Portal de Periódicos da CAPES

Predictive modeling in colorectal cancer: time to move beyond consensus molecular subtypes

2019; Elsevier BV; Volume: 30; Issue: 11 Linguagem: Inglês

10.1093/annonc/mdz412

1569-8041

Anita Sveen, Chiara Cremolini, Rodrigo Dienstmann,

Hepatocellular Carcinoma Treatment and Prognosis

The consensus molecular subtypes (CMS) represent an attempt to define a ‘consensus’ for transcriptomic subtyping of colorectal cancer (CRC), given major inconsistencies among previously reported classification frameworks [1.Guinney J. Dienstmann R. Wang X. et al.The consensus molecular subtypes of colorectal cancer.Nat Med. 2015; 21: 1350-1356Crossref PubMed Scopus (2652) Google Scholar]. The marked interconnectivity between independent gene expression classifiers gave rise to the four CMS groups, which not only reflect cancer cell phenotypes but also microenvironment features present in bulk tumor tissue samples. This rational approach to catalog the intrinsic biological drivers of distinct CRC subtypes and the reproducibility achieved across multiple validation cohorts marked an important step for their clinical translation. The strong prognostic implications of CMS groups in both early-stage and metastatic settings reinforced optimism in the CRC research community that they might also represent biomarkers for therapy selection, inspired by the transcriptomic classification of breast cancer—which has guided drug development and design of clinical trials with targeted agents. Importantly, the drivers of breast cancer molecular subtypes (i.e. hormone-receptor signaling, HER2 activation and cancer cell proliferation) are critical determinants of response to hormonal, targeted and chemotherapy drugs. Correspondingly, the CMS groups have weaker, but still significant, enrichments for EGFR and VEGF signaling, both of which are targets of approved therapies in the metastatic setting. The biological hypothesis is that patients with epithelial-canonical CMS2 tumors have increased benefit with anti-EGFR therapies, while mesenchymal-stromal CMS4 tumors are particularly sensitive to antiangiogenic agents. The benefits of first-line treatment of RAS wild-type metastatic CRC with either cetuximab or bevacizumab are still debated, with discrepant results of two large randomized clinical trials detailed in Table 1. In the CALGB/SWOG 80405 trial (here referred to as the CALGB study), the combination of cetuximab with a doublet of chemotherapy was associated with increased response rates (RR) relative to chemotherapy plus bevacizumab, but no improvements in progression-free survival (PFS) or overall survival (OS) [2.Venook A.P. Niedzwiecki D. Lenz H.J. et al.Effect of first-line chemotherapy combined with cetuximab or bevacizumab on overall survival in patients with KRAS wild-type advanced or metastatic colorectal cancer: a randomized clinical trial.JAMA. 2017; 317: 2392-2401Crossref PubMed Scopus (523) Google Scholar]. In the FIRE3 AIO KRK-0306 trial (the FIRE3 study), FOLFIRI plus cetuximab significantly increased OS when compared with FOLFIRI plus bevacizumab, but no differences in RR or PFS were found between the two treatment arms [3.Heinemann V. von Weikersthal L.F. Decker T. et al.FOLFIRI plus cetuximab versus FOLFIRI plus bevacizumab as first-line treatment for patients with metastatic colorectal cancer (FIRE3): a randomised, open-label, phase 3 trial.Lancet Oncol. 2014; 15: 1065-1075Abstract Full Text Full Text PDF PubMed Scopus (1278) Google Scholar]. However, both studies showed that cetuximab was more effective than bevacizumab for patients with a left-sided primary tumor location, whereas bevacizumab was preferable for right-sided tumors [4.Arnold D. Lueza B. Douillard J.Y. et al.Prognostic and predictive value of primary tumour side in patients with RAS wild-type metastatic colorectal cancer treated with chemotherapy and EGFR directed antibodies in six randomized trials.Ann Oncol. 2017; 28: 1713-1729Abstract Full Text Full Text PDF PubMed Scopus (526) Google Scholar]. To help explain these findings, CALGB and FIRE3 investigators carried out retrospective and exploratory analyses of the CMS groups [5.Lenz H.J. Ou F.S. Venook A.P. et al.Impact of consensus molecular subtype on survival in patients with metastatic colorectal cancer: results from CALGB/SWOG 80405 (Alliance).J Clin Oncol. 