Urinary peptide biomarker panel associated with an improvement in estimated glomerular filtration rate in chronic kidney disease patients

ABSTRACT

Background

An improvement in the glomerular filtration rate (GFR) of chronic kidney disease patients has been an underestimated clinical outcome. Although this may be considered as an unexpected disease course, it may provide some insights into possible mechanisms underlying disease remission and/or regression. Therefore, our aim was to identify urinary peptide biomarkers associated with an improvement in estimated GFR (eGFR) over time and to improve patient stratification.

Methods

Capillary electrophoresis coupled with mass spectrometry (CE-MS) was employed to evaluate the urine peptidome of patients with different types of renal diseases. In total, 376 patients with a slope/year between −1.5% and +1.5% were designated as non-progressors or stable, while 177 patients with a > 5% slope/year were designated as patients with an improved eGFR for state-of-art biomarker discovery and validation.

Results

We detected 384 significant peptide fragments by comparing the CE-MS data of the stable patients and those with improved renal function in our development cohort. Of these 384, a set of 141 peptides with available amino acid sequence information were used to generate a support vector machine-based classification panel. The biomarker panel was applied to our validation cohort, achieving a moderate area under the curve (AUC) value of 0.85 (81% sensitivity and 81% specificity). The majority of the peptides (78%) from the diagnostic panel arose from different types of collagen.

Conclusions

We have developed a panel of urinary peptide markers able to discriminate those patients predisposed to improve their kidney function over time and possibly be treated with more specific or less aggressive therapy.

INTRODUCTION

Chronic kidney disease (CKD) is a long-term condition that affects ∼10% of the world’s population [1]. It is a significant global health problem with an ever-increasing prevalence, especially among the elderly population and individuals suffering from diabetes, hypertension or obesity [1–3].

Increased serum creatinine (SCr) levels, due to a decreased glomerular filtration rate (GFR) and/or the presence of proteinuria, are the current diagnostic parameters employed in the various stages of CKD [4]. As expected, CKD progression is associated with poorer clinical outcomes and a higher risk of mortality, even in the presence of mild to moderate kidney damage [5]. Preventive measures such as dietary modifications, exercise and therapeutic intervention aimed at controlling blood glucose and blood pressure are more efficient if applied in the early stages of disease [6]. During recent decades, CKD treatment has been focused on slowing down the progression towards end-stage renal disease (ESRD) [7]. However, cardiovascular-related comorbidities produce a higher risk of mortality than ESRD progression itself [8–10].

Conversely, a substantial proportion of patients whose renal function (i.e. an increased GFR slope) stabilized or even improved has been reported in several previous studies [11, 12]. Such improvements in GFR over time have also been confirmed by direct measurements of kidney function [13, 14], with improvements or stabilization in renal function being detected in patients with a GFR between 25 and 55 mL/min/1.73 m². This also included patients with underlying diabetic nephropathy (DN), which was previously believed to progress irreversibly towards ESRD [15]. In other retrospective studies, an improvement in GFR [i.e. clear positive estimated glomerular filtration rate (eGFR) slope] was observed in patients with CKD Stages 2–4, those who underwent renal sympathetic denervation and in treated hypertensive CKD patients [14, 16, 17].

Furthermore, continuous interest related to the molecular alterations underlying disease pathogenesis and progression has resulted in the development of several medical treatments [18–20]. Despite the fact that CKD is a life-threatening disease, the biomarkers that are generally employed, such as proteinuria and/or albuminuria, in addition to the high interindividual variability in patients with different CKD stages, can hamper an accurate diagnosis and prognosis of the disease [21]. Therefore, improvements in patient stratification using non-invasive biomarkers are urgently needed in order to guide CKD patient management and to provide targeted and timely intervention [22].

