The MAPK Erk5 is necessary for proper skeletogenesis involving a Smurf-Smad-Sox9 molecular axis

Extracellular signal-regulated kinase 5 (Erk5; Mapk7 – Mouse Genome Informatics) is a member of the mitogen-activated protein kinase (MAPK) family, which includes Erk1 (Mapk3), Erk2 (Mapk1), c-Jun amino-terminal kinase (Mapk8/9) and p38 (Mapk14) (Robinson and Cobb, 1997; Chang and Karin, 2001; Raman et al., 2007). Erk5 is specifically phosphorylated and activated by MAPK/Erk kinase 5 (Mek5; Map2k5) (English et al., 1995; Zhou et al., 1995), and Erk5 knockout is embryonically lethal in mice owing to multiple developmental defects (Regan et al., 2002; Yan et al., 2003). Erk1 and Erk2 exhibit a high degree of similarity, and they are considered functionally equivalent, although isoform-specific differences have been described (Robinson and Cobb, 1997). Erk5 has a larger molecular mass than other Erks (approximately 2-fold larger than Erk1/2) owing to a unique C-terminal extension that encodes two proline-rich regions and a nuclear localization signal (Buschbeck and Ullrich, 2005; Nishimoto and Nishida, 2006). Mouse genetic studies revealed that the Erk1/2 pathway plays an essential role in skeletal development by regulating successive steps associated with chondrocyte differentiation (Matsushita et al., 2009; Nithianandarajah-Jones et al., 2014; Chen et al., 2015). Moreover, mutations in signaling molecules of the Erk1/2 MAPK pathway also cause a number of human skeletal disorders, including Noonan, Costello and cardiofaciocutaneous syndromes (Bentires-Alj et al., 2006; Rodriguez-Viciana et al., 2006; Rauen, 2013). In contrast to Erk1/2, little is known about the role of Erk5 in skeletogenesis despite evidence for its physiological role in regulating a wide variety of cellular functions associated with vasculogenesis, angiogenesis and neurogenesis in different cell lineages (Regan et al., 2002; Sohn et al., 2002; Hayashi et al., 2004; Li et al., 2013).

Bone is formed by two distinct well-organized processes – intramembranous and endochondral ossification – during embryonic and postnatal skeletogenesis (Karsenty et al., 2009; Long and Ornitz, 2013). The flat bones of the skull, clavicle and mandible are ossified through intramembranous ossification, which involves the direct conversion of mesenchymal stem cells into osteoblasts to form bone (Long, 2012). Other bones, such as the long bones in the forelimbs (humerus, radius and ulna), hindlimbs (femur, tibia and fibula), and feet and hands (metatarsal, metacarpal and proximal phalanges), are formed by endochondral ossification (Johnson and Tabin, 1997). In endochondral ossification, mesenchymal stem cells differentiate into chondrocytes from a cartilaginous template. Chondrocytes differentiate, progressing through the resting, proliferating, hypertrophic and calcifying stages, leading to cartilage matrix mineralization around the central region of the rudiment in the area of hypertrophic chondrocytes. After mineralization, most hypertrophic chondrocytes undergo apoptosis, leading to the invasion of capillaries, osteoclasts, bone marrow cells and osteoblasts into the cartilage matrix to produce new bones (Kronenberg, 2003; Maes, 2013).

Paired-related homeobox 1 (Prx1; Prrx1) is expressed in mesenchymal cells in the limb buds and craniofacial mesenchyme (Cserjesi et al., 1992), and genetic research has revealed a crucial role for Erk1/2 in skeletogenesis through its expression in Prx1⁺-derived cells (Matsushita et al., 2009). In this study, we demonstrated that Erk5 plays a pivotal role in skeletogenesis. In particular, we revealed its role in bone compartments, including the metatarsal, metacarpal and proximal phalanges, through its expression in Prx1⁺-derived cells. Subsequent analyses uncovered that Erk5 directly phosphorylates Smad1 at Ser²⁰⁶ in the linker region, which is known to trigger its proteasomal degradation in a Smad-specific E3 ubiquitin ligase 1 (Smurf1)-dependent manner. In addition, Erk5 directly phosphorylates Smurf2 at residue Thr²⁴⁹, leading to the subsequent acceleration of the proteasomal degradation of Smad proteins (Smad1, Smad2 and Smad3). We demonstrated that Smads transcriptionally activate the expression of the sex-determining region within the Y-type high-mobility group box protein 9 (Sox9), which is the principal transcription factor of skeletogenesis in mesenchymal cells. Moreover, using mouse genetic rescue experiments, we revealed that Sox9 is a prominent mediator of Erk5-dependent skeletogenesis. Our results demonstrated that the Smurf-Smad-Sox9 axis is an important pathway through which Mek5-Erk5 signaling regulates skeletal development.

To reveal the in vivo physiological role of Erk5 in skeletal development, we first generated mesenchymal-specific Erk5 knockout mice using Prx1-Cre (Logan et al., 2002). Staining skeletal preparations from embryos at 18.5 days post-coitum (E18.5) with Alcian Blue and Alizarin Red indicated that the long bones in the forelimbs (humerus, radius and ulna) and hindlimbs (femur, tibia and fibula) were wider in Prx1-Cre;Erk5^fl/fl embryos (Fig. 1Aa-Ad,B), and a larger space was identified in the posterior fontanel (space between the parietal and interparietal bones) (Fig. 1Ae,Af). In addition, mineralization was impaired in metatarsal, metacarpal and proximal phalanges, but not in the femur and tibia, in Prx1-Cre;Erk5^fl/fl embryos at E18.5 (Fig. 1Ag-Aj, Fig. S1A).

