Patients were referred for oral rehabilitation at the Reference Center of rare dental diseases (Rothschild Hospital, CRMR O-RARE, Paris, France). Diagnosis of ERS was based on clinical and radiological features (3, 8). Patients and 3 healthy sex- and age-matched controls were recruited following informed consent in accordance with the principles outlined in the declaration of Helsinki. Written informed consent was obtained from probands for the publication of any potentially identifiable images or data included in this article. The samples used were considered as operating waste according to the French law. Samples from probands and controls were harvested during oral rehabilitation and were prepared for histological or cell culture analyses (authorization CODECOH DC-2018-3382).

Control and proband gingival fibroblasts were established by plating small pieces of excised gingival on plastic dishes. Cells, particularly gingival fibroblasts, migrate out of the explant and colonize the petri dish. The flasks were filled with low glucose Dulbecco’s modified Eagle’s medium (DMEM) containing 20% fetal calf serum (FCS), 1% non-essential amino acid, penicillin/streptomycin (100mg/mL) and amphotericin B (2 ng/mL). The flasks were then placed in an incubator programmed at 37°C in a humidified atmosphere with 5% CO₂ and the cell culture medium was changed twice a week until confluence (90% after about 3 weeks). Once at confluence, the gingival fibroblasts (GFs) were trypsinized (Trypsin-EDTA, GIBCO^®, 1 mL at 0.05%) and single-cell suspensions were seeded in 25 cm² flasks containing low glucose DMEM 10% of FCS, passaged by splitting when they reached confluence, and frozen in liquid nitrogen until use. Cells at passages 3 to 6 were used in all experiments. We checked each cell culture for the morphology and the marker of fibroblasts (fibroblast-specific protein 1 [FSP1]; ab27957; Abcam, Cambridge, UK). We confirmed that cell cultures did not exhibit any morphological changes during the passages, and that FSP1 was clearly detected in these cells (data not shown). Each experiment using these cells was repeated at least three times.

A high resolution mass spectrometry (MS)-based approach was used to detect the secreted proteins from GFs cultures. GFs cultures (three controls and four ERS) were seeded and cultured in triplicates, in low glucose DMEM 10% FBS for three days and then in serum-deprived DMEM for two additional days. Each serum free cell supernatant or secretome (Controls, n=9; ERS1, n=3; ERS2, n=3; ERS3, n=3; ERS4, n=2) was analysed in a single-run of LC-MS/MS.

Each culture supernatant (serum free) was precipitated with DOC/TCA (0.1%/10%) to obtain a concentrated protein pool. Protein concentration was estimated using Bradford Assay (Biorad). Based on Bradford results, 25µg of proteins of each sample were loaded into a 7% polyacrylamide gel (Acrylamide/Bis-Acrylamide 30% [29:1], Sigma Aldrich) and an electrophoresis was performed (90 minutes at 10-20mA/gel) to stack all proteins in a small piece of gel. After Coomassie blue staining, the revealed protein bands were excised. Proteins were reduced with 5mM dithiothreitol for 40 min followed by alkylation with 20mM iodoacetamide for 40 min in the dark (all products from Sigma Aldrich). After washing steps with water and acetonitrile (Sigma Aldrich), gel bands were submitted to protein digestion by 1µg of trypsin (Promega). After overnight incubation at 37°C, several steps of peptide extraction were performed with of 0.1% formic acid (FA) in water and acetonitrile solutions. Finally, for each sample, peptide fractions were combined and dried.

For each sample, peptide fractions were solubilized in FA 0.1% (v/v) and analyzed on a LTQ-Orbitrap Elite apparatus coupled to an Easy nanoLC II system (Thermo Scientific). Peptides were injected onto an enrichment column (C18 Pepmap100, Thermo Scientific). The separation was carried out with an analytical column needle (NTCC-360/100-5-153, Nikkyo-Technos). The flow rate was 300 nL/min and the mobile phase composed of H₂O/0.1% FA (buffer A) and ACN/0.1% FA (buffer B). The elution gradient duration was 120 minutes: 0-106 min, 2-40% B; 106-110 min, 40-100% B; 110-120 min, 100% B. The mass spectrometer was operated in positive mode with CID fragmentation. For mass spectrometry settings, the capillary voltage was 1.5 kV and the temperature of the capillary was 275°C. The m/z detection range was 400-1800 in MS scan at a resolution of 60 000. The 20 most intense peptide ions were selected and the fragmentation occurred with a normalized collision energy of 35. Dynamic exclusion of already fragmented precursor ions was applied for 30 seconds.

