A Feed-Forward Mechanosignaling Loop Confers Resistance to Therapies Targeting the MAPK Pathway in BRAF-Mutant Melanoma Free

Reciprocal feedback between the ECM and tumor cells influence the hallmarks of cancer by providing biological abilities to malignant cells that are required for growth, survival, and dissemination. The ECM is a dynamic network of macromolecules with distinctive biochemical and mechanical properties that plays a major role in establishing tumor niches (1). Increased ECM deposition, fiber alignment, and covalent cross-link between collagen molecules lead to tumor stiffening, which has been associated to an elevated risk of cancer and poor clinical outcome in patients with breast or pancreatic cancers (2, 3).

Cancer-associated fibroblasts (CAF) are the main producers of tumorigenic ECM and function like myofibroblasts during wound healing and fibrosis (4). Cells apply contractile forces to sense the physical environmental stiffness through integrin-based focal adhesion (FA) complexes that connect the actin–myosin cytoskeleton with the ECM (2, 5). Matrix rigidity also leads to enhanced nucleus localization and activity of the mechanical-responsive Yes-associated protein (YAP) transcriptional regulator of the Hippo pathway (6). In CAFs, YAP acts as a critical factor regulating force-mediated ECM remodeling towards increased stiffening (7). Similar to YAP, the SRF transcriptional coactivator MRTF is translocated to the nucleus upon actin polymerization and functionally interacts with YAP to coordinate mechanosignaling and CAF contractility (8, 9). Beside, YAP mainly through its interaction with TEAD transcription factors have been shown to promote resistance to RAF/MEK–targeted cancer therapies in tumor cells such as melanoma (10–12).

Because of its resistance to treatment and propensity for metastasis, cutaneous melanoma is one of the most aggressive human cancers (13). Melanoma comprises phenotypically heterogeneous subtypes of cancer cells that can switch between transcriptional programs and differentiation states (14–16). The majority of melanomas display genetic alterations in BRAF or NRAS, leading to constitutive activation of the MAPK pathway. MAPK pathway inhibitors, such as BRAF inhibitors (BRAFi), MEK inhibitors (MEKi), or their combination, achieve significant clinical benefits in patients with BRAFV600-mutant melanoma. However, most patients relapse within months due to the acquisition of drug resistance attributed to intrinsic genetic and nongenetic changes in melanoma cells. Although genetic resistance frequently result from the reactivation of the MAPK pathway through de novo mutations, such as NRAS mutations (17, 18), nongenetic mechanisms involve epigenetic and/or transcriptomic changes in tumor cells during the early phase of treatment (19, 20). Such mechanisms often result in a dedifferentiation cell state characterized by downregulation of the master regulator of melanocyte differentiation microphthalmia-associated transcription factor (MITF) and upregulation of receptor tyrosine kinases (RTK) such as AXL (21–23). In addition, the dedifferentiated resistant MITF^low/AXL^high population was shown to display a mesenchymal invasive phenotype (24–26). Transcriptional reprogramming of proliferative drug-sensitive melanoma cells into invasive drug-resistant cell population is thus a critical event in acquired resistance to targeted therapies.

Beside tumor cell-autonomous events, there is evidence that extrinsic factors derived from the microenvironment contribute to melanoma resistance to MAPK pathway inhibition. Stromal cells including CAFs and macrophages secrete growth and inflammatory factors, and ECM components such as fibronectin, which contribute to drug tolerance (27–31). Interestingly, melanoma cells have the ability to secrete their own matrix, in particular upon cellular transition to a de-differentiated mesenchymal state occurring in response to BRAF inhibition (20, 32, 33). In this study, we asked whether melanoma cell-derived ECM impacts on tumor mechanics and contributes to resistance to targeted therapies. We show that both acquired resistance and early adaptation to MAPK signaling inhibition paradoxically induces a force-mediated ECM reprogramming in melanoma cells that increases intrinsic mechanical sensing properties and alters ECM composition and topography. This fuels a mechanical positive-feedback loop where melanoma cell-derived ECM and YAP/MRTF intracellular pathways play a pivotal role and that could favor the reservoir of therapy-resistant cells.

Melanoma cell lines 501Mel and MNT1 were obtained as described previously (34, 35). 1205Lu cells were from Rockland. Isogenic pairs of vemurafenib-sensitive (P) and -resistant (R) cells (M229, M238, M249) were provided by R. Lo (21). Cells were cultured in DMEM plus 7% FBS (Hyclone). Resistant cells were continuously exposed to 1 μmol/L vemurafenib. Cell lines were used within 6 months between resuscitation and experimentation. Cell lines were authenticated via STR profiling (Eurofins Genomics) and were routinely tested for the absence of Mycoplasma by PCR. For live imaging, M238P, 501Mel, and 1205Lu were transduced with NucLight Red lentivirus reagent (Essen Bioscience) and selected with puromycin (1 μg/mL). Culture reagents were purchased from Thermo Fisher Scientific. BRAFi (PLX4032, vemurafenib), MEKi (GSK1120212, trametinib), and ROCK inhibitor Y27632 were from Selleckem. YAP inhibitor verteporfin was from Sigma.

siGENOME siRNA SMARTpools for YAP1, MRTFa, and nontargeting control were from Dharmacon (Horizon Discovery). Fifty nmol/L of either siRNA pool was transfected using Lipofectamine RNAiMAX (Thermo Fisher Scientific), following the manufacturer's protocol.

