Pluripotent embryonic stem cells (ESCs) generate rostral paraxial mesoderm-like progeny in 5-6 days of differentiation induced by Wnt3a and Noggin (Nog). We report that canonical Wnt signaling introduced either by forced expression of activated β-catenin, or the small-molecule inhibitor of Gsk3, CHIR99021, satisfied the need for Wnt3a signaling, and that the small-molecule inhibitor of BMP type I receptors, LDN193189, was able to replace Nog. Mesodermal progeny generated using such small molecules were chondrogenic in vitro, and expressed trunk paraxial mesoderm markers such as Tcf15 and Meox1, and somite markers such as Uncx, but failed to express sclerotome markers such as Pax1. Induction of the osteochondrogenically committed sclerotome from somite requires sonic hedgehog and Nog. Consistently, Pax1 and Bapx1 expression was induced when the isolated paraxial mesodermal progeny were treated with SAG1 (a hedgehog receptor agonist) and LDN193189, then Sox9 expression was induced, leading to cartilaginous nodules and particles in the presence of BMP, indicative of chondrogenesis via sclerotome specification. By contrast, treatment with TGFβ also supported chondrogenesis and stimulated Sox9 expression, but failed to induce the expression of Pax1 and Bapx1. On ectopic transplantation to immunocompromised mice, the cartilage particles developed under either condition became similarly mineralized and formed pieces of bone with marrow. Thus, the use of small molecules led to the effective generation from ESCs of paraxial mesodermal progeny, and to their further differentiation in vitro through sclerotome specification into growth plate-like chondrocytes, a mechanism resembling in vivo somitic chondrogenesis that is not recapitulated with TGFβ.
Repair of large bone defects and damaged cartilage remain a significant clinical challenge. The combination of therapy with mesenchymal stromal cells (MSCs) and the use of biodegradable biomimetic scaffolds appears to be a promising therapeutic approach, as MSCs from bone marrow or fat tissue can be both osteogenic and chondrogenic under certain conditions in vitro and in vivo. However, the difficulty of the approach lies in controlling the clinical outcome. As bone and cartilage are naturally formed during embryogenesis, we hypothesize that one way to alleviate the difficulty is to use embryonic osteochondro-progenitors instead of adult cells.
For human embryonic cells, pluripotent stem cells (PSCs) [i.e. embryonic stem cells (ESCs) or induced pluripotent stem cells (iPSCs)] are thus far the only practical source. PSC culture is scalable, and in vitro differentiation of PSCs can mimic early embryogenesis (Murry and Keller, 2008; Nishikawa et al., 2007) such that large numbers of a particular type of embryonic cell can be obtained under the appropriate conditions of differentiation to mimic embryogenesis. Furthermore, the wealth of information emanating from developmental biology studies is expected to aid in the discovery and refinement of such conditions. However, although ESCs have been shown to generate mesenchymal progeny capable of developing cartilage in vitro (Nakayama and Umeda, 2011), the resultant chondrogenic activities were generally poor, and with the exception of ours and a few others (Craft et al., 2013; Oldershaw et al., 2010; Tanaka et al., 2009; Umeda et al., 2012), most reports did not define the developmental origin of the mesenchymal cells generated. The osteochondro-progenitors that develop during embryogenesis are limb bud mesenchyme (derived from lateral plate mesoderm) responsible for limb bone and cartilage generation, sclerotome (from somite/rostral paraxial mesoderm) responsible for rib, vertebral joint, intervertebral disc and vertebral body formation, and ectomesenchyme (from cranial neural crest) responsible for craniofacial bone and cartilage generation. These progenitors are differentially specified through the action of protein factors such as Wnt, and members of the transforming growth factor (TGF)β-superfamily: Nodal and bone morphogenetic protein (BMP), and their inhibitors. Therefore, to make use of such knowledge to improve and refine the condition for osteochondro-progenitor development, PSC differentiation should be directed toward a particular lineage.
In the first step toward testing our hypothesis, we have previously demonstrated that somitic/rostral presomitic mesoderm-like progeny (hereafter referred to as ‘rostral paraxial mesoderm’) is specified from mouse and human ESCs and is enriched in the Flk1(Kdr)−Pdgfrα+ cell fraction of embryoid bodies (EBs) generated in a chemically defined medium (CDM) (Tanaka et al., 2009; Umeda et al., 2012). The specification process is strictly dependent on the activation of Wnt signaling and suppression of BMP signaling (Craft et al., 2013; Tanaka et al., 2009; Umeda et al., 2012).
Considering future application to humans, it is important to determine whether the PSC progeny developed in vitro behave in the same way as the in vivo progeny and generate mature progeny. Axial skeletogenesis by somite goes through the osteochondrogenically committed progeny, sclerotome (Chal and Pourquié, 2009; Dockter, 2000), in a process involving platelet-derived growth factor (PDGF) (Hoch and Soriano, 2003; Tallquist et al., 2000), sonic hedgehog (Shh) (Fan et al., 1995; Fan and Tessier-Lavigne, 1994; Johnson et al., 1994; Murtaugh et al., 1999; Zeng et al., 2002), and the BMP inhibitor Noggin (Nog) (Hirsinger et al., 1997; Marcelle et al., 1997; McMahon et al., 1998). The Flk1−Pdgfrα+ rostral paraxial mesoderm from ESCs consistently show chondrogenic activity in vitro (Tanaka et al., 2009). However, it was not clear whether the paraxial mesoderm went through chondrogenesis via sclerotome specification, as occurs in vivo.