Proteomic studies in the context of CKD have focused on investigating molecular alterations present in those patients with rapid disease progression rather than stable patients. However, our study focused on patients presenting an improved GFR, presuming that an early risk stratification in low-risk patients predisposed to an improvement in renal function may avoid unnecessary medical intervention. Additionally, this approach may provide some insight into possible mechanisms of disease remission and/or regression and potentially highlight further key molecules that are involved in the pathogenic pathways of glomerulosclerosis and renal fibrosis.

Analysis of the urinary peptidome using capillary electrophoresis coupled with mass spectrometry (CE-MS) enables the quantification of thousands of peptides in a single sample and has already been extensively used for discovering and validating biomarkers for CKD [23–25]. In this study, we assessed data of urine samples from patients with different types of renal diseases, considering an increasing GFR as an indicator of disease regression [26]. The primary aim of the study was to identify urinary peptides associated with an improvement in eGFR over time. Our secondary aim was to develop a multidimensional classifier that combines urinary peptides detected by CE-MS in order to identify patients whose renal function will improve over time by means of an increased eGFR.

MATERIALS AND METHODS

Patient selection

The Human Urinary Proteome database [27] was used to select and stratify patients with a baseline eGFR ≥15 mL/min/1.73m². CE-MS data of 553 analysed urine samples, deriving from nine different clinical centres, was included: Austin Health and Northern Health (Melbourne, Australia; n = 19), Steno Diabetes Center (Gentofte, Denmark; n = 245), KU Leuven (Leuven, Belgium; n = 196), University Medical Center (Groningen, The Netherlands; n = 72), University Hospital Essen (Essen, Germany; n = 6), Barbara Davis Center for Childhood (Denver, CO; n = 7), RD Nephrologie (Montpellier, France; n = 4), Charles University (Prague, Czech Republic; n = 2) and RWTH University of Aachen (Aachen, Germany; n = 2).

The study was conducted in accordance with the principles of the Declaration of Helsinki. The protocols for previous studies were approved by the local ethics committee in Skopje (09-1047/15) and all data sets were received anonymously.

Study design

In our cohort, all available SCr values measured for each patient during their follow-up period were included. eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation [28] and the slope/year (mL/min/1.73 m²/year) was calculated by linear regression analysis based on follow-up SCr measurements, with each patient having at least 3 years of follow-up and three data points. The percent change of slope/year (% slope/year) was calculated as a percentage change from baseline eGFR and used as a selection criterion for stratifying patients into two groups. In total, 376 patients with a slope/year between −1.5% and +1.5% were designated as non-progressors or stable, while 177 patients with >5% slope/year were designated as patients with an improved eGFR (renal function). From the total number of stratified patients (n = 553), 67% (n = 118 with improved and n = 251 stable renal function) were randomly used and defined as the development cohort while the remaining 33% (n = 59 with improved and n = 125 stable renal function) were used as the validation cohort (Figure 1).

Statistical methods, definition of biomarkers and sample classification

Values expressed as the mean ±  SD and (two-sided) P-values <0.05 were considered significant. Statistical comparison between the groups was performed using chi-square (χ²) test and Mann–Whitney U test or t-test, as appropriate. The selection of biomarkers was performed by using statistical adjustment testing, employing the method described by Benjamini and Hochberg [29]. Ten permutations were performed, each time randomly excluding 50% of both cases and controls. Peptides that were detectable in >30% of patients and reached an adjusted P <0.05 after false discovery rate (FDR) correction in each of the 10 permutations were considered relevant. Generation of the multidimensional classifier was performed using support vector machine (SVM)-based Modular Open Systems Approach Cluster software [30]. The areas under the curve (AUCs) and all further statistical calculations were performed using MedCalc (version 12.7.5.0, MedCalc Software, Mariaakerke, Belgium; www.medcalc.be (1 August 2017, date last accessed)).

