Abstract
Background and objectives Using a discovery/replication approach, we investigated associations between a multiplex panel of 80 circulating proteins associated with cardiovascular pathology or inflammation, and eGFR decline per year and CKD incidence.
Design, setting, participants, & measurements We used two cohorts, the Prospective Investigation of the Vasculature in Uppsala Seniors Study (PIVUS; n=687, mean age of 70 years, 51% women) and the Uppsala Longitudinal Study of Adult Men (ULSAM; n=360 men, mean age of 78 years), with 5-year follow-up data on eGFR. There were 231 and 206 incident cases of CKD during follow-up in the PIVUS and ULSAM studies, respectively. Proteomic profiling of 80 proteins was assessed by a multiplex assay (proximity extension assay). The assay uses two antibodies for each protein and a PCR step to achieve a high-specific binding and the possibility to measure multiple proteins in parallel, but gives no absolute concentrations.
Results In the discovery cohort from the PIVUS Study, 28 plasma proteins were significantly associated with eGFR decline per year, taking into account the multiple testing. Twenty of these proteins were significantly associated with eGFR decline per year in the replication cohort from the ULSAM Study after adjustment for age, sex, cardiovascular risk factors, medications, and urinary albumin-to-creatinine ratio (in order of significance: TNF-related apoptosis-inducing ligand receptor 2*, CD40L receptor, TNF receptor 1*, placenta growth factor*, thrombomodulin*, urokinase plasminogen activator surface receptor*, growth/differentiation factor 15*, macrophage colony-stimulating factor 1, fatty acid-binding protein*, cathepsin D, resistin, kallikrein 11*, C-C motif chemokine 3, proteinase-activated receptor 1*, cathepsin L, chitinase 3-like protein 1, TNF receptor 2*, fibroblast growth factor 23*, monocyte chemotactic protein 1, and kallikrein 6). Moreover, 11 of the proteins predicted CKD incidence (marked with * above). No protein consistently predicted eGFR decline per year independently of baseline eGFR in both cohorts.
Conclusions Several circulating proteins involved in phosphate homeostasis, inflammation, apoptosis, extracellular matrix remodeling, angiogenesis, and endothelial dysfunction were associated with worsening kidney function. Multiplex proteomics appears to be a promising way of discovering novel aspects of kidney disease pathology.
- Adult
- Aged
- Albumins
- apoptosis
- Cardiovascular Diseases
- creatinine
- extracellular matrix
- Fatty Acid-Binding Proteins
- Fibroblast Growth Factors
- Follow-Up Studies
- glomerular filtration rate
- Homeostasis
- Longitudinal Studies
- Phosphates
- Plasminogen
- Prospective Studies
- Proteomics
- risk factors
- Urokinase-Type Plasminogen Activator
Introduction
CKD affects 10% of the general population worldwide (1), but the prevalence is expected to increase even further given the ongoing obesity epidemic and the growing elderly population (2). CKD has a substantial effect on public health, as decline in eGFR is closely associated with the risk of ESRD, cardiovascular disease (CVD), and mortality (2–5). The complex underlying pathophysiology leading to kidney disease is still not completely understood and there is a need for specific, sensitive, and clinically relevant kidney disease biomarkers that may help to improve the identification of individuals at high risk of CKD progression and its consequences.
Advancement of technology that makes it possible to simultaneously measure a large number of proteins (6) provides new opportunities for unbiased discovery of novel pathophysiologic pathways of disease, as well as for the identification of novel disease biomarkers. Yet the utility of proteomic profiling for the development of kidney disease is less studied, and previous studies have generally been performed in small study samples of individuals with CKD or diabetes (7–11). Importantly, community-based data are lacking.
Therefore, we aimed to explore and validate the associations between a multiplex assay panel of 92 plasma proteins involved in CVD pathology or inflammation with eGFR decline per year in two independent community-based cohorts of elderly people. Given the close interplay between kidney disease, inflammation, and CVD (12,13), we hypothesized that this particular proteomics panel could be relevant in this respect.
