BRAF and MEK1/2 inhibitors are effective in melanoma but resistance inevitably develops. Despite increasing the abundance of pro-apoptotic BIM and BMF, ERK1/2 pathway inhibition is predominantly cytostatic, reflecting residual pro-survival BCL2 family activity. Here, we show that uniquely low BCL-XL expression in melanoma biases the pro-survival pool towards MCL1. Consequently, BRAF or MEK1/2 inhibitors are synthetic lethal with the MCL1 inhibitor AZD5991, driving profound tumour cell death that requires BAK/BAX, BIM and BMF, and inhibiting tumour growth in vivo. Combination of ERK1/2 pathway inhibitors with BCL2/BCL- w/BCL-XL inhibitors is stronger in CRC, correlating with a low MCL1:BCL-XL ratio; indeed the MCL1:BCL-XL ratio is predictive of ERK1/2 pathway inhibitor synergy with MCL1 or BCL2/ BCL-w/BCL-XL inhibitors. Finally, AZD5991 delays acquired BRAFi/MEKi resistance and enhances the efficacy of an ERK1/2 inhibitor in a model of acquired BRAFi + MEKi resistance. Thus combining ERK1/2 pathway inhibitors with MCL1 antagonists in melanoma could improve therapeutic index and patient outcomes.

BRAF inhibitors (BRAFi, vemurafenib and dabrafenib) and MEK1/2 inhibitors (MEKi, trametinib) improve the out- comes of patients with BRAFV600E/K-mutant melanoma1–3. However, disease relapse typically occurs due to acquired resis- tance that reinstates ERK1/2 signalling4,5. Although combining BRAFi and MEKi improves responses6,7, acquired resistance and disease relapse are only delayed and again typically arise through reactivation of ERK1/24,5. Thus, strategies are required that boost the primary efficacy of BRAFi and MEKi to delay or prevent theemergence of resistance.The cell intrinsic apoptotic pathway is regulated by BCL2 family proteins, which are frequently deregulated in cancer, typically through aberrant expression, activity or deregulation of oncogenic signalling cascades. ERK1/2 can promote cell survival in BRAF- or KRAS-mutant tumour cells8 by increasing expression of pro- survival BCL2 proteins. ERK1/2 stabilise MCL1 through phos- phorylation of T163 and indirectly through RSK-mediated inhibi- tion of GSK39. ERK1/2 signalling also drives transcription of MCL1, BCL2 and BCL-X 8. In addition, ERK1/2 signalling inhibits the pro- apoptotic BH3-only proteins BIM, BMF, PUMA and BAD. ERK1/2 directly phosphorylate BIM, targeting it for ubiquitination and proteasomal degradation10. ERK1/2 signalling also represses BMF protein expression11, and promotes BAD inactivation through MSK/RSK-catalysed phosphorylation and sequestration by 14-3-3 proteins8. Although ERK1/2 have been proposed to regulate BIK in a manner analogous to BIM, this is disputed12,13. ERK1/2 signalling also represses BIM and PUMA transcription by destabilising FOXO3A14–17. As a result, inhibition of ERK1/2 signalling in tumour cells invariably promotes the expression of pro-apoptotic BIM, BMF and/or PUMA11. Despite this, apoptotic responses to ERK1/2 pathway inhibitors are typically weak because of residual activity of pro-survival BCL2 proteins.Among the agents developed to inhibit pro-survival proteinsand drive tumour cell apoptosis18, drugs that mimic the BH3 domains of BH3-only proteins (BH3-mimetics) are the most advanced. Venetoclax (ABT-199), a BCL2-selective inhibitor, has been approved for clinical use. Navitoclax (ABT-263) and AZD4320 target BCL2, BCL-w and BCL-XL but not MCL1 or A119–22.

AZD4320 has nanomolar affinity for BCL2 and BCL-XL and physicochemical properties suitable for intravenous admin- istration, which may avoid toxicities observed with oral admin- istration of navitoclax21,23,24. BCL2/BCL-w/BCL-XL inhibitors are showing promise in haematological malignancies such as chronic lymphocytic leukaemia (CLL), but will require combination to be effective in solid tumours. Indeed, ERK1/2 pathway inhibitors combine with navitoclax, or the close analogue ABT-737, to induce colorectal cancer (CRC) apoptosis and tumour regression in vivo11,25,26. This combination may also be effective in non- small cell lung cancer (NSCLC) and pancreatic tumours; however, we and others have noted more limited synergy between ERK1/2 pathway inhibitors and navitoclax/ABT-737 in melanoma11,26–28. BCL-XL and MCL1 are the major pro-survival proteins in solid tumours (Cancer Cell Line Encyclopaedia (CCLE; https://portals. broadinstitute.org/ccle) and The Cancer Genome Atlas (TCGA; https://cancergenome.nih.gov/)), but development of MCL1 inhibitors (MCL1i) has lagged behind that of BCL2/BCL-w/BCL- XL inhibitors due to challenges associated with targeting the MCL1 BH3-binding groove8,29,30. However, since potent and selective MCL1i are now in clinical development, including S6384531, AMG 17632 and AZD599133, it is imperative to identify drug combinations and disease stratification criteria to maximise their impact.Here, we show that the pro-survival BCL2 family pool is biasedtowards MCL1 in melanoma compared to CRC, NSCLC and pancreatic tumour lineages, due to low BCL-XL expression. Thus, MCL1 is critical in restraining pro-apoptotic BH3-only proteinsinduced by ERK1/2 inhibition in melanoma. Consequently, com- bined inhibition of ERK1/2 signalling and MCL1 is synthetic lethal, inducing profound, synergistic BAK/BAX-, BIM- and BMF- dependent apoptosis and tumour regression. Finally, combining ERKi and MCL1i overcomes acquired resistance to combined BRAFi + MEKi. Thus, exploiting specific inhibition of ERK1/2 sig- nalling and apoptotic priming in BRAF-mutant cells coupled with the pro-survival bias towards MCL1 could afford a large therapeutic window and further improve patient outcomes in melanoma.

