Objective The mitochondrial apoptosis pathway is controlled by an interaction of multiple BCL-2 family proteins, and plays a key role in tumour progression and therapy responses. We assessed the prognostic potential of an experimentally validated, mathematical model of BCL-2 protein interactions (DR_MOMP) in patients with stage III colorectal cancer (CRC).
Design Absolute protein levels of BCL-2 family proteins were determined in primary CRC tumours collected from n=128 resected and chemotherapy-treated patients with stage III CRC. We applied DR_MOMP to categorise patients as high or low risk based on model outputs, and compared model outputs with known prognostic factors (T-stage, N-stage, lymphovascular invasion). DR_MOMP signatures were validated on protein of n=156 patients with CRC from the Cancer Genome Atlas (TCGA) project.
Results High-risk stage III patients identified by DR_MOMP had an approximately fivefold increased risk of death compared with patients identified as low risk (HR 5.2, 95% CI 1.4 to 17.9, p=0.02). The DR_MOMP signature ranked highest among all molecular and pathological features analysed. The prognostic signature was validated in the TCGA colon adenocarcinoma (COAD) cohort (HR 4.2, 95% CI 1.1 to 15.6, p=0.04). DR_MOMP also further stratified patients identified by supervised gene expression risk scores into low-risk and high-risk categories. BCL-2-dependent signalling critically contributed to treatment responses in consensus molecular subtypes 1 and 3, linking for the first time specific molecular subtypes to apoptosis signalling.
Conclusions DR_MOMP delivers a system-based biomarker with significant potential as a prognostic tool for stage III CRC that significantly improves established histopathological risk factors.
- BCL-2 FAMILY PROTEINS
- COLORECTAL CANCER
- CLINICAL DECISION MAKING
- ADJUVANT TREATMENT
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
The TNM staging system is the primary mean of risk estimation and treatment decision for patients with colorectal cancer (CRC).
Better risk stratification tools are required for the clinical management of CRC.
What are the new findings?
DR_MOMP DR_MOMP identifies high-risk patients with stage III CRC with low overall survival and is superior to clinically established pathological risk factors.
DR_MOMP can further stratify patients identified by supervised gene expression risk scores into low-risk and high-risk categories.
BCL-2-dependent signalling contributes to treatment responses in consensus molecular subtypes 1 and 3, with important implications for the future use of BCL-2 antagonists.
How might it impact on clinical practice in the foreseeable future?
As prognostic tool DR_MOMP has the potential to significantly improve established risk scores for patients with CRC.
Surgical resection is the primary treatment for patients with stages I–III colorectal cancer (CRC). Tumour staging largely dictates whether patients receive adjuvant chemotherapy. In stage III disease, adjuvant treatment with 5-fluorouracil (5-FU)-based chemotherapy with the addition of oxaliplatin (FOLFOX/XELOX regimen) is estimated to result in an approximately 20% improvement in 5-year overall survival (OS).1 However, the majority of patients will relapse or develop distant metastases within 5 years following surgery.2 ,3 Also, adjuvant therapy has not been shown to have a substantial benefit among stage II patients.4–6 Hence, there is an urgent need to improve the selection of high-risk patients with stage II and III CRC.
Deregulation in the apoptotic signalling pathway is a characteristic of carcinogenesis and failure of chemotherapy.7 BCL-2 family proteins are master regulators of apoptosis, and act on mitochondria to induce, or inhibit, the process of mitochondrial outer membrane permeabilisation (MOMP).8 This process is viewed as a key step in chemotherapy-inducing caspase-dependent and caspase-independent cell death.9 Due to the complexity of the BCL-2-controlled mitochondrial apoptosis pathway with multiple interaction partners and signalling redundancies, no individual BCL-2 family protein, or group of proteins, has been shown to reliably predict patient responsiveness to adjuvant chemotherapy in CRC.10 We hypothesised that dynamic computational models which take into account network topology, biochemical interaction data and protein quantities are required to fully understand the regulation of apoptosis at a systems level. We recently established DR_MOMP, an ordinary differential equation-based systems model that addresses the dynamic regulation of MOMP by BCL-2 proteins.11 This model has been validated experimentally in colon cancer cells and calculates the ‘stress dose’ required to induce MOMP as a readout of the sensitivity of cancer cells to genotoxic chemotherapy.11 The aim of this study was to comprehensively test whether DR_MOMP can be employed as a prognostic tool in stage III CRC.
