Hepatocellular Carcinomas Originate Predominantly from Hepatocytes and Benign Lesions from Hepatic Progenitor Cells

1Hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC) are the most lethal primary forms of liver cancers. Rarer benign hepatic tumors include focal nodular hyperplasias and hepatocellular adenomas (HCAs), which may progress to HCC. Regenerative nodules (RN), common in cirrhotic liver, are also likely HCC precursors.

2Hepatocytes represent the most functionally active parenchymal liver cells (>80%). Others are the bile duct epithelial cells or ductular cells (DCs). DCs originate from the canal of Hering and are considered to be the bipotential hepatic progenitor cells (HPCs) also known as transit-amplifying cells or oval cells in rodent models or liver progenitor cells in humans. The canal of Hering is therefore considered the putative hepatic stem cell niche in the adult liver. HPCs have a transcriptomic profile similar to DCs as well as the ability to replenish damaged hepatocytes and cholangiocytes.

3The liver has remarkable plasticity following injury and is the only adult organ with complete regenerative potential. Depending on the injury (acute or chronic), liver regeneration may involve activation of either fully differentiated hepatocytes or HPCs. Fate-tracing studies indicate that hepatocytes are mainly responsible for liver maintenance under normal conditions and responses to acute liver injury. Acute liver damage, partial hepatectomy or chemical damage by carbon tetrachloride (CCl4), may induce replication of mature hepatocytes, which replenishes the liver and maintains its homeostasis.

4Impairment of hepatocyte replicative potential can induce the emergence of HPCs, which are required to continuously supply new hepatocytes and maintain the organ’s functional integrity during chronic liver damage. HPC expansion is mainly due to various pathologies, environmental or genetic etiologies of human HCC, primarily affecting hepatocytes and leading to chronic liver damage in the early stages of tumorigenesis (e.g., fibrosis, cirrhosis, steatosis, inflammation, viral infection, aflatoxin, α-1 antitrypsin deficiency, Wilson’s disease, and hemochromatosis). HCC progression is thus often accompanied by HPC activation, but mechanisms initiated by injured hepatocytes remain still elusive.

5Cancer stem cell markers (EpCam, CD133, and CD44), specific cytokeratins (CKs; 7 and 19), and the sex-determining region Y-box (SRY-box) containing the transcription factor Sox9 aid recognition of DCs and HPCs. DCs and HPCs remain negative for hepatocyte nuclear factor 4 alpha (HNF4α), a marker of mature hepatocytes and hybrid periportal hepatocytes (HybHPs). A recent report indicated that HybHPs expressing both Sox9 and HNF4α may participate in liver regeneration, but not HCC. Long-debated uncertainties are thus whether DCs and HPCs are hepatic cancer stem cells (CSCs), and whether they participate in hepatocarcinogenesis. According to recent proposals, cells with similar transcriptomic profiles to oval cells may be HCC progenitors. Furthermore, liver carcinomas derived from genetically modified bipotential liver progenitors resemble features of the human disease. Recently, hepatocytes were suggested to be the sole cell of origin for HCC, and HPCs or DCs do not produce tumors in genetic and chemical HCC mouse models. Certainly, due to scarcities of animal models recapitulating human liver tumorigenesis, the cells of origin of liver cancer as well as the origin of its heterogeneity (benign lesions, RN-induced cirrhosis, HCAs, and HCCs) are not clear yet and may be context dependent.

6A mouse model (hURI-tetOFFhep) we recently generated expresses hepatocyte-specific human unconventional prefoldin RPB5 interactor (hURI). In this model, NAD+-deficit-induced DNA damage causes multistep liver tumor development somehow mimicking human hepatocarcinogenesis, with development of focal nodular hyperplasia, RN, non-alcoholic steatohepatitis (NASH), HCAs, and HCCs. Using this model and several other models, combined with genetic lineage tracing and in vitro co-culture cell experiments, we evaluate here the contribution of hepatocytes and HPCs in hepatocarcinogenesis. We also report mechanisms emerging from hepatocytes and regulating HPC expansion in the early stages of tumorigenesis, through metabolic paracrine signals, implicating galectin-3 and α-ketoglutarate.

7Human HCC develops via pre-neoplastic lesions that frequently arise within a background of chronic inflammatory liver damage or hepatitis progressing to fibrosis, cirrhosis, and eventually HCC. Premalignant lesions contain Sox9-positive progenitors (considered to be HPCs). Sox9 expression correlates with degrees of liver-damage-associated fibrosis, according to immunohistochemistry (IHC) of human hepatitic samples (Figures 1A and 1B). Thus, Sox9-positive cells (SPCs) expand in early human hepatocarcinogenesis.

