Introduction

During the past 50 years, our knowledge of adipose tissue has evolved, from purely a storage site for lipids to a complex endocrine organ with a potent secretory repertoire, responsive to systemic and local signals. Concomitantly, it also became apparent that the unchecked expansion of extra-medullary adipocytes was associated with a chronic, pro-inflammatory response that was systemic and has been termed ‘metainflammation’1. Similarly, bone marrow adipocytes (once thought to be inert) began receiving increased attention, in part owing to an appreciation for haematopoietic marrow function and its ability to reconstitute the bone marrow and peripheral blood following transplantation2. The most widely held tenet was that bone marrow adipocytes arose by default during mesenchymal progenitor cell (referred to hereafter as skeletal stem cells (SSCs)) differentiation to fill marrow space during loss of haematopoiesis3. SSCs are stem cells and their progeny that are specific to the development and growth of the bone and bone marrow. Gradually, this hypothesis was replaced by the concept of bone marrow adipose tissue (BMAT) as a more dynamic organ, resembling extra-medullary adipose tissue in structure, but with major functional differences4. Discoveries abounded regarding the site of origin of marrow adipocytes, their relationship to other cells in the bone marrow and their unique responsiveness to environmental, genetic and nutritional determinants. Figure 1 shows a timeline of important discoveries in BMAT research.

Fig. 1: A timeline of the evolution of our knowledge on bone marrow adipose tissue.
figure 1

Pathologists noted in the 1880s that abundant adipocytes were found in the marrow of healthy older individuals. Bone marrow adipose tissue (BMAT) was speculated as a nutrient source for haematopoietic elements and a cushion for the skeleton97. In the 1920s, certain blood disorders and heavy-metal intoxication were associated with florid-marrow adiposity and impaired haematopoiesis. The term ‘yellow marrow’ was designated to distinguish tissue from red haematopoietic marrow98. An important report by Tavassoli and Crosby delineated anatomical and functional differences between yellow and red marrow and described, to our knowledge for the first time, the marrow adipocyte99. In the latter part of the twentieth century, investigators noted that marrow ablation following chemotherapy or radiotherapy was associated with a fibrotic response and the appearance of large numbers of adipocytes100. Similarly, others have reported that postmenopausal women with osteoporosis had abundant marrow adipocytes and decreased trabecular bone tissue77. A seminal study on anorexia nervosa has noted an association between an increase in marrow adiposity and skeletal fragility101. With the advent of magnetic resonance spectroscopy and dual-energy CT imaging, a strong negative association between BMAT and bone quantity has been reported102,103. BMAT was then postulated to regulate bone turnover by increasing bone resorption and suppressing bone formation104,105. In vitro studies have confirmed that isolated skeletal stem cells could commit to either the adipocyte or bone lineage. Furthermore, activation of PPARγ shifted skeletal stem cells into an adipocyte-like phenotype. Some in vitro evidence showed that cold temperatures could induce expression of markers of thermogenesis in marrow adipocytes106. Subsequent studies using in vivo models confirmed that lineage allocation enables early progenitors to shift between osteoblasts and adipocytes, providing another mechanism for altered bone remodelling after hormonal, nutritional or environmental injury107. Ablation of BMAT in mice using global and conditional genetic deletions drives increased cortical and trabecular bone, in part because of the absence of growth factors secreted by marrow adipocytes5,23. MALP, marrow adipogenic lineage precursor.

Full size image

Notwithstanding these advances, challenges remain in our understanding of bone marrow adipocytes, foremost of which are the absence of a tractable organoid system and the lack of an in vivo imaging technique to observe real-time changes in lineage and responses to external stimuli. These technological advances are likely to be overcome in the next decade; in the meantime, some important new insights have provided fresh support for the dynamic nature of these cells. For example, the origin of the marrow adipocyte has been better defined by refinements in lineage-tracing studies, particularly in animal models with genetic manipulations5. These techniques and single-cell studies have opened new perspectives on where and how cells originate, as well as their function6. Importantly, these methodological breakthroughs have also led to new hypotheses about BMAT and the skeleton, some of which will be noted later.

In this Review, we examine how bone marrow adipocytes originate, what their function is in normal and pathological states and how these cells influence skeletal remodelling, haematopoiesis and whole-body metabolism during nutritional challenges. We focus on both animal models and human studies to help understand this unique cell type, its response to the external nutrient environment and the resultant effects on the skeleton. We finish by addressing some critical unanswered questions, particularly around weight loss from caloric restriction and its effects on the skeleton.

Origins of bone marrow adipocytes

Overview

The cellular composition of bone marrow is quite heterogeneous, with mesenchymal and haematopoietic cells at different stages of differentiation. SSCs can give rise to chondrocytes, marrow stroma, osteoblasts and adipocytes (Fig. 2), whereas haematopoietic stem cells (HSCs) give rise to all the blood and immune cell lineages. Haematopoietic cell development is supported by a tissue network of marrow stroma, which provides the structure for different niches within the marrow space7. In addition, marrow adipocytes and osteoblast lineage cells regulate haematopoiesis. As an example, niches that support haematopoiesis form at the marrow–endosteal bone interface8,9,10,11,12. The marrow space is highly innervated and has an extensive blood supply. Outside the bone marrow, mesenchymal stem cells can be found in many organs, and extra-medullary white, brown and beige adipocytes derive from mesenchymal stem cells in situ. Marrow adipocytes possess characteristics that are distinct from these other types of adipocytes. In this discussion, SSCs and their progeny are specific to the development and growth of the bone and bone marrow. This refinement helps to define this complex tissue as an organ7. Of note, stem cells from the stromal–vascular fraction of white adipose tissue (WAT) can differentiate into osteoblast-like cells with the appropriate in vitro stimulation, although these cells are not SSCs and their chromatin landscape differs from SSCs13,14,15.

