Induction of Paracrine Signaling in Metastatic Melanoma Cells by PPARγ Agonist Rosiglitazone Activates Stromal Cells and Enhances Tumor Growth Free

Thiazolidinediones (TZD) are synthetic compounds initially described as drug-improving insulin sensitivity, which thereby improve the control of glycemia (1). Two members of this family of compounds, rosiglitazone (RGZ; Avandia) and pioglitazone (Actos) developed in the 1990s, are currently approved for the treatment of type 2 diabetes, a chronic disease characterized by insulin resistance. The best-characterized molecular target of TZDs is the peroxisome proliferator-activated receptor γ (PPARγ), although all TZD effects are not mediated by PPARγ. PPARγ is a member of the nuclear hormone receptor family, best known for its key roles in regulating adipocyte differentiation, energy metabolism, and insulin sensitivity (2). Even though most of the initial studies on TZDs and PPARγ functions focused on these metabolic roles, evidence has accumulated that they are associated with carcinogenesis. TZDs and PPARγ activation exhibit anticancer effects in a variety of cancer cells and are therefore being considered as potential approaches for chemoprevention and treatment of cancers (3). However, such approaches are currently limited by diverse and sometimes discordant findings. Indeed, despite anticancer actions in many cancer cell types (2) and potential efficiency in chemoprevention (3), TZDs and activation of PPARγ have shown little therapeutic efficacy in clinical trials performed over the past 15 years (3). Furthermore, retrospective studies conducted in type 2 diabetic patients chronically treated with TZDs support increased risk for developing bladder cancer associated with pioglitazone (1). An increased risk of non-Hodgkin lymphoma and melanoma was also suspected (4).

Skin melanoma, a malignant neoplasm of melanocytes, is the most aggressive form of skin cancers, responsible for 80% of skin cancer-related deaths. Melanoma incidence and mortality rates are constantly increasing. In 2012, approximately 232,000 new cases of melanoma were diagnosed worldwide (5). Currently, only early diagnosed, noninvasive melanoma can be cured successfully, when local excision is readily achievable. Upon progression, and once melanoma has become metastatic, it remains a tumor of dismal prognosis in spite of the recent development of immuno- and molecularly targeted therapies (6). Melanoma progression and responses to treatments are driven not only by the malignant cells themselves, but also by impaired reciprocal interactions between the malignant cells and the nonmalignant stromal cells of the tumor microenvironment, in particular fibroblasts, endothelial cells, and immune cells. These paracrine interactions are orchestrated by a variety of factors secreted by the melanoma cells, that activate neighboring fibroblasts and blood vessels, and recruit immune cells, which in turn support tumor progression and metastases (7–10).

The co-occurrence of data suggesting tumorigenic effects with data suggesting anticancer effects of TZDs indicates that chronic treatment with this family of compounds has context-dependent impacts, for which the underlying molecular and cellular bases and the involvement of PPARγ still require clarification. These questions are of importance in regard to the growing interest in PPAR activators as therapeutic options in human cancers (11). The present study explores the basis underlying the debated TZD impact in cancers, using melanoma as a model. Our data reveal that RGZ has tumorigenic effects that were not described previously, and which limit its therapeutic potential in cancer. We show that these tumorigenic actions of RGZ may prevail over its anticancer actions in a subset of melanoma and that, therefore, RGZ administration after tumor initiation may be detrimental.

We compared PPARG expression using the GEO2R tool (12) in two melanoma microarray data sets (GSE3189 and GSE46517; refs. 13, 14) selected for reasonably high sample numbers and unequivocal experimental design in the Gene-Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo, accessed on January 19, 2017).

GSE3189 data set (13): In the original study, the samples with less than 50% of melanocytes or melanoma cells were excluded. We analyzed the samples according to the groups “nevi” (n = 18) and “primary melanoma” (n = 45) as in the original data set. We transformed and presented the data in log2. A Mann–Whitney test was performed on log₂-transformed expression values to assess whether PPARG expression significantly differed between nevi and primary melanomas.

GSE46517 data set (14): Clinical information was available only for the samples obtained from the Medical University of Vienna, Austria. Among these samples, we excluded those with less than 75% melanocytes or melanoma cells. We also excluded normal skin samples, given that melanocytes comprise only a minimal fraction of all skin cells. In the absence of clinical information, we included in our analysis all the samples obtained from the Memorial Sloan Kettering Cancer Center, New York, NY, and from the Brigham and Women's Hospital, Boston, MA. We then grouped the data into “nevi” (n = 7), “primary melanoma” (n = 8), and “melanoma metastases” (n = 57), and we transformed and presented the data in log2. A Kruskal–Wallis test was performed followed by a post hoc analysis using Dunn multiple comparisons test on log₂-transformed expression values.

Human metastatic melanoma cell lines A375 and SkMel28 were purchased from CLS Cell Line Service in 2012 and 2011, respectively; normal human melanocytes (NHM) and metastatic melanoma cells 1205Lu were from ATCC in 2010 and 2011, respectively; vertical growth phase (VGP) melanoma cell line WM115 from Rockland (BioConcept) in 2017; pools of primary normal human dermal fibroblasts (HDF) in 2014 from PromoCell, of human endothelial cells human umbilical vein endothelial cells (HUVEC) from Lonza in 2014. Primary melanocyte 13 cells (M13) were kindly provided by Prof. D. Fisher (Massachusetts General Hospital, Cutaneous Biology Research Center, Boston, MA) in 2016; human radial growth phase (RGP) melanoma cell line WM35 and human VGP melanoma cell line WM793 by Dr. A. Mariotti (CEPO, Epalinges, Switzerland) in 2010; the HDF donors 1 and 2 by Prof. P. Dotto (UNIL, Epalinges, Switzerland) in 2015; human melanoma cancer-associated fibroblasts (CAF) by Prof. Mitchell Levesque (University Hospital of Zürich, Switzerland) in 2018. A375, 1205Lu, and SkMel28 melanoma cells as well as HDF cells were maintained in DMEM-GlutaMax (Invitrogen, Thermo Fisher) supplemented with 10% FBS (Gibco, Thermo Fisher); NHM and M13 cells in melanocyte growth medium (PromoCell); WM cells in Tu 2% FBS medium; HUVEC in endothelial cell growth medium 2% FCS (ECGM, ProVitro); melanoma CAFs in RPMI without L-glutamine (Sigma-Aldrich) supplemented with 10% FBS. All cultures were maintained at 37°C in a 5% CO₂ humidified atmosphere. After collection and first thawing, all cell lines were tested for Mycoplasma contamination with PCR Mycoplasma Detection Kit (Promocell). Cells were not genotyped. Melanoma cells were kept in culture for a maximum of 10 passages, and primary cells (HUVEC, CAF, and melanocytes) up to 6 passages. The characteristics of melanoma cells are provided in Supplementary Table S1.

