Article Text


Original article
β-catenin-activated hepatocellular carcinomas are addicted to fatty acids
  1. Nadia Senni1,2,3,4,
  2. Mathilde Savall1,2,3,4,
  3. David Cabrerizo Granados1,2,3,4,
  4. Marie-Clotilde Alves-Guerra1,2,3,4,
  5. Chiara Sartor1,2,3,4,
  6. Isabelle Lagoutte1,2,3,
  7. Angélique Gougelet1,2,3,4,
  8. Benoit Terris1,2,3,4,5,
  9. Hélène Gilgenkrantz1,2,3,4,
  10. Christine Perret1,2,3,4,
  11. Sabine Colnot1,2,3,4,
  12. Pascale Bossard1,2,3,4
  1. 1 INSERM, U1016, Institut Cochin, Paris, France
  2. 2 CNRS, UMR8104, Paris, France
  3. 3 Université Paris Descartes, Sorbonne Paris Cité, Paris, France
  4. 4 Equipe Labellisée Ligne Nationale Contre le Cancer, Paris, France
  5. 5 Pathology Department, APHP, Hôpitaux Universitaires Paris Centre, Hôpital Cochin, Paris, France
  1. Correspondence to Dr Pascale Bossard, Inserm U1016, Institut Cochin, CNRS UMR8104, Université Paris Descartes, Paris75014, France; pascale.bossard{at}


Objectives CTNNB1-mutated hepatocellular carcinomas (HCCs) constitute a major part of human HCC and are largely inaccessible to target therapy. Yet, little is known about the metabolic reprogramming induced by β-catenin oncogenic activation in the liver. We aimed to decipher such reprogramming and assess whether it may represent a new avenue for targeted therapy of CTNNB1-mutated HCC.

Design We used mice with hepatocyte-specific oncogenic activation of β-catenin to evaluate metabolic reprogramming using metabolic fluxes on tumourous explants and primary hepatocytes. We assess the role of Pparα in knock-out mice and analysed the consequences of fatty acid oxidation (FAO) using etomoxir. We explored the expression of the FAO pathway in an annotated human HCC dataset.

Results β-catenin-activated HCC were not glycolytic but intensively oxidised fatty acids. We found that Pparα is a β-catenin target involved in FAO metabolic reprograming. Deletion of Pparα was sufficient to block the initiation and progression of β-catenin-dependent HCC development. FAO was also enriched in human CTNNB1-mutated HCC, under the control of the transcription factor PPARα.

Conclusions FAO induced by β-catenin oncogenic activation in the liver is the driving force of the β-catenin-induced HCC. Inhibiting FAO by genetic and pharmacological approaches blocks HCC development, showing that inhibition of FAO is a suitable therapeutic approach for CTNNB1-mutated HCC.

  • liver metabolism
  • hepatocellular carcinoma
  • lipid oxidation

Statistics from

Significance of this study

What is already known on this subject?

  • The Wnt/β-catenin pathway is the most frequently dysregulated pathway in human hepatocellular carcinoma (HCC). Despite an improvement in the management of HCCs, they rank as the second leading cause of cancer-related deaths as current treatments remain poorly effective.

  • For the past years, targeting metabolic reprogramming has emerged as a promising novel therapeutic strategy, but the metabolic reprogramming of CTNNB1-mutated HCCs has not yet been deciphered.

What are the new findings?

  • β-catenin-induced HCC does not use the Warburg effect described in many tumours.

  • Fatty acid oxidation is the main energy provider of β-catenin-induced HCC.

  • This metabolic rewiring is orchestrated by the transcription factor peroxisome proliferator-activated receptor alpha (PPARα).

  • Fatty acid oxidation is mandatory to the transformation process of β-catenin-induced HCC.

  • Identification of the inhibition of fatty acid oxidation is a potential therapeutic strategy for β-catenin-induced HCC.

How might it impact on clinical practice in the foreseeable future?

  • These findings show that inhibiting fatty acid oxidation could be a pertinent therapeutic approach for β-catenin-induced HCC since such treatment was able to block their development.


Tumourigenesis is a multistep process involving pathways modifications that promote unchecked proliferation and abrogate cell death. Such uncontrolled proliferation requires metabolic reprogramming to provide macromolecules for cell growth and division in order to thrive in a potentially deleterious environment.1 2 Metabolic reprogramming in cancer is under the control of oncogenes, tumour suppressor genes and altered signalling pathways. Moreover, external factors, such as hypoxia and nutrient poor environments, may influence such metabolic editing to allow the survival, growth and migration of tumour cells. The most common metabolic feature observed in tumour cells is high glucose consumption via a glycolytic pathway producing lactate at the expense of the oxidative pathway despite the presence of oxygen.3 This aerobic glycolytic pathway, known as the Warburg effect, has been considered to be the staple of cancer cell metabolism.4 This pathway also contributes to the rerouting of glucose carbons into macromolecule synthesis and production of reduced equivalents.5 Glucose-dependent tumours are often dependent on glutamine, allowing anaplerosis of the Krebs cycle.6 Such glucose addiction is at the root of their detection by positron emission tomography scan fluorodeoxyglucose. However, recent work has highlighted the diversity of nutrients, other than glucose, that contribute to the metabolic functions of cancer cells.7

