Promotion of growth factor signaling as a critical function of β-catenin during HCC progression

Hepatocellular carcinoma (HCC) is the second leading cause of cancer-related deaths worldwide. β-catenin is widely thought to be a major oncogene in HCC based on the frequency of mutations associated with aberrant Wnt signaling in HCC patients. Challenging this model, our data reveal that β-catenin nuclear accumulation is restricted to the late stage of the disease. Until then, β-catenin is primarily located at the plasma membrane in complex with multiple cadherin family members where it drives tumor cell survival by enhancing the signaling of growth factor receptors such as EGFR. Therefore, our study reveals the evolving nature of β-catenin in HCC to establish it as a compound tumor promoter during the progression of the disease.

H epatocellular carcinoma (HCC) is the second cause of cancer-related deaths worldwide 1,2 . Most HCC patients are not eligible for curative surgery due to the advanced stage of the disease at the time of diagnosis. Therapeutic options are limited in their availability and efficacy, with multi-kinase inhibitor Sorafenib showing survival benefit of only 3 months 3 . Therefore, there is a critical need for new HCC treatments. Large sequencing efforts have shown that most recurring driver mutations in HCC (p53, telomerase, β-catenin, etc.) are shared with other cancer types 4 . However, a recent global cancer gene expression analysis also reveals that HCC transcriptome is strikingly different from other cancers 5 . From this discovery emerges the concept that common driver mutations may have unique functional consequences in the context of HCC. However, the mechanisms of action of the principal drivers of HCC initiation and progression remain elusive.
β-Catenin is a protein that serves two distinct roles depending upon its subcellular location. First, β-catenin in the nucleus is a critical effector of the Wnt signaling pathway 6 . In the absence of a Wnt ligand, β-catenin is rapidly phosphorylated by the destruction complex (comprising glycogen synthase kinase 3β (GSK3β), casein kinase Iα (CKIα), Axin, and adenomatosis polyposis coli (APC)) on four key serine/threonine residues, which primes β-catenin for degradation 7 . Wnt ligand binding destabilizes the destruction complex, thus promoting β-catenin stability and its translocation to the nucleus where it serves as a co-factor for TCF/LEF (T cell factor/lymphoid enhancer-binding factor) transcription factors. Second, β-catenin at the membrane is a component of the adherens junction (AJ) complex in association with cadherins. Within the AJ complex, β-catenin acts as a bridge between α-catenin and cadherins to maintain proper cytoskeleton structure, cell-cell interaction, and cell signaling 8 . Earlier works suggest that the AJ complex-associated β-catenin represents a reserve pool of β-catenin that can potentially feed into Wnt signaling activity, thereby allowing E-cadherin to modulate Wnt signaling by physically sequestering β-catenin at the membrane [9][10][11] . However, the mechanism of initial repartition and subsequent transition between these two pools remains poorly understood.
Mutations targeting components of the Wnt pathway are very frequent in colorectal cancer and hepatoblastoma, and aberrant Wnt signaling activation has been established as a dominant oncogenic driver in these tumors [12][13][14] . Similar mutations targeting Wnt pathway components are also frequent in HCC (15-33% of HCC patients carry activating mutations in ctnnb1 (coding for β-catenin) 15 1 β-Catenin is restricted to the membrane until late-stage hepatocellular carcinoma (HCC). a Human HCC patient data from The Cancer Genome Atlas (TCGA) database were stratified by HCC stage (I-III) and respective % of samples with one or more mutations in apc, axin1, and ctnnb1 (or none of the above) were graphed. b β-Catenin localization during HCC progression. A commercial array encompassing the different steps of the disease was stained for β-catenin. A score was attributed to each sample for the intensity of β-catenin expression at the membrane (y-axis), with corresponding p value between the groups of interest displayed on the top of the graph. Black circles represent samples with an absence of nuclear β-catenin. Red circles represent samples with nuclear β-catenin. The number of samples per stage is provided in the table below the graph. The percentage of patients who display nuclear β-catenin for each stage is displayed in red. The p value (=0.002; Tukey's multiple comparisons test) at the bottom of the graph accounts for the difference in the frequency of nuclear β-catenin-positive patients in stage IV vs. earlier stages (stage III and earlier). c Immunohistochemistry (IHC) of human HCC tissue array for β-catenin. One representative image is displayed per stage. Scale bar = 50 μm. Data are represented as mean ± SEM. **p < 0.01. n.s. not significant. See also Supplementary Fig. 1 apc 4,16 ) and have led to the conclusion that these mutations also act as dominant event to trigger oncogenic Wnt/β-catenin signaling in HCC. However, transgenic expression of degradationresistant β-catenin in the liver is not sufficient to initiate HCC 17 , and it is therefore unclear whether β-catenin activation is a driving mutation or a cooperating event that supports HCC progression initiated by other oncogenic events 18,19 . In addition, loss of β-catenin in chemically induced models of HCC increases tumor progression 20,21 , suggesting a paradoxical effect for βcatenin inactivation in HCC. Finally, although the kinetics of Wnt pathway-activating mutations during the course of the disease are poorly known, aberrant Wnt signaling activity is restricted to the advanced stages of the disease 22 . Collectively, these data suggest that the role of β-catenin in HCC is more complex than currently envisioned based on the mutational status of the Wnt pathway in HCC patients.  β-Catenin accumulates at the plasma membrane of HCC cells in the absence of significant Wnt target gene activation 23 , and initial studies drew a correlation between E-cadherin loss of expression and Wnt signaling activation in HCC 24 . These results led to the conclusion that E-cadherin exerts a dual tumor suppressor function in HCC: first by sequestering β-catenin away from Wnt signaling and second by maintaining cell-cell contact to prevent epithelial-mesenchymal transition (EMT) and metastasis 25 . However, this view may be too simplistic as Ecadherin expression in human HCC is largely variable and is elevated in 40% of HCC cases 26,27 . In addition, studies on other types of solid tumors have shown that E-cadherin can exert protumorigenic functions 28,29 . These findings suggest that the role of E-cadherin in HCC is more complex than previously thought.
