Colon adenocarcinoma (COAD) is one of the most common and deadly cancers worldwide. Despite advances in surgery, chemotherapy, radiation, and immunotherapy, patient outcomes still vary widely because COAD is not a single uniform disease but a collection of molecularly distinct subtypes, each behaving differently in response to treatment.
Traditional staging tools like the TNM system, which categorize cancer by tumor size, lymph node involvement, and spread, often fail to capture the full molecular complexity of the disease. This means some patients receive treatments that may be too aggressive or not aggressive enough for their specific tumor biology.
Researchers have been searching for molecular markers that can more accurately predict which patients will do well and which are at high risk of poor outcomes. This study focuses on a cell death pathway called parthanatos as a source of such prognostic information in COAD.
Parthanatos is a distinct form of regulated cell death that differs from the more familiar process of apoptosis. It is triggered when the enzyme PARP1 becomes excessively activated in response to severe DNA damage or oxidative stress, ultimately leading to large-scale DNA fragmentation and cell death in a caspase-independent manner.
The relationship between parthanatos and cancer is complex. On one hand, activating this cell death pathway can eliminate malignant cells undergoing overwhelming DNA stress, potentially enhancing the effects of chemotherapy and radiation therapy. On the other hand, tumors may learn to dysregulate parthanatos-related pathways to evade death and become resistant to treatment.
Aberrant PARP1 activity, impaired translocation of the apoptosis-inducing factor (AIF) from mitochondria to the nucleus, and altered DNA damage responses are all ways cancer cells can subvert parthanatos to their advantage. Emerging research also suggests that parthanatos influences the tumor immune microenvironment through PARP-dependent metabolic and transcriptional reprogramming.
These opposing roles make parthanatos a particularly interesting target for cancer research: understanding which tumors have active parthanatos pathways versus suppressed ones could guide both prognosis and treatment selection. This study set out to systematically characterize parthanatos-associated genes (PAGs) across large COAD patient datasets.
The researchers collected transcriptomic gene expression data and clinical information from multiple large public databases. The TCGA-COAD dataset provided 417 tumor and 42 normal tissue samples, while two GEO datasets (GSE39582 with 556 samples and GSE17538 with 200 samples) provided additional validation cohorts. Batch effects between datasets were corrected using standard bioinformatic tools.
A total of 37 parthanatos-associated genes (PAGs) were retrieved from the GeneCards database. Of these, 11 were found to be significantly differentially expressed between normal colon tissue and COAD tumors, passing thresholds of fold change greater than or equal to 1.5 and adjusted p-value below 0.05. These 11 differentially expressed PAGs (DE-PAGs) formed the foundation for all subsequent analyses.
To build a prognostic scoring model, the researchers used an advanced machine learning framework testing 64 model combinations derived from 10 different algorithms, including LASSO-Cox regression, elastic net, gradient boosting, and support vector machines. All model training used leave-one-out cross-validation to minimize overfitting. The elastic net model with alpha of 0.1 produced the highest concordance index and was selected as the final model.
Additional analyses included protein-protein interaction network construction, copy number variation analysis, somatic mutation profiling, tumor immune microenvironment assessment using multiple algorithms including ESTIMATE and ssGSEA, immunotherapy response prediction using TIDE and immunophenoscore frameworks, and single-cell RNA sequencing analysis of colon tissue samples.
Of the 11 DE-PAGs identified, several showed notable expression patterns. Genes including NAT10, CUL4A, FEN1, TOMM20, GPX4, PARP1, NAMPT, and COL8A1 were significantly overexpressed in COAD tumors compared to normal tissue. In contrast, NCF1, ESR1, and ESR2 were elevated in normal tissue. Protein-protein interaction network analysis confirmed extensive functional connections among these genes.
Using unsupervised consensus clustering across 973 combined COAD samples, two distinct molecular subtypes emerged based on DE-PAG expression profiles, designated PAG subtype A (441 samples) and PAG subtype B (532 samples). These two groups showed clear separation in principal component analysis, confirming their biological distinctiveness.
