YAP1 controls the N-cadherin-mediated tumor-stroma interaction in melanoma progression

Epithelial-to-mesenchymal transition (EMT) is crucial for melanoma cells to escape keratinocyte control, invade underlying dermal tissues, and metastasize to distant organs. The hallmark of EMT is the switch from epithelial cadherin (E-cadherin) to neural cadherin (N-cadherin), allowing melanoma cells to form a homotypic N-cadherin-mediated adhesion with stromal fibroblasts. However, how “cadherin switching” is initiated, maintained, and regulated in melanoma remains unknown. Here, we show that upon Yes-associated protein 1 (YAP1) ablation in cancer-associated fibroblasts (CAFs), the progression of a BRAF-mutant mouse melanoma was significantly suppressed in vivo, and overexpressing YAP1 in CAFs accelerated melanoma growth. CAFs require the YAP1 function to proliferate, migrate, remodel the cytoskeletal machinery and matrix, and promote cancer cell invasion. By RNA-Seq, N-cadherin was identified as a major downstream effector of YAP1 signaling in CAFs. YAP1 silencing led to N-cadherin downregulation in CAFs, which subsequently induced the downregulation of N-cadherin in neighboring melanoma cells. N-cadherin downregulation inhibited the PI3K-AKT signaling pathway in melanoma cells and suppressed melanoma growth in vivo, supporting the role of N-cadherin as an adhesive and signaling molecule in melanoma cells. This finding suggests that YAP1 depletion in CAFs induces the downregulation of p-AKT signaling in melanoma cells through the N-cadherin-mediated interaction between melanoma cells and CAFs. Importantly, our data underscore that CAFs can regulate N-cadherin-mediated interactions with melanoma cells. Thus, disentangling cadherin-mediated cell–cell interactions can potentially disrupt tumor-stroma interactions and reverse the tumor cell invasive phenotype.


Introduction
Despite the advancements and renewed hopes inspired by novel targeted and immune therapies, there is still a long way to go to nd a cure for advanced melanoma. A hallmark of metastatic tumors is their ability to escape primary sites, invade surrounding tissues, and migrate to distant sites [1,2]. The malignant phenotype is achieved by cancer cells at the invasive front to break the surrounding extracellular matrix (ECM) barrier through reciprocal tumor-stroma interactions [3]. During this process, melanoma cells develop a dynamic and mutually supportive relationship with neighboring cells and the matrix in the host microenvironment, which consists of a heterogeneous mix of noncancer cells, including endothelial cells, immune cells, and broblasts. Genetically stable broblasts that surround and in ltrate melanoma stroma, often termed CAFs, remodel the ECM and secrete chemical factors, which together constitute the backbone of a protective "paradise" that allows melanoma cells to grow, migrate and resist therapeutic agents [4]. Conceivably, targeting CAFs could potentially abrogate microenvironmentmediated resistance and improve treatment outcomes in metastatic melanoma.
YAP1 is an oncoprotein and a downstream transcriptional coactivator in the Hippo signaling pathway [5]. YAP1 regulation by the highly conserved Hippo signaling pathway involves a cascade of protein serine kinases, including mammalian sterile twenty-like kinases 1/2 (MST1/2, Hippo homologs; also known as STK3/K4) and large tumor suppressor kinases 1/2 (LATS1/2, Warts homologs) [6]. When Hippo signaling is active, YAP1 is phosphorylated by LATS1/2 for cytoplasmic retention, ubiquitination, and degradation. Thermo Fisher Scienti c (Rochester, NY) unless otherwise stated. The isolation and maintenance of primary human broblasts were approved by the Institutional Review Board and the Institutional Biosafety O ce of the University of Cincinnati. Experimental procedures involving biosafety issues were carried out under the University of Cincinnati Institutional Biosafety Committee protocol 16-08-17-01.

Mice
The α-SMA-CreER T2 mouse strain was generated and authorized by Dr. Pierre Chambon at The Institute of Genetics and Molecular and Cellular Biology (IGBMC) in France and kindly provided by Dr. Richard T.

Mouse melanoma induction
To generate melanomas carrying YAP1-de cient CAFs in mice, as shown in Fig. 1A, broblasts of genotype α-SMA-CreER T2 ; Yap1 loxP/loxP and control α-SMA-CreER T2 ; Yap1 were isolated from 2-or 3-dayold neonatal littermates without induction and validated by immunostaining for the expression of α-SMA, vimentin, S100A4, keratin 14, and TRP1. Subsequently, a mix of 2 x 10 5 D4M melanoma cells and uninduced α-SMA-CreER T2 ; Yap1 loxP/loxP or control α-SMA-CreER T2 ; Yap1 broblasts at a ratio of 1:1 was injected intradermally into the anks of mice possessing the same genotype as injected broblasts. Mice were monitored daily for tumor formation. To induce the activation of Cre recombinase and Yap1 knockout in α-SMA-CreER T2 ; Yap1 loxP/loxP broblasts, all mice underwent intraperitoneal injection of 10 mg/ml tamoxifen (Sigma-Aldrich, St. Louis, MO) in corn oil at 1 mg/g body weight for seven consecutive days after tumors reached a volume of approximately 62.5 cubic millimeters (counted as day 1). Meanwhile, the size of the tumors was measured and recorded for three weeks. Mice were euthanized when the tumor size exceeded 20% of body size, and tumors were harvested for various analyses.
To generate melanomas carrying YAP1-overexpressing CAFs in mice, as shown in Fig. 2A, broblasts of genotype α-SMA-CreER T2 ; Rosa-rtTA; tetO-Yap1 and control α-SMA-CreER T2 ; Rosa-rtTA were isolated from 2-or 3-day-old neonatal littermates without induction and validated by immunostaining for the expression of α-SMA, vimentin, S100A4, keratin 14, and TRP1. Subsequently, a mix of 1X 10 5 D4M melanoma cells and 1X 10 5 uninduced α-SMA-CreER T2 ; Rosa-rtTA; tetO-Yap1 or control α-SMA-CreER T2 ; Rosa-rtTA broblasts was injected intradermally into the anks of mice carrying the same genotype as injected broblasts. Mice were monitored daily for tumor formation. To induce the activation of Cre recombinase and YAP1 overexpression in α-SMA-CreER T2 ; Rosa-rtTA; tetO-Yap1 broblasts, all mice were fed a doxycycline (Dox) diet and underwent intraperitoneal injection of 10 mg/ml tamoxifen (Sigma-Aldrich, St. Louis, MO) in corn oil at 1 mg/g body weight for seven consecutive days after tumors reached a volume of approximately 62.5 cubic millimeters (counted as day 1). Meanwhile, the size of the tumors was measured and recorded for three weeks. Mice were euthanized when the tumor size exceeded 20% of body size, and tumors were harvested for various analyses.