Fig. 2: Skeletal stem cells can differentiate into osteoblasts, chondrocytes, stromal cells or adipocytes.
figure 2

Skeletal stem cells (SSCs) differentiate into early mesenchymal progenitor cells, which can give rise to the adipocyte or osteoblast lineages. One of the earliest progenitors identified that can become an adipocyte is the marrow adipogenic lineage precursor (MALP)27. However, we cannot rule out the possibility that other early progenitor cells can differentiate into marrow adipocytes. Transcriptional, paracrine and endocrine factors ultimately define the fate of progenitors. Of note, with the activation of PPARγ early in the progenitor lineage, adipocyte differentiation occurs at the expense of osteoblastogenesis. The zinc finger proteins (ZF423, ZFP467 and ZFP521) are expressed early in mesenchymal lineage differentiation both in SSCs and in peripheral adipose depots, but have direct lineage regulation32,66. In the pre-osteoblast lineage, RUNX2 is the critical transcriptional regulator that sets the stage for osterix activation, followed by differentiated markers of the terminal osteoblast such as RANKL, osteocalcin and the type 1 collagen COL1A1. Whether terminally differentiated osteoblasts can become adipocytes or differentiated adipocytes can become osteoblasts is unclear (dashed arrow). Endocrine, paracrine or neural factors, or pharmacological agents that exert effects on adipocyte or osteoblast differentiation are shown in the grey boxes. PTH, parathyroid hormone; SNS, sympathetic nervous system.

Full size image

Although the presence of marrow adipocytes has been recognized for more than a century, their origin and differentiation has only been described within the past decade. Early experiments have shown that in vitro culture of plastic-adherent, non-haematopoietic bone marrow cells leads to the development of a discrete fibroblast colony-forming unit (CFU-F), that is, adherent, individual colonies that proliferate in culture, which could differentiate into the adipocyte, osteoblast and chondrocyte lineages8,9,10. These early observations have stood the test of time, being confirmed with our current understanding of lineage allocation and that SSCs do not express haematopoietic or endothelial cell markers10,11,12.

Key findings from mouse studies

The leptin receptor is a marker of some SSCs

SSCs in the bone marrow and the stromal–vascular fraction of other adipose depots have been delineated using fluorescent reporter constructs for lineage tracing, direct analysis of the cell-surface phenotype by fluorescence-activated cell sorting (FACS), and single-cell RNA sequencing (RNA-seq)16,17,18,19 (Table 1). One key study used an antibody to the leptin receptor (LEPR) in conjunction with Lepr-cre:tdTomato conditional reporter mice to show LEPR expression around sinusoids and arterioles throughout the marrow space, consistent with the location of SSCs19. This population of cells were LEPR+CD45−TER119−CD31− and uniformly expressed paired related homeobox transcription factor 1 (Prrx1), PDGFRα and CD51. This population of cells accounted for the majority (94%) of bone marrow CFU-Fs, were able to differentiate into adipocytes and chondrocytes, and were a major source of osteoblasts and marrow adipocytes in adult mice. Moreover, in the bone marrow of adult mice, few CFU-Fs were Nestin+ and able to become osteoblasts. These data indicate that SSCs are not in the haematopoietic (CD45−) or endothelial (CD31−) lineages, are multipotential, are marked by LEPR and trace from very early mesenchymal progenitors as suggested by Prrx1 expression16,17,18,19.

Table 1 Genetically engineered and inbred mouse models to elucidate the effects of diet on marrow adipocytes
Full size table

SSCs express very early mesenchymal markers

Studies using Prrx1-Cre mice showed that Prrx1 is expressed in early mesenchymal progenitors of the head and limbs20 and, thus, only labels bones and adipose depots that are associated with these developmental lineages. For example, Prrx1-Cre labels 96% of white adipocytes in the inguinal depot (iWAT) and >95% of marrow adipocytes in the tibia and femur20,21. By contrast, Prrx1-Cre does not label visceral white adipocytes and brown adipocytes21. Thus, LEPR+ marrow adipocytes, SSCs and iWAT progenitors share some cell-surface markers (LIN−CD34−SCA1+CD29+CD24+PDGFRα+)19,20.

Blocking marrow adipocyte development

The protein encoded by the PPARγ gene (Pparg) is considered a ‘master’ regulator of adipocyte differentiation. Both conditional Pparg deletion in adipocytes and global deletion of Pparg leave mice devoid of adipose tissue5,22,23. To attempt to block BMAT development in long bones of mice, Prrx1-Cre mice were crossed with Ppargfl/fl mice (to attempt to deplete marrow adipocyte progenitors) and the adult offspring were irradiated to induce marrow adipogenesis22,23. As expected, Cre− control mice had striking marrow adipogenesis in both the proximal and the distal tibia and in the femur. By contrast, few if any marrow adipocytes could be seen even after irradiation in the long bones of the Cre+ mice. Of note, few adipocytes are seen in the bone marrow of normal young and adult B6 mice. Importantly, similar numbers of marrow adipocytes could be seen in the caudal vertebra from Cre− and Cre+ mice, suggesting that caudal adipocytes do not express Prrx1 and arise from different progenitors compared with adipocytes in long bones. These data show that Prrx1-Cre traces the majority of marrow adipocyte progenitors in adult mouse tibia and femur24,25.

Regulation of haematopoiesis by SSCs

In addition to differentiating into their specific cell lineages, SSCs also function to regulate haematopoiesis in bone marrow niches26. Stromal cell-derived factor 1 (also known as CXCL12) is secreted by stromal cells and regulates the maintenance of HSCs and lymphoid progenitors27. Osterix (encoded by OSX) is a zinc finger-containing transcription factor specific to osteoblasts that functions early in osteoblast lineage differentiation and is essential for the development of mature, matrix-secreting osteoblasts28. Prrx1-Cre:Cxcl12fl/fl mice were generated as a model of SSC depletion, whereas Osx-Cre;Cxcl12fl/fl mice were a model of early progenitor depletion29. Bone marrow cellularity was reduced by approximately 50% in both models, in part because of the loss of B cells. A statistically significant decrease was observed in the number of HSCs in the Prrx1-Cre:Cxcl12fl/fl mice, whereas only a modest decrease was seen in the Osx-Cre;Cxcl12fl/fl mice29. Similarly, in transplantation studies, a defect occurred in multilineage long-term repopulation in recipient mice receiving bone marrow from Prrx1-Cre:Cxcl12fl/fl mice but not from Osx-Cre;Cxcl12fl/fl mice30. These data suggest that SSCs support HSCs, whereas osteoblast progenitors support the B cell niche and do not notably affect HSCs (Fig. 2).

