A phosphokinome‐based screen uncovers new drug synergies for cancer driven by liver‐specific gain of nononcogenic receptor tyrosine kinases

Genetic mutations leading to oncogenic variants of receptor tyrosine kinases (RTKs) are frequent events during tumorigenesis; however, the cellular vulnerability to nononcogenic RTK fluctuations has not been characterized. Here, we demonstrated genetically that in the liver subtle increases in wild‐type Met RTK levels are sufficient for spontaneous tumors in mice (Alb‐R26Met), conceptually illustrating how the shift from physiological to pathological conditions results from slight perturbations in signaling dosage. By analyzing 96 different genes in a panel of tumor samples, we demonstrated that liver tumorigenesis modeled by Alb‐R26Met mice corresponds to a subset of hepatocellular carcinoma (HCC) patients, thus establishing the clinical relevance of this HCC mouse model. We elucidated the regulatory networks underlying tumorigenesis by combining a phosphokinome screen with bioinformatics analysis. We then used the signaling diversity results obtained from Alb‐R26Met HCC versus control livers to design an “educated guess” drug screen, which led to the identification of new, deleterious synthetic lethal interactions. In particular, we report synergistic effects of mitogen‐activated protein kinase kinase, ribosomal S6 kinase, and cyclin‐dependent kinase 1/2 in combination with Bcl‐XL inhibition on a panel of liver cancer cells. Focusing on mitogen‐activated protein kinase kinase and Bcl‐XL targeting, we mechanistically demonstrated concomitant down‐regulation of phosphorylated extracellular signal–regulated kinase and myeloid cell leukemia 1 levels. Of note, a phosphorylated extracellular signal–regulated kinase+/BCL‐XL+/myeloid cell leukemia 1+ signature, deregulated in Alb‐R26Met tumors, characterizes a subgroup of HCC patients with poor prognosis. Conclusion: Our genetic studies highlight the heightened vulnerability of liver cells to subtle changes in nononcogenic RTK levels, allowing them to acquire a molecular profile that facilitates the full tumorigenic program; furthermore, our outcomes uncover new synthetic lethal interactions as potential therapies for a cluster of HCC patients. (Hepatology 2017;66:1644–1661).

R eceptor tyrosine kinases (RTKs) are frequently mutated in different types of cancer. (1) Through their aberrant activation, RTKs confer upon cancer cells a range of biological advantages and, in some cases, addiction. This is illustrated by the deleterious effects of agents targeting oncogenic RTKs on cancer cells. (1,2) The relevance of RTKs in cancer also derives from bioinformatics methods applied in Abbreviations: Cdk, cyclin-dependent kinase; ERK, extracellular signal-regulated kinase; HCC, hepatocellular carcinoma; HGF, hepatocyte growth factor; HLF, human lung fibroblast; Mcl1, myeloid cell leukemia 1; MEK, mitogen-activated protein kinase kinase; p-, phosphorylated; RSK, ribosomal S6 kinase; RTK, receptor tyrosine kinase. combination with genome-wide profiling and protein network-based studies, which have highlighted RTK signaling as one of the few core pathways that impacts tumor evolution. (3)(4)(5) In this context, we previously showed how alterations in RTK signaling influence other core pathways through interconnecting signaling nodes, which are vulnerable targets in cancer cells. (6)(7)(8) Although genetic strategies have demonstrated how oncogenic RTK signaling mutations found in human patients are capable of triggering cancer, they largely recapitulate situations in which cancer cells are addicted to these drastic genetic oncogenic forms, which, in most cases, act as dominant signaling alterations. Nevertheless, enhanced RTK activation levels are often observed in tumors despite the absence of genetic mutations. Their significance in cancer biology and as a signature remains unclear.
