Are osteoblasts multiple cell types? A new diversity in skeletal stem cells and their derivatives

Abstract

Only in the past decade have skeletal stem cells (SSCs), a cell type displaying formal evidence of stemness and serving as the ultimate origin of mature skeletal cell types such as osteoblasts, been defined. Here, we discuss a pair of recent reports that identify that SSCs do not represent a single cell type, but rather a family of related cells that each have characteristic anatomic locations and distinct functions tailored to the physiology of those sites. The distinct functional properties of these SSCs in turn provide a basis for the diseases of their respective locations. This concept emerges from one report identifying a distinct vertebral skeletal stem cell driving the high rate of breast cancer metastasis to the spine over other skeletal sites and a report identifying 2 SSCs in the calvaria that interact to mediate both physiologic calvarial mineralization and pathologic calvarial suture fusion in craniosynostosis. Despite displaying functional differences, these SSCs are each united by shared features including a shared series of surface markers and parallel differentiation hierarchies. We propose that this diversity at the level of SSCs in turn translates into a similar diversity at the level of mature skeletal cell types, including osteoblasts, with osteoblasts derived from different SSCs each displaying different functional and transcriptional characteristics reflecting their cell of origin. In this model, osteoblasts would represent not a single cell type, but rather a family of related cells each with distinct functions, paralleling the functional diversity in SSCs.

Osteoblasts and other mature skeletal cell types have a limited lifespan, necessitating their continual replacement from a progenitor cell source and the lifelong maintenance of this progenitor. Although many basic concepts in skeletal cellular dynamics remain unresolved, the ultimate source of the mature skeletal cell types that mediate initial bone modeling and remodeling are a class of cells termed skeletal stem cells (SSCs), long-lived, typically slowly cycling cell types that sit at the apex of their differentiation hierarchy. In addition to SSCs playing a fundamental role in generating and maintaining skeletal tissue, increasing evidence indicates that SSC dysfunction is a major basis for many skeletal diseases, including aging-associated osteopenia, impaired fracture repair, and the premature skull fusion seen in craniosynostosis.^1-3 In light of this, identifying and studying SSCs is a central concern for skeletal biology.

Although progenitors for skeletal cells have long been studied, many early studies focused on heterogeneous populations of mixed cell types that lacked formally demonstrated evidence of stemness, such as so-called “mesenchymal stem cells.” Thus, the identity of formally defined SSCs was unclear until a series of reports from Chan and colleagues fractionated skeletal cells to identify specific populations that displayed formal evidence of stemness, including in transplantation-based organoid assays.^4,5 These reports established several important principles for identifying SSCs, including the importance of simultaneous application of a panel of markers to define and study discrete cell types. This represented an important advance over literature that solely utilized cre-based systems, where the cre enzyme is expressed from a locus driving expression in one or more subsets of skeletal cells, with examples including Osterix, Prrx1, Col2a1, or Gli1-cre mouse lines. The cre enzyme then mediates genomic recombination that permanently “turns on” a genetic reporter, typically encoding for a fluorescent reporter protein in the GFP family, in not only cells expressing the cre enzyme, but also in all of the cells produced by the cre-expressing cells. Although cre-based systems remain an essential mainline tool for skeletal biology, their limitation lies in their tendency to label both directly targeted cells and also their derivatives, thus typically can only identify lineages of cells containing multiple differentiation states unless supplementary fractionation approaches are applied in tandem. Additionally, most genes are expressed in multiple cells types, including cell types that do not share a lineage relationship. This by extension results in most cre lines co-labeling multiple lineages thereby potentially obscuring the lineage relationship among labeled cells.

Additional fractionation is typically needed to arrive at the differentiation states or specific cell types within that lineage, though inducible single gene cre markers with extremely restricted expression like PTHrP-creER likely represent an exception, as its activity is restricted to a specific population of cells in the resting zone of the growth plate.⁶ Fractionation and transplantation-based approaches also enabled experimental demonstration of the differentiation hierarchy emanating from each cell type under study and therefore direct determination of which populations display increasingly restricted differentiation potential and are therefore likely to represent non-stem populations. Further, by developing methods for prospective isolation of SSCs, the Chan et al. reports on murine and human SSCs newly enabled investigation into SSC function using in vivo and ex vivo approaches. Ideally, cre-based lineage tracing systems are used in concert with fractionation and transplantation-based methods, with the fractionation-based methods allowing definition of specific cell types and their associated differentiation hierarchy, self-renewal, and function during in vivo tissue formation assays. Meanwhile, cre or other lineage tracing systems allow for complimentary determination of cell fate in an unperturbed system.

