Dualism of FGF and TGF-β Signaling in Heterogeneous Cancer-Associated Fibroblast Activation with ETV1 as a Critical Determinant

Summary

Heterogeneity of cancer-associated fibroblasts (CAFs) can result from activation of distinct signaling pathways. We show that in primary human dermal fibroblasts (HDFs), fibroblast growth factor (FGF) and transforming growth factor β (TGF-β) signaling oppositely modulate multiple CAF effector genes. Genetic abrogation or pharmacological inhibition of either pathway results in induction of genes responsive to the other, with the ETV1 transcription factor mediating the FGF effects. Duality of FGF/TGF-β signaling and differential ETV1 expression occur in multiple CAF strains and fibroblasts of desmoplastic versus non-desmoplastic skin squamous cell carcinomas (SCCs). Functionally, HDFs with opposite TGF-β versus FGF modulation converge on promoting cancer cell proliferation. However, HDFs with increased TGF-β signaling enhance invasive properties and epithelial-mesenchymal transition (EMT) of SCC cells, whereas HDFs with increased FGF signaling promote macrophage infiltration. The findings point to a duality of FGF versus TGF-β signaling in distinct CAF populations that promote cancer development through modulation of different processes.

Graphical Abstract

Highlights

• •FGF and TGF-β signaling exert opposite control over multiple CAF effector genes
• •ETV1 transcription factor mediates FGF effects and suppresses those of TGF-β
• •Modulation of either pathway leads to different tumor-promoting CAF populations
• •TGF-β-activated CAFs promote EMT, but FGF-activated CAFs increase inflammation

Bordignon et al. show that activation of FGF and TGF-β control opposite key CAF effectors. Suppression of one pathway leads to activation of the other and results in tumor-promoting CAF populations that elicit EMT versus inflammation. FGF/TGF-β dualism applies to distinct CAF subsets in invading desmoplastic versus non-desmoplastic skin SCCs.

Introduction

Cancer-associated fibroblasts (CAFs) are a key component of the ecosystem of solid tumors (Gascard and Tlsty, 2016, Kalluri, 2016). A unique molecular definition of CAFs does not exist. The most widely used CAF marker, α-smooth muscle actin (α-SMA), is expressed in only a subset of these cells (Sugimoto et al., 2006). Several other proposed markers also do not identify all CAFs and, in some cases, are not even specific for fibroblastic cells (Erez et al., 2010, Öhlund et al., 2014, Sugimoto et al., 2006). Emphasizing this complexity, evidence has highlighted that CAFs are heterogeneous cell populations (Costa et al., 2018, Öhlund et al., 2017, Su et al., 2018).

Determinants of CAF heterogeneity are not yet understood (Öhlund et al., 2014). Most CAFs derive from tissue resident fibroblasts, but several other sources have been proposed, such as pericytes, bone marrow-derived mesenchymal stromal cells (MSCs), tumor cells undergoing the epithelial-to-mesenchymal transition (EMT), and endothelial cells through the endothelial-to-mesenchymal transition (EndMT) (Gascard and Tlsty, 2016, Kalluri, 2016). Whether there is a connection between different cells of origin and CAF heterogeneity remains unclear, because the fibroblastic cell type is plastic and the specific tissue microenvironment is likely to provide determining instructive cues (Gascard and Tlsty, 2016, Kalluri, 2016, Raz et al., 2018). In this context, multiple signaling pathways activated by exogenous signals have been implicated in conversion of stromal fibroblasts into CAFs, even if their dynamics and interconnected function remain mostly to be established (Albrengues et al., 2014, Biffi et al., 2019, Calon et al., 2012, Gascard and Tlsty, 2016, Kalluri, 2016, Pietras et al., 2008, Yauch et al., 2008).

Transforming growth factor β (TGF-β) signaling is thought to play a central role in CAF activation, with induction of a battery of proteins connected with the fibrotic and wound-healing reaction (David and Massagué, 2018) and enhanced cancer-invasive properties (Calon et al., 2014). However, in keeping with its duality of functions, loss of TGF-β signaling in certain experimental settings may enhance rather than suppress CAF activation. Specifically, mice with conditional deletion of the TGF-β type II receptor (Tgfbr2) gene by mesenchymal expression of a Fsp1-Cre transgene spontaneously develop prostate intraepithelial neoplasia and invasive forestomach squamous cell carcinoma (SCC) (Bhowmick et al., 2004). Proposed mechanisms included increased stromal fibroblast expression of hepatocyte growth factor (HGF), Wnt family members, and several pro-inflammatory and immune-modulatory molecules, which may promote cancer development through indirect mechanisms such as immune suppressor cell infiltration (Achyut et al., 2013, Bhowmick et al., 2004, Li et al., 2008, Placencio et al., 2008). Even in human fibroblast cell lines, downmodulation of TGFBR2 expression was reported to confer tumor-promoting properties through as-yet-undefined mechanisms (Busch et al., 2015).

Another major pathway linked to fibroblast activation is fibroblast growth factor (FGF) signaling. FGF2 has been implicated in multiple fibrotic disorders (Bishen et al., 2008, Inoue et al., 2002, Strutz et al., 2000). However, FGF2 was also reported to suppress myofibroblast activation in skin wounds (Ishiguro et al., 2009), with a potentially favorable impact on hypertrophic scars (Shi et al., 2013). A positive correlation has been established between elevated FGF2 and FGF receptor 1 (FGFR1) expression in CAFs of oral SCCs and of prostate cancers and aggressive tumor behavior (Hase et al., 2006, Musumeci et al., 2011). In the skin, we previously showed that FGF activation can play a positive role in expansion of CAFs through transcriptional repression of TP53 and escape from p53-dependent stroma cell senescence (Procopio et al., 2015). The impact of FGF signaling on other aspects of CAF activation was not assessed.

Here we show that in normal dermal fibroblasts, FGF activation exerts opposite effects to TGF-β on a multiplicity of CAF effector genes, with downmodulation of either pathway inducing expression of genes responsive to the other. We identify the ETV1 transcription factor as a critical determinant of the FGF versus TGF-β duality in CAF activation, which generates distinct CAF populations that converge on promoting cancer development while eliciting different processes: EMT versus macrophage infiltration.

Results

Opposite Impact of FGF and TGF-β Signaling on the Expression of CAF Effector Genes

Conversion of stromal fibroblasts into CAFs involves induction of many genes whose differential expression can contribute to the heterogeneity of CAF populations (Gascard and Tlsty, 2016, Kalluri, 2016). Underlying this diversity, activation of different signaling pathways may be involved. To start testing this possibility, we treated three human dermal fibroblast (HDF) strains with various growth factors/cytokines previously implicated in CAF activation (Kalluri, 2016), followed by expression analysis of key CAF marker/effector genes. Among the tested factors, FGF2 and TGF-β1 caused the greatest modulation of genes, eliciting opposite effects (Figure 1A). These were confirmed by additional experiments, showing that treatment with TGF-β1 and FGF2 caused effective activation of the corresponding canonical pathways (Figures 1B and S7), accompanied by distinct changes in cell morphology (Figure 1C). qRT-PCR, immunoblot, and immunofluorescence analysis confirmed that two sets of CAF effector genes are oppositely regulated by FGF versus TGF-β stimulation (Figures 1D–1G). Genes coding for multiple extracellular matrix (ECM) proteins, growth factors, and previously described markers of CAF activation (ACTA2 and ITGA11) were all upregulated by TGF-β1 and downregulated by FGF2, while FGF activation induced different growth factors, inflammatory cytokines, and ECM remodeling enzymes, all of which were downregulated by TGF-β stimulation (Figures 1D–1G, S1A, and S1B). The same differential effects were found with a second HDF strain, and in mouse dermal fibroblasts, the treatment with the two growth factors exerted an opposite trend (Figures S1C–S1E). Combined treatment of HDFs with the two growth factors affected expression of the various CAF effector genes to a different extent, which likely reflects their prevalent mode of regulation by the two pathways (Figures 1D–1F).

Figure 1.

FGF2 and TGF-β1 Induce Opposite Sets of CAF Effector Genes in Normal HDFs

(A) qRT-PCR analysis of indicated CAF effector genes in HDFs treated with recombinant growth factors and cytokines (10 ng/mL; 72 h, as in the following experiments unless otherwise indicated). Values are expressed as Log10 ratios of treated versus untreated HDFs. n (HDF strains) = 3, mean ± SEM, repeated ANOVA with Dunnett’s multiple comparison test, ^∗∗∗p < 0.001.

(B) Immunoblot analysis of indicated proteins in HDFs treated with either FGF2 or TGF-β1 (60 min). GAPDH was used as loading control. Full scans of this and other blots are shown in Figure S7.

(C) Double staining with phalloidin (F-actin, green) and DAPI (blue) of HDFs plus/minus FGF2 or TGF-β1. Shown are representative images (left) and quantifications of average cell size (right). Scale bar, 25 μm. n (fields per condition) = 4, mean ± SEM, one-way ANOVA/Dunnett’s test, ^∗p < 0.05.

(D and E) qRT-PCR analysis of indicated CAF effector genes in HDFs plus/minus treatment with FGF2, TGF-β1, or both. Values are expressed as Log10 ratios of treated versus mock-treated HDFs. Genes are grouped in TGF-β1-induced genes (D) and FGF2-induced genes (E). n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.001. Results with another HDF strain are shown in Figures S1C and S1D.

(F) Immunoblot analysis of CAF effector proteins in HDFs treated as in (D) and (E). One blot was sequentially probed for HGFα, α-SMA, and GAPDH; the other was sequentially probed for COL1A1, MMP1, GAPDH, and Periostin (POSTN).

