BRAF inhibitors suppress apoptosis through off-target inhibition of JNK signaling

Vemurafenib and dabrafenib selectively inhibit the v-Raf murine sarcoma viral oncogene homolog B1 (BRAF) kinase, resulting in high response rates and increased survival in melanoma. Approximately 22% of individuals treated with vemurafenib develop cutaneous squamous cell carcinoma (cSCC) during therapy. The prevailing explanation for this is drug-induced paradoxical ERK activation, resulting in hyperproliferation. Here we show an unexpected and novel effect of vemurafenib/PLX4720 in suppressing apoptosis through the inhibition of multiple off-target kinases upstream of c-Jun N-terminal kinase (JNK), principally ZAK. JNK signaling is suppressed in multiple contexts, including in cSCC of vemurafenib-treated patients, as well as in mice. Expression of a mutant ZAK that cannot be inhibited reverses the suppression of JNK activation and apoptosis. Our results implicate suppression of JNK-dependent apoptosis as a significant, independent mechanism that cooperates with paradoxical ERK activation to induce cSCC, suggesting broad implications for understanding toxicities associated with BRAF inhibitors and for their use in combination therapies. DOI: http://dx.doi.org/10.7554/eLife.00969.001

Because of the rapid development of these cSCC on BRAFi therapy and the enrichment for RAS mutations, pre-existing genetic lesions are likely present prior to therapy, which are then 'unmasked' following initiation of BRAFi therapy. The fact that many arise in sun-damaged skin suggests that prior chronic UV exposure is an important predisposing event .
We instead hypothesized that vemurafenib and PLX4720 could also affect the susceptibility of cells to apoptosis and in so doing, contribute to the acceleration of tumor development. We studied the acute ultraviolet radiation (UVR) response because this is the most important environmental risk factor in the development of skin cancer and because many BRAFi-induced cSCC arise in sun-damaged areas . PLX4720 and vemurafenib share structural features (Tsai et al., 2008;Bollag et al., 2010) and have similar activities, as is the case in our studies.