Since these reports, it has become increasingly evident that there are additional layers complexity to SSCs, with a range of recent studies indicating that instead of SSCs representing a single monolithic cell type, they are rather a family of related cell types that are distinguished by each having distinct anatomic locations and functional specialization appropriate for those locations. At the same time, these distinct SSCs also share several features among sites, including a shared set of surface markers, parallel differentiation hierarchy, and partially shared transcriptional profiles. These SSC types include growth plate resident PTHrP+ stem cells, periosteal Cathepsin K (CTSK)-lineage stem cells, calvarial CTSK-lineage stem cells, and other stem cell types.^6,7 Recently, we have added to this diversity by identifying a vertebral SSC that is distinct from long bone SSCs and a DDR2 + SSC that works alongside CTSK-lineage SSCs in forming the calvaria.^3,8 In this Perspective, we focus on summarizing how these 2 recent reports, together with previous literature, build a new model for skeletal biology, whereby multiple site-restricted SSCs account for the physiology and pathology associated with distinct skeletal sites.

Diversity within bones: A periosteal SSC

Previously, we identified a periosteal CTSK-lineage SSC present in long bones.⁷ Separating this CTSK-lineage periosteal SSC from endosteal SSCs uncovered a number of shared and distinct elements of these 2 pools of SSCs. First, both endosteal SSCs and periosteal CTSK-lineage SSCs shared a set of defining surface markers matching those previously reported by Chan and colleagues (Lin-CD200 + CD105-Thy-6C3-CD51+).⁴ Within the pool of cells expressing these markers, periosteal CTSK-lineage SSCs were then further discriminated from endosteal cells in total bone digests by a cre-based CTSK-lineage reporter (Figure 1). Whereas the absence of CTSK-lineage status was primarily used to identify endosteal cells in this report, subsequent work has found that pulsed FGFR3-creERT labels endosteal cells, including stem cells, identifying a positive marker of endosteal cells that would enable a more robust comparison between endosteal and periosteal SSC lineages in subsequent studies.⁹ These periosteal and endosteal SSCs both separately display stemness properties in transplantation assays and further display clear functional differences as the CTSK-lineage periosteal SSCs mediate intramembranous ossification, whereas the non-CTSK lineage SSCs mediated endochondral ossification. Altogether, this work set the stage for the studies that followed by establishing that different SSCs populate different regions of bone, with each of these SSC lineages displaying distinct functional specializations that enable the local physiology of their respective regions.

Figure 1.

A model of skeletal stem cell diversity. Long bone harbors a diverse array of skeletal stem cells (SSCs), including periosteal CTSK+ SSCs, endosteal SSCs, and growth plate-resident PTHrP+ SSCs. In turn, a Pax1+, Zic1+ stem cell is present in the vertebral endplate cartilage and generates vertebral osteoblasts and other mature skeletal cell types. This Pax1+, Zic1+ vertebral SSC-lineage is a key driver of the high rate of solid tumor metastasis to the spine, in part by secreting higher levels of a new metastatic tropism factor, MFGE8, than their long bone counterparts.

