Pluripotent stem cells for skeletal tissue engineering

Abstract Here, we review the use of human pluripotent stem cells for skeletal tissue engineering. A number of approaches have been used for generating cartilage and bone from both human embryonic stem cells and induced pluripotent stem cells. These range from protocols relying on intrinsic cell interactions and signals from co-cultured cells to those attempting to recapitulate the series of steps occurring during mammalian skeletal development. The importance of generating authentic tissues rather than just differentiated cells is emphasized and enabling technologies for doing this are reported. We also review the different methods for characterization of skeletal cells and constructs at the tissue and single-cell level, and indicate newer resources not yet fully utilized in this field. There have been many challenges in this research area but the technologies to overcome these are beginning to appear, often adopted from related fields. This makes it more likely that cost-effective and efficacious human pluripotent stem cell-engineered constructs may become available for skeletal repair in the near future.


Introduction
Life means danger: accidents, injury, and disease are inevitable. Unfortunately, human beings do not regenerate such damaged tissues or organs as effectively as lower vertebrates. Amphibians and fish have greater regenerative capacity than mammals and can fully remake limbs, tails, and fins when damaged or lost [1,2]. Although our bones and livers will regenerate to a reasonable extent [3], most skeletal tissues such as articular cartilage instead resort to fibrosis and scar tissue formation in place of the original structure.
However, human scientific ingenuity has resulted in the development of novel tissue engineering (TE) approaches to aid correct cell replacement and generate authentic tissues and organs when our bodies fail to repair on their own. TE is the science of using external materials or constructs to stimulate the innate ability of cells to generate lost or damaged tissue of the right kind in situ, or to use multidisciplinary technologies to create those tissues from cells outside the body [4]. In both cases, the tools available to us are improving rapidly. However, the choice of cell source for TE is of critical importance for success. For skeletal applications, cells may be required to survive in an artificial construct introduced into an inflamed environment. At the same time, they must be able to respond to positive cues determining that they remain as, or develop into, the required cell type. Cells need to be able to respond to biomechanical cues to produce the structure and function of the original tissue or organ, which means they must interact with other cells and migrate appropriately, or avoid migrating, as their destiny dictates. Cells that are plastic and amenable to such cues are likely to be committed progenitors. So where do we obtain such cells? Adult mesenchymal stromal cells (MSCs) have been seen as an attractive source of cells for skeletal tissues. They can be extracted from the bone marrow or peripheral blood, and have multi-lineage differentiation ability into osteogenic, chondrogenic or adipogenic cell types. However, these cells have disadvantages in their reduced differentiation potential on expansion and lack of long-term retention in vivo [5][6][7] combined with the need for several operations for autologous transplantation and challenges in using them allogeneically, which may hinder their widespread clinical application. Notably, they also play immunomodulatory roles. An alternative and attractive cell source for TE is human pluripotent stem cells (hPSCs). These are either human embryonic stem cells (hESCs), derived from the inner cell mass of the human blastocyst surplus to assisted reproduction programs, or human induced pluripotent stem cells (hiPSCs), which are reprogrammed somatic cells. They combine a unique ability to generate virtually any cell type in the body (i.e., pluripotency) with continuous replication without differentiation (i.e., self-renewal) and are becoming increasingly relevant to modern medicine. Importantly, there are now thousands of lines worldwide including some derived and banked at clinical grade [8,9] and therefore usable for cell therapies. With increasingly robust differentiation protocols available, hPSCs hold great promise as biological starting materials for a multitude of applications including regenerative medicine, disease modeling and drug development. Herein we review the available protocols for hPSC skeletogenesis with a focus on cartilage and bone. We discuss some of the most recent advances in enabling technologies and how their integration with hPSC differentiation will eventually lead to a new generation of bioengineered tissues for skeletal regeneration.

Routes for hPSC differentiation into skeletal tissues
Methods to generate skeletal tissues from hPSCs primarily follow one of three routes: (1) Initially nonspecific differentiation, in which the minimal signals required to drive skeletogenesis are applied in a simple and cost-efficient manner; (2) Generation of MSClike cells, serving as an alternative to adult MSCs; (3) Developmentally-guided differentiation, which recapitulates multiple steps of embryonic development in vitro with the aim to generate authentic skeletal cell types of specific regions of the body. These methods are discussed below, and their advantages and limitations are summarized in Table 1.