Results
The melanoma pro-survival BCL2 family pool is MCL1 biased. We first examined RNA-sequencing (RNA-seq) data available in the CCLE for transcripts encoding pro-survival BCL2 proteins34. While MCL1 expression in the CCLE data set was broadly similar in CRC and melanoma cells, and slightly higher in NSCLC and pancreatic, levels of BCL2L1 (encoding BCL-XL) were strikingly lower in melanoma relative to the other lineages (Fig. 1a, b). Consequently, the MCL1:BCL-XL mRNA ratio, encoding the major pro-survival proteins in solid tumours, was two- to four- fold higher in melanoma than in the other lineages (Fig. 1c). Indeed, of all the tumour lineages in the CCLE, melanoma exhibited one of the highest median MCL1:BCL-XL mRNA ratios (Supplementary Fig. 1a). Notably autonomic ganglia tumour cells which, like melanocytes, have neural crest developmental origin also exhibited a high MCL1:BCL-XL ratio. While BCL2 expression was higher in melanoma, BCL2 mRNA levels (RNA-seq read number) were very low in each lineage relative to MCL1 and BCL- XL (Supplementary Fig. 1b). BCL2L2 (encoding BCL-w) levels were similar in each lineage (Supplementary Fig. 1c) and BCL2A1 (encoding A1/BFL1), a MITF target gene35, exhibited melanoma- selective expression (Supplementary Fig. 1d). Similar observations have been made in TCGA patient samples (https://cancergenome. nih.gov/). We confirmed these trends at the protein level in seven melanoma and seven CRC cell lines (Fig. 1d, e). All melanoma cell lines exhibited strikingly lower BCL-XL expression than the CRC cell lines whereas MCL1 was, on the whole, marginally higher in melanoma (Fig. 1d, e). BCL2 expression was markedly elevated in melanoma (Fig. 1d and Supplementary Fig. 2a), consistent with BCL2 being a MITF target gene36.

BCL-w levels were similar between CRC and melanoma (Fig. 1d and Supple- mentary Fig. 2b). We failed to detect A1 protein expression using a variety of commercially available antibodies, even in melanoma cells with high BCL2A1 mRNA. Next, we used recombinant protein standards (SupplementaryFig. 2c, d) to absolutely quantify MCL1 and BCL-XL in 64 cell lines (Supplementary Fig. 2e–h), and thereby determine the MCL1:BCL- XL protein ratio (Fig. 1f–h). MCL1 levels were similar across the four lineages but higher in melanoma (~20 vs. <14 fmol per μg total cellular protein; Fig. 1f and Supplementary Fig. 2f). Consistent withmRNA levels, BCL-XL protein was strikingly lower in melanoma (Fig. 1g and Supplementary Fig. 2f) compared to CRC, NSCLC and pancreatic lineages (median value of <4 fmol vs. >11 fmol per μg total cellular protein). Consequently the MCL1:BCL-XL ratio was substantially biased towards MCL1 in melanoma (Fig. 1h); themedian ratio was 5–6 in melanoma vs. ~1 in CRC and pancreatic tumour cells, and ~2 in NSCLC (Fig. 1h). The same analysis in CRC and melanoma patient-derived xenograft (PDX) samples (Fig. 1i–l) demonstrated markedly low BCL-XL expression in melanoma PDXs vs. CRC (Fig. 1i, k), whereas MCL1 expression was broadly similar (Fig. 1i, j). Consequently, the MCL1:BCL-XL protein ratio was ~5 times higher in melanoma (Fig. 1l). Thus, the ratio of the two major pro-survival proteins in solid tumours, MCL1 and BCL- XL, is substantially biased towards MCL1 in melanoma cell lines and patient samples. BRAFi or MEKi combine with MCL1i to selectively kill mela- noma cells. Combining BRAFi or MEKi with pan-BCL2/BCL-w/ BCL-XL inhibitors causes strong synergistic cell death of CRC, NSCLC and pancreatic cells and tumour regressions11,25,26.However, we and others observed more modest synergy in melanoma11,26–28.

This difference could be explained by the bias towards MCL1 in melanoma (Fig. 1 and Supplementary Figs. 1 and 2). We therefore tested ERK1/2 pathway inhibitors in combinationFig. 1 Low BCL-XL expression in melanoma biases the pro-survival pool towards MCL1. a–c RNA-sequencing mRNA expression data obtained from the Cancer Cell Line Encyclopaedia for MCL1 (a) BCL-XL (BCL2L1) (b) and MCL1:BCL-XL ratio (c) in colorectal cancer (CRC), melanoma, non-small cell lung cancer (NSCLC) and pancreatic tumour cells. Expression values are reads per kilobase million (RPKM) and are presented on a log2 scale. Black lines indicate median values. d, e Seven CRC (left) and seven melanoma (right) cell lines were cultured for 24 h before lysis and western blotting with the indicated antibodies. Quantification (e) of MCL1 and BCL-XL levels relative to A375 cells was performed by quantitative western blotting with fluorescently labelled secondary antibodies. f, g MCL1 (f) and BCL-XL (g) expression were quantified absolutely in CRC, melanoma, NSCLC and pancreatic tumour cell lines using recombinant protein standards and quantitative western blotting with fluorescently labelled secondary antibodies. Each point represents mean expression in a cell line from three independent experiments and is expressed in fmol per µg of total cellular protein (fmol µg−1). Black lines indicate median values. h Ratio of MCL1 to BCL-XL protein calculated from f and g, with black lines indicating median values. i Seven CRC (left) and seven melanoma (right) patient-derived xenograft (PDX) tumours were homogenised in lysis buffer and western blotted with the indicated antibodies.j, k Absolute quantification of MCL1 (j) and BCL-XL (k) expression in CRC and melanoma PDX tumour samples using recombinant protein standards and quantitative western blotting with fluorescently labelled secondary antibodies. Each point represents mean expression in a cell line from at least five technical replicates and is expressed in fmol per µg of total cellular protein (fmol µg−1). Black lines indicate median values.

l Ratio of MCL1 to BCL-XL protein calculated from (j) and (k), with black lines indicating median values. Results (e–h, j–l) are the mean of three or more independent experiments, and error bars indicate SD. P ≤ 0.0001 (****), P ≤ 0.001 (***), P ≤ 0.01 (**), P ≤ 0.05 (*), ns (not significant) as determined by Kruskal–Wallis and Dunn’s multiple comparisons tests (a–c, f–h) or two-tailed Mann–Whitney test (j–l)with the novel MCL1i AZD599133 using high-throughput combi- nation assays (HTCA) in an 8 × 8 concentration matrix and determining Loewe synergy scores in 12 melanoma and CRC cell lines. These responses were compared to those with the BCL2/ BCL-w/BCL-XL inhibitor AZD4320, which has nanomolar affinity for these pro-survival proteins, similar to navitoclax21,23,24. Combination of MEKis trametinib or selumetinib (AZD6244, ARRY-142886) with AZD5991 or AZD4320 was performed in all CRC and melanoma cell lines to allow comparison of BRAFMUT, RASMUT or RAS/RAFWT cells (Supplementary Table 1); combi- nations with the BRAFi vemurafenib (only effective in BRAFV600- mutant cells) or ERK1/2 inhibitor SCH772984 were tested in selected cell lines. Data from these combination assays are shown in Fig. 2a–c and Supplementary Figs. 3–6 where trametinib, selu- metinib or vemurafenib combined with AZD5991 to cause strong cell death in BRAFV600E-mutant A375, SK-MEL-28, WM266-4, A2058 or SK-MEL-5 melanoma cells. MEKis or ERKi, also com- bined well with AZD5991 in NRAS-mutant MEL-JUSO, SK-MEL- 2, SK-MEL-30 and WM852 melanoma cells (Supplementary Figs. 3–6). In these melanoma cells, ERK1/2 pathway inhibitors typically combined less well with AZD4320 than AZD5991, whereas trametinib or selumetinib combined better with AZD4320 than AZD5991 in HCT116 and HT29 CRC cells (Supplementary Fig. 5). Loewe synergy score calculation for all 24 cell lines (Sup- plementary Table 2) revealed that both AZD5991 and AZD4320- synergised (score ≥ 5) with trametinib and selumetinib in the majority of cell lines (Fig. 2d, e).

