The Immune Microenvironment Confers Resistance to MAPK Pathway Inhibitors through Macrophage-Derived TNFα

The MAPK signaling pathway consisting of the RAF–MEK–ERK kinases is hyperactivated in up to 90% of melanomas. The dependence of melanoma cells on this activated pathway has been exploited successfully in the clinic by selectively inhibiting the RAF kinase BRAF, which is mutated in approximately 50% of melanomas (1). The efficacy of these inhibitors is limited, however, by the onset of resistance, and in the majority of cases, this occurs through reactivation of the pathway (2, 3). This is currently addressed by inhibiting the pathway further downstream using MEK inhibitors (MEKi) in combination with BRAF inhibitors (BRAFi; ref. 4).

Other forms of resistance that have been described rely on the activation of additional signaling pathways such as signaling downstream of PI3K, which can be targeted by selective inhibition (5). Another intracellular event that can cause innate and acquired resistance is the high expression of survival factors. One such survival factor, which we have previously identified, is the melanocytic-specific transcription factor MITF (6). MITF-dependent resistance is probably due to its central role in regulating multiple survival and antiapoptotic genes (7). Indeed, the MITF target BCL2A1 has been shown to antagonize BRAF inhibition (8). Furthermore, components of the differentiation program that stimulates upregulation of MITF are also involved in MAPK pathway inhibitor resistance (9).

In addition to these endogenous mechanisms of resistance, secreted factors that originate from the stroma can induce resistance. For instance, stromal fibroblast-derived hepatocyte growth factor causes activation of receptor tyrosine kinases that act to reactivate the pathway by signaling through RAS (10). One important microenvironment-derived cytokine is TNFα, which has been described to block apoptosis in BRAF-depleted melanoma cells (11). TNFα can execute protumorigenic activities in melanoma, such as promoting tumor growth, angiogenesis, and invasion (12, 13). Furthermore, vascular progression and a more metastatic melanoma phenotype correlate with increased activity of NF-κB, a transcription factor that, besides other growth factors, cytokines, or chemokines, is activated by TNFα (14–16). In light of these findings, we wanted to study the role of TNFα in melanoma growth and survival, as well as resistance to MAPK pathway–targeted therapy.

Mice expressing Braf^V600E in the melanocyte lineage develop melanomas with a median latency of 12 months (17), but we found that the lack of TNFα in Braf^V600E/Tnfa^−/− mice significantly delayed the median latency by approximately 6 months (Fig. 1A). Furthermore, when we injected melanoma cells derived from Braf^V600E mice [tumor necrosis factor receptor (TNFR)-expressing 4434 cells; Supplementary Fig. S1A] into syngeneic wild-type (WT) or TNFα^−/− mice, the average tumor size in TNFα-deficient mice was severely reduced (Fig. 1B). These data strongly suggested that TNFα is required for the growth of melanoma cells in vivo. Indeed, TNFα stimulated proliferation of 4434 melanoma cells in vitro (Fig. 1C), induced IκB phosphorylation (pIκB), and protected the cells from cell death when they were unable to adhere to the extracellular matrix (Fig. 1D). One of the key regulators of melanoma cell survival and proliferation is the lineage survival factor MITF. We found that TNFα upregulated MITF expression in Braf^V600E mouse melanoma cells, which correlated with reduced caspase-3 cleavage under anoikis conditions (Fig. 1E). TNFα induced IκB phosphorylation (pIκB), and it also increased MITF expression in human BRAF-mutant TNFR-expressing (Supplementary Fig. S1B) melanoma cells, stimulated their growth (not shown), and protected these cells from anoikis (Fig. 1E–G). Importantly, overexpression of MITF alone significantly reduced cell death and caspase-3 cleavage under anoikis conditions (Fig. 1F and G). On the other hand, counteracting the TNFα-mediated MITF upregulation by RNAi abolished the protective effect of TNFα without affecting pIκB (Fig. 1H), suggesting that MITF contributes to TNFα-mediated survival.

