Plasma Proteins Associated with Whole Grains and Fiber Intake in the Cardiovascular Health Study (CHS) and the Atherosclerosis Risk in Communities (ARIC) Study

Abstract

Background

Limited biomarkers exist for whole grains and fiber intake. Large-scale proteomics can be used to identify objective biomarkers of whole grains and fiber intake.

Objectives

We identified plasma proteins of whole grains and fiber intake in the Cardiovascular Health Study (CHS) and Atherosclerosis Risk in Communities (ARIC) Study.

Methods

Whole grains and fiber intake were quantified from food frequency questionnaire responses in CHS (1989–1990, N = 2411). An aptamer-based technology (SomaLogic) quantified 4979 proteins in stored specimens collected in 1992–1993. Multivariable linear regression identified individual proteins significantly associated with whole grains or fiber intake at a false discovery rate of <0.05. Findings were externally replicated in the ARIC Study. Pathway overrepresentation analysis was conducted for diet-related proteins. Least absolute shrinkage and selection operator (LASSO) regression was used to select a set of proteins associated with whole grains or fiber intake. Prediction statistics assessed the ability of diet-related proteins to predict whole grains and fiber intake beyond participant characteristics.

Results

A total of 68 and 26 proteins were associated with whole grains and fiber intake, respectively, in CHS, with 17 and 20 proteins for whole grains and fiber, respectively, replicated in the ARIC Study. Of these, 5 proteins (beta-glucuronidase, neuropeptide W, inhibin beta C chain, transcobalamin 1, and toll-like receptor 5) were associated with both whole grains and fiber intake. Ascorbate and aldarate metabolism were overrepresented for both whole grains and fiber-related proteins. A total of 13 proteins for whole grains and 19 proteins for fiber intake were selected from LASSO to improve the prediction of these dietary exposures beyond participant characteristics (range of differences in AUCs: 0.022–0.041, P for all tests < 0.05).

Conclusions

We identified proteins and pathways that are robustly associated with whole grains and fiber intake in 2 studies. These proteins may serve as candidate biomarkers if validated in controlled feeding studies.

Introduction

Grain-based foods make up 25% to 30% of the world’s energy source, and in the United States, consumption of grain-based foods (as a proportion of total energy) has increased from 2003 to 2018 [1,2]. The Dietary Guidelines for Americans 2020–2025 recommend adults consume 48 g of grains (at least half of grain consumption as whole grains), and a mean intake of 25–30 g fiber per day [[3], [4], [5]]. Whole grains, which refer to the entire grain composed of the endosperm, germ, and bran [6], and dietary fiber, which is defined as the part of the plant that is indigestible in the human digestive tract [7], are crucial components of a healthy diet as sources of micronutrients and phytochemicals, including vitamins B, E, and iron [8,9]. Diets high in whole grains are associated with lower cholesterol levels [10], risk of cardiovascular disease (CVD) [11], cancer [6], and all-cause mortality [6]. Higher fiber intake increases insulin sensitivity, improves satiety, aids in weight loss, lowers cholesterol, and is associated with a lower risk of CVD and cancer, while also playing a role in the regulation of inflammation and appetite [12].

Although whole grains and fiber intake are associated with health benefits, assessing intake of whole grains and fiber is challenging. Self-reported diet is influenced by systematic errors such as recall bias and measurement error [13]. Objective biomarkers of whole grains and fiber intake can significantly improve dietary assessment. Prior studies examined whether plasma alkylresorcinols, a phenolic lipid group with hydrophobic alkyl chains abundant in wheat and rye, could be considered an objective biomarker of whole grain intake [14]. Besides alkylresorcinols, limited biomarkers of whole grain intake and fiber intake exist [15].

Proteomics offers the opportunity to understand protein expression in relation to dietary intake. Plasma proteins play a role in many biological processes, such as transporting nutrients, acting as enzymes to metabolize nutrients, mediating cellular responses, and activating a cascade of reactions to support immune responses [16]. Previous studies reported that nutrients that are abundant in whole grains (e.g., vitamin E and selenium) are known to be strongly associated with circulating levels of their carrier proteins (e.g., alpha-tocopherol and apolipoprotein C III; selenium and selenoprotein P isoform 1, respectively) [17,18]. Identifying proteins associated with whole grains and fiber intake could help discover candidate biomarkers of dietary intake and allow for exploration of the potential pathways underlying diet–disease associations. To date, there is little evidence of proteomic markers of whole grain intake and fiber intake in humans.

Therefore, we aimed to investigate the association between 1) whole grain intake and 2) fiber intake using large-scale proteomics in the Cardiovascular Health Study (CHS) and to externally replicate findings in the Atherosclerosis Risk in Communities (ARIC) Study.

Methods

Study design and population

The CHS is a multicenter prospective cohort study of older adults in the United States. Participants aged ≥65 y were recruited from 4 sites (Hagerstown, Maryland; Winston-Salem, North Carolina; Sacramento, California; and Pittsburgh, Philadelphia) from 1989 to 1990. Annual follow-up visits occurred for 10 y between 1989 and 1999 and again from 2005 to 2006. Measures such as cardiovascular-related events and baseline medical history were recorded, and blood samples were collected during these visits. Participants were contacted every 6 months via phone to collect information on hospitalizations and cardiovascular-related events [19]. We used dietary data collected at baseline (1989–1990) in the CHS. Plasma specimens used for the proteomics data were collected in year 5 (1992–1993). The median time between dietary intake and proteomic data was 3 y (IQR: 2 y). Of the 5201 CHS participants who completed the dietary assessment between 1989 and 1990, 998 participants with prevalent CVD (myocardial infarction, angina, stroke, transient ischemic attack, claudication, coronary bypass surgery, angioplasty, and carotid endarterectomy) were excluded from the analysis (n = 998) (Supplemental Figure 1). Participants who were missing dietary data or had implausible dietary intake (n = 91), missing covariates (n = 971), or missing proteomics data were also excluded from the analysis (n = 821). Our analytic sample was 2411 participants for the whole grains’ analysis. Additional participants who were missing information related to fiber intake were excluded. Some participants had missing values for fiber intake due to missing information in a small number of foods containing fiber, leaving 2159 participants for our fiber analysis. Although CHS enrolled additional participants (largely Black) in 1992–1993, these participants were not included in our study because they completed dietary assessment (1995–1996) later than proteomic assessment (1992–1993).