Variations in SSC phenotypes

In a different set of experiments, cells were isolated from the femoral growth plate of Rainbow mice (a multicolour Cre recombinase mouse reporter strain that expresses multiple fluorescent proteins from a single genomic locus) and analysed by FACS31. A population of stem cells was isolated (CD45−TER119−TIE2−AlphaV+THY−6C3−CD105−CD200+; referred to as AlphaV+ cells), which gave rise to eight discrete populations in a hierarchical manner. These postnatal stem cells differentiated into bone, marrow stroma and cartilage, but importantly not adipocytes31. Thus, AlphaV+ cells were multipotent and retained their self-renewing capacity in vitro and in vivo. As these SSCs differentiated, their progeny became more restricted in their cell-type specificity, as expected. Some of the differentiated populations were LEPR+ and Nestin+, consistent with some previous data19. As AlphaV+ cells do not give rise to marrow adipocytes, this cell population could be a different SSC population to LEPR+ SSCs.

In a series of papers, cells were isolated from crushed tibiae and femurs of fluorescent reporter mice by collagenase digestion, and then subdivided based on expression of stem cell antigen 1 (SCA1; also known as LY6A) by FACS and expanded in culture30. These plastic-adherent cells also expressed PDGFα30. On the basis of this approach, four distinct non-haematopoietic, non-endothelial populations were identified. The first population was CD45−CD31−SCA1+CD24+ stem cells, which gave rise to adipocytes, osteoblasts and chondrocytes. More cells of this first population resided in the metaphasis than in the diaphysis, not on the endosteum but perivascularly. The second population was CD45−CD31−SCA1+CD24− committed progenitor cells that gave rise unilaterally to adipocytes and were evenly distributed in the metaphysis and diaphysis and perivascularly. The third population was CD45−CD31−SCA1−PDGFRα+ committed progenitor cells that gave rise to osteoblasts and chondrocytes and had little adipogenic potential. The fourth population was CD45−CD31−SCA1−ZFP423+, a committed progenitor or precursor cell population that gave rise to adipocytes. Interestingly, less than 1% of SCA1− cells expressed ZFP423 (a zinc finger protein and transcriptional coregulator of PPARγ that regulates the potential of cells to undergo adipocyte differentiation32), whereas ZFP423 was uniformly expressed by the SCA1+ subset. These data suggested that CD45−CD31−SCA1−ZFP423+ cells were a more mature adipocyte-lineage committed population than the other three subsets. These cells suppressed competitive transplantation assays following lethal irradiation and inhibited fracture repair, supporting the idea that bone marrow adipocytes inhibit bone repair and haematopoiesis.

In a follow-up 2021 study, two distinct populations of SSCs were isolated by cell-surface phenotype: CD45−TER119−Tie2−CD51+THY1−6C3−CD105− cells, which give rise to more restricted stromal and osteochondral lineages, and CD45−CD31−PDGFRα+SCA1+CD24+ cells, which can also give rise to bone, cartilage and, importantly, to adipocytes18,30. As these adipogenic progenitors differentiate, their cell surface takes on a more adipogenic phenotype. These progenitors are thought to be the source of all marrow adipocytes18,30. Importantly, both these SSC subsets expressed LEPR, demonstrating that LEPR is a marker for the majority of SSCs6,18.

Where you get your cells from matters

Where in the bone SSC populations are obtained from can dictate their phenotype and lineage development. This issue was illustrated in experiments that used the Col2-Cre;Rosa26 (Col:Td, tdTomato) fluorescent reporter mouse model, in which all CFU-F, marrow adipocytes and osteocytes were tdTomato+, suggesting that all bone marrow SSCs in this model should be TdTomato+ (ref. 33). In pilot experiments, two populations of bone marrow cells were isolated33. For the first population, the epiphyses were removed from tibias and femurs and the bone marrow flushed from the medullary canal (central bone marrow), which is a standard method of bone marrow collection. We know from our own experience that this method leaves cells adherent to the endosteal surface. For the second population, the exterior of the flushed bones was scraped and treated with collagenase and trypsin to remove periosteal cells. The bones were then cut in half, exposing the trabecular bone and endosteal surface, and were enzymatically digested and the cells collected (endosteal bone marrow). The enzymatic method enabled the capture of cells adherent to endosteal surfaces. The cells isolated from the endosteal bone marrow of tdTomato mice had more SSCs, were more proliferative, gave rise to adipocytes, osteoblasts and chondrocytes, and decreased in number with age compared with cells isolated from the central bone marrow6.

The top 1% of the Tdtomato+ peak in endosteal bone marrow cells were isolated from 1-month-old Col2-Cre;Rosa26 mice by flow cytometry and analysed by single-cell RNA-seq6. Gene expression profiling revealed nine clusters of mesenchymal cells, which included adipocytes, osteoblasts, osteocytes and chondrocytes. Cell trajectory analysis was used to identify cluster 1 cells as the progenitors to the other lineages, which expressed Sca1, Cd34, Thy1 and Lepr, and gave rise to adipocytes and osteoblasts. Compared with the other clusters, which are more lineage restricted, Sca1 alone preferentially marked cluster 1 cells. Using the femurs from 1-month-old Adipoq:Td:Col1a1-GFP mice (an adipocyte-specific reporter mouse with or without Col1a1-GFP that traces osteoblasts), a large cluster of TdTomato+ cells could be seen histologically just below the growth plate of young, but not old, mice6. This population expressed adipocyte markers, including PPARγ, CEBPα, adiponectin and lipoprotein lipase, but notably did not express PLIN1 and FABP4, suggesting that these cells did not contain lipid droplets. Further analysis of this cell population indicated that these cells were a novel population of adipocyte lineage cells that did not contain lipid droplets and were intermediate between SSCs and classic lipid-filled marrow adipocytes. They are known as marrow adipogenic lineage precursors (MALPs)6 (Fig. 2).

MALPs did not express UCP1 (a brown or beige adipocyte marker), TNFRSF9 (a beige adipocyte marker) or leptin (a white adipocyte marker), consistent with the idea that marrow adipocytes are distinct from brown, beige and white adipocytes6,19. Deletion of MALPs in mice caused pathological changes to the marrow vasculature, which become dilated with decreased vessel density6. A distinct increase was observed in trabecular bone mass in the metaphysis extending into the diaphysis, with increased cortical bone thickness consistent with an increase in osteoblasts. Haematopoietic cells in the bone marrow were unaffected. Whether the increase in bone is because of a shift in lineage allocation resulting from the loss of MALPs or is an indirect effect is not clear.

