Single cell dynamics of embryonic muscle progenitor cells in zebrafish

Tissue-resident stem cells, such as muscle stem cells, are crucial for proper organ development and tissue homeostasis. To harness the power of stem cells for treating diseases such as muscular dystrophy, it is critical to understand how muscle progenitor cells (MPCs) behave in vivo. In vertebrates, many skeletal muscles originate from the somites (Saga and Takeda, 2001). As the embryo develops, the ventral somite forms the sclerotome, generating progenitors of the axial skeleton and tendons, whereas the dorsolateral somite forms the dermomyotome. The dermomyotome further splits to form the dermatome and the myotome, which contribute to the skin and skeletal muscles, respectively (Christ and Scaal, 2008; Scaal and Christ, 2004). Thus, the dermomyotome contains embryonic MPCs required for the initial formation and growth of the musculature. Fate mapping and lineage tracing experiments in mouse and chick have demonstrated that the dermomyotome is also the source of adult muscle stem cells known as satellite cells (Gros et al., 2005; Kassar-Duchossoy et al., 2005; Relaix et al., 2005; Schienda et al., 2006).

Both embryonic MPCs in the dermomyotome and satellite cells contribute to myofiber formation, and are marked by the expression of the paired box transcription factors Pax3 and Pax7 (Dumont et al., 2015; Scaal and Christ, 2004). However, the function and dynamics of satellite cells have been more extensively studied than embryonic MPCs. Quiescent satellite cells often display a bipolar morphology and reside beneath the basal lamina on the surface of the myofiber (Scharner and Zammit, 2011; Webster et al., 2016). The extracellular matrix (ECM) surrounding the satellite cell constitutes its niche and has been implicated in the regulation of satellite cell behavior (Baghdadi et al., 2018; Bentzinger et al., 2013b; Fry et al., 2017; Rayagiri et al., 2018; Tierney et al., 2016; Urciuolo et al., 2013). Upon muscle injury, ‘activated’ satellite cells undergo proliferation and initiate the myogenic program. The sequential production of myogenic regulatory factors, including Myf5, MyoD and Myogenin, results in the differentiation of myoblasts that align and form new syncytial myofibers or fuse with existing myofibers. A single transplanted satellite cell is capable of self-renewal and contributes to myofibers (Sacco et al., 2008). Conversely, genetic ablation of satellite cells in adult mice completely abolishes injury-induced muscle regeneration (Lepper et al., 2011; Murphy et al., 2011; Sambasivan et al., 2011), demonstrating a critical role of satellite cells in muscle regeneration.

It has been challenging to visualize the in vivo behavior of satellite cells. Previous work has been mostly inferred from ‘snapshots’ of histological sections or analysis from in vitro cultures (Bentzinger et al., 2014; El Fahime et al., 2000; Jockusch and Voigt, 2003; Kuang et al., 2007; Siegel et al., 2009). However, satellite cells that are separated from their physiological environment are invariably activated, and in vitro approaches do not therefore provide the whole picture of their endogenous behavior. With the advance of intravital imaging, recent work provides the first glimpse of mouse satellite cell behavior in vivo (Webster et al., 2016). Quiescent satellite cells are largely immobile, whereas activated satellite cells proliferate and migrate along the ECM remnants of injured myofibers during regeneration.

The remarkable regenerative ability and the ease of in vivo imaging have made zebrafish a powerful system to study MPC behavior (Ratnayake and Currie, 2017). Muscle injury can be ‘repaired’ by resealing the membrane of the damaged myofiber or ‘regenerated’ by replacing the damaged muscle with a new myofiber generated by MPCs. In this study, we focus on muscle regeneration by MPCs. The zebrafish dermomyotome, also known as the external cell layer (ECL), is marked by the expression of pax3 and pax7, similar to higher vertebrates (Devoto et al., 2006; Feng et al., 2006; Hammond et al., 2007). During somitogenesis, the pax3/pax7⁺ dermomyotome is generated from the anterior somitic compartment via whole-somite rotation and is thought to generate new myofibers during myotome growth (Hollway et al., 2007; Stellabotte et al., 2007). At 4-5 days post-fertilization (dpf), some pax7⁺ MPCs derived from the dermomyotome can be observed deep in the myotome between myofibers (Seger et al., 2011). A sub-population of these pax7⁺ fiber-associated cells is specifically marked by the mouse satellite cell marker cmet. cmet⁺ cells are capable of self-renewal and contribute to muscle regeneration, suggesting that they are functionally equivalent to mammalian satellite cells (Gurevich et al., 2016). Recent work has also shown that Pax3/7-expressing satellite-like cells in adult zebrafish are essential for the regeneration after muscle injury (Berberoglu et al., 2017). Therefore, pax7 labels multiple MPC populations at different developmental stages in zebrafish: (1) embryonic MPCs in the dermomyotome on the surface of the somite; (2) fiber-associated deep myotomal cells (including cmet⁺ cells) in larval zebrafish; and (3) satellite-like cells in adult zebrafish.

To understand how embryonic MPCs contribute to muscle regeneration, we developed new transgenic tools and methods to analyze their dynamics and function at single cell resolution. We identified several ECM genes as new markers of embryonic MPCs. Genetic lineage tracing using col1a2-based transgenic lines demonstrated that col1a2⁺ MPCs contribute not only to embryonic muscle growth but also to muscle regeneration after injury. Combining in vivo imaging and single cell clonal analysis, we describe the dynamics of ‘quiescent’ and ‘activated’ col1a2⁺ MPCs. Together, our study provides a dynamic view of embryonic MPCs during muscle growth and regeneration.