Diversity between bones: A vertebral SSC

If the endosteal and periosteal compartments within a single bone are served by distinct stem cells, then by extension perhaps different bones are formed by distinct stem cells. Vertebral bone was the most attractive region to look for this putative site-specific stem cell as vertebral bone has a distinct evolutionary history from that of long bones, appearing in early jawed fishes as a signature feature of vertebrate life long before the evolutionary appearance of long bones.¹⁰ Additionally, vertebral bone is the product of a very different developmental process than long bones, emerging from a sequence of somite formation and segmentation that has no parallels in the limbs. We hypothesized that the distinct evolutionary and developmental history of vertebral bone would ultimately translate into vertebral bone being both initially formed by and later postnatally maintained by a distinct SSC. Based on the premise that endosteal and periosteal SSCs shared a common immunophenotype based on the Chan et al. markers,^4,7 we considered that this same immunophenotype may putatively mark vertebral SSC candidates. Based on this premise, expression profiling was used to find differentially expressed genes when comparing these candidate vertebral SSCs to a parallel population of established long bone SSCs, finding that vertebral cells postnatally carried forward a number of transcripts best known for their role in somatic development, including genes like Pax1 and Zic1 that have been implicated in disorders of vertebral development in humans. For instance, Pax1 has long been known as important for vertebrae since the identification in 1947 of undulated mice, a mutant mouse line associated with impaired vertebral mineralization and distortion of vertebral column architecture. However, positional cloning of the causative Pax1 mutation in this line didn’t occur until 1988.^11,12 In humans, PAX1 mutations have been associated with both vertebral defects in Klippel-Feil syndrome and with adolescent idiopathic scoliosis.^13,14 Additionally, it is of particular interest that mice with dual deficiency of Zic1 and Pax1 display spine defects very similar to those seen in mice where Zic1 + vSSCs are genetically unable to produce osteoblasts, which raises the possibility that Zic1 and Pax1 cooperate in enforcing vSSC identity and function.¹⁵ Thus, adult vertebral SSCs carry the transcriptional imprint of their embryonic origin, though further study is needed to clarify to what degree phenotypes associated with these somitic genes reflect pre-skeletal somitic patterning versus the direct function of these genes in the vSSC lineage.

Lineage tracing with a Pax1-creERT and Zic1-cre together with direct Zic1 and Pax1 expression analysis found that Pax1-positive cells were a rare population residing in the vertebral endplate cartilage adjacent to the intervertebral disc that co-express Zic1 (Figure 1). With differentiation, these Pax1+ vSSCs silence expression of Pax1 but maintain expression of Zic1. Thus, Pax1 appears to be selective to vSSCs, whereas Zic1 is expressed in the entire vSSC lineage. Accordingly, Zic1-cre labels a wide group of vertebral cell types, including endplate cartilage, osteoblasts, and marrow adipocytes. In line with this, Zic1 + vSSCs display a full formal set of stemness activities, including the capacity for serial bone organoid formation after transplantation, self-renewal both in transplantation systems, and sitting at the apex of their differentiation hierarchy by being able to reconstitute their entire lineage post transplantation. Thus, in both their surface markers and differentiation hierarchy, vSSCs parallel SSC populations in long bones. In a model where different skeletal sites are formed and maintained by different stem cells, it should follow that the signature pathologies of these sites should be closely linked to the properties of these local stem cells. Indeed, the Zic1+ vertebral SSC appears to be a key determinant of the signature diseases of the spine. For instance, it is long appreciated that many solid tumors, especially breast and prostate cancer, display a strong metastatic preference for the spine over other skeletal sites. Interest in the mechanisms underlying this phenomenon is longstanding and notably dates back to the work of Oscar Batson in 1940.^16,17 To understand the metastatic preference of prostate cancer for the spine, he traced the venous drainage of the prostate in dogs. While under normal conditions, little flux from the prostate into the spine was observed, when he cinched a surgical towel around the dog’s waist to mimic a Valsalva state occurring when coughing or defecating, some retrograde blood flow into the vertebrae was observed. Later, it was found that this occurs due to a pressure-based shunt through a vascular plexus investing the vertebral body that eventually wound up bearing his name, Batson’s plexus.¹⁸ This vascular-centric view of skeletal metastatic tropism has remained the dominant explanation that dominated medical thought for nearly 80 yr.