To elucidate the role of β-catenin during HCC progression, we have integrated patient data analysis with mechanistic studies using a pre-clinical model of HCC initiated upon functional inactivation of the Rb family, an almost universal event during HCC development 30,31 . We show that, despite the early occurrence of mutations targeting Wnt signaling components in human HCC, membrane localization of β-catenin is a dominant feature of HCC until the advanced stages of the disease when βcatenin progressively engages in Wnt signaling activation. At the plasma membrane, β-catenin interacts with multiple cadherin family members to promote the signaling of growth factor receptors such as epidermal growth factor receptor (EGFR) and support HCC cell survival. Overall, our study reveals the evolutive nature of β-catenin tumor-promoting functions during HCC progression.
Results β-Catenin is restricted to the membrane until late-stage HCC. Wnt pathway components are mutated in 30-40% of HCC patients 4 . However, when these mutations occur during the course of the disease and whether they directly translate into the activation of the Wnt signaling are not known. To determine the mutational status of Wnt signaling during HCC development, we took advantage of a The Cancer Genome Atlas (TCGA) HCC dataset representing patients from stages I to III. Analysis shows that mutations targeting Wnt signaling are present in~35% of patients, independent of their tumor stage (Fig. 1a), indicating that these mutations represent an early event in HCC and their frequency remains stable thereafter. Next, we sought to address the subcellular localization of β-catenin during HCC progression and performed immunohistochemistry (IHC) for β-catenin in human HCC tissue arrays ranging from pre-malignant conditions to malignant stages (stages I-IV). β-Catenin is predominantly localized at the plasma membrane, but its abundance at the membrane progressively declines beyond stage I. Importantly, it is only at the stage IV that β-catenin is present in the nucleus (54% of the samples) (Fig. 1b, c), which correlates with the latestage activation of Wnt signaling reported in human HCC 32,33 . Therefore, we found a significant absence of correlation between Wnt signaling mutational status and β-catenin nuclear localization during HCC progression until stage IV (Fisher's test, p value <0.05). These results indicate that nuclear localization of βcatenin is not an immediate consequence of mutations targeting Wnt pathway components, suggesting that additional events are required for aberrant Wnt signaling activation.
Membrane-localized β-catenin promotes HCC growth. Localization of β-catenin at the membrane has been observed in multiple mouse models of HCC, either chemically induced by DEN or triggered by single or combined targeted genetic alterations (E2f1, Myc, Tgfβ, Akt, etc.) 17,[34][35][36] . Therefore, human/mouse cross-species analyses show that β-catenin is commonly found at the cellular membrane in HCC.
To determine the mechanisms governing β-catenin localization and its functional consequence for HCC progression, we took advantage of a mouse model that we previously generated 31 . Rbfamily proteins (Rb, p130, and p107) are key regulators of cell cycle activity and their functional alteration is almost universal during HCC initiation. Conditional inactivation of Rb gene family (triple knockout; TKO) in the adult mouse liver generates a welldifferentiated type of HCC (TKO HCC) that recapitulates multiple features of the human disease 30,31 . Whereas β-catenin is expressed in different subcellular compartments of hepatocytes around the central vein (CV) in control liver (CTRL), it is exclusively detected at the membrane of tumor cells in TKO HCC (Fig. 2a), with an increase in β-catenin protein but not mRNA levels (Fig. 2b, c). Serially transplanted TKO HCC tumors exhibit a poorly differentiated morphology compared to the parental TKO HCC. IHC analysis for β-catenin localization showed that subcutaneous tumors display heterogeneous expression of nuclear and membranous β-catenin ( Supplementary Fig. 1A-B). These results indicate that TKO HCC recapitulates the evolution of βcatenin localization observed in human HCC.
β-Catenin does not promote early HCC through Wnt signaling. Minimal amount of β-catenin in the nucleus is sufficient to activate Wnt target genes 38 . Therefore, we cannot rule out the contribution of Wnt signaling in HCC cell survival even though β-catenin is undetectable in the nucleus of early-stage human HCC and TKO HCC cells by IHC. Protein translation inhibition using cycloheximide (CHX) revealed that β-catenin protein expression is stable in TKO HCC cells (Fig. 3a). Sequencing analysis revealed that ctnnb1 is wild type in primary TKO HCC  ). In addition, immunoprecipitation (IP) assay showed that the destruction complex is intact ( Fig. 3c and Supplementary  Fig. 3B). These data suggest that β-catenin, although capable of being degraded by the destruction complex in TKO HCC, evades degradation via alternative means. Accordingly, treatment with XAV939, which stabilizes endogenous Axin 39 , and ectopic Axin expression failed to significantly change β-catenin expression level (Fig. 3d, e). However, stimulation of TKO HCC cells with Wnt3a ligand increased glutamine synthetase (GS; a clinical marker for Wnt activity in the liver 40 ) expression ( Fig. 3f), indicating that Wnt pathway can be activated in TKO HCC.
To determine whether Wnt signaling is active in the context of membrane β-catenin localization, we first assessed the expression of key Wnt target genes in primary tumors. GS is expressed in the pericentral zone of normal liver but is undetectable in TKO HCC (Fig. 3g). In addition, quantitative PCR (qPCR) analysis showed a decreased expression of liver specific (glul, tbx3) and canonical (lgr5, axin2) Wnt target genes in TKO HCC compared to controls (Fig. 3h). However, silencing of β-catenin in TKO HCC cells only repressed the expression of axin2, but not glul and tbx3 (Fig. 3i), suggesting that Wnt signaling is minimally active in TKO HCC.
To determine whether minimal Wnt signaling activity drives TKO HCC cell survival, we tested the transcriptional and functional consequences of suppressing Wnt target gene expression by overexpressing a dominant-negative form of TCF4 (dnTCF4) 41,42 . Ectopic expression of dnTCF4 in TKO HCC cells repressed the expression of axin2 (similar to β-catenin silencing) and GS/glul (albeit this could be through a β-catenin-independent mechanism 43 ) ( Fig. 3j, k), but failed to repress TKO HCC cell and organoid proliferation (Fig. 3l, m). Collectively, these results suggest that, despite the presence of a functional Wnt pathway, Wnt signaling activity is minimal in TKO HCC and does not promote HCC cell survival.
