Urinary proteomic analysis of chronic allograft nephropathy

2.1 Patients and urine samples

Patients were recruited at time of transplantation at Westchester Medical Center and followed prospectively from December 2002 to October 2005. Fresh mid-stream urine samples were collected at routine clinic visits and at the time of biopsy. Urine samples were also collected from 20 healthy volunteers. Urine was stored on ice, and multiple aliquots frozen at −80°C within 4 h of voiding. The urinary proteome has been shown to be stable during that period by Schaub et al. [8]. Samples were thawed just once prior to the analysis as repeated freeze/thawing has been shown to destroy certain proteins [9]. No protease inhibitor was used. Contemporaneous clinical data, including blood pressure and serum creatinine, were abstracted from the patient record. All patients completed a written consent form permitting both urine and data collection in compliance with HIPAA (Health Insurance Portability and Accountability Act). The Institutional Review Board for clinical trials involving human subjects approved the study protocol.

Immunosuppressive protocols were tailored to individual patients as deemed clinically appropriate but usually included a calcineurin inhibitor (tacrolimus (trough level 5–10 ng/mL) or cyclosporin (trough level 50–150 ng/mL)), rapamycin (trough level 10–15 ng/mL)) mycophenolate mofetil (1–2 g/day), and prednisone tapered from 120 to 10 mg/day by 4 wks post-transplant.

2.2 Sample processing and MS

Sample processing and analysis were performed as previously described [10]. In brief, samples were denatured using CHAPS and Urea in a Tris buffer. The denatured sample was then incubated with chip-specific binding buffer (Table 1) for 1 h, washed, and deposited on the chip surface with a Sinapinic acid matrix. SELDI-TOF MS was performed using four different chip types (CM10, a weak cation exchange surface, Q10 a strong anion exchange surface, IMAC 30, a metal affinity binding surface, and H50 which binds hydrophobic proteins). Analysis was performed on a PBS II ProteinChip^™ reader (Ciphergen^®). The read conditions were: detector sensitivity = 7, laser shots per spot = 130, detector voltage = 2850. Calibration was performed using all-in-one-peptide and all-in-one-protein standards (Ciphergen). Two reference samples were used per chip for spectral quality control [11]. Samples were batch analyzed using a filter setting of 0.2 times the expected peak and a peak signal ≥4 times the noise. Baseline subtraction was performed on all spectra and all data were TIC normalized.

Protein purification and identification were performed as follows. One mililiter of urine was dialyzed in a 2 L bath of deionized water for 48 h at 4°C. Sample was concentrated in PEG 20 000 in a 3500Mr cutoff dialysis membrane (Spectropor). The concentrated protein mixture was then run on a 10–20% gradient Tricine gel (Invitrogen) at 120 V for 1 h. Staining was performed using CBB, the protein band identified and excised. Gel bands were destained and in-gel trypsinolysis performed. Briefly, gel bands were reduced by 10 mM DTT and alkylated by 50 mM iodoacetamide. Then gel bands were washed three times with 50:50 water/ACN. After dehydration by ACN, trypsin was added (Promega) and digestion performed at 37°C overnight. Peptides were extracted by 50:50:0.1 water/ACN/trifluroacetic acid, concentrated, and subjected to subsequent LC-MS analysis. Capillary LC-MS/MS (μLC-MS/MS) was used to obtain protein ID information. The experiments were performed on an 1100 series LC/MSD XCT plus IT mass spectrometer (Agilent). The mobile phases were 0.1% formic acid (solvent A) and 0.1% formic acid in 100% ACN (solvent B). An 8 μL aliquot of peptide mixture was injected by an autosampler onto a 0.3 × mm² Zorbax 300SB-C18, 5 μM particle size loading column at a flow rate of 10 μL/min and subsequently separated on 0.075 × mm² Zorbax 300SB-C18, 3.5 μM particle size analytical column at a flow rate of 0.3 μL/min with a gradient of 15–40% solvent B for 50 min and 40–85% solvent B for 30 min. Mass spectra were acquired in automated data dependent MS/MS mode, consisting of a precursor MS scan followed by MS/MS scans on the four most intensive ions. Fragment amplitude was 1.15 V, scan rate was 13 000 m/z/s, capillary voltage is 3500 V, and skimmer voltage was 40 V.

The raw MS/MS spectra were used to identify peptides by database search using Spectrum Mill software (Agilent). Raw data were processed first to extract MS/MS spectra that can be assigned to at least two b or y ion series, and these “good” spectra were searched against the Swiss-Prot database. The important parameters for searching are: minimum matched peak intensity of 50%, precursor mass tolerance of 2.0 Da and product mass tolerance of 0.6 Da. The number of maximum missed cleavages allowed was two and no peptide modification was considered. The threshold used for the identification is a Spectrum Mill score greater than 10 and a SPI % (the percentage of assigned spectrum intensity out of total spectrum intensity) greater than 70%. All matched MS/MS spectra were manually inspected to ensure that the majority of prominent peaks could be assigned to y/b ions, and sufficient number of positive residues to support multiple charged ions.

2.3 Classification algorithms and statistical analysis

A multiple tree-based classification algorithm, randomForest, permits good visualization of SELDI data, and was used as an initial tool [10, 12]. Briefly, this classifies samples using the intensities of a randomly selected, user defined, number of protein peaks (39 were used in this analysis) to create the branches of a tree with each training sample classifying the last branch. The “out of bag” test set (35%) is then passed down the tree with the last branch deciding its classification. Many trees are created (10 000 in this analysis) with each tree voting for a sample to belong to a particular diagnostic group. This was combined with a different form of classification algorithm, Adaboost, a technique that has also been previously shown to be effective for the classification of SELDI analysis, the purpose being to ensure thorough examination of this complex, high dimensional data [13]. This is a method for combining weak classifiers to create a summary and stronger classifier. The basic principle is that after initial selection of the variable most likely to correctly predict the class of a sample, the samples are reweighted with increase in weight applied to the misclassified samples, the next best classifier variable is then selected. This process is then repeated with summation of the classifiers to create a robust and accurate classifier. An out of bag test set of 35% was used for the Adaboost classification. This classification algorithm required similar group sizes to work well. Ten patients in the Banff grade 0 group had a small degree of interstitial fibrosis and tubular atrophy (up to 5% percent is permissible under the classification). Therefore, 13 of the BANFF group 0 patients were used in the classification algorithm with equal sized random subsamples being selected from the larger patient groups. It is important to note that both of these classification methodologies are distribution free. Variable distribution was assessed by kurtosis, skewness, and the Kolmogorov–Smirnov test. Between groups analysis used either the Student’s t-test or one-way analysis of variance (ANOVA) for parametric distributions or Mann–Whitney-U test for nonparametric distributions. Correlations were assessed using Kendall’s Tau (τ) for nonparametric distributions. Differences were considered significant with p<0.05. Analysis was performed using the statistical package for the Social Sciences; randomForest; and Adaboost are available from the R Network for Statistical Computing (www.r-project.org, as packages “RandomForest” and “gbm”).

2.4 ELISA assay for estimation of urinary concentration of endorepellin

High-throughput proteomic techniques such as SELDI are not quantitative. Therefore, a recognized quantitative technique, ELISA, was used to confirm differences in urinary endorepellin between patients groups. An ELISA plate was coated with 1 ug/mL of monoclonal antihuman endorepellin antibody (R&D systems), diluted in PBS and was incubated overnight at 4°C. Blocking was then performed with 1% BSA for 1 h. Urine samples were diluted 20 times with dilution buffer (0.7% BSA in PBS) and 100 uL of each sample and recombinant endorepellin controls were bound to the plate in duplicate for 1 h at room temperature.

Wells were washed, and 100 uL of 0.8 ug/mL of antihuman goat polyclonal antibody to endorepellin (R&D systems) were incubated for 1 h. Washing was repeated and wells incubated with donkey antigoat HRP conjugated antibody for 40 min.