In addition, considering the necessity for culture scale-up for clinical application, there would be a significant cost benefit if the protein factors could be replaced with small-molecule signaling modifiers. Many chemical compounds have been developed that target relevant signaling mechanisms: e.g. the inhibitors of glycogen synthase kinase (Gsk)3, BIO (Sato et al., 2004) and CHIR99021 (Cohen and Goedert, 2004) for mimicking canonical Wnt signaling; SAG1 for mimicking hedgehog signaling (Chen et al., 2002), the dorsomorphin analog LDN193189 for inhibiting Alk1/2 and to a lesser degree Alk3 [type I BMP receptor (BMPR)] signaling (Yu et al., 2008), and SB431542 for inhibition of Alk4/5/7 (type I Nodal/activin/TGFβ receptor) signaling (Inman et al., 2002; Laping et al., 2002).
Here, we describe the establishment of a novel, small-molecule based differentiation method for the specification of chondrogenic rostral paraxial mesoderm from ESCs. We then demonstrate that the paraxial mesodermal progeny respond properly to a set of signals (hedgehog+Nog) known to be essential in vivo for generating sclerotome, and also specify sclerotome-like intermediates before forming chondrocytes in vitro. The same outcome was not recapitulated by TGFβ, which is used widely for the induction of chondrogenesis from various mesenchymal cell types. We have also demonstrated that the cartilage particles developed are mineralized and vascularized, leading to bone and marrow formation in a mouse model.
β-catenin activation without BMP signaling specifies chondrogenic rostral paraxial mesoderm from ESCs
During differentiation of the mouse brachyury (Bry)-GFP ESC in CDM, the presence of Wnt3a and Nog effectively induces rostral paraxial mesoderm but not lateral plate mesoderm in the Bry-GFP+Flk1−Pdgfrα+ cell fraction (Tanaka et al., 2009). Similarly, MIXL1-GFP+KDR−Pdgfrα+ paraxial mesoderm can be specified from Mixl1-GFP human ESCs in CDM in the presence of BIO and Nog (Umeda et al., 2012).
The canonical Wnt signaling mechanism involves suppression of Gsk3 activity and subsequent stabilization/nuclear translocation (hereafter described as ‘activation’) of β-catenin. Therefore, to confirm that canonical Wnt signaling is sufficient to induce paraxial mesoderm in the presence of Nog, we constructed the mouse ESC line, EBRβcatM, expressing the cDNA encoding a 4-hydroxy tamoxifen (OHT)-activatable β-catenin, i.e. Ctnnb1(S33Y)-Mer, in a tetracycline (Tet)-sensitive manner (supplementary material Fig. S1). The protein expression was monitored by the expression of yellow fluorescent protein (YFP). When the expression of Ctnnb(S33Y)-Mer protein was induced by lower concentrations of Tet from one day before differentiation and activation by OHT on day 2 of differentiation (Fig. 1A, supplementary material Fig. S2A), the expression of rostral paraxial mesoderm genes (e.g. Meox1 and Uncx), but not lateral plate/extra-embryonic mesoderm genes (e.g. Foxf1a and Hand2) was induced by day 6 of differentiation (T/OHT, Fig. 1B, supplementary material Fig. S2B,C). The addition of Nog enhanced the level of paraxial mesoderm gene expression (T+N/OHT+N).
Under OHT+Nog, a significant level of YFP+Flk1−Pdgfrα+ progeny was generated (Fig. 1C). Similar to progeny generated under Wnt3a+Nog, the YFP+Flk1−Pdgfrα+ but not the YFP+Flk1−Pdgfrα− cell fraction isolated by fluorescence-activated cell sorting (FACS) contained cells expressing paraxial mesodermal genes such as Meox1 (Fig. 1C, supplementary material Fig. S2D left). Furthermore, cartilage nodule-forming activity was detected in the YFP+Flk1−Pdgfrα+ cell fraction generated under OHT+Nog during 2D micromass culture in the presence of PDGF+TGFβ (Fig. 1D, supplementary material Fig. S2E), as was seen in the Flk1−Pdgfrα+ paraxial mesoderm generated under Wnt3a+Nog (Tanaka et al., 2009). By contrast, the YFP+Flk1−Pdgfrα+ progeny generated with OHT alone or with OHT+BMP4 resulted in significantly fewer cartilage nodules (Fig. 1D), similar to what was observed for the Pdgfrα+ lateral plate/extra-embryonic mesoderm fraction generated with either Wnt3a or BMP4 (Tanaka et al., 2009). Thus, the activated β-catenin is a substitute for Wnt3a signaling in the generation of chondrogenic paraxial mesoderm from ESCs.
Specification of chondrogenic paraxial mesoderm from ESCs by small molecules
Gsk3 occupies a crucial node in the mediation of various cellular processes (Hur and Zhou, 2010). However, the effects of Gsk3 suppression are considered to represent those of canonical, but not noncanonical, Wnt signaling that leads to β-catenin activation. Therefore, to complement the study involving overexpression of activated β-catenin (Fig. 1), the parental ESCs (EBRTcH3) were treated with various concentrations of three Gsk3 inhibitors: BIO, AcBIO or CHIR99021 (CHIR) (Fig. 2B). BIO stimulated Meox1 expression by day 6 in a dose-dependent manner, but its inactive analog (MeBIO) did not (supplementary material Fig. S3E). Inclusion of Nog significantly enhanced the expression of Meox1 in all cases (Fig. 2B left). Of the three inhibitors, CHIR had the strongest effect on the induction of Meox1 transcript (peaking at 5 μM), but at >5 μM induced significant levels of Meox1 even in the absence of Nog (CN, C). Thus, the overexpressed, activated β-catenin could be effectively replaced with either Gsk3 inhibitor, BIO or CHIR.
We next attempted to replace the pan-BMP inhibitor Nog, which prevents BMPs from binding to BMPRs, with the Alk1/2/(3)-inhibitor, LDN193189 (LDN). At about 0.6 μM, LDN augmented the CHIR-induced expression of Meox1 during ESC differentiation (Fig. 2B right). The enhancement of Meox1 expression with Nog or LDN was more pronounced with 5 μM CHIR than with BIO or 2.5 μM CHIR. Furthermore, similar to Wnt3a+Nog (Tanaka et al., 2009), CHIR+LDN also induced the formation of Flk1−Pdgfrα+ progeny from EBRTcH3 and Bry-GFP ESCs. The Flk1−Pdgfrα+ cells were Bry-GFP+ (Tanaka et al., 2009), suggesting mesendodermal origin, and the majority of them were E-cadherin− (Fig. 2C), i.e. mesoderm. Consistently, the E-cadherin−Bry+Flk1−Pdgfrα+ progeny, but not the E-cadherin−Bry+Flk1−Pdgfrα− progeny isolated by FACS were enriched in the rostral paraxial mesoderm transcripts Meox1, Tcf15 and Uncx (see Fig. 3C, supplementary material Fig. S5C), and were also found to be chondrogenic in the subsequent micromass culture (Fig. 2C). Thus, LDN could effectively replace Nog, although LDN preferentially inhibited Alk1/2 instead of the classical type I BMPR Alk3/6 (Yu et al., 2008).
Nodal signaling is involved in gastrulation and germ layer specification in a dose-dependent manner (Schier, 2009). By day 5 of ESC differentiation, Wnt3a, and to a lesser degree BIO, induced expression of the Nodal gene via a pathway independent of BMP signaling, as shown by the lack of effect of Nog (supplementary material Fig. S3B,C right). This finding revealed the potential to optimize differentiation by manipulating Nodal signaling using activin A (activin) and the Nodal/activin/TGFβ receptor inhibitor SB431542 (SB). Attempts to optimize the differentiation method using activin and SB resulted in a parabolic response (Fig. 2D) similar to that seen with EBRβcatM cells (supplementary material Fig. S2D right). The optimal concentrations fell within a small range from 0-5 ng/ml activin to 0-1 μM SB, depending on the type of Wnt activator used. Furthermore, inefficient specification of Meox1-expressing paraxial mesoderm in the presence of activin at higher concentrations (5-20 ng/ml) correlated with the reduction in the Flk1−Pdgfrα+ cell population in EBs formed, and with the weak to nonexistent chondrogenic activity in the Bry+Flk1−Pdgfrα+ progeny (Fig. 2E). Interestingly, in specification of paraxial mesoderm the use of Gsk3 inhibitors gave consistently higher levels of Meox1 transcript than the use of Wnt3a in the presence of Nog (Fig. 2D). However, under the optimal small-molecule condition [5 μM CHIR +0.6 μM LDN (CHIR+LDN)], Meox1 was induced at levels statistically equivalent to those achieved under the protein-based, Wnt3a+Nog condition (supplementary material Fig. S5B).
These results indicate that small-molecule inhibitors of Gsk3 and BMPR successfully replace Wnt3a and Nog for efficient specification of chondrogenic, rostral paraxial mesoderm from ESCs. Manipulation of Nodal/activin/TGFβ signaling is not always necessary, but must be a special consideration when differentiation efficacy is of a critical matter (supplementary material Fig. S4).
Specification of sclerotome from ESCs via a rostral paraxial mesoderm intermediate
Early expression of Pax1 among the mesodermal progeny is largely restricted to the ventral part of sclerotome (Deutsch et al., 1988; Koseki et al., 1993), which is induced by the diffusible factors Shh and Nog, produced from the floor plate and the notochord. The Flk1−Pdgfrα+ rostral paraxial mesoderm induced from ESCs in KSR-based serum-free medium expressed Pax1 mRNA (Tanaka et al., 2009), but those induced in CDM under either Wnt3a+Nog (data not shown) or CHIR+LDN (CL, Fig. 3B,C, supplementary material Fig. S5C) seemed to lack Pax1 mRNA despite Meox1, Tcf15 and Uncx transcripts being highly enriched. We therefore examined the effect of hedgehog on the expression of Pax1 during differentiation of ESCs in CDM. The small-molecule SMO (hedgehog receptor subunit)-agonist SAG1 (SAG) was used to activate hedgehog signaling. The addition of SAG to the (CHIR+LDN)-based differentiation culture on day 4 significantly upregulated Pax1 expression by day 6 (CLS4, Fig. 3B), but had only marginal effects on Meox1 expression and no effect on the progenitor profile distinguished by Flk1 and Pdgfrα expression (data not shown). Consistently, the FACS-purified E-cadherin−Flk1−Pdgfrα+ cell fraction thus generated was enriched in Pax1 mRNA (i.e. sclerotome) and Meox1, Tcf15 and Uncx transcripts (i.e. rostral paraxial mesoderm) (Fig. 3C, supplementary material Fig. S5C). The induction of Pax1 occurred more slowly or to a lesser extent when SAG was added on day 2 instead of day 4 (CLS2, Fig. 3B).
When both Bry+Flk1−Pdgfrα+ and Bry+Flk1−Pdgfrα− progeny were isolated by FACS from EBs developed from Bry-GFP ESCs under CHIR+LDN, and subjected to micromass culture in the presence of Shh+Nog with PDGF, only the Bry+Flk1−Pdgfrα+ cells were capable of inducing the sclerotomal genes Bapx1 (Tribioli et al., 1997) and Pax1 as well as the more committed chondroprogenitor gene Sox9 within 6 days (Fig. 3D). Therefore, the Pax1-expressing sclerotomal cells were probably generated from Pax1-negative Flk1−Pdgfrα+ rostral paraxial mesoderm.
Transient treatment with small-molecule SMO agonist and BMPR inhibitor allows exogenous TGFβ-independent chondrogenesis from the isolated rostral paraxial mesoderm
To demonstrate that the Pax1-positive Flk1−Pdgfrα+ sclerotomal cells are qualitatively more chondrogenic than the immediate precursor cells: i.e. Pax1-negative Flk1−Pdgfrα+ rostral paraxial mesoderm, we first isolated them individually from EBs generated in CDM under CHIR+LDN+SAG (day 4) and CHIR+LDN conditions, respectively, and performed micromass culture in the presence of PDGF, TGFβ and BMP4. However, no significant difference in the efficiency of chondrogenesis was observed (data not shown). The exposure to SAG for 2 days (from day 4 to day 6) during ESC differentiation might be too short to promote commitment of sufficient rostral paraxial mesodermal cells to sclerotome.
Therefore, we isolated the Pax1-negative Flk1−Pdgfrα+ rostral paraxial mesoderm, and addressed whether treatment with Shh+Nog or with their small-molecule analogues SAG+LDN for a longer time would affect the efficiency of chondrogenesis. The 12-day micromass cultures were performed under the following four different conditions, all of which included PDGF (P) as a growth/viability-supporting factor: (1) PHN→PB: Shh+Nog for 6 days (as in Fig. 3D) then BMP4 for 6 days, (2) PSL→PB: SAG+LDN for 6 days then BMP4 for 6 days, (3) PTHN→PTB: condition 1 plus TGFβ, and (4) PTSL→PTB: condition 2 plus TGFβ, with appropriate controls (Fig. 4B). The chondrogenesis cultures with TGFβ (PT→PT), but not those without TGFβ (P→P), generated micromass stained with Alcian Blue. Later (i.e. day 6) addition of BMP4 to the PT culture (PT→PTB) enhanced chondrogenesis as shown by darker blue staining. No such enhancement was observed without TGFβ (P→PB). Interestingly, even in the absence of TGFβ, conditions 1 (PHN→PB) and 2 (PSL→PB) gave rise to a large micromass that stained strongly with Alcian Blue. The micromass accumulated similar amounts of sulfated glycosaminoglycan (sGAG), as judged by the guanidine hydrochloride-extractable Alcian Blue per mass, as that developed under PT→PTB (Fig. 4D). Furthermore, the micromasses developed under PT→PTB and PSN→PB (SAG+Nog followed by BMP4, similar to condition 2) conditions for 18 days contained a similar proportion of active chondrocytes, as indicated by sGAG normalized to DNA, but significantly more than those developed under P→PB conditions (Fig. 4E). The yield of DNA from a micromass reflects cell number, which did not differ significantly at this stage (supplementary material Fig. S6G).
The PSL→PB as well as PSN→PB conditions favor exogenous TGFβ-independent chondrogenesis from the Flk1−Pdgfrα+ rostral paraxial mesoderm, but not from the Flk1−Pdgfrα− progeny (Fig. 4C, supplementary material Fig. S6D). The later treatment with BMP4 proved important. Without it, the efficiency of overall chondrogenesis was reduced significantly (PSL→P, supplementary material Fig. S6E,F). Addition of both SAG and LDN/Nog during the first 6 days of micromass culture was also essential, as treatment with either SAG (PS→PB) or LDN (PL→PB) resulted in zero or only very weak chondrogenesis (Fig. 4C,D, supplementary material Fig. S6C,D,F), with a small change in the growth of micromass (data not shown).
(SAG+Nog/LDN)-induced, but not TGFβ-induced, chondrogenesis from rostral paraxial mesoderm proceeded via sclerotome specification
Real-time RT-PCR analyses demonstrated that SAG+LDN (PSL) stimulation during the first 6 days of micromass culture was crucial for inducing Pax1 and Bapx1 expression from the isolated E-cadherin−Flk1−Pdgfrα+ rostral paraxial mesoderm (Fig. 5B), as was the Shh+Nog stimulation (Fig. 3D). Both transcripts were highly induced at around day 3 by either treatment, but not by treatment with TGFβ (PT, Fig. 5B), LDN (PL) or SAG (PS) alone. The PSL treatment should continue for longer than 3 days before the change to PDGF+BMP4 (PB), and consistent with the notion that BMP signaling counteracts Shh+Nog and inhibits sclerotome induction (Dockter, 2000), an early (day 3) transition led to a somewhat weaker degree of chondrogenesis (supplementary material Fig. S6B). By contrast, expression of Sox9 was strongly induced under PT and somewhat weakly induced under PSL within the first 6 days, but was not induced under PS conditions. In support of differentiation toward chondrocytes, the somite transcript Uncx also disappeared.
These differences in the sclerotomal gene expression during early micromass culture might simply reflect different kinetics of induced chondrogenesis, such that the sclerotomal genes Pax1 and Bapx1 might be expressed at a later time under PT conditions. Therefore, the entire chondrogenesis process was monitored. As LDN shows BMPR subtype specificity (Yu et al., 2008), the SAG+Nog (PSN) condition was included as a control. As shown in Fig. 5C and supplementary material Fig. S7B, under PSL/PSN→PB conditions, Pax1 and Bapx1 expression was induced and peaked within the first 6 days (i.e. before BMP4 stimulation), consistent with the previous results (Fig. 3D and Fig. 5B). Pax1 and Bapx1 expression then significantly decreased or disappeared by day 7 and day 11, respectively, and did not reappear by day 19. Both the type II collagen (Col2a1) and aggrecan (Acan) transcripts, the signs of chondrocyte formation, were detectable from days 11-14 under PSN→PB and from approximately day 6 under PSL→PB. By contrast, under PT→PTB conditions, Pax1 and Bapx1 expression was scarcely detectable until days 19-20, but Col2a1 and Acan transcripts were detectable by day 6, and the levels peaked at around day 11. Sox9 expression was detected under both PSL/PSN→PB and PT→PTB conditions, as with the previous observations (Fig. 3D and Fig. 5B). However, whereas the Sox9 level peaked by day 6 under PT→PTB, Sox9 was expressed over a longer period under PSL/PSN→PB, and especially under PSN→PB for which the peak was delayed to days 7-11 (Fig. 5C, supplementary material Fig. S7B). Concomitantly, Col2a1 and Acan transcripts were induced significantly earlier under PT→PTB than under PSN→PB, but kinetic differences between PT→PTB and PSL→PB conditions were not clear (Fig. 5C, supplementary material Fig. S7C). Thus, the transition from early expression of Pax1 and Bapx1 to later expression of Sox9, Col2a1 and Acan was apparent under the PSL/PSN→PB conditions, while induction of Sox9, Col2a1 and Acan without expression of Pax1 and Bapx1 was observed under PT→PTB condition.
Furthermore, quantitation of sGAG (normalized to DNA) indicated that the micromass developed under both PSL/PSN→PB and PT→PTB conditions accumulated proteoglycan-producing active chondrocytes (Fig. 5D). Consistently under PSN→PB conditions (blue) such active chondrocytes accumulated more slowly than those under PT→PTB (brown) and PSL→PB (green) conditions.
Thus, the ESC-derived rostral paraxial mesoderm can give rise to chondrocytes through two different developmental programs: one initiated by SAG and LDN/Nog (PSL/PSN), which involves Pax1 and Bapx1 induction, i.e. sclerotome specification, and the second by TGFβ (PT), which does not involve sclerotome specification. Interestingly, the PSL→PB condition tended to induce Pax1 and Bapx1 more efficiently and show faster chondrogenesis than the PSN→PB condition, implying that inhibition of Alk1/2-dependent BMP signaling may be the key for sclerotome specification in vitro.
Capacity of the macroscopic cartilage particles developed from rostral paraxial mesoderm via different chondrogenesis mechanisms to mature in vitro and in vivo
Different chondrogenesis programs might lead to different cartilage types. As BMP4 is a robust chondrogenic inducer for many mesenchymal cell types but is also known to facilitate terminal maturation of the developed chondrocytes, we addressed whether chondrocytes developed from the rostral paraxial mesoderm during micromass culture under the conditions in which Pax1 and Bapx1were induced (PSL/PSN→PB) and not induced (PT→PTB) would differentially express the Col10a1 and Alpl transcripts, a sign of hypertrophic differentiation (Fig. 6B, supplementary material Fig. S7B,C). Chondrocytes generated under all conditions expressed Col10a1 and Alpl slightly later than Col2a1 and Acan (Fig. 5C). Similarly, the kinetics of expression under PSL→PB and PT→PTB were faster than under PSN→PB. However, the ratio of the normalized levels of Col2a1 and Col10a1 transcripts accumulated under PT→PTB and PSN→PB conditions did not differ significantly (Fig. 6B upper), suggesting that the rate of maturation of the developed chondrocytes is indistinguishable under both conditions.
In addition, using the 3D pellet culture method, we made cartilage particles under PSN→PB and PT→PTB conditions (Fig. 6C,D). The resulting particles were indistinguishable in size and degree of chondrogenesis, as judged by histological examination. Multiple sections of individual particles demonstrated that all consisted of cartilaginous area that stained uniformly with Toluidine Blue. When such particles were transplanted subcutaneously into immunocompromised mice, both developed in 12 weeks into vascularized and mineralized tissues that contained bone-like structure with hematopoietic marrow. Nog, not LDN, is one of the authentic sclerotome inducers in vivo, and PSN→PB, not PSL→PB, resulted in a significant difference in the kinetics of chondrogenesis compared with PT→PTB (Fig. 5C,D and Fig. 6B). Interestingly, however, the cartilage particles made under PSL→PB conditions yielded essentially the same mineralized cartilage (supplementary material Fig. S8D,E). These observations suggest that the chondrocytes produced from the rostral paraxial mesoderm under both PSL/PSN→PB and PT→PTB conditions resemble growth plate chondrocytes in being able to go through endochondral ossification.
We have demonstrated that small-molecule signaling-modifiers that inhibit Gsk3 and type I BMPR and promote hedgehog signaling effectively specify rostral paraxial mesoderm and then sclerotomal progeny from mouse ESCs as the corresponding protein factors. Therefore, it is now feasible to scale up the culture to obtain a large number of these progeny. We have also demonstrated that chondrocytes can be generated from isolated rostral paraxial mesoderm through two developmental mechanisms (Fig. 7) – the authentic (Shh+Nog)-initiated somitic chondrogenesis mechanism through sclerotome specification, and the TGFβ-induced chondrogenesis mechanism with no sign of sclerotome specification.
In sclerotome, Pax1 and Pax9 interact with Meox1 to induce Bapx1 (Nkx3.2) expression (Rodrigo et al., 2004,, 2003). This leads to activation of the Bapx1-Sox9 positive-feedback mechanism that allows sclerotome to respond to BMP signaling and differentiate into chondrocytes (Kempf et al., 2007; Tribioli and Lufkin, 1999; Yamashita et al., 2009; Zeng et al., 2002). Thus, Bapx1 is crucial for chondrocyte formation for developing vertebral bodies and intervertebral discs (Akazawa et al., 2000; Herbrand et al., 2002; Lettice et al., 1999). The protein (i.e. Shh+Nog)-induced or small-molecule (i.e. SAG+LDN)-induced Pax1 and Bapx1 expression followed by Sox9 expression and BMP4-enhancement of chondrogenesis, therefore, suggests that mouse ESC-derived rostral paraxial mesoderm possesses the capacity of going through the somitic chondrogenesis program as embryonic somitic mesoderm (Dockter, 2000). By contrast, the TGFβ-induction of Sox9 and BMP4-enhancement of chondrogenesis without Pax1 and Bapx1 expression suggests that paraxial mesodermal chondrogenesis can also proceed directly without the Bapx1-Sox9 positive-feedback mechanism.
The conditions in which TGFβ and BMP are involved have been widely used for inducing chondrogenesis in vitro from a variety of mesenchymal cells. In chick embryos, induction of Sox9 expression and chondrogenesis by Shh or by forced Bapx1 expression is observed specifically in paraxial mesoderm (Kempf et al., 2007; Zeng et al., 2002). The same treatments of lateral plate mesoderm fail to induce chondrogenesis. However, forced Sox9 expression allows chondrogenesis from both mesodermal types. These findings suggest that exogenous TGFβ may directly induce Sox9 in many mesenchymal cell types, leading to rapid chondrogenesis in vitro.
The two different mechanisms of chondrogenesis may be mediated by two types of chondrogenic cell present in the rostral paraxial mesoderm fraction, or by one cell type (i.e. sclerotome) with dual capacity. Evidence suggests that the latter is more likely. More than 97% of the cells in the E-cadherin−Flk1−Pdgfrα+ rostral paraxial mesoderm fraction were Bry-GFP+, in which no sign of lateral plate mesoderm and neural crest specification was detected, and both conditions commonly give rise to chondrocytes with the tendency to ossify (Fig. 6).
The question arises as to the importance of the slower and more complex ‘somitic chondrogenesis’ program over the faster and simpler ‘TGFβ-induced chondrogenesis program’. Bone and cartilage formation during embryogenesis is under a tight spatiotemporal control to ensure development of the correct skeletal structure at the correct place. One possible advantage is to make the somitic chondrogenesis process more controllable by extracellular environment. Furthermore, as Bapx1 suppresses Runx2, which is associated with hypertrophic differentiation and subsequent mineralization of chondrocytes (Provot et al., 2006), the Bapx1-Sox9 positive-feedback loop might serve as a mechanism for maintenance of chondrocytes in the nonmineralizing state, as observed in the stable intervertebral disc cartilage. Unfortunately, the results of ectopic transplantation did not support this hypothesis (Fig. 6C,D, supplementary material Fig. S8D,E).
However, the ESC-derived paraxial mesoderm is similar to in vivo paraxial mesoderm, in that both sense the same set of developmental signals (hedgehog/SAG and Nog/LDN) and respond to them in the same way (sclerotomal specification then chondrocyte formation) as in vivo. Such signal-controllable differentiation capacity, together with the fact that many of the skeletogenic signaling mechanisms to which the ESC-derived paraxial mesoderm is expected to respond have already been elucidated, will place the cell system described in an advantageous position over other (adult cell) systems when it comes to finding a suitable condition for the formation and/or regeneration of stable cartilage. As for future therapeutic application, we have already reported the generation of chondrogenic paraxial mesodermal cells from human PSCs using a Gsk3 inhibitor (Umeda et al., 2012). Their capacity to go through a similar somitic chondrogenesis program remains to be determined.