8Using public expression data applied to classify human HCC subtypes (Figures 1 and S1), we identified gene sets that are up- or downregulated in hURI-tetOFFhep livers in early tumorigenesis stages and correlate with therapy resistance and poor prognosis according to gene set enrichment analysis (GSEA). Normalized enrichment scores and false discovery rates indicated that 1-week-old hURI-tetOFFhep mice have a signature very similar to G5 subtype HCC, which is strongly related to β-catenin mutations and thus WNT activation, which is essential for HPC expansion (Figures 1C and 1D). Further evaluation showed that an HPC signature associated with WNT/β-catenin signaling is significantly enriched in 1- and 8-week-old hURI-tetOFFhep mice (Figures 1C and 1E).

9GSEA also detected: significant overlaps between transcriptomic signatures of 8-week-old hURI-tetOFFhep mouse and the S1 HCC subclass, confirming that in early stages the model exhibits an aberrant WNT signaling pathway signature (Figures 1C and S1A); fetal liver and p53 pathway activation signatures, respectively present in G1 and G3 subtypes (Figures 1C and S1B); and a signature associated with transforming growth factor β (TGF-β) signaling pathways that reportedly activate HPCs (Figure 1C).

10Significant enrichment of genes associated with poor patient survival, including stem cell genes, was detected in data from 8-week-old hURI-tetOFFhep mice (Figures 1C and 1F) and a metastasis gene signature predicting tumor relapse in early-stage HCC patients (Figures 1C and S1C). Further analysis revealed that these stem-like liver cancer subtypes represent a class of poorly differentiated HCC (Figures 1C and S1D). Moreover, stem-like liver cancers from relapsing patients showed a signature enriched in 8-week-old hURI-tetOFFhep mice (Figures 1C and 1G). Thus, in early stages, hURI-tetOFFhep livers display a stem cell-like signature found in aggressive human HCC.

11Upregulated proteins identified in published proteomic (isobaric tags for relative and absolute quantification [iTRAQ]) profiles of 8-week-old hURI-tetOFFhep mouse livers included several DC markers and proteins of the extracellular matrix (ECM) (e.g., laminin, vimentin, and transgelin) essential for oval cell expansion (Figure S2A). Notably, hURI was specifically expressed in hepatocytes, but not in DCs (Figure 2A). Thus, hURI-tetOFFhep mice were crossed with Sox9IRES-EGFP mice expressing EGFP from the Sox9 promoter primarily in DCs (Supplemental Experimental Procedures). EGFP/SPCs were isolated by fluorescence-activated cell sorting (FACS). qRT-PCR of isolated EGFP-positive cells and whole mutant livers (hereafter, “mutants” refers to mice ectopically expressing hURI) confirmed that hURI is specifically expressed in hepatocytes (Figure S2B). Interestingly, IHC and western blot (WB) of Sox9 and CK19 markers confirmed the presence of a ductular reaction in mutant livers (Figures 2B, 2C, and S2C). We detected DC expansion in mutant livers when preneoplastic lesions were apparent, in 8- to 24-week-old mutant livers, but not in non-pathological 3-week-old livers expressing hURI (Figure 2B). Importantly, increased laminin was confirmed by IHC (Figures S2D and S2E). SPCs also expanded in 7-week-old C57BL/6 mice treated with the diethylnitrosamine (DEN) carcinogen known to induce HCC (Figures S2F and S2G). Thus, SPCs expand during liver tumorigenesis.

12Co-immunofluorescence (co-IF) using Sox9 and CK19 antibodies in hURI-tetOFFhep liver sections revealed that out of the total number of cells expressing either Sox9 or CK19, ∼15% were positive for only Sox9, 60% were CK19 positive, and 30% were positive for both (Figures S2H and S2I). Thus, SPCs comprise a small subset of the highly heterogeneous DC population.

13We subsequently checked other DC/HPC markers by FACS-sorting EGFP+ SPCs from liver cells of 12-week-old mice generated from an hURI-tetOFFhep and Sox9IRES-EGFP cross (Supplemental Experimental Procedures). The expanded EGFP+ SPCs in mutant mice represented 5.76% ± 2.7% of the liver fraction excluding hepatocytes but only 0.9% ± 1% in their littermates (Figure 2D). EGFP cells were positive for the CSC markers EpCAM, CD133, and CD44 (95.5% ± 1.79%; 94.0% ± 1.51%, and 21.2% ± 3.81%, respectively). However, a small proportion of EGFP+ SPCs was positive for LGR5 (8.23% ± 1.79%) and DLK1 (3.23% ± 1.20%) markers (Figure 2D). SPCs thus represent a heterogeneous DC population with stem cell characteristics and may be considered as hepatic CSCs or HPCs.

