Metabolic Profiling of Human Colorectal Cancer Using High-Resolution Magic Angle Spinning Nuclear Magnetic Resonance (HR-MAS NMR) Spectroscopy and Gas Chromatography Mass Spectrometry (GC/MS) Eric Chun Yong Chan,*,†,§ Poh Koon Koh,‡ Mainak Mal,† Peh Yean Cheah,‡ Kong Weng Eu,‡ Alexandra Backshall,§ Rachel Cavill,§ Jeremy K. Nicholson,§ and Hector C. Keun*,§ Department of Pharmacy, Faculty of Science, National University of Singapore, 18 Science Drive 4, Singapore 117543, Singapore, Colorectal Cancer Research Laboratory, Department of Colorectal Surgery, Singapore General Hospital, Outram Road, Block 6 Level 7, Singapore 169608, Singapore, and Department of Biomolecular Medicine, Division of Surgery, Oncology, Reproductive Biology and Anaesthetics, Faculty of Medicine, Imperial College London, Sir Alexander Fleming Building, South Kensington, London SW7 2AZ, United Kingdom Received August 12, 2008
Current clinical strategy for staging and prognostication of colorectal cancer (CRC) relies mainly upon the TNM or Duke system. This clinicopathological stage is a crude prognostic guide because it reflects in part the delay in diagnosis in the case of an advanced cancer and gives little insight into the biological characteristics of the tumor. We hypothesized that global metabolic profiling (metabonomics/ metabolomics) of colon mucosae would define metabolic signatures that not only discriminate malignant from normal mucosae, but also could distinguish the anatomical and clinicopathological characteristics of CRC. We applied both high-resolution magic angle spinning nuclear magnetic resonance (HR-MAS NMR) and gas chromatography mass spectrometry (GC/MS) to analyze metabolites in biopsied colorectal tumors and their matched normal mucosae obtained from 31 CRC patients. Orthogonal partial leastsquares discriminant analysis (OPLS-DA) models generated from metabolic profiles obtained by both analytical approaches could robustly discriminate normal from malignant samples (Q2 > 0.50, Receiver Operator Characteristic (ROC) AUC >0.95, using 7-fold cross validation). A total of 31 marker metabolites were identified using the two analytical platforms. The majority of these metabolites were associated with expected metabolic perturbations in CRC including elevated tissue hypoxia, glycolysis, nucleotide biosynthesis, lipid metabolism, inflammation and steroid metabolism. OPLS-DA models showed that the metabolite profiles obtained via HR-MAS NMR could further differentiate colon from rectal cancers (Q2> 0.60, ROC AUC ) 1.00, using 7-fold cross validation). These data suggest that metabolic profiling of CRC mucosae could provide new phenotypic biomarkers for CRC management. Keywords: colon • rectum • colorectal cancer • HR-MAS NMR spectroscopy • GC/MS • metabolic profiling • OPLS-DA • chemometric
Introduction Worldwide, colorectal cancer (CRC) is the second most common malignancy and is a leading cause of cancer-associated death in many developed countries. In Singapore, CRC is the leading cancer in males and the second most common cancer in females, accounting for 18.0% of all cancers in males and 14.4% in females. When both genders are combined, CRC is the most common cancer in Singapore. The age-standardized rate (ASR) for mortality from CRC in Singapore over the period * To whom correspondence should be addressed. Asst. Prof. Eric C. Y. Chan: e-mail,
[email protected]; tel, +65 65166137; fax, +65 67791554. Dr. Hector C. Keun: e-mail,
[email protected]. † National University of Singapore. § Imperial College London. ‡ Singapore General Hospital.