The ARIC Study is a multicenter prospective cohort study of predominantly Black and White Americans. Middle-aged adults (45–64 y at baseline) were enrolled into the study from 1987 to 1989 from 4 study sites (Washington County, Maryland; Forsyth County, North Carolina; suburban Minneapolis, Minnesota; and Jackson, Mississippi). Participants returned for study visits periodically. We used dietary data collected at visit 3 (1993–1995), where proteomics data were also assessed [20]. Of the 11,102 ARIC Study participants who completed the dietary assessment using a food frequency questionnaire (FFQ) and had proteomics data, the same exclusion criteria (prevalent CVD, missing dietary data or had implausible dietary intake, or missing covariates) as CHS were applied (Supplemental Figure 2). The final analytic sample for the ARIC Study was 9544. A complete case analysis was conducted for both the CHS and ARIC Study. Although the sample size was larger for the ARIC Study, CHS was selected as the discovery cohort, given that the FFQ used in CHS included more foods that were considered whole grains.

Covariates

In CHS, age, sex, study site, education level, race, and physical activity were self-reported at study enrollment. The estimated glomerular filtration rate (eGFR) was calculated from serum creatinine collected using the Chronic Kidney Disease Epidemiology Collaboration equation [21]. All covariates used were recorded at the time of blood draw (1992–1993), such as smoking status (categorical variable: current, former, and never smoker), alcohol intake (number of alcoholic beverages per week), BMI, hypertension status, diabetes status, and self-reported health (categorical variable: ranging from excellent to poor). Physical activity (excluding chores, quantified as kcal/wk) was assessed using a modified Minnesota Leisure Time Activities Questionnaire [22,23]. Hypertension was defined as elevated blood pressure from the mean of 2 seated resting blood pressure measurements (systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg) or use of antihypertensive medication (yes, no). Diabetes was defined as elevated fasting or nonfasting glucose (fasting glucose ≥126 mg/dL or nonfasting glucose ≥200 mg/dL) or use of diabetes medication (yes, no). BMI was calculated using measured weight and height. We adjusted for eGFR, alcohol intake, physical activity, total energy intake, and BMI as continuous variables.

All covariates in the ARIC Study were harmonized with the covariates in CHS. Age, sex, education, smoking status, and physical activity were self-reported [20]. Given the nonuniform distribution of racial groups in the ARIC Study, a categorical variable combining race and center was used as a covariate. Physical activity was assessed using the modified Baecke questionnaire, which was derived by considering information on frequency, duration, and intensity of leisure-time sports and exercise [24,25].

Whole grain and fiber intake

In the CHS, usual dietary intake was assessed using the validated [26,27] National Cancer Institute 99-item picture-sort FFQ at baseline (1989–1990). The FFQ determined frequency of intake into 5 categories: “almost every day or ≥5 times per week,” “about 1 to 4 times per week,” “about 1to 3 times per month,” “about 5 to 10 times per year,” or “never.” Whole grain intake was determined by combining the intake of dark bread, high fiber cereals, bran, or granola, highly fortified cereals, and cooked cereals. These food groups were considered “whole grain and high fiber grains,” not strictly “whole grains.” For brevity and improved readability, we referred to “whole grain and high fiber grains” as “whole grains” throughout the paper. Fiber intake was derived using the USDA food composition tables based on the responses to the FFQ [28,29]. Energy-adjusted intake of 2 dietary exposures was derived using the residual method [30].

In the ARIC Study, usual dietary intake was assessed using the 66-item modified semiquantitative Willett FFQ [31]. Participants reported the frequency with which they consumed foods and beverages of a particular serving size in the past year, ranging from almost never to >6 times a day [32]. The reliability of the ARIC Study was quantified [33]. Whole grain intake was calculated by summing the intake of cooked cereals and dark or whole grain bread. Nutrient intake, including fiber intake, was determined by multiplying the frequency of consumption and portion size of each food item by the content of each food item. Similar to CHS, energy-adjusted intake of whole grains and fiber was derived using the residual method [30].

Proteomic profiling

Both CHS and the ARIC Study used SOMAScan 5K version 4.0 (SomaLogic) to conduct large-scale proteomic profiling using fasting plasma specimens (stored at −70°C to –80°C after blood draw). The SOMAScan assay uses chemically modified nucleotides (named Slow Off-rate Modified Aptamers “SOMAmers”) to bind to protein epitopes for the measurement of many proteins. The assay then converts the amount of protein present in specimens into relative fluorescent units (approximately proportional to the protein concentrations present) using microarrays. The SOMAScan assay had high reproducibility in CHS participants [34]. Detailed methods and the list of ∼5000 proteins identified using the SOMAScan assay have been described previously [35,36]. In CHS, several quality control steps were taken. Proteins that were indicated as “deprecated” or “nonhuman” proteins were excluded, yielding 4979 aptamers. Adaptive normalization by maximum likelihood was performed to remove biases related to the specimen or assay. No batch effect, clustering based on laboratory handling, or undue influence of outlier values was observed in the specimens [34]. For all analyses, aptamer levels were log transformed and standardized to a mean of 0 and an SD of 1 [34]. To assess the reliability of the SOMAScan platform in CHS, 100 participants for whom plasma specimens were drawn from 1992–1993 and 1997–1998 examinations were examined [34]. The median intraclass correlation coefficient across 4979 aptamers were 0.66 (IQR, 0.35), the median intra-assay coefficient of variation was 3.4%, and the median interassay coefficient of variation was 4.4% among these 100 participants [34]. The correlations between these proteomic results and reference standard measurements for C-reactive protein, cystatin C, and N-terminal pro-B-type natriuretic peptide were all >0.90 [34].

