Introduction

One of the hallmarks of cancer cells is their remarkable ability to adapt to microenvironmental influences, such as the nature of the stroma including the extracellular matrix (ECM) and therapeutic stress (Pickup et al, 2014). This is particularly true for malignant cutaneous melanoma, which is one of the most aggressive and refractory human cancers (Shain & Bastian, 2016). Approximately 50% of melanoma carries activating mutations in the BRAF oncogene, leading to the activation of the mitogen-activated protein kinase (MAPK)/ERK pathway. Inhibition of the BRAFV600E/K oncoprotein by BRAF inhibitors (BRAFi) such as vemurafenib or dabrafenib has markedly improved clinical outcome of patients (Flaherty et al, 2012). Despite this, durable responses are rare as most patients relapse within a year of beginning the treatment. Significant prolonged benefit can be achieved by combining BRAFi and MEK (MAPK/ERK kinase) inhibitors (MEKi) such as cobimetinib or trametinib, yet the development of drug resistance remains the most common clinical outcome (Robert et al, 2019). Acquired resistance to targeted therapies involves genetic alterations in key intracellular regulators of the MAPK signaling pathway. This leads to the restoration of the pathway and non-genetic alterations that are commonly associated with transcriptional reprogramming and phenotype switching from a melanocytic to an invasive undifferentiated mesenchymal-like cell state, which is characterized by lower expression levels of MITF and SOX10 and higher levels of AXL (Muller et al, 2014; Rambow et al, 2019). Such adaptive responses to BRAF oncogenic pathway inhibition are thought to precede mutation-driven acquired resistance (Smith et al, 2016).

However, in addition to mechanisms of resistance intrinsic to cancer cells, dynamic, de novo mechanisms exist, which are orchestrated by the tumor microenvironment and occur during the cancer cell’s adaptation to therapy. Environment-mediated drug resistance (EMDR) thus appears as an important contributor to how cancer cells escape therapies (Meads et al, 2009). This process has initially been described in multiple myeloma and other hematopoietic malignancies and was shown related to minimal residual disease. This phenomenon is gaining importance in the field of melanoma with several studies reporting the involvement of stroma-derived factors in adaptive response and resistance to targeted therapies (Straussman et al, 2012; Fedorenko et al, 2015; Hirata et al, 2015; Kaur et al, 2016; Young et al, 2017). Given the key role of EMDR enabling the emergence of genetic resistance, an understanding and further identification of EMDR mechanisms in melanoma may assist with the development of more effective therapeutic strategies, thereby increasing the efficacy of targeted therapies.

Tumors are complex and adaptive ecosystems that are affected by numerous stromal components, which enhance tumor phenotypes and therapy resistance. Cancer-associated fibroblasts (CAFs) are activated fibroblasts and the primary producers of ECM. The ECM is a highly dynamic structural framework of macromolecules, providing both biochemical and biomechanical cues, which are required for tumor progression (Kalluri, 2016). The ECM is primarily composed of fibrillar and non-fibrillar collagens, hyaluronic acid, proteoglycans, and adhesive glycoproteins such as fibronectin, thrombospondins, and SPARC. It also contains matrix-remodeling enzymes and other ECM-associated proteins and acts as a reservoir for cytokines and growth factors (Hynes & Naba, 2012; Mouw et al, 2014). ECM composition, fiber orientation, and physical characteristics are profoundly altered in the vast majority of solid tumors (Pickup et al, 2014). Interactions between cells and the ECM elicit intracellular signaling pathways and regulate gene transcription, mainly through cell-surface adhesion receptors including integrins and discoidin domain receptors (DDR). DDR1 and DDR2 belong to a unique subfamily of receptor tyrosine kinases and have been identified as non-integrin collagen receptors (Shrivastava et al, 1997; Vogel et al, 1997; Leitinger, 2014). They are distinguished from each other by their relative affinity for different types of collagens, as DDR1 is activated by both fibrillar and non-fibrillar collagens, whereas DDR2 is only activated by fibrillar collagens. Furthermore, their expression and function are associated with fibrotic disease and cancer (Valiathan et al, 2012; Leitinger, 2014). DDR1 or DDR2 is known to control tumor cell proliferation and invasion, depending on the tumor type and the nature of the microenvironment (Valiathan et al, 2012). However, the functional role of DDR activity in mediating sensitivity to anti-cancer therapies and tumor resistance is poorly documented.

Adhesion of tumor cells to the ECM is a key component of EMDR. However, the influence of matrix-mediated drug resistance (MMDR) in response to targeted therapies and the nature of ECM receptors driving the MMDR phenotype in melanoma have not yet been addressed in detail. To model the contribution of the ECM in melanoma cell responses to BRAF and MEK inhibition, we generated fibroblast-derived 3D ECM from melanoma-associated fibroblasts (MAFs), which were isolated from patient-derived biopsies and analyzed the MMDR mechanism with the aim to identify novel opportunities for microenvironment-targeted therapies. Here, we show that DDR1 and DDR2 are key mediators of MMDR in melanoma, through the pro-survival non-canonical NF-κB2 pathway. Our findings reveal a novel role for these collagen-activated tyrosine kinase receptors, in mediating BRAF inhibitor tolerance. These data therefore support the rationale to inhibit DDR1 and DDR2 signaling, to disrupt the therapy-resisting properties conferred by the ECM in the microenvironment. We propose that the use of DDR inhibitors as a novel combinatorial therapeutic strategy may be beneficial for melanoma patients in overcoming resistance to MAPK-targeted therapy.

