2.1 Patients and samples

The 96 KTR patients enrolled in this study are graft recipients in the first-year post-transplantation from two clinical centers in France. The patient cohort from the University Hospital of Strasbourg is divided into two subgroups. The first subgroup consists of 27 patients with various degrees of BK virus-related post-kidney transplantation complications, from whom samples were taken at a single time point after kidney transplantation. The second subgroup from this center comprises 11 KTR patients affected by bone loss due to pre-transplant osteopenia or osteoporosis with samples taken at two different time points: before the initiation and after several months of treatment with bisphosphonates. A third KTR study cohort, composed of 58 patients, is derived from Necker Hospital in Paris, representing a cross-sectional group of patients with (n = 29, eight with antibody-mediated rejections (ABMR), five with T cell-mediated rejections (TCMR), 16 with interstitial fibrosis and tubular atrophy (IFTA)) and without (n = 29) transplantation-related complications.

Bone metabolism is routinely monitored at quarterly intervals during the first post-transplantation year and then once each subsequent year. Monitoring is performed using serum CTX and BAP at the University of Strasbourg, and serum CTX and osteocalcin at Necker Hospital in Paris. Urine samples collected within a time range of less than 15 days from the time of bone metabolism marker assessment were selected from the biobanks of the two hospital centers to perform proteomic analysis by CE-MS. Clinical and demographic characteristics for the KTR patients from the University Hospital of Strasbourg and Necker Hospital in Paris are provided in Table 1.

All patients provided informed consent. The study was approved by the institutional review board of Strasbourg University Hospital under the reference CPP-EST DC-2013-1990 for patients from Strasbourg. For patients from the Necker Hospital the study was approved by the Ethics Committee of Ile-de-France XI (#13016).

Demographic and clinical data was collected from each KTR and their respective kidney allograft donor and included age, sex, height, weight, body surface area (using the Mosteller formula [27]), body mass index (kg/m²).

Spot urine samples were transferred to urine monovettes and immediately frozen without additives at − 20°C. Shipment to the central laboratory performing CE-MS (Mosaiques Diagnostics, Hannover, Germany) was carried out on dry ice to ensure that the samples were thawed only once before CE-MS analysis.

2.2 Immunoassays and laboratory tests

Laboratory markers from the KTR included serum creatinine, estimated glomerular filtration rate (eGFR) using the MDRD formula [28], serum intact PTH (1-84), serum 25(OH) vitamin D, and 1,25(OH)2 vitamin D as determined by the use of standard clinical laboratory methods. For patients from Strasbourg serum CTX-I and BAP were analyzed using the Automated Chemiluminescence Immunoassay (CLIA) (Crosslap, Ostase) on an iSYS automate (IDS-iSYS Multi–Discipline Automated System). In patients from Paris CTX and osteocalcin were analyzed by the Elecsys β-CrossLaps and the N-MID Osteocalcin serum assays respectively (Roche Diagnostics) on the Cobas e411 analyzer.

2.3 Urine sample preparation

Sample preparation for proteomic analysis of urine samples by CE-MS was performed as described elsewhere [23]. Briefly, urine was prepared by 1:2dilution in 2 M urea, 10 mM NH4OH, and 0.02% SDS (pH 10.5). This was followed by centrifugation at 1500 × g through Centrisart centrifugal ultrafiltration units of 20 kDa molecular weight cut-off (Sartorius, Germany) for isolation of the low molecular weight fraction of urine and for clearance of all high molecular weight components including cells, cellular debris and proteins > 20 kDa. A 1.1 mL filtrate was collected and desalted by gravity flow over Sephadex G-25 resin using PD10 columns (GE Healthcare, Sweden) and ultrapure water to remove urea and salts that interfere with CE-MS analysis. Samples were then lyophilized, resuspended in HPLC-grade water to a total peptide concentration of 0.8 μg/μL and a 100 nLaliquot were injected into the CE-MS under constant flow and pressure conditions at a pH of 2.2 to ensure that all peptides are positively charged.

