Abstract

Background. Urine proteomics is one of the key emerging technologies to discover new biomarkers for renal disease, which may be used in the early diagnosis, prognosis and treatment of patients. In the present study, we validated surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS) for biomarker discovery in patients with mild ischaemic kidney injury.

Methods. We used first-morning mid-stream urine samples from healthy volunteers, and from intensive care unit patients we collected urine 12–24 h after coronary artery bypass graft (CABG) surgery. Samples of 50 volunteers were mixed to establish a reference sample (master pool). Urine samples were analysed with constant creatinine levels.

Results. The average intra- and interchip variation was found to be in the normal experimental range (CV of 10 to 30%). Computational analysis revealed (i) low intra-individual day-to-day variation in individual healthy volunteers; (ii) high concordance between the master pool sample and individual samples. Machine learning techniques for classification of CABG condition vs healthy patients showed that (iii) in the 3-20 kDa range, the joint activity of four protein peaks effectively discriminated the two classes, (iv) in the 20–70 kDa range, a single m/z marker was sufficient to achieve perfect separation.

Conclusions. Our results substantiate the effectiveness of Seldi-TOF MS-based computational analysis as a tool for discovering potential biomarkers in urine samples associated with early renal injury.

Introduction

The identification of easily measured and reliable disease biomarkers is an important goal in clinical nephrology. Focus is now shifting from methods that analyse one marker at a time to profiling methods, which allow the simultaneous measurement of a range of markers, i.e. biomarker patterns [1–3]. Urine proteomics techniques in particular allow the identification of relative abundance and post-translational modifications of proteins for biomarker discovery and signal pathway analysis in renal pathophysiology (as reviewed, e.g. [4–8]).

Before the advent of proteomics, several urinary biomarkers were identified by classical biochemistry. Protein biomarkers demonstrated to be very useful in clinical assessment of renal pathology besides the determination of renal function through the analysis of glomerular filtration rate, renal blood flow and gene expression analysis to define relevant genes involved in nephropathy [9,10]. In 1994, Rossi et al. [11] evaluated renal damage induced by ifosfamide/cisplatin using a sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS–PAGE)-based method and reported that an increased level of alpha-1-microglobulin (alpha-1-m) was an important indication for impaired tubular function [11]. In addition, measuring urinary beta-2-microglobulin (beta-2-m) excretion was found to be useful in the identification of patients with membranous nephropathy at risk for renal disease progression [12].

Now, surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (Seldi-TOF MS) profiling provides protein chip-based technology for biomarker discovery [13,14]. Preliminary Seldi-TOF exploratory studies in clinical urine samples with renal pathology revealed new candidate biomarkers. Recently, neuthrophil gelatinase associated lipocalin (NGAL) has been proposed as a marker for acute renal failure after ischaemia/reperfusion injury [15], and cleaved urinary beta-2-m was identified as a potential marker for acute tubular injury in renal allografts [16].

In a review paper, O’Riordan and Goligorsky [17] comment on the development of urine proteomics as a three-stage process. The first phase, the exploratory phase, is now nearly at the end. The next phase is the development of validated prototype test models for urine proteomic profiling of early kidney injury. In this study, we developed and validated such a prototype model to be useful in a third clinical trial phase for proteome-wide screening for renal ischaemic injury [18]. We developed a prototypic experimental approach wherein protein loading was based on constant creatinine levels to minimize variations due to differences in urine flow rate. A defined group of intensive care (IC) unit patients with expected mild kidney injury after ischaemia during coronary artery bypass graft (CABG) surgery was used as a test group and compared with healthy volunteers. We previously demonstrated mild, subclinical, renal damage in this group of patients [19]. A comprehensive statistical methodology was applied, with a focus on finding the most discriminative biomarkers for the disease condition. First, we applied univariate methods for a quantitative assessment of distinctive CABG biomarkers and classification power. Second, we used two explorative graphical display methods based on results of unsupervised multi-variate data analysis techniques, namely principal component analysis and hierarchical clustering analysis. These standard tools, widely available in commercial chemometric packages, allow quick and easy visual inspection of the underlying variance and clusters in the data set. Finally, we applied state-of-the-art supervised data analysis techniques, namely, feature selection and support vector machines (SVM), for substantiating the results of standard chemometric techniques and for identifying robust interrelated potential biomarkers, whose joint action provide improved discrimination of mild kidney injury after CABG surgery.

Subjects and methods

Sample collection of healthy volunteers and CABG patients

First-morning mid-stream urine samples were collected in 50 healthy volunteers after informed consent. The absence of protein, blood and leucocytes in these normal urine samples was checked with the Super Aution EXSA4250 and Uriflet S sticks (Menarini Diagnostics Benelux, Valkenswaard, The Netherlands). After 1 week, 20 out of 50 volunteers returned to donate a second sample for comparison. Freshly collected mid-stream urine samples were briefly centrifuged (10 min, 2000 g) and stored with protease inhibitors [20] in small aliquots at −80°C to minimize freeze-thaw cycles. In 20 patients of the IC unit, urine was collected for 3–4 h through an indwelling catheter 12–20 h after CABG surgery. Included patients had no pre-existing renal insufficiency or post-operative renal failure, as monitored 7 days after surgery by creatinine serum levels.

Quantitative protein-assays for alpha-1-microglobulin and NGAL

Alpha-1-m was measured with immunonephelometry in the Nephelometer Analyzer II (Dade Behring, Marburg, Germany) after incubating samples with polyclonal rabbit anti-human alpha-1-m (DakoCytomation, Clostrup, Denmark). NGAL concentrations in urine samples were determined with an NGAL ELISA kit (kindly provided by AntibodyShop A/S, Gentofte, Denmark), according to the supplier's instructions.

Sample preparation

A master pool reference sample of all healthy volunteers was prepared by mixing together 50 urine samples containing 0.2 mmol creatinine each. For assessment of the biological variation, 13 urine samples (8 male, 5 female) were thawed, vortexed and diluted with ultraPURE™ DNAse/RNAse-free distilled water (Invitrogen, Breda, the Netherlands) to obtain constant creatinine concentrations (2.0 mmol/l). Subsequently, 250 μl urine was 10-times concentrated and desalted with centrifugal ultra-filtration (45–60 min, 13 000 g, 3 kDa Microcon filters, Millipore, Billerica, USA).