Results Fibroblast-derived 3D ECM confers drug-protective action to melanoma cells against anti-BRAFV600E therapies

To investigate the potential contribution of MMDR to targeted therapy in BRAF-mutated melanoma cells, we employed an in vitro model based on live cell-derived 3D ECMs. These matrices mimic many structural and biomolecular features, which are typically found in vivo (Cukierman et al, 2001). We selected human primary fibroblasts obtained from healthy individuals or MAFs isolated from patient metastatic melanoma biopsies (MAFs) in either the skin or lymph nodes (LN). The different fibroblast cultures were functionally tested in a 3D collagen matrix contraction assay, which showed that unlike human dermal fibroblasts (HDF), skin and LN MAF displayed actomyosin contractile activity. Similar to MAFs, LN normal fibroblasts known as fibroblastic reticular cells (FRC) are myofibroblast-like cells (Fletcher et al, 2015), which also showed a high propensity to contract collagen (Fig 1A). Cell-derived matrices were then generated and de-cellularized, and their composition, architecture, and rigidity were analyzed using proteomic and microscopic approaches. In our experimental conditions, compared to HDF, skin and LN MAF, as well as FRC, produced and assembled a dense 3D ECM composed of oriented collagen and fibronectin fibers, as shown by picrosirius red and immunofluorescence staining of the ECMs (Fig 1B). Proteomic analysis of the different fibroblast-derived ECMs further documented the molecular composition of these matrices, showing enrichment for several types of collagens and core matrisome components including glycoproteins, proteoglycans, and ECM regulators and ECM-associated proteins (Appendix Fig S1). Atomic force microscopy (AFM) analysis of ECM stiffness revealed values for MAF and HDF matrices that were within the range of previous observations (Kaukonen et al, 2016; Fig 1C). We noticed that matrices generated from FRC and MAF were stiffer than HDF-derived ECM. Collectively, these observations validate the use of our experimentally derived matrices for functional studies. Next, we tested the effectiveness of the fibroblast-derived ECMs generated from HDF, FRC, or MAF to protect BRAFV600E-mutated melanoma cells against the anti-proliferative effect of MAPK pathway inhibition. We therefore developed a drug-protective assay based on the culture of melanoma cells that stably expressed a fluorescent nuclear label, cultured on top of fibroblast-derived ECMs (Fig 2A). Tumor cells were then treated with drugs targeting the mutant BRAF/MAPK pathway using BRAFi alone or in combination with MEKi (Fig 2B). Cell proliferation was monitored using live cell time-lapse imaging and quantified by counting the number of fluorescent nuclei. Cell growth inhibition induced by BRAFi alone (vemurafenib) or in combination with MEKi (trametinib) was abrogated when 1205Lu melanoma cells were cultured on top of fibroblast-derived ECMs, in sharp contrast to standard cell culture conditions where cells were plated either on plastic or on purified collagen 1 (Coll-1) (Figs 2C and D, and EV1A). In line with the organization of the 3D matrices depicted in Fig 1, drug-protective assays against BRAFi or BRAFi/MEKi combo-therapy revealed that MAF- and FRC-derived matrices display higher protective abilities compared to HDF-derived ECMs and conferred increased protection (Figs 2D and EV1A). Similar protective effects were observed with the SKMEL5 cell line and the MM099 patient-derived short-term melanoma culture when these cells were plated on the different experimental ECM settings (Figs 2E and EV1B). These data suggest that MMDR relies on the topological and molecular features of the ECM. Cell cycle analysis in SKMEL5 and MM099 cells further showed that experimentally produced ECM from HDF, MAF, or FRC prevented the G0/G1 cell cycle arrest, induced by the BRAFi/MEKi cocktail in contrast to the cell culture conditions on plastic or Coll-1-plated dishes (Figs 2F and EV1C). The ECM therefore transmits signals that prevent the cytostatic action of MAPK pathway inhibitors. At the molecular level, ECM-mediated therapeutic escape of SKMEL5, 1205Lu, and MM099 cells from BRAF pathway inhibition was associated with sustained levels of the proliferation markers phosphorylated Rb, cyclin D1, and survivin, and lower levels of the cell cycle inhibitor p27KIP1, although ERK1/2 phosphorylation was similarly decreased in the presence of the targeted drugs in all culture conditions (Figs 2G and EV1D). Immunoblot analysis also suggested that upon BRAF inhibition, the levels of proliferative markers were higher in melanoma cells on MAF- and FRC-derived ECMs than on HDF-derived ECMs. In contrast, no significant changes in p53 levels were observed. Together, these results indicate that fibroblasts assembled and remodeled matrices that provide a drug-tolerant environment for BRAF mutant melanoma cell lines and short-term cultures.