2019; 37: 1876-1885Crossref PubMed Scopus (125) Google Scholar, 6.Stintzing S. Wirapati P. Lenz H.-J. et al.Consensus molecular subgroups (CMS) of colorectal cancer (CRC) and 1st-line efficacy of FOLFIRI plus cetuximab or bevacizumab in the FIRE3 (AIO KRK-0306) trial.Ann Oncol. 2019; Abstract Full Text Full Text PDF PubMed Scopus (87) Google Scholar]; the FIRE3 data are reported in this issue of Annals. Both studies found poor survival for patients with CMS1 tumors, intermediate survival for the CMS4 group, and good outcomes associated with CMS2 across the two treatment arms (Table 1). However, the CALGB study reported significantly worse PFS and OS with cetuximab compared with bevacizumab in the CMS1 population, and improved OS in the cetuximab arm among CMS2 classified cases. In the FIRE3 study, benefit from cetuximab in terms of PFS and OS was only evident in the CMS4 population. Of note, it is unknown whether the CMS associations found in the CALGB study remained significant after adjusting for tumor sidedness, considering that CMS1 and CMS2 are enriched in CRCs with a right- and left-sided primary tumor location, respectively. The FIRE3 study reported a non-significant interaction between CMS and treatment effect in multivariable models, including tumor sidedness. The discordant results between the two clinical trials contrast with other studies investigating the predictive value of CMS groups. In retrospective analyses of the AGITG MAX trial, the PFS rate was significantly increased when adding bevacizumab to chemotherapy among patients with epithelial CMS2/CMS3 tumors (not stratified by RAS/BRAF status) [7.Mooi J.K. Wirapati P. Asher R. et al.The prognostic impact of consensus molecular subtypes (CMS) and its predictive effects for bevacizumab benefit in metastatic colorectal cancer: molecular analysis of the AGITG MAX clinical trial.Ann Oncol. 2018; 29: 2240-2246Abstract Full Text Full Text PDF PubMed Scopus (88) Google Scholar], while the addition of cetuximab to chemotherapy plus bevacizumab increased the OS rate in the CMS2/CMS3 population of the CAIRO trial (RAS/BRAF wild-type) [8.Trinh A. Trumpi K. De Sousa E Melo F. et al.Practical and robust identification of molecular subtypes in colorectal cancer by immunohistochemistry.Clin Cancer Res. 2017; 23: 387-398Crossref PubMed Scopus (107) Google Scholar].Table 1Summary of the results from analysis of CMS groups in relation to survival and treatment outcomes in CALGB/SWOG 80405 and FIRE3 studiesCALGB/SWOG 80405FIRE3KRAS codons 12 and 13 wt (n-=-581)RAS wt (n-=-315)Patient’s and study characteristicsFirst-line therapy (randomization)FOLFOX (75%)/FOLFIRI (25%)+cetuximab or bevacizumabFOLFIRI+cetuximab or bevacizumabAge (median)60.164Sex (male)64%69%ECOG (0, 1, 2)59%, 41%, 0%53%, 45%, 2%Liver metastasis only (yes)36%35%Primary tumor location (right, left)36%, 64%22.5%, 77.5%Primary resected (yes)88%89%Adjuvant chemotherapy (yes)13%18%Diagnosis (synchronous, metachronous)79%, 21%77.5%, 22.2%Microsatellite instability (MSI)12.4 %2.3 %CMS classificationCMS1CMS2CMS3CMS4CMS1CMS2CMS3CMS4Distribution, overall104 (18%)242 (42%)68 (12%)167 (29%)46 (15%)130 (41%)36 (11%)103 (33%)Distribution by primary tumor location (% right- and left-sided)a61%, 32%17%, 75%29%, 69%28%, 65%41%, 59%15%, 85%19%, 81%24%, 76%Classified samples87.6% of the totally 663 with available gene expression data100% of cancers with available gene expression dataSample sourceFFPE samples from primary tumorsFFPE samples from primary tumorsGene expression technologyNanoString and nCounterAlmac Xcel microarrayClassifier genesCustom gene panelNot reportedClassification algorithmCustom multinomial logistic regressionSSP in CMSclassifier R packageMedian PFS in months [95% CI]7.1 [5.7–8.6]13.4 [12.8–15.4]8.7 [7.2–9.8]11.0 [9.7–12.0]7.8 [6.3–9.3]11.8 [10.9–12.6]9.7 [7.4–12.0]9.7 [8.9-10.5]P < 0.001P = 0.008Median OS in months [95% CI]15 [11.7–22.4]40.3 [36.1–43.1]24.3 [16.4–29.0]31.4 [26.3–36.9]14.8 [8.3–21.4]31.9 [25.7–38.1]20.8 [8.6–33.1]23.8 [20.6-27.0]P < 0.001P < 0.001Median PFS (cetuximab versus