Transcriptomic analysis of mouse limb tendon cells during development

Tendons transmit forces generated from muscle to bone in order to facilitate movement. Tendons are mainly composed of type I collagen fibres organised parallel to the axis of the tendon, which provide the tensile strength of tendons. One of the difficulties in studying tendon development is that the main molecular component, type I collagen, is not specific to tendons. The discovery of the basic helix-loop-helix transcription factor scleraxis (Scx) as a specific tendon and ligament marker was an important step in the study of tendon development (Schweitzer et al., 2001). The loss of Scx activity in mice leads to a defect in the differentiation of the force-transmitting and intermuscular tendons, while not affecting the tendons anchoring muscles to the skeleton (Murchison et al., 2007). The type II transmembrane glycoprotein tenomodulin (Tnmd) is considered a marker of differentiated tenocytes. SCX has been shown to be required and sufficient for Tnmd expression (Docheva et al., 2005; Shukunami et al., 2006; Murchison et al., 2007). Two other DNA-binding proteins, namely the zinc-finger protein early growth response 1 (EGR1) and the homeodomain protein mohawk (MKX), have been shown to be involved in tendon formation (Ito et al., 2010; Liu et al., 2010; Lejard et al., 2011; Guerquin et al., 2013). However, these two transcription factors, although important for Col1a1 transcription in tendons, are not specific to tendons.

Experiments in embryology and genetic analyses have shown that limb tendon formation relies on muscle. In the absence of muscles, stylopod (arm) and zeugopod (forearm) tendon development is initiated, but is later arrested, suggesting the requirement of signals from muscles and/or of mechanical forces to complete tendon development (Kardon, 1998; Schweitzer et al., 2001; Edom-Vovard et al., 2002; Bonnin et al., 2005). By contrast, distal (autopod) tendons form independently of muscle (Kardon, 1998; Huang et al., 2013). TGF-β and FGF are the main signalling pathways identified as being involved in stylopod and zeugopod tendon development during the muscle-dependent phase of limb tendon development (Tozer and Duprez, 2005; Schweitzer et al., 2010). However, the TGF-β and FGF signalling pathways have been shown to be involved in limb tendon development in mouse and chick embryos, respectively (Edom-Vovard et al., 2002; Pryce et al., 2009). FGF is also required and sufficient for mouse and chick axial tendon formation (Brent et al., 2003,, 2005; Smith et al., 2005).

In order to identify novel tendon markers and to determine which signalling pathways are involved during tendon development, we undertook a transcriptome analysis of mouse limb tendon cells at different stages of development. We established a list of novel tendon markers. Bioinformatics analysis of the transcriptome identified the TGF-β and MAPK pathways as those most substantially modified in limb tendon cells during development. Bioinformatics data combined with TGF-β and FGF gain- and loss-of-function in mouse limb explants and mesenchymal stem cells showed that the TGF-β/SMAD2/3 and ERK MAPK signalling pathways control the commitment of progenitor cells to enter the tendon lineage.

In order to isolate tendon cells, we took advantage of the Scx-GFP mouse line (Pryce et al., 2007), so that Scx-positive cells could be isolated by flow cytometry based on GFP fluorescence. We chose the E11.5 stage to select limb tendon progenitors and the E14.5 stage to target limb tendon differentiated cells, when tendons are well individualised. We also chose an intermediate time point at E12.5, as the transitory time point between the muscle-independent and -dependent phases of tendon formation. In the absence of muscles, Scx is normally expressed in E11.5 limbs, defining the muscle-independent phase, and is then lost in E14.5 muscleless limbs at the level of the forearm and arm (Schweitzer et al., 2001; Bonnin et al., 2005). In E12.5 muscleless limbs, Scx expression is still present in ventral limb regions but starts to be downregulated in dorsal limb regions, at the level of the forearm (Bonnin et al., 2005; Pryce et al., 2009). We thus consider E12.5 as the transient time point between the muscle-independent and -dependent phases of forearm and arm tendon formation. Forelimbs were dissected from Scx-GFP embryos at E11.5, E12.5 and E14.5 (Fig. 1A,B). Scx-GFP⁺ cells were then separated by flow cytometry (Fig. 1C); 50, 30 and 20 embryos were needed for the E11.5, E12.5 and E14.5 stages, respectively. The dissection and cytometry steps were performed three times for each time point in order to allow triplicate Affymetrix analyses.

We performed three array comparisons: (1) E11.5 versus E12.5, corresponding to the muscle-independent phase of tendon formation; (2) E12.5 versus E14.5, corresponding to the muscle-dependent phase of arm and forearm tendon formation; and (3) E11.5 versus E14.5, corresponding to tendon progenitor cells versus tendon differentiated cells (Table 1). A total of 3282 genes (more than 10% of all transcripts in the genome array) were differentially regulated in limb tendon cells during development, between E11.5 and E14.5 (Table 1). A greater number of genes were differentially regulated during the muscle-dependent phase as compared with the muscle-independent phase (1767 versus 713; Table 1).