Current understanding of lineage development

Our understanding of the lineage development of marrow adipocytes has made notable progress over the past 10 years. Evidence clearly shows that marrow adipocytes derive from a multipotential, self-renewing SSC population in the bone marrow. These cells differentiate to yield more lineage-restricted intermediate progenitor or precursor cells, the result being mature functional marrow adipocytes, osteoblasts, chondrocytes and marrow stroma (Table 1 and Fig. 2). This progression from less restricted to more restricted cell types is similar in kind to the development of white adipocytes and haematopoietic cell lineages. Still, a number of challenges remain.

Many assume that a single true SSC population exists that gives rise to marrow adipocytes. On the basis of the available data, this hypothesis seems unlikely. As an example, in mouse long bones, marrow adipocytes first appear in the distal tibia, with few if any adipocytes in the proximal tibia or femur6,18,19,33. Using Prrx1-Cre:Ppargfl/fl mice, we have shown that the long-bone marrow adipocytes fail to develop if Prrx1+Pparg+ progenitor cells are depleted21,22, suggesting that the progenitor that gives rise to marrow adipocytes in tibias and femurs is Prrx1+. This finding is true regardless of whether marrow adipocytes appear early in life or are induced by various treatments34. By contrast, in adult mice, the progenitor that gives rise to marrow adipocytes in the spine is Prrx1−, therefore, from a different SSC population than in tibiae and femurs22. Various interventions cause increased marrow adipogenesis in mice, including X-irradiation, caloric restriction, methionine restriction, high-fat diet (HFD), treatment with thiazolidinedione (an insulin-sensitizing drug) or physical disruption of the marrow channel, indicating that marrow adipocytes are involved in the repair and recovery of bone marrow function34.

Limitations and variations in phenotypes

The cellular phenotype of SSCs varies depending on the laboratory. As an example, expression of the genes encoding osterix and Gremlin by SSCs has been reported by some but not by others35. Even Lepr expression remains quite variable. The age of mice has a clear effect on the phenotype of the SSC population. Another explanation for these differences could be the starting population of cells used for analysis. In vivo marrow adipogenesis in long bones is most prominent just distal to the growth-plate trabecular bone extending into the bone marrow in the primary spongiosa6. This finding is consistent with the presence of increased marrow adipocyte progenitors in this location. Upon stimulation, marrow adipocytes appear to move distally, filling the medullary canal22,36. Whether this finding simply reflects a need for more space as more cells differentiate or whether progenitors or precursors can be found in the metaphysis, or both mechanisms occur, also remains unresolved. Thus, cells obtained from the endosteal bone marrow, especially in and around the growth plate, might be a superior starting population for lineage analysis than central bone marrow cells flushed from the medullary canal.

Open research questions

Much work still remains to be done on characterizing the interactions of marrow adipocytes and other cells in the marrow. In acute myeloid leukaemia (AML), large numbers of dysfunctional leukaemic blasts accumulate in the bone marrow, resulting in the loss of myelocytes and erythrocytes, causing life-threatening infections. In vivo and in vitro experiments show that AML causes a disruption in the adipocytic niche in the bone marrow, resulting in the failure of myeloid and erythroid lineage development37; thus, studying marrow adipocyte biology during AML is of clinical interest. In mice, HFD feeding induces bone marrow adipogenesis and an increase in the number of Ly6CHigh monocytes in the bone marrow, which was accompanied by a shift in monocyte metabolism, reducing oxidative potential and increasing glycolysis and mitochondrial function38. How a HFD drives bone loss in relation to the haematopoietic niche is critical to fully understand the long-term implications of obesity. Finally, osteoclastogenesis requires colony-stimulating factor 1 (CSF1); however, the cellular source of CSF1 in the bone marrow has been unclear. MALPs express much higher levels of CSF1 than other mesenchymal cells. Deletion of MALPs in mice resulted in increased trabecular BMD, with decreases in TRAP+ osteoclasts39. This finding raises the question of whether altering the fate of marrow adipocytes could be used to treat low-bone-mass syndromes.

Nutrient regulation of BMAT

Adipose tissue is one of the largest organs in the body and 99% of the cells in this tissue are adipocytes40. These cells arise from several locations postnatally and their site of origin (for example, visceral, subcutaneous, brown, beige or breast adipose depots) is thought to determine cellular function. Short-term and long-term changes in dietary regimens affect adipose depots in unique ways that are also location dependent. For example, in humans after a HFD, a modest 1–5% increase in thermogenesis occurs principally from brown adipose tissue because of activation of the sympathetic nervous system, whereas lipid storage ramps up in visceral and subcutaneous depots41. Like other depots, BMAT is unique in its origin and its response to dietary changes. For example, historic research from the 1970s showed that the bone marrow of individuals who have faced famine or starvation becomes laden with adipocytes and, with persistent food deprivation, becomes gelatinous42. Similarly, most starving mammals are known to preserve adipose tissue in two sites: the bone marrow and the lymphatic tissues43,44. As haematopoiesis is such an energy-consuming process, it follows that neighbouring adipocytes could be used as a reservoir for energy during caloric restriction. This paradoxical response to extreme dietary restriction has been the focus of recent research efforts.

In mouse models, nutrient challenges (that is, a HFD and caloric restriction) both cause enhanced marrow adiposity, whereas the extra-medullary responses are quite distinct. Furthermore, oestrogen deficiency drives marrow adiposity in mice and can act as an interactive component associated with dietary changes45,46. In many studies, dietary interventions also cause bone loss, although exceptions do exist. A 2018 study of HFD feeding in adult C57BL/6J mice demonstrated a very pronounced increase in BMAT and bone loss, with evidence suggesting that the bone marrow adipocytes in these mice were senescent and insulin resistant47. The bone loss resulted from reduced bone formation and enhanced adipogenesis from a shift in lineage allocation. Interestingly, those high-calorie-diet-fed mice developed a pro-inflammatory response in peripheral adipose depots, yet such a response was not detected in the bone marrow or adjacent marrow adipocytes47. Preliminary work suggests that after mice were fed a HFD (45% calories from fat) for 8 weeks, expression of inflammatory cytokines was not observed in either marrow adipocytes or marrow macrophages (C.J.R., unpublished observation). Conversely, bone resorption was markedly increased in the HFD-fed mice. Thus, the mechanism of bone loss with HFD might involve uncoupled remodelling, with increased rates of bone resorption and decreased bone formation resulting from greater allocation of SSC into the adipocyte lineage.

