Consensus Molecular Subtypes (CMS) Classification: a progress towards Subtype-Driven treatments in colorectal cancer

Abstract

Background

Colorectal cancer (CRC) is heterogeneous with varied molecular profiles, clinical outcomes, and treatment responses. The Consensus Molecular Subtypes (CMS) classification categorizes CRC into four molecular subtypes (CMS1–4) based on gene expression profiles, aiming to improve prognosis prediction and guide personalized therapy.

Objective

This paper reviews the CMS classification, its prognostic and predictive roles, methods of identification, association with polyps, immunotherapy applications, intratumoral heterogeneity (ITH), and barriers to clinical adoption.

Methods

The review synthesizes data from global studies on CMS, searches were conducted in PubMed, Scopus, and Web of Science, till June 2025, focusing on gene expression profiling, immunohistochemistry (IHC), and image-based CMS (imCMS) classifiers based on computational models (such deep learning). It examines CMS-specific treatment responses, immune profiles, and emerging strategies like single-cell RNA sequencing to address ITH.

Results

CMS classifies CRC into four subtypes: CMS1 (MSI-immune, ~ 15%), CMS2 (canonical, ~ 40%), CMS3 (metabolic, ~ 13%), and CMS4 (mesenchymal, ~ 22%). CMS is prognostic in adjuvant (e.g., PETACC-3, NSABP C-07) and metastatic settings (e.g., CALGB 80405, FIRE-3), with CMS2 linked to the best survival and CMS4 the worst outcomes. CMS1 responds to immunotherapy, CMS2/3 to bevacizumab, and CMS4 to irinotecan. Classification methods include gene expression profiling, immunohistochemistry (IHC), and image-based CMS (imCMS), but intratumoral heterogeneity (ITH) and technical barriers hinder clinical adoption. Solutions like single-cell sequencing and standardized assays are emerging.

Conclusion

CMS classification enhances CRC prognosis and treatment personalization but faces challenges due to ITH and technical limitations. Advances in IHC, imCMS, and targeted gene panels may facilitate broader clinical adoption, improving patient outcomes through tailored therapies.

Highlights

This review synthesizes CMS applications in immunotherapy, polyp linkages, spatial transcriptomics, and ctDNA.

This review identifies the obstacles to CMS adoption: lack of standardization, expense, technological difficulty, suggesting gene panels, and IHC-based techniques. Reiterates CMS’s clinical significance in immunological profiles of early-stage and metastatic colorectal cancer, which goes beyond subtype or single-therapy research.

This review Offers a forward-looking paradigm for precision medicine in colorectal cancer that incorporates current developments and practical clinical solutions.

Introduction

Colorectal cancer (CRC) is one of the most common gastrointestinal malignancies and a leading cause of cancer-related mortality for both men and women worldwide. Despite advances in treatment, metastatic colorectal cancer (mCRC) has a 5-year survival rate of approximately 20%, and about 40% of non-metastatic tumors recur following complete resection ([1] Notably of the main problems with colorectal cancer (CRC) is tumour heterogeneity, which includes genetic, epigenetic, and transcriptome differences. It directly contributes to poor prognosis, drug resistance, recurrence, and metastasis [2, 3]. Histological grading and other traditional techniques are not good predictors [4]. Apart from tumours harbouring BRAF-V600E mutations, targeted therapy for colorectal cancer often have a low response rate (less than 5%) [3]. Thus, combining genetic and histological markers is crucial to creating personalized and precise treatment plans.

At the genomic level, CRC is characterized by two primary mechanisms of genomic instability: chromosomal instability (CIN) and microsatellite instability (MSI), despite a variety of distinct genetic alterations [5]. The Cancer Genome Atlas (TCGA) network introduced the first comprehensive classification model for CRC using multi-omics molecular data. Through extensive transcriptional genome research, six distinct molecular classification methods were brought together to obtain this conclusion. In order to overcome the limitations of inadequately predictive conventional approaches and classify the great heterogeneity of colorectal cancer (CRC), the Consensus Molecular Subtypes (CMS) were developed [6].

This study identified molecular subtypes that aligned with CIN and MSI patterns, integrating epigenetic, mutational, mRNA, and miRNA data. The MSI group was further subdivided based on mutational burden into:

• (A)
  Two groups of highly mutated CRC (16%):

  1. MSI caused by defective mismatch repair (dMMR) in hypermutated cancers (13%).
  2. Ultra mutated cancers (3%) with mutations in the DNA polymerase epsilon (POLE) exonuclease domain, rendering the proofreading function inactive.
• (B)CRC with low mutation rates (84%), which are non-hypermutated, microsatellite stable (MSS), exhibit a higher frequency of DNA somatic copy number alterations (SCNAs) due to CIN, and frequently display deletions or mutations in TP53, APC, KRAS, PIK3CA, and SMAD4 genes [5–7].

This review aims to comprehensively evaluate the Consensus Molecular Subtypes (CMS) classification’s role in guiding colorectal cancer (CRC) prognosis, treatment personalization, and overcoming intratumoral heterogeneity, while proposing novel strategies for clinical integration. CMS provides a strong framework that goes beyond traditional staging and a limited number of molecular markers to improve prognosis and direct the choice of systemic medicines. The current staging and markers such as MMR, RAS, and BRAF status that are used in current molecular practice are insufficient since they do not represent the intrinsic biology of the tumour.

This is perhaps the first study to synthesize CMS applications across polyp associations, immunotherapy, and advanced sequencing technologies, and provides a pioneering framework for addressing CRC’s molecular complexity.

Methods

A systematic literature search was conducted to identify studies relevant to the Consensus Molecular Subtypes (CMS) classification of colorectal cancer (CRC). PubMed, Scopus, Web of Science, and clinical trial registers from inception to June 2025 were searched to retrieve peer-reviewed articles, reviews, and clinical studies. The search strategy employed a combination of keywords and MeSH terms, including “colorectal cancer,” “Consensus Molecular Subtypes,” “CMS classification,” “microsatellite instability,” “chromosomal instability,” “prognosis,” “immunotherapy,” “intratumoral heterogeneity,” and “targeted therapies.” Boolean operators (AND, OR) were used to refine the search, and filters were applied to include English-language publications, and human studies. Additional hand-searching of reference lists from key articles was performed to identify relevant studies not captured in the database searches. Studies were included if they addressed CMS classification, its prognostic or predictive roles, methods of identification, association with polyps, immunotherapy applications, intratumoral heterogeneity, or barriers to clinical adoption. Data were extracted on study design, sample size, CMS subtype characteristics, treatment outcomes, and methodological approaches. The collected evidence was synthesized to provide a comprehensive overview of CMS applications in CRC management.

