Inhibitory SMAD6 interferes with BMP-dependent generation of muscle progenitor cells and perturbs proximodistal pattern of murine limb muscles

As in all tetrapods, mammalian limb musculature is derived from a small number of myogenic progenitor cells (MPCs) that migrate from somites into the developing limb bud, where they expand in number, differentiate and form a multitude of individual muscles (Christ and Brand-Saberi, 2002). Migrating limb MPCs are thought to have no positional information but rather rely on signals from their new environment (Blagden and Hughes, 1999). The cues for muscle patterning reside in the limb mesenchymal cells and are independent of the presence of limb MPCs (Grim and Wachtler, 1991; Vallecillo-García et al., 2017). Increasing evidence suggests that individual muscles are formed when MPCs invade a prepattern that is established by muscle connective tissue (MCT) and controlled by a combination of transcription factors, e.g. Hox (Zakany and Duboule, 2007; Swinehart et al., 2013), TBX3 (Colasanto et al., 2016), TBX4/5 (Hasson et al., 2010), TCF4 (Kardon et al., 2003) and OSR1 (Vallecillo-García et al., 2017). However, the connective tissue of limbs without muscle does not form morphologically distinguishable structures that resemble the pattern of individualized muscles – the muscle-devoid space is instead filled with loosely organized mesenchyme and eventually with fat (Christ et al., 1977). Initial tendon formation also occurs independently of muscle; however, the tendons degenerate secondarily if they do not connect to a muscle (Huang et al., 2015).

Opposing the view of ‘myogenic naivety’, the expression of HOXA11 and HOXA13 proteins has been observed in chicken limb MPCs, suggesting that MPCs acquire positional identity. Interestingly, the spatiotemporal dynamics of Hox expression in chicken MPCs are influenced by cues emanating from the apical ectodermal ridge and the zone of polarizing activity. In addition, ectopic application of factors such as fibroblast growth factors (FGFs) and bone morphogenetic proteins (BMPs) were shown to regulate Hox expression in chicken MPCs. These findings suggest that MPCs may follow similar cues during patterning as the limb mesenchyme (Yamamoto et al., 1998; Hashimoto et al., 1999; Yamamoto and Kuroiwa, 2003).

In mice, muscle patterning starts at embryonic day (E)11.5, with successive splitting of premuscle masses into distinct blocks. Individual muscles become distinguishable from E12.5 onwards, and muscle individualization is complete by E14.5 at the end of embryonic myogenesis (Huang, 2017). Although non-muscle cells drive the limb muscle pattern (Kardon et al., 2002; Tozer et al., 2007), MPCs first need to integrate spatiotemporal information for their appropriate positioning, proliferation and differentiation. The molecular mechanisms driving MPC proliferation and differentiation at the right place and time are not fully understood.

BMPs are involved in embryonic MPC expansion in chicken limbs (Amthor et al., 1998; Wang et al., 2010). Moreover, BMP signaling displays regionalized activity within limb fetal muscles at the muscle and tendon interface level, and fetal MPCs respond to BMP signaling in chicken limbs (Wang et al., 2010), suggesting the appositional growth of limb muscles that is maintained by direct signaling from BMP-expressing tendons. Consistent with this, BMP signaling has been recently shown to promote mesoderm-derived fibroblast transdifferentiation into myoblasts and their incorporation within fetal muscle fibers at the muscle–tendon interface (Esteves de Lima et al., 2021). However, there is a lack of formal proof for whether BMPs act directly on developing limb muscle (thereby activating a BMP-dependent cell-autonomous response), at which developmental stage this interaction takes place, whether it involves BMPs in physiological signaling in orthotopic positions and whether this impacts muscle patterning.

BMPs signal on target cells via transmembrane serine/threonine kinase receptors, which form a ligand-receptor complex that permits the phosphorylation of the type I receptor via the constitutively active type II receptor (Nohe et al., 2002, 2004). The type I receptor in turn phosphorylates the BMP-responsive R-SMAD proteins 1, 5 and 9 (pSMAD1/5/9), which subsequently form complexes with co-SMAD4 and translocate into the nucleus to regulate transcriptional activity of target genes (Miyazawa and Miyazono, 2017). Upon BMP signaling, the inhibitory SMAD6 becomes upregulated as part of a negative feedback loop. SMAD6 interferes with BMP signaling by blocking R-SMAD phosphorylation at the level of the receptor, by antagonizing the pSMAD1/co-SMAD4 complex formation, and by increasing ubiquitin-mediated proteolysis of the BMP signaling components (Goto et al., 2007; Hata et al., 1998; Murakami et al., 2003).

Here, we explored the role of BMP signaling during mouse limb muscle development. We employed overexpression of human SMAD6 (huSMAD6) as a mean to cell-autonomously interfere with BMP signaling. We overexpressed huSMAD6 in embryonic limb MPCs and differentiated limb muscles following Cre-induced recombination by crossing of Rosa26^{LoxP-Stop-LoxP-huSMAD6-IRES-EGFP} mice with Cre-driver mouse lines Lbx1^Cre and HSA-Cre (Miniou et al., 1999; Sieber et al., 2007; Stantzou et al., 2017).

