Targeting ABL1-mediated Oxidative Stress Adaptation in Fumarate Hydratase-Deficient Cancer

SUMMARY

Patients with germline fumarate hydratase (FH) mutation are predisposed to develop aggressive kidney cancer with few treatment options and poor therapeutic outcomes. Activity of the proto-oncogene ABL1 is upregulated in FH-deficient kidney tumors and drives a metabolic and survival signaling network necessary to cope with impaired mitochondrial function and abnormal accumulation of intracellular fumarate. Excess fumarate indirectly stimulates ABL1 activity while restoration of wild-type FH abrogates both ABL1 activation and the cytotoxicity caused by ABL1 inhibition or knockdown. ABL1 upregulates aerobic glycolysis via the mTOR/HIF1α pathway and neutralizes fumarate-induced proteotoxic stress by promoting nuclear localization of the anti-oxidant response transcription factor NRF2. Our findings identify ABL1 as a pharmacologically tractable therapeutic target in glycolytically dependent, oxidatively stressed tumors.

INTRODUCTION

Cancers that depend on altered metabolic programs and on up-regulated stress response pathways need to cope with the consequences of deregulated metabolism. Utilizing genetically defined cancers with well-characterized metabolic adaptations provides a unique opportunity to identify mechanism-based therapeutic interventions. Hereditary kidney cancers provide models that are particularly well-suited for this purpose, as two types of kidney cancer are characterized by mutation of the TCA cycle genes, fumarate hydratase (FH) and succinate dehydrogenase (SDH), respectively. Inactivation of either enzyme disrupts mitochondrial respiration and promotes dependence on aerobic glycolysis, a characteristic of aggressive kidney cancer (Linehan and Rouault, 2013; The Cancer Genome Atlas Research Network, 2013). Mutations in FH are found in the germline of patients with hereditary leiomyomatosis and renal cell carcinoma (HLRCC), a hereditary cancer syndrome in which affected individuals are at risk for developing cutaneous and uterine leiomyomas and a highly aggressive form of type 2 papillary kidney cancer (Grubb, III et al., 2007). No effective form of therapy is currently available for patients with advanced FH-deficient kidney cancer.

FH-deficient kidney cancers and their cell line models are highly glycolytic and display increased glucose dependence, lactate production, elevated levels of the hypoxia-stimulated transcription factor HIF1α and decreased activity of AMP-activated kinase (AMPK) (Yang et al., 2010; Tong et al., 2011; Yang et al., 2012). This metabolic adaptation is, at least in part, a direct consequence of the intracellular accumulation of the oncometabolite, fumarate (Isaacs et al., 2005; Frezza et al., 2011; Tong et al., 2011). By inhibiting HIF prolyl hydroxylase, fumarate stabilizes HIF1α which leads to the transcription of multiple genes, including those that encode for glucose transporters 1 and 4 (Glut1, Glut4), and vascular endothelial growth factor (VEGF).

Although HIF1α transcriptional activity and expression is increased in HLRCC tumors (Koivunen et al., 2007; Isaacs et al., 2005), their high energetic demands and increased glycolytic activity cause redox homeostasis to become unbalanced due to elevated production of reactive oxygen species (ROS) (Sudarshan et al., 2009; Sullivan et al., 2013). To survive this proteotoxic stress, FH-deficient kidney cancer cells utilize the oxidative branch of the pentose phosphate pathway to drive NADPH production and glutathione synthesis (Yang et al., 2013). Excess fumarate also stabilizes the master regulator of the antioxidant response, the transcription factor nuclear factor (erythroid-derived 2)-like 2 (NFE2L2 or NRF2) (Ooi et al., 2011; Adam et al., 2011). NRF2 activation in cancer has been reported to be either beneficial or detrimental, depending on the context and/or the tumor type (Sporn and Liby, 2012). In HLRCC tumors, NRF2 activation appears to be critical for tumor growth and survival (Frezza et al., 2011; Ooi et al., 2011; Adam et al., 2011).

Although FH mutation has been reported almost solely in HLRCC cancers, reduced FH activity has been found in other cancers, including clear cell kidney cancer (Sudarshan et al., 2011), and induction of pseudo-hypoxia and deregulated redox homeostasis are common features of many aggressive epithelial cancers (Denko, 2008; Cairns et al., 2011). While several studies have identified components of these pathways as potential molecular targets in FH-deficient tumors (Xie et al., 2009; Sourbier et al., 2010; Frezza et al., 2011), identification of a clinically tractable therapy for HLRCC cancer patients remains an unmet need. In the current study, we used an unbiased drug screen to identify therapeutic strategies targeting the deregulated metabolism and upregulated stress response pathways of HLRCC cancer.

RESULTS

Drug screening identifies vandetanib as highly cytotoxic for FH-deficient HLRCC kidney cancer cells

In order to uncover therapeutic strategies for patients with highly aggressive HLRCC-associated kidney cancer, we utilized the HLRCC-derived UOK262 cell line cultured in pyruvate-free media to screen a panel of 17 agents targeting diverse signaling pathways (Table S1). The tyrosine kinase inhibitor vandetanib (IC₅₀ = 16 nM) proved to be by far the most potent compound tested. Vandetanib displayed synthetic lethality for FH-deficient cells; stable reintroduction of wild-type FH into two independently derived HLRCC cell lines (UOK262 and UOK268) abrogated its cytotoxicity (Figure 1A).

Figure 1. Vandetanib inhibits ABL1 activity in vitro.