Seldi-MS analysis: preparations and measurement

The preparation procedure was based on protocols from Ciphergen and reports on non-pathological extrinsic/intrinsic factors [21–24]. For analysis, we used an 8-spot weak cation-exchange chip (ProteinChip CM10; Ciphergen Biosystems, Fremont, CA). All spots were pre-treated two times with 200 μl binding buffer (0.1 M ammonium acetate, pH 4.0) for 5 min on a shaking platform (500 rpm) in a bioprocessor. After pre-treatment, 5 μl concentrated urine samples (100 nmol creatinine) were applied on the chip and incubated in the humid chamber for 30 min. Samples were removed with a pipette, and spots were washed three times for 5 min with binding buffer (6 μl). Spots were washed with distilled water and air dried for 10 min. Subsequently, 0.8 µl of a saturated solution of sinapinic acid in 0.5% (v/v) trifluoroacetic acid and 50% (v/v) acetonitrile, used as energy-absorbing matrix, was applied to each spot surface, allowed to air-dry, and reapplied. The chip was analysed in a PBS IIc Seldi mass spectrometer (Ciphergen Biosystems, Fremont, CA). Data were collected with standard operational settings, including starting laser intensity at 200, warming positions with two shots set at laser intensity 205, no warming shots included, high mass at 200 kDa, optimized from 1 kDa to 21 kDa in the low range and optimized from 20 kDa to 71 kDa in the middle range and detector sensitivity at 9. External mass calibration was performed with all-in-one protein standard II proteins (Ciphergen Biosystems). In the range of 3–20 kDa: hirudin BHVK (6964 Da), bovine cytochrome C (12 230 Da), equine myoglobin (16 951 Da). In the range of 20–70 kDa: bovine carbonic anhydrase (29 023 Da), enolase from Saccharomyces cerevisiae (46 671 Da) and bovine albumin (66 433 Da).

Computational data analysis and validation

Ciphergen Express software (version 3.2, Ciphergen Biosystems) was initially used for pre-processing the data, i.e. automatic peak detection after baseline subtraction and adjustment (S/N ratio > 5, cluster mass window 0.3% peak width, peak detection if present in 5% of all spectra). The program also allows basic statistical analysis, including hierarchical clustering, group scatter plots, P-values (t-test statistics) and receiver operating characteristic (ROC) curves. Group differences were tested for statistical significance by a non-parametric t-test (Mann–Whitney U-test). The area under the ROC curve provides a measure of robustness for each biomarker, with a value of 1, indicating a perfect biomarker. Throughout the text, feature, biomarker and peaks are used interchangeably and point to the detected m/z values.

Additionally, multivariate analysis was applied to the data set to identify groups of potential biomarkers and to construct computational models for classification of IC vs healthy patients. We performed principal components analysis (PCA), multivariate feature selection (RELIEF) and classification with linear SVM.

Hierarchical clustering and PCA

For explorative cluster-analysis, hierarchical clustering and PCA are routinely applied in chemometric applications. Hierarchical clustering by the between-group linkage method in a dendrogram graphically shows overall sample similarity based on a Pearson correlation matrix, scoring the ‘overall’ similarity of the profiles. With PCA the data set can be visualized by a cloud of points, where coordinates represent the sample ‘scores’ as a weighted combination of peak value intensities [25–27]. PCA reduces the large number of dimensions (‘peak values’) of a data set into a smaller number of dimensions in such a way that the variance of the data set is maximized in the first principal components (PC). PCA was performed on a mean-centred unscaled data set in Matlab version 6.0 (The MathWorks, Inc., Natick, MA).

RELIEF

RELIEF [28] was used for multivariate feature ranking, using weights associated to features (m/z values, peaks). The rationale behind this algorithm is that a relevant feature (peak) should differentiate between samples from different classes and has the same value for samples of the same class. The algorithm is widely used in machine learning application, because of its simplicity and easy interpretability. In the context of MS data analysis, RELIEF has shown to select reliable features when applied to a MALDI-TOF MS data set obtained by spiking samples [29]. Briefly, the algorithm computes a vector W of feature weights. The vector is initially equal to zero, and is iteratively updated by the following steps: select one sample x from the training set; find its nearest neighbour from the same (x_hit) and from the opposite (x_mis) class; update W by adding abs(x-x_mis) − abs(x-x_hit), where abs denotes the absolute value. At the end of this iterative procedure the values of W provide a measure of feature relevance, where large values correspond to features with a large difference between nearest examples of different classes and a small difference between nearest examples of the same class.

SVM

Linear SVM [30] was used for classification. In short, linear SVM for (binary) classification separates the samples of the two classes (here CABG patients vs healthy volunteers) in the training set by means of a linear decision boundary in such a way that the minimal distance of samples of opposite classes from the decision boundary (the margin) is maximized. Samples nearest to the decision boundary are called support vectors. SVM is considered the state-of-the-art classification technique and is particularly suitable in domain applications where the number of features is very high, like classification with high-throughput MS data sets [31,32].

Validation

Leave-one-out cross validation (loocv) was used to assess the predictive performance of the SVM classification model. In loocv, at each run all samples but one are used for constructing a model, and the left-out sample is used to test the model on new samples. To assess the reliability of the results using a notion of statistical significance, a permutation test [33] was used. Intuitively, the goal is to check whether the observed cross validation (cv) error of the SVM model is obtained by chance, only because the training algorithm identified some pattern in the high-dimensional data that correlate with the class labels as an artefact of a small data set size [34]. To this aim, multiple runs of the cv procedure are performed with data labels permuted, and the distribution of the resulting cv errors is estimated and compared with the observed cv error on the original unpermuted data set.

Finally, in order to assess the reliability of the feature selection technique for classification, the observed cv error is compared with the estimated distribution of cv errors obtained when random feature selection is used. This is achieved by performing multiple runs of the cv procedure; where at each run, features are randomly selected. Note that these techniques are applied to detect over-fitting [32,35], but they do not provide a proof of reproducibility of results. A computational model is said to over-fit when its classification results degrade when the model is applied to new data. In this study, we analyse a data set of small size: dimensionality reduction and permutation tests are applied for preventing and detecting over-fitting, respectively. A more thorough validation of results could be achieved by using a larger data set and a test set preferably from another institution in order to detect bias introduced during the data generation process. However, full confidence on the reliability of results can only be achieved by experimental validation.

Protein identification

Two-dimensional gel electrophoresis was performed as described [36]. An amount of 30 ml urine sample was ultrafiltrated at 4600 g for 15 min through a 5000 NMWL filter (Centricon Plus-20; Millipore). Subsequently, the filter residue was centrifuged at 2000 g for 1 min to collect the ultrafiltrate, and speedvacced to concentrate the sample. An amount of 50 µg protein was loaded on an immobilized pH gradient (IPG) strip (pH 3–10; Immobiline Drystrips) for conventional isoelectric focusing and subsequent second dimension processing, as described [36]. After the second dimension, gels were stained with colloidal Coomassie as described elsewhere and scanned using an Amersham Biosciences Image Scanner. Interesting spots were picked and identified with Fourier Transform Ion Cyclotron Resonance Mass Spectrometry (nano LC-MS/MS) according to [37]. Proteins were identified via automated database searching (Matrix Science, London, UK) of tandem mass spectra against NCBInr database.