In human studies, the effects of obesity on BMAT are not clear, with some reports showing an increase in BMAT, whereas others demonstrating no change or less BMAT43,48. Similar findings have been noted for BMD in individuals with obesity49. In mice, both a HFD and caloric restriction lead to increased BMAT and bone loss47,50 (Table 2). However, the marrow responses that drive adipogenesis differ (see later). Together, most studies demonstrate that the marrow adipocyte response to external stimuli differs from that of extra-medullary adipocytes and that nutritional composition and intake directly affect the fate of marrow adipocyte precursors. Genetically engineered mouse models have provided major insights into these changes.

Table 2 Clinical disorders associated with changes in BMAT volume in humans and their relation to mouse studies
Full size table

Clinical importance

The clinical importance of understanding the effects of nutritional challenges on the skeleton and the bone marrow cannot be emphasized enough48,49,50. Caloric restriction is the most common approach to treating obesity, but bone loss usually occurs with weight loss. For example, in long-term human studies (for example, CALERIE), in which 25% caloric restriction was attempted in healthy volunteers for 2 years, statistically significant bone loss was evident at the spine and femur51. Another effective therapeutic weight-loss strategy in individuals with obesity, gastric bypass, causes bone loss but with variable effects on BMAT52. The most common surgical procedure to treat obesity, vertical sleeve gastrectomy, consistently causes loss of bone mass as measured by dual-energy X-ray absorptiometry, but this effect is not tied temporally to the weight loss or to a decrease in BMAT53. Other weight-loss strategies, such as time-restricted eating, intermittent fasting and treatment with glucagon-like peptide 1 receptor agonists (GLP1RAs), have not been studied relative to the BMAT response, although bone loss has not been reported with GLP1RA use. Of note, the effect on bone marrow adipocytes with weight loss seems to be sex dependent, such that men have a greater increase in BMAT than women52. In human volunteers, after a 10-day fast and notable weight loss, BMAT increased rapidly, more so in men than in women, and was reversed with an ad libitum diet54. By contrast, gastric bypass has been reported to either decrease or increase or to not change BMAT in humans55.

Experimental validation in mice

For C57BL/6J and C3H/HeJ mice, 30% caloric restriction for as short as 4 weeks causes substantial bone marrow adipose infiltration, in both male and female individuals across a wide age range. Cortical and trabecular bone loss and bone marrow adiposity is more pronounced in young C57BL/6J mice that undergo caloric restriction56. However, in mature male and female mice, caloric restriction leads to cortical bone loss, with minimal or no change in trabecular bone volume fraction. Histomorphometric studies in young mice undergoing caloric restriction have revealed a profound suppression in all aspects of bone remodelling, particularly bone formation, which is almost certainly a response to reduced substrate availability57. SSCs isolated from those young mice show an enhanced capacity to differentiate into adipocytes, although increased expression of tissue-nonspecific alkaline phosphatase (TNAP; a key enzyme in skeletal mineralization) has also been noted in these cells (L. Liu, personal communication). In mice undergoing vertical sleeve gastrectomy following a HFD, BMAT declines considerably and this change is associated with statistically significant bone loss58. Cold exposure also causes a similar skeletal phenotype in mice, that is, less BMAT alongside bone loss36. In dams undergoing lactation, bone loss is pronounced and marrow adiposity is reduced59. Thus, in rodent models, nutritional changes in diet can affect both skeletal remodelling and bone marrow adipose infiltration. However, the negative relationship between BMAT and bone loss is not always consistent or reproducible in these studies. In addition, notable sex differences have been observed in skeletal changes between male and female mice undergoing caloric restriction.

Observations from individuals with anorexia nervosa

One human disorder that recapitulates the effect of calorie deprivation in mice is anorexia nervosa. Severe bone loss and fractures are frequent and often difficult to treat in the face of very low body weight60. BMAT is consistently increased in both the peripheral and the axial marrow and this increase has been related to both the duration and the severity of anorexia60. BMAT is reversed by weight gain, but bone mass is less easily restored and is dependent on the age of the individual. Oestrogen treatment can improve bone mass and reduce BMAT but does not fully restore body weight61,62. Anabolic agents, such as parathyroid hormone and insulin-like growth factor 1 (IGF1), have been used to reverse the skeletal phenotype with mixed success63. No histomorphometric data on skeletal remodelling with anorexia have been published, although systemic biochemical markers suggest uncoupling, with suppressed bone formation and increased bone resorption. The paradoxical aspect of anorexia is that as adipose tissue mass declines with caloric restriction in peripheral adipose depots, BMAT increases. The mechanisms responsible for this process are probably multifactorial and include paracrine, endocrine and neural factors. High levels of glucocorticoids, low circulating and marrow levels of IGF1 and changes in central mediators of appetite all probably contribute to the profound suppression in bone formation64,65,66. Whether the high numbers of marrow adipocytes contribute to the increase in bone resorption in humans is unclear, although this has been shown in other animal model systems.

Manipulating BMAT in mouse models

BMAT expansion is induced by injury, for example, radiation, drugs, mechanical ablation or chemotherapy19,67,68. Calorie deprivation could be considered injurious, that is, as an injury to cellular function resulting from substrate deficiency. In experimental mouse models, caloric restriction is one of the most potent inducers of BMAT57. However, whether the marrow adipocyte response to nutrient challenges is primarily because of signalling directly to the adipocyte from neural or skeletal (paracrine) connections, or occurs as a secondary repair or compensatory response that somehow preserves skeletal function, has not been demonstrated. To address that question, several experimental approaches have been used. These include deletion of marrow adipocytes using diphtheria toxin receptor or targeted Pparg deletion, and/or bone marrow adipocyte-targeted deletion (Table 1).

Gene-targeted deletions in mice

The first genetically modified animal model that examined the role of marrow adipocytes was the A-Zip or fat-less mouse, in which expression of the dominant negative A-ZIP/F protein was under the control of the adipose-specific Fabp4 promoter69. This genetic construct suppressed Pparg expression. The mice have reduced adipose depots globally and in the marrow, increased HSC frequency in tail vertebrae and accelerated haematopoietic recovery after irradiation, as well as high bone mass, suggesting that BMAT was a negative regulator of haematopoiesis69. By contrast, a 2017 paper has demonstrated that the presence of marrow adipocytes in long bones might be beneficial for haematopoiesis using Col1a1*2.3-cre;ScfGFP/fl mice70. This study was based on earlier observations in vitro that marrow adipocytes secreted stem cell factor and this was essential for haematopoietic recovery71. Thus, marrow adipogenesis could theoretically be thought of as an emergency response to cytopenia and a rapid way to reconstitute the marrow after injury. A similar compensatory mechanism could be operative in the bone, as noted in a model system of calorie restriction.

Adipocyte-specific deletion models

Several investigators reported that specific deletion of marrow adipocytes in mice could drive high bone mass, supporting the notion that these cells negatively affect the skeleton5,23. Ablation of adiponectin-expressing cells in the bone marrow of mice yielded a rapid and profound increase in systemic bone mass5. This enhancement in bone formation was because of induction of bone morphogenetic protein signalling as a result of the elimination of its inhibitors produced by marrow adipocytes, combined with epidermal growth factor receptor stimulation. The two inhibitors found in marrow adipocytes were chordin 1 and gremlin 1, both expressed fairly early in lineage allocation, consistent with the adiponectin Cre temporal expression5. The genetic modification was accomplished using a conditional expression of the diphtheria toxin receptor (DTR) crossed to an adiponectin Cre mouse. Although extra-medullary adipocytes were also deleted in addition to marrow adipocytes, the authors established that white adipose transplants in those mice did not affect the osteogenic response5. Similarly, transplants of DTR-null adipose tissue into wild-type mice did not affect the skeleton. Notwithstanding, concerns were raised that the whole-body lipoatrophic phenotype and insulin resistance of the conditional-DTR mice might contribute to the skeletal changes. A new model system was developed and findings published in 2022, using the FLPo-dependent-Adipo-Cre (FAC) mouse that deletes genes in marrow adipocytes only, further supported the findings of an earlier work that the FAC mouse model has the potential for providing future insights into the functional importance of BMAT23,70,72.

FAC mice

The creation of the FAC mouse is worth noting. Studies published around 9–10 years ago from several laboratories have shown that Osx traces to osteoblasts and bone marrow adipocytes, but not to white adipocytes28,29. Thus, CRISPR–Cas9 was used to create Osx-FLPo mice, with an in-frame fusion of Osx and optimized FLPo (the flippase enzyme) separated by a P2A self-cleaving sequence, which enabled independent functioning of the two proteins72. Using eGFP, these mice had positive staining in osteocytes, osteoblasts, bone marrow adipocytes and a subset of marrow stromal cells within the bone. FAC mice were then created, which contained an internal ribosome entry sequence followed by FLPo-dependent Cre in reverse orientation within the 3′ untranslated region of endogenous Adipoq23. FLPo expressed from the Osx locus recombines Cre to the correct orientation in progenitors of osteoblasts and bone marrow adipocytes. However, as Adipoq is selectively expressed in adipocytes, Cre was expressed in bone marrow adipocytes, but not in osteoblasts or other adipose depots23. Cre efficiency was ~80% in both male and female mice over 16 weeks of age with one Cre allele and over 90% at 12 weeks of age in mice with two Cre alleles.

In subsequent experiments, FAC mice were then crossed with floxed Pparg mice to delete Pparg in marrow adipocytes only23. The FAC Pparg−/− mice exhibited markedly reduced BMAT but high cortical bone mass, similar in magnitude to what was observed with the targeted DTR5. However, no change was noted in the trabecular compartment23. In a second set of experiments, FAC mice were crossed with ROSA DTA mice harbouring a LoxP-flanked STOP cassette proximal to the DTA sequence. In this model, Cre recombinase excises the STOP cassette to enable expression of cytotoxic DTA in bone marrow adipocytes, thus inducing cell death23,71. In the DTA conditional mice, cortical bone mass and trabecular bone volume fraction were increased and haematopoiesis was markedly impaired. In addition to a reduction, but not loss, of bone marrow adipocytes at the distal and proximal tibia, a statistically significant increase was observed in bone formation. Intriguingly, the DTA mice were resistant to cortical bone loss with 30% caloric restriction. Together, marrow adipocytes clearly have a supportive role for the haematopoietic niche in trabecular bone of the proximal femur, whereas the reduction in marrow adipocytes distally was associated with a concomitant increase in bone formation and mass72.

FAC mice were also used to test the hypothesis that marrow adipocytes might have a supportive role for the skeleton as well as the haematopoietic niche72. Thus, FAC mice were crossed with Pnpla2fl/fl mice to delete adipose triglyceride lipase (not only the rate-limiting enzyme for lipolysis but also a transacylase that drives synthesis of fatty acyl esters of hydroxy fatty acids) only in marrow adipocytes71. FAC Pnpla2−/− mice fed with a regular diet or HFD exhibited a bone marrow adipocyte phenotype comparable with wild-type mice; BMD was also not different in the mutants compared with Pnpla2+/+ mice72. However, during 30% caloric restriction, trabecular bone volume fraction was statistically significantly reduced in male mice, although not in female mice. The latter finding contrasts with the absence of trabecular bone loss in Pnpla2+/+ mice after 30% caloric restriction. Also, haematopoiesis was impaired in mice that had absence of ATGL in bone marrow adipocytes, as was trabecular and cortical bone regeneration using a wound defect in the tibia. Bulk RNA-seq of the bone marrow in male FAC Pnpla2+/+ mice revealed that adipogenic genes were the most upregulated pathway by 30% caloric restriction, but osteogenesis was the second most differentially expressed network72. Conversely, in FAC Pnpla2−/− mice, both pathways were not activated. Thus, 30% caloric restriction probably drives adipogenic and osteogenic responses in bone marrow, the latter of which might be essential for trabecular bone maintenance. Similarly, genes associated with de novo lipogenesis (for example, Fasn, Acaca, Acacb, Echdc2 and Echdc3) were also upregulated with caloric restriction in the FAC Pnpla2+/+ mice, but not in the FAC Pnpla2−/− mice72.

Further support for the tenet that marrow adipocytes are a source of substrate comes from lipid ontology enrichment analysis of lipids in marrow sera after 10 days of fasting in healthy volunteers54 (S. Costa, unpublished observation). Lipid storage and synthesis were the top two networks in marrow sera compared with pre-fasting values. Furthermore, RNA-seq of marrow adipocytes isolated from those fasting individuals revealed that lipoprotein lipase (LPL) was one of the most highly expressed genes with fasting (S. Costa, unpublished observation). Hence, marrow adipocytes could be extracting fatty acids from the circulation for recycling and for later use by other marrow components, such as osteoblasts or haematopoietic cells. Thus, much like the ‘support’ hypothesized for haematopoiesis, increased BMAT might be a compensatory response for maintaining skeletal integrity during nutritional stress. By contrast, under normal conditions, marrow adipocytes secrete factors that tend to suppress bone formation and the osteogenic response72 (Fig. 3).

Fig. 3: Multifunctional marrow adipocytes.
figure 3

Bone marrow adipose tissue (BMAT) is dynamic and responsive to nutrient cues. BMAT can be a source of inhibitory cytokines or a fuel depot for use by haematopoietic and skeletal cells in the niche. In the basal condition with adequate energy, marrow adipocytes restrict osteoblast differentiation and function. During caloric restriction the adipocytes can release free fatty acids and glycerol, which could provide energy for osteoblast differentiation. However, if adipose triglyceride lipase (the rate-limiting lipolysis enzyme) is inhibited, osteoblasts cannot differentiate and bone formation is reduced.

Full size image

Amino acid deficiency and BMAT

Evidence also suggests that amino acid deficiency might phenocopy 30% caloric restriction. First reported in 1992, sulfur-containing amino acid restriction (especially methionine restriction) extends lifespan across many species, from yeast and flies to worms and mice. Mice live up to 45% longer than controls, with less age-related diseases, including adiposity, reduced insulin resistance and cancer, and improved overall metabolic status73. Mice placed on a continuous methionine restriction (0.12%) versus control (0.86%) diet for 8 weeks lose body weight resulting from a loss of subcutaneous and visceral adipose tissue, although this effect is more pronounced in male mice74. Concomitantly, the mice have a striking increase in long-bone marrow adipogenesis, with increased beige adipogenesis in the iWAT depot (M.C.H. and G. P. Ables, unpublished observation). Trabecular and cortical bone mass are significantly decreased and the bones are weaker biomechanically than controls74,75. This phenotype is strikingly similar to that seen with caloric restriction in mice and anorexia nervosa in young women. Importantly, the beneficial effects of methionine restriction, including improved glucose metabolism and protection against HFD-induced obesity, can be achieved by feeding a methionine-restriction diet intermittently, which reduces the increased BMAT and bone-loss phenotype seen with continuous methionine restriction73.

To assess whether humans could benefit from a methionine-restriction diet, healthy men and women were placed on a methionine-restriction diet or a diet restricted in both methionine and cysteine for three 4-week feeding periods, separated by 3–4-week wash-out periods76. Volunteers in the methionine-restricted and cysteine-restricted group had reductions in body weight and plasma levels of cholesterol, low-density lipoprotein, blood urea nitrogen, IGF1 and insulin, and increased FGF21 levels after 4 weeks76. These data support the idea that the beneficial effects of sulfur-containing amino acid restriction, seen in numerous animal models, are translatable to humans.

BMAT and nutrient stress in humans

Since the original description of the relationship between BMAT and the bone by Meunier and colleagues77, subsequent analyses of bone marrow biopsies in postmenopausal women have supported the negative relationship between BMD density and BMAT volume77,78. Within the past two decades, magnetic resonance (MR) spectroscopy has been used to quantify BMAT in the spine and proximal femur in women and to provide relative estimates of saturated fat percentage versus unsaturated fat percentage54,60,62,78,79. Increased fat volume and saturated fat in the marrow determined by MR spectroscopy have both been negatively associated with BMD, trabecular bone volume and cortical bone thickness measured by either areal measurements or volumetric CT78,79. Furthermore, MR spectroscopy provided in vivo evidence that oestrogen supplementation in postmenopausal women rapidly reduced BMAT and ultimately improved bone density80. Subsequent studies also showed that intermittent administration of parathyroid hormone for osteoporosis in men led to a reduction in the number and size of marrow adipocytes, as measured by histomorphometry66. Thus, under certain circumstances in humans, a clear negative relationship exists between BMAT and bone mass. However, those studies do not provide a causal mechanism, and as illustrated in mouse studies, the marrow adipogenic response could conceivably also be supportive of bone formation and serve as a source of fatty acids rather than as a cause of bone loss (Table 2).

Work from our laboratory has lent further credence to the dynamic nature of marrow adiposity in human volunteers, using MR spectroscopy, lipidomics and RNA-seq54. We have also provided a window into possible mechanisms that drive marrow adipogenesis. Healthy young men and women were admitted to the clinical research centre and given a high-calorie diet (HCD) for 10 days. They subsequently went home for 2 weeks and returned for a 10-day fast in the research centre. Bone marrow aspirates were performed pre-HCD and post-HCD and pre-fasting and post-fasting. Not surprisingly, the volunteers gained weight with the HCD and lost weight with fasting54. But after just 10 days, when measured by MR spectroscopy, BMAT had increased statistically significantly with both fasting and HCD, and the increase was greater in men than in women. These increases were reversed after returning to an ad libitum diet after the fast or HCD intervention. Surprisingly, trabecular bone volume measured by micro-CT was higher after the 10-day fast than at baseline, whereas no significant changes were noted in BMD measurements after the HCD. Bulk RNA-seq of marrow adipocytes isolated from the volunteers revealed that the three top gene ontology networks upregulated after fasting were the alternative complement activation pathway, response to stress and regulation of IGF transport54 (S. Costa, unpublished observation). The most upregulated genes were those that encode complement factor D (also known as adipsin) and lipoprotein lipase, and by lipidomic analysis, lipid storage and lipid droplet pathways were enhanced. Similarly, complement factor D has been shown to be markedly upregulated in bone marrow adipocytes from mice undergoing caloric restriction, and global deletion of Cfd results in rescue of bone mass and the diet-induced increase of BMAT81.