In total, 4888 regulated transcripts (probe sets) can be hierarchically clustered (supplementary material Fig. S1A). We then asked whether differentially regulated genes in tendon cells represented specific Gene Ontology (GO) categories, which would highlight differential biological activities during development (Table 2; supplementary material Fig. S1B). GO terms related to extracellular structure organisation, cell adhesion or response to wounding were highly represented in differentiated tendon cells at E14.5, whereas GO terms related to cell cycle were highly represented in tendon progenitor cells at E11.5 (Table 2; supplementary material Fig. S1B). In addition, very high enrichment scores and significant P-values were observed for the GO terms ‘extracellular matrix’, ‘cell adhesion' and ‘collagen' in tendon cells during limb development (Table 2), consistent with the massive increase of matrix synthesis during tendon development.

Few tendon-specific markers have been identified. In addition to the Scx and Tnmd tendon markers, a series of matrix proteins has been described as being expressed or/and associated with tendon development, which we previously attempted to list (Edom-Vovard and Duprez, 2004). The mRNA relative expression of the main tendon collagen, Col1a1, and that of the tendon-associated collagens Col3a1, Col5a1, Col6a1, Col12a1 and Col14a1 was enhanced in tendon cells between E11.5 and E14.5 (Lejard et al., 2011). This is consistent with the high enrichment score of the ‘collagen' biological process during tendon development (Table 2).

One objective was to establish a list of tendon markers during development (supplementary material Table S1). We ordered the top 100 upregulated genes in E14.5 differentiated tendon cells versus E11.5 tendon progenitor cells, from high to low fold change (Table 3). We also analysed the expression of the ordered genes using Eurexpress, a transcriptome altas database for the mouse embryo. We found that the majority of the top 100 upregulated genes were expressed in tendons (Table 3). The known tendon differentiation marker Tnmd was the second most differentially expressed gene on the list, displaying a 376-fold change between E14.5 and E11.5 (Table 3). In order to validate this list of tendon genes, we chose candidates not previously known to be related to tendons, starting with aquaporin 1 (Aqp1), a water channel protein, for which the fold change in expression levels was 57.2 between E11.5 and E14.5 in the array (Table 3). The dramatic increase of Aqp1 mRNA levels was confirmed by RT-q-PCR (Fig. 2A). In situ hybridisation experiments showed that Aqp1 is expressed in mouse forelimb tendons, similar to Scx expression (Fig. 2B-E). We also chose HtrA serine peptidase 3 (Htra3) from the array and demonstrated its expression in E14.5 mouse limb tendons by in situ hybridisation (Fig. 2F-H). Aqp1 and Htra3, which were not previously known to be tendon related, displayed tendon-specific expression in mouse limbs (Fig. 2B-H). It should be noted that Scx did not appear in the array as a significantly upregulated gene, suggesting that Scx expression levels did not change between E11.5 and E14.5, consistent with the little variation in Scx mRNA expression levels in mouse limbs (supplementary material Fig. S4A). We believe that this list of genes (Table 3; supplementary material Table S1) enriched in E14.5 tendon cells constitutes an important inventory of tendon markers.

We next aimed to identify signalling pathways modified in limb tendon cells during development. We first used Genomatix software, which established gene associations with over 400 canonical pathways. In our tendon cell array, TGF-β was the top pathway, displaying the highest number of upregulated genes in the three types of comparisons (Table 4): 59 and 102 genes of the TGF-β pathway (comprising 640 components) were significantly upregulated between E11.5 and E12.5 and between E12.5 and E14.5, respectively (Table 4). Consistent with our analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID) (Table 2), components of the cell cycle were significantly downregulated during tendon cell differentiation between E12.5 and E14.5 (Table 4).

In order to confirm the modification of TGF-β components in tendon cells, we analysed the variation of signal transduction KEGG pathways, which defines pathways from a more conventional point of view than Genomatix. Consistent with the Genomatix analysis (Table 4), TGF-β (KEGG N°4350) was the signalling pathway (among those of the signal transduction group) that displayed the highest P-value in terms of being modified between the three types of comparisons (Table 5). The components of the KEGG TGF-β pathway that display significant upregulation or downregulation in expression between the three time points of comparison are listed in supplementary material Table S2 and illustrated in supplementary material Fig. S2. The mRNA expression levels of TGF-β ligands, receptors and extracellular components that were significantly differentially regulated in the arrays (supplementary material Table S2) were confirmed by RT-q-PCR analyses (Fig. 3A). Notably, all the genes encoding extracellular components [TGF-β, THBS, LTBP, decorin (DCN), TGFβi] of the ‘classical' TGF-β pathway displayed significant upregulation in differentiated tendon cells versus progenitor tendon cells (supplementary material Table S2 and Fig. S2). Two of these extracellular components, namely the thrombospondin genes Thbs2 and Thbs4, were expressed in mouse limb tendons (Fig. 3B-M) and are among the 100 top differentially expressed genes (Table 3).

