WNT–β-catenin signalling in hepatocellular carcinoma: from bench to clinical trials

Abstract

WNT–β-catenin activation is observed in around 50% of all patients with hepatocellular carcinoma (HCC), through either gain-of-function mutations in CTNNB1 (which encodes β-catenin) or loss-of-function mutations in AXIN1 or APC. Currently, first-line therapies for HCC are immune checkpoint inhibitor (ICI) combinations, and β-catenin-active HCCs have garnered increased attention due to their unique tumour immune microenvironment (TIME). This pathway is known to drive an immune-excluded TIME, but clinical investigations have provided a more nuanced perspective, with the emergence of a new ‘immune-like’ subclass of HCC that is paradoxically enriched for CTNNB1 mutations and has high levels of T cell infiltration. As such, patients and animal models with β-catenin activation treated with ICIs exhibit heterogeneous responses. Additionally, these tumours exhibit higher fatty acid oxidation to fuel tumour growth owing to a unique metabolic milieu shaped by zone 3 metabolism, which is a physiological function of WNT–β-catenin signalling in the liver lobule. Biomarkers to detect molecular subclasses of patients for targeted therapies are being developed. In this Review, we discuss advances in our understanding of the TIME and metabolism of β-catenin-active HCC, driven by in vitro and in vivo models and single-cell and spatial sequencing, and their implications for the treatment of a subset of HCCs using precision therapies against WNT–β-catenin signalling.

Introduction

The global burden of hepatocellular carcinoma (HCC) that accounts for ~90% of primary liver cancers remains a significant challenge. It is the sixth most common cancer and the third leading cause of cancer-related mortality worldwide, with men twice as likely to be affected as women^1–3. HCC typically develops on a background of chronic liver disease (CLD) due to either viral aetiologies (for example, hepatitis viruses) or non-viral aetiologies (for example, alcohol and metabolic syndrome). Approximately 1–5% of patients with cirrhosis develop HCC per year^4,5. Viral hepatitis-related HCCs are decreasing in prevalence in Western countries and regions, owing to widespread hepatitis B virus (HBV) vaccination and use of antiviral medications to treat hepatitis C virus (HCV) infection. However, the rise in metabolic syndrome globally, leading to metabolic dysfunction-associated steatotic liver disease (MASLD), has now become the leading cause of HCC in the Western world. For example, since 2002, the prevalence of HCC in patients listed for liver transplantation in the USA with metabolic dysfunction-associated steatohepatitis (MASH) has increased 11.8-fold⁶. Currently, there are no effective treatments to prevent the progression of CLD to HCC, and current surveillance programmes are limited in patients with MASLD, who might not have underlying liver cirrhosis⁷ but are at increased risk of HCC⁸. Screening is critical as surgical options are most effective in early-stage disease.

Specifically, current surveillance guidelines from the American Association for the Study of Liver Diseases and the European Association for the Study of the Liver recommend surveillance using ultrasound with or without serum α-fetoprotein (AFP) biannually for high-risk individuals (that is, individuals with cirrhosis of any aetiology or individuals without cirrhosis with chronic HBV infection)⁹. Once a lesion is identified, contrast-enhanced CT–MRI can detect HCC. HCC is the only solid tumour that does not require biopsy to reach a clinical diagnosis. For 40–50% of patients with detected early-stage lesions (that is, Barcelona Clinic Liver Cancer (BCLC) stage 0 or A), either radiofrequency or microwave ablation, liver resection or transplantation, are curative, improving 5-year survival to >60%¹⁰. Patients with BCLC stage B disease are typically treated with locoregional ablative therapies. In those with advanced HCC (BCLC stage C), systemic agents are indicated. Ultimately, >50% of patients will either be diagnosed with advanced stage disease or have disease recurrence and require systemic management³. With systemic therapy, median overall survival (OS) is <15 months, with 5-year OS <20%. Improved understanding of HCC biology for rational treatment design should improve OS. The activation of the WNT–β-catenin pathway in a notable subset of individuals with HCC is associated with unique tumour biology, distinct therapeutic response to current regimens and opportunities for novel diagnostic approaches and therapeutic interventions, all important topics that are covered in this Review.

Molecular drivers and subclasses in HCC

Over the past two decades, whole-exome sequencing (WES) has paved the path towards precision medicine through profiling the mutational landscape of HCC. The major trunk mutations in HCC are gain-of-function mutations in either the TERT promoter (~50–60%) or CTNNB1 (~30%) (which encodes β-catenin, part of the WNT–β-catenin signalling pathway), or loss-of-function mutations in TP53 (~30%)¹¹. Less frequent mutations in HCC are observed in ARID1A (~10%), TSC1/2 (~10–15%), FGF19 (~10%), AXIN1 (~6–10%), MYC (~6%), APC (~5%), ARID2 (5%), MET (~2%) and others. CTNNB1 and TP53 mutations, and CTNNB1 and AXIN1 mutations, are usually mutually exclusive^12–14. Unfortunately, the more frequent oncogenic drivers in HCC, such as TERT, TP53, CTNNB1, ARID1A/2 and MYC, have been difficult to target. Amplifications of FGF19, VEGFA or MET, or mutations in PI3KCA, HRAS, NRAS and TSC1/2, are currently directly druggable and occur in ~25% of individuals with HCC^12,15. Moreover, mutational frequencies differ by underlying disease aetiology, sex, age and ethnic background. For example, mutations in CTNNB1 are significantly enriched in Hispanic patients¹⁶. Globally, various studies have identified enrichment of CTNNB1 mutations in alcohol-driven, HCV-driven and MASLD-driven HCC^17–19, which are risk factors observed in Black and Hispanic populations, especially in the USA⁵. Interestingly, CTNNB1-mutated HCC can occur in individuals without cirrhosis and in older individuals, and has been associated with MASLD-driven HCC^20,21. In individuals with a non-cirrhotic liver, type 2 diabetes mellitus induces extracellular matrix reorganization through build-up of advanced glycation end products, which promotes HCC in combination with mutant β-catenin signalling²². Lastly, aflatoxin exposure can cause enrichment for mutations in genes encoding WNT scaffolding proteins (for example, AXIN1 and APC)²³. Thus, there exists an intersection between tumour genetics, social determinants of health, and ethnicity. Overall, each combination of driver mutations in a patient ultimately influences the underlying tumour biology, which has formed the foundation for defining unique molecular subclasses in HCC.

There are three major molecular classification systems based on whole-transcriptome sequencing. These molecular classifications are associated with distinct histopathological features²⁴, but have not been used clinically. The Boyault molecular classification categorizes HCC into six molecular subgroups (G1 to G6)²⁵, the Chiang classification categorizes HCC into five molecular subgroups (‘CTNNB1’, ‘proliferation’, ‘interferon’, ‘Poly7’ and ‘unannotated’)²⁶, and the Hoshida scheme categorizes HCC into three subgroups (S1 to S3)²⁷. While β-catenin ‘activation’ is observed in the Boyault G1 and G2 subgroups and the Hoshida S2 subgroup with enrichment of AXIN1 mutations, and the Boyault G2 and G3 subgroups and the Hoshida S1 subgroup are enriched for WNT–TGFβ signalling²⁸, activation of the canonical WNT–β-catenin pathway is evident in the Hoshida S3 subgroup, the Chiang CTNNB1 subgroup, and the Boyault G5 and G6 subgroups, which are enriched in CTNNB1 mutations. Histologically, these are characterized as cholestatic with microtrabecular and pseudoglandular patterns. There is now an improved understanding of how tumour histology, genetics and molecular biology intersect. These molecular subclasses are eventually expected to aid personalized treatments as liquid or tissue biopsy becomes a requirement for clinical trial enrolment. Most pertinent to this Review is HCC that is driven by the activation of the canonical WNT–β-catenin pathway, which is represented by the Hoshida S3 subgroup, the Chiang CTNNB1 subgroup, and the Boyault G5 and G6 subgroups, all of which show specific histological hallmarks and other unique genotype–phenotype correlates.

Current treatment landscape and heterogeneous response to ICIs

The treatment armamentarium for liver oncologists has greatly expanded over the past decade. Initially, since the 2007 approval of the multi-tyrosine kinase inhibitor (TKI) sorafenib as first-line therapy for advanced HCC, very few agents have subsequently been approved²⁹. Nearly 10 years later, other TKIs, including lenvatinib³⁰, regorafenib³¹ and cabozantinib³², were approved for use in the first-line or second-line setting. With the advent of immunotherapy, which harnesses the ability of the patient’s own immune system to recognize and target cancer cells, the efficacy of immune checkpoint inhibitors (ICIs) in HCC has also been investigated. In 2017 and 2018, nivolumab and pembrolizumab monotherapy were given approval as second-line therapy for advanced HCC. Three phase III clinical trials evaluating ICI combinations compared with sorafenib in the first-line setting demonstrated improved OS and progression-free survival (PFS) with these combinations, and led to their FDA approval for the first-line treatment of advanced HCC. The approved combinations are atezolizumab (anti-PDL1) plus bevacizumab (anti-VEGF)^33,34, durvalumab (anti-PDL1) plus tremelimumab (anti-CTLA4)^35,36, and ipilimumab (anti-CTLA4) plus nivolumab (anti-PD1)³⁷. The backbone of anti-PD(L)1 plus either anti-VEGF or anti-CTLA4 is the current standard of care for advanced HCC in the first-line setting. Other phase III clinical trials with ICI combinations have resulted in similar efficacy and safety and tolerability, but have demonstrated worse clinical outcomes irrespective of patient subgroup analyses^38–41. A meta-analysis demonstrated that the combination of atezolizumab plus bevacizumab is associated with superior or similar OS and PFS compared with all combinations⁴².