Figure 1. Composition, topology, and mechanical properties of fibroblast-derived 3D ECMs

Images show collagen matrix gel contraction by HDF (human dermal fibroblasts), skin-MAF (melanoma-associated fibroblasts isolated from skin lesions), LN-FRC (lymph node fibroblast reticular cells), and LN MAF (melanoma-associated fibroblasts isolated from metastatic lymph node). Dashed circles represent the diameter of the gel. Data are representative of n = 3 independent experiments. Immunofluorescence analysis of fibronectin (green) and collagen (red) fibers on de-cellularized ECM produced by human fibroblasts. Fiber orientation was quantified using ImageJ software. Percentages indicate oriented fibers accumulated in a range of ± 21° around the modal angle. Data are represented as mean ± s.d. (n = 10 random fields from 2 independent determinations). Scale bar, 50 µm. A representative image of picrosirius red staining from 10 analyzed images is shown for each condition. Atomic force microscopy (AFM) measurement of the elastic properties (apparent Young’s modulus, Εapp) of fibroblast-derived ECMs. Each dot represents a specific Young's modulus obtained by fitting the corresponding individual force curve acquired on a determined point of the sample. A representative experiment from 2 independent experiments is shown. Scatter plots show mean ± SEM. The black bars represent the median and the interquartile range. ****P < 0.0001, two-tailed Mann–Whitney test.

Figure 2. Fibroblast-derived 3D ECM confers drug-protective action to melanoma cells against anti-BRAFV600 therapies

A. Scheme of the ECM-mediated drug-protective assay. B. Illustration of the BRAFV600E pathway and the MAPK pathway inhibitors used in the study. C. Time-lapse imaging of proliferation of NucLight-labeled 1205Lu cells plated on plastic (left panel), Coll-1 (collagen 1; middle panel), or FRC-derived ECM (right panel) and treated with vehicle, 5 µM BRAFi, or 2 µM BRAFi plus 0.01 µM MEKi using the IncuCyte ZOOM system. Each data point represents the mean of NucLight red nuclear objects per field ± SEM. ****P < 0.0001, two-way ANOVA followed by Dunnett’s multiple comparisons test. Data are representative of n = 3 independent experiments. D, E. Quantification of proliferation of 1205Lu (D) and SKMEL5 (E) cells plated for 48 h on plastic, Coll-1, or the indicated fibroblast-derived ECMs prior to a 96-h treatment with vehicle, 5 µM BRAFi, or 2 µM BRAFi plus 0.01 µM MEKi. Cells were counted by Hoechst-labeled nuclei staining. Data are represented as bar plots with mean ± SEM normalized to vehicle of 3 independent experiments. (D) **P = 0.015, ***P = 0.0009, and ****P < 0.0001; and (E) *P = 0.0204, ***P = 0.0006, and ****P < 0.0001, the Kruskal–Wallis test followed by Dunn’s multiple comparisons test. F. Flow cytometry analysis of cell cycle distribution of SKMEL5 cells cultured on plastic, Coll-1, or the indicated fibroblast-derived ECMs and treated with vehicle or 2 µM BRAFi combined with 0.01 µM MEKi. The percentage of cells in different phases of the cell cycle is indicated. G. Immunoblotting of protein extracts from SKMEL5 cells cultivated as described above on plastic, Coll-1, or the indicated fibroblast-derived ECMs in the presence or not of BRAFi or BRAFi/MEKi for 96 h, using antibodies against P-ERK1/2, ERK2, or cell cycle markers (P-Rb, Rb, cyclin D1, p27KIP1, survivin, and p53). HSP60, loading control.

Source data are available online for this figure.

Click here to expand this figure.

Figure EV1. Fibroblast-derived 3D ECM confers drug-protective action to melanoma cells against anti-BRAFV600E therapies

Time-lapse imaging of proliferation of NucLight-labeled 1205Lu cells using the IncuCyte ZOOM system. Cells were plated for 48 h on HDF- or MAF-derived ECMs prior to a 96-h treatment with vehicle, 5 µM BRAFi, or 2 µM BRAFi plus 0.01 µM MEKi. Each data point represents the mean of NucLight red nuclear objects per field ± SEM. **P = 0.0079 (left panel), **P = 0.0012 (right panel), and ****P < 0.0001, two-way ANOVA followed by Dunnett’s multiple comparisons test. Data are representative of n = 3 independent experiments. Quantification of proliferation of MM099 short-term melanoma cell cultures plated for 48 h on plastic, Coll-1, or the indicated fibroblast-derived ECMs prior to a 96-h treatment with vehicle, 5 µM BRAFi, or 2 µM BRAFi plus 0.01 µM MEKi. Cells were counted by Hoechst-labeled nucleus staining. Data are represented as bar plots with mean ± SEM normalized to vehicle of 3 independent experiments. *P = 0.0416 and ****P < 0.0001, two-way ANOVA followed by Dunnett’s multiple comparisons test. Flow cytometry analysis of cell cycle distribution of MM099 cells cultured on indicated substrates and treated with vehicle, 5 µM BRAFi, or 2 µM BRAFi plus 0.01 µM MEKi. The percentage of cells in different phases of the cell cycle is indicated. Immunoblotting of protein extracts from 1205Lu cells (left panel) and MM099 cells (right panel) cultivated on indicated substrates in the presence or not of 5 µM BRAFi or 2 µM BRAFi plus 0.01 µM MEKi for 96 h, using antibodies against P-ERK1/2, ERK2, P-Rb, Rb, E2F1, survivin, p27KIP1, cyclin D1, and p53. HSP60, loading control. Expression of the collagen receptors DDR1 and DDR2 in melanoma

