The mevalonate pathway is a crucial regulator of tendon cell specification

Tendons and ligaments are key components of the musculoskeletal system, transmitting and stabilizing high tensile forces. Exposed to such forces, adult tendons are often injured during overuse, trauma or age-related decline (Jozsa and Kannus, 1997). Knowledge of tendon fate determination has important implications for developing regenerative approaches to treating tendon injuries. Although the basic helix-loop-helix transcription factor scleraxis (Scx) provides a marker of early tendon and ligament progenitors (Schweitzer et al., 2001), the pathways leading to tendon cell fate induction remain largely unknown.

Scx marks tendon and ligament cells in all anatomical locations (Brent et al., 2003; Grenier et al., 2009; Schweitzer et al., 2001), and is required for the development of the limb force-transmitting tendons (Murchison et al., 2007). Scx also regulates the expression of extracellular matrix proteins – including Col1a1 – through its proximal promoter (Léjard et al., 2007). The identification of Scx as a marker of early tendon progenitors opened the door to studies of factors involved in tendon specification. TGFβ signals are sufficient to induce Scx expression in mesenchymal tissues (Havis et al., 2014; Pryce et al., 2009). However, mouse genetic loss-of-function studies indicate a role for this pathway in tendon cell maintenance, rather than induction (Oka et al., 2008; Pryce et al., 2009; Tan et al., 2020). An important role for FGF signaling was identified in the specification of axial tendon progenitors in the chick (Brent et al., 2003; Brent and Tabin, 2004), yet the tendon-promoting effect of FGF signaling is not conserved in mouse, as highlighted in studies showing that early FGF/ERK signaling represses Scx expression in mouse (Havis et al., 2014, 2016).

Although the crucial pathways responsible for tendon cell specification remain unknown, studies examining neighboring tissues provide insights into the interactions required for their specification. In the craniofacial, axial and limb regions, tendon and ligament cells form in close association with skeletal cells and are thought to arise from common progenitors (Soeda et al., 2010). The cranial skeletal and tendon progenitors are derived from the cranial neural crest (CNC) (Chen and Galloway, 2014; Couly et al., 1993; Kontges and Lumsden, 1996; Le Douarin and Kalcheim, 1999; Le Lievre, 1978; Lumsden et al., 1991; Noden, 1978; Schilling and Kimmel, 1994); in the limb, they are derived from the lateral plate mesoderm (Hurle et al., 1989, 1990; Kieny and Chevallier, 1979; Pearse et al., 2007; Ros et al., 1995; Saunders, 1948; Wortham, 1948); and in axial regions, tendon progenitors, which form the syndetome, are derived from the early sclerotome, a somitic compartment (Brand-Saberi and Christ, 2000; Brent et al., 2003; Monsoro-Burq and Le Douarin, 2000). Sox9 is expressed in chondrogenic cells, and common Sox9/Scx co-expression domains are found at tendon-bone attachment sites (Blitz et al., 2013; Sugimoto et al., 2013). Furthermore, loss of transcription factors important for cartilage differentiation results in ectopic expression of tendon markers within the axial rib cartilage structures (Brent et al., 2005). Although this effect was not observed in cranial or limb regions, mammalian in vitro studies suggest that tendon and cartilage are mutually exclusive lineages of common progenitor populations (Havis et al., 2014).

To discover new pathways that govern tendon cell induction, we turned to the zebrafish, a system that is highly amenable to high-throughput small molecule screening (Mathias et al., 2012). Previous zebrafish chemical screens have expanded our understanding of developmental processes and identified compounds of potential therapeutic benefit (Goessling and North, 2014; North et al., 2007), highlighting the conservation between zebrafish and humans. Zebrafish tendons share many similarities with those in mammals, including gene expression, morphology, collagen ultrastructural arrangement and developmental regulation (Chen and Galloway, 2014; Subramanian and Schilling, 2014). We therefore reasoned that chemical screens of biologically annotated bioactives in zebrafish could reveal new pathways regulating tendon development.

In a small-molecule whole-embryo screen, we identified statins as regulators of early tendon specification. Statins are a class of 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) inhibitors that are prescribed as cholesterol-lowering agents (Istvan and Deisenhofer, 2001). In addition to preventing cardiovascular disease, statins and HMGCR pathway inhibition regulate cell proliferation in tumors (Jones et al., 1994) and cell migration during heart and germ cell development (D'Amico et al., 2007; Santos and Lehmann, 2004; Thorpe et al., 2004; Yi et al., 2006). Using chemical and genetic approaches, we show that Hmgcr inhibition expands the number of craniofacial and pectoral fin tendon progenitors in the zebrafish embryo. The expansion is not a consequence of increased proliferation of a pre-existing scleraxisa-positive (scxa⁺) cell population, but rather is caused by recruitment of additional neural crest sox9a⁺/sox10⁺ skeletal progenitors towards tendon fates. We further demonstrate through loss-of-function studies that the interaction is mediated by inhibition of the geranylgeranylation branch of the mevalonate pathway, and specifically Rac GTPases. Taken together, we establish that the mevalonate pathway via Rac activity is a crucial regulator of tendon fate specification.

To identify new pathways regulating tendon development in zebrafish, we conducted a chemical screen using a known bioactive compound library: NINDS Custom Collection 2. Wild-type embryos were incubated with individual chemicals from 32 to 56 h post-fertilization (hpf) and examined at 56 hpf by whole-mount in situ hybridization for altered craniofacial expression of scxa (a scx paralogue in zebrafish) (Fig. 1A). These stages were chosen to identify pathways affecting tendon cell induction, as scxa is first detected in the zebrafish pharyngeal arches by 40 hpf (Chen and Galloway, 2014) (Fig. S1). Our screen identified lovastatin and simvastatin (Fig. S2A-D) as agents that increase craniofacial scxa expression. Additional statins, atorvastatin and fluvastatin, similarly expand craniofacial scxa expression (Fig. S2E-L). All statin compounds used in this study inhibit HMGCR and have some differences in their metabolism and pharmacokinetic properties (Klotz, 2003). Atorvastatin caused the most robust phenotype in the cranial and fin regions (Fig. 1B,C,F,G), and most analysis was performed with this compound. Atorvastatin treatment from 48 to 72 hpf resulted in a similar scxa expansion at 72 hpf when assessed by whole-mount in situ hybridization (Fig. 1D,E,H,I). Analysis of fluvastatin, which has the shortest half-life among statins (Igel et al., 2001), demonstrated the 32-48 hpf treatment window was sufficient to expand craniofacial scxa expression (Fig. S2M-R). The statin-mediated upregulation of scxa was further confirmed by qPCR at both stages (Fig. 1J). Quantification of scxa⁺ cells using whole-mount in situ hybridization or Tg(scxa:mcherry) zebrafish (Fig. S3I-Q) verified that statin treatment increases the number of scxa⁺ cells in the craniofacial (Fig. 1L) and pectoral fin (Fig. 1M, Fig. S3B) regions compared with controls. At the stages examined, the Tg(scxa:mcherry) line was found to recapitulate endogenous scxa expression (Fig. S1). Although early (32-56 hpf) and later (48-72 hpf) treatments resulted in expanded scxa expression, it is possible that the cellular mechanism underlying expansion is not identical between the stages. Because the early treatment was better tolerated by embryos and resulted in less toxicity, we focused on dissecting the effects of statin treatment in the 32-56 hpf stages.

To test whether statin expands expression of additional tendon genes, as opposed to simply upregulating scxa expression, we examined expression of col1a2, which is expressed during tendon differentiation and is a major component of the tendon extracellular matrix (Brent et al., 2003; Chen and Galloway, 2014). We found that embryos incubated with atorvastatin also have expanded col1a2 expression (Fig. S2W-X). Likewise, statin causes >2-fold increase in the number of craniofacial col1a2⁺ tendon progenitors compared with controls (Fig. 1N) and increases col1a2 expression by qPCR (Fig. 1K). As forming bone is rich in type I collagen, we examined expression of runx2a, a marker of osteoblasts (Burns et al., 2002), following statin treatment. We found that runx2a expression was not qualitatively expanded at 56 hpf and was decreased at 72 hpf (Fig. S2S-V), suggesting that the increase in col1a2 expression is due to expanded tendon fates. Besides the head and fin, scxa⁺ tendon progenitors in zebrafish exist in the myosepta, connecting the muscles of the trunk. In contrast to the pectoral fin and craniofacial region, atorvastatin does not affect the quantity of scxa⁺ (Fig. S2Y) or scxa:mcherry⁺ (Fig. S2Y-C′) cells in the myosepta. Furthermore, atorvastatin treatment from 32-56 hpf reduced myoseptal col1a2 expression at 74 hpf (Fig. S2D′-E′). Although the differences in response to statin treatment in the different anatomical locations may be derived from tissue-specific differences in drug efficaciousness, we hypothesize the difference is more reflective of region-specific programs of tendon development (Brent et al., 2003; Chen and Galloway, 2014).

In principle, the increased quantity of tendon progenitors could arise from increased proliferation of scxa⁺ cells. To determine whether statin increases tendon cell proliferation, we performed EdU labeling (Salic and Mitchison, 2008) and phospho-histone H3 (PH3) staining. EdU pulse-labeling at 4 or 5 h intervals of chemically treated embryos transgenic for a neural crest and cranial cartilage reporter, sox10:mcherry (Fig. 2A-B′), indicated that atorvastatin does not increase the number of proliferative cells in anatomically defined regions compared with corresponding regions in controls (Fig. 2C). Rather, we observed a trend of decreased proliferation following atorvastatin treatment at 43 hpf, 52 hpf and 56 hpf, suggesting that the mevalonate pathway regulates proliferation in the craniofacial region. The mevalonate pathway negatively affects proliferation in other cell types (Corsini et al., 1993; Gong et al., 2019). Given that subtle changes in proliferation within the scxa⁺ domain may not be detected with this analysis, we performed two sets of experiments to address this concern. First, we quantified the number of PH3⁺ mitotic cells in atorvastatin-treated and control embryos. Consistent with our EdU findings, we determined that atorvastatin does not promote proliferation in the craniofacial or fin regions (Fig. S3A,C). More importantly, we did not detect any increase in the ratio of PH3⁺/scxa⁺ cells in either region (Fig. 2D,E,H and Fig. S3D,I,L-Q). Second, we hypothesized that if the statin-mediated expansion were acting through increased cell proliferation, an antimitotic agent would abrogate its affect. We analyzed the effect of aphidicolin, a DNA polymerase inhibitor (Spadari et al., 1984), on proliferation and craniofacial scxa expression with and without statin treatment. The concentration of aphidicolin used reduces the quantity of PH3⁺ cells (Fig. 2D,F and Fig. S3A) (Ladstein et al., 2010) as well as the total number of scxa⁺ cells in DMSO-treated embryos (Figs 1L, 2D,F,I and Fig. S3E,F). This indicates that proliferation contributes to the expansion of scxa⁺ cells during normal development in control DMSO-treated embryos. Strikingly, atorvastatin and aphidicolin co-treatment from 32 to 56 hpf caused >2-fold expansion in craniofacial scxa⁺ cells (Fig. 2E,G,I and Fig. S3G,H), compared with aphidicolin treatment alone, indicating that inhibiting proliferation has no effect on the ability of statin to expand tendon cell number.

As increased proliferation cannot explain the statin-mediated tendon cell expansion, we explored the possibility that cells from adjacent tissues are recruited to tendon fates following statin treatment. In the craniofacial region, skeletal and tendon cells arise from the CNC and interact with the pharyngeal mesoderm-derived muscle to form the craniofacial musculoskeletal system (Chen and Galloway, 2014; Schilling and Kimmel, 1994). To determine whether statin alters musculoskeletal development, we examined muscle and cartilage tissues by whole-mount in situ hybridization. Atorvastatin treatment from 32 to 56 hpf altered the spatial expression patterns of sox9a (Fig. 3A,B), an early marker of skeletal progenitors and differentiating chondrocytes, col2a1 (Fig. 3C,D), a component of the cartilage matrix (Yan et al., 2002), and myod1 (Fig. 3E,F), a marker of myoblast progenitors (Lin et al., 2006). Specifically, the expression of col2a1 (Fig. 3C,D, arrows) and myod1 (Fig. 3E,F, white arrow) appear to mark groups of cells that are not properly organized in the midline compared with controls. In the pectoral fin, the expression of sox9a, col2a1, sox10:eGFP and myod1 (Fig. S4A-H) was present, but with an altered morphology following statin treatment. Quantification by qPCR showed increased expression of sox9a and myod1 and no significant effect on col2a1 expression (Fig. 3G). Together, these results suggest that statin alters the morphology of the forming cartilage and muscle but does not cause a loss in their respective gene expressions.