In the ARIC Study, the same proteomic platform was used to conduct proteomic profiling in fasting plasma specimens [20]. In the ARIC Study, proteins with a coefficient of variation >50% and proteins with small variance (<0.01 on a log scale) were excluded. Outliers were capped at $\pm$5 SD. After quality control, we analyzed the proteins using the same procedure as CHS. For all analyses, after log transformation, aptamer levels were standardized to have a mean of 0 and an SD of 1.

Statistical analysis

Baseline characteristics of the participants and nutritional characteristics of diets were examined according to quartiles of whole grains and fiber intake. Means (SDs) were used for continuous variables, and proportions were used for categorical variables.

Multivariable linear regression models were used to assess the association between whole grain intake (independent variable, per serving higher) and individual proteins (response variable). We used 2 nested models: Model 1 was adjusted for age, sex, race, study site, total energy intake, education level, physical activity, smoking status, alcohol intake, and eGFR. Model 2 was adjusted for covariates in model 1 and BMI, hypertension, diabetes, and dietary factors that might be correlated with whole grain intake (fruit, vegetable, and refined grain intake). Proteins with a false discovery rate (FDR) at <0.05 were considered significant. We considered model 1 (sociodemographic characteristics, health behaviors, and eGFR) without potential mediators to be our main model. Then, significant proteins were assessed for external replication in the ARIC Study using the same set of models. Proteins that met FDR <0.05 and had the same direction of association were considered replicated. Analyses were repeated for fiber intake. As sensitivity analyses, among the externally replicated proteins from our main model, we 1) performed another set of models (“model 1 + clinical covariates”) which included all covariates in model 1 and only the clinical characteristics (i.e., fruit, vegetable, and refined grain intake were not adjusted) and 2) compared beta coefficients from the main model that includes compared with excludes those with hypertension or diabetes, to further reduce the influence of medication use on the plasma proteome.

Next, to identify specific biologic pathways that were impacted by whole grain intake and fiber intake, pathway overrepresentation analysis was conducted for significant whole grain- and fiber-related proteins using the Web-based gene set analysis toolkit (WebGestalt) [37]. Pathway overrepresentation test is a statistical method that tests whether proteins of interest (i.e., diet-related proteins) are overrepresented in a predefined biological pathway more than expected by chance, relative to a background list of proteins (4979 proteins) [38]. We included proteins that were significant at FDR <0.05 and were adjusted for age, sex, race, study site, and total energy intake in CHS. We used this minimally adjusted model to maximize the opportunity to identify overrepresented pathways. Proteins were annotated using the Kyoto Encyclopedia of Genes and Genomes database with a reference set of 4979 proteins. Pathways with <5 or >2000 pathways were excluded from our analysis. We included information from the top 5 pathways (significant at P value <0.05 only).

Lastly, we aimed to assess whether a smaller set of whole grains and fiber-related proteins can improve the prediction of these 2 dietary exposures beyond covariates in model 1. Least absolute shrinkage and selection operator (LASSO) regression was performed using embedded feature selection. All covariates from model 1 were included without being penalized. The full set of predictors used the 17 externally replicated whole grain–associated proteins and 20 fiber-associated proteins. A 5-fold cross-validation was conducted to tune the regularization parameter. Proteins with a nonzero coefficient were recorded.

To predict whole grain intake and fiber intake, defined as the highest quartile of intake compared with the lower 3 quartiles, we evaluated 3 models: 1) using only proteins, 2) using only model 1 covariates, and 3) using both covariates and proteins. We evaluated the ability of all selected proteins from LASSO regression to assess whether the inclusion of proteins significantly improved prediction over clinical covariates alone using DeLong’s test.

Results

Characteristics of CHS and ARIC participants

Participants in the CHS had a mean of 1.3 servings (range: 0–4) of whole grains per d and 18 g (range: 4.5–44) of fiber per day (Table 1). Participants in the ARIC Study had 1 serving (range: 0–12) of whole grain per day and 17 g (range: 1.5–120) of fiber per day. Compared with participants in the ARIC Study, participants in CHS were older, more likely to be female, White, and to have more than a high school education. Participants in the ARIC Study were more likely to have better kidney function, higher BMI, and diabetes. Compared to those in the lowest quartile of whole grain intake, those in the highest quartile were more likely to be female, had an education higher than high school, had lower amounts of alcohol intake, and were less likely to have diabetes and hypertension (Table 2). Participants in the highest compared with lowest quartiles of whole grains intake had higher intake of carbohydrate as a percent of total energy, fiber-rich foods (cereal, fruit, and vegetable), fruits, vegetables, and micronutrients (calcium, phosphorus, vitamins A, E, thiamin, riboflavin, vitamin B12, and β-carotene), and ratio of polyunsaturated to saturated fat and lower saturated fat, refined grains, and dietary cholesterol. Similar patterns were also observed with the highest compared with the lowest quartile of fiber intake (Supplemental Table 1).

TABLE 1.

Baseline characteristics of participants in the Cardiovascular Health Study (CHS) and the Atherosclerosis Risk in Communities (ARIC) Study

N	CHS	ARIC Study
	2411	9544
Calculated age at baseline (y), mean (SD)1	71.7 (4.8)	59.9 (5.7)
Sex, n (%)
Female	1539 (63.8)	4065 (42.6)
Male	872 (36.2)	5479 (57.4)
Race, n (%)
White	2285 (94.8)	7589 (79.5)
Black	115 (4.8)	1927 (20.2)
American Indian/Alaskan Native	4 (0.2)	7 (0.1)
Asian/Pacific Islander	3 (0.1)	21 (0.2)
Other	4 (0.2)	0 (0)
Education (y), n (%)
<12	538 (22.3)	1792 (18.8)
12	710 (29.4)	4064 (42.6)
>12	1163 (48.2)	3688 (38.6)
Smoking status, n (%)
Never smoked	1197 (49.6)	4052 (42.5)
Former smoker	957 (39.7)	3845 (40.3)
Current smoker	257 (10.7)	1647 (17.3)
Physical activity3	1983.0 (2112.2)	2.5 (0.8)
Total energy intake	1812 (645)	1615 (591)
Alcoholic beverages/wk	2.8 (6.4)	3.1 (8.1)
Estimated GFR, CKD-EPI formula	63.7 (13.8)	89.9 (14.2)
BMI (kg/m2)	26.5 (4.4)	28.5 (5.5)
Diabetes, n (%)
No	2134 (88.5)	8131 (85.2)
Yes	277 (11.5)	1413 (14.8)
Hypertension, n (%)
No	979 (40.6)	5763 (60.4)
Yes	1432 (59.4)	3781 (39.6)
Whole grain intake, energy-adjusted2	1.3 (0.9)	1.0 (1.0)
Fiber intake, energy-adjusted	18.0 (5.4)	17.6 (8.4)