Previous studies have demonstrated the critical role of ECM receptors belonging to the integrin family in drug resistance (Seguin et al, 2015). Moreover, BRAF inhibition has been described to generate a drug-protective stroma with high β1 integrin/FAK signaling, as a result of the paradoxical action of BRAFi on MAF (Hirata et al, 2015). Yet, in our experimental conditions, we were unable to show a significant implication of the β1 integrin/FAK axis in drug protection conferred by fibroblast-derived ECMs. Indeed, the addition of blocking β1 integrin antibodies and the depletion of FAK both failed to prevent the protective properties of fibroblast-derived ECM against the growth and survival inhibitory signals induced by BRAFi (Appendix Fig S2).

This finding prompted us to interrogate the contribution of other ECM receptors in targeted therapy resistance. Keeping in mind the elevated levels of fibrillar collagens found in fibroblast-derived ECMs (Fig 1B; Appendix Fig S1), we examined the functional implication of the collagen tyrosine kinase receptors DDR1 and DDR2 (Shrivastava et al, 1997; Vogel et al, 1997). The analysis of TCGA datasets for cutaneous melanoma showed that the DDR1 and DDR2 genes were genetically altered in 20% and 13% of melanoma cases, respectively. Interestingly, a significant fraction of melanomas was found to be associated with an amplification of DNA copy number and higher mRNA levels of DDR1 and DDR2 (respectively 13% and 10% of samples) (Fig 3A). This is consistent with the notion that these collagen receptors may play an important role in melanoma pathogenesis. Immunohistochemical analysis of DDR1 and DDR2 expression in benign nevi and malignant primary and metastatic melanocytic skin lesions further showed that DDR1 and DDR2 levels significantly increased during melanoma progression, indicating that DDR1 and DDR2 may represent novel prognostic factors for melanoma (Fig 3B; Appendix Fig S3). We next examined the levels of DDR1 and DDR2 in a collection of melanoma cell lines and short-term melanoma cultures in relation to the cell state differentiation markers MITF, SOX10, and AXL. DDR1 and DDR2 were both expressed in melanoma cell lines regardless of their differentiation of cell phenotype (Fig 3C). In patient-derived short-term melanoma cultures, higher DDR1 and DDR2 protein levels were detected in BRAF mutant MM099 and MM029 and NRAS mutant MM165 cells with the MITFlow, SOX10low, and AXLhigh de-differentiated phenotype signature (Fig 3D). Moreover, higher levels of DDR2 were found to be associated with lower levels of the melanocytic marker MITF and higher levels of the drug-resistant marker AXL in de-differentiated mesenchymal-like BRAFi-resistant M229R, M238R, and UACC62R cells compared to their parental counterparts (Nazarian et al, 2010; Girard et al, 2020; Misek et al, 2020) (Fig 3E). The examination of public gene expression datasets of the melanoma differentiation signature confirmed that DDR1 and DDR2 levels were increased in the undifferentiated (U) and neural crest-like (NC) cell subpopulations from the TSOI signature (Fig 3F; Tsoi et al, 2018) and in the invasive MITFlow cells from the HOEK signature (Appendix Fig S4; Widmer et al, 2012). De-differentiated/undifferentiated melanoma cells display intrinsic resistance to MAPK pathway inhibition (Muller et al, 2014; Tsoi et al, 2018; Rambow et al, 2019). In line with this notion, we found that DNA amplification and elevated mRNA levels of DDR1 negatively correlated to the activity of BRAF and MEK inhibitors in melanoma cell lines from the GDSC (Genomic of Drug Sensitivity in Cancer) (Appendix Fig S5). Together, these observations associate DDR expression with melanoma progression and with the invasive and therapy-resistant phenotype.

