Colorectal Cancer Heterogeneity: Evolving Classification Schemes for Prognostication and Prediction for Response to Therapy

Colorectal Cancer Heterogeneity: Evolving Classification Schemes for Prognostication and Prediction for Response to Therapy


Dr. Kit Man Wong

Dr. William M. Grady

By Kit Man Wong, MD, and William M. Grady, MD

Article Highlights

  • Colorectal cancers (CRCs) can be classified into three subgroups based on their underlying genomic instability or epigenomic instability.
  • Right-sided CRCs have worse prognosis and distinct molecular characteristics compared to left-sided CRCs. Given our current understanding of anatomic location and prognosis, CRC sidedness should be routinely documented as part of the standard tumor profile.
  • The Consensus Molecular Subtypes classify CRCs based on their inherent gene expression patterns.
  • The Immunoscore is a CRC stratification approach that aims to describe the immune contexture of tumors.
  • These classifications have provided insights into the molecular and immune/stromal heterogeneity of CRCs and demonstrated prognostic value based on preliminary data.

Using whole genome sequencing, epigenetic analysis, and gene expression analysis to characterize colorectal cancer (CRC) has led to a greater understanding of this disease and has revealed its remarkable molecular heterogeneity. Although the TNM staging system for CRC established by the American Joint Committee on Cancer is currently the standard classification approach used in clinical practice to determine prognosis,1 a significant variability in patient outcomes is observed within the same TNM stage.2 This highlights the reality that other factors are relevant beyond the extent of disease in determining the prognosis of patients with CRC. It is widely believed that the most accurate CRC classification system will include measures of the intrinsic biology of the tumor to not only provide an accurate prognostic tool but also a predictive tool for response to specific targeted therapies.

CRC has been previously subdivided at a molecular level in different ways. Many of the existing classification schemes are based on inherent tumor cell characteristics. However, with the increasing appreciation for the role that the tumor microenvironment and host immune system play in tumorigenesis,3 these extrinsic factors have been integrated into recently proposed classification systems.

Tumor Cell–Based Molecular Profiling

The genome-wide assessment of recurrent somatic mutations in key oncogenes and tumor suppressors in CRC has been made possible by recent advances in DNA sequencing technology. The Cancer Genome Atlas (TCGA) network conducted a comprehensive analysis of somatic alterations in more than 250 colorectal tumors, which revealed the most commonly mutated genes, including APC (81%), TP53 (60%), KRAS (43%), PIK3CA (18%), and NRAS (9%) in the non-hypermutated class of CRCs.4 In contrast, BRAF V600E mutation was frequently associated with hypermutated CRCs (46%). The accurate characterization of the frequency and nature of mutations in these genes is particularly significant, as they function in critical signaling pathways that control the hallmark behaviors of CRC.3 In addition, some of these pathogenic mutations have well-established prognostic or predictive value. For example, activating mutations of KRAS/NRAS in the MAPK pathway predict a lack of response to anti-EGFR agents, while BRAF V600E is a known poor prognostic factor in CRC.5 Presently in clinical practice, extended RAS mutation analysis is done by targeted sequencing, and the results are used to guide the treatment of patients with metastatic CRC.

TCGA analysis of genome-wide alterations also confirmed that CRCs can be classified into three subgroups based on their underlying genomic instability or epigenomic instability. The classes include: chromosomal instability (CIN; 84%), microsatellite instability (MSI) due to mismatch repair (MMR) deficiency and characterized by hypermutation (16%), and CpG island methylator phenotype (CIMP; 20%).6,7

These subgroups have distinct features. Most CRCs have CIN as marked by chromosome and copy-number changes, and this is thought to result in genetic alterations that activate oncogenes (e.g., KRAS and PIK3CA) and inhibit or delete tumor suppressor genes (e.g., APC and TP53).6 CIMP-positive CRCs, which are defined by an excessively high frequency of aberrantly methylated DNA loci, are believed to arise from serrated polyps and are associated with increased age, female gender, proximal tumors, poor cell differentiation, high microsatellite instability (MSI-H) secondary to epigenetic repression of MLH1, and BRAF V600E mutations.8 In fact, there is substantial overlap between sporadic MSI-H and CIMP-positive CRCs. Of note, hypermutated CRCs can also arise from germline mutations in MMR genes associated with Lynch syndrome (i.e., MLH1, MSH2, MSH6, and PMS2), double somatic mutations in these MMR genes, and somatic mutations in the proofreading polymerases POLE and POLD1.9,10

