Proteomic Markers of Dietary Patterns, CKD Progression, and Mortality in the CRIC Study

Visual Abstract

Abstract

Key Points

• Using untargeted proteomics, we identified 199 plasma protein markers of four healthy dietary patterns in adults with CKD.

• Twenty-one diet-related proteins were associated with CKD progression and 30 proteins were associated with all-cause mortality.

• These proteins may represent biologic mechanisms through which diet modifies disease course in CKD.

Background

Healthy dietary patterns reduce the risk of CKD progression and mortality in people with CKD. Identifying protein biomarkers of diet, and their associations with these outcomes, can elucidate biologic mechanisms through which diet improves prognosis.

Methods

Using data from the Chronic Renal Insufficiency Cohort study of adults with CKD (n=2217, mean age 59 years), we examined cross-sectional associations between 4954 plasma proteins and four dietary patterns: Healthy Eating Index-2020, Alternative Healthy Eating Index-2010, Dietary Approaches to Stop Hypertension, and alternate Mediterranean diet. Relative values of proteins were determined using an aptamer-based assay. Dietary intake was assessed using a food frequency questionnaire. We used multivariable linear regression to identify proteins associated with diet, and Cox proportional hazards regression to assess longitudinal associations between diet-related proteins, CKD progression, and mortality. Elastic net regression was used to select subsets of proteins that are associated with these outcomes.

Results

At a false discovery rate-adjusted P < 0.05, 199 proteins were associated with ≥1 dietary pattern and 18 were associated with all patterns. Over 7 years of median follow-up, 824 CKD progression events occurred. Twenty-one proteins were associated with CKD progression at P < 2.5×10⁻⁴ (=0.05/199), of which eight were selected in elastic net regression: follistatin-related protein 3, glutaredoxin-1, asialoglycoprotein receptor 1, extracellular superoxide dismutase [Cu-Zn], zinc-α-2-glycoprotein, IL-18 receptor 1, ecto-ADP-ribosyltransferase 3, and ephrin type-A receptor 1. All were inversely associated with healthy dietary patterns and associated with higher risk of CKD progression. Thirty proteins were associated with all-cause mortality, of which 14 were selected by elastic net regression.

Conclusions

Large-scale proteomics analyses identified potential protein biomarkers of healthy dietary patterns that were associated with CKD progression and mortality in adults with CKD. Their functions, including regulating blood lipids, insulin sensitivity, vascular homeostasis, inflammation, and oxidative stress, may represent mechanisms through which diet improves disease course.

Introduction

Dietary patterns rich in fruits, vegetables, whole grains, legumes, nuts, and seeds and low in red and processed meat, sodium, and refined sugars have been associated with a lower risk of kidney disease progression and mortality in adults with CKD.^1,2 Foods emphasized in healthy dietary patterns are generally base-producing and provide fiber and phytochemicals that favorably alter gastrointestinal microbiota composition and serve as antioxidants.^3–5 Thus, hypothesized mechanisms linking diet with kidney disease prognosis include modulation of metabolic acidosis, uremic toxin generation, oxidative stress, inflammation, and insulin resistance.^6,7

Proteins play key roles in building and repairing tissues, serve as enzymes and hormones, participate in signal transduction, and transport nutrients and other molecules throughout the body, among other functions. Examining proteomic links between diet and disease may elucidate processes through which diet modulates disease progression. Plasma protein correlates of several dietary patterns have been identified and mapped to pathways involved in cellular proliferation, cellular adhesion, immune function, and inflammation,^8–11 which may explain the inverse associations between adherence to healthy dietary patterns and adverse health outcomes.

Proteomic profiles of healthy dietary patterns have not yet been examined in people with CKD, in whom changes in protein filtration, reabsorption, and production associated with kidney function decline may alter the plasma proteome.^12,13 Identifying proteomic markers of diet and their biologic functions in people with kidney disease could help to guide risk assessment and inform interventions to improve disease prognosis. In this study, we aimed to (1) identify plasma proteins associated with adherence to four healthy dietary patterns in adults with CKD and (2) examine longitudinal associations between diet-related proteins and risk of CKD progression and all-cause mortality.

Methods

Study Population

The Chronic Renal Insufficiency Cohort (CRIC) study is an ongoing multicenter prospective cohort study that enrolled 3939 adults aged 21–74 years with reduced eGFR (20–70 ml/min per 1.73 m²) from seven US clinical centers between 2003 and 2008.¹⁴ By design, approximately half of enrolled participants had diabetes and half identified as non-Hispanic Black race.¹⁵ Participants are examined annually at in-person follow-up visits and followed by telephone at 6-month intervals between visits. The study protocol was approved by institutional review boards at participating institutions, participants provided written informed consent, and research procedures complied with ethical principles of the Declaration of Helsinki.

Our study used blood samples collected at the year 1 clinic visit. We included 2217 participants free of ESKD at the year 1 clinic visit who had plasma protein measurements that met predefined quality control standards, completed the baseline diet assessment, and were not missing covariates (Supplemental Figure 1). For the prospective analysis of CKD progression, 88 participants were excluded because they did not have any eGFR recorded after year 1 to assess decline in kidney function.

Dietary Pattern Assessment

Diet was assessed at baseline and year 2 using a 124-item food frequency questionnaire developed by the National Cancer Institute, the Diet History Questionnaire-1¹⁶ (Supplemental Methods). We averaged responses from baseline and year 2 for participants who completed both diet assessments (n=1625, 73%). For the remaining participants who did not complete a reliable year 2 diet assessment or who experienced CKD progression, died, or were lost to follow-up before year 2, dietary intake was represented by baseline responses.

