Bone marrow adipose tissue as an endocrine organ: close to the bone?

Adiponectin

We first investigated the potential endocrine function of MAT in response to two surprising observations. The first of these is the so-called ‘adiponectin paradox’: despite being produced almost exclusively by adipose tissue, circulating adiponectin concentrations inversely correlate with peripheral adiposity. Thus, hypoadiponectinaemia occurs in obese and insulin-resistant subjects but hyperadiponectinaemia exists in states of leanness, including patients with anorexia nervosa and in animal models of CR [21], [24], [25], [26]. Hypoadiponectinaemia in obesity likely results from decreased production from WAT, but the mechanisms underlying hyperadiponectinaemia in CR have remained elusive [26]. Why might circulating adiponectin increase in states of leanness? A potential answer emerged in light of a second surprising observation: in stark contrast to WAT, MAT increases during CR in animals and humans [8], [27], i.e. the very states when circulating adiponectin is also increased. Moreover, both MAT and circulating adiponectin are increased in many other clinical conditions, including oestrogen deficiency, type I diabetes, ageing, and in response to pharmacological agents such as glucocorticoids, thiazolidinediones and fibroblast growth factor-21 [4], [25], [28], [29], [30], [31], [32], [33]. Our findings have since extended this list, revealing that both MAT expansion and hyperadiponectinaemia occur in cancer patients undergoing chemotherapy or radiotherapy [8]. These observations led us to ask the question: does MAT contribute to circulating adiponectin?

Adiponectin production by MAT

Previous studies have demonstrated adiponectin expression in whole BM of the long bones of mice [34], [35], [36], [37], [38]; in adipocytes differentiated in vitro from mouse or human BM mesenchymal stromal cells (MSCs) [39], [40]; in BM adipocytes isolated from mouse tibiae and femurs [28], [41]; and in human femoral BM adipocytes, both in situ [42] and after isolation and culture ex vivo [43] (Table 1). We extended these findings by demonstrating that adiponectin is expressed in intact MAT depots, including caudal vertebrae of mice and in tibial MAT of rabbits and humans [8] (Table 1). We found that, relative to other typical adipocyte transcripts or proteins, adiponectin expression in these MAT depots is greater than that in WAT [8]. This finding is inconsistent with other results suggesting that adiponectin expression in adipocytes isolated from mouse or human BM is lower than that in WAT adipocytes [28], [61]. However, this discrepancy may reflect site-specific differences in MAT characteristics: while our initial analysis revealed greater adiponectin expression in tibial MAT than in WAT of humans [8], we have since found that, in human femoral MAT, adiponectin expression is not always higher than in WAT [26]. Indeed, we recently revealed that MAT exists in distinct subtypes, designated constitutive MAT (cMAT) and regulated MAT (rMAT), with different characteristics [62]. The former develops in the earlier postnatal stages and exists in more distal skeletal sites, such as the distal tibia and caudal vertebrae of rodents. This cMAT also contains few visible haematopoietic cells, thereby appearing histologically similar to WAT. In contrast, rMAT exists in more proximal regions, including the proximal femur, and is comprised of adipocytes interspersed with haematopoietic populations. Lipid composition and response to external stimuli also differs between rMAT and cMAT [62]. Our findings in human tibial and femoral MAT thus suggest that these MAT subtypes may also differ in adiponectin expression [26].

Table 1:

Endocrine and paracrine factors expressed and/or secreted by BM adipocytes.

Protein expression is based on immunoblots or immunohistochemistry. Transcript expression is based on the indicated methods. Boxes are shaded in green or turquoise if expression or secretion, respectively, has been reported. If expression or secretion remains unknown, boxes are shaded in light green or light blue, respectively. ‘Other factors’ lists microarray-detected transcripts, some of which encode proteins that are better known by the following names: Cfd, Adipsin; Apln, apelin; Ctsk, cathepsin K; Nampt, visfatin; Ccl2, MCP-1; Ptgs2, COX-2; Cxcl8, interleukin 8; Fgf7, keratinocyte growth factor; Serpinf1, pigment epithelium-derived growth factor.

