After advances in multiplexed genotyping and massively parallel sequencing technologies, drug development of targeted agents in CRC is moving from the “one test-one drug” paradigm to a “genomically-stratified” model.²³ This model engenders an up-front multicategorical approach that can be used to decide among different potential targeted treatments. As summarized in Table 1, genomically driven clinical trials in advanced CRC are selecting patients according to KRAS, BRAF, NRAS and PIK3CA mutation status. Based on preclinical data and early clinical trials, KRAS wild-type and quadruple-negative tumors are particularly sensitive to dual EGFR targeting (ERBB tyrosine kinase inhibitors added to anti-EGFR mAbs) and this strategy is undergoing clinical validation.²⁴ The important role of compensatory PI3K pathway activation and HER3 signaling in the development of resistance to anti-EGFR mAbs has also been translated into clinical trials.²⁵ In the setting of acquired resistance to cetuximab and panitumumab, second-generation anti-EGFR mAbs with increased receptor internalization/degradation potency and agents engineered to induce enhanced antibody dependent cell-mediated cytotoxicity (ADCC) are now available for clinical testing.^26,27 Other promising strategies under investigation include HER2- and MET-targeted therapies in the context of acquired receptor amplification, or combination of anti-EGFR mAbs with MEK inhibitors to delay the emergence of KRAS mutated clones.^9-12

TABLE 1. Selected Genomically Driven Clinical Trials in Advanced Colorectal Cancer

View larger version (1374K)

Targeting KRAS-mutated CRC has proven particularly challenging. MEK inhibitors are the mainstay of therapy and combinations with anti-EGFR/HER3 or anti-insulin growth factor 1 receptor mAbs (IGFR1) are being explored, based on preclinical data showing that MEK inhibitor-induced activation of PI3K/AKT results from hyperactivation of these receptor tyrosine kinases.²⁸ Hopefully, this vertical dual targeting approach will yield more promising clinical results than those provided by early trials with PI3K pathway inhibitors in combination with MEK inhibitors in KRAS mutated CRC.^29,30 However, not all hotspot mutations in the KRAS gene may have the same biologic behavior. KRAS G13D allele is reported to bear a weaker transforming and pathway activation potency, and retrospective data suggest that, as opposed to the other active mutations, it may not confer resistance to cetuximab-based therapy.^31,32 In addition, as ADCC induced by IgG1 mAbs is influenced by Fc gamma receptor (FcγR) polymorphisms, specific genotypes of FcγRIIa have been linked to clinical benefit with cetuximab even in KRAS mutated tumors.³³ Prospective validation of these findings is underway.

BRAF inhibitors have demonstrated limited antitumor activity as single agents in BRAF V600E-mutated CRC.³⁴ In combination with MEK inhibitors, clinical benefit increases slightly but is still low compared with the efficacy reported in melanomas.³⁵ In CRC models, both in vitro and in vivo, inhibition of EGFR has a strong synergistic effect to BRAF targeting.^36,37 The results of clinical trials investigating BRAF inhibitors combined with anti-EGFR mAbs with or without a third agent (MEK or PI3K pathway inhibitors) are highly anticipated. With regards to NRAS-mutated CRC, clinical investigation is centered on MEK inhibitors as single agents or in combination with PI3K pathway inhibitors, following the same direction of NRAS-mutated melanoma.³⁸ In addition, for PIK3CA-mutated CRC without concurrent RAS/RAF mutations, PI3K pathway inhibitors plus anti-EGFR mAbs are under investigation.

Blockade of programmed death 1 (PD-1) and its receptor, an inhibitory receptor expressed by T cells, overcomes immune evasion and has demonstrated impressive clinical benefit in a variety of solid tumors. Interestingly, one of the longest complete remissions achieved with anti-PD-1 mAb therapy was documented in a patient with advanced MSI CRC.³⁹ The link between MSI/hypermutated phenotype/immune infiltration and benefit with immune checkpoint inhibitors is intellectually sound and is undergoing clinical validation. Of note, emerging data indicate a poor prognosis for MSI colorectal tumors that progress to stage IV disease.^40,41 The predominant role of BRAF mutations in this scenario is likely to be the main contributor of this worse outcome.

