Fgf-driven Tbx protein activities directly induce myf5 and myod to initiate zebrafish myogenesis

Sarcomeric muscle arose early in animal evolution and is today regulated by similar gene families in Drosophila and vertebrates e.g. mef2 genes (Taylor and Hughes, 2017). Skeletal myogenesis is initiated during gastrulation. Yet conserved bilaterian pathways leading specifically to skeletal (as opposed to cardiac or visceral) muscle have been hard to discern, perhaps because new regulatory steps have evolved in each phylum since their divergence.

A key step in vertebrate evolution was the chordate transition through which animals acquired notochord, post-anal tail, gill slits and dorsal neural tube, facilitating swimming (Brunet et al., 2015; Gee, 2018; Gerhart, 2001; Satoh et al., 2012). Throughout vertebrates, the notochord patterns the neural tube and paraxial mesodermal tissue by secreting Hedgehog (Hh) signals that promote motoneuron and early muscle formation (Beattie et al., 1997; Blagden et al., 1997; Du et al., 1997; Münsterberg et al., 1995; Roelink et al., 1994). Nevertheless, in the absence of either notochord or Hh signalling, muscle is formed in vertebrate somites (Blagden et al., 1997; Dietrich et al., 1999; Du et al., 1997; Grimaldi et al., 2004; Zhang et al., 2001). How might deuterostome muscle have formed prior to evolution of the notochord?

A change in function of the Tbxt gene, a T-box (Tbx) family paralogue, occurred during chordate evolution such that Tbxt now directly controls formation of posterior mesoderm, notochord and post-anal tail in vertebrates (Chiba et al., 2009; Showell et al., 2004). Hitherto, Tbxt may have distinguished ectoderm from endoderm and regulated formation of the most posterior mesendoderm (Arenas-Mena, 2013; Kispert et al., 1994; Woollard and Hodgkin, 2000; Yasuoka et al., 2016). In zebrafish, Tbxt genes also promote slow myogenesis (Coutelle et al., 2001; Halpern et al., 1993; Martin and Kimelman, 2008; Weinberg et al., 1996). Other Tbx genes, such as Tbx1, Tbx4, Tbx5, Tbx16 and Tbx6, also influence sarcomeric muscle development in both fish and amniotes (Chapman and Papaioannou, 1998; Griffin et al., 1998; Hasson et al., 2010; Kimmel et al., 1989; Manning and Kimelman, 2015; Nandkishore et al., 2018; Weinberg et al., 1996; Windner et al., 2015). For example, the Tbx6 family is implicated in early stages of paraxial mesoderm commitment, somite patterning and myogenesis (Bouldin et al., 2015; Chapman and Papaioannou, 1998; Kimmel et al., 1989; Manning and Kimelman, 2015; Nikaido et al., 2002; White et al., 2003; Windner et al., 2012). It is unclear, however, whether the Tbx genes promote myogenesis directly, and/or are required for earlier events in mesoderm development that are necessary for myogenesis.

In vertebrates, a key essential step in skeletal myogenesis is activation of myogenic regulatory factors (MRFs) encoded by the myf5 and myod genes (Hinits et al., 2009, 2011; Rudnicki et al., 1993). In mice, distinct myogenic cell populations initiate myf5 and myod expression in different ways, the genes being induced by distinct signals through different cis-regulatory elements in different skeletal muscle precursor cells. Once expressed, these MRF proteins have two functions: to remodel chromatin and directly enhance transcription of muscle genes (reviewed by Buckingham and Rigby, 2014). In zebrafish, as the anteroposterior axis forms and extends, de novo induction of myf5 and myod mRNAs in slow and fast muscle precursors occurs in tissue destined to generate each successive somite (Coutelle et al., 2001). Zebrafish myogenesis begins at about 75% epiboly stage when adaxial cells that flank the shield/organiser/nascent notochord (hereafter called pre-adaxial cells; see schematic in Fig. 1A), begin MRF expression (Hinits et al., 2009; Melby et al., 1996). Pre-adaxial cells express both myf5 and myod, converge to form two rows of adaxial cells flanking the notochord, become incorporated into somites and differentiate into slow muscle fibres (Coutelle et al., 2001; Devoto et al., 1996; Weinberg et al., 1996). Loss of either MRF alone is not sufficient to prevent slow myogenesis, but loss of both completely inhibits adaxial slow muscle formation (Hinits et al., 2009, 2011). Dorsal tissue immediately lateral to the pre-adaxial cells, the paraxial mesoderm (Fig. 1A), expresses myf5 and subsequently generates fast muscle once somites have formed, upregulating myod in the process. A key to understanding myogenesis in both paraxial and adaxial cells is thus the mechanism(s) by which myf5 and myod expression is regulated.

Intrinsic factors such as Tbx proteins likely interact with extrinsic positional cues within the embryo to pattern myogenesis. Fgf and Tbx function have long been known to interact to drive early mesendoderm patterning, but how directly they control early embryonic myogenesis remains unclear (Kimelman and Kirschner, 1987; Showell et al., 2004; Slack et al., 1987). Various Fgf family members are expressed close to myogenic zones during vertebrate gastrulation (Isaacs et al., 2007; Itoh and Konishi, 2007; Wilkinson et al., 1988). In zebrafish, Fgf signalling is required for mesendoderm formation, tailbud outgrowth and normal fast myogenesis (Draper et al., 2003; Griffin et al., 1995; Groves et al., 2005; Reifers et al., 1998; Yin et al., 2018). Fgf signalling is also involved in early expression of myf5 and myod in pre-adaxial cells, but the mechanism of initial induction of myf5 and myod is unknown (Ochi et al., 2008). Expression of several Fgfs has been detected in the chordoneural hinge (CNH; Fig. 1A) adjacent to pre-adaxial cells (Draper et al., 2003; Groves et al., 2005; Thisse and Thisse, 2005). Subsequently, Hh signalling from the ventral midline maintains MRF expression and progression of the pre-adaxial cells into terminal slow muscle differentiation (Coutelle et al., 2001; Hirsinger et al., 2004; Lewis et al., 1999). Here, we show how both Fgf and Hh extracellular signals cooperate with Tbx genes to control fast and slow myogenesis. In the trunk, Fgf signalling is required for the initiation of myogenesis and acts in cooperation with Tbx16/Tbxta function directly on myf5 and myod. In the tail, by contrast, direct MRF gene induction by Fgf is not required and the evolutionary novelty of midline-derived Hh signalling accounts for adaxial myogenesis.