Color development was with Sure-Blue/TMB (KPL) and reaction was stopped with TMB Stop Solution. Absorbance was read at 450 nm. The urinary endorepellin concentration was calculated from a standard curve. Urinary creatinine was measured with a creatinine assay kit from Cayman. Recombinant Endorepellin was gift from Dr. Iozzo.

2.5 Morphologic and immunofluorescence studies

Seventy-five patients underwent renal transplant biopsy due to a decline in renal function and/or emergence of proteinuria (>1 g/day). All biopsies were evaluated prospectively according to the Banff classification by an experienced renal transplant pathologist (PNC).

Perlecan and endorepellin staining were performed on native renal biopsies from “control” patients (n = 15), which included patients with hypertensive nephropathy (n = 2), FSGS (n = 3), diabetic nephropathy (n = 2), membranous nephropathy (n = 2) and others (n = 6), patients with acute rejection (n = 18), and patients with CAN of severity Banff grade II or III (n = 16). Frozen sections from human renal biopsies were washed with PBS and blocked with 20% goat serum (Sigma) for 1 h at room temperature. After washing in PBS the sections were incubated with a polyclonal rabbit antiperlecan domain V (endorepellin) antibody at a dilution of 1/250 and monoclonal mouse antiperlecan domain III antibody (Zymed), dilution 1/50, at 4°C overnight [14]. Sections were washed three times with PBS before incubation with AlexaFluor^™ 594 goat antirabbit antibody (Molecular Probes) and FITC goat antimouse antibody (Jackson Immunoresearch) for 1 h. Sections were then rinsed with PBS and coverslipped. Quantitative image analysis was performed by digital imaging of the entire section using a TE2000-U microscope (Nikon) equipped with a CCD digital camera (Diagnostic Instruments) and appropriate dichroic mirrors to detect red and green fluorescence and quantifying the intensity of red, green, and yellow (overlap of green and red) fluorescence using the pixel counting function in Photoshop^™(Adobe). Image collection and quantification were performed by an investigator to whom the tissue diagnosis was unknown to exclude observer bias.

3.1 Demographics and clinical parameters

Seventy-five patients and 20 healthy volunteers are included in this analysis. All had proteomic analysis of urine and the patient group had detailed histological analysis of biopsy material. The patient demographics are outlined in Table 2. There was an excess of males (male/female ratio 2.6:1) probably due to the increased risk of CAN in male patients [15].

Serum creatinine increased with rising Banff score as expected. The correlation (Kendall’s Tau-b (τ)) with the Banff score was strongest for the creatinine level measured at last follow-up, a finding congruent with the previously described correlation between graft outcome and histological findings [16, 17]. Correlations were as follows; creatinine nadir after transplantation, τ = 0.25 (p<0.01), creatinine at biopsy, τ = 0.24 (p<0.01), and creatinine at last follow-up, τ = 0.35, (p<0.001). Serum creatinine was higher at the time of biopsy 3.1 ± 1.7 (biopsies were performed for clinical reasons) and at last follow-up 3.3 ± 2.2 versus nadir after transplantation, 1.9 ± 1.0 for all patients. Acute and chronic Banff scores for the patient population are detailed in Table 3. The histological diagnosis for graft dysfunction in the Banff CAN = 0 group included, drug toxicity (n = 8), acute tubular necrosis (n = 6), acute rejection (n = 5), thrombotic microangiopathy (n = 2) focal infarction (n = 1), and diabetic glomerulosclerosis (n = 1). Blood pressure did not differ significantly between the groups (p = 0.78) (KS = 0.69, ANOVA) although there was a trend toward higher values in the patients with CAN. Mean patient time from transplantation to biopsy averaged 612 ± 910 days while time from sampling to last follow-up was 203 ± 176 days. Collectively, these data are comparable to other reported series (18–20) and provide the background to the analyses performed.

3.2 Proteomic data visualization

A total of 1222 peaks were detected on the four chip surfaces. There were 381 peaks on the CM10 surface, 347 on H50, 313 on IMAC30, and 181 on Q10. Initially, the distribution of the patients was assessed by multidimensional scaling (MDS) plots of urinary proteomic data created by the randomForest proximity matrix (Fig. 1). The urine samples of healthy subjects form a tight cluster clearly distinct from the urine of transplant patients. There is good separation of those with CAN Banff grade 0 versus those with CAN Banff grades 1–3, suggesting there is sufficient structure within the data to identify protein peaks that distinguish the two patient groups.