14Next, we tracked SPCs during liver tumorigenesis by crossing Sox9IRES-CreERT2 and reporter R26-stop-EYFP. In this context, SPCs express an inducible Cre recombinase, which specifically deletes the loxP-flanked stop cassette in the Rosa26R reporter locus. This occurs in genomic DNA, so all descendants of labeled cells inherit enhanced yellow fluorescent protein (EYFP) expression. For confirmation, 5-week-old mice were fed a tamoxifen diet for 2 weeks and then transferred to chow (Figure 3A). Tamoxifen-activated Cre recombinase permanently EYFP-labeled the emerging SPCs (Figure 3B). Importantly, the tamoxifen dose (400 mg/kg supplied in food) and exposure time (2 weeks) did not induce the hepatocyte toxicity known to increase Sox9 levels in hepatocytes as previously reported (Figure 3B). Co-IF of Sox9 and EYFP revealed a 44.2% ± 7.7% recombination efficiency (Figure 3C). Therefore, only about 40% of the SPCs are genetically tracked in this model. Importantly, after 2 weeks of tamoxifen treatment, we did not detect any cells positive for Sox9 and HNF4α (a marker of mature hepatocytes) in Sox9IRES-creERT2; R26-stop-EYFP mice (Figures 3D and 3E). Confocal microscopy and quantification confirmed that EYFP and HNF4α did not co-localize (Figures 3F and 3G). Additionally, Sox9 expression was solely restricted to SPCs in the hURI-tetOFFhep model after 2 weeks of tamoxifen treatment (Figures 3H and 3I). Importantly, we did not detect the presence of hybrid periportal hepatocytes (HybHPs) positive for both Sox9 and HNF4α, which have been reported to participate in liver regeneration, but not HCC. Therefore, only cells positive for Sox9 but negative for HNF4α were tracked in this study. Moreover, Sox9 induction due to hepatocyte toxicity was definitely excluded.

15Crosses of hURI-tetOFFhep and Sox9IRES-CreERT2; R26-stop-EYFP lines showed that at 12 weeks of age, SPCs gave rise to periportal hepatocytes expressing HNF4α and EYFP in control mice (Figures 3J and S3A), suggesting that SPCs supply new periportal hepatocytes. In 12-week-old mutants, newly formed hepatocytes (HNF4α and EYFP positive) invaded the parenchyma (Figure 3J). At 35 weeks of age, we observed EYFP-positive anisokaryotic hepatocyte clusters in the liver periportal region (Figure S3B). Notably, in mutant mice, EYFP-positive hepatocytes derived from SPCs expanded to the parenchyma, and those reaching the central vein expressed cytochrome P450 2E1, indicating that SPCs produce fully differentiated and metabolically functional hepatocytes (Figure S3C). In 65-week-old mice, some tumors were completely EYFP positive (8%), suggesting SPC origin, but in some, EYFP signals were heterogeneous (46%) or non-detectable (46%) (Figures 3K and 3L, tumors 1–3). Thus, SPCs are not the only cells at the origin of liver cancer.

16The above observations, together with previous findings showing that genotoxic stress in hepatocytes initiates hepatocarcinogenesis in hURI-tetOFFhep and DEN-treated mice, suggest that some tumors in these models may derive from hepatocytes. Thus, we followed hepatocyte fate during hepatocarcinogenesis using mice obtained by crossing serum albumin (SA)CreERT2 and R26-stop-EYFP reporter lines. In this context, Cre recombinase activation enables expression of EYFP in all hepatocyte progeny. Recombination efficiency was verified after tamoxifen treatment (Figure 4A). IHC revealed that the EYFP signal was restricted to all hepatcoytes (Figure 4B). No signal was observed in DCs or HPCs (Figure 4B). Co-IF of EYFP and the hepatocytic marker HNF4α revealed a recombination efficiency of 99.95% ± 0.05% (Figures 4C and 4D), suggesting almost complete hepatocyte labeling.