Results

Assessment of experimental methodology with the master pool sample

Healthy volunteers were instructed to follow a clean-catch mid-stream protocol. In line with previous reports [23], we avoided well-known confounding external factors (blood contamination, first void urine). Blood contamination in CABG patients, possibly by catheterization, was avoided by collecting urine the morning after surgical intervention. Microscopy was employed to detect the presence of squamous epithelial cells and non-mid-stream urine was excluded. Urines with a protein concentration of more than 0.5 mg/ml were excluded as well, and the number of erythrocytes was counted with a ×20 objective and should not exceed 10 per microscopic field. After optimizing intrinsic experimental factors (energy-absorbing matrix, laser intensity, pH set to 4.0 for CM10 chips, urine dilution, storage at −80°C), we evaluated the intra- and interchip variation with the master pool sample (Figure 1). We found variation coefficients between 13 and 34% with similar values in the range of 3–20 kDa and 20–70 kDa (Table 1). The addition of protease inhibitors to the master pool resulted in less peptide fragments, as assessed by nano-liquid chromatography coupled Fourier Transform Ion Cyclotron Resonance Mass Spectrometry (data not shown).

Fig. 1.

Seldi-TOF spectra showing the interday, interchip and inter 3 kD ultrafiltration variation of the masterpool; (A) Peaks within the 3–20 kDa range are shown of the master pool sample, processed in quadruplicate on CM10 chips. (B) Gel view of the same spectra.

Fig. 1.

Seldi-TOF spectra showing the interday, interchip and inter 3 kD ultrafiltration variation of the masterpool; (A) Peaks within the 3–20 kDa range are shown of the master pool sample, processed in quadruplicate on CM10 chips. (B) Gel view of the same spectra.

Table 1.

Assessment of the technical reproducibility of Seldi-TOF analysisa

m/z range: 3–20 kDa 20–70 kDa 
Intrachip 13 20 
Interchip 27 26 
Interday (interchip; independent 3 kDa ultrafiltration) 30 34 
m/z range: 3–20 kDa 20–70 kDa 
Intrachip 13 20 
Interchip 27 26 
Interday (interchip; independent 3 kDa ultrafiltration) 30 34 

aThe master pool sample was processed with CM10 chips. The average inter-, and intra-chip and inter-day variation is presented, expressed as Coefficient of Variation.

Urine profiles by SELDI-TOF MS in individual urine samples and the master pool

The protein composition of the master pool sample is reflected in the m/z profile. Comparison of individual urine samples, collected with a 1-week time interval in three healthy volunteers revealed consistent co-clustering (Figure 2). Profiles of normal individuals revealed no overall clustering of samples in the same gender or age category, though m/z markers can be found for these categories (data not shown). A gel view of spectra of individual healthy subjects, the master pool and CABG patients shows a good agreement with the individual healthy subjects and the master pool and high concordance within the groups, with clearly more peaks appearing in urines of CABG patients (Figure 3).

Fig. 2.

Hierarchical clustering of the biological interday variation in three healthy volunteers (B12; B11; B17). Urines were collected at day 1 (d1) and 1 week later (d8) and processed for Seldi-TOF MS analysis (in duplicate). Hierarchical clustering shows clear co-clustering of the individual samples, revealing modest individual daily variation in the proteome.

Fig. 2.

Hierarchical clustering of the biological interday variation in three healthy volunteers (B12; B11; B17). Urines were collected at day 1 (d1) and 1 week later (d8) and processed for Seldi-TOF MS analysis (in duplicate). Hierarchical clustering shows clear co-clustering of the individual samples, revealing modest individual daily variation in the proteome.

Fig. 3.

Gel view of spectra in the m/z range 3–20 kDa (A) and the m/z range 20–70 kDa (B) from normal and CABG urine samples. Spectra are shown for the master pool (MP, row 1), normal samples (N; rows 2–5), and CABG patient samples with increasing alpha-1-m levels, namely 7.9, 27.9, 38.8, 62.8 and 116.1 mg alpha-1-m/g creatinine, (respectively IC13, IC21, IC34, IC24, IC22; rows 6–10).

Fig. 3.

Gel view of spectra in the m/z range 3–20 kDa (A) and the m/z range 20–70 kDa (B) from normal and CABG urine samples. Spectra are shown for the master pool (MP, row 1), normal samples (N; rows 2–5), and CABG patient samples with increasing alpha-1-m levels, namely 7.9, 27.9, 38.8, 62.8 and 116.1 mg alpha-1-m/g creatinine, (respectively IC13, IC21, IC34, IC24, IC22; rows 6–10).

Univariate statistical analysis of urine profiles by SELDI-TOF MS in CABG patients

No evidence of acute renal insufficiency was found in these patients, in only one patient serum creatinine increased by >25% (Table 2). However, most CABG patients showed mild tubular injury, as indicated by elevated levels of alpha-1-m (n = 19) and NGAL (n = 7; Table 2). We selected three patients in the high and middle alfa-1-m excretion range, and two patients with normal alfa-1-m excretion for further analysis. We compared urine profiles of patients, normal individuals and the master pool with Ciphergen Express Software, and we performed a univariate t-test and ROC-analysis to interpret significance of biomarkers in CABG patients.

Table 2.

Laboratory characteristics of the master pool and CABG patientsa

ID Gender (M/F) Age (years) Serum creatinine (µmol/l) Urine creatinine (mmol/l) Alpha-1-m (mg/g creatinine) NGAL (ng/g creatinine)b Total protein (g/l) 
   pre-surgery post-surgery     
MP M (25)/ F (25) 20–70 nd nd 6.5 6.7 0.6 nd 
IC7 66 75 101 11.8 55.4 nd 0.27 
IC8 75 90 81 9.7 207.8 70.8 0.39 
IC10 55 94 82 16.9 41.3 nd 0.28 
IC12 52 82 79 7.8 38.5 nd nd 
IC13 64 89 110 18.0 7.9 17.0 0.22 
IC14 78 106 100 16.9 99.9 74.0 0.51 
IC16 52 94 76 12.8 27.6 nd undetectable 
IC18 66 125 104 10.9 30.0 nd undetectable 
IC20 48 102 73 7.8 62.3 nd undetectable 
IC21 41 85 80 7.3 27.9 7.3 undetectable 
IC22 71 100 112 8.3 116.1 nd 0.31 
IC24 58 70 60 3.8 62.8 54.2 undetectable 
IC27 60 83 60 8.0 15.5 78.2 undetectable 
IC28 58 79 62 6.2 28.5 nd undetectable 
IC30 88 80 97 3.8 76.8 nd undetectable 
IC33 40 91 72 7.2 68.8 nd undetectable 
IC34 62 103 89 18.0 38.8 11.0 0.23 
IC36 52 85 72 14.0 50.5 nd 0.19 
IC37 58 73 73 10.8 39.3 nd undetectable 
IC40 75 71 66 3.7 33.4 nd undetectable 
ID Gender (M/F) Age (years) Serum creatinine (µmol/l) Urine creatinine (mmol/l) Alpha-1-m (mg/g creatinine) NGAL (ng/g creatinine)b Total protein (g/l) 
   pre-surgery post-surgery     
MP M (25)/ F (25) 20–70 nd nd 6.5 6.7 0.6 nd 
IC7 66 75 101 11.8 55.4 nd 0.27 
IC8 75 90 81 9.7 207.8 70.8 0.39 
IC10 55 94 82 16.9 41.3 nd 0.28 
IC12 52 82 79 7.8 38.5 nd nd 
IC13 64 89 110 18.0 7.9 17.0 0.22 
IC14 78 106 100 16.9 99.9 74.0 0.51 
IC16 52 94 76 12.8 27.6 nd undetectable 
IC18 66 125 104 10.9 30.0 nd undetectable 
IC20 48 102 73 7.8 62.3 nd undetectable 
IC21 41 85 80 7.3 27.9 7.3 undetectable 
IC22 71 100 112 8.3 116.1 nd 0.31 
IC24 58 70 60 3.8 62.8 54.2 undetectable 
IC27 60 83 60 8.0 15.5 78.2 undetectable 
IC28 58 79 62 6.2 28.5 nd undetectable 
IC30 88 80 97 3.8 76.8 nd undetectable 
IC33 40 91 72 7.2 68.8 nd undetectable 
IC34 62 103 89 18.0 38.8 11.0 0.23 
IC36 52 85 72 14.0 50.5 nd 0.19 
IC37 58 73 73 10.8 39.3 nd undetectable 
IC40 75 71 66 3.7 33.4 nd undetectable 