Computational protease prediction

Naturally occurring (endogenous) peptides from urine are products of enzymatic proteolytic cleavages performed by proteases. Prediction of their activity in patients with improved kidney function was calculated by Proteasix bioinformatics software [31]. Generally, proteases previously identified by Proteasix were considered with high confidence based on the proteolysis of the N- or C-termini of the peptides present in the biomarker panel. In addition, protease prediction was evaluated against a probability threshold generated by the MEPROS database, which contains >6000 randomly mapped octapeptide sequences and the frequency of each amino acid at certain positions. Using the intensities of the individual peptide markers from patients with stable and improved kidney function, altered protease activity was estimated. For more in-depth investigation, proteases with at least two cleavage sites and a high-confidence score were considered [32, 33]. Selection of the identified proteases was based on Mann–Whitney test, with adjusted P-values <0.05 when the groups were compared.

RESULTS

The mean age of the selected 553 patients was 49.8 ± 15.4 years, with 283 (51.2%) females. The mean value of baseline urinary albumin excretion (UAE) was 16.3 ± 46.9 mg/day and baseline eGFR was 72.8 ± 13.7 mL/min/1.73 m². The distribution among the CKD Stages was 9.0% Stage 1, 75.4% Stage 2, 14.6% Stage 3 and 0.9% stage 4, according to the Kidney Disease: Improving global Outcomes guidelines of 2012 [34]. There was no significant difference between the groups with regard to baseline age, eGFR and UAE, whereas the stable group consisted of predominantly male and diabetic patients. Their baseline characteristics are shown in Table 1.

Biomarker identification

Statistical analysis of the CE-MS data deriving from the urine samples assigned to the development cohort resulted in the identification of 384 peptides that were significantly different between the stable patients and those with improved renal function. Of these 384 peptides, 141 sequenced peptide fragments were used to generate an SVM-based classifier (Table 2). The compiled contour plots of the urine polypeptides for each patient cohort (i.e. stable versus improvement) are shown in Figure 2. After total cross validation in the development cohort, the biomarker panel was subsequently applied to an independent validation cohort, resulting in an AUC value of 0.85 along with a sensitivity and specificity of 81% (Figure 3A). In addition, subanalysis of the peptide panel was performed using a subgroup of patients with an eGFR <60 mL/min/1.73m² at baseline (n = 32) who were already used for validation. Similarly, classification employing the 141 peptides resulted in an AUC value of 0.80 [95% confidence interval (CI) 0.62–0.91; P = 0.0006], differentiating stable patients from those with improved renal function in CKD Stages 3 and 4 (Supplementary data, Figure S1).

Furthermore, the diagnostic performances of the biomarker panel were compared with the available clinical features (e.g. age, eGFR and UAE; Figure 4). When applied in the same validation cohort, the outcome based on the clinical parameters demonstrated much poorer diagnostic capabilities, with AUC values ranging from 0.56 to 0.59 (Figure 4). In fact, these results underline the importance of developing such a panel of markers and further support its validity as a single-step procedure that could be employed in clinical settings.

Using tandem mass spectrometry (MS/MS), sequence information of the peptides included in the diagnostic biomarker panel was obtained (Supplementary data, Table S1). The majority (78%) belonged to different forms of collagen and, in fact, 66 were downregulated and 44 were upregulated.

Apart from the collagen-related peptides, 31 non-collagen peptides belonging to 25 other proteins were identified. Among these 25 proteins, 7 had already been associated with different renal diseases and CKD progression. In the study presented here, however, they were regulated in an opposing manner (Supplementary data, Table S2).

In silico protease identification

Identification of the proteases responsible for generating the urinary peptide fragments included in our panel was performed based on the available sequence information. The mean peptide intensities in both the groups (e.g. non-progressors and those with improved renal function) revealed 11 significant proteases with altered cleavage activity, with the majority of them belonging to the matrix metalloproteinases (MMPs) group of enzymes. Of note, increased activity of MMP-8, MMP-14, prolyl endopeptidase (PREP), kallikrein 2 (KLK 2) and cathepsin L1 (CTSL 1) and decreased activity of MMP-3, MMP-9, MMP-12, MMP-20, tripeptidyl-peptidase 1 (TPP1) and membrane metallo-endopeptidase (MME) was observed (Table 3).