First, we identified whether limb myogenic cells respond to BMP signaling. We monitored the nuclear accumulation of BMP-induced phosphorylated SMAD proteins using double immunofluorescence against pSMAD1/5/9 (pSMADs) and myogenic markers in mouse forelimbs at different developmental stages. In E10.5 limb buds, migrating MPCs expressed the transcription factor PAX3 (Fig. 1), whereas the PAX7 and MYOD transcription factors were not detected, thus reproducing previously published data (Lepper and Fan, 2010; Wood et al., 2013). Surprisingly, all PAX3⁺ MPCs accumulated high levels of pSMADs, whereas non-myogenic mesenchymal cells showed no or, if any, very weak levels of pSMADs (Fig. 1). One day later, at E11.5, PAX3⁺ MPCs rapidly lost BMP signaling responsiveness during lineage progression. Emerging MYOD⁺ cells showed pSMADs in varying levels, some were negative and others showed a continuum from faintly to strongly positive. PAX7⁺ cells, however, were rarely pSMAD⁺ (Fig. 1). Of note, pSMAD⁺ non-myogenic cells were also found in the progress zone of E11.5 limb buds (Fig. 1).

By the end of the embryonic period, at E14.5, pSMADs were enriched at the tips of the muscle fibers abutting tendons (Fig. S1A). Double labeling of pSMADs with either PAX7, MYOD or myosin heavy chain (MHC) antibodies showed active BMP signaling in MYOD⁺ myonuclei at the myotendinous junctions and notably not in tendons (Fig. S1A). Low levels of pSMADs were also detected in rare PAX7⁺ MPCs at the muscle tips (Fig. S1A). This confirms, in mouse, the presence of BMP-responsive myonuclei and MPCs at the tips of primary myofibers facing tendons, reminiscent of previous work in chick (Esteves de Lima et al., 2021).

By the end of the fetal period, at E18.5, the pSMAD expression pattern was reversed. Indeed, the tips of fetal muscle fibers were devoid of pSMADs, which had now accumulated in the nuclei of non-muscle cells at the muscle–tendon interface (Fig. S1A). MCT cells, labeled by the marker TCF4, were rarely pSMAD⁺ (Fig. S1B). Furthermore, PAX7⁺ and MYOD⁺ myogenic cells occasionally accumulated pSMADs (Fig. S1B), consistent with the role of BMP signaling in postnatal satellite cells (Stantzou et al., 2017).

We abrogated BMP signaling in limb MPCs by crossing Lbx1^Cre mice (Sieber et al., 2007) with Rosa26^{LoxP-Stop-LoxP-huSMAD6-IRES-EGFP} (RS6) animals (Stantzou et al., 2017). In the resulting Lbx1^Cre;RS6 embryos, activation of the Lbx1 promoter in migrating limb MPCs induced Cre-mediated excision of the LoxP-Stop-LoxP cassette, leading to the expression of the inhibitory human SMAD6 (huSMAD6) and enhanced green fluorescent protein (EGFP). The Lbx1^Cre;RS6 genotype was detected at the expected frequency up to the fetal stages. However, new-born Lbx1^Cre;RS6 mice rarely survived, and the very few that did had severe growth retardation (Fig. S10A). We validated the activation of the transgenes in Lbx1^Cre;RS6 embryos using EGFP as a marker of successful recombination. EGFP fluorescence was detected in cells from the proximal central field of E10.5 forelimb buds and was absent from the RS6 controls (Fig. 2A). When compared with whole mount in situ hybridization (WISH) against Lbx1, the position of the EGFP fluorescence corresponded to that of migrating MPCs that populated the limb mesenchyme (Fig. 2A). In the forelimb buds of E12.5 Lbx1^Cre;RS6 embryos, EGFP was present in areas corresponding to the position of premuscle masses, as indicated by Myod (also known as Myod1) mRNA expression (Fig. 2A).

As the EGFP fluorescence was quite weak after cryosectioning, we used Ai9 mice, a Cre recombinase-dependent tandem dimer Tomato (tdTomato) reporter strain (Madisen et al., 2010), to generate Lbx1^Cre;RS6/LoxP-Stop-LoxP-tdTomato (Lbx1^Cre;RS6/Ai9) embryos. All tdTomato⁺ MPCs were also positive for EGFP and for PAX3, allowing the tracing of limb MPCs, which were depleted of BMP signaling during limb mesenchyme invasion (Fig. 2B). We did not find any tdTomato⁺/EGFP⁺ MPC accumulation in somites at limb level, nor aberrant migration into the anterior/posterior/distal limb margins (Fig. 2B). At E18.5, there was strong EGFP and tdTomato fluorescence in the limb muscles of Lbx1^Cre;RS6/Ai9 fetuses (Fig. 2C). TdTomato was present in all myofibers of E18.5 forelimbs, indicating high recombination efficiency (Fig. 2C). TdTomato expression was observed exclusively in developing muscles, indicating that MPCs depleted of BMP activity differentiated exclusively into muscle cells (Fig. 2C).

As Lbx1^Cre represents a loss-of-function allele due to insertion of the Cre transgene into the Lbx1 exon 1 coding sequence (Sieber et al., 2007), we determined whether heterozygous Lbx1^Cre mice show signs of haploinsufficiency. Myogenic marker WISH revealed a similar expression pattern in Lbx1 and Pax3 at E10.5 or Myod at E11.5 in Lbx1^Cre limbs compared with that in the RS6 controls (Fig. S10B). Furthermore, the Lbx1^Cre mice had normal viability and reproduction rates. We concluded that the loss of one functional Lbx1 allele did not cause haploinsufficiency, allowing us to use RS6 and Lbx1^Cre as controls for experiments with Lbx1^Cre;RS6 mutants.

In summary, these results show that following Cre-recombination, the huSMAD6-IRES-EGFP cassette was expressed exclusively in MPCs and their progeny in developing limbs of Lbx1^Cre;RS6 mice, allowing permanent overexpression of the BMP signaling inhibitor SMAD6 in cells of the myogenic lineage.

We confirmed via RT-qPCR that huSMAD6 was upregulated (3.7-fold) in the E18.5 forelimb muscles of Lbx1^Cre;RS6 fetuses compared with that of the RS6 controls (Fig. 2D). Next, we determined whether huSMAD6 overexpression caused cell-autonomous abrogation of BMP signaling in the myogenic lineage. Indeed, we observed the absence of pSMADs in PAX3⁺ MPCs in E10.5 Lbx1^Cre;RS6 limb buds compared with the RS6 controls (Fig. 3A). In addition, the presence of pSMADs at the tips of E14.5 muscle fibers was also lost (Fig. 3B). Moreover, WISH revealed that Lbx1 and Pax3 expression was strongly reduced in E10.5 Lbx1^Cre;RS6 limb buds compared with that in the RS6 controls (Fig. 3C). Residual Lbx1 and Pax3 transcripts were found in the proximal part of the limb buds. Similarly, Myod expression was strongly reduced in E11.5 and E12.5 limb buds from Lbx1^Cre;RS6 embryos compared with that from the RS6 controls (Fig. 3C). However, using WISH, we were unable to discriminate whether the decreased gene expression was due to a decrease in cell number or in the transcript number per cell.

We transversely cryosectioned E10.5 embryos at limb level, allowing for a proximodistal sectioning plane of the developing limb bud. Double immunofluorescence for PAX3 and the proliferation marker KI67 revealed ∼40% reduction of the entire PAX3⁺ cell population and a decline in the PAX3⁺/KI67⁺ subpopulation in Lbx1^Cre;RS6 embryos, suggesting reduced MPC proliferation after the inhibition of BMP signaling (Fig. 4A-C). The cell death marker cleaved Caspase-3 was absent in E10.5 limb mesenchyme in both genotypes, whereas it was present at trunk level and, as expected, at interdigital positions of E12.5 autopods (Fig. S2A). In addition, we analyzed the proximodistal distribution of the PAX3⁺ cell population in the E10.5 limb buds and found that total cell numbers were significantly reduced in the middle and distal parts of the limb bud in Lbx1^Cre;RS6 embryos compared with that in the RS6 controls (Fig. 4D,E). As total PAX3⁺ cell number in Lbx1^Cre;RS6 limbs was lower than in RS6 limbs, we also analyzed the normalized distribution of PAX3⁺ cells along the proximodistal axis. Such analysis revealed a decreased presence of normalized PAX3⁺ cell numbers in the distal parts of the limb, whereas there was a tendency towards increased cell numbers in the proximal parts (Fig. S2B). Next, we determined the distribution of PAX3⁺ cell in dorsal and ventral premuscle masses. We found a ∼40% reduction in cell numbers within the dorsal and ventral premuscle masses when comparing Lbx1^Cre;RS6 limbs with RS6 limbs (Fig. S2C), which accords with the loss in total PAX3⁺ cell number in Lbx1^Cre;RS6 limbs (compare with Fig. 4B). Cell numbers, however, were similar when comparing dorsal and ventral muscle masses of the same genotype (Fig. S2C). Together, these data suggest that the lack of BMP signaling in MPCs attenuated their proliferation and distal migration, and argue against a loss of MPCs by apoptosis or a rerouting of migration.

In the embryonic limb, PAX3 controls the entry of MPCs into the myogenic program (Relaix et al., 2005; Lagha et al., 2008). In E11.5 RS6 forelimbs, we observed a transition from PAX3 to PAX7 and MYOD expression. PAX7⁺ cells emerged in proximal pre-muscle masses (Fig. 5A). PAX3⁺ cells were located closer to the ectoderm, whereas MYOD⁺ cells were present closer to the core of the limb bud (Fig. 5B), consistent with the myogenic lineage progression from the peripheral towards central limb mesenchyme observed in developing chicken limbs (Amthor et al., 1998).

