Identification of Protein Biomarkers of the Dietary Approaches to Stop Hypertension Diet in Randomized Feeding Studies and Validation in an Observational Study

Abstract

Background

The Dietary Approaches to Stop Hypertension (DASH) diet is recommended for cardiovascular disease prevention. We aimed to identify protein biomarkers of the DASH diet using data from 2 randomized feeding studies and validate them in an observational study, the ARIC (Atherosclerosis Risk in Communities) study.

Methods and Results

Large‐scale proteomic profiling was conducted in serum specimens (SomaLogic) collected at the end of 8‐week and 4‐week DASH diet interventions in multicenter, randomized controlled feeding studies of the DASH trial (N=215) and the DASH‐Sodium trial (N=396), respectively. Multivariable linear regression models were used to compare the relative abundance of 7241 proteins between the DASH and control diet interventions. Estimates from the 2 trials were meta‐analyzed using fixed‐effects models. We validated significant proteins in the ARIC study (N=10 490) using the DASH diet score. At a false discovery rate <0.05, there were 71 proteins that were different between the DASH diet and control diet in the DASH and DASH‐Sodium trials. Nineteen proteins were validated in the ARIC study. The 19 proteins collectively improved the prediction of the DASH diet intervention in the feeding studies (range of difference in C statistics, 0.267–0.313; P<0.001 for both tests) and the DASH diet score in the ARIC study (difference in C statistics, 0.017; P<0.001) beyond participant characteristics.

Conclusions

We identified 19 proteins robustly associated with the DASH diet in 3 studies, which may serve as biomarkers of the DASH diet. These results suggest potential pathways that are impacted by consumption of the DASH diet.

Registration

URL: https://www.clinicaltrials.gov; Unique identifiers: NCT03403166, NCT00000608.

Nonstandard Abbreviations and Acronyms

AFM

afamin

ARIC

Atherosclerosis Risk in Communities

ASAH2

neutral ceramidase

CCL25

C‐C motif chemokine 25

CD163

scavenger receptor cysteine‐rich type 1 protein M130

CTHRC1

collagen triple helix repeat‐containing protein 1

DASH

Dietary Approaches to Stop Hypertension

EGFR

epidermal growth factor receptor

FOLR2

folate receptor β

GGH

γ‐glutamyl hydrolase

INHBA

inhibin β A chain

INHBC

inhibin β C chain

KREMEN1

kremen protein 1

LYVE1

lymphatic vessel endothelial hyaluronic acid receptor 1

MET

hepatocyte growth factor receptor

PCOLCE

procollagen C‐endopeptidase enhancer 1

PDRG1

p53 and DNA damage‐regulated protein 1

PGLYRP2

N‐acetylmuramoyl‐L‐alanine amidase

RBP4

retinol‐binding protein 4

S100A14

protein S100‐A14

VAT1

synaptic vesicle membrane protein VAT‐1 homolog

Clinical Perspective.

What Is New?

• The Dietary Approaches to Stop Hypertension (DASH) diet is recommended by the Dietary Guidelines for Americans for the prevention of cardiovascular disease.

• We identified circulating protein biomarkers of the DASH diet in 2 randomized feeding studies and validated them in an observational study.

• Nineteen circulating proteins were robustly associated with the DASH diet, which reflected dietary intake, lower inflammation, and collagen deposition.

What Are the Clinical Implications?

• These proteins may serve as objective biomarkers of the DASH diet, addressing important gaps related to self‐reported diet in nutritional epidemiologic research.

• The 19 proteins associated with the DASH diet can help further our understanding on the mechanisms through which the DASH diet is associated with cardiometabolic diseases.

The Dietary Approaches to Stop Hypertension (DASH) diet is recommended by the Dietary Guidelines for Americans for the prevention of cardiovascular disease. ¹ The DASH diet is high in fruits, vegetables, and low‐fat dairy foods; moderate in fish, poultry, nuts, and seeds; and low in sugar‐sweetened beverages, sweets, and red and processed meats. ² In the original DASH trial, a randomized controlled feeding study, the DASH diet significantly lowered blood pressure in adults with pre‐ or stage 1 hypertension. ² Subsequently, in the DASH‐Sodium trial, the DASH diet combined with low dietary intake of sodium significantly lowered blood pressure compared with the DASH diet higher in sodium or the control diet. ³ Since then, many observational studies have identified associations of the DASH diet with a lower risk of hypertension, diabetes, kidney and cardiovascular diseases, and other diseases. ⁴ , ⁵ , ⁶ , ⁷

Assessment of adherence to the DASH diet within free‐living populations relies on self‐reported data. However, self‐reported data are prone to measurement error and systematic biases such as recall and socially desirable biases. ⁸ , ⁹ Objective biomarkers of the DASH diet that are not influenced by these factors can complement or replace existing approaches to dietary assessment. ¹⁰

Proteomics, a method that screens thousands of proteins in a single biospecimen, may be used to identify functional biomarkers of dietary intake. ¹¹ These proteins may offer insights on the underlying mechanism(s) through which diet is associated with cardiometabolic outcomes, given that proteins are integral to many biological processes. ¹¹ , ¹² , ¹³

Few prior studies identified protein biomarkers of dietary patterns. These studies used empirically derived dietary patterns or established diet indices such as the DASH diet score, Alternative Health Eating Index, and Mediterranean‐style diet. ¹⁴ , ¹⁵ These studies were highly promising in identifying novel biomarkers but were limited by a lack of validation and use of self‐reported diet. To minimize spurious associations, it is important to validate biomarkers in independent study populations and across study designs. ¹⁶

Therefore, the present study aimed to (1) identify protein biomarkers of the DASH diet using data from 2 randomized controlled feeding studies and (2) validate these proteins using data from a large observational cohort study.

Methods

Study Design and Population

Study data of the DASH trial, DASH‐Sodium trial, and the ARIC (Atherosclerosis Risk in Communities) study are available through the National Heart, Lung, and Blood Institute Biologic Specimen and Data Repository Information Coordinating Center. Interested researchers may also contact the ARIC study coordinating center to access data. The DASH and the DASH‐Sodium trials were multicenter, randomized controlled feeding studies that aimed to test the effect of the DASH diet on blood pressure. Details on the design of these studies have been published. ² , ³ , ¹⁷ In the DASH trial, after a 3‐week run‐in period, participants were randomly assigned to the DASH diet, fruit and vegetables diet, or the control diet, which lasted 8 weeks (Figure S1). ² Individuals who were included in the DASH trial were adults (≥22 years of age) with systolic blood pressure <160 mm Hg and diastolic blood pressure 80 to 95 mm Hg.

The DASH‐Sodium trial was designed to examine the combined effect of the DASH diet and sodium (low, medium, and high) on blood pressure. In the DASH‐Sodium trial, after a 2‐week run‐in period, participants were randomly assigned to either the DASH diet or control diet for 12 weeks (parallel arm design). ³ Within each diet, participants consumed low, medium, and high levels of sodium in random order for 4 weeks each (crossover design), and each sodium period was separated by up to 5 days. ¹⁸ Carryover from one sodium period to the next was expected to be minimal. ¹⁸ Eligibility criteria for the DASH and DASH‐Sodium trials were similar. Procedures were followed in accordance with the ethical standards of the institutional review boards at all study centers, and written informed consent was obtained from participants. The DASH and DASH‐Sodium trials were registered at http://clinicaltrials.gov/ as NCT03403166 and NCT00000608, respectively. The first author had full access to all of the data in the study and takes responsibility for its integrity and the data analysis.