(G) Double immunofluorescence analysis of α-SMA (green) and MMP1 (red) in HDFs treated as in (C). Shown are representative images. Scale bar, 50 μm. Further images and quantifications are shown in Figures S1A and S1B.

We compared the effects elicited by FGF2 with members of three other subfamilies of paracrine FGFs: FGF5 (FGF4/5/6 subfamily), FGF10 (FGF3/7/10/22 subfamily), and FGF9 (FGF9/16/20 subfamily). As shown in Figure S1F, treatment with FGF9 affected the expression of various CAF effector genes in the same way as FGF2 but to a lesser extent, whereas FGF5 and FGF10 treatment had no such effects.

Loss-of-Function Studies: Inhibition of a Pathway Results in Induction of Genes Responsive to One Another

The differential impact of exogenous FGF2 versus TGF-β1 stimulation may reflect an intrinsic difference between the two pathways, but it may also result from experimental conditions, such as chosen doses and times of treatment. Accordingly, to validate the findings, we employed a genetic approach that does not depend on growth factor treatment and is instead based on expression of dominant-negative mutants of FGFR1 and TGFBR2 receptors (Figures 2A and 2B). Functionality of constructs was verified by the strongly reduced phosphorylation of ERK1/2 or SMAD2/3 in HDFs expressing dominant-negative FGFR1 (DNFGFR1) or dominant-negative TGFBR2 (DNTGFBR2), respectively, in response to the corresponding growth factors (Figures 2C and 2D). In parallel, FGF2 stimulation failed to induce MMP1 and suppressed α-SMA expression to a substantially lesser extent in DNFGFR1-expressing HDFs than in controls (Figure 2E), while induction of α-SMA by TGF-β1 was significantly reduced in DNTGFBR2-expressing HDFs (Figure 2F).

Figure 2.

Loss-of-Function Studies Highlight the Opposite Impact of FGF versus TGF-β1 Pathways

(A and B) Schematic illustrations of the mechanism of action of DNFGFR1 (A) and DNTGFBR2 (B).

(C–F) Immunoblot analysis of the indicated proteins in HDFs expressing hemagglutinin (HA)-tagged DNFGFR1 or DNTGFBR2 versus controls plus/minus FGF2 or TGF-β1 treatment for 15 min (C), 60 min (D), or 72 h (DNFGFR1 in E and DNTGFBR2 in F).

(G and H) qRT-PCR analysis of indicated CAF effector genes in HDFs expressing DNFGFR1 (G) or DNTGFBR2 (H) versus controls. Values are expressed as Log10 ratios of dominant-negative-expressing versus control HDFs. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.001.

(I and J) qRT-PCR analysis of indicated CAF effector genes in HDFs after treatment with FGFR inhibitor BGJ398 versus vehicle (DMSO) (I) or TGFBR1 inhibitor EW-7197 versus vehicle (EtOH) (J). n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗∗p < 0.0001.

Even without exogenous growth factor stimulation, HDFs expressing DNFGFR1 showed lower levels of FGF2-dependent genes (Figure 2G), whereas fibroblasts with DNTGFBR2 presented a reduction of TGF-β1-controlled genes (Figure 2H). CAF effector genes under positive TGF-β1 control were all induced in DNFGFR1-expressing HDFs (Figure 2G), and conversely, FGF2-stimulated genes were all upregulated in HDFs with DNTGFBR2 expression (Figure 2H).

The mirroring effects resulting from genetic abrogation of either pathway were also elicited by pharmacological inhibitors under clinical trials. Treatment of HDFs with BGJ398, a selective inhibitor of FGFRs (Guagnano et al., 2011), while suppressing the expression of FGF-dependent genes, induced TGF-β-related genes (Figure 2I). Similar effects were observed upon treatment with PD184352, a specific MEK inhibitor, consistent with the involvement of mitogen-activated protein kinase (MAPK)/ERK signaling downstream of FGFR activation (Turner and Grose, 2010) (Figure S2A). Converse effects were observed upon treatment of HDFs with EW-7197, a TGFBR1 inhibitor (Jin et al., 2014), which suppressed TGF-β-related genes and upregulated those responsive to FGF (Figure 2J).

The opposing effects of FGF versus TGF-β suppression were not limited to CAF effector gene expression but also extended to proliferation. We previously reported that FGF activation in HDFs and CAFs suppresses p53 expression and activity, while treatment of these cells with FGFR inhibitors causes p53-dependent cellular senescence (Procopio et al., 2015). Consistent with these findings, DNFGFR1 expression resulted in reduced proliferation of HDFs, while DNTGFBR2 expression increased it (Figure S2B). DNFGFR1 effects on CAF effector genes were similarly observed in HDFs with CRISPR-mediated TP53 gene deletion, in which induction of the p53 target gene CDKN1A was blocked (Figure S2C).

Global Gene Expression Analysis Points to the Transcription Factor ETV1 as a Critical Determinant of FGF-TGF-β Dualism

For further mechanistic insights, we examined the global gene expression profile induced by FGF activation in HDFs and compared it with that of TGF-β (Table S1). Gene set enrichment analysis (GSEA) showed that genes positively controlled by FGF2 were significantly enriched for signatures related to DNA replication, cell cycle, and proliferation, while downregulated genes correlated with gene sets related to collagen formation and extracellular matrix organization (Figures S3A and S3B). There was also an inverse correlation with gene signatures of previously reported TGF-β-induced and TGF-β-suppressed genes in skin fibroblasts (Bhattacharyya et al., 2016) (Figure S3C; Table S2).

We previously established a set of 165 CAF signature genes (Kim et al., 2017), bearing on various aspects of CAF activation. Intersection analysis of the RNA sequencing (RNA-seq) expression profiles showed that none of these genes was similarly induced by activation of FGF versus TGF-β signaling (Figure 3A).

Figure 3.

Global Gene Expression Analysis Implicates ETV1 in FGF-TGF-β Dualism

(A) CAF effector genes upregulated by FGF2 versus TGF-β1 treatment as per RNA-seq expression profiles of HDFs plus/minus FGF2 compared with published RNA-seq data of TGF-β-treated HDFs (GEO: GSE79621). A complete list of up- and downregulated genes in FGF2-treated HDFs is shown in Table S1.

(B) GSEA of expression profiles of DNFGFR1-expressing HDFs (left) or DNTGFBR2-expressing HDFs (right) versus controls against signatures of genes upregulated by TGF-β1 or FGF2 treatment as identified in (A). The complete lists of genes is shown in Table S2.

(C and D) Similar GSEA to that in (B) against the Hallmark gene set collection (Broad Institute). Shown is the positive versus negative enrichment of the EMT gene signature in DNFGFR1- versus DNTGFBR2-expressing HDFs (C) and the inverse pattern for the interferon alpha gene signature (D). A list of gene signatures positively associated with either type of HDF is shown in Figure S3F.

(E) qRT-PCR analysis of ETV1 in HDFs treated with FGF2 or TGF-β1 and HDFs expressing DNFGFR1 or DNTGFBR2. Values are expressed as Log10 ratios of treated HDFs versus controls. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05, ^∗∗p < 0.005.

(F) Immunoblot analysis of ETV1 in HDFs plus/minus FGF2 or TGF-β1 treatment. The membrane used in Figure 1F (top) was subsequently reprobed for ETV1 and GAPDH.

(G) Immunofluorescence analysis of ETV1 (red) coupled with phalloidin staining (cyan) of HDFs treated as in (F). Shown are representative images (left) and quantification (right) of average ETV1 signal intensity per cell per field. Scale bar, 50 μm. n (fields per condition) = 15, mean ± SEM, one-way ANOVA/Dunnett’s test, ^∗∗p < 0.005.

(H and I) qRT-PCR analysis (H) and immunoblot (I) of ETV1 and indicated CAF effectors in ETV1-expressing HDFs versus controls. (H) Values are expressed as Log10 ratios over control. n (independent experiments) = 3, mean ± SEM, two-tailed paired t test, ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.001.

(J and K) qRT-PCR analysis (J) and immunoblot (K) of ETV1 and indicated CAF effectors in HDFs infected with two ETV1-silencing lentiviruses (shETV1-3 and shETV1-5) versus empty-vector controls. (J) Values are expressed as Log10 ratios over control. n (independent experiments) = 3, mean ± SEM, two-tailed paired t test using the average of shETV1-3 and shETV1-5, ^∗p < 0.05, ^∗∗p < 0.005.

(L and M) qRT-PCR analysis (L) and immunoblot (M) of ETV1 and indicated CAF effectors in HDFs infected with an ETV1-silencing lentivirus (shETV1-3) versus empty-vector control plus/minus FGF2 treatment. (L) n (independent experiments) = 3, mean ± SEM, two-tailed paired t test, ^∗p < 0.05, ^∗∗p < 0.005.

(N and O) qRT-PCR analysis (N) and immunoblot (O) of ETV1 and indicated CAF effectors in ETV1-expressing HDFs versus controls plus/minus TGF-β1 treatment. (N) n (independent experiments) = 4, mean ± SEM, two-tailed paired t test, ^∗∗p < 0.005, ^∗∗∗∗p < 0.0001.

