Tumors with mutant BRAF are dependent on the RAF/MEK/ERK signaling pathway for their growth1-3. We found that ATP-competitive RAF inhibitors inhibit ERK signaling in cells with mutant BRAF, but unexpectedly enhance signaling in cells with wild-type BRAF. Here we demonstrate the mechanistic basis for these findings. We employed chemical genetic methods to show that drug-mediated transactivation of RAF dimers is responsible for paradoxical activation of the enzyme by inhibitors. Induction of ERK signaling requires direct binding of the drug to the ATP-binding site of one kinase of the dimer and is dependent on RAS activity. Drug binding to one member of RAF homo-(CRAF/CRAF) or heterodimers (CRAF/BRAF) inhibits one protomer, but results in transactivation of the drug-free protomer. In BRAFV600E tumors, RAS is not activated, thus transactivation is minimal and ERK signaling is inhibited in cells exposed to RAF inhibitors. These results imply that RAF inhibitors will be effective in tumors in which BRAF is mutated. Furthermore, since they do not inhibit ERK signaling in other cells, the model predicts that they would have a higher therapeutic index and greater antitumor activity than MEK inhibitors, but could also cause toxicity due to MEK/ERK activation. These predictions have been borne out strikingly in a recent clinical trial of the RAF inhibitor PLX40324-5. Finally, the model suggests that promotion of RAF dimerization by elevation of wild-type RAF expression or RAS activity could lead to drug resistance in mutant BRAF tumors. In agreement with this prediction, RAF inhibitors do not inhibit ERK signaling in cells that coexpress BRAFV600E and mutant RAS.
Six distinct ATP-competitive RAF inhibitors induced ERK activation in cells with wild-type BRAF, but inhibited signaling in mutant BRAFV600E cells (Fig. 1a, b, Supplementary Fig. 2a, b, Data Not Shown (DNS), structures of compounds shown in Supplementary Fig. 3, except that of PLX4032, which is unavailable). PLX47206, and its analog in clinical trial PLX4032, were studied in more detail. PLX4032 inhibited ARAF, BRAF and CRAF immunoprecipitated from 293H cells (Supplementary Fig. 4) and purified catalytic domains of BRAFV600E, wild-type BRAF and CRAF (IC50s: 35, 110 and 48nM) (Supplementary Table 1). PLX4032 was assayed against 62 additional kinases that span the kinome, and had IC50s of 1μM-10μM against eight of these and greater than 10μM against the rest (G.B., unpublished data). Induction of ERK signaling by PLX4720 was rapid (Fig. 1c), reversible (Fig. 1d), and associated with increased phosphorylation of the ERK substrate RSK (Fig 1b). MEK and ERK phosphorylation were induced at intermediate concentrations of RAF inhibitor, and inhibited at much higher doses (Fig. 1a).
Physiologic induction of ERK signaling depends on upstream activation of RAS by receptor-induced signaling7-8. PLX4032 induced ERK signaling in SKBR3 breast cancer cells, in which RAS activation is HER2-dependent9. The HER2 inhibitor Lapatinib abolished basal and PLX4032-induced ERK signaling in these cells (Supplementary Fig. 5a). In 293H cells, induction of MEK and ERK phosphorylation by either PLX4032 or PLX4720 was barely detectable (PLX will refer to data obtained with both compounds). HA-tagged wild-type RAS overexpression resulted in enhanced MEK/ERK activation by RAF inhibitor, which was more pronounced when mutant RAS was overexpressed (Fig. 2a and Supplementary Fig. 5b). The results suggest that RAS activity is required for MEK/ERK activation by RAF inhibitors. In contrast, in 293H cells expressing FLAG-tagged BRAFV600E, ERK signaling was inhibited by PLX4032 (Supplementary Fig. 5c). These results suggest that RAF inhibitors will inhibit the growth of tumors with mutant BRAF, but not those with wild-type BRAF, including those with RAS mutation. This is indeed the case: MEK-dependent tumors with RAS mutation are unaffected by PLX4032 (unpublished data).
BRAF and CRAF kinases form homo- and heterodimers upon RAS activation10-12. PLX induced pronounced phosphorylation of MEK and ERK in wild-type MEFs and BRAF (−/−) MEFs. The response was diminished markedly in CRAF (−/−) MEFs (Fig. 2b, Supplementary Fig. 6a). Coexpression of CRAF and active RAS in CRAF (−/−) MEFs reconstituted the wild-type phenotype (Supplementary Fig. 6b, c). We conclude that BRAF is dispensable for MEK/ERK activation by PLX, and that CRAF expression is required for significant induction. We therefore investigated the mechanism of CRAF-dependent induction of ERK signaling in response to the drug.
