Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Research Article Cancer Epidemiology, Biomarkers Urinary Metabolomics to Identify a Unique & Prevention Biomarker Panel for Detecting Colorectal Cancer: A Multicenter Study Lu Deng1, Kathleen Ismond1,2, Zhengjun Liu3, Jeremy Constable4, Haili Wang2, Olusegun I. Alatise5, Martin R. Weiser4, T.P. Kingham4, and David Chang1,2
Abstract
Background: Population-based screening programs are paring the metabolomic profiles from colorectal cancer versus credited with earlier colorectal cancer diagnoses and treat- controls. Multiple models were constructed leading to a good ment initiation, which reduce mortality rates and improve separation of colorectal cancer from controls. patient health outcomes. However, recommended screen- Results: A panel of 17 metabolites was identified as possible ing methods are unsatisfactory as they are invasive, are biomarkers for colorectal cancer. Using only two of the select- resource intensive, suffer from low uptake, or have poor ed metabolites, namely diacetylspermine and kynurenine, a diagnostic performance. Our goal was to identify a urine predictor for detecting colorectal cancer was developed with an metabolomic-based biomarker panel for the detection of AUC of 0.864, a specificity of 80.0%, and a sensitivity of colorectal cancer that has the potential for global popula- 80.0%. tion-based screening. Conclusions: We present a potentially "universal" metabo- Methods: Prospective urine samples were collected from lomic biomarker panel for colorectal cancer independent of study participants. Based upon colonoscopy and histopathol- cohort clinical features based on a North American popula- ogy results, 342 participants (colorectal cancer, 171; healthy tion. Further research is needed to confirm the utility of the controls, 171) from two study sites (Canada, United States) profile in a prospective, population-based colorectal cancer were included in the analyses. Targeted liquid chromatogra- screening trial. phy-mass spectrometry (LC-MS) was performed to quantify Impact: A urinary metabolomic biomarker panel was iden- 140 highly valuable metabolites in each urine sample. Poten- tified for colorectal cancer with the potential of clinical tial biomarkers for colorectal cancer were identified by com- application.
Introduction programmatic rather than opportunistic to ensure a high rate of compliance (2). Such programs have been instituted nationally or Colorectal cancer is the third most commonly diagnosed regionally within many countries in Europe (e.g., UK, Ireland, malignancy and the fourth leading cause of cancer-related deaths Germany, France), United States, Japan, and Australia as reviewed in the world. On the basis of 2018 estimates, the 2040 incidence by Navarro and colleagues (3). rates for colorectal cancer are projected to increase by 72% to 3.1 The most commonly used population-based screening modal- million new cases, while mortality rates will increase by 82% to ities are the fecal immunochemical test (FIT) and colonoscopy (4). 1.5 million deaths (https://gco.iarc.fr/tomorrow). Mortalities due FIT detects hidden blood in stool that occurs mostly in the later to colorectal cancer are largely preventable through regular screen- stages of cancer and has low sensitivity for detecting the precursors ing and early detection using fecal-based tests and colonosco- to colorectal cancer, adenomatous polyps (9). A new fecal DNA py (1). To be effective, population-based screening must be test detects DNA mutations in addition to hidden blood in stool with improved sensitivity (5), but it is costly and only available in a few countries. To date, fecal-based tests are limited to colorectal 1Metabolomic Technologies Inc., Edmonton, Alberta, Canada. 2Department of cancer detection not prevention, and have low adherence rates 3 Medicine, University of Alberta, Edmonton, Alberta, Canada. Department of due to the need for stool collection and manipulation (6–10). Mathematical and Statistical Sciences, University of Alberta, Edmonton, Alberta, Colonoscopy has a superior sensitivity and specificity to nonin- Canada. 4Department of Surgery, Memorial Sloan Kettering Cancer Center, New York, New York. 5Department of Surgery, Obafemi Awolowo University and vasive screening tests, but is costly in terms of direct and indirect Obafemi Awolowo University Teaching Hospitals Complex, Ile-Ife, Nigeria. health care dollars, has a higher risk of procedural-related com- plications, and, like fecal-based tests, has low rates of screening Note: Supplementary data for this article are available at Cancer Epidemiology, Biomarkers & Prevention Online (http://cebp.aacrjournals.org/). compliance (11). To increase screening compliance rates, programs have largely Corresponding Author: Lu Deng, Metabolomic Technologies Inc., 132, 9650 focused on colorectal cancer education and sending reminders to 20th Avenue, Edmonton, Alberta T6N 1G1, Canada. Phone: 587-772-1684; Fax: 780-492-8121; E-mail: [email protected] eligible participants (12, 13). An alternative approach for improv- ing colorectal cancer screening rates is to use a biosample other Cancer Epidemiol Biomarkers Prev 2019;28:1283–91 than stool (14). A blood-based screening test has been shown to doi: 10.1158/1055-9965.EPI-18-1291 have higher patient uptake than FIT (15), but its cost-effectiveness 2019 American Association for Cancer Research. is debatable for population-based screening (16). Urine is
www.aacrjournals.org 1283
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Deng et al.
