|Home | About | Journals | Submit | Contact Us | Français|
Resistance and partial responses to targeted monotherapy are major obstacles in cancer treatment. Systematic approaches to identify efficacious drug combinations for cancer are not well established, especially in the context of genotype. To address this, we have tested pairwise combinations of an array of small molecule inhibitors on early passage melanoma cultures using combinatorial drug screening. Results reveal several inhibitor combinations effective for melanomas with activating RAS or BRAF mutations, including mutant BRAF melanomas with intrinsic or acquired resistance to vemurafenib. Inhibition of both EGFR and AKT sensitized treatment-resistant BRAF-mutant melanoma cultures to vemurafenib. Melanomas with RAS mutations were more resistant to combination therapies relative to BRAF mutants, but were sensitive to combinations of statins and cyclin-dependent kinase inhibitors in vitro and in vivo. These results demonstrate the utility of combinatorial drug screening for discovering unique treatment regimens that overcome resistance phenotypes of mutant BRAF and RAS driven melanomas.
Inhibition of oncogenic drivers by targeted agents improves patient response rates (1–3), but primary and secondary resistance to these drugs is common (4–6). In melanoma, about half of patients harbor activating mutations in the BRAF oncogene (7). Most of these patients respond to the mutant BRAF inhibitor vemurafenib (PLX4032/Zelboraf), however some patients have primary resistance to therapy (8–9). Acquired resistance occurs over a period of months in nearly all patients that do respond (8–9). A large number of resistance mechanisms have been identified (10–18), and combinatorial therapies with vemurafenib will be needed in order to prevent disease progression and improve patient survival.
Another roadblock to progress in melanoma therapeutics is the lack of effective therapies for mutant RAS-driven melanomas, which constitute up to one-fifth of melanoma patients (7,19). RAS mutations in general (NRAS, KRAS, and HRAS) occur in up to a third of all human cancers (20) and are associated with relatively poorer prognoses in many cancer types, including lung and melanoma (21–22). Furthermore, activating mutations in NRAS are one mechanism for secondary (acquired) vemurafenib-resistance in a subset of mutant BRAF melanomas (14). Thus, there is an urgent need to identify effective drug combinations that attack mutant RAS-driven tumors and prevent treatment resistance. To date, there have been no reports describing any systematic method for identifying the most effective drug combinations for either mutant RAS or BRAF melanomas. Therefore we have utilized a high-throughput combinatorial drug screening approach to evaluate the selectivity of drugs, alone and in pairs, in the context of BRAF or RAS activating mutations using early-passage melanoma cultures.
We identify several unique cytotoxic combinations with agents commonly used in the clinic that have pronounced genotype-selectivity for RAS or BRAF mutant melanomas, including those with primary and acquired resistance to vemurafenib. Further, we confirm the effectiveness of select combinations in vitro and in vivo and provide mechanistic reasoning for their cooperation. The results show that combinatorial drug screening illuminates positive drug interactions in melanoma otherwise obscured with single-agent screening alone, and that these findings may guide the implementation of new combinatorial drug trials.
To address the impact of genotype on response to a range of anti-cancer agents, we curated a panel of 150 small molecule compounds including traditional chemotherapies and targeted agents (Table S1) and tested the concentration-response behaviors in early-passage patient-derived melanomas characterized for activating mutations in BRAF, RAS, or those wildtype for both (hereafter, WT) (Table S2). Concentration-response curves generally exhibited sigmoidal or exponential behavior and many agents elicited incomplete growth inhibition (Fig. S1A), suggesting that additional agents are needed for full growth inhibition. Concentrations required to inhibit 50% of cell growth (GI50) were variable among cell lines (Table S3). Compounds arrayed in duplicate on different plates yielded similar concentration-dependent responses (Fig. S1B), as did compounds in related pharmacological classes (Fig. S1C).
Unsupervised clustering of single agent efficacies (maximal growth inhibition achieved) was performed to identify agents that selectively inhibit growth of genotypic subsets. Drugs partitioned into three clusters including agents lacking effect (Fig. 1A, Cluster 1), agents with uniform high efficacy (Cluster 2) or variable efficacies across cell lines (Cluster 3). Re-clustering of drugs showing variable efficacies confirmed mutant BRAF-selectivity of vemurafenib, PLX4720, and GDC-0879 as well as clustering of other agents belonging to the same specific classes, including ErbB, SRC/ABL, and MEK inhibitors (Fig. 1B). In agreement with previous studies, growth-stimulatory effects of vemurafenib were seen in some mutant RAS and WT melanomas and incomplete growth inhibition was observed for all mutant BRAF lines at concentrations up to 10µM vemurafenib (Fig. 1C) (23–24). Approximately 25% of mutant BRAF cultures were innately resistant to vemurafenib, showing a GI50 beyond 3µM (Fig. 1D), as observed elsewhere (25). It has been suggested that PTEN deficiency may hinder vemurafenib-induced apoptosis (16), while co-occurring PTEN and RB1 deficiency may attenuate growth inhibition by vemurafenib (18). Although we did not evaluate the contribution of apoptosis to growth inhibition in the limited number of lines deficient in PTEN (YUGEN8, YUMUT, and YUKADI) or RB1 (YUSAC2), we observed growth inhibition resistance to vemurafenib in the setting of normal PTEN and RB1 expression (Fig. 1E and Table S2), suggesting that loss of either or both is not essential for primary vemurafenib-resistance. Indeed, others have observed that primary resistance to vemurafenib can occur by alternative mechanisms, including increased MET/HGFR signaling (30, 31).