However, there are several reasons to revisit this model. First, it seems implausible that only brief moments when a Valsalva maneuver occurs in everyday life are relevant to metastatic seeding, while that vast majority of normal anteriorgrade blood flow is irrelevant. Second, at least in the case of prostate cancer, studies using somatic mutations to infer the history of cells in spine metastases raise the possibility that some skeletal metastases may be secondary metastases from other distant sites, rendering the venous drainage of the prostate irrelevant.¹⁹ However, it is noted that this may not hold true for breast cancer.²⁰ In line with this, despite recapitulating a spine metastatic preference with multiple breast cancer cell lines in mice, we did not observe evidence of a greater vascular flux to vertebrae over long bones. Thus, other unrecognized biologic factors must explain this vertebral metastatic preference. Indeed, when bone organoids were formed in contralateral legs from isolated and transplanted long bone SSCs versus vSSCs, allowed to mineralize and then challenged with breast cancer tumor lines via the caudal artery, the vSSC-derived organoids displayed metastatic outgrown nearly twice as often as the long bone SSC-derived organoids. Thus, the vertebral metastatic preference is intrinsic to the lineage of the specific vSSC transplanted and independent of confounding differences in anatomic context or architecture between vertebrae and long bones. Likewise, conditionally targeting vSSCs to impair their osteoblast differentiation capacity selectively impaired spine metastases in a hematogenous seeding model. Thus, this Zic1+ vSSC is both necessary and sufficient to recapitulate the clinically observed metastatic preference for the spine over long bones, arguing that site-to-site differences in skeletal cellular composition are the key determinant of the tropism for bone metastases. However, reflecting our current lack of genetic tools that enable functionally targeting stem cells without necessarily impacting their mature derivatives, it is currently unclear the degree to which these metastatic phenotypes reflect the contributions of vSSCs themselves versus reflecting the function of the entire vSSC lineage, including vSSC-derived osteoblasts. Analysis of the physical cell–cell interactions comprising the early bone metastatic niche using mouse lines that allow for proximity-based labeling of cells near a cell type of interest may allow for progress in addressing this question despite this challenge.²¹

Multiple stem cells collaborate to mediate calvarial mineralization

Another, separate line of investigation emerging from the identification of CTSK+SSCs in the calvaria and long bone periosteum was whether, if CTSK+SSCs form a large portion of the calvarial osteoblasts, defects in CTSK+SSCs represent a convergent etiology for multiple craniofacial disorders of the calvaria. To test this, we conditionally deleted Twist1, a gene mutated in the craniosynostosis disorder Saethre-Chotzen Syndrome,²² in CTSK+SSCs. The resulting mice did display suture fusion; however, contrary to our expectation that osteogenic differentiation of CTSK+SSCs would be enhanced to mediate the fusion observed, we instead observed marked dropout of CTSK+SSCs at sutures destined to undergo fusion. Instead, an alternative non-CTSK lineage markedly expanded in response to the absence of CTSK+SSCs. Inspired by work identifying the transmembrane collagen receptor DDR2 as marking a population of long-lived suture cells, we found that DDR2 marked an orthogonal SSC lineage not overlapping with that of CTSK+SSCs²³ (Figure 2). Through secretion of IGF1 and other mediators, CTSK+SSCs suppressed the proliferation of DDR2 + SSCs. Thus, CTSK+SSCs and DDR2 + SSCs compete for a partially shared niche, and the expansion and activation of DDR2 + SSCs that drive suture fusion is a maladaptive response to the absence of CTSK+SSCs. In contrast to the intramembranous specialization of the CTKS+SSC lineage, DDR2 + SSC lineage mediates endochondral ossification. Accordingly, CTSK+SSC dropout triggered inappropriate endochondral ossification in the calvarial suture, a phenomenon we also observed in a series of sporadic cases of human craniosynostosis.

Figure 2.

A summary of calvarial skeletal stem cells. Under physiologic conditions, CTSK+ skeletal stem cells (SSCs) suppress the expansion and activation of DDR2 + SSCs. In the presence of mutations or other conditions that cause a dropout of CTSK+SSCs, DDR2 + SSCs are de-repressed and expanded to mediate endochondral ossification in the calvarial suture, ultimately resulting in suture fusion. Image created with Biorender.com.