Wider long bones (Fig. S1Cc-Cf, Fig. S2A-C) and defects of metatarsal mineralization (Fig. S1B,Ci-Cj) were also observed in Prx1-Cre;Erk5^fl/fl embryos at E16.5. Moreover, wider long bones were also observed in Prx1-Cre;Erk5^fl/fl embryos at E13.5 and E14.5 (Fig. S2A-C). In addition to femur and tibia, the width of the metatarsals was also increased in Prx1-Cre;Erk5^fl/fl embryos at E14.5, E16.5 and E18.5 (Fig. S2A,D).

Histological analyses also revealed a marked increase in long both width (Fig. 1C,D). In addition, almost all chondrocytes in the metatarsals of Prx1-Cre;Erk5^fl/fl embryos expressed collagen type II α1 (Col2a1), a marker of non-hypertrophic chondrocytes, whereas the middle portion of the metatarsals in control embryos did not express this marker (Fig. 1Eg,Eh). Moreover, the level of collagen type X (Col10a1), a hypertrophic chondrocyte marker, was reduced markedly in Prx1-Cre;Erk5^fl/fl embryos relative to control mice (Fig. 1Ei,Ej). These abnormalities in chondrocyte differentiation in the metatarsals were confirmed by Hematoxylin and Eosin (H&E) and Safranin O staining (Fig. 1Ea-Ed). Consistent with delays in chondrocyte hypertrophy and mineralization as determined by von Kossa staining, matrix metalloproteinase-13 (Mmp13) expression level was reduced in Prx1-Cre;Erk5^fl/fl embryos (Fig. 1Ee,Ef,Ek,El).

To examine further the physiological role of Erk5 in skeletal development in vivo, we generated cell-specific Erk5 knockout mice using Col2a1-Cre and osterix (Osx; Sp7)-Cre drivers (Terpstra et al., 2003). Skeletal preparations and histological analyses revealed that Col2a1-Cre;Erk5^fl/fl embryos at E18.5 did not exhibit the abnormalities in space in the posterior fontanel or mineralization of the metatarsal, metacarpal and proximal phalanges noted in Prx1-Cre;Erk5^fl/fl embryos (Fig. S3A). However, these mutant mice recapitulated the wider long bone phenotype in the forelimbs and hindlimbs of Prx1-Cre;Erk5^fl/fl embryos at E18.5 (Fig. S2B,C, Fig. S3A-C). Moreover, wider long bones were also observed in Col2a1-Cre;Erk5^fl/fl embryos at E16.5 but not at E12.5, E13.5 or E14.5 (Fig. S2A-C).

Contrary to observations at embryonic stages, 3-week-old Prx1-Cre;Erk5^fl/fl mice displayed skeletal abnormalities, including deformities of the metatarsal bones in the feet and shorter and thinner long bones, whereas Col2a1-Cre;Erk5^fl/fl mice did not exhibit gross skeletal abnormalities at this age (Fig. S4A-C). Because the Osx-Cre transgene itself caused craniofacial bone development defects (Wang et al., 2015), we were unable to compare embryonic phenotypes in the space of the posterior fontanel between Osx-Cre and Osx-Cre;Erk5^fl/fl embryos. However, Osx-Cre;Erk5^fl/fl embryos did not exhibit abnormalities in metatarsal, metacarpal and proximal phalanx mineralization, nor did they affect the width of long bones in the forelimbs and hindlimbs relative to Osx-Cre embryos (Fig. S4D).

Collectively, among the phenotypic changes in Erk5-deficient mice generated by three different Cre drivers, the production of wider long bones was a common outcome following Erk5 deletion by both Prx1-Cre and Col2a1-Cre drivers, whereas phenotypic changes involving a larger space in the posterior fontanel and impairment of mineralization of metatarsal, metacarpal and proximal phalanges were specific phenotypes related to Erk5 deletion in Prx1⁺ cells.

To reveal whether skeletal abnormalities associated with Erk5 deficiency were derived from abnormalities in the chondrogenic differentiation of mesenchymal cells, micromass cultures of limb bud cells prepared from Erk5^fl/fl and Prx1-Cre;Erk5^fl/fl embryos were differentiated for 6 days and matured for 12 days, then stained with Alcian Blue and Alizarin Red. A marked increase in the size of the Alcian Blue-positive area was observed in Erk5-deficient cells at day 6, whereas the Alizarin Red intensity in the area was markedly decreased in Erk5-deficient cells at day 12 (Fig. 2A, Fig. S5A,B). In addition, the micromass diameter was significantly increased in Erk5-deficient cells at day 6 (Fig. S5C), suggesting that cell proliferation is altered in Erk5-deficient cells. In Erk5-deficient cells, significant increases were observed in the mRNA levels of Col2a1, aggrecan (Acan) and Sox9 (Fig. 2B), whereas Col10a1 and Mmp13 levels were significantly decreased (Fig. 2C). Moreover, Sox9 protein levels were increased in Erk5-deficient cells despite no change in Runx2 and osterix levels (Fig. 2D).

We next investigated whether the Mek5-Erk5 pathway controlled the chondrogenic differentiation of mesenchymal cells in vitro. Alcian Blue-positive area and Alizarin Red intensity were inhibited by retroviral introduction of Erk5(WT) in control and Erk5-deficient cells (Fig. 2Ea-Ed,Eg-Ej). Erk5 phosphorylation was enhanced in cells infected with Erk5(WT) alone (Fig. 2F). In contrast, retroviral introduction of Erk5(DN) – the dominant-negative form of Erk5 that lacks the two phosphorylation sites for Mek5 and cannot phosphorylate target genes – led to an increase in the Alcian Blue-positive area and a decrease in the Alizarin Red intensity in control cells (Fig. 2Ea,Ee,Eg,Ek). This finding was consistent with previous observations in Erk5-deficient cells. No further changes in the differentiation and maturation of chondrocytes were noted when Erk5(DN) was introduced into Erk5-deficient cells (Fig. 2Eb,Ef,Eh,El). Moreover, retroviral introduction of Mek5D, the constitutively active form of Mek5, inhibited the differentiation and maturation of control cells but not Erk5-deficient cells (Fig. 2G).

