The Osteocyte Transcriptome: Discovering messages buried within bone

Abstract

Purpose of the review:

Osteocytes are cells embedded within the bone matrix, but their function and specific patterns of gene expression remain only partially defined; this is beginning to change with recent studies using transcriptomics. This unbiased approach can generate large amounts of data and is now being used to identify novel genes and signalling pathways within osteocytes both at baseline conditions, and in response to stimuli. This review outlines the methods used to isolate cell populations containing osteocytes, and key recent transcriptomic studies that used osteocyte-containing preparations from bone tissue.

Recent findings:

Three common methods are used to prepare samples to examine osteocyte gene expression: digestion followed by sorting, laser capture microscopy and the isolation of cortical bone shafts. All these methods present challenges in interpreting the data generated. Genes previously not known to be expressed by osteocytes have been identified and variations in osteocyte gene expression have been reported with age, sex, anatomical location, mechanical loading, and defects in bone strength.

Summary:

A substantial proportion of newly identified transcripts in osteocytes remain functionally undefined but several have been cross-referenced with functional data. Future work and improved methods (e.g. scRNAseq) are likely provide useful resources for the study of osteocytes and important new information on the identity and functions of this unique cell type within the skeleton.

1. Introduction

The skeleton is a highly dynamic organ with several important functions. The skeleton supports the whole body and facilitates movement, stores large amounts of calcium and inorganic phosphate and releases them when required for calcium homeostasis, protects vital organs, hosts hematopoiesis, and contributes to endocrine regulation through hormone production. These functions are accomplished through the coordinated activities of osteoblasts, which build bone, osteoclasts, which resorb bone, and osteocytes which orchestrate the process of bone remodelling.

Osteoblasts, located on the bone surface, deposit new bone through the secretion of extracellular matrix proteins (principally collagen type I) and the production of non-collagenous proteins that regulate osteoid mineralisation. When osteoblasts secrete the extracellular matrix, a portion of them become embedded in it, and differentiate into osteocytes. Once embedded, osteocytes are terminally differentiated, and long lived, and constitute >90% of adult bone cells (for reviews, see [1–3]). Osteocytes live entombed inside bone within spaces called lacunae; as they differentiate within the matrix, they form many cell membrane projections from their cell body, known as dendrites, which extend through a network of inter-connected canals. The combination of cells, canals and lacunae forms the lacuno-canalicular network. Osteocyte dendritic processes establish contact with dendrites from adjacent osteocytes, and extend to other cell types including cells on the surface of bone, and endothelial cells lining the intra-cortical capillaries that feed the bone tissue [4, 5]. The highly interconnected osteocyte network is ideally positioned to sense both mechanical stimuli applied to the skeleton and microcrack damage to bone, and to coordinate appropriate anabolic or catabolic responses by the cells on the bone surface [1, 6].

Osteocytes are critical to the maintenance of bone mass and its mineralisation. Prior to extensive transcriptomic studies, Phex, E11 and Dmp1 were known marker genes associated with early embedding osteocytes, while Mepe, Sost and Fgf23 were known marker genes for late, mature osteocytes (for comprehensive reviews, see [3, 7]). Osteocytes promote mineralization of the bone matrix through their production of Phex, Dmp1 and Mepe. These cells also control systemic phosphate homeostasis via their production of FGF23 [8–10]. Osteocytes are also mechanosensory cells, and mechanical stimulation reduces Sost expression in osteocytes, thus releasing its inhibition of Wnt signalling and thereby stimulating bone formation [3].

While the study of the molecular and biological characteristics of osteocytes has been limited by significant technical challenges, much progress has been made in recent years. Although most of this work is limited to rodent studies due to the difficulty of obtaining and purifying human osteocytes, osteocyte transcriptomic studies have already provided new information about osteocyte identity and cellular functions. Here, we will review what is known about the osteocyte transcriptome in vivo by focusing on cell preparations made from bone tissue, discuss recent findings, and highlight future research directions.

2. Technical obstacles and challenges to obtain a pure osteocyte preparation

When characterising cell populations and interrogating the molecular pathways used by those cells, it is important to obtain a pure population of the intended cell type. However, since osteocytes are buried deep within the bone, the isolation of a pure population is inherently difficult because of the presence of other cell types within cortical bone and on the bone surface. Several strategies have been used to prepare samples for analysing the osteocyte transcriptome; these are discussed below and illustrated in Figure 1.

Figure 1: Methods commonly used to isolate and study the osteocyte transcriptome.

