Precision targeting of β-catenin induces tumor reprogramming and immunity in hepatocellular cancers

Abstract

First-line immune checkpoint inhibitor (ICI) combinations show responses in subsets of hepatocellular carcinoma (HCC) patients. Nearly half of HCCs are Wnt-active with mutations in CTNNB1 (encoding for β-catenin), AXIN1/2, or APC, and demonstrate heterogeneous and limited benefit to ICI due to an immune excluded tumor microenvironment. We show significant tumor responses in multiple β-catenin-mutated immunocompetent HCC models to a novel siRNA encapsulated in lipid nanoparticle targeting CTNNB1 (LNP-CTNNB1). Both single-cell and spatial transcriptomics reveal cellular and zonal reprogramming, along with activation of immune regulatory transcription factors IRF2 and POU2F1, re-engaged type I/II interferon signaling, and alterations in both innate and adaptive immunity upon β-catenin suppression with LNP-CTNNB1 at early- and advanced-stage disease. Moreover, ICI enhances response to LNP-CTNNB1 in advanced-stage disease by preventing T cell exhaustion and through formation of lymphoid aggregates (LA). In fact, expression of an LA-like gene signature prognosticates survival for patients receiving atezolizumab plus bevacizumab in the IMbrave150 phase III trial and inversely correlates with CTNNB1-mutatational status in this patient cohort. In conclusion, LNP-CTNNB1 is efficacious as monotherapy and in combination with ICI in CTNNB1-mutated HCCs through impacting tumor cell-intrinsic signaling and remodeling global immune surveillance, providing rationale for clinical investigations.

Fig. 1. RNAi-mediated β-catenin inhibition is efficacious in multiple immunocompetent CTNNB1-mutated HCC mouse models in early-stage disease setting.

a Schematic of CTNNB1-mutated patient-derived HCC organoid LNP treatment (20 nM). b (Left) Brightfield images at baseline, 48-, and 72-h. Scale bar represents 200 μm. (Right) Quantification of number of multi-cellular organoids per high powered field (HPF) and organoid diameter (n = 3 biological replicates per treatment and time point). c LNP treatment scheme in β-catenin-Nrf2 (β-N) model. Mice received once weekly intravenous (I.V.) injections at 1 mg/kg dosage starting at 5-weeks post-hydrodynamic tail vein injection (HDTVi). d Representative gross liver images of LNP-CTRL and LNP-CTNNB1 (1 mg/kg) treated β-N animals at 10.5-week timepoint. e, f Liver weights (p = 0.0008) and liver weight/body weight (LW/BW) (p = 0.0002) comparing LNP-CTRL (n = 6) and LNP-CTNNB1 (n = 4; 1 mg/kg) treated β-N animals. g (Left) Representative images of immunohistochemistry (IHC) for glutamine synthetase (GS)/Ki67 co-stain comparing LNP-CTRL and LNP-CTNNB1 (1 mg/kg) treated β-N animals. (Right) Quantification of %GS+ area comparing LNP-CTRL (n = 5) and LNP-CTNNB1 (n = 4). h LNP treatment scheme in β-catenin-hMet (β-M) model. Mice received once weekly I.V. injections at 1 mg/kg dosage starting at 3-weeks post-HDTVi. i Representative gross liver images of LNP-CTRL and LNP-CTNNB1 (1 mg/kg) treated β-M animals at 8.5-week timepoint. j, k Liver weights and LW/BW comparing LNP-CTRL (n = 3) and LNP-CTNNB1 (n = 7; 1 mg/kg) treated β-M animals. l (Left) Representative images of IHC for GS/Ki67 co-stain comparing LNP-CTRL and LNP-CTNNB1 (1 mg/kg) treated β-M animals. (Right) Quantification of %GS+ area comparing LNP-CTRL (n = 3) and LNP-CTNNB1 (n = 6). m LNP treatment scheme in β-catenin-Nrf2-hMet (β-N-M) model. Mice received once weekly I.V. injections at 1 mg/kg dosage starting at 3-weeks post-HDTVi. n Representative gross liver images of LNP-CTRL and LNP-CTNNB1 (1 mg/kg) treated β-N-M animals at 7.5-week timepoint. o, p Liver weights (p = 0.01) and LW/BW (p = 0.0098) comparing LNP-CTRL (n = 4) and LNP-CTNNB1 (n = 3; 1 mg/kg) treated β-N-M animals. q (Left) Representative images of IHC for GS/Ki67 co-stain comparing LNP-CTRL and LNP-CTNNB1 (1 mg/kg) treated β-N-M animals. (Right) Quantification of %GS+ area comparing LNP-CTRL (n = 3) and LNP-CTNNB1 (n = 3) (p = 0.0003). Created in BioRender. Lehrich (2025) https://BioRender.com/smaki7g. Lehrich (2025) https://BioRender.com/858w237. Lehrich (2025) https://BioRender.com/6ce7tob. Lehrich (2025) https://BioRender.com/9127h6y. For (b), (e–g), (j–l), and (o–q), data presented as mean values ± standard deviation (SD) and P-values calculated by unpaired two-tailed Student’s t-test. Source data are provided as a Source Data File. For (g), (l), (q), scale bar indicates magnification. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 2. Earliest biological response to RNAi-mediated β-catenin inhibition is observed 3-days following LNP treatment.

