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.

   Subject terms: Cancer microenvironment, Targeted therapies,
   Hepatocellular carcinoma
     __________________________________________________________________

   Gain-of-function mutations in CTNNB1 (encoding for b-catenin) leading
   to deregulated Wnt/β-catenin signaling are frequently observed in
   patients with hepatocellular carcinoma (HCC). Here the authors show
   that inhibiting b-catenin with lipid nanoparticles encapsulating siRNA
   targeting CTNNB1 impairs tumor growth and promotes anti-tumor immunity
   in preclinical HCC models.

Introduction

   Hepatocellular carcinoma (HCC) is the third leading cause of
   cancer-related death globally^[93]1. Despite the shift in therapeutic
   management of advanced disease over the last five years from
   multi-tyrosine kinase inhibitors (TKIs) (e.g., sorafenib) to
   immunotherapy with immune-checkpoint inhibitor (ICI) combinations
   (e.g., atezolizumab plus bevacizumab), objective response rates (ORRs)
   remain low at ~30% with overall survival <2 years^[94]2–[95]5.
   Preclinical and clinical studies investigating molecular correlates of
   ICI response have yielded novel insights into potential mechanisms of
   resistance, including but not limited to immune exclusion, with
   Wnt/β-catenin activation contributing to this phenotype^[96]6–[97]8.
   Wnt/β-catenin pathway activity is observed in up to 50% of tumors from
   patients with HCC, with mutations mostly occurring in CTNNB1 (26–37%),
   AXIN1/2 (8–10%), and APC (3–5%)^[98]9–[99]12. Gain-of-function (GOF)
   mutations in CTNNB1 (encoding for β-catenin) are one of the major trunk
   mutational events in HCC and occur mostly as missense mutations in exon
   3 at serine and threonine residues or the ubiquitination destruction
   motif, which interfere with its degradation, leading to constitutive
   β-catenin activation and target gene transcription^[100]13,[101]14.
   Patients with CTNNB1-mutated HCC have upregulation of known
   Wnt/β-catenin target genes, including GLUL, AXIN2, LGR5, and
   TBX3^[102]11. In fact, glutamine synthetase (GS; encoded by GLUL)
   immunohistochemistry is used as a biomarker for patients with
   CTNNB1-mutated HCC^[103]15. However, targeting these downstream Wnt
   target genes has revealed novel negative feedback loops in the
   Wnt/β-catenin oncogenic circuit^[104]16,[105]17, necessitating the need
   to focus on targeting β-catenin directly for precision therapy.

   Despite improved molecular stratification of HCC over the last decade,
   with recognition of Wnt/β-catenin-driven tumors overlapping with
   Hoshida S3^[106]18 or Boyault G5/G6 subclasses^[107]19, these different
   molecular stratifications have not yielded prognostic implications due
   to a lack of clinically approved targeted or biomarker-driven precision
   therapeutics. β-catenin has traditionally been an undruggable target,
   despite preclinical studies elucidating the molecular and metabolic
   addiction to β-catenin oncogenic signaling in CTNNB1-mutated
   HCC^[108]20–[109]23. Thus, β-catenin is a prime target for precision
   therapy. Advances in RNAi technology over the last two decades have
   resulted in multiple approved RNAi therapies^[110]24, and RNAi-mediated
   gene silencing has proven to be an excellent tool for targeting the
   traditionally undruggable, especially in hepatic tissue^[111]25.

   In this work, we investigate the relevance of RNAi-mediated β-catenin
   inhibition in patient-derived CTNNB1-mutated HCC organoids and multiple
   humanized mouse models of CTNNB1-mutated HCC at different treatment
   windows and elucidate the underlying mechanisms of response in both
   hepatic and immune compartments through both single-cell and spatial
   transcriptomic approaches. Our findings provide a mechanistic basis for
   clinical investigations of this RNAi therapeutic targeting β-catenin
   for HCC treatment as an innovative treatment paradigm in the form of
   monotherapy and/or in combination with ICI in human subjects belonging
   to the Wnt-β-catenin active-HCC subclass.

Results

RNAi-mediated β-catenin inhibition results in potent CTNNB1 knockdown in
vitro and in vivo

   To study the effects of RNAi-mediated inhibition in β-catenin-mutated
   HCC, we utilized a novel siRNA that targets the CTNNB1 gene, with both
   mouse and human specificity, encapsulated in a lipid nanoparticle
   (referred hereafter as LNP-CTNNB1). We first assessed whether
   LNP-CTNNB1 affected growth in a patient-derived HCC organoid (23277)
   with known mutation in CTNNB1^[112]26. After 48- and 72-h treatment
   with LNP-CTNNB1 at 20 nM concentration, we observed a significant
   decrease in both the numbers and the size of the organoids compared to
   treatment with the LNP-CTRL (Fig. [113]1a, b). Thus, LNP-CTNNB1
   demonstrates efficacy in mutant-CTNNB1 human HCC organoid cultures.

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

   [114]Fig. 1
   [115]Open in a new tab

   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) [116]https://BioRender.com/smaki7g. Lehrich (2025)
   [117]https://BioRender.com/858w237. Lehrich (2025)
   [118]https://BioRender.com/6ce7tob. Lehrich (2025)
   [119]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.

   Next, to assess its pharmacodynamic effects, we first delivered
   LNP-CTNNB1 via tail vein intravenous (I.V.) injection to mouse livers
   which were transfected with human S45Y-mutant-CTNNB1 gene (S45Y-hCTNNB1
   mice) via sleeping beauty-hydrodynamic tail vein injection (SB-HDTVi)
   system. We have previously reported that mouse hepatocytes
   overexpressing mutant-β-catenin alone via SB-HDTVi do not develop
   HCC^[120]27, but require a secondary driver like hMet, mutant-Kras, or
   mutant-Nrf2, to induce HCC^[121]20,[122]27,[123]28. After 4 treatments
   at 3 mg/kg dosing in S45Y-hCTNNB1 mice (Supplementary Fig. [124]1a), we
   observed an appreciable decrease in liver weight and liver weight to
   body weight ratio (LW/BW), which is consistent with the role of
   β-catenin in regulating liver growth and size (Supplementary
   Fig. [125]1b–d)^[126]29,[127]30. Additionally, expression of two
   well-known β-catenin target genes GS and Cyclin D1 (CCND1) via
   immunohistochemistry (IHC) were absent throughout the liver lobule
   indicating high mCTNNB1 gene knockdown (Supplementary Fig. [128]1e, f).
   Concomitantly, Myc-tag (present on the S45Y-hCTNNB1 plasmid) positive
   cells were absent in the LNP-CTNNB1 treated mice compared to islands of
   Myc-tag positive cells in the LNP-CTRL mice (Supplementary
   Fig. [129]1e, f). Thus, LNP-CTNNB1 targets both endogenous mouse and
   mutant human CTNNB1 with high potency and specificity in vivo.

   Prior to testing efficacy of siRNA-mediated CTNNB1 knockdown in HCC, we
   assessed whether there were any effects of the LNP itself on the tumor
   immune microenvironment (TIME). We treated mice injected with
   T41A-β-catenin-G31A-Nrf2 (β-N model) with either PBS, LNP-CTRL, or
   LNP-CTNNB1 (Supplementary Fig. [130]1g) utilizing a similar frequency
   and dosage scheme as in Supplementary Fig. [131]1a. β-N model has been
   previously shown to represent 9–12% of all human HCC^[132]28. Following
   treatment, we observed a decrease in liver weight and LW/BW in
   LNP-CTNNB1 treated mice (Supplementary Fig. [133]1h, i) with no
   differences in liver serum biochemistries (Supplementary Fig. [134]1j).
   Next, we performed bulk RNA-sequencing across all 3 treatment groups,
   and observed that PBS and LNP-CTRL treated animals are
   transcriptionally very similar, and distinct from the LNP-CTNNB1
   treated group (Supplementary Fig. [135]1k). Additionally, gene set
   variation analysis using gene ontology (GO) pathways demonstrated that
   the immune phenotype is similar between PBS and LNP-CTRL treated mice,
   suggesting that the LNPs themselves do not influence the TIME in the
   CTNNB1-mutated HCC (Supplementary Fig. [136]1l).

RNAi-mediated β-catenin inhibition impairs tumor growth in multiple
immunocompetent CTNNB1-mutated and non-CTNNB1-mutated HCC mouse models with
durable response in early-stage disease setting

   We next evaluated the in vivo efficacy of LNP-CTNNB1 in CTNNB1-mutated
   and non-mutated HCC models. We first performed a dose titration study
   to determine the lowest efficacious dose in the β-N model. We
   administered once weekly I.V. injections at 3, 1, 0.3, 0.1, and
   0.03 mg/kg dosages over 6 weeks of LNP-CTNNB1 starting at 5-weeks
   post-HDTVi, which we previously determined as a timepoint when
   microscopic tumor foci are already established^[137]28 (Supplementary
   Fig. [138]2a). There were significant reducations in the tumor burden
   across a wide dose range of the LNP-CTNNB1 (3, 1, 0.3, and 0.1 mg/kg),
   as evident grossly, by liver weight, and by LW/BW (Supplementary
   Fig. [139]2b–d, Fig. [140]1c–f). However, at 3 mg/kg dosage, following
   the 4th dose, we observed mortality in one of four mice, which was
   likely due to the high LNP dose and frequency. Additionally, the 0.3,
   0.1, and 0.03 mg/kg LNP-CTNNB1 dosages resulted in partial responses,
   with remnant microscopic tumor foci observed in 0.3 and 0.1 mg/kg
   treated animals (Supplementary Fig. [141]2e) and macroscopic tumor
   nodules present in animals treated with 0.03 mg/kg (Supplementary
   Fig. [142]2b, e). At the 1 mg/kg LNP-CTNNB1 dosage there was no
   morbidity or mortality observed in these mice. Additionally, there were
   no gross phenotypic changes to other organs including lungs, spleen,
   intestine, and heart. H&E of the spleens across a wide dose range did
   not demonstrate any microscopic changes (Supplementary Fig. [143]2f).
   We did not observe any gross neurological, gastrointestinal,
   genitourinary, cardiovascular, or respiratory deficits and/or distress
   following the once weekly treatments over 6 weeks, similar to what has
   been noted in rodents previously across a broad dose range, along with
   absence of any adverse signs or toxicity including any alterations in
   body weight or liver function tests^[144]31. Significant tumor
   responses were observed at 1 mg/kg LNP-CTNNB1 dosage as noted via H&E,
   IHC for Myc-tag and GS/Ki67, and magnetic resonance imaging (MRI)
   (Fig. [145]1g; Supplementary Fig. [146]3a–d). As a result, we utilized
   the 1 mg/kg LNP-CTNNB1 dose for treatment of β-catenin-mutated HCC
   preclinical models.

   To extrapolate our findings to additional β-catenin-mutated HCC
   preclinical models that we have previously reported, we next tested
   LNP-CTNNB1 in the S45Y-mutant-β-catenin-Met (β-M) model, which
   represents 11% of human HCC^[147]27. Treatment was initiated at 3-weeks
   post-HDTVi when microscopic tumor foci are already established and
   based on the more aggressive tumor phenotype in this model, as
   determined by us previously^[148]27. Once weekly I.V. administration at
   1 mg/kg LNP-CTNNB1 dosage over 6 weeks led to a significant decrease in
   tumor burden grossly (Fig. [149]1h–k), and histologically as observed
   via H&E, Myc-tag, and GS/Ki67 IHC (Fig. [150]1l; Supplementary
   Fig. [151]3e, f). Lastly, starting at 3-weeks post-HDTVi, a timepoint
   with known microscopic tumor burden^[152]32, we tested 1 mg/kg
   LNP-CTNNB1 dosage in a third CTNNB1-mutated model, the
   S45Y-mutant-β-catenin-Nrf2-Met (β-N-M) model, which represents another
   unique ~5% subset of human HCC^[153]32. Following a similar treatment
   protocol, we again observed significant tumor responses grossly and
   microscopically (Fig. [154]1m–q; Supplementary Fig. [155]3g, h).

   Next, we wanted to assess response to LNP-CTNNB1 in models that were
   not CTNNB1-mutated, due to the general mitogenic function of
   Wnt/β-catenin signaling pathway in the liver^[156]33. β-Catenin
   suppression by LNP-CTNNB1 in the Nrf2-hMet (N-M) model at 8-weeks
   post-HDTVi, a timepoint with known microscopic tumor burden^[157]32,
   led to a decrease in liver weight and LW/BW and in macroscopic disease
   (Supplementary Fig. [158]4a–d), although there was persistence of some
   microscopic nodules, which depicted inferiority in response when
   compared to mutant-β-catenin-driven tumors (Supplementary
   Fig. [159]4e–g). This decrease in tumor burden was observed despite HCC
   nodules in this model not being homogenously positive for the bonafide
   Wnt target GS (Supplementary Fig. [160]4e). We have also previously
   reported that c-Met/sgAxin1 tumors require intact β-catenin to initiate
   tumorigenesis^[161]34. We also tested dependence on β-catenin in
   another independent non-CTNNB1-mutated HCC model using a genetic
   approach (Supplementary Fig. [162]4h). β-Catenin deletion in SB-HDTVi
   induced Akt-NRas HCC in β-catenin floxed mice through simultaneous
   administration of pCMV-cre or control led to a significant improvement
   in overall survival and less tumor burden in pCMV-Cre compared to
   control, although tumors still persisted (Supplementary Fig. [163]4i).
   Thus, overall, we observed that β-catenin inhibition alone for
   CTNNB1-mutated HCC is most effective in early-stage disease setting as
   evident through significant tumor responses in multiple models of
   CTNNB1-mutated HCC, and as partial responses in β-catenin non-mutated
   HCC models.