BRAFi suppress stress-induced, JNK-dependent apoptosis
We performed our initial studies using cSCC (SRB1, SRB12, COLO16) and keratinocyte (HaCaT) cell lines. Cells treated with 1 kJ/m 2 of UVB (FS40 lamp) undergo apoptosis within 24 hr ( Figure 1A-D). eLife digest Over 50% of melanomas, a highly lethal form of skin cancer, carry mutations in a gene called BRAF. The BRAF gene encodes an enzyme that helps to regulate the proliferation of cells, but mutations in this gene lead to the excessive proliferation that is seen in cancer. Clinical trials have shown that a drug called vemurafenib can be used to treat patients who carry the mutated BRAF genes and go on to develop melanoma, but around one fifth of these patients developed another type of skin cancer called cSCC (cutaneous squamous cell carcinoma).
The cSCC tumors often develop in areas where the sun has damaged the patient's skin, and it is thought that their growth is then accelerated by vemurafenib activating another enzyme, ERK, which causes the excessive proliferation of skin cells. Vin et al. have now found that vemurafenib might also cause cSCC tumors by blocking another signaling pathway. The experiments were performed in human cells and also in mice, and the results were then verified in human cSCC samples.
Cells that are exposed to UV radiation usually die, but when treated with vemurafenib, some 70% of the cells that would have died instead survived. The stress from the UV radiation activates the JNK signaling pathway, which causes the irradiated cells to die. However, Vin et al. found that cSCC cells had very low levels of JNK signaling because treatment with vemurafenib had the unintended effect of inhibiting three enzymes that are needed to fully activate the JNK signaling pathway.
Vin et al. estimate that suppression of JNK signaling and cell death is responsible for about 17.6 to 40% of the effect on cSCC growth seen in melanoma patients, with activation of the ERK pathway accounting for the rest. These unexpected findings suggest that combining vemurafenib treatment with radiation or chemotherapy should be done with caution as these effects could affect their efficacy. It also suggests that future drugs should be designed in a way that avoids these types of effects by making sure they do not inhibit important 'off-target' enzymes. Surprisingly, this apoptosis was suppressed by at least 70% in cells concomitantly treated with 1 μM PLX4720 ( Figure 1A-D) compared to control DMSO-treated cells as measured by FACS for Annexin V+; TMRE (tetramethylrhodamine)-low cells ( Figure 1E, Figure 1-figure supplement 1A-C). Similar results were obtained using doxorubicin as the inducer of apoptosis, and similar suppression of apoptosis was obtained using 1 μM PLX4720 in all cells (Figure 1-figure supplement 2A,B). Importantly, these cells have no oncogenic RAS or BRAF mutations (Table 1), and PLX4720 conferred no significant proliferative advantage to the tested cells ( Figure 1-figure supplement 3) even when used at concentrations that inhibit the proliferation of BRAF V600E melanoma cell lines (Tsai et al., 2008).
Because the p38 and JNK stress-activated MAP kinases are well-established critical mediators of UV-induced apoptosis (Derijard et al., 1994;Chen et al., 1996;Tournier et al., 2000;Hildesheim et al., 2004), we explored the status of JNK and p38 activation by assessing phospho-JNK and phospho-p38 levels by Western blot ( Figure 1F). Phospho-JNK levels in particular were highly upregulated upon UV irradiation and were significantly suppressed by treatment post-radiation with 1 μM PLX4720 in cSCC and HaCaT cell lines ( Figure 1F). Similar effects were seen with 1 μM vemurafenib (data not shown) and in cells stressed with doxorubicin ( Figure 1-figure supplement 2C). Importantly, ERK signaling remained intact, as evidenced both by the paradoxical activation of ERK (upregulation of phospho-ERK) and by the failure to upregulate BIM levels ( Figure 1F,G). This pro-apoptotic BCL2 family member is upregulated by inhibition of ERK signaling (Collins et al., 2005) and in BRAF V600E melanoma cells treated with vemurafenib (Paraiso et al., 2011). Since NOXA is a downstream effector of UV-induced apoptosis (Naik et al., 2007), we examined its expression and found that NOXA expression is induced by UV irradiation and suppressed by PLX4720 in all cell lines ( Figure 1G), suggesting that inhibition of NOXA expression may be a mechanism of PLX4720-induced suppression of apoptosis. Finally, we examined the expression of the antiapoptotic BCL2 family member MCL1 because it is downregulated by UV exposure ( Figure 1G), but as previously reported (Paraiso et al., 2011), unaffected by PLX4720 ( Figure 1G).
To test the generality of these effects in cells in which ERK activity is suppressed by BRAFi, we extended our analysis to the BRAF V600E melanoma cells A375 and WM35. As expected, phospho-ERK expression was strongly suppressed by PLX4720 ( Figure 1H). Phospho-JNK and phospho-p38 were significantly upregulated following UV-irradiation ( Figure 1H), showing that signaling to JNK and p38 is intact in BRAF V600E melanoma cells. Here again, there was significant suppression of both phospho-p38 and phospho-JNK induction by PLX4720 ( Figure 1H), and similar effects were seen with vemurafenib (data not shown).
We next examined the responses of primary normal human epidermal keratinocytes (NHEKs) to vemurafenib. UV-induced apoptosis was significantly suppressed (approximately 70%) by vemurafenib at least 70% suppression of apoptosis in the presence of PLX4720 as measured by FACS for Annexin V+, TMRE-low cells (n = 6 for each cell line, '*' denotes statistical significance at p<0.05). (E) A representative FACS plot for COLO16 is shown. Annexin V+, TMRE-low cells are contained in the upper left quadrant (boxed), which was significantly populated in UV-irradiated cells, but not in the absence of UV, or in the presence of PLX4720. (F) Western blots probed for the MAP kinases demonstrated strong phospho-JNK and phospho-p38 induction following irradiation and significant suppression by PLX4720. Phospho-ERK was slightly induced following irradiation, and at 24 hr, paradoxical hyperactivation in the presence of PLX4720 was observed, particularly in SRB1 and HaCaT cells. (G) Western blots showed that BIM was not upregulated in these BRAF-wild-type cells, consistent with intact ERK signaling. MCL1 was downregulated by irradiation and not modulated by PLX4720, whereas NOXA expression was strongly induced in irradiated cells and suppressed by PLX4720. (H) Western blots of BRAF V600E melanoma cell lines, A375 and WM35, demonstrated suppression of UV-mediated induction of phospho-JNK and phospho-p38 by PLX4720 at 24 hr. As expected, phospho-ERK is shut down in PLX4720-treated cells. DOI: 10.7554/eLife.00969.003 The following figure supplements are available for figure 1:   Figure 1-figure supplement 1D), and the UV-induced upregulation of phospho-JNK and phospho-p38 was likewise suppressed most significantly at 6 and 24 hr ( Figure 2B). As in the cSCC and HaCaT cell lines, activation of ERK was observed following exposure to vemurafenib ( Figure 2B). The presence of cleaved caspase-3 correlated with high levels of apoptosis in the UV-treated cells and its absence with rescue by vemurafenib at 24 hr post-irradiation ( Figure 2C). In probing members of the BCL2 family, we found similar results to those in the cSCC and HaCaT cell lines. BIM and MCL1 were unaffected by vemurafenib but NOXA induction at 24 hr post-UV irradiation was diminished by vemurafenib ( Figure 2C). The advantage of using primary cells is that p53 is intact. In NHEKs, p53 is stabilized by 24 hr post-UV irradiation and this is unaffected by vemurafenib ( Figure 2C). However, since BCL2 family members can be modulated by JNK (Tournier et al., 2000;Haeusgen et al., 2011) and p53 (Oda et al., 2000) in apoptosis, the inhibition of NOXA expression by PLX4720 and vemurafenib ( Figures 1G and 2C) likely reflects p53-independent regulation of NOXA given that  We conclude from our results that vemurafenib and PLX4720 suppress UV-induced apoptosis by inhibiting JNK signaling and NOXA induction in BRAF and RAS WT cells.