3.3 Data interrogation using the classification algorithm Adaboost

This section outlines the classification accuracy of the urinary proteome when applied to transplant recipients with and without CAN. Analysis of the urinary proteome of those with normal graft histology (Banff = 0, n = 13) versus an equal sized sample of randomly selected patients from both the Banff grade 2 and Banff grade 3 subgroups of patients performed well and is outlined in Table 4. The corresponding bar charts depicting the most important protein peaks with the length of the bars representing the degree of significance are shown in Fig. 2.

3.4 Protein identification

Proteins of masses ~9.9, ~13.9, ~20.9, and ~43.4.0 kDa were selected based on the their utility in distinguishing patients with CAN Banff grade 0 versus CAN Banff grade II or III (Figs. 2a and b). The reasons for the selection of these masses included their appearance in both lists and the exclusion of molecular masses that are either difficult to identify using tryptic digest PMF (small molecular masses), or those likely to belong to proteins that are abundant in urine and are elevated nonspecifically in disease, e.g., albumin ~66 kDa. Gel views of the SELDI data are provided in Supporting Information. Protein identification was performed using MS/MS, as detailed in Section 2. The identified proteins were β2-micro-globulin, prealbumin, α-2 micro-globulin/bikunin precursor (AMBP), and a C-terminal perlecan fragment, as detailed in Table 5. As the first three candidates lacked specificity for, or had been previously identified as markers of CAN, we focused attention on perlecan, a ubiquitous basement membrane heparan sulfate [18]. This candidate seemed to be especially relevant due to the recent demonstration that the C-terminal domain V, endorepellin, has antiangiogenic properties [19]. Therefore, the abundance and distribution of perlecan and endorepellin were further investigated to determine if changes in tissue expression paralleled the proteomic findings.

3.5 ELISA detection of endorepellin

A quantitative analysis of urinary endorepellin was required to confirm this as a novel biomarker of CAN. An ELISA was developed and urinary endorepellin measured in samples from healthy volunteers (n = 8) and patients with either Banff grade 0 or 1 (n = 19) and patients with Banff grade 2 or 3 (n = 17). There was no statistical difference between mean urinary endorepellin/creatinine ratios for healthy volunteers (0.0026 ± 0.0017) and renal transplant recipients with CAN grades 0 and 1 (0.00545 ± 0.005) (p = 0.283, Mann–Whitney) However, there was significantly increased levels of urinary endorepellin in those with more advanced grades of CAN (Banff grades 2 and 3) (0.016 ± 0.014) versus those with normal histology or low grade CAN (Banff grades 0 and 1) (0.00545 ± 0.005) (p<0.01, Mann–Whitney) (Fig. 3). Urinary endorepellin concentration did not correlate with serum creatinine (p = 0.7, Kendall’s tau). The sensitivity, specificity, positive, and negative predictive values were respectively, 65, 79, 73, and 71%. The area under the curve for the receiver operator characteristic curve is 0.774 (Supporting Information).

3.6 Tissue expression

Tissue expression of perlecan and its C-terminal fragment were determined using antibodies raised against domain III (whole molecule) and the domain V (endorepellin). Representative images are shown in Fig. 4. The results showed a remarkable overlap between perlecan and endorepellin staining in control biopsies without CAN. In contrast there was a significant increase in the ratio of domain V staining to domain III staining (red to green) in patients with CAN compared with patients experiencing acute rejection, or disease control (p<0.01, Kruskal–Wallis) (Fig. 4). The linear staining pattern observed in uncomplicated control biopsies was changed in CAN, as it was in acute rejection: staining has become punctated and usurated. In CAN biopsies, endorepellin distribution exceeded that of the perlecan parent molecule, suggesting that the former was diffusing from the basement membranes (Fig. 5).