A transcriptomic pan-cancer signature for survival prognostication and prediction of immunotherapy response based on endothelial senescence.

BACKGROUND

The microvascular endothelium inherently controls nutrient delivery, oxygen supply, and immune surveillance of malignant tumors, thus representing both biological prerequisite and therapeutic vulnerability in cancer. Recently, cellular senescence emerged as a fundamental characteristic of solid malignancies. In particular, tumor endothelial cells have been reported to acquire a senescence-associated secretory phenotype, which is characterized by a pro-inflammatory transcriptional program, eventually promoting tumor growth and formation of distant metastases. We therefore hypothesize that senescence of tumor endothelial cells (TEC) represents a promising target for survival prognostication and prediction of immunotherapy efficacy in precision oncology.

METHODS

Published single-cell RNA sequencing datasets of different cancer entities were analyzed for cell-specific senescence, before generating a pan-cancer endothelial senescence-related transcriptomic signature termed EC.SENESCENCE.SIG. Utilizing this signature, machine learning algorithms were employed to construct survival prognostication and immunotherapy response prediction models. Machine learning-based feature selection algorithms were applied to select key genes as prognostic biomarkers.

RESULTS

Our analyses in published transcriptomic datasets indicate that in a variety of cancers, endothelial cells exhibit the highest cellular senescence as compared to tumor cells or other cells in the vascular compartment of malignant tumors. Based on these findings, we developed a TEC-associated, senescence-related transcriptomic signature (EC.SENESCENCE.SIG) that positively correlates with pro-tumorigenic signaling, tumor-promoting dysbalance of immune cell responses, and impaired patient survival across multiple cancer entities. Combining clinical patient data with a risk score computed from EC.SENESCENCE.SIG, a nomogram model was constructed that enhanced the accuracy of clinical survival prognostication. Towards clinical application, we identified three genes as pan-cancer biomarkers for survival probability estimation. As therapeutic perspective, a machine learning model constructed on EC.SENESCENCE.SIG provided superior pan-cancer prediction for immunotherapy response than previously published transcriptomic models.

CONCLUSIONS

We here established a pan-cancer transcriptomic signature for survival prognostication and prediction of immunotherapy response based on endothelial senescence.