17Next, we crossed hURI-tetOFFhep and SACreERT2; R26-stop-EYFP lines (hereafter “crosses”) (Figure 4E). At 65 weeks of age, immunofluorescence (IF) of liver sections from crosses revealed that the whole tissue (and tumors) was EYFP-positive (Figure 4F, tumor 1; 17% of all cases), suggesting that hepatocytes contribute to hepatocarcinogenesis. However, in some cases, while peritumoral counterparts were EYFP positive, tumors were highly heterogeneous mixtures of EYFP-positive and EYFP-negative hepatocytes (Figure 4F, tumor 2; 49% of all cases), some were even completely EYFP-negative (Figure 4F, tumor 3; 34% of all cases). We detected similar trends in DEN-treated 35-week-old SACreERT2; R26-stop-EYFP mice (Figures S4A and S4B). Notably, recombination efficiency in DEN-treated SACreERT2; R26-stop-EYFP mice was about 100% (98.1% ± 1.6%) (Figures S4C and S4D). Therefore, hepatocytes participate in liver tumorigenesis but are not the exclusive origin of liver tumors.

18We also checked whether EYFP and Sox9 costained in 12-week-old hURI-tetOFFhep and SACreERT2; R26-stop-EYFP mice. No co-IF was detected in liver sections from crosses, indicating that hepatocyte dedifferentiation may not give rise to SPCs (Figures S4E and S4F).

19To determine the contributions of hepatocytes and HPCs to HCC and/or HCA, we evaluated the proportions and types of tumors they predominantly formed. Tumors derived from SACreERT2; R26-stop-EYFP or Sox9IRES-CreERT2; R26-stop-EYFP crosses were first histopathologically classified as either HCC or HCA according to World Health Organization criteria. Interestingly, hepatocytes and HPCs participate in liver tumor heterogeneity, giving rise to RNs, HCAs, and HCCs (Figures 5A and S5A–S5C). Normalization of number of tumors to the percentage of recombination indicated that 66% of EYFP-positive tumors originated from labeled hepatocytes. Interestingly, 69% and 31% of EYFP-positive tumors (66% of total) derived from hepatocytes were HCC and HCA, respectively (Figure 5B). Half of the EYFP-negative tumors were HCC and half were HCA, suggesting that they have distinct cell origins, possibly some from SPCs. Notably, 73% and 27% of all HCCs analyzed from SACreERT2; R26-stop-EYFP crosses originated from hepatocytes and non-hepatocytic cells, respectively (Figure 5C). Almost similar percentages of the HCAs originated from hepatocytic and non-hepatocytic cells (Figure 5C). Thus, hepatocytes principally contribute to HCCs.

20Most EYFP-positive lesions (54% of total) derived from HPCs were benign (Figure 5D), including dysplasia (9%), RNs (63%), and HCA (14%). Surprisingly, only 14% HCCs were detected, possibly originating directly from HPCs or transiting from the RNs or adenomas known to be HCC precursors. Notably, 56% of examined EYFP-negative tumors halved into HCC and benign lesions, including dysplasia (7%), RNs (17%) and HCA (23%) (Figure 5D), suggesting that these EYFP-negative tumors may derive from hepatocytes and/or Sox9-negative cells. Thus, HPCs contribute to liver tumorigeneity (Figure S5C).

21In order to evaluate the aggressiveness of the lesions across different tumor types and to further classify HCC variants obtained from hepatocyte and HPC tracing, resected tumors were further analyzed by IHC, a “HAGCKS” (HSP70, AFP, GS, CK19, Ki67, and Sox9) score with arbitrary units was created to classify tumors aggressiveness, as previously reported (Supplemental Experimental Procedures; Figures S5D–S5V). Fundamentally, aggressive tumors irrespective of their origin have high HAGCKS scores. Expectedly, HCCs had the highest score, whereas HCAs, dysplasia, and RNs scored lower (Figures 5E and 5F). Notably, dysplasia and RNs had apparently a higher HAGCKS score than HCAs, suggesting that these lesions in URI mice are HCC precursors (Figure 5F). Histopathology indicated that HCCs emerging from hepatocytes were heterogeneous and resembled the spectrum of primary human HCC subtypes. Trabecular (41%), mixed (33%), glandular (19%), and solid HCCs (7%) emanated from hepatocyte tracing (Figures 5G). Only trabecular HCCs originated from HPCs (Figure 5G). Interestingly, solid HCCs had a higher HAGCKS score than other HCC variants (Figure 5H), suggesting that HCCs originating from hepatocytes may be slightly more aggressive than HCCs derived from HPCs. Therefore, in the URI model, which recapitulates several stages of human hepatocarcinogenesis, hepatocytes and HPCs contribute to liver tumor heterogeneity.