commonly used for many clinical tests, can be readily collected, Ethics approval was obtained from the Health Research and is more acceptable to patients (17, 18). Recently, putative Ethics Boards at the University of Alberta (Pro0000514 and biomarkers of colorectal cancer were identified in urine in the Pro00074045) and MSKCC (IRB catalog nos. 06-107 and 15-209). forms of volatile organic compounds (19), modified cytosine nucleosides (20), and polyamines (21, 22). As well, we have Metabolite analysis reported a urine-based screening test specific for colorectal ade- Targeted liquid chromatography-mass spectrometry (LC-MS) nomatous polyps (23, 24) developed in a Canadian population was performed to quantify urinary metabolites in each sample and its subsequent validation in a homogenous Asian cohort to using the LC-MS kit TMIC00UJ designed and prepared by The demonstrate its clinical relevance transcending both diet and Metabolomics Innovation Centre (TMIC) at the University of ethnicity (25). Alberta in Edmonton, Alberta. Calibration solutions (Cal 1–Cal In the current multicenter study, the potential utility of urine- 7), isotopically labeled standard mix, quality control solutions based metabolomics for detecting colorectal cancer was investi- (QC 1–QC 3), LC-MS methods, and standard operating proce- gated. This was done by analyzing metabolites in urine samples dures were provided by TMIC. The TMIC00UJ kit was a combi- from colonoscopy- and histopathology-confirmed cases of colo- nation of three assays to identify 140 unique urinary metabolites rectal cancer and healthy controls (e.g., polyp- and colorectal (see Supplementary Table S1) indexed by the Human Metabo- cancer-free). Our findings highlight the predictive potential of lome Database (www.hmdb.ca). The phenyl isothiocyanate urinary metabolites for colorectal cancer and we discuss the (PITC) assay quantified 47 biologic amines in the LC mode while clinical relevance of a proposed screening test. 75 lipids were semiquantified in the flow injection analysis (FIA) mode. The organic acid assay quantified 17 compounds while ascorbic acid was quantified independently. The TMIC00UJ kit components were run on an API4000 Qtrap Materials and Methods tandem mass spectrometry instrument (AB Sciex) coupled with a Study participants and sample collection Waters UPLC system (Waters Limited). Urine samples were Adult patients with newly diagnosed colorectal cancer (based thawed on ice, vortexed, then centrifuged at 13,000 g. Each on preoperative imaging, colonoscopies, and pathology reports plate contained 82 unique urine samples as well as 1 solvent blank of biopsies) were eligible for study inclusion provided they had solution, 3 matrix solutions, 7 calibration solutions (Cal 1–Cal 7), not received colorectal cancer–related treatment. Canadian and 3 quality control (QC) samples. PBS (1 , pH 7.4) was used as recruitment (October 2008–2010) was conducted at four tertiary the matrix solution. Metabolite quantification was achieved using hospitals in the Edmonton region (Grey Nuns Hospital, Miser- the AB Sciex Analyst software, version 1.6.2. During quantifica- icordia Hospital, University of Alberta Hospital, and the Royal tion, each metabolite was identified using the internal standard Alexandra Hospital) and included patients from across the and compared against the established calibration curve. The lower prairie provinces (i.e., CRC-CADcohort).Americanpatients limits of detection (LLOD) were calculated as three times the value were recruited (February–July 2018) from the Memorial Sloan of the matrix solutions. The upper limit of detection was not Kettering Cancer Center (MSKCC) in New York City, New York reached for any metabolite. (i.e., CRC-MSKCC cohort). Patients diagnosed with colorectal cancer provided a urine Statistical analysis sample prior to any operation, chemotherapy, radiation, or other Data preprocessing was performed using code written in R, cancer-related treatment. Clinical features, such as age, gender, version 3.4.3. Metabolites that were lower than the LLOD or not and smoking status, were also collected at this time. Each urine detected in more than half of the urine samples were removed sample was transferred to labeled 1 mL tubes (5 ) and frozen at from the initial list of 140 metabolites. For the remaining meta- 80 C within 1 hour of collection. Frozen urine was shipped on bolites on the list, if a sample had a metabolite concentration that dry ice in a standard insulated Styrofoam shipper and immedi- was less than the LLOD, it was replaced with half the value of the ately transferred to a 80 C freezer upon arrival at the University LLOD. Statistical analyses were conducted with MetaboAnalyst, of Alberta in Edmonton, Alberta. Pathology reports were reviewed version 4.0 for the web (29). Metabolite concentration was to abstract cancer stage. normalized against creatinine, log-transformed, and auto-scaled. The healthy controls were selected from a previous population- Potential biomarkers for colorectal cancer were identified (30) by based study (n ¼ 1,000) called Stop COlorectal cancer through comparing the metabolomic profiles of the colorectal cancer and Prevention and Education (SCOPE; refs. 23, 26–28) The SCOPE control groups for both fold-change analyses and Student t tests. program, regional colon cancer screening program (Edmonton, One-way ANOVA was performed on the independent sample Alberta, Canada) where over 1,000 urine samples were collected groups (e.g., CRC-CAD, CRC-MSKCC, and control) to identify from April 2008 to October 2009. Study participants (40–74 years statistically significant metabolite differences (31–33). The meta- of age) of average or increased risk for colorectal cancer were bolites with concentration changes in the same direction for both recruited. On day of entry, participants provided informed written the CRC-CAD and CRC-MSKCC groups were considered consis- consent, a midstream urine sample, and completed a demograph- tent colorectal cancer markers. Furthermore, multivariate models, ic survey. Urine was aliquoted and frozen at 80 C within 1 hour using principal component analysis (PCA), partial least squares of collection. Colonoscopy was performed 2–6 weeks after the discriminant analysis (PLS-DA), and sparse PLS-DA (sPLS-DA; urine collection confirmed that the individuals were classified as ref. 34) were constructed. Finally, predictors were built using the normal based upon endoscopy findings and pathology reports. logistic regression with selected biomarkers. Leave-out approach Urine samples from the healthy controls were matched 1:1 to the was used to evaluate the built models. A total 171 controls were colorectal cancer cases based on gender. A study design chart was randomized to form 121 controls for training and 50 controls for shown in Supplementary Fig. S1 (Supporting Information). testing with balanced age, gender, and smoking status. A total of