Cell lysates were subjected to immunoblot analysis as described previously (35). The following antibodies were used at dilution of 1:1,000, unless otherwise stated: type I collagen and smooth muscle actin-α (αSMA) (Abcam); TAGLN2 (Genetex); PDGFRβ (Cohesion Biosciences); EGFR and LOXL2 (Bio-Techne); LOX (Novus Biological); MITF (Thermo Fisher Scientific); fibronectin, thrombospondin (TSP1), β1 integrin, FAK, paxillin, FAP, and MRTF (BD Biosciences); SPARC (Haematologic Technologies); ERK1/2, HSP90, HSP60, MLC2 (Santa Cruz Biotechnology); AXL, YAP, phospho-paxillin (Y118), phospho-ERK1/2 (T202/Y204), phospho-Rb (S807/811), Rb, p27KIP1, caveolin-1, survivin, and tubulin (Cell Signaling Technology).

Three-dimensional (3D) ECMs were generated as described previously (36). Briefly, gelatin-coated culture dishes were seeded with cells and cultured for 8 days in complete medium, supplemented with 50 μg/mL ascorbic acid every 48 hours. Cell cultures were then washed with PBS and matrices were denuded following a 2-minute treatment with prewarmed extraction buffer (PBS 0.5% Triton X-100, 20 μmol/L NH₄OH). Matrices were then gently washed several times with PBS. For drug-protection assays, melanoma cells were seeded onto decellularized matrices for 24 hours, and cultured for another 48-hour period in presence or not of indicated drugs.

Cell-cycle profiles were determined by flow cytometry as described previously (34). Proliferation was measured by a MTS conversion assay (34) or followed by live imaging of NucLight Red-stained cells using the IncuCyte ZOOM system (Essen BioScience) or by nuclei quantification of Hoescht-stained cells.

A total of 5 × 10⁴ melanoma cells were embedded in 100 μL of collagen I/Matrigel and seeded on a glass bottom 96-well plate (MatTek). Once the gel was set (1 hour at 37°C), cells were maintained in DMEM 10% FBS with or without indicated drugs. Gel contraction was monitored at day 3. The gel area was measured using ImageJ software and the percentage of contraction was calculated using the formula 100 × (well diameter − gel diameter)/well diameter as described (37).

Contractile forces were assessed by traction force microscopy (TFM) as described (38) using collagen-coated polyacrylamide hydrogels with shear modulus of 4 kPa coated with red fluorescent beads (SoftTrac; Cell Guidance Systems). Cells were plated on bead-conjugated gels for 48 hours. Images were acquired before and after cell removal using a fluorescence microscope (Leica DMI6000, ×10 magnification). Tractions exerted by cells were estimated by measuring beads displacement fields, computing corresponding traction fields using Fourier transformation and calculating root-mean-square traction using the particle image velocity plugin on ImageJ. The same procedure was performed on a cell-free region to measure baseline noise.

Cells were grown on collagen-coated polyacrylamide/bisacrylamide synthetic hydrogels with defined stiffness as described (39), then rinsed, fixed in 4% formaldehyde, and incubated in PBS 0.2% saponin 1% BSA in PBS for 1 hour with 1:100 dilution of the indicated primary antibodies. Following incubation with Alexa Fluor-conjugated secondary antibodies (1:1,000), hydrogels were mounted in Prolong antifade (Thermo Fisher Scientific). F-actin was stained with Texas Red-X or Alexa Fluor-488 phalloidin (1:100; Thermo Fisher Scientific). Nuclei were stained with DAPI. Images were captured on a widefield microscope (Leica DM5500B, ×40 magnification). Cell area and roundness and orientation of fibronectin fibers were assessed on immunofluorescence images using ImageJ. Nuclear/cytosolic ratio of YAP or MRTF was assessed by measuring the fluorescence intensity of nucleus and cytosol and quantified using ImageJ. The corresponding DAPI staining image was used to delimit nuclear versus cytosolic regions.

Collagen deposition and organization were visualized by standard Masson's trichrome staining or picrosirius red staining accordingly to (see Supplementary Materials and Methods for details; ref. 40). Second harmonic generation (SHG) and multiphoton-fluorescence images were acquired on a Zeiss 780NLO (Carl Zeiss Microscopy) with Mai Tai HP DeepSee (Newport Corporation). Acquisitions were achieved simultaneously in backward through 10× dry NA 0.45 objective and forward through condenser NA 0.55. Each side is equipped with dual NDD GaAspP detectors (BiG) with 440/10 (for SHG forward and backward) and 525/50 filter (for autofluorescence). Transmission images were acquired with 514 nm laser through the 525/50 filter.

