Plasma Metabolomic Signatures of Healthy Dietary Patterns in the Chronic Renal Insufficiency Cohort (CRIC) Study

ABSTRACT

Background

In individuals with chronic kidney disease (CKD), healthy dietary patterns are inversely associated with CKD progression. Metabolomics, an approach that measures many small molecules in biofluids, can identify biomarkers of healthy dietary patterns.

Objectives

We aimed to identify known metabolites associated with greater adherence to 4 healthy dietary patterns in CKD patients.

Methods

We examined associations between 486 known plasma metabolites and Healthy Eating Index (HEI)-2015, Alternative Healthy Eating Index (AHEI)-2010, the Dietary Approaches to Stop Hypertension (DASH) diet, and alternate Mediterranean diet (aMED) in 1056 participants (aged 21–74 y at baseline) in the Chronic Renal Insufficiency Cohort (CRIC) Study. Usual dietary intake was assessed using a semiquantitative FFQ. We conducted multivariable linear regression models to study associations between healthy dietary patterns and individual plasma metabolites, adjusting for sociodemographic characteristics, health behaviors, and clinical factors. We used principal component analysis to identify groups of metabolites associated with individual food components within healthy dietary patterns.

Results

After Bonferroni correction, we identified 266 statistically significant diet-metabolite associations (HEI: n = 60; AHEI: n = 78; DASH: n = 77; aMED: n = 51); 78 metabolites were associated with >1 dietary pattern. Lipids with a longer acyl chain length and double bonds (unsaturated) were positively associated with all 4 dietary patterns. A metabolite pattern low in saturated diacylglycerols and triacylglycerols, and a pattern high in unsaturated triacylglycerols was positively associated with intake of healthy food components. Plasmalogens were negatively associated with the consumption of nuts and legumes and healthy fat, and positively associated with the intake of red and processed meat.

Conclusions

We identified many metabolites associated with healthy dietary patterns, indicative of food consumption. If replicated, these metabolites may be considered biomarkers of healthy dietary patterns in patients with CKD.

Introduction

Diet is a modifiable risk factor for the progression of chronic kidney disease (CKD), cardiovascular disease, and mortality (1–3). The current nutrition guideline for individuals with, or at risk of, CKD recommends lowering their dietary intake of protein, reducing dietary acid load, and consuming a diet that is high in fruits and vegetables (4). In line with these recommendations, recent evidence suggests that a greater adherence to healthy dietary patterns [e.g. Healthy Eating Index (HEI)-2015, Alternative Healthy Eating Index (AHEI)-2010, the Dietary Approaches to Stop Hypertension (DASH) diet, and alternate Mediterranean (aMED) diet] was inversely associated with a lower risk of CKD progression and all-cause mortality in individuals with CKD (5).

A vast body of literature on the relation of diet with health outcomes has relied on self-reported dietary measures, including the FFQ that was developed for large epidemiologic studies (6). Still, assessment of diet is challenging as it is prone to measurement error and bias (7, 8). Biomarkers that are not influenced by recall bias, social desirability bias, and other systematic errors related to self-reported measures, have the benefit of being objective measures of nutritional status (9, 10) and can replace or supplement self-report of dietary intake.

Advances in metabolomics—a high-throughput approach to detect and quantify a wide range of small molecules to characterize complex biosystems—provide an opportunity to assess the proximal physiologic effects of diet for the discovery of novel biomarkers of dietary intake (11). Using the same data set in which we documented an inverse association between dietary patterns and clinical outcomes in individuals with CKD, i.e. the Chronic Renal Insufficiency Cohort (CRIC) Study, we aimed to identify metabolites associated with the same 4 healthy dietary patterns using plasma metabolomics. Our secondary objective was to identify plasma metabolites associated with food components within these healthy dietary patterns.

Methods

Study design and study population

We used data from the CRIC Study, an ongoing multicenter prospective cohort of 3939 individuals with CKD (aged 21–74 y at baseline). The CRIC Study recruited men and women with reduced kidney function [estimated glomerular filtration rate (eGFR) 20–70 mL ⋅ min⁻¹ ⋅ 1.73 m⁻² based on the Modification of Diet in Renal Disease (MDRD) Study equation] from 2003 to 2008 from 7 clinical sites across the USA (12, 13). Participants were not eligible for the study if they could not provide informed consent, or if they were institutionalized, pregnant, or had severe chronic conditions (12). Participants returned for annual in-person follow-up visits and telephone interviews between the annual visits. The study protocol was approved by Institutional Review Boards at all clinical sites. All participants provided informed consent.

Among the 3939 CRIC Study participants, a random sample of 1800 individuals underwent metabolomics assays who were the study population for this analysis. Then, we excluded participants who did not complete an FFQ (n = 427), those with implausible dietary intake (women: <500 or 3500 kcal/d; men: <700 or 4500 kcal/d) (n = 13), those with insufficient data to calculate dietary patterns (n = 202), and those with missing covariates (n = 102) (Supplemental Figure 1). Our final analytic sample was 1056.

Diet assessment

At baseline and year 2, dietary intake was assessed using the National Cancer Institute's 124-item Diet History Questionnaire (DHQ). This semiquantitative FFQ has been validated in a nationally representative sample of US adults (14). Participants reported the frequency and amount of foods and beverages consumed in the past year. Participants were able to choose from 10 frequency options ranging from “never” to “6 or more times per day,” and had 3 options for amount depending on the food item (e.g. <1 apple, 1 apple, >1 apple). Nutrient intakes were quantified using the Diet*Calc Software (Diet*Calc Analysis Program, Version 1.5.0, National Cancer Institute). To increase precision for dietary intake estimates, we averaged dietary intake at baseline and year 2 (n = 826). For 230 participants without dietary intake data at visit 2, we used their baseline data. Pearson's correlations between diet scores at baseline and visit 2 was moderate to high, ranging from 0.63 (aMED) to 0.76 (DASH).

Participants were not advised to follow a certain dietary pattern. We used participants’ responses on the DHQ to calculate adherence to 4 healthy dietary patterns (HEI-2015, AHEI-2010, DASH, and aMED). Details on the calculation of these scores in the CRIC Study have been described previously (5). Briefly, the HEI-2015 assesses adherence to the Dietary Guidelines for Americans 2015–2020, has 13 components, and ranges from 0 to 100 (15). The AHEI-2010 is similar to the HEI-2015 but has been modified to include foods and nutrients strongly associated with chronic diseases based on prior literature (16). The AHEI has 11 components and ranges from 0 to 110. The DASH score has been developed based on the results from the original DASH feeding trial, has 8 components, and ranges from 8 to 40 (17, 18). The aMED score has been designed to measure alignment to a Mediterranean-style dietary pattern in US populations (19, 20). The aMED score has 9 components, and ranges from 0 to 9.

