ANOVA results comparing nAUC between cell lines.

Detailed methods on cell culture, cell viability assays, RNAseq, western blot, and Elastic net regression analysis.

Relative expression of CD44, ITGA3, and Apo genes.

Demographics and clinical characteristics of exploratory patient cohort.

Primer sequences for genes used for RT-qPCR

qRT-PCR validation of gene expression representing relevant biological pathways after 90Y microsphere treatment in the most sensitive (PLC/PRF/5), resistant (SK-Hep1), and intermediate (SNU-398) cell lines. Select genes involved in (A) interferon stimulation and antigen presentation were mostly upregulated in both SNU-398 and SK-Hep1 after treatment, with the exception of GBP1 in SK-Hep1 and MX1 and IFI27 in SNU-398 which did not have reliable qPCR readouts. TXNIP was strongly upregulated in SK-Hep1 after treatment, consistent with increased oxidative stress signaling. (B) SNU-398 demonstrated very strong upregulation of inflammatory response genes, in particular CCL5 and TNFAIP2. C, Distinct patterns of DNA damage and cell stress genes were seen across cell lines, with SK-Hep1 showing upregulation of BRCA1 and downregulation of BNIP3. D, ECM genes CD44 and ITGA3 were strongly upregulated in SNU-398 after treatment, suggesting stress-induced EMT acquisition. Although there was slight downregulation of these genes in SK-Hep1, they remained at highest abundance in this line consistent with an EMT-associated expression profile. All experiments were performed in technical triplicate and error bars represent SEM when >1 biological replicate was performed. nd, no reliable qRT-PCR readout.

Differential expression and GSEA after 90Y treatment in individual cell lines.

Heterogeneity of response to 90Y microsphere treatment across human liver cancer cell lines. A, Dose–response curves of cell viability after 10-day treatment to escalating 90Y microsphere activities (0–20 MBq/mL) in each of 10 cell lines. Each point represents the mean surviving fraction relative to untreated baseline control across all independent experiments (error bars: SEM). B, For each experiment, the area under the dose–response curve was calculated and normalized to yield nAUC (0 = sensitive and 1 = resistant). Cell lines are ordered left to right by decreasing nAUC (increased sensitivity). Horizontal bar indicates mean nAUC for each cell line across experiments. Group differences were assessed by one-way ANOVA with a Tukey multiple comparisons test (*, P < 0.05; **, P < 0.01; ***, P < 0.001). C, Relationship between response to 90Y and established HCC transcriptomic subtypes. Cell lines were assigned to select HCC transcriptomic subtypes by nearest template prediction. nAUC distributions differed by subtype, with Hoshida S1 and C1 (cholangiocarcinoma-like) subtypes associated with 90Y resistance (P < 0.05, Kruskal–Wallis rank-sum test). No correlation with the hepatoblastoma HB-16 signature was observed. D, PCA of RNA baseline expression profiles of all cell lines demonstrates clustering of the five most resistant cell lines by nAUC (red: SK-Hep1, SNU-449, SNU-475, SNU-387, and SNU-423) along PC2/PC3 (13.1%/8.2% variance), with clear separation of the three most 90Y-sensitive cell lines (yellow: PLC/PRF/5, Hep3B, and HepG2) along PC2.