After MS analysis, raw data were imported in Progenesis LC-MS software (NonLinear Dynamics, Newcastle, UK). To perform quantification, one sample was set as a reference and retention times of all other samples were aligned. After alignment and normalization, statistical analysis was performed using the inbuilt Progenesis statistical box called ‘one-way ANOVA’. MS/MS spectra were then exported for peptide identification with Mascot (Matrix Science, version 2.6.0). Database searches were performed with the following parameters: taxonomy: human (22,244 sequences); 1 missed cleavage; variable modification: carbamidomethyl of cysteine and oxidation of methionine. Mass tolerances for precursor and fragment ions were 10 ppm and 0.35 Da respectively. False discovery rates were calculated using a decoy-fusion approach in Mascot. Identified spectrum matches with -10logP value of 20 or higher were kept, at a FDR threshold of 5%. Mascot search results were imported into Progenesis. For each condition, the total cumulative abundance of protein was calculated by summing abundances of peptides. Proteins identified with less than 2 peptides were discarded from further analysis.

Secreted proteins identified by MS were entered in STRING database (string-db.org) to create a protein-protein association network (25). Network nodes represent all the proteins produced by a single protein-coding gene locus noting that splice isoforms or post-translational modifications are collapsed. Edges represent protein-protein associations that are specific and meaningful, such as proteins that jointly contribute to a shared function (note that this does not necessarily mean they physically bind each other). Thickness of network edges indicates the strength of data support based on text mining, experiments, databases, co-expression, neighbourhood, gene fusion and co-occurrence. The minimum required interaction score was set to a high confidence level of 0.7. The exported network image was further refined using Cytoscape 3.8.2 (https://cytoscape.org). Significantly deregulated proteins identified in CM secretome were also entered in ToppFun (ToppGene Suite: https://toppgene.cchmc.org/) that detects functional enrichment of the query list based on transcriptome, proteome, regulome (transcription factors and miRNAs) and ontologies (GO, pathway) amongst other features (26). Finally, fold of functional enrichment was determined using the GO resource powered by geneontology.org (PANTHER16.0).

For RT-qPCR and Western blot analyses cells were seeded at 7000 cells per cm² surface area in 6 well plates and in 100 mm diameter dishes respectively. For immunofluorescence cells chamber slides (Nunc Lab-Tek, Thermofisher) were used. GFs from 3 control and 4 ERS subjects were used for all the experiments. Three replicates of each experiment were performed for each test to ensure reproducibility.

For TGF-β1 treatment, recombinant human TGF-β1 at 5 ng/ml was used (R&D Systems, MN, USA). Prior to TGF-β1 treatment, nearly confluent cells were serum-starved in low-glucose DMEM for 24h and washed with serum-free DMEM. Immediately after, cells were treated with TGF-β1 for 6 hours. Unless otherwise stated ‘ERS GFs’ or ‘mutant CM’ refer to untreated cells/CM.

Western blotting was performed as previously described (8). In brief, GFs were washed twice with PBS and protein was isolated in RIPA buffer (Sigma Aldrich) containing protease (Roche Diagnostics) and phosphatase inhibitors (Calbiochem) cocktails. Before loading, protein concentration was determined by Pierce^® BCA Protein assay kit (Pierce; Waltham, MA, USA). A 25 µg quantity of protein from each sample was separated by sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to nitrocellulose membranes. Conditioned medium made from GF cultures were directly loaded on the gel. Membranes were washed with Tris-buffered saline containing 0.05% Tween-20 (TBS-T) and blocked with 5% dried milk in TBS-T. Primary antibodies for phosphorylated-SMAD3 (p-SMAD3) (ser423/425) (ab52903; Abcam; 1:1000), Periostin (EPR19934; Abcam; 1:500), FAM20A (OACD03385; Aviva; 1:500), and GAPDH (MAB374; Millipore; 1:2000) were used to incubate the membranes for 12 hours. Detection was with appropriate peroxidase-conjugated secondary antibodies (Jackson ImmunoResearch; West Grove, PA, USA; 1:2000), which were developed with SuperSignal Western Pico Chemiluminescence Substrate (Pierce).