Nonspecific initial differentiation
Nonspecific initial differentiation protocols for hPSC skeletogenesis follow one of four strategies: (A) Monolayer differentiation; (B) Formation of embryoid bodies; (C) hPSC micromass cultures, in each case followed by chondro-or osteo-genic differentiation or; (D) hPSC differentiation through co-culture with adult target cell types ( Figure 1).
Nonspecific differentiation as a monolayer ( Figure  1(A)) is a simple strategy that involves the treatment of hPSC monolayers with a differentiation medium in order to generate a population of differentiated cells in a single step, without sequentially driving multiple stages of development. Promising hPSC differentiation protocols using this method are scarce, but in recent years protocols for osteogenesis have been reported [10,11]. One method involves the direct application of osteogenic medium [Minimum Essential Medium Eagle alpha modification, fetal bovine serum (FBS), nonessential amino acids, b-mercaptoethanol, ascorbic acid, sodium glycerophosphate and dexamethasone (DEX)] for 35 days. Despite enhanced expression of osteogenic markers, this strategy resulted in a highly heterogeneous population of cells, and low differentiation efficiency, evidencing the need for more specific methods [10]. A different strategy uses osteogenic medium supplemented with retinoic acid (RA), resulting in the formation of calcified osteogenic nodules, from which osteogenic cells can be isolated. The use of RA was shown to stimulate BMP and WNT signaling, allowing cells to differentiate further when compared to osteogenic medium alone, and triggering the expression of osteocyte markers such as PHEX and sclerostin [11].
The use of embryoid bodies is a common strategy to promote hPSC differentiation into cells equivalent to the ectoderm, mesoderm, and endodermthe three  Figure 1. Strategies for nonspecific differentiation of hPSCs. (A) Nonspecific differentiation as a monolayer through treatment of PSCs with differentiation medium is an uncommon strategy but has been applied to osteogenic differentiation [10,11]. (B) Embryoid body (EB) formation enables spontaneous differentiation of the three embryonic germ layers. Continuous culture of EBs within chondrogenic differentiation medium supplemented with GFs may lead to generation of cartilaginous cell types [12,13]. Alternatively, EB culture with osteogenic medium leads to generation of mineralizing osteoblastic-like cells [14].
(C) Micromass culture involves condensation of PSCs within a small volume of chondrogenic growth medium in the presence of GFs. Chondrogenic-like cells then can generate cartilaginous tissue within scaffold-free pellet culture [15]. (D) Co-culture systems require the presence of adult primary cells to influence PSC differentiation through paracrine cues. PSCs can be separated from primary cells via a porous membrane that enables the passage of chemical cues (top) or mixed within micromass cultures micromass cultures (bottom), which requires a selection method to remove primary cells post-differentiation [16][17][18]. Days noted represent period of time spent in appropriate culture as illustrated.
embryonic germ layers. After embryoid body formation, different levels of control over differentiation of the mesoderm toward skeletal lineages can be exerted. The simplest methods involve treatment with chondrogenic, or osteogenic medium, which can be supplemented with growth factors (GFs), particularly those of the TGFb superfamily for chondrogenesis, or ascorbic acid, b-glycerophosphate, and DEX for osteogenesis (Figure 1(B)) [12,13]. Multi-stage differentiation protocols using embryoid bodies, but aiming to recapitulate different stages of development, or including a mesenchymal stem cell-like stage, have also been established (Sections "hPSC-derived MSC-like Cells" and "Developmentally-guided Differentiation").
Similarly to embryoid bodies, differentiation can be promoted in hPSC micromass cultures by treatment with chondrogenic GFs such as BMP-4, which was shown to result in the generation of cells expressing collagen type-II, a key cartilage ECM component ( Figure  1(C)) [15]. However, the efficiency of these systems is often low, with one study reporting approximately 10% of cells in the micromass expressing COL2A1 promoterdriven GFP. Thus, cell sorting was required to obtain appropriate cells for further chondrogenic differentiation, first as a monolayer treated with FBS, and then as pellet cultures exposed to TFGb-3 [15].
A third method of differentiation is the use of co-culture systems, in which adult primary cells, or cell lines, promote differentiation of hPSCs into the cell type of interest. These systems have been developed using chondrocytes [16,18], or limb bud progenitor cells from mouse embryos [17]. In order to ensure that the resulting cell population is homogeneous and consists of hPSC-derived progeny, the cells used as a co-culture can be separated in the culture well by a porous membrane ( Figure 1(D top)) [16], or cell division of the cocultured cells can be inactivated through gamma irradiation [18,19]. Alternatively, hPSCs can be transfected with genes for antibiotic resistance and selected after exposure to the antibiotic to purify the final hPSC-generated population (Figure 1(D bottom)) [17].
Despite their ease of use, nonspecific differentiation methods often have low efficiencies, and the mechanisms guiding differentiation are less defined. Co-culture systems rely on the availability and homogeneity of differentiated cell types, which likely promote differentiation through the release of paracrine factors or cell-cell signaling, but may be affected by batch-tobatch or donor variability. However, these systems can still be valuable due to their simplicity and used to further interrogate the signals needed for hPSC differentiation. The identification of the paracrine factors released by cells in co-culture and their mode of action may lead to the generation of new methods to differentiate hPSCs into skeletal lineages, by replacing the coculture with a medium containing key GFs identified, or small molecules.

