Proteins Associated with Risk of Kidney Function Decline in the General Population

Significance Statement

Proteomic profiling may allow identification of plasma proteins associated with subsequent changes in kidney function, elucidating biologic processes that underlie CKD. The authors used large-scale proteomic profiling to evaluate the association of 4877 plasma proteins with the development of adverse kidney outcomes in a cohort of 9406 middle-aged adults with a median follow-up of 14.4 years, and, in a subset of 4378 adults at a later time point, with a median follow-up of 4.4 years. They found 13 proteins for which higher levels were associated with greater kidney risk at both time periods, and 12 replicated in at least one external cohort. Although most proteins appeared to be markers, rather than causes, of kidney damage, genetic evidence suggested a causal role for lectin mannose-binding 2 protein (LMAN2).

Abstract

Background

Proteomic profiling may allow identification of plasma proteins that associate with subsequent changesin kidney function, elucidating biologic processes underlying the development and progression of CKD.

Methods

We quantified the association between 4877 plasma proteins and a composite outcome of ESKD or decline in eGFR by ≥50% among 9406 participants in the Atherosclerosis Risk in Communities (ARIC) Study (visit 3; mean age, 60 years) who were followed for a median of 14.4 years. We performed separate analyses for these proteins in a subset of 4378 participants (visit 5), who were followed at a later time point, for a median of 4.4 years. For validation, we evaluated proteins with significant associations (false discovery rate <5%) in both time periods in 3249 participants in the Chronic Renal Insufficiency Cohort (CRIC) and 703 participants in the African American Study of Kidney Disease and Hypertension (AASK). We also compared the genetic determinants of protein levels with those from a meta-analysis genome-wide association study of eGFR.

Results

In models adjusted for multiple covariates, including baseline eGFR and albuminuria, we identified 13 distinct proteins that were significantly associated with the composite end point in both time periods, including TNF receptor superfamily members 1A and 1B, trefoil factor 3, and β-trace protein. Of these proteins, 12 were also significantly associated in CRIC, and nine were significantly associated in AASK. Higher levels of each protein associated with higher risk of 50% eGFR decline or ESKD. We found genetic evidence for a causal role for one protein, lectin mannose-binding 2 protein (LMAN2).

Conclusions

Large-scale proteomic analysis identified both known and novel proteomic risk factors for eGFR decline.

The application of proteomics to the study of kidney disease can inform pathophysiology and identify novel biomarkers for risk assessment and management. Different approaches may yield distinct insights. Multivariable-adjusted observational analyses may identify biomarkers that herald higher risk of adverse outcomes, but do not necessarily cause the outcome. Additional evaluation of the genetic determinants of protein levels and kidney phenotypes may help distinguish those proteins in causal pathways.^1,2

Previous work in the genetics and proteomics of kidney disease have suggested that inflammatory components may be central in CKD.^3,4 The TNF receptor superfamily members 1a (TNFRSF1A) and TNFRSF1B have been associated with progression of kidney disease in several cohorts with diabetes, suggesting possible activation of the TNF signaling pathway.⁵ In two cohorts of patients with type 1 and 2 diabetes, investigators developed a kidney risk inflammatory signature comprising 17 circulating proteins involved in inflammation that were associated with subsequent development of ESKD.⁶ Others have applied a similar approach to urine proteomics, developing a composite of 273 urine peptides to predict early CKD, with many of the peptide components representing proteins involved in inflammation and tissue repair.⁷

The aims of this study were to uncover associations of proteins, eGFR decline, and ESKD in a population-based cohort with long-term follow-up; to replicate the findings in external cohorts also followed for eGFR decline and ESKD; and to provide insight into potential causality of identified proteins using genetic association studies. With the hypothesis that large-scale proteomic approaches can identify novel and potential causal markers of kidney disease, we evaluated the association between 4877 proteins quantified using a modified aptamer technology (SomaScan) and the composite end point of ≥50% decline in eGFR or ESKD.

Methods

Study Population

The study population was drawn from the Atherosclerosis Risk in Communities (ARIC) Study, an ongoing cohort of individuals recruited from four US communities: suburbs of Minneapolis, Minnesota; Jackson, Mississippi; Forsyth County, North Carolina; and Washington County, Maryland.⁸ Enrollment occurred between 1987 and 1989, with subsequent visits in 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), 2011–2013 (visit 5), 2016–2017 (visit 6), and 2018–2019 (visit 7). Participants who attended visits 3 and 4, were free of ESKD, had nonmissing eGFR and albuminuria measures, and agreed to participate in cardiovascular research were included (n=9406; termed “visit 3 cohort”). For the purpose of confirmation, the subset of participants who attended visit 5 and met the same inclusion criteria were studied (n=4378; termed “visit 5 cohort”) (Supplemental Figure 1). Study protocols were approved at the institutional review board of each participating center.

For validation, we evaluated associations in two external cohorts: the Chronic Renal Insufficiency Cohort (CRIC) and the African American Study of Kidney Disease and Hypertension (AASK). CRIC is a study of patients with CKD, recruited between 2003 and 2008, followed for >14 years for clinical events.⁹ To be included in the validation study, CRIC participants had to be event-free at the 1-year visit (considered “baseline” for this study, because proteomic profiling was performed on plasma samples drawn at that visit), with nonmissing covariates and available proteomic data (n=3249). For AASK, a study of Black patients with CKD attributed to hypertension, we included participants with serum proteomic profiling available at the baseline visit (n=703) and evaluated protein associations with doubling of creatinine or ESKD.¹⁰