We found that other drugs eliciting selectively higher growth inhibition in mutant BRAF melanomas included the SRC/ABL inhibitor bosutinib, the FGFR inhibitor dovitinib and the EGFR inhibitor gefitinib (Fig. S1D). No drugs were selective exclusively for the WT melanomas, likely due to the greater genetic heterogeneity of this group. MEK inhibitors U0126 and CIP-1374 were selective as single agents for both the mutant BRAF and RAS groups (Fig. S1D), but were more effective for mutant BRAF lines. Lines with primary resistance to vemurafenib were also less sensitive to MEK blockade (Fig. 1B), in agreement with other reports (26). Interestingly, the HMG-CoA reductase inhibitor simvastatin, which indirectly interferes with post-translational processing of RAS proteins (27), was the only drug to trend towards higher degree of growth inhibition in the mutant RAS group (Fig. 1F). Collectively, these data reveal a limited number of single agents with genotype-selective efficacy (Table S4). Moreover, unless used at high concentrations (10 µM or greater), these single agents inhibited cell growth incompletely, which likely reflects that these agents are incompletely cytotoxic at lower concentrations.
The limited responses to single agents prompted us to determine whether more defined genotype-selective patterns and higher efficacies could be observed using combinatorial high-throughput drug screening (cHTS). Forty representative agents with high or variable efficacies (Fig. 1A, Clusters 2 and 3) were chosen for cHTS (Table S5). As a group, these agents inhibit many of the signaling pathways important in cancer (Fig. S2A). Since melanoma lines varied in their sensitivities to each agent, we chose three consensus concentrations based on corrected median values for the GI50, GI25, and GI10 effect levels across all melanoma lines (Fig. S2B), together giving nine pair-wise combinations per drug pair. Nineteen melanoma lines including eight mutant BRAF, six mutant RAS (five NRAS mutants and one HRAS mutant), and five WT lines were screened against more than 7,000 pair-wise combinations in parallel with single agent controls.
Relationships between genotype and drug combination efficacy were first assessed by unsupervised clustering of the most effective combination out of nine tested for each drug pair per cell line (Fig. 2A). Two clusters are apparent, with one dominated by mutant BRAF melanomas that are sensitive to a substantial number of combinations, and the other dominated by mutant RAS and WT melanomas, which were generally less sensitive to the combinations. To ascertain genotype-selective combinations, we filtered data for drug pairings that yielded an average 15% or greater growth inhibition exclusively in a genotypic group. Mutant BRAF-selective drug pairs greatly outnumbered drug pairs selective for mutant RAS or WT groups (1021, 75, and 22 combinations, respectively). Agents most frequently paired in combinations selective for the mutant BRAF group (“sensitizers”) included the FGFR inhibitor PD173074, the pan-BCL2 family inhibitor obatoclax, vemurafenib, bosutinib, the c-MET inhibitor Pha665752, the ErbB inhibitor lapatinib, and the AKT inhibitor MK-2206 (Fig. 2B). All of these agents were selective for BRAF-mutant cultures at multiple concentration combinations (Table S6), reinforcing their cooperative effect. Vemurafenib was not the most frequent sensitizer, possibly due to the lower concentrations used in combination screening (Table S5) as well as the presence of two innately vemurafenib-resistant lines used in the screens (YUKSI and 501Mel, Fig. 1D). At the concentrations tested in cHTS, none of these sensitizers were exclusively selective for mutant BRAF melanomas in single-agent screening.