The multitude of signaling alterations orchestrated by deregulated RTKs have been largely explored by focusing on putative candidate signaling circuits. (1) Recent large-scale screens have been carried out in an attempt to gain broader knowledge regarding altered signal transduction systems operating in cancer cells and identify vulnerable nodes to target. For example, as determined using cancer cell lines characterized by oncogenic Met addiction, the Ras and phosphoinositide 3-kinase pathways are required to sustain addiction. (9) Another elegant example of synergistic drug discovery in RTK-driven tumorigenesis involved a drug screen in combination with network modeling, leading to the identification of synergistic responses targeting cyclin-dependent kinase 4 (Cdk4) with insulin-like growth factor 1 receptor, epidermal growth factor receptor, or Akt in dedifferentiated liposarcoma. (10) Alternatively, pooled short hairpin RNA screening strategies permitted vulnerable points in drug resistance to be uncovered, exemplified by identification of the ability of p38a mitogenactivated protein kinase to elevate mitogen-activated protein kinase kinase (Mek)-extracellular signal-regulated kinase (Erk) signaling in sorafenib-resistant cells. (11) Although all of these studies highlight examples of critical RTK pathways that participate in mediating the oncogenic properties of cells, major limitations are associated with the biological systems that are often used: the addiction of cancer cells to oncogenic RTK forms and/or downstream signaling component(s). Little is known about the signaling platforms originated by and operating in cancer cells with subtle, yet still functionally relevant, increases in RTK inputs as well as whether these platforms possess vulnerabilities. Such knowledge deficiencies are also attributable to a lack of proper genetic systems that model the biological context of nonchronic RTK activation.
We recently explored the ability of cells to handle subtle changes in RTK levels in vivo using conditional transgenic mice in which the expression of wild-type Met is slightly enhanced above endogenous levels (R26 stopMet mice) (12) and demonstrated the vulnerability of restricted cell types during embryogenesis. (13) In the present study, we used the R26 stopMet system to genetically explore cell vulnerability versus resilience to enhanced RTK levels in adulthood. Although most cells are resistant, liver cells (Alb-R26 Met ) are sensitive to enhanced wild-type Met levels and develop spontaneous hepatocellular carcinoma (HCC). We discovered new, deleterious synthetic lethal interactions for a panel of HCC cells. We also established the clinical relevance of our findings by identifying a signaling signature characterizing a subgroup of HCC patients with poor prognosis.

Materials and Methods
More detailed procedures can be found in the Supporting Information.

WESTERN BLOTTING
Protein extracts were prepared, and western blot analysis was performed as described. (7,13,18) The antibodies used are reported in Supporting Table S1.

STATISTICAL ANALYSIS
Results are expressed as the median or as the mean 6 standard error of the mean, according to sample distributions. Statistically significant differences were estimated by applying unpaired Student t tests to data showing normal distributions and Mann-Whitney tests in all other situations (e.g., for xenografts studies). P values are indicated in figures. The cumulative overall survival and disease-free survival rates were calculated using the Kaplan-Meier method. P < 0.05 was considered significant. All tests were two-sided. All analyses were performed with Stata 13.

ENHANCED NONONCOGENIC RTK Met LEVELS IN THE LIVER CAUSE HCC PATHOGENESIS
To enhance Met-RTK levels in all tissues, we crossed R26 stopMet with Deleter-Cre transgenics to obtain Deleter-Cre;R26 Met mice (referred to as Del-R26 Met ), in which the Met transgene (Met tg ) is expressed in all tissues following LacZ-stop cassette excision (Supporting Fig. S1A). Whereas fully recombined Del-R26 Met mice at birth exhibit a hyperflexed forelimb and weak hind limb phenotype, (13) pups with a mosaic deletion of the LacZ-stop cassette appeared normal. However, some died beginning at 25 weeks of age (Supporting Fig. S1B). A proportion of Del-R26 Met mice spontaneously developed liver-specific tumors (Supporting Fig. S1C), which were identified as HCC through histopathological analysis (Supporting Fig. S1D). Based on these findings, whereas most tissues are competent to buffer enhanced RTK Met levels, liver cells are sensitive. Moreover, the wild-type form of Met, not just its oncogenic forms, is sufficient to induce spontaneous tumorigenesis when its expression is enhanced above endogenous levels.
