A Ferroptosis–Heat Shock Response Prognostic Signature in Hepatocellular Carcinoma: Cross-Platform Validation and Independence from HSPB1–GPX4 Axis Activity

Background Hepatocellular carcinoma (HCC) remains a leading cause of cancer mortality, and molecular biomarkers are urgently needed to refine risk stratification beyond conventional staging. Although ferroptosis and the heat shock response have each been proposed as therapeutic vulnerabilities in HCC, their prognostic interplay has not been systematically evaluated. Specifically, it is unknown whether a ferroptosis-derived gene signature retains predictive value across varying levels of heat shock response activity, or whether the established HSPB1–GPX4 cytoprotective axis overrides ferroptosis-based prognostic information. Methods We analyzed RNA sequencing data from 365 HCC patients in TCGA-LIHC and constructed a LASSO-Cox prognostic model from 83 candidate genes spanning ferroptosis and heat shock response pathways. The resulting five-gene signature was validated internally (70/30 split with bootstrap resampling) and externally on two independent cohorts profiled on technically distinct microarray platforms: GSE14520 (n = 221, Affymetrix U133 Plus 2.0) as primary external validation and GSE116174 (n = 64, Affymetrix HT HG-U133 + PM) as cross-platform secondary validation. We performed stratified survival analyses by HSPB1 expression and quantified the HSPB1–GPX4 correlation across risk groups, survival outcomes, and tissue types. A nomogram integrating the signature with TNM stage was developed, and decision curve analysis assessed net clinical benefit. Results The signature comprising SLC2A1, NFS1, AIFM2, HILPDA, and SLC7A11 achieved a C-index of 0.67 (95% CI: 0.62–0.72) in TCGA-LIHC (training HR = 5.32, P = 0.008; validation HR = 12.55, P = 0.010). On GSE14520, a reduced two-gene model achieved significant risk separation (log-rank P = 0.0056; HR = 1.89). On GSE116174, all five genes were detected and showed directionally concordant estimates (HR = 1.21), confirming cross-platform technical feasibility despite limited statistical power at n = 64. Critically, the signature stratified patients independently of HSPB1 expression (high-HSPB1 subgroup HR = 4.65, P = 0.044; low-HSPB1 subgroup HR = 9.75, P = 0.002; interaction P = 0.84). HSPB1 correlated positively with GPX4 across all subgroups (Spearman ρ = 0.415, P < < 1×10⁻¹¹), yet adding HSPB1 to the model did not improve discrimination (ΔAIC = + 4.0; C-index unchanged). The nomogram combining the risk score with TNM stage achieved a C-index of 0.70 (95% CI: 0.64–0.76) and demonstrated net clinical benefit across threshold probabilities of 10–80%. Conclusions This study establishes a five-gene ferroptosis–heat shock response prognostic signature for HCC that generalizes across RNA-seq and two disparate microarray platforms. The consistent prognostic performance across HSPB1-defined subgroups—and the observation that robust HSPB1–GPX4 co-expression does not translate into incremental prognostic value—demonstrates that ferroptosis-based prognostic information is largely independent of HSPB1–GPX4 axis activity. High-risk patients, particularly those with elevated SLC2A1 or SLC7A11 expression, may be prioritized for intensified surveillance or enrollment in trials testing ferroptosis-inducing agents, without requiring ancillary HSPB1 testing. The accompanying nomogram provides a readily applicable bedside risk-stratification tool.