a LNP treatment scheme in β-catenin-Nrf2 (β-N) model. b (Left) Representative gross liver images and (Right) liver weight/body weight (LW/BW) comparing LNP-CTRL (n = 4) and LNP-CTNNB1 (n = 4; 1 mg/kg) β-N treated animals 3-days post 1st LNP treatment. c qPCR RNA expression levels of Ctnnb1 and β-catenin target genes (Glul, Ccnd1, Lect2, Rgn) between LNP-CTRL (n = 3) and LNP-CTNNB1 (n = 3) β-N treated animals. d, e Representative immunohistochemistry (IHC) images for glutamine synthetase (GS), Ki67, and TUNEL comparing LNP-CTRL and LNP-CTNNB1 treated β-N animals 3-days post 1st LNP treatment. Scale bar represents 100 μm. f Quantification of number of Ki67- and TUNEL-positive cells across multiple high-power fields (HPF) between LNP-CTRL (n = 3) and LNP-CTNNB1 (n = 3) treated β-N animals 3-days post 1st LNP treatment (p = 0.0006). g Principal component analysis of bulk RNA-sequencing transcriptomic profiles of β-N and β-M model treated with LNP-CTRL or LNP-CTNNB1 and harvested 3-days post-LNP treatment, using all genes (n = 3–4 per condition and model). h Venn diagram highlighting number of common downregulated differentially expressed genes (DEGs) (n = 230) between β-N and β-M models treated with LNP-CTRL or LNP-CTNNB1 3-days post-LNP treatment. i Venn diagram highlighting number of common upregulated DEGs (n = 73) between β-N and β-M models treated with LNP-CTRL or LNP-CTNNB1 3-days post-LNP treatment. DEGs defined by FDR = 0.05 and fold change >1.5. j Heatmap of selected common downregulated and upregulated genes demonstrating normalized z-score expression value in each model with each LNP treatment condition from (h) and (i). k Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway gene set enrichment analysis (GSEA) in β-N model comparing LNP-CTRL and LNP-CTNNB1 treated animals. l KEGG pathway GSEA in β-M model comparing LNP-CTRL and LNP-CTNNB1 treated animals. For (k, l), normalized enrichment score (NES) was normalized from ES, which was calculated by Kolmogorov–Smirnov-like statistic and the p-value was calculated using the one-tailed empirical permutation test procedure. Created in BioRender. Lehrich (2025) https://BioRender.com/kymtir0. For (b), (c), and (f), data are presented as mean values ± standard deviation (SD) and P-values calculated by unpaired two-tailed Student’s t-test. Source data are provided as a Source Data File. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 3. Integrated single-cell analyses reveal de novo formation of reprogrammed hepatocytes within remnant tumor nodules 3-days post-LNP-CTNNB1 treatment.