To investigate if MAT also secretes adiponectin, we cultured explants of MAT and WAT from rabbits and humans, ex vivo, and analysed adiponectin secretion into the explant-conditioned media. This revealed that, in both species, secretion of adiponectin was far greater from MAT than from WAT, with enhanced secretion from human explants also noted for LMW, MMW and HMW adiponectin multimers [8]. Importantly, these increases are not attributable to a general increase in MAT protein secretion or breakdown of the MAT explants [8]. These observations build on previous studies demonstrating that adiponectin is secreted from adipocytes differentiated ex vivo from human BM [46] and from primary adipocytes isolated from human femurs [43], [49] (Table 1). Indeed, measurement of adiponectin secretion has recently been suggested as a marker of adipogenesis in cultured BM-derived mesenchymal stem cells, further underscoring the capacity of BM adipocytes to secrete this adipokine [47]. Although the mechanisms underlying enhanced adiponectin secretion from MAT relative to WAT remain to be determined, there are many possibilities and technical approaches that are poised to shed further light on this issue [26].

While the above studies confirm that MAT expresses and secretes adiponectin, additional evidence suggests that MAT also influences circulating adiponectin levels. We originally investigated this role of MAT through studies in Ocn-Wnt10b mice, in which a Wnt10b transgene is overexpressed from the osteocalcin promoter. We found that Ocn-Wnt10b mice resist both CR-associated MAT expansion and hyperadiponectinaemia, demonstrating that MAT expansion is required for the full increases in circulating adiponectin [8]. Other recent studies further support this conclusion. For example, separation-based anorexia, a unique CR model, causes many effects typical of CR, including weight loss and hypoleptinaemia, but without leading to hyperadiponectinaemia [63]. Notably, MAT expansion also fails to occur in this model (Christophe Chauveau, personal communication). Similarly, we recently revealed that CR in rabbits leads to decreased body mass, WAT mass, and circulating leptin, but without hyperadiponectinaemia or MAT expansion [55]. Thus, findings in three distinct animal models collectively suggest that MAT expansion is required for hyperadiponectinaemia during CR, supporting the conclusion that MAT contributes to increased circulating adiponectin in this context (Figure 1, Table 1).

Figure 1:

Endocrine and paracrine functions of bone marrow adipose tissue.

BM adipocytes and/or intact MAT can express and/or secrete a diverse range of secreted factors. In some cases expression has been confirmed but secretion has not yet been demonstrated; these factors are shown in thin-bordered boxes at the end of arrows with dotted borders. Those factors for which secretion has been demonstrated in vitro or ex vivo are shown in thick-bordered boxes at the end of arrows with solid borders. At present, adiponectin is the only endocrine factor for which MAT has been shown to influence circulating concentrations; hence, adiponectin is highlighted in a black box. MAT-derived adiponectin is implicated in regulating CR-associated adaptations in skeletal muscle, suggestive of effects on mitochondrial function. Other putative systemic effects of adiponectin or other MAT-derived endocrine factors remain to be determined, although it is possible that MAT may have indirect effects by regulating production of osteoblast-secreted factors such as osteocalcin, SPARC or osteopontin. It is unclear how local production of leptin or adiponectin might influence osteoblasts and osteoclasts in vivo, although positive and negative effects for other MAT-secreted paracrine factors have been demonstrated more conclusively. These positive and negative effects are indicated by red or green circles containing ‘+’ or ‘–’, respectively. Inconclusive effects are indicated by question marks in grey circles. Few of the studies supporting these findings have yet distinguished between rMAT and cMAT, and therefore future research must investigate if these paracrine and endocrine functions of MAT vary in a site-specific manner. Abbreviations are defined in the text.

Endocrine functions of MAT-derived adiponectin

What is the endocrine impact, if any, of MAT as a source of circulating adiponectin? Our studies in Ocn-Wnt10b mice revealed that CR-associated adaptations in skeletal muscle are impaired when MAT expansion and hyperadiponectinaemia are suppressed (Table 2). Specifically, we found that skeletal muscle expression of transcripts with roles in mitochondrial function, including Ppargc1a, Tfam and Acadm, is increased during CR in wild-type mice, but this response is completely absent in Ocn-Wnt10b mice [8]. This effect is consistent with adiponectin promoting mitochondrial biogenesis and function in skeletal muscle [87] (Figure 1). In contrast, skeletal muscle AMPK activity is elevated during CR in Ocn-Wnt10b mice but not in wild-types, which is at odds with the reported ability of adiponectin to activate AMPK [88]. Thus, it remains unclear if these systemic effects of impaired MAT expansion result from relative adiponectin deficiency, or if other MAT-derived endocrine factors are playing a role.

Table 2:

Clinical observations, in vitro and in vivo evidence for or against the putative endocrine and paracrine functions of MAT.

Green and red boxes denote evidence for or against each putative function, respectively. Decreased haematopoiesis in Ocn-Wnt10b mice is based on unpublished observations by Erica Scheller.

It is also possible that, as a source of adiponectin, MAT has systemic impacts beyond skeletal muscle. Indeed, adiponectin has diverse systemic actions, including increased systemic glucose tolerance, enhanced hepatic insulin sensitivity and beta-cell function, improved cardiovascular function, anti-inflammatory and anti-cancer effects [22] (Figure 1). We found that the ability of CR to improve glucose tolerance and modulate hepatic transcription did not differ between wild-type and Ocn-Wnt10b mice, despite resistance of the latter to MAT expansion and hyperadiponectinaemia [8]. However, although CR-associated MAT expansion and hyperadiponectinaemia are blunted in Ocn-Wnt10b mice, it is important to emphasise that these responses still occur [8]. It is therefore possible that the degree of hyperadiponectinaemia in Ocn-Wnt10b mice, while diminished, remains sufficient to impact glucose tolerance, hepatic transcription, and other systemic targets. This possibility underscores the need to develop improved animal models that more robustly resist MAT expansion, or in which MAT formation is prevented entirely.

Leptin

Another possibility is that MAT exerts systemic effects independently of adiponectin, via production of other endocrine factors. While our research has focused primarily on adiponectin, BM adipocytes have also been found to produce leptin, which may extend the endocrine repertoire of MAT. Laharrague et al. were the first to report leptin expression and secretion by adipocytes differentiated in vitro from human BM MSCs [51], an observation since repeated elsewhere [46], [52], [53]. It remains unclear if such BM-derived adipocytes faithfully recapitulate the characteristics of BM adipocytes in vivo; however, it is worth noting that leptin is also secreted by human primary BM adipocytes in culture [43], [54]. Moreover, microarray analyses by Liu et al. have detected leptin expression in isolated BM adipocytes from mice [28], while we recently revealed leptin transcript expression in distal tibial MAT of rabbits [55] (Table 1). These latter findings demonstrate that leptin production by BM adipocytes is not limited to ex vivo culture, supporting the possibility that MAT contributes to circulating leptin in vivo. However, this possibility presents a paradox: if MAT does indeed contribute to circulating leptin, why does hypoleptinaemia occur during CR, when MAT volume is increased? Our recent findings provide a potential answer in that, as in WAT, leptin expression in MAT is decreased during CR [55]. This suggests that during CR, MAT expansion may be offset by decreased leptin production, thus ensuring that increased MAT does not undermine the hypoleptinaemia that is important for homeostatic adaptations to CR.

The above findings demonstrate that BM adipocytes express leptin and that, as in WAT, this is suppressed during CR. Laharrague et al. found that the pro-inflammatory cytokines IL-1β, IL-6, tumour necrosis factor-α (TNF-α) and interferon-γ reduce both expression and secretion of leptin from cultured BM adipocytes, further demonstrating that these cells produce leptin in a regulated manner [52]. These authors previously suggested that, given the relatively large amount of MAT in humans, the degree of leptin secretion detected from BM adipocytes is likely sufficient to contribute to circulating leptin concentrations [51]; however, in contrast to adiponectin, this remains to be determined (Figure 1). One possibility is that rather than exerting systemic endocrine functions, MAT-derived leptin acts predominantly within the BM as an autocrine and paracrine factor. Indeed, numerous studies report local effects of leptin within the skeleton, with potential consequences on skeletal remodelling, haematopoiesis and tumour progression (Figure 1). These possibilities are discussed further below (see ‘Close to the bone: paracrine functions of MAT’).

Other adipokines

The identification of leptin and adiponectin presaged the discovery of numerous additional adipokines capable of exerting diverse local and systemic effects. Notable examples include retinol-binding protein 4 (RBP4), adipsin, resistin, chemerin, and cytokines such as tumour necrosis factor-alpha (TNF-α), interleukin 6 (IL-6), monocyte chemoattractant protein-1 and plasminogen activator inhibitor-1 (PAI-1) (Table 1); the latter is implicated with elevated atherogenic risk (Figure 1). More recent findings suggest that fatty acid-binding protein 4 (FABP4), also known as adipocyte protein 2 (aP2), is an adipokine that can mediate systemic effects [89], [90]. Microarray analyses confirm that primary BM adipocytes express each of these adipokines, with Il6 and Tnf transcripts expressed more highly in adipocytes from MAT than in those from WAT [28]. Another microarray study of whole mouse vertebrae further confirms that intact MAT expresses RBP4 and adipsin [50], while transcripts for chemerin and other secreted factors are expressed in BM MSC-derived adipocytes [45], [56] (Table 1). Moving beyond mRNA, several studies demonstrate expression and/or secretion of these factors at the protein level (Table 1). For example, we found that FABP4 is expressed in lysates of intact rabbit MAT [8], while and TNF-α and PAI-1 are expressed and secreted by primary BM adipocytes in culture [49], [48]. Finally, BM MSC-derived adipocytes can also secrete chemerin [56], which, like TNF-α, can modulate inflammation within WAT (Figure 1). It remains to be determined if BM adipocytes secrete these adipokines in vivo and if this influences their concentrations in the circulation (Table 1); hence, the relevance of these adipokines to MAT’s putative endocrine function requires further study. However, many of these factors have been suggested to mediate paracrine effects within the BM, highlighting another potentially important function of MAT as a secretory organ (Figure 1). This possibility is further discussed later.

Lipid species

A hallmark of adipocytes is their ability to catabolise their stores of triacylglycerol in a process known as lipolysis. This leads to adipocyte release of glycerol, which is used by the liver for gluconeogenesis; and free fatty acids (FFAs), which are used in other tissues to produce ATP via beta-oxidation. However, the FFAs released by adipocytes can serve as more than just metabolic substrates: they are also able to play a signalling role to impact metabolic states, for example by causing changes in insulin signalling. It has been shown that toll-like receptor 4, which normally functions as a lipopolysaccharide receptor, can also respond to the presence of elevated levels of FFAs in the circulation, causing increased inflammatory signals in WAT [91] (Figure 1). Similarly, palmitoleate release from WAT has been suggested to inhibit hepatic steatosis and to promote insulin sensitivity in skeletal muscle [92], suggesting that adipose-derived circulating lipids can exert endocrine effects. Earlier research in dog models shows that MAT can release FFAs into the circulation in vivo in response to perfusion of epinephrine or isoproterenol, albeit to a lesser extent than omental WAT [64]. Follow-up research in dog tibiae also confirmed the release of FFA from MAT was regulated, in part, due to an interplay between α2- and β-adrenoreceptors [93]. Thus, MAT-derived FFA might contribute to circulating lipids and thereby act as a peripheral energy supply (Figure 1). However, other studies suggest that BM adipocytes do not undergo lipolysis even during acute starvation [66], raising questions over whether MAT contributes to circulating lipids (Table 2).

The lipid composition of MAT is a subject that is attracting increasing attention. Early chromatographic analyses of human BM revealed that it consists almost exclusively of triglycerides, with free cholesterol and phospholipids the only other significant proportions [94]. A more recent study using ¹H NMR spectroscopy showed very similar degrees of fatty acid saturation between WAT and tibial MAT in humans [95], suggesting that the lipid content of MAT is similar to that of WAT. However, it has since been revealed that there are site-specific differences in MAT lipid composition, with the proportion of unsaturated fatty acids, such as palmitoleate, being greater in cMAT than in rMAT [62]. This is consistent with earlier observations [96] and suggests that the potential for MAT to exert lipid-mediated endocrine effects may vary among MAT subtypes. Increasing clinical evidence further underscores the potential importance of MAT lipid composition. For example, MRI studies show that increased fatty acid saturation in MAT can be used as a predictor for increased bone fracture risk [74], and that the degree of unsaturation is lower in patients with type 2 diabetes [97]. Thus, increased saturation in MAT of type 2 diabetics might contribute to increased fracture risk in type 2 diabetes. In addition, saturated fatty acids can suppress the expression and secretion of adiponectin [98]; hence, from an endocrine perspective, differential lipid saturation may influence adipokine production from rMAT and cMAT.

Collectively, these observations underscore the potential interplay between MAT location, lipid content, endocrine function and human health; however, as for the other adipokines discussed above, whether MAT contributes to systemic lipid metabolism, or if MAT-derived lipid species exert endocrine effects, remains to be firmly established (Figure 1). The accumulation of MAT during CR suggests that it is resistant to stimuli that drive lipolysis in WAT, consistent with the work of Bathija et al. [66]. Nevertheless, MAT breakdown can occur if the degree and duration of CR is sufficient [55], [99], [100], [101], [102], [103], suggesting that BM adipocytes can undergo lipolysis in certain contexts. Therefore, the possibility that MAT contributes to circulating lipids remains an intriguing possibility.

MAT and skeletal remodelling

MAT is often increased in conditions of decreased bone mass and increased fracture risk [4], [107], underscoring the possibility that MAT directly promotes bone loss and/or inhibits bone formation [107], [108]. Thus, many researchers have investigated the hypothesis that BM adipocytes directly regulate the formation and/or function of osteoblasts and osteoclasts. For example, several studies demonstrate that, during in vitro co-culture, primary or MSC-derived BM adipocytes stimulate osteoclastogenesis [58], [59], [60], [44], [57] and compromise osteoblast formation, proliferation or function [67], [68]. Treatment with BM adipocyte-conditioned media also enhances osteoclastogenesis [57]. These in vitro results suggest that BM adipocytes secrete factors that directly promote bone loss (Table 2). Potential candidates include conventional MAT-derived adipokines, cytokines, lipid species and other secreted factors. For example, both leptin and adiponectin are implicated with the regulation of skeletal remodelling, and receptors for each of these adipokines are expressed by osteoblasts [34], [109], osteoclasts [35] or osteoclast precursors [110]. However, while some studies find that leptin or adiponectin stimulate the formation or function of osteoblasts and inhibit these processes for osteoclasts, other experiments show the opposite effects [104], [106], [110], [111], [112], [113]; hence, precisely how these adipokines impact bone formation remains controversial (Figure 1). Other adipokines, including resistin, chemerin, apelin and visfatin, can also regulate the differentiation, function and/or survival of osteoclasts or osteoblasts [56], [114], [115], [116], but it remains unknown if MAT secretes these factors in vivo (Table 1). Similarly, FFA release from BM adipocytes may directly impair skeletal integrity by inhibiting the proliferation, differentiation or survival of osteoblasts while stimulating these processes for osteoclasts [106], [69], [117], [70] (Figure 1). This is consistent with the aforementioned finding that fatty acid saturation in MAT is positively associated with fracture risk in type 2 diabetic patients, supporting the possibility that BM adipocyte-derived fatty acids may directly impact bone integrity [74]. However, it remains unknown if this correlation reflects any causal role for MAT in compromising bone strength, a hypothesis that would benefit from further mechanistic studies in vivo (Table 2).

Beyond these adipokines, many cytokines also profoundly affect the formation and function of osteoblasts and osteoclasts. One essential regulator of osteoclastogenesisis is receptor activator of nuclear factor κB ligand (RANKL), a member of the TNF family that signals via its receptor, RANK. Osteoblasts secrete RANKL to stimulate osteoclastogenesis from osteoclast precursors, and therefore RANKL production is strongly linked with bone resorption, osteoporosis and arthritis [118], [119], [120]. RANKL transcripts are upregulated during adipogenesis of BM MSCs, and pro-resorptive stimuli such as dexamethasone and TNF-α significantly increase RANKL mRNA in cultures of human primary BM adipocytes [58], [59], [60]. Similarly, one microarray has detected RANKL transcripts in mouse primary BM adipocytes [28]. While these findings are limited to BM adipocytes in culture, another report reveals that BM of aged mice has increased RANKL expression [44], suggesting that BM adipocytes may also express RANKL in vivo. Functional studies in vitro further demonstrate that, in co-culture with osteoclast precursors, human primary BM adipocytes stimulate osteoclastogenesis in a RANKL-dependent manner [58], [59]; however, BM adipocyte production of RANKL was not assessed. MAT might further regulate RANKL activity by production of osteoprotegerin (OPG), an osteoblast-secreted protein that acts as a decoy receptor for RANKL, thereby inhibiting osteoclast-mediated bone resorption [118]. As for RANKL, primary BM adipocytes from mice and humans express OPG transcripts (Table 1). Some studies find that dexamethasone increases OPG expression in BM adipocytes [60], whereas others find no effect [58]; however, dexamethasone and TNF-α consistently increase the RANKL/OPG transcript ratio in human BM adipocytes [58], [59], [60], and one report suggests downregulation of OPG during BM MSC adipogenesis [44]. Thus, increased BM adiposity may augment the local RANKL/OPG balance and thereby contribute to osteoclast-mediated bone resorption (Figure 1).

In addition to RANKL, the other key regulator of osteoclastogenesis is macrophage colony-stimulating factor (M-CSF) [118]. Whereas RANKL stimulates osteoclast precursors to undergo osteoclastogenesis, M-CSF supports the proliferation and survival of these precursors, as well as upregulating their expression of RANK [118]; the latter is a defining feature of osteoclast precursor identity. As for RANKL and OPG, primary BM adipocytes from mice and humans express M-CSF mRNA (Table 1), suggesting that MAT may impact bone resorption via M-CSF secretion (Figure 1). The ability of BM adipocytes to secrete M-CSF awaits confirmation.

Other BM adipocyte-derived secreted factors may regulate skeletal remodelling by influencing RANKL or OPG production by neighbouring cell types. For example, in osteoblasts adiponectin can stimulate RANKL production and inhibit OPG expression [121]. TNF-α or IL-6 can also stimulate osteoclastogenesis and bone resorption by modulating the RANKL/OPG pathway [105] (Figure 1), and therefore it is notable that primary BM adipocytes express these factors more highly than adipocytes from WAT [28] (Table 1). Other MAT-derived cytokines may also promote bone resorption. For example, BM MSC-derived adipocytes secrete chemokine (C-X-C motif) ligand 1 (CXCL1) and chemokine (C-X-C motif) ligand 2 (CXCL2) to promote osteoclastogenesis, at least during in vitro co-culture with osteoclast precursors [57]. In addition to these cytokines, BM adipocytes express cathepsin K [28], a protease implicated in bone resorption [118], while oxidised lipids can also stimulate RANKL production by T lymphocytes [122]. More recently, BM adipocytes were found to secrete extracellular vesicles allowing transfer of adipogenic RNAs to osteoblasts, which may impair osteoblast function [71]. Collectively, these findings support the possibility that BM adipocytes impact osteoblasts and osteoclasts and stimulate bone resorption by secretion of cytokines, proteases, lipid species and adipogenic RNAs (Figure 1); however, as above for RANKL, OPG and M-CSF, whether this occurs in vivo is unclear (Table 2).

MAT and haematopoiesis

It has been known for over 130 years that the fundamental function of BM is haematopoiesis [125]. This initial discovery motivated further studies of BM anatomy, which firmly established the existence of adipocytes within BM [77], [126]. It is therefore unsurprising that the earliest studies of MAT focused on its relationship with haematopoiesis [99], [77], [80], [127], [128]. Although MAT accumulation is a normal developmental process, clinical evidence clearly demonstrates that BM adiposity associates negatively with haematopoiesis. For example, MAT increases in aplastic anaemia [81] but regresses in response to increased erythropoietic demand, such as during hypertensive heart failure or haemolytic conditions [80], [129]. Radiation treatment for BM transplantation is also associated with marked MAT expansion that subsides only after development of haematopoietic precursors [5], suggesting that the former may somehow stimulate the latter. Indeed, MAT expansion has often been reported as ‘fatty degeneration’ of the red marrow [130], reflecting the concept that increased MAT must necessarily compromise BM haematopoietic potential. While this could simply result from the gross balance between the amounts of haematopoietic ‘red marrow’ and non-haematopoietic ‘yellow marrow’ [79], it has also been proposed that BM adipocytes directly modulate haematopoietic stem cell (HSC) differentiation through paracrine effects [107], [108].

Several lines of evidence support such a function. For example, many adipocyte-derived secreted factors can regulate haematopoiesis, including adiponectin, leptin, prostaglandins and sex steroids [51], [131], [132]. One earlier in vitro study suggested that leptin drives haematopoietic precursors toward myelopoiesis and away from other lineages [131], while subsequent work in leptin-deficient ob/ob mice suggests that leptin sustains both myelopoiesis and lymphopoiesis in vivo [133]. Adiponectin can also promote HSC proliferation in vitro, while HSCs cultured with adiponectin display enhanced reconstitution capacity after transplantation in vivo [36]. Fatty acid metabolism also plays a crucial role in HSC proliferation and function [134], [135], and therefore it has been proposed that BM adipocytes may influence haematopoiesis by acting as a local source of fatty acids. This is supported by the observations of Tavassoli and Scheller, showing that fatty acid content varies between adipocytes located within haematopoietic or non-haematopoietic BM (i.e. rMAT and cMAT, respectively) [62], [96], [129].

As for skeletal remodelling, co-culture approaches have been used to further investigate if BM adipocytes can directly impact HSC function. One study found that primary human BM adipocytes suppress granulopoiesis of CD34⁺ HSCs, albeit only when co-culture allowed cell-to-cell contact; when cells were co-cultured using transwells, this effect did not occur [54]. While this is consistent with the concept that MAT suppresses haematopoiesis, it argues against a role for BM adipocyte-derived secreted factors. Similarly, an earlier co-culture study allowing direct cell-to-cell contact found that human BM MSC-derived adipocytes influenced haematopoiesis of CD34⁺ progenitors, although in this case the adipocytes supported, rather than suppressed, both myelopoiesis and lymphopoeisis [53]. Thus, direct contact between BM adipocytes and HSCs may influence haematopoiesis, although whether this is stimulatory or suppressive remains unclear (Figure 1).

