Sage Journals: Discover world-class research

Abstract

BACKGROUND:

African colorectal cancer (CRC) rates are rising rapidly. A low-cost CRC screening approach is needed to identify CRC from non-CRC patients who should be sent for colonoscopy (a scarcity in Africa).

OBJECTIVE:

To identify urinary metabolite biomarkers that, combined with easy-to-measure clinical variables, would identify patients that should be further screened for CRC by colonoscopy. Ideal metabolites would be water-soluble and easily translated into a sensitive, low-cost point-of-care (POC) test.

METHODS:

Liquid-chromatography mass spectrometry (LC-MS/MS) was used to quantify 142 metabolites in spot urine samples from 514 Nigerian CRC patients and healthy controls. Metabolite concentration data and clinical characteristics were used to determine optimal sets of biomarkers for identifying CRC from non-CRC subjects.

RESULTS:

Our statistical analysis identified N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippurate, $p$ -hydroxyhippurate, and glutamate as the best metabolites to discriminate CRC patients via POC screening. Logistic regression modeling using these metabolites plus clinical data achieved an area under the receiver-operator characteristic (AUCs) curves of 89.2% for the discovery set, and 89.7% for a separate validation set.

CONCLUSIONS:

Effective urinary biomarkers for CRC screening do exist. These results could be transferred into a simple, POC urinary test for screening CRC patients in Africa.

Keywords

Colorectal cancer screening cancer biomarkers urine test liquid chromatography-mass spectrometry metabolomic profiling logistic regression

1. Introduction

Colorectal cancer (CRC) is the third most common cancer in the world with an estimated 1.93 million new cases diagnosed in 2020 [1]. CRC rates are rapidly increasing worldwide with about 2.4 million cases/yr being expected by 2035 [2]. Until recently, CRC was a rare disease in Africa with rates as low as 1.2 per 100,000 (versus 38.7 per 100,000 in North America) [3]. However, this is changing rapidly. By 2035, the entire WHO African Region is anticipated to have the second highest rate of CRC growth [2]. The rising incidence of CRC in resource poor, low- and middle-income countries (LMICs) in Africa is attributed in part to the adoption of the Western diet and Western lifestyle [4, 5].

CRC disease management and presentation differ in continental Africa compared to Western countries. The average age of incidence is younger (43–46 years versus 78–85 years) [6, 7], the percentage of rectal cancer (compared to colon cancer) is higher in most African regions [8], and the metastatic patterns are different [9]. Furthermore, the African region has no population-wide CRC screening programs. As a result, up to 99% of individuals in Africa diagnosed with CRC have late-stage disease, where a cure is often not possible and the 1-year survival is just 47% [9]. In comparison, the 5-year survival rate in North America and Europe is 60.3% [10].

There are a number of methods that can be used for population-wide screening of CRC. These include fecal immunochemical tests (FIT) [11], stool DNA (sDNA [12]) tests such as Cologuard ${}^{\@setsize{\scriptsize}{8pt}{\viipt}{\@viipt}\textregistered}$ , blood-based DNA tests such as Septin 9 (SEPT9) [13], urinary mass spectrometry (MS) tests, such as PolypDX™ [14] and colonoscopy, the gold standard. The FIT and sDNA tests, with 92% and 74%, sensitivity for detecting CRC, respectively [15], have limited compliance and very low uptake in Western countries. For Africans, fecal donations may be forbidden, and therefore, many would be unable to provide samples for fecal FIT or sDNA tests. The PolypDX™ test is a fast and simple urine assay with greater sensitivity than the FIT. While acquiring a urine sample is more straightforward, specialized MS-instruments are required to analyze the urine samples. While these MS-based tests are easily accessible in Western countries, MS-instruments are not widely available in Africa. Colonoscopy is the most widely used method for CRC screening but compliance for colonoscopy is low. In a German study of colonoscopy compliance, 63% of patients refused to undergo colonoscopy [16].

For those in LMICs, colonoscopy is neither widely available nor affordable. Moreover, healthcare resources are not available to implement population-based CRC screening programs [6, 17, 18]. In response to increasing CRC rates in LMICS, a surveillance symptom-based model [17] was developed with the intent to reduce mortality from CRC. This point-of-care (POC) symptom-based test uses clinical characteristics of rectal bleeding, weight loss and changes in bowel habits. Such a test was found to be easily delivered and interpreted by healthcare professionals after minimal training. However, its sensitivity and specificity were of moderate quality: only 56% of high-risk patients had CRC (Stage II or III). Other simple, fast, easy-to-use methods are needed to compliment this surveillance symptom-based test to identify and prioritize those with CRC among high-risk individuals.

One fast, easy-to-use complementary approach would be to use metabolite detection for CRC screening. Many metabolomic studies have identified a plethora of metabolite biomarkers for CRC diagnosis from human biofluids, including urine [19, 20, 21, 22, 23, 24, 25, 26]. The close coupling between colonic metabolism and urinary metabolites suggests that urine could be a very informative medium for identifying easy-to-detect CRC biomarkers. Indeed, a recent meta-analysis examined 29 CRC cohort studies where quantitative metabolomic analyses of controls, CRC (stages 1–4), adenomas and polyps were examined [19]. This study identified 6 upregulated (3-hydroxybutyric acid, L-dopa, L-histidinol, and N ${}^{1}$ , N ${}^{12}$ -diacetylspermine) and 4 downregulated (pyruvic acid, hydroquinone, tartaric acid, and hippuric acid) urinary metabolites of clinical significance. Several metabolites were sufficiently abundant and easy to detect (via colorimetric, fluorometric or ELISA-like assays) which suggests that they could be translated to low-cost biomarker POC tests. However, none of these CRC metabolomic studies were conducted on an African population, which is known to have significant differences in CRC presentation, location and type.

The purpose of this study was to discover and validate a combination of urinary metabolite biomarkers for the detection of CRC in symptomatic Nigerian patients. We chose liquid chromatography tandem mass spectrometry (LC-MS/MS) and direct injection tandem mass spectrometry (DI-MS/MS), to help us identify and quantify a large pool of potential biomarkers. Our goal was to use the power of LC-MS to obtain a clinically relevant set of CRC predictive biomarkers that are sufficiently abundant and water-soluble, to make them translatable into a low-cost POC testing device. Coupling a low-cost urinary POC test with the clinical symptom-based test could further improve CRC screening performance. Indeed, when metabolite profiles and clinical characteristics are combined, the sensitivity and specificity of many metabolite tests and models increases [20, 27].

Here, we report the identification of four, high-abundance urinary metabolites that could serve as robust biomarkers of CRC in the Nigerian populations. We also show that by combining these urinary metabolites with clinical characteristics, a urine-based screening test, with good potential for conversion to a POC screening test, can be developed that has high sensitivity to discriminate between symptomatic patients at higher risk for CRC versus others.

