Blockade of pro-fibrotic response mediated by the miR-143/-145 cluster prevents targeted therapy-induced phenotypic plasticity and resistance in melanoma

Lineage dedifferentiation towards a mesenchymal-like state is a common mechanism of adaptive response and resistance to targeted therapy in melanoma. Yet, the transcriptional network driving this phenotypic plasticity remains elusive. Remarkably, this cellular state displays myofibroblast and fibrotic features and escapes MAPK inhibitors (MAPKi) through extracellular matrix (ECM) remodeling activities. Here we show that the anti-fibrotic drug Nintedanib/BIBF1120 is active to normalize the fibrous ECM network, enhance the efficacy of MAPK-targeted therapy and delay tumor relapse in a pre-clinical model of melanoma. We also uncovered the molecular networks that regulate the acquisition of this resistant phenotype and its reversion by Nintedanib, pointing the miR-143/-145 pro-fibrotic cluster as a driver of the therapy-resistant mesenchymal-like phenotype. Upregulation of the miR-143/-145 cluster under BRAFi/MAPKi therapy was observed in melanoma cells in vitro and in vivo and was associated with an invasive/undifferentiated profile of resistant cells. The 2 mature miRNAs generated from this cluster, miR-143-3p and miR-145-5p collaborated to mediate phenotypic transition towards a drug resistant undifferentiated mesenchymal-like state by targeting Fascin actin-bundling protein 1 (FSCN1), modulating the dynamic crosstalk between the actin cytoskeleton and the ECM through the regulation of focal adhesion dynamics as well as contributing to a fine-tuning of mechanotransduction pathways. Our study brings insights into a novel miRNA-mediated regulatory network that contributes to non-genetic adaptive drug resistance and provides proof-of-principle that preventing MAPKi-induced pro-fibrotic stromal response is a viable therapeutic opportunity for patients on targeted therapy.

Because of its high mutational burden, metastasis propensity, and resistance to 26 treatment, cutaneous melanoma is one of the most aggressive human cancers and the 27 deadliest form of skin cancer (1). Melanoma is a non-epithelial tumor that originates from 28 neural crest-derived and pigment producing melanocytes in the skin. Genetic alterations in 29 the BRAF, NRAS, or NF1 genes define melanoma subtypes and lead to the MAPK pathway 30 hyperactivation (2, 3). Current therapeutic options for BRAF V600E/K metastatic melanoma 31 include MAPK-targeted therapies, which show remarkable efficacy during the first months of 32 treatment (4, 5). However, the majority of patients treated with a combination of BRAF 33 inhibitor (BRAFi) and MEK inhibitor (MEKi) inevitably relapse within months (6). Genetic 34 mechanisms of resistance cannot singly explain the acquisition of therapy resistance in 35 melanoma and non-genetic heterogeneity actively participates in drug tolerance (7,8). 36 Extensive studies have been carried out to dissect the non-mutational mechanisms of 37 resistance (9, 10). Genetic and non-genetic mechanisms of resistance are frequently linked 38 and not mutually exclusive (8). Non-genetic resistance is due to the intrinsic melanoma cell 39 phenotypic plasticity, i.e., ability to undergo transcriptional and epigenetic reprogramming in 40 response to environmental challenges or upon therapy (11). These adaptive mechanisms 41 exploit the developmental plasticity of melanoma cells and often result in an undifferentiated 42 state characterized by upregulation of receptor tyrosine kinases (RTK) like AXL, 43 downregulation of melanocyte differentiation transcription factors MITF and SOX10 (12) 44 and acquisition of mesenchymal and invasive features (9,10,(13)(14)(15)(16)(17)(18). 45 Tumors are shaped dynamically by reciprocal crosstalk between cancer cells and the 46 ECM through cellular-ECM interactions and stromal matrix remodeling. Recent findings 47 indicated that elevated ECM production and remodeling contribute to adaptive and acquired 48 4 resistance to BRAFi therapy by conferring a drug-protective niche to melanoma cells (19)(20)(21)(22). 49 Moreover, we recently reported that undifferentiated mesenchymal-like BRAFi-resistant cells 50 exhibit myofibroblast/cancer associated fibroblast (CAF)-like features leading to pro-fibrotic 51 ECM reprogramming in vitro and in vivo (22,23). Cell autonomous ECM deposition and 52 remodeling abilities adopted by melanoma cells after MAPKi treatment results in cross-53 linked collagen matrix and tumor stiffening fostering a feedforward loop dependent on the 54 mechanotransducers YAP and MRTFA and leading to therapy resistance (22). Thus, this pro-55 fibrotic-like response, typical of the early adaptation and acquired resistance to MAPK 56 inhibition, provides a therapeutic escape route through the activation of alternative survival 57 pathways mediated by cell-matrix communications. However, the signaling networks 58 underlying the acquisition of this undifferentiated, mesenchymal-like melanoma cell state and 59 drug resistant behavior remain unclear. 60 We reasoned that therapeutic approaches aimed at preventing this targeted therapy-61 induced abnormal pro-fibrotic-like response could represent rationale combination strategies 62 to normalize the fibrous stroma and overcome non-genetic resistance in BRAF V600E -mutant 63 melanomas. We show here that the anti-fibrotic drug Nintedanib (BIBF1120) improves the 64 response of the BRAFi/MEKi targeted therapy in a pre-clinical model of melanoma as well as 65 in BRAF-mutated cell lines by preventing MAPKi-induced lineage dedifferentiation, ECM 66 reprogramming and mesenchymal traits. We also identified the master regulator associated 67 with the acquisition of this pro-fibrotic and dedifferentiation program, pointing the miR-143/-68 145 cluster as a driver of the phenotype switching to a drug resistant mesenchymal-like cell 69 state.

