Rapid Diagnosis and Staging of Colorectal Cancer
Rapid Diagnosis and Staging of Colorectal Cancer
The 44 study participants included 26 males (59%) and 18 females (41%), with a median age of 72 years (range, 56–87). These patients were consecutively enrolled to the study where possible to minimize selection bias. During the study time frame, a total of 64 potentially eligible consecutive patients were identified. Reasons for study omission for 20 of these patients were as follows: (1) enrolment to FOxTROT study, precluding opening of surgical specimen (n = 7); (2) patient decision not to take part in study (n = 4); (3) no pathologist available outside of standard working hours to evaluate and sample resection specimen (n = 4); and (4) inability to sample rectal cancer resection specimen due to potential for compromising circumferential resection margin assessment (n = 5). This information is summarized in the REMARK flow diagram (Figure 1).
All surgical resection specimens were confirmed as adenocarcinoma on pathological assessment. Tumors were located in the transverse colon (n = 3; 7%), descending colon/proximal sigmoid colon (n = 5; 11%), ascending colon (n = 6; 14%), rectosigmoid (n = 8; 18%), cecum (n = 10; 23%), and rectum (n = 12; 27%) in order of increasing prevalence. A summary of demographic and clinicopathological characteristics is provided in Table 2 .
A total of 171 CPMG spectral profiles were generated from tissue samples obtained from Ct (n = 88) and healthy appearing mucosa 5 cm from the tumor margin Hm (n = 83). Representative annotated HR-MAS NMR spectra for Ct and Hm samples are shown in Figure 2A and B respectively. Spectra were imported into SIMCA and MATLAB software packages for analysis using unsupervised principal component analysis (figures not shown) and supervised OPLS-DA multivariate approaches. The OPLS-DA model generated revealed clear separation between Ct and Hm tissue samples based on metabolic profiles (RY = 0.94; QY = 0.72; Fig. 3A). The corresponding ROC curve analysis constructed using 7-fold cross-validated predicted Y values further confirmed the robustness of the OPLS-DA model in distinguishing between cancerous and noncancerous tissue samples, with an AUC of 0.98 (Fig. 3B). The loadings plot (Fig. 4) revealed the key metabolic discriminators to be lactate, taurine, isoglutamine, glycine, and scyllo-inositol (found in significantly greater concentrations in Ct than in Hm) and lipid/triglyceride species (found in significantly greater concentrations in Hm than Ct). These results are in keeping with previous findings reported by our group.Table 3 summarizes the observed metabolic differences between Ct and Hm tissue samples. Biochemical species are numerically assigned in the table for cross-referencing with representative spectra shown in Figure 2.
(Enlarge Image)
Figure 2.
Annotated HR-MAS NMR spectral profiles from center of tumor (Ct) (A) and healthy appearing mucosa 5 cm from tumor margin (Hm) (B). 1, lipids/triglycerides (-CH3); 2, lipids/triglycerides (-CH2-); 3, Lipids/triglycerides (C=C-CH2); 4, Lipids/triglycerides (=CH-); 5, Lactate (-CH3); 6, Lactate (-CH-); 7, Valine (-CH3); 8, Alanine (-CH3); 9, Leucine (β-CH2γ -CH); 10, Acetate (-CH3); 11, Isoglutamine (β-CH2); 12, Creatine (-CH3); 13, Creatine (-CH2-); 14, Glycerophosphoryl-choline [N(CH3)3]; 15, Taurine (-CH2-NH); 16, Taurine (CH2SO3); 17, Scyllo-inositol (-O-H); 18, Glycine (-CH2); 19, β-glucose (1-CH); 20, α-glucose (1-CH); § residual water signal.
(Enlarge Image)
Figure 3.
A, Scatter plot demonstrating separation between Ct (green) and Hm (blue) tissue samples based on metabolite-driven OPLS-DA model. B, Corresponding ROC curve confirming the robustness of the OPLS-DA model at distinguishing between Ct and Hm samples (AUC = 0.98).
(Enlarge Image)
Figure 4.
Loadings plot allowing identification of the main metabolites responsible for discrimination between Ct and Hm tissue classes in generated OPLS-DA model. ANOVA indicates analysis of variance.