Survival analysis revealed a statistically significant difference between the two subtypes: patients in PAG subtype B had significantly better overall survival than those in subtype A, with a hazard ratio of 1.32 (95% CI: 1.05 to 1.68, p = 0.020). This finding indicates that parthanatos gene activity patterns correlate meaningfully with patient prognosis in COAD.
Gene set variation analysis based on KEGG pathways revealed pronounced functional differences between subtypes. PAG subtype A, associated with worse outcomes, showed dysregulation in drug metabolism pathways. PAG subtype B showed suppression of tumor progression pathways including the TGF-beta signaling pathway, along with suppression of immune-related pathways such as leukocyte transendothelial migration and ECM-receptor interaction.
Comprehensive immune infiltration analysis using ESTIMATE algorithms revealed that PAG subtype B, the better-prognosis group, had significantly lower immune scores, stromal scores, and ESTIMATE scores compared to subtype A, along with higher tumor purity. This suggests that subtype B tumors have a fundamentally different relationship with the surrounding immune and stromal environment.
Single-sample gene set enrichment analysis quantifying 23 immune cell types showed that most immune cell populations, including activated B cells, CD4+ T cells, CD8+ T cells, and activated dendritic cells, were significantly reduced in PAG subtype B. While counterintuitive, lower immune infiltration in a better-prognosis subtype may reflect less immune-mediated tumor promotion rather than less anti-tumor immunity.
Immunotherapy response prediction using the TIDE framework showed that PAG subtype B had lower immune evasion scores, suggesting a potentially enhanced likelihood of responding to immune checkpoint inhibitor therapy. Immunophenoscore analysis also indicated higher predicted responsiveness to both CTLA-4 and PD-1 checkpoint blockade therapies in subtype B.
These immune findings reveal that parthanatos molecular subtypes are deeply interconnected with the tumor immune microenvironment. The two subtypes not only differ in their parthanatos gene expression but also in how immune cells infiltrate the tumor and how likely the tumor is to respond to modern immunotherapy.
From the 871 differentially expressed genes between PAG subtypes, the elastic net machine learning model selected seven key prognostic variables for the final PAG score. Of these, SLC2A3 (solute carrier family 2 member 3) exhibited the highest risk coefficient, marking it as the most prominent prognostic gene in the model. SLC2A3 encodes a glucose transporter protein, linking cancer metabolism to tumor aggressiveness.
Patients were stratified into high and low PAG score groups based on the median score. Kaplan-Meier survival curves demonstrated that patients with low PAG scores consistently showed significantly better overall survival in both the training cohort and the independent GSE17538 validation cohort, confirming the model's robustness across independent datasets.
Multivariate Cox regression analysis confirmed that the PAG score was an independent prognostic factor in COAD, separate from clinical staging variables. In the training cohort, hazard ratio for the PAG score was 2.224 (95% CI: 1.605 to 3.081, p less than 0.001). Time-dependent receiver operating characteristic analysis showed AUC values of 0.722, 0.733, and 0.720 for predicting 1-, 3-, and 5-year overall survival in the validation cohort, indicating strong predictive performance.
PAG scores were significantly elevated in patients with advanced tumor stage, T stage, and N stage, while no significant differences were observed across age or sex subgroups. This staging correlation validates that the PAG score captures biologically meaningful disease progression information rather than simply reflecting demographic factors.
To move beyond computational analysis, the researchers conducted laboratory experiments to directly test whether SLC2A3 functionally promotes colon cancer behavior. Western blot analysis confirmed that SLC2A3 protein expression was significantly higher in the colon cancer cell line SW480 compared to the normal colorectal epithelial cell line NCM460, consistent with the bioinformatic findings.
Using small interfering RNA (siRNA) technology, the researchers knocked down SLC2A3 expression in SW480 colon cancer cells. CCK-8 proliferation assays showed that reducing SLC2A3 levels significantly slowed cancer cell growth compared to control cells, with effects becoming more pronounced over the 96-hour observation period.