Mouse experiments were carried out in accordance with the Institutional Animal Care and the local ethical committee (CIEPAL-Azur agreement NCE/2014-179). A total of 1 × 10⁶ melanoma cells were subcutaneously implanted into both flanks of 6-week-old female athymic nude nu/nu mice (Envigo). When tumor reached 100 mm³, mice were randomly grouped into control and test groups. The BRAFi group received six intraperitoneal injections of vemurafenib (35 mg/kg) over a period of 2 weeks. Verteporfin was delivered intraperitoneally three times per week at 45 mg/kg. Mice in the control group were treated with vehicle. At the end of the experiment, mice were sacrificed, tumors were dissected, weighed, and either snap frozen in liquid nitrogen (for mRNA and protein analysis), in Tissue-Tek O.C.T. (VWR; for AFM analysis) or formalin fixed and paraffin embedded for picrosirius red or Masson's trichrome staining, SHG analysis, and IHC.

Patient-derived xenograft (PDX) models treated or not with the BRAFi–MEKi combination as described previously (see Supplementary Materials and Methods for details; ref. 41) were established by TRACE (PDX platform; KU Leuven) using tissue from melanoma patients undergoing surgery at the University Hospitals KU Leuven. Written informed consent was obtained from all patients and all procedures were approved by the UZ Leuven Medical Ethical Committee (S54185/S57760/S59199) and carried out in accordance with the principles of the Declaration of Helsinki. All procedures involving animals were performed in accordance with the guidelines of the Institutional Animal Care and Use Committee of KU Leuven and within the context of approved project applications P147/2012, P038/2015, P098/2015, and P035/2016. Formalin-fixed and paraffin-embedded tumor biopsies were sectioned for picrosirius red staining.

Mechanical properties of tumor sections were analyzed by atomic force microscopy (AFM) as described previously (42) with a Bioscope Catalyst operating in Point and Shoot (Bruker Nano Surfaces), coupled with an inverted optical microscope (Leica DMI6000B; Leica Microsystems Ltd.). The apparent Young's modulus (E_app) was measured on unfixed frozen tumor sections using a Borosilicate Glass spherical tip (5 μm of diameter) mounted on a cantilever with a nominal spring constant of 0.06 N/m (Novascan Technologies). The force–distance curves were collected using a velocity of 2 μm/s, in relative trigger mode and by setting the trigger threshold to 1 nN. E_app values were presented in a boxplot using GraphPad Prism (GraphPad software).

Public datasets of human melanoma cell lines developing drug resistance to vemurafenib (M229R and SKMel28R) and double resistance to vemurafenib and selumetinib (M229DDR and SKMel28DDR) were used to analyze gene levels compared with drug-naive parental cell lines (GSE65185; ref. 19). Differential gene expression was also examined in datasets derived from tumor biopsies from melanoma patients before and after development of drug resistance to BRAFi, MEKi, or BRAFi/MEKi combination [GSE50535 (25); Tirosh and colleagues (15)]. Normalized data were prepared using MeV software.

Statistical analysis was performed using GraphPad Prism. Unpaired two-tailed Mann–Whitney test were used for statistical comparisons between two groups and Kruskal–Wallis test with Dunn posttests or two-way analysis of variance test with Bonferroni posttests to compare three or more groups. Error bars are ±SD.

To investigate whether acquired resistance to BRAFi modifies mechanosensing pathways, we exploited models of isogenic pairs of parental (P) and resistant (R) melanoma cells showing either reactivation of MAPK pathway through NRAS mutation (M249R) or upregulation of AXL, EGFR, and PDGFRβ RTKs associated with low levels of MITF and reduced differentiation of melanoma cells (M229R, M238R; Supplementary Fig. S1; ref. 20). Cells were cultured on collagen-coated hydrogels with stiffness ranging from 0.2kPa (low), 4kPa (medium) to 50kPa (high; ref. 39). In contrast to parental sublines, a dramatic modification of M238R (Fig. 1A and B) and M229R (Supplementary Figs. S2A and S2B) cell morphology measured by actin reorganization, cell roundness, and area was noticeable upon increased substrate stiffness. In contrast, the shape and actin cytoskeleton of the NRAS-mutated M249R subline and its parent M249P showed no significant changes in response to mechanical stimulation (Fig. 1C and D; Supplementary Figs. S2C and S2D). Importantly, MITF^low/AXL^high BRAFi-resistant M229R or M238R cells, but not NRAS-mutated M249R cells, exhibited an enhanced capacity to proliferate on a collagen-coated stiff substrate (Fig. 1E; Supplementary Fig. S2E).

The β1 integrin/FA pathway is essential for ECM mechanosignaling (43). Consistently, when compared with drug-sensitive cells, M238R and M229R cells expressed higher levels of β1 integrin and increased phosphorylation of FA components, including FAK, p130Cas, and paxillin (Supplementary Fig. S1). In addition, M238R cells displayed higher number of FAs upon increased matrix rigidity compared with parental cells (Supplementary Fig. S3).