2. Materials and methods

2.1 Study design

This was a prospective, observational study designed to develop a urine-based screening test using metabolomics to discriminate between those at higher-risk for CRC from other non-CRC conditions. Study participants provided informed written consent prior to enrolment in accordance with the Obafemi Awolowo University Teaching Hospital (OAUTH) Ethics and Research Committee, which approved the study (IRB/IEC/ 0004553). The study conforms with The Code of Ethics of the World Medical Association (Declaration of Helsinki). Ethics approvals covering sample processing were received by the Health Research Ethics Board at the University of Alberta (Pro00074045). The study was registered with ClinicalTrials.gov as NCT03173729 and has an earlier submission of NCT03032874.

2.2 Study population

Recruitment of eligible patients, which was managed by the African Research Group for Oncology (ARGO) Consortium, occurred at one of five Nigerian hospitals: OAUTH, Ibadan University Teaching Hospital, University Ilorin Teaching Hospital, Federal Medical Center, Owo, and Ladoke Akintola University of Technology (LAUTECH). The catchment area of the hospital network includes all the Osun, Ekiti, and Ondo states and part of the Oyo, Kwara, Kogi, Lagos, and Edo states providing care to over 15 million people. Eligible study participants were accrued from one of two populations: adults ( $\geqslant$ 18 yrs) presenting with rectal bleeding symptoms and patients ( $\geqslant$ 18 yrs) with a biopsy-confirmed diagnosis of CRC. A total of 512 participants, including 225 healthy controls, 77 participants with confirmed colonic polyps, and 212 with confirmed CRC (determined by pathological findings – see Section 2.3) were identified and included in the study.

2.3 Sample and clinical data collection

At recruitment, all participants completed a questionnaire (see Table S1) concerning their symptoms and medical history and provided a mid-stream spot urine sample. Urine samples were frozen at $-$ 80 ${}^{\circ}$ C within 1 hour of collection. Next, all participants had a standard colonoscopy, examining from the ileocecal valve to the anus. The extent and quality of the colonoscopy was documented to ensure a standardized quality of colonoscopy. Data points from the colonoscopy included the presence of internal or external hemorrhoids, quality of the bowel preparation, presence and location of polyps, CRC, and diverticulum, source of bleeding, complications of colonoscopy, and cancer stage if present (see Table S2). In addition, if a patient was diagnosed with CRC, they were counseled about a treatment plan by the investigators at each of the study hospitals.

Pathologic samples were read at OAU, Ibadan, and Ilorin for all patients with biopsies of CRC or precancerous polyps (adenomatous, hyperplastic, hamartomatous, or inflammatory). A pathologist, Dr. Vakiani, at Memorial Sloan Kettering Cancer Center, performed a central pathologic review remotely for pathologic quality assurance and accuracy of the reference standard results. Prior to conducting the MS-based metabolomic analyses, the urine samples were divided into two batches. The first batch included 413 samples (consisting of 169 biopsy-confirmed CRC cases, 50 polyp cases and 194 healthy controls) which were assigned to the discovery set. The second batch of the remaining 101 samples (consisting of 43 biopsy-confirmed CRC cases, 27 polyp cases and 31 healthy controls) were assigned to the validation set.

2.4 Chemicals, reagents and materials for mass spectrometry-based metabolomic assays

Optima™ LC/MS grade formic acid and high performance liquid chromatography (HPLC) grade water were purchased from Fisher Scientific (Ottawa, CA). Pure reference standard compounds, except 3-(3-hydroxyphenyl)-3-hydroxypropionic acid (HPHPA), were purchased from Sigma-Aldrich (Oakville, CA). ${}^{2}$ H-, ${}^{13}$ C- and ${}^{15}$ N-labelled compounds (except ${}^{13}$ C-labelled HPHPA), which were used as internal quantification standards for amino acids, biogenic amines, carnitines and derivatives, phosphatidylcholines and their derivatives, were purchased from Cambridge Isotope Laboratories, Inc. (Tewksbury, MA, USA). HPHPA and ${}^{13}$ C-labelled HPHPA were synthesized in-house as described previously [28]. Optima™ LC/MS grade ammonium acetate, phenylisothiocyanate (PITC), 3-nitrophenylhydrazine (3-NPH), 1 -ethyl-3-(3-dimethyl-aminopropyl) carbodiimide (EDC) and butylated hydroxytoluene (BHT), HPLC grade pyridine, HPLC grade methanol, HPLC grade ethanol and HPLC grade acetonitrile (ACN) were purchased from Sigma-Aldrich (Oakville, CA). Nunc ${}^{\@setsize{\scriptsize}{8pt}{\viipt}{\@viipt}\textregistered}$ 96 DeepWell™plates were purchased from Sigma-Aldrich (Oakville, CA).

2.5 Stock solutions, internal standard mixture, and calibration curve standards for metabolomic assays

All solid chemicals were carefully weighed on a CPA225D semi-microelectronic balance (Sartorius, NY, USA) with a precision of 0.0001 g. Stock solutions of each compound were prepared by dissolving the accurately weighed solids in double-distilled water. Calibration curve standards were obtained by mixing and diluting the corresponding stock solutions with double-distilled water. For amino acids, biogenic amines, carbohydrates, carnitines and derivatives, phosphatidylcholines and their derivatives, stock solutions of isotope-labelled compounds were prepared in the same way. A working internal standard (ISTD) solution mixture was also made by mixing all the prepared isotope-labeled stock solutions together in water. For organic acids, stock solutions of isotope-labelled compounds were prepared by dissolving the accurately weighed solids in 75% aqueous methanol. A working ISTD solution mixture was made by mixing and diluting all the isotope-labelled stock solutions in 75% aqueous methanol. All standard solutions were aliquoted and stored at $-$ 80 ${}^{\circ}$ C until further use.

2.6 Sample preparation and liquid chromatography and direct injection mass spectrometry for metabolomic assays

A targeted, quantitative MS-based metabolomics approach was used to analyze the urine samples using a combination of DI-MS and reverse-phase high performance liquid chromatography tandem mass spectrometry (MS/MS). MS/MS analysis of the endogenous metabolites in the urine samples, including amino acids, organic acids, biogenic amines, acylcarnitines, glycerophospholipids, sphingolipids and sugars, was performed on an HPLC (Agilent 1260 HPLC, Agilent Technologies, Santa Clara, USA) equipped Qtrap ${}^{\@setsize{\scriptsize}{8pt}{\viipt}{\@viipt}\textregistered}$ 4000 tandem mass spectrometry instrument (Applied Biosystems/MDS Analytical Technologies, Foster, NE, USA). An Agilent reversed-phase Zorbax Eclipse XDB C18 column coupled to the Qtrap ${}^{\@setsize{\scriptsize}{8pt}{\viipt}{\@viipt}\textregistered}$ 4000 was used for the LC separations for LC-MS/MS analysis. The methods for urine sample preparation and MS analysis have been described in our previous publication [29].