   Lastly, we assessed the long-term durability of the significant tumor
   responses and overall survival (OS) in both the β-N and β-M models
   following LNP-CTNNB1 treatment at 1 mg/kg dosage. Following the same
   treatment protocol in β-N (Fig. [164]1c) and β-M (Fig. [165]1h) models,
   we then withdrew LNP-CTNNB1. In the β-N model, following treatment
   cessation, mice were moribund by ~22.5-weeks post-LNP-CTNNB1 treatment,
   with gross tumor burden becoming equivalent to the tumor burden
   observed in LNP-CTRL treated mice at ~10.5 weeks, which is a lethal
   timepoint in β-N model (Supplementary Fig. [166]5a, b). Thus, with
   LNP-CTNNB1 treatment in β-N model, OS was significantly extended by ~12
   weeks (p < 0.01) (Supplementary Fig. [167]5c). The nodules that
   reappeared at the ~22.5-week timepoint were positive for both GS and
   Nqo1 (Nrf2-target) (Supplementary Fig. [168]5d). Similarly, in the β-M
   model, following treatment cessation, mice were moribund by ~16.5-weeks
   post-LNP-CTNNB1 treatment, with gross tumor burden becoming equivalent
   to the tumor burden observed in LNP-CTRL treated mice at ~7.5 weeks,
   which is a lethal timepoint in β-M model (Supplementary Fig. [169]5e,
   f). Thus, LNP-CTNNB1 treatment in the β-M model extended OS by ~9 weeks
   (p < 0.001) (Supplementary Fig. [170]5g). The nodules that reappeared
   at ~16.5-week timepoint in β-M model were also positive for GS and
   V5-tag (present on hMet plasmid) (Supplementary Fig. [171]5h). Overall,
   LNP-CTNNB1 treatment as monotherapy more than doubled the OS of mice in
   both HCC models although tumors recurred after treatment cessation.
   These recurring tumors appear to be mutant-β-catenin-driven and not due
   to appearance of any de novo resistant clones.

Earliest biological response to RNAi-mediated β-catenin Inhibition observed
at 3-days following initial LNP-CTNNB1 treatment

   Given the robust tumor responses following LNP-CTNNB1 treatment, we
   investigated the earliest biological response observed following
   β-catenin knockdown within the tumor cells. In the β-N model, we
   followed mice over a 3-week treatment course (LNP-CTNNB1 injected
   weekly ×3) and sacrificed mice at 1-, 3-, 5-, 7-, 14-, and 21-days post
   the first LNP treatment (Fig. [172]2a). Over this 21-day treatment time
   course, the visible tumor foci or LW/BW progressively trended lower in
   the LNP-CTNNB1 group although differences were insignificant (except
   day 5) when compared to time-matched LNP-CTRL group (Supplementary
   Fig. [173]6a; Fig. [174]2b). However, at 3-days after a single
   LNP-CTNNB1 dose, RNA expression of Ctnnb1, along with Wnt target genes,
   Glul, Ccnd1, Lect2, and Rgn were significantly decreased in LNP-CTNNB1
   compared to LNP-CTRL mice (Fig. [175]2c). Additionally, GS protein
   visualized via IHC was decreased in tumor nodules but retained in
   pericentral hepatocytes at the 3-day timepoint, while it was absent
   also in the pericentral hepatocytes by 14-days in the LNP-CTNNB1 group
   (Fig. [176]2d; Supplementary Fig. [177]6b). Ki67 and TUNEL IHC also
   demonstrated significantly decreased tumor cell proliferation and trend
   toward increased cell death, respectively, at the 3-day timepoint,
   which was not observed at the 1-day timepoint (Fig. [178]2e, f;
   Supplementary Fig. [179]6c, d). Given these results, we administered a
   single LNP treatment to β-M animals and sacrificed mice at 3-days
   post-treatment (Supplementary Fig. [180]7a). While there was no
   significant difference in gross tumor burden (Supplementary
   Fig. [181]7b), a single dose of LNP-CTNNB1 significantly decreased
   liver weight and LW/BW (Supplementary Fig. [182]7c–e), decreased
   intra-tumoral GS expression but retained V5-tag expression
   (Supplementary Fig. [183]7f, g). Also, there were significantly less
   intra-tumoral Ki67-positive cells and significantly more TUNEL-positive
   cells (Supplementary Fig. [184]7h, i). Thus, the earliest evident
   biological response following RNAi-mediated β-catenin inhibition in
   both models occurred at 3-days post-LNP treatment.

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

   [185]Fig. 2
   [186]Open in a new tab

   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) [187]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.

   To understand the transcriptional consequences of β-catenin knockdown
   in HCC, we performed bulk RNA-sequencing (RNA-seq) on both the β-N and
   β-M models treated with either LNP-CTRL or LNP-CTNNB1 at the 3-day
   timepoint. Each model clustered distinctly with LNP-CTNNB1 groups for
   each model clustering independently from the LNP-CTRL groups as shown
   via PCA analysis (Fig. [188]2g). Differential gene expression analysis
   comparing LNP-CTRL vs LNP-CTNNB1 demonstrated 455 upregulated and 628
   downregulated genes in the β-N model, and 608 upregulated and 634
   downregulated genes in the β-M model, with 230 common downregulated and
   73 common upregulated genes (Fig. [189]2h, i). Common downregulated
   genes included Wnt/β-catenin target genes and pericentral hepatocyte
   markers (e.g., Glul, Axin2, Lgr5, Notum, Lect2, Ccnd1, Cyp2e1, Cyp1a2,
   and Oat), and common upregulated genes were midzonal and periportal
   hepatocyte markers (e.g., Hamp2, Cyp8b1, and Cyp2f2) (Fig. [190]2j).
   From both models, gene set enrichment analysis (GSEA) using Kyoto
   Encyclopedia of Genes and Genomes (KEGG) pathways demonstrated relative
   positive enrichment of metabolic and tumor microenvironment pathways,
   along with relative negative enrichment of cell cycle, Wnt signaling,
   and xenobiotic metabolism pathways (Fig. [191]2k, l). Thus, we inferred
   β-catenin mutations in HCC confer most profound effects on tumor cell
   growth/proliferation, tumor metabolism, and tumor microenvironment.

Integrated single-cell analyses reveal de novo formation of reprogrammed
hepatocytes within remnant tumor nodules

   To further interrogate tumor cell-intrinsic biological effects that
   occurred at the 3-day timepoint, we administered LNP-CTRL or LNP-CTNNB1
   at 5-weeks post-HDTVi to β-N model mice and performed single-cell
   RNA-sequencing (scRNA-seq) analysis on a hepatocyte-enriched
   single-cell population following whole liver perfusion. In total,
   94,650 single cells were sequenced with 26,851 in the LNP-CTRL group
   and 67,799 in the LNP-CTNNB1 group. Unbiased clustering on the
   integrated dataset resulted in 10 unique cell populations
   (Supplementary Fig. [192]8a), annotated as (a) Dying/injured
   hepatocytes, (b), Hepatic stellate cells, (c) Kupffer cells, (d)
   Erythroid cells, (e) Endothelial cells, (f) Low-quality hepatocytes,
   (g) Reprogrammed hepatocytes (expressing both zone 1 & 2 markers Arg1,
   Ass1, Pck1, Hal, Hamp2, with Nrf2 tumor targets Prdx2, Prdx5, Gstm1,
   Gpx1), (h) Zone 1 CTNNB1 WT (GS−) hepatocytes, (i) Zone 1/2 CTNNB1 MUT
   (GS+) hepatocytes, and (j) Zone 3 CTNNB1 WT & MUT (GS+) hepatocytes
   based on differential gene expression analysis per cluster
   (Supplementary Fig. [193]8b, c). KEGG pathway enrichment analysis
   comparing each cluster to all other clusters revealed that top pathways
   for Zone 3 CTNNB1 WT & MUT (GS+) hepatocytes were bile acid secretion,
   drug metabolism – cytochrome P450, and fatty acid metabolism, which are
   all known hallmarks of CTNNB1-mutated HCC (Supplementary
   Fig. [194]8d)^[195]35. Zone 1/2 CTNNB1 MUT (GS+) hepatocytes and Zone 1
   CTNNB1 WT (GS−) hepatocytes were interestingly enriched for arginine
   biosynthesis and amino acid biosynthesis (Supplementary Fig. [196]8e,
   f), which are known metabolic hallmarks of zone 1 metabolism^[197]35.
   This KEGG pathway enrichment analysis reveals the metabolic
   heterogeneity of tumor cells along the portal-central axis.

   Cell-type proportion analysis comparing LNP-CTRL and LNP-CTNNB1
   demonstrated less Zone 3 CTNNB1 WT & MUT (GS+) hepatocytes along with
   de novo appearance of reprogrammed hepatocytes following LNP-CTNNB1
   treatment (Fig. [198]3a, b). KEGG and GO pathway enrichment analysis on
   the reprogrammed hepatocytes demonstrated enrichment of pathways across
   all zones, including biosynthesis of cofactors (Zone 1), amino acid
   catabolism (Zone 1), arginine biosynthesis (Zone 1), glutamate
   metabolism (Zone 3), glycolysis/TCA cycle (Zone 3), along with fatty
   acid metabolism, a hallmark of CTNNB1-mutated hepatocellular cancers
   (Supplementary Fig. [199]9a–d). Cell cycle phase-specific gene
   expression analysis on hepatocyte clusters importantly demonstrated
   that tumor cells (both Zone 3 CTNNB1 WT & MUT [GS+] and Zone 1/2 CTNNB1
   MUT [GS+] hepatocytes) were the most proliferative, while reprogrammed
   hepatocytes and Zone 1 CTNNB1 WT (GS−) hepatocytes were the least
   proliferative with proportionally fewer cells in S and G2M phases of
   the cell cycle (Fig. [200]3c). In fact, reprogrammed hepatocytes and
   Zone 1 CTNNB1 WT (GS−) were the two enriched hepatocyte populations
   following LNP-CTNNB1 treatment. Interestingly, Zone 1/2 CTNNB1 MUT
   (GS+) hepatocytes were the most proliferative tumor cell population,
   with the most cells in S and G2M cell cycle phases (Fig. [201]3c). We
   next performed pseudotime analysis on all the hepatocyte populations in
   the dataset which demonstrated the intermediate cell state of the
   reprogrammed hepatocytes occurring along the trajectory of Zone 3
   CTNNB1 WT & MUT (GS+) hepatocytes to Zone 1 CTNNB1 WT (GS−) hepatocytes
   (Fig. [202]3d). Thus, reprogrammed hepatocytes are an intermediate cell
   phenotype, reflecting tumor cell differentiation to normal
   hepatocyte-like cells and contributing to the rapid cell turnover
   observed following LNP-CTNNB1 treatment.

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

   [203]Fig. 3
   [204]Open in a new tab

   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.

   Next, to confirm the spatial identity of the reprogrammed hepatocyte
   population, we performed single-cell spatial transcriptomics using
   Molecular Cartography^TM platform on tissue sections from the 3-day
   timepoint with LNP-CTRL or LNP-CTNNB1 treatment in the β-N model. The
   100-gene panel consisted of markers specific for Wnt/β-catenin targets,
   metabolic zonation, and nonparenchymal cell types (Supplementary
   Table [205]7). Following data pre-processing and automatic cell
   segmentation, in total, 19,301 single cells were sequenced from
   multiple regions of interest (ROIs) with 10,227 cells across 6 ROIs in
   LNP-CTRL group and 9074 cells across 5 ROIs in LNP-CTNNB1 group.
   Unbiased clustering on all 100 genes resulted in 9 unique cell
   populations, annotated as (a) H1: Zone 3 CTNNB1 MUT (GS+), (b) H2: Zone
   3 Central Vein (CV) CTNNB1 WT (GS+), (c) H3: Zone 3 CTNNB1 WT
   (GS-negative), (d) H4: Zone 2–3 CTNNB1 WT (GS-negative), (e) H5: Zone 1
   CTNNB1 WT (GS-negative), (f) H6: Reprogrammed hepatocytes, (g) Hepatic
   stellate cells (HSCs), (h) Immune cells, and (i) Endothelial cells
   (ECs), based on marker gene expression per cluster (Supplementary
   Fig. [206]10a–d). Clustering by treatment condition demonstrated
   enrichment of reprogrammed hepatocytes (46.3% vs 2.7%) with loss of H1:
   Zone 3 CTNNB1 MUT (GS + ) hepatocytes (1.2% vs 29.0%) in LNP-CTNNB1
   group (Fig. [207]3e, f), similar to the scRNA-seq analysis
   (Fig. [208]3a, b). Spatial plots confirmed the tumoral origin of the H6
   cluster representing the reprogrammed hepatocytes (Fig. [209]3g, h). In
   fact, spatial visualization and quantification of Wnt target genes in
   tumoral and non-tumoral regions, including both hepatocyte and
   nonparenchymal cell populatons, revealed that β-catenin-mutated tumor
   cells are defined by expression of bonafide Wnt targets Glul, Tbx3,
   Axin2, Lgr5, Lect2, and Ccnd1 (Supplementary Fig. [210]11a–c) with
   their identity intimately linked to zone 3 metabolic genes (and
   processes), including Cyp2e1, Cyp1a2, and Oat, with exclusion of zone 1
   metabolic genes (and processes), including Cyp2f2, Ass1, and Arg1
   (Supplementary Fig. [211]12a, b). However, with LNP-CTNNB1 treatment,
   tumor cells begin to express Cyp2f2, Arg1, and Ass1 along with
   diminished expression of Cyp2e1, Cyp1a2, Oat and others (Supplementary
   Fig. [212]12a, b). IHC validated the sc-Spatial transcriptomic findings
   and confirmed decreased expression of zone 3 markers CYP2E1 and OAT,
   with increased expression of zone 1 markers ARG1 and CYP2F2
   (Supplementary Fig. [213]12c). Additionally, pseudotime analysis on the
   sc-Spatial transcriptomic data confirmed the intermediary phenotype of
   the H6: reprogrammed hepatocytes (Fig. [214]3i), similar to the
   scRNA-seq data (Fig. [215]3d). Lastly, for validation, cell cluster
   quantification was performed within tumoral and non-tumoral regions
   (using Glul as tumoral landmark) (Supplementary Fig. [216]13a, b),
   which revealed a significant decrease in cell density of clusters with
   active β-catenin signaling, and significant increase in cell density of
   the reprogrammed hepaotcytes, which occurred mostly in tumoral regions
   following LNP-CTNNB1 treatment (Supplementary Fig. [217]13c). Overall,
   this integrated single-cell analysis revealed that β-catenin-mutated
   tumors are exclusively zone 3 and respond to β-catenin suppression by
   turning off expression of these genes while differentiating towards
   zone 1/2 hepatocyte-like cells, thus reprogramming their overall
   metabolic machinery.

