Introduction

The last two decades of research provided emerging evidence that bone marrow adipose tissue (BMAT), also called yellow marrow as opposed to the hematopoietic red marrow, is a unique fat depot. However, although remarkable advances have been performed in the understanding of its developmental origin, metabolism and function, the physiological roles of BMAT as well as its contribution to the etiology of bone diseases remain to be clarified.

Bone marrow adipocytes have been previously considered as inert space filling bystanders but this concept has radically changed. Currently, it is widely accepted that BMAT constitutes a metabolically active depot that contributes in modulating the bone marrow microenvironment, with putative subsequent impact on skeletal health [1, 2]. An excessive expansion of BMAT is definitely associated with the loss of bone quality observed in aging, unloading, and in pathological conditions such as osteoporosis, type 1 diabetes, and anorexia nervosa [3•].

Through the release of cytokines, adipokines, and free fatty acids into the bone marrow microenvironment, bone marrow adipocytes may indeed influence the biology of their neighboring cells. Depending on the pathophysiological context, BMAT could be a source of energy for bone cells, or conversely, when present in excess, it could disturb surrounding bone cell function and, therefore, interferes with skeletal homeostasis [3•].

At birth, marrow adipocytes are extremely sparse, but their exponential accumulation into the bone marrow is promptly observed during childhood and is followed by their gradual accretion throughout the adult life [4]. BMAT follows a centripetal development, from distal to proximal extremities of the bones, showing a similar pattern in all vertebrates. In humans, the middle phalanges of the toes are fully converted to BMAT by 1 year of age, while in 7-year-old humans, the entirety of feet and hand bones is filled by BMAT [4, 5] In 25-year-old humans, BMAT occupies about 70% of the bone marrow volume of the appendicular skeleton, the hematopoietic red bone marrow being restricted to the axial skeleton, ribs, and sternum [5].

The aim of the present review is to update the knowledge related to the molecular phenotype of BMAT.

Is BMAT Equivalent to White, Brown, or Beige Adipose Tissue?

In order to understand the function and the regulation of the BMAT, many investigators tried to classify bone marrow adipocytes among the two well-known different subtypes of adipose tissue. This is either white adipose tissue (WAT) which takes over the body energy storage and supply, and participates to the inflammatory process or brown adipose tissue (BAT), which dissipates energy in the form of heat in order to maintain the body temperature [6].

In contrast to the extramedullary adipocytes, lineage tracing studies performed in mice have shown that bone marrow adipocytes express Osterix, a transcriptional factor indispensable for osteoblast differentiation in mammals [7]. This observation suggests that bone marrow adipocytes and osteoblasts arise from a common progenitor cell, which is distinct from the progenitor of the adipocytes constituting the WAT and the BAT. It is now generally accepted that bone marrow adipocytes originate from skeletal stem cells (SSC), which are multipotent progenitor cells able to differentiate into several cell types such as chondrocytes, osteoblasts, and adipocytes [8, 9]. Nevertheless, bone marrow adipocytes present some similarities with the adipocytes located in the WAT and in the BAT.

Although displaying a smaller diameter than white adipocytes, the histological appearance of bone marrow and white adipocytes is quite similar. However, the BMAT does not present the homogeneous organization observed in WAT, since bone marrow adipocytes are scattered within the bone marrow and nested among the hematopoietic cells. Moreover, the accumulation of BMAT in the bone marrow is due to a combination of hypertrophy and hyperplasia of adipocytes whereas the enlargement of WAT is mostly the result of cellular hypertrophy, hyperplasia occurring uniquely in case of extreme obesity [10,11•]. BMAT also differs from WAT with respect to the impact of the nutritional status. Unlike WAT, the amount of BMAT is not strictly associated with body mass index (BMI) or with body fat. For example, in humans, obesity is not always linked to an increase of medullar adiposity [12, 13], whereas patients suffering from anorexia nervosa exhibit an excess of marrow adipocytes contrasting with the leanness of the subjects [14]. In vitro studies have revealed that the ability to secrete cytokines such as IL-6, IL-8, and macrophage-colony stimulating factor (M-CSF) is similar between human SSC-derived adipocytes, primary adipocytes isolated from bone marrow and adipocytes isolated from the WAT [15,16••]. By contrast, several studies have shown that human marrow adipocytes produce low quantities of TNFα and IL-1β compared to WAT [16••,17], although in a murine model, these cytokines are secreted by marrow adipocytes in larger amounts than by epididymal adipocytes (WAT) [18]. Of note, in mice, marrow adipocytes are characterized by a high expression of inflammatory genes and a weak expression of adipogenic genes [18].