To establish the mechanism of TNFα-mediated MITF regulation, we analyzed MITF mRNA expression in different melanoma cell lines. This revealed that TNFα regulates MITF at the transcriptional level (Fig. 2A), which was further confirmed by an MITF promoter analysis (Fig. 2B). Whereas TNFα efficiently activated a −2.3-kb promoter fragment that contains a potential NF-κB binding site at −1870/−1879, it failed to elicit a response from a −1.8-kb promoter fragment that lacked the site, or when the potential site was mutated (Fig. 2B and Supplementary Fig. S2A and S2B). A chromatin immunoprecipitation confirmed that NF-κB/p65 binds to the MITF promoter (Fig. 2C). Although TNFα stimulated IκBα phosphorylation and nuclear translocation of NF-κB/p65 in melanoma cells, basal activation of NF-κB signaling was detectable in the absence of exogenous TNFα (Fig. 2D–F). Inhibition of IKK activity using BMS-345541 (IKKα and IKKβ inhibitor) or SC-514 (IKKβ-specific inhibitor) was able to efficiently block p65 nuclear translocation, led to a reduction in pIκBα, and decreased both protein and mRNA expression of MITF (Fig. 2D–G). This indicates that TNFα and IKK–NF-κB signaling contribute to the regulation of MITF expression in BRAF-mutant melanoma cells. In line with this finding, along with diminished MITF expression, IKK inhibition in BRAF-mutant melanoma cells resulted in reduced CDK2 and BCL2 expression, whereas p27 was upregulated (Fig. 2H). These are well-characterized MITF target genes (7), and using RNAi we confirmed that MITF regulates the expression of these cell-cycle and survival proteins in melanoma cells (Fig. 2I and Supplementary Fig. S2C).

We next wished to identify the source of TNFα expression, and found an average 2- to 5-fold increase in TNFα mRNA throughout a panel of 16 melanoma cell lines compared with normal human melanocytes (NHM; Fig. 3A). However, A375 and WM266-4 cells do not express significant amounts of TNFα, which suggests that the basal IKK/NF-κB activation we observed might be due to other mechanisms such as autocrine signaling through CXCL1, PI3K–AKT signaling, or loss of p16^INK4A (16). Also, Braf^V600E-4434 cells do not express any TNFα (Supplementary Fig. S3A), which is in agreement with the reduced tumor growth in TNFα-deficient mice (see Fig. 1B). We therefore analyzed stromal cells, including fibroblasts, keratinocytes, and also macrophages, as they are a major source of TNFα (18). Macrophages can polarize into the classically activated M1 and the alternatively activated M2 phenotype (19), and these phenotypes can be generated in vitro by differentiating and polarizing monocytic THP-1 cells through treatment with specific cytokines (Supplementary Fig. S3B). We found that both M1 and M2 macrophages were indeed the highest TNFα-expressing cells (Fig. 3A).

In accordance with the TNFα mRNA expression, soluble TNFα was detectable in the medium of M1- and M2-polarized macrophages (Fig. 3B), and treatment of WM266-4 cells with conditioned media from either M1 or M2 macrophages led to increased IκBα phosphorylation and increased MITF expression at the protein and mRNA levels (Fig. 3B and C and Supplementary Fig. S3C). The major driver of the macrophage-induced MITF upregulation was secreted TNFα, as conditioned media no longer induced MITF expression after the addition of a TNFα-blocking antibody (Fig. 3C).

Exposure of melanoma cells to conditioned medium from M1 macrophages for 3 weeks had a slight growth-promoting effect, but growth was suppressed when TNFα action was inhibited by a blocking antibody (Fig. 3D). On the other hand, M2 macrophage–derived conditioned medium stimulated growth (Fig. 3D). However, depletion of TNFα using a blocking antibody significantly reduced this growth-promoting effect (Fig. 3D). Importantly, similar results were obtained when using human peripheral blood monocyte–derived macrophages (Fig. 3E and Supplementary Fig. S3D). On the other hand, keratinocytes and fibroblasts, which express 5- to 10-fold more TNFα than melanoma cells, but 10- to 80-fold less than macrophages (Fig. 3A), did not support melanoma cell growth in a TNFα-dependent manner (Supplementary Fig. S3E).

Macrophage recruitment to melanoma is well documented and has been linked to UV-induced melanomagenesis in mice (20). Using publicly available gene-expression datasets (21, 22), we found that the expression of macrophage markers was significantly upregulated during melanoma progression (Supplementary Fig. S4A–S4C), indicating the availability of this potential TNFα source in the tumor microenvironment.

To assess the importance of macrophage-derived TNFα for melanoma growth in vivo, we used LysM-Cre/Tnfα^F/F mice, in which Cre-mediated recombination results in the loss of TNFα expression in the myeloid cell lineage (23). Remarkably, the conditional ablation of TNFα resulted in significant growth retardation of Braf^V600E-4434 melanoma allografts (Fig. 3F). When we analyzed the tumors for the presence of macrophages using a pan–macrophage anti-CD68 antibody, we found that the presence of CD68-positive cells within the tumor was significantly reduced in LysM-Cre/Tnfα^F/F mice, which was accompanied by significantly decreased TNFα expression (Fig. 3G).