Hepatocellular carcinoma (HCC) is a highly heterogeneous disease that mostly arises in the context of chronic liver inflammation associated with hepatitis B or C infections, alcohol consumption, obesity or genotoxic exposure.8 9 Genomic approaches have allowed their classification, drawing their genetic landscape and identifying the altered molecular pathways.10–12 These studies led to the classification of HCCs into two large subclasses: one proliferative subclass associated with high chromosomal instability, enrichment of signals associated with cell proliferation and cell cycle progression and an aggressive phenotype. Some are steatotic. The first class contains HCC with mutations in TP53 or AXIN1,11 whereas the second class includes tumours with activating mutations in CTNNB1, encoding β-catenin.13 14 Mutant β-catenin translocates to the nucleus and activates the β-catenin transcription programme, promoting tumourigenesis. This class of HCCs follows a peculiar pattern of pathogenesis as they mostly conserve the molecular traits of differentiated hepatocytes, exhibit a high chromosomal stability, lower proliferation index and a very distinctive metabolic morphotype. There are often cholestatic and infrequently steatotic.15

We investigated the metabolic adaptations triggered by β-catenin oncogenic activation in hepatocytes and asked whether this metabolic reprogramming is a driving force of β-catenin-induced HCC tumourigenesis. We used a mouse model recapitulating CTNNB1-mutated HCC16 and first showed that the metabolic reprogramming of β-catenin-activated HCC does not conform to the Warburg effect but leads to the induction of fatty acid oxidation (FAO). Indeed most of the CTNNB1-mutated HCC expressed a FAO programme. Importantly, we showed that both the initiation and progression of β-catenin-induced HCC were dramatically impaired following the genetic deletion of peroxisome proliferator activated receptor alpha (Pparα), a master inducer of FAO. Finally, treatment with the FAO inhibitor, etomoxir, halted HCC development, highlighting a new therapeutic tool for CTNNB1-mutated HCC.


β-catenin-activated tumours do not conform to the ‘Warburg effect’

We investigated metabolic reprogramming in β-catenin-activated HCC using the previously described (and experimental procedures) ApcTumLIV mouse, which mimics CTNNB1-mutated HCC tumourigenesis.16–19

Tumour development was followed by echography (online supplementary figure S1A). We detected the first tumours 5 months after Apc deletion. Tumour penetrance was 100% by 7 months (online supplementary figure S1B). As expected, all Apc-HCCs in the ApcTumLIV mice showed a high level of nuclear β-catenin accumulation and glutamine synthetase (Glul) immunostaining, a surrogate marker for β-catenin-activated HCC. In contrast, non-tumour hepatic tissues showed a normal Glul expression pattern, restricted to hepatocytes surrounding the centrolobular vein17 20 (online supplementary figure S1C).

Supplementary file 1

Apc-HCC metabolism was analysed when tumours reached an average diameter of 7 mm. We first analysed the fate of glucose and observe no modification in lactate production by the Apc-HCC explants compared with the non-tumour explants (figure 1A). Furthermore, glucose oxidation rates between tumour and adjacent non-tumour tissues were similar when cultured in presence of [U-14C]glucose (figure 1A). These results show that the Apc-HCCs were not glycolytic. Moreover, [U-14C]glucose incorporation into lipids was much lower in the tumour than non-tumour tissues, indicating diminished lipogenesis rates from glucose in Apc-HCC (figure 1A). Overall, our results show that Apc-HCCs do not use glycolysis as an energy source and do not exhibit the classical Warburg effect described for many tumours.

Figure 1

No Warburg effect and a sharp increase in FAO in β-catenin-activated tumours. (A) Tumour (T) and non-tumour (NT) explants from ApcTumLIV mice were cultured for 2 hour with 25 mM [U-14C]glucose and glucose metabolism assayed. Far left panel: scheme of the fate of [14C]glucose (in red), left panel: lactate production, middle panel: glucose oxidation rates into [14CO2] and right panel: esterification rates, (n=3). Six tumour loads and their non-tumour counterparts were analysed in three independent experiments. (B) Oil Red O staining of neutral lipids on sections encompassing NT and T explants. (C) NT and T tissue explants were cultured for 2 hours in media containing 0.3 mM [1-14C]oleate. Far left panel: scheme of the fate of [14C]oleate (in red), left panel: oleate oxidation into [14C] CO2, middle panel: [14C] ASP and right panel: ketone bodies. (D) Left panel: esterification rates, middle panel: tissue explants triglycerides content, right panel: iodine staining of a TLC plate of NT and T explants lipid extracts. L, control lipid mixture; PL, phospholipids; TAG, triacylglycerol. Seven tumour loads and their non-tumour counterparts were analysed in three independent experiments. Results are presented as the means±SEM of triplicate flasks from three independent experiments. ASP, acid soluble products; DAG, diacylglycerol; FAO, fatty acid oxidation; TLC, thin layer chromatography.

β-catenin-activated tumours intensively oxidise fatty acids

One of the hallmark features of CTNNB1-mutated HCC is the very infrequent presence of steatosis.15 Apc-HCC showed very low lipid accumulation during the fasting state, whereas lipid droplets were clearly visible in the non-tumour tissue of the liver (figure 1B). Indeed, during fasting periods, lipolysis of adipose tissue provides the liver with non-esterified fatty acids, which are oxidised in the mitochondria, excess NEFA being temporarily stored as triglycerides in small lipid droplets characteristic of fasting livers (figure 1B). We evaluated oleate metabolism by measuring its oxidation into CO2, its reorientation into ketone bodies and its esterification into lipids in Apc-HCC tumour and non-tumour explants cultured with albumin conjugated [1-14C]oleate. Examination of [14C]oleate oxidation in tumour explants showed a higher increase in [14CO2] production (two fold) and a higher level of [14C]acid soluble products (ASP) production than in adjacent non-tumour explants (figure 1C). Ketogenesis, the synthesis of ketone bodies from the acetyl-Coenzyme A produced by FAO, occurs only in hepatocytes and reflects active β-oxidation in these cells.21 Ketone body production was significantly higher in the tumour than the non-tumour explants (figure 1C). However, oleate esterification rates into lipids were much lower in tumour than non-tumour tissues (figure 1D). Overall, there was considerable rerouting of fatty acids into an oxidative pathway at the expense of esterification in Apc-HCC. Accordingly, Apc-HCC contained 10-fold lower triglycerides than non-tumour tissues (figure 1D). Overtime, the continuous daily cycle of higher FAO, lower lipogenesis and lower esterification rates led to the overall lack of a steatotic phenotype in Apc-HCC (figure 1B).