Embryonic MPCs in the dermomyotome in zebrafish are traditionally labeled by the expression of pax7a (Devoto et al., 2006; Feng et al., 2006; Hammond et al., 2007). Using double fluorescent in situ hybridization, we identified several extracellular matrix (ECM) genes, including col1a2 (collagen type 1 α2), col5a1 (collagen type 5 α1) and cilp (cartilage intermediate layer protein), that showed co-expression with pax7a on the outer surface of the somite (Fig. 1A). To visualize embryonic MPC dynamics in vivo, we generated a col1a2:Gal4 transgenic line by BAC recombineering (Fig. 1B). The col1a2:Gal4 line can be crossed with different UAS effector lines to label, ablate or lineage trace MPCs. Co-labeling using col1a2 and kaede probes in col1a2:Gal4; UAS:Kaede (col1a2^Kaede) embryos revealed that the col1a2:Gal4 reporter largely recapitulated the endogenous col1a2 expression pattern (Fig. S1). Similarly, col1a2:Gal4; UAS:NTR-mCherry (col1a2^NTR-mCherry) labeled the outer surface of the somite external to α-actin:GFP-expressing muscle cells, corresponding to the anatomical location of the dermomyotome (Fig. 1C; Movie 1). To determine whether the col1a2:Gal4 line labels embryonic MPCs, we performed immunostaining using an anti-Pax7 antibody in col1a2^NTR-mCherry embryos. At 2 dpf, Pax7 antibody labels both MPCs (weak staining) and xanthophores, neural crest-derived pigment cells (strong staining). Indeed, all mCherry⁺ cells on the lateral surface of the somite were weakly Pax7⁺ (Fig. 1D), indicating that col1a2^NTR-mCherry specifically labels MPCs in the dermomyotome but not xanthophores. On average, there were 24.6 Pax7⁺ MPCs on the superficial surface of each somite, of which ∼87% were also mCherry⁺ (Fig. S2A). The incomplete labeling might be because gene expression driven by the Gal4-UAS system is variegated (Akitake et al., 2011), or alternatively, the col1a2:Gal4 line might label a sub-population of Pax7⁺ MPCs. At 5 dpf, Pax7⁺ fiber-associated MPCs (about 9.3 per somite) started to emerge in the deep myotome, of which about 39% were weakly mCherry⁺, probably due to the perdurance of the mCherry protein (Fig. 1D; Fig. S2B). Based on marker expression and lineage analysis (shown later), we termed Pax7⁺col1a2⁺ cells on the superficial surface of the somite ‘col1a2⁺ MPCs’. In this study, we focused on the behavior and function of such col1a2⁺ MPCs.

To determine whether the dermomyotome on the surface of muscles persists in adult zebrafish, we imaged vibratome sections of col1a2^NTR-mCherry fish at 22 mm standard length (SL). We identified three layers of col1a2⁺ cells external to the muscles, corresponding to scales (the most outer layer), skin (the middle layer) and the presumptive dermomyotome (the innermost layer external to muscles) (Fig. 1E). The presence of a few weakly mCherry⁺ myofibers adjacent to the presumptive dermomyotome suggests that col1a2⁺ cells contribute to muscle growth in adult zebrafish.

Using the col1a2:Gal4 line, we determined the distribution, morphology and dynamics of col1a2⁺ MPCs. First, we quantified the distribution of col1a2⁺ MPCs in col1a2^NTR-mCherry embryos stained with the Pax7 antibody at 2 dpf. Most col1a2⁺ MPCs were located along the myosepta with 42% along the vertical myoseptum (VM) and 15% along the horizontal myoseptum (HM) (Fig. 1D; Fig. 2A,B). Second, by taking advantage of some highly mosaic col1a2^Kaede embryos (which probably result from transcriptional silencing of UAS repeats over many generations; Akitake et al., 2011), we were able to visualize the morphology of individual col1a2⁺ MPCs at 3 dpf (Fig. 2C). They were relatively flat cells sandwiched between the epidermis and myofibers. Individual col1a2⁺ MPCs appeared to either ‘float’ in between the two somite boundaries or ‘anchor’ their cell bodies along the myoseptum. They always exhibited a ramified morphology with multipolar lamellipodia-like cellular projections (Fig. 2C). Thin filopodia-like projections (up to 25 µm) extending out of these lamellipodia can often be observed in col1a2⁺ MPCs, suggesting potential long-range cell-cell communications or the ability to detect distant injuries. Quantification showed that col1a2⁺ MPCs along the HM displayed fewer filopodia compared with cells along the VM, whereas centrally located MPCs had shorter filopodia than cells along the VM and HM (Fig. S2C,D). By combining two UAS reporters, we next visualized the dynamics of col1a2⁺ MPCs in col1a2:Gal4; UAS:NTR-mCherry; UAS:Kaede embryos (Fig. 2D). The mosaic nature of these transgenes allowed us to label a large number of col1a2⁺ MPCs in different colors. col1a2⁺ MPCs appeared to cover the surface of the somite with each cell occupying a non-overlapping territory. Time-lapse imaging showed that after cell division, daughter cells regained the ramified morphology and maintained the similar territory previously occupied by the mother cell (Fig. 2D; Movie 2). Together, our results demonstrate that the new col1a2:Gal4 line can be utilized to visualize the dynamics of embryonic MPCs at single cell resolution.

To determine whether col1a2⁺ MPCs contribute to muscle growth, we performed Cre-mediated lineage tracing to determine the fate of col1a2⁺ MPCs. We generated the col1a2:Gal4; UAS:Cre-ERT2 (col1a2^Cre-ERT2) transgenic line to express tamoxifen-inducible Cre recombinase in col1a2⁺ cells, and utilized the ubi:loxP-EGFP-loxP-mCherry (ubi:Switch) line (Mosimann et al., 2011) as the lineage reporter. Induction of Cre activity by 4-hydroxytamoxifen (4-OHT) results in the excision of the EGFP cassette and permanent expression of the mCherry protein in the Cre-expressing cell and its progeny (Fig. 3A). col1a2^Cre-ERT2; ubi:Switch embryos were pulsed with 4-OHT for 2 h at 3 dpf to mosaicly label col1a2⁺ MPCs, and the mosaicism allows for reliable cell tracing over multiple time points. At 24 h after the 4-OHT pulse, a small fraction of col1a2⁺ MPCs on the surface of the somite was labeled by the mCherry expression, but no myofibers were labeled (Fig. 3B). Tracing of mCherry⁺ MPCs for 4 consecutive days revealed that new mCherry⁺ myofibers started to emerge at 48 h after 4-OHT pulse, and increased in number at later time points. Similar results were obtained with photoconversion-based lineage tracing experiments in col1a2^Kaede embryos (Fig. S3A,B). These results indicate that col1a2⁺ cells, most likely col1a2⁺ MPCs, contribute to the generation of new myofibers during larval muscle growth. Long-term tracing of col1a2⁺ cells resulted in substantial labeling of the skin and muscles in fish at 4 weeks post-fertilization (wpf) and 1 year post-fertilization (ypf) (Fig. 3C). Interestingly, almost all mCherry⁺ myofibers were also GFP⁺. By contrast, ‘switched’ skin cells were mostly mCherry⁺GFP⁻. This observation suggests that col1a2⁺ MPCs contribute to muscle growth by cell fusion in adult zebrafish.