We found a precocious conversion of PAX3⁺ cells towards PAX7⁺ and MYOD⁺ cells in E11.5 Lbx1^Cre;RS6 limbs: the total number of PAX3⁺ cells decreased by 85%, whereas the total number of PAX7⁺ cells increased by 64% and the MYOD⁺ cells by 46% (Fig. 5A-E), thus the total PAX3/PAX7/MYOD population remained stable. In addition, the PAX3⁻/PAX7⁺ and PAX3⁻/MYOD⁺ cell population ratios increased by 68% and 61%, respectively, compared with that in the RS6 controls (Fig. 5F,G). These results suggest accelerated myogenic lineage progression in Lbx1^Cre;RS6 MPCs due to the absence of BMP signaling, which is similar to that shown in embryonic chicken limbs (Amthor et al., 1998).

In E12.5 Lbx1^Cre;RS6 limbs, the accelerated lineage progression was associated with a loss in total number of MYOD⁺ cells (42%) and PAX7⁺ cells (47%) (Fig. S3A-D). Furthermore, we detected a decline in PAX7⁺/KI67⁺ and PAX7⁺/MYOD⁺ cell populations, whereas the proportion of MYOD⁺/MYOG⁺ cells increased, confirming the shift of myogenic lineage progression towards differentiating myoblasts at the expense of proliferating precursors (Fig. S3B,E-G).

We analyzed the consequences of decreased MPC generation on primary myofiber formation after abrogation of BMP signaling by visualizing myofibers on transverse sections at the end of the embryonic period (E14.5) of mouse forelimb development. The zeugopod muscles of Lbx1^Cre;RS6 embryos were significantly smaller and contained about half the number of primary myofibers compared with the RS6 controls (Fig. 6). At this stage, we observed defective muscle patterning in the Lbx1^Cre;RS6 embryos. Whereas muscle pattern was normal at stylopod level, certain zeugopod muscles were either completely absent (supinator, extensor pollicis, flexor digitorum superficialis) or fused (extensor carpi radialis longus and brevis), whereas the remaining zeugopod muscles were remarkably hypoplastic (Fig. 6). At the autopod level, only a few remnant MHC-expressing cells were observed, and autopod muscles were entirely absent (Fig. 6). The anatomical changes in muscle pattern seen at the end-embryonic stage (E14.5) persisted during the fetal stage (Fig. 7).

We wanted to determine whether defective muscle patterning was also caused by the abrogation of BMP signaling in differentiated muscle cells. We used HSA-Cre driver mice to conditionally direct recombination in differentiated muscle cells. We first performed a time course to determine the spatiotemporal occurrence of HSA-Cre-driven recombination in HSA-Cre;Ai9 crosses by following the onset of tdTomato expression. In E10.5 and E11.5 embryos, tdTomato was found in somites but not in limb buds. TdTomato was present in developing limb muscles from E12.5 onwards, which is consistent with the emergence of primary myofibers at this stage (Fig. S4). We then generated HSA-Cre;RS6 mice to overexpress huSMAD6 exclusively in terminally differentiated muscles; a mouse model we have validated previously (Stantzou et al., 2017). The forelimbs of E18.5 HSA-Cre;RS6 fetuses developed normally and no change was detected in the muscle pattern (Fig. 7). These results may indicate that the information for the future muscle pattern is already present in MPCs before their differentiation. An alternative explanation may be that sufficient MPCs reached their destination (as migration and/or proliferation were not affected), allowing them to be exposed to patterning cues. Of note, MCT did not increase at the expense of skeletal muscle, as the pattern of collagen 12 expression in the HSA-Cre;RS6 fetuses was similar to that of the controls despite the smaller muscles (Fig. 7).

The observed changes in the muscle pattern of Lbx1^Cre;RS6 mutants (Fig. 7) resembled those previously observed in Hoxa11/d11 double mutants (Swinehart et al., 2013), raising the question of the intrinsic positional information of myogenic cells and putative regulation by BMP signaling.

Indeed, at E10.5, PAX3⁺ MPCs, which had left the dermomyotome and migrated into the limb bud, expressed HOXA11 protein. Notably, HOXA11 levels in the MPCs were higher than in the surrounding limb mesenchymal cells (Fig. 8A). As early as 1 day later, at E11.5, most PAX3⁺ cells had lost the high HOXA11 protein levels (Fig. 8B). In the absence of BMP signaling in Lbx1^Cre;RS6 embryonic limbs, the MPCs failed to accumulate high levels of HOXA11 protein (Fig. 8C compared with 8A; Fig. S5A).

To gain a global vision of Hox gene expression at single-cell resolution, we analyzed open-access single-cell RNA-sequencing (scRNA-seq) datasets of early chicken and mouse whole limb buds (Esteves de Lima et al., 2021; Rouco et al., 2021). Chicken and mouse limb buds have comparable Hox patterns in limb mesenchyme and myogenic differentiation (Pownall et al., 2002; Sundin et al., 1990; Yakushiji-Kaminatsui et al., 2018).