BRAFi suppress JNK activity through off-target inhibition of ZAK, MKK4, MAP4K5 Although BRAFi-induced JNK inhibition is observed in BRAF-WT as well as BRAF V600E cells ( Figure 1F,H, 2B), with opposite effects on ERK signaling, we sought to further demonstrate that JNK inhibition and paradoxical ERK activation are independent and separable. We treated SRB1 and NHEK cells with the MEK inhibitor (MEKi) AZD6244 and PLX4720 singly and in combination with and without UV irradiation. While MEKi effectively abrogated ERK phosphorylation and activation ( Figure 2D,E), this left PLX4720mediated suppression of UV-induced JNK activation ( Figure 2D,E) and apoptosis ( Figure 2F,G) unaffected in both SRB1 and NHEK cells. Because JNK and p38 isoforms are not significantly inhibited by PLX4720 or vemurafenib directly, (Tsai et al., 2008;Bollag et al., 2010) we probed the phosphorylation status of MKK4 and MKK7 (MAP2K7), the two proximal kinases that synergistically phosphorylate JNK and that are required for JNK activation (Tournier et al., 2001;Haeusgen et al., 2011). The phosphorylation of both MKK4 and MKK7, corresponding to their activation, was significantly upregulated in control UV-irradiated cells and inhibited by PLX4720 in all cSCC cell lines, HaCaT, and NHEK cells ( Figure 2H,I).
We then performed a kinome screen of PLX4720 and vemurafenib against a panel of 38 kinases reported to be upstream of JNK (Keshet and Seger, 2010;Haeusgen et al., 2011) and other kinases previously tested against PLX4720 and vemurafenib (Tsai et al., 2008;Bollag et al., 2010) using a quantitative competitive binding assay  at four concentrations (50 nM, 200 nM, 1 μM, 10 μM). We extended previously reported results obtained on this platform  by testing a wider concentration range and by additionally testing vemurafenib. Reported biochemical analysis and protein extracts 24 hr later. (A) Apoptosis was significantly suppressed (70%) in the presence of vemurafenib as measured by FACS for Annexin V+, TMRE-low cells (n = 6, '*' denotes statistical significance at p<0.05). (B) Western blot analysis showed induction of phospho-JNK and phospho-p38 within 1 hr following irradiation, which persisted for at least 24 hr and which was suppressed by vemurafenib at all time points. (C) MCL1 and BIM expression was not significantly modulated by vemurafenib; however, NOXA induction, which occurred at 24 hr, was reduced by vemurafenib. In these primary cells, p53 protein was stabilized by 24 hr and vemurafenib did not affect overall levels. Suppression of apoptosis, as measured by cleaved caspase-3 levels, was observed in the presence of vemurafenib-treated irradiated cells, consistent with the FACS results. To test the relevance of MEK signaling, cSCC (SRB1) and NHEK cells were irradiated with 1 kJ/m 2 of UVB in the absence ('o', 1:2000 DMSO) or presence of 1 μM PLX4720 singly or in combination with 0.6 μM (NHEK) or 1.2 μM (SRB1) AZD6244 (MEKi) and isolated for FACS analysis and protein extracts 24 hr later. (D) SRB1 and (E) NHEK cells showed induction of phospho-JNK at 24 hr following irradiation, by Western in the presence (lane 7) and absence (lane 5) of MEKi. The addition of MEKi to PLX4720 did not affect the suppression of JNK activation (compare lanes 6, 8) despite potent suppression of phospho-ERK. (F) SRB1 and (G) NHEK cells exhibited a strong suppression of UV-induced apoptosis by PLX4720 (Annexin V+, TMRE-low cells; n = 6, '*' denotes statistical significance at p<0.05) that was likewise unaffected by the addition of MEKi. To test whether upstream kinases in the JNK pathway were inhibited, MKK4 and MKK7 activation was probed in cells. (H) Both phospho-MKK4 and phospho-MKK7 were induced in HaCaT and NHEK cells following irradiation, and this was suppressed in the presence of 1 μM PLX4720 and vemurafenib, respectively. (I) In all cSCC cell lines, SRB1, SRB12, COLO16, phospho-MKK4 and phospho-MKK7 are strongly induced following irradiation, and this is suppressed in all lines by 1 μM PLX4720. DOI: 10.7554/eLife.00969.008 The following figure supplements are available for figure 2:  IC50s for vemurafenib  and PLX4720 (Tsai et al., 2008) against multiple kinases including BRAF V600E , MAP4K5, SRMS, and BRK were quantitatively similar to the estimated K d , confirming the validity of this assay (Tables 2 and 3). We confirmed that ZAK and MKK4 (MAP2K4) have high binding affinities comparable to that of the intended target, BRAF (estimated K d below 50 nM) for both PLX4720  and vemurafenib, and confirmed activity against MAP4K5   (Tables 2 and 3). To demonstrate an effect on activity, in vitro kinase assays were performed ( Figure 3A-C) and revealed biochemical IC50s of 187 ± 5 nM, 460 ± 41 nM, and 354 ± 26 nM for ZAK, MKK4, and MAP4K5, respectively. All of these values are within the range of reported correspondences between binding assays and activity-based assays and with reported data (Anastassiadis et al., 2011;Davis et al., 2011). Importantly, at 1 μM vemurafenib used in our experiments, the residual activity of ZAK, MKK4, and MAP4K5 kinases, was 18.9 ± 0.5%, 29.6 ± 1.1%, and 25.7 ± 0.6%, respectively.
To examine the requirements for ZAK, MAP4K5, and MKK4 in activating JNK activation and apoptosis more directly, we performed lentiviral shRNA knockdown experiments in HaCaT cells. HaCaT cells with knockdown of ZAK ('shZAK2') showed a strong suppression of ZAK protein expression ( As knockdown of ZAK alone can account for up to 93.7% of the effect of PLX4720 treatment, we conclude that the potent inhibition of JNK activation and resultant apoptosis by PLX4720 and vemurafenib is due to the off-target inhibition of ZAK principally, with smaller additional contributions from inhibition of MKK4 and MAP4K5, which abrogates the activation of the two kinases essential for JNK phosphorylation and activation: MKK4 and MKK7 ( Figures 2H-I,3D-E, Figure 3-figure supplement 4). Consistent with our findings, ZAK has been shown to be critically important for JNK activation upstream of MKK4 and MKK7 (Wang et al., 2005) and doxorubicin-induced apoptosis (Sauter et al., 2010;Wong et al., 2013).