1284 Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 Cancer Epidemiology, Biomarkers & Prevention
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Urinary Biomarker Panel for Detecting Colorectal Cancer
121 CRC-CAD and 121 controls were used as training set to build Table 1. Patient characteristics a model, and 50 CRC-MSKCC and 50 controls were used as testing CRC Cases set to validate the model. Controls CRC-All CRC-CAD CRC-MSKCC All code for statistical analyses was also written in R, version Mean age, years 58.9 (5.6) 66.4 (11.5) 67.4 (10.9) 63.8 (12.5) (SD) 3.4.3 (https://www.R-project.org). The glmnet package was used Gender, n (%) for logistic regression (35). ROC (12) curves were generated and Male 100 (58.5%) 89 (52.0%) 68 (53.7%) 24 (48.0%) reported using the ROCR package. Female 71 (41.5%) 82 (48.0%) 59 (46.3%) 26 (52.0%) Smoking, n (%) Current 12 (7.0%) 29 (17.0%) 24 (19.8%) 5 (10.0%) Results Prior 66 (38.6%) 56 (32.7%) 38 (31.4%) 18 (36.0%) Never 87 (50.9%) 86 (50.3%) 59 (48.8%) 27 (54.0%) Patient characteristics By Stage, n (%) In Canada, a total of 161 participants were enrolled of which 40 0 – 3 (1.8%) 3 (2.5%) 0 (0.0%) were excluded due to missing clinical information. A total of 50 I – 30 (17.5%) 16(13.2%) 14 (28.0%) samples were collected from patients at MSKCC and used for this II – 50(29.2%) 30 (24.8%) 20 (40.0%) study. The 171 colorectal cancer samples were matched with 171 III – 57 (33.3%) 51(42.1%) 6 (12.0%) – urine samples from colonoscopy-confirmed healthy controls. IV 31 (18.1%) 21 (17.4%) 10 (20.0%) Total, n 171 171 121 50 See Table 1 for a summary of clinical characteristics for the participants. Statistical analysis was performed on control group versus colorectal cancer group. The P value for gender was 0.63 consistent markers for colorectal cancer. For each of the 17 indicating there was no significant difference in gender between metabolites, the concentration change in either colorectal cancer colorectal cancer and controls. The P value for smoking was 0.02 group (e.g., CRC-CAD or CRC-MSKCC) compared with the control with more current smokers in the colorectal cancer group. The P group were analyzed. Diacetylspermine (Fig. 1A), proline, kynur- value for age was 2.83 10 13 indicating there was a significant enine, and glucose were upregulated in both colorectal cancer difference in age between colorectal cancer and controls where the groups compared with controls and classified as consistent bio- mean age in colorectal cancer group was approximately 7 years markers. Although they were identified as potential markers accord- older than the control group. ing to the volcano plot for colorectal cancer cases versus controls, the concentrations of 3-(3-Hydroxyphenyl)-3-hydroxypropanoic Metabolite analysis acid (HPHPA, Fig. 1B), beta-hydroxybutyric acid, 3,4-dihydroxyl A total of 140 metabolites were quantified in each urine sample phenylalanine (DOPA), 4-hydroxyproline, aminoadipic acid, by three LC-MS assays. In the PITC assay, a total of 47 biologic putrescine, indole acetic acid, hippuric acid, citric acid, and sarco- amines were quantified in LC mode and a total of 75 lipids were sine did not significantly change when CRC-CAD were compared semiquantified in the FIA mode. In the organic acid assay, a total with controls. Similarly, there were no significant changes in the of 17 valuable organic acids were quantified. Ascorbic acid was concentrations of Tetradecenoyl carnitine (C14:1), and aspartic quantified using a specific assay. For each assay, a total of 382 acid (Fig. 1C), and sarcosine when CRC-MSKCC was compared samples including both the colorectal cancer and control samples with controls. When compared against the control group, the were randomized and analyzed using 5 plates in 96-well plate concentration of butyric acid (Fig. 1D) increased in CRC-CAD and format. For each plate, a set of calibration curves was generated decreased in CRC-MSKCC. The concentration changes of 13 meta- and used for quantification. Linear regression (R2) for the cali- bolites were dependent on the cohort rather than colorectal cancer bration curves of each metabolite were >0.99 for all plates. For status and were discarded from future analyses (Table 2). each plate, the LLODs were calculated to be three times the values of the matrix solutions and an average of LLODs from 5 plates Prediction models were reported in Supplementary Table S1 and used for later To construct an effective diagnostic model for colorectal cancer, analysis. Metabolites concentration that is lower than the LLOD we conducted multivariate analysis using MetaboAnalyst. Among was unreliable and classified as missing value. A total of 46 the PCA, PLS-DA, and sPLS-DA model options, sPLS-DA provided metabolite features (including methyl-histidine, propionic acid, the best separation between the groups with the least number of isobutyric acid, and 43 lipids) were removed as >50% of the metabolites. Figure 2A shows the separation plot from sPLS-DA information was missing (Supplementary Table S1). Three QC with component 1 and component 2. The classification error rate samples at different concentration levels were included in each was 11.4%. The metabolites selected by the sPLS-DA model for 96-well plate to assess the coefficient of variation (CV%) across component 1 and component 2 with their loading value are the 5 different plates. The CV% of QC samples for each metabolite shown in Fig. 2B and C. Notably, diacetylspermine, proline, was calculated as the SD divided by the average. Notably, the CV% kynurenine, and glucose were among the top six selected features for each metabolite across was <15% indicating a robust analytic based on loading values for component 1. This confirms their method. selection as consistent markers. Finally, logistic regression models were constructed in R with Potential biomarkers for colorectal cancer selected metabolites. We used a leave-out approach to build and Potential biomarkers for colorectal cancer were identified by evaluate models as it is most rigorous. A total of 121 CRC-CAD comparing the metabolomic profile from colorectal cancer versus and 121 controls were used as training set to build a model and 50 controls for both the fold change (FC) analyses and t-tests. A total CRC-MSKCC and 50 controls were used as testing set to validate of 17 metabolites were identified by volcano plot with a threshold the model. The first model (I) used the 17 metabolites listed for FC either >2or<0.5 and P < 0.05 (Table 2). Results from the in Table 2 selected according to the volcano plot of colorectal one-way ANOVA analyses for the three study groups identified cancer versus control. This model had an AUC value of 0.967 for
www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 1285
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Deng et al.