In chicken forelimb buds, scRNA-seq showed the expression of genes of the HOXA and HOXD clusters in mesenchymal cells and in muscle cluster cells (Figs S5B-H and S6). As an example, HOXA11 transcripts were detected in the majority (69%) of muscle cluster cells in E4 limbs (E10.5 mouse stage equivalence) and more rarely (37% of cells) by E6 (E12.5 mouse stage equivalence). In contrast, HOXD13 transcripts were not detected before E6, at which point its expression was limited to a few muscle cluster cells (Fig. S5B). HOXA11 was expressed at all successive steps of the myogenic process: first in PAX7⁺ and MYOD⁺ MPCs at E4 and E6, and then in MYOG⁺ myoblasts at E6 (Fig. S5C-G). There was nonetheless a drop in HOXA11 expression during myogenic lineage progression, given that at E6, 39% of PAX7⁺ muscle cluster cells co-expressed HOXA11, whereas only 15% of MYOG⁺ cells co-expressed HOXA11 (Fig. S5H). Interestingly, we found a heterogeneous combinatorial expression of HOXA genes in single muscle cluster cells, suggesting Hox-dependent positional information in chicken limb MPCs (Fig. S6A,B).

We next analyzed whether BMP response correlates with HOXA gene expression in muscle cluster cells and found that HOXA⁺ cells (expressing one or several genes of the HOXA cluster) expressed BMP downstream effector genes ID2 and ID3, but not ID1, in higher proportion compared with HOXA⁻ cells (Fig. S6C-E). Consistently, the BMP score, which is the corrected average expression of the ID1, ID2 and ID3 genes, was significantly higher in HOXA⁺ cells compared with HOXA⁻ cells (Fig. S6F).

scRNA-seq of E12.5 mouse forelimbs showed very similar results compared with those in chicken: (1) muscle cluster cells expressed genes of the Hoxa and Hoxd clusters (Fig. S7A); (2) a subset of Hoxa11⁺ muscle cluster cells co-expressed Pax3, Pax7, Myf5 and Myod, but only rarely Myog (Fig. S8); (3) muscle cluster cells showed large heterogeneity in the expression of genes of the Hoxa cluster (Fig. S7B); and (4) there was Id1, Id2 and Id3 expression in a higher proportion of Hoxa⁺ cells than Hoxa⁻ cells (Fig. S7C-E).

The current paradigm of limb muscle patterning considers limb MPCs as naïve, where they develop individual muscles by invading a prepattern established by MCT (Kardon et al., 2003). Our results contribute to this concept by showing that BMP signaling (produced by limb connective tissue cells surrounding developing muscles; Amthor et al., 1998; Esteves de Lima et al., 2021) is necessary for the generation of MPCs responsive to BMP, thereby establishing the necessary cellular source for limb muscle pattern. We used overexpression of huSMAD6 as an experimental tool to abrogate BMP signaling, which caused precocious loss of PAX3 in MPCs and accelerated myogenic lineage progression. MPCs advanced less distally, as expected, as PAX3 is a prerequisite for myogenic migration (Bober et al., 1994), likely causing a mismatch between their distal progression and the local connective tissue, and thus responsible for the observed defects in proximodistal muscle pattern.

It has been shown that impaired distal MPC migration can cause varying degrees of limb muscle defects (Brohmann et al., 2000; Shin et al., 2016; Vasyutina et al., 2005). A detailed anatomical analysis of these mouse mutants would be required to determine whether different signaling cues, e.g. scatter factor/hepatocyte growth factor (SF/HGF) as compared to BMPs, exert distinctive roles during muscle patterning. In the absence of such comparative anatomical analysis, however, we cannot exclude the possibility that migration defects, independently of the underlying molecular mechanisms, result in a generic patterning defect.

We demonstrated that limb MPCs expressed Hox genes in mouse as well as in chicken embryos. scRNA-seq revealed: (1) a high proportion of Hox-expressing MPCs in early limb buds; (2) heterogeneity of Hox gene expression in the MPCs; (3) their sequential upregulation; and (4) their downregulation during myogenic lineage progression. Immunohistochemistry confirmed the transcriptome data: upon leaving the dermomyotome and entering limb bud mesenchyme, PAX3⁺ migrating limb MPCs produced high HOXA11 protein levels, and this was dependent on BMP signaling. Interestingly, the Lbx1^Cre;RS6 mutants resembled the muscle pattern defect observed in Hoxa11^−/−/d11^−/− dKO mutants (Swinehart et al., 2013). Swinehart et al. showed that Hoxa11 was not expressed by differentiated muscle cells at E14.5, but in cells surrounding primary muscle fibers, such as TCF4⁺ connective tissue cells. Whether HOXA11 colocalizes with MPCs (which also surround primary muscle fibers), however, was not investigated (Swinehart et al., 2013). It has also been shown that Hoxa13^−/− KO and Hoxa13^−/−/d13^−/− dKO disturb autopod development (Fromental-Ramain et al., 1996). Here, we found both Hoxa13 and Hoxd13 being expressed by MPCs. However, we examined scRNA-seq data sets from whole limb buds, which did not allow us to specify which MPC subpopulation (e.g. autopod MPCs) expressed which Hox code. Thus, it remains to be determined whether Hox gene expression in MPCs follows the collinearity in the developing limb. We can therefore only speculate about the exact role of Hox genes in developing muscle, and how their expression relates to BMP signaling. Previous work on chick limb MPCs showed that Hoxa11 and Hoxd13 blocked expression of MyoD, and that Hox gain-of-function experiments resulted in distorted limb muscle patterning (Yamamoto and Kuroiwa, 2003). Many questions, however, remain unresolved. Does the Hox code control MPC proliferation, myogenic lineage progression and muscle splitting? Do MPCs, through Hox code, acquire positional identity and establish a muscle pre-pattern? Alternatively, herein observed loss in specific muscles may simply result from a tissue default that is caused by insufficiently generated precursors.