a Uniform manifold approximation and projection (UMAP) visualization of single-cell RNA-seq data following liver perfusion and enrichment of hepatocyte cell populations from LNP-CTRL and LNP-CTNNB1 treated β-catenin-Nrf2 (β-N) animals 3-days post-LNP treatment. UMAP split by treatment condition with 94,650 cells total across both treatment conditions. LNP-CTRL (n = 2) has 26,851 cells in the library; LNP-CTNNB1 (n = 3) has 67,799 cells in the library after data integration. b Pie chart of cell-type proportions between LNP-CTRL and LNP-CTNNB1 treatment conditions from (a). c Cell cycle regression scoring visualized via pie charts depicting cell cycle phase proportions in each of the indicated hepatocyte cell clusters. Each pie slice represents a group of cells colored by whether the RNA expression fits single cells belonging to G1 (red), S (green), or G2M (blue) phases of the cell cycle. d Pseudotime trajectory analysis on UMAP plot subset to only hepatocyte specific cell populations using the Zone 3 CTNNB1 WT and MUT (GS+) cell cluster as the root. e UMAP visualization of single-cell spatial transcriptomic data via Molecular Cartogrpahy^TM platform taken from frozen liver tissue sections of LNP-CTRL (n = 1) and LNP-CTNNB1 (n = 1) treated β-N animals 3-days post-treatment. UMAP generated based on expression of 100 genes. UMAP split by treatment condition with 19,301 cells total across both treatment conditions (LNP-CTRL library has n = 6 regions of interest (ROIs) with 10,227 cells total; LNP-CTNNB1 library has n = 5 ROIs with 9074 cells total). Labeled cell populations indicated by color. f Pie chart of cell-type proportions between LNP-CTRL and LNP-CTNNB1 treatment conditions from (e). g Spatial plots of LNP-CTRL and LNP-CTNNB1 regions of interest demonstrating visualization of certain cell populations by color from (e, f) on a virtual tissue section. h Dot plot visualization of various zonated marker gene expression (for all zones 1–3) for each hepatocyte cluster from (e, f). i Pseudotime trajectory analysis on Uniform manifold approximation and projection (UMAP) plot using the H1: Zone 3 CTNNB1 MUT (GS+) cluster as the root/origin. For (b), (c), and (f) labeled cell populations indicated by color and Source data are provided as a Source Data File.

Fig. 4. RNAi-mediated β-catenin inhibition induces infiltration of pro-inflammatory M1-like macrophages due to re-engaged interferon signaling.

a UMAP visualization of integrated single-cell RNA-seq data following liver perfusion and enrichment of immune populations from LNP-CTRL (n = 3) and LNP-CTNNB1 (n = 3; 1 mg/kg) treated β-catenin-Nrf2 (β-N) animals 3-days post-LNP treatment. UMAP split by LNP treatment condition. b Stacked bar plot of cell-type proportions between LNP-CTRL (n = 8499 cells) and LNP-CTNNB1 (n = 11,736 cells) from (a). Labeled cell populations indicated by color for (a, b). c Dot plot visualization of expression of M1-like and M2-like macrophage phenotype markers. d Bar plot comparing percentage of M1-like macrophages amongst the total cell population in the LNP-CTRL (n = 3) and LNP-CTNNB1 (n = 3; 1 mg/kg) groups (p = 0.0653). e Gene Ontology (GO) pathway gene set enrichment analysis (GSEA) in the M1-like macrophage population comparing LNP-CTRL and LNP-CTNNB1 groups. f Stacked horizontal bar plot comparing relative information flow from CellChat between LNP-CTRL and LNP-CTNNB1 (1 mg/kg) groups. Boxed pathways show 100% information flow in LNP-CTNNB1 (1 mg/kg) group. IFN-II signaling highlighted in light blue show 100% enriched in LNP-CTNNB1 (1 mg/kg) group. g Chord diagram for IFN-II pathway in LNP-CTNNB1 (1 mg/kg) group demonstrating information flow from proliferative T cells to macrophage populations. No information flow in LNP-CTRL treated animals. h IFNγ treatment schematic in β-catenin-hMet (β-M) model. Mice received multi-weekly intra-peritoneal (I.P.) injections of IFNγ at 1 × 10⁶ IU/ml dosage or vehicle control starting at 3-weeks post-hydrodynamic tail vein injection (HDTVi). Mice were sacrificed at 7.5-weeks post-HDTVi. i, j Representative gross liver images and liver weight/body weight (LW/BW) comparing β-M animals treated with either vehicle control (n = 3) or IFNγ (n = 8) at 7.5-week timepoint (p = 0.0086). k Representative immunohistochemistry (IHC) images for S100A8/9 comparing Vehicle and IFNγ treated β-M animals. 10× objective lens, scale bar is 100 μm. IHC was repeated at least twice on tissue sections from multiple animals. Created in BioRender. Lehrich (2025) https://BioRender.com/9a8wkfh. For (d, j), data presented as mean values ± standard deviation (SD) and P-values calculated by unpaired two-tailed Student’s t-test. For (e), normalized enrichment score (NES) was normalized from ES, which was calculated by Kolmogorov–Smirnov-like statistic and the adjusted p-value was calculated using the one-tailed empirical permutation test procedure. Source data are provided as a Source Data File. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 5. IRF2 and POU2F1 repression by mutated-β-catenin is a major tumor cell-intrinsic mechanism of immune exclusion in CTNNB1-mutated HCC.