However, synergy with AZD5991 was typically far stronger in melanoma: median synergy scores for trametinib or selumetinib combined with AZD5991 were >3 times higher in melanoma than CRC, whereas combination with AZD4320 was more effective than with AZD5991 in CRC (>2-fold higher median synergy score).We confirmed these high-throughput results in selected cell lines by examining cell cycle profiles. Vemurafenib (Fig. 2f) or selumetinib (Supplementary Fig. 7a) resulted in a strong G1 arrest in BRAF-mutant A375 cells but only limited cell death (<10% sub- G1 DNA fraction) at higher concentrations. 1 μM AZD5991 hadno effect on cell cycle or cell death; however, combination ofAZD5991 with 1 μM vemurafenib or selumetinib elicited 65–75% cell death (Fig. 2f and Supplementary Fig. 7a). In A375 cells, 10 μM AZD5991 had no effect on cell cycle profile, whereas concentra-tions as low as 100 nM combined with selumetinib to induce substantial synergistic cell death (Supplementary Fig. 7b). Very similar results were observed in NRASQ61K-mutant SK-MEL-30 melanoma cells (Supplementary Fig. 7c, d). Further experiments focused on annexin V-positive, apoptotic cells. In A375 cells both vemurafenib and selumetinib combined synergistically withAZD5991 to drive ~80% apoptosis following 48 h treatment (Fig. 2g, h); similar results were observed with vemurafenib in BRAF-mutant WM266-4 (Supplementary Fig. 7e) and selumetinib in NRAS-mutant SK-MEL-30 cells (Supplementary Fig. 7f). BRAFi (vemurafenib), MEKi (selumetinib) or ERKi (SCH772984) com- bined to kill BRAF-mutant A375 and WM266-4 cells in a caspase- dependent fashion (Fig. 2i) but only selumetinib or SCH772984 were effective in NRASQ61K-mutant SK-MEL-30 cells, consistent with vemurafenib efficacy being confined to BRAF-mutant tumour cells. In contrast, and consistent with results from the HTCA experiments, A375 (Fig. 2j, k) and SK-MEL-30 cells (Supplemen- tary Fig. 7g, h) both exhibited markedly weaker apoptotic responses when selumetinib was combined with the BCL2/BCL- w/BCL-XL inhibitor AZD4320 rather than AZD5991.The inverse was true for the CRC cell lines COLO205 and SW620. While selumetinib and AZD5991 combined to induce apoptosis in COLO205 cells (Emax ~ 45%), far stronger synergistic apoptosis was apparent with AZD4320 (Emax ~ 80%) (Fig. 2l, m).

SW620 cells were an even more striking example: combination of selumetinib with 10 μM AZD5991 resulted in <10% apoptosis, whereas ~90% apoptosis was achieved when selumetinib was combined with AZD4320 (Fig. 2n, o). Indeed, in COLO205 and SW620, concentrations of AZD4320 >1 μM elicited considerable apoptosis alone (Fig. 2m, o), suggesting a stronger dependency upon BCL2, BCL-w and/or BCL-XL for basal viability in CRC. Combination of navitoclax with selumetinib promoted only modest apoptosis at near-identical potency and maximal response as AZD4320 in A375 and SK-MEL-30 (Supplementary Fig. 7i, j), consistent with AZD4320 and navitoclax having similar pharmacology21,23,24.To investigate the role of the individual pro-survival proteins inhibited by AZD4320/navitoclax in melanoma, we used selective inhibitors of BCL2 (venetoclax, ABT-199) and BCL-XL (A- 1155463)37. Selumetinib combined weakly with A-1155463 in A375 or SK-MEL-30 (Fig. 2p, q) to induce a similar apoptotic response as with AZD4320 or navitoclax (Fig. 2k and Supple- mentary Fig. 7h–j). Despite higher levels of BCL2 in melanoma relative to CRC (Fig. 1d and Supplementary Fig. 1b) venetoclax monotherapy had no effect on viability and did not combine with selumetinib to induce apoptosis, although venetoclax marginally enhanced the combination effect of A-1155463 and selumetinib in A375 cells (Fig. 2p and Supplementary Fig. 7k, l). Thus, melanoma cells were not addicted to BCL2 for viability when the ERK1/2 pathway was inhibited. BCL2 inhibition may, however, slightly enhance the effect of BCL-XL inhibition when the ERK1/2 pathway is inhibited in some melanoma cells. These results indicate that different tumour lineages exhibit selective dependencies upon distinct BCL2 proteins for survival, with melanoma exhibiting a strong bias towards MCL1 and CRC an apparent bias towards BCL-XL.Combining vemurafenib or selumetinib with AZD5991 also inhibited the clonogenic survival of A375 cells (Fig. 3a, b).

However, combined vemurafenib and AZD4320 or navitoclax did not significantly inhibit colony formation (Fig. 3c and Supple- mentary Fig. 8a).The striking synergy between ERK1/2 pathway inhibition and AZD5991 in melanoma was also observed in vivo. Whileselumetinib moderately inhibited the growth of A375 tumour xenografts, and AZD5991 had no effect, selumetinib + AZD5991 caused regression of established tumours and durable full growth inhibition (Fig. 3d, e and Supplementary Fig. 8b, c). Given that AZD5991 is cleared quickly from blood plasma, within 2–6 h33, this efficacy is particularly striking in the context of a once weekly AZD5991 dosing schedule that likely results in MCL1 inhibition for just a few hours per week. Throughout dosing there was littlechange in mouse body mass indicating that these treatments were well tolerated (Supplementary Fig. 8d).We next examined BRAFV600E-mutant primary patient- derived melanoma cells. Consistent with the results describedabove, trametinib and AZD5991 combined to kill MEDI-MEL-10 and MEDI-MEL-07 primary melanoma cells (Fig. 4a–d), whereas combination with AZD4320 or A-1155463 was far less effective (Fig. 4e, f). Thus, dependency on MCL1 when the ERK1/2pathway was inhibited was observed in melanoma cell lines and primary patient-derived melanoma cells.BRAFi or MEKi combinations with AZD5991 were also effective in BRAF-mutant PDX models (Fig. 4g–n andSupplementary Fig. 9a–f). While the BRAFi vemurafenib induced moderate initial regressions of BRAFV600E-mutant MEDI-MEL- 02 PDX tumours, combination with AZD5991 caused sustained near-complete tumour regressions (Fig. 4g, h and SupplementaryFig. 9a).