Abbreviations: ARIC, Atherosclerosis Risk in Communities; CHS, Cardiovascular Health Study; CKD-EPI, Chronic Kidney Disease Epidemiology Collaboration equation; GFR, glomerular filtration rate.

¹Continuous variables are expressed as means and SDs, and categorical variables are expressed as sample size (proportions. Sample size and proportions were reported for categorical variables.

²Nutrients were recorded as a daily average over the whole year and energy adjusted.

³In CHS, physical activity was assessed with the validated Minnesota Leisure-Time Activities questionnaire, and unit was kcal/d. In ARIC Study, physical activity ranged from 1 to 5, which was derived by considering information on frequency, duration, and intensity of leisure-time sports and exercise.

TABLE 2.

Baseline characteristics according to quartiles of whole grains intake in the Cardiovascular Health Study (N = 2411)¹

N	Quartile 1	Quartile 2	Quartile 3	Quartile 4	P value
	603	603	603	602
Calculated age at baseline (y), mean (SD)	71.7 (4.8)	72.0 (4.8)	71.5 (4.7)	71.7 (4.9)	0.27
Sex, n (%)
Female	357 (59.2)	379 (62.9)	391 (64.8)	412 (68.4)	0.009
Male	246 (40.8)	224 (37.1)	212 (35.2)	190 (31.6)
Race, n (%)
White	557 (92.4)	579 (96.0)	573 (95.0)	576 (95.7)	0.088
Black	42 (7.0)	23 (3.8)	28 (4.6)	22 (3.7)
American Indian/Alaskan Native	2 (0.3)	1 (0.2)	1 (0.2)	0 (0.0)
Asian/Pacific Islander	0 (0.0)	0 (0.0)	1 (0.2)	2 (0.3)
Other	2 (0.3)	0 (0.0)	0 (0.0)	2 (0.3)
Study site, n (%)
Bowman Gray	130 (21.6)	155 (25.7)	148 (24.5)	142 (23.6)	0.034
Davis	137 (22.7)	146 (24.2)	174 (28.9)	169 (28.1)
Hopkins	146 (24.2)	135 (22.4)	140 (23.2)	144 (23.9)
Pittsburgh	190 (31.5)	167 (27.7)	141 (23.4)	147 (24.4)
Education (y), n (%)
<12	175 (29.0)	121 (20.1)	121 (20.1)	121 (20.1)	<0.001
12	185 (30.7)	184 (30.5)	183 (30.3)	158 (26.2)
>12	243 (40.3)	298 (49.4)	299 (49.6)	323 (53.7)
Smoking status, n (%)
Never smoked	280 (46.4)	295 (48.9)	314 (52.1)	308 (51.2)	0.005
Former smoker	236 (39.1)	246 (40.8)	225 (37.3)	250 (41.5)
Current smoker	87 (14.4)	62 (10.3)	64 (10.6)	44 (7.3)
Total kcals physical activity, kcal/d	1912.9 (2222.3)	1935.1 (2051.9)	2044.4 (2154.5)	2039.9 (2015.5)	0.59
Total energy intake, kcal/d	1865 (713)	1765 (616)	1702 (604)	1916 (628)	<0.001
Number of alcoholic beverages/wk	3.8 (8.3)	3.1 (6.7)	2.3 (5.4)	1.8 (4.3)	<0.001
eGFR, CKD-EPI formula	63.1 (14.5)	63.2 (13.5)	63.9 (14.2)	64.5 (13.0)	0.23
BMI from imputed height (kg/m2)	26.7 (4.2)	26.6 (4.5)	26.5 (4.4)	26.1 (4.5)	0.096
Diabetes, n (%)
No	527 (87.4)	511 (84.7)	550 (91.2)	546 (90.7)	<0.001
Yes	76 (12.6)	92 (15.3)	53 (8.8)	56 (9.3)
Hypertension, n (%)
No	226 (37.5)	236 (39.1)	250 (41.5)	267 (44.4)	0.083
Yes	377 (62.5)	367 (60.9)	353 (58.5)	335 (55.6)
Whole grains, servings/d	2.4 (2.0)	6.2 (1.9)	9.3 (2.0)	15.3 (3.5)	<0.001
Refined grain, servings/d	1.02 (0.67)	0.87 (0.66)	0.78 (0.64)	0.94 (0.75)	<0.001
Fruit, servings/d	1.9 (1.1)	2.1 (1.0)	2.2 (1.1)	2.4 (1.1)	<0.001
Vegetables, servings/d	2.3 (1.3)	2.4 (1.3)	2.5 (1.2)	2.9 (1.4)	<0.001
Beef or pork, servings/d	1.1 (0.8)	0.8 (0.6)	0.7 (0.6)	0.7 (0.6)	<0.001
Fiber, g/d	14.9 (4.7)	17.2 (5.0)	18.8 (4.9)	21.1 (5.0)	<0.001
Cereal fiber, g/d	1.3 (1.1)	3.4 (1.3)	4.9 (1.4)	7.4 (1.7)	<0.001
Fruit fiber, g/d	4.7 (2.7)	5.1 (2.6)	5.4 (2.7)	5.8 (2.7)	<0.001
Vegetable fiber, g/d	6.6 (3.7)	6.6 (3.4)	6.7 (3.4)	7.7 (4.0)	<0.001
Dietary fiber (daily average over whole mean) (g)	15.7 (6.7)	17.1 (6.0)	18.4 (6.3)	22.3 (6.9)	<0.001
% kcal from carbohydrates	42.7 (7.6)	46.6 (7.7)	49.1 (7.9)	51.4 (7.1)	<0.001
% kcal from proteins	17.4 (3.3)	17.5 (2.8)	17.7 (2.7)	17.8 (2.6)	0.13
% kcal Saturated fat	13.4 (2.9)	12.0 (2.9)	11.2 (2.9)	10.4 (2.7)	<0.001
Polyunsaturated/saturated fat ratio	0.36 (0.25)	0.40 (0.28)	0.41 (0.26)	0.36 (0.20)	<0.001
Saturated fat (g)	14.9 (3.2)	13.4 (3.3)	12.5 (3.2)	11.6 (3.0)	<0.001
Calcium (mg)	433.3 (142.1)	481.0 (154.3)	516.5 (153.0)	524.1 (130.4)	<0.001
Phosphorous (mg)	685.5 (113.3)	731.1 (120.9)	767.1 (121.7)	799.7 (112.7)	<0.001
Iron (mg)	7.2 (1.3)	7.9 (1.3)	8.5 (1.5)	9.7 (2.0)	<0.001
Sodium (mg)	1681.4 (337.8)	1743.1 (345.0)	1769.3 (327.5)	1851.4 (328.1)	<0.001
Potassium (mg)	1722.4 (323.5)	1852.0 (353.1)	1933.0 (339.3)	1947.7 (307.8)	<0.001
Vitamin A (IU)	6915.9 (3329.6)	7506.6 (3366.5)	7745.6 (3183.2)	8227.1 (2966.7)	<0.001
Vitamin B12 (mg)	6.8 (3.7)	7.4 (3.7)	7.4 (3.4)	8.6 (3.5)	<0.001
Thiamin (mg)	0.7 (0.1)	0.8 (0.1)	0.8 (0.2)	1.0 (0.2)	<0.001
Riboflavin (mg)	1.0 (0.3)	1.1 (0.3)	1.2 (0.3)	1.4 (0.3)	<0.001
Beta-carotene (mg)	2579.0 (1507.2)	2974.6 (1838.8)	3.099.7 (1675.0)	3153.3 (1590.5)	<0.001
Niacin (mg)	10.5 (2.2)	11.1 (2.1)	11.7 (2.3)	13.1 (2.5)	<0.001
Vitamin C (mg)	105.1 (51.5)	117.8 (52.3)	123.2 (57.7)	120.91 (47.6)	<0.001
Vitamin E (mg)	4.6 (1.6)	5.5 (2.3)	6.3 (3.3)	8.3 (4.3)	<0.001
Oleic acid (gm)	14.9 (3.1)	13.4 (3.2)	12.5 (3.0)	11.7 (2.9)	<0.001
Linoleic acid (gm)	7.9 (2.5)	7.4 (2.5)	7.1 (2.4)	6.9 (2.2)	<0.001
Cholesterol (mg)	209.0 (90.0)	179.0 (72.1)	164.1 (70.7)	152.5 (68.6)	<0.001

