Answers are grounded in this paper and cite their sources.
Transcriptomic classes of hepatocellular carcinoma (HCC) are assigned from bulk tissue, which mixes tumour epithelial cells with immune, endothelial and stromal cells. This study examines which cellular compartments supply the expression underlying Hoshida subclass calls and the hepatic-metabolic (H) and proliferation-related (P) readings of our previous bulk-RNA study (doi:10.5281/zenodo.23214311), and how tissue composition and the profiles used for standardization affect their interpretation. We analysed 272,444 cells or nuclei from 38 tumours (HBV 14, HCV 5, non-viral 13, not stated 6) across seven public single-cell or single-nucleus RNA-sequencing datasets, with the patient as the unit of inference. Compartment pseudobulk profiles were used to distinguish shares of captured transcripts from expression rates per unit of captured RNA, decompose subclass decision margins into gene terms annotated by the compartment that supplied most of each gene's captured counts, and compare all-cell readings with those of the marker-defined tumour epithelial compartment (TEC). Expected directions derived from these analyses were specified before testing in 348 tumours of The Cancer Genome Atlas liver cohort (TCGA-LIHC). In the four larger single-cell cohorts, comprising 30 tumours (HBV 11, HCV 5, non-viral 12, not stated 2), the S1-over-S3 margins of the all-cell S1 calls received approximately equal support from S1-template genes, most often myeloid-captured, and from lower all-cell expression of S3-template genes, most often TEC-captured. For S1-template genes, compartment-to-TEC expression-rate ratios, summarized by taking the median across genes within each tumour and then across tumours, exceeded 1 in each of the five annotated non-epithelial compartments, and those for S3-template genes were below 1 (0.11 to 0.33). Non-tumour cells supplied most of the S1-template transcripts in each of the 13 all-cell S1 tumours (HBV 5, HCV 2, non-viral 6). When classified among the TEC profiles of their own cohorts, the TECs of 9 of these 13 tumours were called S2 or S3, including four calls with a false discovery rate of at least 0.05. The calls depended on the profiles used for standardization: standardized with the all-cell profiles, the TEC of every all-cell S1 tumour was called S3 (10) or S2 (3), and standardized with the TEC profiles, the all-cell pseudobulk was S1 in all 30 tumours. In a post hoc comparison within each of the four cohorts, every all-cell S1 tumour had a lower TEC share of the captured RNA than every all-cell S3 tumour (14 tumours: HBV 4, HCV 3, non-viral 5, not stated 2). The H target genes were enriched in the TEC, whereas T/NK cells expressed the P targets at higher rates than the TEC in 22 of 27 evaluable tumours (HBV 10, HCV 4, non-viral 11, not stated 2). In TCGA-LIHC (aetiology not recorded in the tables used), S1 tumours had lower purity and higher immune and stromal proxies than S3 tumours. Among the 174 tumours with ABSOLUTE (DNA-based) purity estimates, H rose with purity (Spearman rho 0.35, 95% interval 0.21 to 0.46), whereas P did not (-0.09, -0.26 to 0.07). In these tumours, the bulk-like Hoshida S1-S3 axis followed the composition of the tissue, not only the state of its tumour cells. The H targets were enriched in the tumour epithelium, whereas the P targets were shared with immune cells. The patient interpretation profiles (Supplementary Data 1) set the all-cell labels of each tumour beside their cellular sources and the subclass of its tumour epithelium. Files: the manuscript (PDF and DOCX), the Supplementary Methods and Results and the Supplementary Tables (DOCX), Supplementary Data 1 (HTML) and 2 (TSV), the graphical abstract (PNG and PDF), the figures (PNG, zip), the source core (scripts, run, environment and verification records, manuscript sources with the index of result files, search and download records, gene-list registry and reference tables; zip) and the result tables (zip). CS10-README.md describes the files; SHA256SUMS gives their checksums. Third-party materials (the GEO series, TCGA-LIHC data from the Genomic Data Commons, ICGC LIRI-JP, MSigDB gene sets, the published gene lists, Ensembl and HGNC reference tables and the repository and literature-database records) remain subject to their original licences and repository terms.
No takes yet. Share an insight, caveat, or question.
Young Min Park (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: