Bone Marrow Adiposity: Basic and Clinical Implications

Abstract

The presence of adipocytes in mammalian bone marrow (BM) has been recognized histologically for decades, yet, until recently, these cells have received little attention from the research community. Advancements in mouse transgenics and imaging methods, particularly in the last 10 years, have permitted more detailed examinations of marrow adipocytes than ever before and yielded data that show these cells are critical regulators of the BM microenvironment and whole-body metabolism. Indeed, marrow adipocytes are anatomically and functionally separate from brown, beige, and classic white adipocytes. Thus, areas of BM space populated by adipocytes can be considered distinct fat depots and are collectively referred to as marrow adipose tissue (MAT) in this review. In the proceeding text, we focus on the developmental origin and physiologic functions of MAT. We also discuss the signals that cause the accumulation and loss of marrow adipocytes and the ability of these cells to regulate other cell lineages in the BM. Last, we consider roles for MAT in human physiology and disease.

Essential Points

• Marrow adipocytes increase with age and other physiologic cues, representing a major population of bone marrow cells, especially in adult humans

• Bone marrow is unique in that it is the only tissue where adipocytes and bone cells are in close juxtaposition

• Although marrow adipocytes arise from a mesenchymal stem cell, there appears to be a small number of downstream progenitor cells with the capacity to differentiate into mature adipocytes

• Although marrow adipocytes possess some characteristics of white adipocytes, they appear to be a distinct (i.e., fourth) population of fat cells, different from white, brown, and beige adipocytes; thus, bone marrow is a previously unrecognized fat depot

• Marrow adipose tissue is an important endocrine organ that can regulate systemic metabolism

• Marrow adipocytes can have both positive and negative effects on bone density and hematopoiesis, depending on the specific model, by mechanisms that are only partially understood

Mammalian bone marrow (BM) is a heterogeneous tissue, located in the medullary canal of the tibia, femur, and humerus, as well as in flat bones such as the sternum and iliac crest. Cell types that constitute BM are thought to be predominantly mesodermal in origin (1, 2). These include myeloid, lymphoid, and erythroid cells, which make up the hematopoietic lineage, as well as adipocytes, osteoblasts, chondrocytes, and marrow stromal cells (stroma). Nerve fibers and vasculature that invade the marrow space compose the remainder of BM cells (1, 2).

Marrow adipose tissue (MAT) has been recognized histologically for more than a century (3). In humans, marrow adipocytes rapidly increase with age and become a prominent component of BM. However, their ability to regulate other cells, either by direct interaction or through the secretion of soluble factors (adipokines), has been largely unexplored. Because MAT resides in bone, it is much more difficult to study than the adipose tissue outside bone. This difficulty contributes to the paucity of data and has prompted the development of new tools to study MAT.

Because of this new effort, it is now recognized that, in addition to age, a variety of conditions [e.g., high-fat diet feeding, peroxisome proliferator–activated receptor (PPAR)γ agonist treatment, and irradiation] can be used to induce BM adipogenesis in mice. These methods, coupled with new transgenic reporter mouse strains and the ability to quantify MAT in vivo, have made MAT amenable to experimental analyses that historically have been exceedingly labor intensive or simply not possible. Owing to these advances, the functional distinctions between marrow adipocytes and their white, brown, and beige counterparts are beginning to be understood. It is also clear that MAT is an important endocrine organ that regulates systemic metabolism (4). Indeed, excessive MAT is associated with several conditions marked by metabolic derangement, including obesity, diabetes and, paradoxically, anorexia nervosa (5). Moreover, it is known that in many model systems, increased MAT is associated with decreased bone mass and osteoporosis (5–7). A role for the marrow adipocyte lineage in hematopoiesis is emerging as well and may have implications for understanding age-related deficits in bone health and fracture repair (5, 8).

Although largely intractable for many decades, MAT is finally giving up some of its long-held secrets. In this review, we critically evaluate our current understanding of MAT biology, as well as discuss areas of speculation where future work is required. Our working definition of MAT is detailed in the following section.

Marrow adipose tissue

Marrow adipocytes arise from mesenchymal lineage cells within the BM. Mature marrow adipocytes are unilocular and can be found, for the most part, in bones that contain BM. In long bones, marrow adipocytes can be seen just below the growth plate and can extend into the metaphysis and diaphysis. MAT in mice and humans increases with age, and in mice, cell number can be increased by exposure to a group of “stressors” (e.g., x-irradiation, thiazolidinediones, high-fat diet). Increased MAT is also associated with a number of pathophysiologic conditions. Marrow adipocytes secrete adipokines that can function locally, affecting other BM cells, and at distant sites functioning in an endocrine manner (e.g., adiponectin). Marrow adipocytes express numerous characteristics that distinguish them from white, brown, and beige adipocytes.

White, Brown, Beige, and Marrow Adipose Tissues

White adipose tissue

White adipose tissue (WAT) is indispensable for maintaining proper metabolic homeostasis. It is the major hub for lipid stockpiling and distribution, and functions as an endocrine organ to regulate systemic insulin sensitivity (9–13), as well as feeding behavior (14, 15) (Fig. 1). In mice, white adipocytes do not fill with lipid until after birth and are derived predominantly from progenitors that originate in the somites or lateral plate mesoderm (16). White adipocytes contain a single unilocular lipid droplet (17, 18). In humans, the final morphological and functional characteristics of white adipocytes are similar to rodents. However, WAT emerges in the second trimester of gestation rather than postnatally (19). In the adult body WAT is anatomically divided into various spatially distinct depots that are broadly classified as either subcutaneous (just beneath the skin) or visceral (within the abdominal cavity). In women, WAT mass is preferentially distributed subcutaneously around the hips and thighs, whereas men have comparatively reduced subcutaneous WAT and a proportionally similar amount of visceral WAT (VWAT) (20).

Figure 1.

Phenotypes of white, brown, beige, and marrow adipocytes. A plus sign indicates that the gene/protein is expressed; a minus sign indicates that the gene/protein is not expressed; a plus/minus sign indicates a mixed population of cells; a question mark indicates that expression is unknown; an up arrow indicates increased numbers of cells; and a down arrow indicates decreased numbers of cells. Note that the VWAT includes perigonadal, retroperitoneal, and mesenteric WAT. asWAT, anterior subcutaneous WAT; psWAT, posterior subcutaneous WAT; vWAT, visceral WAT.

Unsurprisingly, sex hormones play a significant role in body fat distribution; indeed, visceral adiposity and subcutaneous adiposity are sometimes referred to as android and gynoid, respectively. Consistent with this terminology, elevated free testosterone is associated with an increased waist-to-hip ratio (a metric for visceral adiposity) in women with obesity (21), and postmenopausal women have an increased proportion of VWAT relative to premenopausal women as measured by dual-energy x-ray absorptiometry (22). Moreover, ovariectomized mice fed a high-fat diet display a male-like pattern of adipogenesis, and male mice treated with estradiol have increased adipogenesis in subcutaneous WAT relative to male animals without estradiol treatment (23). Interestingly, WAT distribution in obesity is a predictor of the metabolic syndrome independent of sex (24, 25), with excessive visceral adiposity more strongly associated with cardiometabolic disease risk than excessive subcutaneous adiposity (26, 27). Thus, WAT distribution is markedly sex-specific, and distinct functional and pathophysiological roles exist for WAT depots.

Brown and beige adipose tissue

Unlike WAT, in mice, brown adipose tissue (BAT) is formed prior to birth and is mainly located in one well-defined depot between the shoulder blades (intrascapular). Brown adipocytes, in contrast to white adipocytes, are multilocular, having multiple small lipid droplets within the cytoplasm. Brown adipocytes have large numbers of mitochondria, facilitating one of their major functions, which is to actively metabolize fatty acids to generate heat (28) (Fig. 1). In humans, BAT is present embryonically and was thought to be lost after childhood and absent in adults. However, within the last few years, BAT has been confirmed to exist in adult humans through modern imaging methods, being located in the supraclavicular region and in the neck, and it can be increased with cold exposure (29, 30). Interscapular BAT is derived from Myf5⁺ progenitor cells, and brown adipogenesis requires the activity of Ebf2, Prdm16, and BMP7 (18, 30, 31).