YAP and MRTF are critical transcriptional mediators of mechanical signals through partially overlapping signaling pathways and target genes (6, 7, 9, 44). Immunofluorescence analysis of melanoma cells plated on soft or rigid substrates revealed that in contrast to M238P cells, M238R cells showed higher levels of nuclear YAP (Fig. 2A and B) on low stiffness substrate (0.2kPa). Nuclear YAP and MRTF markedly increased in M238R cells plated on medium (4kPa) and high (50kPa) substrate stiffness, whereas translocation of YAP and MRTF was only apparent when parental cells were plated on stiff substrate. Consistently, expression of shared YAP/MRTF target genes paralleled increasing collagen rigidity in M238R, but not M238P cells (Fig. 2C). Furthermore, impairment of the actomyosin cytoskeleton with the ROCK inhibitor Y27632 reduced the nuclear localization of YAP and MRTF in M238R cells plated on high stiffness substrate (Fig. 2D). Accordingly, ROCK inhibition abrogated the expression of two shared YAP/MRTF target genes CTGF and CYR61 activated in M238R cells on stiff substrate (Fig. 2E).

Finally, to evaluate the potential contribution of ECM stiffness-induced YAP/MRTF activation in MITF^low/AXL^high associated resistance, M238R cells cultured on rigid collagen hydrogels were transfected with siRNA pool targeting YAP or MRTF and treated with increasing doses of BRAFi (vemurafenib). The sensitivity of M238R cells to BRAFi-induced cell proliferation arrest was partially restored upon YAP or MRTF knockdown, suggesting that collagen stiffening through YAP and MRTF activation contributes to acquired resistance (Fig. 2F). Together, these results indicate that the dedifferentiated MITF^low/AXL^high-resistant cell state is associated with a mechanophenotype.

Further functional analysis of the dedifferentiated MITF^low mesenchymal resistant state revealed that M229R and M238R cells were characterized by high expression levels of typical CAF markers such as caveolin-1 (CAV1), myosin light-chain 2 (MLC2), αSMA, fibroblast activation protein (FAP), transgelin-2 (TAGLN2), in addition to ECM proteins collagen 1 (COL1) and fibronectin (Fig. 3A). In contrast, parental and mutant NRAS-driven resistance M249R cell lines showed low or no expression of such markers. We thus examined whether MITF^low/AXL^high-resistant cells display CAF-associated features such as ROCK-dependent actomyosin contractility and force-mediated ECM remodeling leading to fibers organization (7, 37). We first compared traction stresses generated by sensitive and BRAFi-resistant cells using TFM and observed that M238R cells applied stronger forces on collagen-coated stiff matrices than their drug-sensitive parental counterparts (Fig. 3B). Next, we performed collagen gel contraction assays to assess cell contractility. Contractility in 3D collagen was observed for M238R, but not for M238P cells. Inhibition of ROCK by Y27632 or YAP by Verteporfin reduced the capacity of M238R cells to contract collagen gels to levels that were observed for drug-sensitive M238P cells (Fig. 3C). Moreover, siRNA-mediated knockdown of YAP or MRTF abrogated the contractile activity of drug-resistant M238R cells (Fig. 3D).

Given that increased cellular forces lead to matrix fiber organization and that BRAFi-resistant mesenchymal cells secrete high levels of ECM proteins (20, 21), we analyzed the topography of the fibronectin and collagen network generated by this resistant cellular state. We compared ECM proteins differentially produced and deposited by M238P and M238R cells. Cell-derived 3D matrices were generated, denuded of cells, and analyzed by quantitative mass spectrometry. Compared with M238P cells, M238R cells assembled a matrix that was enriched in ECM glycoproteins (fibronectin, fibrilin-1, thrombospondin-1, and fibulin-1/2), collagens, proteoglycans (versican and biglycan), as well as collagen-modifying enzymes such as transglutaminase 2 and LOXL2 (Supplementary Table S1). Furthermore, in contrast to parental cells, M238R cells assembled fibronectin and collagen fibers oriented in parallel patterns that resembled those produced by TGFβ-activated fibroblasts (Fig. 3E). Fibronectin fibers organization was quantified by measuring the relative orientation angle of fibers. The percentages were 16.5%, 23.7%, and 27.8% for M238P, M238R, and fibroblasts 3D ECM, respectively (Fig. 3F). Importantly, the lower degree of ECM production by parental cells was not due to a difference in proliferation as evidenced by nuclear and fibronectin stainings of M238P and M238R cell cultures before the decellularization process (Supplementary Fig. S4). Together, these results suggest that MITF^low/AXL^high BRAFi-resistant cells display increased traction forces and contractility, leading to aligned organization of ECM fibers.

Given our observations so far, we explored publicly available expression array studies searching for mechanosignaling, and cell contractility gene expression in drug-resistant human melanoma cells. Data extracted from the GEO database (GSE65185; ref. 19) showed increased levels of several YAP/MRTF target genes (THBS1, CYR61, CTGF, AMOTL2, ANKRD1, and SERPINE1) together with high levels of ECM genes (COL1A1, COL1A2, and FN1) and mesenchymal markers (PDGFRB, MYL9, ACTA2, FAP, and TAGLN) in MITF^low/AXL^high cells developing drug resistance to vemurafenib (BRAFi; M229R and SKMel28R) and double resistance to vemurafenib and selumetinib (BRAFi + MEKi; M229DDR and SKMel28DDR) as compared with drug-sensitive parental cells (Fig. 3G). Moreover, analysis of gene expression on tumor biopsies from patients progressing during therapy with BRAFi and/or MEKi [GSE50535 and Tirosh and colleagues (Supplementary Information); refs. 15, 25] revealed that expression of ECM and mechanosignaling genes markedly increased in a subset of relapsing patients with MITF^low/AXL^high expression (Fig. 3G).

