Article pubs.acs.org/jpr
Urinary Metabotyping of Bladder Cancer Using Two-Dimensional Gas Chromatography Time-of-Flight Mass Spectrometry Kishore Kumar Pasikanti,†,§ Kesavan Esuvaranathan,*,‡ Yanjun Hong,†,§ Paul C. Ho,† Ratha Mahendran,‡ Lata Raman Nee Mani,‡ Edmund Chiong,‡ and Eric Chun Yong Chan*,† †
Department of Pharmacy, Faculty of Science, National University of Singapore, 18 Science Drive 4, Singapore 117543 Department of Surgery, National University Health System, 5 Lower Kent Ridge Road, Singapore 119074
‡
S Supporting Information *
ABSTRACT: Cystoscopy is the gold standard clinical diagnosis of human bladder cancer (BC). As cystoscopy is expensive and invasive, it compromises patients’ compliance toward surveillance screening and challenges the detection of recurrent BC. Therefore, the development of a noninvasive method for the diagnosis and surveillance of BC and the elucidation of BC progression become pertinent. In this study, urine samples from 38 BC patients and 61 non-BC controls were subjected to urinary metabotyping using two-dimensional gas chromatography timeof-flight mass spectrometry (GC×GC−TOFMS). Subsequent to data preprocessing and chemometric analysis, the orthogonal partial least-squares discriminant analysis (OPLS-DA, R2X = 0.278, R2Y = 0.904 and Q2Y (cumulative) = 0.398) model was validated using permutation tests and receiver operating characteristic (ROC) analysis. Marker metabolites were further screened from the OPLS-DA model using statistical tests. GC×GC−TOFMS urinary metabotyping demonstrated 100% specificity and 71% sensitivity in detecting BC, while 100% specificity and 46% sensitivity were observed via cytology. In addition, the model revealed 46 metabolites that characterize human BC. Among the perturbed metabolic pathways, our clinical finding on the alteration of the tryptophanquinolinic metabolic axis in BC suggested the potential roles of kynurenine in the malignancy and therapy of BC. In conclusion, global urinary metabotyping holds potential for the noninvasive diagnosis and surveillance of BC in clinics. In addition, perturbed metabolic pathways gleaned from urinary metabotyping shed new and established insights on the biology of human BC. KEYWORDS: bladder cancer, two dimensional gas chromatography, metabolomics, metabonomics, metabotyping, GC×GC−TOFMS,
■
INTRODUCTION
pertinent for the future management of the progression and recurrence of BC. Metabonomics measures the dynamic multiparametric response of the systems biological metabolome to genetic modifications or pathophysiological stimuli.6,7 In contrast to classical biochemical approaches that focus on a single metabolite or family of metabolites and their associated metabolic reactions,8 metabonomics allows scientists to survey global changes in metabolic pathways and gain a holistic understanding of the changes in a biological system.9,10 Fluctuations in the metabolome are a result of a complex interaction between the genome, epigenome, transcriptome, proteome, and the environment.8,9 As a result, metabonomics has shown increasing applications in biomarker discovery for the diagnosis and elucidation of a number of pathologies11−13 and assessment of the exposure of biological systems to xenobiotics.14 Our group demonstrated for the first time that urinary metabolic profiling using gas chromatography−mass spectrometry (GC−MS) holds potential to diagnose BC and additionally provided information on the classification of bladder
Bladder cancer (BC) is the second most common genitourinary malignant disease in the United States with estimated 70,530 newly diagnosed cases in 2010.1 Cystoscopic examination and histopathologic evaluation of the biopsied tissue remain the gold standard for BC diagnosis.2 Cystoscopy is invasive and may be associated with a small but definite risk of morbidity. Due to the relatively high frequency of BC recurrence, cystoscopic surveillance is essential after tumor resection.3 The schedule recommended by the European Organization for Research and Treatment of Cancer (EORTC) for patients with low- to intermediate-risk disease is cystoscopy at 3-month intervals for the first 2 years, at 4-month intervals for the next 2 years, and yearly thereafter.3 Due to the demand for life-long surveillance, the per-patient cost of BC management is highest among all other cancers, with total estimated cost of $3.98 billion for year 2010 in USA.4 In recent years, numerous urinebased bladder tumor markers (UBBTMs) have been evaluated to determine whether less invasive follow-up of patients with BC is feasible.5 However, UBBTMs do not have the specificity and sensitivity of cystoscopy. As such, the development of alternative noninvasive diagnostic methods for BC becomes important. In addition, the elucidation of BC biology is © 2013 American Chemical Society