A. Cell viability following vandetanib treatment of FH-deficient cells (UOK262, UOK268) and their paired molecularly restored counterparts (UOK262WT and UOK268WT). B. Effect of vandetanib (5 nM) on UOK262 invasiveness monitored in real-time using the xCELLingence platform. C. HIF1α expression after vandetanib treatment or siRNA-mediated silencing of ABL1 (16 hr in each case) was assessed by ELISA in UOK262 and UOK268 cells (MSD Technology, 20 μg of protein/sample). D. Phosphorylation of mTOR and S6K is decreased following vandetanib treatment and siRNA-mediated silencing of ABL1 as visualized by immunoblotting. Phosphorylation of CRKII is used as a marker of ABL1 activity. Vandetanib (VAN) was used at 20 nM; imatinib (IMA) was used at 200 nM. E-F. ABL1 kinase activity was determined by in vitro assay using ABL1 immunoprecipitated for either UOK262 or UOK268 cells (E) or purified protein (F). Immunoprecipitated ABL1 or purified protein was incubated for 1 hr at 30°C with vandetanib (VAN; 20 nM) or imatinib (IMA; 200 nM) in presence of ATP (10 μM) and ABL1 substrate (see Experimental Procedures). ABL1 kinase activity was measured by quantifying substrate phosphorylation. G. Steady-state phosphorylation of ABL1 in cells following vandetanib treatment was visualized by immunoblotting. VAN, vandetanib; IMA, imatinib; * p<0.05. See also Figure S1 and Table S1. Data are displayed as the mean ± SD.

We reported previously that HIF1α expression is necessary to maintain the invasive phenotype of HLRCC cells (Tong et al., 2011). Vandetanib fully inhibited the invasiveness of UOK262 cells (Figure 1B, Figure S1A) and markedly decreased their HIF1α protein expression to levels comparable to those seen in wild-type FH-restored cells (Figure 1C), but its mechanism of action involved neither protein degradation nor modulation of HIF1α mRNA levels (Figure S1B-C). These data suggested that vandetanib may affect HIF1α translation, a process regulated by the protein kinase mTOR (Powis and Kirkpatrick, 2004). Indeed, we found that steady-state phosphorylation of mTOR and its downstream target S6 kinase, were markedly decreased following vandetanib treatment (Figure 1D).

Since ABL1 is reported to regulate mTOR (Cilloni and Saglio, 2012; Kim et al., 2005; Markova et al., 2010) and vandetanib potently inhibits ABL1 in vitro (Davis et al., 2011; Karaman et al., 2008), we asked whether vandetanib-mediated inhibition of the mTOR/HIF1α pathway might be phenocopied by ABL1 knockdown. Using UOK262 cells in which ABL1 was silenced with siRNA, we confirmed that reduced expression of ABL1, like vandetanib treatment, inhibited HIF1α expression and mTOR activity (Figure 1C-D, Figure S1D). Next, we confirmed that vandetanib and the ABL1 inhibitor imatinib inhibited the in vitro kinase activity of endogenous ABL1 that was immunopurified from both FH-deficient renal cell carcinoma cell lines (UOK262 and UOK268), as well as the activity of purified ABL1 protein (Figure 1E and 1F, respectively). These data were confirmed by assessing the phosphorylation status of CRKII, a substrate of ABL1 (Figure 1D), and by determining the phosphorylation status of ABL1 immunopurified from UOK262 and UOK268 cell lines previously treated for 4 hours with vandetanib (Figure 1G). Further, ABL1-specific siRNA reduced HIF1α expression in two HLRCC-derived cell lines to levels comparable to those achieved with vandetanib treatment or by restoration of wild-type FH (Figure 1C). Finally, nilotinib and ponatinib, two structurally distinct ABL1 inhibitors were equipotent with vandetanib in inhibiting the growth of UOK262 cells and their toxicity was markedly ameliorated by restoration of wild-type FH (Figure S1E). Notably, the growth inhibitory activity of these ABL1 inhibitors was reduced in medium containing pyruvate (Figure S1F). Finally, using an in vitro kinase assay, we found that these three TKIs displayed similar potency in inhibiting the activity of purified ABL1 (vandetanib IC₅₀ = 15 nM, nilotinib IC₅₀ = 4 nM, ponatinib IC₅₀ = 17 nM; Figure S1G), consistent with our hypothesis that ABL1 may be an important target in FH-deficient tumor cells. However, we cannot exclude the possibility that an off-target effect of vandetanib may fortuitously contribute to its activity in this tumor model.

ABL1: a therapeutic target in FH-deficient tumor cells

The Abelson (ABL) family of nonreceptor tyrosine kinases, comprised of ABL1 and ABL2, affects diverse signaling pathways involved in cell growth, invasion and migration. Activation of ABL1 and ABL2 have been detected in a number of different cancers (Greuber et al., 2013). Although the role of oncogenic BCR-ABL fusion protein has been extensively studied in hematologic malignancies (Cilloni and Saglio, 2012), much less is known about the physiological and/or pathological role of wild-type ABL1. Based on the preceding data, we immunoprecipitated ABL1 from protein lysates of 5 HLRCC tumors surgically removed from 3 patients, as well as from lysates of normal tissues, and we compared ABL1 phosphorylation normalized to total immunoprecipitated ABL1 protein as an indicator of its activation state (Figure 2A). The ABL1 activation state in 5 of 5 tumor specimens was significantly greater than that in the normal tissues. We performed a similar comparison in 2 HLRCC-derived cell lines and in their wild-type FH-restored counterparts (Figure 2B). Constitutive ABL1 phosphorylation, while clearly evident in both HLRCC-derived cell lines, was markedly reduced upon restoration of wild-type FH (total and phospho-ABL1 levels in HEK293 kidney cells are included for comparison), although this did not affect the expression of total ABL1 protein. We also assessed the phosphorylation status of ABL1 in skin and uterine leiomyomas, two classical manifestations of HLRCC disease, compared with their normal counterparts (Figure S2A-B). ABL1 phosphorylation was elevated in skin leiomyomas but not in uterine leiomyomas, suggesting that treatment of these manifestations might require alternative targeting approaches.

Figure 2. Anti-tumor activity of vandetanib is ABL1-dependent.