Stage III cohort
Data were obtained from an independent cohort of n=134 patients with stage III CRC treated with 5-FU-based chemotherapy, specifically FOLFOX or XELOX. Formalin-fixed paraffin-embedded (FFPE) primary tumour samples were collected and stored from three centres during 2005–2013: Beaumont Hospital (RCSI, Ireland), Queen's University Belfast (UK) and Paris Descartes University (France). Pathological stage was determined using the American Joint Committee on Cancer (AJCC) tumour, node, metastases (TNM) staging system. Clinical follow-up was obtained through medical record review. All centres provided ethical approval for this study. Informed consent was obtained from all participants. Patients' clinical characteristics are provided in online supplementary table S1.
We accessed data from The Cancer Genome Atlas (TCGA) (23 July 2015) on patient follow-up as well as level 3 protein from n=331 (reverse phase protein array, RPPA) and mRNA expression values for n=457 (mRNA SeqV2) patients with stage II–IV Colon adenocarcinoma (COAD). Protein expression data were available for BAK, BAX, BCL2 and BCL(X)L. Patient clinical characteristics are provided in online supplementary table S2. Stage I patients (n=50), patients with residual tumour or unknown status (n=90) and patients with a follow-up time of 0 days (n=64) were excluded from the RPPA dataset.
Protein extraction from FFPE tumour samples
Proteins from our stage III cohort were extracted as previously described.12 Briefly, all tumour samples were histologically confirmed by a pathologist to have at least 70% tumour cell content. Three to twenty serial 20 μm sections of tumour were adhered to uncharged slides using nuclease-free water. One additional adjacent 5 μm section was stained for H&E. The tumour hotspot region was outlined by an expert histopathologist. Each slide from the block was overlaid on the H&E-stained slide and oriented according to the features of the section. The area surrounding the tumour-dense target region was removed using a sterile razor blade. After deparaffinisation using xylene/ethanol, the tissue samples were incubated with an extraction buffer (20 mM Tris buffer (pH 9) containing 2% sodium dodecyl sulfate (SDS) and protease inhibitors) for 2 hours at 80°C.
Reverse phase protein array
RPPAs for the FFPE tumour samples from our stage III cohort were carried out as previously described.12 Protein extraction lysates were normalised to a 1.5 μg/μL as assessed by a bicinchoninic acid assay (DC Protein Assay—Bio-Rad, California, USA). Three parts of cell lysates were mixed with one part SDS buffer (40% Glycerol, 8% SDS, 0.25 M Tris-HCL, pH 6.8 plus Bond-Breaker TCEP Solution (Pierce Biotechnology, Illinois, USA) at 1/10th of the volume) and boiled. Lysates were manually diluted in fourfold serial dilutions with lysis buffer. A QArray 2 arrayer (Molecular Device, UK) created 378 sample arrays on Oncyte Avid nitrocellulose-coated slides (Grace Bio-Labs, Oregon, USA). The slides were stored with desiccant (Drierite, Ohio, USA) at −20°C prior to immunostaining. Immunostaining was performed on an automated slide stainer (Dako Link 48—Dako, California, USA) according to the manufacturer's instructions (CSA kit—Dako). Each slide was incubated with a single primary antibody (see online supplementary table S3) at room temperature for 30 min. All antibodies used were listed in the Standard Antibody List 298 and validated by the MD Anderson Cancer Center as well as internally (Bak antibody).