Results

The search yielded 1,234 articles after removing duplicates. Following title and abstract screening, 142 articles were selected for full-text review, and 70 studies were included in the narrative synthesis based on relevance to CMS classification and CRC. These studies encompassed molecular profiling, clinical trials, and reviews, covering CMS1–4 subtypes, their prognostic and predictive roles, and methods like gene expression profiling, immunohistochemistry, and image-based CMS classifiers. Key findings confirmed CMS2’s association with the best overall survival and CMS4’s poor prognosis, with CMS1 showing responsiveness to immunotherapy. Intratumoral heterogeneity was prevalent in over 55% of tumors, complicating classification. Technical and financial barriers, such as complex RNA analysis and lack of standardization, were identified as major obstacles to clinical adoption. Emerging strategies, including single-cell RNA sequencing and spatial transcriptomics, were highlighted as potential solutions to address heterogeneity and improve CMS utility.

Consensus molecular subtypes (CMS) classification (Table 1)

Table 1.

Consensus molecular subtype (CMS) groups [11]

Characteristic	CMS1 (MSI-Immune)	CMS2 (Canonical)	CMS3 (Metabolic)	CMS4 (Mesenchymal)
Prevalence	15%	40%	13%	25%
Genetic Instability	¾ MSI	CIN	¾ CIN	CIN
Cell Signature	-	Epithelial	Epithelial	Mesenchymal
Cell Line	Undifferentiated	Colon-like	Colon-like	Undifferentiated
Mutations	BRAF mut	-	RAS mut	-
Pathways and Gene expression profiles	MAPK, RTK, Jak-Stat and Caspases	WNT/MYC. High expression of EGFR/SRC and Her2/Cyclin upregulation	MAPK, RTK, Epithelial and metabolism signatures (glutaminolysis, lipidogenesis	TGF-β, Angiogenesis, Matrix Remodeling, Stromal Infiltration, Complement Inflammatory System, Epithelial-mesenchymal transition
Characteristics	Immune infiltration	EGFR/ligands high	Metabolic	Stromal infiltration
Immune Phenotype	Immune-activated	Immune-desert	Immune-mixed	Immunosuppression
Right/Left-Sided Tumors	Right-sided	Left-sided	Right-sided	Both sides
Prognostic Value	Good prognosis in early stage, Poor prognosis after recurrence	Highest OS	Lowest OS	Worst prognosis

(MSI Microsatellite Instability, CIN Chromosomal Instability, Mut Mutation, TGF-β Transforming Growth Factor Beta, BRAF v-raf murine sarcoma viral oncogene homolog B1, EGFR Epidermal Growth Factor Receptor)

The CMS classification is based on extensive gene expression profiling from 18 worldwide studies, utilizing bulk transcriptome data clustering from 4,151 CRC patients. Guinney et al. (2015) [3] categorized tumors according to their molecular enrichments and primary biological characteristics, which may not be exclusive to a specific CMS group. Through transcriptome profiling, tumors exhibiting CIN were further classified into CMS2–4, all of which displayed significant levels of SCNAs, while hypermutated malignancies were categorized as CMS1 [8, 9] (Fig. 1).

Fig. 1.

CMS classification of colorectal cancer and its key characteristics

CMS1 (MSI-Immune)

CMS1 tumors, comprising 15% of CRC cases, are defined by MSI due to dysfunctional mismatch repair (MMR) pathways. Promoter hypermethylation silences the MLH1 gene, impairing the MMR pathway. These tumor cells exhibit a high mutation rate, including single-base mismatches or mutations in microsatellite regions, resulting from an inability to correct DNA replication-associated errors. Most spontaneous hypermutated tumors are CpG Island Methylator Phenotype (CIMP)-high and harbor BRAF V600E or related mutations. These tumors are characterized by strong immune activity, particularly PD-1 and natural killer (NK) cell activation, with notable tumor-infiltrating lymphocytes and a tendency to be right-sided CRCs. Patients with CMS1 tumors have a very low survival rate after relapse [9, 10].

CMS2 (Canonical)

The most prevalent CRC subtype (40%), CMS2 represents canonical CRC carcinogenesis. These tumors exhibit epithelial signatures with activation of Wnt and Myc pathways, as well as high EGFR/ligand expression. They frequently show loss of tumor suppressor genes and copy number gains of oncogenes compared to other subtypes. CMS2 tumors are more commonly left-sided and associated with a higher post-relapse survival rate than other subtypes [9, 10].

CMS3 (Metabolic)

Comprising 13% of CRC cases, CMS3 tumors display predominantly epithelial signatures with frequent KRAS mutations and a higher prevalence of CIMP-low. They exhibit dysregulated metabolism across various pathways, fewer SCNAs, and more hypermutated tumors compared to CMS2 and CMS4. These tumors are more commonly right-sided [9, 10].

CMS4 (Mesenchymal)

CMS4 tumors, representing 25% of CRC cases, are characterized by mesenchymal signatures and prominent Transforming Growth Factor (TGF)-β/integrin pathways. They express genes associated with stromal invasion, angiogenesis, matrix remodeling, and complement-associated inflammation. Patients with CMS4 tumors have the worst overall survival (OS) and relapse-free survival (RFS) compared to other molecular subtypes. These tumors can be either right- or left-sided [9, 10].

Mixed

A minority (7%) of tumors exhibit “mixed” characteristics and may not be classified into any of the four molecular subtypes. This group may represent intra-tumoral heterogeneity or a “transition phenotype” [9, 10].

Methods of CMS classification

Gene expression profiling and RNA analysis are the primary techniques used to identify CRC subtypes [4]. Comprehensive genomic sequencing is performed on tumor samples, typically obtained via biopsy, to classify tumors into one of the four CMS subtypes [3]. However, the widespread clinical adoption of RNA analysis faces significant challenges. Accurate data interpretation requires specialized knowledge and bioinformatics expertise, and the process is associated with high technical complexity and costs. These factors collectively limit its routine use in clinical settings [4].

Trinh et al. developed a 5-marker immunohistochemistry (IHC) panel (ZEB1, HTR2B, CDX2, FRMD6, and cytokeratin) to eliminate the need for molecular profiling and create a CMS classification method suitable for diagnostic pathology labs. This panel complements standard MSI/dMMR testing, achieving 87% concordance with the gold-standard transcriptomic CMS classification. MSS tumors were divided into two classes using four IHC markers: epithelial (CMS2/3 combined) or mesenchymal (CMS4), with pan-cytokeratin IHC used to normalize epithelial content. CMS1 was defined using dMMR/MSI status. The authors noted the inability to separate CMS2 and CMS3, which are driven by distinct biological signaling in the original CMS study. Nevertheless, the IHC approach for CMS classification has not been widely adopted [12].

