RNF43/ZNRF3 loss predisposes to hepatocellular-carcinoma by impairing liver regeneration and altering the liver lipid metabolic ground-state

RNF43/ZNRF3 negatively regulate WNT signalling. Both genes are mutated in several types of cancers, however, their contribution to liver disease is unknown. Here we describe that hepatocyte-specific loss of Rnf43/Znrf3 results in steatohepatitis and in increase in unsaturated lipids, in the absence of dietary fat supplementation. Upon injury, Rnf43/Znrf3 deletion results in defective hepatocyte regeneration and liver cancer, caused by an imbalance between differentiation/proliferation. Using hepatocyte-, hepatoblast- and ductal cell-derived organoids we demonstrate that the differentiation defects and lipid alterations are, in part, cell-autonomous. Interestingly, ZNRF3 mutant liver cancer patients present poorer prognosis, altered hepatic lipid metabolism and steatohepatitis/NASH signatures. Our results imply that RNF43/ZNRF3 predispose to liver cancer by controlling the proliferative/differentiation and lipid metabolic state of hepatocytes. Both mechanisms combined facilitate the progression towards malignancy. Our findings might aid on the management of those RNF43/ZNRF3 mutated individuals at risk of developing fatty liver and/or liver cancer.

Rnf43 and Znrf3 are two negative-feedback regulators of the WNT pathway 17,18 . Both are E3-ubiquitin ligases that specifically ubiquitinate the cytoplasmic loops of the FZDs receptors, which induces their rapid endo-lysosomal degradation 17,18 . Conversely, R-spondins (RSPO), ligands of LGR4-5-6 receptors, also interact with RNF43/ZNRF3, which reverses the RNF43/ZNRF3-mediated membrane clearance of FZDs boosting WNT signal strength and duration 19 . Mutations in RNF43 and/or ZNRF3 have been observed in human cancers 20 . We had reported that intestinal deletion results in adenoma formation 18 . In primary liver cancer, ZNRF3 is mutated in hepatocellular carcinoma (HCC) 21 , while RNF43 in intrahepatic cholangiocarcinoma 22 . However, their role in liver disease has not been explored yet. Ubiquitous (wholebody) deletion of both Rnf43/Znrf3 results in some liver metabolic changes 12 ; however, whether these genes exert a direct regulation on liver metabolism or whether these result from secondary effects on non-liver tissues has not been elucidated yet.
Here, we describe that specific deletion of RNF43/ZNRF3 in adult hepatocytes results in liver degeneration, steatohepatitis, increased unsaturated lipids and altered liver lipid zonation. Concomitantly to these alterations, Rnf43/Znrf3 loss increases hepatocyte proliferation. Both alterations are explained, at least in part, by cell-autonomous mechanisms, since Rnf43/Znrf3-null liver epithelial organoids exhibit significant intracellular lipid droplet accumulation and reduced differentiation capacity. Upon chronic injury, the lipotoxicity and imbalance in proliferation/ differentiation states contribute to a defective regeneration, extensive fibrosis and tissue damage, which, with time, progresses to malignancy, either HCC or mixed HCC/CC cancer. Notably, our mutant mouse recapitulates the signatures of HCC patients with mutations in RNF43 and/or ZNRF3, which also present lipid metabolic alterations and poorer prognosis. Our observations imply that RNF43/ZNRF3 predispose to liver cancer, at least in part, by altering the hepatocyte metabolic ground-state while, in parallel, preventing the tissue from completely regenerating upon damage. Our results are in agreement with our previously published preprint manuscript (Mastrogiovanni et al. 23 ) and the recent publication by Sun et al., published while our manuscript was under revision 24 .

Results
Rnf43/Znrf3 loss results in hyperplasia and tissue degeneration.
Following recombination, we collected Rnf43/Znrf3 del and control livers at 1 week and 3 months after deletion (Fig. 1a). Rnf43/Znrf3 del mice exhibited significant hyperplasia and hepatomegaly at 1 week (Fig. 1b), which persisted throughout the course of the study, resembling the hepatocyte-specific Apc mutant mice 16 . This correlated with a significant increase in the number of proliferating hepatocytes (Ki67 + ) ( Supplementary  Fig. 1d, e). At the histological level, we noted that at 3 months, but not at 1 week, double Rnf43/Znrf3 del mutant livers presented signs of cellular degeneration (hepatocyte ballooning, decellularization) ( Supplementary Fig. 1d, haematoxylin and eosin (H&E) panel arrowhead), which coincided with a significant increase in the number of apoptotic (Cleaved Caspase-3+) and necroptotic (p-MLKL+) positive hepatocytes ( Supplementary Fig. 1f). In addition, we observed a significant increase in innate (NK and macrophages) as well as adaptative (T cells, CD3+) inflammatory cells ( Supplementary Fig. 1g). Notably, the mutant livers presented foci of fat deposition, as assessed by Oil Red-O staining (Fig. 1c). Tissue degeneration, inflammation, fat deposition and ballooned hepatocytes are observed in a variety of acute and chronic liver diseases 26 , including steatohepatitis/nonalcoholic steatohepatitis (NASH; inflamed fatty liver) [26][27][28] . Hence, we formally determined the levels of steatohepatitis in our mice by obtaining the non-alcoholic fatty liver disease (NAFLD) Activity Score (NAS score, Supplementary Fig. 2a) 29 . NAS score indicated that double mutant mice presented a probable/borderline score, with 1 mouse scoring >5 (steatohepatitis) ( Fig. 1d and Source data). Individual homozygous mutants and wild-type (WT) littermate controls did not present any signs of hepatomegaly, neither discernible histological architecture changes (inflammatory foci, fat accumulation, tissue degeneration) nor steatohepatitis (NAS score <2) ( Supplementary Fig. 1h-j). Hence, we opted to not study the single mutants any further.
The significant increase in the number of proliferating hepatocytes, combined with tissue damage features (inflammation and cell death) and probable steatohepatitis at 3 months post deletion, prompted us to investigate the impact of long-term depletion of both genes. We opted to investigate a maximum of 7 months post-deletion, since this represented a total mouse age of a maximum of 10 months and prevents the occurrence of confounding factors due to ageing, and the well-documented agedinduced senescence changes that arise in middle-aged and old-aged mice 30 . At this later time point, Rnf43/Znrf3 del livers presented exacerbated hepatocyte proliferation (Ki67) and cellular degeneration with clear cell nodules formed by hepatocytes with nuclear alterations (crenation and karyorrhexis) and ballooned, and a significant number of apoptotic (Cleaved Caspase-3+), necroptotic (p-MLKL+) and senescent (P21+) hepatocytes, in agreement with the significant elevation of the liver serum parameters aspartate transaminase (AST) and alkaline phosphatase (ALP) (Fig. 1e-i). In addition, 50% of the mice exhibited steatohepatitis (NAS score >5) while the remainder scored probable/borderline (NAS score 3-4) ( Fig. 1d and Supplementary Fig. 2a). Of note, NASH in humans refers to a disease that develops in the context of obesity and systemic dysmetabolic alterations. However, none of the mice exhibited any of these features. Instead, body weight, triglycerides (TG) and cholesterol levels were under the normal range at 7 months post deletion (Supplementary Table 1). Remarkably, at this later time point, 10% of the mice (2 out of 20) had developed adenomas ( Supplementary Fig. 2b), suggesting progression towards malignancy. As expected, mutant livers also presented a marked increase in the Wnt/β-catenin liver zonated genes (GS and CYP2E1) ( Supplementary Fig. 2c, d) and higher levels of β-CATENIN around the central vein area (Supplementary Fig. 2e). None of these histological changes was detected in any of the control WT littermates (non-tamoxifen or control-virus treated) ( Supplementary Fig. 2f).
