Understanding paraxial mesoderm development and sclerotome specification for skeletal repair

Pluripotent stem cells (PSCs) are attractive regenerative therapy tools for skeletal tissues. However, a deep understanding of skeletal development is required in order to model this development with PSCs, and for the application of PSCs in clinical settings. Skeletal tissues originate from three types of cell populations: the paraxial mesoderm, lateral plate mesoderm, and neural crest. The paraxial mesoderm gives rise to the sclerotome mainly through somitogenesis. In this process, key developmental processes, including initiation of the segmentation clock, formation of the determination front, and the mesenchymal–epithelial transition, are sequentially coordinated. The sclerotome further forms vertebral columns and contributes to various other tissues, such as tendons, vessels (including the dorsal aorta), and even meninges. To understand the molecular mechanisms underlying these developmental processes, extensive studies have been conducted. These studies have demonstrated that a gradient of activities involving multiple signaling pathways specify the embryonic axis and induce cell-type-specific master transcription factors in a spatiotemporal manner. Moreover, applying the knowledge of mesoderm development, researchers have attempted to recapitulate the in vivo development processes in in vitro settings, using mouse and human PSCs. In this review, we summarize the state-of-the-art understanding of mesoderm development and in vitro modeling of mesoderm development using PSCs. We also discuss future perspectives on the use of PSCs to generate skeletal tissues for basic research and clinical applications. A deeper understanding of skeletal tissue development and improvements in tissue engineering will help pluripotent stem cell (PSC) therapies to reach their full potential for skeletal repair. The paraxial mesoderm, an embryonic germ layer, is crucial to the formation of healthy axial skeleton. Shoichiro Tani at the University of Tokyo, Japan, and co-workers reviewed current understanding of paraxial mesoderm development and studies involving in vitro PSC skeletal modeling. The formation of the paraxial mesoderm and associated connective tissues comprises multiple stages, and studies in vertebrate embryos have uncovered critical signaling pathways and cellular components important to PSC modeling. Although many individual cellular components can now be modeled, it remains challenging to recreate three-dimensional skeletal tissues. Such an achievement would facilitate a functioning model of bone metabolism, the next step in achieving skeletal regeneration.


Introduction
Although the continued elongation of life expectancy is generally a positive development for mankind, it also has certain unintended consequences, including the numerous social and medical problems associated with an aging society. Prominent among these problems are skeletal disorders, such as sarcopenia, osteoporosis, fragile bone fractures, and osteoarthritis. Bone formation and resorption, i.e., bone remodeling, dynamically occur in the whole body throughout the whole life span 1 . Although bone has the potential to repair itself, this ability is limited. Thus, trauma and skeletal disorders in elderly individuals cause critical unrepaired bone defects, resulting in a compromised quality of life 2 .
Because there are remaining unmet needs that cannot be solved by the conventional treatments of skeletal disorders, regenerative medicine has been attracting attention for the past several decades 3 . The emerging technology of induced pluripotent stem cells (PSCs) has expanded our strategies in both basic research and clinical applications, enabling the modeling of human development and genetic diseases, the establishment of drug screening systems, and clinical trials for the transplantation of differentiated cells into patients 4,5 . In this review, we focus on the paraxial mesoderm, which is one of the main origins of skeletal tissues. We summarize the latest understanding of mesoderm development, which is mainly based on studies in mammals and birds. We also introduce recent studies concerned with the in vitro modeling of mesoderm development using PSCs, and discuss future perspectives on the use of PSCs for basic research and clinical applications.

Overview of bone development
Bones originate from three types of cells: the paraxial mesoderm, lateral plate mesoderm, and neural crest 6 . The paraxial mesoderm gives rise to the axial skeleton. The lateral plate mesoderm gives rise to the appendicular skeleton. The neural crest originates from the ectoderm and gives rise to the craniofacial skeleton 6 .
There are two processes involved in bone formation: intramembranous ossification and endochondral ossification. While endochondral ossification occurs via cartilage formation and is followed by bone formation, intramembranous ossification occurs through direct transition from mesenchymal cells to bone-forming osteoblasts 6 . Although the neural crest forms most of the craniofacial skeleton by intramembranous ossification (with exceptions such as the skull base), most mesodermderived bones are formed through the endochondral ossification process 6 . Because of the large contribution of the mesoderm population to skeletal development, we focused on the paraxial mesoderm lineages in this review.

Development of the mesoderm
Overview of the mesodermal derivatives ( Fig. 1) The mesoderm initially forms in the primitive streak during gastrulation and later continues developing in the tail bud. First, the chordamesoderm forms a notochord that expands beneath the neural tube, as observed in human embryos 7 . The mesoderm lies along the notochord and divides into the paraxial mesoderm, intermediate mesoderm, and lateral plate mesoderm 8 . Mesodermal subtypes are specified along the mediolateral axis depending on the activities of BMPs 9 . A study with chick embryos demonstrated that Noggin, a BMP inhibitor expressed first in the notochord and then in the somatic mesoderm, creates the BMP gradient that specifies the mesodermal subtypes 10 . Another study with mouse embryos demonstrated that the notochord gives rise to the nucleus pulposus, which later forms vertebral discs 11 .
Paraxial mesoderm development is composed of several stages: presomitic mesoderm specification, somitogenesis, and somite specification 12 . Mature somites contain two major populations: the sclerotome and dermomyotome. The sclerotome gives rise to the vertebrae and associated ribs, tendons, and other tissues, such as vascular cells of the dorsal aorta, intervertebral blood vessels, and meninges 12,13 . The dermomyotome produces two components: the myotome and the dermatome. The myotome gives rise to the musculature of the back, rib cage, ventral body wall, and limbs. The dermatome gives rise to the dermis of the back, although the term dermomyotome is sometimes used to describe this region because a recent study showed that this central region of the dermomyotome also gave rise to muscles in chick embryos 14 .
The lateral plate mesoderm forms the splanchnic mesoderm, somatic mesoderm, and extraembryonic membranes, as evidenced by a study of chick embryos 15 . The splanchnic mesoderm gives rise to 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, with the exception of the muscles, which are derived from the dermomyotome 14,16 . The intermediate mesoderm forms the urogenital system, including the kidneys and gonads 8 .

Specification of the presomitic mesoderm
The early paraxial mesoderm is referred to as the presomitic mesoderm, and consists of bilateral streaks of mesenchymal cells adjacent to the notochord 17 . The presomitic mesoderm is derived from the primitive streak or neuromesodermal progenitors in the tail bud, as shown in studies with mouse and bird embryos 18,19 . In these steps, Noggin produced by the notochord protects the paraxial mesoderm from lateralization by BMPs produced by the intermediate mesoderm and lateral plate mesoderm 9,10 . This gradient is crucial for mesodermal cell fate determination 9,10 . When Noggin-expressing cells were implanted into the presumptive lateral plate region, somitic tissues were formed in the original lateral plate territory of chick embryos 10 . This demonstrates that the paraxial mesoderm and lateral plate mesoderm share common precursors in the primitive streak, and that the cell fate is plastic, depending on the gradients of BMP activity.
Wnt signaling is another crucial pathway in these processes. Wnt3a is widely expressed in the primitive streak and tail bud, as revealed in a study of mice 18 . Loss of function of Wnt3a or Ctnnb1, which is a gene encoding β-catenin, led to a loss of paraxial mesoderm progenitors and their derivatives, the presomitic mesoderm and somites in the mouse embryos 18 . Key transcriptional regulators in presomitic mesoderm specification, including Brachyury (T), Tbx6, and Mesogenin1 (Msgn1), are known to be downstream factors of Wnt signaling 20 .
T, the T-box transcription factor 21 , is expressed in the primitive streak, the tail bud, the early mesoderm and primitive ectoderm next to the primitive streak, the notochordal plate, and the notochord, as shown by studies of mouse embryos 22 . Classical genetic analyses using spontaneous mutant mice revealed that T is essential for mesoderm formation 22,23 . Loss of function of T caused a disturbance of primitive streak formation and insufficient mesoderm formation in mouse embryos 23 .
Tbx6, the T-box transcription factor, is expressed initially in the primitive streak, and later in the tail bud and presomitic mesoderm 24 . Studies with mouse embryos demonstrate that Tbx6 expression in the paraxial mesoderm is restricted to the presomitic mesoderm and is rapidly downregulated as somite forms 24 . Thus, the expression of Tbx6 overlaps that of T in the primitive streak and tail bud, although T is expressed at an earlier point in the primitive streak 24 . Loss of function of Tbx6 in mice resulted in conversion of the presumptive presomitic mesoderm into neural tissues 25 . In Tbx6-knockout mice, Sox2, a member of the Sry-related high mobility group box containing genes, was ectopically expressed in a presumptive presomitic mesoderm region; Sox2 was not expressed in that region in wild-type mice 25 . Given that Sox2 is known to be a critical factor for neuroectodermal development, this indicates that Tbx6 promotes presomitic mesoderm specification by repressing Sox2 expression and neural fates 25 .
Msgn1, the basic helix-loop-helix (bHLH) transcription factor, is expressed in the paraxial mesoderm from gastrulation until somite formation 26 . Overexpression of Msgn1 in mice expanded Tbx6-expressing regions through the trunk, resulting in expansion of the presomitic mesoderm region into the anterior 27 . The loss of function of Msgn1 reduced Tbx6 expression in the presomitic mesoderm 27 , and caused a complete failure of somite formation and segmentation of the body trunk and tail in mice 26 . These results indicate that Msgn1 is another determinant of presomitic mesoderm fate specification.

