LXRα activation and Raf inhibition trigger lethal lipotoxicity in liver cancer

The success of molecular therapies targeting specific metabolic pathways in cancer is often limited by the plasticity and adaptability of metabolic networks. Here we show that pharmacologically induced lipotoxicity represents a promising therapeutic strategy for the treatment of hepatocellular carcinoma (HCC). LXRα-induced liponeogenesis and Raf-1 inhibition are synthetic lethal in HCC owing to a toxic accumulation of saturated fatty acids. Raf-1 was found to bind and activate SCD1, and conformation-changing DFG-out Raf inhibitors could disrupt this interaction, thereby blocking fatty acid desaturation and inducing lethal lipotoxicity. Studies in genetically engineered and nonalcoholic steatohepatitis-induced HCC mouse models and xenograft models of human HCC revealed that therapies comprising LXR agonists and Raf inhibitors were well tolerated and capable of overcoming therapy resistance in HCC. Conceptually, our study suggests pharmacologically induced lipotoxicity as a new mode for metabolic targeting of liver cancer. Zender, Dauch and colleagues demonstrate that pharmacologically induced lipotoxicity by activating LXRα and Raf-1 inhibition provides a metabolic therapeutic strategy for hepatocellular carcinoma.

T umors need to adapt metabolic signaling to meet their biosynthetic demands and to counteract an overload of toxic metabolites. Metabolic reprogramming may render tumor cells dependent on specific metabolic cascades that could be exploited therapeutically 1,2 ; however, the development of cancer therapies aimed at the inhibition of specific biosynthetic routes is often impaired because of high inter-and intratumoral metabolic heterogeneity as well as the enormous plasticity of metabolic networks [1][2][3] .
HCC accounts for 85-90% of all cases of primary liver cancer and represents a leading cause of cancer-related death worldwide (with more than 750,000 deaths per year) 4 . It is expected that the incidence of HCC will continue to rise owing to steadily increasing rates of fatty liver disease 5,6 . HCCs have primary resistance to cytotoxic therapies 7 , and the lack of common driver mutations has impeded the development of targeted therapies 4 . Therefore, systemic therapy for HCC currently relies on the use of multikinase inhibitors such as sorafenib, regorafenib, lenvatinib and cabozantinib that inhibit multiple pro-proliferative and anti-apoptotic signaling pathways simultaneously 8,9 . However, the use of these therapies in first-and second-line settings results in only moderately increased survival 8,9 . Phase 3 trial data regarding the use of immune checkpoint inhibitory antibodies in HCC are heterogeneous. While combination of the anti-PD-L1 antibody atezolizumab and the vasoendothelial growth factor receptor (VEGFR)-inhibiting antibody bevacizumab looked promising (NCT03434379), a recent phase 3 trial exploring the anti-PD-1 antibody pembrolizumab yielded negative results (NCT02702401) [9][10][11] .
Here we report pharmacologically induced lipotoxicity as a promising therapeutic strategy for the treatment of HCC. Induction of liver X receptor α (LXRα) signaling via small molecules resulted in increased synthesis of free fatty acids (FFAs). Concomitant inhibition of Raf led to lipotoxicity and cell death of therapy-resistant HCC cells in vitro and in vivo, as Raf-1 directly binds stearoyl-CoA desaturase-1 (SCD1) and enhances its function. Conformation-changing (DFG-out) Raf inhibitors prevented Raf-mediated SCD1 stabilization and thus reduced SCD1 activity, resulting in a toxic accumulation of saturated fatty acids.

NaTuRE CaNCER
in survival (Extended Data Fig. 1a-c), which was greater than the increase in survival recently reported in Nras G12V ; Cdkn2a ARF-/liver carcinomas 12 . To characterize the response of Myc OE ; Nras G12V HCCs to sorafenib, we analyzed their transcriptomes after 3 weeks of sorafenib therapy ( Fig. 1a and Supplementary Table 1). Ingenuity Pathway Analysis (IPA, canonical pathways) revealed differential regulation of several signal transduction pathways in sorafenib-treated as compared to carrier-treated Myc OE ; Nras G12V -driven HCCs ( Fig. 1b and Supplementary Table 2). The most pronounced finding was induction of LXR in sorafenib-treated HCCs (z score = 2.985, P < 0.0001; Fig. 1b and Supplementary Table 2).
LXR is a nuclear receptor and transcription factor that heterodimerizes with retinoid X receptor (RXR) 16 . In line with the proposed LXR-RXR activation, central LXR downstream factors [17][18][19][20] were also found to be transcriptionally upregulated in sorafenib-treated HCCs ( Fig. 1c and Supplementary Table 1).
To functionally test whether sorafenib activates LXR in mouse HCCs, we engineered Myc OE ; Nras G12V -driven HCC co-expressing an LXR response element (LXRE) linked to GFP (an RFP cDNA under the control of a constitutively active promoter served as a control; Fig. 1d). Strikingly, when tumors stably expressing the reporter construct were treated with sorafenib, pronounced induction of the GFP signal was observed, whereas RFP expression remained unchanged (Fig. 1d,e).
Molecular modeling analyses suggested direct binding of sorafenib to the ligand pocket of LXR proteins, resulting in stabilization of the active form of LXR (Fig. 1f,g and Extended Data Fig. 1d,e). To probe for direct interaction of sorafenib and LXRα, 3 H-labeled sorafenib was incubated with recombinant LXRα or, as a control, GFP. Upon pulldown of polyhistidine-tagged proteins, the bound fractions of [ 3 H]sorafenib were quantified (Fig. 1h). Our analyses revealed an increased 3 H signal upon LXRα pulldown and thus demonstrate direct binding of sorafenib to LXRα (Fig. 1i). This interaction could be outcompeted by a natural LXR ligand (24(S)-hydroxycholesterol, or 24(S)-HC 21 ), suggesting binding of sorafenib to the ligand pocket of LXRα (Fig. 1i). An LXRα binding saturation assay confirmed binding of sorafenib to LXRα with a moderate binding affinity (K d = 455 nM; Fig. 1j and Extended Data Fig. 1f).
To address whether sorafenib activates LXR signaling in human HCCs, we performed transcriptome analyses on tumor biopsies from six patients with HCC before and after sorafenib treatment (clinical study: e:Med-HCC-1, NCT02372162; Fig. 2a). Of these, two patients showed partial tumor remission, two showed stable disease and two had tumor progression under sorafenib treatment (Supplementary Table 3). IPA (canonical pathways) of mRNA expression data revealed strong induction of LXR-RXR signaling by sorafenib in the tumors of patients with stable disease (Fig. 2b and Supplementary Table 4), whereas no induction of LXR was seen in sorafenib-unresponsive tumors as well as in the remaining cells of tumors that had undergone partial remission upon sorafenib exposure (Fig. 2b). These data indicate that sorafenib activates LXR in therapy-responsive human HCC cells and that LXR activity correlates with the therapeutic efficacy of sorafenib.
Sorafenib is a multikinase inhibitor that exerts its therapeutic effects via inhibition of several targets 22,23 . To functionally test whether sorafenib-mediated activation of LXR contributes to the antitumorigenic properties of sorafenib, we antagonized LXR activity in sorafenib-treated Myc OE ; Nras G12V tumors by means of the LXR inverse agonist SR9238 (ref. 24 ; Fig. 2c,d). SR9238-mediated inhibition of LXR strongly reduced the therapeutic efficacy of sorafenib (Fig. 2e,f). SR9238 inhibits both LXR isoforms (LXRα and LXRβ) 24 . To test whether the effect of sorafenib is LXRα dependent, we performed CRISPR-mediated depletion of the LXRα gene Nr1h3 in Myc OE ; Nras G12V tumors, using an established protocol for direct gene editing in murine hepatocytes in vivo 25 . We co-delivered pT-CaMIN, pSB13 and a plasmid encoding Cas9, GFP and two independent guide RNAs (gRNAs) targeting exon 6 of Nr1h3 (pX-458-GFP-gNr1h3) into the livers of wild-type mice, resulting in full LXRα loss in >70% of all tumor nodules ( Fig. 2g and Extended Data Fig. 1g). Strikingly, Nr1h3-depleted HCCs showed a strongly diminished response to sorafenib treatment, thus verifying that sorafenib-mediated induction of LXRα signaling contributes to its therapeutic efficacy (Fig. 2h,i).

