Proteomic Analysis of Urine in Kidney Transplant Patients with BK Virus Nephropathy
Timo Jahnukainen*,
David Malehorn,,
Mai Sun,,
James Lyons-Weiler,
William Bigbee,,
Gaurav Gupta*,
Ron Shapiro||,
Parmjeet Singh Randhawa¶,
Richard Pelikan**,
Milos Hauskrecht,** and
Abhay Vats*
* Department of Pediatrics, Children’s Hospital of Pittsburgh; Clinical Proteomics Facility; University of Pittsburgh Cancer Institute, University of Pittsburgh; || Department of Surgery, University of Pittsburgh School of Medicine, ¶ Department of Pathology, University of Pittsburgh Medical Center; ** Department of Computer Science; and Department of Pathology, Cancer Biomarkers Laboratory, Center for Pathology Informatics, Benedum Oncology Informatics Center, University of Pittsburgh, Pittsburgh, Pennsylvania
Address correspondence to: Dr. Abhay Vats, Department of Pediatrics, Division of Pediatric Nephrology, Children’s Hospital of Pittsburgh, 3705 Fifth Avenue, Pittsburgh, PA 15213. Phone: 412-692-6120; Fax: 412-692-8569; abhay.vats{at}chp.edu
Received for publication May 5, 2006.
Accepted for publication September 11, 2006.
The differentiation of BK virus–associated renal allograftnephropathy (BKVAN) from acute allograft rejection (AR) in renaltransplant recipients is an important clinical problem becausethe treatment can be diametrically opposite for the two conditions.The aim of this discovery-phase biomarker development studywas to examine feasibility of developing a noninvasive methodto differentiate BKVAN from AR. Surface-enhanced laser desorption/ionization(SELDI) time-of-flight mass spectrometry analysis was used tocompare proteomic profiles of urine samples of 21 patients withBKVAN, 28 patients with AR (Banff Ia to IIb), and 29 patientswith stable graft function. SELDI analysis showed proteomicprofiles that were significantly different in the BKVAN groupversus the AR and stable transplant groups. Peaks that correspondedto m/z values of 5.872, 11.311, 11.929, 12.727, and 13.349 kDwere significantly higher in patients with BKVAN. Bioinformaticsanalyses allowed distinction of profiles of patients with BKVANfrom patients with AR and stable patients. SELDI profiles alsoshowed a high degree of reproducibility. Proteomic analysisof urine may offer a noninvasive way to differentiate BKVANfrom AR in clinical practice. The identification of individualproteomic peaks can improve further the clinical utility ofthis screening method.
BK virus (BKV) infection is common in the general population(1,2). The importance of BKV infection in healthy individualsis unclear, but it has emerged to be a significant problem inimmunocompromised patients, such as bone marrow transplant patientsand patients with solid-organ allograft (1–3). BKV renalallograft nephropathy (BKVAN) now is recognized to have an importantrole in development of renal allograft dysfunction (3–6).In recent studies, the prevalence of asymptomatic BK viruriain kidney transplant patients has been reported to be on theorder of 30% (7). Approximately 6 to 10% of these patients developBKVAN, and the reported graft loss rate in this group has beenas high as 50% (6,7).
BKVAN can resemble acute allograft rejection (AR) both clinicallyand histologically (3–5). Differentiation between BKVANand AR is important, however, because the treatment is diametricallyopposite for the two conditions. In general, immunosuppressionneeds to be reduced in patients with BKVAN, whereas it is increasedin AR. Currently, these two clinical conditions cannot be differentiatedin a reliable way on the basis of clinical and laboratory findings,and a definitive diagnosis of BKVAN requires allograft biopsy.Even the histologic differentiation of BKVAN from AR can bedifficult unless viral inclusions are seen on allograft biopsy(4,5).
Early detection of patients with increased risk for BKVAN islikely to improve their ultimate outcome. However, currently,no noninvasive methods are available for this purpose. Duringthe past few years, proteomic profiling of blood and urine sampleshas been used to develop noninvasive biomarkers for severalpathologic states, including various cancers (8–12). Recently,such techniques also have been used in developing diagnosticalgorithms for AR in kidney transplant patients (13–17).In this study, we report that proteomic analysis of urine potentiallymay be a noninvasive way to differentiate patients with BKVANfrom patients with stable graft function or AR.
These studies were reviewed and approved by Institutional ReviewBoards of the University of Pittsburgh Medical Center and Children’sHospital of Pittsburgh.
