Modeling of skeletal development and diseases using human pluripotent stem cells

Abstract

Human skeletal elements are formed from distinct origins at distinct positions of the embryo. For example, the neural crest produces the facial bones, the paraxial mesoderm produces the axial skeleton, and the lateral plate mesoderm produces the appendicular skeleton. During skeletal development, different combinations of signaling pathways are coordinated from distinct origins during the sequential developmental stages. Models for human skeletal development have been established using human pluripotent stem cells (hPSCs) and by exploiting our understanding of skeletal development. Stepwise protocols for generating skeletal cells from different origins have been designed to mimic developmental trails. Recently, organoid methods have allowed the multicellular organization of skeletal cell types to recapitulate complicated skeletal development and metabolism. Similarly, several genetic diseases of the skeleton have been modeled using patient-derived induced pluripotent stem cells and genome-editing technologies. Model-based drug screening is a powerful tool for identifying drug candidates. This review briefly summarizes our current understanding of the embryonic development of skeletal tissues and introduces the current state-of-the-art hPSC methods for recapitulating skeletal development, metabolism, and diseases. We also discuss the current limitations and future perspectives for applications of the hPSC-based modeling system in precision medicine in this research field.

Introduction

Skeletal development is initiated by 3 cellular origins located at distinct positions in mammalian embryos: the neural crest, paraxial mesoderm, and somatic lateral plate mesoderm^1,2 The neural crest produces the facial bones, including the frontal bone, mandible, maxilla, and nasal bone, and fills the pharyngeal arches that ultimately generate the cartilage and bones of the jaws, middle ear, and neck.³ The paraxial mesoderm produces the axial skeleton and contributes to the neurocranium formation. The lateral plate mesoderm forms the appendicular skeleton and dorsal root of the aorta, the latter of which is a potential source of skeletal cells.⁴ There are primarily 2 modes of ossification: intramembranous and endochondral. In intramembranous ossification, during the embryonic stages, condensed mesenchymal cells directly differentiate into bone-forming osteoblasts when forming the skull and facial bones, whereas in endochondral ossification, mesenchymal cells condense and differentiate into a cartilage template that is subsequently replaced by bone, generating most of the axial and appendicular skeleton. Extensive evidence suggests more complex modes between the 2 forms. Lineage-tracing studies have demonstrated transdifferentiation from hypertrophic chondrocytes to osteoblasts during endochondral ossification.⁵ As every part of the skeleton is formed, unmineralized fetal cartilage has been reported to undergo dissolution and be remodeled into intramembranous bone.⁶ The periosteal collar of endochondral bones is also similar to intramembranous bone. Periosteal expansion occurs through direct deposition of osteoid without a cartilage template. This mode of ossification is an intramembranous-like process, although the persistent population of periosteal progenitor cells is clearly distinct from osteoblast progenitors during development.⁷

Various differentiation methods using human pluripotent stem cells (hPSCs) have been developed for better understanding of human skeletal development and generating skeletal cell types. In early attempts, traditional differentiation media for each target cell type were used; nevertheless, the efficacy and accuracy of induction were insufficient. To achieve more effective physiological differentiation, researchers began to consider the recapitulation of skeletal development. As every skeleton is formed by a specific combination of 3 origins and 2 modes of ossification at different locations, various differentiation methods have been developed based on the context of the target tissues. This is also motivated by the evidence that regenerative capacity varies depending on the origin of skeletal elements.⁸ In this review, we first briefly summarize our current understanding of the embryonic development of skeletal tissues through the 3 origins and then introduce the current state-of-the-art hPSC differentiation methods for skeletal cell types, primarily chondrocytes and osteoblasts.

Understanding the development of the 3 origins of skeletal elements

Mesodermal development

The mesoderm, along with the ectoderm and endoderm, is generated during gastrulation.^9,10 In mice, the commencement of gastrulation is marked by the formation of the primitive streak (PS) in the epiblast region.¹¹ During this process, the mesendoderm, consisting of uncommitted epiblast cells, migrates through the PS and exits as either the mesoderm or definitive endoderm, where the distinct subpopulations of the mesoderm are specified temporally and spatially (Figure 1A). The first epiblast cells to migrate through the posterior PS produce the extraembryonic mesoderm. As gastrulation proceeds, cells migrate through more anterior parts of the PS. From the posterior to anterior position, the lateral plate mesoderm, the intermediate mesoderm, the paraxial mesoderm, and the axial mesoderm are formed accordingly. The axial mesoderm generates the notochordal mesoderm.¹³

Figure 1.

Mesoderm development. (A) Schematic of the specification of the mesoderm. From the posterior to anterior position, the lateral plate mesoderm, the intermediate mesoderm, the paraxial mesoderm, and the axial mesoderm are formed. (B) Schematic of cell specification during paraxial mesoderm development. The paraxial mesoderm produces somites that contain the sclerotome and dermomyotome; the former produces the sclerotome and syndetome, and the latter produces the myotome and dermatome. Sonic hedgehog (Shh) secreted from the notochord is essential for sclerotome induction. Wnt secreted from the neural tube induced the dermomyotome and maintained the epithelial state in somites. The notochord also produces noggin, which inhibits BMPs secreted from the lateral plate mesoderm and supports sclerotome induction. These figures are modified versions of images from a previous review.¹²

Table 1.

Summary of protocols for generating human pluripotent stem cell-derived chondrocytes.