Expression of gatekeeper mutant ZAK reverses BRAFi-mediated apoptosis suppression
Our biochemical and shRNA data showed that knockdown of ZAK suppressed phospho-JNK activation and apoptosis ( Figure 3D-E, Figure 3-figure supplements 1,3) and that the degree of knockdown correlated with the degree of JNK and apoptosis suppression. To show that PLX4720 suppresses apoptosis primarily through direct action on ZAK in cells, we employed a chemical-genetic approach by engineering a gatekeeper mutant ZAK (T82Q). Gatekeeper mutant kinases, in which the threonine (T) is replaced by a larger amino acid, in our case glutamine (Q), are often rendered insensitive to small molecule inhibitors and are an important mechanism of drug resistance (Daub et al., 2004;Whittaker et al., 2010). We overexpressed equivalent amounts of ZAK (T82Q) wild-type ZAK (WT) in HaCaT cells ( Figure 3F), and compared their UV responses.

Vemurafenib suppresses JNK activity and apoptosis in cSCC arising in treated patients
We then explored whether vemurafenib or PLX4720-mediated suppression of JNK and apoptosis is relevant in vivo. We first examined cSCC arising in patients treated with vemurafenib and compared them to sporadic cSCC that were histologically similar, arising in individuals never treated with vemurafenib ( Figure 4A-E). Phospho-JNK and cleaved caspase-3 expression were assessed by immunohistochemistry and then quantified following normalization by unit area (mm 2 ) of tumor tissue (malignant keratinocytes) only ( Figure 4A-D, Figure 4-figure supplement 1). Sporadic cSCC arising in patients never treated with vemurafenib (n = 15) contained substantially greater expression of phospho-JNK (p=0.013; Figure 4A,E) and cleaved caspase-3 (p=0.042; Figure 4C,E) as compared to lesions arising in vemurafenib-treated patients (n = 16; Figure 4B,D,E). Therefore, we found significant reductions in phospho-JNK and cleaved caspase-3 expression in human cSCC suggesting that suppression of JNK activity and apoptosis occur in vivo in patients treated with vemurafenib.

BRAFi suppress acute UVR-induced epidermal apoptosis in vivo
We then probed the acute, in vivo, short-term UV response in skin by pre-treating C57BL/6 mice with PLX4720 administered by oral gavage 40-80 mg/kg twice a day for 2-4 days (Tsai et al., 2008). Following depilation, mice were irradiated once using a solar simulator (Oriel) with 10 kJ/m 2 UVB. The skin was harvested at 1 hr, 6 hr, and 24 hr post-irradiation. Consistent with our other results, we found significant apoptosis of epidermal keratinocytes in irradiated mouse skin that was suppressed by PLX4720 treatment (12.7 ± 0.4 apoptotic cells/mm vs 4.9 ± 0.3 apoptotic cells/mm skin, [n = 3 pairs], p<10 −5 ; Figure 4F,G), a finding corroborated by cleaved caspase-3 levels, which were induced within 6 hr of irradiation and suppressed in PLX4720-treated mice ( Figure 4H). As expected, phospho-JNK and phospho-p38 were significantly upregulated following UV irradiation and phospho-JNK was significantly suppressed by PLX4720 ( Figure 4H). The upstream kinases MKK4 and MKK7 were likewise activated by UV radiation and suppressed by PLX4720 at 1 and 6 hr post-irradiation, confirming the importance of this mechanism of PLX4720-induced JNK signaling suppression in vivo. Finally, Noxa mRNA expression, as measured by qPCR, was strongly induced by UV exposure, a response significantly dampened by PLX4720 treatment ( Figure 4I).