RNF43/ZNRF3 control liver lipid metabolic gene expression. We next performed global gene expression analysis (RNA sequencing (RNAseq)) of Rnf43/Znrf3 del and corresponding controls (Rnf43/Znrf3 flox ) at several time points after gene deletion ( Fig. 3b and Supplementary Data 2). Of note, pericentral and periportal genes were upregulated and downregulated, respectively, compared to WT mice ( Supplementary  Fig. 3c), suggesting that liver zonation was inverted and confirming the reported role of canonical WNT signalling in the maintenance of liver zonation 12,15 . We then overlapped our DE gene list with a published list of liver TCF4 targets obtained by chromatin immunoprecipitation-sequencing 15 . We found that 11.6% of our DE genes (388 of 3334 genes) were canonical liver WNT/TCF4 targets, such as Axin2, Sp5, Cyp2e1 and Tnfrsf19, among others (Fig. 2c, Supplementary Fig. 3d and Supplementary Data 3_S3 and S9). More interestingly, both IPA and GO analysis revealed that a significant proportion of the upregulated genes were involved in lipid/phospholipid metabolism, being "lipid metabolic process" the topmost significantly enriched category at 3-months post-deletion (Fig. 2b, Supplementary Fig. 3a and Supplementary Data 1 and 2). To further investigate the lipid metabolic changes observed, we generated a gene list of lipid metabolic genes (see methods) and overlapped it to our DE gene lists (3334 genes). We found that~12.5% of all the DE genes (416 out of 3334) were involved in lipid metabolic processes (Fig. 2c). Within these, we found genes encoding for enzymes involved in the synthesis and remodelling of fatty acids including Fasn (fatty acid synthase), Fads1 and Fads2, which regulate unsaturation of fatty acids through the introduction of double bonds on the fatty acyl chain and elongases, like Elovl2, Elovl3 and Elovl6, which add 2 carbons to the long-and very-long-chain fatty acids. In addition, we found significantly upregulated genes involved in diacylglycerol (DAG) biosynthesis (i.e. Mogat1, Lpin2 and Lpin1) and in the biosynthesis of TG and phospholipids, such as Dgat2 (synthesis of TG from DAG) and Etnk2 (which catalysers the first step of phosphatidylethanolamine (PE) biosynthesis). Also, genes involved in lipid transfer (Pltp and Lpl) and TG hydrolysis (Pnpla2, which generates DAG from TG) were also significantly upregulated. Last, we found upregulation in the expression of genes involved in cholesterol and bile acid metabolism, such as the oxysterol sensor Liver X receptor alpha (Nr1h3) and its target genes Cyp7a1 (the rate-limiting step enzyme in cholesterol conversion into bile acids), Abcg8 (cholesterol secretion in bile), as well as the phospholipid transporter Abcb4 (i.e. MDR2). Conversely, genes involved in phospholipid oxidation (Pla2g4a) and cholesterol hydroxylation (Hsd11b1) were within the most downregulated genes (Fig. 2d, Data 4). These results suggested the remodelling of the fatty acyl chain composition of phospholipids and diacylglycerides and were in agreement with the gene expression signatures (Fig. 2), that had shown a significant upregulation of the genes involved in TG, DAG and PC and PE biosynthesis and elongases (e.g. Dgat2, Etnk2, Fads1 and Fads2, Elovl2, among others).
Finally, IF analysis of LPCAT2 (an enzyme that incorporates 20:4 into PC), revealed that the intensity and spatial area of LPCAT2 were increased in Rnf43/Znrf3 del ( Supplementary Fig. 4b).
Taken together, the lipidomic and lipid imaging analyses point towards a clear alteration in the liver lipid composition and lipid distribution across the tissue, with a significant increase in the abundance of unsaturated lipids, likely caused by upregulated lipid desaturase enzymes.
To investigate whether part of the lipid alterations observed was due to a cell-autonomous effect of Rnf43/Znrf3 in hepatocyte lipid metabolism, we studied the impact of deleting both genes in hepatic cells ex vivo. For that, we took advantage of liver organoid cultures that enable the expansion of mutant hepatocytes derived from mouse embryonic 32 and adult 33 tissue in the absence of stromal niche. Hence, we isolated Rnf43/ Znrf3 flox hepatocytes or hepatoblasts and then deleted Rnf43/ Znrf3 to generate Rnf43/Znrf3 del mutant cultures and corresponding parental (non-deleted) controls ( Fig. 3g and methods) and analysed their lipid content by staining for neutral lipids. In both models, we observed a significant increase in the number and size of the lipid droplets in the Rnf43/Znrf3 mutant cultures, which was completely reverted upon treatment with WNT inhibitors (Fig. 3h, i and Supplementary Fig. 4c, d). Notably, treatment of parental WT (Rnf43/Znrf3 flox ) hepatocyte organoids with WNT3A ligand also resulted in a significant increase in the number and size of lipid droplets (Fig. 3j, k).
In summary, the combination of the transcriptional, lipidomics and lipid imaging analysis and organoid data reveals a role for RNF43/ZNRF3 as gatekeepers of the lipid metabolic ground-state in hepatocytes, through the combination of both cell-autonomous (direct intracellular accumulation) as well as indirect (e.g., microenvironmental/zonation changes) mechanisms, such as the expansion of the metabolic programmes of the pericentral hepatocytes.
Rnf43/Znrf3 mutant mice exhibit regeneration defects. Our observations that Rnf43/Znrf3 livers exhibit a significant degree of parenchymal degeneration, steatohepatitis and increased unsaturated lipids-also associated with inflammation and NASH 34prompted us to hypothesise that the homoeostatic turnover of the hepatic parenchyma might be impaired in Rnf43/Znrf3 del mice. Hence, we decided to investigate the role of RNF43/ZNRF3 during damage-regenerative response. For that, we challenged Rnf43/Znrf3 del mice with repeated doses of carbon tetrachloride (CCl 4 ) 35 , in order to induce chronic liver damage ( Fig. 4a and Supplementary Fig. 5a).
As a proxy for liver damage, we quantified the levels of transaminases in serum before the first dose of CCl 4 (day −40), during damage (day −12, middle of the regimen, after 7× doses of CCl 4 ) and at several time points after the last dose (day 4, 35 and 50 of recovery, which correspond to 1.4, 2.5 and 3 months post-deletion). We observed significantly higher levels of ALT, AST and ALP in our mutant mice compared to WT littermates undergoing the same damage regimen, which persisted even when control mice had fully recovered ( Supplementary  Fig. 5b). In addition, mutant livers were significantly bigger and presented multiple regenerative and pre-neoplastic nodules not observed in non-damage controls ( Supplementary Fig. 5d, e). Histological analysis revealed hallmarks of prominent tissue damage, including decellularized regions, significant fibrosis and extensive ductular reaction (Fig. 4b, c and Supplementary  Fig. 5f, g). Senescence (p21), apoptotic (cleaved caspase-3) and damage (HMGB1) markers confirmed that CCl4-treated Rnf43/ Znrf3 del livers were extensively damaged (Fig. 4d, e and Supplementary Fig. 5h). Expectedly, we also found a significant increase in the GS staining area after 50 days of recovery, but not after 4 days, when the cells expressing this marker are highly damaged by the CCl 4 ( Supplementary Fig. 5i). These prominent tissue damage alterations (decellularization, fibrosis, cytoplasmic HMGB1 staining) were not detected in control littermates, which had mostly recovered already at day 4 after the last CCl 4 injection ( Fig. 4b-e and Supplementary Fig. 5b, d-i) neither in mutant livers not exposed to damage at the same time point (3M, Supplementary Fig. 1).
To gain further insights, we performed RNAseq analysis of the CCl 4 -damaged Rnf43/Znrf3 del and Rnf43/Znrf3 flox livers at days 4 and 50 of recovery. We found a significant number of genes DE (2989 at day 4 and 1438 at day 50) including several metalloproteases (Mmp7, Mmp11, Mmp2, Mmp23) and components of the caspase-mediated apoptosis signalling pathway (Casp12, Pycard) (Supplementary Data 1_S16-S19). In addition, we found an increased number of proliferative (Ki67+) cells (Supplementary Fig. 6a-b), in agreement with the reports of active β-catenin 36 and liver-specific APC mutant mice 37 . GO term and gene set expression analysis (GSEA) confirmed that many biological processes involved in proliferation were activated at both time points, including "regulation of cell growth", "proliferation" and "wound healing". Interestingly, we found significant enrichment for liver cancer data sets, as well as data sets associated with tissue and DNA damage and inflammation (Fig. 4f, g and Supplementary Data 1_S38-41).