Patients and Sample Collection
Urine and blood samples were collected from all kidney transplantpatients at the University of Pittsburgh Medical Center whoconsented. As a part of the study protocol, the initial sampleswere obtained within 48 h after transplantation and thereafterduring each routine follow-up visit. Patients who were undergoingkidney biopsy also provided a voided urine sample <24 h beforethe biopsy. One portion of each sample was sent for routineexamination, and an aliquot was centrifuged at 2000 rpm for10 min at 4°C and stored at –80°C until furtheranalyses without any protease inhibitors.
Kidney transplant patients with biopsy-proven BKVAN and theirclinical information were retrieved from an electronic database.Urine samples were available from 21 of these patients. Furthersearches were performed to identify the two comparison groupsof 28 patients with AR and 29 patients with stable graft function.Acute and chronic AR was classified according to the Banff classification(18).
Detection of Viral Infections
Each study patient was screened for BKV viruria and viremiausing a quantitative PCR assay (19). BKVAN was diagnosed byhistopathologic examination, which typically showed viral inclusionsand positive immunohistochemistry and/or in situ hybridization(19). Urine samples additionally were screened for cytomegalovirus(CMV), adenovirus, and human herpesvirus 6 (HHV-6) using PCRassays. CMV screening also was performed by detection of pp65antigenemia.
Surface-Enhanced Laser Desorption/Ionization Time-of-Flight Mass Spectrometry Analyses
Urine samples were thawed and centrifuged at 15,000 x g for5 min at 4°C. Protein content was measured using the Bradfordassay. Urine sample aliquots totaling 10 µg of proteinwere dispensed, adjust to 160 µl of final volume withPBS, then flash-frozen again and stored at –80°C beforeuse on surface-enhanced laser desorption/ionization (SELDI)chips. Samples were analyzed using two chip types (IMAC30 andCM10) on three replicate spots, with total protein of roughly1.7 µg per spot.
The IMAC30 ProteinChip arrays were preactivated using CiphergenBioprocessor (Ciphergen Biosystems, Fremont, CA) by loadingwith 50 µl of 100 mM CuSO4 on each spot. The chips wereshaken on a tube mixer (Tomy Seiko Co., Ltd., Tokyo, Japan)for 5 min (speed form 20, amplitude 7). Each array spot wasrinsed with 200 µl of HPLC-grade water and aspirated beforeaddition of 50 µl of sodium acetate (100 mM, pH 4). Thechips then were equilibrated twice for 5 min with 200 µlof binding buffer (PBS + 0.2 M NaCl). For preactivation of CM10ProteinChip, the arrays were equilibrated for 5 min with 200µl of binding buffer (100 mM sodium acetate, pH 4).
The urine samples were thawed on ice and denatured with 60 µlof 9 M Urea/4% CHAPS in 40 mM Tris (pH 9) and then further diluted1:5 in IMAC30 and CM10 binding buffers. The samples were arrayedin a blinded layout of combined case/control samples, togetherwith a standard pooled sample (one spot on each array for qualityassurance purposes). Arrays were incubated overnight with shakingat 4°C, then rinsed twice for 5 min with binding buffer,then twice for 1 min with water, and air-dried for 20 min. Energy-absorbingmatrix (sinapinic acid) in 50% acetonitrile/0.5% trifluoroaceticacid was applied to arrays using automated rapid spotting onBiomek2000 robotic workstation (Beckman-Coulter, Fullerton,CA) in two 1-µl applications, separated by 5 min. Chipswere air-dried for 1 h and stored in the dark at room temperatureuntil SELDI analysis.
The reacted ProteinChip Arrays were analyzed using the PBSIIcChipReader mass spectrometer. The spotted chips were read, inrandom order, in an uninterrupted run using the Ciphergen ChipReaderAutoLoader device. The SELDI time-of-flight mass spectrometry(SELDI-TOF-MS) spectra (0 to 100 kD) were collected by the accumulationand averaging of 192 laser shots from 16 positions across thewidth of the ProteinChip Array spot. A laser intensity of 180was used in a positive ion mode, ensuring that transient shotintensities were below saturation of the detector, with a detectorsensitivity setting of 9 (2900 V) and a focus lag time of 900ns, using mass deflection at 1500 Da. The protein masses werecalibrated externally using the 7-in-1 purified peptide molecularmass standard (Ciphergen Biosystems).
Data Processing and Statistical Analysis
The spectra that were generated by SELDI-MS-TOF were analyzedusing two independent software tools: (1) Ciphergen ProteinChipwith CiphergenExpress and (2) proteomic data analysis package(PDAP). PDAP is a data analysis program developed at the ComputerScience department of the University of Pittsburgh and implementedin MATLAB (MathWorks, Natick, MA) (20).