Metabolomic profiling

Metabolomic profiling was conducted by the Broad Institute using fasting plasma samples (>8 h) collected at year 1. Plasma EDTA samples were frozen on the day of collection and were shipped frozen to the central laboratory within a month of collection. All samples were stored at −80°C until they were assayed. Metabolites were measured using Nexera X2 U-HPLC systems (Shimadzu Scientific Instruments) coupled with Q Exactive/Exactive Plus orbitrap mass spectrometers (Thermo Fisher Scientific). Details on metabolomic profiling have been previously published (21). Briefly, for positively charged polar analytes [hydrophilic interaction liquid chromatography (HILIC)-pos], 10 μL of plasma was precipitated using 9 volumes of 74.9:24.9:0.2 v/v/v acetonitrile/methanol/formic acid containing stable isotope-labeled internal standards (valine-d8, Isotec; phenylalanine-d8, Cambridge Isotope Laboratories). Then, samples were centrifuged (9000 × g; 10 min; 4°C) and the supernatants were injected onto a 150 × 2 mm Atlantis HILIC column (Waters). The column was eluted with 5% mobile phase A (10 mM ammonium formate and 0.1% formic acid in water) for 1 min then 40% mobile phase B (acetonitrile with 0.1% formic acid) over 10 min. For negatively charged polar analytes (HILIC-neg), 30 μL of plasma was precipitated using 4 volumes of methanol containing inosine-15N4, thymine-d4, and glycocholate-d4 internal standards (Cambridge Isotope Laboratories). Samples were centrifuged (9000 × g; 10 min; 4°C), and supernatants were injected onto a 150 × 2 mm Luna NH2 column (Phenomenex). The column was eluted with 10% mobile phase A (20 mM ammonium acetate and 20 mM ammonium hydroxide in water) and 90% mobile phase B (10 mM ammonium hydroxide in 75:25 v/v acetonitrile/methanol), then a linear gradient to 100% mobile phase A. For lipids (C-8 pos), 10 μL of plasma was precipitated with 190 μL of isopropanol containing 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphocholine as an internal standard (Avanti Polar Lipids). Samples were centrifuged (9000 × g; 10 min; 4°C), and supernatants were injected onto a 100 × 2.1 mm ACQUITY BEH C8 column (1.7 μm; Waters). Then, the column was eluted with 80% mobile phase A (95:5:0.1 v/v/v 10 mM ammonium acetate/methanol/acetic acid) for 1 min, then a linear gradient to 80% mobile phase B (99.9:0.1 v/v methanol/acetic acid) over 2 min, a linear gradient to 100% mobile phase B for 7 min, followed by 3 min at 100% mobile phase B. Raw data were processed using Trace Finder version 3.1 and 3.3 (Thermo Fisher Scientific) and Progenesis CoMet software (version 2.0, Nonlinear Dynamics). Metabolites were identified by matching molecular weights and retention times to reference metabolites included in each batch and the Broad Institute's internal database. Internal standard peak areas were monitored for quality control and to identify any outlier samples requiring reanalysis. In addition, a pooled plasma reference sample was analyzed at intervals of 20 study samples and was used to standardize data using a “nearest neighbor” approach; standardized values were calculated using the ratio of the value in each sample over the nearest pooled reference multiplied by the median value measured across the pooled references. A total of 258, 92, and 294 known metabolites were quantitated by the HILIC-pos, HILIC-neg, and C8-pos methods, respectively.

Metabolites were excluded if >80% were missing across samples. Then, for the remaining metabolites, missing values were imputed with the lowest detectable value for that metabolite. Metabolites were then rescaled to a median of 1 by dividing by the median, and log transformed with log base 2. Next, metabolites with variance on log scale <0.01 or missing variance were excluded. Outliers were capped at mean ± 5 SDs. Our analyses were restricted to known metabolites, which resulted in 255 metabolites from HILIC-pos, 86 metabolites from HILIC-neg, and 293 metabolites from C8-pos. In total, 634 metabolites were measured from these 3 complementary methods. Out of 634 metabolites, 148 metabolites were duplicates [64 metabolites with the same metabolite name across different methods and 84 metabolites measured as 2 separate ions (e.g. C44:1 triacylglycerol and C44:1 triacylglycerol + NH₄)]. In total, 486 metabolites were nonoverlapping.

We classified metabolites according to metabolic pathway or category by merging their unique Human Metabolome Database (HMDB) identification number with the metabolomics data from Metabolon and by searching the HMDB website (22). For lipids without a HMDB identification number, we classified metabolites based on their names (e.g. C16:0 cholesterol ester was classified as cholesterol esters). Lipids were identified by total acyl carbon number and total number of double bonds.

Covariates

At baseline, participants reported age, sex, race/ethnicity, education, and income. At baseline and every subsequent in-person visit, data on health behaviors (smoking status, drinking status), and medical history (history of cardiovascular disease) were collected. Anthropometric measurements (height, weight), blood pressure (3 measurements in a seated position), 24-h urine protein, and blood samples were obtained at in-person visits as well. Details on collection and analysis of biospecimens are reported elsewhere (23). Briefly, urine protein was analyzed using the Roche/Hitachi Modular P Chemistry Analyzer. Glucose concentration was measured using the Hitachi Vitros 950 AT. Serum creatinine was measured using the Hitachi Vitros 950 AT, and cystatin-C was measured using a Siemens BNII instrument. Plasma HDL cholesterol was measured using an enzymatic colorimetric method. Height and weight were used to calculate BMI (kg/m²). Participants with mean systolic blood pressure ≥140 mmHg, mean diastolic blood pressure ≥90 mmHg, or use of antihypertensive medications were considered to have hypertension. Participants with fasting plasma glucose concentration ≥126 mg/dL, nonfasting plasma glucose concentration ≥200 mg/dL, or using antidiabetes mellitus medications were considered to have diabetes mellitus. Age, sex, race, blood creatinine concentration, and plasma cystatin concentration were used to estimate GFR using a CRIC-specific equation (23). Physical activity was assessed using the Multi-Ethnic Study of Atherosclerosis (MESA) Typical Week Physical Activity Survey only at baseline, and was calculated as metabolic equivalent tasks per week (24). We used all covariates assessed at year 1 to align covariates with the time of metabolomic profiling, except for physical activity and sociodemographic factors (sex, race/ethnicity, education, income), which were assessed at baseline, not year 1.

Statistical analyses

We summarized baseline characteristics of the study sample according to tertiles of all healthy dietary patterns using means (SDs) for continuous variables and proportions for categorical variables.

We used multivariable linear regression models to evaluate the associations between 1-unit higher score in healthy dietary patterns and individual metabolites, adjusting for sociodemographic characteristics (age, race/ethnicity, sex, clinical center, education, income), health behaviors (physical activity, smoking status), clinical factors (BMI, eGFR, proteinuria, HDL cholesterol, diabetes status, history of cardiovascular disease, hypertension status), and total energy intake. In analyses of HEI and DASH dietary patterns, we additionally adjusted for alcohol drinking status because alcohol was not a component within these indices. To account for multiple statistical testing, we used Bonferroni-adjusted P values of 2.57 × 10⁻⁵ [0.05/(4 dietary patterns × 486 nonoverlapping metabolites)]. We plotted β-coefficients according to acyl chain length and double bonds for all lipids that passed this Bonferroni threshold, given the large number of statistically significant lipids associated with each dietary pattern.

Then, we used principal component analysis (PCA) of the statistically significant diet-related metabolites to examine the association between metabolite patterns and food components within each dietary pattern. We selected principal components using scree plots and eigenvalues >1. Then, we restricted our analyses to principal components with ≥3 metabolites with factor loadings ≥|0.5|. This criterion was chosen to improve interpretability and identify metabolites which contribute substantially to principal components. Based on these criteria, we retained the first 4 principal components for HEI (63% of variance explained) and aMED (67% of variance explained), the first 5 principal components and principal component 12 for AHEI (69% of variance explained), and the first 5 principal components for DASH (64% explained). In AHEI, principal component 6 through principal component 11 had fewer than 3 metabolites with loadings ≥|0.5| or were not interpretable, thus were not retained for further analyses. After selecting principal components, we calculated scores for each principal component for each participant by summing metabolite patterns. Higher scores represented greater adherence to principal components. We used these scores as dependent variables and food components as independent variables in multivariable linear regression models. In these multivariable linear regression models, we adjusted for the same set of covariates as our analyses on overall healthy dietary patterns. We used Bonferroni-adjusted P threshold of 2.58 × 10⁻⁴ [0.05/194 statistical tests (13 HEI components × 4 principal components; 11 AHEI components × 6 principal components; 8 DASH components × 5 principal components; 9 aMED components × 4 principal components)] for the analyses of food components.