A. Assessment of ABL1 activation status in HLRCC tumor specimens was performed by immunoblotting for phosphorylated ABL1 following immunoprecipitation of ABL1 protein from tissue lysates. The intensity ratio of phosphorylated ABL1 to total ABL1 was assessed by densitometric analysis. B. Assessment of ABL1 activation status in a panel of cell lines was performed as in (A). C. Rescue with various infected murine ABL proteins was performed 72 hr after lentiviral infection of UOK262 cells with ABL-targeted miRNA (miABL). Scrambled miRNA (miSCR) was used as an miRNA control and empty pBABE vector (BABE) served as the control for infection with the various murine ABL contructs. Murine ABL constructs included wild type (WT), kinase-dead (K290M), kinase inhibitor-resistant (T315I), and constitutively active ABL (P131I). Cell viability in presence and absence of vandetanib was assessed (24 hr, 50 nM; VAN: vandetanib). D. Immunoblot showing reduction in ABL1 protein expression 72 hr post miABL infection, and subsequent expression level of re-introduced murine ABL proteins. mTOR phosphorylation status is correlated with expression of competent ABL1 protein. E. Silencing of ABL1 with siRNA is cytotoxic for FH-deficient cells. F. Importance of ABL1 for soft agar colony growth of FH-deficient cells was assessed by measuring the number of colonies visible 4 weeks after seeding with cells treated as shown. VAN, vandetanib; IMA, imatinib; * p<0.05, ** p<0.001. Data are displayed as the mean ± SD. See also Figure S2.

Given these findings, we investigated next whether the cytotoxicity of vandetanib in FH-deficient tumor cells was ABL1-dependent. Classical rescue experiments were performed following ABL1 knockdown in UOK262 cells by lentiviral infection of microRNA (miRNA) targeting human ABL1 (miABL1; a scrambled miRNA, miSCR, was used as an miRNA control) (Yogalingam and Pendergast, 2008). Various lentiviral murine ABL constructs, not recognized by miABL1, were then infected and their ability to rescue endogenous ABL1 knockdown or its pharmacologic inhibition was examined. These constructs included wild-type ABL1 (WT), kinase-dead ABL1 (K290M), tyrosine kinase inhibitor (TKI) resistant ABL1 (T315I) and constitutively active ABL1 (P131L) (Figure 2C-D) (Smith-Pearson et al., 2010; Yogalingam and Pendergast, 2008). Infection with empty pBABE plasmid (‘BABE’) served as a control. Using purified proteins in an in vitro kinase assay, we confirmed that ABL1-T315I was resistant to vandetanib (Figure S2C). Knockdown of endogenous ABL1 upon infection with miRNA significantly decreased cell viability (approximately 50%) at 24 hr, while simultaneous expression of WT, T315I and P131L murine ABL proteins restored cell viability to control (or miSCR-infected) levels (Figure 2C). In contrast, kinase-dead ABL1 expression did not restore cell viability, supporting the ABL dependence of this phenotype. Importantly, vandetanib did not cause additional toxicity in miABL1-treated cells expressing either empty plasmid, kinase-dead ABL1, TKI-resistant ABL1, or constitutively active ABL1. However, vandetanib produced significant cytotoxicity in cells infected with miSCR, and in cells infected with both miABL1 and wild-type murine ABL1. Consistent with these data, the silencing of endogenous ABL1 in UOK262 cells with two distinct siRNAs dramatically decreased cell viability in the absence of vandetanib (Figure 2E). Finally, like vandetanib cytotoxicity, the cytotoxicity incurred by ABL1 knockdown was abrogated by restoration of wild-type FH in UOK262 cells (Figure S2D-E). In both UOK262 and HEK293 cells, ABL1 knockdown upon miABL1 infection markedly reduced mTOR phosphorylation. Re-expression of wild-type, TKI-resistant or constitutively active ABL1, but not kinase-dead ABL1 was able to rescue this phenotype (Figure 2D and Figure S2F). Importantly, the infected murine ABL1 proteins were expressed at levels similar to endogenous ABL1 in both cell lines.

To assess the importance of ABL1 in maintaining the tumorigenic phenotype of FH-deficient cells, we measured their anchorage-independent growth and invasiveness after ABL1 inhibition or silencing. These data show that ABL1 expression and activity are necessary to sustain the clonogenicity and invasive potential of FH-deficient tumor cells (Figure 2F, Figure S1A).

ABL1 promotes aerobic glycolysis

Based on our data linking ABL1 with HIF1α expression (Figure 1C), we investigated whether ABL1 inhibition affected the aerobic glycolysis on which FH-deficient cells depend for generation of ATP and cellular building blocks (Yang et al., 2010; Tong et al., 2011). First, we found that vandetanib significantly decreased expression of the glucose transporters Glut1 and Glut4 in UOK262 cells (Figure 3A), concomitant with decreased glucose uptake and lactate secretion (Figure 3B and 3C, respectively). Transient silencing of ABL1 or HIF1α decreased glucose uptake to a similar degree (Figure 3B, Figure S3A). To confirm that ABL1 promotes aerobic glycolysis, we assessed the impact of ABL1 modulation on the extracellular acidification rate (ECAR, a surrogate for lactate secretion). In agreement with our earlier results, silencing or pharmacologic inhibition of ABL1 significantly decreased ECAR in UOK262 cells (Figure 3D). Consistent with a relationship between ABL1 activity and glycolysis, vandetanib concomitantly inhibited both ABL1 activity (steady-state ABL1 phosphorylation) and glucose uptake in UOK150, a glycolytic clear cell kidney cancer cell line established from a non-HLRCC kidney cancer harboring an inactivating VHL mutation (Sourbier et al., 2012) (Figure S3B-C).

Figure 3. ABL1 supports aerobic glycolysis in FH-deficient tumor cells.