2.4 CE-MS analysis and proteomic data processing

CE-MS sample acquisition was performed using a P/ACE MDQ capillary electrophoresis system (Beckman Coulter, Fullerton, USA) on-line coupled to a Micro-TOF MS (Bruker Daltonic, Bremen, Germany). Prerequisite for coupling of the CE with the MS was that the ESI sprayer (Agilent Technologies, Palo Alto, USA) was grounded, the ion spray interface potential set between −4.0 and −4.5 kV and contact-close-relays were used to control MS data acquisition automatically by the CE. Spectra were accumulated every 3 s over a m/z range of 350–3000.

For raw proteomic data processing, the in-house software MosaiquesVisu [29] was used to deconvolute mass spectral ion peaks representing identical molecules at different charge into single masses. MosaiquesVisu uses both isotopic distribution as well as conjugated masses for charge-state determination of peptides with charge states > 1 and signal-to-noise ratios > 4 observed in a minimum of three consecutive spectra. The software generates a list of peptides in the sample characterized by molecular mass, CE-migration time and ion signal intensity, which were subsequently normalized by global and local linear regression using internal standard peptides with FT-ICR MS-determined exact molecular mass ensuring inter-comparability of CE migration times and molecular masses across measurements. In order to compensate for dilution effects due to different hydration status of the urine samples, the calibration method for peptide signal intensities established by Jantos-Siwy et al. [30] was used. This method is based on the relative quantification of 29 internal collagen peptides of high abundance in human urine. In subsequent studies with large numbers of patient samples and covering a broad range of disease etiologies [31-34], this method has been demonstrated to be better suited than urinary creatinine specifically for comparative analysis of CE-MS peptide profiles. This remains true, even if some of the peptides used for calibration are affected by group differences. The normalized peptide profiles were deposited, matched, and annotated in a Microsoft SQL database, allowing further analysis and comparison of multiple samples. Peptides were considered identical within different samples, when mass deviation was lower than 50 ppm for peptides less than 2000 Da or 75 ppm for larger peptides. After calibration, deviation of migration time was controlled to be < 0.45 min.

2.5 Peptide sequence analysis

For amino acid sequence identification, CTX-associated urinary peptides derived from the CE-MS spectra, based on their MS-detected molecular mass and characteristic migration time in the CE, were searched against Mosaiques’ in-house human urinary peptide database containing sequences from both LC- and CE-MS/MS using a Dionex Ultimate 3000 RSLS nano flow system (Dionex, Camberly, UK) for LC- and a Beckman CE/Orbitrap Q Exactive plus combination for CE-MS/MS (Thermo Scientific, Waltham, MA) [35].

2.6 In silico protease mapping

In silico protease maps were generated using Proteasix (available online at http://proteasix.org/), which allows retrieving protease/cleavage sites (CS) associations based on the MEROPS (https://www.ebi.ac.uk/merops/) and the SWISS-PROT database (http://www.uniprot.org/) [36]. N- and C-terminal cleavage sites were retrieved for the sequenced BR-associated peptides, based on an alignment to the respective substrate sequence information. The association of proteases with their cleavage sites is based on octapeptide consensus sequences. A list of 6000 random CS sequences was used as reference to determine the specificity of prediction against false-positive associations. Only proteases/CS associations which were previously observed or predicted with high/medium confidence level and having at least two cleavage site associations were considered.

2.7 Statistical analysis and definition of peptide markers

Statistical analyses were carried out using the statistical programming language R and the statistical software MedCalc version 12.7.5.0 (MedCalc Software; Mariakerke, Belgium).

Signal intensities of each of the 3555 annotated peptides were correlated with serological laboratory markers using Spearman's rank correlation for univariate analysis. To reduce the risk of false discovery, the patient cohorts from Strasbourg and Paris were studied independently from each other. Statistical adjustment of p-values for multiple testing was performed by the method of Benjamini and Hochberg [37] and by taking into account the zero-inflated distribution of the peptide's intensity values [38].

For the selection of age- and gender-matched normal individuals, as well as age-, gender-, and eGFR-matched CKD patients as reference groups to compare signal intensities of selected peptides, the R-script “MatchIt” [39] was used, which conducts nearest neighbor matching with logistic regression to estimate the propensity score. These urinary peptide reference profiles were from normal individuals and CKD patients of previously described study populations recorded by CE-MS under identical operating conditions and instrument settings [40-42].