To generate hepatocyte-specific conditional R26 Met mice, we crossed Alb-Cre transgenics with R26 stopMet mice (referred to as Alb-R26 Met ). The specificity of LacZ-stop cassette deletion in the liver was supported by the use of mice carrying a modified version of the R26 stopMet construct, in which Met tg was followed by an internal ribosome entry site-luciferase reporter (R26 stopMet-IRES-Luciferase , referred to as R26 stopMet-Luc ; Supporting Fig. S2A). In vivo imaging of Alb-R26 Met-Luc mice confirmed liver-specific luciferase expression (Supporting Fig. S2B). Western blot analysis revealed increased Met tg expression in Alb-R26 Met livers over time after birth (Supporting Fig. S2C), consistent with the reported Alb-Cre expression. Macroscopic tumors were observed in 42%, 66%, and 79% of Alb-R26 Met mice at the ages of 40-48, 49-67, and >67 weeks, respectively (Fig. 1A,B). According to anatomopathological analysis, the majority of tumors range from poorly differentiated to well-differentiated HCC (Supporting Fig. S2D,E and Table S2). Histological analyses showed that Alb-R26 Met tumors express Met tg and contain a proportion of cells with active Met (Supporting Fig. S3A). We compared Met expression levels in Alb-R26 Met (n 5 32) and human (n 5 249; from Tao et al. (19) ) liver tumors as well as in Alb-R26 Met (n 5 6; see below) and human (n 5 8) HCC cell lines. Met levels in the Alb-R26 Met genetic setting (3. 16 Table S4). Based on these findings, Alb-R26 Met tumors belong to the so-called proliferative HCC group; and within this group, they correspond to the progenitor subclass rather than to the Wnt subgroup. (20) Next, we performed a series of biochemical studies to characterize Met-signaling levels. In Alb-R26 Met tumors, Met tg is expressed, and its expression correlates with a concomitant increase in Met phosphorylation levels on Tyr 1234-1235 , Tyr 1003 , and Tyr 1349 (Fig. 1G). (21,22) This is accompanied by increased phosphorylation of Erk, but not Akt, compared to control livers (Fig. 1G). We then explored quantitative and qualitative Met levels in cells prior to tumorigenesis using primary hepatocytes from E15.5 wild-type and Del-R26 Met livers. Quantification analysis indicated an approximately 3-fold increase in Met expression levels in Del-R26 Met hepatocytes compared to controls (Fig. 1H), consistent with quantitative RT-PCR data shown in Fig. 1F and with Fan et al. (13) Interestingly, qualitative analysis of the phosphorylation status of Met revealed (1) high phosphorylation levels of Tyr 1234-1235 , which are not further enhanced upon hepatocyte growth factor (HGF) stimulation, and (2) that phosphorylation of Tyr 1349 and Tyr 1356 is dependent on HGF stimulation (Fig. 1I,J). Thus, subtle increases in Met expression in hepatocytes (3-fold) confer a poised state upon the receptor: Met is active (as shown by high phosphorylated Tyr 1234-1235 [pTyr 1234-1235 ] levels) but not fully competent for signaling, which is conditioned by HGF stimulation. Based on these data, Alb-R26 Met mice represent a genetic model of tumorigenesis in which vulnerable cells are challenged by subtle increases in RTK levels rather than the overexpression of an oncogenic form.

IS NOT Met-ADDICTED
To assess whether Alb-R26 Met HCC correspond to Met-addicted tumorigenesis, we generated several HCC cell lines from individual Alb-R26 Met tumors (referred to as Alb-R26 Met HCC cells; Fig. 2A). Alb-R26 Met HCC cells are of liver origin and express albumin and alpha-fetoprotein in contrast to mouse embryonic fibroblasts, which were used as a negative control ( Fig. 2B; Supporting Table S5). A consistent 3-fold increase in Met expression; the up-regulation of Mki67, Prom1, Cdh1, Igf2bp3, Cdc20, Nrcam, and Cd24; and the down-regulation of Lrg5, Tbx3, Glul,  (21,22) ) is dependent on HGF stimulation.  Table  S6), similar to Alb-R26 Met tumors (Fig. 1F). Western blot analyses confirmed Met expression and phosphorylation of Met as well as downstream signals in Alb-R26 Met HCC cells (Fig. 2D). According to anchorage-independent growth assays and xenografts in nude mice, Alb-R26 Met HCC cells retain tumorigenic properties (Fig. 2E). We then explored whether Alb-R26 Met HCC cells are addicted to Met and found that none of them are fully sensitive to Met inhibitors ( Fig. 2F; Supporting Fig. S3B). This is consistent with our genetic strategy, which was based on use of the wild-type, rather than the oncogenic, form of Met tg as well as a subtle increase in Met expression levels. Given these results, the tumorigenesis modeled by Alb-R26 Met mice is characterized not by Met addiction but rather by a signaling context originating from, and associated with, subtle increase in RTK levels.