Materials and Methods
The Prospective Investigation of the Vasculature in Uppsala Seniors Study
All 70-year-old men and women living in Uppsala, Sweden, between the years 2001 and 2004 were eligible for the Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) Study (described in detail at http://www.medsci.uu.se/pivus/pivus.htm) (14). Of 2025 invited individuals, 1016 agreed to participate. A second examination cycle was performed between 2006 and 2009, when participants were 75 years old. Of 964 invited participants, 827 participated (86%). Of these, 687 individuals had valid measurements of the proteomic assay, eGFR at baseline and at follow-up, and covariates, and thus constitute the present sample.
The Uppsala Longitudinal Study of Adult Men
The Uppsala Longitudinal Study of Adult Men (ULSAM) was initiated in 1970. All 50-year-old men born in 1920–24 and living in Uppsala, Sweden, were invited to participate in a health survey, focusing on identifying cardiovascular risk factors (described in detail at http://www.pubcare.uu.se/ulsam/) (15). This study used the fourth examination cycle as baseline, when participants were approximately 77 years old (1998–2001). Of 1398 invited men, 838 (60%) participated, and data on the proteomics assay were available in 786 individuals. We used the fifth examination cycle (2003–2005), when participants were approximately 82 years old, as a follow-up examination. To this examination, 952 men still living in Uppsala were invited and 530 men (56%) participated. After exclusion of individuals without data on covariates and eGFR at both examinations, the present sample is comprised of 360 men.
All participants in the PIVUS and ULSAM studies gave written informed consent and the Ethics Committee of Uppsala University approved the study protocols. Both studies were conducted according to the Declaration of Helsinki.
Baseline and Follow-Up Investigations
The investigations in the PIVUS and ULSAM studies were performed using similar standardized methods, including anthropometrical measurements, BP, blood sampling, and questionnaires regarding socioeconomic status, medical history, smoking habits, medication, and physical activity level (14,15). In both cohorts, venous blood samples were drawn in the morning after an overnight fast and stored at −70°C until analysis.
In the ULSAM Study, eGFR was analyzed from serum cystatin C using latex-enhanced reagent (N Latex Cystatin C; Dade Behring, Deerfield, IL) on a BN ProSpec analyzer (Dade Behring), and in the PIVUS Study by an assay from Gentian (Gentian, Moss, Norway). As the analyses of cystatin C were performed before the development of the international reference standard for cystatin C (16), we used eGFR equations that were specifically developed for the particular cystatin C assays at our laboratory (ULSAM Study: eGFR=77.24×cystatin C−1·2623; PIVUS Study: eGFR=79.901×cystatin C−1.4389). Both formulae for eGFR are closely correlated with iohexol clearance (17,18).
In the ULSAM Study, a 24-hour collection of urine was made and stored at −70°C until analysis. No urine was collected at baseline in the PIVUS Study. Urine albumin was measured by nephelometry (Reagent OSAL15 and OUMS65; Dade Behring, Deerfield, IL) using a BN ProSpec analyzer (Dade Behring). Urinary creatinine was analyzed with a modified kinetic Jaffe reaction on an Architect Ci8200 analyzer (Reagent 3L81; Abbott, Abbot Park, IL) and urinary albumin-to-creatinine ratio was calculated.
Diabetes mellitus was diagnosed as fasting plasma glucose ≥7.0 mmol/L (≥126 mg/dl) or use of antidiabetic medication (19).