Melanoma cDNA array containing sets of 43 melanoma tumors surgically resected from patients, covering four disease stages (21 stage III, 19 stage IV) was purchased from OriGene (#MERT501). cDNA extracted from primary melanocytes from four different donors were used as controls.

Monocytes and T cells were isolated from healthy donor human peripheral blood mononuclear cells (HPBMC) with CD14, CD4, or CD8-conjugated microbeads (Miltenyi Biotec). Buffy coats were purchased from the local Blood Transfusion Center, Lausanne, Switzerland, and all subjects gave informed consent. HPBMCs were purified using Leucosep tubes according to the manufacturer's instructions (Greiner Bio-One). In brief, 15 mL of Ficoll-Plaque was preloaded in a 50-mL Leucosep tube by centrifugation for 1 minute at 1,000 × g at room temperature. The whole-blood samples were diluted with equal volumes of PBS, and 30 mL of the diluted blood was added to a Leucosep tube. The cell separation tubes were centrifuged for 10 minutes at 1000 × g without braking at room temperature. The cell suspension was collected and the cells were washed three times in PBS (for 10 minutes at 300 × g and 10 minutes at 200 × g for the two last washes) and resuspended in PBS before counting. Once counted, CD14⁺ monocytes were isolated from PBMCs by magnetic labeling using mAb CD14 conjugated microbeads (Miltenyi Biotec) followed by separation using magnetic columns according to the manufacturer's instructions. The purity of isolated monocytes was >97%, as determined by CD14 staining. CD4⁺ and CD8⁺ T cells were isolated from the CD14-negative fraction by magnetic cell sorting using CD4 and CD8 microbeads (Miltenyi Biotec) as described by the manufacturer. The purity of isolated CD4⁺ and CD8⁺ T cells was >96% and >90%, respectively.

Human CD14⁺ monocytes or T cells were seeded in 48 wells at a cell density of 1 × 10⁶ cells/mL in RPMI 1640 medium (Thermo Fisher) with 2 mmol/L L-glutamine (Thermo Fisher) supplemented with 10% human AB serum (Biowest), 100 U/mL penicillin (BioConcept), 100 μg/mL streptomycin (BioConcept), 1 mmol/L sodium pyruvate (Thermo Fisher), 0.1 mol/L nonessential amino acids (Thermo Fisher) and 0.05 mmol/L 2-mercaptoethanol (Thermo Fisher). Monocytes were left unstimulated, whereas T-cell proliferation was stimulated with anti-CD3 and anti-CD28–coated beads (Miltenyi Biotec).

Melanoma-derived CAFs were isolated according to the methods previously published in (15). Briefly, surplus melanoma biopsies were obtained from consenting patients at the University Hospital of Zurich using a protocol approved by the Institutional Review Board (Ek.647/800), in accordance with the Human Research Laws of Switzerland and the Declaration of Helsinki. CAFs were removed according to the selective adherence protocols (15) and frozen at early passage cultures once the absence of melanoma cells was confirmed by morphologic analysis and Sanger sequencing of the known oncogene mutations.

A375 cells were treated for 24 hours with 5 μmol/L rosiglitazone maleate (RGZ, Avandia, Lucerna-Chem), or 2 μmol/L T0070907 (T007, Enzo Life Sciences), or a combination of both. All treatments were diluted in DMSO as vehicle (minimum dilution 1:10,000).

Conditioned media (CM) preparation: Cells were treated for 24 hours with 5 μmol/L RGZ; then washed twice in PBS and cultured in fresh medium (7 mL: 1.10⁶ plated cells; FBS-free; RGZ-free) for 72 hours. The medium conditioned for 72 hours was collected and filtered with a 0.45-μm filter and was either used immediately or frozen at −80°C for future experiments. Cells were harvested for RNA extraction.

Transwell assay (TW): A375 cells were plated into 6-well companion plates (BD Falcon Corning, Vitaris) and fibroblasts into culture inserts (BD Falcon, High Density, Translucent PET Membrane 3-μm pore). RGZ (5 μmol/L) was added into the wells for 72 hours.

Human normal fibroblasts, melanoma-derived CAFs, endothelial cells, and immune cells were stimulated with CM mixed with corresponding complete medium in 1:1 (v:v) mix.

For gene-expression analyses, normal fibroblasts, melanoma-derived CAFs, and endothelial cells were stimulated for 72 hours and then harvested for RNA extraction.

For secretome analyses, fibroblasts were stimulated for 24 hours, and then washed twice in PBS and cultured in fresh medium 5% FBS. The medium was collected 48 hours later, filtered with a 0.45-μm filter, and was either used immediately or frozen at −80°C for future experiments.

For differentiation analyses, the unstimulated CD14⁺ monocytes, the stimulated CD4⁺ T cells, and the stimulated CD8⁺ T cells were cultured with CM for 5 days or left untreated. As control, M1 and M2 lineages were activated with the phenotype M1 (10 ng/mL LPS + 50 ng/mL IFNγ) or M2 (20 ng/mL IL4 + 20 ng/mL IL13) stimuli. To check for differentiation phenotype, we measured the expression of several markers by flow cytometry.

HDFs were pretreated with complete medium containing either unrelated mock antibodies as control (anti-BAFF antibody belimumab, registered trade name Benlysta, 50 ng/mL + anti-RANKL antibody denosumab, registered trade name Xgeva, 50 ng/mL), or with a combination of recombinant human IL1R antagonist (registered trade name Anakinra) 1 μg/mL and anti-IL6R antibody tocilizumab (registered trade name Actemra) 100 ng/mL for 1 hour. The pretreatment medium was then replaced with RGZ-treated A375 melanoma cell CM mixed with corresponding complete medium in 1:1 (v:v) mix and supplemented with the same amount of either mock antibodies or with the same combination of recombinant human IL1R antagonist Anakinra and anti-IL6R antibody tocilizumab as pretreated. After 72 hours of stimulation, HDFs were harvested for RNA extraction.

Cell and tissue RNA were extracted using TRIzol (Life Technologies, Thermo Fisher) or peqGOLD TriFast (Peqlab, Axonlab AG) according to the manufacturer's instructions. RNA quality was checked using Fragment Analyzer (Advanced Analytical). RNAs (500 ng–2 μg) were reverse-transcribed with an iScript cDNA Synthesis Kit (Bio-Rad) according to the manufacturer's instructions. cDNA quantifications were performed with GoTaq qPCR Master Mix (Promega). No template control and no reverse transcriptase enzyme sample were used as negative controls. Gene expression was normalized to housekeeping gene expression (USP16 for human genes, Rpl27 for mouse genes). In all experiments, control treatments were set at one. Geo-mean fold change for each treatment is reported. Primer sequences are accessible in the Supplementary Material.