Specification of neuromesodermal progenitors
In the posterior region of the embryo, the paraxial mesoderm is derived from a cell population called neuromesodermal progenitors, which have the bipotential to differentiate into both mesodermal and ectodermal cell types, as demonstrated in a mouse study 28 . The cell fate is determined by several morphogens expressed along the anterior-posterior axis. Histological studies with mouse embryos demonstrated that Fgf8 and Wnt3a are highly expressed in the vertebrate tail bud, whereas the retinoic acid (RA) gradient is produced from the somite and the neural plate 29 . Fgf8 and Wnt3a were found to upregulate both Msgn1 and Tbx6, resulting in the promotion of presomitic mesoderm specification by suppressing Sox2 expression and neural cell fate specification in mouse embryos 25,30 .
Loss of function analysis with aldehyde dehydrogenase 1 family member A2 (Raldh2), which functions as a catalyst of RA synthesis, revealed that the Raldh2 −/− mouse embryos exhibited increased Fgf8 expression in the anterior part of the embryo 29 . The deficient mice also displayed impaired formation of somites with a decrease in Sox2-positive and Sox1-positive neuroectodermal progeny, and an increase in Tbx6-positive presomitic mesodermal progeny 29 . The phenotype of impaired somite formation in Raldh2 −/− mice was rescued by treatment, with an FGF receptor antagonist 29 . These findings indicate that RA directly represses the expression of Fgf8 and Tbx6, resulting in cell fate specification to neural cell types with upregulation of Sox2. They further suggest that the balance of signaling activities between these opposing morphogens is a key determinant of neural and mesodermal cell fates 29,30 .

Somitogenesis
The somite is derived from the anterior presomitic mesoderm through a series of dynamic morphogenetic events that involve cyclical signaling. The periodicity of somitomere formation is produced by the segmentation clock that operates in the presomitic mesoderm. A study with mouse embryos demonstrated that this segmental prepattern is defined at the "determination front", which creates future somitic boundaries 31 . This process proceeds according to a "clock and wavefront model": a clock determines the time, and a wavefront determines the place for the segmentation 32 . Mesenchymal-epithelial transition (MET) is another essential process for somitogenesis, as it is involved in epithelial somite formation 33 . Studies with mouse embryos demonstrate that during these processes, Msgn1 is downregulated, but several other markers, including Mesp2, Paraxis, Pax3, Foxc1/2, and Meox1/2, are upregulated 9,34-36 .

Segmentation clock
The major signaling pathways in the segmentation clock are the Notch, Wnt/b-catenin, and FGF pathways, which integrate to form a molecular network and generate a traveling wave of gene expression along the embryonic axis. Global gene expression analysis in mice revealed that Notch-and FGF-related cyclic genes oscillated mostly in the opposite phase of the Wnt-cyclic genes, suggesting crosstalk between these signaling pathways 37,38 . In mice, the clock in each region of the presomitic mesoderm is well understood to be a negative feedback mechanism centered on the activities of the transcription factor Hes7 39,40 . Hes7 is initially activated by FGF signaling, and then it is controlled by Notch activity 40 . Hes7 suppresses its own transcription to generate an oscillating pattern of expression 39 . Notch signaling activates mesodermal posterior 2 (Mesp2), a bHLH transcription factor, which suppresses the Notch pathway via lunatic fringe (L-fng) 41 . Thus, Notch activity oscillates in the presomitic mesoderm as a "Notch clock oscillator" 42 . FGF signaling also oscillates via the phosphorylation of extracellular signal-regulated kinase, an FGF signaling downstream molecule, which was also demonstrated in mice 43 .