Based on bioinformatics analyses, RT-q-PCR and in situ hybridisation data, we concluded that the TGF-β pathway is the most active pathway in tendon cells during mouse limb development.

Bioinformatics analyses highlighted that, in addition to being upregulated during tendon cell differentiation between E14.5 and E12.5, components of the TGF-β pathway were also significantly upregulated between E12.5 and E11.5, i.e. during the muscle-independent phase of tendon formation (Tables 4 and 5). This suggested an involvement of the TGF-β pathway at an earlier stage of the tendon program than previously believed (Pryce et al., 2009). In order to validate our bioinformatics analysis of the tendon transcriptome, we employed an ex vivo system based on mouse forelimb explants. Forelimb bud explants were dissected either at E9.5/E10.5 to target the initial phase of tendon formation or at E12.5 to target the differentiation phase of tendon formation, and were incubated for 24 h in the presence of TGFβ2 ligand for TGF-β gain-of-function experiments or specific TGF-β inhibitors for TGF-β loss-of-function experiments. We chose TGFβ2 ligand, as opposed to TGFβ1 or TGFβ3, for the gain-of-function experiments because Tgfb2 displayed higher levels of endogenous expression than Tgfb1 and Tgfb3 in mouse limbs (supplementary material Fig. S4C).

TGFβ2 was sufficient to increase Scx expression in E9.5/E10.5 mouse forelimb explants (Fig. 4A). Expression of the tendon-related Col1a1, Col1a2, Thbs2 and Thbs4 genes was also upregulated after TGFβ2 application (Fig. 4A). In E9.5/E10.5 mouse limbs, the expression levels of Tnmd and Aqp1 were considered undetectable (above 31 PCR cycles; supplementary material Fig. S4B), and TGFβ2 was unable to increase their expression in early E9.5/E10.5 mouse limb explants cultured for 24 h (data not shown). In order to test the requirement of TGF-β signalling for the initiation of tendon gene expression in forelimb buds, we blocked the TGF-β signalling pathway by applying specific TGF-β inhibitors (supplementary material Fig. S5). In the presence of the SB43 inhibitor, which blocks the TGF-β pathway at the level of the ALK4, ALK5 and ALK7 (ACVR1B, TGFBR1 and ACVR1C, respectively – Mouse Genome Informatics) receptors (Inman et al., 2002), Scx, Col1a1, Col1a2 and Thbs2 gene expression was significantly downregulated (Fig. 4A). Blockade of the SMAD2/3 intracellular pathway using the SIS3 inhibitor (Jinnin et al., 2006) also diminished the relative expression levels of Scx, Col1a1, Col1a2, Thbs2 and Thbs4 compared with control limbs in E9.5/E10.5 mouse limb explants (Fig. 4A). Consistently, application of the SIS3 inhibitor abolished Scx expression in E9.5 mouse limb explants (Fig. 4B). This showed that TGF-β was sufficient and required via the SMAD2/3 intracellular pathway for Scx, Col1a1, Col1a2, Thbs2 and Thbs4 expression in E9.5/E10.5 mouse forelimbs.

From E12.5, tendon differentiation is concomitant with important transcriptomic changes in mouse limb tendon cells (Table 1) and with an enriched expression of matrix and tendon genes (Tables 2 and 3). Consistently, the relative endogenous expression levels of Col1a1, Col1a2, Thbs2, Thbs4, Tnmd and Aqp1 were significantly upregulated in E12.5 limbs compared with E9.5 or E10.5 limbs (supplementary material Fig. S4A,B). In E12.5 limb explants, TGFβ2 was sufficient to increase the relative mRNA levels of Scx, Col1a1, Thbs2, Thbs4 and Tnmd, but not of Aqp1 (Fig. 4C). However, the blockade of TGF-β receptors (SB43) or SMAD2/3 activity (SIS3) in E12.5 limb explants did not affect Scx, Col1a2, Thbs4, Tnmd or Aqp1 gene expression, while decreasing that of Col1a1 and Thbs2 (Fig. 4C). This showed that in E12.5 mouse limbs TGF-β is sufficient for the expression of the tendon markers Scx, Col1a1, Thbs2, Thbs4 and Tnmd, while being required only for Col1a1 and Thbs2 expression.