Dietary patterns were scored according to four predefined indices: the healthy eating index (HEI)-2020,¹⁷ the alternative healthy eating index (AHEI)-2010,¹⁸ the dietary approaches to stop hypertension (DASH)-style diet,¹⁹ and the alternate mediterranean diet (aMed; Supplemental Methods).²⁰ Scoring criteria have been published²¹ and applied in CRIC.²² Higher scores represent better alignment with healthy dietary patterns.

Proteomic Measurements

Plasma proteins were profiled using the SomaScan v4.0 assay (SomaLogic, Boulder, CO). The SomaScan platform uses chemically modified single strands of DNA as protein binding agents and quantifies relative protein abundance in relative fluorescence units using standard DNA microarrays.^23,24 Standard normalization and calibration procedures were performed by SomaLogic to reduce systematic biases in raw protein measurements.²⁵ Protein levels were log₂-transformed and winsorized at five standard deviations from the mean. A total of 5284 protein targets were quantified. After excluding 330 targets with a Bland Altman coefficient of variation (CV) >50%,²⁶ variance (on log scale) <0.01, missing Uniprot ID, binding Fc mouse, contaminant, or nonproteins, 4954 protein targets (representing 4711 unique proteins) remained for analysis.

The median Bland-Altman CV for 129 quality control samples was 4.8% including outliers and was 4.6% after excluding one outlier. The median interassay Bland Altman CV for 44 masked duplicate pairs was 8.7% and the median Pearson correlation was 0.89. After removing one outlier, the remaining 43 pairs had a median CV of 1.1% and Pearson correlation of 0.92.

Clinical Outcomes

The primary outcome was CKD progression, defined as ≥50% decline in eGFR from baseline. eGFR was calculated based on age, sex, and serum creatinine using the 2021 CKD Epidemiology Collaboration Equation without race.²⁷ Time to eGFR halving was estimated by assuming a linear decline between clinic visits.²⁸ An assumed eGFR value of 10.9 ml/min per 1.73 m² was used when participants initiated KRT (dialysis or transplant).²⁸ Self-reported initiation of KRT was confirmed by review of dialysis unit or hospital records and supplemented by data from the US Renal Data System.²⁹

We also investigated associations with all-cause mortality. Deaths were ascertained by report from next of kin, death certificates, hospital records, and linkage with the Social Security Death Master File.

Covariates

Participants self-reported sociodemographic information (age, sex, race, ethnicity, education, income), smoking status, and medications using questionnaires. Physical activity was self-reported using the Multi-Ethnic Study of Atherosclerosis Typical Week Physical Activity Survey, which assesses time spent in various activities during a typical week over the past month, and expressed in total weekly metabolic equivalent of task (MET) scores.³⁰ Body mass index (BMI), systolic BP, and 24-hour urine protein were measured according to standard study protocols. Prevalent diabetes was defined as self-reported use of insulin or medications to lower blood sugar, fasting blood glucose >126 mg/dl, or nonfasting glucose >200 mg/dl.

Sociodemographic information was collected at study enrollment. All other covariates except physical activity were assessed at year 1. Physical activity at year 1 was represented by averaged responses from baseline and year 2.

Statistical Analyses

We used multivariable linear regression to assess cross-sectional associations between dietary pattern scores (per 1 SD) and relative abundance of log₂-transformed plasma proteins. Dietary pattern scores were standardized to facilitate comparison of β coefficients, as the indices are scored on different scales. β coefficients represent doubling in protein levels per 1 SD higher diet score. We adjusted for age, sex, race and ethnicity, clinical center, total energy intake, education, income, smoking status, physical activity, BMI, eGFR (linear splines with one knot at 45 ml/min per 1.73 m²), 24-hour urine protein, systolic BP, prevalent cardiovascular disease, prevalent diabetes, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker medication use, and lipid-lowering medication use. Statistical significance of diet-protein associations was determined at a false discovery rate (FDR)–adjusted P < 0.05. Dose-response relationships were also assessed by examining protein levels across quartiles of diet scores, using the median score per quartile to test for trend. We evaluated nonlinearity of diet-protein associations using likelihood ratio tests to compare model fit with diet scores modeled linearly versus restricted cubic splines with four knots (at 5th, 35th, 65th, and 95th percentiles).

We analyzed pathway overrepresentation using the Web-based Gene Set Analysis Toolkit (Web-Gestalt) platform for proteins that were significantly associated with at least one dietary pattern (FDR-adjusted P < 0.05).³¹ To understand the biologic functions of significant proteins, we mapped them to pathways using the Kyoto Encyclopedia of Genes and Genomes functional database.³² We excluded pathways with <5 or more than 2000 proteins. Enriched pathways were considered significant at P < 0.05.

To explore whether specific food groups or nutrients were driving associations with dietary patterns, we examined associations between their scoring components and proteins that were associated with overall dietary patterns.

For proteins that were significantly associated with at least one dietary pattern, we assessed longitudinal associations between protein levels and CKD progression and all-cause mortality using Cox proportional hazards regression and adjusted for the same covariates as cross-sectional models. Person-years were calculated from the time of proteomic profiling (year 1) until an event or censoring by withdrawal, death (for CKD progression), or administrative censoring (December 31, 2021). We also evaluated a model accounting for death as a competing risk for CKD progression.