Comparing performance with the CKD273 biomarker classifier

In order to evaluate whether the identified peptide markers indicated different physiological processes and/or altered molecular changes in patients with improved kidney function compared with CKD progressors, the CKD273 classifier, a urinary-based biomarker panel for the detection and prognosis of CKD [35–37], was applied to our validation cohort (n = 59 with improved and n = 125 stable renal function). This yielded an AUC of 0.62, showing a sensitivity and specificity of 63% and 52%, respectively (Figure 3B). The classifier with 141 peptides was also applied in patients, including those with a −5% slope/year GFR decline (n = 109), that were not used in our study (cohort generated from the Human Urinary Proteome database), and those with an improved GFR, that were used in our validation set (n = 59). This resulted in an AUC of 0.82 (95 CI% 0.75–0.87; P < 0.0001) (Supplementary data, Figure S2).

Furthermore, we compared the peptide fragments from the 141 biomarkers included in our classifier with those from CKD273. Twenty-nine peptides were found to be common to both classifiers (20 with an opposing regulation), with 27 of them being collagen fragments (Supplementary data, Table S3) [37].

DISCUSSION

Optimal management of CKD patients requires early detection and prognostic evaluation of the disease in order to possibly prevent, or at least delay, its progression. On the other hand, an improvement of the kidney function, or a regression of CKD, has been underestimated as a possible clinical outcome and the underlying reasons or predictors for such an event remain unclear. Therefore we investigated the urinary peptidome of >500 patients and developed a prognostic classifier composed of 141 urinary peptides, which may indicate patients predisposed to improved renal function.

One of the crucial findings in our study is that the majority of peptides with significantly altered levels in patients with improving kidney function were collagen fragments. This could possibly indicate molecular changes in the cleavage of extracellular matrix (ECM) during the remodelling processes undertaken by different types of proteases. It was previously shown that the downregulation of different collagen fragments was associated with CKD progression [37], suggesting their accumulation in the intrarenal extracellular matrix. Nevertheless, it should be pointed out that the collagen fragments included in our panel differ from those associated with the process of progression [37]. Thus, while comparing the sequence data, we observed an overlap of only 27 collagen fragments (Supplementary data, Table S2), which may indicate that those patients with improving renal function undergo different physiological processes. As expected from this observation, the CKD273 classifier poorly identified patients with an improved eGFR. In a similar manner, the clinical parameters showed weak performance in the diagnosis of patients with improved kidney function. Therefore the SVM-based classifier (Classifier 141) developed in this study appears specific for improving renal function, which, at a molecular level, is indicative of CKD regression.

Based on the observed differences in peptide intensities in patients with an improved eGFR over time, we hypothesized that these patients undergo molecular changes different from those specific for stable patients or those patients with disease progression. To our knowledge, the enzymatic activity of in silico predicted metalloproteinases (MMP-9, MMP-3, MMP-8, MMP-12, MMP-20, MMP-14) along with TTP1, PREP, KLK2, MME and CTSL differs from that of CKD progressors, where the increased activity of MMP-2, MMP-3, MMP-7, MMP-8, MMP-9 was observed [38]. In addition, proteases such as thrombin, adamts5, plasmin and granzyme A or TTP 1, PREP, KLK 2 and CTSL 1 did not overlap between the studies, suggesting that patients with improved renal function undergo different biological and/or molecular processes compared with those patients with reduced kidney function. Further molecular analyses are required in order to provide a complete insight into this process, but this may hopefully lead to the identification of new molecular targets for CKD remission.