Determination front and segmentation
Mesp2 is a master regulator of the onset of segmentation 35 . Mesp2 is expressed at the initial stage of segmentation in the presomitic mesoderm 35 , and its expression is restricted to the rostral compartment by the oscillators of the Notch and FGF pathways, as evidenced by a study of mouse embryos 43 . Mesp2 expression is activated by the Notch pathway in the anterior part of the presumptive somite 44 , whereas it is suppressed by FGF signaling in the posterior part, resulting in the formation of anterior and posterior borders 35,45 . This model is supported by several lines of evidence. First, Mesp2 expression was strongly suppressed in Notch mutant mouse embryos, such as Dll1-null and RBP-jk-null embryos 40 . Second, the Mesp2expressing domain was shifted into the posterior presomitic mesoderm in the absence of FGF signaling in mouse embryos 43 .
As described earlier, Mesp2 plays a crucial role in the formation of the border of somite segments and in establishing the rostrocaudal patterning of each somite 35,42 . Mesp2-null mouse embryos were shown to have a nonsegmented somite with completely caudalized somite derivatives 35 . Notch activity is required for caudalization of the somite, since its absence in the caudal compartment resulted in a rostralized phenotype in mice 46 . With respect to the mechanism of Mesp2mediated somite patterning, a study in mouse embryos demonstrated that Mesp2 suppresses Notch activity in the rostral compartment by destabilizing Mastermind-like 1, one of the core components of the nuclear Notch intercellular domain complex 47 . This leads to rostrocaudal formation via differential Notch activity 47 .
A study with mouse embryos demonstrated that Mesp2 activates its target Ripply2, which suppresses Mesp2 expression via the inhibition of Tbx6 in a negative feedback loop, leading to formation of the next segmental border 48 .
Another research in mice has demonstrated that Mesp2 also upregulates Eph in the anterior portion of somitomeres, which is followed by the upregulation of ephrin in the opposing posterior half of the more anterior somitomere 49 . Then, separation of the somite from the anterior end of the presomitic mesoderm occurs at the border between ephrin-and Eph-expressing cells 49 . This pattern is repeated sequentially in the process of somitogenesis 42 .

Mesenchymal-epithelial transition
MET is necessary to form the epithelial layer of the somite during somitogenesis, since the presomitic mesoderm is composed of only mesenchymal cells. A study with mouse embryos demonstrated that without MET, neither the epithelial somite nor the dermomyotome can properly form; the absence of MET leads to abnormalities of the axial skeleton, such as numerous patterning defects of the musculature in the axial skeleton, limbs, and body wall 33 . During MET in the future somatic boundaries, the outer epithelial layer assumes apical-basal polarity and expresses N-cadherin, β-catenin, and F-actin in apical adherens junctions 50 . This process is intimately regulated in a spatial and temporal manner along the anterior-posterior axis, as evidenced by a study in birds 50 .
Paraxis, a bHLH transcription factor, is expressed in the presomitic mesoderm and somites. Paraxis is indispensable for epithelialization in the developmental process of somite 33 . Paraxis-null mice had no epithelial somites, although the somites were segmented into loose mesenchymal units of approximately the correct size and periodicity as somites in the paraxial mesoderm 33 . The mutants also displayed skeletal abnormalities, such as caudal agenesis 33 . These facts suggest that Paraxis is required for the formation of the epithelial somite but not for segmentation of the paraxial mesoderm.
Somitic MET in the mouse and chick paraxial mesoderm is dependent on Wnt signaling from the overlying surface ectoderm 51,52 . Although segmentation of the paraxial mesoderm was maintained even with the removal of the surface ectoderm, somitic MET did not occur in mice 52 . Loss of Wnt signaling caused loss of Paraxis expression and somitic MET in mice 52 . In addition, Wnt6 expression in the surface ectoderm induced somitic MET, and ectopic Wnt6 expression substituted for a lack of surface ectoderm and β-catenin dependent processes in chick embryos 53 . Moreover, forced expression of paraxis rescued somite epithelialization in the absence of Wnt signaling in chick embryos 54 . On the other hand, paraxis −/− mouse embryos showed decreased expression of downstream genes in the Wnt and Notch signaling pathways, as well as decreased expression of Meox1/2 and Pax1, which are required for proper somite formation and specification, respectively 55 . These facts suggest that paraxis participates in Wnt signaling-mediated epithelialization in PSM 55 .
A previous report demonstrated that Meox1 and Meox2 are coexpressed in the epithelial somites, sclerotome, and limb buds, whereas the dermomyotome only expresses Meox1 56 . Meox1-null mutant mice had defects in axial skeletal development but not muscle development 57 , whereas Meox2-null mutant mice displayed a lack of limb muscles, as well as a generally reduced muscle mass, but no abnormality in the axial skeleton 56 . These results suggest that Meox1 substitutes for Meox2 in the sclerotome, but not the myotome and that Meox2 compensates for the lack of Meox1 in the myotome, but not the sclerotome.
Foxc1 and Foxc2, members of the winged helix transcription factors, are expressed in many tissues forming the somites, head mesoderm, and endothelial and mesenchymal cells of the developing heart and blood vessels, as shown in mouse embryos 36 . Mouse embryos lacking both Foxc1 and Foxc2 had no epithelial somite or morphological segmentation of the paraxial mesoderm 36 . Paraxis was undetectable in the presomitic mesoderm and somite region in the mutant, suggesting that Foxc1 and Foxc2 are upstream of paraxis during the somite formation processes 36 . Another report showed that the paraxial mesoderm in Foxc1/2 mutant mice was respecified into the intermediate mesoderm, which expresses Pax2; Pax2 is a major transcription factor in the intermediate mesoderm 58 . However, no significant change was detected in the expression of either Bmp4 or Noggin, which can regulate mesodermal fates 58 . In addition, misexpression of Foxc1 or Foxc2 in the presumptive intermediate mesoderm of mouse embryos resulted in conversion of the cell fates from the intermediate to paraxial mesoderm and somite, but not to the lateral plate mesoderm 58 . These results suggest plasticity of the cell fates between the intermediate mesoderm and the paraxial mesoderm, with Foxc1 and Foxc2 contributing to somite segmentation in the paraxial mesoderm. Taken together, these results show that Foxc1 and Foxc2 are essential for paraxial mesoderm differentiation and fate determination.

