Muscle precursor cell movements in zebrafish are dynamic and require Six family genes

Long-range cell migrations are vital to producing the vertebrate body plan. Migratory muscle precursors (MMPs) are specified in the trunk and then migrate to populate hypaxial skeletal muscles including the limb, tongue, neck, chest and diaphragm (Vasyutina and Birchmeier, 2006). We refer to these collectively as ‘MMP-derived muscles’. Mammalian MMPs are specified within the trunk dermomyotome by transcription factors, including Pax3, Six1, Six4, and the Hox genes (Alvares et al., 2003; Bober et al., 1994; Grifone et al., 2005). After specification, MMPs detach from the somite and migrate to their destinations. The limb bud attracts MMPs (Chevallier et al., 1977; Christ and Brand-Saberi, 2004; Christ et al., 1977; Dietrich et al., 1998; Hayashi and Ozawa, 1995) and is required for formation of pectoral girdle and pelvic girdle muscles (Evans et al., 2006; Masyuk et al., 2014; Valasek et al., 2011). Other cues may guide MMPs to non-limb bud locations; for example, occipital MMPs generate a subset of neck and tongue muscles (Dietrich, 1999). The occipital MMPs form far from the limb bud in somites 1-5 (Adachi et al., 2018; Huang et al., 1999; Lours-Calet et al., 2014; Mackenzie et al., 1998; Parada et al., 2012) and do not respond to several molecules that guide limb-level MMPs (Lours-Calet et al., 2014). In zebrafish embryos, three distinct MMP streams originate in somites 1-5, with the second stream populating the fin bud (Haines et al., 2004; Minchin et al., 2013; Neyt et al., 2000). Because all zebrafish MMPs originate near the fin bud, it was unclear whether zebrafish MMPs are guided primarily by fin bud-derived signals, by fin bud-independent mechanisms, or both. Here, we investigate the molecular mechanisms that specify MMPs, facilitate MMP motility, and organize MMP streams in zebrafish.

In mammals, two homeobox genes, Six1 and Six4, are required for formation of the hypaxial dermomyotomal lip, the structure that generates MMPs, and are broadly required for MMP-derived muscle formation (Grifone et al., 2005, 2007; Laclef et al., 2003; Ozaki et al., 2001). Six1 and Six4 are expressed in all muscle precursors and are required for expression of at least three MMP-specification genes: Met, Pax3 and Lbx1 (Grifone et al., 2005). Lbx1 encodes a homeodomain protein present in amniote MMPs (Dietrich, 1999; Mennerich et al., 1998) that is required for MMP migration into limbs and for tongue muscle size (Brohmann et al., 2000; Gross et al., 2000; Masselink et al., 2017; Schäfer and Braun, 1999). Pax3 is a paired homeobox-containing gene required for MMP migration in the mouse (Bober et al., 1994; Daston et al., 1996; Relaix, 2004). Met is a cell surface receptor expressed in MMPs that is essential for mammalian MMP migration (Bladt et al., 1995; Dietrich et al., 1999). The Met ligand, Hgf, is expressed along MMP migratory routes and is essential for MMP migration in mammals (Bladt et al., 1995; Dietrich et al., 1999; Scaal et al., 1999). Thus, Six1 and Six4 are upstream of several genes vital to MMP migration. However, because mammalian Six1 and Six4 function is required to generate the tissue that produces MMPs (Grifone et al., 2005, 2007), it remained unclear whether these genes function in MMPs during migration.

MMP-derived muscle defects are observed in zebrafish injected with a six1b morpholino (MO) (Lin et al., 2009; Nord et al., 2013), met MO (Haines et al., 2004), Hgf antibody (Haines et al., 2004), lbx1a MO, lbx2 MO (Ochi and Westerfield, 2009), or a combination of both pax3a and pax3b MOs (Minchin et al., 2013). These phenotypes resemble those in mouse knockouts and suggest that MMP specification pathways are conserved among bone-forming vertebrates. Early limb development is also conserved between fish and amniotes (Mercader, 2007). Fin bud specification in zebrafish begins during gastrulation, when retinoic acid (RA) induces the fin field (Grandel et al., 2002), a structure that will later develop into the fin bud. RA signaling indirectly activates Fgf signaling in the fin field, leading to fin bud formation (Cunningham et al., 2013; Gibert et al., 2006; Grandel and Brand, 2011). Similarly, in the mouse, RA and Fgf together induce the limb bud (Cunningham et al., 2013; Zhao et al., 2009). Within the zebrafish fin bud, Fgf signaling acts again to generate the fin apical ectodermal ridge, leading to fin bud outgrowth (Masselink et al., 2016; Mercader et al., 2006). RA is also required for sonic hedgehog (shh) expression in the fin field (Alexa et al., 2009). Hedgehog is required for fin bud patterning, but not for fin bud specification (Ahn and Ho, 2008; Neumann et al., 1999), similar to Hedgehog function in mammalian limb development (Robert and Lallemand, 2006). Thus, zebrafish appear to have well-conserved MMP-specification and fin-specification pathways, making them an excellent model to investigate how the fin bud influences MMP migration.