To corroborate the reprogramming of CTNNB1-mutated HCC towards FAO, we assessed the expression of an FAO programme using the human dataset GSE62232, which includes 81 human tumours annotated for their genetic alterations.22 23 Unsupervised hierarchical clustering showed that 20 of 26 CTNNB1-mutated HCCs were gathered in a single cluster expressing the FAO programme at the highest level (figure 2A).

Figure 2

Human CTNNB1-mutated HCCs express a FAO programme. (A) Non-supervised hierarchical clustering of the GSE62232 data set of human HCC. Hierarchical clustering was performed using Genesis software. Euclidean distance and average linkage arrangement for both axes was used and shows the FAO gene signature described in Ingenuity Pathway Analysis (IPA). Gene expression data were normalised using the z-score transformation. CTNNB1 and AXIN1 mutations are shown by orange and blue marks, respectively. The genes known to be involved in FAO are highlighted with a black mark. (B) mRNA relative expression of fatty acid catabolism enzymes in 21 mutated CTNNB1 HCC (M), 21 non-mutated CTNNB1 HCC (NM) or in six non-tumour liver tissues (NT). (C) Hepatoma cell lines were cultured for 2 hours in a media containing 0.3 mM [1-14C]oleate. Left panel: PPARα mRNA expression in hepatoma cell lines; middle panel: oleate oxidation into [14C] CO2; right panel: [14C] ASP production. n=2 independent experiments in triplicates. ASP, acid soluble products; FAO, fatty acid oxidation; HCC, hepatocellular carcinoma; PPARα, peroxisome proliferator-activated receptor alpha.

We confirmed these results using an independent cohort of human HCC mutated (21 samples) or non-mutated for CTNNB1 (21 samples) and six non-tumour samples. We analysed in this cohort the expression levels of several genes from the FAO programme. We observed an overall increase of the FAO programme in the CTNNB-mutated HCCs compared with non-mutated HCC (figure 2B).

We then analysed FAO in hepatoma cells with or without CTNNB1 mutations. HepG2 and SNU 398 are two hepatoma cell lines with gain-of-function mutations of CTNNB1 (p.25_140 deletion and p.Ser37Cys, respectively, both in the DNA sequence encoding for β-catenin degradation domain), while the Huh7 cell line carries no CTNNB1 mutation. We observed a significantly higher FAO activity in both CTNNB1 mutated cell lines (HepG2 and SNU 398) compared with the non-mutated cells (Huh7) (figure 2C).

All the AXIN1-mutated HCC were excluded from the CTNNB1 cluster (figure 2A), displaying lower levels of FAO. This correlates with our recent data showing that, although surprising, AXIN1-mutated HCCs express a genetic programme highly divergent from that of CTNNB1-mutated HCC, which is frequently not linked to the Wnt canonical pathway10 24 (figure 2A).

We experimentally validated these data by characterising the metabolic reprogramming of tumours developed in a model in which HCC is induced following the deletion of Axin1 in hepatocytes, called Axin1-HCC.24 Axin1 deletion does not lead to β-catenin activation in these tumours24 (online supplementary figure SI1D). Axin1-HCC exhibited the Warburg effect, with decreased glucose oxidation and increased lactate production (online supplementary figure SI2A). In addition, FAO in Axin1-HCC was much lower than in non-tumour tissues with higher incorporation into triglycerides (online supplementary figure SI2B). Elevated lipogenesis and reduced FAO in Axin1-HCC tumours may thus contribute to their steatotic phenotype.

Supplementary file 2

Overall, our results demonstrate that the aberrant β-catenin oncogenic programme is associated with metabolic reprogramming, using FAO as the energy source. Such reprogramming was found in most CTNNB1-mutated HCC.

FAO is the driving force for HCC development in Apc-HCC

We investigated whether the peculiar metabolic profile found in Apc-HCC was acquired during transformation or was a cell-autonomous driving force of tumourigenesis. We thus analysed the metabolic consequences of Apc loss in pretumoural hepatocytes. We obtained panlobular β-catenin oncogenic activation in hepatocytes following acute Apc deletion. This pretumoural model is called ApchepKO and has already been described.16 18

We used primary hepatocyte culture and explants of ApchepKO and control livers to evaluate glucose metabolism by measuring [U-14C]glucose oxidation into [14CO2]. ApchepKO hepatocytes produced the same amount of [14CO2] per mole of glucose as the control (figure 3A). There was also no difference in pyruvate production and only slightly less lactate production (figure 3A). Accordingly, there was no difference in lactate dehydrogenase activity or extracellular glucose concentrations between ApchepKO and control cells (online supplementary figure SI3A). Pyruvate entry into the Krebs cycle relies on its conversion into acetyl-CoA by pyruvate dehydrogenase of which the expression and activity were not affected in ApchepKO cells (online supplementary figure SI3B), indicating that mitochondrial acetyl-CoA production rates from glucose did not differ between ApchepKO and control hepatocytes.