Beige adipocytes are similar to brown adipocytes in that they are multilocular, are rich in mitochondria, and metabolize fatty acids (Fig. 1). Unlike BAT, beige adipocyte progenitors are found interspersed within WAT depots, with more in the inguinal WAT (IWAT) depot than in the visceral depots (32, 33). Although beige progenitor cells are Myf5⁻ in inguinal fat, similar to brown adipocytes, they can be induced to differentiate in vivo by cold exposure or treatment with β-adrenergic agents and feeding mice a methionine-restricted diet (33, 34). Because beige adipocyte precursors can be stimulated to differentiate in vivo to mature fatty acid–metabolizing cells, much interest has been focused on this process as a potential therapeutic target to increase energy expenditure and combat obesity. However, beigeing can be detrimental in hypermetabolic states, such as cancer and burns, leading to cachexia [(35–38); for review see (39)].

Marrow adipose tissue

Anatomic distribution and bone and inductive marrow adipogenesis

In the young, BM is rich with hematopoietic, osteogenic, and erythroid cells and is red in appearance. One of the most prominent characteristics of MAT is that it increases with age in humans and mice, resulting in the development of yellow fatty marrow (40, 41). In adult C57BL/6 (B6) mice, a small number of marrow adipocytes can be seen histologically distributed throughout the medullary canal between the tibia–fibular junction and the growth plate. These cells cannot be detected by osmium staining (see “Measurement of mouse MAT in vivo” below). In contrast, a small number of marrow adipocytes can be imaged by osmium staining just below the growth plate in 8-week-old mice (42). MAT can also be detected above the growth plate, in the secondary center of ossification at 4 weeks of age.

Although MAT does not have the same spatially distinct depots as WAT, marrow adipocytes appear in the distal tibia developmentally well in advance of large numbers of adipocytes in the proximal tibia (42). MAT can be seen in caudal (tail) vertebrae, but not in thoracic or lumbar vertebrae (6). The reason for this uneven distribution in vertebrae is unclear but may relate to temperature differences in the tissue (43). The increase in marrow adipocytes with age correlates with age-related bone loss. This correlation is often cited to indicate that marrow adipocytes are a negative regulator of bone mass. However, this correlation does not uniformly hold. B6 mice have one of the lowest trabecular and cortical bone densities of any mouse strain and very low numbers of marrow adipocytes in their long bones. In contrast, C3H/HeJ (C3H) mice (a distinct strain) have one of the highest bone densities and higher numbers of marrow adipocytes than do B6 mice (44). Interestingly, marrow adipocytes appear in the distal tibia (distal to the tibia–fibular junction extending to the end of the tibia) as early as 4 weeks of age in both B6 and C3H mice and fill the marrow space (42). Whereas few marrow adipocytes can be detected in the proximal tibia of B6 mice, C3H mice have measurable marrow adipocytes in their proximal tibia. The mechanism by which these marrow adipocytes arise is unknown, and a comprehensive analysis of marrow adipocyte distribution by mouse strain remains to be done. Nonetheless, these observations suggest that differentiation of marrow adipocytes is regulated, at least in part, genetically. Therefore, an analysis of MAT using Collaborative Cross mice or Diversity Outbred mice, which are the most genetically diverse mouse resources available, would provide valuable data about the genetics of MAT (45, 46).

Independent of mouse strain, however, marrow adipocytes can be induced in the proximal tibia and femur by a variety of methods. For example, placing mice on 6 weeks of a rosiglitazone (PPARγ agonist)-containing diet potently induces marrow adipogenesis in these bones. Marrow adipocytes extend from below the growth plate, through the metaphysis, and into the diaphysis as a result of this treatment. Increased marrow adipocytes can also be observed above the growth plate in the secondary center of ossification. Eight weeks of a methionine-restricted diet or whole-body x-irradiation are also potent inducers of MAT (8, 47). In contrast to methionine restriction inducing MAT, it causes a marked reduction in IWAT and VWAT mass due to an impaired lipid cycle. Interestingly, although a methionine-restricted diet induces beige adipogenesis with increased mitochondrial biogenesis and amplified expression of uncoupling protein 1 as well, methionine restriction–induced marrow adipocytes do not express uncoupling protein 1, indicating that these cells are not thermogenic in this context (48–51). However, it has been reported recently that treatment of B6 or C3H mice with CL316,243, a β-adrenergic receptor agonist, caused the formation of a small number of multilocular cells in the distal tibia. Nonetheless, these cells did not resemble classic beige adipocytes (52). Further work is required to identify whether marrow adipocytes can take on thermogenic properties in vivo.

Measurement of mouse MAT in vivo

Because of the inability to accurately quantitate marrow adiposity in mice, we developed a method to measure MAT in vivo (53). In brief, mouse tibiae or femurs are isolated and fixed in formalin overnight. The bones are gently decalcified in EDTA for 20 days, washed, and then stained with osmium tetroxide. Osmium stains lipid in the adipocytes, and because it is a heavy metal, it can be imaged by micro-CT (Fig. 2). Histologic sectioning of osmium-stained bones shows that >95% of the stain is present in marrow adipocytes. Moreover, the amount and position of the marrow adipocytes measured by this method correspond directly with the histology (Fig. 2). This method provides a reproducible, volumetric (three-dimensional, adipocyte volume/total volume) measurement of MAT, including in the distal tibia. Measurements can be obtained above the growth plate in the secondary center of ossification, but because of the variability of structure at this site, comparisons between bones can be difficult.

Figure 2.

(Left) The tibia from B6 or C3H mice. The tibia was isolated, fixed in 10% formalin, and decalcified for 20 d in 4% EDTA. The bones were washed and stained with osmium tetroxide for 48 h, washed, and MAT was imaged by micro-CT. The tibia from B6 mice had low MAT in the proximal tibia with significant MAT visible in the distal tibia and above the growth plate. In contrast, the tibia from C3H mice had extensive MAT throughout the medullary canal. Even the fibula can be imaged in the C3H mice because of MAT filling the marrow space. (Right) A 5-μm thin histologic section of osmium-stained distal tibia (below the tibia–fibular junction) is shown. Marrow adipocytes fill the BM space from endosteum to endosteum.

The BM Adipocyte Lineage

The Cre/Lox system

Modern lineage-tracing experiments invariably depend on the Cre/Lox system (54). It therefore merits a brief description and a word of caution relating to its effective use in defining bona fide cell lineage relationships. In the conventional Cre/Lox system, Cre recombinase is expressed under the control of a tissue-specific promoter to permanently activate a reporter gene that functions to mark the original Cre-expressing cell population and all daughter cells thereof. To accurately interpret any Cre/Lox-based lineage tracing experiment, one must have a thorough understanding of the Cre’s spatiotemporal expression domain. Otherwise, erroneous conclusions regarding the tissue of origin may be put forward. Equally important is the nature of the reporter gene used. For example, adipocytes possess little cytoplasm relative to other cell types, and therefore cytoplasmic reporters such as LacZ are suboptimal in this context. Instead, membrane-targeted reporters such as membrane-targeted dTomato (mT)/membrane-targeted eGFP (mG) are preferred (55, 56). For extensive discussions on this topic, see Jeffery et al. (57) and Sanchez-Gurmaches et al. (58).

Several groups have performed lineage tracing in vivo using the fluorescent mT/mG reporter mouse in concert with various mouse models driving cre-recombinase from lineage-specific promoters. The expression of cre-recombinase induces the permanent excision of the upstream cassette encoding an mT reporter protein to allow expression of a downstream cassette encoding an mG reporter protein. In this system, cells expressing cre-recombinase “flip” from being dTomato⁺ to being eGFP⁺ (Fig. 3). Importantly, expression of eGFP reflects the expression of the gene of interest during the life of the cell and will be maintained in daughter cells (Fig. 3). Therefore, this system allows us to infer progenitor cell lineage (59). In our experiments, BM adipogenesis was induced by either whole-body x-irradiation or feeding mice a rosiglitazone-containing diet. BM plugs were collected from freshly isolated femurs by removing the femoral head, and sliding a 20-gauge needle through the marrow canal from the proximal to the distal end of the femur, and the cellular fluorescence was analyzed by confocal microscopy within hours of isolating the BM. Thus marrow adipogenesis can be analyzed in vivo.

Figure 3.

(Top) Schematic of the mT/mG double reporter system. (Bottom) Adiponectin-cre:mT/mG double reporter mice were fed a rosiglitazone-containing diet for 6 wk. The femur was isolated and the femoral head was removed. A 20-gauge needle was slipped down the medullary canal and punched out through the distal epiphysis. The needle was attached to a 5-cc syringe, and the BM was extruded onto a microscope slide. The BM was stained with LipidTox (a lipophilic dye), washed, immersed in Fluoromount, and coverslipped, and fluorescence was viewed by confocal microscopy. LipidTox (left) was used to identify marrow adipocytes. All of the marrow adipocytes were traced with eGFP, (center left), whereas none was with dTomato⁺ (center right). These data indicate that all marrow adipocytes and their progenitors express adiponectin.

BM adipocyte precursor identity

It is generally thought that adipocytes, regardless of type, arise from precursors within their specific depot. In an effort to identify the adult marrow adipocyte precursor, B6 or mT/mG mice were lethally irradiated and reconstituted with either mT/mG or B6 BM respectively. Marrow adipocytes in B6 mice reconstituted with mT/mG BM showed no fluorescence, whereas marrow adipocytes from irradiated mT/mG mice reconstituted with B6 BM were uniformly dTomato⁺. Marrow adipogenesis could be observed as early as 5 days after irradiation. Importantly, in a separate set of experiments, mice were sub-lethally irradiated and were not reconstituted with BM cells. Tracing of marrow adipocytes was similar to the reconstituted mice. These data suggest that marrow adipocytes arise from a radio-resistant progenitor cell population that is present in adult BM (47).

Hematopoietic stem cells (HSCs) reside in BM, and it is possible that marrow adipocytes arise from these cells, although HSCs are uniquely radiosensitive due to their rapid proliferation. Regardless, it has been reported that WAT can develop from HSCs (60). However, lineage tracing using Vav1-cre:mT/mG mice, which traces more than 95% of HSCs and their progeny, failed to mark white adipocytes residing in inguinal subcutaneous WAT (59). Marrow adipocytes induced by a rosiglitazone diet were also not traced using this Vav1-Cre tracing paradigm, indicating that marrow adipocytes do not arise from hematopoietic cells (47).

In mice, brown adipocytes are traced in the Myf5-cre:mT/mG system, indicating that brown adipocytes arise from Myf5⁺ progenitors. In contrast, marrow adipocytes are untraced in the same Myf5-cre:mT/mG mice (47). In addition to this apparent developmental difference, mature marrow adipocytes are unilocular whereas brown adipocytes are multilocular. In our experience of examining numerous histologic sections of mouse long bones (tibias, femurs, and humeri) from different mouse models, in which MAT was induced or not induced, we have never observed a multilocular mature marrow adipocyte. Taken together, these data suggest that marrow adipocytes are developmentally and morphologically distinct from brown and beige adipocytes.

Compared with WAT and BAT, less is known about the embryonic origin of MAT. Regardless, it would be surprising if the marrow adipocyte lineages did not follow known developmental trajectories for other BM cell types, at least until the final stages of differentiation. As mentioned previously, marrow adipocytes are closely related to osteoblasts ontogenetically. To explore this relationship, we performed lineage tracing of marrow adipocytes using two mouse reporter stains, which label mesenchymal/osteogenic cells. Paired related homeobox 1 is a protein that in humans is encoded by the PRRX1 (Prx1) gene. Prx1 expression is restricted to limb bud mesenchyme during development, and Prx1 and Prx2 are expressed in mesenchymal tissue in adult mice (61). In Prx1-cre:mT/mG mice that were either irradiated or fed a rosiglitazone diet, all marrow adipocytes were eGFP⁺, indicating that Prx1 traced all of the marrow adipocytes and that these cells arise from lateral plate mesoderm (similar to most other BM cells) (51). Prx1 has a variable expression pattern in anterior subcutaneous WAT, whereas 96% of adipocytes in posterior WAT, including beige adipocytes, are traced by Prx1. In contrast, Prx1 does not trace most BAT or any VWAT (retroperitoneal WAT, mesenteric WAT, or perigonadal WAT) (62). These data indicate that Prx1 is an early marker of marrow adipocytes. Consistent with the expression of Prx1 in limb bud mesenchyme, cells on bone surfaces, osteocytes embedded deep in bone, and articular cartilage were also traced with Prx1 (data not shown). Because Prx1 expression is more restricted, in whole-body adipose tissue, than many other Cre drivers (e.g., adiponectin-cre), it is useful in adipocyte lineage tracing and gene deletion. Additionally, it has been shown that Osx⁺ cells give rise to the osteolineage in developing (embryonic day 13.5) fetal BM (63). Osx can also be detected in osteocalcin (osteoblast marker)-negative BM stromal cells, suggesting that Osx marks nonosteolineage cells as well. In adult BM Osx continues to mark osteolineage cells, but the Osx population of stromal cells is lost. A separate marker, Nestin, labels mesenchymal progenitor cells that are present in developing mouse BM. Nestin⁺ cells appear to overlap with CD45⁻CD90⁻CD51⁺CD105⁺ and Prx1^lin+PDGFRα⁺Sca1⁺ cell populations (64, 65). Notably, most (85%) of the Osx⁺ BM cells are also nestin⁺ and highly enriched in colony forming units–fibroblasts. This population of cells is able to differentiate to yield osteoblasts and chondrocytes. Importantly, irradiation-induced adipocytes are Osx⁺ (63).

In a separate set of experiments, cells expressing the leptin receptor (LepR⁺) accounted for most colony forming units–fibroblasts (94%). A LepR-Cre traced most marrow adipocytes and osteoblasts in adult BM but not in developing BM (8). Interestingly, the LepR^lin+ BM population highly overlaps with the Nestin-marked BM population (66). These data suggest that Nestin⁺/LepR⁺ progenitor cells can give rise to marrow adipocytes in adult BM. In addition to these reports, it has been shown that Gremlin1-expressing BM cells, which are capable of self-renewal, can also give rise to osteoblasts, chondrocytes, and marrow stromal cells, but not adipocytes (67). This suggests Gremlin1⁺ precursors represent a more lineage-restricted cell population than Nestin⁺/LepR⁺ precursors and may be useful for distinguishing adipogenesis-competent cells from incompetent cells in the BM.

Intermittent PTH administration stimulates new bone formation. Although PTH increases the activity of mature osteoblasts, it may also recruit osteoblast–marrow adipocyte progenitor cells into the osteoblast lineage. To evaluate the role of PTH, or its related protein PTHrP, on mesenchymal stem cell fate, Prx1-cre:PTH1R^fl/fl mice were generated to delete the PTH/PTHrP receptor from Prx1⁺ mesenchymal progenitor cells (68). These mice had a bone phenotype characterized by increased bone resorption, low bone mass, severely deformed tibias, increased numbers of tartrate-resistant acid phosphatase–positive (TRAP⁺ osteoclasts) cells, and increased MAT. RT-PCR analysis showed that the marrow adipocytes from the Cre⁺ cells but not the Cre⁻ cells expressed RANKL and are likely responsible for the increased numbers of osteoclasts and bone loss. Some of the RANKL⁺ cells were also preadipocyte factor-1 (Pref-1)⁺. Additionally, PTH treatment reduced marrow adipocyte numbers in humans by 27% after 18 months (68). These data show that the anabolic effect of PTH on bone can be attributed, at least in part, to regulation of osteoblast/adipocyte lineage allocation. In a separate set of experiments, mice were placed on a calorie-restricted or control diet and treated with PTH or vehicle. A calorie-restricted diet induces bone loss and increases MAT in mice and humans (69, 70). In contrast, PTH treatment significantly increased bone mass with increased numbers of osteoblasts in calorie-restricted and control groups and reduced marrow adipocyte number (71). These data are consistent with the experiments deleting the PTH/PTHrP receptor, in that PTH signaling can suppress the development of marrow adipocytes through a mechanism that likely alters lineage allocation away from adipogenesis and toward osteoblastogensis.

