CD532

A MYC–aurora kinase A protein complex represents an actionable drug target in p53-altered liver cancer

MYC oncoproteins are involved in the genesis and maintenance of the majority of human tumors but are considered undruggable. By using a direct in vivo shRNA screen, we show that liver cancer cells that have mutations in the gene encoding the tumor suppressor protein p53 (Trp53 in mice and TP53 in humans) and that are driven by the oncoprotein NRAS become addicted to MYC stabilization via a mechanism mediated by aurora kinase A (AURKA). This MYC stabilization enables the tumor cells to overcome a latent G2/M cell cycle arrest that is mediated by AURKA and the tumor suppressor protein p19ARF. MYC directly binds to AURKA, and inhibition of this protein–protein interaction by conformation-changing AURKA inhibitors results in subsequent MYC degradation and cell death. These conformation-changing AURKA inhibitors, with one of them currently being tested in early clinical trials, suppressed tumor growth and prolonged survival in mice bearing Trp53-deficient, NRAS-driven MYC-expressing hepatocellular carcinomas (HCCs). TP53-mutated human HCCs revealed increased AURKA expression and a positive correlation between AURKA and MYC expression. In xenograft models, mice bearing TP53-mutated or TP53-deleted human HCCs were hypersensitive to treatment with conformation-changing AURKA inhibitors, thus suggesting a therapeutic strategy for this subgroup of human HCCs.

The closely related MYC (also known as c-MYC), MYCN and MYCL transcription factors have been implicated in tumorigenesis. Of these, MYC represents a critical node downstream of signaling networks that are frequently activated in human solid tumors1,2; its involvement in both tumor initiation and maintenance makes MYC an ideal target for cancer treatment3–6. However, MYC does not harbor any cavities into which small molecules can easily bind, thereby complicating the design of direct inhibitors. One strategy to indirectly target MYC is to target the bromodomain and extraterminal (BET) subfamily of bromodomain- containing proteins, which regulate MYC transcription. Bromodomain- containing 4 (BRD4), the most well-studied BET protein, has been implicated in a number of hematological and solid tumors7,8. BRD4 inhibitors seem to be particularly efficient in the treatment of leuke- mia and lymphoma, and clinical trials have been initiated using them (https://clinicaltrials.gov: NCT01713582, NCT01949883).Another strategy to indirectly inhibit MYC is suggested by the recent finding that proteolytic turnover of MYCN is regulated by a kinase-independent function of AURKA in neuroblastoma9–11. Inhibitors that disrupt the native conformation of AURKA and sub- sequently induce degradation of the MYCN protein were suggested as a therapeutic option for neuroblastoma and other MYCN-driven cancers, such as medulloblastoma and neuroendocrine prostate can- cer10–13. However, the most frequently occurring solid tumors are driven by MYC, not MYCN, and to date it remains unknown whether MYC–AURKA protein complexes exist in solid tumors.HCC represents one of the most frequently occurring solid tumors (resulting in 700,000 deaths per year worldwide), and no effective treatment options exist for patients with advanced liver cancer14,15. Despite the availability of an effective vaccine for hepatitis B virus (HBV) and the reachable goal of eradicating hepatitis C virus (HCV) in industrialized countries through new antiviral therapies16, the incidence of liver disease and liver cancer will continue to rise, in particular because of a western lifestyle and the associated increases in the incidence of non-alcoholic fatty liver disease (NAFLD) and non-alcoholic steatohepatitis17,18.
HCCs are dependent on MYC and frequently harbor genetic altera- tions in TP53 that are associated with a dismal prognosis, as compared to HCCs with wild-type (WT) TP53 (refs. 4,19,20). p53 restricts cell proliferation in response to DNA damage or the aberrant activation of oncogenes21,22. After oncogenic stress, p53 is activated by the tumor suppressor protein p14ARF (p19ARF in mice, which is encoded by the Cdkn2a locus)22, which has been found to be inactivated by promoter methylation in some cases of HCC23. p14ARF antagonizes the ubiquitin ligase MDM2, which normally negatively regulates p53, and thus suppresses tumorigenesis by increasing p53 levels. Although p14ARF physically associates with proteins other than MDM2, the p53-independent tumor suppressive roles of p14ARF remain poorly understood24.
Here we used a direct in vivo shRNA screen and show that Nras- driven TP53-altered (deleted or mutated) liver cancer cells become addicted to MYC stabilization and that this is mediated by AURKA. MYC directly binds to AURKA, and inhibition of this protein– protein interaction by conformation-changing AURKA inhibitors results in subsequent MYC degradation and death of TP53-altered HCC cells, thus suggesting a novel therapeutic strategy for this geneti- cally defined subgroup of human HCCs.

RESULTS
p19ARF prevents transformation of Trp53-deficient hepatocytes To study the role of the p19ARF–p53 tumor suppressor network in HCC, we took advantage of a well-established transposon-based mouse model of liver cancer25–29, in which intrahepatic deliv- ery of oncogenic NrasG12V generates HCCs that show increased RAS–mitogen-activated protein kinase (MAPK) signaling, a fea- ture of more than 50% of all human HCCs30–32. NrasG12V was stably delivered into hepatocytes of Cdkn2aARF−/−, Trp53−/− or Cdkn2aARF−/−;Trp53−/− double-knockout (dko) livers via hydro- dynamic tail-vein injection (HDI) of the transposon vectors (pCaN) into mice (Supplementary Fig. 1a). Immunostaining for NRAS revealed equal numbers of NRAS-expressing hepatocytes in livers from mice of the different genotypes (Supplementary Fig. 1b). We observed rapid outgrowth of multifocal and lethal HCCs in Cdkn2aARF−/− and Cdkn2aARF−/−;Trp53−/− dko mice but not in Trp53−/− mice Fig. 1a,b). Quantification of NRAS+ cells in liver sections at various time points after intrahepatic delivery of NrasG12V revealed steady numbers of NRAS-expressing Trp53−/− hepatocytes over time, whereas there were rapid increases in the number of NRAS-expressing Cdkn2aARF−/− hepatocytes within 1–3 weeks (Fig. 1c,d).