Altogether, the serum, histological and transcriptional findings indicated that at the molecular and histological level, Rnf43 and Znrf3 (R&Z) null hepatocytes present lipid metabolic transcriptional changes. a-f Liver tissues from 1-3-and 7-month-old R&Z del mice and R&Z flox littermates were collected and processed for RNAseq analysis. Differentially expressed (DE) gene profiles were obtained as described in "Methods." a Principal component analysis (PCA) of R&Z flox and R&Z del . Each data point represents one sample. Note that PC1 is strongly correlated with gender, whereas PC3 separates R&Z flox mice from R&Z del mice. PC2 corresponded to the two batches (batch1 and batch 2) from which the data were generated. b Graphs showing top 3 gene ontology (GO) terms significantly enriched for genes upregulated (red) and downregulated (purple) in R&Z del compared to R&Z flox . Full list in Supplementary Data 1. c Venn diagram showing a correlation between genes involved in lipid metabolism (brown), TCF4 target genes (green) and genes DE at any time point in R&Z del livers (purple). The numbers denote the number of genes in each comparison. Details are given in Supplementary Data 3. d, e Heatmaps of the lipid metabolism genes DE in R&Z del livers. d Genes DE averaged between all mice per time point and ranked by fold change with respect to their respective R&Z flox control. e DE genes showing some representative lipid metabolic genes. Likelihood ratio test and Benjamini-Hochberg correction; Fasn, 3M *p = 0.0446, 7M ***p = 0.0001; Fads1, 3M ***p = 0.0006; Fads2, 1M ***p = 0.0038, 3M ***p = 0.0001, 7M *p = 0.0282; Elovl2, 3M *p = 0.0392, 7M **p = 0.0088; Elovl6, 3M **p = 0.0020, 7M **p = 0.0088. f Heatmap of the RPKM values (raw z-scored) of DE lipid metabolic genes found to be TCF4 targets. Each column, one independent biological replicate. Source data are provided as a Source data file.
Rnf43/Znrf3 del livers were still undergoing a damage-regenerative response at day 50 of recovery, when the WT tissue had fully regenerated. Notably, a single acute dose of CCl 4 resulted in similar levels of damage in both genotypes ( Supplementary  Fig. 5c), suggesting that at baseline both genotypes have a similar response to damage.
Rnf43/Znrf3 mutants present defective proliferation/differentiation states. We next studied the mechanism for defective regeneration. We investigated whether loss of Rnf43/Znrf3, and the subsequent WNT activation, directly impact the proliferation/ differentiation state of hepatocytes or whether the impaired regeneration observed was a consequence of the damage     (Fig. 5c), supporting a scenario whereby the absence of Rnf43 and Znrf3 "locks" hepatocytes in a proliferative state. RNAseq analysis on Rnf43/Znrf3 flox and Rnf43/Znrf3 del livers at 7 and 21 days after PHx revealed that many of the genes altered by the damage corresponded to molecular signatures of cell proliferation as well as liver cancer ( Supplementary Fig. 6e, f), with canonical WNT signalling pathway as the top upstream regulator ( Supplementary Fig. 6g). When integrating the data from both damage models (PHx and CCl4-CD) we identified 92 unique genes that were commonly upregulated only during recovery (day 21 after PHx and CD after 4 days), whose GO terms corresponded to cell cycle, proliferation and cell division, confirming that the absence of Rnf43/Znrf3 results in a hyperproliferative state in hepatocytes, regardless of the damage type (Fig. 5d, e and Supplementary Data 5).
Taken together the data suggested a model whereby Rnf43/ Znrf3 mutant hepatocytes once they activate a proliferative state cannot exit from it to complete differentiation and maturation, which, consequently, impairs the tissue to terminate the regenerative process. To investigate this hypothesis, we performed GSEA between CCl 4 -damaged Rnf43/Znrf3 del livers and a signature of mature hepatocytes 38,39 (Supplementary Data 1_S37). We observed a significant negative enrichment against hepatocyte signatures, indicating that the hepatocyte cell fate was downregulated ( Fig. 5f and Supplementary Fig. 6h). Detailed inspection of the significantly DE genes revealed that several wellknown hepatocyte markers were significantly downregulated in CCl 4 -treated mutants compared to controls, including genes involved in the complement complex (C8a, C8b), coagulation factors (F7 and F9) and basic hepatocyte functions (Baat, Tat, Arg1, Alb, Hal), among others. Notably, similar results were obtained when analysing the livers of mutant mice at 7 days and 21 days after PHx (Fig. 5g), a model of compensatory hyperplasia not associated with massive necrosis but to the activation of hepatocyte proliferation 40 . Both results combined indicate that it is the imbalance between proliferation/differentiation in Rnf43/ Znrf3 mutant hepatocytes that explains the defective termination of the regenerative programme and persistence of proliferative hepatocytes, far beyond the end of regeneration phase is terminated in WT cells.
To determine whether the differentiation capacity of the Rnf43/Znrf3 del hepatocytes was reduced due to a cell-autonomous effect, we studied the impact of deleting Rnf43/Znrf3 in hepatocyte cells in vitro, in organoid cultures. For that, we deleted both genes in organoids derived from either adult hepatocytes 33 or our organoid culture system whereby progenitors, either hepatoblasts 32 or adult liver ductal cells differentiate into hepatocyte-like cells 41 (Fig. 5h-l and Supplementary Fig. 6i, j). In agreement with the mouse data, Rnf43/Znrf3 del organoids exhibited increased proliferation and reduced differentiation capacity than control WT organoids, as assessed by quantitative polymerase chain reaction (qPCR) analysis of common hepatocyte markers such as Ttr, Cyp3a11, Pparg, Alb and Hnf4a and WNT target genes such as Lgr5 and Axin2, or the proliferation marker mKi67 (Fig. 5j). To investigate the direct role of WNT hyperactivation in the differentiation defect, we treated Wild-Type organoids with either Wnt-inhibitors (IWP2 42 and iCRT3 43 ) or WNT3A ligand, to completely inactivate or mimic the hyperactivation of the pathway. The blockade of WNT signal resulted in an increased differentiation ability of the adult hepatocytes. Conversely, supplementing WNT3A in the media promoted an undifferentiated and highly proliferative state, mimicking the Rnf43/Znrf3 del phenotype (Fig. 5k, l). Similarly, Wnt inhibition of mutant organoids also partially rescued the differentiation defect ( Supplementary Fig. 6j). As expected, Wntinhibition treatment abolished the expression of the Wnt targets Lgr5 and Axin2, while Wnt3a upregulated them (Fig. 5k).
Collectively, these results imply that loss of RNF43/ZNRF3 activates the proliferative programme of hepatocytes while at the same time impairs their ability to terminate differentiation, which combined with the extensive tissue degeneration and changes in the cellular microenvironment, could lead to eventual liver fibrosis and the degenerative tissue phenotype observed upon chronic injury.
RNF43/ZNRF3 deletion predisposes to liver cancer. To assess whether the steatohepatitis/lipotoxicity and malfunctioning regenerative capacity of Rnf43/Znrf3 del mice would progress into a malignant HCC we subjected the mice to chronic injury and collected the livers~5 months later (170 days after the last dose of  :2)]. Data are represented as mean + /− SD of n = 3 (3M) and n = 4-5 (7M) mice per group. Unpaired two-tail t test. 3M, *p = 0.0103; 7M, **p = 0.0059. Right, the correlation between PC Fads2 index and Fads2 RNA expression. Pearson correlation, r = 0.69, CI 95% (0.2597-0.8952), two-tailed p = 0.005. f Mass spectrometric imaging analysis shows increased intensity and expansion of area covered by PC(36:4) relative to PC(32:1) and PC(34:2) in R&Z del liver. g-k Isolated adult hepatocytes derived from either AlbCreERT-R&Z flox or R&Z flox livers were expanded and treated or not with hydroxy-tamoxifen (OH-Tam) or AAV8-TGB-Cre infection to induce Rnf43/ Znrf3 deletion in vitro. The derived organoid lines and parental non-recombined lines were treated with Wnt inhibitors (Wnti, h, i) or WNT3A conditioned media (Wnt3a, j, k) and analysed for lipid droplet content by Bodipy staining. g Experimental design. h-k Bodipy staining and quantification of R&Z flox and R&Z del hepatocyte organoids grown in control medium or medium supplemented with Wnti. h Representative pictures from n = 2 independent experiments. Scale bar, 100 μm. i Image analysis quantification of the intracellular lipid droplets size and content (n = 5). Two-way ANOVA with Tukey's multiple comparisons. Small particles, R&Z flox vsR&Z del ***p = 0.0001; R&Z del vsR&Z del + Wnti ***p = 0.0001. Big particles, R&Z del vsR&Z del + Wnti *p = 0.0198. j, k Bodipy staining and quantification of R&Z flox wild-type hepatocyte organoids grown in control or WNT3A media. j Representative pictures from n = 2 independent experiments. Scale bar, 100 μm. k Image analysis quantification of the intracellular lipid droplets size and content (n = 5), two-way ANOVA, *p = 0.0000. Source data provided in Source data file.