2.7 Statistical analysis

Numeric and categorical clinical variables were analyzed by Mann-Whitney rank sum tests and $\chi^{2}$ tests with a $p$ -value threshold at 0.05 using the R statistical programming language (R 3.6.3) [30]. Recommended statistical procedures for standard quantitative metabolomic analysis were followed [31]. Metabolites with $>$ 80% of missing values (in all groups) were removed from further analysis. For metabolites missing $<$ 80% of their concentrations, values were imputed by using 1/5 of the minimum detectable concentration value for that metabolite. The raw concentrations were first normalized against the urinary creatinine level of the same sample. Further processing, including transformation, and auto-scaling (mean-centered and divided by the standard deviation of each variable) were applied for further data scaling and normalization. Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were performed by using MetaboAnalyst [32]. A 1000-fold permutation test was performed to assess the probability that the observed separation of the PLS-DA was due to chance.

Logistic regression with a least absolute shrinkage and selection operator (LASSO) feature selection algorithm was used to develop diagnostic models of CRC using both metabolite and clinical variables. Logistic regression and LASSO feature selection were performed by using R 3.6.3 [30]. Optimal regression models for CRC diagnosis were first identified on the discovery sets. Then the validation datasets, which were acquired from samples of the patients recruited independently from the discovery cohort, were used to validate the regression models. The area under the receiver-operator characteristic curves (AUC), sensitivities/specificities at selected cut-off points and the 95% confidence intervals were calculated for the discovery and the validation sets for all models using the pROC R package [33]. Cut-off points were selected by calculating the Youden Index (J $=\text{max}$ {Sensitivity $+$ Specificity $-$ 1}).

3. Results

3.1 Study population and statistical data processing

Table 1
Summary of grouping of samples

Group	Number of samples	Age at consent			Gender
		Range	Median	Mean	Male	Female
Discovery set
Colorectal cancer (CRC)	169	19–73	55	51.9	95	74
Polyps	50	30–75	56	57.2	34	16
Healthy controls	194	21–74	56	55.3	125	69
Total	413	19–75	56	54.1	254	159
Validation set
CRC	43	20–75	52	51.9	20	23
Polyps	27	43–74	56	57.8	19	8
Healthy controls	31	28–72	54	52.5	19	12
Total	101	20–75	55	53.6	58	43

A complete data set (survey, urine samples, and pathology reports) was collected from 514 recruited participants (Table 1). The mean age was 53.9 years (range: 19–75 yrs) of which 60.7% were male (data obtained from study questionnaire, Table S1). As noted earlier, the samples were divided into two batches, with 413 samples assigned to the discovery set and 101 samples assigned to the validation set. The discovery and validation sets were built using a ratio of 80.4%:19.6% of participants. Clinical information including age, biological sex, weight, height, personal smoking history [Y/N], cancer history [Y/N], blood in stool [Y/N], hemorrhoids [Y/N], weight loss in 6 months [Y/N] were used as the clinical variables for further statistical analysis. As described in Table S1, additional information on bowel habits and stool features were also included. The feature “change in caliber of stool” was classified as “pellet-like [Y/N]”, “watery [Y/N]”, “diarrhea [Y/N]”, and “alternating constipation with diarrhea/constipation [Y/N]”. All descriptions were introduced into the statistical analysis as independent clinical variables. Colonic polyps detected by colonoscopy are not considered cancerous (they may be precancerous) and are frequently detected in healthy individuals. As a result, patients with colonic polyps were combined with healthy controls as one “Normal” group for the preliminary statistical analysis of the clinical variables. To confirm that the same patient demographics were present in the CRC and the Normal group, Mann-Whitney rank sum tests and $\chi^{2}$ tests were performed. No statistical significance was observed in either the age and biological sex distribution, demonstrating that the CRC, polyps, and healthy control groups were age and gender matched (Table S3). To determine which clinical variables were statistically significant between the CRC and Normal group, $\chi^{2}$ tests of the other clinical variables were performed. As shown in Table S3, except for “constipation”, “personal smoking history” and “personal cancer history”, the remaining clinical variables (weight loss, blood in stool, etc.) were statistically different between the CRC and Normal groups. This suggested that the clinical features of weight loss, blood in stool and the four changes in stool calibre (pellet-like, watery, diarrhea, alternating constipation/diarrhea) could be used to generate our logistic models in combination with the quantified urinary metabolites.

Figure 1.

(A) Partial least squares discriminant analysis (PLS-DA) 2D-scores plot of colorectal cancer (CRC) patients ( $n=$ 169) and healthy controls ( $n=$ 194) using metabolite profiles only; (B) variable importance of projection (VIP) scores plot of important urinary metabolite coefficients. Abbreviations: healthy controls (HC); 3-(3-hydroxyphenyl)-3-hydroxypropanoic acid (HPHPA); lysophosphatidylcholine 28:1 (LYSOC28:1).

Using LC-MS/MS and DI-MS/MS analysis, the absolute concentrations of 142 urinary metabolites were identified and quantified in all samples using our previously published, targeted urinary assay [29]. The identified metabolites include 28 amino acids and derivatives, 17 organic acids, 22 biogenic amines and derivatives, 40 acylcarnitines, 34 lipids, and glucose/hexose. Data processing of the measured metabolites was first performed to identify metabolites with missing values. Thirty-eight metabolites had a high ( $>$ 80%) fraction of missing values and were excluded from the further analysis as they had urinary concentrations below the assay’s limit of detection (LOD). To ensure there were no batch effects between the discovery and validation sets, a principal component analysis (PCA) was performed. As shown in Fig. S1, no significant batch effect was detected. Therefore, the second batch of urine samples could be used to validate the models generated from the first batch used with the discovery set.

3.2 Statistical analysis: CRC patients versus healthy controls

To determine if there was a difference between the urinary metabolite profiles of the CRC and healthy control groups (excluding those patients with polyps), partial least squares discriminant analysis (PLS-DA) was performed using the remaining 104 metabolites. Using the discovery set, a clear two-dimensional separation in the PLS-DA was achieved between the CRC and healthy control groups (Fig. 1A). Using permutation testing, we confirmed that the observed separation between the cases and controls was not due to chance ( $p$ -value $<$ 0.001, Fig. S2A). Figure 1B displays the results of the overall coefficient scores from the PLS-DA. The most important metabolites that contributed to the separation included N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid, and glutamic acid.

Figure 2.

(A) PLS-DA separation of CRC ( $n=$ 169) and normal ( $n=$ 194) using metabolite profiles plus clinical features; (B) VIP scores plot of important urinary metabolite coefficients for separation.

Figure 3.

Receiver operating characteristic (ROC) curves of the logistic regression models for CRC versus healthy controls where (A) shows the results for metabolites only and (B) shows the results for metabolites and weight loss.