aCharacteristics of first morning urines of all CABG patients and master pool samples are given, including pre- and post-surgery serum creatinine values. Levels of known renal biomarkers, namely alpha-1-m and NGAL, were corrected for high variability in urinary creatinine concentration. bNGAL was determined solely in samples of which enough material was left. Nd, not determined.

As shown in Table 3, in the range of 3–20 kDa, 28 out of 70 automatically detected m/z values showed significantly different mean peak intensities in the patient samples as compared with normal samples (t-test, P < 0.01). A large proportion (22/28) was up-regulated. At least four of these (m/z 4546, 4715, 11 732 and 11 925 kDa) proved very good biomarkers for the CABG-condition (area under the ROC curves >0.98). The 11 732 kDa peak showed a 7-fold increase in average peak intensity compared with controls and represents beta-2-m, a well-known marker of proximal tubular injury. The other peaks remain to be identified. In the range of 20–70 kDa, 28 out of 35 protein peaks were significantly different, all up-regulated in CABG patients. At least 10 of these peaks are also very good classifiers for the CABG-condition (area under the ROC curve >0.98). We observed m/z values, potentially corresponding with known biomarkers retinol binding protein (RBP; m/z 20 900, 21 times up-regulated) and NGAL (m/z 28 749, 13 times up-regulated), a recently established marker for ischaemic injury [15]. Prominent, yet unidentified, peaks in the high m/z range were found at 35 222 kDa (16 times up-regulated) and 58 434 kDa (12 times up-regulated).

Table 3.

Univariate t-test and RELIEF-based rankings of top 28 features (m/z peaks) with smallest P-valuea

CM10 range 3–20 kDa 
m/z Peak rank P AURC Average fold ratio 
 t-test RELIEF    
4715 15 0.00017 0.99 
4546 0.00022 0.99 
9884 13 0.00022 0.04 0.25 
11 732 0.00029 0.99 
11 925 11 0.00029 0.99 
6087 18 0.00029 0.95 27 
5985 0.00038 0.95 31 
9781 0.00038 0.04 0.25 
7635 21 0.00051 0.95 
4651 10 17 0.00051 0.95 
10 276 11 29 0.00051 0.95 15 
8595 12 0.00066 0.95 13 
13 300 13 66 0.00066 0.95 
9121 14 23 0.00066 0.04 0.25 
4396 15 19 0.00086 0.95 
13 724 16 35 0.00086 0.91 
10 342 17 14 0.00086 0.91 12 
8496 18 0.00112 0.95 
4735 19 20 0.00144 0.91 
9943 20 28 0.00144 0.08 0.5 
15 783 21 30 0.00184 0.91 
5525 22 44 0.00230 0.15 0.13 
4027 23 36 0.00377 0.87 
5684 24 26 0.00377 0.87 25 
6283 25 34 0.00473 0.15 0.33 
3089 26 12 0.00592 0.87 373 
4256 27 10 0.00737 0.87 
13860 28 46 0.00737 0.83 
CM10 range 3–20 kDa 
m/z Peak rank P AURC Average fold ratio 
 t-test RELIEF    
4715 15 0.00017 0.99 
4546 0.00022 0.99 
9884 13 0.00022 0.04 0.25 
11 732 0.00029 0.99 
11 925 11 0.00029 0.99 
6087 18 0.00029 0.95 27 
5985 0.00038 0.95 31 
9781 0.00038 0.04 0.25 
7635 21 0.00051 0.95 
4651 10 17 0.00051 0.95 
10 276 11 29 0.00051 0.95 15 
8595 12 0.00066 0.95 13 
13 300 13 66 0.00066 0.95 
9121 14 23 0.00066 0.04 0.25 
4396 15 19 0.00086 0.95 
13 724 16 35 0.00086 0.91 
10 342 17 14 0.00086 0.91 12 
8496 18 0.00112 0.95 
4735 19 20 0.00144 0.91 
9943 20 28 0.00144 0.08 0.5 
15 783 21 30 0.00184 0.91 
5525 22 44 0.00230 0.15 0.13 
4027 23 36 0.00377 0.87 
5684 24 26 0.00377 0.87 25 
6283 25 34 0.00473 0.15 0.33 
3089 26 12 0.00592 0.87 373 
4256 27 10 0.00737 0.87 
13860 28 46 0.00737 0.83 

aAURC: Area under the ROC-curves for CABG patients vs normal and average fold increase in peak intensity with CABG patients vs normal range. Twenty-eight out of 70 m/z values have P-values <0.01, 22 are up-regulated.

Hierarchical clustering and principal component analysis

We applied two explorative data analysis methods, which allow visual inspection of clusters and underlying variance in the data set. In the range of 3–20 kDa, hierarchical cluster analysis almost perfectly discriminates CABG from healthy samples (Figure 4). The first and the second PC explained >70% of the variance in the data (explained variance: PC 1 = 58%; PC 2 = 16%; PC 3 = 6%; PC 4 = 5%; PC 5 = 4%). With only minor variation in the normal healthy samples, the first component clearly represents the difference between healthy volunteers and IC patients (Figure 5A). Seldi-TOF spectra of CABG patients with high alpha-1-m values (patient numbers 8, 14, 22 and 34, Table 2) indeed cluster together (Figure 4A). An exception is CABG patient 24, with high alpha-1-m values, which does not cluster with these samples. The urine of this patient had a high salt content, which complicated the process of ultrafiltration. During salt precipitation, proteins may have precipitated as well. The variation explained in the second component is dominated by polarity in two clusters of m/z markers, distinguishing patients 13, 24 and 27 from the rest (Figure 5B). This polarity in CABG patients is also reflected in the branch split in hierarchical cluster analysis (Figure 4A).

Fig. 4.

Hierarchical clustering of spectra in the m/z range 3–20 kDa (A) and 20–70 kDa (B). In the lower range there is distinctive clustering of CABG patients, whereas in the higher range there is only clustering of CABG patients with high urinary protein concentrations (>0,22 g/l; i.e. IC8, IC14, IC22, IC34). Mpm, masterpool sample; b, normal sample; IC, CABG patient sample.

Fig. 4.