Specification of the somite
The sclerotome is derived from a ventromedial part of the somite and is formed by epithelial-mesenchymal transition, whereas the dermomyotome is derived from the epithelial dorsolateral part of the somite 59 . The sclerotome is a mesenchymal tissue in which key regulators, including Pax1, Pax9, Nkx3.2 (Bapx1), and Sox9, are specifically expressed 60 . On the other hand, Pax3 and Myf5 are upstream factors of MyoD that are involved in muscle development, as evidenced by studies of mouse embryos 61 . Pax3 is initially expressed in the forming somite, but its expression is downregulated during specification in the sclerotome, whereas it remains expressed in the dermomyotome 62 .
Sonic hedgehog (Shh) is secreted from the notochord and floor plate of the neural tube 63 . Studies with mouse and bird embryos demonstrate that Shh functions as a crucial molecule in sclerotome formation 62,63 . Shh mutant mice lacked vertebral columns, and only a few rudimentary rib cartilages were formed 64 . Mouse embryos with deletion of both Gli2 and Gli3, the downstream factors of Shh, exhibited severely reduced expression of Pax1 and Pax9; further, Sox9 expression was undetectable in somites 65 . However, Shh-null mice still showed transient Pax1 expression. These results imply that Shh is a crucial inducer for Pax1, Pax9, and Sox9 through Gli2 and Gli3 65 , although other signals may also be involved in the induction 64 . Studies in mice and birds showed that Noggin was expressed in the node, notochord, and dorsal somite and that it inhibited BMP4 activity during sclerotome specification 63,66 . Shh also competed with Wnt signaling from the roof plate and surface ectoderm, and Wnt functioned in these locations to maintain the somite epithelial state and induce the dermomyotome in chick embryos 62 . Collectively, these results show that high levels of Shh activation and low levels of Wnt and BMP signaling are required to determine sclerotomal fate.
Pax1 and Pax9, transcription factors of the Pax family, are specifically expressed in the large part of sclerotomes. Homozygous Pax1-null newborn mice showed severe abnormalities in the axial skeleton 67 . On the other hand, homozygous Pax9 mutant mice showed skeletal defects in the limbs and the skull, but exhibited no obvious defects in the axial skeleton 68 . In addition, Pax1/Pax9 double mutant mice showed much more severe phenotypes than Pax1 single homozygous mutants; Pax1/Pax9 double mutants completely lack tissues derived from the medial part of the sclerotome, such as the vertebral bodies, the intervertebral discs, and the proximal parts of the ribs 69 . The condensation of the ventromedial sclerotome around the notochord was also prevented in the double mutants, resulting in an impairment of chondrogenesis and vertebral formation 69 . Moreover, a rescue experiment in mice showed that Pax1 compensated for Pax9 function, whereas Pax9 did not compensate for Pax1 function 69 .
Nkx3.2 (Bapx1) is a homeobox-containing transcription factor expressed in the sclerotome during early embryonic mouse development 70 . A targeted disruption of the Nkx3.2 gene in mice resulted in lethal skeletal dysplasia with abnormal development of the vertebral column and craniofacial bones 70 , and a failure of cartilage development with decreased expression of Sox9 and type II collagen 69 . In addition, condensation of sclerotomal cells in the vertebral anlagen around the notochord was completely lost during early embryogenesis in Nkx3.2 mutant mice 71 . Analyses of mouse embryos with a double mutation of Pax1/Pax9 revealed that Nkx3.2 expression in the sclerotome required the presence of both Pax1 and Pax9 in a dose-dependent manner 72 . In the same study, Nkx3.2 was found to be induced by the overexpression of Pax1 even without Shh. Furthermore, Pax1 and Pax9 transactivated the Nkx3.2 promotor through direct interaction with DNA, indicating that Nkx3.2 is a direct target of Pax1 and Pax9 72 . These results suggest that Nkx3.2 functions downstream of Pax1 and Pax9, and plays crucial roles in chondrogenesis and vertebral development 71 .
Meox1 and Meox2 are essential for somite formation, as described above, and they contribute to somite development. Meox1/2 double mutant mice lacked Pax1 expression in the paraxial mesoderm, and they exhibited an attenuation of Pax3 and Pax7 expression, resulting in a failure of dermomyotome differentiation 34 . These deficiencies led to abnormalities in the axial skeleton, such as a lack of normal vertebral columns 34 , as well as major defects in the development of the somite-derived skeletal musculature 34 . In addition, the expression of Nkx3.2 in mice with mutations in both Meox1 and Meox2 was much more severe than it was in mice with a single Meox1 mutation 34 . These results suggest that Meox1 and Meox2 coordinate with and compensate for each other during sclerotome and dermomyotome development.