Melanoma cell isolation
To isolate melanoma cells from mouse tumors, we labeled D4M melanoma cells with red uorescent protein (RFP) for easy identi cation and isolation. Mouse melanomas carrying YAP1-de cient CAFs were generated as described above. Fresh melanomas were excised from mice on day 21. After removing necrotic and connective tissues, the remaining tissues were minced using a scalpel blade into small pieces and incubated with 5 ml of 2.5 mg/ml collagenase IV at 37°C with stirring for one hour to generate a single-cell suspension. After digestion, the mixture was ltered through a 40 µm cell strainer to remove tissue debris and cell clumps. The ltered cell suspension was centrifuged at 1500 rpm for 10 minutes.
Afterwards, cell pellet was washed with DMEM medium with 10% FBS twice and then suspended in PBS solution with 2% FBS for uorescence-activated cell sorting (FACS) to isolate RFP-tagged D4M cells.
Histology, immuno uorescence staining, and immunohistochemistry Tumor tissues were xed in 10% formalin overnight at 4°C and embedded in para n. Five-micron-thick para n-embedded tumor tissue sections were prepared for hematoxylin and eosin (H&E) staining and immunostaining as described previously [16,17]. For histological analysis, para n sections were stained using a standard H&E staining protocol. Slides were mounted using VectaMount permanent mounting medium (Vector lab, Burlingame, CA) and analyzed using a bright eld microscope.
After incubation with primary antibodies, slides were washed with PBS three times, incubated with the corresponding biotin-conjugated secondary antibodies at room temperature for one hour, and then with either uorochrome-conjugated streptavidin for immuno uorescence or VECTASTAIN Elite ABC Reagents (Vector Laboratories, Burlingame, CA) for immunohistochemistry. Nuclei were counterstained with DAPI (blue) for immuno uorescence or hematoxylin (blue) for immunohistochemistry. Images were taken using a Nikon Eclipse 80i uorescence microscope. The number of positively stained cells in each highpower eld (40X) was counted using the particle analysis function of ImageJ software (NIH). The number of positive cells per square millimeter was calculated by multiplying the number of cells counted in each eld by 4.5.

Collagen staining
The collagen content in melanoma tissues was evaluated using the Picrosirius Red staining kit (American MasterTech, Lodi, CA) according to the manufacturer's instructions. After staining, the slides were washed with 1% acetic acid for one minute, dehydrated in 100% ethanol, cleared using xylene, and mounted using VectaMount permanent mounting medium.

Quanti cation of collagen content
To quantify collagen content in melanoma tissues, a Sirius Red/Fast Green Collagen Staining Kit (Chondrex, Redmond, WA) was used according to the manufacturer's instructions. Brie y, after the slide was incubated with the dye solution at room temperature for 30 minutes, the dye solution was removed, and the slide was rinsed with distilled water until the water became colorless. One milliliter of dye extraction buffer was added to each slide to elute the dye from dyed tissues. A 200 µl dye extraction solution from each slide was collected for absorbance measurements at 540 nm and 605 nm using a microplate reader. The collagen content in each sample was normalized to the total protein content.

Generation of inducible YAP1-de cient CAFs
To ablate YAP1 expression in CAFs, M27 and M50 were transduced with doxycycline-dependent inducible lentivirus expressing shRNAs that speci cally target YAP1 expression (V3SH7669-225152043, V3SH7669-225222498, V3SH7669-226435710 from GE Dharmacon, Lafayette, CO). A nontargeting shRNA lentivirus was used to generate control M27 and M50. The inducible lentiviral shRNA vector, which uses the Tet-On inducible system, only allows the expression of either YAP1-targeting shRNA or nontargeting shRNA when cells are treated with doxycycline. The expression of GFP is driven by the same tetO promoter so that the transduction and shRNA expression upon doxycycline treatment can be visually tracked by green uorescence. Brie y, CAFs were seeded in 6-well tissue culture plates. When the cells reached 50% con uence, lentiviral particles were added and incubated with the cells for 16 hours. The medium containing viral particles was then replaced with fresh DMEM supplemented with 0.5% FBS. To select transduced broblasts, medium containing 10 µg/ml puromycin was added and maintained for three days. To assess whether puromycin selection was complete, 500 ng/ml doxycycline (Fisher Scienti c, Pittsburgh, PA) was added for 72 hours to induce the expression of shRNAs and GFP. The e ciency of inhibiting YAP1 expression by shRNA was determined by Western blotting and qPCR.