For the present proteomics study, stored serum specimens were obtained from the National Heart, Lung, and Blood Institute Biologic Specimen and Data Repository Coordinating Center. ¹⁹ To characterize the proteome of the DASH diet, we compared serum specimens collected at the end of the (1) DASH diet and control diet interventions in the DASH trial and (2) DASH high‐sodium and control high‐sodium phases in the DASH‐Sodium trial. We selected the same high‐sodium phases in the DASH‐Sodium trial to provide a similar comparison of diet interventions as the DASH trial. For the present study, we used the same exclusion criteria for the DASH and DASH‐Sodium trials. Of the participants who participated in the DASH diet or control diet interventions (ie, excluding those randomized to the fruits and vegetables arm in the DASH trial; n_DASH=305; n_{DASH‐Sodium}=412), we excluded participants who did not consent to use of their stored biospecimens or had no available stored serum specimens (n_DASH=88; n_{DASH‐Sodium}=16) and participants with a flagged specimen by the laboratory (SomaLogic, Boulder, CO) suspected of poor sample quality (n_DASH=2; n_{DASH‐Sodium}=0) (Figure S2). After these exclusions, the analytic sample was 215 participants for the DASH trial and 396 participants for the DASH‐Sodium trial.

DASH Diet

The DASH diet emphasizes greater intake of fruits, vegetables, and low‐fat dairy foods; is low in red and processed meats, sugar‐sweetened beverages, and sweets; and includes different sources of proteins such as fish, poultry, nuts, and beans. ² , ¹⁷ The DASH and DASH‐Sodium trials used a 7‐day menu cycle (21 meals). The nutrient composition of the diet interventions in both trials have been published previously. ² , ³ , ¹⁷ , ¹⁸ Compared with the typical American diet (control diet), the DASH diet was designed to be higher in protein and fiber and lower in cholesterol and fat, especially saturated fat. Potassium, magnesium, and calcium intake was at the 75th percentile of consumption in the United States in the 1970s to 1980s. The control diet mirrored a typical American diet in that the macronutrient profile and fiber intake was close to average US consumption, whereas potassium, magnesium, and calcium intakes were at the 25th percentile of US consumption. The sodium content of the DASH diet and the control diet was similar (≈3000 mg) in the DASH trial. Sodium content was divided into 3 levels in the DASH‐Sodium trial. In the present proteomics study, we profiled samples from the high‐sodium phases (≈3600 mg/d of sodium) in the DASH‐Sodium trial to be as similar as possible to the sodium intake in the DASH trial and to be similar to average sodium intake in the United States. Considering the similarity in sodium content between the 2 feeding studies and between DASH and control interventions in both studies, it was expected that protein biomarkers identified would represent aspects of the DASH diet unrelated to sodium.

During the intervention periods, participants' body weight was measured each weekday, and total energy intake was adjusted to keep body weight stable. All meals were prepared in research kitchens using standardized menus. During weekdays, participants ate either lunch or dinner at study sites. Participants received cooked meals and snacks to be consumed on the weekends and offsite.

Protein Measurements in the DASH and DASH‐Sodium Trials

From the DASH and DASH‐Sodium trials, fasting (>8 hours) serum specimens, which were stored at −70 °C, were sent to SomaLogic. A slow off‐rate modified aptamer–based capture array named SOMAScan version 4.1 was used to identify >7000 protein targets and protein complexes. Details on the SOMAScan methods have been published previously. ²⁰ , ²¹ , ²² Briefly, the SOMAScan assay uses slow off‐rate modified aptamers, chemically modified, short single‐stranded DNA molecules as binding reagents to identify proteins and protein complexes. The SOMAScan platform has high sensitivity, similar to that of immunoassays. ²¹ , ²² Each serum specimen is mixed with thousands of slow off‐rate modified aptamers, then undergoes a capture, washing, and recapture process to identify cognate slow off‐rate modified aptamer and protein complexes with high specificity. Proteins that were captured during these steps can be quantified using standard DNA microarrays in relative fluorescence units. Then, relative abundance of all proteins was standardized and normalized to remove variations within an assay run and between runs. The SOMAScan assay had low intra‐ and interrun coefficient variation (5%) and high intraclass correlation coefficients (0.9). ²³ , ²⁴ Details on quality control in the DASH and DASH‐Sodium trials are found in Data S1.

The SOMAScan assay identified 7596 proteins in the DASH and DASH‐Sodium trials. In both trials, we excluded proteins that bound to Fc‐fusion mouse (n=233), bound to a contaminant (n=0), nonproteins that were hybridization control elution or nonhuman proteins (n=73), and proteins that were missing UniProt ID (n=0). Then, of the 7290 remaining proteins in both trials, we excluded proteins with coefficient of variation >50% (n_DASH=32, n_{DASH‐Sodium}=20) and proteins with low variance (<0.01 on the log₂ scale) (n_DASH=0, n_{DASH‐Sodium}=14). After these quality control steps, in the DASH and DASH‐Sodium trials, 7258 proteins and 7256 proteins remained, respectively. Of these, we focused on 7241 proteins that were in common between the DASH and DASH‐Sodium trials.

Validation: The ARIC Study

The ARIC study is a community‐based prospective study of predominantly Black and White men and women in the United States. ²⁵ In 1987 to 1989 (Visit 1), the ARIC study enrolled 15 792 adults between 45 and 64 years of age from 4 communities (Minneapolis, MN; Jackson, MS; Forsyth County, NC; and Washington County, MD).

Dietary assessment and protein measurement was conducted at Visit 3 (1993–1995). Trained interviewers collected data on usual intake of foods and beverages, using a 66‐item semiquantitative food frequency questionnaire. Participants reported the frequency with which they consumed foods and beverages of a certain serving size in the past year. Visual aids helped participants estimate the serving size. The responses on the food frequency questionnaire were used to calculate the DASH diet score. Detailed description and calculation of the DASH diet score in the ARIC study has been published. ⁶ , ²⁶ Briefly, the DASH diet score had 8 components. Higher intake of 5 components (fruits, vegetables, whole grains, nuts and legumes, and low‐fat dairy) received higher scores. Higher intake of 3 components (red and processed meats, sugar‐sweetened beverages, and sodium) received lower scores. For each component, participants were ranked based on their reported intake and could receive a score between 1 and 5 points. The theoretical range of the DASH diet score was 8 to 40, with a higher score representing greater adherence to the DASH diet.

In the ARIC study, large‐scale protein profiling using fasting plasma specimens was conducted by SomaLogic, the same biotechnology company that measured proteins in the DASH and DASH‐Sodium trials, and identified 5284 proteins and protein targets. Detailed methods and quality control of the proteomic data in the ARIC study have been published and are available in Data S1. ²⁷ , ²⁸

Of the 11 478 participants who attended Visit 3 and had complete proteomics data, we excluded participants based on the following criteria: (1) missing >10 items on the food frequency questionnaire (n=107), (2) those with implausible total energy intake (women <500 or >3500 kcal/d and men <600 or >4500 kcal/d, n=269), (3) those who were neither White nor Black (ie, Asian or American Indian), and Black individuals from 2 sites (Minneapolis, MN and Washington County, MD) because of the small sample size (n=67), (4) those with missing covariates (n=429), and (5) those with incomplete responses on dietary data that precluded calculating the DASH diet score (n=116). The analytic sample for the validation study was 10 490 participants.