Strikingly, 52% of mutant RAS-selective drug combinations involved simvastatin (Fig.2C), the only agent with a lower median GI50 in the mutant RAS group in single agent screening (Fig. 1F). Despite the low number of mutant RAS-selective combinations, 25 other drugs paired with simvastatin showed greater effectiveness in the mutant RAS group while only four drugs in combination with simvastatin were more effective in mutant BRAF lines, and were all general sensitizers of this group (Table S7). 62/64 combinations that elicited greater than 50% inhibition in the mutant RAS group included simvastatin (Table S8). Agents that combined with simvastatin at different concentrations selective for the mutant RAS group included the XIAP inhibitor embelin, the novel allosteric MEK inhibitor CIP-1374, the IGF1R non-specific inhibitor BMS-536924, the HSP90 inhibitor 17-DMAG, and the pan-CDK inhibitor flavopiridol (Table S6). None of these agents showed greater potency exclusively in the mutant RAS group with single agent screening (Fig. 1B and Table S4).
Finally, no drug combination partners were selective for the WT group of melanomas. Some combinations were effective (≥ GI75) for all genotypic groups, although many included compounds that have not been utilized in the clinic, or were tested at higher concentrations of clinically-tested agents shown to be genotype-selective in combinations. Table S8 collectively lists the genotype-selective and non-selective combinations.
Drug combinations were analyzed for additive, synergistic, or antagonistic interactions. Single agent effects assessed in cHTS were plotted in relation to theoretical combinatorial effects using the Bliss independence model (28) and to their measured effects for each of the nine concentration combinations per drug pair, depicted globally as drug “interaction signatures” (Fig. 3A). The majority of mutant BRAF melanomas shared a pattern of extensive synergies, whereas combinations in mutant RAS lines were more often antagonistic (i.e. less effective than predicted additivity) (Fig. 3B). To identify genotype-selective synergistic drug pairs, we performed unsupervised clustering of average synergy values obtained from the 9 concentration combinations for each drug pair. Drugs that frequently appeared in combinations showing significantly higher synergies included lapatinib, vemurafenib, bosutinib, nilotinib, and PD173074 (Fig. 3C), and were usually observed in BRAF-mutant melanomas. In contrast, these combinations were often antagonistic in mutant RAS melanomas (Fig. 3D). Lapatinib was the most frequent synergizing partner showing specificity for mutant BRAF lines (Fig. 3E).
A significant fraction of patients with BRAF-mutated melanoma do not respond to vemurafenib (9). YUKSI and 501Mel cell lines harbor activating BRAF mutations but show primary resistance to vemurafenib (Fig. 1D). These lines also showed higher relative resistance against many drugs paired with vemurafenib, even at the highest concentrations tested (Fig. 3F). Nevertheless, both resistant lines were more sensitive to the same set of combinations that were effective for vemurafenib-sensitive melanomas, including vemurafenib paired with bosutinib, lapatinib, obatoclax, PD173074, or Pha665752 (Fig. 3F, highlighted area). Combinations that overlapped in our queries for highest average synergy and average efficacy specific to mutant BRAF lines often included lapatinib and MK-2206 (summarized in Table 1). This suggests a shared mechanistic susceptibility in BRAF mutants to inhibition of the targets of these agents, regardless of vemurafenib sensitivity. MK-2206 and lapatinib were not found to be selective for mutant BRAF melanomas in single-agent screens, indicating that the genotype-selectivity of these agents in combination, as based on BRAF or RAS mutation status, was not predictable by single drug screening alone. Combinations most selective for RAS mutated melanomas were mainly additive and included simvastatin with the pan-CDK inhibitor flavopiridol or the HSP90 inhibitor 17-DMAG.
We next confirmed cytotoxic and synergistic properties of select drug pairs that overlapped in our analyses for highest average synergy and efficacy (Table 1). For mutant BRAF-specific combinations, we chose the AKT inhibitor MK-2206 paired with lapatinib or bosutinib (Fig. 3G, S3A, and S3D). For mutant RAS-specific combinations, we chose simvastatin with flavopiridol or with 17-DMAG (Fig. 3H, S3B, and S3E). Finally, vorinostat with flavopiridol was selected as a genotype-independent combination (Fig. S3C). Isobologram analyses (29) using a broader range of concentrations than those used in cHTS confirmed the overall synergistic nature of these drug interactions as determined by combination index calculations (Fig. 4A and Table S9). Lapatinib or bosutinib in combination with MK-2206, were highly synergistic on the representative mutant BRAF line YUMAC. Simvastatin with flavopiridol or 17-DMAG were also additive on the mutant NRAS line YUGASP at high concentrations, but synergistic at concentrations well below the range used in cHTS.
Lapatinib and MK-2206 individually induced approximately 20% cell death each (normalized to vehicle control) by three days in the BRAF mutant line YUMAC (Fig. 4B and 4C). The combination of these two drugs at the same concentrations increased cell death by 50% (87% actual, 36% Bliss model prediction), consistent with their synergy seen in cHTS and isolobologram analyses. These agents alone or in combination lacked appreciable cytotoxicity in the mutant NRAS line YUGASP. Conversely, simvastatin paired with flavopiridol was more cytotoxic in YUGASP (80% actual, 63% predicted), compared to the BRAF mutant line YUMAC (36% actual, 35% predicted). The combinations of simvastatin with 17-DMAG, or bosutinib with MK-2206 similarly resulted in genotype-selective cytotoxicity in the mutant RAS and BRAF lines, respectively (Fig. 4C and S3F). The genotype-unbiased combination of vorinostat and flavopiridol was also cytotoxic to both RAS and BRAF lines at high concentrations.