Autoinhibition of RAF by its N-terminal domain13 is relieved upon binding to activated RAS7. We asked whether overexpression of an N-truncated form of CRAF would bypass the requirement for RAS activity. In 293H cells expressing the catalytic domain of CRAF (catC), PLX caused dramatic induction of MEK and ERK phosphorylation (Fig. 2a, Supplementary Fig. 7a). We focused mechanistic investigations on catC, in which PLX-induced MEK/ERK activation is RAS-independent. To test whether direct binding of PLX to CRAF is required for induction of signaling, we generated a catC carrying a mutation at the gatekeeper position (T421) in the kinase domain (mutations used and their properties are in Supplementary Fig. 1a). Structural studies6 predict that the T421M mutation should prevent drug binding and catCT421M was indeed resistant to inhibition by PLX in vitro (Supplementary Fig. 8a, b). ERK signaling was not induced by PLX in cells expressing catCT421M (Fig. 2a, Supplementary Fig. 7b). Thus, activation of MEK/ERK by PLX depends on its direct binding to the RAF kinase active site. Sorafenib inhibited catCT421M in vitro (Supplementary Fig. 8c) and induced ERK signaling in cells expressing catCT421M (Fig. 2c), demonstrating that this mutant is capable of inhibitor-induced MEK/ERK activation. Thus, direct binding of an ATP-competitive inhibitor to CRAF is required for induction of ERK signaling.
Recent work shows that binding of ATP-competitive inhibitors to AKT and PKC inhibits their activity, but induces the active, phosphorylated state of these kinases14-15. Washed catC immunoprecipitated from PLX-treated cells was more active than that isolated from untreated cells (Fig. 3a, Supplementary Fig. 9a). The same was true for endogenous BRAF and CRAF immunoprecipitated from treated Calu-6 cells (Fig. 3b, Supplementary Fig. 9b). Phosphorylation of CRAF at S338 and S621 has been correlated with its activation7. PLX caused increased phosphorylation of both sites on wild-type and kinase-dead CRAF in 293H cells. In contrast, it did not affect the phosphorylation of the PLX-resistant CRAFT421M mutant. (Fig. 3c, Supplementary Fig. 9c). All RAF inhibitors tested induced phosphorylation at p338 of endogenous CRAF (Fig. 3d). The data suggest that binding of PLX to CRAF induces activation of the enzyme and, subsequently, ERK signaling. The result seems paradoxical: binding of ATP-competitive inhibitors to the catalytic domain of CRAF activate its function.
RAF isoforms form dimers in cells10-12,16. Since binding of both the drug and ATP to the catalytic domain would be required for activation and cannot occur simultaneously on the same molecule, we hypothesized that RAF inhibitors activate CRAF dimers in trans. (Supplementary Fig. 1b). To test this model, we generated mutant catCS428C that binds to quinazoylacrylamide-based inhibitors17, whereas catC does not. Two inhibitors JAB1317 and JAB34 (PD-168393)18 both inhibited catCS428C, but up to 30μM had no effect in vitro on catC (Supplementary Fig. 10a, b). JAB13 and JAB34 selectively affected ERK signaling in cells expressing catCS428C, and were inactive in those expressing catC (Supplementary Fig. 11). Like the other inhibitors (Fig. 1a), lower doses (40nM - 1μM) induced ERK signaling (Supplementary Fig. 11), whereas higher doses (10μM) inhibited (Fig. 4a). The specificity of this system allows us to test the dimer transactivation model. We coexpressed a V5-tagged, JAB-sensitive, kinase-dead catC, V5-catCS428C/D486N and FLAG-catC in 293H cells. V5-catCS428C/D486N is deficient in catalytic activity; it can bind to the inhibitor (JAB34) but cannot phosphorylate MEK, while FLAG-catC is catalytically active, but cannot bind JAB34. Treatment of cells expressing both constructs with a concentration of JAB34 that inhibited ERK signaling in cells expressing catCS428C alone (10μM JAB34, Fig. 4a) resulted in marked induction of ERK signaling (Fig. 4b, lanes 5-6). Thus, binding of JAB34 to kinase-dead, V5-catCS428C/D486N transactivated the catalytically competent FLAG-catC. When the catalytically active drug-binding mutant V5-catCS428C is coexpressed with catalytically inactive catC (FLAG-catCD486N), 10μM JAB34 inhibited, rather than activated ERK signaling. (Fig. 4b, lanes 9-10). When both constructs were insensitive to JAB, JAB34 had no effect on ERK signaling (Fig. 4b, lanes 1-2). When both constructs were catalytically active, we observed moderate MEK/ERK activation, likely resulting from inhibition of V5-catCS428C and transactivation of FLAG-catC (Fig. 4b, lanes 3-4).
Transactivation from CRAF to BRAF can occur as well. JAB34 activated ERK signaling in cells coexpressing FLAG-BRAF with V5-catCS428C/D486N (Fig. 4c). Finally, JAB34 induced ERK activation in cells coexpressing full-length V5-CRAFS428C/D486N and wild-type FLAG-CRAF, confirming that our model is valid in the context of full-length CRAF (Supplementary Fig. 12).
