The Hoshida S1 class of hepatocellular carcinoma spans proliferative and hepatocyte-like tumour states with different niche contributions

The Hoshida S1 class of hepatocellular carcinoma (HCC) is associated with immune context, yet S1 tumours vary in hepatic differentiation and proliferation. This study examines how hepatic-lineage and tissue-associated expression combine in an S1 call, and whether S1 contains biologically different tumour states. It extends our previous bulk-RNA study (https://doi.org/10.5281/zenodo.23214311) and our single-cell study of the cellular sources of Hoshida calls (https://doi.org/10.5281/zenodo.23593492). In 348 TCGA-LIHC tumours, the S1 classification margins were decomposed into exact gene contributions, with genes annotated for cellular source using a fixed single-cell atlas. The components were compared with DNA-based purity, the DNA-methylation leukocyte fraction and same-portion histology. The calculation was repeated in HEPTROMIC (204 tumours with paired RNA and DNA methylation) and LIRI-JP (49 HCC with whole-genome consensus purity). The analyses were exploratory or post hoc. Only 8 of 235 S1 genes were predominant in the tumour epithelial compartment (TEC). The combined contribution of 74 non-TEC-predominant S1 genes tracked DNA-based tissue measures in all three cohorts (TCGA purity rho -0.73, leukocyte fraction 0.85; HEPTROMIC leukocyte-associated methylation 0.59; LIRI-JP whole-genome purity -0.56). In HEPTROMIC, the correlation with the leukocyte-associated methylation surrogate exceeded that of the below-reference TEC-predominant S3 component by 0.36 (95% CI 0.20 to 0.52). Excluding the 83 non-TEC-predominant template genes and renormalizing the classifier retained 98 of 108 TCGA S1 calls. S1 encompassed two configurations. On average, in tumours whose nearest competing class was S2, reduced expression of the TEC-predominant S3 hepatocyte programme contributed more to the S1 call than the non-epithelial S1 genes. This programme includes urea-cycle, gluconeogenic, bile-acid and drug-metabolism genes. In S3-nearest tumours, the balance reversed, and these tumours had higher hepatic and lower proliferation-related expression. Both patterns recurred across the three cohorts. In TCGA-LIHC, S2-nearest tumours more often carried TP53 mutations (29 of 62 against 5 of 35; difference 32.5 percentage points, 95% CI 13.3 to 47.1) and grade 3 histology (38 of 70 against 10 of 38), whereas S3-nearest tumours had more stroma on histology (median 17.5% against 10.0%). All five hepatic-metabolic-dominant grade 3 S1 tumours were S3-nearest: high histological grade coexisted with near-reference hepatic expression and below-reference proliferation-related expression. In a pooled analysis of TCGA-LIHC grade 3 tumours, TP53 mutation was associated with proliferation-related expression after accounting for Hoshida class and S1 configuration (odds ratio 2.30 per unit of the standardized proliferation-related score z(P), 95% CI 1.29 to 4.10). Applied to the 105 candidate genes of the previous bulk study, the single-cell atlas placed 18 in the TEC-predominant, 18 in the endothelial-predominant and 50 in the context-dependent category. Co-expression was compared between 43 P-dominant grade 3 S1 tumours and 113 grade 1-2 S3 H-dominant tumours. P-dominant denotes a higher standardized proliferation-related score than hepatic-metabolic score (z(P) > z(H)). Eight of 32 candidate pairs had correlation differences with pointwise 95% intervals excluding zero. These included stronger correlations for the endothelial pairs TIE1-ESAM and TIE1-CLEC14A and lower correlations for five pairs involving PGF or PECAM1, linking the earlier candidates to a vascular context. The same S1 designation thus covers a proliferative, TP53-associated configuration with reduced hepatocyte-programme expression and a hepatocyte-like configuration with a larger contribution from its niche. Studies of immunotherapy and targeted therapy in HCC can evaluate the two configurations separately, alongside hepatic-lineage, proliferation and tissue-context measures. Pooling these biologically different S1 tumours may obscure response heterogeneity or dilute biomarker associations. Files include the manuscript (PDF and DOCX), Supplementary Methods and Results with Supplementary Figures S1-S8 (DOCX), Supplementary Tables S1-S40 (DOCX), figures (PNG ZIP) and source data (ZIP). The source data include the inputs of the Hoshida call not deposited earlier, derived per-tumour and per-gene values, gene and sample mappings of the external cohorts, and the source data of every figure, together with SHA-256 values and a list of inputs held in the two earlier records. Analysis and drawing scripts are available from the author on request. CS11-README.txt describes the files; SHA256SUMS gives their checksums. Third-party materials (TCGA-LIHC data from the Genomic Data Commons and PanCanAtlas, the GEO series of HEPTROMIC, the PCAWG/ICGC LIRI-JP data, published gene lists and annotation packages) remain subject to their original licences and repository terms.