3.1 Urinary peptides with association to CTX, BAP, and osteocalcin

In a first step, association of urinary peptides associated with bone metabolism in the setting of kidney transplantation was investigated. For this purpose, we analyzed 85 KTR from Strasbourg and Paris for their individual peptide profiles by CE-MS. The 27 patients from Strasbourg are a highly selective group of KTR patients affected by BK-virus infection, whereas the 58 patients from Paris represent a cross-sectional group of whom 29 are with (8 ABMR, 5 TCMR, 16 IFTA) and 29 are without transplantation-related complications Demographic and clinical data for the two patient groups from Strasbourg and Paris are presented in Table 1. As indicated by this table, both cohorts are comparable to the specific presented clinical and demographic characteristics. For most of the patients, samples were taken within the first 2 years after transplantation. The only differences were higher prevalence of steroid treatment in the Paris group and serum PTH being higher in the Strasbourg group.

Applying a minimum frequency threshold of 30%, 3555 urinary peptides were selected and evaluated in the 85 kidney transplantation patients. The 30% lower limit was selected as this was determined in previous studies using CE-MS [43-45] to be an adequate data foundation and optimal tradeoff to obtain statistically meaningful results in multivariate analysis and to compensate for biological variability between individuals in respect to protein turnover and proteolytic environment, for example, distinct protease profiles and enzyme activities. In previous biomarker studies, the use of 30% threshold enabled establishment of biomarker patterns for different disease conditions [43-45]. The signal intensities of the 3555 annotated peptides in each individual peptide profile were then correlated to the serum levels of the bone metabolism laboratory markers CTX, osteocalcin, and bone alkaline phosphatase (BAP) by Spearman rank sum analysis. As shown in the volcano plot in Figure 1A, in the case of the Strasbourg cohort of 27 patients this resulted in the identification of 92 peptides with significant correlation (FDR < 0.05) to CTX of which 87 (95%) showed positive correlation with Spearman's rho coefficients in the range of 0.58–0.82 and only five (5%) showed inverse correlation with Spearman rho coefficients in the range of −0.59 to −0.69. As shown in the Venn diagram of Figure 1B, 43 (47%) of these 92 peptides were also significantly correlated with serum levels of BAP as second bone metabolism marker. Of note, all 43 peptides were positively correlated to both CTX and BAP. In the case of the Paris group of 58 patients, 590 urinary peptides are significantly associated with CTX. A volcano plot of Spearman's rho values for urinary peptides of the Paris group is presented in Figure 1C. From these, 387 (66%) showed a positive correlation (range of Spearman rho values: 0.34–0.75), whereas 203 (34%) were found inversely regulated (range of Spearman rho values: −0.33 to −0.74). In the case of the Paris group, osteocalcin was available as a second bone marker. As shown in the Venn diagram in Figure 1D, an overlap of 455 (77%) peptides was observed showing a significant correlation in the same direction to both CTX and osteocalcin. Out of these, 293 demonstrated positive and 162 inverse correlations to both bone metabolism markers.

Eighty-two out of the pool of BR-associated urinary peptide candidates were found to be significantly correlated with CTX in both the Strasbourg and Paris patient groups, which represents 89% (82 out of 92) of the total number of CTX-correlated peptides in the Strasbourg group. The 82 peptides are presented in Table S2 by their CE-MS characteristics and group-specific signal distributions and are also labelled with their respective peptide IDs in the volcano plots of Figure 1A for the patients from Strasbourg and in Figure 1C for the patients from Paris. As presented in Figure 2A, congruence between the Spearman Rho correlation factors of CTX and BAP was observed for the Strasbourg cohort. The same applies to the Spearman Rho correlation factors of CTX and osteocalcin in the Paris cohort (Figure 2B). Together this indicates that the 82 selected peptides are not only correlated to CTX, but also to other bone metabolism markers irrespective if these are resorption or formation markers. Conversely, none of these peptides were correlated with serum PTH and creatinine levels, as well as with the estimated glomerular filtration rate, neither in the Strasbourg cohort nor in the Paris cohort.