Since col1a2⁺ MPCs contribute to muscle growth, we next asked whether they also contribute to muscle regeneration. We performed needle injury experiments on muscles of col1a2^NTR-mCherry; α-actin:GFP embryos (Fig. 4A). Fish were injured within a 1-2 somite area at 3 dpf and imaged every 24 h for 3 days. The injured area, as indicated by the absence of α-actin:GFP expression at 1 hpi (hours post-injury), became substantially smaller at 24 hpi. This was accompanied by the emergence of elongated mCherry⁺ cells in the GFP⁻ area, suggesting that col1a2⁺ cells were recruited to the injury site. By 48 hpi, the injury area was completely replaced by newly regenerated muscles, as indicated by higher level of α-actin:GFP expression compared with uninjured regions. At 72 hpi, on average 71% of the regenerated area was also labeled by mCherry expression (Fig. 4A; Fig. S4A), suggesting that col1a2⁺ cells contribute to newly formed fibers. Thus, needle injury experiments demonstrate that small muscle injuries in zebrafish can be quickly regenerated within 48 h, with potential contributions from col1a2⁺ MPCs.

We further investigated the expression kinetics of different markers during the entire process of muscle regeneration. Fish were injured by needle stabbing at 3 dpf and fixed at different time points (7, 24, 48, 76 hpi) for in situ analysis (Fig. 4B; Fig. S4B). The expression of MPC marker pax7a reached the highest level at the injury site at 24 hpi. The elevated pax7a expression remained at 48 hpi before returning to the basal level by 76 hpi. Analysis of several ECM genes (col5a1, col1a2, col1a1a, cilp, postnb and sparc) revealed similar expression kinetics as pax7a (Fig. 4B; Fig. S4B). By contrast, myogenic markers (myoD and myogenin) displayed kinetics slightly lagging behind pax7a and ECM markers. Their expression initiated at 24 hpi, peaked at 48 hpi, and returned to the basal level by 76 hpi. Double fluorescent in situ hybridization on injured embryos at 24 hpi revealed that high-level col5a1 expression at the injury site was accompanied by the elevated expression of pax7a, myogenin and myoD (Fig. 4C). Together, the time course of marker expression is consistent with the timing of muscle regeneration from the activation of MPCs to the differentiation of new myofibers (Fig. 4A). Our results suggest that col1a2⁺ MPCs are not only recruited to the injury site to generate new myofibers, but also upregulate ECM gene expression perhaps to facilitate the regenerative process.

We have shown that col1a2⁺ MPCs contribute to new myofibers during both muscle growth and regeneration. To define the dynamic behavior of individual col1a2⁺ MPCs in vivo, we performed single cell clonal analysis by taking advantage of the photoconvertible Kaede (Ando et al., 2002) and the mosaic nature of the col1a2^Kaede line. Photoconversion of one isolated Kaede^green MPC allowed us to visualize its cellular behavior and trace all of its Kaede^red descendants. Briefly, we screened col1a2^Kaede fish at 3 dpf and identified embryos with mosaic labeling of MPCs. Single isolated Kaede^green cells were photoconverted with one cell per somite (1-4 cells per embryo) to ensure accurate cell tracing across multiple time points. We then used the 750 nm laser to introduce targeted muscle injury near one Kaede^red cell. Kaede^red MPCs in laser-ablated somites were designated as cells under ‘injured’ condition whereas photoconverted cells in uninjured somites were defined as ‘control’ cells. Individual embryos were imaged at 1, 24, 48 and 72 hpi to capture the behaviors of individual Kaede^red MPCs and their descendants (Fig. 5A). Since we found that col1a2⁺ MPCs along the HM rarely divide (as shown previously by Nguyen et al., 2017), we focused our analysis on centrally located and VM-associated MPCs. Together, we followed 133 col1a2⁺ MPCs in 56 embryos.

Based on cell behaviors of each clone, we categorized the response of col1a2⁺ MPCs into four different groups (Fig. 5B). In the type I response, cells did not proliferate and maintained the ramified morphology throughout 72 h. In the type II response, cells underwent one or more cell divisions, but all daughters maintained the ramified morphology, suggesting a quiescent state. By contrast, in the type III response, cells generated small elongated cells, which were usually bi-polar with processes extending along myofibers. This is markedly distinct from the multipolar ramified morphology of quiescent col1a2⁺ MPCs, suggesting an activated state. Lastly, in the type IV response, cells first gave rise to small elongated cells similar to those in the type III response, some of which later generated one or more Kaede^red myofibers by 72 hpi. As Kaede protein appeared to concentrate in the nuclei of muscle cells (Fig. S5A,B), new myofibers generated from Kaede^red cells can be easily identified based on the stronger Kaede^red signal in the oval-shaped nucleus with a weaker and diffusive signal in the cytoplasm spanning the width of a somite. Antibody staining revealed that the Kaede⁺ nucleus of a newly generated muscle no longer expressed Pax7 (a feature typical of differentiated muscles), whereas small elongated cells in between myofibers remained Pax7⁺ (Fig. S5C). Together, under uninjured conditions, col1a2⁺ MPCs generated predominantly type I (58%, 38/65 cells) or type II (40%, 26/65 cells) responses, but rarely type III response (2%, 1/65 cells) and never a type IV response (Fig. 5C). By contrast, cells under injured conditions exhibited all four types of responses, with a combined 35% (24/68 cells) in the type III and IV categories. Similarly, 24% (16/68 cells) of col1a2⁺ MPCs in injured conditions generated clones of at least 3 cells compared with only 2% (1/65 cells) in control conditions (Fig. S5D), suggesting an increase in cell proliferation during muscle regeneration. These results suggest that type III/IV behaviors represent the muscle regenerative response of activated col1a2⁺ MPCs. Since the formation of new myofibers was always preceded by small elongated cells (Fig. 5B), the type III response likely represents a transitional phase before the type IV response.

Since not all col1a2⁺ MPCs would respond to the injury in the same somite (Fig. 5B,C), we asked whether the initial position of the cell influences its behavior. We found that the distance from the injury to the center of the labeled MPCs did not correlate with the type of response (Fig. 5D). For example, some cells 24 µm away failed to respond to the injury whereas other cells >80 µm away became activated. This result suggests that col1a2⁺ MPCs can detect and respond to an injury at a long distance away from the cell body. Interestingly, col1a2⁺ MPCs located along the VM were slightly more likely to generate a type III or IV response (52%, 12/23 cells) than centrally located cells (29%, 13/45 cells) (Fig. 5E), suggesting that the local niche might influence the behavior of MPCs or that there are multiple types of MPCs that differentially respond to injury.

Our single cell clonal analysis reveals that activated col1a2⁺ MPCs go through a series of stereotypic phases to generate new myofibers. To further test this, we carried out confocal time-lapse imaging to visualize the entire process of muscle regeneration. First, we imaged the ‘early phase’ of muscle regeneration at 0-23 hpi (Fig. 6A). Mosaic col1a2^Kaede embryos were injured by needle stabbing at 59 hpf. Individual Kaede^green MPCs were photoconverted to facilitate cell tracking (Fig. 6B; Movie 3). Within the first few hours, the activated col1a2⁺ MPC started to project polarized cellular processes along myofibers. This bi-polar and elongated morphology was maintained even after cell division. Consistent with our clonal analysis, new myofibers were never generated during this time interval. Thus, the early phase of muscle regeneration is characterized by the morphological changes and proliferations of activated col1a2⁺ MPCs.

Next, we imaged the ‘late phase’ of muscle regeneration in mosaic col1a2^Kaede embryos at 29-48 hpi (Fig. 6C). Remarkably, all new myofibers formed in a similar manner: a small elongated Kaede⁺ MPC visible at one time point disappeared by the next time point (8-min intervals), with simultaneous emergence of Kaede⁺ myofiber characterized by Kaede^strong nucleus and Kaede^weak cytoplasm (Fig. 6D; Movie 4). The rapidity of this event suggests that new myofibers are formed through cell fusions between a Kaede⁺ MPC and an existing non-labeled myofiber. To test this model further, we performed time-lapse imaging on col1a2^NTR-mCherry; α-actin:GFP embryos. Similar cytoplasmic spread of the mCherry signal was observed after the fusion of an mCherry⁺ MPC with a neighboring GFP⁺ myofiber (Movie 5).

To better characterize the behavior of col1a2⁺ MPCs, we performed similar cell tracking experiments at high temporal resolution (5-min intervals) using col1a2^Kaede; ubi:Zebrabow embryos. Nuclear localization of the Kaede protein in myofibers allowed us to track the nucleus even after the cell fusion. ubi:Zebrabow was used to visualize myofibers as it was expressed strongly in muscles but at a very low level in col1a2⁺ MPCs. We observed a total of 31 fusion events from 12 embryos, all of which displayed similar kinetics to those described above (Fig. 6E; Fig. S6; Movie 6). The disappearance of cell processes was accompanied by simultaneous spreading of the Kaede signal in the cytoplasm of an RFP⁺ myofiber. Of all fusion events, we identified 3 pairs of Kaede⁺ sibling cells from 3 different embryos. Intriguingly, in all 3 cases, the two sibling cells fused sequentially to the same muscle, resulting in the addition of two new nuclei to one existing myofiber. This result suggests that sequential fusion of sibling MPCs is one mechanism to grow existing myofibers to replace injured muscles.

Since we never observed any de novo fiber formation in our time-lapse movies, we asked whether muscles are formed exclusively through cell fusions. To answer this question, we performed Cre-mediated lineage tracing in col1a2^Cre-ERT2; ubi:Switch fish (Fig. 7A). Embryos at 3 dpf were pulsed with 4-OHT for 3.5 h to mosaicly label col1a2⁺ MPCs. Fish were then injured by needle stabbing and imaged at 75 hpi to quantify newly formed mCherry⁺ myofibers. Since only a small fraction of col1a2⁺ MPCs were switched and labeled by mCherry, there are three possible modes of generating a new myofiber (Fig. 7B): (1) fusion of an mCherry⁺ MPC with an existing EGFP⁺ myofiber (MPC-muscle fusion); (2) fusion of an mCherry⁺ MPC with one or more EGFP⁺ MPCs (MPC-MPC fusion); and (3) de novo formation of a nascent fiber from an mCherry⁺ MPC. The first two cases (cell fusions) would result in an EGFP⁺mCherry⁺ myofiber, while the third scenario (de novo formation) would result in an EGFP⁻mCherry⁺ myofiber. In uninjured controls, 97% of new myofibers (37/38) were formed via cell fusion, whereas only 3% of fibers (1/38) were generated de novo (Fig. 7C,D). Interestingly, in injured embryos, de novo fiber formation increased about 5-fold to 14% (14/98) at the muscle injury site. Combined with our time-lapse imaging data, this result suggests that during larval growth, new myofibers are generated predominantly through fusion of col1a2⁺ MPC descendants with existing myofibers; however, under injury conditions, more muscles are regenerated through de novo fiber formation.

To test whether col1a2⁺ MPCs are essential for embryonic muscle regeneration, we partially ablated col1a2⁺ cells using a nitroreductase (NTR)-based system. The NTR enzyme converts the harmless prodrug metronidazole (MTZ) into a cytotoxic compound that induces rapid cell death of NTR-expressing cells (Curado et al., 2008; Pisharath et al., 2007). We selected col1a2^NTR-mCherry embryos with a modest expression level to avoid high perduring NTR-mCherry expression in muscles. Compared with water-treated controls, MTZ-treated col1a2^NTR-mCherry embryos showed a 59% reduction in the number of Pax7⁺ cells, including both superficial MPCs and deep fiber-associated MPCs (Fig. 7E,F). We then analyzed muscle regeneration in embryos with partially depleted col1a2⁺ MPCs. mCherry⁺ embryos and mCherry⁻ controls from a col1a2^NTR-mCherry outcross were treated with MTZ at 2-3 dpf, injured at 3 dpf and fixed at 24 hpi for in situ analysis. Similarly to our previous observations (Fig. 4; Fig. S4B), most mCherry⁻ controls showed substantial upregulation of both myogenin and myoD at the injury site at 24 hpi (Fig. 7G). By contrast, the majority of ablated embryos showed no or only modest upregulation of myogenic markers at the injury site, suggesting a compromised regenerative response. This result suggests that col1a2⁺ MPCs are required for efficient regeneration of muscle injury in zebrafish embryos.

Our study provides a dynamic view of embryonic MPCs in the zebrafish dermomyotome during muscle growth and regeneration. First, col1a2⁺ MPCs display a unique ramified morphology and express many ECM genes. Second, lineage tracing experiments show that col1a2⁺ MPCs contribute to normal muscle growth as well as muscle regeneration. Third, col1a2⁺ MPCs undergo a series of stereotypical steps to generate new muscle fibers, predominantly by cell fusions with existing myofibers.

Dermomyotome is an evolutionarily conserved structure on the external surface of the myotome present in most vertebrates (Devoto et al., 2006). Using a novel col1a2 transgenic line, we identified a population of col1a2⁺ cells on the surface of the somite that co-expresses the MPC marker Pax7. The location, distribution and marker expression of col1a2⁺ MPCs are consistent with what has been described as the dermomyotome, or the external cell layer, in zebrafish (Devoto et al., 2006; Feng et al., 2006; Hammond et al., 2007). Previous work in mouse and chick has shown that satellite cells originate from the embryonic dermomyotome (Gros et al., 2005; Kassar-Duchossoy et al., 2005; Relaix et al., 2005; Schienda et al., 2006). Similarly, using lineage tracing and time-lapse imaging, we showed that col1a2⁺ MPCs contribute not only to new myofibers but also to small Pax7⁺ fiber-associated cells, which likely correspond to larval MPCs (Gurevich et al., 2016; Nguyen et al., 2017; Pipalia et al., 2016; Roy et al., 2017; Seger et al., 2011). In amniotes, the dermomyotome has also been shown to give rise to the dermis (Christ and Scaal, 2008; Stellabotte and Devoto, 2007). However, previous short-term cell tracing experiments with dye labeling in zebrafish have not reached an agreement on whether the anterior somitic compartment (precursors of the dermomyotome) contributes to the dermis (Hollway et al., 2007; Stellabotte et al., 2007). Using Cre-mediated lineage tracing, we found that although short-term lineage tracing of col1a2⁺ MPCs (3 days) showed no contribution to the skin, long-term tracing for more than a month revealed substantial labeling of the skin. Our finding is consistent with a previous report that the zebrafish dermis is acellular during the first 3-4 weeks of development (Le Guellec et al., 2004). Together, our results suggest that col1a2⁺ MPCs are functionally equivalent to the amniote dermomyotome, contributing to both the dermis and muscles. Several scenarios might explain our observations. First, superficial col1a2⁺ cells on the somite at 3 dpf might be bi-potential and thus capable of generating both dermal and muscle lineages. Second, these superficial col1a2⁺ cells might be heterogeneous, containing distinct dermal progenitors and muscle progenitors. Lastly, in an unlikely scenario, col1a2⁺ cells deep around the notochord could contribute to dermal progenitors that migrate mediolaterally through muscles and populate the dermis at 4 wpf. Future experiments on single cell lineage analysis will help distinguish the first two more likely possibilities.

Using the same col1a2 reporter, we found that the superficial layer of col1a2⁺ cells immediately adjacent to muscles persists in adult zebrafish, likely corresponding to the presumptive dermomyotome described previously (Hollway et al., 2007). This observation is in contrast with the situation in the amniote dermomyotome, which is a transient embryonic structure (Christ and Scaal, 2008; Stellabotte and Devoto, 2007). The persistence of the dermomyotome in adult zebrafish provides another source of MPCs in addition to satellite-like cells (Berberoglu et al., 2017) for the continuous growth of the fish.

Using col1a2 transgenic lines, we discovered some unique features of col1a2⁺ MPCs. First, col1a2⁺ MPCs display a ramified morphology with dynamic cellular processes. In the quiescent state, these cells are not polarized, showing lamellipodia-like structures all around the cell. This morphology contrasts with that of mouse satellite cells, which usually display a bipolar morphology with short processes extending along the myofiber (Webster et al., 2016). Although cell bodies of col1a2⁺ MPCs are mostly stationary in uninjured conditions, their cellular processes are quite dynamic, constantly extending and retracting. Each cell and its processes occupy a largely non-overlapping area, reminiscent of the tiling of neuronal dendrites (Grueber and Sagasti, 2010). During cell divisions, the mother cell retracts all its processes, but immediately after the division, the two daughter cells would extend new processes to reclaim the similar surface area. These dynamic behaviors might ensure the complete coverage of the somitic surface and detect potential muscle injuries. Moreover, we often observed long filopodia-like structures extending from col1a2⁺ MPCs. Intriguingly, our single cell clonal analysis revealed that the distance between the cell body of a col1a2⁺ MPC and the injury site is not a reliable predictor of whether the MPC would generate a regenerative response. It thus raises the possibility that the long cellular projections might allow col1a2⁺ MPCs to detect muscle injury at a significant distance from the cell body. Interestingly, dermomyotome cells in chick display similar filopodia-like protrusions, facilitating the interaction with the overlying ectoderm during somite development (Sagar et al., 2015).

Second, we found that embryonic MPCs in the dermomyotome express a number of ECM genes, including col1a2, col5a1 and cilp. Similarly, col1a1 and cilp are expressed in what is described as a dermomyotome-like domain in trout embryos (Ralliere et al., 2015; Rescan et al., 2005). Moreover, we found that the expression of these ECM genes is dynamically induced during muscle regeneration. Our observations in embryonic MPCs in zebrafish show strong similarities to gene expression in mouse MPCs. For example, fetal muscle stem cells in mice show high-level expression of several ECM molecules, such as tenascin-C (TnC), fibronectin (Fn1) and collagen VI (Col6) (Tierney et al., 2016). Similarly, recently identified Twist2⁺ MPCs in mouse show enrichment of many ECM genes, including Col1a1, Col1a2 and Cilp (Liu et al., 2017). Interestingly, although quiescent adult satellite cells do not express high levels of ECM genes (Liu et al., 2017), activated satellite cells show upregulation of many collagen genes, such as Col1a1 and Col1a2 (Pallafacchina et al., 2010). Thus, dynamic regulation of ECM gene expression might be a common feature of MPCs.

Does the ECM play an active role in regulating the function of MPCs? We envision two non-mutually exclusive scenarios. First, the ECM might be an integral part of the niche that maintains the self-renewing property of MPCs. Recent work has implicated many ECM components, including fibronectin, tenascin-C, laminins and collagens, as critical niche factors that modulate satellite cell function (Bentzinger et al., 2013b; Fry et al., 2017; Rayagiri et al., 2018; Tierney et al., 2016; Urciuolo et al., 2013). In particular, it has been shown that collagen V produced by adult muscle satellite cells is an essential component of the quiescent niche, as deletion of the Col5a1 gene results in depletion of the stem cell pool (Baghdadi et al., 2018). The close association of most col1a2⁺ MPCs with the ECM-rich myoseptum (Charvet et al., 2011) suggests that the myoseptum might function as the niche for embryonic MPCs in zebrafish. Second, the induction of ECM gene expression upon muscle injury might provide the appropriate scaffold for the recruitment and/or migration of activated MPCs. Consistent with this idea, intravital imaging of muscle regeneration in mice reveals that ECM remnants from injured myofibers provide cues to regulate satellite cell behavior (Webster et al., 2016). Future studies will need to determine whether loss of any ECM component would compromise the regenerative capacity of col1a2⁺ MPCs in zebrafish.

Previous studies using different pax7 reporter lines have led to different conclusions on the role of MPCs during muscle regeneration. For example, it has been shown that pax7a⁺ MPCs do not participate in regenerating a small focal muscle injury, although they do contribute to new myofibers during the regeneration of a large injury (Knappe et al., 2015). This observation is in contrast with the work by Pipalia et al., which has shown that both pax7a⁺pax7b⁻ and pax7b⁺ MPCs contribute to new myofibers during muscle regeneration (Pipalia et al., 2016). These studies rely on GFP perdurance in new myofibers to infer that they originate from GFP⁺ MPCs. Differences in GFP stability in different transgenic lines could complicate lineage interpretations. To definitively address this issue, we performed Cre-mediated lineage tracing of col1a2⁺ MPCs. We demonstrated that col1a2⁺ MPCs not only contribute to new myofibers during normal larval development but also generate new muscles during regeneration of small injuries. Long-term tracing of descendants of embryonic col1a2⁺ MPCs revealed substantial contribution to adult muscles, suggesting that col1a2⁺ MPCs also give rise to larval/adult MPCs to sustain the growth of skeletal muscles.

Another question that needs to be clarified is the functional requirement of different MPC populations during muscle regeneration. Pipalia et al. have shown that muscle regeneration occurs normally despite the ablation of pax7b⁺ MPCs, which include superficial MPCs in the dermomyotome and deep fiber-associated MPCs (Pipalia et al., 2016). By contrast, depletion of a small subset of deep myotomal cells co-labeled by cmet and pax7a results in a failure to initiate regeneration (Gurevich et al., 2016). Although the lineage relationship between pax7b⁺ and cmet⁺ MPCs remains to be established, it raises the question which population of MPCs is essential for muscle regeneration. In our studies, partial depletion of col1a2⁺ MPCs resulted in significantly blunted regenerative response. Since col1a2⁺ MPCs are likely to include both pax7a⁺ and pax7b⁺ progenitors, one parsimonious explanation is that pax7a⁺ MPCs might play a more prominent role in mounting the regenerative response.

Our newly developed col1a2:Gal4 transgenic line provides a useful alternative to existing pax7-based transgenes in studying the dynamics of embryonic MPCs. By taking advantage of the variegated nature of the col1a2^Kaede line and the photoconvertible Kaede, we performed single cell clonal analysis of col1a2⁺ MPCs. A quiescent col1a2⁺ MPC (type I/II responses) maintains its ramified morphology even after occasional cell divisions. By contrast, an activated col1a2⁺ MPC (type III/IV responses) undergoes several stereotypic steps to generate new myofibers (Fig. 7H). First, it changes from its resting ramified morphology to an elongated and polarized morphology, usually within the first 3-5 h after injury. The cell extends long cellular projections along the longitudinal axis of neighboring myofibers. Next, the activated MPC undergoes several rounds of cell divisions generating a clone of small polarized daughter cells by 12-24 hpi. Finally, by 72 hpi, new myofibers emerge at the injury site, predominantly formed through fusions between descendants of col1a2⁺ MPCs and existing myofibers (discussed below). The dynamic behaviors of an activated col1a2⁺ MPC, such as the initial polarization phase, are reminiscent of myf5⁺ fiber-associated MPCs during muscle regeneration (Gurevich et al., 2016). However, unlike myf5⁺ MPCs, col1a2⁺ MPCs never cross the myoseptum to regenerate muscle injury in neighboring somites. Quantification of the clonal analysis further revealed that although the distance to the injury did not enable us to predict the cellular responsiveness, col1a2⁺ MPCs located along the VM were slightly more likely to respond to the injury than centrally located MPCs. One explanation is that VM-associated MPCs are better equipped to sense and respond to injury with slightly longer filopodia compared to central MPCs. Alternatively, the ECM-enriched myoseptum might provide the scaffold to facilitate the migration of activated col1a2⁺ MPCs.

Two independent lines of evidence indicate that col1a2⁺ MPCs generate new myofibers primarily through cell fusion. First, time-lapse imaging of the regeneration process at high temporal resolution in different col1a2 reporter lines showed that the emergence of a new Kaede⁺ or mCherry⁺ myofiber always appeared to be completed between two time points (5- to 8-min intervals), accompanied by the simultaneous disappearance of the col1a2⁺ MPC. The absence of intermediate steps, such as searching for attachment sites along the myoseptum, suggests that new myofibers are formed by cell fusion rather than de novo fiber formation. Consistent with this result, our Cre-mediated lineage tracing experiments showed that most new myofibers expressed both EGFP and mCherry, suggesting that cell fusion events between a ‘switched’ mCherry⁺ MPC and an ‘un-switched’ EGFP⁺ myofiber occurred. Indeed, long-term tracing of col1a2⁺ MPCs showed that almost all mCherry⁺ myofibers were also EGFP⁺, suggesting that fusion is also the predominant mode of muscle growth in adult zebrafish. Intriguingly, during muscle regeneration, a 5-fold increase in de novo fiber formation (14%) was observed compared with uninjured conditions (3%). This result suggests that de novo fiber formation is specifically induced during muscle regeneration to combat the increasing demand to replace damaged fibers. This might be achieved by differential activation of sub-populations of col1a2⁺ MPCs. Pipalia et al. (2016) have previously shown that pax7b⁺ MPCs contribute to muscle growth and regeneration by cell fusions, whereas pax7a⁺pax7b⁻ MPCs predominantly generate nascent fibers. Considering our findings, one possibility is that normal muscle growth is primarily contributed by pax7b⁺ MPCs, whereas muscle injury activates both pax7b⁺ and pax7a⁺pax7b⁻ MPCs.

In summary, our work provides a dynamic view of col1a2⁺ MPCs during muscle growth and regeneration. It also raises additional questions for future investigations. For example, what is the injury signal that activates col1a2⁺ MPCs? Tissue injury is often associated with an elevated level of reactive oxygen species (ROS) and the recruitment of patrolling immune cells such as macrophages and neutrophils (Love et al., 2013; Niethammer et al., 2009). It is therefore plausible that ROS and/or cytokines secreted by immune cells might be the trigger to activate MPCs. Indeed, mouse and human data have implicated macrophages and other immune cells as the critical regulators of satellite cell functions (Bentzinger et al., 2013a; Saclier et al., 2013).

Zebrafish strains were maintained and raised according to standard protocols (Westerfield, 2000). All animal research was conducted in accordance with the principles outlined in the current Guidelines of the Canadian Council on Animal Care. All protocols were approved by the Animal Care Committee at the University of Calgary (#AC17-0128). Embryos were grown at 28.5°C and staged as previously described (Kimmel et al., 1995). Fish older than 24 hpf were treated with 1-phenyl 2-thiourea (PTU) to prevent pigmentation. TL and TL/AB wild-type strains were used in this study along with the following transgenic lines: α-actin:GFP (Higashijima et al., 1997), col1a2:Gal4 (Ma et al., 2018), ptc2:Kaede (Huang et al., 2012), UAS:Kaede (Scott et al., 2007), UAS:NTR-mCherry (Davison et al., 2007), UAS:Cre-ERT2, ubi:loxP-EGFP-loxP-mCherry (ubi:Switch) (Mosimann et al., 2011), and ubi:Zebrabow (Pan et al., 2013). The mosaic col1a2:Gal4; UAS:Kaede line was maintained by selectively growing embryos with more mosaic Kaede expression.

UAS:Cre-ERT2 transgenic fish were generated by standard Tol2-mediated transgenesis (Suster et al., 2009). To generate col1a2:Gal4 transgenic line, BAC (bacteria artificial chromosome) clone zC122K13 from the CHORI-211 library that contains col1a2 genomic region with 78 kb upstream and 58 kb downstream regulatory sequences was selected for bacteria-mediated homologous recombination following the standard protocol (Bussmann and Schulte-Merker, 2011). Briefly, the pRedET plasmid was first transformed into BAC-containing bacteria. Second, an iTol2_amp cassette containing two Tol2 arms in opposite directions flanking an ampicillin resistance gene was recombined into the vector backbone of zC122K13. Lastly, a cassette containing the Gal4-VP16 with a kanamycin-resistant gene was recombined into zC122K13-iTol2_amp to replace the first coding exon of the col1a2 gene. After each round of recombination, successful recombinants were confirmed by PCR analysis. The final col1a2:Gal4 BAC was then co-injected with tol2 transposase mRNA into UAS:Kaede embryos at one-cell stage. Positive transgenic lines were identified by screening Kaede expression in F1 embryos from injected founders.

Two methods were employed to generate small focal muscle injury at specific locations in larval zebrafish. In needle injury, we used a sharp injection needle to stab muscles near the end of yolk extension (somites 17-19) so the injury site could be located easily during the lineage tracing. Alternatively, to introduce muscle injury at a more precise location, laser ablation was performed with the 750 nm laser and 25× objective on the Leica TCS SP8 multi-photon microscope. A region of interest (ROI) at a desired location was selected, zoomed in to the maximum (48×), and scanned with 100% 750 nm laser once. The laser-induced injury could be readily visualized in the bright field after scanning.

col1a2^Kaede embryos at appropriate stages were anesthetized with tricaine and mounted in 0.8% low melting point agarose in a glass-bottom dish (MatTek). Photoconversion experiments were performed using the 405 nm laser and the 20× objective on the Olympus FV1200 confocal microscope. For photoconversion of large areas (∼5 to 6 somite region), 50% laser power was used to scan the desired ROI for two frames at a dwell time of 200 µs per pixel. For single cell photoconversions, 2% laser power was used to scan a small ROI (10×10 pixels) with the Tornado mode at a dwell time of 100 µs per pixel for a total of 1-2 s. After photoconversion, embryos were released from the agarose, transferred to fish water to recover in the dark, and analyzed at desired stages.