Immunostaining for phosphorylated MAPK3 and MAPK1 (also known as ERK1 and ERK2, respectively; phos- phorylated at Thr202 and Tyr204) revealed efficient activation of RAS downstream signaling in hepatocytes from Cdkn2aARF−/− and Trp53−/− mice (Supplementary Fig. 1c,d). Moreover, negative TUNEL staining in Trp53−/− livers excluded the possibility of apop- totic cell death (Supplementary Fig. 1c,e), and only a few p21Cip1+ (a marker for oncogene-induced senescence) and senescence- associated -galactosidase (SA–-Gal)+ cells were found in Trp53- deficient mouse livers after NrasG12V delivery, suggesting that cellular senescence does not account for the observed proliferation arrest of NrasG12V;Trp53−/− hepatocytes (Fig. 1c,e). Figure 1 Trp53-deficient hepatocytes undergo a G2/M cell cycle harrest after activation of RAS–MAPK signaling. (a) Representativephotographs of intrahepatic tumor burden (top) and H&E staining of tumor tissue (bottom) 5 weeks after transposons encoding oncogenic NrasG12V were delivered into livers of Cdkn2aARF−/− (left), Trp53−/− (middle) or Cdkn2aARF−/−;Trp53−/− (right) miceby hydrodynamic injection (HDI) (n = 4 mice per group). Scale bars, 1 cm (top) and 50 m (bottom). (b) Survival curves for Cdkn2aARF−/− (n = 9), Trp53−/− (n = 12) and Cdkn2aARF−/−;Trp53−/− (n = 5) mice after injection of transposons encoding oncogenicNrasG12V. (c) Representative photographs of liver sections stained for Cdkn2aARF–/– iNRAS, p21Cip1 or SA–-Gal 6 d and 21 d after intrahepatic delivery of oncogenic NrasG12V into Cdkn2aARF−/− (top) and Trp53−/− (bottom) mice (n = 3 mice per group). Scale bars, 50 m. (d,e) Quantification of NRAS+ (d) or p21Cip1+ (e) cells in liver sections from Cdkn2aARF−/− mice (NRAS, n = 15 mice;p21Cip1, n = 17 mice) and Trp53−/− mice (NRAS, n = 14 mice; p21Cip1, n = 16 mice) at the indicated time points after injection of a transposon expressing NrasG12V. (f) Representative cell cycle analysis of gated RFP+ (NrasG12V+) hepatocytes after liver perfusion (NrasG12V was co-delivered with RFP) (n = 3 mice per group). (g) Representative images of liver sections from Cdkn2aARF−/− (left) or Trp53−/− (right) mice stained for NRAS (green) and either cyclin B1 (red) (top) or p-histone H3 (Ser10) (red) (bottom) 12 d after HDI of a transposon expressing NrasG12V (n = 3 mice per group). White arrow indicates a NRAS+cyclin B1+ cell. Scale bar, 50 m. (h,i) Quantification of NRAS+ cells that were also cyclin B1+ (h) or p-histone H3 (Ser10)+ (i) in liver sections from Cdkn2aARF−/− (cyclin B1, n = 17 mice; p-histone H3 (Ser10), n = 16 mice) or Trp53−/− (cyclin B1, n = 14 mice; p-histone H3 (Ser10), n = 16 mice) mice at different time points after HDI of NrasG12V. In d,e,h,i, data are mean  s.d. Statistical significance was calculated by log-rank test (b) or Student’s t-test (d,e,h,i).Figure 2 In vivo RNAi screening identifies an AURKA-dependent G2/M arrest in Trp53−/− hepatocytes. (a) Schematic of transposable elements used for coexpression of NrasG12V, marker genes and miR30-based shRNAs.