A375 melanoma cells were treated for 24 hours with 5 μmol/L RGZ, 2 μmol/L T0070907, or a combination of both; three independent experiments were performed, and in each of these three experiments, RNA samples from technical replicates of the same condition were pooled. RNA-seq experiment was conducted at the Lausanne Genomic Technologies Facility (Lausanne, Switzerland) according to an in-house pipeline. Briefly, RNA samples were extracted from treated A375; RNA quality was assessed by spectrophotometry (NanoDrop 8000, Fisher Scientific) and electrophoresis (Bioanalyzer 2100, Agilent Technologies). cDNA libraries were prepared and sequenced on a HiSeq 2500 machine (Illumina). Quality controls were applied for cleaning data for adapters and trimming of low-quality sequence ends. Cleaned data were aligned and read counts computed using Tophat (version 2.0.9) and HTSeq (version 0.5.4p3), which generate gene counts. Additional quality controls were done using R (version 3.0.2) for inspecting the sample counts summary, pairwise sample correlations, relative log expression, clustering, and sample principal component analysis. Lowly expressed genes (fewer than 50 read counts summed across all samples) were removed before data processing. Statistical analysis was performed using the R packages edgeR 3.4.2 (16) for normalization and limma 3.26.9 (17, 18) for differential expression. Linear model fits were extracted for contrasts, which were tested for significance using a moderated t test; P values were adjusted with the Benjamini–Hochberg method (19). Enrichment analysis was performed using MetaCore (version 6.19 build 65960; GeneGo) and WebGestalt (20). Primary data accession number is GSE115221.

Cells were lysed in RIPA buffer supplemented with 1% protease and phosphatase inhibitor cocktail (Cell Signaling Technology CST Bioconcept) on ice for 15 minutes, then spinned 10,000 × g for 10 minutes. Supernatants were collected and proteins were quantified using DC protein assay (Bio-Rad). Proteins (15 μg) were loaded on a 10% SDS-PAGE. After migration with a Mini-PROTEAN system (Bio-Rad) and transfer with PerfectBlue semidry Electro Blotter (Peqlab) in Towbin buffer, membranes were processed as follows: BSA 5% for 1 hour at room temperature; anti-PPARγ antibody (CST, #2435, 1:1000 in BSA 5%) O/N at 4°C; T-TBS washes; anti-rabbit IgG-HRP (Promega, #W4011, 1:30,000 in BSA 1%) 1 hour at room temperature; T-TBS washes. Signals were revealed using WesternBright Quantum HRP substrate (Advansta, Witec) and Fusion FX imager (Witec). GAPDH was used as loading control following membrane stripping as follows: Restore Western Blot Stripping Buffer (Thermo Fisher) 5 minutes at room temperature; milk 5%; anti-GAPDH (CST, #2118, 1:5,000 in milk 1%) 1 hour at room temperature; anti-rabbit IgG-HRP (1:30,000 in milk 1%) 1 hour at room temperature.

Forty milliliters of phenol-free conditioned medium from A375 cells was 0.2 μm filtered and ultra-centrifuged 10,000 × g for 30 minutes at 4°C. Thirty milliliters of media was then concentrated with Amicon Ultra-15, 3kD (Merck) in 3 rounds: 10 mL, 3,214 × g for 45 minutes at 4°C. At last, two washes were done with 10 mL of freshly prepared 100 mmol/L ammonium bicarbonate (Sigma-Aldrich). Concentrated media (400–500 μL) were analyzed by the Proteomic Analysis Facility of University of Lausanne as described in the Supplementary Material. Enrichment analyses of secreted proteins were performed using the online software Reactome (version 63). Primary data accession number is PXD009108.

Secreted proteins from A375-treated media, WM793-treated media, and from conditioned medium-stimulated HDF media were analyzed by an ELISA kit according to the manufacturer's instructions: ANGPTL4 (Human Angiopoietin-like 4 DuoSet ELISA, R&D Systems), IL1β, IL6, IL8, and GM-CSF (Human IL1B, IL6, IL8 or GM-CSF ELISA Ready-SET-Go!, Affymetrix eBioscience), and CXCL-3 (Human CXCL3 ELISA). Optical density was analyzed with a BioTek PowerWave microplate reader or an Epoch2 microplate spectrophotometer (BioTek).

Iced 96-well plates were coated with 40 μL of growth factor–reduced Matrigel (Corning, VWR) and incubated 30 minutes at 37°C. HUVECs (8000) were plated onto Matrigel in complete ECGM (80% confluency). One hundred milliliters of RGZ-supplemented medium or of conditioned medium was added on top, in triplicate conditions. Tubes were allowed to form for 24 hours at 37°C, and then living cells were stained with 5 nmol/L calcein (eBioscience) 30 minutes at 37°C. A total of nine representative pictures (three per well) were taken with a motorized AxioVert microscope (magnification, ×50; Carl Zeiss). Pictures were analyzed using ImageJ as described in (21). Vessel length and number of junctions are reported.

Spheroids containing 500 HUVECs were drop-formed in spheroid medium (0.25% methycellulose in ECGM) for 24 hours at 37°C. Spheroids were collected, washed, and suspended in a 1:1 mix (conditioned medium: collagen solution). Immediately after suspension, spheroids were transferred into prewarmed 24-well plates. Each treatment was performed in triplicate. Twenty-four hours after incubation at 37°C, living spheroid cells were stained with 5 μmol/L calcein 30 minutes at 37°C. Fifteen spheroids per condition were analyzed with a motorized AxioVert microscope (magnification, ×50). Invasion and spheroid areas were measured with ImageJ distribution Fiji software, and the percentage of invasion is reported. VEGF 25 ng/mL was used as a positive control.

On day 5 of culture of CD14⁺ monocytes, cells were detached from the culture plates by subsequent trypsinization (Gibco, Tryple Express enzyme, phenol red). Cells were then washed with PBS-2 mmol/L EDTA and incubated 10 minutes on ice with FCR blocking reagent (Miltenyi Biotec). Cells were then incubated 30 minutes on ice with the following mAbs: HLA-DR, CD209, CD86, CD206, CD80 (BioLegend), CD14, CD16, CD11b (BD Biosciences), CD163 (R&D Systems), CD64, CD83 (Beckman Coulter). Intracellular CD68 (BioLegend) staining was performed with the Foxp3 Staining Set (eBioscience/Thermo Fisher). On day 5, cultured CD4⁺ and CD8⁺ T cells were collected, washed, and incubated with the following mAbs: CCR7, CD3, CD4, CCR6, CXCR3, CD8, CD127 (BioLegend), CD25, and CD45RA (BD Biosciences). Data were acquired on the LSRFortessa (BD Biosciences) cytometer and analyzed using FlowJo Software (Tree Star).