Sirinukunwattana et al.’s image-based CMS (imCMS) deep learning classifier accurately identified the four CMS classes (AUC = 0.84 in TCGA samples and AUC = 0.85 in rectal biopsies) [4]. The imCMS method can distinguish all four CMS classes without merging CMS2 and CMS3, unlike the IHC approach, and achieves comparable alignment with transcriptional CMS. As it uses diagnostic hematoxylin and eosin (H&E) images, it does not require concurrent MSI/dMMR testing or IHC staining. On a technical level, transcriptional subtypes can be predicted from standard histology sections using computational models (such deep learning), avoiding the high expense and technical complexity of traditional RNA analysis [4].

A novel PDS classification identifies three novel subtypes (PDS1-PDS3) by leveraging pathway-level data, successfully sub-stratifying the epithelial-rich CMS2 tumours into two distinct prognostic groups, PDS1 (good prognosis) and PDS3 (worst prognosis). PDS3 tumours represent a slow-cycling subset with increased differentiated lineages and a low-MYC-high-PRC target expression profile, showing biology previously overlooked by existing gene-level classifiers like iCMS. This PDS/iCMS approach allows a more comprehensive view of tumour biology, as iCMS captures the core epithelial intrinsic components (iCMS2/iCMS3 dichotomy) while PDS uncovers nuanced stem-differentiation dynamics, demonstrating that combined subtyping offers greater mechanistic understanding for precision oncology [13].

Unsorted cells from colorectal cancer (CRC) can be analyzed using single cell RNA (scRNA)-sequence. Cancer cells exhibit genetic changes and lineage-dependent gene programs that support an immunosuppressive milieu. Mechanistic insights for developing immuno-oncology therapies are provided by this cellular and immunological landscape [14].

The original four Consensus Molecular Subtypes (CMS) are refined by new classifications such as CRC Intrinsic Subtypes (CRIS) and Pathway-Derived Subtypes (PDS1-3). These novel models show a range of intrinsic biology and distinct stem cell phenotypes (LGR5 + vs. ANXA1+) using both bulk and single-cell RNA data. A developing area for improved risk assessment and individualized treatment is the application of immunohistochemistry and digital pathology [15].

Morphologic features facilitate clinical application by acting as affordable and easily available stand-ins for intricate CMS classification. In particular, there is a strong correlation between the aggressive CMS4 subtype and increased tumour budding, suggesting high-risk biology. CMS groups can be predicted by measuring the extracellular mucin area; CMS2 tumours have very little of it, but CMS1, CMS3, and CMS4 cancers have substantial levels. Additionally, the existence of Stroma A reactive Invasion Front Areas (SARIFA), which are characterized by direct interaction between the tumour and adipocytes, is a biomarker for prognosis that is closely linked to survival outcomes and shares genetic similarities with aggressive CMS1/4 (Table 2). For more precise risk classification in situations where complicated RNA sequencing is not feasible, it is essential to use these standard histological characteristics and cutting-edge digital pathology techniques to determine CMS status [16].

Table 2.

Diagnostic platforms, accuracy and reproducibility

Platform	Assay Characteristics	Diagnostic Accuracy, Reproducibility & Validation	Cost	Feasibility for Clinical Implementation
RNA-based Assays (RNA-seq, Gene Expression Profiling)	Gold Standard. n. Nanostring: A targeted RNA-based assay (gene panel).	Highest Reproducibility/Gold Standard.. Limitation: Can result in unclassified results in upto 13% of cases due to mixed subtypes.	High.	Low/Limited. Considerable technical complexity, specialized knowledge, and bioinformatics expertise.
Immunohistochemistry (IHC) Classifiers	Uses CMS-associated markers (e.g., CDX2, KRT20) on formalin-fixed, paraffin-embedded (FFPE) tissue.	Moderate Reproducibility. Aims to recapitulate the prognostic and predictive value of transcriptomic CMS.	Low	Moderate/High. More accessible and cost-effective than RNA-seq.
Image-based CMS (imCMS)	Deep Learning Model. Computational approach that predicts transcriptional CMS subtypes directly from standard H&E stained whole-slide histology images.	High Diagnostic Accuracy (Emerging). Accurately classified slides in external validation datasets (AUC 0.84–0.85). Reproduces similar prognostic associations as transcriptomic CMS. Validated using three independent cohorts (FOCUS, TCGA, GRAMPIAN). Benefit: Can classify samples previously unclassifiable by RNA profiling.	Low (Leverages existing H&E slides and digital scanners).	High/Emerging. Aims to be a simple, cheap, and reliable stratification tool integrated into routine pathology workflows.
Emerging Liquid/Spatial Approaches (cfDNA, Spatial Transcriptomics)	cfDNA (Cell-free DNA): Analyses molecular markers (mutations, methylation patterns) in circulating tumour DNA from a blood sample. Spatial Transcriptomics (ST): Maps gene expression and cellular composition while preserving spatial location within the tissue.	cfDNA (Diagnostic): Shows high diagnostic accuracy for CRC screening (Pooled sensitivity: 0.81, Specificity: 0.93). cfDNA (CMS): Still primarily investigational for broad CMS classification; more often used for specific mutations (e.g., KRAS, BRAS). ST (Research): Used to decipher intratumoral heterogeneity.	cfDNA: Moderate/Lower per-test cost than RNA-seq. ST: Very High (Currently a research-grade technology).	cfDNA: Moderate (High compliance, non-invasive). ST: Low/Research. Critical for fundamental insights into tumor microenvironment (TME) and heterogeneity, which informs future CMS-guided therapies.

CMS and polyps

Approximately 5% of CRC cases are caused by inherited syndromes, including Lynch syndrome and familial adenomatous polyposis (FAP). The APC germline mutation in FAP-associated adenomas activates WNT/β-catenin-mediated transcription, leading to the development of precursor lesions such as tubular or villous adenomas. FAP syndrome shares molecular traits with CRC exhibiting CIN [17]. Germline mutations in MMR genes (MLH1, MSH2, MSH6, PMS2) are associated with Lynch syndrome-related CRC, with these mutations appearing in more advanced premalignant lesions. Only 50% of polyps with Lynch syndrome exhibit the MSI phenotype.