A. Isolation of cells by enzymatic digestion: Bone tissue (e.g. calvaria) is cut into pieces and enzymatically digested (e.g. collagenase, dispase, etc.), the samples can then be processed for fluorescence-activated cell sorting (FACS) based on fluorescent reporters (lineage tracing) or by antibody staining. B. Laser capture microscopy: Bone tissue is fixed and prepared for cryosectioning. Samples are then observed under a microscope and a laser is used to cut out the desired regions. C. Isolation of the cortical bone shaft: Bones are dissected free from muscle and periosteum; the epiphyses are removed, and the bone marrow is removed either by flushing the cavity with phosphate buffered solution or by centrifugation. The remaining bone shaft appears white in colour and is snap frozen in liquid nitrogen for preservation. The sample is then homogenised to extract RNA. D: Following isolation of RNA using methods (A, B or C), libraries are prepared, sequenced using RNAseq. Alternatively, RNA can be amplified and evaluated using a microarray with known probes.

One method that has been quite successful in generating purified osteocytes from murine bone samples is lineage tracing and cell sorting. Lineage tracing is a method to identify and track cells within transgenic mice generated so that specific cell populations express a reporter gene (e.g. a fluorescent protein) driven either by cell-specific Cre/loxP recombination or under the guidance of a cell-specific gene promoter. This method depends on the availability of genetically modified mice to label specific populations. It was first used by the Kalajzic group to isolate osteoblasts and osteocytes that had been released from calvaria of 5-8-day-old mice by 4 sequential digestions with trypsin/EDTA and collagenase P [11]. Two transgenic mouse lines were crossed to generate double-transgenic mice where osteoblasts and osteocytes were labelled with different fluorescent reporter constructs. These were the Col1(2.3kb)-GFPCyan mouse line, in which mature osteoblast lineage cells were labelled with a fluorescent blue (Cyan) reporter construct using the type I collagen (Col1a1) promoter, and the Dmp1-GFPTopaz mouse line, in which osteocytes were labelled with a fluorescent yellow (Topaz) reporter construct targeted using a portion of the osteocyte-specific Dentin matrix protein 1 (Dmp1) promoter. Calvarial cells were separated into 3 populations by fluorescence-activated cell sorting (FACS) based on their GFP expression: (1) GFP negative cells (neither osteoblasts nor osteocytes), (2) osteoblasts (Col2.3kb-Cyan+, Dmp1Topaz-), and (3) osteocytes (Dmp1Topaz+). Comparison of these three populations by microarray data analysis identified 385 genes that were differentially expressed between osteocytes and osteoblasts. In osteocytes, 136 genes were downregulated (i.e ≤ 0.5-fold change) and 249 genes were upregulated (i.e. ≥ 2 fold change) compared to Col2.3kbCyan+ osteoblasts. As expected, Dmp1 gene expression was higher in Dmp1Topaz+ cells [11]. Several genes not previously associated with osteoblast lineage cells were identified and validated by qPCR to be uniquely expressed in osteoblasts or osteocytes. For example, osteoblasts, but not osteocytes expressed Keratocan (Kera), a proteoglycan previously only associated with corneal tissue [12] and conversely, osteocytes but not osteoblasts expressed Neuropeptide Y (Npy) and Reelin (Reln), two important proteins for central nervous system development/function. The functional significance of two of these genes in bone has been confirmed by subsequent studies showing that Kera promotes osteoblast differentiation [13] and that Npy inhibits osteoblast differentiation and mineral deposition [14–16]. This foundational study of osteocyte transcriptomics is important because it successfully identified the above novel genes (Kera, Reln, Npy) within the osteoblast lineage that distinguish osteoblasts and osteocytes. This has informed the bone field, with several groups using this information to confirm cell identities [17–19] indicating that the analysis of the transcriptome can provide useful new information and novel insights on genes that had previously been primarily studied in non-bone tissues.

The methods described above using reporter genes and digestion of young calvarial bone are highly dependent on the labelling/gating methods used and are likely to contain contaminant cells. This was demonstrated when haematopoietic and endothelial cells were later identified within the GFP+ fraction of calvarial cells derived from Dmp1-GFPTopaz mice [17]. Only after the depletion of haematopoietic and endothelial lineage cells using FACS, was it possible to obtain a more purified population of osteocytes (Lin-GFP+). In addition to osteocyte-specific genes (Dmp1 and Sost), Lin-GFP+ osteocytes expressed genes normally thought to be specific to osteoclasts, bone resorbing cells of haemopoietic origin. These osteoclast-type genes expressed included cathepsin K (Ctsk), tartrate-resistant acid phosphatase (Acp5) and calcitonin receptor (Calcr) which are associated with bone resorption but did not include other osteoclast markers (Dcstamp, Oscar) associated with multinucleation [17]. However, while lineage depletion improved the purity of the isolated osteocytes, providing clarification of some genes expressed in osteocytes, it also resulted in very few cells for analysis making this a challenging method for regular use.