Before analyses, all spectra were preprocessed using the followingpreprocessing steps: Variance stabilization, baseline correction,calibration/rescaling, smoothing, and profile alignment. Spectrawith the total ion current normalization factor of >2 SDover the mean were considered bad spots and excluded from thesubsequent analyses. After preprocessing, profile peaks wereidentified. All subsequent data analysis was performed on thesepeaks.
Univariate Analysis
The three patient groups were compared for demographic and otherclinical characteristics using the t test for continuous variablesand 2 test for categorical variables. For the SELDI data, CiphergenProteinChip software was used to detect peaks, and the correspondingpeak intensities were exported to Microsoft Excel. The significanceof differentially expressed peaks among three groups was studiedusing the Kruskal-Wallis one-way ANOVA on ranks. Pairwise comparisonsbetween groups that were found to be significantly differenton Kruskal-Wallis test were done using t test (SPSS 13.0 statisticalsoftware; SPSS, Chicago, IL).
Multivariate Analysis
The discriminative potential of peak combinations in proteomicprofiles was analyzed using multivariate statistical predictionmodels. Three prediction tasks were considered in multivariateanalyses: (1) BKVAN versus AR, (2) BKVAN versus stable transplant,and (3) AR versus stable transplant. Evaluation of predictionperformance was conducted using repeated random subsamplingof the data in the study into multiple training and test subsets(21).
Statistical machine learning algorithms were used to evaluatethe ability of profile features (peaks) to discriminate betweensamples that belonged to the various patient groups. Three classificationalgorithms were examined for potential disease prediction models:(1) Support vector machine (SVM) (22), (2) Classification andRegression Tree (CART) (23), and (3) Random Forest (RF) (24).All three models were trained on peaks that were identifiedby the PDAP program (20). In addition, the SVM algorithm wasapplied to the subset of the top 100 differential peaks accordingto the P value of the t test (25).
Each learning model was evaluated by dividing the data multipletimes into training and test sets using the repeated randomsubsampling approach (21) with 70:30 training/testing splits.The classification model was always learned on the trainingset and evaluated on the test set. The performance statisticsthat were used to evaluate the learning models were AchievedClassification Error (ACE), sensitivity, and specificity ofthe model on test sets. The performance statistics of each modelwere obtained by averaging the results over multiple train/testdivisions. The CART model in the Ciphergen-based software wasevaluated on three training/testing splits. All other classificationmethods were evaluated on 40 different training/testing datasplits.
Given the small size of the study samples, an analysis was conductedto test whether the classification results might represent afinding that may occur by chance (26). The analyses, calledPermutation Achieved Classification Error (PACE), evaluateswhether the classification performance that is observed on thetrue data could be achievable on randomly regrouped data (27).Briefly, PACE computes the empirical distribution of classificationerrors under null random conditions and compares it with theerror on the original data. In each permutation, a sample israndomly reassigned to one of the disease categories (BKVAN,AR, or control). For example, the sample (and its profile) thatbelonged to a patient with BKVAN was randomly reassigned toeither the AR, the control, or the BKVAN group. This processwas repeated for each sample until all 78 samples were reassignedrandomly into three groups. After that, the entire model buildingand evaluation, with multiple train/test iterations, is performedby PACE. This is repeated for 1000 permutations to define thedistribution of errors. The results of PACE are expressed interms of the mean of the classification error distribution (MACE)along with the 95th and 99th percentiles of this distribution.The difference between error distribution and the ACE is representativeof the nonrandom nature of the results that are obtained bythe various disease prediction modeling algorithms.
Patient Characteristics
The clinical and demographic characteristics of the variousgroups are shown in Table 1. All of the patients in the stabletransplant group had stable graft function at least 6 wk beforeand after sample collection. The time from transplantation tosample collection was significantly shorter in stable transplantpatients compared with patients with AR (P < 0.05). Noneof these patients had a history of delayed graft function. Patientswith AR and BKVAN had significantly higher mean serum creatininelevels (P < 0.05) than stable transplant patients (Table 1).Otherwise, there were no significant demographic differencesbetween study groups (Table 1).
The histologic findings in the AR group showed acute rejectionscores of IA to IIB. Also, 16 (57%) of the patients with ARhad low-grade (IA) chronic rejection changes in the allograftbiopsy. In the BKVAN group, 15 (71%) of the patients were consideredto have changes that would be compatible with grade IA to IIAacute rejection scores, and a similar proportion also showedfeatures of chronic rejection. The acute and chronic rejectionscores for patients with AR and BKVAN are shown in Table 2.Urinary viral PCR studies showed BKV load that ranged from 30,500copies to >2 x 1010 copies/ml in all of the BKVAN urine samples.HHV-6 PCR was positive in two patients in the stable transplantgroup and in one patient each in both AR and BKVAN groups. Onepatient with AR also had simultaneously positive adenovirus,HHV-6, and CMV PCR tests, whereas two patients were positivefor simultaneous adenovirus and CMV. In the BKVAN group, therewere two patients with positive adenovirus PCR in urine.