EMT and adhesion pathways associated with 90Y resistance. A, EN regression analysis identified 18 protein-encoding genes with nonzero coefficients for which expression was correlated with 90Y resistance across all cell lines at baseline (EN score >0.7). Among these, ITGA3 (EN = 0.911, R = 0.79) encodes the α subunit of the α3β1 integrin heterodimer, previously reported to influence HCC tumor progression and immune checkpoint expression. B, Differential expression analysis of RNA expression between 90Y-resistant (SK-Hep1, SNU-449, SNU-475, SNU-387, and SNU-423) and -sensitive (PLC/PRF/5, Hep3B, and HepG2) cell lines. Groupings defined a priori by nAUC Z-scores and baseline PCA. Volcano plot of log2 FC vs. −log10P value (FDR adjusted) of genes upregulated (red) and downregulated (blue) in 90Y-resistant vs. -sensitive cell lines. Genes involved in the extracellular matrix (ITGA3) and cancer stemness (CD44) were significantly upregulated in 90Y-resistant cell lines. C, GSEA of Hallmark pathways demonstrates strong upregulation of the EMT pathway in 90Y-resistant cell lines (mean log FC 8.9), which contains CD44 and ITGB1, the counterpart of ITGA3 in the a3b1 integrin heterodimer. Numbers next to each gene set bar represent FDR. D, Consistent with Hallmark EMT enrichment, KEGG and Reactome pathways associated with extracellular matrix and integrin cell surface interactions are enriched in resistant cell lines. E, qPCR (mean log2 FC, error bars represent SEM, with n = 2 biological replicates) and Western blot validation of ITGA3/a3b1 and CD44 confirming elevated expression of these genes in the most 90Y-resistant (SK-Hep1) vs. 90Y-sensitive (PLC/PRF/5) and intermediate (SNU-398) cell lines, consistent with an EMT/adhesion phenotype associated with 90Y resistance. F, qPCR of tumor vs. normal CD44 expression demonstrates a trend toward higher CD44 expression in those with IR or OFP (n = 5) vs. SR (n = 12), although not powered for statistical significance (P = 0.43 Mann–Whitney). Color circle indicates treatment intent: blue, radiation segmentectomy; orange, multicompartment dosimetry (MCD) with TAD > 205 Gy. MIRD, medical internal radiation dose. Relative expression from qPCR data = 2−(Cttarget − Cthousekeeping), in which Ct is the detection crossing threshold.

Molecular signatures predict prognosis in hepatocellular carcinoma, but their relevance to transarterial radioembolization (TARE) with yttrium-90 (90Y) is unknown. We aimed to identify baseline and treatment-induced pathways associated with response and nominate biomarkers. Ten transcriptomically diverse human liver cancer cell lines were exposed to escalating activities of glass 90Y microspheres for 10 days. Normalized AUC values quantified sensitivity. Whole-transcriptome RNA sequencing at baseline and after treatment was analyzed with elastic net regression and gene set enrichment. Findings were corroborated by qRT-qPCR and exploratory analysis of pretreatment tumor samples from patients undergoing TARE. Liver cancer cell line responses to 90Y were heterogeneous, with resistance aligning to Hoshida S1 and cholangiocarcinoma-like subtypes. Epithelial-mesenchymal transition (EMT) and adhesion pathways were enriched in resistant lines, with CD44 and ITGA3/α3β1 emerging as candidate markers, corroborated by RNA and protein expression. After 90Y exposure, resistant lines upregulated IFNγ/α, TNFα/inflammatory, and antigen presentation-related pathways, whereas sensitive lines downregulated these pathways along with DNA repair and oxidative phosphorylation. In an exploratory patient cohort, higher tumor CD44 expression trended with early progression. In conclusion, liver cancer cell lines display marked biological heterogeneity in response to 90Y. Baseline EMT/adhesion signatures and stress response pathways nominate CD44 and ITGA3/α3β1 as candidate biomarkers of resistance. These findings delineate molecular programs of β-emitter radioresistance and identify candidate pathways for future targeting. SIGNIFICANCE:TARE with 90Y is widely used for liver cancer, yet its molecular determinants of response are poorly understood. Using a diverse panel of liver cancer cell lines, we identify EMT, adhesion, and stress response pathways associated with resistance. These findings highlight candidate biomarkers and molecular vulnerabilities that may guide future therapeutic strategies and patient selection.

Details of liver cancer cell lines

Supplemental Figure S1. Perinatal deletion of intestinal HuR in the Apc min/+ background does not alter tumorigenesis.