Sessile serrated adenomas (SSAs) and hyperplastic polyps (HPs) represent serrated pathways, accounting for 33% of CRC cases. SSAs are primarily right-sided and exhibit high CpG island methylation [18]. The BRAF V600E mutation is a key mechanism in both sporadic and hereditary CRC. Chang et al. applied CMS stratification to 11 GEO datasets and original cohorts of familial and sporadic CRC. Using gene-set enrichment analysis (GSEA) based on biological pathways and expression signatures associated with CRC carcinogenesis, they found that CMS1-related polyps were enriched in stromal and immune infiltration and immune cytotoxicity pathways, with strong JAK-STAT and MAPK signaling activation. WNT and MYC pathways, traditional drivers of CRC carcinogenesis, were notably enriched in CMS2 polyps. A small proportion of polyps were classified as CMS3. CMS4-like polyps, though fewer in number, showed significant mesenchymal and stromal signatures linked to TGF-β activation. CMS1 (MSI-Immune) phenotypes were observed in polyps resembling Lynch syndrome, while most FAP-related and sporadic adenomatous polyps exhibited CMS2 status. SSAs and HPs were enriched for CMS1 or CMS4 (mesenchymal-like) phenotypes [18].

Prognostic and predictive role of CMS

CMS can be used to assess prognosis. In a meta-analysis, TenHoorn et al. [19]. evaluated the clinical predictive relevance of CMS, demonstrating that CMS2 tumors were associated with the best prognosis. CMS2 tumors were linked to higher survival rates compared to CMS1 tumors in patients with mCRC. In contrast, patients with localized CRC who had CMS4 tumors exhibited poorer prognostic outcomes, including overall survival (OS), relapse-free survival (RFS), and survival after recurrence (SAR), compared to those with CMS1 or CMS2 tumors. CMS2 was associated with the longest OS, and even in cases of recurrence, CMS2 patients had significantly higher post-recurrence survival rates [20].

When cytotoxic chemotherapy is combined with targeted therapies, such as anti-angiogenic drugs, anti-EGFR antibodies, or immunotherapy, the median OS for patients with incurable mCRC can reach up to 30 months [21]. Current first-line treatment for mCRC typically involves fluoropyrimidine-based cytotoxic chemotherapy combined with a biological or targeted drug, tailored to the patient’s molecular profile [22]. To enhance the efficacy of cytotoxic drugs in mCRC, biological agents such as anti-VEGF monoclonal antibodies (e.g., bevacizumab), anti-EGFR monoclonal antibodies (e.g., cetuximab or panitumumab), anti-VEGF recombinant fusion proteins (e.g., aflibercept), or ramucirumab are often added [23].

Pembrolizumab, nivolumab, and cemiplimab, which block the PD-1 receptor, along with larotrectinib and entrectinib, which target solid tumors with NTRK (neurotrophic receptor tyrosine kinase) gene fusions, are examples of “agnostic” antibodies with histology-independent growth models recently approved for mCRC treatment. Additionally, CTLA-4 inhibitors, such as ipilimumab, which inhibit T-cell activation, have been approved for CMS1 tumors [24]. Patients with CMS4 CRC treated with irinotecan-based chemotherapy showed improved responses (survival metrics and objective response) compared to those treated with oxaliplatin-based chemotherapy. Among the four subtypes, patients with CMS1 and CMS2 tumors treated with cytotoxic chemotherapy (predominantly irinotecan-based) combined with anti-EGFR monoclonal antibodies exhibited the worst OS and the highest progression-free survival (PFS), respectively [23, 25] (Table 3).

Table 3.

Clinical trials using CMS classifiers studies focusing on role of CMS as prognostic (predicts outcome regardless of treatment) or predictive (predicts benefit from a specific treatment)

Study	Treatment Setting	Clinical Analysis Type	Sample (n)	Study Design	Major Findings	Efficacy	Platform
PETACC-3 [9] (Van Cutsem et al.)	Adjuvant	Correlative analysis of Phase III trial	688	5-FU + Leucovorin ±Irinotecan	CMS is prognostic	Prognostic: CMS status significantly predicts overall outcome (likely including Recurrence-Free Survival and OS) independent of treatment.	Custom Nanostring FFPE
NSABP C-07 [21] (Song et al.)	Adjuvant	Correlative analysis of Phase III trial	1,729	5-FU + Leucovorin ±Oxaliplatin	CMS is prognostic; CMS2 enterocyte sub-subtype linked to oxaliplatin benefit	Predictive/Prognostic: CMS status is prognostic. The CMS2 enterocyte sub-subtype showed improved efficacy (better PFS/OS) from the addition of Oxaliplatin.	Almac Xcel FFPE
PETACC-8 [22] (Laurent-Puig P.)	Adjuvant	Correlative analysis of Phase III trial	1,779	FOLFOX ±Cetuximab	CMS is prognostic	Prognostic: CMS status significantly predicts overall outcome (likely including Recurrence-Free Survival and OS).	IHC FFPE

Stinzing et al. [26] conducted a sub-analysis of the FIRE-3 trial to evaluate the effectiveness of FOLFIRI combined with bevacizumab or cetuximab in 592 patients with KRAS exon 2 wild-type mCRC. The results indicated that CMS was a prognostic factor independent of the medication used. Although there was a trend toward improved survival outcomes across all CMS subtypes in patients treated with cetuximab compared to bevacizumab, only CMS4 showed a statistically significant association in RAS wild-type mCRC. Similarly, a sub-analysis of the CALGB 80,405 phase III trial, involving 581 mCRC patients treated with FOLFOX or FOLFIRI combined with bevacizumab or cetuximab as first-line therapy, was conducted by Lenz et al. [27]. The study found that CMS subtypes were strongly predictive of survival metrics, including OS and PFS. Patients with CMS1 CRC who received bevacizumab had significantly longer OS compared to those treated with cetuximab, while the opposite was observed in the CMS2 subtype. A follow-up investigation revealed that the enterocyte subset of CMS2 (based on the CRC Assigner classifier) was sensitive to oxaliplatin, whereas other subtypes were highly resistant. In contrast, irinotecan appeared to have a greater impact on CMS4-like tumors in metastatic disease [28, 29] (Table 4).

Table 4.

