4-Gene Polyamine Signature Predicts Liver Cancer Survival and Treatment Response

Liver cancer, predominantly hepatocellular carcinoma (HCC), kills nearly 830,000 people annually worldwide and carries a dismal 5-year survival rate of just 18% overall. Identifying molecular markers that stratify patients by prognosis and predict treatment response is critical for advancing precision oncology in this disease. This study focused on polyamines — small positively charged molecules including putrescine, spermidine, and spermine — whose metabolism has established roles in cell proliferation, immune suppression, and tumor progression, yet whose prognostic value in HCC had not been systematically characterized.

The researchers assembled a large multi-cohort dataset incorporating transcriptome, clinical, and somatic mutation data from TCGA (371 tumor samples) and GEO dataset GSE14520 (225 tumor samples), with four additional external validation cohorts: LIRI-JP, CHCC-HBV, GSE54236, and GSE116174. Twenty-two polyamine-related genes (PRGs) were sourced from the Molecular Signatures Database. Consensus clustering of TCGA HCC samples using these PRGs identified two distinct molecular subtypes — Cluster A and Cluster B. Cluster A was characterized by poor prognosis, elevated immune cell infiltration, and reduced metabolic pathway activity, while Cluster B showed relatively better survival outcomes and higher metabolic activity.

Differentially expressed genes (DEGs) between the two PRG clusters were identified using adjusted P < .05 and |logFC| > 1 thresholds. These DEGs were then subjected to univariate Cox regression, followed by LASSO regularization to prevent overfitting, and finally multivariate Cox regression. This pipeline distilled a 4-gene prognostic signature comprising KIF20A, SPP1, PON1, and SERPINE1, each weighted by its regression coefficient to produce a continuous risk score. Patients were split at the median risk score into high- and low-risk groups. In the training cohort, high-risk patients had significantly worse overall survival, a finding consistently replicated across the internal test cohort, the full merged cohort, and all four external validation cohorts — a robust cross-dataset confirmation.

The signature demonstrated superior predictive accuracy compared to four previously published HCC prognostic signatures, with higher concordance indices (C-index), improved time-dependent AUC values at 1-, 3-, and 5-year timepoints, and better-calibrated nomogram predictions. Multivariate Cox analysis confirmed the risk score was an independent prognostic factor after adjusting for age, sex, tumor grade, and TNM/T stage. High-risk patients exhibited a significantly higher TP53 mutation rate, distinct tumor immune microenvironment features including differential infiltration of CD8+ T cells, macrophages, and regulatory T cells, and higher TIDE scores — indicating greater likelihood of immune dysfunction and poorer immunotherapy response. Predicted sensitivity analysis using the GDSC database and sorafenib/TACE treatment cohorts (GSE109211 and GSE104580) further suggested the high-risk group would respond worse to standard-of-care systemic and locoregional therapies.

Expression of all four signature genes was confirmed as aberrant in HCC relative to normal liver tissue at the mRNA level across ten GEO datasets, at the protein level via UALCAN, and at the single-cell level via the TISCH database. KIF20A and SPP1 were upregulated in tumor tissue, while PON1 was markedly downregulated, consistent with its known hepatoprotective detoxification role. SERPINE1 upregulation aligned with its established pro-metastatic and fibrinolysis-inhibiting functions. These multi-level validations strengthen confidence in the biological plausibility of the signature. Limitations include the retrospective, bioinformatics-driven design, reliance on public datasets of varying quality, and the absence of prospective clinical validation.