Secondary antibodies were goat anti-rabbit IgG (1:5000) (Vector Laboratories, California, USA) or rabbit anti-mouse IgG (1:10) (Dako). Dako secondary antibodies were used as a starting point for amplification via horseradish peroxidase-mediated biotinyl tyramide with chromogenic detection (diaminobenzidine) according to the manufacturer's instructions (Dako).
Scanned TIFF images of slides were analysed using MicroVigene software V.5.1 (VigeneTech, Massachusetts, USA) to generate spot signal intensities.13 Instead of generating multiple linear regression curves for data quantification over each series of serial dilutions, the RPPA module of MicroVigene uses a four-parameter logistic-log model (‘SuperCurve’ algorithm14 with all spots within one array employed to form a sigmoid antigen-binding kinetic curve. The spots were normalised to β-actin to compensate variances in protein loading.
To calculate the absolute protein concentrations of the RPPA data of our stage III cohort, protein expression levels were normalised to known protein concentrations in HeLa cells (1109, 519, 239, 110 and 83 nM for BAK, BAX, BCL2, BCL(X)L and MCL1, respectively.11 HeLa cells were spotted in parallel with the patient sample onto the RPPA.
A MassARRAY system (Sequenom) was used to detect somatic point mutations of KRAS. Protocols are available on request.
Calculation of TCGA protein profiles
To calculate the absolute protein concentrations based on the RPPA and mRNA SeqV2 Data from the TCGA database, linear regression was used to normalise the expression levels to the 25th and 75th percentile of a fresh-frozen CRC cohort.11 We used 67, 139, 162, 2 and 4 nM and 580, 488, 967, 76 and 29 nM for the 25th and 75th percentiles for BAK, BAX, BCL2, BCL(X)L and MCL1, respectively. The Matlab functions polyfit and polyval were used for linear fitting. As no expression levels were available for MCL1 in the RPPA dataset, MCL1 concentrations were set to the mean value of 23.1 nM found in fresh frozen (FF) samples of patients with stage II and III CRC.11
To calculate the sensitivity of patients' cancer cells to undergo apoptosis, the mathematical model DR_MOMP was used.11 DR_MOMP was executed with Matlab R2007b (V.18.104.22.1682; The MathWorks, USA). To normalise the model predicted stress dose required to induce MOMP (η, eta), we calculated the z-score of η (zη, zeta) without extreme outliers and used this value as risk score. Extreme outliers were defined as samples whose η were greater or less than three times the IQR plus/minus the first and third quartiles, respectively. Patients with zη>0 were labelled as high-risk, and those with zη≤0 were labelled as low-risk.
Computation of risk scores
To determine the Almac, Genomic Health risk scores, the gene expression from the RNASeqV2 genome analyser (GA) and HiSeq dataset were accessed (23 July 2015). The Almac risk score was approximated based on the online supplementary methods from Kennedy et al15 by means of gene expression data of the RNASeqV2 dataset from the TCGA COAD cohort. Due to limitations of the RNASeqV2 dataset, only 453 unique genes instead of the original 634 genes from the original Affymetrix scanner were used. Patients with favourable and unfavourable outcome were classified according to the cut-off point <0.465.15 The Genomic Health score was approximated based on the methods from Park et al.16 ,17 Patients with favourable, intermediate and unfavourable outcome were classified according to the cut-off points <30, 30–40 and ≥41, respectively.16 ,17 To merge HiSeq and GA datasets, linear regression was used to transform the HiSeq dataset scores to the GA dataset scores. Consensus molecular subtypes (CMS) for the TCGA COAD cohort were accessed from the online supplementary information provided by Guinney et al.18
Survival analyses were performed using the ‘survival’ package (V.2.38–3) in R (V.3.2.2, The R Foundation). We used Kaplan-Meier estimates to compare differences between survival curves. Statistically significant differences between OS and disease-free survival (DFS) curves were determined using log-rank tests. We computed HRs and 95% CIs using unadjusted and multivariate Cox proportional hazards models to estimate the relative risks of recurrence and death associated with high-risk versus low-risk DR_MOMP risk scores. Statistical significance was determined using Wald tests (p<0.05).