Clinical studies, completed and ongoing in CRC

Study/Trial Name (NCT ID)	Status	Treatment Setting	Sample (n)	Study Population	Study Design/Regimens	CMS Subtype Focus	Primary Finding/Efficacy	Platform
Completed Correlative Trials
PETACC-3 (Van Cutsem et al.)[9]	Completed	Adjuvant	688	N/A	5-FU + Leucovorin ±Irinotecan	All CMS	Prognostic: CMS classification is prognostic for overall outcome.	Custom Nanostring FFPE
NSABP C-07 (Song et al.)[28]	Completed	Adjuvant	1,729	N/A	5-FU + Leucovorin with Oxaliplatin	All CMS, specific to CMS2	Predictive/Prognostic: CMS status is prognostic; CMS2 enterocyte sub-subtype linked to benefit from Oxaliplatin.	Almac Xcel FFPE
CALGB 80,405 (Lenz et al.) [27]	Completed	1 st line Metastatic	581	KRAS wild-type	FOLFOX with Cetuximab vs. FOLFOX with Bevacizumab	CMS1, CMS2	Predictive (OS): CMS1 had better OS with Bevacizumab; CMS2 had better OS with Cetuximab.	Custom Nanostring FFPE
FIRE-3 (Stintzing et al.)[26]	Completed	1 st line Metastatic	438	RAS wild-type	FOLFIRI withCetuthximab vs. FOLFIRI with Bevacizumab	CMS4	Predictive (OS): CMS4 showed better OS in the Cetuximab arm (in the RAS wild-type population).	Almac Xcel FFPE
CAIRO2 (Trinh et al.)[12]	Completed	1 st line Metastatic	311	All-comers	CAPOX-Bevacizumab vs. CAPOX-Bevacizumab-Cetuximab	CMS2/CMS3	Predictive (OS): CMS2/CMS3 showed improved OS when Cetuximab was added.	IHC FFPE
MAX (Mooi et al.)[30]	Completed	1 st line Metastatic	237	All-comers	Capecitabine with Mitomycin with Bevacizumab	All CMS	Predictive (PFS): CMS2/CMS3 showed better PFS with the addition of Bevacizumab; less effect observed in CMS1 and CMS4.	Almac Xcel FFPE
Japan (Okita et al.)[9]	Completed	1 st line Metastatic	193	All-comers	Oxaliplatin vs. Irinotecan regimens	CMS4	Predictive (PFS/OS): CMS4 achieved better PFS and OS with an Irinotecan-based regimen compared to an Oxaliplatin-based regimen.	Agilent FF
KEYNOTE-177 (NCT02563002)	Completed	1 st line Metastatic	307	MSI-H/dMMR	Pembrolizumab vs. Chemotherapy with Targeted Agents	CMS1 (by MSI-H/dMMR status)	Predictive/Standard of Care: Pembrolizumab (ICI) significantly prolonged PFS (16.5 mo vs. 8.2 mo) vs. chemotherapy.	N/A
EPOC1503/SCOOP (N/A)	Completed	Metastatic (Refractory)	55 (Total)	MSS/MSI-H	Napabucasin + Pembrolizumab	All CMS (exploratory)	Exploratory CMS Finding: Immune-related Objective Response Rate (ORR) was 33.3% across MS1, MS3, and MS4 subsets (CMS2: 0%) in the MSS cohort.	N/A
Ongoing/Future Prospective Strategies
Prospective Immunotherapy Trials	Ongoing/Future Direction	Metastatic	N/A	N/A	ICIs with Novel Agents (e.g., POCHI Trial)	CMS1 (Immune)	Next-Gen Immunotherapy: Trials testing ICI combinations to enhance response, even in the MSS subset with high immune infiltrate.	N/A
KRYSTAL-10 (NCT04793958)	Active, not recruiting	N/A	N/A	N/A	Adagrasib (MRTX849) plus Cetuximab	CMS3 (Metabolic)	KRAS-G12C & Metabolic Inhibitors: Targets the specific $KRAS$ G12C mutation (common in CMS3) to overcome anti-EGFR resistance. Future focus on general metabolic inhibitors.	N/A
M7824 Trial (NCT03436563)	Active	Metastatic	N/A	N/A	Anti-PD-L1/TGFbeta RII fusion protein(M7824)	CMS4 (Mesenchymal)	TME Disruption& TGF- beta Inhibition: Directly targets the TGF- $\beta$ pathway to disrupt the immunosuppressive and mesenchymal phenotype.	N/A
FoCus (NCT06191120)	Recruiting (Diagnostic)	N/A	N/A	N/A	FAPI molecular imaging	CMS4 (Mesenchymal)	FAP-Targeting: Diagnostic study to quantify Fibroblast Activation Protein (FAP) expression, paving the way for targeted FAP therapies.	N/A

Mooi et al. [30] demonstrated that the CMS classifier could guide treatment and prognosis decisions. Although standard adjuvant treatment (FOLFOX) is recommended for CMS4 patients at stage III, systemic adjuvant therapy offers no benefit for these patients. CMS4 patients with metastatic disease, regardless of KRAS mutation status, are resistant to anti-EGFR therapy [31].

The combination of oxaliplatin and a fluoropyrimidine (e.g., 5-FU or capecitabine, such as FOLFOX or CAPOX) is the standard chemotherapy regimen for CRC in the adjuvant setting, particularly for proficient mismatch repair (pMMR) high-risk stage II and stage III disease. Song et al. [28] reported that adjuvant conventional chemotherapy is beneficial only for patients with epithelial CMS2-like subtypes and not for those with CMS4-like subtypes. According to Stahler et al.‘s Xelaviri trial, CMS2 in conjunction with RAS/BRAF wild-type status was a significant predictor of overall and progression-free survival benefit after initial combinationtherapy. The study came to the conclusion that CMS2 might be used as an extra biomarker to help choose the initial combination of of drugs (fluoropyrimidine + irinotecan with bevacizumab). for patients with RAS/BRAF wild-type mCRC [32].

In CMS4, FOLFIRI outperformed FOLFOX in two separate clinical studies [33]. According to Allen et al., adjuvant chemotherapy was significantly utilized by CRC patients classified in the CMS2 subtype in both stage II and III (p = 0.02 and p < 0.001, respectively). However, only stage III patients with the CMS3 subtype benefited from drug therapy (p = 0.001) [34].

In early-stage CRC, the tumor microenvironment and cytotoxic T-cell infiltration were better predictive factors (10%) than CMS or MSI (< 5%) based on multivariate analysis of prognostic factors following pathological TNM staging [35]. Marisa et al. found that more heterogeneous tumors are associated with a worse prognosis, particularly when undifferentiated subtypes, such as CMS4 and CMS1, are present [36]. CMS specific treatment response are detailed in Table 5.

Table 5.