The Proseek Multiplex Panel
The Olink Proseek Multiplex Cardiovascular 1 96×96 kit was used to measure proteins in plasma by real-time PCR using the Fluidigm BioMark HD real-time PCR platform, as described previously (6,20). Of the 96 wells, one is a negative control and three are positive controls (spiked in IL-6, IL-8, and VEGF-A), resulting in 92 measured proteins. Each sample includes two incubations, one extension, and one detection control used to determine the lower detection limit and normalize the measurements. The resulting relative values obtained are log2-transformed for subsequent analysis. Twelve proteins where <85% of the individuals had a valid measurement of that protein were removed from further analysis (Supplemental Table 1), Hence, 80 proteins were included in the statistical analysis. Individuals with excess missing protein data were also excluded (>5% and >3% in the PIVUS and ULSAM studies, respectively). Values below the lower limit of detection (LOD) were replaced with LOD/2. Each protein was normalized by plate (by setting the mean=0 and SD=1 within each plate) and storage time (correction on the basis of the observed values and predicted values from a spline model). In a previous validation study of the proteomics assay that included 90 proteins and seven samples analyzed, the mean intra-assay coefficient of variation was found to be 8% and the mean interassay coefficient of variation was 12% (6). A more detailed information on the coefficients of variation of specific proteins can be found on the Olink website (www.olink.com).
We used the Protein Annotation Through Evolutionary Relationship Pathways annotation data set and literature review to find common biologic processes between the replicated proteins (21).
Outcomes
Data on eGFR was available both at baseline and at a reinvestigation after approximately 5 years in both cohorts. eGFR decline per year was calculated and used as the primary end point. Incident CKD was defined as having eGFR<60 ml/min per 1.73 m2 at the follow-up investigation in those with eGFR≥60 ml/min per 1.73 m2 at baseline.
Statistical Analyses
For the primary analysis, the PIVUS Study was used as the discovery cohort and the ULSAM Study was used for replication (22). For discovery, the association between each of the proteins and eGFR decline per year was investigated using multivariable linear regression models adjusting for age, sex, LDL and HDL cholesterol, lipid-lowering treatment, systolic BP, specified antihypertensive drug classes (angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, β-blockers, calcium antagonists, and diuretics [mainly thiazides]), body mass index, fasting glucose, oral antidiabetic treatment, insulin treatment, and smoking, all assessed at baseline. We chose this approach as we primarily wanted to identify proteins that were associated with eGFR decline independently of established risk factors for cardiovascular or kidney disease. The proteins showing a false discovery rate <0.05 were taken further to linear regression analyses in the ULSAM Study replication cohort. In the replication step, baseline urinary albumin-to-creatinine ratio was also added to the multivariable model. A nominal P value of <0.05 for the multivariate-adjusted analysis was considered statistically significant for the analyses in the ULSAM Study replication cohort.
Secondary Analyses
In secondary analyses, we added baseline eGFR to the aforementioned multivariable models. We performed subgroup analyses in participants without diabetes at baseline (n=613 for the PIVUS Study and n=320 for the ULSAM Study). We also investigated the association between the proteins that were replicated and incident CKD in participants with eGFR>60 ml/min per 1.73 m2 at baseline (n=660 for the PIVUS Study and n=319 for the ULSAM Study). Finally, we investigated the cross-sectional association between the proteins and baseline eGFR in both cohorts. STATA version 14 was used for calculations (StataCorp., College Station, TX).
Results
Baseline characteristics for the covariate factors in the PIVUS and ULSAM studies are shown in Table 1.
Characteristics in the PIVUS and ULSAM studies
eGFR Decline per Year in the PIVUS Study: Discovery
The multivariate β coefficients and P values for the linear association between all proteins and eGFR decline per year in the PIVUS Study are shown in Figure 1. A total of 28 proteins were significantly associated with eGFR decline per year when using a false discovery rate of 5% (Figure 1).
Twenty-eight proteins were significantly associated with eGFR decline per year in the Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) Study: multivariate linear regression. Data are regression coefficients (β) and 95% confidence intervals (95% CIs).
eGFR Decline per Year in the ULSAM Study: Replication
In the ULSAM Study replication cohort, 20 of these proteins were significantly associated with eGFR decline per year (Figure 2A, Table 2).