HPSC-derived MSC-like cells
The use of MSCs is a widely adopted strategy for skeletal TE. HPSCs can serve as a cell source for the generation of MSC-like cells, often called induced mesenchymal stem cells (iMSCs) [20]. Despite their origin, iMSCs meet some of the minimal criteria for MSC classification as defined by the International Society for Cellular Therapy [21] although adipogenic differentiation is not a requirement for cells to give skeletal lineages.
The generation of iMSCs is one of the simplest and most widely used methods for the generation of cartilage and bone from hPSCs ( Figure 2). Through a sequential multistage process, hPSCS are initially differentiated either through embryoid bodies [22,23], or monolayer culture. Further culture with medium containing FBS leads to the formation of MSC-like cells. Finally, chondro-or osteogenic differentiation is induced in a similar manner to that for adult MSCs. Chondrogenic differentiation can be achieved through the use of commercially available chondrocyte differentiation medium [24], or through supplementation with chondrogenic GFs such as TGFb-3 [25], TGFb-1 [22], or BMP-2 [26]. Osteogenesis can be promoted by the use of commercially available osteogenic medium [27], or media supplemented with dexamethasone, b-glycerophosphate, and ascorbic acid-2-phosphate [22,28,29]. Indeed, although less investigated, tenocytes for tendon repair are also obtainable through stimulation with FGF2 and TGFb [20,30].
When compared to adult MSCs, the generation of iMSCs requires an additional mesenchymal differentiation step. However, their use can offer multiple advantages. Unlike adult MSCs, hPSCs can self-renew in vitro for multiple passages, overcoming supply limitations [31,32]. In addition, hiPSC generation from peripheral blood [33] or skin cells [34] can enable the generation of high iMSC cell numbers, with reduced donor site morbidity and variability [35]. When considering the use of iMSCs for chondrogenesis, it is important to note that the resulting cell type is likely representative of a prehypertrophic growth plate chondrocyte, characterized by expression of markers such as SOX9, IHH, RUNX2/3, collagen type II, and type X [26,36]. Growth plate chondrocytes are known to undergo hypertrophy and participate in the process of bone formation by endochondral ossification, being replaced by osteoblasts but also contributing directly to osteogenesis [37][38][39]. However, this means that iMSCs' use for articular cartilage (AC) applications may be more limited.

Developmentally-guided differentiation
In vivo, skeletal tissues can have neuroectodermal or mesodermal origin, depending on their anatomic location ( Figure 3(A)). It is important to understand how these tissues develop, and to be aware of the signals needed for cell lineage commitment, in order to inform selection of the best differentiation protocol. The craniofacial skeleton is mainly an ectodermal neural crest derivative, but most bone and cartilaginous tissues in the body originate from the mesoderm. The remaining axial skeleton originates predominantly from the paraxial mesoderm through the somites, whereas the lateral plate mesoderm (LPM) gives rise to the cartilage, tendons, and bone of the appendicular skeleton [49]. Within the fetal limb, different types of cartilage develop: growth plate cartilage, which serves as a template for bone development, and AC, which lines the ends of long bones at the joints.
Most developmentally-guided hPSC differentiation methods use combinations of GFs and small molecules to stimulate or inhibit cell-signaling pathways that play a crucial role in development. The most widely used stimulatory/inhibitory factors are summarized in Table 2 and can be used by tissue engineers as a toolbox to guide hPSC differentiation into skeletal lineages. By using these factors, developmentally-guided hPSC differentiation protocols have been developed, following the three main routes: neural crest, paraxial mesoderm, and lateral plate mesoderm. The most important signals needed to promote hPSC differentiation toward these lineages are outlined in Figure 3(B), and can serve as a guide toward the development of improved methods of differentiation.
Ectoderm: Neural crest Gastrulation specifies pluripotent cells from the inner cell mass into the three germ layers: the ectoderm, mesoderm, and endoderm. The first step in gastrulation to key papers that illustrate derivation or identification of specific skeletal lineages from PSCs. Activation or inhibition of cell signaling pathways required to drive differentiation of each lineage outlined as described in key references [40][41][42][43][44][45][46][47][48].
is the formation of the primitive streak, a transient structure through which mesoderm and endodermal precursors are internalized under the prospective ectoderm [66,67]. At this stage, pluripotent cells commit to differentiation toward ectoderm or mesendodermal fates.
The neurectoderm gives rise to the neural crest, which generates skeletal components of the head and neck ( Figure 3(A)) [68]. HPSCs can be differentiated into skeletal tissue types following the neural crest and ectomesenchymal cell stages. One method to generate neural crest precursor cells is through their isolation from neural rosettes followed by FACS sorting (precursors HNK1 þ p75 þ ) [69]. A different strategy uses WNT stimulation through GSK3 beta inhibition, and inhibition of BMP and Activin/Nodal [70][71][72][73]. Neural crest cells can then be differentiated into mesenchymal cells through culture in an FBS-rich medium or in a neural crest induction medium. Finally, mesenchymal cells can be further differentiated using chondrogenic medium supplemented with TGFb-3 or osteogenic media [69,70]. Alternatively, neural crest cells can be directly treated with osteogenic medium supplemented with BMP-2, FGF9, rapamycin, and WNT3a followed by osteogenic medium alone for generation of osteoprogenitor cells without the need for an additional differentiation step [73].