Abbreviations: CKD-EPI, Chronic Kidney Disease Epidemiology Collaboration equation; eGFR, estimated glomerular filtration rate.

¹Continuous variables are expressed as means and SDs, and categorical variables are expressed as sample size (proportions). Sample size and proportions were reported for categorical variables. Nutrients were recorded as a daily average over the whole year and energy-adjusted as per 1000 kcal.

Proteins associated with whole grains and fiber intake

In CHS, out of 4979 proteins, 68 proteins were associated with whole grain intake in our main model (model 1) at FDR <0.5 (Supplemental Table 2). Fatty acid–binding protein, heart (FABP3), and inhibin beta C chain (INHBC), negatively associated with whole grain intake, were among the proteins with the smallest P values. Synaptic vesicle membrane protein VAT-1 homolog (VAT1) was also inversely associated with whole grain intake. When adjusting for additional clinical covariates and dietary factors that correlated with whole grain intake, no significant proteins were associated with whole grains.

We found that 26 proteins were significantly associated with fiber intake at FDR <0.05. When additional clinical factors and dietary factors (fruit, vegetable, and refined grain intake) were adjusted for, 1 protein, kynureninase (KYNU), was inversely associated with fiber intake ($\beta$for per gram higher in fiber = −0.35; 95% confidence interval: −0.49, −0.20). Twelve proteins were associated with both whole grain and fiber intake in model 1.

External replication in ARIC

Seventeen (out of 68 proteins) proteins associated with whole grain intake in CHS were replicated in the ARIC Study at FDR <0.05 (Figure 1; Supplemental Table 2). Twenty (of 26 proteins) proteins associated with fiber intake were replicated in the ARIC study at FDR <0.05 (Figure 1). Apolipoprotein C3 (APOC3) was positively associated with fiber intake, whereas FABP3 was negatively associated with fiber intake in both CHS and ARIC. Five proteins [beta-glucuronidase (GUSB), neuropeptide W (NPW), INHBC, transcobalamin 1 (TCN1), and toll-like receptor 5 (TLR5)] were inversely associated with whole grain intake and fiber intake in both CHS and ARIC. The direction of associations remained the same for all proteins for model 2.

FIGURE 1.

Seventeen proteins associated with whole grain intake and 20 proteins associated with fiber intake in CHS and ARIC Study. Multivariable linear regression models were used to investigate the association between (A) whole grain intake and (B) fiber intake and 4979 proteins, adjusting for age, sex, race, study site, total energy intake, education level, physical activity, smoking status, alcohol intake, and estimated glomerular filtration rate in CHS. Seventeen proteins for whole grain intake and 20 proteins for fiber intake that were externally replicated in the ARIC study at FDR <0.05 are shown in this plot. Full names of each protein can be found in Supplemental Table 2. ∗Indicates proteins also significant in model 2 at FDR <0.05, adjusting for all covariates presented earlier and BMI, hypertension, diabetes, and additional dietary factors that may be correlated with whole grain or fiber intake (fruit, vegetable, and refined grain intake). The dots on the horizontal lines represent the beta coefficient, and the horizontal black lines indicate the 95% confidence interval. ARIC, Atherosclerosis Risk in Communities; CHS, Cardiovascular Health Study; FDR, false discovery rate.