Supplementary file 3

Figure 3

β-catenin overactivation does not alter glycolytic rates in pretumoural hepatocytes. Primary hepatocytes isolated from Apc lox/lox/TTR-CreTam (ApchepKO) or Apc lox/lox (control) mice 6 days after tamoxifen injection were cultured for 24 hours with 25 mM glucose and 100 nM insulin. (A) [U-14C] glucose was added for the last 3 hours of culture. Left panel: glucose oxidation rates into [14CO2]; middle panel: lactate and pyruvate concentrations in the culture media; right panel: quantification of esterification rates into TAG, DAG and PL after thin layer chromatography (TLC) of lipid extracts. Far right panel: [14C] incorporation into RNA (n=4 independent experiments in triplicate). (B) Left panel: immunoblot analysis of cellular extracts using specific antibodies for Acss2 and Gapdh; right panel: [1–2 14C] acetate was added for the last 2 hours of culture to assay de novo lipid synthesis from acetyl-CoA. After TLC, [14C] incorporation into NEFA, TAG, DAG and PL was quantified. n=2 independent experiments in triplicate. (C) Left panel: immunoblot analysis of cellular extracts using specific antibodies for Acac, Fasn, Lpin1 and γ tubulin; right panel: malonyl-CoA concentrations. Results are presented as the means±SEM of triplicate flasks from independent experiments. Acac, acetyl-CoA carboxylase; DAG, diacylglycerol; Fasn, fatty acid synthase; Lpin1, lipin1; NEFA, non-esterified fatty acid; PL, phospholipids; TAG, triacylglycerol.

We also investigated [14C] incorporation into RNA. The proportion of [14C]RNA relative to the total amount was significantly higher in ApchepKO cells, indicating that part of the [U-14C]glucose was incorporated into RNA and likely contributed to biomass synthesis (figure 3A).

After entering the Krebs cycle, through lipogenesis, glucose carbons are then rerouted into fatty acids that are esterified into lipids. We investigated the incorporation of glucose carbons into [14C]triacylglycerol (TAG), [14C]diacylglycerol (DAG) and [14C]phospholipids (PL). The levels of [14C]DAG and [14C]TAG were significantly lower (1.5-fold and 2-fold, respectively) in ApchepKO than control hepatocytes, whereas the level of [14C]PL was slightly but significantly higher (1.3-fold) (figure 3A). Glucose carbon rerouting into PL is associated with increased expression of choline/ethanolamine phosphotransferase 1, an enzyme that catalyses the first step of PL synthesis (Sartor et al in preparation). However, examination of total esterification showed that fewer lipids were synthesised from glucose in ApchepKO than control hepatocytes.

We confirmed dysregulation of de novo lipogenesis using [1–2 14C]acetate. Levels of synthesised [14C]NEFA were at least twofold lower in ApchepKO than control hepatocytes, and the esterification profiles into labelled TAG, DAG, and PL were similar to those observed with [14C]glucose (figure 3B). Cytoplasmic acetyl-CoA synthesis from acetate is catalysed by acyl-CoA synthetase short chain 2 (Acss2) and then used as a substrate by the acetyl-CoA carboxylase (Acac) and fatty acid synthase (Fasn). The level of Acss2 expression did not differ between ApchepKO livers and controls, but expression of the three lipogenic enzymes, Acac, Fasn and lipin1 (Lpin1) was lower leading to a diminished flux of acetyl-CoA through the lipogenesis pathway in ApchepKO livers (figure 3B and C). Accordingly, malonyl-CoA levels were significantly lower in ApchepKO livers than controls (figure 3C).

Studies have shown that tumours can also use glutamine as a lipogenic substrate.25 We therefore evaluated [14C]glutamine metabolism by measuring its oxidation into [14CO2] and [14C] incorporation into lipids. There was significantly lower glutamine oxidation and lower incorporation of its carbons into triglycerides in ApchepKO than control cells (figure 4A,B). Altogether, depressed lipogenesis from both glucose and glutamine carbons contributes to the marked drop in triglyceride content observed in ApchepKO hepatocytes (figure 4B,C) and the lower amount of long-chain fatty acids present in the fed state that we observed in our previous lipidomic analysis.19 At the same time, ApchepKO livers show an accumulation of glutamine,19 linked to high levels of Glul, the surrogate marker of β-catenin signalling activation in hepatocytes.15 16 20 Consequently, poor or absent glutamine catabolism in ApchepKO hepatocytes was not unexpected. Glutamine anabolism from glucose has already been described using [U-13C]glucose, in other tumour models overexpressing β-catenin and Glul in the liver.25 To support this hypothesis, we assessed the fate of pyruvate in mitochondria by following pyruvate oxidation into the Krebs cycle using [2-14C]pyruvate. [2-14C]Pyruvate only generates [14CO2] during the second round of the Krebs cycle. [2-14C]Pyruvate oxidation was significantly lower in ApchepKO than control hepatocytes, indicating that part of the pyruvate was redirected to another pathway, likely glutamine synthesis, in addition to the Krebs cycle (figure 4D).