We next questioned whether adaptive response to MAPK pathway inhibition involves mechanosensing pathways and ECM remodeling. BRAF-mutant melanoma cells (1205Lu and M238P) were plated on collagen-coated hydrogels and treated with the BRAFi vemurafenib or the MEKi trametinib (Fig. 4A; Supplementary Fig. S5). In both cases, drug-treated cells displayed pronounced morphological and actin cytoskeleton changes that were accompanied by increased YAP and MRTF nuclear localization (Fig. 4A; Supplementary Figs. S5A and S5B), and transcriptional activation of the YAP/MRTF shared target gene CYR61 relative to untreated cells (Supplementary Fig. S5C). Moreover, drug-treated M238P and 1205Lu cells displayed significantly higher number of FAs with larger size as compared with control cells (Supplementary Fig. S6). We further confirmed that a short-term treatment of cells with either BRAFi or MEKi increased the expression of collagen 1 (COL1) and fibronectin and of the YAP/MRTF target thrombospondin-1 (TSP1), along with reduced phosphorylation of RB and increased expression of p27KIP1, two cell-cycle markers that are modulated by MAPK pathway inhibition (Fig. 4B). Importantly, short-term treatment with BRAFi or MEKi was sufficient to increase the contractile activity of 1205Lu cells embedded in collagen gels (Fig. 4C). Consistently, when cultivated one week in the presence of vemurafenib, 1205Lu cells assembled an organized ECM composed of collagen and fibronectin fibers that were anisotropically oriented, as compared with untreated cells (Fig. 4D). A further indication of the involvement of mechanopathways in adaptation of melanoma cells to MAPK inhibition was brought by the observation that M238P, 1205Lu, and 501Mel cells cultivated on stiff collagen-coated substrates were significantly more resistant to increasing doses of BRAFi as compared with cells cultivated on soft substrates (Fig. 4E). Together, these results demonstrate that melanoma cells rapidly adapt to MAPK pathway inhibition by acquiring an ECM-remodeling contractile phenotype associated with increased mechanosignaling pathways.

The findings described above support the notion that a subset of BRAF-mutant melanoma cells in response to early and late MAPK pathway inhibition acquire the capacity to produce and remodel a matrix reminiscent to CAF-derived ECM. Because ECM plays a major role in mediating drug resistance, we hypothesized that melanoma cell-derived matrix functions as a supporting niche for melanoma cell behavior. To investigate the effect of melanoma cell-derived ECM on survival and resistance to targeted therapies, drug naïve BRAF-mutant melanoma cells were plated on 3D matrices generated from parental cells (M238P and M229P) or their BRAFi-resistant counterparts (M238R and M229R), and treated or not with vemurafenib alone or the combination vemurafenib/trametinib (Fig. 5; Supplementary Fig. S7). Time-lapse monitoring of 501Mel proliferation revealed that matrices derived from MITF^low/AXL^high mesenchymal BRAFi-resistant cells significantly reduced the proliferation arrest induced by MAPK pathway inhibition in contrast to ECMs from BRAFi-sensitive cells, which had no impact on the cytostatic action of BRAF and MEK inhibition (Fig. 5A and B; Supplementary Figs. S7A and S7B). Cell-cycle analysis further confirmed the protective action of matrices from mesenchymal resistant, but not parental cells, over the G₀–G₁ cell-cycle arrest induced by BRAFi on drug-naive 501Mel and MNT1 cells (Fig. 5C; Supplementary Figs. S7C and S7D). At the molecular level, matrix-mediated therapeutic escape from BRAF inhibition was associated in both 501Mel and MNT1 cells with sustained levels of the proliferation marker phosphorylated-RB and of survivin, low levels of the cell-cycle inhibitor p27KIP1 together with maintained phosphorylation of ERK1/2 in presence of the drug (Fig. 5D; Supplementary Figs. S7E and S7F). Importantly, similar biochemical events were promoted in 501Mel cells escaping from the combination of BRAFi and MEKi upon adhesion to M238R-derived, but not M238P-derived ECM (Fig. 5E). Next we wondered if short-term MAPK pathway inhibition fosters a drug-protective ECM program in melanoma cells. 501Mel cells were plated on matrices generated from vehicle or vemurafenib-treated 1205Lu cells, and treated with or without BRAFi. Cell cycle and biochemical analysis showed that BRAF inhibition rapidly promoted the production by 1205Lu cells of an ECM that significantly counteracted the cytostatic action of vemurafenib in 501Mel cells (Fig. 5F and G).