Curiously, we show that MPCs were the only limb cells that showed robust BMP-dependent pSMAD expression at early limb bud stages, implying a high dependency of MPCs on BMP signaling. However, we neither explored the source of BMPs, nor which ligands of the BMP family signal to limb MPCs. In previous work, early migrating MPCs were found surrounded by BMP2/4/7-expressing cells at limb margins and ectopically applied BMP altered the positioning of premuscle masses in chick embryos (Amthor et al., 1998). Similar expression of BMPs in limb margins was also observed in mouse embryos (Michos et al., 2004). It remains to be determined whether long-range BMP signaling from limb margins could regulate MPCs. Alternative sources, including expression by MPCs themselves, must be considered. Of note, triple knockout of BMP2/4/7 in the apical ectodermal ridge caused polydactyly and does not affect limb outgrowth, whereas overexpression of the BMP antagonist gremlin in entire limb mesenchyme prevented limb outgrowth altogether (Choi et al., 2012; Norrie et al., 2014). However, muscle development has not been analyzed in these mutants.

The majority of MPCs is derived from migratory and Pax3-dependent MPCs of somite origin. However, recent work demonstrated a dual origin of MPCs in the developing limb: a small population of MCT cells is integrated into myotubes at muscle tips close to tendons in chicken and mouse muscles (Esteves de Lima et al., 2021; Yaseen et al., 2021), a process being promoted by BMP signaling (Esteves de Lima et al., 2021). BMP gain- and loss-of-function experiments in chicken embryos demonstrated that BMP signaling balances the fibroblast-myoblast conversion and consequently the muscle pattern (Esteves de Lima et al., 2021). Here, we show that BMP signaling also regulates the somite-derived Pax3-dependent MPC lineage in mouse limbs. Cell-autonomous inhibition of BMP signaling in somite-derived MPCs caused absence of entire muscles. Therefore, MCT depends on the presence of somite-derived MPCs and are lost secondarily when muscle fails to develop. Further, the generation of the somite-independent muscle lineage depends on the presence of somite-derived muscle.

We found that the patterning defect in Lbx1^Cre;RS6 limbs persisted from embryonic to fetal stages, showing that secondary myogenesis cannot compensate for embryonic muscle defects, and remaining muscles continue to grow despite persistent inhibition of BMP signaling.

We would like to emphasize that our results support the MCT prepattern model (see Fig. S9). We believe that mesenchymal cells that form future MCT are the source of cues, including BMPs, that inform MPCs of where to migrate and proliferate. MCT and MPCs could be mutually dependent on each other to establish the muscle pattern. Indeed, a defined MCT pattern resembling a muscle pattern failed to develop in muscle-devoid limbs (Christ et al., 1977). Whereas tendons initially developed autonomously in lack of muscles; they degenerated secondarily (Christ et al., 1977; Schweitzer et al., 2010).

We here employed the overexpression of huSMAD6 as a mean to test the cell-autonomous effect of abrogating BMP signaling. SMAD6 inhibits Smad signaling by the BMP type I receptors ALK-3/6 subgroup and only weakly inhibits TGF-β/activin signaling via the BMP type I receptors ALK-1/2 subgroup, the latter being a preferential target of SMAD7 (Goto et al., 2007; Miyazawa and Miyazono, 2017). As SMAD6 is not a direct component of the BMP signaling cascade, further work is required to substantiate our results, such as performing specific BMP receptor knockout. In previous work, we showed that satellite cell-specific overexpression of SMAD6 or knockout of Alk3, or overexpression of the BMP antagonist Nog in postnatal mice decreased proliferation of satellite cells, diminished their accretion during myofiber growth and retarded muscle growth, whereas overexpression of SMAD6 exclusively in terminally differentiated myofibers did not affect satellite cell-dependent muscle growth (Stantzou et al., 2017). Together with herein presented results, this confirms that BMP signaling acts in a similar cell-autonomous manner in MPCs during prenatal and postnatal development.

In conclusion, our data suggest that BMP signaling controls embryonic limb MPCs to maintain PAX3-expressing precursor status, coordinates MPC migration, proliferation and myogenic lineage progression, thereby providing the cellular source that is required for building the correct muscle pattern. The expression of the Hox code in MPCs may indicate that positional identity is established before the splitting of premuscle masses into individual muscles. Future loss- and gain-of-function experiments are required to directly test the function of Hox gene expression in MPCs.

We conducted all animal experiments according to national and European legislation as well as institutional guidelines for the care and use of laboratory animals as approved by the French government.

The following mouse lines have been previously described: Lbx1^Cre mice (Sieber et al., 2007), HSA-Cre transgenic mice (Miniou et al., 1999), Rosa26^{LoxP-Stop-LoxP-huSMAD6-IRES-EGFP} mice (i.e. RS6) (Stantzou et al., 2017), and Ai9 mice, which contain an insertion in the Rosa26 locus of a strong and ubiquitous CAG promoter, followed by a floxed-Stop cassette-controlled tdTomato (Madisen et al., 2010).