To further quantify the effects of statin on the numbers of chondrocyte and tendon cells, we performed flow cytometry on dissociated Tg(scxa:mcherry;col2a1a:eGFP) embryos treated with DMSO or statin (Dale and Topczewski, 2011). Consistent with our col2a1 whole-mount in situ hybridization results, the morphology of col2a1a:eGFP⁺ cranial chondrocytes appeared altered (Fig. 3H-I′) compared with controls. Using flow cytometry, we were able to identify singly positive col2a1a:eGFP⁺ or scxa:mcherry⁺ cell populations from dissociated transgenic embryos. Statin had no effect on the percentage of col2a1:eGFP⁺ cells, but significantly increased the percentage of scxa:mcherry⁺ cells (Fig. 3J-M). This approximate 2-fold expansion corresponds to the increase detected by qPCR and cell quantification analysis of stained embryos. Although a scxa:mcherry⁺/col2a1a:eGFP⁺ cell population was found in controls, this represented less than 0.1% of the total cells and changes to this population upon statin treatment were not statistically significant.

We next tested whether the increased craniofacial scxa⁺ cells of statin-treated embryos derive from the neural crest. We used expression of the Tg(sox10:kaede) (Dougherty et al., 2012) as a reporter to mark cells derived from the sox10:kaede⁺ lineage. Confocal images of craniofacial region of DMSO-treated (n=4 per timepoint) and atorvastatin-treated (n=8 per timepoint) embryos were analyzed for colocalized Kaede and scxa expression. By analyzing the optical sections from confocal images, we found that scxa⁺ cells co-expressed the Kaede protein in all Tg(sox10:kaede) embryos examined following DMSO and atorvastatin treatment from 32 to 56 hpf (Fig. S4I-L). This demonstrates that the statin-mediated expanded scxa⁺ domain in the craniofacial region is exclusively derived from the CNC. Given the significant increase in scxa⁺ cells and expression in the craniofacial region, we focused on the craniofacial tendons in subsequent analyses.

Previous studies using genetic lineage tracing in mice have shown that tendon and ligament cells are descendants of Sox9⁺ cells (Huang et al., 2019; Soeda et al., 2010). Zebrafish sox9a marks migrating CNC as well as later skeletal cell types (Yan et al., 2005). Therefore, we sought to explore the possibility that the statin phenotype results from more CNC sox9a⁺ progenitors being diverted to scxa⁺ tendon fates. As our results show statin increases sox9a transcripts (Fig. 3G), we wanted to determine whether the expanded scxa⁺ cells co-expressed sox9a. To examine co-expression of scxa and sox9a, we performed fluorescent whole-mount in situ hybridization for transcripts of sox9a and mcherry for identifying scxa:mcherry⁺ tendon progenitors. For most regions in the lower jaw, sox9a and mcherry occupied spatially distinct domains in DMSO-treated Tg(scxa:mcherry) embryos at 48 hpf. In these control embryos, limited co-expression of mcherry and sox9a in some tendon attachment regions, specifically near the forming palatoquadrate and maxilla were observed (Fig. 4A,A′, yellow arrows). In contrast, we observed significant and consistent co-expression of sox9a and mcherry in statin-treated Tg(scxa:mcherry) embryos at 48 hpf (Fig. 4B,B′, yellow arrows).

sox9a and sox10:eGFP are expressed in CNC cells by 26 hpf, but their later expression becomes predominantly restricted to the forming cartilage elements by 48 hpf (Carney et al., 2006; Yan et al., 2005). Based on this, we wanted to determine whether statin causes sox10:eGFP⁺ CNC progenitors to adopt a tendon fate. To test this possibility, we performed live imaging of Tg(scxa:mcherry;sox10:eGFP) embryos treated with DMSO or statin from 32 to 53 hpf. Analysis of live images shows sox10:eGFP⁺ cells in and around Meckel's cartilage co-expressing scxa:mcherry in statin-treated embryos (Fig. 4F-H,O-T‴, arrows) whereas this was not observed in controls (Fig. 4C-E,I-N‴, arrows). Taken together, these results indicate that the statin-mediated increase in scxa⁺ cranial tendon cells results from the conversion of CNC sox9a⁺/sox10⁺ progenitors to scxa⁺ cells.

One potential mechanism by which multipotent CNC cells might be diverted to a different cell fate would be if the patterning of the craniofacial region were altered. To test this, we examined known markers of anterior-posterior and dorsal-ventral arch polarity. Despite statin-mediated scxa expansion (Fig. S5A,B), we did not identify any major patterning alterations in the expression of hoxa2b, hoxb2a, hand2, bapx1, shha, patched1 or gsc at the stages examined (Hunter and Prince, 2002; Miller et al., 2003), although the expression levels of some genes appeared reduced upon chemical treatment compared with controls (Fig. S5C-M,R). This suggested that dorsal-ventral and anterior-posterior arch polarity are not affected by statin. A second possibility was that statins could interact with and act through pathways known to maintain or promote tendon fate. We interrogated constituents of the BMP (bmp4), TGFβ (tgfbr2a and tgfbi), FGF (pea3) and Wnt (fzd7a and fzd7b) pathways (Brent and Tabin, 2004; Pryce et al., 2009; ten Berge et al., 2008; Yamamoto-Shiraishi and Kuroiwa, 2013), and observed no increase in expression of any markers in atorvastatin-treated embryos compared with controls (Fig. S5N-Q,S-X). Analysis of embryos transgenic for 6xTcf/LefBS-miniP:d2eGFP (Shimizu et al., 2012) and bre:egfp (Laux et al., 2011), which mark Wnt/β-catenin-mediated Tcf and BMP/Smad1/5 transcriptional activity, respectively, did not show expanded egfp expression in atorvastatin-treated embryos compared with controls (Figs S5Y-B′ and S6A-F). Although some differences in bre:egfp expression were observed near the mouth and posterior pharyngeal arch regions of statin-treated embryos, these alterations are consistent with a developmental delay in bre:egfp expression based on previous descriptions (Laux et al., 2011). To examine TGFβ signaling and its interaction with the statin-mediated scxa expansion, we treated embryos with an ALK5 inhibitor (SB-431542) (Inman et al., 2002), alone and in combination with atorvastatin. We found SB-431542 significantly decreased scxa expression and increased sox9a expression at 56 hpf compared with controls (Fig. S6G). In embryos co-treated with SB-431542 and atorvastatin, scxa expression is significantly increased compared with SB-431542 treatment alone to a level comparable to that of DMSO controls. The finding that statin rescues scxa expression in the presence of TGFβ inhibition suggests that the pathway targeted by statin acts independently of TGFβ signaling. Furthermore, we performed staining for phospho-Smad3 and observed differences in phospho-Smad3 staining surrounding the cartilage regions of statin-treated embryos compared with controls. Although this result indicates altered TGFβ signaling in statin-treated embryos, these phospho-Smad3⁺ regions did not co-express scxa:mcherry (Fig. S6H-S′). Together, these results suggest that statin may alter the activity of some signaling pathways in the craniofacial region, but it is unclear how these changes could directly cause the increase in tendon cells.

Taken together, our results suggest that statins expand the progenitor pool by altering CNC cell specification through a pathway previously unexplored in this context. To confirm the chemical inhibition results, we established that the expanded tendon cell phenotype results from statin inhibition of Hmgcr, the rate-limiting enzyme of the mevalonate pathway (Fig. 5A). Zebrafish have two HMGCR orthologs, hmgcra and hmgcrb. hmgcra is expressed in the larval liver and intestine; hmgcrb is maternally transcribed and ubiquitously expressed during early development (Thorpe et al., 2004). As hmgcrb is expressed during the spatial-temporal domain of interest, we focused on hmgcrb for further analysis. To test whether the statin-mediated expansion in craniofacial tendon progenitors is due to inhibition of Hmgcrb, we examined scxa and col1a2 expression in hmgcrb-deficient embryos by whole-mount in situ hybridization. In the hmgcr1b^s617 mutants (D'Amico et al., 2007; Mapp et al., 2011), identified by their severe pericardial edema, scxa expression appears expanded at 60 hpf, and in a subset at 76 hpf, the pattern of col1a2 is expanded compared with controls (Fig. S7A-D, arrow). The remaining mutants had extreme morphological abnormalities and did not express col1a2. Morpholino-mediated knockdown of the maternal and zygotic hmgcrb in zebrafish resulted in severe morphological defects and death prior to the onset of craniofacial scxa expression. Consistent with the zebrafish phenotype, the Hmgcr knockout in mouse is embryonic lethal (Ohashi et al., 2003), suggestive of an essential role for HMGCR/Hmgcr in early mouse and zebrafish development. For this reason, we injected a sub-lethal concentration of morpholino directed against hmgcrb. We next exposed the same sensitized embryos to a dose of atorvastatin that alone does not cause a tendon phenotype from 32 to 56 hpf. In this manner, we were able to induce loss of Hmgcrb within a specified developmental time window and to circumvent its earlier embryonic requirement. We observed an expansion of scxa in such doubly treated embryos at 56 hpf compared with controls (Fig. S7E-H, arrow), suggesting an essential role for Hmgcrb inhibition in the statin-mediated expansion of craniofacial scxa. To obtain unequivocal evidence that inhibition of the mevalonate pathway is responsible for the phenotype observed, we rescued the expansion in craniofacial scxa expression to wild-type levels at 56 hpf with injection of mevalonic acid, the immediate metabolite targeted by statin (Fig. 5B-D). Together, these data establish the inhibition of Hmgcrb as the mechanism by which statin promotes expansion of craniofacial scxa⁺ tendon progenitors in the zebrafish.

The mevalonate pathway metabolizes the biosynthesis of cholesterol and isoprenoids (Goldstein and Brown, 1990). Isoprenoids are covalently attached to proteins by post-translational farnesylation or geranylgeranylation. Farnesylation is the addition of CaaX by farnesyltransferase (FTase); geranylgeranylation is the addition of CAAL by geranylgeranyltransferase I (GGTase I) or GGTase II (McTaggart, 2006). To test the requirement of the cholesterol and/or prenylation branches of the mevalonate pathway in scxa expansion, we incubated embryos with inhibitors of GGTase I (GGTI-286) (Lerner et al., 1995), FTase (FTI-277) (Lerner et al., 1995) and 2,3-oxidosqualene:lanosterol cyclase (Osc; RO48-8071) (Morand et al., 1997). Inhibition of all three branches in combination causes scxa expansion at 56 hpf, which phenocopies that of embryos treated with atorvastatin alone (Fig. S7I,K, arrow). Conversely, microinjection of farnesyl pyrophosphate, a branch-point intermediate for the cholesterol and prenylation branches, rescued scxa expression at 56 hpf following atorvastatin treatment (Fig. 5E).

To dissect the pathways involved in the statin-mediated tendon cell expansion, we targeted specific branches of the mevalonate pathway using genetic and chemical loss- and gain-of-function approaches. To test the requirement of the prenylation branch in the expansion of scxa, we knocked down essential regulators of prenylation, FTase beta (fntb) and protein GGTase I beta (pggt1b) (Mapp et al., 2011). We found that loss of both fntb and pggt1b resulted in an expansion of craniofacial scxa at 60 hpf compared with controls (Fig. 5G,H), suggesting that inhibition of prenylation sufficiently expands craniofacial scxa expression. This expansion of scxa is found in pggt1b morphants alone (Fig. 5I), whereas scxa is reduced in fntb morphants (Fig. 5J), indicating that loss of GGTase I can elicit a tendon cell expansion. Furthermore, microinjection of geranylgeranyl pyrophosphate, an intermediate unique to the geranylgeranylation branch (McTaggart, 2006), rescued scxa expression at 56 hpf following atorvastatin treatment (Fig. 5F). To completely exclude the possibility that the cholesterol branch could be involved in expanding scxa expression, we inhibited cholesterol production. We found RO48-8071 had no significant effect on scxa, sox9a and myod1 transcripts, and significantly decreased col2a1 expression at 56 hpf compared with controls (Fig. 5K), suggesting that inhibition of the cholesterol biosynthesis branch does not expand craniofacial scxa⁺ tendon progenitors. Together, our data indicate that inhibition of geranylgeranylation alone and GGTase I specifically is required for the statin-mediated expansion of craniofacial scxa⁺ tendon progenitors in zebrafish.

GGTase I catalyzes the geranylgeranylation of the Rho family of GTPases, regulators of extracellular signaling that control actin cytoskeleton organization, transcriptional activation and mitogenesis (McTaggart, 2006). We examined the role of the Rho family members (Rho, Rac and Cdc42) in patterning the craniofacial musculoskeletal tissues using chemical inhibitors specific to each pathway. ROCK/Rho inhibition (Y-27632) (Uehata et al., 1997) significantly reduced col2a1, while scxa, sox9a and myod1 were not significantly changed at 56 hpf (Fig. 5L). Cdc42 inhibition (ML-141) (Hong et al., 2013) significantly reduced scxa and col2a1, but had no significant effect on sox9a and myod1 expression at 56 hpf (Fig. 5M). In contrast, Rac GTPase inhibition (EHT-1864) (Shutes et al., 2007) caused a qualitative expansion of scxa expression by whole-mount in situ hybridization (Fig. 5N) and significant quantitative increase in scxa⁺ cells (Fig. 5O) at 56 hpf. Expression of scxa and sox9a was also significantly increased, whereas col2a1 and myod1 were not significantly altered by qPCR upon EHT-1864 treatment at 48 hpf (Fig. 5P). Therefore, the two processes observed upon statin treatment, the expansion of craniofacial scxa⁺ tendon progenitors and an increase in sox9a expression, can be reproduced through EHT-1864 inhibition of Rac GTPase activity. Although a similar morphological result from pathway inhibition may not represent a true phenocopy, we believe the similarity in gene expression changes caused by Rac inhibition strongly suggests that inhibition of mevalonate through Rac GTPases, results in scxa expansion.

The mechanisms underlying tendon progenitor specification in vertebrates have remained elusive. Although TGFβ and FGF can promote tendon differentiation, no major signaling pathways have been identified to regulate tendon induction (Brent et al., 2003; Pryce et al., 2009; Havis et al., 2016). The small-molecule screen described here identified a new pathway as a crucial regulator of this process. We propose that inhibition of Hmgcr expands the craniofacial tendon progenitor pool through recruitment of neighboring sox9a⁺ CNC cells, diverting them from a sox10:eGFP⁺ skeletal fate (Fig. 6). Moreover, our data suggest that the mevalonate pathway acts through Rac GTPases in this context, thereby implicating a unique role for the mevalonate pathway and Rac GTPases in the formation of CNC tendon progenitors.

CNC cells are a multipotent population (Le Douarin and Kalcheim, 1999), although the time at which these cells become restricted and cues directing the cells towards individual fates are not well understood. Subpopulations of the CNC were postulated to become limited in potential prior to their ventral migration (Le Douarin and Smith, 1988). However, other studies have demonstrated that migrating CNC cells contain both pluripotent and lineage-restricted populations when cultured in vitro (Baroffio et al., 1991). Consistent with these findings, CNC cell fates can be altered in response to different signaling environments, indicating the cells are more plastic than originally believed (Trainor and Krumlauf, 2000). A single cell profiling analysis of mouse neural crest cells showed that cell fate decisions are made through the sequential stages of co-activation of transcriptional programs of competing fates, followed by transcriptional biasing, and finally by resolution towards one of the fates (Soldatov et al., 2019). Within the mesenchymal lineage, the CNC contains osteo-chondrogenic progenitors that become segregated into two distinct fates, cells that give rise to the bones that form either by endochondral ossification or intramembranous ossification (Akiyama et al., 2005; Mori-Akiyama et al., 2003; Nakashima et al., 2002). These studies, along with in vitro differentiation experiments showing osteogenic and chondrogenic differentiation potential of cells derived from embryonic calvarial bone (Grigoriadis et al., 1990; Stringa and Tuan, 1996), support the existence of bi-potential CNC-derived skeletal progenitors. In our studies, we observe changes in scxa⁺, scxa⁺/sox9a⁺ and scxa⁺/sox10⁺ cells in embryos treated with statin after the completion of CNC ventral cell migration. Our results are consistent with a model in which the post-migratory CNC cells contain multipotent bone-cartilage-tendon-forming progenitors rather than fate-restricted cells committed to one lineage. Studies have suggested that such multipotent progenitors capable of becoming ligament or bone exist in the zebrafish in a specific craniofacial region in the hyoid arch (Nichols et al., 2016). In our work, expression analysis of tendon markers revealed that the statin-mediated expansion of the craniofacial tendon program is localized to defined domains of the pharyngeal arches, as we never observed the entire skeletal-forming region converting to tendon fates. The restriction in the expansion of the tendon program indicates that either not all populations of CNC-derived mesenchyme are competent to activate a tendon program upon inhibition of the mevalonate pathway, or that skeletal-tendon progenitors are equivalent in their potential throughout the craniofacial region, but a subset of these sit in a microenvironment permissive for statin-mediated activation of tendon fates. These possibilities of cell competence to form tendon or site-specific signaling differences could explain why the axial tendon cells are not expanded upon statin treatment. In considering CNC tendon specification as it relates to other regions of tendon formation, cranial and limb/fin tendon progenitors have been previously suggested to have similar modes of regulation, as these regions do not require muscle for their specification whereas the axial tendon progenitors do (Brent et al., 2005; Edom-Vovard and Duprez, 2004; Grenier et al., 2009). Based on this, it is interesting to note that statin-mediated expansion of scxa expression was observed in fin and craniofacial regions, but not in axial locations. Further experiments are necessary to discern whether this represents a broader similarity in specification programs between the two regions.

In addition to increased numbers of scxa⁺ cells, our study demonstrates that statin promotes a broader scxa⁺/sox9a⁺ domain. Previous mouse lineage-tracing studies in the limb have shown that Sox9⁺ cells contribute to tendon tissues (Huang et al., 2019; Soeda et al., 2010). In the zebrafish head, sox9a and sox10 are expressed throughout the post-migratory CNC cells at 26 hpf prior to the initiation of cranial scxa expression. Previous work and the lineage-tracing studies in this paper show that cranial scxa⁺ cells descend from a CNC sox10 lineage. Therefore, it is likely that the cranial tendons also form from sox9a⁺ cells despite there being few scxa⁺ cells co-expressing either sox9a or sox10 when scxa initiates expression after 40 hpf. This likely reflects the restriction in differentiation of these cells to sox9a⁺/sox10:eGFP cartilage or scxa⁺ tendon fates. We observe a few scxa⁺/sox9a⁺ cells located at tendon-bone interfaces for Meckel's adductor tendon and palatoquadrate adductor tendon (Fig. 4A-C′, yellow arrow) (McGurk et al., 2017). The localization of scxa⁺/sox9a⁺ cells in wild-type embryos could indicate a similarity to the mammalian tendon-bone attachment unit, which also co-expresses these markers (Blitz et al., 2013; Sugimoto et al., 2013). Therefore, the statin-mediated increase in scxa and sox9a transcript levels along with a broadened scxa⁺/sox9a⁺ region could be interpreted in two ways: statin results in the persistent expression of sox9a and expansion of scxa, resulting in the conversion of CNC multipotent skeletal progenitors to tendon fates; or there is an increase in tendon-bone attachment unit cell types following statin treatment. We believe our data more strongly support the former interpretation, as our live imaging shows sox10:eGFP⁺ cells turning on scxa:mcherry expression upon statin treatment, which is not observed in wild-type embryos at this stage. In addition, we observe persistent expanded scxa expression upon statin treatment, whereas in mouse attachment units, Scx expression is not maintained at later stages of development (Blitz et al., 2013). Although our live imaging could suggest a direct transdifferentiation event from sox10:eGFP-expressing chondrocytes to scxa:mcherry-expressing cells, we favor a model where statin mediates the conversion of multipotent sox9a⁺/sox10:eGFP⁺ progenitors to tendon fates prior to the initiation of col2a1a:eGFP. Consistent with this interpretation, we do not observe a decrease in col2a1a:eGFP⁺ cells or cartilage-associated transcripts (col2a1) by qPCR (Fig. 3). This interpretation is also consistent with single cell analysis of murine neural crest cells, showing co-activation of competing fate programs prior to the commitment to a specific cell fate (Soldatov et al., 2019). Notwithstanding, given that statin increases craniofacial sox9a expression (Fig. 3A,B,G) and we did not examine col2a1 expression at later developmental stages, we cannot formally exclude the possibility that statin-mediated modulation of CNC cell fate specification may also direct cells towards a chondrocyte fate. We did observe abnormal cartilage morphology similar to previous reports of statin treatment (Signore et al., 2016). Altered scxa expression pattern was also observed for many of the chemical treatments, which could result from the chemicals targeting pathways associated with cell migration (Eisa-Beygi et al., 2014; Thorpe et al., 2004) and the cytoskeleton (Giger and David, 2017; Kardash et al., 2010; Xu et al., 2012). Interestingly, we found many of the expanded scxa-expressing cells within the periphery of the cartilage elements, corresponding to perichondrium, suggesting the expanded scxa cells arise from perichondrial cells. However, the lack of specific perichondrial markers prior to their morphological appearance makes it difficult to directly test this hypothesis.

Despite the consistent statin-mediated increase in scxa⁺ cells, we observed differing results upon inhibition of branch points downstream in the mevalonate pathway. Inhibition of Rac GTPases alone could produce similar increases in scxa and sox9a transcripts, and had no significant effect on col2a1 expression. In contrast, inhibition of cholesterol synthesis and ROCK/Rho GTPases had no significant effect on scxa and sox9a expression. Inhibition of cholesterol synthesis, ROCK/Rho or Cdc42 had a negative effect on col2a1 expression. Consistent with our findings, conditional knockout of Cdc42 using Prx1-Cre resulted in abnormal chondrogenesis in mice (Aizawa et al., 2012). We also found that ML141-mediated inhibition of Cdc42 reduced scxa expression, whereas EHT1864-mediated inhibition of Rac increased scxa expression and quantity of scxa⁺ tendon progenitors (Fig. 5N-P), potentially indicating a binary relationship in which Cdc42 is required for and Rac restricts tenogenic fate. Our finding that Cdc42 inhibition reduces scxa expression is consistent with a requirement for Rho/ROCK signals in the expression of tendon markers in mammalian mesenchymal cells in vitro (Maharam et al., 2015).

Previous studies suggest a role for Rac signaling in regulating the cytoskeleton and cell-cell contacts during mesenchymal condensation, perhaps providing a permissive or restrictive environment for cell type specification (Woods et al., 2007). Condensation of mesenchymal progenitors is essential for cartilage formation and precedes their commitment to the chondrogenic lineage (DeLise and Tuan, 2002; Oberlender and Tuan, 1994). Indeed, cell morphology influences the differentiation of mesenchyme towards chondrogenic, osteogenic and adipogenic lineages (McBeath et al., 2004; Solursh et al., 1982). In tenocytes, cell morphology influences gene expression, such as that of collagen type I (Li et al., 2008; von der Mark et al., 1977), and the actin cytoskeleton functions in proper organization of the collagenous extracellular matrix (Canty et al., 2006). This model suggests a cell-autonomous role for the mevalonate pathway and Rac GTPases in tendon fate regulation, but, as these pathways can affect multiple aspects of cell behavior and signaling, it is possible that the mevalonate pathway and Rac inhibition work indirectly on neighboring cells and tissues to alter unknown pathways that result in increased tendon specification events. Nevertheless, it is tempting to speculate that Rac inhibition results in a permissive signaling environment that directly or indirectly leads to mesenchymal progenitors adopting a tendon fate.

Using a zebrafish small-molecule screen, we established that the mevalonate pathway is involved in regulating specification of tenogenic fate based on expansion of a tendon marker domain. Our analyses of the cellular processes governing tendon cell induction provide evidence that, besides cell migration and proliferation, the mevalonate pathway has an important role in specifying cell fate in a developing embryo. Furthermore, we propose that the emergence of the tendon lineage from a common progenitor population occurs through multipotent sox9a⁺/sox10⁺ progenitors and is mediated by inhibition of Rac activity. These results could significantly impact the field of regenerative medicine in guiding approaches aimed at generating tendon tissues in vitro. Ultimately, our studies highlight the advantages of the zebrafish system and chemical screening for discovering novel genetic pathways involved in developmental processes.

Zebrafish were staged and maintained as described previously (Kimmel et al., 1995; Westerfield, 2000). All experiments were approved by the HMS and MGH IACUC. sox10:kaede (Dougherty et al., 2012), sox10:mcherry (Kamel et al., 2013), sox10:eGFP (Curtin et al., 2011), col2a1a:eGFP (Dale and Topczewski, 2011), 6xTcf/LefBS-miniP:d2eGFP (Shimizu et al., 2012) and bre:egfp (Laux et al., 2011) were used. hmgcr1b^s617 was obtained from Masazumi Tada (University College London, UK) (D'Amico et al., 2007). The scxa:mcherry transgene was generated using a bacterial artificial chromosome (BAC) containing the entire scxa genomic locus (CHORI BACPAC Resources). Using constructs and protocols provided by Stefan Schulte-Merker and Koichi Kawakami (Bussmann and Schulte-Merker, 2011; Suster et al., 2011), mcherry was recombineered into the first exon of the scxa BAC (CH211 251g8) and tol2 sites were introduced into the vector backbone. Properly targeted BACs were injected, embryos were raised to adults, and scxa:mcherry-expressing offspring were identified and raised to establish the line (fb301Tg).

Wild-type stage-matched embryos were arrayed into 48-well plates (∼10 embryos per well) of individual compounds and exposed from 32 to 56 hpf. The NINDS Custom Collection 2 library (1040 known bioactives) was obtained from the Institute of Chemistry and Chemical Biology, Longwood Screening Facility. The compounds were dissolved in DMSO and subsequently diluted in E3 buffer to a final concentration of ∼13 μM. Each compound was screened in duplicate; DMSO and SU5402 were negative and positive controls, respectively. Whole-mount in situ hybridization for scxa expression was performed to assess craniofacial tendon progenitors. Qualitative scoring (number of embryos with altered scxa expression per number scored) was conducted using the following criteria: normal/unchanged, decreased/absent and increased. Annotations were recorded regarding any changes in the spatial domain of scxa. Most compounds (97.69%; 1016/1040) failed to alter scxa expression, although two (0.19%) were toxic and resulted in death, 10 (0.96%) resulted in developmental delay and unaltered scxa expression, three (0.29%) resulted in general and non-specific morphological abnormalities, two (0.29%) increased scxa expression, six compounds were commercially unavailable and therefore could not be validated, and one (0.10%) decreased scxa expression.

Whole zebrafish embryos (n=15-125) were collected at 56 hpf following atorvastatin treatment (32-56 hpf) and dissected craniofacial regions of zebrafish embryos (n=15-45) were collected at 48 hpf following EHT-1864 treatment (32-48 hpf), at 56 hpf following RO488071, Y27632, ML141 or atorvastatin±SB-431542 treatments (32-56 hpf), and at 72 hpf following atorvastatin (48-72 hpf) or fluvastatin (32-48 hpf) treatments. Owing to high rate of toxicity, embryos collected following fluvastatin were confirmed to have a beating heart and increased scxa:mcherry expression using a dissecting microscope. Embryos were pooled per condition for RNA extraction (Qiagen, 74104). After DNase treatment (Qiagen, 79254), cDNA was synthesized (Roche, 04379012001; Invitrogen, 18091200 and Fisher Scientific, 18-091-050). For TaqMan Fast Universal gene expression assays (Applied Biosystems, 4352042), the FAM-dye probes used were: scxa (Dr03104896) and β-actin (Dr03432610). For Fast SYBR Green PCR assays (Applied Biosystems, 4385618), the primers used were: scxa forward, 5′-GGCGTCCAGTTACATCTCT-3′; scxa reverse, 5′-GTCTTTGCTCATTTTCCTCTGGTT-3′; sox9a forward, 5′-GGCCGACCAGTACCCGCACCTC-3′; sox9a reverse, 5′-GATTCGCTCTGGCCGTTCTTCACC-3′; col2a1 forward, 5′-TGCTGCAGGGCTCCAATGATGT-3′; col2a1 reverse, 5′-CTGCGCCAATGTCCACTCCGAACT-3′; myod1 forward, 5′-GAGGGAGAGGAGGCGACTGAGC-3′; myod1 reverse, 5′-CTGACACGTTGGGCCCATAAAATC-3′; col1a2 forward, 5′-GCCCCGCTGGTAAAGATGGTTCAA-3′; col1a2 reverse, 5′-ACGGCAAGTACGAGCGGGGTTCT-3′); β-actin forward, 5′-GATGCCCCTCGTGCTGTTTTC-3′; β-actin reverse, 5′-TCTCTGTTGGCTTTGGGATTCA-3′).

No-reverse transcriptase controls were used for each sample. The ratio of the target transcript to β-actin reference was quantified and normalized to the control sample. Statistical analysis was conducted with significance defined as P≤0.05 (Welch's two-sample two-tailed t-test) and error bars represent standard deviation based on all technical replicates and all biological replicates.

Morpholinos (Gene Tools) were injected as described previously: fntβ ATG (Mapp et al., 2011), hmgcr1b ATG (D'Amico et al., 2007) and pggt1β ATG (Mapp et al., 2011). Morpholino concentrations were: fntβ (1.4 ng alone and 2.1 ng in fntβ-pggt1β morphants); pggt1β (5.5 ng alone and 2.8 ng in fntβ-pggt1β morphants); and1.0 ng hmgcr1b. Embryos were incubated in the following compounds: atorvastatin, lovastatin, fluvastatin, FTI-277 and SU5402 (Tocris Bioscience); simvastatin, aphidicolin and GGTI-286 (Sigma-Aldrich); and RO48-8071, EHT-1864, ML-141, Y-27632 and SB-431542 (Cayman Chemical). In the metabolite rescue experiments, the pharyngeal cavity was injected at 28-30 hpf with 1-2 nl of the following compounds (Sigma-Aldrich): 2.6 mM geranylgeranyl pyrophosphate, 10 mM farnesyl pyrophosphate, 1.0 M mevalonolactone and DMSO or ethanol (vehicle control).