Finally, we investigated the involvement of the mechanoresponsive YAP and MRTF transcriptional pathways in ECM-mediated drug protection. 501Mel cells were cultured on matrices prepared from parental M238P or drug-resistant M238R cells and the subcellular location of YAP and MRTF was examined by immunofluorescence microscopy. In contrast to ECM from M238P cells, matrices derived from M238R cells promoted the nuclear translocation of YAP and MRTF (Fig. 6A), and their transcriptional activation as indicated by the increased expression of ANKRD1 and SERPINE1 genes (Fig. 6B). Consistently, drug protective action provided by matrices derived from therapy-resistant M238R cells against BRAFi or the combination BRAFi/MEKi was dramatically reduced in 501Mel cells in which either YAP (Fig. 6C) or MRTF (Fig. 6D) expression was knocked-down. Depletion of YAP or MRTF enhanced the efficacy of MAPK pathway inhibition as shown by reduced levels of phosphorylation of ERK1/2 and RB and increased expression of p27KIP1 (Fig. 6E and F). These suggest that melanoma cell-derived ECM mediates drug protection through YAP and MRTF regulation.

Collectively, our findings demonstrate that both early and late adaptation to MAPK pathway inhibition involves the mechanical reprogramming of melanoma cells, leading to the assembly of an organized matrix that confers de novo resistance to targeted therapies in a YAP and MRTF-dependent manner.

Exposure of BRAF-mutant melanoma cells to MAPK pathway inhibition promotes a mechanophenotype associated with drug tolerance in vitro, which could have important outcomes for disease progression in vivo. To address this, we first explored whether BRAF inhibition induces ECM remodeling in human melanoma xenograft models. BRAF-mutant melanoma cells 1205Lu or M229P were xenografted into nude mice (melanoma CDX), which were treated blindly with either vehicle or vemurafenib (Supplementary Fig. S8A). As expected, BRAF targeting induced a strong inhibition of tumor growth (Supplementary Figs. S8B and S8C). Histologic, transcriptomic, and biophysic analyses were then performed at the experiment end point. Vemurafenib treatment triggered a profound remodeling of the 1205Lu (Fig. 7A) and M229P (Supplementary Fig. S8D) tumor stroma, with a marked increase of collagen fibers area and thickness, as measured by polarized light of picrosirius red-labeled tumors and SHG microscopy (Fig. 7A and B; Supplementary Fig. S8D). We then examined gene expression on BRAFi-treated melanoma tumors by performing RT-qPCR analysis using human and mouse probes. Consistent with a previous study (31), vemurafenib was found to significantly activate tumor-associated host stromal cells. However, compared with untreated tumors, tumors exposed to BRAFi also dramatically upregulated human mesenchymal and ECM genes, including genes for collagens (COL1A1, COL3A1, COL5A1, COL15A1), fibronectin (FN1), collagen-modifying enzyme (LOX), and myofibroblast markers (SPARC, ACTA2), as well as YAP and/or MRTF target genes, such as AXL, CYR61, SERPINE1, AMOTL2, and THBS1 (Fig. 7C). This observation supports the notion that BRAF inhibition can promote a cancer cell-autonomous mechanism of ECM production in vivo. Consistent with the changes in ECM composition and assembly, vemurafenib treatment significantly increased tumor elastic modulus in the two CDX models when measured by AFM (Fig. 7D; Supplementary Fig. S8E), suggesting that ECM stiffening constitute an adaptive response of melanoma cells to MAPK pathway inhibition in vivo. We next wished to validate these observations in melanoma PDXs. PDX exposed or not to the combination of BRAFi and MEKi were stained with picrosirius red (Fig. 7E). Combined BRAF and MEK inhibition also resulted in a marked accumulation of collagen fibers in the tumor stroma of melanoma PDX (Fig. 7E and F). Finally, Verteporfin, a FDA-approved drug used in photodynamic therapy for macular degeneration and a known inhibitor of YAP was used to interrogate if YAP contributes to BRAFi-induced collagen remodeling and therapy response in vivo. Although Verteporfin alone did not affect 1205Lu tumor growth, cotreatment with vemurafenib plus Verteporfin had a greater antitumor effect than vemurafenib alone after 17 days of drug regimens (Fig. 7G and H). Thus, combined Verteporfin and vemurafenib therapy enhanced vemurafenib response in a preclinical melanoma model. Furthermore, Masson's trichrome and picrosirius red stainings revealed that Verteporfin treatment abrogated the accumulation of collagen fibers induced by BRAF inhibition in the stroma of melanoma xenografts (Fig. 7I and J). Together these data suggest that YAP mechanosensing pathway contributes to collagen reorganization in response to MAPK pathway inhibition and support the concept of a combinatorial approach to overcome ECM-mediated therapy resistance in BRAF-mutated melanoma models.