MATERIALS AND METHODS
Cells, factors and antibodies
Mouse E14 ESCs and green fluorescence protein (GFP)-Bry ESCs kindly provided by G. Keller (Toronto, Canada) were maintained on mouse embryonic fibroblast feeders as described previously (Nakayama et al., 2003, 1998). EBRTcH3 ESCs carrying IRES-Bsd in the Oct3/4 locus (i.e. Blasticidin resistant) were maintained as described (Masui et al., 2005), except that the medium contained monothioglycerol (MTG, Sigma) in place of β-mercaptoethanol and was supplemented with 2 μg/ml Blasticidin S (InvivoGen) (hereafter EBR medium). The EBRβcatM cells were generated by inserting the Ctnnb1(S33Y)-Mer fusion gene (provided by T. Nakano, Osaka, Japan) i.e. the S33Y mutant β-catenin open reading frame fused with the FLAG-tag sequence at the N-terminus and the mutant estrogen receptor (Mer) sequence at the C-terminus, into the Rosa26 locus via Cre recombinase-mediated site-directed recombination (supplementary material Fig. S1A). The EBRβcatM cells were maintained in EBR medium supplemented with 0.5 μg/ml puromycin (InvivoGen) and 1 μM tetracycline (Tet; Sigma-Aldrich). Expression of the Ctnnb1(S33Y)-Mer fusion gene was induced by removal or reduction of Tet, which can be monitored by the emergence of YFP (Masui et al., 2005). Two clones (EBRβcatM Cl24-5 and Cl3-8) were used for biological studies. The protein growth and differentiation factors and protein inhibitors, media and supplements, buffers and specialized tissue culture plates, and antibodies for FACS, were sourced as described (Nakayama et al., 2003, 1998, 2000; Tanaka et al., 2009).
Differentiation induction in a chemically defined medium
ESCs were subjected to the EB-forming culture in a CDM-based semisolid medium as described (Tanaka et al., 2009). In the case of EBRβcatM cells, the pre-culture medium included 1.5 μg/ml puromycin and 10 nM Tet, and the differentiation medium was supplemented with 3-4.5 μg/ml puromycin and 4 nM Tet. The high concentration of puromycin with a reduced concentration of Tet during EB formation was used to maximize the yield of viable YFP+ progeny: i.e. viable progeny in which the Rosa26 locus was not silenced. The differentiation medium was supplemented on day 2 of differentiation with Wnt3a (Tanaka et al., 2009) at 100 ng/ml, BMP4 (R&D Systems) at 2 ng/ml, Noggin (Noggin-Fc, R&D Systems) at 100-500 ng/ml, BIO, Acetoxime-BIO (AcBIO), 1-methyl BIO (MeBIO) (all from Calbiochem), and CHIR99021 (BioVision) at 5 μM, and/or LDN193189 (Stemgent) at 0.6 μM, unless otherwise stated. activin A (PeproTech) was added at 1-25 ng/ml, and SB431542 (Tocris Bioscience) at 1-3 μM. Activation of the expressed Ctnnb(S33Y)-Mer protein was achieved by addition on day 2 of 4-hydroxytamoxifen (OHT, Sigma) at 0.9 μM.
Flow cytometry on EB cells
Flow cytometric analysis of EB cells and fractionation by FACS were performed as described (Tanaka et al., 2009), except that the viable cells were gated using DAPI (Sigma), and the cell sorter used was Aria or Aria II (BD Biosciences). Since the (Bry+)Pdgfrα+ cell fraction may contain endoderm, and since E-cadherin is a marker for endoderm (and ectoderm), E-cadherin (clone Decma1, Sigma) was added to the FACS strategy to ensure removal of the E-cadherin+ endoderm within the Pdgfrα+ fraction.
Two-dimensional (2D)-micromass culture was initiated by spotting 1×105 cells in 5 μl to a well of a 24-well plate and maintained in the chondrogenesis medium supplemented with 40 ng/ml PDGF (PDGF-BB, R&D Systems) as described (Tanaka et al., 2009). When indicated, TGFβ (TGFβ3, R&D Systems) was added at 10 ng/ml on day 1. When the EBRβcatM clone-derived YFP+P+ EB cells were used, Tet was included at 1 μM to suppress the accumulation of Ctnnb(S33Y)-Mer protein. Furthermore, for induction of the somitic chondrogenesis program, the chondrogenic medium with PDGF was further supplemented with either 300 ng/ml Shh (SHH-N, R&D Systems) or 100 nM SAG1 (Calbiochem), together with either 0.3-0.6 μM LDN193189 or 100-200 ng/ml Nog, both of which were replaced on day 6 with 50 ng/ml BMP4 (R&D Systems). On day 12 to 20 of culture, the developed micromass was either fixed with 10% (v/v) buffered Zn-formalin (Z-Fix, American MasterTech), and stained with 1% (w/v) acid Alcian Blue (Sigma) pH 1, as described (Nakayama et al., 2003), or subjected to analyses of DNA, RNA and sulfated glycosaminoglycans (sGAGs). The stained micromasses were then treated with 400 μl/well of 4 M guanidine hydrochloride in Tris.HCl-EDTA pH 7.5 (Hoemann, 2004), and the absorbance of the extract was measured at 600 nm (OD600) using SpectraMax M2 (Molecular Devices) to determine the relative amount of bound sGAGs per micromass. The 3D pellet culture was carried out in the serum-free chondrogenesis medium with PDGF, as described (Tanaka et al., 2009). The concentrations of protein factors and timings of addition were the same as for the 2D micromass culture. Developed cartilage particles were fixed on days 21-30 of culture by Z-Fix, paraffin sectioned, and stained primarily with 0.1% (w/v) Toluidine Blue (Sigma).