Similarly, AZD5991 substantially enhanced tumour growth inhibition by trametinib (Fig. 4i, j and Supplementary Fig. 9b). In another PDX model, BRAFV600E-mutant MEDI- MEL-10, AZD5991 again enhanced the efficacy of vemurafenib or trametinib despite this model being refractory to these ERK1/2 pathway inhibitors as monotherapies (Fig. 4k–n). These PDX models represented extremes of response to ERK1/2 pathway inhibition; MEDI-MEL-02 tumour growth was halted by vemurafenib or trametinib, whereas MEDI-MEL-10 tumours were unresponsive (Fig. 4g–n). However, AZD5991 was able to combine with BRAFi or MEKi to regress or inhibit tumour growth and these combinations were well tolerated (Supplementary Fig. 9g–j).Altogether these results indicate that the melanoma bias towards MCL1 provides a therapeutic opportunity for combina- tion with multiple ERK1/2 pathway inhibitors, including the clinically approved BRAFV600-mutant selective inhibitor vemur- afenib and MEKi such as selumetinib or trametinib, thereby promoting striking tumour cell apoptosis and enhancing tumour growth inhibition in vivo.BRAFi or MEKi prime melanoma for MCL1i-induced death. To define the mechanisms underpinning synergy between ERK1/2 pathway inhibition and AZD5991 we assessed the kinetics of the apoptotic response. Addition of AZD5991 to A375 cells pre-treated with vemurafenib or selumetinib for 24 h resulted in a time- dependent increase in annexin V-positive cells (Fig. 5a and Sup- plementary Fig. 10a). This rapid induction of apoptosis was even more apparent at the level of DNA cleavage, suggesting that DNA fragmentation preceded phosphatidylserine membrane leaflet flip- ping (Fig. 5b). Similar rapid apoptosis also occurred following addition of AZD5991 to WM266-4 or SK-MEL-30 cells pre-treated with vemurafenib or selumetinib, respectively (Fig. 5c, d).Accumulation of annexin V-positive cells was preceded by caspase activation. Selumetinib pre-treatment of A375 cells for 24 h, or treatment with AZD5991 alone, had little effect on caspase-3/-7 activity.

However, in cells pre-treated with selume- tinib, caspase-3/-7 activity increased ~5-fold after 2 h AZD5991 treatment, reaching a plateau (~20-fold increase) after 8 h (Fig. 5e). Caspase activation was preceded by activation of the key apoptotic effector BAX, assessed using the 6A7 antibody that recognises an amino-terminal epitope exposed only when BAX is activated (Fig. 5f). In A375 cells pre-treated with selumetinib for 24 h, AZD5991 caused activation of BAX within 15 min that reached a plateau after 2–4 h (Fig. 5f). Thus, ERK1/2 pathway inhibitors primed ERK1/2-dependent tumour cells to undergo an immediate activation of the canonical cell intrinsic apoptotic pathway following MCL1 inhibition.To examine whether apoptosis required BAK, BAX or both, we utilised isogenic HCT116 cells lacking either BAK, BAX or both (BAK BAX DKO)38 (Fig. 5g). Although HCT116 is a CRC cellline, a combination effect with AZD5991 is still observed, albeit more modest than that observed with A375 or SK-MEL-30. BAK knock-out (KO) cells showed no apoptotic defect in response to either selumetinib or combination with AZD5991 (Fig. 5h), whereas cells lacking BAX exhibited a 60% reduction in apoptosis. Finally, BAK BAX double knock-out (DKO) cells were completely resistant to the modest pro-apoptotic effects of selumetinib or the more striking response to combined selumetinib + AZD5991 (Fig. 5h). So while BAX was the more important effector in HCT116, there was some redundancy with BAK. Regardless, the combination killed cells in a BAK/BAX-dependent manner, confirming the anticipated on-target mechanism of this drug combination. In addition, this combination did not increase the fraction of necrotic cells, only apoptotic cells, and BAX knock-out also prevented the accumulation of cells with sub-G1 DNA (Fig. 5i and Supplementary Fig. 10b).

BRAFi or MEKi plus MCLi drives BIM- and BMF-dependent apoptosis. We next examined how ERK1/2 pathway inhibition and/or AZD5991 influenced expression of the BCL2 protein family, the upstream regulators of BAK and BAX. Treatment of A375 cells with vemurafenib (Fig. 6a) or selumetinib (Supple- mentary Fig. 11a) increased expression of the pro-apoptotic BH3- only proteins BIM and BMF in a concentration-dependent manner as expected11. Inhibition of ERK1/2 also abolished expression of NOXA, consistent with observations that NOXA mRNA expression is promoted by ERK1/2 signalling39,40. Similar results were observed in SK-MEL-30 cells with the MEKi selu- metinib (Fig. 6b) and these cells also exhibited PUMA induction. BRAFi or MEKi had more modest effects on other pro-apoptotic and pro-survival BCL2 proteins (Fig. 6a, b and Supplementary Fig. 11a). In both cell lines, 1 μM AZD5991 increased the abun- dance of its target, MCL1, and caused loss of NOXA expression (Fig. 6a, b). The increase in MCL1 expression likely reflects its stabilisation because MULE, an E3 ubiquitin ligase for MCL1, binds to MCL1 via a BH3 domain41 and so will be displaced by AZD5991; thus, increases in MCL1 expression serve as a good marker for MCL1 engagement by AZD5991. The reduction in NOXA may also reflect its displacement from MCL1; certainly some BH3-only proteins are intrinsically disordered and undergo proteasome-dependent turnover when displaced from their binding partners42,43. Combination of vemurafenib (Fig. 6c) or selumetinib (Supplementary Fig. 11b) with increasing con- centrations of AZD5991 promoted PARP cleavage indicative of caspase-3 activation, and also reduced expression of BIM and BMF slightly compared to ERK1/2 pathway inhibitor alone. In A375 and SK-MEL-30 cells, MCL1 expression increased with AZD5991 concentrations as low as 10 nM (Fig. 6c, d and Sup- plementary Fig. 11b).These results suggested that BIM and/or BMF might be responsible for initiating apoptosis in response to combinedinhibition of MCL1 and ERK1/2 signalling so we used CRISPR/ Cas9 gene editing to generate BIM KO, BMF KO or double knock-out (BIM BMF DKO) A375 cells (Fig. 6e).