Table 2

Logistic regression based optimal model for CRC diagnosis (CRC vs healthy controls)

	Metabolites only		Metabolites $+$ clinical features
	$p$ -value	Odds	$p$ -value	Odds
Summary of each feature
Diacetylspermine	3.41 $\times$ 10 ${}^{-8}$	2.593	5.60 $\times$ 10 ${}^{-5}$	3.034
$p$ -Hydroxyhippuric acid	6.18 $\times$ 10 ${}^{-4}$	1.850	2.55 $\times$ 10 ${}^{-3}$	2.693
Glutamic acid	1.03 $\times$ 10 ${}^{-3}$	1.645	3.83 $\times$ 10 ${}^{-3}$	2.202
Hippuric acid	2.80 $\times$ 10 ${}^{-3}$	1.715	3.99 $\times$ 10 ${}^{-3}$	2.568
Weight loss [Y/N]	NA	NA	1.17 $\times$ 10 ${}^{-8}$	10.857
Model performance
AUC (95% CI)	87.77% (84.20%–91.33%)		89.72% (86.41%–93.02%)
Sensitivity (95% CI)	83.43% (77.51%–88.76%)		81.66% (76.33%–87.57%)
Specificity (95% CI)	79.38% (73.20%–85.05%)		86.59% (81.44%–91.24%)

NA- not applicable. CI indicates the confidence interval.

Figure 4.

(A) PLS-DA 2D-scores plot of CRC ( $n=$ 169) and healthy controls $+$ polyps ( $n=$ 244) using metabolite profiles only; (B) VIP scores plot of important urinary metabolite coefficients. Abbreviations: healthy controls $+$ polyps (HC $+$ P).

To determine if clinical features combined with metabolite profiles could improve the separation of the CRC patients and healthy controls, further evaluation was performed by PLS-DA. Figure 2A shows that separation between the CRC versus healthy control groups was achieved. Figure 2B shows the most important contributors to the group separation. As before, a permutation test was also performed to ensure that the separation was not due to chance (Fig. S2B). When clinical variables were included in the PLS-DA separation, the same four metabolites once again topped the list and included (in decreasing order of importance) N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid and glutamic acid (Fig. 2B). Among the clinical features that were investigated, weight loss was the most useful and ranked fourth in the variable importance in projection (VIP) plot (Fig. 2B) while pellet-like stool ranked much lower (12 ${}^{\text{th}})$ .

These findings suggest that a robust set of urinary metabolites (alone) exist that could serve as biomarkers in a high-performing CRC test among African subjects. Therefore, we performed logistic regression modelling to identify the best combination of urinary metabolites that could distinguish CRC patients from healthy controls. To find the most effective predictors, a LASSO feature selection algorithm was performed. The logistic regression models were first built and optimized using a discovery cohort of 169 CRC patients and 194 healthy controls (Table 1). Using only the metabolomics profile, the following optimum logistic regression model was generated that calculates the odds or likelihood (specifically the log odds) of a patient having CRC (P): $\log$ (P/(1 $-$ P)) $=-$ 0.114 $+$ 0.953 $\times$ [Diacetylspermine] $+$ 0.615 $\times$ [ $p$ -Hydroxyhippuric acid] $+$ 0.498 $\times$ [Glutamic acid] $+$ 0.539 $\times$ [Hippuric acid]. Note: the concentration of each named metabolite in the equation is the value that results after normalization of the LC-MS/MS-determined concentrations against each sample’s urinary creatinine level, followed by log transformation and auto-scaling. The exact scaling equations for these concentration adjustments are provided in Table S4. A receiver-operating characteristic (ROC) curve of the model with 95% confidence interval (CI) was built using this equation and is shown in Fig. 3A. This model was validated with a validation (hold-out) set that included 43 CRC patients and 31 healthy controls (Table 1). The AUC of the ROC curves for the discovery and validation sets were 87.8% and 83.4%, respectively (Fig. 3A) with the model yielding 83.4% sensitivity and 79.4%, specificity. Other details of the model are listed in Table 2.

When clinical variables were introduced into the logistic regression modelling, an optimum modified model to predict if a patient has CRC was generated: $\log$ (P/(1 $-$ P)) $=-$ 1.204 $+$ 0.737 $\times$ [Diacetylspermine] $+$ 0.594 $\times$ [ $p$ -Hydroxyhippuric acid] $+$ 0.466 $\times$ [Glutamic acid] $+$ 0.557 $\times$ [Hippuric acid] $+$ 1.763 $\times$ Weight loss. Here, the clinical feature of weight loss was introduced as a binary variable (Yes $=$ 1, No $=$ 0). Again, the numeric values of the named metabolites in the equation are the concentrations resulting from the normalization process described above. The ROC curves of the models for the discovery set and the validation set are both shown in Fig. 3B. Other details are listed in Table 2 as well. As shown in Fig. 3B and Table 2, the performance of the models for both sets were very close with AUCs of 89.7% and 89.2% for the training and validation sets, respectively. Thus, the addition of the clinical variable “weight loss” improved the model’s performance by 6–10%. Compared to the metabolite only models, the sensitivity of the model slightly decreased while the specificity increased (Table 2).

Table 3

Logistic regression based optimal model for CRC diagnosis (CRC vs healthy controls $+$ polyps)

	Metabolites only		Metabolites $+$ clinical features
	$p$ -value	Odds	$p$ -value	Odds
Summary of each feature
Diacetylspermine	1.14 $\times$ 10 ${}^{-8}$	3.359	2.98 $\times$ 10 ${}^{-5}$	2.771
$p$ -Hydroxyhippuric acid	7.09 $\times$ 10 ${}^{-4}$	2.396	1.79 $\times$ 10 ${}^{-3}$	2.520
Glutamic acid	4.48 $\times$ 10 ${}^{-3}$	1.979	2.22 $\times$ 10 ${}^{-2}$	1.909
Hippuric acid	1.20 $\times$ 10 ${}^{-4}$	2.642	1.11 $\times$ 10 ${}^{-3}$	2.617
Weight loss [Y/N]	NA	NA	3.52 $\times$ 10 ${}^{-9}$	10.247
Model performance
AUC (95% CI)	86.91% (83.29%–90.54%)		89.50% (86.38%–92.62%)
Sensitivity (95% CI)	81.07% (75.15%–86.98%)		79.88% (73.96%–85.80%)
Specificity (95% CI)	82.79% (78.28%–87.70%)		85.66% (81.15%–89.75%)

NA- not applicable. CI indicates the confidence interval

Figure 5.

(A) PLS-DA separation of CRC ( $n=$ 169) and normal ( $n=$ 244) using metabolite profiles plus clinical features; (B) VIP scores plot of important urinary metabolite coefficients for separation.

Figure 6.

ROC curves of the logistic regression models for CRC vs healthy controls $+$ polyps where (A) using only the metabolomics data and (B) includes both metabolomics and weight loss data.