SIGNALING PERTURBATIONS IN RTK-DRIVEN TUMORIGENESIS IDENTIFIED IN A PHOSPHOKINOME SCREEN, LEADING TO THE DISCOVERY OF NEW SYNTHETIC LETHAL INTERACTIONS
To undertake an unbiased analysis of hundreds of signaling proteins in tumors linked to subtle enhanced RTK levels, we employed a phosphokinome screen based on protein antibody microarrays (Kinex Antibody Microarray). We performed these experiments using liver tumors dissected from Alb-R26 Met mice rather than Alb-R26 Met HCC cells to identify signaling changes occurring in vivo and to take into account the context of intrinsic tumor heterogeneity (Fig. 3A). The signaling profiles of Alb-R26 Met tumor (n 5 19) versus control (n 5 7) pools were analyzed by comparing the levels of phosphorylated signals (300 epitopes) and kinase expression levels (500). Among 118 identified changes, 43 were top-ranked for further validation based on their fold change, low error rate, and high antibody quality (Supporting Table S7). By performing a Kinetworks custom multiantibody screen, we confirmed changes in 26 signals between control and tumor pools by western blotting (Supporting Table S7). Analyses of expression and/or phosphorylation levels of 23 signals in individual tumor and control samples revealed consistent trends in the changes for distinct signals (using Kinetworks custom multisample comparison; Supporting Fig. S4A and Table S8). When we bioinformatically analyzed interactions between the 43 top-ranked signaling components, the majority formed an interactome characterized by multiple links between signals, a "signaling node" (p53), and several components of the Ras pathway (Fig. 3B).
We next asked whether the Alb-R26 Met interactome could be used to predict vulnerable signaling points for HCC cells, taking into account three hypotheses. (1) The intense network of interactions between components may reflect the redundancy of certain signals during inhibition, and it may therefore be necessary to target several points to destabilize the interactome. (2) The interactome may be composed of a very restricted set of sensitive points required for the tumorigenic properties of cells among the most abundant changes that may be either irrelevant or linked to specific biological properties. (3) As tumor cells demonstrate a greater dependence on stress support pathways than normal cells, it may be necessary to simultaneously destabilize the interactome while targeting stress support pathways. Taking into account these three hypotheses, we designed an "educated guess" drug screen and assessed the effects of 42 treatments (as single drugs or two-drug combinations; Supporting Tables S9 and S10). Alb-R26 Met HCC1 cells (not responding to Met inhibitors) were used for an unbiased search of drugs targeting HCC modeled by the Alb-R26 Met genetic setting. A dose-response screen was performed by administering drugs under optimal culture conditions (10% serum) for 48 hours; cell numbers were measured in a metabolic activity-based cell viability assay. Two out of the 42 treatments were deleterious for cells: the combined inhibition of Mek (by PD184161) or Cdk1/2 (by JNJ-7706621) together with the mitochondrial stress support pathway (by ABT-737) was not permissive for Alb-R26 Met HCC cell viability, whereas single treatments did not have significant effects (the Mek inhibitor elicited only a partial response at higher doses; Fig. 3C,D; Supporting Fig. S4B,C). Importantly, combined treatments of ABT-737 and Fas ligand or tumor necrosis factor-a did not induce Alb-R26 Met HCC cell death, excluding the possibility that ABT-737 sensitizes cells to any type of death-inducing signal (Supporting Fig. S5A).