IR, inverted repeats; DR, direct repeats; IRES, internal ribosome entry site;Caggs, synthetic CAG promoter. (b) Representative bioluminescence imaging photographs of WT mice 3 d after delivery of pCaNIL-shControl (left) or pCaNIL-shLuc (middle), or without transposon delivery (negative control) (right) (n = 3 mice per group). Colored bar represents counts of bioluminescent signals. Scale bar, 1 cm. (c) Outline of the RNAi screen. (d) Quantification of tumors in Trp53−/− livers after delivery of NrasG12V and the shRNA library pools (n = 4 mice per group). Horizontal lines represent means. Statistical significance was calculated by Student’s t-test. (e) Sequencing pie chartsof shAurka+ NrasG12V;Trp53−/− tumors (indicated by numbers) and shRNA library pool 2 (shPool2). (f) Representative images (of n = 4 mice per group) of tumor burden (top) and GFP expression (bottom) 9 weeks after delivery of pCaNIG-shAurka or pCaNIG-shNC into Trp53−/− mice. Scale bars, 1 cm.(g) Survival analysis of Trp53−/− mice after delivery of pCaNIG-shAurka or pCaNIG-shNC (shNC, n = 12 mice; shAurka-568, n = 4 mice; shAurka- 701, n = 4 mice; shAurka-1,054, n = 8 mice). Statistical significance was calculated by a log-rank test. (h) Representative bioluminescence imaging photographs of mice of the indicated genotypes 3 d after delivery of Aurka promoter–Luc and either NrasG12V or NrasG12V,D38A (n = 3 mice per group). Scale bar, 1 cm. (i) Survival analysis of Cdkn2aARF−/− mice expressing NrasG12V and Aurka (n = 6 mice) or NrasG12V and GFP (n = 8 mice). Statistical significance was calculated by a log-rank test. (j) Quantification of NRAS+ hepatocytes that are also cyclin B1+ or p-histone H3 (Ser10)+ 12 dafter delivery of pCaNIG (n = 3 mice) or pCaNIAU (n = 4 mice) into Cdkn2aARF−/− mice (left), or delivery of pCaNIG-shAurka-1,054 (n = 4 mice) or pCaNIG-shNC (n = 4 mice) into Trp53−/− mice (right). Data are mean  s.d. Statistical significance was calculated by Student’s t-test.

To further characterize the cell cycle arrest of NrasG12V;Trp53−/− hepatocytes, we delivered a construct co-expressing NrasG12V and the gene encoding red fluorescent protein (RFP) (pCaNIR; Supplementary Fig. 2a) into Trp53- or Cdkn2aARF-deficient mouse livers and performed flow-cytometry-based cell cycle analyses on sorted RFP+ hepatocytes (Supplementary Fig. 2b–d). NRASG12V- expressing Trp53−/− hepatocytes showed twice the DNA content of NrasG12V;Cdkn2aARF−/− hepatocytes, indicating that NRASG12V-posi- tive Trp53−/− hepatocytes might be arrested in the G2/M phase of the cell cycle (Fig. 1f). NRAS+ Trp53−/− hepatocytes showed two other features of G2/M cell cycle arrest—the presence of cyclin B1 and absence of phosphorylated histone H3 at Ser10 (Fig. 1g–i)—thus con- firming that NRASG12V-expressing Trp53-deficient cells are arrested in the G2/M transition of the cell cycle. We recently developed transposon vectors to stably express micro- RNA (miRNA)-based shRNAs (shRNAmir; hereafter referred to as shRNAs)27,29 in livers or tumors, and we sought to apply these tools to functionally identify the factors that are involved in the p19ARF-medi- ated G2/M arrest of Trp53−/− hepatocytes. We reasoned that positive- selection RNA interference (RNAi) screening would identify shRNAs that knock down potential mediators of the G2/M arrest in Trp53−/− mice after NrasG12V delivery.To ensure shRNA expression in the transposon-based mouse model of liver cancer, we took advantage of transposable elements that co- express oncogenic NrasG12V, genes encoding for the marker proteins green fluorescent protein (GFP) or luciferase (Luc), and an shRNA on a single transcript (pCaNIG-shRNA (for the expression of GFP) andpCaNIL-shRNA (for the expression of Luc); Fig. 2a)29. High knock- down efficiency in hepatocytes was confirmed by an in vivo biolu- minescence assay (Fig. 2b). In a proof-of-concept experiment, we delivered NrasG12V together with either a Cdkn2aARF-specific shRNA (pCaNIG-shCdkn2aARF) or a Cdkn2a-specific shRNA (pCaNIG- shCdkn2a; which leads to the knockdown of both Cdkn2aARF and Cdkn2aINK4A expression) into Trp53−/− hepatocytes and found that expression of either shRNA triggered the outgrowth of multifocal GFP+ HCCs (Supplementary Fig. 3a–c). In contrast, only weak back- ground tumorigenesis was observed after delivery of a non-targeting control shRNA (pCaNIG-shNC).

To evaluate the maximum complex- ity of shRNAs that can be screened for each individual mouse liver, a pCaNIG-shCdkn2a transposon was co-injected with pCaNIG-shNC at different ratios into Trp53−/− mice. Because a Cdkn2a-specific hair- pin that was diluted 1:48 was still efficient at inducing tumor devel- opment (Supplementary Fig. 3d), we reasoned that shRNA pools comprising up to 50 shRNAs could be screened in a single mouse.We compiled a miRNA-based shRNA library comprising shR- NAs that target the 60 most-upregulated genes in Trp53−/− versus Cdkn2aARF−/− mouse livers after NrasG12V delivery (mRNA expression array; Supplementary Table 1). We also included shRNAs targeting the mRNAs for 29 published p19ARF-interaction partners24, as well as for 12 genes whose products are involved in sumoylation, as p19ARF can promote sumoylation24 (Supplementary Fig. 4a). The custom- ized shRNA library (which targets 101 genes; Supplementary Table 2) was divided into six low-complexity pools (n = 41–53 shRNAs per pool), and each pool was delivered together with NrasG12V into Trp53−/− mice (n = 4 mice per pool). Seven weeks after transposon delivery we assessed tumor burden in the mice that were injected with the different shRNA library pools (Fig. 2c). Although most of the shRNA library pools did not trigger tumor development over background levels, mice that were injected with shRNA library pool 2 showed a significantly increased tumor incidence, with, on average,14.8 tumors per mouse (Fig. 2d and Supplementary Fig. 4b). Tumor nodules were excised from tumor-bearing livers of the mice in this group, and genomic DNA was isolated to PCR-amplify the shRNA cassettes.