Figure 3. Expression of DDR1 and DDR2 in human melanoma

A. Meta-analysis of 363 cutaneous melanoma from TCGA (skin cutaneous melanoma, PanCancer Atlas) (http://www.cbioportal.org/) showing the percentage of samples with genetic alterations in DDR1 and DDR2. Cases with missense (green) and truncating (blue) mutations, amplification (red), and mRNA overexpression (pink) are indicated; gray, individual cases. B. Immunohistochemical analysis of DDR1 and DDR2 levels on human melanoma tissue microarrays. Representative IHC images and quantification (right bar histograms) of DDR1 and DDR2 expression in normal skin, nevus, primary melanoma (PM), and lymph node melanoma metastases (MM). Scale bar, 100 µm. Histological scoring of the samples was performed in a blinded fashion. Samples were scored as low, medium, or high for DDR1 or DDR2 expression (nevus, n = 12; PM, n = 30; and MM, n = 20). C–E. Immunoblotting of equal amounts of protein extracts from melanoma cell lines (C), patient-derived short-term cell cultures (D), or isogenic pairs of parental-sensitive and BRAFi-resistant cell lines (E) using antibodies against DDR1, DDR2, or markers of the melanoma cell differentiation AXL, MITF, or SOX10. ERK2, loading control. F. DDR1 and DDR2 levels increase in de-differentiated melanoma cells. Box-and-whisker plots show DDR1, DDR2, MITF, AXL, and SOX10 expression among four differentiation melanoma cell states (U, undifferentiated, n = 10; NC, neural crest-like, n = 14; T, transitory, n = 12; and M, melanocytic, n = 17) (GSE80829). The expression of AXL, MITF, and SOX10 is shown as control markers of cell differentiation. n, number of cell lines representative of each cell state. Central bars represent the median, and the whiskers, the 10th to 90th percentile of the boxplot. Multiple comparisons were performed using ordinary one-way ANOVA.

Source data are available online for this figure.

Targeting DDR impairs ECM-mediated resistance to oncogenic BRAF pathway inhibition

Next, we compared how fibroblast-derived ECMs modulate DDR phosphorylation, an event linked to their activity (Shrivastava et al, 1997; Vogel et al, 1997). Immunoblot analysis of lysates from 1205Lu and MM099 cells plated on MAF- and FRC-derived ECMs indicated that MAF-derived matrices have a stronger ability to increase the levels and phosphorylation of DDR1 and/or DDR2 compared to HDF-derived matrices (Fig EV2A), supporting a functional implication of DDR in melanoma drug tolerance. Interestingly, while BRAF and/or MEK inhibition had no significant effect on DDR phosphorylation in 1205Lu cells cultivated on MAF-derived ECM, MEK inhibition seemed to increase DDR1 expression levels (Appendix Fig S6). To address the contribution of DDR in MMDR, a siRNA approach was then used to target DDR1, DDR2, or both in melanoma cells cultured on MAF- or FRC-derived ECMs, in the presence of BRAFi alone or in combination with MEKi. Immunoblot analysis showed specific DDR1 and DDR2 protein reduction after siRNA transfection using two different targeted sequences in BRAFi-treated 1205Lu cells cultured on MAF- or FRC-derived ECMs (Figs 4A and EV2B). Compared to the single knockdown, the simultaneous knockdown of DDR1 and DDR2 overcame MMDR to BRAF-targeted therapy as revealed by decreased levels of cell proliferation markers, including phosphorylated Rb, survivin, and E2F1, in 1205Lu and SKMEL5 cells (Fig 4A and B). Importantly, depletion of both DDR1 and DDR2 enhanced the cytotoxic activity of co-targeting BRAF/MEK as shown by the increased cleavage of apoptotic caspase-3 that was detected in SKMEL5 and MM099 melanoma cells (Figs 4B and EV2C). DDR are druggable receptors targeted by imatinib, a tyrosine kinase inhibitor (TKI) initially developed as an ABL inhibitor, which also inhibits DDR activity with high efficacy (Day et al, 2008). Imatinib belongs to therapeutic molecules used in the clinic for the treatment of chronic myeloid leukemia and acute lymphoblastic leukemia with the Philadelphia chromosome (Druker et al, 2001). Enzymatic activities of DDR were also inhibited by other small molecules including DDR1-IN-1 (Kim et al, 2013). We first confirmed that imatinib and DDR1-IN-1 efficiently inhibited type I collagen-induced DDR1 and DDR2 tyrosine phosphorylation in 1205Lu cells (Fig EV2D). Inhibition of DDR1/2 kinases by imatinib or DDR1-IN-1 suppressed the protective action of MAF- and FRC-derived ECMs against BRAF inhibition, as evidenced by the significant decrease in 1205Lu cell proliferation, which was observed after co-treatment with BRAFi and DDR1/2 inhibitors (Figs 4C and EV2E). Similar anti-proliferative effects of DDR inhibitors were observed during drug-protective assays performed in SKMEL5 and MM099 cells on MAF- or FRC-derived ECMs (Fig EV2F and G). Mutant BRAF and DDR1/2 co-targeting in the three melanoma cell lines plated on FRC- or MAF-derived ECMs decreased levels of phosphorylated Rb, E2F1, and survivin, and induced caspase-3 cleavage (Figs 4D and EV3A and B). Similar biochemical events were promoted by BRAFi or with the combined BRAFi and MEKi treatment, in the presence of nilotinib, a next-generation BCR-ABL inhibitor that is approved for the treatment of imatinib-resistant patients and targeting DDR1/2 (Day et al, 2008; Fig EV3C). Induction of apoptosis in co-treated melanoma cells was further confirmed using flow cytometry analysis of cell death markers (Figs 4E and EV3D).