352 Journal of Proteome Research 2009, 8, 352–361 Published on Web 12/08/2008
from 2002 to 2006 was 18.1 per 100 000 per year in males and 12.5 in females. The ASR for CRC mortality in the United States and United Kingdom in 2002 according to the World Health Organization was 14.8 and 17.3 per 100 000 per year, respectively (http://www.who.int/healthinfo). Strategies to improve CRC-related mortality and morbidity include presymptomatic screening of at-risk population and risk stratification of affected individuals for further adjuvant treatment with chemo- or radiotherapy. Presymptomatic screening aims to detect early stage CRC or its precursor lesions for improved cancer-specific survival and reduced treatment-related morbidity. Unfortunately, only about 37% of CRC remain localized at the time of diagnosis.1 Present diagnostic and screening methods for CRC, such as colonoscopy, flexible sigmoidscopy, double contrast barium 10.1021/pr8006232 CCC: $40.75
2009 American Chemical Society
research articles
Metabolic Profiling of Human CRC Using HR-MAS NMR and GC/MS enema, virtual colonoscopy, fecal occult blood test (FOBT), fecal based DNA test, and serum based protein markers such as carcinoembryonic antigen (CEA), have certain limitations and shortfalls in their own ways. While genomic and proteomic techniques have led to the identification of many serum, tissue and fecal-based biomarkers of CRC, only serum CEA and fecalbased genetic markers are useful clinically.2-4 Although CEA is easy to detect using a sample of blood, it suffers from a low specificity and is therefore not useful as a screening or prognostic test. Fecal DNA testing while more specific, is cumbersome to administer leading to poor patient compliance. Patients at high risk for cancer recurrence following surgery may be given adjuvant treatment to improve their 5-year survival rate (YSR). The 5-YSR for patients with stage II CRC is 75%, the majority of patients being cured by surgery alone. However, 25% of these patients will develop tumor recurrence within their lifetime. A recent study by Ribic et al. showed that fluorouracil-based adjuvant therapy benefited patients with stage II or stage III CRC with stable or low microsatellite (MSI) instability but not those whose tumors exhibited high MSI instability.5 Therefore, some patients may not derive significant clinical benefits when given adjuvant treatment and may in fact suffer from the side effects of treatment. As tumor typing using genetic markers for MSI is labor intensive and costly, metabolite profiling of tumors may provide a rapid and costeffective alternative means of identifying patients who are more likely to benefit from adjuvant therapy and select out those who are likely to be adversely affected by the treatment. CRCrelated metabolic biomarkers of good prognostic value would potentially enable oncologists to tailor individualized treatment regimens according to tumor metabolic profiles and optimize the current clinical strategies for CRC management.6-8 The current strategy for staging and prognostication of CRC relies mainly upon the TNM or Duke system. This clinicopathological staging is derived mainly from the histological assessment of depth of tumor invasion through the bowel wall and the extent of lymph nodal spread. However, histological staging is in reality a reflection of the time point in which surgical intervention took place and represents a “snap shot” of the tumor morphology at the time of diagnosis and surgical intervention. It is a crude guide to prognostication because it reflects the delay in diagnosis in the case of an advanced cancer but gives no insight into the molecular characteristics of the tumor. Surrogate markers such as angiolymphatic and perineural invasion provide some indication of the ability of the tumor to metastasize. However, histological staging cannot explain why certain small tumors metastasize while other tumors grow to a large size with invasion of surrounding structures and yet remain localized. In an effort to fill this knowledge gap, genomics and proteomics have emerged as possible tools to provide further insight into the true biological potential of the tumor. While genomics and proteomics target profiling at the genetic and protein levels, metabonomics encompasses the comprehensive and simultaneous systematic profiling of multiparametric metabolic changes that occur in living systems in response to pathological, environmental or lifestyle factors.9 Metabonomics complements and enhances the information provided by genomics and proteomics10 and has already shown promise in identifying metabolite-based biomarkers in ovarian, brain and liver cancers.11-13 An early 1H NMR spectroscopy study demonstrated significant increases in taurine, choline-containing compounds and lipid resonances in malignant colon
14
mucosa. More recent proteomic work has indicated that alterations of metabolic pathways such as purine and pyrimidine metabolism, glycolysis, gluconeogenesis, glucoronic acid synthesis and tricarboxylic acid cycle also occur in CRC.15 Such alterations in different metabolic pathways are expected to result in associated changes in metabolic profiles that, if identified with the aid of metabonomics, may be explored as potential biomarkers for CRC detection, staging, prognostication, and treatment stratification.16 In this study, we hypothesized that the global analysis of metabolites in colon mucosae would define metabolic signatures that discriminate tumor from normal mucosae and correlate to the anatomical and clinicopathological characteristics of CRC. To test the hypothesis, we used high-resolution magic angle spinning nuclear magnetic resonance (HR-MAS NMR) and gas chromatography mass spectrometry (GC/MS) metabonomics to profile biopsied tumor specimens and their matched normal mucosae. This report compares and contrasts the ability of the two analytical platforms to generate diagnostic profiles and describes biological significance of the marker metabolites that most strongly influence the classification of tumor and normal mucosae. Our results illustrate the potential for metabolic profiling to provide biomarkers for staging, prognosis and therapy selection in CRC.