Statistical Analysis

Baseline characteristics of the study population in the DASH and DASH‐Sodium trials were examined using means and SDs for continuous variables and proportions for categorical variables. Characteristics of the study population were also examined by diet interventions (DASH versus control diets) within each trial.

To examine the association between the DASH diet and 7241 serum proteins in the DASH and DASH‐Sodium trials, we conducted multivariable linear regression models using randomized diet interventions as the independent variable and proteins as the response variable. In each study, we adjusted for age, sex, race and ethnicity, and body mass index. Although participants were randomly assigned to diet interventions, we adjusted for baseline covariates, because the present proteomics study used data from a subset of participants and to improve precision. Then, estimates from multivariable linear regression models were meta‐analyzed using fixed‐effects models across the 2 trials, using the metafor package and inverse variance weights for calculation of standard errors. We selected a fixed‐effects model, given the similarities of the study design, conduct of the study, study setting, study populations, procedures, and dietary interventions between the DASH and DASH‐Sodium trials. Additionally, the Cochran Q test was not statistically significant for 6952 (96%) of the total of 7241 proteins (P>0.05). Proteins were considered statistically significant at a false discovery rate (FDR) <0.05. ²⁹

Then, we conducted validation analyses in the ARIC study. Only the proteins that were significantly different between the DASH and control diets from the DASH and DASH‐Sodium trials were considered for validation analyses. In the ARIC study, multivariable linear regression models were used to assess the association between 1 SD higher in the DASH diet score (independent variable) and log₂‐transformed proteins (response variable), adjusting for sociodemographic characteristics (age, sex, race, study center, education), total energy intake, health behaviors (alcohol intake, smoking, physical activity), and clinical factors (estimated glomerular filtration rate, diabetes, and body mass index). We adjusted for more covariates than the DASH and DASH‐Sodium trials, given that the ARIC study is an observational study. Of the 71 proteins that were associated with the DASH diet in the 2 trials, 50 proteins were available in the ARIC study. We considered proteins to be validated if proteins were statistically significantly associated with the DASH diet score at an FDR threshold <0.05 to account for multiple comparisons and had a consistent direction of association with the DASH and DASH‐Sodium trials.

Next, we calculated C statistics to assess whether the validated proteins individually and collectively predicted the DASH diet intervention in the DASH trial and the DASH‐Sodium trial beyond age, sex, race and ethnicity, and body mass index (covariates used in regression models). Similarly, in the ARIC study, we examined whether the validated proteins predicted those in the highest quartile compared with 3 lower quartiles of the DASH diet score beyond the participant characteristics (covariates in the regression models in the ARIC study). The highest quartile of the DASH diet indicates the highest adherence to the DASH diet. Greater adherence to the DASH diet was associated with a lower risk of important chronic diseases. ⁴ , ⁵ , ⁶ , ⁷ Lastly, we conducted sex‐stratified analyses of the validated proteins in the DASH and DASH‐Sodium trials and tested for interaction by sex. Estimates from sex‐stratified analyses were meta‐analyzed using a fixed‐effects model across the 2 trials. All analyses were conducted using R software version 4.1.0 (R Foundation for Statistical Computing).

Results

Participant Characteristics

Baseline characteristics were generally similar in the DASH and DASH‐Sodium trials (Table 1). In both trials, more than half of the participants were between the ages of 31 and 50 years and identified as belonging to a racial minority group. In both trials, approximately half of the participants were women. Those in the DASH trial were more likely to be in the middle‐income category ($30 000–$59 999). Other characteristics, such as education, smoking, and body mass index were similar across the DASH and the DASH‐Sodium trials. In the DASH‐Sodium trial, systolic blood pressure and diastolic blood pressure levels were slightly higher, which led to a greater proportion of participants classified as having hypertension (40.7% in the DASH‐Sodium trial compared with 26.0% in the DASH trial). These characteristics were similarly distributed when participants were stratified by the randomized diet interventions (Table S1).

Table 1.

Characteristics of the Participants in the DASH and DASH‐Sodium Trials^*

Characteristic	DASH, N=215	DASH‐Sodium, N=396
Age, y, n (%)
≤30	26 (12.1%)	12 (3.0%)
31–50	119 (55.3%)	233 (58.8%)
≥51	70 (32.6%)	151 (38.1%)
Women, n (%)	102 (47.4%)	225 (56.8%)
Minority race, n (%)†	123 (57.2%)	227 (57.3%)
Income, n (%)
<$29 999	71 (33.0%)	126 (31.8%)
$30 000–$59 999	95 (44.2%)	144 (36.4%)
≥$60 000	47 (21.9%)	117 (29.5%)
Not answered	2 (0.9%)	9 (2.3%)
Education, n (%)
High school graduate or less	33 (15.3%)	65 (16.4%)
College graduate	59 (27.4%)	94 (23.7%)
Postgraduate work/degree	79 (36.7%)	145 (36.6%)
Some college	44 (20.5%)	91 (23.0%)
Current smoker, n (%)	18 (8.4%)	42 (10.6%)
Weight, kg	82.7 (14.8)	84.0 (15.2)
Height, cm	170.9 (9.1)	169.7 (9.2)
BMI, kg/m2	28.2 (3.9)	29.2 (4.7)
SBP, mm Hg	130.8 (10.5)	134.8 (9.5)
DBP, mm Hg	84.6 (4.4)	85.7 (4.5)
Hypertensive status, n (%)‡	56 (26.0%)	161 (40.7%)
Ever used BP medication, n (%)	46 (21.4%)	70 (17.7%)

BMI indicates body mass index; BP, blood pressure; DASH, Dietary Approaches to Stop Hypertension; DBP, diastolic blood pressure; and SBP, systolic blood pressure.

^*Values are n (%) for categorical variables and mean (SD) for continuous variables.

^† The trials collected information on race (minority race vs nonminority race). Minority race refers to Black individuals.

^‡ Hypertensive status was defined as SBP ≥140 mm Hg or DBP ≥90 mm Hg.

Individual Proteins Differed Significantly Between the DASH and Control Diets in the DASH and DASH‐Sodium Trials

Out of 7241 proteins detected, relative abundance of 71 proteins was significantly different between the DASH and control diets in the DASH and DASH‐Sodium trials at FDR <0.05 after meta‐analysis (Table 2; Figure 1). Abundance of 41 proteins was higher, whereas abundance of 30 proteins was lower for participants consuming the DASH diet compared with the control diet. Proteins that were positively associated with the DASH diet included RBP4 (retinol‐binding protein 4), FOLR2 (folate receptor β), and EGFR (epidermal growth factor receptor). These proteins were not statistically different between diet interventions in the DASH trial, but were significantly different between diet interventions after meta‐analysis. Proteins that were negatively associated with the DASH diet included AFM (afamin), PGLYRP2 (N‐acetylmuramoyl‐L‐alanine amidase), GGH (γ‐glutamyl hydrolase), and INHBA (inhibin β A chain). Of these, GGH was the only protein that was not statistically different between the DASH and control diets in the DASH trial. Three proteins (CD209 antigen, AFM, and activin A) had duplicate protein targets (2 duplicates for CD209 antigen, and 1 duplicate for AFM and activin A). These duplicates were associated with the DASH diet in the same direction, but 1 duplicate protein target for CD209 antigen did not meet FDR <0.05 criterion after meta‐analysis.

