Use of multivariate analysis to suggest a new molecular classification of colorectal cancer

Enric Domingo, Rajarajan Ramamoorthy, Dahmane Oukrif, Daniel Norris Rosmarin, Michal Presz, Haitao Wang, Hannah Pulker, Helen Lockstone, Tarjei S. Hveem, Treena Cranston, Håvard E. Danielsen, Marco Novelli, Brian R Davidson, Zheng‐Zhou Xu, Peter L. Molloy, Elaine Johnstone, Chris Holmes, Rachel Midgley, David Kerr, Oliver M. Sieber, Ian TomlinsonThe Journal of Pathology 2012 · DOI

TL;DR

This study proposes a novel, more granular molecular classification for colorectal cancer (CRC) by using advanced multivariate analyses to uncover complex interdependencies between genetic mutations and chromosomal instability. This new classification better reflects underlying tumor biology and shows potential for improved prognostic prediction compared to existing methods.

What Problem Does This Paper Address?

Current molecular classifications of colorectal cancer (CRC) primarily rely on broad categories like microsatellite instability (MSI), KRAS/BRAF mutations, and chromosomal instability (CIN). While useful, these categories are thought to be too simplistic and may not fully capture the diverse molecular subtypes of CRC, leading to a "residual heterogeneity" in understanding tumor behavior and prognosis. The authors aim to refine this classification by identifying primary molecular drivers and their complex relationships, moving beyond simple pairwise associations.

Methods

The researchers employed a multi-pronged analytical approach on a large dataset of CRC samples:

  • Multivariate Logistic Regression: Used to identify primary associations between molecular alterations and clinico-pathological variables, helping to disentangle direct effects from indirect ones.
  • Bayesian Network Analysis: A powerful tool for modeling conditional dependencies between multiple random variables. This allowed the authors to construct a probabilistic model representing the relationships between genetic mutations (KRAS, NRAS, BRAF, PIK3CA, TP53) and CIN/MSI, identifying primary positive and negative associations.
  • Unsupervised Hierarchical Clustering: Applied to group tumors based on their molecular profiles, providing a visual and quantitative way to confirm the distinct molecular subgroups identified by the other methods.

These methods were used in conjunction to identify the "primary drivers" of associations, moving beyond simple pairwise correlations that might be misleading due to confounding factors.

Datasets

The study utilized data from the VICTOR clinical trial, which included 906 stage II/III colorectal cancer patients. The dataset comprised:

  • Somatic Mutations: Analysis of 11 specific somatic genetic alterations, including KRAS, NRAS, BRAF, PIK3CA, TP53, and FBXW7/CDC4.
  • Chromosomal Instability (CIN): Assessed through various methods.
  • Microsatellite Instability (MSI): Assessed through various methods.
  • Loss of Heterozygosity (LOH): Analyzed for specific chromosomal regions (5q, 17p, 18q) in a subset of patients.
  • Clinico-pathological Variables: Data on stage, gender, tumor location, differentiation, age, and treatment arm were collected.

This represents a large, well-characterized cohort with comprehensive molecular and clinical data, which is crucial for robust multivariate analysis.

Key Insights and Ideas

  • Complex Interdependencies: The study revealed that many previously observed associations between molecular alterations in CRC are not direct but are secondary to primary drivers. For example, the relationship between CIN and BRAF is mediated by MSI.
  • Direct Negative KRAS-TP53 Association: A significant finding is the confirmation of a direct negative association between KRAS and TP53 mutations, which counterbalances a weaker, indirect positive association mediated through a cascade of other alterations (TP53-CIN-MSI-BRAF-KRAS). This challenges simpler models of tumorigenesis.
  • Heterogeneity within CIN+ Cancers: The authors identified two distinct subtypes of CIN+ colorectal cancers, challenging the conventional view that CIN+ tumors are a monolithic group. One subtype was associated with distal location and male gender, while the other was linked to higher stage but not gender or location.
  • Prognostic Significance of New Groups: The proposed molecular classification demonstrated potential for improved prognostic prediction, with one specific group (KRAS/PIK3CA-mutant, CIN+, TP53-wild-type) showing significantly worse disease-free survival.