We conclude that TGF-β is sufficient for the expression of Scx and tendon-associated genes at different stages of limb development from E9.5 to E12.5, whereas the intracellular SMAD2/3 pathway is required for Scx and tendon-associated gene expression in early E9.5/E10.5 limbs.

In addition to the TGF-β pathway, the MAPK signalling pathway (KEGG N°4010) also appeared to be significantly differentially regulated between E11.5 and E12.5 and between E11.5 and E14.5 (Table 5; supplementary material Fig. S3 and Table S3). MAPK pathways are activated by receptor tyrosine kinases, including FGF receptors (Mason et al., 2006). FGF signalling has been shown to positively regulate Scx expression in chick limb tendons, chick and mouse axial tendons and intermuscular tendons of chick stomach (Edom-Vovard et al., 2002; Brent and Tabin, 2004; Brent et al., 2005; Le Guen et al., 2009). No such evidence of FGF sufficiency exists during mouse limb tendon development. An observation from the bioinformatics data is that several FGF ligands appeared significantly upregulated in mouse tendon cells during limb development (supplementary material Table S3), although none of these specific FGF ligands has been reported to be linked with tendon development. Another striking observation is that most of the MAP kinases displaying significant variation in the transcriptome showed a significant decrease in expression, such as Map3k4, Map2k6, Mapk8 (Jnk) and Mapk12 (p38) (supplementary material Table S3). In addition, the MAP kinase phosphatase Dusp6, which is known to be a readout of active ERK1/2 (MAPK3/1) signalling during development (Lunn et al., 2007), was clearly downregulated in tendon cells during development (supplementary material Table S3). This decrease in the expression of Dusp6 and of genes encoding MAP kinases in tendon cells suggested a diminution of ERK MAPK activity in tendon cells during mouse development.

In order to validate this bioinformatics data, we blocked the ERK MAPK pathway in mouse limb explants using the PD18 inhibitor, which is known to prevent ERK phosphorylation (Bain et al., 2007). Pea3 (Etv4) and sprouty 2 (Spry2) genes are transcriptional targets of ERK MAP kinases and are considered a readout of ERK activity (O'Hagan et al., 1996; Mason et al., 2006). The dramatic loss of Pea3 and Spry2 expression showed that ERK signalling activity was downregulated in the presence of PD18 (Fig. 5A). Consistent with the bioinformatics data, ERK inhibition led to a significant activation of the expression of Scx, Col1a1, Col1a2 and Thbs2 in E9.5 mouse limb explants (Fig. 5A). FGF4 application to E9.5 mouse limb explants did not modify tendon gene expression (Fig. 5A). We conclude that inhibition of the ERK MAPK signalling pathway was sufficient to activate the expression of tendon genes, including Scx, in early mouse limbs.

Positive and negative cross-talk between the ERK and SMAD intracellular pathways have been highlighted in several cellular and in vivo contexts (Massague, 2012) (supplementary material Fig. S5). In epithelial cells, ERK1/2 activation inhibits SMAD3 transcriptional activity via phosphorylation in its linker region (Kretzschmar et al., 1999; Matsuura et al., 2005; Wrighton et al., 2009). Since mutations of these phosphorylation sites increase SMAD3 activity (Wrighton et al., 2009), we hypothesise that ERK blockade could activate SMAD3 transcriptional activity in tendon cells. To assess activity of the SMAD3 pathway we used Smad7, a negative-feedback regulator that is considered a general TGF-β transcriptional target gene (Massague, 2012). We did not observe any increase in Smad7 expression following PD18 application (Fig. 5A). Smad7 expression was also unchanged after FGF4 application (Fig. 5A), indicating an absence of cross-talk between the FGF/ERK and SMAD3 pathways in the E9.5 mouse limb context. This indicated that tendon gene activation following ERK inhibition was not a consequence of SMAD3 activation.

In order to determine whether the positive effect of TGF-β and of ERK inhibition on tendon gene expression could be additive, we systematically compared the TGF-β effect with that of simultaneous TGF-β+PD18 application on the mRNA levels of tendon genes in E9.5 mouse limb explants (Fig. 5B). We did not observe any significant increase in Smad7 expression levels in the presence of TGF-β+PD18 versus TGF-β alone (Fig. 5B), confirming the absence of any increase in SMAD3 transcriptional activity in the context of ERK inhibition (Fig. 5A). TGF-β receptors can activate various MAP kinases, including ERK (supplementary material Fig. S5) (Massague, 2012). However, in early mouse limb explants TGF-β did not activate Pea3 or Spry2 expression (Fig. 5B) and did not prevent ERK inhibition based on similar Pea3 and Spry2 downregulation with PD18 (Fig. 5A) and TGF-β+PD18 (Fig. 5B). Regarding tendon genes, there was no difference in Scx mRNA levels following TGF-β+PD18 application versus TGF-β (Fig. 5B). However, we did observe a significant difference in Col1a1 expression levels, and a non-significant tendency to increase for Col1a2 and Thbs2 expression levels following TGF-β+PD18 co-treatment versus TGF-β alone in E9.5 mouse limb explants (Fig. 5B). This showed that there was no obvious additive effect on Scx expression following TGF-β application and ERK inhibition in E9.5 mouse limbs.