Authors	Source	Route	Differentiation stages	Differentiation reagents and procedures	In vitro assessments	In vivo assessments	Ref
Umeda et al. (2012)	hESC	Paraxial mesoderm	Stage 1: EB - mesoderm	BIO, Activin A or SB431542; Cell sorting (KDR-; PDGFRa+)	Gene expression Flow cytometry Histology	None	60
			Stage 2: Chondrogenesis	PDGF, TGFβ3, BMP4
Dicks et al. (2020)	COL2A1-GFP reporter hiPSCs	Paraxial mesoderm	Stage 1: Anterior PS	Activin, CHIR99021, FGF-2	Gene expression Histology scRNA-seq	None	61
			Stage 2: Paraxial mesoderm	SB-505124, CHIR99021, FGF, dorsomorphin
			Stage 3: Early somite	SB-505124, dorsomorphin, Wnt-C59, PD173074
			Stage 4: Sclerotome	purmorphamine, Wnt-C59
			Stage 5: Chondroprogenitor	BMP-4; Cell sorting (CD146+; CD166+; PDGFRβ+; CD45-)
Craft et al. (2015)	hESC, hiPSC	Paraxial mesoderm	Stage 1: EB formation	BMP4	Gene expression Flow cytometry Histology	Subcutaneous implantation Histology	62
			Stage 2: PS	Activin A, BMP4, FGF
			Stage 3: Paraxial mesoderm	Dorsomorphin, FGF
			Stage 4: Chondroprogenitors	TGFβ3
			Stage 5: Chondrogenesis	TGFβ3 or BMP4
Xi et al. (2017)	hESC, hiPSC	Paraxial mesoderm	Stage 1: PS, Posterior presomitic mesoderm	CHIR99021	Gene expression Bulk RNA-seq Flow cytometry Histology	None	18
			Stage 2: Anterior presomitic mesoderm/Somite	LDN193189, SB431542
			Stage 3: Sclerotome	SAG, FGF2
			Stage 4: Chondrogenesis	StemPro Chondrogenesis Differentiation medium
Loh et al. (2016)	hESC, hiPSC	Paraxial mesoderm	Stage 1: Anterior PS	Activin A, CHIR99021, FGF2, PIK90	Gene expression Bulk RNA-seq scRNA-seq ATAC-seq Flow cytometry Histology	Subcutaneous implantation Histology	63
			Stage 2: Paraxial mesoderm	A-83–01, CHIR99021, LDN-193189 or DM3189, FGF2
			Stage 3: Early somite	A-83–01, LDN-193189, C59, PD0325901
			Stage 4: Sclerotome	21K, C59
			Stage 5a: Chondrogenesis	BMP4
Wu et al. (2021)	hiPSC	Paraxial mesoderm	Stage 1: Anterior PS	Activin A, CHIR99021, FGF2	Gene expression Bulk RNA-seq scRNA-seq Flow cytometry Histology	Implantation to osteochondral defects in mice Histology	64
			Stage 2: Paraxial mesoderm	SB-505124, CHIR99021, FGF2, dorsomorphin
			Stage 3: Early somite	SB-505124, dorsomorphin, Wnt-C59, PD173074
			Stage 4: Sclerotome	purmorphamine, Wnt-C59
			Stage 5a: Chondrogenesis	BMP4
Yamada et al. (2021)	PRRX1-tdTomato- hiPSC, hESC	Lateral plate mesoderm	Stage 1: Mid PS	Activin A, hBMP-4, CHIR99021, FGF2, PIK90, Y-27632	Gene expression Bulk RNA-seq Flow cytometry Histology Transmission electron- microscopy	Implantation to osteochondral defects in mice and rats Subcutaneous implantation with scaffold Histology	92
			Stage 2: Lateral Platre Mesoderm	A-83-01, hBMP-4, Wnt-C59, Y-27632
			Stage 3: Limb-bud Mesenchyme	A-83-01, LDN193189, CHIR99021, Vismodegib, Y-27632
			Stage 4: Expandable Limb-bud Mesenchyme	A-83-01, CHIR99021, hFGF-2, hEGF, Y-27632
			Stage 5: Chondrogenesis step1	Ascorbic acid, ITS-G, CHIR99021, hFGF-2
			Stage 6: Chondrogenesis step 2	Ascorbic acid, ITS-G, hBMP-4, hTGF-β1, hGDF-5, hFGF-2
			Stage 7: Chondrogenesis step 3	Ascorbic acid, ITS-G, hBMP-4, hTGF-β1, hGDF-5
Smith et al. (2023)	hESCs	Lateral plate mesoderm	Stage 1: Mid PS	Activin A, BMP2, CHIR99021, FGF2, PIK90	Gene expression Bulk RNA-seq Flow cytometry Histology	None	65
			Stage 2: Lateral Platre Mesoderm	BMP2, FGF2, C59, SB431542
			Stage 3: Limb-bud Mesoderm	BMP2, CHIR99021, FGF2
			Stage 4: Chondro-progenitor	FGF2, C59, GDF5
			Stage 5: Prechondrocyte	FGF2, GDF5
Umeda et al. (2015)	hESC, hiPSC	Neural crest	Stage 1: Neural Crest	SB431542	Gene expression Bulk RNA-seq Flow cytometry Histology	Subcutaneous implantation Histology	66
			Stage 2: Expandable Ectomesenchymal Cell	FGF2, SB431542; Cell sorting (CD271high;CD73-)
			Stage 3: Chondrocyte progenitor	FGF2 (+TGFβ)
			Stage 4: Chondrocyte	PDGF, TGFβ, BMP
Shen et al. (2022)	hESC, hiPSC	Neural crest	Stage 1: Neural Crest	BMP4, SB431542	Gene expression Bulk RNA-seq scRNA-seq Flow cytometry Histology	Subcutaneous implantation with or without scaffold Implantation into a cartilage defect model in rats Histology	67
			Stage 2: Expandable Ectomesenchymal Cell	CHIR99021, SB431542, SAG, EGF, FGF2, DMH1

The list includes the first author, source of cells, route of differentiation, stages of differentiation, differentiation reagents and procedures, in vitro assessments, and in vivo assessments. Protocols that induced skeletal cells through specific routes were selected.

Although Wnt signaling is essential for all mesodermal subtypes, the coordination of BMP and Nodal signaling is crucial in the specification of the subtypes. BMP4 is the primary BMP family member responsible for the mesodermal patterning of the PS in tandem with region-specific concentrations of Nodal.¹⁴ A gradient of these signaling molecules is crucial in the anterior-to-posterior axis; the anterior PS receives high Nodal signals,¹⁵ whereas the posterior PS receives high BMP signals.¹⁶

Paraxial mesoderm development

Development of the paraxial mesoderm comprises several stages, including presomitic mesoderm specification, somitogenesis, and somite specification.¹² Wnt signaling positively impacts presomitic mesoderm differentiation^17,18 and induces the expression of transcriptional regulators, including Brachyury (T), T-box transcription factor 6 (Tbx6), and Mesogenin 1 (Msgn1).^17,19,20 Noggin, a BMP inhibitor that is secreted from the adjacent notochord, prevents the lateralization of the presomitic mesoderm.²¹

Somite formation begins as paraxial mesoderm cells become organized into whorls of cells known as somitomeres.¹² In chick embryos, cranial somitomeres take shape along Hensen’s node until 7 pairs have formed. Cells from these somitomeres contribute to craniofacial skeletal elements in the head. Caudal to the seventh somitomere, they become compacted and bound together by an epithelium and eventually separate from the presomitic paraxial mesoderm to form individual somites. During somitogenesis, the presomitic mesoderm forms somites through the segmentation clock, determination front, and mesenchymal–epithelial transition.¹² The segmentation clock is regulated by fibroblast growth factor (FGF) and Notch signaling.²² In mice, somite formation is regulated by FGF8 and Wnt in a posterior–anterior gradient, which is established by a messenger RNA decay mechanism with a retinoic acid (RA) gradient in the opposite direction.²³ FGF signals control the pace of segmentation in a phase opposite to that of Wnt/β-catenin.

