Activation of BRAF by mutation occurs in approximately 8% of human cancers including the majority of melanomas (Davies et al., 2002
, Badalian-Very et al., 2010
; Brose et al., 2002
; Nikiforova et al., 2003
; Schiffman et al., 2010
; Tiacci et al., 2011
). Recently, ATP-competitive inhibitors of RAF kinase have been shown to be extremely effective in the treatment of melanomas with mutant BRAF (Chapman et al., 2011
). This is thought to occur because these drugs inhibit ERK signaling only in tumors with mutant BRAF, whereas they induce ERK in other tumors and normal cells (Hatzivassiliou et al., 2010
; Heidorn et al., 2010
; Joseph et al., 2010
; Poulikakos et al., 2010
Induction occurs because RAF inhibitors cause transactivation of Ras-dependent RAF dimers (Poulikakos et al., 2010
). However, BRAFV600E
signals as a functional monomer and RAF inhibitors inhibit ERK signaling in this setting. We now show that Ras activity is extremely low in BRAFV600E
melanomas. This finding confirms that BRAFV600E
functions in a Ras-independent fashion in these cells. The questions arising now are why Ras activity is low, and whether there a causal relationship that explains why a RAF mutant that signals as a monomer is prevalent in tumor cells with low Ras activity? It is possible that physiologic levels of Ras-GTP are low in the normal precursor cells from which melanomas develop. RAF mutants that require Ras dependent dimerization would have low activity in these cells and there would be a strong selection for a RAF mutant capable of signaling as a monomer. Alternatively, ERK activation induces feedback inhibition of upstream signaling, which may be sufficient to potently suppress Ras activation. Here we have demonstrated the latter to be the case. Inhibition of ERK signaling with either RAF or MEK inhibitors significantly induced Ras activation in these tumors.
This induction is likely multifactorial with contributions from the various components of ERK feedback, such as direct phosphorylation of SOS and EGFR, as well as overexpression of Spry. Here we show that knockdown of Spry in BRAFV600E cells increased Ras and RAF activation, and decreased the sensitivity of the pathway to RAF inhibitors. Spry proteins, however, do not affect the direct inhibition of SOS and CRAF by ERK, and therefore, even though Spry knockdown enables signaling from RTKs to SOS, loss of Spry alone cannot account for the full effect of ERK-dependent feedback.
Because physiologic activation of ERK is self-limited in extent and duration (Courtois-Cox et al., 2006
), one may ask how oncoproteins cause sufficient activation of ERK output at all? We believe that activation of ERK output requires selection of oncoproteins that have decreased sensitivity to feedback, or second mutations that inactivate the feedback apparatus. In fact, we have previously shown that whereas ERK transcriptional output is quite elevated in tumors with mutant BRAF or mutant Ras; it is only marginally elevated in tumor cells with mutant EGFR or amplified HER2 (Pratilas et al., 2009
). In these tumors, ERK pathway feedback is intact and levels of Ras activation are low. In contrast, the mutant Ras protein is constitutively activated (McCormick, 1993
) and it is thus refractory to feedback inhibition of upstream signaling.
We propose that there is a powerful selection for the BRAFV600E mutation because it signals as a Ras-independent monomer that is insensitive to feedback. This results in marked elevation of ERK output, with consequent feedback inhibition of Ras-GTP. In agreement with this idea, inhibition of ERK signaling relieves this feedback, and causes induction of Ras activation. Ras activation is associated with a rebound in ERK phosphorylation and output. This rebound is Ras- and SOS-dependent, and more importantly, is CRAF-dependent. Therefore, while the rebound may be potentiated by the loss of ERK phosphatases following RAF inhibition, these findings are consistent with the idea that rebound requires reactivation of upstream signaling and induction of RAF dimers that are refractory to RAF inhibitors but sensitive to MEK inhibition.
If RAF inhibitors cause the Ras-dependent formation of active RAF dimers that are refractory to RAF inhibition, why do these drugs work at all? The induction of Ras-GTP is variable in different melanoma cell lines. It tends to be modest, however, reaching levels that are still significantly below those found in RTK-driven tumor cells. This results in a concomitant modest increase in ERK phosphorylation and in ERK output. In most melanomas, this reactivation is not sufficient to cause resistance. We believe, however, that it can attenuate the effects of therapy, as we find that combining RAF inhibitor with low dose MEK inhibitor causes greater inhibition of pERK and ERK output than either drug alone, and enhanced antitumor activity in vivo in melanoma xenograft models. Thus, the variability observed in the degree of BRAFV600E melanoma response in patients treated with RAF inhibitors may be due in part to variable relief of feedback. This suggests that combined inhibition could increase the degree or duration of response obtained with RAF inhibition alone.