Hierarchical clustering of spectra in the m/z range 3–20 kDa (A) and 20–70 kDa (B). In the lower range there is distinctive clustering of CABG patients, whereas in the higher range there is only clustering of CABG patients with high urinary protein concentrations (>0,22 g/l; i.e. IC8, IC14, IC22, IC34). Mpm, masterpool sample; b, normal sample; IC, CABG patient sample.

Fig. 5.

PCA score plot of PCA on the mean centered peak intensities in the m/z range 3–20 kDa. (A) PC1 vs PC2 reveals distinctive clustering of CABG patients compared with normal healthy controls: 58% of the variance in the data is explained by component 1, reflecting the CABG vs normal condition. (B) PCA loadings plot, showing the m/z values contributing the most to PC1 and PC2. Stars represent robust features also selected by SVM–RELIEF (4546, 11732, 8595, 8495). In addition, four other features (5985, 9780, 11473, 19114) were selected (cross) for classifcation. Mpm, masterpool sample; b, normal sample; IC, CABG patient sample.

Fig. 5.

PCA score plot of PCA on the mean centered peak intensities in the m/z range 3–20 kDa. (A) PC1 vs PC2 reveals distinctive clustering of CABG patients compared with normal healthy controls: 58% of the variance in the data is explained by component 1, reflecting the CABG vs normal condition. (B) PCA loadings plot, showing the m/z values contributing the most to PC1 and PC2. Stars represent robust features also selected by SVM–RELIEF (4546, 11732, 8595, 8495). In addition, four other features (5985, 9780, 11473, 19114) were selected (cross) for classifcation. Mpm, masterpool sample; b, normal sample; IC, CABG patient sample.

In the range of 2–70 kDa, the overall similarity of CABG patients, measured by hierarchical clustering of 35 peaks in the range of 20–70 kDa, tends to be less different from the normal samples than the profile in the lower range (compare Figure 4A and B), except for the four patients with very high alpha-1-m values. However, a limited number of markers can explain most of the variance in the data as shown by PCA (data not shown). Here, the first and the second PC explained 90% of the variance (explained variance: PC 1 = 67%; PC 2 = 23%; PC 3 = 5%; PC 4 = 3%; PC 5 = 1%). Most peaks do not add much weight in explaining the variance. Peaks with high weight in PC 1 (>0.1) are 35 227, 33 354 (double charged albumin), 21 089, 20 703, 20 507, 20 789, 20 900 (RBP), 66 288 (albumin) and 23 467 (Figure 6D). The latter peak adds most weight towards CABG identification (‘PC 1’). The variation within CABG patients (dominant in ‘PC 2’) is mainly explained by the albumin and RBP peaks.

Fig. 6.

SVM classification of CABG vs normal in the m/z range 3–20 kDa. (A) Model optimization with a variable number of peaks: Loocv error vs number of selected peaks of SVM with RELIEF-based feature selection. (B) SVM classification with four peaks: selected m/z values (x-axis) vs loocv runs (y-axis). The selected m/z markers are indicated with arrows and their m/z value.

Fig. 6.

SVM classification of CABG vs normal in the m/z range 3–20 kDa. (A) Model optimization with a variable number of peaks: Loocv error vs number of selected peaks of SVM with RELIEF-based feature selection. (B) SVM classification with four peaks: selected m/z values (x-axis) vs loocv runs (y-axis). The selected m/z markers are indicated with arrows and their m/z value.

Models for classification of CABG vs healthy condition

We applied machine learning techniques, SVM and RELIEF-based feature selection for constructing computational diagnostic models for the CABG condition. These techniques have been shown to be beneficial in analysis of high-throughput MS data sets [29,32]. In the range of 3–20 kDa, perfect predictive performance (zero loocv error) was achieved by SVM with all the 70 peaks. Permutation test (1000 runs) of the class labels (CABG patients, controls, master pool) resulted in a P-value = 0. This indicated that SVM performance is related to the structure of the data set and not to the complexity of the type of model. Application of SVM with four peaks selected by RELIEF achieved zero loocv error (Figure 6A and B). Permutation tests (1000 runs) of class labels (P-value = 0) and results of random feature selection (P-value = 0.035) substantiated the relationship between the results and structure of the data. A high consensus of features selected over the loocv runs indicated the robustness of the selected m/z markers. In particular, m/z 4546 and 11 732 (beta-2-m) were selected at each of the 22 loocv runs, 8495, 8595 were almost always selected, and the remaining markers were 19 114, 5985, 9780 and 11 473. Figure 6B shows the selected m/z markers, indicated with arrows.

In the 20–70 kDa range, SVM applied to all 35 peaks resulted in 1 loocv error (sample IC21). Permutation test of class labels (1000 runs) showed that SVM captures the structure present in the data set (P-value = 0). Zero loocv error was obtained by SVM with single markers selected by RELIEF, for m/z values 20 900, 21 089 and 44 242. Random feature selection (1000 runs) achieved 0 loocv error in 35 out of 1000 selections, indicating significance of the selected peaks. SVM-based models constructed with more than one feature achieved slightly worse performance (1 loocv error, sample IC21). These results show that in the low molecular mass range, the combined action of interrelated m/z values provides improved separation of the CABG condition, while in the high molecular mass range each of the single selected m/z values is sufficient for the separation. This seems to indicate that for the type of data considered in this article multivariate feature selection for classification is especially beneficial for biomarker detection performed on a low molecular mass range.

Finally, we performed an additional study using two-dimensional gel electrophoresis and nano LC-MS/MS to further identify potential biomarkers. After loading an equal amount of protein on the gel, a clear up-regulation in protein expression in patients after CABG was found as compared with the MP (Figure 7). Subsequently, 27 abundantly present proteins were picked from the gel and identified using nano LC-MS/MS (Table 4). Among these proteins were NGAL (lipocalin), alpha-2-glycoprotein and RBP.

Fig. 7.

Proteome map of the masterpool (A) and three CABG patients (24, 8 and 14; B–D, respectively). The proteins were separated by 2D-PAGE based on their differential pH value for the isoelectric point (pI; x-axis) and molecular weights (y-axis). The protein spots were excised and underwent in-gel tryptic digestion followed by nano LC-MS/MS. Arrows point to the most abundant differences between the masterpool and patient samples. Number labelling in the figure corresponds to the numbers in Table 4.

Fig. 7.

Proteome map of the masterpool (A) and three CABG patients (24, 8 and 14; B–D, respectively). The proteins were separated by 2D-PAGE based on their differential pH value for the isoelectric point (pI; x-axis) and molecular weights (y-axis). The protein spots were excised and underwent in-gel tryptic digestion followed by nano LC-MS/MS. Arrows point to the most abundant differences between the masterpool and patient samples. Number labelling in the figure corresponds to the numbers in Table 4.

Table 4.