Most target classes were represented by only one compound in cHTS, so we verified that the genotype-selective effects produced by these combinations are related to their target classes. The EGFR inhibitors gefitinib or afatinib/BIBW2992 paired with the AKT inhibitors MK-2206 or GSK692094 produced similar combination responses to lapatinib and MK-2206, and were more effective in mutant BRAF lines, including the line most intrinsically resistant to vemurafenib (YUKSI) (Fig. 4D and S4A). Likewise, the combination of statins including lovastatin and atorvastatin combined with the pan-CDK inhibitors flavopiridol or AT7519 produced similar responses that were somewhat effective in mutant BRAF lines, but more effective in NRAS and HRAS mutant lines (Fig. 4D and S4B).
We further evaluated the efficacy of the lapatinib/MK-2206 combination with flow cytometry, clonogenic assays, and soft agar assays on mutant BRAF melanomas sensitive or resistant to vemurafenib using repeated administration of lower concentrations of these agents to minimize off-target effects (see Methods). This treatment strategy was minimally effective for patient-derived lines with the greatest primary resistance to vemurafenib including YUKSI and YUKOLI, as indicated by maintained cell viability and colony numbers with reduced colony size (Fig. 5A and 5B), but relatively more effective for the vemurafenib-sensitive lines including YULAC (Fig. S5A). Vemurafenib-resistant cell lines selected from YULAC and YUCOT lines (Fig. S5B), hereafter, YULAC-R and YUCOT-R, respectively, were also less sensitive to this regimen (Fig. 5B and S5C). At 500 nM and above, lapatinib and MK-2206 effectively suppressed their targets p-EGFR and p-AKT, respectively (Fig. 5C), despite moderate impact on viability and clonogenicity of these agents singly and in combination.
We next assessed the interaction of vemurafenib with lapatinib or MK-2206. Dual combination of vemurafenib with either of these agents moderately increased efficacy relative to single-agent treatments in primary resistant lines YUKSI and YUKOLI (Fig. 5D and 5E) and in acquired resistant lines YULAC-R and YUCOT-R (Fig. 5E, S5D, and S5E). Vemurafenib and MK-2206 together were more effective in acquired resistant lines; however the triple combination of lapatinib, MK-2206, and vemurafenib greatly enhanced cytotoxicity and abolished colony growth in both primary and secondary resistance settings. Importantly, the dual and triple combinations did not induce substantial cytotoxicity in the mutant NRAS line YUGASP and only moderately reduced colony formation (Fig. S5F).
Receptor tyrosine kinase activation often contributes to vemurafenib-resistance in mutant BRAF melanomas (13–14,30–31). In primary vemurafenib-resistant YUKSI and YUKOLI BRAF-mutant melanoma cells, inhibition of AKT with MK-2206 single-agent treatment effectively suppressed p-AKT, but increased levels of p-EGFR and p-ERK by 24 hours (Fig. 5C, 5F, S6A, and S6B). Conversely, vemurafenib or lapatinib alone did not change p-AKT levels in these lines. In the triple combinations, the activity of MAPK and PI3K/mTORC pathways were effectively suppressed, as suggested by strong reduction in p-ERK, p-AKT, and p-P70S6K (Fig. 5F), although some rescue of p-ERK signal was observed. This was likely due to cross-pathway activation of ERK upon p-AKT inhibition with MK-2206, as all dual and triple combinations with MK-2206 partially restored p-EGFR and p-ERK.
The impact of these agents on MAPK and PI3K signaling in acquired vemurafenib-resistant lines (YULAC-R and YUCOT-R) was similar to primary vemurafenib-resistant lines. Vemurafenib or MK-2206 alone depleted or partially restored p-ERK levels, respectively (Fig. 5G and S6C). In contrast, single-agent lapatinib treatment in acquired vemurafenib-resistant lines resulted in initial suppression, but subsequent elevation of p-ERK and p-AKT levels relative to baseline (Fig. 5G), whereas it had little impact on p-AKT and p-ERK in primary vemurafenib-resistant lines. Nevertheless, the three combined agents effectively suppressed both p-AKT and p-ERK levels in the vemurafenib-sensitive (Fig. S6D and S6E) and vemurafenib-resistant lines tested.