Table 2. Potential colorectal cancer biomarkers Metabolite concentration change relative to controls Metabolite HMDB ID FC P CRC-CAD CRC-MSKCC Consistent biomarker 1. Diacetylspermine HMDB02172 10.75 3.61E–31 þþ Yes 2. Proline HMDB00162 2.53 4.04E–31 þþ Yes 3. C14:1 HMDB62588 3.20 3.19E–22 þ NC No 4. Kynurenine HMDB00684 3.50 6.53E–16 þþ Yes 5. Glucose HMDB00122 3.06 1.90E–15 þþ Yes 6. HPHPA HMDB02643 0.33 9.44E–11 NC No 7. Aspartic acid HMDB00191 0.32 5.73E–10 NC No 8. Beta-hydroxybutyric acid HMDB00357 17.56 2.55E–09 NC þ No 9. DOPA HMDB00181 14.63 5.57E–09 NC þ No 10. 4-Hydroxyproline HMDB00725 2.53 1.31E–08 NC þ No 11. Aminoadipic acid HMDB00510 0.47 2.70E–08 NC No 12. Putrescine HMDB01414 3.78 1.36E–05 NC þ No 13. Indole acetic acid HMDB00197 0.21 2.06E–04 NC No 14. Hippuric acid HMDB00714 0.39 4.42E–04 NC No 15. Citric acid HMDB00094 3.07 1.18E–03 NC þ No 16. Sarcosine HMDB00271 14.68 1.82E–03 NC NC No 17. Butyric acid HMDB00039 0.19 9.72E–03 þ No NOTE: "þ" indicates a significant metabolite concentration increase; "–" indicates a significant metabolite concentration decrease; "NC" means that the metabolite concentration was not significantly changed.
training set and 0.868 for testing set (Fig. 3IA and B). At specificity cancer biomarkers from the ANOVA analysis. The model had an of 80%, the model's sensitivity were 99.2% for training set and AUC of 0.903 for training set and an AUC of 0.873 on testing set 74.0% for testing set, respectively (Table 3). The second model (II) (Fig. 3IIA and B) with a training sensitivity of 82.6% and a testing was limited to the four metabolites (e.g., proline, diacetylsper- sensitivity of 72.% at specificity of 80% (Table 3). The last logistic mine, kynurenine, and glucose) identified as robust colorectal regression model (III) incorporated only diacetylspermine and
A Normalized concentrations B Normalized concentrations
3 3
2 2 1 1 0 0 −1 −1 −2 −2 Figure 1. Controls CRC-CAD CRC-MSKCC Controls CRC-CAD CRC-MSKCC Normalized concentrations of metabolites for controls, CRC-CAD, and CRC-MSKCC study groups for C Normalized concentrations D Normalized concentrations diacetylspermine (A); HPHPA (B); aspartic acid (C); and butyric 4 1.5 acid (D).
1.0 3
0.5 2 0.0 1 −0.5 0 −1.0
−1.5 −1
Controls CRC-CAD CRC-MSKCC Controls CRC-CAD CRC-MSKCC
1286 Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 Cancer Epidemiology, Biomarkers & Prevention
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Urinary Biomarker Panel for Detecting Colorectal Cancer
A Scores plot With an AUC of 0.868 on training set and an AUC of 0.851 on CRC Normal testing set (Fig. 3IIIA and B), model III has the least AUC drop from training to testing among 3 models that confirmed the robustness of the selected biomarkers. At specificity of 80%, model III's sensitivity were 80.0% for training set and 74.0% for testing set, respectively (Table 3).