All experiments involving animals were approved by the Veterinary Office of the Canton Vaud (Switzerland) in accordance with the Federal Swiss Veterinary Office Guidelines and conform to the Commission Directive 86/609/EEC. NOD SCID gamma (NSG) mice were bred and maintained locally in the SFP animal facility of the University of Lausanne, Lemanic animal facility network. Experiments were performed in the conventional animal facility of the University of Lausanne. A375-tdTomato fluorescent cells were injected (1.10⁶ cells) into the left flank of 6-week-old or older mice in compliance with the University of Lausanne Institutional regulations. Six days after injection (palpable tumors), mice were fed either a control (n = 12) or a RGZ-supplemented (n = 12) diet (12 mg/kg/d). Mice behavior and tumor growth were monitored twice a week. One of the criteria for withdrawal from the experiment was tumor volume of 1,000 mm³.

After sacrifice, the tumors were cut in four equal parts: a quarter was fixed in zinc and paraffin-embedded for histochemistry analyses; two quarters were frozen for further gene-expression analyses. The last quarter of tumor and the left lung were red-fluorescent analyzed with the stereomicroscope for fluorescence Leica M205FA (Leica) and the LAS AF6000 software (Leica). Quantification of the metastases was performed with ImageJ distribution Fiji software in a blinded fashion.

Tumors were fixed in zinc solution (BD Biosciences) for 24 to 48 hours, then PBS washed and paraffin-embedded. Four-micrometer sections were rehydrated and processed as follows.

Sections were stained in Mayer's hematoxylin for 5 minutes, washed in H₂O₂, differentiated in alcohol acid for 10 seconds, washed in H₂O₂, stained in eosin alcohol for 45 seconds, and dehydrated in ethanol 70% to 100%. A European board-certified veterinary pathologist performed the histopathologic evaluation in a blinded fashion. The necrotic area per section was measured in μm for each animal using a magnification of ×100.

Sections were blocked in NSG 2.5% for 30 minutes, incubated with anti-Ki67 (Life Technologies, #41-5698-80, 1:60), anti-cleaved caspase-3 (Cell Signaling Technology, #9661S, 1:100), or anti-Meca32 Ab (Novus, #NB100-77668, 1:100) in NSG 2.5% O/N at 4°C; washed in PBS; goat anti-rat Ab (for Ki67 and Meca32: Mol. Probes, #A11077, 1:100; or cleaved caspase-3: goat anti-rabbit Mol.Probes Ab A21069 in NSG 2.5%) 30 minutes at room temperature, washed in PBS, counterstained with DAPI, and embedded in MOWIOL. Pictures were taken with a motorized AxioImager M1 microscope and the AxioVision software (Carl Zeiss), using a magnification of ×200. Quantification of blood vessels, Ki67, and cleaved caspase-3–positive cells was performed with ImageJ distribution Fiji software in a blinded fashion.

To compare two independent groups, we used the Student t test or Mann–Whitney test, depending on homogeneity of variances; to compare one independent variable in more than two groups, we used the one-way ANOVA or the Kruskal–Wallis test, depending on the homogeneity of variances; to compare two independent variables in more than two groups, we used two-way ANOVA with Tukey post hoc comparisons; to compare two paired groups we used paired Student t test. Statistics on the survival curves were performed using a Mantel–Cox log-rank test.

Gene-expression levels in cells, as quantified by RT-qPCR: for each gene, control and treated expression levels were all normalized to the control (hence, control fold change is 1). Log₂-transformed fold changes were compared with 0 (the control Log₂-fold change) using a one-sample t test.

Each in vitro experiment was performed in technical duplicate or triplicate.

All statistical analyses were performed using Prism GraphPad (v7).

As a first step toward assessing the impact that TZDs and PPARγ activation may have in existing melanoma tumors, we quantified PPARγ isoform 1 expression in patient-resected tumors of melanocyte origin and in a variety of melanoma cell lines. Using two publicly available NCBI GEO microarray data sets (GSE3189 and GSE46517, chosen for high sample numbers and unequivocal experimental design; refs. 13, 14; Fig. 1A and B), as well as melanoma of commercially available melanoma cDNA arrays (Fig. 1C), we found that except in primary tumors, the RNA expression level of PPARγ was highly variable. Indeed, we detected a difference of up to 64-fold between the lowest and highest expression of PPARG in melanoma metastases (Fig. 1C). We observed no significant difference in comparing PPARG RNA levels in patient-resected nevi, primary, or metastatic melanoma, suggesting no association between PPARG expression levels and the stages of the disease (Fig. 1A–C). We next found that PPARγ RNA and protein levels were also highly variable in NHM and in a variety of human melanoma cell lines derived from RGP (WM35), VGP (WM115 and WM793), and metastatic tumors (1205Lu, SkMel28, and A375) representative of the disease progression (Fig. 1D). Although we observed a trend toward increased expression of PPARG RNA, we found no clear correlation between PPARγ protein levels and the progression of melanoma cells toward a metastatic phenotype (Fig. 1D).

To probe into the transcriptional impact of RGZ in melanoma, we next performed a global gene-expression analysis in melanoma cells using an RNA-seq approach. We chose the A375 human metastatic melanoma cell line, as a model for human melanoma tumors expressing high levels of PPARγ and exhibiting the typical BRAFV600E mutations found in approximately 50% of patient tumors (6). We focused our study on genes whose expression was regulated by RGZ in a PPARγ-dependent manner, based on the results of combined treatments with RGZ and the PPARγ-specific antagonist T0070907 (T007). With cutoff values for adjusted P value (Benjamini–Hochberg method; ref. 19) and absolute fold change set at 0.05 and 1.5, respectively, we obtained a list of 313 upregulated and 350 downregulated PPARγ-regulated genes (Supplementary Fig. S1A; Supplementary Table S2 for the top 10 upregulated and downregulated genes). Half of the top 20 significantly deregulated pathways were related to interleukin secretion or immune response (Table 1). In particular, three IL1-related pathways of relevance for cancer progression were upregulated, namely, “Immune response–IL1 signaling,” “Immune response–CCL2 signaling” and “Immune response–IL18 signaling” (Table 1). The upregulated expression of IL1A, IL1B, and IL6, three major proinflammatory actors in these pathways, mediated by RGZ in independent experiments was in agreement with the RNA-seq analysis (Supplementary Fig. S1B). Both basal and RGZ-induced expression of these cytokines was downregulated by the PPARγ antagonist T0070907, suggesting PPARγ-dependent regulation (Supplementary Fig. S1B). Furthermore, the RNA levels of IL1A and IL1B were significantly upregulated in A375 cells treated with ciglitazone and pioglitazone, two other members of the TZD family, and with GW1929, a non-TZD PPARγ agonist (Supplementary Fig. S1C), whereas IL6 expression was upregulated by treatment with GW1929 only. Finally, CRISPR-mediated downregulation of PPARγ expression partially prevented the upregulation of IL1A and IL1B, but not that of IL6 (Supplementary Fig. S1D). In summary, RNA-seq complemented by RT-qPCR analyses revealed that RGZ-mediated activation of PPARγ increases the expression of genes encoding secreted proteins in melanoma cells. Although the regulation of IL1A and IL1B expression by RGZ is clearly PPARγ dependent, that of IL6 seems to be more complex and to involve other RGZ targets.