Characteristics of Urine Proteome Analyzed by SELDI-TOF-MS
The Ciphergen peak detection software detected a total of 158peak clusters. Eighty of these peaks were found in IMAC30 data,and 78 were found in the CM10 data. We assayed 27 samples intriplicate in two sessions separated by 3 mo (a total of sixprofiles for each of the 27 samples for a total of 162 profiles).The profiles were reproducible to a very significant extentdespite the relatively low protein amount (<2 µg) reactedwith each ProteinChip spot and a gap of 3 mo (Figure 1).
Figure 1. Surface-enhanced laser desorption/ionization (SELDI) analysis of urine samples from patients with BK virus–associated renal allograft nephropathy (BKVAN; A), patients with allograft rejection (AR; B), and patients with stable graft function (C) analyzed in triplicate (1 through 3) at onset of study (O) and the same samples analyzed on separate experiment 3 mo later (+3). Proteomic profile of the samples is shown as peak intensities (I) and as false-gel image (II).
Visual analysis of the urinary profile data was done using Ciphergensoftware to identify peaks that were different between the studygroups. Urine samples of stable transplant patients showed peakswith significantly higher peak intensities at the m/z valuesof 4.755, 6.245, 6.440, 7.672, 8.012, 8.230, 9.636, 9.870, 10.067,10.569, and 16.918 kD compared with the samples of patientswith AR or BKVAN on the analyses of both chip data. Figure 2shows an overview of the SELDI profiles of the three groupsin false gel views, and Table 3 shows further details of thecharacteristics of some of the differentially expressed peaks.A profile with a combination of four closely clustered peaks(IMAC30) between 6.072 and 6.440 kD together with separate peaksat 9.870 and 10.569 kD was found to be relatively constant andwas present in the urine samples of 25 (86%) of the stable transplantpatients (Figure 2B, frames 1 and 3). This peak pattern wasseen in a significantly smaller subset of 35% of the patientswith AR and 24% of the patients with BKVAN. The only peak thatwas seen more frequently and with a higher mean peak intensityin the AR group than in the other two groups was a peak locatedat m/z value of 8.854 kD in the IMAC30 data set (Figure 2B,frame 2, Table 3).
Figure 2. False-gel image of urine protein profiles corresponding to m/z 5 to 25 kD using the CM10 chip data set (A) and the IMAC30 chip data set (B). The framed m/z areas show examples of regions with significant differences between the various study groups. (A) Frame 1 = 5.872 kD (upregulated in BKVAN), frame 2 = 12.727 to 13.349 (upregulated in BKVAN), and frame 3 = 16.918 kD (upregulated in stable). (B) Frame 1 = 6.072 to 6.440 kD (upregulated in stable), frame 2 = 8.854 kD (upregulated in AR), frame 3 = 9.636 to 10.569 kD, and frame 4 = 23.482 kD (upregulated in BKVAN).
Table 3. Mean peak intensities of the most significant peaks differentiating BKVAN from other patient groups
Univariate analyses indicated that the most significant peaks(m/z ratio [chip type]) that differentiated BKVAN samples fromthe other two groups were 5.872 (CM10, IMAC30), 11.311 (CM10,IMAC30), 11.929 (CM10), 12.727 (CM10), and 13.349 (CM10, IMAC30)kD peaks. All of these peaks were higher in patients with BKVANand were seen more frequently in this group (Figures 2 and 3,Table 3). The analysis of IMAC30 data also showed significantlyhigher peak intensity in BKVAN versus stable patients for aprotein with the m/z value of 23.482 kD (P = 0.005; Figures 2Band 3). The peak at 11.311 kD (Figure 3, panels 1 and 2) wasone of the most significant peaks that differentiated betweenpatients with BKVAN from patients with both stable graft function(P < 0.001) and patients with AR (P = 0.002).
Figure 3. (1) False-gel image of 11.311-kD peak showing significant upregulation in BKVAN versus AR. (2 through 5) Scatter plots of the various significant classifiers in the two protein chips that differentiated BKVAN from AR and controls (IMAC30 11.3 kD [2], CM10 11.9 kD [3], CM10 12.7 kD [4], and IMAC30 23.4 kD [5]).