Early β-catenin suppression induces an innate immune response characterized
by type I/II interferon network signaling

   Spatial plots also revealed a significant increase in the immune cell
   cluster intratumorally in the LNP-CTNNB1 group compared to the LNP-CTRL
   group (Fig. [218]3g), which was also quantified (Supplementary
   Fig. [219]13c). To further investigate alterations in the immune
   landscape in an unbiased manner at the 3-day time point following
   LNP-CTNNB1 treatment, scRNA-seq was performed on an immune-enriched
   single-cell suspension from β-N treated animals. In total, 20,235
   single cells were sequenced with 8499 cells across 3 individual
   biological replicates in the LNP-CTRL group and 11,736 cells across 3
   individual biological replicates in the LNP-CTNNB1 group. Unbiased
   clustering on the integrated dataset across all cells resulted in 9
   cell populations, which were annotated as: (a) T cells, (b) NK cells,
   (c) Myeloid cells, (d) B cells, (e) Dendritic cells, (f) Stellate
   cells, (g) Endothelial cells, (h) Hepatocytes, and (i) Proliferative T
   cells, based on known marker gene expression for each of these cell
   types (Supplementary Fig. [220]14a, b). Across the 3 biological
   replicates for each treatment condition, the majority cell populations
   that were detected were T cells, B cells, NK cells, and myeloid cells
   (Supplementary Fig. [221]14c–e). We further subclustered and annotated
   these populations to better understand the T cell and myeloid cell
   functional states using marker genes previously described^[222]36
   (Supplementary Fig. [223]15a, b; Fig. [224]4a–c). The major difference
   observed following treatment was a 3-fold enrichment of M1-like
   pro-inflammatory macrophages in the LNP-CTNNB1 group (12.4%) compared
   to LNP-CTRL group (4.1%), which was trending toward significant
   enrichment when averaged across the 3 samples per LNP treatment groups
   (p = 0.0653) (Fig. [225]4b, d). However, at this 3-day time point
   following LNP-CTNNB1 treatment, we did not observe any significant
   differences in CD4+ T cell subpopulations in the β-N model from the
   scRNA-seq analysis (Supplementary Fig. [226]15c), or the sc-spatial
   transcriptomic analysis (Supplementary Fig. [227]15d, e). Additionally,
   in the the β-M model, IHC for CD4 did not reveal differences at the
   3-day timepoint following LNP-CTNNB1 treatment (Supplementary
   Fig. [228]15f). CD8+ T cell subpopulations were not significantly
   altered following LNP-CTNNB1 treatment from this early scRNA-seq
   analysis (Supplementary Fig. [229]15c), although we did detect
   increased CD8a expression in tumoral compartment from the sc-Spatial
   Transcriptomic analysis (Supplementary Fig. [230]15d, e). Thus, innate
   immunity via myeloid cells, appears to be the predominant cell
   population which shifts at 3-days post-treatment (Fig. [231]4d).

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

   [232]Fig. 4
   [233]Open in a new tab

   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 × 10^6 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)
   [234]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.

   To investigate functional changes within the M1-like macrophage
   population, we performed differential gene expression comparing the
   M1-like macrophages from LNP-CTRL and LNP-CTNNB1 treatment. GO pathway
   GSEA demonstrated enrichment of both response to type I/II interferon
   (IFN) and IFN alpha/beta pathways following LNP-CTNNB1 treatment
   (Fig. [235]4e). CellChat analysis, which suggests cell signaling
   pathway level changes based on gene expression of ligands and cognate
   receptors^[236]37, showed enrichment of IFN-II and TNF signaling in the
   M1-like macrophage population following LNP-CTNNB1 treatment
   (Fig. [237]4f). Specifically, this analysis shows increased probability
   of cell communication via Ifng from proliferative T cells to Ifngr1 and
   Ifngr2 on M1-like macrophages, and other macrophage cell populations
   solely in the LNP-CTNNB1 group (Fig. [238]4g). Thus, increased type
   I/II IFNs released from the immune compartment (likely from T cells,
   macrophages, and dying tumor cells) following LNP-CTNNB1 treatment,
   engage with macrophages in the TIME milieu, and in part contribute
   towards polarizing them towards a pro-inflammatory anti-tumor
   phenotype. To validate our findings that a type I/II IFN response is,
   in part, a driver of the anti-tumor immune response following
   LNP-CTNNB1 treatment (Fig. [239]4f, g), we treated β-M mice with IFNγ
   3× weekly for 5 weeks, which led to a significant decrease in tumor
   burden (Fig. [240]4h–j) accompanied by increases in S100A8/9-positive
   cells, a marker for M1-like macrophages, when compared to vehicle
   control group (Fig. [241]4k). Thus, early β-catenin suppression in
   β-catenin-mutated tumors induces local IFN release which is likely
   recruiting and reprogramming intra-tumoral myeloid cells to drive an
   anti-tumor immune response.

Mutated-β-catenin represses a module of transcription factors which drives
immune exclusion in CTNNB1-mutated HCC

   Given the amplified IFN response early after LNP-CTNNB1 treatment, we
   next investigated potential tumor cell-intrinsic molecular mechanisms
   driving this phentoype following β-catenin suppression. Also,
   β-catenin-mutated HCCs are well known for an immune cell excluded
   phenotype^[242]7. Analysis of TCGA-LIHC revealed CTNNB1-mutated HCCs
   downregulate pathways involved in type I/II IFN signaling, T cell
   activation, and chemokine signaling (Supplementary Fig. [243]16a, b).
   To identify potential mechanisms, we utilized bulk RNA-seq datasets
   which contained the transcriptome of multiple β-catenin-mutated HCC
   mouse models ([244]GSE125336) and β-catenin knockout mouse livers
   ([245]GSE68779) and performed transcription factor enrichment analysis
   on the 162 common genes downregulated in β-catenin-mutated HCC and
   upregulated in β-catenin knockout livers. We identified multiple
   transcription factors (TFs), including Irf2 (p = 0.0052) and Pou2f1
   (p = 0.0023), as candidate TFs with known binding to the upregulated
   genes in β-catenin knockout livers (Fig. [246]5a). Interestingly,
   further analysis of Zone 3 CTNNB1 WT & MUT (GS+) hepatocytes from the
   scRNA-seq dataset (Fig. [247]3a) revealed upregulation of Irf2 and
   Pou2f1 target genes, inferred from an unbiased analysis (Fig. [248]5a),
   following LNP-CTNNB1 treatment (Fig. [249]5b). To further confirm
   hepatocytes as the cell source of Irf2 and Pou2f1, which would
   potentially drive an immune response upon β-catenin knockdown or loss,
   we investigated IRF2/Irf2 and POU2F1/Pou2f1 expression in both human
   and mouse liver scRNA-seq datasets^[250]38 ([251]GSE192742). We
   observed IRF2/Irf2 and POU2F1/Pou2f1 expression in hepatocyte cell
   populations in both human and mouse livers (Supplementary
   Fig. [252]16c, d), suggesting β-catenin-mediated IRF2/POU2F1
   suppression is hepatocyte-intrinsic. Likewise, analysis of the
   TCGA-LIHC cohort revealed IRF2 and POU2F1 expression was not
   significantly different between patients with or without Wnt/β-catenin
   activity, rather, the target genes of IRF2/POU2F1 were significantly
   downregulated (p = 0.009) in patients with either CTNNB1, AXIN1, or APC
   mutations compared to other patients (Fig. [253]5c, d). High expression
   of IRF2/POU2F1 target genes was associated with improved disease-free
   survival (p = 0.01) in all TCGA-LIHC patients and in those with
   Wnt/β-catenin activating mutations (p = 0.065) (Supplementary
   Fig. [254]16e, f). Thus, we hypothesized that mutated-β-catenin is
   repressing a module of TFs driving immune exclusion and limiting an
   anti-tumor immune response.

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

   [255]Fig. 5
   [256]Open in a new tab

   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. [257]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)
   [258]https://BioRender.com/25qvbwj. Lehrich (2025)
   [259]https://BioRender.com/qqvvb4x. Lehrich (2025)
   [260]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.

   To validate that repression of IRF2 and POU2F1 are driving immune
   exclusion in β-catenin-mutated HCC, we first overexpressed either pT3
   (empty vector) or Irf2 (β-M-IRF2) in the β-M model (Fig. [261]5e). We
   observed a decrease in overall gross tumor burden and significant
   decrease in LW/BW in β-M-IRF2 mice at 7.5-weeks post-HDTVi
   (Fig. [262]5f–h). RNA-seq confirmed the overexpression of Irf2 in the
   β-M-IRF2 mice at the 7.5-week timepoint where less tumor burden was
   evident (Supplementary Fig. [263]17a). Expectedly, given the known
   immunomodulatory roles of IRF2 and its involvement in type I/II IFN
   signaling^[264]39, we observed increased immune aggregates as evident
   by IHC for CD45 (Supplementary Fig. [265]17b). The composition of the
   immune infiltrate was examined with fluorescence-activated cell sorting
   (FACS) on isolated immune cells from β-M-pT3 and β-M-IRF2 HCC, which
   demonstrated significant increases in total CD4+ T cells with decreases
   in T regulatory cell populations in the β-M-IRF2 mice (Supplementary
   Fig. [266]17c; Supplementary Fig. [267]18a). Next, we overexpressed
   either pT3 (empty vector) or POU2F1 (β-N-POU2F1) in the β-N model
   (Fig. [268]5i). We also observed a significant decrease in overall
   gross tumor burden at 10.7-weeks post-HDTVi in POU2F1-overexpression
   β-N model (Fig. [269]5j–l) and via histology (Supplementary
   Fig. [270]19a). These findings were also validated in the β-M model
   where significant reductions in tumor burden were observed at 7.7-weeks
   post-HDTVi (Supplementary Fig. [271]19b–f). IHC for CD4, CD8, and CD20
   revelaed T and B cells in aggregates in the TIME in the β-N-POU2F1 mice
   (Fig. [272]5m). RNA-seq confirmed the overexpression of POU2F1 in the
   β-M-POU2F1 mice at the 7.7-week timepoint, along with decreased
   enrichment for published mutated-β-catenin gene signature (MBGS)
   (Supplementary Fig. [273]19g–i)^[274]32. Additionally, GSEA with GO
   pathways demonstrated enrichment of both T and B cell activation and
   proliferation (Fig. [275]5n). Lastly, given the less well characterized
   role of POU2F1 mediating an immune response, as compared to
   IRF2^[276]39,[277]40, we administered IgG/αCD3 to deplete CD3+ immune
   cells from β-M-POU2F1 mice (Supplementary Fig. [278]20a), which was
   confirmed via IHC for CD3 on spleens (Supplementary Fig. [279]20b). At
   8.3-weeks post-HDTVi, there was an increase in gross tumor burden and
   significant increase in liver weight, LW/BW, and spleen weight in
   β-M-POU2F1 + αCD3 versus β-M-POU2F1 + IgG animals (Supplementary
   Fig. [280]20c–f), suggesting an immune-dependent role for
   POU2F1-mediated tumor regression in CTNNB1-mutated HCC. Overall,
   mutated-β-catenin represses IRF2, POU2F1, and likely other TFs, which
   limits transcription of key signaling molecules (i.e., cytokines and
   chemokines) important for priming lymphocyte recruitment needed for
   effective anti-tumor immunity.

RNAi-mediated β-catenin inhibition impairs tumor growth in multiple
immunocompetent CTNNB1-mutated HCC mouse models in advanced-stage disease
with lack of response due to expansion of clones derived from secondary
driver

   To further assess translatability, we evaluated the in vivo activity of
   LNP-CTNNB1 in advanced-stage disease in β-M and β-N HCC models. First,
   we assessed response to LNP-CTNNB1 in the β-M model with once weekly
   I.V. treatments starting at 6-weeks post-HDTVi (Fig. [281]6a). We
   observed a heterogeneous response with 5/8 animals responding and 3/8
   animals demonstrating minimal/poor response to LNP-CTNNB1 at 10.5-weeks
   post-HDTVi (Fig. [282]6b, c; Supplementary Fig. [283]21a). Next, we
   assessed response in the β-N model where we administered once weekly
   I.V. LNP treatments starting at 8-weeks post-HDTVi (Fig. [284]6d).
   Similar to the β-M model, after 6 cycles we observed a heterogeneous
   response with 5/8 animals responding and 3/8 animals demonstrating
   minimal/poor response to LNP-CTNNB1 at 13.5-weeks post-HDTVi
   (Fig. [285]6e, f; Supplementary Fig. [286]21b). Thus, LNP-CTNNB1
   appears to be efficacious in advanced-stage disease setting, with a
   subset of animals responding less optimally.

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.

   [287]Fig. 6
   [288]Open in a new tab

   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) [289]https://BioRender.com/63585xs. Lehrich (2025)
   [290]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.