We next sought to complement gene-expression analyses by a comprehensive comparison of the proteins secreted by control- and RGZ-treated A375 melanoma cells. Media were collected from A375 melanoma cells treated with vehicle or RGZ for analysis by tandem mass spectrometry. Efficient PPARγ activation with RGZ was first confirmed by significant upregulation of the expression of the canonical PPARγ target gene FABP4 (Fig. 1E). An initial proteomic analysis of the media obtained in three independent experiments was performed with a cutoff P of 0.05. Out of a total of 1,699 proteins detected, 279 were secreted proteins, among which 19 were significantly over-secreted and 21 were significantly under-secreted upon RGZ treatment (Fig. 1F). A pathway enrichment analysis was conducted using the Reactome database and revealed that the secreted proteins affected by RGZ mainly belonged to “Immune system,” “Innate Immune System,” “Cytokine Signaling in Immune system,” and “Signaling by Interleukins” pathways (Table 2 for the top 20 pathways). The interleukin family members also highlighted by our RNA-seq analysis are small proteins, which are represented by just a few peptides, if not one, in mass spectrometry analyses. They may therefore remain undetectable or underestimated. To improve the detection of such small proteins, we performed a deeper analysis of one replicate pair of samples. With a cutoff value for absolute fold change of 1.5, we obtained a list of 44 upregulated and 159 downregulated secreted proteins exhibiting RGZ-dependent regulation (data set identifier PXD009108). Reinforcing our initial observation and in line with the increased RNA expression of cytokines and chemokines described earlier, pathway enrichment analysis revealed that “immune system,” “cytokine signaling in immune system,” and “signaling by interleukins” were among the four pathways most affected by RGZ (Supplementary Table S3). Though improved with the deeper mass spectrometry analysis, the identification and quantitation of cytokines remained based on very few peptides. We thus selected proteins playing leading roles in the pathways significantly affected by RGZ, and we used ELISA assays in three independent experiments to confirm increased secretion of the PPARγ target ANGPTL4 (Fig. 1G), the inflammatory cytokines IL1β and IL6 (Fig. 1H), the chemokine GM-CSF (Fig. 1I), as well as the angiogenic and neutrophil attractant chemokine IL8 (Fig. 1J). By contrast, IL1α and CXCL3 were increased by RGZ at the RNA level only, with no significant changes in secretion of the proteins (Fig. 1H and I).

We next used a panel of cancer cells to address whether this action of RGZ was PPARγ dependent, and whether it extended beyond the A375 metastatic cell line. To do so, we quantified the expression of canonical PPARγ target genes and of interleukins IL1A, IL1B, and IL6 as flagship markers, respectively. The combined treatment of various melanoma cell lines with RGZ and the PPARγ antagonist T0070907 suggested a PPARγ-dependent action of RGZ only in A375-M2 (a more severe form of the metastatic A375; Supplementary Fig. S2A). The expression of at least one of the three cytokines IL1A, IL1B and IL6 was significantly increased by RGZ in locally invasive melanoma (WM793; Supplementary Fig. S2B, left), in metastatic melanoma (A375-M2; Supplementary Fig. S2C), in metastatic human bladder cancer cells (T24; Supplementary Fig. S2D), and we observed a similar trend in human skin squamous cell carcinoma (SCC13; Supplementary Fig. S2E). In the locally invasive melanoma cell (WM793), this was accompanied by increased IL6 secretion (Supplementary Fig. S2B, right). In contrast, cytokine expression was not affected by RGZ in noninvasive melanoma cells (WM35; Supplementary Fig. S2F), locally invasive melanoma cells (WM115; Supplementary Fig. S2G) or in a human cervix cancer cell line (HeLa; Supplementary Fig. S2H).

Collectively, these data suggest that RGZ activates a cell-to-cell communication program involving the secretion of inflammatory interleukins, chemokines, and angiogenic factors in a subset of human metastatic melanoma cells, which may affect the tumor cells and their microenvironment. Moreover, we show that this action of RGZ is likely not limited to melanoma cells, and that the involvement of PPARγ is context dependent.

Having documented the impact of RGZ treatment on gene expression and protein secretion by A375 metastatic melanoma cells, we next addressed the functional relevance of these changes for the tumor cells and stromal cells of the tumor microenvironment. Secreted factors, in particular those in IL1-related pathways, can promote autocrine activation of proliferation and inhibition of apoptosis in melanoma cell (22). These signals are transmitted through membrane receptors: IL1R1, which is activated by IL1α and IL1β, IL1R2, their decoy receptor, and IL6R, activated by IL6. We detected no modification in the membrane expression of IL1R1, IL1R2, or IL6R in RGZ compared with vehicle-treated A375 cells (Supplementary Fig. S3A), and no impact of RGZ treatment on melanoma cell proliferation (Supplementary Fig. S3B), cell cycle (Supplementary Fig. S3C), apoptosis (Supplementary Fig. S3C, sub-G₁ cell population), or senescence (Supplementary Fig. S3D).

We next tested the hypothesis that the secretome of human melanoma cells treated with RGZ would affect nonmalignant stromal cells of the tumor microenvironment in a paracrine way. In particular, fibroblasts, endothelial and immune cells receiving signals from melanoma cells can acquire a tumor-friendly phenotype and can thereby promote tumor growth (7–9). To address this question, we collected the secretome of A375 melanoma cells exposed to RGZ or vehicle for 24 hours and then cultured for an additional 3 days in fresh serum-free and RGZ-free medium. This medium containing the secretome of A375 melanoma cells exposed to RGZ or vehicle is referred to as “conditioned medium”. We then compared the response of human primary fibroblasts, human melanoma-derived CAFs, endothelial and immune cells to the CM collected from RGZ- or vehicle-treated A375 cells.