72
Nintedanib/BIBF1120 prevents MAPKi-induced pro-fibrotic-like response, enhances 73 targeted therapy efficiency and delays tumor relapse 74 In order to limit ECM reprogramming and collagen remodeling associated with 75 therapy resistance and relapse in melanoma, we tested the effect of the anti-fibrotic drug 76 Nintedanib (BIBF1120), a triple inhibitor of PDGFR, VEGFR and FGFR used to treat  Supplementary Fig. 1B). Thus, combination of targeted therapy 94 with the anti-fibrotic drug Nintedanib prevents the appearance of a pro-fibrotic matrix 95 6 observed upon MAPK-targeted therapy exposure and significantly delays the onset of 96 resistance in vivo. 97 We next examined the impact of Nintedanib on ECM reprogramming and cell 98 phenotype switching in the context of early adaptation and resistance to MAPK targeted 99 therapy in human BRAF V600E mutated melanoma M238P cells. BIBF1120 strongly attenuated 100 targeted drugs-induced ECM/myofibroblast-related signatures, prevented the undifferentiated 101 AXL high MITF low phenotype switch ( Supplementary Fig. 1C) and potentiated the effect of the 102 BRAFi/MEKi cocktail on M238P cell viability ( Supplementary Fig. 1D). The efficacy of the 103 described treatment to reduce upregulation of Fibronectin (FN1) and LOXL2 expression was 104 confirmed at protein levels by Western Blot analysis. Of note, a strong activation of AKT 105 induced by the BRAFi/MEKi cocktail was fully inhibited by BIBF1120, suggesting that the 106 anti-fibrotic drug is able to counteract the rewiring of alternative survival pathway observed 107 upon MAPK oncogenic pathway inhibition ( Supplementary Fig. 1E) (17). 108 We finally evaluated the effect of BIBF1120 on the undifferentiated mesenchymal-  (Fig. 1I), but also significantly decreased cell viability 117 and resistance to BRAFi (Fig. 1J). These findings indicate that an anti-fibrotic therapy is able  Table 1). The predicted targets for each 214 11 of the mature miRNAs were significantly overrepresented among the downregulated genes in 215 response to the corresponding mimics transfection (Fig. 4B). A first set of target candidates 216 were identified by crossing these predicted targets and the genes shown experimentally to be 217 downregulated in resistant M238R cells compared to parental M238P cells (Fig. 4C). 218 Second, RNAs from cells stably overexpressing the miR-143/-145 cluster were 219 analyzed by RNA-sequencing and processed through Ingenuity Pathway Analysis (IPA) to 220 identify the common regulators (transcription factors, growth factors, cytokines, 221 transmembrane receptors, kinases, and phosphatases) between parental cells overexpressing 222 the cluster and resistant cells (Fig. 4D). These analyses notably highlighted changes related to 223 decreased cell proliferation, increased cell invasion and fibrotic pathways activation. To 224 narrow the best target candidates, we finally compared the best-predicted targets based on the 225 two different gain-of-function approaches (Supplementary Table 2  exposure is due to increased expression of miR-143-3p and miR-145-5p. 256 To evaluate the influence of FSCN1 downregulation among the various cellular 257 effects mediated by miR-143-3p and miR-145-5p, we then performed a loss-of-function 258 experiment using FSCN1 specific siRNAs in BRAF-mutant parental melanoma cells. mechanosignaling as well as drug resistance (22,23). Importantly, we provided evidence that 332 the triplet combination BRAFi/MEKi/Nintedanib is active to normalize the fibrous collagen 333 network, delay the onset of resistance and improve mice survival. We also confirmed the 334 efficacy of this therapeutic combination in human BRAF V600E mutant melanoma cells and 335 described its potential to impair phenotype switching and improve response to MAPK 336 targeted therapy (Fig. 8). demonstrates that Nintedanib has anti-fibrotic but also anti-inflammatory and anti-angiogenic 347 activity, the exact contribution of inhibition of specific kinases to the activity of the drug in 348 IPF has not been established and its precise anti-fibrotic mechanism(s) of action is not 349 known. 350 In melanoma, the effects of Nintedanib are likely achieved through the normalization 351 of the fibrotic and drug-protective ECM generated upon MAPK-targeted therapy exposure. 352 We found that combined administration of Nintedanib and MAPK-targeted therapy dampens inflammatory pathways, and actin-SRF regulatory network that need to be fully investigated 427 in this context. 428 We conclude that our work opens new therapeutic avenues to prevent or delay the 429 onset of targeted therapy resistance in melanoma. Our findings provide a rationale for