To obtain mosaic labeling, col1a2:Gal4; UAS:Cre-ERT2; ubi:Switch embryos were pulsed with 10 µM 4-hydroxytamoxifen (4-OHT) for 2-3 h at desired stages. After treatment, 4-OHT was rinsed off with fish water three times, and embryos were recovered in fish water for analysis at appropriate stages.

Mosaic col1a2^Kaede embryos at appropriate stages were selected for single cell tracing experiments. Individual isolated col1a2⁺ MPCs were photoconverted to Kaede^red. To ensure reliable cell tracing over time, a maximum of one cell per somite and four cells per embryo were photoconverted and traced. Immediately after photoconversion, muscle injury was introduced by laser ablation near one Kaede^red cell. Images were taken before and after the photoconversion and then every 24 h until 72 hpi to trace individual Kaede^red cells and their progeny. For quantification in Fig. 5D, distance was measured from the center of the photoconverted cell to the center of the muscle injury.

Whole-mount in situ hybridization and antibody staining were performed according to the standard protocols (Thisse et al., 2004). The following antisense probes were used in this study: cilp, col1a1a, col1a2, col5a1, kaede, myoD, myogenin, pax7a (Seo et al., 1998), postnb and sparc. Double fluorescent in situ hybridizations were performed using different combinations of digoxigenin (DIG)- and dinitrophenyl (DNP)-labeled probes. For antibody staining, the following primary antibodies were used: rabbit polyclonal antibody to Kaede (1:1000; MBL, PM012), rabbit polyclonal antibody to RFP (1:1000; MBL, PM005) and mouse monoclonal antibody to Pax7 [1:10; Developmental Studies Hybridoma Bank (DSHB), PAX7-s]. For fluorescence detection of antibody labeling, appropriate Alexa Fluor-conjugated secondary antibodies (1:500; Thermo Fisher) were used.

Adult zebrafish were fixed in 4% paraformaldehyde (PFA) with 1% DMSO overnight at 4°C. After fixation, fish were washed in PBS and decalcified in 0.5 M EDTA for 3 days at room temperature. Following decalcification, the trunks were dissected, embedded in 20% gelatin and fixed overnight in 4% PFA at 4°C. Gelatin blocks were then washed with PBS and mounted in a Leica VT1000S vibratome to obtain transverse sections at 150 µm thick. Sections were mounted on slides in 50% glycerol for confocal imaging. Confocal tile scans were stitched together by the Olympus Fluoview software to generate the full view of the transverse section.

To ablate col1a2⁺ MPCs, col1a2^NTR-mCherry embryos at 2 dpf were treated with water (control group) or 5 mM metronidazole (MTZ, experimental group) for 24 h. Embryos were then washed with fish water two or three times and recovered at 28.5°C. To check ablation efficiency, embryos were fixed at 4 dpf and stained with anti-mCherry and anti-Pax7 antibodies. Pax7⁺ MPCs, including superficial MPCs in the dermomyotome and deep muscle fiber-associated cells, were counted for a 4-somite region of each embryo. To examine muscle regeneration, mCherry⁺ and mCherry⁻ embryos from a cross of col1a2^NTR-mCherry fish were treated with MTZ from 2 to 3 dpf, needle injured after the MTZ treatment, recovered, and fixed at 4 dpf for in situ analysis.

Embryos were anesthetized with tricaine and embedded in 0.8% low melting point agarose on a glass-bottom dish (MatTek). Fish were imaged with an Olympus FV1200 confocal microscope using a 20× objective. For time-lapse imaging of muscle regeneration, embryos were first injured at 2 or 3 dpf by needle stabbing. Injured embryos were then imaged laterally starting at appropriate stages at 5- to 8-min intervals for 19-24 h. All the confocal images were analyzed and quantified using Fiji software (Schindelin et al., 2012). Brightness and contrast were adjusted for better visualization. To generate color-coded depth projections, confocal z-stacks were processed with Fiji using the ‘temporal color code’ function.

To quantify the number of col1a2⁺ MPCs, col1a2^NTR-mCherry fish were stained with anti-Pax7 antibody at 2 dpf. Cells in each somite were counted based on col1a2 expression and Pax7 staining: all Pax7⁺ cells, Pax7⁺Col1a2⁺ cells and Pax7⁺Col1a2⁻ cells. Note that xanthophores, which have strong Pax7 staining, were not counted. Each data point in the graph represents the average from five somites of one embryo. Pax7⁺Col1a2⁺ cells were further categorized according to their position in a somite: near the vertical myosepta (VM), near the horizontal myoseptum (HM) or in between somite boundaries (central). To quantify the number of fiber-associated MPCs, col1a2^NTR-mCherry fish were stained with anti-Pax7 antibody at 5 dpf. Cells deep in each somite between myofibers were counted based on col1a2 expression and Pax7 staining: all Pax7⁺ cells, Pax7⁺Col1a2⁺ cells and Pax7⁺Col1a2⁻ cells.

All the graphs were generated in GraphPad Prism software. Data were plotted as mean±s.e.m. indicated. Significance between two samples was calculated by performing the Mann–Whitney U-test with two-tailed P-values: *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001.

We thank the zebrafish community for providing probes and reagents; Holger Knaut for BAC clones; Jason Berman for UAS:NTR-mCherry fish; Roger Ma for help with video annotation; Sarah Childs and members of the Huang laboratory for discussion; and Paul Mains and James McGhee for critical comments on the manuscript.