A. Expression of the glucose transporters Glut1 and Glut4 following vandetanib treatment (24 hr) in UOK262 cells. B. Glucose uptake was measured in UOK262 cells using the non-degradable fluorescent glucose analog 2-NBDG (20 μM) after vandetanib (VAN; 20 nM, 16 hr) or imatinib (IMA; 200 nM, 16 hr) treatment, or after silencing either ABL1 or HIF1α. C. Lactate secretion after vandetanib treatment (20 nM, 16 hr). D. The effect of ABL1 silencing or inhibition on aerobic glycolysis in UOK262 cells was measured by recording the extracellular acidification rate (ECAR, a surrogate for lactate secretion) in a Seahorse Bioanalyzer. VAN, vandetanib; IMA, imatinib; * p<0.05. Data are displayed as the mean ± SD. See also Figure S3.

Vandetanib inhibits ABL1, glycolysis and HLRCC tumor growth in vivo

In order to visualize the dynamic impact of vandetanib on HLRCC metabolism in vivo, we used magnetic resonance spectroscopic imaging (MRSI) to monitor the metabolism of intravenously injected hyperpolarized [1-¹³C] pyruvate, as previously described (Golman et al., 2006; Chen et al., 2007). MRSI detects increased conversion of pyruvate to lactate in tumors, most notably in those characterized by a high rate of glycolysis and impaired oxidative phosphorylation. As shown in Figure 4A, in vivo imaging of untreated UOK262 xenografts (day 0) revealed a high lactate/pyruvate ratio, whereas vandetanib treatment (day 2) significantly suppressed the conversion of pyruvate to lactate in tumors without affecting tumor perfusion (Figure 4A-B), suggesting that, in agreement with ourin vitro data, vandetanib inhibits tumor glycolysis in vivo.

Figure 4. Vandetanib inhibits HLRCC tumor growth, ABL1 and glycolysis in vivo.

A. ¹³C-hyperpolarized imaging of mice harboring UOK262 xenografts before and 2 days after treatment with vandetanib. The images on the left display representative NMR images, the right set of images display representative images of the lactate/pyruvate ratio. B. Quantification of lactate/pyruvate ratios from data obtained in (A). Averaged measurements obtained from 5 animals in each group were used. C. Growth of UOK262 xenografts treated once weekly with vehicle (DMSO/PBS) or vandetanib (100 mg/kg). Right panel: Averaged tumor volume of 2 independent experiments taken after 7 weeks of treatment. D. ABL1 kinase activity was measured by kinase assay after immunopurification of ABL1 protein from tumor tissues excised from vehicle-treated and vandetanib-treated mice 2 days post-treatment. ** p<0.01; * p<0.05; VAN, vandetanib. Data are displayed as the mean ± SD. See also Figure S4.

Next, we assessed the antitumor efficacy of vandetanib in the same HLRCC murine xenograft model. As shown in Figure 4C, vandetanib (100 mg/kg, once weekly, i.p.) caused marked tumor regression in 80% (12 of 15, two independent experiments) of treated mice. When we interrogated the molecular effects of vandetanib in vivo, we found that the activation status of ABL1 was significantly reduced after 2 days in tumors excised from vandetanib-treated mice compared with vehicle-treated animals (Figure 4D), as well as after 10 weeks of treatment (Figure S4B). Also, similar to earlier in vitro observations, activity of the mTOR/HIF1α pathway in these tumors (including HIF1α expression, Glut1 expression, VEGF-A level in tumor tissue and in plasma, and phospho-S6K) was decreased in vandetanib-treated mice after 2 days (Figure S4A) as well as after 10 weeks of treatment (Figure S4C-F). After 10 weeks of treatment, tumors from vandetanib-treated mice were also more apoptotic and displayed less activation of EGFR family and VEGF receptor kinases compared with tumors excised from vehicle-treated animals (Figure S4G-H).

Fumarate-mediated ABL1 activation upregulates NRF2 nuclear localization and transcriptional activity in vitro and in vivo

The constitutively elevated level of phosphorylated ABL1 in HLRCC cell lines is reversed by restoration of wild-type FH expression (see Figure 2B), and the in vitro kinase activity of ABL1 immunopurified from HLRCC cells is significantly greater than that of the kinase immunopurified from FH-restored HLRCC cells (Figure 5A). Therefore, we examined whether intracellular accumulation of fumarate influenced ABL1 kinase activity, either directly or indirectly. We observed that treatment of cells expressing functional FH protein (HEK293 or UOK262WT) with a cell-permeable form of fumarate (dimethylfumarate, DMF) resulted in increased ABL1 phosphorylation (Figure 5B). However, neither fumarate nor DMF were able to directly stimulate ABL1 in vitro, suggesting that fumarate-mediated ABL1 activation in cells is indirect (Figure S5A). Since ABL1 is activated by oxidative stress and excess intracellular fumarate increases ROS (Sudarshan et al., 2009; Sullivan et al., 2013), fumarate-induced ABL1 activation may be ROS-dependent. Consistent with this model, treatment of UOK262 cells with the ROS scavenger N-acetylcysteine (NAC) concurrently reduced both steady-state ABL1 phosphorylation and cellular ROS level (Figure S5B-C). However, we cannot exclude the possibility that additional factors may contribute to the increased activation state of ABL1 in these tumor cells.

Figure 5. ABL1 coordinates NRF2 antioxidant response in vitro.