We have previously shown that elevated MITF expression provides resistance to MEKi-induced cell death (6). Because TNFα induces MITF expression, it was not surprising to see that it suppressed MEKi-induced caspase-3 cleavage (Fig. 4A). Importantly, this response was dependent on MITF, because MITF depletion through RNAi resulted in loss of the TNFα-mediated protective effect (Fig. 4A).

We then assessed whether macrophages can protect melanoma cells from MEKi-induced apoptosis. For this, we cocultured melanoma cells with macrophages using a Transwell technique (Fig. 4B). This approach enabled the isolation of both melanoma cells and macrophages for separate analysis, and also excluded any macrophage-derived phagocytic activity. We found that the presence of either M1 or M2 macrophages significantly protected melanoma cells from MEKi exposure (Fig. 4C and Supplementary Fig. S5A). Analysis of TNFα expression and secretion by the macrophages showed no change when treated with MEKi (Fig. 4D and Supplementary Fig. S5B), confirming that TNFα was available for the melanoma cells under these conditions. Moreover, MITF expression in the melanoma cells was elevated in the presence of macrophages even in the presence of MEKi (Supplementary Fig. S5C). Most importantly, however, when a TNFα-blocking antibody was added during drug treatment, the protective function of macrophages toward MEKi-induced cell death was lost (Fig. 4E).

We next wanted to assess the effect of MEKi treatment on the presence of macrophages in the tumor environment in vivo. Histologic sections of Braf^V600E-4434 melanoma allografts from MEKi-treated immunocompetent mice showed increased staining for the pan–macrophage marker CD68 when compared with tumors from vehicle-treated mice (Fig. 7C). This increase in CD68 expression was confirmed at the mRNA level (Fig. 5A and Supplementary Fig. S6A), indicating that MEKi treatment enhances macrophage accumulation within the tumor microenvironment. We also found significantly increased expression of the monocyte and macrophage marker F4/80 (not shown), the M1 macrophage marker CD86, the M2 marker CD163, and TNFα in tumors from MEKi-treated over vehicle-treated mice (Fig. 5A and Supplementary Fig. S6A).

To validate the relevance of these findings for targeted therapy in melanoma, we examined paired BRAF^V600E-positive tumor biopsies from 11 patients before treatment, and after 10 to 14 days of treatment with either BRAFi alone or a BRAFi–MEKi combination (for detailed patient data, see Supplementary Table S1).

We found a significant increase in the macrophage marker CD68 (Fig. 5B–D), as well as the M1 marker CD86 and the M2 marker CD163, in all patients in response to the treatment with BRAFi and MEKi (Fig. 5D and Supplementary Fig. S6B), indicating an accumulation of M1- and M2-polarized macrophages. Moreover, TNFα expression was upregulated in response to treatment (Fig. 5D and Supplementary Fig. S6B), and the increase in TNFα expression significantly correlated with enhanced MITF expression in the tumors (Pearson correlation: r = 0.717, P = 0.019; Fig. 5E), supporting the idea that TNFα contributes to MITF expression in these patients during drug treatment. Importantly, there was no difference between patients on BRAFi monotherapy and patients on BRAFi–MEKi combination therapy (Supplementary Fig. S6C), suggesting that inhibition of MEK does not alter the effect of BRAF inhibition on macrophage accumulation in vivo.

Our findings suggest that inhibition of IKK represents a possible strategy to overcome the TNFα/MITF–mediated survival signals that protect melanoma cells from MEK inhibition. We therefore assessed whether IKKi-mediated reduction in MITF expression can synergize with MEKi to induce cell death in melanoma cells. Indeed, at conditions in which neither IKKi nor MEKi treatment alone was able to elicit apoptosis in 501mel cells, the combination of both inhibitors was able to induce cleavage of caspase-3 (Fig. 6A). In line with this finding, IKK inhibition using BMS-345541 in the presence of MEKi reduced the EC₅₀ approximately 26-fold (from 3.45 μmol/L to 132 nmol/L; Fig. 6B). Furthermore, as expected, TNFα protected 501mel melanoma cells from MEKi-induced cell death, but IKK inhibition was able to counteract the protective effect of TNFα by reducing the EC₅₀ approximately 51-fold (from 11.2 μmol/L to 214 nmol/L; Fig. 6B). Similar results were found with another cell line, WM266-4, where the combined treatment led to a dose-dependent reduction in the EC₅₀ of approximately 14-fold at 0.1 μmol/L and approximately 24-fold at 0.25 μmol/L IKKi, respectively (Fig. 6C).