Figure 4

β-catenin-activated hepatocytes are not addicted to glutamine. (A) Primary mouse hepatocytes isolated from Apc lox/lox/TTR-CreTam (ApchepKO) or Apc lox/lox (control) mice 6 days after tamoxifen injection were cultured for the last 3 hours in the presence of [14C] glutamine, and glutamine oxidation into [14CO2] was analysed. n=3 independent experiments in triplicate. (B) Left panel: quantification of [14C] incorporation from glutamine into TAG, DAG and PL after TLC separation and right panel: intracellular triglycerides content at the end of the experiment. Results are presented as the means±SEM of triplicate flasks from three independent experiments. (C) Neutral lipid Oil Red O staining of frozen sections from control and ApchepKO livers. (D) Oxidation rates of [2-14C] pyruvate into [14CO2] in primary hepatocytes isolated from Apcko (black) or control mice (white). n=3 independent experiments with triplicates. DAG, diacylglycerol; PL, phospholipid; TAG, triacylglycerol; TLC, thin layer chromatography.

Elevated FAO is a prominent feature of Apc-HCC. We thus evaluated [14C]oleate and [14C]palmitate metabolism in primary hepatocytes isolated from ApchepKO livers. Their oxidation to [14CO2] and incorporation into ASP were significantly higher than in control hepatocytes (figure 5A and online supplementary figure SI4A). Etomoxir, an irreversible inhibitor of carnitine palmitoyltransferase 1 (Cpt1) (a FAO rate-limiting enzyme), efficiently blunted FAO in ApchepKO hepatocytes (online supplementary figure SI4B). We observed similar elevated FAO rates in cultured ApchepKO liver explants (online supplementary figure SI4C). In addition, the overall esterification rates of [14C]oleate were significantly lower in ApchepKO hepatocytes, indicating the metabolic reorientation of long-chain fatty acid towards oxidation at the expense of esterification (figure 5A and online supplementary figure SI4D). Moreover, FAO still occurred at a higher rate in ApchepKO hepatocytes under lipogenic conditions (ie, 25 mM glucose and 100 nM insulin) of malonyl-CoA synthesis, the allosteric Cpt1 inhibitor, whereas FAO in control hepatocytes was repressed, as expected (figure 5B).

Supplementary file 4

Figure 5

β-catenin-activated hepatocytes preferentially oxidise fatty acids. Primary hepatocytes isolated from Apc lox/lox/TTR-CreTam (ApchepKO) or Apc lox/lox (control) mice 6 days after tamoxifen injection were cultured for 24 hours with 10 mM glucose (A–C) or 25 mM glucose and 100 nM insulin. (B) [1-14C]Oleate (0.3 mM) was added for the last 2 hours of culture. (A) Left panel: oleate oxidation into [14CO2] and [14C] ASP and right panel: quantification of esterification rates into TAG, DAG, PL and cholesteryl esters after TLC of the lipid extracts. n=4 independent experiments with triplicates. (B) Oleate oxidation into CO2 and [14C] ASP in the presence of 25 mM glucose and 100 nM insulin. n=3 independent experiment with triplicates. (C) Left panel: ketone body production with or without 0.3 mM oleate, n=3 independent experiments with triplicates. Right panel: immunoblot analysis of cellular extracts using specific antibodies for Hmgcs2, Bdh1 and γ tubulin. ASP, acid soluble products; Bdh1, β-hydroxybutyrate dehydrogenase; DAG, diacylglycerol; PL, phospholipid; TAG, triacylglycerol; TLC, thin layer chromatography.

We assayed ketone body production. Acetoacetate and β-hydroxybutyrate production were significantly induced in ApchepKO hepatocytes in presence of exogenous oleate, consistent with the [14C]-labelled ASP profile (figure 5C). Accordingly, the expression of two ketogenic enzymes, HMG-CoA synthase 2 and β-hydroxybutyrate dehydrogenase, was elevated in ApchepKO livers (figure 5C). Ketone body production was still significantly higher in ApchepKO hepatocytes, even in the absence of exogenous oleate, indicating active FAO fuelled by endogenous fatty acids. This production was blunted by an adipose triglyceride lipase (Atgl) inhibitor, indicating that the increased expression of Atgl observed in ApchepKO also contributes to the increased ketone body production from endogenous lipids (figure 5C and SI4E).

We next assessed the overall adiposity of the ApchepKO mice, having observed the strong lipidic phenotype with increased FAO and decreased lipogenesis. In the days following tamoxifen injection, when ApchepKO and control mice recovered a similar food intake, mutant mice kept loosing adiposity (Fig. SI4F).

Overall, our results show that the metabolic reprogramming in ApchepKO hepatocytes is similar to those observed in Apc-induced HCC and thus could be a driving force for HCC development. It consists of oxidative metabolism, using fatty acids as fuel and an overall decrease in lipogenesis, with some rerouting towards phospholipids synthesis.

FAO fuels oxidative phosphorylation and promotes ApcKO hepatocyte proliferation

Mitochondria are the principal cellular energy source, through glucose or FAO, driving ATP production via the oxidative phosphorylation pathway. We therefore addressed the impact of β-catenin overexpression on mitochondrial substrate utilisation and oxygen consumption. Basal and maximal oxygen consumption rates were significantly higher in ApchepKO hepatocytes with both the glutamate/malate or succinate substrates as well as the fatty acid, palmitoyl-carnitine (figure 6A), whereas non ATP-generating respiration (leak) was unaffected. Therefore, β-catenin-activated hepatocytes have an enhanced cell respiratory capacity, which is very efficiently fed by FAO.