22Furthermore, hepatocytes and HPCs were tracked in other hepatocarcinogenesis models described previously (Figure S5C). Hepatocytes tracking in R26-mTOM-stop-mGFP mice infected with AAV8-TgbCre and treated with DEN and CCl4 showed that hepatocytes were the HCC cell of origin (Figures 5E and S5C). This was confirmed in Mdr2 knockout (Mdr2KO) mice crossed with R26-stop-ZsGreen line and infected with AAV8-TgbCre (Figures 5E and S5C). All tumors obtained in these mice were HCCs GFP positive from trabecular and mixed types (Figure S5W). However, HCAs or other lesions were not detected (Figures 5E and S5C).

23Surprisingly, tracking of HPCs in Sox9IRES-CreERT2; R26-stop-EYFP mice treated with DEN or in CK19IRES-CreERT2; R26-mTOM-stop-mGFP mice treated with DEN and CCl4 revealed that all HCCs and HCAs obtained were EYFP or GFP negative (Figures 5F and S5C), suggesting that tumors do not originate from HPCs in these models and confirming that tamoxifen treatment of Sox9IRES-CreERT2; R26-stop-EYFP mice does not lead to Sox9 expression in hepatocytes. Thus, different liver cancer mouse models, possibly due to the type of damage (oncogene expression, location, and duration) and their capacity to mimic the human disease may have different impact on liver tumorigenesis.

24The above findings suggest that transformed hepatocytes (e.g., oncogenic URI expression) may non-cell-autonomously instruct and determine the fate of HPCs toward oncogenesis. HPCs expand during tumorigenesis and when liver damage impairs hepatocyte turnover. hURI expression leads to NAD+-depletion-induced DNA damage, triggering hepatocyte damage and apoptosis via the p53-DNA damage response pathway. Thus, relieving the p53-dependent proliferative break in hepatocytes would stop HPC expansion. In order to verify this, we inactivated p53 in hURI-tetOFFhep mice by crossing them with p53ERTAM mice, in which p53 activation requires ectopic 4-hydroxytamoxifen. Surprisingly, while p53 was inactivated, IHC and WB showed that HPC expansion and increased Ki67 was independent of p53 (Figures S6A–S6E). p19ARF was also markedly expressed in hURI-tetOFFhep liver sections. However, p19ARF (Cdkn2a) deletion did not affect HPC expansion and increased Ki67 in hURI-tetOFFhep mice (Figures S6A–S6D and S6F), suggesting that HPCs expand independently of hepatocytes’ proliferative status.

25We next checked several inflammation-regulated factors described to induce oval cell expansion. mRNA levels of cytokines were not altered (Figure S6G). Additionally, levels of TWEAK (tumor necrosis factor-like weak inducer of apoptosis) did not change in early tumorigenesis stages (Figure S6H), and WNT/β catenin target genes were not upregulated in HPCs from hURI-tetOFFhep and Sox9IRES-EYFP crosses (Figure S6I). Finally, treating 3-week-old hURI-tetOFFhep mice with anti-inflammatory sulindac did not abolish oval cell expansion (Figures S6J–S6L). Thus, inhibition of HPC expansion is independent of inflammatory cues in our model.

26Switching off hURI expression with doxycycline significantly reduced HPC proliferation (Figures 6A and 6B), suggesting that continuous hURI expression is essential for oval cell expansion. Abrogation of DNA damage by supplying the NAD+ booster nicotinamide riboside (NR) reduced HPC proliferation in 8-week-old mice (Figures 6C–6E), suggesting that the NAD+-deficit-induced DNA damage axis is essential for oval cell expansion.

27Previous iTRAQ analysis revealed that galectin-1 and galectin-3 were highly upregulated in 8-week-old hURI-expressing livers (Figure S6M). Galectins are extracellular β-galactoside-binding lectin, which bind to glycoproteins such as laminin and integrins (also expressed in mutant livers; Figures S2A, S2D, and S2E), to regulate and remodel the ECM and promote integrin signaling and fibrillogenesis, allowing HPC expansion during chronic liver injury. WB confirmed that galectin-3 was strikingly enhanced in 12-week-old mutant livers, but galectin-1 was only modestly increased (Figure S6N). WB and IF of 8-week-old hURI-tetOFFhep livers confirmed that galectin-3 was upregulated in hepatocytes (Figures 6E and S6O). Abrogation of DNA damage by NR reduced galectin-3 levels (Figure 6E), suggesting that hepatocytic NAD+-deficit-induced DNA damage may be involved in galectin-3 secretion. Suppression of glycosylation-dependent surface glycoprotein by 4-fluoro-N-acetylgalactosamine (4-F-GalNAc), which inhibits the actions of galectin, significantly reduced HPC expansion in hURI-tetOFFhep livers (Figures S6P–S6R). Thus, transformed hepatocytes are the source of galectin-3, which is essential for HPC expansion and tumorigenesis, highlighting a crosstalk between hepatocytes and HPCs.