BRAFi accelerates UVR-driven cSCC development in Hairless mice
We also used the Hairless mouse model of squamous cell carcinoma to assess whether PLX4720 would affect UV-driven tumor development. This is particularly relevant since it appears that UV exposure is   an important initiating event in BRAFi-accelerated cSCC . Unlike the DMBA/TPA model, in which lesions almost universally harbor Hras mutations (Brown et al., 1990), the Hairless model has a very low frequency of Ras mutation in papillomas and carcinomas (van Kranen et al., 1995), more similar to sporadic human cSCC. The cohorts (n = 5 each) were identically irradiated thrice weekly (12.5 kJ/m 2 per week UVB) for 72 days before starting on PLX4720 treatment vs vehicle control. Within 20 days of administration of drug, hyperkeratotic papules were visible on the backs of PLX4720treated animals ( Figure 5A,B), which steadily grew into cSCC over the following several weeks ( Figure  5C,D). Within this period of 150 days (78 days of drug treatment), control-treated mice had not yet developed any visible lesions ( Figure 5E). When we quantified the effects of each of these drug treatments, we found significant decreases in both phospho-JNK expression (p=0.046; Figure 5F,G,J) and cleaved caspase 3 expression (p=0.019; Figure 5H-J) in PLX4720-treated mice as compared to control-treated mice. Importantly, we sequenced the entire coding regions for Ras (Hras, Kras, Nras) and found no mutations in any of the tumors in PLX4720-treated mice, as compared to one of 14 papillomas and carcinomas in a cohort of control-treated chronically-irradiated Hairless mice ( Figure 5-figure  supplement 1).

Paradoxical ERK activation and off-target JNK inhibition cooperate to accelerate tumor growth
While the effects of these BRAFi on JNK-dependent apoptosis is clear and independent of ERK activity, the relative contribution of paradoxical ERK activation vs JNK pathway inhibition to tumorigenesis has not been precisely addressed ( Figure 6A). To accomplish this, we took advantage of the fact that paradoxical ERK activation requires intact CRAF (Hatzivassiliou et al., 2010;Heidorn et al., 2010;Poulikakos et al., 2010) ( Figure 6A). We used isogenic, matched WT and Craf-deficient (Craf−/−) mouse embryonic fibroblasts (MEFs) and transformed them with adenovirus E1A and human HRAS G12V to enable anchorage-independent growth ( Figure 6-figure supplement 1). These Craf−/− cells do not exhibit strong paradoxical MEK or ERK activation, consistent with previous reports (Poulikakos et al., 2010) (Figure 6-figure supplement 1A). Wild-type and matched Craf-deficient MEFs were plated in soft agar assays  and treated with PLX4720. Both WT and Craf-deficient MEFs exhibited a significant colony formation advantage in the presence of drug ( Figure 6B-D). Based upon this analysis, we estimated that the effect of paradoxical ERK activation to be 60% and other effects, including inhibition of JNK activity, to account for the rest (40%) of the total colony growth advantage ( Figure 6D). To assess the role of JNK signaling directly, we used HRAS G12V -transformed HaCaT cells with ('TKD') and without ('SCR') triple lentiviral knockdown of ZAK, MAP4K5, and MKK4 ( Figure 3D,E), to perform similar colony formation assays to assess responses to PLX4720 treatment ( Figure 6E). Drug treatment conferred a significant colony formation advantage in both sets of cells, which exhibit equivalent paradoxical ERK activation ( Figure 6-figure supplement 1B). Yet, untreated TKD HaCaT cells produced more colonies than SCR HaCaT cells suggesting that JNK pathway suppression results in an advantage in the absence of drug and paradoxical ERK activation ( Figure 6E). Drug-treated SCR and TKD HaCaT cells, had elevated colony counts to similar levels, as expected, because both lines would experience similar degrees of both paradoxical ERK activation and JNK inhibition, and TKD cells (knocked down for ZAK, MAP4K5, MKK4) are unlikely to experience any further suppression of JNK signaling ( Figure 3D). Based on this, we estimated the effect of JNK pathway inhibition to be 17.6% ( Figure 6E). When combined with the MEF experiment, we estimate that the effect of JNK inhibition contributes approximately 17.6-40% of the total effect of PLX4720accelerated colony formation ( Figure 6D,E). Importantly, although we can quantify these individual contributions, it is clear in many contexts in cancer that hyperproliferation and inhibition of apoptosis are highly cooperative (Hanahan and Weinberg, 2011), and our data do not preclude the possibility that one or both are individually required.