Figure 6

Fatty acid oxidation fuels oxidative phosphorylation and promotes Apcko hepatocyte proliferation. (A) Mitochondrial respiration was determined under basal conditions or in the presence of oligomycin (1 mg/mL; leak) or increasing amounts of CCCP (1–20 mM) to determine the maximal respiration rate. Left panel: in the presence of succinate, right panel: in the presence of glutamate/malate, lower panel: in the presence of palmitoyl-carnitine. n=3. Results are presented as the means±SEM of triplicate from three independent experiments. (B) ATP production rates, over a 3 min period by hepatocytes incubated with either 0.3 mM oleate or 25 mM glucose. n=2. Results are presented as the means±SEM of quadruplicates from two independent experiments. (C) Doubling time analysed by impedance measurement in real time over a 30-hour period, in presence of either 0.3 mM oleate or 0.3 mM oleate and 1 mM etomoxir. n=3 of quadruplicates. ADP, adenosine diphosphate; ATP, adenosine triphosphate; CCP, carbonyl cyanide m-chloro phenyl hydrazone.

We then investigated whether fatty acids are the preferential substrates used by ApchepKO livers. We assessed ATP production rates in hepatocytes cultured in the presence of either oleate or glucose. The rate of ATP production from oleate was higher in ApchepKO hepatocytes than control cells and was blunted by etomoxir. ATP production rates from glucose were lower in ApchepKO hepatocytes, supporting that part of the glucose did not fuel the Krebs cycle and was likely incorporated into glutamine (figure 6B).

We investigated the functional relevance of boosted FAO in ApchepKO hepatocytes. In presence of oleate, the rate of ApchepKO hepatocyte proliferation was twofold greater than that of control cells. This was inhibited by etomoxir, indicating that FAO provides a proliferative advantage to β-catenin-activated hepatocytes (figure 6C).

PPARα-induced FAO is mandatory for HCC development led by aberrant β-catenin signalling

Our data showed that Apc-HCC are oxidative tumours that use fatty acid as fuel. One of the key regulators of FAO in the liver is PPARα. Indeed, Pparα knockout (PPARαKO) mice have reduced hepatic β-oxidation that, over time, leads to the development of steatosis.26 27 In accordance with our previously published results showing PPARα to be a direct β-catenin target gene, PPARα is very likely a transcriptional regulator responsible for the FAO programme observed in CTNNB1-mutated HCC19 (figure 2A). Accordingly, PPARα mRNA and protein expression was higher in ApchepKO mouse livers than controls (figure 7A and online supplementary figure SI5A). Moreover, transfection experiments with a luciferase reporter gene under the control of the peroxisome proliferator responsive element (PPRE) of the peroxisomal bifunctional enzyme showed a greater PPRE activity in ApchepKO hepatocytes than controls (online supplementary figure SI5). Finally, reverse transcription quantitative PCR showed a higher level of expression of several PPARα target genes in ApchepKO livers than in control livers, and this expression was diminished in PPARαKO livers (online supplementary figure SI5A). Moreover, a gene set enrichment analysis confirmed the presence of a Pparα signature in ApchepKO (figure 7A). In human, the analysis of PPARα expression in CTNNB1-mutated HCC in the GSE62232 data set of human HCC22 23 showed that PPARα mRNAs were more abundant in CTNNB1-mutated HCC than in non-CTNNB1-mutated HCC (figure 7B). Finally, in an independent cohort of human HCC mutated (21 samples) or non-mutated for CTNNB1 (21 samples) and six non-tumour samples, PPARα mRNA expression was higher in CTNNB1-mutated HCC compared with non-mutated and non-tumour samples (figure 7B).

Supplementary file 5

Figure 7

FAO is mandatory to β-catenin tumourigenesis. (A) Left panel: immunoblot analysis of nuclear extracts from Apc lox/lox/TTR-CreTam (ApchepKO), Apclox/lox (control) and PparαKO mouse livers 6 days after tamoxifen injection using specific antibodies for PPARα and p84. Right panel: β-catenin-activated genes are enriched in a PPARα signature. Normalised enrichment ratio=1.49; p value 0.0 (<0.001). (B) Left panel: transcript abundance (probe intensity) of PPARα in human HCC was obtained from the GSE62232 data set.22 23 Right panel: mRNA relative expression of PPARα in 21 mutated CTNNB1 HCC (M), 21 non-mutated CTNNB1 HCC (NM) or in 6 non-tumour liver tissues (NT). (C) Apc lox/lox/TTR-CreTam (ApchepKO), Apc lox/lox (control), PPARαKO and Apc lox/lox/TTR-CreTam/PPARαKO (DKO) liver explants 6 days after tamoxifen injection were cultured for 2 hours with 0.3 mM [1-14C]oleate and oleate oxidation into CO2 quantified. (D) Longitudinal analysis of β-catenin tumourigenesis in PPARαKO livers. Left panel: tumourous load was quantified at the time of sacrifice. Right panel: tumourous penetrance was analysed as the percentage of mice presenting a tumour at each time point. (n=20 for ApcTumLIV and n=18 for ApcTumLIV/PPARαKO mice). (E) Tumour growth rate in ApcTumLIV mice fed a control diet (control) or a diet containing an FAO inhibitor, etomoxir (treated) at 0.2 g/kg of the diet for 3 weeks. FAO, fatty acid oxidation; HCC, hepatocellular carcinoma; PPARα, peroxisome proliferator-activated receptor alpha.