We used elastic net regression with ten-fold cross-validation to identify a weighted subset of diet-related proteins that were consistently associated with CKD progression and mortality. Proteins that were significantly associated with CKD progression or all-cause mortality at P < 2.5×10⁻⁴ (=0.05/199 proteins) were inputted as predictors in outcome-specific elastic net regression models, along with all covariates included in the Cox regression models. For each of ten iterations (per outcome), a model was generated based on a randomly selected 90% of the study sample and then validated in the remaining 10% of participants. We forced models to retain all covariates. Proteins that were consistently selected in seven or more of the ten iterations were used to build final models. We calculated C-statistics comparing models that included proteins plus all covariates and compared them to models including only covariates to determine whether the proteins improved prediction of CKD progression and mortality.

We explored the consistency of cross-sectional and prospective associations across subgroups defined by sex, baseline eGFR (<45 versus ≥45 ml/min per 1.73 m²), diabetes, and proteinuria (<1.5 versus ≥1.5 g/d). Effect modification was evaluated using likelihood ratio tests at P < 3.7×10⁻⁵ (=0.05/341 diet-protein associations/four subgroups) for cross-sectional diet-protein associations; P < 6.0×10⁻⁴ (=0.05/21 proteins/four subgroups) for CKD progression; and P < 4.2×10⁻⁴ (=0.05/30 proteins/4 subgroups) for all-cause mortality.

Results

Of 2217 participants in our study sample, 47% were female and 39% were non-Hispanic Black (Table 1). The mean age was 59 (SD 11) years, the mean BMI was 31.9 (SD 7.7) kg/m², and the mean eGFR was 44 (SD 16) ml/min per 1.73 m². The mean diet scores were 65 (SD 10) for HEI-2020, 53 (SD 10) for AHEI-2010, 24 (SD 5) for DASH, and 28 (SD 6) for aMed. Pearson correlation coefficients between diet scores ranged from 0.68 (HEI-2020 and aMed) to 0.77 (HEI-2020 and DASH; all P < 0.001). Higher diet scores were associated with older age, lower BMI, and higher eGFR (Supplemental Table 1). People with higher diet scores were more likely to be college graduates and less likely to ever smoke. Those with higher AHEI-2010 and DASH scores were more likely to identify as non-Hispanic White.

Table 1.

Chronic Renal Insufficient Cohort participant characteristics (n=2217)

Characteristic	Mean (SD) or n (%)
Age, yr	59 (11)
Female, n (%)	1047 (47)
Race and ethnicity, n (%)
Non-Hispanic Black	863 (39)
Non-Hispanic White	1169 (53)
Other	185 (8)
HEI-2020 (range 0–100)	65 (10)
AHEI-2010 (range 0–110)	53 (10)
DASH diet score (range 8–40)	24 (5)
aMed (range 9–45)	28 (6)
Total energy intake, kcal	1780 (739)
Education, n (%)
< high school	268 (12)
High school graduate	397 (18)
Some college	667 (30)
College graduate or higher	885 (40)
Income, n (%)
≤$20,000	510 (23)
$20,001–50,000	554 (25)
$50,001–100,000	508 (23)
>$100,000	303 (14)
Don't wish to answer	342 (15)
Smoking status, n (%)
Never	1018 (46)
Former	946 (43)
Current	253 (11)
Physical activity, METs/wk	203 (125)
BMI, kg/m2	31.9 (7.7)
BMI, n (%)
<25 kg/m2	371 (17)
25–<30 kg/m2	655 (30)
30–<35 kg/m2	571 (26)
35–<40 kg/m2	334 (15)
≥40 kg/m2	286 (13)
eGFR, ml/min per 1.73 m2	44 (16)
eGFR, n (%)
≥60 ml/min per 1.73 m2	353 (16)
45–<60 ml/min per 1.73 m2	661 (30)
30–<45 ml/min per 1.73 m2	738 (33)
<30 ml/min per 1.73 m2	465 (21)
Urine protein, g/d	0.8 (1.8)
Urine protein, n (%)
<0.10 g/d	914 (41)
0.10–<0.50 g/d	658 (30)
0.50–<1.50 g/d	308 (14)
≥1.50 g/d	337 (15)
Systolic BP, mm Hg	126 (21)
Systolic BP, n (%)
<120 mm Hg	965 (44)
120–129 mm Hg	474 (21)
130–139 mm Hg	304 (14)
≥140 mm Hg	474 (21)
Prevalent cardiovascular disease, n (%)	747 (34)
Prevalent diabetes, n (%)	1000 (45)
ACEi/ARB medication use, n (%)	1551 (70)
Lipid-lowering medication use, n (%)	1427 (64)

ACEi, angiotensin-converting enzyme inhibitor; AHEI, alternative healthy eating index; aMed, alternate mediterranean diet; ARB, angiotensin II receptor blocker; BMI, body mass index; HEI, healthy eating index; MET, metabolic equivalent of task.

Among the 1625 participants with complete baseline and year 2 dietary data, diet scores were generally consistent across the two assessments (Supplemental Figure 2). The median (25th–75th percentile) 2-year diet score changes for HEI-2020 were 0 (−5 to 5), 0 for AHEI-2010 (−5 to 5), 0 for DASH (−2 to 2), and 0 for aMed (−3 to 3).

Dietary Pattern Associations with Proteins

A total of 341 diet-protein associations were significant at FDR-adjusted P < 0.05, after adjusting for covariates (HEI-2020: 76; AHEI-2010: 59; DASH: 173; aMed: 33; Figure 1, Supplemental Table 2). The number of inverse and positive associations were as follows: HEI-2020: 46 inverse, 30 positive; AHEI-2010: 35 inverse, 24 positive; DASH: 93 inverse, 80 positive; and aMed: 18 inverse, 15 positive). Most of these proteins also exhibited a significant trend across quartiles of diet scores (HEI-2020: 52 of 76; AHEI-2010: 43 of 59; DASH: 84 of 173; aMed: 20 of 33), consistent with a dose-response relationship (Supplemental Table 2). At a Bonferroni-corrected threshold of P < 1.5×10⁻⁴ (=0.05/341 diet-protein associations), there were no significant deviations from linearity (Supplemental Table 2). Associations were consistent across sex, baseline eGFR, diabetes status, and proteinuria subgroups (Supplemental Table 3).