3.3 Statistical analysis: CRC versus healthy controls including polyps

A total of 15% of study participants had colorectal polyps. To ensure that our models could distinguish CRC from polyps in symptomatic individuals, we assessed the model’s performance comparing CRC patients versus healthy controls plus those with polyps. The influence of including the clinical features (Figs 4A and 5A, respectively) resulted in similar separations in the PLS-DA plots. Combining healthy controls with individuals having polyps into a single “healthy” group led to a few more outliers when compared to healthy controls without polyps (Fig. 1A). Using urinary metabolic profiles only (Fig. 4B), the VIP plot yielded the same order of the top performing metabolites (N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid) for group separation. This was identical to what was observed when metabolites from healthy controls without polyps were compared to metabolites from CRC patients (Fig. 1B). Glutamic acid was still highly ranked (fifth compared to fourth, Fig. 1B), while aspartic acid became more important for separation, moving from eighth to fourth position (Figs 1B and 4B, respectively). When the metabolite profiles and clinical features were combined, weight loss remained the most effective clinical indicator (Fig. 5B) while stool appearance dropped to tenth position. N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid, and glutamic acid were listed among the top 5 metabolite factors with the highest overall coefficient factors. Other contributing metabolites (such as aspartic acid, kynurenine, 3-(3-hydroxyphenyl)-3-hydroxypropionic acid (HPHPA), and betaine) appear consistently in both VIP plots (Figs 1B and 5B). The $p$ -values of the permutation tests for each of the PLS-DAs described above were both smaller than 0.001, which demonstrated that the observed separations were not due to chance (Fig. S2C and S2D). The results of the PLS-DA analysis suggested that both the healthy group and the polyps group had similar urinary metabolic profiles, which were significantly different from patients with CRC.

Using only the urinary metabolite profiles, the optimal logistic regression model was generated using the combination of N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid, and glutamic acid (equation: $\log$ (P/(1 $-$ P)) $=-$ 0.474 $+$ 0.893 $\times$ [Diacetylspermine] $+$ 0.549 $\times$ [ $p$ -Hydroxyhippuric acid] $+$ 0.400 $\times$ [Glutamic acid] $+$ 0.639 $\times$ [Hippuric acid]). As before, the numeric values of the named metabolites in the equation are the concentrations resulting from the normalization procedure described above. Compared with the performance of the model shown in Fig. 3A (no clinical variable included), the AUC of this model’s training set dropped slightly from 87.8% to 86.9% while the AUC of the validation set remained the same (Fig. 6A). (Other details of the logistic regression modeling are shown in Table 3). After adding the clinical variables, the logistic regression model predicting the likelihood of having CRC was: $\log$ (P/(1 $-$ P)) $=-$ 1.516 $+$ 0.686 $\times$ Diacetylspermine $+$ 0.562 $\times$ $p$ -Hydroxyhippuric acid $+$ 0.345 $\times$ Glutamic acid $+$ 0.597 $\times$ Hippuric acid $+$ 1.736 $\times$ Weight loss. As before, the numeric values of the named metabolites in the equation are the concentrations resulting from the normalization procedure described earlier (See Table S4). The AUC of the ROC curve for this model using the discovery set was 89.5% (95% CI: 86.4%–92.6%; Fig. 6B). This result remained the same even when only healthy subjects were included. Validation of the model using the validation (hold-out) set returned a slightly higher AUC value at 92.1%. (All details of this model are included in Table 3). This model’s sensitivity and specificity diminished slightly but remained $>$ 80%.

4. Discussion

In this prospective study, urinary metabolomic profiles were used to build a clinically relevant diagnostic model to discriminate between symptomatic patients with CRC versus those with non-CRC conditions (e.g., inflammatory bowel disease, hemorrhoids or polyps). Urinary metabolite concentrations were analyzed via quantitative MS-based techniques and normalized to creatinine levels. Separate discovery and validation cohorts were used to prevent overtraining and any unintended bias in the results. Two different metabolite-only and two different metabolite $+$ weight loss models were developed and independently validated to identify CRC patients from healthy controls with or without polyps. While most of these models achieved AUCs $>$ 0.8–0.9, the model built with healthy controls plus those with polyps forming a single “healthy” group achieved the highest AUC in the validation set. With a diagnostic performance of $>$ 80% sensitivity and $>$ 80%specificity, our models perform sufficiently well to effectively screen and prioritize symptomatic patients at highest risk for CRC for colonoscopy To the best of our knowledge, this is the first CRC metabolomics study conducted in a sub-Saharan African LMIC.

Several metabolomic studies have identified urinary biomarkers for CRC [19, 20, 21, 22, 23, 24, 25, 26]. These studies were limited to populations in high income, northern climate countries with very different diet, lifestyle, colonic microflora, climate conditions and lifetime exposures than those from low or middle-income countries. This necessitated a detailed study of CRC urinary biomarkers focused on individuals from LMIC’s. We determined that N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid, and glutamic acid were the most important urinary metabolites to discriminate between CRC and non-CRC. Consistent with our results, other published studies have detected metabolite changes in response to CRC observing increased N ${}^{1}$ , N ${}^{12}$ -diacetylspermine [20, 21], decreased hippuric acid [22, 23, 24, 25], decreased $p$ -hydroxyhippuric acid [26], and increased glutamic acid [25]. A meta-analysis of case-controlled urinary targeted CRC studies identified 5 upregulated metabolites: N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, D-glucose, L-kynurenine, L-proline, and putrescine and 5 downregulated metabolites: creatinine, citric acid, indole-3-acetic acid, hippuric acid, and phenol [19]. This meta-analysis also identified N ${}^{1}$ , N ${}^{12}$ -diacetylspermine and hippuric acid as the more reliable urinary metabolites that are altered in CRC, which agree with our study. Many of remaining consensus compounds identified by the meta-analysis differed from those we identified. Presumably this was attributed to the differences in CRC trajectories and environments between the LMICs and higher income countries including Canada [20], United States [20] China [34] and Korea [23].

Our CRC diagnostic models were built using metabo-lite-only and metabolite plus clinical characteristics (such as weight loss, rectal bleeding and change in stool habits) to improve the assays specificity and sensitivity. Weight loss was the feature that contributed the most to our predictive models. Rectal bleeding lacked importance as a clinical variable likely because non-cancer related illnesses or conditions such as hemorrhoids and diverticulitis [17] can cause rectal bleeding. A CRC rectal bleeding symptom-screening model developed by Alatise et al. [17], using weight loss and change in bowel habits as the symptoms, found that 56% of patients with these three symptoms were identified with CRC Stages II to IV while 44% did not have CRC. Although change in bowel habits was an important characteristic in the symptom-screening model [17], this feature had a low predictive value when included with our logistic regression models. Instead, we found that the urinary metabolites N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid and glutamic acid combined with weight loss were much more important indicators of CRC in Nigerian patients.