It has been observed that patients with CIMP-positive, microsatellite stable (MSS) CRC have worse survival outcomes.11 In contrast, multiple studies have demonstrated the favorable prognosis associated with MSI-H tumors.9 The assessment of MMR protein expression by immunohistochemistry is now a standard component of the pathologic examination of CRCs, as the MSI status carries prognostic and predictive value and is used to screen for Lynch syndrome. MSI-H predicts a lack of benefit with adjuvant 5-fluorouracil in resected stage II CRC,12 whereas in the metastatic setting, it correlates with response to immune checkpoint blockade therapies, presumably because these hypermutable CRCs have increased tumor mutational burden and consequent neo-antigen load, making them more likely to be recognized by the adaptive immune system.13

Furthermore, using a combination of these genomic and epigenomic biomarkers, CRC subtypes have been defined with different prognoses.11,14 For example, a study demonstrated that stage III CRCs can be categorized into five distinct groups based on their MSI status and the presence of mutant KRAS or BRAF:

  • MSS and BRAF mutation (6.9%),
  • MSS and KRAS mutation (35%),
  • MSS without KRAS or BRAF mutation (49%),
  • MSI-H and BRAF mutation or MLH1 hypermethylation (6.8%), and
  • MSI-H without BRAF mutation or MLH1 hypermethylation.14

MSS tumors with BRAF V600E mutation were more likely to be proximal and high grade and to have nodal involvement. The disease-free survival was significantly shorter in stage III CRCs with proficient MMR and mutation in KRAS or BRAF. Moreover, CIMP was shown to interact with MMR and KRAS/BRAF status in determining CRC prognosis.11 Particularly, CRCs with a MSI-H, non-CIMP, and wild-type KRAS/BRAF profile had a more favorable prognosis compared to other tumors (HR 0.30, 95% CI [0.14, 0.66]).

Immune Profiling

The immune system plays a central role in tumor biology. The abnormal balance between immune cell activation caused by cancer antigens and immune cell suppression induced by cancer cells is an important determinant of the malignant phenotype.15 This concept has been exploited in the successful use of immunotherapies and immune checkpoint blockade to elicit the host anti-tumor immune response in several malignancies, especially those with a high rate of mutations.13

Consequently, the immunologic aspects of the tumor microenvironment, including the number and types of infiltrating immune cells, will likely be an important part of the tumor profile in the future and may be used to distinguish CRC subtypes. This is best exemplified by the Immunoscore, an immune-based CRC stratification approach that aims to describe the immune contexture of tumors.15 Using immunohistochemistry and digital pathology for quantification, the Immunoscore is a composite evaluation of immune cell type (CD3, CD8, and CD45RO), density (cells/mm3), and location (tumor core and invasive margin), yielding a score from 0 (low cell densities in both locations) to 4 (high cell densities in both locations).15,16 This new classification has been validated in independent CRC cohorts. Indeed, it showed superior prognostic performance compared to the TNM system regardless of stage.16-18

Efforts by an international consortium of 17 countries to validate the Immunoscore in 1,336 patients with stage I to III CRCs demonstrated significantly longer time to recurrence in patients with high compared to low scores (HR 0.54, 95% CI [0.34, 0.84]; p = 0.006). Moreover, it effectively differentiated stage II CRCs into low-risk (high score) and high-risk (low score) groups (time to recurrence: HR 0.46, 95% CI [0.24, 0.87]; p = 0.014).19

Although further validation by other teams of investigators is needed, the Immunoscore has the potential to incorporate into TNM staging to better risk-stratify CRCs. Another possible and useful application is to predict response to immune checkpoint inhibitors (e.g., anti–PD-1 and anti–PD-L1 inhibitors) in metastatic CRC.20

The Consensus Molecular Subtypes of CRC

Several molecular classifications of CRCs derived from global gene expression analysis were previously published but yielded inconsistent results, likely due to variations in study population, assay platform, and data processing methods.21 In order to reconcile these discrepancies, the CRC Subtyping Consortium combined genomic data and compared the classification algorithms used across six different studies that included 4,151 tumors.21 The analysis showed high correlation between these classifications and defined four consensus molecular subtypes (CMS) for CRC, each associated with a distinct profile of biologic features, pathway activity, and cellular processes. These subtypes are:

  • CMS1 (MSI immune, 14%), which encompass increased mutations, MSI, and immune activation;
  • CMS2 (canonical, 37%), which has epithelial gene expression with strong WNT and MYC signaling;
  • CMS3 (metabolic, 13%), which is characterized by epithelial gene expression and dysregulation of metabolic pathways; and
  • CMS4 (mesenchymal, 23%), which includes activated TGF-b and features of stromal invasion and angiogenesis.