Figure 1.

Plasma proteins associated with healthy dietary patterns in the CRIC study. (A) HEI-2020, (B) AHEI-2010, (C) DASH, (D) aMed. Multivariable linear regression models adjusted for age, sex, race and ethnicity, clinical center, total energy intake, education, income, smoking status, physical activity, BMI, eGFR, 24-hour urine protein, systolic BP, prevalent cardiovascular disease, prevalent diabetes, ACEi/ARB medication use, and lipid-lowering medication use. β-coefficients represent doubling of log₂-transformed protein levels per 1 SD difference in dietary pattern score. Horizontal dashed line represents FDR-adjusted P value < 0.05. ACEi, angiotensin-converting enzyme inhibitor; AHEI, alternative healthy eating index; aMed, alternate mediterranean diet; ARB, angiotensin II receptor blocker; BMI, body mass index; DASH, dietary approaches to stop hypertension; FDR, false discovery rate; HEI, healthy eating index.

A total of 199 proteins were associated with at least one dietary pattern. Of the 199 proteins, 41% (82 proteins) were associated with greater than one dietary pattern. Eighteen proteins were associated with all four dietary patterns, and the direction of associations was consistent across diets (Table 2).

Table 2.

Plasma proteins consistently associated with healthy eating index-2020, alternative healthy eating index-2010, dietary approaches to stop hypertension, and alternate mediterranean diet dietary patterns

UniProt ID	Entrez Gene Symbol	Protein Name	Direction
Q9UKU7	ACAD8	Isobutyryl-CoA dehydrogenase, mitochondrial	↓
P07306	ASGR1	Asialoglycoprotein receptor 1	↓
O75354	ENTPD6	Ectonucleoside triphosphate diphosphohydrolase 6	↓
P09681	GIP	Gastric inhibitory polypeptide	↓
P35754	GLRX	Glutaredoxin-1	↓
Q92743	HTRA1	Serine protease HTRA1	↓
Q13478	IL18R1	IL-18 receptor 1	↓
P18510	IL1RN	IL-1 receptor antagonist protein	↓
P08476	INHBA	Inhibin β A chain	↓
Q08999	RBL2	Retinoblastoma-like protein 2	↓
P05546	SERPIND1	Heparin cofactor 2	↓
P20061	TCN1	Transcobalamin-1	↓
P58335	ANTXR2	Anthrax toxin receptor 2	↑
Q8IWV2	CNTN4	Contactin-4	↑
Q6B8I1	DUSP13	Dual specificity protein phosphatase 13 isoform A	↑
P08833	IGFBP1	Insulin-like growth factor-binding protein 1	↑
Q9BRK3	MXRA8	Matrix-remodeling-associated protein 8	↑
P04155	TFF1	Trefoil factor 1	↑

Listed proteins were significantly associated with all dietary patterns in the same direction (false discovery rate-adjusted P < 0.05). Multivariable linear regression models adjusted for age, sex, race and ethnicity, clinical center, total energy intake, education, income, smoking status, physical activity, body mass index, eGFR, 24-hour urine protein, systolic BP, prevalent cardiovascular disease, prevalent diabetes, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker medication use, and lipid-lowering medication use. Arrows indicate direction of diet-protein association as positive (↑) or inverse (↓). AHEI, alternative healthy eating index; aMed, alternate mediterranean diet; DASH, dietary approaches to stop hypertension; HEI, healthy eating index.

Some proteins were uniquely associated with a single dietary pattern. The DASH dietary pattern was exclusively associated with the most proteins (93 proteins), followed by HEI-2020 (11 proteins) and AHEI-2010 (10 proteins). Three proteins were uniquely associated with the aMed pattern.

Diet-related proteins were mapped to pathways involved in cell adhesion, AA metabolism, retinol metabolism, lysosome activity, ferroptosis, metabolism of xenobiotics, and central carbon metabolism in cancer (Supplemental Table 4).

Dietary Component Associations with Proteins

Sugar-sweetened beverage intake was independently associated with the greatest number of proteins in the AHEI-2010 (Figure 2B) and DASH (Figure 2C) dietary patterns. Total fruit intake was associated with the greatest number of proteins related to the HEI-2020 (Figure 2A) and aMed (Figure 2D) dietary patterns. Greater fruit intake was consistently associated with higher trefoil factor 1 and dual specificity protein phosphatase 13 isoform A and lower ectonucleoside triphosphate diphosphohydrolase 6 in all dietary patterns. Greater vegetable intake was consistently associated with lower inhibin β A chain. Higher component scores for red meat intake, which signify lower red meat intake, were associated with higher matrix-remodeling-associated protein 8 (MXRA8).

Figure 2.

Plasma proteins associated with components of healthy dietary patterns in the CRIC study. (A) HEI-2020, (B) AHEI-2010, (C) DASH, (D) aMed. Colored swatches represent β-coefficient for significant associations between individual dietary components and plasma proteins. Multivariable linear regression models adjusted for age, sex, race and ethnicity, clinical center, total energy intake, education, income, smoking status, physical activity, BMI, eGFR, 24-hour urine protein, systolic BP, prevalent cardiovascular disease, prevalent diabetes, ACEi/ARB medication use, and lipid-lowering medication use. Statistical significance of dietary component associations was evaluated at P < 0.05/number of dietary components/number of proteins associated with the overall dietary pattern. MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; SFA, saturated fatty acids; SSBs, sugar-sweetened beverages.