N-cadherin siRNA silencing
Silencer siRNAs targeting N-cadherin (siRNA ID: s2771) and scramble Silencer® siRNA control were purchased from Thermo Fisher Scienti c (Rochester, NY). Melanoma cells and CAFs were seeded in 6 cm dishes at an initial cell density of 1 x 10 5 cells, cultured for 24 hours until 60-70% con uence, and then transfected with siRNA using Lipofectamine RNAiMAX (Thermo Fisher Scienti c, Rochester, NY). According to the manufacturer's protocol, N-cadherin siRNA (10 µM) was diluted in 250 µl of Opti-MEM I reduced serum medium (Thermo Fisher Scienti c, Rochester, NY) and mixed with 15 µl of Lipofectamine RNAiMAX in 250 µl of Opti-MEM I Reduced Serum Medium. After incubation at room temperature for 10 minutes, the mixture was added to the cells and incubated for three days. Afterward, the medium containing siRNA and RNAiMAX was replaced with regular DMEM culture medium.
Lentivirus transduction was carried out as described above. RFP expression and transduction e ciency were con rmed by ow cytometry.
Chamber slide immuno uorescence staining 1.5 x 10 4 cells were counted and seeded in one well of an 8-well Nunc™ Lab-Tek™ II chamber slide (Thermo Fisher Scienti c, Rochester, NY). The cells were xed in 4% paraformaldehyde for 10 minutes at room temperature and then permeabilized using 0.05% Triton-100 for 10 minutes on ice for immuno uorescence staining. After permeabilization, the cells were washed three times with PBS and blocked in 10% normal goat serum for 1 hour at room temperature. Primary antibodies recognizing Factin (Thermo Fisher, R415, 1:60), paxillin (BD Biosciences, 610051, 1:200), bronectin (Sigma, F3648, 1:200), α-SMA (Thermo Fisher, 14-9760-82, 1:200), S100A4 (Thermo Fisher, 16105-1-AP, 1:200), Ncadherin (Abcam, ab18203, 1:100), MYH10 (Cell Signaling Technology, 3404, 1:100), and MLC2 (Cell Signaling Technology, 3672, 1:100) were then added to each speci c chamber and incubated overnight at 4°C. The next day, after washing with PBS three times, an Alexa Fluor 488-or 594-conjugated secondary antibody (Thermo Fisher Scienti c, Rochester, NY) was added to the corresponding well for a one-hour incubation at room temperature. The slides were mounted with VECTASHIELD Antifade Mounting Medium (Vector Laboratories, Burlingame, CA) and coverslipped. Images were acquired using a Nikon Eclipse 80i uorescence microscope. For EdU staining, a Click-iT™ Plus EdU Cell Proliferation Kit was used for imaging according to the manufacturer's instructions. The number of EdU + cells and total cell number in each high-power eld (40X) were counted using the particle analysis function of ImageJ software. The TUNEL assay was performed using an In Situ Cell Death Detection Kit following the standard protocol provided by the manufacturer (Sigma, St. Louis, MO).

Cell viability and proliferation assays
For cell number counting, 1 x 10 5 cells were seeded in one 6 cm dish and cultured for seven days. The cells were collected on days 1, 3, 5 and 7 for cell number counting using a hemocytometer. At least three repeats were performed for each indicated cell line per assay, and cell number counting was performed a minimum of three times. The MTT assay was performed as we previously published [17].
Generation of melanoma cell-conditioned culture medium To prepare melanoma cell-conditioned culture medium, A375 or SK-MEL-24 melanoma cells were seeded in 10 cm dishes. After the cells reached con uency, the medium was replaced with serum-free medium for another 24-hour culture. Afterwards, the medium was collected and centrifuged at 1000 × g for 10 minutes to remove cell debris for future use.

Transwell migration assay
To perform the transwell migration assay, 3.5 x 10 5 CAFs were seeded on the insert with a permeable membrane (Greiner Bio-One, Kremsmünster, Austria) in a 6-well plate. To assess the response and migratory ability of CAFs to different chemoattractants with and without YAP1 expression, melanoma cell-conditioned medium was added to the bottom well or melanoma cells were seeded in the well, ensuring that CAFs on the membrane were in contact with the medium. After 48 hours, CAFs that migrated through the membrane were stained and imaged. Because CAFs were transduced with GFP, migrated CAFs on the opposite side of the insert were rst captured using a Cytation 1 cell imaging multimode reader and then xed with paraformaldehyde for crystal violet staining. The migration area was de ned as the area occupied by green uorescent CAFs and was quanti ed using ImageJ.
Collagen gel contraction assay A collagen gel contraction assay was performed as reported previously [10,16,17]. After 72 hours of doxycycline induction, 1 x 10 5 CAF cells were resuspended in 500 µl of 1 mg/ml collagen solution and transferred into one well of a 24-well tissue culture plate. After a 30-minute incubation in a humidi ed incubator at 37°C, one ml of fresh medium was added on top of the gel for a 72-hour incubation. Afterwards, gels were detached from the wall of each well and allowed to contract as indicated. Pictures of the gels were taken using a Nikon digital camera every 24 hours, and ZEN 2.3 software was used to measure the diameters. The gel contraction percentage was calculated by dividing the difference in gel diameters between 0 hours and 72 hours by the diameter at 0 hours.