Received: January 16, 2013 Published: July 25, 2013 3865
dx.doi.org/10.1021/pr4000448 | J. Proteome Res. 2013, 12, 3865−3873
Journal of Proteome Research
Article
tumors.13,15 Subsequently, Issaq et al. reported the feasibility of using liquid chromatography−mass spectrometry (LC−MS) to analyze urinary metabolites to discriminate between BC patients and control subjects.16 A subsequent study identified diagnostic and prognostic markers of BC using LC−MS and highlighted mechanisms associated with the silencing of xenobiotic metabolism.17 Srivastava et al. utilized nuclear magnetic resonance (NMR) spectroscopy to probe the metabolic perturbations occurring in non-muscle invasive urinary BC and identified taurine as a possible biomarker associated with non-muscle invasive BC.18 Collectively, these studies underscore the complementary advantage of adopting multiple analytical platforms for the metabotyping of BC. Nevertheless, it is important to recognize that only a fraction of the metabolic space is currently captured by each analytical platform. While an expansion of the metabolic space coverage may yield novel biomarkers related to pathogenesis and disease progression,19 capturing the complete metabolome remains challenging due to the large and diverse metabolic space and wide dynamic range of metabolite concentrations.20 Comprehensive two-dimensional gas chromatography timeof-flight mass spectrometry (GC×GC−TOFMS) is noted for its ability to analyze multiple analytes in complex mixtures21 and has been expectedly explored in metabonomics in recent years.22−24 The principles and instrumentation of GC×GC− MS are presented in greater details in the review papers by Cortes et al. and Ong et al.25,26 In comparison to onedimensional GC−MS and NMR spectroscopy, GC×GC− TOFMS offers several advantages.22,24,27 For example, our group recently published the tissue metabolic profiling of human colorectal cancer using two-dimensional gas chromatography time-of-flight mass spectrometry (GC×GC− TOFMS) and demonstrated its superior performance in terms of metabolic space coverage, screening of marker metabolites and robustness in model prediction as compared to GC−MS and high-resolution magic angle spinning NMR spectroscopy.28 A research question thus arose whether the augmented metabolic space coverage provided by GC×GC−TOFMS would enrich the metabotyping of human BC and yield additional biomarkers associated with the pathology. The overarching aim of this study was therefore to apply GC×GC− TOFMS-based urinary metabolic profiling of human BC and investigate the marker metabolites associated with the disease. This study is significant as it facilitates a deeper mechanistic understanding of human BC and further reinforces the role of GC×GC−TOFMS in clinical metabonomics.
■
Table 1. Summary of Clinicopathogical Characteristics of Bladder Cancer (BC) Patients and Non-BC Subjects characteristics no. of subjects age (mean ± SD) gender male female cancer stage and gradea PUNLMP Ta-LG Ta-HG T1-LG T1-HG CIS/T1-HG others citizenship Chinese Indian othersb smoking habit nonsmokers ex-smokers smokers information not available
BC patients
non-BC controls
38 68.3 ± 10.9
61 60.5 ± 12.7
32 (84.2%) 6 (15.8%)
36 (59.0%) 23 (41.0%)
4 9 4 1 9 4 7
(10.5%) (23.7%) (10.5%) (2.6%) (23.7%) (10.5%) (18.4%)
28 (73.7%) 6 (15.8%) 4 (10.5%)
46 (75.4%) 11 (18.0%) 4 (6.6%)
12 (31.6%) 12 (31.6%) 10 (26.3%) 4 (10.5%)
31 (50.8%) 4 (6.6%) 8 (13.1%) 18 (29.5%)
a
Tumors were classified according World Health Organization (WHO)/International Society of Urological Pathology (ISUP) classification criteria. Low grade and high grade are represented as LG and HG, respectively. bOther citizenships include Caucasian and Malaysian.