Identified proteins up-regulated in CABG patients

Spot nr. Identified protein 
Retinol binding protein; Agrine 
Retinol binding protein; Regenerating protein 
Lipocalin; Retinol binding protein 
Lipocalin 
Lipocalin 
Lipocalin 
Zn-alpha-2-glycoprotein; gelsolin 
Retinol binding protein; Basement membrane; Agrine precursor 
Lipocalin; Retinol binding protein 
10 Lipocalin; Retinol binding protein 
11 Lipocalin; Actin 
12 Lipocalin; Zn-alpha-2-glycoprotein 
13 Zn-alpha-2-glycoprotein; Haptoglobin 
14 Zn-alpha-2-glycoprotein 
15 Zn-alpha-2-glycoprotein 
16 Zn-alpha-2-glycoprotein; Carbonic anhydrase; Lipocalin 
17 Zn-alpha-2-glycoprotein; Carbonic anhydrase 
18 Perlecan; Alpha-actin 
19 Zn-alpha-2-glycoprotein; Perlecan; Immunoglobulin; Prostaglandin 
20 Immunoglobulin 
21 Amylase 
22 Alpha-amylase; Retinol binding protein 
23 Alpha-amylase; Retinol binding protein 
24 Immunoglobulin M (FAB); Alpha-amylase 
25 Transferrin; Bence–Jones protein 
26 Transferrin 
27 Cystatin; Transferrin; Alpha-amylase 
Spot nr. Identified protein 
Retinol binding protein; Agrine 
Retinol binding protein; Regenerating protein 
Lipocalin; Retinol binding protein 
Lipocalin 
Lipocalin 
Lipocalin 
Zn-alpha-2-glycoprotein; gelsolin 
Retinol binding protein; Basement membrane; Agrine precursor 
Lipocalin; Retinol binding protein 
10 Lipocalin; Retinol binding protein 
11 Lipocalin; Actin 
12 Lipocalin; Zn-alpha-2-glycoprotein 
13 Zn-alpha-2-glycoprotein; Haptoglobin 
14 Zn-alpha-2-glycoprotein 
15 Zn-alpha-2-glycoprotein 
16 Zn-alpha-2-glycoprotein; Carbonic anhydrase; Lipocalin 
17 Zn-alpha-2-glycoprotein; Carbonic anhydrase 
18 Perlecan; Alpha-actin 
19 Zn-alpha-2-glycoprotein; Perlecan; Immunoglobulin; Prostaglandin 
20 Immunoglobulin 
21 Amylase 
22 Alpha-amylase; Retinol binding protein 
23 Alpha-amylase; Retinol binding protein 
24 Immunoglobulin M (FAB); Alpha-amylase 
25 Transferrin; Bence–Jones protein 
26 Transferrin 
27 Cystatin; Transferrin; Alpha-amylase 

Discussion

Several proteomic approaches are now applied in the search for disease biomarkers [38]. These techniques hold great promise, but care is needed for accurate data interpretation. We validated a standardized creatinine based Seldi-TOF MS analysis using a reference sample (master pool). In addition, we have applied the technique in a comparison between healthy controls and a test group of CABG patients, and have evaluated different computational methods. Based on the evidence presented in this study, we emphasize that a creatinine-corrected Seldi-TOF MS analysis of urine samples could potentially be useful for the discovery of new urinary biomarkers after early stage kidney injury. This has also been concluded from a recent study in children after cardiac surgery [39] and in acute renal transplant rejection [40], although different biomarkers were detected in each study. This discrepancy may be due to variations in patient group studied, and also due to variations in urine collections and sample preparations. In nephrology, a SELDI-TOF MS-based spectrum test is not used for diagnostics so far, although in other clinical disciplines the proteomics technique has shown to be of great value. For example, it was recently shown that hepcidin can be determined quantitatively in urine using SELDI-TOF MS [41], and this is now applied on a routine basis. Major advantages of using the technique for urine proteomics are that the method is rapid, reliable and suitable for high-throughput profiling of urine samples. On the other hand, the use of the technique is limited by the lack of possibilities of protein identification and biomarker verification.

Experimental validation and development of a creatinine-based Seldi-TOF MS analysis

As with all clinical samples, urine should be obtained in a uniform reproducible and representative manner for protein profiling. We designed a protocol to minimize protein degradation and to increase experimental reproducibility. We developed a standardized creatinine-corrected procedure in morning urine with acceptable reproducibility for semi-quantitative protein profiling (variation <30%) for biomarker lead finding. We optimized the instrumental settings and urine dilution mainly with weak cation-exchange chips to yield spectra with many high quality peaks (S/N > 5). The choice of energy-absorbing matrix (SPA) favours detection of more proteins in the higher m/z range (>20 000).

Ultrafiltration was chosen as the sample preparation method for desalting and concentrating urine samples, which allows for versatile down-stream proteomics applications. This is in good agreement with optimal spot patterns obtained with 2D gel electrophoresis after urine ultrafiltration [42]. Creatinine-correction in this study adjusts for variation in urine flow. Other protein loading methods based on volume or based on total protein concentration (standard procedure in the literature) are biased by the variability in urine flow rate.

Assessing the master pool sample: a high quality normal reference sample

Profiling urines from healthy volunteers revealed a high concordance with the master pool. Hierarchical cluster analysis confirmed the master pool as a suitable reference for normal healthy volunteers of any sex and age. A reference master pool sample will facilitate and standardize comparison with pathological samples, especially in expensive and time-consuming techniques, like 2D differential gel electrophoresis. Furthermore, the biological inter-day variation was assessed. Cluster analysis showed overall minimal influence of inter-day fluctuations on urine profiling.

Profiling of CABG urine samples: a model for subclinical ischaemic kidney injury

We have selected patients without overt renal insufficiency. A previous study showed evidence of only subtle tubular injury in these patients as reflected by measurement of glutathione-S-transferases [19]. The protein pattern found in the present study early after CABG is no proof of tubular injury, however, the elevated levels of tubular injury markers are in support of mild renal damage. Alpha-1-m was increased in all but one patient sample and renal tubular injury is evident if levels exceed 20 mg/g creatinine. In this study, in CABG urine samples alpha-1-m levels ranged from 8 to 208 mg/g creatinine, indicating that tubular injury was absent to severe. Furthermore, the CABG-samples measured here have significantly elevated levels of the recently established ischaemia biomarker, NGAL, for which a level exceeding 10 ng/g creatinine may reflect acute renal failure [15].

Computational diagnostics and biomarker discovery in CABG patients

The Seldi-TOF MS technique proved very sensitive in detecting changes in the urinary proteome profile of CABG patients. Univariate and multivariate techniques revealed the presence of potential markers for CABG.

Univariate analysis indicated significant difference between Seldi-TOF protein profiles of IC unit patients after CABG surgery and healthy controls, primarily by an overall increase of proteins. Renal injury markers likely dominate the CABG profiles in the present study. Peaks in the Seldi-TOF profiles at m/z 11 732, 20 900, representing beta-2-m and RBP, respectively, indicate the presence of renal tubular damage [43,44]. Furthermore, the prominent m/z marker 8595, a robust classifier for the CABG condition, may represent ubiquitin. Ubiquitinated proteins are especially present in exosomes, which are vesicles present in urine that are secreted by kidney cells [45], and indicate enhanced degradation processes.