Lastly, we assessed the single, dual, and triple-agent treatments on preventing the long-term emergence of resistant colonies (Fig. 5H). As expected, the parental line YULAC readily formed colonies with lapatinib or MK-2206 alone or combined (data not shown). Dual-agent combinations with vemurafenib performed more effectively, however a lower number of resistant clones did emerge. Only the vemurafenib/lapatinib/MK-2206 triple combination was found to impede the emergence of resistant colonies completely.
Statins interfere with isoprenyl RAS modifications required for plasma membrane localization and activity (32–34). Indeed, membrane localization of NRAS was nearly eliminated with simvastatin treatment (Fig. 6A), confirming an association between loss of membrane-anchored NRAS and induction of cytotoxicity. Still, a large number of other proteins, including cancer-relevant RAC and RHO proteins, are also isoprenylated and may be concomitantly affected by statins. We therefore performed RNAi-mediated knockdown experiments in order to compare the effects of NRAS reduction to statin-induced NRAS inhibition. NRAS siRNA treatments resulted in nearly complete loss of mutant NRAS protein, greatly reduced p-ERK, and moderately reduced p-AKT in mutant NRAS YUGASP cells (Fig. 6B). The combination of NRAS siRNA and flavopiridol treatment fully suppressed p-ERK levels, and substantially increased BIM and PARP cleavage by 96 hours. This was associated with increased cytotoxicity relative to either treatment alone as measured by flow cytometry (Fig. 6C). Similar to siNRAS, repeated administration of lower concentration simvastatin reduced activity of MAPK and PI3K pathways. This effect was enhanced in the presence of flavopiridol in NRAS and HRAS mutant lines (Fig. 6D, S7A, and S7B), along with a significant reduction in viability and clonogenicity (Fig. 6E, S7C, and S7D). In contrast, cytotoxicity and clonogenic inhibition following treatment with these agents was not remarkable in mutant BRAF lines resistant to vemurafenib (Fig. S7E and S7F). Collectively, these data demonstrate enhanced effects of statins combined with CDK inhibitors in NRAS or HRAS mutant melanomas that may in part be mediated by reduction of NRAS activity.
Finally, we tested the impact of the simvastatin/flavopiridol combination in reducing tumor growth using a pre-clinical xenograft model. YUGASP cells were injected subcutaneously into immunocompromised mice and a dose-escalation study was performed to ensure tolerability up to pre-defined maximum doses of the two drugs combined. No toxicity was observed with single agents or dual agents in the ranges of doses tested (data not shown). Simvastatin and flavopiridol modestly reduced tumor growth as single agents, with flavopiridol having a greater effect (Fig. 6F and 6G). The combination of simvastatin and flavopiridol significantly reduced tumor growth and resulted in initial tumor regression within the first week of treatment. The combination was well-tolerated (Fig. S7G) and histological assessment at the two-week mark indicated a trend towards increased cell death as indicated by pyknotic cell index and reduced mitotic index in the combination treatment group relative to single-agent and mock treatment arms (Fig. S7H).
Resistance to therapies in cancer is a major clinical hurdle and there is a compelling need for the discovery of more effective combinations of agents that are currently available. We present results from a drug combinatorial screen designed to probe positive drug interactions in mutation-defined subgroups of cancer. This has identified previously undescribed drug interaction patterns and several combinations with potential for high efficacy in melanoma (Table 1). Among these, statins with CDK inhibitors were validated as selective for NRAS and HRAS mutants, and a triple combination consisting of vemurafenib, EGFR and AKT inhibitors was selective for BRAF mutant lines including those with primary or in vitro-selected resistance to vemurafenib.
In patient-derived BRAF mutant melanomas with vemurafenib resistance, the EGFR inhibitors gefitinib, lapatinib, and afatinib were minimally cytotoxic when used alone, as were the AKT inhibitors MK-2206 and GSK692094, but more effective when combined at high concentrations than at lower concentrations. Yet, combinations of lapatinib and MK-2206 at reduced concentrations still effectively sensitized resistant cells to vemurafenib, giving credence to the idea that the major targets of these agents are critical mediators of vemurafenib resistance.
Cross-pathway feedback control between the MAPK and PI3K pathways is an important feature of many cancers (35–36). In all vemurafenib-sensitive or resistant BRAF mutant lines tested, we found an increase in ERK activity upon AKT inhibition with MK-2206. These effects were robust enough in primary vemurafenib-resistant lines to partially rescue p-ERK levels in spite of vemurafenib treatment. Thus, the consequences of MK-2206 treatment are likely two-pronged: suppression of the survival role of the AKT pathway and alteration of feedback inhibition resulting in increased MAPK signaling, potentially through receptor kinases.