Lbx1^Cre, HSA-Cre and RS6 mice were interbred to obtain Lbx1^Cre;RS6 and HSA-Cre;RS6 embryos. Lbx1^Cre and Ai9 mice were interbred to obtain heterozygous Lbx1^Cre;Ai9 mice, which were crossed with RS6 mice to obtain Lbx1^Cre;RS6/Ai9 embryos.

Genomic DNA isolated from ear clippings postnatally, or from yolk sacs or parts of the non-limb tissues prenatally, were genotyped. The PCR primers are described in Table S1.

Embryos and fetuses were collected in ice-cold phosphate-buffered saline (PBS) at different stages (plug date: E0.5). Embryos used for WISH experiments were fixed overnight in 4% paraformaldehyde (PFA) at 4°C, washed twice in PBS-T (PBS + 0.1% Tween-20, P9416, Sigma-Aldrich) and dehydrated using a methanol series of 50% methanol (15 min×2) and 100% methanol (15 min), after which they were stored at −20°C for WISH.

For immunostaining, forelimbs from E10.5 and E11.5 embryos were dissected as pairs connected with the rostral (thoracic) body segment to preserve the structure of the forelimbs and forelimb-level somites. At the later stages, the forelimbs were individually dissected. The tissues were fixed at 4°C in either 1% PFA for 1 h (E10.5-E12.5) or in 4% PFA for 2 h (E14.5 and E18.5), washed thrice for 10 min and then dehydrated overnight at 4°C in either 15% sucrose (E10.5-E12.5) or 30% sucrose (E14.5 and E18.5). The forelimbs were embedded in Optimum Cutting Temperature compound (Qpath) in disposable plastic molds (Dutscher), frozen in liquid nitrogen and stored at −80°C for sectioning.

Total RNA from frozen E18.5 forelimb muscle tissue was extracted using TRIzol (Life Technologies Ambion) in combination with an RNeasy Mini kit (Qiagen). Traces of DNA in the RNA extract were removed with an RNase-Free DNase Set (Qiagen). The isolated RNA was quantified using a NanoVue Plus GE HealthCare spectrophotometer (Dutscher). Next, complementary DNA (cDNA) was synthesized using reverse transcriptase (SuperScript™ III First-Strand Synthesis SuperMix kit, Invitrogen). RT-qPCR was performed according to the SYBR Green protocol (Bio-Rad) in triplicate on a CFX96 Touch Real-Time detection system (Bio-Rad) using iTaq Universal SYBR Green Supermix (Bio-Rad) and primers for huSMAD6 and the housekeeping gene Gapdh as described previously (Stantzou et al., 2017).

WISH with digoxigenin-labeled probes was used for visualizing the expression of Lbx1, Pax3 or Myod. WISH was performed as previously described (Murgai et al., 2018; Tajbakhsh et al., 1997).

Serial sections of frozen forelimbs on SuperFrost Plus adhesion slides (Thermo Fisher Scientific) were obtained at 10-µm thickness using a cryostat (Leica, CM3050S) at −24°C. E10.5 and E11.5 forelimbs were longitudinally sectioned, which allowed 2D visualization of forelimb sections in the proximodistal and dorsoventral axes. E12.5, E14.5 and E18.5 forelimbs were sectioned in the transverse plane (except where stated otherwise), which allowed 2D visualization of forelimb sections in the dorsoventral and anteroposterior axes. The forelimb sections on the slides were directly used for immunofluorescence staining experiments or were stored at −80°C for future use.

Immunofluorescence staining was performed using the following protocols: (1) rehydration of slides in PBS for 5 min; (2) permeabilization with 0.1% (E10.5-E12.5) or 0.5% (E14.5 and E18.5) Triton X-100 (Sigma-Aldrich) (in the case of nuclear protein staining, e.g. PAX3, PAX7, MYOD, MYOG, pSMAD1/5/9, HOXA11, KI67, Caspase-3) or with methanol (for non-nuclear proteins, e.g. MHC, laminin alpha 2, DsRed, collagen 12) at −20°C; (3) three 5-min washes in PBS; (4) antigen retrieval by 20-min immersion of the slides in boiled 10 mM citric acid solution kept at 60°C in a water bath; the slides were cooled at room temperature in the citric acid solution, and three PBS washes were performed (only for E14.5 and E18.5 nuclear staining); (5) up to 1.5-h blocking with 10% normal goat serum (Abcam); (6) overnight incubation with primary antibodies (dilutions prepared in blocking solution, Table S2) at 4°C; (7) three 5-min washes in PBS; (8) up to 1.5-h incubation with secondary antibodies (dilutions prepared in blocking solution, Table S3); (9) three 5-min washes in PBS; (10) incubation with DAPI for nuclear staining (10 min; dilution 1:5000); (11) washing in PBS for 5 min; (12) coverslip mounting with Fluoromount-G (Southern Biotech). Tables S2 and S3 detail the primary and secondary antibodies used in this study.