A. Basal in vitro ABL1 kinase activity in UOK262WT and UOK262 cells (normalized to total protein). B. Effect of intracellular fumarate accumulation on ABL1 activity in HEK293 and UOK262 cells (DMF, dimethylfumarate; Fum, fumarate; 0.25 mM, 4 hr). C-E. The impact of vandetanib on NRF2 nuclear/cytoplasmic distribution was assessed by immunoblotting of both nuclear and cytoplasmic extracts (C) and by confocal immunofluorescence microscopy (D-E). Original magnification x63; bar scale 25μm. F. NRF2 transcriptional activity was assessed in HEK293 cells following treatment with vandetanib (VAN, 50 nM, 4 hr) and/or DMF (*: significance compared to CTL; #: significance compared to DMF-induced NRF2 transcriptional activity) using a NRF2-responsive luciferase reporter assay and normalized to co-transfected Renilla luciferase. G. The role of ABL1 in regulating NRF2 transcriptional activity was assessed as in (F) following transient silencing of ABL1 using siRNA. H. NQO1 expression in UOK262 cells after vandetanib or ABL1 silencing. VAN, vandetanib; IMA, imatinib; ^#, * p<0.05. Data are displayed as the mean ± SD. See also Figure S5.

Accumulation of intracellular fumarate has been shown to stabilize expression of the antioxidant response transcription factor NRF2 via inactivation of its endogenous inhibitor KEAP1 (Adam et al., 2011; Ooi et al., 2011). Since the Src kinase family is reported to modulate NRF2 nuclear translocation (Jain and Jaiswal, 2006; Shelton and Jaiswal, 2013), we examined whether vandetanib affects this process in FH-deficient tumor cells. By evaluating nuclear and cytoplasmic fractions prepared from drug-treated and control cells for NRF2 content, we observed that vandetanib caused a time-dependent redistribution of NRF2 from nucleus to cytosol that was complete within 6 hours (Figure 5C). These data were further confirmed by confocal immunofluorescence microscopy performed 6 hours after vandetanib (20 nM) treatment (Figure 5D-E). Redistribution of NRF2 from nucleus to cytosol should result in decreased transcriptional activity. Using a luciferase-based NRF2 reporter assay, we found that DMF stimulated, while vandetanib inhibited NRF2 transcriptional activity (Figure 5F). Importantly, NAC treatment that reduced ROS level and ABL1 activity (see Figure S5B, C) also caused a significant reduction in DMF-stimulated NRF2 transcriptional activity (Figure S5D). Using the same assay we confirmed that silencing of ABL1 abrogated DMF-induced NRF2 transcriptional activity (Figure 5G). Taken together, these data support a role for ABL1 in supporting NRF2-mediated transcription in response to elevated intracellular fumarate levels, and for ROS in mediating the fumarate-dependent increase in NRF2 activity.

HLRCC tumor cells constitutively express high levels of ROS, and ROS-inducing agents are particularly cytotoxic to these cells (Sourbier et al., 2010). Because NRF2 plays a critical role in cellular defense against ROS (Sporn and Liby, 2012), we tested whether the vandetanib sensitivity of these cells might be affected by cellular ROS level. We pretreated UOK262 cells with NAC for 2 hr prior to treatment with vandetanib for an additional 16 hr. NAC pre-treatment, did not itself affect cell viability, but it significantly protected the cells from vandetanib toxicity (Figure S5E). Notably, the growth inhibitory activity of ABL1-targeting TKIs is also markedly compromised by cell culture in pyruvate-containing medium (Figure S1F), and several reports identify pyruvate as a ROS scavenger (Sudarshan et al., 2009; Babich et al., 2009). Consistent with these findings, cell culture in pyruvate-containing medium significantly reduced DMF-induced NRF2 activity (data not shown).

Since ABL1 also promotes HIF1α-mediated glycolysis (vide supra), we asked whether inhibition of HIF1α per se contributed to cell death in this model. First, based on data from our cell screen, both LY294002 and rapamycin were significantly less effective in inhibiting cell growth or causing cytotoxicity than was vandetanib (Table S1). Further, NAC reduced HIF1α expression as effectively as did vandetanib, but with no toxicity, indicating that inhibition of NRF2 without relief from the underlying oxidative stress is at least a necessary, if not sufficient determinant of vandetanib toxicity (Figure S5E-F). This possibility is consistent with our earlier findings that DMF-stimulated NRF2 activity is at least partially ROS-dependent. In support of this hypothesis, both ABL1 inhibition and silencing decreased expression of the endogenous NRF2 transcriptional target NQO1 (Figure 5H), which is strongly upregulated in HLRCC tumor tissue (Figure S5H; (Adam et al., 2011)). Finally, silencing of NRF2 with two independent siRNAs caused significant cytotoxicity in UOK262 cells, a phenotype that was reversed in UOK262WT (Figure S5G).

To assess whether these findings are recapitulated in vivo, we examined the effect of vandetanib on NQO1 expression in HLRCC xenografts. Its expression was significantly decreased in UOK262 xenografts excised from mice treated with vandetanib (Figure S5I; tissues from the animal study described in Figure 4).

AMPK activation inhibits NRF2 activity additively with vandetanib to inhibit FH −/− RCC tumor growth

SIRT1-mediated deacetylation of NRF2 is reported to inhibit its transcriptional activity (Kawai et al., 2011). Since AMPK positively regulates SIRT1 and is constitutively hypoactivated in FH-deficient tumors (Tong et al., 2011; Fulco and Sartorelli, 2008), we asked whether pharmacologic activation of AMPK in these cells might enhance SIRT1 deacetylase activity to inhibit NRF2 independently from vandetanib. First, we assessed the effect of AMPK modulation on NRF2 transcriptional activity using a luciferase-reporter assay. The AMPK activators metformin and AICAR decreased DMF-stimulated NRF2 activity in HEK293 and UOK262 cells, while knockdown/inhibition of SIRT1 with siRNA or nicotinamide, respectively, had the opposite effect (although the increased NRF2 activity remained sensitive to vandetanib, Figure 6A, Figure S6A-B). Consistent with these data, metformin treatment decreased NRF2 acetylation in UOK262 cells while silencing SIRT1 had the opposite effect (Figure 6B).

Figure 6. Metformin sensitizes HLRCC xenografts to vandetanib.