Confocal re ection microscopy (CRM)
CRM was performed as reported previously [10,16,17]. After a 72-hour incubation, collagen ber distribution was assessed using a Zeiss LSM 710 confocal microscope at 40X magni cation. Images were acquired at a depth of at least 100 mm inside the gel to avoid edge effects. Images of at least ten areas of each gel were randomly captured for 3D reconstruction of the matrix using ImageJ. Fiber connectivity was calculated using ImageJ with the BoneJ plugin (http://bonej.org).

3D invasion assay
Collagen gels were prepared as described in the gel contraction assay. 100 µl of gel mixture was placed on the transwell insert in a 24-well plate and allowed to solidify for 30 minutes. Afterwards, 5 x 10 4 CAFs were resuspended in 300 µl of gel mixture and added to the top of the solidi ed gel, and 1 ml of medium was added for incubation overnight. After 24 hours, the medium was removed, and 5 x 10 4 A375 melanoma cells suspended in 100 µl medium were added on top of collagen gel embedded with CAFs. To allow melanoma cells to invade into the underlying gel, gel assemblies carrying CAFs and A375 melanoma cells in 24-well plates were placed in an incubator for up to 15 days. On days 10 and 15, collagen gels were removed and xed with 4% paraformaldehyde overnight at 4°C. Fixed gels were collected for para n section preparation and histological analysis to determine the distance of melanoma cell invasion in the matrices. The collagen content was determined as described above. The distance of melanoma cell invasion was measured using ImageJ.
Quantitative real-time PCR assay RNA was extracted from cultured cells and melanoma tissues using a PurelinkTM RNA Mini kit (Thermo Fisher, Waltham, MA) according to the manufacturer's protocol. For 3D coculture spheroids, RNA was extracted using an RNeasy Plus Micro Kit (Qiagen, Hilden, Germany). Brie y, frozen tumor tissues were added to an RNase-free mortar containing liquid nitrogen and ground thoroughly into powder using an RNase-free pestle before transferring to an RNase-free microcentrifuge tube. Based on the weight of the powder, enough lysis buffer was added for the subsequent homogenization and RNA extraction steps. For cultured cells and spheroids, samples were collected and washed once using ice-cold PBS for RNA extraction. RNA concentration was determined using a NanoDrop spectrophotometer. cDNAs were generated by reverse transcription using a SuperScript IV rst-strand synthesis system (Thermo Fisher, Waltham, MA). qPCRs were performed using SYPR green master mix power on a StepOnePlus real-time PCR system (Applied Biosystems, Waltham, MA). qPCR primers for bronectin, tenascin C, talin, and Ncadherin were purchased from realtimeprimers.com (Philadelphia, PA). The relative expression level of each gene was normalized to the level of GAPDH. The data shown were generated from at least three independent repeats. 3D spheroid cell coculture Cocultured cell spheroids were formed by mixing 5000 RFP-tagged melanoma cells with 5000 CAFs in a total volume of 100 µl in 96-well plates (Thermo Fisher Scienti c, Rochester, NY) with a low cell-adhesion surface. Fluorescent images of spheroids were taken every 24 hours for up to 72 hours using a Cytation 1 cell imaging multimode reader [10]. The RFP area and intensity were recorded each time. To count the RFP + melanoma cell number in the spheroids, 12 spheroids from each group were collected and digested using 2 mg/ml collagenase IV (Thermo Fisher Scienti c, MA) for 30 minutes at 37°C with stirring to generate a single cell suspension. A Countess II Automated Cell Counter (Thermo Fisher Scienti c, Rochester, NY) was used to quantify the RFP + melanoma cell number. The average melanoma cell number in each spheroid was calculated by dividing the total RFP + melanoma cell number by 12. For immuno uorescence staining, spheroids were washed once with cold PBS, xed in 4% paraformaldehyde, and dehydrated using ethanol. After dehydration, spheroids were embedded in 1% agarose gel for para n section preparation and immunostaining.

RNA-Seq
Total RNA was extracted from GFP/M50 and YAP1-GFP/M50 cells after a 3-day doxycycline induction using a PureLink ™ RNA Mini Kit. RNA-Seq was performed as single-end sequencing at the University of Cincinnati Genomics, Epigenomics and Sequencing core using the Illumina NextSeq 550 system. Generated fastq les were validated, processed, and analyzed using the UCSC human hg19 reference genome on the A.I.R. platform (Sequentia Biotech, Barcelona, Spain). Differentially expressed genes were identi ed using the DESeq2 package [18]. Genes with an adjusted p value < 0.05 and a fold change > or < 2 were considered signi cant. Heatmaps of differentially expressed genes were generated using A.I.R. A volcano plot to visualize signi cant gene expression was generated using the OmicStudio tools [19]. Gene Ontology (GO) functional enrichment analysis was performed using A.I.R. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was performed using DAVID Bioinformatics Resources 6.8 [20]. The most affected sets of genes in YAP1-GFP/M50 upon YAP1 ablation were determined using the GSEA software package 4.3.1 [21].

Statistical analysis
All quantitative results were obtained from a minimum of three independent experiments. Data were analyzed using the GraphPad Prism 9 software package (GraphPad Software Inc., San Diego, CA) and expressed as the mean ± SD. The mean difference was determined by Student's t tests and considered statistically signi cant at P < 0.05.