These observations suggest that craniosynostosis may involve a complex interplay of simultaneous loss-of-function and gain-of-function, each in distinct SSC lineages that actively interact with each other. In this example, loss-of-function mapped to the CTSK+ SSC lineage and a secondary gain-of-function mapped to the DDR2 + SSC lineage, though it is plausible that in other scenarios primary dysfunction could map to DDR2 + SSCs or to as of yet undiscovered populations. In this respect, it is notable that a number of genes causatively mutated in craniosynostosis show strong skewing in expression to either CTSK+SSCs or DDR2 + SSCs. For instance, Rab23, Fgfr1, Smad6, and Il11ra demonstrate preferential expression in CTSK+SSCs, whereas Fgfr2, Fgfr3, Msx2, Efnb1, and Tcf12 exhibit preferential expression in DDR2 + SSCs. We hypothesize that the craniosynostosis disorders associated with mutations in these genes will in turn be driven by primary dysfunction in the SSC lineage displaying preferential expression of this gene. If substantiated, this could ultimately provide the basis for a mechanistic nosology of craniosynostosis disorders, allowing the classification of whether those disorders are driven by primary defects in CTSK+SSCs vs DDR2 + SSCs. Moreover, it raises the possibility that craniosynostosis-associated genes may include receptor-ligand pairs involved in crosstalk between these cell types, with the receptor in one of these stem cells and the cognate ligand in the other, as shown by the example of IGF1 expression in CTSK+SSCs and IGFR1 expression in DDR2 + SSCs. In line with the simultaneous loss and gain-of-function in calvarial mineralization seen with Twist1 deletion targeted to CTSK+SSCs, a combination of hypomineralization (loss-of-function) and suture fusion (gain-of-function) is commonly seen in both human craniosynostosis and mouse models. This includes hypomineralization centered on the anterior fontanelle region in a variety of germline and conditional Twist1 loss-of-function models, including those also displaying fusion.^24-26 Similar concurrent hypomineralization and fusion have been seen in Gli3 loss-of-function mice,²⁷  ALX4-deficient patients,^28,29  FGFR2 mutations associated with Bent Bone Dysplasia,³⁰ human Apert syndrome,³¹ both human hypophosphatasia and mouse Alpl^−/− models of hypophosphatasia,^32-34 X-linked hypophosphatemic rickets or nutritional rickets,^35-37 and with MSX2 mutations in mice and humans.^38,39 Overall, it seems that concurrent hypomineralization and fusion are broadly observed in many forms of craniosynostosis, and the model here offers a potential means to explain this finding, with the loss-of-function and gain-of-function elements of the phenotype each mapping to distinct stem cell lines.

How many SSCs?

The studies of the vertebral stem cell and the 2 stem cells involved in mediating craniosynostosis indicate that distinct skeletal regions are formed by distinct SSC lineages, each displaying functional specializations that enable the local physiology of their respective sites. In turn, these same functional specializations provide a basis for site-specific skeletal pathologies, such as craniosynostosis or the high rate of spine metastases.

Though the hematopoietic system displays a large number of possible mature cell fates, each of these hematopoietic cell types is classically linked by a single differentiation hierarchy that operates throughout the entire body. Whereas the cellular basis of bone formation was once considered simple in comparison—consisting largely of osteoblasts, osteocytes, and chondrocytes, the emerging evidence summarized here paints a much more complex picture. Instead of a single shared differentiation tree, each SSC maintains its own differentiation hierarchy. These separate differentiation hierarchies each have a similar and parallel overall structure, but with specific differentiation pathways latent or suppressed in stem cells not competent to generate specific cell fates. For instance, adipogenic and chondrogenic differentiation pathways are normally suppressed or latent in CTSK-lineage periosteal stem cells. In this respect, comparison of the cell types comprising the lineage of SSCs able to generate a mature cell type and SSCs unable to form that cell type may represent an opportunity to identify as yet elusive cellular points of lineage commitment to osteoblasts, marrow adipocytes, or other skeletal cell types.

This increasing complexity of skeletal cell types naturally invites the question of how many distinct stem cell types are present and thereby what are the limits of SSC complexity. One attractive possibility is that there is an evolutionary basis for the diversity of SSCs. In this model, each time evolution elaborates a new and distinct skeletal structure, a new class of SSCs will arise to orchestrate the formation and physiology of that site. This suggests that lesser studied bones, such as the clavicle or possibly the ossicles, with a unique evolutionary origin will be of interest for possible housing yet undiscovered distinct SSC types.

Another related lens through which to anticipate the ultimate degree of diversity in SSCs is the view that each skeletal tissue represents the sum of the tissue-forming activities of the stem cells comprising that tissue. Thus, careful evaluation of the tissue production capacity of each stem cell lineage can reveal cell areas where known cell types are inadequate to fully explain the formation of that region of the skeleton and therefore infer the likely presence of additional stem cells. For instance, neither CTSK+SSCs nor DDR2 + SSCs in the calvaria can form a bone marrow niche, suggesting the existence of yet a third calvarial stem cell.