Similar transcriptomic analysis was performed with HDFs expressing either DNFGFR1 or DNTGFBR2 versus control counterparts. The profile of upregulated genes in HDFs with DNFGFR1 was positively associated with the gene signature of TGF-β-induced genes and inversely related to that of TGF-β-suppressed genes (Figures 3B and S3D). In contrast, the profile of upregulated genes in HDFs with DNTGFBR2 correlated positively with the signature of FGF-induced genes and negatively with that of FGF-suppressed genes (Figures 3B and S3D). In DNFGFR1-expressing HDFs, there was an upregulation of TGF-β ligands, specifically TGFB3, which may contribute to enforcement of TGF-β signaling in these cells, even if expression of the endogenous TGF-β receptors was downmodulated. In DNTGFBR2-expressing cells, both FGF2 and its receptor FGFR1 were downregulated, indicating that an autocrine mechanism of FGFR stimulation is unlikely (Figure S3E).

Further GSEA of the profiles of HDFs plus/minus DNFGFR1 versus DNTGFBR2 expression using a collection of publicly available gene sets (Broad Institute) showed that a signature associated with the onset of EMT, comprising multiple cytokines, growth factors, and matrix proteins, was the one most significantly enriched in DNFGFR1-expressing HDFs (Figures 3C and S3F). Conversely, inflammatory-related gene signatures, such as those related to interferon alpha and gamma response, were selectively enriched in DNTGFBR2-expressing HDFs (Figures 3D, S3F, and S3G). All these gene sets were inversely correlated in HDFs expressing the opposite dominant-negative receptor (Figures 3C, 3D, and S3G).

Opposite control of CAF effector gene transcription by FGF versus TGF-β signaling may involve differential modulation of specific transcription factors. MetaCore analysis of gene expression profiles pointed to a specific ETS family member, ETV1, as the most upregulated transcription factor in FGF2-treated HDFs (Table S1). qRT-PCR, immunoblot, and immunofluorescence analysis showed that ETV1 is strongly upregulated by FGF2 treatment of HDFs and downmodulated by TGF-β1, with opposite modulation in HDFs expressing DNFGFR1 versus DNTGFBR2 (Figures 3E–3G). ETV1 expression was not affected by HDF treatment with other growth factors (Figure S3H), and it was suppressed to a similar extent by DNFGFR1 expression in HDFs plus/minus CRISPR-mediated TP53 deletion (Figure S3I).

To assess whether modulation of ETV1 is of functional significance, we resorted to lentivirus-mediated overexpression and silencing. Increased ETV1 expression in HDFs was sufficient to strongly upregulate all FGF2-stimulated CAF effector genes and significantly reduced those induced by TGF-β1 (Figures 3H and 3I). The opposite pattern was found when ETV1 was silenced via short hairpin RNAs (shRNAs) (Figures 3J and 3K). To assess the extent to which increased ETV1 expression mediates FGF2 stimulatory activity, we evaluated the effects of this growth factor on HDFs plus/minus ETV1 silencing. As shown in Figures 3L and 3M, the ability of FGF2 to induce positive target genes like MMP1 and HGF was abolished by ETV1 silencing. The converse ability of TGF-β1 to induce genes such as ACTA2 or COL1A1 was abrogated in HDFs with elevated ETV1 expression (Figures 3N and 3O).

FGF/TGF-β Dualism in Fibroblast and CAF Heterogeneity

Substantial differences can exist among gene expression programs of cells from different individuals (Menietti et al., 2016, Storey et al., 2007). We assessed whether this inter-population heterogeneity applies to the dichotomy of ETV1 and FGF versus TGF-β signaling by analyzing levels of corresponding signature genes in transcriptomic profiles of early-passage foreskin HDFs derived from 65 donors. We found a significant anti-correlation between genes induced by FGF2 and TGF-β, and a strong positive correlation of genes within each group (Figure 4A). Similar analysis was applied to gene expression profiles of 8 skin SCC-derived CAF strains. One strain, CAF8, was characterized by concordant high levels of ETV1 and FGF-induced genes and low expression of those that were TGF-β controlled, while the opposite was found with another strain, CAF10 (Figure 4B). The other CAF strains showed more varied patterns, with an overall trend toward TGF-β-induced genes (Figure 4B). Differences in expression of ETV1 and FGF versus TGF-β-responsive genes were relatively stable as they were validated by qRT-PCR analysis of independent cultures of the same CAF strains at a later passage (Figure 4C).

Figure 4.

FGF-TGF-β Dualism in Heterogeneity of Dermal Fibroblasts and CAFs

(A) HDFs from 65 healthy individuals were analyzed by RNA-seq. Log₂ reads per kilobase of transcript per million mapped reads (RPKM) values for the indicated genes were employed to perform Pearson’s cross-correlation analysis. Shown is the resulting plot with the gene matrix ordered by unsupervised hierarchical clustering.

(B) Gene expression profiling of 8 CAF strains (passages 1–2) from different skin SCCs was analyzed by Clariom D GeneChip hybridization. Log₂ expression values for the indicated genes were used to build a heatmap with CAF strains ordered by unsupervised hierarchical clustering.

(C) qRT-PCR validation of indicated CAF effector genes in independent cultures of the same CAF strains described in (B) at a different passage (passage 3).

(D–J) Triple immunofluorescence analysis for vimentin (VIM, cyan), ETV1 (red), and α-SMA (green) of desmoplastic and non-desmoplastic SCCs. (D) Full scans of a desmoplastic versus non-desmoplastic SCC with squared areas used for high-magnification image analysis and quantification. Scale bars, 500 μm. Other two SCCs are shown in Figures S4D and S4E. Analysis of parallel SCC sections showed that most VIM-positive cells, at the chosen signal intensity threshold, were negative for keratin expression (Figures S4F and S4G).

(E and F) Representative high-magnification confocal images of stromal areas of two desmoplastic SCCs, SCC1 (E) and SCC7 (F), showing enrichment of VIM-positive cells with high α-SMA and low ETV1 expression (arrowheads). Scale bars, 10 μm.

(G and H) Representative images of stromal areas of two non-desmoplastic SCCs, SCC3 (G) and SCC14 (H), showing enrichment of VIM-positive cells with low α-SMA and high ETV1 expression. Scale bars, 10 μm.

(I and J) Quantification of the immunofluorescence analysis in 7 desmoplastic versus 7 non-desmoplastic SCCs plotted as α-SMA (I) and ETV1 (J) signal intensity values per individual VIM-positive cells in arbitrary units (dots). For each lesion, >500 cells were counted. n (fields per lesion) >20, mean ± SEM, two-tailed unpaired t test, ^∗∗∗∗p < 0.0001.

An important question was whether different levels of FGF versus TGF-β signaling can also be found in CAFs in vivo. Immunofluorescence analysis of multiple premalignant actinic keratoses (AKs) showed differences among the individual lesions in fibroblast expression of α-SMA and ETV1 (Figures S4A–S4C). Similar analysis was conducted on cutaneous SCCs with a more compact versus infiltrating desmoplastic growth pattern. The number of fibroblasts (vimentin positive and keratin negative) with elevated α-SMA expression was greater in the latter type of lesions, as would be expected for pro-fibrotic stromal cells with activated TGF-β signaling, while the number of ETV1-positive fibroblasts, as a marker of FGF signaling, was greater in non-desmoplastic lesions (Figures 4D–4J and S4D–S4G). These differences were paralleled, in desmoplastic SCCs, by cancer cells with more invasive behavior (Figure S4H) and, in non-desmoplastic SCCs, by greater macrophage infiltration (Figure S4I).

HDFs with Loss of Either FGF or TGF-β Signaling Promote Cancer Cell Growth and Tumorigenesis Impinging on Two Distinct Processes: EMT versus Inflammation

HDFs with loss of FGF versus TGF-β signaling are characterized by opposite regulation of various CAF effector genes, raising the question of how they affect neighboring cancer cells. We addressed this question by multiple in vitro and in vivo approaches. EdU (5-ethynyl-2′-deoxyuridine) labeling and sphere-forming ability of two squamous carcinoma cell lines (SCC13 and CAL27) were significantly enhanced in co-culture assays with HDFs with either DNFGFR1 or DNTGFBR2 expression (Figures 5A–5D, S5A, and S5B). Proliferation of SCC cells was increased to a similar extent when the cells were cultured with conditioned media derived from DNFGFR1- or DNTGFBR2-expressing HDFs (Figures S5C and S5D) or plated on matrix produced by these cells (Figure S5E). The effects of conditioned media from DNFGFR1- or DNTGFBR2-expressing HDFs on normal HDFs were also tested; however, no major changes in CAF effector gene expression were observed (Figure S5F). This similarity of growth-stimulatory effects can be attributed to the shared pro-mitogenic properties of different growth factors, cytokines, and matrix remodeling proteins expressed by the two types of fibroblasts (Figure S5G).

Figure 5.

HDFs with Loss of Either FGF or TGF-β Signaling Enhance Proliferation of SCC Cells In Vitro, Impinging on Different Processes

(A) 2D co-cultures of SCC13 cells in a 1:1 ratio with DNFGFR1- or DNTGFBR2-expressing HDFs versus controls. Cancer cell proliferation was assessed by EdU incorporation (cyan) and double immunofluorescence analysis for pan-keratin (red) and vimentin (green) for SCC13 and HDF cell identification. Shown are representative images (left) and quantification (right) of EdU-positive SCC13 cells. Scale bar, 200 μm. More than 1,000 cells were assessed from 15 fields per condition. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗∗p < 0.005.

(B) Similar 2D co-culture experiments to the ones performed in (A) but using CAL27 cells. More than 300 cells were assessed from 15 fields per condition. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05, ^∗∗p < 0.005. Representative pictures are shown in Figure S5A.

(C) Sphere-forming assays of SCC13 cells admixed in a 1:1 ratio with DNFGFR1- or DNTGFBR2-expressing HDFs versus controls and grown in Matrigel. Shown are representative images (left) and quantification (right) of the number of spheres per field. Scale bar, 500 μm. 6 fields per condition. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05, ^∗∗p < 0.005.