Other CRC models built using urinary metabolites with and without clinical characteristics have been described for North American, Chinese and Korean patients. Deng et al. [20] developed a North American CRC diagnostic model using N ${}^{1}$ , N ${}^{12}$ -diacetylspermine and kynurenine as CRC biomarkers. They observed the biggest change in these biomarkers from CRC Stages 0-I with N ${}^{1}$ , N ${}^{12}$ -diacetylspermine levels increasing as the disease progressed. Kim et al. [23] studied urine NMR profiles in patients with colorectal neoplasia (advanced adenomas and CRCs at various stages) and found that taurine, alanine and 3-aminoisobutyrate could discriminate CRC patients from healthy controls. Other models have been described, identifying patients with or without colonic adenomatous polyps. Using NMR-based urinary metabolomics, Wang et al. [35] used 17 metabolites to predict Canadian patients with adenomatous polyps. They found that butyrate topped the list of the most important predictive metabolites. An earlier study using NMR-based urinary metabolomics that combined clinical characteristics acquired from patient questionnaires found 4 metabolites (methanol, trigonelline, acetone and tyrosine) and 4 clinical features (age, sex, smoking and fecal occult blood) predicted patients with adenomatous polyps who should be screened further by colonoscopy [36]. While this latter urine metabolomic screening test has greater sensitivity and specificity than the FIT tests, it would not be useful in screening symptomatic individuals in our Nigerian population since the urinary metabolomic profiles were similar between healthy controls and polyps.

We observed statistically significant changes in urinary N ${}^{1}$ , N ${}^{12}$ -diacetylspermine, hippuric acid, $p$ -hydroxyhippuric acid and glutamic acid concentrations in CRC patients compared to healthy controls. Hippuric acid has been reported to be decreased in several CRC biomarker studies [22, 23, 25]. While increased urinary hippuric acid correlates with healthy gut microbial diversity [37], decreased urinary hippuric acid has been negatively associated with blood pressure, non-alcoholic fatty liver disease, visceral fat mass and Crohn’s disease [38, 39, 40, 41]. The change in hippuric acid in all these diseases could result from alterations in the microbiome, which may be similar to the gut microbiome alterations of CRC patients. The association of increased levels of N ${}^{1}$ , N ${}^{12}$ -diacetylspermine with CRC, even in its earliest developmental stage of adenomatous polyps, has been reported previously [20, 21, 42]. However, N ${}^{1}$ , N ${}^{12}$ -diacetylspermine is not specific to CRC as it has been shown to be elevated in other malignancies (e.g., breast, prostate, testicular, and renal cancers) [43].

While the performance of our metabolite-only or metabolite $+$ weight loss for diagnosing CRC is encouraging, these models still need to be further validated on even larger, more diverse cohorts. This large-scale validation will begin in mid-2022. The inclusion of individuals with other diseases (i.e., tape worm or amebiasis) in a separate control group would also help determine whether these metabolite markers are specific to CRC or are predictors for a range of intestinal conditions. The inclusion of other clinical variables beyond weight loss, such as nutrition, place of residency, occupation, chronic and infectious disease history, and current medications could be used to assist in identifying those with CRC. This non-invasive urinary metabolomic strategy could be used in combination with the symptom-risk model [17] to screen for CRC in Nigerian (and potentially other African) patients.

While our metabolite or metabolite $+$ weight loss model could be easily converted to an MS-based diagnostic test, in resource poor countries such as Nigeria, the cost, time and expertise required to perform MS-based tests is far too prohibitive. However, we were fortunate to have discovered urinary metabolites in the Nigerian CRC patients that are abundant and water-soluble and for which simple colorimetric or fluorometric tests exist [44, 45, 46, 47, 48]. This suggests that the metabolite markers we have identified could be readily translated to a colorimetric POC device. Indeed, we have already developed colorimetric assays for these metabolites and have already constructed such a colorimetric POC device. This device could achieve our goal of offering a low-cost, fast, easy to use, non-invasive urinary test to detect and screen for CRC in symptomatic Nigerian patients.

Author contributions

Conception: D.S. Wishart, O.I Alatise, T.P. Kingham.

Methodology: L. Zhang, J. Zheng, D.S. Wishart.

Data collection: L. Zhang; J. Zheng, K.P. Ismond, S. MacKay, J. Constable, O.I. Alatise, T.P. Kingham, D.S. Wishart.

Interpretation or analysis of data: L. Zhang, J. Zheng.

Preparation of the manuscript: L. Zhang, K.P. Ismond, M. LeVatte, D.S. Wishart.

Revision for important intellectual content: L. Zhang; K.P. Ismond, S. MacKay, M. LeVatte, J. Constable, O.I. Alatise, T.P. Kingham, D.S. Wishart.

Supervision: D.S. Wishart.

Supplementary data

The supplementary files are available to download from http://dx.doi.org/10.3233/CBM-220034.

sj-doc-1-cbm-10.3233_CBM-220034.doc - Supplemental material

Supplemental material, sj-doc-1-cbm-10.3233_CBM-220034.doc

Footnotes

Acknowledgments

We thank the staff at The Metabolomics Innovation Centre (TMIC) for the assistance in processing the samples and operating the metabolomics equipment used for this study. This work was funded by the National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health (NIH) (grant number UG3EB024965). Financial support from Genome Canada, the Canada Foundation for Innovation (CFI) and the Canadian Institutes of Health Research (CIHR) is gratefully acknowledged.

References

and Xu

, Global colorectal cancer burden in 2020 and projections to 2040, Transl Oncol 14 (2021), 101174. doi: 10.1016/J.TRANON.2021.101174.

Douaiher

Ravipati

Grams

Chowdhury

Alatise

and A. C, Colorectal cancer-global burden, trends, and geographical variations, J Surg Oncol 115 (2017), 619–630. doi: 10.1002/JSO.24578.

Irabor

Arowolo

and Afolabi

, Colon and rectal cancer in Ibadan, Nigeria: an update, Color Dis 12 (2010), e43–e49. doi: 10.1111/J.1463-1318.2009.01928.X.

Arnold

Sierra

Laversanne

Soerjomataram

Jemal

and Bray

, Global patterns and trends in colorectal cancer incidence and mortality, Gut 66 (2017), 683–691. doi: 10.1136/GUTJNL-2015-310912.

Torre

Siegel

Ward

and Jemal

, Global cancer incidence and mortality rates and trends – An update, CancerEpidemiol Biomarkers Prev 25 (2016), 16–27. doi: 10.1158/1055-9965.EPI-15-0578.

Irabor

, Emergence of colorectal cancer in West Africa: Accepting the inevitable, Niger Med J 58 (2017), 11–16. doi: 10.4103/0300-1652.234076.

Davidson

K.W.

Barry

M.J.

Mangione

C.M.

Cabana

Caughey

A.B.

Davis

E.M.

Donahue

K.E.

Doubeni

C.A.

Krist

A.H.

Kubik

Ogedegbe

Owens

D.K.