The positive effect of ERK inhibition on Scx expression observed in E9.5 limb explants was lost in E10.5 limb explants, whereas the Col1a1 and Thbs2 expression levels were still significantly elevated, following PD18 application at E10.5 (supplementary material Fig. S6A). FGF4 application did not have any significant effect on tendon gene expression in E10.5 mouse limbs, as in E9.5 limbs (supplementary material Fig. S6A). Lastly, in E12.5 mouse limb explants, ERK inhibition did not activate the expression of any tendon genes and even significantly inhibited Col1a2 expression. Consistently, FGF ligand application activated Col1a2 expression levels in E12.5 mouse limb explants (supplementary material Fig. S6B).

We conclude that FGF does not positively regulate Scx expression in early and late mouse limbs and that ERK inhibition is sufficient to enhance the expression of tendon genes independently of the activation of TGF-β signalling in early E9.5 mouse limbs.

TGF-β and MAPK signalling pathways were modified in limb tendon cells during development (Tables 4 and 5), and TGFβ2 application or ERK MAPK inhibition was sufficient to activate tendon gene expression in mesodermal limb cells (Figs 4 and 5). In order to determine whether TGF-β ligands or ERK inhibition could also drive stem cells towards the tendon lineage, we utilised mesenchymal stem cells. These cells can differentiate into various tissues of mesodermal origin, when cultured in appropriate differentiation media (Caplan, 2007).

TGF-β has been shown to have a protenogenic or prochondrogenic effect depending on cell type or culture conditions (Lorda-Diez et al., 2009, 2013; Pryce et al., 2009). We used the multipotent murine C3H10T1/2 mesenchymal stem cells (Reznikoff et al., 1973). TGFβ2 has been shown to activate Scx expression in C3H10T1/2 stem cells (Pryce et al., 2009). Consistently, the relative levels of Scx and Col1a1 gene expression were significantly elevated in the presence of TGFβ2 or TGFβ3 (Fig. 6A). In contrast to the tendon markers, the relative levels of the cartilage marker Sox9 were significantly decreased in the presence of TGF-β ligands (Fig. 6A). We did not observe any increase in tendon or cartilage marker expression after simultaneous addition of TGFβ2 and TGFβ3 ligands compared with either ligand alone (Fig. 6A). The TGF-β effect on tendon and cartilage marker expression was reduced in the presence of SB43 inhibitor, which blocks the TGF-β pathway at the level of the receptors, and in the presence of SIS3 inhibitor, which blocks the SMAD2/3 intracellular pathway (Fig. 6B). We conclude that TGF-β ligand has the ability to direct mouse mesenchymal stem cells towards the tendon lineage (Scx) at the expense of the cartilage lineage (Sox9) via the SMAD2/3 pathway.

The outcome of FGF treatment on Scx expression in stem cells differs between studies. FGF2 treatment in mouse stem cells activated the expression of Scx (Ker et al., 2011), whereas FGF4 treatment in mouse tendon progenitor cells inhibits Scx expression (Brown et al., 2014). We assessed the efficiency of FGF4 and ERK inhibition (PD18) in C3H10T1/2 cells by Pea3 upregulation in the presence of FGF4 and Pea3 downregulation with PD18 (Fig. 6C). Consistent with the PD18 effect in E9.5 mouse explants (Fig. 5), ERK inhibition activated Scx and Col1a1 gene expression in C3H10T1/2 cells (Fig. 6C). PD18 treatment did not affect Smad7 expression in C3H10T1/2 cells (Fig. 6C), indicating that SMAD2/3 activity was not modified in this experimental design. ERK inhibition did not affect Sox9 expression in C3H10T1/2 cells (Fig. 6C). FGF4 had the opposite effect and led to a significant inhibition of Scx and Col1a1 gene expression in C3H10T1/2 cells (Fig. 6C).

We conclude that TGF-β ligand has the ability to direct mouse mesenchymal stem cells towards the tendon lineage at the expense of cartilage. ERK inhibition activates Scx expression, whereas FGF appeared to have an anti-tenogenic effect on mouse mesenchymal stem cells.