Table 2.

Seventy‐One Proteins Significantly Different Between the DASH Diet Compared With the Control Diet in the DASH and DASH‐Sodium Trials^*

Uniprot ID	Entrezgene symbol	Protein name	DASH trial		DASH‐Sodium trial		Meta‐analysis
			β	P value	β	P‐value	β	P value	FDR‐adjusted P value
Q9NTN9	SEMA4G	Semaphorin‐4G	0.1317	2.07 E‐03	0.1399	5.93 E‐07	0.1374	2.57 E‐09	1.86 E‐05
Q9NNX6	CD209	CD209 antigen	−0.2122	4.66 E‐05	−0.1667	1.38 E‐04	−0.1858	1.84 E‐08	6.66 E‐05
P09110	ACAA1	3‐ketoacyl‐CoA thiolase, peroxisomal	0.3399	1.62 E‐03	0.2901	7.94 E‐06	0.3034	3.27 E‐08	6.69 E‐05
Q96IY4	CPB2	Carboxypeptidase B2	0.1030	7.84 E‐05	0.0703	1.16 E‐04	0.0812	3.70 E‐08	6.69 E‐05
O95445	APOM	Apolipoprotein M	−0.1489	3.26 E‐05	−0.0819	1.66 E‐03	−0.1055	3.96 E‐07	5.74 E‐04
P08476	INHBA	Activin A	−0.1708	1.98 E‐02	−0.1888	1.35 E‐05	−0.1842	6.03 E‐07	6.98 E‐04
Q96KN2	CNDP1	β‐Ala‐His dipeptidase	0.2006	1.56 E‐03	0.1857	1.68 E‐04	0.1913	6.75 E‐07	6.98 E‐04
Q9NNX6	CD209	CD209 antigen	−0.1624	4.94 E‐03	−0.1585	6.50 E‐05	−0.1598	7.94 E‐07	7.09 E‐04
Q15113	PCOLCE	Procollagen C‐endopeptidase enhancer 1	0.0846	2.33 E‐02	0.1258	1.16 E‐05	0.1106	8.81 E‐07	7.09 E‐04
O94819	KBTBD11	Kelch repeat and BTB domain‐containing protein 11	0.0818	1.43 E‐02	0.0850	3.39 E‐05	0.0842	1.13 E‐06	8.19 E‐04
P08476	INHBA	Inhibin β A chain	−0.1948	3.60 E‐04	−0.1155	6.48 E‐04	−0.1378	1.31 E‐06	8.61 E‐04
Q9BZR6	RTN4R	Reticulon‐4 receptor	0.0854	3.28 E‐02	0.1196	1.74 E‐05	0.1086	1.59 E‐06	9.57 E‐04
Q96PD5	PGLYRP2	N‐acetylmuramoyl‐L‐alanine amidase	−0.1030	9.89 E‐04	−0.0827	6.97 E‐04	−0.0905	2.02 E‐06	1.09 E‐03
Q86VB7	CD163	Scavenger receptor cysteine‐rich type 1 protein M130	0.1705	7.84 E‐03	0.1673	1.09 E‐04	0.1683	2.12 E‐06	1.09 E‐03
Q9UPZ6	THSD7A	Thrombospondin type‐1 domain‐containing protein 7A: thrombospondin type‐1 domain 6	0.2159	2.39 E‐03	0.1881	3.66 E‐04	0.1980	2.36 E‐06	1.14 E‐03
Q9Y624	F11R	Junctional adhesion molecule A	0.0306	6.12 E‐01	0.1165	2.79 E‐06	0.1043	4.29 E‐06	1.89 E‐03
Q9HCY8	S100A14	Protein S100‐A14	0.0638	5.04 E‐02	0.0891	3.34 E‐05	0.0815	4.43 E‐06	1.89 E‐03
P04003	C4BPA	C4b‐binding protein α chain	0.1143	2.39 E‐04	0.0688	4.23 E‐03	0.0860	4.88 E‐06	1.92 E‐03
Q9NUG6	PDRG1	p53 and DNA damage‐regulated protein 1	−0.1765	1.42 E‐05	−0.0739	2.58 E‐02	−0.1159	5.03 E‐06	1.92 E‐03
Q96CG8	CTHRC1	Collagen triple helix repeat‐containing protein 1	0.0772	2.06 E‐02	0.1066	1.05 E‐04	0.0947	6.53 E‐06	2.29 E‐03
Q96KN2	CNDP1	β‐Ala‐His dipeptidase	0.1677	8.21 E‐03	0.1671	3.23 E‐04	0.1673	6.65 E‐06	2.29 E‐03
P03951	F11	Coagulation factor XI	0.1109	5.82 E‐03	0.0991	4.81 E‐04	0.1031	7.35 E‐06	2.42 E‐03
P43652	AFM	Afamin	−0.0987	3.42 E‐03	−0.0770	7.67 E‐04	−0.0838	7.85 E‐06	2.47 E‐03
P35858	IGFALS	Insulin‐like growth factor‐binding protein complex acid labile subunit	0.0969	7.33 E‐04	0.0695	4.93 E‐03	0.0813	1.17 E‐05	3.53 E‐03
O15031	PLXNB2	Plexin‐B2	0.1179	3.16 E‐03	0.0786	1.90 E‐03	0.0899	2.23 E‐05	6.45 E‐03
Q99536	VAT1	Synaptic vesicle membrane protein VAT‐1 homolog	−0.4338	4.92 E‐04	−0.1611	2.18 E‐03	−0.2031	2.37 E‐05	6.59 E‐03
P00797	REN	Renin	0.2458	9.20 E‐03	0.2482	1.06 E‐03	0.2473	2.46 E‐05	6.60 E‐03
P02753	RBP4	Retinol‐binding protein 4	0.0704	1.46 E‐01	0.1181	7.36 E‐05	0.1052	2.91 E‐05	7.54 E‐03
P08476	INHBA	Activin A	−0.1413	2.37 E‐02	−0.1325	6.05 E‐04	−0.1349	3.48 E‐05	8.70 E‐03
O00161	SNAP23	Synaptosomal‐associated protein 23	−0.0444	6.74 E‐01	−0.1286	4.28 E‐05	−0.1219	4.33 E‐05	1.04 E‐02
P43652	AFM	Afamin	−0.0778	2.80 E‐02	−0.0801	7.77 E‐04	−0.0794	5.21 E‐05	1.22 E‐02
O75356	ENTPD5	Ectonucleoside triphosphate diphosphohydrolase 5	0.1311	9.12 E‐04	0.0682	1.13 E‐02	0.0884	6.23 E‐05	1.41 E‐02
P49221	TGM4	Protein‐glutamine gamma‐glutamyltransferase 4	−0.1439	1.01 E‐03	−0.1000	2.21 E‐02	−0.1222	6.71 E‐05	1.47 E‐02
Q9Y5C1	ANGPTL3	Angiopoietin‐related protein 3	−0.1086	8.54 E‐03	−0.1008	3.57 E‐03	−0.1041	7.74 E‐05	1.65 E‐02
Q96MU8	KREMEN1	Kremen protein 1	0.0624	3.67 E‐02	0.0714	9.42 E‐04	0.0683	8.39 E‐05	1.74 E‐02