As a sensitivity analysis, we excluded 22 individuals who developed cardiovascular disease between year 1 and year 2 and repeated our main analyses on dietary patterns and individual metabolites. All analyses were conducted using R software version 3.6.2 (R Foundation for Statistical Computing) and Stata software version 15.0 (StataCorp).

Results

Baseline characteristics according to tertiles of healthy dietary patterns

In our study sample, mean BMI was 32, 82% and 44% of the participants were overweight, and had diabetes, respectively. Participants in the highest tertiles of healthy dietary patterns were more likely to be women, college graduates, drinkers, to have an income ≥$50,000, to not be overweight or obese, and were less likely to be current smokers compared with those in the lowest tertiles (Table 1). Those in the highest tertiles of AHEI and DASH were more likely to be white. Total energy intake was lower for those in the highest tertile of HEI and DASH, but higher for AHEI and aMED. Participants in the highest tertiles of all healthy dietary patterns had slightly higher HDL cholesterol, slightly higher eGFR, slightly lower concentrations of 24-h urine protein, and were more likely to have a history of CVD compared with those in the lowest tertiles.

TABLE 1.

Baseline characteristics of Chronic Renal Insufficiency Cohort (CRIC) Study participants by healthy dietary patterns¹

	HEI-2015		AHEI-2010		DASH		aMED
	Tertile 1 (n = 352)	Tertile 3 (n = 352)	Tertile 1 (n = 352)	Tertile 3 (n = 352)	Tertile 1 (n = 394)	Tertile 3 (n = 293)	Tertile 1 (n = 387)	Tertile 3 (n = 296)
Median score (range)	55.3 (29.6–61.8)	78.3 (72.6–92.2)	33.6 (16.5–40.4)	58.7 (52.0–85.4)	20 (9–22)	30 (28–37)	2 (0–3)	7 (6–9)
Age, y	56.5 ± 11.8	60.6 ± 9.9	56.8 ± 11.5	59.7 ± 10.2	56.0 ± 11.5	61.0 ± 9.4	57.1 ± 11.6	60.2 ± 10.1
Women, %	37.2	58.0	42.0	53.1	35.0	61.4	46.3	50.3
Race/ethnicity, %
White	59.4	53.7	48.6	63.4	47.0	59.7	54.8	54.1
Black	33.5	37.2	42.0	28.7	43.9	31.8	36.7	38.2
Other	7.1	9.1	9.4	8.0	9.1	8.5	8.5	7.8
Education, %
<High school	15.1	11.1	14.5	11.4	13.2	12.3	15.5	10.1
High school graduate	18.2	12.8	21.3	10.8	19.8	13.0	18.9	9.5
Some college	29.3	29.8	32.4	28.4	32.7	8.3	30.5	33.8
College graduate	37.5	46.3	31.8	49.4	34.3	46.4	35.1	46.6
Income ≥$50,000, %	38.1	41.8	33.2	44.3	36.5	39.6	34.9	43.9
Physical activity, METs/wk	205 ± 124	203 ± 128	201 ± 130	218 ± 139	209 ± 129	208 ± 127	195 ± 128	217 ± 147
Current smoker, %	16.8	6.5	13.4	6.0	17.8	4.8	15.8	5.7
Drinker, %	19.0	26.4	14.5	30.7	20.3	22.9	18.6	27.7
Total energy intake, kcal/d	1875 ± 767	1678 ± 643	1624 ± 655	1944 ± 713	1796 ± 717	1699 ± 634	1587 ± 645	1963 ± 695
BMI, kg/m2	32.2 ± 8.7	31.5 ± 7.8	32.1 ± 8.3	31.5 ± 7.5	31.8 ± 8.0	31.8 ± 7.9	31.9 ± 8.2	31.6 ± 7.5
Not overweight or obese (<25), %	16.7	21.3	16.7	19.3	16.2	20.8	18.1	18.2
Overweight (25 to <30), %	31.3	25.3	31.5	28.7	30.9	25.3	30.8	29.1
Obese (≥30), %	51.9	53.4	51.7	51.9	52.8	53.9	51.2	52.7
HDL cholesterol, mg/dL	47.3 ± 15.6	50.2 ± 13.9	47.4 ± 15.8	51.2 ± 14.7	46.6 ± 15.4	51.3 ± 15.4	49.1 ± 16.8	50.3 ± 14.4
Urinary protein, g/24 h	1.1 ± 2.4	0.7 ± 1.6	0.9 ± 2.0	0.7 ± 1.6	1.1 ± 2.3	0.6 ± 1.5	1.0 ± 2.2	0.6 ± 1.3
eGFR, mL · min−1 · 1.73 m−2	43.6 ± 17.4	44.7 ± 17.7	43.0 ± 17.0	45.7 ± 17.1	43.8 ± 17.1	43.9 ± 16.5	43.0 ± 17.5	44.6 ± 16.9
History of CVD, %	30.4	34.7	31.7	31.8	30.5	34.5	32.0	33.1
Diabetes, %	43.5	44.3	38.4	46.0	37.6	49.8	44.2	43.2
Hypertension, %	87.5	84.4	88.4	84.5	90.1	83.6	87.6	88.5

¹Values are mean ± SD for continuous variables and percentage for categorical variables. Plasma HDL cholesterol was measured using an enzymatic colorimetric method. Urine protein was analyzed using the Roche/Hitachi Modular P Chemistry Analyzer. Glucose was measured using the Hitachi Vitros 950 AT. Serum creatinine was measured using the Hitachi Vitros 950 AT, and cystatin-C was measured using a Siemens BNII instrument. Age, sex, race, blood creatinine concentration, and plasma cystatin concentration were used to estimate GFR using a CRIC-specific equation (23). AHEI-2010, Alternative Healthy Eating Index; aMED, alternate Mediterranean diet; CVD, cardiovascular disease; DASH, Dietary Approaches to Stop Hypertension; eGFR, estimated glomerular filtration rate; HEI-2015, Healthy Eating Index; MET, metabolic equivalent of task.

Metabolites associated with healthy dietary patterns

A total of 266 statistically significant diet-metabolite associations were identified (HEI: n = 60; AHEI: n = 78; DASH: n = 77; aMED: n = 51) (Table 2; Supplemental Table 1). More than half of these significant metabolites were triacylglycerols (range: 46–55%); followed by a combination of phosphatidylcholines, phosphatidylethanolamines, plasmalogens (range: 10–13%); diacylglycerols (range: 5–8%); and xenobiotics (range: 4–12%) (Figure 1).

TABLE 2.

Full list of known metabolites significantly associated with healthy dietary patterns¹