Dabrafenib and vemurafenib differ significantly in their off-target effects and risk of cSCC
The recently updated combination trial of the BRAFi dabrafenib and MEKi trametinib shows a low 7% cSCC rate in 54 patients, (Flaherty et al., 2012) suggesting that combined MEK inhibition can reduce, but not eliminate cSCC formation, nevertheless reinforcing a role for paradoxical ERK activation . However, clinical trial data on dabrafenib alone at 150 mg PO BID, shows an overall aggregated cSCC rate of 6.1% (Falchook et al., 2012;Hauschild et al., 2012;Long et al., 2012) vs 22% The difference between colony formation advantages conferred by 1.0 μM PLX4720 in WT vs Craf−/− MEFs was interpreted to reflect the contribution of paradoxical ERK signaling (red arrow), which depends upon Craf, and is 60% of the total effect (black arrow), with the remainder composed of other effects including JNK inhibition (blue arrow). All differences between each MEF population were significant ('***', p<0.001) (E) The fold-change in colony counts of transformed HaCaT cells with Figure 6. Continued on next page for vemurafenib at 960 mg PO BID in several hundred patients (Flaherty et al., 2010;Chapman et al., 2011;Sosman et al., 2012;Menzies et al., 2013). We interpreted this as being reflective of differences between vemurafenib and dabrafenib, as opposed to unequivocal proof that paradoxical ERK activation is the only mechanism involved. To explore this further, we used similar assays to assess the effects of dabrafenib on apoptosis, JNK signaling, and colony formation. In stark contrast to vemurafenib, dabrafenib has little effect on apoptosis and JNK signaling at doses that are biologically equivalent based upon growth inhibition of BRAF V600E melanoma cells and human pharmacokinetic data (Falchook et al., 2012;Gowrishankar et al., 2012) (Figure 6-figure supplement 2A). Peak serum concentrations of dabrafenib at 150 mg PO BID in humans (Falchook et al., 2012) are over 50-fold lower (1.55 μM) than mean sustained serum levels of vemurafenib (86 μM) at 960 mg PO BID (Flaherty et al., 2010), and the GI 50 for the A375 melanoma cell line is less than 0.01 μM for dabrafenib (Greger et al., 2012) vs 0.50 μM for PLX4720 (Tsai et al., 2008). Even at 0.05 μM, dabrafenib did not significantly impact UV-induced phospho-JNK upregulation or apoptosis in HaCaT and SRB1 cells ( Figure 6-figure supplement 2B-E). We profiled dabrafenib activity against ZAK, MKK4, and MAP4K5, and found that ZAK is a significant off-target kinase for dabrafenib as well, but at 0.01 μM, over 64% of activity is retained ( Figure 6-figure supplement 2F). Neither MKK4 nor MAP4K5 is substantially inhibited by dabrafenib up to 1 μM ( Figure 6-figure supplement 2F).
Using transformed WT and Craf-deficient MEFs in soft agar assays, we also showed that dabrafenib enhanced colony formation in WT MEFs, but not in Craf-deficient MEFs ( Figure 6-figure supplement 3). Our results suggest that while both dabrafenib and vemurafenib cause equivalent paradoxical ERK activation in BRAF-wild-type cells (Figure 6-figure supplement 1A-B), only vemurafenib confers a significant colony formation advantage in Craf-deficient cells that have no significant paradoxical MEK/ERK activation, implicating off-target effects as a key difference between the two drugs with respect to cSCC development (Menzies et al., 2013).