A major resistance program in melanomas exposed to MAPK-targeting therapies is linked to a dedifferentiated, mesenchymal transcriptional cell state characterized by low levels of the melanoma differentiation factor MITF and high levels of AXL (15, 19, 21–24). MITF^low/AXL^high-resistant cells exhibit multiple traits of the Hoek's invasive gene signature (14), including prominent expression of ECM proteins (20, 21). Here we showed that this resistant cell population also exhibits key aspects of CAFs involved in ECM remodeling: they acquired a mechanical phenotype associated with an actomyosin/YAP/MRTF-dependent contractile activity, and the ability to deposit ECM to create a tumor-permissive environment. In contrast, drug-naive cells and a population of MITF^high/NRAS-mutant resistant cells displayed no such mechanoresponsive features and ECM remodeling activities. Importantly, we also found that early adaptation to MAPK pathway inhibition promotes de novo acquisition of a CAF-like phenotype, leading to biomechanical reprogramming both in vitro and in vivo. We thus uncover a previously unidentified feed-forward loop between drug-exposed or resistant MITF^low/AXL^high melanoma cells and ECM remodeling to increase tumor tissue stiffness, mechanosensing and resistance through YAP and MRTF regulation (Fig. 7K).

Short-term treatment of melanoma cells with targeted drugs induced actin dynamics, mechanosensitive regulation of YAP and MRTF and increased cell contractility. This differs from another early adaptation state to BRAF inhibition characterized by the emergence of a slow-cycling NGFR/CD271^high persistent cell population (20). However, our results are in line with the observation that BRAFi modulates actin reorganization and YAP/TAZ activation (11) as well as Rho GTPase signaling (45). Thus, our findings underscore the exquisite phenotypic plasticity of melanoma cells and the notion that their biomechanical reprogramming may actively participate to intratumor heterogeneity and therapeutic escape.

Another indication of the ability of targeted therapies to switch melanoma cells towards a CAF-like phenotype is based on our findings that BRAFi induces melanoma cells to autonomously remodel a fibrillar and drug-protective ECM, an additional trait typical of CAFs. A previous study has shown that short-term BRAF inhibition upregulates adhesion signaling and drug tolerance in BRAF-mutant/PTEN-null melanoma cells (46). Extending this observation, our data demonstrate that short-term MAPK pathway inhibition induces the assembly by melanoma cells of an aligned ECM containing collagens, fibronectin and thrombospondin-1, indicating that targeted therapies have the capacity to rapidly exacerbate the intrinsic ability of melanoma cells to produce a pro-invasive ECM (32, 47). Vemurafenib treatment was shown to activate CAFs to generate a drug-tolerant niche through fibronectin-mediated integrin β1/FAK signaling (31). In this study, cell death following BRAF inhibition was reduced when melanoma cells were cultured on stiff substrates containing the combination of fibronectin, thrombospondin-1, and tenascin-C (31). A part from ECMs assembled by therapy-activated fibroblasts, our study reveals a crucial role of fibronectin and collagen-rich ECMs derived from either drug-resistant or drug-exposed melanoma cells in driving tolerance. The protection against the cytostatic effect of MAPK inhibition brought by melanoma-derived matrices is evidenced by the persistence of cycling cells, with sustained levels of proliferative markers and YAP/MRTF nuclear translocation. Remarkably, tolerance to BRAFi was achieved when BRAF-mutant melanoma cells were plated on collagen-coated stiff matrices, supporting the notion that, in addition to fibronectin (31, 46), the collagen network and ECM stiffening are major mediators of melanoma drug resistance. Interestingly, previous studies with bioengineered materials have shown the impact of substrate stiffness on targeted drugs responses in melanoma (48) and carcinoma cell lines (49).

YAP-TEAD and MRTF-SRF pathways functionally interact to coordinate mechanosignaling required for the maintenance of the CAF phenotype in solid tumors (7–9, 50). Similarly, we showed that the contractile behavior of the dedifferentiated resistant melanoma cells requires YAP and MRTF expression. Importantly, we found that YAP and MRTF are activated upon mechanical stress and contribute to ECM-mediated drug resistance. This is in agreement with recent reports demonstrating the contribution of the YAP pathway in BRAFi resistance (10–12). However, these studies were conducted on rigid plastic dishes that do not reflect tissue mechanical compliance. In contrast, we demonstrated the exacerbated ability of dedifferentiated resistant and BRAFi-exposed melanoma cells to adapt to substrate rigidity using cell-derived 3D ECMs and collagen-coated hydrogels with defined stiffness, which model more accurately the activation of YAP and MRTF mechanosensors. In contrast to YAP-TEAD pathway, the role of MRTF-SRF pathway in melanoma therapeutic resistance remains less defined. MRTF controls several cytoskeletal genes, including αSMA and MLC2 (8) that we found enriched in the MITF^low/AXL^high resistant cells and in MITF^low tumor biopsies from progressing patients with melanoma. Moreover, several components of the matrisome from MITF^low resistant cells, such as tenascin-C, CYR61, thrombospondin-1, and serpine1 are known YAP and/or MRTF targets (9). Remarkably, a YAP1 enrichment signature has also been identified as a driver event of melanoma acquired resistance (19). This is in line with our in silico gene expression analyses that revealed a similar trend towards an increased expression of YAP/MRTF target genes in MITF^low tumor biopsies from patients relapsing from therapy. Of note, a recent study identified AXL, a RTK required to maintain the resistant phenotype in melanoma (24), in a YAP/TAZ target gene signature (51). Accordingly, we found several YAP/MRTF target genes including AXL induced upon BRAFi treatment in our xenograft model. This raises the possibility that the reservoir of AXL^high resistant cells is promoted by biomechanical adaptation of melanoma cells to oncogenic BRAF inhibition. Interestingly, collagen stiffening has been recently shown to promote melanoma differentiation via YAP/PAX3-mediated MITF expression (52). This study and our present report support the emerging notion that collagen density and rigidity is a key microenvironmental factor that governs melanoma cell plasticity and intratumor heterogeneity. How YAP and MRTF actually coordinate mechanical signals from tumor microenvironments to drive melanoma differentiation, invasive behavior or drug resistance is currently unknown and requires further investigations.

