Colorectal Cancer Detection Using Targeted ... - ACS Publications

Aug 4, 2014 - Indiana University Melvin and Bren Simon Cancer Center, 535 Barnhill Drive, Indianapolis, Indiana 46202, United States. ⊥. Public Heal...
13 downloads 10 Views 2MB Size
Article pubs.acs.org/jpr

Colorectal Cancer Detection Using Targeted Serum Metabolic Profiling Jiangjiang Zhu,†,# Danijel Djukovic,†,# Lingli Deng,†,‡ Haiwei Gu,† Farhan Himmati,† E. Gabriela Chiorean,§,∥ and Daniel Raftery*,†,⊥ †

Northwest Metabolomics Research Center, Department of Anesthesiology and Pain Medicine, University of Washington, 850 Republican Street, Seattle, Washington 98109, United States ‡ Departments of Electronic Science and Communication Engineering, State Key Laboratory for the Physical Chemistry of Solid Surfaces, Xiamen University, 422, South Siming Road, Xiamen 361005, China § University of Washington, 825 Eastlake Ave East, Seattle, Washington 98109, United States ∥ Indiana University Melvin and Bren Simon Cancer Center, 535 Barnhill Drive, Indianapolis, Indiana 46202, United States ⊥ Public Health Sciences Division, Fred Hutchinson Cancer Research Center, 1100 Fairview Avenue North, Seattle, Washington 98109, United States S Supporting Information *

ABSTRACT: Colorectal cancer (CRC) is one of the most prevalent and deadly cancers in the world. Despite an expanding knowledge of its molecular pathogenesis during the past two decades, robust biomarkers to enable screening, surveillance, and therapy monitoring of CRC are still lacking. In this study, we present a targeted liquid chromatography−tandem mass spectrometry-based metabolic profiling approach for identifying biomarker candidates that could enable highly sensitive and specific CRC detection using human serum samples. In this targeted approach, 158 metabolites from 25 metabolic pathways of potential significance were monitored in 234 serum samples from three groups of patients (66 CRC patients, 76 polyp patients, and 92 healthy controls). Partial least-squares−discriminant analysis (PLS−DA) models were established, which proved to be powerful for distinguishing CRC patients from both healthy controls and polyp patients. Receiver operating characteristic curves generated based on these PLS−DA models showed high sensitivities (0.96 and 0.89, respectively, for differentiating CRC patients from healthy controls or polyp patients), good specificities (0.80 and 0.88), and excellent areas under the curve (0.93 and 0.95). Monte Carlo cross validation was also applied, demonstrating the robust diagnostic power of this metabolic profiling approach. KEYWORDS: metabolomics, colorectal cancer, polyps, LC−MS/MS, targeted metabolic profiling, serum metabolites, diagnostic biomarkers



INTRODUCTION Colorectal cancer (CRC) is one of the most prevalent types of cancer worldwide and a major cause of human morbidity and mortality.1 According to the American Cancer Society, CRC is the third most common type of cancer in the US (136 830 new cases estimated for 2014) and the third most common cause of cancer death (50 310 deaths estimated).2 The American College of Gastroenterology guidelines for CRC screening (2008)3 suggest several preventive screening and detection methods for CRC, including fecal occult blood test (FOBT), fecal immunochemical test (FIT), colonoscopy/sigmoidoscopy, and family history-based risk assessment. FOBT is one of the most commonly used screening methods for CRC diagnosis; however, it has low sensitivity (43%), especially for early-stage CRC.4 FIT has several advantages over FOBT, such as requiring less restricted diet for test preparation and less demanding sample collection procedures; however, the sensitivity is still not ideal (65.8−81.8%).5 Colonoscopy and © 2014 American Chemical Society

sigmoidoscopy remain the gold standards for screening and detection of CRC, but their major disadvantages include invasiveness, potential risks of complications, and high cost.1 Compliance rates are less than ideal (∼48%),6 with the result that only 40% of CRC patients are diagnosed and treated with early-stage, localized disease (Stages I and II), which have relatively high (80−90%) 5-year survival rates.7 Therefore, developments of new screening methods that are highly sensitive, specific, and noninvasive are critically needed for the early diagnosis and timely treatment of CRC. A number of new CRC detection methods are being developed particularly based on stool samples, including stool DNA (sDNA)8 and microRNA (miRNA)9 testing, that have shown evidence of possible noninvasive detection of CRC. The sDNA test detected four methylated genes and a mutant form Received: May 19, 2014 Published: August 4, 2014 4120

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

robust diagnostic power of this targeted serum metabolic profiling approach.

of V-Ki-ras2 Kirsten rat sarcoma viral oncogene homologue (KRAS) and correctly identified 85% of patients with CRC with a specificity of 90%.8 A very recent version has been performed on almost 10 000 patients and shows improved results.10 Fecal miRNA measurements showed higher expression of miR-21 and miR-106a in patients with CRC compared with individuals free of colorectal neoplasia; therefore, they might also be used as diagnostic biomarkers.9a An important characteristic of cancer is its abnormal metabolism. Altered levels of metabolites from important metabolic pathways have been the focus of many cancer studies.11 Metabolomics, the comprehensive study of smallmolecular-weight metabolites and their dynamic changes in biological systems, provides advanced methods to identify changing metabolite levels, resulting in rapid progress in disease biomarker discovery over the past decade.12 Mass-spectrometry-based metabolic profiling has been proven to be a promising tool for analyzing metabolic alterations due to various cancers and therefore can provide sensitive and valuable diagnostic information, pathogenesis clarification, and potential therapeutic targets for clinical treatments.13 Previous studies to identify metabolite biomarkers for CRC have been performed using gas chromatography−mass spectrometry (GC−MS),14 nuclear magnetic resonance (NMR) spectroscopy,15 flow injection analysis tandem mass spectrometry (FIA−MS/MS), and liquid chromatography−time-of-flight−mass spectrometry (LC− TOF−MS)14a,c,16 by comparing the global metabolic profiles from CRC patients to healthy controls. Blood based biomarkers from several chemical classes (organic acid, amino acid, monosaccharide, pyrimidine nucleoside, etc.) were reported from these studies, such as lactate, glucose, proline uridine, 2hydroxybutyrate, aspartic acid, fumarate, and tryptophan. On the basis of these selected metabolites, their statistical model performance varies from area under receiver operating characteristic curve (AUROC) 0.88 to 0.97. Fourier transform ion cyclotron resonance mass spectrometry (FTICR−MS) has also been used for biomarker discovery, and a group of 10 lipid metabolites were chosen for the best separation between CRC patients and healthy controls.17 In another FTICR−MS study, a panel of 28−36 carbon-containing hydroxylated polyunsaturated ultralong-chain gastrointestinal tract fatty acids (GTA), identified from the results of a comprehensive global metabolomics analysis, has been shown to occur at reduced levels in CRC patient serum compared with those in diseasefree subjects,18 with GTA-446 as its most promising representative. Low levels of GTA-446 in serum may be considered to be a significant risk factor for CRC and a possibly sensitive predictor for the disease. While the results to date are promising, limited work has focused on targeted serum metabolic profiling of CRC. Meanwhile, discussion of the metabolic differences between CRC patients and polyp patients, in addition to healthy controls, has an important clinical impact on correct CRC diagnosis, however this information is rather scarce. In this study, a targeted metabolic profiling approach focused on the reliable detection of 158 metabolites from 25 metabolic pathways of potential significance is presented for the discovery of sensitive and specific CRC metabolic biomarkers. A total of 234 serum samples from three groups of subjects were analyzed, and potential biomarkers were selected from univariate and multivariate statistical analysis. Furthermore, cross-validation steps were performed to demonstrate the