A. NRF2 transcriptional activity was assessed in HEK293 cells following treatment with the AMPK activators metformin (METF; 1 mM, 6 hr), and AICAR (50 μM, 6 hr) in the presence and absence of DMF (0.25 mM). B. Metformin and SIRT1 knockdown exert opposing effects on NRF2 acetylation status as visualized by immunoblotting with an antibody recognizing acetylated lysine (“AcK”) following immunoprecipitation of NRF2. C. Effect of vandetanib (5 nM) and metformin (5 μM) treatment, singly or together, on NRF2 transcriptional activity. D. Effect of combining metformin (METF, 5 μM) and vandetanib (VAN, 5 nM) on UOK262 viability after 24 hr. E. Effect of metformin and vandetanib alone or in combination on UOK262 xenograft growth in mice. This study used 1/10^th of the vandetanib dose used in the experiment shown in Figure 4. F. Averaged data from two independent experiments identical to the experiment shown in (E) (8 animals per group). G. Tumor regressions achieved with vandetanib/metformin 8-week treatment regimen are durable. Graph shows percent survival over 13 months of mice treated for 8 weeks with the vandetanib/metformin combination as shown in (E) and (F), compared to vehicle-treated animals. ^ø denotes a mouse that died following pelvic prolapse. No tumor was found at necropsy. VAN, vandetanib; DMF, dimethylfumarate; Fum, fumarate; ^#,* p<0.05. Data are displayed as the mean ± SD. See also Figure S6.

Next, we examined whether combining metformin and vandetanib caused additive inhibition of NRF2. Data obtained using a luciferase-reporter in HEK293 cells and by monitoring endogenous NQO1 expression in UOK262 cells suggested that this is the case (Figure 6C and Figure S6C). Supporting these observations, we found that metformin, at a concentration lacking single agent toxicity, significantly decreased the viability of UOK262 cells when combined with a sub-optimal concentration of vandetanib (Figure 6D). Since both vandetanib and metformin have multiple effects in cells, we confirmed that their additive impact on NRF2 was due to ABL1 inhibition combined with AMPK activation. Thus, both metformin and AICAR cooperated with knockdown of ABL1 to markedly reduce cell viability, while, in the absence of ABL1 silencing, neither drug was effective. Further, nilotinib displayed greater cytotoxicity at lower concentrations in combination with metformin compared with its single agent activity (Figure S6D-F). Finally, in vivo analysis confirmed that low dose vandetanib (10 mg/kg, once weekly; 1/10th of the dose used in the experiment shown in Figure 4) combined with metformin (5 mg/kg, once weekly) induced complete tumor regression within 8 weeks of treatment initiation in 100% of treated mice (Figures 6E-F, 12 of 12 mice from 2 independent experiments). No tumor regressions were seen in mice treated with the single agents at these doses. Nine of ten mice treated with the vandetanib/metformin combination remained tumor-free for more than 13 months after treatment cessation, while vehicle-treated mice all had to be sacrificed by 8 weeks due to tumor size (Figure 6G).

DISCUSSION

Loss of FH activity leads to fumarate accumulation that sustains both aerobic glycolysis and NRF2-mediated antioxidant responses (Isaacs et al., 2005; Ooi et al., 2011; Adam et al., 2011). Although both pathways are critical for HLRCC cancer growth in vivo, a strategy to simultaneously target these processes has eluded investigators. Here, we show that ABL1, a non-receptor tyrosine kinase involved in numerous biological processes including cell proliferation, cell migration, metabolism and apoptosis (Greuber et al., 2013), is indirectly activated by increased fumarate, likely as a consequence of elevated ROS levels, and supports both aerobic glycolysis and NRF2 activity in HLRCC tumor cells (Figure 7). Using an unbiased screen, we identified the tyrosine kinase inhibitor vandetanib as having particular potency in an HLRCC-derived cell line in vitro and we found that vandetanib efficacy was compromised by restoration of wild-type FH activity in these cells. Although vandetanib inhibits several other kinases in addition to ABL1, we found that ABL1 knockdown using two different siRNAs as well as a human ABL1-targeted miRNA mimicked vandetanib cytotoxicity, as did several other TKIs that share ABL1 inhibitory activity with vandetanib. In addition, re-expression of active murine ABL1 rescued cells from these effects, and re-expression of TKI-resistant ABL1 abrogated vandetanib toxicity. Taken together, these data suggest that ABL1 inhibition accounts for much of the vandetanib effect in HLRCC cells, although we cannot exclude the possibility that an off-target effect contributes to the potency of vandetanib in this tumor model.

Figure 7. ABL1 is a key regulator of both aerobic glycolysis and the antioxidant stress response in HLRCC tumors.

The model depicts a proposed dual role for ABL1 in supporting FH-deficient tumor cell viability and energy metabolism. By stimulating mTOR activity to increase HIF1α translation, ABL1 promotes aerobic glycolysis, thus enabling the tumor cells to meet their energetic needs. Simultaneously, ABL1 also facilitates NRF2 nuclear localization, thus supporting its transcriptional activity, which allows tumor cells to buffer the high oxidative stress that is a consequence of excess fumarate accumulation. See also Figure S7.

A common characteristic of advanced cancer is enhanced dependence on glucose and on glycolytic metabolism (Vander Heiden et al., 2009). In this regard, FH-deficient tumor cells, with their disrupted TCA cycle, are examples of the Warburg effect and represent a human-derived, genetically defined tumor model with which to evaluate therapeutic strategies designed to interdict these pathways. Although targeting aerobic glycolysis in cancer has yielded promising results in vitro, in vivo efficacy has been limited, often due to systemic toxicities. Our current findings reveal a potential role for ABL1 in modulating mTOR-dependent HIF1α expression and demonstrate that both vandetanib and ABL1 knockdown phenocopy several consequences of HIF1α silencing in HLRCC cells, including reduced HIF1α protein expression and transcriptional activity, reduced aerobic glycolysis (in vitro and in vivo) and reduced cell invasiveness (Tong et al., 2011).