Overall, our phosphokinome screen combined with functional analyses identified Mek or Cdk1/2 together with the stress support pathway as a druggable synthetic lethal interaction.  Next, we explored the net contribution of distinct components of the Ras and stress support pathways in a panel of Alb-R26 Met HCC cell lines. In terms of the Ras pathway, the inhibition of Mek and ribosomal S6 kinase (Rsk), but not Raf and Erk2, together with the stress support pathway is deleterious for all Alb-R26 Met HCC cell lines tested ( Fig. 4A; Supporting Fig. S5B). Notably, the Alb-R26 Met HCC cell lines are resistant to sorafenib, a standard care for HCC (Fig. 4A). Regarding the mitochondrial stress support pathway, we compared the effectiveness of ABT-737 (targeting Bcl-2, Bcl-XL, Bcl-w) with that of ABT-199 (targeting Bcl-2) and WEHI-539 (targeting Bcl-XL). The inhibition of Bcl-XL, but not Bcl-2, together with Mek or Cdk1/2 was detrimental for Alb-R26 Met HCC cells (Fig. 4B). Thus, Bcl-XL inhibition confers Alb-R26 Met HCC cell sensitivity to Mek, Rsk, or Cdk1/2 targeting. Based on the Chou-Talalay additivity-based combination index score, combinations of Mek, Rsk, and Cdk1/2 with Bcl-XL inhibition resulted in strong synergistic interactions (Fig. 4C,D).
Finally, we focused on the synergistic effects of Mek and Bcl-XL inhibition to evaluate in vivo the effectiveness of this drug combination on tumor growth by performing xenografts of Alb-R26 Met HCC cells in nude mice. Tumor volumes as well as tumor volume changes were significantly reduced in mice treated with a combination of Mek and Bcl-XL inhibitors (selumetinib and ABT-737, respectively; Fig. 4E,F; Supporting  Fig. S5C). The combination of selumetinib with ABT-737 doses used in vivo was not toxic, as revealed by no significant changes in mouse weight during the treatment (Supporting Fig. S5D). Thus, the concomitant inhibition of Mek and Bcl-XL is deleterious for Alb-R26 Met HCC cells both in vitro and in vivo.
We next explored the effectiveness of these new synthetic lethal interactions on human HCC cells characterized by distinct molecular features and observed a range of sensitivity to these drug combinations. Cell numbers were drastically reduced when HepG2, HLF, and Hep3B cells were treated with Mek or Cdk1/2 together with Bcl-XL inhibitors compared to other human cell lines ( Fig. 5A; Supporting Fig. S6), indicating a selective sensitivity of human HCC cell types to the identified treatments. The in vitro tumorigenesis of HepG2 cells was impaired when Mek and Bcl-XL were inhibited, whereas tumorigenesis was only partially reduced following Rsk or Cdk1/2 and Bcl-XL inhibition (Fig. 5B). Moreover, the tumorigenesis of HLF cells was drastically affected by the inhibition of Mek, Rsk, or Cdk1/2 and Bcl-XL (Fig.  5C). Notably, the combined treatments we discovered elicit beneficial effects for cells that do not respond to sorafenib (Fig. 5B,C). Finally, in vivo combined inhibition of Mek and Bcl-XL significantly reduced tumor formation compared to vehicle or single treatment (Fig. 5D,E; Supporting Fig. S7A), without showing toxic effects (Supporting Fig. S7B). These findings highlight the potential of combining Mek, Rsk, or Cdk1/2 together with Bcl-XL inhibition to counteract HCC cell tumorigenesis and raise the possibility of applying these new synthetic lethal interactions to human HCC subgroups.