A barcoded deep-sequencing approach was then applied to quantify the abundance of the individual shRNAs in these tumors.The majority of tumors mostly contained one or two shRNAs, sug- gesting that the tumors expanded clonally from single cells (with usu- ally one or two integrated transposable elements). Notably, whereas sequencing of shRNA pool 2 (before injection) revealed an almost equal distribution of all of the 45 shRNAs in this pool, the majority of tumors that were triggered by injection of this shRNA pool showed very high enrichment (often close to 100%) of different shRNAs tar- geting aurora kinase A (Aurka) (Fig. 2e), suggesting that AURKA has a central role in p19ARF-mediated tumor suppression in Trp53−/− hepatocytes. AURKA is an important regulator of cell cycle progres- sion and mitosis33, and it has been described as an oncoprotein and therapeutic target in various types of advanced tumors, including HCC9,10,34,35. However, impaired AURKA function can also result in increased tumorigenesis in mouse models and in humans36,37. In line with the latter observations, our data here suggest a context-depend- ent tumor-suppressive function of AURKA in the early stages of liver tumor development.We therefore tested the effect of independent Aurka-specific shR- NAs on NrasG12V-driven hepatocarcinogenesis in Trp53−/− livers. Three individual transposable elements encoding NrasG12V (pCaNIG; Fig. 2a) and different, validated Aurka-specific shRNAs (two scoring shRNAs from the screen plus one additional shRNA; Supplementary Fig. 4c) were stably delivered into Trp53-deficient mice. Trp53−/− mice that received the Aurka-specific shRNAs showed multifocal tumor development and a significantly reduced survival rate as com- pared to the control group (pCaNIG-shNC; Fig. 2f,g).We next set out to characterize how AURKA expression is regu- lated in NRASG12V-expressing Trp53-deficient hepatocytes. mRNA expression profiling revealed elevated Aurka mRNA levels in NrasG12V;Trp53−/− livers (mRNA expression array; Supplementary Table 1), and we validated these findings by quantitative PCR. NRASG12V expression resulted in a strong induction of Aurka mRNA expression in the livers of Trp53−/− but not of Cdkn2aARF-deficient or dko, mice; low Aurka expression levels were observed in all genetic backgrounds after delivery of an Nras allele (NrasG12V,D38A) that encodes an effector loop mutant incapable of signaling to RAS down- stream targets26,38 (Supplementary Fig. 5a).

To address whether the induction of Aurka mRNA in NrasG12V-expressing Trp53-deficient cells is mediated by transcriptional activation of Aurka, we took advantage of a vector that expresses Luc from the endogenous Aurka promoter39. This reporter construct was delivered hydrodynami- cally, together with NrasG12V (or with NrasG12V; D38A as a control), into Cdkn2aARF−/−, Trp53−/− or Cdkn2aARF−/−;Trp53−/− mice, and bioluminescence intensity was quantified after 3 d. We observed strong bioluminescence emission in livers from Trp53−/− mice but not in those from Cdkn2aARF−/− or dko mice after NrasG12V delivery (Fig. 2h). Because we also observed a strong induction of the p19ARF protein in Trp53−/− livers in parallel to activation of the Aurka pro- moter (Supplementary Fig. 5b), our data suggest that RAS–MAPK signaling induces p19ARF in Trp53-altered hepatocytes and that a p19ARF-dependent transcriptional activation of Aurka suppresses development of liver cancer.To further corroborate the tumor-suppressive role of AURKA in theliver, we next took a reverse approach and stably co-delivered NrasG12V and a full-length Aurka transgene (pCaNIAU; Supplementary Fig. 5c,d) into hepatocytes of Cdkn2aARF−/− mice. Exogenous Aurka expression reduced tumor growth and significantly prolonged sur- vival of tumor-bearing mice (Fig. 2i and Supplementary Fig. 5e). Exogenous expression of Aurka in Cdkn2aARF-deficient mouse livers also blocked mitotic entry, as determined by a significantly increased number of cyclin B1+ cells in NRAS- and AURKA-expressing Cdkn2aARF−/− liver cells than in NRAS- and GFP-expressing hepato- cytes (Fig. 2j and Supplementary Fig. 5f). Conversely, shRNA-medi- ated knockdown of Aurka in NRAS-expressing Trp53−/− hepatocytes significantly decreased the number of G2/M-arrested cells (cyclin B1+/phospho-histone H3−), as compared to those in shNC-expressing cells, 12 d after injection (Fig. 2j and Supplementary Fig. 5f).MYC mediates cell cycle re-entry of Trp53-altered hepatocytes The majority of human HCCs develop in a background of chronic liver damage, resulting in repetitive cycles of hepatocyte death and com- pensatory hepatocyte proliferation (reviewed in ref. 40). Therefore, we aimed to address whether the identified G2/M cell cycle arrest in Trp53-deficient hepatocytes is stable under conditions of chronic liver damage or whether enforced liver regeneration after damage would allow an escape from the arrest, with subsequent cell cycle re-entry.