Colony formation assays further confirmed that SLC2A3 knockdown reduced the ability of cancer cells to form colonies, a measure of long-term proliferative capacity. Transwell invasion assays showed that SLC2A3 knockdown also significantly impaired the ability of SW480 cells to invade through Matrigel-coated membranes, suggesting that SLC2A3 contributes to tumor invasiveness.
These experimental findings, while preliminary and requiring validation in additional models, provide biological evidence supporting the computational prediction that SLC2A3 is a functionally important gene in colon cancer. As a glucose transporter, SLC2A3 may fuel the high metabolic demands of rapidly growing and invading cancer cells, connecting the Warburg effect to tumor aggressiveness in COAD.
High PAG scores were associated with advanced microsatellite instability (MSI) status. Tumors with MSI-high status, which are known to carry more mutations and often respond better to immunotherapy, showed distinct PAG score distributions compared to microsatellite stable tumors. This association reinforces the connection between the parthanatos pathway and genomic instability in COAD.
Tumor mutation burden (TMB) analysis revealed that high PAG score patients tended to carry greater mutational loads. Higher TMB is associated with increased neoantigen production and, in many cancer types, improved response to immune checkpoint inhibitors. Together with the immunophenoscore findings, this suggests that high-PAG-score patients might paradoxically benefit from immunotherapy despite their worse prognosis under standard treatment.
Drug sensitivity prediction using the GDSC database indicated differential sensitivity to various chemotherapy agents across PAG score subgroups. These predictions, if validated clinically, could help guide treatment selection, offering the possibility of matching specific chemotherapy regimens to the molecular profile of individual patients' tumors.
Single-cell RNA sequencing analysis of three normal and three tumor colon tissue samples provided cellular-resolution insight into PAG expression. The analysis revealed broad expression of PAG-related signatures across multiple cell types within the tumor microenvironment, suggesting that parthanatos-related biology is not confined to cancer cells alone but extends to immune and stromal cell populations.
This cellular heterogeneity in PAG expression helps explain why bulk transcriptomic analyses capture different signals than single-cell approaches. Tumors with high overall PAG scores may have elevated expression in multiple cell compartments simultaneously, reflecting a systemic shift in parthanatos-related biology throughout the tumor ecosystem rather than a purely cancer-cell-intrinsic phenomenon.
The finding that SLC2A3, a glucose transporter, is the top risk gene in the parthanatos prognostic model opens an interesting biological connection. Cancer cells have long been known to preferentially consume glucose through aerobic glycolysis, a phenomenon called the Warburg effect. SLC2A3-mediated glucose uptake may provide the metabolic fuel that enables cancer cells to maintain the high energy demands associated with both rapid proliferation and resistance to programmed cell death including parthanatos.
This study establishes parthanatos as a clinically relevant regulatory axis in colon adenocarcinoma. By integrating transcriptomic data from nearly 1,200 patients across multiple cohorts, the researchers identified two biologically distinct COAD subtypes defined by parthanatos gene expression, each with different immune landscapes, mutation profiles, and survival outcomes.
The seven-gene PAG prognostic score provides an independent predictor of patient survival that complements existing clinical staging information. Validated across multiple independent cohorts with AUC values consistently above 0.70, this score represents a potentially clinically useful tool for identifying high-risk patients who may need more aggressive treatment or closer monitoring.
SLC2A3 emerges as the most prominent actionable target identified by this framework. Its high expression in colon cancer compared to normal tissue, combined with experimental evidence showing that reducing its expression slows cancer cell growth and invasion, positions it as a candidate for future therapeutic targeting. Inhibiting glucose transport in high-SLC2A3 tumors could potentially deprive cancer cells of the metabolic resources they need to survive and spread.
The study also highlights important limitations: the in vitro SLC2A3 experiments used a single cell line and require validation in more diverse models and eventual clinical studies. The prognostic model also needs prospective validation before clinical implementation. Future research integrating multi-omics data with single-cell and spatial transcriptomics could further refine the parthanatos-based framework and identify additional therapeutic vulnerabilities in COAD.