In addition, CD45⁻CD29⁺Sca1⁺CD105⁺Ptprc⁺Itgb⁺Ly6a⁺Eng⁺ mesenchymal stem cells isolated from limb buds were subject to monolayer culture (2D micromass culture) and chondrogenic pellet culture (3D pellet culture), followed by Alcian Blue staining (Fig. S5D). Alcian Blue staining was enhanced in Erk5-deficient CD45⁻CD29⁺Sca1⁺CD105⁺ mesenchymal stem cells (Fig. S5E), whereas the population of CD45⁻CD29⁺Sca1⁺CD105⁺ cells in the limb buds of Prx1-Cre;Erk5^fl/fl embryos was indistinguishable from that of control embryos (Fig. S5F).

Sox9 protein and mRNA levels were significantly increased in the forelimb buds of Prx1-Cre;Erk5^fl/fl embryos but not Col2a1-Cre;Erk5^fl/fl embryos (Fig. 2H,I, Fig. S5G). Erk5 protein and mRNA levels were significantly diminished in the forelimb buds of Prx1-Cre;Erk5^fl/fl embryos, whereas their levels were significantly but modestly decreased in those of Col2a1-Cre;Erk5^fl/fl embryos (Fig. 2H,J, Fig. S5H). Immunohistochemical analysis also revealed that Sox9 protein expression was markedly higher in the forelimb buds of Prx1-Cre;Erk5^fl/fl embryos at E12.5 and the metatarsals at E16.5, whereas increased Sox9 expression was not detected in Col2a1-Cre;Erk5^fl/fl embryos (Fig. S5I,J).

We next attempted to identify the factors responsible for altering the chondrogenic differentiation of mesenchymal cells and increasing Sox9 expression in Erk5-deficient cells. We monitored the activity of several reporters in Erk5-deficient cells. Among the tested reporters, BRE-luc (Smad1/5/8-dependent) and SBE4-luc (Smad2/3-dependent) displayed significantly increased expression in Erk5-deficient cells compared with controls (Fig. 3A). Consistent with the induction of Sox9 expression in Erk5-deficient cells (Fig. 2D), Sox9 promoter activity (Sox9-luc) and Sox9-dependent transcriptional activity (4x48-luc) were increased significantly in Erk5-deficient cells, whereas no changes were noted in the Runx2 promoter (Runx2-luc) or Runx2-dependent transcriptional activity (6xOSE2-luc) (Fig. 3A). We next attempted to elucidate the mechanism underlying the induction of Smad1/5/8- and Smad2/3-dependent transcriptional activities in Erk5-deficient cells. Smad mRNA levels were comparable between Erk5-deficient and control cells (Fig. 3B), whereas the protein levels of Smad1, Smad2 and Smad3, but not Smad5 or Smad8 (also known as Smad9), were increased in Erk5-deficient cells (Fig. 3C). Moreover, phosphorylation of Smad1/5/8 and Smad2 was increased in Erk5-deficient cells (Fig. 3C), in line with the results of the luciferase assays (Fig. 3A). Consistent with this finding, we observed that Smad1, Smad2 and Smad3 levels were elevated in the nuclei of Erk5-deficient cells along with the induction of Sox9 (Fig. 3D). These results suggest that Erk5 selectively regulates particular Smads at the protein, but not mRNA, level.

We next sought to elucidate whether Erk5 could selectively regulate Smad degradation. Erk5-deficient cells were treated with cycloheximide (CHX), a protein synthesis inhibitor, and Smad1, Smad2 and Smad3 levels were assayed. Smad1, Smad2 and Smad3 rapidly degraded in CHX-treated control cells, whereas their protein levels were more stable in Erk5-deficient cells (Fig. 3E, Fig. S6A-C). We next tested whether the Mek5-Erk5 pathway enhanced Smad1, Smad2 and Smad3 turnover through its ability to promote ubiquitylation. HA-tagged ubiquitin and Flag-tagged Smad1, Smad2 or Smad3 were co-introduced in the presence or absence of Mek5D and Erk5(WT) in HEK293 cells. In the absence of Mek5D and Erk5(WT), no ubiquitination was observed for any of the Smads examined. By contrast, ubiquitin-conjugated Smad products of high molecular weight (exceeding 80 kDa) were observed for Smad1, Smad2 and Smad3 in the presence of Mek5D and Erk5, whereas no high molecular weight products were observed when Mek5D and/or Erk5(WT) were transfected without Smads in HEK293 cells (Fig. 3F, Fig. S6D,E).

These results indicate that the Mek5-Erk5 pathway preferentially targets Smad1, Smad2 and Smad3 for ubiquitylation and proteasome-mediated degradation to regulate Smad1/5/8- and Smad2/3-dependent transcriptional activity.

To investigate the mechanism by which Mek5-Erk5 signaling stimulates the ubiquitylation and proteasome-mediated degradation of Smads, we examined whether Erk5 could phosphorylate and interact directly with Smad1, Smad2 and Smad3. An immunoprecipitation (IP) assay using recombinant proteins revealed that Erk5 directly interacted with Smad1 but not with Smad2 or Smad3 in vitro (IP with anti-Smad antibodies and probed with anti-Erk5 antibody) (Fig. 4A-C). Moreover, an in vitro kinase assay revealed that of the Smad proteins tested only Smad1 was phosphorylated by Erk5 (Fig. 4D-F).

Seven candidate serine or threonine sites in mouse Smad1 were identified from computer alignment with an optimal Erk5 substrate motif (Ser¹¹, Ser¹³², Ser¹⁸⁷, Ser¹⁹⁵, Ser²⁰⁶, Ser²¹⁴ and Ser⁴⁵⁶). To examine which candidate sites could be phosphorylated by Erk5, we generated seven site-directed mutant constructs of Smad1 (S11A, S132A, S187A, S195A, S206A, S214A and S456A), in which the serine was replaced by alanine to prevent phosphorylation. We then performed an in vitro kinase assay. Smad1 phosphorylation by Erk5(WT) was abolished by Smad1(S206A) but not by the other mutants (Fig. 4G). These results indicate that Erk5 directly phosphorylates Smad1 at Ser²⁰⁶, a residue known to be phosphorylated by Erk2, leading to the proteasomal degradation of Smad1 in a Smurf1-dependent manner (Sapkota et al., 2007).

To investigate alternative mechanisms that could be involved in the ubiquitylation and proteasome-mediated degradation of Smads induced by the Mek5-Erk5 axis, we next investigated whether Erk5 could interact directly with Smurf2, another ubiquitin E3 ligase responsible for the ubiquitin-dependent degradation of Smads (Lin et al., 2000; Zhang et al., 2001). An IP assay revealed that Erk5 interacted with Smurf2 in HEK293 cells (IP with anti-Flag antibody and probed with anti-HA antibody), in vitro (cell-free system; IP with anti-Smurf2 antibody and probed with anti-Erk5 antibody) and in primary mesenchymal cells (IP with anti-Erk5 antibody and probed with anti-Smurf2 antibody) (Fig. 5A-C), indicating that Erk5 physically interacts with Smurf2.

Among the four candidate serine or threonine sites conserved in different species based on computer alignment to an optimal Erk5 substrate motif, one potential Erk5 phosphorylation site emerged in mouse Smurf2 at Thr²⁴⁹ (Fig. 5D, Fig. S7A). In addition, substrate sequence motifs indicated that Erk1/2 was unlikely to phosphorylate Smurf2 at Thr²⁴⁹ (Fig. S7A). To establish the role of this potential phosphorylation site in Smad degradation, we generated a site-directed mutant construct of Smurf2, Smurf2(T249A), in which the threonine was replaced by alanine to prevent phosphorylation. An in vitro kinase assay revealed that bacterially expressed recombinant Smurf2(WT) was phosphorylated by Erk5(active) but not by Erk5(inactive), whereas Smurf2 phosphorylation by Erk5 was stopped completely by Smurf2(T249A) (Fig. 5E), indicating that Erk5 directly phosphorylated Smurf2 at Thr²⁴⁹.

To investigate further the role of Smurf2 phosphorylation at Thr²⁴⁹ in Smad degradation, an in vitro ubiquitylation assay was performed with recombinant Smurf2(WT), Smurf2(T249A) and Smurf2(T249E), in which threonine was replaced with glutamate to mimic the phosphorylated state, as well as Erk5(active) and Smads. Incubation of Smurf2(WT) with ubiquitin, E1 and UbcH5c caused Smad protein ubiquitylation in the presence of Erk5 despite the weak ubiquitylation signal noted in the absence of Erk5. The reaction with Smurf2(T249A) produced a faint ubiquitylation signal in the presence of Smad1, Smad2 and Smad3, irrespective of the presence of Erk5, whereas extensive ubiquitylation was observed when Smads were incubated with Smurf2(T249E) in the absence of Erk5. The addition of Erk5 did not increase ubiquitylation (Fig. 5F, Fig. S7B). Moreover, in the absence of Smad proteins, a faint ubiquitylation signal above 100 kDa was detected after adding Smurf2, irrespective of the presence or absence of Erk5, indicating ubiquitination of Smurf2 (Fig. S7C).

Finally, we determined whether Smurf2 phosphorylation at Thr²⁴⁹ was required for Erk5-dependent chondrogenic differentiation. Retroviral infection of Smurf2(WT) and Smurf2(T249E) decreased chondrocyte differentiation in control and Erk5-deficient cells (Fig. 5Ga,Gb,Gd-Gf,Gh), whereas Smurf2(T249A) increased the differentiation of control cells but not Erk5-deficient cells (Fig. 5Ga,Gc,Ge,Gg). Consistent with these results, but in contrast to the increase in activity following Smurf2(T249A) transfection, BRE-luc activity was lower in cells transfected with Smurf2(WT), with further decreases induced by Smurf2(T249E) transfection (Fig. 5H).

Collectively, these results suggest that the Mek5-Erk5 pathway regulates the stability of Smads by enhancing the ubiquitin E3 ligase activity of Smurf2 through direct phosphorylation at Thr²⁴⁹ and by modulating the Smurf1-dependent proteasomal degradation of Smad1 through direct phosphorylation at Ser²⁰⁶.