Our study focuses on the formation of four MMP-derived muscles in zebrafish. These muscles are, in anterior-to-posterior order, the sternohyoideus muscle (SHM), the abductor pectoral fin muscle (AbFM), the adductor pectoral fin muscle (AdFM) and the posterior hypaxial muscle (PHM) (Haines et al., 2004; Masselink et al., 2016; Minchin et al., 2013; Neyt et al., 2000; Ochi and Westerfield, 2009). These four muscles are predominantly composed of fast-twitch muscle fibers (Minchin et al., 2013; Patterson et al., 2008). The SHM is homologous to hypobranchial muscles in amniotes, which include the tongue and some neck muscles (Lours-Calet et al., 2014; Minchin et al., 2013; Okamoto et al., 2017). The muscles that lift (AbFM) and lower (AdFM) the pectoral fin are homologous to mammalian forelimb muscles (Mercader, 2007). The PHM is an early-forming ventral body wall muscle, located immediately posterior to the fin and positioned like a chest muscle (Haines et al., 2004; Windner et al., 2011). These four MMP-derived muscles form via three seemingly separate MMP streams (Haines et al., 2004; Masselink et al., 2016; Minchin et al., 2013; Neyt et al., 2000). Previous fate-mapping experiments indicated that somites 1-3 contribute to the SHM, somites 2-4 contribute to fin muscles, and somites 5 and 6 contribute to the PHM (Haines et al., 2004; Hollway and Currie, 2003; Li et al., 2017; Minchin et al., 2013; Neyt et al., 2000). In this work, we use high-resolution imaging to refine and clarify existing zebrafish MMP fate maps and migratory routes.

In this study, we investigate zebrafish MMP migration in vivo. As expected, the anterior and posterior MMP streams contribute to the SHM and PHM, respectively. We find that the middle (second) stream, previously shown to contribute to the AbFM and AdFM fin muscles (Haines et al., 2004; Masselink et al., 2016; Neyt et al., 2000), also contributes to the PHM. When we block fin bud formation, the precursors that would normally form fin muscles instead shift posteriorly and contribute to the PHM. The SHM and PHM streams eventually form even when fin field development is blocked, revealing that they do not require the fin bud. We show that the zebrafish homologs of Six1 and Six4 (six1a, six1b, six4a and six4b) are redundantly required for cell motility in all MMP streams, acting upstream of met. The six1/six4 genes and met also function in the fin bud, where they are required to partition fin bud MMPs into discrete AbFM and AdFM populations. Together, these findings demonstrate a conserved pathway for MMP specification, reveal flexibility in MMP movements, and show how a contiguous precursor population in anterior somites generates four separate muscles.

Although previous studies have investigated MMP migration in zebrafish, none examined how the streams form and interact with one another during migration and differentiation. To visualize muscle migration at high resolution, we generated a BAC transgenic line TgBAC(six1b:lyn-GFP)oz5 (hereafter: six1b:lyn-GFP), which expresses membrane-tethered GFP in muscle precursors and mature muscle cells (Fig. 1). This line also labels other six1b-expressing tissues, including the ear and lateral line (Fig. 1, Fig. S1). To distinguish myoblast precursors from maturing fast muscle fibers, we combined the six1b:lyn-GFP transgene with a fast muscle-specific transgene, Tg(mylpfa:mCherry)cz3327 (hereafter: mylpfa:mCherry) (Ignatius et al., 2012). Using these lines, we followed MMP migration and differentiation in fixed samples (Fig. 1A-D′) and in time-lapse of live embryos (Fig. 1E-I′, Movies 1-3). The three MMP streams form between 24 and 36 hours post-fertilization (hpf) (Fig. 1A-B′). As expected, the first stream, emanating from somites 1-3, generates the SHM (Fig. 1A-E, Movie 1). The second stream, which emanates primarily from somite 4, generates the AbFM and AdFM pectoral fin muscles (Fig. 1A-E, Movies 1-3). At 24 hpf, a few cells that will give rise to the AbFM are adjacent to somite 3 (Movie 1); it is presently unclear whether these cells originated in somite 3 or whether they originated in somite 4 and had already migrated adjacent to somite 3. After populating the fin bud, the remaining second stream MMPs alter their course and contribute to the anterior edge of the PHM (Fig. 1E′-I′, Movies 1, 2). Individual MMPs from somite 4 sometimes generate progeny that contribute to both the AdFM and PHM (green lines in Fig. 1E′, green dots in Movies 1, 2). The third stream, which emanates from somite 5, will form the posterior portion of the PHM (Fig. 1E-I′, Movie 1). Later, somite 6 produces ventral extensions that connect the posterior end of the PHM to the trunk (Fig. 1D,D′, Movie 1). The dynamic direction of MMP movement suggests that MMPs may respond to guidance cues during migration. At high resolution, long filopodia extend and retract from every MMP (Movie 3) and resemble filopodia that are specialized to detect guidance cues (Fairchild and Barna, 2014). These filopodia make contact between adjacent streams (Movie 3), suggesting that MMP streams coordinate their movements. Thus, our findings map MMP migration and MMP-derived muscle formation at cellular resolution, suggest that cells are actively guided to their destinations, and reveal that somite 4 MMPs contribute to both fin and PHM.