Raf counteracts the therapeutic efficacy of LXR activation in liver cancer.
To compare sorafenib's ability to induce LXR with that of designed LXR ligands, we carried out reporter assays in liver cancer cells (pL-LXRE-GFP reporter construct; Fig. 3a). The reporter construct was lentivirally delivered into mouse Mycand Nras G12V -driven HCC cells (Myc OE ; Nras G12V ; Cdkn2a ARF-/-) 13 . Upon selection, these cells were treated with DMSO, sorafenib, the well-established LXRα agonist T0901317 (refs. 18,26 ) or, as a control, the LXR-inactivating compound SR9238 for 3 d followed by GFP quantification in individual cells ( Fig. 3a and Extended Data Fig. 2a,b). While SR9238 reduced the activity of LXR in HCC cells, dose-dependent induction of LXR activity was observed upon administration of sorafenib. As expected, treatment with the LXR ligand T0901317 resulted in much stronger LXR activation than sorafenib treatment ( Fig. 3b and Extended Data Fig. 2c). To quantify intratumoral LXR induction by T0901317 or sorafenib in vivo, we generated Myc OE ; Nras G12V + LXRE-GFP HCCs and found increased GFP expression in T0901317-treated HCCs than in sorafenib-treated tumors ( Fig. 3c and Extended Data Fig. 2d). study analyzing the therapy response of Myc OE ; Nras G12V liver carcinomas to sorafenib in a transposon-based HCC mouse model. b, IPA of mRNA-seq data from Myc OE ; Nras G12V HCCs that were treated for 3 weeks with sorafenib or carrier. Shown are all significantly changed canonical pathways (P < 0.05, Fisher's exact test) with a ratio ≥0.2 (values represent z scores, n = 2 mice per group; exact P values are provided as source data and in Supplementary Table 2). c, Heat map comparing mRNA levels of established transcriptional target genes of LXR in Myc OE ; Nras G12V HCCs after 3 weeks of treatment with sorafenib or carrier (mRNA-seq data, n = 2 mice per group). d, Schematic outline of an in vivo reporter assay to analyze LXR activation by sorafenib in a transposon-based HCC mouse model. e, Representative photographs and quantification of tumor sections stained for native GFP and native RFP (+ DAPI) after intrahepatic delivery of pT-CaM-LXRE-G, pT-CaRIN and pSB13 and treatment with sorafenib or carrier for 3 weeks (values represent the mean ± s.d., n = 3 mice per group, statistical significance was calculated by two-tailed Student's t test). Scale bars, 100 µm. f, Molecular modeling to investigate the binding of sorafenib to the ligand pocket of LXRα. An overview of the LXRα structure is shown with co-activator peptide (NCOA1, dark blue) in the AF2 region and sorafenib docked in the ligand-binding pocket (orange). g, Suggested binding mode of sorafenib (proposed by molecular modeling) in the ligand-binding pocket of LXRα (PDB 3IPS). Residues contributing to the interaction are colored by property: hydrophobic side chains are purple and hydrogen bonds are cyan. h, Schematic outline of an LXRα-sorafenib binding assay applying pulldown of LXRα and GFP upon incubation with radioactive [ 3 H]sorafenib. i, Quantification of beta signal upon pulldown of LXRα and GFP in samples that were previously incubated with [ 3 H]sorafenib ± 24(S)-HC (values represent the mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). j, Quantification of beta signal upon pulldown of LXRα in samples that were previously incubated with different concentrations of [ 3 H]sorafenib (values represent the mean ± s.d., n = 3 independent experiments; specific binding, one-site specific binding curve). Numerical source data are provided.

NaTuRE CaNCER
We hypothesized that the superior LXR-agonistic properties of T0901317 might result in stronger antitumor effects than with sorafenib. However, when mouse liver cancer cells (Myc OE ; Nras G12V ; Cdkn2a ARF-/-) were treated with T0901317, we found only weak therapeutic activity (Fig. 3d). We therefore reasoned that LXRα induction only exerts antitumorigenic effects when a second target is inhibited simultaneously. Besides VEGFR, major targets of sorafenib include B-Raf and Raf-1 (refs. 22,23 ), two key kinases within the Ras-MEK-ERK pathway. To test whether Raf inhibition is required to unlock the antitumor activ-ity of LXRα ligands, we tested the LXR ligand T0901317 in combination with the specific B-Raf and Raf-1 inhibitor BI-882370 (ref. 27 ) or sorafenib administered at a dose sufficient to inhibit Raf without pronounced LXR activation. As a control, we tested T0901317 in combination with sunitinib, a multikinase inhibitor that inhibits many sorafenib targets (VEGFRs, PDGFRs, Kit, RET and CD135) but does not show efficient Raf inhibition 28 . While BI-882370 and sorafenib strongly increased the therapeutic effect of T0901317, no combinatorial effect was observed with sunitinib treatment (Fig. 3e)

NaTuRE CaNCER
To functionally test whether the synergistic antitumor efficacy of T0901317 and Raf inhibition is dependent on LXRα, we treated Myc OE ; Nras G12V ; Cdkn2a ARF-/-HCC cells harboring CRISPR-mediated gene deletions of either Nr1h3 or Nr1h2 (Extended Data Fig. 2e-g). While loss of Nr1h3 (LXRα) strongly diminished the therapeutic efficacy of T0901317 and sorafenib, knockout of Nr1h2 (LXRβ) did not impact the treatment response (Extended Data Fig. 2h). In line with these genetic experiments, we found that an additional synthetic LXR agonist (GW3965) and the natural LXR activator 24(S)-HC strongly synergized with sorafenib, while no pronounced therapeutic synergy was seen when the LXRβ-specific compound LXR-623 (ref. 29 ) was tested in combination with sorafenib in HCC cells (Extended Data Fig. 2i). Taken together, these data show that LXRα induction and Raf inhibition have synergistic antitumor activity in HCC (Fig. 3f).  . c, mRNA expression analysis of Abcg5, Srebf1 and Scd1 in Myc OE ; Nras G12V HCCs in vivo upon treatment with sorafenib ± the reverse LXR agonist SR9238 (3 d, values represent the mean ± s.d., n = 6 mice per group, statistical significance was calculated by two-tailed Student's t test; the analysis was repeated twice with similar results). d-f, Treatment of Myc OE ; Nras G12V HCCs with sorafenib or SR9238 alone, a combination thereof or the corresponding carrier. Representative pictures of livers following 5 weeks of treatment (e; n = 3 mice per group) and survival of mice under treatment (f; Kaplan-Meier curve, statistical significance was calculated by log-rank test, n = 6 mice per group). Scale bar, 1 cm. g-i, CRISPR-mediated editing of Nr1h3 in Myc OE ; Nras G12V HCCs and treatment of Myc OE ; Nras G12V + gNr1h3 or Myc OE ; Nras G12V + gNC HCCs with carrier or sorafenib. Representative pictures of livers following 5 weeks of treatment (h; n = 3 mice per group) and survival of mice under treatment (i; Kaplan-Meier curve, statistical significance was calculated by log-rank test, n = 6 mice per group). Scale bar, 1 cm. Numerical source data are provided.

