HOXD9 prioritization within developmental regulator-related genes suggests stage-adjusted prognostic relevance in hepatocellular carcinoma

Background: Prognostic stratification in hepatocellular carcinoma (HCC) remains difficult because standard staging systems do not fully capture the biological heterogeneity of the disease. We hypothesized that developmental regulators could provide a biologically relevant framework for prioritizing prognostic markers.

Methods: RNA sequencing (RNA-seq) and clinical data from The Cancer Genome Atlas liver hepatocellular carcinoma (TCGA-LIHC) cohort were analyzed. Tumor vs. normal differential expression analysis was performed using differential expression analysis for sequence count data 2 (DESeq2). Developmental regulator-related candidate genes were prioritized through a stepwise framework integrating differential expression, clinicopathological association, Kaplan-Meier survival analysis, and Cox regression. A main prognostic model was constructed from the final independent factors and evaluated using a nomogram, calibration analysis, and time-dependent receiver operating characteristic (ROC) analysis. The incremental prognostic value of HOXD9 beyond pathologic stage was assessed using the likelihood ratio test, Harrell’s concordance index (C-index), bootstrap-based delta C-index analysis, and time-dependent area under the curve (AUC) comparison. External validation was performed in the LIRI-JP (Liver Cancer-RIKEN, Japan) cohort.

Results: Among the screened developmental regulator-related genes, HOXD9 and MESP2 showed the most consistent signals across differential expression, clinicopathological association, and survival analyses. In multivariable Cox regression in the TCGA-LIHC cohort, only pathologic stage [hazard ratio (HR) =1.565, 95% confidence interval (CI): 1.266–1.935; P<0.001] and HOXD9 expression [HR =1.165, 95% CI: 1.035–1.312; P=0.01] remained independently associated with overall survival (OS). The main prognostic model integrating pathologic stage and HOXD9 yielded AUC values of 0.707 at 1 year and 0.701 at 3 years. Compared with the stage-only model, the stage plus HOXD9 model showed improved model fit (P=0.004), a higher C-index (0.651 vs. 0.601; P=0.004), and a higher 1-year AUC (0.707 vs. 0.656; P=0.02). These findings indicated incremental prognostic value of HOXD9 in the TCGA-LIHC training cohort. In the LIRI-JP cohort, HOXD9 significantly stratified OS (P=0.01), and the main external validation model yielded AUC values of 0.819 at 1 year and 0.739 at 3 years. However, HOXD9 was not independently associated with OS after adjustment for tumor-node-metastasis (TNM) stage in the external multivariable model (HR =1.027, 95% CI: 0.763–1.383; P=0.86).

Conclusions: HOXD9 emerged as a key prognostic candidate and showed incremental prognostic value beyond pathologic stage in the TCGA-LIHC training cohort. However, this stage-adjusted independent association was not reproduced in the LIRI-JP multivariable validation model. HOXD9 should therefore be considered a candidate prognostic marker requiring further validation.