It has been established in the mouse model that during embryogenesis joint cartilage is generated from a specialized progenitor cell type, distinct from that responsible for the formation of growth plate cartilage. We recently found that mesodermal progeny of human pluripotent stem cells gave rise to two types of chondrogenic mesenchymal cells in culture: SOX9+ and GDF5+ cells. The fast-growing SOX9+ cells formed in vitro cartilage that expressed chondrocyte hypertrophy markers and readily underwent mineralization after ectopic transplantation. In contrast, the slowly growing GDF5+ cells derived from SOX9+ cells formed cartilage that tended to express low to undetectable levels of chondrocyte hypertrophy markers, but expressed PRG4, a marker of embryonic articular chondrocytes. The GDF5+-derived cartilage remained largely unmineralized in vivo. Interestingly, chondrocytes derived from the GDF5+ cells seemed to elicit these activities via non-cell-autonomous mechanisms. Genome-wide transcriptomic analyses suggested that GDF5+ cells might contain a teno/ligamento-genic potential, whereas SOX9+ cells resembled neural crest-like progeny-derived chondroprogenitors. Thus, human pluripotent stem cell-derived GDF5+ cells specified to generate permanent-like cartilage seem to emerge coincidentally with the commitment of the SOX9+ progeny to the tendon/ligament lineage.
Healthy joint cartilage, such as articular and meniscal cartilage, is stably maintained throughout life, but once damaged is not spontaneously repaired, especially in large animals and humans, leading to severe wear of the surrounding cartilage and consequent osteoarthritis (Buckwalter et al., 2014). One of the major challenges in the cell-based strategy for repairing cartilage lesion or regenerating lost cartilage is to prevent the in situ-regenerated cartilage or transplanted tissue-engineered cartilage from being committed to endochondral ossification (Somoza et al., 2014), which results in hypertrophic differentiation (Zhang et al., 2012), mineralization/calcification and the death of chondrocytes, and, ultimately, degradation of repaired cartilage tissue (Wang et al., 2016). These problems have not yet been completely resolved using adult mesenchymal stromal cells (MSCs) or expanded articular chondrocytes as reparative cells.
Joint articular cartilage is developed during embryogenesis by a specialized progenitor cell type, named the joint progenitor or interzone cell (Rux et al., 2019), and fetal articular cartilage possesses self-repairing activity even in large animal models (Namba et al., 1998; Ribitsch et al., 2018). We have therefore hypothesized that embryonic chondrocytes and/or chondroprogenitors may direct proper regeneration of articular cartilage even in the adult joint environment. For humans, however, pluripotent stem cells (PSCs), including embryonic stem cells (ESCs) and induced PSCs (iPSCs), are the only practical source of embryonic cells to test. Chondroprogenitors and MSCs have already been developed from mouse (m) and human (h) PSCs (reviewed by Nakayama et al., 2020). Many of them have been shown to preferentially generate growth plate-like, endochondral ossification-prone cartilage (hereafter, called ‘transient cartilage’), and some have been shown to repair articular cartilage defects (Cheng et al., 2014; Gardner et al., 2019; Toh et al., 2010). However, recently, we and others have successfully generated from hPSCs cartilage that expresses low-to-no hypertrophic chondrocyte markers in vitro and tends to stay in an unmineralized state when transplanted to ectopic sites in immunocompromised rodents, as did a piece of articular cartilage (hereafter, called ‘permanent cartilage’) (Craft et al., 2015; Lee et al., 2018; Umeda et al., 2015; Yamashita et al., 2015). Such cartilage was obtained by differentiation of hPSCs toward mesoderm without extensive expansion of the mesodermal progeny. However, the question of whether the permanent cartilage-forming cells will demonstrate better, longer-term repair of articular cartilage defects than the transient cartilage-forming cells has not yet been pursued in depth.
To obtain enough permanent cartilage-forming cells and transient cartilage-forming cells to address the question and to elucidate mechanistic basis of the permanent cartilage-forming activity in chondroprogenitors, we set out to develop methods to expand mesodermal and neural crest-like progeny of hPSCs. Our studies identified Sry-box 9 (SOX9)+ chondroprogenitors that can be expanded without loss of their chondrogenic activity in defined media containing fibroblast growth factor (FGF) and transforming growth factor-beta (TGFβ) receptor inhibitor (Lee et al., 2018; Umeda et al., 2015). SOX9 is the master transcription factor for chondrocytes (Lefebvre, 2019). Forced overexpression of SOX9 induced permanent chondrocytes from dermal fibroblasts (Hiramatsu et al., 2011). However, the expanded SOX9+ cells failed to show permanent cartilage-forming activity. Rather, they preferentially formed transient cartilage that readily developed into a bony tissue in vivo (Lee et al., 2018).
Here, we report the generation and characterization of novel growth and differentiation factor 5 (GDF5) gene-expressing progenitors, generated from SOX9+ chondroprogenitors derived from paraxial mesodermal progeny of hPSCs. Unlike SOX9+ cells, the GDF5+ cells preferentially formed permanent cartilage, potentially via non-cell-autonomous mechanisms. Comparative transcriptome analyses suggest a potential link between the acquisition of permanent cartilage-forming activity and that of teno/ligamento-genic potential during the genesis of GDF5+ cells from SOX9+ cells.
Generation and expansion of paraxial mesodermal progeny of hPSCs
As we and others have previously demonstrated (reviewed by Nakayama et al., 2020), paraxial mesoderm has been specified from PSCs commonly by stimulation of canonical WNT signaling using WNT3a or the glycogen synthase kinase (GSK) 3 inhibitor CHIR99021, and inhibition of bone morphogenetic protein (BMP) signaling using inhibitors of BMP or the BMP receptor, such as noggin or LDN193189, respectively. Such initial treatment was then followed by use of the activin/TGF-β/nodal receptor inhibitor SB-431542, FGF2 or the FGF receptor inhibitor, and/or platelet-derived growth factor (PDGF). We improved the differentiation method (Fig. 1A) by two additions: (1) BMP4 and FGF2 for the first 24 h, which synergistically enhanced mesendoderm specification (Fig. 1B, Fig. S1A,B), and (2) the retinoid X receptor agonist bexarotene for the last 48 h, which increased both the expression of paraxial mesoderm genes, such as MEOX1 (Dixon et al., 2017) (Fig. 1C,F, Fig. S1C), and the yield of KDR−/loPDGFRα+ cells (Fig. 1D,E) enriched in the paraxial mesodermal progeny (Fig. 1F).
On day 6-7 of differentiation, the KDR−/loPDGFRα+ paraxial mesoderm fraction was isolated by fluorescence-activated cell sorting (FACS) (Fig. 1D), and cultured in mesoderm growth medium (Fig. 2A), i.e. chemically defined medium (CDM) supplemented with FGF2, PDGF, SB-431542 and CHIR99021 (Fig. S2A, gray line). Small mesenchymal cells grew out 5-10 days later. Further growth of the cells appeared to depend primarily on FGF2, as they remained small (Fig. 2F, SOX9+ cells) and grew well without PDGF [i.e. in CDM with FGF2+SB-431542+ CHIR99021 (FSbC); light green in Fig. 2B and Fig. S2A). However, further changes in the medium, such as removal of CHIR99021 (red; FSb), replacement of FGF2 with PDGF (green; PSbC), or simultaneous removal of CHIR99021 and replacement of FGF2 with PDGF (purple; PSb), resulted in significantly slowed, or stopped cell proliferation.
Generation of GDF5+ cells and SOX9+ cells from the paraxial mesoderm cell fraction
The original KDR−/loPDGFRα+ paraxial mesoderm fraction contained cells that expressed SOX9 (‘PM’ in Fig. 2C). However, when the KDR−/loPDGFRα+ cells were grown in culture, the level of SOX9 transcripts increased, especially after the medium was shifted to ‘CDM with FGF2+SB-431542+CHIR99021’, and was maintained for over 30 days [light green (FSbC) in Fig. 2C and Fig. S2B]. In contrast, other media failed to maintain the expression of SOX9. Interestingly, however, when the culture medium was changed to CDM supplemented with PDGF and SB-431542, GDF5 mRNA became detectable and the level increased as the SOX9 mRNA level decreased [purple (PSb), Fig. 2C]. Furthermore, cells grown in another PDGF-based medium, in which SB-431542 was replaced with noggin [i.e. CDM with PDGF+noggin (PN)], showed a more rapid increase in GDF5 expression [blue (PN) compared with purple (PSb) in Fig. 2D], a decrease in SOX9 expression (Fig. 2E, Fig. S2D), and a slowed rate of cell proliferation, which resulted in large mesenchymal cells (GDF5+ cells; Fig. 2F). Both PDGF-based media supplemented with CHIR99021 (C) supported improved cell viability but delayed conversion of the SOX9+GDF5− state to the SOX9loGDF5+ state [green (PSbC) compared with purple (PSb) in Fig. 2C; light blue (PNC) compared with blue (PN) in Fig. 2D]. Therefore, the emergence of GDF5+ cells from the SOX9+ cell population required a change of culture medium from the FGF-based ‘CDM with FGF2+SB-431542+CHIR99021’ to the PDGF-based ‘CDM with PDGF+noggin+CHIR99021’, then to ‘CDM with PDGF+noggin’ (hereafter called ‘PNC/PN treatment’). Inhibition of the BMP receptor with LDN193189 had the same effect as inhibition of BMP with noggin (PLd; Fig. S2C).
Interestingly, GDF5+ cells thus generated also induced the expression of scleraxis (SCX), the master transcription factor gene of tendon/ligament precursor cells (Subramanian and Schilling, 2015) [blue (PN) in Fig. 2D and Fig. S2E]. In contrast, GDF5+ and SOX9+ cells generated and maintained in media containing SB-431542 failed to express SCX [green (FSbC) and purple (PSb) in Fig. 2D]. These observations were consistent with the notion that SCX expression is enhanced by TGFβ (Pryce et al., 2009) and inhibited by BMP (Bi et al., 2007; Schweitzer et al., 2001).
In agreement with the mRNA analyses, immunofluorescence (Fig. 2G,H, Fig. S2F-J) demonstrated that 67.7±7.4% (mean±s.e.m.) of the SOX9+ cells strongly (SOX9hi; Fig. 2H, green bars) and 24.7±6.1% weakly (SOX9lo) expressed SOX9 protein. Therefore, over 90% (67.7+24.7=92.4%) were SOX9+. However, SOX9 protein was largely undetected in GDF5+ cells (SOX9−; 94.7±1.2%; Fig. 2H, blue bars) and where detected was expressed only weakly (SOX9lo; 4.5±0.96%). In contrast, also consistent with the mRNA analyses, the immunofluorescence signal of GDF5 protein was weakly, but seemingly uniformly, detected among the GDF5+ cells (Fig. 2G), but never detected in the SOX9+ cell population.
These observations indicate that the SOX9+GDF5− cells expanded in ‘CDM with FGF2+SB-431542+CHIR99021’ (FSbC) medium are mostly SOX9+GDF5−, and that the SOX9loGDF5+ cells generated by PNC/PN-treatment are mostly SOX9−GDF5lo cells, possibly including SCX+ tendon/ligament precursor cells. Because only the differential expression of SOX9 protein was clear, the SOX9+GDF5− cells and SOX9loGDF5+ are hereafter designated as SOX9+ cells and GDF5+ cells, respectively.
Types of cartilage preferentially generated by SOX9+ cells and GDF5+ cells in vitro
Next, we tested the chondrogenic activity of the SOX9+ cells and GDF5+ cells using our standard pellet chondrogenesis culture, which involves TGFβ3 treatment followed by TGFβ3+BMP4 and then TGFβ3+GDF5 treatments (Lee et al., 2018; Umeda et al., 2015, 2012) (Fig. 3A). Even without further purification, both cell types showed robust chondrogenic activity (Fig. 3B,C). The BMP treatment (of either or both BMP4 and GDF5) was included to ensure the reproducible generation of large translucent hyaline cartilaginous pellets (Fig. 3D) that expressed only a low level of the fibrochondrogenesis marker the type I collagen (COL1) gene (COL1A1) (T versus TB and TG in Fig. 3E and Fig. S3B).
Pellets generated from the SOX9+ cells showed a marked increase in the level of the chondrocyte-specific COL2A1 mRNA from around day 10 and the hypertrophic chondrocyte-specific type X collagen (COL10A1) mRNA from approximately day 20 (Fig. 3B, green). At days 28-32 of culture, expression of RUNX2, an essential transcription factor for COL10A1, became obvious, and that of COL2A1 and COL10A1 persisted (Fig. 3C, green). These data collectively suggest that the SOX9+ cell-derived cartilage contained chondrocytes committed to hypertrophic differentiation.
Compared with the cartilage pellets produced by SOX9+ cells, those produced by GDF5+ cells were slightly larger (Fig. 3D, green versus blue) and expressed in vitro slightly lower levels of COL2A1, significantly lower levels of RUNX2 and very low to undetectable levels of COL10A1 (Fig. 3B,C, blue), suggesting that the GDF5+ cell-derived cartilage consisted mostly of non-hypertrophic chondrocytes. Consistently, the GDF5+ cell-derived cartilage also expressed higher levels of PTHLH, encoding parathyroid hormone-related peptide (PTHrP), and PRG4, encoding proteoglycan 4, also known as lubricin/superficial zone protein (Jay and Waller, 2014). PTHrP functions as an anti-hypertrophic differentiation factor for chondrocytes (Kronenberg et al., 2009; Zhang et al., 2012). PRG4 is a primitive/embryonic articular chondrocyte marker (Decker et al., 2017; Kozhemyakina et al., 2015). PRG4 was not expressed significantly in either SOX9+ cells or GDF5+ cells prior to induction of chondrogenesis (Fig. 2E, Fig. S3A). The level of COL1A1 mRNA did not differ significantly between the SOX9+ and GDF5+ cell-derived cartilage (Fig. 3C).
In keeping with such transcript analyses, the SOX9+ cell-derived, COL2-positive (Fig. 3F) cartilage pellets consisted of larger (i.e. morphologically hypertrophic) chondrocytes (Fig. 3G) that produced immunohistologically detectable levels of COL10 protein (Fig. 3H). The GDF5+ cell-derived, COL2-positive cartilage pellets (Fig. 3F) consisted of smaller chondrocytes (Fig. 3G) that produced a very low level of COL10 (Fig. 3H), but accumulated higher levels of sulphated glycosaminoglycan (sGAG) matrices (Fig. 3I).
Thus, both SOX9+ cells and GDF5+ cells are chondrogenic, but they differ in the type of chondrocytes they preferentially form. The SOX9+ cell-derived chondrocytes seem prone to BMP-stimulated hypertrophic differentiation, but the GDF5+ cell-derived chondrocytes seem less sensitive, or relatively resistant, to such an effect of BMP, at least at 28-32 days of pellet culture.
Difference in the in vivo stability of cartilage pellets formed from SOX9+ cells and GDF5+ cells
The ectopic transplantation model has been used to determine the degree of commitment of chondrocytes developed in vitro to endochondral ossification (Pelttari et al., 2006; Scotti et al., 2010). We previously applied the model to examine the property of chondrocytes generated from hPSC-derived ectomesenchymal cells (Lee et al., 2018). In the same way, we compared the difference in the in vivo stability of the GDF5+ cell-derived chondrocytes with that of the SOX9+ cell-derived chondrocytes (Fig. 4, Fig. S4). Metachromatic staining with Toluidine Blue (Tb; purple in Fig. 4) implies the presence of sGAG, an indicator of active chondrocytes, whereas the von Kossa staining (Vk; black) demonstrates mineral (i.e. calcium) deposition, indicating endochondral ossification. Fig. 4A,E shows typical SOX9+ cell-derived cartilage pellets recovered from mice, which lack cartilaginous sGAG matrices (Tb−) and have a mineralized interior with a marrow-like space (Vk+) (hereafter, called ‘bony pellet’). Fig. 4B,D shows typical GDF5+ cell-derived cartilage pellets in that the cartilaginous matrices remain uniformly distributed (Tb+) with minimal to no signs of mineralization (Vk−/lo) (hereafter called ‘full cartilage’). Interestingly, cartilage pellets generated with a 1:1 mix of SOX9+ and GDF5+ cells tended to resemble the GDF5+ cell-derived cartilage (Fig. 4C).
Quantitative analyses demonstrated that 53.0±9.7% (mean±s.e.m.) of the GDF5+ cell-derived cartilage pellets recovered from mice were Tb+Vk−/lo full cartilage and 26.0±14.4% were cartilage that was fully or partly Tb+ (Tb+/lo) but partly mineralized (Vklo) (hereafter called ‘mixed cartilage’; Fig. 4F,G,H, blue bars). In contrast, the SOX9+ cell-derived cartilage pellets recovered from mice were neither full nor mixed cartilage (Fig. 4H, green bars). Furthermore, 75.0±15.9% of them were fully mineralized, Tb−Vk+ bony pellets. In contrast, only 14.5±5.4% of the GDF5+ cell-derived cartilage pellets were classified as bony pellet (P=0.011). Some cartilage pellets showed a Tb−Vk− phenotype, suggesting the potential loss of chondrocytes in vivo.
Thus, SOX9+ cell-derived cartilage is mostly transient cartilage that tends to be extensively mineralized and lose the cartilaginous matrix in vivo. In contrast, GDF5+ cell-derived cartilage is mostly permanent cartilage that is either unmineralized or only partly mineralized and maintains cartilaginous matrix in vivo. Given that the former, but not the latter, expressed significant levels of hypertrophic chondrocyte markers prior to transplantation, the differences in the pre-transplantational state of chondrocytes seem to correlate well with their 8-week stability in ectopic sites in mice, as previously demonstrated with ectomesenchymal cells (Lee et al., 2018).
Demonstration of interaction between SOX9+ cells and GDF5+ cells during chondrogenesis, potentially through a non-cell-autonomous mechanism
To elucidate how GDF5+ cells and SOX9+ cells give rise to different types of chondrocytes, we first performed studies in which GDF5+ and SOX9+ cells were mixed in different ratios, similar to a previous study using MSCs and articular chondrocytes (Giovannini et al., 2010), and then subjected to a chondrogenic pellet culture (Mixed cell pellet, Fig. 5A). If both cell types were equally viable and able to differentiate independently into chondrocytes of their favored type (i.e. non-hypertrophic and hypertrophic chondrocytes, respectively), the level of hypertrophic chondrocyte marker gene expression (e.g. COL10A1, RUNX2) would be proportional to the SOX9+ cell content. However, that is not what we found. Instead, as the proportion of GDF5+ cells in the mix was increased to 30%, expression of RUNX2 and COL10A1 was knocked down (by 80% and 99%, respectively) and PRG4 expression increased (by 5-fold), to the levels achieved in pellets generated with 100% GDF5+ cells (Fig. 5B). When the 1:1 mixed cell pellet culture was performed using fluorescent SOX9+ cells derived from hiPSCs [i.e. SOX9-ZsGreen (GFP) cells] and non-fluorescent (NF) GDF5+ cells derived from hESCs for 5 days (which were pre-chondrogenic because TGFβ3 was not added until day 6) and analyzed by FACS (Fig. 5C), the SOX9-GFP+ cells comprised 50.3±4.2% (mean±s.e.m.; n=4) of total pellet cells, suggesting that SOX9+ cells were stably integrated into the pellet with GDF5+ cells. These results imply that GDF5+ cell-derived chondrocytes may non-autonomously control the expression level of COL10A1 and PRG4 in the nearby SOX9+ cell-derived chondrocytes.
Consistent with the reverse transcription-polymerase chain reaction (RT-PCR) and immunostaining results (Fig. 2E,G,H), GDF5+ cells derived from hiPSCs showed weaker SOX9-GFP signals than SOX9+ cells (Fig. 5C). However, during pellet culture, GDF5+ cells, but not SOX9+ cells, significantly upregulated the expression of SOX9-GFP over 5 days (Fig. S5C). Interestingly, the SOX9+ cells in mixed cell pellets seemed to weakly but consistently (n=4) upregulate the level of SOX9-GFP signal (Fig. S5B,C). Therefore, GDF5+ cells may affect nearby SOX9+ cells from an early stage of pellet culture, directing them to give rise to COL10A1−/loPRG4hi chondrocytes.
To seek evidence to support the proposed non-cell-autonomous mechanism, we co-cultured pre-formed pellets derived from SOX9+ cells and GDF5+ cells from day 5 or 6 of chondrogenesis culture (Fig. 5A). In one approach, a SOX9+ cell-derived pellet and a GDF5+ cell-derived pellet were combined to form a single gourd-shaped cartilage pellet (pellet contact co-culture). In a second approach, SOX9+ cell-derived pellets and GDF5+ cell-derived pellets were subjected to insert co-culture whereby both pellets were physically separated and never fused (pellet insert co-culture). The gourd-shaped fused pellets from the contact co-culture showed significantly lower levels of COL10A1 mRNAs than the control (combining a SOX9+ cell-derived pellet and a GDF5+ cell-derived pellet formed individually, right before RNA preparation) on days 30-32 (Fig. 5D), as was found for the mixed cell pellet analysis (Fig. 5B). In contrast, the insert co-culture showed no reduction in COL10A1 expression levels, but a significant increase in PRG4 expression levels was observed in a SOX9+ cell-derived pellet co-cultured with a GDF5+ cell-derived pellet (Fig. 5E, green). No significant change in COL10A1 and PRG4 expression levels were observed in the GDF5+ cell-derived pellets (P=0.39 and 0.85, respectively; Fig. 5E, blue).
These results suggest that GDF5+ cells may generate COL10A1−/loPRG4hi permanent-like chondrocytes by eliciting transmembrane signaling via the production of secreted molecules and/or membrane proteins, which may also work in trans to affect nearby SOX9+ cells during chondrogenesis, which become similar permanent-like chondrocytes as a result (i.e. non-cell-autonomous mechanism). Furthermore, whereas trans-suppression of COL10A1 expression seemed to be mediated by a secreted molecule that can travel only a short distance or by cell-to-cell contact, enhancement of PRG4 expression seemed to be achieved by a secreted molecule that can affect distantly located SOX9+ cell-derived chondrocytes.
Signaling analyses to investigate the hypothetical non-cell-autonomous mechanism
It is known that PTHrP is secreted from articular chondrocytes (Fischer et al., 2010), and suppresses chondrocyte hypertrophy by elevating the intracellular levels of cyclic adenosine monophosphate (cAMP) (Kronenberg et al., 2009). Given that the PTHrP receptor gene, PTH1R, was expressed in both SOX9+ and GDF5+ cells (Fig. S2D) and expression of PTHLH was increased in the GDF5+ cell-derived cartilage pellets (Fig. 3C), we addressed whether PTHrP might be one of the critical components of the hypothetical non-cell-autonomous mechanism, using the PTHrP agonist PTHrP(1-34) and the cAMP inducer forskolin (Lee et al., 2018). Both forskolin and PTHrP(1-34) effectively suppressed the COL10A1 expression in the SOX9+ cell-derived cartilage pellets (green compared with light green in Fig. 6A,B and Fig. S6A-C). However, although PRG4 is known to be a cAMP-inducible gene in chondrocytes (Ogawa et al., 2014), forskolin and PTHrP(1-34) treatments failed to elevate the expression level of PRG4, regardless of the type of BMP used for chondrogenesis (Fig. 6A-C, green compared with light green). Furthermore, we also failed to detect significant changes in the level of COL10A1 and PRG4 expression in cartilage pellets generated from GDF5+ cells when the PTHrP antagonist (Asn10,Leu¹¹,D-Trp¹²)PTHrP(7-34) (Fischer et al., 2014) was added (Fig. 6B, Fig. S6B). Therefore, GDF5+ cell-derived chondrocytes may utilize cAMP signaling to suppress hypertrophic differentiation (i.e. COL10A1 expression) of themselves as well as nearby SOX9+ cell-derived chondrocytes, but potentially via a molecule other than PTHrP. However, activation of cAMP signaling alone is insufficient for inducing PRG4 expression during chondrogenesis.
PRG4 expression in articular chondrocytes is also enhanced by TGFβ1/2/3 and BMPs (Niikura and Reddi, 2007). Because TGFβ3 is the standard component of pellet chondrogenesis media, we addressed whether a particular type of BMP is an essential component of the putative non-cell-autonomous mechanism that leads to enhancement of PRG4 expression. As noted, our standard chondrogenesis culture involves BMP4 and GDF5 for hyaline cartilage formation. Under these conditions, the level of PRG4 mRNA in SOX9+ cell-derived pellets was approximately 10% of that in GDF5+ cell-derived pellets (Fig. 6A,C, top). However, when only GDF5 (but no BMP4) was used, PRG4 expression in the SOX9+ cell-derived pellets increased (Fig. S3A, TG) and reached a level similar to that in the GDF5+ cell-derived pellets (Fig. 6C, bottom). Therefore, use of GDF5 and not exposing to BMP4 are both needed for upregulation of PRG4 expression in SOX9+ cell-derived cartilage pellets. Simple removal of BMP4 after the initial BMP4 treatment (Fig. 6D, TB→T) also failed to upregulate PRG4 expression in SOX9+ cell-derived pellets, although it somewhat reduced the COL10A1 levels (P=0.055) (Fig. 6D, green; Std compared with TB→T). In contrast, the expression of PRG4 in GDF5+ cell-derived pellets was enhanced by their prior exposure to BMP4, which was subsequently replaced with GDF5 (Fig. 6D, blue; Std). Furthermore, the anti-GDF5 antibody reported to neutralize GDF5 activity (Margheri et al., 2012) significantly reduced the level of PRG4 expression in GDF5+ cell-derived pellets treated with BMP4 alone (Fig. S6G, TB+Ab), suggesting that GDF5 secreted from GDF5+ cell-derived chondrocytes may support the PRG4 expression in GDF5+ cell-derived cartilage, even in the presence of BMP4. Therefore, effects of GDF5 on the expression of PRG4 seem to be differentially influenced by BMP4 in GDF5+ cell-derived chondrocytes and SOX9+ cell-derived chondrocytes. Thus, despite the GDF5+ cell-derived cartilage pellets expressing significantly higher levels of GDF5 than the SOX9+ cell-derived pellets (Fig. S6D), considering that GDF5+ cell-derived cartilage was able to affect in trans SOX9+ cell-derived cartilage to upregulate expression of PRG4 under the standard chondrogenesis condition, which includes a BMP4 treatment step (Fig. 5B,E), GDF5 secreted from GDF5+ cell-derived chondrocytes is unlikely to be the sole factor responsible for the non-cell-autonomous effects.
FGF18 and FGF2 signaling likely elicits anabolic and catabolic effects, respectively, on chondrocytes (Ellman et al., 2013), and FGF18 and its preferred receptor, FGFR3, are expressed in articular chondrocytes (Mori et al., 2014; Ornitz and Marie, 2019). However, the catabolic FGF2 was expressed in both SOX9+ cell-derived and GDF5+ cell-derived cartilage pellets (Fig. S6D,E), whereas expression of the anabolic FGF18 was undetectable. Interestingly, treatment with FGF18 from day 10 of pellet culture strongly suppressed PRG4 expression in both the SOX9+ cell- and GDF5+ cell-derived cartilage pellets (Fig. 6E), despite FGF signaling being known to increase PRG4 (lubricin) production from articular chondrocytes (Khalafi et al., 2007). In contrast, addition of a pan-FGFR inhibitor, either BGJ398 or AZD4547, from day 11 enhanced PRG4 expression (Fig. 6F, Fig. S6F). The FGF treatment also increased COL10A1 expression in GDF5+ cell-derived pellets (Fig. 6E, light blue), but decreased it in SOX9+ cell-derived pellets (light green). However, the FGFR inhibitor treatment weakly inhibited COL10A1 expression in both cartilage pellets (Fig. 6F). Therefore, GDF5+ cell-derived chondrocytes may also suppress FGF signaling in a non-cell-autonomous fashion.
In conclusion, signaling mechanisms that activate GDF5 and cAMP signaling and counteract FGF signaling mimic the proposed non-cell-autonomous effects of GDF5+ cell-derived chondrocytes. However, the molecular identities of the involved components remain unknown.
Comparative gene expression profile analyses of SOX9+ cells and GDF5+ cells
Aiming to elucidate the cell type of GDF5+ cells, we performed genome-wide, comparative RNA-sequencing (RNA-seq) analyses of SOX9+ cells and GDF5+ cells. Although both were chondrogenic, results indicated many differentially expressed gene (DEG) sets between them, suggesting that SOX9+ and GDF5+ cells may be distinct types of chondrogenic cells (Fig. 7A). The genes more strongly expressed in SOX9+ cells than in GDF5+ cells were relevant to growing chondroprogenitors: e.g. proliferation genes: LIN28A/B, TRIM71 (LIN41) and MYCN (Tsialikas and Romer-Seibert, 2015); cell cycle/DNA replication genes: MCM2-7, ORC1, CDT1, CDC7 and CDC25A (Sclafani and Holzen, 2007); chondrocytic genes: SOX5/6/8/9 and COL2A1; embryonic progenitor cell genes: SOX4/12 (Lefebvre, 2019) and SALL4; and SOX9 regulator genes, such as MAF and ZBTB16, similar to the ectomesenchymal cells (Umeda et al., 2015) (Fig. 7B, Table S3). In contrast, the slowly growing GDF5+ cells expressed higher levels than SOX9+ cells of cell cycle inhibitor genes: CDKN1A/2B/2C/2D; marker genes of articular chondrocytes and their precursors: COL22A1 (Feng et al., 2019), ITGA4 (CD49D) (Ferguson et al., 2018), GREM1 and DKK1 (Leijten et al., 2012), NFATC2 (Caldwell and Wang, 2015) and RUNX1 (Yano et al., 2019); genes encoding secreted factors involved in joint cartilage formation: TGFB1/2, GDF5/6/7 and WNT9A (Salva and Merrill, 2017); and genes of tenocytes/ligamentocytes and their precursors: COL1A1/2, SCX, MKX, TNMD and LOX (Subramanian and Schilling, 2015) (Fig. 7C, Table S4). Interestingly, GDF5+ cells also expressed the osteoblast gene RUNX2 at higher levels. Consistent RT-PCR results for SOX9, GDF5, COL2A1, RUNX2 and SCX are shown in Fig. 2C-E and Fig. S2B-E.
In addition, gene ontology (GO) enrichment analysis of DEGs indicated that SOX9+ cells would likely be involved in ‘DNA replication’, ‘palate development’, ‘nervous system development’ (Fig. S7B, Table S1) and ‘chondrocyte differentiation’ (Table S2). In contrast, the same analysis indicated that GDF5+ cells were likely associated with ‘cell adhesion’, ‘cell migration’, ‘angiogenesis’ (Fig. S7B, Table S1), ‘skeletal system development’, ‘osteoblast differentiation’ and ‘chondrocyte differentiation’ (Table S2). Further analyses, namely cell type deconvolution (Fig. S7A) and relative abundance (Fig. 7D), predicted that SOX9+ cells were likely composed of cells for which gene expression is characteristic of iPSCs, neural cells, smooth muscle cells and chondrocytes, whereas GDF5+ cells consisted predominantly of MSC-like cells with gene expression characteristic of ESCs and osteoblasts, and, less likely, of tissue stem cells.
These results prompted us to compare the transcriptomes of SOX9+ and GDF5+ cells with those of mouse interzone cells (GSE51098, GSE110281) (Feng et al., 2019; Jenner et al., 2014), mouse tendon/ligament precursor cells (GDS5642, GSE129820) (Eyal et al., 2019; Havis et al., 2014) and human neural crest-derived ectomesenchymal cells (GSE64752) (Umeda et al., 2015), in order to determine their potential cell types (Fig. 7E). Our data indicated that SOX9+ cells showed a strong correlation with the ectomesenchymal cells, as expected from the DEG list and enriched GO terms (Fig. 7B, Fig. S7B, Tables S1-S3) and the cell type deconvolution analysis (Fig. 7D, Fig. S7A). In contrast, GDF5+ cells showed a correlation with tendon/ligament precursor cells (especially in the GDS5642 dataset), consistent with the results from the DEG analyses (Fig. 7C, Tables S4-S7), but only a weak correlation with interzone cells (especially the GSE51098 dataset).
Cell surface molecule analyses of SOX9+ cells and GDF5+ cells
The DEG analyses have also predicted potential cell surface receptors/markers that distinguish, and could be used to purify, SOX9+ cells and GDF5+ cells, or to determine heterogeneity in them: e.g. CD271 (NGFR) and CD200 for SOX9+ cells, and CD44, CD49D (ITGA4), CD73 (NT5E), CD105 (ENG), CD127 (IL7R), CD146 (MCAM) and CD166 (ALCAM) for GDF5+ cells (Tables S3, S4), which largely overlap with the MSC markers (Lv et al., 2014). In order to validate the RNA-seq data further, we conducted FACS analyses on SOX9+ cells and GDF5+ cells using antibodies against these markers and confirmed that differences in their protein expression levels between SOX9+ cells and GDF5+ cells correlated perfectly with differences in their gene expression levels (Fig. 7F,G, Fig. S7C-E) with the exception of CD127, which FACS failed to detect. Interestingly, heterogeneity detected by FACS in the GDF5+ cell population was minimal, and was only apparent based on CD200 expression (Fig. 7F, Fig. S7D), but heterogeneity in the SOX9+ cell population was apparent by CD44, CD73, CD146 and CD166 expression (Fig. 7F,G, Fig. S7D,E).
We have demonstrated that: (1) paraxial mesodermal progeny of hPSCs can give rise to a GDF5+(SOX9lo) chondrogenic cell population in culture, probably through SOX9+ (GDF5−) chondroprogenitors, consistent with the observation that Gdf5 cells are found in the joint area between developing vertebral processes in the mouse (Settle et al., 2003); (2) hyaline cartilage pellets generated from SOX9+ cells under our standard chondrogenesis condition, which involves BMP treatments, tend to express higher levels of hypertrophic chondrocyte markers and lower levels of PRG4 in vitro, whereas those from GDF5+ cells express low levels, if any, of the hypertrophic markers and higher levels of PRG4; (3) when ectopically transplanted, the SOX9+ cell-derived cartilage pellets are unstable and readily mineralized, whereas the GDF5+ cell-derived cartilage pellets are stable and largely unmineralized (i.e. permanent cartilage); (4) GDF5+ cells appear to generate PRG4hi hyaline permanent cartilage through non-cell-autonomous mechanisms that also appear to work on nearby chondrocytes generated from SOX9+ cells; (5) such effects can be mimicked by activation of cAMP signaling and selected BMP (i.e. GDF5 but not BMP4) signaling, and/or suppression of FGF signaling; and (6) the genome-wide transcriptome of GDF5+ cells differs from that of SOX9+ cells, despite both cells being chondrogenic, and resembles the transcriptome of developing tendon/ligament (precursor) cells.
GDF5 is a member of the BMP family, and is involved in the genesis and maintenance of articular cartilage (Lyons and Rosen, 2019; Thielen et al., 2019) and tendons/ligaments (Harada et al., 2007; Wolfman et al., 1997). Gdf5 is also a marker gene for the interzone cells (or embryonic joint progenitors) in developing limb and digit joints, which not only contribute to the formation of joint articular cartilage but also to that of other joint tissues, such as meniscal cartilage, intra-joint ligaments, and synovial lining and capsule during mouse embryogenesis (Koyama et al., 2008; Shwartz et al., 2016). The interzone cells are recruited from Sox9+ cells (Rux et al., 2019) and require noggin for their formation (Lyons and Rosen, 2019). They are mostly non-growing cells (Ray et al., 2015), but show a short-term articular cartilage repair activity (Feng et al., 2019). Furthermore, articular chondrocytes non-autonomously prevent growth plate chondrocytes (D'Angelo and Pacifici, 1997; Jikko et al., 1999) or MSC-derived chondrocytes (Fischer et al., 2010; Giovannini et al., 2010) from hypertrophic differentiation.
These properties of interzone cells and their progeny (articular chondrocytes) resemble those of GDF5+ cells and the chondrocytes derived from them. However, bioinformatic analyses of the RNA-seq data indicated that the GDF5+ cells might relate more closely to developing tenocytes/ligamentocytes than to interzone cells (Fig. 7B,D). In support of this suggestion, the genes preferentially expressed in Scx+ tendon precursor cells from the embryonic day 14.5 mouse limb seem to correlate better with the GDF5+ cell-restricted genes (seven genes in the SOX9+ cell DEGs versus 40 genes in the GDF5+ cell DEGs; Table S7) than those associated with the interzone cells (21 genes versus 32 genes; Table S5). Furthermore, the GDF5+ cells did not express LGR5, but expressed COL22A1 and SCX at higher levels than SOX9+ cells (Figs 2D and 7C, Table S4). The Lgr5+(Col22a1−Scxlo) subpopulation of the Gdf5+ mouse interzone cells represent all joint progenitor activities, and upregulation of Col22a1 and Scx expression is associated with their commitment to chondrocytes and ligamentocytes/tenocytes, respectively (Feng et al., 2019). Progenitor cells contributing to the formation of tendon enthesis are also Gdf5+ (Dyment et al., 2015, 2014) and Sox9+Scx+ (Blitz et al., 2013; Sugimoto et al., 2013). Therefore, the hPSC-derived GDF5+ cells might be a mixture of both chondroprogenitors and ligament/tendon (enthesis) progenitors. Such cellular heterogeneity might help in the generation of the PRG4hi permanent cartilage-forming activity, but might in turn cause the variations in the type of cartilage recovered from mice (Fig. 4H, blue). However, FACS analyses have provided only limited support for such to be the case (Fig. 7F,G, Fig. S7C-E). In addition, the GDF5+ cells might represent a differentiation intermediate between sclerotomal chondrogenic cells (SOX9+) and syndetomal teno-/ligamento-genic cells (SCX+) of the somite (Fig. 8). Mechanisms associated with such differentiation may be linked to those that direct chondrogenesis to give rise to PRG4hi permanent cartilage.
The RNA-seq analyses indicated similarity between SOX9+ cells derived from paraxial mesodermal progeny and ectomesenchymal cells derived from neural crest-like progeny (Fig. 7E), and the presence of cells related to neurons in the SOX9+ cell population (Fig. 7D, Fig. S7A). These results suggest that the paraxial mesodermal progeny we generated and isolated from hPSCs contained cells that were not fully determined, so that the expansion culture conditions may direct such undetermined cells to become SOX9+ cells, similarly to ectomesenchymal cells arising from neural crest-like progeny. Alternatively, the isolated mesoderm fraction may be contaminated with cells committed to neural crest such that expansion culture resulted in a significant proportion of ectomesenchymal cells within the resulting SOX9+ cells. Recently, Wu et al. (2021) demonstrated that the in vitro differentiation pathway from hPSCs to paraxial mesodermal progeny and to chondrocytes seems to go through an intermediate with neurogenic potential, observations that seem to support the former possibility.
Finally, the successful generation of two types of chondrogenic mesenchymal cells, SOX9+ (GDF5−) cells and GDF5+ (SOX9lo) cells, which are committed to giving rise preferentially to distinct types of chondrocytes indicates that the post-transplantational fate of chondrocytes can be controlled, even at the mesenchymal cell stage. Therefore, although establishing whether the GDF5+ cells more faithfully reproduce long-lasting articular cartilage in the joint than other chondroprogenitors is important to the development of hPSC-based cartilage regenerative therapy, mechanistic studies of how such commitment is made in chondroprogenitors are equally important for improving available adult stem cell-based therapies for articular cartilage repair. In this sense, it is worth mentioning that the current DEG analyses (Tables S3 and S4) may have already nominated candidate signaling mechanisms, e.g. those for activin, connective tissue growth factor, epidermal growth factor, interleukin 11, netrin 4, endothelin 1 and angiopoietin-like 1, that may be involved in reprogramming SOX9+ cells and even adult stem/progenitor cells to generate articular-like permanent chondrocytes preferentially during in vitro chondrogenesis or after transplantation, a possibility that is currently under investigation.
MATERIALS AND METHODS
H9 hESCs from WiCell and CY2;SOX9-2A-ZsGreen-2A-Puro (herein abbreviated as SOX9-GFP) hiPSCs from the former NIH Center for Regenerative Medicine were tested for contamination and maintained in Essential-8 medium on a vitronectin-coated plate (Invitrogen), as described (Lee et al., 2018; Umeda et al., 2015). The neomycin/G418-resistance gene cassette was removed from the original SOX9-GFP hiPSCs using the TAT-CRE protein transfection kit (SCR508, EMD Millipore) according to the manufacturer's recommendation.
Generation and isolation of paraxial mesoderm from hPSCs
Human PSCs were differentiated three-dimensionally using chemically defined medium-based embryoid body (EB)-forming culture, as described (Lee et al., 2018; Umeda et al., 2015). Briefly, EB culture was initiated in chemically defined EB medium (CDEBM): Iscove modified Dulbecco's Medium (IMDM): Ham's F12 (1:1) (Invitrogen), 5 mg/ml fatty acid-free bovine serum albumin (BSA; Sigma-Aldrich), 2% (v/v) chemically defined lipid concentrate (Invitrogen), 2 mM GlutaMax (Invitrogen), 100 µg/ml holo-transferrin (Sigma-Aldrich), 20 µg/ml insulin (Sigma-Aldrich), 0.45 mM monothioglycerol (MTG; Sigma-Aldrich), 0.17 mM ascorbic acid-2-phosphate (AA2P; Sigma-Aldrich), 2.5 µg/ml catalase (CAT; Sigma-Aldrich), 5 mM proline (Pro; Sigma-Aldrich) and 1.5 µg/ml reduced glutathione (G-SH; Sigma-Aldrich), supplemented with protein and small molecule factors 5 µM Y27632 (EMD Millipore), 5 ng/ml BMP4 (hBMP4, R&D Systems), 5 µM CHIR99021 (Tocris and Selleck) and 20-25 ng/ml FGF2 (hFGF2-IS, Miltenyi Biotec) at 37°C under 5% CO2/5% O2. On day 1 (24 h later), medium was changed to CDEBM supplemented with 5 µM CHIR99021 and 100 ng/ml noggin (m/hNoggin-Fc chimera, R&D Systems). Where indicated, medium was changed on day 1 to CDEBM supplemented with CHIR99021 and on day 2, noggin was added (day-2 noggin addition). On day 3, EBs were transferred to CDEBM containing 0.9% (w/v) methylcellulose (Methocel A4M, Dow Chemical) with 100 ng/ml noggin, 0.5 µM CHIR99021, 0.2 µM PD173074 (FGFR1 inhibitor, Tocris), 3 µM SB-431542 and 5 ng/ml PDGF (hPDGF-BB; R&D Systems). On day 4 (or 5), 1 µM bexarotene (RXR agonist; Sigma-Aldrich) was added, and on day 6 (or 7), single EB cells were obtained by treatment of EBs with TrypLE Select (Invitrogen) for 3-5 min at 37°C and resuspended at 5×106 cells/ml or less in 0.5% (w/v) BSA (Sigma-Aldrich)-containing PBS without Mg2+Ca2+ (D-PBS, Invitrogen). The cells were then stained with mouse anti-hPDGFRα (IgG2a, 556001, BD Biosciences) and anti-KDR (IgG1, 101-M20 ReliaTech) monoclonal antibodies, followed by biotin-conjugated goat anti-mouse IgG2a (1080-08), then with phycoerythrin (PE)-conjugated goat anti-mouse IgG1 antibodies (1070-01) from SouthernBiotech, and finally with allophycocyanin (APC)-conjugated streptavidin (554067, BD Biosciences). The KDR−/loPDGFRα+ cell population was isolated by FACS, as described (Umeda et al., 2012).
Generation and isolation of lateral plate/extra-embryonic mesoderm from hPSCs
As indicated in Fig. S1D, delay in noggin addition to day 2 allowed leakage of lateral plate mesoderm marker expression during EB culture of hPSCs. To suppress expression of paraxial mesoderm markers completely and maximize that of lateral plate mesoderm markers (i.e. FOXF1 and PRRX1), we omitted the noggin treatment. Therefore, we initiated the EB culture in CDEBM supplemented with 5 µM Y27632, 5 ng/ml BMP4, 5 µM CHIR99021 and 20-25 ng/ml FGF2, at 37°C under 5% CO2/5% O2. On day 1 (24 h later), medium was changed to CDEBM containing 5 µM CHIR99021, and on day 3, EBs were transferred in 0.9% (w/v) methylcellulose containing CDEBM, supplemented with 15 ng/ml WNT3a (hWNT3a; R&D Systems), 10 ng/ml vascular endothelial growth factor (hVEGF; R&D Systems), 10 ng/ml FGF2 and 50 ng/ml stem cell factor (hSCF; StemGen/Amgen).
FACS analysis was performed on an LSR II (BD Biosciences) or Cytek Aurora (Cytek Biosciences). Cell sorting was carried out with a FACS Aria II (BD Biosciences) as described (Umeda et al., 2015, 2012). Viable single cells were gated using 0.5 ng/ml DAPI (4′,6-diamidino-2-phenylindole) (Sigma-Aldrich) for LSR II and Aria II and Sytox Blue (Invitrogen) at 1/2000 for Cytek Aurora. Sorted cells ranged in purity from 90 to 95%. The Cytek Aurora is a spectral flow cytometer so raw data acquired were subjected to ‘unmixing’ using SpectroFlo software (Cytek Bio).
Generation and expansion of SOX9+ cells from hPSC-derived paraxial mesodermal cells
The FACS-isolated KDR−/loPDGFRα+ mesodermal cells were cultured on a gelatin/fibronectin-coated plate in mesoderm growth medium, i.e. chemically defined medium [CDM; the same as CDEBM without G-SH but with 2 µg/ml heparin (Sigma-Aldrich)], supplemented with protein and small molecule factors (5-10 ng/ml FGF2, 10-15 ng/ml PDGF, 7.5 µM SB-431542 and 2 µM CHIR99021) at 37°C under 5% CO2/5% O2 and passaged every 2-3 days (when nearly confluent), at 3×104 cells/cm2 using TrypLE Select. Passage 1-3 cells were stored frozen. For generating SOX9+ cells, medium was changed to CDM with FGF2, SB-431542 and CHIR99021 (FSbC: the mesoderm growth medium without PDGF). Inclusion of SB-431542 in the growth medium was based on our previous success in the genesis of chondrogenic ectomesenchymal cells from hPSC-derived neural crest-like progeny and their long-term maintenance without loss of chondrogenicity using SB-431542 (Umeda et al., 2015).
Generation of GDF5+ cells from hPSC-derived paraxial mesodermal cells
The standard PNC/PN treatment method
FACS-isolated mesodermal cells maintained in mesoderm growth medium and passaged once in CDM with FGF2, SB-431542 and CHIR99021 (FSbC medium) (i.e. SOX9+ cells) were passaged again at 3×104 cells/cm2 and 5-15 h later the medium was changed to CDM supplemented with 15 ng/ml PDGF, 50-100 ng/ml noggin and 2 µM CHIR99021 (PNC medium). The cells were maintained in PNC for 7-10 days (and passaged when nearly confluent, usually once at 1:3). Between 5 and 15 h after the second passage, the medium was changed to CDM with PDGF+noggin (PN medium), and the culture was maintained for 3-6 days (until nearly confluent). PLd medium, consisting of CDM with PDGF plus 0.1-0.3 µM LDN193189, also worked in place of PN (Fig. S2C). In some experiments, PNC culture was continuously maintained, or ‘CDM with FGF2, SB-431542 and CHIR99021’ (i.e. FSbC)-cultured SOX9+ cells were directly transferred into PN medium.
In some experiments (Fig. 2, Fig. S2), SOX9+ cells grown in mesoderm growth medium and in CDM with FGF2, SB-431542 and CHIR99021 were treated as described above except that the PNC medium was replaced with CDM supplemented with 15 ng/ml PDGF, 7.5 µM SB-431542 and 2 µM CHIR99021, and the PN medium was replaced with CDM supplemented with PDGF+SB-431542.
Mesenchymal cells were maintained by passaging 2-3×104 cells/cm2 every 2-5 days. The total number of cells in each passage was calculated as a ratio of harvested cell number to seeded cell number, multiplied by the calculated cell number of the previous passage.
Scaffold-free cartilage formation: pellet culture
To induce chondrogenesis from hPSC-derived chondroprogenitors, aliquots of 2×105 to 4×105 cells were centrifuged at 200 g for 6 min in a 15-ml conical tube (BD Biosciences) to form a pellet, and cultured in 0.5 ml of the incomplete serum-free chondrogenic media: Dulbecco's Modified Eagle Medium (DMEM, high glucose; Invitrogen), 1 mM sodium pyruvate (Invitrogen), 1% (v/v) ITS+ (BD), 50 nM dexamethasone (Sigma-Aldrich), 0.17 mM AA2P, 0.35 mM Pro, 2 mM GlutaMax and 50 µM MTG, supplemented with 40 ng/ml PDGF. On day 5 or 6 of pellet culture, 10 ng/ml TGFβ3 (hTGFβ3, R&D Systems) was added to induce chondrogenesis; on day 10, 50 ng/ml BMP4 was added; and then on day 17, the BMP4 was replaced with 50 ng/ml GDF5 (hGDF5; R&D Systems): i.e. d5/6 TGFβ3→d10 ‘TGFβ3+BMP4’→d17 ‘TGFβ3+GDF5’. The pellet cultures were maintained at 37°C under 7.5% CO2 and generally ended on days 28-32. Alternatively, at the times indicated, pellets were treated with only one type of BMP, such as BMP4 (TB conditions: d5/6 TGFβ3→d10 ‘TGFβ3+BMP4’) or GDF5 (TG conditions: d5/6 TGFβ3→d10 ‘TGFβ3+GDF5’), or with no BMP (T conditions: TGFβ3) from day 10 to the end of culture. In some experiments, PDGF was replaced with 40 ng/ml FGF18 (R&D Systems) on day 10, or one of the following was added on day 11: forskolin, an activator of adenylyl cyclase, at 25 µM (Tocris) (Lee et al., 2018); PTHrP(1-34) (Bachem) at 1 µM (Lee et al., 2018); the PTHrP antagonist (Asn10,Leu¹¹,D-Trp¹²)PTHrP(7-34) (Bachem) at 1-2 µM (Fischer et al., 2014); the pan-FGF receptor inhibitor BGJ398 (Selleck Chemicals) (Chen et al., 2014) or AZD4547 (Selleck) at 0.5 µM (Mack et al., 2018); or an anti-hGDF5 antibody (PA5-29775, Invitrogen) at 1-3 µg/ml (Margheri et al., 2012). The concentrations used were in a published range that had been successful for various cell-based assays. Some cartilage pellets were then fixed with Zinc-Formalin (Z-Fix, Anatech) for 1 day, paraffin-embedded, sectioned (5 µm), deparaffinized, rehydrated and stained with 1 mg/ml Toluidine Blue (Sigma-Aldrich), 1 mg/ml urea (Invitrogen) and 30% (v/v) ethanol (Millipore), or immunostained with anti-COL1, -COL2 and/or -COL10 antibodies as described in the relevant section below.
SOX9+ cell and GDF5+ cell staining for FACS analyses
For SOX9+ and GDF5+ mesenchymal cell staining, cells were collected using TrypLE Select (Invitrogen), resuspended in 0.5% BSA D-PBS, and then stained with Alexa Fluor 647-conjugated mouse anti-human CD271 (560326) from BD Biosciences (at 2 µl per 100 µl cell suspension), and with PE-conjugated mouse anti-human CD44 (338807), CD49D (304303), CD73 (344004), CD105 (323205), CD127 (351303), CD146 (342003), CD166 (343903) or CD200 (329205) from BioLegend (at 1 µl per 100 µl cell suspension), at 0°C for 30 min.
Pellet cell preparation for FACS analysis
Day-5 pellets formed in the incomplete serum-free chondrogenic media were treated for 1 h at 37°C with 4 mg/ml collagenase [D (Roche)+XI (Sigma-Aldrich) 1:1 mix] in IMDM, followed by TrypLE Select (Invitrogen) treatment for 10 min at 37°C. Cell aggregates/clumps were removed with a 40-µm mesh (BD Biosciences) prior to FACS analysis.
Pellet contact co-culture and pellet insert co-culture
Pellet contact co-culture was initiated by transferring a GDF5+ cell-derived pellet into the culture tube of a SOX9+ cell-derived pellet on day 5 or 6 of pellet culture, and the culture was continued as described in the previous section. Pellet insert culture was performed using a non-stick 24-well plate for suspension cell culture (Sarstedt) and polyethylene terephthalate (PET)-based hanging tissue culture (TC) inserts (Sarstedt, pore size 0.4 µm). On day 6 of pellet culture, a GDF5+ cell-derived pellet was transferred to a well of the non-stick 24-well plate, and a SOX9+ cell-derived pellet was transferred into a PET-TC insert. The SOX9+ cell-derived pellet-containing insert was then placed in the well containing GDF5+ cell-derived pellets to initiate insert culture. Both SOX9+ cell-derived pellets and GDF5+ cell-derived pellets were also individually transferred to the wells of 24-well plate as non-co-culture control. The culture was continued as described in the previous section. However, culture volume was changed to 1.5 ml/well and the gas conditions were changed to 7.5% CO2/3% O2.
Isolation and quantification of DNA, RNA and sGAG from cartilage pellets
Cartilage pellets were collected and submerged in liquid N2, manually cracked into small pieces with a liquid N2-cooled mortar and a pellet pestle, and homogenized in lysis buffer (RLT, Qiagen). The cleared lysates were then subjected to DNA, RNA and protein isolation using the AllPrep DNA/RNA/Protein mini kit (Qiagen). The purified RNA was used for real-time RT-PCR analysis. The isolated proteins were subjected to papain digestion for 20-24 h at 60°C [125 µg/ml papain, 10 mM cysteine in sodium phosphate-ethylenediamine tetraacetic acid (EDTA) pH 6.5 (all from Sigma-Aldrich)], and released sGAG was quantified by the 1,9-dimethyl Methylene Blue [DMMB, 16 µg/ml in glycine-NaCl pH 3 (Sigma-Aldrich)] serial-dilution assay, using bovine tracheal chondroitin 4-sulfate (Biocolor) as standard (Hoemann, 2004). The OD590-530 was measured with SpectraMax M2 (Molecular Devices). The DNA isolated with the AllPrep kit was quantified by a Hoechst 33258 (Sigma-Aldrich; 0.2 µg/ml in Tris·HCl-EDTA-NaCl pH 7.5) serial-dilution assay, using bovine thymus DNA (Sigma-Aldrich) as standard. The fluorescence (emission 460 nm, excitation 360 nm, 420 nm cutoff) was measured with SpectraMax M2. The total sGAG and DNA amounts per pellet, along with comparative ratios of the sGAG and DNA, were then calculated. Results are presented as mean values with s.e.m. shown by thin error bars.
Gene expression profiling
The isolated RNA was reverse transcribed using a Superscript III kit (Invitrogen) and real-time RT-PCR was performed using the Taqman Gene Expression Assay and ABI7900 (Applied Biosystems) or CFX Connect (Bio-Rad). The expression levels of individual genes from duplicate or triplicate reactions were normalized to that of EEF1A1 transcript (2−ΔCt×100) and averaged to obtain relative expression, as described (Wang and Nakayama, 2009). The RT-PCR results are presented as mean relative expression levels with s.e.m. shown by the thin error bars, unless indicated otherwise: e.g. s.d. Undetectable levels of all the genes tested lie in the relative expression range of 0.001 to 0.0001. When indicated as ‘% COL2A1’, for example, the expression level of a gene has been normalized to that of COL2A1 and multiplied by 100.
For RNA-seq, total RNAs from four independent sets of SOX9+ cells (passage 4-6 in FSbC medium, derived from H9 hESCs), and four independent sets of GDF5+ cells (generated by PNC/PN-treatment, derived from H9 hESCs) were extracted with an RNeasy mini kit (Qiagen). Poly (A)-tailed messenger RNA was enriched and the RNA-seq library was prepared in the UTHealth Cancer Genomics Core following the instructions of the KAPA mRNA HyperPrep Kit (KK8581, Roche Holding AG) and KAPA Unique Dual-indexed Adapter kit (KK8727, Roche). RNA-seq was performed using the Illumina Nextseq550 system with the 75 bp pair-ended running mode. Raw mRNA sequence reads were pre-processed using Cutadapt (v1.15) to remove bases with quality scores <20 and adapter sequences (Martin, 2011), followed by alignment of clean RNA-seq reads to GRCh38.83 with STAR (v2.5.3a) (Dobin et al., 2013). Uniquely mapped reads overlapping genes were counted by HTseq-count with default parameter using annotation from ENSEMBL v83. Only genes with more than five reads in at least one sample were retained. The raw read counts of retained genes were submitted for differential expression analysis of cases compared with controls with DESeq2 software (Anders and Huber, 2010), which uses a model based on the negative binomial distribution. Resulting P-values were adjusted using the Benjamini and Hochberg approach (Benjamini and Hochberg, 1995) to control for false discovery rate (FDR). Genes with fold change (FC)>2 (or FC<0.5) and FDR<0.05 were assigned as DEGs. Standard gene set enrichment analysis was performed with a hypergeometric test using RDAVID WebService (v1.19.0) (Fresno and Fernandez, 2013). The resulting P-values were also adjusted using the Benjamini and Hochberg approach. The analyzed data are summarized in Tables S1-S7.
Cell type deconvolution analysis
Cell type composition scores were computed using CIBERSORT v1.04 (Newman et al., 2015) to deconvolute the data and predict cell type compositions. CIBERSORT takes a reference single-cell expression panel and implements a support vector regression (SVR)-based machine learning approach to estimate the composition of each cell type. Here, the reference set was downloaded from SingleR software (Aran et al., 2019). Cell type composition score was compared between two groups (SOX9+ cells and GDF5+ cells) by unpaired, two-tailed Student's t-test.
Correlation analysis of transcriptome data sets
We downloaded Gene Expression Omnibus transcriptome data sets GDS5642, GSE110281, GSE129820, GSE51098, GSE64752. Gene expression levels for Entrez or HGNC symbols were assigned to the corresponding homologous gene ENSEMBL identifier. Genes with >1 count per million in 10% of the samples in each data set were retained. Then, common genes identified across all datasets were kept. All expression values were log2 converted followed by quantile normalization. Uniform manifold approximation and projection (UMAP) analysis (McInnes et al., 2018) was performed using UMAP function in R, based on all genes in the merged data set.
Subcutaneous transplantation and histological analysis of cartilage pellets
The subcutaneous transplantation of cartilage pellets was performed as described (Lee et al., 2018; Umeda et al., 2015) under the regulation of IACUC for the University of Texas Health Science Center at Houston. In preparation for cartilage transplantation, 7- to 12-week-old female immunocompromised NSG mice (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ; #005557, The Jackson Laboratory) were anesthetized with isoflurane. After they had lost the pedal withdrawal reflex, buprenorphine was injected subcutaneously near the proposed site of incision, followed by clipping of back hair, skin disinfection with chlorhexidine and 70% (v/v) alcohol, and placement of a sterile drape around the area of incision. Mice were placed on a heated pad during the procedure to preserve body temperature. Two mid-longitudinal skin incisions of approximately 1 cm were made on the dorsal neck area of each mouse, and subcutaneous pockets formed by blunt dissection. In vitro-prepared cartilage pellets (of approximately 1-5 mm ‘wet’ diameter) were individually placed into each pocket. Each mouse received one or two transplants. Incisions were closed with skin adhesive. After 8 weeks, the transplanted mice were euthanized and cartilage pellets were harvested, fixed with Z-Fix for 4 days, embedded in plastic, sectioned (5 µm), deplastified in 1-acetoxy-2-methoxyethane (Sigma-Aldrich) for 30 min, rehydrated, and stained with von Kossa then counterstained with van Gieson, or with Toluidine Blue. Control experiments using a piece of articular cartilage surface of 2-year-old bovine knee were performed as previously described (Lee et al., 2018).
Immunofluorescence staining of SOX9+ and GDF5+ mesenchymal cells
The FSbC-expanded SOX9+ cells and PNC/PN-treated GDF5+ cells were transferred to gelatin-coated 24-well plates and cultured for 3 days in FSbC and PN medium, respectively. Cultures were stopped and fixed with Z-fix for 20-30 min at 4-8°C, then immunostained initially with the goat anti-hSOX9 polyclonal antibody (3 μg/ml, AF3075, R&D Systems) and the rabbit anti-SCX polyclonal antibody (1:150, HPA043183, Sigma-Aldrich), and then with biotinylated donkey anti-goat IgG (705-065-147, Jackson ImmunoResearch) and PE-conjugated donkey anti-rabbit IgG (406421, BioLegend), followed by Alexa Fluor 488-conjugated streptavidin (S32354, Invitrogen). The stained plates were treated with DAPI and inspected with a TE2000-E inverted fluorescence microscope (Nikon). Alternatively, the SOX9+ cells and GDF5+ cells were transferred to an 8-well chamber slide (Nunc) coated with gelatin and cultured for 3 days in the corresponding media. Cultures were stopped and fixed with Z-fix, then immunostained initially with the rabbit anti-hSOX9 polyclonal antibody (1:1500, AB5535, EMD Millipore) and the rabbit anti-hGDF5 polyclonal antibody (1:100, PA5-29775, Invitrogen), then with biotinylated donkey anti-rabbit IgG (711-065-152, Jackson ImmunoResearch), and finally with AF594-conjugated streptavidin (S32356, Invitrogen). The stained slides were inspected with a BX61 fluorescence microscope (Olympus) 1 day after being mounted with the DAPI-containing mounting solution, ProLong Gold anti-fade mounting media (Molecular Probes/Invitrogen). The percentage of SOX9+ nuclei/DAPI+ nuclei was calculated as the average ratio between the number of SOX9hi, SOX9lo or SOX9− nuclei and that of DAPI+ nuclei counted from three or four different areas per slide and multiplied by 100.
Immunofluorescence staining of cartilage pellets
The sections of in vitro-derived, paraffin-embedded cartilage pellets were deparaffinized with xylene, rehydrated, heat-treated in antigen-retrieval solution (Dako), blocked with the blocking buffer (Dako), and subjected to immunofluorescence detection of COL2 (for detecting chondrocytes) and COL1 (for detecting mesenchymal cells or osteoblasts) or COL10 (for detecting hypertrophic chondrocytes). Primary antibodies were: rabbit anti-COL1 antibody (1:50, NB600-408, Novus Biologicals), biotinylated goat anti-COL2 antibody (1:100, NBP1-26546, Novus Biologicals), rabbit anti-COL10 antibody (1:100, AB58632, Abcam) and rabbit anti-PRG4 antibody (1:100, HPA028523, Sigma-Aldrich). Secondary reagents were goat anti-rabbit IgG-Alexa Fluor 488 (A11034) and streptavidin-Alexa Fluor 594 (S11227) from Molecular Probes. The slides were washed in PBS and mounted with ProLong Gold anti-fade mounting media.
Statistical difference between groups was calculated by Student's t-test (two categories, two-tailed) or one-way ANOVA (for more than two categories) followed by the Student–Newman–Keuls multiple comparison by Excel (Microsoft) and KaleidaGraph (Synergy) software. n=number of independent experiments (i.e. biological replicates). All quantitative data presented in a graph are mean values of independently generated samples of three or more batches (n≥3 in general), with s.e.m. shown by thin error bars unless stated otherwise. P<0.05 was considered to be statistically significant and marked with an asterisk.
We would like to acknowledge A. Hazen and A. Blancas for cell sorting and B. Burke and E. Creissen for FACS analysis using Cytek Aurora; Z. Mao and M. Starbuck for histological analyses; Z. Mao for immunofluorescence analyses; C. Yang for RNA-seq library construction and sequencing; and S. Easton for additional hiPSC experiments during the revision of the manuscript.
Conceptualization: N.N.; Methodology: A.P., B.E.S., B.K.A., N.M., G.P., Q.Y., Z.Z., N.N.; Software: G.P., Z.Z.; Validation: A.P., B.E.S., B.K.A., N.M., G.P., Z.Z., N.N.; Formal analysis: A.P., B.E.S., B.K.A., N.M., G.P., N.N.; Investigation: A.P., B.E.S., B.K.A., N.M., G.P., Q.Y., N.N.; Resources: B.R.D., Z.Z., J.H., N.N.; Data curation: B.E.S., B.K.A., N.M., G.P., N.N.; Writing - original draft: N.N.; Writing - review & editing: G.P., B.R.D., J.H., Z.Z., N.N.; Supervision: Z.Z., N.N.; Funding acquisition: B.R.D., J.H., Z.Z., N.N.
This work was supported by the Annie and Bob Graham Distinguished Chair in Stem Cell Biology (N.N.), the Cancer Prevention and Research Institute of Texas (RP180734 to Z.Z.), the National Institutes of Health (R01AR077045 to N.N., R21AR079075 to N.N. and J.H., R01LM012806 to Z.Z., R01HL139876 to B.R.D., R21AR072870, R21AR073509, R21AR075997 and R21AR074132 to J.H.), and the U.S. Department of Defense (N00014-18-RFI-0014 to J.H.). Deposited in PMC for release after 12 months.
RNA-seq data have been deposited in Gene Expression Omnibus under accession number GSE145048.
The authors declare no competing or financial interests.