Proteomic Profiling

The relative concentrations of 4877 plasma proteins or protein complexes were quantified using a Slow Off-rate Modified Aptamer–based capture array and plasma collected at visits 3 and 5. In brief, 250 µl of plasma stored at −80°C was shipped to SomaLogic (Boulder, CO) for identification and quantification of the low-abundance plasma proteins by SomaScan, a technology that uses easily quantifiable, chemically modified oligonucleotides as a binding reagent for proteins and protein complexes.¹¹ Samples from visits 5 and 3 were run as separate batches in 2018 and 2019, respectively, using the V4 platform. Quality control was run separately for each visit: 422 samples from visit 3 and 197 samples from visit 5 were run in duplicate, in addition to standard SomaLogic reference standards run on every plate. The median intra-assay coefficient of variation for visit 3 was 8.3%, and was 1.2% for visit 5. The protein-wise, between-batch reliability for 26 pooled samples run in both 2018 and 2019 was 0.85, with a coefficient of variation of 8.0%. Quality control on all proteins is listed in Supplemental Table 1. For the purposes of analyses, all proteins, reported in relative fluorescence units, were log₂ transformed because of skewed distributions, and values outside of 5 SDs on the log₂ scale were winsorized. Similar procedures, also using SomaScan, were applied in the two external validation cohorts. In CRIC, the same platform (V4) was run on plasma samples in 2019; in AASK, the expanded V4.1 platform was run on serum samples in 2021.

Outcomes

The study outcome was a composite of decline in eGFR by ≥50% or ESKD. In ARIC, eGFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration equation that incorporated both serum creatinine and cystatin C. Creatinine was measured using the Roche enzymatic method at visits 3, 5, and 6, and by the modified kinetic Jaffe method at visit 4. Cystatin C was measured using the Roche Cobas 6000 chemistry analyzer. For the visit 3 cohort, time at risk began at visit 4 (when urine albumin-creatinine ratio [ACR] was measured). Thus, the ≥50% decline was assessed at visit 5 in relation to the eGFR measured at visit 4; events were recorded if they occurred before the date of visit 5 or death. For the visit 5 cohort, time at risk began at visit 5 and continued until the first among study outcome, death, or December 31, 2018. ESKD was assessed continuously through linkage to the United States Renal Data System (USRDS). In CRIC, similar procedures were used, except that the eGFR equation incorporating both creatinine and cystatin C was developed within CRIC itself, as used in previous studies, and time at risk began at year 1, the first available visit with proteomic profiling. In AASK, the primary outcome was doubling of serum creatinine (equivalent to an approximately 57% decline in eGFR) or ESKD.

Covariate Definitions

For each cohort, covariates were assessed concomitant with the proteomic profiling. In ARIC, these included age, sex, self-reported race/ethnicity, study center, systolic BP, antihypertensive medication use, diabetes, history of atherosclerotic cardiovascular disease (ASCVD), smoking history (current, former, never), eGFR, and HDL cholesterol level. The one exception was log-transformed ACR for the visit 3 cohort, which was first measured at visit 4. Systolic BP was estimated using the average of the second and third measurements with a random-zero sphygmomanometer after 5 minutes of quiet rest. Antihypertensive medications were assessed during the study visit, when study participants were asked to bring with them all medications taken in the last 2 weeks. History of ASCVD and smoking were self-reported. HDL was assessed using the enzymatic method after precipitation with dextran sulfate–magnesium. Albumin in urine samples was quantified using a nephelometric method with a Dade Behring BN100 or on the Beckman Image Nephelometer.

Statistical Analyses

Covariates were summarized for each cohort using mean and SD for variables with normal distributions. Cox proportional hazards regression was used to relate each protein to risk of the study outcome. Sequential adjustment models were used: model 0 adjusted for age, sex, and race/center; model 1 additionally adjusted for systolic BP, antihypertensive medications, diabetes, history of ASCVD, smoking, eGFR based on creatinine and cystatin C (eGFR_crcys), and HDL levels; and model 2 additionally adjusted for log-transformed ACR. A false discovery rate of 5% was applied to the P value to determine statistical significance.

For evaluation of consistency at middle and older age, analyses were performed separately for all proteins in the ARIC visit 3 and 5 cohorts. We evaluated correlations between the proteins that had statistically significant associations with the outcome in both cohorts (adjusted for model 2 covariates), and the difference in c-statistic over a fully adjusted model (model 2 with all 13 significant proteins versus model 2 without proteins). We also built a multiprotein model using forward stepwise regression in each cohort, including model 2 covariates as mandatory and the 13 significant proteins as optional; tested the association between protein and outcome after additionally adjusting for the known kidney biomarkers kidney injury molecule-1 (KIM-1), trefoil factor 3 (TFF3), and β-trace protein (PTGDS); and evaluated whether there was an interaction between each significant protein and eGFR with the outcome by including the product term.

For evaluation in the validation cohorts, the 13 proteins (15 aptamers, with two proteins measured by two different aptamers) with significant associations in both visits 3 and 5 were evaluated using model 2 covariates for adjustment. Statistical significance was determined using a Bonferroni correction (P<0.05/13). In both CRIC and AASK, urine protein-creatinine ratios were used in place of ACR. In AASK, no participants had diabetes and all were using antihypertensive medications; thus, these variables were not included as covariates. In addition, measured GFR rather than eGFR was used for adjustment.