To begin investigating the genetic basis of MMP migration, we re-examined the function of Six family genes. Morpholino-knockdown of six1b causes loss of MMP-derived muscles (Lin et al., 2009; Nord et al., 2013). Surprisingly, when we characterized six1b mutants, we found no muscle defects (Fig. 2A, Fig. S2). We are confident that the zebrafish six1b lesions are null, because the frameshifting allele (six1b^oz1) and the deletion allele that eliminates the coding sequence (Δsix1b^oz34) give the same phenotype. The six1b mutants show ear cristae defects that resemble cristae defects in the Six1 hypomorphic mouse (Bosman et al., 2009), suggesting that six1b null alleles only partially eliminate total Six1 function. Embryos double mutant for both zebrafish six1 genes (six1a^−/−;six1b^−/−) have stronger ear defects and partial loss of MMP-derived muscles (Fig. 2A-C, Fig. S2), consistent with phenotypes in the mouse Six1 knockout (Laclef et al., 2003; Zhang et al., 2017). MMP-derived muscles are lost in the mouse Six1;Six4 double mutant (Grifone et al., 2005), so we knocked out all four zebrafish six1 and six4 genes (six1a, six1b, six4a and six4b) (Fig. 2A). Through combinatorial mutant analysis, we determined that all four genes participate in MMP patterning (Fig. 2A). We mainly used deletion alleles that remove the two bi-gene clusters: six1a-six4a on chromosome 13 and six1b-six4b on chromosome 20 (Fig. 2D). These deficiency alleles are officially named Df(Chr13:six1a,six4a)^oz27 and Df(Chr20:six1b,six4b)^oz16, and as Δsix1a;4a and Δsix1b;4b for simplicity. Δsix1a;4a;Δsix1b;4b mutant embryos lack nearly all MMP-derived muscle fibers (Fig. 2E,F). The AbFM, AdFM and SHM muscles are consistently absent and the PHM muscle is consistently reduced to a few short muscle fibers (Fig. 2E-G). Despite the severe MMP-derived muscle defects, larval trunk muscle appears normal in Δsix1a;4a;Δsix1b;4b mutants (Fig. 2F, Fig. S3A-L). Δsix1b;4b mutants and Δsix1a;4a;Δsix1b;4b mutants also have severe ear and lateral line defects (Fig. 2A,F, Fig. S3E-T), consistent with sensory defects in the mouse Six1;Six4 mutant (Moody and LaMantia, 2015). These findings indicate that the redundant requirement for Six1 and Six4 family gene function in MMP-derived muscles and sensory tissues is conserved in zebrafish and mouse.

The severe Δsix1a;4a;Δsix1b;4b mutant MMP-derived muscle defect suggests that six1/six4 genes are required to generate normal MMPs. However, at the beginning of MMP migration (24 hpf), cells closely resembling MMPs are seen in Δsix1a;4a;Δsix1b;4b mutants (Fig. 3A,B), in the zebrafish equivalent of the hypaxial dermomyotomal lip (Devoto et al., 2006; Feng et al., 2006; Hammond et al., 2007; Stellabotte and Devoto, 2007). High-resolution time-lapse imaging at 30 hpf indicates that these cells have active filopodia but do not migrate (Movie 4). Consistent with a role for six1 and six4 genes in cell migration, we note that migration of the primary lateral line (PrimI) fails in Δsix1b;4b mutants and in Δsix1a;4a;Δsix1b;4b mutants (Movies 4, 5). As Δsix1a;4a;Δsix1b;4b mutants have MMP-like cells that extend filopodia, yet do not migrate, we hypothesize that these cells are only partially specified. The Δsix1a;4a;Δsix1b;4b mutant MMP-like cells also appear to have impaired cell proliferation and survival (Movies 4, 5), which likely contributes to the final phenotype, near-complete absence of MMP-derived muscle. In the mouse, Six1 and Six4 are redundantly required for activation of several MMP-specific genes, including Met, Pax3 and Lbx1 (Grifone et al., 2005). At the beginning of migration, zebrafish MMPs express pax3a, pax3b and lbx2 (Minchin et al., 2013; Ochi and Westerfield, 2009). In contrast, lbx1a and met expression is restricted to an MMP subset, with lbx1a expressed in MMPs at somites 1-4 and met expressed in MMPs at somites 4-5 (Fig. 3C-L, Fig. S4A-J). By 36 hpf, met is also expressed by MMPs in the first migratory stream (Figs S4M, S5), which emanates from somites 1-3. In Δsix1a;4a;Δsix1b;4b mutant embryos, MMP-like cells express lbx2, pax3a and pax3b, but not met or lbx1a (Fig. 3C-L, Fig. S4A-J), even at later times when MMP migration would normally be well underway (36 hpf) (Fig. S4K-N). Thus, we propose that zebrafish six1 and six4 genes stimulate precursor migration by activating a subset of MMP genes.

We hypothesized that Δsix1a;4a;Δsix1b;4b mutant phenotypes may be explained at least in part by the loss of met expression, consistent with previous mouse Met mutant and zebrafish met knockdown analyses (Bladt et al., 1995; Haines et al., 2004). To block Met function during MMP migration, but after MMP formation, we treated embryos with a potent and specific Met protein inhibitor, SGX523 (Buchanan et al., 2009), at the onset of MMP streaming (24 hpf). SGX523 treatment causes delayed MMP streaming (Fig. 4A-D, Movie 6), and reduced size of all MMP-derived muscles, especially the AbFM (Fig. 4E,F). We also generated a frame-shifting met allele, and show that, like in SGX523-treated embryos, met mutant MMP-derived muscles are shortened (Fig. 4G-K) and the AbFM is especially affected (Fig. 4G,H inset, and K). Similarly, by examining an ‘allelic series’ of Δsix1a;six4a and Δsix1b;six4b heterozygous and mutant combinations, we find that progressive loss of six1 and six4 genes causes delayed MMP streaming, reduced size of all three MMP streams, and reduced size of all MMP-derived muscles (Fig. 5, Figs S2, S5, Movie 5). Furthermore, like met mutants, the AbFM is more sensitive to partial loss of six1/six4 gene function than the AdFM (Fig. 5K,L,P, Fig. S2D,H). These findings support our model that six1 and six4 genes activate MMP migration in part by activating met expression.

Like AbFM/AdFM separation defects in met mutant zebrafish, limb abductor and adductor streams fail to partition in the Smo mutant mouse (Anderson et al., 2012). To test whether Smoothened function is required for AbFM/AdFM separation in fish, we treated embryos with the Smoothened inhibitor cyclopamine beginning at the onset of MMP migration, 24 hpf. During normal development, MMPs have begun to partition within the fin bud by 36 hpf (Fig. 6A); however, partitioning is impaired in cyclopamine-treated embryos (Fig. 6B, Movie 7). The fin muscle defect becomes more pronounced by 48 hpf in cyclopamine-treated embryos, in which AbFM precursors are reduced compared with AdFM precursors (Fig. 6C,D), similar to met mutants. We thus reasoned that Smoothened might control AbFM/AdFM separation by activating Hgf/Met signaling. Consistent with this idea, we find that the fin buds that form in cyclopamine-treated embryos lack expression of hgfa (Fig. 6E,F) at 36 hpf. Thus, the AbFM/AdFM separation defect in Smoothened-inhibited embryos may be explained by a fin bud-specific loss of Hgf/Met signaling.

MMPs enter the fin bud in met mutant embryos and Smoothened-inhibited embryos (Fig. 6), suggesting that other molecules, such as Fgfs, guide and/or recruit MMPs to the fin bud. To investigate how the fin bud influences MMP movements, we blocked fin bud formation by treating embryos with the Fgf/Vegf inhibitor SU5402 at 24 hpf, after MMP formation. Even though fin bud formation is blocked, second stream MMPs initially extend anteriorly towards the area where the fin bud would normally form; later, they change direction and join the third stream (Fig. 7A-E, Movie 8). We narrowed the critical window for Fgf/Vegf function to 24-36 hpf (Fig. 7B,D,E), the period during which the fin bud usually forms and becomes populated with MMPs. To confirm that phenotypes after SU5402 treatment reflect loss of Fgf (not Vegf) signaling, we activated a heat-shock-inducible dominant negative Fgf receptor (dnFgfr) at 24 hpf using the Tg(hsp70l:dnFgfr1-GFP)pd1 transgenic line. As expected, dnFgfr induction at 24 hpf results in pectoral fin loss (Fig. S5A-C), loss of fin muscle, and an expanded PHM stream (Fig. 7F,G). By 72 hpf, the length of the PHM in wild-type and dnFgfr-expressing embryos is equivalent, but PHM girth is larger in wild-type than dnFgfr-expressing embryos (Fig. 7H,I). Similarly, by 72 hpf, the PHM of SU5402-treated embryos is narrower than controls (not shown). Thus, although the PHM stream transiently expands owing to addition of extra MMPs, this expansion does not increase the size of the PHM after myofiber formation. These findings show that Fgfr function is not required for MMP migration and suggest that the fin bud recruits early-migrating second stream MMPs, preventing them from contributing to the PHM.

Surprisingly, MMPs transiently visit the fin-forming region when we block fin bud development. We hypothesized that MMP movements might be influenced by the fin field, an early-forming precursor population that will later become the fin bud. To test this idea, we examined MMP migration in aldh1a2 mutants (Fig. 8), which are deficient in RA synthesis, lack pectoral fin fields, and ultimately fail to form pectoral fins (Fig. S5D, supplementary Materials and Methods; Grandel et al., 2002). At 36 and 48 hpf, six1b-expressing MMPs are present in two small streams in aldh1a2 mutants (Fig. 8A-D) and express the MMP markers lbx1a and met (Fig. S5E-H). The aldh1a2 mutant SHM- and PHM-progenitor streams grow slowly (Fig. 8C-H), and eventually give rise to small SHM and PHM muscles (Fig. 8G,H). The MMPs that emanate from somite 4 in aldh1a2 mutants, which in wild-type embryos would contribute to pectoral fin and PHM muscles, appear to contribute to the PHM. These findings support the hypothesis that the fin field is not required for MMP specification and show a role for aldh1a2 in MMP stream growth.