CCl 4 , 7 months after deletion) (Fig. 6a). We noted the appearance of tumoral nodules of different sizes already at the macroscopic examination of the tissue (Supplementary Fig. 7a). Histopathological analysis confirmed that these were HCC (HCC) or HCC with ductal features (CHC, mixed subtype), with a combined penetrance of 89% (8 out of 9 mice) ( Fig. 6b and Supplementary   Fig. 7b-d). HCC lesions were characterised by small cells with high nuclei/cytoplasmic ratio, a compact pattern without the trabecular formation and 2-3 cell thick plates, fat accumulation, a significant number of proliferative (Ki67 + ) cells, neovascularisation and disruption of tissue histoarchitecture as evidenced by CD34 staining, Collagen IV deposition and stroma   Fig. 2b).
To better characterise the HCC lesions, we microdissected the tumour and either processed them for molecular analysis or cultured them in our previously described human primary liver cancer organoid method 44 . Molecular analysis indicated that the HCC tumours were indeed devoid of Rnf43/Znfr3 and expressed high levels of Axin2, confirming the Rnf43/Znrf3 del origin of the tumour lesion ( Supplementary Fig. 7i). Interestingly, HCC lesions but not the background liver expanded as tumour organoids in vitro by forming compact structures composed of tumoural cells that filled in the lumen of the organoid and expressed high levels of WNT target genes (Axin2 and Sp5) ( Fig. 6d and Supplementary Fig. 7j). Of note, Rnf43/Znrf3 del tumour organoids, but not WT controls, grew long-term (up to passage 10) in the absence of RSPO1, confirming their origin from Rnf43/ Znrf3 del cells (Fig. 6e).
RNAseq analysis identified 214 DE genes between mutants and WT littermates subjected to the same regimen and recovery time. GO and GSEA analysis revealed enrichment on fatty acid metabolism, and several cancer signatures, including tumours with β-catenin activation 45 , hepatoblastoma 46 , cancer patients with NASH 47 as well as advanced cancer and metastatic cancer data sets, and inflammation and cholesterol uptake data sets (Fig. 6f-h and Supplementary Data 1_S18). Similar enrichment was also detected at an earlier time point (at 50 days recovery) ( Supplementary Fig. 7l).
Altogether, this data indicated that RNF43/ZNRF3 loss predisposes to cancer due to altered lipid metabolism and defective tissue regeneration (hepatocyte maturation) upon cumulative rounds of tissue damage.
RNF43/ZNRF3 patients present altered lipid metabolism and poorer prognosis. To determine the relevance of our findings in human liver cancer, we next assessed the effect of RNF43/ZNRF3 mutations in human primary liver tumours taking advantage that the projects LIRI-JP and LICA_FR from the ICGC collection 48,49 , contained genomic and expression data of the same tumours (Fig. 7a). Within the somatic variant mutations that passed our filtering criteria, we found mutations in both coding as well as non-coding regions (mainly intron variants, see methods and Supplementary Data 6_S1). We used these patients to study the prognostic value of the two genes, for those samples where survival and clinical information was available (n = 257 donors from LIRI-JP data set, "Methods"). Using multivariate survival analysis accounting for WNT mutation, RNF43 or ZNRF3 mutation, double RNF43 /ZNRF3 mutation, donor sex and clinical tumour stage, we observed a significant difference in the effect of RNF43 or ZNRF3 mutation on survival depending on whether or not a mutation in WNT was also present (interaction term p value 0.049, Supplementary Data 6_S13). Hence, we next considered the effect of RNF43/ZNRF3 conditional on WNT mutation status. We separated the groups by those containing WNT mutations and those not (-CTNNB1, APC, AXIN1 mutations, see "Methods"). Interestingly, for those without WNT mutations, we observed that patients bearing ZNRF3 tumours had a significantly poorer prognosis than patients with WT tumours (p = 0.026). This result was independent of clinical features, sex and tumour stage. A similar trend was observed when evaluating patients harbouring either ZNRF3 or RNF43 mutations, although this was not statistically significant (p = 0.06) ( Fig. 7b and Supplementary Data 6_S14-S15). Conversely, neither ZNRF3 alone nor ZNRF3 or RNF43 showed altered prognosis when patients had WNT mutations ( Supplementary Fig. 8b). For single RNF43 or double mutations, the number of patients was too low to provide meaningful conclusions. The effect of WNT mutation across all patients was not significant (p value 0.32) consistent with what had been previously reported 21 (Supplementary Data 6_S13). Of note, we did not find reported liver cancer driver genes shared between all RNF43/ZNRF3 patients (Supplementary Data 6_S12). We next performed differential expression analysis of the livers of patients harbouring WNT mutations (WNT), RNF43 and ZNRF3 (R&Z) and RNF43 or ZNRF3 (RorZ) compared to liver cancer patients with WT alleles for these. Among the DE genes, we found lipid metabolic genes, including LPCAT2 and APOA4 (Supplementary Data 6_S16-S18). GSEA analysis of the DE genes for the RNF43 and/or ZNRF3 groups revealed enrichment in lipid metabolism. In addition, the group containing only RNF43 or ZNRF3 mutations was also enriched on gene sets of "NASH", "simple steatosis", "Fatty acid metabolism", "cholesterol biosynthesis" and "adipogenesis" (Supplementary Fig. 8c and Supplementary Data 6_S25-S30). Interestingly, we noted that the group of patients bearing WNT mutations, but not mutations in RNF43 or ZNRF3, also had a significant enrichment in some of these lipid metabolic pathways (Supplementary Fig. 8c, WNT). Hence, we investigated the potential interaction between WNT mutations and RNF43 or ZNRF3 using a multiple regression model. The results indicated that there is an impact of WNT mutations in the Fig. 4 Liver-specific Rnf43/Znrf3 deletion leads to defective tissue regeneration after chronic CCl 4 treatment. a-g Adult AlbCreERT2xRnf43/Znrf3 flox mice received tamoxifen to generate Rnf43/Znrf3 null livers. After a washout period, CCl 4 was injected twice a week for a total of 6 weeks and mice were collected at 4, 35 and 50 days after the last injection. These time points are equivalent to 1.5, 2.5 and 3 months post recombination, and hence comparable to the 1 month and 3 months undamaged mutant mice. Tamoxifen was re-administered to prevent expansion from un-recombined hepatocytes, which ensured a >80% recombination of both alleles (See Supplementary Fig 5). a Experimental design. T tamoxifen, C CCl 4 . b Representative pictures of H&E staining of n = 4 (CD 4d) and n = 3 (CD 50d) independent replicates. Histopathological analysis revealed increased tissue damage as evidenced by the presence of giant hepatocytes (arrowhead), de-cellularised areas (arrow) and regenerative nodules (dashed line) in Rnf43/Znrf3 mutant livers compared to littermate controls. Scale bar, 100 μm (top) and 50 μm (bottom). c Collagen deposition (Pico-Sirius red staining) in mutant livers after damage indicate mild fibrosis. Representative pictures of n = 4 (CD 4d) and n = 3 (CD 50d) independent biological replicates. Scale bar, 500 μm. Quantification of the fibrotic area is given in Supplementary Fig 5f. d, e Cleaved caspase-3 (d) and senescence marker p21 (e) stainings at 50 days of recovery from chronic damage. Representative images of n = 3 independent replicates. Scale bar, 100 μm. f Graph showing selected gene ontology (GO) terms significantly enriched for genes upregulated (red) and downregulated (purple) in R&Z del compared to R&Z flox after damage. Numbers on the bars denote the number of genes associated with each term. Full list in Supplementary Data 1_S29-S32. g Heatmap of the normalised enrichment score (NES) of selected GSEA data sets at 4 and 50 days of recovery. Red, gene sets of apoptosis and senescence. Full list in Supplementary Data 1_S38-S45. Source data in Source data file. expression pattern of RNF43 or ZNRF3 (Fig. 7c, compare RorZ vs RorZ+Wnt, Supplementary Data 6_S21-S23), similar to our observation with the survival analysis. Hence, to gain an understanding of the impact of RNF43/ZNRF3 in lipid metabolism in human patients without the cofounding factor of other WNT mutations, we re-analysed the differential expression of  "unsaturated fatty acids", "Acyl-CoA biosynthesis" and "NASH" were now also significantly enriched in the patients harbouring mutations in both genes (RNF43 and ZNRF3 group) (Fig. 7d).