Genetic Association Studies

To provide orthogonal evidence of aptamer specificity for the target protein, we performed genome-wide association studies (GWAS) to assess for protein quantitative trait loci (pQTL) in the cis-region of the corresponding gene, defined as the transcription start site ±1 Mb. We also assessed whether genetic models of proteins of interest were associated with GWAS associations with eGFR. Genetic analyses were performed in White participants attending visit 3 (n=7213) to be consistent with existing summary statistics for eGFR performed by the CKD Genetics Consortium. Genotyping procedures in ARIC have previously been described^12–14; briefly, genotyping was performed on the Affymetrix 6.0 DNA microarray with exclusion of single-nucleotide polymorphisms (SNPs) with call rates <95%, Hardy–Weinberg equilibrium P<0.001, or minor allele frequencies <1%. Data were then imputed to a common set of SNPs using Trans-Omics for Precision Medicine (Freeze 5b).¹³ GWAS were performed using the inverse-rank normalized residuals of the log-transformed protein regressed on age, sex, the first ten genetic principal components, and the first ten protein principal components, with the latter included to remove variation due to latent factors, such as technical variance in the proteomic data.¹⁵ To build genetic models of proteins, individual SNPs associated with the protein in the cis-gene region (P<0.01) were combined using elastic net regression.^16(preprint) The model was then applied to published summary statistics from the CKD Genetics Consortium of a GWAS on eGFR.¹⁷ Statistical significance was determined using a Bonferroni correction. We also evaluated the index SNPs within the cis-region and used GeneAtlas,¹⁸ PhenoScanner,^19,20 and SNiPA²¹ to evaluate published associations with expression QTLs (eQTLs), pQTLs, metabolite QTLs, and phenotypes. Results were restricted to those with a P value of <5 × 10⁻⁵.

Gene Expression

Gene expression analyses for the genes coding for proteins associated with kidney function decline were examined using data from the Genotype-Tissue Expression (GTEx) project.^22,23 Using the participants in Genotype-Tissue Expression V8, we evaluated eQTLs for index SNPs for the proteins that had support for a causal association from the genetic instrument studies. Associations were evaluated in whole blood (n=670).

Results

The 9406 participants included in the ARIC visit 3 cohort had a mean (SD) age of 60.0 (5.7) years, and 54.6% were women (Table 1). Mean (SD) baseline eGFR_crcys was 85 (15.3) ml/min per 1.73 m², median ACR was 4 mg/g, and 14.1% of participants had diabetes. The 4378 participants in the visit 5 cohort had a mean (SD) age of 75.7 (5.2) years, 54.9% were women, their mean (SD) baseline eGFR_crcys was 65 (17.8) ml/min per 1.73 m², with a median ACR of 11 mg/g, and 30.0% of participants had diabetes. In the visit 3 cohort, there were 595 outcome events of 50% eGFR decline or ESKD over a median follow-up of 14.4 years; in the visit 5 cohort, there were 69 events over a median follow-up of 4.4 years. The CRIC validation cohort included 3249 participants with a mean (SD) age of 59.1 (10.7) years, 44.5% were women, and the mean (SD) baseline eGFR_crcys was 43 (16.8) ml/min per 1.73 m². The median urine protein-creatinine ratio was 13 mg/g, and 49% of participants had diabetes. There were 1171 outcome events occurring over a median follow-up of 6.0 years. The AASK validation cohort included 703 participants with a mean (SD) age of 54.5 (10.7) years, 38.5% were women, and their mean (SD) baseline measured GFR was 42.6 (15.0) ml/min per 1.73 m². Median urine protein-creatinine ratio was 81 mg/g, and 49% of participants had diabetes. There were 274 events occurring over a median of 7.5 years of follow-up.

Table 1.

Characteristics of visit 3 (1993–1995) and visit 5 (2011–2013) participants of the ARIC Study at visit 3 and 5, respectively, and of the two external validation cohorts

Characteristic	ARIC Visit 3 Cohort	ARIC Visit 5 Cohorta	CRIC Validation Cohort	AASK Validation Cohort
N	9406	4378	3249	703
Age (yr), mean (SD)	60.0 (5.7)	75.7 (5.2)	59.1 (10.7)	54.5 (10.7)
Female, n (%)	5133 (54.6)	2402 (54.9)	1447 (44.5)	271 (38.5)
Self-reported Black race, n (%)	1748 (18.6)	823 (18.8)	1344 (41.4)	703 (100)
Center, n (%)
Forsyth	2392 (25.4)	1064 (24.3)	NA	NA
Jackson	1543 (16.4)	756 (17.3)	NA	NA
Minneapolis	2767 (29.4)	1331 (30.4)	NA	NA
Washington County	2704 (28.8)	1227 (28.0)	NA	NA
Antihypertension medication, n (%)	2820 (30.0)	2956 (67.5)	2974 (91.9)	703 (100)
Diabetes, n (%)	1323 (14.1)	1314 (30.0)	1591 (49.0)	0 (0)
History of ASCVD, n (%)	739 (7.9)	811 (18.5)	750 (23.1)	358 (50.9)
Current smoker, n (%)	1553 (16.5)	242 (5.5)	396 (12.2)	205 (29.2)
eGFRcrcys (ml/min per 1.73 m2), mean (SD)b	85.2 (15.3)	65.0 (17.8)	42.6 (16.8)	45.7 (14.3)
Albuminuria (mg/g), median (IQR)c	4 (2–8)	11 (6–23)	13 (5–65)	81 (28–373)
Systolic BP (mm Hg), mean (SD)	124.0 (18.5)	130.5 (18.1)	126.8 (21.6)	151.1 (24.8)
Cholesterol (mmol/L), mean (SD)	209.4 (37.8)	181.0 (42.1)	181.8 (43.6)	211.9 (45.7)
HDL cholesterol (mmol/L), mean (SD)	52.7 (18.3)	51.9 (14.0)	48.5 (15.8)	48.3 (16.0)
Body mass index (kg/m2), mean (SD)	28.5 (5.4)	28.8 (5.6)	32.15 (7.72)	30.5 (6.4)
Follow-up variablesd
Number of events	595	69	1171	274
Median follow-up time, years	14.4	4.4	6.0	7.5

NA, not applicable; IQR, interquartile range.

^aA total of 3663 participants in the visit 5 cohort were also present in the visit 3 cohort.

^bValue in ARIC is calculated using the Chronic Kidney Disease Epidemiology Collaboration equation, value from CRIC is calculated using the CRIC equation, and the value in AASK is the iothalamate-measured GFR.