MATERIALS AND METHODS

Clinical Samples

Patient recruitment and sample collection protocols were approved by the Purdue University and Indiana University School of Medicine Institutional Review Boards. Informed consent was provided from all subjects in the study according to institutional guidelines. All consenting participants (both healthy and diseased subjects) undergoing colonoscopy or CRC surgery were evaluated, and blood samples from the patients were obtained after overnight fasting and bowel preparation prior to their procedure. Healthy or polyp status was determined after colonoscopy. In total, 234 subjects were recruited in this study and were grouped into CRC patients (n = 66), polyp patients (n = 76), and healthy controls (n = 92) based on the analysis of biopsied tissue. Patients were age- and gender-matched in each group such that the gender/age group p values >0.05 using the Mann−Whitney U test, suggesting there was no statistical significance between groups. Patient demographical and clinical information is shown in Table 1. Each blood sample was allowed to clot for 45 min and then centrifuged at 1500g for 10 min. All samples were stored at −80 °C until experiments were performed. Table 1. Summary of Clinical and Demographic Characteristics of Human Subjects Included in This Study

age

gender cancer stage

diagnosis smoking status

alcohol status

total n = 234

CRC n = 66

polyps n = 76

healthy controls n = 92

median min max male female stage I/II stage III stage IV colon cancer rectal cancer nonsmoker some days everyday no alcohol occasionally at least 1 drink/day

58 27 88 30 36 21 17 28 39 27 19 42 5 3 29 34

56 37 86 37 39

57 18 80 45 47

14 61 1 2 37 37

34 49 9 9 41 42

Reagents

LC−MS-grade acetonitrile, ammonium acetate, and acetic acid were all purchased from Fisher Scientific (Pittsburgh, PA). Standard compounds corresponding to the measured metabolites (Supplemental Table S1 in the Supporting Information) were purchased from Sigma-Aldrich (Saint Louis, MO) or Fisher Scientific (Pittsburgh, PA). Stable isotope-labeled tyrosine and lactate internal standards (L-tyrosine-13C2 and sodium-L-lactate-13C3) were purchased from Cambridge Isotope Laboratories, (Tewksbury, MA). The purities of nonlabeled standards were >95−99%, whereas the purities of the two 13C labeled compounds were >99%. 4121

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

Sample Preparation

Mass Spectrometry Conditions

Frozen samples were first thawed at room temperature (25 °C) for ∼45 min, and 50 μL of each serum sample was placed in a 2 mL Eppendorf vial (Fisher Scientific). The initial step for protein precipitation and metabolite extraction was performed by adding 150 μL of methanol; the mixture was then vortexed for 2 min and stored at −20 °C for 20 min. Next, the sample was centrifuged at 20 800g for 10 min, and the supernatant was collected into a new Eppendorf vial. To the first vial containing the pellet, another 300 μL of methanol was added, and the mixture was vortexed for 10 min to allow thorough metabolite extraction. After centrifuging this mixture at 20 800g for 10 min, the supernatant was collected in the same vial that contained the previous supernatant. The resulting supernatants from two rounds of extractions were dried using a Vacufuge Plus evaporator (Eppendorf, Hauppauge, NY). The dried samples were stored at −20 °C and were reconstituted in 500 μL of 5 mM ammonium acetate in 40% water/60% acetonitrile +0.2% acetic acid containing 5.13 μM L-tyrosine-13C2 and 22.5 μM sodium-L-lactate-13C3 (Cambridge Isotope Laboratory). The two isotope-labeled internal standards were added to each sample to monitor the system performance. The samples were filtered through 0.45 μm PVDF filters (Phenomenex, Torrance, CA) prior to LC−MS analysis. A pooled sample, which was a mixture of serum from CRC patients, polyp patients, and healthy controls, was extracted using the same procedure as previously described. This sample was used as the qualitycontrol (QC) sample and was analyzed once every 10 patient samples.

After the chromatographic separation, MS ionization and data acquisition were performed using an AB Sciex QTrap 5500 mass spectrometer (AB Sciex, Toronto, ON, Canada) equipped with an electrospray ionization (ESI) source. The instrument was controlled by Analyst 1.5 software (AB Sciex, Toronto, ON, Canada). Targeted data acquisition was performed in multiple-reaction-monitoring (MRM) mode. We monitored 99 and 59 MRM transitions in negative and positive modes, respectively (158 transitions in total). The source and collision gas was N2 (99.999% purity). The ion source conditions in negative/positive mode were: curtain gas (CUR) = 25 psi, collision gas (CAD) = high, ion spray voltage (IS) = −3.8/3.8 kV, temperature (TEM) = 500 °C, ion source gas 1 (GS1) = 50 psi, and ion source gas 2 (GS2) = 40 psi. The optimized MS compound conditions are shown in Supplemental Table S1 in the Supporting Information. The extracted MRM peaks were integrated using MultiQuant 2.1 software (AB Sciex). Data Analysis, Model Development, and Cross Validation