Discussion
We have discovered an unexpected and novel effect of the BRAFi PLX4720 and vemurafenib in inhibiting apoptosis in vitro and in vivo through the ERK-independent suppression of JNK signaling (Tournier et al., 2000). Our studies implicate the off-target binding and inhibition of these compounds to ZAK primarily (Figure 3-figure supplements 1-3), with additional contributions of MKK4 (MAP2K4) and MAP4K5 inhibition, thus implicating inhibition of JNK signaling at all three upstream tiers of MAP kinase signaling (Figure 3-figure supplement 4). Although MKK4 knockdown alone could suppress UV-induced apoptosis and phospho-JNK induction by up to 27.3% (Figure 3-figure supplement 2), this is expected, given that ZAK signals through MKK4 and MKK7 (Gross et al., 2002) (Figure 2H,I,3E,4H) and MKK4 is important (with MKK7) for full JNK activation (Tournier et al., 2001;Haeusgen et al., 2011). Additionally, UV-mediated induction of NOXA is suppressed in cell lines, primary NHEKs, and ('TKD') and without ('SCR') triple lentiviral shRNA knockdown of ZAK, MAP4K5, and MAP2K4, show significant differences between 1.0 μM PLX4720-treated and control-treated conditions ('****', p<10 −10 ). Importantly, untreated TKD cells had a significant advantage over untreated SCR HaCaT cells ('**', p<0.01), which we interpreted to be the contribution of JNK signaling inhibition, of 17.6% (blue arrow). Drug-treated SCR and TKD cells both had a similar degree of total colony formation advantage (averaged as black arrow), as expected, since the TKD cells are not expected to have any additional suppression of JNK signaling in the presence of drug ('NS', p=0.17, Figure 3D). Therefore, the colony counts for these two distinct systems (D and E), when taken together, show that JNK pathway inhibition accounts for approximately 17.6-40% and paradoxical ERK activation accounts for approximately 60-82.4% of the total effects of PLX4720 on tumor growth. DOI: 10.7554/eLife.00969.022 The following figure supplements are available for figure 6: in vivo, indicating that this BCL2 family member may be a critical effector of apoptosis in this context ( Figures 1G and 4I). In chronically-irradiated Hairless mice, development of well-differentiated papillomas and cSCC is substantially accelerated by PLX4720 treatment without the need for Ras mutation and with a dramatic reduction in latency by at least 10 weeks ( Figure 5E, Figure 5-figure  supplement 1).
While there is enrichment for RAS mutations in human cSCC arising in vemurafenib-treated patients vs controls (Oberholzer et al., 2011;, up to 30-40% of these lesions do not have RAS mutations. Our novel mechanism of BRAFi-mediated apoptosis suppression is the off-target inhibition of several kinases in the JNK pathway, which is independent from, and compatible with, paradoxical ERKdependent mechanisms   (Figure 6A). Importantly, our approach ( Figure 6B-E) has not only allowed us to quantify the contribution of the effect on apoptosis (17.6-40%) vs paradoxical ERK activation (60-82.4%), but also shows that the growth advantage conferred by BRAFi in BRAF-WT cells is not accounted for entirely by paradoxical ERK activation. Our results have also shown that there are significant differences between the BRAFi vemurafenib and dabrafenib ( Figure 6-figure  supplements 2,3) with respect to these off-target effects in cells (even though their relative selectivities for BRAF over ZAK are similar) and this may, in part, explain why they differ in rates of cSCC (Flaherty et al., 2010;Chapman et al., 2011;Falchook et al., 2012;Hauschild et al., 2012;Long et al., 2012;Sosman et al., 2012). At present it is unclear why ZAK appears to be a common off-target kinase and whether structural similarities with other kinases may explain this (Sauter et al., 2010;Wong et al., 2013).
Because off-target kinases in the JNK pathway are affected by vemurafenib/PLX4720, one expects that these kinases would be affected in all cells regardless of BRAF status. Indeed, melanoma cells expressing BRAF V600E also exhibit suppression of JNK activity following irradiation ( Figure 1H). However, in BRAF V600E -expressing melanoma cells, the effect of blocking BRAF activity alone clearly dominates, because these cells are exquisitely dependent upon BRAF activity (Tsai et al., 2008). Therefore, although off-target kinases are inhibited, the cellular context of dependence on particular kinases is still highly relevant and likely dictates the outcome.
Our findings suggest a tumor suppressive role for JNK signaling in the context of drug-induced cSCC, though the role of JNK in cancer is highly context-dependent and is partly related to differing functions of the individual isoforms and partial redundancy (Tournier et al., 2000). Nonetheless, there is ample in vivo evidence showing that JNK can function in a tumor suppressive role. Genetically-engineered mice lacking Jnk1 and Jnk2 have increased (She et al., 2002) and decreased (Chen et al., 2001) susceptibility, respectively, to chemical carcinogenesis in skin, though these mice also have opposite defects in epidermal differentiation (Weston et al., 2004). In mouse models, lack of Jnk1/2 activity suppresses Ras-driven tumorigenesis in lung (Cellurale et al., 2011) and promotes it in Ras-driven and Trp53-deficient breast cancer models (Cellurale et al., 2010(Cellurale et al., , 2012. In the context of Pten-deficiency, loss of Jnk1/2 or Mkk4/Mkk7 promotes aggressive prostate adenocarcinoma (Hubner et al., 2012).
Importantly, the effects of JNK on cancer are not always tumor cell autonomous, as JNK activity supports a pro-tumorigenic inflammatory microenvironment in hepatocellular carcinoma .
Our results have important clinical implications and suggest careful consideration of combining certain BRAFi with therapeutic modalities that induce apoptosis such as radiation or chemotherapy, particularly with respect to off-target tissues (keratinocytes in skin). We have shown that off-target inhibition of kinases, even at higher IC50s, can contribute biologically significant effects, particularly if they are in the same pathway. Finally, our results show that kinase inhibitors must be considered in terms of their entire spectrum of activity, which can dramatically affect pathways distinct from those affected by inhibition of the intended target.

Ethics statement
All studies were conducted under institutionally-approved IRB (LAB08-0750) and ACUF (06-09-06332) protocols for the protection of human and animal subjects, respectively.