^cUrine ACR was measured 3 years after the other variables (at visit 4) for the visit 3 cohort. Values in the validation cohorts reflect urine protein-creatinine ratios.

^dFollow-up for the visit 3 cohort ran from visit 4 to visit 5; for the visit 5 cohort, from visit 5 to 6.

In the visit 3 cohort, 64 proteins had statistically significant associations (false discovery rate P<0.05) with the study outcome of eGFR decline by ≥50% or ESKD in the fully adjusted model, with more significant proteins exhibiting a positive rather than negative association (Figure 1A, Supplemental Table 2). Fewer proteins in the visit 5 cohort showed negative associations with the study outcome, and the associations were generally stronger than in the visit 3 cohort (Figure 1B, Supplemental Table 3). A total of 15 aptamers, corresponding to 13 proteins, had significant associations in both cohorts in the fully adjusted model (Table 2). Of these, all exhibited positive associations with the composite end point. Four were known biomarkers of CKD, including TFF3, TNFRSF1A, TNFRSF1B, and PTGDS. Another known biomarker of kidney disease, KIM-1 (HAVCR), was associated with the study outcome in the visit 3 but not visit 5 cohort.

Figure 1.

Proteins associated with eGFR decline of 50% of ESKD in the ARIC visit 3 and visit 5 cohorts. β Coefficients and P values for the association of 4877 proteins with eGFR decline by 50% or ESKD in the (A) visit 3 and (B) visit 5 cohort. Top 20 (ranked by P value) positively and negatively associated proteins reaching statistical significance (false discovery rate [FDR] P<0.05) are labeled on the volcano plots. β Coefficients are per doubling of protein level.

Table 2.

Proteins associated with eGFR decline by 50% or ESKD in both visit 3 and visit 5 ARIC cohorts (FDR P<0.05 in model 2)

Symbol	Protein Name	HR (95% CI)
		Visit 3 Cohort (n=9046)			Visit 5 Cohort (n=4378)
		Model 0	Model 1	Model 2	Model 0	Model 1	Model 2
CLMP	CXADR-like membrane protein	1.16 (1.10 to 1.22)	1.14 (1.07 to 1.21)	1.14 (1.07 to 1.22)	1.53 (1.39 to 1.68)	1.41 (1.20 to 1.64)	1.39 (1.17 to 1.66)
DCTN2	Dynactin subunit 2	1.13 (1.05 to 1.21)	1.14 (1.06 to 1.23)	1.14 (1.06 to 1.23)	1.35 (1.21 to 1.50)	1.41 (1.21 to 1.65)	1.36 (1.15 to 1.60)
DLK2	Protein δ homolog 2	1.18 (1.10 to 1.27)	1.17 (1.09 to 1.26)	1.15 (1.06 to 1.24)	1.97 (1.72 to 2.25)	1.71 (1.40 to 2.08)	1.58 (1.26 to 1.98)
DSC2	Desmocollin-2	1.23 (1.14 to 1.32)	1.16 (1.06 to 1.26)	1.17 (1.08 to 1.27)	1.64 (1.49 to 1.81)	1.35 (1.14 to 1.61)	1.40 (1.17 to 1.69)
FSTL3	Follistatin-related protein 3	1.42 (1.31 to 1.54)	1.22 (1.12 to 1.34)	1.18 (1.07 to 1.29)	3.24 (2.65 to 3.96)	2.51 (1.85 to 3.41)	2.08 (1.50 to 2.89)
LMAN2	Vesicular integral-membrane protein VIP36	1.40 (1.29 to 1.52)	1.29 (1.18 to 1.41)	1.20 (1.10 to 1.31)	3.09 (2.54 to 3.76)	2.38 (1.76 to 3.23)	2.00 (1.47 to 2.73)
NBL1	Neuroblastoma suppressor of tumorigenicity 1	1.29 (1.22 to 1.35)	1.24 (1.17 to 1.32)	1.21 (1.14 to 1.29)	2.33 (2.05 to 2.65)	2.06 (1.71 to 2.47)	1.86 (1.52 to 2.28)
PTGDS	Prostaglandin-H2 D-isomerase	1.27 (1.18 to 1.38)	1.23 (1.13 to 1.35)	1.17 (1.07 to 1.29)	2.62 (2.19 to 3.13)	1.87 (1.43 to 2.43)	1.73 (1.33 to 2.27)
TFF3a	Trefoil factor 3	1.11 (1.02 to 1.20)	1.17 (1.08 to 1.28)	1.19 (1.10 to 1.30)	1.69 (1.51 to 1.89)	1.68 (1.39 to 2.03)	1.63 (1.38 to 1.92)
TFF3a	Trefoil factor 3	1.14 (1.06 to 1.24)	1.20 (1.10 to 1.30)	1.21 (1.11 to 1.32)	1.95 (1.72 to 2.23)	1.57 (1.34 to 1.84)	1.73 (1.42 to 2.11)
TNFRSF1A	TNF receptor superfamily member 1A	1.49 (1.38 to 1.60)	1.28 (1.17 to 1.40)	1.20 (1.09 to 1.31)	3.27 (2.66 to 4.01)	2.67 (1.89 to 3.78)	2.16 (1.51 to 3.09)
TNFRSF1Ba	TNF receptor superfamily member 1B	1.33 (1.25 to 1.42)	1.22 (1.13 to 1.31)	1.19 (1.09 to 1.30)	1.52 (1.40 to 1.65)	1.33 (1.15 to 1.52)	1.36 (1.17 to 1.57)
TNFRSF1Ba	TNF receptor superfamily member 1B	1.46 (1.35 to 1.58)	1.25 (1.14 to 1.36)	1.18 (1.09 to 1.28)	1.58 (1.44 to 1.73)	1.28 (1.12 to 1.47)	1.31 (1.14 to 1.51)
WFDC2	WAP four-disulfide core domain protein 2	1.34 (1.25 to 1.45)	1.26 (1.15 to 1.38)	1.19 (1.08 to 1.31)	2.82 (2.31 to 3.46)	2.02 (1.50 to 2.72)	1.77 (1.29 to 2.43)
ZHX3	Zinc fingers and homeoboxes protein 3	1.25 (1.18 to 1.32)	1.20 (1.12 to 1.29)	1.19 (1.10 to 1.28)	1.31 (1.22 to 1.40)	1.29 (1.16 to 1.44)	1.22 (1.09 to 1.37)