Finally, we systematically tested the status of the key Wnt pathway components to elucidate the mechanism of progressive nuclear translocation of β-catenin upon TKO HCC serial transplantation. We found that ctnnb1 is wild type and βcatenin is phosphorylated by the destruction complex in serially transplanted TKO HCC tumors (Supplementary Fig. 3C and Fig. 3n, top panel). In contrast, serially transplanted tumors displaying nuclear β-catenin expression had lost Axin1 expression ( Supplementary Fig. 3D). Functionally, we found a correlation between the progression of TKO HCC and the onset of metastasis markers (Fig. 3n). Therefore, these data recapitulate the observations made in advanced human HCC 22 and suggest that the activation of Wnt signaling in late-stage HCC plays a critical role in the initiation of metastasis.
E-cadherin recruits β-catenin away from the Wnt pathway. Increased β-catenin stability in the presence of a functional destruction complex suggests that it escapes degradation. Previous studies have demonstrated that the recruitment of β-catenin by the AJ complex results in delayed β-catenin protein turnover 11 . To test this possibility, we first confirmed co-localization of β-catenin and E-cadherin in TKO HCC cells (Fig. 4a) and in various human HCC cell lines ( Supplementary Fig. 4A) by IF. IP assay showed that β-catenin interacts with both E-cadherin and α-catenin in TKO HCC and this interaction was not altered upon stimulation with Wnt3A ligand (Fig. 4b, c). In addition, we observed increased E-cadherin protein and cdh1 mRNA expression in TKO HCC compared to the control liver, similar to what has been reported in human HCC 26 (Fig. 4d, e). Cdh1 expression is predominantly regulated by a family of transcription factors involved in the EMT, such as Slug, Snail, and Twist 25 , and we found decreased expression of twist1 in TKO HCC (Fig. 4e). To determine the relevance of β-catenin recruitment by the AJ complex, we exploited TCGA HCC patient dataset and identified a correlation between β-catenin and E-cadherin protein (but not mRNA) expression levels in HCC patients, independent of the tumor stage (Fig. 4f). Despite having identical residue sequences, exogenous Flag-β-catenin had a shorter half-life than endogenous β-catenin ( Supplementary Fig. 4B-C), suggesting that the recruitment of β-catenin by the AJ complex is not a posttranslational mechanism. Collectively, these data suggest that increased E-cadherin expression promotes the recruitment of βcatenin to the plasma membrane by the AJ complex to prevent its degradation by the destruction complex in HCC.
To determine the consequence of β-catenin recruitment by the AJ complex for Wnt signaling in HCC, we used a set of established Wnt target genes (glul, tbx3, lgr5, and axin2) as a readout for Wnt signaling activity in the TCGA dataset of HCC patients and found that E-cadherin expression negatively correlates with Wnt signaling activity (Fig. 4g). Using a similar approach for TCGA datasets for three other gastro-intestinal tract cancers, we failed to observe a similar correlation (Fig. 4h). Collectively, these results suggest that the dominant recruitment of β-catenin within the AJ complex to the detriment of Wnt signaling is a unique feature of HCC. Fig. 3 β-Catenin does not promote early hepatocellular carcinoma (HCC) through Wnt signaling. a Immunoblot for β-catenin in triple knockout (TKO) HCC cells and Wnt3A-inducible mouse fibroblasts used as controls for non-active vs. active Wnt signaling (L cells; parental or Wnt3a+) treated with 25 μg/ml cycloheximide (CHX) or dimethyl sulfoxide (DMSO) for the time indicated (0-24 h). b Comparison of ctnnb1 complementary DNA (cDNA) sequence in control liver (CTRL) and TKO HCC. Regions highlighted in blue correspond to the phosphorylation sites. c Immunoprecipitation (IP) of immunoglobulin G (IgG) and Axin in TKO HCC cells. The presence of Axin, glycogen synthase kinase 3β (GSK3α/β), and phospho-β-catenin in the pull-down fraction was determined by immunoblot. d TKO HCC cells and as L cells (parental or Wnt3a+) were treated with 5 μM XAV939 for 48 h. The expression levels of Axin and β-catenin were determined by immunoblot. e Axin cDNA was overexpressed in TKO HCC cells and L cells (parental or Wnt3a+). The expression levels of Axin and β-catenin were determined by immunoblot. f Immunoblot for β-catenin and glutamine synthetase (GS) in TKO HCC cells treated with Wnt3A media. g Immunohistochemistry (IHC) for GS in control liver (CTRL) and TKO HCC. Black arrowheads indicate central vein (CV). The tumor zone in TKO HCC is delineated by a white dotted line. Scale bar = 50 μm. h Quantitative reverse transcription PCR (RT-qPCR) analysis for glul, tbx3, lgr5, and axin2 in control liver (CTRL; n = 4) and TKO HCC (n = 9). i RT-qPCR analysis for glul, tbx3, lgr5, and axin2 in TKO HCC cells upon β-catenin knockdown (KD) with shβcat3-4 (n > 3). j Immunoblot for β-catenin and GS in TKO HCC cells overexpressing dominant-negative of TCF4 (dnTCF4). k RT-qPCR analysis for glul, tbx3, lgr5, and axin2 in TKO HCC cells upon dnTCF4 overexpression (n = 3). l Growth curve of TKO HCC cells upon dnTCF4 overexpression (n = 2). m Fold difference in the number of TKO HCC organoids upon dnTCF4 overexpression (n = 2). n Immunoblot of phospho β-catenin (Ser33/37/Thr41) and epithelial-mesenchymal transition (EMT) markers (Zeb1, Vimentin, and Claudin7) in control (CTRL) livers (n = 3), primary TKO HCC tumors (n = 4) and serially transplanted TKO HCC (Serial) tumors (n = 4). Data are represented as mean ± SEM. *p < 0.05, **p < 0.01, and ***p < 0.001. n.s. not significant. See also Supplementary Fig. 3 Compound integration of β-catenin in the AJ complex in HCC.