The somite contains 2 major populations: the sclerotome, which produces the sclerotome and syndetome, and the dermomyotome, which produces the myotome and dermatome¹² (Figure 1B). The sclerotome produces the vertebrae and associated ribs, whereas the syndetome produces tendons. The myotome produces the musculature of the back, rib cage, ventral body wall, and limbs, whereas the dermatome generates the dermis.

Table 2.

Summary of protocols for generating human pluripotent stem cell-derived osteoblasts.

Authors	Source	Route	Differentiation stages	Differentiation reagents and procedures	In vitro assessments	In vivo assessments	Ref
Xi et al. (2017)	hESC, hiPSC	Paraxial mesoderm	Stage 1: PS, Posterior presomitic mesoderm	CHIR99021	Gene expression Bulk RNA-seq Flow cytometry Histology	None	18
			Stage 2: Anterior presomitic mesoderm/Somite	LDN193189, SB431542
			Stage 3: Sclerotome	SAG, FGF2
			Stage 4: Osteogenesis	StemPro Osteogenesis Differentiation medium
Kidwai et al. (2020)	hESC-RUNX2-YFP, hiPSC	Paraxial mesoderm	Stage 1: Primitive streak	CHIR99021	Gene expression Bulk RNA-seq Flow cytometry Histology	Subcutaneous implantation with scaffold Histology	68
			Stage 2: Paraxial mesoderm	LDN193189, SB431542
			Stage 3: Osteogenesis	BMP2, FGF9, Wnt3a, Rapamycin
		Lateral plate mesoderm	Stage 1: Primitive streak	CHIR99021
			Stage 2: Lateral plate mesoderm	BMP4, VEGF
			Stage 3: Osteogenesis	BMP2, FGF9, Wnt3a, Rapamycin
		Neural crest	Stage 1: Neural crest	SB431542, CHIR99021
			Stage 2: Osteogenesis	BMP2, FGF9, Wnt3a, Rapamycin
Tani et al. (2023)	hESC, hiPSC	Paraxial mesoderm	Stage 1: Primitive streak	CHIR99021	Gene expression Bulk RNA-seq Flow cytometry Histology scRNA-seq	Renal capsule implantation Histology scRNA-seq scMultiome	69
			Stage 2: Paraxial mesoderm	CHIR99021, A83-01, LDN193189
			Stage 3: Somitic mesoderm	C59, A83-01, LDN193189
			Stage 4: Sclerotome	SAG, C59, LDN193189
			Stage 5: Osteogenesis	Renal capsule implantation
Lamandé et al. (2023)	hiPSC	Paraxial mesoderm	Stage 1: Primitive streak	Activin A, CHIR99021, FGF2, PIK90	Gene expression Bulk RNA-seq scRNA-seq Flow cytometry Histology	Subcutaneous implantation Histology	70
			Stage 2: Paraxial mesoderm	A-83–01, CHIR99021, LDN-193189, FGF2
			Stage 3: Somitic mesoderm	A-83–01, LDN-193189, C59, PD0325901
			Stage 4: Sclerotome	Purmorphamine, C59
			Stage 5: Chondrogenesis	FGF2
			Stage 6: Chondrocyte hypertrophy	T3
			Stage 7: Osteogenesis	Osteogenic medium
Smith et al. (2023)	hESCs	lateral plate mesoderm	Stage 1: Mid PS	Activin A, BMP2, CHIR99021, FGF2, PIK90	Gene expression Bulk RNA-seq Flow cytometry Histology	None	65
			Stage 2: Lateral Platre Mesoderm	BMP2, FGF2, C59, SB431542
			Stage 3: Limb-bud Mesoderm	BMP2, CHIR99021, FGF2
			Stage 4: Osteoblast	CHIR99021

The list includes the first author, source of cells, route of differentiation, stages of differentiation, differentiation reagents and procedures, in vitro assessments, and in vivo assessments. Protocols that induced skeletal cells through specific routes were selected.

Sclerotome induction from the somite depends on sonic hedgehog (Shh), secreted from the notochord and floor plate of the neural tube in mice,²⁴ resulting in the expression of Pax1, Nkx3.2, and Sox9.²⁵ Sclerotome specification initially requires the antagonism of BMP by continuous notochordal noggin activity²⁶; however, subsequent chondrogenesis is maintained by BMP signaling.²⁵ BMPs induce Sox9, a master regulator of chondrogenesis, whose activity is essential throughout chondrocyte development.²⁷ Further signaling and gene regulatory mechanisms in endochondral ossification have been described in previous reviews.^28,29

The paraxial mesoderm contributes to craniofacial skeletal elements in mice, particularly the parietal, occipital, temporal, and sphenoid bones.³⁰ The intermediate mesoderm produces the urogenital system, consisting of the kidneys, gonads, and their respective duct systems. The lateral plate mesoderm forms the splanchnic mesoderm, somatic mesoderm, and extraembryonic membranes, as shown in a study on chick embryos.³¹ The splanchnic mesoderm produces components of the circulatory system, such as the heart, blood vessels, and blood cells, whereas the somatic mesoderm forms the pelvic skeleton and mesodermal components of the limbs, except for muscles derived from the dermomyotome.³²

Lateral plate mesoderm development

As explained previously, specification of the mesoderm is regulated by the BMP and Nodal gradient, which coordinate the patterning of the anterior–posterior and dorsoventral axes.³³ High levels of BMP signaling in the ventral domain specify the lateral plate mesoderm in embryos. In early appendicular skeletal development, the somatic lateral plate mesoderm produces the limb bud mesenchyme.³⁴ The identity of the forelimb and hindlimb is regulated by the antagonistic gradient of RA and FGF signaling.³⁴ RA is required for the expression of Tbx5, which is essential for the specification and formation of the forelimb.³⁵  Gdf11 expression is involved in hindlimb positioning by regulating hindlimb-related genes such as Tbx4 and Pitx1.^36,37

Limb patterning and outgrowth are regulated by the apical ectodermal ridge (AER), where several members of the FGF family are required in a time- and dose-dependent manner.³⁸ An antagonistic feedback loop is established between the proximal RA cue AER-FGF8 and the non-AER ectoderm-derived WNT3. In cooperation with the BMP signal, the antagonistic feedback loop drives the outgrowth of the limb bud.