forms to be participants in the study. Ethical permission to perform the study was obtained from National University of Singapore Institutional Review Board (NUS-IRB, DSRB-B/07/ 192). All urine samples were stored in aliquots at −80 °C until sample preparation for GC×GC−TOFMS analysis. Chemicals
MSTFA (N-methyl-N-(trimethylsilyl)trifluoroacetamide)) with 1% (trimethylchlorosilane) and MOX (methoxamine) were purchased from Pierce (Rockford, IL,USA). 4-[2-(N,NDimethylamino)ethylaminosulfonyl]-7-(2-aminoethylamino)2,1,3-benzoxadiazole (DAABD-AE), N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride (EDC), 4(dimethylamino)pyridine (DMAP), dansyl chloride, creatinine, L-tryptophan, kynurenine, kynurenic acid, xanthurenic acid, 3hydroxyanthranilic acid, 3-hydroxykynurenine, alkane standard mixture (C10 to C40), and sodium sulfate (anhydrous) were obtained from Sigma-Aldrich (St. Louis MO, USA). LTryptophan-d5 was purchased from Cambridge Isotope Laboratories (Tewksbury, MA, USA). All other chemicals were of analytical grade.
MATERIALS AND METHODS
Clinical Population
The clinicopathological data of BC and non-BC groups are summarized in Table 1. The presence of tumors in bladder was initially visualized using cystoscopy. Subsequently, histopathology examination of the transurethrally resected specimens was used to diagnose BC and determine the stage and grade of these tumors. Patients whose histology did not show any malignancy were classified as non-BC subjects. Both stage and grade of tumors were determined according to the World Health Organization (WHO)/International Society of Urological Pathology (ISUP) classification criteria.29 All volunteers completed confidential health and lifestyle related questionnaires, reported gender and age, and signed informed consent
Metabolic Profiling
The overall workflow of GC×GC−TOFMS-based urinary metabolic profiling utilized in this study is similar to the previously reported workflow.13 Sample Preparation
All urine samples were prepared according to the sample preparation protocol described previously.13 Quality control (QC) samples were prepared by mixing equal amounts of urine samples from 5 BC patients and 5 non-BC subjects. The urine samples were thawed at room temperature (23 ± 3 °C). Briefly, 3866
dx.doi.org/10.1021/pr4000448 | J. Proteome Res. 2013, 12, 3865−3873
Journal of Proteome Research
Article
200 μL of each urine samples was subjected to urea depletion using 100 U of urease enzyme. Subsequently, each urine sample was extracted using methanol, evaporated to dryness at 40 °C, and subjected to methoximation followed by MSTFA derivatization. A 1 μL portion of each derivatized sample was subjected to GC×GC−TOFMS analysis.
chemometric analysis. In addition, peaks that were not present in at least 50% of the samples were removed from the data table. Total area normalization was performed to correct minor variations due to sample preparation and analysis. Chemometric Data Analysis
Normalized data was exported to SIMCA-P (version 12.0, Umetrics, Umeå, Sweden) to perform principal component analysis (PCA) where grouping trends and outliers were observed. Prior to PCA, GC×GC−TOFMS data was meancentered and unit variance-scaled. DModX plot was calculated to check for any outliers. Subsequently, the data were subjected to partial least-squares discriminant analysis (PLS-DA) where a model was built and utilized to identify marker metabolites that accounted for the differentiation of BC from non-BC phenotypes. The chemical identities of the marker metabolites were confirmed by cross-referencing with the Golm metabolite library30 and the HMDB database.31 Key marker metabolites were further confirmed using standards. In addition, metabolic pathway interpretation of marker metabolites was performed using the KEGG database.32
GC×GC−TOFMS Analysis
A Pegasus 4D GC×GC−TOFMS (LecoCorp., St. Joseph, MI, USA) was utilized for the analysis. The instrument was equipped with an Agilent 7890 GC and Pegasus IV TOFMS (LecoCorp., St. Joseph, MI, USA) featuring a dual-stage, quadjet thermal consumable free modulator (−80 °C) and an independently temperature-controlled secondary oven. Helium was used as the carrier gas at 1.5 mL/min in corrected constant flow mode. A 30 m × 250 μm (i.d.) × 0.25 μm DB-1 (Agilent J&W Scientific, Folsom, CA) and a 1.5 m × 100 μm (i.d.) × 0.100 μm Rxi-17 (Restek Corp., Bellefonte, PA, USA) fused silica capillary columns were used as the primary and secondary columns, respectively. Primary oven temperature was programmed at 70 °C for 0.2 min and increased at 10 °C/min to 270 °C where it was held for 7.5 min. Secondary oven temperature was always maintained at 10 °C higher than the primary oven temperature. The thermal modulator was set to 20 °C higher relative to the primary oven. A modulation time of 3.5 s with hot pulse of 0.8 s was used. The injector, transfer line, and ion source temperatures were maintained at 220, 200, and 250 °C, respectively, throughout each analysis. Injector split ratio was set to 1:50. The MS was operated in EI mode (70 eV) at the detector voltage of 1600 V. Data acquisition was performed in the full scan mode from m/z 40 to 600 with an acquisition rate of 100 Hz.