To search for potential CRC diagnostic serum biomarkers, we performed metabolite selection, model building, and cross validation, and the data analysis steps are shown using a simplified flowchart shown in Supplemental Figure S1 in the Supporting Information. After exporting from MultiQuant software, we normalized spectral data using average values from the data of QC injections (at least 5 in each batch, 26 QC samples in total). The distribution of the correlation of variation (CV) values of all measured metabolites as well as two example metabolites (tyrosine and lactate) showing the variation of measurement during our QC injections can be seen in Figure S2 in the Supporting Information. Mann− Whitney U tests, generation of receiver operating characteristics (ROC) curves, and calculation of sensitivity, specificity, and AUROC were conducted using JMP Pro10 (SAS Institute). Partial least-squares−discriminant analysis (PLS−DA) and Monte Carlo cross validation19 (MCCV, developed using inhouse scripts) were performed using Matlab software (Mathworks, Natick, MA) installed with the PLS toolbox (Eigenvector Research, Wenatchee, WA). For each iteration of MCCV, the n samples were first randomly split into two parts, the training set (Xtrain, ytrain) and testing set (Xtest, ytest). Then, a PLS−DA model was fit using the training set of samples to obtain a fitted response (ŷtrain), which was then evaluated using the testing set of samples, that is, (ŷtest). The MCCV procedure was repeated N times (i.e., N = 100), and the average and distribution of predictive performance was calculated (i.e., AUC using ŷtest). In this study, MCCV combined with ROC curves was used to estimate the performance of the PLS−DA model using a selected set of metabolites described below, and MCCV was applied using 70% of the data as the training set, while the remaining 30% served as the testing set, and employing 100 iterations. For each iteration, three specificities of the training set, 0.95, 0.85, and 0.75, were used to determine the thresholds of PLS−DA predicted Y values. The same thresholds were then applied to the test set to determine sensitivities and specificities. The sample classification can be correctly assigned, termed “true class,” or the sample class information can be randomly permuted, which is referred to as a “random permutation.”

Liquid Chromatography Conditions

The LC system was composed of two Agilent 1260 binary pumps, an Agilent 1260 autosampler, and Agilent 1290 column compartment containing a column-switching valve (Agilent Technologies, Santa Clara, CA). Each sample was injected twice, 10 μL for analysis using negative ionization mode and 2 μL for analysis using positive ionization mode. Both chromatographic separations were performed using hydrophilic interaction chromatography (HILIC) on two SeQuant ZIC-cHILIC columns (150 × 2.1 mm, 3.0 μm particle size, Merck KGaA, Darmstadt, Germany) connected in parallel. Our setup allows one column to perform the separation, while the other column is reconditioned and readied for the next injection. The flow rate was 0.300 mL/min, the autosampler temperature was kept at 4 °C, the column compartment was set at 40 °C, and the total separation time for both ionization modes was 20 min. The mobile phase was composed of Solvents A (5 mM ammonium acetate in 90% H2O/10% acetonitrile +0.2% acetic acid) and B (5 mM ammonium acetate in 90% acetonitrile/ 10% H2O + 0.2% acetic acid). The gradient conditions for both separations were identical and are shown in Supplemental Table S2 in the Supporting Information. The metabolite identities were confirmed by spiking the pooled serum sample used for method development with mixtures of standard compounds (each mixture contained five standard metabolites). However, some metabolites that could not be well separated and had similar m/z values ( 1 metabolite can be seen in Figure 1. On the basis of the VIP selection, a second PLS−DA model was built using only these metabolites that had VIP scores greater than 1. To evaluate the diagnostic power of the potential metabolic markers, ROC curves (Supplemental Figure S6 in the Supporting Information) were generated. For differentiating CRC patients from healthy controls, our VIP metabolites model showed a sensitivity of 0.80, specificity of 0.83, and AUROC of 0.89 (R2X = 0.95, R2Y = 0.48, and Q2Y = 0.35). For differentiating CRC patients from polyps patients, sensitivity of 0.92, specificity of 0.86, and AUROC of 0.94 from the ROC were obtained (R2X = 0.95, R2Y = 0.59, and Q2Y = 0.44). Also, to examine the robustness of our PLS−DA-based CRC diagnostic models, MCCV21 was applied to compare the PLS−DA models using the true sample classifications to those with randomly permuted sample class information; superior sensitivity observed in the true sample classifications showed the robust diagnostic power of this metabolic profiling approach (Supplemental Figure S6 in the Supporting Information).

Targeted Metabolic Profiles of CRC versus Polyp Patients and Healthy Controls

In the current study, we used a targeted LC−MS/MS approach for comprehensive CRC serum metabolic profiling. Using this metabolic profiling system, targeted analysis of 158 MRM transitions was achieved for metabolites of 20 different chemical classes (such as amino acids, carboxylic acids, pyridines, etc.), which are located in 25 important metabolic pathways (e.g., TCA cycle, amino acid metabolism, glycolysis, purine and pyrimidine metabolism, urea cycle), in both positive and negative ionization modes (Supplemental Table S1 in the Supporting Information). Two additional stable isotope-labeled internal standards (L-tyrosine-13C2 and sodium-L-lactate-13C3) were also monitored to ensure instrument performance. A CV of 7.6% for L-tyrosine-13C2 and CV of 5.8% for sodium-Llactate-13C3 was obtained in the QC samples; when looking at the ratio between unlabeled metabolites versus labeled internal standards, the CVs were even better: 4.4% for 12C/13C tyrosine and 2.5% for 12C/13C lactate in all QC runs, Supplemental Figure S2 in the Supporting Information. A heat map based on z scores showing the biological variation of the metabolite levels across three groups can also be seen in Figure S3 in the Supporting Information. In total, we reliably detected 113 metabolites out of 158 targeted MRM transitions, with a median QC CV of 8% (ranging from 5 to 31%, with ∼80% metabolites having CV < 15%; see Supplemental Figure S2 in the Supporting Information for CV distributions). As shown in Table 2, 42 of these metabolites showed statistical significance between CRC patients and healthy controls, 48 showed statistically significant differences between CRC and polyp patients, and 8 showed statistically significant differences between healthy controls and polyp patients based on the Mann−Whitney U-test with a p < 0.05. Large fold changes (FCs), calculated based on mean ratios for CRC/healthy, CRC/polyps, or healthy/polyps as appropriate were also observed. Eleven metabolites had p < 0.001(with FC ranging from 0.75 to 2.73) when comparing the CRC patients to healthy controls, and 13 metabolites had p < 0.001(with FC ranging from 0.77 to 3.22) when comparing the CRC patients to polyp patients. Biomarker Selection, Model Setup, and Cross Validation

Initially, individual metabolites that had p < 0.05 were selected as potential biomarker candidates. AUROC, sensitivity, and specificity values for each metabolite were calculated while comparing CRC patients with healthy controls and CRC patients with polyp patients, respectively, and these values are listed in Supplemental Tables S3 and S4 in the Supporting Information. As evidenced in these two tables, no single metabolite proved to be sufficiently sensitive and specific by itself to distinguish CRC patients from either healthy controls or polyp patients. (In general, the AUROC values are below 0.7 for each metabolite.) PLS−DA models established by leave one out cross validation were then applied to identify groups of 4124