Melanomas with primary vemurafenib resistance were found to have higher baseline EGFR levels, similar to colorectal cancers insensitive to this agent (37–38). In contrast to these studies, we found that combined EGFR and mutant BRAF inhibition was not effective for primary vemurafenib-resistant melanoma cells, or in preventing the emergence of resistant clones in vemurafenib-sensitive cells. Previous work in breast cancer demonstrated upregulation of multiple RTKs, including EGFR, upon AKT inhibition with MK-2206 (36). Thus it is plausible that primary resistant melanomas will require AKT inhibition as a prerequisite for enhancing reliance upon MAPK signaling through RTKs and mutant BRAF. Additionally, the partial reduction in p-ERK with EGFR inhibition seen in these cells may increase sensitivity to further MAPK inhibition by vemurafenib treatment. More generally, this suggests that combinations of specific RTK inhibitors with AKT inhibitors could likewise re-sensitize acquired vemurafenib resistant lines to vemurafenib, and indeed, we found that the triple-agent combination was superior in these cases. Moreover, this regimen also blocked the long-term development of resistant clones in parental vemurafenib-sensitive cells, suggesting a route for prevention of vemurafenib resistance, which otherwise develops over a period of months (8–9).
The effects of the vemurafenib/lapatinib/MK-2206 combination were minor on mutant NRAS cells, reaffirming its selectivity for lines with BRAF mutations. Generally, RAS-mutant melanomas were resistant to most single-agent and dual-agent treatments in comparison to mutant BRAF melanomas and melanomas with wild-type BRAF and RAS, which is consistent with the slightly poorer clinical prognosis associated with NRAS (22). Many drug combinations synergistic in mutant BRAF cells tested in cHTS, including RTK inhibitor combinations, were more often antagonistic in mutant RAS cells. This likely reflects the greater pleiotropy of RAS-dependent signaling since RAS activates a variety of effector pathways including MAPK, JAK-STAT, RAL-GDS, and PI3K signaling (20,34).
Mutant RAS-selective combinations detected in cHTS often involved the HMG-CoA reductase inhibitor simvastatin. Clinical studies have not supported use of statins as cancer monotherapies (39), and epidemiological studies have not conclusively substantiated reduced cancer risk of individuals chronically treated with statins at the somewhat lower concentrations used for hypercholesterolemia control (39–40). Still, it remains to be seen if statins are more functional for the specific prevention or treatment of RAS-mutant cancers.
Here we find that the impact of statins on NRAS mutant melanomas may be mediated through direct interference with the function of NRAS, as we confirmed that it is relocalized from the plasma membrane under these conditions, an important component of RAS signaling (41). Moreover, NRAS knockdown yielded similar biological phenotypes to statin treatment, both alone, and in combination with flavopiridol. Nonetheless, NRAS knockdown was more effective in suppression of p-ERK than was simvastatin, and it is possible that the myriad of proteins reliant on isoprenoid modifications, including additional GTP-binding proteins important in cancer such as RAC and RHO, and more global effects on lipid metabolism contribute to these differences.
Concentrations of statins used in combinations to elicit complete cytotoxicity were by themselves only partially cytotoxic. Moreover, tumor growth inhibition with simvastatin alone in pre-clinical xenografts was not remarkable. RNAi-mediated knockdown of oncogenic NRAS elicited only partial cytotoxicity, similar to results seen with KRAS mutant cancers and mutant NRAS melanomas with or without activating BRAF mutations (14,42–44). However, cHTS identified multiple second agents combined with simvastatin that were more effective in mutant RAS melanomas. Thus, these data demonstrate that loss of mutant NRAS, whether induced by small-molecule agents or by synthetic oligonucleotides, is mainly a critical priming event for cytotoxicity induced by a second agent such as a HSP90, MEK, or CDK inhibitor.
The pan-CDK inhibitor flavopiridol alone was not selective for mutant RAS melanomas, so this drug in combination with statins is presumably working to forestall residual cell cycle activation through RAS-dependent pathways or through pathways acting in parallel (45). In support of this idea, synthetic lethality has been shown elsewhere with combinatorial knockdown of mutant KRAS and CDK4 in NSCLC mouse models (46). One phase II trial of flavopiridol as monotherapy in stage IV melanoma patients resulted in stable disease in approximately half of patients, for up to half a year or more (47). Candidate oncogenic drivers were not evaluated and so the efficacy of CDK inhibition in conjunction with oncogenic driver blockade, such as mutant RAS, remains unknown in humans. Our in vivo mouse studies demonstrate the tolerability and enhanced efficacy of flavopiridol in combination with simvastatin for mutant NRAS melanoma. This combination may also prove superior for preventing or treating vemurafenib-resistant BRAF V600* melanomas that acquire de novo mutations in RAS (14,48).