One major concern with studies that rely on dissociation of cells from bone is that the digestion methods themselves may induce changes to the transcriptome. While suggested in earlier studies [20–22], this was most eloquently demonstrated by O’Flanagan et al. (2019) who compared the effects on primary solid tumour tissue of a collagenase/hyaluronidase digestion at 37°C with a gentler method (a serine protease at 6°C) [23]. They identified 512 differentially regulated genes conserved across a range of cancer tissues that were associated with heat shock and stress responses following the 30-minute digestion with collagenase compared to the cold-activated protease [23]. Similarly, longer digestion times also led to an increase in the number of differentially expressed transcripts [23]. This is particularly important for studies of osteocytes in long bone since the thick cortical bone requires an extended time in digestion, compared to the thinner calvarial bone, to release the cells [24]. Digestion will preferentially release recently embedded osteocytes, whilst more mature, deeply embedded osteocytes may be retained within pieces of bone. Other approaches to osteocyte isolation such as homogenization of cortical bone would both avoid digestion-induced changes to the transcriptome and would obtain the transcriptome of osteocytes embedded deeper within the bone.

Laser capture microscopy is a second method used to isolate osteocytes based on their location within bone. In this procedure, samples are fixed, processed for frozen tissue sectioning, and then under a microscope small sections are cut using an infrared laser for subsequent RNA isolation. The advantage of this approach is that it provides spatial data for the cell types isolated and can facilitate comparisons from cells isolated between different parts of the bone. Nioi et al. (2015) compared gene expression by osteoblasts (cells adjacent to labelled mineralising bone surfaces), lining cells (flat cells on unlabelled bone surfaces) and osteocytes (cells embedded within bone matrix) [25]. This study aimed to understand how these mature cell types of the osteoblast lineage respond to a single dose of sclerostin antibody treatment, a potent stimulator of bone formation [25]. They found that all three cell types (osteoblasts, osteocytes, and lining cells) had similar anabolic transcriptomic responses to anti-sclerostin, suggesting that osteocytes and lining cells may also participate in bone formation. There are limitations to this method too. The fixation and processing required results in lower RNA yield and quality compared to other methods such as digesting and isolating cortical bone shafts [26]. The isolation of specific cell populations within the heterogeneous bone marrow environment can also be technically challenging. This is less of an issue for bone-embedded osteocytes which are less likely to be adjacent to other cell types.

Since osteocytes are the predominant cell type within cortical bone [1–3], the majority of osteocyte transcriptomic studies have assessed RNA isolated from cortical bone preparations [22, 27, 28, 19]. Briefly, the excision of long bones (e.g. femur/tibia) is followed by the removal of periosteum and muscle (by scraping the exterior surface of the bone), cutting of the epiphyses and elimination of the bone marrow (by flushing or spinning out). The remaining cortical bone shaft is then processed to isolate RNA. Speed in the processing of these samples is essential to generate high quality RNA with little degradation. Consistency of sample preparation is also important, since inadequate removal of bone marrow, periosteum, or muscle would change the transcriptomic profile. Even with these precautions, other cell types are present within cortical bone, including osteoblast progenitors [29], osteoclasts and their progenitors [30], vascular and perivascular cells [30]. These would also be captured within these samples and represent contaminants when interpreting what is often termed “osteocyte-enriched” populations. In samples with altered cortical bone features (e.g. increased vasculature within cortical bone), their presence within the samples may also lead to misleading interpretations of differences in gene expression.