Model 0 was adjusted for age, sex, and race/center; model 1 additionally adjusted for systolic BP, antihypertension medications, diabetes, history of ASCVD, smoking, eGFR_crcys, and HDL levels; and model 2 was additionally adjusted for log-transformed ACR. An FDR of 5% was applied to the P value to determine statistical significance. All hazard ratios are expressed per doubling of protein level. FDR, false discovery rate; HR, hazard ratio.

^aTwo different aptamers separately quantified these proteins.

Correlations between the 13 proteins varied (Figure 2). The proteins identified by two different aptamers were highly correlated (TFF3, 0.97; TNFRSF1B, 0.83); most other correlations were moderately positive. CLMP, DCTN2, TFF3, and ZHX3 were the most distinct, demonstrating smaller correlations with each other and the other proteins. A model of the visit 3 cohort with all 13 proteins slightly improved the prediction of the composite end point, compared with a fully adjusted model (Harrell c-statistic, 0.77 versus 0.76; for difference, P=1.23 × 10⁻⁵). Forward selection, using the fully adjusted model and the visit 3 cohort, selected CLMP, DCTN2, TFF3, ZHX3, and NBL1 as additional variables from the 13 proteins; each retained statistical significance with hazard ratios per doubling ranging from 1.08 to 1.18. All but FSTL3 and WFDC2 remained significant after adjusting for the known kidney biomarkers TFF3, KIM-1, and PTGDS. Most of the top proteins exhibited an interaction with eGFR, such that there was a stronger association between protein and the study outcome in the setting of lower eGFR (Supplemental Table 4).

Figure 2.

Correlations between the 15 proteins associated with 50% decline in eGFR or ESKD and age, eGFR, and log-transformed urine ACR in the visit 3 cohort. eGFR_crcys was measured at visit 3; log-ACR at visit 4.

In the CRIC validation cohort, all but one of the 13 proteins (DCTN2) was significantly associated with the composite end point (P<0.05/13). Hazard ratios were all consistent in direction with those in the ARIC cohorts, with generally similar or stronger magnitudes compared with the visit 5 cohort. In AASK, nine of the proteins replicated (Table 3).

Table 3.

Evaluation of the top proteins in external cohorts: CRIC and AASK

Symbol	Protein Name	CRIC Validation Cohort (n=3249)		AASK Validation Cohort (n=709)
		Model 2, HR (95% CI)a	P Value	Model 2, HR (95% CI)a	P Value
CLMP	CXADR-like membrane protein	1.37 (1.18 to 1.59)	3.54 × 10−5	1.85 (1.20 to 2.87)	5.50 × 10−3
DCTN2	Dynactin subunit 2	1.01 (0.82 to 1.24)	0.93	1.11 (0.61 to 2.00)	0.74
DLK2	Protein δ homolog 2	1.56 (1.34 to 1.82)	9.48 × 10−9	2.05 (1.44 to 2.91)	6.18 × 10−5
DSC2	Desmocollin-2	1.68 (1.36 to 2.07)	1.31 × 10−6	1.21 (0.75 to 1.96)	0.43
FSTL3	Follistatin-related protein 3	2.52 (1.95 to 3.24)	1.10 × 10−12	3.54 (1.93 to 6.50)	4.45 × 10−5
LMAN2	Vesicular integral-membrane protein VIP36	1.59 (1.26 to 2.00)	7.89 × 10−5	3.46 (1.80 to 6.66)	2.06 × 10−4
NBL1	Neuroblastoma suppressor of tumorigenicity 1	1.46 (1.29 to 1.64)	3.31 × 10−10	2.04 (1.37 to 3.04)	4.39 × 10−4
PTGDS	Prostaglandin-H2 D-isomerase	1.44 (1.20 to 1.74)	1.23 × 10−4	2.30 (1.34 to 3.94)	2.58 × 10−3
TFF3b	Trefoil factor 3	1.31 (1.15 to 1.49)	5.30 × 10−5	1.50 (1.16 to 1.95)	2.05 × 10−3
TFF3b	Trefoil factor 3	1.22 (1.08 to 1.38)	1.66 × 10−3	1.50 (1.15 to 1.95)	2.57 × 10−3
TNFRSF1A	TNF receptor superfamily member 1A	2.02 (1.65 to 2.40)	1.19 × 10−11	2.90 (1.69 to 4.97)	1.16 × 10−4
TNFRSF1Bb	TNF receptor superfamily member 1B	1.57 (1.33 to 1.86)	1.17 × 10−7	2.31 (1.50 to 3.58)	1.61 × 10−4
TNFRSF1Bb	TNF receptor superfamily member 1B	1.49 (1.27 to 1.74)	6.82 × 10−7	2.31 (1.43 to 3.71)	5.58 × 10−4
WFDC2	WAP four-disulfide core domain protein 2	2.19 (1.79 to 2.68)	2.17 × 10−14	2.56 (1.54 to 4.24)	2.70 × 10−4
ZHX3	Zinc fingers and homeoboxes protein 3	1.30 (1.09 to 1.55)	3.37 × 10−3	1.98 (1.11 to 3.53)	2.05 × 10−2

All hazard ratios are expressed per doubling of protein level.