PPARγ coactivator 1α (PGC-1α) regulates cell fate allocation and is expressed preferentially in BAT compared with WAT. PGC-1α is highly expressed in mesenchymal stem cells, and its expression decreases with age in mice and humans (72, 73). Pgc1α global knockout mice have decreased femoral bone mineral density (BMD) with an associated loss of osteoblast number and bone formation rate. The bone loss was associated with a marked increase in marrow adipogenesis (74). To determine the role of Pgc-1α on the mesenchymal lineage, Prx1-cre:Pgc1α^fl/fl mice were used to delete PGC-1α from cells expressing Prx1. In ovariectomized Prx1-cre:Pgc-1α^fl/fl mice, BMD and bone volume were decreased relative to sham-operated controls. In the proximal tibia of the Prx1:Pgc-1α^fl/fl mice, marrow adipocytes extend from just below the growth plate through the metaphysis into the diaphysis as measured by osmium staining. These data suggest that the loss of PGC-1α in mesenchymal precursor cells results in increased bone loss and marrow adipogenesis during ovariectomy-induced osteoporosis. Interestingly, LepR-cre:Pgc-1α^fl/fl mice had a bone and marrow fat phenotype similar to the Prx1-cre:Pgc-1α^fl/fl mice. This is important because LepR is expressed on the mesenchymal progenitor cells that give rise to marrow adipocytes and osteoblasts in adult mice (8). Thus, expression of Pgc1α functions to regulate the balance between marrow adipocytes and osteoblast lineage allocation (75).

With menopause there is a decrease in bone mass and a concomitant increase in subcutaneous, visceral, and BM adiposity. Treating these conditions, especially together, has been difficult. During this time FSH levels rise, whereas estrogen levels are relatively stable (76). A polyclonal antibody directed at the β-subunit of Fsh was used in vivo to determine the role of FSH in regulating obesity and bone mass. Injection of the anti-FSH antibody in wild-type male and female mice that were on a high-fat diet showed a significant reduction in visceral and subcutaneous adiposity with an increase in lean mass compared with an isotype (IgG) control antibody (77). Antibody-treated mice also had a significant increase in BMD. Using FSH receptor–deficient mice (Fshr^+/−), the lack of actions of the antibody were observed, confirming the effect of the antibody in Fshr-replete mice. Ovariectomized and sham-operated control mice treated with antibody also showed a reduction in fat mass and an increase in lean mass. Importantly, ovariectomized mice treated with antibody showed a significant decrease in MAT (77). Interestingly, mice placed on a high-fat diet and treated with antibody showed increased numbers of multilocular, Ucp1⁺ staining adipocytes in IWAT, consistent with increased beige adipogenesis. These data provide evidence that interfering with FSH signaling in vivo can reduce whole-body WAT, with an increase in metabolically active beige adipocytes, while increasing bone mass. Thus, suppressing FSH may provide a new approach to treat a variety of pathologic conditions associated with high fat.

Distinguishing cells according to cell surface marker profile has also proven to be an effective strategy for identifying adipogenesis-competent BM-derived cells. CD45⁻CD31⁻PdgfRα⁺Sca1⁺ BM cells have been shown to be highly adipogenic but have limited osteochondrogenic potential in vitro. In contrast, CD45⁻CD31⁻PdgfRα⁺Sca1⁻ cells do not differentiate into adipocytes in vitro but have a strong osteochondrogenic potential. The CD45⁻CD31⁻PdgfRα⁺Sca1⁺ population can be further separated based on CD24 expression. The CD45⁻CD31⁻PdgfRα⁺Sca1⁺CD24⁺ population is able to differentiate into osteoblasts, chondrocytes, and adipocytes, whereas the CD45⁻CD31⁻PdgfRα⁺Sca1⁺CD24⁻ population gives rise only to adipocytes (7).

Whereas MAT can be induced by external stressors, it is regulated by the expression of cell-intrinsic molecules. Thy-1 (CD90) is a glycosylphosphatidyl-anchored cell surface protein belonging to the immunoglobulin superfamily and is expressed on mesenchymal stem cells (78, 79). Thy-1 regulates a variety of cell functions, including the balance between proliferation and apoptosis (80). Mesenchymal stem cells from Thy-1–deficient (knockout) mice cultured in vitro had decreased osteoblast differentiation with a concomitant increase in marrow adipogenic differentiation compared with wild-type control mice (81). Thy-1^−/− mice exhibited decreased bone volume, bone formation rate, and increased cortical porosity, resulting in biomechanically weaker bones. Interestingly, subcutaneous WAT (SWAT), GWAT, and MAT were increased in the Thy-1^−/− mice compared with controls. These data suggest that similar to expression of Pgc1α, Thy1 expression can regulate lineage allocation between marrow adipocytes and osteoblasts.

When marrow adipogenesis is stimulated in long bones (e.g., by rosiglitazone feeding), marrow adipocytes form just below the growth plate through the metaphysis and into the diaphysis. Whether marrow adipocyte precursors are concentrated in the primary spongiosa or distributed throughout the BM is unclear. It has been reported that marrow adipocyte precursors are not located on the endosteum but rather in the BM close to the bone surface and perivascularly, being evenly distributed over the metaphysis and diaphysis (7). However, as previously discussed, the identity of these cells remains vaguely defined. In contrast, osteogenic progenitors were found on the endosteum with more in the metaphysis than in the diaphysis (7). This raises the interesting possibility that distinct marrow adipocyte precursor niches exist in bone. In keeping with this idea, mature marrow adipocytes can directly interact by cell-to-cell contact with other BM cells and with the sympathetic nervous system, which may provide another pathway to regulate marrow adipocytes (82).

Taken together, it is apparent that marrow adipocytes are derived from BM-resident mesenchymal progenitor cells. However, a precise lineage hierarchy within BM for marrow adipocytes and the molecular mechanisms by which they differentiate remain unclear. Further investigation is required to determine whether marrow adipocytes arise from a single population or from a small number of different progenitors in the BM. Additionally, the lineage relationships between marrow adipocytes and white, brown, and beige adipocytes has yet to be elucidated.

BM adipocyte metabolism

The intracellular metabolism of marrow adipocytes is largely unknown due to technical challenges associated with their isolation. Nonetheless, the defining feature of these cells is their ability to store lipids as large, unilocular lipid droplets, sometimes referred to as lipid bodies or adiposomes. The presence of these lipid droplets suggests that marrow adipocytes experience similar cellular processes responsible for the storage, mobilization, and metabolism of fatty acids as in other metabolically active adipose depots. Given this defining organelle, any gaps in knowledge relative to marrow adipocyte metabolism can be, to a certain extent, extrapolated from other classic fat depots. This section therefore aims to review these processes and how they likely contribute to BM adipocyte metabolism at a cellular level.

Fatty acid uptake and lipid storage

MAT lipid droplets are composed of a core of neutral lipids, including triacylglycerol (TAG) and cholesteryl esters, surrounded by a phospholipid monolayer that includes embedded proteins (83–85). As such, marrow adipocytes must first accumulate fatty acid substrates that can then be esterified and stored. Similar to other adipose depots, these fatty acids are thought to be acquired either through uptake from the circulation or through de novo synthesis from glucose substrates.

Upon entry into the cell, fatty acids destined for storage must be esterified to glycerol or cholesterol via a two-step, ATP-dependent reaction catalyzed by acyl–coenzyme A synthetase (ACS) that also involves ACSL1, ACSL3, and ACSL4 (85). Once activated, these fatty acids can begin to be incorporated into the lipid droplet core. Although the relative amounts of TAGs and cholesteryl esters can vary, TAGs are by far the most predominant lipid found in MAT (83–85). As the neutral lipid core continues to expand, the lipid droplet membrane must be synthesized as well. Aside from the phospholipids, lipid droplet membranes may contain lipid droplet–associated proteins, including the PAT proteins [perilipin A, adipose differentiation-related protein (ADRP), and tail-interacting protein of 47 kDa (TIP47)]. The PAT family of proteins or perilipins consists of PLIN1, PLIN2, PLIN3, PLIN4, and PLIN5 (86). PLIN1 has been identified as a component of marrow adipocytes both through immunohistochemistry and western blot analysis (52). Interestingly, in other adipose depots, it is thought that small lipid droplets are first coated with PLIN2 and PLIN3 (87), and as the droplet enlarges they are replaced by PLIN1 (88); however, other members of the PAT family have yet to be identified on marrow adipocytes. Pertinent to this observation, researchers have often studied MAT by biochemical fractionation owing to their lipid droplet–mediated buoyancy (89). However, it is reasonable to speculate that marrow adipocytes with smaller, denser lipid droplets are not included in this fraction, and therefore are not represented (90). Consequently, this technique is expected to bias data toward large lipid droplets/mature marrow adipocytes, which could confound any interpretation. Therefore, although marrow adipocytes are unique and much remains to be learned about marrow adipocyte lipid droplets, their biogenesis appears to involve common machinery found in classic white adipocytes.

Lipolysis and fatty acid mobilization

Once lipid droplets have accumulated in marrow adipocytes, they are speculated to serve as a source of stored energy or ATP. When cellular energy is in demand, either by the marrow adipocytes themselves or other cells, lipid droplets are presumed to be broken down to free fatty acids via lipolysis. Lipolysis is a generalized term describing the catabolism of lipids via mechanisms involving (i) lipid droplet–associated cytoplasmic lipases, and/or (ii) autophagosome sequestration of neutral lipids and subsequent degradation by the lysosome, termed lipophagy.

The more commonly described lipolysis involves cytoplasmic lipases acting on the lipid droplet in three distinct, sequential steps: (i) adipose triglyceride lipase hydrolyzes TAGs to release a fatty acid, yielding diacylglycerol (DAG) (91); (ii) hormone-sensitive lipase then converts DAG to monoacylglycerol (92), again releasing a fatty acid, while (iii) the last fatty acid is liberated from glycerol via monoglyceride lipase (93). Importantly, note that hormone-sensitive lipase can hydrolyze TAGs as well; however, its activity is markedly higher for DAGs. Until recently, MAT lipolysis has remained underappreciated and rarely discussed. For example, histological analysis of MAT would often shed some light on the size or volume of marrow adipocytes/lipid droplets, but owing to the nature of lipid flux, one could not conclude whether this was due to increased storage of fatty acids or diminished lipolysis. This point is further underscored by the fact that although MAT was recently shown to undergo cytoplasmic lipase-mediated lipolysis, this unique depot appears to be less responsive to β-adrenergic stimulation compared with WAT (52). Therefore, although the process of lipolysis appears to occur in marrow adipocytes similar to other adipose depots, the stimulus and/or responsiveness may be unique to this depot. These data provide an interesting clue about how MAT metabolism is regulated and indicate that continued investigation is required.

“Although the association between MAT and BMD is clear, what is not clear is the mechanism of action.”

Aside from cytoplasmic lipase lipolysis, lipophagy has been shown to be a crucial means by which white adipocytes can mobilize lipids as well. Macroautophagy, referred to hereafter as autophagy, is a multistep cellular process that involves initiation, membrane nucleation, phagophore formation, sequestration, and autophagosome formation, followed by autophagosome–lysosome fusion (94). As such, lipophagy invokes a similar series of steps; however, instead of the indiscriminate engulfment of cytoplasmic organelles, this process selectively targets lipid droplets (95–97). Although the process of lipophagy has been described in white and brown adipocytes, to date, we are unaware of any studies that have investigated lipolysis via lipophagy in MAT; however, this remains an interesting prospect for future studies. Whether cytoplasmic-mediated lipolysis and/or lipophagy are initiated in marrow adipocytes, the end product of free fatty acids and glycerol remains conserved between pathways.

Fatty acid oxidation and ATP generation

Aside from MAT’s presumed ability to mobilize fatty acid stores to other cells (i.e., osteoblasts, osteoclasts, and their precursors), marrow adipocytes may also use the fatty acids for their own ATP generation. Similar to the esterification that takes place during lipid droplet biogenesis, fatty acids would first need to be activated in the cytoplasm via ACS, prior to entry into either the mitochondria or peroxisome for β-oxidation (98, 99). More specifically, MAT in both the distal tibia and proximal tibia and femur contains a large proportion of long-chain fatty acids and, therefore, primarily β-oxidation is thought to occur (83, 100). Acetyl–coenzyme A can now enter into the tricarboxylic acid cycle where it undergoes oxidative phosphorylation, generating ATP. This ATP is now capable of suppling the bone with the energy demands required for cellular function and maintenance.

MAT in Humans

Measurement of MAT in humans

Imaging techniques have allowed for the noninvasive quantification of MAT in humans. The imaging methods used to quantify MAT include magnetic resonance spectroscopy (MRS) (101, 102), T1-weighted MRI (103, 104), a modified Dixon method (also known as fat fraction MRI or the water-fat MRI) (105, 106), as well as CT (107, 108). MRS and MRI are the preferred methods for measurement of MAT in humans given that CT imaging involves radiation exposure. These methods all differ slightly in their methods of quantitating MAT. For example, MRS quantification is limited to a specifically demarcated region and may be more technically challenging as compared with T1-weighted MRI, which is a more standard procedure and therefore may be more feasible (105). T1-weighted MRI, alternatively, uses a threshold method to distinguish MAT and therefore is considered a semiquantitative method, as only pixels reaching a given threshold are quantitated as MAT (105). Despite these differences, these methods have been compared with each other as well as to histologic examination of MAT and, importantly, they have been shown to produce highly concordant results. For example, Shen et al. (105) compared MAT quantification in the L3 vertebra in postmenopausal women using MRS, T1-weighted MRI, and a modified Dixon method and found strong concordance among the three techniques (0.78 to 0.88). Bredella et al. (108) found strong concordance between dual-energy CT and MRS quantification of MAT. Arentsen et al. (109) measured vertebral MAT using the Dixon method as well as dual-energy CT in human cadavers (<24 hours postmortem) and found a high concordance between these imaging methods (r = 0.88). They then compared histologic sections taken from the imaged region of interest and found that adipose volume/tissue volume (AV/TV) (the histologic quantification of MAT) was consistent with the imaging methods (r = 0.77 to 0.80) (109). Therefore, all of these methods provide an accurate, noninvasive means of quantifying MAT.

MAT in Healthy Populations

Development

As in mice, at birth, the composition of BM is predominantly red with hematopoietic/osteogenic cells (110). Thereafter, MAT rapidly begins to accumulate in the human skeleton (110), and this accumulation occurs more rapidly in the distal bones as compared with the proximal bones (111). This accumulation occurs during the time of peak bone acquisition and by the age of 25 years, 70% of marrow in humans contains MAT (111). MAT continues to increase with age in both males and females (102, 103, 112, 113), yet after the age of 55 years, although MAT continues to increase steadily in males, it increases dramatically in females, likely due to the estrogen deficiency state characteristic of menopause (114). In vitro studies demonstrate that 17β-estradiol decreases PPARγ agonist-induced adipocyte differentiation of human mesenchymal stem cells (115). In murine models, estrogen deficiency, for example through ovariectomy, also enhances adipocyte formation in BM, and estrogen supplementation can prevent increases in marrow adiposity in these estrogen-deficient animals (116). Similarly, in women, marrow AV/TV and marrow adipocyte number increase after menopause, and AV/TV decreases with estrogen replacement (117), thereby demonstrating the effects of estrogen deficiency on MAT accumulation in humans. Levels of endogenously produced estradiol have also been associated with MAT in older adult men, with higher levels of estradiol being associated with lower vertebral MAT (118). Another factor that may influence accumulation of MAT during development is exercise. In young, school-aged children, individuals who participated in 20 minutes of moderate physical activity 3 days per week for 10 weeks had a significant decrease in femoral MAT as compared with children in the control group whose femoral MAT did not change (119). In mice, exercise reduces not only VWAT but also the accumulation of MAT induced by rosiglitazone feeding and a high-fat diet (120, 121) [for a review, see (122)]. Therefore, systemic factors likely play an important role in the accumulation or loss of MAT throughout human development and the lifespan.

Relationship with BMD

Although MAT replaces red or hematopoietic/osteogenic marrow in the long bones during peak bone acquisition (40, 41), suggesting a positive association between bone mass and MAT, most studies demonstrate an inverse relationship between MAT and BMD, including in children and young individuals. For example, Shen et al. (123) demonstrated a significant inverse association between pelvic MAT and pelvic BMD in children 5 to 17 years of age. Similarly, Di Iorgi et al. (107) have demonstrated this inverse association between BMD and MAT in individuals between the ages of 15 and 25 years. This inverse association has also been demonstrated in many groups of healthy adults, including white women (18 to 88 years of age) (103) and middle-aged African American and white women and men (38 to 52 years of age) (104). Therefore, the preponderance of data in humans demonstrate that across the age spectrum, in healthy populations, there is a strong inverse association between BMD and MAT. Because MAT has also been associated with measures of decreased bone integrity, including vertebral compression fractures (113), the inverse association between BMD and MAT may have important structural consequences and, importantly, marrow adiposity has been associated with fractures. In postmenopausal women with osteoporosis, vertebral fractures were associated with higher MAT volume and larger adipocytes independent of BMD (124).

The association between MAT and BMD is further strengthened by longitudinal studies demonstrating changes in both BMD and MAT in response to treatments for low bone mass. Treatment of postmenopausal women with teriparatide (recombinant human PTH) for 1 year resulted in decreased vertebral MAT concomitant with increases in lumbar spine BMD (125). Similarly, in postmenopausal women treated with estrogen replacement (see “Development” above), a decrease in AV/TV was observed after 12 months of treatment (117). In contrast, although an in vivo rodent model demonstrated a significant decrease in MAT in ovariectomized rats treated with raloxifene (a selective estrogen receptor modulator used in the treatment of osteoporosis), postmenopausal women treated with raloxifene for 2 years did not demonstrate significant changes in marrow adipocyte volume or number (124, 126). Whether the differences observed with raloxifene (125) vs estrogen replacement (117) are due to the time course of treatment (i.e., perhaps changes would have been observed after 12 months of treatment with raloxifene but not at the 2-year time point) or due to the fact that estrogen replacement has been shown to have greater effects on BMD as compared with raloxifene after 12 months of treatment (127), and therefore may have greater effects on MAT, is not known. Pioglitazone, an insulin sensitizer that activates PPARγ, has also been shown to influence levels of MAT (128). In a 12-month study in subjects with obesity with metabolic syndrome, subjects randomized to pioglitazone had a 4.1% increase in femoral MAT and a 1.4% decrease in total hip BMD (128). Importantly, in this study there was a statistically significant correlation between the change in femoral MAT and decrease in hip BMD, strengthening the association between BMD and MAT in humans (128).

Although the association between MAT and BMD is clear, what is not clear is the mechanism of action. One likely possibility is that marrow adipocytes inhibit osteoblast differentiation through effects on lineage allocation. Alternatively, marrow adipocytes could affect mature osteoblast and or osteoclast function. It is also likely that other molecules, yet to be discovered, will have significant effects on the relationship of marrow adipocytes and bone.

“…women and adolescents with anorexia nervosa have more MAT compared with normal-weight controls….”

Composition of MAT

The degree of saturated fatty acid composition (saturation) of MAT has also been examined in a number of studies. In healthy populations, the degree of unsaturated fatty acid composition (unsaturation) has been shown to increase with age (129, 130) and is higher in women as compared with men (129). Importantly, the composition of MAT has been associated with BMD. In MAT at the lumbar spine and iliac crest, lower amounts of unsaturation have been associated with osteopenia and osteoporosis in postmenopausal women (131, 132). Some (133), but not all (132), studies have demonstrated that individuals with fractures have lower proportions of unsaturated lipids as compared with those without fractures. Similarly, in women with anorexia nervosa, a psychiatric disease characterized by low body weight due to self-induced starvation (see “MAT in anorexia nervosa” below), higher estimates of saturation in MAT at the femoral diaphysis are associated with lower BMD (134), suggesting that the composition of MAT may be an important factor contributing to the association between MAT and BMD.

In considering the degree of unsaturation in MAT in humans, it is important to consider the location of MAT measurement. We have shown that in humans, MAT in the distal tibia has an increased degree of unsaturation relative to MAT in the more proximally located femur (42). This difference in degree of unsaturation at different MAT sites may explain the differences observed in studies looking at the association between MAT composition and metabolic parameters. For example, one study in humans found that a higher degree of unsaturation in MAT from the tibia was associated with increasing insulin resistance (135), whereas other studies examining MAT in the lumbar spine and femur found that those with type 2 diabetes mellitus (DM) had a significantly lower proportion of unsaturated lipids compared with nondiabetics (136, 137). Therefore, in humans MAT may have different metabolic functions depending on its site of accumulation.

MAT in hematopoiesis

Marrow adipocytes are generally considered to be negative regulators of hematopoiesis (6, 7). However, marrow adipocytes have been shown to promote hematopoietic stem cell survival (138) and support the regeneration of hematopoietic stem cells following irradiation or 5-fluorouracil treatment by the secretion of stem cell factor (139). Acute myeloid leukemia is characterized by uncontrolled proliferation of myeloid progenitor cells, resulting in fatal infections due to suppressed differentiation of myeloid and erythroid cells. MAT has been shown to promote the proliferation of tumor cells in acute myeloid leukemia (140), and as tumor cells increase in the BM there is a correlated loss of marrow adipocytes, whereas osteoblast numbers are unaffected (141). This suggests that the effect of tumor is marrow adipocyte specific. In vitro culture experiments demonstrate that marrow adipocytes support myeloerythroid differentiation, and administration of PPARγ agonists increased marrow adipogenesis and reduced leukemia progenitor pools.

Marrow adipocytes may also play a role in the progression of other human tumors. Postmenopausal women with breast cancer have more MAT than do age- and body mass index–matched controls, with significant positive associations observed between MAT and tumor size and histologic grade (142). In vitro studies have also demonstrated preferential migration of breast cancer cells toward adipocytes isolated from marrow (143). Similarly, in prostate cancer, marrow adipocytes have been shown to promote the invasiveness of prostate cancer cells in bone, in part by driving increased expression of fatty-acid binding protein 4 on tumor cells (144). Therefore, MAT may play a critical role in the development or progression of skeletal metastases in human malignancy.

MAT and Metabolic Disease

MAT in obesity

In women who are overweight/obese, MAT is positively correlated with VWAT, such that individuals with higher VWAT have higher MAT (145). This is in direct contrast to what is observed in women who are underweight, in which VWAT and MAT are not correlated (70) (see “MAT in anorexia nervosa” below). This positive correlation between VWAT and MAT in individuals who are overweight/obese but not in individuals who are lean suggests that MAT may serve different functions depending on an individual’s nutritional status. As in healthy individuals, the inverse association between MAT and BMD is also observed in women who are overweight/obese (134, 145). Similarly, in men who are obese, an inverse association between MAT and parameters of bone microarchitecture has been reported (146). Similar to humans, mice fed a high-fat diet develop increased VWAT and SWAT with associated insulin resistance and metabolic syndrome. The mice also develop increased MAT, although the time of onset (12 to 17 weeks) is variable for reasons that are unclear (147, 148).