Materials and Methods Clinical Population. Clinical data such as age, gender, ethnicity, location of primary tumor, histological staging and grade were obtained from a prospectively maintained computerized database from the Singapore Polyposis Registry & the Colorectal Cancer Research Laboratory, Department of Colorectal Surgery, Singapore General Hospital. The clinicopathological characteristics of the CRC patients are summarized in Table 1. The study population comprised 31 patients with a mean age of 67 ( 13 years at the time of cancer diagnosis. Three patients were younger than 50 years old (T11, T13 and T21). There were 18 males (58%) and 13 females (42%). The majority (87%) of the patients were Chinese (n ) 27), while the remaining comprised 2 Indians (T5 and T27), one Malay (T7) and one of other ethnicity (T16, T23; n ) 1). The CRC anatomical site, tissue histology, tumor grade, TNM and Duke stages are presented in Table 1. There were 22 left-sided tumors (defined as those arising distal to the splenic flexure) of which 14 were in the rectosigmoid or rectum. The tumors were predominantly (81%) moderately differentiated (n ) 25), and the remaining 6 comprise 2 each of mucinous, well and poorly differentiated tumors. Although not presented in Table 1, tumor invasion of neighboring organs, lesion nature and dimension, and presence of angiolymphatic or perineural invasion were also noted. Tissue Collection. This study was approved by the institutional review board at the Singapore General Hospital (IRB reference number 260/2007). Matched CRC and normal mucosae (n ) 63) were obtained from the 31 CRC patients during surgery. Among these subjects, one patient provided two matched pairs of tissues (M23, T23, M16 and T16), while another only provided the normal mucosa (M19). None of the patients received neoadjuvant chemotherapy or radiotherapy prior to surgical excision. Fresh tumor tissue and matched normal mucosa were snap-frozen immediately following excision of the specimen at surgery, then stored at -80 °C until processing. Tumor specimens were carefully microdissected to ensure that at least 90% of the analyzed tissue contained cancer Journal of Proteome Research • Vol. 8, No. 1, 2009 353
research articles
Chan et al.