Embryos were dissected and whole limbs in the unfixed state were immediately imaged for native EGFP and tdTomato fluorescence using a stereomicroscope (SteREO Lumar.V12, Zeiss). Native EGFP and tdTomato fluorescence was subsequently imaged on fresh unfixed cryosections, and fluorescence immunohistochemistry was captured under 20×, 40× or 63× objective using a fluorescence microscope (Zeiss Axio Imager) with an Orkan camera (Hamamatsu). Images were acquired using AxioVision software. Mosaic images of immunostained whole limbs were obtained after stitching together multiple individual images captured with a 20× objective of all the different regions in the whole limb for all fluorescence channels of interest. Masson's trichrome staining images were acquired using a digital slide scanner (Leica) and analyzed with ImageScope software. Images were exported and saved as TIFF files for further analyses or for illustration in the figures.

The captured fluorescent images were analyzed by applying morphometric studies using ImageJ (Schneider et al., 2012). The populations of cell and nuclear markers (PAX3, MYOD, PAX7, MYOG, KI67, HOXA11, DAPI) were quantified on immunostained cryosections as detailed above by superimposing fluorescence channels to visualize signal colocalization.

Proximodistal migration of PAX3-expressing MPCs in E10.5 forelimbs was quantified by dividing the forelimb into ten equally sized proximodistal zones and counting the PAX3⁺ cells in each zone. Myofibers on the transverse sections of E14.5 forelimb zeugopods were quantified following co-immunostaining against laminin alpha 2 and MHC. Total muscle cross-sectional area (CSA) was determined as the sum of the CSA of all individual zeugopod muscles.

Morphometric studies at E10.5, E11.5 and E12.5 were conducted on three consecutive sections of each forelimb, and n=5 forelimbs were analyzed for each genotype, except where stated otherwise. Morphometric studies at E14.5 were conducted on one transverse section through the proximal region of forelimb zeugopods, assuring measurements at the maximal size of the zeugopod muscle.

The scRNA-seq protocol for E12.5 mouse whole limb cells is described by Rouco et al. (2021); that for chicken whole limb cells is described by Esteves de Lima et al. (2021). Briefly, scRNA-seq datasets were generated from whole forelimbs from two different E4 embryos and three different E6 embryos using a 10x Chromium Chip (10x Genomics) followed by sequencing with a High Output Flow Cell using an Illumina Nextseq 500 and by sequence analysis with Cell Ranger Single Cell Software Suite 3.0.2 (10x Genomics). Only mononucleated muscle cells were included in the datasets, as plurinucleated myotubes were excluded by the single-cell isolation protocol. Downstream clustering analysis of scRNA-seq data was performed using the Seurat package (v3.0) (Stuart et al., 2019) under R (‘The R Project for Statistical Computing’, v3.6.1) (Macosko et al., 2015). We then extracted the clusters identified as muscle clusters by the differential expression of the classical myogenic markers (PAX7, MYOD, MYOG) and performed the remaining analysis on these muscle clusters only. Gene expression was defined by ‘gene log-normalized count >0’. The scRNA-seq datasets were analyzed using Seurat tools: FeaturePlot and Violin plots. Custom feature plots highlighting gene co-expression were generated using the R package ggplot2 v3.3.3 (Wickham, 2016). Population intersection plots were generated with the R package UpSetR v1.4.0 (Conway et al., 2017).

Within the muscle clusters, cells were grouped according to two identities, i.e. whether or not they expressed HOXA (HOXA⁺ and HOXA⁻, respectively). The HOXA⁺ identity was defined by the expression (i.e. gene log-normalized count >0) in a cell of at least one of the seven HOXA genes found in the muscle clusters (HOXA4, HOXA5, HOXA6, HOXA7, HOXA9, HOXA10, HOXA11). HOXA⁻ identity was conferred to cells that expressed none of the seven HOXA genes. A BMP score was calculated using the AddModuleScore function for the well-characterized BMP transcriptional read-out genes ID1, ID2 and ID3. Response to BMP signaling was then compared between these two identities using the Seurat tool Violin plots and the ggplot2 tool boxplots.

Both chicken and mouse scRNA-seq datasets used are deposited in the National Center for Biotechnology Information Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/), respectively under accession numbers GSE166981 and GSE168633.

Numerical data are presented as the mean±s.d. The probability for statistical differences between experimental and control groups was determined by calculating the exact P-value using the non-parametric two-tailed Mann–Whitney U-test. GraphPad Prism Software version 7.00 for Windows (www.graphpad.com) was used for all statistical analyses and graphs.

We thank Prof. Dr Carmen Birchmeier (Developmental Biology and Signal Transduction Group, Max Delbrück Center for Molecular Medicine, Berlin, Germany) for providing the Lbx1^Cre mice. Some of the text and figures in this paper formed part of Dr Hasan Asfour's PhD thesis in the department ‘UFR Simone Veil – Santé’ and in the unit ‘UVSQ, Inserm, END-ICAP’, at Paris-Saclay University, France, in 2021. We thank the animal and imaging facility of the UFR Simone Veil - Santé.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201504.reviewer-comments.pdf.