Click here to expand this figure.

Figure EV2. Knockdown and pharmacological inhibition of DDR1 and DDR2 abrogate ECM-mediated resistance to BRAFV600E pathway inhibition

Immunoblotting of protein extracts from 1205Lu and MM099 cells cultivated on HDF- or MAF-derived matrices using antibodies against P-DDR1, P-DDR1/P-DDR2, DDR1, and DDR2. β-actin, loading control. Immunoblotting of protein extracts of siCTRL-, siDDR1#2-, siDDR2#2-, or siDDR1#2/siDDR2#2-transfected 1205Lu cells plated on FRC-derived ECM and treated or not with 5 µM BRAFi for 96 h, using antibodies against DDR1, DDR2, P-MEK1/2, P-ERK1/2, ERK2, P-Rb, and survivin. HSP60, loading control. Immunoblotting of protein extracts of siCTRL-, siDDR1#1/siDDR2#1, or siDDR1#2/siDDR2#2-transfected MM099 melanoma short-term cultures plated on FRC-derived ECM and treated with vehicle or 2 µM BRAFi combined with 0.01 µM MEKi for 96 h, using antibodies against DDR1, DDR2, P-ERK1/2, ERK2, and cleaved caspase-3. HSP60, loading control. Immunoblot analysis of collagen I-induced DDR1 and DDR2 tyrosine phosphorylation. 1205Lu cells were incubated with 10 µg/ml of Coll-I (collagen I) in the presence or not of 7 µM imatinib or 1 µM DDR1-IN-1 for 18 h. After cell lysis, DDR2 phosphorylation was analyzed with anti-P-DDR2 following immunoprecipitation (IP) with anti-DDR2 antibodies. DDR1 phosphorylation was analyzed in total cell lysates with anti-P-DDR1. HSP60, loading control. Time-lapse imaging of proliferation of NucLight-labeled 1205Lu cells plated for 48 h on FRC- or MAF-derived ECMs prior to treatment with 5 µM BRAFi in the presence or not of 7 µM imatinib (left panels) or 1 µM DDR1-IN-1 (right panels) for the indicated times. Each data point represents the mean of NucLight red nuclear objects per field ± SEM. One experiment representative of 3 independent experiments is shown. ****P < 0.0001, 2-way ANOVA followed by Dunnett’s multiple comparisons test. Quantification of proliferation of SKMEL5 cells plated for 48 h on FRC- (left panel) or MAF-derived ECMs (right panel) prior to a 96-h treatment with vehicle or 5 µM BRAFi in the presence or not of 10 µM imatinib or 5 µM DDR1-IN-1. Cells were counted by Hoechst-labeled nucleus staining. Data are represented as bar plots with mean ± SEM normalized to vehicle. ***P = 0.0002, the Mann–Whitney test (n = 3). Quantification of proliferation of MM099 melanoma short-term cultures plated for 48 h on FRC- (left panel) or MAF-derived ECMs (right panel) prior to a 96-h treatment with vehicle or 5 µM BRAFi in the presence or not of 10 µM imatinib or 3 µM DDR1-IN-1. Cells were counted by Hoechst-labeled nucleus staining. Data are represented as bar plots with mean ± SEM normalized to vehicle. ***P = 0.0002, the Mann–Whitney test (n = 3).

Figure 4. Inhibition of DDR1 and DDR2 by genetic or pharmacological approaches abrogates ECM-mediated resistance to BRAFV600 pathway inhibition

Immunoblotting of protein extracts from 1205Lu cells transfected with a siRNA control (CTRL) or two different sequences of siRNA (#1 and #2) directed against DDR1 or DDR2 alone or in combination prior to being cultivated on MAF-derived ECM and treated or not with 5 µM BRAFi for 96 h, using antibodies against the indicated proteins. HSP60, loading control. Immunoblotting of protein extracts from SKMEL5 cells transfected with siCTRL or the combination of siDDR1#2 and siDDR2#2 prior to being cultivated on FRC- or MAF-derived ECMs (left and right panels, respectively) and treated with vehicle, 5 µM BRAFi, or 2 µM BRAFi plus 0.01 µM MEKi, using antibodies against the indicated proteins. HSP60, loading control. Quantification of the time-lapse imaging of the proliferation of NucLight-labeled 1205Lu cells using the IncuCyte ZOOM system. Cells were plated for 48 h on FRC- or MAF-derived ECMs prior to a 96-h treatment with vehicle or 5 µM BRAFi in the presence or not of 7 µM imatinib or 1 µM DDR1-IN-1. The bar plots represent the mean normalized to vehicle of NucLight red nuclear objects per field ± SEM from 3 independent experiments performed in triplicate. ****P < 0.0001, two-way ANOVA followed by Sidak’s multiple comparisons test. Immunoblotting of protein extracts from 1205Lu, SKMEL5, and MM099 cells plated for 48 h on FRC-derived ECM prior to a 96-h treatment with vehicle or 5 µM BRAFi in the presence or not of imatinib (7 µM for 1205Lu, 10 µM for SKMEL5 and MM099) or DDR1-IN-1 (1 µM for 1205Lu, 5 µM for SKMEL5, and 3 µM for MM099), using antibodies against the indicated proteins. HSP60, loading control. Flow cytometry analysis of cell death (Annexin V/PI labeling) in 1205Lu cells plated on FRC-derived ECM and treated as above. Right bar plots show the distribution of cells (% of total) across the different forms of death.