Notably, those DE genes were enriched in TCF4 targets, with the patients bearing double mutations exhibiting a stronger enrichment compared to the patients bearing single mutations, when compared to non-WNT mutant patients ( Supplementary Fig. 8d). These results were in agreement with our mouse transcriptomics (Figs. 2 and 6g and Supplementary Fig. 3a) and lipidomics data (Fig. 3). Further, we found a strong significant enrichment between the human signatures and the DE genes in our mutant mice (Fig. 7e), when considering both canonical WNT signature (TCF4 targets) as well as TCF4-independent signatures, suggesting that WNT and additional, non-WNT mechanisms may contribute to the mutant phenotype, both in human and mouse. To investigate whether lipid accumulation was also cellautonomous in cancer patients, we analysed the lipid droplet content in human liver cancer tumour organoids mutated in these genes. In our collection, only one line (CHC2 line) 44 harboured a homozygous (deleterious) RNF43 mutation (Leu418-Met). We found that RNF43 mutant organoids accumulated intracellular lipid droplets, while organoids derived from liver cancer patients without mutations in any of the two genes (CHC-1) or donor-derived organoids (derived from non-cancer tissue) presented fewer intracellular lipid droplets (Fig. 7f). These results suggested that, at least in part, lipid droplet accumulation is cellautonomous in human patients.
In summary, our results indicated that human liver cancer patients bearing ZNRF3 and/or RNF43 mutations present altered liver lipid metabolism, in agreement with our findings in the Rnf43/Znrf3 mutant mouse model, and identified ZNRF3 and/or RNF43 as potential prognostic biomarkers for liver cancer.

Discussion
Cancer genomic studies have identified RNF43 and ZNRF3 as WNT pathway genes mutated in liver cancer 21,22,50 . However, their implication in liver regeneration and or liver disease remained unexplored. Here we find that loss of RNF43/ZNRF3 predisposes to liver cancer by controlling liver lipid metabolic ground-state and the balance between proliferation/differentiation in hepatocytes. Hence, both homologues behave as landscaper genes, which contribute to neoplastic transformation by altering the tissue environment 51 . Rnf43/Znrf3 mutant mice presented steatohepatitis and aberrant liver lipid composition, with an increase in unsaturated lipid species, in the absence of dietary fatty acid supplementation. Concomitantly, mutant livers exhibited increased proliferation and impaired regeneration capacity, which following chronic injury, progressed to liver cancer, both HCC and mixed eHCC/CC.
Our findings might seem at odds with reported results whereby WNT3A injection prevents lipid accumulation in mice with active non-canonical WNT through an LRP6 receptor mutation (LRP6 R611C ) 52,53 and with the phenotype of the core canonical WNT component APC in the liver, which results in HCC without the need of any additional hit or tissue damage 37 . However, recent reports have demonstrated a link between WNT signal strength and pathway outcome, suggesting that the strength of activation of the pathway as well as the nuclear betacatenin fold change, rather than absolute levels, are critical for target gene activation 54 , stem cell behaviour or malignant transformation [55][56][57] . Our studies where WT organoids are supplemented with WNT3A ligand (Figs. 3j, k and 5k, l) suggest that the lipid droplet accumulation and defective regeneration of the mutant mice is the result of increased sensitivity to WNT activation, which in vivo could be mediated by an increased sensitisation of mutant hepatocytes to WNT ligands secreted by the cellular microenvironment, as suggested by Sun et al. 24 . However, our organoid studies in mutant organoids (Fig. 5j) and mutant cancer organoids (Fig. 6e) show that mutated cells proliferate in the absence of WNT (EM) and WNT and RSPO ligands (DM condition and cancer organoids). This observation prompts us to an additional --yet complementary--interpretation; in vitro, RNF43/ZNRF3 mutant cells exhibit an intrinsic, cell-autonomous activation of their proliferative state and a reduced differentiation capacity (Figs. 3h, i and 5i, j). These results combined with our observations that the 7 months mutant livers present signs of tissue degeneration (necro and apoptosis) suggest a scenario whereby the cooperation between different mechanisms, increased susceptibility to WNT ligand from the cell microenvironment, intrinsic changes to the differentiation and metabolic states and different susceptibility to rounds of damage, explains the lipid metabolic alterations and the defective hepatocyte differentiation and regeneration, which would eventually lead to increased susceptibility to liver cancer.  5 Rnf43/Znrf3 loss impacts the regenerative capacity of hepatocytes by altering their proliferation/differentiation states. a-g Adult R&Z flox mice were treated with tamoxifen or AAV8-TGB-Cre to generate R&Z del . Controls were treated with vehicle or AAV8-null virus. Induced mice underwent partial hepatectomy (PHx) and samples were collected at the indicated time points. a, b Experimental design (a) and timeline (b). c Fold change of the number of proliferating (Ki67+) cells in R&Z del livers with respect to corresponding controls in homoeostasis (3 or 7 months of deletion), after CCl 4 (4 or 50 days recovery) or PHx (7 or 21 days recovery) damage. Violin plot showing median, IQR and full distribution of a minimum of 10 FOV per mouse (n = 3 or n = 4, CD 4d). Two-way ANOVA with Sidak multiple comparisons. Homeostasis, **p = 0.0023; PHx, ***p = 0.0003. d, e Common signature of 92 genes upregulated in the mutant upon PHx and CCl 4 chronic damage. d Venn diagram. e Top 10 GO biological processes. Details in Supplementary Data 5. f, g Hepatocyte markers are significantly downregulated in the R&Z del vs WT upon CCl 4 (f) and PHx (g) damage. f Left, GSEA analysis between hepatocyte signature and CCl 4 -damaged. Details in Supplementary Data 1_S37. Right, DE genes in R&Z del vs R&Z flox after chronic CCl 4 treatment. Heatmaps present the Top500 (FDR < 10%) ranked by Log2FC and averaged between all mice. g Heatmaps of DE genes upon PHx ranked by Log2FC and p value <0.01 and averaged between all mice (Likelihood ratio test and Benjamini-Hochberg correction). Number, the number of genes. h-l Wild-type and R&Z del adult liver ductal, hepatocyte or hepatoblasts organoids were generated as detailed in "Methods" and grown in expansion media (EM, devoid of WNT3a) or differentiation media (DM, devoid of WNT3a and RSPO1) or treated with Wnt inhibitors (Wnti) or WNT3a media before collection. Note that, for WNT3a, two conditions were used: WNT3a in expansion medium (Wnt3a) and Wnt3a in the absence of CHIR 99021 and RSPO (Wnt3a −CHIR/RSPO). h Experimental design. i Representative pictures of hepatoblast organoids stained for Albumin (magenta). j, k qPCR expression analysis on the indicated genes in R&Z del vs R&Z flox in differentiated cholangiocyte (left) or hepatoblast (middle) or adult hepatocyte (right) organoids grown in the indicated media (j) or R&Z flox adult hepatocyte organoids treated or not with WNT3A or Wnti. Heatmap represents the fold change values of R&Z del vs R&Z flox (j) or R&Z flox treated vs untreated (k). Each column is a biological replicate. nd not detected. l Albumin secretion in R&Z flox hepatocyte organoids in WNT3A in the absence of CHIR and RSPO. Data represents mean + /− SEM from n = 2 replicates from n = 3 independent experiments. Paired two-tail t test was used, *p = 0.0135. Source data in Source data file.