Twenty proteins were significantly associated with eGFR decline per year in both PIVUS and ULSAM studies: multivariate linear regression. Almost all proteins were also significant in both PIVUS and ULSAM studies in participants without diabetes. Data are regression coefficients (β) and 95% confidence intervals (95% CIs). (A) Whole cohort. (B) Participants without diabetes. PIVUS, the Prospective Investigation of the Vasculature in Uppsala Seniors Study; ULSAM, the Uppsala Longitudinal Study of Adult Men.
List of the 20 proteins that were significantly associated with eGFR decline in both the PIVUS and ULSAM studies
When excluding participants with a diabetes diagnosis at baseline, the associations were similar, except for chitinase 3-like protein 1, which was found to be of borderline significance in the ULSAM Study (P=0.07; Figure 2B).
Secondary Analyses
eGFR Decline per Year in the PIVUS and ULSAM Studies: Adjustment for Baseline eGFR, Nonsteroidal Anti-Inflammatory Drug Use, or Mineral Metabolism Factors.
None of the 20 proteins were consistently associated with eGFR decline per year in both cohorts after further adjustment for baseline eGFR. Cathepsin L was significantly associated with eGFR decline per year in the PIVUS Study (0.20 ml/min per 1.73 m2 per year faster decline per SD log-transformed cathepsin L; 95% confidence interval, 0.04 to 0.35; P=0.01), as was CD40L receptor in the ULSAM Study (0.44 ml/min per 1.73 m2 per year faster decline per SD log CD40L; 95% confidence interval, 0.02 to 0.85; P=0.04).
CKD Incidence in the PIVUS and ULSAM Studies.
In the subset free from CKD at baseline, 231 of 660 participants in the PIVUS Study and 206 of 319 participants in the ULSAM Study developed incident CKD at follow-up. Eleven of the abovementioned 20 proteins consistently predicted incident CKD in subgroup analyses in those with eGFR>60 ml/min per 1.73 m2 at baseline in both cohorts (Figure 3, Table 2).
Eleven proteins predict incident CKD in subgroup analyses in those with eGFR>60 ml/min per 1.73 m2: multivariate logistic regression. Data are odds ratios (ORs) and 95% confidence intervals (95% CIs). PIVUS, the Prospective Investigation of the Vasculature in Uppsala Seniors Study; ULSAM, the Uppsala Longitudinal Study of Adult Men.
Cross-Sectional Analyses
Please see Supplemental Tables 2 and 3 for the cross-sectional association between the proteins and baseline eGFR in both cohorts. With the exception of a nonsignificant association between cathepsin L and baseline eGFR in the PIVUS Study, higher levels of 19 of the 20 abovementioned proteins were associated with lower baseline GFR in both cohorts in cross-sectional multivariable linear regression analyses. Moreover, a total of 60 proteins in the PIVUS Study and 58 proteins in the ULSAM Study were significantly associated with baseline eGFR.
Discussion
Principal Findings
Using a multiplex proteomics assay primarily designed to be relevant for CVD pathology and inflammation, we discovered and replicated independent associations between 20 circulating proteins with kidney function decline (Table 2). These associations were generally robust in individuals without diabetes. More than half of the proteins also predicted incident CKD, suggesting that pathways that are generally considered to be important primarily in the later stages of CKD, such as a disturbed phosphate homeostasis or apoptosis, are also involved in the early stages of progressive kidney disease. An impaired phosphate homeostasis appeared to be a particularly important common pathway for the proteins associated with kidney function decline because half of the proteins were involved in this pathway, according to the functional analyses of our findings. Of all of the individual proteins, TRAIL-R2 was most strongly associated with eGFR decline per year.