Despite these significant advances, not all patients demonstrate a clinical response, and the majority of studies have found an objective response rate of ~30%. Additionally, HCC is currently the only solid tumour type with no FDA-approved tissue-based or liquid-based biomarkers to predict response to ICIs. However, biopsy is being integrated more into clinical management, and is improving our understanding of the pretreatment tumour immune microenvironment (TIME)⁴³. Studies have suggested that tertiary lymphoid structures (TLS) could predict responses to ICIs in multiple solid tumours⁴⁴. The role of TLS in HCC prognostication remains unclear despite their importance in generating potent antitumour immune responses^45,46. Artificial intelligence (AI)-based algorithms trained on haematoxylin and eosin (H&E)-stained slides in combination with ‘omics’ might aid identification of novel tissue biomarkers in ICI-treated patients^47,48. Retrospective analyses using transcriptomics, comparing responders to non-responders from various ICI trials, have derived response signatures, but have not been prospectively validated^49–51. Blood-based biomarkers utilizing AFP in conjunction with other markers (for example, cell-free DNA (cfDNA), cytokines) are being used to predict immune activity⁵².

WNT–β-catenin pathway activation has received increased attention due to its association with an immune-exclusion phenotype in HCC and other tumours^53–55. For ICIs to be effective, the presence of functional immune cells, particularly T cells, within the tumour microenvironment is critical. Globally, β-catenin-active tumours have been considered ‘immune-excluded’ owing to various biological mechanisms that contribute to these tumours not being optimally responsive to ICIs⁵⁴. However, there are conflicting reports of WNT–β-catenin activation influencing responses to ICIs in HCC, although each study utilized different ICI treatment approaches and some lacked a comparator arm^{15,50,51,56–67} (Table 1). There are other limitations to these studies, including being insufficiently powered, using different definitions of WNT–β-catenin activity (that is, mutational or immunohistochemical), and inconsistent reporting of clinical outcomes. In this Review, we focus on preclinical and clinical investigations related to the activation of WNT–β-catenin signalling, emphasizing cell-intrinsic biology modulating tumour metabolism and TIME. We attempt to reconcile conflicts in the findings of studies evaluating mutated β-catenin as a biomarker of resistance to ICIs, and discuss future strategies for targeting the WNT–β-catenin pathway in HCC based on an understanding of the tumour biology.

Table 1 |.

Clinical studies evaluating WNT–β-catenin as a biomarker of ICI response in HCC

Study	ICI treatment	Comparator arm	Patients with WNT alteration (n)	Phenotype reported
Tissue-biopsy-based CTNNB1 detection
Ruiz de Galarreta et al. (2019)56	Nivolumab monotherapy	NA	15 total: 3 with CTNNB1 mutation 8 with wild-type CTNNB1 4 unknown status	1 of 3 with CTNNB1 mutation responded 4 of 8 with wild-type CTNNB1 responded 1 of 4 unknown status responded
Harding et al. (2019)15	Anti-CTLA4 monotherapy (n = 1) Anti-PD(L)1 monotherapy (n = 25) Anti-PD(L)1 + anti-CTLA4 (n = 1) Anti-PD(L)1 + anti-LAG3 (n = 2) Anti-PD(L)1 + anti-KIR (n = 2)	NA	127 total 35.7%, 6.3%, 0.05% and 5.2% had mutations in CTNNB1, AXIN1, AXIN2 and APC, respectively 31 treated with immunotherapy (10 had WNT alteration; 27 evaluable)	0 of 10 with WNT alteration had progressive disease 9 of 17 without WNT alteration had durable stable disease WNT-activating tumours associated with shorter PFS (2.0 vs 7.4 months) WNT-activating tumours associated with shorter OS (9.1 vs 15.2 months)
Morita et al. (2021)61	Anti-PD1 monotherapy not specified (n = 34)	NA	34 with HCC (14 of 34 (41.2%) had WNT-β-catenin activation) Activation determined by β-catenin nuclear and GS staining IHC	WNT-β-catenin activation had a modest negative association with disease control (P = 0.0509 by multivariate logistic regression) WNT-β-catenin activation associated with significantly worse OS and PFS (P<0.001)
Zhu et al. (2022)50	Atezolizumab + bevacizumab (n = 85)	Sorafenib (n = 35)	130 profiled with whole-exome sequencing (48 with CTNNB1 mutation)	CTNNB1 mutation no difference in OS (HR 0.49, P = 0.2) or PFS (HR 1.1, P = 0.51)
Kuwano et al. (2022)57	Atezolizumab + bevacizumab (n = 24)	NA	15 profiled had WNT-β-catenin activation, while 9 had inactivation	WNT-β-catenin active: CR 0, PR 7, SD 3, PD 5, CR + PR 46% WNTT-β-catenin inactive: CR 1, PR 2, SD 2, PD 4, CR + PR 33%
Hong et al. (2022)60	Pembrolizumab (n = 60) after sorafenib failure	NA	10 out of the 47 profiled had CTNNB1 mutation (21.3%)	All patients with CTNNB1 mutations were non-responders to pembrolizumab
Haber et al. (2023)51	Nivolumab (n = 67) Pembrolizumab (n = 14) Tislelizumab (n = 2)	NA	83 available with transcriptomic analysis 28 (first-line, 23 with CTNNB1 mutation) 55 (second-line or third-line, 44 with CTNNB1 mutation)	4 of 11 (36%) responders had CTNNB1 mutation 6 of 12 (50%) non-responders had CTNNB1 mutation No difference in PFS or OS
Kurebayashi et al. (2023)58	Atezolizumab + bevacizumab (n = 33)	NA	13 of 33 (39.4%) had nuclear β-catenin and diffuse GS positivity on IHC	WNT-β-catenin subclass not significant prognosticator for PFS (HR 0.71, P = 0.46) or OS (HR 0.57, P = 0.41)
Kuwano et al. (2024)67	Tremelimumab plus durvalumab (n = 15)	NA	WNT-β-catenin active signalling defined by β-catenin nuclear staining in ≥5% of tumour cells or diffuse GS staining in ≥50% of tumour cells	4 patients had WNT-β-catenin active signalling and low CD8+ T cell infiltration; all these patients had no response (2 SD, 2 PD) 9 patients were GS-positive or showed nuclear β-catenin staining (2 PR, 3 SD, 4 PD)
Liquid-biopsy-based CTNNB1 detection
von Felden et al. (2021)59	Nivolumab or pembrolizumab (n = 38 overall, 97% nivolumab)	NA	6 patients with WNT alteration (4 CTNNB1-mutant); profiled by circulating tumour DNA	No significant difference in PFS (P = 0.32) or OS (P = 0.57) comparing patients with and without WNT alterations
Matsumae et al. (2022)62	Atezolizumab + bevacizumab (n = 85)	NA	CTNNB1 mutations by profiling cfDNA were found in 13 patients (15%)	No significant difference in PFS (P = 0.968) or OS (P = 0.733) comparing patients with CTNNB1 mutation and patients with wild-type CTNNB1
Ogawa et al. (2022)64	Atezolizumab + bevacizumab (n = 33)	NA	8 patients (24.2%) had mutations in exon 3 of CTNNB1	ORR was 12.5% in patients with CTNNB1 mutation and 8.0% in patients with wild-type CTNNB1 (P > 0.999) Disease control rate was 87.5% in patients with CTNNB1 mutation and 68.0% in patients with wild-type CTNNB1 (P = 0.394) PFS was 7.6 months in patients with CTNNB1 mutation and 6.6 months in patients with wild-type CTNNB1 (P = 0.772) OS was 12.3 months in patients with CTNNB1 mutation and 13.6 months in patients with wild-type CTNNB1 (P = NS)
Mohamed et al. (2024)63	Nivolumab (n = 44)	NA	CTNNB1 mutations by profiling circulating tumour DNA were detected in 16 patients (36.4%)	No significant difference in OS (P = 0.37) but significant difference in PFS (P = 0.04): patients with CTNNB1 mutation had longer median PFS than patients with wild-type CTNNB1 (7.42 vs 3.45 months)
Campani et al. (2024)65	Atezolizumab + bevacizumab (n = 35)	NA	CTNNB1 mutations by profiling cfDNA (high) were detected in 8 patients (22.9%) Analysis was done based on median cfDNA concentration	No significant differences in OS (P = 0.572) and PFS (P = 0.116)
Terashima et al. (2025)66	Atezolizumab + bevacizumab (n = 52)	NA	CTNNB1 mutations by profiling cfDNA (high) were detected in 24 patients (46%)	ORR, tumour control rate and PFS in patients with CTNNB1 mutations vs patients with wild-type CTNNB1 were 37.5% vs 28.6%, 79.2% vs 96.4% and 10.3 vs 5.6 months, respectively

cfDNA, cell-free DNA; CR, complete response; GS, glutamine synthetase; HCC, hepatocellular carcinoma; ICI, immune checkpoint inhibitor; IHC, immunohistochemistry; NA, not applicable; NS, not significant; ORR, overall response rate; OS, overall survival; PD, progressive disease; PFS, progression-free survival; PR, partial response; SD, stable disease.