To determine the role of the AJ complex in HCC, we silenced Ecadherin expression in TKO HCC cells using two independent hairpins (shEcad1-2). Efficient E-cadherin KD did not repress the proliferative capacity of TKO HCC cells, despite an alteration in cell morphology and cell-cell contact at low cell density ( Fig. 5a and Supplementary Figu. 5A). IF analysis showed that β-catenin maintains membrane surface localization despite E-cadherin loss, suggesting a compensatory mechanism for E-cadherin loss of expression (Fig. 5b). To determine this compensatory mechanism, we analyzed β-catenin interactome in TKO HCC by IP followed by mass spectrometry (MS) and found that most βcatenin interactions occur with either the AJ complex or the degradation complex (Fig. 5c and    unbiased approach showed that, in addition to E-cadherin, βcatenin also interacts with N-and K-cadherin, which was confirmed by IP (Fig. 5d). We failed to detect by IP the interaction between β-catenin and K-cadherin upon N-cadherin KD (shNcad1), which suggests that N-cadherin may mediate β-catenin/K-cadherin binding. Next, we performed E/N-cadherin single KD or double KD (DKD) in TKO HCC cells and found that silencing of both cadherins leads to reduced half-life and decreased expression of β-catenin (Fig. 5e, f and Supplementary  Fig. 5B), without any effect on Wnt target gene (glul, tbx3, and axin2) expression ( Supplementary Fig. 5C). Combined E-and Ncadherin repression led to decreased TKO HCC cell growth in vitro with altered cell morphology and impaired cell-cell contact ( Fig. 5g and Supplementary Fig. 5D). Of note, N-cadherin expression did not increase in TKO HCC compared to controls, suggesting a mechanism specific for E-cadherin up-regulation in TKO HCC ( Supplementary Fig. 5E-F).
These data indicate that compound integration of β-catenin by several members of the cadherin family in the AJ complex is crucial for β-catenin retention at the membrane and HCC cell survival.
To determine the long-term consequences of disrupting the AJ complex in HCC development, TKO HCC cells with E/Ncadherin single KD or DKD were subcutaneously transplanted into the flank of immunocompromised nonobese diabetic/severe combined immunodeficiency (nod-scid) mice. In contrast to the growth inhibition observed in an in vitro short-term context, DKD did not impair tumor formation in vivo (Fig. 5h) and actually lead to increased cell cycle activity ( Supplementary  Fig. 5G). Histological and IHC analyses revealed that individual clones lacking detectable β-catenin expression progressively emerge from double E/N-cadherin-deficient tumor masses ( Fig. 5i and Supplementary Fig. 5H). While there was no difference in tumor grade among the different conditions, cells within the E/Ncadherin DKD tumors displayed heterogeneous nuclear size and ploidy ( Fig. 5j and Supplementary Fig. 5I-J). In addition, IHC analysis revealed a mosaic GS expression in the E/N-cadherin DKD tumors, indicative of heterogeneous Wnt signaling activity (Fig. 5k). Interestingly, we could not find co-occurrence of increased nuclear size and GS expression, suggesting that Wnt signaling activation and DNA content alteration are mutually exclusive, in line with previous observations 16,34 . Collectively, these data suggest that the long-term consequences of disrupting the AJ complex, which include increased proliferation, heterogeneous nuclear size, and ploidy, as well as Wnt signaling activation, are beneficial for tumor growth.
EGFR signaling promotes HCC cell survival. To determine the mechanism by which the AJ complex supports HCC cell survival, we systematically investigated the consequences of targeting each of its established functions (cytoskeleton maintenance, cell-cell interaction, and promotion of growth factor signaling) in TKO HCC. We first disrupted AJ complex interaction with the intracellular cytoskeleton by repressing the expression of α-catenin, which acts as a bridge between the AJ complex and F-actin filaments 8 . α-Catenin KD altered the morphology (but not the cell-cell contact) of TKO HCC cells ( Supplementary Fig. 6A), but did not impact cell growth nor β-catenin expression (Fig. 6a), indicating that the AJ complex does not promote HCC cell survival by preserving the integrity of the cytoskeleton. We next disrupted the cell-cell adhesion activity of the AJ complex by overexpressing a dominant-negative form of E-cadherin (DNE) subcloned into MigR1-GFP (co-expressing the GFP protein as a selection marker) 44 . DNE lacks most of the ectodomain of Ecadherin (required for the maintenance of cell-cell adhesion), but retains the transmembrane domain (for anchoring E-cadherin to the plasma membrane) and the cytoplasmic domain (Supplementary Fig. 6B), allowing DNE to compete with endogenous Ecadherin, and in particular its slow migrating form, for β-catenin binding (Fig. 6b, c). IF showed that DNE expression increases βcatenin expression and re-directs its subcellular localization in TKO HCC and in human HCC cells ( Fig. 6d and Supplementary  Fig. 6C-D). DNE expression did not impair β-catenin half-life or TKO HCC growth in vitro ( Fig. 6e and Supplementary Fig. 6E) and wound healing assays showed no difference in the migratory capacity of TKO HCC cells upon DNE expression (Supplementary Fig. 6F). DNE-expressing TKO HCC cells were subfractionated based on their GFP expression, which showed a correlation between DNE expression and the disrupted cell-cell adhesion without any consequences on their growth (Supplementary Fig. 6F-G). Collectively, these data indicate that βcatenin within the AJ complex supports HCC cell survival by mechanisms independent of cytoskeletal maintenance or cell-cell adhesion.