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

performance than the VIP metabolite model alone (Figure S6 in the Supporting Information), which suggests that inclusion of clinical factors could improve an already well-performed VIP metabolite model, therefore increasing the diagnostic power of this targeted serum metabolic profiling approach for CRC. MCCV was again applied, and the advanced performance of the true class models over the random permutation model was obtained as anticipated (Figure 2, sensitivity values after MCCV were shown with average and standard deviation), indicating the robustness of this combined metabolite and clinical model approach. After the enhanced metabolite-based prediction model was established, subgroups of CRC patients in this study were analyzed using the previously described models to evaluate the diagnostic power for specific CRC disease type and stage. As can be seen in Table 3 for the enhanced VIP metabolite model, all AUROCs were equal to or greater than 0.93. The models have slightly better diagnostic power in colon cancer detection compared to rectal cancer and also have varying performances depending on different stages of CRC, with the highest performance seen for stage-IV CRC diagnosis.



DISCUSSION During the past decade, interest has grown in applying massspectrometry-based metabolic profiling for analyzing and monitoring cancer-related metabolic alterations, and, in particular, to thereby provide sensitive and valuable diagnostic information.12a,23 In the current investigation, we explored the combination of targeted metabolic profiling with multivariate statistical analysis for the discovery of sensitive and specific metabolite biomarkers for CRC detection. We have used this particular method to monitor 158 metabolites from 25 metabolic pathways of potential significance by LC−MS/MS using both positive and negative ionization modes and MRM methods. The 158 metabolites were selected according to the established knowledge from previous studies24 and our own previous work with regard to key metabolites of interest from important biological pathways as well as a consideration of the detection ability and measurement reliability for each of the metabolites using our particular LC−MS instrument. On the basis of our multiple step biomarker selection, model construction, and cross validation, we successfully demonstrated the robust diagnostic power of this metabolic profiling approach in this study composed of 234 patients. To date, a number of studies have performed massspectrometry-based methods (such as GC−MS and LC− QTOF−MS) for detecting the serum metabolic alterations from CRC patients.14b,c,25 These studies have typically used global metabolic profiling methods to measure as many features that can be captured by the analytical platform, which can make them less reliable and robust. In contrast, the very reproducible targeted LC−MS/MS metabolite profiling approach we applied in this current study has a median CV value of ∼8% and has not been reported in any previous CRC metabolic profiling study. Additionally, instead of applying database searches for compound annotation,14b,c we tested all targeted metabolites included in this study with pure standard compounds. The possibility of an unknown compound with a similar m/z and RT being detected simultaneously is quite low, although it still exists. It is also worth noting that there are only a few studies available so far regarding the comparison of metabolic shifts from healthy controls to polyp patients and then to CRC patients,26 and none of these studies used serum samples. In

Figure 1. Bar graphs of metabolites with PLS−DA VIP scores >1 in the comparison of (A) CRC vs healthy controls and (B) CRC vs polyp patients. (Error bars show standard error of the mean.)

Clinical factors, such as gender, age, medication, and substance status have often been incorporated to build predictive or diagnostic clinical models, and such variables have recently been used to enhance metabolite biomarker models.22 To enhance our current VIP metabolite model, four general clinical factors (age, gender, smoking, and alcohol status) were chosen to be candidates for inclusion in the model. The enhanced metabolite model (Figure 2) showed excellent AUROC (0.93) for differentiating CRC patients from healthy controls, with sensitivity of 0.96 and specificity of 0.80 (R2X = 0.95, R2Y = 0.57, and Q2Y = 0.40). An improved AUROC (0.95) for differentiating CRC patients from polyp patients was also obtained, with sensitivity of 0.89 and specificity of 0.88 (R2X = 0.83, R2Y = 0.60, and Q2Y = 0.41). The model incorporating these four clinical parameters showed better 4125

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

Figure 2. Left: ROC curves for the enhanced PLS−DA model combining metabolites (p < 0.05 and VIP score >1) and clinical parameters (age, gender, smoking status, and alcohol status). Right: Monte Carlo cross-validation (MCCV) results of enhanced PLS−DA models. True: true class model; random: random permutation model. For the MCCV, sensitivities were calculated for test specificities of 0.95, 0.85, and 0.75. (A) CRC vs healthy controls, AUROC = 0.93. (B) CRC vs polyp patients, AUROC = 0.95.

To understand the possible connections among these serum metabolites, we constructed metabolic pathway maps based on information obtained from the Kyoto Encyclopedia of Genes and Genomes Web site (www.genome.jp/kegg/), and these maps are shown in Figures 3 and 4. For example, in examining central carbon metabolism, including glycolysis, the tricarboxylic acid (TCA) cycle, and other related pathways, 10 metabolites were altered significantly (Figure 3). Mean glucose levels from CRC patients are significantly higher than in healthy controls, which has previously been related to a higher risk of CRC,27 and significantly impaired glucose metabolism has also been previously reported in CRC cases.28 Meanwhile, significantly increased pyruvate and lactate levels in CRC patients were also detected in our study, which matched previous reports.14a,c Increased glycolysis is proposed to be associated with many tumors and with cancer cell growth and forms part of the well-known Warburg effect.29 Three TCA cycle metabolites were detected as having significant differences in the pairwise comparison of CRC with the other two groups; 2-oxoglutarate was decreased in CRC patients, indicating that the TCA cycle may be impaired and leading to reduced mitochondrial respiration. Fumarate and oxaloacetate were, however, found to be slightly increased in CRC patients compared with either healthy controls or polyp patients. Interestingly, significantly increased fumarate levels were also reported by a previous metabolic study,30 which was suggested as part of a typical metabolic fingerprint of hypoxic cells. The authors from that study also proposed that so-called fumarate respiration, which is a known activity of some parasites and bacteria, contributes greatly to the energy generation of cancer