(D) Similar sphere-forming assays to the ones described in (C) but using CAL27 cells. 7 fields per condition. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05. Representative pictures are shown in Figure S5B.

(E) Immunofluorescence analysis for vimentin (green) of SCC13 cells cultured for 7 days on top of ECM derived from DNFGFR1- or DNTGFBR2-expressing HDFs versus controls. Shown are representative images (left) and quantification of vimentin-positive SCC cells (right). The same pictures have been assessed for EdU incorporation in Figure S5E. Scale bar, 200 μm. More than 500 cells were assessed from more than 10 fields per condition. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗p < 0.05.

(F) SCC13 cells cultured as in the previous panel, followed by qRT-PCR analysis of indicated EMT genes. n (independent experiments) = 3, mean ± SEM, repeated ANOVA/Dunnett’s test, ^∗∗p < 0.005, ^∗∗∗∗p < 0.0001.

(G) Organotypic assays performed by adding a 1:1 mix of CAL27 cells and DNFGFR1- or DNTGFBR2-expressing HDFs versus controls on top of a collagen gel for 7 days, followed by immunofluorescence analysis for pan-keratin (red) and vimentin (green) of gel sections. Shown are representative images (left) and quantification (right) of CAL27 cell invasion. Scale bar, 100 μm. 6 fields per condition. n (independent experiments) = 3. repeated ANOVA/Dunnett’s test, ^∗∗∗p < 0.001.

(H and I) Macrophage migration studies performed by seeding phorbol 12-myristate 13-acetate (PMA)-differentiated THP-1 cells on top of DNFGFR1- or DNTGFBR2-expressing HDFs versus controls via a transwell system. (H) Diagram representing the setup of the experiment. (I) Representative images (left) and quantification (right) of THP-1 macrophages that migrated through the pores (stained by DAPI and phalloidin). Scale bar, 200 μm. 20 fields per condition. n (independent experiments) = 3. repeated ANOVA/Dunnett’s test, ^∗∗∗p < 0.001.

Contrasting the similar pro-proliferative effects, SCC cells plated on matrix produced by DNFGFR1-expressing HDFs exhibited a scattered pattern of growth, while SCC cells plated on matrix produced by DNTGFBR2-expressing HDFs formed large expanding colonies (Figures 5E and S5E). In parallel, only SCC cells plated on matrix produced by DNFGFR1-expressing HDFs, or cultured with their conditioned media, showed increased expression of vimentin and other EMT markers, such as SNAI1/2 and ZEB1 (Figures 5E, 5F, and S5H). EMT has been previously associated with cancer cell invasion (Nieto et al., 2016). Accordingly, in a skin organotypic assay, SCC cells grown on collagen gels with DNFGFR1-expressing HDFs exhibited an increased invasive capability, which was not observed with control or DNTGFBR2-expressing HDFs (Figures 5G and S5I).

The preceding results are consistent with the finding that the most significant gene signature associated with DNFGFR1 expression was related to EMT (Figure 3C). By contrast, several inflammation-related signatures were associated with DNTGFBR2 expression (Figures 3D, S3F, and S3G), including genes for multiple chemokines with macrophage-recruiting activity (e.g., CXCL1, CXCL10, and CXCL11). Transwell migration assays were used to assess the functional significance of the finding. DNTGFBR2-expressing HDFs elicited strong recruitment of THP-1-derived macrophages, with no such effects exerted by control or DNFGFR1-expressing HDFs (Figures 5H and 5I).

To assess the in vivo significance of the results, we employed an orthotopic skin cancer model based on parallel mouse ear injections of fluorescently labeled cells (Procopio et al., 2015). SCC cell expansion was significantly enhanced by HDFs with either DNFGFR1 or DNTGFBR2 expression versus controls (Figures 6A, S6A, and S6B). Analogous tumor-enhancing properties were observed with SCC cells admixed with HDFs with either overexpression or silencing of ETV1 (Figure 6B). Histological and immunofluorescence analysis at the end of experiments showed markedly increased tumor cell density in lesions formed in the presence of DNFGFR1- or DNTGFBR2-expressing HDFs, or fibroblasts with modulated ETV1 expression, versus controls (Figures 6C, 6D, and S6C–S6H).

Figure 6.

HDFs with Loss of Either FGF or TGF-β Signaling Enhance Tumorigenic Expansion of Neighboring SCC Cells In Vivo

(A) RFP-expressing CAL27 cells were admixed with either DNFGFR1- or DNTGFBR2-expressing HDFs versus controls, followed by parallel co-injection into mouse ears. Shown are representative images of injected ear pairs at day 20 after injection (left) and quantifications of relative red fluorescence intensity values normalized to day 1 values (right). Scale bar, 2 mm. DNFGFR1, n (injected ear pairs) = 16. DNTGFBR2, n (injected ear pairs) = 8. Mean ± SEM, two-tailed paired t test, ^∗p < 0.05.

(B) Similar tumorigenicity assays to the ones shown in (A) were performed with RFP-expressing CAL27 admixed with HDFs infected with an ETV1-silencing virus (left) or ETV1-expressing virus (right) versus empty-vector controls. Shown are quantifications of SCC expansion by in vivo fluorescence imaging as in (A). shETV1-3, n (injected ear pairs) = 6. ETV1 expression, n (injected ear pairs) = 8. Mean ± SEM, two-tailed paired t test, ^∗p < 0.05.

(C) H&E staining of parallel ear lesions formed by CAL27 cells co-injected with DNFGFR1-expressing HDFs versus controls. Scale bar, 500 μm.

(D) Immunofluorescence analysis for pan-keratin (red) of lesions formed by CAL27 cells admixed with DNFGFR1- or DNTGFBR2-expressing HDFs. Shown are representative images (left) and quantifications expressed as number of pan-keratin-positive CAL27 cells per square millimeter (right). Scale bar, 100 μm. DNFGFR1, n (tumor pairs) = 4. DNTGFBR2, n (tumor pairs) = 3. Mean ± SEM, two-tailed paired t test, ^∗p < 0.05, ^∗∗p < 0.005.

(E and F) Double immunofluorescence analysis for pan-keratin (red) and KI67 (green) of nodules formed by parallel back-skin injections of SCC13 cells admixed in Matrigel with DNFGFR1-expressing HDFs (E) or DNTGFBR2-expressing HDFs (F) versus controls. Shown are representative images (left panels), together with quantification of the fraction of KI67 and pan-keratin double-positive cells (right panels). Scale bar, 100 μm. n (nodules per condition, >800 cells per nodule, 4 fields per nodule) = 5, mean ± SEM, two-tailed paired t test, ^∗p < 0.05.

(G and H) Double immunofluorescence analysis of the same nodules in the (E) and (F) for pan-keratin (red) and keratin 10 (K10, green). Shown are representative images and quantification of the fraction of K10 and pan-keratin double-positive cells (left and right panels, respectively), with DNFGFR1-expressing (G) and DNTGFBR2-expressing (H) HDFs versus controls. Scale bar, 100 μm. n (nodules per condition, >700 cells per nodule, 4 fields per nodule) = 5, mean ± SEM, two-tailed paired t test, ^∗∗p < 0.005, ^∗∗∗p < 0.001.

(I and J) Double immunofluorescence analysis of the same nodules in the previous panels for pan-keratin (red) and CD31 (green). Shown are representative images and quantification of CD31-positive area with DNFGFR1-expressing (I) and DNTGFBR2-expressing (J) HDFs versus controls. Scale bar, 100 μm. n (nodules per condition, 4 fields per nodule) = 4, mean ± SEM, two-tailed paired t test, ^∗p < 0.05, ^∗∗p < 0.005.

The similar endpoint of SCC tumor formation in the presence of DNFGFR1- or DNTGFBR2-expressing HDFs may result from initially different effects on cancer cells and/or their microenvironment. To investigate this possibility, we used an alternative skin tumorigenicity assay based on intradermal back-skin injections of cells in Matrigel, with immediate formation of nodules at time of injection and retrievable any time thereafter (Al Labban et al., 2018). Matrigel nodules with SCC cells plus/minus DNFGFR1- or DNTGFBR2-expressing HDFs versus controls were analyzed 1 week after injection. Immunofluorescence analysis revealed that DNFGFR1- or DNTGFBR2-expressing HDFs already enhanced proliferation and reduced differentiation of SCC13 cells, as assessed by staining for KI67 and keratin 10 (K10) differentiation marker, respectively (Figures 6E–6H). Angiogenesis (measured by CD31 expression) was also commonly enhanced (Figures 6I and 6J).

However, there were important differences. The α-SMA and POSTN markers were markedly induced in lesions formed with DNFGFR1-expressing HDFs cells and significantly reduced in those formed with DNTGFBR2-expressing HDFs (Figures 7A–7D). An opposite modulation was found with the FGF-responsive MMP1 (Figures 7E and 7F). Paralleling these differences, there was an increased percentage of vimentin-positive SCC13 cells only in the presence of DNFGFR1-expressing HDFs (Figures 7G and 7H), while enhanced infiltration of macrophages (F4/80-positive cells) with M2 marker expression (CD206-positive cells) was found in lesions with DNTGFBR2-expressing HDFs (Figures 7I–7L).

Figure 7.