   To investigate the basis of the observed heterogeneous response, we
   first utilized the 10× Visium platform to perform unbiased spatial
   transcriptomics on an LNP-CTRL treated β-M HCC (β-M Control), 2
   LNP-CTNNB1 treated β-M HCC showing minimal/poor response (β-M NR-1; β-M
   NR-2), and an LNP-CTNNB1 treated β-M HCC showing response (β-M R-1). In
   total, we sequenced 17,685 spots across the 4 slides, with 4461 spots
   in β-M Control, 4331 spots in β-M NR-1, 4842 in β-M NR-2, and 4051
   spots in β-M R-1. After integrating the transcriptomes from all slides,
   unbiased clustering revealed 17 clusters conserved across the different
   conditions with changes in cluster proportions and spatial patterns of
   such changes (Supplementary Fig. [291]22a, b). Spatial plots and
   cluster proportion analyses revealed β-M R animals had increases in
   clusters 1, 2, 13, and 14, while β-M NR animals had increases in
   cluster 3 (Fig. [292]6g, h). We also performed 10× Visium spatial
   transcriptomics on an LNP-CTRL treated β-N HCC (β-N Control), 2
   LNP-CTNNB1 treated β-N HCC showing minimal/poor response (β-N NR-1; β-N
   NR-2), and two LNP-CTNNB1 treated β-N HCC showing response (β-N R-1;
   β-N R-2) to validate findings from the β-M model. Here, we sequenced
   17,130 spots across the 5 slides, with 3884 spots in β-N Control, 3624
   spots in β-N NR-1, 4037 spots in β-N NR-2, 3390 spots in β-N R-1, and
   2195 spots in β-N R-2. Integration and unbiased clustering across the
   β-N transcriptomes revealed 13 conserved clusters with changes in
   cluster proportions and spatial patterns of such changes (Fig. [293]6i;
   Supplementary Fig. [294]22c, d). β-N R animals showed increases in
   clusters 3, 4, and 5, while β-N NR animals showed increases in cluster
   11 (Fig. [295]6j).

   To ascertain mechanisms of response to LNP-CTNNB1, we first assessed
   whether there were cell cycle differences between NR and R animals in
   the β-M model. Utilizing gene signatures for S and G2M phases, spots
   belonging to specific clusters were categorized as belonging to G1, S,
   or G2M phases of the cell cycle (Supplementary Fig. [296]23a). There
   was a graded decrease in expression of S and G2M signature genes from
   control to NR to R animals (Supplementary Fig. [297]23b). Spatial
   mapping and quantification of the spots belonging to G1, S, or G2M
   phases revealed majority of spots corresponding to tumoral regions had
   increased expression of S and G2M signature genes in both β-M Control
   and NR animals, while β-M R animals showed majority of cell clusters to
   be in G1 phase (Supplementary Fig. [298]23c–e). Furthermore, RNA in
   situ expression of Mki67 and validation with IHC for Ki67 expressed
   throughout the cell cycle, confirmed active proliferation in tumor
   nodules of β-M NR animals (Supplementary Fig. [299]23f, g). These
   results were also observed in the β-N model (Supplementary
   Fig. [300]24a–g). Thus the NR phenotype in both models demonstrates
   active cellular proliferation likely due to insufficient β-catenin
   knockdown.

   We next assessed whether the residual proliferating tumor nodules in
   the NR mice in both models were Wnt/β-catenin active. We evaluated
   expression of our previously reported mutated-β-catenin gene signature
   (MBGS)^[301]32, which detects Wnt activation in whole and spatial
   transcriptomic datasets with high sensitivity and specificity. MBGS had
   graded decrease in expression from control to NR to R phenotype in both
   models (Supplementary Fig. [302]25a, b) and highest expression in
   clusters 4, 9, and 12 in β-M model and in cluster 7 in β-N model
   (Supplementary Fig. [303]25c, d). As expected, we observed that the
   majority of tumor nodules which were proliferating in the NR phenotype
   were MBGS-high (Supplementary Fig. [304]25e, f). However, a subset of
   tumor nodules in the NR were MBGS-low and proliferating, implying
   active tumor growth despite sufficient β-catenin knockdown. These were
   cluster 3 in the β-M model and cluster 11 in the β-N model
   (Fig. [305]6g, i). Spatial pseudotime analysis demonstrated these
   MBGS-low/Mki67-high nodules were insufficiently reprogrammed from zone
   3 tumor to zone 1/2 hepatocytes (Supplementary Fig. [306]26a–d), and
   transcriptional trajectory indicating tumoral origin. Thus, a subset of
   tumor nodules in the NR phenotype may have persisted due to
   insufficient dosing (i.e., incomplete β-catenin suppression), yet some
   tumor nodules that expanded appear to be biologically distinct.

   To identify biological origin of these persistent tumor nodules, we
   performed differential gene expression analysis across each cluster in
   the β-M (Supplementary Fig. [307]27a) and β-N models (Supplementary
   Fig. [308]28a). Cluster 3 in β-M model revealed upregulation of KEGG
   and GO pathways involved in PI3K-Akt signaling and actin cytoskeleton
   remodeling (Supplementary Fig. [309]29a). We then performed promoter
   enrichment analysis on the top DEGs from cluster 3 and identified MET
   as one of the top kinases predicted to be enriched (Fig. [310]6k). This
   was validated with V5-tag IHC where we observed persistence of V5-tag+
   clones in NR phenotype (Supplementary Fig. [311]29b). Additionally,
   cluster 11 in the β-N model revealed upregulation of KEGG and GO
   glutathione metabolism and reactive oxygen species metabolic pathways
   (Supplementary Fig. [312]30a). Promoter enrichment analysis on the top
   DEGs from cluster 11 identified NFE2L2 as one of the top transcription
   factors predicted to be enriched (Fig. [313]6l). We also validated this
   finding with NQO1 IHC where we observed persistence of NQO1+ clones in
   NR phenotype (Supplementary Fig. [314]30b). Thus, MBGS-low nodules,
   which expanded in the NR phenotype, were likely derived from clonal
   expansion of tumor cells which were predominant upregulators of
   pathways specific to the secondary driver as a means to escape
   β-catenin suppression. Overall, heterogeneity in advanced-stage disease
   response appears to be due in part to insufficient β-catenin
   suppression and in part due to an escape mechanism.

Response to RNAi-mediated β-catenin inhibition in advanced-stage disease is
associated with restored adaptive immunity

   Given the known role of β-catenin in promoting a non-T cell-inflamed
   TIME^[315]7, we examined whether LNP-CTNNB1 sufficiently reprogrammed
   the adaptive immune response in advanced-stage HCC. First, we observed
   increased expression of T cell markers, including Cd3d, Cd3e, Cd3g,
   Cd4, and Cd8a in cluster 14 in the β-M model, (Supplementary
   Fig. [316]31a), which had expanded in the R phenotype. Additionally,
   cluster 14 in the β-M model showed increased expression of a T cell
   module score, which was also enriched in β-M R animals (Supplementary
   Fig. [317]31b, c). Interestingly, we also observed enrichment of these
   T cell markers in MBGS-high tumor-specific clusters (clusters 9 and 12
   in β-M model) within R animals, demonstrating active T cell
   infiltration within residual tumor nodules, which was likely driving
   anti-tumor immunity (Fig. [318]7a; Supplementary Fig. [319]31d). A
   similar amplified T cell response was observed in the β-N model with
   increased T cell marker expression within the MBGS-high cluster 7 in R
   animals (Supplementary Fig. [320]32a–d), although not as pronounced as
   the β-M model. In the β-M model, GSEA on clusters 9 and 12 comparing R
   to control demonstrated enrichment of GO pathways involved in
   leukocyte-mediated immunity, response to IFNγ, leukocyte cell adhesion,
   cell killing, and T cell proliferation (Supplementary Fig. [321]33a,
   b). Increased IFNγ expression was observed at the RNA and protein level
   following LNP-CTNNB1 treatment (Supplementary Fig. [322]33c, d). We
   validated the increased T cell infiltrate via IHC in the β-M model
   demonstrating increased CD3+ and CD4+ cells within tumor nodules with
   organization into lymphoid aggregates (LA) in the β-M R animals
   (Fig. [323]7b, c). We observed the LAs to be also composed of B cells,
   as noted by increased B cell markers gene expression in clusters 13 and
   14, the MBGS-high tumor-specific clusters (clusters 9 and 12),
   specifically enriched in β-M R animals, along with increased CD20+
   cells via IHC (Supplementary Fig. [324]34a–e). To further characterize
   the enhanced adaptive immune response in β-M R animals, we performed
   spatially enhanced CellChat^[325]37 analysis to investigate
   ligand-receptor interactions between different clusters across the
   responders. This analysis revealed enrichment of MHC-II signaling with
   antigen communication from most clusters to CD4+ cells in cluster 12
   specifically in β-M R animals compared to both β-M Control and β-M NR
   animals (Supplementary Fig. [326]35a–d). Morevoer, an IRF2/POU2F1
   target gene module score showed increased expression in cluster 12 of
   β-M R compared to NR and control animals (Supplementary Fig. [327]35e),
   further suggesting an IRF2-mediated IFN response upon β-catenin
   suppression. Overall, the R phenotype demonstrated reinvigorated and
   persistent adaptive immune surveillance with active T and B cell
   infiltration, T cell proliferation, and engaged IFN signaling in the
   intra-tumoral compartment, which was not observed in the NR phenotype
   in the advanced-stage disease setting.

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

   [328]Fig. 7
   [329]Open in a new tab

   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. [330]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)
   [331]https://BioRender.com/z4h1j67. Lehrich (2025)
   [332]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.

Immunotherapy enhances response to RNAi-mediated β-catenin inhibition in
advanced-stage disease setting in CTNNB1-mutated HCC mouse model

   Due to the known role of β-catenin promoting and IRF2 inhibiting
   transcription of Cd274 (encoding PD-L1 on tumor cells)^[333]41–[334]43,
   a classical immune checkpoint mediating T cell exhaustion, we
   hypothesized that the NR phenotype may be driven by T cell exhaustion
   through increased expression of Cd274. Indeed, we observed increased
   Cd274 and Pdcd1 (encoding PD1 on T cells) expression in β-M NR and
   control compared to R animals (Fig. [335]7d; Supplementary
   Fig. [336]31e), implying lack of sustained and active lymphocyte
   effector function with increased exhaustion in NR, likely due to
   insufficient β-catenin suppression. This led us to hypothesize that
   administration of ICI would enhance LNP-CTNNB1 response in
   advanced-stage disease through limiting T cell exhaustion via PD1/PD-L1
   axis, thus promoting sustained anti-tumor immunity. IgG or α-PD1 was
   added 3-days after the first LNP dose with continued LNP treatment
   scheme as for the advanced-stage disease in the more aggressive β-M
   model. We harvested mice by 10.5-week timepoint or when moribund to
   assess response and performed a survival study to determine long-term
   anti-tumor immunity (Fig. [337]7e). By 10.5-weeks, LNP-CTRL mice were
   all moribund with α-PD1 alone suggesting no impact on tumor burden,
   whereas the combination of LNP-CTNNB1 + α-PD1 resulted in enhanced
   efficacy with absence of any NR compared to LNP-CTNNB1 + IgG treated
   animals (Fig. [338]7f, g). Additionally, MRI demonstrated less
   hyperintense foci in LNP-CTNNB1 + α-PD1 compared to LNP-CTNNB1 + IgG
   treated mice (Fig. [339]7h). Decrease in hCTNNB1, indicative of tumor
   burden driven by mutant-CTNNB1, was enhanced in the LNP-CTNNB1 + α-PD1
   compared to LNP-CTNNB1 + IgG treated mice (p = 0.0395) suggesting an
   augmented response to LNP-CTNNB1 with α-PD1 (Fig. [340]7i). To
   determine whether there was enhanced effector T cell function in
   animals treated with LNP-CTNNB1 + α-PD1, we performed IHC for granzyme
   B (GZMB), a marker for cytotoxic T cells, which is known to be
   expressed at higher levels in tertiary lymphoid structures
   (TLS)/LA^[341]44. We observed an increased trend in the quantity of
   GZMB+LA within remnant tumor nodules in LNP-CTNNB1 + α-PD1 compared to
   LNP-CTNNB1 + IgG treated mice (p = 0.0989) (Fig. [342]7j, k).
   Concomitantly, mice receiving LNP-CTNNB1 + α-PD1 survived significantly
   longer than those receiving LNP-CTNNB1 + IgG (p = 0.019)
   (Fig. [343]7l), supporting an enhanced response to LNP-CTNNB1 with
   α-PD1. Overall, administration of α-PD1 augmented LNP-CTNNB1 response
   and restored adaptive immunity.

TLS/LA are correlated with atezolizumab plus bevacizumab responders and
inversely correlated with CTNNB1 mutational status in the IMbrave150 trial

   Given the enhanced recruitment of both T and B cells into organized LA
   following LNP-CTNNB1 treatment in mice, we were interested in examining
   such a relationship between TLS/LA, CTNNB1 mutation, and ICI response
   in patients. In the IMbrave150 phase III clinical trial, unresectable
   HCC patients received atezolizumab and bevacizumab in one arm versus
   sorafenib in another^[344]6. Atezolizumab and bevacizumab arm showed
   improved response rates, overall survival (OS) and progression-free
   survival (PFS). Accessing this information, 174 HCC patients were
   scored by a clinical pathologist for the presence of immune
   infiltration ranging from TLS, LA, diffuse infiltrate [DI], or none
   from the hematoxylin & eosin (H&E) slides. Overall, majority of the
   patients, irrespective of treatment arm, had LA (n = 71/174), while
   fewer had TLS (n = 8/174) or DI (n = 8/174) (Fig. [345]8a).
   Interestingly, among the responders, those in the atezolizumab plus
   bevacizumab arm tended to be enriched for presence of TLS/LA, which was
   not observed in the sorafenib arm (Fig. [346]8b). Additionally,
   patients with TLS/LA correlated with improved PFS and OS, which was
   more pronounced in the atezolizumab plus bevacizumab arm
   (Fig. [347]8c). Moreover, patients with TLS/LA had significantly
   increased expression of a LA-like gene signature, which consisted of
   previously reported marker genes for B cells, CD4 T[conv] cells, and
   CD4 T[fh] cells^[348]45 compared to the patients with DI/None
   (Fig. [349]8d, e). Interestingly, high expression of the LA-like gene
   signature was associated with improved PFS (p = 0.0639) and OS
   (p = 0.000697) in the atezolizumab plus bevacizumab arm (Fig. [350]8f).
   Importantly, we observed that CTNNB1-mutated patients had significantly
   lower expression of the LA-like gene signature compared to
   CTNNB1-wild-type patients (Fig. [351]8g). This was also confirmed in
   the TCGA-LIHC cohort (Supplementary Fig. [352]36a–c). Thus,
   CTNNB1-mutated patients express fewer genes important for TLS/LA
   formation compared to CTNNB1-wild-type patients, ultimately influencing
   response to combination ICIs.