P00533	EGFR	Epidermal growth factor receptor	0.0564	2.12 E‐02	0.0544	1.99 E‐03	0.0551	1.03 E‐04	2.08 E‐02
Q9UGP4	LIMD1	LIM domain‐containing protein 1	0.0692	2.11 E‐01	0.1089	2.65 E‐04	0.1000	1.24 E‐04	2.37 E‐02
Q8IVM0	CCDC50	Coiled‐coil domain‐containing protein 50	0.0706	3.00 E‐01	0.1909	8.71 E‐05	0.1507	1.25 E‐04	2.37 E‐02
P14207	FOLR2	Folate receptor β	0.0686	6.06 E‐02	0.0868	8.76 E‐04	0.0807	1.30 E‐04	2.42 E‐02
Q7L266	ASRGL1	Isoaspartyl peptidase/L‐asparaginase	−0.0547	4.29 E‐02	−0.0408	1.19 E‐03	−0.0432	1.34 E‐04	2.42 E‐02
Q5H8A3	NMS	Neuromedin‐S	0.0298	3.48 E‐01	0.1108	3.61 E‐05	0.0774	1.40 E‐04	2.48 E‐02
Q96CA5	BIRC7	Baculoviral inhibitor of apoptosis domain repeat‐containing protein 7 isoform β	−0.0208	2.54 E‐01	−0.0559	1.03 E‐04	−0.0425	1.48 E‐04	2.55 E‐02
Q9Y5Y7	LYVE1	Lymphatic vessel endothelial hyaluronic acid receptor 1	0.0949	4.35 E‐02	0.0972	1.50 E‐03	0.0965	1.52 E‐04	2.55 E‐02
Q86U17	SERPINA11	Serpin A11	−0.1544	1.51 E‐02	−0.0913	3.06 E‐03	−0.1033	1.76 E‐04	2.90 E‐02
P48304	REG1B	Lithostathine‐1‐β	0.0765	2.17 E‐01	0.1521	2.73 E‐04	0.1287	1.83 E‐04	2.90 E‐02
Q96AH0	NABP1	Sensor of single‐strand DNA complex subunit B2	0.0813	6.48 E‐02	0.0970	1.21 E‐03	0.0921	1.84 E‐04	2.90 E‐02
P11226	MBL2	Mannose‐binding protein C	−0.3164	6.48 E‐03	−0.2090	8.86 E‐03	−0.2437	1.93 E‐04	2.98 E‐02
P08581	MET	Hepatocyte growth factor receptor	0.0737	2.38 E‐02	0.0798	3.54 E‐03	0.0773	2.06 E‐04	3.11 E‐02
Q5UCC4	EMC10	UPF0510 protein INM02	−0.0909	1.23 E‐04	−0.0298	1.03 E‐01	−0.0531	2.14 E‐04	3.17 E‐02
P01042	KNG1	Kininostatin	−0.1196	3.62 E‐01	0.3739	2.09 E‐06	0.2456	2.35 E‐04	3.40 E‐02
Q13093	PLA2G7	Platelet‐activating factor acetylhydrolase	−0.1118	7.82 E‐02	−0.1169	1.48 E‐03	−0.1156	2.56 E‐04	3.63 E‐02
Q96GK7	FAHD2A	Fumarylacetoacetate hydrolase domain‐containing protein 2A	0.1745	1.23 E‐03	0.0927	5.10 E‐02	0.1288	2.73 E‐04	3.72 E‐02
O14793	MSTN	Growth/differentiation factor 8	−0.1161	9.58 E‐03	−0.0622	6.19 E‐03	−0.0733	2.74 E‐04	3.72 E‐02
P05451	REG1A	Lithostathine‐1‐α	0.1306	4.87 E‐02	0.1446	2.43 E‐03	0.1398	2.77 E‐04	3.72 E‐02
Q9NR71	ASAH2	Neutral ceramidase	0.1354	4.93 E‐02	0.1430	2.58 E‐03	0.1406	2.95 E‐04	3.88 E‐02
P22352	GPX3	Glutathione peroxidase 3	0.0789	7.37 E‐03	0.0521	1.33 E‐02	0.0612	3.19 E‐04	4.08 E‐02
Q16787	LAMA3	Laminin subunit α‐3	−0.1318	3.00 E‐03	−0.0613	1.59 E‐02	−0.0788	3.21 E‐04	4.08 E‐02
P55103	INHBC	Inhibin β C chain	−0.2466	1.51 E‐04	−0.0764	1.17 E‐01	−0.1388	3.33 E‐04	4.16 E‐02
Q96FC7	PHYHIPL	Phytanoyl‐CoA hydroxylase‐interacting protein‐like	0.1956	3.55 E‐04	0.0737	1.20 E‐01	0.1267	3.62 E‐04	4.44 E‐02
Q86Z14	KLB	β‐klotho	0.1431	5.27 E‐03	0.0758	1.52 E‐02	0.0942	3.82 E‐04	4.61 E‐02
O15444	CCL25	C‐C motif chemokine 25	−0.2705	1.24 E‐06	−0.0355	2.99 E‐01	−0.1025	3.92 E‐04	4.65 E‐02
Q9H4D0	CLSTN2	Calsyntenin‐2	0.0021	9.71 E‐01	−0.1071	1.04 E‐04	−0.0876	4.02 E‐04	4.70 E‐02
P13725	OSM	Oncostatin‐M	−0.1646	5.45 E‐03	−0.0680	9.63 E‐03	−0.0841	4.31 E‐04	4.90 E‐02
P62072	TIMM10	Mitochondrial import inner membrane translocase subunit Tim10	0.1451	1.27 E‐02	0.0864	9.22 E‐03	0.1009	4.33 E‐04	4.90 E‐02
Q15884	FAM189A2	Protein FAM189A2	0.1753	1.28 E‐02	0.1236	1.17 E‐02	0.1405	4.41 E‐04	4.90 E‐02
Q92820	GGH	γ‐glutamyl hydrolase	−0.0143	6.58 E‐01	−0.0953	6.71 E‐05	−0.0669	4.48 E‐04	4.90 E‐02
Q9Y279	VSIG4	V‐set and immunoglobulin domain‐containing protein 4	0.0898	7.43 E‐02	0.1153	2.52 E‐03	0.1060	4.53 E‐04	4.90 E‐02
P04114	APOB	Apolipoprotein B	−0.0069	9.75 E‐01	−0.2855	2.49 E‐04	−0.2550	4.65 E‐04	4.90 E‐02
O15444	CCL25	C‐C motif chemokine 25	−0.2144	1.02 E‐04	−0.0565	1.22 E‐01	−0.1056	4.67 E‐04	4.90 E‐02
P03952	KLKB1	Plasma kallikrein	0.0526	6.14 E‐02	0.0718	3.02 E‐03	0.0636	4.86 E‐04	4.96 E‐02
P04217	A1BG	α‐1B‐glycoprotein	−0.0484	1.37 E‐01	−0.0716	1.45 E‐03	−0.0641	4.87 E‐04	4.96 E‐02

DASH indicates Dietary Approaches to Stop Hypertension; FDR, false discovery rate; and ID, identification.