Table 3. Performance of PLS−DA Prediction Models for Different CRC Diagnostic Groups and Cancer Stages CRC compared to healthy controls

colon cancer

rectal cancer

AUROC sensitivity specificity CRC compared to polyp patients

0.96 0.95 0.88 colon cancer

0.93 0.93 0.82 rectal cancer

stage I/II

AUROC sensitivity specificity

0.96 0.92 0.91

0.95 0.89 0.95

0.97 0.95 0.92

stage I/II 0.93 0.95 0.82

stage III stage IV 0.93 0.76 0.95

0.99 0.94 0.94

stage III stage IV 0.94 0.94 0.82

0.99 1.00 0.96

our current study, we performed pairwise comparisons of serum metabolites from CRC patients, polyp patients, and healthy controls and observed significant alterations in a variety of the metabolites detected (e.g., amino acids, carboxylic acid, fatty acids, and nucleosides; see Supplemental Table S5 in the Supporting Information for detailed metabolite classifications). Furthermore, significantly altered serum metabolites with p < 0.05 (Mann−Whitney U test) and VIP > 1 in the first PLS−DA model were selected in this study and compared between different groups (Figure 1). Meanwhile, efforts were also made in this study to look for possible enhancements to the VIPbased metabolite model using four clinical factors, including age, gender, smoking, and alcohol status. After adding these clinical factors to the metabolites selected by VIP, improved AUROC, sensitivity, or selectivity was observed in the cross validated PLS−DA model. 4126

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

Figure 3. Metabolic network of significantly changed metabolites in central carbon metabolism (glycolysis, TCA, and other related pathways). Bar chart left to right: CRC (blue bar), healthy controls (red bar), and polyp patients (green bar); the Y axis represents relative abundance of MS signals (normalized to the highest peaks in comparison). Dashed lines surrounding compounds mean measured but not significant between any two groups. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

as significantly decreased in CRC patients compared with both healthy controls and polyp patients, while orotate was measured to be higher in CRC than polyp patients. Most of the key serum metabolite biomarker candidates (determined by the criteria of both p < 0.05 and VIP score >1) discovered in this study are of biological importance and have been proposed as CRC-related compounds. For example, glycocholate and glycochenodeoxycholate, two intermediate metabolites between primary bile synthesis and secondary bile synthesis, have significantly higher concentrations in CRC patients compared with healthy controls or polyp patients (in agreement with a previous report36), suggesting significant increases in primary and secondary bile acids in CRC patients. Down regulation of histidine was observed in our study and by others,14b,c,37 and this down regulation may be due to the acceleration of decarboxylation from histidine to histamine in CRC patients, which is caused by the increased activity of histidine decarboxylase.31 Increased concentrations of hydroxylproline was also observed in CRC patients, and a previous study suggested that the excessive degradation of collagen in these patients may be the cause.30 Nevertheless, we discovered some new potential CRC serum biomarkers, including glyceraldehyde, glycocholate, linolenic acid, and leucic acid, that have not been previously reported. Further study is needed to identify the biological roles of these metabolites in CRC. Besides the diagnostic power of metabolite biomarkers for comparing CRC patients with healthy controls and polyp patients, we also carefully examined the metabolite changes in CRC patients with different disease stages and observed that three significantly altered serum metabolites, namely, glutamic acid, adenosine, and aspartic acid, consistently changed over the different cancer stages. (See Figure 5.) These metabolites could

cells under conditions of glucose deprivation and severe hypoxia.31 Oxaloacetate has been reported to contribute significantly to aspartic acid production by transamination,14b while an increase in aspartic acid levels was reported in various studies and was proposed as one of the nutrients that cancer cells prefer.14d Amino acid, purine, and pyrimidine metabolism pathways were also significantly impacted by CRC, as can be seen in Figure 4. Cancer cells are known to use some amino acids as an energy source;32 alterations of amino acid levels therefore can be indicative of cancer cell activity. For example, serum alanine, glutamine, lysine, creatinine, asparagine, and tryptophan levels decreased significantly, while levels of serum glutamate, proline, asparate, and hydroxylproline increased significantly in CRC patients compared with either healthy control or polyp patients in our study, which is in agreement with previous serum studies.14b,c,25 It is interesting to note that these observations are somewhat different from a CRC study focused on tissue samples. In that study, most of the free amino acids were higher in CRC due to possible up-regulation of cell amino acid biosynthesis and cell autophagy.30 Metabolite level changes in tissue and serum are not always correlated.33 Altered purine metabolism has been reported in other types of cancers, such as liver cancer, and enzyme pattern imbalances and other changes in purine metabolism have been linked to disease progression.34 On the basis of the observed significant changes in adenosine, urate, adenylosuccinate, and allantoin levels between CRC and the other two patient groups, the impact of CRC on purine metabolism can be observed. Pyrimidine metabolism, which has close connection to glutamine metabolism (Figure 4), can also be influenced by CRC.14d,35 In our study, several pyrimidine metabolites, such as uridine and 2-deoxyuridine, were detected 4127

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

Figure 4. Metabolic network of the significantly changed metabolites involved in amino acid, purine, and pyrimidine metabolisms. Bar charts left to right: CRC (blue bar), healthy controls (red bar), and polyp patients (green bar); the Y axis represents relative abundance of MS signals (normalized to the highest peaks in comparison). Dashed lines surrounding compounds indicates metabolites that were measured but not significant between any of two groups. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

Figure 5. Box plots of metabolites that significantly changed (p < 0.05) over different CRC stages.



be further explored in the future for the potential differentiation between early-stage and late-stage CRC. To the best of our knowledge, this is the first report in which an LC−MS/MS targeted serum metabolic profiling approach has been applied for the comparison of CRC patients to both healthy controls and polyp patients, and our results demonstrate a panel of 13 serum metabolites for the differentiation of CRC patients and healthy controls and 14 for the differentiation of CRC and polyp patients. With the inclusion of four clinical factors (age, gender, smoking, and alcohol status), this metabolic profile can potentially serve as a novel disease biomarker panel for CRC diagnosis.

ASSOCIATED CONTENT

S Supporting Information *

List of targeted metabolites in this study, their optimized MS parameters, and associated major metabolic pathways. LC gradient conditions. Diagnostic performance of metabolites with p < 0.05. Percentages of different classes of metabolites that have significant alterations (p < 0.05). Flowchart describing biomarker selection, model development, and validation. Distribution for CV values of all measured metabolites along with variations of two example metabolites. Z-score heatmap showing biological variation. ROC curves of PLS−DA models using all metabolites with U-test p < 0.05 between healthy controls and polyp patients. PLS−DA VIP plots indicate important metabolite biomarker candidates. ROC curves of 4128