The significant resemblance between the mouse and human expression signatures, combined with the altered lipid metabolism and potential as prognostic biomarkers, suggests that our mouse findings are relevant to explain the pathogenesis of the disease in ZNRF3/RNF43 mutant patients. Hepatic long-chain fatty acid composition has been linked to inflammation and NASH development 34,58 . Both mice and patients present increased signatures of unsaturated fatty acid biosynthesis and NASH. This is in contrast to Sun et al. 24 , who observe increased fatty acid signatures, adipogenesis and cholesterol but not steatohepatitis in the mutant mice. The reason for this difference is unknown but is likely due to differences in genetic backgrounds  Several outstanding questions arise from our investigations. First, we found a clear interaction between mutations in WNT (APC, AXIN, CTNNB1) and RNF43/ZNRF3 mutations, both in prognosis and gene expression, supporting the notion that WNT and ZNRF3/RNF43 mutations might be acting through slightly different mechanisms. Both types of patients exhibited clear canonical WNT activation and lipid metabolic alterations; however, only RNF43/ZNRF3 with no other WNT mutations (APC/ AXIN/CTNNB1) present poorer survival, suggesting that additional non-WNT mechanisms might contribute to tumorigenesis in RNF43/ZNRF3 patients. In that regard, the transcriptomic analysis excluding TCF4 targets indicates that both, mouse and patients present non-TCF4 (non-canonical WNT) signatures that contribute to tumorigenesis. Along these lines, a recent report demonstrates a non-WNT role for RNF43 in DNA-damage response in gastric cancer cells 59 . Second, one intriguing observation from the human studies was that many of the somatic variant mutations for ZNRF3, contrary to RNF43, were in noncoding regions. Variants in introns can affect splice sites and also cause loss of regulatory repressor elements, as described for BRCA2 in Fanconi anaemia 60 . Third, we found no significant differences in the expression of any of both genes in mutant patients (Supplementary Data 6_S44), suggesting that, in humans, the mechanism is not related to transcriptional changes but potentially changes in protein function. In that regard, at least for RNF43, point mutations can result in dominant-negative forms [61][62][63] . Hence, our results call for future studies to assess the effect of individual RNF43/ZNRF3 mutations in human liver tumorigenesis.
In summary, our studies provide a framework to start to understand the role of the tumour suppressors RNF43/ZNRF3 in liver cancer. We find that the combination of mechanisms including (1) intrinsic predisposition to intracellular fat accumulation, (2) a disrupted liver lipid composition and zonation, (3) an increased hepatocyte proliferation and reduced differentiation capacity with subsequent inability to terminate regeneration and (4) an increased sensitivity to WNT ligand, they all coalesce in the perfect mix to promote malignancy. More studies are needed to characterise the role of RNF43/ZNRF3 in NAFLD development and cancer progression. With the alarming increase in the consumption of fat and sugar worldwide, recognising those individuals with mutations in these E3 ligases or any other canonical WNT signalling component might facilitate the discovery of populations at risk of developing liver steatosis, NASH, cirrhosis or cancer, and facilitate their therapeutic intervention and management of the disease.

Animals
Husbandry and gene recombination. All mouse experiments were conducted in accordance with procedures approved by the UK Home Office relating to the use of animals in research (Animals Act 1986). All mice were maintained under standard dark/light cycle, temperature and humidity conditions and both male and female adult (8-16 weeks old) mice were used in this study. Conditional knockout (KO) mice for Rnf43 and Znrf3 (Rnf43/Znrf3 flox ) were previously generated and described in Koo et al. 18 . Briefly, exons encoding for the ring domain of RNF43 and ZNRF3 were flanked by LoxP sites to drive excision in presence of a CRE protein.
Rnf43/Znrf3 flox mice were also crossed with either Alb1-CreERT2 25 or Sox9-CreERT2 (Jackson Laboratory JAX-http://jaxmice.jax.org/) to generate Alb1-CreERT2-R&Z flox (used for the in vivo studies) and Sox9-CreERT2-R&Z flox (used to generate R&Z del organoids from liver progenitors). The Cre enzyme was induced by intraperitoneal (IP) injection of 200 mg/kg tamoxifen (Sigma-Aldrich) dissolved in sunflower oil. Sunflower oil only was injected to R&Z flox mice, as littermate controls. Alternatively, deletion was performed after tail vein injections of 2.5 × 10 11  Liver damage models. For the chronic liver damage model, mice received IP injections of CCl 4 (Sigma-Aldrich) at 0.5 μl/g (diluted in corn oil) multiple times according to the protocol (Figs. 4a and 6a). For the acute liver damage model, mice received a single IP injection of a lethal CCl 4 (Sigma-Aldrich) dose diluted in corn oil at 2 μl/g (diluted in corn oil). In both cases, unrecombined, control littermates, received corn oil vehicle only.
Partial hepatectomy was performed as described in ref. 64 and mice were allowed to recover for 7 or 21 days before liver collection. In all cases, mice were sacrificed by either cervical dislocation or carbon dioxide overdose. Blood was collected by cardiac puncture or tail vein prick and samples were submitted for the analysis of liver transaminases levels (Clinical Pathology Laboratory, The Queen's Veterinary School Hospital, Cambridge, UK).
Primary hepatocyte isolation, FACS sorting and analysis. For adult hepatocyte organoid cultures, mice were culled by cervical dislocation and isolation of primary hepatocytes was performed by liver perfusion as described in ref. 65 . For hepatocyte collection and cell sorting of the different cell compartments, mice were culled by carbon dioxide intoxication and the head was removed to allow exsanguination. Livers were perfused with 10 ml of perfusion buffer (0.5 mM EGTA in PBS), then with 10 ml of perfusion buffer with bovine serum albumin (BSA; Sigma-Aldrich) and then with 10 ml of pre-warmed 0.5 mg/ml Collagenase A (Sigma-Aldrich) solution (HBSS 1×, 5 mM CaCl 2 , 20 mM HEPES). After 2 min of incubation, livers were carefully removed, place in STOP solution (HBSS 1×, 5 mM CaCl 2 , 20 mM HEPES, 0.5% BSA) and disaggregated using tweezers. The mixture was washed several times in STOP solution (50 G for 2 min) and filtered with a 100 μm strainer. The pellet containing hepatocytes was resuspended in RLT lysis buffer (Qiagen) and frozen at −80°C for RNA extraction. The supernatant was centrifuged again at 300 × g for 5 min to collect non-parenchymal cells. Pellet was divided into two fractions and stained either with EpCAM-APC (eBioscience) and CD31-PE-Cy7 (AbCAM) antibodies or EpCam-APC and CD11b-PE-Cy7 (BD Bioscience) antibodies (see Supplementary Data 7). Positive cells were sorted using a Flow Sorter MoFlo (Dako Colorado, Inc.), resuspended in RLT lysis buffer and stored at −80°C. For analysis of the immune compartment by flow cytometry, livers were isolated and digested as previously published 41 . Single cells were subsequently stained with the corresponding labelled antibodies (Supplementary Data Set 5, 30 min, ice) before analysis in a SONY SH800S cell sorter. Liver single cells were selected by forward/scatter exclusion and the different immune cell types were determined according to the expression of the described cell markers using FlowJO_V10 (Supplementary Fig. 1g).
Histology. Tissues were dissected and then fixed overnight in 10% buffered formalin (Sigma-Aldrich). For paraffin embedding, tissues were dehydrated in increasing concentrations of ethanol (70, 95 and 100% for 2 h each) ending in xylene and then incubated overnight in paraffin (Sigma-Aldrich). The following day, tissues were embedded in paraffin blocks and subsequently sectioned at a thickness of 5 μm using a microtome. Sections were mounted onto SuperFrost Plus slides (ThermoFisher) and incubated from 3 to 12 h at 60°C before staining. H&E (Sigma-Aldrich), Oil Red O (Sigma-Aldrich) and PicroSirius Red (AbCAM) staining were performed according to the manufacturer's instruction (Supplementary Data 7). For immunohistochemistry, sections were deparaffinised in two changes of xylene, 5 min each and rehydrated in descending grades of ethanol (100% 2×, 95%, 70%), 3-5 min each ending in water. Antigen retrieval was performed either enzymatically or heat-mediated based on primary antibody (see Supplementary Data 7). For enzymatic retrieval, sections were incubated 10 min at 37°C and 10 min at room temperature (RT) with TE buffer containing 0.5% Triton X-100 and 1/1000 proteinase K (800 U/ml). For heat-mediated antigen retrieval, samples were incubated in either citrate buffer (10 mM Sodium Citrate, pH 6) or Tris-EDTA buffer (10 mM Tris base, 1 mM EDTA solution, 0.05% Tween 20, pH 9.0), heated in a microwave to 95-98°C for 20 min and then cooled at RT for 20 min. Afterwards, samples were blocked and permeabilized for 1.5 h using TBS buffer containing 2% foetal bovine serum (FBS), 1% bovine serum albumin (BSA) and 1 or 0.1% of Triton x-100 based on antibody (see Supplementary Data 7 for details). Samples were then incubated overnight with primary antibody at an antibody-dependent concentration (see Supplementary Data 7). The following day, samples were washed in TBS and endogenous peroxidases were inactivated by incubating for 15 min in methanol containing 3% of hydrogen peroxide. Poly-HRP-GAMs/Rb IgG (Immunologic) secondary antibody was incubated for 30 min (Supplementary Data 7). After washing, the signal was developed using Bright Dab substrate kit (Immunologic) using the manufacturer's instruction. Sections were counterstained in haematoxylin for 5 min, dehydrated and mounted using xylenebased DPX mounting media (ThermoFisher). Slides were analysed using optical microscopy. Pictures were taken in a DM4 Leica microscope and images were quantified in ImageJ (Fiji) by a researcher that was blinded to the experimental cohorts. H&E samples were evaluated for NAS score as previously described 29 .