Importantly, our data reveal a targetable vulnerability of vemurafenib-induced mechanical reprograming of melanoma in vivo. Tumors treated with BRAFi or combined BRAFi/MEKi therapy displayed an intense remodeling of the tumor niche associated to increased collagen fibers organization and YAP/MRTF-mediated gene expression. Earlier studies have underscored the critical role of CAFs activated by BRAF inhibition for the development of resistant niches (27, 28, 31, 53). Accordingly, we found that host stromal cells that likely include fibroblasts produce some ECM genes in response to vemurafenib. However, we demonstrated that the molecular changes associated with the dramatic remodeling of the tumor niche in response to MAPK pathway inhibition also results from the activation of human melanoma cells, thereby promoting an autocrine production of a rigid ECM enriched in collagen fibers. In line with the key role of the YAP pathway during melanoma relapse (19) and phenotypic heterogeneity (12), we found that YAP-TEAD inhibition by Verteporfin reverses vemurafenib-induced excessive collagen deposition. Consequently, treatment with Verteporfin cooperated with vemurafenib to reduce melanoma growth. Whether targeting MRTF-SRF signaling pathway may also demonstrate therapeutic efficiency is currently under investigation.

In conclusion, our findings disclose a novel mechanism of BRAF-mutant melanoma cells adaptation to MAPK-targeted therapies through the acquisition of an auto-amplifying CAF-like phenotype in which melanoma cell-derived ECM modulates mechanosensing pathways to promote tumor stiffening. In addition to therapy-induced tumor secretomes (54), therapy-induced mechanical phenotypes could endow cancer cells with unique cell-autonomous abilities to survive and differentiate within challenging tumor-associated microenvironments, thereby contributing to drug resistance and relapse. Our results suggest that cancer cell-ECM interactions and tumor mechanics provide promising targets for therapeutic intervention aimed at enhancing targeted therapies efficacy in melanoma.

No potential conflicts of interest were disclosed.

Conception and design: C.A. Girard, R. Ben Jouira, J.-C. Marine, M. Deckert, S. Tartare-Deckert

Development of methodology: C.A. Girard, M. Lecacheur, R. Ben Jouira, I. Berestjuk, S. Diazzi, C. Gaggioli, M. Deckert

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Ben Jouira, V. Prod'homme, A. Mallavialle, F. Larbret, M. Gesson, S. Schaub, S. Pisano, S. Audebert, B. Mari, E. Leucci, J.-C. Marine

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.A. Girard, M. Lecacheur, R. Ben Jouira, V. Prod'homme, M. Gesson, S. Schaub, S. Audebert, B. Mari, M. Deckert, S. Tartare-Deckert

Writing, review, and/or revision of the manuscript: C.A. Girard, M. Lecacheur, R. Ben Jouira, J.-C. Marine, M. Deckert, S. Tartare-Deckert

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Deckert, S. Tartare-Deckert

Study supervision: M. Deckert, S. Tartare-Deckert

We thank R.S. Lo for melanoma cells. We acknowledge the iBV, IRCAN, and C3M partners of “Microscopie Imagerie Côte d'Azur” (MICA) GIS-IBISA multisites platform supported by the GIS IBiSA, Conseil Départemental 06, and Région PACA. We also acknowledge the C3M animal facility and we thank TRACE (PDX platform at the KULeuven University) for providing the PDX models. This work was supported by institutional funds from Institut National de la Santé et de la Recherche Médicale (Inserm), Université Côte d’Azur, the Ligue Contre le Cancer (Equipe labellisée Ligue Contre le Cancer 2016 to S. Tartare-Deckert), and Institut National du Cancer (INCA_12673 to S. Tartare-Deckert). Funding from the Fondation ARC, National Research Agency (#ANR-18-CE14-0019-01 to M. Deckert), ITMO Cancer Aviesan within the framework of the Cancer Plan, and the French Government through the “Investments for the Future” LABEX SIGNALIFE (#ANR-11-LABX-0028-01) is also acknowledged. We also thank financial supports by Conseil Départemental 06 and Canceropôle PACA. R. Ben Jouira was a recipient of a doctoral fellowship from Fondation ARC. I. Berestjuk is a recipient of a doctoral fellowship from La Ligue Contre le Cancer.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.