Finally, cadherins also interact with several receptor tyrosine kinases (RTKs), with varying and sometimes conflicting consequences 45 . Among these RTKs, E-cadherin interacts with EGFR through its ectodomain 46 . EGFR is overexpressed in 40-70% of HCC and treatment of HCC cell lines, xenograft, and mouse models with EGFR inhibitors represses cell growth 47,48 , indicating that EGFR signaling is critical for HCC cell proliferation. To determine whether EGFR signaling supports TKO HCC cell survival, we first silenced EGFR expression by short hairpin RNA (shRNA). Repression of EGFR expression decreases TKO HCC cell growth in a dose-dependent manner. Analysis of other EGFR family members shows no compensation by EGFR2, while EGFR3 was undetectable by qPCR and immunoblot analysis (Fig. 6f). Next, we tested the capacity of EGFR inhibitors to Fig. 4 E-cadherin recruits β-catenin away from the Wnt pathway. a Immunofluorescence (IF) for β-catenin (red), E-cadherin (green), and 4′,6-diamidino-2phenylindole (DAPI) (blue) in triple knockout hepatocellular carcinoma (TKO HCC) and Wnt3a+ L cell (a positive control for nuclear β-catenin). Scale bar = 25 μm. b Immunoprecipitation (IP) for immunoglobulin G (IgG) and β-catenin in TKO HCC cells. The presence of β-catenin, E-cadherin, and α-catenin were detected in the pull-down fraction by immunoblot. c IP for IgG and β-catenin in TKO HCC cells treated with Wnt3A media. The presence of β-catenin and E-cadherin were detected in both unbound and pull-down fraction by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. d Immunohistochemistry (IHC) for E-cadherin in control liver (CTRL) and TKO HCC. The tumor zone in TKO HCC is delineated by a white dotted line. Scale bar = 100 μm (top), 25 μm (bottom). e Quantitative reverse transcription PCR (RT-qPCR) analysis in control livers (CTRL, n = 4) and TKO HCC (n = 8) for cdh1, snai2, snai1, and twist. f Correlation between E-cadherin vs. β-catenin (left top; p value = 2.509e−12), cdh1 vs. ctnnb1 (left bottom; p value = 4.861e−3), cdh1 vs. E-cadherin (right top; p value = 2.426e−16) and ctnnb1 vs. β-catenin (right bottom; p value = 0.691) expressions in HCC patients at different stages of the disease (CR = correlation ratio; Pearson's product-moment correlation). g, h Analysis of protein and mRNA expression from The Cancer Genome Atlas (TCGA) patient dataset for g HCC and h other gastro-intestinal cancers (colorectal, stomach, and esophageal). Samples were separated based on E-cadherin protein expression (low vs. high). Expression of glul, tbx3, lgr5, and axin2 were assessed between these two groups to determine the correlation (non-parametric t test) between E-cadherin expression and Wnt target gene expressions in different cancers. Data are represented as mean ± SEM. *p < 0.05 and **p < 0.01. n.s. not significant. See also Supplementary Fig. 4 repress TKO HCC survival. We treated TKO HCC cells and organoids derived from primary TKO HCC tumors with Erlotinib ® (EGFR inhibitor 49 ), Afatinib ® (EGFR family inhibitor 50 ), and Sorafenib ® (a multi-kinase inhibitor used as a control 51 ). In these conditions, we found that EGFR inhibition by both Erlotinib and Afatinib trigger apoptosis of TKO HCC cells and TKO HCC organoids (Fig. 6g, h). Interestingly, significant effects were already observed at 1 μM Afatinib vs. 10 μM Erlotinib, suggesting that targeting multiple EGFR family members may be more potent to trigger TKO HCC cell death. Collectively, these results show that EGFR inhibition impairs TKO HCC growth.  The AJ complex enhances EGFR signaling. To elucidate the potential regulation of EGFR signaling by the AJ complex in TKO HCC, we first determined whether the components of the AJ complex interact with EGFR. IP assay showed that EGFR interacts with both E-and N-cadherin but not with β-catenin (Fig. 7a), suggesting an indirect interaction between EGFR and β-catenin.
While β-catenin interacts with both migrating forms of E-cadherins, we observed that EGFR only interacts with the slow migrating form of E-cadherin, suggesting two pools of E-cadherin/β-catenin complex in the cell. Silencing E-cadherin repressed EGFR expression (Fig. 7b). In addition, silencing β-catenin expression led to a significant decrease in E-cadherin and EGFR expression and delayed EGFR activation upon EGF stimulation (Fig. 7c). Repression of EGFR expression upon β-catenin silencing appears to be due to post-translational mechanism since egfr mRNA was not altered upon silencing β-catenin (Fig. 7d). Interestingly, although silencing β-catenin repressed the expression of both E-cadherin and EGFR, it did not impair EGFR-E-cadherin interaction (Fig. 7e). Phosphorylation of β-catenin by multiple tyrosine kinases on residue Tyr654 disrupts β-catenin interaction with E-cadherin 52 . Treatment of TKO HCC cells with orthovanadate salt, a general phosphotyrosyl phosphatase inhibitor 53 , disrupted the interaction between β-catenin and the slow migrating form of E-cadherin and decreased the expression of β-catenin, E-cadherin, and EGFR (as illustrated by the difference of signal intensity in the input lanes) (Fig. 7f). Overall, our results from genetic (shRNA silencing of AJ complex components) and pharmacological disruption of the βcatenin/cadherins interaction indicate that the AJ complex interacts with EGFR to promote the receptor stability and enhance EGFR signaling. Analysis of Tyr654 phosphorylation status shows decreased phosphorylation in TKO HCC (two out of nine) compared to controls (five out of six) ( Supplementary  Fig. 7A). However, pTyr654 β-catenin was undetectable in serially transplanted tumors, suggesting that different post-translational modifications 11 may regulate AJ complex stability during TKO HCC evolution. Analysis of the control liver, and primary and serially transplanted TKO HCC samples indicates that EGFR expression decreases in TKO HCC vs. control livers (likely as a consequence of decreased mRNA abundance; Supplementary Fig. 7B) and its activation (phosphorylation of the Tyr1068 residue as a readout) becomes heterogeneous. However, EGFR and activated EGFR become undetectable in subcutaneous tumors characterized by a progressive shift in β-catenin localization, further supporting the concept that an intact AJ complex is required for EGFR stability and signaling (Fig. 7g). Interestingly, we also observed a similar expression pattern for EGFR2, further suggesting that the AJ complex may stabilize multiple growth factor receptors in HCC.