Over the last two decades, the discovery of novel targeted agents that inhibit the oncogenic drivers of cancers has paved the way for more favorable patient outcomes, and importantly, more tolerable therapies. However, even the most precisely targeted therapies when used alone are limited in their ability to promote cytotoxicity in some cancer cells (28). Poor-prognosis cancers, such as advanced-stage melanomas, will require combination therapies to obstruct the outgrowth of resistant cells (49). Combinatorial drug screening has allowed for the discovery and experimental confirmation of a number of effective combination regimens that in the correct genotypic setting may prove more effective and tolerable in patients, whilst avoiding selection for resistance through inadequate dosing.
Patient-derived melanoma lines, with the exception of 501Mel, were collected as previously described (24). Cell lines were derived directly from human melanoma metastases using a Yale institutional review board-approved protocol with informed consent. Tumor lines were further confirmed by expression profiling and Sanger sequencing for BRAF and NRAS mutational status (Table S2) and compared with the independent clinical genetic evaluation of the pathological resection specimens in many cases. These studies served to validate that the short-term cultures expressed melanoma markers and were of correct designated genotype. Cell lines were cultured in basal medium [OptiMEM (Invitrogen) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin (P/S)] and maintained in 37°C incubator maintained at 5% CO2. Acquired vemurafenib-resistant lines (YUCOT-R and YULAC-R) were generated by exposing parental lines to 3µM vemurafenib every two days for approximately 10–12 weeks.
Cells were deposited into 384-well microtiter plates at 750 cells/well using a multidrop dispenser (Thermo) in 20µl basal medium. Drug stock plates for 1000X single agents were created by serial 1:2 dilution typically from 10mM (Table S1), using an expandable multichannel pipette (MatrixTechCorp). A PlateMatePlus automated instrument (MatrixTechCorp) was used for pin transfer of 20nl drug volume from drug stock plates into 384-well microtiter cell plates. 0.1% and 20% DMSO was used as negative and positive controls, respectively. For dual-agent screens, 96-deepwell plates were used for single-agent concentration stocks and pairwise combinations of drugs were generated using hit-picking automation (Freedom EVO, Tecan). 780 unique drug pairs at nine concentration combinations each were generated spanning twenty-two 384-well microtiter stock plates. All experiments were performed in triplicate at the Yale Center for Molecular Discovery (YCMD). Cells were exposed to drug for 72 hours followed by growth inhibition measurement with the CellTiterGlo® ATP detection assay (Promega) based on amenability for HTS (50). Only experiments with high Z-factor quality indices (>0.5) were analyzed.
Data were compiled into a relational database using PostgreSQL RDBMS (PostgreSQL.org.). Statistics were computed using built-in functions of PostgreSQL or with R (R-project.org). In single-agent studies, concentration-effect curves were computed using Michaelis-Menten or four-parameter logistic curve fitting in R. Growth inhibition (GI) values were only interpolated. Maximum GI and GI50 effects achieved with single agents were hierarchically clustered by Ward’s method based on Euclidean distance using R. Kruskal-Wallis test was used to compare genotype groups with regards to efficacy differences. For dual-agent efficacy, unsupervised hierarchical clustering was performed using heatmap.2 function from the gplots library in R. For synergy and antagonism studies, unsupervised clustering was performed and analyzed by ANOVA and Tukey-Kramer pairwise testing. Identification of drug synergy was assessed using the Bliss independence model (28), or the Chou-Talalay method (29) using normalized isobologram analyses.
Cells were plated at 1×105 cells/well in 6-well format and allowed to adhere overnight. Lifted cells were combined with adherent cells harvested by trypsinization. Repeated administration of agents was performed every other day for three total treatments using fresh medium and drugs, and analyzed by flow cytometry the following day. Cells were stained using the BD Pharmingen Apoptosis Detection Kit II according to the manufacturer's protocol (BD). Samples were analyzed with the BD LSRII flow cytometer with recording of at least ten-thousand events per sample. Cells were analyzed by doublet discrimination and compensation was applied for each experiment. Each line was treated independently and gates fixed based on negative control signals. Plots were generated using FlowJo 7.6.5.
Cells were plated on 8-chamber culture slides (BD Falcon) at 1000 cells/well and allowed to adhere overnight before drug treatments. Cells were fixed using 4% paraformaldehyde at room temperature for 15 minutes, followed by quenching with 100mM glycine, permeabilization with 0.1% Triton X-100/PBS, and blocked with 5%BSA/PBST for 30 minutes at 37°C. The anti-NRAS antibody (Santa Cruz, clone F155), which recognizes both mutant and wildtype NRAS, was used at 1:100 in 100µl blocking buffer overnight at 4°C, followed by AlexaFluor594 secondary antibody (Invitrogen) at 1:1000 for 2 hours, followed by wash and mounting with Prolong Gold (Invitrogen). IgG1 species-matched antibody was used for isotype negative control. An Olympus IX71 scope was used for fluorescence microscopy.