Alb-R26 Met HCC CELL DEATH
We next focused on the lethal synergistic effects of Mek and Bcl-XL inhibition on Alb-R26 Met HCC cells and investigated the underlying mechanism of action. Treatment of cells with ABT-737 led to a strong increase in Erk phosphorylation levels, which was abrogated in the presence of selumetinib ( Fig. 6A;  Supporting Fig. S7C). Concomitantly, ABT-737 predisposed cells to apoptosis as revealed by cleaved Cas-pase3, levels of which were significantly enhanced by selumetinib cotreatment (Fig. 6A; Supporting Fig.  S7C). Analysis of several proapoptotic and antiapoptotic regulators revealed drastic down-regulation of the antiapoptotic myeloid cell leukemia 1 (Mcl1) protein following combined ABT-737 and selumetinib treatment, whereas other Bcl-2 family members were not affected ( Fig. 6A; Supporting Fig. S7C). Next, we analyzed the impact of drug treatments on Alb-R26 Met HCC cell cycle progression. The percentage of cells in G 0 /G 1 phase was increased at the expense of those in S phase in the presence of selumetinib or ABT-737 plus selumetinib but not of ABT-737 alone (Fig. 6B). Thus, this new synthetic lethal interaction is likely to overexpression interfered with Caspase3 activation and cell death triggered by ABT-737 plus selumetinib (Fig. 6C-E; Supporting Fig. S7D).

Alb-R26 Met TUMORS AND A SUBGROUP OF HCC PATIENTS
High levels of pMEK1 have been detected in 49% of HCC samples. (23) High MCL1 levels have been reported in approximately 50% of HCC patients, with an intriguing significant correlation with BCL-XL expression. (24) However, a putative correlation between high levels of MCL1/BCL-XL and MEK/ERK deregulation has never been explored. Remarkably, Alb-R26 Met tumors were characterized by increased levels of pMek, pErk, Bcl-XL, and Mcl1 (Fig. 7A). Comparable levels of two other antiapoptotic members of the Bcl-2 family, Bcl-2 and Bcl-w, in control livers and tumors corroborate the impact of targeting Bcl-XL (Fig. 7A). Next, we asked whether there is a clinical correlation between high levels of pMek/pErk, Bcl-XL, and Mcl1 in human HCC by analyzing a total of 116 patients. pERK, rather than pMEK, levels were analyzed as antibodies were of superior quality and reliability for sample evaluation by immunostaining (both pMek and pErk are up-regulated in Alb-R26 Met tumors). Eighty-one HCC samples were positive for MCL1 (70%), 86 for BCL-XL (74%), and 62 for pERK (53%; Fig. 7B,C). When considering triplepositives for pERK 1 /BCL-XL 1 /MCL1 1 , we identified 38 HCC patients (33%; Fig. 7B,C). Next, we looked at the association of triple-positives with Ki67 index and the main recognized molecular categories of HCC (p53 1 , glutamine synthetase (GS)/b-catenin 1 , double-positive p53 and GS/b-catenin, and null; Supporting Fig. S8). For triple-positive cases, 42% are also positive for Ki67, although this correlation is not statistically significant. In addition, triple-positive cases had an association with poor HCC differentiation (grade 3-4 according to Edmondson; P 5 0.008). Moreover, there is a statistically significant positive correlation with p53 1 HCC and a negative correlation with null phenotype (Supporting Table S11). Remarkably, there is a striking correlation between pERK 1 / BCL-XL 1 /MCL1 1 triple cases with poor overall survival (hazard ratio, 2.12; 95% confidence interval, 1.10-4.07; P 5 0.023) and with disease-free survival (hazard ratio, 2.08; 95% confidence interval, 1.23-3.49; P 5 0.006; Fig. 7D,E). These results are supported by bioinformatics integrative analyses using outcomes from microarray studies revealing that 27% and 22% of patients (in GSE14323 and GSE14520, respectively) are MCL1 1 /BCL-XL 1 , and all of them (except one patient in GSE14323) are positive also for ERK1/ERK2 (Supporting Table S12). Thus, the pERK 1 /BCL-XL 1 /MCL1 1 signature characterizes a subgroup of HCC patients with poor prognosis.