CMS Subtype-Specific treatment responses and therapeutic considerations [11, 19, 24, 25], 37– [44]

Characteristic	CMS1	CMS2	CMS3	CMS4
Adjuvant Chemotherapy (Stage 2/3)	No clear benefit overall. Stage 2 MSI: no adjuvant chemo. Stage 3 MSI: Responsive to FOLFOX, not 5FU monotherapy.	Most beneficial for OS. Addition of oxaliplatin to 5FU increases RFS in stage 3.	Most beneficial for OS.	No benefit from systemic adjuvant treatments.
First-line Metastatic Chemotherapy	Survival advantage with irinotecan-based regimens. Bevacizumab more effective than cetuximab.	Preferential benefit from bevacizumab with capecitabine-based chemotherapy.	Preferential benefit from bevacizumab with capecitabine-based chemotherapy.	Irinotecan-based regimens improve PFS/OS over oxaliplatin-based. More sensitive to cisplatin than CMS2.
Anti-EGFR	Survival on bevacizumab is twice that of cetuximab.	Benefit in KRAS wild-type mCRC, more effective with oxaliplatin.	Not sensitive due to high KRAS mutation frequency. May be useful if no KRAS/BRAF/PIK3CA mutations.	Resistant independent of KRAS mutation. Detrimental effect of oxaliplatin-based anti-EGFR in left-sided tumors.
Anti-VEGF	More effective than cetuximab.	PFS benefits when combined with capecitabine.	PFS benefits when combined with capecitabine.	Less effect of bevacizumab in some studies. Needs biomarker/immunotherapy integration.
Immunotherapy	Highly responsive due to MSI-H/dMMR and high neoantigen. BEACON regimen effective for BRAF-mutant mCRC. PD-1/PD-L1 inhibitors (pembrolizumab, nivolumab, cemiplimab). CTLA-4 inhibitors (ipilimumab).	Not primary target, considered “cold tumors.”	Not primary target.	New approaches to enhance immunotherapy role are being explored.
Emerging Strategies	CAR-T cell therapy, TME modulation, napabucasin (STAT3 inhibitor) – EPOC1503/SCOOP trial.	Ribosomal Modulating Agents (RMAs) – ZKN-157, which selectively inhibits protein translation in cells with high MYC & WNT activity.	Metabolic inhibitors (e.g., CBS inhibitors) for KRAS-mutant CRC. HER2 TKIs for HER2 overexpression.	TME targeting (hypoxia, ECM, CAFs, immunosuppression). HIF inhibitors, HAPs, LOX inhibitors, FAP targeting, EMT inhibitors, novel immunotherapies (BiTEs, CAR T-cells, TAM-directed strategies).

(CBS Cystathionine β-synthase, TME Tumor Microenvironment, HIF Hypoxia-Inducible Factors, HAPs 4-Hydroxyacetophenone)

CMS and immunotherapy

Patients with CRC have benefited from novel therapeutic approaches over the years. Although checkpoint inhibitors have produced remarkable results in mCRC, they are primarily effective in patients with MSI-high (MSI-H) or dMMR tumors [45]. CMS subtypes are associated with specific immune infiltration characteristics and immune escape mechanisms, reflecting the latest molecular characterization of CRC [46].

Thorsson et al. [47] analyzed gene expression patterns from over 10,000 individuals with TCGA cancers across 33 non-hematological types, developing a comprehensive immunological classification of solid tumors, including CRC. They identified six immunological subtypes (ISs):

• C1: High expression of angiogenic genes.

• C2: IFN-γ dominant.

• C3: High Th17.

• C4: Lymphocyte depleted.

• C5: Poorest lymphocyte and highest macrophage responses.

• C6: TGF-β dominant.

Soldevilla et al. found that C1 and C2 were prevalent across all CMS subgroups, with C1 most prevalent in CMS2 (91%) and C2 in CMS1 (53%). Other immune subtypes were sporadically identified in CMS1 and CMS2. C6 was exclusively present in CMS4, while CMS3 and CMS4 had distinct immunological landscapes enriched in C3 and C4. IS was a predictor of survival independent of other prognostic markers, such as age, stage at diagnosis, or primary tumor site [48].

As the majority of MSI/highly mutated cancers are CMS1 (MSI-like immune), these patients are considered the most likely to benefit from immunotherapy. However, not all CMS1 tumors are MSI, and a significant fraction of MSI tumors are found in other CMS subgroups (e.g., CMS3). MSI is the only biomarker routinely used in clinical practice to identify immunogenic CRC and predict the success of checkpoint inhibitors. Soldevilla et al. noted that CMS1’s unique immunological landscape may influence treatment efficacy [48] (Fig. 2).

Fig. 2.

Illustrating the pathways in CMS1 and the drugs available to treat with targets

Early-stage CMS1 tumors have a very good prognosis because of their signature hypermutation/MSI-H, which produces many neoantigens and makes them extremely sensitive to ICIs. The aggressiveness of MSI-H, which overwhelms the host immune response at distant sites, and certain BRAF mutations may be the cause of their extremely poor prognosis despite this initial favorable outlook after relapse or metastasis [49].

Due to faulty DNA proofreading, POLE-mutant colorectal tumors (POLE-CRCs) have an ultramutated phenotype with a very high tumor mutation burden (TMB). Despite frequently being microsatellite-stable (MSS), POLE-CRCs are highly immunogenic and functionally comparable to the CMS1 subtype because of their high TMB. This connection is important because POLE-CRCs show increased sensitivity to Immune Checkpoint Inhibitors (ICIs), similar to CMS1/MSI-H tumors, undermining the conventional supremacy of MSI status as the only predictor for immunotherapy in CRC. In order to effectively classify these patients for the best individualized treatment, deep learning methods are now being developed to detect POLE mutations based solely on histopathological scans [50, 51].

A strategy for choosing patients to receive TROP2-targeted antibody-drug conjugates (ADCs) is provided by CMS categorization. Microsatellite stable (MSS) cancers and an aggressive phenotype are typically linked to the target TROP2. The CMS system can theoretically target TROP2-ADC therapy on MSS tumors while de-prioritizing CMS1 (MSI-H) cancers because CMS2, CMS3, and CMS4 are the most common MSS subtypes (representing 75% of CRCs). Specifically, considering CMS4’s poor prognosis and strong resistance to traditional therapies, the use of TROP2-ADCs such as sacituzumab govitecan is a viable approach [52].

The main therapeutic approach for CMS1 (Immune) centers on its hypermutated phenotype and robust immune activation. These tumors are very susceptible to Immune Checkpoint Inhibitors (ICIs), such as Pembrolizumab (anti-PD-1) and Nivolumab (anti-PD-1), due to the high burden of neoantigens. ICIs “release the brakes” on the pre-existing anti-tumor immune response by blocking inhibitory molecules such as PD-1/PD-L1, according to the molecular reasoning. The accepted standard of therapy for MSI-H metastatic colorectal cancer (mCRC) is this method, which has a lot in common with CMS1. Since the BRAF-V600E mutation is frequently found in CMS1, the BEACON regimen—which consists of cetuximab, binimetinib, and encorafenib—has also demonstrated efficacy in treating BRAF-mutant mCRC. Adjuvant chemotherapy is not clearly linked to improved overall survival in CMS1 malignancies as compared to standard chemotherapy ([53]– [54]).