Key Results

  • Primary Associations Identified: The Bayesian network analysis identified key primary positive associations: CIN and TP53; MSI and BRAF; KRAS and PIK3CA. Primary negative associations included MSI with CIN and NRAS, and KRAS with BRAF, NRAS, and TP53.
  • Proposed 7-Group Classification: Based on these primary associations, a new classification was proposed, encompassing 7 groups, with 5 main molecularly defined groups (MSI+/BRAFmut; CIN+/TP53mut with WT KRAS/PIK3CA; KRAS/PIK3CAmut with CIN+ and TP53WT; KRAS/PIK3CAmut with CIN- and TP53WT; NRASmut; no mutations; others). This classification covered over 80% of the studied CRCs.
  • Clinico-Pathological Correlations: Group 1 (MSI+/BRAFmut) was strongly associated with proximal location, poor differentiation, and female gender, as expected. However, the CIN+ groups showed distinct associations: Group 2 with distal location and male gender, and Group 3 with higher stage.
  • Prognostic Impact: Group 3 (KRAS-/PIK3CA-mutant, CIN+, TP53-wild-type) was identified as an independent predictor of poor disease-free survival (hazard ratio = 1.59, p=0.008). Stage III was also a significant predictor.

Limitations and Future Work

  • Sample Size for Specific Groups: Some of the proposed groups, particularly Group 5 (NRAS-mutants), were small, which may limit the statistical power for drawing definitive conclusions about their associations.
  • Refinement and Replication: The authors acknowledge that the proposed classification requires further replication and refinement in independent cohorts.
  • Emerging Complexity with Next-Gen Sequencing: The authors note that as next-generation sequencing uncovers more low-frequency mutations (like FBXW7), fitting them into established pathways will become more challenging, suggesting that future classifications may need to be even more complex.
  • Further Investigation of MSI Status: While MSI was a borderline significant indicator of good prognosis, the authors suggest further investigation into its independent prognostic value, especially in relation to BRAF-mutant MSI cancers.
  • Understanding FBXW7: The study found FBXW7/CDC4 mutations to be largely random and unassociated with other molecular or clinico-pathological features, suggesting it might mutate early in tumorigenesis or play a role independent of the major pathways studied. Further research is needed to clarify its role.

Summary updated

Abstract

Molecular classification of colorectal cancer (CRC) is currently based on microsatellite instability (MSI), KRAS or BRAF mutation and, occasionally, chromosomal instability (CIN). Whilst useful, these categories may not fully represent the underlying molecular subgroups. We screened 906 stage II/III CRCs from the VICTOR clinical trial for somatic mutations. Multivariate analyses (logistic regression, clustering, Bayesian networks) identified the primary molecular associations. Positive associations occurred between: CIN and TP53 mutation; MSI and BRAF mutation; and KRAS and PIK3CA mutations. Negative associations occurred between: MSI and CIN; MSI and NRAS mutation; and KRAS mutation, and each of NRAS, TP53 and BRAF mutations. Some complex relationships were elucidated: KRAS and TP53 mutations had both a direct negative association and a weaker, confounding, positive association via TP53-CIN-MSI-BRAF-KRAS. Our results suggested a new molecular classification of CRCs: (1) MSI(+) and/or BRAF-mutant; (2) CIN(+) and/or TP53(-) mutant, with wild-type KRAS and PIK3CA; (3) KRAS- and/or PIK3CA-mutant, CIN(+) , TP53-wild-type; (4) KRAS(-) and/or PIK3CA-mutant, CIN(-) , TP53-wild-type; (5) NRAS-mutant; (6) no mutations; (7) others. As expected, group 1 cancers were mostly proximal and poorly differentiated, usually occurring in women. Unexpectedly, two different types of CIN(+) CRC were found: group 2 cancers were usually distal and occurred in men, whereas group 3 showed neither of these associations but were of higher stage. CIN(+) cancers have conventionally been associated with all three of these variables, because they have been tested en masse. Our classification also showed potentially improved prognostic capabilities, with group 3, and possibly group 1, independently predicting disease-free survival.
BibTeX
@article{domingo2012use,
  author = {Enric Domingo and Rajarajan Ramamoorthy and Dahmane Oukrif and Daniel Norris Rosmarin and Michal Presz and Haitao Wang and Hannah Pulker and Helen Lockstone and Tarjei S. Hveem and Treena Cranston and Håvard E. Danielsen and Marco Novelli and Brian R Davidson and Zheng‐Zhou Xu and Peter L. Molloy and Elaine Johnstone and Chris Holmes and Rachel Midgley and David Kerr and Oliver M. Sieber and Ian Tomlinson},
  title = {{Use of multivariate analysis to suggest a new molecular classification of colorectal cancer}},
  journal = {The Journal of Pathology},
  year = {2012},
  doi = {10.1002/path.4139},
}
Figure