Mesoderm
During gastrulation, mesendodermal precursors formed in the primitive streak are able to generate both mesodermal and endodermal progenitors [74]. This process, replicated in vitro, forms the basis for differentiation protocols that follow a two or three-step strategy, differentiating hPSCs first through a primitive streak-like stage via epithelial-mesenchymal transition (EMT), with formation of mesendoderm and mesoderm and then toward a chondrogenic phenotype [55,59,62]. In vivo studies of embryonic development have uncovered the crucial importance of TGFb, WNT, FGF, and BMP signaling in primitive streak induction [75][76][77][78]. Strategies focused on in vitro differentiation of hPSCs into primitive streak-like cells often employ GFs or small molecules that stimulate these pathways.
Pioneering work has promoted mesendoderm induction using combinations of WNT, Activin, and BMP stimulation [79]. Then, a mesoderm phenotype was achieved through the use of BMP-4 (and in later work BMP-2), while follistatin was employed to inhibit endoderm formation. Finally, chondrogenesis was promoted using a switch from BMP-4/2 to GDF-5 [55,59]. More recently, a simpler method was developed, using only CHIR99021, a WNT activator, and TTNPB, a retinoic acid receptor pan-agonist. The resulting cells expressed key chondrogenic markers, both in vitro and in vivo after subcutaneous implantation in mice [62]. The exact mechanism of action of TTNPB remains unclear, but it may act as an epigenetic modulator.
As different regions of the mesoderm can generate skeletal tissues of distinct regions of the body, acquisition of a particular mesoderm phenotype is important. The specification of a paraxial vs. lateral plate phenotype is heavily reliant on a balance between WNT and BMP signaling. WNT stimulation promotes paraxial mesoderm whereas BMP leads to lateral plate mesoderm. Conversely, inhibition of either of these signals can induce the generation of the opposite mesodermal fate [43,73].

Paraxial mesoderm
After paraxial mesoderm induction, hPSC-derived cells can be directly differentiated into osteoprogenitor cells All-trans RA, TTNPB AGN193109 [24,48,62] through treatment with osteogenic medium [73]. Alternatively, paraxial mesoderm-like cells can be differentiated into an early somite phenotype through WNT stimulation combined with inhibition of FGF, TGFb, and BMP signaling. Then, WNT and hedgehog signaling drive the bifurcation between ventral (sclerotome) and dorsal (dermomyotome) portions of the somites. By stimulating WNT and inhibiting hedgehog signaling, sclerotome-like cells can be produced, which differentiate into skeletal fates including osteo-and chondroprogenitors. Sclerotome progenitor implantation in vivo leads to ectopic bone formation whereas stimulation with BMP can generate a pool of highly chondrogenic cells [43]. Additional treatment with WNT inhibitor C59 has been shown to further enhance chondrogenesis using this system [80]. However, despite the development of highly efficient, multi-step differentiation protocols, generation of homogeneous cell populations is still challenging. Single-cell transcriptomic analysis has revealed that this protocol still generates some off-target cell types such as neural cells and melanocytes during such paraxial chondrogenic differentiation [80].

Lateral plate mesoderm (LPM)
Efforts to map LPM development using hPSCs were pioneered by Loh and colleagues, with the identification of WNT stimulation as key to induce a limb bud-like phenotype, with increased expression of PRRX1 and HOXB5, whereas WNT inhibition resulted in cardiac differentiation [43]. When WNT stimulation is combined with sonic hedgehog activation, PSC-derived LPM cells can be differentiated into tracheal cartilage and smooth muscle cells [42]. However, when RA is applied under WNT inhibition, different splanchnic lateral mesoderm progeny can be formed while inhibiting the formation of anterior LPM-derived cardiomyocytes [41]. In vitro, RA signaling modulation has been used to specify forelimb vs. hindlimb phenotype using mouse PSC aggregates in suspension culture. Exposure to RA resulted in upregulation of forelimb marker Tbx5, whereas the use of a RA antagonist resulted in increased expression of hindlimb-specific Tbx4 and Pitx1 [48]. Despite significant progress in the field, the full range of necessary signals for the generation of skeletal tissues from LPM and limb-bud intermediates using hPSCs is still unclear and single-cell analysis has not been published. However, recent work has shown the generation of osteoprogenitor cells from an LPM-like intermediate through treatment with osteogenic medium [73], and chondro-and osteogenic cells from a limb bud-like intermediate through the activation of BMP and TGFb, or WNT signaling respectively [47].

Enabling technologies
The stem cell engineering field is rapidly evolving, thanks to a clearer understanding of the temporal dynamics of stimulation and inhibition of signaling pathways that enable the generation of desired cell types. However, current methods for hPSC skeletal differentiation are still heavily reliant on the use of GFs, which are costly, suffer from batch to batch variation and have short shelf lives and half-lives in vitro.
Protocols are frequently conducted in 2D, or in simple 3D models such as embryoid bodies or cell spheroids of limited size. In contrast, technological innovations have emerged in closely related fields such as biomedical engineering, nanotechnology, and synthetic biology. Even though some of these have found applications in the field of TE, their potential has not yet been fully explored in the generation of skeletal tissues from hPSCs. The opportunities they offer for improving methods of generation of cartilage and bone from hPSCs are summarized in Figure 4.