Protein Associations with CKD Progression

Over a median follow-up period of 7 years (25th–75th percentile: 3–13 years), 824 CKD progression events occurred. Of 199 proteins associated with at least one dietary pattern, 21 were significantly associated with risk of CKD progression at P < 2.5×10⁻⁴ (=0.05/199; Table 3). Higher relative abundance of all 21 proteins was associated with higher risk of CKD progression. Accounting for the competing risk of death attenuated the association between some proteins and CKD progression, particularly for follistatin-related protein 3 (FSTL3; Supplemental Table 5).

Table 3.

Plasma protein associations with dietary patterns, CKD progression, and all-cause mortality

Entrez Gene Symbol	Protein Name	HEI-2020	AHEI-2010	DASH	aMed	CKD Progression HR (95% CI)	All-Cause Mortality HR (95% CI)
FSTL3	Follistatin-related protein 3a				↓	5.08 (3.71 to 6.95)	1.97 (1.47 to 2.64)
ASGR1	Asialoglycoprotein receptor 1a	↓	↓	↓	↓	2.02 (1.65 to 2.46)	1.51 (1.26 to 1.81)
ART3	Ecto-ADP-ribosyltransferase 3a			↑		1.92 (1.59 to 2.32)	0.61 (0.51 to 0.73)
SLITRK4	SLIT and NTRK-like protein 4			↑		1.78 (1.49 to 2.14)
RGMB	RGM domain family member B	↑	↑	↑		2.71 (1.97 to 3.73)
PTGDS	Prostaglandin-H2 D-isomerase			↑		2.04 (1.62 to 2.58)
GLRX	Glutaredoxin-1a	↓	↓	↓	↓	1.43 (1.27 to 1.62)
AZGP1	Zinc-α-2-glycoproteina		↑			2.02 (1.56 to 2.62)
CADM2	Cell adhesion molecule 2			↑		1.62 (1.35 to 1.93)
SOD3	Extracellular superoxide dismutase [Cu-Zn]a			↑		1.42 (1.24 to 1.62)
GUCA2B	Guanylate cyclase activator 2B			↑		1.34 (1.20 to 1.51)
EPHA1	Ephrin type A receptor 1a		↓	↓		1.42 (1.23 to 1.63)
RTN4R	Reticulon-4 receptor	↓	↓	↓		1.58 (1.30 to 1.91)
PCDH10	Protocadherin-10			↑		1.49 (1.26 to 1.77)
VASN	Vasorin			↑	↑	2.01 (1.48 to 2.72)
MAGI2	Membrane-associated guanylate kinase, WW and PDZ domain-containing protein 2	↓		↓		1.47 (1.24 to 1.73)	1.42 (1.21 to 1.66)
ADAMTSL1	ADAMTS-like protein 1			↑		1.53 (1.26 to 1.87)
IL18R1	IL-18 receptor 1a	↓	↓	↓	↓	1.41 (1.20 to 1.65)	1.37 (1.20 to 1.57)
FAM20A	Pseudokinase FAM20A			↓		1.62 (1.29 to 2.05)
CRHBP	Corticotropin-releasing factor-binding protein			↓		1.66 (1.30 to 2.11)
CNTFR	Ciliary neurotrophic factor receptor subunit α			↑		1.54 (1.23 to 1.94)
EGFR	Epidermal growth factor receptorb	↑	↑	↑			0.29 (0.21 to 0.41)
THBS2	Thrombospondin-2b			↓			1.34 (1.22 to 1.47)
CSF1	Macrophage colony-stimulating factor 1b	↓	↓				1.66 (1.40 to 1.95)
CTSD	Cathepsin Db			↓			1.49 (1.30 to 1.70)
ADAMTSL2	ADAMTS-like protein 2			↓			1.54 (1.33 to 1.79)
SMEK1	Serine/threonine-protein phosphatase 4 regulatory subunit 3Ab		↓				1.43 (1.26 to 1.63)
CLEC3B	Tetranectinb			↑			0.50 (0.39 to 0.65)
ANTXR2	Anthrax toxin receptor 2b	↑	↑	↑	↑		0.69 (0.59 to 0.79)
GHR	Growth hormone receptorb			↓			0.70 (0.61 to 0.81)
IGFALS	Insulin-like growth factor-binding protein complex acid labile subunitb			↑			0.75 (0.67 to 0.84)
DNAJB9	DnaJ homolog subfamily B member 9b	↓	↓	↓			1.48 (1.26 to 1.74)
ASPRV1	Retroviral-like aspartic protease 1b			↑			0.70 (0.60 to 0.81)
MYL6B	Myosin light chain 6Bb			↑			1.29 (1.16 to 1.44)
HTRA1	Serine protease HTRA1	↓	↓	↓	↓		1.57 (1.29 to 1.91)
HPGDS	Hematopoietic prostaglandin D synthaseb	↑					0.72 (0.62 to 0.84)
SIGLEC12	Sialic acid-binding Ig-like lectin 12b		↓	↓			0.72 (0.62 to 0.84)
CECR1	Adenosine deaminase CECR1		↓				1.25 (1.12 to 1.40)
MXRA8	Matrix-remodeling-associated protein 8	↑	↑	↑	↑		0.60 (0.47 to 0.78)
CCDC126	Coiled-coil domain-containing protein 126		↑	↑			0.77 (0.67 to 0.88)
HS6ST2	Heparan-sulfate 6-O-sulfotransferase 2			↑			1.32 (1.15 to 1.53)
TCN1	Transcobalamin-1	↓	↓	↓	↓		1.18 (1.08 to 1.28)
CHI3L1	Chitinase-3-like protein 1		↓	↓			1.12 (1.05 to 1.18)
CD5L	CD5 antigen-like		↓	↓			1.23 (1.10 to 1.36)
NEGR1	Neuronal growth regulator 1	↑	↑	↑			0.57 (0.42 to 0.76)
CDH3	Cadherin-3			↓			0.71 (0.59 to 0.85)