Discussion We have identified a discrete subset of common urinary meta-
Component 2 (5.7%) 2 Component bolites that may serve as potential biomarkers for colorectal 2 0 2 2 0 2 6 4 − cancer when used in combination based upon modeling to separate colorectal cancer and control samples. An sPLS-DA
model with two components was built with a classification error 4 4 − rate of 11.4%. For logistic models, the AUC varied from 0.965 to 0.868 highlighting the predictive power of urinary metabolomics −4 −2 0 2 4 for colorectal cancer screening. However, given the sample size Component 1 (9.1%) (n ¼ 342), one needs to be conscientious about error due to fi B over tting the model. To guard against this, further analyses were CRCNormal performed by building a model that only used consistent bio-
Diacetylspermine markers regardless of the cohorts. Finally, a metabolomic predic- tor for colorectal cancer was built with two metabolites: diace- Proline tylspermine and kynurenine. At its optimal cut-off value of 0.498, C14.1 High the predictor's specificity and sensitivity values were 90.6% and Kynurenine 74.3%, respectively. C5MDC The mechanism of diacetylspermine and kynurenine being
Glucose colorectal cancer markers still needs to be investigated. Here, we plotted the trend of their changes from control, to stage 0, to stage C14.1OH I, to stage II, to stage III, to stage IV in Fig. 4. For both diacetyl- Low C4 spermine and kynurenine, the biggest change was observed from Choline control to stage 0 confirming the usage of these two markers for
C5.1 early screening. There was a continuous increase in diacetylsper- mine as the cancer progresses. The final metabolites, diacetylsper-
0.1 0.2 0.3 0.4 0.5 0.6 mine and kynurenine, have been associated with cancer detection Loadings 1 in the past. For instance, increased urinary kynurenine concentra- tions were first identified in patients with different malignancies C by Spacek in 1955 (37). Urine samples were collected without CRCNormal dietary modifications, and the kynurenine levels increased from 1- total.DMA to 7-fold in patients with colorectal cancer. Several teams have Glucose identified diacetylspermine's presence in urine in association with hepatocellular carcinoma (sensitivity of 65.5%, specificity versus C5.1 High cirrhosis of 76.0%; ref. 38), breast and colorectal cancers (sensi- Ac.Om tivity was 60.2% and 75.8%, respectively; ref. 39), pancreatobili- Homovanillic acid ary cancer (sensitivity of 75%; ref. 40), and non–small cell lung Ascorbic acid cancer recurrence following resection (sensitivity 62.2%; ref. 41). Aspartic acid In spite of its utility, urinary diacetylspermine was unable to
C4.1 Low discriminate between patients with and without bladder can- cer (42). Enrichment of proline has been identified as a biomarker HPHPA for colorectal cancer based upon serum, tissue, urine, exhaled C5OH breath, and plasma (43). The urinary metabolite glucose is typically associated with reduced concentrations in samples from 0.1 0.2 0.3 0.4 0.5 0.6 patients with cancer compared with healthy controls while we Loadings 2 report increased levels in both colorectal cancer groups (43). Although our approach to diagnose colorectal cancer is novel Figure 2. Results showing separation plot from sPLS-DA with component 1 and and promising, there were several limitations to this study. component 2 (A); variables selected by the sPLS-DA model for component 1 Smoking and age are known contributors to colorectal can- (B); and variables selected by the sPLS-DA model for component 2 (C). cer (44). As such, we tried to match controls and colorectal cancer cases based upon smoking status and age; however, this was not kynurenine. Proline and glucose were excluded due to their possible due to higher than expected rates of smoking (current, potential association with diet (36), a feature that was not prior, never) and age in the two colorectal cancer groups. This may controlled during the 24 hours prior to urine sample collection. have impacted the selection of metabolites in a negative way.
www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 1287
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Deng et al.
Figure 3. The ROC curve of Model I on training set using 17 metabolites (IA), Model I on testing set using 17 metabolites (IB), Model II on training set using 4 metabolites (IIA), Model II on testing set using 4 metabolites (IIB), Model III on training set using diacetylspermine and kynurenine (IIIA), and Model III on testing set using diacetylspermine and kynurenine (IIIB).
Table 3. The AUC of the ROC curve, sensitivity, and specificity for each model Logistic AUC Sensitivity at specificity of 80% regression Delta Delta models Features Train Test (Train-Test) Train Test (Train-Test) I Proline, diacetylspermine, C14.1, kynurenine, glucose, aspartic acid, Glutamate, 0.967 0.868 0.099 99.2% 74.0% 25.2% Beta-Hydroxybutyric acid, HPHPA, DOPA, c4-OH, proline, putrescine, indole acetic acid, citric acid, hippuric acid, sarcosine, and butyric acid II Proline, diacetylspermine, kynurenine, and glucose 0.903 0.873 0.030 82.6% 72.0% 10.6% III Diacetylspermine and kynurenine 0.864 0.851 0.013 80.0% 74.0% 6.0%
1288 Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 Cancer Epidemiology, Biomarkers & Prevention
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Urinary Biomarker Panel for Detecting Colorectal Cancer
ABNormalized concentration Normalized concentration
3 3
2 2 1 Figure 4. 1 The normalized concentration trend 0 for diacetylspermine (A)and kynurenine (B) from controls, to 0 −1 stage 0, to stage I, to stage II, to stage III, and to stage IV. − −1 2
−3
−2
Stage I Stage
Stage I Stage
Control
Control
Stage II Stage Stage 0 Stage
Stage II Stage
Stage 0 Stage
Stage III Stage
Stage III Stage
Stage IV Stage Stage IV Stage