Therefore, we assessed whether increased FAO of ApchepKO livers is linked to PPARα activation. We crossed Apc lox/lox/TTR-CreTam mice with PPARα KO mice and analysed FAO in double knockout ApchepKO/PPARαKO (DKO) liver explants. DKO livers showed activation of β-catenin signalling, as expected (online supplementary figure SI5B). We first analysed the level of FAO in livers of PPARα KO, control, ApchepKO and DKO mice. FAO was lower in PPARα KO livers than controls. In contrast, FAO was higher in ApchepKO than controls, but was blunted in absence of PPARα in the DKO liver, to levels similar to those observed in the control mice (figure 7C). This suggests that boosted FAO in ApchepKO livers is mostly mediated directly or indirectly by PPARα. We then investigated the role of PPARα-induced FAO in β-catenin-driven tumourigenesis. We monitored Apc-HCC progression in the context of PPARα deficiency and found that DKO mice had markedly fewer HCC than ApcTumLIV mice with tumour penetrance of <20% and less than two tumours per mouse (figure 7D). Moreover, the HCC appeared 4 months later (10 months vs 6 months), showing that Pparα deletion decreased both HCC initiation and progression in ApcTumLIV mice (figure 7D).

We then examined the consequences of direct inhibition of FAO on β-catenin-activated HCC development. ApcTumLIV mice were fed a diet containing 0.2 g/kg etomoxir for 3 weeks from the moment the first tumours were detected. After 3 weeks of treatment, the tumour growth rate of the etomoxir-treated tumour-bearing mice dropped to less than 3%, whereas the rate remained at 12% in non-treated controls (figure 7E), showing that pharmacological inhibition of FAO was able to efficiently block β-catenin-induced HCC development.

Overall, our results demonstrate that FAO plays a critical role in vivo in HCC development in ApcTumLIV mice at all steps of tumour development, including initiation and progression. Treatment of ApcTumLIV mice with a FAO antagonist markedly inhibited HCC tumour growth, strongly suggesting that targeting FAO is a suitable therapeutic strategy for CTNNB1-mutated HCC.


We have unravelled the specific metabolic programming of HCC induced on β-catenin oncogenic activation (figure 8). This is a frequent event in human HCC as CTNNB1-mutated HCC represents 18%–40% of all HCC. Our main findings are: (1) β-catenin-activated HCCs are not glycolytic, do not follow the Warburg effect but use fatty acids as an energy source and exhibits reduced lipogenesis. which is rerouted towards phospholipid synthesis; (2) their principal metabolic programme is FAO, which is enriched in CTNNB1-mutated HCC and under the direct or indirect control of PPARα; and (3) FAO is required for the transformation of β-catenin-induced HCC and is a suitable therapeutic target, as treatment with an FAO inhibitor was able to block tumour development of β-catenin-induced HCC.

Figure 8

Model summarising β-catenin-driven metabolic reprogramming. Substrates are shown in green, fatty acid fate in red and glucose fate in blue. β-catenin oncogenic activation leads to increased FAO. Extrahepatic and intrahepatic fatty acids are rerouted to an oxidative pathway at the expense of esterification. Higher rates of FAO are also sustained by higher rates of ketogenesis. Moreover, lipogenesis from both glucose and glutamine decreases, leading to a decrease in the amount of malonyl-CoA, the allosteric inhibitor of Cpt1, allowing FAO even in the presence of glucose. FAO then efficiently fuels ATP production through oxidative phosphorylation. Bold line: boosted metabolic pathway, dotted line: decreased metabolic pathway. ATP, adenosine triphosphate; Cpt1, carnitine palmitoyltransferase 1; DAG, diacylglycerol; FA, fatty acid; FAO, fatty acid oxidation; PL, phospholipid; PPP, pentose phosphate pathway; TAG, triacylglycerol.

The Warburg effect has long been considered to be a staple of tumour metabolism. However, it has been shown that tumour cells find other ways to survive and proliferate other than the classic increased aerobic glycolysis.25 28 29 FAO is used by malignancies such as glioblastoma, triple negative breast cancers or leukaemia cells.30 31 In these tumours, however, FAO is used for different purposes, not always linked to ATP production by the electron transfer chain. Indeed, in acute leukaemia myeloid cells, FAO is uncoupled from ATP production but is required for tumour cell survival, as it is involved in the control of the mitochondrial permeability transition with energy production coming from the Warburg effect.31 In contrast, in the c-myc overexpressing triple-negative breast tumour model, FAO is, similarly to our model, the leading force behind energy production.30 Indeed, in β-catenin-activated HCC, FAO efficiently feeds the electron transfer chain, producing ATP with no uncoupling, rendering them independent from glucose for ATP production. Aside from providing an obvious energy source, boosting FAO confers other advantages to tumour cells, and recent research has highlighted the role of FAO in nicotinamide adenine dinucleotide phosphate (NADPH)homeostasis32 33 and epigenetics modifications.34 Indeed, it has been shown that lipid-derived acetyl-CoA is the main substrate for histone acetylation leading to epigenetic alterations.34 In addition, FAO has been suggested to be a pathway that allows NADPH homeostasis for tumour cells under conditions of energetic stress, such as during phases of nutrient deprivation.35 36 In addition to its key functions in antioxidant responses, NADPH is also the cofactor of anabolic enzymes, and maintaining NADPH is therefore key to tumour cell survival.

In the recent years, with the growing epidemic of obesity and diabetes, non-alcoholic fatty liver disease-related HCCs have been on the rise.37 Unfortunately, these two pathologies are often associated with an increased peripheral insulin resistance and higher rate of circulating free fatty acids, the main substrate of hepatic FAO. Therefore, this metabolic context could also contribute to the development of the CTNNB1-mutated HCC.