^aModel 2, adjusted for age, sex, race (CRIC only), systolic BP, antihypertension medications (CRIC only), diabetes (CRIC only), history of ASCVD, smoking, eGFR_crcys, HDL levels, and log-transformed urine protein-creatinine ratio.

^bTwo different aptamers separately quantified this protein.

All identified proteins represented genes that are widely expressed, including in the kidney (Supplemental Figure 2). Hierarchic modeling on the basis of gene expression patterns clustered TFF3, WFDC2, DLK2, and DSC2 at the opposite end of the spectrum from the other genes.

In ARIC, GWAS identified a pQTL for all but one of the proteins, and an available cis-genetic model for seven of the proteins (Supplemental Figure 3). Other studies provided additional orthogonal data for aptamer specificity for all target proteins but DCTN2 (Supplemental Table 5). Applying the genetic model of the protein level to summary statistics from the CKD Genetics Consortium eGFR GWAS showed a strong association between SNPs for LMAN2 and eGFR (P=5.62 × 10⁻⁵), all of the others were not significant (Table 4). The index cis-SNP for LMAN2 was associated with differential gene expression in whole blood (P=0.00055). It also had metabolite QTLs of creatinine, leucylphenylalanine, prolylproline, isoleucylvaline, and glutamine, and was reported to be associated with urea, cystatin C, and urate in the UK Biobank (Supplemental Tables 6 and 7).

Table 4.

Associations of cis-instrument models of top proteins with protein and eGFR

Gene Symbol	Chromosome; Transcription Start Site	Heritability of the Gene (%)	R2 for Protein Model (%)	P Value for Model Association with Protein	P Value for Model Association with eGFR
CLMP	11; 123069872	8.0	10.5	1.29 × 10−175	0.13
DSC2	18; 31058840	15.7	10.7	2.36 × 10−179	0.36
LMAN2	5; 177331567	6.5	2.3	2.62 × 10−38	5.62 × 10−5
NBL1	1; 19596979	3.6	3.7	6.72 × 10−62	0.35
TNFRSF1A	12; 6328757	2.8	1.6	8.56 × 10−27	0.22
TNFRSF1B	1; 12166991	11.3	13.6	3.39 × 10−232	0.95
WFDC2	20; 45469753	1.8	0.9	9.42 × 10−17	0.32

Discussion

This study of 9406 adults identified 13 proteins that were associated with the development of ESKD or a 50% decline in eGFR, both in midlife and older age. All of the proteins represented genes that were widely expressed, including in the kidney. Some of the proteins were known biomarkers of kidney disease, including TNFRSF1A and 1B, TFF3, and PTGDS, providing proof-of-concept support for the untargeted proteomic profiling approach. One of the proteins, LMAN2, had genetic determinants that were also associated with eGFR, providing support for elevated LMAN2 levels as a cause, rather than merely a marker, of CKD progression. Another implicated protein, WFDC2, had previously been identified as a promoter of extracellular matrix deposition in animal studies and is implicated in the development of renal fibrosis.^24,25 Taken together, this study suggests that untargeted, large-scale proteomic profiling can identify antecedents of kidney function decline, some of which may lie in the causal pathway and, thus, represent potential therapeutic targets, whereas others may be risk markers of progressive kidney damage.

All identified proteins had positive associations with kidney function decline, with higher levels associated with worse prognosis. The 13 proteins were significantly associated even after taking into account both eGFR and albuminuria, and 12 and nine proteins replicated in CRIC and AASK, respectively. Given the differences in these cohorts (ARIC is a general population cohort; CRIC enrolled patients with diabetic and nondiabetic nephropathy; AASK included Black participants with hypertension-attributed nephropathy), many of the observed protein effects appear highly generalizable. Indeed, two of the proteins, TNFRSF1A and TNFRSF1B, have been the subject of multiple recent investigations in children and adults with diverse disease etiologies.^5,26,27 Both TNFRSF1A and TNFRSF1B are involved in the inflammatory response; TNFSFR1A helps mediate caspase-dependent apoptosis, and TNFSFR1B mediates the metabolic effects of TNF-α. The latter is already used in clinical medicine: the extracellular ligand-binding portion of TNFRSF1B, linked to an Ig Fc chain, is marketed as etanercept and used to treat rheumatoid arthritis. Interestingly, etanercept and other biologic agents used in rheumatoid arthritis were associated with lower risk of incident CKD and GFR decline in an observational study.²⁸ In our study, genetic prediction did not support a causal relationship between TNFSFR1A or TNFSFR1B and the development of all-cause kidney disease, suggesting genetically determined levels may not cause disease; however, the instrument for TNFSFR1A, in particular, was relatively weak. In addition, if pleiotropy exists, some mechanisms for elevating TNFSR1B would be harmful, and others would not.

Other identified proteins included PTGDS and TFF3. Both have been investigated as biomarkers of kidney disease. PTGDS is a slightly larger molecule than creatinine and cystatin C, and has been proposed as an alternate filtration marker, particularly among transplant recipients, where change in muscle mass and/or use of immunosuppressive medications may confound the relationship between the more traditional biomarkers and kidney function.²⁹ TFF3, on the other hand, was noted by regulatory bodies as an indicator of drug-induced kidney injury in preclinical drug development.³⁰ Urinary TFF3 levels have also been associated with future kidney disease risk in humans, heralding higher risk of incident CKD, even with adjustment for eGFR and albuminuria.³¹ Previous work identified rare variants in LRP2 which were both associated with urinary TFF3 levels and with eGFR, supporting a causal relationship.³² Our GWAS of TFF3 did not identify variants in LRP2 as a major modulator of serum levels; however, we did not evaluate SNPs with minor allele frequencies <1%. Serum levels of TFF3 have been associated with more advanced kidney disease,³³ and immunostaining suggests TFF3 is located in renal tubular epithelial cells, with higher mRNA expression corresponding to higher degrees of tubulointerstitial fibrosis and urinary TFF3 secretion.³⁴