Others have noted that ERK rebound is greater in BRAFV600E
thyroid (J. Fagin, unpublished data) and colon (Corcoran et al., 2012
) carcinomas and is associated with resistance to the RAF inhibitor. Recent studies show that rebound in colorectal tumors may be associated with feedback reactivation of EGFR function (Corcoran et al., 2012
; Prahallad et al., 2012
). This may explain why RAF inhibitors have been much less effective in the treatment of BRAFV600E
colorectal cancer than they are in melanoma.
Prahallad et al. report that RAF inhibitors induce EGFR activation by inhibiting the ERK-dependent CDC25C phosphatase and thus activating EGFR signaling in colorectal cancer cells (Prahallad et al., 2012
). Our data suggest that ERK dependent feedback is complex and that relief of feedback and rebound in ERK activity is due to multiple mechanisms. In melanomas, we did not observe an association between ERK rebound and sustained induction of EGFR phosphorylation. Corcoran et al. also demonstrated that ERK phosphorylation rapidly rebounds after initial inhibition by RAF inhibitors in colorectal cancer (Corcoran et al., 2012
). They also find that this rebound is EGFR dependent and associated with Ras activation, but not with induction of EGFR phosphorylation. Here, we demonstrate that relief of ERK dependent feedback by RAF inhibitors results in Ras activation, induction of CRAF-containing dimers, and RAF inhibitor-resistant ERK rebound. In contrast to our findings, Corcoran et al. do not observe Ras reactivation or ERK rebound in melanomas. This is probably because the degree of rebound is greater in colorectal cancer than it is in melanoma, in which it is more difficult to appreciate. We believe that potent ERK-dependent feedback inhibition of signaling is a general phenomenon in tumors with BRAFV600E
and that the antitumor effects of drugs that inhibit ERK signaling will be diminished by relief of this feedback.
It is clear that the degree of rebound varies among individual tumors within lineages and that the rebound is greater on the average in some lineages (e.g. colorectal), than in others (e.g. melanoma). Although it is unlikely that this is a simple process dependent on reactivation of a single receptor, it appears that the process may be preferentially dependent on activation of a particular receptor in some lineages (e.g. EGFR in colorectal carcinoma (Corcoran et al., 2012
; Prahallad et al., 2012
)). Our findings show that signaling from many receptors is suppressed by ERK-dependent feedback in melanomas and reactivated when feedback is relieved by ERK inhibition. It must be kept in mind that as receptor activation of ERK increases, feedback increases and receptor signaling declines. Each tumor reaches a new steady state of ERK activity after RAF inhibition that must be dependent on the level of ERK output required to induce feedback. If feedback mechanisms are sensitive to induction by low levels of ERK output, rebound will be modest. If high levels of ERK output are required to reinitiate feedback, marked ERK rebound will occur and the tumor will be resistant. Future progress will depend on determining the lineage-dependent and tumor specific factors responsible for the new steady state.
Our data show that BRAFV600E melanomas are characterized by high levels of ERK-dependent feedback that operates globally to regulate oncogenic signaling. These cells have markedly decreased sensitivity to extracellular ligands. Indeed, the transduction of signals from activated RTKs, a cellular property that we have termed `signalability’, is markedly suppressed in BRAFV600E melanomas. After ERK inhibition, however, the ERK-dependent negative feedback is lost; and the ability of ligands to activate signaling is markedly enhanced. This is our key finding: at baseline these tumors are relatively insensitive to the effects of secreted growth factors, because the ability of such ligands to induce signaling is disabled. After administration of drugs that effectively inhibit ERK signaling, feedback is reduced and growth factors can signal. Thus, they may attenuate or prevent the antitumor effects of the inhibitor. The signaling network is radically changed and reactivated as an adaptation to inhibition of ERK signaling ().
Recently several reports have shown that ligands, particularly HGF, can cause resistance to RAF inhibitors (Straussman et al., 2012
; Wilson et al., 2012
). Induction of signalability when ERK-dependent feedback is relieved requires the presence of active RTKs. We show here that multiple ligands contribute to ERK rebound in melanomas exposed to RAF inhibitors. However, receptor activation is permissive for induction of signalability, i.e. necessary, but not sufficient. Rebound in ERK signaling is due to relief of feedback inhibition of signal transduction when ERK activation is inhibited.
In order to understand how the tumor adapts to pathway inhibition and design more effective therapies, it will be necessary to identify the pathways that become reactivated in patients, as it is not clear that preclinical models are useful in this regard. This will require comparison of pre-treatment biopsies with biopsies obtained hours after treatment and the development of new technologies to determine which ligands are present and which pathways have become reactivated. This will allow the development of rational combination therapies aimed at inhibiting the adaptation of the tumor to the targeted therapy.