We have established the first transcriptome of mouse tendon cells during development. Bioinformatics analyses highlighted the TGF-β and MAPK signalling pathways as being the main signalling pathways regulated in limb tendon cells during development. Modification of these pathways in mouse limb explants or mesenchymal stem cells showed that TGF-β signalling was sufficient and required via SMAD2/3 to drive mouse mesodermal stem cells towards the tendon lineage ex vivo and in vitro. FGF did not have a tenogenic effect and the inhibition of the ERK MAPK signalling pathway was sufficient to activate Scx in mouse limb mesodermal progenitors and mesenchymal stem cells.

We have established a list of genes that are differentially expressed in limb tendon cells during development. Differentially expressed genes displayed very high fold changes (ranging from 533 to 14 for the top 100 genes). The second most differentially expressed gene of the ordered list is Tnmd, which is known to be involved in tendon formation (Docheva et al., 2005; Shukunami et al., 2006). Among the top 100 enriched genes, those encoding matrix components were found to be drastically upregulated in tendon cells during development, including Col14a1, Fmod, Dcn and Bgn. Mice mutant for these matrix genes display a tendon phenotype (Ameye et al., 2002; Ansorge et al., 2009). Among this top list, Aqp1 and Htra3 (which were previously not known to be related to tendon) were identified as expressed in E14.5 mouse limb tendons. The AQP1 hydrophobic transmembrane water channel protein is known to be responsible for the rapid response of cell volume to changes in plasma tonicity and is involved in cell proliferation, migration and adhesion processes of many cell types (Benga, 2012). However, AQP1 function in tendon development is not yet known. The serine protease gene Htra3, which is known to act as a tumour suppressor (Skorko-Glonek et al., 2013), has until now never been associated with tendon formation. Lastly, among the top enriched genes, we found components of the TGF-β pathway, including the thrombospondin genes Thbs2 and Thbs4, which were expressed in mouse limb tendons. Mice that lack Thbs2 display connective tissue abnormalities, including in tendons (Kyriakides et al., 1998). The Drosophila equivalent, Thrombospondin (Tsp), is produced by tendon cells and is essential for the formation of the integrin-mediated myotendinous junction (Subramanian et al., 2007). We believe that these genes that are differentially expressed in limb tendon cells during mouse development (listed in Table 3; supplementary material Table S1) constitute an important list of tendon markers.

Bioinformatics data highlighted a TGF-β activity in limb tendon cells during the muscle-independent phase (before E12.5) and the differentiation phase (after E12.5) of tendon formation. We confirmed that TGF-β was sufficient for tendon gene expression in early mouse limb explants. In addition, the SMAD2/3 intracellular pathway was required for Scx expression in early mouse limb explants. In early mouse limbs, Scx is also expressed in cartilage progenitors of the entheses (tendon attachment sites to bone) (Blitz et al., 2013; Sugimoto et al., 2013) and TGF-β signalling has been shown to be required for the specification of Scx⁺ enthesis progenitors (Blitz et al., 2013). The onset of Scx expression in mouse limbs is at E10 (Schweitzer et al., 2001; Sugimoto et al., 2013) and Scx expression was disrupted only at E12.5 in limb buds of Tgfb2^−/−;Tgfb3^−/− and conditional Tgfbr2 mutant mice (Pryce et al., 2009). However, Scx and Col1a1 gene expression and tendon matrix organisation are altered in adult tendons of Smad3^−/− mice (Berthet et al., 2013) and it was shown that SMAD3 is recruited to Scx regulatory regions in adult mouse tendons (Berthet et al., 2013), suggesting a direct role for SMAD signalling in Scx expression. Because we found a loss of Scx expression when SMAD2/3 was inhibited in mouse limb explants, we propose that in the Tgfb2^−/−;Tgfb3^−/− and conditional Tgfbr2 mutant mice, the endogenous SMAD2/3 intracellular pathways may be activated by alternative TGF-β superfamily ligands or receptors to initiate Scx expression. Moreover, it has been shown that the SMAD intracellular pathways can also be activated by alternative ligands (Guo and Wang, 2009; Massague, 2012), suggesting that non-canonical TGF-β signals or receptors could drive Scx expression in early mouse limbs. Based on Scx expression, TGF-β was also able to drive mesenchymal stem cells toward the tendon lineage, via SMAD2/3. These data converge on the idea that the TGF-β signalling pathway is sufficient and required via SMAD2/3 to initiate the commitment of undifferentiated mesodermal cells towards the tendon lineage ex vivo and in vitro, but direct targeting of SMAD genes in vivo will be required to determine if, during mouse development Scx expression is indeed dependent exclusively on SMAD2/3 signalling.