Pbert

Silverstein

Stevermer

Tseng

C.W.

and Wong

J.B.

, Screening for colorectal cancer: US preventive services task force recommendation statement, JAMA 325 (2021), 1965–1977. doi: 10.1001/jama.2021.6238.

Ibrahim

Anjorin

Afolayan

and Badmos

, Morphology of colorectal carcinoma among Nigerians: a 30-year review, Niger J Clin Pract 14 (2011), 432–435. doi: 10.4103/1119-3077.91750.

Saluja

Alatise

Adewale

Misholy

Chou

Gonen

Weiser

and Kingham

, A comparison of colorectal cancer in Nigerian and North American patients: is the cancer biology different?, Surgery 156 (2014), 305–310. doi: 10.1016/J.SURG.2014.03.036.

10.

Jiang

Yuan

Shen

Tuo

and Xiang

, Global pattern and trends of colorectal cancer survival: a systematic review of population-based registration data, Cancer Biol Med 19 (2021), 175–186. doi: 10.20892/j.issn.2095-3941.2020.0634.

11.

Lee

Liles

Bent

Levin

and Corley

, Accuracy of fecal immunochemical tests for colorectal cancer: systematic review and meta-analysis, Ann Intern Med 160 (2014), 171–181. doi: 10.7326/M13-1484.

12.

Imperiale

Ransohoff

Itzkowitz

Levin

Lavin

Lidgard

Ahlquist

and Berger

, Multitarget stool DNA testing for colorectal-cancer screening, N Engl J Med 370 (2014), 1287–1297. doi: 10.1056/NEJMOA1311194.

13.

Warren

Xiong

Bunker

Vaughn

Furtado

Roberts

Fang

Samowitz

and Heichman

, Septin 9 methylated DNA is a sensitive and specific blood test for colorectal cancer, BMC Med 9 (2011), 133. doi: 10.1186/1741-7015-9-133.

14.

Products

|

Metabolomic Technologies Inc., (n.d.). https://www.mtidx.com/our-work/products (accessed September 1, 2021).

15.

Carethers

, Fecal DNA testing for colorectal cancer screening, Annu Rev Med 71 (2020), 59–69. doi: 10.1146/ANNUREV-MED-103018-123125.

16.

Adler

Geiger

Keil

Bias

Schatz

DeVos

Dhein

Zimmermann

Tauber

and Wiedenmann

, Improving compliance to colorectal cancer screening using blood and stool based tests in patients refusing screening colonoscopy in Germany, BMC Gastroenterol 14 (2014), 183. doi: 10.1186/1471-230X-14-183.

17.

Alatise

Ayandipo

Adeyeye

Seier

Komolafe

Bojuwoye

Afuwape

Zauber

Omisore

Olatoke

Akere

Famurewa

Gonen

Irabor

and Kingham

, A symptom-based model to predict colorectal cancer in low-resource countries: Results from a prospective study of patients at high risk for colorectal cancer, Cancer 124 (2018), 2766–2773. doi: 10.1002/CNCR.31399.

18.

Laiyemo

Brawley

Irabor

Boutall

Ramesar

and Madiba

, Towards colorectal cancer control in Africa, Int J Cancer 138 (2016), 1033-1034. doi: 10.1002/IJC.29843.

19.

Mallafré-Muro

Llambrich

Cumeras

Pardo

Brezmes

Marco

and Gumà

, Comprehensive volatilome and metabolome signatures of colorectal cancer in urine: A systematic review and meta-analysis, Cancers (Basel) 13 (2021), 2534. doi: 10.3390/CANCERS13112534.

20.

Deng

Ismond

Liu

Constable

Wang

Alatise

O.I.

Weiser

M.R.

Kingham

T.P.

and Chang

, Urinary metabolomics to identify a unique biomarker panel for detecting colorectal cancer: A multicenter study, Cancer Epidemiol Biomarkers Prev 28 (2019), 1283–1291. doi: 10.1158/1055-9965.EPI-18-1291.

21.

Nakajima

Katsumata

Kuwabara

Soya

Enomoto

Ishizaki

Tsuchida

Mori

Hiwatari

Soga

Tomita

and Sugimoto

, Urinary polyamine biomarker panels with machine-learning differentiated colorectal cancers, benign disease, and healthy controls, Int J Mol Sci 19 (2018), 756. doi: 10.3390/IJMS19030756.

22.

Cheng

Xie

Chen

Qiu

Zou

Zheng

Tan

, Feng

Dong

Zhao

Zhang

and Jia

, Distinct urinary metabolic profile of human colorectal cancer, J Proteome Res 11 (2012), 1354–1363. doi: 10.1021/PR201001A.

23.

Kim

Kwon

Nam

Kim

Park

and Kim

Y.-H.

, Urine-NMR metabolomics for screening of advanced colorectal adenoma and early stage colorectal cancer, Sci Rep 9 (2019), 4786. doi: 10.1038/s41598-019-41216-y.

24.

Wang

Lin

Liang

Huang

Liu

and Yang

,NMR-based metabolomic techniques identify potential urinary biomarkers for early colorectal cancer detection, Oncotarget 8 (2017), 105819-105831. doi: 10.18632/ONCOTARGET.22402.

25.

Qiu

Cai

Chen

Liu

Zhao

, Cai

and Jia

, Urinary metabonomic study on colorectal cancer, J Proteome Res 9 (2010), 1627–1634. doi: 10.1021/PR901081Y.

26.

Liesenfeld

Habermann

Toth

Owen

Frei

Böhm

Schrotz-King

Klika

K.K.

and Ulrich

, Changes in urinary metabolic profiles of colorectal cancer patients enrolled in a prospective cohort study (ColoCare), Metabolomics 11 (2015), 998–1012. doi: 10.1007/S11306-014-0758-3.

27.

Deng

Chang

Foshaug

Eisner

Tso

Wishart

and Fedorak

, Development and validation of a high-throughput mass spectrometry based urine metabolomic test for the detection of colonic adenomatous polyps, Metabolites 7 (2017), 32. doi: 10.3390/METABO7030032.

28.

Khaniani

Lipfert

Bhattacharyya

Perez Pineiro

Zheng

and Wishart

, A simple and convenient synthesis of unlabeled and 13 C-labeled 3-(3-hydroxyphenyl)-3-hydroxy- propionic acid and its quantification in human urine samples, Metabolites 8 (2018), 80. doi: 10.3390/METABO8040080.

29.

Zheng

Zhang

Johnson

Mandal

and Wishart

D.S.

, Comprehensive targeted metabolomic assay for urine analysis, Anal Chem 92 (2020), 10627–10634. doi: 10.1021/ACS.ANALCHEM.0C01682.

30.

R: The R Project for Statistical Computing, (n.d.). https://www.r-project.org/ (accessed September 14, 2021).

31.