NaTuRE CaNCER
Combined LXR activation and Raf blockade overcome therapy resistance in HCC in vivo. We next sought to probe our combination therapy in therapy-resistant liver carcinomas in vivo. Activation of the phosphatidylinositol-3-OH kinase (PI(3)K)-Akt signaling pathway has been reported to mediate sorafenib resistance in HCC 30 . We thus generated multifocal HCCs driven by Myc and a constitutively active form of Akt1 (Akt1 Myr ) by hydrodynamic delivery of a transposable element encoding both oncogenes (pT-CaMIA 15 + pSB13; Extended Data Fig. 3a).
In strong contrast to Myc OE ; Nras G12V -driven HCCs (compare Extended Data Fig. 1a-c), Myc OE ; Akt1 Myr -driven HCCs had primary resistance to sorafenib monotherapy (Extended Data Fig. 3a-c). Interestingly, the therapy resistance of Akt1 Myr -driven HCCs was not due to decreased LXRα expression, as Myc OE ; Akt1 Myr HCCs even showed higher Nr1h3 expression levels than Myc OE ; Nras G12V tumors (Extended Data Fig. 3d and Supplementary Table 5). However, despite the higher abundance of Nr1h3 transcripts, expression of LXRα target genes independent of Srebp1c (which is directly activated by the Akt1-mTORC pathway 31 ) was not found (Extended Data Fig. 3d and Supplementary Table 5).
To measure LXRα-dependent transcriptional activity in Myc OE ; Akt1 Myr HCCs, we delivered pL-LXRE-GFP into tumor cells. Despite robust LXRα protein expression, we detected markedly reduced reporter activity in Myc OE ; Akt1 Myr ; Cdkn2a ARF-/cells as compared to Myc OE ; Nras G12V ; Cdkn2a ARF-/cells (Extended Data Fig. 3e,f). It was recently shown that the Nrf2 transcription factor blocks LXRα function by activating farnesoid X receptor (FXR) 32 . In line with these data, we found increased Nrf2 (Nfe2l2) mRNA and protein expression in Akt1 Myr -driven HCCs (Extended Data Fig. 3g,h). Furthermore, short hairpin RNA (shRNA)-mediated knockdown of Nfe2l2 partly restored LXRα activity in Akt1 Myr -driven HCCs (Extended Data Fig. 3i,j). These data suggest that the weak LXR-agonistic properties of sorafenib might not be sufficient to overcome Nrf2-mediated LXRα inhibition in Akt1 Myr -driven HCC. Indeed, in contrast to sorafenib, the designed compound T0901317 that strongly activates LXRα was able to induce LXRα activity in Myc OE ; Akt1 Myr ; Cdkn2a ARF-/cells (Extended Data Fig. 3k).
To test whether the combination of a potent LXRα activator and a Raf inhibitor allows therapy resistance to be overcome in HCC, we combined T0901317 with sorafenib or the specific Raf inhibitor BI-882370 in Myc OE ; Akt1 Myr -driven HCCs (Fig. 4a). While T0901317 monotherapy only moderately decreased tumor development, treatment with the T0901317-BI-882370 or T0901317sorafenib combination strongly reduced intrahepatic tumor burden and significantly prolonged the survival of tumor-bearing mice (Fig. 4b,c and Extended Data Fig. 4a). Importantly, this therapeutic an LXR reporter assay in mouse or human HCC cells using a lentiviral vector encoding LXRE linked to GFP (p-LXRE-GFP). b, LXRE reporter assay in Myc OE ; Nras G12V ; Cdkn2a ARF-/-+ p-LXRE-GFP cells that were treated for 3 d with DMSO, sorafenib, T0901317 or SR9238 (GFP measurements in individual cells, values represent the mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). c, Reporter assay to compare LXR activation by sorafenib and T0901317 in vivo. Quantification was performed of tumor sections stained for native GFP and native RFP after intrahepatic delivery of pT-CaM-LXRE-G, pT-CaRIN and pSB13 and treatment with carrier, sorafenib or T0901317 (3 d, values represent the mean ± s.d., n = 4 mice per group, statistical significance was calculated by two-tailed Student's t test). d, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with sorafenib or T0901317 (quantification of viable cells by crystal violet staining, values represent mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). e, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with DMSO, BI-882370, sorafenib or sunitinib ± T0901317 (quantification of viable cells by crystal violet staining, values represent the mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). f, Schematic overview of treatment responses in HCC using either an LXR agonist or a Raf inhibitor alone or a combination of both compounds. Numerical source data are provided.

NaTuRE CaNCER
efficacy was lost in tumors with CRISPR-mediated depletion of the LXRα gene Nr1h3, confirming that the treatment response to the T0901317-sorafenib combination in vivo is LXRα dependent (Extended Data Fig. 4b-d).
To test the potency of our established combination therapy in liver carcinomas of a different genotype, we generated Yap1-driven, Trp53-deficient liver carcinomas. We co-delivered a transposon encoding Yap1 and Nras G12V (pT-CaYIN) and pPGK-cre into the hepatocytes of Trp53 fl/fl mice and commenced therapy with the T0901317-sorafenib combination, carrier or the corresponding monotherapies upon tumor formation (Fig. 4d). While these aggressive tumors showed only a weak response to sorafenib monotherapy and no response to T0901317 alone, the T0901317sorafenib combination strongly reduced tumor development and increased the survival of the animals (Fig. 4e,f). Taken together, these data suggest that strong LXRα activation with simultaneous Raf inhibition can overcome primary sorafenib resistance in HCCs of different genotypes.

NaTuRE CaNCER
Besides primary therapy resistance, acquired (secondary) therapy resistance often represents a major problem in clinical oncology 33 . To address whether the T0901317-sorafenib combination allows acquired therapy resistance to be overcome, we generated Myc OE ; Nras G12V HCCs (hydrodynamic delivery of pT-CaMIN + pSB13), which have primary sensitivity to sorafenib (Extended Data Fig. 1a-c). Tumor-bearing mice were treated with sorafenib monotherapy for 4 weeks (Fig. 4g). The 4-week treatment interval triggered acquired sorafenib resistance, as persistent treatment of these tumors with sorafenib monotherapy did not result in a survival benefit in comparison to mice where sorafenib was discontinued (Fig. 4h,i). Similarly, these tumors did not respond to T0901317 monotherapy. However, application of the T0901317sorafenib combination to mice that had previously received 4 weeks of sorafenib monotherapy resulted in strikingly reduced tumor development and significantly prolonged median survival time (Fig. 4h,i). These data indicate that T0901317-sorafenib combination therapy represents a powerful treatment regimen that allows primary and secondary therapy resistance to be overcome in liver cancer.

Combination of LXRα activation and Raf inhibition results in a toxic accumulation of saturated fatty acids in liver cancer cells.
Histopathological analyses of tumor-bearing livers (Myc OE ; Akt1 Myr , Yap1; Nras G12V ; Trp53 -/and Myc OE ; Nras G12V HCCs) after T0901317-sorafenib treatment revealed an accumulation of lipid droplets in liver tumors (Fig. 5a,b and Extended Data Fig. 5a,b). Because of the central role of LXRα in the transcription of hepatic genes involved in de novo fatty acid synthesis 18,34 , we reasoned that pharmacological LXRα activation might modify lipogenesis in HCC cells. Indeed, we identified elevated expression of full-length and cleaved Srebp1c as well as acetyl-CoA carboxylase 1 (ACACA), two key factors of fatty acid synthesis, in mouse HCC cells (Myc OE ; Nras G12V ; Cdkn2a ARF-/-) under T0901317-sorafenib therapy (Extended Data Fig. 5c).
We hypothesized that LXRα-mediated de novo fatty acid synthesis is crucial for the antitumor effect of the T0901317-sorafenib combinatorial therapy. To functionally test this, we combined T0901317-sorafenib treatment with the therapeutically active ACACA inhibitor ND-630 (ref. 35 ; Extended Data Fig. 5d) and, in fact, found that ACACA inhibition diminished the therapeutic efficacy of the T0901317-sorafenib combination (Fig. 5c).
Notably, LXRs also have a major role in cholesterol transport, and it was shown that LXRβ-mediated efflux of cholesterol can induce death of brain and prostate tumor cells with high cholesterol dependencies 29,36,37 . However, culture medium that contained high levels of cholesterol (complexed to methyl-β-cyclodextrin to enable cholesterol to permeate cells) 29 did not rescue Myc OE ; Nras G12V ; Cdkn2a ARF-/cells from T0901317-sorafenib-induced death (Extended Data Fig. 5e). This shows that LXR-mediated cholesterol efflux is not relevant for the therapeutic efficacy of the T0901317-sorafenib combination.
To understand how activation of fatty acid synthesis in combination with Raf inhibition kills HCC cells, we conducted lipidomic analyses in mouse liver cancer cells (Myc OE ; Nras G12V ; Cdkn2a ARF-/-). First, we measured the abundance of FFAs upon treatment with the T0901317-sorafenib combination or the corresponding monotherapies. We observed greatly increased levels of FFAs upon T0901317-sorafenib treatment, while only a moderate change in FFA abundance was seen with T0901317 monotherapy (Fig. 5d). We thus hypothesized that Raf regulates the processing of FFAs upon activation of fatty acid synthesis. We compared the distribution of all major long-chain fatty acids (≥16 carbons) upon LXRα activation with or without Raf inhibition (Fig. 5e). While T0901317 monotherapy resulted in an increased fraction of monounsaturated fatty acids in comparison to DMSO-treated cells, the T0901317sorafenib combination resulted in an increased fraction of saturated fatty acids in HCC cells (Fig. 5e). These data suggest that Raf is involved in the transformation of saturated fatty acids into monounsaturated fatty acids by an as-yet-unknown mechanism (Extended Data Fig. 5f) and that Raf suppression during activated fatty acid synthesis results in an intracellular accumulation of saturated FFAs in HCC cells.
Saturated FFAs are known to be toxic metabolites in different cell types, resulting in metabolic stress (such as oxidative stress and endoplasmic reticulum (ER) stress), cellular dysfunction and cell death [38][39][40] . To compare the cytotoxic effects of saturated and monounsaturated FFAs in HCC cells, we incubated mouse liver cancer cells (Myc OE ; Nras G12V ; Cdkn2a ARF-/-) in medium containing a high concentration of either a saturated fatty acid salt (sodium palmitate, 16:0) or a monounsaturated fatty acid salt (sodium oleate, 18:1). While sodium palmitate triggered massive death of HCC cells, no cell death was observed upon treatment with sodium oleate (Fig. 5f). To address whether the response to T0901317-sorafenib therapy is due to the altered ratio of saturated fatty acids to monounsaturated fatty acids, we replenished HCC cells with sodium oleate during treatment. Interestingly, this regimen strongly diminished the therapeutic efficacy of the T0901317-sorafenib combination (Fig. 5g).