In addition to being sufficient for Scx expression during the early stages of mouse limb development, TGFβ2 was also sufficient for tendon gene expression in late mouse limb explants. This indicated that a TGF-β signal was involved in the positive regulation of late tendon marker expression during the tendon differentiation process. Because Tnmd mRNA expression levels were barely detectable before E12.5 in mouse limbs, we believe that a TGF-β signal participates in the initiation of Tnmd expression in E12.5 limbs. However, the intracellular SMAD2/3 pathway was only required for Col1a1 and Thbs2 expression and not for Scx, Thbs4 or Tnmd expression in E12.5 mouse limbs. This showed that other signalling pathways were required for the expression of these tendon markers during the tendon differentiation process. The calcium pathway, being the third most differentially regulated (Table 5), is a good candidate pathway. The absence of Aqp1 gene regulation by TGFβ2 also indicated the involvement of other signalling pathways during tendon cell differentiation. It is likely that mechanical forces and downstream signalling pathways are important for limb tendon differentiation after E12.5.

Bioinformatics analyses identified the MAPK pathway as being significantly modified in tendon cells during mouse limb development. The significant decrease in expression of MAP kinases and the phosphatase Dusp6 in the array suggested a reduction of MAP kinase activity in tendon cells during development. This tendency was surprising given the positive effect of FGF signal on Scx expression in chick limbs and chick and mouse somites (Edom-Vovard et al., 2002; Brent et al., 2003,, 2005). In addition, the positive effect of FGF on Scx expression has been shown to occur via the Ets transcription factor PEA3 and the ERK MAP kinases in chick axial tendons (Brent and Tabin, 2004; Smith et al., 2005). In contrast to the chick model and mouse somites, we found that FGF does not activate Scx expression in mouse limb explants (Fig. 5; supplementary material Fig. S6) nor in mouse mesenchymal stem cells (Fig. 6). We even observed a significant reduction of Scx and Col1a1 expression in mouse C3H10T1/2 cells in the presence of FGF4 (Fig. 6). This decrease in Scx expression is consistent with that observed in mouse tendon progenitor cells upon FGF4 treatment (Brown et al., 2014). Moreover, the inhibition of ERK MAPK signalling appeared to be sufficient for inducing Scx expression in E9.5 mouse limb mesodermal progenitors and in mesenchymal stem cells. This result is consistent with the requirement of FGF loss to promote cell differentiation in many tissues (Mathis et al., 2001; ten Berge et al., 2008; Chang et al., 2013). The ability of ERK inhibition to activate Scx expression was only observed in E9.5 limb explants and not at later stages. E9.5 corresponds to the developmental time when limb mesodermal progenitor cells will commit to the tendon lineage based on Scx expression.

To date, we conclude that ERK inhibition is sufficient to prime mouse stem cells for the tendon lineage and that the FGF signalling pathway has a different role in mouse limb tendon development than that in chick limb tendon development. Experiments are underway with the aim of furthering our understanding of the differences in the involvement of the FGF/ERK pathway in limb tendon development between the chick and mouse models.

In summary, we have established a list of genes enriched in limb tendon cells during mouse development. We have shown that TGF-β signalling is sufficient and required via SMAD2/3 to drive mouse mesodermal stem cells towards the tendon lineage ex vivo and in vitro. In contrast to chick, in the mouse FGF does not have a tenogenic effect and inhibition of the ERK MAPK signalling pathway is sufficient to activate Scx in mouse limb mesodermal progenitors and mesenchymal stem cells.

Scx-GFP (Pryce et al., 2007) or wild-type (Janvier, France) mouse embryos were collected after natural overnight matings. For staging, fertilisation was considered to take place at 12.00 a.m.

Forelimbs from E11.5, E12.5 and E14.5 Scx-GFP embryos were collected and dissociated. Cell suspensions were subjected to FACS using a MoFlo XDP flow cytometer (Beckman Coulter) with the Dako-Moflo Summit software (Dako, Agilent Technologies) or using a VantageTM SE option DiVa flow cytometer (Becton-Dickinson; laser 488 nm). The GFP⁺ fractions were collected in PBS containing 2 mM EDTA and 20% foetal calf serum. RNA quantity was monitored on Agilent RNA Pico LabCHips.

Fragmented biotin-labelled cRNA samples were hybridised on Affymetrix GeneChip Mouse Genome 430 2.0 arrays that contain 45,000 probe sets. Each probe set consists of 22 probes of 25 bp with 11 perfect matches and 11 mismatches. For each experimental group (E11.5, E12.5 and E14.5), three biological replicates were hybridised. Microarray analysis was performed using a high-density oligonucleotide array (Affymetrix) on the ProfileXpert core facility. Total RNA (100 ng) was amplified and biotin-labelled using GeneChip 3′ IVT Express target labelling, control reagents and procedures from Affymetrix. Before amplification, spikes of synthetic mRNA at different concentrations were added to all samples; these positive controls were used to ascertain the quality of the process. Biotinylated antisense cRNA for microarray hybridisation was prepared. After final purification using magnetic beads, cRNA quantification was performed with a NanoDrop (Thermo Scientific) and quality checked with an Agilent 2100 Bioanalyzer.