Adaxial myogenesis is driven by successive Fgf and Hh signals. When Hh signalling was prevented with the Smoothened (Smo) antagonist cyclopamine (cyA) from 30% epiboly, myf5 and myod mRNAs in pre-adaxial cells were unaffected at 90% epiboly (Fig. 1A). In contrast, when Fgf signalling was inhibited with SU5402 from 30% epiboly both myf5 and myod mRNAs were lost (Fig. 1A) (Ochi et al., 2008). Persistence of the anterior mesodermal marker aplnrb (Zeng et al., 2007) showed that the lack of MRFs was not due to failure of gastrulation caused by SU5402 treatment (Fig. 1A, Fig. S1; see Table S1 for quantification and Table S5 for a summary checklist of results of this and subsequent experiments). In SU5402-treated embryos, aplnrb mRNA revealed the normal invagination of mesoderm and aplnrb-expressing cells flanking the organiser. However, both downregulation of aplnrb mRNA in paraxial trunk mesoderm that normally expresses myf5 alone, and pre-adaxial aplnrb upregulation were absent in SU5402-treated embryos (Fig. 1A). Thus, early Fgf signalling is required for the initiation of skeletal myogenesis in future trunk regions.

As trunk myogenesis proceeds, Hh signalling becomes essential for adaxial slow myogenesis. Blockade of Hh signalling with cyA or in smo mutants (which lack Smoothened, an essential component of the Hh signal transduction pathway), led to loss of myod mRNA from adaxial slow muscle at the 6 somite stage (ss), whereas paraxial fast muscle precursors were unaffected (Fig. 1B). Nevertheless, myod mRNA transiently accumulated in pre-adaxial cells of presomitic mesoderm (PSM) destined to make trunk somites, but was then lost in anterior PSM (Fig. 1B) (Barresi et al., 2000; Lewis et al., 1999; Osborn et al., 2011; van Eeden et al., 1996, 1998). Thus, as suggested previously (Coutelle et al., 2001; Ochi et al., 2008), in the wild-type situation trunk pre-adaxial myod expression is maintained and enhanced by Hh. In contrast, during tail myogenesis at 15ss and thereafter, no pre-adaxial myod expression was detected in smo mutants or cyA-treated embryos (Fig. 1B, Fig. S2). These data suggest that whereas Hh is necessary for induction of adaxial myogenesis in the tail, Fgf-like signals initiate myod expression in trunk pre-adaxial cells.

Additional evidence emphasises the greater reliance on Hh in tail myogenesis. In shha mutants at 24 hours post-fertilisation (hpf), slow muscle was lost from tail but remained in trunk somites, suggesting that slow muscle in tail is more sensitive to reduction in Hh activity (Fig. 2A,B). Moreover, injection of myod or myog mRNA into embryos lacking Hh signalling was able to rescue slow myogenesis in trunk but not in tail (Fig. S3). Similarly, absence of notochord-derived signals in noto (flh) mutants, in which the nascent notochord loses notochord character and converts to muscle (Coutelle et al., 2001; Halpern et al., 1995), ablated tail but not trunk slow muscle (Fig. 2C). Treatment of noto mutants with cyA shows that myod expression is initiated in trunk pre-adaxial cells adjacent to the transient pre-notochordal tissue, but fails to be maintained owing to the blockade of floorplate-derived Hh signals (Fig. 2D). Taken together, these data show that Hh can initiate and then maintain MRF gene expression, but that other signals initiate slow myogenesis in the trunk.

We next tested whether Hh-independent myod expression and myf5 upregulation in trunk pre-adaxial cells requires Fgf signalling. Treatment with cyA left residual pre-adaxial myod and myf5 mRNA flanking the base of the notochord at trunk levels, but ablated adaxial expression in slow muscle precursor cells (Fig. 2E). The residual expression was ablated when, in addition to cyA, SU5402 was used to block Fgf signalling from 30% epiboly (Fig. 2E). Application of SU5402 alone diminished myf5 and myod mRNA accumulation up to the tailbud stage, but caused little if any reduction of adaxial myf5 and myod mRNAs in the tailbud region at 6ss, after midline shha function had commenced (Fig. 1B) (Krauss et al., 1993). Nevertheless, SU5402 greatly diminished myod expression in somitic fast muscle precursors and reduced the extent of myf5 expression in paraxial PSM (Fig. 1B) (Groves et al., 2005; Reifers et al., 1998).

To identify candidate Fgf regulators of pre-adaxial myogenesis, the expression patterns of fgf3, fgf4, fgf6a and fgf8a were investigated in wild-type embryos (Fig. S4A). As reported previously, fgf4, fgf6a and fgf8a mRNAs were all detected in the posterior dorsal midline at 80% epiboly, followed by fgf3 in the chordoneural hinge (CNH) and posterior notochord (Fig. S4A) (Kudoh et al., 2001; Thisse and Thisse, 2005; Yamauchi et al., 2009). These Fgfs are candidate regulators of myf5 and myod.

Each Fgf was knocked down with previously validated antisense morpholino oligonucleotides (MOs) in wild-type embryos (Fig. S4B). At 80% epiboly, there was little or no decrease of myf5 or myod mRNA in individual Fgf morphants or fgf8a mutant embryos (Fig. S4B). Combinatorial knockdown of several Fgfs led to progressively more severe loss of myod mRNA and reduction of the raised pre-adaxial and paraxial levels of myf5 mRNA (Fig. 3A, Fig. S4C,D) and pre-adaxial aplnrb mRNA (Fig. S4E). Thus, specific Fgfs collaborate to drive the initial expression of myod and myf5 in pre-adaxial and paraxial cells.