Fig. 5 | LXRα activation and Raf inhibition result in a toxic accumulation of saturated fatty acids in liver cancer cells. a,b,
Representative pictures of mouse livers bearing Myc OE ; Akt1 Myr (a) or Myc OE ; Nras G12V (b) tumors that were treated for 7 weeks (Myc OE ; Akt1 Myr ) or 3 weeks (Myc OE ; Nras G12V ) with the T0901317-sorafenib combination, T0901317, sorafenib or carrier and stained with hematoxylin and eosin (H&E; n = 3 mice per group). Scale bars, 100 µm. c, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with the T0901317-sorafenib combination together with the ACACA inhibitor ND-630 (quantification of viable cells by crystal violet staining, values were normalized to the effect of ND-630 or DMSO alone, values represent the mean ± s.d., n = 6 independent experiments, statistical significance was calculated by two-tailed Student's t test). d, Quantification of FFAs in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after 3 d of treatment with DMSO, T0901317, sorafenib or a combination thereof (values represent the mean ± s.d., n = 3 cell cultures per condition, statistical significance was calculated by two-tailed Student's t test, the analysis was repeated twice with similar results). e, Evaluation of fatty acid distribution (≥16 carbons) in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after 3 d of treatment with DMSO, T0901317, sorafenib or a combination thereof (shown are pie charts of mean values, n = 3 cell cultures per condition, the analysis was repeated twice with similar results). f, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with sodium oleate or sodium palmitate (quantification of viable cells by crystal violet staining, values represent the mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). g, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with the T0901317sorafenib combination ± 200 µM sodium oleate (quantification of viable cells by crystal violet staining, values represent the mean, n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). h, Quantification of oxidative stress in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells, after 3 d of treatment with sorafenib, T0901317 or a combination thereof (quantification of green fluorescence, values represent the mean, n = 3 cell cultures per condition, statistical significance was calculated by two-tailed Student's t test, the analysis was repeated twice with similar results). i, Representative western blot analysis of eIF2α, phosphorylated eIF2α (S51), PERK, phosphorylated PERK (T982), GADD34, CHOP and cleaved caspase-3 in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after 3 d of treatment with T0901317, sorafenib or a combination thereof (cropped blot images, n = 2 independent experiments). α-Tubulin and vinculin were used as loading controls. j, Proposed model for the therapeutic response to a combination of LXRα activation and Raf-1 inhibition in liver cancer. Unprocessed images of the blots and numerical source data are provided.

NaTuRE CaNCER
We next tested whether the T0901317-sorafenib combination causes oxidative stress. In fact, we observed strongly increased levels of reactive oxygen species (ROS) in mouse HCC cells upon T0901317-sorafenib exposure (Fig. 5h), also leading to lipid peroxidation (Extended Data Fig. 6a). We also analyzed the activity of ER stress response factors and found activation of PERK and its downstream targets eIF2α, GADD34 and CHOP (GADD153) upon T0901317-sorafenib or T0901317-BI-882370 treatment, while ER stress pathways, which are mainly involved in the unfolded protein response (Atf6 pathway, IRE1α pathway), remained inactive ( Fig. 5i and Extended Data Fig. 6b,c). The PERK signaling pathway can either induce tissue repair or trigger apoptosis upon prolonged exposure to cellular stress 41 . Because PERK-induced apoptosis can be triggered by CHOP-mediated inhibition of Bcl-2 (ref. 42 ), we reasoned that elevated CHOP activity results in ER-stress-induced apoptosis. In fact, we observed increased levels of cleaved caspase-3 upon treatment with our combination therapy (Fig. 5i), while activation of markers related to necroptosis (MLKL, RIPK3) or caspase-independent apoptosis/oxeiptosis (AIF) 43 was not observed (Extended Data Fig. 6d). Flow cytometry analysis of in vivo samples identified increased infiltration and maturation of macrophages but not Ly6c -F4/80 + Kupffer cells in T0901317-sorafenib-treated tumors (Extended Data Fig. 6e-j), and inhibition of Bcl-2 (using ABT-199; ref. 44 ) increased cell death upon LXRα induction (Extended Data Fig. 6k).
Collectively, our data show that LXRα-mediated fatty acid production in combination with Raf inhibition results in an intracellular accumulation of saturated fatty acids. This accumulation leads to oxidative stress, to activation of a critical ER stress response and subsequently to apoptosis of liver cancer cells (Fig. 5j).