Luciferase assay:
Molecular constructs for luciferase assay were made in psiCHECK-2 vectors from Promega by cloning upstream of the Renilla luciferase gene annealed oligonucleotides based on the 3'UTR of target genes. HEK239 cells were plated on 96-well plates and co-transfected with 0.2 μg of psiCHECK-2 plasmid constructs and 10 nM of pre-miRNAs (miR-143-3p, miR-145-5p) or control pre-miRNA. Transfections were performed using Lipofectamine 3000, following the manufacturer's instructions. Firefly and Renilla luciferase activities were measured using the Dual-Glo Luciferase assay kit by Promega 48 hours after transfection.

Conditioned medium preparation:
Medium conditioned by melanoma cells was harvested, centrifuged for 5 min at 2,500g and filtered with 0.22 μM filters to eliminate cell debris.

Tumors and cells RNA extraction:
Total RNA was extracted from tumors and cell samples with the miRNeasy Mini Kit (Qiagen) according to the manufacturer's instructions.

Real-time quantitative PCR:
Gene expression: Protocol using the Step One (Applied Biosystem): 1 μg of extracted RNA was reverse transcribed into cDNA using the Multiscribe reverse transcriptase kit provided by Applied Biosystems. Primers were designed using PrimerBank or adopted from published studies.
Gene expression levels were measured using Platinum SYBR Green qPCR Supermix (Fisher Scientific) and Step One thermocycler. Results from qPCR were normalized using the reference gene RPL32 and relative gene expression was quantified with the ΔΔCt method.
Heatmaps describing gene expression fold changes were prepared using MeV software. were purchased by Qiagen.
Information on primer sequences used in this study is provided in table S4 and S5.

Immunoblot analysis and antibodies:
Whole-cell lysates were prepared using RIPA buffer supplemented with protease and phosphatase inhibitor (Pierce, Fisher Scientific), briefly sonicated and centrifuged for 20 min,

Fibrillar collagen imaging:
Collagen in paraffin-embedded tumors was stained with picrosirius red using standard protocols. Tumor sections were analyzed by polarized light microscopy as described [26].
Images were acquired under polarized illumination using a light transmission microscope (Zeiss PALM, at 10x magnification). Fiber thickness was analyzed by the change in polarization color. Birefringence hue and amount were quantified as a percent of total tissue area using ImageJ software.

Viability assay:
After the indicated treatments, cells were stained with 0.04% crystal violet, 20% ethanol in PBS for 30 min. Following accurate washing of the plate, representative photographs were taken. The crystal violet dye was solubilized by 10% acetic acid in PBS and measured by absorbance at 595 nm.

