Development and validation of an HDAC-related prognostic signature for hepatocellular carcinoma: a single-cell transcriptomic and machine learning study

Hepatocellular carcinoma (HCC) remains a health concern due to its heterogeneity, high mortality, and poor prognosis. Histone deacetylases (HDACs) influence tumor progression, but their role in HCC tumor microenvironment (TME) remains underexplored. This study aimed to characterize HDAC-related cell subtypes in HCC using single-cell RNA sequencing (scRNA-seq) data and to develop a robust prognostic signature suitable for clinical application. We analyzed scRNA-seq data (GSE242889) from HCC and adjacent non‑tumor tissues. Malignant cells were identified using inferCNV, and HDAC subtypes were classified by non-negative matrix factorization (NMF). A machine learning-based prognostic signature (HDAC-related Signature) was then developed and validated in independent cohorts. Thirteen differentially expressed HDACs were identified, and malignant cells showed reduced HDAC enrichment scores. NMF distinguished three HDAC subtypes (HDAC-C1, -C2, -C3), each with distinct metabolic and transcriptional profiles. The HDAC-C1 subtype, associated with copper detoxification and poor prognosis, was used to develop the HDAC-related Signature, incorporating six genes ( SQSTM1 , ADM , HDAC2 , LGALS1 , HDAC1 , NDUFS5 ). This signature stratified patients into high- and low-risk groups. The high-risk group exhibited worse overall survival (OS), elevated tumor mutational burden, and an immunosuppressive microenvironment, whereas the low-risk group was characterized by enhanced infiltration of anti-tumor immune effectors (CD8 + T cells and NK cells). The signature consistently maintained strong discriminative power across independent validation cohorts. This study delineates the heterogeneity of HDAC-related malignant subtypes in HCC and establishes the HDAC-related Signature as a robust Class C1 prognostic tool within the ESMO EBAI (2025) framework. The signature accurately stratifies baseline patient survival and reflects distinct immune microenvironment phenotypes, providing a reproducible prognostic framework for clinical risk assessment.