NaTuRE CaNCER
Raf impacts lipid metabolism through binding and stabilization of SCD1. To characterize how Raf influences intracellular fatty acid levels upon LXRα activation, we investigated whether Raf affects the activity of SCD1, the crucial enzyme in the conversion of saturated C16/C18 fatty acids into monounsaturated fatty acids 45 . LXRα is known to induce transcription of SCD1 (ref. 46 ), and, in line with this, we observed elevated Scd1 mRNA levels following treatment with T0901317 or T0901317 together with sorafenib (Fig. 6a). However, increased SCD1 protein expression was only seen with T0901317 monotherapy and not following T0901317-sorafenib treatment ( Fig. 6b and Extended Data Fig. 7a), indicating that sorafenib-mediated Raf inhibition negatively impacts SCD1 protein abundance. In a time course experiment, sorafenib strongly reduced SCD1 protein levels in T0901317-treated mouse liver cancer cells within 3 h, while no reduction in Scd1 mRNA levels was seen (Fig. 6c,d and Extended Data Fig. 7b). Because sorafenib was not observed to have an impact on SCD1 protein levels in the presence of a proteasome inhibitor (MG-132; Fig. 6e), we reasoned that Raf protects SCD1 from proteasomal degradation and thereby delays turnover of this short-lived protein (half-life of 2 to 4 h) 47,48 . These results were confirmed in a panel of different human HCC cell lines (Hep3B, HuH7, HLF and HLE; Fig. 6f-h and Extended Data Fig. 7c) and in mouse HCCs in vivo (Fig. 6i).
To test whether SCD1 protein abundance is also influenced by other Raf inhibitors, we analyzed the protein levels of SCD1 in cells treated with a panel of different commercially available Raf inhibitors over time. While BI-882370 (ref. 27 ), LY-3009120 (ref. 49 ), RAF-265 and RAF-709 (ref. 50 ) resulted in the same phenotype as sorafenib (Fig. 6j,k and Extended Data Fig. 7d,e), no degradation of SCD1 protein was seen in the presence of dabrafenib 51 or SB590885 (ref. 52 ) ( Fig. 6l and Extended Data Fig. 7f). Because dabrafenib and SB590885 block Raf kinase activity similarly to other Raf inhibitors (as demonstrated by reduced phosphorylation of MEK1 and MEK2 (MEK1/2); Extended Data Fig. 7g), we concluded that Raf stabilizes SCD1 independently of its kinase activity and activation of MEK1/2. In line with this, MEK1/2 inhibition (using AZD-6244; ref. 53 ) did not influence SCD1 protein levels (Fig. 6m), and we observed colocalization of SCD1 and Raf-1 in the cytoplasm of mouse liver tumors in vivo (Fig. 6n).
While all SCD1-modifying compounds represent classical DFG-out Raf inhibitors, both dabrafenib and SB590885 trigger a DFG-in conformational change in Raf 27,49,50,54 . The exclusive effect of DFG-out Raf inhibitors on SCD1 levels suggests that Raf in the DFG-in conformation influences SCD1 protein turnover via a physical protein-protein interaction. To confirm this hypothesis, we performed co-immunoprecipitation experiments using antibodies specific to Raf-1 and B-Raf in mouse liver cancer cells with transgenic expression of Scd1 (retroviral delivery of pWZL-Scd1 HA into Myc OE ; Nras G12V ; Cdkn2a ARF-/cells; Fig. 6o). These experiments revealed binding of SCD1 predominantly to Raf-1 (Fig. 6p). The Raf-1-SCD1 protein-protein interaction could also be confirmed in human HCC cells (Fig. 6q). To verify binding of the Raf-1 kinase domain to SCD1, we ectopically expressed either a truncated Raf-1 346-648 isoform (comprising only the Raf-1 kinase domain) or full-length Raf-1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/-; Scd1 HA cells (retroviral delivery of pBABE-Raf1 346-648-FLAG , pBABE-Raf1 WT-FLAG or pBABE-empty as a control). Co-immunoprecipitation experiments using an anti-FLAG antibody revealed binding of Raf-1 WT and truncated Raf-1 346-648 to SCD1, indicating that indeed the C-terminal kinase domain of Raf-1 interacts with SCD1 (Fig. 6r,s).
We next tested whether DFG-out Raf inhibitors are able to disrupt the Raf-1-SCD1 protein-protein interaction in Scd1 HA ; Raf1 WT-FLAG cells. To control for potential off-target effects, the assay was performed in parallel in Scd1 HA cells that expressed a kinase-defective Raf-1 mutant, which impedes binding of DFG-out Raf inhibitors (retroviral delivery of pBABE-Raf1 K375A-FLAG ). Both cell lines were treated with sorafenib or DMSO and the proteasome inhibitor MG-132 and subsequently subjected to co-immunoprecipitation experiments with a FLAG-specific antibody (Fig. 7a). We found sorafenib-mediated disruption of the Raf-1-SCD1 complex in Raf1 WT but not Raf1 K375A cells (Fig. 7b). In line with these results, ectopic expression of Raf1 K375A prevented sorafenib-mediated degradation of SCD1 in HCC cells (Fig. 7c and Extended Data Fig. 7h),

Fig. 6 | Raf-1 influences intracellular lipid composition in HCC cells by stabilizing SCD1 protein levels. a, mRNA expression analysis of Scd1 in
Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after 3 d of treatment with sorafenib, T0901317 or a combination thereof (values represent the mean ± s.d., n = 6 cell cultures per condition, statistical significance was calculated by two-tailed Student's t test, the analysis was repeated twice with similar results). NS, not significant. b, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after 3 d of treatment with sorafenib, T0901317 or a combination thereof (cropped blot images, n = 6 independent experiments). α-Tubulin was used as a loading control. c, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after treatment with T0901317 (2 d) and sorafenib (1-3 h) (cropped blot images, n = 6 independent experiments). α-Tubulin was used as a loading control. d, mRNA expression analysis of Scd1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after treatment with T0901317 (2 d) and sorafenib (1-3 h) (values represent the mean ± s.d., n = 3 cell cultures per condition, statistical significance was calculated by two-tailed Student's t test, the analysis was repeated twice with similar results). e, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after treatment with T0901317 (2 d) and sorafenib (1-3 h) in the presence of the proteasome inhibitor MG-132 (cropped blot images, n = 2 independent experiments). α-Tubulin was used as a loading control. f-h, Representative western blot analysis of SCD1 in Hep3B (f), HuH7 (g) and HLF (h) cells after treatment with T0901317 (2 d) and sorafenib (1-3 h) (cropped blot images, n = 3 independent experiments). α-Tubulin was used as a loading control. i, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V tumors after treatment with T0901317 (2 d) and sorafenib (3-8 h) (cropped blot images, n = 2 mice per condition, the western blot was repeated twice with similar results). α-Tubulin was used as a loading control. j-l, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after treatment with T0901317 (2 d) and the DFG-out RAF inhibitors BI-882370 (j) and Ly-3009120 (k) or the DFG-in RAF inhibitor dabrafenib (l) for 1-3 h (cropped blot images, n = 2 independent experiments). α-Tubulin was used as a loading control. m, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after treatment with T0901317 (2 d) and the MEK1/2 inhibitor AZD-6244 for 1-3 h (cropped blot images, n = 2 independent experiments). α-Tubulin was used as a loading control. n, Representative pictures of mouse Myc OE ; Akt1 Myr tumors after treatment with T0901317 (2 d) that were stained for SCD1 (green), Raf-1 (red) and DAPI (blue) (n = 3 mice per group, the staining was repeated twice with similar results). Scale bars, 20 µm. o, Schematic outline of immunoprecipitation (IP) experiments to analyze a potential protein-protein interaction of Raf and SCD1. p, Representative western blot analysis of lysates from Myc OE ; Nras G12V ; Cdkn2a ARF-/-; Scd1 HA cells that were immunoprecipitated with control IgG or antibodies specific to Raf-1 or B-Raf (cropped blot images, n = 3 independent experiments). q, Representative western blot analysis of lysates from Hep3B cells that were immunoprecipitated with control IgG or antibody specific to SCD1 (cropped blot images, n = 3 independent experiments). r,s, Schematic outline of immunoprecipitation experiments (r) and representative western blot analysis of lysates from Myc OE ; Nras G12V ; Cdkn2a ARF-/-; Scd1 HA ; Raf1 WT-FLAG or Raf1 346-648-FLAG cells that were immunoprecipitated with a FLAG-specific antibody (s; Myc OE ; Nras G12V ; Cdkn2a ARF-/-; Scd1 HA cells after infection with pBABE-empty served as a control, cropped blot images, n = 3 independent experiments). Unprocessed images of the blots and numerical source data are provided.
To investigate the relevance of the Raf-1-SCD1 interaction for the antitumor effects mediated by LXRα in HCC, we next combined T0901317 with either DFG-in Raf inhibitors (dabrafenib and SB590885) or the DFG-out Raf inhibitor sorafenib. In line with the hypothesis that only DFG-out Raf inhibitors are able to disrupt the interaction between Raf-1 and SCD1 and thus destabilize SCD1, we did not detect cytotoxicity with dabrafenib or SB590885 but could confirm a strong synergism between T0901317 and the DFG-out Raf-1 inhibitor sorafenib (Fig. 7g). To address the role of SCD1, we tested whether direct inhibition of SCD1 mimics the effect of DFG-out Raf inhibitors. We combined T0901317 with the SCD1-inhibitory compound A939572 (ref. 55 ) or with MEK1/2 inhibitors (AZD-6244 or PD0325901; refs. 53,56 ). While SCD1 blockade strongly improved the efficacy of T0901317, no therapeutic synergism was seen with MEK1/2 inhibition (Fig. 7h,i and Extended Data Fig. 7j).
Taken together, our data show that the SCD1 protein is stabilized through physical interaction with the Raf-1 protein. Accordingly, elevated Raf-1 expression in tumors results in accelerated transformation of saturated fatty acids into monounsaturated fatty acids, which protects HCC cells from LXRα-induced lipotoxicity (Fig. 7j). Inhibitors that induce a DFG-out conformation of Raf disrupt this interaction and induce proteasomal degradation of SCD1, which in turn leads to a toxic accumulation of saturated FFAs in HCC cells with activated lipogenesis (Fig. 7k-m). representative western blot analysis of lysates from Myc OE ; Nras G12V ; Cdkn2a ARF-/-; Scd1 HA ; Raf1 WT-FLAG or Raf1 K375A-FLAG cells that were treated with sorafenib or DMSO and immunoprecipitated with a FLAG-specific antibody (b; Myc OE ; Nras G12V ; Cdkn2a ARF-/-; Scd1 HA cells after infection with pBABE-empty served as a control, cropped blot images, n = 3 independent experiments). c, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/-; Raf1 K375A cells after treatment with T0901317 (2 d) and sorafenib for 1-3 h (cropped blot images, n = 2 independent experiments). α-Tubulin was used as a loading control. d-f, Representative western blot analysis of SCD1 in Myc OE ; Nras G12V ; Cdkn2a ARF-/-+ shNC (d) or Myc OE ; Nras G12V ; Cdkn2a ARF-/-+ shRaf1 (e,f) cells after treatment with T0901317 (pretreatment for 2 d) and doxycycline (Dox; 1-3 d) (cropped blot images, n = 3 independent experiments). α-Tubulin was used as a loading control. g, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with DMSO, sorafenib, dabrafenib or SB590885 ± T0901317 (quantification of viable cells by crystal violet staining, values represent the mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). h, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with carrier, T0901317, the SCD1 inhibitor A939572 or a combination thereof (quantification of viable cells by crystal violet staining, values represent the mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). i, Treatment of Myc OE ; Nras G12V ; Cdkn2a ARF-/cells with carrier, T0901317, the MEK1/2 inhibitor AZD-6244 or a combination thereof (quantification of viable cells by crystal violet staining, values represent the mean ± s.d., n = 3 independent experiments, statistical significance was calculated by two-tailed Student's t test). j,k, Schematic representations of the proposed Raf-1-SCD1 interaction and its impact on fatty acid distribution in HCC cells. l,m, Quantification of 16:0 (l) and 18:0 (m) fatty acids in Myc OE ; Nras G12V ; Cdkn2a ARF-/cells after 3 d of treatment with DMSO, sorafenib, BI-882370 or dabrafenib ± T0901317 (peak intensity was normalized to total protein levels in individual samples, values represent the mean ± s.d., n = 3 cell cultures per condition, statistical significance was calculated by two-tailed Student's t test). Unprocessed images of the blots and numerical source data are provided.