bevacizumab)HR-=-2.28, P<0.001HR-=-0.91, P-=-0.52HR-=-1.10, P-=-0.74HR-=-0.87, P-=-0.44HR-=-1.05, P-=-0.87HR-=-1.04, P-=-0.82HR-=-0.82, P-=-0.59HR-=-0.67, P-=-0.048P (interaction)-=-0.003P (interaction)-=-0.07Median OS (cetuximab versus bevacizumab)HR-=-2.34, P<0.001HR-=-0.62, P-=-0.005HR-=-1.09, P-=-0.76HR-=-1.04, P-=-0.83HR-=-0.83, P-=-0.57HR-=-0.86, P-=-0.44HR-=-0.57, P-=-0.15HR-=-0.57, P-=-0.008P (interaction) < 0.001P (interaction)-=-0.05aPrimary tumors in transverse colon not included among right- or left-sided tumors in CALGB. Open table in a new tab aPrimary tumors in transverse colon not included among right- or left-sided tumors in CALGB. There are important differences between the CALGB and FIRE3 studies that help interpret these results, as summarized in Table 1. Firstly, the patient cohorts are diverse. FIRE3 included only RAS wild-type metastatic CRCs, while the CALGB cohort included KRAS wild-type and had a higher prevalence of right-sided and microsatellite instable (MSI) tumors. Of particular relevance, the majority of CMS1 tumors were right-sided in CALGB and left-sided in FIRE3. Furthermore, the predictive effect in disfavor of cetuximab for CMS1 tumors in CALGB can also be linked to MSI, as a separate retrospective analysis in this study showed that patients with MSI tumors had a survival advantage when receiving bevacizumab rather than cetuximab [9.Innocenti F. Ou F.S. Qu X. et al.Mutational analysis of patients with colorectal cancer in CALGB/SWOG 80405 identifies new roles of microsatellite instability and tumor mutational burden for patient outcome.J Clin Oncol. 2019; 37: 1217-1227Crossref PubMed Scopus (162) Google Scholar]. Secondly, technical variations in CMS classification between the two studies are likely to have impacted on the subtype determinations and distributions. The CMS classification framework was originally developed from microarray-based gene expression profiles or RNA sequencing of fresh-frozen primary CRC samples, mostly (>90%) from patients with non-metastatic disease [1.Guinney J. Dienstmann R. Wang X. et al.The consensus molecular subtypes of colorectal cancer.Nat Med. 2015; 21: 1350-1356Crossref PubMed Scopus (2652) Google Scholar]. The current lack of a standard assay for classification of formalin-fixed paraffin-embedded (FFPE) tumor samples imposes several non-trivial challenges, which were approached differently in FIRE3 and CALGB with respect to the choice of gene expression platform, size and composition of the template gene sets, as well as the classification algorithm. Furthermore, unclassified samples with a ‘mixed’ CMS phenotype represented 12% of the samples in CALGB, while no samples were unclassified in FIRE3. Finally, the accuracy of CMS3 classification is generally low, and this may be increased in an RAS wild-type population that is underrepresented for CMS3 [1.Guinney J. Dienstmann R. Wang X. et al.The consensus molecular subtypes of colorectal cancer.Nat Med. 2015; 21: 1350-1356Crossref PubMed Scopus (2652) Google Scholar], highlighting inherent difficulties with correlative analyses in this subtype. Thirdly, the differences in the combination chemotherapy backbones between CALGB and FIRE3 might have influenced the results, as proposed by the authors [10.Aderka D. Stintzing S. Heinemann V. Explaining the unexplainable: discrepancies in results from the CALGB/SWOG 80405 and FIRE-3 studies.Lancet Oncol. 2019; 20: e274-e283Abstract Full Text Full Text PDF PubMed Scopus (31) Google Scholar]. There are known interactions between biologics and chemotherapies used (irinotecan versus oxaliplatin) with a potential impact on treatment efficacy. Importantly, these interactions may be influenced by the tumor microenvironment, which is highly heterogeneous among the CMS groups: undifferentiated CMS1 and CMS4 tumors are ‘immune-stromal rich’, while differentiated CMS2 and CMS3 tumors are ‘immune-stromal desert’. One intriguing hypothesis is that a potentially synergistic effect between cetuximab and oxaliplatin, used in the vast majority of patients included in the CALGB study, is limited to tumors with a fibroblast-low microenvironment (CMS2/3). In this context, it is important to note inconsistent results in studies assessing CMS groups as predictors of benefit with oxaliplatin and irinotecan in the early-stage and metastatic settings, respectively [11.Pogue-Geile K. Andre T. Song N. et al.Association of colon cancer (CC) molecular signatures with prognosis and oxaliplatin prediction-benefit in the MOSAIC Trial (Multicenter International Study of Oxaliplatin/5FU-LV in the Adjuvant Treatment of Colon Cancer).J Clin Oncol. 