NrasG12V and GFP (via pCaNIG) were co-delivered into Trp53- deficient mouse livers to induce the formation of G2/M-arrested hepatocytes. As a control we co-delivered NrasG12V, D38A in combi- nation with GFP. After 3 weeks, we subjected both groups of mice to repetitive cycles of carbon tetrachloride (CCl4) treatment, a well- established model of inducing chronic liver damage and fibrosis Figure 3 MYC induction during liver regeneration allowsNrasG12V; Trp53−/− hepatocytes to bypass a G2/M cell cycle arrest. (a) Schematic outline for CCl4 treatment of livers harboring G2/M-arrested NrasG12V; Trp53−/− hepatocytes or, as a control, non-arrested NrasG12V,D38A; Trp53−/− hepatocytes. (b) Representative images showing tumor burden (top) and GFP expression (bottom) in livers from Trp53−/− mice that were injected with a transposon expressing either oncogenic NrasG12V or non-oncogenic that shares many features of chronic liver disease in humans41 (Fig. 3a and Supplementary Fig. 6a). NrasG12V;Trp53−/− mice that were treated with CCl4 harbored more tumors than carrier-treated NrasG12V;Trp53−/− mice or CCl4-treated NrasG12V, D38A;Trp53−/− mice (Fig. 3b). These data suggest that liver regeneration triggered by chronic liver damage overrides the p19ARF- and AURKA-medi- ated G2/M arrest in Trp53-altered hepatocytes. Immunoblot analy- ses revealed that tumors derived from NrasG12V;Trp53−/− hepatocytes maintained high p19ARF and AURKA levels (Fig. 3c). Because MYC is a well-established key factor for regulating hepatocyte proliferation dur- ing liver regeneration42,43, we also analyzed MYC expression levels in CCl4-treated NrasG12V;Trp53−/− tumors and found higher amounts of MYC in these tumors than in NrasG12V;Cdkn2aARF−/− tumors (Fig. 3c and Supplementary Fig. 6b,c). Additionally, Trp53-altered tumors that spontaneously arose without CCl4 treatment maintained high levels of MYC (Supplementary Fig. 6d), suggesting that these tumors may depend on sustained MYC activity for tumor maintenance.We then asked whether increased MYC levels could account for the observed bypass of the G2/M cell-cycle-arrest phenotype in NrasG12V;Trp53−/− hepatocytes. We again delivered oncogenic NrasG12V (pCaN or pCaNIG) into Trp53-deficient mice to allow for the formation of G2/M-arrested hepatocytes. Three weeks later, we hydrodynamically delivered transposons encoding oncogenic Myc (pCaMIG or pCaM; Fig. 3d and Supplementary Fig. 6e), which do not trigger tumor development by themselves (Supplementary Fig. 6f).

Notably, we observed strong tumor development after injec- tion of the Myc-expressing transposons into NrasG12V;Trp53−/− mice, whereas there were only low tumor loads in the NrasG12V;Trp53−/− mice that had received an empty transposon construct (Fig. 3e,f). Of note, NrasG12V- and Myc-driven tumor nodules were all GFP+, regardless of whether GFP was delivered along with NrasG12V or Myc (Fig. 3e), indicating that the resulting tumors originated from previously G2/M- arrested hepatocytes.AURKA stabilizes MYC to promote tumor cell survivalWe next aimed to address whether Trp53-altered liver carcinomas with high levels of AURKA and MYC harbor vulnerabilities that could be exploited for therapeutic intervention. Owing to their elevated AURKA levels, we reasoned that these tumors might be hypersensitive to AURKA inhibition. We tested the effect of four different AURKA inhibitors10,11,44,45 (one of them, MLN8237, is currently being tested in clinical trials (for example, https://clinicaltrials.gov: NCT02444884, NCT01799278)) on NrasG12V- and Myc-expressing (NrasG12V;MycOE) Trp53−/− cells or, as a control, on Cdkn2aARF−/−and Cdkn2aARF−/−; Trp53−/− cells (all of the cell lines showed similar proliferation rates; Supplementary Fig. 7a). All of the inhibitors were applied in doses that allowed for an equal and efficient inhibition of AURKA’s kinase activity (as demonstrated by reduced autophosphorylation of AURKA at Thr288; Fig. 4a). However, two of the tested inhibitors (Aurora A inhibitor I (also known as S1451)44 and PHA-739358 (also known as danusertib45)) did not show any growth-suppressive effect in the tested cell lines (Fig. 4b,c). In contrast, two other inhibitors (MLN8237 and CD532)10,11 suppressed the growth of NrasG12V;MycOE;Trp53−/− cells but had minimal effect on NrasG12V;MycOE;Cdkn2aARF−/− and NrasG12V;MycOE;Cdkn2aARF−/−;Trp53−/− cells (Fig. 4b,c). To address whether off-target activities of MLN8237 and CD532 might account for the observed effects in NrasG12V;MycOE;Trp53−/− liver cancer cells, we tested RNAi-mediated suppression of Aurka expression in NrasG12V;MycOE;Cdkn2aARF−/−, Trp53−/− or Cdkn2aARF−/−;Trp53−/− double-knockout cells. All of the cell lines were retrovirally trans- duced with tetracycline (tet)-responsive Aurka-specific shRNAs (or control shRNAs). The resulting cell populations were injectedsubcutaneously into syngeneic immunocompetent mice, and after tumor formation, doxycycline was administered via drinking water to induce intratumoral shRNA expression. Aurka knockdown diminished the growth of NrasG12V;MycOE;Trp53−/− tumors but not of NrasG12V; MycOE;Cdkn2aARF−/− or Cdkn2aARF−/−;Trp53−/− double-knockout tumors (Supplementary Fig. 7b–d), indicating that AURKA is crucial for tumor maintenance in Trp53-deficient HCCs.