Table 1. Summary of Anatomical and Clinicopathological Characteristics of CRC Patients normala d
M23 M16 M7 M4d M2 M1d M8 M30d M31d M19 M6d M33d M24d M28d M29d M32d M27d M17 M3d M15d M11d M25d M10d M5 M12d M26d M22d M9d M13d M20d M21d M18d
CRCa b,d
T23 T16b T7d T4d T2 T1 T8 T30d T31d T6 T33d T24d T28d T29d T32d T27d T17d T3d T15 T11 T25d T10d T5d T12 T26d T22d T9d T13 T20d T21d T18
CRC anatomical site
histology
Cecum Cecum Cecum Ascending colon Hepatic flexure Transverse colon Transverse colon Transverse colon Descending colon Descending colon Sigmoid colon Sigmoid colon Sigmoid colon Sigmoid colon Sigmoid colon Sigmoid colon Sigmoid colon Sigmoid colon Rectosigmoid colon Rectosigmoid colon Rectosigmoid colon Rectum Rectum Rectum Rectum Rectum Rectum Rectum Rectum Rectum Rectum Rectum
Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma High-grade dysplasia Adenocarcinoma Mucinous Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Adenocarcinoma Mucinous Adenocarcinoma Adenocarcinoma
gradec
MD MD MD MD MD MD MD MD MD PD PD MD MD MD MD WD M MD MD MD WD MD MD MD MD MD MD MD MD M MD MD
stage
Duke
T3N0M0 T3N0M0 T3N2M0 T3N1M0 T2N0M1 T3N2M0 T3N2M0 T3N0M0 T3N0M0 T4N1M0 T2N1M0 T3N1M0 T3N1M0 T3N0M0 T2N0M0 T4N0M0 T3N1M0 T3N1M0 T2N0M1 T1N0M0 T3N1M0 T3N2M0 T3N1M1 T2N0M0 T2N0M0 T3N0M0 T3N0M0 T2N0M0 T4N0M1 T3N1M0 T3N0M0 T3N0M1
B B C C D C C B B C C C C B A B C C D A C C D B A B A B D C B D
a For each T (CRC) sample, matched M (normal) mucosa was provided, with the exception of M19 (i.e., no matched T19). b M16, T16, M23 and T23 were obtained from one patient. c MD, WD, M and NA are moderately differentiated, well-differentiated, mucinous and not applicable, respectively. d These samples were analyzed by HR-MAS NMR in addition to GC/MS.
cells. Matched normal mucosae were taken at least 5-10 cm away from the edges of the tumor. All CRC tissues and matched normal mucosae were cut and weighed accurately where approximately 10 and 20 mg of each tissue were reserved for HR-MAS NMR spectroscopy and GC/MS, respectively. The samples were kept at -80 °C until analysis. Because of the limited size of each tissue block, all the tissues (n ) 63) were analyzed using GC/MS, while 47 tissues were analyzed using HR-MAS NMR spectroscopy, of which 18 pairs were matched samples. The remaining specimens were preserved in formalin and submitted for routine histological examination with hematoxylin and eosin staining by a gastrointestinal pathologist to determine the tumor stage and differentiation. HR-MAS NMR Spectroscopy. Each accurately weighed intact tissue (10 mg) was bathed in D2O solution for 15 s. The tissue was inserted into a zirconium oxide 4 mm outer diameter rotor with an additional drop of D2O to provide a field-frequency lock for the NMR spectrometer. An insert was placed into the rotor to make a spherical sample volume of 25 µL. A cap was finally added as a closure of the rotor and the assembled device was used immediately for NMR analysis. All samples were randomized during analysis to reduce any potential systematic errors. All 1H NMR spectra were recorded on a Bruker AV-600 NMR spectrometer (Rheinstetten, Germany) operating at 600.11 MHz for 1H, equipped with a high-resolution magic angle spinning 354
Journal of Proteome Research • Vol. 8, No. 1, 2009
probe at a spin rate of 5000 Hz. Sample temperature was regulated using cooled N2 gas at 283 K during the acquisition of spectra to minimize spectral degradation. Two different types of 1H NMR spectra were collected for each sample, a onedimensional (1D) NOESY spectrum with water suppression and a 1D Carr-Purcell-Meiboom-Gill (CPMG) spin-echo spectrum. Because the CPMG spin-echo experiment gave the clearest signature of metabolic changes between the cancer and normal mucosae, with little extra information contained in the higher molecular weight components, only results obtained from the CPMG experiments were used for further data analysis. CPMG spin-echo spectra were obtained using the pulse sequence [RD-90°-(τ-180°-τ)n - acquire FID], with a spin-spin relaxation delay, 2nτ, of 240 ms. The recycle delay (RD) was 2 s. The 90° pulse length was 6.9-9.0 µs. A total of 256 transients were collected into 32 K data points with a spectral width of 20 ppm. 1H MAS NMR spectra of tissues were manually phased and baseline corrected using XwinNMR 3.5 (Bruker Analytik, Rheinstetten, Germany). The 1H NMR spectra were referenced to the methyl resonance of alanine at δ 1.47. The total analysis time (including sample preparation, optimization of NMR parameters and data acquisition) of HR-MAS NMR spectroscopy for each sample was approximately 40 min. Although a greater mass of the tissue (>20 mg) is recommended to fill the rotor space for MAS analysis, the ∼10 mg tissue used in our experiments generated good signal sensitivity and resolution. For assignment purposes, 2D correlation spectros-