Urinary metabolites are waste products, it is unclear the upstream that is readily available, straightforward to collect as part of any metabolic role of either diacetylspermine or kynurenine in cancer physician's clinic visit, and acceptable to patients in most cultures. pathogenesis. Knowing more about the metabolic cycles and Further supporting the use of urine is availability of collection, degradation pathways involved in colorectal cancer will be help- handling, shipping, and storage protocols many of which have ful to identify additional biomarkers. The specificity of the met- been instituted by major biobanks and repositories. A 2018 abolic profile must also be evaluated by comparing with samples systematic review of 16 urinary metabolomic studies in colorectal from patients with other cancer types. Although promising results cancer listed metabolites independently reported three or more were obtained, the metabolomic profile obtained cannot yet be times (47); none of which were the same as those we reported. As considered definitive and need to be tested in clinical setting, the largest, multicenter urine-based metabolomics study con- ideally within a pragmatic study setting to make the findings ducted to date (43, 47), there were insufficient samples at each relevant and generalizable to others. Testing the predictive per- cancer stage to analyze them independently or in sequence to formance of the metabolite profile against other cancers is espe- understand the disease trajectory. Larger datasets supported by cially relevant as diacetylspermine has been included in many comprehensive clinicodemographic characteristics will be valu- noncolorectal cancer panels. In addition, it may be beneficial to able to discern the discrete shifts in metabolites associated with make a more comprehensive metabolomic assessment. This real-time changes in cellular metabolism associated with disease. could be done using additional analytic assays, such as gas This will also facilitate external validation of putative biomarker chromatography–mass spectrometry, which will enable the detec- panels such as that reported herein. tion of more metabolites (45). A more comprehensive metabo- lomic profile may improve diagnostic accuracy. It is possible that Disclosure of Potential Conflicts of Interest we could derive a better understanding of the underlying meta- L. Deng is a Senior Scientist at Metabolomic Technologies, Inc. No potential bolic processes associated with colorectal cancer. We intentionally conflicts of interest were disclosed by the other authors. did not have patients follow a controlled diet or fast before providing a urine sample. Appreciating the diurnal changes in Authors' Contributions urinary metabolite concentrations (46), all collections were com- Conception and design: L. Deng, H. Wang, O.I. Alatise, M.R. Weiser, pleted during daytime business hours. Dietary controls place T.P. Kingham, D. Chang unreasonable burdens on patients and believed that this would Development of methodology: L. Deng, H. Wang, O.I. Alatise, D. Chang decrease the value of this or any urinary biomarker panel intended Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Deng, J. Constable, H. Wang, O.I. Alatise, for use as a screening tool for colorectal cancer. Furthermore, it is T.P. Kingham highly probably that differences in the intestinal microbiota Analysis and interpretation of data (e.g., statistical analysis, biostatistics, between healthy individuals and those with colorectal cancer computational analysis): L. Deng, K. Ismond, Z. Liu, O.I. Alatise, M.R. Weiser impact urinary metabolites more so than diet (47). A limitation Writing, review, and/or revision of the manuscript: L. Deng, K. Ismond, of any large multicenter study is the need to handle, ship, and J. Constable, H. Wang, O.I. Alatise, M.R. Weiser, T.P. Kingham store the biosamples over time. To minimize metabolite degra- Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Deng, K. Ismond, J. Constable, O.I. Alatise dation, all specimens were handled similarly regardless of collec- Study supervision: L. Deng, O.I. Alatise, M.R. Weiser, D. Chang tion date and aliquoting prior to the first freeze at 80 C pre- – vented exposure to multiple freeze thaw cycles (48, 49). Acknowledgments In conclusion, this metabolomic-based predictor for colorectal We would like to express our deep gratitude to Dr. Richard N. Fedorak, who cancer has potential clinical application for population-based contributed to this project and passed away on Nov. 8, 2018. This work was colorectal cancer screening using urine; a preferred biosample funded, in part, by the National Institute of Biomedical Imaging and
www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 1289
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Deng et al.
Bioengineering (NIBIB), NIH (K. Ismond, O.I. Alatise, and T.P. Kingham are advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate supported by grant number UG3EB024965), and Mitacs (IT10425, to Z. Liu). this fact.
The costs of publication of this article were defrayed in part by the Received January 14, 2019; revised March 29, 2019; accepted May 28, 2019; payment of page charges. This article must therefore be hereby marked published first May 31, 2019.
References 1. Levin TR, Corley DA, Jensen CD, Schottinger JE, Quinn VP, Zauber AG, et al. 20. Guo C, Xie C, Chen Q, Cao X, Guo M, Zheng S, et al. A novel malic acid- Effects of organized colorectal cancer screening on cancer incidence and enhanced method for the analysis of 5-methyl-2'-deoxycytidine, 5-hydro- mortality in a large community-based population. Gastroenterology 2018; xymethyl-2'-deoxycytidine, 5-methylcytidine and 5-hydroxymethylcyti- 155:1383–91. dine in human urine using hydrophilic interaction liquid chromatogra- 2. Dube C. Organized screening is better than opportunistic screening phy-tandem mass spectrometry. Anal Chim Acta 2018;1034:110–8. at decreasing the burden of colorectal cancer in the United States. 21. Nakajima T, Katsumata K, Kuwabara H, Soya R, Enomoto M, Ishizaki T, Gastroenterology 2018;155:1302–4. et al. Urinary polyamine biomarker panels with machine-learning differ- 3. Navarro M, Nicolas A, Ferrandez A, Lanas A. Colorectal cancer population entiated colorectal cancers, benign disease, and healthy controls. Int J Mol screening programs worldwide in 2016: an update. World J Gastroenterol Sci 2018;19. doi: 10.3390/ijms19030756. 2017;23:3632–42. 