Listed proteins were significantly associated with at least one dietary pattern (false discovery rate-adjusted P < 0.05) and CKD progression or all-cause mortality (P < 2.5×10⁻⁴=0.05/199). Arrows indicate direction of diet-protein association as positive (↑) or inverse (↓). AHEI, alternative healthy eating index; aMed, alternate mediterranean Diet; CI, confidence interval; DASH, dietary approaches to stop hypertension; HEI, healthy eating index; HR, hazard ratio; RGM, repulsive guidance molecule.

Potentially unclear. Suggest repeat note for all-cause mortality, e.g.,

^aSymbol identifies proteins selected in elastic net regression as predictors of CKD progression.

^bSymbol identifies proteins selected in elastic net regression as predictors of all-cause mortality.

Elastic net regression models consistently selected eight proteins associated with CKD progression: FSTL3, glutaredoxin-1 (GLRX), asialoglycoprotein receptor 1 (ASGR1), extracellular SOD [Cu-Zn] (SOD3), zinc-α-2-glycoprotein, IL-18 receptor 1 (IL18R1), ecto-ADP-ribosyltransferase 3, and ephrin type-A receptor 1. Inclusion of all eight proteins along with covariates significantly improved prediction of CKD progression (C-statistic: 0.853), compared with covariates only (C-statistic: 0.843; difference in C-statistics 0.010, 95% confidence interval: 0.004 to 0.015; P = 3.19×10⁻⁴).

Protein Associations with Mortality

Over 14 years median follow-up (25th–75th percentile: 8–16 years), 918 deaths occurred. Of 199 proteins associated with at least one dietary pattern, 30 were associated with mortality at P < 2.5×10⁻⁴ (=0.05/199). Higher relative abundance of 17 proteins was associated with higher risk of mortality; the remaining 13 were inversely associated with mortality (Table 3).

Elastic net regression consistently selected 14 proteins associated with mortality (Table 3). Inclusion of all 14 proteins significantly improved prediction of mortality (C-statistic: 0.787) over a covariates-only model (C-statistic: 0.769; difference in C-statistics 0.019, 95% confidence interval: 0.012 to 0.025; P = 7.25×10⁻⁹).

Subgroup Analyses

Associations between several proteins and CKD progression statistically significantly differed by diabetes and proteinuria but not by eGFR or sex (Supplemental Table 6). ASGR1, FSTL3, and repulsive guidance molecule domain family member B were more strongly associated among people with diabetes than those without diabetes. IL18R1 was associated with CKD progression among those with <1.5 g/d urinary protein.

Associations between proteins and all-cause mortality did not significantly differ by sex, eGFR, diabetes, or proteinuria (Supplemental Table 7).

Discussion

In a cohort of US adults with CKD, we identified 199 plasma proteins that were associated with adherence to at least one healthy dietary pattern. Many of these proteins (41%) were associated with more than one dietary pattern, and 18 were associated with all four dietary patterns. Diet-related proteins were mapped to various pathways involved in cell adhesion, AA metabolism, retinol metabolism, lysosome activity, ferroptosis, xenobiotic metabolism, and central carbon metabolism in cancer. Of the 199 diet-related proteins, 21 were prospectively associated with CKD progression and 30 were associated with all-cause mortality. Elastic net regression identified a subset of eight diet-related proteins that were consistently associated with CKD progression and 14 that were consistently associated with mortality. These may represent proteins on biologic pathways linking diet to clinical outcomes in CKD or could be markers of progressive disease that may be modifiable by diet.

Proteomic correlates of healthy dietary patterns have been identified in general population cohorts. Du et al. used the same SomaScan platform as our study to identify 282 proteins associated with dietary patterns in the Atherosclerosis Risk in Communities study, including 96 proteins that were associated with dietary patterns in our study.¹⁰ Nearly half of the proteins that were associated with HEI-2020 (35/76=46%) and DASH (84/173=49%) in our study, one third of those associated with AHEI-2010 (18/59=31%), and 4 of 33 (12%) associated with aMed replicated observed associations in Atherosclerosis Risk in Communities. Proteomic associations with dietary patterns were also investigated in the Framingham Offspring Study, Jackson Heart Study, and the Singapore Multi-Ethnic Cohort using earlier SomaScan platforms.^8,33 Comparing these studies with ours, some of the most consistent diet-protein associations were for epidermal growth factor receptor (EGFR) and contactin-4.^8,10,33 The association between EGFR with a DASH dietary pattern was also supported by two rigorously controlled feeding trials.⁹ EGFR is a tyrosine kinase involved in cell signaling pathways that promote cell growth, proliferation, differentiation, and apoptosis.³⁴ EGFR is expressed in various cell types throughout the kidney and is involved in electrolyte handling and tissue repair.³⁵ Sustained overactivation of the EGFR pathway has been linked to renal fibrosis.^34,35 Soluble EGFR may function as a decoy receptor, serving to regulate the activity of membrane-bound EGFR.³⁶ Higher plasma EGFR was associated with lower risk of CKD progression and mortality in our study. CNTN4 is a member of the Ig superfamily of neuronal cell adhesion molecules that is primarily expressed in the brain but also expressed by the thyroid, small intestine, testes, and uterus.³⁷ Although the role of this protein outside the central nervous system is not well elucidated, the CNTN4 gene has been associated with variation in serum uric acid³⁸ and thromboxane A₂ formation.³⁹ We also replicated its association with greater whole grain intake,³³ suggesting that this dietary component may underlie the dietary patterns' association with CNTN4.