Fibroblasts in tumors, called CAFs, express specific markers and are an important source of cytokines and chemokines. When exposed to the conditioned medium collected from RGZ-treated A375 cells, normal HDFs isolated from two individual donors (HDF donor 1 and HDF donor 2) failed to significantly increase the expression of α-smooth muscle actin (ACTA2) or tenascin C (TNC), two of the most widely used markers for CAFs (Fig. 2A and B, left). In contrast, they demonstrated an increase in the expression of the proinflammatory cytokines IL1A, IL1B, and IL6 (Fig. 2A and B, middle). Although the secretion of IL1α and IL1β remained below detection levels by ELISA assay, the increased expression of IL6 was accompanied by an increased secretion of the protein in both HDF donors 1 and 2 (Fig. 2A and B, right). Besides increased expression and secretion of proinflammatory interleukins, we observed an increased expression and secretion of the chemokines GM-CSF and of the proangiogenic chemokine IL8 (CXCL8; Fig. 2A and B). Reinforcing these data, a commercial pool of normal HDFs responded similarly with a strong increase in the expression of the same inflammatory markers when exposed to RGZ-treated melanoma cell conditioned medium (Fig. 2C), as well as when receiving signals from RGZ-treated melanoma cells in a transwell coculture assay (Fig. 2D). Of all secreted proteins that were significantly affected upon RGZ treatment of melanoma cells, the inhibition of IL1 and IL6 with clinically used inhibitors (Anakinra and tocilizumab, respectively) was sufficient to attenuate the expression of IL1A, IL1B, IL6, and CXCL8 in HDFs (Fig. 2E). Finally, although CAFs are considered as already activated cells (10), exposure of CAFs derived from a patient melanoma tumor to the conditioned medium collected from RGZ-treated A375 cells further enhanced the expression of IL1A, IL1B, CSF2, and CXCL8 (Fig. 2F). Importantly, direct exposure to RGZ had no such impact on any of the three normal primary fibroblast cells (Supplementary Fig. S3E–S3G).

Because angiogenesis is key to support melanoma progression and metastatic progression of melanoma, we next performed similar experiments to address the hypothesis that human primary endothelial cells may be activated by the secretome of the PPARγ-activated melanoma cells. Exposure of HUVEC to RGZ-treated melanoma cell conditioned medium induced an increase in the expression of the VEGF receptor KDR, of the cell–cell adhesion protein ICAM1, and of the cytokines IL1B and CCL2, four markers known to be upregulated in endothelial cell in angiogenic and/or inflammatory conditions (Fig. 3A; refs. 23–26). Direct treatment of HUVEC with RGZ did not affect these markers (Supplementary Fig. S3H, left). A Matrigel-based tube formation assay and a collagen-based invasion assay were used to evaluate angiogenic capacities of the endothelial cells. As judged by the total length of branches, as well as by the number of nodes quantified in the Matrigel-based tube formation assay (Fig. 3B and C), RGZ-treated melanoma cell conditioned medium enhanced the angiogenic capacities of HUVEC compared with control-treated cells. In line with this finding, RGZ-treated melanoma cell conditioned medium also enhanced the ability of HUVEC to invade collagen (Fig. 3D and E), whereas direct treatment of HUVEC with RGZ did not affect their angiogenic capacities (Supplementary Fig. S3H, middle and right).

As the function of immune cells can also be affected by signals secreted by cancer cells, we performed similar experiments using HPBMCs isolated from healthy donors. We observed that the exposure of unstimulated CD14⁺ monocytes to the conditioned medium of RGZ-treated melanoma cells induced their differentiation into macrophages, with a clear polarization toward M2, procarcinogenic phenotype (Fig. 4A). Such polarization was not observed in monocytes directly exposed to RGZ (Supplementary Fig. S3I). Moreover, the conditioned medium of RGZ-treated melanoma cells completely prevented the differentiation of monocytes into cells exhibiting antigen-presenting cells (APC) markers (Fig. 4B). In contrast, the phenotype of T cells (Fig. 4C) was not affected by melanoma cell conditioned medium.

Finally, we tested the hypothesis that the impact of melanoma cells on stromal cells was not restricted to the metastatic melanoma cell line A375, and that it was proportional to the melanoma cell response to RGZ. We collected the secretome of two other melanoma cell lines, the RGP WM35 and the VGP WM793 cell lines. WM35 cells were chosen as a model of noninvasive melanoma cells nonresponsive to RGZ as indicated by interleukin expression (Supplementary Fig. S2F). In line with that observation, we observed no significant activation of human endothelial cells (HUVEC) or HDFs (donors 1 and 2; Supplementary Fig. S4A) by the condition medium collected from RGZ-treated WM35. WM793 cells were chosen as a model of malignant, locally invasive melanoma cells in which exposure to RGZ induced a moderate, lower increase in interleukin markers compared with the metastatic A375 cells (Supplementary Fig. S2B and Fig. 1H). Although this trend did not reach statistical significance, the conditioned medium collected from these cells slightly activated HUVEC (as indicated by an increased expression of KDR, ICAM1, IL1B, and CCL2 molecular markers; Supplementary Fig. S4B, left) but not primary fibroblasts (Supplementary Fig. S4B, middle and right).

Altogether, these findings support the hypothesis that RGZ is able to alter the secretome of human melanoma cells in a cell context–dependent manner. The complex blend of paracrine signals produced by the melanoma cells, in which IL1 and IL6 seem to play a major role, can in turn affect the phenotype of nonmalignant fibroblasts, endothelial and immune cells. When exposed to the conditioned medium of melanoma cells responsive to RGZ, fibroblasts developed a proinflammatory phenotype, endothelial cells exhibited increased angiogenic capacities, and monocytes tended to polarize toward an M2, procarcinogenic macrophage phenotype (Fig. 5A).

Collectively, our study of RGZ direct action in human melanoma cells and indirect action on nonmalignant stromal cells highlights a tumorigenic action of RGZ. We next addressed the impact of RGZ chronic treatment on the progression of existing human melanoma tumors, using xenografts of human melanoma cells in immuno-deficient mice as an in vivo model. A375 cells were injected subcutaneously into the flanks of NSG immunodeficient mice. Once the tumors were established, two groups of mice (each n = 12) were fed either a control or a RGZ-supplemented diet. The growth of primary tumors was significantly enhanced upon RGZ chronic feeding compared with control, as shown by a significantly larger tumor volume at 2 and 3 weeks after tumor cell injection (Fig. 5B). Furthermore, RGZ-fed mice fulfilled the criteria for withdrawal from the experiment (tumor volume of 1,000 mm³) and were thus humanely euthanized earlier than control-fed mice (Fig. 5C). These results confirm a tumorigenic role of chronic RGZ treatment on preexisting melanoma originating from human melanoma cells expressing high levels of PPARγ.