For two-dimensional (2–D) clonogenic assays, cells were plated at 5×103 cells/well in 6-well tissue culture-treated plates and grown for 72 hours in 2 ml basal medium. Drug treatments were performed at 72 hours and then every other day for a total of 4 treatments for 2-D colony assays or 6 treatments for soft agar assays, each time replenished with an additional 1ml medium containing fresh drug(s). Cells were allowed to recover for 2 weeks for 2-D colonies, or 3–4 weeks for soft agar. 2-D colonies were fixed in ice-cold 100% methanol for 15 minutes and stained for 20 minutes with 0.05% crystal violet followed by de-staining with water. For soft agar, the base layer consisted of 1ml 1.6% low gelling-temperature agarose (Sigma, #A9414) combined with 1ml 2X RPMI medium with 20% fetal bovine serum and 2% P/S (2X RPMI complete). The top layer consisted of 750µl 0.6% agarose combined with 750µl 2X RPMI complete medium containing 1×104 cells/well in 6-well plate format. Colony spheres were fixed and stained with 0.01% crystal violet [50ml MeOH, 10ml glacial acetic acid, 4ml 0.5% crystal violet, 36ml dH20] for 2 hours at RT followed by de-staining with water. 2-D clonogenic plates were scanned with a VersaDoc Model 3000 imager (Bio-Rad) and QuantityOne software. Soft agar plates were photographed. Both 2-D and soft agar colonies were enumerated using ImageJ software version 1.46r.
Immunoblots were performed with the following primary antibodies, all at 1:1000: NRAS (catalog #sc-31), GAPDH (sc-25778), and EGFR (sc-03) (Santa Cruz Biotechnology), and phospho-AKT (Ser473) (#4060), AKT (#9272), phospho-p42/44 MAPK (Thr202/Tyr204) (#9106), p42/44 MAPK (#4695), phospho-p70S6K (Thr421/Ser424) (#9204), p70S6K (#9202), phospho-EGFR (Tyr1068) (#3777), BIM (#2819), PARP (#9542), β-actin (#4970), PTEN (#9559) and Rb1 (#9309) (Cell Signaling Technology), and ImmunoPure donkey anti-rabbit (#31458) and goat anti-mouse (#31432) HRP-conjugated secondary antibodies (Thermo).
The NRAS siRNA ON-TARGET plus SMART Pool (Dharmacon) was used for in vitro knockdown experiments. 10µl FuGENE HD transfection reagent (Promega) was added to 190µl of basal media without P/S, mixed by inversion and allowed to equilibrate for 10 minutes. In parallel, 10µl siRNA was added to 190µl of basal media without P/S at 500nM and combined 1:1 with the transfection reagent/basal media mixture, resuspended, and added to cells at 30–50% confluence in 6-well plate format in 1600µl media (final siRNA 50nM). Fresh siRNA/transfection reagent mixture was added again 24 hours later as described above and cells were collected for protein or flow cytometry after 96 hour total exposure to siRNA. Flavopiridol (0.1µM) was added 48 hours before protein or cell collection where indicated.
5×106 cells were resuspended in basal medium and injected subcutaneously into both flanks of 5–6 week old NCr Nude mice (Taconic). Tumors were grown until palpable at 200mm3 before treatments. Simvastatin was slowly dissolved in pre-warmed PBS and delivered daily by oral gavage at 10mg/kg. Flavopiridol was dissolved in pre-warmed 0.1% saline solution and delivered intraperitoneally every other day. Tumors were measured by digital caliper every other day and volume estimated using the equation Volume = (width)2 × length/2.
We have used drug combinatorial screening to identify effective combinations for mutant BRAF melanomas, including those resistant to vemurafenib, and mutant RAS melanomas that are resistant to many therapies. Mechanisms governing the interactions of the drug combinations are proposed and in vivo xenografts demonstrate the enhanced benefit and tolerability of a mutant RAS-selective combination, which is currently lacking in the clinic.
We thank Ruth Halaban for her leadership of the Yale SPORE in Skin Cancer, and her generosity in providing materials for this study. We are grateful to Harriet Kluger, Mario Sznol, Karen Anderson, Yung-Chi Cheng, Gil Mor, Rick Bucala, and members of the Stern lab for helpful discussions and generous provision of some agents used in this work. Special thanks to Yale Center for Molecular Discovery for HTS expertise. This work was supported by a grant from an Anonymous Foundation to MWB and DFS, the Harry J. Lloyd Charitable Trust (MAH, CGL, JTP, DFS), USPHS R01CA45708 (DFS), and the Yale SPORE in Skin Cancer, funded by the NCI (P50 CA121974; R. Halaban, PI).
Conflicts of Interest: None declared.