We selected three independent clones for each genotype and also three wild- type clones that had been through CRISPR/Cas9 editing with guides for BIM, BMF or both but were negative clones, retaining normal expression of the targeted proteins and serving as controls for off-target effects. These clones were compared with parental A375 cells (Fig. 6f). We found that the striking cell death induced by selumetinib + AZD5991 was inhibited ~50% by knock-out of BIM or BMF whereas BIM BMF DKO cells exhibited a 65%reduction in cell death (Fig. 6f). Thus, BIM and BMF make significant contributions to cell death arising from selumetinib + AZD5991, likely acting through a shared mechanism. The remaining cell death response might reflect contributions from PUMA and BAD, which exhibited modest increases in abundance in response to vemurafenib (Fig. 6a) or selumetinib (Supplemen- tary Fig. 11a). A very similar dependency upon BIM and BMF was observed in SK-MEL-30 cells. Knock-out of BIM and BMF (BIM BMF DKO) (Fig. 6g) inhibited apoptosis induced by selumetinib + AZD5991 by 60–70% compared to parental cells and wild-type clones that, as above, retained normal BIM andBMF expression and induction following the CRISPR/Cas9 process (Fig. 6g, h).Binding of AZD5991 to MCL1 should displace BH3-only proteins such as BIM and BMF from MCL1, allowing them to inhibit other pro-survival proteins to drive apoptosis.

To test this we treated A375 cells with vemurafenib, AZD5991 or the combination and then immunoprecipitated (IP) either MCL1 or BCL-XL before blotting for the presence of BIM and BMF (Fig. 7a). Vemurafenib increased expression of BIM and BMF and while these bound to both MCL1 and BCL-XL, far more bound to MCL1 than to BCL-XL (Fig. 7a), consistent with the high MCL1:BCL-XL ratio (~8) in these cells (Fig. 1 and Supplementary Fig. 2). AZD5991 increased MCL1 abundance, as seen in all experiments, but did not affect BIM or BMF expression. Combination with AZD5991 slightly reduced the vemurafenib-driven increase in BIM and BMF abundance. Most strikingly, vemurafenib + AZD5991 promoted a loss of BIM and BMF from MCL1 and their binding to BCL-XL (Fig. 7a); given the decrease in total BIM and BMF this is likely an underestimate of the redistribution of these proteins to BCL- XL. Very similar results were observed when selumetinib and AZD5991 were combined in A375 and SK-MEL-30 cells (Supplementary Fig. 11c, d). Thus, AZD5991 acted on-targetto drive redistribution of BIM and BMF from MCL1 on to other pro-survival BCL2 proteins, including BCL-XL, which are not themselves targets of AZD5991. Notably, since BIM may activate BAX directly as well as indirectly through inhibition of pro-survival proteins44,45, combination of vemurafenib or selumetinib with AZD5991 also increased the amount of free unbound BIM (Fig. 7a and Supplementary Fig. 11c, d).In contrast to MCL1 inhibition by AZD5991, we observed minimal redistribution of BIM and BMF when BCL-XL was inhibited by A-1155463 in conjunction with ERK1/2 signalling in A375 cells, consistent with the high MCL1:BCL-XL ratio in these cells (~8) (Figs. 1 and 7b and Supplementary Fig. 2). The BCL-XL inhibitor A-1155463 displaced BIM and BMF from BCL-XL, but caused little accumulation on to MCL1 or increase in the fraction of free BIM and BMF not bound to BCL-XL (Fig. 7b). This was not due to loss of BIM and BMF expression in the presence of the combination; indeed, BIM and BMF expression was only reduced in combinations that caused strong PARP cleavage and cell death (Fig. 7a, b and Supplementary Fig. 11c, d). Thus, this is consistent with a more minor role for BCL-XL in these melanoma cells.

Taken together these results indicate AZD5991 acts on-target in melanoma to inhibit MCL1 and strongly synergises with ERK1/2 pathway inhibitors that drive BIM and BMF expressionto promote BAK/BAX-dependent apoptosis; this arises through redistribution of BIM and BMF from MCL1 on to other pro- survival proteins and an increase in free BIM, which may directly activate BAX.ERKi plus MCL1i overcomes BRAFi and/or MEKi resistance. Our results indicated that AZD5991 would combine very well with BRAFi and/or MEKi as a primary, up-front drug combi- nation but we also assessed activity in models of acquired resistance. BRAFi + MEKi is now the standard for con- sistency care for melanoma with BRAFV600E/K 6,7; however, acquired resistance still emerges, typically through reactivation of ERK1/2, and this is fuelling the development of ERK1/2 inhibitors46. We generated A375 sub-lines with acquired resistance to PLX4720 (a vemurafenib analogue) (Supplemen- tary Fig. 12a), selumetinib (Supplementary Fig. 12b) or the combination of PLX4720 + selumetinib at five- or 10-times the IC50 concentrations (Fig. 8a and Supplementary Fig. 12c). Next- generation sequencing revealed that PLX4720/selumetinib- resistant A375 sub-lines exhibited NRASQ61R and MEK1Q56P mutations (Supplementary Fig. 12d–g). Whereas BRAF copy number and expression were unchanged in both PLX4720/ selumetinib-resistant A375 sub-lines, selumetinib-resistant A375 cells harboured BRAF amplification and upregulation of BRAF expression, whereas PLX4720-resistant cells expressed a putative splice variant of BRAFV600E (Supplementary Fig. 13a–f); these are well defined mechanisms of resistance to MEKi or BRAFi, respectively5,47–49. In each case these mechanisms reinstated ERK1/2 signalling in the presence of the respective inhibitor (Supplementary Fig. 13e); however, ERK1/ 2 signalling in these resistant sub-lines remained sensitive to the selective ERK1/2 inhibitor SCH772984 (Supplementary Fig. 13f).