Mechanisms of WNT–β-catenin activation in HCC

β-Catenin levels are tightly regulated in a cell through interaction with the ‘destruction complex’ that consists of adenomatous polyposis coli (APC), AXIN1/2, glycogen synthase kinase 3β (GSK3β) and casein kinase 1α (CK1α). Phosphorylation at serine and/or threonine residues at codons 33, 37, 41 or 45 by GSK-3β allows ubiquitin-mediated proteasomal degradation of β-catenin. In the presence of WNT ligands that bind to canonical Frizzled receptors and LDL receptor-related protein co-receptors, β-catenin translocates to the nucleus and associates with the T cell factor–lymphoid enhancer factor (TCF–LEF) family of transcription factors to activate expression of target genes. This process of β-catenin nuclear translocation is facilitated by recruitment of dishevelled to the cell membrane, with subsequent AXIN1 and GSK3β detachment from the destruction complex. Constitutive activation of WNT–β-catenin signalling in HCC occurs through one of three molecular events: gain-of-function somatic mutations in exon 3 of CTNNB1 leading to impairment in serine/threonine phosphorylation and ubiquitination; loss-of-function somatic mutations in APC; or loss-of-function somatic mutations in AXIN1. Activation of WNT–β-catenin signalling leads to key target gene expression, including GLUL (encoding glutamine synthetase (GS)), CCND1 (encoding cyclin D1), TBX3, AXIN2 and LGR5. For example, when β-catenin is mutated, GS, which is normally present in just one or two hepatocyte layers surrounding the central vein, is seen diffusely, albeit heterogeneously, throughout the tumour nodule, allowing GS staining to sensitively and specifically diagnose CTNNB1-mutated HCC^68–70.

Loss-of-function mutations in APC (1.2–5.5%) or AXIN1 (8–10%), genes encoding key components of the β-catenin degradation complex, augment β-catenin activity through stabilization and nuclear translocation^16,21,28. APC and AXIN1 facilitate the ability of GSK3β to phosphorylate and subsequently degrade β-catenin⁷¹. Despite their similar roles, loss-of-function mutations in either lead to distinct HCC biology. In mice, APC deletion alone (through CRISPR–Cas9 single-guide RNA (sgRNA)) does not promote tumorigenesis and, only in combination with MET, constitutively active mutant of TAZ (Taz^S89A) or YAP (Yap^S127A) induces β-catenin nuclear translocation and HCCs that are strongly GS-positive⁷². However, in mice, expression of Axin1-targeting sgRNA plus MET led to weak β-catenin activation and absence of tumoural GS staining compared with β-catenin and MET co-expression, despite the requirement of β-catenin to develop HCC⁷³. Similarly, patients with HCC with loss-of-function mutations in AXIN1 do not show upregulation of canonical WNT–β-catenin target genes, such as GLUL, but rather belong to a proliferation subclass with activation of YAP–NOTCH signalling^74,75. These differences are likely due to the degree of β-catenin activation and/or differential interactions of mutated β-catenin or stabilized β-catenin due to varying mechanisms with specific transcription factors or histone acetyl transferases. Future studies in preclinical models will help to establish the functional relevance of β-catenin activation in these disparate scenarios, including any differential susceptibility to ICI-based therapy due to levels of immune exclusion that might in turn be dependent on the extent of β-catenin activation. In summary, activation of the WNT–β-catenin pathway in HCC could be due to multiple genetic events, which can lead to heterogeneity in the magnitude of β-catenin activation or differential protein–protein interactions, with an eventual disparity in target gene expression.

Genotype–phenotype correlation due to CTNNB1 mutations

Using human specimens, many groups have used WES to profile the HCC genomic landscape. Mutations in CTNNB1 in HCC often coexist with mutations in the TERT promoter, NFE2L2, ARID1A, ARID2, CDKN2A or APOB^12,13,76. For example, CTNNB1^T41A can occur with NFE2L2 mutations or NRF2 activation in general⁷⁶, which was recapitulated in mice by co-expression of Ctnnb1^T41A and Nfe2l2^G31A, which led to well-differentiated HCC in 10–12 weeks⁷⁷. The spectrum of CTNNB1 mutations is broad, with one study identifying 14 of 75 tumours (19%) as CTNNB1-mutated, with six tumours mutated at S37, T41 or S45, six mutated at D32 or G34 and two showing deletions affecting exon 3⁷⁸. In a previous study, 9 of 32 HCC tumours (28%) were CTNNB1-mutated, 8 of 9 at D32/S37 or tabl). Various mutants of CTNNB1 have been shown to differ in relative β-catenin activity. Rebouissou et al. demonstrated the existence of three subgroups: group 1 were ‘high’ activators (large exon 3 deletion, D32/S37), group 2 were ‘moderate’ activators (T41) and group 3 were ‘weak’ activators (S45, K335 and N387), although S45 mutants had gene duplication and overall showed high activity⁷⁶.

Liver tumours with moderate or high β-catenin activation tended to be HCC, while liver tumours with weak activity tended to be hepatocellular adenomas⁷⁶. In mice, this was addressed by co-expressing human MET with either CTNNB1^S33Y, which led to the development of well-differentiated HCC by 9–9.5 weeks, or with CTNNB1^S45Y, which led to the development of HCC by 6–7 weeks⁷⁹. Accelerated tumorigenesis was also evident when KRAS^G12D was the second common hit, and led to well-differentiated HCC by 8 weeks when co-expressed with CTNNB1^S33Y compared with 7 weeks when co-expressed with CTNNB1^S45Y (ref. 80). Structural biology-based studies might help us to understand the basis of underlying differences in activity of the various β-catenin mutants.

In vitro and in vivo models of CTNNB1-mutated HCC

Several models that display activation of the WNT–β-catenin pathway and that have helped in understanding tumour biology and therapeutic targeting in this molecular subtype of HCC are available (Table 2).

Table 2 |.

Comparison of in vitro and in vivo models to study β-catenin-mutated HCC

Model	Mechanism of β-catenin activation	Correlation with human HCC	Ability to study effects of tumour microenvironment	Strengths	Weaknesses	Correlation with drug response
In vitro
2D human cell lines	HepG2, SNU-398 and Huh6 have mutations in β-catenin	Correlates most with Hoshida S3 subclass	2D and 3D culture systems are limited by lack of stromal (e.g. endothelial and hepatic stellate cells) and immune (e.g. macrophages and lymphocytes) cells; can co-culture with different cell types 2D and 3D culture systems do not account for background liver injury where HCC develops in background of chronic liver disease	Commercially available, efficient and easy to culture; cost-effective; can be used to study human tumour biology, and for large-scale drug and CRISPR screens	Not truly representative of non-proliferative subclass of tumours; not ideal for biomarker identification	Various tyrosine kinases, mTOR inhibitors and anti-β-catenin drugs have been tested in these systems, yet effects limited to study direct effects on cell growth and metabolism
3D patient-derived organoids	Patient-derived organoids have known mutations in CTNNB1, AXIN1 or APC from parental tumour	High expression of WNT-β-catenin targets, enrichment for Chiang CTNNB1 and Hoshida S3 subclass		Retains in vivo tumour histology; reflects genetic diversity observed in human HCC; derived from patients with various aetiologies of HCC; reflects true intratumoural heterogeneity	Requires significant expertise; low yield; not commercially available; expensive due to materials cost
In vivo
DEN-PB mouse model (C3H/He strain only)	β-Catenin mutations in >80% of mouse tumours in codons 32, 33, 37 and 41	Mouse tumours show biological characteristics of β-catenin-mutated HCC, including being GS-positive, and expression of cytochrome P450 superfamily enzymes	Can study effects on mouse stromal and immune cells	Somewhat spontaneous development of tumours; intact β-catenin mutations with ability to study tumour microenvironment	Transcriptomic characterization comparisons to human HCC have not been performed; requires use of chemical carcinogens; these chemical carcinogens are not reflected in aetiology of human HCC; HCC induction is protracted (up to 1 year) and has low penetrance	Various anti-β-catenin drugs have been tested in these systems with effects mostly focused on cell growth and metabolism
HDTVI with sleeping beauty transposon-transposase	Plasmids are introduced to overexpress human forms of mutated β-catenin, or engineered sgRNAs with Axin1-targeting or Apc-targeting sgRNAs	Molecular characterization has shown 70–80% similarity to human HCC with similar perturbations		Efficient (tumour develops in 6–8 weeks after HDTVI); cost-effective; reproducible	HCC develops in background of normal liver without liver injury; mouse and human stromal and immune cells may differ	Various tyrosine kinases, mTOR inhibitors and anti-β-catenin drugs have been tested in these systems to study effects on cell growth, metabolism and immune microenvironment; most drug testing in preclinical setting uses this model
Patient-derived xenograft	Tumours are derived from patient tumours with known mutations in CTNNB1, AXIN1 or APC	Patient tumours reflect parental histology once explanted and retain intrinsic molecular biology	Requires immunodeficient mouse strains, so cannot study effects on mouse stromal and immune cells Humanized mice, however, allow study of tumour microenvironment and in background of chronic liver injury (e.g. Western diet or alcohol liver injury)	Can be option for drug screening in vivo due to retention of same genetic alterations as primary tumour With liver damage, the effects on tumour microenvironment more closely recapitulate human HCC and disease progression	Patient-derived xenografts in immunodeficient mouse strains may be heterotopic or orthotopic models with each reflecting distinct limitations Humanized models are costly and time-consuming, and require expertise and not commercially available	Various tyrosine kinase inhibitors and anti-β-catenin drugs have been tested in these systems with effects mostly focused on cell growth and metabolism

DEN, diethylnitrosamine; GS, glutamine synthetase; HCC, hepatocellular carcinoma; HDTVI, hydrodynamic tail vein injection; PB, phenobarbital; sgRNA, single-guide RNA.