Anteroposterior patterning and outgrowth of the limb bud mesenchyme are driven by the zone of polarizing activity, which is the Shh signaling center. The BMP antagonist gremlin-1 is initially induced by BMP4, and its expression is improved by Shh, resulting in the restriction of BMP activity to the distal edge of the limb bud. Gremlin-1 also promotes the upregulation of FGF expression in the AER, generating an Shh/AER-FGF/gremlin-1 feedback loop.^39,40 This inhibitory loop terminates the outgrowth of the AER, which probably determines the tissue size.⁴¹ After the removal of the inhibitory loop, high BMP activity promotes the transition of mesenchymal progenitor cells into proliferating prechondrocytes, which undergo endochondral ossification.⁴² Lineage-tracing analysis in mice indicated that PRRX1-positive cells produce Sox9-positive skeletal progenitors, which further differentiate into chondrocytes, osteoblasts, tenocytes, and synovial cells.^43,44

Single-cell RNA sequencing (scRNA-seq) can be used to determine limb mesenchymal progenitors. Markman et al.⁴⁵ conducted scRNA-seq analyses of mouse limb buds and identified Msx1-positive cells as naive-stage progenitors of the limb mesenchyme. He et al.⁴⁶ conducted scRNA-seq analyses of human embryonic limbs and identified putative embryonic skeletal stem/progenitor cells, which were marked by the adhesion molecule CADM1 and highly enriched with the FOXP1/2 transcriptional network. Further integrative analyses of rodents and humans will provide a fundamental understanding of gene expression profiles during limb development.

Neural crest development (Figure 2)

Figure 2.

Neural crest development. (A) Schematic of neural crest development, including neural plate border specification, neural crest specification, neural crest epithelial–mesenchymal transition (EMT), and neural crest migration. Activities of key signaling pathways (fibroblast growth factor, BMP, and Wnt) for neural plate border specification are shown. (B) The neural crest is divided into 4 subpopulations along with the anteroposterior body axis: cranial, vagal, trunk, and sacral. These figures are modified versions of images from a previous review.⁴⁷

The neural crest is derived from the ectoderm and produces various cell types, including cells for the facial skeleton and the sensory and sympathetic nervous systems, adrenomedullary cells, and tissue pigment cells.^48,49 Neural crest formation is initiated during gastrulation. The presumptive neural crest territory forms at the neural plate border between the future neural and non-neural ectoderms. Wnt, BMP, and FGF signaling specify this border; the morphogen gradient generated by the antagonists of BMP and Wnt and FGF activation play important roles in the specification.⁵⁰ Wnt and BMP signaling regulate the expression of genes that specify the neural plate border, such as Zic1, Msx1, and Tfap2,⁵¹ after which the expression of neural crest specifiers, including Foxd3, Snai1/2, and Sox8/9/10, is activated.⁵⁰  Foxd3 and Sox10 are bona fide neural crest markers.⁵⁰ After neural tube closure/cavitation, neural crest cells (NCCs) undergo epithelial–mesenchymal transition and delaminate from the ectodermal neuroepithelium, migrating to a multitude of embryonic locations and contributing to different morphogenetic processes depending on their anterior–posterior axial location.⁵² NCC patterning is primarily controlled by homeobox (Hox) family transcription factors, which commit NCCs into region-specific categories, such as cranial, trunk, cardiac, and vagal NCCs.^48,49

Cranial NCC (CNCC) derivatives produce the facial skeleton, including the frontal bone, mandible, maxilla, and nasal bone. Most components of the craniofacial skeleton are formed through intramembranous ossification, in which CNCC-derived progenitors proliferate, condense, and directly differentiate into Runx2- and Sp7-positive osteoblasts. Runx2 and Sp7 are master transcriptional regulators in osteoblasts.^53,54 The facial bones primarily undergo intramembranous ossification. Other cranial bones, such as bones in the cranial base, are formed by endochondral ossification, in which cartilage is initially produced by CNCC-derived Sox9-positive chondrocytes and is gradually replaced by bone.⁵⁵ Some cranial cartilages do not undergo direct replacement by bone but instead undergo a regression process,⁶ whereas other cartilage, including part of the nasal septum, does not undergo mineralization.

Signaling pathways such as BMP, FGF, and Wnt signaling also play crucial roles in the formation of cranial bone and cartilage. An appropriate amount of BMP signaling is essential for proper craniofacial morphogenesis. Increased BMP signaling causes the formation of ectopic cartilage in the craniofacial region.⁵⁶ FGF signaling is a positive regulator in osteogenesis, wherein activation of FGF signaling in cells of the frontal bones mediates the proliferation and differentiation of osteoblasts.⁵⁷ FGF2 stimulates BMP2 expression in osteoblasts and modulates Wnt/β-catenin signaling during cranial development.⁵⁸ Wnt/β-catenin signaling is essential for the development of CNCC-derived skeletal elements.⁵⁹

In vitro differentiation methods of skeletal cell types from hPSCs

hPSC technology has been widely utilized to model human development and diseases in vitro. Recapitulating skeletal development is key to inducing skeletal cells, where skeletal tissues of different origins undergo distinct developmental phases. Methods for the stepwise and stage-specific induction of hPSCs into skeletal cell types of different origins have been developed (Figure 3). Human skeletal diseases, especially genetic diseases, have been modeled using patient-derived induced pluripotent stem cells (iPSCs), providing insights into molecular pathogenesis and drug screening platforms. Hereafter, we review the hPSC techniques used for introducing skeletal cell types and recapitulating skeletal development, metabolism, and disease.

Figure 3.

Overview of osteoblast and chondrocyte induction from human pluripotent stem cells (hPSCs). Major routes of stepwise induction from hPSCs to skeletal cells, including osteoblasts and chondrocytes, are shown. The differentiation protocols for each cell lineage are summarized in Tables 1 and 2.

In addition to induction protocols, methods for evaluating cell differentiation have been developed. In the initial stages of research, marker gene expression and morphological evaluations, such as chondrocyte morphology and bone mineralization, were the standard criteria. Later, RNA-seq analysis enabled the transcriptome evaluation of gene expression profiles, and flow cytometry was used to determine the efficiency of the induced cells. In addition, scRNA-seq analysis can reveal the gene expression and induction efficiency at the single-cell level. Further omics analyses, including epigenomics and proteomics, have provided valuable information regarding cell states. In addition to profiling induced cells, the implantation of cells in animals and the evaluation of in vivo tissue formation have been applied to test the differentiation capacities of induced progenitors. Therefore, it is important to quantify the induction methods using several criteria to understand the methods that fit the purpose of the study.

In vitro differentiation of paraxial mesoderm-derived skeletal cells

Considering that mesoderm-derived tissues originate from the PS,⁹ the initial step in inducing a mesodermal derivative from hPSCs is PS induction. This induction step has been developed by Gordon Keller’s group, who conducted a comparative analysis between mouse embryos and in vitro culture of embryonic stem cells (ESCs) using reporter lines that visualize cell populations expressing the PS marker, Brachyury (T), and those highly expressing the anterior PS marker, Foxa2.⁷¹ Activin-A induces the endoderm from the anterior PS,⁷² whereas inhibition of TGFβ signaling suppresses endoderm formation and promotes mesodermal phenotypic changes in hPSCs.⁶³ The combined stimulation of TGFβ/Nodal, BMP, and FGF signaling promotes the posterior primitive differentiation of hPSCs. Wnt signaling activators, such as CHIR99021, induce the PS and presomitic mesoderm, whereas coactivation of FGF and Wnt signaling improves the efficacy of induction.^18,73,74 BMP signaling promotes the formation of the lateral plate mesoderm but suppresses the formation of the paraxial mesoderm due to its crucial role in the mediolateral axis through an activity gradient.⁶³

In the subsequent steps of PS induction, various stepwise methods have been developed for derivatives of the paraxial or lateral plate mesoderm. For example, Darabi et al.⁷⁵ demonstrated a cell-sorting strategy to enrich paraxial mesoderm populations. In addition, as extensively reviewed by Humphreys et al.⁷⁶ and De Kinderen et al.⁷⁷, several protocols have been established for generating derivatives of the paraxial mesoderm. These protocols are essential for the sequential recapitulation of embryonic development, including the development of the anterior presomitic mesoderm, somites, and sclerotome. As described earlier, the anterior presomitic mesoderm in the determination front forms somites according to the oscillation of Notch signaling and the antagonizing effects of Wnt/FGF. Several research groups have demonstrated somite induction by inhibiting both Wnt and FGF/ERK signaling,^63,73,78 although other groups have demonstrated that Wnt inhibition alone was sufficient for somite induction.^79,80 The BMP gradient is a crucial factor determining the mediolateral axis during mesoderm development and somite specification.^21,81 Consistent with this finding, some studies have shown that BMP inhibition improves somite induction, possibly by promoting the medial fate and protecting the mesodermal population from lateralization.^18,63,73,79 Activation of hedgehog (Hh) signaling is critical for sclerotome induction.⁸² Shh, which is produced by the notochord and floor plate, plays a vital role in somite specification in the sclerotome, as evidenced by studies using in vivo animal models.^81,82

Several induction methods have been established for the in vitro differentiation of chondrocytes from the hPSC-derived paraxial mesoderm.^76,77 Cell-sorting strategies have been used to improve the efficacy of chondrocyte induction. After induction of the somitic paraxial mesoderm, Umeda et al.⁶⁰ isolated cells negative for kinase insert domain receptor 2 (KDR2) and positive for platelet-derived growth factor receptor a1 (PDGFRa1). They found that the isolated cell population had a higher potential to generate a hyaline-like cartilage matrix without inducing chondrocyte hypertrophy after treatment with 3 chondrogenic factors, PDGF, TGFβ, and BMP. Similarly, Dicks et al.⁶¹ identified CD146-positive, CD166-positive, PDGFRβ-positive, and CD45-negative chondroprogenitors, which exhibited higher expression of chondrogenic marker genes (SOX9, COL2A1, and ACAN), in the population induced from the hPSC-derived paraxial mesoderm. The sorted cells further differentiated and induced type X collagen, a matrix protein associated with hypertrophic chondrocytes, suggesting that these cells further differentiate into hypertrophic chondrocytes and undergo endochondral ossification.⁶¹

Craft et al.⁶² developed an embryoid body culture to induce the paraxial mesoderm, followed by micromass culture with either BMP4 or TGFβ3. They found that BMP4 treatment induced chondrocyte hypertrophy, which initiated endochondral ossification in vivo. In contrast, TGFβ3 treatment imparted articular chondrocyte-like properties in vitro and in vivo. Importantly, no calcification or type X collagen expression was observed in the induced cartilage in vivo. This group further investigated gene expression and chromatin accessibility profiles in the induced cells.⁸³ By integrating the datasets with the profiles of mouse embryonic chondrocytes, they revealed cell type–specific gene regulatory networks.⁸³ For example, they validated the enrichment of RUNX2 and RELA activities in the growth plate and articular chondrocytes, respectively.⁸³ Considering that articular chondrocytes possess limited healing potential and are therefore highly relevant to unmet clinical needs, induction protocols for articular chondrocytes appear promising for a better understanding of articular development and potential clinical applications.

Xi et al.¹⁸ developed a method for inducing chondrocytes from the hPSC-derived somitic mesoderm in 2D culture. After specification of the sclerotome from the somitic mesoderm by treatment with smoothened agonist (SAG) and FGF2, cells were differentiated into chondrocytes via 4-week pellet culture in a commercially available chondrogenic medium. This method resulted in the upregulation of chondrogenic marker genes along with the production of cartilage matrix, including type II collagen. Approximately 60% of the cells were positive for SOX9 during induction. COL10A1 was also upregulated, possibly inducing endochondral ossification.

Loh et al.⁶³ proposed a comprehensive strategy for inducing hPSCs using several derivatives of the paraxial and lateral plate mesoderm. They identified signaling combinations that control mesodermal fate specification by promoting the desired fate and blocking the undesired fate. Importantly, RNA-seq and ATAC-seq analyses revealed the sequential differentiation of hPSCs into mesodermal derivatives in terms of gene expression and chromatin profiles. scRNA-seq analysis also confirmed the efficacy of the induction. Recently, Yamanaka et al.⁸⁴ and Miao et al.⁸⁵ independently reported their methods for recapitulating human 3D somitogenesis in vitro. In both methods, the paraxial mesoderm was initially induced from hPSCs and then cultured in Matrigel for the formation of spheroids, which elongated and subsequently formed somite-like structures. Importantly, the oscillatory dynamics of the segmentation clock were recapitulated, as evidenced by HES7 activity, leading to the formation of anterior and posterior somite compartments. In addition to the scRNA-seq analysis used to validate the gene expression profiles, spatial transcriptomics further suggested the existence of a Hox code in the induced somite-like structure.⁸⁴ Although these methods have not been extended to skeletal formation following somitogenesis, they are promising tools for the spatiotemporal recapitulation of the development of the paraxial mesoderm.

Subsequently, Loh et al.⁶³ showed that after the stepwise induction of the sclerotome through the hPSC-derived paraxial mesoderm in 2D culture, BMP4 treatment resulted in the differentiation of the induced sclerotome into chondrocytes in vitro. When the chondrocytes were subcutaneously implanted into immunodeficient mice, they became hypertrophic, resulting in endochondral ossification. Wu et al.⁶⁴ improved the induction method described by Loh et al.⁶³ by reducing off-target differentiation of the paraxial mesoderm into neural cells and melanocytes during chondrocyte differentiations. Using scRNA-seq data, they found that Wnt signaling was produced by off-target cells and that Wnt signaling induced chondrocyte hypertrophy. The use of a Wnt inhibitor augments chondrocyte properties from hypertrophy during induction.⁶⁴ When the induced chondrocyte pellets were subcutaneously implanted with the Wnt inhibitor into immunodeficient mice, no hypertrophic chondrocytes were induced in vivo within 14 days.

A few studies have reported osteoblast differentiation in the hPSC-derived paraxial mesoderm. For example, Xi et al.¹⁸ developed a method to induce osteoblasts from the hPSC-derived somitic mesoderm in a 2D culture. After specification of the sclerotome, cells were differentiated into an osteoblast lineage via a 28-day culture in a commercially available osteogenic medium. This method induced approximately 40%-50% of RUNX2-positive cells as well as mineralization.

Kidwai et al.⁶⁸ proposed another method for osteoblast differentiation from 3 skeletal origins: NCC, the paraxial mesoderm, and the lateral plate mesoderm. Each induced cell was differentiated into osteoblasts in a common osteogenic medium supplemented with WNT3A, BMP2, FGF9, and rapamycin in 2D culture. They also subcutaneously implanted the induced cells with ceramic particles into immunodeficient mice to evaluate their osteogenic potential in vivo. Interestingly, the NCC-derived cell population underwent intermembranous ossification, whereas the mesoderm-derived cell population formed endochondral bones.⁶⁸ Bulk RNA-seq analysis revealed origin-specific gene expression profiles that may contribute to different outcomes.⁶⁸

Exploiting the in vivo implantation strategies and single-cell multiome analyses, we recently developed a method to induce 3D endochondral bone-like tissue formation from hPSCs and investigated the molecular mechanisms underlying human skeletal development.^69,86 We induced a sclerotome population and implanted it with Matrigel beneath the renal capsules of immunodeficient mice. Radiopaque tissues were examined 8 weeks after implantation using histological and scRNA-seq analyses, which revealed that these tissues were endochondral bones composed of hPSC-derived skeletal cells and mouse circulatory cells. Integrative analysis using scRNA-seq datasets from human embryonic long bones at 8 weeks postconception⁴⁶ showed that the skeletal cell types and their trajectories in the induced tissues were similar to those of human embryos. We evaluated the dynamics of chromatin accessibility and transcription factor activity in this model. Considering that chromatin accessibility defines the DNA binding of transcription factors through specific motifs,⁸⁷ we predicted the dynamics of gene regulatory networks caused by the activity of cell type–specific transcription factors in endochondral ossification. Importantly, the following key transcription factors, identified by mouse genetic studies and genomic analyses, were enriched in the predicted networks: FOXA2⁸⁸ for chondrocyte hypertrophy, DLX5-SP7⁸⁹ for osteoblast specification, and RUNX2⁹⁰ for both osteoblasts and hypertrophic chondrocytes. We identified ZEB2 as a novel osteogenic factor involved in endochondral ossification.⁶⁹ Therefore, these datasets may be useful for clarifying the gene regulatory mechanisms underlying human skeletal development.

Similarly, Lamandé et al.⁷⁰ developed a model of endochondral ossification through the paraxial mesoderm. In their protocol, chondroprogenitors derived from the hPSC-derived sclerotome were further differentiated into articular chondrocytes by TGFβ3 treatment, whereas the chondroprogenitors underwent hypertrophy by treatment with triiodothyronine (T3, a thyroid hormone) in the pellet culture. Strikingly, hypertrophic chondrocytes were transdifferentiated into osteogenic cells via a 3-week culture in conventional osteogenic medium. This phenomenon is consistent with the findings from a mouse genetic study,⁹¹ which suggested that the transdifferentiation of hypertrophic chondrocytes into osteoblasts contributes to endochondral ossification. This method partially recapitulates the ossification process without the use of in vivo conditions. Complete in vitro modeling of endochondral ossification accompanied by bone collar formation and blood vessel invasion is the next step in this field.

In vitro differentiation of lateral plate mesoderm-derived skeletal cells

Loh et al.⁶³ successfully differentiated the hPSC-derived lateral plate mesoderm into a PRRX1-positive limb mesenchyme by suppressing the cardiac fate using FGF and Wnt signals. Yamada et al.⁹² improved a developmental method for generating the limb bud mesenchyme using PRRX1 reporter hiPSCs.⁹³ The induced cells expanded and demonstrated excellent chondrogenic properties without chondrocyte hypertrophy. Through stepwise chondrogenic differentiation in pellet culture, the cells formed hyaline cartilage–like tissues in vitro and the induced chondrocytes contributed to cartilage repair in a rat model of articular cartilage defect. Importantly, no hypertrophic chondrocytes were observed in the implantation assay using the induced cartilage. Smith et al.⁶⁵ developed a protocol for inducing skeletal cell types, including prechondrocytes and preosteoblasts, from the lateral plate mesoderm, wherein the former underwent chondrocyte differentiation without chondrocyte hypertrophy in pellet culture and the latter underwent osteoblast differentiation in 2D culture.

In addition to protocols for generating a specific skeletal cell type, Mori et al.⁹⁴ reported a method for reconstructing 3D limb bud tissues using mouse ESCs (mESCs). They first generated limb bud–like mesenchymal–epithelial complex tissues from mESCs. The induced cells selectively differentiated into forelimb- or hindlimb-type mesenchyme depending on the RA concentration. The thickness of the epithelial structure, which is similar to that of the AER, is regulated by BMP signaling. An implantation assay further showed the contribution of the induced cells to skeletal formation. Further studies, particularly those using hPSCs, will be valuable for modeling the development of human limbs.