Colorimetric whole-mount in situ hybridization was performed as described previously (Brent et al., 2003). Probes include scxa (AL923903 and AL921296), col1a2 (DY559926), tnmd (BC155615 and EV754577), myod1 (BC078421), sox9a (Yan et al., 2005), col2a1 (Yan et al., 1995), runx2a (Burns et al., 2002), hoxa2b (Kong et al., 2014), hoxb2a (Kong et al., 2014), hand2 (Foote et al., 2016), bapx1 (Kamel et al., 2013), tgfbi (CB361081), tgfbr2a (Thisse et al., 2001), pea3 (Münchberg et al., 1999), fzd7a (Kamel et al., 2013), fzd7b (Rochard et al., 2016), shha (Piotrowski and Nüsslein-Volhard, 2000), patched1 (Lewis et al., 1999), bmp4 (Gonzalez et al., 2000), ednraa (Nair et al., 2007) and gsc (Erter et al., 1998). Fluorescent whole-mount in situ hybridization and antibody staining were performed as described previously (Chen and Galloway, 2017). Primary antibodies were 1:100 anti-PSmad3 (Abcam, ab52903); 1:500 anti-kaede (MBL, PM012M); and 1:200 anti-phospho-Histone H3 (Millipore, 06-570). The sensitivity of the anti-kaede antibody marks the descendants of the sox10:kaede⁺ lineage that have turned off sox10:kaede expression. Secondary antibodies were 1:500 goat anti-mouse IgG2b-HRP (Southern Biotech); 1:200 donkey anti-rabbit Alexa Fluor 488 and 1:500 donkey anti-rabbit IgG-HRP (Jackson ImmunoResearch Laboratories); 1:400 goat anti-mouse Alexa Fluor 488 (Invitrogen); and 1:400 donkey anti-rabbit Alexa Fluor 647 (Abcam). All images captured were of representative embryos for each condition.

Tg(sox10:mcherry) embryos were chemically treated at 32 hpf, pulsed with EdU for 20 min at 4 h or 5 h intervals, and chased with E3 buffer for 30 min prior to fixation. Embryos were counterstained with Hoechst33342 (Invitrogen). For quantification, we used confocal microscopy and counted the number of proliferative cells in optical sections spanning the depth of a defined craniofacial region of the embryos processed by Click-iT EdU Alexa Fluor 488 Imaging Kit (Invitrogen, C10337). Statistical analysis was conducted with significance defined as P≤0.05 (Welch's two-sample two-tailed t-test) and error bars are standard deviation.

For quantification of craniofacial scxa- and col1a2-expressing cells, we counted the number of cells in optical confocal sections spanning the area of the craniofacial region of embryos stained for the transcript of interest by fluorescent whole-mount in situ hybridization and counterstained with Hoechst33342. For quantification of pectoral fin and myoseptal scxa-expressing cells, we counted the number of mcherry⁺ cells in maximum-intensity projection images spanning the width of the pectoral fin or trunk region, respectively, of Tg(scxa:mcherry) embryos. For quantification of cells co-expressing scxa transcript and PH3 protein (Fig. S3I,L-Q), we counted the number of cells in optical sections spanning the depth of a defined region of the embryos processed by fluorescent whole-mount in situ hybridization and immunohistochemistry. Statistical analysis was conducted using the Mann–Whitney-Wilcoxon test with significance defined as P≤0.05.

Embryos were imaged using a Zeiss AxioZoom V16 with ApoTome2 or Nikon Eclipse 80i, and images were acquired using Zeiss Zen or NIS Elements. Confocal time-course images were taken using a Nikon ECLIPSE Ti-E and confocal images were taken using a Zeiss LSM 710 NLO or Nikon ECLIPSE Ti-E. Confocal images were acquired with Zeiss Zen or ImageJ software using the maximum-intensity projection (MIP) feature applied to z-stacks and processed uniformly with Photoshop. Tg(bre:egfp), Tg(scxa:mcherry), Tg(col2a1a:eGFP; scxa:mcherry) and Tg(6XTcf/LefBS-miniP:d2eGFP) embryos were mounted in 0.8-1.0% low melting-point agarose (RPI, 9012-36-6) in E3 buffer. Embryos for live imaging were anesthetized with 0.015-0.2% tricaine in E3 buffer. Embryos, positioned with the ventral side of the head facing the objective, were imaged every 1 h using a z-stack module. Post-imaging, the embryos resumed normal development at 28.5°C in fresh E3 buffer.

56 hpf embryos were fixed in 4% PFA overnight and then rinsed in PBST (PBS+0.15% Triton X-100) for 3×10 min at room temperature (RT). Embryos were next permeated in 20 μg/ml Proteinase K (Thermo Fisher Scientific, 507203027) for 25 min at room temperature and rinsed in PBST for 3×10 min. Afterwards, the embryos were blocked with 4% BSA (ThermoFisher Scientific, BP1605-100) in PBST for 2 h at room temperature and subsequently incubated with anti-Smad3 antibody overnight at 4°C. Embryos were washed with PBST for at least 2 h, with the PBST changed every 30 min, and then incubated with donkey anti-rabbit Alexa Fluor 647 overnight at 4°C. Hoechst (ThermoFisher Scientific, H3570) was added to the secondary antibody solution for nuclear staining.

Fluorescent whole-mount in situ hybridization was performed as previously described (Chen and Galloway, 2017) using DIG- (Roche, 11277073910) and fluorescein- (Roche, 11685619910) labeled RNA probes against sox9a and mcherry. Probes were synthesized using Sp6 or T7 RNA polymerase (Roche, 10810274001 or 10881767001). Signals were visualized using a TSA Plus Fluorescein Evaluation Kit (PerkinElmer, NEL741E001KT) and TSA Cyanine 5 Evaluation Kit (PerkinElmer, NEL705A001KT).

The Tg(col2a1a:eGFP; scxa:mcherry) embryos were treated with DMSO or atorvastatin from 32 to 56 hpf. At 56 hpf, 15-20 pooled heads from each group were collected for single cell dissociation as previously described (Bresciani et al., 2018). The ratio of each population: scxa:mcherry⁺ cells, col2a1a:eGFP⁺ cells and scxa:mcherry⁺/col2a1a:eGFP⁺ cells were analyzed using a BD FACS Aria 3 Cell Sorter equipped with ultraviolet, green and red lasers. Gates were set using positive (single transgene for scxa:mcherry and col2a1a:eGFP) and negative (wild-type) control cell populations.

We thank Caroline Shamu, Jennifer Smith, David Wrobel and Doug Flood, and personnel at the ICCB Longwood, for assistance and guidance with the screen. We are grateful to Jessica Lehoczky, Mor Grinstein, Heather Dingwall, the BIDMC and MGH Simches fish facilities, Trista North, Wolfram Goessling, and North and Goessling lab members.