Thus, activation of RAF by ATP-competitive inhibitors can be explained by transactivation: binding of drug to one RAF in the dimer activates the other. This is consistent with the enhancement of induction by active RAS, which promotes homo- and hetero-dimerization of BRAF and CRAF10,12. Our model suggests that transactivation will be dependent on formation of RAF dimers. A side-to-side dimer of the kinase domain is observed in crystal structures of BRAF11 and the residues at the dimer interface are conserved in all RAF isoforms. Based on the BRAF crystal structures, we identified a conserved Arg (R509) at the center of the dimer interface. Structural analysis predicts that mutation of R509 will diminish contacts between the two interacting proteins and reduce dimer formation, as also recently reported19. In that study, mutation of BRAF at R509 to Histidine resulted in dramatic loss of activity. The corresponding mutation in catC (R401H) results in severe loss of both expression and activity (DNS). We therefore mutated R401 to Alanine in V5-catCS428C and FLAG-catC. This mutation diminished dimerization (Supplementary Fig. 13), but retained expression and activity. In cells coexpressing these mutants, JAB34 failed to induce ERK signaling (Fig. 4b, lanes 7-8). Thus, a mutation that affects dimerization prevents transactivation.
The transactivation model explains the observation that inhibitors of RAF activate ERK signaling at low concentrations, but inhibit at higher concentrations in BRAFwild-type cells. Binding of an ATP-competitive inhibitor to one protomer within a RAF dimer results in both abolition of the catalytic activity of the inhibitor-bound RAF and transactivation of the other. Transactivation of RAF homo- and heterodimers is therefore responsible for induction of MEK/ERK phosphorylation by RAF inhibitors in cells with wild-type BRAF. Our model explains the paradoxical phenomenon of ERK activation by RAF inhibitors, previously reported by others20-22. Other kinases that exist in dimeric or multimeric complexes may behave in a similar manner. Recently, another model to explain these phenomena has been proposed23. They report that only selective BRAF inhibitors activate CRAF and ERK signaling, whereas pan-RAF inhibitors do not. Our data that all RAF inhibitors activate ERK signaling at low concentrations, that the phenomenon occurs in BRAF-null cells and that binding to CRAF activates CRAF and BRAF-dependent ERK signaling render that model unlikely.
Nevertheless, the clinical utility of these inhibitors depends on their inhibition of ERK signaling in tumor cells with BRAFV600E. Since transactivation of wild-type RAF requires dimerization and depends on RAS activity, we hypothesized that the levels of RAS activity in BRAFV600E mutant tumors may not be sufficient to support transactivation. If so, activation of RAS in BRAFV600E cells should prevent inhibition of ERK signaling by RAF inhibitors. In 293H cells overexpressing BRAFV600E and in HT29 tumor cells with endogenous BRAFV600E, ERK signaling was inhibited by either PLX or a MEK inhibitor. In contrast, when mutant RAS was coexpressed with BRAFV600E in either cell, ERK signaling became resistant to PLX, but remained sensitive to the MEK inhibitor (Fig. 4d, Supplementary Fig. 14a, b).
The data are consistent with the idea that RAF inhibitors suppress ERK signaling in BRAFV600E tumors because the level of RAS activation in these cells is insufficient to support transactivation of wild-type RAF and inhibition of BRAFV600E activity becomes the dominant effect of the drug. The findings suggest that increases in RAS activation or RAF dimerization may be sufficient to cause drug resistance.
The clinical implications of these findings are profound. BRAFV600E tumors and some with RAS mutation are dependent on ERK signaling. However, in clinic, a MEK inhibitor had only a 12% response rate in melanomas with BRAF mutation24. MEK inhibitors block ERK signaling in all tumor and normal cells and the dose of the drug that can be administered is limited by toxicity. RAF inhibitors and MEK inhibitors might have been expected to have similar biologic effects. Our findings show otherwise. RAF inhibitors will be useful for the treatment of tumors driven by BRAFV600E, but could have deleterious effects in some contexts due to ERK activation. However, the absence of ERK inhibition in normal cells may allow administration of high doses of RAF inhibitors and thus more complete inhibition of ERK signaling in BRAFV600E tumors, than is possible with MEK inhibitors.
The recent phase I clinical trial of PLX4032 in metastatic melanoma strikingly confirmed these predictions4-5. High serum levels of drug were achieved with modest toxicity and resulted in profound inhibition of ERK signaling in tumors. Tumor regression was observed in more than 90% of patients with BRAFV600E mutation, with 64% achieving a partial response by RECIST criteria. We believe that the remarkable activity of this drug, compared to that of MEK inhibitors, is due to its ability to inhibit ERK signaling in tumors more completely because of the absence of ERK inhibition in normal tissue.
Resistance to PLX4032 does develop, with a median time to disease progression of 8-9 months5. Potential mechanisms include gatekeeper mutations in BRAF and activating mutations in parallel signaling pathways. Our results suggest the possibility of novel mechanisms as well. Lesions that activate RAS or, as recently reported, overexpression of wild type RAF isoforms25 could result in inability of RAF inhibitors to suppress ERK signaling in the tumor and thus lead to resistance.