Subsets of the sclerotome
Because of its location, the sclerotome is in contact with various cell populations that produce different signaling molecules, resulting in the establishment of various subpopulations along the ventral-dorsal, medial-lateral, and anterior-posterior axes 12 . These conclusions were drawn mainly from studies on birds.
The center part of the sclerotome forms the mesenchymal core of the epithelial somite, which mainly contributes to the formation of intervertebral discs and joints of the vertebral column 73 . Thus, Christ named this sclerotomal subdomain the "arthrotome" 12 . The dorsal part of the sclerotome is located close to the myotome. FGF ligands, such as Fgf8, are secreted from the myotome and induce scleraxis expression 74 . These signaling activities give rise to the syndetome population, which is a precursor of the vertebral tendons and ligaments 74 . In the dorsomedial and lateral parts of the sclerotome, Pax1 expression is downregulated by BMP4 from the dorsal neural tube, intermediate mesoderm, and lateral plate mesoderm, leading to the development that is independent of notochordal signals 75,76 . The dorsomedial part of the sclerotome forms the spine and arch, whereas the lateral part of the sclerotome forms the distal ribs 75 . This depends on Bmp4 secreted from the dorsal neural tube and lateral plate mesoderm 75 . Moreover, these two parts are devoid of Pax1 expression 75 . The medial sclerotome located adjacent to the lateral surface of the neural tube was identified, as a population giving rise to the blood vessels and meninges of the spinal cord 77 . In response to signaling molecules secreted from the notochord, the ventromedial sclerotome strongly expresses Pax1 and migrates medially toward the notochord. These cells form the perinotochordal tube, which gives rise to the vertebral bodies and intervertebral discs 78 . Last, the ventral-posterior sclerotome with endothelial precursor potential was recently named the endotome 13,79 . This part of the sclerotome migrates and differentiates into vascular cells of the dorsal aorta and intervertebral blood vessels, as shown in studies with chicks 79,80 .

Induction of mesodermal populations in vitro (Figs. 2 and 3)
By utilizing knowledge of mesoderm development, researchers have attempted to recapitulate the in vivo development processes using in vitro settings. Next, we will summarize key reports about these in vitro models, and there will be a particular focus on studies, using PSCs from mice and humans.