		HEI (n = 60)		AHEI (n = 78)		DASH (n = 77)		aMED (n = 51)
Method2	Metabolite	β	P value	β	P value	β	P value	β	P value
HILIC-neg	2-hydroxy-3-methylbutyrate/hydroxyisovalerate	—	—	—	—	−0.0207	6.68 × 10−6	—	—
HILIC-neg	2-hydroxy-3-methylpentanoate/hydroxyisocaproate	—	—	−0.0073	1.16 × 10−5	−0.0198	1.34 × 10−6	—	—
HILIC-pos	3-methylxanthine	−0.0259	1.50 × 10−5	—	—	—	—	—	—
HILIC-neg	4-pyridoxate	—	—	0.0189	7.12 × 10−6	0.0567	4.99 × 10−8	—	—
HILIC-pos	4-pyridoxate	—	—	0.0185	4.39 × 10−6	0.0539	4.99 × 10−8	—	—
HILIC-neg	α-keto-β-methylvalerate/α-ketoisocaproate	—	—	−0.0066	4.58 × 10−6	—	—	—	—
C8-pos	C16:0 CE	—	—	0.0032	3.50 × 10−6	0.0079	2.94 × 10−6	—	—
C8-pos	C18:2 CE	—	—	0.0034	5.72 × 10−6	0.0079	2.16 × 10−5	—	—
C8-pos	C20:5 CE	—	—	0.0085	2.39 × 10−6	0.0201	6.10 × 10−6	—	—
C8-pos	C22:6 CE	0.0119	6.47 × 10−8	0.0132	1.65 × 10−10	0.0289	1.68 × 10−8	0.0604	3.67 × 10−6
C8-pos	C16:0 Ceramide (d18:1)	—	—	—	—	−0.0142	2.82 × 10−6	−0.0325	2.44 × 10−5
C8-pos	C22:0 Ceramide (d18:1)	—	—	—	—	−0.0163	5.14 × 10−6	—	—
C8-pos	C24:1 Ceramide (d18:1)	—	—	−0.0064	4.58 × 10−6	−0.0177	3.17 × 10−7	−0.0391	8.23 × 10−6
C8-pos	C32:0 DAG	—	—	−0.0117	4.99 × 10−6	−0.0305	1.07 × 10−6	—	—
C8-pos	C32:1 DAG	—	—	−0.0134	1.42 × 10−7	−0.0278	1.03 × 10−5	—	—
C8-pos	C34:0 DAG	−0.0136	6.04 × 10−7	−0.0142	4.58 × 10−8	−0.0373	4.84 × 10−9	−0.0795	1.14 × 10−6
C8-pos	C34:0 DAG + NH4	−0.0140	7.06 × 10−7	−0.0143	9.05 × 10−8	−0.0383	5.44 × 10−9	−0.0820	1.11 × 10−6
C8-pos	C34:1 DAG	—	—	−0.0087	1.34 × 10−6	—	—	—	—
HILIC-pos	C34:1 DAG or TAG fragment	−0.0137	8.66 × 10−6	−0.0154	1.25 × 10−7	−0.0342	2.08 × 10−6	−0.0938	2.95 × 10−7
C8-pos	C36:0 DAG	−0.0095	3.30 × 10−7	−0.0084	1.98 × 10−6	−0.0247	1.19 × 10−8	−0.0502	5.75 × 10−6
C8-pos	C36:1 DAG	−0.0113	9.30 × 10−6	−0.0138	8.16 × 10−9	−0.0327	3.10 × 10−8	−0.0720	1.74 × 10−6
HILIC-pos	C16:1 LPC	—	—	−0.0074	3.01 × 10−8	—	—	—	—
C8-pos	C16:1 LPC	—	—	−0.0083	2.56 × 10−8	—	—	—	—
HILIC-pos	C22:4 LPC	−0.0110	1.33 × 10−7	−0.0140	8.56 × 10−13	—	—	−0.0639	2.17 × 10−7
HILIC-pos	C22:5 LPC	—	—	−0.0067	1.46 × 10−5	−0.0257	1.29 × 10−7	—	—
HILIC-pos	C24:0 LPC	0.0095	7.26 × 10−10	0.0089	1.08 × 10−9	0.0213	3.49 × 10−9	0.0531	6.50 × 10−9
HILIC-pos	C20:1 LPE	—	—	−0.0086	5.93 × 10−10	—	—	−0.0427	1.15 × 10−6
C8-pos	C20:4 LPE	—	—	−0.0059	1.53 × 10−5	—	—	—	—
HILIC-pos	C20:4 LPE	—	—	−0.0055	2.17 × 10−5	—	—	—	—
C8-pos	C30:1 PC	—	—	−0.0141	2.43 × 10−5	—	—	—	—
C8-pos	C36:1 PC	—	—	−0.0066	4.16 × 10−7	—	—	−0.0384	2.69 × 10−6
C8-pos	C38:3 PC	—	—	−0.0075	4.36 × 10−7	—	—	—	—
C8-pos	C38:6 PC	0.0067	2.31 × 10−6	—	—	—	—	—	—
C8-pos	C40:9 PC	0.0050	4.94 × 10−7	—	—	—	—	—	—
C8-pos	C34:5 PC plasmalogen	—	—	—	—	−0.0264	1.59 × 10−6	—	—
C8-pos	C36:4 PC plasmalogen	—	—	–0.0046	2.12 × 10−6	−0.0146	1.37 × 10−9	—	—
C8-pos	C36:5 PC plasmalogen-B	—	—	—	—	−0.0117	1.03 × 10−5	—	—
C8-pos	C38:4 PC plasmalogen	−0.0071	7.63 × 10−9	−0.0069	2.55 × 10−9	−0.0183	1.98 × 10−10	−0.0426	5.06 × 10−9
HILIC-pos	C38:7 PC plasmalogen	0.0073	4.21 × 10−7	0.0065	2.23 × 10−6	—	—	—	—
C8-pos	C36:0 PE	—	—	−0.0063	2.05 × 10−5	—	—	−0.0475	3.35 × 10−7
HILIC-pos	C38:6 PE	0.0076	2.63 × 10−6	—	—	—	—	—	—
C8-pos	C36:2 PE plasmalogen	—	—	—	—	−0.0183	3.13 × 10−7	—	—
C8-pos	C36:5 PE plasmalogen	—	—	—	—	−0.0180	7.79 × 10−6	—	—
C8-pos	C38:3 PE plasmalogen	—	—	—	—	−0.0217	7.63 × 10−6	−0.0560	5.12 × 10−6
HILIC-pos	C38:5 PE plasmalogen	—	—	−0.0045	1.65 × 10−6	−0.0138	2.21 × 10−9	−0.0249	2.22 × 10−5
C8-pos	C38:5 PE plasmalogen	—	—	—	—	−0.0211	1.73 × 10−9	—	—
C8-pos	C38:6 PE plasmalogen	—	—	—	—	−0.0155	3.72 × 10−6	—	—
HILIC-pos	C38:7 PE plasmalogen	0.0054	1.73 × 10−6	0.0053	7.64 × 10−7	—	—	0.0305	5.37 × 10−6
C8-pos	C18:1 SM	—	—	—	—	—	—	−0.0320	2.10 × 10–5
C8-pos	C44:0 TAG	−0.0198	1.06 × 10−5	—	—	−0.0448	1.95 × 10−5	—	—
C8-pos	C44:1 TAG	−0.0196	1.49 × 10−5	−0.0187	1.24 × 10−5	−0.0449	2.03 × 10−5	—	—
C8-pos	C44:2 TAG	−0.0199	9.45 × 10−6	−0.0181	2.08 × 10−5	−0.0449	1.87 × 10−5	—	—
C8-pos	C45:1 TAG	−0.0182	5.30 × 10−6	—	—	—	—	−0.1014	2.03 × 10−5
C8-pos	C46:1 TAG	−0.0186	6.79 × 10−7	−0.0195	3.66 × 10−8	−0.0433	7.48 × 10−7	−0.1075	1.40 × 10−6
C8-pos	C46:2 TAG	−0.0166	2.73 × 10−6	−0.0175	1.79 × 10−7	−0.0378	4.84 × 10−6	−0.0895	2.13 × 10−5
C8-pos	C46:3 TAG	−0.0155	8.35 × 10−6	−0.0160	1.25 × 10−6	−0.0353	1.46 × 10−5	—	—