The potentiating effect of IKK inhibition on the efficacy of the MEKi was further confirmed in other melanoma cell lines, including Braf^V600E-4434 melanoma cells (Fig. 6D and Supplementary Fig. S7A–S7C). Moreover, when we analyzed the effect of IKK inhibition on the efficacy of the BRAFi vemurafenib, we found that IKK inhibition significantly synergized with vemurafenib in cell killing (Fig. 6E and Supplementary Fig. S8A and S8B). According to MITF's protective function, there was a trend of a more efficient response to the inhibitors when MITF expression was lower (Fig. 6E and Supplementary Fig. S8C).

To test whether our findings about the combination of MEKi and IKKi in vitro apply to the in vivo situation, we treated Braf^V600E-4434 allograft–bearing immunocompetent mice with MEKi, either alone or in combination with the IKKi BMS-345541. Tumors from mice treated with the inhibitor combination showed significantly reduced growth compared with either the single MEKi or IKKi treatment (Fig. 7A), which is in agreement with the effects of the MEKi–IKKi combination on 4434 cells in vitro (see Fig. 6D) and clearly demonstrates that inhibition of IKK sensitizes melanoma cells to MEK inhibition also in vivo.

To assess the consequences of the MEKi–IKKi combination treatment on the tumor microenvironment, we analyzed the tumors for macrophage markers. This analysis revealed a reduction in the expression of not only the pan–macrophage marker CD68, but also the M1 and M2 macrophage markers CD86 and CD163, in the tumors from mice treated with the MEKi–IKKi combination when compared with the MEKi-and IKKi-only treatment (compare Fig. 7B with Fig. 5A and Supplementary Fig. S9). This finding suggested that the MEKi-induced effect on macrophage numbers is inhibited by the IKKi, which was confirmed when we assessed the presence of CD68-positive cells within the tumors (Fig. 7C and D).

Most strikingly, the expression of TNFα in the tumors was reduced below basal level (=1) when the IKKi was present (Fig. 7B), suggesting that, in addition to decreasing macrophage numbers, IKK inhibition directly affects TNFα expression in the microenvironment, most probably the macrophages. Indeed, when we analyzed the effect of IKK inhibition on TNFα expression in isolated macrophages in vitro, we observed a strong suppression in response to the inhibitor (Fig. 7E). Finally, in correlation with the severely reduced TNFα expression in the MEKi–IKKi-treated tumors (Fig. 7B), MITF mRNA levels had also dropped (Fig. 7F). Thus, inhibition of IKK signaling suppresses not only stromal-derived TNFα levels but also MITF expression in melanoma cells, which creates an advantageous environment to increase the efficacy of MEKi activity (Fig. 7G).

Targeting the MAPK pathway has become a powerful therapeutic approach in melanoma. Nevertheless, the inevitable development of resistance demands further improvement, which could come from combination therapies that tackle the mechanisms contributing to resistance. Furthermore, the combination of targeted approaches with recently developed immune therapies is considered an attractive novel strategy. However, initial attempts indicate that we are yet to completely understand the interplay of the immune microenvironment and targeted therapy in melanoma (24). The full impact of targeted therapy on the immune response is not clear, which challenges the ability to predict how interfering with both simultaneously will affect the overall treatment outcome.

We found that MAPK pathway inhibition directly affects the tumor immune microenvironment by increasing the number of macrophages, and that this can create a source for resistance to BRAFi and MEKi. We identify TNFα as a potentially crucial factor in this resistance due to its ability to enhance the expression of the melanoma survival factor MITF (Fig. 7G).

Although originally identified as an antitumorigenic factor, TNFα and its downstream effectors IKK and NF-κB are now well-accepted players in inflammation-driven tumorigenesis (18, 25). As such, in mice, TNFα is required for skin or liver carcinogenesis (26, 27), and IKK activity is essential for colitis-associated cancer (28). Moreover, the depletion of the IKK subunit IKKβ protects from oncogenic Hras–induced melanoma development in mice (29, 30). We now demonstrate a clear dependence of Braf^V600E-driven melanoma growth on TNFα in vivo, and we show that MITF contributes to survival signaling downstream of TNFα in both mouse and human BRAF-mutant melanoma cells.

Downstream of TNFα, IKK activity is required for the expression of MITF and its target genes CDK2, CDK4, or BCL2. This regulation seems to occur also in vivo, because reduced expression of these target genes is seen in Ras-transformed melanocytes of mice with conditional deletion of Ikkb(30). Thus, although NF-κB can regulate many important cell-cycle and survival genes directly, in melanoma, MITF seems to contribute to this regulation, thereby acting downstream of TNFα.