In vitro differentiation of neural crest-derived skeletal cells

In early studies, researchers attempted to induce NCCs from hPSCs via neural precursor cells (NPCs), which subsequently produced NCC subpopulations. Several NPC induction methods have been proposed, including coculture with stromal cells,⁹⁵ neurosphere culture,⁹⁶ defined monolayer induction of neural rosettes,⁹⁷ and manual dissection of hESC-derived neural rosettes.⁹⁸ However, the induction efficacy of these methods has been proven to be insufficient, possibly because NCCs are a small subset of NPCs.⁹⁹ To improve the induction efficacy, direct induction methods have been developed using a combination of Wnt activation and TGFβ inhibition with small molecules of BIO or CHIR99021 (Wnt signaling activators) and SB431542 (a TGFβ signaling inhibitor).^100–102 Some studies have demonstrated that Wnt activation without TGFβ inhibition is sufficient to induce NCCs under specific conditions.^101,103

Several methods have been reported for inducing skeletal cell types. For instance, Umeda et al.⁶⁶ reported a differentiation method for long-term expandable SOX9-positive chondrogenic ectomesenchymal cells purified by cell sorting. CD271-positive, PDGFRα-positive, and CD73-positive chondrogenic ectomesenchymal cells were generated from PAX3-positive, SOX10-positive, and FOXD3-positive neural crest-like progeny via treatments with a TGFβ inhibitor and FGF. Remarkably, these ectomesenchymal cells were expandable without the loss of chondrogenic potential for at least 16 passages. When the cells were implanted subcutaneously into immunocompromised mice, all cells were mineralized, suggesting that the ectomesenchymal cell-derived chondrocytes matured into hypertrophic chondrocytes.⁶⁶

Shen et al.⁶⁷ developed a 2-step method to induce NCC-derived ectodermal chondrogenic cells (ECCs), which are self-renewable over long periods. In their method, NCCs were first induced using an induction medium supplemented with BMP4 and SB431542. In the second step, the induced NCCs were differentiated into ECCs using a medium supplemented with CHIR99021, SB431542, SAG, EGF, FGF2, and the BMP receptor inhibitor DMH1. The induced ECCs could be stably maintained with their self-renewal properties, and when the cocktail for the self-renewal condition was removed, the cells stopped proliferating and differentiated into a homogenous chondrocyte population.⁶⁷ The 3D culture of the induced chondrocytes in collagen sponge demonstrated that the induced cartilage produced proteoglycan-rich cartilage-like extracellular matrix, although some hypertrophic chondrocyte markers were also induced. In vivo assessments by subcutaneous implantation of the induced cartilage revealed that the cartilage tissue expressed collagen II, collagen X, aggrecan, and RUNX2. scRNA-seq analysis confirmed that gene expression profiles were comparable between chondrocytes induced from NCCs using their method and those induced from the hPSC-derived paraxial mesoderm.⁶⁷ It was also confirmed that these induced cells could regenerate defects in the mandibular condylar cartilage in vivo.⁶⁷ Therefore, these induction methods for expandable NCCs are promising to obtain sufficient progenitors for regenerative therapy of the facial skeleton. Further comparative analyses using human embryos will help clarify the integrity of the induction methods.

A few studies have reported induction protocols for NCC-derived osteoblasts. For example, Kidwai et al.⁶⁸ developed induction methods for osteoblast progenitors derived from skeletal elements of 3 origins: NCC, the paraxial mesoderm, and the lateral plate mesoderm. Osteoblasts were subsequently induced via a 6-day culture in an osteogenic basal medium supplemented with WNT3A, BMP2, FGF9, and rapamycin and continued to differentiate further in an osteogenic medium without any osteogenic mediators until day 28. Importantly, despite using the same osteogenic induction method, the biological signatures of the osteoblasts derived from the 3 origins were different; for example, NCC-derived osteoblast progenitors were strongly associated with the FGF1–RUNX2 axis.⁶⁸ Kobayashi et al.¹⁰⁴ developed an induction method for odontoblastic cells using CNCCs. They found that the combination of FGF4 and FGF9 improved the differentiation of CNNs into odontoblast-like cells under osteogenic/odontogenic differentiation conditions.¹⁰⁴

Disease modeling and drug screening

Advances in human stem cell technology and the clustered regularly interspaced short palindromic repeats (CRISPR) system have enabled evaluation of the effects of genetic mutations on skeletal formation in patient-derived hPSCs and differentiated cells. For example, fibrodysplasia ossificans progressiva (FOP) has been investigated extensively using iPSCs. FOP is a rare genetic disease characterized by congenital bone malformations and heterotopic ossification of soft tissues, including skeletal muscles, tendons, and ligaments.¹⁰⁵ It is caused by hyperactive mutations in the BMP type 1 receptor ACVR1/ALK2.¹⁰⁵ Matsumoto et al.¹⁰⁶ demonstrated that iPSCs derived from patients with FOP exhibited improved chondrogenesis and mineralization in an in vitro pellet culture and 2D culture in an osteogenic medium, respectively. They found that DMH1, a small-molecule inhibitor of BMP signaling, suppressed mineralization in mutant cells, suggesting the involvement of BMP signaling in the disease.¹⁰⁶ Hino et al.¹⁰⁷ explored the molecular mechanisms underlying FOP¹⁰⁷ and found that FOP-ACVR1 abnormally transduced BMP signaling in response to activin-A, a molecule that normally transduces TGF-β signaling but not BMP signaling. Activin-A improved the chondrogenesis of induced mesenchymal stromal cells derived from FOP-iPSCs in vitro using pellet culture, and FOP-iPSC-derived cartilage pellets were spontaneously calcified in vivo when they were subcutaneously implanted into immunodeficient mice. Hino et al.¹⁰⁸ performed drug screening using FOP-iPSCs and identified mTOR signaling as a crucial pathway in FOP pathogenesis. Rapamycin, an mTOR inhibitor, suppressed ectopic bone formation in an in vivo animal model. Takeyari et al.¹⁰⁹ generated iPSCs from patients with osteogenesis imperfecta (OI), a heritable brittle bone disease primarily caused by mutations in 2 type I collagen genes, COL1A1 and COL1A2. They found abnormalities in type I collagen synthesis in patient-derived fibroblasts and impaired mineralization of OI osteoblasts induced by patient-derived iPSCs through neural crest differentiation. They further found that 4-phenylbutyric acid improved these abnormalities in the culture.