These genetic data support the idea that NrasG12V;MycOE;Trp53−/− liver carcinomas are sensitive to Aurka suppression and raise the question of why pharma- cological inhibition with MLN8237 and CD532, but not with S1451 and PHA-739358, kills NrasG12V;MycOE;Trp53−/− cells.To explain the different effects of the tested AURKA inhibitors, we studied their pharmacological properties. The two inefficient inhibitors (S1451 and PHA-739358) represent classical type I inhibi- tors, which, after binding, do not induce conformational changes of the AURKA protein46,47. On the contrary, MLN8237 and CD532 share some scaffolding functions with type II inhibitors and have been reported to induce conformational changes in the AURKA pro- tein after binding10,11. In neuroblastoma cells these conformational changes inhibit formation of a complex between AURKA and MYCN, resulting in proteasomal degradation of MYCN9–11.Because MYCN and MYC share a high degree of sequence homol- ogy and also many biochemical properties, we explored the notions of whether a protein complex between AURKA and MYC exists in TP53- altered liver carcinomas and whether AURKA is important for stabiliz- ing MYC levels in these tumors. Indeed, after immunoprecipitation of AURKA or MYC from different TP53-altered mouse or human HCC cells, an interaction between AURKA and MYC could be identified (Fig. 4d and Supplementary Fig. 8). To address whether MYC pro- tein degradation can be triggered by conformation-changing AURKA inhibitors, we incubated NrasG12V;MycOE;Trp53−/− mouse HCC cells with MLN8237 or CD532. These inhibitors induced a marked decrease in the abundance of MYC protein (Fig. 4e) but not of Myc mRNA (Supplementary Fig. 9a,b). Incubation of the same cells with S1451 did not affect MYC levels, and treatment of NrasG12V;MycOE; Cdkn2aARF−/− mouse hepatoma cells with the conformation-changing inhibitors only induced a marginal decrease in MYC protein levels after 3 d (Fig. 4f and Supplementary Fig. 9c). To ensure that reduced MYC expression in NrasG12V;MycOE;Trp53−/− cells is not the result of a reduced proliferation rate, we treated these cells with nocodazole to block them in the G2/M phase of the cell cycle; this did not reduce MYC protein levels (Supplementary Fig. 9d). Together, these data suggest that conformation-changing AURKA inhibitors induce MYC protein degradation in Trp53-altered mouse HCCs.Because MYC protein degradation occurs after 2 d of treatment and therefore earlier than therapy-induced cell death (which occurs after 4–5 d of treatment) (Fig. 4c and Supplementary Fig. 9e), we rea- soned that cell death might be preceded by changes in the cell cycle.

We performed DNA staining and FACS-based cell cycle analysis and observed that treatment of NrasG12V;MycOE;Trp53−/− cells with con- formation-changing AURKA inhibitors induced a transient G2/M cell cycle arrest, with a peak that occurred after 2 d of treatment (Fig. 4g and Supplementary Fig. 9f).We recently showed that an NrasG12V-driven transposon-based genetically engineered HCC mouse model closely resembles therapy resistance in human HCC29. To test whether conformation-changing AURKA inhibitors harbor therapeutic potential for the treatment of therapy-refractory liver carcinomas in vivo, we set out to conduct pre- clinical treatment studies in such a model. The conformation-changing AURKA inhibitor MLN8237 is clinically the most advanced35, and it is well tolerated in humans48 and in Trp53-deficient mice that are subjected to cycles of a 5-d treatment period followed by a 2-d recov- ery period (Supplementary Fig. 10). NrasG12V (via pCaN) and Myc (via pCaMIG) were hydrodynamically delivered into Trp53- and Cdkn2aARF-deficient mouse livers to induce HCC formation, and treatment with MLN8237 or the carrier was started 1 week after stable intrahepatic transposon delivery (a time point at which small clusters of transformed cells can be detected). MLN8237 monotherapy showed marked therapeutic efficacy to NrasG12V-driven Trp53-deficient mouse HCCs, allowing even for long-term survival in 50% of the mice. In contrast, only a modest gain in survival, comparable to the gain in sur- vival induced by the multikinase inhibitor sorafenib in this model29, was observed after MLN8237 was administered to mice harboring NrasG12V-driven Cdkn2aARF-deficient liver carcinomas (Fig. 4h,i).The turnover of MYC is regulated by its phosphorylation at Thr58, which triggers its ubiquitination by F-box and WD-40 domain protein 7 (FBXW7) and subsequently its proteasomal degradation49. To test whether the decrease in MYC levels after treatment with conforma- tion-changing AURKA inhibitors is dependent on Thr58 phosphor- ylation and proteasomal degradation, we generated Trp53-deficient liver tumors which were driven by NrasG12V and a mutated Myc that is incapable of being phosphorylated at Thr58 (MycT58A; Fig. 5a).