In β-catenin-activated hepatocytes, FAO was also boosted through several pathways that feed into it or enhance it. Acetyl-CoA, the product of FAO, exerts a feedback control loop on FAO activity38; thus, enhanced ketogenesis contributes to the increased rate of FAO by condensing acetyl-CoA into ketone bodies. Moreover, ketogenesis allows the recycling of CoenzymeA, which can then be reused by various metabolic pathways including FAO. Thus, decreased malonyl-CoA concentrations, increased rates of endogenous lipid hydrolysis and boosted ketogenesis contribute to higher FAO in β-catenin-activated HCC, providing the ATP, NADH and FADH2 necessary for cancer cell survival and proliferation. Finally, one of the regulators of this enhanced FAO is PPARα, itself induced by β-catenin. Interestingly, a single-cell spatial reconstruction of the liver lobule has recently shown that the PPARα pathway is more active in the perivenous region where the β-catenin pathway is activated.39

One of the main goals of such metabolic reprogramming is to promote unchecked proliferation, which requires energy to fuel their growth and macromolecule synthesis. This raises questions concerning the ability of β-catenin-activated hepatocytes to generate their biomass, at least in terms of membranes. The decreased lipogenesis observed in β-catenin-activated HCCs may appear counterproductive, but the newly synthesised fatty acids are rerouted into phospholipids, and there is also increased esterification of cholesterol by acyls. Phospholipids and cholesterols are two major components of cellular membranes. Moreover, our previous studies showed that β-catenin-activated HCC produce high levels of glutamine,19 20 which is most certainly produced from glucose as observed in other models of β-catenin-induced HCC.25 However, glutamine is well known to mainly feed into the biosynthetic pathway of purine and pyrimidine bases in several malignancies,25 40 corroborating our previous work, showing high expression of pyrimidine and purine pathway enzymes.19 Thus, although glucose anabolism is not primarily directed towards lipids synthesis, glucose carbons are still used for the ribose and glutamine synthesis required to meet the nucleotide requirement for cell growth.

From a therapeutic perspective, tumour plasticity is one of the main obstacles in treating tumours with metabolic inhibitors.41 However, β-catenin-activated tumours present a strong metabolic inflexibility as decreasing FAO was sufficient to abolish tumour initiation and slow tumour progression. Therefore, the use of FAO inhibitors to treat CTNNB1-mutated HCC appears to be a pertinent therapy. Clinical trials have been performed with etomoxir, but they were stopped.42 More recently, Teglicar, a reversible inhibitor specific to the hepatic isoform of Cpt1, has proven to be efficient in preclinical tests against leukaemias, and it is in a phase II clinical trial for treating diabetes.43 Such molecules could be considered for therapeutic approaches to drastically slow the progression of CTNNB1-mutated HCCs.

Experimental procedures


Mice carrying two floxed alleles on the 14th exon of the Apc gene16 (here referred to as Apc lox/lox) were previously interbred with TTR-CreTam mice44 generating Apc lox/lox/TTR-CreTam mice and their controls, Apc lox/lox. All mice were backcrossed onto the C57Bl6N background. The mouse modelling both pretumoural and tumour development of CTNNB1-activated human tumours have been already described.16–19 Male Apc lox/lox/TTR-CreTam mice that are 2–4 months old, intraperitoneally injected with 1.5 mg tamoxifen (MP Biomedicals), constitute the so-called ‘pretumoural’ model16 and are referred to here as ApchepKO mice. This results in a deletion of Apc in all hepatocytes and the early death of the animals (mostly prior to 20 days) from major metabolic defects.16 17 The controls consisted of male Apc lox/lox littermates that received one injection of 1.5 mg tamoxifen. To obtain liver tumours, Apc deletion should occur in less than 70% of hepatocytes to prevent animal death.16 This can be obtained, either by unique injection of a lower dose of tamoxifen (0.75 mg) or intravenous injection of AdCre. Partial Apc inactivation in the hepatocytes with AdCre injection was the most efficient method to obtain a reproducible model of liver tumour development. Thus, liver tumours were obtained by intravenous injection of 0.5×109 Pfu AdCre into 2-month-old male Apc lox/lox/TTR-CreTam, Apc lox/lox mice and/or Apc +/lox/TTR-CreTam, Apc +/lox mice as controls. These mice are referred to here as ApcTumLIV mice, and the HCCs that develop in these mice are called Apc-HCCs. Tumour development was followed twice monthly by two-dimensional ultrasound. Additional details regarding methods can be found in the online supplementary information.

Supplementary file 6


We are very thankful to the ‘Mitomaniac’ team led by F Bouillaud for enlightening all discussions on the wonders of mitochondria and A Lombes for her patience. We would like to thank Sandra Rebouissou for the kind gift of the SNU 398 cells. We wish to thank Nadia Boussetta for caring for the mice and managing our little colony. We are also very thankful to Julien Planchais for his very helpful discussions concerning PPARαKO mice.


View Abstract


  • CP and SC contributed equally.

  • Contributors PB: conceived and supervised the study, analysed the data, carried out the experiments, wrote the manuscript and funding. NS: designed and carried out experiments and analysed the data. MS, DCG and CS: performed experiments. AG: technical advises and reading of the manuscript. IL: acquisition of ultrasound data. HG: axin1 animal model and critical reading of the manuscript. BT: provided the human samples. CP: funding, heatmap generation and analysis, critical revision of the data and critical reading of the manuscript. SC: funding, GSEA analysis and critical reading of the manuscript.

  • Funding NS is a recipient of Ligue Nationale Contre le Cancer (LNCC) and ARC foundation scholarships. MS is a recipient of the French Ministry of Research scholarship. This study was supported by grants from the ANR (Agence Nationale de la Recherche), the AFEF (Association Française d’Hépatologie) and the LNCC.

  • Competing interests None declared.

  • Patient consent Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles