ABSTRACT
Body skeletal muscles derive from the paraxial mesoderm, which forms in the posterior region of the embryo. Using microarrays, we characterize novel mouse presomitic mesoderm (PSM) markers and show that, unlike the abrupt transcriptome reorganization of the PSM, neural tube differentiation is accompanied by progressive transcriptome changes. The early paraxial mesoderm differentiation stages can be efficiently recapitulated in vitro using mouse and human pluripotent stem cells. While Wnt activation alone can induce posterior PSM markers, acquisition of a committed PSM fate and efficient differentiation into anterior PSM Pax3+ identity further requires BMP inhibition to prevent progenitors from drifting to a lateral plate mesoderm fate. When transplanted into injured adult muscle, these precursors generated large numbers of immature muscle fibers. Furthermore, exposing these mouse PSM-like cells to a brief FGF inhibition step followed by culture in horse serum-containing medium allows efficient recapitulation of the myogenic program to generate myotubes and associated Pax7+ cells. This protocol results in improved in vitro differentiation and maturation of mouse muscle fibers over serum-free protocols and enables the study of myogenic cell fusion and satellite cell differentiation.
INTRODUCTION
Skeletal muscles represent a major derivative of the embryonic paraxial mesoderm. Presomitic mesoderm (PSM) cells, first produced in the primitive streak and later the tail bud, express specific sets of genes such as mesogenin 1 (Msgn1) and experience periodic signaling driven by the segmentation clock (Hubaud and Pourquié, 2014). As the cells pass the so-called determination front in the anterior PSM, they acquire their segmental identity and activate expression of genes such as Pax3 that control their subsequent differentiation. Epithelial somites form at the anterior tip of the PSM and somitic cells soon activate myogenic and chondrogenic differentiation programs (Chal and Pourquié, 2009, 2017). At the trunk level, the dorsal region of epithelial somites forms the dermomyotome, which contains the Pax3-expressing myogenic precursors. A subset of these precursors located in the dermomyotome lips first activates Myf5 and then Mrf4, Myod1 and Myog, and gives rise to mononucleated post-mitotic myocytes that form the myotome (Kassar-Duchossoy et al., 2004). Precursors of the limb and girdle muscles delaminate from the ventro-lateral lip of the dermomyotome to migrate to their final locations, where they activate the myogenic program. Myocytes subsequently fuse in a highly patterned manner to first generate myotubes, which further mature into myofibers. Thereafter, muscle fibers continue to be formed during embryogenesis from a pool of proliferating precursors expressing Pax3 and Pax7 (Hutcheson et al., 2009). These cells ultimately form the embryonic, fetal and adult muscle fibers, and the satellite cells (Biressi et al., 2007).
We recently identified Rspo3 as a secreted protein expressed in the posterior PSM able to induce the differentiation of embryonic stem cells (ESCs) towards a Msgn1+ posterior PSM fate when combined with BMP inhibitors in chemically defined conditions (Chal et al., 2015). These progenitors can further differentiate into anterior PSM Pax3+ fates and can be used to subsequently generate large amounts of muscle cells in vitro and in vivo. Here, we describe a robust, serum-containing protocol based on Wnt signaling activation and BMP inhibition to efficiently produce PSM-like cells from mouse ESCs and human induced pluripotent stem cells (iPSCs). We show that the transcriptome of in vitro differentiating mouse ES cells exhibits conserved kinetics of gene activation with mouse PSM cells in vivo. Furthermore, these induced PSM cells can engraft into adult injured muscles and generate large amounts of immature skeletal muscle fibers, supporting true paraxial mesoderm commitment. Finally, with this serum-containing protocol, the PSM precursors could be further differentiated in vitro into muscle fibers that are more mature and can be cultured for longer than those generated in serum-free conditions, thus providing an ideal system for studying myogenesis in vitro using mouse ESCs.
RESULTS
Comparison of the transcriptional landscape between the PSM and posterior neural tube in the mouse embryo
We have previously reported the generation of microarray series of consecutive micro-dissected fragments of the E9.5 mouse PSM (Chal et al., 2015) and used them to identify lists of 40-50 highly specific signature genes for the posterior and anterior PSM domains. This strategy identified members of the Wnt, FGF and Notch signaling pathways known to be involved in PSM specification and differentiation (Chal et al., 2015). Many other PSM signature genes were associated with signaling pathways that have not been as thoroughly studied in the paraxial mesoderm (Table S3, S4). To characterize further their expression pattern, we performed systematic whole mount in situ hybridization (Fig. 1A). Genes strongly expressed in the posterior PSM include carbohydrate (N-acetylglucosamino) sulfotransferase 7 (Chst7), the gene regulated by estrogen in breast cancer product (Greb1), tropomyosin alpha 1 (Tpm1), EGF domain-specific O-linked N-acetylglucosamine (GlcNAc) transferase (Eogt), the previously identified mouse cyclic gene Tnfrsf19 (Troy), as well as the apelin receptor (Aplnr/Apj) (Kälin et al., 2007) and the sphingosine phosphate receptors (S1pr3 and S1pr5) (Ohuchi et al., 2008) (Fig. 1A, data not shown). Some genes were expressed in the entire PSM but not in somites, and included Greb1l and the interferon induced transmembrane protein 1 (Ifitm1, fragilis 2) (Tanaka and Matsui, 2002). Genes enriched in the anterior PSM included adducin gamma (Add3), the lipoma HMGIC fusion partner-like 2 (Lhfpl2) and fibrillin 2 (Fbn2) (Fig. 1A). A subset of genes was expressed as stripes in the anterior PSM and comprised myocardin (Myocd), vitronectin (Vtn), Shroom3 (Sousa-Nunes et al., 2003), and a number of genes with unknown function in the PSM including Abca1, Arg1 (Hou et al., 2007), Ism1 (Tamplin et al., 2008) and Pgm5 (Fig. 1A, data not shown). Thus, our data identify a novel set of molecular players with tightly restricted spatio-temporal expression in the mouse PSM.
We next compared the transcriptional program of differentiating PSM tissue with that of the adjacent neural tube. Microarray series of consecutive micro-dissected fragments of the posterior neural tube region adjacent to the PSM were generated from two different E9.5 mouse embryos. Each series comprised six contiguous ∼100 µm long neural tube fragments spanning from the tail bud to the level of the newly formed somite (S0) (Fig. 1B). As for the PSM, RNA extracted from each fragment was used to hybridize a single Affymetrix microarray. We first confirmed that known genes activated during neural tube differentiation, including Pax6, Neurog1, Nrcam, Nkx6.1 and Nkx1.2, showed the expected expression gradients in the neural tube, thus validating the neural tube series (Fig. 1C). Using clustering analysis, we have previously identified an unbiased molecular subdivision in the PSM series that corresponds to the determination front at which level the segmental pattern is first established (Chal et al., 2015). Such a defined demarcation was not observed when a similar analysis was performed for the neural tube series, suggesting that progressive transcriptional changes accompany early differentiation of the neural lineage (Fig. 1B, Table S5, S6). Thus, our data identify novel PSM marker genes and argue for a different mode of transcriptome regulation during contemporary stages of paraxial mesoderm and neural tube development.
R-spondin/Wnt signaling activation in combination with BMP inhibitors promotes posterior PSM differentiation of mouse ES and human iPSCs
In vivo, the first stage of paraxial mesoderm cell differentiation is characterized by the activation of the gene mesogenin 1 (Msgn1), which codes for a basic helix-loop-helix transcription factor specifically expressed in the posterior PSM (Yoon et al., 2000). Examination of the PSM microarray series allowed the identification of the secreted Wnt agonist R-spondin3 (Rspo3), which is strongly expressed in the posterior PSM (Chal et al., 2015). Treatment of monolayers of the Msgn1-repV mES reporter cells (Chal et al., 2015) with Rspo3 in fetal bovine serum (FBS)-containing medium, strongly induced Msgn1-Venus+ cells compared with control medium, reaching up to 70% after 4 days (Fig. 2A). Dimethyl sulfoxide (DMSO) has been shown to promote differentiation of several embryonic cell types (McBurney et al., 1982; Chetty et al., 2013). We found that addition of DMSO synergized with Rspo3 to induce Venus+ cells in both serum-containing and serum-free media (Fig. S1, data not shown). Thus, 0.5% DMSO was systematically added to the differentiation medium.
To confirm the PSM identity of induced cells, we generated microarrays from FACS-sorted Msgn1-Venus+ cells cultured for 3 and 4 days in Rspo3-containing medium (RD) and compared them with our previous PSM arrays. At day 3, differentiated ESCs expressed many posterior PSM marker genes, including Tbx6 (Fig. 2B and data not shown). Unexpectedly, Venus+ cells differentiated for 4 days in the presence of Rspo3 were also enriched for Bmp4, which is expressed in the lateral plate mesoderm in the posterior region of the embryo (Fig. 2B). Accordingly, Msgn1-repV+ cells induced in RD media also expressed the lateral plate marker Foxf1a (Fig. 2B). This suggests that treatment with Rspo3 alone leads cells to acquire a mixed identity between paraxial mesoderm and lateral plate.
In order to prevent cells from acquiring a lateral plate identity, we added the BMP inhibitor Noggin to the culture medium. Microarray and qPCR analysis indicated that Noggin-treated cells downregulated Bmp4 and Foxf1a, and upregulated Tbx6 (Fig. 2B). Importantly, after 4 days in culture, Noggin-treated cells upregulated the anterior PSM markers, Ripply2 and Pax3 (Fig. 2B). Stronger inhibition of Bmp4 activity was observed with the chemical BMP inhibitor LDN 193189 (hereafter named Ldn) (Cuny et al., 2008). RT-qPCR analysis on FACS-sorted Msgn1-Venus+ cells revealed that Ldn applied together with Rspo3 on Msgn1-repV cells acts in a dose-dependent fashion, inhibiting Foxf1a and Bmp4 expression, and promoting expression of Tbx6, Hes7, Msgn1 and Pax3 much more efficiently than Noggin (Fig. 2C). Addition of Noggin or Ldn to the medium was found to lead to lower levels of induction of Venus+ cells from the Msgn1-repV cultures (data not shown). Because Ldn was much more effective than Noggin at repressing lateral plate mesoderm fate, we systematically used it for myogenic differentiation thereafter. Substituting Rspo3 with Rspo2 also led to similar results (Fig. S1). Addition of Dkk1 inhibited the induction of Venus+ cells from Msgn1-repV, arguing that Rspo3 action is downstream of the canonical Wnt/β-catenin pathway (Fig. 2D). Accordingly, substituting Rspo3 with the GSK3-inhibitor and canonical Wnt activator CHIRON 99021 (Chir) led to comparable induction level of the Msgn1-repV+ population (Fig. 2E, Table S7 and data not shown). These data demonstrate that Rspo3 efficiently promotes the induction of a Msgn1-positive cell population from mouse ESCs by activating the canonical Wnt/β-catenin pathway.
We next analyzed the in vitro induction of the posterior PSM fate from human iPSCs using a Mesogenin1-Venus iPS reporter line generated by knock-in of a MSGN1-Venus fusion using CRISPR/Cas9 (Fig. 3A). Human MSGN1-Venus reporter cells were differentiated in serum-free medium containing Chir and Ldn (CL medium) for 4-5 days to induce paraxial mesoderm differentiation (Chal et al., 2016, 2015). By day 3, Venus expression was detected and the MSGN1 transcript was almost exclusively found in the Venus-positive (hM+) cell fraction (Fig. 3B). By day 4, flow cytometry analysis demonstrated that up to 95% of the differentiating human iPSCs were Venus+, both in CL and in Chir-only (C) conditions (Fig. 3C, data not shown). Venus+ cells induced in the presence of Ldn also expressed the PSM marker TBX6, validating the specificity of the reporter and the efficiency of the protocol to induce posterior PSM fate in human cells (Fig. 3D).
Next, we compared the gene expression profile of FACS-sorted MSGN1-Venus+ cells generated in serum-free CL medium with undifferentiated hPSCs (Fig. 3E, Table S8). Venus+ cells strongly upregulated posterior PSM marker genes, including MSGN1, RSPO3, TBXT (brachyury), HES7 and LFNG (Fig. 3E). To evaluate the impact of BMP inhibition on the gene signature of induced hM+ cells, microarrays from FACS-sorted hM+ cells cultured in serum-free CL or Chir-only (C) medium were compared (Fig. 3F). Genes of the Wnt (AXIN, DKK1, DKK2, RSPO3, LEF1, WNT3A), Notch (DLL1, DLL3, LFNG, HES7) and FGF (DUSP1, FGF10, FGF8, FGF9) signaling pathways, which are important for posterior PSM development, were found to be upregulated in Venus+ cells induced under CL conditions (fold change >2 and false discovery rate <10%). By contrast, hM+ cells induced in the presence of the WNT activator alone upregulated lateral plate and cardiac markers, including FOXF1, BMP4, TBX3, HAND1, HAND2 and CXCR7, while posterior PSM markers were expressed at a lower level compared with cells differentiated in CL medium (Fig. 3F,G, Table S9). Strikingly, Pax3 was found to be unexpectedly enriched in absence of LDN (Fig. 3F). Although this could reflect early neuromesodermal progenitors having differentiated to a non-paraxial mesoderm fate (Pax3+ neural, neural crest) in the absence of BMP inhibition, we cannot exclude that Pax3 is also expressed in a subset of lateral plate/intermediate mesoderm derivatives in humans. Together, these data suggest that these cells exhibit a mixed paraxial/lateral plate mesoderm identity. Thus, as for mouse ESCs, efficient induction of paraxial mesoderm from human iPSCs requires BMP inhibition to prevent early paraxial mesoderm cells from drifting to a lateral plate mesoderm fate.
We next compared human MSGN1-Venus+ progenitors induced in vitro by Wnt activation and BMP inhibition with their mouse Msgn1-Venus+ counterparts (Chal et al., 2015). Both populations upregulated a large number of posterior PSM signature genes, including MSGN1, TBX6, TBXT, WNT5A, RSPO3, CDX2 and EVX1 (Fig. 3H; Table S7, S8). Interestingly, human MSGN1-Venus+ progenitors expressed genes such as DKK2 and CXCR4, which were detected at significantly lower levels in mouse Msgn1-Venus+ cells, whereas mouse M+ cells showed enriched expression of some FGF pathway components (Dusp2, Dusp4, Fgf3, Fgf15) over their human counterparts. Altogether, our data indicate that, as reported for mouse ESCs, in vitro differentiation of human posterior PSM-like cells can be induced in the presence of a WNT activator and a BMP inhibitor. Moreover, our data suggest that the transcriptomes of mouse and human posterior PSM-like cells are highly similar.
Generation of Pax3+ anterior PSM-like precursors from ESCs in vitro
During paraxial mesoderm maturation in vivo, PSM cells downregulate Msgn1 and activate Pax3 as they become located in the anterior PSM domain. FACS-sorted Msgn1-Venus+ mouse cells obtained after 3 days of differentiation in medium containing Rspo3 or CHIR, plus DMSO and Ldn (RDL or CDL, respectively), strongly expressed the posterior PSM markers Tbx6 and Hes7 compared with Msgn1-Venus− cells (Figs 2E and 4A). Msgn1-Venus+ cells obtained from 4-day-old cultures in RDL medium activated the anterior PSM and somitic markers Pax3, Meox1/2, Tcf15, Foxc1/2, Tbx18 and Uncx4.1 (Fig. 4A-C). Co-expression of these genes with the Venus protein is most likely due to the stability of the reporter that persists for some time in cells that have ceased expressing Msgn1. Msgn1-Venus+ cells induced in basal (FBS 15%, F15) medium without Ldn strongly upregulated BMP4 at day 4-5 and did not transit to an anterior PSM fate efficiently (Fig. 4A-C). This transition to an anterior PSM fate was highly dependent on the concentration of Ldn, with 100 nM leading to the highest induction of anterior PSM markers and the lowest expression levels of lateral plate markers (Fig. 4B). Chir (CDL medium) was found to be as efficient as R-spondin3 at inducing the anterior PSM fate (Fig. 4C), whereas cells differentiated in base F15 medium failed to activate the anterior PSM program. Therefore, our data suggest that mouse ESCs differentiated in serum-containing medium supplemented with a WNT activator and a BMP inhibitor are able to efficiently progress to an anterior PSM fate and to activate the early somitic differentiation program, as reported in serum-free conditions (Chal et al., 2015).
Next, we took advantage of the mouse Pax3-GFP ES reporter line to characterize anterior PSM/somitic precursors generated in vitro (Chal et al., 2015). When these cells were differentiated for 5 days in F15 or in RD medium, only about 1% of GFP+ cells were detected (Fig. 5A). While addition of Noggin (RDN) led to similar results, in contrast, in Ldn-containing RDL medium, up to 20-40% of Pax3-GFP+ cells were observed by day 5 of differentiation (Fig. 5A). Maximal induction of Pax3-GFP+ cells was observed when Ldn was added in the differentiation media from day 0 (Fig. 5B). Furthermore, these Pax3-GFP+ cells expressed the Pax3 protein (Fig. 5C). While in vivo, Pax3 is expressed both in the paraxial mesoderm and in neural precursors, the Pax3-GFP+ cells induced in the RDL/CDL conditions exhibited a mesenchymal aspect, distinct from the rosette-forming Pax3+ neural precursors found in neural-inducing conditions (Fig. 5D) (Chambers et al., 2009). These Pax3-GFP+ cells were also found to be essentially negative for the neural marker Sox2 but expressed the anterior PSM/ somitic markers Pax3, Uncx, Meox1 and Foxc2 at levels comparable to those detected in vivo (Fig. 5E and data not shown). Thus, differentiation of ESCs in serum-containing RDL or CDL media was found to efficiently recapitulate the major stages of PSM differentiation in vitro.
Validation of the paraxial mesoderm identity of mouse ESCs differentiated in vitro
The presence of PDGFRα (CD140a) and the absence of VEGFR2 (also named Flk1, KDR or CD309) have been used to identify the paraxial mesoderm lineage in differentiated ESCs in vitro (Darabi et al., 2008; Nishikawa et al., 1998; Sakurai et al., 2006, 2009). In vivo, however, these surface markers are not specific to paraxial mesoderm and their expression largely overlaps with other mesodermal populations such as the lateral plate (Ding et al., 2013; Ema et al., 2006; Motoike et al., 2003). FACS analysis of the Msgn1-repV cultures differentiated for 3-4 days in RDL medium with or without a pre-differentiation step in N2B27 medium supplemented with 1% knockout serum (thereafter, NK1) medium (Chal et al., 2015) revealed that more than 90% of the Msgn1-repV+ (M+) cells were recognized by the PDGFRα antibody (Fig. 6A). However, 25% of these M+ cells were also found to express VEGFR2 (Fig. 6B). Moreover, PDGFRα was not specific to the Msgn1-repV+ population as it also marked 75% of the Msgn1-repV− (M−) cells (Fig. 6B), 45% of which also expressed VEGFR2 (Fig. 6A,B). We also analyzed the expression of the surface marker CXCR4 (CD184; Borchin et al., 2013) in relation to M+ cells and found that although most of the M− were CXCR4+, about 30-50% of the M+ cells were also CXCR4+, suggesting that CXCR4 cannot discriminate for PSM fate in vitro (Fig. S2). Furthermore, while CXCR4 was found expressed on hMSGN1-Venus+ progenitors, this was true under both Chir-only and Chir/Ldn conditions, suggesting that CXCR4 cannot discriminate between lateral plate mesoderm- and paraxial mesoderm-fated cells (Fig. S3).
Next, we characterized by RT-qPCR the FACS-sorted Msgn1-Venus+ subpopulations based on their PDGFRα and VEGFR2 expression (as indicated in Fig. 6A). We found that the Msgn1-Venus+ PDGFRα+ (M+Pα+) fraction was strongly enriched for the paraxial mesoderm markers Msgn1, Tbx6 and Pax3 (Fig. 6C). In contrast, the M-Pα+ population showed very low levels of Tbx6, suggesting that it does not contain PSM cells (Fig. 6C). These M-Pα+ cells expressed Sox10 together with Pax3 but lack Sox2 expression, suggesting that this fraction contains neural crest cells (Fig. 6C). Furthermore, while the M+Pα+ VEGFR2− (M+Pα+V−) subpopulation expressed the highest levels of anterior PSM/somitic markers, the triple positive M+Pα+V+ population also expressed a significant level of anterior PSM/somitic markers (Fig. 6D). Interestingly, the M−Pα+V− subpopulation also expressed comparable level of anterior PSM/somitic markers, suggesting that, by day 4, a fraction of cells transited to anterior PSM/ somitic fate and downregulated the Msgn1-repV reporter. Moreover, Pax3-GFP+ cells differentiated for 6 days in RDL medium with or without pre-differentiation in NK1 medium (Chal et al., 2015) were essentially negative for VEGFR2 while the majority were PDGFRα+ (Fig. 6E). Nevertheless, Pax3-GFP+ PDGFRα− accounted for about 15 to 35% of the total Pax3-GFP+ population. Moreover, staining for CXCR4 showed that Pax3-GFP+ were negative for CXCR4 (Fig. S2). Altogether, these observations show that PDGFRα, VEGFR2 and CXCR4 surface expression lacks tissue specificity and does not fully capture paraxial mesoderm identity.
Transcriptomic analysis of Msgn1-Venus+ and Pax3-GFP+ cells
We next compared the transcriptomes of in vitro-differentiated mouse PSM-like cells with their mouse PSM in vivo counterparts. Microarrays were generated for Msgn1-repV+ and Pax3-GFP+ cells isolated by FACS after 3, 4 or 5 days of differentiation in serum-containing RDL and CDL media. Their gene signatures were compared with those of the anterior and posterior PSM transcriptional domains (Chal et al., 2015) and with the neural tube array series (Fig. 7, Tables S4, S5). As the differentiating ESCs transited from a Msgn1+ to a Pax3+ stage, they downregulated a large number of posterior-specific PSM genes, including Dusp4, Rspo3, Evx1 and Fgf8, as observed in vivo (Fig. 7A,C). Interestingly, although many posterior PSM signature genes were also shared with the posterior-most neural tube, it was not the case for the anterior PSM signature genes. In parallel, progressive activation of a large fraction of the anterior PSM-specific genes, including Ripply2, Mesp2, Nkx3-1 and Tbx18, was observed in differentiating PSM-like cells (Fig. 7B,D). Thus, in vitro differentiation of mouse ESCs in the presence of a WNT activator and a BMP inhibitor can recapitulate the early differentiation stages of the paraxial mesoderm in vivo.
In vivo analysis of the myogenic potential of the ES-derived PSM-like cells
Trunk skeletal muscle tissue is generated exclusively from somitic paraxial mesoderm (Sambasivan et al., 2011). Properly specified PSM-like cells should therefore have the unique potential to generate skeletal muscle. The myogenic potential of the mouse PSM-like cells differentiated in vitro was assessed by transplanting them into injured adult muscle in vivo. Msgn1-repV+ and Pax3-GFP+ cells were differentiated for 3-4 days and 5-6 days in RDL medium, respectively, and isolated by FACS. Next, they were permanently labeled with a lentivirus driving ubiquitous expression of cytoplasmic GFP in order to monitor the fate of the cells following their transplantation. Fifty thousand to 100,000 labeled Msgn1-repV+ or Pax3-GFP+ cells were transplanted into cardiotoxin-injured tibialis anterior muscle of adult Rag2−/−: γc−/− mice (Fig. 8A). Freshly isolated Pax7-nGFP+ adult satellite cells, transduced with the GFP lentivirus, were used as a positive control. One month post-transplantation, Msgn1-repV+ and Pax3-GFP+ donor cells reconstituted large GFP+ areas filled with small, poorly organized, striated dystrophin-positive muscle fibers, as well as occasionally other derivatives, such as fibroblasts, chondrogenic nodules or epithelial cells forming large cysts (Fig. 8B,C and data not shown). GFP+ muscle fibers expressed embryonic, slow or perinatal/fast isoforms of myosin heavy chain (MyHC), indicating that they span a large array of myogenic differentiation stages (Fig. 8D). Thus, when transplanted in vivo into adult injured muscles, PSM-like cells derived in vitro from ESCs are able to continue their differentiation towards the myogenic lineage.
Differentiated PSM-like cells can recapitulate skeletal myogenesis in vitro
We next sought to define conditions in which the anterior PSM-like cells differentiated from ESCs could be reproducibly induced to generate skeletal muscle in vitro. In the embryo, FGF and Wnt signaling are downregulated in the anterior PSM prior to activation of the myogenic program (Aulehla and Pourquie, 2009). Following 4-6 days in serum-containing RDL medium, we exposed the differentiating cultures for 2 days in a medium lacking the Wnt activator and containing the FGF inhibitor PD173074, while maintaining the BMP inhibition with LDN193189 (thereafter referred as PDL medium). Cultures were subsequently transferred to a medium containing 2% horse serum, in which they were maintained healthily for more than 2 months, which could not be achieved in serum-free conditions (Table S1). In these conditions, mononucleated Myog-Venus+ myocytes (Chal et al., 2015), with a spindle shape reminiscent of primary myotomal cells, were visible after 1 week in culture (Fig. 9A). Furthermore, pre-differentiating the mouse ESC culture for 2 days in NK1 medium prior to exposure to RDL medium led to a more robust and homogeneous activation of the Myog-repV reporter in culture (Table S1). Substituting R-spondin3 by Chir also led to efficient induction of Myog-repV+ myocytes. These myocytes were not observed when cells were maintained in RDL (or CDL) after day 6 or differentiated in base FBS 15% media (data not shown). The number of myocytes steadily increased over time, progressively covering the entire surface of the wells (Fig. 9B, data not shown). Elongated slow and perinatal/fast MyHC-positive fibers progressively formed from the Myog-Venus+ myocyte cells (Fig. 9C,D). Immature myogenic cells expressing MyoD, with a subset of them also expressing Myf5, but negative for Myog were also found in the cultures, whereas myotubes myonuclei were both MyoD+ and Myog+ (Fig. 9E and data not shown). After 2 weeks of differentiation, a large number of multinucleated fast MyHC+ myofibers were observed in the cultures (Fig. 9F-H). Around 10 to 15,000 such muscle fibers could be obtained within 3 to 5 weeks in wells seeded with 20,000-30,000 ESCs, reaching densities of up to 80 fibers per mm2 in places (Fig. 9F-H, data not shown). These fibers were between 1 and 3 mm in length, 10-20 µm in width and contained up to 50-100 myonuclei each, which are values comparable with those of mouse perinatal fibers and exceeded those obtained in serum-free culture conditions (Fig. 9G,L,M, Table S2) (White et al., 2010). These mature fibers exhibited highly organized striations, as evidenced by anti-perinatal MyHC immunolabeling and a sarcomeric length of around 2.5 µm (Fig. 9H,M). Further evidence of maturation was their expression of dystrophin in sub-sarcolemmal position, in some instances the presence of a single cluster of acetylcholine receptors in equatorial position (Fig. 9I,J) and the presence of continuous basal lamina individually surrounding fibers (Fig. 9K). Unlike in serum-free conditions, most of the differentiated fibers differentiated and maintained in serum-containing conditions contracted spontaneously indicating the functionality of the contractile apparatus (Movies 1 and 2, data not shown). Together, these data suggest that our in vitro culture conditions are able to recapitulate a myogenic differentiation sequence resembling that observed during normal embryogenesis in the mouse.
Next, mature ES-derived myogenic cultures differentiated as described above were analyzed by RT-qPCR for expression of myogenic markers and compared with undifferentiated ESCs, E11.5 trunk (primary) muscles and E17.5 (fetal) back muscles (Fig. 9N). While Myod1 and Myog were clearly detected in all myogenic samples, Pax3 was only detected in the E11.5 primary muscles. In contrast, Pax7 was robustly expressed in both muscle samples and at a lower level in the differentiated ESC myogenic cultures. Furthermore, the fetal muscle marker Nfix (Messina et al., 2010) and Myh2 (Fast2A MyHC) were not detected in E11.5 muscles but were significantly enriched both in the E17.5 muscle sample and in the 3-week-old cultures (Fig. 9N). Thus, myogenic cultures differentiated from mouse ESCs can undergo in vitro a maturation reminiscent of the primary and secondary (fetal) skeletal myogenesis, ultimately resulting in the production of striated contractile muscle fibers exhibiting a phenotype similar to early post-natal fibers.
Production of ES-derived Pax7+ satellite-like precursor cells in vitro
We next examined the presence of Pax7+ progenitors in the long-term cultures of differentiated ESCs described above. By 2-3 weeks of differentiation, large streams of Pax7+ cells were found interspersed with newly formed myocytes, suggesting that these correspond to myogenic progenitors (Fig. 10A,G). Between 3 and 4 weeks, these populations quickly resolved to generate skeletal myocytes and aligned muscle fibers. After 4 weeks of differentiation, the number of Pax7+ cells in culture decreased drastically. Nevertheless, a small fraction was found in close association with large myofibers, a topography strikingly reminiscent of the in vivo situation where adult Pax7+ satellite cells are found in contact and aligned with fully differentiated muscle fibers (Yin et al., 2013) (Fig. 10B,C). A fraction of Pax7+ progenitors becomes quiescent, as suggested by the loss of Ki67 expression (Fig. 10D-F,H). Furthermore, isolated Pax7+ cells were found adjacent to the basal lamina of fibers and occasionally in sublaminar position, as evidenced by confocal microscopy (Fig. 10I-K). Thus, our experiments suggest that differentiation of satellite-like Pax7+ myogenic progenitor cells can also be obtained in vitro using a serum-based differentiation strategy.
DISCUSSION
Here, we describe a novel protocol for highly efficient in vitro myogenic differentiation of mouse ESCs and for long term culture of the resulting myofibers. This serum-based protocol is based on an initial step involving activation of Wnt/β-catenin signaling combined with BMP inhibition followed by FGF and Wnt inhibition, thus recapitulating the signaling sequence to which the PSM is exposed in vivo. It allows for more effective differentiation and maturation of striated myofibers than the chemically defined protocols recently published (Chal et al., 2015; Lee et al., 2016; Sakurai et al., 2012; Shelton et al., 2014). While posterior PSM cells were generated faster in vitro than in vivo (4 days versus E7.5-E8.0), the first myocytes were detected after 7 days, soon followed by the first myofibers (10-12 days), a timing comparable with embryonic development. The bulk of myotube formation seen in vitro after 2 weeks could also match the Pax7+-driven, massive myogenic fetal growth. The myotubes generated in vitro acquired characteristics of perinatal muscle fibers after 3-4 weeks, which is roughly on par with the 3-week gestation of the mouse. However, these fibers do not appear to progress much further than stage, possibly owing to the lack of neural input (Sanes and Lichtman, 1999).
Using fluorescent reporter lines, we demonstrate that activation of the Wnt/β-catenin pathway with Rspo3 or CHIR-99021 can induce up to 70-95% of cells to activate expression of Msgn1, a posterior PSM-specific marker, in mouse and human ES/iPSCs. Such a requirement of Wnt/β-catenin signaling for paraxial mesoderm differentiation has been well established both in vivo in mice with mutations in the Wnt pathway (Dunty et al., 2008; Galceran et al., 2004; Takada et al., 1994; Yamaguchi et al., 1999), and in vitro in differentiating ESC cultures (Borchin et al., 2013; Chal and Pourquié, 2013; Gouti et al., 2014; Mendjan et al., 2014; Shelton et al., 2014). We further demonstrate that while Wnt activation alone is sufficient to activate expression of Msgn1 in mouse and human ES/iPSCs, these cells start to express Bmp4 and progressively drift to a lateral plate fate, expressing markers such as Foxf1 or Hand1 and Hand2. In vivo, Bmp4 has been shown to divert early paraxial mesoderm progenitors toward a lateral plate fate (Tonegawa et al., 1997). When mouse ES or human iPSCs were treated with Rspo3 or CHIR-99021 alone, we noticed an activation of Bmp4 in the Venus-positive cells together with expression of lateral plate markers such as Foxf1. This raised the possibility that early Msgn1-repV cells exhibit a mixed unresolved paraxial mesoderm/lateral plate identity. Alternatively, Msgn1-repV+ cells may constitute an heterogeneous population of paraxial-fated cells and of cells that only transiently express Msgn1 (but retain the stable fluorescent Venus), while subsequently differentiating into lateral plate. Because Bmp4 has been shown to activate its own expression (Adelman et al., 2002; Blitz et al., 2000; Rojas et al., 2005; Schuler-Metz et al., 2000), we treated the cells with BMP inhibitors such as Noggin and a BMP type I receptor inhibitor to block this endogenous BMP production. This led to an active downregulation of Bmp4 and to the dose-dependent induction of a paraxial mesoderm fate. Although Noggin could repress lateral plate mesoderm fate in the short term, the small molecule LDN-193189 was much more effective in the long term in promoting paraxial mesoderm differentiation. These differences may be due to the nature and respective stability of these two types of inhibitors or, alternatively, to their precise mechanism of action on the BMP signaling pathway. BMP modulation has also been shown to be important for intermediate mesoderm and chondrogenic mesoderm induction (Craft et al., 2013, 2015; Morizane et al., 2015; Tanaka et al., 2009; Umeda et al., 2012; Zhao et al., 2014).
Differentiation of mouse and human ES/iPSCs in monolayers combined with modulators of Activin, BMP and Wnt pathways has been reported to induce a paraxial mesoderm fate. However, these methods lead to limited induction of paraxial mesoderm precursors requiring FACS enrichment to generate limited amounts of paraxial mesoderm derivatives such as muscle or cartilage (Sakurai et al., 2009, 2012; Tanaka et al., 2009). Furthermore, these studies largely relied on expression of PDGFRα and VEGFR2 to identify paraxial mesoderm cells. Here, we show that these markers are not only restricted to subpopulations of paraxial mesoderm but also identify many other non-paraxial mesoderm cell populations both in vitro and in vivo, and thus cannot be used alone as the identification criterion for this lineage (this study; Ding et al., 2013; Ema et al., 2006; Motoike et al., 2003). Most studies published so far on the differentiation of paraxial mesoderm cells and their derivatives from ESCs, have relied on this identification method (Chan et al., 2016; Darabi et al., 2008; Filareto et al., 2012; Hwang et al., 2014; Magli et al., 2013; Sakurai et al., 2006, 2008, 2012; Tanaka et al., 2009) and thus the populations described therein as paraxial mesoderm likely correspond to a mixture of fates.
Finally, we generated a microarray series of microdissected embryonic neural tube fragments and compared it with the PSM series (Chal et al., 2015). Clustering analysis revealed that, unlike the PSM, early neural tube differentiation is a progressive process with no abrupt transcriptional fate specification (Oginuma et al., 2017; Olivera-Martinez et al., 2014; Ozbudak et al., 2010). Interestingly, the gene signature of the posterior-most neural tube domain was largely overlapping with the tail bud domain of the PSM series, supporting the idea that the progenitors of both lineages share a common program/origin (Henrique et al., 2015; Tzouanacou et al., 2009).
MATERIALS AND METHODS
Mouse ESC paraxial mesoderm and skeletal muscle serum-based differentiation
Mouse ES lines Msgn1-repV, Pax3-GFP, Myog-repV and Pax7-GFP have been described, and were maintained as previously described (Chal et al., 2015). For PSM-like differentiation, mouse ESCs were plated at ∼10,000-20,000 cells/cm2 on gelatin-coated dishes in DMEM-based medium with 15% FBS supplemented with 10 ng/ml Rspo3 (Peprotech, R&D Biosystems), 0.5% DMSO (Sigma) and 0.1 µM LDN-193189 (Ldn, Tocris, Stemgent) for 2 days (RDL medium). Alternatively, Rspo3 was replaced with the GSK3-β inhibitor CHIRON99021 (Chir; Tocris, Stemgent) at 1-3 µM (CDL medium). After 2 days, medium was changed to a reduced-serum medium, typically 1% FBS, 14% knock-out serum replacement (KSR, Gibco) supplemented as indicated above. Alternatively, mouse ESCs were pre-differentiated in serum-free N2B27 medium supplemented with 1% KSR thereafter (NK1 medium) for 2 days (Table S1). Cells were then changed to the RDL (or CDL) medium described above. PSM-like differentiations were performed at least 30 times independently on four different mouse ESC lines. For skeletal muscle differentiation, PSM-like cultures at day 6 of differentiation were changed to reduced-serum medium supplemented with 0.25 µM of the MEK inhibitor PD173074 (Stemgent) and 0.1 µM Ldn (PdL medium) for 2 days (Table S1). After day 8 of differentiation, media were changed to 2% horse serum (HS2%) and changed every other day. Myogenic differentiations were performed at least 20 times independently on four different mouse ESC lines. Further details on the maintenance, and PSM-like fate, skeletal muscle and neural differentiation can be found in the supplementary Materials and Methods.
Human iPSC paraxial mesoderm serum-free differentiation
Human PS cell lines H9 (WA09) and hiPS11a (HSCI) cells were cultured on a Matrigel (Corning)/mTESR1 system (StemCell Technologies) and were differentiated according to Chal et al. (2016). Briefly, hPS were plated at approximatively at 15,000-30,000 cells/cm2 and, after recovery, medium was changed to a DMEM-based medium supplemented with insulin-transferrin-selenium (ITS, Gibco), 3 µM CHIRON99021 (Axon MedChem, Tocris) and 0.5 µM LDN-193189 (Axon MedChem, Stemgent) (CL medium). On day 3, 20 ng/ml FGF2 (R&D Systems) was added for an additional 3 days. Differentiation experiments were performed at least 15 times independently on two unrelated human iPS lines. Further details on the maintenance of human iPSCs and their differentiation can be found in the supplementary Materials and Methods.
Human hMSGN1-Venus hiPS reporter line generation
The MSGN1 locus was targeted in the hiPS11a line by the CRISPR-Cas9 method (Cong et al., 2013; Ran et al., 2013) to produce a cleavable fusion protein hMSGN1-2A-nuclearVenus. Clones were verified for correct integration and three clones were further validated for correct expression and differentiation. Further details can be found in the supplementary Materials and Methods.
Flow cytometry, immunophenotyping, RT-qPCR and sorting
Dissociated cells were incubated for 20 min at 4°C with mouse anti-PDGFRα (CD140a, clone APA5)-APC and mouse anti-VEGFR2 (Flk-1/KDR/CD309, clone Avas12a)-PE (eBioscience) antibodies at 1 µg and 0.5 µg per million of cells, respectively. Next, cells were washed with PBS, resuspended in PBS/2% FBS and analyzed or sorted for RT-qPCR analysis. Flow cytometry analyses were performed at least three times independently. Further details can be found in the supplementary Materials and Methods.
Microarray generation and analysis
Microarrays were generated and analyzed essentially as described previously (Chal et al., 2015). Mouse neural tube series and mouse ESCs samples were hybridized on GeneChip Mouse Genome 430 2.0 arrays. hiPS samples were hybridized on GeneChip Human Genome U133 2.0 arrays (Affymetrix). Further details can be found in the supplementary Materials and Methods.
PSM-like cells preparation for transplantation into injured tibialis anterior (TA) muscle
Msgn1-repV or Pax3-GFP mouse ESCs were differentiated for 4-6 days and pretreated with 10 µM ROCK inhibitor (Y-27632, Tocris) 1 day before being trypsinized. The positive fraction was sorted by FACS and permanently labeled with a CAG-GFP lentivirus prior injection at 50,000 to 100,000 cells in the cardiotoxin-preinjured tibialis anterior muscles of a cohort of Rag2−/−: γc−/− mice essentially as previously described (Gayraud-Morel et al., 2012). Engraftment was analyzed 1-2 months post-transplantation. Transplantation experiments were carried out at least three times independently on cohorts of three to five animals. Further details can be found in the supplementary Materials and Methods.
Gene signature lists (GSL) method
Microarrays of wild-type mouse tissues deposited in GEO for the Affymetrix Mouse Genome 430 2.0 were used. Arrays were collected, normalized and the median expression value for each probeset was defined as the reference value for that particular probeset or gene. Signature gene for a given conditions were any probeset whose normalized expression value was 10 times higher than the corresponding reference value. Further details can be found in the supplementary Materials and Methods.
In situ hybridization
Whole mount in situ hybridization was carried out as described previously (Henrique et al., 1995). Further details can be found in the supplementary Materials and Methods.
Immunohistochemistry
Cell cultures were fixed in 4% formaldehyde followed by incubation with primary antibodies overnight at 4°C. The next day, cultures were washed three times and incubated with secondary antibodies (1:500) for at least 6 hours. For dissected TA muscles, muscles were cryosectioned and incubated overnight with primary antibodies, followed by secondary antibodies conjugated with AlexaFluor (Molecular Probes) at 1:500. Further details can be found in the supplementary Materials and Methods.
Image acquisition and processing
Samples were acquired either on a Zeiss Axiovert or Evos FL. Images were processed using Adobe Photoshop and Fiji. Movies of muscle contractions were taken with a Leica DMRB microscope using a photometric FX camera. Further details can be found in the supplementary Materials.
Quantifications and statistical analysis
Image quantifications were carried out using Fiji (Schindelin et al., 2012) as described previously (Chal et al., 2015). Hierarchical clustering of array data and differential expression analysis were carried out as described (Chal et al., 2015).
Acknowledgements
We thank Christopher Henderson for critical reading of the manuscript. We are grateful to Jennifer Pace and Tania Knauer-Meyer for their help, and to Laurent Bianchetti for Bioinformatic support. We thank the cytometry (Claudine Ebel), microarray (Christelle Thibault-Carpentier) and cell culture facilities at the IGBMC, and the Cytometry and Animal Facilities at the Pasteur Institute for assistance.
Footnotes
Author contributions
Conceptualization: J.C., O.P.; Methodology: J.C., Z.A.T., M.O., P.M., B. Gobert, A.M., A.H., J.-M.G., M.K., B. Gayraud-Morel, S.T., O.P.; Software: P.M., O.T.; Validation: J.C., Z.A.T.; Formal analysis: J.C., P.M., O.P.; Investigation: J.C., Z.A.T., M.O., B. Gobert, A.M., G.G., A.B., O.S., L.K., B. Gayraud-Morel; Resources: O.T., A.H., J.-M.G., B. Gayraud-Morel, S.T., O.P.; Data curation: J.C., P.M., O.T.; Writing - original draft: J.C., O.P.; Visualization: J.C., Z.A.T., P.M., O.T.; Supervision: J.C., O.P.; Project administration: J.C., M.K., O.P.; Funding acquisition: O.P.
Funding
This work was supported by an advanced grant from the European Research Council (ERC-2009-AdG 249931 to O.P.), by a Seventh Framework Programme grant (Plurimes 602423) and by a strategic project from the Association Française contre les Myopathies (AFM-Téléthon) to O.P.
Data availability
Microarray data have been deposited in GEO database under accession number GSE39615.
References
Competing interests
The work described in this article is partially covered by patent application PCT/EP2012/066793 (publication number WO2013030243 A1). O.P., J.C. and M.K. are co-founders and shareholders of Anagenesis Biotechnologies, a start-up company specializing in the production of muscle cells in vitro for cell therapy and drug screening.