It is clear that IKKβ and NF-κB are activated in cancer cells, and in melanoma, enhanced NF-κB signaling has been correlated with progression (16). The source of TNFα to stimulate this signaling can be the cancer cells themselves, leading to autocrine TNFα signaling (18). Approximately 50% of the melanoma cells we analyzed displayed increased TNFα expression compared with melanocytes, and this might contribute to enhanced basal NF-κB activation in these cells. On the other hand, paracrine signaling derived from the microenvironment clearly also plays an important role in IKK–NF-κB activation in cancer cells, and TNFα produced by myeloid cells, particularly macrophages, can promote tumor growth in vivo and stimulate tumor cell invasion in vitro (27, 31, 32). We found that TNFα produced by myeloid cells was crucial for melanoma growth in vivo. Although we could recapitulate the growth-promoting effect of macrophages in vitro, this also revealed that TNFα acts in conjunction with other macrophage-derived factors, and it is the overall balance of tumor-promoting and tumor-inhibiting factors that will produce the net effect of growth.

Our in vitro data suggest that TNFα directly acts as a growth and survival factor in melanoma cells, but the reduced number of macrophages within the tumors grown in LysM-Cre/Tnfα^F/F mice indicates that TNFα also affects immune cell recruitment. Such a role for TNFα has been described previously (18), and reduced immune cell recruitment would result in a tumor microenvironment containing fewer tumor-promoting cytokines and hence a less favorable milieu for tumor growth. Importantly, we observe a significant effect on macrophage numbers when we inhibit IKKs, which, as we show, reduces TNFα dramatically. In line with our observations, Ikkb deletion from myeloid cells using LysM-Cre mice in a colitis-associated cancer model results in reduced expression in paracrine-acting cytokines and reduced tumor growth (28).

We found that differentiated macrophages are able to protect melanoma cells from MEKi–induced apoptosis in vitro, and that this protection is dependent on TNFα and MITF. We and others have demonstrated the relevance of MITF in resistance to MAPK pathway inhibitor treatment, i.e., BRAFi and MEKi, in single and combination therapies (6, 8, 9). This MITF-dependent increased survival is probably due to its central role in regulating multiple antiapoptotic genes, such as BCL2, BCL2A1, and ML-IAP (8, 33, 34). We now show that targeting IKKs acts on the cell-autonomous resistance by diminishing MITF expression in melanoma cells, thereby rendering them more sensitive to MAPK pathway inhibition. Moreover, the advantage of targeting IKKs lies in the concomitant inhibition of the external activation of the TNFα pathway stimulated by the stroma. Unfortunately, so far, preclinical data using IKK inhibitors have not successfully been translated into the clinic due to toxicity issues (35), but our data suggest that when used in combination therapies, lower, and therefore less toxic, doses of IKK inhibitors could produce synergistic effects. In an approach to target TNFα directly, we had trialed a combination treatment with Enbrel (etanercept) and MEKi, but we did not observe any synergy (not shown). Besides scheduling and drug penetrance issues, we think that a reason for this observation could be that directly blocking TNFα action will have a broader impact, because it will inhibit all routes of signaling downstream of TNFR (including MKK7–JNK and MKK3 signaling). At the same time, contrary to IKK inhibition, etanercept will not target the TNFα-independent basal IKK–NF-κB activation found in melanoma cells.

An important finding of our study is that the number of macrophages within the tumor is increased in patients in response to BRAFi and MEKi treatment, and this is correlated with a significant increase in TNFα expression in the tumor microenvironment. However, despite this increase in cytokine production, for the development of novel strategies combining MAPK pathway–targeted therapy with adoptive immunotherapy, it will be crucial to fully understand the impact of BRAF and MEK inhibition on cytokine function. Interestingly, an increase in serum TNFα in patients during MAPK pathway inhibitor treatment has also been described in a study that showed that overall immunity is not perturbed during treatment (36). Furthermore, T-cell infiltration and clonality are enhanced in patients on MAPK pathway–targeted therapy, and BRAF inhibition enhances adoptive T-cell transfer therapy in mice (37–39). Although, in contrast to BRAF inhibition, MEK inhibition can affect viability and function of dendritic cells in vitro (40, 41), in patients, T-cell recruitment and clonality are still increased in the presence of MEKi (37, 42–44). The exact impact of combined BRAF and MEK inhibition on the activity of the individual immune-cell populations within the tumor remains to be investigated, but we find that in vitro macrophages protect melanoma cells in the presence of MEKi, and the inhibitor does not affect the expression of TNFα or the ability of macrophages to stimulate MITF expression in melanoma cells (Supplementary Fig. S5). Moreover, the majority of patients in our study who displayed increased macrophage numbers had been on BRAFi–MEKi combination therapy.