Given that Erk5 deficiency induces Sox9 mRNA transcription and Erk5 activates Smad proteasomal degradation by Smurf2, we tested whether Smads could activate Sox9 expression at the transcriptional level in mesenchymal cells. Introduction of Smad1 and Smad4 (Smad1/Smad4), Smad2 and Smad4 (Smad2/Smad4), and Smad3 and Smad4 (Smad3/Smad4) induced Sox9 protein (Fig. 6A) and mRNA expression (Fig. 6B). We computationally analyzed the 5′-flanking region of mouse Sox9 and human SOX9 and identified at least one putative bone morphogenic protein responsive element (BRE) and one putative Smad-binding element (SBE) in the 5′-flanking region, which were highly conserved between mouse and human SOX9 (Fig. 6C). A reporter assay performed using primary mesenchymal cells revealed that introduction of Smad1/Smad4, Smad2/Smad4 and Smad3/Smad4 significantly activated Sox9 promoter activity (Fig. 6D). Moreover, a chromatin IP assay verified that Smad1 recruitment to the Sox9 promoter regions encompassing the BRE was enhanced in Erk5-deficient cells compared with control cells without a significant change in Smad2 and Smad3 recruitment (Fig. 6E). By contrast, recruitment of Smad2 and Smad3, but not Smad1, to the Sox9 promoter regions encompassing the SBE was significantly increased in Erk5-deficient cells (Fig. 6F).

Our results suggest that Erk5 deficiency influences skeletogenesis and chondrogenesis through the activation of Sox9 expression in vitro and in vivo. To investigate this genetically in vivo, we generated Erk5^−/− embryos lacking one allele at Sox9 in mesenchymal cells. Prx1-Cre;Erk5^fl/fl;Sox9^fl/+ embryos and mice displayed partial normalization of abnormalities of the metatarsal and proximal phalanges at E18.5 (Fig. 7Am-Ap, Fig. S8A) and bone formation deformities in the feet at 3 weeks old (Fig. 7Aq-At). Moreover, Sox9 heterodeficiency rescued the larger posterior fontanel identified in Prx1-Cre;Erk5^fl/fl embryos (Fig. 7Ai-Al), whereas the width of the long bones in the forelimbs and hindlimbs was comparable between Prx1-Cre;Erk5^fl/fl and Prx1-Cre;Erk5^fl/fl;Sox9^fl/+ embryos (Fig. 7Ae-Ah). Moreover, abnormalities of differentiation and maturation in Erk5-deficient cells, defined as Alcian Blue-positive areas (Fig. 7Ba-Bd,C) and Alizarin Red intensity (Fig. 7Be-Bh,D), respectively, were rescued by Sox9 heterodeficiency in mesenchymal cells in vitro. This result is correlated with the reduction of Sox9 protein levels in Erk5^−/−;Sox9^+/− cells (Fig. S8B). Taken together, these results provide genetic evidence that Erk5 regulates skeletogenesis in part through its ability to inhibit Sox9 expression and function in mesenchymal cells.

Previous in vitro studies revealed the role of Erk5 as a negative regulator of chondrocyte differentiation (Bobick et al., 2010; Wang et al., 2016). However, Erk5 knockout in mice is embryonically lethal at E9.5-10.5 (Regan et al., 2002; Yan et al., 2003); therefore, the in vivo physiological role of Erk5 in skeletal development has remained largely unclear. The present findings reveal that Erk5 deficiency in mesenchymal cells leads to abnormal skeletogenesis in vivo. Moreover, several lines of evidence indicate that Mek5-Erk5 signaling controls chondrogenic differentiation by modulating the protein stability of Smads (Smad1, Smad2 and Smad3), which activate Sox9 expression at the transcriptional level through at least two pathways. First, Erk5 directly phosphorylates Smad1 at Ser²⁰⁶ to trigger its proteasomal degradation in a Smurf1-dependent manner. Second, Erk5 directly phosphorylates Smurf2 at Thr²⁴⁹ to activate its ubiquitin E3 ligase ability to degrade Smad proteins (Smad1, Smad2 and Smad3). Although we should emphasize that this cascade is not activated in all skeletal elements and we have not linked their mechanistic in vitro studies to the phenotypic alternations by examining the relevant players in vivo, to our knowledge, this is the first direct demonstration of the role for the Mek5-Erk5 pathway in skeletogenesis in vivo in modulating chondrogenic differentiation through the Smurf-Smad-Sox9 cascade (Fig. 7E).

The possibility should be mentioned that Erk5 controls chondrogenic differentiation through its expression in perichondrium cells, which are targeted by Prx1-Cre and Col2a1-Cre (Ono et al., 2014). However, Osx-Cre;Erk5^fl/fl embryos did not exhibit any abnormalities in skeletal phenotypes (Fig. S4) despite the activity of Osx-Cre in the perichondrium (Rodda and McMahon, 2006). Thus, although further studies are needed, these results suggest that Erk5 does not directly control growth plate cartilage development or endochondral ossification through its expression in the perichondrium.

It should also be noted that the data from two different phenotypes of Prx1-Cre;Erk5^fl/fl embryos (wider bones and mineralization defect/delay) should be interpreted as two different developmental processes. The mineralization area was markedly decreased in the metatarsals, but not in tibia and femur, in Prx1-Cre;Erk5^fl/fl embryos at E16.5 and E18.5, suggesting the mineralization defect or delay is specific to the development of distal skeletal elements, although it cannot be ruled out that a transient delay in hypertrophic differentiation/mineralization may occur in the proximal skeletal elements at an earlier developmental time point. Many genetic diseases with brachydactylies, which are inheritable conditions typified by short fingers and toes, has been reported to be related to malfunction of BMP/TGFβ and hedgehog signaling that affect distal chondrogenic events with delayed mineralization of digit cartilage elements (Gao et al., 2001, 2009; Lehmann et al., 2003). To identify the processes or molecular signals involved in the distal preference of mineralization defects or delays in Prx1-Cre;Erk5^fl/fl embryos, we should test whether the Mek5-Erk5 pathway is a downstream or upstream effector of BMP/TGFβ or hedgehog signaling in mesenchymal cells. However, the possibility of a more general effect of Erk5 on fibroblast differentiation or proliferation, with chondrogenesis being a downstream consequence, should be considered, although the fact that Erk5-deficient CD45⁻CD29⁺Sca1⁺CD105⁺ mesenchymal stem cells exhibited higher chondrogenic differentiation capacity could indicate a direct role of Erk5 in chondrogenic differentiation through its expression in mesenchymal progenitor cells.

