Fibroblast Gene Signature Predicts Survival in Colorectal Cancer

Seven-gene, fibroblast-focused panel improves risk stratification and may guide therapy choices

Researchers report a new fibroblast-derived prognostic signature that stratifies patients with colorectal cancer (CRC) by survival risk and enhances the performance of standard clinical models.

About the disease

Colorectal cancer is among the most common cancers worldwide and a leading cause of cancer-related deaths. While surgery, chemotherapy and targeted therapies have advanced care, outcomes for patients diagnosed at advanced stages remain poor. Tumor heterogeneity, variable therapy response and emerging drug resistance are key barriers—shifting attention from cancer cells alone to the tumor microenvironment (TME), where fibroblasts play an outsized role.

Why fibroblasts?

Cancer-associated fibroblasts (CAFs) release growth factors and cytokines, remodel the extracellular matrix and promote spread. Single-cell studies show fibroblasts frequently dominate cell-to-cell communication in solid tumors. Building on this, the team asked a translational question: Can fibroblast-specific genes outperform conventional models in predicting patient outcomes?

Study at a glance

• 
  Samples & data: 306 tumor-core specimens; 448,255 single cells profiled by scRNA-seq
• 
  Network insight (CellChat): fibroblasts identified as principal signaling hubs in CRC tissue
• 
  Feature pool: 435 fibroblast marker genes
• 
  Machine-learning pipeline: 101 model combinations evaluated with leave-one-out cross-validation
• 
  Final 7-gene signature: CSRP2, DBN1, FSTL3, GPX3, PAM, RGS16, CXCL14

Performance

• 
  Risk discrimination: C-index 0.65 (TCGA-COAD, n = 351) and 0.63 (GSE17536, n = 177)
• 
  Hazard separation: high-risk patients show ~2.4× greater hazard of death in both datasets
• 
  Time-dependent accuracy: AUC ~0.65–0.68 at 1, 3 and 5 years
• 
  Added clinical value: integrating the signature into a nomogram with routine variables boosts overall C-index to 0.81

Biological interpretation

• 
  Fibroblasts emit the highest number and strength of outgoing signals across cell types, reinforcing regulatory dominance.
• 
  Signature-linked genes are enriched for cytokine–receptor and chemokine signaling pathways.
• 
  High-risk tumors show increased macrophage and NK-cell infiltration alongside T-cell dysfunction and immune evasion—an immune landscape consistent with poorer outcomes.

Therapeutic signals to explore

• 
  Topoisomerase-I inhibitors: modelled sensitivity is higher in low-risk patients (e.g., camptothecin/irinotecan).
• 
  Immunotherapy context: high-risk profiles correlate with greater stromal content and immunosuppression—features often linked to reduced checkpoint blockade efficacy.
  Note: These insights are computational and require prospective clinical testing.

Why this study matters

• 
  Focus on the stroma: a concise, fibroblast-centric panel matches or exceeds larger, tumor-cell-centric signatures.
• 
  Single-cell grounding: deconvolution of ~448k cells yields a biologically coherent feature set for model building.
• 
  Clinical integration: remains an independent predictor and improves existing prognostic tools.
• 
  Actionable hypotheses: supports treatment stratification concepts for chemotherapy and immunotherapy.

Practical implications and next steps

• 
  Laboratory use: the 7-gene panel is small enough for qPCR or robust multiplex assays.
• 
  Clinical use: incorporation into standard nomograms could refine decisions around adjuvant chemotherapy or trial enrollment.
• 
  Future work: functional validation (in vitro and in vivo) to test causality and whether modulating fibroblast activity alters drug response; prospective, multi-center validation to confirm generalizability.

Conclusion

Centering analysis on fibroblasts delivers a scientifically grounded, clinically relevant seven-gene prognostic tool for CRC. By translating single-cell insights into a practical assay, this approach has the potential to outperform traditional metrics and inform more effective, personalized treatment decisions.

More information:

Ning Zhang et al, Machine learning integration of single-cell and bulk transcriptomics identifies fibroblast-driven prognostic markers in colorectal cancer. Biomol Biomed [Internet]. 2025 Apr. 22 [cited 2025 Oct. 29];26(1):115–131.

Available from: https://doi.org/10.17305/bb.2025.12038

Journal information: Biomolecules and Biomedicine

Provided by: Association of Basic Medical Sciences of FBIH

Provided by Association of Basic Medical Sciences of FBIH