Source data are available online for this figure.

Click here to expand this figure.

Figure EV3. Pharmacological inhibition of DDR abrogates ECM-mediated resistance to BRAFV600E pathway inhibition and induces cell death

Immunoblotting of protein extracts from 1205Lu cells cultivated on MAF-derived ECM, treated or not with 5 µM BRAFi and/or 7 µM imatinib using antibodies against P-Rb, P-ERK1/2, survivin, or cleaved caspase-3. HSP60, loading control. Immunoblotting of protein extracts from SKMEL5 cells (left panel) and MM099 melanoma short-term cultures (right panel) cultivated on MAF-derived ECM, treated or not with 5 µM BRAFi in combination with imatinib (10 µM) or DDR1-IN-1 (5 µM for SKMEL5 and 3 µM for MM099) using antibodies against P-ERK1/2, ERK2, P-Rb, Rb, E2F1, survivin, or cleaved caspase-3. HSP60, loading control. Immunoblotting of protein extracts from 1205Lu cells cultivated on FRC-derived ECM for 96 h in the presence of 5 µM BRAFi, 0.01 µM MEKi, or the combination of 2 µM BRAFi and 0.01 µM MEKi, in the presence or not of 10 µM imatinib or 5 µM nilotinib using anti-P-MEK1/2, P-ERK1/2, P-Rb, E2F1, survivin, or cleaved caspase-3 antibodies (n = 2). HSP60, loading control. Flow cytometry analysis of cell death (Annexin V/PI labeling) in SKMEL5 cells (left) and MM099 cells (right) plated on FRC-derived ECM and treated by the indicated drugs as described above. Data show the percentage of the different forms of cell death based on Annexin V/PI positivity.

Recent studies described that collagen binding to DDR leads to their activation and clustering into filamentous membrane structures that are associated with collagen fibrils (Yeung et al, 2019). We thus examined the clustering and spatial distribution of phosphorylated DDR in melanoma cells plated on collagen I-coated plastic dishes or on 3D ECMs, in response to oncogenic BRAF pathway inhibition. To detect DDR1 and/or DDR2, we used an antibody that specifically recognizes phosphorylated DDR1 at Y792 and an antibody that not only recognizes phosphorylated DDR2 at Y740 but also cross-reacts with the phosphorylated DDR1 (Yeung et al, 2019). Immunofluorescence staining of P-DDR1 and P-DDR1/2 revealed that 1205Lu cells, which were cultured on purified collagen I, displayed a globular dot-like distribution of these two receptors, whereas cells seeded on FRC-derived 3D ECMs exhibited a fraction of P-DDR distributed into linear membrane clusters (Fig 5A and B). Importantly, cell exposure to BRAFi or to BRAFi/MEKi combo-therapy dramatically increased the proportion of P-DDR1 and P-DDR1/2 containing linear clusters, on FRC-derived 3D ECM, as well as F-actin remodeling (Fig 5A and B).

Figure 5. Interaction of melanoma cells with 3D ECMs induces the linear clustering of phosphorylated DDR upon BRAFi/MEKi treatment

Representative images of 1,205 cells cultivated on collagen I (Coll-I) or FRC-derived ECM for 48 h prior to treatment with vehicle or 5 µM BRAFi or 2 µM BRAFi plus 0.01 µM MEKi for 96 h. Immunofluorescence for phospho-DDR1 (P(Y792)-DDR1) (red; left panels) and phospho-DDR1/2 (P(Y796)-DDR1/P(Y740)-DDR2) (red; right panels), F-actin (green), and nuclei (blue) is shown. Enlarged images of P-DDR1 and P-DDR1/2 immunostaining are shown. White arrows indicate P-DDR1 and P-DDR1/2 cell membrane linear clustering. Scale bar, 25 µm (enlarged images: scale bar, 10 µm). Quantification of globular versus linear clusters of phospho-DDR1 (left panels) and phospho-DDR1/2 (right panels) from immunofluorescence staining shown in (A) using ImageJ software. Prior to the quantification of DDR clusters, a “subtract background” function of ImageJ has been applied to all images. In order to quantify clusters, the IsoData threshold has been used. Clusters with circularity 0.3–1 have been defined as “globular”, and clusters with circularity 0–0.29 have been defined as “linear”. Data are from > 20 individual cells (n = 3). Error bars reflect mean ± s.d. Values for each treated condition are compared to the vehicle control. 2-way ANOVA followed by Dunnett’s multiple comparisons test. Quantification of globular versus linear clusters of Phospho-DDR1 and Phospho-DDR1/2 from immunofluorescence staining shown in Fig EV4A of 1205Lu cells cultivated on HDF- or MAF-derived ECM and treated with the indicated targeted drugs as described in (A). Data are from > 20 individual cells. Error bars reflect mean ± s.d. Values for each treated condition are compared to the vehicle control. 2-way ANOVA followed by Dunnett’s multiple comparisons test.