C8-pos	C47:0 TAG + NH4	−0.0163	1.76 × 10−5	—	—	−0.0385	1.43 × 10−5	—	—
C8-pos	C47:1 TAG	−0.0212	9.48 × 10−7	−0.0198	1.27 × 10−6	−0.0446	1.03 × 10−5	−0.1326	2.30 × 10−7
C8-pos	C47:2 TAG	−0.0139	1.12 × 10−6	−0.0139	2.49 × 10−7	−0.0307	4.38 × 10−6	−0.0848	5.48 × 10−7
C8-pos	C48:0 TAG	−0.0192	6.91 × 10−6	−0.0185	5.62 × 10−6	−0.0494	7.23 × 10−7	−0.1163	5.64 × 10−6
C8-pos	C48:1 TAG	−0.0148	2.14 × 10−5	−0.0162	9.11 × 10−7	−0.0367	6.21 × 10−6	−0.0875	2.39 × 10−5
C8-pos	C48:2 TAG	−0.0130	7.88 × 10−8	−0.0151	4.33 × 10−11	−0.0302	9.97 × 10−8	−0.0785	5.09 × 10−8
C8-pos	C48:3 TAG	−0.0105	2.12 × 10−5	−0.0129	2.71 × 10−8	−0.0246	1.95 × 10−5	−0.0975	8.40 × 10−8
C8-pos	C49:0 TAG + NH4	−0.0195	7.60 × 10−7	−0.0195	1.78 × 10−7	−0.0497	5.87 × 10−8	−0.1191	3.67 × 10−7
C8-pos	C49:1 TAG + NH4	−0.0154	1.58 × 10−6	−0.0172	1.30 × 10−8	−0.0377	4.68 × 10−7	−0.1010	1.06 × 10−7
C8-pos	C49:2 TAG	−0.0147	1.88 × 10−6	−0.0148	3.60 × 10−7	—	—	−0.0975	8.40 × 10−8
C8-pos	C49:3 TAG	−0.0094	1.18 × 10−6	−0.0116	2.37 × 10−10	−0.0211	3.24 × 10−6	−0.0653	1.24 × 10−8
C8-pos	C50:0 TAG	−0.0149	1.00 × 10−6	−0.0145	6.18 × 10−7	−0.0382	7.28 × 10−8	−0.0886	1.29 × 10−6
C8-pos	C50:1 TAG	−0.0085	2.65 × 10−6	−0.0096	1.79 × 10−8	−0.0227	6.77 × 10−8	—	—
C8-pos	C50:2 TAG	—	—	−0.0079	7.93 × 10−8	–0.0158	1.47 × 10−5	—	—
C8-pos	C50:3 TAG	—	—	−0.0066	1.13 × 10−9	–0.0117	1.22 × 10−5	–0.0289	2.07 × 10−5
C8-pos	C50:4 TAG	—	—	−0.0062	1.43 × 10−5	—	—	—	—
C8-pos	C51:0 TAG	—	—	−0.0122	2.74 × 10−6	−0.0297	3.34 × 10−6	−0.0719	1.00 × 10−5
C8-pos	C51:1 TAG	−0.0110	7.59 × 10−7	−0.0127	1.47 × 10–9	−0.0304	4.52 × 10−9	−0.0753	1.12 × 10−8
C8-pos	C52:0 TAG	−0.0181	1.10 × 10−7	−0.0180	2.82 × 10−8	−0.0471	3.17 × 10−9	−0.1067	1.65 × 10−7
C8-pos	C52:1 TAG	−0.0086	8.69 × 10−7	−0.0101	1.26 × 10−9	−0.0237	6.02 × 10−9	−0.0574	3.60 × 10−8
C8-pos	C52:2 TAG	—	—	−0.0047	2.58 × 10−6	—	—	—	—
C8-pos	C53:2 TAG	—	—	−0.0069	2.28 × 10−7	—	—	—	—
C8-pos	C54:1 TAG	−0.0148	1.90 × 10−7	−0.0148	3.28 × 10−8	−0.0379	9.48 × 10−9	−0.0868	2.58 × 10−7
C8-pos	C54:2 TAG	—	—	−0.0061	4.49 × 10−6	—	—	—	—
C8-pos	C55:2 TAG	—	—	−0.0106	2.87 × 10−7	−0.0254	5.70 × 10−7	−0.0597	3.87 × 10−6
C8-pos	C56:7 TAG	0.0068	2.37 × 10−7	0.0057	4.75 × 10−6	—	—	0.0385	7.15 × 10−7
C8-pos	C56:8 TAG	0.0095	6.10 × 10−10	0.0090	6.17 × 10−10	0.0166	4.08 × 10−6	0.0535	4.15 × 10−9
C8-pos	C56:9 TAG	0.0108	1.24 × 10−6	0.0089	2.14 × 10−5	—	—	0.0591	6.71 × 10−6
C8-pos	C58:9 TAG	0.0167	1.66 × 10−10	0.0158	1.68 × 10−10	0.0313	3.34 × 10−7	0.0868	2.28 × 10−8
C8-pos	C58:10 TAG	0.0136	1.35 × 10−8	0.0132	4.41 × 10−9	0.0243	1.46 × 10−5	0.0736	2.06 × 10−7
C8-pos	C58:11 TAG	0.0181	1.06 × 10−6	0.0171	9.14 × 10−7	—	—	0.0971	9.46 × 10−6
C8-pos	C60:12 TAG	0.0278	2.37 × 10−9	0.0281	1.58 × 10−10	0.0530	1.15 × 10−6	0.1367	7.59 × 10−7
HILIC-pos	Cinnamoylglycine	—	—	0.0227	1.57 × 10−5	0.0637	9.14 × 10−7	—	—
HILIC-pos	Citrulline	0.0052	1.12 × 10−5	—	—	—	—	0.1647	2.11 × 10−7
HILIC-neg	CMPF	0.0284	1.07 × 10−7	0.0317	3.12 × 10−10	0.0541	1.55 × 10−5	—	—
HILIC-neg	Glycerate	0.0094	7.95 × 10−9	—	—	0.0190	5.71 × 10−7	—	—
HILIC-neg	Hippurate	—	—	0.0168	4.54 × 10−6	0.0479	1.08 × 10−7	—	—
HILIC-pos	Hippurate	0.0132	1.23 × 10−5	0.0133	2.70 × 10−6	0.0368	1.57 × 10−7	—	—
HILIC-neg	Indole-3-propionate	0.0209	2.42 × 10−6	—	—	0.0511	7.77 × 10−7	—	—
HILIC-neg	Inositol	0.0072	8.09 × 10−7	—	—	—	—	—	—
HILIC-neg	Malondialdehyde (MDA)	0.0096	5.27 × 10−8	—	—	0.0225	4.41 × 10−8	—	—
HILIC-pos	N-acetylornithine	0.0148	5.43 × 10−9	0.0133	2.73 × 10−8	0.0339	9.93 × 10−9	0.0828	3.62 × 10−8
HILIC-pos	N1-methyl-2-pyridone-5-carboxamide	—	—	—	—	0.0300	1.18 × 10−6	—	—
HILIC-pos	N-methylproline	0.0291	5.07 × 10−8	—	—	—	—	—	—
HILIC-pos	N-methylmalonamic acid (NMMA)	0.0047	1.80 × 10−6	0.0057	4.59 × 10−10	0.0102	7.47 × 10−6	—	—
HILIC-neg	Oxalate	0.0138	2.80 × 10−6	—	—	0.0300	1.25 × 10−5	—	—
HILIC-neg	Pantothenate	—	—	—	—	0.0338	1.48 × 10−7	—	—
HILIC-pos	Pantothenate	0.0118	1.95 × 10−5	0.0121	3.10 × 10−6	0.0362	1.55 × 10−8	—	—
HILIC-pos	Proline-betaine	0.0384	6.94 × 10−13	—	—	0.0535	2.17 × 10−5	0.1527	1.69 × 10−6
HILIC-neg	Quinate	—	—	—	—	0.1364	2.27 × 10−7	—	—
HILIC-pos	Thiamine	—	—	—	—	0.0564	4.80 × 10−8	—	—
HILIC-neg	Uracil	0.0063	4.00 × 10−8	0.0054	6.34 × 10−7	0.0118	1.14 × 10−5	0.0334	8.38 × 10−7
HILIC-neg	Uridine	0.0065	2.14 × 10−7	0.0057	1.81 × 10−6	—	—	0.0340	5.24 × 10−6

