This study proposes four consensus molecular subtypes (ECMS) for esophageal squamous cell carcinoma (ESCC) and develops imECMS, a deep learning classifier based on H&E images, enabling accurate prediction of prognosis and treatment response, thereby advancing the development of personalized therapy for ESCC.
Literature Overview
This article, 'The consensus molecular subtypes of esophageal squamous cell carcinoma,' published in the journal Signal Transduction and Targeted Therapy, reviews and summarizes the complexity of molecular heterogeneity in esophageal squamous cell carcinoma (ESCC). By systematically integrating eight existing molecular classification systems, the study defines four consensus molecular subtypes (ECMS) with distinct biological characteristics and clinical relevance through network analysis. Furthermore, the research develops imECMS, a deep learning classification tool based on H&E whole-slide images, enabling precise molecular subtyping without high-throughput sequencing, significantly enhancing the clinical accessibility of this classification system. This work provides a robust framework for precise classification, prognosis assessment, and personalized treatment of ESCC. The findings have been validated across multiple independent cohorts, demonstrating high reproducibility and clinical translational potential.Background Knowledge
Esophageal squamous cell carcinoma (ESCC) is a highly prevalent malignant tumor worldwide, particularly in East Asia, where its incidence remains high. The prognosis is extremely poor, with a five-year survival rate of less than 20%. Although recent multi-omics studies have revealed the molecular heterogeneity of ESCC and proposed various molecular classification systems, most of these systems are based on single-omics data and show significant inconsistencies across studies, severely limiting their clinical application. The lack of a unified, standardized molecular classification system results in imprecise treatment strategies, making individualized therapy difficult for patients. Additionally, the high cost of omics sequencing hinders the widespread adoption of molecular subtyping in routine pathology. Therefore, establishing a robust, reproducible, and clinically feasible consensus molecular classification system has become a critical unmet need. In recent years, the successful application of deep learning in digital pathology has provided new opportunities to extract deep molecular information from routine H&E images. This study, set against this backdrop, aims to develop a novel classification framework that combines biological depth with clinical utility by integrating multi-omics data and pathological images, overcoming the limitations of existing classification systems and providing new tools for precision medicine in ESCC.
Research Methods and Experiments
The research team first collected a multi-omics cohort (SXM-I) comprising 152 ESCC patients, integrating genomic, transcriptomic, and methylomic data. Using the Similarity Network Fusion (SNF) method for multi-omics clustering, they defined four multi-omics subtypes (MESCC). Subsequently, they applied network clustering to integrate eight existing ESCC classification systems—including MESCC—constructing a subtype association network and ultimately identifying four consensus molecular subtypes (ECMS1–4) using the Markov Clustering Algorithm (MCL). Based on differentially expressed genes, a random forest classifier was developed for transcriptomic data classification. Furthermore, the team employed deep learning to build the ESCC-SPA tissue segmentation model, which automatically identifies regions in H&E whole-slide images and extracts multi-scale spatial tissue features (SOFs). Using these SOFs, an image classifier named imECMS was developed via the extremely randomized trees algorithm to predict ECMS subtypes. The classifier was trained, validated, and prospectively tested across multiple independent cohorts to evaluate its concordance with molecular subtypes, prognostic value, and ability to predict treatment response.Key Conclusions and Perspectives
Research Significance and Prospects
The greatest significance of this study lies in providing a standardized, reproducible molecular classification framework (ECMS) for ESCC, resolving the inconsistency among previous classification systems and laying the foundation for cross-study comparisons and clinical applications. It not only deepens our understanding of ESCC’s molecular heterogeneity but also offers potential therapeutic strategies for different subtypes—for example, ECMS3-IM may benefit more from immunotherapy, while ECMS2-CLS may be sensitive to CDK inhibitors.
More importantly, the development of the imECMS classifier represents a groundbreaking contribution. It successfully translates complex molecular subtyping into routine pathological diagnosis based on standard H&E images, greatly reducing the cost and technical barriers to precision subtyping, making widespread adoption across healthcare institutions feasible. This provides a practical tool toward achieving precision medicine in ESCC. Future studies should focus on validating imECMS performance in larger, multicenter prospective cohorts and exploring its real-world value in guiding clinical treatment decisions.
Conclusion
This study establishes four consensus molecular subtypes (ECMS) of esophageal squamous cell carcinoma through systematic integration of multiple molecular classification systems, providing a unified biological framework for understanding its complex tumor heterogeneity. Each subtype possesses distinct molecular features and microenvironment landscapes, and is associated with specific clinical outcomes and treatment responses, offering crucial guidance for personalized therapeutic strategies. Another major highlight is the development of the deep learning–based imECMS classifier, which enables accurate prediction of molecular subtypes directly from routine H&E pathology images, overcoming the clinical accessibility limitations imposed by costly omics testing. With high accuracy, low cost, and ease of use, imECMS holds strong promise for clinical translation and could become a standard component of future ESCC pathological diagnosis. Overall, the ECMS/imECMS system provides the most robust and practical framework to date for precise subtyping, prognosis evaluation, and treatment selection in ESCC, marking a critical step toward true precision medicine, with profound clinical significance and broad application potential.