In vitro systems

Many 2D in vitro cell lines are available to study HCC of different Hoshida molecular subclass (S1 (SNU-387, SNU-475 and SNU-449), S2 (Huh7, Hep3B and HepG2) and S3-like (HepG2, SNU-398 and Huh6)) based on various point mutations in CTNNB1 (SNU-398 (CTNNB1^S37C) and Huh6 (CTNNB1^G34V)), CTNNB1 exon 3 deletions (HepG2) or AXIN1 (Hep3B, SNU-475 and SNU-449)⁸¹. In an analysis of 25 human hepatoma cell lines, S1 and S2 subtypes were largely represented, with only one cell line (SNU-398) having a mutation in exon 3 of CTNNB1, suggesting that the S3 subclass is not well recapitulated in vitro⁸¹. Interestingly, other studies have also shown that the frequency of CTNNB1 mutations in HCC cell lines is lower than in primary HCC^82,83. The available HCC cell lines with CTNNB1 mutations tend to represent the proliferative subclass, rather than the non-proliferative subclass to which the majority of this HCC subset belongs⁸². The HepG2 cell line is widely used to study WNT–β-catenin-active HCC due to deletion of exon 3 in CTNNB1 and high GLUL expression, yet there is increased expression of S2 genes and a more hepatoblastoma-like phenotype⁸¹. There are no truly representative in vitro cell lines to study the non-proliferative class of tumours, encompassing Hoshida S3 and Boyault G5–G6 tumours, which are predominantly CTNNB1-mutated. It is thought that in vitro modelling of the S3 subtype might require a microphysiological system that supports pericentral hepatocyte growth, as current in vitro conditions probably select for clones with greater proliferative capacity^81,84,85. As a result, new patient-derived HCC cell lines with confirmed CTNNB1 mutations have been developed to better recapitulate this subclass, including seven additional primary HCC patient-derived lines with different CTNNB1 point mutations⁸⁶. Qiu and colleagues developed the Liver Cancer Model Repository (LIMORE), which profiled genomic and transcriptomic data and drug synergistic responses from 50 new cell lines derived from patients with HCC and 31 commercially available lines, including 5 newly profiled CTNNB1-mutated cell lines⁸³. Alternatively, using CRISPR–Cas9, WNT–β-catenin activating mutations can be introduced into human hepatocyte cell lines⁸⁷, HCC mouse or human cell lines without WNT–β-catenin activity (for example, Hep3B)⁸⁸, or induced pluripotent stem cells (iPSCs). As an example, Akasu et al. engineered human (Huh7) and mouse (3H3) HCC cell lines with exon 3 skipping of CTNNB1, which resulted in a 40-fold increase in WNT–β-catenin target gene expression⁸⁹.

Others have pioneered the use of 3D spheroid or organoid cultures using hepatoma cell lines (for example, HepG2 and Huh7)^90–92 or primary liver organoids or iPSCs transfected with specific oncogenes^93,94. Primary patient-derived organoids (PDOs) and 3D tumour spheroids from patient tumour tissue retain the in vivo tumour histology and parental genetic profile compared with 2D cultures. Two primary CTNNB1-mutated HCC PDOs were used to test the β-catenin inhibitor WNTinib⁸⁶. Broutier et al. described derivation of a CTNNB1-mutated HCC PDO from resected explanted tumour tissue after liver transplantation⁹⁵. Interestingly, this HCC PDO had high expression of WNT–β-catenin target genes, enrichment for Chiang CTNNB1 and Hoshida S3 signatures, and retained the pseudoglandular rosette histology of the parental tumour⁹⁵. PDOs have also been developed from needle biopsies⁹⁶. Of 38 HCC biopsies, 10 (26%) led to stable organoids. These were all poorly differentiated tumours, probably due to mutation in CTNNB1, with concomitant mutations in ARID1A and TSC2 (ref. 96). Ji et al. described the generation of a biobank of 65 liver cancer organoids from 57 patients⁹⁷, of which only one PDO had a confirmed CTNNB1 mutation⁹⁷. PDOs, like any other model, have limitations, as they do not recapitulate the tumour heterogeneity and microenvironment that is observed in vivo, lacking immune elements (for example, macrophages, T cells and B cells) and stromal elements (for example, endothelial and hepatic stellate cells)^97,98. Integrating 3D liver tumour organoids with stromal elements through either co-culture or liver-on-chipm⁹⁹ systems may better recapitulate the in vivo architecture.

In vivo systems

Over the past two decades, many animal models of HCC have been described to be β-catenin-active. Here, we focus on preclinical models that display either gain-of-function mutations in CTNNB1 or loss-of-function mutations in APC and AXIN1. Various groups have utilized the following types of models to study WNT–β-catenin activity in HCC: chemical carcinogen-driven or diet-driven models; subcutaneous or orthotopic injection of human hepatoma cell lines, patient-derived xenografts (PDX) or PDO xenografts (PDOX); transgenic mouse lines or genetically engineered mouse models (GEMMs); or humanized HCC mouse models.

Diethylnitrosamine (DEN) was the first described chemical carcinogen that initiates HCC in mice. DEN administration to B6C3F1 mice led to HCC with nuclear β-catenin staining in >50% of hepatocytes and mutations in 7 of 13 tumours in codon 41 of CTNNB1 after 1.5 years¹⁰⁰. Addition of the tumour promoter phenobarbital in this model led to HCC, with mutations in β-catenin in >80% of tumours in codons 32, 33, 37 and 41 (ref. 101). Without phenobarbital, DEN alone tends to induce HCC with Hras mutations¹⁰¹. Additionally, >90% of these tumours are GS-positive and have increased expression of cytochrome P450 (CYP) superfamily enzymes^102,103. A second study replicating the first demonstrated these tumours to be sensitive to β-catenin inhibition⁸⁸. This phenotype is strain-specific, as C57BL/6 mice develop HCC without Ctnnb1 mutations with either DEN or DEN with phenobarbital¹⁰⁴. As a result, the liver-specific Ctnnb1-knockout mice with the C57BL/6 background showed no decrease in development of liver tumours in response to DEN or DEN with phenobarbital. They displayed exacerbated hepatocarcinogenesis due to increased fibrosis, injury and inflammation, supporting the relevance of β-catenin suppression only in HCCs driven by active β-catenin^104,105. Likewise, Ctnnb1-knockout C3H/He mice showed exacerbated carcinogenesis following DEN exposure and reduced cancer after DEN plus phenobarbital treatment¹⁰⁶. With increased incidence of MASLD and MASH-related HCC, there is an interest in developing relevant preclinical models. In a study reported in a non-peer-reviewed preprint, C57BL/6 mice fed a high-fat diet (HFD), with 60% of its calories from fats, demonstrated upregulation of WNT–β-catenin targets at 15 months, although the mutational status of CTNNB1 was not confirmed¹⁰⁷. An HFD increases β-catenin expression in the liver, and CTNNB1 knockdown protects against hepatic steatosis and diet-induced obesity^108–110. Additionally, HFD plus DEN led to early hepatocarcinogenesis with upregulation of WNT–β-catenin targets in C57BL/6 mice^108,111. Thus, both chemical carcinogen-driven and diet-driven HCC models can be useful to study β-catenin activation in HCC, even though these models can be expensive, require extended study times and can have lower penetrance than other models.

PDX or PDOX models are also widely used to study HCC. PDX and PDOX models can be generated more quickly but require immunodeficient mouse strains to prevent immune rejection — with the exception of syngeneic models using mouse HCC cells such as Hepa1–6 with CTNNB1^S45P mutations¹⁷. Various immunodeficient strains have been used, such as NOD-SCID, Rag1/2-knockout and NSG mice¹¹². HepG2, Hep3B, Huh7 and Huh6 have been used to generate models to study responses to anti-β-catenin agents^113,114. Primary HCC and organoid lines have also been used to develop PDX and PDOX models. Gu et al. derived 65 HCC PDXs from 254 Chinese patients and identified 8 (12.3%) with mutations in CTNNB1, 8 in AXIN1 (12.3%) and 1 (2%) in APC¹¹⁵. Interestingly, a PDX with an AXIN1 mutation (LI-03–0020) showed partial response to sorafenib and complete response to lenvatinib¹¹⁵.

CTNNB1 mutations alone are insufficient to initiate HCC in mice¹¹⁶. Hepatic overexpression of wild-type or mutant β-catenin through Cre-recombinase, sleeping beauty transposon system hydrodynamic tail veil injection (SB-HDTVI) or adenovirus, were all unable to stimulate HCC development by 1 year. To model this unique molecular subclass of HCC in mice, we and others have utilized the SB-HDTVI system to deliver CTNNB1 in combination with another oncogene, or with a tumour suppressor gene via CRISPR–Cas9 with sgRNAs, in a subset of hepatocytes^73,79,117. This technique is efficient, cost-effective and reproducible. Candidate genes are selected through interrogation of genomic data sets, such as The Cancer Genome Atlas. Once integrated into the genome of 1–5% of hepatocytes, well-differentiated HCC develops through clonal expansion within 6–12 weeks of HDTVI. We and others have demonstrated synergy in HCC development between mutant CTNNB1 and mutant NFE2L2 (ref. 77), KRAS⁸⁰, NRAS¹¹⁸, MET⁷⁹, MYC^56,119, AKT¹¹⁷ and TERT¹¹⁹. Additionally, loss of Axin1 and Apc demonstrate cooperation with MET^72,73. Lastly, challenging these mice with various forms of liver injury or environmental exposures might help to further recapitulate clinical HCC development in this otherwise reductionist model¹²⁰.