28To further test galectin-3-mediated activation of HPCs by transformed hepatocytes, mouse bipotential murine oval liver (BMOL) cells were cultured with conditioned media from AML-12 (alpha mouse liver 12) cells (nontumorigenic murine hepatocytes) stably expressing either EGFP or hemagglutinin (HA) URI. A concomitant knockdown of galectin-3 in AML-12 cells overexpressing HA-URI significantly reduced BMOL cell numbers (Figures 6F–6H). Similar results were obtained when hepatocytes were transformed by oncogenic c-MYC (data not shown), which is known to be implicated in hepatocarcinogenesis, suggesting that transformed hepatocytes, through DNA damage, talk to HPCs via galectin-3 secretion.

29Next, hURI-tetOFFhep mice crossed with heterozygous galectin-3 knockout mice had significantly reduced HPCs and dysplastic lesions and number of hepatic tumors (Figures 6I–6N). Notably, levels of galectin-3 were significantly much higher in the URI model than in DEN-treated mice (Figures S6S and S6T). Galectin-3 is thus determinant for HPC expansion and their tumorigenic fate in the hURI-tetOFFhep, but not in DEN-treated mice.

30To estimate the clinical relevance of our findings, we interrogated the galectin family gene signature using gene expression dataset from a cohort of 221 HCC patients (GEO: GSE14520). Multivariate Cox regression indicated that the overall survival of HCC patients was significantly associated with LGALS3, LGALS4, and LGALS7 gene signatures (p = 0.025) (Figure S6U), but not with all combined LGALS genes (data not shown). Importantly, LGALS3 gene expression positively correlated with patient poor prognosis (Figure 6O), whereas LGALS4 and LGALS7 gene expression did not show significant differences (Figures S6V and S6W). Furthermore, the multivariate Cox regression suggested a significant association between URI (C19orf2) or c-MYC and LGALS3 expression and predicted poor patient survival (p = 0.027 for URI and p = 0.023 for c-MYC) (Figures 6P and S6X). Thus, galectin-3 activates HPC, negatively affecting overall patient survival.

31To probe signaling changes that may explain why HPCs do not immediately differentiate into hepatocytes while proliferating, we further analyzed our previously reported iTRAQ data. Several glycolysis enzymes, including hexokinase (HK1), glyceraldehyde-3-phosphate dehydrogenase, and pyruvate kinase, were upregulated in early tumorigenesis in hURI-tetOFFhep hepatocytes (Figures 7A and S7A), suggesting a pseudo-Warburg-like effect. Accordingly, pyruvate dehydrogenase kinase, which inactivates pyruvate dehydrogenase (catalyzing pyruvate to acetyl-coenzyme A conversion), was downregulated (Figures 7A and S7A). Many other enzymes involved in the tricarboxylic acid (TCA) cycle acting downstream of α-ketoglutarate were also downregulated (Figures 7A and S7A). Thus, glycolytic flux may result in α-ketoglutarate accumulation, maintaining the undifferentiated state of HPCs in the liver.

32In accordance with our iTRAQ data, α-ketoglutarate was significantly elevated in mutant livers (Figure 7B) but decreased following NR treatment, suggesting that hepatocytic NAD+-depletion-induced DNA damage is essential for α-ketoglutarate elevation (Figure 7B). Further, 4-F-GalNAc treatment did not affect the α-ketoglutarate levels (Figure 7C), suggesting that α-ketoglutarate accumulation is independent of changes in glycan structures. Treatment of BMOL cells with dimethyl 2-oxoglutarate (or dimethyl-α-ketoglutarate), a precursor for the glutamine synthesis, significantly increased expression of stem cell markers such as Cd44, Sox9, or Epcam, but expression of Hnf4a was reduced, as shown by qRT-PCR (Figure S7B). This suggests that increased α-ketoglutarate maintains HPC undifferentiated state. Addition of dimethyl succinate and dimethyl 2-oxoglutarate further enhanced the expression of stem cells markers (Figure S7B), while dimethyl succinate favored an intermediate state, confirming previous work and showing that the α-ketoglutarate/succinate ratio suppresses stem cell differentiation. Additionally, media from AML-12 cells stably expressing HA-URI and treated with bis-2-(5-phenylacetamido-1,3,4-thiadiazol-2-yl)ethyl sulfide (BPTES), a selective inhibitor of glutaminase GLS1 reducing α-ketoglutarate levels, decreased BMOL cell numbers (Figures 7D and 7E). Notably, media from HA-URI-expressing AML-12 cells had higher levels of α-ketoglutarate (Figure 7F). However, in vitro, only few stem cell markers were reduced (Figure 7G), suggesting that other paracrine signals (e.g., galectin-3) emerging from transformed hepatocytes contribute to preserve the HPC stemness.