Results
Stromal YAP1 signaling contributes to melanoma progression in vivo To assess the contribution of YAP1 signaling in CAFs to BRAF-mutant melanoma progression in vivo, we generated a transgenic mouse model, α-SMA-CreER T2 ; Yap1 loxP/loxP , which allows the inducible ablation of YAP1 expression in CAFs [22,23]. In the α-SMA-CreER T2 transgene, the expression of sequestered Cre recombinase is driven by a α-SMA promoter [24], which is known to be a major gene expressed in CAFs [25]. Yap1 loxP/loxP harbors homozygous loxP-anked Yap1 alleles, which can be recombined into Yap1null alleles by activated Cre recombinase [23]. To induce melanomas, isolated broblasts of genotype either α-SMA-CreER T2 ; Yap1 loxP/loxP (YAP1 group) or α-SMA-CreER T2 ; Yap1 (control group) were mixed with oncogenic D4M melanoma cells, which carry the Braf V600E activating mutation and are PTEN-de cient, and injected intradermally into the anks of recipient mice of the same genotype as CAFs (Fig. 1A). When the tumors reached an approximate size of 62.5 mm 3 , the mice were administered tamoxifen by intraperitoneal injection to activate sequestered Cre recombinase and knock out the Yap1 gene in CAFs ( Fig. S1A-B).
CAFs are the major producers of collagen and ECM proteins in the tumor stroma. As shown in Fig. 1J-K and M-N, the levels of collagen bers and bronectin were signi cantly reduced in YAP1 KO melanomas. Quanti cation of collagen contents further con rmed the reduction in collagen production in melanomas carrying YAP1-de cient CAFs (Fig. 1L). Consistent with the reduced YAP1 KO tumor size and weight, the number of melanoma cells (a-SMA-negative, α-SMA-) that were Ki67-positive (Ki67+) and the number of Cyclin D1-positive (Cyclin D1+) cells were both decreased in YAP1 KO melanomas (Fig. 1O-T). The data suggest that YAP1 activity is important for CAFs to support melanoma cell proliferation in vivo.

YAP1 overexpression in CAFs accelerates tumor progression
Next, we addressed the impact of upregulating YAP1 signaling in CAFs on melanoma progression. As shown in Fig. 2A, we generated a triple transgenic mouse strain, α-SMA-CreER T2 ; Rosa-rtTA; tetO-Yap1, that inducibly overexpresses YAP1 protein in α-SMA + broblasts and determined whether YAP1overexpressing α-SMA + broblasts could accelerate melanoma progression. In the tetO-Yap1 transgene, the expression of YAP1 is controlled by a doxycycline-inducible tetO promoter [26] and requires the presence of reverse tet transactivator (rtTA) and doxycycline. The expression of rtTA by the Rosa-rtTA transgene is prevented by a loxP-anked STOP cassette [27]. Thus, the generation of rtTA can only occur when active Cre recombinase is present to remove the STOP cassette, which is provided by the α-SMA-CreER T2 transgene when tamoxifen is injected, leading to the overexpression of YAP1 protein in α-SMA + cells (Fig. S1C-D).
Interestingly, we observed that melanomas carrying YAP1-overexpressing broblasts (named YAP1) grew more quickly than control tumors carrying wild-type broblasts (Fig. 2B-C). On day 21, when the tumors were collected, the average size and weight of YAP1 melanomas (1287.99 ± 286.13 mm 3 , 1.30 ± 0.23 g) exceeded those of control tumors carrying wild-type broblasts (829.98 ± 136.48 mm 3 , 0.96 ± 0.28 g) ( Fig. 2D-E). Histological analysis showed that YAP1 melanomas had a more compact internal structure than control tumors (Fig. 2F). The number of α-SMA + CAFs per mm 2 was increased from 314 ± 35 in control melanomas to 450 ± 44 in YAP1 melanomas (Fig. 2G-I). The contents of collagen (Fig. 2J-L) and bronectin ( Fig. 2M-N) were also higher in YAP1 melanomas than in control melanomas. In contrast to the YAP1 KO mouse model, as shown in Fig. 2O-Q, overexpressing YAP1 in CAFs increased the number of Ki67 + melanoma cells per mm 2 to 468 ± 33, while control melanomas had an average of 350 ± 35.
Similarly, the number of Cyclin D1 + melanoma cells was also increased in YAP1 melanomas ( Fig. 2R-T).
The data are consistent with increased tumor size and weight upon YAP1 overexpression in CAFs and further con rm that YAP1 indeed plays a crucial role in CAFs to support melanoma growth.