Lastly, humanized HCC mouse models comprise humanized liver parenchymal, non-parenchymal and immune cell types. This model utilizes Fah^−/−;Rag2^−/−;Il2rgc^−/− (FRG) immunodeficient mice, which allow engraftment of donor human hepatocytes and hematopoietic progenitor cells. Zhu et al. developed a protocol to generate a humanized HCC mouse model through injection of constitutively activated CTNNB1 (mutated at residues S33, S37, T41 and S45) and subjected these mice to alcohol and/or a Western diet to induce HCC development¹²¹. These models are well suited to study the TIME, test novel ICI combinations and understand the mechanisms involved in the ICI response¹²². However, these models are costly and time-intensive, and require significant expertise. Next-generation versions of humanized HCC models will ultimately incorporate patient tumours to generate patient ‘biomimetic twins’ for testing of targeted therapies to gain insights into therapeutic efficacy before systemic administration.

Unique tumour cell-intrinsic metabolism

There is a decreasing gradient of WNT2 and WNT9B expression in endothelial cells from zone 3 to zone 1 in a liver lobule. Due to its paracrine effect, β-catenin is most active in hepatocytes closest to the central vein^123,124. β-Catenin-mutated HCCs share a metabolic identity linked to pericentral or zone 3 metabolism¹²⁵.

β-Catenin-mutated HCC tumours in preclinical models and in individuals with HCC are simultaneously positive for GS and mTOR phosphorylated at Ser2448, a downstream effector of mTORC1 signalling. Glul, which encodes GS, is a transcriptional target gene of the WNT–β-catenin pathway in the liver and is responsible for synthesis of glutamine from ammonia and glutamate, leading to higher intracellular glutamine content¹²⁶. High intracellular glutamine has been shown to directly phosphorylate mTOR at Ser2448 in the lysosome¹²⁷. As a result, preclinical models of β-catenin-mutated HCC are sensitive to the mTOR inhibitors rapamycin, everolimus, RMC-6272 (representative of the investigational drug RMC-5552) and MLN0128 (sapanisertib), with their efficacy potentiated by addition of a MET inhibitor^77,123,128. This is relevant as there is potential to repurpose mTOR inhibitors in biomarker-driven trials for the treatment of β-catenin-active HCC. This subset of HCCs is independent of an additional 5–10% of HCCs that show mTOR activation due to mutations in genes encoding TSC1/2, which are associated with the tuberous sclerosis complex and negative regulators of mTOR signalling¹²⁹. Indeed, these studies have now led to an investigator-initiated clinical trial for use of sapanisertib with or without cabozantinib in the second-line setting in CTNNB1-mutated HCC (NCT06811116) as part of the Experimental Therapeutics Clinical Trial Network at the Cancer Therapy Evaluation Program (CTEP), National Cancer Institute. Additionally, GS has been shown to induce autophagic flux, which mediates sorafenib sensitivity in β-catenin-mutated HCC¹³⁰. β-Catenin-mutated HCCs are relatively sensitive to sorafenib, which explains in part the fewer treatment effects observed in patients with CTNNB1 mutations in the IMbrave150 trial¹³¹. The neurotransmitter γ-aminobutyric acid (GABA) has also been shown to directly bind and stabilize β-catenin, leading to increased metastasis and upregulation of SLC6A12 to allow more GABA re-entry into HCC cells¹²⁶. Moreover, xenobiotic metabolism, another hallmark of pericentral hepatocytes¹³² and of β-catenin-mutated HCCs, due to transcription control of expression of the glutathione (GSH) S-transferase Mu (GSTM) family of genes (that is, GSTM2 and GSTM3), makes these tumours susceptible to GST inhibitors such as ethacrynic acid in preclinical models, as discussed in a non-peer-reviewed preprint¹³³. A β-catenin-mutated preclinical HCC model was also shown to be sensitive to acetaminophen due to higher expression of Cyp2e1 and Cyp1a2, both zone 3 target genes, which facilitate formation of a toxic intermediate that leads to oncotic death of tumour cells¹³⁴. Aside from targeting the GS–mTORC1 axis, which is now being clinically evaluated, targeting xenobiotic metabolism, either through GSTM or the CYP family of enzymes, in CTNNB1-mutated HCC has not yet been translated to the clinic.

Lastly, despite the prevailing dogma that fatty acid β-oxidation occurs in periportal zone 1 hepatocytes, β-catenin-mutated HCCs are uniquely fatty acid oxidation (FAO)-addicted instead of being glycolytic, which is typically observed in zone 3 hepatocytes¹³⁵. As a result, these tumours utilize fatty acids as energy substrates and are infrequently steatotic¹³⁵. β-Catenin-active tumours are not sensitive to fatty acid synthase inhibition with TVB3664 (ref. 136). Instead, as shown in preclinical models, FAO could be targeted through Ppara ablation or treatment with etomoxir¹³⁶. Intriguingly, PPARα inhibition via TPST-1120 combined with an ICI is in clinical development for the treatment of patients with HCC (NCT03829436 and NCT06680258). APC loss-of-function mutations recapitulate similar tumour biology, but not tumours induced by loss-of-function mutations in AXIN1, underscoring divergent biology¹³⁵. It should be noted that the unique zone 3 glycolytic phenotype in β-catenin-mutant HCC cells leads not only to susceptibility to glycolysis inhibition but also to a switch from predominantly aerobic to anaerobic metabolism under ischaemic conditions¹³⁷. This was posited to be the mechanism underlying WNT–β-catenin-active tumours showing an improved response to transarterial chemoembolization¹³⁷. Overall, β-catenin-mutated HCCs exhibit a metabolic phenotype similar to that of hepatocytes residing in zone 3 of the liver lobule including glutamine metabolism, xenobiotic metabolism and FAO¹⁰⁷.

Role of WNT target genes in CTNNB1-mutant HCC

We and others have recently described the development of a ten-gene mutated-β-catenin gene signature (MBGS), which included AXIN2, GLUL, LGR5, NKD1, NOTUM, RHBG, SLC13A3, SP5, TCF7 and TNFRSF19, and identified patients with CTNNB1 mutations in HCC with receiver operating characteristic area under the curve of 0.90 to 0.94 in large patient cohorts¹³⁸. Other direct WNT–β-catenin targets previously considered to influence tumorigenesis include CCND1, TBX3, MYC and LECT2. The majority of these genes have fundamental roles in driving β-catenin-mutated HCC tumorigenesis (Fig. 1). For example, inhibition of SLC13A3, a known GSH transporter, via genetic and pharmacological approaches depletes GSH, inactivates mTOR signalling, induces ferroptosis and decreases tumour burden in β-catenin-mutated HCC preclinical models¹³⁹. Under low autophagic conditions, cyclin D1 (encoded by CCND1) functions as a tumour oncogene through mediating cell cycle dysregulation, cell proliferation and HCC development¹⁴⁰; however, cyclin D1 is not essential for β-catenin-mutated tumour development, as cyclin D2 can functionally compensate in its absence¹⁴¹.

Fig. 1 |. Multiple roles of β-catenin in hepatocellular carcinoma.

WNT–β-catenin target genes involved in each facet of tumour biology are shown. ECM, extracellular matrix; TF, transcription factor.

Not all downstream WNT–β-catenin targets directly facilitate tumorigenesis. Some targets can have paradoxical and unexpected roles and even serve as ‘tumour suppressors’ in the context of β-catenin-active HCCs. Glul knockdown paradoxically aggravates tumorigenesis by inducing glutamate-derived non-essential amino acid production and mTORC1 activation¹⁴². Unpublished studies from our laboratory have demonstrated a unique tumour cell–macrophage–T cell cross-talk as another mechanism of tumour promotion upon Glul loss in β-catenin-mutated HCC cells (S.P.M., unpublished observations). Ablation of Tbx3, a transcriptional repressor, accelerates tumorigenesis through activation of YAP–TAZ signalling via phospholipase D1 (ref. 143). NOTUM, a secreted WNT inhibitor and target, might influence metabolic activity when β-catenin is active¹⁴⁴. It is likely that the non-proliferative and less aggressive nature of β-catenin-mutated HCC might be driven in part by expression of metabolic genes and in part by upregulation of WNT-driven tumour suppressors.

β-Catenin modulation of the tumour immune microenvironment

The HCC TIME is heterogeneous, with five main subtypes: immune-active (inflamed; 11–15%), immune-exhausted (inflamed; 7–8%), immune-like (inflamed; 15%), immune-intermediate (non-inflamed; 43%) and immune-excluded (non-inflamed; 20%)^52,53. Montironi et al. described WNT–β-catenin-active tumours as both immune-like (~30%) and immune-excluded (~70%)⁵² (Fig. 2). The immune-like subtype was associated with Hoshida S2 and Chiang CTNNB1 classes, and the immune-exhausted subtype was associated with the Hoshida S1 or the WNT–TGFβ class. The immune-like tumours had high T cell infiltration, antigen presentation and expression of IFN signatures associated with antitumour immune response. However, comparing the immune-like and immune-excluded genomic landscape, there were no differences in the proportion of CTNNB1 mutations conferring moderate or high activity^52,76. Aoki et al. also described non-inflamed (n = 33, 82.5%) and inflamed (n = 7, 17.5%) subtypes of WNT–β-catenin-mutant HCC¹⁴⁵. The non-inflamed subtype tended to be of viral aetiology, with high expression of AXIN2 and GLUL with low T cell infiltration, while the inflamed subtype tended to be alcohol-related HCC but with low GS, and increased T cell, myeloid-derived suppressor cell and regulatory T cell infiltration. It is likely that in the latter study by Aoki et al., the immune-exhausted WNT–TGFβ class with M2 macrophage infiltration as described by Montironi et al. was captured^52,145. Additionally, WNT activation via AXIN1 and APC mutations were observed in 3 of 7 (42.9%) of the inflamed tumours, compared with 3 of 33 (9.1%) of the non-inflamed tumours (P = 0.01)¹⁴⁵. Overall, these studies provide a more nuanced appreciation of the immune phenotypes of β-catenin-active, CTNNB1-mutant HCC, and how subsets of these tumours might be susceptible to ICI treatment, requiring deeper mechanistic evaluation and biomarker discovery.