Kynureninase (KYNU) was inversely and hepatocyte growth factor receptor (MET) was positively associated with AHEI-2010, HEI, and DASH dietary patterns in prior studies but not replicated in ours.^8,10,33 KYNU is an enzyme in the tryptophan metabolic pathway. In CKD, tryptophan conversion to kyurenine increases and kyurenine metabolites accumulate due to decreased renal excretion.⁴⁰ In a rat model of CKD, KYNU activity decreased proportionally with reduced kidney function.⁴¹ MET is a tyrosine kinase receptor for hepatocyte growth factor that promotes cell growth, regeneration, and survival. MET and its ligand, hepatocyte growth factor, increase after AKI to promote regeneration and repair.⁴² In CKD, an initial increase in hepatocyte growth factor is followed by a decline as disease progresses.⁴³ Changes in KYNU and MET resulting from kidney disease may have obscured their associations with diet in our study relative to previous studies conducted in generally healthy populations.

Our study also identified diet-related proteins that were associated with risk of CKD progression and all-cause mortality. Three proteins—ASGR1, GLRX, and IL18R1—were inversely associated with all four dietary patterns, associated with a higher risk of CKD progression, and were among the 8 proteins selected by elastic net. Asialoglycoprotein receptor and IL18R1 were also associated with higher risk of death. Higher ASGR1 has previously been associated with eGFR decline⁴⁴ and albuminuria in CKD.⁴⁵ Inhibition of ASGR1 promotes cholesterol efflux, reduces blood lipid levels, and may improve insulin sensitivity.^46,47 It is plausible that lower ASGR1 associated with healthy diets could protect against CKD progression and death by reducing these cardiometabolic risk factors. GLRX maintains redox balance by catalyzing reversible glutathionylation and deglutathionylation reactions.⁴⁸ Higher GLRX may signify greater oxidative stress, prognostic of CKD progression.⁴⁹ Lower GLRX associated with healthy dietary patterns may be related to higher intake of dietary antioxidants.⁵⁰ An association between GLRX and healthy dietary patterns has not previously been reported and may be evident in this cohort due to elevated oxidative stress associated with CKD.⁵¹ IL18R1 is a receptor for cytokine IL-18, a product of the NLRP3 inflammasome, which is implicated in CKD pathogenesis.⁵² Higher plasma IL-18 and IL18R1 have been associated with more severe disease and poor prognosis in CKD.^53–57 Fatty acids may modulate NLRP3 inflammasome activity, with saturated fats promoting and unsaturated fats inhibiting activation.⁵⁸ Better fat composition of healthy dietary patterns could underlie lower IL18R1, consistent with our observed association between higher intake of nuts (high in unsaturated fats) with lower IL18R1. Lower levels of these proteins associated with healthy dietary patterns may reflect a better cardiometabolic risk profile (ASGR1), antioxidant (GLRX), and anti-inflammatory (IL18R1) mechanisms through which diet improves kidney disease prognosis.

In addition to ASGR1 and IL18R1, HTRA1 and TCN1 were inversely associated with all diets and associated with higher risk of death. Both proteins' associations with diet were previously observed in healthy adults.¹⁰ HTRA1 is a protease that targets extracellular matrix proteins, insulin-like growth factors, and TGF-ß family proteins.⁵⁹ It is abundantly expressed in vascular tissues and regulates angiogenesis and smooth muscle cell phenotype.⁶⁰ While understood to play a protective role in vascular homeostasis,⁶⁰ higher levels have been associated with worse glycemic control and higher serum triglycerides,⁶¹ possibly representing a response to vascular injury or inflammation and thus serving as a biomarker of disease severity. TCN1, also known as haptocorrin, carries vitamin B12 in plasma.⁶² Elevated TCN1 has been observed in CKD,⁶³ and high plasma vitamin B12 has been associated with all-cause mortality,⁶⁴ although the mechanisms underlying these associations are not well understood. MXRA8 and ANTXR2 were positively associated with all dietary patterns and associated with lower risk of all-cause mortality. Increased expression of MXRA8, also known as dual Ig domain-containing cell adhesion molecule, suppressed inflammation in a cell model of AKI, whereas MXRA8 knockdown increased inflammation and apoptosis.⁶⁵ Higher red and processed meat component scores, representing lower meat intakes, were associated with higher MXRA8. Similarly, higher MXRA8 was previously associated with plant-based dietary patterns, providing concordant evidence that lower meat intake could explain the association.⁶⁶ ANTXR2, also known as capillary morphogenesis gene 2, helps to regulate extracellular matrix degradation and remodeling and has been implicated in BP regulation.^67,68 Higher ANTXR2 was inversely associated with risk of secondary cardiovascular events among people with CKD.⁶⁹

Strengths of our study include, first, use of a large-scale untargeted assay to identify thousands of proteins, which optimized discovery of novel protein associations. Second, the diverse study cohort, including men and women in varying stages of CKD with approximately 50% Black participants by design, enhances the generalizability of our findings. Third, we characterized healthy dietary patterns according to four definitions, with secondary assessment of specific dietary components, to help discern influential foods or nutrients that may explain associations. Finally, outcomes were thoroughly ascertained based on eGFR, calculated from routinely measured creatinine, and linkage to external records, which strengthens the validity of our results.