Ayturk et al (2013) tested three strategies for identifying potential contaminant mRNAs in cortical bone preparations while identifying differentially expressed transcripts among control (Lrp5+/+), LRP5 deficient mice (Lrp5−/−), and mice with high bone mass due to an LRP5 knock-in mutation (Lrp5p.A214V/+)[22]. First, they prepared and analysed the transcriptome of known bone-contaminating tissues, such as muscle, blood and bone marrow, to identify transcripts with at least two-fold greater abundance in these tissues compared to cortical bone, so that this information could be used for later bioinformatic exclusion from bone analyses. This approach identified ~6000 potential transcripts that may represent contaminant genes, however, removal of these genes from the cortical bone transcriptome reduced the transcriptome by 43%. This sizeable filtration of data would make it difficult to obtain a complete picture of the osteocyte transcriptome as there are likely some genes also expressed by other tissues that may have roles in osteocytes even when expressed at lower levels. Thus, other complementary approaches described next were required to refine this initial gene list. Second, in an attempt to produce a more purified osteocyte population, they cut diaphyseal bone shafts into small pieces and digested them with collagenase to remove any remaining cells from the bone surface prior to processing the remaining bone chips for RNA. This reduced the presence of contaminating cells/mRNAs compared to homogenised cortical bone samples. However, the use of collagenase caused significant variation in >700 genes when transcriptomes from digested and fresh-frozen bone were compared; this far exceeded the variation expected based on the simple removal of “contaminant non-bone” genes defined in the first experiment. A refined list of potential contaminant genes was subsequently generated based on transcripts identified in both experiments. A third method to identify contaminants compared RNA-seq data from a single animal’s right and left tibia. The assumption being investigated was that cortical bone genes would be directly correlated (R²~1), while contaminant genes would have the greatest variance (i.e. have the greatest effect on the R² value) since their presence would be inconsistent between samples. Genes identified by this third strategy had previously also largely been identified by the first two strategies. Technically, this third approach is least useful for large scale studies because it requires twice the amount of sequencing (two samples per biological sample) to generate these comparisons, and it provided little new information compared to the refined gene list generated from experiments 1 and 2. Ideally, independent studies should be used to confirm the accuracy of this gene list, although this information has already been applied by other groups to filter results [31, 28], and was effective in removing highly abundant genes associated with blood (e.g. haemoglobin) and some muscle genes.

Choosing a method for examining the osteocyte transcript requires careful consideration of the experimental question proposed. The isolation of cortical bone shafts, with or without bioinformatic correction for likely contaminants, is likely the best and most amenable method since it yields sufficient high-quality RNA and can be used with bones of all ages. However, if a specific population or region of the bone is being examined, the digestion method or laser capture microscopy may be more desirable.

3. Inferring osteocyte identity and function based on transcriptomic analyses

Pathway analysis is commonly applied to transcriptomic data to identify patterns of gene expression that are associated with known signalling pathways and cellular functions. This is useful for understanding and interpreting possible functions associated with cell identities and changes in gene expression. Two frequently used approaches are gene set enrichment analysis (GSEA) and gene ontology (GO) analyses. GSEA ranks known pathways/gene sets against an entire data set to determine the pathways that may be enriched (i.e. more strongly represented) in one biological sample compared to another whereas GO analyses are used to associate differentially expressed genes with a known molecular function (e.g. Wnt signalling), cellular component (e.g. ribosomal), or biological process (e.g. osteoblast differentiation). Both methods depend on evidence gleaned from the literature. Given that osteocytes descend from the osteoblast lineage, it is not surprising that these analyses generally show the osteocyte transcriptome is associated with pathways such as skeletal development and morphogenesis, osteoblast differentiation, osteogenesis, and extracellular matrix. However, there are currently no gene ontology (GO) terms specifically for or containing the word “osteocyte”. This represents a significant limitation of using pathway analysis on bone samples. For instance, pathway analyses can be quite useful for identifying changes in known signalling pathways following an intervention, e.g. mechanical loading, however the bias for selecting pathways already known to be associated with osteoblast function may exclude specific genes that could provide new insights into osteocyte identify or function.

A different approach used recently was a Bayesian network-based clustering method to group genes that were co-expressed, associate these clusters with GO terms to identify possible functions, and use this to identify and select novel genes that clustered with known bone genes with the hypothesis that they would most likely have a causal influence on bone homeostasis. [32]. Clusters were then screened for associations with colocalising eQTLs from human genome-wide association studies (GWAS) of bone mineral density. This led to further work examining the roles of SERTA domain-containing protein 4 (Sertad4) and Glycosyltransferase8 domain containing 2 (Glt8d2) in bone metabolism [32]. Although pathway analysis and network-based clustering serve different purposes, they are useful tools for examining the whole transcriptomic data and then using it to identify key pathways or gene clusters associated with potential functions.

Youlten et al (2021) recently described an “osteocyte gene signature” based on a comparison between isolated cortical bone shafts and cortical bone shafts in which bone marrow had been retained. This comparison was used to examine the distribution of gene enrichment and calculating a threshold for osteocyte-enrichment. Additional genes identified in blood and muscle by Ayturk et al (2013) were then filtered from this gene set since they are likely contaminants. This led to a collection of 1239 genes that were more highly expressed in samples of cortical bone and attributed to osteocytes. They were confirmed by cross-referencing of this gene set with an independent study that isolated osteoblasts, osteocytes, and lining cells by laser capture microscopy [33]. Within this signature were genes known to be expressed by osteocytes such as Dmp1, Sost and Mepe, but also genes not previously associated with bone such as Auts2 and Cttnbp2. Whether these are expressed by osteocytes themselves has not been confirmed, and will be important to do, since murine cortical bone also contains blood vessels, and novel cell types [29, 30]. Functional analysis of these genes specifically within osteocytes remains to be performed. Using the Origin of Bone and Cartilage (OBCD) database, 64 genes within the osteocyte gene signature had global knockouts available and 26 of those exhibited a skeletal phenotype. Further work is required to confirm whether these skeletal changes are a direct result of lost osteocyte gene expression or result from lack of expression by other cell types.