33Multivariate Cox regression of a gene set of intermediate metabolism downregulated in URI model revealed a nonsignificant association between expression of IDH1, IDH2, IDH3A, IDH3B, IDH3G, OGDH, SCS, SDH, or PC3 and overall patient survival (Figure S7C). Expression analysis of a single gene from intermediate metabolism also failed to classify the patients with poor prognosis (data not shown). We therefore checked whether galectin-3 was sufficient to sustain the undifferentiated state of HPCs. BMOL cells cultured with media from AML-12 cells expressing HA-URI or His-c-MYC and depleted for galectin-3 showed a significant reduction of all stem cell markers (Figures 7H, 7I, and S7D–S7G). Galectin-3 secreted by transformed hepatocytes may thus play a prominent role in maintaining HPC stemness, expansion, and aggressiveness (Figure 7J).

34Different murine models, possibly due to their capacity of mimicking the human disease or the type of damage (oncogene expression, location, and duration) impact differently on tumorigenesis. In the hURI-tetOFFhep mouse model, which recapitulates some features and stages of the human hepatocarcinogenesis, hepatocytes and HPCs contribute to liver tumor heterogeneity. Supporting this, transplanted HPCs from normal mice regenerate hepatocytes and restore liver parenchyma in mice with hepatocytic deletion of the E3 ubiquitin ligase Mdm2-induced apoptosis, necrosis, and foremost senescence. Consistently, in the hURI-tetOFFhep model, HPCs clearly supply new mature hepatocytes, even though we do not completely exclude that hepatocytes can also derive from pre-existing hepatocytes. Interestingly, hepatocytes expressing URI show high levels of replicative stress and undergo into senescence, arguing that the type of hepatocyte damage may determine HPC fate during tumorigenesis. HPCs are therefore a potential future alternative to hepatocyte or liver transplantation for liver disease, depending on the damage, and can participate in hepatocarcinogenesis.

35Proliferation block or apoptosis of hepatocytes triggers growth of HPCs. In our study, HPC proliferation seems to depend on the hepatocyte’s oncogenic status rather than its proliferative capacity. Transformed hepatocytes instruct the neighboring HPCs through paracrine signal mediated by NAD+-deficit-induced DNA damage, leading to galectin-3 and α-ketoglutarate secretion, to maintain an HPC undifferentiated state and activation. In this regard, galectins are implicated in fibrosis of multiple organs. Specifically, galectin-3 is reportedly activated in cirrhotic liver and HCC, and high circulating levels are associated with poor prognosis in patients with primary HCC, corroborating our results.

36Activated HPCs contribute to the formation of RNs, HCAs, and HCCs. Notably, shared mutations by HPCs and RNs putatively indicate common origins. Moreover, RNs mainly occur in cirrhotic livers and define histological development of cirrhosis, a potent risk factor for HCC. RNs are thus likely HCC precursors. Thus, HPC activation may engender cirrhosis and subsequently, through selective pressure, HCC formation.

37HCA is a rare benign liver tumor primarily found in women using hormonal contraceptives. The etiology is uncertain, but direct linkage between estrogen exposure and HCA growth is suspected, and estrogens reportedly contribute to the proliferative responses of biliary epithelial cells. Interestingly, HCC may result from malignant transformation of HCA, even though this transformation does not occur in non-fibrotic livers. Finally, patients with trabecular HCCs often have NASH. Accordingly, hURI-tetOFFhep mice develop NASH and trabecular HCCs.