2019; 37Crossref Google Scholar, 12.Okita A. Takahashi S. Ouchi K. et al.Consensus molecular subtypes classification of colorectal cancer as a predictive factor for chemotherapeutic efficacy against metastatic colorectal cancer.Oncotarget. 2018; 9: 18698-18711Crossref PubMed Scopus (91) Google Scholar]. In practice, the independent versus combinatorial effect of different chemotherapy agents and targeted drugs in the CMS framework is difficult to disentangle. Finally, factors related to CRC biology and tumor heterogeneity are, in our opinion, the strongest determinants of the lack of reproducible associations between CMS and benefit from specific therapies, as illustrated in Figure 1. Despite significant enrichments for EGFR and VEGF pathway activities among CMS groups, these bioinformatical associations do not represent ‘target dependencies’ that are unique to one subtype. Pathway addictions defined by gene expression are never an ‘on–off’ phenomenon or a ‘black–white’ phenotype, but a dynamic state with various ‘shades of grey’. In contrast to the situation in breast cancer, we lack targetable alterations that are hallmarks of a single transcriptomic subtype of CRC. To illustrate this point, the VEGF pathway activation and proangiogenic features found in CMS4 tumors are accompanied by a surprisingly small benefit from bevacizumab-containing regimens in this subtype in the CALGB, FIRE3 and MAX studies. This seemingly paradoxical effect has been proposed to be linked to a microenvironment rich in cancer-associated fibroblasts and macrophages that release multiple proangiogenic factors promoting VEGF-independent angiogenesis in CMS4 tumors [10.Aderka D. Stintzing S. Heinemann V. Explaining the unexplainable: discrepancies in results from the CALGB/SWOG 80405 and FIRE-3 studies.Lancet Oncol. 2019; 20: e274-e283Abstract Full Text Full Text PDF PubMed Scopus (31) Google Scholar, 13.Francia G. Emmenegger U. Kerbel R.S. Tumor-associated fibroblasts as “Trojan Horse” mediators of resistance to anti-VEGF therapy.Cancer Cell. 2009; 15: 3-5Abstract Full Text Full Text PDF PubMed Scopus (38) Google Scholar]. An additional complicating factor in the biological diversity of the CMS groups is the strong evidence that most CRC tumors do not have one unique and ‘clonal’ CMS assignment. Intra-tumor CMS heterogeneity has been demonstrated both by bioinformatic modeling of bulk tumor tissue samples [14.Laurent-Puig P. Marisa L. Ayadi M. et al.Colon cancer molecular subtype intratumoral heterogeneity and its prognostic impact: an extensive molecular analysis of the PETACC-8.Ann Oncol. 2018; 29Abstract Full Text Full Text PDF PubMed Google Scholar] and by multi-region sampling of primary tumors. This heterogeneity may be at least partly related to tumor infiltration of nonmalignant cells [15.Dunne P.D. Alderdice M. O'Reilly P.G. et al.Cancer-cell intrinsic gene expression signatures overcome intratumoural heterogeneity bias in colorectal cancer patient classification.Nat Commun. 2017; 8: 15657Crossref PubMed Scopus (57) Google Scholar]. This ‘subclonality’ may help understand primary-metastasis heterogeneity in CMS classification of patient-matched samples, which may be as high as 40% [16.Piskol R. Huw L.Y. Sergin I. et al.A clinical applicable gene expression classifier reveals intrinsic and extrinsic contributions to consensus molecular subtypes in primary and metastatic colon cancer.Clin Cancer Res. 2019; 25: 4431-4442Crossref PubMed Scopus (31) Google Scholar]. However, a dedicated translation of CMS classifiers to colorectal tumors from different metastatic organs is still pending, complicated by the strong and organ-specific influence of gene expression signals from the tumor microenvironment. Accordingly, the use of primary tumors as the sample source for CMS classification in both the CALGB and FIRE3 studies indicate vulnerability to the effects of tumor heterogeneity. A marked shift in the subtype distribution is expected in metastatic CRC, related to the aggressive biology of CMS4 tumors, the lower prevalence of MSI in metastatic disease, and a potential effect of standard chemotherapies on the subtype of progressing cancers [17.Fontana E. Eason K. Cervantes A. et al.Context matters - consensus molecular subtypes of colorectal cancer as biomarkers for clinical trials.Ann Oncol. 