Upon its stimulation, EGFR is internalized to be either degraded through the ubiquitin/lysosome pathway or to be recycled back to the membrane 54 . To determine the fate of EGFR upon disruption of its stabilizing interaction with the AJ complex, we treated TKO HCC cells with CHX for increasing amount of time and found that repression of β-catenin increased the degradation of EGFR, suggesting that EGFR is degraded and not recycled upon disruption of the AJ complex (Fig. 7h). Accordingly, exogenous expression of Flag-Ubiquitin followed by the IP of the Flag peptide showed an increased presence of EGFR in the pull-down fraction upon β-catenin silencing (Fig. 7i). Similarly, IP of EGFR showed an increased presence of ubiquitin chains upon silencing of β-catenin (Fig. 7j). Collectively, these results indicate that disruption of the interaction between EGFR and the AJ complex promotes EGFR degradation by the ubiquitin/ lysozyme pathway.

Correlation between E-cadherin and EGFR in HCC patients.
To investigate the relationship between EGFR signaling and AJ complex in human HCC, we interrogated the TCGA dataset. Analysis of EGFR and AJ complex protein expression in human HCC identifies a significant correlation between total EGFR and either E-cadherin or β-catenin expression (Fig. 8a). A positive correlation was also observed between activated EGFR and Ecadherin, but failed to reach significance. When repeated on a gender basis, we found a significant correlation within the male, but not the female group. Further analysis of the female group identified a subgroup of outlier patients with low EGFR expression (denoted by black dots) who display decreased expression of the EGF ligand (fold repression: 3.45 log 2 , p value < 0.004; DESeq2 Analysis) (Supplementary Data File), and exclusion of these outlier patients restores a significant correlation between activated EGFR and E-cadherin within the female group (Fig. 8b). Although the mechanism underlying the gender exclusive decreased EGFR/EGF expression is not known, these results establish a correlation between EGFR expression/signaling activity and AJ complex abundance in human HCC. In addition, detection of EGFR and E-cadherin expression in multiple epithelial cancers reveals a significant correlation for the following types of cancers (head and neck, colorectal cancer, and nonsmall-cell lung carcinoma (NSCLC)) out of seven tested Fig. 5 Compound integration of β-catenin in the adherens junction (AJ) complex in hepatocellular carcinoma (HCC). a Growth curve of triple knockout HCC (TKO HCC) cells upon E-cadherin knockdown (KD) with shEcad1-2 (n = 4). E-cadherin protein level is determined by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. b Immunofluorescence (IF) for E-cadherin (magenta), β-catenin (green), and 4′,6-diamidino-2phenylindole (DAPI (blue) in TKO HCC cells expressing an empty vector (control (CTRL)) or shEcad2 (E-cadherin KD). Scale bar = 25 μm. c Mass spectrometry (MS) analysis identifies the interactome of β-catenin in TKO HCC cells upon E-cadherin KD (shEcad2), relative to the control (CTRL). d Immunoprecipitation (IP) of immunoglobulin G (IgG) and β-catenin in TKO cells expressing an empty vector (CTRL), shEcad2 (E-cadherin KD), shNcad1 (N-cadherin KD), or shEcad2 + shNcad1 (E/N-cadherin double knockdown (DKD)). Endogenous β-catenin was pulled down and β-catenin, E-cadherin, Ncatenin, and K-cadherin were detected by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. e β-Catenin, E-cadherin and Ncadherin expression in TKO HCC cells expressing an empty vector (CTRL), shEcad2 (E-cadherin KD), shNcad1 (N-cadherin KD), or shEcad2 + shNcad1 (E/ N-cadherin DKD). Values in the lower panel represent the intensity of corresponding bands, normalized by actin. f IF for β-catenin (green) and DAPI (blue) in TKO HCC cells expressing an empty vector (CTRL), shNcad1 (N-cadherin KD), or shEcad2 + shNcad1 (E/N-cadherin DKD). g Growth curve of TKO HCC cells expressing an empty vector (CTRL) or shEcad2 + shNcad1 (E/N-cadherin DKD). (n = 3). Scale bar = 25 μm. h Subcutaneous tumors from nonobese diabetic/severe combined immunodeficiency (nod/scid) mice injected with TKO HCC cells expressing an empty vector (CTRL), shEcad2 (E-cadherin KD), shNcad1 (N-cadherin KD), or shEcad2 + shNcad1 (E/N-cadherin DKD). i Immunohistochemistry (IHC) for β-catenin in subcutaneous tumors. j Quantification of nuclei size in subcutaneous tumors from TKO HCC cells, control (CTRL) or E/N-cadherin DKD (n = 3). Scale bar = 200 μm, 60 μm (enlarged). k IHC for glutamine synthetase (GS) in subcutaneous tumor from TKO HCC cells with E/N-cadherin DKD. Cells with enlarged nuclei (arrowheads) do not express GS, indicating that Wnt activity is not required to trigger genomic instability. Scale bar = 50 μm. Data are represented as mean ± SEM. **p < 0.01 and ***p < 0.001. n.s. not significant. See also Supplementary Fig. 5 ( Supplementary Fig. 8A). These cancers display mutations of the egfr gene in 4-14% of the cases, suggesting that AJ complexmediated stabilization and genetic alteration may not represent mutually exclusive mechanisms to promote EGF signaling in cancer (Supplementary Fig. 8B). Finally, the increased apoptotic activity of Afatinib vs. Erlotinib and the correlation between EGFR2 expression and β-catenin presence at the membrane during TKO HCC progression suggest that the AJ complex may expand its stability promoting functions to other EGFR family members. Accordingly, analysis of human HCC also indicates a  strong correlation between the expression of EGFR2 and EGFR3 and E-cadherin (Fig. 8c). Collectively, our results identify the promotion of epidermal growth factor signaling through cadherin maintenance as a critical tumor-promoting function of membranous β-catenin in HCC.