The CMS2 “canonical” subtype, comprising 37% of CRC cases, consists of CIN tumors with epithelial differentiation markers and upregulated WNT and MYC signaling pathways. CMS2 exhibits the fewest MSI tumors among the four CMS groups [32]. Described as an “immune desert,” CMS2 tumors have few immune cells, primarily naïve CD4 T cells, B cells, or resting NK cells, which are unable to facilitate active anti-tumor immunity [3, 54] (Fig. 3).

Fig. 3.

Illustrating wnt/APC pathway followed by these tumors with available drugs and their targets

The two main pathways targeted by treatment approaches for CMS2 (Canonical) are WNT/MYC Signaling and EGFR Signaling. Because EGFR and its ligands (AREG/EREG) are highly expressed in CMS2 tumours, EGFR-targeting drugs like cetuximab and panitumumab (anti-EGFR medicines) work well. In KRAS wild-type CMS2 metastatic colorectal cancer (mCRC), the anti-EGFR advantage is most noticeable, and its effectiveness is greatly increased when paired with an oxaliplatin chemotherapy backbone. The significant activation of the WNT and MYC pathways is the other essential feature of this subtype. As a natural next step for therapeutic research, experimental medicines such as WNT/MYC inhibitors (e.g., ZKN-157) are being developed to directly target these major active carcinogenic pathways [52, 54].

The CMS3 “metabolic” subtype, representing 13% of CRC cases, includes more MSI tumors (16% of CMS3) and frequent KRAS mutations. Its immune landscape is “immune excluded,” with a dormant immune microenvironment, low infiltration of monocytes, lymphocytes, and myeloid cells, but higher content of Th17 cells, T cells, PD-1, naïve B cells, and resting T cells [37, 55] (Fig. 4).

Fig. 4.

Illustrating RAS/RAF/MEK and PI3K/AKT pathway followed by CMS3 tumors with available drugs and their targets

The targeted solutions for CMS3 (Metabolic) are mostly focused on tackling its two main features: metabolic dysregulation and the high frequency of the KRAS mutation. Although KRAS mutations have historically resulted in resistance to anti-EGFR therapy, KRAS-G12C inhibitors are novel experimental drugs being investigated to directly target this extremely widespread mutation and overcome resistance. The frequent KRAS-G12C mutation is a crucial target. At the same time, metabolic inhibitors like CBS inhibitors are being investigated due to the subtype’s distinctive metabolic dysregulation, which includes glutaminolysis and lipidogenesis. addresses KRAS-dependent resistance by focusing on the underlying metabolic processes that characterize CMS3 and promote chemoresistance. HER2 Tyrosine Kinase Inhibitors (TKIs), such as dacomitinib and neratinib, can be used to treat HER2 overexpression, possibly in conjunction with trastuzumab [20, 53].

CMS4 tumors are characterized by epithelial-mesenchymal transition (EMT), TGF-β signaling activation, and angiogenesis traits. Compared to CMS1, CMS4 tumors exhibit more regulatory T cells (Tregs) and fewer CD8 and CD4 T cells due to their inflammatory profile, marked by accumulation of complement components, high macrophage numbers, and infiltrating lymphocytes [3] (Fig. 4).

It was proposed that imatinib, a PDGFR/KIT inhibitor, might be useful as a CMS4-targeting medication that may reduce the aggressive mesenchymal features and possibly change the tumor to a less aggressive CMS2-like phenotype in primary colon cancer in a small proof of concept study involving five patients [55] (Fig. 5).

Fig. 5.

Illustrating TGF and BMP pathway followed by CMS4 tumors with available drugs and their targets

Clinical translation

1. Immunogenicity in CMS1 (Immune)

   Justification: CMS1 has a significant burden of neoantigens due to its hypermutated phenotypic MSI-H/dMMR status.

   Therapy: Immune Checkpoint Inhibitors (ICIs) work well in this situation. Anti-PD-1 monotherapy was made the conventional first-line treatment for MSI-H metastatic colorectal cancer (CRC) by trials such as KEYNOTE-177 (pembrolizumab). This effect is further enhanced by dual checkpoint blockage (e.g., Nivolumab + Ipilimumab in CheckMate-142).

   Result: Very effective, producing long-lasting reactions [24].

2. Targeting Oncogenic Addiction in CMS2 (Canonical)

   Justification: WNT and MYC signaling activation, which promotes proliferation and an addiction to high protein translation, are characteristics of CMS2.

   Therapy: Although CMS2 is sensitive to anti-EGFR treatment when paired with an oxaliplatin backbone, new drugs target the protein’s primary carcinogens. The development of investigational drugs such as the Ribosome Modulating Agent (RMA) ZKN-157 aims to specifically block translation pathways that are increased by MYC [54].

3. Overcoming KRAS Resistance in CMS3 (Metabolic)

   Justification: The metabolic abnormalities and high frequency of KRAS mutations (up to 73.9%) that give resistance to anti-EGFR therapies characterize CMS3.

   Treatment: To combat EGFR feedback activation, new KRAS-G12C inhibitors are being developed, such as Adagrasib in the KRYSTAL-10 trial (NCT04793958), which is frequently used in conjunction with cetuximab. Metabolic modulators will be investigated in future research. To combat KRAS-dependent chemoresistance, use CBS inhibitors.

4. Changing the Protective Microenvironment CMS4 (Mesenchymal)

5. Justification: Based on its extremely active TME (high CAFs, ECM, hypoxia, TGF-beta activation), which serves as a barrier to medications and immune cells, CMS4 is the most treatment-resistant subtype.

   Therapy: Techniques that focus on TME regulation TGF-beta Inhibition: To reverse immunosuppression and T-cell exclusion, trials such as M7824 NCT03436563 specifically target the TGF-beta system.

   CAF Targeting: Because FAP is highly expressed on CAFs, FAP-targeting diagnostics (NCT06191120) and treatments are essential.

   Chemotherapy: The preferred cytotoxic backbone over oxaliplatin is regimens based on irinotecan [20, 52] (Fig. 6).

Fig. 6.

Image showing integrated CMS based targeting of CRC

CMS and intratumoral heterogeneity (ITH)

Intratumoral heterogeneity (ITH) is a widespread issue in CRC, with more than 55% of tumors containing combinations of at least two CMS subtypes. This suggests that a single biopsy may not fully capture a tumor’s CMS profile, potentially leading to misclassification or incomplete understanding of the tumor’s biological makeup. Discordant CMS profiles are often observed between primary tumors and paired metastases. This intrinsic heterogeneity within and between tumor sites complicates CMS-targeted therapeutic strategies, as a drug effective for one CMS component may be ineffective for another within the same tumor or at a metastatic site. ITH is associated with lower overall survival (OS) and disease-free survival (DFS), with hazard ratios of 1.34 (95% CI: 1.12–1.59) for DFS and 1.40 (95% CI: 1.14–1.71) for OS [55, 56].