The plasma protein with genetic evidence of a causal relationship, LMAN2, or vesicular integral-membrane protein VIP36, was not previously appreciated as a risk factor for CKD progression. LMAN2 is a transmembrane protein that is thought to aid with sorting, transport, and modification and metabolism of high mannose–type glycoproteins. Shown to have high expression in the kidneys and an apical-predominant distribution, which facilitates apical glycoprotein secretion,³⁵ LMAN2 expression is differentially expressed during acute renal allograft rejection.³⁶ Our GWAS of the LMAN2 protein showed associations in distinct genetic regions: the index cis-SNP on chromosome 5 has been linked to creatinine, isoleucylphenylalanine, glutamine, and histidine in metabolite GWAS; and to expression of LMAN2, FGFR4, NSD1, FAM153A, RAB24, MXD3, and PRELID1 in multiple tissues.^37–40 In addition, the index SNP was linked to cystatin C, urea, and urate in the UK Biobank, providing further evidence of a causal effect on kidney function.²¹

Although we found no genetic evidence of a causal association for WFDC2 and eGFR, the genetic instrument was relatively weak. WFDC2 is a known prognostic marker in multiple diseases, including cancer.^41,42 Also known as HE4, WFDC2 is a secreted, type 4 glycoprotein thought to be involved in inhibiting protease activity.⁴³ Relevant to our work, WFDC2 has been used as a predictor of the development of lupus nephritis among patients with systemic lupus erythematosus.⁴⁴ In cancer, it promotes tumor growth and migration of malignant cells.⁴³ Recent work has also linked WFDC2 to immune-related gene expression in PBMCs.⁴⁵ Genes stimulated by the protein include CSF3, IL6, CCL20, IL1A, CXCL1, CSF2, CCL18, CCL4, IL18, PTGS2, CXCL2, IL10, CXCL8 (IL8), CXCL5, CCR7, TNF, IDO1, and CD274, all with at least ten-fold increase in gene expression, with some via activation of STAT3 signaling. Gene expression of WFDC2 is also upregulated in human fibrotic kidneys, and a mouse model of renal fibrosis showed that pretreatment with a neutralizing antibody for WFDC2 attenuates the formation of fibrosis.⁴⁶ Taken together, these data suggest WFDC2 may have a causal role in the development of renal fibrosis and the progression of CKD.

Other novel kidney disease–related proteins that replicated in external cohorts included CLMP, DLK2, DSC2, FSTL3, and NBL1. Most were strongly associated with eGFR_crcys, with correlations ranging from −0.25 to −0.41. CLMP, a membrane protein thought to be involved in cell-cell adhesion and adipocyte differentiation, has been associated with proteinuria in children with CKD. CLMP colocalizes with the tight junction protein ZO-1 on the slit diaphragm.⁴⁷ DSC2 is also involved in cell adhesion, acting as a calcium-dependent glycoprotein in the desmosome to hold adjacent plasma membranes together, and has been associated with eGFR decline in both the Jackson Heart Study (JHS) and the Framingham Heart Study (FHS).⁴⁸ Although not statistically significant at a level <5 × 10⁻⁵, the index SNP for DSC2 was associated with renal failure in the UK Biobank (P=0.003).¹⁸ FSTL3 is an extracellular matrix protein that binds morphogens to regulate activity during development, and was also associated with eGFR decline in JHS and FHS.⁴⁸ DLK2 is involved in calcium ion binding and regulates adipogenesis. NBL1 is also involved in morphogen activity. Variants in the NBL1 gene have been associated with height, white blood cell count, and several urine metabolites, including glycerophosphorylcholine.⁴⁹

This study has strengths and limitations. It employs large-scale profiling of the proteome, allowing for hypothesis-generating findings. However, not all Slow Off-rate Modified Aptamer–quantified proteins are perfect correlates of those identified through targeted assays.⁵⁰ Thus, we used GWAS to evaluate for expected associations with protein levels, providing orthogonal support for aptamer specificity. We evaluated associations between cis-instruments for seven of the protein levels and eGFR to provide supportive evidence for a causal relationship. Although this method minimizes the possibility of pleiotropy, it is limited in evaluating only the genetically determined levels of proteins, and does not rule out a causal relationship for proteins with missing or weak genetic instruments, or causal proteins with an environmental interaction. The genetic models were performed only in the White population due to a paucity of summary GWAS data from genetically diverse populations. Our discovery approach used the same individuals from a single population-based study in two different risk periods. Participants in the visit 5 cohort were much older and represent a subset of those who survived and attended a subsequent study visit from the visit 3 cohort. This approach may have resulted in identification of only a small subset of proteins; however, the initial discoveries replicated well in two CKD cohorts. Finally, the associations between proteins and kidney outcomes appeared stronger in lower eGFR, lending support for the need to perform additional “omic” studies in CKD populations.

In summary, large-scale proteomic analysis identified 13 known and novel proteins associated with eGFR decline in a general population cohort, of which 12 were also associated in at least one independent CKD cohort. Follow-up studies should continue to evaluate these proteins, particularly LMAN2, CLMP, and WFDC2, which have some evidence of a causal role in promoting eGFR decline.