We also demonstrated that Thr²⁴⁹ is required for the E3 ubiquitin ligase activity of Smurf2. Replacing this residue with alanine (Smurf2-T249A) attenuated the ability of Smurf2 to degrade Smad proteins, whereas replacement with glutamate (Smurf2-T249E) led to greater degradation (Fig. 5, Fig. S7). Although Akt phosphorylates Smurf2 at unknown residues to induce Smurf2 degradation (Choi et al., 2014), our study clearly identified the particular phosphorylation site of Smurf2 (Thr²⁴⁹) and its target proteins (Smads) for regulating cellular function (chondrogenic differentiation) and uncovered the E3 ubiquitin ligase activity. Among the large number of target proteins of Smurf2 (Lin et al., 2000; Zhang et al., 2001; David et al., 2013), we focused on Smad1, Smad2 and Smad3, because Erk5 deficiency led to their upregulation at the protein but not mRNA level (Fig. 3). Smurf2 is known to directly and indirectly induce the ubiquitylation and degradation of multiple target proteins implicated in chondrogenesis, including the type I receptor of TGFβ, which is upstream of Smad2 and Smad3, as well as glycogen synthase kinase 3β and β-catenin (Kavsak et al., 2000; Han et al., 2006; Wu et al., 2009). Erk5 positively affects TGFβ-dependent Smad3 transcriptional activity via acetylation (Kim et al., 2013). We therefore cannot exclude the possibility that another pathway could also contribute to the regulation of Erk5-dependent skeletogenesis.

Moreover, for the first time, we revealed that Smads (Smad1/Smad4, Smad2/Smad4 and Smad3/Smad4) directly activated Sox9 expression at the transcriptional level in mesenchymal cells. As it has been demonstrated that Smad2 and Smad3 physically interact with Sox9, leading to the enhancement of its transcriptional activity (Furumatsu et al., 2005; Song et al., 2009), it is possible that Smads could regulate chondrogenic differentiation following Erk5-dependent induction by binding to Sox9 to enhance its transcriptional activity. This would induce the expression of chondrocyte-specific genes, including Col2a1 and Acan, carrying Sox9-binding sites in their promoters (Lefrebvre and de Crombrugghe, 1998; Akiyama et al., 2002), in addition to directly inducing Sox9 transcription in mesenchymal cells. Alternatively, Erk5 might control Sox9 levels through Smad-independent mechanisms, for example, through the control of Sox9 RNA stability.

In conclusion, the Mek5-Erk5 pathway plays a crucial role in skeletogenesis in vivo by modulating the protein stability of Smads (Smad1, Smad2 and Smad3) via Smurfs (Smurf1 and Smurf2) through Smad expression in mesenchymal cells (Fig. 7E). Our findings improve our understanding of the molecular mechanisms underlying skeletal development, and they may lead to the development of novel drugs targeting cartilage diseases associated with abnormal chondrocyte differentiation and maturation in humans. Indeed, the Smurf-Smad cascade is associated with various human diseases, including cancer and aging (Fukuchi et al., 2002; Zhang and Cohen, 2004; Jin et al., 2009). Therefore, components of the Mek5-Erk5-Smurf-Smad-Sox9 cascade could represent novel targets for drug development to treat various human diseases.

Erk5^fl/fl (Newbern et al., 2011) and Sox9^fl/fl (Akiyama et al., 2002) mice were crossed with either Prx1-Cre (Logan et al., 2002), Col2a1-Cre (Terpstra et al., 2003) or Osx-Cre (Takarada et al., 2016). These mutant mice were backcrossed for more than five generations with C57BL/6J. Mice were bred under standard animal housing conditions at 23±1°C with humidity of 55% and a 12 h light/dark cycle, with free access to food and water. Genotyping was performed by PCR using tail genomic DNA. The study protocol meets the guidelines of the Japanese Pharmacological Society and was approved by the Committee for Ethical Use of Experimental Animals at Kanazawa University, Kanazawa, Japan. The numbers of animals used per experiment are stated in the figure legends.

Embryos were eviscerated and the skins were removed, then embryos were fixed in 95% ethanol overnight. Embryos were then immersed in Alcian Blue solution overnight, and were transferred in a solution of 2% KOH overnight, followed by staining in Alizarin Red solution overnight. Finally, the skeletons were cleaned with 1% KOH/20% glycerol for 1 week and then stored in 50% ethanol/50% glycerol (Takahata et al., 2012). For micro computed tomography (μCT) 3D images of the foot were obtained using a VivaCT 40 mCT system (Scanco Medical) as described previously (Kajimura et al., 2011).

Mouse tibia, femurs and feet were fixed with 10% formalin neutral buffer solution overnight, embedded in paraffin, and then sectioned at a thickness of 5 μm. Sections were stained with H&E, Safranin O and von Kossa.