This ERK1/2 inhibitor again combined well with AZD5991 to kill parental A375 cells (Fig. 8a and Supplementary Fig. 12a–c). Significantly, this cell death response was fully maintained in PLX4720/selumetinib-resistant A375 cells (Fig. 8a and Supplementary Fig. 12c), as well as the cell lines resistant to PLX4720 (Supplementary Fig. 12a) or selumetinib (Supplementary Fig. 12b) monotherapy.To see if this activity of ERKi + MCL1i was maintained in vivowe employed tumour xenografts using the PLX4720/selumetinib- resistant A375 cells. For these studies we used the ERKi AZ619750, due to the poor pharmacokinetic properties of SCH772984. AZD5991 had no activity as monotherapy against the PLX4720/selumetinib-resistant A375 cells (Fig. 8b), as was seen in cell culture studies (Fig. 8a). AZ6197 caused significant inhibition of tumour growth, consistent with ERKi being able to overcome acquired resistance to BRAFi + MEKi, but this effect was transient. AZD5991 combined with AZ6197 to cause a more profound and durable tumour growth inhibition (Fig. 8b). Again this was striking given that AZD5991 was dosed once weekly, a schedule that likely results in MCL1 engagement for just a few hours per week. AZ6197 alone, or in combination with AZD5991, was well tolerated (Supplementary Fig. 13g). These results indicate that, as well as use in an up-front combination with BRAFi and/or MEKi, AZD5991 can be used in combination with an ERKi to overcome acquired resistance to BRAFi and/or MEKi driven by a variety of mutations in ERK1/2 pathway components. Despite the efficacy of AZD5991 in these combination studies experience shows that acquired resistance is inevitable. To model this we generated A375 cells adapted to grow in vemurafenib (Vem-R), AZD5991 (5991-R) or the combination (Vem/5991-R) in three independent experiments. Cumulative population doublings showed that AZD5991 had no effect on A375 cell proliferation (Fig. 8c). Vemurafenib caused a complete cessationof A375 proliferation for ~2.5-weeks before resistant cells started to emerge and proliferate whereas combination with AZD5991 doubled the time to resistance, with Vem/5991-R-resistant cells emerging after 5–10 weeks (Fig. 8c, d).

Vem-R cells exhibited a previously described BRAF splice variant that likely allowed them to maintain p-MEK1/2 and p-ERK1/2 in the presence of vemurafenib49 (Fig. 8e). Interestingly 5991-R cells exhibited higher levels of MCL1, though this was reduced slightly following withdrawal of AZD5991, suggesting that this might simply represent AZD5991 binding and stabilisation of MCL1 as described above. Vem/5991-R A375 cells did not express any apparent BRAF splice variant, suggesting that this drug combination selected for reinstatement of ERK1/2 signalling through another mechanism (Fig. 8e). Significantly, all resistant cells, whether to monotherapy or combination, retained p-MEK1/2 and p-ERK1/2, and all Vem-R, 5991-R and Vem/5991-R remained sensitive to ERKi SCH772984 (Fig. 8d and Supplemen- tary Fig. 14a, b). Thus, adaptation or acquired resistance to vemurafenib, AZD5991 or vemurafenib + AZD5991 could be overcome by combination with an ERK1/2 inhibitor.Combination with AZD5991 also doubled the time taken for vemurafenib resistance to emerge in WM266-4 cells (Fig. 8f), and increased the time to selumetinib resistance in SK-MEL-30 cells by ~50% (Fig. 8g and Supplementary Fig. 14c). As with A375 cells, SK-MEL-30 cells acquired resistance to selumetinib monotherapy and selumetinib combination with AZD5991 by reinstating p-MEK1/2 and p-ERK1/2 in the presence of inhibitor (Supplementary Fig. 14d), and similarly acquired resistance to selumetinib or selumetinib + AZD5991 could be overcome using the ERK1/2 inhibitor SCH772984 (Supplementary Fig. 14c). Thus, AZD5991 can delay acquired resistance to BRAFi or MEKi in BRAF-mutant or NRAS-mutant melanoma cells.

Discussion
While the BRAFi + MEKi combination has transformed the treatment of BRAFV600E/K-mutant melanoma, clinical responses are transient due to the emergence of resistance. This reflects the fact that even strong ERK1/2 inhibition exerts a predominantly cytostatic effect despite causing a substantial increase in pro- apoptotic BIM, BMF and in some cases PUMA. This is also true in other tumour lineages, such as CRC; however, whereas increased BIM and BMF primed CRC cells to undergo substantial apoptosis following treatment with the BCL2/BCL-w/BCL-XL inhibitor navitoclax/ABT-26311,25,26, this combination was notably less effective in melanoma11,27,28. This prompted us to examine the expression of pro-survival proteins. Interrogation of CCLE and TCGA data sets indicated that the median MCL1:BCL- XL mRNA ratio was two- to fourfold higher in melanoma than in CRC, NSCLC and pancreatic cancer lineages. This was confirmed by western blotting and absolute protein quantification, which revealed that low BCL-XL expression biased the pro-survival protein pool strongly towards MCL1 in melanoma.Our results reveal that this lineage-selective MCL1 bias predicts the response to the MCL1i AZD5991, but only in combination with agents that inhibit ERK1/2 signalling to drive BIM and BMF expression. MCL1i monotherapy, even at higher concentrations, induced little apoptosis of melanoma (or CRC) tumour cells in vitro and did not inhibit cell division during long-term culture, while in vivo AZD5991 alone did not inhibit the growth of melanoma xenografts. Therefore, melanoma tumour cells are not dependent on MCL1 for survival per se; rather they are dependent on MCL1, to a far greater extent than the other pro- survivals, when the ERK1/2 pathway is inhibited. Presumably, without induction of BIM and/or BMF the basal level of otherpro-survival proteins represents a sufficient barrier to prevent apoptosis when MCL1 is inhibited.

These results suggest that melanoma represents a unique opportunity as a disease in which two broad therapeutic windows (ERK1/2 addiction and MCL1 addiction) are inter-dependent,with MCL1 addiction only manifest when ERK1/2 signalling is inhibited. This synthetic lethality will be best leveraged using BRAF inhibitors such as vemurafenib, which specifically inhibit ERK1/2 signalling in BRAFV600-mutant tumour cells, but may also transform the potential of MEKi and ERKi inERK1/2-addicted melanoma tumour cells. This is reflected in striking synergistic cell death in vitro and tumour regression or enhanced growth inhibition in vivo when BRAFi or MEKi and AZD5991 are combined as an up-front combination. Mechanistic studies demonstrated that ERK1/2 inhibition primes melanoma cells to undergo apoptosis following sub- sequent MCL1 inhibition. This apoptosis required increased expression of BIM and BMF, which appear to play overlapping roles; the remaining 20–25% cell death in BIM BMF DKOs may reflect contributions from BAD and PUMA, both of which increased in expression following ERK1/2 inhibition. Apoptosis following addition of AZD5991 to primed cells was rapid, with maximal BAX activation after 2–4 h, underscoring the extent to which these cells are addicted to MCL1 to restrain the BIM and BMF induced by ERK1/2 inhibition. Consistent with this, when MCL1 was inhibited we observed redistribution of BIM and BMF from MCL1 on to BCL-XL and an increase in free unbound BIM that may activate BAX directly44,45. NOXA exhibits highly selective binding to MCL1 (and A1) vs. BCL2, BCL-w or BCL- X 11,51,52, and consistent with previous reports was dramatically lost when the ERK1/2 pathway was inhibited (Fig. 6). This loss of NOXA presumably dampens the apoptotic response to ERK1/2 pathway inhibitor monotherapy but further primes MCL1 with BIM and BMF. Release of this BIM and BMF upon inhibition of MCL1 may therefore accentuate the synergistic effect of com- bined ERK1/2 pathway and MCL1 inhibition.Critically, AZD5991 was also active in melanoma cells with acquired resistance to BRAFi + MEKi, the current standard of care for BRAFV600E/K melanoma. These cells acquired resistance through the emergence of NRAS and MEK1 mutations to reinstate ERK1/2 signalling and remained sensitive to ERKi.