Isolation and quantification of DNA, RNA and sulfated glycosaminoglycan from micromass
The micromass was lysed using the guanidine isothiocyanate buffer (RLT, Qiagen), insoluble materials separated, and then DNAs, RNAs and proteins were individually isolated using AllPrep DNA/RNA/Protein Mini Kit (Qiagen). The purified RNAs were used for real-time RT-PCR analyses. The isolated proteins and the insoluble materials were both subjected to papain digestion for 18-20 h at 60°C [125 μg/ml papain, 10 mM cysteine in sodium phosphate-EDTA pH 6.5 (all from Sigma)], and released sGAGs were quantified with 1,9-dimethyl methylene blue [DMMB (Sigma), 16 μg/ml in glycine-NaCl pH 3], as described (Hoemann, 2004), using bovine tracheal chondroitin-4-sulfate (Biocolor) as standard. The OD590-530 was measured with SpectraMax M2. The DNAs isolated with the AllPrep kit and DNAs present in the papain-digest of insoluble materials were quantified using Hoechst 33258 (Sigma, 0.2 μg/ml in Tris.HCl-EDTA-NaCl pH 7.5), as described (Hoemann, 2004). The fluorescence (emission 460 nm, excitation 360 nm) was measured with SpectraMax M2 or Infinite M1000 (Tecan). The amounts of sGAG and DNA from a soluble extract and the corresponding insoluble material were combined to obtain total sGAG and DNA per micromass (μg sGAG/mass, μg DNA/mass), and the amount of sGAG was normalized to the amount of corresponding DNA to determine the active chondrocyte population within each micromass (μg sGAG/μg DNA).
Subcutaneous transplantation of cartilage particles
One to three cartilaginous particles (2-3 mm ‘wet’ diameter), formed by 3D pellet culture for 21-28 days using freshly sorted Flk1−Pdgfrα+ rostral paraxial mesoderm cells were wrapped in one 0.5 cm×1 cm Gelfoam (Pharmacia Upjohn) and transplanted to one or two sites under the dorsal skin of NSG mice. Ten to 12 weeks later, the cartilage/bone particles were harvested, fixed with Z-fix for 4 days, embedded in plastic, sectioned (5 μm), and stained with von Kossa counterstained with van Gieson, or with acid T-Blue [i.e. 0.1% (w/v) Toluidine Blue pH 3.7, 0.1% (w/v) urea (Sigma)]. The transplantation experiments were performed under the regulation of the IACUC for the University of Texas Health Science Center at Houston.
Real-time PCR for gene expression profiling
Total RNAs were purified with RNeasy Micro Kit or AllPrep RNA/DNA/Protein Mini Kit (Qiagen). Reverse transcription (RT) performed as described (Nakayama et al., 1998; Wang and Nakayama, 2009) was followed by the real-time quantitative PCR analysis using the ABI 7900 system (Applied Biosystems). All experiments were performed in duplicate or triplicate using either ABI SYBR Green Master mix and gene-specific primers, or ABI TaqMan Mix and TaqMan Gene Expression Assay (Applied Biosystems). The gene-specific oligonucleotide primers were purchased from Realtimeprimers.com or synthesized by IDT (supplementary material Table S1) (Tanaka et al., 2009). The eukaryotic translation elongation factor 1 alpha 1 (Eef1a1) gene was used as a reference gene, and the relative expression values were calculated as described (Wang and Nakayama, 2009). All RT-PCR results are presented as average relative expression levels with the s.d. shown as a thin vertical line. The P-value was determined by the Student's t-test using Excel or KaleidaGraph software.
We acknowledge T. Nakano for providing the Ctnnb1(S33Y)-Mer DNA constructs, Q. Yan for assistance in RT-PCR analyses, N. Matthias for transplantation, A. Hazen, D. Havilland and A. Fryga for cell sorting, S. Amra and M. Starbuck for histological sections and staining, and E. Zsigmond and A. Mitchell for proofreading.
J.Z., S.L., S.T., M.T., V.J.-J. and B.W. performed experiments and analysed data. H.N. constructed and provided the complete EBRTcH3 cell system. N.N. designed the research, performed additional experiments, analysed data and completed the preparation of the manuscript.
This work was supported partly by the Haemopoiesis Therapeutic Focus Fund [P017, P009 to N.N.] from the Australian Stem Cell Centre, by a Startup Fund (N.N.) from IMM UT Health, by the Rolanette and Berdon Lawrence Bone Disease Program of Texas Research Award (N.N.) and by the Innovative Research Grant of Arthritis Foundation (N.N.).
The authors declare no competing financial interests.