Our work has mapped MMP movements over time at single cell resolution in the living embryo and supports a new working model for the control of MMP migration during development. Using time-lapse analysis, we clarified MMP migratory routes. By examining the effect of the fin bud on MMP migratory routes, we identified flexibility between fin muscle and PHM fates. We show that embryos mutant for all six1/six4 genes (six1a, six1b, six4a and six4b) form partially specified MMP-like cells that are unable to migrate, indicating that MMP movements require six1/six4 function. Consistent with previous mammalian findings (Grifone et al., 2005), we show that six1/six4 genes are essential for expression of met and lbx1a. However, in contrast to mouse (Grifone et al., 2005), zebrafish pax3a and pax3b expression does not require six1/six4 function. As expected, the six1/six4 target gene met is essential for normal MMP migration. Although all MMP streams are reduced, some MMPs do, surprisingly, migrate into the fin bud in the zebrafish met mutant. Within the fin bud, met function is required for normal fin muscle formation, especially for the AbFM. Below, we integrate our findings with other studies to offer a new view of MMP migration.

Our time-lapse analyses reveal similarities and differences in some MMP stream origins and fates compared with previous fate-mapping studies that indicated that MMPs from somites 1-3 give rise to the SHM (Minchin et al., 2013), from somites 2-4 to fin muscle (Minchin et al., 2013; Neyt et al., 2000), and from somites 5-6 to the PHM (Neyt et al., 2000) (reviewed by Li et al., 2017). Using time-lapse imaging beginning just prior to MMP migration, we find that MMPs from somites 1-3 give rise to the SHM, somites 3-4 to fin muscle, somites 4-5 to the PHM, and somite 6 and possibly somite 7 contribute to the portion of PHM that connects to myotome either by short-range migration or myotomal extension. Thus, our time-lapse imaging is largely consistent with and extends previous fate-mapping studies, with some differences. For example, we did not observe MMPs from somite 2 cells contributing to fin muscle. Importantly, like Neyt et al. (2000), we confirm that MMPs from somite 4 are the major source of fin muscle. Notably, we find that the second MMP stream, which emanates from somite 4, contributes to three muscles. Second stream MMPs initially populate the AbFM and AdFM fin muscles whereas later-migrating second stream MMPs contribute to the PHM, a previously unrecognized contribution. The slight differences among previous and current work may be due to the different experimental methodologies employed. It is also possible that MMPs can originate in one somite but exit from another. Future work using 3D tracking of all cells in the streams may provide additional insight into MMP origins and cell dynamics during migration.

Using cell tracking, we observed that a single migrating MMP can generate daughters that contribute to different muscles (fin and PHM), showing that MMP fates are not fixed before migration. Thus, although MMP position is a strong indicator of future muscle fate, MMPs retain some plasticity. During normal development, somite 4 MMPs contribute to the fin muscles and to the PHM. However, when fin field or fin bud development is blocked genetically or with signaling inhibitors, all MMPs from the fourth somite default to the PHM fate. For example, in aldh1a2 mutants, which lack the fin field, somite 4 MMPs appear to contribute solely to the PHM. When we block fin bud formation late (e.g. by SU5402 treatment at 24 hpf), somite 4 MMPs move into the fin-forming region, but ultimately contribute to an expanded PHM stream. In wild-type embryos, in which somite 4 MMPs contribute first to the limb muscles and then to the PHM, this shift may be controlled by temporal cues rather than by absolute MMP cell number. We find that somite 4 MMPs shift their migration to the PHM at about the same time even in Δsix1b;4b mutant and Met-inhibited embryos, which populate the fin bud with fewer MMPs than in wild-type embryos. Thus, we propose that MMP guidance cues dynamically change with developmental time, first attracting MMPs to the fin bud and later inactivating or overriding those cues to enable PHM formation.

In addition to fin field- and fin bud-derived guidance cues, our findings also suggest that cell-intrinsic processes could influence MMP stream identity. Prior to migration, we find that lbx1a and met are expressed in a partially overlapping fashion that roughly corresponds to the three MMP streams (Fig. 3). In limb bud-deficient chicken embryos and fin field-deficient medaka embryos, limb-level Lbx1 is still expressed, indicating that the limb is not required for MMP specification (Alvares et al., 2003; Tani-Matsuhana et al., 2018). However, these studies did not test whether Lbx1-positive cells ultimately contribute to muscle in the absence of limb buds. We show that lbx1-positive, met-positive cells are present in zebrafish embryos that lack the fin field. We also show that these cells are true MMPs and that they migrate to produce SHM and PHM muscles. Similarly, in fin field-deficient tbx5a mutants, two MMP-derived muscles, the SHM and PHM, are present (Valasek et al., 2011). This idea is further supported by work in Xenopus laevis tadpoles, which form SHM-like and PHM-like muscles several days before limb bud formation (Martin and Harland, 2001, 2006). Because fin field-deficient zebrafish generate separate SHM and PHM streams, specification of these progenitor populations must not require the fin field. We hypothesize that before migration begins, cell-intrinsic processes distinguish the first stream from the second stream. This model explains why the second and third somites, which are close to the fin bud, generate MMPs that enter the first stream and populate the SHM, whereas the fin bud is populated by the second stream originating from the more distant somite 4. Further work, such as MMP transplantation between anterior somites, could be used to demonstrate the extent that intrinsic and extrinsic processes, and timing, control muscle formation.

It remains unclear which signals directly guide MMPs. Fgf has been described as a potential limb-derived guidance cue in some studies (Itoh et al., 1996; Masselink et al., 2017), but not others (Flanagan-Steet et al., 2000). We find that MMPs transiently visit the fin field when we block Fgf receptor function, indicating that signals other than Fgf draw MMPs toward the fin bud. As in mouse, our work implicates Hedgehog as a potential fin/limb-specific signal. When we inhibit the Hedgehog signal transducer Smoothened, the SHM, PHM and AdFM form, but AbFM formation is disrupted. In mammalian limbs, Hedgehog has also been proposed to guide MMPs and Smoothened is required for abductor and adductor stream separation (Anderson et al., 2012). We also find that the AbFM and AdFM fail to correctly partition in the fin bud of Smoothened-inhibited zebrafish embryos. In mouse, the adductor MMPs are particularly affected in Smo mutants, and distal muscle is lost (Anderson et al., 2012; Hu et al., 2012) whereas in zebrafish the abductor MMPs, which form distal to somites compared with the adductor muscle, are most affected. In both mouse and fish, the limb MMP populations that migrate most distally from the trunk are most sensitive to Smo loss, suggesting conserved function. The effect of Smoothened-inhibition could be explained by fin bud-specific loss of Hgf/Met signaling, or by effects outside the fin bud. Future experiments placing Hgf-soaked beads into the fin bud of Smoothened-inhibited embryos could test the effect of Hgf more directly. Another ligand/receptor pair, Sdf1 (Cxcl12)/Cxcr4_, is required to guide MMPs from the limb to the amniote chest (Griffin et al., 2010; Masyuk et al., 2014). Sdf1/Cxcr4 could also potentially control some MMP movements in zebrafish, such as the movement of second stream MMPs into the PHM. Another ligand/receptor pair EphrinA5/EphA4 has been proposed to repulse MMPs from non-myogenic portions of the limb in chick (Swartz et al., 2001). Future work will identify molecules that directly guide zebrafish MMPs to their destinations.

Our findings expand the role of Six1 and Six4 in vertebrate muscle formation. In mouse, Six1/Six4 function is necessary and sufficient for fast-twitch muscle specification (Grifone et al., 2004; Niro et al., 2010; Richard et al., 2011; Wu et al., 2013); and zebrafish six1a and six1b morpholino studies showed a similar result (Bessarab et al., 2008; O′Brien et al., 2014). However, we show here that although deletion of all four six1 and six4 genes (Δsix1a;4a;Δsix1b;4b mutants) severely affects MMPs, and thus fast-twitch MMP-derived muscles, fast muscle in the trunk forms normally. This observation suggests that other genes specify fast-twitch trunk muscles in zebrafish. Similar to mammalian studies (Grifone et al., 2005, 2007), we find that zebrafish six1/six4 genes are required for MMP migration, which happens well before fiber type differentiation. In zebrafish Δsix1a;4a;Δsix1b;4b mutants, MMP-like cells express pax3a, pax3b and lbx2, but not met or lbx1a, and thus are partially specified. Because Δsix1a;4a;Δsix1b;4b mutant MMP-like cells fail to migrate, we propose that zebrafish six1/six4 genes are needed to transition hypaxial muscle precursors from a non-migratory state to a migratory state (MMPs). We hypothesize that six1/six4 genes activate MMP motility in part by activating met, a gene thought to promote MMP motility (Birchmeier et al., 2003; Dietrich et al., 1999; Haines et al., 2004; present study), and in part by activating lbx1a. In the mouse, Lbx1 is required for MMP migration into the limb bud and for normal tongue muscle size (Schäfer and Braun, 1999). six1/six4 genes may also act upstream of other conserved targets (in addition to met and lbx1a) that are important for MMP migration, such as cell guidance receptors. Future work to identify differentially expressed genes in wild-type versus MMP-like cells sorted from Δsix1a;4a;Δsix1b;4b mutants may identify new genes that are vital for MMP migration in both fish and mammals.

Our findings are consistent with previous work implicating Met in cell motility, rather than cell guidance (Birchmeier et al., 2003; Dietrich et al., 1999; Haines et al., 2004; present study). We propose that Met activates MMP cell motility in somites, initiating stream movement and then acts again in the fin bud to promote AbFM/AdFM separation. Because the Met ligand gene hgfa is expressed in the trunk and fin bud (see Fig. 6E), rather than along the entire migratory route, our findings suggest that there are two phases of Met activity. Additionally, we find that all MMP-derived muscles are reduced in met mutants, with the AbFM particularly affected. Time-lapse analysis of Met-inhibited embryos shows delayed streaming of all MMPs, followed by a failure to partition the AbFM from the AdFM within the fin bud. These Met-inhibited phenotypes are surprisingly subtle because Met is considered to be absolutely essential for MMP migration in the mouse limb (Bladt et al., 1995). The subtle defects could be explained if the met^+13-3 lesion is not completely null, consistent with a previous study showing that met morphants have a more severe loss of hypaxial muscle (Haines et al., 2004) than what we observe in mutants. Despite the difference in severity, both mutants and morphants have partial loss of MMP-derived muscle (Haines et al., 2004; present study). Disparity between mutant and morphant phenotypes are sometimes explained by morpholino toxicity, maternal contribution, genetic compensation in the mutant, or other factors independent of whether the mutation is null (El-Brolosy et al., 2019; Kok et al., 2015; Rossi et al., 2015; Stainier et al., 2017). However, two lines of evidence suggest that the met^+13-3 allele we use in our study is a strong loss-of-function allele. First, the lesion truncates Met prior to the transmembrane and intracellular domains; second, met mutants appear identical to SGX523 Met inhibitor-treated embryos. We suggest that the anterior origins of zebrafish MMPs may explain why met may be only partially required for MMP motility. In mammals, MMPs in the most anterior (occipital) somites do not require several genes, including Met, that are essential for migration into limb buds (Lours-Calet et al., 2014). For instance, in the Met mutant mouse, MMP migration into the limb and diaphragm fails entirely, but Met mutant MMPs do migrate a short distance from the most anterior (occipital) somites to form a subset of neck and tongue muscles (Bladt et al., 1995; Lours-Calet et al., 2014). Thus, we propose that zebrafish met is necessary for some, but not all, MMP motility because the zebrafish MMPs arise from anterior somites and, like amniote occipital MMPs, they are capable of short-range migration independently of met function.

Our findings on zebrafish MMP development have broader implications. Because the fin bud recruits MMPs from somite 4, which also contributes to the PHM, we suggest that these fin/PHM MMPs are partially naïve of their destination when they exit the somite. Such flexibility in development may have provided MMPs with plasticity in form during evolution. Further work is needed to understand how both cell-extrinsic and cell-intrinsic processes produce the vast array of MMP-derived muscles found in vertebrates. Such studies may also shed light on how other cell types migrate, such as the lateral line that we show requires six1b/six4b function for motility. MMP studies may also contribute to understanding of disease; for example, tumor cells undergo epithelial-mesenchymal transitions and migrate in response to SIX1 and MET overexpression (Birchmeier et al., 2003; Christensen et al., 2008; Micalizzi et al., 2009; Xu et al., 2014). Thus, our work on muscle precursor migration may have broad applications to the study of development, evolution and medicine.

Zebrafish transgenic and mutant strains were maintained on the AB wild-type background, except for the met^+13-3 mutants and their control siblings, which were maintained on the TU wild-type background. Standard husbandry and stock maintenance procedures were followed (Westerfield, 2007). All experiments were conducted before zebrafish sex becomes apparent and were replicated on at least two clutches. All animal protocols are approved by the Ohio State University Institutional Animal Care and Use Committee (OSU IACUC) or, for the met mutant experiments, by the Monash University Animal Services Animal Ethics committee. The transgenic lines used are Tg(mylpfa:mCherry)cz3327 (Ignatius et al., 2012), Tg(hsp70l:dnfgfr1-GFP)pd1 (Lee et al., 2005), Tg(smyhc1:lyn-tdTomato)oz29 and Tg(six1b:lyn-GFP)oz5. Transgenic construction used established methods (Suster et al., 2011; Wang et al., 2011); see supplementary Materials and Methods for details. The mutant lines used are aldh1a2ⁱ²⁶ (Begemann et al., 2001), six1a^oz9 (Talbot and Amacher, 2014), met^+13-3, six1b^oz1, Δsix1b^oz34, six4a^oz40, six4b^oz31, six4b^oz35, Df(Chr13:six1a,six4a)^oz27 and Df(Chr20:six1b,six4b)^oz16. Fish were genotyped using PCR of extracted DNA, followed in some cases by restriction enzyme cleavage. See supplementary Materials and Methods for details on genotyping and mutant construction of the latter eight mutant lines (Talbot and Amacher, 2014; Xiao et al., 2013).

Fluorescent RNA in situ hybridization was performed using established protocols (Talbot et al., 2010), available in detail on ZFIN (https://wiki.zfin.org/x/0wHI). Non-fluorescent RNA in situ hybridization was also performed using established protocols (Jowett, 1999). We mounted embryos for imaging as described previously (Morrow et al., 2017). For all in situ experiments, to prevent pigment formation, embryos were treated with 0.003% N-phenylthiourea (PTU) beginning at 24 hpf. In situ hybridization, coloration and imaging was performed side-by-side for compared images. We used previously published probes for lbx1a (Lukowski et al., 2011), lbx2 (Ochi and Westerfield, 2009), hgfa (Haines et al., 2004), met (Haines et al., 2004), and pax3b (Minchin et al., 2013). We generated a new probe covering 1 kb of the pax3a transcript and find expression patterns matching those previously described (Minchin et al., 2013). The six4a probe was kindly provided by Lindsey Barske and Gage Crump (University of Southern California, CA, USA). We generated new probes to six1a, six1b and six4b, complementary to approximately 1 kb of each transcript and validated them by comparison to previously published Six family probes (Bessarab et al., 2004; Kobayashi et al., 2000; Nord et al., 2013); further details on probe construction are available in the supplementary Materials and Methods.

Immunolabeling used antibodies Rbfox1l (1:500) (Berberoglu et al., 2017), F310 (1:100, Developmental Studies Hybridoma Bank) and A4.1025 (1:1000, Developmental Studies Hybridoma Bank). F-Actin was labeled using Alexa Fluor-conjugated Phalloidin (1:50, Thermo Fisher).

To induce dnFgfr expression, embryos from an outcross of Tg(hsp70l:dnFgfr1-GFP)pd1 (Lee et al., 2005) were transferred to 37°C system water for 30 min. After heat shock, embryos were placed in dishes of 28.5°C system water containing 0.003% PTU to prevent pigment formation. Embryos carrying the Tg(hsp70l:dnFgfr-GFP)pd1 transgene were separated from non-transgenic sibling controls by expression of ubiquitous GFP after heat shock.

For small molecule treatments, we transferred embryos to new dishes containing either the indicated concentration of chemical inhibitor or the equivalent concentration of solvent as the treatment control. Three protein inhibitors were used in this study: the Fgf/Vegf inhibitor SU5402 (Sun et al., 1999), the Smoothened inhibitor cyclopamine (Chen et al., 2002) and the Met inhibitor SGX523 (Buchanan et al., 2009). Working concentration was 100 µM, except for SU5402, which was 10 µM for Movie 8 and 17 µM for Fig. 7. For SU5402 24-36 hpf treatments (Fig. 7C-E), we transferred embryos to inhibitor-containing water at the beginning of the pulse, and then washed them rapidly and thoroughly in fresh system water at the end of the pulse. For all other inhibitor experiments, embryos were transferred into inhibitor at 24 hpf until fixation.

Brightfield images were acquired using an upright Axioplan 2 microscope using Axiovision software. For the met/sibling images (Fig. 4G,H), we used a Zeiss LSM 710 Live Duo confocal microscope with Zen 2011 software. All other confocal images were collected using an inverted Nikon TiE microscope equipped with an Andor Revolution WD spinning disk confocal system. Images were analyzed using Metamorph software, and measurements were taken using ImageJ (Abràmoff et al., 2004). Cell tracking was performed manually using the MTrackJ plugin in ImageJ software (Meijering et al., 2012). All compared images were taken with the same image settings and processed identically. For time-lapse imaging, we mounted embryos in system water containing 0.1% agarose (GeneMate, E-3120-500) that had been boiled and cooled to 42°C before Tricaine was added (0.019% final concentration). Approximately 4 ml of the agarose solution was transferred to 35 mm glass-bottom Petri dishes (World Precision Instruments, FD35-100). Embryos were oriented in the dishes, and the agarose was allowed to set 1 hour before plates were imaged. During imaging, an Okogawa-heated stage and lens collar was used to maintain a temperature of 28.5°C. Petri dishes were covered with lids during imaging to minimize evaporation. Additional time-lapse details are provided in movie legends. For Movie 7, we labeled nuclei by injecting H2B-CFP mRNA into embryos at the one-cell stage (Megason, 2009).

Embryos were selected randomly for confocal imaging, and we included all imaged embryos in statistical analysis. We measured muscles as described in the supplementary Materials and Methods. All statistical analysis was performed using JMP v 11. Significance was defined using Tukey–Kramer Honestly Significantly Different (HSD) comparisons after ANOVA, with a significance threshold of <0.01. n represents the number of fish measured. Graphs of quantification represent data as box and whisker plots, with boxes showing the first through third quartiles, whiskers representing 1.5 times the interquartile range, and outlier data shown as dots.

We thank the Amacher lab fish facility staff for excellent fish husbandry and the OSU Neuroscience Imaging Core for equipment and advice. We thank the FishCore zebrafish facility (Australian Regenerative Medicine Institute, Monash University) for care and maintenance of the met^+13-3 mutant line. We thank Zachary Morrow for imaging many of the embryos shown in Fig. S4 and for genotyping assistance. We thank Ken Poss for sharing the Tg(hsp70l:dnfgfr1-GFP)pd1 line, David Langenau for the Tg(mylpfa:mCherry)cz3327 line, Tom Schilling for the aldh1a2ⁱ²⁶ line, Phil Ingham for the plasmid used to generate the Tg(smyhc1:lyn-tdTomato)oz29 line, and Gage Crump, Andrew Waskiewicz and Monte Westerfield for the six4a, lbx1a and lbx2 in situ plasmids, respectively. We thank Zachary Morrow, Anna Newman-Griffis, Lindsey Barske, Kiel Tietz, Joy-El Talbot and Frank Tulenko for comments on this manuscript. The Australian Regenerative Medicine Institute is supported by grants from the State Government of Victoria and the Australian Government.