Clinical advances in metastatic CRC have not been easily translated to early stage disease. Examples include the lack of benefit of anti-EGFR mAbs in the adjuvant setting, irrespective of KRAS mutation status.⁴² Prospective observational studies and clinical trials have demonstrated that regular use of aspirin reduces the risk of developing adenomas and carcinomas.⁴³ In addition, recent data from large observational studies and retrospective analysis of randomized clinical trials suggest a survival benefit of regular aspirin among patients with established CRC restricted to PIK3CA-mutated tumors.^44,45 The mechanisms of the effect of aspirin on CRC development and progression, as well as the interaction with tumoral PIK3CA mutations, are poorly understood.⁴³

Numerous gene expression signatures predicting high risk of recurrence in early stage CRC have been developed. These prognostic classifiers are the result of supervised techniques that train models in predefined groups of patients, mainly stage II disease (relapsed vs. nonrelapsed). Oncotype DX (expression of 12 genes measured in fixed paraffin-embedded tumor tissue and scores defining tumors as low, intermediate and high risk of recurrence) and ColoPrint (18 genes assessed in fresh frozen tumor tissue and a low- vs. high-risk classification system) have been validated in large datasets of patients with CRC. A common finding of these studies is the high degree of discordance in risk stratification between the gene expression profile and conventional prognostic clinicopathological factors.^46,47 As expected, the majority of patients with MSI tumors were identified as low risk by gene expression profile. The performance of ColoPrint as a prognostic classifier in stage II CRC is being prospectively validated.

Recently, independent scientific groups applied unsupervised clustering methods to genome-wide data to define the intrinsic CRC subtypes under the hypothesis that the categorization of CRC based on molecular features would improve our understanding of the disease.^6,15-22 Using diverse classification algorithms in different datasets, these studies identified up to six molecular subtypes of CRC. Some groups assigned names to each subtype according to histopathological or pathway enrichment correlates, as seen in Table 2.^6,17,19-22 As depicted in Fig. 2, at least three subtypes were repeatedly identified: (1) MSI immune-activated, hypermutated (enriched for BRAF mutations) and CIMP-H; (2) CIN with epithelial proliferative features, upregulation of Wnt pathway and CIMP negative or low; (3) CIN with mesenchymal/invasive stromal phenotype, enriched in cancer stem cells.^6,18,22 A critical cross-comparison of the different subtypes identified by each group reveals significant overlap in key clinical and molecular correlates. However, attempts to discern the underlying somatic drivers in each biologically distinct subtype have not been successful. For example, KRAS mutations are shared out among all subtypes, supporting the idea that KRAS-mutated CRC is highly heterogeneous at gene expression level. In addition, MSI and BRAF mutation status proved helpful for highlighting subtype characteristics, but are not sufficient to define boundaries between subtypes. Regarding prognosis, the MSI enriched clusters were repeatedly associated with better disease-free survival^16,21,22 and the mesenchymal/invasive subtypes were correlated with advanced stage at diagnosis, poor differentiation of the tumor and significantly worse survival outcomes.^16-22 TGF-β activation in the tumor microenvironment appears to be one of the most relevant programs in these tumors, confirming previous findings on its influence in stromal cells and metastasis initiation.⁴⁸ Of note, most mesenchymal/stromal enriched tumors were classified as Oncotype Dx high risk.^18,20 Furthermore, activation of a gene expression signature of epithelial-mesenchymal transition (where mesenchymal and epithelial markers are overexpressed and underexpressed, respectively) correlated with reduced benefit to adjuvant chemotherapy and anti-EGFR mAbs.^18,19,22 Application of this molecular stratification to CRC cell line panels and overlaying pharmacologic response data for targeted therapies showed significant differences in sensitivity across cell lines assigned to different subtypes. Examples of the differences include increased response to aurora kinase inhibitors in epithelial proliferative CRC cells, MET inhibitors in those with mesenchymal features, and SRC/PI3K pathway inhibitors in cell lines with a MSI phenotype.^16,19