Presomitic mesoderm induction from PSCs via the primitive streak and neuromesodermal progenitor
Following the revelation of the pivotal roles of Wnt signaling in development, previous reports showed that Wnt activators, such as CHIR99021, a GSK3 inhibitor, induced cell types of the primitive streak and neuromesodermal progenitor from mouse and human PSCs in vitro 30,[81][82][83][84][85] . In addition, several papers showed that combinatorial activation of Wnt and FGF signaling enhanced the induction efficiency of the primitive streak and neuromesodermal progenitors [81][82][83][85][86][87] . The requirements for TGFβ signaling differ among the differentiation stages. Nodal, a TGFβ activator, has been shown to enhance induction of the anterior primitive streak, which gives rise to neuromesodermal progenitors 27,82,85,88 . In contrast, the combination of a TGFβ inhibitor and a Wnt activator enhanced induction of the presomitic mesoderm, which possibly occurred through induction of the primitive streak 83,86 . Given that Wnt activation may be a trigger for endogenous FGF and Nodal activation in vitro 83,85,89,90 , Wnt activation is likely sufficient. FGF and Nodal may accelerate Wnt-mediated cell fate commitment or differentiation in the primitive streak and neuromesodermal progenitors.
In the differentiation from the primitive streak/neuromesodermal progenitor into the presomitic mesoderm, activation of Wnt signaling with CHIR was reported to promote paraxial mesoderm cell fate, and to upregulate Tbx6 and Msgn1 expression 82,83,86,91 . The combination of a BMP inhibitor, LDN-193189, and CHIR improved the induction efficiency of the paraxial mesoderm 82,83,[85][86][87]89 , possibly because the fate of mesodermal progenitors can be changed from the paraxial mesoderm to the lateral plate mesoderm according to the gradient of BMP activity. Conversely, exogenous BMP activation promoted lateral plate mesoderm fate, but suppressed paraxial mesoderm fate 82 .
One difference has been observed between humans and other model organisms during paraxial mesoderm differentiation in vivo: TGFβ signaling was downregulated during presomitic mesoderm induction in humans, but not in animals 83 . Consistent with this finding, some reports showed that the combination of a TGFβ inhibitor with a BMP inhibitor increased the efficacy of the induction from human PSCs to the paraxial mesoderm or anterior presomitic mesoderm/somite 82,83,[85][86][87] .

Somite induction and sclerotome specification
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. Some reports have tried to recapitulate this signaling transition in vitro by removing Wnt and FGF/ ERK signaling effects 82,85,89 . However, activation of Wnt signaling, regardless of the presence or absence of FGF, was sufficient to recapitulate this transition in both mouse and human PSCs 86,91 . Molecular patterns similar to the in vivo segmentation program were observed in hPSCs 82,86,89,91 : downregulation of Msgn1, transient upregulation of Mesp2, and induction of Pax3 82, [85][86][87]89,91 . These findings suggest that once the cells respond to those key signaling factors, they are set to undergo the sequences of differentiation in a cell autonomous manner.
The BMP gradient is a critical factor in the determination of the mediolateral axis during mesoderm development and in somite specification 10,63 . In line with this, some reports showed that a BMP inhibitor improved somite-sclerotome induction, possibly by both promoting medial fate and protecting the mesodermal population from lateralization 10,[82][83][84][85][86] .
For sclerotome induction, the activation of Hh signaling seems the most important 62,64,65 . Although various protocols were established for sclerotome induction, all of them used Hh agonists [82][83][84]86 . This is due to the indispensable roles of Shh in the developmental stage, which were demonstrated by several studies in mice and birds: Shh is produced by the notochord and floor plate; it is essential for somite specification into the sclerotome in vivo 62,63 . In addition, the combination of an Hh agonist with a Wnt inhibitor recapitulated the competing relationship of Shh and Wnt 62,82 ; the combination of an Hh agonist with a BMP inhibitor recapitulated Noggin activity to inhibit BMP signals from the intermediate mesoderm or lateral plate mesoderm, and to promote sclerotome differentiation [83][84][85][86] . In any case, the cell fate commitment to the sclerotome was achieved by the expression of specific transcription factors, such as PAX1, PAX9, SOX9, and NKX3.2 [82][83][84][85][86] . Based on what has been learned about mouse and bird embryo development 63,66 , the combination of Hh signaling and BMP inhibition may be sufficient for sclerotome induction.
Two concerns that remain to be resolved with respect to in vitro differentiation are (1) heterogeneity of the induced cells and (2) differences in differentiation protocols among species. First, as described above, although several protocols were established to induce paraxial mesoderm derivatives from PSCs, there is no protocol that can exclusively induce generation of the desired population and exclude unwanted populations. The remaining undifferentiated cells or wrongly differentiated cell populations can cause oncogenicity, which leads to considerable risk in a clinical context. Heterogeneous populations may also cause less efficiency of the induction of wanted cell populations. Thus, manipulating the activities of combinatorial signaling pathways in a stepwise manner will be required not only for proper modeling of development, but also for clinical usage. Second, recent studies have shown divergence between the developmental process of human and mouse embryos 83,92 . This is consistent with in vitro studies that show that an optimized differentiation protocol to induce a cell type in mice is not always optimal for induction of the same cell type in humans; different combinations of The paraxial mesoderm forms in the primitive streak and from NMPs. Through the elongation of the embryo axis, the PSM is located from posterior to anterior according to the body axis produced by competing gradients of Wnt, FGF, and RA. The anterior PSM forms somite at the determination front depending on the FGF gradient and Notch signaling. b A flowchart of the in vitro differentiation process from PSCs to sclerotomes, with critical inducers and specific markers in each stage. In the future, a next step will be to differentiate the induced sclerotome into sclerotome derivatives, such as bone, tendon, and vessels. RA retinoic acid, NMP neuromesodermal progenitor, PS primitive streak, aPS anterior primitive streak, pPSM posterior presomitic mesoderm, aPSM anterior presomitic mesoderm, nt neural tube, n notochord, LPM lateral plate mesoderm. Asterisks in each of the differentiation stages in a correspond to the stages highlighted by asterisks in b. signaling activities are required for different species to some extent 83,93 . Thus, further comparative analysis among different species will provide clues for establishing an optimized protocol for the differentiation of human PSCs into specific cell types.