External Validation 1
To evaluate the predictive ability of the orthogonal PLS-DA (OPLS-DA) model with respect to human BC, an external sample set “A” containing seven BC patients and 10 non-BC subjects was tested. The clinical characteristics of the patients used for external validation are summarized in Table 2. Table 2. Clinicopathogical Characteristics of Bladder Cancer (BC) Patients and Non-BC Subjects Utilized for External Validation of the OPLS-DA Model (External Validation Set “A”)
Data Preprocessing
Each chromatogram obtained from GC×GC−TOFMS analysis of urine samples was processed for baseline correction, smoothing, noise reduction, deconvolution, library matching, and area calculation using the ChromaTOF software (version 4.21, LecoCorp.). The area of each peak was calculated using the unique mass of each derivatized metabolite. Only peaks with signal-to-noise ratio greater than 200 were selected for further analysis. All GC×GC−TOFMS detected peaks were identified by comparing both the MS spectra and the retention indices (RI) with those available in the NIST mass spectral library (Wiley registry) and our in-house spectral library. Subsequently, the Statistical Compare feature of ChromaTOF software was utilized to generate data table in which all of the peak information from different chromatograms were aligned. Statistical Compare utilized a mass spectral match criterion of 60% when aligning the multidimensional peak data comprising sample names, metabolites, retention time (RT), mass, and integrated peak area. Quantitation mass for each peak in the data table was selected from the unique mass that was common to all matching peaks within the RT window (2 times of peak width). Subsequently, this unique mass was used to calculate the peak areas from each chromatogram. Similarly, the best quality peak from matching peaks was then selected whereby the putative identity was used as the name of the metabolite in the data table. The resulting data table comprised observations (samples and controls) and variables (peak areas) aligned according to their RT and unique mass pair as identifiers. The data table was exported as a .csv file and was processed subsequently using in-house Matlab (Mathworks, Natick, MA, USA) scripts to retain only observations and variables for
characteristics no. of subjects age (mean ± SD) gender male female cancer stage and gradea PUNLMP Ta-LG Ta-HG T1-LG T1-HG CIS/T1-HG others citizenship Chinese Indian othersb smoking habit nonsmokers ex-smokers smokers information not available
BC patients
non-BC controls
7 63.6 ± 16.4
10 58.4 ± 12.1
7 (100%) 0 (0%)
7 (70.0%) 3 (30.0%)
0 4 1 0 2 0 0
(0%) (57.1%) (14.3%) (0%) (28.6%) (0%) (0%)
4 (57.1%) 1 (14.3%) 2 (28.6%)
9 (90.0%) 1 (10.0%) 0 (0%)
2 2 2 1
4 1 1 4
(28.6%) (28.6%) (28.6%) (14.3%)
(40.0%) (10.0%) (10.0%) (40.0%)
a Tumors were classified according World Health Organization (WHO)/International Society of Urological Pathology (ISUP) classification criteria. Low grade and high grade are represented as LG and HG, respectively. bOther races include Caucasian and Malaysian.
3867
dx.doi.org/10.1021/pr4000448 | J. Proteome Res. 2013, 12, 3865−3873
Journal of Proteome Research
Article
Figure 1. (A) OPLS-DA scores plot obtained from the GC×GC−TOFMS-based urine analysis of bladder cancer (BC) and non-BC (H) subjects. (B) Validation plot obtained from 100 permutation tests. (C) Receiver operating characteristic (ROC) calculated using cross validated Y-predicted values of PLS-DA model and (D) T-predicted scatter plot of OPLS-DA model obtained from prediction of independent set of BC and H urine samples.
