Clear cell renal cell carcinoma (ccRCC) is the most common and aggressive type of kidney cancer in adults. One-third of patients already have metastatic disease at the time of diagnosis, and treatment options for advanced cases, while improving, remain limited. Kidney cancer also has the highest death rate among urological cancers.
Programmed cell death (PCD) refers to the many controlled ways in which the body removes damaged or unwanted cells, including apoptosis, ferroptosis, autophagy, and others. Cancer cells often find ways to disable these death programs, allowing them to survive and grow unchecked.
Mitochondria, the structures that power cells, also regulate PCD. When mitochondria are dysfunctional, they can disrupt cell death signals, promote immune evasion, and reshape the tumor environment. However, how mitochondrial function and programmed cell death interact in kidney cancer, and what this means for patient survival, had not been well characterized before this study.
The team collected a list of 1,575 genes related to 19 different programmed cell death pathways and 1,136 mitochondrial-function genes from the MitoCarta3.0 database. They overlapped these with genes that were differentially expressed in ccRCC tumors compared to normal kidney tissue from the TCGA-KIRC dataset, yielding 192 mitochondrial and 429 cell-death-related differentially expressed genes.
Next, Pearson co-expression analysis identified 292 genes that are active together across both categories. Survival analysis across three independent patient cohorts (TCGA-KIRC, E-MTAB-1980, and GSE167573) narrowed this to 17 prognostically relevant genes with a consistent risk direction in all three groups.
These 17 genes were then used to build a model called mpMLDPS (machine learning-derived prognostic signature) using 76 different combinations of 10 machine learning algorithms, including LASSO regression, random survival forest, Cox regression variants, and others. The combination of LASSO and random survival forest (RSF) achieved the highest average C-index of 0.863, making it the optimal model.
In the TCGA-KIRC training dataset, the mpMLDPS model achieved AUC values of 0.968, 0.980, and 0.983 for predicting 1-, 3-, and 5-year overall survival, respectively. In the external validation cohorts, performance remained strong, with AUC values up to 0.900 in one cohort and 0.980 in another.
Patients were divided into high and low mpMLDPS groups based on the median score. Kaplan-Meier survival analysis consistently showed that those in the high-score group had significantly shorter overall survival across all three cohorts. The model also outperformed traditional clinical measures like tumor stage and pathology grade in predicting survival.
A nomogram (a visual tool doctors use to estimate individual patient risk) was constructed by combining mpMLDPS with patient age, the two factors that independently predicted survival in multivariate analysis. This tool could help clinicians communicate personalized prognosis estimates to patients and guide treatment planning.
Using the CIBERSORT immune deconvolution algorithm, the team found that high-mpMLDPS tumors had significantly more regulatory T cells (Tregs) and activated CD4+ T cells. Tregs typically suppress the immune system, which can protect tumors from attack. Higher stromal and immune scores in this group suggested greater infiltration by non-tumor cells, consistent with a more immunosuppressive environment.
Analysis of immune checkpoint gene expression revealed higher levels of CD28, CTLA4, PDCD1 (PD-1), and TIGIT in high-mpMLDPS patients, all markers associated with immune exhaustion. The TIDE algorithm predicted a significantly lower immunotherapy response rate in the high-mpMLDPS group (34%) versus the low group (43%).
Drug sensitivity analysis using CCLE, CTRP, and PRISM databases identified 12 candidate drugs potentially more effective for high-mpMLDPS patients. These included compounds targeting cancer metabolism, nuclear transport, and DNA repair pathways, offering leads for patients who may not respond well to standard immunotherapy.
The final eight genes in the mpMLDPS model are: PIF1, TIMP1, PLK1, E2F2, BCL2A1, CHDH, AGXT2, and AUH. Five of these (PIF1, TIMP1, PLK1, E2F2, BCL2A1) are overexpressed in tumor tissue compared to normal kidney, while three (CHDH, AGXT2, AUH) are underexpressed in tumors. High expression of the first group predicts worse survival; low expression of the second group also predicts worse survival.
PIF1 stood out because it had the highest hazard ratio among all eight genes and showed the largest expression difference between tumor and normal tissue. Gene ontology and pathway enrichment analysis linked PIF1 to cell cycle regulation, chromosome stability, DNA repair, and energy metabolism, positioning it as a gene at the crossroads of mitochondrial function and programmed cell death.
PIF1 expression also increased progressively with more advanced clinical stage and higher tumor grade in the TCGA-KIRC dataset. Tissue staining (immunohistochemistry) and protein analysis (Western blot) on actual patient samples confirmed that PIF1 protein levels were significantly higher in kidney cancer tissue compared to adjacent normal tissue, validating the computational findings in a real clinical setting.
To test whether PIF1 actually drives kidney cancer behavior, the researchers used a lentiviral system to knock down PIF1 expression in two kidney cancer cell lines: Caki-1 and 786-O. These lines were chosen because they showed the highest PIF1 protein levels among all tested cancer cell lines.
After PIF1 silencing, CCK-8 proliferation assays showed significantly reduced cell growth. Colony formation assays confirmed that cells had greatly diminished ability to form new colonies. Wound healing and Transwell migration assays showed that the cells also moved more slowly, indicating that PIF1 promotes both the growth and the spread of kidney cancer cells.
On the molecular level, PIF1 knockdown reduced mitochondrial membrane potential and increased reactive oxygen species (ROS), markers of mitochondrial stress. Cell cycle analysis by flow cytometry showed that cells accumulated in the G0/G1 phase (a growth pause) and fewer cells entered S-phase (DNA replication). Key proteins confirmed these effects: CDK2 and CDK4 (cell cycle drivers) were reduced, while p21 (a brake on cell division) and Cleaved Caspase-9 (a signal of programmed cell death) were increased.
The mpMLDPS signature represents a significant improvement over existing clinical tools for predicting outcome in clear cell renal cell carcinoma. Built from genes at the intersection of mitochondrial function and programmed cell death, it captures biological complexity that traditional staging systems miss. Its consistent performance across multiple independent patient cohorts makes it a strong candidate for clinical validation.
The discovery of PIF1 as a high-impact driver gene within this signature opens a new avenue for targeted therapy. PIF1 is a DNA helicase that normally protects cells from DNA damage during replication. In kidney cancer, its overexpression appears to help tumor cells survive and proliferate by maintaining mitochondrial health and suppressing apoptosis. Blocking PIF1 could disrupt these protective mechanisms.
The study also highlights the broader principle that mitochondrial dysfunction and immune evasion are deeply connected in ccRCC. Patients with high mitochondrial-related stress in their tumors tend to have more immunosuppressive environments and respond less well to current immunotherapies. Understanding and targeting this axis could help a subgroup of kidney cancer patients who currently have limited treatment options.