These inconsistent in vitro findings raise the question of whether MAT directly modulates haematopoiesis in vivo. One notable study in mice found that, post-irradiation, BM engraftment is enhanced when MAT formation is impaired [79]. This observation was based on two mouse models: A-ZIP/F1 mice, which, as mentioned above, lack the capacity to form adipose tissue; and mice treated with bisphenol A diglycidyl ether (BADGE), which inhibits adipogenesis in BM and other adipose depots. However, it is worth emphasising that neither the A-ZIP/F1 transgene nor BADGE treatment is specific for MAT, such that each also affects WAT formation and other non-adipose cell types. This could have clear confounding effects, for example via markedly decreased circulating leptin concentrations. Thus, these models alone cannot conclusively establish whether MAT directly modulates haematopoiesis in vivo. Further uncertainty arises when other in vivo models are considered. For example, increased MAT in Ebf1-knockout mice is not associated with impaired haematopoiesis [82]. Conversely, Ocn-Wnt10b mice have decreased rMAT [26] but impaired haematopoiesis within the BM, as indicated by increased extramedullary haematopoiesis in the spleen (Erica Scheller, unpublished observations). While impaired haematopoiesis in Ocn-Wnt10b mice is likely influenced by their decreased BM volume, these observations demonstrate that increases or decreases in BM adiposity are not always associated with opposite effects on haematopoietic function in vivo (Table 2). Thus, the potential impact of MAT on haematopoiesis, including the contribution of BM adipocyte-derived secreted factors, remains to be firmly established.

MAT and tumour progression

The above studies underscore the longstanding interest in MAT as a potential regulator of skeletal remodelling and haematopoiesis. More recently, researchers have begun to investigate the impact of BM adipocytes on tumour progression within bone, with most studies focusing on the hypothesis that MAT promotes tumour development and invasiveness [105]. This concept has support from some clinical observations. For example, MAT is increased with ageing and, in some cases, during obesity [136], both of which are risk factors for aggressive prostate cancer (as discussed in [105]) (Table 2). Mechanistically, BM adipocytes may influence tumour progression by stimulating osteoclast activity, which is closely associated with the development of skeletal metastases [137]. In support of this hypothesis, increased BM adiposity is associated with elevated levels of pro-inflammatory factors such as CCL2, COX-2, and the chemokines CXCL1 and CXCL2, each of which is implicated in osteoclastogenesis and tumour growth within bone [105], [57], [138]. Based on these findings, Hardaway et al. further investigated the role of BM adipocyte-derived CXCL1 and CXCL2 in prostate tumour-associated bone resorption. They found that, during transwell-based co-culture between BM MSC-derived adipocytes and prostate tumour cells, the adipocytes’ expression and secretion of CXCL1 and CXCL2 was increased [57]. Thus, unlike many of the other putative regulators of bone remodelling, there is direct evidence that BM adipocytes can secrete CXCL1 and CXCL2, at least in vitro (Table 1; Figure 1). These authors further revealed that BM adipocyte-conditioned media stimulates osteoclastogenesis in vitro, an effect prevented by the inclusion of neutralising antibodies against CXCL1 or CXCL2, or by blocking CXCR2, the receptor for these chemokines [57]. This study therefore provides compelling in vitro evidence that BM adipocyte-derived chemokines may promote bone resorption and prostate tumour invasiveness (Figure 1).

In addition to these effects on osteoclast activity and bone resorption, BM adipocytes can also promote metastatic progression through direct effects on tumour cells. Fatty acid oxidation is a key pathway in prostate tumours [139], and therefore the local secretion of fatty acids from MAT may help to sustain their growth and proliferation within bone. Consistent with this, BM adipocytes can transfer lipids to prostate tumour cells during in vitro co-culture [85], [83] and this is associated with increased tumour cell invasiveness and FABP4 expression [83]. Notably, prostate tumour cells in mouse models and clinical samples also display increased FABP4 expression, particularly in those cells within adipocyte-rich regions of the BM [83]. This is important, as it supports a role for this pathway in the progression of prostate tumour metastases within bone in vivo (Table 2; Figure 1).