HDFs with Opposite Modulation of FGF versus TGF-β Signaling Stimulate In Vivo SCC Cell Proliferation, Eliciting Different Tumor Microenvironments

(A–F) Immunofluorescence analysis of the same nodules as in Figures 6E–6J for α-SMA (DNFGFR1 in A and DNTGFB2 in B), Periostin (POSTN) (DNFGFR1 in C and DNTGFB2 in D), or MMP1 (DNFGFR1 in E and DNTGFB2 in F) with concomitant staining for E-cadherin or pan-keratin for SCC cell visualization as indicated. Shown are representative images (left panels) and quantifications (right panels) of α-SMA-, Periostin-, or MMP1-positive areas. Scale bar, 100 μm. n (nodules per conditions, 4 fields per nodule) = 4, mean ± SEM, two-tailed paired t test, ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.001.

(G and H) Double immunofluorescence analysis of the same nodules as in the previous panels for pan-keratin (red) and vimentin (green). Shown are representative images (left panels) and quantification (right panels) of the fraction of pan-keratin and vimentin double-positive cells with DNFGFR1-expressing (G) and DNTGFBR2-expressing (H) HDFs versus controls. Scale bar, 100 μm. n (nodules per conditions, 4 fields per nodule) = 5, mean ± SEM, two-tailed paired t test, ^∗p < 0.05.

(I–L) Immunofluorescence analysis of the same nodules as in the previous panels for F4/80 (DNFGFR1 in I and DNTGFBR2 in J) or CD206 (DNFGFR1 in K and DNTGFBR2 in L). Shown are representative images (left panels) and quantifications (right panels) of F4/80- and CD206-positive areas. Scale bar, 100 μm. n (nodules per conditions, 4 fields per nodule) = 4, mean ± SEM, two-tailed paired t test, ^∗p < 0.05, ^∗∗p < 0.005.

(M and N) Fluorescence-guided LCM of HDFs labeled with anti-PDGFRα-fluorescein isothiocyanate (FITC)-conjugated antibodies of consecutive sections of the same nodules as in the previous panels followed by qRT-PCR analysis of indicated genes. (M) Representative images of a PDGFRα-FITC-stained section with propidium iodide for nuclear visualization before capturing and of fluorescence and phase contrast images of captured cells on an LCM cap. (N) Heatmap of qRT-PCR analysis of ETV1 and indicated CAF effector genes in the captured fibroblasts, with values expressed as Log10 ratios of DNFGFR1- or DNTGFBR2-expressing HDFs versus controls. n (nodules per conditions) = 3. Scale bars, 150 μm (M, left and center) and 750 μm (M, right).

(O) Fluorescence-guided LCM of SCC13 cells labeled with anti-pan-keratin-Alexa 488-conjugated antibodies of consecutive sections of the same nodules as in the previous panels, followed by qRT-PCR analysis of indicated EMT genes. Results are shown as a heatmap of qRT-PCR results with values expressed as Log10 ratios of DNFGFR1- or DNTGFBR2-expressing HDFs versus controls. n (nodules per conditions) = 3.

Immunofluorescence results were validated by fluorescence-guided laser capture microdissection (LCM) of stromal fibroblasts and cancer cells. Divergent modulation of CAF effector genes and ETV1 was observed in the platelet-derived growth factor receptor alpha (PDGFRα)-positive fibroblasts in lesions formed by DNFGFR1- versus DNTGFBR2-expressing HDFs (Figures 7M and 7N). A similar opposition of EMT marker expression was found in pan-keratin-positive SCC cells in lesions formed in the presence of the two types of fibroblasts (Figure 7O).

Thus, similar enhancement of SCC proliferation and tumorigenicity by HDFs with increased FGF and ETV1 versus TGF-β signaling is accompanied by distinct effects on EMT versus inflammation.

Discussion

Multiple signaling pathways have been implicated in CAF activation through various paracrine and autocrine mechanisms (Gascard and Tlsty, 2016, Kalluri, 2016). We previously showed that increased FGF signaling plays a selective role in suppressing p53-mediated stromal fibroblast senescence, which occurs as a fail-safe mechanism against CAF expansion (Procopio et al., 2015). We have reported here that FGF activation induces several key CAF effector genes while suppressing others under positive TGF-β control. Increased TGF-β signaling exerts opposite functions, and a specific transcription factor, ETV1, functions as a critical determinant of the balance between FGF- versus TGF-β-controlled genes. The findings point to the importance of reciprocal-negative regulation of signaling pathways in homeostatic control of CAF activation, with modulation of different CAF effector genes with shared pro-mitogenic properties converging on promoting cancer cell proliferation, while expression of others has diverging consequences on induction of EMT versus macrophage infiltration.

An antagonistic role of FGF and TGF-β signaling has been previously reported in the context of lung fibrosis (Shimbori et al., 2016). The two pathways have instead been commonly associated with CAF activation, but a systematic comparison of their impact in this context was missing (Cirri and Chiarugi, 2011, Gascard and Tlsty, 2016, Kalluri, 2016). We have found that FGF activation in multiple strains of primary HDFs induced a consistent set of genes coding for pro-inflammatory cytokines, mitogenic growth factors, and secreted metalloproteases; and downmodulated expression of others, coding mostly for pro-fibrotic and cancer-associated matrix proteins, while TGF-β activation had an opposite impact. The dual effects of growth factor stimulation were mirrored by a loss-of-function approach, based on expression of dominant-negative receptor mutants or treatment with pharmacological inhibitors, which not only blocked the effects of one pathway but also induced those of the other.

Mechanistically, this dual impact on CAF effector genes is amenable, at least partly, to a specific transcription factor of the ETS/PEA3 family, ETV1 (Sizemore et al., 2017). Although ETV1 has been previously implicated in tumorigenesis of prostate cancer and Ewing’s sarcoma cells (de Launoit et al., 2006, Heeg et al., 2016, Oh et al., 2012), its involvement in CAF activation was unknown. Normal dermal fibroblasts are characterized by low basal levels of ETV1 expression that is under opposite FGF and TGF-β control. Upregulation of this transcription factor was sufficient to induce most CAF effectors upregulated by FGF, while its silencing suppressed these genes and induced those under positive TGF-β control. Further work will be required to elucidate the biochemical mechanism of action of ETV1 in dual control of CAF gene transcription and how expression of this transcription factor is in turn regulated. Here we focused on the in vivo significance of the results, finding that both up- and downregulation of ETV1 expression in HDFs promoted tumor growth, recapitulating the converging effects of FGF versus TGF-β activation.

Proliferation and self-renewal of SCC cancer cells was enhanced by co-culture with HDFs with opposite modulation of either FGF or TGF-β pathway, and similar effects were elicited by a conditioned medium or extracellular matrix derived from these cells. A distinct spectrum of secreted molecules and matrix proteins produced by HDFs with enhanced TGF-β signaling (by DNFGFR expression) can explain their ability to induce scattered growth of SCC cells, with associated expression of EMT markers and invasive properties. Conversely, only HDFs with increased FGF signaling (by DNTGFBR2 expression) induced recruitment of macrophages, another important determinant of tumor properties (Mantovani et al., 2017). This is of possible clinical significance, because CAFs with high TGF-β markers (α-SMA) and low FGF markers (ETV1) were preferentially found in skin SCCs with desmoplastic invasive features, while CAFs with inverse expression of the two markers were found in non-desmoplastic SCCs.

Although multiple studies have implicated TGF-β-activated fibroblasts as pro-tumorigenic and important for tumor progression (Calon et al., 2012, Casey et al., 2008), other reports have shown that loss of TGFBR2 in fibroblasts can promote tumor formation (Bhowmick et al., 2004, Cheng et al., 2008, Fang et al., 2011). These apparently paradoxical effects could be explained by our findings, suggesting that when TGF-β signaling is suppressed, FGF effects are unleashed. Specifically, a mechanism proposed for the tumor-promoting role of fibroblasts with loss of Tgfbr2 was enhanced production of HGF (Bhowmick et al., 2004, Cheng et al., 2008), a CAF effector gene that we have found to be under positive FGF and ETV1 control. Increased infiltration of inflammatory cells was also implicated as possible mechanism for the pro-tumorigenic effects of fibroblasts with compromised TGF-β signaling (Achyut et al., 2013), which is also what we have found with fibroblasts with FGF activation, with chemokines like CXCL1, CXCL5, and CXCL8 as possible mediators.

Larger clinical studies will be required to establish whether the FGF- and TGF-β-signature genes that we have identified could be useful for a CAF-based classification of tumors. Elucidation of distinct mechanisms for CAF activation could also be of relevance for the use of specific drugs aimed at normalization of the cancer stroma. In this respect, the development of an ETV1 small molecular weight inhibitor (Pop et al., 2014) could be used in combination with FGFR and TGFBR inhibitors for stroma-focused cancer prevention and treatment approaches.