Cell selection
Spontaneous differentiation of hPSCs and poor efficiency of differentiation protocols can lead to unpredictable and heterogeneous off-target populations of non-chondrogenic or osteogenic cells. Skeletal progenitors are cells capable of selectively giving rise to bone, cartilage, and stromal cells. Efforts to identify them have been made in recent years [81][82][83][84]. However, a gold standard combination of markers has not been definitively agreed upon. Researchers have mainly investigated such markers in rodents [82,83,85] but occasionally human fetal limb [86] or adult joints [87] have been the source. A list of potential markers for the identification of overall skeletal progenitors as well as osteogenic and chondrogenic progenitors is reported in Table 3. The need for combinations of markers is very clear. One goal has been to use selection methods to isolate such progenitors for further differentiation thus improving the purity level of mature skeletal populations. The selection process has to retain cell viability and progenitor characteristics in order to continue to allow healthy culture and further differentiation after sorting. The latter has to avoid damage to the cells precluding the use of intracellular markers which would require fixation and permeabilisation. However, intracellular markers are still exploitable if gene-reporter lines can be engineered without interfering with cell phenotype. Currently, Fluorescent-Activated Cell Sorting (FACS), the gentler Magnetic-Activated Cell Sorting (MACS) and Microfluidic cell sorting technologies (MST) are the main techniques adopted for cell selection (Table 4). In FACS, cells are bound to fluorescent antibodies to particular lineage-selective markers (or engineered with fluorescent lineage-selective promoterreporters) and sorted according to label intensity with gating selected by the operator [92]. MACS still requires the use of antibodies, but uses magnetic particle-bound antibodies. Cells positive for the selected antigen bind to the antibody and through the use of a strong magnet are retained, whereas negative cells can be washed away. Alternatively to FACS and MACS, in MST, antibodies are not required, and the label-free cells are sorted based solely on size and morphology. This was Control of cell signaling mechanisms may enable more precise direction of hPSC differentiation in 2D or 3D systems. Optogenetic technologies enable rapid, reversible and spatiotemporal control of cell behavior and physiology with light stimulation. Release of GFs through controlled delivery systems can improve GF half-life, prevent premature degradation and specify spatiotemporal signals within 3D constructs. (C) Cell characterization is critical to analyze differentiation efficiency and ensure the presence of desired cell types. Characterization methods can also be used to identify markers that can subsequently be used to enrich specific populations through cell selection techniques.
illustrated in a recent publication, where undifferentiated MSCs were sorted through the use of a spiral microfluidic chip into three different subpopulations according to size [93]. Employing gene expression and staining techniques, the authors were able to identify medium and large size cells as being more chondrogenic (pre and post differentiation) and osteogenic, respectively. Single-cell RNAseq analysis may confirm and extend such marker selections [80,91].

Biofabrication strategies
Biofabrication is the automated generation of living functional products [94] and comprises top-down or bottom-up strategies. Top-down strategies are characterized by the fabrication of temporary structures (i.e., scaffolds) to support and guide tissue formation by seeded cells. Bottom-up approaches require the use of cells as building blocks (i.e., spheroids, sheets, cell-laden hydrogels) to generate tissue constructs with high degree of complexity through automated assembly processes. Top-down strategies such as the use of nanofibrous scaffolds have been shown to improve osteogenic differentiation from iPSC-derived embryoid bodies [95,96] and iMSCs [97]. Similarly, culture of iMSCs within decellularized bone scaffolds has resulted in improved osteogenesis in vitro and in vivo, with enhanced tissue maturation after implantation [98]. Bottom-up approaches such as 3D bioprinting have also been used for hPSC skeletogenesis. Nanofibrillated Table 3. Surface markers for the identification of skeletal progenitor cells.

Markers
Differentiating fate Cell source Additional information Reference

Osteogenic, Chondrogenic, Adipogenic
Minimal criteria for defining MSCs [21] CD166 low/neg CD146 low/neg Mouse Limb Mouse skeletal stem cells. Potential to form cartilage, bone and stroma. [83] Mouse Limb Mouse osteoprogenitors, chondroprogenitors, stromal progenitors Only osteogenic differentiation in vitro and in vivo. [91] cellulose (NFC)-based hydrogels were developed as bioinks to promote the chondrogenic differentiation of encapsulated iPSCs. Comparison between different bioinks suggested increased levels of cell proliferation when NFC was blended with alginate compared to blending with hyaluronan, with cells retaining the expression of chondrogenic markers in 3D culture [18].
Despite these examples, the use of biomaterials and biofabrication technologies for skeletal TE applications using hPSCs has been limited thus far, and their full potential is yet to be explored. In order to achieve control over PSC differentiation in biofabricated constructs, stringent design and manufacturing requirements need to be considered and can be found elsewhere [99]. However, 3D bioprinting can offer multiple advantages, such as: (1) Control over macroarchitecture, allowing the fabrication of patient-specific constructs. (2) Control of microarchitecture, regarding pore size, shape and interconnectivity, and ability to harness the extrusion process to orient microfibers within a bioink [100]. (3) Integration of multiple materials within a single construct, providing the ability to mechanically reinforce weak hydrogels with stiffer materials such as thermoplastics [101]. (4) Ability to integrate multiple cell types and generate heterogeneous tissues, such as the osteochondral interface [102]. Due to these advantages, the integration of biofabrication technologies with hPSC differentiation methods can offer innovative ways to control cell differentiation, reproduce cell niches, and maintain cells with a stable phenotype to generate three-dimensional skeletal tissue constructs.