^* $\mathrm{\beta }$ and P values were calculated using multivariable linear regression models that were adjusted for age, sex, race and ethnicity, and body mass index in each study. Study‐specific $\mathrm{\beta }$ coefficients and P values were meta‐analyzed across the DASH and DASH‐Sodium trials using a fixed‐effects model. Only proteins that were at FDR <0.05 after meta‐analysis are presented. Positive $\mathrm{\beta }$ coefficients indicate that the abundance of protein was higher in participants consuming the DASH diet relative to the control diet. Negative $\mathrm{\beta }$ coefficients indicate that the abundance of protein was lower in participants consuming the DASH diet relative to the control diet.

Figure 1. Volcano plots of $\mathbf{\beta }$ coefficients and false discovery rate (FDR)–adjusted P values for proteins differing significantly between the Dietary Approaches to Stop Hypertension (DASH) diet compared with the control diet in the DASH and DASH‐Sodium trials.

The dashed horizontal line represents FDR <0.05, and the vertical dashed line is set at the $\mathrm{\beta }$ coefficient of 0. Proteins located to the right of the vertical line indicate that the abundance of proteins was higher in participants consuming the DASH compared with the control diet, and proteins located to the left of the vertical line indicate that the abundance of proteins was lower in participants consuming the DASH compared with the control diet. $\mathrm{\beta }$ and P values were calculated using multivariable linear regression models that were adjusted for age, sex, race and ethnicity, and body mass index in each study. Estimates were meta‐analyzed across the DASH and DASH‐Sodium trials using fixed‐effects models. Study‐specific $\mathrm{\beta }$ coefficients, P values, and the full name of all of these proteins are presented in Table 2.

Proteins That Were Validated in the ARIC Study

Out of 71 proteins, 50 proteins were available in the ARIC study. Of the 50 proteins analyzed, 19 proteins were validated in the ARIC study at a statistically significant threshold of FDR <0.05 and with the same direction of association as the DASH and DASH‐Sodium trials (Figure 2; Table S2). The magnitude of the association between proteins and the DASH diet score was smaller in the ARIC study compared with the DASH trials (Figure 2). Eleven proteins (CD163 [scavenger receptor cysteine‐rich type 1 protein M130], ASHA2 [neutral ceramidase], PCOLCE [procollagen C‐endopeptidase enhancer 1], RBP4 [retinol‐binding protein 4], LYVE1 [lymphatic vessel endothelial hyaluronic acid receptor 1], CTHRC1 [collagen triple helix repeat‐containing protein 1], S100A14 [protein S100‐A14], FOLR2, MET [hepatocyte growth factor receptor], KREMEN1 [kremen protein 1], and EGFR) were positively associated with the DASH diet score, and 8 proteins (GGH, AFM, PGLYRP2, CCL25 [C‐C motif chemokine 25], PDRG1 [p53 and DNA damage‐regulated protein 1], INHBA, INHBC [inhibin β C chain], and VAT1 [synaptic vesicle membrane protein VAT‐1 homolog]) were negatively associated with the DASH diet score.

Figures 2. $\mathbf{\beta }$ coefficients and 95% CIs of 19 proteins that were significantly different between the Dietary Approaches to Stop Hypertension (DASH) diet and the control diet in the trials (DASH trial [N=215] and DASH‐Sodium trial [N=396]) (left panel) and were validated using the DASH diet score in the ARIC (Atherosclerosis Risk in Communities) study (N=10 490) (right panel).

In the DASH and DASH‐Sodium trials, $\mathrm{\beta }$ coefficients and 95% CIs were calculated using multivariable linear regression models, adjusting for age, sex, race and ethnicity, and body mass index. Estimates were meta‐analyzed across the 2 trials using fixed‐effects models. In the ARIC study, $\mathrm{\beta }$ coefficients and 95% CIs were calculated using multivariable linear regression models that were adjusted for age, sex, race, study center, education level, alcohol drinking status, cigarette smoking status, physical activity score, body mass index, diabetes status, and estimated glomerular filtration rate. AFM indicates afamin; ASAH2, neutral ceramidase; CCL25, C‐C motif chemokine 25; CD163, scavenger receptor cysteine‐rich type 1 protein M130; CTHRC1, collagen triple helix repeat‐containing protein 1; EGFR, epidermal growth factor receptor; FOLR2, folate receptor β; GGH, γ‐glutamyl hydrolase; INHBA, inhibin β A chain; INHBC, inhibin β C chain; KREMEN1, kremen protein 1; LYVE1, lymphatic vessel endothelial hyaluronic acid receptor 1; MET, hepatocyte growth factor receptor; PCOLCE, procollagen C‐endopeptidase enhancer 1; PDRG1, p53 and DNA damage‐regulated protein 1; PGLYRP2, N‐acetylmuramoyl‐L‐alanine amidase; RBP4, retinol‐binding protein 4; S100A14, protein S100‐A14; and VAT1, synaptic vesicle membrane protein VAT‐1 homolog.

Prediction of the DASH Diet in the DASH and DASH‐Sodium Trials and the ARIC Study

PCOLCE, INHBA, VAT1, and AFM significantly improved the prediction of the DASH diet beyond the participant characteristics in both trials (range of difference in C statistics, 0.037–0.082; P<0.001 for all tests) (Table 3). Collectively, the 19 validated proteins significantly improved the prediction of the DASH diet in the DASH trial (difference in C statistics, 0.313; P<0.001), DASH‐Sodium trial (difference in C statistics, 0.267; P<0.001), and modestly in the ARIC study (C statistic for participant characteristics, 0.698; C statistic for 19 proteins and participant characteristics, 0.715; difference in C statistics, 0.017; P<0.001).

Table 3.

Prediction of the DASH Dietary Pattern Using 19 Proteins That Were Statistically Significant in the DASH and DASH‐Sodium Trials and Replicated in the ARIC Study^*