MLN8237 treatment did not alter growth of NrasG12V;MycT58A;Tr p53−/− tumors (Fig. 5b,c). Similarly, neither MLN8237 nor CD532 treatment induced MYC protein degradation in NrasG12V;MycT58A;T rp53−/− liver cancer cells (Fig. 5d).We then blotted AURKA immunoprecipitate lysates with antibodies specific for phospho-MYC and detected predominantly doubly phos- phorylated MYC (at Thr58 and Ser62) in the complex with AURKA (Fig. 5e). To further corroborate this finding, we performed pulldown assays with phosphorylated (at Thr58, or at Thr58 and Ser62) or non- phosphorylated MYC peptides and indeed could show that only phos- phorylated MYC peptides efficiently pull down endogenous AURKA from protein lysates of HLF liver cancer cells (Fig. 5f).Unexpectedly, we did not observe a disruption of preformed MYC– AURKA complexes even after prolonged incubation with CD532, as determined by immunoprecipitation assays (Fig. 5e,g). In silico Figure 5 Conformation-changing AURKA inhibitors prohibit the formation of p-MYC–AURKA protein complexes. (a) Schematics of transposable elements for intrahepatic co-expression of oncogenic NrasG12V and MycT58A. (b) Representative images for intrahepatic tumor burden and GFP expression 3 weeks after delivery of NrasG12V (pCaN) and MycT58A (pCaMT58A) into Trp53−/− mice and treatment of the mice with MLN8237 or carrier (n = 4 mice per group). Scale bar, 1 cm. (c) Survival analysis of Trp53−/− mice treated as in b (n = 5 mice per group). Statistical significance was calculated by a log-rank test; n.s., not significant. (d) Representative western blot analysis of NrasG12V;MycT58A;Trp53−/− cells after 1, 2, or 3 d of treatment with MLN8237 or CD532 (n = 2 biological replicates per condition). (e) Representative western blot analysis after immunoprecipitationof HLF cell lysates with an AURKA-specific antibody after treatment with MLN8237, CD532, or the corresponding DMSO concentrations (n = 2 independent experiments). (f) Representative western blot analysis after pulldown of MYC46–74 peptides that are differentially phosphorylated at Thr58 and Ser62 (n = 2 independent experiments). (g) Representative western blot analysis of HLF cell lysates that were immunoprecipitated with an AURKA- specific antibody (or IgG) and treated with either CD532 or DMSO (n = 2 independent experiments). (h) Binding cavities (gray mesh surface) in the proposed MYC–AURKA protein complex (Protein Data Bank (PDB) IDs: 4J8M (AURKA in complex with CD532) and 2J50 (AURKA in complex withPHA-739358)). (i) Schematic outline of the p-MYC–AURKA protein complex after treatment with different kinds of AURKA inhibitors.

Conformation- changing AURKA inhibitors prevent the de novo formation of MYC–AURKA complexes but do not disrupt preformed complexes.of treatment with MLN8237, S1451, or the corresponding DMSO concentrations (n = 3 biological replicates per condition). Scale bar, 1 cm.(d)Representative western blot analysis of MYC protein levels in cells with mutated TP53 (PLC/PRF/5, HLE, and HLF) or WT TP53 (HepG2 and SMMC-7721) after 1, 2, or 3 d of treatment with MLN8237 (n = 2 biological replicates per condition). -actin was used as a loading control.(e–h) Tumor development in mice after subcutaneous injection of Hep3B (MLN8237-treated, n = 4 mice; carrier-treated, n = 4 mice) (e), PLC/PRF/5 (MLN8237-treated, n = 5 mice; carrier-treated, n = 5 mice) (f), HepG2 (MLN8237-treated, n = 4 mice; carrier-treated, n = 6 mice) (g), or SMMC-7721 (MLN8237-treated, n = 6 mice; carrier-treated, n = 4 mice) (h) human HCC cells and treatment with either MLN8237 or the corresponding carrier.Data are mean  s.e.m. Statistical significance was calculated by Student’s t-test; n.s., not significant. modeling confirmed an efficient binding of AURKA to only doubly phosphorylated MYC (Supplementary Fig. 11) and furthermore revealed that conformation-changing AURKA inhibitors (such as CD532) cannot access their binding site on the AURKA protein when AURKA is bound to MYC (in contrast to classical type I AURKA inhibitors such as PHA-739358; Fig. 5h). Furthermore, a Glide scor- ing function indicated very high stability for the MYC–AURKA complex (docking score −11.5 kcal mol−1 for residues 55–68), thus providing a possible explanation for why both compounds are not capable of disrupting preformed complexes. Taken together these data suggest that conformation-changing AURKA inhibitors are efficient in preventing de novo formation of MYC–AURKA complexes but are incapable of disrupting preformed complexes (Fig. 5i). These findings provide an explanation for why it takes more than 1 d after exposure to MLN8237 or CD532 until MYC levels in NrasG12VMycOE;Trp53−/− cancer cells are diminished (Fig. 4e).We next explored whether TP53-altered human HCCs exist that harbor high AURKA and MYC levels and that may thus represent candidates for treatment with conformation-changing AURKA inhibitors. We first analyzed AURKA mRNA levels in a cohort of human HCC samples that have a defined TP53 mutational status and found a strong correlation between TP53 status and AURKA mRNA expression levels (Fig. 6a). Notably, this correlation was still significant when AURKA levels were normalized to the proliferation rate of the tumor (as determined by expression of marker of proliferation Ki-67 (MKI67)), thus ruling out the possibility that TP53-altered tumors have higher AURKA levels due to higher proliferation rates (Supplementary Fig. 12a). Furthermore, we detected a correlation between AURKA mRNA abundance and the abundance of mRNA transcripts for genes that are transcriptionally activated by MYC (such as BUB1B (also known as BUBR1), NCL, CDK4, PTMA and CAD2,50,51) in human HCCs(Fig. 6b and Supplementary Fig. 12b–e), indicating that human HCC samples with mutated TP53 and high levels of AURKA and MYC exist.