3.2 Identification of urinary peptides related to bone metabolism

From these 82 CTX-associated urinary peptides, 59 (72%) could be resolved by their amino acid sequence. As evident from Table 1, most of the CTX-associated peptides, namely 46, originate from alpha-1 type I collagen (COL1A1). From the remaining 13 sequenced peptides, nine were derived from other collagen chains (COL1A2, n = 1; COL3A1, n = 2; COL4A1, n = 1; COL4A2, n = 1; COL4A5, n = 2; COL7A1, n = 1; COL8A2, n = 1) and four from the non-collagen proteins cadherin-1 (CADH1), α−1-antitrypsin (A1AT), EMI domain-containing protein 1 (EMID1), and gelsolin (GSN). All COL1A1 peptides show a positive correlation with bone metabolism markers, whereas the COL3A1 peptides represent two out of three peptides that are inversely correlated (with the third not resolved by sequence).

Alignment of the 46 COL1A1peptide fragments to the linear sequence of COL1A1, as shown in Figure 3, indicates that the peptides originate from certain specific regions of the full–length protein. This observation may be interpreted as a sign of regions of increased susceptibility to proteolytic cleavage. To investigate this hypothesis, a search was carried out for observed and in silico predicted endoproteinase-specific cleavage sites in the immediate vicinity of these hot spots. Octapeptide sites matching with the consensus recognition sites of the proteases Cathepsin K (CTSK) and metalloproteinase 9 (MMP9) were identified by in silico-prediction using the software tool Proteasix, as indicated in Figure 3.

3.3 Identification of the CTX-associated marker peptides in plasma

From the 59 out of the 82 urinary bone remodeling markers for which sequence information is available, 36 were also detectable in human plasma [26]. as presented in Table S2. Except for the Gelsolin peptide 605-WVGTGASEAEKTGAQEL-621, all of these peptides are derived from collagen chains and here especially from COL1A1 (n = 30). In contrast, when comparing all 3648 peptide sequences in our urinary sequence database with the 3035 peptide sequences identified so far in plasma samples, an overlap of only 143 peptides were detected which corresponds to 3.9% agreement (143 out of 3648). This limited overlap between peptides in human urine and plasma was reported previously by Magalhães et al. [26] and Parker et al. [46]. Comparing this proportion to the 61% overlap (36 out of 59) in case of the bone remodeling markers, this amounts to a 15-fold enrichment for the bone-specific clearance process that was highly significant (p < 0.000001) in Fisher's exact test.

3.4 Effect of bisphosphonate therapy on the CTX-associated urinary peptides

In order to evaluate the impact of therapeutic interventions against bone loss on the 82 CTX-associated urinary peptides, the peptides abundance before and after bisphosphonate treatment was determined in a separate group of 11 KTR recruited from the University Hospital of Strasbourg. For bisphosphonate therapy, all but one patient received 70 mg oral alendronic acid once a week, whereas the remaining patient received 35 mg oral risedronate once a week. From the set of 82 peptides, 17 showed a significant change after bisphosphonate treatment, consistently a decrease, indicating bisphosphonate treatment adherence and possibly reduced bone resorption following treatment (Table 2).

We compared the excretion levels of the 17 bisphosphonate-sensitive peptides of our KTR patient groups with those found in age- and gender-matched hindividuals, as well as age-, gender-, and eGFR-matched chronic kidney disease patients. Since we were unable to match normal individuals for eGFR aswell, we included CKD patients into this comparative analysis in order to investigate the impact of eGFR on the peptide's excretion levels. As presented in Figure 4, overall excretion of these 17 peptides is significantly higher in KTR patients not treated with bisphosphonate compared to normal individuals but also CKD patients. After bisphosphonate therapy the regulation direction is reversed, now being significantly lower in bisphosphonate treated KTR patients compared to both the normal individuals and CKD patient reference groups.

3.5 Involvement of CTX-associated urinary peptides in CKD

To investigate if the 82 CTX-associated urinary peptides are also associated with CKD, these peptides were compared with the peptides (N = 273) contained in the CKD273 urinary classifier, which was developed on a large set of CKD patients and disease-matched controls using the same CE-MS peptide annotation database [23]. By this comparison it was found that 17 of the bone metabolism–related markers are also included in the CKD273 model. The expected chance overlap was calculated to be 2.3%, since 82 differentially regulated peptides were identified in a data space of 3555 urinary peptides. Correlating this to the observed overlap of 6.2% (17 out of 273) yielded a significant difference compared to the random condition (p = 0.001, Fisher's exact test).