In sensitivity analyses, of the 17 externally replicated whole grain–related proteins and 20 externally replicated fiber-related proteins, the estimates were not substantially different between the fully adjusted model and the model that did not adjust for fruits, vegetables, and refined grains (Supplemental Table 3). Beta coefficients were similar in the main model when we included compared with excluded those with hypertension or diabetes (Supplemental Table 4).

Pathway overrepresentation analysis

For both whole grains and fiber intake in CHS, the ascorbate and aldarate metabolism was overrepresented (P < 0.05 for all tests) (Table 3). For whole grain–related proteins, salivary secretion and glycosaminoglycan biosynthesis were overrepresented. For fiber-related proteins, histidine metabolism, glycosaminoglycan degradation, thiamine metabolism, and beta-alanine metabolism were overrepresented.

TABLE 3.

Pathway overrepresentation analysis for proteins significantly associated with whole grains and fiber intake in the Cardiovascular Health Study (CHS)¹

Dietary intake	KEGG ID	Pathway	Enrichment ratio	Number of proteins in pathway	P	Matched proteins2
Whole grains	hsa04970	Salivary secretion	5.482	28	0.0054	CST3, AMY2A, ATP1B2, CAMP
	hsa00053	Ascorbate and aldarate metabolism	8.527	9	0.0214	GUSB, KL
	hsa00533	Glycosaminoglycan biosynthesis	6.977	11	0.0316	B4GALT2, ST3GAL1
Fiber	hsa00340	Histidine metabolism	13.430	8	0.0089	ALDH2, CNDP1
	hsa00053	Ascorbate and aldarate metabolism	11.938	9	0.0112	ALDH2, GUSB
	hsa00531	Glycosaminoglycan degradation	10.744	10	0.0139	HEXB, GUSB
	hsa00730	Thiamine metabolism	9.767	11	0.0168	ALPP, ALPG
	hsa00410	Beta-alanine metabolism	8.953	12	0.0199	ALDH2, CNDP1

Abbreviations: GUSB, beta-glucuronidase; KEGG, Kyoto Encyclopedia of Genes and Genomes.

¹We used 127 and 78 significant proteins associated with whole grain intake and fiber intake, respectively, adjusting for age, sex, race, study site, and total energy intake at a false discovery rate <0.05. As the reference set, we used 4979 proteins. Web-based gene set analysis toolkit (Webgestalt) was used for pathway analysis. Proteins were annotated to the KEGG pathway. Pathways with <5 and >2000 proteins were excluded. The full protein names can be found in the supplemental material.

²Beta-1, 4-galactosyltransferase 2, B4GALT2; cystatin-C, CST3; pancreatic alpha-amylase, AMY2A; Sodium/potassium-transporting ATPase subunit Beta-2, ATP1B2; antibacterial protein LL-37, CAMP; klotho, KL; CMP-N-acetylneuraminate-beta-galactosamide-alpha-2,3-sialyltransferase 1, ST3GAL1; aldehyde dehydrogenase, mitochondrial, ALDH2; beta hexosaminidase subunit beta, HEXB; alkaline phosphatase, placental type, ALPP; alkaline phosphatase, placental-like, ALPG; beta-ala-his dipeptidase, CNDP1.

Prediction of whole grain intake and fiber intake

LASSO regression further narrowed down the externally replicated proteins that were associated with whole grain intake or fiber intake. Thirteen out of 17 proteins for whole grain intake and 19 of 20 proteins for fiber intake were selected. For example, procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2 (PLOD2), thrombospondin-2 (THBS2), neural cell adhesion molecule 1 (NCAM1), and 13 other proteins were selected for whole grain intake, and Malignant T-cell- amplifed sequence 1 reinitiation and release factor (MCTS1), IL-1 receptor antagonist (IL1RN), and 18 other proteins were selected for fiber intake (Table 4). Altogether, 13 proteins significantly improved the prediction of whole grain intake (AUC for covariates only = 0.600, AUC for covariates + proteins = 0.622, difference in AUC covariates only and covariates + proteins = 0.022, P value of 0.012). Nineteen proteins altogether significantly improved the prediction of the highest quartile of fiber intake (AUC for covariates only = 0.641, AUC for covariates + proteins = 0.682, difference in AUC covariates only and covariates + proteins = 0.041, P value <0.001).

TABLE 4.

AUC for selected proteins to predict the highest vs. lower 3 quartiles of adherence to whole grains intake and fiber intake

Dietary intake	Proteins	AUC for proteins only	AUC for covariates only1	AUC for covariates + proteins	Difference in AUC for the model with covariates only vs. covariates + proteins	P value for difference in AUC
Whole grain (n = 13 proteins)	TCN1, VAT1, TLR5, LECT2, PLOD2, CNTN4, NPW, NTRK3, THBS2, NCAM1, INHBC, NEGR1, NCAN	0.590	0.600	0.622	0.022	0.012
Fiber (n = 19 proteins)	TPP1, CYB5R3, NAMPT, TCN1, PLBD1, IGFALS, TLR5, WNT5A, NPW, KYNU, APOC3, FABP3, DUSP13, MCTS1, PCOLCE, IL1RN, IGFBP5, PTPN11, INHBC	0.643	0.641	0.682	0.041	<0.001