Advanced bioreactors
During development, both biomolecular cues and biomechanical forces are instrumental in influencing skeletal cell fate and differentiation. Cell response to load through integrins, mechanoreceptors, and primary cilia are critical for endochondral skeleton development [103,104] and articular cartilage homeostasis [105,106]. In 3D cartilage constructs and stem cell protocols, moderate cyclic hydrostatic pressure (around 3-10 MPa and 1 Hz) has been shown to increase collagen II and GAG deposition and the chondrogenic phenotype in newly forming cartilage [107][108][109]. While bone is mainly exposed to cyclic compression, articular cartilage also has to withstand shear stresses in a dynamic manner. The use of advanced bioreactors offers the ability to replicate these conditions in vitro through the application of multiple physical stimuli including compression, shear and hydrostatic pressure (HP). These stimuli can be applied through the solid (e.g., scaffold) or liquid phases (e.g., cell culture medium), either separately or in combination. Compression and shear stresses are generally deployed on the solid construct by direct mechanical loading, whilst HP and fluid shear can be applied to cells by tuning the flow parameters of the liquid medium (i.e., velocity, pressure, etc.).
Application of mechanical stimuli on hPSC skeletal differentiation has shown promising results in guiding cell fate. For cartilage TE applications, the use of cyclic compression can limit hypertrophic differentiation of iMSCs embedded in 3D hydrogels, reducing collagen type-X [54]. A distinct application of bioreactors is its miniaturization with the integration of microfluidics and organ-on-a-chip technologies. Microfluidic bioreactors enable cell culture with smaller medium volumes, reducing the high cost of supplements for hPSC differentiation. High-throughput systems can make it easier to evaluate the effects of different concentrations or combinations of biochemical compounds [110]. In addition, the ability to perform mechanical stimulation within microfluidic systems has made it possible to study the response of chondrocytes at the onset of osteoarthritis to compression [111]. Harnessing the full potential of bioreactor technologies by integrating them with hPSC differentiation systems may lead to new ways of guiding cell differentiation toward specific developmental lineages, and to a greater understanding of mechanotransduction processes.

Characterization methods
Classically, in vitro characterization of skeletogenesis in scaffold-based or scaffold-free systems has involved the use of histochemical stains such as Safranin O or Alcian/Toluidine blue for sulfated GAGs, Picrosirius red for collagen fibers and Alizarin red for mineralized tissue [112], together with immunocytochemistry of fixed tissue for key proteins. Picrosirius red staining is best viewed under polarizing light microscopy where the birefringence reflects collagen bundle alignment. These are still the cornerstone of the characterization repertoire but recently many more techniques have emerged. It is now standard to assess gene expression using quantitative reverse transcription PCR (RT-qPCR) for transcription factor and matrix molecule genes expressed during chondrocyte development [43,55,113] or in mature chondrocytes or osteoblasts [114,115]. This can be extended to assessment of the whole transcriptome through RNAseq [116], e.g., comparison with developing limb tissue [88], osteogenesis compared to calvaria [115] and evaluation of chondrogenesis-associated microRNAs [116,117]. More recently it has been possible to delve into the nature of the different cells developing to form scaffold-free cartilage using singlecell RNAseq [80]. This allows the identification of off-target cells that would be detrimental to the generation of homogeneous engineered tissue. It also allows detailed cell evaluation in comparison to native tissues in which a number of cell types are present, such as in bone.
Mass spectrometry (MS)-based methods allow for evaluation of bimolecular sample composition, making it possible to conduct proteomic or lipidomic studies. The use of proteomics has been applied to analyze the composition of cartilage [118] and bone [119], as well as their changes with disease such as in osteoarthritis [120,121]. The main advantage of these techniques over most methods is the untargeted analysis, and the ability to identify post-translational modifications such as phosphorylation, important in the regulation of protein function [122]. MS-based proteomics has been applied to study bone and cartilage, both at the tissue and cellular levels. Using this technology it is possible to identify both intracellular and extracellular proteins such as collagens, proteoglycans, osteocalcin and BMPs [123]. The first proteomic map of human articular chondrocytes in culture resulted in the identification of key proteins with roles in cellular organization, metabolic activity, ECM production and remodeling [118]. Studies with hESCs have revealed important insights about how they change during differentiation [124]. LC-MS/ MS has also been applied to characterize the secretome of hESC-derived iMSCs, allowing the identification of 247 proteins differentially present in conditioned medium, which might provide valuable insight about their paracrine effects [125]. Recent developments in mass spectrometry have enabled application in the analysis of tissue sections with spatial resolution. This technology made it possible to analyze the protein or lipid distribution without the need for targeted approaches such as the use of antibodies [126,127]. Despite limited applications in skeletal TE [127], mass spectrometry imaging may become a powerful technology for sample analysis, with particular benefits when considering interfacial tissues such as the osteochondral interface.
Mechanical characterization tools provide vital information on the physical properties of TE constructs regulating skeletogenesis, and determine their suitability to support tissue growth under physiological loads. Therefore, characterizing the mechanical properties of skeletal TE implants is an important parameter in their development. Mechanical characterization typically involves the use of static mechanical force testers to measure the compressive or tensile properties of the scaffold (e.g., Young's modulus, tensile strength, compressive modulus or compressive strength) [96,[128][129][130][131]. Compressive modulus is frequently used as an indirect method of measuring the scaffold degradation rate, by measuring repeatedly at time intervals [128,129,132]. While cartilage applications typically focus on compression testing, tensile testing is an important parameter for bone constructs in order to prevent bone fracture. Rheological measurements (i.e., viscosity, recovery, storage, and loss modulus, etc.) are also very common, especially for cartilage applications where polymeric hydrogels are increasingly used to mimic the viscoelastic nature of the native tissue [133,134].