It is also possible that even within classes of skeletal elements such as the long bones of the limb or vertebral bone, there could be further distinctions among SSCs along a craniocaudal or proximal to distal axis. Thus, the cervical spine and lumbar spine Zic1 + SSCs may display functional and transcriptional differences. Relevant to this model, Hox genes display continual postnatal expression and function in skeletal progenitors,^40,41 raising the possibility that a postnatal Hox code is further overlaid on the SSC types above to create a matrix whereby both the Hox gene and SSC type as framed by the examples of Zic1 + SSCs or CTSK+SSCs above combine to determine regional cell identity. However, we anticipate that with increasingly fine local “resolution” of regional cell identity will in turn likely correspond to increasingly subtle differences between these cell types.

Lastly, we contend that this expanding complexity of SSC types highlights the need for a shared and standardized nomenclature in denoting these cell types, as it has become progressively challenging for those not actively involved in the field to identify and define skeletal cell types to understand which papers refer to overlapping versus independent populations, which studies only identity lineages of cells versus resolving specific cell states within that lineage, and how these studies differ in their evidence for the stemness or other functional properties of these cells. While such a suggested nomenclature will necessarily be incomplete, contentious, and initially include inaccuracies that require continual revision in light of new studies, this nevertheless represents a necessary step to enabling the broader field of skeletal biology to incorporate the rapid advances in studying skeletal progenitors into studying other aspects of skeletal biology.

Is there no single unified osteoblast cell type?

This new appreciation of diversity in the SSC compartment raises the question of whether this diversity extends throughout these lineages to mature cells, especially as osteoblasts are increasingly appreciated to be derived from a diverse set of SSC sources.⁴² Does each of these SSCs produce osteoblasts that converge on a similar phenotype or are each of the pools of osteoblasts produced by these SSC lineages functionally and transcriptionally distinct? While this question remains not fully resolved, existing evidence supports the model that each SSC produces distinct pools of osteoblasts. For instance, in the Sun et al. report, the Emb + fraction of cells enriched for osteoblasts in both the long bone SSC and Zic1 vSSC lineages are transcriptionally distinct, with the vSSC-lineage Emb + cells retaining expression of genes classically associated with somitic development, such as Mesp2 and Paraxis (Tcf15).^43,44 This argues that osteoblasts are likely to retain the imprint of their SSC of origin. Thus, we propose that there may be no such thing as a single unified osteoblast cell type existing throughout the skeleton. Rather, osteoblasts are likely a family of related cell types, each distinct based on their SSC of origin. This in turn opens a new mechanism by which site-specific SSCs can display distinct functions, as these SSCs each imprint, likely via lineage-specific epigenetic differences, transcriptional and corresponding functional differences to the mature effector cells, such as osteoblasts, within their lineages. Thus, SSCs are likely not only important as a continual source of osteoblasts facilitating cellular turnover, but rather as the determinants of lineage specific functional properties. In this model, it is likely important not only to understand the number of osteoblasts present, such as in traditional histomorphometry, but to moreover understand the stem cell lineage of these osteoblasts and how this in turn shapes their identity and function.

Author contributions

Seoyeon Bok (Conceptualization, Writing—original draft, Writing—review & editing), Jun Sun (Conceptualization, Writing—original draft, Writing—review & editing), and Matthew B. Greenblatt (Conceptualization, Project administration, Writing—original draft, Writing—review & editing)

Funding

J.S. is supported by a fellowship from the Children’s Tumor Foundation (CTF-2023-01-005, https://doi.org/10.48105/CTF.CTF-2023-01-005.pc.gr.172007). S.B. is supported by the Arthritis National Research Foundation, a Jumpstart award from Weill Cornell, a Fellowship from the MOGAM Science Scholarship Foundation, and a Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2021R1A6A3A14038667). M.B.G. is supported by the NIH through R01HD115274, R01 CA282815, R01AR075585, R01AR083462 and a developmental project award through 5P50CA211024, a MIND Prize from the Pershing Square Foundation, a Mary Kay Ash Foundation Award, an Innovator Award from the Marfan Foundation, a Burroughs Wellcome Career Award for Medical Scientists, and a research award from the William Rhodes and Louise Tilzer-Rhodes Center for Glioblastoma. All opinions expressed are solely those of the authors.

Conflicts of interest

None declared.

Data availability

No new data were generated or analyzed in support of this research.