Our finding of possible immune-promoted resistance to MAPK pathway inhibitors has important implications for clinical strategies, because it means that we have to consider all components of the immune microenvironment in combination therapies. Targeting myeloid cell infiltration seems to be an attractive option, and indeed the colony-stimulating factor (CSF)-1R inhibitor PLX3397 has been shown to reduce myeloid cell infiltration and enhance adoptive cell transfer immunotherapy in Braf^V600E-driven melanomagenesis in mice (45). Not surprisingly, a clinical trial combining PLX3397 with vemurafenib in melanoma has recently been initiated. In summary, our data suggest that using drug combinations that affect both the tumor cells and tumor microenvironment–derived survival signals will increase the responsiveness to MAPK pathway inhibitors in melanoma and may have a greater chance of creating more durable responses.

The A375, WM266-4, SKMel28, and SKMel2 cells were purchased from the ATCC, and the 501mel and 888mel cells were a gift from Steve Rosenberg (NCI, Bethesda, MD); all cells were obtained in 2008. Additional cell lines in the panel used for RNA extraction were a gift from Imanol Arozarena (University of Manchester). All cell lines were authenticated in house by short tandem repeat profiling before and during the study; the last authentication was carried out in 2014. These cell lines were grown in DMEM/10% FCS (PAA). The 4434 melanoma cells were isolated from a Braf^V600E mouse (46) and were grown in RPMI/10% FCS. THP1 cells were a gift from Adam Hurlstone (University of Manchester) and grown in RPMI/10% FCS (PAA). Cell survival was measured as the optical density at 540 nm of solubilized toluidine blue from formalin-fixed cells. Anoikis assays were performed by culturing 10,000 cells in nonadhesive plates in DMEM/2% FCS for 48 hours. Viable cells were assessed by trypan blue exclusion.

PD184352 was obtained from Axon Medchem, selumetinib (AZD6244) from Selleck Chemicals, and BMS-345541 and SC-514 from Sigma. Human recombinant TNFα, IL4, CSF-1, and IL13, as well as mouse recombinant TNFα, were from PreproTech.

All procedures involving animals were approved by the Animal Ethics Committees of the Institute of Cancer Research and The Cancer Research UK Manchester Institute in accordance with National Home Office regulations under the Animals (Scientific Procedures) Act of 1986 and according to the guidelines of the Committee of the National Cancer Research Institute. C57J/B6 mice were purchased from Charles River, and LysM-Cre mice (B6.129P2-Lyz2tm1(cre)Ifo/J) were from The Jackson Laboratory. Tnfα^−/−, LysM-Cre, and Tnfα^F/F have been described previously (23, 47, 48). For long-term survival tests, Braf^V600E;Tyr::CreERT2 and Braf^V600E;Tnfα^−/−;Tyr::CreERT2 mice were treated with tamoxifen, as described previously (17). Controls were either ethanol-treated Braf^V600E;Tnfα^−/−;Tyr::CreERT2 mice or tamoxifen-treated Tyr::CreERT2 mice. For allografts, 5 × 10⁶ 4434 melanoma cells were injected subcutaneously into the flank of immunocompetent mice, and tumor growth was monitored. For drug treatments, the tumors were allowed to establish, and mice were dosed daily by oral gavage with vehicle (5% DMSO), PD184352 (25 mg/kg/day), BMS-345541 (40 mg/kg/day), or PD184352 (25 mg/kg/day) plus BMS-345541 (40 mg/kg/day). Tumor size was determined by caliper measurements of tumor length, width, and depth, and volume was calculated as follows: volume = 0.5236 × length × width × depth (mm).

Patients with BRAF^V600-positive metastatic melanoma were treated with either a BRAFi or a combination of BRAFi and MEKi (for patient characteristics, see Supplementary Table S1). All patients gave their consent for tissue acquisition according to an Institutional Review Board–approved protocol. Tumor biopsies were obtained before treatment (day 0), at 10 to 14 days on treatment, and/or at the time of progression if applicable. Patient cDNA samples were preamplified using the PreAmp Master Mix Kit (Applied Biosystems) according to the manufacturer's instructions. Real-time qPCR conditions and primer sequences are described in Supplementary Data.