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

   [353]Fig. 8
   [354]Open in a new tab

   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).

Discussion

   We report strong in vitro and vivo efficacy of a novel LNP-formulated
   siRNA targeting CTNNB1 mRNA transcript for treatment of
   β-catenin-mutated HCC as monotherapy in early-stage disease or in
   combination with ICI in advanced-stage disease. We identified through
   unbiased bulk RNA-seq, scRNA-seq, and spatial transcriptomic approaches
   tumor-cell-intrinsic roles of β-catenin-mediated IRF2 and POU2F1
   repression driving an immune-excluded TIME and inert type I/II IFN
   responses in β-catenin-mutated HCC with in vivo validation.
   Additionally, we demonstrate upon β-catenin suppression,
   β-catenin-mutated tumor cells reprogram towards zone 1/2
   hepatocyte-like cells, revealing the unique role of mutated-β-catenin
   in driving zone 3 (pericentral) tumor metabolism. Our work demonstrates
   that β-catenin is now directly targetable in murine HCC and supports
   the high impact development of clinical investigations utilizing
   LNP-CTNNB1 as monotherapy or in combination with ICI to achieve
   therapeutic benefit in HCC patients with Wnt/β-catenin activation.

   β-Catenin is most active in the pericentral (zone 3) region in the
   hepatic lobule with hepatocytes in each of the three zones of the
   hepatic lobule expressing genes important for different metabolic
   functions, termed liver metabolic zonation^[355]35. Given the
   localization of β-catenin to zone 3, it is no surprise that
   β-catenin-mutated tumors preferentially originate and clonally expand
   from hepatocytes residing within zone 3, and these tumors share unique
   metabolic addictions to processes canonically identified in zone 3. In
   fact, we have previously shown that CTNNB1-mutated HCC is addicted to
   glutamine synthesis^[356]46, as part of β-catenin-GS-mTOR axis^[357]21.
   Additionally, CTNNB1-mutated HCCs demonstrate addiction to xenobiotic
   metabolism through GSTM3^[358]47. However, surprisingly, tumors with
   β-catenin oncogenic activation are not glycolytic (zone 3 metabolism),
   but are fatty acid oxidative (zone 1 metabolism) addicted^[359]48.
   Here, we show that β-catenin-mutated tumors residing specifically in
   zone 3 are metabolically wired to perform canonical zone 3 metabolic
   processes with a focus on fatty acids as substrates, while
   β-catenin-mutated tumor cells in zone 1 are metabolically wired to
   perform canonical zone 1 metabolic processes with a focus on arginine
   metabolism and amino acid biosynthesis. We have also uniquely
   demonstrated that β-catenin-mutated tumor cells in zone 1 possess the
   highest proliferative capacity compared to those in zone 3, suggesting
   that despite β-catenin-mutated HCCs being well-differentiated, less
   proliferative tumors, in ectopic regions of absent Wnt signals or in
   presence of normal zone 1 signals, proliferation may be favored over
   metabolic homeostasis. Whether zone 1 β-catenin-mutated HCCs in current
   model are due to clonal expansion, evolution, or budding from zone 3
   tumors to eventually establish in zone 1, or an artifact of plasmid
   transfection in rare hepatocytes in zone 1 requires further
   investigation. However, despite these tumor cell-intrinsic pathways,
   the overall tumor biology and metabolism may also be regulated by local
   zonal environment and signals. Overall, we demonstrate that suppressing
   β-catenin in CTNNB1-mutated tumors reprograms zone 3 tumors towards a
   zone 1/2 metabolic phenotype as early as 3-days post-LNP treatment,
   which contributes to the phenotypic differentiation and metabolic
   rewiring, loss of tumor nodules, and normalization of hepatic
   parenchyma and liver mass. Such reprogramming may yield unique
   metabolic vulnerabilities to be exploited for additional therapies in
   the future.

   Cancers with Wnt/β-catenin activation are considered non-T
   cell-inflamed across a variety of tumors, including HCC, melanoma,
   esophageal, and others^[360]6,[361]7,[362]49,[363]50. This has been
   associated with mixed ICI responses, specifically of the
   anti-PD1/anti-PD-L1 agent class^[364]49. In HCC, many chemokines are
   lowly expressed in CTNNB1-mutated patients, which directly influence
   immune cell recruitment, infiltration, and function^[365]7. For
   example, upon re-expression of CCL5, survival of β-catenin-mutated
   tumors was prolonged, along with increased antigen-specific CD8+ T
   cells^[366]7. Analogously, in KRAS-mutated colorectal cancer, where ICI
   is ineffective, expression of key chemokines involved in IFN network
   signaling, such as CXCL3, are downregulated when KRAS is mutated due to
   direct interaction with IRF2^[367]39. In the context of
   β-catenin-mutated tumors, previous work has described that β-catenin
   sequesters and inhibits p65 (a subunit of NF-κB) leading to reduced p65
   binding, target gene expression, and inflammation^[368]51–[369]53.
   Here, our unbiased bioinformatic analysis identified both IRF2 and
   POU2F1 as candidate TFs whose target genes are downregulated when
   β-catenin is active. We demonstrated that β-catenin suppression
   directly increases IFN signaling molecules and antigen presentation
   machinery components in β-catenin-mutated HCC murine models, with
   forced expression of either IRF2 or POU2F1 in two β-catenin-mutated HCC
   models sufficient to delay tumor development and convert a non-T
   cell-inflamed to a T cell-inflamed tumor. Thus, in β-catenin-mutated
   HCC, we posit that mutated-β-catenin is directly interacting with an
   immune-regulatory module of key TFs, which are normally active when
   β-catenin is absent from the nucleus, influencing expression of
   essential factors necessary for immune cell recruitment and function.
   Additionally, β-catenin has been shown to promote expression of PD-L1
   on tumor cells by regulating expression of
   Cd274^[370]41,[371]42,[372]54. On the contrary, IRF2 has been shown to
   inhibit transcription of Cd274 on tumor cells^[373]43. Adequate
   β-catenin loss thus prevents expression of this classical immune
   checkpoint mediating T cell exhaustion dually, to thus synergize with
   ICIs and promote anti-HCC effect which can have wider therapeutic
   implications. Overall, pharmacological targeting of β-catenin likely
   has clinical implications across a broad spectrum of tumor types to
   enhance ICI clinical efficacy in part through modulation of key TFs
   involved in priming immune cell recruitment and re-engaging global
   immune surveillance.

   We have shown here that targeting β-catenin directly impacts both tumor
   cell-intrinsic biology and simultaneously reprograms the TIME from
   non-T cell-inflamed to T cell-inflamed, with innate immune remodeling
   occurring as early as 3-days post-LNP treatment. This innate immune
   remodeling coincided with first observed biological effect of β-catenin
   knockdown at 3-days. Biological effects due to siRNA knockdown are
   usually observed within hours in vitro^[374]55, yet we observed a
   protracted time course in vivo, likely due to the systemic delivery
   method. Additionally, prior work has illustrated that adaptive immune
   surveillance begins to remodel at least 7–10 days following oncogene
   withdrawal, which explains the significant adaptive immune effects we
   observed studying advanced-stage response after 6 weeks of LNP
   treatment^[375]56. However, the profound anti-tumor effects we observed
   here likely would not be so pronounced through targeting downstream
   effector molecules of the Wnt/β-catenin signaling pathway.
   Specifically, we and others have previously shown that genetic deletion
   or pharmacologic inhibition of downstream effectors of
   β-catenin-TCF/LEF interactions, such as cyclin D1 (encoded by
   CCND1)^[376]57, GS^[377]16, mTORC1^[378]21, TBX3^[379]17,
   AXIN2^[380]58, or TNFRSF19^[381]59 either result in partial tumor
   responses or compensatory negative feedback loops leading to enhanced
   tumorigenesis. For example, it has been shown that hepatocarcinogenesis
   is not dependent on cyclin D1 as β-catenin-mutated tumors induced in
   Ccnd1-null background mice still develop through compensatory cyclin D2
   expression^[382]57. Additionally, conditionally deleting TBX3 or GS in
   mice with β-catenin-mutated HCC exacerbates tumorigenesis through
   YAP/TAZ inhibition or nitrogen metabolic rewiring,
   respectively^[383]16,[384]17. Moreover, our group has previously
   identified metabolic addiction to β-catenin-GS-mTOR axis in
   β-catenin-mutated HCC and evaluated mTOR inhibitor (e.g., rapamycin,
   everolimus) response in multiple preclinical models of
   β-catenin-mutated HCC. However, response to LNP-CTNNB1 results in more
   consistent, robust, and durable responses in preclinical
   models^[385]21,[386]28. Lastly, targeting solely TNFRSF19 will likely
   impact expression of chemokines involved in immune cell recruitment,
   yet there would be minimal impact on tumor cell-intrinsic
   biology^[387]59. Thus, targeting β-catenin directly is a holistic and
   rational strategy leading to durable anti-tumor immune responses
   through inhibiting multiple mechanisms hitting a truncal event, and
   impacting not only tumor cell-intrinsic biology, but also
   simultaneously remodeling the TIME architecture to promote long-lasting
   anti-tumor immunity.

   Therapeutic targeting of Wnt/β-catenin oncogenic signaling has been
   pursued over the last two decades with no therapeutic agent ultimately
   resulting in translation to the clinic. First, given the ubiquitous
   role of β-catenin in many cell types, translation of many agents has
   been limited due to on-target, off-tumor effects^[388]49,[389]60.
   Small-molecule inhibitors which limit interactions between β-catenin
   and TCF/LEF or β-catenin and cAMP response element–binding protein
   (CREB)–binding protein (CBP), or repurposed drugs against Wnt activity
   have shown in vitro inhibitory effects, yet lack strong in vivo
   efficacy, likely due to alternative escape mechanisms^[390]9.
   Alternative methods of Wnt/β-catenin inactivation have investigated
   porcupine (PORCN), tankyrase (TNKS), or Frizzled (FZD) receptor
   inhibitors, however, these are ineffective and far too upstream in the
   pathway for treating tumors with GOF CTNNB1 mutations due to subsequent
   independence of Wnt/FZD receptor binding^[391]9. Thus, RNAi- or
   antisense-mediated gene silencing approaches have proven to be an
   effective therapeutic approach to reduce CTNNB1 mRNA levels in tumors.
   Efficacy has previously been shown by our group and others across a
   variety of different tumor types^[392]20,[393]22,[394]23,[395]61. Our
   work here builds upon these previous findings and demonstrates that
   RNAi-mediated β-catenin inhibition via LNP for HCC results in minimal
   off-target effects with strong and durable on-target effects.

   In summary, we propose a synergistic two-part working mechanism of
   response to RNAi-mediated β-catenin inhibition in preclinical
   CTNNB1-mutated HCC models (Fig. [396]7m). First, response to LNP-CTNNB1
   treatment includes cessation of tumor cell proliferation and
   concomitant metabolic zonal reprogramming with zone 3 tumor cells
   converting to zone 1/2 hepatocyte-like cells. Second, there is
   remodeling of the TIME with macrophages repolarizing towards a
   M1-like-phenotype, along with restored adaptive immune surveillance.
   This is driven by IRF2/POU2F1 re-engagement upon β-catenin knockdown,
   which act as mediators of enhanced IFN network signaling, suppression
   of PD-L1 expression, and priming of lymphocyte (T and B cell)
   recruitment and infiltration. All these tumor cell-intrinsic and TIME
   remodeling mechanisms ultimately drive enhanced efficacy of LNP-CTNNB1
   to α-PD1 in advanced-stage disease. Based on our findings,
   RNAi-mediated inhibition of β-catenin may have the potential to provide
   anti-tumor effects as a monotherapy in early-stage disease or in
   neoadjuvant setting in patients with Wnt-β-catenin active liver tumors.
   These proof-of-concept studies also support the clinical investigation
   of RNAi therapeutic approaches targeting β-catenin in combination with
   ICI in advanced-stage Wnt-β-catenin active-HCC patients.

Methods

Mice for tumor study

   Approval for all animal husbandry and experimental procedures,
   including animal housing and diet, complies with all relevant ethical
   regulations, and performed under the guidelines and approval of the
   National Institutes of Health and the University of Pittsburgh School
   of Medicine Institutional Animal Use and Care Committee (IACUC). All
   FVB/NJ male mice were purchased from the Jackson Laboratory (Bar
   Harbor, ME). Ctnnb1^f/f (on a C57Bl6 background) male and female mice
   were used for this study as well and purchased form the Jackson
   Laboratory (Bar Harbor, ME). For β-catenin-mutated HCC studies, we used
   6–8-week-old FVB/NJ male mice. Male mice were utilized given the
   increased prevalence of liver cancer in males. Mice were fed a standard
   chow diet ad libitum, with access to water, enrichment, and exposed to
   12 h light/dark cycles in ventilated cages. Mice were housed to
   individual cages with each having 3–4 mice. Body weights were monitored
   once weekly. As tumors developed in mice over time, they were monitored
   for signs of morbidity, including increased abdominal distension/girth,
   grimace, and body weights not to exceed 20–25% of aged-matched
   controls, and were appropriately euthanized and considered moribund.
   Mice were fasted 4–6 h prior to sacrifice. At time of sacrifice, gross
   images of the liver were recorded with a handheld camera, along with
   documenting the body and liver weights, and gross morphology of the
   mouse livers with a ruler measurement. At time of sacrifice, whole
   blood was drawn from inferior vena cava via insulin syringe and
   expunged into microcentrifuge tubes to be spun at 2500 × g for
   10–15 min at room temperature to collect serum. Serum was then
   collected, processed, and sent to be analyzed for liver function tests,
   including aspartate aminotransferase (AST), alanine aminotransferase
   (ALT), and alkaline phosphatase via UPMC Clinical Pathology
   Laboratories.

Plasmids

   For mutant-β-catenin constructs, we utilized the S45Y-hCTNNB1-Myc-tag
   and T41A-hCTNNB1-Myc-tag plasmids, which are human constructs of the
   gene^[397]27,[398]28. Additionally, we utilized the hMet-V5-tag and
   G31A-hNFE2L2 plasmids, which are also human constructs of the
   genes^[399]27,[400]28. Both the T41A-hCTNNB1-Myc-tag and G31A-hNFE2L2
   plasmids were subcloned into the pT3-EF1αh destination vector, while
   the S45Y-hCTNNB1-Myc-tag and hMet-V5-tag plasmids were subcloned into
   the pT3-EF5αh destination vector^[401]27,[402]28. We also utilized the
   myr-AKT and N-RasV12 plasmids known to induce Akt/NRas HCC^[403]62. We
   purchased the mouse pFUW-tetO-Irf2 plasmid from Addgene (plasmid
   #139814) and pENTR223-POU2F1 plasmid from DNASU (plasmid #
   HsCD00505520). We then subcloned these plasmids into the pT3-EF1αh
   plasmid using Gateway PCR cloning technology (Invitrogen, Carlsbad, CA)
   to make a pT3-EF1αh-Irf2 and pT3-EF1αh-POU2F1 plasmids. The pCMV-pT3,
   pCMV-SB transposase, and pCMV-Cre plasmids have been described
   previously^[404]27. The plasmid constructs were purified using
   Endotoxin-Free Maxiprep kit (NA 0410, Sigma-Aldrich, St. Louis, MO) for
   eventual hydrodynamic delivery. For hydrodynamic delivery, plasmids
   were diluted into 0.9% normal saline (NaCl) purchased from TEKNOVA
   (#S5815).

HCC animal models developed via sleeping beauty-hydrodynamic tail vein
injection

   We utilized the sleeping beauty-hydrodynamic tail vein injection
   (SB-HDTVi) model, which has been extensively described by our
   group^[405]20,[406]27,[407]28. We utilize 5 HCC mouse models in this
   study: (1) T41A-hCTNNB1/G31A-hNFE2L2 (β-N), (2) S45Y-hCTNNB1/hMet
   (β-M), (3) S45Y-hCTNNB1/G31A-hNFE2L2/hMet (β-N-M), (4)
   G31A-hNFE2L2/hMet (N-M), and (5) Akt/NRas models. For the β-N model,
   20 μg of pT3-EF1αh-T41A-hCTNNB1-Myc-tag and 20 μg of NFE2L2-plasmid
   (pT3-EF1αh-G31A-hNFE2L2) were mixed. For the β-M model, 10 μg of
   pT3-EF5αh-S45Y-hCTNNB1-Myc-tag and 10 μg of hMET-plasmid
   (pT3-EF5αh-hMet-V5-tag) were mixed. For the β-N-M model, 10 μg of
   pT3-EF5αh-S45Y-hCTNNB1-Myc-tag, 10 μg of NFE2L2-plasmid
   (pT3-EF1αh-G31A-hNFE2L2), and 10 μg of hMET-plasmid
   (pT3-EF5αh-hMet-V5-tag) were mixed. For the N-M model, 20 μg of
   NFE2L2-plasmid (pT3-EF1αh-G31A-hNFE2L2) and 20 μg of hMET-plasmid
   (pT3-EF5αh-hMet-V5-tag) were mixed. For the Akt/NRas model, 10 μg
   pT3-EF1α-HA-myr-AKT and 10 μg of pT2-Caggs-N-RasV12 were mixed with
   pCMV-pT3 (60 μg; empty vector; solely the backbone plasmid) or pCMV-Cre
   plasmids (60 μg) to delete endogenous genes in Ctnnb1 floxed mice
   background. For β-M-pT3 and β-M-IRF2 models, we mixed 20 μg of
   pT3-EF5αh-S45Y-hCTNNB1-Myc-tag, 20 μg of hMET-plasmid
   (pT3-EF5αh-hMet-V5-tag), and either 20 μg of pCMV-pT3 or Irf2 plasmid
   (pT3-EF1αh-Irf2). For the β-M-pT3 and β-M-POU2F1 models, we mixed 20 μg
   of pT3-EF5αh-S45Y-hCTNNB1-Myc-tag, 20 μg of hMET-plasmid
   (pT3-EF5αh-hMet-V5-tag), and either 60 μg of pCMV-pT3 or POU2F1 plasmid
   (pT3-EF1αh-POU2F1). For the β-N-pT3 and β-N-POU2F1 models, we mixed
   20 μg of pT3-EF1αh-T41A-hCTNNB1-Myc-tag, 20 μg of Nrf2-plasmid
   (pT3-EF1αh-G31A-hNFE2L2), and either 60 μg of pCMV-pT3 or POU2F1
   plasmid (pT3-EF1αh-POU2F1). pCMV-Sleeping Beauty transposase (pCMV/SB)
   plasmid is mixed with each of the oncogenic plasmids at a concentration
   of 25:1 in 2 mL normal saline (0.9% NaCl) and filtered through 0.22 μm
   filter (Millipore) for hydrodynamic injection. For hydrodynamic
   injection, 2 mL volume is injected in the lateral tail vein in 5–7 s.

LNP-siRNA synthesis

   Chemically modified and functionalized siRNAs for this study were
   synthesized and characterized by Alnylam
   Pharmaceuticals^[408]63,[409]64. The lipid nanoparticles (LNPs)
   containing encapsulated siRNAs were formulated and characterized using
   validated techniques^[410]65,[411]66 with the biodegradable ionizable
   lipid bis(3-pentyloctyl)
   9-((4-(dimethylamino)butanoyl)oxy)heptadecanedioate. The LNPs used in
   this study size range from 60 to 80 nm with siRNA encapsulation
   efficiency of ~95% (PDI = 0.045) with quality control performed by
   Alnylam Pharmaceuticals^[412]31. LNP formulations from Alnylam
   Pharmaceuticals have demonstrated a rapid accumulation of the parent
   lipids in the liver and plasma of mice with elimination from plasma by
   8-h, liver by 48-h, and spleen by 96-h post systemic administration via
   tail vein injection^[413]31. Biodistribution of the LNPs in mice have
   also been reported to accumulate primarily in the liver and
   spleen^[414]31. A control siRNA was synthesized which is not predicted
   to bind to any location in the transcriptome and the lack of activity
   was confirmed by RNA-sequencing (RNA-seq) analysis performed by Alnylam
   Pharmaceuticals as part of quality control. The siRNA targeting CTNNB1
   has demonstrated ability to reduce levels of both mouse and human
   CTNNB1 transcripts. In this study we refer to the LNP encapsulating
   control siRNA as LNP-CTRL and LNP encapsulating siRNA targeting CTNNB1
   transcript as LNP-CTNNB1.

siRNA sequences

   siCTNNB1: sense – UACUGUUGGAUUGAUUCGAAA,

   antisense - UUUCGAAUCAAUCCAACAGUAGC

   siControl: sense ACCCAGAUAUUUUUAUCGCGU,

   antisense – ACGCGAUAAAAAUAUCUGGGUCG

Patient-derived organoid cultures

   CTNNB1-mutated patient-derived HCC organoid 23277, which has a S33F
   mutation in exon 3 of CTNNB1, was kindly provided by Ernesto
   Guccione^[415]26. These organoids were cultured and expanded in 24-well
   plates with tumor organoid medium^[416]26. Organoids were treated with
   LNP-CTRL or LNP-CTNNB1 at 20 nM concentration and imaged at 48-h and
   72-h post-LNP treatment. Experiments were performed in triplicate.

Various therapeutic interventions and dosing

   The therapeutic formulations of LNP-CTRL and LNP-CTNNB1 were diluted
   from their stock formulations into phosphate buffered saline (PBS) to
   reach final concentrations of either 3, 1, 0.3, 0.1, and 0.03 mg/kg
   dosages. The PBS/LNP-CTRL/LNP-CTNNB1 formulations were administered
   once weekly at the indicated time points below via lateral tail vein
   (intravenous; I.V.) injection using insulin syringe as the delivery
   method. For early-stage intervention in the β-N model, dosing started
   at 5-weeks post-HDTVi. For early-stage intervention in the β-M model,
   dosing started at 3-weeks post-HDTVi. For early-stage intervention in
   the β-N-M model, dosing started at 3-weeks post-HDTVi. For early-stage
   intervention in the N-M model, dosing started at 8-weeks post-HDTVi.
   For late-stage intervention in the β-N model, dosing started at 8-weeks
   post-HDTVi. For late-stage intervention in the β-M model, dosing
   started at 6-weeks post-HDTVi. For combination therapy study of
   LNP-CTNNB1 with α-PD1 in β-M model at late-stage, mice were randomized
   to receive either 200 μg of either IgG control antibody (BioXcell;
   InVivoMAb rat IgG2a isotype control; Catalog #BE0089) or α-PD1 antibody
   (BioXcell; InVivoMAb anti-mouse PD1 (CD279); Catalog #BE0146) starting
   3-days after the 6-weeks post-HDTVi timepoint when given
   LNP-CTRL/LNP-CTNNB1 treatments. The animals received IgG/α-PD1 2×/week
   over 2-weeks. The IgG/α-PD-1treatments were given via intra-peritoneal
   (I.P.) injection. For IFNγ (Miltenyi; #130-105-774) treatments, mice
   were treated at the indicated time points at a concentration of 1 × 104
   IU starting 3-weeks post-HDTVi. For α-CD3 immune depletion studies,
   β-M-pT3 or β-M-POU2F1 mice were treated with either IgG isotype control
   (BioXcell; InVivoMAb rat IgG2a isotype control; Catalog #BE0089) or
   InVivoMAb α-mouse CD3 (Catalog #BE0002, clone 17A2) at 200 μg
   concentration at the indicated time points.

Sample collection and processing

   At the time of animal sacrifice, whole blood was collected via either
   drainage from IVC or retroorbital blood draw. Serum was processed
   through centrifugation at 3000 × g for 10 min at cold 4 °C temperature
   (supernatant contained the serum). Liver tissue from all samples were
   either fixed in 10% neutralized formalin (Fisher Chemicals) at room
   temperature for 48–72 h, followed by transfer into 70% ethanol,
   dehydration, and paraffin embedded for histology analysis, or flash
   frozen in liquid nitrogen and stored long-term at −80 °C.

Hematoxylin and Eosin (H&E) staining

   Formalin, fixed paraffin embedded (FFPE) liver tissue was sectioned at
   4 μm thick slices using a standard microtome. Liver tissue sections
   were deparaffinized in xylene and graded ethanol followed by
   Hematoxylin (Fisher Chemical Harris Modified Method Hematoxylin Stains
   #SH26-500D) and Eosin (Eosin Y, #23-314-63) and dehydrated, mounted,
   and cover-slipped. Slides were imaged on Zeiss Axioskop microscope.

Immunohistochemistry (IHC)

Single-stain for IHC

   FFPE sections were deparaffinized in xylene and graded ethanol (10%,
   95%, 90%) and rinsed with PBS or water. For antigen retrieval, samples
   were placed in either Citrate Buffer (0.01 M, pH 6.0), Tris-EDTA
   (0.01 M, pH 9.0) or DAKO (Agilent Technologies; S236784-2) and heated
   in a steamer or a pressure cooker for 20 min. Next, samples were placed
   in 3% H[2]O[2] solution (Fisher Chemicals) for 10 min to quench
   endogenous peroxide activity. After appropriates washing in saline,
   slides were blocked with Super Block (ScyTek Laboratories) for 10 min
   to prevent non-specific antibody binding. Next, the sections were
   incubated with the primary antibodies diluted in PBS at 4 °C overnight
   or at room temperature. A list of primary antibodies and the specific
   antigen retrieval conditions for each is provided in Supplementary
   Table [417]1. This was followed by incubation with species-specific
   biotinylated secondary antibodies for 15–30 min at room temperature.
   Next, the sections were washed with 1× PBS and incubated with ABC
   reagent (Vectastain ABC Elite kit, Vector Laboratories) for 5–10 min
   followed by washing with 1× PBS. The signal was detected using DAB
   Peroxidase Substrate Kit (Vector Laboratories) using the positive
   control to determine timing of signal. Appropriate negative controls
   were used. Finally, slides were counterstained with hematoxylin (Thermo
   Fisher Scientific), dehydrated, mounted, and cover-slipped. 5× and 10×
   views were imaged on Zeiss Axioskop microscope. Whole-slide tiled
   images were taken with a Fritz Precipoint microscope (Nikon).

Dual-stain for IHC

   For co-staining of glutamine synthetase (GS)/Ki67 via IHC, slides were
   initially incubated with GS (and subsequently with species-specific
   biotinylated secondary antibody) and developed using ABC-AP reagent
   (Vectastain ABC Elite kit, Vector Laboratories) for 30 min and
   incubated with Vector Red reagent (Vector Laboratories) and developed
   in the dark for 15 min. On the second day, slides were incubated with
   Ki67 (and subsequently with biotinylated secondary antibody) and
   developed using ABC reagent (Vectastain ABC Elite kit, Vector
   Laboratories) for 10 min and incubated with DAB reagent (Vector
   Laboratories) and developed in the light for 1–2 min. Finally, slides
   were counterstained with hematoxylin (Thermo Fisher Scientific),
   dehydrated, mounted, and cover-slipped. 5× and 10× views were imaged on
   Zeiss Axioskop microscope (EMD Millipore). Whole-slide tiled images
   were taken with a Fritz Precipoint microscope (Nikon).

Quantification of % area for IHC

   IHC Images stained with indicated markers and quantified with %Area of
   indicated marker were exported as.TIFF files from the Zeiss Axioskop
   microscope and loaded into ImageJ. A color deconvolution plugin was
   downloaded
   ([418]https://github.com/landinig/IJ-Colour_Deconvolution2/blob/main/co
   lour_deconvolution2.jar) and used to deconvolute the different color
   channels. The brown stain channel was used to color threshold based on
   the intensity of the DAB staining on IHC. The same threshold parameters
   were used for each indicated stain comparing LNP-CTRL and LNP-CTNNB1.
   Raw data was exported to an Excel file and analyzed in GraphPad Prism.
   Technical tissue replicates were averaged across to determine the %Area
   for each biological replicate value. Thus, individual data points are
   reported as an average of the technical replicates.

RNA extraction and cDNA preparation

   Whole liver tissue chunks were homogenized in TRIzol (Thermo
   Scientific) and RNA was extracted with RNeasy Micro Kit (Qiagen)
   following the manufacturer instructions. RNA was reverse transcribed
   into cDNA using SuperScript III (Invitrogen). cDNA was diluted in
   nuclease free water and stored long-term at −20 °C.

Quantitative reverse transcriptase polymerase chain reaction (qRT-PCR)

   Real-time PCR was performed in technical duplicates on a StepOnePlus
   Real-Time PCR System (Applied Biosystems) using the Power SYBR Green
   PCR Master Mix (Applied Biosystems). Target gene expression was
   normalized to the housekeeping genes Rn18s, and fold change was
   calculated utilizing the ΔΔ-Ct method. Primers used are listed in
   Supplementary Table [419]2.

qRT-PCR for mouse-specific Ctnnb1 and human-specific CTNNB1

   This analysis was performed by Alnylam Pharmaceuticals. Whole liver was
   homogenized with a tissue grinder and approximately 10 mg of tissue was
   dissolved in QIAzol (Qiagen; 217061). RNA was extracted with miRNeasy
   96 kit (Qiagen; 74004) following manufacturer instructions. RNA was
   reverse transcribed into complementary DNA (cDNA) using Taqman advanced
   fast polymerase chain reaction (PCR) mix (Thermo Scientific; 4374967).
   Real-time PCR was performed in technical duplicate on a Quantstudio
   quantitative PCR (qPCR) System (Thermo Scientific; A28135) using the
   Taqman advanced fast PCR mix (Thermo Fisher; 4444557). Target gene
   expression was normalized to housekeeping gene GAPDH, and fold change
   was calculated utilizing the ΔΔ-Ct method. Primers used are listed in
   Supplementary Table [420]3.

IFN-γ levels in liver tissue (Enzyme-Linked Immunosorbent Assay; ELISA)

   Liver tissue was weighed and homogenized in ice cold, sterile filtered
   1× phosphate buffered saline (PBS), pH 7.4 (10 µL per mg tissue).
   Lysate was incubated on ice for 4 h and centrifuged 2500 × g for 5 min
   at 4 °C. Supernatant was collected and used directly. IFN-γ was
   measured in liver lysate following manufacturers instructions (DY485,
   R&D Systems, Minneapolis, MN). Briefly, Microplates (96 wells) were
   coated with working dilution of capture antibody overnight and blocked
   with 1% bovine serum albumin (BSA) diluted in sterile filtered 1× PBS
   pH 7.4. Samples and detection antibody were incubated for 2 h at room
   temperature with slight agitation. Working concentration of
   streptavidin-horse radish peroxidase (HRP) and color substrate mix
   (DY999, R&D Systems, Minneapolis, MN) were incubated at room
   temperature for 20 min with slight agitation. Plate was washed with 1×
   PBS pH 7. 4 with 0.05% Tween-20 after all steps except color substrate
   incubation. Product development was stopped with 1 N HCl and absorbance
   was measured at 450 and 540 nm on BioTek Synergy XTS Microplate Reader
   (Agilent Technologies, Santa Clara, CA). IFN-γ levels were calculated
   following manufacturer recommendation and converted into pg IFN-γ per
   mg of liver tissue.

Magnetic resonance imaging (MRI) acquisition and analysis

   In vivo MRI was performed using a Bruker AV3HD 9.4 T/30 cm scanner,
   equipped with a BGA-12S HP gradient insert and a 40 mm transmit/receive
   coil, running ParaVision 6.0.1 (Bruker BioSpin, Billerica MA). Mice
   were anesthetized using 1–1.5% Isoflurane via a nose cone, with
   respiration continuously monitored and temperature maintained at 37 °C
   with warm air (SA Instruments, Stoney Brook, NY) during image
   acquisition. Following positioning and pilot scans, T2-weighted axial
   images of the liver were acquired using a Rapid Acquisition with
   Relaxation Enhancement (RARE) sequence, with the following parameters:
   Echo Time/Repetition Time (TE/TR) = 20/4000 ms, averages = 12, a field
   of view (FOV) of 36 × 24 mm, 192 × 128 matrix, 59 slices with a 0.5 mm
   slice thickness and a RARE factor = 8. The T2-weighted images were
   visualized using DSIstudio ([421]https://dsi-studio.labsolver.org).

Bulk RNA-sequencing and analysis pipeline

   Standard pipelines for transcriptome sequencing, quality control, data
   pre-processing, and mapping to mouse reference transcriptome were
   performed^[422]28. There were 5 different mouse whole transcriptome
   datasets analyzed in this study. First, RNA-seq analysis was performed
   on n = 3 PBS vehicle, n = 3 LNP-CTRL, and n = 3 LNP-CTNNB1 at 3 mg/kg
   dosage treated animals from the β-N model after 4 LNP dosages. The
   RNA-seq data for this analysis is deposited to Gene Expression Omnibus
   (GEO) under accession number: [423]GSE290449. Second, RNA-seq analysis
   was performed on n = 3 LNP-CTRL and n = 3–4 LNP-CTNNB1 at 1 mg/kg
   dosage treated animals from each β-catenin-mutated HCC model (β-N & β-M
   models) after 3-days post-treatment. The RNA-seq data for this analysis
   is deposited to GEO under accession number: [424]GSE270414. Third,
   RNA-seq analysis was performed on publicly available transcriptomic
   data from β-catenin-mutated HCC mouse models ([425]GSE125336) and
   microarray gene expression data β-catenin knockout mouse livers
   ([426]GSE68779). Fourth, RNA-seq was performed on liver tumors from
   β-M-pT3 (n = 2) and β-M-IRF2 (n = 4) animals. The RNA-seq data for this
   analysis is deposited to GEO under accession number: [427]GSE270415.
   Fifth, RNA-seq was performed on liver tumors from β-M-pT3 (n = 3) and
   β-M-POU2F1 (n = 3) animals. The RNA-seq data for this analysis is
   deposited to GEO under accession number: [428]GSE290444. For each of
   these different transcriptomic datasets, raw FASTQ files were first
   preprocessed by FastQC for quality control, Trimmomatic for filtering
   out low-quality reads^[429]67, and STAR for reads alignment and gene
   count quantification^[430]68. After pre-processing data normalization
   was performed followed by principal component analysis (PCA). Then,
   based on the gene count data, differential gene expression analysis was
   performed using the R package DEseq2^[431]69 to identify differentially
   expressed genes (DEGs). DEGs (upregulated and downregulated) were
   selected based on filtering criteria of absolute fold change ≥1.5 and
   FDR = 0.05. Volcano plots were used to depict upregulated and
   downregulated genes using EnhancedVolcano package in R. Gene set
   enrichment analysis (GSEA) with clusterProfiler R package using ranked
   genes was performed using Gene Ontology: Biological Processes (GO:BP)
   or Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways from
   Molecular Signatures Database (MsigDB). To discover regulatory
   transcription factors (TFs) of common differentially expressed genes,
   an enrichment test was performed to select TFs with known downstream
   regulatory genes which overlapped with the common differentially
   expressed genes using the JASPAR database^[432]70,[433]71. Heatmap of
   normalized expression values for genes was used to depict expression of
   individual genes between relative conditions. Waterfall plot was used
   to demonstrate positive or negative enrichment of certain pathways
   colored by p-value for each pathway.

Whole liver perfusion for single-cell isolation

   Cells from the liver were isolated using a two-step collagenase
   perfusion^[434]72. Under anesthesia, a 27.5” catheter was used to
   circulate 0.3 mg/mL of collagenase II (Worthington, Lakewood, NJ)
   through the liver via the inferior vena cava. Livers were then
   collected in Dulbecco’s modified Eagle’s medium/F12 supplemented with
   15 mM of HEPES (Corning, Corning, NY) + 5% fetal bovine serum (FBS;
   Biowest, Bradenton, FL). After manual digestion and agitation, cell
   suspensions were passed through a 100- then 70-μm cell strainer and
   washed twice using low-speed centrifugation (50 × g for 5 min) to
   separate out nonparenchymal cells. Hepatocyte viability, determined by
   trypan blue staining, was typically >80%.

   Hepatic lymphocytes and monocytes were enriched from the nonparenchymal
   fraction using a Percoll gradient combined with centrifugation^[435]73.
   Briefly, the nonparenchymal fraction from liver perfusions were
   centrifuged (805 × g for 10 min) followed by 36% Percoll separation in
   Roswell Park Memorial Institute 1640 medium supplemented with +5% FBS.
   Erythrocytes were removed using Ammonium-Chloride-Potassium Lysing
   Buffer (Thermo Fisher Scientific). Isolated cells were counted for
   viability, which was typically >90%, using trypan blue.

Fluorescence-activated cell sorting and gating strategy

   Enriched lymphocytes from liver perfusions were stained with
   fluorescent-conjugated antibodies (Supplementary Table [436]4) and a
   Fixable Dead Cell Stain (Thermo Fisher Scientific). Data was acquired
   using Cytek Aurora (Cytek Biosciences, Bethesda, MD) equipped with five
   lasers and analyzed using FlowJo software v10.1 (Treestar, Ashland,
   OR).

Single-cell RNA-sequencing and analysis

   Hepatocytes isolated and purified from whole liver perfusions were
   sequenced along with the enriched nonparenchymal cell fraction (mostly
   lymphocytes/monocytes) were used for tandem scRNA-sequencing coupled
   with immune profiling separately. For the hepatocyte-enriched cell
   population sequenced (GEX; [437]GSE270714), cells isolated from 2–3
   individual animals were pooled together in each treatment condition and
   sequenced in separate wells on the flow cell. For the immune-enriched
   population (scRNA-seq with immune profiling; [438]GSE270974), cells
   isolated from 6 individual animals (3 LNP-CTRL and 3 LNP-CTNNB1) were
   feature barcoded, hashed, and then pooled to be sequenced in one well
   on the flow cell. Specifically, the nonparenchymal cell fraction was
   feature barcoded and marked using TotalSeq™-C Mouse Universal Cocktail
   (Biolegend, San Diego, CA) following manufacturer’s recommendations,
   along with BCR and TCR library preparation. At minimum 10,000
   hepatocytes were sequenced (non-multiplexed wells) and 30,000 immune
   cells (multiplexed wells) were sequenced with at least 20,000 reads per
   cell in the non-multiplexed wells and 5000 reads per cell in the
   multiplexed wells. Multi-plexing was performed according to standard
   CellPlex protocols. Single-cell sequencing was performed on NovaSeq
   2000 instrument and library prep with 5PrimeV2 was performed according
   to standard protocols. The raw FASTQ files for the pooled library were
   first processed using CellRanger for alignment against mm10 mouse
   genome using standard CellRanger multi pipeline with the 5p hashtag
   demultiplexing configuration file. Then, the demultiplexed aligned BAM
   files were re-formatted into FASTQ files per sample. Lastly, the
   CellRanger multi was applied to individual samples to align the data to
   mouse genome by integrating multi-omics data: gene expression, antibody
   capture, VDJ-T and VDJ-B. Ultimately, we generated the gene count
   matrices on a per-sample and per-cell basis for downstream analysis.

Analysis pipeline

   We used standard Seurat^[439]74 pipelines for quality control of the
   data at cell and gene level. We made sure to not include empty cells,
   doublets/multiplets, or dead/dying cells. This was determined by using
   nFeature_RNA > 500 & nFeature_RNA < 2000, nCount_RNA ≥ 0 &
   nCount_RNA < 10,000, and percent.mt ≥ 0 & percent.mt <10 for analysis
   of the GEX dataset ([440]GSE270714); and, nFeature_RNA > 200 &
   nFeature_RNA < 5000, nCount_RNA ≥ 0 & nCount_RNA < 12,000, and
   percent.mt ≥ 0 & percent.mt <10 for analysis of the scRNA-seq with
   immune profiling dataset ([441]GSE270974). Quality control was
   visualized with violin plots of each of these parameters. After this
   step, we resulted in 26,851 high-quality cells from LNP-CTRL and 67,799
   high-quality cells from LNP-CTNNB1 treatment for GEX dataset
   ([442]GSE270714). Additionally, we resulted in 8499 high-quality cells
   from LNP-CTRL and 11,736 high-quality cells from LNP-CTNNB1 treatment
   for scRNA-seq with immune profiling dataset ([443]GSE270974).

   For each of the matrices, we integrated the LNP-CTRL and LNP-CTNNB1
   datasets together using standard Seurat v4 pipeline of
   SelectIntegrationFeatures, PrepSCTIntegration, FindIntegrationAnchors,
   and IntegrateData using n = 3000 features. Then, dimensionality
   reduction method of principal component analysis (PCA) and uniform
   manifold approximation and projection (UMAP) was performed^[444]75.
   Initial cluster calls were generated using FindNeighbors and
   FindClusters function with dims = 1:30 and resolution = 0.2. Then, the
   dataset was swapped to RNA assay for visuals and variable features were
   identified, along with normalization and scaling of the data using
   FindVariableFeatures, NormalizeData, and ScaleData functions. Red blood
   cell contamination was inspected via Feature plots and Violin plots of
   Hba-a1 and Hbb-bs genes. To identify specific cell types, present in
   each cluster, the FindAllMarkers function was used to determine
   differentially expressed genes per cluster compared to all other
   clusters. This identified 16 clusters in total for the
   hepatocyte-enriched dataset and 22 clusters in total for the immune
   cell-enriched dataset. We utilized the marker genes for different cell
   types present in Supplementary Table [445]5 and Supplementary
   Table [446]6, which is based on the literature and manual
   curation^[447]36. Clusters were then annotated and cell-type
   proportions per treatment condition were determined for each of the two
   datasets and visualized with either pie chart or stacked bar graphs.

   Differential gene expression analysis, comparing each cluster to all
   other clusters, was performed on the integrated datasets for each cell
   type/cluster using the Seurat FindMarkers function to identify genes
   differentially expressed for each cell type irrespective of treatment
   condition. Differential gene expression within each cluster between
   treatment conditions was performed using the Seurat FindMarkers
   function and renaming the idents such that cells from each treatment
   condition were a different ident in the comparison. Differentially
   expressed genes (DEGs) were identified based on cut-off thresholds of
   adjusted p-value of <0.05 and absolute log2 Fold Change >0.25, or more
   stringent cutoff used in case of many DEGs. We utilized volcano plots
   to illustrate specific DEGs using the EnhancedVolcano R package with
   p-value cutoff of p = 0.05. Pathway enrichment analysis was performed
   using clusterProlifer R package using the DEGs identified and
   referencing GO:BP or KEGG pathways from MSigDB. Top pathways were
   visualized using dot plot function in clusterProlifer R package.

Cell cycle regression analysis

   Cell cycle scoring was performed on the hepatocyte-enriched dataset
   using Seurat CellCycleScoring function. The expression matrix was
   retrieved from the source vignette and the S phase and G2M phase genes
   were extracted and converted to mouse orthologs from human genes. Cells
   that are not in S or G2M phase are classified as G1 phase. Pie charts
   were used to visualize distribution of percentage of cells in each
   phase of the cell cycle for each cluster.

Pseudotime analysis

   Pseudotime trajectory analysis was performed on the integrated
   hepatocyte-enriched dataset following subsetting for only the normal
   hepatocyte or tumor hepatocyte clusters (Zone 3 CTNNB1 WT and MUT
   (GS+), Zone 1/2 CTNNB1 MUT (GS+), Zone 1 CTNNB1 WT (GS−), Reprogrammed
   Hepatocytes) using the monocle3 R package and standard
   pipelines^[448]76. First, a Cell Data Set (CDS) object was created,
   which transferred the UMAP cluster labels and cell metadata. Second,
   pseudotime analysis was performed and included calculations of
   potential cell trajectories using the published algorithm. The root
   node was chosen as the Zone 3 CTNNB1 WT and MUT (GS+) cell cluster. The
   results were visualized using the plot_cells function.

Cell-cell interaction analysis

   Cell-to-cell interaction analysis on the immune cell-enriched
   population was performed using the CellChat V1 R package^[449]37. This
   package allows for analysis and inference of cell-to-cell communication
   using relative ligand-receptor expression. For visualization of
   specific communication pathways (i.e., ligand-receptor pairs), we used
   all the standard pipelines built in the CellChat R package^[450]37.

Single-cell spatial transcriptomics using Resolve Molecular Cartography^TM

   Mouse liver tissue was preserved frozen using OCT media following
   submersion in cold 2-methyl-butane. Frozen mouse liver tissue was
   sectioned with a cryostat (Leica, CM 1850-3-1) at a temperature of
   −17 °C. Tissue sections were 10μm thick and placed on the Resolve
   Biosciences slide also cold at −17 °C. Following sectioning and
   placement, the slide was stored at −80 °C overnight, then shipped to
   Resolve BioSciences on dry ice for analysis. Probes were devised using
   Resolve BioSciences’ proprietary algorithm^[451]77. The 100-gene panel
   used in this study is described in Supplementary Table [452]7. Resolve
   BioSciences performed the tissue fixation, 100-plex priming,
   hybridization, slide imaging, and fluorescent-tagging using their
   standard pipelines^[453]77. Then, regions of interest were imaged with
   fluorescent signals removed during a decolorization step to achieve a
   unique combinatorial code for each target gene in the panel^[454]77.
   The final images of the Molecular Cartography^TM signals could then be
   viewed on Resolve BioSciences website which has a plugin tool for
   viewing the signals of each of the 100 genes at different magnification
   scales. We used tissue from one animal per LNP treatment group.
   Ultimately, 5-6 regions of interest (ROIs) were selected per tissue
   section for analysis. Single-cell spatial transcriptomic analysis was
   performed by quantifying gene counts per cell using automatic cell
   segmentation via QuPath software^[455]78. The libraries from each
   treatment condition (LNP-CTRL and LNP-CTNNB1) were then integrated
   together using the R package Seurat^[456]74,[457]79. Analyzed cells
   included those filtered for greater than or equal to 10 gene counts per
   cell. Following these quality control steps, we performed unbiased
   clustering on all 100 genes using the dimensionality reduction method
   of principal component analysis (PCA) and uniform manifold
   approximation and projection (UMAP)^[458]75. Clusters corresponding to
   different cell types, including stellate, endothelial, and immune
   cells, and zones were identified and annotated from the UMAP based on
   expression of different landmark genes described in Supplementary
   Table [459]8^[460]77. The cluster proportions per treatment condition
   were depicted as pie charts or stacked bar plots.

   Next, using Seurat, feature plots and violin plots were used to
   visualize gene expression across different clusters and between
   treatment conditions. Additionally, dot plots were used to visualize
   expression of specific genes within each cluster in terms of number of
   cells expressing that gene in each cluster and the expression level.
   Normalized expression values of each of the 100 genes were also
   visualized with heatmap across each of the different clusters. Lastly,
   once clusters were annotated using Supplementary Table [461]8 landmark
   genes, cells corresponding to these specific clusters were mapped back
   onto the virtual slide using SpatialPlot function to visualize spatial
   location of specific clusters using a different color for each cluster,
   which corresponded to the same cluster color on the UMAP plot.
   Moreover, pseudotime trajectory analysis to define cell trajectories
   and cell states was performed on the integrated dataset using all
   clusters with the monocle3 R package^[462]76. The same pipelines for
   this were performed as described above in the single-cell
   RNA-sequencing dataset. The root node was chosen as the H1: Zone 3
   CTNNB1 MUT (GS+) cell cluster. Lastly, within each ROI, tumor nodules
   were outlined based on the expression of Glul as a landmark gene for
   β-catenin-mutated tumor nodules. Then, gene counts and expression
   density (gene counts per area) were quantified within each tumor
   boundary and averaged across the defined regions (tumor or non-tumor).
   This was also performed on a cell basis to determine cell density per
   region (tumor or non-tumor). The single-cell spatial transcriptomic
   data from this analysis is deposited to Gene Expression Omnibus (GEO)
   under accession number: [463]GSE270708.

10× visium spatial transcriptomics

   Tissue sections of 10 μm thickness were cut from mouse liver tumor FFPE
   blocks and were deparaffinized, H&E stained, then decrosslinked using
   standard 10× genomics protocols. Regions of interest (ROIs) were
   selected by outlining ~6.5 mm^2 areas. Standard Visium CytAssist
   protocols were used for FFPE tissue for generation of sequence ready
   libraries. Quality control parameters were a DV200% ≥30%. For 10×
   Visium spatial transcriptomics bioinformatic analysis, sequencing reads
   from 55 µm tissue regions (spots) were first preprocessed by 10× Space
   Ranger software for alignment and spatial gene count quantification.
   The processed data were then imported into R using the package
   Seurat^[464]74. For the β-M model, spots across 4 samples (β-M LNP-CTRL
   [BL-131], 2 β-M LNP-CTNNB1 NR [BL-146, BL-147], and 1 β-M LNP-CTNNB1 R
   [BL-152]) were normalized and integrated. For the β-N model, spots
   across 5 samples (β-N LNP-CTRL [BL-551], 2 β-N LNP-CTNNB1 NR [BL-584,
   BL-587], and 2 β-N LNP-CTNNB1 R [BL-585, BL-586]) were normalized and
   integrated. For both datasets, following normalization and integration,
   dimensionality reduction via principal component analysis and uniform
   manifold approximation was performed. K-means clustering identified 17
   clusters in the β-M model and 13 clusters in the β-N model, which were
   overlaid on hematoxylin and eosin sections using the SpatialPlot
   function in Seurat. Differential gene expression was used to identify
   up- and downregulated genes in each cluster using the FindAllMarkers
   function in Seurat with return.thresh = 0.05. We utilized volcano plots
   to illustrate specific DEGs (defined by adjusted p-value < 10e-32 and
   absolute log2FC > 0.5) using the EnhancedVolcano R package. Cell-type
   composition in each cluster was inferred based on the expression of
   marker genes in Supplementary Table [465]9 from literature and manual
   curation^[466]36. Marker genes were visualized in dot plot format
   across clusters and by treatment condition. Gene set enrichment
   analysis was performed with the clusterProfiler R package^[467]80 using
   ordered ranked gene lists with gene sets from GO:BP from MSigDB
   collection^[468]81, with a p-value threshold of 0.05. Pathway
   enrichment analysis was performed with the clusterProfiler R
   package^[469]80 using DEGs (defined by adjusted p-value < 0.05 and
   absolute log2FC > 0.25) and gene sets from GO:BP and KEGG from MSigDB.
   Upstream regulator analysis was performed using Qiagen software with
   the DEGs from each cluster. Cell cycle regression and pseudotime
   analysis were performed as described in the “Single-cell RNA-sequencing
   and analysis” section. Version 2 of the CellChat R package^[470]82 was
   used to determine probabilities of spatial cell-cell communication
   between clusters, based on expression of ligands and receptors in
   spatially proximal spots. The 10× Visium transcriptomic data from this
   analysis is deposited to Gene Expression Omnibus (GEO) under accession
   number GSE270997 (subseries: [471]GSE270975 and [472]GSE290445).

TCGA-LIHC human HCC data mining

   The Cancer Genome Atlas (TCGA) RNA-seq whole exome sequencing and
   transcriptomic data were downloaded from Genomic Data Commons (GDC)
   through the R Bioconductor package GenomicDataCommons. We selected
   patients from TCGA with Liver Hepatocellular Carcinoma (TCGA-LIHC) for
   analysis. This dataset contains 424 cases with 50 of these being
   adjacent normal liver tissue. Raw gene counts were normalized and used
   for downstream analyses. We identified that 98 out of the 374 TCGA-LIHC
   patients had CTNNB1 mutations, 18 had AXIN1 mutations, and 3 had APC
   mutations. Differential gene expression was performed to identify
   pathways differentially regulated between CTNNB1-mutated and normal
   patients and CTNNB1-mutated and CTNNB1-wild-type patients using gene
   set enrichment analysis with GO_BP pathways from MsigDB. IRF2 and
   POU2F1 target genes plotted in the heatmap are provided in
   Supplementary Table [473]10. The 33 IRF2 and 10 POU2F1 target genes
   which were intersected for common overlap and plotted (39 common
   genes). The 39 genes identified were inferred IRF2/POU2F1 target genes
   retrieved from [474]https://maayanlab.cloud/chea3/ and which overlapped
   with the 162 common genes which were upregulated in β-catenin knockout
   livers and downregulated in β-catenin-mutated HCC mouse livers. A
   heatmap of z-scored values was generated to overlap patients with and
   without tumor, and with and without CTNNB1, AXIN1, or APC mutations,
   and the relative normalized expression scores of IRF2/POU2F1 target
   genes. Boxplots were used to visualize expression of IRF2, POU2F1, or a
   module score of the IRF2/POU2F1 target genes. Module score was
   calculated based on normalized mean expression of the common 39 target
   genes. Survival data was downloaded directly from cBioPortal website
   and integrated with the expression data.

IMbrave150 phase 3 retrospective clinical trial analysis

   We retrospectively analyzed genomic and transcriptomic data (Whole
   Exome Sequencing and RNA-seq data) from the IMbrave150 trial^[475]6 and
   associated this data with CTNNB1 mutational status, tertiary lymphoid
   structure/lymphoid aggregate (TLS/LA)-like signature expression score,
   TLS/LA/diffuse infiltrate (DI) presence, and clinical parameters
   (overall and progression-free survival [OS/PFS] and clinical response
   using mRECIST criteria). First, TLS/LA/DI/ presence was assessed by
   Genentech Pathologist (author H.K.) and associated with clinical
   response based on mRECIST criteria (CR/PR, SD/PD, or NA)^[476]6.
   Second, TLS/LA/DI presence was associated with PFS and OS in each
   treatment arm (atezolizumab plus bevacizumab or sorafenib). Third, a
   previously described B and CD4 Tconv and CD4 Tfh cell signature
   expression score was merged and overlapped with genes sequenced in
   TCGA-LIHC, resulting in 140 genes, which was associated with TLS/LA/DI
   presence, clinical response based on mRECIST criteria, and CTNNB1
   mutational status^[477]45. The source data has been previously
   deposited in the European Genome-Phenome Archive under accession no.
   EGAS00001005503.

Public single-cell human and mouse liver analysis

   Single-cell data from both human and mouse livers were accessed from
   the LiverCellAtlas website directly
   ([478]https://www.livercellatlas.org/). The human data was accessed
   here and images of the UMAP plots were taken
   ([479]https://www.livercellatlas.org/umap-humanAll.php). The mouse data
   was accessed here and images of the UMAP plots were taken
   ([480]https://www.livercellatlas.org/umap-ststmouseAll.php). Expression
   of IRF2/Irf2 and POU2F1/Pou2f1 were queried in these datasets to
   investigate cell-type specific expression of these transcription
   factors in normal human/mouse liver.

Statistical analysis

   The data presented throughout the manuscript is represented as
   mean ± standard deviation (SD) for each bar plot. The indicated
   statistical tests were performed in either R or Prism 10.3.1 software
   (GraphPad Software Inc., Boston, Massachusetts, USA,
   [481]www.graphpad.com). For our study, P < 0.05 was considered
   statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001,
   ****p < 0.0001).

Reporting summary

   Further information on research design is available in the [482]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [483]Supplementary Information^ (20.8MB, pdf)
   [484]Reporting Summary^ (1.2MB, pdf)
   [485]Transparent Peer Review file^ (4.1MB, pdf)

Source data

   [486]Source Data^ (2.4MB, xlsx)

Acknowledgements