For OCT embedding, livers were previously cryopreserved in 30% sucrose overnight. OCT blocks were subsequently sectioned at 6 μm in a cryostat for immunofluorescence staining of the corresponding antibodies (see Supplementary Data 7). Briefly, sections were washed in PBS and incubated in blocking solution (2% NDS, 1% Triton x-100 and 5% DMSO in PBS) for 2 h at 4°C before staining with the corresponding antibodies (see Supplementary Data 7) overnight at 4°C. After 3 washes with PBS, fluorescent-labelled secondary antibodies (see Supplementary Data 7) were eventually added in blocking buffer for 2 h at RT. Nuclei were stained by Hoechst 33342 (Invitrogen) 1:1000 in PBS for 5 min before mounting in Vectashield mounting medium. Slides were analysed in an Olympus Fluoview FV1000.
Lipidomics analysis. Lipids were extracted from liver tissue using the Folch method. Briefly, liver tissue was homogenised in chloroform:methanol (2:1, 1 ml) using TissueLyzer (Qiagen Ltd., Manchester, UK). Deionized water (400 μl) was added, and the samples were well mixed. Separation of the aqueous and organic layers was carried out following centrifugation (12,000 × g, 10 min). The lipidcontaining organic extract was evaporated in a vacuum centrifuge, without heat, and stored at −80°C until analysis. Details on lipidomics analysis can be found in Supplementary Methods.
Mass spectrometric imaging. Ten-μm frozen liver sections were placed on glass microscope slides (VWR) using a cryostat. Adjacent sections were stained with H&E (Sigma Aldrich). Matrix solution (5 mg/ml) of norharmane (Sigma-Aldrich, St. Louis, MO) was prepared in 2:1:0.8 methanol:chloroform:water (v/v) and administered to the tissue surface using a SMALDIPrep sprayer (TransMIT GmbH, Giessen, Germany). Imaging experiments were performed using an AP-SMALDI10 source (TransMIT GmbH) coupled to an Orbitrap Q-Exactive (Thermo Fisher Scientific) at 10 μm increments across the tissue, using 2D-line mode. Spectra were acquired in positive ion mode from 250 to 1000 m/z at 70,000 resolution. Data were converted to imzML format 66  RNAseq and analysis. RNA libraries were prepared for sequencing using the Smart-Seq2 protocol. Details on sequencing and analysis can be found in Supplementary Methods.
TCF4-binding analysis. The location of TCF4 peaks was taken from a previous study 15 and annotated using ChIPseeker (version 1.12.1) 68 in R on the mm9 genome. Genes were filtered to include those with a peak within 5 kb upstream or downstream of their transcriptional start site. We defined a lipid metabolism set of genes by combining genes contained in the GSEA hallmark fatty acid metabolism gene set, the Reactome cholesterol biosynthesis gene set, Wang classic adipogenic targets of Pparg and the GO lipid metabolic process (including child terms) set. Random peak sets were generated by shuffling the peak locations using ChIPseeker. This was done 1000 times to generate empirical p values for the enrichment of lipid metabolism genes with TCF4-binding sites. Venn diagrams were generated using the web tool Venny 2.1 (http://bioinfogp.cnb.csic.es/tools/venny/).
RNAseq differential expression analysis. For the homoeostasis model (R&Z del vs R&Z flox at 1 month, 3 months and 7 months of deletion), reads were filtered to keep genes with counts per million >1 in at least 3 samples. A multiple regression model was fitted with three terms, one a factor representing the condition (WT or KO) and the time-point of the sample (1 month, 3 months or 7 months), one denoting the sex of the mouse and the third denoting the batch of the sample as the data were generated in two separate batches. Normalisation was done using calc-NormFactors and variance estimators calculated using estimateDisp in edgeR. Differential expression between WT and KO mice was calculated by fitting the appropriate contrast. For example, a contrast of the difference between KO at 1M and WT at 1M. Significantly DE genes for GO and GSEA were defined as having a false discovery rate (FDR) <10% and an absolute log 2 FC >1 (Supplementary Data 1). For TCF4 and lipid metabolic analysis differential expression of R&Z del livers was defined as p value <0.05 and an absolute log 2 FC >0.4 to increase the depth of the analysis (Supplementary Data 3). Empirical p values for enrichment of TCF4 regulated genes in those showing differential expression was similarly calculated using the random peak sets used in the TCF4 peak analysis. For partial hepatectomy differential expression analysis, a PCA plot was drawn following rlog transform (DESeq2) for all samples. The PCA plot was coloured according to the genotype and time-point and split the data on the first principal component. For the differential expression analysis, all 24 samples from the PCA analysis were used. This comprised 8 mice that had undergone damage and 6 nondamage mice at 1 month included as additional controls. The model included a term relating to the condition (either WT or KO) and the time point (either 7 or 21 days for the damaged mice and 1M for the non-damaged). This combined condition and time point factor also accounts for the difference between damage and non-damage as all non-damage mice were at 1M. Based on the PCA results, a factor for the sex of the mouse was also included in the regression model. Ensembl IDs were mapped to HUGO gene symbols using biomaRt (version 2.42.0) in R and the org.Mm.eg.db (version 3.10.0) database. The contrasts fitted were between KO and WT damaged mice at 7 days and at 21 days. Significantly DE genes for GO and GSEA were defined as having a p value <0.05 and an absolute log 2 FC >1 (Supplementary Data 5). The non-damage mice were not used in the differential expression calculation.
For the chronic damage model, comparison expressed genes were defined as having counts per million of at least 1 in half of the samples. Normalisation was done using calcNormFactors and variance estimators calculated using estimateDisp in edgeR. Differential expression analysis was performed using edgeR (version 3.18.1) and significantly DE genes were defined as having a FDR <10% and an absolute log 2 FC >1 (Supplementary Data 1) while for comparison between chronic damage and partial hepatectomy DE genes were defined as either FDR < 0.1 or p value <0.05 and a log 2 FC >1 depending on the number of significantly expressed genes (Supplementary Data 5).
IPA analysis. After statistical analysis, DE genes within groups were studied using the IPA (Qiagen). We imputed in IPA the whole transcriptome and then filtered for analysis focused on only statistically significant (p < 0.05) items with Log2CPM > 0.05 and −0.585 < Log 2 FC > 0.585. Pathways and "Upstream Regulators" showing relationships and interactions between DE genes and others that functionally interact with them, were generated and ranked in terms of the significance of participating genes (p < 0.05) and activation status (Z-score). A comparison analysis was then performed to focus only on those pathways significantly enriched. We considered "biologically relevant" only those genes that are statistically significant (p < 0.05), and enriched in "significantly modulated" pathways in the comparative analysis, and/or those genes with a FDR < 0.05. See details in Supplementary Data 2.
Liver hepatocyte organoids. To obtain adult hepatocyte organoids, livers were perfused as described above. The resulting hepatocyte suspension was then filtered through a strainer (pore size of 100 μm) followed by centrifugation for 5 min at 50 × g and at 4°C. After washing the obtained cell pellet with suspension buffer and centrifuging as before, the resulting pellet was resuspended in suspension buffer by gentle inversion for cell quantification and viability assessment. Five thousand viable hepatocytes were eventually plated in 20 μl matrigel and cultured in expansion media (Hepatocyte medium as described in 33 ). R&Z del hepatocytes were generated by either viral infection using AAV8-TBG-Cre particles (MOI = 100, 60 min at 37°C) before cells were plated in matrigel for organoid culture establishment or treatment with 6 μM 4-hydroxy-tamoxifen (Sigma) for 48 h after plating. After deletion, cells were allowed to recover for 7 days in expansion media before splitting for experiments. For bodipy, R&Z flox and R&Z del organoids were grown in expansion media or in the presence of 2 Wnt-inhibitors: IWP2 (3 μM) and iCRT3 (25 μM) (Wnti medium), or 30% Wnt3a conditioned media (+Wnt3a) for 7 days before collection. For assessment of differentiation, R&Z flox and R&Z del organoids were grown either in expansion media or DM for 7 days before collection. Additionally, R&Z flox organoids were also treated with Wnti in expansion media (with CHIR99021 and RSPO1) or 30% Wnt3a conditioned media in expansion media, or in expansion media without other WNT activators (no CHIR and no RSPO1) for 7 days before collection. Finally, 24 h before collection, media was changed and the eventual supernatant was collected for measuring albumin secretion using the Mouse Albumin AssayMax ELISA Kit (Assay Pro).