MAT in diabetes mellitus

Individuals with type 1 DM have a significantly increased risk of low bone mass, impaired bone microarchitecture, and an increased risk of fractures (149, 150). Mouse models of insulin deficiency, including the spontaneous diabetic NOD mouse, and streptozotocin-induced B6 mice (151, 152) demonstrate increased expression of proadipogenic genes, including PPARγ and aP2 (fatty-acid binding protein 4), in long bones concomitant with decreased osteocalcin expression (151, 152). Whereas treatment with a PPARγ antagonist (bisphenol A diglycidyl ether) decreased MAT in the streptozotocin-induced diabetic mice, it did not reverse loss of bone mass (153). In humans, individuals with type 1 DM have comparable amounts of MAT as compared with nondiabetic controls, although MAT is inversely associated with BMD in this population (154). Although MAT did not correlate with glycemic control, as estimated by hemoglobin (HbA1c), in type 1 DM, a significant correlation was observed with serum lipid levels (155). In postmenopausal women with type 2 DM and in women who are morbidly obese with type 2 DM, MAT is also inversely associated with BMD, but in this group, HbA1c levels are positively associated with MAT (136, 137). Therefore, in both type 1 DM and type 2 DM, there are significant associations between MAT and metabolic risk markers, providing evidence of a functional role for MAT in systemic metabolism.

Similarly, in a pediatric population, MAT at the lumbar spine has been positively associated with hepatic fat content in adolescent boys with nonalcoholic fatty liver disease (155), an entity associated with increased metabolic risk. However, in young, healthy individuals between 16 and 25 years of age, an association between femoral MAT and metabolic risk markers, including serum lipids, measures of insulin resistance, carotid intima-media thickness, waist-to-hip ratio, and blood pressure, was not found (156). In contrast to what has been observed in women who are overweight and obese (145), in this population of young, healthy individuals with a mean body mass index of <24 kg/m², an association between VWAT and MAT was not observed (156), suggesting that MAT may be a more important factor in systemic metabolism in individuals who are overweight/obese and/or those with metabolic disease as compared with young, healthy individuals.

In individuals who are morbidly obese, changes in MAT after the significant weight loss associated with Roux-en-Y gastric bypass (RYGB) appear to depend on the presence or absence of type 2 DM, again arguing for a role for MAT in metabolic disease and health. Whereas women with type 2 DM lose MAT after RYGB, in those without type 2 DM, levels of MAT remained stable (157). In individuals who are morbidly obese undergoing sleeve gastrectomy, levels of MAT increased 12 months postsurgery, and changes in MAT and VWAT were positively correlated as compared with individuals undergoing RYGB in whom MAT did not change, despite that both groups lost comparable amounts of weight (158). As the mechanism of weight loss differs in these two surgical procedures, with malabsorption being a component of RYGB, the divergence in changes in MAT postsurgery suggests that factors other than weight loss alone may play an important role in MAT homeostasis.

MAT in anorexia nervosa

On the other end of the weight spectrum, anorexia nervosa is a psychiatric disorder predominantly affecting females characterized by self-induced starvation and low body weight. Individuals with anorexia nervosa have low SWAT and VWAT stores (70). Despite having low SWAT and VWAT compared with normal-weight controls, women and adolescents with anorexia nervosa have more MAT compared with normal-weight controls, as measured by MRI and MRS (70, 159, 160). The findings of these noninvasive studies are substantiated by a morphometric study of BM aspirate and biopsy specimens in 44 individuals with anorexia nervosa (161). Thirty-nine percent of the subjects had hypoplastic or aplastic marrow with an increase in the fat fraction and nearly 50% of subjects had an increase in the proportion of adipocytes (161). Importantly, after recovery from anorexia nervosa, levels of MAT normalize. Abella et al. (161) found normalization of the marrow in a subset of subjects who had repeat biopsies after treatment, and using a noninvasive imaging modality (MRS), we have shown that women with a history of anorexia nervosa who have recovered (defined as normal weight and achievement of eumenorrhea) have levels of vertebral MAT that are comparable to normal-weight controls and significantly lower than levels in women with active anorexia nervosa (162).

Potential hormonal determinants of MAT in anorexia nervosa

Estrogen

A number of hormonal adaptations occur in anorexia nervosa to minimize energy expenditure in this state of chronic starvation. Functional hypothalamic amenorrhea, a result of disruption of GnRH secretion (163, 164), is one of those adaptations that allows for a decrease in energy expenditure on the costly process of reproduction during periods of nutritional stress. In postmenopausal women, marrow AV/TV and marrow adipocyte number increase after menopause, and AV/TV decreases with estrogen replacement (117) (see “Development” above). Whether estrogen is also a mediator of elevated MAT in premenopausal populations is not known but is an area of active research.

“BM should be considered a new fat depot and MAT the fourth type of fat.”

Leptin

Girls and women with anorexia nervosa have low levels of leptin, an adipokine secreted primarily by SWAT (165, 166). In animal models, subcutaneous or intracerebroventricular administration of leptin decreases marrow adiposity (167–169). In women with anorexia nervosa, leptin levels are inversely associated with vertebral MAT (170), but longitudinal studies are necessary to definitively determine whether low leptin levels are a determinant of MAT in anorexia nervosa.

Cortisol

Hypothalamic–pituitary–adrenal axis activation in states of physiologic stress results in the release of cortisol, and women with anorexia nervosa have higher levels of cortisol than do normal-weight controls (171, 172). In vitro studies demonstrate that cortisol may be necessary for marrow adipocyte differentiation (173). The effects of cortisol as a potential determinant of MAT have been evaluated in a study comparing women with anorexia nervosa to women with Cushing syndrome who have frankly elevated cortisol levels. Using quantitative CT, women with anorexia nervosa were found to have higher levels of vertebral adipose tissue compared with healthy controls, whereas the patients with Cushing syndrome had similar levels of marrow adiposity compared with healthy controls (174). In contrast, in a more recent study, patients with Cushing syndrome had higher levels of MAT at the vertebra and femur as compared with healthy controls (175). Therefore, further studies are necessary to evaluate the possible role of cortisol as a determinant of MAT in humans.

IGF-1

Although IGF-1 is secreted by the liver in response to GH, in states of undernutrition, IGF-1 levels are low despite normal or elevated GH levels (176). This state of GH resistance minimizes energy expenditure on growth during periods of undernutrition. In cross-sectional studies, low IGF-1 levels have been associated with elevated levels of vertebral MAT in women with anorexia nervosa (170), and therefore IGF-1 may be a potential determinant of MAT in this population.

Pref-1

Pref-1, a member of the epidermal growth factor family of proteins, is an inhibitor of both adipocyte and osteoblast differentiation (177). Despite that Pref-1 inhibits adipocyte differentiation, circulating levels of Pref-1 have been positively associated with femoral MAT in women with anorexia nervosa and inversely associated with lumbar spine BMD (170). In adolescent girls with anorexia nervosa, Pref-1 levels decrease in response to transdermal estrogen treatment, and this decrease is directly associated with an increase in BMD (178), suggesting that Pref-1 may play a role in mediating changes in BMD observed in response to estrogen replacement. Whether Pref-1 also mediates the changes in MAT observed in response to estrogen therapy in women is unknown.