Pharmacologically induced lipotoxicity is effective in NASH-induced-and human HCCs and is well tolerated by mice.
To determine the translational potential of the above-outlined lipotoxic therapy, we sought to evaluate potential side effects of our T0901317-sorafenib combination therapy. We first commenced 3 weeks of therapy in healthy C57BL/6 mice (Extended Data Fig. 8a) and observed a slight increase in the liver/body weight ratio in comparison to carrier-treated mice (Extended Data Fig. 8b). A detailed histopathological evaluation revealed microvesicular steatosis of 20-40% of hepatocytes in treated mice; however, no changes in overall liver architecture were observed (Extended Data Fig. 8c). Importantly, we did not detect increased hepatocyte apoptosis or the development of liver fibrosis (Extended Data Fig. 8c). To evaluate the consequences of long-term therapy, we next treated healthy mice for 6 months with the T0901317-sorafenib combination or carrier. This long-term therapy resulted in a liver/body weight ratio that was comparable to that in mice treated for 3 weeks, with only slightly increased steatosis (Extended Data Fig. 8d-f). Furthermore, no increased cell death or fibrosis was found (Extended Data Fig. 8f). In line with this, mice did not lose body weight over time (Extended Data Fig. 8g,h).
To address whether a therapy that exploits metabolic stress can be applied to patients already suffering from a metabolic liver disease, we tested the T0901317-sorafenib therapy in a well-established mouse model of nonalcoholic steatohepatitis (NASH) and NASH-induced HCC 57 . Wild-type C57BL/6 mice were fed a choline-deficient high-fat diet (CD-HFD) for 12 months and subsequently treated with T0901317-sorafenib therapy or carrier (Extended Data Fig. 9a,b). While our lipotoxic therapy efficiently blocked NASH-induced tumor development (Extended Data Fig. 9c), only a slight increase in the liver/body weight ratio was observed in comparison to carrier-treated NASH mice (Extended Data Fig. 9d). Histopathological analysis revealed severe NASH in all mice (formation of lipid droplets, increased levels of cell death and liver fibrosis); no difference was seen between carrier-treated and T0901317-sorafenib-treated animals (Extended Data Fig. 9e). After a moderate weight reduction within the first 6 d of treatment (13%), the body weight development of T0901317-sorafenib-treated animals was comparable to that of carrier-treated mice (Extended Data Fig. 9f). Taken together, our data suggest that a therapeutic window for combined LXRα activation and Raf inhibition exists in liver cancer.
We also aimed to test the therapeutic efficacy of our combination therapy against human HCC in vivo. To do so, we generated orthotopically xenografted human HCCs by injecting Hep3B cells into the left liver lobe of immunodeficient CB17.Cg-Prkdc scid Lyst bg/Crl mice. After tumor formation, mice were subjected to treatment with the T0901317-sorafenib combination, carrier or the corresponding monotherapies (Fig. 8a). While sorafenib increased the median survival of tumor-bearing mice only marginally, the combination of T0901317 and sorafenib resulted in a pronounced antitumorigenic effect (Fig. 8b,c). T0901317-sorafenib therapy also induced death of Hep3B cells in vitro, whereas no synergistic effect was seen upon administration of T0901317 together with DFG-in Raf inhibitors such as dabrafenib or SB590885 (Fig. 8d).
As human HCCs are heterogeneous 4 , we next applied our combination therapy to a wider panel of different human HCC cell lines (HuH7, HLE, HLF, PLC/PRF/5). These cell lines differ substantially with respect to their metabolic and genetic makeup 58 and show different Akt-1, Nrf2 and LXRα protein expression levels (Extended Data Fig. 10a). Thus, we found varying LXRα activities and different responses to sorafenib monotherapy in these cells (Fig. 8e-h and Extended Data Fig. 10b). Importantly, the combination of LXRα activation and Raf inhibition triggered a pronounced antitumor effect and an ER stress response (activation of PERK and GADD34) in all tested cell lines (Fig. 8e-l).
Finally, we used HuH7 cells to generate orthotopically xenografted human HCC with marked primary resistance to sorafenib monotherapy (Fig. 8e,m and Extended Data Fig. 10a,b). These therapy-resistant HuH7 HCCs strongly responded to a T0901317sorafenib or T0901317-BI-882370 combination, and the treated mice showed significantly increased survival in comparison to animals that received carrier or monotherapy treatment (Fig. 8n,o and Extended Data Fig. 10c).

Discussion
Our data revealed that LXRα activation with concomitant Raf inhibition is lethal to HCC cells in vitro and in vivo, owing to an accumulation of toxic fatty acids. This combinatorial effect was simultaneously found and validated by another research group (Wangensteen, Kaestner and colleagues) 59 . We designated our therapeutic approach 'induced lipotoxicity' and showed that it is able to overcome the therapy resistance of genetically engineered mouse as well as xenografted human HCCs in vitro and in vivo and is well tolerated by mice. Because it is expected that HCC incidence will rise owing to increasing rates of fatty liver diseases in the future 5,6 , we also addressed whether this therapy can be applied to mice with a fatty liver disease such as NASH. Importantly, our therapy efficiently reduced NASH-driven HCC development and did not negatively impact the histological NASH phenotype in the liver. Notably, a certain weight reduction of obese NASH mice under T0901317sorafenib treatment was detected throughout the course of the experiment. However, carrier-treated mice also showed weight loss at this stage of NASH development. In summary, we believe that a therapeutic window for exploiting this approach for the treatment of human HCC exists.
While DFG-out Raf inhibitors such as sorafenib are already used for the treatment of patients with HCC, synthetic LXRα agonists are not yet available for clinical use 17 . However, as pharmacological LXR activation has been suggested for the treatment of different kinds of diseases (for example, Alzheimer's disease, atherosclerosis, hypercholesterolemia, retinopathy or brain cancer), LXR agonists are under development 17 . Some of them, such as LXR-623 (ref. 29 ) and BMS-779788 (ref. 60 ), have already been tested in phase 1 clinical trials (NCT00379860, NCT00366522, NCT00836602) and showed safety and tolerability in humans. However, to avoid potential toxic effects due to LXRα-induced liponeogenesis, such compounds were designed to preferentially induce LXRβ rather than LXRα (refs. 17,29,60 ). On the basis of the data presented herein, the development of potent and specific LXRα agonists and the use of such compounds as cancer therapeutics in combination with Raf inhibitors are warranted.