Fig. 2 |. Theranostic work-up to assess CTNNB1-mutated HCC for precision therapy.

Tissue or liquid biopsy can detect WNT–β-catenin active hepatocellular carcinoma (HCC) to guide therapeutic management based on emerging precision medicine-based clinical trials in HCC. cfDNA, cell-free DNA; ctDNA, circulating tumour DNA; H&E, haematoxylin and eosin; ICI, immune checkpoint inhibitor; TKI, tyrosine kinase inhibitor.

Preliminary preclinical and clinical studies have identified various secreted proteins that might modulate the TIME. Immune evasion was linked to β-catenin-mediated inhibition of NKG2D ligand expression through TCF4 binding to the promoter of ULBP1/2 genes¹⁴⁶. Others have observed no difference in NKG2D ligand expression between inflamed and non-inflamed classes⁵². Activation of TNF receptor superfamily member 19 (TNFRSF19), which was also a marker in our MBGS panel, was shown to reduce the expression of senescence-associated secretory phenotype-like cytokines (for example, IL-6, IL-8 and CXCL8)¹⁷. Absence of LECT2, another WNT–β-catenin target, has been shown to alter the TIME through inducing infiltration of immunosuppressive monocytes¹⁴⁷. Lastly, matrix metalloproteinase 9 (MMP9) has been shown to be upregulated in CTNNB1-mutated HCC, which leads to diminished cytotoxic functionality of CD8⁺ T cells, with MMP9 inhibition shown to synergize with ICIs in a preclinical model¹⁴⁸. Past studies have shown that β-catenin binds directly to inflammation-regulating factors such as p65 and increased β-catenin levels inhibit their activity¹⁴⁹. More recent studies have shown that IRF2 and POU2F1 activity are suppressed in β-catenin-mutated HCC via undetermined mechanisms⁴⁴. Whether these same immune-regulatory mechanisms observed in β-catenin-mutated HCC preclinical models are also relevant in patient tumours remains to be examined. Such studies might allow therapeutic exploitation via delivery of exogenous immune excitatory factors that activate transcription factors, such as p65, IRF2 and POU2F1.

Biomarkers to identify patients with CTNNB1-mutated HCC

As precision therapy is gradually integrated into clinical practice, patient selection will be important for HCC management. As CTNNB1-mutated and WNT–β-catenin-active HCCs in general exhibit distinct tumour biology and responses to existing standard of care treatments, it is important to review the latest advances in identifying this unique subset of HCCs.

Imaging and radiomics

Biannual surveillance using ultrasound is recommended in high-risk individuals, with CT–MRI reserved for confirming suspicious lesions.

On MRI, peritumoural enhancement is associated with CTNNB1-mutated HCC, with 47.0% and 84.6% sensitivity and specificity, respectively¹⁵⁰. WNT–β-catenin-active HCCs also show increased uptake of ¹⁸F-fluorocholine, but not ¹⁸F-fluorodeoxyglucose, on PET imaging, and ¹⁸F-fluorocholine PET is currently being evaluated in a clinical trial for predicting response to ICI therapy (NCT04965454)¹⁵¹. Gadoxetic acid-enhanced MRI (EOB-MRI) has 78.9% and 81.7% sensitivity and specificity, respectively, in predicting Chiang CTNNB1 and Hoshida S3 HCC^152,153. The organic anion transporting polypeptides 1B1 (OATP1B1) and OATP1B3 are the transporters for gadoxetic acid, and their expression is increased with active WNT–β-catenin signalling¹⁵². Also, hyperintensity in the hepatobiliary phase of EOB-MRI is associated with an immune-excluded TIME infiltrated with tumour-associated macrophages, but there are conflicting reports on the correlation with response to ICI therapy^153–155. In a single-centre study, using a combination of Liver Imaging Reporting and Data System (LI-RADS) descriptors and extracted radiomic features to develop a machine-learning algorithm, Arefan and colleagues were able to predict β-catenin-mutated HCC with high accuracy (area under the curve 0.88; P = 0.009)¹⁵⁶. AI-based radiomics might have an important role in the stratification of patients with HCC in the future.

Tissue biopsy

Tissue biopsy in HCC management has not been widely implemented, but there are specific circumstances necessitating biopsy⁴³. Classifying tumours into histomolecular subtypes has prognostic utility and is useful for guiding targeted therapy selection in the era of precision medicine. Despite immunohistochemistry (IHC) for GS being the current gold standard to identify CTNNB1-mutated HCC, it is not useful for treatment selection as there are currently no anti-β-catenin therapies available. However, even if implemented, GS IHC can have a high false-positive rate^68,70. While it can be used in retrospective studies as a method to identify patients belonging to WNT–β-catenin subclasses associated with ICI response⁵⁸, a more sensitive tissue-based biomarker method such as RNA probe-based detection platforms might have higher sensitivity and specificity⁴³. Indeed, our ten-gene MBGS panel demonstrated high accuracy in detecting CTNNB1-mutated HCC¹³⁸. Tissue biopsy is also relevant in defining the pretreatment TIME, utilizing various T cell, macrophage, dendritic cell and B cell marker multiplex panels, with patients with CTNNB1 mutations showing mostly immune exclusion. Lastly, AI-based algorithms have promise for predicting molecular alterations from histology images with high accuracy, sensitivity and specificity. For example, Chen et al. described prediction of CTNNB1 mutation status from whole-slide images of H&E-stained HCC tumour tissue using a neural network model, and found 91%, 89.5% and 81% accuracy, precision and recall, respectively¹⁵⁷. Future iterations of these algorithms are expected to integrate spatial transcriptomic data and predict intratumoural pathway-level changes to guide synergistic drug combinations.

Liquid biopsy

There has been much development in the use of liquid biopsy to detect HCC. Compared with tissue biopsy, liquid biopsy provides the opportunity for serial sample assessment during ICI treatment progression⁶⁵. This platform also allows the evaluation of tumours non-invasively, of clonal evolution of tumours after treatment, of intertumoural and intratumoural heterogeneity, and for the presence of minimal residual disease¹⁵⁸. AFP is the most widely utilized serum biomarker for diagnosing HCC, but its interpretation is limited owing to various confounders and its low expression in WNT–β-catenin-active HCC^24–26,43. There are now various commercial liquid biopsy platforms utilizing a combination of markers, demonstrating variable levels of accuracy⁴³. Also, various novel liquid biopsy analytes are being tested for clinical utility, including circulating tumour cells, circulating tumour DNA (ctDNA), cfDNA, cell-free RNA and extracellular vesicles. Secreted factors including NOTUM and TNFRSF19 from the MBGS panel might diagnose CTNNB1-mutated HCC and require direct investigation¹³⁸.

Various studies involving patients with HCC have identified mutations in CTNNB1, AXIN1 and APC in cfDNA or ctDNA in blood or urine^159–161, with high concordance in tissue–blood mutational profiling¹⁵⁹. A combination of both liquid and tissue mutational profiling increases the sensitivity in detecting CTNNB1 mutations in HCC^159,162. Liquid biopsy profiling for WNT–β-catenin-activating mutations might also inform treatment prognosis. Ogawa et al. profiled cfDNA from 33 patients with HCC treated with atezolizumab plus bevacizumab, and identified eight patients (24.2%) with CTNNB1 mutations⁶⁴. Matsumae et al. profiled cfDNA from 85 patients with HCC treated with atezolizumab plus bevacizumab, and identified 13 patients (15%) with CTNNB1 mutations⁶². Interestingly, both studies found no differences in OS or PFS between patients with wild-type and patients with mutated CTNNB1 profiled by cfDNA⁶⁴. However, both retrospective studies lacked a comparator treatment arm and compared only patients with wild-type and patients with mutated CTNNB1. As more anti-β-catenin therapies become available, tissue and/or liquid biopsy might be a prerequisite for enrolment in randomized biomarker-driven clinical trials.

Single-cell and spatial transcriptomics to study CTNNB1-mutated HCC

Single-cell RNA sequencing (scRNA-seq) and single-nucleus RNA sequencing (snRNA-seq) have revolutionized the study of HCC, leading to the discovery of novel cell types and allowing investigation of tumour heterogeneity. Coupling these techniques with cellular spatial coordinates has also permitted localization of cell types along the hepatic lobule or in peritumoural or intratumoural regions (Fig. 3). A snRNA-seq study revealed intratumoural heterogeneity along the portal–central axis, and zonal metabolic phenotypes of different β-catenin-mutated HCC populations, in a novel β-catenin-mutated preclinical model¹³³. Intriguingly, an ectopic population of CTNNB1-mutated tumour cells was found in zone 1 and zone 2. Further analysis might reveal whether this population is metabolically similar to zone 3 CTNNB1-mutated tumour cells and whether the metabolic phenotype is acquired based on spatial origin or from the tumour driver. In addition to studying the heterogeneity of tumoural subpopulations, scRNA-seq performed on CD45⁺ cells from a β-catenin-mutated HCC model revealed low CD8⁺CXCR3⁺ effector T cells and high CD8⁺PDCD1⁺ exhausted T cells¹⁴⁸. A scRNA-seq study comparing two preclinical models with differential responses to ICIs concluded that progenitor exhausted CD8⁺ T cells were enriched in non-responder animals¹⁶³. These preclinical findings were translatable to the clinic, as scRNA-seq profiling of peripheral blood mononuclear cells (PBMCs) from ten patients before and after treatment with pembrolizumab identified an enrichment of CTNNB1 mutations in non-responders, with these patients having fewer cytotoxic CD8⁺ T cells than responders⁶⁰. scRNA-seq profiles of PBMCs from patients with HCC before and after treatment with an ICI showed that the CXCR3⁺CD8⁺ effector memory T cells mediate the response¹⁶⁴. Another study further clarified that tumoural PDL1⁺CXCL10⁺ macrophages secreting CXCL9, CXCL10 and CXCL11 attract peripheral CXCR3⁺CD8⁺ effector memory T cells and mediate durable responses to atezolizumab plus bevacizumab¹⁶⁵. Thus, scRNA-seq of tumoural and peripheral immune compartments allows a deeper understanding of various cell populations associated with response to ICIs.