Comparison with Other Studies
Prior studies trying to identify novel kidney disease biomarkers have generally been experimental colocalization studies or animal studies (23–26), but in recent years there have also emerged some promising data suggesting that a panel of 273 urinary proteins (CKD273) could improve the prediction of kidney function decline and CKD progression (7,8,10,11). Studies evaluating the association between circulating proteomics and kidney function decline are more rare. A recent nested case-control study in patients with diabetes reported that a panel of 14 serum biomarkers substantially increased the discrimination of individuals with a rapid decline in eGFR from those with a stable eGFR (9). It should be noted that most prospective proteomics studies conducted so far have had small to moderate sample sizes, and have lacked external validation (27). Importantly, we are aware of no previous studies reporting the association between circulating proteomics and kidney function decline in the general population. However, some of the individual proteins that we investigated in this study have previously been shown to be associated with eGFR decline: U-PAR (28), GDF-15 (29), CD40L (30), PlGF (29), RETN (31), sTNFR1 and sTNFR2 (32,33), and mixed results have been reported for FGF-23 (34,35). For the remaining 12 proteins, including TRAIL-R2, we have not been able to find previous longitudinal studies investigating the association between plasma levels of these proteins and kidney function decline, emphasizing the potential utility of circulating multiplex proteomics for biomarker discovery in kidney disease.
Potential Mechanisms Explaining Our Main Findings
The 20 proteins share several common biologic pathways that are important for the progression of kidney disease: phosphate homeostasis (FGF-23, PAR-1, FABP4, CD40L, U-PAR, CCL3, sTNFR1, sTNFR2, CHI3L1, and GDF-15), inflammation (sTNFR1, sTNFR2, U-PAR, hK11, CD40L, FGF-23, GDF-15, CSF-1, CCL3, CHI3L1, and MCP-1), apoptosis (TRAIL-R2, GDF-15, sTNFR1, sTNFR2, U-PAR, PAR-1, CD40L, and CCL3), proteolysis and extracellular matrix remodeling (U-PAR, CTSD, CTSL, CHI3L1, hK11, and KLK6), angiogenesis (PlGF, CTSD, and CHI3L1), endothelial function (PlGF, GDF-15, and PAR-1), and thrombosis (TM, PAR-1, and CD40L).
Specific effects of individual proteins, such as fatty acid transport (FABP4), lipoprotein formation (RETN), or antigen presentation (CD40L), could also provide explanations for these associations.
In addition to the aforementioned mechanisms of the different proteins, many may have pleiotropic effects and there could also be kidney-specific mechanisms that are not yet identified. Further experimental studies are warranted to explain the role of the proteins in kidney disease progression. Our data provide a basis that may help guide future directions of such experimental research.
For TRAIL-R2, there are some mechanisms that may explain why it was the protein that was most closely associated with eGFR decline. TRAIL-R2 is expressed in many tissues in the body, including in convoluted tubules and in the loop of Henle, but not in the glomeruli or in renal vasculature (36). The overall main function of TRAIL-R2 is to penetrate cell membranes and to induce an intracellular reaction leading to apoptotic cell death. The TRAIL-induced cell death could play an important role in the progression of diabetic nephropathy (37). Renal TRAIL expression has been shown to be consistently higher in animals with diabetes than in those without (36), and circulating TRAIL levels have also been shown to be higher in patients with diabetes and albuminuria than in patients with normoalbuminuria (38). TRAIL has also been suggested to be involved in tubular and glomerular injury and its expression to be induced after exposure to toxic metals (36). The TRAIL-R2 receptor is activated by TRAIL, and in a previous study in patients with CKD, lower levels of soluble TRAIL were associated with worsening CKD categories in cross-sectional analyses (39). Our data confirm and extend the notion that the interplay between TRAIL and its receptors is an important factor contributing to kidney disease progression.
There are several common underlying risk factors in the development of kidney disease and CVD (12), such as diabetes (40), hypertension (41), dyslipidemia (42), inflammation (42), and albuminuria (42). However, the fact that the association between the proteins and eGFR decline was independent of all aforementioned risk factors suggests that these are not major mediators of the present associations.