Abbreviations: APOC3, apolipoprotein C3; FABP3, fatty acid–binding protein, heart; IL1RN, IL-1 receptor antagonist; INHBC, Inhibin beta C chain; KYNU, kynureninase; NCAM, neural cell adhesion molecule; NPW, neuropeptide W; PLOD2, procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2; TCN1, transcobalamin 1; THBS2, thrombospondin-2; TLR5, toll-like receptor 5; VAT1, synaptic vesicle membrane protein VAT-1 homolog; MCTS1, malignant T-cell-amplified sequence 1; PCOLCE, procollagen C-endopeptidase enhancer 1; NAMPT, nicotinamide phosphoribosyltransferase; WNT5A, protein wnt-5a; IGFALS, insulin-like growth factor-binding protein complex acid labile subunit; IGFBP1, insulin-like growth factor-binding protein 1; NCAN, neurocan core protein; TPP1, tripeptidyl-peptidase 1; CYB5R3, NADH-cytochrome b5 reductase 3; IGFBP5, insulin-like growth factor-binding protein 5; DUSP13, dual specificity protein phosphatase 13 isoform A; LECT2, leukocyte cell-derived chemotaxin-2; CNTN4, contactin-4; PTPN11, tyrosine-protein phosphatase non-receptor type 11; NCAM1, neural cel adhesion molecule 1; PLBD1, phospholipase B-like 1; NEGR1, neuronal growth regulator 1.

¹Thirteen proteins were selected to predict whole grain intake, and 19 proteins were selected to predict fiber intake from least absolute shrinkage and selection operator (LASSO) regression. Covariates were age, sex, race, study site, total energy intake, education level, physical activity, smoking status, alcohol intake, and estimated glomerular filtration rate (eGFR).

Discussion

In 2 prospective cohort studies, 17 proteins and 20 proteins were robustly associated with whole grain intake and fiber intake, respectively. Thirteen out of 17 whole grain–related proteins and 19 of 20 fiber-related proteins, selected from LASSO models, significantly improved the prediction of each of these dietary exposures beyond participant characteristics. Five of these proteins (GUSB, NPW, INHBC, TCN1, and TLR5) were common for whole grains and fiber intake in both CHS and ARIC, which suggests that they might be considered candidate biomarkers of whole grains and fiber intake. We found several pathways relevant to whole grains and fiber intake. These findings contribute evidence on potential pathways through which whole grain and fiber intake are associated with a lower risk of chronic conditions.

Out of the 17 proteins associated with whole grain intake and the 20 proteins associated with fiber intake in CHS and the ARIC Study, many proteins (Phospholipase B-like 1, PLBD1; VAT1; NADH-cytochrome b5 reductase 3, CYB5R3; APOC3; FABP3; Procollagen C-endopeptidase enhancer, PCOLCE; Nicotinamide phosphoribosyltransferase, NAMPT; insulin-like growth factor-binding protein 1, INHBC; IGFBP1; TLR5; KYNU; NPW; NCAN; and neuronal growth regulator 1, NEGR1) have previously been shown in other studies to be associated with healthy dietary patterns, with consistent direction of association [20,[39], [40], [41]]. These results are not surprising, given that whole grains and dietary fiber are important components of high-quality diets. Some of these proteins deserve mention. CYB5R3 is an enzyme involved in fatty acid elongation and cholesterol biosynthesis [40]. Greater amounts of CYB5R3 in the liver could support cell homeostasis, which can be crucial in the regulation of cholesterol and lipid levels in the body [40]. APOC3 and FABP3 are proteins that encode and are involved in the transport of plasma lipoproteins, and APOC3 is known to additionally be involved in the transport of vitamin E [[42], [43], [44]]. In a previous study of 500 Nepalese children, APOC3 was strongly correlated with plasma $\alpha$-tocopherol levels, a lipoprotein antioxidant [17]. Consistent with the prior study in Nepal, in our study, those with greater intake of fiber had higher vitamin E intake. PCOLCE is a secretory protein and regulator of collagen fibrillogenesis, a key process in the development of CVD [45]. Evidence suggests that increased collagen production in cardiac muscle can create cardiac fibrosis and result in abnormalities in heart function [20,[45], [46], [47], [48], [49]]. Taken together, these results highlight multiple biological processes such as lipid metabolism and transport, inflammation, and collagen binding activity, which might be impacted by whole grains and dietary fiber intake.

Further, we found inverse associations between whole grain or fiber intake and INHBC and NPW, consistent with a previous study [50]. INHBC is an inhibin that is involved in insulin secretion [51,52]. Inhibin proteins can limit the activity of follicle-stimulating hormone, which in turn decreases the amount of insulin secreted [53]. NPW is a protein involved in energy homeostasis and food satiation signals, and can act as a modulator of glucose-induced insulin release in animal models [54], corroborating that greater intake of fiber is associated with increased satiety [55]. In our study, KYNU was the only protein that remained significantly associated with fiber intake after adjusting for clinical factors and additional dietary factors. In the Framingham Heart Offspring Study, KYNU was inversely associated with the Alternative Healthy Eating Index (AHEI), Dietary Approaches to Stop Hypertension (DASH), and Mediterranean Diet Score [56], consistent with the direction observed in our study. KYNU is a protein involved in inflammatory pathways [57]. KYNU’s role in inflammation supports that a diet high in fiber can decrease inflammation [29,58,59]. Whole grains and fiber intake can lead to greater insulin sensitivity [60] through glucose oxidation and insulin clearance [61].

Multiple proteins predictive of whole grains and fiber intake in our study have been shown to be associated with cancer, highlighting potential mechanisms through which these dietary exposures are inversely associated with total cancer risk [62,63]. PLOD2, THBS2, and NCAM1 were predictive of whole grains in our study and were associated with both proliferation and suppression of cancer [20,[64], [65], [66], [67], [68]]. PLOD2 and NCAM1 are proteins that have been shown to have a higher presence in cancer cells and play a role in advancing disease progression [[69], [70], [71]]. MCTS1, predictive of fiber intake, is involved in RNA binding activity and has been found to be associated with lower cancer survival time [72,73]. IL1RN is a protein involved in inflammatory cytokines and has been found to be associated with colorectal cancer incidence [74].