Future perspectives
As we described in this review, skeletal elements are derived from the neural crest, lateral plate mesoderm, and paraxial mesoderm. How does the difference in origin affect the properties of skeletal cell types in terms of their differentiation efficacy and regenerative capacity? Because protocols for the differentiation of skeletal cells through the neural crest and lateral plate mesoderm were recently proposed 94,95 , it may be possible to perform such comparative analyses in the future. This will provide insights into the origin-distinct process of skeletal development and will generate a strategy of skeletal regeneration.
As we have shown in the above review, the current stepwise protocols enable the generation of sclerotome cell types. Recent studies have further established protocols that recapitulate the segmentation clock in mouse gastruloids 96 , and in the somite induced from human and mouse PSCs 85,87 . The next step will be further optimization of these protocols to induce sclerotome derivatives, including bone, cartilage, and tendon. Some groups have proposed protocols to induce osteoblasts, chondrocytes, and tenocytes from mouse or human PSCs via the sclerotome 82,83,85,86 .
Previous reports have established protocols for directing the differentiation of PSCs toward three-dimensional organoids of various tissues, such as retina 97 and limb bud 95 . In addition, some of these protocols have already been applied in clinical trials 5 . In contrast, it remains challenging to generate three-dimensional skeletal tissues even though we can induce every cell component of the tissue, including osteoblasts and chondrocytes. To do this, we may need not only a deeper understanding of skeletal development, but also tissue engineering approaches that take advantage of biomaterials. We recently reported that differentiation of osteoblasts from mouse ES cells with a three-dimensional atelocollagen scaffold enhanced osteoblast maturation 98 . Several biomaterials and threedimensional culture systems have been developed 95,99 , although further optimization is needed. In addition to enabling the modeling of skeletal development, the threedimensional system will provide a model that recapitulates bone metabolism under physiological conditions: remodeling by coupling of bone-forming osteoblasts and boneresorbing osteoclasts, and responses of the cells upon mechanical stresses. The three-dimensional system will thus be useful for drug screening and disease modeling.
For clinical settings, the transplantation of human PSCderived tissues is one of the goals in translational studies, but several issues remain to be solved-namely, the quality, quantity, safety, and availability of the resulting tissues. Thus, when we implant tissues derived from PSCs into patients, the tissues that are produced must meet several criteria: they must have sufficient function, sufficient size or volume, little or no immune response, and no oncogenicity. Researchers have tried to satisfy these criteria by using iPS cells derived from the patients themselves or from donors with the same haplotype of HLA 100 , reducing or removing xenogeneic components, such as fetal bovine serum and Matrigel 101 , and removing undifferentiated iPS cells by a specific reagent 102 . Further clinical studies, especially those focusing on safety, will be needed in the future.
In addition, several attempts have been made to utilize human PSCs in clinical settings. Disease modeling and drug discovery are attractive applications for the treatment of various diseases. Recent reports have presented various models of skeletal diseases using patient-derived iPS cells, such as models of fibrodysplasia ossificans progressiva, FGFR3-related skeletal dysplasia, and osteogenesis imperfecta. The pathological features of these diseases were recapitulated well by chondrocytes and osteoblasts differentiated from patient-derived iPS cells; accordingly, these differentiation platforms have been utilized for drug screening 4,86,103,104 . In addition, some reports have demonstrated human PSC-derived skeletal stem/progenitor cells and mesenchymal stem/stromal cells that have multipotent capacity to differentiate into several lineages of skeletal components 105,106 . These cells may be valuable cell sources for cell therapies for skeletal diseases. Taken together, these findings show that the potential applications for human PSCs have been expanded in both basic science and clinical medicine. An improved understanding of the development of paraxial mesoderm derivatives and further optimization of the protocols for the differentiation of human PSCs into skeletal tissues will provide both insights into human skeletal development and valuable tools for clinical settings.