Given these data it is perhaps surprising that LY294002 and rapamycin were not more active in our in vitro growth inhibition/cytotoxicity screen (see Table S1). However, HIF modulation alone would not be expected to fully correct either the underlying defect in fumarate metabolism or the resultant accompanying oxidative stress, phenomena identified in our current study as key contributors to the cytotoxicity of both vandetanib and ABL1 knockdown. We previously reported that HLRCC-derived cell lines express abundant HIF1α as a consequence of both direct fumarate-mediated inhibition of HIF prolyl hydroxylase and indirect effects of elevated ROS levels (Isaacs et al., 2005; Klimova and Chandel, 2008; Sudarshan et al, 2009; Tong et al., 2011). Highly glycolytic cancers, including HLRCC-associated kidney cancer, generate excessive ROS and depend on an efficient oxidative stress response transcriptional program, suggesting that disruption or overwhelming of this machinery may provide an alternative treatment strategy (Sourbier et al., 2010). NRF2 is the master transcriptional regulator of the anti-oxidant response. Under non-stress conditions, it is sequestered in the cytoplasm by the adaptor protein, KEAP1, which also facilitates NRF2 ubiquitination and its subsequent proteasome-mediated degradation. In the presence of oxidative stress, cysteine residues in KEAP1 are oxidized to inactivate the protein, promoting stabilization of NRF2 in the cytoplasm. In HLRCC cells, elevated intracellular fumarate inactivates KEAP1 by covalently modifying its cysteine residues in a process termed succination, resulting in constitutive stabilization of NRF2 (Adam et al., 2011; Ooi et al., 2011). NRF2 is also regulated by the protein deacetylase SIRT1, and Ooi and colleagues recently identified SIRT1 mutations as contributors to NRF2 activation in papillary type II renal cancer, a sporadic cancer that is histologically similar to HRLCC (Ooi et al., 2013), suggesting that stabilization and activation of NRF2-driven anti-oxidant responses may be important factors in kidney cancers.

ABL1 is an important, SIRT1-independent determinant of NRF2 nuclear localization and transcriptional activity both in vitro and in vivo (Figure 7). Our data are consistent with previous studies in other tumor models reporting that ABL1 is activated by oxidative stress and DNA damage, and affects the NRF2-dependent antioxidant response (Sun et al., 2000; Greuber et al., 2013; Li et al., 2004). Since aberrant activation of NRF2 signaling occurs in multiple tumor types (Figure S7A), it is reasonable to speculate that targeting ABL1 might be a useful therapeutic strategy for such tumors in addition to HLRCC. Preliminary data obtained using a panel of lung cancer cell lines support this hypothesis (Figure S7B-C), and suggest that expression of NQO1 may be a predictive biomarker of response.

Finally, we found that upregulation of AMPK (constitutively hypoactive in HLRCC) (Tong et al., 2011), either as a consequence of mitochondrial complex I inhibition with metformin or following treatment with the AMPK activator AICAR, was additive with ABL1 inhibition in maximally inhibiting NRF2 activity in vitro. The sensitivity of NRF2 to AMPK activation is likely a consequence of AMPK-dependent SIRT1 activation, since metformin decreased nicotinamide-sensitive NRF2 acetylation. Notably, in an HLRCC xenograft model, inclusion of metformin (at a clinically achievable dose) reduced the effective vandetanib concentration by 90% (compared to single agent vandetanib administration). Weekly administration of this drug combination for 8 weeks caused complete tumor regression in 100% of treated mice and extended tumor-free survival for more than one year after cessation of treatment.

In summary, our data show that ABL1 is a key, pharmacologically tractable modulator of the cellular stress response to excess fumarate accumulation. Since ABL1 also supports HIF-dependent glycolytic metabolism, inhibiting this kinase provides a clinically viable strategy to simultaneously interfere with aerobic glycolysis and the anti-oxidant response pathway, upon which highly glycolytic, oxidatively stressed tumors depend for survival.

EXPERIMENTAL PROCEDURES

Cell lines and cell culture

UOK262, UOK268, UOK262WT, UOK268WT, and UOK150 cell lines were established in the Urologic Oncology Branch from surgically resected tumor specimens (National Cancer Institute, Bethesda, MD) (Yang et al., 2010; Anglard et al., 1992; Yang et al., 2012). All other cell lines were purchased from ATCC (Manassas, VA). Cells were cultured in high glucose DMEM without pyruvate supplemented with 10% FBS. The cells were harvested or treated when they reached 70-80% confluence.

Chemical agents

Vandetanib was generously provided by Astra Zeneca (Wilmington, DE). PTHrP-neutralizing antibody was from Bachem (Torrance, CA). All other compounds used were from Sigma-Aldrich (St. Louis, MO) or Selleck Chemicals (Houston, TX).

Human Tissues

All human tissues were obtained with informed consent from patients enrolled on a Urologic Oncology Branch Clinical Protocol approved by the Institutional Review Board of the National Cancer Institute, National Institutes of Health.

ABL1 kinase activity assay

Five nanograms of purified ABL1 protein (Life Technologies, Carlsbad, CA) and the peptide substrate paxillin (Santa-Cruz Biotechnology, Dallas, TX) were incubated for 1 hr at 30°C with 10 μM ATP in kinase assay buffer. The reaction was stopped by adding denaturing sample buffer and phosphorylation of the recognition motif I/V/L-Y-X-X-P/F was assessed by immunoblot analysis. Alternatively, the reaction was stopped by adding ADP-Glo™ reagent (ADPGlo™ kinase assay kit, Promega, Madison, WI) and kinase activity was assessed following the manufacturer's instructions.