MAT in lipodystrophy

Lipodystrophy consists of a number of disorders, including both congenital and acquired forms, that are characterized by partial or complete absence or loss of adipose tissue. Congenital generalized lipodystrophy (CGL, also known as Berardinelli–Seip syndrome) is an autosomal recessive form of lipodystrophy with four subtypes. CGL1 is the subtype due to mutations in the gene encoding 1-acyl-gycerol-3-phosphate O-acyltransferase 2, an enzyme required for TAG synthesis (179). CGL2 is due to a mutation in the Berardinelli–Seip congenital lipodystrophy 2 gene, whose function remains largely unknown, although it appears to play a role upstream of PPARγ to allow for normal adipogenesis (180). Although there are phenotypic differences between these two subtypes, including the presence of subcutaneous fat in the scalp, retro-orbital space, palms, and soles of the feet in patients with CGL1 but not in patients with CGL2, both subtypes lack MAT (181). Importantly, Simha and Garg (181) suggested that the adipose tissue present in CGL1 patients is “mechanical” (i.e., serves as structural or protective purpose) rather than being metabolically active fat, thus providing support for the concept that MAT is metabolically important in humans. CGL3, due to a mutation in the CAV1 gene, encodes a structural protein present in plasma membrane, is known as caveolin and is characterized by the presence of both MAT and mechanical fat (182). CGL4 is the result of a mutation in polymerase I and transcript release factor, and it is closely associated with CGL3. Because the polymerase I and transcript release factor is important for the formation of caveolae (the invaginations in plasma membrane containing caveolin) (183), it is associated with a much milder metabolic phenotype (183), again suggesting that MAT may be a metabolically important form of adipose tissue.

Patients with HIV often develop a form of acquired lipodystrophy typically associated with their antiretroviral therapy. HIV-associated lipodystrophy is characterized by loss of subcutaneous fat, particularly in the extremities, as well as an increase in VWAT and development of a posterior cervical pad sometimes termed a “buffalo hump” (184, 185). HIV-infected men were found to have significantly lower MAT as compared with noninfected controls, and levels were lowest in those with lipodystrophy (185). Lumbar spine BMD was also significantly lower in HIV-infected men as compared with noninfected controls, suggesting a positive association between BMD and MAT. However, when only subjects in the HIV-infected group were assessed, there was a strong inverse association between BMD and MAT (186). Therefore, although both BMD and MAT are reduced in HIV-infected males as compared with noninfected controls, the inverse association between BMD and MAT remains intact in this population.

Perspective and Future Directions

Prior to the last 8 to 10 years, the developmental origin and physiologic roles of marrow adipocytes were obscure. However, it is now known that these cells arise from the embryonic mesoderm and, in adult mice, marrow-resident progenitors with proadipogenic properties can be distinguished from other BM cells according to their cell-surface marker profile. Historically, MAT has been observed histologically just below the growth plate in long bones, indicating that progenitor cells reside in these locations. Nonetheless, induction of marrow adipocyte hyperplasia by x-irradiation or rosiglitazone results in MAT formation first in the primary spongiosa, then the metaphysis and, finally, the diaphysis. This suggests that progenitors are located throughout the medullary canal. Alternatively, marrow adipocyte progenitors may only reside in intratrabecular marrow spaces and, upon induction, migrate down the medullary canal before differentiating into mature adipocytes. Moreover, although it is known that at least some progenitors reside in the marrow space, lineage tracing experiments suggest that these cells may also be found on the endosteum. Thus, additional work is required to resolve these points.

It is becoming clear that marrow adipocytes can regulate other cell types (e.g., bone and hematopoietic) in the BM by either direct cell-to-cell contact or the secretion of adipokines (Fig. 4). Whether marrow adipocytes can regulate their own precursors in such ways remains to be shown. Moreover, marrow adipocytes can secrete adipokines that can function on cells and tissues outside the BM in an endocrine manner. Whether this regulation results in a positive or negative effect on the target cells appears to be context-dependent. As an example, in most circumstances, increased MAT correlates with decreased bone mass. Although it is likely that the mechanism involves a shift in lineage allocation away from osteoblast differentiation and toward adipocyte differentiation, it is just as likely that this is not the only mechanism. As such, it is worth a note of caution: consistent with much of the available data, it is difficult to know whether the effects that marrow adipocytes exert on other cells types, particularly in the BM, are caused by marrow adipocytes or are correlated with marrow adipocytes.

Figure 4.

Marrow adipogenesis can be increased in long bones of B6 mice by a variety of external inducers, including x-irradiation, feeding mice a diet containing rosiglitazone, a methionine-restricted diet, or a high-fat diet, or physically disrupting the BM. Alternatively, loss of certain genes [e.g., early B cell factor 1 (Ebf1), PTH/PTHrP receptor (PTH1R)] can also result in increased marrow adipogenesis; however, this occurs in the absence of any exogenous induction. This increased marrow adipogenesis occurs above the growth plate, in the secondary center of ossification, and just below the growth plate, in the intratrabecular spaces. Adipocytes can extend down the medullary canal through the metaphysis and into the diaphysis. Nerves that intercalate the marrow channel can contact these adipocytes directly. Marrow adipocytes may regulate bone cells (osteoblast and osteoclast precursors) as well as hematopoietic cells, either by direct cell contact or the secretion of adipokines in a paracrine manner. It is also possible that mature adipocytes regulate the development of adipocyte precursors in the marrow. Lastly, this ability of marrow adipocytes to regulate other cells is not restricted to BM. Secretion of adipokines, such as adiponectin, can regulate cells outside the BM in an endocrine manner. AP, adipocyte progenitor; HFD, high-fat diet; MR, methionine-restricted diet; rosi, rosiglitazone; TZD, thiazolidinedione; X-ray, x-irradiation.

It is well documented that MAT forms in response to a variety of external treatments in mice (i.e., x-irradiation, radiomimetic drugs such as 5-fluorouracil, methionine-restricted or high-fat diets) or as a result of aging and disease in humans (e.g., estrogen-deficient osteoporosis and anorexia nervosa). All of these conditions can be viewed as “trauma” to the BM. This suggests that the induction of MAT, in response to these traumas, is part of repairing the damage to the BM. Importantly, the genetic and biochemical pathways leading to increased MAT are unknown. Do these traumas use common pathways, or is the pathway to adipogenesis and repair trauma type–dependent? Certainly, in adult mice, the mechanisms that leads to marrow adipogenesis are suppressed. Loss of specific genes (e.g., Ebf1, Pgc1α, PTH1R, Fsh) and their downstream signaling results in florid marrow adipogenesis. This suggests that marrow adipogenesis is tightly regulated and mediates important functions. This idea is supported by the demonstration that adipogenesis is driven, at least in part, by repression of transcription factors expressed in BM stromal cells (human) that function as brakes on adipogenesis (187). Consequently, the loss of this suppression could account for the age-dependent increase in MAT. Understanding how these conditions induce MAT formation is critical to develop strategies for therapeutically manipulating MAT.

As we have gained a better understanding of MAT, the data show that the origin, development, and function of marrow adipocytes distinguish them from white, brown, and beige adipocytes. Additionally, because of the regulatory activity of these cells, the association with clinically relevant diseases that affect large numbers of people (i.e., osteoporosis and obesity), and their association with bone and BM repair and location within bone, they are unique. As such, BM should be considered a new fat depot and MAT the fourth type of fat.

Glossary

Abbreviations:

ACS

acyl–coenzyme A synthetase

AV/TV

adipose volume/tissue volume

B6

C57BL/6

BAT

brown adipose tissue

BM

bone marrow

BMD

bone mineral density

CGL

congenital generalized lipodystrophy

C3H

C3H/HeJ

DAG

diacylglycerol

DM

diabetes mellitus

HbA1c

hemoglobin A1c

HSC

hematopoietic stem cell

IWAT

inguinal WAT

LepR

leptin receptor

MAT

marrow adipose tissue

mG

membrane-targeted eGFP

MRS

magnetic resonance spectroscopy

mT

membrane-targeted dTomato

PGC-1α

PPARγ coactivator 1α

PLIN

perilipin

PPAR

peroxisome proliferator–activated receptor

Pref-1

preadipocyte factor-1

PTHrP

PTH-related protein

RYGB

Roux-en-Y gastric bypass

SWAT

subcutaneous WAT

TAG

triacylglycerol

VWAT

visceral WAT

WAT

white adipose tissue