Identification and verification of immunogenic cell death-related signatures derived from bone metastasis of prostate cancer based on multi-omics and machine learning algorithms
TL;DR
This model provides a valuable tool for characterizing TIME in metastatic PCa and demonstrates significant clinical utility for prognostic assessment and prediction of immunotherapy response and demonstrates robust prognostic accuracy for both the clinical outcomes and pathological features.
Abstract
Background
Prostate cancer (PCa) is a prevalent urological malignancy in men, with bone metastasis occurring in the majority of patients, often leading to poor clinical outcomes. Despite its clinical significance, reliable prognostic biomarkers for metastatic PCa remain limited. Immunogenic cell death (ICD), a regulated form of cell death, has emerged as a key modulator of the tumor immune microenvironment (TIME) and a potential determinant of the immunotherapy response. However, the functional role of ICD in PCa progression, particularly in bone metastasis, remains poorly understood. This study aimed to elucidate the prognostic implications of ICD in PCa and develop an ICD-correlated signature (ICDCS) model to improve risk stratification and therapeutic decision-making.
Materials And Methods
We developed the ICDCS model by integrating multi-omics data, including single-cell transcriptomics, and leveraging computational approaches, such as AddModuleScore, WGCNA, ssGSEA, and 10 machine-learning algorithms (with 98 combinations). Its diagnostic and prognostic performances were rigorously assessed in the training cohort and two independent validation sets, offering a clinically applicable tool for outcome prediction. To gain deeper insights into prognostic features, we performed functional enrichment, immune infiltration, and immunotherapy response analyses. Additionally, we evaluated differential responses to immunotherapy across risk subgroups and identified potential personalized therapeutics. Finally, the vitro experiment validation of the key genes further strengthened the reliability of our findings.
Results
By integrating single-cell and bulk transcriptomic datasets, we identified 49 ICD-related genes, including nine linked to disease-free survival. By using genes common to both the training and validation sets, the final 8 key genes (MBNL1, IRAK3, CD59, LPP, TACC1, PRNP, CDC42EP3, and MYADM) were incorporated into 98 machine-learning computational frameworks for constructing ICDCS, and the Enet [α = 0.3] algorithm was finally chosen to build the model. In vitro validation confirmed the reduced expression of all eight genes in PCa cells, which was consistent with their putative protective roles. The ICDCS model exhibited robust prognostic accuracy for both the clinical outcomes and pathological features. Risk stratification based on ICDCS scores revealed distinct biological pathways, mutational profiles, and TIME characteristics between subgroups. Notably, patients in the high-risk subgroup showed an enhanced responsiveness to immunotherapy.
Conclusion
In this study, we developed an ICDCS for PCa bone metastases. This model provides a valuable tool for characterizing TIME in metastatic PCa and demonstrates significant clinical utility for prognostic assessment and prediction of immunotherapy response.