A total of 88 CPMG spectral profiles were generated from center of colon tumor(n = 49) and center of rectal tumor (CtR; n = 39) tissue samples. The "colon" class comprised samples obtained from cecum, ascending colon, transverse colon, descending colon, and proximal sigmoid/sigmoid colon. Tumor tissue was classified as "rectal" if retrieved from rectosigmoid or rectal cancers. Figure 5A shows the OPLS-DA model scores generated using this classification approach (RY = 0.94; QY = 0.33) with corresponding 7-fold cross-validated ROC curve analysis shown in Figure 5B (AUC = 0.86). The model allows metabolite-based distinction to be made between tumor samples according to location, with colon cancer tissue seen to harbor significantly greater concentrations of acetate and arginine than rectal cancers (Figure 6A). Rectal tumors in comparison were found to have increased concentrations of lactate (Figure 6A). The lactate peak at 1.33 is difficult to visualize in Figure 6A because the strong lipid presence overwhelms this signal. A statistical total correlation spectroscopy analysis driven by the apex of the -CH3 signal for lactate at 4.1 was performed to clarify this further, and permitted clear identification of the main lactate peak (Figure 6B). To strengthen the validity of the identified metabolic differences between colon and rectal cancer tissue, we performed additional OPLS-DA and ROC curve analysis after exclusion of cases having received neoadjuvant (NA) chemo- and/or radiotherapy. Supplementary Figure 1A, available at http://links.lww.com/SLA/A413 shows the OPLS-DA model scores generated after exclusion of these samples (RY = 0.80; QY = 0.21) and reveals clear distinction between colon and rectal cancer tissue samples with corresponding 7-fold cross-validated ROC curve analysis shown in Supplementary Figure 1B (AUC = 0.83), available at http://links.lww.com/SLA/A413.
(Enlarge Image)
Figure 5.
A, OPLS-DA scatter plot demonstrating metabolite-based separation between CtC and CtR tissue samples. B, Corresponding ROC curve analysis confirms that this model is good at distinguishing between classes (AUC = 0.86). CtC indicates center of colon tumor; CtR, center of rectal tumor.
(Enlarge Image)
Figure 6.
A, Loadings plot revealing the main metabolites responsible for discrimination between classes in "CtC vs. CtR" OPLSDA model. B, STOCSY analysis driven from apex of lactate signal at 4.1ppm demonstrating main lactate peak at 1.33ppm that appears hidden within a strong lipid signal in Figure A. STOCSY indicates statistical total correlation spectroscopy; CtC, center of colon tumor; CtR, center of rectal tumor; ANOVA, analysis of variance
We compared CPMG spectral profiles from Ct tissue samples obtained from T1/2 tumors (12 patients; 23 spectra), T3 tumors (20 patients; 41 spectra), and T4 tumors (12 patients; 24 spectra) to determine whether metabolic characteristics could be identified with which to determine local tumor extent (T-stage). Figure 7A shows the OPLS-DA model scores plot generated from this classification approach, with visible clustering of samples according to T-stage (T1/2 in blue, T3 in green, T4 in red). In Figure 7B the misclassification table shows that this metabolite-driven means of determining local tumor stage was able to correctly assign samples as T1/2, T3, or T4 in 91%, 90%, and 75% of cases, respectively. Corresponding 7-fold cross-validated ROC curve analysis (see Supplementary Figure 2, available at http://links.lww.com/SLA/A414) further confirmed the robustness of this model for classification of tumor tissue samples according to T-stage, based on metabolic profiles.
(Enlarge Image)
Figure 7.
A, OPLS-DA scatter plot demonstrating clustering of metabolic activity among Ct tissue samples taken from T1–T2 (blue), T3 (green), and T4 (red) cancers. B, Misclassification table demonstrating robustness of model at correctly assigning samples according to T-stage.
Biochemical differences between the different T-stage tumors were explored by performing OPLS-DA analysis between groups (T1/2 vs T3, T3 vs T4). This approach revealed specific metabolic phenotypes associated with each stage of local tumor development (Figure 8). We found T3 tumor tissue to have significantly greater concentrations of lipids/triglycerides (P < 0.05) and acetate (P < 0.05) than T1/2 tumors. In contrast, T1/2 tumors contained greater concentrations of glycerophosphoryl-choline (P < 0.05) than T3 tumors. In comparison, T4 tumors had a unique metabotype of their own, marked by a general depleted biochemical state, with significantly reduced levels of lipids/triglycerides, acetate, and succinate relative to T3 cancers. These metabolic fluxes are summarized in Figure 8.
(Enlarge Image)
Figure 8.
Distinct HR-MAS NMR spectroscopy–based metabolic phenotypes according to T-stage.