Fig. 3 |. Single-cell and spatial transcriptomic workflows to study CTNNB1-mutated HCC.

Single-cell and spatial transcriptomic workflows, and how they can be integrated to reveal spatial β-catenin cell-to-cell signalling, and unique tumour-intrinsic and tumour-extrinsic biology. FFPE, formalin-fixed paraffin-embedded; HCC, hepatocellular carcinoma; TIME, tumour immune microenvironment.

Spatial transcriptomics might also inform clinical outcomes based on spatial patterns of tumour, stroma and immune cellular networks. Cancer-associated fibroblast populations that form a physical barrier limiting immune infiltration and impairing ICI response have been found using this modality^166,167, and it has also been used to map bulk transcriptomic signatures to different spatial cell populations. Lu et al. mapped the spatial localization of Hoshida subclass S1–S3 signatures, and found that the S3 signature maps to tumour parenchyma, whereas the S1 signature maps to tumour endothelium, showing higher vascular density in the S1 subclass of HCC¹⁶⁸. Similarly, we and others have shown that the previously described immune ‘hot’ gene signatures differ in expression within the intratumoural, peritumoural and adjacent normal compartments in tumours with high MBGS expression¹³⁸. Similarly, expression of various chemokines has been mapped spatially in HCC: CCL15 in the tumour core recruits M2-like macrophages, while CCL19 and CCL21 in an immune-infiltrated region recruit CD3⁺ T cells to the TIME¹⁶⁹. Ultimately, these chemokines might be upregulated in the tissue, but their spatial location dictates function. Thus, spatial transcriptomics demonstrates the ability to characterize transcriptional differences within cellular ecosystems and the ability to understand tumour–immune and tumour–stroma interactions. However, it is important to note the nuances of using bulk RNA-seq-derived signatures on scRNA-seq or spatial transcriptomics, as the use of bulk RNA does not allow differences in cell composition of tissues used to derive transcriptomic signatures to be taken into account.

Precision medicine pipeline: targeting oncogenic WNT–β-catenin signalling

Given the heterogeneous ICI responses in patients with CTNNB1 mutations, a precision medicine-based approach that not only inhibits oncogenic dependencies in CTNNB1-mutated cancers but simultaneously reshapes the TIME and augments the efficacy of current ICI regimens is required. Here, we discuss various approaches that are being trialled to target oncogenic β-catenin signalling for personalized HCC therapy. It should be noted that several inhibitors of the WNT–β-catenin pathway have been identified that act at various levels within the signalling cascade^170,171 (Fig. 4). Here, we discuss those with most relevance based on the previously discussed modes of pathway activation unique to HCCs. Various clinical trials targeting the WNT–β-catenin signalling pathway in cancer are summarized in Table 3.

Fig. 4 |. Therapies for CTNNB1-mutated HCC for precision therapy.

Components of the WNT–β-catenin pathway that are critical for hepatocellular carcinoma (HCC) biology along with various inhibitors that have demonstrated efficacy in preclinical studies, shown limited benefit in patients, or are in clinical trials. APC, adenomatous polyposis coli; CBP, CREB-binding protein; GSK3β, glycogen synthase kinase 3β; LEF, lymphoid enhancer factor; LNP, lipid nanoparticle; siRNA, small interfering RNA; TCA, tricarboxylic acid; TCF, T cell factor.

Table 3 |.

Clinical trials for precision medicine approaches to treat WNT–β-catenin-active tumours

Study title	NCT identifier	Phase	Agents utilized and mechanism of action	Status
DKN-01 Inhibition in Advanced Liver Cancer	NCT03645980	I/II	DKN-01 (monoclonal antibody targeting DKK1)	Unknown
A Study of TPST-1120 With Atezolizumab Plus Bevacizumab in Patients With Unresectable or Metastatic HCC Not Previously Treated With Systemic Therapy	NCT06680258	III	TPST-1120 (small-molecule antagonist inhibiting PPARα)	Active, not yet recruiting
A Study to Evaluate ALN-BCAT in Patients With Hepatocellular Carcinoma	NCT06600321	I	ALN-BCAT (siRNA encapsulated in LNP targeting CTNNB1)	Active, recruiting
Testing the Addition of an Anticance,r Drug, Sapanisertib, to the Usual Chemotherapy Treatment (Cabozantinib) in Metastatic Liver Cell Cancer With a Change in Genes for the Protein β-Catenin, The SAPHIRE Trial	NCT06811116	I/II	Sapanisertib (mTOR inhibitor) along with cabozantinib in second line for CTNNB1-mutated HCC	Active, recruiting
Study of E7386 in Participants With Selected Advanced Neoplasms	NCT03264664	I	E7386 (inhibitor of β-catenin and CBP)	Active, not recruiting
A Study of E7386 in Combination With Pembrolizumab in Previously Treated Participants With Selected Solid Tumors	NCT05091346	I/II	E7386 (inhibitor of β-catenin and CBP interaction) Pembrolizumab (anti-PD1 therapy)	Completed
FOG-001 in Locally Advanced or Metastatic Solid Tumors	NCT05919264	I/II	FOG-001 (β-catenin-TCF antagonist)	Active, recruiting
A Study of LGK974 in Patients With Malignancies Dependent on Wnt Ligands	NCT01351103	I	LGK974 (PORCN inhibitor)	Completed
Phase 1 Study to Evaluate the Safety, Tolerability and Pharmacokinetic Profile of XNW7201 in Subjects With Advanced Solid Tumors	NCT03901950	I	XNW7201 (PORCN inhibitor)	Completed
A Phase 1–2 of ST316 With Selected Advanced Unresectable and Metastatic Solid Tumors	NCT05848739	I	ST316 (antagonist of the interaction between β-catenin and its co-activator BCL-9)	Active, recruiting
A First-in-Human Safety Trial of MTX-463	NCT06401213	I	MTX-463 (antibody against WISP1)	Completed
CGX1321 in Subjects With Advanced Solid Tumors and CGX1321 With Pembrolizumab or Encorafenib + Cetuximab in Subjects With Advanced GI Tumors (Keynote 596)	NCT02675946	I	CGX1321 (selective inhibitor of PORCN)	Active, unknown status
Study to Evaluate the Safety and Tolerability of RXC004 in Advanced Malignancies	NCT03447470	I	RXC004 (small molecule inhibitor of PORCN)	Completed
Safety and Efficacy Study of PRI-724 in Subjects With Advanced Solid Tumors	NCT01302405	I	PRI-724 (a selective inhibitor of WNT-β-catenin-CBP signalling)	Terminated
A Dose Escalation Study of OMP-54F28 in Subjects With Solid Tumors	NCT01608867	I	OMP-54F28 (fusion protein with the extracellular part of FZD8 receptor bound to a human IgG1 Fc fragment which binds WNTs)	Completed
A Study of LY2090314 in Patients With Advanced or Metastatic Cancer	NCT01287520	I	LY2090314 (GSK3 inhibitor)	Completed

Clinical trials registered on ClinicalTrials.gov. CBP, CREB-binding protein; GSK3, glycogen synthase kinase 3; HCC, hepatocellular carcinoma; LNP, lipid nanoparticle; PORCN, porcupine; siRNA, small interfering RNA; TCF, T cell factor.

TKIs

There are various TKIs that have shown some efficacy in β-catenin-active HCCs. Sorafenib has shown small-to-modest responses in CTNNB1-mutated hepatoma cell lines¹³⁰ and in patients^50,138. CTNNB1-mutated HCCs have shown sensitivity to MET inhibitors such as cabozantinib, which are currently second-line or third-line therapy¹²⁸. Rialdi et al. described development of WNTinib, a TKI with specificity for oncogenic β-catenin signalling⁸⁶. TKIs might modulate the TIME, making CTNNB1-mutated HCC more susceptible to ICIs¹⁶³. Clinical trials testing lenvatinib (NCT03006926) or regorafenib (NCT03347292) in combination with pembrolizumab showed good ORR, although their efficacy in CTNNB1-mutated HCC has not been specifically examined. However, this effect might be unique to certain TKIs, as cabozantinib has been shown not to modulate the TIME or synergize with ICI therapy¹²⁸.

Small-molecule inhibitors

Small molecule inhibitors have been developed to ‘drug’ CTNNB1-mutated cancers, within two broad categories: those targeting upstream of β-catenin, and those targeting downstream of β-catenin. In the former category are small-molecule inhibitors targeting WNT activity, such as porcupine (PORCN) inhibitors (for example, IWP12). These inhibitors have made it to clinical trials, but are unlikely to affect β-catenin-mutated HCCs, which are independent of WNT ligands¹⁷². However, HCCs driven by APC or AXIN1 loss-of-function mutations might show responses, as β-catenin activation may be partially WNT-dependent. Further testing in Axin1-targeting sgRNA–MET and Apc-targeting sgRNA–MET models and in synergy with an ICI would be relevant to explore such precision therapeutic targeting. In the category of downstream β-catenin inhibition are small-molecule inhibitors that target β-catenin interactions with transcriptional co-factors, such as TCF–LEF or CREB-binding protein (CBP). ICG-001, PMED-1, E7386 and PRI-724 inhibit interactions between β-catenin–TCF and CBP, and have shown success in preclinical models, with E7386 being tested in combination with pembrolizumab in a phase I/II trial in selected solid tumours including HCC (NCT05091346)^173–175. Tankyrase inhibitors target β-catenin through stabilization of the AXIN destruction complex through inhibiting ADP ribosylation. The tankyrase inhibitors XAV939 and WXL-8 have shown promise in preclinical models¹⁷⁶. Another tankyrase inhibitor G007-LK has shown synergy with cabozantinib in the Axin1-targeting sgRNA–MET model¹⁷⁷.