PCA revealed multiple subclasses of the CABG condition, primarily based on the variation in 10 potential biomarkers. Half of the CABG patients (IC8, 14, 34, 22) consistently co-clustered in the CABG spectrum over the whole m/z range. The other half revealed a less typical spectrum and tends to have a more normal profile. IC24 and IC27 showed a clear CABG signature in the high kDa range, but showed a more normal expression pattern in the low range. IC13 and IC21 were close to normal in the range of 20–70 kDa. In the low range, also IC21 was not different from samples of healthy volunteers, explaining why this patient sample is the most difficult to classify with supervised machine learning techniques.

SVM and RELIEF were applied to substantiate and enrich the results obtained by standard chemo-informatics techniques, in particular, for identifying groups of interrelated potential biomarkers whose joint activity provides a stronger degree of separation of the CAGB condition. In the range of 3–20 kDa, the joint effect of more peaks turned out to improve predictive performance. Especially linear discriminant models with four peaks achieved zero loocv error. In the range of 20–70 kDa, three single peaks achieved improved predictive performance. These results seem to indicate that univariate and multivariate techniques yield equivalent performance results when applied to high molecular mass data, while multivariate techniques improve separation when applied to small molecular mass data by means of models based on a panel of peaks [44]. However, we stress that the results reported here are based on the analysis of a small data set, and no validation on an external test set has been performed. Hence, accuracy results are possibly an optimistic estimate of the true discriminatory power of the detected protein signatures.

In conclusion, results of the computational analysis by means of multiple techniques indicate the high potential of urine samples processed by Seldi-TOF MS for identifying proteomic signatures of CABG. The detection of a number of well-known biomarkers relevant for the CABG condition shows that the experimental protocol used in this study generates reliable results, substantiated by computational analysis.

Perspectives

Based on the evidence presented in this study, we emphasize that a creatinine-corrected Seldi-TOF MS analysis of urine samples could potentially be useful for the discovery of new urinary biomarkers after early-stage kidney injury. Our method results in representative protein patterns. Further study is needed to identify and validate candidate biomarkers (e.g. the robust CABG biomarker m/z 4546) and to assess the reliability of the other markers here identified, in particular m/z 21 089 and 44 242.

The reproducibility can be improved by automation and uniform sample treatment. Further statistical and biological validation on larger data sets can fine-tune the relationship of CABG biomarkers with ischaemia, dissecting renal vs cardiovascular ischaemic damage, and document influences of pre-surgical ischaemic conditions or pharmacotherapy. We expect our prototypic tool to prove valuable for defining characteristic sets of proteins that are differentially expressed in different types of renal injury (e.g. ischaemic and toxic renal injury) and to assist in the diagnosis, and thereby prognosis of patients suffering from renal damage.

Acknowledgement

The Dutch Kidney Foundation (projectnr C04.2088) supported this study. The authors gratefully thank Marieke Menkhorst for her help with the two-dimensional gel electrophoresis.

Conflict of interest statement. None declared.