22. Venalainen MK, Roine AN, Hakkinen MR, Vepsalainen JJ, Kumpulainen 4. Schreuders EH, Ruco A, Rabeneck L, Schoen RE, Sung JJ, Young GP, et al. PS, Kiviniemi MS, et al. Altered polyamine profiles in colorectal cancer. Colorectal cancer screening: a global overview of existing programmes. Gut Anticancer Res 2018;38:3601–7. 2015;64:1637–49. 23. Wang H, Tso V, Wong C, Sadowski D, Fedorak RN. Development and 5. Imperiale TF, Ransohoff DF, Itzkowitz SH, Levin TR, Lavin P, Lidgard GP, validation of a highly sensitive urine-based test to identify patients with et al. Multitarget stool DNA testing for colorectal-cancer screening. N Engl J colonic adenomatous polyps. Clin Transl Gastroenterol 2014;5:e54. Med 2014;370:1287–97. 24. Deng L, Chang D, Foshaug RR, Eisner R, Tso VK, Wishart DS, et al. 6. Quintero E, Castells A, Bujanda L, Cubiella J, Salas D, Lanas A, et al. Development and validation of a high-throughput mass spectrometry Colonoscopy versus fecal immunochemical testing in colorectal-cancer based urine metabolomic test for the detection of colonic adenomatous screening. N Engl J Med 2012;366:697–706. polyps. Metabolites 2017;7:32. 7. van Roon AH, Goede SL, van Ballegooijen M, van Vuuren AJ, Looman CW, 25. Deng L, Fang H, Tso VK, Sun Y, Foshaug RR, Krahn SC, et al. Clinical Biermann K, et al. Random comparison of repeated faecal immunochem- validation of a novel urine-based metabolomic test for the detection of ical testing at different intervals for population-based colorectal cancer colonic polyps on Chinese population. Int J Colorectal Dis 2017;32:741–3. screening. Gut 2013;62:409–15. 26. Tso V ER, Macleod S, Ismond KP, Foshaug RR, Wang H, Joseph R, et al. 8. Zubero MB, Arana-Arri E, Pijoan JI, Portillo I, Idigoras I, Lopez-Urrutia A, Consistency of metabolite determination from NMR spectra over time and et al. Population-based colorectal cancer screening: comparison of two between operators. Metabolomics 2015;5:151. fecal occult blood test. Front Pharmacol 2014;4:175. 27. Eisner R, Greiner R, Tso V, Wang H, Fedorak RN. A machine-learned 9. Singh H, Bernstein CN, Samadder JN, Ahmed R. Screening rates for predictor of colonic polyps based on urinary metabolomics. colorectal cancer in Canada: a cross-sectional study. CMAJ Open 2015; Biomed Res Int 2013;2013:303982. 3:E149–57. 28. Wong CK, Fedorak RN, Prosser CI, Stewart ME, van Zanten SV, Sadowski 10. Singal AG, Corley DA, Kamineni A, Garcia M, Zheng Y, Doria-Rose PV, et al. DC. The sensitivity and specificity of guaiac and immunochemical fecal Patterns and predictors of repeat fecal immunochemical and occult blood occult blood tests for the detection of advanced colonic adenomas and test screening in four large health care systems in the United States. Am J cancer. Int J Colorectal Dis 2012;27:1657–64. Gastroenterol 2018;113:746–54. 29. Chong J, Soufan O, Li C, Caraus I, Li S, Bourque G, et al. MetaboAnalyst 4.0: 11. Church J. Complications of colonoscopy. Gastroenterol Clin North Am towards more transparent and integrative metabolomics analysis. 2013;42:639–57. Nucleic Acids Res 2018;46:W486–W94. 12. Dougherty MK, Brenner AT, Crockett SD, Gupta S, Wheeler SB, Coker- 30. Altobelli E, Angeletti PM, Latella G. Role of urinary biomarkers in the Schwimmer M, et al. Evaluation of interventions intended to increase diagnosis of adenoma and colorectal cancer: a systematic review and meta- colorectal cancer screening rates in the United States: a systematic review analysis. J Cancer 2016;7:1984–2004. and meta-analysis. JAMA Intern Med 2018;178:1645–58. 31. Qiu G, Zheng Y, Wang H, Sun J, Ma H, Xiao Y, et al. Plasma metabolomics 13. Cossu G, Saba L, Minerba L, Mascalchi M. Colorectal cancer screening: the identified novel metabolites associated with risk of type 2 diabetes in two role of psychological, social and background factors in decision-making prospective cohorts of Chinese adults. Int J Epidemiol 2016;45:1507–16. process. Clin Pract Epidemiol Ment Health 2018;14:63–9. 32. Stoessel D, Stellmann JP, Willing A, Behrens B, Rosenkranz SC, Hodecker 14. Osborne JM, Flight I, Wilson CJ, Chen G, Ratcliffe J, Young GP. The impact SC, et al. Metabolomic profiles for primary progressive multiple sclerosis of sample type and procedural attributes on relative acceptability of stratification and disease course monitoring. Front Hum Neurosci 2018; different colorectal cancer screening regimens. Patient Prefer Adherence 12:226. 2018;12:1825–36. 33. Delplancke TDJ, de Seymour JV, Tong C, Sulek K, Xia Y, Zhang H, et al. 15. Liles EG, Coronado GD, Perrin N, Harte AH, Nungesser R, Quigley N, et al. Analysis of sequential hair segments reflects changes in the metabolome Uptake of a colorectal cancer screening blood test is higher than of a fecal across the trimesters of pregnancy. Sci Rep 2018;8:36. test offered in clinic: a randomized trial. Cancer Treat Res Commun 2017; 34. L^e Cao KA, Boitard S, Besse P. Sparse PLS discriminant analysis: biologically 10:27–31. relevant feature selection and graphical displays for multiclass problems. 16. Lamb YN, Dhillon S. Epi proColon 2.0 CE: a blood-based screening test BMC Bioinformatics 2011;12:253. for colorectal cancer. Mol Diagn Ther 2017;21:225–32. 35. Friedman JH, Hastie T, Tibshirani R. Regularization paths for generalized 17. Anabtawi A, Mathew LM. Improving compliance with screening of diabetic linear models via coordinate descent. J StatSoft 2010;33:1–22. patients for microalbuminuria in primary care practice. ISRN Endocrinol- 36. Cifuentes A. Foodomics: advanced mass spectrometry in modern food ogy 2013;2013:893913. science and nutrition. Hoboken, NJ: John Wiley & Sons, Inc.; 2013. 18. Oboler SK, Prochazka AV, Gonzales R, Xu S, Anderson RJ. Public expecta- 37. Spacek M. Kynurenine in disease, with particular reference to cancer. tions and attitudes for annual physical examinations and testing. Can Med Assoc J 1955;73:198–201. Ann Intern Med 2002;136:652–9. 38. Enjoji M, Nakamuta M, Arimura E, Morizono S, Kuniyoshi M, Fukush- 19. Widlak MM, Neal M, Daulton E, Thomas CL, Tomkins C, Singh B, et al. Risk ima M, et al. Clinical significance of urinary N1,N12-diacetylspermine stratification of symptomatic patients suspected of colorectal cancer using levels in patients with hepatocellular carcinoma. Int J Biol Markers faecal and urinary markers. Colorectal Dis 2018;20:O335–O42. 2004;19:322–7.