Source data are available online for this figure.

Notably, the formation of linear clusters of P-DDR1 and P-DDR1/2 upon MAPK pathway inhibition is also significantly increased in cells plated on MAF-derived 3D ECMs compared to cells cultured on HDF-derived ECMs (Figs 5C and EV4A), and these clusters co-localized with collagen fibers (Fig EV4B). Similar observations on phosphorylated DDR clustering were obtained with SKMEL5 melanoma cells (Fig EV4C and D). These data suggest that therapy-induced cytoskeletal changes drive a linear clustering of phosphorylated DDR along collagen fibers and subsequently cause MMDR. These findings therefore suggest that DDR1 and DDR2 determine BRAF mutant melanoma cell responsiveness to BRAF-targeted therapy, with regard to ECM features and that the drug-tolerant action of DDR is dependent on their enzymatic activities and correlates with the formation of filamentous membrane structures.

Click here to expand this figure.

Figure EV4. Interaction of melanoma cells with 3D ECM induces the clustering of phosphorylated DDR upon BRAFi or BRAFi/MEKi treatment

Representative images of 1205Lu cells cultivated on HDF- or MAF-derived ECM for 48 h prior to treatment with vehicle, 5 µM BRAFi, or 2 µM BRAFi combined with 0.01 µM MEKi for 96 h. Immunofluorescence for phospho-DDR1 (P(Y792)-DDR1) (red; left panels) or phospho-DDR1/2 (P(Y796)-DDR1/P(Y740)-DDR2) (red; right panels) is shown. Nuclei (blue) were stained with DAPI. Enlarged images of P-DDR1 and P-DDR1/2 immunostaining are shown. White arrows indicate P-DDR1 and P-DDR1/2 cell membrane linear clustering. Scale bar = 25 µm (enlarged images, scale bar = 10 µm). Analysis of co-localization of phospho-DDR with collagen 1 in 1205Lu cells cultivated on MAF-derived ECM and treated with BRAFi/MEKi. Immunofluorescence for phospho-DDR1 (P(Y792)-DDR1) (red; upper panels) and phospho-DDR1/2 (P(Y796)-DDR1/P(Y740)-DDR2) (red; lower panels), collagen 1 (green), and nuclei (blue) is shown. Enlarged images are shown. White arrows indicate co-localization (yellow fluorescence). Images were captured on Nikon Eclipse Ti confocal microscope at 60x magnification. Scale bar, 20 µm. Representative images of SKMEL5 cells cultivated on collagen I (Coll-I) or on indicated fibroblast-derived ECMs for 48 h prior to treatment with vehicle or 5 µM BRAFi or 2 µM BRAFi combined with 0.01 µM MEKi for 96 h. Immunofluorescence for phospho-DDR1 (P(Y792)-DDR1) (red; left panels) or phospho-DDR1/2 (P(Y796)-DDR1/P(Y740)-DDR2) (red; right panels) is shown. Nuclei (blue) were stained with DAPI. Enlarged images of P-DDR1 and P-DDR1/2 immunostaining are shown. Scale bar = 25 µm (enlarged images, scale bar = 10 µm). Quantification of globular (left panels) and linear (right panels) clusters of phospho-DDR1 and phospho-DDR1/2 from immunofluorescence staining shown in (B) using ImageJ software. Prior to the quantification of DDR clusters, a “subtract background” function of ImageJ has been applied to all images. In order to quantify clusters, the IsoData threshold has been used. Clusters with circularity 0.3–1 have been defined as “globular”, and clusters with circularity 0–0.29 have been defined as “linear”. Data are from > 20 individual cells. Error bars reflect mean ± s.d. Values for each treated condition are compared to the vehicle control. 2-way ANOVA followed by Dunnett’s multiple comparisons test. Pharmacological inhibition of DDR by imatinib improves targeted therapy efficacy, counteracts drug-induced collagen remodeling, and delays tumor relapse

The anti-tumor activity of imatinib combined with the BRAFi vemurafenib was assessed in a preclinical xenograft melanoma model. BRAF-mutated melanoma cells (1205Lu) were subcutaneously xenografted into nude mice (CDX model), which were exposed to BRAFi, imatinib, or BRAFi plus imatinib (Fig 6A). As expected, B