UniProt ID	Entrezgene symbol	Protein name	DASH trial† , ‡			DASH‐Sodium trial† , ‡
			C statistics for protein+participant characteristics	Difference in C statistics	P value	C statistics for protein+participant characteristics	Difference in C statistics	P value
Q15113||	PCOLCE||	Procollagen C‐endopeptidase enhancer 1||	0.608||	0.044||	<0.001||	0.653||	0.063||	<0.001||
Q9NUG6	PDRG1	p53 and DNA damage‐regulated protein 1	0.678	0.103	<0.001	0.604	0.028	0.12
P08476||	INHBA||	Inhibin β A chain||	0.651||	0.077||	<0.001||	0.621||	0.045||	<0.001||
Q96CG8	CTHRC1	Collagen triple helix repeat‐containing protein 1	0.603	0.043	0.08	0.645	0.054	<0.001
P14207	FOLR2	Folate receptor β	0.598	0.030	0.2	0.625	0.044	<0.001
P02753	RBP4	Retinol‐binding protein 4	0.580	0.027	0.18	0.635	0.053	0.06
P55103	INHBC	Inhibin β C chain	0.638	0.071	<0.001	0.588	0.020	0.24
Q96MU8	KREMEN1	Kremen protein 1	0.625	0.049	0.06	0.623	0.039	0.06
Q9HCY8	S100A14	Protein S100‐A14	0.618	0.041	0.1	0.652	0.069	<0.001
Q99536||	VAT1||	Synaptic vesicle membrane protein VAT‐1 homolog||	0.666||	0.082||	<0.001||	0.614||	0.037||	<0.001||
P00533	EGFR	Epidermal growth factor receptor	0.598	0.034	0.06	0.613	0.035	0.14
O15444	CCL25	C‐C motif chemokine 25	0.669	0.091	<0.001	0.583	0.009	0.46
P08581	MET	Hepatocyte growth factor receptor	0.611	0.040	0.08	0.624	0.043	0.1
Q9Y5Y7	LYVE1	Lymphatic vessel endothelial hyaluronic acid receptor 1	0.593	0.038	0.06	0.611	0.030	0.06
Q9NR71	ASAH2	Neutral ceramidase	0.594	0.035	0.12	0.619	0.039	<0.001
P43652||	AFM||	Afamin||	0.640||	0.069||	<0.001||	0.618||	0.041||	<0.001||
Q86VB7	CD163	Scavenger receptor cysteine‐rich type 1 protein M130	0.618	0.057	0.1	0.623	0.045	<0.001
Q96PD5	PGLYRP2	N‐acetylmuramoyl‐L‐alanine amidase	0.645	0.089	0.04	0.629	0.044	0.06
Q92820	GGH	γ‐glutamyl hydrolase	0.559	0.012	0.58	0.633	0.055	0.06
All 19 proteins§			0.867	0.313	<0.001	0.837	0.267	<0.001

ARIC indicates Atherosclerosis Risk in Communities; DASH, Dietary Approaches to Stop Hypertension; ID, identification; and VAT1, synaptic vesicle membrane protein VAT‐1 homolog.

^*C statistics for covariates only were 0.559 for the DASH trial and 0.577 for the DASH‐Sodium trial. We used logistic regression models to predict DASH diet intervention compared with the control diet intervention (response variable) using participant characteristics (age, sex, race and ethnicity, and body mass index) first, and investigated whether adding each of the proteins separately on top of the model with only the covariates improved the prediction of the DASH diet intervention.

^† Difference in C statistics was calculated using 1000 bootstrapped samples and represents the difference in C statistics for a model with only the participant characteristics and a model with proteins and participant characteristics.

^‡ P value comparing difference in C statistics for a model with only the participant characteristics and a model with proteins and participant characteristics.

^§ All 19 proteins were added to the model to assess whether they collectively improved the prediction of the DASH diet intervention.

^|| Indicates the 4 proteins that improved the prediction of the DASH diet in both the DASH and DASH‐Sodium trials.

The direction and magnitude of the association between the 19 validated proteins and the DASH diet were similar for women and men, and there was no statistical evidence of interaction by sex (P for interaction >0.05 for all tests; Table S3).

Discussion

In a large‐scale proteomic analysis of 2 randomized feeding studies of adults with pre‐ or stage 1 hypertension (systolic blood pressure <160 mm Hg and diastolic blood pressure 80–95 mm Hg), 71 proteins were significantly different between the DASH and control diets. In an observational study, 19 of the 50 available proteins were validated as being associated with adherence to the DASH diet score, which represented the usual intake of the DASH diet. Rather than a single biomarker, we identified a panel of proteins in the blood that better capture the multidimensional nature of the DASH diet, supported by the greater magnitude of increase in C statistic for the 19 proteins collectively versus individual proteins. These 19 proteins significantly improved the prediction of the DASH diet in the DASH and DASH‐Sodium trials, and modestly in the ARIC study. If replicated in other study populations, these results support the use of 19 proteins as objective biomarkers of the DASH diet.

Two of the 19 validated proteins (RBP4 and FOLR2), which were positively associated with the DASH diet, have well‐established relations with dietary intake. Retinoic acid (vitamin A), which can be derived from fruits and vegetables (carotenoids), directly binds to its cognate circulating protein (RBP4) and transthyretin, distributing vitamin A for cellular uptake. ³⁰ In a sample of 500 children in Nepal, plasma retinol concentration was strongly correlated with RBP4. ³⁰ FOLR2 binds to folate and reduces folic acid derivatives, and delivers 5‐methyltetrahydrofolate and folate analogs into the interior of cells. ³¹ Compared with the control diet, the DASH diet was higher in vitamin A and folate. Furthermore, serum folate increased significantly after the DASH diet intervention, whereas serum folate declined after the control diet intervention in the DASH trial. ³² Positive associations between the DASH diet, RBP4, and FOLR2 suggest that the DASH diet improved vitamin A and folate status.

Prior studies reported an association between AFM and clinical outcomes. AFM is a vitamin‐E binding protein expressed in the liver, and is known as a carrier of hydrophobic molecules. ³³ However, plasma levels of AFM had significant or null associations with vitamin E status. ³⁴ , ³⁵ Several studies reported that circulating levels of AFM were positively associated with adiposity, metabolic syndrome, greater hepatic lipid content, and prevalent and incident type 2 diabetes. ³³ , ³⁶ , ³⁷ , ³⁸ The DASH diet had favorable effects on blood lipids and was associated with a lower risk of metabolic syndrome and type 2 diabetes. ³⁹ , ⁴⁰ , ⁴¹ , ⁴² Consistent with these observations, we found an inverse association between the DASH diet and AFM.

Furthermore, EGFR has been associated with dietary intake and clinical outcomes. In 1662 participants of the Framingham Heart Study, greater adherence to self‐reported DASH diet, Alternative Healthy Eating Index, and Mediterranean‐style diet was positively associated with plasma EGFR levels, consistent with the findings from our study. ¹⁵ EGFR is a membrane‐bound protein involved in cell metabolism and cell adhesion. Plasma EGFR has been associated with more stable plaque phenotype given its role in stimulation of smooth muscle cells, and was inversely associated with incident atrial fibrillation and coronary events. ⁴³ , ⁴⁴