Metabolic Profiling of Human CRC Using HR-MAS NMR and GC/MS copy (COSY) and J-resolved (JRES) NMR spectra were acquired on selected samples. HR-MAS NMR Data Analysis. The spectra over the range of δ -1.0 to 10.0 were imported into Matlab using in-house software developed by Dr. Rachel Cavill, Dr. Tim Ebbels and Dr. Hector Keun (Version 7, The Mathworks, Inc., Natwick, MA). All spectra were “binned” into 0.01 ppm regions. Probabilistic quotient normalization of the spectra using the median spectrum to estimate of the most probable quotient was carried out before chemometric and statistical analyses.17 The resulting binned data comprising 48 samples names (observations) and 1101 variables were analyzed using orthogonal partial leastsquares discriminant analysis (OPLS-DA) (SIMCA-P software, Umetrics, Umeå, Sweden).18 All processed data were mean centered and scaled (as indicated in the relevant figure legend) during chemometric data analysis. The residual water resonance signal (δ 4.50-5.19) and the spectral region (δ -1.0 to 0.5) were removed prior to analysis. The peak intensities of the spectra region related to the marker metabolites were integrated in Matlab using the full-resolution data and a local linear baseline correction. An independent t test with Welch’s correction was used for the comparison of the marker metabolite levels to determine their significant differences between the CRC and normal mucosa groups (p < 0.05 was considered to be statistically significant). Receiver operating characteristic (ROC) analysis was performed to validate the robustness of some of the OPLS-DA models using the cross-validated (7-fold) predicted Y-values. Cross-validation is a procedure where multiple models are generated each excluding a different portion of the data (in our case every seventh sample), such that all samples are excluded once and once only. Predicted classifications (Y-values) were then generated for each set of excluded samples using the appropriate model, in order to estimate the predictive performance of the classification algorithm. The area under the ROC curve denoted as AUC was calculated using the trapezoidal rule. GC/MS Analysis. Twenty milligrams of each tissue was transferred to a 15 mL glass centrifuge tube. One milliliter of a monophasic mixture of chloroform-methanol-water (2:5:2, v/v/v) was added to each sample and the mixture was ultrasonicated at ambient temperature (24-28 °C) for 100 min and then vortex-mixed for another 2 min. The samples were subsequently centrifuged at 18 000g for 3 min and 0.8 mL of the supernatant was collected separately from each sample into a 15 mL glass tube. The collected supernatant was concentrated to complete dryness in a Turbovap nitrogen evaporator using a temperature of 50 °C for 30 min. A total of 100 µL of toluene, kept anhydrous with sodium sulfate, was added to each of the dried tissue extracts, vortex-mixed for 1 min and again evaporated to complete dryness in a Turbovap nitrogen evaporator in order to eliminate any trace of water which may interfere with GC/MS analysis. The dried samples were then derivatized to increase the volatility of polar metabolites by adding 100 µL of N-methyl-N-trifluoroacetamide (MSTFA) with 1% trimethylchlorosilane (TMCS) to each sample. The samples were then vortex-mixed for 1 min and incubated at 70 °C for 30 min. After incubation, samples were again vortex-mixed for 1 min and then transferred to vials for GC/MS analysis.19-22 Analysis was performed on a Shimadzu QP2010 GC/MS system (Shimadzu, Kyoto, Japan). A HP-5MS 30m × 250 µm (i.d) fused silica capillary column (Agilent J&W Scientific, Folsom, CA), chemically bonded with a 5% diphenyl, 95% dimethylpolysiloxane cross-linked stationary phase (0.25 µm