Discussion
Modeling human tumors in mice in combination with genome-wide screening and bioinformatics enables the highly precise tracking of molecular mechanisms underlying tumorigenesis. Genetically modified mice carrying loss-of-function mutations in key tumor suppressors and/or overactivated forms of oncogenes mimic certain types of human tumors. However, such acute manipulations generally do not recapitulate the physiopathology observed in patients, in whom progressive, subtle alterations impact tumor initiation, maintenance, and evolution. Here, we show that nononcogenic RTKs are competent to trigger tumorigenesis when their expression is moderately enhanced above endogenous levels. However, vulnerability to enhanced RTK inputs is restrained to very specific cell types, as observed in Del-R26 Met mice. These results demonstrate the unique sensitivity of each tissue to subtly increased RTK levels, which discriminates resilience versus vulnerability during tissue homeostasis perturbation and cell transformation. Also, they show how slight changes in nononcogenic signaling inputs may have profound consequences in cells, which become pathological by triggering the full tumorigenic program in the liver. Based on our biochemical studies in hepatocytes, the Met receptor exists in a poised state: the strong phosphorylation status of the two tyrosine residues in the kinase domain indicates that Met is active, although not fully signaling-competent, as phosphorylation of the multifunctional docking site (21,22) is conditioned by HGF stimulation. This poised state of Met likely challenges basal signaling mechanisms in hepatocytes, generating instability over time as revealed by large cell dysplasia foci at early stages. The tumorigenic program in Alb-R26 Met mice likely results from the reprogramming of molecular settings triggered by enhanced RTK levels rather than from the acquisition of Met addiction. The competence of Met to reprogram signaling and biological events is consistent with its pleiotropic functions in developmental programs (15,22,(25)(26)(27)(28)(29) as well as in tumor evolution/resistance to therapies. (30)(31)(32) The implications for oncogenic Met in liver cancer are supported by reports showing: (1) Met activation in approximately 50% of HCC patients, correlating with poor prognosis, (33,34) although Met genetic mutations are rare (35) ; (2) a cell-cell crosstalk in which deregulated vascular endothelial growth factor A in HCC cells signals to HGF-producing macrophages to trigger tumor cell proliferation (36) ; (3) tumorigenesis in mice carrying either oncogenic Met forms or concomitant alterations of HGF/Met with other genes (19,(37)(38)(39)(40) ; (4) accelerated chemically induced liver neoplasia in Met mutants. (41) Compounds targeting Met are currently being explored in clinical trials. The 2-fold to 4-fold increase in Met levels in Alb-R26 Met mice is comparable to that in 20% of HCC patients. Therefore, the Alb-R26 Met genetic setting demonstrates the dramatic consequence of moderately increased levels of wild-type Met for triggering the tumorigenic program. Moreover, in the context of high HCC heterogeneity, (42) Alb-R26 Met mice exemplify HCC patients corresponding to the progenitor cell group of the proliferative subset, which is associated with poor outcomes. (20) The challenge to identify cancer cell vulnerabilities is evidenced by their resilience to the blockade of several operating signaling circuits and by their capacity to acquire drug resistance over time. These features are critical limits to the broadening of anticancer-targeted treatments. In most cases, the effectiveness of molecular therapies is conditioned by the identification of oncogenic alterations that contribute to the addiction of cancer cells. Exceptional cases of effectiveness are represented by the inhibition of BCR-ABL for the treatment of chronic myeloid leukemia, ERBB2 for breast cancer, ERBB1 for non-small-cell lung cancer, and B-RAF for metastatic melanoma. However, in most other types of cancer, the long-term efficiency of targeted treatment remains unsatisfactory for several reasons, including the underestimated relevance of certain protein functions compared to others, the existence of redundancy and crosstalk between pathways, and the cytostatic rather than cytotoxic effects elicited by drugs. Several treatments involving drug combinations lead effective responses in cancer cells. (43,44) A promising strategy to broaden the use of molecularly targeted therapies for cancer treatment is inspired by the concept of synthetic lethality, which reflects the deleterious effects of simultaneously targeting separate signals that are individually nonlethal on cells. The synthetic lethality approach may be particularly promising for the treatment