Results
Patient Demographics and Tumor Characteristics
The 44 study participants included 26 males (59%) and 18 females (41%), with a median age of 72 years (range, 56–87). These patients were consecutively enrolled to the study where possible to minimize selection bias. During the study time frame, a total of 64 potentially eligible consecutive patients were identified. Reasons for study omission for 20 of these patients were as follows: (1) enrolment to FOxTROT study, precluding opening of surgical specimen (n = 7); (2) patient decision not to take part in study (n = 4); (3) no pathologist available outside of standard working hours to evaluate and sample resection specimen (n = 4); and (4) inability to sample rectal cancer resection specimen due to potential for compromising circumferential resection margin assessment (n = 5). This information is summarized in the REMARK flow diagram (Figure 1).
All surgical resection specimens were confirmed as adenocarcinoma on pathological assessment. Tumors were located in the transverse colon (n = 3; 7%), descending colon/proximal sigmoid colon (n = 5; 11%), ascending colon (n = 6; 14%), rectosigmoid (n = 8; 18%), cecum (n = 10; 23%), and rectum (n = 12; 27%) in order of increasing prevalence. A summary of demographic and clinicopathological characteristics is provided in Table 2 .
Determining Metabolic Differences Between Cancerous and Adjacent Healthy Colorectal Mucosa Using HR-MAS NMR Spectroscopy
A total of 171 CPMG spectral profiles were generated from tissue samples obtained from Ct (n = 88) and healthy appearing mucosa 5 cm from the tumor margin Hm (n = 83). Representative annotated HR-MAS NMR spectra for Ct and Hm samples are shown in Figure 2A and B respectively. Spectra were imported into SIMCA and MATLAB software packages for analysis using unsupervised principal component analysis (figures not shown) and supervised OPLS-DA multivariate approaches. The OPLS-DA model generated revealed clear separation between Ct and Hm tissue samples based on metabolic profiles (RY = 0.94; QY = 0.72; Fig. 3A). The corresponding ROC curve analysis constructed using 7-fold cross-validated predicted Y values further confirmed the robustness of the OPLS-DA model in distinguishing between cancerous and noncancerous tissue samples, with an AUC of 0.98 (Fig. 3B). The loadings plot (Fig. 4) revealed the key metabolic discriminators to be lactate, taurine, isoglutamine, glycine, and scyllo-inositol (found in significantly greater concentrations in Ct than in Hm) and lipid/triglyceride species (found in significantly greater concentrations in Hm than Ct). These results are in keeping with previous findings reported by our group.Table 3 summarizes the observed metabolic differences between Ct and Hm tissue samples. Biochemical species are numerically assigned in the table for cross-referencing with representative spectra shown in Figure 2.
(Enlarge Image)
Figure 2.
Annotated HR-MAS NMR spectral profiles from center of tumor (Ct) (A) and healthy appearing mucosa 5 cm from tumor margin (Hm) (B). 1, lipids/triglycerides (-CH3); 2, lipids/triglycerides (-CH2-); 3, Lipids/triglycerides (C=C-CH2); 4, Lipids/triglycerides (=CH-); 5, Lactate (-CH3); 6, Lactate (-CH-); 7, Valine (-CH3); 8, Alanine (-CH3); 9, Leucine (β-CH2γ -CH); 10, Acetate (-CH3); 11, Isoglutamine (β-CH2); 12, Creatine (-CH3); 13, Creatine (-CH2-); 14, Glycerophosphoryl-choline [N(CH3)3]; 15, Taurine (-CH2-NH); 16, Taurine (CH2SO3); 17, Scyllo-inositol (-O-H); 18, Glycine (-CH2); 19, β-glucose (1-CH); 20, α-glucose (1-CH); § residual water signal.
(Enlarge Image)
Figure 3.
A, Scatter plot demonstrating separation between Ct (green) and Hm (blue) tissue samples based on metabolite-driven OPLS-DA model. B, Corresponding ROC curve confirming the robustness of the OPLS-DA model at distinguishing between Ct and Hm samples (AUC = 0.98).
(Enlarge Image)
Figure 4.
Loadings plot allowing identification of the main metabolites responsible for discrimination between Ct and Hm tissue classes in generated OPLS-DA model. ANOVA indicates analysis of variance.
Determining Metabolic Differences Between Colon and Rectal Cancer Tissue Using HR-MAS NMR Spectroscopy
A total of 88 CPMG spectral profiles were generated from center of colon tumor(n = 49) and center of rectal tumor (CtR; n = 39) tissue samples. The "colon" class comprised samples obtained from cecum, ascending colon, transverse colon, descending colon, and proximal sigmoid/sigmoid colon. Tumor tissue was classified as "rectal" if retrieved from rectosigmoid or rectal cancers. Figure 5A shows the OPLS-DA model scores generated using this classification approach (RY = 0.94; QY = 0.33) with corresponding 7-fold cross-validated ROC curve analysis shown in Figure 5B (AUC = 0.86). The model allows metabolite-based distinction to be made between tumor samples according to location, with colon cancer tissue seen to harbor significantly greater concentrations of acetate and arginine than rectal cancers (Figure 6A). Rectal tumors in comparison were found to have increased concentrations of lactate (Figure 6A). The lactate peak at 1.33 is difficult to visualize in Figure 6A because the strong lipid presence overwhelms this signal. A statistical total correlation spectroscopy analysis driven by the apex of the -CH3 signal for lactate at 4.1 was performed to clarify this further, and permitted clear identification of the main lactate peak (Figure 6B). To strengthen the validity of the identified metabolic differences between colon and rectal cancer tissue, we performed additional OPLS-DA and ROC curve analysis after exclusion of cases having received neoadjuvant (NA) chemo- and/or radiotherapy. Supplementary Figure 1A, available at http://links.lww.com/SLA/A413 shows the OPLS-DA model scores generated after exclusion of these samples (RY = 0.80; QY = 0.21) and reveals clear distinction between colon and rectal cancer tissue samples with corresponding 7-fold cross-validated ROC curve analysis shown in Supplementary Figure 1B (AUC = 0.83), available at http://links.lww.com/SLA/A413.
(Enlarge Image)
Figure 5.
A, OPLS-DA scatter plot demonstrating metabolite-based separation between CtC and CtR tissue samples. B, Corresponding ROC curve analysis confirms that this model is good at distinguishing between classes (AUC = 0.86). CtC indicates center of colon tumor; CtR, center of rectal tumor.
(Enlarge Image)
Figure 6.
A, Loadings plot revealing the main metabolites responsible for discrimination between classes in "CtC vs. CtR" OPLSDA model. B, STOCSY analysis driven from apex of lactate signal at 4.1ppm demonstrating main lactate peak at 1.33ppm that appears hidden within a strong lipid signal in Figure A. STOCSY indicates statistical total correlation spectroscopy; CtC, center of colon tumor; CtR, center of rectal tumor; ANOVA, analysis of variance
Determining Metabolic Differences Between T1/T2, T3, and T4 Colorectal Tumors Using HR-MAS NMR Spectroscopy
We compared CPMG spectral profiles from Ct tissue samples obtained from T1/2 tumors (12 patients; 23 spectra), T3 tumors (20 patients; 41 spectra), and T4 tumors (12 patients; 24 spectra) to determine whether metabolic characteristics could be identified with which to determine local tumor extent (T-stage). Figure 7A shows the OPLS-DA model scores plot generated from this classification approach, with visible clustering of samples according to T-stage (T1/2 in blue, T3 in green, T4 in red). In Figure 7B the misclassification table shows that this metabolite-driven means of determining local tumor stage was able to correctly assign samples as T1/2, T3, or T4 in 91%, 90%, and 75% of cases, respectively. Corresponding 7-fold cross-validated ROC curve analysis (see Supplementary Figure 2, available at http://links.lww.com/SLA/A414) further confirmed the robustness of this model for classification of tumor tissue samples according to T-stage, based on metabolic profiles.
(Enlarge Image)
Figure 7.
A, OPLS-DA scatter plot demonstrating clustering of metabolic activity among Ct tissue samples taken from T1–T2 (blue), T3 (green), and T4 (red) cancers. B, Misclassification table demonstrating robustness of model at correctly assigning samples according to T-stage.
Biochemical differences between the different T-stage tumors were explored by performing OPLS-DA analysis between groups (T1/2 vs T3, T3 vs T4). This approach revealed specific metabolic phenotypes associated with each stage of local tumor development (Figure 8). We found T3 tumor tissue to have significantly greater concentrations of lipids/triglycerides (P < 0.05) and acetate (P < 0.05) than T1/2 tumors. In contrast, T1/2 tumors contained greater concentrations of glycerophosphoryl-choline (P < 0.05) than T3 tumors. In comparison, T4 tumors had a unique metabotype of their own, marked by a general depleted biochemical state, with significantly reduced levels of lipids/triglycerides, acetate, and succinate relative to T3 cancers. These metabolic fluxes are summarized in Figure 8.
(Enlarge Image)
Figure 8.
Distinct HR-MAS NMR spectroscopy–based metabolic phenotypes according to T-stage.