¹β and P values were calculated from multivariable linear regression models which adjusted for sociodemographic characteristics (age, race/ethnicity, sex, clinical center, education, income), health behaviors (physical activity, smoking status), clinical factors (BMI, estimated glomerular filtration rate, proteinuria, HDL cholesterol, diabetes status, history of cardiovascular disease, hypertension status), and total energy intake. For HEI-2015 and DASH dietary patterns, models also adjusted for drinking status because alcohol consumption was not part of the scores. P values were considered to be statistically significant at a Bonferroni-adjusted level of 2.57 × 10⁻⁵ [0.05/(486 metabolites × 4 dietary patterns)] for dietary patterns. AHEI-2010, Alternative Healthy Eating Index; aMED, alternate Mediterranean diet; CE, cholesterol esters; CMPF, 3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid; DAG, diacylglycerols; DASH, Dietary Approaches to Stop Hypertension; HEI-2015, Healthy Eating Index; LPC, lysophosphatidylcholines; LPE, lysophosphatidylethanolamines; PC, phosphatidylcholines; PE, phosphatidylethanolamines; SM, sphingomyelins; TAG, triacylglycerols.

²HILIC-pos indicates hydrophilic interaction liquid chromatography (HILIC) analyses of positively charged polar analytes in the positive ionization mode. HILIC-neg indicates HILIC analyses of negatively charged polar analytes in the negative ionization mode. C8-pos indicates that lipids were analyzed using positive ion mode (C8-pos). Details on these methods can be found in a prior publication (21).

— indicates that the association was not statistically significant at the Bonferroni threshold for the specific metabolite and dietary pattern.

FIGURE 1.

Distribution of metabolites significantly associated with healthy dietary patterns by metabolite categories. Numbers in the graph are n (%). “Identified metabolites” indicates the distribution of known metabolites in the present study, i.e. those metabolites detected in the data set. Metabolite categories which were <1% were not included. The sum of percentages may not equal to 100% due to rounding error. PC/PE/PS includes phosphatidylcholines, phosphatidylcholines plasmalogen, phosphatidylethanolamines, phosphatidylethanolamines plasmalogen, and phosphatidylserine. Details on the methods used to profile metabolites can be found in a prior publication (21). AHEI, Alternative Healthy Eating Index; aMED, alternate Mediterranean diet; DAG, diacylglycerols; DASH, Dietary Approaches to Stop Hypertension; HEI, Healthy Eating Index; TAG, triacylglycerols.

Of 266 statistically significant diet-metabolite associations, we identified 108 nonoverlapping metabolites. Of these 108 unique metabolites, 72% of metabolites (n = 78) were associated with >1 dietary pattern (Supplemental Figure 2). Across all 4 dietary patterns, 30 metabolites were common, most of which were lipids [triacylglycerols (n = 20), diacylglycerols (n = 3), cholesterol esters (n = 1), lysophosphatidylcholines (n = 2), diacylglycerol or triacylglycerol fragment (n = 1), plasmalogen (n = 1)], an amino acid (N-acetylornithine), and a metabolite with an unidentified pathway (uracil). Other nonlipid metabolites (4-pyridoxate, glycerate, pantothenate, proline-betaine) and 3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid (CMPF) were associated with 2 or more healthy dietary patterns. The DASH diet had the highest number of unique metabolites (n = 12), followed by the AHEI diet (n = 11), HEI (n = 6), and aMED (n = 1).

All diacylglycerols and most triacylglycerols were negatively associated with healthy dietary patterns (Supplemental Figure 3A–D). Amino acids and a few cholesterol esters were positively associated with healthy dietary patterns.

Among the lipids that were significantly associated with dietary patterns, lipids with a longer acyl chain length were inversely associated with dietary patterns (Figure 2A; Supplemental Figure 4A). This trend was evident for triacylglycerols. However, longer triacylglycerols, i.e. triacylglycerols with carbon number ≥58, were positively associated with all dietary patterns. Similarly, triacylglycerols and plasmalogens with ≥7 double bonds, and phosphatidylcholines and phosphatidylethanolamines with ≥6 double bonds were positively associated with all dietary patterns (Figure 2B; Supplemental Figure 4B).

FIGURE 2.

β-Coefficients for all significant lipids according to acyl chain carbon number (A) and double bonds (B) for participants in the Chronic Renal Insufficiency Cohort (CRIC) Study. Dashed line is set at a β-coefficient of zero. Models were adjusted for sociodemographic characteristics (age, race/ethnicity, sex, clinical center, education, income), health behaviors (physical activity, smoking status), clinical factors (BMI, estimated glomerular filtration rate, proteinuria, HDL cholesterol, diabetes status, history of cardiovascular disease, hypertension status), and total energy intake. PC/PE includes phosphatidylcholines, phosphatidylcholines plasmalogen, phosphatidylethanolamines, and phosphatidylethanolamines plasmalogen. Details on the methods used to profile metabolites can be found in a prior publication (21). AHEI, Alternative Healthy Eating Index; aMED, alternate Mediterranean diet; DAG, diacylglycerols; DASH, Dietary Approaches to Stop Hypertension; HEI, Healthy Eating Index; TAG, triacylglycerols.

Metabolites associated with food components

Some principal components were similar across dietary patterns with respect to the metabolites represented and the magnitude of their factor loadings (Supplemental Table 2). For instance, the first principal component of all dietary patterns, which explained 37–41% of the variance, was low in diacylglycerols and triacylglycerols (i.e. low factor loadings for diacylglycerols and triacylglycerols with 0 to 3 double bonds). The second principal component of HEI and AHEI and the fourth principal components of the DASH diet, which explained 5–14% of variance, reflected triacylglycerols with ≥7 double bonds. Principal component 5 of AHEI, principal component 2 of DASH, and principal component 4 of aMED were low in plasmalogens.

There was a positive association between components with low factor loadings for saturated diacylglycerols and triacylglycerols (with 0 to 3 double bonds) and healthy food components [total vegetables, seafood, or plant protein, (MUFA + PUFA/SFA in HEI; PUFA in AHEI; total vegetable and nuts and legumes in DASH; vegetable intake in aMED)] (Table 3). Components with high factor loadings for triacylglycerols with ≥7 double bonds were positively associated with healthy fat intake [(MUFA + PUFA)/SFA within HEI]. Similarly, components with low factor loadings for triacylglycerols with ≥7 double bonds were negatively associated with healthy fat intake (ω-3 fatty acids in AHEI) and total vegetable intake in DASH. A metabolite pattern low in plasmalogens (i.e. principal component with low factor loadings for plasmalogens) was positively associated with consumption of nuts and legumes and negatively associated with red and processed meat intake within AHEI. Similarly, a metabolite pattern low in plasmalogens was positively associated with MUFA:SFA ratio in aMED and negatively associated with red and processed meat intake within the DASH diet.

TABLE 3.