Quantification of BCL-2 protein levels
We determined BCL-2 protein profiles in primary tumour samples obtained from patients with CRC acquired from Beaumont Hospital (RCSI, Ireland), Queen's University Belfast (UK) and Paris Descartes University (France). Patient characteristics are summarised in table 1. The cohort included 134 patients with stage III CRC treated with FOLFOX/XELOX in the adjuvant setting. Among those patients, the 5-year survival rate was 83.0% (95% CI 74.5% to 88.9%) and the median follow-up time was 65.1 months (95% CI 59.2 to 70.8). In 37 patients, the regimen was discontinued or altered due to side effects. Discontinuation of the regimen did not significantly increase the mortality risk in this cohort (see online supplementary table S4).
Protein amounts of BAK, BAX, BCL2, BCL(X)L and MCL1 were quantified in lysates of FFPE primary tumour samples by RPPAs. To calculate absolute BAK, BAX, BCL2, BCL(X)L and MCL1 protein concentrations, we normalised to known absolute protein concentrations in HeLa cells spotted in parallel onto the array (figure 1A, B).11 The proteins were heterogeneously expressed, with higher concentrations of proapoptotic BAK and BAX, compared with the anti-apoptotic BCL-2 proteins. We observed a median concentration of 509, 208, 201, 63 and 15 nM for BAK, BAX, BCL2, BCL(X)L and MCL1, respectively. BCL(X)L levels above the mean were strongly associated with unfavourable clinical outcome of the patients (HR 7.2, 95% CI 2.4 to 21.5, Wald p<0.001; online supplementary table S5). Other BCL2 family protein levels were not significantly associated with unfavourable clinical outcome.
We employed DR_MOMP to calculate the stress dose required to induce MOMP (η, eta; figure 1C) by means of the BCL-2 protein profiles. We modelled reversible protein binding based on mass action kinetics among proapoptotic and anti-apoptotic proteins BAK, BAX, BIM, BCL2, BCL(X)L, MCL1, PUMA, NOXA and VDAC2. BIM and PUMA were modelled to activate BAK and BAX which subsequently bind to homodimers and higher oligomers. Pores were defined as BAK and BAX protein complexes with six or more proteins. MOMP was considered to occur once 10% of total BAK and BAX were associated in pores. Protein profiles were used as initial protein levels in each patient. The DR_MOMP score is the calculated amount of stress producing the minimal required amount of BIM, PUMA and NOXA to lead to MOMP. Detailed modelled interactions and kinetics can be found in the online supplementary methods of Lindner et al.11
We determined a median value of 121 nM. We also performed a data normalisation step by calculating the z-score of η (zη, figure 1D). This was performed to overcome potential differences in tissue preparation and tissue fixation, and to adopt the methodology to other types of protein and gene expression analysis.
DR_MOMP identifies high-risk patients with stage III CRC with low OS and is superior to clinically established pathological risk factors
We examined whether DR_MOMP is a prognostic tool in stage III CRC. Using DR_MOMP, we found a higher stress dose in stage III patients that had poorer OS. We observed statistically significant differences for OS curves when comparing those who were high- and low-risk (p<0.001; figure 2A). In unadjusted analyses, high-risk patients had an approximately fivefold increased risk of death (HR 5.2, 95% CI 1.8 to 15.7, p<0.01) compared with low-risk patients. Similar results were obtained when analysing DFS (HR 1.8, 95% CI 1.0 to 3.5, p=0.06; online supplementary figure S1).
In this cohort of patients with CRC, we also compared DR_MOMP model outputs with known prognostic factors such as T-stage, N-stage and lymphovascular invasion. In univariate analyses (see online supplementary table S6), DR_MOMP (p<0.01), the tumour extent (p=0.3) and lymphovascular invasion (p=0.02) were significantly related to the OS, but none of the other analysed factors. Of note, the BCL-2 systems signature ranked highest among all molecular and pathological features analysed.
In a multivariate analysis (see online supplementary table S6), DR_MOMP was the only factor associated with mortality risk and was associated with an approximately fivefold increased risk of death compared with low-risk patients (HR 4.9, 95% CI 1.4 to 17.9, p=0.02; figure 2B). Although there were no statistically significant associations for other factors in this multivariate analysis, the univariate analysis suggested a strong influence of regional lymph node involvement and lymphovascular invasion on OS and DFS (HR>2.0). In exploratory analyses, we therefore evaluated associations between DR_MOMP and OS after stratifying by subgroups for these factors. Among patients where lymphovascular invasion was absent, we observed a difference between OS curves when comparing high-risk and low-risk patients (p=0.02, figure 2C), despite a high overall 5-year survival rate of 93.9% (95% CI 82.3% to 98.0%) among those patients. Among patients where lymphovascular invasion was present, we also found a difference between OS curves when comparing high-risk and low-risk patients (p=0.04, figure 2C). Among patients with high regional lymph node involvement (N2, >3 nodes), we observed a significant difference when comparing high-risk and low-risk patients (p<0.01, figure 2D). DR_MOMP also stratified patients by tumour location (figure 3), tumour extent and KRAS mutation status (see online supplementary figure S2). Collectively, these findings suggested that DR_MOMP delivers a system-based biomarker that significantly improves established histopathological risk factors.
Validation of DR_MOMP in an external colon cancer cohort
To validate our findings in an external cohort, we used publicly available protein expression data of 331 patients from the TCGA COAD cohort (see online supplementary table S2). To calculate absolute protein levels, we used linear regression to normalise the TCGA protein levels to 25th and 75th percentiles of the protein levels from our previous cohort (see online supplementary figure S3A).11 From 331 patients with available protein expression data, we excluded patients with residual tumour or unknown status (n=90, online supplementary figure S3B), stage I patients (n=50, online supplementary figure S3C), and patients with a follow-up time of 0 days (n=64). Among the remaining 156 patients (table 1), the 5-year survival rate was 66.4% (95% CI 43.8% to 81.7%) with a median follow-up time among censored participants of 8.6 months (95% CI 6.7 to 11.0). In contrast to our stage III cohort, no protein of the BCL2 family of proteins was associated with unfavourable outcome in the TCGA cohort (see online supplementary table S5).
We did not observe a statistical difference in DR_MOMP predictor zη between stage II, III and IV patients (Kruskal-Wallis H test p=0.76, figure 4A), and there was no interaction between stage and DR_MOMP stress dose (interaction p=0.99). We did not find a significant difference in the OS in stage II and III patients individually possibly due to the small sample size (see online supplementary figure S4). However, when comparing high-risk and low-risk patients among the different stages (II–IV), we observed statistically significant differences in OS curves (p=0.03; figure 4B). High-risk patients had an approximately fourfold increased risk of death (HR 4.1, 95% CI 1.1 to 15.6, p=0.04) compared with low-risk patients. We further included the age, sex, stage, microsatellite stable and instable (MSS/MSI) status and tumour location in univariate and multivariate analyses (see online supplementary table S7). The results of these analyses yielded only DR_MOMP and age as significant variables. In the adjusted model, we found that high-risk patients had a more than eightfold increased risk of death compared with low-risk patients (HR 8.4, 95% CI 1.2 to 57.7, p=0.03).
DR_MOMP stratifies patients identified by CMS or supervised gene expression risk scores
The relationship between DR_MOMP model outputs and survival outcomes could differ within subpopulations of patients in which resistance to chemotherapy is not caused by a deregulation of apoptosis. To explore this assumption, we interrogated the TCGA COAD cohort to examine whether this relationship differed by the CMS for CRC established by Guinney et al.18
We did not find that the predictors zη were significantly different among the CMS (figure 5A, Kruskal-Wallis H test p=0.24). However, the median stress doses of CMS1 (0.14) and CMS3 (0.13) were >0, while the medians of CMS2 (–0.18) and CMS3 (–0.28) were <0 (Wilcoxon rank-sum test, p=0.06). Interestingly, among a subgroup of patients with CMS 1 and 3, we observed a difference between OS curves when comparing those with high-risk and low-risk model outputs (p=0.03; figure 5B). This was not observed among those with CMS 2 and 4 (p=0.94), suggesting that apoptotic signatures were particularly important for patient outcome in CMS 1 and 3. The survival of low-risk and high-risk patients in the individual CMS is depicted in online supplementary figure S5.
Since CMS1 is associated with MSI,18 we were interested whether we would observe the same finding in MSI patients. The predictors zη in high MSI (MSI-H) patients were significantly higher compared with low MSI (MSI-L; Dunn’s test p=0.01) and MSS (figure 5C, Dunn’s Test p<0.001, Kruskal-Wallis H test p<0.01). Among a subgroup of patients with MSS and MSI-L, we observed a difference between OS curves when comparing those with high-risk and low-risk model outputs (p=0.04; figure 5D). We did not observe a significant difference in OS among those with MSI-H (p=0.23).
Finally, we were interested how DR_MOMP performed within other established, supervised risk scores, in particular whether it could further stratify patients identified with unfavourable outcome by these scores. We approximated Almac15 and Genomic Health16 ,17 risk scores, based on the RNA SeqV2 gene expression data of the TCGA project. Neither estimates of the risk scores were associated with increased mortality in our TCGA cohort (see online supplementary table S8). Subsequently, we further analysed patients who were notionally classified with unfavourable outcome by the approximated Genomic Health risk score, to test if DR_MOMP can stratify these patients into low- and high-risk. We observed a significant difference between OS curves when patients were further stratified into low and high risk by DR_MOMP (p=0.02; HR 5.1, 95% CI 1.1 to 24.6, p=0.04; figure 6A). Among patients classified with unfavourable outcome by the approximated Almac score, we also observed a significant difference between OS curves when patients were further stratified by DR_MOMP (p<0.01; HR 11.4, 95% CI 1.4 to 92.6, p=0.02; figure 6B). DR_MOMP's prognostic performance was similar to its performance in the univariate analysis, after adjusting for the Almac and Genomic Health risk score (HR 4.6, 95% CI 1.2 to 17.8, p=0.03; online supplementary table S8). There was no interaction between the DR_MOMP and the Almac or Genomic Health risk score (interaction p values=0.92 and 0.99, respectively).
In this comprehensive analysis of patients with CRC, we have shown that a decreased sensitivity of cancer cells to undergo mitochondrial apoptosis was highly prognostic of adverse clinical outcomes. We demonstrate that as a clinical decision tool, DR_MOMP is effective at stratifying patients into low-risk and high-risk groups across different AJCC TNM stages, and may significantly improve established pathological risk factors or gene expression risk scores. DR_MOMP delivers a single, patient-specific clinical value that is able to inform the pathologist and oncologist about the likelihood of risk of recurrence and death of patients with CRC. We also found that CMS 1 and 3 are specifically linked to defective apoptosis signalling in CRC, demonstrating the biological relevance and potential clinical benefits of molecular tumour subtyping.