External Validation 2
where chromatographic regions belonging to artifacts were demarcated. One key advantage of GC×GC−TOFMS was that the artifacts were chromatographically resolved from the metabolites in the second dimension chromatography due to different polarity of the secondary column, and hence, the noise peaks could be demarcated efficiently. Subsequent to the removal of artifacts, peak information belonging to different chromatograms was aligned in a data table. Due to the varied number of peaks and inconsistencies in assignment of compound name, aligning peaks in a suitable matrix format was particularly challenging for 2D chromatographic data. A few strategies had been reported to address this challenge.22,24,33 However, these strategies typically required manual data processing in conjunction with existing instrument software24 or development of new software.33 This challenge was suitably addressed in our study by using the Statistical Compare feature of the ChromaTOF software34 where a large percentage of data was observed to be aligned accurately. Nevertheless, a small subset of metabolites was found to be represented multiple times in the data table. These multiply represented peaks were counter-checked and corrected in the raw data such that only a single analyte name was retained in the data table for each metabolite. Subsequent to peak alignment, the final data table comprising 533 peaks was subjected to chemometric data analysis.
To further confirm the perturbation of the tryptophan metabolic pathway in BC, a liquid chromatography tandem mass spectrometry (LC−MS/MS) method was developed to quantitate creatinine, tryptophan, kynurenine, kynurenic acid, 3-hydroxykynurenine, xanthurenic acid, and 3-hydroxyanthranilic acid in urine samples obtained from an external validation set “B” comprising 32 BC and 31 non-BC subjects. The details of the LC−MS/MS method are summarized in the Supporting Information and Supplementary Table 2. The characteristics of the subjects used for external validation are summarized in Supplementary Table 3. The urinary levels of tryptophan and its metabolites were normalized against the urinary concentration of creatinine. Statistical t test was performed using Prism 5.0 (GraphPad Software, San Diego, CA) to compare the urinary levels of metabolites between the BC and non-BC cohorts.
■
RESULTS
Data Preprocessing
Subsequent to peak deconvolution, an average of 930 derivatized peaks were detected in each chromatogram. Among these peaks, a number of artifact peaks, derived from derivatizing reagent or column bleeding, were detected. Such artifacts, if not removed, would significantly interfere with the subsequent chemometric data analysis.23 To remove these artifacts consistently from all of the chromatograms, the classification feature of ChromaTOF software was utilized
Unsupervised and Supervised Chemometrics
Analytical performance was validated via PCA of QC samples, a strategy utilized for the validation of GC−TOFMS metabo3868
dx.doi.org/10.1021/pr4000448 | J. Proteome Res. 2013, 12, 3865−3873
Journal of Proteome Research
Article
Table 3. Differentiating Marker Metabolites between Bladder Cancer (BC) and Non-BC Subjects by GC×GC−TOFMS-Based Urinary Metabonomic Profiling subject
RIa
metabolite identity
fold changeb
t test
welch t test
QC (% CV)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
1495.2 1605.9 1841 1912.1 1708.5 1382.8 1544.8 1240.3 1600 1267 1905.7 1988.2 2006.6 1288 2120.2 1065.1 1450.4 1055.9 1520 2798.9 1680 1139.8 1435.1 2447.9 1745.8 1805.7 1580 1756.9 1535.4 1929.4 1213.6 1565.1 1130.1 2401.6 1882.8 1368.9 1627.5 1162.7 1905.7 1545 1386.8 1186.7 1111.9 1692 1530.2 1708.7
adipic acid anthranilic acid citric acid coumaric acid derivative cyclopentane-1,2-diamine dihydroxyacetone erythritol erythro-pentonic acid ethyl tartrate ethylmalonic acid gluconic acid gluconic acid derivative gluconic acid derivative glycerol heptadecanoic acid hydroxybutyric acid itaconic acid lactic acid levulinic acid enol melibiose N-acetyl-anthranilic acid p-cresol pinene pseudouridine ribitol ribonic acid sebacic acid sugar (unknown) Sumiki’s acid talonic acid unknown unknown unknown uridine vanillylmandelic acid 2,3,4,5-tetrahydroxypentanoic acid-1,4-lactone 2,5-furandicarboxylic acid 2-aminoisobutyric acid 2-butanedioic acid 2-hydroxyglutaric acid 2-hydroxymalonic acid 2-pentadecanol 3,4-dihydroxyphenylpyruvate 3-hydroxysebacic acid 3-methyladipic acid 4-methoxycinnamic acid
1.4 1.7 0.8 1.4 1.8 0.7 1.2 2.1 0.4 1.6 0.5 1.3 2 0.7 1.3 1.2 1.8 1.7 0.4 1.3 1.9 1.7 0.9 1.6 0.6 0.5 0.7 0.4 0.6 1.3 2.5 1.7 1.3 1.2 1.3 1.7 0.5 1.5 0.6 0.7 1.7 1.7 1.4 2.2 1.4 1.6