Production of extracellular matrix transcripts by osteocytes

As osteocytes represent terminally differentiated osteoblasts, it is unclear to what extent certain transcriptional programs are still shared between the two cell types, and which are unique particularly under different contexts (e.g. treatment, disease). As an example, multiple transcriptomic analyses have indicated that osteocytes express high mRNA levels of extracellular matrix (ECM) components. This is somewhat surprising, since osteocytes are not generally considered to deposit osteoid to the same extent as osteoblasts. When the Dmp1Topaz model was used to study the young murine calvarial transcriptome, more than half the genes coding for secreted proteins, including components of basement membranes and ECM, showed different expression levels between osteoblasts and osteocytes [11]. For instance, osteocytes expressed higher levels of type IV collagen (a key component of all basement membranes) while osteoblasts expressed higher levels of collagens usually associated with cartilage, such as types II and IX. Some other less abundant collagens were also differentially expressed, but type I collagen (Col1a1), the most abundant and arguably the most important collagen for the skeleton, was not differentially expressed. Laser capture microscopy analysis of rat long bone response to anti-sclerostin antibody also found similar levels of Col1a1 and osteocalcin (Bglap) in osteoblasts and osteocytes [25]. However, although osteoblasts are the major cells involved in bone matrix production, it was surprising that sclerostin antibody treatment caused a greater upregulation of these transcripts in osteocytes than in osteoblasts. This suggests osteocytes are still very capable of responding to an anabolic stimulus, however it is still unclear if they can produce new matrix and if so, under which conditions. It is not likely to occur in response to anti-sclerostin treatment, since osteocyte lacunar size is unaltered with sclerostin treatment/deficiency [34, 35]. Consistent with these findings, Zimmerman et al. also observed that, in an osteocyte-enriched RNA preparation from murine long bones, the genes encoding for Col1a1, Bglap and osteonectin (Sparc) were in the top 1% most abundant osteocyte transcripts [19].

The ability of osteocytes to contribute to bone matrix gene expression is further supported by Youlten et al. (2021) showing that both type one collagen transcripts (Col1a1, Col1a2) and several other genes known to be involved in intracellular collagen processing, folding and secretion (e.g. Crtap, P3h1, Plod2, Fkbp10) were highly enriched in their osteocyte transcriptome signature genes [28]. Most of these genes were thought to be primarily expressed by osteoblasts and many are associated with congenital forms of bone brittleness, including osteogenesis imperfecta (OI) (see below). Instead, their expression level was similar, if not higher, in osteocytes compared to osteoblasts, suggesting that osteocytes produce collagen type I, the major constituent of osteoid. Although all these studies point to production of collagen type I by osteocytes, this remains to be proven at a protein or functional level.

Are all osteocytes the same?

It has been known for many years that osteocytes embedded within woven bone have a different appearance and different number of cellular processes compared to osteocytes within lamellar bone [36]. Also, early in situ hybridization studies showed that early osteocytes closer to the bone surface express higher levels of some transcripts (e.g. osteocalcin, RANKL, and PTHrP)[37–39], while more deeply embedded osteocytes express other proteins like sclerostin [40]. Therefore, since variation is present between osteocytes, there are also differences within the osteocyte transcriptome. Some of these differences in the osteocyte transcriptome are associated with bone development and maturation, sex, and the skeletal site examined. These findings are discussed below.

Youlten et al. (2021) examined RNA isolated from cortical bone shafts from male and female mice aged 4, 10, 16 and 26 weeks old [28]. When compared using principal component analyses, the transcriptomes of samples from growing 4-week-old mice were distinct from those derived from skeletally mature mice at 10, 16 and 26 weeks of age [28]. This demonstrates age-related differences within the osteocyte transcriptome. These differences may be related to bone composition since cortical bone of younger mice has more woven bone and contains mineralised cartilage [41]. The extent of differences in the osteocyte transcriptome between woven and lamellar bone has not been studied.

While Youlten et al. observed no difference in the osteocyte transcriptome between sexes at 4 and 10 weeks of age, differences between sexes were observed at 16 and 26 weeks [28]. For example, genes associated with bone resorption such as tartrate resistant acid phosphatase (Acp5) and cathepsin K (Ctsk) were higher in female bone compared to male, although no TRAP-positive osteoclasts were identified within histological sections of the prepared cortical bone shafts [28]. Previous studies consistently identified these transcripts in osteocytes isolated from calvarial bone by other methods [42, 11, 17]. Since this gene set was also shown to be differentially regulated in a transcriptomic study examining bones from lactating mice, they may be associated with the removal of bone matrix by osteocytes, known as osteocytic osteolysis [42].

The osteocyte transcriptome also differs between anatomical sites [28]. Wang et al. (2020) identified 32 genes differentially expressed between adult cortical bone from calvaria and tibiae across 3 species (mouse, rat, and macaque), including multiple Hox genes (Hoxc8, Hoxc9, Hoxc10, Hoxb5, Hoxb7, Hoxc4, Hoxc5) [43]. Hox genes are differentially expressed across the body and involved in defining regions of the body during embryonic development [44]. Several genes associated with bone formation and bone mineral density were enriched in tibial samples compared to calvaria (Bmp7, Dkk1, Fgf1, FrzB, Hoxa11, Hoxa7, Mepe, Sost, and Tnfsf10); conversely, FosB and Zic1 were enriched in calvarial samples [43]. Differential gene expression between calvaria and tibiae is likely associated with bone function/response, for example, osteocytes within the tibia experience greater mechanical load than osteocytes in the calvarium [45]. Youlten et al, studying murine cortical bone shafts from femur, tibia and humerus also detected differential expression of Hox genes and Hox anti-sense long coding RNAs between fore- and hind-limbs. Some of these differences between humerus and hindlimbs were predicted from previous literature: (e.g. forelimb-specific expression of homeobox-d9 (Hoxd9) and T-box 5 (Tbx5) [46] and hind limb-specific expression of Pitx1 [47]). While Hox genes and Pitx1 have long established roles in skeletal patterning during development [46–48], there is also evidence for Hox gene activity within mesenchymal stromal cells and haematopoietic cells of the adult skeleton [44]. The presence of Hox transcripts in osteocytes within the adult skeleton raises the question of whether they are functionally active and how they contribute to osteocyte function.

Neuronal gene expression in osteocytes

During the osteoblast-to-osteocyte transition, cells progressively change their morphology from a cuboidal to a stellate shape, consisting of a smaller cell body with many dendritic projections, resembling neurons. This has led many to hypothesise that osteocytes and neurons might share processes including dendrite formation, intracellular vesicle transport and cytoskeletal organization. Indeed, genes that were first discovered to have a neuronal function have now been identified in the osteocyte transcriptome [11, 28, 19]. Indeed, neuropeptide Y (Npy) was identified by Paic et al in the osteocyte transcriptome, and later functional studies showed that Npy reduces bone formation and has broader expression within the osteoblast lineage [49, 14, 50, 16]. Also, of interest in the context of osteocyte transcriptomics are the genes and pathways associated with axon development and growth such as Ephrin-Eph signalling, netrins, semaphorins, and Slit-Robo signaling. These might provide novel information about the formation and function of dendrites in osteocytes.

Semaphorins, along with their receptor components, Plexins and Neuropilin, are consistently identified in transcriptomic analyses of osteocytes [28, 19]. This network plays a role in axon guidance and has been implicated in the regulation of bone mass (although with some controversy) [51]. Early work using a global Sema3A loss of function mouse showed that Sema3A protects bone mass by suppressing osteoclastic bone resorption and increasing osteoblastic bone formation [52]. Others demonstrated that while neuronal Sema3a knockout recapitulated this phenotype, targeted life-long deletion of Sema3a in the osteoblast lineage, using Osx-Cre or Col1a1-Cre, had no effect on bone mass, suggesting that the global Sema3 knockout bone phenotype was due to neuronal depletion [53]. Further elegant work by Hayashi et al (2019) depleted Sema3a in the osteoblast lineage postnatally, to exclude possible influences of the innervation process, and this led to lower bone formation and higher bone resorption indicating a requirement for postnatal cell-autonomous Sema3a in bone forming cells [54]. Similar effects on bone were seen by mutating Neuropilin 1, a component of the Sema3a receptor complex [54]. The same authors also showed that Sema3 / Neuropilin 1 signalling promoted mature osteocyte survival. This indicates that genes associated initially with axon guidance in neurons can regulate a different cell function, i.e. cell survival, in osteocytes. This is an example of how great care must be taken when interpreting the functional significance of genes identified within transcriptomic data as their function within osteocytes may differ from their previously identified function.