Cryosections of mouse or human tumors were permeabilized in a solution of 0.1% Trition and 1% saponin in PBS for 15 minutes. Sections were blocked in 10% BSA at 37°C for 30 minutes and incubated overnight at 4°C with primary CD68 antibody in 10% BSA PBS. The anti-mouse CD68 antibody (FA-11) was from Abcam. The anti-human CD68 monoclonal antibody (KP1) was from DAKO. Stained sections were washed in PBS and then incubated with secondary antibody for 2 hours at room temperature and mounted using Vectashield.

Cells were lysed in SDS sample buffer and analyzed by standard Western blotting protocols. The antibodies used were as follows: MITF clone C5 from Neomarkers/Lab Vision, and CDK2 (D-12), CDK4 (H-22), and ERK2 (C-14) from Santa Cruz Biotechnology. Antibodies against p65, cleaved caspase-3, and pIκBα were from Cell Signaling Technology and those against BCL2 and p27 were from BD Biosciences. Anti–phospho-ERK was from Sigma.

RNA from cell lines was isolated with TRizol, and selected genes were amplified by real-time qPCR using SYBR green (Qiagen). RNA was similarly isolated from frozen sections of mouse tumor left in TRizol for 2 hours.

Primers used in the qPCR gene-expression analyses were for human sequences: MITF: CCGTCTCTCACTGGATTGGT and TACTTGGTGGGGTTTTCGAG; GAPDH: CAATGACCCCTTCATTGACC and GACAAGCTTCCCGTTCTCAG; ACTB: GCAAGCAGGAGTATGACGAG and CAAATAAAGCCATGCCAATC; primers for mouse sequences were Cd68: GCTACATGGCGGTGGAGTACAA and ATGATGAGAGGCAGCAAGATGG; Cd86: TGCTCATCTATACACGGTTAC and TTTCTTGGTCTGTTCACTCTC; Cd163: ACATAGATCATGCATCTGTCATTTG and CATTCTCCTTGGAATCTCA CTTCTA; Tnfα: GACGTGGAAGTGGCAGAAGAG and TGCCACAAGCAGGAATGAGA; Gapdh: TCTCCCTCACAATTTCCATCCCAG and GGGTGCAGCGAACTTTATTGATGG. Qiagen QuantiTect primers were used for TNFα: QT0002916; IL1β: QT00021385, and YM1: QT00068446.

Conditioned medium was collected from macrophages and analyzed using a TNFα ELISA kit from PreproTech according to the manufacturer's instructions. The TNFα-blocking antibody (Ab6671; Abcam) was used at a concentration of 5 ng/mL.

siRNAs were transfected using INTERFERin siRNA-transfection reagent (Polyplus) according to the manufacturer's instructions. MITF target sequences were MI#1, GAACGAAGAAGAAGAUUUAUU; MI#2, AAAGCAGUACCUUUCUACCAC; and MI#3: GACCUAACCUGUACAACAAUU.

Melanoma cells (1 × 10⁵) were plated into 10-cm dishes and allowed to adhere overnight. The next day the medium was replaced with control medium or medium derived from M1- or M2-polarized macrophages containing no antibody or a TNFα-blocking antibody (5 ng/mL). The medium was exchanged every 48 hours for 3 weeks, after which cells were fixed, stained, and quantified.

Publicly available Oncomine datasets used in this article were the Haqq Melanoma dataset (21), containing 37 samples: 3 skin, 8 nonneoplastic nevi, and 25 melanomas (6 primary and 19 metastases); and the Riker Melanoma dataset (accession: GSE7553; ref. 22), containing 72 samples: 4 skin samples, 1 normal epidermal melanocyte culture, 2 melanoma in situ, 14 primary melanomas, and 40 metastatic melanomas as deposited in Oncomine. The datasets were analyzed in Oncomine, and the results were exported and further analyzed using GraphPad Prism. Alternatively, heat maps were exported as publication-quality graphic (SVG).

Chromatin immunoprecipitation assays, using control IgG (Santa Cruz Biotechnology) or antibodies specific for p65 (Ab7970; Abcam), were performed as described previously (49). Primers for the M-MITF promoter were ACTGTCTGTGTTGTCAGGCA and ACATTCCCTTGGAGATAGCCT; for the negative control (MITF coding region): ACCACATACAGCAAGCCCAA and TCCCTCTTTTTCACAGTTGGAGT; and for the positive control (GROa): CGTCGCCTTCCTTCCGGACTCG and GCTCTCCGAGATCCGCGAACCC.