We also acknowledge limitations of our study. First, declining kidney function may affect concentrations of plasma proteins. While we adjusted for eGFR, differences in baseline kidney function that are not captured by eGFR may contribute to observed associations.⁷⁰ Second, proteins were only measured at a single time point, precluding assessment of temporal changes in protein levels in response to diet or CKD progression. Third, the CKD progression was defined only by eGFR and did not consider albuminuria, which can be an earlier marker of progression.⁷¹ Fourth, dietary intake was not assessed at the same time point when plasma for proteomic measurements was collected. However, we averaged available dietary assessments from 1 year prior and 1 year after proteomic measurement, which may be a reasonable estimate of participants' usual intakes. Fifth, we scored adherence to healthy dietary patterns using self-reported dietary intake, which is prone to random and systematic error. However, replication of diet-protein associations from other studies increases confidence in our findings. Finally, though our multivariable models adjusted for several sociodemographic, behavioral, and clinical confounders, residual confounding may influence observed associations. Socioeconomic status, in particular, affects the ability to afford, access, and prepare healthy foods⁷² and has been linked to a distinct proteomic profile, which could confound observed associations with diet.⁷³ Controlled-feeding interventions, where diet is fully elucidated, randomly assigned, and temporality and is established by preintervention and postintervention measures, would provide greater assurance that diet is causally linked to protein levels.

In conclusion, our study identified proteins related to healthy dietary patterns that are associated with CKD progression and all-cause mortality in US adults with CKD. Diet-related proteins that were associated with CKD progression and mortality are involved in regulation of blood lipids, insulin sensitivity, vascular homeostasis, inflammation, and oxidative stress, which may represent biologic mechanisms through which diet improves kidney disease prognosis. If further research confirms the causality of diet-protein associations, proteins may be valuable to assess adherence to recommended diets or monitor response to dietary interventions in CKD.

Supplementary Material

cjasn-21-553-s001.pdf ^{(1.9MB, pdf)}

cjasn-21-553-s002.xlsx ^{(183.3KB, xlsx)}

cjasn-21-553-s003.pdf ^{(438.8KB, pdf)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/CJN/C546.

Author Contributions

Conceptualization: Casey M. Rebholz, Valerie K Sullivan.

Formal analysis: Jingsha Chen.

Funding acquisition: Casey M. Rebholz.

Methodology: Casey M. Rebholz, Valerie K Sullivan.

Project administration: Casey M. Rebholz.

Supervision: Casey M. Rebholz.

Visualization: Jingsha Chen.

Writing – original draft: Valerie K Sullivan.

Writing – review & editing: Lawrence J Appel, Jing Chen, Jingsha Chen, Mirela Dobre, Jiang He, Paul L Kimmel, Nishigandha Pradhan, Panduranga Rao, Casey M. Rebholz, Ana C. Ricardo, Hernan Rincon-Choles, Sarah Schrauben.

Funding

C.M. Rebholz: National Heart, Lung, and Blood Institute (R01 HL153178), Johns Hopkins University (U54 DK137331). This study was supported by National Institute of Diabetes and Digestive and Kidney Diseases (U01DK108809, U01DK060990, U01DK060984, U01DK061022, U01DK061021, U01DK061028, U01DK060980, U01DK060963, U01DK060902 and U24DK060990). National Center for Advancing Translational Sciences (UL1TR000003, UL1TR000439, 1UM1TR004926). Johns Hopkins University (UL1 TR-000424). University of Maryland (GCRC M01 RR-16500). Clinical and Translational Science Collaborative of Cleveland, School of Medicine, Case Western Reserve University. Michigan Institute for Clinical and Health Research (UL1TR000433). University of Illinois at Chicago (CTSA UL1RR029879). Tulane University (P20 GM109036). Kaiser Permanente (NIH/NCRR UCSF-CTSI UL1 RR-024131). School of Medicine, University of New Mexico (NM R01DK119199).

Declarative Statements

This study includes clinical experimentation and received Institutional Review Board or Ethics Committee approval. The need to obtain informed patient consent was waived. This study includes clinical experimentation and complies with the Declaration of Helsinki.

Data Availability Statements

Original data generated for the study are or will be made available in a repository subject to controlled access. Data Type: Aggregated Data. Repository Name: NIDDK Repository. Linkable Citation: https://doi.org/10.58020/6dxf-ed78. Reason for Restriction: Data from the CRIC study are available for request at the NIDDK Central Repository (NIDDK-CR) Website (https://doi.org/10.58020/6dxf-ed78) and proteomic data will available for request at dbGAP (Study Accession: phs000524).

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/CJN/C547 and http://links.lww.com/CJN/C548.

Supplemental Figure 1. Derivation of study sample.

Supplemental Figure 2. Kernel density plots of dietary pattern scores at baseline and year 2.

Supplemental Table 1. Participant characteristics according to extreme quintiles of dietary pattern scores.

Supplemental Table 2. Associations between dietary pattern scores and plasma proteins.

Supplemental Table 3. Subgroup associations between dietary patterns and plasma proteins.

Supplemental Table 4. Pathway overrepresentation analysis for 199 proteins associated with≥1 dietary pattern.

Supplemental Table 5. Association between diet-related proteins and CKD progression, accounting for competing risk of death.

Supplemental Table 6. Subgroup associations between diet-related proteins and CKD progression.

Supplemental Table 7. Subgroup associations between diet-related proteins and all-cause mortality.