Biotin-labelled cRNA samples (15 µg) were fragmented, denatured and hybridised on Affymetrix arrays for 16 h at 45°C with constant mixing by rotation at 60 rpm in a GeneChip hybridisation oven 640 (Affymetrix). After hybridisation, arrays were washed and stained with streptavidin-phycoerythrin (Invitrogen) in a Fluidic Station 450 (Affymetrix) according to the manufacturer's instruction. The arrays were read with a confocal laser (GeneChip scanner 3000, Affymetrix). Then, CEL files were generated using Affymetrix GeneChip Command Console software 3.0. The array has been submitted to the GEO repository with accession number GSE54207.

The microarray data were normalised with Affymetrix Expression Console software using the MAS5 statistical algorithm. Normalised data were compared and filtered using Partek Genomic Suite software 6.5. Pairwise comparisons were performed between each developmental stage (E11.5, E12.5 and E14.5). Each sample from one group was compared with each sample from the other group and only genes showing a variation of 1.5-fold were considered significantly differentially regulated.

DAVID was used to identify enriched GO terms. Genomatix software was used to identify signalling pathways based on literature data mining. Consequently, a Genomatix pathway includes a larger number of components than canonical pathways. The KEGG signal transduction pathways are a collection of manually drawn pathway maps representing current knowledge on the molecular interaction and reaction networks for a wide range of biological processes. DAVID was used to identify regulated KEGG pathways.

Limb buds were dissected from E9.5, E10.5 and E12.5 mouse embryos, embedded in collagen and cultured at 37°C in 5% CO₂ in Optimem medium (Diez del Corral et al., 2003). Explants were treated with recombinant human TGFβ2 (R&D Systems) at 20 ng/ml or with FGF4 (R&D Systems) at 200 ng/ml, for 24 h. The TGFβ2 signalling pathway was blocked using SB431542 (SB43, Selleck Chemicals) or SIS3 (Merck) chemical inhibitors; the ERK signalling pathway was blocked using PD184352 (PD18) chemical inhibitor (Axon Medchem). All inhibitors were diluted in DMSO (Fluka) and added to the medium for 24 h at 10 µM (SB43), 20 µM (SIS3) or 3.3 µM (PD18). Media with buffers only were used as controls. After treatments, explants were fixed and processed for RT-q-PCR or in situ hybridisation.

Total RNAs were extracted from forelimb FACS-sorted Scx-GFP cells at different developmental stages, mouse C3H10T1/2 cells or mouse limb explants. RNA (300 ng to 1 μg) was reverse transcribed using the High Capacity Retrotranscription Kit (Applied Biosystems). RT-q-PCR was performed using SYBR Green PCR Master Mix (Applied Biosystems). Primer used for RT-q-PCR are listed in supplementary material Table S4. Relative mRNA levels were calculated using the 2^−ΔΔCt method (Livak and Schmittgen, 2001). The ΔCts were obtained from Ct normalised with Gapdh, Hprt or 18S levels in each sample. For mRNA level analyses of Scx-GFP cells at different developmental stages, three independent RNA samples originating from three FACS-sorted experiments were analysed in triplicate. For mRNA level analyses of C3H10T1/2 cell cultures, five independent RNA samples were analysed in duplicate. For mRNA level analyses in mouse limb explant cultures, 8-12 independent RNA samples were analysed in duplicate. For E9.5, E10.5 and E12.5 mouse limb explants, we pooled 14, 11 and 6 limb buds, respectively, to obtain enough material in RNA samples. Data were analysed by unpaired Student's t-test using Microsoft Excel.

Forelimbs from E14.5 wild-type mouse embryos were fixed in Farnoy and processed for in situ hybridisation using 8 μm wax tissue sections. Mouse limb explants were fixed in 4% formaldehyde. The digoxigenin-labelled mRNA probe for mouse Scx was used as described (Lejard et al., 2011). cDNAs for Aqp1, Thbs2 and Thbs4 were cloned by PCR in pCRII-TOPO (Invitrogen). Htra3 cDNA was cloned by PCR in pBluescript KS (Addgene). The probes were prepared by plasmid linearisation with BamHI and probe synthesis with T7 RNA polymerase for Aqp1 and Thbs4, plasmid linearisation with NotI and probe synthesis with Sp6 RNA polymerase for Thbs2, and plasmid linearisation with SalI and probe synthesis with T7 RNA polymerase for Htra3.

We thank Sophie Gournet for illustrations and Sébastien Dussurgey for assistance with cell sorting and FACS illustrations (SFR Lyon Biosciences Gerland, UMS3444/US8). We thank Estelle Hirsinger and Claire Fournier-Thibault for reading the manuscript.