Cells were transfected with plasmid DNA using Attractene (Qiagen) and analyzed for luciferase activity 24 hours after treatment with forskolin or TNFα using an reporter lysis buffer (RLB)–based luciferase assay kit (Promega). Data were normalized to Renilla luciferase activity. The −2.3-kb M-MITF promoter fragment (−2293 bps to +120 bps) and the truncated promoter (−333) cloned into pGL2 (Promega) were described previously (49). The −1.8-kb M-MITF promoter fragment was created by deleting a 5′ KpnI/AvrII fragment from the −2.3-kb construct. The NF-κB mutation (Supplementary Fig. S2B) was created by site-directed mutagenesis.

For immunofluorescence, a Zeiss Axioskop2 plus equipped with epifluorescence was used, and images were taken at room temperature by a Photometrics Cool Snap HQ CCD camera driven by Metamorph software (Universal Imaging). Image analysis was performed using ImageJ software. All Western blot analyses were carried out using Photoshop CS5.1.

THP1 cells were differentiated in Transwell inserts (BD Biosciences). To differentiate THP1 cells into M2-activated macrophages, THP1 cells were treated with 10 ng/mL of 12-O-tetradecanoylphorbol-l3-acetate (TPA) for 24 hours and subsequently with 20 ng/mL of IL4 and 20 ng/mL of IL13 for 48 hours. Alternatively, TPA-treated THP1 cells were stimulated with 15 ng/mL of lipopolysaccharide (LPS) to differentiate them into activated M1 macrophages. After differentiation, the inserts were washed in RPMI three times before being placed in wells with preplated melanoma cells. Experiments using drugs and/or blocking antibodies were performed for 48 hours by adding the respective reagents to the wells, so that both cell populations were exposed to the same conditions.

Peripheral blood mononuclear cells (PBMC) were isolated from leukocyte cones obtained from healthy donors (provided by NHS Blood and Transplant) by density gradient centrifugation using Ficoll Paque Plus (GE Healthcare) for 50 minutes at 400× g. PBMCs were transferred to flasks in serum-free RPMI-1640 Glutamax media (Life Technologies) to allow enrichment for peripheral blood monocytes by adherence to the tissue culture plastic for 1 hour at 37°C. After thorough washing, adhered monocytes were incubated for 6 days in RPMI/10% FCS and 1% penicillin/streptomycin solution (Sigma) supplemented with 100 ng/mL human M-CSF (Peprotech) to stimulate macrophage differentiation. Macrophages were washed and primed by incubating with RPMI media supplemented with 100 ng/mL of IFNγ (Peprotech) or 100 ng/mL of IL4 and IL13 (Peprotech) for 24 hours to drive M1 or M2 polarization, respectively. Unprimed macrophages were incubated with nonsupplemented RPMI media. By adding 20 ng/mL of LPS to media containing priming stimuli for a further 24 hours, M1 and M2 macrophages were activated. Cells were thoroughly washed in PBS before incubating for a further 24 hours in nonsupplemented RPMI media to produce conditioned media to be used in subsequent in vitro assays.

If not indicated otherwise, data represent the results for assays performed in triplicate, with error bars representing SDs or errors from the mean. Predominantly the Student t test and one-way ANOVA with the Tukeys post hoc tests were used and performed using GraphPad Prism version 4.00 for Mac OS. Pearson correlation was used to analyze associated gene expression.

J.A. Wargo has received honoraria from the speakers' bureau of Dava Oncology. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.P. Smith, K.T. Flaherty, C. Wellbrock

Development of methodology: M.P. Smith, H. Young, C. Wellbrock

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.P. Smith, B. Sanchez-Laorden, K. O'Brien, J. Ferguson, H. Young, N. Dhomen, K.T. Flaherty, D.T. Frederick, Z.A. Cooper, R. Marais

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.P. Smith, B. Sanchez-Laorden, H. Brunton, K.T. Flaherty, J.A. Wargo, R. Marais, C. Wellbrock

Writing, review, and/or revision of the manuscript: M.P. Smith, K.T. Flaherty, D.T. Frederick, Z.A. Cooper, J.A. Wargo, C. Wellbrock

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Dhomen, D.T. Frederick, C. Wellbrock

Study supervision: C. Wellbrock

The authors thank Adam Hurlstone for help with the THP1 system and Imanol Arozarena for providing melanoma cell lines.

This work was supported by funding from Cancer Research UK (C11591/A16416, to C. Wellbrock; C15759/A12328 and C107/A10433, to R. Marais), a Wellcome Trust Institutional Strategic Support Fund (ISSF) award (097820/Z/11/B) to the University of Manchester, and an NCI/NIH U54CA163125 grant to J.A. Wargo and K.T. Flaherty.