Methods
Vector design. The pL-LXRE-GFP (pGF-LXRE-GFP) plasmid was obtained from System Bioscience. The Sleeping Beauty transposase (pSB13), pPGK-cre, pT-CaMIN and pT-CaMIA transposon plasmids have been described previously 15,61 . To generate the pT-CaYIN transposon, a previously described Yap1 cDNA 62 was shuttled to pT-CaG-IRES-Nras G12V using NotI and AgeI restriction sites. The pT-CaM-LXRE-G transposon plasmid was generated by blunt-end cloning using NotI, AgeI and ClaI restriction sites. LXRE-G was obtained from the pGF-LXRE-GFP plasmid and cloned into the pT-CAG-Myc 12 transposon construct. The pT-CaRIN plasmid was cloned using NotI and AgeI restriction sites. The pWZL-blast plasmid was described previously 63 and obtained from M. Eilers (Theodor Boveri Institute, Biocenter, University of Wuerzburg, Wuerzburg, Germany). pWZL-Scd1 HA -blast was cloned via PCR using endogenous mouse Scd1 as a template. Full-length Raf1 WT-FLAG and Raf1 K375A-FLAG DNA fragments were ordered from IDT, and the Raf1 346-648-FLAG DNA fragment was generated via PCR. Raf1 FLAG constructs were cloned into pBABE-puro (Addgene) using BamHI and EcoRI restriction sites.
Nfe2l2 and Raf1 shRNAs were designed using splashRNA 64 and ordered as full-length oligonucleotides from Sigma-Aldrich. They were cloned into pRT3GEPIR 65 (obtained from J. Zuber, Research Institute of Molecular Pathology (IMP), Vienna Biocenter, Vienna, Austria) using XhoI or EcoRI restriction sites and into pT-CaR using XhoI and MluI/AscI restriction sites. A non-targeting Articles NaTuRE CaNCER control shRNA (shRen) was described previously 12 . gRNAs against mouse Nr1h3 or Nr1h2 were designed using CRISPRdirect 66 . A non-targeting control gRNA was described previously 67 . For each gRNA, two complementary oligonucleotides were ordered (Sigma-Aldrich), annealed and cloned into pSpCas9(BB)-2A-GFP (pX-458-GFP) using BbsI restriction sites 68 . The sequences of the PCR primers and shRNA and gRNA oligonucleotides are listed in Supplementary Table 6. Cloning was performed using SnapGene Viewer v2.8.2 software.
Animal strains and methods. All animal experiments were approved by committees of the regional authority of the state of Baden-Wuerttemberg (Regierungspraesidium Tuebingen, authorization numbers M15/18G, M3/18 and M1/15) as appropriate. All mice were housed and maintained under pathogen-free conditions (~22 °C ambient temperature, 50-55% humidity and 12-h dark/12-h light cycle) in accordance with the institutional guidelines of the University Hospital Tuebingen.
Hydrodynamic tail vein injection was performed in 5-to 6-week-old male or female wild-type C57BL/6N or B6.129P2-Trp53 tm1Brn /J (Trp53 fl/fl ) mice. Animals were injected with 25 µg of transposon plasmids, 5 µg of pSB13, 10 µg of pPGK-cre or 60 µg of pX-458-GFP diluted in 0.9% NaCl solution to a final volume of 10% of the mouse's body weight. DNA for hydrodynamic tail vein injection was generated using the Qiagen EndoFreeMaxi kit. Animals that were not efficiently injected within 10 s were excluded from the experiments. Generation of NASH in mouse livers was performed as described 57 , by feeding 5-week-old male wild-type C57BL/6N mice with CD-HFD for 12 months. Xenograft tumors of human Hep3B or HuH7 cells were established in 7-to 8-week-old male CB17.Cg-Prkdc scid Lyst bg/ Crl mice (Charles River). Cells were washed with PBS and injected into the left liver lobe (5 million cells per liver). In vivo treatment studies were performed in randomized groups with 100 mg per kg body weight of sorafenib (Bayer), BI-882370, SR9238 (both generated by S.L. and M.F. as previously described 24,27 ) and/or T0901317 (Cayman Chemicals) every second day. Drugs were administered via oral gavage dissolved in a cremophor:ethanol:water (12.5%:12.5%:75%) solution. The treatments were started 7 d after hydrodynamic tail vein injection, 10 d after cell transplantation or after 12 months of CD-HFD consumption. Treatment of healthy mice was started in 6-to 7-week-old male or female C57BL/6N mice. The group sizes were chosen on the basis of our experience with treatment studies in transposon-based and xenograft mouse models.
Cell culture. Myc OE ; Nras G12V ; Cdkn2a ARF-/cells and Myc OE ; Akt1 Myr ; Cdkn2a ARF-/cells were generated and described previously 12,13 . PLC/PRF/5, Hep3B, HEK293T and Phoenix packaging cells were obtained from the American Type Culture Collection; HuH7, HLE and HLF cells were obtained from the Japan Water-soluble cholesterol (Sigma-Aldrich) was dissolved in water and added to the cells at a final concentration of 0.75 µg ml −1 (as described previously 29 ). Cell viability was determined as described before 69 after 4 d (mouse cells) or 5 d (human cells) of treatment. Viable cells were stained with crystal violet and quantified with a plate reader (Tecan) by reading the absorbance at 590 nm (Tecan i-control v3.9.1 software). Production of lentiviral particles was carried out using HEK293T cells, and retroviral particles were generated using Phoenix packaging cells. The cell lines were transfected with lentiviral or retroviral DNA via calcium phosphate transfection, and the virus-containing supernatant was applied to target cells (supplemented with polybrene (5 µg ml −1 ) to enhance infection efficiency). Infected cells were selected with puromycin (3 µg ml −1 ) or blasticidin (3 µg ml −1 ). Expression of retrovirally delivered shRNAs and GFP was induced with 5 µg ml −1 doxycycline (Sigma). Transient transfection of cells was performed using Lipofectamine 3000 reagent (L3000008) from Thermo Fisher Scientific.
In vitro reporter assays and flow cytometry analysis. In vitro LXRE reporter assays were performed via GFP measurement in individual cells using the Guava Easy Cite Plus (see Extended Data Fig. 2a,b for plots exemplifying the gating strategy). Oxidative stress was determined using CellROX Green Reagent for oxidative stress detection from Thermo Fisher Scientific (C10444), and lipid peroxidation was measured with the Image-iT Lipid Peroxidation kit for live-cell analysis (C10445) from Thermo Fisher Scientific. Green and red fluorescence intensity was measured with the Guava Easy Cite Plus (CytoSoft v5.2 software).
Histopathology and immunohistochemistry. Histopathological evaluation of mouse livers and liver carcinomas was performed on H&E-stained paraffin sections by an experienced pathologist (M.T.R.). Visualization of fibrotic scars by sirius red staining was performed on paraffin-embedded liver sections as described previously 59 . A positive control was generated by induction of fibrotic scars via biweekly intraperitoneal injections of 0.25 ml per kg body weight of CCl 4 (Sigma) for 6 weeks. Apoptosis was analyzed on paraffin-embedded liver sections using terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) from Roche. A positive control was generated by induction of apoptosis via intraperitoneal injection of 0.4 µg per g body weight of Jo2 antibody. Immunohistochemistry for Ki67 (Abcam, ab15580; 1:450) or cleaved caspase-3 (Cell Signaling, 9661; 1:200) was performed on paraffin-embedded liver sections with standard protocols and hematoxylin counterstaining. Co-immunohistochemistry for SCD1 (Abcam, ab19862 (CD.E10); 1:250) and Raf-1 (Abcam, ab137435; 1:250) was performed on paraffin-embedded liver sections with standard protocols and DAPI staining. Native GFP and native RFP imaging was performed on sections (10 μm) of snap-frozen tissue that were previously fixed for 4 h with 4% paraformaldehyde. Oil red O staining was performed on sections (10 μm) of snap-frozen tissue as described previously 70 using Oil red O solution from Sigma-Aldrich. Microscopic analyses were performed using the Olympus BX63 microscope (CellSens Dimension v1.17 software). GFP and RFP intensity was quantified using ImageJ software (v1.47/v1.51n).
Lipid analysis. FFAs were quantified using the Free Fatty Acid Quantification kit from Abcam (ab65341). Colorimetric quantification was performed with a plate reader (Tecan), and values were normalized to the number of cells (counted by Neubauer chamber). Total fatty acid profiling was performed with liquid chromatography followed by mass spectrometry (LC-MS) analysis using the Q Exactive instrument from Thermo Fisher (Xcalibur v4.0 software) and the DIONEX Ultimate 3000 UPLC system (Thermo Fisher Scientific). Separation was achieved by C8 column (Thermo Fisher Scientific) over an acetonitrile:water:formic acid gradient. Spectra were analyzed using Tracefinder software (v3.3, Thermo Fisher Scientific). Lipid extraction and purification were performed with established protocols 71 . mRNA expression analyses. RNA was isolated from liver tumor samples via TRIzol (Invitrogen) and the RNeasy kit (Qiagen). For quantitative PCR (qPCR), mRNA was reverse transcribed into cDNA via the TaqMan kit (Applied Biosystems) using random hexamer primers. qPCR was performed using SYBR Green Master Mix (Applied Biosystems or Takara) and the Applied Biosystems 7500 or Bio-Rad CFX Real-Time PCR devise (using AB 7500 v2.0.6 or Bio-Rad CFX Maestro v1.1 software). Values were normalized to β-actin (Actb). The sequences for the primers used are listed in Supplementary Table 6.
Gene expression analyses (mRNA-seq) of mouse and human samples were performed in the c.ATG facility of Tuebingen University using a HiSeq 2500 or NextSeq 500 75-bp high-output run. Heat maps were generated with heatmapper 72 from counts per million values. Transcriptome data were analyzed with IPA (389077M/27821452) from Qiagen using standard parameters. Articles NaTuRE CaNCER solely on defined inclusion and exclusion criteria, which can be found at https:// clinicaltrials.gov/ct2/show/NCT02372162. Patients were treated with 800 mg sorafenib per day (because of adverse effects of the drug, P1 and P2 temporarily received 600 mg per day). Response monitoring was performed radiologically, and the best response according to mRECIST criteria was determined for each patient.
Transcriptome analysis of HCC tissues was performed using tumor samples of patients from whom we were able to obtain tumor biopsies before and after the primary treatment interval (n = 6 of 21 sorafenib-treated patients). The clinical characteristics and time points of the second biopsy are shown in Supplementary  Table 3.
Protein expression analyses. Protein was extracted from tissues and cells with RIPA extraction buffer or an established protocol to obtain nuclear or cytoplasmic extracts 73 . Protein (30-50 µg LXR-sorafenib binding assay. 500 ng of recombinant human LXRα (ab81921) or GFP (ab119740) from Abcam was incubated with 10 µM of [ 3 H]sorafenib (Hartmann Analytic) ± 100 µM of 24(S)-HC (Sigma-Aldrich) or an equal volume of 70% ethanol in reaction buffer (1% BSA in PBS, total volume 250 µl) for 1 h. Subsequently, the reaction was incubated with 100 µl equilibrated PureCube His affinity agarose (Cube Biotech) for 1 h in a rotating shaker. The bound fraction was washed twice with 1.5 ml wash buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 20 mM imidazole, pH 8.0) and eluted in elution buffer (1% SDS, 10 mM DTT in PBS) at 70 °C for 30 min. 10 µl of the eluted fraction was mixed with 4 ml UltimaGold (PerkinElmer) and measured with a beta counter (Packard, using QuantaSmart v2.11 software). The saturation binding assay (Fig. 1j) was performed in the same way with different concentrations of [ 3 H]sorafenib ± 320 µM unlabeled sorafenib (Bayer).