Gene and RNAi-mediated therapies

Various platforms to directly knockdown CTNNB1 mRNA levels in β-catenin-mutated HCC have been used in many preclinical studies. For example, our group demonstrated for the first time the therapeutic efficacy of directly suppressing β-catenin in vivo using a locked nucleic acid antisense in a β-catenin-active preclinical mouse model of HCC driven by DEN–PB exposure⁸⁸. Additionally, RNA interference (RNAi)-mediated inhibition of CTNNB1 mRNA levels through small interfering RNA (siRNA) therapeutics has demonstrated considerable success in preclinical models. This method typically uses a packaging vehicle, such as a manufactured lipid nanoparticle (LNP). Another study from our group demonstrated therapeutic efficacy in the S33Y–β-catenin–Kras-G12D model using a EnCore LNP encapsulating a Dicer substrate siRNA targeting CTNNB1 (ref. 80). Ganesh et al. demonstrated therapeutic efficacy of the same agent in a CTNNB1^S45Y;KRAS^G12D and an HCC PDX model¹⁷⁸. β-Catenin mRNA silencing with this agent has shown synergism with both trametinib, a MEK inhibitor, and with an ICI in multiple WNT–β-catenin-driven tumour models^179,180. Synergy of an ICI with an siRNA targeting CTNNB1 encapsulated in extracellular vesicles in the ΔN90–β-catenin–MET model has also been described¹⁸¹. We and others used a novel LNP–siRNA against CTNNB1 to demonstrate its therapeutic efficacy in multiple CTNNB1-mutated HCC models¹³¹, which laid the groundwork for an ongoing precision medicine-guided clinical trial in HCC (NCT06600321). After intravenous administration, the EnCore LNP binds to apolipoprotein E in the circulation and is internalized in hepatocytes (including tumour cells) via the LDL receptor to release the CTNNB1 siRNA payload through endosomal processing^131,182.

Other novel therapies

The potential of using vesicular stomatitis virus oncolytic therapy, which expresses tumour antigens (VSV-HCC_TAA), to treat preclinical β-catenin-mutated HCC was shown by Webb et al.¹⁸³. This treatment leads to induction of anti-HCC_TAA T cells, which expand the antiviral T cell response when combined with an ICI¹⁸³. Preclinical testing of the AFP vaccine with an ICI demonstrated synergy in the β-catenin–MET model¹⁸⁴. Interestingly, AFP immunization alone is ineffective, despite inducing a strong antitumour T cell shift in the TIME, yet combination with an ICI affects tumour progression and growth¹⁸⁴. However, it is unclear whether this type of therapy has the potential to be translated to the clinic, as β-catenin-mutated HCC tend to be AFP-negative.

Feasibility of β-catenin suppression in patients with HCC

The WNT–β-catenin pathway is critical for the development of many organs. In adults, this pathway is also critical in tissues that are highly dependent on an active stem cell compartment in homeostasis, such as the skin, gut and haematopoietic tissue¹⁸⁵. As a result, as with many other targeted therapies and chemotherapy, suppression of this pathway to treat cancer would have to be carefully monitored. It is likely that the tumours driven by chronically high levels of WNT–β-catenin pathway activation for tumour metabolism and growth would be susceptible to low doses of inhibitors, which might not affect the low level of pathway activity essential for homeostasis in some tissues, and thus lack any major untoward adverse effects. Additionally, we expect these regimens to be administered in cycles, which might permit recovery of tissues dependent on WNT–β-catenin signalling for homeostasis. Finally, targeted delivery of inhibitors to tumour cells within the liver by packaging in specific LNPs or by conjugating to N-acetylgalactosamine would be ideal to prevent off-target adverse effects^{80,131,186,187}.

It is also important to realize the dual role of β-catenin both as an effector of the WNT signalling pathway and as an important component of adherens junctions, where it bridges the cytoplasmic tail of E-cadherin to α-catenin, thus having a role in cell–cell adhesion¹⁸⁸. While not relevant in other scenarios, this complex becomes relevant when β-catenin itself is targeted for the treatment of HCC, with the concern being a potential loss of cell junctions that could promote local and distant invasion of tumour cells. Fortunately, we identified a tissue-agnostic compensatory molecular switch that allows maintenance of adherens junctions upon β-catenin loss or suppression¹⁸⁸. As previously identified by others in the heart and skin, we identified an increase in γ-catenin, normally present in desmosomes, to complex with E-cadherin and α-catenin and maintain adherens junctions upon β-catenin knockdown in hepatocytes^189–191. This compensation only occurs at cell–cell junctions and is truly functional, as dual loss of β-catenin and γ-catenin in livers led to notable morbidity and mortality in mice associated with progressive cholestatic injury^192,193. We showed that γ-catenin stabilization upon β-catenin suppression is dependent on protein kinase A, supporting the importance of sparing this pathway when β-catenin is being directly inhibited for HCC treatment¹⁹⁴.

In summary, like any oncological therapeutics, suppression of the WNT–β-catenin pathway might be associated with certain adverse effects, but with dose adjustments, cyclical regimens and tissue-specificity, these effects could be mitigated or reduced in severity. In addition, owing to redundancy, β-catenin suppression should not have any deleterious effect on adherens junctions and thus should not promote any local or distant metastases.

Future perspectives

Research on the intricacies of WNT–β-catenin signalling in liver pathobiology for over 25 years has eventually led to the successful development of therapeutics that are now being tested in patients. In clinical trials in HCC, this will be the first precision medicine-centred approach. There continues to be room for further optimization in terms of understanding the mechanisms, patient selection, durability, optimum combination therapy and adverse effects. As heterogeneity in β-catenin-active tumours is addressed through the use of unbiased single-cell and spatial multi-omics techniques that might reveal new target genes, interactomes, metabolic dependencies and cellular ‘neighbourhoods’, new targets might emerge for further therapeutic optimization. In addition, there might be opportunities to identify new biomarkers that could be used for patient stratification for further personalized medicine in this molecular subset of HCC. A clear identification of a prognostic biomarker is needed to stratify patients successfully, so as to avoid untoward adverse effects of first line-treatments and to improve objective response rates. Understanding the molecular basis for poor responses to ICIs based on underlying mechanisms, disease aetiology and genetics using multi-omics, multiscale modelling, machine learning and other modalities, might aid in optimal patient selection. Similarly, further understanding the cellular and molecular underpinnings of immune cell exclusion in some but not all β-catenin-active tumours would be critical to either modulate those pathways to induce responses or for patient exclusion. As patients are being treated with β-catenin-directed therapies or for metabolic dependencies owing to mutant CTNNB1, the durability and resistance mechanisms would need to be investigated alongside any unexpected or severe adverse reactions to these novel therapies. Collection of patient tissues and other biospecimens are now standard for clinical trials and would allow both identification of biomarkers of response or resistance and further understanding of tumour biology of this molecular subset.

Conclusions

WNT–β-catenin-active HCCs are a heterogeneous group of tumours owing to their molecular underpinnings, which include diverse mutations in CTNNB1, AXIN and APC or activation of TGFβ and other mechanisms. These tumours, while dependent on β-catenin activation, still demonstrate heterogeneity, probably according to the extent of β-catenin expression, protein–protein interactions, and eventually target gene expression. Specifically, CTNNB1-mutated HCCs express high levels of GS, eventually leading to unique mTOR-dependency of the tumour cells, making them susceptible to mTOR inhibitors. Further, subsets of WNT–β-catenin-active tumours display an immune-excluded TIME, and as a result show suboptimal responses to ICIs. As a result, negative selection of such patients for ICI treatment would help to improve response rates and prevent unnecessary adverse effects. WNT–β-catenin-directed therapies are now a reality, with various modulators with various modes of action now in different stages of preclinical and clinical investigations. Overall, we are witnessing the start of a new era of precision-medicine for treating HCC, with the WNT–β-catenin pathway as the primary focus for future opportunities. We are cautiously optimistic regarding the effect of selective β-catenin inhibition in the first-line and second-line settings of HCC neoadjuvant or adjuvant treatment over the next 5 years.

Key points.

• The WNT–β-catenin pathway regulates metabolic zonation in the liver lobule and regulates expression of genes in hepatocytes residing in zone 3.

• Activation of the WNT–β-catenin pathway is evident in a notable subset of hepatocellular carcinomas (HCC) due to multiple mechanisms including missense mutations in CTNNB1.

• HCCs driven by mutations in CTNNB1 exhibit unique zone 3 metabolic phenotypes and tumour cell metabolism.

• β-Catenin-active HCCs have a unique tumour immune microenvironment due to both tumour cell-intrinsic and tumour cell-extrinsic mechanisms.

• Preclinical models and multi-omics clinical studies have been key to understanding the unique biology of the molecular subset of HCCs with WNT–β-catenin activation.

• Current precision medicine-based clinical trials are exploiting novel metabolic vulnerabilities or directly targeting the WNT–β-catenin pathway for HCC treatment.