research articles
film thickness), was used with open split interface. Helium was used as the carrier gas at 1.2 mL/min and the injector split ratio was set to 1:5. An injection volume of 1 µL was used and the solvent cutoff time was 5 min. The injector and source temperatures were kept at 250 and 200 °C, respectively. Oven temperature was kept at 60 °C for 3 min, increased at 7 °C/ min to 140 °C where it was held for 4 min and further increased at 5 °C/min to 310 °C where it remained for 6 min. The MS was operated in electron impact (EI) ionization mode at 70 eV. Data acquisition was performed in the full scan mode from m/z 50-650 with a scan time of 0.5 s. To detect and eliminate retention time shifts, standard alkane series mixture (C-10 to C-40) was injected periodically into the GC-MS system during analysis of each batch of samples. For every sample, a fresh liner and vial was taken to avoid sample carryover and crosscontamination. Chromatogram acquisition and preliminary compound identification by the National Institute of Standards and Technology (NIST) and Wiley EI mass spectral library search were performed using the Shimadzu GCMSsolution (Version 2.5) software. Retention time (RT) correction of peaks based on retention time of standard alkane series mixture (C10 to C-40) was performed using the AART (Automatic Adjustment of Retention Time) function of the Shimadzu GCMSsolution software. Peaks with similarity index more than 70% were assigned compound names, while those having less than 70% similarity were listed as unknown metabolites. Identities of selected metabolites preliminarily identified by NIST mass spectral library were further confirmed by comparison of their mass spectra and retention times with those obtained using commercially available reference standards (L-alanine, L-valine, glycine, L-threonine, L-proline, L-phenylalanine, L-tyrosine, myo-inositol, uridine, uracil, D-glucose, D-mannose, D-galactose, L-(+)-lactic acid, fumaric acid, D-(+)-malic acid, arachidonic acid, phosphoric acid, boric acid, formaldehyde and cholesterol). The chromatograms were subjected to noise reduction and baseline correction using metAlign software (www.metalign.wur.nl) prior to peak area integration. Integrated peak areas of multiple derivative peaks belonging to the same compound were summed and considered as single compound. All known artifact peaks, such as peaks due to column bleed and MSTFA artifact peaks, were not considered in the final data analyses. Normalization to a constant sum of the chromatographic peak area was carried out before chemometric and statistical analyses. The total analysis time (including sample preparation and data acquisition) for each sample was approximately 200 min. Although overnight methoximation of samples using methoxyamine hydrochloride in pyridine prior to derivatization with MSTFA is considered a conventional step in sample preparation protocol for metabonomic studies, it was found to offer no clear advantage in our case and so was eliminated from our sample preparation protocol. As a result, multiple peaks were generated for the monosaccharides and their summed peak areas were used in the data analysis. During method development, it was determined that at least 20 mg of the tissue sample was needed to generate reasonably sensitive profiling of the endogenous metabolites. GC/MS Data Analysis. The resulting data comprising of peak number (RT-mass pair), sample name, and ion intensity were analyzed similarly to the HR-MAS NMR data using OPLS-DA. All processed data were mean centered and pareto scaled during chemometric data analysis. Journal of Proteome Research • Vol. 8, No. 1, 2009 355
research articles
Chan et al.
Figure 1. (A) HR-MAS NMR spectra and histological sections (inset) of representative CRC and normal mucosae (from left to right). The marker metabolites that were compared and found to be elevated in CRC and normal mucosae are labeled in the respective spectra. (B) OPLS-DA scores plot discriminating CRC from normal mucosae based on selected marker metabolites detected by HR-MAS NMR (Pareto scaled data was used). (C) ROC curve determined using the cross-validated predicted Y-values of the HR-MAS NMR OPLS-DA model.