Discussion
In this study, we provide evidence that β-catenin in the AJ complex supports HCC cell survival by promoting EGFR signaling during the earlier stages of HCC, and that its Wnt signaling associated role is restricted to late-stage HCC. Our report challenges the conventional view that β-catenin is a dominant oncogenic driver and instead highlights β-catenin as a tumor promoter with evolving functions during HCC development (Fig. 8d). The AJ complex is best known to regulate cell-cell interaction and maintain F-actin-composed cytoskeleton, but here we show that the AJ complex also supports HCC through EGFR signaling. E-cadherin binds to EGFR in various solid tumors to modulate its activation but the pro-and anti-tumorigenic consequences appear to be context-dependent 45,55 . In the context of HCC, our data indicate that cadherins stabilize EGFR, and AJ complex disruption impairs EGFR stability, signaling, and tumor cell survival. EGFR signaling is often hyperactive in HCC, and its inhibition is detrimental for HCC cell survival 47,56 . However, EGFR inhibition in HCC does not impair tumor progression as the significant level of acute tumor cell death associated with EGFR inhibition induces compensatory proliferation and fosters tumor evolution 57 . Supportive of this, DEN-and DEN/Phenobarbital-induced HCC develop faster in the context of β-catenin deficiency, suggesting a paradoxical role of β-catenin in HCC 20,21 . In these experimental models, β-catenin is located at the plasma membrane and its deficiency impairs EGFR expression. Therefore, our data elucidate the mechanism of paradoxical tumor progression upon β-catenin deficiency/inhibition by establishing the connection between the AJ complex and EGFR signaling in HCC.
The fact that mutations targeting β-catenin, Axin, and APC occur as early as stage I and their frequency remains constant throughout HCC development suggests that Wnt signaling may nevertheless exert a pro-tumorigenic function at the early stage of HCC development. Our data determine that β-catenin-driven promotion of cell survival is independent of Wnt signaling, but limited data suggest that Wnt signaling could support more discrete function(s) such as escaping immune surveillance 58 . Despite the early occurrence of Wnt-activating mutations, it is not until stage IV that we observe a drastic shift in β-catenin localization, indicating that β-catenin localization is not solely dependent on the mutational status of the Wnt pathway. The mechanisms that ultimately trigger β-catenin cellular redistribution in advanced HCC remain unresolved, although the effect of orthovanadate treatment on β-catenin/E-cadherin interaction suggests that increased tyrosine phosphorylation represents an important mechanism to disrupt AJ complex stability.
Our data indicate that AJ complex disruption has multiple beneficial consequences for advanced HCC progression. Although AJ complex disruption is detrimental for tumor growth in the short-term (as illustrated by decreased cell growth in vitro), longterm consequences are multiple and appear to be mutually exclusive such as altered DNA content and Wnt signaling activation. In contrast to colorectal cancer and hepatoblastoma, where Wnt signaling exerts a well-established driver role [12][13][14] , the functional consequences of aberrant Wnt signaling for HCC remain obscure, as illustrated by its association with either good or bad prognosis 59 . Our data, as well as others, indicate that aberrant Wnt signaling is a feature of advanced HCC, characterized by the activation of multiple other oncogenic pathways (PI3k, Myc, Tgfβ, etc.) 22 . This molecular context raises the question of the exact contribution of Wnt signaling for late-stage HCC maintenance and progression. In particular, the mosaic expression of GS (Fig. 5k), a key marker of Wnt signaling in HCC 60 , supports the concept that aberrant Wnt signaling is not a dominant oncogene in late-stage HCC. Based on Fig. 3n and previous reports 22 , we propose that activation of Wnt signaling in advanced-stage HCC promotes specific features such as the initiation of metastasis.
In conclusion, our results challenge the current view on Wnt/βcatenin signaling in HCC and reveal an unexpected role for βcatenin in HCC development, with emphasis on its complex and evolving tumor-promoting function during the course of the disease. Future research will specifically focus on targeting AJ complex-associated tumor-promoting functions of β-catenin to impact HCC development.

Methods
Mice. Rb-family cTKO mice were maintained on a mixed 129Sv/J;C57/BL6 background 31 . All experiments with mice were approved by the Children's Hospital of Philadelphia Institutional Animal Care and Use Committee (CHOP IACUC) (protocol #969). All animal studies were conducted in compliance with the ethical regulations for animal testing and research. For intrasplenic injection, mice were anesthetized and surgically opened on the upper left flank to inject adenovirus into the spleen. For subcutaneous serial transplantation, independent HCC primary tumors from the parental TKO mice were dissociated into single cells. The tumor cells (ranging 1 × 10 3 -10 6 ) were resuspended in Matrigel and injected subcutaneously into the flanks of nod/scid mice. The resulting subcutaneous tumors [Generation 1] were harvested, dissociated, and subcutaneously transplanted [Generation 2]. For subcutaneous TKO HCC cell injection, TKO cells (ranging 5 × 10 4 -10 5 ) were sorted and injected subcutaneously into nod/scid mice. Experiment endpoint was determined by the first appearance of tumors >1 cm 2 .