YAP1 regulates the biological properties of CAFs
To further evaluate the mechanism that controls CAF phenotypes after YAP1 loss, we established inducible YAP1-de cient CAFs using shRNA in different human melanoma-derived CAF lines, including M27 and M50 [17]. Three different YAP1-targeting shRNAs (Fig. S2A-D) were evaluated for their e ciencies in silencing YAP1 expression. YAP1-GFP/Fb-3 shRNA (V3IHSHEG-6373360) was found to have the highest inhibitory e ciency and was selected for subsequent experiments (Fig. S2E), although both siRNAs showed signi cant effects in reducing cell viability (Fig. S2F). CAFs transduced with shRNA (V3IHSHEG-6373360) were named YAP1-GFP/M50 and YAP1-GFP/M27. Nontargeting shRNA-transduced M27 and M50 were used as the controls and named GFP/M50 and GFP/M27, respectively. YAP1 depletion in M27 and M50 was con rmed by Western blotting and qPCR ( Fig. 3A and S2G-H). The loss of YAP1 expression suppressed the proliferation of M50 and M27 cells (Fig. 3B and S3A) but did not cause increased cell apoptosis, which was con rmed by EdU staining and TUNEL assay (Fig. 3C-G and S3B-C).
CAFs exhibit remodeled cytoskeletal structure with increased intracellular tension. As shown in Fig. 3N-P and S3J-O, YAP1 de ciency led to reduced levels of stress ber F-actin, myosin heavy chain MYH10, and myosin light chain MLC2 in CAFs. In addition, the expression of the focal adhesion proteins paxillin and talin, which are involved in cytoskeletal regulation, was suppressed (Fig. 3Q-S and S3P-Q). The data suggest that YAP1 function is involved in the control of cytoskeletal machinery.
CAFs require YAP1 to maintain matrix remodeling ability Migration, matrix contraction, and ECM remodeling require cytoskeletal contraction and remodeling [28,29]. Since YAP1 de ciency led to reduced expression of cytoskeletal proteins, we assessed the contribution of YAP1 to the migratory ability of CAFs using the transwell migration assay. As shown in Gel contraction assays showed that the ability of CAFs to remodel the collagen matrix was reduced after YAP1 depletion (Fig. 4F-J and S3S-V). Next, to visualize and quantify collagen ber alignment and reorganization, collagen ber distribution was assessed using confocal re ection microscopy (CRM). The density of collagen bers in the gel was signi cantly lower in collagen gel embedded with YAP1-de cient CAFs than in collagen gel embedded with wild-type CAFs (Fig. 4K-L). CRM analysis revealed decreased connectivity and increased ber spacing in collagen gel embedded with YAP1-de cient CAFs (Fig. 4M-N).
The CRM data correlated well with the results shown by the collagen gel contraction assays, highlighting the importance of YAP1 for the ability of CAFs to remodel the ECM.
To understand how the ECM remodeled by CAFs in uences the invasion of melanoma cells, we performed a 3D collagen gel invasion assay. GFP/M50 or YAP1-GFP/M50 cells were embedded in collagen gel and seeded onto 24-well transwell inserts. Melanoma cell A375 was added on top of the collagen gel embedded with CAFs and placed in a 24-well plate for culture for up to 15 days. Gel samples were collected on days 10 and 15. Histological staining showed a signi cant reduction in melanoma cell invasion in collagen gel populated with YAP1-de cient CAFs (Fig. 4O-R). On day 15, the distance of melanoma cells invading in gel embedded with CAFs was four times greater than that of YAP1-de cient CAFs (to 1.34 ± 0.28 mm vs. 0.33 ± 0.14 mm in Fig. 4S), suggesting that CAFs require YAP1 functions to make the gel accessible for A375 to invade.

N-cadherin is a YAP1 target in CAFs
To obtain a global picture of the underlying mechanisms by which YAP1 regulates the functional properties of CAFs and identify YAP1-regulated genes that are involved in a CAF-elicited melanoma program, we performed RNA-Seq to compare the gene expression pro les between GFP/M50 and YAP1-GFP/M50. As shown in Fig. 5A, a list of differentially expressed genes was generated by high-stringency and comparative analysis of RNA-Seq data using the Air platform (https://transcriptomics.sequentiabiotech.com/). A total of 1147 genes that were upregulated at least twofold and 1140 genes that were downregulated at least twofold in YAP1-GFP/M50 are shown in a volcano plot (Fig. 5B). We are particularly interested in understanding the genes that were downregulated in YAP1-GFP/M50, as they may be relevant for identifying the YAP1-mediated mechanism by which CAFs interact with melanoma cells. Therefore, KEGG pathway analysis and Gene Ontology (GO) enrichment analysis were performed [30][31][32]. As shown in Fig. 5C, KEGG enrichment analysis of the top 300 downregulated genes revealed that proteoglycans in cancer, regulation of actin cytoskeleton, and focal adhesion are the most affected pathways in CAFs after the loss of YAP1 expression. The most enriched subclasses by GO enrichment analysis of all 1140 downregulated genes were cytoskeletal anchoring at the plasma membrane, stress ber, and cytoskeletal protein binding (Fig. 5D). Gene set enrichment analysis (GSEA) showed that the expression levels of the genes in the KEGG pathways of regulation of actin cytoskeleton and focal adhesion were indeed suppressed in YAP1-GFP/M50 compared to those of GFP/M50 (Fig. 5E-F). Interestingly, among the downregulated genes, CDH2, the gene encoding Ncadherin, appears to be the most signi cantly downregulated gene in CAFs upon YAP1 ablation, as shown in the volcano plot (Fig. 5B), and could potentially function as a key downstream effector of YAP1 signaling in CAFs due to its known role in tumor cell-broblast adhesion.