Total RNA was isolated using commercially available kits according to manufacturer guidelines (RNeasy Mini, Qiagen) and measured (Nanodrop, Peqleb). One μg was used in a reverse transcription reaction (SuperScript First strand synthesis, Thermofisher). Quantitative-PCR was performed using Quantifast SYBR Green PCR Kit (Qiagen), reactions were performed in triplicate. Transcript levels were calculated using the standard curves generated using serial dilutions of cDNA obtained by reverse transcription of control RNA samples then normalized to HPRT. Primer sequences were listed in Supplementary Table 1. Amplification specificities were assessed by melting curve analyses and amplicons were sequenced. Values correspond to the mean of 3 independent experiments in triplicates of three control cultures and the four ERS patient cultures. Data represent mean fold gene expressions ± s.d. relative to control (without TGFβ1). Data were analyzed via two-way ANOVA with Bonferroni multiple comparisons test (*p<0.05, **p<0.01, ***p<0.001).

Gingival biopsy: Approximately, 1 cm³ gingival samples from patients were fixed for 24 h at 4°C in 4% paraformaldehyde and then embedded in paraffin wax. After sectioning, epitope retrieval was achieved by heat. Sections were incubated overnight at 4°C with primary antibodies; rabbit anti-Collagen VI (EPR17077; Abcam; 1:250), rabbit anti-phosphoSMAD3 (ab52903; Abcam; 1:100) and mouse anti-Netrin-1 (ALX-522-100; Enzo Life; 1:100).

GF culture: Cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and blocked with 1% bovine serum albumin. Primary antibodies used were rabbit anti-FAM20A (OACD 03385; Aviva; 1:250) and P-SMAD3 (ab52903; Abcam; 1:100). Golgi staining was achieved by incubating cultures with lectin HPA from Helix pomatia Alexa 647-conjugated (Thermo Fisher Scientific; 1:200).

Secondary antibodies used were Alexa 488- or Cy3-conjugated donkey anti-rabbit (Jackson Immunoresearch Laboratories, West Grove, PA; 1:500), and Alexa 488-conjugated donkey anti-mouse (Thermo Fisher Scientific; 1:200). Nuclear staining was achieved by 20 min incubation at room temperature in Hoechst 33342 (Thermo Fisher Scientific). No cellular autofluorescence and no nonspecific labeling were detected in these conditions. Images were collected by confocal microscopy (Zeiss LSM8) and processed using ZEN (Zeiss) and ImageJ softwares. ERS photomicrographs in Figure 4 are a representative of all ERS cultures.

Statistical analysis was by one-way or two-way ANOVA, as appropriate, followed by Bonferroni multiple comparisons test with Graphpad Software version 5 (Graphpad Software; La Jolla, CA, USA) (p < 0.05 was considered significant). Data are expressed as the mean ± standard deviation of 3 or 5 individual experiments with independent primary cultures from different subjects. Individual experiments included three replicates.

Primary cultures from GFs were obtained from three unrelated controls and four ERS patients carrying distinct FAM20A mutations (Table 1). We previously showed that the c.358C>T mutation (ERS1) resulted in a null protein, undetectable in ERS1 GFs (8). FAM20A could readily be detected in GFs derived from the other three patients and control subjects (Supplementary Figure 1). However, whereas in control GFs FAM20A was essentially localized in discoidal vesicles, most likely secretory ones, in the mutant GFs FAM20A was exclusively detected in HPA positive, cis-Golgi structures (Supplementary Figures 1A–D). Western blot analysis of the CM failed to detect secreted FAM20A in either control or ERS GFs (Supplementary Figure 1E).

To gain insight into the pathogenesis of gingival fibromatosis we analyzed the proteome of control and ERS-derived CM using nano-LC-MS/MS. In order to evaluate the reproducibility of the experiments, different linear regressions were performed by plotting the logarithm of protein intensities for the different samples of the same group (as mentioned in the Methods section). The averaged regression coefficient measured to evaluate the robustness of the technical scenario between biological replicates for the different groups of samples was estimated as follows, R_Ctls = 0.9869, R_ERS1 = 0.9568, R_ERS2 = 0.9095, R_ERS3 = 0.9376, R_ERS4 = 0.9670. Without any filtering criteria, nano-LC-MS analyses allowed to identify 1061 proteins (Supplementary Table S2). After applying classical proteomic filters (at least two unique matched peptides for each protein) this number was decreased to 534 proteins which are further described below. Out of these proteins, 520 were predicted to be secreted (96%), including 187 classically (as based on the presence of a signal peptide using Uniprot; https://www.uniprot.org) and 333 non-classically (Supplementary Table S3). The latter also included proteins found in extracellular vesicles (exosomes, ectosomes and apoptotic bodies) as defined in Vesiclepedia (27); (http://microvesicles.org/index.html) and Exocarta (28); (http://exocarta.org/index.html). The 17 remaining hits were most likely representing membrane shed peptides.