Guided by our molecular and functional experimental data, we next compared RGZ-exposed and control primary melanoma tumors obtained in our in vivo setting, with particular attention to the levels of inflammatory markers, to tumor-associated vessels, and to cell proliferation and apoptosis. In agreement with our prior data, we detected significantly higher expression of the canonical PPARγ target gene FABP4, and of human ANGPTL4, IL1B, IL6, CSF2 (GM-CSF), and CXCL8 (IL8) in RGZ-exposed tumors compared with the tumors of the control group (Fig. 5D). Immunostaining revealed an increased density of Meca-32 positive microvessels, indicating enhanced angiogenesis in the tumors exposed to RGZ (Fig. 5E). Consistent with increased blood supply, RGZ-exposed tumors exhibited reduced relative necrotic area (Fig. 5F). Ki67 and cleaved caspase-3 immunostaining revealed that proliferation and apoptosis, respectively, did not significantly differ in RGZ-exposed and control tumors (Fig. 5G). Finally, whereas primary human fibroblasts and endothelial cells did not increase interleukin or angiogenic marker expression in response to RGZ, murine fibroblasts and primary endothelial cells did respond to direct RGZ exposure (Supplementary Fig. S4C–S4D). However, the magnitude of murine endothelial cells' and fibroblasts' responses to RGZ-treated human melanoma cell conditioned medium was greater (Supplementary Fig. S4E–S4F), suggesting that the impact of RGZ in vivo is mainly due to its action on human melanoma cells, associated with a mild direct effect on murine stromal cells.

All together, these results reveal that enhanced primary melanoma tumor growth in RGZ-fed mice in vivo is accompanied by increased inflammation and angiogenesis, two favorable conditions enabling tumor growth.

We have chosen to study RGZ as a currently approved drug for the treatment of type 2 diabetic patients and assessed its impact on melanoma, the most severe form of skin cancers. The major finding from our study is that when exposed to RGZ, a subset of human metastatic melanoma cells activates a complex blend of paracrine signals including cytokines, chemokines, and angiogenic factors, which in turn activate nonmalignant human fibroblasts, endothelial and immune cells in a tumor-friendly manner, and enhances the expression of inflammatory markers in patient-derived melanoma associated fibroblasts (Fig. 5A). Accordingly, we reveal that an RGZ-supplemented diet sustains the growth of established tumors originating from these human metastatic melanoma cells in immunodeficient mice.

Given that malignant melanoma, like other cancers, are highly heterogeneous tumors (27), it is expected that melanoma cells originating from different tumors also show variability at the molecular level. Accordingly, the secretome of melanoma cells is highly complex, and variations among individual melanoma cell lines were reported (28). In full agreement with the intrinsic variability of melanoma cells, we detected large differences in PPARγ protein levels in a variety of cell lines representative of the disease progression and in PPARG RNA levels in patient tumors. Besides variable PPARγ availability, TZDs are known to exert PPARγ off-target actions, including in melanoma cells (29). Given the complexity of TZD actions, it is not surprising that RGZ exhibits cell-specific effects that do not necessarily correlate with PPARγ expression levels and/or activity. In our study, this translates into heterogeneous responses to RGZ in terms of activation of inflammatory cytokines expression in melanoma cells, which do not necessarily involve PPARγ, and which range from no response to high amplitude responses. Importantly, we show that RGZ ability to activate the expression of proinflammatory cytokines was not restricted to melanoma cells, because a bladder cancer metastatic cell line and a cutaneous squamous cell carcinoma cell line responded in the same way.

Among the cell lines we tested, we identified a subset of human melanoma cells that responded to RGZ by activating tumorigenic paracrine signaling with their microenvironment. In light of recent research, tumors are viewed as complex ecosystems, in which the tumor microenvironment plays important roles in tumor growth and responses to therapies. Notably, dynamic interactions between tumor cells, fibroblasts, endothelial cells, and immune cells present in the stroma, mediated by a remarkable complex blend of paracrine factors, can sustain tumor growth and affect responses to treatments in melanoma and other cancers (10, 30–32). Moreover, it has been generally accepted that chronic inflammation is a condition that contributes to tumor development, including melanomagenesis (33). In our study, we selected a model of metastatic melanoma cells exhibiting a PPARγ-dependent and great increase in proinflammatory interleukin secretion in response to RGZ for further study. Using this model, we show that exposing cancer cells displaying such a profile to RGZ is at risk (e.g., metastatic melanoma or bladder cancer cells in our study), as we detected a significantly increased secretion of the cytokines IL1β and IL6, of the chemokine GM-CSF, of the proangiogenic factor IL8, and of the matrix-associated protein ANGPTL4. All these factors are implicated in a number of facets of tumor progression, like promoting invasiveness and metastases, suppressing the immune response, and sustaining angiogenesis, and they were associated with melanoma progression (33–36). In addition to these factors, our proteomic analysis revealed hundreds of dysregulated secreted proteins in these melanoma cells exposed to RGZ, underlying the remarkable complexity of reciprocal interactions that these cells can maintain with nonmalignant stromal cells of their microenvironment.

In line with the activation by RGZ of the secretion of paracrine factors with a well-established link to cancer, the conditioned medium of RGZ-exposed melanoma cells was able to induce changes in the phenotype of HDFs, of patient-derived melanoma-derived CAFs, of endothelial cells and of immune cells, all stromal cell types playing important regulatory roles in tumor development (10, 30). CAFs represent a major and highly heterogeneous cell population in the tumor stroma, where they largely contribute to tumorigenesis and affect resistance to therapies (10). In response to the secretome of RGZ-treated melanoma cells, human primary fibroblasts adopt a phenotype, which is reminiscent of CAFs, notably by expressing enhanced amounts of IL6 and IL8 (10). Furthermore, the cancer-friendly phenotype of patient-derived melanoma associated fibroblasts was enhanced. The secretome of RGZ-treated melanoma cells also enhances the angiogenic capabilities of human endothelial cells, thereby stimulating angiogenesis, the best characterized among all enabling conditions for tumor growth (37). Finally, this secretome was also able to accentuate the polarization of macrophages toward the M2 phenotype, which is described to support tumor growth, to promote an immunosuppressive microenvironment, and to enhance invasion and angiogenesis (38). In particular, increased presence of M2 macrophages in the tumor correlates with poor prognosis and short survival in patients suffering from malignant melanoma (39).