TABLE 2. Intrinsic Subtypes of Colorectal Cancer Based on Gene Expression Profile

View larger version (1095K)

FIG 2. Summary of CRC subtypes according to gene expression profiles and comprehensive genomic characterization efforts. The three “consensus” subgroups (MSI, CNI epithelial/proliferative, and CIN mesenchymal/invasive) are described and uncertainty about possible additional subtypes is also outlined.

Abbreviation: EMT: epithelial mesenchymal transition.

Nevertheless, to guide clinical practice, there is a growing need to reconcile and confirm the existence of these subtypes. Members of the CRC research community have joined together in a consortium (Colorectal Cancer Subtyping Consortium—CRCSC) to assess and establish the molecular subtypes of CRC. This effort, coordinated by Sage Bionetworks, is motivated by the recent advances in this field, including the availability of several large public and nonpublic patient cohorts with molecular and clinical annotations. The consortium members have agreed to contribute and share their respective datasets to achieve the following goals: (1) to compare and validate the major published CRC subtypes; (2) to conduct an integrative analysis across the pooled datasets to describe a robust consensus of molecular subtypes; (3) to define the clinical and molecular hallmarks of these subtypes; (4) to correlate stable and recurrent subtypes with survival endpoints and response to adjuvant therapy; and (5) to establish a new paradigm within the research community for collaborative subtyping. The results are eagerly anticipated.

The research community must make sure that important findings from large clinical trials are getting across to medical oncologists treating patients with CRC. First, rather than assessing only KRAS (exon 2) mutations in the metastatic setting, clinicians should adopt a broader mutation profiling strategy upfront. There are clear reasons for doing so: (1) addition of panitumumab or cetuximab to first-line chemotherapy in anti-EGFR naive patients whose tumors harbored KRAS mutations beyond codons 12 and 13 or NRAS mutations was associated with a detrimental effect^49,50; and (2) with regards to BRAF mutations as predictor of response to anti-EGFR mAbs, the evidence for lack of benefit in the first-line setting is not definitive,^49,51 but the same detrimental outcome was seen with panitumumab added to chemotherapy in the second-line setting.⁵² Second, genomically-driven clinical trials with matched targeted therapies for KRAS, NRAS, BRAF and PIK3CA mutations offer promising treatment opportunities for patients with CRC. Third, moving to the adjuvant setting, two large studies linked aspirin and improved outcomes primarily in PIK3CA-mutated tumors.^44,45 Finally, as an alternative to tumor biopsies, ctDNA profiling will likely be integrated into clinical trials. These circulating “liquid biopsies” may help identify primary or acquired resistance mechanisms to targeted agents and usually precede radiological evidence of tumor growth, which will hopefully assist inpatient care.^11,12

Our current understanding of the genomic and epigenomic features of CRC, together with the identification of distinct molecularly defined subtypes based on gene expression profiles, highlights the complexity and heterogeneity of the disease. Each of these classification modalities provides different perspective on the same underlying biologic reality. The challenge now is to translate all these findings into a comprehensive, robust and reproducible CRC classification system linking molecular features to precision medicine.

To conclude, the successful clinical translation of prognostic gene expression signatures and molecular subtyping in breast cancer should serve as an example to the CRC community. Nevertheless, considering the heterogeneity of CRC, it is expected that the “one size fits all” prognostic signature approach may be difficult to implement. In addition, the differences among all independent unsupervised subtyping studies in CRC and the lack of unique driver events in each cluster suggest that common genomic aberrations already in clinical use will not accurately distinguish these intrinsic subtypes. The integrative analysis being performed by the CRCSC will hopefully conciliate disparities, integrate information from various platforms, and establish a robust consensus of molecular subtypes that becomes a tool for clinical use, providing a guide for patient stratification and for rational design of drugs targeting specific pathways.