Immunotherapy - particularly immune checkpoint inhibitors that unleash the body's T cells to attack cancer - has revolutionized treatment for many cancers. However, pancreatic cancer remains largely resistant to immunotherapy, with response rates below 5% in most clinical trials.
A major reason for this resistance is that pancreatic tumors undergo extensive metabolic reprogramming - they rewire their energy metabolism in ways that simultaneously fuel their own growth and create a hostile, immunosuppressive microenvironment. This metabolic transformation depletes nutrients that immune cells need and produces toxic byproducts that paralyze T cells.
The researchers hypothesized that characterizing the metabolic landscape of pancreatic cancer systematically could reveal biomarkers that predict immunotherapy response and identify patients most likely to benefit - a critical unmet need in personalizing pancreatic cancer treatment.
The researchers characterized the metabolic landscape of pancreatic cancer across 112 distinct metabolic pathways, using transcriptomic data from 1,188 patients with pancreatic cancer. This large, multi-dataset approach was designed to capture the full diversity of metabolic states across different patient populations.
A metabolism-related signature (MBS) was constructed using machine learning integration of multiple datasets - a consensus approach that combines results across different algorithms to produce a robust, reproducible model. This 'consensus' methodology reduces the risk of overfitting to any single dataset.
The predictive performance of MBS was evaluated across 11 independent immunotherapy cohorts, examining both bulk RNA sequencing data and single-cell RNA sequencing data. The model was further validated using laboratory experiments including immunohistochemistry, western blotting, and colony formation assays.
MBS demonstrated strong predictive power for immunotherapy response, outperforming 66 previously published cancer signatures in head-to-head comparisons. High MBS scores were associated with poor immunotherapy response, marking patients unlikely to benefit from checkpoint inhibitor treatment.
Mechanistically, high MBS was found to suppress anti-tumor immunity while simultaneously promoting cancer stemness (the ability of cancer cells to regenerate and resist treatment) and intratumoral heterogeneity (diversity within the tumor that helps it adapt to therapies).
MBS was also identified as a robust prognostic biomarker independent of immunotherapy, with high-MBS patients showing significantly shorter overall survival. This dual utility - predicting both response to immunotherapy and overall prognosis - makes MBS particularly clinically valuable.
For patients classified as high-MBS (and therefore unlikely to benefit from standard immunotherapy), the study used computational drug repurposing to identify alternative therapeutic options. This approach screens thousands of existing drugs for those whose mechanism of action could counteract the high-MBS metabolic state.
Two drugs emerged as promising candidates for high-MBS pancreatic cancer patients: dasatinib (a tyrosine kinase inhibitor originally approved for leukemia) and epothilone B (a microtubule stabilizer). Laboratory experiments confirmed that both drugs showed superior activity against high-MBS cancer cells compared to standard therapies.
This personalized approach - using a biomarker to both predict who won't respond to immunotherapy AND to identify what alternative drug they should receive instead - represents a more complete clinical solution than simply predicting treatment failure without offering an alternative.
The MBS framework could be implemented as a molecular test on tumor tissue or potentially blood samples to stratify patients before treatment initiation. Low-MBS patients would be prioritized for immunotherapy trials, while high-MBS patients would be directed toward metabolism-targeting drugs or the identified alternatives.
The single-cell validation of MBS adds an important dimension - it demonstrates that the metabolic signature captures real heterogeneity within tumors at the level of individual cells, not just average tumor gene expression. This suggests MBS reflects genuine tumor biology rather than technical artifact.
The study opens a new paradigm: targeting tumor metabolism not just as a way to starve cancer cells, but as a strategy to simultaneously reverse immunosuppression and make tumors more responsive to immune-based therapies. Combining metabolism-targeting drugs with immunotherapy could potentially break the immunotherapy resistance barrier in pancreatic cancer.
This study establishes tumor metabolism as a master regulator of immunotherapy response in pancreatic cancer and provides a validated biomarker (MBS) to guide treatment decisions. The consensus machine learning approach produces a more robust and generalizable signature than single-algorithm methods.
The identification of specific therapeutic alternatives for MBS-high patients transforms MBS from a predictive test into a treatment guidance tool. This companion diagnostic concept is increasingly important in oncology as personalized treatment becomes the standard of care.
Future work should include prospective clinical trials using MBS to select patients for immunotherapy versus alternative drug regimens, with overall survival as the primary endpoint. Integration of MBS with other biomarkers such as tumor mutational burden and microsatellite instability could further refine patient selection.