In situ hybridization was performed as previously described (Wang et al., 2005). In brief, the metatarsals were dissected in a cryostat to yield frozen sections with a thickness of 10 μm. Sections mounted on glass slides were fixed with 4% paraformaldehyde and successively treated with 0.2 M HCl for 10 min and 10 μg/ml proteinase K for 5 min. Sections were then subjected to acetylation in 0.1 M triethanolamine with 0.25% acetic anhydride for 10 min. After prehybridization, sections were covered with digoxigenin-labeled cRNA probes at 65°C for 16 h. Sections were then washed and treated with 4 μg/ml RNase A for 30 min. Sections were further incubated with anti-digoxigenin-ALP Fab fragments for 16 h. After washing, sections were treated with nitro blue tetrazolium chloride/5-bromo-4-chloro-3-indolyl phosphate for different periods. Plasmid vectors for generating cRNA probes were kindly provided by Dr Nishimura (Osaka University, Osaka, Japan) and Dr Ducy (Columbia University, NY, USA).

For immunohistochemistry, 5-μm-thick sections were deparaffinized in xylene and rehydrated in ethanol. Antigen retrieval was performed in citrate buffer (pH 6.0). After treatment with blocking buffer (PBS containing 0.5% Triton X-100 and 5% normal goat serum) for 1 h, the sections were incubated with anti-Sox9 antibody (EMD Millipore) in blocking buffer for 24 h. The sections were further incubated with anti-rabbit IgG (Alexa Fluor 488) in blocking buffer for 2 h.

For whole-mount immunohistochemistry, limb buds were fixed with methanol:dimethyl sulfoxide (4:1) overnight, and were washed three times with 1% Triton X-100/PBS. After treatment with blocking buffer (PBS containing 1% Triton X-100 and 5% normal goat serum) for 1 h, limb buds were incubated with anti-Sox9 antibody in blocking buffer for 24 h, followed by incubation with anti-rabbit IgG (Alexa Fluor 488) in blocking buffer for 48 h.

To generate the mutant Smurf2 constructs [pMX-Smurf2(T249A) and pMX-Smurf2(T249E)], site-directed mutagenesis was performed using the PrimeSTAR Max Mutagenesis Basal Kit (Takara Bio) with specific primers (Table S1). pMX-Erk5(WT) and pMX-Erk5(DN) vectors were generated by subcloning into pMX vector from pcDNA3-Erk5(WT) and pcDNA3-Erk5AEF vectors, respectively, which were generously provided by Dr Jiing-Dwan Lee (Scripps Research Institute, CA, USA). These vectors were then transfected into PLAT-E cells using the calcium carbonate method. Virus supernatant was collected 48 h after transfection and cells were then infected with this viral supernatant for 8 h in the presence of 4 μg/ml polybrene.

Whole limb bud mesenchymal cells from E12.5 were treated with 0.1% collagenase and 0.1% dispase for 1.5 h at 37°C, and subsequently re-suspended in DMEM/F12 medium containing 10% fetal bovine serum (FBS) at 1.5×10⁷ cells/ml. Droplets were carefully placed in each well of a 4-well plate. Cells were allowed to adhere at 37°C for 1.5 h, followed by the addition of chondrogenic medium (DMEM/F12 medium containing 10% FBS and 50 μg/ml ascorbic acid). Medium was changed every 2 days, and Alcian Blue and Alizarin Red staining was performed on day 6 and day 12, respectively.

Limb bud mesenchymal cells from E12.5 were digested and re-suspended in PBS containing 2% FBS, followed by incubation for 30 min at 4°C with FITC anti-CD45 antibody (1:40, 30-F11, BD Pharmingen), APC anti-CD105 antibody (1:40, MJ7/18, eBioscience), PE anti-CD29 antibody (1:40, HMβ1-1, BioLegend), PE-Cy7 anti-Ly6A/E antibody (1:40, D7, BD Pharmingen) and DAPI for dead cell exclusion. Immunostained cells were analyzed on FACS AriaII cell sorter (BD Biosciences), and subsequently FITC-negative and APC-, PE-, PE-Cy7-positive cells were sorted, followed by re-suspension in DMEM/F12 medium containing 10% FBS at 1.5×10⁷ cells/ml. Cells were allowed to adhere at 37°C for 1.5 h, followed by the addition of chondrogenic medium (DMEM/F12 medium containing 10% FBS, 50 μg/ml ascorbic acid, 0.1 M dexamethasone, 1 mM sodium pyruvate and 20 ng/ml TGFβ).

Total RNA was extracted from cells, followed by the synthesis of cDNA using reverse transcriptase and oligo-dT primer. The cDNA samples were then used as templates for real-time PCR analysis, which was performed on an MX3005P qPCR System (Agilent Technologies) using specific primers for each gene (Table S2). The levels of the genes examined were normalized using Actb as an internal control for each sample.

Cells were solubilized in lysis buffer containing 1% Nonidet P-40. Samples were then subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), followed by transfer to polyvinylidine difluoride membranes (PVDF) and subsequent immunoblotting. The primary antibodies used were: anti-Runx2 (1:1000, #8486), anti-Erk5 (1:1000, #3372), anti-Smad1 (1:1000, #9743), anti-Smad2 (1:1000, #5339), anti-Smad3 (1:1000, #9523), anti-phospho-Smad1/5/8 (1:1000, #13820), anti-phospho-Smad2/3 (1:1000, #8828) and anti-phospho-Erk5 (1:1000, #3371) (all from Cell Signaling Technologies); anti-β-actin (1:2000, C4), anti-Smad5 (1:1000, D-20), anti-Smad8 (1:1000, R-64) and anti-Smurf2 (1:1000, H-50) (all from Santa Cruz Biotechnology); anti-Sox9 (1:2000, #AB5535, EMD Millipore); and anti-Osx (1:1000, #ab94744, Abcam).