a Schematic of pipeline comparing whole transcriptomes of β-catenin-mutated HCC to β-catenin knockout livers. b Dot plot highlighting Irf2 and Pou2f1 target genes within the Zone 3 CTNNB1 WT and MUT (GS+) cell population between LNP-CTRL and LNP-CTNNB1 from Fig. 3a, b. c Heatmap of z-scored expression values of IRF2/POU2F1 target genes in TCGA-LIHC patients (n = 374) and adjacent normal (n = 50). Data stratified by CTNNB1- (n = 98), AXIN1- (n = 18), and APC-mutated patients (n = 3). d Boxplots of normalized expression of IRF2, POU2F1, and IRF2/POU2F1 target genes stratified by adjacent normal (n = 50), Wnt/β-catenin-mutant (n = 119), and -wild-type (n = 255). e Schematic of β-catenin-hMet (β-M) animals co-injected with pT3 or IRF2 at time of hydrodynamic tail vein injection (HDTVi) and sacrificed at 7.5-weeks post-HDTVi. f Representative gross liver images from β-M-pT3 and β-M-IRF2 animals. Scale bar indicates 1 centimeter (cm). g, h Liver weights (p = 0.0008) and liver weight/body weight (LW/BW) (p = 0.0003) comparing β-M-pT3 (n = 7) and β-M-IRF2 (n = 12) animals at 7.5-week timepoint. i Schematic of β-catenin-Nrf2 (β-N) animals co-injected with pT3 or POU2F1 at time of HDTVi and sacrificed at 10.7-weeks post-HDTVi. j Representative gross liver images from β-N-pT3 and β-N-POU2F1 animals. k, l Liver weights (p = 0.0013) and LW/BW (p = 0.0005) comparing β-N-pT3 (n = 4) and β-N-POU2F1 (n = 4) animals at 10.7-week timepoint. m Representative IHC images for CD4, CD8, and CD20 comparing β-N-pT3 and β-N-POU2F1 animals at 10.7-week timepoint. n GO pathway gene set enrichment analysis comparing β-M-POU2F1 to β-M-pT3. Created in BioRender. Lehrich (2025) https://BioRender.com/25qvbwj. Lehrich (2025) https://BioRender.com/qqvvb4x. Lehrich (2025) https://BioRender.com/ecs8omd. For (g, h), (k, l), data presented as mean values ± standard deviation (SD) and P-values calculated by unpaired two-tailed Student’s t-test. For (d), the center line shows the median, the box limits show the interquartile range (IQR; the range between the 25th and 75th percentile) and the whiskers show 1.5× IQR. For (d), One-way ANOVA with Tukey-HSD post-hoc adjusted p-values comparing Wnt/β-catenin-mutant vs wild-type are: p = 0.80, p = 0.54, and p = 0.009 for IRF2, POU2F1, and IRF2/POU2F1 target gene expression, repsectively. For (n), NES was normalized from ES, which was calculated by Kolmogorov–Smirnov-like statistic and the adjusted p-value was calculated using the one-tailed empirical permutation test procedure. Source data are provided as a Source Data File. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 6. Lack of response to RNAi-mediated β-catenin inhibition in CTNNB1-mutated HCC mouse models in advanced-stage disease setting is due to expansion of clones derived from the secondary driver.