Methods and research directions to address ITH include:

1. Single-Cell RNA Sequencing (scRNA-seq): Enables transcriptional profiling of individual cells within a tumor [57].

2. Weighted In Silico Pathology (WISP) Algorithm: Quantifies multiple CMS signatures within a single bulk sample, providing a better understanding of the tumor’s molecular landscape [55]. WISP precisely measures the mixing proportions of the four CMS subtypes in a single tumor sample in colorectal cancer (CRC). This helps to improve patient prognosis and better direct individualized treatment plans by addressing the clinical problem of ITH, where more than 55% of tumors are combinations of CMS subtypes.

3. Spatial Transcriptomics: Bridges bulk RNA sequencing and single-cell sequencing, demonstrating interactions between CMS populations and the tumor microenvironment, such as immune (CMS1) and stromal (CMS4) cells [58].

4. Image-based CMS (imCMS): Uses deep learning algorithms to predict CMS subtypes from H&E-stained tissue sections, identifying regions within the same slide corresponding to different CMS subtypes [4].

5. Circulating Tumor DNA (ctDNA): Enables real-time monitoring of tumor changes and clonal evolution, reflecting the dynamic nature of ITH. While it does not directly classify CMS, it can reveal emerging resistance mechanisms or shifts in dominant molecular features, indirectly reflecting changes in CMS-like characteristics over time or under therapy-induced selective pressure [59].

Colorectal cancer (CRC) has an unusually high degree of morphologic heterogeneity as well; on routine slides, the majority of tumors exhibit numerous unique morphotypes. The ratios of particular morphotypes present, rather than heterogeneity per se, dictate prognosis. Aggressive disease is indicated by a large percentage of the desmoplastic (DE) morphotype, which is associated with an advanced stage and a shorter survival time (OS/RFS). On the other hand, the papillary (PP) morphotype is associated with a better prognosis and earlier stages. According to these findings, in order to accurately predict clinical risk, morphotypes must be quantified by AI-based analysis and comprehensive tumor collection [60].

Barriers to CMS adoption in clinical practice

Despite providing a comprehensive understanding of CRC biology with implications for prognosis and personalized treatment, technical and financial barriers hinder the widespread clinical adoption of CMS classification. Technical challenges include:

1. The complexity of gene expression analysis, which requires specialized expertise and is not readily available in standard pathology labs, along with lengthy turnaround times for bioinformatics analyses that may delay critical treatment decisions.

2. Sample quality issues, as formalin-fixed paraffin-embedded (FFPE) tissues, commonly used in clinical practice, often yield degraded RNA.

3. Significant molecular intratumoral heterogeneity and the lack of a universally accepted standardized methodology for classification across platforms and laboratories.

4. The need for advanced molecular biology laboratories equipped with RNA sequencing and specific gene panel assays.

5. The requirement for further clinical trials to validate CMS utility in diverse patient populations.

6. High upfront costs for equipment, reagents, and tests involving RNA extraction, sequencing/array, and bioinformatics analysis [3, 61–65].

Potential solutions include using tailored gene panels (e.g., Nanostring-based assays or custom PCR panels) to reduce technological complexity, turnaround time, and costs. A “40-gene ColoType signature” has been developed using frozen tissue-specific RNA sequences or FFPE, alongside genome-wide tests. Research is ongoing to create IHC-based panels as surrogate markers for CMS classification, which are widely accessible, cost-effective, and require less expertise. Centralized complex gene expression analyses in specialized reference laboratories, globally accepted guidelines, and established bioinformatic protocols could minimize resource utilization and reduce per-test costs. Extensive prospective clinical studies are needed to demonstrate CMS classification’s clinical relevance and identify scenarios where it provides the greatest prognostic or predictive value. Advances in sequencing technologies and bioinformatics tools are essential to reduce costs and turnaround times [4, 66–69].

Challenges

Although CMS classification is currently the most robust CRC categorization method, it faces several challenges and criticisms. The original sample size included fewer stage IV tumors, suggesting underrepresentation of advanced disease biology. Additionally, rectal cancers were less represented and often treated with neoadjuvant therapies, limiting the generalizability of findings to rectal cancers [70].

While some genomic and epigenomic markers are enriched in CMS groups, their associations are weak, limiting the ability to classify gene expression subtypes accurately. This supports the notion that transcriptional fingerprints enable more precise disease subclassification than current biomarkers. For example, RAS wild-type tumors, despite being present across multiple CMS groups with significant biological differences, are treated as a homogeneous entity in advanced therapeutic settings, despite expected variations in treatment response.

Oversimplifying CRC biology into four CMS groups may hinder the translation of biological insights into clinical and therapeutic applications. Intra-tumor heterogeneity is evident in 57% of samples, which contain multiple subclones or more than one CMS subtype with a weight above 20% [36]. The tumor center and invasive fronts of the same lesion exhibit distinct CMS characteristics, and this heterogeneity extends to primary tumors and their metastases. CMS4 weightage increases as early-stage CRC progresses to metastatic disease, while other subtype clones diminish. Thus, CMS classifiers, while valuable, face technical and heterogeneity challenges that are as critical as the disease context itself.

Conclusions

The CMS classification system represents a significant advancement in understanding CRC’s molecular heterogeneity, offering a framework to predict prognosis and tailor therapies. CMS1 tumors benefit from immunotherapy due to their immune-activated profile, while CMS2 tumors show superior survival with oxaliplatin-based regimens. CMS3 and CMS4 present unique challenges, with CMS4’s poor prognosis and resistance to anti-EGFR therapies highlighting the need for novel strategies. Intratumoral heterogeneity complicates accurate CMS classification, necessitating advanced techniques like single-cell RNA sequencing and spatial transcriptomics. Despite its potential, CMS adoption in clinical practice is hindered by technical complexity, high costs, and lack of standardized methodologies. Simplified approaches, such as IHC-based panels and centralized molecular testing, could overcome these barriers. Future research should focus on validating CMS in diverse populations and integrating it into clinical algorithms to optimize personalized treatment, ultimately improving survival and quality of life for CRC patients.

Authors’ contributions

AK: literature review, study selection, writing of draft manuscriptDD: literature review, writing of draft manuscript, illustrationsEP: literature review, study selection, writing of draft manuscriptTK: Concept and design, visualization, writing of draft manuscriptMP: Concept and design, visualization, validation, supervision, editing of final manuscriptAll authors read and approved the final manuscript.

Funding

None.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not required.

Competing interests

The authors declare no competing interests.