The set of the 187 classically secreted proteins were the most abundantly represented in all samples. We used Gene Ontology (GO) analysis to evaluate the molecular functions, biological processes and cellular components, related to these proteins. All p-values were adjusted with Bonferroni corrections. The most enriched “GO Molecular Function” represented binding and signalling of ECM structural constituents such as collagens, glycosaminoglycans or integrins (Table 2). ECM organization, collagen trimerization, and collagen fibril organization were accordingly the major “GO Biological Process” identified (Table 2).

Research of the most enriched pathways (Table 3) identified 140 out of 187 secreted proteins as belonging to the “Ensemble of genes encoding extracellular matrix and extracellular matrix-associated proteins”, “Ensemble of genes encoding core extracellular matrix including ECM glycoproteins, collagens and proteoglycans” and “Ensemble of genes encoding ECM-associated proteins including ECM-affiliated proteins, ECM regulators and secreted factors” pathways (Table 3).

In total, 149 out of the 187 proteins were predicted to be linked to structure and organization/remodeling of the ECM (29) (Figure 1A) including thirteen different types of Collagens (I, VI, VIII or XII) and several proteoglycans of the small leucine-rich repeat proteins family (Podocan, Biglycan, Decorin, Fibromodulin and Lumican, as well as Versican and the HSPG2 encoded Perlecan). With the exception of Podocan, all the above small leucine-rich repeat members were previously localized in fibroblasts from human gingiva (30) confirming the accuracy of GFs CM model. Among the other proteins, we identified the ECM regulators Fibronectin, Osteonectin (SPARC), Laminins and BHG3, the collagenolytic enzymes Matrix metallopeptidases I and II, Cathepsins as well as the so-called ‘ECM-affiliated proteins’ Annexins, Galectins, and Glypicans. Furthermore, our analysis pinpointed several secreted growth factors such as Follistatin, Follistatin-like protein 1, Angiopoietin 2 and Transforming growth factor beta 2 (TGFβ2; Figure 1A). In addition to the above 149 proteins, we found 7 proteins involved in calcium binding (Calreticulin, Calumenin, Annexins A1 and A6) or calcium homeostasis (Fetuin A, Stanniocalcin 1 and 2). Six of them were forming an interaction network as revealed by String analysis (Figure 1B). A second interactome was formed among the 9 proteins of the complement pathway (Figure 1C). In addition, 13 out of the 22 remaining secreted proteins were forming a third protein-protein association network (Figure 1D), related to lipid metabolism and iron transport and homeostasis.

The abundance of the 333 non-classically secreted proteins was lower; only 7 were relatively well represented (cytoplasmic Actin, Pyruvate Kinase, Prelamin A/C, Glyceraldehyde-3-phosphate dehydrogenase, Filamin-A, 78 kDa glucose-regulated protein, and Endoplasmin).

The high regression coefficient of the samples analysed allowed a reliable comparison of the control and mutant secretomes. Principal component analysis clearly demonstrated a segregation between control and ERS CM (Figure 2A). At the protein level, the volcano plot representation for all the identified proteins revealed a clear separation between more abundant and less abundant proteins (Figure 2B). It also allowed to visualize the109 differentially regulated proteins, 69 being more abundant (upper right) and 40 being less abundant (upper left) in CM from ERS compared to controls (Supplementary Tables S5, S6). For the purpose of this work, we decided to focus on the 50 classically secreted, differentially expressed proteins; i.e 38 over-represented and 12 under-represented proteins in the mutant CM (Tables 4, 5).