of HCC, one of the most heterogeneous cancers characterized by poor clinical outcomes and a lack of effective therapies. In contrast to other carcinomas, in which tumor initiation and progression are triggered by mutations in a subset of oncogenes and/or tumor suppressors, a wide range of (epi)genetic mutations have been identified in HCC. How to translate this knowledge into therapeutic interventions remains unclear. The identification of functionally relevant targets in HCC is complicated because signals such as RTK pathways are rarely genetically mutated, although their activation is observed in a high proportion of HCC. (20) Mice carrying genetic alterations found in human HCC have been instrumental for validating the roles of certain signaling mechanisms in this disease. (45) However, agents targeting these pathways have been unsatisfactory in clinical trials due to their predominantly cytostatic rather than cytotoxic effects. The unbiased phosphokinome screen we performed in Alb-R26 Met mice highlighted vulnerable nodes and identified synthetic lethal interactions between Mek, Rsk, or Cdk1/2 and Bcl-XL inhibition in HCC. These combined treatments elicit strong synergistic effects, whereas single blockage induces minimal effects. Based on our mechanistic studies, combined Mek and Bcl-XL inhibition suppresses Erk and Mcl1 signaling, whereas individual targeting exerts different effects (Fig. 8). Bcl-XL inhibition alone primes cells toward cell death, as shown by slight Mcl1 down-regulation and moderate Cas-pase3 activation, although proliferating cells have an enhanced pErk. Instead, Mek inhibition alone impairs Erk signaling and affects cell cycle progression, without influencing cell death. Thus, the combined effects of Mek and Bcl-XL inhibition are effective in HCC cells through the triggering of cytotoxic effects. Our results are consistent with studies showing that antibodies targeting the calcium channel a2d1 subunit interfere with tumor-initiating properties through pErk and Bcl-2 down-regulation. (46) The functional relevance of the deregulated Ras pathway and Bcl-XL inputs in the context of Mcl1 overexpression in HCC biology is consistent with their concomitant alteration in a subgroup of HCC patients with poor prognosis. Thus, the synergistic effects of Mek and Bcl-XL inhibition previously observed in K-Ras mutant cancer cell lines (47) may represent a therapeutic strategy that is also effective for the treatment of HCC subgroups.
According to our studies, inhibition of the Ras pathway at distinct vertical levels does not exert equivalent effects, at least in terms of HCC tumorigenesis modeled by the Alb-R26 Met system. This is consistent with knowledge that the Ras pathway is not a simple linear cascade and that each component receives distinct regulatory inputs and is exposed to compensatory circuits. (48) Sorafenib, which targets Raf-1 and B-Raf as well as several RTKs such as vascular endothelial growth factor receptor, Kit, Flt3, and platelet-derived growth factor receptor, is a standard treatment for HCC. (49) In light of our findings, it may be relevant to assess the effectiveness of selumetinib or sorafenib treatment in combination with Bcl-XL inhibition in enriched trials.
The synthetic lethal interaction we identified reveals how the inhibition of multiple pathways with combined drug cocktails may be a successful strategy for HCC treatment and deserves further studies to optimize doses and to establish the potential (and limits) for the treatment of HCC subgroups. Drugs targeting vulnerable points emerging from the Alb-R26 Met cancer model are very promising anticancer agents for molecularly targeted therapies and should be actively explored in clinical trials. Therefore, our discoveries may widen their use as synergistic treatments for HCC patient subgroups. It will be essential to stratify HCC patients who may benefit from such synergistic treatments, although patient stratification is not a simple issue. Sensitivity versus resistance to this treatment is likely linked not only to the presence of the pErk 1 / Bcl-XL 1 /Mcl1 1 signature, found in Alb-R26 Met mice and in a proportion of HCC patients, but also to other signaling regulators of HCC biology. In conclusion, our studies illustrate how the use of cancer mouse models such as Alb-R26 Met mice, in which cell homeostasis is challenged by signaling fluctuations over time, uncovers "model-guided signatures" for patient stratification as a complementary strategy to broad sequencing screens in human cancer. In addition to recapitulating the molecular mechanisms underlying tumorigenesis and highlighting vulnerable signals, defined genetic settings may be valuable systems to explore signaling cooperation during the transition from healthy cells to transformation.