Four novel proteins (MET, ASAH2 [neutral ceramidase], PGLYRP2, CCL25] associated with the DASH diet in our study have a role in inflammation, suggesting an anti‐inflammatory effect of the DASH diet. MET regulates many physiological processes, including proliferation, scattering, morphogenesis, survival, and is essential for tissue repair. ⁴⁵ Interestingly, circulating MET levels have been proposed as biomarkers of atherosclerosis and stroke. ⁴⁶ , ⁴⁷ However, this protein is released in response to endothelial tissue injury and is considered both anti‐inflammatory and antifibrotic. ⁴⁸ In animal models, MET signaling regulated normal heart function by reducing oxidative stress. ⁴⁹ ASAH2, a secreted protein that hydrolyzes ceramides and dietary sphingolipids in the small intestine, was positively associated with the DASH diet in our study. ⁵⁰ One study of rat insulinoma cell line found that ASAH2 was released through exomes from insulinoma cells when concentration of cytokines was low, but ASAH2 release was inhibited when concentration of cytokines was high. ⁵¹ CCL25, a chemokine secreted by various cells, was negatively associated with the DASH diet. CCL25 is known to play a role in immune regulation and inflammatory processes, and circulating levels of chemokines were proposed as biomarkers of cardiovascular disease. ⁵² , ⁵³ Our findings for MET, ASAH2, and CCL25 supports the potential anti‐inflammatory effect of the DASH diet. PGLYRP2, also known as peptidoglycan recognition protein 2, is produced in the liver and is secreted into the blood. ⁵⁴ PGLYRP2 is thought to hydrolyze proinflammatory peptidoglycan. ⁵⁴ Plasma protein abundance of PGLYRP2 declined in mice after being fed high‐fat diet regimens (high in saturated fat, high in polyunsaturated fat) for 6 weeks. ⁵⁵ The results of our study are in contrast with this prior study in mice, given that blood PGLYRP2 concentrations were lower after the DASH diet intervention, which was lower in saturated fat than the control diet. However, the effect of PGLYRP2 on inflammation is inconclusive. ⁵⁶ The direction of association between the healthy dietary patterns and PGLYRP2 requires further investigation.

We identified 3 additional proteins that were positively associated with the DASH diet (CTHRC1, LYVE1, and PCOLCE), which supports that the DASH diet lowers arterial stiffness, ⁵⁷ potentially by influencing collagen deposition. CTHRC1 is secreted in response to endothelial injury and acts as a negative regulator of collagen matrix deposition. ⁵⁸ , ⁵⁹ LYVE1 is a transporter of hyaluronic acid. In mouse and human aorta, macrophages that expressed LYVE1 degraded collagen in vascular smooth muscle cells. ⁶⁰ PCOLCE is a secretory glycoprotein that enhances procollagen C‐proteinase activity. ⁶¹ Although PCOLCE has been suggested as a biomarker of fibrosis, PCOLCE is also involved in repairing injured tissue and plays an important role in regulation of collagen accumulation. ⁶¹ The DASH diet is high in fruits and vegetables (rich in polyphenols), soy (isoflavones), and low‐fat dairy, which has been associated with lower arterial stiffening in humans. ⁶²

Our finding of a positive association between CD163 and the DASH diet is inconsistent in its direction of association with previous research. CD163 is a transmembrane scavenger receptor protein that is expressed in macrophages and monocytes. ⁶³ During macrophage activation, CD163 sheds a soluble form. Circulating CD163 levels were associated with prevalent diabetes, prevalent obesity, and incident type 2 diabetes. ⁶⁴ , ⁶⁵ , ⁶⁶ In a case–control study of Spanish adults with and without type 2 diabetes (N=514), serum CD163 level was inversely associated with coffee and tea consumption. ⁶⁶ In Taiwanese adults (n=166, 84% with nonalcoholic fatty liver disease), greater intake of refined carbohydrates (noodles, desserts) and low intake of vegetables, dairy, seafood, and soy was positively associated with serum CD163. ⁶⁷ Differences in study results may be attributable to differences in diet exposures, study populations, and possible residual confounding in prior studies. To our knowledge, this is the first study that reported the effect of randomized diets on blood CD163. Future studies on the mechanism through which the DASH diet is associated with circulating CD163 is warranted.

The primary strengths of our study are related to the study design. We used stored specimens from rigorously conducted feeding studies. In the DASH and DASH‐Sodium trials, participants received all of their meals, ate 1 meal onsite on weekdays, and adherence to the interventions was high as documented with urinary excretion of potassium, phosphorus, and urea nitrogen. ² , ³ Participants were randomly assigned to diet interventions, which would have distributed measured and unmeasured confounders equally across the diet interventions. In addition, participants' weight was maintained stable, minimizing confounding because of weight loss. Therefore, serum proteins that were significantly different between the DASH diet and the control diet likely reflect the true biologic effect of diet, making these trials an ideal setting for discovery of the protein biomarkers of the DASH diet. Furthermore, validation of the proteomic biomarkers of the DASH diet score in a large independent study population of Black and White adults (the ARIC study) demonstrates the robustness of the findings. In the ARIC study, the magnitude of the difference based on C statistics was modest, which may be attributable to the use of DASH diet score versus dietary interventions in the feeding trials (DASH and DASH‐Sodium). Nonetheless, it is encouraging that the protein biomarkers discovered in randomized feeding studies validated in a free‐living population, who were not recommended to follow the DASH diet. Our results suggest the broader use of these biomarkers in the general population. Although the cost of untargeted proteomic profiling remains a barrier, in the future, targeted assays of a selected panel of proteins may be used to assess the degree of adherence to the DASH diet.

The study also has limitations. The DASH and DASH‐Sodium trials used a limited set of menus (7‐day cycle, repeated each week), which could have led to fewer identified proteins had the study been conducted in free‐living people eating a much greater variety of foods. The DASH trial, DASH‐Sodium trials, and the ARIC study were conducted in the 1990s, which may not reflect the current food supply. It would be important to validate these proteins in modern studies. Although all biospecimens were frozen and never thawed, biospecimens were in storage for >20 years. Degradation of proteins is possible, although it would be similar by diet interventions in the feeding studies and adherence to the DASH diet in the observational study. The number of proteins available in the ARIC study was smaller (≈5000 proteins) than the DASH and DASH‐Sodium trials (>7000 proteins). As a result, of the 71 proteins that were significantly different between the DASH diet and control diet interventions, 50 proteins were available in the ARIC study. We may have found a greater number of validated proteins if the number of available proteins was comparable across the studies. Furthermore, this proteomics assay provides only relative abundance, and there are no data on absolute concentration of the validated proteins. We used archived serum specimens from the DASH and DASH‐Sodium trials, and plasma specimens from the ARIC study. However, there is no evidence that differences in sample matrices would confound the findings.

In conclusion, our study addresses important gaps in nutritional epidemiologic research by providing candidate objective biomarkers of dietary intake to replace or complement self‐reported dietary intake. We identified 19 proteins robustly associated with the DASH diet in the DASH trial, DASH‐Sodium trial, and ARIC study, which may serve as biomarkers of the DASH diet. Our proteomics approach provides complementary information to serum and urine metabolomic markers of the DASH diet that were previously reported in feeding trials and observational studies. ⁸ , ⁶⁸ , ⁶⁹ , ⁷⁰ , ⁷¹ , ⁷² Our findings not only directly reflect dietary intake and represent dietary biomarkers, but also highlight potential mechanisms (eg, lowering inflammation and collagen deposition) that underlie the associations between the DASH diet and cardiometabolic disease.

Sources of Funding

This research was supported by the National Heart, Lung, and Blood Institute (R01 HL153178). The ARIC study was supported by the National Heart, Lung, and Blood Institute, National Institutes of Health, and U. S. Department of Health and Human Services (HHSN268201700001I, HHSN268201700002I, HHSN268201700003I, HHSN268201700004I, and HHSN268201700005I). C.M.R. was additionally supported by a grant from the National Institute of Diabetes and Digestive and Kidney Diseases (R03 DK128386). O.T. was supported by a grant from the National Heart, Lung, and Blood Institute (T32HL007024).

Disclosures

None.

Supporting information

Data S1

Tables S1–S3

Figures S1–S2