Molecular modeling.
Three structures of LXRα (5HJS, 3IPS and 3LOE) and two structures of LXRβ (1PQ6 and 1PQ9) were selected on the basis of the chemical similarity of their co-crystallized ligands with sorafenib and were retrieved from the PDB server. The selected proteins were prepared by adding hydrogen atoms, adjusting the protonation states of amino acids and fixing missing side-chain atoms and protein loops (PrepWiz v2018, Maestro v2018.2). All co-crystallized ligands and sorafenib were redrawn with the 2D-sketcher (Maestro v2018.2), with stereochemistry determined from the literature, and were minimized with force-field OPLS3e followed by correction of their protonation state using LigPrep.
Molecular docking was performed within a determined 10-Å grid around the co-crystallized compound in the ligand-binding pocket or, in the case of the AF-2 region, by residues interacting with the co-crystallized peptide of nuclear receptor co-activator 1 (NCOA1). For the AF-2 region, only X-ray crystal structures with co-crystallized co-activator peptide were considered (3IPS and 3LOE). For each compound, 80 docking runs were carried out, using the default settings of the Glide program (Glide v7.7, Maestro v2018.2) in extra-precision mode 74 . Amino acid residues were considered rigid during the calculation, but hydroxyl groups were allowed to rotate. Redocking of co-crystallized ligands in the ligand-binding pocket resulted in top-ranked docking poses with heavy atom root-mean-square deviation below 1 Å in comparison to the original conformation, for all PDB structures used. Docking poses were clustered by root-mean-square deviation, and a representative pose from the most populated cluster and the top-ranked pose by docking score were visually inspected. Selected docking poses for sorafenib within the ligand-binding pocket of LXRα and LXRβ, as well as the AF-2 region of LXRα, underwent molecular dynamics simulation to evaluate ligand stability within the site and to analyze interactions. Figures

October 2018
-Histone H3; Company: Abcam ab1791; LOT: GR3237728-1. Application: Western blotting in mouse cells. The antibody was validated by the manufacturer for the mentioned application in the mentioned species. Data are available on the manufacturer's website: https://www.abcam.com/histone-h3-antibody-nuclear-marker-and-chip-grade-ab1791.html -Ki67; Company: Abcam ab15580; LOT: GR180659-1. Application: Immunohistochemistry in mouse tissue sections. The antibody was validated by the manufacturer for the mentioned application in the mentioned species. Data are available on the manufacturer's website: https://www.abcam.com/ki67-antibody-ab15580.html -p-IRE1α (S724); Company: Novus Biologicals NB100-2323; LOT: AW. Application: Western blotting in mouse cells. The antibody was validated by the manufacturer for the mentioned application in the mentioned species. Data are available on the manufacturer's website: https://www.novusbio.com/products/ire1-alpha-antibody_nb100-2323 -LXRα; Company: Novus Biologicals NB300-612; LOT: UD286022. Application: Western blotting in mouse and human cells. The antibody was validated by the manufacturer for the mentioned application in the mentioned species. Data are available on the manufacturer's website: https://www.novusbio.com/products/lxr-alpha-nr1h3-antibody_nb300-612 -LXRβ; Company: Novus Biologicals NB100-74457; LOT: UJ289833. Application: Western blotting in mouse cells. The antibody was validated by the manufacturer for the mentioned application in the mentioned species. Data are available on the manufacturer's website: https://www.novusbio.com/products/lxr-beta-nr1h2-antibody_nb100-74457 -Nrf2; Company: Novus Biologicals NBP1-32822; LOT: 43054. Application: Western blotting in mouse and human cells. Flow Cytometry Plots Confirm that: The axis labels state the marker and fluorochrome used (e.g. CD4-FITC).
The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysis of identical markers).
All plots are contour plots with outliers or pseudocolor plots.
A numerical value for number of cells or percentage (with statistics) is provided.
Tumors: Liver tumours were chopped into small ~1 mm3 pieces and then enzymatically digested in a medium composed of equal volume of DMEM and HBS supplemented with 0.5 mg ml-1 Collagenase (Serva Collagenase NB 4G) for 30 min at 37°C. The enzymatic reaction was stopped using cold medium and the liver suspension was meshed through a 70 μm nylon mesh (Falcon). After centrifugation erythrocytes were lysed using an ACK buffer (150 mM NH4Cl, 10 mM KHCO3, and 0. Tumors: Samples were gated on viable leukocytes by DAPI exclusion and doublets were excluded using height versus area dot plots. Positive populations were defined using not stained cells as reference. Isotype controls were used to confirm the specificity of the staining. A detailed gating strategy is provided in Extended Data Fig. 6f-h.
Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information.