The relevance of KRAS or other somatic mutations as predictive or prognostic biomarkers in stage II and III CRC is still controversial. Tumours with proficient mismatch repair, KRAS or BRAF (KRAS/BRAF) mutations have been associated with a high risk of recurrence and lower survival rates;19 ,20 however, these markers have not been shown to predict response to 5-FU/oxaliplatin.21 For risk stratification of patients, multiparametric gene expression assays are increasingly used. Such gene expression-based assays complement prognostic information22 but were shown to only marginally improve prediction of recurrence and death based on established risk factors such as TNM staging.23 In this study, we demonstrate that as a clinical decision tool, DR_MOMP is effective at stratifying patients into low-risk and high-risk groups across different TNM stages and histopathological risk factors. Indeed, multivariate analysis in our stage 3 cohort indicated that the BCL-2 systems signature ranked highest among all risk factors analysed. DR_MOMP also has the potential to be employed together with established gene expression signatures, such as the risk score of Almac15 and Genomic Health,16 ,17 to significantly improve the stratification of patients into low-risk and high-risk groups.
Implementation of DR_MOMP into clinical histopathological settings requires protein profiling of five BCL-2 proteins, for which routine assessment methods exist such as multiplex ELISA, Mesoscale platforms or Taqman protein PCR assays. BH3-peptide profiling or exposure of isolated cancer cells to BCL-2 antagonists was shown to identify patients with low DFS and to predict drug response in multiple myeloma, myelogenous and lymphoblastic leukaemia, and ovarian cancer24 ,25 demonstrating the predictive power of the BCL-2 signalling network in the setting of cancer therapy. However, living cancer cells are required for such tests to identify if a patient's cancer cell is primed for cell death and sensitive to chemotherapy. DR_MOMP has the key advantage that it can be employed on FF or FFPE tissues, and is employable in the routine diagnostics and histopathology setting. Importantly, by identifying a mean ‘risk score’ value, we are able to stratify low-risk and high-risk patients using tissue material derived from FF or FFPE specimen, and are able to adopt the methodology to different types of protein level analysis.
CRC is a heterogeneous disease. Recently, an international CRC Subtyping Consortium established a robust consensus for molecular subtypes (CMS 1–4).18 CMS 4 resulted in worse OS and DFS compared with CMS 1–3.18 However, little is known whether CMS is predictive for patients' response to 5-FU-based chemotherapy or what CMS means in terms of the underlying tumour biology. To study the influence of molecular subtypes on apoptosis signalling, we analysed DR_MOMP output among CMS.18 When comparing high-risk and low-risk groups, we observed a difference in OS among those with CMS 1 (‘immune’) and 3 (‘metabolic’), though not among those with CMS 2 (‘WNT/MYC signalling’) and 4 (‘mesenchymal/stromal infiltration’), suggesting that impairment of apoptosis in CMS 2 and 4 does not contribute to disease progression or response to therapy. Of note, these findings may also have implications for the use of BCL-2 antagonists as novel adjuvant therapy in the setting of metastatic or advanced CRC. Our data indicate that patients with CMS 2 and 4 might not benefit from treatment with BCL-2 antagonists, but that those with CMS 1 and 3 and a high DR_MOMP model predictor should be stratified for future clinical trials.
The authors sincerely thank the stage III patients who participated in this study. The authors thank Dr Triona Ni Chonghaile for critically reading this manuscript. The results published here are in part based on data generated by the TCGA Research Network: http://cancergenome.nih.gov which we also gratefully acknowledge.
Contributors AUL, MaS, NM, MiS and AR prepared figures. AUL, MaS, NM, MiS, AR and JHMP analysed and interpreted the data. AUL, AR and JHMP wrote and edited the manuscript. CM, MC, SC, NM, ST, ROB, ROB, LF, RW, PGJ and MST performed acquisition and processing of data. SCB. DAMN, EWK, BRH, PLP, SVS and JHMP supervised the project. All authors read, reviewed and approved the final manuscript.
Funding This work has received funding from a Science Foundation Ireland Investigator Award to JHMP (13/IA/1881) and the European Union's Seventh Framework Programme (FP7 APO-DECIDE) under contract No. 306021.
Competing interests AUL and JHMP filed a patent application at the EPO (Appl.No. EP20120166187 and EP20130728324), USPTO (Appl.No. 14/397697) and WIPO (Appl.No. PCT/EP2013/059051).
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.