Among the patients with improved renal function, we identified downregulated peptide fragments originating from seven proteins that had been previously reported with a contrary regulation in the context of CKD progression [36, 37] (fibrinogen α chain and haemoglobin subunit β; Ig κ chain C region; plasma protease C1 inhibitor). Peptide fragments of Mimecan were identified in patients with autosomal dominant polycystic kidney disease (ADPKD) [39] and, together with some fragments of plasma protease C1 inhibitor, were used as biomarkers to predict the risk of progression to ESRD in patients with ADPKD [32]. Peptide fragments from haemoglobin subunit β were identified as biomarkers for anti-neutrophil cytoplasmic antibody–associated vasculitis [40] and IgA nephropathy [41], whereas the Ig κ chain C region was found to be increased in the urine of early DN patients [42, 43].

Osteopontin (OPN) is the main effector in direct stone-forming processes, playing an important role in the prevention of calcium oxalate monohydrate kidney stones formation and progressive renal calcification [44]. On the other hand, OPN was associated with a higher risk of subsequent coronary heart disease and calcification of atherosclerotic plaques in patients with CKD [45].

Peptide fragments of α₂-HS-glycoprotein (fetuin-A) were found to be upregulated in CKD patients [46] and associated with CKD progression, demonstrating a negative correlation with GFR reduction [37]. Higher urinary excretion of fetuin-A is a risk factor for both microalbuminuria and a GFR decrease in DN [47]. In patients with DN and diabetic patients that progressed from normo- to microalbuminuria, higher concentrations of this protein were identified, positively correlating with UAE [48]. Several other studies reported that α₂-HS-glycoprotein has been associated with inflammation and tubular damage in patients with diabetes [43].

Notwithstanding this information, the presented data have some potential limitations. We could not evaluate the effect of the therapeutic treatment used in the management of CKD patients, such as angiotensin-converting enzyme inhibitors or angiotensin receptor blockers, that might have been associated with improved kidney function. In addition, the estimation of GFR based on SCr could be biased due to nutritional modifications and changes in body composition. Nevertheless, those factors seemed to be constant among the patients from the African American Study of Kidney Disease and Hypertension group, where an improvement in kidney function was observed [14].

In summary, we have developed a panel of urinary peptide markers that detects patients predisposed to improved kidney function with good accuracy. This could enable the selection of more personalized, less aggressive treatment. In the context of CKD, MS-based proteomics and peptidomics has been implemented as the primary analytical approach, providing novel molecular information regarding disease mechanisms that could be used to better define intervention strategies. In particular, knowing that the proteins associated with impaired kidney function are consequences of significant changes at the proteomic level and understanding their role in disease can enable the identification of biomarkers ideal for patient management. In fact, today almost all drugs are developed based on targeted proteins and a combination of therapeutic targets, and biomarkers identified by proteomic/peptidomic analysis could be a promising strategy for monitoring disease activity and drug response.

Based on these findings, we hypothesize that downregulation of the aforementioned CKD-related peptides in patients with improved renal function indicates the presence of reduced inflammation and renal and/or tubular damage at baseline. Hence our approach, employing a multimarker panel, may be implemented in the routine clinical practice and management of advanced CKD stages, accepting that an improved eGFR may be a possible clinical outcome. Long-term follow-up in a larger patient cohort may support these observations.

SUPPLEMENTARY DATA

Supplementary data are available online at http://ndt.oxfordjournals.org.

AUTHORS’ CONTRIBUTIONS

H.M. and G.S conceived and designed the experiments. K.M. and M.P. performed the experiments and analysed the data. K.M., M.P., C.P., P.Z., L.C., A.S., J.M.S., O.S.T., M.P. and F.M. contributed reagent, materials and/or analysis tools. K.M., M.P., H.M. and G.S. wrote the article.

FUNDING

The research presented in this article was supported by Clinical and System -omics for the Identification of the Molecular Determinants of Established Chronic Kidney Disease (iMODE CKD, PEOPLE-ITN-GA-2013-608332).

CONFLICT OF INTEREST STATEMENT

H.M. is founder and co-owner of Mosaiques Diagnostics, who developed the CE-MS technology. M.P., C.P. and P.Z. are employees of Mosaiques Diagnostics. The remaining authors have no conflict of interests to declare.

REFERENCES

Author notes