STAR★Methods

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-α-SMA	Santa Cruz Biotechnology	Cat# sc-32251; RRID: AB_262054
Rat monoclonal anti-CD31	BD Biosciences	Cat# 553370; RRID: AB_394816
Mouse monoclonal anti-CD68	Agilent	Cat# M0876; RRID: AB_2074844
Goat polyclonal anti-CD206	R and D Systems	Cat# AF2535; RRID: AB_2063012
Mouse monoclonal anti-COL1A1	Santa Cruz Biotechnology	Cat# sc-293182; RRID: AB_2797597
Rabbit polyclonal anti-Erk1	Santa Cruz Biotechnology	Cat# sc-93; RRID: AB_631453
Rabbit polyclonal anti-Erk2	Santa Cruz Biotechnology	Cat# sc-154; RRID: AB_631459
Rabbit polyclonal anti-ETV1	GeneTex	Cat# GTX129202; RRID: AB_2797598
Rat monoclonal anti-F4/80 (Alexa Fluor 488-conjugated)	BioLegend	Cat# 123120; RRID: AB_893479
Rabbit polyclonal anti-GAPDH	Santa Cruz Biotechnology	Cat# sc-25778; RRID: AB_10167668
Mouse monoclonal anti-HA-probe	Santa Cruz Biotechnology	Cat# sc-7392; RRID: AB_627809
Mouse monoclonal anti-HGF	Santa Cruz Biotechnology	Cat# sc-374422; RRID: AB_10989095
Rabbit polyclonal anti-Keratin 10	BioLegend	Cat# 905404; RRID: AB_2616955
Rabbit polyclonal anti-Ki67	GeneTex	Cat# GTX20833; RRID: AB_367700
Rabbit polyclonal anti-MMP1	GeneTex	Cat# GTX100534; RRID: AB_1950926
Mouse monoclonal anti-Pan-Cytokeratin	BMA Biomedicals	Cat# T-1302; RRID: AB_1227343
Mouse monoclonal anti-Pan-Cytokeratin (Alexa Fluor 488-conjugated)	BioLegend	Cat# 628608; RRID: AB_2616664
Mouse monoclonal anti-PDGFR-alpha (FITC-conjugated)	Santa Cruz Biotechnology	Cat# sc-21789; RRID: AB_626904
Rabbit polyclonal anti-Periostin	Santa Cruz Biotechnology	Cat# sc-67233; RRID: AB_2166650
Rabbit polyclonal anti-Phospho-Erk1/2	Cell Signaling Technology	Cat# 9101; RRID: AB_331646
Rabbit monoclonal anti-Phospho-Smad2/3	Cell Signaling Technology	Cat# 8828; RRID: AB_2631089
Rabbit monoclonal anti-Smad2/3	Cell Signaling Technology	Cat# 8685; RRID: AB_10889933
Rat monoclonal anti-Vimentin	R and D Systems	Cat# MAB2105; RRID: AB_2241653
Biological Samples
HDFs: Foreskin and abdominoplasty skin samples from Lausanne University Hospital	CHUV, Lausanne, Switzerland	N/A
CAFs: surgically excised skin SCCs from Massachusetts General Hospital	MGH, Boston, MA	N/A
AKs and SCCs: from University of Zürich Biobank	USZ, Zurich, Switzerland	N/A
SCCs: from University of Tübingen	Tübingen, Germany	N/A
Chemicals, Peptides, and Recombinant Proteins
FGFR inhibitor BGJ398	Selleck Chemicals	Cat# S2183; CAS: 872511-34-7
TGFBR1 inhibitor EW-7197	AdooQ BioScience	Cat# A14213; CAS: 1352608-82-2
MEK inhibitor PD184352	Selleck Chemicals	Cat# S1020; CAS: 212631-79-3
Phorbol 12-myristate 13-acetate (PMA)	Sigma-Aldrich	Cat# P8139; CAS: 16561-29-8
Phalloidin CruzFluor 633 Conjugate	Santa Cruz Biotechnology	Cat# sc-363796
Matrigel (Growth Factor Reduced)	Corning	Cat# 354230
Collagen Type I, rat tail	Ibidi	Cat# 50202
Recombinant human FGF2	Immunotools	Cat# 11343625
Recombinant human FGF9	Immunotools	Cat# 11343634
Recombinant human FGF10	Immunotools	Cat# 11343603
Recombinant human TGFβ1	Immunotools	Cat# 11343160
Recombinant human PDGF-BB	Immunotools	Cat# 11343672
Recombinant human SHH	Immunotools	Cat# 11344070
Recombinant human IL6	Immunotools	Cat# 11340060
Recombinant human LIF	Immunotools	Cat# 11344252
Recombinant human FGF5	Cell Guidance Systems	Cat# GFH500-10
Critical Commercial Assays
Arcturus PicoPure RNA Isolation Kit	Thermo Fisher Scientific	Cat# KIT0204
Click-iT Plus EdU Alexa Fluor 647 Imaging Kit	Thermo Fisher Scientific	Cat# C10640
Direct-zol RNA Miniprep Kit	Zymo Research	Cat# R2050
Agilent RNA 6000 Nano Kit	Agilent	Cat# 5067-1511
Deposited Data
Raw and analyzed data	This paper	GEO: GSE122372
Experimental Models: Cell Lines
Human: Primary dermal fibroblasts	This paper	N/A
Mouse: Primary dermal fibroblasts	This paper	N/A
Human: Primary SCC-derived CAFs	This paper	N/A
Human: SCC13, SCC cell line	Rheinwald and Beckett, 1981	RRID: CVCL_4029
Human: CAL27, SCC cell line	Gioanni et al., 1988	RRID: CVCL_1107
Human: THP-1, monocytic cell line	Tsuchiya et al., 1980	RRID: CVCL_0006
Human: HEK293T	ATCC	RRID: CVCL_0063
Experimental Models: Organisms/Strains
Mouse: NOD.CB17-Prkdcscid/J	The Jackson Laboratory	RRID: IMSR_JAX:001303
Oligonucleotides
Human qPCR Primers, see Table S3	This paper	N/A
Mouse qPCR Primers, see Table S3	This paper	N/A
Primers for DNFGFR1-HA-tag Gateway Entry Clone, see Table S3	This paper	N/A
Primers for DNTGFBR2-HA-tag Gateway Entry Clone, see Table S3	This paper	N/A
Recombinant DNA
LentiCRISPRv2-TP53	Procopio et al., 2015	N/A
pLKO.1 human shETV1-3	Dharmacon	TRCN0000013923
pLKO.1 human shETV1-5	Dharmacon	TRCN0000013925
pLX304 empty vector	Yang et al., 2011	Addgene plasmid #25890
pLX304-ETV1	Broad Institute	CloneID: ccsbBroad304_00520
pLenti-CMV-Blast-DEST empty vector	Campeau et al., 2009	Addgene plasmid #17451
pLenti-CMV-Blast-DNFGFR1-HA	This paper	Addgene plasmid #130887
pLenti-CMV-Blast-DNTGFBR2-HA	This paper	Addgene plasmid #130888
Software and Algorithms
ImageJ	NIH	https://imagej.nih.gov/ij/
Prism 7	GraphPad software	https://www.graphpad.com
Gene set enrichment analysis (GSEA v3.0)	Subramanian et al., 2005	http://software.broadinstitute.org/gsea/
Gene set association analysis (GSAA v2.0)	Xiong et al., 2014	http://gsaa.unc.edu/

Lead Contact and Materials Availability

Plasmids generated in this study have been deposited to Addgene (#130887 and #130888). Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, G. Paolo Dotto (paolo.dotto@unil.ch).

Experimental Model and Subject Details

Human Cells and Samples

Primary human fibroblasts were derived from foreskins of circumcised 1-year-old to 5-year-old healthy males and from abdominoplasty samples of adult patients obtained from surgical procedures at the Lausanne University Hospital (CHUV, Lausanne, Switzerland) with institutional approvals and informed consent (UNIL: CER-VD 222/12).

For the isolation of HDFs, excess of connective tissue and fat was removed leaving only epidermal and dermal layers, which were separated by incubating with a 0.25% Trypsin-EDTA solution (Thermo Fisher Scientific) overnight at 4°C followed by epidermis peeling off. The dermis was then minced into small pieces and further digested using 1% (w/v) Collagenase (Sigma-Aldrich) solution in HBSS (Thermo Fisher Scientific) at 37°C for 1 hour. The reaction was stopped by adding fibroblast complete medium (DMEM enriched in Glutamine, Pyruvate and high Glucose level (Thermo Fisher Scientific) supplemented with 10% Fetal Bovine Serum (FBS, Thermo Fisher Scientific) and 1% Antibiotic Cocktail (Bio Concept)), and cell suspension filtered through a 40 μm mesh and centrifuged at 300 g. The pellet was resuspended in complete medium and plated in one 10 cm dish. After 24 hours, floating cells were aspirated and attached fibroblasts washed extensively with warm PBS before medium change. One strain of HDFs (HDF-PB1) was used for the majority of experiments. Other strains were employed to validate the findings in some specific studies. In details, in Figure 1A, results with three strains were averaged (HDF-PB1, HDF-GP, HDF-CF2); in Figures S1C and S1D, independent experiments were performed with a second strain (HDF-CF2); for RNA-seq analysis of FGF2-treated HDFs, two strains were used (HDF-GP, HDF-KL) as well as for GeneChip arrays analysis of dominant negative-expressing HDFs (HDF-GP, HDF-PB1).

Skin CAF strains, from surgically excised skin SCCs obtained from the Department of Dermatology of Massachusetts General Hospital (Boston, MA, USA) with institutional approval and informed consent (MGH: 2000P002418), were derived as follows. The tumor samples were cut in 1-2 mm pieces after removal of fat excess, followed by incubation in 2 mL of PBS containing 0.25 mg/ml of Liberase TL (Roche) for 40 minutes, at 37°C. An equal volume of FBS was added to stop the enzymatic reaction, and the dissociated tissue was filtered through a 70 μm sieve. Derived cells were centrifuged and washed three times with complete DMEM before being seeded in a 10 cm dish. All fibroblasts and CAFs were cultured in complete DMEM at 37°C and 5% CO₂, changing medium every 48-72 hours.

AK and SCC sections were obtained from the University of Zürich Biobank (Zurich, Switzerland) and the department of dermatology of the Eberhard Karls University of Tübingen (Tübingen, Germany), with institutional review board approvals and informed consent (Zurich: EK No. 647, Tübingen: 568/201BO1).