Significantly ERKi still combined with MCL1i to kill BRAFi + MEKi-resistant tumour cells in vitro and inhibit their growth as tumour xenografts in vivo.Finally, AZD4320 (a BCL2/BCL-w/BCL-XL inhibitor) syner- gised with ERK1/2 pathway inhibitors to induce apoptosis more strongly than AZD5991 in CRC, a disease with common KRAS or BRAF mutations that is as yet poorly served by ERK1/2 pathway inhibition. The reciprocal was true in melanoma, where combi- nation with AZD5991 generally resulted in greater synergy and apoptosis than combination with AZD4320. Despite high expression of BCL2 in melanoma compared to other lineages, the response to AZD4320 was not replicated by selective BCL2 inhibition with venetoclax, but rather was recapitulated using the selective BCL-XL inhibitor A-1155463. Thus, although either BCL2/BCL-w/BCL-XL or MCL1 inhibitors can synergise to someextent with ERK1/2 pathway inhibitors in a broad range of lineages11,26,27,53,54, the ability to target individual pro-survival proteins as BH3-mimetic-specificity continues to be refined will allow antagonism of only those critical pro-survival proteins that limit apoptotic responses to ERK1/2 pathway inhibitors in a particular lineage, thereby improving the therapeutic window of these combinations. Whereas BRAFi and MEKi are approved for use in melanoma, current clinical studies are investigating how best to combine BRAFi and/or MEKi with other targeted agents in CRC and other solid tumours such as NSCLC and pancreatic55,56. Our results suggest that in CRC ERK1/2 inhibi- tion may combine better with antagonists of BCL-XL than MCL1 due to their relative expression, further underlining the potential for tumour lineage-selective differences that will drive BH3- mimetic combination choices; as we show here, the ratio of pro- survival proteins will be informative in this choice, and represent a potential clinical biomarker.In summary, our results demonstrate that targeting melano-ma’s MCL1 bias unleashes the potential of BRAF and ERK1/2 pathway inhibitors by transforming cytostatic responses into striking apoptotic cell death. This approach results in superior tumour regressions or growth inhibition and could afford a broad therapeutic window in patients. Since BRAFi and MEKi are already approved for treatment of BRAFV600E/K-positive mela- noma, combination with MCL1i may provide an opportunity to promptly improve patient outcomes.Reagents and resources.

A full list of reagents and resources used in this study is provided in Supplementary Table 3. Further information and requests for resources and reagents should be directed to Simon Cook ([email protected]).Cell lines and culture. Human colorectal cancer (CRC) cell lines Caco-2, COLO205, DLD-1, HT29, LS174T, LS411N, NCI-H747, SK-CO-1, SW48, SW480,SW620, SW1116 and SW1417 were from ATCC (LGC Standards, Middlesex, UK). HCA7, HCT8 and SW837 cells were from ECACC (Culture Collections, Public Health England, Salisbury, UK). CO115 cells were provided by Dr. Richard Hamelin (INSERM, Paris, France) and LoVo by Professor Kevin Ryan (The Beatson Institute for Cancer Research, Glasgow, UK). HCT116 cells and isogenic derivatives lacking BAK and/or BAX were provided by Bert Vogelstein and Richard Youle (NIH, Bethesda, Maryland, USA).Human melanoma cell lines A2058, CHL-1, HMCB, MeWo, PMWK, RPMI- 7951, SK-MEL-2, SK-MEL-5, SK-MEL-28 and SK-MEL-31 were from ATCC. SK-MEL-30 cells were from DSMZ (Leibniz Institute DSMZ-German Collection of Microorganisms and Cell Cultures, Braunschweig, Germany). WM266-4 and WM852 cells were from ESTDAB (European Searchable Tumour Cell and Data Bank, Centre for Medical Research (ZMF), Tübingen, Germany). A375 cells were provided by Richard Marais (Cancer Research UK Manchester Institute, Manchester, UK). MEL-JUSO cells were provided by Professor Judith Johnson(Institute of Immunology, University of Munich, Munich, Germany). MEDI-MEL- 07 and MEDI-MEL-10 primary patient-derived human melanoma cells were provided by AstraZeneca (Gaithersburg, Maryland, USA).Human non-small cell lung cancer (NSCLC) cell lines A549, Calu-6, NCI-H520, NCI-H727, NCI-H838, NCI-H1299, NCI-H1437, NCI-H1793, NCI-H1869, NCI-H2122 and NCI-H2286 were from ATCC. HARA cells were from JCRB (Japanese Collection of Research Bioresources Cell Bank, National Institutes of Biomedical Innovation, NIBIO, Osaka, Japan) and HCC15 cells were from DSMZ. Human pancreatic adenocarcinoma cell lines AsPC-1, BxPC3, Capan-1, CFPAC-1, HPAC, HPAF-II, Hs 700T, MIA PaCa-2, Panc 02.03, Panc 03.27, Panc 04.03, Panc 08.13,Panc 10.05, PANC-1 and PSN1 were from ATCC. HuP-T4 and PSN1 cells were from ECACC, and YAPC cells were from DSMZ.

Cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) (A375, A549, A2058, CO115, DLD-1, HCA7, HCT116, Hs 700T, LoVo, MEDI-MEL-07, MEDI-MEL-10, MIA PaCa-2, PANC-1), IMDM (CFPAC-1), Leibovitz’s L-15 (SW48, SW480, SW620, SW837, SW1116, SW1417), McCoy’s 5A (HT29), MEM alpha (Calu-6, HMCB, HPAF-II, HuP-T4, LS174T, MeWo, PMWK, RPMI-7951, SK-MEL-2, SK-MEL-5, SK-MEL-28, WM266-4, WM852) or RPMI 1640 (AsPC-1, BxPC3, COLO205, HARA, HCC15, HCT8, LS411N, NCI-H520, NCI-H727, NCI- H747, NCI-H838, NCI-H1299, NCI-H1437, NCI-H1793, NCI-H1869, NCI-H2122, NCI-H2286, PSN1, SK-MEL-30, YAPC) media supplemented with 10% (v/v) fetal bovine serum, penicillin (100 U mL−1), streptomycin (100 mg mL−1) and 2 mM glutamine. DMEM for MEDI-MEL-07 and MEDI-MEL-10 cells was additionally supplemented with 0.1 mM non-essential amino acids and 25 mM HEPES. Leibovitz’s L-15 media was additionally supplemented with 0.75 mg mL−1 sodium bicarbonate. Caco-2 cells were grown in MEM alpha, 20% (v/v) fetal bovine serum, penicillin (100 U mL−1), streptomycin (100 mg mL−1) and 2 mM glutamine. Capan-1 cells were grown in IMDM, 20% (v/v) fetal bovine serum, penicillin (100 U mL−1), streptomycin (100 mg mL−1) and 2 mM glutamine.HPAC cells were grown in 1:1 DMEM/Ham’s F12 medium mix, 0.002 mg mL−1insulin (Sigma-Aldrich, Dorset, UK), 0.005 mg mL−1 transferrin (Sigma-Aldrich), 40 ng mL−1 hydrocortisone (Sigma-Aldrich), 10 ng mL−1 human recombinant epidermal growth factor (Sigma-Aldrich) and 5% fetal bovine serum, penicillin (100 U mL−1), streptomycin (100 mg mL−1) and 2 mM glutamine. Panc 02.03,Panc 03.27, Panc 04.03, Panc 08.13 and Panc 10.05 cells were grown in RPMI 1640 medium, 15% (v/v) fetal bovine serum, 1 U mL−1 insulin, penicillin (100 U mL−1), streptomycin (100 mg mL−1) and 2 mM glutamine. SK-MEL-31 cells were grown in MEM alpha medium, 15% (v/v) fetal bovine serum, penicillin (100 U mL−1), streptomycin (100 mg mL−1) and 2 mM glutamine. Cells were incubated in a humidified incubator at 37 °C and 5% (v/v) CO2. All cell lines were authenticated by short tandem repeat (STR) profiling and confirmed negative for mycoplasma prior to experiments commencing.

Unless stated otherwise, the above reagents were from Gibco, Thermo Fisher Scientific (Paisley, UK).A375 cells with acquired resistance to PLX4720 (A375 PLX-R), selumetinib (A375 Sel-R), or combined PLX4720 and selumetinib (A375 PLX/Sel-R1 and A375 PLX/Sel-R2) were provided by AstraZeneca. Cells were grown in RPMI 1640 media supplemented with 10% (v/v) fetal bovine serum, penicillin (100 U mL−1), streptomycin (100 mg mL−1), 2 mM glutamine and containing 0.66 μM PLX4720 (A375 PLX-R), 0.5 μM selumetinib (A375 Sel-R), 0.33 μM PLX4720 plus 0.25 μM selumetinib (A375 PLX/Sel-R1) or 0.66 μM PLX4720 plus 0.5 μM selumetinib (A375 PLX/Sel-R2).Mice. All animal studies were conducted in accordance with the National Institute of Health guidelines for animal research or with UK Home Office regulations, UK Animals (Scientific Procedures) Act of 1986 and approved by AstraZeneca’s Institutional Animal Care and Use Committee (IACUC). Animals were kept in a pathogen-free, AAALAC (Association for the Assessment and Accreditation of Laboratory Animal Care) or UK Home Office approved facilities on a 12 h light/ dark cycle. The ambient temperature was 21 ± 2 °C and the humidity was 55 ± 10%. Mice were housed at 3–5 animals per cage with air exchange and cages were changed weekly. Food and water were provided ad libitum.Inhibitor compounds and cell culture treatments. AZD4320 (CAS: 1357576–48–7), AZD5991 (CAS: 2143010–83–5), AZ6197 (compound 35)50 and selumetinib (CAS: 606143–52–6) were supplied by AstraZeneca. A-1155463 (CAS: 1235034–55–5) and Q-VD-OPh (CAS: 1135695–98–5) were purchased from Adooq Bioscience (Irvine, California, USA). PLX4720 (CAS: 918505–84–7), SCH772984 (CAS: 942183–80–4),trametinib (CAS: 871700–17–3) and vemurafenib (CAS: 918504–65–1) were pur- chased from Selleck Chemicals (Houston, Texas, USA). Venetoclax (CAS: 1257044–40–8) was purchased from MedChemExpress (New Jersey, USA). All compounds were dissolved in dimethyl sulfoxide (DMSO) to a stock concentration of 10 mM.

For inhibitor compound treatments, cells were seeded into dishes or plates in their normal growth medium (containing the relevant inhibitors for cells with acquired resistance) and allowed to settle for 24 h. For experiments with resistant cells, all treatment groups were washed with complete media only and then treated with fresh media containing the indicated compounds or media containing vehicle- only. Vehicle DMSO concentrations were normalised so that they were equivalent for all treatments.Generation of PLX4720- and/or selumetinib-resistant cells. To generate A375 cells with acquired resistance to either PLX4720 or selumetinib, cells at 50% confluence were treated with a GI50 concentration of PLX4720 (66 nM) or selu- metinib (50 nM). Upon confluence, the cells were split and the inhibitor con- centration increased a 1 × GI50 step (i.e., from 66 to 132 nM PLX4720 and 50 to 100 nM selumetinib). This was continued for ~3–4 months until the cells grew normally in 10 × GI50 PLX4720 (0.66 μM; A375 PLX-R) or selumetinib (0.5 μM; A375 Sel-R).To generate cell lines in parallel to those above with acquired resistance to both PLX4720 and selumetinib, A375 cells at 50% confluence were treated with a GI50 concentration of selumetinib (50 nM). Upon confluence the cells were split and additionally treated with 1 × GI50 PLX4720 (66 nM). Compound concentrations were alternately increased a 1 × GI50 step following splitting until after 4 months the cells grew normally in 5 × GI50 PLX4720/selumetinib (0.33 μM PLX4720 and0.25 μM selumetinib; A375 PLX/Sel-R1). The same procedure was used to generate an independent cell line in parallel with resistance to 10 × GI50 PLX4720/ selumetinib (0.66 μM PLX4720 AZD5991 and 0.5 μM selumetinib; A375 PLX/Sel-R2).