The results of the disease prediction modeling analyses thatwere performed using three different models (SVM, RF, and CART)are shown in Table 4. The ACE, sensitivity, and specificityof test sets depended on the chip type and on the classificationmodel used. The best performance characteristics with the leastclassification error (16.6%), maximum sensitivity (79.4%), andmaximum specificity (86.5%) for differentiating BKVAN from ARgroup was obtained by SVM algorithm that was performed on thetop 100 classifier peaks for the IMAC30 chip data set. The PACEanalysis that was performed on two of the classification algorithms(SVM and RF) indicated that the actual classification error(ACE) results were much lower than the mean error (MACE) expectedby chance alone. The resulting 95th and 99th percentiles ofthis distribution along with the MACE are shown in Table 5.
Table 4. Performance of various prediction models optimized for IMAC30 and CM10 chips for the BKVAN versus AR groups, BKVAN versus stable transplant groups, and AR versus stable transplant groupsa
We show that proteomic profiling of urine samples may offera noninvasive way to differentiate BKVAN from other conditionsas has been shown for several other diseases (8–12,28–30).All studied patients with BKVAN had significant viremia, viruria,and biopsy-proven nephropathy at the time of sample collection.The proteomic profile of the patients with BKVAN had similaritieswith the AR pattern. However, we were able to detect severalpeaks that were differentially expressed in the BKVAN groupcompared with both the AR and stable function groups.
Differentiation of BKVAN from AR can be challenging both athistologic and molecular levels. A recent study by Mannon etal. (31) showed significant similarity of transcriptional expressionof molecules associated with inflammation and fibrosis betweenBKVAN and AR. We found a similar overlap in the proteomic profilesof BKVAN and AR. This probably is due to the similarity of theinflammatory response and leakage of inflammation related smallmolecular weight proteins into urine in both conditions. However,several differentially expressed proteins were identified inthe urine of patients with BKVAN in our study. Our studies canbe an initial but promising phase in noninvasive biomarker developmentfor BKVAN.
The influence of viral infections on the urine proteomics hasnot been studied previously in a systematic manner, and no previousstudies have been reported for BKVAN. However, bacterial urinarytract infection (UTI) reportedly can have a unique proteomicprofile that can be different from AR (15,16). Schaub et al.(15) were able to differentiate urine samples (n = 5) of kidneytransplant patients with UTI from both AR and stable transplantsamples using SELDI. Another study also showed the proteomicprofiles of five of seven UTI samples were different from controland AR samples (16). In our studies, some of the samples alsowere positive for CMV, HHV-6, or adenovirus. These samples didnot seem to have any different or additional peaks that couldbe attributed to presence of the additional virus besides BKV.Because the progression of disease that is associated with BKVgenerally is less aggressive (32) as compared with bacterialinfections, development of noninvasive monitoring methods forBKV infection may be a clinically relevant and useful approach.
One of the advantages of SELDI is that it is inexpensive andis suitable for screening a large number of samples. However,this technique has some perceived inherent disadvantages. Oneof the major issues with SELDI is its relative lack of reproducibility,at least in different centers. However, in our studies, we foundexcellent reproducibility of the proteomic profiles when theexperiments were repeated on approximately 30% of the samplesafter 3 mo. The reasons underlying the variable results of differentcenters may be due to different sample-processing protocolsor approaches to the data analysis. Another disadvantage ofSELDI-MS-TOF has been that it does not offer easy identificationof the biomarker candidates. Therefore, studies with largernumbers of samples followed by peptide identification and immunologicassay development may be warranted.
Although no previous studies on BKVAN urinary proteomics areavailable, there are several reports now on urinary proteomicprofiling to differentiate patients with AR from stable patients,as summarized in Table 6 (13–17). Schaub et al. (33) alsorecently showed that cleaved 2-microglobulin (11.731 kD) maybe a potential marker for tubular injury in AR. The cleavageproducts were seen in urine proteome as three separate peakclusters (5.270 to 5.550, 7.050 to 7.360, and 10.530 to 11.100kD). However, in other studies, including ours, the most significantpeaks or peak clusters that differentiated patients with ARfrom stable transplant patients were different from these peaks(13,14,16,17). We found some similarities in the proteomic profileof our stable transplant patients and patients with AR comparedwith some of the other studies as shown in Table 6. The reasonsunderlying the different results of these studies probably aremultifactorial. Urine is a highly complex mixture of varioussolutes, and its composition is influenced by diet, hydration,smoking, medication etc. We aimed to study the different clinicalsyndromes using urine samples that typically would be obtainedin a standard clinical setting. We did not consider dietaryfactors or medications that are difficult to control. Besidesthese clinical variables, protein chip type and normalizationapproaches can be other potential confounding factors. In ourstudy, the urine samples were normalized by loading an equalamount of protein. This same method has been used in some otherstudies (10,17).