ABL1 constructs and infection

Murine ABL1 constructs and miRNA targeting human ABL1 (not recognizing murine ABL) were introduced into HEK293 and UOK262 cells following a previously described protocol (Smith-Pearson et al., 2010; Yogalingam and Pendergast, 2008) with the following modifications. Briefly, lentiviruses were generated in HEK293T cells as previously described (Smith-Pearson et al., 2010; Yogalingam and Pendergast, 2008). Supernatant containing the viral particles was collected 48 hr post-transfection, aliquoted and stored at -80^oC until usage. Five thousand UOK262 or HEK293 cells were plated in blackview 96-well plates and 50,000 cells were seeded in 6-well plates. The former were used for viability determination and the later were used for Western blotting (see text). The day after plating, cells were infected with the different viruses using 1:30 v/v miRNA/media and 1:30 v/v vector/media per well in the presence of polybrene (8 μg/mL). The media were changed 24 hr post-infection. Forty-eight hours post-infection, cells were treated as indicated in the figures. Seventy-two hours post-infection, protein lysates were harvested for Western blotting or cell viability assays were performed.

Animal study

All animal experiments were performed on approved protocols and in accordance with the guidelines of the Animal Care and Use Committee of the National Institutes of Health.

¹³C MRI of hyperpolarized ¹³C-labeled pyruvate metabolism

Samples of [1-¹³C] pyruvic acid (30 μL) containing 15 mM triarylmethyl radical (TAM) and 2.5 mM gadolium chelate ProHance (Bracco Diagnostics, Milano, Italy) were polarized at 3.35 T and 1.4 K in a Hypersense DNP Polarizer (Oxford Instruments), according to the manufacturer's instructions. After 40-60 min, the hyperpolarized sample was rapidly dissolved in 4.5 mL of a superheated alkaline buffer comprising 40 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 30 mM NaCl and 100 mg/L ethylendiaminetetraacetic acid (EDTA). NaOH was added to the dissolution buffer to adjust pH to 7.4 after mixture with [1-¹³C] pyruvic acid. Hyperpolarized [1-¹³C] pyruvate solution (12 μL/g body weight) was intravenously injected through a catheter placed in the tail vein of the mouse. Hyperpolarized ¹³C MRI studies were performed on a 7 T scanner (Bruker Bio-Spin MRI GmbH) using a 17 mm home-built ¹³C solenoid coil placed inside of a saddle coil for ¹H. The ¹³C two-dimensional spectroscopic images were acquired 30 seconds after the start of pyruvate injection from a 28 × 28 mm field of view in a 8 mm coronal slice through the tumor, with matrix size of 16 × 16, spectral width of 8 kHz, repetition time (TR) 78 ms, 0.2 ms Gaussian excitation pulse with a flip angle of 10°. The total time required to acquire an image was 20 seconds.

ATP assay

ATP levels were determined using the ATPLite assay (PerkinElmer, Shelton, Connecticut), following the manufacturer's protocol.

Immunofluorescence

Immunofluorescence was performed as previously described (Sourbier et al., 2013). Briefly, 5000 UOK262 cells were plated in 2-well chamber-slides (Nunc., Sigma-Aldrich) and treated for 6 hr with vandetanib (20 nM) before fixation with 4% paraformaldehyde. Cells were blocked for 1 hr with BSA (3%) and permeabilized with Triton (0.5%). Mouse anti-NRF2 antibody (dilution 1:100; dilution; Abcam # 62352) was added and chamber slides were incubated overnight at 4°C in a humidified atmosphere. After 3 washes with TBST buffer, slides were incubated 1 hr with secondary antibody coupled to Alexa455 (dilution 1:1000), washed and mounted. Nuclei were stained with DAPI (Cell Signaling Technology). Images were captured with a confocal microscope (Zeiss NLO510). Nuclear localization was quantified by measuring the corrected total cell fluorescence (CTCF) in the nuclei of the cells. Briefly, the intensity of 5 nuclei from 10 different fields per condition were measured using Image J. CTCF was then calculated with the following formula: CTCF = Integrated Density – (Area of selected cell × Mean fluorescence of background readings).

NRF2 reporter assay

NRF2 transcriptional activity was measured using the pGL4.37[luc2/ARE/Hygro] vector (Promega, Madison, WI), following the manufacturer's protocol. Briefly, 5000 HEK293 cells were plated in white-clear view plates (Perkin-Elmer) and were transfected the following day with 100 ng of DNA per well using XTreme Gene HP transfection reagent (Roche Applied Science, Indianapolis, IN). After 24 hr, cells were treated as indicated in Figure Legends. Luminescence was measured as an indicator of NRF2 transcriptional activity using the Dual-Glo Luciferase Assay System (Promega). The Renilla luciferase plasmid was co-transfected as recommended by the manufacturer to normalize the results.

Statistics

All values are expressed as mean ± standard error. All experiments were performed three times, with exception of the animal study which was performed two times. Values were compared using the Student-Newman-Keul's test. P < 0.05 was considered significant.

SIGNIFICANCE.

Patients with germline fumarate hydratase (FH) mutation are predisposed to develop aggressive kidney cancer with few treatment options and poor therapeutic outcomes. An unbiased drug screen using FH-deficient tumor-derived cell lines identified the tyrosine kinase inhibitor vandetanib to be highly cytotoxic both in vitro and in vivo. Mechanistic studies revealed that vandetanib-mediated cytotoxicity is ABL1-dependent. ABL1 is activated in both FH-deficient kidney tumor specimens and tumor-derived cell lines, and simultaneously promotes aerobic glycolysis and NRF2-mediated antioxidant defense. In animal xenograft studies, an 8-week regimen of drug treatment resulted in 13 months of tumor-free survival. Thus, inhibiting ABL1 may provide a clinically feasible strategy for treating patients with highly aggressive FH-deficient kidney cancer and perhaps additional glycolytic, oxidatively stressed tumors.