Disclosures

D. E. Arking reports serving on the scientific advisory board for the Association for the Eradication of Heart Attack. C. M. Ballantyne reports having consultancy agreements with Abbott Diagnostics, Althera, Amarin, Amgen, Arrowhead, AstraZeneca, Corvidia, Denka Seiken, Esperion, Genentech, Gilead, Matinas BioPharma Inc, New Amsterdam, Novartis, Novo Nordisk, Pfizer, Regeneron, Roche Diagnostic, Sanofi-Synthelabo; receiving research funding from Abbott Diagnostics, Akcea, Amgen, Esperion, Ionis, Novartis, Regeneron, Roche Diagnostic; and serving as a scientific advisor for, or membership of, Amarin, Amgen, Arrowhead, AstraZeneca, Corvidia, Esperion, and Matinas BioPharma. E. Boerwinkle reports ownership interest in Codified Genomics. T. K. Chen reports receiving research funding from the National Institutes of Health (NIH)/National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) and Yale University. J. Coresh reports having ownership interest in Health.io; having consultancy agreements with Healthy.io, Kaleido, and Ultragenyx; serving as a scientific advisor for, or member of, Healthy.io and NKF; and receiving research funding from National Kidney Foundation (NKF; which receives industry support) and NIH. H. Feldman reports serving as editor in chief of the American Journal of Kidney Disease (AJKD), on the steering committee for the CRIC Study, and as a member of the advisory board for NKF; having consultancy agreements with DLA Piper LLP, InMed Inc., Kyowa Hakko Kirin Co. Ltd. (ongoing), and NKF (ongoing); and receiving honoraria from Rogosin Institute (invited speaker). P. Ganz reports having consultancy agreements with, and serving on the medical advisory boards of, Itamar and SomaLogic (no money is received from either company). M. E. Grams reports serving as a scientific advisor for, or member of, AJKD, CJASN, JASN (as editorial board member), Kidney Disease Improving Global Outcomes (KDIGO; on the executive committee), NKF (on the scientific advisory board), and USRDS (on the scientific advisory board); receiving the American Society of Nephrology for Young Investigator Award, and honoraria from academic institutions for giving grand rounds; receiving travel support from DCI to speak at the annual meeting, and support from KDIGO for participation in scientific meetings and the executive committee; and having other interests in/relationships with NKF, which receives funding from Abbvie, Relypsa, and Thrasos. C. M. Rebholz reports serving as an editorial board member for Diabetes Care. P. Welling reports serving as a scientific advisor for, or member of, American Journal of Physiology (on the renal editorial board), American Physiological Society (as chair of the finance committee), Kidney Molecular Biology and Development (as chair), and the NIH; receiving honoraria from American Physiological Society; and receiving research funding from LeDucq Foundation and NIH. Z. Zheng reports having consultancy agreements with Akebia Therapeutics Inc. All remaining authors have nothing to disclose.

Funding

The ARIC Study has been funded, in whole or in part, by federal funds from the National Heart, Lung, and Blood Institute (NHLBI), under contract numbers HHSN268201700001I, HHSN268201700002I, HHSN268201700003I, HHSN268201700004I, HHSN268201700005I, R01HL087641, R01HL059367, and R01HL086694; National Human Genome Research Institute contract U01HG004402; and NIH contract HHSN268200625226C. Funding for laboratory testing and biospecimen collection at ARIC visit 6 was supported by NIDDK grant R01DK089174. Infrastructure was partly supported by NIH grant UL1RR025005 (a component of the NIH Roadmap for Medical Research). This study was also supported by the CKD Biomarkers Consortium via NIH grants U01 DK106981 (principal investigator [PI], E. P. Rhee), U01 DK085689 (PI, J. Coresh), R01DK108803, and R01DK124399 (PI, M. E. Grams). Funding for the CRIC Study was obtained under a NIDDK cooperative agreement, under grants U01DK060990, U01DK060984, U01DK061022, U01DK061021, U01DK061028, U01DK060980, U01DK060963, U01DK060902, and U24DK060990. In addition, this work was supported, in part, by National Center for Advancing Translational Sciences grants UL1TR000003 (via the Perelman School of Medicine at the University of Pennsylvania Clinical and Translational Science Award), UL1 TR-000424 (via Johns Hopkins University), UL1TR000439 (from the NCATS component of the NIH and NIH roadmap for Medical Research), UL1TR000433 (via the Michigan Institute for Clinical and Health Research), UL1RR029879 (via the University of Illinois at Chicago Clinical and Translational Science Award), P20 GM109036 (via the Tulane COBRE for Clinical and Translational Research in Cardiometabolic Diseases), UL1 RR-024131 (via Kaiser Permanente NIH/National Center for Research Resources University of California San Francisco–Clinical and Translational Science Institute); University of Maryland General Clinical Research Center grant M01 RR-16500; Clinical and Translational Science Collaborative of Cleveland; NIDDK grant R01DK119199 (via the Department of Internal Medicine, University of New Mexico School of Medicine Albuquerque). This work was also supported, in part, by NHLBI grant R01 HL134320.

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2020111607/-/DCSupplemental.

Supplemental Figure 1. Flow chart of the ARIC Study.

Supplemental Figure 2. Gene expression patterns for the 13 proteins associated with 50% decline in eGFR or ESKD in ARIC visit 3 and ARIC visit 5 cohorts.

Supplemental Figure 3. GWAS of proteins of interest (adjusted for age, sex, genetic PCs 1-10, protein PCs 1-10) among Caucasian participants of the Atherosclerosis Risk in Communities study (visit 3 cohort).

Supplemental Table 1. Quality control metrics in the ARIC visit 3 and visit 5 cohort for all 4877 SOMA proteins.

Supplemental Table 2. Proteins with FDR p<0.05 in the ARIC visit 3 cohort, model 2.

Supplemental Table 3. Proteins with FDR p<0.05 in the ARIC visit 5 cohort, model 2.

Supplemental Table 4. Top proteins in 1) multivariable model, 2) when adjusting for known kidney biomarkers, and 3) ARIC.

Supplemental Table 5. Orthogonal validation of aptamer specificity for top proteins.

Supplemental Table 6. Annotation of index SNPs for top proteins.

Supplemental Table 7. Summary of gene expression and protein function for top proteins.