Cell culture. Rb-family TKO 1.1 and 2.1 cell lines were derived from independent TKO HCC tumors 30 . Experiments have been performed in both cell lines with similar results and they are collectively referred to as TKO HCC cells. TKO HCC, HepG2 (ATCC ® HB-8065), Hep3B (ATCC ® HB-8064), and Huh-7 (JCRB0403 HuH-7) cells were cultured in DMEM (Mediatech), supplemented with 10% fetal bovine serum (FBS)/1% penicillin-streptomycin-glutamine (Invitrogen). SNU449 (ATCC ® CRL-2234) cells were cultured in RPMI, supplemented with 10% FBS/1% Fig. 6 Epidermal growth factor receptor (EGFR) signaling promotes hepatocellular carcinoma (HCC) cell survival. a Growth curve of triple knockout (TKO) HCC cells upon α-catenin KD with shαcat1-2 (n = 6). α-Catenin and β-catenin protein levels upon expression of shαcat1-2, as detected by immunoblot. b Immunoblot for E-cadherin and N-cadherin in TKO HCC cells infected with empty MigR1 (control (CTRL)) or MigR1 dominant-negative form of Ecadherin (DNE). White arrowhead indicates DN-Ecad. c Immunoprecipitation (IP) of immonoglobulin G (IgG) and β-catenin in TKO HCC cells expressing control (CTRL) or DNE. The expressions of β-catenin, E-cadherin, and DNE were detected by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. d IF for E-cadherin (green), β-catenin (red), and 4′,6-diamidino-2-phenylindole (DAPI (blue) in TKO HCC cells and HEPG2 cells expressing control (CTRL) or DNE. Scale bar = 25 μm. e Immunoblot for β-catenin in TKO HCC cells expressing control (CTRL) or DNE. Cells were treated with 25 μg/ ml cycloheximide (CHX) for the time indicated (0-24 h). f Growth curve of TKO HCC cells upon EGFR KD with shEGFR(A-B) (n = 3). EGFR and EGFR2 (HER2) protein levels upon expression of shEGFR(A-B), as detected by immunoblot. g Representative fluorescence-activated cell sorting (FACS) plot for AnnexinV-FITC expression on TKO HCC cells upon 0-50 μM Afatinib, Erlotinib, and Sorafenib treatment for 24 h. Dosage conditions that have significant effect on cells are highlighted in red. h Representative phase contrast images of TKO HCC organoids upon 0-50 μM Afatinib, Erlotinib, and Sorafenib treatment for 10 days. Dosage conditions that have significant effect on organoids are highlighted in red. Data are represented as mean ± SEM. ***p < 0.001. n.s. not significant. See also Supplementary Fig. 6 a  Fig. 7 The adherens junction (AJ) complex enhances epidermal growth factor receptor (EGFR) signaling. a Immunoprecipitation (IP) for immunoglobulin G (IgG) and EGFR in triple knockout hepatocellular carcinoma (TKO HCC) cells. The presence of EGFR, E-cadherin, N-cadherin, and β-catenin were detected in both unbound and pull-down fractions by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. b Immunoblot for E-cadherin and EGFR in TKO HCC cells expressing control (CTRL) or shEcad1-2 (E-cadherin KD). c Serum-starved TKO HCC cells expressing control (CTRL) or shβcat3 (β-catenin KD) were treated with EGF (20 ng/ml) for the time indicated (0-60 min). β-Catenin, E-cadherin, P-EGFR(Tyr1068, activated EGFR), and EGFR were detected by immunoblot. d Quantitative reverse transcription PCR (RT-qPCR) analysis for egfr mRNA levels in TKO HCC cells expressing control (CTRL) or shβcat3 (β-cadherin KD). e IP for IgG and EGFR in TKO HCC cells expressing control (CTRL) or shβcat3 (β-cadherin KD). EGFR, E-cadherin, and β-catenin were detected in both unbound and pull-down fractions by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. f IP of IgG and β-catenin in TKO HCC cells treated with 0.5 mM orthovanadate. β-catenin, EGFR, and E-cadherin were detected by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. g P-EGFR (Tyr1068, activated EGFR), EGFR, and EGFR2 were detected in control livers (1-6), TKO HCC tumors (7)(8)(9)(10)(11)(12)(13)(14)(15), and serially transplanted TKO HCC (Serial) tumors (16-23) by immunoblot. β-Catenin immunohistochemistry (IHC) is provided as a reference for β-catenin subcellular localization. Ponceau serves as a loading control. h Immunoblot for EGFR in TKO HCC cells expressing control (CTRL) or shβcat3 (β-catenin KD). The cells were treated with 25 μg/ml cycloheximide (CHX) in the presence (upper panel) or absence (lower panel) of 20 ng/ml EGF for the time indicated (upper: 15-60 min; lower: 0-24 h). Ponceau serves as a loading control. i IP of Flag epitope in TKO HCC cells expressing control (CTRL) or shβcat3 (β-cadherin KD) as well as ectopic Flag-Ubiquitin. EGFR, E-cadherin, and β-catenin were detected by immunoblot. j IP of EGFR and Ecadherin in TKO HCC cells expressing control (CTRL) or shβcat3 (β-cadherin KD), as well as ectopic Flag-Ubiquitin. EGFR, E-cadherin, and Flag were detected by immunoblot. Double arrowheads indicate two migrating forms of E-cadherin. n.s. not significant. See also Supplementary Fig. 7 penicillin-streptomycin-glutamine. L cells (ATCC ® CRL-2648) were cultured in DMEM with 10% FBS (with the addition of 0.4 mg/ml G-418, in case of L-Wnt3A (ATCC ® CRL-2647)). Cell lines were tested for mycoplasma detection using an inhouse PCR-based methods. For organoid culture, primary TKO HCCs were harvested and digested with an enzymatic cocktail to obtain a single-cell suspension. Organoids were generated and maintained as recently described 37 . For viral infection, shRNA sequences were generated using psicoligomaker [61][62][63] and cloned into the pSicoR-PGK-RFP and psicoR-CMV-GFP vectors. Viral infection was performed as previously described 31 . To determine protein stability, cells were treated with 25 μg/ml CHX (Millipore) 24 h after plating, and harvested after the indicated treatment time period prior to immunoblotting. For Axin stabilization, TKO cells were treated with 5 μM XAV939 (Selleckchem) for 48 h 39 . For Wnt3A   media treatment, the recipient cells were treated with media from L-Wnt3A cells mixed with fresh media at 1:1 ratio. For EGF treatment, serum-starved cells were stimulated with 20 ng/ml EGF for 15 min-1 h. For EGFR inhibition/multi-kinase inhibition, the cells were plated in a 6-well plate at 5 × 10 5 cells/well in the presence of DMSO or 0.1, 1, 10, or 50 μM and Afatinib, Erlotinib, or Sorafenib for 24 h and were harvested for analysis. For orthovanadate treatment, the cells were treated with 0.5 mM orthovanadate for 2 h and were harvested for analysis. To generate the growth curves, the cells were plated in triplicate in a 6-well plate at 5 × 10 4 cells/ well and counted every 24 h for 5 days. Cells collected on days 0 and 5 were processed for RNA analysis and immunoblotting. For AnnexinV staining, the cells were plated in triplicate in a 6-well plate at 1 × 10 5 cells/well and cultured for 3 days prior to AnnexinV (Roche) analysis. For BrdU labeling, BrdU (1:100) were added to the media 1 h prior to harvest.