By tailbud stage, however, fgf4+fgf8a MO treatment alone had little effect on myod mRNA accumulation, presumably due to the presence of Hh in the midline (Fig. 3B). Congruently, cyA-treatment reduced anterior adaxial myod mRNA, but pre-adaxial expression persisted after blockade of Hh signalling (Fig. 3B). Pre-adaxial myod mRNA was ablated in cyA-treated embryos injected with fgf4+fgf8a MO (Fig. 3B). Thus, expression of Fgf4 and Fgf8a in the shield, CNH and posterior notochord, provides a spatiotemporal cue for pre-adaxial myogenic initiation in the tailbud.

To test the ability of Fgfs to promote myogenesis further, we generated ectopic Fgf signals by injection of fgf4 or fgf6a mRNA into wild-type embryos, and analysed myf5 and myod mRNA at 80% epiboly (Fig. 3C). Both myod and myf5 mRNAs were upregulated in more ventral regions at levels comparable to those in adaxial cells of control embryos, despite the absence of Hh signalling in these regions. Overexpression of fgf4 mRNA upregulated myf5 mRNA in an initially uniform band around the embryo that extended towards the animal pole for a distance similar to that of myod mRNA in adaxial cells of controls (Fig. 3C). myod was less easily induced, but was upregulated in similar regions of the mesoderm. Fgf4-injected embryos became ovoid, with germ ring constriction that stretched and broadened the notochord (although the numbers of DAPI-stained notochord cells appeared normal). There was a positive correlation between the extent of myf5 and myod mRNA upregulation and the extent of deformation. aplnrb mRNA persisted in animal regions of the mesoderm, but was downregulated where myf5 mRNA was induced nearer to the margin, revealing that anterior/cranial mesoderm is present but resistant to Fgf-driven MRF induction and aplnrb suppression (Fig. 3D). Thus, Fgf4 dorsalised the embryo, converting the entire posterior paraxial and ventral mesoderm to a myogenic profile with some regions expressing only myf5 and others expressing also myod, particularly around the germ ring. Fgf6a overexpression also induced ectopic MRFs in cells around the germ ring, which then appeared to cluster (Fig. 3C). Taken together, these data show that posterior/dorsal Fgf signals initiate MRF expression in both pre-adaxial slow and paraxial fast muscle precursors in pre-somitic mesoderm.

Zebrafish Tbx genes, including tbxta and tbx16 [formerly called no tail (ntla) and spadetail (spt), respectively], potentially mediate Fgf signalling in gastrulating embryos (Amaya et al., 1993; Griffin et al., 1995; Smith et al., 1991; Sun et al., 1999). tbx16 is suppressed by a dominant negative Fgf receptor (FgfR) (Griffin et al., 1998). However, whether tbxta and tbx16 activities are altered by SU5402 treatment, which also blocks FgfR function, is unclear (Rentzsch et al., 2004; Rhinn et al., 2005). Wild-type embryos at 30% epiboly were therefore exposed to SU5402 and subsequently fixed at 80% epiboly or 6ss to investigate expression of tbxta and tbx16 (Fig. 4A). Application of 10 μM SU5402 diminished tbxta mRNA in notochord, and tbxta and tbx16 mRNAs in the germ ring (particularly in dorsal paraxial regions) at 80% and in the tailbud at 6ss (Fig. 4A), and 30 μM SU5402 abolished tbxta and tbx16 mRNAs throughout the trunk (Fig. 4A). Thus, Fgf-like signalling is required for normal Tbx gene expression in mesoderm.

Tbxta is required for normal myf5 and myod expression in posterior regions during tailbud outgrowth, partly due to loss of notochordal Hh signalling, which normally maintains a high level of tbx16 mRNA accumulation in adaxial cells in anterior PSM (Fig. S5) (Coutelle et al., 2001; Weinberg et al., 1996). To test whether Tbx genes are required for MRF expression at earlier stages, each Tbx gene was knocked down and MRF and Fgf expression analysed at 80% epiboly. Tbxta knockdown reduced expression of dorsal midline Fgfs, ablated myod mRNA and reduced myf5 mRNA accumulation in pre-adaxial cells to the level in paraxial regions (Fig. 4B). However, Tbxta knockdown had little effect on either fgf8a or myf5 mRNAs in more lateral paraxial mesoderm or the germ ring (Fig. 4B). This correlation raised the possibility (addressed below) that Tbxta may drive MRF expression through induction of Fgf expression. Tbx16 knockdown, on the other hand, ablated pre-adaxial myod mRNA and reduced both pre-adaxial and paraxial myf5 mRNA without reduction of Fgf expression (Fig. 4B). Indeed, both germ ring fgf8a mRNA at 80% epiboly and dorsal midline fgf3, fgf4 and fgf8a mRNAs in the tailbud at 6ss appeared to be increased (Fig. 4B), as previously described (Warga et al., 2013). As Tbx16 expression persists in tbxta mutants (Amack et al., 2007; Griffin et al., 1998), these data raise the possibility that Tbx16 is required to mediate the action of Fgf signals on myogenesis.

tbx16 null mutants show a failure of convergent migration of mesodermal cells into the paraxial region (Ho and Kane, 1990; Molven et al., 1990), which, by reducing mesodermal cells flanking the CNH, may contribute to the reduction in MRF mRNAs observed at 80% epiboly. Nevertheless, tbx16 mutants generate enough paraxial mesoderm that reduced numbers of both paraxial fast muscle and adaxially derived slow muscle fibres arise after Hh signalling commences (Honjo and Eisen, 2005; Weinberg et al., 1996). To investigate whether Tbx16 is required for initial induction of myf5 and/or myod expression in pre-adaxial cells, we titrated MO to reduce tbx16 function to a level that did not prevent accumulation of significant numbers of trunk mesodermal cells and examined myf5 and myod expression at 6ss (Fig. 4C). In tbx16 morphants, myod mRNA was readily detected in adaxial cells adjacent to notochordal Hh expression (Fig. 4C, cyan arrowheads). Treatment of these tbx16 morphants with cyA to block Hh signalling, however, ablated adaxial myod expression, leaving only weak myod in paraxial somitic fast muscle precursors (Fig. 4C, cyan and pink arrowheads). In contrast, treatment of control embryos with cyA left pre-adaxial myod mRNA intact (Fig. 4C, black arrowheads). These findings show that Fgf-driven pre-adaxial myod expression flanking the CNH requires Tbx16 function.

Adaxial myf5 expression also requires Tbx16. Tbx16 knockdown reduced myf5 mRNA accumulation in the posterior tailbud, and also diminished the upregulation of myf5 mRNA in pre-adaxial and adaxial cells (Fig. 4C, orange arrowheads). Addition of cyA to tbx16 morphants had little further effect on myf5 expression (Fig. 4C, white arrowheads). In contrast, cyA treatment alone reduced adaxial myf5 mRNA in the anterior PSM, but did not affect the myf5 upregulation in pre-adaxial cells or tailbud myf5 expression (Fig. 4C, white arrowheads). Additional knockdown of Tbx16 in cyA-treated embryos prevented pre-adaxial myf5 upregulation (Fig. 4C, white arrowheads). Thus, Tbx16 is required for Fgf to upregulate both myf5 and myod in pre-adaxial cells.

Both pre-adaxial and anterior PSM adaxial myod expression were also absent in tbxta morphants, but recovered in somites, again due to midline-derived Hh signalling (Fig. 4C) (Coutelle et al., 2001). In marked contrast, Tbxta knockdown upregulated myf5 mRNA in the tailbud (Fig. 4C, asterisks), presumably reflecting loss of tailbud stem cells that lack myf5 mRNA. Taken together, the data strongly suggest that tbx16 is required for midline-derived Fgf signals to induce myod and upregulate myf5 in pre-adaxial cells in the tailbud. In contrast, the loss of MRF expression in tbxta mutants could be simply explained by loss of midline-derived Fgf signals, and/or might require some other Tbxta-dependent process.

To test rigorously whether Tbx16 is required for Fgf to induce MRFs, fgf4 mRNA was injected into embryos from a tbx16 heterozygote incross. Whereas Fgf4 upregulated myf5 and myod mRNAs all around the germ ring in siblings, in sequence-genotyped tbx16 mutant embryos no upregulation was detected (Fig. 5A). It is clear that mesoderm was present in tbx16 mutants because the mRNAs encoding Aplnrb, Tbxta, Tbx16 and Tbx16-like (formerly known as Tbx6 and then Tbx6-like) are present in tbx16 mutants (Fig. S6; Griffin et al., 1998; Morrow et al., 2017). The effect of Fgf4 does not appear to act by radically altering tbxta or tbx16 gene expression (Fig. 5B). Thus, Tbx16 is required for Fgf-driven expression of MRFs in pre-somitic mesoderm.

The results so far show that tbx16 function is necessary for Fgf to induce myf5 and myod (Fig. 5A). To determine whether increased Tbx16 activity is sufficient to induce MRFs, Tbx16 was overexpressed. Injection of tbx16 mRNA caused ectopic expression of both myf5 and myod in the germ marginal zone (Fig. 5C). Notably, Tbx16 overexpression induced myf5 mRNA in a much broader region than was observed for myod mRNA.

We next tested whether Tbx16 could induce expression of myf5 or myod in the absence of Fgf signalling. Exposure to a high dose of SU5402, which downregulates endogenous tbx16 and tbxta mRNA (Fig. 4A), prevented MRF induction by injection of tbx16 mRNA (Fig. 5D). Nevertheless, when tbx16 mRNA was injected into low dose (10 µM) SU5402-treated embryos, which normally have reduced MRF expression, the level of myf5 and myod mRNAs was rescued (Fig. 5C). However, tbx16 mRNA was less effective at ectopic MRF induction in the presence of SU5402 (Fig. 5C). These results demonstrate that Fgf signalling cooperates with Tbx16 activity in inducing expression of myf5 and myod in pre-adaxial cells at gastrulation stages. Moreover, Tbx16 requires Fgf-like signals to induce MRF gene expression.

In order to determine further the regulatory relationship between Tbx16 and myf5 and myod, we interrogated our previously published chromatin immunoprecipitation and sequencing (ChIP-seq) experiments for endogenous Tbx16 at 75-85% epiboly stage (see Materials and Methods) (Nelson et al., 2017). Our analyses revealed a highly significant peak at −80 kb upstream (myf5 distal element, 5DE) and two peaks proximal to myf5 (proximal elements, 5PE1,5PE3) (Fig. 6A,B; Table S3). Cross-referencing to published histone modification ChIP-seq data (Bogdanovic et al., 2012) revealed that 5DE and 5PE1 overlapped significant H3K27ac and H3K4me1 peaks (Table S3), suggesting that these are functionally active enhancers. Tbx16 ChIP-qPCR confirmed the validity of the 5DE and 5PE1 ChIP-seq peaks (Fig. 6C). These putative enhancers are likely to regulate myf5, the promoter of which has a H3K4me3 mark, rather than the adjacent myf6 gene, which is not expressed at 80% epiboly and does not have a H3K4me3 mark. Each peak showed significant conservation to regions adjacent to myf5 in other fish species (Fig. 6A,B). Thus, ChIP-seq peaks corresponding to histone marks indicative of enhancer activity suggest evolutionarily conserved mechanisms of myf5 regulation in fish.

We next tested whether Tbx16 is able to positively regulate myf5 directly by using a dexamethasone (DEX)-inducible system to activate Tbx16 in the absence of translation (Kolm and Sive, 1995; Martin and Kimelman, 2008) (Fig. 6D). Ectopic expression of myf5 around the germ ring, and of myod in a narrower domain flanking the base of the notochord, was observed upon Tbx16 activation (Fig. 6E). Interestingly, across the set of cycloheximide (CHD)+DEX-treated embryos, ectopic myf5 mRNA was induced to a higher level than elsewhere (as assessed by staining intensity) in a similar region to ectopic myod mRNA, suggesting that Tbx16 was able to induce two aspects of pre-adaxial character (myod expression and upregulation of myf5) directly in this region of the embryo. To confirm this result, we tested whether Myf5 is required for Tbx16 to induce myod expression. When tbx16 mRNA was injected into myf5 mutant or heterozygote embryos, ectopic myod mRNA was observed flanking the base of the notochord in about 50% of mutant or heterozygote embryos, but appeared to be more readily induced in wild-type siblings (Fig. 6F). Thus, Tbx16 is necessary for MRF expression and can induce both myf5 and myod independently, as long as Fgf signalling is active. Moreover, a feed-forward mechanism operates by which Tbx16 induction of myf5 augments, but is not essential for, induction of myod. In summary, Tbx16 directly induces MRF expression in gastrulating mesoderm and is particularly potent in the pre-adaxial region that normally retains high tbx16 expression.

Whereas the entire paraxial PSM expresses myf5, pre-adaxial cells upregulate myf5 and are the first cells to express myod. Tbxta and Tbx16 have similar DNA-binding recognition sequences (Garnett et al., 2009; Nelson et al., 2017). Congruently, we observed a prominent Tbxta ChIP-seq peak at the 5DE −80 kb site upstream of myf5, and minor peaks at the proximal sites (Fig. 6A,B; Table S3). Because of the role of Tbx16 and Tbxta in myod expression, we also examined the myod locus for Tbx protein binding. We found multiple sites occupied by Tbxta and Tbx16 either individually or in combination (Table S3). Notably, only one site (DDE3) displayed strongly significant H3K4me1 and H3K27ac peaks and this was only occupied by Tbxta and not by Tbx16 (Fig. S7, Table S3). However, an additional site (DDE1) showed significant occupancy by Tbx16 and Tbxta concurrent with H3K4me1 but not H3K27ac (Fig. S7, Table S3). These findings indicate that differential direct binding of Tbxta and Tbx16 may control both myf5 and myod expression at the inception of skeletal myogenesis.

Is Tbxta also required for MRF expression in response to Fgf? Overexpression of Fgf4 in tbxta mutants successfully induced myf5 mRNA and suppressed aplnrb mRNA widely in the posterior mesoderm except in a widened dorsal midline region, showing that the introduced Fgf4 was active (Fig. 7A,B). However, myod expression was not rescued in tbxta mutants in the dorsal pre-adaxial region of Fgf4-injected embryos, or elsewhere around the germ ring (Fig. 7A). Moreover, even an increased dose of 225 pg fgf4 mRNA/embryo failed to rescue myod mRNA in tbxta mutants. Importantly, tbxta heterozygotes showed significantly less extensive induction of myod mRNA in response to Fgf4 than did their wild-type siblings (P=0.0001 χ²-test; Fig. 7C; Table S4). Therefore, Tbxta is essential for myod induction in pre-adaxial cells independently of its role in promoting expression of midline Fgfs.

Although Fgf4 injection did not radically alter the location of tbx16 or tbxta mRNA (Fig. 5B), we noticed that the higher level of tbx16 mRNA in adaxial compared with paraxial cells was not obvious in Fgf4-injected wild-type embryos, with high levels present at all dorsoventral locations, presumably because pre-adaxial character was induced widely in posterior mesoderm (Fig. 5B). Nevertheless, as Fgf4 injection into tbxta mutants induced myf5 but not myod mRNA (Fig. 7A), it seems that Tbxta is essential for progression from myf5 to myod expression.

Two hypotheses could explain the lack of myod expression in Fgf4-injected tbxta mutants. First, despite the apparent lack of Tbxta protein in adaxial cells (Ochi et al., 2008; Schulte-Merker et al., 1994a), Tbxta might act directly on myod. Alternatively, Fgf-driven Tbxta activity might act indirectly in pre-adaxial cells to upregulate Tbx16 and thereby drive myod expression. We therefore examined the ability of Fgf4 to upregulate Tbx16 in tbxta mutants. Fgf4 enhanced tbx16 mRNA throughout the posterior mesoderm in siblings, with the exception of the widened notochordal tissue that contained nuclear Tbxta protein and failed to upregulate MRFs (Figs 7D and 8D). In tbxta mutants, Fgf4 also enhanced tbx16 mRNA in the ventral mesoderm, but a broader dorsal region did not express tbx16. Moreover, the level of tbx16 mRNA appeared to be lower than in siblings (Fig. 7D). Thus, Fgf4-injected tbxta mutants lack both Tbxta and Tbx16 upregulation in pre-adaxial cells. We conclude that, whereas Fgf-driven induction of lateral myogenic tissue requires Tbx16 but not Tbxta, induction of pre-adaxial character (marked by upregulated myf5 and myod mRNAs) requires both Tbx genes.

Finally, we examined the wider effect of Fgf signalling when skeletal muscle cannot form in the absence of tbx16 function. Excess Fgf action in embryos causes gross patterning defects (Kimelman and Kirschner, 1987; Slack et al., 1987). When Fgf4 was overexpressed in tbx16 heterozygote incross lays, some anterior mesodermal tissue formed and head tissues, such as eye and brain, were apparent, but trunk and tail mesoderm was grossly deficient (Fig. 8A). In siblings overexpressing Fgf4, some muscle was formed and truncated embryos were observed to twitch at 24 hpf. In contrast, no muscle was detected in tbx16 mutants overexpressing Fgf4 (Fig. 8B). Moreover, the residual expression of mutant tbx16 mRNA at 90% epiboly observed in uninjected embryos was lost upon Fgf4 overexpression (Fig. 8C). This suggests that the cells with tailbud character normally accumulating in tbx16 mutants were missing. Instead, widespread expression of Tbxta protein in nuclei far from the germ ring suggested that the entire posterior (but not anterior) mesoderm had converted to notochord, the most dorsal posterior mesoderm fate (Fig. 8D). Indeed, shha mRNA, a marker of notochord, was found to be broadly but weakly upregulated around the embryo in tbx16 mutant embryos injected with fgf4 mRNA, but not in their siblings (Fig. 8E). The data suggest that Fgf4 drives early involution of all posterior mesoderm precursors, leaving none to form a tailbud. In the presence of Tbx16, Fgf4 also dorsalises most involuted trunk mesoderm to muscle, whereas in the absence of Tbx16 Fgf4 converts most of the trunk mesoderm to notochord precursors.

The current work contains four main findings. First, that Tbx16 directly binds and activates myf5 regulatory elements to initiate skeletal myogenesis. Second, that Fgf signalling acts through Tbx16 to drive the initial myogenic events in the adaxial cell lineage, which subsequently require Hh signalling to complete myogenesis. Third, that Tbxta, the dorsal-most/posterior Tbx factor, binds directly to myod regulatory elements and also promotes dorsal midline expression of Fgfs, which subsequently cooperate to drive dorsal myogenesis through Tbx16. Fourth, that Fgf action through Tbx16 suppresses the dorsoposterior axial fate induced by Tbxta. Overall, Tbx transcription factors provide a crucial link between mesoderm induction and the initiation of myogenesis, which has profound implications for understanding the evolution of vertebrates.

Building on previous evidence that Tbx16 upregulates myf5 expression (Garnett et al., 2009; Mueller et al., 2010), our findings show that myf5 and myod genes are direct targets of Tbx16. We also present evidence that myf5 and myod are direct targets of Tbxta. As MRF gene activity drives commitment to skeletal myogenesis in vertebrates, our findings place Tbx protein activity at the base of skeletal myogenesis in zebrafish (Fig. S8).

Before myogenesis, Tbx16 is required for migration of most trunk PSM cells away from the ‘maturation zone’ immediately after their involution (Griffin and Kimelman, 2002; Row et al., 2011). Our analysis of aplnrb-expressing mesodermal cells shows that most anterior (i.e. head) and posterior ventral (i.e. ventral trunk) mesoderm involution and migration occurs normally in both tbx16 and tbxta mutants. Indeed, some PSM eventually yielding muscle is formed in tbx16 mutant trunk (Amacher et al., 2002; Kimmel et al., 1989). PSM is more severely lacking in tbx16;tbxta or tbx16;tbx16l double mutants (Amacher et al., 2002; Griffin et al., 1998; Morrow et al., 2017; Nelson et al., 2017) or after Tbxtb knockdown in the tbxta mutant (Martin and Kimelman, 2008). Cooperation of Tbx proteins in PSM formation also occurs in Xenopus (Gentsch et al., 2013). It is likely, therefore, that all PSM formation and its accompanying myf5 expression requires Tbx proteins, which may help explain why tbx16 mutants have increased pronephric mesoderm (Warga et al., 2013). As Tbx16 is required for direct induction of myf5 and for pcdh8, msgn1, mespaa and tbx6 expression in PSM (Fior et al., 2012; Goering et al., 2003; Griffin and Kimelman, 2002; Lee et al., 2009; Morrow et al., 2017; Yamamoto et al., 1998), the data support previous proposals (Amacher and Kimmel, 1998; Griffin and Kimelman, 2002) of a role for Tbx16 in promotion of the earliest step in PSM formation, en route to myogenesis. These early actions of Tbx16 and Tbxta proteins have previously masked their direct myogenic actions in mutants.

Our data argue that once posterior (i.e. trunk) mesoderm forms, Tbx proteins are still required for MRF expression and normal myogenesis. Hh signalling from notochord acts to maintain adaxial MRF expression in wild type and, if Tbx-driven initiation fails, Hh can initiate myod and upregulate myf5 expression, thereby driving slow myogenesis (Blagden et al., 1997; Coutelle et al., 2001; Du et al., 1997). When Hh signalling is prevented, both Tbx16 and Tbxta are essential for initial pre-adaxial myod transcription. Conversely, Tbx16-induced ectopic myod expression is restricted to a narrower mesodermal region flanking the pre-adaxial cells, likely due to the restricted expression of smarcd3 (Baf60c) in this region (Ochi et al., 2008). Nevertheless, myod is induced by Tbx16 in the absence of Myf5, probably through direct binding to regulatory elements in the myod locus. It is possible that changes in chromatin structure in myf5 and myod loci accompanying posterior mesoderm formation facilitate Tbx16 access to its binding sites in these MRF genes.

Adaxial slow and paraxial fast myogenesis differ (Blagden et al., 1997; Devoto et al., 1996). Paraxial PSM expresses myf5 after involution, which requires Tbx16, but not Tbxta. However, fast myogenesis is delayed until somites form, perhaps through Tbx6 action (Windner et al., 2015). In contrast, pre-adaxial cells that generate slow muscle within PSM require both Tbxta and Tbx16 function for upregulation of myf5 and initiation of myod expression. Thus, distinct Tbx proteins are required for normal adaxial and paraxial myogenesis.

Residual muscle in tbx16 mutants is likely driven by Tbx16l (Morrow et al., 2017). However, as tbx16;tbx16l double mutants continue myod mRNA expression at a reduced level throughout the axis at 24 hpf (Morrow et al., 2017), we predict this is in adaxially derived slow muscle induced by Hh signalling.

To identify likely MRF enhancers at 80% epiboly, we have largely restricted our focus to robust Tbx16 and Tbxta ChIP-seq peaks that are co-incident with established histone marks indicative of active enhancers, H3K4me1 and H3K27ac. As many transcription factor-binding events may be non-functional, not all enhancers have H3K27ac (Pradeepa, 2017; Pradeepa et al., 2016) and a minority of embryonic cells are myogenic, additional Tbx16 and Tbxta ChIP-seq peaks beyond 5DE, 5PE1 and DDE3 may mark functionally important enhancers regulating myf5 and myod. Of particular note DDE1, −31 kb upstream of myod, with Tbx16 and Tbxta ChIP-seq peaks that colocalise with significant H3K4me1, may be functionally important.

Understanding of the elements driving specific aspects of zebrafish myf5 expression is limited. As with murine Myf5 (Buckingham and Rigby, 2014), our data suggest that elements ∼80 kb upstream of the transcription start site are required for myf5 expression. A BAC transgenic encompassing 5DE drives GFP expression in muscle, although analysis of shorter constructs has been confounded by cloning artefacts (Chen et al., 2001, 2007). We also observe Tbx-binding peaks far upstream of zebrafish myod. Upstream elements are known to initiate murine Myod (Myod1) expression in some embryonic regions, but whether these elements drive the earliest myotomal regulation of Myod is unknown (Chen and Goldhamer, 2004). Mouse Tbx6 and Tbxt family genes are also required for trunk/tail, but not cranial, myogenesis (Nandkishore et al., 2018). Moreover, other Tbx proteins bind similar DNA motifs and are required to pattern cranial, cardiac and limb muscle (Don et al., 2016; Knight et al., 2008; Lu et al., 2017; Papaioannou, 2014; Valasek et al., 2011). The extent to which Tbx genes act through similar binding elements to initiate MRF expression and myogenesis across vertebrates remains to be determined.

Tbx16 is required for Fgf signalling to induce myf5 (Fig. 5A, Fig. S8). In its absence, Fgf drives all posterior mesoderm towards a notochord-like fate (Fig. 8D,E), probably via activation of Tbxta. Fgf signalling is required for expression of myf5 in tailbud paraxial PSM (Groves et al., 2005). Here, we show that Fgf is also required for the earliest myf5 expression in involuting trunk mesoderm and for the initiation of myf5 and myod expression in pre-adaxial cells destined to form the slow muscle of anterior somites. Our data provide further evidence that this MRF expression is subsequently maintained by Hh signalling, as the shield/tailbud-derived sources of Fgf recede from the adaxial cells (Coutelle et al., 2001; Osborn et al., 2011) (Fig. S8). We find that Fgf action is stronger in trunk (as opposed to tail) somites, consistent with (1) dwindling Fgf mRNA levels as tailbud outgrowth slows and (2) the unresolved issue of Hh-independent initiation of MRF expression in the anterior-most somites of murine smoothened mutants (Zhang et al., 2001). We suggest this MRF expression is triggered by Fgf in mouse, as in zebrafish.

Our data argue strongly that Fgf signalling not only promotes tbx16 expression, but also enhances the activity of the Tbx16 protein, constituting a feed-forward mechanism. The MRF-inducing activity of Tbx16 is suppressed by inhibition of Fgf signalling (Fig. 5C,D). Bearing in mind the existence of Tbx16l and Tbxta, this result is consistent with the finding that tbxta or tbx16 mutation sensitises embryos to Fgf inhibition (Griffin and Kimelman, 2003). As Tbx16 overexpression can expand PSM fates and reverse the effect of partial Fgf inhibition, a primary role of Fgf signalling is to cooperate with Tbx16 to drive expression of its target genes, including myf5. This understanding provides mechanistic insight into how the effects of Fgf on gastrulation movements and histogenesis are separated, as originally proposed (Amaya et al., 1993). Interestingly, Tbx16 overexpression rescues myf5 mRNA preferentially on the dorsal side of the embryo, suggesting that BMP and/or other signals continue to suppress PSM fates ventrally, and thus that Tbx16 does not act by simply suppressing the inhibitory effect of such signals.

Tbxta and Fgf appear to act in a positive-feedback loop to both maintain tailbud character and diversify PSM into pre-adaxial and paraxial. Tbxta is required for normal Fgf signalling from the midline to promote pre-adaxial myogenesis. Fgf overexpression drives myf5 in tbxta mutants, but cannot drive myod. ChIP shows sites upstream of myod preferentially bound by Tbxta. As Tbxta is not readily immunodetectable in these pre-adaxial cells (Hammerschmidt and Nüsslein-Volhard, 1993; Odenthal et al., 1996; Schulte-Merker et al., 1994a), Tbxta may either open the myod locus, act indirectly, or remain tightly bound despite low free concentration. In the absence of Tbx16, Fgf drives the entire germ ring towards notochord fate, preventing continued tailbud outgrowth. It seems, therefore, that Fgf/Tbx16 interaction is required to maintain tailbud stem cells and promote PSM formation. In a chordate ancestor, the evolution of interaction between Fgf/Tbx6/16-dependent muscle-forming tissue and Fgf/Tbxta-dependent dorsal organiser might be a key innovation leading to both tailbud stem cells and notochord.

As efficient motility driven by sarcomeric muscle is found throughout triploblasts, it is likely that mesodermal striated muscle existed in the common ancestor of deuterostomes and protostomes. There is consensus that the appearance of neural crest, notochord and a post-anal tail were significant evolutionary steps for chordates (Gee, 2018). Already in cephalochordates at least two kinds of striated muscle had evolved in anterior somites (Devoto et al., 2006; Lacalli, 2002). Our evidence that initiation of both slow and fast myogenesis in the most anterior trunk is driven by Fgf/Tbx signalling indicates that a major function of this early mesodermal inducer was induction of trunk striated myogenesis, which may constitute an ancestral chordate character. Once Hh is expressed in maturing midline tissues, it triggers terminal differentiation of muscle precursors into functional muscle through a positive-feedback loop (Coutelle et al., 2001; Osborn et al., 2011) (Fig. S8). Parallel diversification of neural tube cells, also regulated by Hh (Placzek and Briscoe, 2018), may have generated matching motoneural and muscle fibre populations that enhanced organismal motility. With the evolution of a tbxta-dependent tailbud destined to make the post-anal tail and notochord, our data suggest that weakening Fgf signalling continued to induce myf5 expression and paraxial mesoderm character through tbx16, but was insufficient to induce adaxial myogenesis. The presence of Hh and Tbxta plus Tbx16, however, ensure that adaxial slow muscle is initiated in the zebrafish tail.

In the anterior somites of amniotes, as in zebrafish, Hh signalling maintains, rather than initiates, myf5 expression (Zhang et al., 2001). In more posterior somites of zebrafish, Xenopus and amniotes, Hh signalling drives myf5 initiation (Borycki et al., 1999; Grimaldi et al., 2004). Compared with zebrafish, however, murine Myf5 induction is further delayed until after somitogenesis, when Gli3 repressive signals in PSM have diminished (McDermott et al., 2005). In mouse, Tbxt (known as brachyury) is required for myogenesis and binds 20 kb downstream of Myod, but does not obviously control its expression (Lolas et al., 2014). No clear orthologue of tbx16 exists in mammals, although it clusters by sequence with Tbx6 genes. Mammalian Tbx6 suppresses neurogenesis in posterior paraxial mesoderm, suggesting that additional mechanisms have evolved that suppress early pre-somitic Myf5 expression (Chapman and Papaioannou, 1998). Indeed, possible low level Myf5 expression in PSM has long been a source of controversy (George-Weinstein et al., 1996; Gerhart et al., 2004). Thus, there has been diversification in how these Tbx genes regulate somitic myogenesis.

The ancestral situation seems clearer. In amphioxus, Tbx6/16 is expressed in tailbud and PSM (Belgacem et al., 2011). In Ciona, knockdown of Tbx6b/c/d leads to reduced MyoD expression, loss of muscle and paralysis (Imai et al., 2006). In Xenopus, both Tbx6 and VegT are implicated in early myogenesis (Callery et al., 2010; Fukuda et al., 2010; Tazumi et al., 2010), although some mechanisms may differ from those in zebrafish (Maguire et al., 2012). By adding our zebrafish findings, we show that in all major chordate groups Tbx-dependent gene regulation is central to skeletal myogenesis. The conserved involvement, yet divergent detail, of how Tbx16, Tbx6 and Tbxt genes regulate somitic myogenic diversity along the body axis are consistent with selective pressures on these duplicated Tbx gene families playing a key role in the diversification of myogenesis in the vertebrate trunk and tail, characters that gave these chordates their predatory advantage.

Mutant lines fgf8a^ti282a (Reifers et al., 1998), notoⁿ¹ (Talbot et al., 1995), smo^b641 (Barresi et al., 2000), tbxta^b195 and tbx16^b104 (Griffin et al., 1998) are likely nulls and were maintained on King's wild-type background. Staging and husbandry were as described previously (Westerfield, 2000). All experiments were performed under licences awarded under the UK Animal (Scientific Procedures) Act 1986 and subsequent modifications.

In situ mRNA hybridisation (ISH) for myf5 and myod was performed as described previously (Hinits et al., 2009). Additional probes were: fgf3 (Maroon et al., 2002), fgf4 (IMAGE: 6790533, https://www.ncbi.nlm.nih.gov/nuccore/BC128817), fgf6a (Thisse and Thisse, 2005), fgf8a (Reifers et al., 1998), tbxta (Schulte-Merker et al., 1994b) and tbx16 (Griffin et al., 1998; Ruvinsky et al., 1998). Anti-Ntla immunostaining was performed after ISH using rabbit anti-Ntla (Schulte-Merker et al., 1992; 1:2000) and goat anti-rabbit IgG-HRP-conjugated secondary antibody (PI-1000-1, Vector Laboratories).

Embryos were injected with MOs (GeneTools LLC) as indicated in Table S2 to fgf3, fgf4, fgf6a, fgf8a, tbxta (Feldman and Stemple, 2001) or tbx16 (Bisgrove et al., 2005). Controls were vehicle or irrelevant mismatch MO. Cyclopamine (100 µM in embryo medium), SU5402 (at the indicted concentration in embryo medium) and vehicle control were added at 30% epiboly to embryos whose chorions had been punctured with a 30G hypodermic needle. A PCR product of fgf4 (IMAGE: 6790533) was cloned (primers in Table S2) into the SacI/SalI sites of pβUT3 to make mRNA for overexpression. One-cell-stage embryos were injected with 100-220 pg fgf4 mRNA (made with messageMachine), 50 pg fgf6a mRNA, 150 pg tbx16 mRNA (Griffin et al., 1998) or 150 pg tbx16-GR mRNA (Jahangiri et al., 2012). For hormone-inducible Tbx16 activation, mRNA corresponding to Tbx16 fused to the hormone-binding domain of glucocorticoid receptor (GR) was overexpressed in wild-type embryos. The resulting protein is held in the cytoplasm until DEX from shield stage stimulates GR nuclear translocation. In the presence of the translation inhibitor CHD (added prior to DEX at the germ ring stage), increased nuclear Tbx16 is expected to induce only direct targets of Tbx16 in a mosaic fashion. Embryos were treated with 10 μg/ml final concentration of CHD 2 h prior to collection at 75-80% epiboly. CHD caused ∼5% delay in epiboly, showing that it was active. After 30 min, 20 μM DEX was added for the remaining 1.5 h.

ChIP-seq data from GSE84619 and GSE32483 were analysed as reported previously (Nelson et al., 2017). Multiz Alignments & Conservation from UCSC Genome Browser (Haeussler et al., 2019) are shown beneath schematics. ChIP-seq peak height was measured in reads per million reads (RPM). ChIP-qPCR experiments were performed as previously reported (Jahangiri et al., 2012) using the primers in Table S2.

We are grateful to many members of the Hughes lab for advice and to Bruno Correia da Silva and his staff for care of the fish.