2019; 30: 520-527Abstract Full Text Full Text PDF PubMed Scopus (62) Google Scholar]. In summary, the CMS classification does not currently provide a rationale for therapy selection in metastatic CRC. The CALGB and FIRE3 investigators should be commended for their efforts to provide the data needed to reach this conclusion. On top of technical issues leading to misclassification of CRC samples, biological factors related to spatial and temporal intra-tumor heterogeneity may impact on the results of studies assessing interactions between CMS groups and treatment effects. For a gene expression classifier to be of clinical utility, it has to be accessible, reliable, consistent, economical and able to improve the accuracy of validated predictive factors, including genomic and clinical markers (i.e. left- versus right-sided tumor location with regard to anti-EGFR agents). Benchmark metrics essential to robust NanoString® assay development were recently illustrated, including the representativeness of smaller template gene sets, cross-platform performance, reproducibility in replicate analyses and robustness to sample source (fresh frozen versus FFPE) and bioinformatic class prediction methods [18.Ragulan C. Eason K. Fontana E. et al.Analytical validation of multiplex biomarker assay to stratify colorectal cancer into molecular subtypes.Sci Rep. 2019; 9: 7665Crossref PubMed Scopus (27) Google Scholar]. In order to better exploit CMS for therapy selection, we should exit the paradigm of CMS as a ‘one marker fits all’ tool to optimize the use of existing drugs. Different research groups are actively working to refine the transcriptomic analysis of CRC from the viewpoint of treatment efficacy prediction, such as the combination of CMS classification with a signature of DNA damage repair efficiency to model oxaliplatin benefit in the adjuvant setting [11.Pogue-Geile K. Andre T. Song N. et al.Association of colon cancer (CC) molecular signatures with prognosis and oxaliplatin prediction-benefit in the MOSAIC Trial (Multicenter International Study of Oxaliplatin/5FU-LV in the Adjuvant Treatment of Colon Cancer).J Clin Oncol. 2019; 37Crossref Google Scholar]. The CMS classification should be a starting point to deepen our knowledge about CRC biology and the ‘drivers’ of metastatic progression that could support drug discovery and rational combination therapies. The transcriptomic subtyping of tumors that become resistant to approved targeted therapies is of particular interest, and a recent study described subtype switching with stromal remodeling and increased cytotoxic immune infiltration linked to PD-L1 and LAG3 checkpoint expression in metastatic CRC with acquired resistance to cetuximab [19.Woolston A. Khan K. Spain G. et al.Genomic and transcriptomic determinants of therapy resistance and immune landscape evolution during anti-EGFR treatment in colorectal cancer.Cancer Cell. 2019; 36: 35-50Abstract Full Text Full Text PDF PubMed Scopus (126) Google Scholar]. Additional efforts to optimize the classification of CRC, including the ‘functional’ characterization of cancer cell and/or microenvironment-derived signaling dependencies, may overcome some of the limitations of CMS groups when developing biomarkers for novel targeted therapies and immunotherapy combinations. The work of AS is partly funded by Norwegian Cancer Society under grant number 6824048-2016. RD has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement no. 754923 (COLOSSUS—Advancing a Precision Medicine Paradigm in metastatic Colorectal Cancer: Systems based patient stratification solutions). The materials presented and views expressed here are the responsibility of the authors only. The EU Commission takes no responsibility for any use made of the information set out.