38Transplantation of DC-induced organoids has been recently proposed to promote liver regeneration, especially for patients with resected HCC. Since HPCs give rise to RN-induced cirrhosis, the re-implantation strategy of HPCs from HCC patients should be approached cautiously. Liver transplantation remains the only curative option for some patients, but pharmacological blockage of galectin-3 may offer elegant approaches for inhibiting cirrhosis progression and HCC, thereby reducing needs for transplantation. Finally, HCC is an intractable primary liver cancer, certainly due to the recurrence of multiple secondary liver tumors after surgical interventions or first therapies. HPCs predict recurrence in HCC patients after curative resection. Our findings may thus provide a disease management strategy after primary surgical resection to circumvent tumor recurrence, facilitating the design of efficient therapies against liver cancer and recurrences.

39Materials and antibodies were purchased from different manufacturers as detailed in Supplemental Experimental Procedures.

40Mice were generated, handled, and treated as previously described and detailed in Supplemental Experimental Procedures. All mice have been housed in pathogen-free conditions. All experiments were approved by the CNIO-ISCIII Ethics Committee and performed in accordance with the guidelines for ethical conduct in the care and use of animals as stated in the international guiding principles for biomedical research involving animals, developed by the Council for International Organizations of Medical Sciences. Littermates were always used as controls. No gender differences were observed and age/developmental stage of mice was included appropriately in the text and the figure legends. Food (Harlan Laboratories and Research Diets Inc.) and water were provided ad libitum.

41Cell culture, transfection, and cell number counts were performed as detailed in Supplemental Experimental Procedures.

42Six images per slide were taken at 10× and 20× magnifications and quantified using color de-convolution, a colocalization finder, and image analysis tools available in Fiji/ImageJ software (http://fiji.sc/).

43Consecutive liver sections were stained by immunohistochemistry for HSP70, AFP, GS, CK19, Ki67, and Sox9 (HAGCKS). Differential expression levels (0–3) of GS and HSP70 as well as Sox9 were evaluated using previously reported classifications. CK19 and AFP were only classified in positive (1) or negative (0) cases. Individual cases were considered immunoreactive (IR) for individual antigens when more than 5% of cells were IR. IR cases were further sub-classified as follows: 5%–10% IR cells (1), 11%–50% IR cells (2), and more than 50% IR cells (3). HAGCKS score was determined by adding the grading of immunohistochemistry signal intensity for each individual marker and scored in Figure 5.

44Total α-ketoglutarate levels were measured according to the manufacturer’s instruction provided by a kit purchased from Sigma-Aldrich (MAK054A). For liver tissue, the protocol was scaled up to two times.

45Serum TWEAK levels were measured according to the manufacturer’s instructions using a Mouse TWEAK/TNFSF12 ELISA kit purchased from R&D Systems (DY1237).

46Flow cytometry of HPCs was performed as detailed in Supplemental Experimental Procedures.

47Immunoblotting was performed as previously described elsewhere.

48Immunohistochemistry and immunofluorescence were performed as previously described.

49Genotyping and qRT-PCR (including primer sequences) were performed as detailed in Supplemental Experimental Procedures.

50Statistical analyses were performed using GraphPad Prism V5.0 software. Statistical significance (∗p ≤ 0.05, ∗∗p ≤ 0.01, and ∗∗∗p ≤ 0.001) between the means of a minimum of three groups was determined using unpaired two-tailed Student’s t test or linear regression analysis. Results are expressed as the mean value ± SEM or SD, as indicated. All results, including WB, are representative of at least three independent experiments. Statistical parameter false discovery rate estimates the probability of a gene set with false-positive finding. Normalized enrichment score allows to compare enrichment analysis results across gene sets. Tumor numbers normalization in the Sox9IRES-CreERT2 model was done according to Cre recombination efficiency, and contingency analysis was performed using two-sided Fischer’s exact tests (results are displayed in Figures S5A and S5B). Survival and Kaplan-Meier analyses for individual genes were performed using publically available human HCC patient gene expression datasets (GEO: GSE14520). Multivariate Cox regression analyses were performed using SPSS software (version 19).

51K.S.T. designed performed and analyzed most of the in vivo experiments. M.B. designed performed and analyzed in vitro experiments and experiments with the Gal3KO mice. K.S.T., M.B., A.T., and C.P. performed and analyzed HAGCKS scores. C.P. analyzed all tissues histology. O.G. performed the bioinformatics analyses. R.F.S. provided some paraffin-embedded blocks used in HAGCKS scoring and previously reported. K.S.T., M.B., and N.D. analyzed all the data. N.D. designed the experiments and conceived, developed, and wrote the project and the manuscript. Funding was secured by N.D.