The function of Ephrin-Eph signaling in the osteoblast lineage, including osteocytes, has been studied using a transcriptomic approach to uncover how osteocytes may influence bone structure and function [55, 27]. Targeted depletion of one member of this family, EphrinB2, in osteocytes using Dmp1-cre led to increased mineralization and a brittle bone phenotype without any change in bone mass [27]. Transmission electron microscopy showed no changes to dendritic morphology compared to control mice. However, RNAseq analysis of cortical bone shafts isolated from the EphrinB2 deficient mice identified differential regulation of genes primarily associated with autophagy, a process not previously associated with EphrinB2. The work suggested that osteocyte autophagy regulates secondary bone mineralization, and thereby controls strength independent of bone mass. Autophagy is also important in the nervous system; its dysregulation within neurons is associated with neurodegenerative diseases (e.g. Parkinson’s disease) [56, 57]. Thus, long-lived and terminally differentiated cells, such as neurons and osteocytes, may share an increased requirement of autophagic processes to maintain cell function with an aging proteome.

How do osteocytes respond to load?

Osteocytes sense and respond to mechanical stimuli that are applied to the skeleton. One of the first studies of the bone transcriptome response to load tested the effect of daily compressive loading to the right ulna of adult female Lewis rats compared to the contra-lateral unloaded ulna [58]. RNAs were analysed at 11 time points, from 4 hours to 32 days post loading, using an Affymetrix GeneChip Rat Exon array (>8000 genes) to identify patterns of change over the time course and with a clustering algorithm and functional characterisation using Ingenuity Pathway Analysis. This approach confirmed many genes already known to be important in loading-induced bone formation, including genes related to AP-1 in the early response group, matrix-related genes in the up-regulated cluster and canonical Wnt signalling pathway inhibitors in the downregulated cluster. Those associated with Wnt canonical signalling and Nerve growth factor (Ngf) were already known to have functional roles in the physiological response to mechanical loading in bone [59, 60]. However, several other gene groups were regulated (e.g. chemokine- and cytokine-related, and solute carriers) and some of these genes such as Capn6, Il4 and Kcne3 had not previously been associated with bone regulation.

More recent studies have confirmed changes in Wnt signalling with mechanical load and identified additional regulated genes. Kelly et al. (2016) assessed the transcriptional profile of cortical and cancellous bone following a single session of tibial compression in 10-week-old female C57BL/6 mice [31]. Three hours after loading, more than double the number of genes were differentially regulated in cortical bone compared to trabecular bone, suggesting a more sensitive and acute response of cortical bone to loading. Wnt signalling was upregulated in both cortical and cancellous bone in response to loading, but its upregulation persisted longer in cortical bone. Additional differentially expressed genes known to affect bone remodelling included Ngf, Ptgs2, Bmp2, Dmp1 and Tnfrsf11b. Using a different method (laser capture followed by microarray), Harris and Silva (2021) compared RNA from loaded tibial cortical bone to the unloaded contralateral control [26]. They identified over 150 genes that were differentially regulated 4 hours after a single loading bout or after 5 days of daily loading. Only 10 genes were differentially expressed in response to loading at both time points, including Ngf and Wnt1 as identified in the previous studies. Multiple transcriptomic studies of response to mechanical loading have consistently shown involvement of Ngf and Wnt signalling, however, other genes (e.g. Dot1l, Kcne3, etc.) identified in these studies remain to be interrogated for their functional significance.

The response of bone to mechanical loading diminishes with age and at least two studies have reported a reduced transcriptomic response. Galea et al. (2016) compared transcriptomic profiles in 19-week-old and 19-month-old mice subjected to a single session of tibial compression and collected at 1-, 3-, 6-, 12-, 18- or 24-hours post-loading, using unloaded contralateral tibias as controls [61]. Temporal analysis of the pathways following mechanical loading suggested that aged mice show similar transcriptomic activation to younger mice at early time points, but their response is brief compared to younger mice. Similar observations were made when loading responses were compared between 5 month and 22-month-old female C57BL/6N mice [62]. The number of differentially expressed genes in 5-month-old mice, compared to the unloaded contralateral tibia, progressively increased with time after loading (from day 1 to day 5), but the older mice had fewer differentially expressed genes at each time point. This confirmed that with mechanical loading, age is associated with a lower level and shorter duration of transcriptional response. Besides emphasizing that age influences the mechanical loading response of osteocytes, this study also highlighted a few candidate genes that may be targets for the restoration of bone mechano-responsiveness in old mice, such as Ngf, Notum, prostaglandin signalling, Nell-1, and genes within the AP-1 family of transcription factors.