Proliferation assay:
For real-time analysis of cell proliferation, 3x10 4 cells were plated in complete medium in triplicates on 12-well plates. The Incucyte ZOOM imaging system (Essen Bioscience) was used. Phase-contrast pictures were taken every hour. Proliferation curves were generated using the IncuCyte cell proliferation assay software based on cell confluence.

Cell cycle analysis:
Cell cycle analysis was performed by flow cytometry analysis of cells stained with propidium iodide. After fixation in ice-cold 70% ethanol, cells were stained with 40 μg/mL propidium iodide in PBS with 100 μg/mL RNAse A. The samples were then analyzed on a BD FACSCanto cytometer.

Migration and invasion assays:
Migration properties of melanoma cells were tested using Boyden chambers containing polycarbonate membranes (8 μm pores transwell from Corning). After overnight starvation, 1x10 4 cells were seeded on the upper side of the chambers placed on 24 well plates containing 10% FBS medium for 24 h, unless otherwise stated, at 37°C in 5% CO2. At the end of the experiment, cells migrated on the lower side of the chambers were fixed in 4% paraformaldehyde, stained for 15 min with Hoechst and imaged at the microscope (5 random fields per well). Nuclei counting was performed using the ImageJ software. To assess invasion properties of melanoma cells, transwells were coated with Matrigel (1 mg/mL) and cell solution was added on the top of the matrigel coating to simulate invasion through the extracellular matrix.

Immunofluorescence analysis:
Cell area was measured on cells stained for F-Actin using ImageJ. The nuclear/cytosolic ratio of YAP or MRTF was quantified by measuring the nuclear and cytosolic fluorescence intensity using ImageJ. The Hoechst staining was used to define nuclear versus cytosolic regions. Focal adhesions were quantified using ImageJ. Pictures were subjected to background subtraction (rolling: 10) before analysis, then "default threshold" was applied, followed by "analyze particles of object with a size 0.20 and infinity" to analyze the number of objects and their area. The number of focal adhesions was normalized to the total cell area.

Microarray gene expression analysis:
Total RNA integrity was tested with the Agilent BioAnalyser 2100 (Agilent Technologies). After labeling RNA samples with the Cy3 dye using the low RNA input QuickAmp kit (Agilent) following the manufacturer's instruction, labeled cRNA probes were hybridized on 8x60K high-density SurePrint G3 gene expression human Agilent microarrays.

RNA-sequencing:
Short reads: Libraries were generated from 500ng of total RNAs using Truseq corrected reads of all samples sharing same unique splice junctions, by selecting for each group a representative isoform with confident TSS/TES and supported by more than 3 reads.
Selected isoforms were quantified using minimap2 in each sample. Differential isoform expression and alternative splicing events significance were tested without replicates using ad-hoc scripts provided on the Brook's lab Github (https://github.com/BrooksLabUCSC/FLAIR).
Statistical analysis and Biological Theme Analysis: Microarray data analyses were performed using R (http://www.r-project.org/). Quality control of expression arrays was performed using the Bioconductor package arrayQualityMetrics and custom R scripts.
Additional analyses of expression arrays were performed using the Bioconductor package limma. Briefly, data were normalized using the quantile method. No background subtraction was performed. Replicated probes were averaged after normalization and control probes removed. Statistical significance was assessed using the limma moderated t-statistic Quality control of RNA-seq count data was assessed using in-house R scripts. Normalization and statistical analysis were performed using Bioconductor package DESeq2. All P-values were adjusted for multiple testing using the Benjamini-Hochberg procedure, which controls the false discovery rate (FDR). Differentially expressed genes were selected based on an adjusted p-value below 0.05. Enrichment in biological themes (Molecular function, Upstream regulators and canonical pathways) were performed using Ingenuity Pathway Analysis software (http://www.ingenuity.com/).

miRNA targets analysis:
MiRonTop is an online java web tool (http://www.genomique.info/) [31] integrating whole transcriptome expression data to investigate if specific miRNAs are involved in a specific biological system. MiRonTop classifies transcripts into two categories ('Upregulated' and 'Downregulated'), based on thresholds for expression level, differential expression, and statistical significance. It then analyzes the number of predicted targets for each miRNA, according to the prediction software selected (Targetscan, exact seed search, TarBase).