Organoid staining. Organoids were collected using a plastic Pasteur pipette, washed with cold PBS several times to remove matrigel or BME carefully to not disrupt their 3D structure. Subsequently, organoids were fixed for 40 min on ice with 10% Formalin. Fixed organoids were stored at 4°C. For staining, fixed organoids were blocked with 1% bovine serum albumin (BSA), 2% donkey serum, 1% DMSO in PBS for 2 h at RT. Primary antibody was incubated o/n at 4°C in 0.05% BSA in PBS. The secondary antibodies were added at a concentration of 1:150 for 2 h at RT. Nuclei were stained with Hoechst 33342 (Invitrogen) at a dilution 1:1000 for 15 min at RT. To assess lipid accumulation, cells were stained using BODIPY 493/503 dye (2 μM in PBS). BODIPY was added after the incubation with the membrane dye phalloidin. BODIPY was incubated for 3 h at RT protected from light. Imaging was performed using an SP8 White Light inverted confocal microscope or an Olympus Fluoview FV1000 confocal. The images were edited and analysed using the Leica Application Suite X software (LAS X, v1.5.1.1387) and ImageJ (v1.51j8).
ZNRF3/RNF43 human mutation survival and expression analysis. To assess the effect of RNF43/ZNRF3 mutations in human primary liver tumours we took advantage of the publicly available ICGC data collections, precisely, from Japan (LIRI-JP) and LICA-FR studies from the ICGC database 48,49 . We downloaded the RNAseq data for both studies. These data sets also included WGS of patients and corresponding survival data. Details on filtering methods and mutation and expression analysis can be found in Supplementary Methods. Survival analysis. For the survival analysis donors were filtered to include those passing the quality control of the PCAWG 69 study and containing associated tumour stage information. There were 257 donors in the LIRI-JP study and 1 in LICA-FR meeting the above-selected criteria. As the LICA-FR study only included 1 donor meeting our criteria the remaining survival analysis was performed using the LIRI-JP data set only.
We included a combined tumour stage covariate in the survival analysis based on the tumour stage at diagnosis information in the clinical data. Patients were split into two groups, those whose tumour stage was either 0 or 1 and those whose tumour stage was 2, 3 or 4. The sex of the donor was also used as a covariate in the model. As covariates, we also considered the effect of having a ZNRF3 and RNF43 mutation (ZandR). That left a sample size of 119 donors in total (no WNT mutations), 48 of these had ZNRF3 or RNF43 mutants with 38 having a mutation in ZNRF3 and 71 patients with no mutations (WT tumours). The number of patients with single RNF43 mutations or double mutation and fulfilling the criteria of no other mutations in the WNT pathway was too low to provide meaningful conclusions. To compare the double mutation to a single mutation in at least one of ZNRF3 or RNF43, we included in one model a separate term for each of ZNRF3 and RNF43 single mutations. In a second model, we included a combined ZNRF3 or RNF43 mutation factor (ZorR), in which a donor had at least one mutation in ZNRF3 or RNF43.
To test for differences in the effect of ZorR mutation on survival depending on the presence of Wnt mutations, we also fitted an interaction term between ZorR and Wnt.
The assumption of the proportional hazards used in the Cox regression model was tested using the Cox proportional hazards test (cox.zph) in the survival package in R. We accepted the null hypothesis of the proportional hazards model for all covariates except Wnt mutation status (p value <0.1) when using the ZorR covariate in the interaction model. Therefore, the data were split according to Wnt status.
For the donors with no mutation in Wnt, we fitted two multiple cox regression models using factors ZandR, sex and the combined tumour stage factor in each. We then included either two factors one for each ZNRF3 or RNF43 mutation in one model (Single model). In the second model, we included a combined OR factor for ZNRF3/RNF43 mutation status that was 1 if the patient had a mutation in at least one of ZNRF3 or RNF43, 0 otherwise (Combined model).
In the Wnt WT model, there were 15 with RNF43, 38 with ZNRF3 mutations, 48 with a mutation in RNF43 or RNF43 and 5 with a mutation in both. In the Wnt mutant group, there were 138 patients in total, 25 with double mutation and 54 with ZNRF3 and 32 with RNF43 mutation. Differences between the survival curves were tested using the Wald statistic and Kaplan-Meier curves plotted using the survival (version 3.2-7) and GGally (version 2.1.1) packages in R.
Gene expression analysis. For the differential gene expression analysis, we used RNAseq samples from the LIRI-JP and LICA-FR data sets. We included patients who also had corresponding mutation data from WGS, excluding those where WES was used. In total, 282 donors had RNAseq data with matching WGS available with 240 from LIRI-JP and 42 from LICA-FR. Genes were filtered to remove those not expressed, where expressed genes are defined as having counts per million >1 in >150 samples across the two data sets. Mutation status was determined as detailed above. A multiple regression model was fitted in R using edgeR. The regression model contained four terms, one for each ZNRF3 OR RNF43 mutation status, ZNRF3 AND RNF43 mutation status, WNT mutation status and the fourth for the batch of origin of the sample LIRI-JP or LICA-FR. The normalisation of RNAseq samples was done using calcNormFactors and variance estimates calculated using estimateDisp in edgeR.
Differential gene expression analysis was calculated between WT and ZNRF3 OR RNF43 patients (having a mutation in at least one of ZNRF3 or RNF43, RorZ), between WT and double ZNRF3/RNF43 mutation (R&Z) and finally between WT and WNT mutation (WNT). There were 67 WT donors and 123 RorZ patients; 33 patients had a mutation in both R&Z and 177 patients had a mutation in one of the WNT genes. Genes were defined as DE between conditions if the FDR was <10%. GSEA against a defined set of lipid metabolism gene set (Supplementary Data 6_S24) was performed using the GSEA 3.0 software downloaded from http://software.broadinstitute.org/gsea/index.jsp.
For TCF4 enrichment in human gene expression data, we ranked the differential expression of three factors (Wnt mutation status, ZNRF3 or RNF43 mutation status and ZNRF3 and RNF43 mutation status) versus WT from the multiple regression model. A set of genes containing a TCF4 binding site within 5 kb (189 genes) was used to perform GSEA against a ranked list of DE genes in both models (vs all (left) and vs non-WNT mutants (right)). The top (first ranked) position had the gene with the highest increase in expression compared to the WT for each group. The ranked list was compared to the set of TCF4 genes (Supplementary Data 6_S45) and the maximum absolute enrichment score was calculated for each gene set. A positive enrichment score indicative of enrichment for TCF4 genes was upregulated (increase in expression) when compared to the WT.
For the comparison between the gene expression of our mutant mice and the human gene expression of HCC patients mutated in RNF43 and/or ZNRF3, GSEA was performed using genes DE between human patients with ZNRF3 and/or RNF43 mutations compared to non-WNT mutant patients and human patients with WNT vs non-WNT mutant patients regardless of their ZNRF3/RNF43 mutation status.
Statistical analysis. All statistical methods are reported in the figure legends. Panels where several data points were quantified per mouse were plotted as a mean showing the distribution of all data points and analysed using a t test of means. When required, the data were tested and confirmed to be normally distributed by performing a normality test (Shapiro-Wilk test).
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27923-z ARTICLE Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All mouse gene expression data generated in this study have been deposited in the Gene Expression Omnibus (GEO) under accession code GSE133213. LIPID MAPS® structure was obtained from the owner's webpage [https://www.lipidmaps.org/data/structure/]. TCF4 targets were determined using a publicly available data set [https:// www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE41284]. LICA-FR and LIRI-JP data sets were obtained from the ICGC database [https://daco.icgc.org/]. Source data are provided with this paper.