Associations between principal components and food components within dietary patterns¹

Dietary pattern	Principal component	Description of metabolites represented in the principal component	Associated food components
HEI	1	Low in DAGs and TAGs with 0–3 double bonds	Total vegetables (0.11), seafood, or plant protein (0.08), (MUFA + PUFA)/SFA (0.07)
	2	High in TAGs with ≥7 double bonds	(MUFA + PUFA)/SFA (0.04)
	3	Low in amino acids and cofactors	Total vegetables (−0.09), greens and beans (−0.07), refined grains (−0.06),2 added sugars (−0.05)2
AHEI	1	Low in DAGs and TAGs with 0–3 double bonds	PUFA (0.06)
	2	Low in TAGs with ≥7 double bonds	ω-3 fatty acids (−0.08)
	3	High in LPEs and LPCs	PUFA (–0.05)
	4	Low in LPEs and LPCs	PUFA (0.05)
	5	Low in plasmalogens	Nuts and legumes (0.03), red and processed meats (0.04)2
DASH	1	Low in DAGs and TAGs with 0–3 double bonds	Total vegetables (0.11), nuts and legumes (0.09)
	2	Low in plasmalogens	Red and processed meats (0.19)2
	3	Low in cofactors and vitamins	Sugar-sweetened beverages (–0.10)2
	4	Low in TAGs with  ≥7 double bonds	Total vegetables (–0.09)
aMED	1	Low in DAGs and TAGs with 0–3 double bonds	Vegetable intake (0.25)
	4	Low in plasmalogen	MUFA:SFA ratio (0.30)

¹Only statistically significant associations between food components and principal components at the Bonferroni threshold of 2.58 × 10⁻⁴ [0.05/194 tests (13 HEI food components × 4 principal components; 11 AHEI food components × 6 principal components; 8 DASH food components × 5 principal components; 9 aMED food components × 4 principal components)] are presented. Values in parentheses indicate β-coefficients calculated from multivariable linear regression models which adjusted for sociodemographic characteristics (age, race/ethnicity, sex, clinical center, education, income), health behaviors (physical activity, smoking status), clinical factors (BMI, estimated glomerular filtration rate, proteinuria, HDL cholesterol, diabetes status, history of cardiovascular disease, hypertension status), and total energy intake. For HEI-2015 and DASH diets, drinking status was also adjusted because alcohol consumption was not part of the scores. AHEI, Alternative Healthy Eating Index; aMED, alternate Mediterranean diet; DAG, diacylglycerols; DASH, Dietary Approaches to Stop Hypertension; HEI, Healthy Eating Index; LPC, lysophosphatidylcholines; LPE, lysophosphatidylethanolamines; TAG, triacylglycerols.

²For these score components, individuals with lower intakes receive higher scores.

Sensitivity analyses

We repeated our main analyses on dietary patterns and individual metabolites after excluding 22 individuals with incident cardiovascular disease in year 2. Results were very similar to the main analyses, with similar magnitude and the same direction of association.

Discussion

In this sample of individuals with CKD, we identified 266 significant diet-metabolite associations (HEI: n = 60; AHEI: n = 78; DASH: n = 77; aMED: n = 51); 78 of which were associated with >1 dietary pattern. Nearly one-third of metabolites (out of 108 nonoverlapping metabolites) replicated across all dietary patterns, most of which were lipids. Lipids with longer acyl chains and more double bonds (unsaturated) were positively associated with dietary patterns. We found that saturated triacylglycerols and diacylglycerols were inversely associated with healthy food intake (vegetables, seafood, or plant protein, MUFA, PUFA, nuts, and legumes), and unsaturated triacylglycerols (≥7 double bonds) were positively associated with healthy fat and vegetable intake. Plasmalogens were negatively associated with nuts and legumes and MUFA and positively associated with red and processed meat intake.

Our finding that there was a sizable overlap of metabolites across healthy dietary patterns is in line with other studies. Only 1 study examined metabolomics of multiple dietary patterns using Broad Institute's metabolomics platform (25). In the Framingham Offspring Study, 27 out of 65 significant diet-related metabolites (42%) were associated with 3 healthy dietary patterns (AHEI, DASH, and Mediterranean Diet Score) (25). In studies using different metabolomic platforms (e.g. Metabolon), similar observations were noted (26–28), suggesting that metabolites that are associated with multiple dietary patterns may reflect similar food components of healthy diets. In our study, most of the metabolite-diet associations that replicated across dietary patterns were lipids (triacylglycerols, diacylglycerols, cholesterol esters, lysophosphatidylcholines, diacylglycerol or triacylglycerol fragment, and CMPF). The list of metabolites common to multiple dietary patterns in our study (n = 78) was more than the number of overlapping metabolites reported from the Framingham Offspring Study and different from previous studies using Metabolon's metabolomics platform. Such differences may be due to an emphasis on lipid metabolites by the Broad Institute's metabolomics platform, the identification and analysis of more metabolites in our study, and differences between study populations.

Prior research on biomarkers of dietary patterns have typically examined metabolomic signatures of a single healthy dietary pattern and have been conducted in both observational studies and feeding studies. These studies focused on diverse groups of individuals, including the general US and UK populations (26, 29), health care professionals (30, 31), smokers (27), postmenopausal women (28), and Mediterranean populations (32). However, to our knowledge, no previous study has examined metabolites associated with healthy dietary patterns in individuals with CKD.

Despite the differences between our study and previous research, we found that some metabolites reported as candidate biomarkers of healthy dietary patterns in previous studies replicated in our study. For instance, N-methylproline which was positively associated with HEI in our study has been reported as a candidate biomarker of HEI in cross-sectional studies in the USA (26), and as a biomarker of the DASH diet in the DASH feeding trial (33) and the DASH-Sodium feeding trial (34). Of note, N-methylproline has been documented as a biomarker of healthy dietary patterns in serum (26, 27, 33, 35), in urine (34), and now in plasma in our study. N-acetylornithine, 4-pyridoxate, glycerate, pantothenate, CMPF, and proline-betaine (also known as stachydrine), which were positively associated with 2 or more healthy dietary patterns in our study, have previously been reported as biomarkers of healthy diets in serum (26–28, 35) and, in some cases, in urine (34). Such replication across different study designs, study populations, sample matrices, and dietary patterns provides greater support for these metabolites as objective biomarkers of healthy dietary patterns.

We observed that triacylglycerols with a longer acyl chain length and lipids (specifically, diacylglycerols, triacylglycerols, and plasmalogens) with ≥7 double bonds were positively associated with all healthy dietary patterns. In contrast, triacylglycerols, diacylglycerols, and plasmalogens with fewer double bonds were inversely associated with healthy dietary patterns. These findings are consistent with previous studies of HEI, AHEI, DASH, and Mediterranean Diet Adherence Screener (MEDAS) in a general population (25), a Mediterranean population (30), and health professionals (30, 31). However, considering that our study population was comprised of individuals with CKD with a high mean BMI and a large proportion of which had diabetes, it is possible that our findings may in part reflect their health status, instead of food intake. For example, higher carbon and double bond content in triacylglycerols has been highlighted as a signature of insulin sensitivity (36), whereas lower carbon and double bond content in triacylglycerols were associated with a higher BMI, higher waist circumference, insulin resistance, and cardiovascular disease (37, 38). Further, a study of individuals with CKD found that triacylglycerols, diacylglycerols, and phosphatidylcholamines with a shorter acyl chain length and fewer double bonds were relatively more abundant in earlier CKD stages (stages 2–4 consistent with our study population) than CKD stage 5 (39). In addition, our multivariable regression analyses adjusted for BMI and diabetes status. Our findings taken into context with prior literature suggest that lipids (specifically, triacylglycerols, diacylglycerols, and plasmalogens) with a longer acyl chain length and more double bonds may be used as nutritional biomarkers in CKD populations as well as in the general population.