Clinical Implications
Since the 1990s, the global number of deaths due to CKD has almost doubled (43), and the incidence and prevalence of CKD is expected to increase even further. In this moving landscape, an improved understanding of the underlying causes of CKD and the identification of high-risk individuals for CKD progression will be increasingly important. Although 20 proteins were identified as potential risk markers in this study, none of them were consistently associated with eGFR decline after taking into account the baseline eGFR. From a mechanistic viewpoint, adjustment for baseline eGFR may not necessarily be appropriate because lower baseline eGFR could be regarded as an intermediate state along the causal pathway for further eGFR decline, and consequently constitutes an overadjustment that hides true associations. The strong cross-sectional associations between the proteins and baseline eGFR support this notion. However, from a clinical perspective, it is essential that a novel kidney disease biomarker provides prognostic information on CKD progression beyond the baseline eGFR. Thus, our proteomic profiling did not identify a promising biomarker that seems to have clinical utility for risk prediction purposes.
Strengths and Limitations
Strengths of our investigation include the longitudinal study design and the replication of findings in an independent cohort. Limitations include the unknown generalizability to other age and ethnic groups, and that there may be healthy cohort effects at play in population-based invited investigations of elderly individuals. The healthy cohort effects may be of particular importance in this study, given that only those surviving 5 years or more could be included. It is reasonable to assume that those with the most rapid decline in eGFR did not attend the re-examination to the same extent as those with a more stable eGFR. Another limitation is the fact that our study used single assessments of eGFR and the proteomic assay. However, the potential misclassification because of short-term variability would likely only result in more conservative estimates. Limitations of the proteomics assay include that no absolute levels of the proteins are obtained, which makes comparisons between studies or defining relevant cut-off limits difficult, and that the selection process of proteins on the assay was on the basis of both the availability of high-quality antibodies and the limits of what range of concentrations could be measured using the same assay. Moreover, the coefficients of variation of the proteomics assay were on the basis of data from a previous publication (6) and on information provided by the manufacturer Olink (www.olink.com). Thus, the performance of the proteomics assay in the clinical practice setting warrants further study.
Conclusions
The vast number of proteins independently associated with decline in kidney function show CKD to be a multifactorial and highly complex disease, involving an impaired phosphate homeostasis, inflammation, apoptosis, increased extracellular matrix remodeling, a disturbed angiogenesis, and endothelial dysfunction. The strongest association was found for TRAIL-R2, a protein that has not been associated with eGFR decline previously. Multiplex proteomics appears to be a promising way of discovering novel aspects of kidney disease pathology. Additional studies are needed to identify kidney disease biomarkers with potential for clinical utility.
Disclosures
The manufacturer of the protein assay, Olink Biosciences, had no input in the study design, analysis, or manuscript preparation. E.I. is an advisor and consultant for Precision Wellness, Inc., and advisor for Cellink and Olink Proteomics. J.S. has an advisory board membership for Itrim. J.Ä. has received lecturing fees from AstraZeneca.
Acknowledgments
J.Ä. is the guarantor of this work, had full access to all of the data, and takes full responsibility for the integrity of data and the accuracy of data analysis.
This study was supported by The Swedish Research Council, the Swedish Heart-Lung Foundation, the European Union Horizon 2020 (grant number 634869), the Marianne and Marcus Wallenberg Foundation, Dalarna University, and Uppsala University.
The funding sources did not play any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.
Footnotes
Published online ahead of print. Publication date available at www.cjasn.org.
See related editorial, “The Possibilities to Improve Kidney Health with Proteomics,” on pages 1206–1208.
This article contains supplemental material online at http://cjasn.asnjournals.org/lookup/suppl/doi:10.2215/CJN.08780816/-/DCSupplemental.
- Received August 18, 2016.
- Accepted April 17, 2017.
- Copyright © 2017 by the American Society of Nephrology