Flow Cytometry
TMRE (Invitrogen) was used as a measure of mitochondrial membrane potential, Annexin V-FITC or Annexin V-APC (Invitrogen) as a probe for apoptosis, and Sytox Blue (Invitrogen) as an indicator for dead cells. At 24 hr post-irradiation, floating and adherent cells were collected and stained with TMRE, Annexin V and Sytox Blue. Data was collected and analyzed using a flow cytometer (Fortessa, Becton Dickinson) and FlowJo Software (Tree Star). Data were calculated and charts were plotted using GraphPad Prism 5 software.

Western blot analysis
Cell were lysed in standard buffers with protease inhibitors (Roche) and phosphatase inhibitors (Santa Cruz) with extracts run on SDS/polyacrylamide gels and transferred to Immobilon-P transfer membrane (Millipore). Blots were blocked in TBST (10 mM Tris-HCL pH8, 150 mM NaCl, 0.5% Tween) with milk or BSA, probed with primary antibodies, corresponding HRP-conjugated secondary antibodies, and signals detected using ECL kit (Amersham).

Immunohistochemistry and histology
Cutaneous squamous cell carcinomas biopsied from patients treated with or without BRAF-inhibitor were obtained either under clinical trials (Roche) or separate IRB approval (LAB08-0750). Staining levels were quantified by counting positively labeled cells and dividing by the total area of the tumor tissue within each sample. To measure tumor areas, all samples were photographed, tumor cells outlined, and total pixel numbers calculated using included image analysis tools in Adobe Photoshop and standardized to a hemacytometer to convert to mm 2 . To measure apoptosis in irradiated skin, pyknotic or dyskeratotic epidermal keratinocytes were counted and normalized to length (mm) of epidermis.
Kinase activity profiling PLX4720 and vemurafenib were prepared in DMSO and tested in duplicate at four concentrations (50 nM, 200 nM, 1000 nM, 10 μM) against a panel of 38 kinases using a quantitative competitive binding assay (KINOMEscan, San Diego, CA). Average percent inhibition was reported. Estimated K d values were derived by averaging pointwise estimates calculated using a transformed Hill equation at each concentration of drug. In vitro kinase assays were performed using human full-length ZAK (MBP substrate, ATP 2.5 μM), amino acids 33-end MKK4 (JNK1 substrate, ATP 0.1 μM), and full length MAP4K5 (MBP substrate, ATP 10 μM) (Reaction Biology). Assays for BRAF V600E and ASK1 against vemurafenib were run in parallel revealing IC50s of 31.6 ± 2.9 nM for BRAF V600E and no significant inhibition of ASK1, as previously reported .

Lentiviral knockdown experiments
Lentiviral shRNA knockdown was accomplished using standard lentiviral methods using 293T cells and psPAX2/pVSV.G packaging plasmids. shRNA clones against ZAK (clones V2LHS_239842, V3LHS_336769), MKK4 (clones V3LHS_646205, V3LHS_386825A), and MAP4K5 (clones 196277A, 334084), as well as a non-silencing shRNA were obtained from Open Biosystems in the GIPZ vector. Following transduction, cells were puromycin-selected and FACS sorted to obtain cells with high-level suppression. Degree of mRNA suppression was quantified by qPCR using Taqman probes using internally controlled (2-color, same well) GAPDH probes to ensure proper normalization.

ZAK overexpression
ZAK (T82Q) mutant was generated in the pcDNA3 mammalian expression vector. HaCaT cells were electroporated using the Neon transfection system 24 hr prior to irradiation. Transfection efficiencies were estimated to be 70-80% by GFP fluorescence.

Mouse experiments
Wild-type C57BL/6 mice, 5-8 weeks old, were pretreated with PLX4720 in 5% DMSO in 1% methylcellulose for 2-4 days at 40-80 mg/kg twice a day or control 5% DMSO in 1% methylcellulose by oral gavage. The mice were shaved and depilated (Nair) 24 hr prior to irradiation with a solar simulator (Oriel) dosed at 10 kJ/m 2 of UVB. Epidermis was harvested and protein extracts run on Western blots and probed as above. For chronically-irradiated Hairless mice, 3-4 week old males were irradiated thrice weekly for a total weekly dose of 12.5 kJ/m 2 UVB (solar simulator, Oriel). At 72 days, PLX4720 treatment was started using drugimpregnated chow (Plexxikon) with vehicle chow in the control cohort.

Soft agar assays
Following plating of bottom agar (0.6% Bacto Agar) with media and appropriate amounts of drug, 2500 to 10,000 cells per well (transformed WT, Craf−/− MEFs; transformed HaCaT SCR and TKD cells) were embedded in top agar (0.3%) and plated in 24-well plates. Control or drug-treated media was replaced every 48 hr for 4-6 weeks. The plates were stained with 1% crystal violet and colonies counted by bright-field microscopy.

Statistical analysis
All data are represented as means ± SEM. All experiments were performed in triplicate at least. Student's t-test was used for comparison between two groups. p≤0.05 was considered significant.