Our findings from the PCA on patterns or principal components of metabolites associated with food components within healthy dietary patterns add to the literature on diet biomarkers. A metabolite pattern low in saturated diacylglycerols and triacylglycerols was positively associated with intake of vegetables, seafood, and plant proteins, ratio of MUFA and PUFA to SFA, PUFA, and nuts and legumes across all dietary patterns. A related finding was that unsaturated triacylglycerols were positively associated with healthy fats such as ω-3 fatty acids and vegetables. In the Nurses’ Health Study and Health Professionals Follow-up Study, a metabolite pattern high in unsaturated triacylglycerols was positively associated with PUFA and ω-3 fatty acids and negatively associated with trans fat intake within the AHEI (31). Similar to our study, another metabolomics study conducted in the PREvención con DIeta MEDiterránea (PREDIMED) trial reported that concentrations of triacylglycerols with a higher number of double bonds increased more among participants who were assigned to a Mediterranean diet intervention supplemented with nuts than participants in the control group (32). In a feeding study, greater consumption of long-chain PUFAs (salmon) led to an increase in long-chain ω-3 concentrations after 6 wk (40), suggesting that our findings are biologically plausible. Importantly, our metabolite patterns of triacylglycerols and diacylglycerols were also associated with vegetable intake, highlighting that not only fats but other nutrients (e.g. flavonoids) may play important roles in influencing plasma triacylglycerol and diacylglycerol levels (41, 42).

Plasmalogens, except for those with ≥7 double bonds, were inversely associated with healthy dietary patterns. In the analyses of food components, a metabolite pattern low in plasmalogens was positively associated with intake of nuts and legumes and MUFA and negatively associated with red and processed meats. In the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial, plasmalogens were inversely associated with habitual intake of red meat and processed meat, separately (43). Plasmalogens are components of cellular membrane structure and play important roles in cellular function such as cell signaling (44). Plasmalogens have been considered a biomarker of oxidative stress in individuals with CKD (45). It has also been hypothesized that lower concentrations of plasmalogens in individuals with advanced CKD may be due to chronic inflammation, implicating the role of inflammation in circulating plasmalogens (46). One randomized crossover study in patients with coronary artery disease found that plasma concentrations of plasmalogens decreased after a 4-wk lacto-ovo-vegetarian diet intervention and concentrations of plasmalogens increased after a 4-wk meat diet intervention (42). Taken together, these results suggest that plasmalogens may represent a metabolic pathway through which healthy dietary patterns low in red and processed meats are associated with a lower risk of clinical outcomes in individuals with CKD.

Our study has limitations. Although we averaged dietary assessment conducted at 2 time points (baseline and year 2) in order to improve the precision of usual dietary intake, metabolomic profiling was conducted using plasma specimens collected at year 1. Therefore, it is uncertain if participants maintained the same dietary intake at the time when their biospecimens were collected as when dietary intake was assessed. However, the diet history questionnaire was administered twice within 2 y and participants were asked to report average intake over the past year; thus, it is reasonable to conclude that this dietary assessment captured usual dietary intake. Although the robust lipid coverage is a strength of the Broad Institute's metabolomics platform (47), lipids do not constitute the only metabolic pathway through which healthy dietary patterns are associated with a lower risk of CKD progression and all-cause mortality in CKD patients. Previous studies have shown that dietary acid load may be an important biological mechanism that explains the association between healthy dietary patterns and CKD progression and end-stage renal disease (5, 48). However, metabolomics may not capture all relevant aspects of diet, such as cation intake (Na⁺, K⁺, Ca⁺, Mg⁺) or dietary acid load. Further, the present metabolomics data set provides lipid data at the sum compositional level. Thus, for lipids comprised of >1 acyl chain (such as triacylglycerol and diacylglycerol species), it is important to note that multiple isomers that sum to the same total acyl carbon number and double bond content can be subsumed by the same lipid measurement. This makes it difficult to make a direct link between our results and specific fatty acids available in the food supply and dietary assessments. For this reason, we are also uncertain if lipids reported in our study are the same as previously reported lipids in a sample of the general population, without knowing more details. Additionally, the metabolomics methods utilized in this study provide relative metabolite abundances, but do not provide information on absolute concentrations. Next, we focused only on known metabolites in the present study. In future studies, identification of metabolites in greater detail (acyl chain lengths, location of double bonds, absolute concentrations), and investigation of unknown compounds may be worthwhile to identify novel markers of healthy dietary patterns. Lastly, there is a possibility of residual confounding due to incorrectly measured or unmeasured covariates. However, we adjusted for the most important covariates which were carefully collected by trained study staff.

Nonetheless, to our knowledge, this is the first study to use untargeted metabolomics to identify biomarkers of healthy dietary patterns in a racially diverse sample of individuals with CKD. We leveraged repeated dietary assessments to improve precision of dietary intake. We used predefined diet indices in order to facilitate comparisons with other studies on biomarkers of dietary patterns.

In summary, we identified a total of 266 significant diet-metabolite associations, including 78 metabolites that were associated with >1 dietary pattern and 30 metabolites that were common across all dietary patterns, a majority of which were lipids. Our results suggest that acyl chain length and double bonds of triacylglycerols, diacylglycerols, and plasmalogens provide important information indicative of food consumption in CKD populations. Unsaturated triacylglycerols were associated with healthy foods (vegetables, PUFAs) and plasmalogens were associated with unhealthy foods (red and processed meat). If replicated, these metabolites may be considered biomarkers of healthy dietary patterns in individuals with CKD.

Notes

Funding for the CRIC Study was obtained under a cooperative agreement from National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (U01DK060990, U01DK060984, U01DK061022, U01DK061021, U01DK061028, U01DK060980, U01DK060963, U01DK060902, and U24DK060990). In addition, this work was supported in part by: the Perelman School of Medicine at the University of Pennsylvania Clinical and Translational Science Award NIH/National Center for Advancing Translational Sciences UL1TR000003, Johns Hopkins University UL1 TR-000424, University of Maryland General Cilnical Research Center M01 RR-16500, Clinical and Translational Science Collaborative of Cleveland, UL1TR000439 from the NCATS component of the NIH and NIH roadmap for Medical Research, Michigan Institute for Clinical and Health Research (MICHR) UL1TR000433, University of Illinois at Chicago Clinical and Translational Science Award UL1RR029879, Tulane Center of Biomedical Research Excellence for Clinical and Translational Research in Cardiometabolic Diseases P20 GM109036, Kaiser Permanente NIH/National Center for Research Resources University of California, San Francisco-Clinical and Translational Science Institute UL1 RR-024131, Department of Internal Medicine, University of New Mexico School of Medicine Albuquerque, NM R01DK119199, and a grant from the Mid-Atlantic Nutrition Obesity Research Center (NORC) under NIH award number P30DK072488. This work was partially supported by the Chronic Kidney Disease Biomarker Consortium funded by NIDDK U01 DK106981 (PI: EPR) and U01 DK085689 (PI: JC). CMR is supported by grants from the NIDDK (K01 DK107782, R03 DK128386) and grants from the National Heart, Lung, and Blood Institute (NHLBI; R21 HL143089, R56 HL153178). ZZ is supported by the Ruth L. Kirschstein Predoctoral Individual National Research Service Award (F31 DK-122683) from the National Institute of Diabetes and Digestive and Kidney Diseases. The findings do not necessarily reflect the opinions of the National Institute of Diabetes and Digestive and Kidney Diseases, the NIH, the Department of Health and Human Services, or the government of the USA.

Author disclosures: The authors report no conflicts of interest.

Supplemental Figures 1–4 and Supplemental Tables 1 and 2 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/jn/.

Abbreviations used: AHEI, Alternative Healthy Eating Index; aMED, alternate Mediterranean diet; CKD, chronic kidney disease; CMPF, 3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid; CRIC, Chronic Renal Insufficiency Cohort; DASH, Dietary Approaches to Stop Hypertension; DHQ, Diet History Questionnaire; eGFR, estimated glomerular filtration rate; HEI, Healthy Eating Index; HILIC, hydrophilic interaction liquid chromatography; HMDB; Human Metabolome Database; MDRD; Modification of Diet in Renal Disease; PCA, principal component analysis.