Table 6. Summary of the recent studies reporting proteomics biomarker candidates differentiating patients with acute AR and stable graft function and patients with various types of cancer from healthy individualsa
The statistical analyses, including univariate and multivariatetesting, of our data have shown intriguing results. Univariateanalysis evaluates the ability of peaks to discriminate betweentwo conditions, and several such peaks were identified. However,the limitation of this analysis is that peaks are analyzed inisolation. A much better discriminative model possibly can beconstructed if several peaks are analyzed together. Therefore,to analyze the multivariate discriminative potential of variouspeaks, we evaluated various statistical prediction models: SVM,CART, and RF (22–24). In addition, the SVM model was combinedwith two feature selection strategies that aimed to prefilterpeaks before classifiers were learned, thereby reducing potentialthreat of overfit. Using this approach, we found that SVM modelingof the top 100 peaks provided the best discrimination betweenBKVAN and AR using the IMAC30 data set. However these resultscome with a caveat, because one potential weakness of our studyis that the sample size is not large. The various disease-predictionmodeling algorithms may have limited utility in analyses ofa data set of our sample size and also of others published beforethis study (10–12,14). Given these limitations, our analysesof SELDI profile data can be interpreted only to suggest thatthere is a diagnostic signal for BKVAN versus AR or controlsin the SELDI analyses performed even on the limited sample sizeof our study. Our interpretation is reinforced further by thenonparametric PACE analysis (27), which also suggested thatthe potential diagnostic signal that was observed in patientdata is significantly different from randomly reassigned databecause of the marked difference of the ACE from the MACE andthe 99th percentiles of error distribution curves. However,all of our analyses were based on a limited sample size, andour results on the sensitivity and the specificity of the variousalgorithms should be interpreted with caution. A true assessmentof sensitivity and specificity of the SELDI technique and thevarious models tested in this report cannot be determined untilan independent validation set that is derived from another setof patients is assessed. Our results, therefore, may providethe basis for development of noninvasive marker tests for BKVANif further validation studies on independent data sets couldconfirm our observations.
Proteomic marker(s) profiles, together with plasma and urineBKV PCR and clinical information, may help in making differentiationof BKVAN from AR in a noninvasive manner. Histologic verificationof BKVAN probably will continue to be required for the foreseeablefuture, but it is likely that proteomic biomarkers could beused in deciding when a biopsy is necessary. Further studieson a larger number of patients are needed to validate our findingsand to detect the identity of the significantly different peaksto develop robust, noninvasive methods for BKVAN diagnostics.
Acknowledgments
This study was supported by the Finnish Pediatric Research Foundationand Finnish Cultural Foundation to T.J. and National Institutesof Health awards R01AI060602 to A.V. and R01AI051227 to P.S.R.The Proteomics and Bioinformatics Core Facilities and personnel(J.L.W., M.H., W.B., R.P., and D.M.) are supported by a Telemedicineand Advanced Technology Research Center/Department of Defense,Prime Award (W81XWH-05-2-0066; R. Herberman, Principal Investigator)to the University of Pittsburgh.
Portions of these studies were presented in abstract form atthe World Transplant Congress; Boston, MA; July 22 to 27, 2006.
Footnotes
T.J. and D.M. contributed equally to this work.
Published online ahead of print. Publication date availableat www.jasn.org.
Gardner SD, MacKenzie EF, Smith C, Porter AA: Prospective study of the human polyomaviruses BK and JC and cytomegalovirus in renal transplant recipients.
J Clin Pathol 37
: 578
–586, 1984[Abstract/Free Full Text]
Hirsch HH, Brennan DC, Drachenberg CB, Ginevri F, Gordon J, Limaye AP, Mihatsch MJ, Nickeleit V, Ramos E, Randhawa P, Shapiro R, Steiger J, Suthanthiran M, Trofe J: Polyomavirus-associated nephropathy in renal transplantation: Interdisciplinary analyses and recommendations.
Transplantation 79
: 1277
–1286, 2005[CrossRef][Medline]
Hariharan S: BK virus nephritis after renal transplantation.
Kidney Int 69
: 655
–662, 2006[CrossRef][Medline]
Pappo O, Demetris AJ, Raikow RB, Randhawa PS: Human polyoma virus infection of renal allografts: Histopathologic diagnosis, clinical significance, and literature review.
Mod Pathol 9
: 105
–109, 1996[Medline]
Randhawa PS, Demetris AJ: Nephropathy due to polyomavirus type BK.
N Engl J Med 342
: 1361
–1363, 2000[Free Full Text]
Randhawa PS, Finkelstein S, Scantlebury V, Shapiro R, Vivas C, Jordan M, Picken MM, Demetris AJ: Human polyoma virus-associated interstitial nephritis in the allograft kidney.