The ascorbate and aldarate metabolism, a carbohydrate metabolic pathway that carries out redox reactions [75], and the glycosaminoglycan (degradation and biosynthesis) pathway were overrepresented for both whole grains and fiber-related proteins. In mice, the ascorbate and aldarate metabolism has been shown to protect cells from oxidative damage, an important contributor to inflammation and risk of chronic diseases [75]. Whole grains and fiber intake have been shown to reduce oxidative stress, supporting the significance of this pathway [63,76]. The glycosaminoglycan pathway is involved in the production of glycosaminoglycans, polysaccharides that include hyaluronic acid and heparin [77]. For fiber-related proteins, histidine metabolism was overrepresented. Histidine, an essential amino acid, is found in whole grains such as wheat and quinoa [78]. Therefore, these pathways represent nutrient intake (ascorbate, histidine) and pathways through which whole grains and fiber intake are associated with a lower risk of chronic conditions.

The present study has several strengths. This study identified candidate protein biomarkers of whole grains and fiber intake. The FFQ used in CHS captured a wide range of foods rich in whole grains and fiber, allowing us to comprehensively characterize participants’ dietary intake. Large-scale proteomics of nearly 5000 proteins provided an opportunity to discover potential protein biomarkers of whole grain intake and fiber intake. After discovery, we replicated our findings in an independent population. In addition, proteins that were significantly associated with whole grains and fiber intake in our study were consistent with previous studies of dietary intake and plasma proteins, strengthening our study’s findings.

Our study has several limitations. First, the timing of dietary assessment and proteomic profiling in CHS was not concordant. Dietary intake was assessed at baseline (1989–1990), and proteomic profiling was conducted using blood specimens collected in 1992–1993. It is possible that dietary intake may have changed over time. However, dietary assessment and proteomic measurement were concordant in the ARIC Study (cross-sectional analysis), and it is encouraging that many proteins identified in CHS replicated in the ARIC Study with the same direction of association. Second, self-reported diet is subject to measurement error and systematic biases such as social desirability bias [79]. Nonetheless, FFQ used in the CHS was validated [26,27], and trained interviewers administered FFQ in the ARIC Study with visual guides to estimate portion sizes, which can reduce measurement error. Further, the reliability of FFQ in the ARIC Study was quantified [33]. Third, although the FFQ used in the CHS and ARIC Study asked about a wide range of foods rich in whole grains and fiber, there are likely modern foods that may not have been captured. Fourth, whole grains and fiber can vary widely throughout a country and represent cultural differences; therefore, our findings require replication in different populations. Fifth, study participants in CHS and ARIC were comprised primarily of White and Black participants, which can limit the generalizability. In addition, participants in CHS are older than those in the ARIC Study, which could have limited replication of dietary exposures. However, it is difficult to identify cohorts with similar dietary intake and use the same proteomics platform. Sixth, our dietary exposure, “whole grains and high fiber grains,” included dark bread; high fiber cereals, bran, or granola; highly fortified cereals; and cooked cereals. However, these foods do not necessarily indicate that the grains are whole grains. Seventh, the SOMAScan assay, which is considered to be the most comprehensive way to quantify circulating proteins compared with other proteomics platforms, is an aptamer-based method. Thus, the targeted approach may limit the discovery of proteins that are associated with whole grain or fiber intake [80].

In our study, we identified and externally replicated proteins that are representative of whole grains and fiber intake, and pathways that were involved in the metabolism of these dietary exposures. These findings inform potential proteomic biomarkers of whole grains and fiber intake and improve understanding of the pathways underlying these dietary exposures and chronic conditions.

Author contributions

The authors’ responsibilities were as follows – AF: conducted statistical analysis and drafted the paper; TRA, AGV: contributed to statistical analysis, interpretation, and revision of the manuscript; JAB, KLW, RNL, AMF, DCH, NS, JSF, CMR, BMP: contributed to important intellectual content during manuscript drafting or revision; HK: conducted replication analysis and involved in all aspects of the study, from study conception to study design, analyses, and writing; AF, HK: took responsibility for the design, writing, and final content of the paper; and all authors: read and approved the manuscript.

Data availability

Cardiovascular Health Study (CHS) and Atherosclerosis Risk in Communities (ARIC) Study data are available through the National Heart, Lung, and Blood Institute Biologic Specimen and Data Repository Information Coordinating Center. Interested researchers may also contact the CHS and ARIC Study Coordinating Centers to access data and study materials. Analytic code will be made available upon reasonable request.

Declaration of Generative AI and AI-assisted Technologies in the Writing Process

The authors declare that no generative AI or AI-assisted technologies were used in the writing of this manuscript.

Funding

HK is supported by a grant from the National Heart, Lung, and Blood Institute (NHLBI, K01 HL168232). CHS was supported by contracts HHSN268201200036C, HHSN268200800007C, HHSN268201800001C, N01HC55222, N01HC85079, N01HC85080, N01HC85081, N01HC85082, N01HC85083, N01HC85086, 75N92021D00006, and grants U01HL080295, U01HL130114, R01HL144483, and R01HL172803 from the NHLBI, with additional contributions from the National Institute of Neurological Disorders and Stroke (NINDS). Additional support was provided by R01AG023629 from the National Institute on Aging (NIA). A full list of principal CHS investigators and institutions can be found at CHS-NHLBI.org. Proteomics in CHS was additionally supported by R01HL144483 and R01H-L172803. The Atherosclerosis Risk in Communities study has been funded in whole or in part with federal funds from the NHLBI, NIH, Department of Health, and Human Services, under Contract nos. (75N92022D00001, 75N92022D00002, 5N92022D00003, 75N92022D00004, 75N92022D00005). SomaLogic Inc. conducted the SOMAScan assays in exchange for the use of ARIC data. This work was supported in part by NIH/NHLBI grant R01 HL134320. All papers will be shared with SomaLogic by the ARIC publications committee (aricjhu@jhu.edu). The funders had no role in study design, analysis, writing, and decision to submit the manuscript for publication.

Conflict of interest

The authors report no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: