary acid load was significantly associated with lower serum levels of 11 out of 12 metabolites. Higher levels of N-methylproline, a marker of citrus fruit, were significantly and inversely associated with incident CKD after adjustment of demographic characteristics, socioeconomic status, health behaviors, health status, total energy intake, and eGFR and after meta-analyzing across the two samples. Our study also demonstrated that the dietary acid load-related metabolites significantly improved the prediction of elevated levels of dietary acid load, either measured by PRAL or NEAP, relative to covariates alone.

Our results in a general population with preserved kidney function replicated findings from a previous metabolomics study of dietary acid load biomarkers conducted in two kidney disease populations.^10,26 The findings replicated in terms of statistical significance and direction, i.e., all metabolites were inversely associated with dietary acid load. Our study documented that these candidate blood biomarkers that were previously related to dietary acid load estimated using urine measures are also related to dietary acid load estimated using dietary assessment.

The 12 dietary acid load-related metabolites analyzed in our study represent a wide range of metabolic pathways and categories including amino acids (n=4), cofactors and vitamins (n=2), lipid (n=1), and xenobiotics (n=5). Among amino acid metabolites, indolepropionate is consumed from red meats and eggs,²⁷ indolepropionylglycine is ubiquitous in tryptophan-rich foods (e.g. poultry, nuts, seeds, dairy)²⁸, N-methylproline is consumed from fruits,²⁷ and N-δ-acetylornithine is consumed from legumes and plants.²⁹ Threonate is consumed from multivitamins,²⁷ and it is also a product of ascorbic acid metabolism^30,31 and from plant-based foods, such as green leafy vegetables.³² Oxalate is another metabolite found in many plant-based foods (e.g. leafy vegetables, okra),³³ in addition to nuts and chocolate. Chiro-inositol, a lipid, may be derived from citrus fruits, beans, nuts, and grains.²⁷ Xenobiotics include food components (methyl glucopyranoside, stachydrine), a bacterial/fungal compound (tartronate), and compounds involved in benzoate metabolism (catechol sulfate, hippurate). Catechol sulfate has been detected in coffee,³⁴ hippurate is consumed from fruits and whole grains,^35,36 and stachydrine is a known biomarker of citrus fruits and plants.²⁷ Methyl glucopyranoside is common in essential oils and citrus foods^27,37-39 and an additive found in cleaning supplies.⁴⁰ Tartronate is a degradation product from the metabolism of ascorbic acid.³⁰ This broad array of dietary acid load-related metabolites reflects the holistic nature of dietary acid load determined by intake of all types of food, particularly plant sources of food, as suggested by prior literature on food sources and the inverse association of metabolites with dietary acid load.

Our study extends previous research by examining the potential role of these candidate biomarkers on CKD risk. In particular, N-methylproline was inversely associated with dietary acid and inversely associated with risk of incident CKD. This well-recognized marker of citrus fruit may more generally represent plant-based food consumption. N-methylproline serves as an osmolyte scavenger of reactive oxygen species, inhibiting apoptotic cellular death, replenishing NADP+ supply, and preventing cellular damage in saline-rich or other stressful environments.^41,42 Proline accumulates under osmotic stress, drought, and high saline conditions, thereby protecting plants from damage.^43-46 In a previous study, N-methylproline was identified as a biomarker of the Dietary Approaches to Stop Hypertension (DASH) diet, which is low in dietary acid load due to higher consumption of fruits and vegetables.⁴⁷ It may be plausible that consuming fruits, rich in N-methylproline, would decrease the risk of incident CKD.

N-δ-acetylornithine is consumed from legumes and plant sources, and was inversely associated with dietary acid load in sample 2 (it was not detected in sample 1), although it was not statistically significant. N-δ-acetylornithine may increase amino acids from animal-based foods involved with nitrogen storage, the urea cycle, and defense-related functions in plants,²⁹ which may be influenced by the gut microbiome. We found that N-δ-acetylornithine was significantly and positively associated with incident CKD in sample 2. Previous work has demonstrated that NAT8 is a locus for chronic kidney disease and a mutation in NAT8 is significantly associated with higher N-acetylornithine and higher risk of incident CKD among African Americans ^48,49.