Plasmids BRE-luc (#45126) (artificial sequences) and SBE4-luc (#16495) (conserved sequences between human and mouse) were obtained from Addgene, and plasmids Runx2-luc and 6xOSE2-luc were kindly provided by Dr Ducy (Columbia University, NY, USA). Plasmid 4x48-luc was generously provided by Dr Crombrugghe (University of Texas, M.D. Anderson Cancer Center, TX, USA). For the luciferase assay, cells were transfected with reporter vectors using the lipofection method as previously described (Hinoi et al., 2012), followed by the preparation of cell lysates and subsequent determination of luciferase activity using specific substrates in a luminometer (ATTO). Transfection efficiency was normalized by determining lacZ activity by using an ONPG assay.

ChIP experiments were performed following the protocol provided with the ChIP assay kit (EMD Millipore). Cells were treated with formaldehyde to induce cross-linking and were subsequently subjected to sonication in lysis buffer. IP was performed with the anti-Smad1, anti-Smad2 or anti-Smad3 antibodies used at 1:50 dilution, followed by PCR with specific primers (Table S3).

Recombinant GST-tagged human Smad1, Smad2 and Smad3 proteins and recombinant full-length Erk5 active were purchased from Sigma-Aldrich, and recombinant Erk5 inactive was obtained from Carna Biosciences. The mutant Smad1 constructs were generated with site-directed mutagenesis using the PrimeSTAR Max Mutagenesis Basal Kit with specific primers (Table S4). Human SMAD1 (WT and mutants), and SMURF2 (WT and mutants) cDNA were inserted into pCold II vector (Takara Bio) with N-terminal GST-tag. GST-tagged vectors were expressed in Escherichia coli strain BL21 cultured with 50 μM isopropyl-β-D-thiogalactopyranoside at 15°C for 24 h. The bacterial cells were centrifuged at 3000 g and the supernatant removed. The bacterial pellets were subjected to freeze/thaw treatment three times and then incubated in lysis buffer (50 mM Tris-HCl, 500 mM NaCl, 10% glycerol, 1% Triton X-100, pH 8.0 at 4°C) containing 1 mg/ml lysozyme for 30 min, followed by sonication. GST-tagged proteins in the supernatant were purified by a column packed with Glutathione Sepharose 4 Fast Flow (GE Healthcare). The GST-tagged proteins were eluted with the buffer (100 mM Tris-HCl, 20 mM glutathione, pH 8.0 at 4°C). The eluted proteins were dialyzed for 24 h at 4°C against repeated changes of dialysis buffer (10 mM Tris-HCl, pH 7.5).

Plasmids Flag-Smurf2 (#11746, deposited by Jeff Wrana) and HA-Erk5 (#31817, deposited by Astar Winoto) were obtained from Addgene. HEK293 cells were transiently transfected with expression vectors using the calcium phosphate transfection method, and cells were then solubilized in lysis buffer containing 1% Nonidet P-40, followed by incubation with an antibody for 1 h at 4°C and subsequent IP with protein G-Sepharose. Recombinant proteins were incubated with an antibody in binding buffer containing 0.5% Nonidet P-40 for 1 h at 4°C and subsequent IP with protein G-Sepharose. Immunoprecipitates were washed three times with lysis buffer or binding buffer and boiled in SDS sample buffer. Samples were then separated by SDS-PAGE, followed by transfer to PVDF membranes and subsequent immunoblotting, as previously described (Nakamura et al., 2013).

Recombinant Erk5 active or recombinant Erk5 inactive (0.3 μM) were incubated with GST-Smad1(WT and mutants), GST-Smad2(WT), GST-Smad3(WT) or GST-Smurf2(WT and mutants) (1 μM) and 0.5 mM ATP in 30 μl kinase buffer (2.5 mM MOPS, pH 7.2, 2.5 mM MgCl₂, 1.25 mM glycerol 2-phosphate, 0.25 mM DTT, 0.05% bovine serum albumin) for 30 min at 30°C. The reaction was stopped by adding SDS sample buffer. Phosphorylated proteins were analyzed by Phos-tag PAGE. For Phos-tag immunoblotting, 0.05 mM Phos-tag (Wako) and 0.1 mM MnCl₂ were added to conventional SDS polyacrylamide gels according to the manufacturer's protocol.

In vitro ubiquitylation assay was performed with the E2-Ubiquitin Conjugation Kit (Abcam) following the protocol recommended by the manufacturer. GST-Smad1, GST-Smad2 or GST-Smad3 (1 μM) were incubated with GST-Smurf2(WT and mutants) (0.1 μM) and GST-Erk5 active (0.05 μM) in 1×Ubiquitylation Buffer [0.1 μM E1, 0.5 μM UbcH5c (E2), 5 mM Mg-ATP, 1 mM DTT, 2.5 μM biotinylated ubiquitin] for 1 h at 37°C. The reaction was stopped by adding 2×Non-reducing gel loading buffer. Ubiquitylated proteins were analyzed by SDS-PAGE and horseradish peroxidase-conjugated streptavidin detection system.

For immunocytochemical analysis, cells were plated on a chamber slide coated with poly-l-lysine. Cultured cells were fixed with 4% paraformaldehyde in PBS, followed by washing and subsequent treatment with normal goat serum in PBS containing 0.1% Triton X-100. Cells were then incubated with a primary antibody diluted in PBS containing normal goat serum and 0.3% Triton X-100. Finally, cells were reacted with the corresponding secondary antibody and chamber slides were mounted using FluorSave (EMD Millipore).

All results are expressed as the mean±s.e.m., and statistical significance was determined using two-tailed, unpaired Student's t-test or one-way analysis of variance with Bonferroni post-hoc test.

We wish to thank Dr W. D. Snider (University of North Carolina) and Dr T. Kitamura (Tokyo University) for kindly providing Erk5^fl/fl mice and PLAT-E cells, respectively.