4. Osteocyte transcriptome in rare bone diseases

Mutations in genes that are typically expressed by osteocytes have been associated with rare bone diseases. For instance, congenital mutations in the SOST gene, which encodes sclerostin, an inhibitor of the bone anabolic Wnt signaling pathway, cause the high bone mass conditions of sclerosteosis and Van Buchem disease [63–66]; mutations in PHEX, a transmembrane endopeptidase involved in bone mineralization and renal phosphate handling, cause X-linked hypophosphatemic rickets [67]; mutations in WNT1 are associated with early-onset osteoporosis when monoallelic or with a recessive form of osteogenesis imperfecta (OI) in biallelic mutations [68–70]. This underscores an important contribution of osteocytes in the etiology of rare genetic bone diseases. However, whether these mutations have an overall impact on the osteocyte transcriptome is less clear.

Initial evidence on this aspect has come from the study of OI, a bone fragility disorder most commonly caused by mutations or alterations in type I collagen [71]. Zimmerman et al have shown that the osteocyte transcriptome of long bones from two mouse models of OI (the CrtapKO and the oim/oim mutant) was similarly and strongly dysregulated, with as many as 281 differentially expressed shared transcripts compared to wild type controls [19]. Not surprisingly, these transcripts encode genes involved in processes such as collagen processing, cell adhesion, cell signaling pathways (including many genes involved in Wnt signaling and targets of the Tgfb signaling pathway), and others. Another study reported similar findings using the heterozygous Jrt mouse model of OI in addition to oim/oim mice in the calvarial bone transcriptome and identified 4 genes (Cgref1, Slc13a5, Smpd3, and Ifitm5) that were commonly upregulated in all three mouse models of OI used in these studies [72]. However, the calvarial samples were not prepared to enrich for osteocyte cells, suggesting contaminants may have been present. These examples provide initial evidence that the osteocyte transcriptome can be significantly impacted by congenital mutations in genes known to regulate bone structure and strength. However, whether the observed changes in transcription are the result of primary osteocyte cell-autonomous changes or are secondary passive adaptation to or interaction with a modified environment (e.g. the abnormal ECM of OI bone) remains to be established. In the specific example of OI, the direct or indirect effects of type I collagen mutations on osteocytes are still incompletely understood.

5. Conclusions, Limitations and Future Research Directions

In vertebrates, few genes, if any, are expressed by a single cell type but, if RNA expression levels correlate with function, then transcriptomic studies such as those performed in osteocytes can provide an unbiased picture of their transcriptional program under baseline conditions and in response to specific stimuli; such findings will point to unique and shared cellular activities and the signalling pathways that control them. Considering this, transcriptomic analyses of osteocytes have identified many genes associated with osteocyte identity and function. Several have been cross-referenced with functional data [32, 28], but since a substantial proportion remain functionally undefined in osteocytes, caution is needed when deducing conclusions from comparative transcriptomics since differential gene expression does not necessarily translate into causation. Clearly there is much more functional analysis to be done.

One method to achieve this will be to use cell type-specific loss-of-function studies to understand the role of many genes expressed by osteocytes. Current Cre-drivers (e.g. Dmp1-Cre, Sost-Cre) may be used in conjunction with existing loxP mouse lines, or novel CRISPR interference strategies to assess the function of genes in osteocytes [73]. Although these current Cre-drivers suffer from lack of specificity [74–77, 29, 78, 79], future identification of osteocyte-specific transcripts may lead to the development of improved Cre-driver systems for targeting these cells; in this way transcriptomics provide both new targets and new tools for functional studies.

Significant advancements in cell isolation methods for transcriptomics have progressed over recent years for other tissues including cold protease digestion [23], spatial transcriptomics [80], and single cell sequencing has been applied to a range of bone cells [81–84], see [20] for an in-depth review. However, osteocyte transcriptomics at spatial or single cell levels have not yet been achieved due to the difficulty of working with calcified tissue (as outlined above). The optimisation of these methods to mineralized tissues may provide new possibilities for examining osteocyte heterogeneity in the future and lead to further discoveries about the function of these cells that are so central to bone biology.

Future research on the impact of common or rare skeletal diseases on the osteocyte transcriptome will be important for the discovery of the cell autonomous contribution of these cells to disease manifestations. Such studies may be difficult to perform in patient samples and will require the generation of mouse models that accurately mimic the human condition. Osteocytes may indeed offer an untapped source of therapeutic options for bone diseases [85].

Finally, current transcriptomic approaches, by identifying a gene signature of osteocytes should be able to provide specific osteocyte markers. If specific osteocyte markers, or a combination or markers, can be found that do not overlap with osteoblasts or with other cell types found within bone, such as neuronal and endothelial cells, this will enable the purification and study of osteocytes from humans and other species. Such an approach could lead to new understandings of how osteocytes contribute to human skeletal disease and will be particularly useful in multifactorial conditions such as osteoporosis.