References

1
Johann
DJ
Jr
McGuigan
MD
Patel
AR
, et al.  . 
Clinical proteomics and biomarker discovery
Ann NY Acad Sci
 , 
2004
, vol. 
1022
 (pg. 
295
-
305
)
2
Espina
V
Mehta
AI
Winters
ME
, et al.  . 
Protein microarrays: molecular profiling technologies for clinical specimens
Proteomics
 , 
2003
, vol. 
3
 (pg. 
2091
-
2100
)
3
Bischoff
R
Luider
TM
Methodological advances in the discovery of protein and peptide disease markers
J Chromatogr B Analyt Technol Biomed Life Sci
 , 
2004
, vol. 
803
 (pg. 
27
-
40
)
4
Thongboonkerd
V
Malasit
P
Renal and urinary proteomics: current applications and challenges
Proteomics
 , 
2005
, vol. 
5
 (pg. 
1033
-
1042
)
5
Knepper
MA
Proteomics and the kidney
J Am Soc Nephrol
 , 
2002
, vol. 
13
 (pg. 
1398
-
1408
)
6
Thongboonkerd
V
Proteomics in nephrology: current status and future directions
Am J Nephrol
 , 
2004
, vol. 
24
 (pg. 
360
-
378
)
7
Thongboonkerd
V
Proteomic analysis of renal diseases: unraveling the pathophysiology and biomarker discovery
Expert Rev Proteomics
 , 
2005
, vol. 
2
 (pg. 
349
-
366
)
8
Hewitt
SM
Dear
J
Star
RA
Discovery of protein biomarkers for renal diseases
J Am Soc Nephrol
 , 
2004
, vol. 
15
 (pg. 
1677
-
1689
)
9
Fleck
C
Sutter
L
Appenroth
D
, et al.  . 
Use of gene chip technology for the characterisation of the regulation of renal transport processes and of nephrotoxicity in rats
Exp Toxicol Pathol
 , 
2003
, vol. 
54
 (pg. 
401
-
410
)
10
Luhe
A
Hildebrand
H
Bach
U
Dingermann
T
Ahr
HJ
A new approach to studying ochratoxin A (OTA)-induced nephrotoxicity: expression profiling in vivo and in vitro employing cDNA microarrays
Toxicol Sci
 , 
2003
, vol. 
73
 (pg. 
315
-
328
)
11
Rossi
RM
Kist
C
Wurster
U
Kulpmann
WR
Ehrich
JH
Estimation of ifosfamide/cisplatinum-induced renal toxicity by urinary protein analysis
Pediatr Nephrol
 , 
1994
, vol. 
8
 (pg. 
151
-
156
)
12
Branten
AJ
du Buf-Vereijken
PW
Klasen
IS
, et al.  . 
Urinary excretion of beta2-microglobulin and IgG predict prognosis in idiopathic membranous nephropathy: a validation study
J Am Soc Nephrol
 , 
2005
, vol. 
16
 (pg. 
169
-
174
)
13
Issaq
HJ
Veenstra
TD
Conrads
TP
Felschow
D
The SELDI-TOF MS approach to proteomics: protein profiling and biomarker identification
Biochem Biophys Res Commun
 , 
2002
, vol. 
292
 (pg. 
587
-
592
)
14
Clarke
W
Chan
DW
ProteinChips: the essential tools for proteomic biomarker discovery and future clinical diagnostics
Clin Chem Lab Med
 , 
2005
, vol. 
43
 (pg. 
1279
-
1280
)
15
Mishra
J
Dent
C
Tarabishi
R
, et al.  . 
Neutrophil gelatinase-associated lipocalin (NGAL) as a biomarker for acute renal injury after cardiac surgery
Lancet
 , 
2005
, vol. 
365
 (pg. 
1231
-
1238
)
16
Schaub
S
Wilkins
JA
Antonovici
M
, et al.  . 
Proteomic-based identification of cleaved urinary beta2-microglobulin as a potential marker for acute tubular injury in renal allografts
Am J Transplant
 , 
2005
, vol. 
5
 (pg. 
729
-
738
)
17
O’riordan
E
Goligorsky
MS
Emerging studies of the urinary proteome: the end of the beginning?
Curr Opin Nephrol Hypertens
 , 
2005
, vol. 
14
 (pg. 
579
-
585
)
18
Espina
V
Dettloff
KA
Cowherd
S
Petricoin
EF
III
Liotta
LA
Use of proteomic analysis to monitor responses to biological therapies
Expert Opin Biol Ther
 , 
2004
, vol. 
4
 (pg. 
83
-
93
)
19
Eijkenboom
JJ
van Eijk
LT
Pickkers
P
Peters
WH
Wetzels
JF
van der Hoeven
HG
Small increases in the urinary excretion of glutathione S-transferase A1 and P1 after cardiac surgery are not associated with clinically relevant renal injury
Intensive Care Med
 , 
2005
, vol. 
31
 (pg. 
664
-
667
)
20
Thongboonkerd
V
McLeish
KR
Arthur
JM
Klein
JB
Proteomic analysis of normal human urinary proteins isolated by acetone precipitation or ultracentrifugation
Kidney Int
 , 
2002
, vol. 
62
 (pg. 
1461
-
1469
)
21
Klasen
IS
Reichert
LJ
de Kat Angelino
CM
Wetzels
JF
Quantitative determination of low and high molecular weight proteins in human urine: influence of temperature and storage time
Clin Chem
 , 
1999
, vol. 
45
 (pg. 
430
-
432
)
22
Rogers
MA
Clarke
P
Noble
J
, et al.  . 
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
 , 
2003
, vol. 
63
 (pg. 
6971
-
6983
)
23
Schaub
S
Wilkins
J
Weiler
T
Sangster
K
Rush
D
Nickerson
P
Urine protein profiling with surface-enhanced laser-desorption/ionization time-of-flight mass spectrometry
Kidney Int
 , 
2004
, vol. 
65
 (pg. 
323
-
332
)
24
West-Nielsen
M
Hogdall
EV
Marchiori
E
Hogdall
CK
Schou
C
Heegaard
NH
Sample handling for mass spectrometric proteomic investigations of human sera
Anal Chem
 , 
2005
, vol. 
77
 (pg. 
5114
-
5123
)
25
Jackson
JE
A User's Guide to Principal Components
 , 
1991
New York
John Wiley and Sons
(pg. 
1
-
592
)
26
Jolliffe
IT
Principal Component Analysis
 , 
2002
New York
Springer-Verlag
(pg. 
1
-
478
)
27
Krzanowski
WJ
Principles of Multivariate Analysis
 , 
1988
Oxford
Oxford University Press
(pg. 
1
-
608
)
28
Kira
K
Rendell
LA
Sleeman
D
Edwards
P
A practical approach to feature selection
1992
Proceedings of the Ninth international Workshop on Machine Learning (Aberdeen, Scotland, United Kingdom)
San Francisco
Morgan Kaufmann Publishers
(pg. 
249
-
256
)
29
Marchiori
E
Heegaard
NH
Jimenez
C
West-Nielsen
M
Feature Selection for Classification with Proteomic Data of Mixed Quality
Proc IEEE Symp Computat Intel Bioinf Computat Biol
 , 
2005
(pg. 
385
-
391
)
30
Vapnik
VN
The Nature of Statistical Learning Theory
 , 
1998
New York
John Wiley and Sons, Inc.
(pg. 
1
-
314
)
31
Good
PI
Permutation Tests: A Practical Guide to Resampling Methods for Testing Hypotheses
 , 
1994
New York
Springer-Verlag
32
Hilario
M
Kalousis
A
Pellegrini
C
Muller
M
Processing and classification of protein mass spectra
Mass Spectrom Rev
 , 
2006
, vol. 
25
 (pg. 
409
-
449
)
33
Pavlidis
P
Wapinski
I
Noble
WS
Support vector machine classification on the web
Bioinformatics
 , 
2004
, vol. 
20
 (pg. 
586
-
587
)
34
Golland
P
Liang
F
Mukherjee
S
Panchenko
D
Permutation tests for classification
Learning Theory, Proceedings
 , 
2005
, vol. 
3559
 (pg. 
501
-
515
)
35
Simon
R
Radmacher
MD
Dobbin
K
McShane
LM
Pitfalls in the use of DNA microarray data for diagnostic and prognostic classification
J Natl Cancer Inst
 , 
2003
, vol. 
95
 (pg. 
14
-
18
)
36
Farhoud
MH
Wessels
HJ
Wevers
RA
van Engelen
BG
van den Heuvel
LP
Smeitink
JA
Serial isoelectric focusing as an effective and economic way to obtain maximal resolution and high-throughput in 2D-based comparative proteomics of scarce samples: proof-of-principle
J Proteome Res
 , 
2005
, vol. 
4
 (pg. 
2364
-
2368
)
37
Tjalsma
H
Pluk
W
van den Heuvel
LP
Peters
WH
Roelofs
R
Swinkels
DW
Proteomic inventory of “anchorless” proteins on the colon adenocarcinoma cell surface
Biochim Biophys Acta
 , 
2006
, vol. 
1764
 (pg. 
1607
-
1617
)
38
Vlahou
A
Fountoulakis
A
Proteomic approaches in the search for disease biomarkers
J Chrom B-Anal Technol Biomed Life Sci
 , 
2005
, vol. 
814
 (pg. 
11
-
19
)
39
Nguyen
MT
Ross
GF
Dent
CL
Devarajan
P
Early prediction of acute renal injury using urinary proteomics
Am J Nephrol
 , 
2005
, vol. 
25
 (pg. 
318
-
326
)
40
Clarke
W
Silverman
BC
Zhang
Z
Chan
DW
Klein
AS
Molmenti
EP
Characterization of renal allograft rejection by urinary proteomic analysis
Ann Surg
 , 
2003
, vol. 
237
 (pg. 
660
-
664
)
41
Kemna
E
Tjalsma
H
Laarakkers
C
Nemeth
E
Willems
H
Swinkels
D
Novel urine hepcidin assay by mass spectrometry
Blood
 , 
2005
, vol. 
106
 (pg. 
3268
-
3270
)
42
Tantipaiboonwong
P
Sinchaikul
S
Sriyam
S
Phutrakul
S
Chen
ST
Different techniques for urinary protein analysis of normal and lung cancer patients
Proteomics
 , 
2005
, vol. 
5
 (pg. 
1140
-
1149
)
43
Blaikley
J
Sutton
P
Walter
M
, et al.  . 
Tubular proteinuria and enzymuria following open heart surgery
Intensive Care Med
 , 
2003
, vol. 
29
 (pg. 
1364
-
1367
)
44
Hampel
DJ
Sansome
C
Sha
M
Brodsky
S
Lawson
WE
Goligorsky
MS
Toward proteomics in uroscopy: urinary protein profiles after radiocontrast medium administration
J Am Soc Nephrol
 , 
2001
, vol. 
12
 (pg. 
1026
-
1035
)
45
Pisitkun
T
Shen
RF
Knepper
MA
Identification and proteomic profiling of exosomes in human urine
Proc Natl Acad Sci USA
 , 
2004
, vol. 
10
 (pg. 
13368
-
13373
)

Comments

0 Comments