Can the tumor transcriptomic profile predict immunotherapy efficacy?

  The emergence of immunotherapies, particularly immune checkpoint inhibitors (ICIs), has revolutionized the management of many cancers. However, only a minority of patients derive durable benefit. Current biomarkers such as PD-L1 expression, tumor mutational burden (TMB), and microsatellite instability (MSI) have significant limitations in sensitivity and specificity. To overcome these challenges, large-scale transcriptomic approaches offer a new opportunity to precisely characterize tumor immune phenotypes. This study proposes and evaluates pan-cancer transcriptomic classifiers capable of predicting response to ICIs, based on the integrative analysis of multiple databases and clinical trials.   Can transcriptomic signatures outperform PD-L1 and TMB? The authors analyzed 47 gene expression signatures across more than 10,000 tumors from 33 cancer types using data from The Cancer Genome Atlas (TCGA). These signatures were grouped into functional categories (adaptive immunity, inflammation, angiogenesis, etc.) and integrated into supervised models using machine learning algorithms such as random forest and elastic net regression. Three main classifiers were developed:

• Pan-Immune Classifier (PIC): based on 11 signatures related to immune activation.
• Tumor Microenvironment Classifier (TMC): based on 14 signatures related to stroma, angiogenesis, and metabolism.
• Combined Classifier (PIC + TMC): an integrated approach capturing both immune and non-immune dimensions of the tumor ecosystem.

These models were tested and validated in seven independent clinical cohorts treated with ICIs (including MCC, BLCA, NSCLC, melanoma, etc.), comprising over 1,500 patients. Both the PIC and the combined model demonstrated superior predictive performance compared to PD-L1, TMB, and classical IFNγ-based classification, achieving an area under the curve (AUC) of up to 0.79 in some cohorts. For instance, in the IMvigor210 cohort (urothelial carcinoma), patients classified as “high PIC” had an objective response rate (ORR) of 53%, compared with 12% in the “low PIC” group. Similar trends were observed in the MCC cohort, where highly classified patients experienced a clear improvement in overall survival. The study also showed that these classifiers were robust to technical variability and could be applied to standard RNA-seq data from FFPE samples, reinforcing their clinical applicability. Furthermore, the analyses revealed that the models capture signals beyond the mere presence of immune cells, incorporating the functional state of the tumor microenvironment.   Toward a universal transcriptomic score to guide immunotherapy? The findings suggest that multi-signature transcriptomic classifiers provide a finer stratification of patients eligible for immunotherapy than traditional biomarkers. Integrating the PIC, TMC, and their combined model into clinical decision-making algorithms could transform patient selection, preventing unnecessary, costly, and ineffective treatments. However, several challenges remain: these models need prospective validation, analytical standardization, and integration into clinically accessible platforms. Further research is also required to combine transcriptomic data with genomic and imaging biomarkers, paving the way for multimodal predictive models. In summary, this study represents a major step toward precision immuno-oncology, enabling clinicians to predict, select, and tailor immunotherapeutic treatments based on the patient’s global transcript