a LNP treatment schematic in β-catenin-hMet (β-M) model for advanced-stage disease. Mice received once weekly intravenous (I.V.) injections at 1 mg/kg dosage starting at 6-weeks post-hydrodynamic tail vein injection (HDTVi) and sacrificed at 10.5-weeks post-HDTVi. b Representative gross liver images of LNP-CTNNB1 treated β-M animals at 10.5-week timepoint demonstrating non-responders (NR) and responders (R) compared to LNP-CTRL β-M animals when moribund at ~8.5-weeks. c Liver weight/body weight (LW/BW) comparing LNP-CTRL (n = 4) and LNP-CTNNB1 (n = 8) treated β-M animals at 8.5-week and 10.5-week timepoint, respectively (p = 0.0373). d LNP treatment schematic in β-catenin-Nrf2 (β-N) model for advanced-stage disease. Mice received once weekly intravenous (I.V.) injections at 1 mg/kg dosage starting at 8-weeks post-HDTVi. e Representative gross liver images of LNP-CTNNB1 treated β-N animals at 13.5-week timepoint demonstrating NR and R compared to LNP-CTRL β-N animals when moribund at ~10.5-weeks. f LW/BW comparing LNP-CTRL (n = 6) and LNP-CTNNB1 (n = 8) treated β-N animals at 10.5-week and 13.5-week timepoint, respectively (p = 0.0005). g (Top) Spatial plot of all 17 clusters from the uniform manifold approximation and projection (UMAP) visualization showing the spatial localization on the H&E tissue section. (Bottom) Spatial plot of cluster 3 ident on the H&E tissue section. h Stacked bar chart of cluster proportions for the 3 LNP response (Control, NR, and R) groups. Asterisk denotes cluster 3. i (Top) Spatial plot of all 13 spot clusters from the UMAP visualization showing the spatial localization on the H&E tissue section. (Bottom) Spatial plot of cluster 11 ident on the H&E tissue section. j Stacked bar chart of cluster proportions for the 3 LNP response (Control, NR, and R) groups. Asterisk denotes cluster 11. k Waterfall plot of predicted kinases upstream of the differentially expressed genes (DEGs) in cluster 3 from the β-M model. l Waterfall plot of predicted transcriptional regulators upstream of the DEGs in cluster 11 from the β-N model. Created in BioRender. Lehrich (2025) https://BioRender.com/63585xs. Lehrich (2025) https://BioRender.com/6yyrkql. For (c, f), data presented as mean values ± standard deviation (SD) and P-values calculated by unpaired two-tailed Student’s t-test. Source data are provided as a Source Data File. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 7. ICI enhances response to RNAi-mediated β-catenin inhibition in β-M CTNNB1-mutated mouse model in advanced-stage disease setting.