N-cadherin de ciency in CAFs leads to the downregulation of Ncadherin in melanoma cells
N-cadherin is a transmembrane protein that is known to function in cell-cell adhesion [33]. Importantly, the E-cadherin to N-cadherin switch is known as a major part of the EMT event in tumor progression and metastasis [34]. Western blotting and immunostaining demonstrated the loss of N-cadherin after YAP1 silencing in CAFs (Fig. 6A-C and S4A-C). N-cadherin is expressed in both the mouse melanoma cell line D4M and the human melanoma cell lines A375 and SK-MEL-24 ( Fig. 6D-F). We then checked the expression of N-cadherin in our mouse melanoma model (Fig. 1A), in which YAP1 is ablated in CAFs.
Surprisingly, as shown in Fig. 6H, the number of CAFs was dramatically decreased in the tumor, as shown by α-SMA staining, and the expression of N-cadherin was markedly reduced, especially in melanoma cells (α-SMA-), suggesting that N-cadherin downregulation caused by YAP1 de ciency in CAFs induced Ncadherin downregulation in melanoma cells. To validate this nding, we performed a qPCR analysis of Ncadherin expression using tumor tissues isolated from control and YAP1 KO tumors. As shown in Fig. 6I, the expression of N-cadherin in YAP1 KO tissues was three times lower than that in Ct tissues, indicating that N-cadherin expression was indeed downregulated in melanoma cells upon YAP1 ablation in CAFs.
Next, we cocultured A375 cells with either GFP/M50 or YAP1-de cient YAP1-GFP/M50 cells using an in vitro 3D coculture system and assessed N-cadherin expression. The spheroids formed by A375 and YAP1-GFP/M50 appeared to have more interspaces and lower levels of N-cadherin expression (Fig. 6K, M) than the spheroids formed by A375 and GFP/M50 (Fig. 6J, L), which was con rmed by qPCR analysis (Fig. 6N). The observation of a loose spheroid structure is consistent with what we observed in mouse melanoma carrying YAP1-de cient CAFs (Fig. 1F). Because A375 was tagged with red uorescence, the number of A375 cells in each spheroid was counted and compared using a Countess II Automated Cell Counter. As expected, the spheroids formed by A375 and YAP1-de cient M50 contained a lower number of A375 than the ones formed by A375 and M50 (Fig. 6O). The ndings were con rmed using M27 and the melanoma cell line SK-MEL-24 ( Fig. S4D-I). The reduction in melanoma cells in spheroids also explained a potential reason why the spheroids formed by A375 and YAP1-de cient M50 appeared loose.
Next, to con rm that ablating N-cadherin expression in CAFs does have similar effects as YAP1 depletion and to substantiate our nding that N-cadherin is the major downstream target of YAP1, we used siRNA to silence N-cadherin expression in GFP/M50 cells. After depleting N-cadherin expression, YAP1 was still expressed in CAFs (Fig. 6P, R-S), suggesting that N-cadherin is indeed a downstream effector of YAP1. However, the loss of N-cadherin in M50 suppressed cell proliferation (Fig. 6Q), partially mirroring the effects of YAP1 on CAF proliferation shown in Fig. 3B. We utilized an in vitro 3D coculture system to determine whether N-cadherin de ciency in CAFs can lead to a change in N-cadherin in melanoma cells, similar to YAP1 de ciency. As shown in Figs. 6X and S4M-N, depleting N-cadherin in CAFs led to the downregulation of N-cadherin in the melanoma cell lines A375 and SK-MEL-24. Furthermore, the spheroids formed by A375 and N-cadherin-de cient M50 (Fig. 6U), SK-MEL-24 and N-cadherin-de cient M27 (Fig. S4K), and N-cadherin-de cient A375 and N-cadherin-de cient M50 (Fig. 6V) displayed loose structure compared to M50 + A375 spheroids (Fig. 6T) and M27 + SK-MEL-24 spheroids (Fig. S4J).
Spheroids were digested using collagenase to make single cell suspensions for cell counting. The number of A375 and SK-MEL-24 cells in the spheroids was signi cantly reduced after the loss of Ncadherin expression in CAFs but close to the number of A375 cells in the spheroids formed by N-cadherinde cient A375 and N-cadherin-de cient M50 cells ( Fig. 3Z and S4O), suggesting that YAP1 de ciency and N-cadherin de ciency have similar effects in regulating tumor-stroma interactions.

N-cadherin loss in melanoma cells downregulates p-AKT signaling
It was reported previously that N-cadherin contributes to the proliferation of different cell types via the PI3K/AKT signaling pathway [35]. We examined the proliferation of melanoma cells after N-cadherin ablation and found that N-cadherin depletion in A375 and SK-MEL-24 indeed led to reduced cell proliferation ( Fig. 7A and S4P). As shown in Fig. 7B, S4Q, and S5, p-AKT was signi cantly decreased in Ncadherin-de cient melanoma cells, suggesting that N-cadherin in melanoma cells regulates the activation of the PI3K/AKT signaling pathway. Furthermore, AKT signaling was signi cantly suppressed in D4M melanoma cells when YAP1 expression was silenced in CAFs in the melanoma stroma (Fig. 7C-E). However, an increased number of p-AKT-positive (p-AKT+) cells was found in the melanomas containing YAP1-overexpressing CAFs (Fig. 7F-H). To con rm that AKT signaling was indeed downregulated in melanoma cells upon YAP1 ablation in CAFs, tumor cells were isolated from melanomas containing YAP1-de cient CAFs for AKT Western blotting. As shown in Fig. 7I, both N-cadherin and p-AKT were reduced in melanoma cells, suggesting that AKT signaling in BRAF-mutant melanoma cells was regulated through the N-cadherin-N-cadherin interaction between melanoma cells and CAFs.