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

(11) (a) Jain, M.; Nilsson, R.; Sharma, S.; Madhusudhan, N.; Kitami, T.; Souza, A. L.; Kafri, R.; Kirschner, M. W.; Clish, C. B.; Mootha, V. K. Metabolite Profiling Identifies a Key Role for Glycine in Rapid Cancer Cell Proliferation. Science 2012, 336 (6084), 1040−1044. (b) Sreekumar, A.; Poisson, L. M.; Rajendiran, T. M.; Khan, A. P.; Cao, Q.; Yu, J.; Laxman, B.; Mehra, R.; Lonigro, R. J.; Li, Y.; Nyati, M. K.; Ahsan, A.; Kalyana-Sundaram, S.; Han, B.; Cao, X.; Byun, J.; Omenn, G. S.; Ghosh, D.; Pennathur, S.; Alexander, D. C.; Berger, A.; Shuster, J. R.; Wei, J. T.; Varambally, S.; Beecher, C.; Chinnaiyan, A. M. Metabolomic profiles delineate potential role for sarcosine in prostate cancer progression. Nature 2009, 457 (7231), 910−914. (c) Wise, D. R.; Thompson, C. B. Glutamine addiction: a new therapeutic target in cancer. Trends Biochem. Sci. 2010, 35 (8), 427− 433. (d) Munoz-Pinedo, C.; El Mjiyad, N.; Ricci, J. E. Cancer metabolism: current perspectives and future directions. Cell Death Dis. 2012, 3, e248. (e) Gross, S.; Cairns, R. A.; Minden, M. D.; Driggers, E. M.; Bittinger, M. A.; Jang, H. G.; Sasaki, M.; Jin, S.; Schenkein, D. P.; Su, S. M.; Dang, L.; Fantin, V. R.; Mak, T. W. Cancer-associated metabolite 2-hydroxyglutarate accumulates in acute myelogenous leukemia with isocitrate dehydrogenase 1 and 2 mutations. J. Exp. Med. 2010, 207 (2), 339−344. (12) (a) Patti, G. J.; Yanes, O.; Siuzdak, G. Innovation: Metabolomics: the apogee of the omics trilogy. Nat. Rev. Mol. Cell Biol. 2012, 13 (4), 263−269. (b) Gu, H.; Gowda, G. A. N.; Raftery, D. Metabolic profiling: are we en route to better diagnostic tests for cancer? Future Oncol. 2012, 8 (10), 1207−1210. (c) Gowda, G. A. N.; Zhang, S.; Gu, H.; Asiago, V.; Shanaiah, N.; Raftery, D. Metabolomicsbased methods for early disease diagnostics. Expert Rev. Mol. Diagn. 2008, 8 (5), 617−633. (d) Scalbert, A.; Brennan, L.; Fiehn, O.; Hankemeier, T.; Kristal, B.; Ommen, B.; Pujos-Guillot, E.; Verheij, E.; Wishart, D.; Wopereis, S. Mass-spectrometry-based metabolomics: limitations and recommendations for future progress with particular focus on nutrition research. Metabolomics 2009, 5 (4), 435−458. (e) Nicholson, J. K.; Holmes, E.; Kinross, J. M.; Darzi, A. W.; Takats, Z.; Lindon, J. C. Metabolic phenotyping in clinical and surgical environments. Nature 2012, 491 (7424), 384−392. (f) Fan, T.-M.; Lane, A. NMR-based stable isotope resolved metabolomics in systems biochemistry. J. Biomol. NMR 2011, 49 (3−4), 267−280. (g) Reaves, M. L.; Rabinowitz, J. D. Metabolomics in systems microbiology. Curr. Opin. Biotechnol. 2011, 22 (1), 17−25. (h) Bain, J. R.; Stevens, R. D.; Wenner, B. R.; Ilkayeva, O.; Muoio, D. M.; Newgard, C. B. Metabolomics Applied to Diabetes Research: Moving From Information to Knowledge. Diabetes 2009, 58 (11), 2429−2443. (i) Yanes, O.; Tautenhahn, R.; Patti, G. J.; Siuzdak, G. Expanding Coverage of the Metabolome for Global Metabolite Profiling. Anal. Chem. 2011, 83 (6), 2152−2161. (13) Spratlin, J. L.; Serkova, N. J.; Eckhardt, S. G. Clinical Applications of Metabolomics in Oncology: A Review. Clin. Cancer Res. 2009, 15 (2), 431−440. (14) (a) Qiu, Y.; Cai, G.; Su, M.; Chen, T.; Zheng, X.; Xu, Y.; Ni, Y.; Zhao, A.; Xu, L. X.; Cai, S.; Jia, W. Serum Metabolite Profiling of Human Colorectal Cancer Using GC−TOFMS and UPLC− QTOFMS. J. Proteome Res. 2009, 8 (10), 4844−4850. (b) Nishiumi, S.; Kobayashi, T.; Ikeda, A.; Yoshie, T.; Kibi, M.; Izumi, Y.; Okuno, T.; Hayashi, N.; Kawano, S.; Takenawa, T.; Azuma, T.; Yoshida, M.; Novel Serum, A. Metabolomics-Based Diagnostic Approach for Colorectal Cancer. PLoS One 2012, 7 (7), e40459. (c) Tan, B.; Qiu, Y.; Zou, X.; Chen, T.; Xie, G.; Cheng, Y.; Dong, T.; Zhao, L.; Feng, B.; Hu, X.; Xu, L. X.; Zhao, A.; Zhang, M.; Cai, G.; Cai, S.; Zhou, Z.; Zheng, M.; Zhang, Y.; Jia, W. Metabonomics Identifies Serum Metabolite Markers of Colorectal Cancer. J. Proteome Res. 2013, 12 (6), 3000−3009. (d) Denkert, C.; Budczies, J.; Weichert, W.; Wohlgemuth, G.; Scholz, M.; Kind, T.; Niesporek, S.; Noske, A.; Buckendahl, A.; Dietel, M.; Fiehn, O. Metabolite profiling of human colon carcinoma - deregulation of TCA cycle and amino acid turnover. Mol. Cancer 2008, 7 (1), 72. (15) Chan, E. C. Y.; Koh, P. K.; Mal, M.; Cheah, P. Y.; Eu, K. W.; Backshall, A.; Cavill, R.; Nicholson, J. K.; Keun, H. C. Metabolic Profiling of Human Colorectal Cancer Using High-Resolution Magic

PLS−DA models using all metabolites with both U-test p < 0.05 and VIP scores >1. MCCV results of proposed PLS−DA models. This material is available free of charge via the Internet at http://pubs.acs.org.



AUTHOR INFORMATION

Corresponding Author

*Tel: 206-543-9709. Fax: 206-616-4819. E-mail: draftery@uw. edu. Author Contributions #

J.Z. and D.D. contributed equally to this project.

Notes

The authors declare the following competing financial interest(s): Daniel Raftery reports holding equity and an executive position at Matrix-Bio, Inc.