1290 Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 Cancer Epidemiology, Biomarkers & Prevention
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Urinary Biomarker Panel for Detecting Colorectal Cancer
39. Hiramatsu K, Takahashi K, Yamaguchi T, Matsumoto H, Miyamoto H, 44. Liang PS, Chen TY, Giovannucci E. Cigarette smoking and colorectal cancer Tanaka S, et al. N(1),N(12)-Diacetylspermine as a sensitive and specific incidence and mortality: systematic review and meta-analysis. Int J Cancer novel marker for early- and late-stage colorectal and breast cancers. 2009;124:2406–15. Clin Cancer Res 2005;11:2986–90. 45. Bathe OF, Shaykhutdinov R, Kopciuk K, Weljie AM, McKay A, Sutherland 40. Yamaguchi K, Nakamura M, Shirahane K, Konomi H, Torata N, Hamasaki FR, et al. Feasibility of identifying pancreatic cancer based on serum N, et al. Urine diacetylspermine as a novel tumour maker for pancreato- metabolomics. Cancer Epidemiol Biomarkers Prev 2011;20:140–7. biliary carcinomas. Dig Liver Dis 2005;37:190–4. 46. Ni Y, Xie G, Jia W. Metabonomics of human colorectal cancer: new 41. Takahashi Y, Sakaguchi K, Horio H, Hiramatsu K, Moriya S, Takahashi K, approaches for early diagnosis and biomarker discovery. J Proteome Res et al. Urinary N1, N12-diacetylspermine is a non-invasive marker for the 2014;13:3857–70. diagnosis and prognosis of non-small-cell lung cancer. Br J Cancer 2015; 47. Erben V, Bhardwaj M, Schrotz-King P, Brenner H. Metabolomics biomar- 113:1493–501. kers for detection of colorectal neoplasms: a systematic review. Cancers 42. Stejskal D, Humenanska V, Hanulova Z, Fiala R, Vrtal R, Solichova P, et al. (Basel) 2018;10:1–24. Evaluation of urine N1,N12-Diacetylspermine as potential tumor marker 48. Rotter M, Brandmaier S, Prehn C, Adam J, Rabstein S, Gawrych K, et al. for urinary bladder cancer. Biomed Pap Med Fac Univ Palacky Olomouc Stability of targeted metabolite profiles of urine samples under different Czech Repub 2006;150:235–7. storage conditions. Metabolomics 2017;13:4. 43. Zhang F, Zhang Y, Zhao W, Deng K, Wang Z, Yang C, et al. Metabolomics for 49. Laparre J, Kaabia Z, Mooney M, Buckley T, Sherry M, Le Bizec B, et al. Impact biomarker discovery in the diagnosis, prognosis, survival and recurrence of of storage conditions on the urinary metabolomics fingerprint. Anal Chim colorectal cancer: a systematic review. Oncotarget 2017;8:35460–72. Acta 2017;951:99–107.
www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 28(8) August 2019 1291
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research. Published OnlineFirst May 31, 2019; DOI: 10.1158/1055-9965.EPI-18-1291
Urinary Metabolomics to Identify a Unique Biomarker Panel for Detecting Colorectal Cancer: A Multicenter Study
Lu Deng, Kathleen Ismond, Zhengjun Liu, et al.
Cancer Epidemiol Biomarkers Prev 2019;28:1283-1291. Published OnlineFirst May 31, 2019.
Updated version Access the most recent version of this article at: doi:10.1158/1055-9965.EPI-18-1291
Supplementary Access the most recent supplemental material at: Material http://cebp.aacrjournals.org/content/suppl/2019/05/31/1055-9965.EPI-18-1291.DC1
Cited articles This article cites 47 articles, 5 of which you can access for free at: http://cebp.aacrjournals.org/content/28/8/1283.full#ref-list-1
E-mail alerts Sign up to receive free email-alerts related to this article or journal.
Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department Subscriptions at [email protected].
Permissions To request permission to re-use all or part of this article, use this link http://cebp.aacrjournals.org/content/28/8/1283. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.
Downloaded from cebp.aacrjournals.org on September 29, 2021. © 2019 American Association for Cancer Research.