Results HR-MAS NMR Analysis. In total, 22 tumor samples and 25 normal mucosae from 29 CRC patients were analyzed by MAS NMR (see Table 1 for patient characteristics). The representative HR-MAS NMR spectra of CRC and normal mucosae are shown in Figure 1A. The resulting spectra were processed and converted into 1101 integral regions of 0.01 ppm width (“binned”) as described in Materials and Methods. The binned data were initially analyzed by OPLS-DA, generating a model classifying normal from tumor (3 LV, R2Y and Q2(cum) were 0.843 and 0.653, respectively) and the correlated loadings were used to help identify the spectral regions discriminating between these groups. LV, R 2Y and Q2(cum) are the latent variables, fraction of the sum of squares of all Y-values explained by the current latent variable and cumulative Q2 for the extracted latent variable, respectively. Q2 is given by the expression Q2 ) 1 -∑(Ypredicted - Ytrue)2/∑Y2true. The overall explained variance, R 2Y, is the same expression as for Q2 but calculated for a model generated with all the training data.23 These differential spectra regions were subsequently confirmed via visual inspection of the NMR spectra. The marker metabolites were subsequently assigned (Figure 1A) and the peak intensities of the spectra region related to the marker metabolites were integrated from the full-resolution data. An independent t test with Welch’s correction was used for the comparison of the marker metabolite levels to determine the statistical significance of differences between the tumor and normal mucosa groups. 356
Journal of Proteome Research • Vol. 8, No. 1, 2009
Table 2 shows the marker metabolites that were responsible for the separation of tumor specimens from their matched normal mucosae in the OPLS-DA model. Lipids, polyethylene glycol (PEG) and glucose were present at higher levels in normal mucosae compared to CRC tissues, while choline-containing compounds (ChoCC), taurine, scyllo-inositol, glycine, phosphoethanolamine (PE), lactate and phosphocholine (PC) were present at higher levels in the CRC specimens. The percentage changes of the metabolites in cancer compared to the normal mucosae were broadly consistent for the different resonances of the same metabolites (e.g., lipids, taurine and glucose) and further validated our assignment of the biomarkers. All marker metabolites were found to be statistically different between the test groups (p < 0.05) except for glycine and PE. However, these two metabolites were included in further analysis as their levels were clearly different between tumor and normal mucosae in the loadings of the OPLS-DA model. A secondary OPLS-DA model was created using the marker metabolite intensities as variables except for PEG which was an exogenous compound. Receiver operating characteristic (ROC) analysis using the cross-validated predicted Y (predicted class) values was performed to validate the robustness of the OPLS-DA model in discriminating the CRC specimens from the normal muscosae. Figure 1B shows the OPLS-DA model scores generated using the selected marker metabolites. (3 LV, R2(Y) and Q2(cum) were 0.622 and 0.518, respectively). The sensitivity and specificity trade-offs were summarized for each variable using the area under the ROC curve denoted as AUC and
Metabolic Profiling of Human CRC Using HR-MAS NMR and GC/MS Table 2.
research articles
1
H Chemical Shifts of Marker Metabolites Found in HR-MAS NMR Spectra of Human CRC and Normal Colon Mucosae
metabolite
Lipids
ChoCCd Taurine Scyllo-inositol Glycine PEGd PEd Lactate PCd Glucose
δ 1H (ppm)
group
multiplicitya
0.90 2.00 5.28-5.44 3.21 3.25 3.42 3.34 3.55 3.70 3.99 4.11 4.19 4.64 5.23
CH3 CH2-CdC –CHdCH– N(CH3)3 NCH2 SCH2 Half δ-CH2 CH2 CH2 OCH2 R-CH OCH2 1-CH 1-CH
m m m s (multiple) t t s s s m q t d d
observed
1D
1D, COSY 1D, JRES, COSY 1D, 1D, 1D, 1D, 1D, 1D, 1D, 1D,
JRES JRES JRES JRES, JRES, JRES, JRES, JRES,
COSY COSY COSY COSY COSY
% change of cancer from normalb
pc
-83.3 -48.0 -84.5 82.7 115.8 152.3 39.1 24.4 -58.6 46.0 65.0 75.6 -45.8 -63.3