Moreover, there is a greater rate of BRAF mutations in CMS1 CRCs and KRAS mutations in CMS3 CRCs.21 Thus, the CMS stratify CRCs based on their inherent gene expression patterns that reflect fundamental aspects of their tumor biology. The observation that the CMS classes exhibit different clinical features and prognosis supports the concept that they represent truly unique CRC subtypes and not simply artifacts of the clustering methods used. In further support of the clinical significance of the CMS classes, CMS1 tumors tend to occur in females with right-sided colon cancer and correlate with higher grade and poorer prognosis after relapse. The mesenchymal CMS4 CRCs are more strongly associated with stage III-IV disease and younger age at diagnosis, as well as worse overall and relapse-free survival compared to the other subtypes.21

A recent transcriptomic analysis demonstrated that the CMS4 CRCs have distinct gene signatures of the tumor microenvironment, which encompasses the immune, inflammatory, angiogenic, and fibroblastic aspects.22 For example, cytotoxic lymphocyte–specific genes are upregulated in CMS1 CRCs, whereas CMS4 CRCs are marked by a strong immunosuppressive, angiogenic, and inflammatory environment. In contrast, CMS2 and CMS3 subtypes display intermediate profiles. Therefore, this study adds further support for the robustness of the CMS classification. Whether the CMS class will predict response to immune- or stromal-targeting strategies (e.g., anti-angiogenics) remains to be seen.

Distinction by Anatomic Location

Right- and left-sided colon cancers arise from the embryological midgut and hindgut, respectively, and exhibit clinically different features.23 Evidence has shown that they also have distinct gene expression profiles and pathway activity.24 Multiple retrospective and prospective studies comparing the survival of patients with right- and left-sided CRC have been published with variable results, although most of these studies point to a poorer prognosis for right-sided CRC. A meta-analysis of 15 studies including 108,474 patients confirmed worse overall survival (OS) in right- versus left-sided CRC independent of stage (HR 1.14, 95% CI [1.06, 1.22], p < 0.01). In fact, there was a survival advantage for left- over right-sided CRC even in the setting of stage I disease (5-year OS, 84% vs. 78%; p = 0.01).25

The effects of tumor location on prognosis and treatment outcomes were highlighted in a recent retrospective analysis of the CALGB/SWOG 80405 trial of chemotherapy in combination with cetuximab or bevacizumab in the first-line treatment of metastatic CRC.26 This study confirmed worse OS (HR 1.56, 95% CI [1.32, 1.84]; p < 0.0001) and progression-free survival (HR 1.25, 95% CI [1.08, 1.46]; p = 0.002) for right-sided CRC. Importantly, although the CALGB/SWOG 80405 trial showed no difference between the use of cetuximab or bevacizumab in the overall study population, re-analysis of the data by tumor location revealed longer median OS with cetuximab in left-sided CRC (37.5 months vs. 32.1 months; log rank p = 0.04) and with bevacizumab in right-sided CRC (24.5 months vs. 16.4 months; log rank p = 0.03). Indeed, tumor sidedness had the most profound impact on the efficacy of cetuximab: For right-sided CRC, median OS was 16.4 months compared with 37.5 months for left-sided CRC (HR 1.97, 95% CI [1.56, 2.48]).26 This variation in outcomes might be related to the association of right-sided CRC with CIMP-high, mutant BRAF, CMS1, and CMS3, as well as epigenetic downregulation of EGFR ligands.27

Further work is needed to better understand the molecular and immune differences between right- and left-sided CRC, and how to best incorporate tumor location in disease management. Given our current understanding of anatomic location and prognosis, CRC sidedness should be routinely documented as part of the standard tumor profile.

Future Directions

The availability of genomic tools and large-scale collaborative efforts have led to significant advances in our understanding of the common molecular features of CRC. The emerging molecular landscape of CRC is more complex than previously appreciated and will likely continue to increase in complexity as our understanding of non-coding RNAs and other molecular features matures. The CMS and Immunoscore classifications have provided insights into the molecular and immune/stromal heterogeneity of CRCs and have demonstrated prognostic value based on preliminary data.

Further research is needed to validate these classification systems in prospective studies and to perhaps integrate molecular and immune signatures, as both play important roles in determining the tumor phenotype. In addition, an ideal classification system would not only enable the accurate risk stratification of patients but would also serve as a predictive tool for treatment decisions driven by the biology of individual tumors. Thus, the predictive value of the current and future CRC molecular classification systems appears promising and should be further investigated in clinical trials.  

About the Authors: Dr. Wong is with the Division of Medical Oncology, Department of Medicine, at the University of Washington School of Medicine. Dr. Grady is with the Division of Gastroenterology, Department of Medicine, at the University of Washington School of Medicine.