Although not formally shown here, it is conceivable that such human HCCs might also develop after a MYC-enabled escape from a p14ARF- and AURKA-mediated G2/M arrest.Because of the strong correlation observed between TP53 muta- tional status and AURKA expression level on the one hand and AURKA expression and MYC target gene expression on the other hand, we hypothesized that sensitivity of human HCC toward conformation- changing AURKA inhibitors might be predictable on the basis of the mutational status of TP53. To address this question experimentally, we subjected a panel of human HCC cell lines with defined TP53 mutational status to treatment with MLN8237 or, as a control, to the non-conformation-changing AURKA inhibitor S1451. Our human HCC panel comprised cells with a TP53 deletion (Hep3B), a TP53 mutation (PLC/PRF/5, HLE, HLF and Huh7) or a WT TP53 status (HepG2, Huh6, SMMC-7721 and JHH-1). These cells were treated with DMSO, S1451 or MLN8237. Whereas only a weak and nonse- lective therapeutic efficacy was observed after treatment with S1451, treatment with MLN8237 almost completely suppressed growth of the TP53-altered human HCC cells. In contrast, HCC cells carrying a WT TP53 gene were barely affected by MLN8237 treatment (Fig. 6c). We also determined MYC expression at the protein and mRNA lev- els in MLN8237-treated human HCC cell lines and observed that MLN8237 treatment induced degradation of MYC protein in TP53- altered human HCC cells (Fig. 6d and Supplementary Fig. 13a–c). In contrast, HepG2 and SMMC-7721 cells (HCC cell lines with WT TP53 and elevated MYC expression) showed no or only a mild decrease in MYC protein levels after treatment, indicating that these cell lines have acquired other strategies to maintain elevated MYC levels (Fig. 6d and Supplementary Fig. 13d,e). Consistent with this, we were not able to detect an interaction between MYC and AURKA in these cells (Supplementary Fig. 13f).We next sought to determine whether the observed hypersensi- tivity of TP53-altered human hepatomas toward MLN8237 would translate into a therapeutic efficacy of the drug to xenografted human HCCs. TP53-altered human HCC cells (Hep3B and PLC/PRF/5) and HCC cells with a WT TP53 status (HepG2 and SMMC-7721) were subcutaneously injected into the rear flanks of nude mice, and after tumor formation (77.8 mm2 average tumor size at the start of treat- ment), mice were subjected to treatment with MLN8237 or carrier. Pronounced antitumorigenic effects, with marked tumor remission and almost complete suppression of tumor growth, were observed in tumors derived from cells with altered TP53 after MLN8237 treatment (Fig. 6e,f). In contrast, liver tumors harboring a WT TP53 gene were marginally affected by MLN8237 treatment (Fig. 6g,h).

DISCUSSION
Our study provides an example for a p53-independent function of p19ARF and explains how the understanding of such mechanisms can be translated into therapeutic strategies. p19ARF is induced upon onco- genic stress in Trp53-altered hepatocytes, and it mediates G2/M cell cycle arrest via transcriptional activation of Aurka. Tumorigenesis from G2/M-arrested Trp53-altered hepatocytes required selection for high MYC expression, and survival of the resulting tumors was depend- ent on AURKA-mediated MYC stabilization. The MYC oncogene is upregulated in more than 50% of all human cancers; however, owing to its structure, MYC is regarded as undruggable. Our data show that MYC and AURKA, two proteins that have been described to regulate each other on a transcriptional level52, also interact physically in TP53- altered cells and that treatment of cells with conformation-changing AURKA inhibitors efficiently prevents formation of a MYC–AURKA complex, thereby resulting in subsequent degradation of MYC.

In the past decade, all efforts to identify effective targeted molecular therapies for HCCs have failed. By applying state-of-the-art in vivo shRNA-based screening technology7,27,29,53, we recently identified the induction of MAPK14 (p38a) signaling as a mechanism by which HCCs become resistant to sorafenib, which currently represents the only systemic therapy for patients with advanced liver cancer, and showed that pharmacological inhibition of MAPK14 was able to restore sorafenib sensitivity. However, here we show that AURKA inhibitors, even when administered as a monotherapy, suppress tumor growth and promote long-term survival of 50% of treated mice that harbor aggressive liver carcinomas. To the best of our knowledge there is no other therapy with comparable therapeutic efficacy in mouse models that reflect therapy resistance of human HCCs. Notably, our xenograft data suggest that AURKA inhibitors may show efficacy to human HCCs that have TP53 mutations.From epidemiological data it is well established that chronic liver damage with subsequent liver cirrhosis development represents the most important risk factor for the development of liver cancer.However, from a mechanistic point of view it is still poorly under- stood how chronic liver damage and compensatory liver regeneration fuel the development of liver cancer. Our study provides a molecular mechanism—liver damage with compensatory liver regeneration pro- motes tumorigenesis via induction of MYC, which in turn allows the bypass of a p19ARF- and AURKA-mediated G2/M cell cycle arrest in Trp53-altered hepatocytes. On CD532 the basis of the data presented here, we suggest that MLN8237, or other conformation-changing AURKA inhibitors, should be tested in clinical trials for the treatment of advanced TP53-altered HCCs.