ACKNOWLEDGMENTS This work was supported by AMRMC grant W81XWH-10-10540. We thank Dr. Li Yuan Bermel for assistance with the CCE project sample collection bank. The China Scholarship Council is also gratefully acknowledged (Grant to L.D.).



REFERENCES

(1) Weitz, J.; Koch, M.; Debus, J.; Höhler, T.; Galle, P. R.; Büchler, M. W. Colorectal cancer. Lancet 2005, 365 (9454), 153−165. (2) Siegel, R.; Ma, J.; Zou, Z.; Jemal, A. Cancer statistics, 2014. CaCancer J. Clin. 2014, 64 (1), 9−29. (3) Rex, D. K.; Johnson, D. A.; Anderson, J. C.; Schoenfeld, P. S.; Burke, C. A.; Inadomi, J. M. American college of gastroenterology guidelines for colorectal cancer screening 2008. Am. J. Gastroenterol. 2009, 104 (3), 739−750. (4) Bond, J. H. Colon polyps and cancer. Endoscopy 2003, 35 (01), 27−35. (5) Levin, B.; Lieberman, D. A.; McFarland, B.; Smith, R. A.; Brooks, D.; Andrews, K. S.; Dash, C.; Giardiello, F. M.; Glick, S.; Levin, T. R.; Pickhardt, P.; Rex, D. K.; Thorson, A.; Winawer, S. J. Screening and surveillance for the early detection of colorectal cancer and adenomatous polyps, 2008: A joint guideline from the American Cancer Society, the US Multi-Society Task Force on Colorectal Cancer, and the American College of Radiology. Ca-Cancer J. Clin. 2008, 58 (3), 130−160. (6) Taylor, D. P.; Cannon-Albright, L. A.; Sweeney, C.; Williams, M. S.; Haug, P. J.; Mitchell, J. A.; Burt, R. W. Comparison of compliance for colorectal cancer screening and surveillance by colonoscopy based on risk. Genet. Med. 2011, 13 (8), 737−743. (7) Cancer Facts & Figures 2013; American Cancer Society: Atlanta, GA, 2013. (8) Ahlquist, D. A.; Zou, H.; Domanico, M.; Mahoney, D. W.; Yab, T. C.; Taylor, W. R.; Butz, M. L.; Thibodeau, S. N.; Rabeneck, L.; Paszat, L. F.; Kinzler, K. W.; Vogelstein, B.; Bjerregaard, N. C.; Laurberg, S.; Sørensen, H. T.; Berger, B. M.; Lidgard, G. P. NextGeneration Stool DNA Test Accurately Detects Colorectal Cancer and Large Adenomas. Gastroenterology 2012, 142 (2), 248−256. (9) (a) Link, A.; Balaguer, F.; Shen, Y.; Nagasaka, T.; Lozano, J. J.; Boland, C. R.; Goel, A. Fecal MicroRNAs as Novel Biomarkers for Colon Cancer Screening. Cancer Epidemiol., Biomarkers Prev. 2010, 19 (7), 1766−1774. (b) Ahmed, F. E.; Ahmed, N. C.; Vos, P. W.; Bonnerup, C.; Atkins, J. N.; Casey, M.; Nuovo, G. J.; Naziri, W.; Wiley, J. E.; Mota, H.; Allison, R. R. Diagnostic MicroRNA markers to screen for sporadic human colon cancer in stool: I. Proof of principle. Cancer Genomics Proteomics 2013, 10 (3), 93−113. (10) Imperiale, T. F.; Ransohoff, D. F.; Itzkowitz, S. H.; Levin, T. R.; Lavin, P.; Lidgard, G. P.; Ahlquist, D. A.; Berger, B. M. Multitarget Stool DNA Testing for Colorectal-Cancer Screening. N. Engl. J. Med. 2014, 370, 1287−1297. 4129