SCC13 cells (Rheinwald and Beckett, 1981), provided by James Rheinwald, and CAL27 cells (Gioanni et al., 1988), provided by Genrich Tolstonog, were cultured in DMEM/10% FBS. THP-1 cells (Tsuchiya et al., 1980), provided by Fabio Martinon, were cultured in RPMI/10% FBS. All cell lines have been routinely tested to exclude mycoplasma contamination via PCR and fluorescence microscopy upon Hoechst staining.

Mouse Cells and In Vivo Animal Studies

Primary mouse dermal fibroblasts were isolated from the back skin of wild-type mice (C57BL/6J) at postnatal day 2-3 as previously indicated for HDFs. Both male and female mice were used for the experiments.

In vivo animal studies were carried out in 6- to 10-week-old NOD.CB17-Prkdcscid/J mice (The Jackson Laboratory). In all the experiments each treatment group had a similar number of male and female mice.

All mouse work was carried out according to Swiss guidelines for the use of laboratory animals, with protocols approved by the veterinary office of Canton Vaud (animal license No. 1854.4e).

Method Details

Cell Manipulations

To study long-term gene expression changes, HDFs were treated with growth factors or cytokines by adding them in complete DMEM once and collecting RNA or protein samples 72 hours after treatment, unless otherwise indicated. After resuspension with H₂O, 10 ng/mL was used as standard concentration for the treatment with FGF2, TGFβ1, PDGF-BB, SHH, IL-6, LIF, FGF5, FGF9 and FGF10 (ImmunoTools).

To study the effectiveness of FGF2 and TGFβ1 in inducing respectively phospho-ERK1/2 and phospho-SMAD2/3 levels, HDFs were starved overnight in serum-free DMEM followed by treatment with growth factors in starvation medium for 15 or 60 minutes, as indicated.

For drug treatment, HDFs were treated 24 hours after seeding with 500 nM / 1 μM BGJ398 or PD184352 (Selleckchem) versus DMSO as vehicle control and EW-7197 (also called Vactosertib, Adooq) versus EtOH as vehicle control, collecting RNA samples 72 hours after treatment.

HDFs were genetically manipulated by infection with multiple lentiviral vectors described below, followed after 48 hours by antibiotic selection with 1 μg/mL of Puromycin for 3 days (for silencing vectors), or with 7.5 μg/mL of Blasticidin for 5 days (for expression vectors).

For the design of DNFGFR1-expressing vector, the first 1002 bp of mouse Fgfr1 c-isoform cDNA (variant 3, lacking immunoglobulin (Ig)-like domain I, GenBank: NM_001079909.2) were amplified from a plasmid received from Sabine Werner (Werner et al., 1993) fusing an HA-tag sequence at the 3′, cloned into the pENTR-D-TOPO entry vector (Thermo Fisher Scientific) and finally transferred trough the Gateway LR Clonase system (Thermo Fisher Scientific) into the pLenti-CMV-Blast-DEST (Addgene plasmid #17451) (Campeau et al., 2009).

Similarly, for the DNTGFBR2-expressing vector, the first 573 bp of human TGFBR2 cDNA (variant 2, GenBank: NM_003242.5) were amplified from cDNA previously derived from the retrotranscription through oligo(dT) primers of HDF-derived mRNA, with an HA-tag sequence in frame at the 3′. As for DNFGFR1, the amplified sequence was then cloned into the destination vector pLenti-CMV-Blast-DEST. The empty pLenti-CMV-Blast-DEST vector was used to produce control lentiviruses. Both vectors contained Blasticidin as antibiotic selection marker. The exact sequences of the PCR primers are indicated in Table S3.

Upon infection with dominant negative expressing lentiviruses, RNA or protein samples were collected 10 days after infection.

The lentiviral vector for the overexpression of human ETV1 was obtained from the CCSB-Broad lentiviral expression library (original ID: ccsbBroad304_00520). The ORF was inserted in the pLX304 vector, leading to the expression of the protein fused with a V5-tag, and containing Blasticidin as antibiotic selection marker. The pLX304 empty vector was used to produce control lentiviruses (Addgene plasmid #25890) (Yang et al., 2011). RNA or protein samples were collected 10 days after infection.

Two different shRNAs directed against human ETV1 in the pLKO.1 lentiviral vector (Dharmacon, clone IDs: TRCN0000013923, TRCN0000013925) were used to silence ETV1 gene expression. All the vectors contained Puromycin as selection marker. The sequences of the shRNA vectors are reported in Table S3. RNA or protein samples were collected 8 days after infection.

For the deletion of human TP53 gene via CRISPR-Cas9 system, HDFs were lentivirally infected with a vector expressing, together with Cas9 nuclease, the sgRNA sequence (5′-ACTTCCTGAAAACAACGTTC-3′) targeting the end of exon 2 of human TP53 gene, as previously described (Procopio et al., 2015).

For mouse ear injections, SCC13 and CAL27 cells were lentivirally infected with an RFP-expressing vector and selected with 1 mg/mL of G418 Sulfate (Corning).

Cell Assays

For EdU incorporation studies, Click-iT EdU Alexa Fluor 647 imaging kit (Thermo Fisher Scientific) was used following manufacturer’s instructions. In details, for the analysis of fibroblast proliferation, 5000 cells per condition were plated in 12-well plates on glass coverslips. 24 hours later, HDFs were incubated with 10 μM Click-iT EdU reagent for 3 hours, followed by fixation with formalin. In co-culture experiments, 2500 HDFs admixed with 2500 cancer cells (SCC13 or CAL27 cells) were plated and cultured together for 72 hours, at which point 10 μM Click-iT EdU reagent was added for 2 hours before fixation of the cells. To determine the percentage of cancer cells positive for EdU, cells were subsequently immunostained for pan-keratin and vimentin, with pan-keratin expressing cells screened for EdU positivity. Cells were counterstained with DAPI to visualize the nuclei, and slides were imaged using a Zeiss upright Axio-Imager Z1 microscope.

For sphere formation assays of fibroblasts-cancer cells co-cultures, cells were plated onto 8-well chamber slides (Corning) pre-coated with Matrigel (Corning). In brief, chambers were coated with 80 μL Matrigel per well and incubated for 1 hour at 37°C to polymerize. 1000 fibroblasts admixed with 1000 cancer cells (SCC13 or CAL27) were plated in each well. The number of spheroids was assessed 7 days after plating through an EVOS Cell Imaging System (Thermo Fisher Scientific).

To test the effects of HDF-derived conditioned medium on SCC13 cells, the different fibroblasts were seeded in 10 cm culture dishes at approximately 70% confluency and grown for 24 hours, at which point they were washed and serum-free DMEM was added. Conditioned media were collected after 24 hours, filtered with a 0.45 um filter, aliquoted, and stored at −20°C. Finally, 5000 SCC13 cells per condition were plated in 12-well plates on glass coverslips and, 24 hours after, medium was changed with conditioned medium that was replaced every 48 hours. After 7 days, 10 μM Click-iT EdU reagent was added for 2 hours and proliferation studied as described above.

To assess the effects of HDF-produced ECM on SCC13 cells, the different fibroblasts were seeded in 12-well plates on glass coverslips at confluency and incubated in complete medium for 3 days. After washing three times with PBS, cells were removed by treating with 20 mM ammonium hydroxide for 10 minutes with gentle shaking, leaving ECM intact, followed by thorough washes with deionized H₂O. 5000 SCC13 cells were then plated on each coverslip in complete DMEM and cultured for 7 days changing medium every 48 hours. EdU incorporation assays were performed as for conditioned medium experiments.

To analyze the effects of fibroblasts on macrophage recruitment, transwell migration assays were performed using 12-well hanging cell culture inserts with 8 μm pore size (Merck-Millipore) as follows. 30,000 HDFs were plated in 12-well plates, in the lower chamber. 24-hours later, their culture medium was changed in serum free RPMI and 10,000 THP-1 cells, previously differentiated into macrophages by the treatment with 10 ng/ml phorbol 12-myristate 13-acetate (PMA, Sigma-Aldrich) for 72 hours, were seeded in serum free RPMI in each top chamber. After incubation for 24 hours, cells on the upper surface of the membrane were wiped off with a cotton swab. The cells that migrated to the lower surface were stained with DAPI and Phalloidin and were counted using a Zeiss upright Axio-Imager Z1 microscope.

Organotypic assays to study cancer cell invasive properties were performed using 3D collagen gels made with type I collagen from rat tail (Ibidi) following manufacturer’s instructions. In details, 1 mL collagen gels were prepared in 24-well plates at a final concentration of 1.5 mg/ml in complete DMEM. After solidification, a mix of 200,000 CAL27 cells and 200,000 HDFs were seeded on top of the gels and left to adhere for 24 hours. At this point the gel was transferred on a sterilized metal grid pre-covered by a nylon filter, leaving the gels on top of complete DMEM, at the liquid-air interface. The organotypic cultures were allowed to develop for 7 days before fixation in 4% PFA, followed by paraffin embedding and further processing similar to tumor samples, as described below.

Gene/Protein Expression Studies

For RT-qPCR analysis, total RNA was extracted and purified from different cell types with TRI-reagent (Sigma-Aldrich) following manufacturer’s instructions. 1 μg of total RNA was reverse-transcribed to cDNA with random primers using RevertAid H Minus reverse transcriptase (Thermo Fisher Scientific). Real Time qPCR was performed with a LightCycler 480 System (Roche) using SYBR Green (Bioline) for detection. Each sample was tested in triplicate with gene-specific primers (see Table S3 for list of used primers) and results were normalized using amplification of the same cDNAs with 36β4 (RPLP0) or GAPDH primers for human samples, and Actb for mouse samples.

