Programmed cell death encompasses multiple mechanisms including apoptosis, ferroptosis, autophagy, necroptosis, and related pathways. Dysregulation of these processes in clear cell renal cell carcinoma (ccRCC) allows cancer cells to evade death signals and sustain uncontrolled proliferation.
The tumor microenvironment (TME) of ccRCC is characterized by complex immune cell infiltration patterns that interact with cell death pathways to influence disease progression and treatment resistance. Understanding which programmed cell death related genes (PCDRGs) drive prognosis could reveal new therapeutic targets.
This study applied 10 machine learning algorithms to a curated set of PCDRGs from the ImmPort database to build a prognostic signature for ccRCC, then validated it in independent cohorts and explored its immune landscape and drug sensitivity implications.
The primary dataset was TCGA-KIRC with 537 ccRCC tumor samples and 60 normal kidney samples. An independent validation cohort of 101 ccRCC samples from ArrayExpress provided external performance testing. Starting from 581 PCDRGs identified through ImmPort, differential expression analysis narrowed the candidate list to 218 genes significantly altered in ccRCC.
Ten machine learning algorithms were applied in combination to generate 101 distinct model configurations. These algorithms included LASSO regression, Random Forest, Support Vector Machine, elastic net, and others. Each configuration was evaluated by C-index on the training cohort.
The Lasso+SuperPC combination was selected as the optimal model. SuperPC (Supervised Principal Components) identifies latent structure in high-dimensional gene expression data while accounting for survival outcomes, and its combination with LASSO's regularization yielded the best prognostic discrimination.
The final model used 17 genes: bcl2, bcl3, bid, cdkn1a, chac1, csf2, cx3cl1, bdh, ddias, erbb3, il4, ltbr, pdcd5, pink1, plaur, ybx3, and prkag3. Patients were stratified into high-risk and low-risk groups based on the model's risk score.
The 17-gene prognostic model achieved AUC values of 0.787, 0.753, and 0.757 for 1-year, 3-year, and 5-year overall survival prediction respectively in TCGA-KIRC. In the independent ArrayExpress validation cohort, corresponding AUC values were consistent, confirming the signature's generalizability.
High-risk patients defined by the model showed significantly worse overall survival compared to low-risk patients across all validation cohorts. Multivariate Cox regression confirmed that the risk score remained an independent prognostic factor after adjusting for clinical variables including tumor stage and grade.
CIBERSORT immune deconvolution revealed that high-risk group tumors were enriched for regulatory T cells (Tregs) and M0 macrophages, while low-risk tumors showed higher infiltration of M1 macrophages. This immunosuppressive TME pattern in high-risk patients aligns with known mechanisms of immune evasion in aggressive ccRCC.
Drug sensitivity analysis using TIDE scores showed that high-risk patients predicted by the model were more sensitive to rapamycin and vinblastine, suggesting potential therapeutic vulnerabilities that could be exploited in this patient subgroup. Low-risk patients showed differential responses to immune checkpoint therapies.
Single-cell RNA sequencing (scRNA-seq) analysis of ccRCC tumor microenvironment data identified monocytes as the key cell type differentially expressing the 17 signature genes, linking the prognostic signal to a specific TME cellular population rather than the tumor cells themselves.
Functional in vitro validation demonstrated that CSF2, one of the 17 signature genes, inhibited ccRCC cell proliferation, migration, and invasion when experimentally manipulated, providing mechanistic confirmation that the signature genes are functionally relevant to tumor biology rather than merely correlative markers.
The identification of Treg enrichment and M0 macrophage dominance in high-risk tumors has direct implications for immunotherapy selection. Tumors with immunosuppressive microenvironments may benefit from combination approaches targeting regulatory T cells or macrophage repolarization alongside PD-1/PD-L1 checkpoint inhibition.
The drug sensitivity predictions suggest that mTOR inhibitors like rapamycin may be particularly effective in high-risk ccRCC patients, consistent with the established role of mTOR pathway dysregulation in RCC biology. This provides a rationale for biomarker-guided mTOR inhibitor use.
The TIDE score integration links the prognostic signature to immune dysfunction and tumor immune escape, offering a framework for simultaneously stratifying prognosis and predicting immunotherapy benefit. This dual utility is particularly valuable given the growing importance of immune checkpoint blockade in advanced ccRCC treatment.
This study demonstrates that systematic application of multiple machine learning algorithms to programmed cell death gene expression data can identify robust prognostic signatures in ccRCC that capture both tumor intrinsic and microenvironment-mediated survival signals.
The 17-gene Lasso+SuperPC signature integrates diverse cell death mechanisms including apoptosis (BCL2, BID), ferroptosis (CHAC1), and mitophagy (PINK1), reflecting the multi-pathway nature of cell death dysregulation in ccRCC progression.
Future work should prospectively validate this signature in clinical trial cohorts, assess its performance in the context of current standard-of-care treatment combinations, and explore whether risk-stratified treatment allocation based on this model improves patient survival outcomes.