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130

Journal of Proteome Research

Article

Angle Spinning Nuclear Magnetic Resonance (HR-MAS NMR) Spectroscopy and Gas Chromatography Mass Spectrometry (GC/ MS). J. Proteome Res. 2008, 8 (1), 352−361. (16) Ma, Y.-L.; Qin, H.-L.; Liu, W.-J.; Peng, J.-Y.; Huang, L.; Zhao, X.-P.; Cheng, Y.-Y. Ultra-High Performance Liquid Chromatography− Mass Spectrometry for the Metabolomic Analysis of Urine in Colorectal Cancer. Dig. Dis. Sci. 2009, 54 (12), 2655−2662. (17) Li, F.; Qin, X.; Chen, H.; Qiu, L.; Guo, Y.; Liu, H.; Chen, G.; Song, G.; Wang, X.; Li, F.; Guo, S.; Wang, B.; Li, Z. Lipid profiling for early diagnosis and progression of colorectal cancer using directinfusion electrospray ionization Fourier transform ion cyclotron resonance mass spectrometry. Rapid Commun. Mass Spectrom. 2013, 27 (1), 24−34. (18) (a) Ritchie, S.; Ahiahonu, P.; Jayasinghe, D.; Heath, D.; Liu, J.; Lu, Y.; Jin, W.; Kavianpour, A.; Yamazaki, Y.; Khan, A.; Hossain, M.; Su-Myat, K.; Wood, P.; Krenitsky, K.; Takemasa, I.; Miyake, M.; Sekimoto, M.; Monden, M.; Matsubara, H.; Nomura, F.; Goodenowe, D. Reduced levels of hydroxylated, polyunsaturated ultra long-chain fatty acids in the serum of colorectal cancer patients: implications for early screening and detection. BMC Med. 2010, 8 (1), 13. (b) Ritchie, S. A.; Tonita, J.; Alvi, R.; Lehotay, D.; Elshoni, H.; Myat, S.; McHattie, J.; Goodenowe, D. B. Low-serum GTA-446 anti-inflammatory fatty acid levels as a new risk factor for colon cancer. Int. J. Cancer 2013, 132 (2), 355−362. (19) Xu, Q.-S.; Liang, Y.-Z. Monte Carlo cross validation. Chemom. Intell. Lab. Syst. 2001, 56 (1), 1−11. (20) Chong, I.-G.; Jun, C.-H. Performance of some variable selection methods when multicollinearity is present. Chemom. Intell. Lab. Syst. 2005, 78 (1−2), 103−112. (21) (a) Rocha, C. u. M.; Carrola, J.; Barros, A. n. S.; Gil, A. M.; Goodfellow, B. J.; Carreira, I. M.; Bernardo, J. o.; Gomes, A.; Sousa, V.; Carvalho, L.; Duarte, I. F. Metabolic Signatures of Lung Cancer in Biofluids: NMR-Based Metabonomics of Blood Plasma. J. Proteome Res. 2011, 10 (9), 4314−4324. (b) Wei, S.; Suryani, Y.; Gowda, G. A. N.; Skill, N.; Maluccio, M.; Raftery, D. Differentiating Hepatocellular Carcinoma from Hepatitis C Using Metabolite Profiling. Metabolites 2012, 2 (4), 701−716. (22) Rhee, E. P.; Ho, J. E.; Chen, M.-H.; Shen, D.; Cheng, S.; Larson, M. G.; Ghorbani, A.; Shi, X.; Helenius, I. T.; O’Donnell, C. J.; Souza, A. L.; Deik, A.; Pierce, K. A.; Bullock, K.; Walford, G. A.; Vasan, R. S.; Florez, J. C.; Clish, C.; Yeh, J. R. J.; Wang, T. J.; Gerszten, R. E. A Genome-wide Association Study of the Human Metabolome in a Community-Based Cohort. Cell Metab. 2013, 18 (1), 130−143. (23) (a) Benjamin, D. I.; Cravatt, B. F.; Nomura, D. K. Global Profiling Strategies for Mapping Dysregulated Metabolic Pathways in Cancer. Cell Metab. 2012, 16 (5), 565−577. (b) Gu, H.; Pan, Z.; Xi, B.; Asiago, V.; Musselman, B.; Raftery, D. Principal component directed partial least squares analysis for combining nuclear magnetic resonance and mass spectrometry data in metabolomics: Application to the detection of breast cancer. Anal. Chim. Acta 2011, 686 (1−2), 57−63. (24) (a) Bajad, S. U.; Lu, W.; Kimball, E. H.; Yuan, J.; Peterson, C.; Rabinowitz, J. D. Separation and quantitation of water soluble cellular metabolites by hydrophilic interaction chromatography-tandem mass spectrometry. J. Chromatogr. A 2006, 1125 (1), 76−88. (b) Yuan, M.; Breitkopf, S. B.; Yang, X.; Asara, J. M. A positive/negative ionswitching, targeted mass spectrometry-based metabolomics platform for bodily fluids, cells, and fresh and fixed tissue. Nat. Protoc. 2012, 7 (5), 872−881. (c) Wei, R.; Li, G.; Seymour, A. B. High-Throughput and Multiplexed LC/MS/MRM Method for Targeted Metabolomics. Anal. Chem. 2010, 82 (13), 5527−5533. (25) Leichtle, A.; Nuoffer, J.-M.; Ceglarek, U.; Kase, J.; Conrad, T.; Witzigmann, H.; Thiery, J.; Fiedler, G. Serum amino acid profiles and their alterations in colorectal cancer. Metabolomics 2012, 8 (4), 643− 653. (26) (a) Ong, E. S.; Zou, L.; Li, S.; Cheah, P. Y.; Eu, K. W.; Ong, c. N. Metabolic profiling in colorectal cancer reveals signature metabolic shifts during tumorigenesis. Mol. Cell. Proteomics 2010, DOI: 10.1074/ mcp.M900551-MCP200. (b) Eisner, R.; Greiner, R.; Tso, V.; Wang,

H.; Fedorak, R. N. A Machine-Learned Predictor of Colonic Polyps Based on Urinary Metabolomics. BioMed. Res. Int. 2013, 2013, 11. (27) Schoen, R. E.; Tangen, C. M.; Kuller, L. H.; Burke, G. L.; Cushman, M.; Tracy, R. P.; Dobs, A.; Savage, P. J. Increased blood glucose and insulin, body Size, and incident colorectal cancer. J. Natl. Cancer Inst. 1999, 91 (13), 1147−1154. (28) Ehrmann-Jósko, A.; Siemińska, J.; Górnicka, B.; ZiarkiewiczWróblewska, B.; Ziółkowski, B.; Muszyński, J. Impaired glucose metabolism in colorectal cancer. Scand. J. Gastroenterol. 2006, 41 (9), 1079−1086. (29) (a) Warburg, O. On the Origin of Cancer Cells. Science 1956, 123 (3191), 309−314. (b) Vander Heiden, M. G.; Cantley, L. C.; Thompson, C. B. Understanding the Warburg Effect: The Metabolic Requirements of Cell Proliferation. Science 2009, 324 (5930), 1029− 1033. (30) Hirayama, A.; Kami, K.; Sugimoto, M.; Sugawara, M.; Toki, N.; Onozuka, H.; Kinoshita, T.; Saito, N.; Ochiai, A.; Tomita, M.; Esumi, H.; Soga, T. Quantitative Metabolome Profiling of Colon and Stomach Cancer Microenvironment by Capillary Electrophoresis Time-of-Flight Mass Spectrometry. Cancer Res. 2009, 69 (11), 4918−4925. (31) Garcia-Caballero, M.; Neugebauer, E.; Campos, R.; Nunez de Castro, I.; Vara-Thorbeck, C. Increased histidine decarboxylase (HDC) activity in human colorectal cancer: Results of a study on ten patients. Agents Actions 1988, 23 (3−4), 357−360. (32) Argilés, J.; Azcón-Bieto, J. The metabolic environment of cancer. Mol. Cell. Biochem. 1988, 81 (1), 3−17. (33) Huang, Q.; Tan, Y.; Yin, P.; Ye, G.; Gao, P.; Lu, X.; Wang, H.; Xu, G. Metabolic Characterization of Hepatocellular Carcinoma Using Nontargeted Tissue Metabolomics. Cancer Res. 2013, 73 (16), 4992− 5002. (34) Weber, G. Enzymes of purine metabolism in cancer. Clin. Biochem. 1983, 16 (1), 57−63. (35) Israel, M.; Schwartz, L. The metabolic advantage of tumor cells. Mol. Cancer 2011, 10 (1), 70. (36) Tocchi, A.; Basso, L.; Costa, G.; Lepre, L.; Liotta, G.; Mazzoni, G.; Sita, A.; Tagliacozzo, S. Is there a causal connection between bile acids and colorectal cancer? Surg. Today 1996, 26 (2), 101−104. (37) Qiu, Y.; Cai, G.; Su, M.; Chen, T.; Liu, Y.; Xu, Y.; Ni, Y.; Zhao, A.; Cai, S.; Xu, L. X.; Jia, W. Urinary Metabonomic Study on Colorectal Cancer. J. Proteome Res. 2010, 9 (3), 1627−1634.

4130

dx.doi.org/10.1021/pr500494u | J. Proteome Res. 2014, 13, 4120−4130