Beyond chemokines and lipid species, classical adipokines such as leptin and adiponectin may also contribute to BM adipocytes’ impact on skeletal tumours. In vitro, leptin can stimulate prostate cancer cell growth by modulating MAPK activity (as discussed in [105]), and BM adipocyte-derived leptin might also promote metastatic growth within bone by stimulating bone resorption. However, it is worth emphasising that pro-proliferative effects of leptin have not been observed in all studies [140] and that, as noted above, leptin’s impact on skeletal remodelling remains controversial [104]. A recent report by Templeton et al. sheds further light on the relationship between leptin, MAT and tumour progression in humans. These authors found that conditioned media from human bone samples, which were highly enriched for MAT, stimulates migration of a breast cancer cell line and that the degree of migration is positively associated with leptin concentrations in the conditioned media [84]. This supports the possibility that BM adipocytes might promote breast cancer metastasis to bone via leptin production (Figure 1). However, the impact of MAT becomes less certain when adiponectin is also considered. Unlike leptin, most studies find that adiponectin suppresses proliferation and induces apoptosis in cancer cells in vitro, suggesting a role in limiting tumour progression [23] (Figure 1). Consistent with this, circulating adiponectin concentrations are inversely associated with the risk of developing endometrial cancer, postmenopausal breast cancer, colon cancer, renal cancer, leukaemia and other haematological malignancies [23]. As mentioned above, we recently revealed that both MAT and circulating adiponectin increase in cancer patients undergoing chemo- or radio-therapy [8], supporting not only the concept that MAT influences circulating adiponectin, but also raising another possibility: might MAT expansion actually exert anti-cancer effects?

MAT as a paracrine and endocrine organ: why context matters

This question highlights an important point: the influence of MAT on tumour progression, as well as skeletal remodelling and haematopoiesis, is likely to be highly context dependent. For example, while ageing is associated with increases in both MAT and cancer risk, other conditions featuring MAT accumulation, such as CR, are associated with decreased cancer risk [86]. Microarray analyses suggest that BM adipocytes are inherently more pro-inflammatory than adipocytes in WAT, and that ageing or diet-induced obesity alters the global characteristics of these cells [28], [41]. Such changes have clear potential to impact MAT’s ability to influence skeletal remodelling, haematopoiesis and tumourigenesis. For example, ageing is associated with decreased adiponectin expression in BM adipocytes [28], which might lessen their ability to suppress local tumour development and thereby contribute to age-associated increases in cancer risk. Conversely, we have shown that CR leads to decreased leptin expression in MAT [55], which may help to limit the progression of bone metastases during CR. As discussed above, rMAT and cMAT may have distinct secretory properties, with present observations suggesting increased adiponectin expression in cMAT [8], [26]. Thus, it is notable that bone metastases typically arise in the cMAT-deficient axial skeleton, whereas metastases in the ambulatory skeleton, which is cMAT-rich, seldom occur. Whether this relates to distinct secretory functions of rMAT vs cMAT, or to other context-dependent changes in BM adipocyte function, remains unknown. Future research must therefore address these important issues.

Endocrine consequences of MAT’s paracrine actions

While many questions remain to be addressed, interest in the potential paracrine actions of MAT continues to increase. However, an unexplored possibility is that these local effects might have endocrine consequences and thereby mediate systemic effects. For example, it is now known that the skeleton serves not only a structural role, but also produces endocrine factors such as osteocalcin, which has systemic effects on insulin sensitivity, glucose homeostasis and male fertility [141], [142]. In addition to osteocalcin, the bone-derived proteins SPARC and osteopontin also play important endocrine roles; SPARC (secreted protein acidic and rich in cysteine; also called osteonectin) influences bone mineralisation but can also have broad systemic effects, including modifying cancer progression and wound healing [143], [144], [145], [146]. Osteopontin is expressed by osteoblasts, osteocytes and other cell and tissue types and, like SPARC, can alter bone mineralisation [147]; however, osteopontin can also act at an endocrine level, with effects on immune function, inflammation and tumour metastasis [148], [149], [150]. Thus, osteocalcin, SPARC and osteopontin have diverse systemic effects. Bone might also exert endocrine effects via secretion of factors such as bone morphogenetic proteins, or as an important site for systemic glucose disposal [142]. In vitro co-culture studies have shown that BM adipocytes can suppress osteocalcin expression in osteoblasts [68], [70], [71], which suggests that MAT might influence endocrine pathways as a result of its local effects within bone (Figure 1). This is an intriguing possibility that awaits further study.