Unbiased approaches using tools such as single-cell and single-nucleus RNA-seq, as well as lipidomics and proteomics, can provide novel insights into the importance of BMAT relative to the marrow during states of nutritional stress82. However, single-cell RNA-seq studies of bone marrow adipocytes are challenging, owing to cell fragility. Single-nucleus RNA-seq has been accomplished in adipocytes and provides a window into lineage trajectories, although this analysis has not been accomplished yet in marrow adipocytes. Nevertheless, despite the nearly identical cellular phenotype of high BMAT with fasting or HCD, functional differences clearly exist in the bone marrow adipocyte. Importantly, caloric restriction drives an injury response in mice that can have major implications for both haematopoiesis and skeletal remodelling72. The recruitment of skeletal progenitors to both adipocytes and osteoblasts also occurs through IGF1 and its carrier proteins in the IGF-binding protein regulatory system83,84. High expression of LEPR after fasting in humans possibly reflects a recruitment of progenitors in adults; whether these cells are destined to become osteoblasts or adipocytes is not clear, but is consistent with mouse studies of caloric restriction19,72.

Conclusions

Although progress has been made in understanding the origin of marrow adipocytes, as well as the major regulatory determinants of BMAT, many unanswered questions remain about the functionality of marrow adipocytes under distinct conditions. For example, is long-term caloric restriction detrimental to the skeleton because of increased marrow adiposity and, if so, what is the mechanism? If in fact a detrimental effect does occur on the skeleton from long-term intermittent fasting, time-restricted eating or sustained caloric restriction, how does this effect manifest clinically? Studies of patients who underwent gastric bypass have shown that both bone loss and fractures can occur late after surgery85,86. Hence, longer-term calorie-restriction trials are needed with BMD and fractures as primary or secondary end points, respectively. Mechanistically, we still do not know the signal that drives the adipogenic and the osteogenic response to caloric restriction. For example, does a centrally mediated signal exist or does it arise locally from the haematopoietic marrow or possibly from compromised bone cells? In other words, what comes first: the adipogenic response to caloric restriction driven by a hormonal or neural signal, or a paracrine factor originating from a nutrient-compromised osteoblast or haematopoietic cell?

Another key question is what is the colour and function of the marrow adipocyte: beige, brown, white or other? Marrow adipocytes have a phenotype that distinguishes them from WAT and brown adipose tissue. However, whether beige adipocyte precursors reside in the bone marrow and can differentiate into functional cells is unclear, as is the unlikely possibility that the precursor of marrow adipocytes can also give rise to beige adipocytes. Beige precursor cells can be found embedded in WAT depots, most notably in the inguinal depot, whereas they are almost never seen in the perigonadal depot in mice87.

Only a small handful of reports suggest that the human bone marrow can beige88,89. For example, a bone marrow biopsy was taken from a 74-year-old man with untreated lymphoplasmacytic lymphoma. The marrow contained large numbers of lymphoma cells and “a large area of loosely aggregated adipose cells containing multiple cytoplasmic vacuoles, suggestive of brown adipocytes”. Whether these cells were brown or beige adipocytes is unknown. Moreover, whether these cells arose from the bone marrow or migrated from another site is equally unknown. The authors conclude that “this is an incidental intraosseous hibernoma”88. Other investigators have used in vitro culture systems and, in some experiments, gene expression has been used to show the presence of genes associated with beige adipocytes in bone marrow90,91. However, none of these experiments have used histology of freshly prepared bone to determine whether multilocular cells (indicative of brown or beige adipocytes with multiple small lipid droplets) are present in the bone marrow92.

Regulation of the beige adipocyte phenotype is complex. Inactivation of ZFP423 in mature adipocytes triggers a conversion of differentiated adiponectin-expressing iWAT and gonadal WAT (gWAT) or white adipocytes into beige adipocytes32. We have taken advantage of this effect to determine whether marrow adipocytes have the capacity to express a beige-like phenotype similar to that seen in iWAT and gWAT in this model. A terminally differentiated adipocyte-specific inducible Zfp423-knockout mouse model was generated32. Marrow adipogenesis after a sub-lethal dose of whole-body X-irradiation was measured in the tibia and femur of Zfp423-knockout and control mice using osmium staining with micro-CT32. As expected, irradiated Zfp423-knockout mice and control mice had a marked increase in marrow adipocytes. However, no multilocular marrow adipocytes could be observed in either the tibia or femurs of irradiated Zfp423-knockout mice. Irradiated control mice also had no multilocular cells (M.C.H. and R. K. Gupta, unpublished observation). By contrast, the loss of Zfp423 in iWAT and gWAT results in the appearance of numerous beige adipocytes in those depots32.

Preliminary findings have been obtained in adipocyte-specific Zfp423-knockout mice that were irradiated and then treated with CL-316243 (a potent β3-adrenergic agonist) to induce beiging. This protocol induces obvious beiging in iWAT and even in gWAT. However, no multilocular cells were observed in marrow adipocytes (M.C.H. and R. K. Gupta, unpublished observation). Thus, using the same inducible adipocyte-specific Zfp423-knockout mouse model, even under the most forceful conditions of beige-cell induction, did not allow marrow adipocytes to develop multilocular beige adipocytes.

One of the molecular hallmarks of beige and brown adipocytes is the expression of UCP1 (ref. 93). A fluorescent double reporter mT/mG mouse was used to trace Ucp1-expressing cells, demonstrating that Ucp1 is not expressed in marrow adipocytes94. Together, these data strongly suggest that mature marrow adipocytes in the mouse tibia and femur are unable to undergo beiging. These data also suggest that beige precursors are not present in the mouse bone marrow and support the idea that marrow adipocytes are distinct from the other adipocyte lineages.

Another provocative question to answer is whether BMAT has any importance or relevance in human evolution. It is generally accepted that the consumption of meat from large animals was instrumental in human evolution, enabling the development of large brains in early hominins, such as the australopithecines 3.4 million years ago95. These early ancestors of ours probably gathered meat from the kills of predatory animals because hunting large animals, with few tools, was dangerous. Scavenging carcasses would enable the hominins to break open the bones to access the nutrient-rich fatty bone marrow (which was not available to most other carnivores) and help support their high metabolism, body heat and brain development95,96. Although we have made great progress in understanding the origin, development, physiology and function of marrow adipocytes, we still retain some of the same traits of our hominin ancestors. Nowadays, we do not have to root through carcasses, and bone marrow is now better prepared, but we are still eating BMAT after millions of years of human evolution. As we look back across centuries, much progress has been made in understanding the bone marrow milieu and so, the marrow adipocyte can rightly take its place as an important functional component of the bone marrow.