ECM and collagen fibril organization (Table 6, highlighted in blue) as well as angiogenesis (Table 6, highlighted in yellow) were the most enriched biological processes related to the group of over-expressed proteins (Table 6 and Supplementary Table S7). GO analysis of the molecular functions showed a very strong enrichment in the ECM structural constituents conferring ‘compression resistance’ and ‘tensile strength’. Molecular functions related to collagen, integrin and glycosaminoglycan binding were also significantly enriched (Table 7 and Supplementary Table S7). As expected, in this group of proteins, all the enriched pathways converged to the pathway “Ensemble of genes encoding extracellular matrix and extracellular matrix-associated proteins” (Supplementary Table S7; 15 over 21 pathways). GO disease analysis showed that among the five diseases associated to the over-expressed proteins (Supplementary Table S7) tumor angiogenesis (p-value 2.747E-8) and fibrosis, liver (p-value 1.229E-5) were highly significant. In this context, it is interesting to note that the number and size of gingival vessels and fibrotic modifications are typically observed ERS gingival tissue (8, 31).

GO analysis of the 12 under-expressed proteins showed an enrichment in ‘regulation of coagulation’, ‘ECM organization’, and ‘regulation of wound healing’ biological processes (Supplementary Table S8). Enzymatic regulation by (endo)peptidase activity (Serpine1, Serpine2 and ESM1) and calcium binding (ANXA2, ANXA5 and ANXA6) were the most significant molecular functions. Not surprisingly the most enriched pathways were “Ensemble of genes encoding extracellular matrix and extracellular matrix-associated proteins “and “Dissolution of fibrin clot” (Supplementary Table S8). Tumor angiogenesis (p-value 8.847E-5) and idiopathic pulmonary fibrosis (p-value 2.188E-3) were significantly associated diseases (Supplementary Table S8).

Moreover, 33 out of the 38 over-expressed proteins were structural (HAPLN1, Lumican, Decorin, Perlecan and Collagens type 8, type 6 and type 14) or regulating/remodeling factors (Fibronectin, BMP1, Thrombospondin 2, Nidogen 2, IGFBP7, EDIL3, QSOX1, FAP, SerpinF1, PLOD3, Cathepsin Z, TIMP2, MMP2 and Gremlin 1). STRING analysis revealed interactions among all these proteins (Figure 2C).

Ten out of the 12 under-represented proteins were ECM-associated proteins (Table 5). This set comprised the membrane traffic proteins, Annexin 2 (ANXA2) and 5 (ANXA5) involved in calcification and fibrosis (32–34), the serine protease inhibitors Serpine1 and Serpine2, Pentraxin 3 (PTX3) and the SLRP family member Biglycan (BGN) involved in ECM organization. Interactions between ANXA2, ANXA5, ANXA6, Serpine1, Serpine2, BGN, PTX3 are shown in Figure 2D.

We used lysates of GFs cultured under standard conditions to analyse the mRNAs levels of ten significantly over-represented proteins (TGFB2, Gremlin 1, Collagen alpha (1) type VIII, Collagen alpha (2) type VI, Collagen alpha (3) type VI, Matrix Metallopeptidase 2, EGF Like Repeats and Discoidin Domains 3, Fibronectin, Calumenin and Stanniocalcin 1) and of four significantly under-represented ones: PTX3, BGN, ANXA2 and Serpine1 (blue columns in Figures 3A, B). Increased mRNA levels were identified for all the over-represented proteins, suggesting that increased protein synthesis may at least partly explain their abundance in the mutant CM (Figure 3A, blue columns). Similarly, the transcripts encoding the selected under-represented proteins were significantly downregulated, with a dramatic decrease in PTX3 mRNA level (5-fold; Figure 3B, blue columns).

TGFB2, a multi-functional TGFB isoform with pro-fibrotic and pro-calcific functions was significantly increased in the mutant CM and cell lysates at the protein and mRNA levels respectively (Table 4 and Figure 3A, blue column). The TGFB2 gene was recently identified as a key factor in drug-induced gingival overgrowth and autocrine TGFβ2 signaling could contribute to the pathogenesis of hereditary or pharmacological-induced gingival fibromatosis (23, 35–37). Additionally, TGFβ2 was shown to favor ectopic calcification in various cell types including vascular smooth muscle cells, dermal fibroblasts and trabecular meshwork cells (38, 39). To exert these effects TGFβ2 employed the canonical Smad-signaling pathway, a pathway also shown to induce Gremlin1 or COL8A1 expression in various cell types (39–43).

We therefore hypothesized that impaired TGFβ signaling could contribute to the dysregulation of the ERS secretome and the pathogenesis of the ERS gingival phenotype.