For immunoblot analysis, proteins were extracted from whole cell lysates with ice-cold RIPA lysis buffer (50 mM Tris-Cl pH 8.0, 2 mM EDTA, 150 mM NaCl, 1% (v/v) Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM NaF) supplemented with protease inhibitor cocktail (Thermo Fisher Scientific) for 30 minutes on ice. Total protein concentrations were measured through Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). 20-40 mg of proteins were loaded per sample on 8%–15% SDS-PAGE gels. After separation through Bio-Rad equipment for protein electrophoresis, proteins were transferred onto PVDF membranes (Immobilon-P, Merck-Millipore) and blocked in 5% non-fat milk in PBS/Tween-20. Detection was performed by peroxidase-conjugated secondary antibodies, and blots developed with SuperSignal West Pico (Thermo Fisher Scientific) or Luminata Crescendo (Merck-Millipore) Western chemiluminescent substrates through direct exposure to Fuji Medical X-ray films (Fujifilm). Some blots were stripped before further incubation with new primary antibodies by incubating membranes with a stripping buffer (50mM Tris-Cl pH 6.8, 2% SDS, 0.7% (v/v) β-mercaptoethanol) for 30 minutes at 50°C.

For immunofluorescence analysis of adherent cells, the latter were washed once with PBS and fixed with formalin for 15 minutes at room temperature. For tissues, samples were embedded in OCT compound and frozen, to be subsequently cryosectioned with a cryotome into 7-8 μm sections. Before staining, sections were dried for 30 minutes at room temperature and then fixed in paraformaldehyde (4% w/v in PBS) for 20 minutes. After fixation, cells and tissues were permeabilized for 15 minutes in PBS with 0.1%–0.3% Triton X-100, blocked for 1 hour in blocking solution (PBS with 5% donkey/goat serum and 1% BSA), and incubated overnight at 4°C with primary antibodies in blocking solution. The day after, cells or tissues were washed three times in PBS and incubated for 2 hours with secondary antibodies conjugated with Alexa Fluor fluorescent dyes (Alexa488, Alexa568, Alexa594, Alexa647, Thermo Fisher Scientific) and DAPI for nuclear counterstaining. After 3 washes, samples were mounted with Fluorescence Mounting Medium (Agilent). Immunofluorescence images were acquired with Zeiss upright Axio-Imager Z1 microscope and with Zeiss LSM880 confocal microscope. For quantifications, digitally acquired images were processed using ImageJ software.

For immunofluorescence-guided LCM, freshly cut frozen sections mounted on membrane-coated glass slides (Thermo Fisher Scientific) were immediately fixed with 75% ethanol for 30 s followed by blocking with 10% serum in PBS for 2 minutes. Sections were then incubated with FITC-conjugated antibodies against PDGFRα or Alexa Fluor 488-conjugated antibodies against pan-keratin, in combination with propidium iodide for 1 minute, subsequently washed in PBS and air-dried before processing. Stained cells were captured using ArcturusXT Microdissection System (Thermo Fisher Scientific) followed by RNA extraction with Arcturus PicoPure RNA isolation kit (Thermo Fisher Scientific) according to manufacturer’s recommendations.

For immunohistochemistry analysis, SCC sections were deparaffinized through washes in xylene and serial ethanol washes (from 100% to 80%), followed by blocking of endogenous peroxidases by peroxidase-blocking solution (3% H₂O₂ in PBS) for 10 minutes and heat-induced antigen retrieval with sodium citrate 10 mM, pH 6.0 for 15 minutes. After washing, tissues were further blocked in 10% goat serum and 1% BSA in PBS + 0.5% Tween-20 for 10 minutes. Slides were therefore incubated with primary antibodies for 1 hour at room temperature, washed with PBS + Tween-20, then incubated with secondary antibodies for 30 minutes at room temperature. DAB kit (Agilent) was applied for 3 minutes, followed by counterstaining for 4 s with Harris Hematoxylin, dehydration and mounting.

Tumorigenicity Assays

Intradermal ear injection assays were carried out in 6- to 10-week-old NOD/SCID mice as follows. 1 × 10⁵ RFP-expressing SCC13 or CAL27 cells were admixed with an equal number of HDFs expressing either DNFGFR1, DNTGFRB2, ETV1 or an shRNA against ETV1, versus respective control HDFs, followed by parallel injections of the two combinations into left and right ears of each mouse to minimize individual animal variation. In all cases, cell mixtures were concentrated by centrifugation and resuspended in 3 μL of HBSS per ear injection. Intradermal injections were performed using a 33-gauge microsyringe (Hamilton). Live images were taken every 5-7 days with a fluorescence stereomicroscope (Leica M205F) starting 24 hours post-injection. Quantification of red fluorescence intensity values (of RFP-SCC13 and RFP-CAL27 cells) was achieved by ImageJ software analysis of digitally acquired images. Mice were sacrificed for tissue analysis 2-3 weeks after injection.

Intradermal back injection assays were carried out in 6- to 10-week-old NOD/SCID mice as follows. 2.5 × 10⁵ SCC13 cells were admixed with an equal number of HDFs plus/minus either DNFGFR1 or DNTGFBR2 expression. After centrifugation, cell mixtures were re-suspended with 70 μL of Matrigel solution (Corning) per injection and intradermally injected in parallel into the left and right side of mouse back skin with 29-gauge syringes. Mice were sacrificed and Matrigel nodules retrieved for tissue analysis 1 week after injection.

Transcriptomic Analysis

For RNA-seq analysis, total RNA was extracted from HDFs through Direct-zol RNA MiniPrep kit (Zymo Research) with on-column DNase treatment and its quality was verified by Bioanalyzer (Agilent Technologies). Samples with an RNA integrity number (RIN) ≥ 8 were further processed. 500 ng (for RNA-seq of FGF2-treated HDFs) or 1 μg (for RNA-seq of HDFs from 65 different healthy donors) of total RNA per sample was used for library preparation using a NEBNext Ultra DNA Library Prep Kit for Illumina (New England Biolabs). Single read was performed on Illumina HiSeq 2500 sequencer at the Genomic Technologies Facility of the University of Lausanne (Lausanne, Switzerland). Reads were trimmed by using Trimmomatic (v0.22), followed by mapping to human hg19 reference genome using TopHat2 (v2.0.8b). Gene expression levels were then evaluated using HTSeq package (release 0.5.4p1). R package DeSeq2 was used to normalize count data, estimate biological variance and determine differential expression in FGF2-treated HDFs, generating fold-changes and p values for each class comparison. MetaCore (Clarivate Analytics) analysis was performed for further molecular function analysis of RNA-seq expression profiles.

For GeneChip array analysis, RNA was extracted from HDFs and CAFs through Direct-zol RNA MiniPrep kit (Zymo Research) coupled with DNase treatment and its quality verified by Bioanalyzer (Agilent Technologies). 50 ng (for GeneChips of SCC-derived CAFs) and 100 ng (for GeneChips of HDFs plus/minus DNFGFR1/DNTGFBR2) of total RNA was used as input for the preparation of single-strand cDNA using the GeneChip WT PLUS Reagent Kit (Thermo Fisher Scientific). Targets were then fragmented and labeled with the GeneChip WT Terminal Labeling Kit (Thermo Fisher Scientific) and hybridized on Human Clariom D arrays (Thermo Fisher Scientific) according to recommendations. GeneChips of SCC-derived CAFs were performed by Thermo Fisher Scientific (Santa Clara, CA, USA), while GeneChips of HDFs plus/minus DNFGFR1- / DNTGFBR2- expression were performed at the iGE3 Genomics Platform of the University of Geneva (Geneva, Switzerland). Data obtained for each set of samples were normalized by applying the SST-RMA algorithm using the TAC software (v4.0) (Thermo Fisher Scientific).

Gene set enrichment analysis (GSEA) for RNA-seq expression profiles was performed using GSAASeqSP software (gene set association analysis for RNA-seq data with sample permutation) (Xiong et al., 2014) from GSAA platform (GSAA_v2.0, http://gsaa.unc.edu/). GSEA for GeneChip microarray data were conducted using GSEA software (Subramanian et al., 2005). Curated gene sets were obtained from the Molecular Signatures Database (MSigDB v5.2, http://www.broadinstitute.org/gsea/msigdb/).

Quantification and Statistical Analysis

Statistical testing was performed using Prism 7 (GraphPad Software) as indicated in the figure legends. Data are presented as mean ± SEM or ratios among experimental groups and controls ± SEM. When two experimental conditions were compared, statistical significance was calculated by two-tailed t tests. For multiple comparison of more than two conditions, one-way ANOVA was employed, with Dunnett’s multiple comparison test to compare differential treatments to the same control. ^∗ p < 0.05, ^∗∗ p < 0.005, ^∗∗∗ p < 0.001, ^∗∗∗∗ p < 0.0001. For tumorigenicity assays, individual animal variability issue was minimized by contralateral injections in the same animals of control versus experimental combinations of cells. No statistical method was used to predetermine sample size in animal experiments and no exclusion criteria were adopted for studies and sample collection. No randomization was used, and the researchers were not blinded.

Data and Code Availability

The accession number for the transcriptomic data reported in this paper is GEO: GSE122372.

Published: August 27, 2019

Supplemental Information

Table S1. FGF2 Effects on HDF Gene Expression, Related to Figures 3A, 3B, and S3A–S3D

List of significant DEGs from RNA-seq profiles of FGF2-treated HDFs.

Table S2. FGF2 and TGF-β1 Signatures for GSEA, Related to Figures 3B, S3C, and S3D

FGF2 and TGFβ1 gene signatures used for GSEA.

Table S3. List of Primers, Related to STAR Methods

List of primers used for RT-qPCR and cloning.