Wishart

, Computational approaches to metabolomics in: Bioinformatics Methods in Clinical Research. Methods in Molecular Biology, vol. 593. Matthiesen

, (ed), Humana Press, 2010, pp. 283–313. doi: 10.1007/978-1-60327-194-3_14.

32.

Chong

Soufan

Caraus

Bourque

Wishart

and Xia

, MetaboAnalyst 4.0: towards more transparent and integrative metabolomics analysis, Nucleic Acids Res 46 (2018), W486–W494. doi: 10.1093/NAR/GKY310.

33.

Robin

Turck

Hainard

Tiberti

Lisacek

Sanchez

J.-C.

and Müller

, pROC: an open-source package for R and S+ to analyze and compare ROC curves, BMC Bioinformatics 12 (2011), 77. doi: 10.1186/1471-2105-12-77.

34.

Deng

Fang

Tso

Sun

Foshaug

Krahn

Zhang

Yan

Chang

Zhang

and Fedorak

, Clinical validation of a novel urine-based metabolomic test for the detection of colonic polyps on Chinese population, Int J Colorectal Dis 32 (2017), 741–743. doi: 10.1007/S00384-016-2729-9.

35.

Wang

Tso

Wong

Sadowski

and Fedorak

, Development and validation of a highly sensitive urine-based test to identify patients with colonic adenomatous polyps, Clin Transl Gastroenterol 5 (2014), e54. doi: 10.1038/CTG.2014.2.

36.

Eisner

Greiner

Tso

Wang

and Fedorak

R.N.

, A machine-learned predictor of colonic polyps based on urinary metabolomics, Biomed Res Int 2013 (2013), 303982. doi: 10.1155/2013/303982.

37.

Brial

Chilloux

Nielsen

Vieira-Silva

Falony

Andrikopoulos

Olanipekun

Hoyles

Djouadi

Neves

Rodriguez-Martinez

Mouawad

Pons

Forslund

Le-chatelier

Lay

Nicholson

Hansen

Hyötyläinen

Clément

Oresic

Bork

Ehrlich

Raes

Pedersen

Gauguier

and Dumas

M.-E.

, Human and preclinical studies of the host-gut microbiome co-metabolite hippurate as a marker and mediator of metabolic health, Gut 70 (2021), 2105–2114. doi: 10.1136/GUTJNL-2020-323314.

38.

Dumas

M.-E.

Barton

Toye

Cloarec

Blancher

Rothwell

Fearnside

Tatoud

Blanc

Lindon

Mitchell

Holmes

McCarthy

Scott

Gauguier

and Nicholson

, Metabolic profiling reveals a contribution of gut microbiota to fatty liver phenotype in insulin-resistant mice, Proc Natl Acad Sci USA 103 (2006), 12511–12516. doi: 10.1073/PNAS.0601056103.

39.

Holmes

Loo

Stamler

Bictash

Yap

Chan

Ebbels

De Iorio

Brown

Veselkov

Daviglus

Kesteloot

Ueshima

Zhao

Nicholson

and Elliott

, Human metabolic phenotype diversity and its association with diet and blood pressure, Nature 453 (2008), 396–400. doi: 10.1038/NATURE06882.

40.

Pallister

Jackson

M.A.

Martin

T.C.

Glastonbury

C.A.

Jennings

Beaumont

Mohney

R.P.

Small

K.S.

MacGregor

Steves

C.J.

Cassidy

Spector

T.D.

Menni

and Valdes

A.M.

, Untangling the relationship between diet and visceral fat mass through blood metabolomics and gutmicrobiome profiling, Int J Obes 41 (2017), 1106–1113. doi: 10.1038/ijo.2017.70.

41.

Williams

Cox

Walker

Cobbold

Taylor-Robinson

Marshall

and Orchard

, Differences in gut microbial metabolism are responsible for reduced hippurate synthesis in Crohn’s disease, BMC Gastroenterol 10 (2010), 108. doi: 10.1186/1471-230X-10-108.

42.

Hiramatsu

Takahashi

Yamaguchi

Matsumoto

Miyamoto

Tanaka

Tamamori

Imajo

Kawaguchi

Toi

Mori

and Kawakita

, N(1)N(12)-Diacetylspermine as a sensitive and specific novel marker for early- and late-stage colorectal and breast cancers, Clin Cancer Res 11 (2005), 2986–2990. doi: 10.1158/1078-0432.CCR-04-2275.

43.

Sugimoto

Hiramatsu

Kamei

Kinoshita

Hoshino

Iwasaki

and Kawakita

, Significance of urinary N1,N8-diacetylspermidine and N1,N12-diacetylspermine as indicators of neoplastic diseases, J Cancer Res Clin Oncol 121 (1995), 317–319. doi: 10.1007/BF01209602.

44.

Zinalibdin

M.R.

and Yacob

A.R.

, Detection of hippuric acid: A glue solvent metabolite, using a mobile test kit, Arab J Chem 6 (2013), 115–120. doi: 10.1016/j.arabjc.2010.09.029.

45.

Boomsma

and Schalekamp

M.A.D.H.

, Evaluation of a test kit for the rapid and simple colorimetric measurement of angiotensin I-converting enzyme in serum, J Clin Chem Clin Biochem 21 (1983), 845–849. doi: 10.1515/CCLM.1983.21.12.845.

46.

Koçdor

M.A.

Astarcioğlu

and Fadiloğlu

, Serum tumor necrosis factor-alpha, glutamate and lactate changes in two different stages of mechanical intestinal obstruction, Turkish J Gastroenterol 14 (2003), 115–120. https://pubmedncbi.nlm.nih.gov/14614638/.

47.

Bakke

Shimoji

and Kajiyama

, N1, N12-diacetylsper- mine, oxidase from Debaryomyces hansenii T-42: purification, characterization, molecular cloning and gene expression, Biochim Biophys Acta 1774 (2007), 1395–1401. doi: 10.1016/j.bbapap.2007.08.010.

48.

Moriya

S.S.

Hiramatsu

Kimura

Matsumoto

and Kawakita

, Construction of an immunochromatographic determination system for N1, N12-diacetylspermine, J Clin Lab Anal 28 (2014), 452–460. doi: 10.1002/JCLA.21709.

Identification of urinary biomarkers of colorectal cancer: Towards the development of a colorectal screening test in limited resource settings

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Materials and methods

2.1 Study design

2.2 Study population

2.3 Sample and clinical data collection

2.4 Chemicals, reagents and materials for mass spectrometry-based metabolomic assays

2.5 Stock solutions, internal standard mixture, and calibration curve standards for metabolomic assays

2.6 Sample preparation and liquid chromatography and direct injection mass spectrometry for metabolomic assays

2.7 Statistical analysis

3. Results

3.1 Study population and statistical data processing

Table 1 Summary of grouping of samples

4. Discussion

Author contributions

Supplementary data

sj-doc-1-cbm-10.3233_CBM-220034.doc - Supplemental material

Footnotes

Acknowledgments

References

Table 1
Summary of grouping of samples