Furthermore, several bone and cartilage diseases have been modeled using iPSCs and differentiation strategies, including cleidocranial dysplasia,¹¹⁰ Gorlin syndrome,¹¹¹ thanatophoric dysplasia and achondroplasia,^112,113 type II collagenopathy,^92,114,115 and familial osteochondritis dissecans.^116,117 These studies are extensively summarized in 2 previous review papers.^77,118

Current challenges and future perspectives

Knowledge of developmental biology has supported the development of hPSC-based modeling of human skeletal development, and disease modeling has provided a promising platform for the screening and identification of drug candidates. Nonetheless, several other factors must be considered. First, current modeling strategies lack standard assessments of induced cells, and different protocols have been evaluated using various methods. Essential assessments should include characterization at the molecular, morphological, and functional levels. In vivo morphological and functional assessments are the gold standard for regenerative therapy research. The former includes microscopic assessments using histological sections and macroscopic assessments, including radiological analysis; the latter includes biomechanical and metabolism evaluations in formed tissues. Marker gene/protein expressions have been widely used for molecular assessments. Recently developed omics analyses have enhanced comprehensive molecular characterization, which helps evaluate induction protocols and further understand cell differentiation mechanisms. Progress in scRNA-seq and bioinformatics will enable integrative analyses with several datasets of human skeletal tissues that have been generated and deposited in public databases.^{46,83,119–121} It is also important to obtain benchmark data, such as comprehensive gene expression profiles in primary skeletal cells. Moreover, because of the accumulated studies of mouse genetics and other species,^122,123 cross-species analysis will help elucidate the key biological features conserved among species. In addition, epigenome profiling, including chromatin accessibility, histone modification, and DNA methylation, will be useful for a more precise evaluation that represents distinct cell-type signatures more specifically than gene expression.¹²⁴

Second, the modeling of skeletal development requires further improvement. Currently, the ossification of bone tissue primarily depends on the implantation of progenitors into mice. Although one study demonstrated the transdifferentiation of hypertrophic chondrocytes into osteoblasts in vitro, no bone collar formation was observed in vitro.⁷⁰ The recapitulation of bone collar formation and subsequent blood vessel invasion is the next step in this field. Therefore, there exists a need for organoid methods involving multiple cell types. Bioengineering technologies are promising, particularly organ-on-a-chip systems, which have been developed to construct relevant in vivo microenvironments.^125–128 Additional challenges in the field include controlling the shapes of skeletal elements. The skeletal tissue dynamically changes shape during development in a spatiotemporal manner. Recapitulation of the morphogen gradient is crucial. As described for the human 3D somitogenesis method,^84,85 strategies for the self-assembly of cells in organoids can be extended to skeletal development, which may establish a morphogen gradient and regulate the shape of skeletal tissues in a cell-autonomous manner. Another approach involves the use of biomaterials. Koh et al.¹²⁹ developed a cube-based 3D scaffold that allowed for the generation of a concentration gradient of growth factors.

Third, in the case of disease modeling, it is necessary to consider genetic and epigenetic backgrounds. Extensive research has shown that the processes involved in iPSC reprogramming and expansion can cause the acquisition of genomic aberrations, including chromosomal aneuploidy, subchromosomal copy number variants, and point mutations.¹³⁰ Moreover, the patient-specific epigenetic landscape can be partially retained even after reprogramming, which is known as “epigenetic memory,” and can cause significant variations in cell differentiation among cell lines.¹³¹ To avoid these risks, screening of genetic integrity should become a standard procedure to screen iPSC lines with high mutations; not only karyotyping but also DNA-based fingerprinting, exome sequencing, and epigenetic analysis will be beneficial.^132,133 Generating several lines from a patient and testing the consistency between clones is also vital. Generating isogenic controls is essential to correctly evaluate genotype-phenotype relationships. The CRISPR system is often used to correct a specific mutation from patient-derived iPSCs. This renders the genomic background, except for the specific mutation, consistent between the control and experimental groups, providing us with precise evaluations of the mutation.

Fourth, modeling of complex diseases with common genetic variants, such as osteoarthritis and osteoporosis, is challenging. Most skeletal disease models target monogenic mutations with high penetrance and significant phenotypic effects. Several studies have generated iPSC lines using osteoarthritic patient-derived fibroblasts. For example, one study demonstrated that iPSC lines harboring variants associated with osteoarthritis exhibited differences in their chondrogenic capacity in micromass culture in vitro.¹³⁴ Another study demonstrated that compared with chondrocytes from osteoarthritic patient-derived iPSCs, those derived from healthy iPSCs were enriched in molecular pathways associated with energy metabolism and epigenetic regulation.¹³³ These findings suggest the presence of genetic links between the properties of chondrocytes and osteoarthritis, although the number of pathological conditions that are reproduced in these contexts remains unclear. The establishment of organoids for joint metabolism will be essential for recapitulating these phenomena. Integrative analysis of iPSC arrays with genome-wide association studies has been recently reported in other research fields.^135,136 These integrative studies of iPSC technologies and genomics are promising to gain a better understanding of the molecular mechanisms underlying pathological conditions.

Finally, we should consider the trade-off between simplicity and complexity in the modeling of organ development, metabolism, and pathogenesis.¹³⁷ An appropriate model must be selected based on the research questions and purposes. Induction protocols for single-cell types require a relatively simple method for providing more robust and reproducible outcomes. In contrast, although methods using self-assembly mechanisms, multicellular organoids, or complicated bioengineering approaches provide more physiologically relevant outcomes, these protocols tend to be more complicated and their robustness and reproducibility are generally limited. Investigating different methods with various degrees of complexity may help with optimization. For example, in the case of high-throughput screening, sequential screening may be useful. The first screening would use a robust and simple method, whereas the second screening would employ physiologically relevant and complicated methods. Using only one system is often insufficient to draw conclusions, and verification by several experimental systems is required to obtain concrete conclusions.

Author contributions

Hironori Hojo (Conceptualization, Data curation, Funding acquisition, Writing—original draft), Shoichiro Tani (Conceptualization, Data curation, Writing—review & editing), and Shinsuke Ohba (Conceptualization, Data curation, Funding acquisition, Writing—review & editing)

Funding

This work was supported by grants-in-aid for Scientific Research from the Japan Society for the Promotion of Science (JSPS: 20H03885, 21H03142, and 21 K19589), Rising Star Award from American Society for Bone and Mineral Research, the Japan Agency for Medical Research and Development (AMED; JP21bm0704071), the Japan Science and Technology Agency (JST) FOREST program (JPMJFR225N) and JST ERATO program (JPMJER2401).

Conflicts of interest

The authors declare no competing interests.

Data availability

There is no new data associated with this article.