More recently, characteristics specific for BAT were attributed to the adipocytes of the BMAT. In mice, marrow adipocytes express thermogenic genes such as deiodinase 2 (Dio2) and the peroxisome proliferator-activated receptor-gamma coactivator 1-alpha (PGC1α) [19]. Nevertheless, the expression of these genes is reduced with aging and the uncoupling protein 1 (UCP-1) is scarcely present compared to BAT. However, another study showed the presence of UCP-1 in marrow adipocytes derived from lumbar vertebrae of 3 weeks-old mice as well as from vertebrae of cold-exposed healthy humans [20]. Some authors also suggest that BMAT may exhibit properties specific to beige (or bright) adipocytes, which derive from a subtype of white adipocytes that are able to reversibly differentiate in brown adipocytes and to express UCP-1 [21].

In conclusion, regarding its expression of specific marker genes, its anatomical distribution and its response to the nutritional status, it turns out that BMAT is considered as a unique and typical fat depot, exhibiting its own specific properties while possessing some WAT- and BAT-like features as well.

Is BMAT Constituted by Two Sub-populations of Adipocytes?

Inspired by the study of Tavassoli et al. suggesting that adipocytes located in the red bone marrow are phenotypically different from those situated in the yellow marrow [22], Scheller and colleagues recently used osmium tetroxide staining combined with μCT visualization to demonstrate that in mice, marrow adipocytes may exist in two sub-populations, constituting a regulated and a constitutive type of BMAT (rBMAT and cBMAT, respectively). These types of BMAT are located in specific areas of the bone and revealed distinct temporal development as well as lipidomic profiles [23••].

The adipocytes of the cBMAT are formed early during development, present with a large size and are highly enriched in unsaturated lipids. Such an enrichment may be associated with an elevated expression of several desaturase enzymes, including the two isoforms of stearoyl-CoA desaturase, SCD-1 and -2, and of fatty acid desaturase, FADS-1 and -2. The adipocytes of the rBMAT are characterized by a smaller diameter and contain high levels of saturated lipids.

cBMAT and rBMAT might also follow a preferential distribution within the bone marrow. cBMAT seems mostly located in the distal area within the yellow marrow, whereas rBMAT may be found essentially in the proximal area, among the red marrow enriched in hematopoietic cells, and could be characterized by a high bone turnover. Extrapolating this BMAT distribution in humans, the shift in marrow fat composition described in diseases associated with high fracture risk, such as osteoporosis [24, 25•] and type 2 diabetes [26], as discussed below, and characterized by a rise of saturated lipids and a decrease of unsaturated lipids in the BMAT, could be linked to a prevailing development of rBMAT. Taking these findings together, the existence of cBMAT and rBMAT is still under debate, both in humans and in mice.

Does BMAT Present a Specific Secretory Profile?

When compared to white and brown adipocytes, the secretory profile of bone marrow adipocytes remains largely unexplored [6]. However, since co-culture experiments have revealed that marrow adipocytes exert paracrine effects that are deleterious for osteoblasts [27, 28] while beneficial for osteoclasts [29••], an increased number of studies now focuses on the characterization of the BMAT secretory profile.

Abdallah et al. have recently shown that mouse bone marrow adipocytes inhibit osteoblastic differentiation of bone marrow SSC by blocking BMP2-induced osteoblastogenesis and by activating the proinflammatory NF-κB signaling pathway [30•]. Moreover, marrow adipocytes secrete frizzled-related protein 1 (sFRP-1), an inhibitor of the Wnt signaling pathway and it was shown in co-culture experiments that depletion of sFRP-1 abolishes this anti-osteoblastic effect [31].

Interestingly, in humans, marrow adipocytes display the particularity to express RANKL and OPG, and the RANKL/OPG ratio increases following dexamethasone treatment [32, 33], a situation close resembling that observed in mature osteoblasts. In mice, RANKL expression rises while OPG is reduced throughout adipogenesis and aging, an observation further supporting the link between adipogenesis and osteoclastogenesis [34]. This is reinforced by the recent findings of Fan et al. demonstrating that murine marrow adipocytes, unlike extra-medullar adipocytes, secrete RANKL and that deletion of the parathyroid hormone receptor gene rises RANKL production, further supporting osteoclastogenesis [35].

Leptin and adiponectin are two well-studied adipokines produced by the WAT. Both are also secreted by marrow adipocytes, although to a lesser extent, and it has been shown that bone cells express the receptors of these two adipokines [16••, 19]. Released by white adipocytes in response to triglycerides storage and cell hyperplasia, leptin regulates body mass by modulating food intake and energy expenditure [36]. It also controls the skeletal metabolism by exerting two opposite actions: locally, leptin stimulates bone formation [37, 38] whereas via its sympathetic activity, the adipokine favors bone resorption by suppressing osteoblastic activity [39]. Nevertheless, human epidemiological studies attempting to correlate peripheral leptin levels and bone mass or fracture risk remain inconclusive [40, 41].

Adiponectin can also modulate bone metabolism by two opposite mechanisms which partially counteract the actions of leptin [42]. Locally, adiponectin blocks osteoblastic proliferation while it induces bone formation via an inhibitory effect on the sympathetic tone. In human and murine models, caloric restriction drastically increases adiponectinemia and is associated with BMAT accumulation despite scarce extramedullary fat depots [43]. The inhibition of BMAT expansion via Wnt10b overexpression is sufficient to abolish hyperadiponectinemia. Additionally, in a rabbit model of caloric restriction in which BMAT does not display expansion, the circulating level of adiponectin is not increased [43]. Altogether, these observations suggest that circulating adiponectin largely originates from marrow adipocytes, and that BMAT is required for caloric restriction-induced hyperadiponectinemia. In contrast to leptin, adiponectin seems to be a reliable marker of bone integrity. Prior clinical studies described that elevated circulating adiponectin levels are correlated with low bone mineral density (BMD) in both genders [44] and with a higher fracture risk in men [41].

Chemerin, a more recently identified adipokine, is also secreted by marrow adipocytes. Its pro-adipogenic and anti-osteoblastic effects were highlighted in mice by the group of Muruganandan et al. [45]. They established that chemerin promotes adipocyte function and differentiation whereas it negatively regulates osteoblastogenesis.

More recently, Martin et al. demonstrated that human bone marrow adipocytes influence the medullar microenvironment through the secretion of extracellular vesicles containing adipogenic mRNAs (i.e., PPARγ, CEBPα, and CEBPδ mRNA) [46•], which could be incorporated and translated by osteoblasts, supporting the idea that marrow adipocytes may modulate the osteoblastic phenotype [47].

As discussed above (see section “Is BMAT Equivalent to White, Brown, or Beige Adipose Tissue?”), adipocytes of the BMAT also secrete a wide range of cytokines such as IL-1β, IL-6, IL-8, and TNFα, depending on the animal species studied. In addition to the release of cytokines and adipokines, in vitro co-culture experiments have shown that human SSC-derived adipocytes release free fatty acids [48]. Depending on their type, they may differently affect skeletal health (see below).

Lipid Composition of BMAT

The concept that not only the amount but also the nature of the lipids contained in marrow adipocytes may be relevant for skeletal health originates from in vitro studies showing the deleterious impact of saturated fatty acids on the bone forming cells and on theirs progenitors, the SSC [48, 49]. Conversely, unsaturated fatty acids, particularly the monounsaturated ones, are beneficial for osteoblasts and prevent lipotoxicity [50•,51]. Interestingly, fatty acids seem to exert opposing actions on osteoclasts, the bone-resorbing cells; the saturated fatty acid palmitate is not toxic for osteoclasts and enhances osteoclastogenesis [29••, 52] whereas the monounsaturated fatty acid oleate counteracts palmitate-induced osteoclast differentiation [29••]. Thus, it was proposed that depending on the nature of the lipids present in the BMAT, bone marrow adipocytes could either prevent or support bone remodeling and, therefore, could impact BMD and skeletal integrity.

Since the conversion of red hematopoietic bone marrow to yellow fatty bone marrow was firstly described by histomorphometric studies of human iliac crest biopsies [53, 54], the progress of medical imaging now allows to distinguish the two types of bone marrow non-invasively. Magnetic resonance imaging (MRI) is currently the method of choice to carry out these studies and recent biochemical data obtained by magnetic resonance (MR)-spectroscopy support the above hypothesis [55, 56]. The non-invasive adiposity assessment by MR-spectroscopy is based on the presence or absence of single or double hydrogen bonds in a bone marrow volume of interest and allows quantification of the saturated and unsaturated fractions of bone marrow fat [57]. Using that method, Patsch et al. showed for the first time that BMAT of type 2 diabetes patients with prevalent fragility fractures is depleted in unsaturated fatty acids and enriched in saturated fatty acids, when compared to control subjects [26]. However, no differences were detected in the total marrow fat content between the two groups. Type 2 diabetes patients are characterized by the paradoxical combination of a normal or increased BMD and a rise in fracture risk. Indeed, bone strength is not only determined by BMD but also depends strongly on bone microarchitecture [58]. Along the same lines, the recent pilot study of Bredella et al. demonstrated that, when compared to non-diabetic obese controls, the femoral neck of type 2 diabetes subjects displaying morbid obesity is characterized by a higher BMD although its BMAT content is elevated and impoverished in unsaturated lipids [59•]. Bredella’s research group also used MR-spectroscopy to investigate the BMAT composition in patients suffering from anorexia nervosa, a metabolic disease characterized by an extremely low BMI combined with an excessive BMAT development and a poor bone quality [60]. They demonstrated that the saturation level in the fatty acids BMAT was inversely correlated with BMD in anorexic subjects [61]. However, despite its higher total lipid content, the BMAT of the anorexic subjects displays a similar unsaturation index compared to normal-weight control.

Very recently, data obtained by high-resolution magic angle spinning (HRMAS) MR-spectroscopy of ex vivo punctures of the iliac crest, revealed that the BMAT of osteoporotic and osteopenic patients is characterized by low levels of unsaturated and high levels of saturated fatty acids [25•], which is in accordance with the in vivo MR-spectroscopy study conducted by Yeung et al. [24].

In light of these findings, some authors suggested that lipid composition, rather than lipid quantity, of the BMAT is correlated with bone integrity and may serve as a biomarker for bone quality [55].

To precisely characterize the nature of the lipids composing the interstitial compartment surrounding bone marrow cells, the composition of the bone marrow interstitial fluid has gained considerable interest. For this purpose, bone marrow samples obtained from the iliac crest are nowadays centrifuged and the supernatant, referred to as the bone marrow supernatant fluid (BMSF), is collected and analyzed. It is anticipated that the lipid concentration and composition of the BMSF will reflect better the physiological lipid profile of the marrow microenvironment than values found in blood, and consequently, will be more representative for the activity of marrow adipocytes. Miranda et al. showed that the fatty acid composition of the BMSF differed from that of the circulation, displaying an enrichment in saturated fatty acids and a reduction in unsaturated fatty acids [62••]. They also observed modifications of the fatty acid composition of the BMSF obtained from osteoporotic women suffering from hip fracture. The content of saturated fatty acids decreased whereas the unsaturated ones were on the rise, suggesting that the fatty acid profile of the BMSF is dynamic and directly related to the local release of marrow adipocytes-derived lipids. The authors proposed that following a hip fracture, the metabolism of BMAT adipocytes is modified to provide energy to osteoblasts for injury repair, or to suppress excessive inflammation [56].

Also characterized by excessive BMAT accumulation, osteonecrosis of the femoral head (ONFH) is a painful disorder attributed to necrosis of the osteomedullary elements of the hip [63]. We analyzed the BMSF isolated from those patients and we revealed major modifications of their bone marrow microenvironment compared to healthy volunteers [64••]. In osteonecrotic patients, the amounts of saturated, monounsaturated, and polyunsaturated fatty acids of the BMSF are severely increased; the concentration of the saturated fatty acid palmitate rises by 50%, whereas the concentrations of the monounsaturated fatty acid oleate and of the polyunsaturated fatty acid linoleate tripled compared to BMSF of healthy volunteers. Interestingly, prior studies correlated an increased food intake or blood value of linoleate with a higher fracture risk in elderly and osteoporotic subjects [65, 66]. Of note, the study of Gillet et al. also highlighted the presence of cis-vaccenic acid in the BMSF of osteonecrotic patients whereas this fatty acid is absent in the BMSF of control subjects [64••]. An epidemiological study performed in Inuit women revealed that bone strength is negatively correlated with the content of saturated fatty acids and cis-vaccenic acid in the erythrocyte membrane phospholipids, and conversely, bone strength is positively correlated with their oleic acid content [67]. Finally, the modifications of the lipid profile observed in the BMSF of patients suffering from ONFH were associated with dysfunctions of SSC, suggesting that marrow adipocyte enlargement plays a role in the pathogenesis of the disease [64••].

Despite the growing interest in the physiological and pathological roles of BMAT, analyses of the BMSF are still sparse and, up to now, no study has been published related to the BMSF assessment of obese or type 2 diabetes subjects.

Conclusions

Through the growing interest of investigators, the secretory and lipidomic profiles of the BMAT have become partially established during the last decade. The use of animal models has allowed for great progress in understanding the multiple functions of bone marrow adipocytes. However, in the context of the study of bone-fat interactions, one should keep in mind that a large diversity of bone responses may be observed depending on the animal models used. Data should thus be interpreted cautiously [68] and, as notably highlighted by Lecka-Czernik et al., genetic background, age, gender, animal species, and duration of the study are essential criteria to consider [69••]. As alluded to before, in mice, two sub-populations of adipocytes presenting distinct bone distribution, function and lipidomic profiles may constitute the BMAT.

In humans, advanced imaging techniques and BMSF assessment allow for indirect evaluation of the adipose lipid content. Current evidence reveals a dynamic lipid composition of BMAT, varying according to physiological and pathological situations. Having highlighted the impact of excessive BMAT enlargement on skeletal health, the research field is now focusing on the characterization of the qualitative aspects of marrow adiposity.

However, several questions are still unresolved. What is the physiological lipid composition of the BMAT? In which circumstances and upon which modifications of its composition does BMAT become pathological? Why is the increase of BMAT beneficial for bone formation during childhood? Why is the presence of excessive BMAT deleterious for skeletal integrity in adults?

In that respect, a better knowledge of BMAT and all its aspects will improve the understanding of the pathophysiological mechanisms of bone diseases characterized by an excessive accumulation of BMAT, such as osteoporosis and ONFH, and therefore, allow the development of new therapeutic approaches.