Catechol sulfate, which is found in caffeinated coffee, was associated with higher risk of incident CKD in sample 2 only. Although coffee consumption is part of a healthy diet and was associated with lower risk of incident CKD,⁵⁰ it may also increase exposure to certain chemicals, such as catechol sulfate.^32,34 The differences in the results between the two samples may be due to differences in coffee preparation (roasting, filtering, processing, etc.) which could vary by geography and racial/ethnic groups.

There are a few important study limitations. Self-reported dietary intake is subject to measurement error and recall bias. Nonetheless, using self-reported dietary intake data, we were able to replicate most of the metabolites previously associated with dietary acid load estimated using urine measures.¹⁰ As such, these findings provide support for future nutritional metabolomics research for the discovery of diet biomarkers using dietary data. The dietary data were collected in 1987-1989 and, as such, dietary acid load estimated in this dataset may not reflect contemporary estimates of dietary intake in the U.S. As an observational cohort study, we cannot rule out the possibility of residual confounding due to imprecise measurement (e.g. socioeconomic status) being partly responsible for the observed findings. A diet intervention study of dietary acid load would be ideal for the documentation of diet biomarkers. There were differences between the two samples with respect to race, center, and metabolomic platforms in terms of breadth of metabolites detected.⁵¹ It is yet unknown whether our findings could be generalized to other racial/ethnic groups. Our research was focused on biomarkers of dietary acid load. Further research is necessary to investigate biomarkers of specific sources of protein and plant-rich dietary patterns, and to examine whether these diet biomarkers provide insight on metabolic pathways linking these aspects of diet to kidney disease risk. Given the lack of information on etiology of CKD within the ARIC study, we were unable to ascertain whether associations vary based on this factor.

There are several noteworthy strengths. Given the large, multi-center, prospective nature of the ARIC study, these findings can be broadly generalizable to a large segment of the US population – middle-aged, African-American and Caucasian men and women without underlying CKD. Our study also demonstrated consistent associations in the general population for dietary acid load estimated using data derived from a FFQ. In addition, the blood metabolomic profiling allowed for the measurement of a wide range of metabolites, representing many pathways, thereby providing a comprehensive, unbiased analysis of participants’ dietary intake. By leveraging the metabolomics data and long-term follow-up for incident CKD, our study was able to provide insights about food consumption, metabolism, and disease development. In addition, estimating dietary acid load using two established equations (i.e., NEAP and PRAL)⁷ with consistent results in terms of direction, significance, and magnitude, enhances the validity of the findings.

In summary, 11 biomarkers of dietary acid load were replicated: indolepropionylglycine, indolepropionate, N-methylproline, threonate, oxalate, chiro-inositol, catechol sulfate, hippurate, methyl glucopyranoside (α + β), stachydrine, and tartronate. Our study extends prior work in kidney disease populations to a general population and also extends the prior work using urine-based estimates of dietary acid load to diet-based (FFQ) estimates. These findings were consistent in the two samples within the ARIC study and were consistent across two estimates of dietary acid load (NEAP, PRAL). These metabolites are robust candidate diet biomarkers that may be useful in future studies to objectively assess the level of dietary acid load. N-methylproline may also represent an important metabolic pathway linking dietary acid load to CKD. Citrus fruit, represented by N-methylproline, and plant-derived foods in general, may be particularly useful in lowering dietary acid load and preventing kidney disease.

PRACTICAL APPLICATION.

This study replicates 11 biomarkers of dietary acid load which may be used to objectively assess the level of dietary acid load. Citrus fruit, represented by N-methylproline, may be useful in lowering dietary acid load and preventing kidney disease.