Discussion
Cutaneous melanoma arises from the malignant transformation of skin melanocytes and is known for its high propensity to metastasize [36]. Formation of a "reactive" tumor stroma is a wound-like reaction of the stroma to tumor cells, in which dense brosis develops because of reciprocal interactions between tumor cells and local dermal broblasts [37,38]. CAFs are a major cell type in the complex multicell-type and immunosuppressive microcosmos, which creates a supportive tumor niche and constitutes a protective shell that blocks drug access and nurtures drug resistance against targeted therapies [4].
Unsurprisingly, CAFs are associated with poor prognosis in many cancer types [39][40][41]. One interesting and important characteristic of CAFs that has been described by Calvo et al. is the contribution of YAP1 as a master driver of the CAF phenotype [13]. Since Calvo et al. outlined the role of YAP1 in breast cancerderived CAFs, nuclear YAP1 expression has been detected in CAFs of a broad spectrum of cancers, suggesting that its expression is one of their universal features [42][43][44]. It has become clear that for CAFs to have a tumor-supporting function, they require YAP1 [10,45].
YAP1 is the main target of the Hippo pathway, whose role is to inactivate the nuclear localization and transcriptional coactivator activity of YAP1 [46]. However, it is unclear whether impaired Hippo signaling is a main promoter of nuclear YAP1 in CAFs because Calvo et al. did not nd any evidence for decreased Hippo activity in this cell type. We previously reported that YAP1 is a part of the b-catenin signaling axis in CAFs that controls CAF behavior and demonstrated that loss of either of them is detrimental to CAF function [10,16,17]. In this study, two mouse models carrying either YAP1-de cient CAFs or YAPoverexpressing CAFs showed opposite effects on in vivo melanoma growth, re ecting the importance of YAP1 in CAFs and of CAFs for melanoma growth. Furthermore, we con rmed that YAP1 is essential for modulating cytoskeletal changes in CAFs and their ability to remodel the matrix. Coincidently, RNA-Seq revealed that YAP1 modulates CAF functions by controlling major sets of genes that are associated with cytoskeletal functions. The gene that stands out and is most downregulated after YAP1 loss is Ncadherin, a molecule that spearheads CAF-melanoma interactions and invasion after melanoma cells have undergone the switch from epithelial cadherin (E-cadherin) to neural cadherin (N-cadherin) [47]. Ncadherin is known to be expressed on CAFs and associated with the actin cytoskeleton through catenin molecules [48]. We observed that YAP1-de cient CAFs lose the ability to remodel the cytoskeleton and matrix and that N-cadherin is a downstream effector of YAP1 signaling, potentially suggesting that Ncadherin may indeed act as a YAP1-controlled molecular regulator of the actin cytoskeleton in CAFs.
Our results of YAP1 regulating N-cadherin have also been documented in other cell types. For example, suppressing YAP1 expression reduces N-cadherin expression in glioma cells [49]. One component of Hippo signaling, AJUBA, has a similar effect on N-cadherin as YAP1 [50]. However, it is unclear whether AJUBA requires YAP1 to activate N-cadherin. It was only shown that AJUBA can cooperate with the transcription factor Twist to activate the N-cadherin promoter. In addition to AJUBA/Twist, a plethora of other factors, including SET8 [51], FOXD1 [52], ZNF532 [52], CSNK2 [53], MZF1 [54], TRIM28 [55], activated Notch1 signaling [56], GSK3 inhibition [57], Accordingly, reducing N-cadherin expression in human melanoma cell lines leads to decreased invasive capabilities [77]. Different mechanisms that control the E-to N-cadherin switch in tumor cells have been reported [47], including the PI3K/PTEN pathway that transcriptionally regulates the 'cadherin switch' through Twist and Snail [78]. However, it is unclear whether CAFs have any role in regulating this switch and in uencing melanoma progression through the N-cadherin interaction.
Interestingly, we observed that the loss of N-cadherin expression in CAFs reduces N-cadherin in melanoma cells and their proliferative phenotype. N-cadherin, E-cadherin, and other adhesion receptors are continuously turned over at the membrane and contribute to the dynamic assembly and disassembly of adhesive junctions [79][80][81]. Upon internalization in response to different signals, cadherin molecules are either recycled or degraded. However, it remains unclear how the loss of N-cadherin in CAFs leads to the loss of N-cadherin in melanoma cells. One possible reason could be that the direct interaction between melanocytes and broblasts via N-cadherin is closely regulated. When CAFs lose N-cadherin molecules, homotypic N-cadherin interactions can no longer be maintained. Thus, melanoma cells initiate the degradation process to remove unneeded N-cadherin and downregulate N-cadherin expression.
Another explanation could be that YAP1-de cient or N-cadherin-de cient CAFs are unable to produce cytokines or molecules that are necessary to maintain N-cadherin expression in melanoma cells.
Potentially, ablating N-cadherin in CAFs could constitute a negative signal and reverse the E-and N-cadherin switch in melanoma cells since cadherin-mediated cell-cell adhesion needs to be maintained and is central to bridging neighboring cells and the cytoskeleton for sensing and responding to physical and mechanical changes in the stroma.
How changes in N-cadherin expression bring about these profound changes in cellular behavior in melanoma cells is partially understood. We now at least understand that N-cadherin is not only involved in cell-cell adhesion [33].

Conclusion
In conclusion, we present novel insights into the reciprocal interaction between melanoma cells and CAFs and how their interactions promote melanoma cell proliferation and tumor progression. We demonstrate that YAP1 is a key activator of N-cadherin expression in CAFs and that the loss of N-cadherin in CAFs results in a loss of N-cadherin in melanoma cells. However, it remains to be further studied whether the Eto N-cadherin switch and EMT in melanoma cells could be reversed by targeting N-cadherin in CAFs and/or melanoma cells. In addition, the mechanism that controls the recycling, internalization, degradation, and expression of N-cadherin in melanoma cells by N-cadherin-negative CAFs needs to be further elucidated. Nevertheless, further exploration of YAP1 and N-cadherin as therapeutic targets to inhibit melanoma metastasis and improve targeted therapies and immunotherapies is warranted.

Competing interests
The authors declare that they have no competing interests.
Author's contributions Y.Z. designed the experiments; Y.X. and L.Z. performed the experiments and collected data; Y.X., L.Z., T.A., and Y.Z. analyzed the data; Y.X., Y.Z. and T.A. wrote the main manuscript text; Y.X. and Y.Z. prepared gures; Y.X., L.Z., Y.Z., and T.A. reviewed and edited the manuscript. All authors reviewed and approved the nal version to be published. All authors agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved and declare to have con dence in the integrity of the contributions of their coauthors.