Our data demonstrate that in the context of a tumor, the concerted action of cancer cells exposed to RGZ and of consequently activated fibroblasts, endothelial cells, and M2 macrophages may support tumor growth, by promoting angiogenesis and inflammation. Although the absence of an immune system in in vivo models of human cancer cell xenografts represents a limitation, we show that RGZ accelerates the development of established human melanoma tumor xenografts. In this in vivo setting, it is possible that the impact of RGZ is mediated by the effect of RGZ not only on cancer cells, but also by a direct effect on fibroblasts and endothelial cells. However, we observed a mild nonsignificant impact of RGZ direct treatment on the phenotype of isolated murine fibroblasts and endothelial cells. We therefore propose that the enhanced growth of melanoma tumors observed in immunodeficient RGZ-fed mice is primarily due to the tumorigenic effect of RGZ on metastatic melanoma cells, likely associated with modest direct effects of RGZ on stromal cells. A similar RGZ tumorigenic activity was recently observed in a model of mouse melanoma tumors, in which RGZ accelerated tumor growth originating from murine metastatic melanoma cells grafted in young animals (40), although the underlying mechanism was not explored and the involvement of stromal cells remained unknown. In perfect agreement with our experiments performed in isolated cells, the expression of human ANGPTL4, IL1B, IL6, CSF2 (GM-CSF), and CXCL8 (IL8) was upregulated in the tumors of RGZ-fed compared with control-fed mice. These human secreted factors were shown to have the capacity of activating receptors in mouse cells (41–46). Consistently, these tumors exhibited enhanced vascularization and, accordingly, reduced necrotic areas as compared with control tumors.

PPARγ agonists of the TZD family of compounds have shown anticancer properties in most isolated cancer cells, including melanoma (2). Although a consensus as to how they affect tumorigenesis at the molecular level is lacking, TZDs and PPARγ activation have generally been associated with promotion of terminal differentiation, inhibition of cell growth, and induction of apoptosis in a wide range of human cancer cell lines (2). In line with these data, several reports suggested that TZDs and activation of PPARγ might be used to prevent melanoma development (47). Yet, TZDs and activation of PPARγ have shown very little therapeutic efficacy in clinical trials performed over the past 15 years (3, 48). Several reasons have been cited for explaining these observations, including context- and compound-specific effects; the requirement for activation of PPARγ or other targets; the stage of the tumor at the time of drug exposure; and, as suggested by a recent preclinical study, the age of the animal developing the tumor (3, 40). Our study shows that the impact of TZDs on the paracrine activity of cancer cells should also be taken into consideration. Although the vast majority of studies have focused on the cell-autonomous, anticancer effects of TZDs and PPARγ activation, their impact on cancer cell paracrine activity has been neglected. In the context of a tumor, the tumorigenic paracrine consequences of exposure to RGZ that we describe in the present study may prevail over its cell-autonomous anticancer effects, the final outcome being pro- or anticancer, depending on the dominant action of RGZ in each tumor. In the same vein as our study of melanoma, the effect of pioglitazone on the progression of lung cancer is another example of such an interplay between anticancer and tumorigenic actions, in cancer or in stromal cells. Indeed, this compound exhibits an anticancer action in the lung cancer cells while being tumorigenic in macrophages, the balance being in favor of lung cancer progression (49).

In addition to revealing a mode of action of TZDs neglected so far, our study has an impact on preclinical and clinical research. It suggests that RGZ tumorigenic activity may prevail over its anticancer actions, and that administration of TZDs after tumor initiation may be detrimental. Importantly, similar concerns regarding established bladder cancers were also discussed (3). Thus, whereas studies conducted on cohorts of diabetic patients report overall preventive benefits of TZDs (with the exception of pioglitazone, which appears to slightly increase the risk of bladder cancers), there is a potential risk associated with TZDs in established melanoma and other cancers.

P. Romero is Editor-in-Chief at Journal for Immunotherapy of Cancer, reports receiving a commercial research grant from Roche pRED - Zurich, and is a consultant/advisory board member for Immatics Biotechnologies and Matwin. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Pich, P. Meylan, P. Romero, L. Michalik

Development of methodology: C. Pich

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Pich, B. Mastelic-Gavillet, J. Hafner, M.P. Levesque, C. Jandus

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Pich, P. Meylan, B. Mastelic-Gavillet, T.N. Nguyen, B.K. Trang, H. Moser, C. Goepfert, P. Romero, C. Jandus, L. Michalik

Writing, review, and/or revision of the manuscript: C. Pich, P. Meylan, B. Mastelic-Gavillet, R. Loyon, J. Hafner, M.P. Levesque, P. Romero, C. Jandus, L. Michalik

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Pich, T.N. Nguyen, R. Loyon, H. Moser, C. Moret, C. Goepfert, L. Michalik

Study supervision: C. Pich, L. Michalik

Other (designed, performed and analyzed the experiments for Figure 4A and C. Revision of the manuscript): B. Mastelic-Gavillet

We thank the Lausanne Genomic Technologies Facility and the Lausanne Protein Analysis Facility (University of Lausanne, Switzerland) for performing the RNA-seq and the proteomics analyses, respectively.

We thank Dr. Frédéric Schütz and Dr. Linda Dib (Center for Integrative Genomics, Swiss Institute of Bioinformatics, University of Lausanne, Switzerland) for expert advice on statistical analyses, and Dr. Manfredo Quadroni (Protein Analysis Facility, University of Lausanne, Switzerland) for valuable input in the design and the interpretation of the proteomic analysis. We are grateful to Dr. Pascal Schneider (Department of Biochemistry, University of Lausanne) for expert advice and support regarding inhibition of interleukins, and to Prof. Béatrice Desvergne (Center for Integrative Genomics, University of Lausanne, Switzerland) for critical reading of the manuscript.

We thank Prof. David Fisher (Massachusetts General Hospital, Cutaneous Biology Research Center, Boston, MA) for primary human melanocytes; Prof. Paolo Dotto and Dr. Karine Lefort (Department of Biochemistry, University of Lausanne) for primary human fibroblasts; Dr. Agnese Mariotti (CEPO, Epalinges, Switzerland) for WM35 and WM793 cell lines; Prof. Daniel Constam (EPFL, Lausanne, Switzerland) for C8161 cell line; Prof. Tatiana Petrova (UNIL, Epalinges, Switzerland) for primary murine lung endothelial mLuEC cells.

We thank M. Husson, L. Mury, the Animal Facility (University of Lausanne, Switzerland), the Mouse Metabolic Evaluation Facility (University of Lausanne, Switzerland), and the Cellular Imaging Facility (University of Lausanne, Switzerland) for their expert technical assistance.

This work was supported by the Etat de Vaud, by grants from the Swiss National Science Foundation (FN31003A_169232 to L. Michalik), from the Swiss Cancer League (KFS-2900-02-2012 to L. Michalik), from the Fondation Pierre Mercier pour la Science (to L. Michalik) and the Swiss Foundation for Excellence and Talent in Biomedical Research (to C. Pich), from the Swiss National Science Foundation (31003A_156469 to P. Romero), and from the Swiss National Science Foundation grant (Ambizione PZOOP3_1614590 to C. Jandus).

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