a Dot plot of T cell markers (Cd3d, CD3e, Cd3g, Cd4, Cd8a) in clusters 9 and 12 stratified by LNP treatment response (LNP-CTRL [blue]; non-responder [NR; red], responder [R; green]) in β-catein-hMet (β-M) from Fig. 6g, h. b, c Representative immunohistochemistry (IHC) images for CD3 and CD4 in β-M animals treated with LNP-CTRL or LNP-CTNNB1. 5× objective magnification. Scale bar indicates 200 μm. d Dot plot of Cd274 expression by LNP treatment response in β-M. e Schematic of LNP + IgG/α-PD1 in β-M model. Mice received weekly LNP injections at 1 mg/kg dosage starting at 6-weeks post-hydrodynamic tail vein injection (HDTVi) with twice weekly injections of IgG/α-PD1 (200 μg) for two weeks starting 3-days after first LNP treatment. f Representative gross liver images of LNP-CTNNB1 ± IgG/α-PD1 treated β-M animals at 10.5-week timepoint compared to LNP-CTRL ± IgG/α-PD1 when moribund. g Liver weight/body weight (LW/BW) comparing LNP-CTNNB1 ± IgG/α-PD1 (n = 8/n = 8) treated β-M animals at 10.5-week timepoint to LNP-CTRL ± IgG/α-PD1 (n = 4/n = 4) when moribund. h Magnetic resonance images (MRI) of LNP-CTNNB1 ± IgG/α-PD1 treated β-M animals at 10.5-week timepoint. i RNA expression levels by qPCR of mCTNNB1 and hCTNNB1 in β-M animals treated with LNP-CTNNB1 ± IgG/α-PD1 (n = 8). j Representative tiled IHC images for granzyme B (GZMB) from LNP-CTNNB1 ± IgG/α-PD1 treated β-M animals at 10.5-week timepoint. Scale bar indicates magnification. k Violin plot showing quantification of number of GZMB+ lymphoid aggregates per tissue section in tumoral regions from LNP-CTNNB1 ± IgG/α-PD1 (n = 8) treated β-M animals at 10.5-week timepoint. l Kaplan–Meier overall survival (OS) curve comparing β-M animals treated with LNP-CTRL ± IgG/α-PD1 (n = 3/n = 4) or LNP-CTNNB1 ± IgG/α-PD1 (n = 10/n = 8) (p = 0.0188). m Schematic of two-part mechanistic working model for LNP-CTNNB1 response in β-catenin-mutated HCC. Created in BioRender. Lehrich (2025) https://BioRender.com/z4h1j67. Lehrich (2025) https://BioRender.com/hjjd4zn. For (g, i), data presented as mean values ± standard deviation (SD). For (g), P-values calculated by one-way ANOVA with Tukey-HSD post-hoc correction. For (i, k), P-values calculated by unpaired two-tailed Student’s t-test. For (l), P-value calculated by log-rank test to compare difference in mean OS time between LNP-CTNNB1 + IgG and LNP-CTNNB1 + α-PD1. Source data are provided as a Source Data File. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Fig. 8. Lymphoid aggregates are prognostic in hepatocellular carcinoma and negatively correlated with CTNNB1 mutational status.

a Stacked bar plot of number of patients in IMbrave150 phase III trial having either tertiary lymphoid structure (TLS), lymphoid aggregate (LA), diffuse infiltrate (DI), or none (n = 174). b Stacked bar plot of number of patients in IMbrave150 phase III trial having either TLS, LA, DI, or None stratified by treatment group and clinical response: atezolizumab + bevacizumab (AtezoBev; n = 116) versus sorafenib (Sor; n = 58). c (Top) Kaplan–Meier progression-free survival (PFS) and overall survival (OS) curves comparing patients with TLS/LA in atezolizumab + bevacizumab versus sorafenib arms. (Bottom) Kaplan–Meier PFS and OS curves comparing patients with DI/None in atezolizumab + bevacizumab versus sorafenib arms (n = 174; AtezoBev, n = 116; Sor, n = 58). d Expression of a “LA-like” gene signature stratified by whether patients in IMbrave150 phase III trial had TLS/LA or DI/None (n = 174; p = 0.025). e Expression of a “LA-like” gene signature stratified by whether patients in IMbrave150 phase III trial had TLS, LA, DI, or None (n = 174; p = 0.0095). f (Left) Kaplan–Meier PFS curve comparing patients with high LA-like gene signature in atezolizumab + bevacizumab versus sorafenib arms (n = 88; AtezoBev, n = 59; Sor, n = 29). (Right) Kaplan–Meier OS curve comparing patients with high LA-like gene signature in atezolizumab + bevacizumab versus sorafenib arms (AtezoBev, n = 59; Sor, n = 29). g Box plot depicting “LA-like” gene signature stratified by CTNNB1 mutational status in all patients and within each of the two treatment arms from IMbrave150 phase III trial (n = 129; p = 0.002). For (c, f), hazard ratios (HRs) and 95% confidence intervals (CIs) were determined using a univariate Cox model. Kaplan–Meier log-rank test was used to compare differences in PFS/OS outcomes. For the boxplots in (d, e, g), the center line shows the median, the box limits show the interquartile range (IQR; the range between the 25th and 75th percentile) and the whiskers show 1.58× IQR. For (d, g), P-values were calculated using two-tailed unpaired t-test. For (e), P-value was calculated using one-way analysis of variance (ANOVA).