Transplantation 67
: 103
–109, 1999[CrossRef][Medline]
Hirsch HH, Knowles W, Dickenmann M, Passweg J, Klimkait T, Mihatsch MJ, Steiger J: Prospective study of polyomavirus type BK replication and nephropathy in renal-transplant recipients.
N Engl J Med 347
: 488
–496, 2002[Abstract/Free Full Text]
Petricoin EF, Ardekani AM, Hitt BA, Levine PJ, Fusaro VA, Steinberg SM, Mills GB, Simone C, Fishman DA, Kohn EC, Liotta LA: Use of proteomic patterns in serum to identify ovarian cancer.
Lancet 359
: 572
–577, 2002[CrossRef][Medline]
Qu Y, Adam BL, Yasui Y, Ward MD, Cazares LH, Schellhammer PF, Feng Z, Semmes OJ, Wright GL Jr: Boosted decision tree analysis of surface-enhanced laser desorption/ionization mass spectral serum profiles discriminates prostate cancer from noncancer patients.
Clin Chem 48
: 1835
–1843, 2002[Abstract/Free Full Text]
Rogers MA, Clarke P, Noble J, Munro NP, Paul A, Selby PJ, Banks RE: Proteomic profiling of urinary proteins in renal cancer by surface enhanced laser desorption ionization and neural-network analysis: Identification of key issues affecting potential clinical utility.
Cancer Res 63
: 6971
–6983, 2003[Abstract/Free Full Text]
Vlahou A, Schellhammer PF, Mendrinos S, Patel K, Kondylis FI, Gong L, Nasijm S, Wright GL Jr: Development of a novel proteomic approach for the detection of transitional cell carcinoma of the bladder in urine.
Am J Pathol 158
: 1491
–1502, 2001[Abstract/Free Full Text]
Vlahou A, Schorge JO, Gregory BW, Coleman RL: Diagnosis of ovarian cancer using decision tree classification of mass spectral data.
J Biomed Biotechnol 5
: 308
–314, 2005
Clarke W, Silverman BC, Zhang Z, Chan DW, Klein AS, Molmenti EP: Characterization of renal allograft rejection by urinary proteomic analysis.
Ann Surg 237
: 660
–665, 2003[CrossRef][Medline]
O’Riordan E, Orlova TN, Mei JJ, Butt K, Chander PM, Rahman S, Mya M, Hu R, Momin J, Eng EW, Hampel DJ, Hartman B, Kretzler M, Delaney V, Goligorsky MS: Bioinformatic analysis of the urine proteome of acute allograft rejection.
J Am Soc Nephrol 15
: 3240
–3248, 2004[Abstract/Free Full Text]
Schaub S, Rush D, Wilkins JA, Gibson IW, Weiler T, Sangster K, Nicolle L, Kaprinski M, Jeffery J, Nickerson P: Proteomic-based detection of urine proteins associated with acute allograft rejection.
J Am Soc Nephrol 15
: 219
–227, 2004[Abstract/Free Full Text]
Wittke S, Haubitz M, Walden M, Rohde F, Schwartz A, Mengel M, Mischak H, Haller H, Gwinner W: Detection of acute tubulointerstitial rejection by proteomic analysis of urinary samples in renal transplant recipients.
Am J Transplant 5
: 2479
–2488, 2005[CrossRef][Medline]
Voshol H, Brendlen N, Muller D, Inverardi B, Augustin A, Pally C, Wieczorek G, Morris RE, Raulf F, van Oostrum J: Evaluation of biomarker discovery approaches to detect protein biomarkers of acute allograft rejection.
J Proteome Res 4
: 1192
–1199, 2005[CrossRef][Medline]
Racusen LC, Solez K, Colvin RB, Bonsib SM, Castro MC, Cavallo T, Croker BP, Demetris AJ, Drachenberg JC, Fogo AB, Furness P, Gaber LW, Gibson IW, Glotz D, Goldberg JC, Grande J, Halloran PF, Hansen HE, Hartley B, Hayry PJ, Hill CM, Hoffman EO, Hunsicker LG, Lindblad AS, Yamaguchi Y: The Banff 97 working classification of renal allograft pathology.
Kidney Int 55
: 713
–723, 1999[CrossRef][Medline]
Vats A, Shapiro R, Randhawa PS, Scantlebury V, Tuzuner A, Saxena M, Moritz ML, Beattie TJ, Gonwa T, Green MD, Ellis D: Quantitative viral load monitoring and cidofovir therapy for the management of BK virus-associated nephropathy in children and adults.
Transplantation 75
: 105
–112, 2003[CrossRef][Medline]
Weiss SM, KulJkowski CA:
Computer Systems that Learn: Classification and Prediction Methods from Statistics, Neural Nets, Machine Learning, and Expert Systems, San Mateo, Morgan Kaufmann, 1990
Vapnik VN:
The Nature of Statistical Learning Theory, New York, Springer-Verlag, 1995
Breiman L: Random forests of random features.
Machine Learning 45
: 5
–32, 2001[CrossRef]
Breiman L, Friedman JH, Olshen RA, Stone CJ:
Classification and Regression Trees, Belmont, Wadsworth, 1984
Baldi P, Long AD: A Bayesian framework for the analysis of microarray expression data: Regularized t-test and statistical inferences of gene changes.
Bioinformatics 17
: 509
–519, 2001[Abstract/Free Full Text]
Good P:
Permutation Tests: A Practical Guide to Resampling Methods for Testing Hypotheses, 2nd Ed., New York, Springer-Verlag, 2000
Lyons-Weiler J, Pelikan R, Zeh HJ 3rd, Whitcomb DC, Malehorn DE, Bigbee WL, Hauskrecht M: Assessing the statistical significance of the Achieved Classification Error of classifiers constructed using serum peptide profiles, and a prescription for random sampling repeated studies for massive high-throughput genomic and proteomic studies.
Cancer Informatics 1
: 53
–77, 2005
Haubitz M, Wittke S, Weissinger EM, Walden M, Rup-precht HD, Floege J, Haller H, Mischak H: Urine protein patterns can serve as diagnostic tools in patients with IgA nephropathy.
Kidney Int 67
: 2313
–2320, 2005[CrossRef][Medline]
Vlahou A, Fountoulakis M: Proteomic approaches in the search for disease biomarkers.
J Chromatogr B 814
: 11
–19, 2005
Nguyen MT, Ross GF, Dent CL, Devarajan P: Early prediction of acute renal injury using urine proteomics.
Am J Nephrol 25
: 318
–326, 2005[CrossRef][Medline]
Mannon RB, Hoffmann SC, Kampen RL, Cheng OC, Kleiner DE, Ryschkewitsch C, Curfman B, Major E, Hale DA, Kirk AD: Molecular evaluation of BK polyomavirus nephropathy.
Am J Transplant 5
: 2883
–2893, 2005[CrossRef][Medline]
Carbone M, Burck C, Rdzanek M, Rudzinski J, Cutrone R, Bocchetta M: Different susceptibility of human mesothelial cells to polyomavirus infection and malignant transformation.
Cancer Res 63
: 6125
–6129, 2003[Abstract/Free Full Text]
Schaub S, Wilkins JA, Antonovici M, Krokhin O, Weiler T, Rush D, Nickerson P: Proteomic-based identification of cleaved urinary beta2-microglobulin as a potential marker for acute tubular injury in renal allografts.
Am J Transplant 5
: 729
–738, 2005[CrossRef][Medline]
This article has been cited by other articles:
D. L. Bohl and D. C. Brennan BK Virus Nephropathy and Kidney Transplantation
Clin. J. Am. Soc. Nephrol.,
July 1, 2007;
2(Supplement_1):
S36 - S46.
[Abstract][Full Text][PDF]
,
Top
Abstract
Introduction
2 test for categorical variables. For the SELDI data, Ciphergen ProteinChip software was used to detect peaks, and the corresponding peak intensities were exported to Microsoft Excel. The significance of differentially expressed peaks among three groups was studied using the Kruskal-Wallis one-way ANOVA on ranks. Pairwise comparisons between groups that were found to be significantly different on Kruskal-Wallis test were done using t test (SPSS 13.0 statistical software; SPSS, Chicago, IL). 
2-microglobulin (11.731 kD) may be a potential marker for tubular injury in AR. The cleavage products were seen in urine proteome as three separate peak clusters (5.270 to 5.550, 7.050 to 7.360, and 10.530 to 11.100 kD). However, in other studies, including ours, the most significant peaks or peak clusters that differentiated patients with AR from stable transplant patients were different from these peaks (13,14,16,17). We found some similarities in the proteomic profile of our stable transplant patients and patients with AR compared with some of the other studies as shown in Table 6. The reasons underlying the different results of these studies probably are multifactorial. Urine is a highly complex mixture of various solutes, and its composition is influenced by diet, hydration, smoking, medication etc. We aimed to study the different clinical syndromes using urine samples that typically would be obtained in a standard clinical setting. We did not consider dietary factors or medications that are difficult to control. Besides these clinical variables, protein chip type and normalization approaches can be other potential confounding factors. In our study, the urine samples were normalized by loading an equal amount of protein. This same method has been used in some other studies (10,17). 