In canonical TGFβ signaling binding of TGFβ1 or TGFβ2 to and activation of TGFβ- receptors results in the phosphorylation of the intracellular effectors, the cytoplasmic SMAD2 and SMAD3 proteins. The ‘activated’ phosphorylated SMAD2/3 complex translocates to the nucleus and modulates the expression of genes regulated by TGFβ (44).

To further investigate whether TGFβ signalling was intrinsically activated we analysed control and mutant GFs treated or not with recombinant TGFB1 (5ng/ml, 6h). In control GFs, TGFβ1 exposure significantly upregulated the transcription level of the selected genes; a dramatic increase (4-fold) was observed for MMP2 and Fibronectin (FN1) (Figure 3A, red columns). It is interesting to note that the levels of MMP2, Col6A2, Col6A3, FN1, EDIL3, Calumenin and Staniocalcin1 mRNA in treated control cells were similar to those of untreated ERS cells (Figure 3A, compare red Ctl to blue ERS columns). Exposure of the ERS GFs to TGFβ1 further and significantly increased the expression of TGFβ2, MMP2 and FN1 mRNA (Figure 3A). This observation may suggest that aberrant autocrine activation of the TGFβ pathway could contribute to the gingival phenotype of ERS patients.

The TGFβ1 treatment had contrasting effects on the gene expression of under-represented proteins (Figure 3B). Compared to the untreated controls, PTX3 mRNA level was dramatically decreased; Serpine1 mRNA level was significantly increased while BGN and ANXA2 mRNA levels were unchanged after treatment (Figure 3B, blue and red Ctl columns). Exposure of ERS cultures to TGFβ1 did not change the levels of PTX3, BGN or ANXA2 mRNA. We only observed an increase in Serpine1 mRNA levels (Figure 3B). It is interesting to note however that the mRNA levels of PTX3 and Serpine1 were very similar in treated control and ERS-derived GFs (Figure 3B, red Ctl and ERS columns). This set of results furthers supports the role of TGFβ and may reflect the contribution of additional pathways.

In agreement with the proteomic and quantitative RT-PCR data suggesting activation of the TGFβ pathway the levels of the effector protein phospho-SMAD3 (p-SMAD3) was significantly increased in treated control (Figures 4A, B red column) as well as treated and untreated ERS GFs (Figures 4A, B blue and red ERS columns). Immunomorphological data showing nuclear accumulation of p-SMAD3 in untreated ERS GFs (Figure 4D) and in treated control and ERS GFs (Figures 4E, F) further suggested that this pathway was intrinsically activated in ERS GFs. The ratio p-SMAD3-positive nuclei/total number of nuclei was significantly increased in untreated ERS GFs compared to untreated control GFs (Figure 4G, blue columns). Upon TGFβ1 exposure a tenfold increase in the ratio of p-SMAD3-positive nuclei to the total number of nuclei in control GFs was seen (Figures 4G, Ctl). In treated ERS GFs, a further increase in this ratio was seen but it was not significantly different from that of treated control GFs (Figure 4G).

To determine whether the results obtained in vitro reflected what occurs in the ERS gingiva we analysed the distribution of p-Smad3 and the TGFβ targets Netrin1 and COL6A in the gingival connective tissue of unaffected subjects and ERS patients (Figure 5 and Supplementary Figure 2). P-Smad3 was only occasionally seen in control gingiva (Figure 5A and Supplementary Figure 2A) but was readily observed within the nuclei of ERS fibroblasts (Figures 5B, C and Supplementary Figures 2B, C). Netrin1, a laminin-like protein involved in angiogenesis, tumor progression and fibrosis (45, 46) exhibited a scattered distribution within fibroblasts of the control gingiva (Figure 5D). In agreement with the mass spectrometry data the Netrin-1 staining was much stronger in the ERS gingiva and Netrin1 positive puncta could be seen along the entire ERS fibroblasts (Figures 5E, F). Netrin1 has previously been involved in the intercellular cross-talk among bone cells (47); its strong expression may reflect the osteogenic potential of ERS GFs (8). Finally, the expression of the fibrillar COL6A, low in the control gingiva, was patchy and strongly decorated the disorganized collagen fibres of the ERS gingiva (Figures 5G–I and Supplementary Figures 2D–F). This set of results clearly supports the hypothesis that aberrant activation of the TGFβ pathway may contribute to the ERS gingival phenotype.