In silico models
Despite the importance of in vitro and in vivo studies to assess the performance of hPSC-derived cells for skeletogenesis, the use of computational in silico models has been gaining relevance as a tool to predict cell behavior and quantitatively simulate experimental scenarios that might lead to an accelerated route to market with increased speed and reduced costs.
In a two-part study, Campbell and coworkers propose a reaction-diffusion mathematical model to investigate the effect of GFs (BMP-2 and FGF-1) and coimplantation (i.e., autologous chondrocytes and MSCs) on cartilage regeneration after cell therapy [135,136]. The simulation results showed differences in matrix production, distinct effects of the different GFs, and suggested an optimal cell ratio for co-implantation of autologous chondrocytes and MSCs in order to accelerate cartilage repair and lead to higher matrix densities.
Employing a mathematical model based on ordinary differential equations, Gaspari et al. investigated the complex paracrine mechanisms associated with mesendodermal differentiation of hPSCs. The authors suggested that at least three paracrine factors, LEFTY1, CER1, and an undefined activator must be included In the simulation to accurately replicate the in vitro differentiation kinetics of hPSCs [137].
A different model [138] recapitulated in vitro stem cell morphogenesis in silico and demonstrated the possibility of controlling multicellular patterning toward the generation of human organoids and tissues. In an attempt to overcome some of the limitations associated with the in vitro culture of hPSCs and to scale-up their bioprocessing, Manstein and colleagues have combined instrumented stirred tank bioreactor technology with in silico modeling [139]. The authors demonstrate that a 70-fold cell expansion in 7 days (independently of the cell line) is achievable while pluripotency, differentiation potential and karyotype remain unaffected. The proposed strategy also allows for significant economic savings (over 75% reduction in medium consumption) which, in combination with the large cell volumes yielded, represent an important step toward the mass production of hPSCs for clinical implantation. Taken together, the above studies demonstrate how in silico modeling can be used to inform hPSC technology and eventually drive the future of regenerative medicine therapies. However, this approach has yet to be applied to differentiation of hPSCs toward skeletal lineages or the application of chondro-or osteoprogenitors in tissue repair.

Stimulation of growth factor signalling
Growth factors are critical components of hPSC-skeletal engineering strategies due to their key role in directing cell fate. However, the clinical application of GFs is limited by their short half-life and rapid denaturation in vivo, which leads to poor local retention and consequent supraphysiological dose requirements [140]. The dosage-related adverse effects of GFs used for bone or cartilage repair have included osteoclast activation [141], ectopic bone formation [142], adipogenesis [142], fibrosis and hypertrophic scars [143]. When using scaffolds for tissue repair, the controlled and sustained delivery of GFs to cells within any 3D construct is crucial. This has driven the development of controlledrelease biomaterial-delivery systems designed to retain active GFs for longer and allow specific GF targeting while avoiding the adverse effects of inappropriate concentrations.
Skeletogenic delivery systems have incorporated different GFs namely BMP2 [144], BMP7 [145], TGFb-3 [128,134,145], VEGF, and PDGF [146] to promote cartilage or bone regeneration and the vascularization of the latter. GF release profiles and encapsulation efficiency are most commonly measured using an Enzyme-Linked Immunosorbent Assay (ELISA) [128,134,144,145,147]. Direct physical encapsulation of the GF within the scaffold matrix is the simplest delivery strategy. In such systems release kinetics are determined by the biodegradation of hydrophobic polymers (e.g., poly-lactide-co-glycolide (PLGA) [148], polycaprolactone (PCL) [149]), or diffusion of GF through porous hydrogel networks such as chitosan [150], and alginate [151]. The latter is used frequently, but limited by the high porosity and hydrophilicity of the hydrogel matrix, resulting in poor retention of the GF and early burst release which leads to, e.g., chondrocyte hypertrophy [152].
Triggered delivery mechanisms via electro-responsive BMP-2 release and TGFb immobilized magnetic beads have been developed for greater spatiotemporal control over the release profile [144,153]. Nanoparticle tethering and microsphere encapsulation are increasingly used to obviate burst release kinetics by acting as secondary carriers [128,134,[145][146][147][148]154]. Indeed, Zhou et al. reported <0.35% release of TGFb-3 from graphene oxide nanoflakes after 72 h [134]. Dual delivery systems have also been developed to better mimic the stimulation of parallel complex signaling pathways which drive the development of skeletal tissues in vivo [145,146]. Currently, the spatiotemporal delivery of GFs to encapsulated cells within 3D constructs has yet to be explored with pluripotent cell types.

Optogenetic technologies
Although tissue morphogenesis is a dynamic process, in vitro recapitulation of developmental processes during hPSC differentiation is limited by the intrinsic properties of stimulation by GFs (Section "Stimulation of growth factor signalling"). Addition of exogenous factors, including GFs and small molecule agonists, cannot accurately reflect the dynamic signaling cues that cells are subject to in vivoboth in time and space. The emergence of optogenetic approaches provides a means to develop synthetic photoreceptors that enable dynamic manipulation of cell signaling pathways with spatiotemporal precision [155]. Optogenetics describes combining lightsensitive molecules with cell signaling machinery that then renders cell signaling activity under control of a specific light wavelength (Figure 4). Using this approach light can be used to drive the signaling pathways normally triggered selectively by, e.g., specific chondro-or osteogenic growth factors. Further details on optogenetic technologies and their developmental implementation can be found elsewhere [156][157][158][159][160].
Incorporation of optogenetic tools within TE is currently in its infancy, but there is great potential for combining technologies to enhance current approaches (reviewed [161,162]). Signaling pathways crucial for skeletal development and differentiation of hPSCs/MSCs include the TGFb and BMP signaling pathways (Sections "Developmentallyguided Differentiation" and "Stimulation of Growth factor Signalling"). Targeted optogenetic manipulation of both pathways to investigate early developmental processes in vivo [163,164], which control intracellular signaling dynamics [165] and elicit downstream transcriptional activity in hPSCs [166], has been described in recent years. Additionally, optogenetic induction of BMP-2 gene expression within MSCs has been demonstrated to enable control of osteogenic fate in vitro and fine-tune bone regeneration in vivo [167]. Such reports illustrate the potential for integrating optogenetic technologies within skeletal differentiation approaches. Their future use will improve hPSC differentiation consistency and enable control of regionalized cell signaling cues within 3D-stratified constructs.

Conclusions and outlook
The advent of stem cell technologies has led to major breakthroughs in modern medicine, with promising results in the regeneration of multiple tissue types, including the skeletal system. Multiple stem cell-based therapies are now reaching clinical trials [168], including adult MSCs for the treatment of osteoarthritis and bone lesions [169].
The establishment of methods to efficiently differentiate hPSCs into skeletal lineages such as chondrogenic, osteogenic or tenogenic-like cells has led to their identification as a promising cell source for orthopedic applications due to their availability, expansion potential, and potency when compared to adult MSCs or alternative cell types. From a technical perspective, and although many differentiation protocols are available, the reproducibility, specificity, scalability and associated costs of using hPSCs still limit their clinical translation. In order to generate cell types that closely resemble native tissues, hPSC differentiation protocols can be designed to follow the developmental routes that generate the authentic tissue types in the embryo and fetus. This strategy is likely to result in improved differentiation stability and more homogeneous cell populations. However, developmentally-guided differentiation protocols are still heavily reliant on costly reagents such as GFs. The use of improved delivery methods and the identification of suitable alternatives such as small molecule analogues, or the use of innovative technologies such as cell sorting or optogenetics, is expected to lead to cost reductions and improved efficiencies for TE.
In order to repair relevant clinical defects, large-scale 3D tissues are needed. However, most hPSC differentiation protocols have been developed as 2D monolayer cultures and lack validation in 3D. The integration of biofabrication technologies to reproduce the structural and functional organization of skeletal tissues in 3D will be crucial to translate the potential of hPSC-derived cells to the clinic. For that purpose, the rapidly-advancing field of biofabrication technologies is likely to offer unprecedented benefits to generate highly complex, patient-specific models. Their integration with advanced bioreactors capable of providing dynamic culture conditions resembling the in vivo micromechanical environment with appropriate stimuli such as compression, tension, and shear, combined with computational modeling will likely facilitate improved methods to scale up the production of large batches of cells. Together, these different advances will reduce costs of cellular therapies, and accelerate the route of hPSC skeletal therapies to the clinic.

Disclosure statement
No potential conflict of interest was reported by the author(s).